xref: /dragonfly/sys/kern/kern_sig.c (revision e8c03636)
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 
83 static int	filt_sigattach(struct knote *kn);
84 static void	filt_sigdetach(struct knote *kn);
85 static int	filt_signal(struct knote *kn, long hint);
86 
87 struct filterops sig_filtops =
88 	{ 0, filt_sigattach, filt_sigdetach, filt_signal };
89 
90 static int	kern_logsigexit = 1;
91 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
92     &kern_logsigexit, 0,
93     "Log processes quitting on abnormal signals to syslog(3)");
94 
95 /*
96  * Can process p, with pcred pc, send the signal sig to process q?
97  */
98 #define CANSIGNAL(q, sig) \
99 	(!p_trespass(curproc->p_ucred, (q)->p_ucred) || \
100 	((sig) == SIGCONT && (q)->p_session == curproc->p_session))
101 
102 /*
103  * Policy -- Can real uid ruid with ucred uc send a signal to process q?
104  */
105 #define CANSIGIO(ruid, uc, q) \
106 	((uc)->cr_uid == 0 || \
107 	    (ruid) == (q)->p_ucred->cr_ruid || \
108 	    (uc)->cr_uid == (q)->p_ucred->cr_ruid || \
109 	    (ruid) == (q)->p_ucred->cr_uid || \
110 	    (uc)->cr_uid == (q)->p_ucred->cr_uid)
111 
112 int sugid_coredump;
113 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
114 	&sugid_coredump, 0, "Enable coredumping set user/group ID processes");
115 
116 static int	do_coredump = 1;
117 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
118 	&do_coredump, 0, "Enable/Disable coredumps");
119 
120 /*
121  * Signal properties and actions.
122  * The array below categorizes the signals and their default actions
123  * according to the following properties:
124  */
125 #define	SA_KILL		0x01		/* terminates process by default */
126 #define	SA_CORE		0x02		/* ditto and coredumps */
127 #define	SA_STOP		0x04		/* suspend process */
128 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
129 #define	SA_IGNORE	0x10		/* ignore by default */
130 #define	SA_CONT		0x20		/* continue if suspended */
131 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
132 #define SA_CKPT         0x80            /* checkpoint process */
133 
134 
135 static int sigproptbl[NSIG] = {
136         SA_KILL,                /* SIGHUP */
137         SA_KILL,                /* SIGINT */
138         SA_KILL|SA_CORE,        /* SIGQUIT */
139         SA_KILL|SA_CORE,        /* SIGILL */
140         SA_KILL|SA_CORE,        /* SIGTRAP */
141         SA_KILL|SA_CORE,        /* SIGABRT */
142         SA_KILL|SA_CORE,        /* SIGEMT */
143         SA_KILL|SA_CORE,        /* SIGFPE */
144         SA_KILL,                /* SIGKILL */
145         SA_KILL|SA_CORE,        /* SIGBUS */
146         SA_KILL|SA_CORE,        /* SIGSEGV */
147         SA_KILL|SA_CORE,        /* SIGSYS */
148         SA_KILL,                /* SIGPIPE */
149         SA_KILL,                /* SIGALRM */
150         SA_KILL,                /* SIGTERM */
151         SA_IGNORE,              /* SIGURG */
152         SA_STOP,                /* SIGSTOP */
153         SA_STOP|SA_TTYSTOP,     /* SIGTSTP */
154         SA_IGNORE|SA_CONT,      /* SIGCONT */
155         SA_IGNORE,              /* SIGCHLD */
156         SA_STOP|SA_TTYSTOP,     /* SIGTTIN */
157         SA_STOP|SA_TTYSTOP,     /* SIGTTOU */
158         SA_IGNORE,              /* SIGIO */
159         SA_KILL,                /* SIGXCPU */
160         SA_KILL,                /* SIGXFSZ */
161         SA_KILL,                /* SIGVTALRM */
162         SA_KILL,                /* SIGPROF */
163         SA_IGNORE,              /* SIGWINCH  */
164         SA_IGNORE,              /* SIGINFO */
165         SA_KILL,                /* SIGUSR1 */
166         SA_KILL,                /* SIGUSR2 */
167 	SA_IGNORE,              /* SIGTHR */
168 	SA_CKPT,                /* SIGCKPT */
169 	SA_KILL|SA_CKPT,        /* SIGCKPTEXIT */
170 	SA_IGNORE,
171 	SA_IGNORE,
172 	SA_IGNORE,
173 	SA_IGNORE,
174 	SA_IGNORE,
175 	SA_IGNORE,
176 	SA_IGNORE,
177 	SA_IGNORE,
178 	SA_IGNORE,
179 	SA_IGNORE,
180 	SA_IGNORE,
181 	SA_IGNORE,
182 	SA_IGNORE,
183 	SA_IGNORE,
184 	SA_IGNORE,
185 	SA_IGNORE,
186 	SA_IGNORE,
187 	SA_IGNORE,
188 	SA_IGNORE,
189 	SA_IGNORE,
190 	SA_IGNORE,
191 	SA_IGNORE,
192 	SA_IGNORE,
193 	SA_IGNORE,
194 	SA_IGNORE,
195 	SA_IGNORE,
196 	SA_IGNORE,
197 	SA_IGNORE,
198 	SA_IGNORE,
199 	SA_IGNORE,
200 
201 };
202 
203 static __inline int
204 sigprop(int sig)
205 {
206 
207 	if (sig > 0 && sig < NSIG)
208 		return (sigproptbl[_SIG_IDX(sig)]);
209 	return (0);
210 }
211 
212 static __inline int
213 sig_ffs(sigset_t *set)
214 {
215 	int i;
216 
217 	for (i = 0; i < _SIG_WORDS; i++)
218 		if (set->__bits[i])
219 			return (ffs(set->__bits[i]) + (i * 32));
220 	return (0);
221 }
222 
223 /*
224  * No requirements.
225  */
226 int
227 kern_sigaction(int sig, struct sigaction *act, struct sigaction *oact)
228 {
229 	struct thread *td = curthread;
230 	struct proc *p = td->td_proc;
231 	struct lwp *lp;
232 	struct sigacts *ps = p->p_sigacts;
233 
234 	if (sig <= 0 || sig > _SIG_MAXSIG)
235 		return (EINVAL);
236 
237 	lwkt_gettoken(&p->p_token);
238 
239 	if (oact) {
240 		oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
241 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
242 		oact->sa_flags = 0;
243 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
244 			oact->sa_flags |= SA_ONSTACK;
245 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
246 			oact->sa_flags |= SA_RESTART;
247 		if (SIGISMEMBER(ps->ps_sigreset, sig))
248 			oact->sa_flags |= SA_RESETHAND;
249 		if (SIGISMEMBER(ps->ps_signodefer, sig))
250 			oact->sa_flags |= SA_NODEFER;
251 		if (SIGISMEMBER(ps->ps_siginfo, sig))
252 			oact->sa_flags |= SA_SIGINFO;
253 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDSTOP)
254 			oact->sa_flags |= SA_NOCLDSTOP;
255 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDWAIT)
256 			oact->sa_flags |= SA_NOCLDWAIT;
257 	}
258 	if (act) {
259 		/*
260 		 * Check for invalid requests.  KILL and STOP cannot be
261 		 * caught.
262 		 */
263 		if (sig == SIGKILL || sig == SIGSTOP) {
264 			if (act->sa_handler != SIG_DFL) {
265 				lwkt_reltoken(&p->p_token);
266 				return (EINVAL);
267 			}
268 		}
269 
270 		/*
271 		 * Change setting atomically.
272 		 */
273 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
274 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
275 		if (act->sa_flags & SA_SIGINFO) {
276 			ps->ps_sigact[_SIG_IDX(sig)] =
277 			    (__sighandler_t *)act->sa_sigaction;
278 			SIGADDSET(ps->ps_siginfo, sig);
279 		} else {
280 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
281 			SIGDELSET(ps->ps_siginfo, sig);
282 		}
283 		if (!(act->sa_flags & SA_RESTART))
284 			SIGADDSET(ps->ps_sigintr, sig);
285 		else
286 			SIGDELSET(ps->ps_sigintr, sig);
287 		if (act->sa_flags & SA_ONSTACK)
288 			SIGADDSET(ps->ps_sigonstack, sig);
289 		else
290 			SIGDELSET(ps->ps_sigonstack, sig);
291 		if (act->sa_flags & SA_RESETHAND)
292 			SIGADDSET(ps->ps_sigreset, sig);
293 		else
294 			SIGDELSET(ps->ps_sigreset, sig);
295 		if (act->sa_flags & SA_NODEFER)
296 			SIGADDSET(ps->ps_signodefer, sig);
297 		else
298 			SIGDELSET(ps->ps_signodefer, sig);
299 		if (sig == SIGCHLD) {
300 			if (act->sa_flags & SA_NOCLDSTOP)
301 				p->p_sigacts->ps_flag |= PS_NOCLDSTOP;
302 			else
303 				p->p_sigacts->ps_flag &= ~PS_NOCLDSTOP;
304 			if (act->sa_flags & SA_NOCLDWAIT) {
305 				/*
306 				 * Paranoia: since SA_NOCLDWAIT is implemented
307 				 * by reparenting the dying child to PID 1 (and
308 				 * trust it to reap the zombie), PID 1 itself
309 				 * is forbidden to set SA_NOCLDWAIT.
310 				 */
311 				if (p->p_pid == 1)
312 					p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
313 				else
314 					p->p_sigacts->ps_flag |= PS_NOCLDWAIT;
315 			} else {
316 				p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
317 			}
318 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
319 				ps->ps_flag |= PS_CLDSIGIGN;
320 			else
321 				ps->ps_flag &= ~PS_CLDSIGIGN;
322 		}
323 		/*
324 		 * Set bit in p_sigignore for signals that are set to SIG_IGN,
325 		 * and for signals set to SIG_DFL where the default is to
326 		 * ignore. However, don't put SIGCONT in p_sigignore, as we
327 		 * have to restart the process.
328 		 *
329 		 * Also remove the signal from the process and lwp signal
330 		 * list.
331 		 */
332 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
333 		    (sigprop(sig) & SA_IGNORE &&
334 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
335 			SIGDELSET(p->p_siglist, sig);
336 			FOREACH_LWP_IN_PROC(lp, p) {
337 				spin_lock(&lp->lwp_spin);
338 				SIGDELSET(lp->lwp_siglist, sig);
339 				spin_unlock(&lp->lwp_spin);
340 			}
341 			if (sig != SIGCONT) {
342 				/* easier in ksignal */
343 				SIGADDSET(p->p_sigignore, sig);
344 			}
345 			SIGDELSET(p->p_sigcatch, sig);
346 		} else {
347 			SIGDELSET(p->p_sigignore, sig);
348 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
349 				SIGDELSET(p->p_sigcatch, sig);
350 			else
351 				SIGADDSET(p->p_sigcatch, sig);
352 		}
353 	}
354 	lwkt_reltoken(&p->p_token);
355 	return (0);
356 }
357 
358 int
359 sys_sigaction(struct sigaction_args *uap)
360 {
361 	struct sigaction act, oact;
362 	struct sigaction *actp, *oactp;
363 	int error;
364 
365 	actp = (uap->act != NULL) ? &act : NULL;
366 	oactp = (uap->oact != NULL) ? &oact : NULL;
367 	if (actp) {
368 		error = copyin(uap->act, actp, sizeof(act));
369 		if (error)
370 			return (error);
371 	}
372 	error = kern_sigaction(uap->sig, actp, oactp);
373 	if (oactp && !error) {
374 		error = copyout(oactp, uap->oact, sizeof(oact));
375 	}
376 	return (error);
377 }
378 
379 /*
380  * Initialize signal state for process 0;
381  * set to ignore signals that are ignored by default.
382  */
383 void
384 siginit(struct proc *p)
385 {
386 	int i;
387 
388 	for (i = 1; i <= NSIG; i++)
389 		if (sigprop(i) & SA_IGNORE && i != SIGCONT)
390 			SIGADDSET(p->p_sigignore, i);
391 }
392 
393 /*
394  * Reset signals for an exec of the specified process.
395  */
396 void
397 execsigs(struct proc *p)
398 {
399 	struct sigacts *ps = p->p_sigacts;
400 	struct lwp *lp;
401 	int sig;
402 
403 	lp = ONLY_LWP_IN_PROC(p);
404 
405 	/*
406 	 * Reset caught signals.  Held signals remain held
407 	 * through p_sigmask (unless they were caught,
408 	 * and are now ignored by default).
409 	 */
410 	while (SIGNOTEMPTY(p->p_sigcatch)) {
411 		sig = sig_ffs(&p->p_sigcatch);
412 		SIGDELSET(p->p_sigcatch, sig);
413 		if (sigprop(sig) & SA_IGNORE) {
414 			if (sig != SIGCONT)
415 				SIGADDSET(p->p_sigignore, sig);
416 			SIGDELSET(p->p_siglist, sig);
417 			/* don't need spinlock */
418 			SIGDELSET(lp->lwp_siglist, sig);
419 		}
420 		ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
421 	}
422 
423 	/*
424 	 * Reset stack state to the user stack.
425 	 * Clear set of signals caught on the signal stack.
426 	 */
427 	lp->lwp_sigstk.ss_flags = SS_DISABLE;
428 	lp->lwp_sigstk.ss_size = 0;
429 	lp->lwp_sigstk.ss_sp = NULL;
430 	lp->lwp_flags &= ~LWP_ALTSTACK;
431 	/*
432 	 * Reset no zombies if child dies flag as Solaris does.
433 	 */
434 	p->p_sigacts->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
435 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
436 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
437 }
438 
439 /*
440  * kern_sigprocmask() - MP SAFE ONLY IF p == curproc
441  *
442  *	Manipulate signal mask.  This routine is MP SAFE *ONLY* if
443  *	p == curproc.
444  */
445 int
446 kern_sigprocmask(int how, sigset_t *set, sigset_t *oset)
447 {
448 	struct thread *td = curthread;
449 	struct lwp *lp = td->td_lwp;
450 	struct proc *p = td->td_proc;
451 	int error;
452 
453 	lwkt_gettoken(&p->p_token);
454 
455 	if (oset != NULL)
456 		*oset = lp->lwp_sigmask;
457 
458 	error = 0;
459 	if (set != NULL) {
460 		switch (how) {
461 		case SIG_BLOCK:
462 			SIG_CANTMASK(*set);
463 			SIGSETOR(lp->lwp_sigmask, *set);
464 			break;
465 		case SIG_UNBLOCK:
466 			SIGSETNAND(lp->lwp_sigmask, *set);
467 			break;
468 		case SIG_SETMASK:
469 			SIG_CANTMASK(*set);
470 			lp->lwp_sigmask = *set;
471 			break;
472 		default:
473 			error = EINVAL;
474 			break;
475 		}
476 	}
477 
478 	lwkt_reltoken(&p->p_token);
479 
480 	return (error);
481 }
482 
483 /*
484  * sigprocmask()
485  *
486  * MPSAFE
487  */
488 int
489 sys_sigprocmask(struct sigprocmask_args *uap)
490 {
491 	sigset_t set, oset;
492 	sigset_t *setp, *osetp;
493 	int error;
494 
495 	setp = (uap->set != NULL) ? &set : NULL;
496 	osetp = (uap->oset != NULL) ? &oset : NULL;
497 	if (setp) {
498 		error = copyin(uap->set, setp, sizeof(set));
499 		if (error)
500 			return (error);
501 	}
502 	error = kern_sigprocmask(uap->how, setp, osetp);
503 	if (osetp && !error) {
504 		error = copyout(osetp, uap->oset, sizeof(oset));
505 	}
506 	return (error);
507 }
508 
509 /*
510  * MPSAFE
511  */
512 int
513 kern_sigpending(struct __sigset *set)
514 {
515 	struct lwp *lp = curthread->td_lwp;
516 
517 	*set = lwp_sigpend(lp);
518 
519 	return (0);
520 }
521 
522 /*
523  * MPSAFE
524  */
525 int
526 sys_sigpending(struct sigpending_args *uap)
527 {
528 	sigset_t set;
529 	int error;
530 
531 	error = kern_sigpending(&set);
532 
533 	if (error == 0)
534 		error = copyout(&set, uap->set, sizeof(set));
535 	return (error);
536 }
537 
538 /*
539  * Suspend process until signal, providing mask to be set
540  * in the meantime.
541  *
542  * MPSAFE
543  */
544 int
545 kern_sigsuspend(struct __sigset *set)
546 {
547 	struct thread *td = curthread;
548 	struct lwp *lp = td->td_lwp;
549 	struct proc *p = td->td_proc;
550 	struct sigacts *ps = p->p_sigacts;
551 
552 	/*
553 	 * When returning from sigsuspend, we want
554 	 * the old mask to be restored after the
555 	 * signal handler has finished.  Thus, we
556 	 * save it here and mark the sigacts structure
557 	 * to indicate this.
558 	 */
559 	lp->lwp_oldsigmask = lp->lwp_sigmask;
560 	lp->lwp_flags |= LWP_OLDMASK;
561 
562 	SIG_CANTMASK(*set);
563 	lp->lwp_sigmask = *set;
564 	while (tsleep(ps, PCATCH, "pause", 0) == 0)
565 		/* void */;
566 	/* always return EINTR rather than ERESTART... */
567 	return (EINTR);
568 }
569 
570 /*
571  * Note nonstandard calling convention: libc stub passes mask, not
572  * pointer, to save a copyin.
573  *
574  * MPSAFE
575  */
576 int
577 sys_sigsuspend(struct sigsuspend_args *uap)
578 {
579 	sigset_t mask;
580 	int error;
581 
582 	error = copyin(uap->sigmask, &mask, sizeof(mask));
583 	if (error)
584 		return (error);
585 
586 	error = kern_sigsuspend(&mask);
587 
588 	return (error);
589 }
590 
591 /*
592  * MPSAFE
593  */
594 int
595 kern_sigaltstack(struct sigaltstack *ss, struct sigaltstack *oss)
596 {
597 	struct thread *td = curthread;
598 	struct lwp *lp = td->td_lwp;
599 	struct proc *p = td->td_proc;
600 
601 	if ((lp->lwp_flags & LWP_ALTSTACK) == 0)
602 		lp->lwp_sigstk.ss_flags |= SS_DISABLE;
603 
604 	if (oss)
605 		*oss = lp->lwp_sigstk;
606 
607 	if (ss) {
608 		if (ss->ss_flags & ~SS_DISABLE)
609 			return (EINVAL);
610 		if (ss->ss_flags & SS_DISABLE) {
611 			if (lp->lwp_sigstk.ss_flags & SS_ONSTACK)
612 				return (EINVAL);
613 			lp->lwp_flags &= ~LWP_ALTSTACK;
614 			lp->lwp_sigstk.ss_flags = ss->ss_flags;
615 		} else {
616 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
617 				return (ENOMEM);
618 			lp->lwp_flags |= LWP_ALTSTACK;
619 			lp->lwp_sigstk = *ss;
620 		}
621 	}
622 
623 	return (0);
624 }
625 
626 /*
627  * MPSAFE
628  */
629 int
630 sys_sigaltstack(struct sigaltstack_args *uap)
631 {
632 	stack_t ss, oss;
633 	int error;
634 
635 	if (uap->ss) {
636 		error = copyin(uap->ss, &ss, sizeof(ss));
637 		if (error)
638 			return (error);
639 	}
640 
641 	error = kern_sigaltstack(uap->ss ? &ss : NULL,
642 	    uap->oss ? &oss : NULL);
643 
644 	if (error == 0 && uap->oss)
645 		error = copyout(&oss, uap->oss, sizeof(*uap->oss));
646 	return (error);
647 }
648 
649 /*
650  * Common code for kill process group/broadcast kill.
651  * cp is calling process.
652  */
653 struct killpg_info {
654 	int nfound;
655 	int sig;
656 };
657 
658 static int killpg_all_callback(struct proc *p, void *data);
659 
660 static int
661 dokillpg(int sig, int pgid, int all)
662 {
663 	struct killpg_info info;
664 	struct proc *cp = curproc;
665 	struct proc *p;
666 	struct pgrp *pgrp;
667 
668 	info.nfound = 0;
669 	info.sig = sig;
670 
671 	if (all) {
672 		/*
673 		 * broadcast
674 		 */
675 		allproc_scan(killpg_all_callback, &info);
676 	} else {
677 		if (pgid == 0) {
678 			/*
679 			 * zero pgid means send to my process group.
680 			 */
681 			pgrp = cp->p_pgrp;
682 			pgref(pgrp);
683 		} else {
684 			pgrp = pgfind(pgid);
685 			if (pgrp == NULL)
686 				return (ESRCH);
687 		}
688 
689 		/*
690 		 * Must interlock all signals against fork
691 		 */
692 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
693 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
694 			if (p->p_pid <= 1 ||
695 			    p->p_stat == SZOMB ||
696 			    (p->p_flags & P_SYSTEM) ||
697 			    !CANSIGNAL(p, sig)) {
698 				continue;
699 			}
700 			++info.nfound;
701 			if (sig)
702 				ksignal(p, sig);
703 		}
704 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
705 		pgrel(pgrp);
706 	}
707 	return (info.nfound ? 0 : ESRCH);
708 }
709 
710 static int
711 killpg_all_callback(struct proc *p, void *data)
712 {
713 	struct killpg_info *info = data;
714 
715 	if (p->p_pid <= 1 || (p->p_flags & P_SYSTEM) ||
716 	    p == curproc || !CANSIGNAL(p, info->sig)) {
717 		return (0);
718 	}
719 	++info->nfound;
720 	if (info->sig)
721 		ksignal(p, info->sig);
722 	return(0);
723 }
724 
725 /*
726  * Send a general signal to a process or LWPs within that process.
727  *
728  * Note that new signals cannot be sent if a process is exiting or already
729  * a zombie, but we return success anyway as userland is likely to not handle
730  * the race properly.
731  *
732  * No requirements.
733  */
734 int
735 kern_kill(int sig, pid_t pid, lwpid_t tid)
736 {
737 	int t;
738 
739 	if ((u_int)sig > _SIG_MAXSIG)
740 		return (EINVAL);
741 
742 	lwkt_gettoken(&proc_token);
743 
744 	if (pid > 0) {
745 		struct proc *p;
746 		struct lwp *lp = NULL;
747 
748 		/*
749 		 * Send a signal to a single process.  If the kill() is
750 		 * racing an exiting process which has not yet been reaped
751 		 * act as though the signal was delivered successfully but
752 		 * don't actually try to deliver the signal.
753 		 */
754 		if ((p = pfind(pid)) == NULL) {
755 			if ((p = zpfind(pid)) == NULL) {
756 				lwkt_reltoken(&proc_token);
757 				return (ESRCH);
758 			}
759 			lwkt_reltoken(&proc_token);
760 			PRELE(p);
761 			return (0);
762 		}
763 		lwkt_gettoken(&p->p_token);
764 		if (!CANSIGNAL(p, sig)) {
765 			lwkt_reltoken(&p->p_token);
766 			PRELE(p);
767 			lwkt_reltoken(&proc_token);
768 			return (EPERM);
769 		}
770 
771 		/*
772 		 * NOP if the process is exiting.  Note that lwpsignal() is
773 		 * called directly with P_WEXIT set to kill individual LWPs
774 		 * during exit, which is allowed.
775 		 */
776 		if (p->p_flags & P_WEXIT) {
777 			lwkt_reltoken(&p->p_token);
778 			PRELE(p);
779 			lwkt_reltoken(&proc_token);
780 			return (0);
781 		}
782 		if (tid != -1) {
783 			lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
784 			if (lp == NULL) {
785 				lwkt_reltoken(&p->p_token);
786 				PRELE(p);
787 				lwkt_reltoken(&proc_token);
788 				return (ESRCH);
789 			}
790 		}
791 		if (sig)
792 			lwpsignal(p, lp, sig);
793 		lwkt_reltoken(&p->p_token);
794 		PRELE(p);
795 		lwkt_reltoken(&proc_token);
796 		return (0);
797 	}
798 
799 	/*
800 	 * If we come here, pid is a special broadcast pid.
801 	 * This doesn't mix with a tid.
802 	 */
803 	if (tid != -1) {
804 		lwkt_reltoken(&proc_token);
805 		return (EINVAL);
806 	}
807 	switch (pid) {
808 	case -1:		/* broadcast signal */
809 		t = (dokillpg(sig, 0, 1));
810 		break;
811 	case 0:			/* signal own process group */
812 		t = (dokillpg(sig, 0, 0));
813 		break;
814 	default:		/* negative explicit process group */
815 		t = (dokillpg(sig, -pid, 0));
816 		break;
817 	}
818 	lwkt_reltoken(&proc_token);
819 	return t;
820 }
821 
822 int
823 sys_kill(struct kill_args *uap)
824 {
825 	int error;
826 
827 	error = kern_kill(uap->signum, uap->pid, -1);
828 	return (error);
829 }
830 
831 int
832 sys_lwp_kill(struct lwp_kill_args *uap)
833 {
834 	int error;
835 	pid_t pid = uap->pid;
836 
837 	/*
838 	 * A tid is mandatory for lwp_kill(), otherwise
839 	 * you could simply use kill().
840 	 */
841 	if (uap->tid == -1)
842 		return (EINVAL);
843 
844 	/*
845 	 * To save on a getpid() function call for intra-process
846 	 * signals, pid == -1 means current process.
847 	 */
848 	if (pid == -1)
849 		pid = curproc->p_pid;
850 
851 	error = kern_kill(uap->signum, pid, uap->tid);
852 	return (error);
853 }
854 
855 /*
856  * Send a signal to a process group.
857  */
858 void
859 gsignal(int pgid, int sig)
860 {
861 	struct pgrp *pgrp;
862 
863 	if (pgid && (pgrp = pgfind(pgid)))
864 		pgsignal(pgrp, sig, 0);
865 }
866 
867 /*
868  * Send a signal to a process group.  If checktty is 1,
869  * limit to members which have a controlling terminal.
870  *
871  * pg_lock interlocks against a fork that might be in progress, to
872  * ensure that the new child process picks up the signal.
873  */
874 void
875 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
876 {
877 	struct proc *p;
878 
879 	/*
880 	 * Must interlock all signals against fork
881 	 */
882 	if (pgrp) {
883 		pgref(pgrp);
884 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
885 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
886 			if (checkctty == 0 || p->p_flags & P_CONTROLT)
887 				ksignal(p, sig);
888 		}
889 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
890 		pgrel(pgrp);
891 	}
892 }
893 
894 /*
895  * Send a signal caused by a trap to the current lwp.  If it will be caught
896  * immediately, deliver it with correct code.  Otherwise, post it normally.
897  *
898  * These signals may ONLY be delivered to the specified lwp and may never
899  * be delivered to the process generically.
900  */
901 void
902 trapsignal(struct lwp *lp, int sig, u_long code)
903 {
904 	struct proc *p = lp->lwp_proc;
905 	struct sigacts *ps = p->p_sigacts;
906 
907 	/*
908 	 * If we are a virtual kernel running an emulated user process
909 	 * context, switch back to the virtual kernel context before
910 	 * trying to post the signal.
911 	 */
912 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
913 		struct trapframe *tf = lp->lwp_md.md_regs;
914 		tf->tf_trapno = 0;
915 		vkernel_trap(lp, tf);
916 	}
917 
918 
919 	if ((p->p_flags & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
920 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
921 		lp->lwp_ru.ru_nsignals++;
922 #ifdef KTRACE
923 		if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
924 			ktrpsig(lp, sig, ps->ps_sigact[_SIG_IDX(sig)],
925 				&lp->lwp_sigmask, code);
926 #endif
927 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
928 						&lp->lwp_sigmask, code);
929 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
930 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
931 			SIGADDSET(lp->lwp_sigmask, sig);
932 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
933 			/*
934 			 * See kern_sigaction() for origin of this code.
935 			 */
936 			SIGDELSET(p->p_sigcatch, sig);
937 			if (sig != SIGCONT &&
938 			    sigprop(sig) & SA_IGNORE)
939 				SIGADDSET(p->p_sigignore, sig);
940 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
941 		}
942 	} else {
943 		lp->lwp_code = code;	/* XXX for core dump/debugger */
944 		lp->lwp_sig = sig;	/* XXX to verify code */
945 		lwpsignal(p, lp, sig);
946 	}
947 }
948 
949 /*
950  * Find a suitable lwp to deliver the signal to.  Returns NULL if all
951  * lwps hold the signal blocked.
952  *
953  * Caller must hold p->p_token.
954  *
955  * Returns a lp or NULL.  If non-NULL the lp is held and its token is
956  * acquired.
957  */
958 static struct lwp *
959 find_lwp_for_signal(struct proc *p, int sig)
960 {
961 	struct lwp *lp;
962 	struct lwp *run, *sleep, *stop;
963 
964 	/*
965 	 * If the running/preempted thread belongs to the proc to which
966 	 * the signal is being delivered and this thread does not block
967 	 * the signal, then we can avoid a context switch by delivering
968 	 * the signal to this thread, because it will return to userland
969 	 * soon anyways.
970 	 */
971 	lp = lwkt_preempted_proc();
972 	if (lp != NULL && lp->lwp_proc == p) {
973 		LWPHOLD(lp);
974 		lwkt_gettoken(&lp->lwp_token);
975 		if (!SIGISMEMBER(lp->lwp_sigmask, sig)) {
976 			/* return w/ token held */
977 			return (lp);
978 		}
979 		lwkt_reltoken(&lp->lwp_token);
980 		LWPRELE(lp);
981 	}
982 
983 	run = sleep = stop = NULL;
984 	FOREACH_LWP_IN_PROC(lp, p) {
985 		/*
986 		 * If the signal is being blocked by the lwp, then this
987 		 * lwp is not eligible for receiving the signal.
988 		 */
989 		LWPHOLD(lp);
990 		lwkt_gettoken(&lp->lwp_token);
991 
992 		if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
993 			lwkt_reltoken(&lp->lwp_token);
994 			LWPRELE(lp);
995 			continue;
996 		}
997 
998 		switch (lp->lwp_stat) {
999 		case LSRUN:
1000 			if (sleep) {
1001 				lwkt_token_swap();
1002 				lwkt_reltoken(&sleep->lwp_token);
1003 				LWPRELE(sleep);
1004 				sleep = NULL;
1005 				run = lp;
1006 			} else if (stop) {
1007 				lwkt_token_swap();
1008 				lwkt_reltoken(&stop->lwp_token);
1009 				LWPRELE(stop);
1010 				stop = NULL;
1011 				run = lp;
1012 			} else {
1013 				run = lp;
1014 			}
1015 			break;
1016 		case LSSLEEP:
1017 			if (lp->lwp_flags & LWP_SINTR) {
1018 				if (sleep) {
1019 					lwkt_reltoken(&lp->lwp_token);
1020 					LWPRELE(lp);
1021 				} else if (stop) {
1022 					lwkt_token_swap();
1023 					lwkt_reltoken(&stop->lwp_token);
1024 					LWPRELE(stop);
1025 					stop = NULL;
1026 					sleep = lp;
1027 				} else {
1028 					sleep = lp;
1029 				}
1030 			} else {
1031 				lwkt_reltoken(&lp->lwp_token);
1032 				LWPRELE(lp);
1033 			}
1034 			break;
1035 		case LSSTOP:
1036 			if (sleep) {
1037 				lwkt_reltoken(&lp->lwp_token);
1038 				LWPRELE(lp);
1039 			} else if (stop) {
1040 				lwkt_reltoken(&lp->lwp_token);
1041 				LWPRELE(lp);
1042 			} else {
1043 				stop = lp;
1044 			}
1045 			break;
1046 		}
1047 		if (run)
1048 			break;
1049 	}
1050 
1051 	if (run != NULL)
1052 		return (run);
1053 	else if (sleep != NULL)
1054 		return (sleep);
1055 	else
1056 		return (stop);
1057 }
1058 
1059 /*
1060  * Send the signal to the process.  If the signal has an action, the action
1061  * is usually performed by the target process rather than the caller; we add
1062  * the signal to the set of pending signals for the process.
1063  *
1064  * Exceptions:
1065  *   o When a stop signal is sent to a sleeping process that takes the
1066  *     default action, the process is stopped without awakening it.
1067  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
1068  *     regardless of the signal action (eg, blocked or ignored).
1069  *
1070  * Other ignored signals are discarded immediately.
1071  *
1072  * If the caller wishes to call this function from a hard code section the
1073  * caller must already hold p->p_token (see kern_clock.c).
1074  *
1075  * No requirements.
1076  */
1077 void
1078 ksignal(struct proc *p, int sig)
1079 {
1080 	lwpsignal(p, NULL, sig);
1081 }
1082 
1083 /*
1084  * The core for ksignal.  lp may be NULL, then a suitable thread
1085  * will be chosen.  If not, lp MUST be a member of p.
1086  *
1087  * If the caller wishes to call this function from a hard code section the
1088  * caller must already hold p->p_token.
1089  *
1090  * No requirements.
1091  */
1092 void
1093 lwpsignal(struct proc *p, struct lwp *lp, int sig)
1094 {
1095 	struct proc *q;
1096 	sig_t action;
1097 	int prop;
1098 
1099 	if (sig > _SIG_MAXSIG || sig <= 0) {
1100 		kprintf("lwpsignal: signal %d\n", sig);
1101 		panic("lwpsignal signal number");
1102 	}
1103 
1104 	KKASSERT(lp == NULL || lp->lwp_proc == p);
1105 
1106 	/*
1107 	 * We don't want to race... well, all sorts of things.  Get appropriate
1108 	 * tokens.
1109 	 *
1110 	 * Don't try to deliver a generic signal to an exiting process,
1111 	 * the signal structures could be in flux.  We check the LWP later
1112 	 * on.
1113 	 */
1114 	PHOLD(p);
1115 	lwkt_gettoken(&p->p_token);
1116 	if (lp) {
1117 		LWPHOLD(lp);
1118 		lwkt_gettoken(&lp->lwp_token);
1119 	} else if (p->p_flags & P_WEXIT) {
1120 		goto out;
1121 	}
1122 
1123 	prop = sigprop(sig);
1124 
1125 	/*
1126 	 * If proc is traced, always give parent a chance;
1127 	 * if signal event is tracked by procfs, give *that*
1128 	 * a chance, as well.
1129 	 */
1130 	if ((p->p_flags & P_TRACED) || (p->p_stops & S_SIG)) {
1131 		action = SIG_DFL;
1132 	} else {
1133 		/*
1134 		 * Do not try to deliver signals to an exiting lwp.  Note
1135 		 * that we must still deliver the signal if P_WEXIT is set
1136 		 * in the process flags.
1137 		 */
1138 		if (lp && (lp->lwp_mpflags & LWP_MP_WEXIT)) {
1139 			if (lp) {
1140 				lwkt_reltoken(&lp->lwp_token);
1141 				LWPRELE(lp);
1142 			}
1143 			lwkt_reltoken(&p->p_token);
1144 			PRELE(p);
1145 			return;
1146 		}
1147 
1148 		/*
1149 		 * If the signal is being ignored, then we forget about
1150 		 * it immediately.  NOTE: We don't set SIGCONT in p_sigignore,
1151 		 * and if it is set to SIG_IGN, action will be SIG_DFL here.
1152 		 */
1153 		if (SIGISMEMBER(p->p_sigignore, sig)) {
1154 			/*
1155 			 * Even if a signal is set SIG_IGN, it may still be
1156 			 * lurking in a kqueue.
1157 			 */
1158 			KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1159 			if (lp) {
1160 				lwkt_reltoken(&lp->lwp_token);
1161 				LWPRELE(lp);
1162 			}
1163 			lwkt_reltoken(&p->p_token);
1164 			PRELE(p);
1165 			return;
1166 		}
1167 		if (SIGISMEMBER(p->p_sigcatch, sig))
1168 			action = SIG_CATCH;
1169 		else
1170 			action = SIG_DFL;
1171 	}
1172 
1173 	/*
1174 	 * If continuing, clear any pending STOP signals.
1175 	 */
1176 	if (prop & SA_CONT)
1177 		SIG_STOPSIGMASK(p->p_siglist);
1178 
1179 	if (prop & SA_STOP) {
1180 		/*
1181 		 * If sending a tty stop signal to a member of an orphaned
1182 		 * process group, discard the signal here if the action
1183 		 * is default; don't stop the process below if sleeping,
1184 		 * and don't clear any pending SIGCONT.
1185 		 */
1186 		if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0 &&
1187 		    action == SIG_DFL) {
1188 			lwkt_reltoken(&p->p_token);
1189 			PRELE(p);
1190 		        return;
1191 		}
1192 		SIG_CONTSIGMASK(p->p_siglist);
1193 		p->p_flags &= ~P_CONTINUED;
1194 	}
1195 
1196 	if (p->p_stat == SSTOP) {
1197 		/*
1198 		 * Nobody can handle this signal, add it to the lwp or
1199 		 * process pending list
1200 		 */
1201 		if (lp) {
1202 			spin_lock(&lp->lwp_spin);
1203 			SIGADDSET(lp->lwp_siglist, sig);
1204 			spin_unlock(&lp->lwp_spin);
1205 		} else {
1206 			SIGADDSET(p->p_siglist, sig);
1207 		}
1208 
1209 		/*
1210 		 * If the process is stopped and is being traced, then no
1211 		 * further action is necessary.
1212 		 */
1213 		if (p->p_flags & P_TRACED)
1214 			goto out;
1215 
1216 		/*
1217 		 * If the process is stopped and receives a KILL signal,
1218 		 * make the process runnable.
1219 		 */
1220 		if (sig == SIGKILL) {
1221 			proc_unstop(p);
1222 			goto active_process;
1223 		}
1224 
1225 		/*
1226 		 * If the process is stopped and receives a CONT signal,
1227 		 * then try to make the process runnable again.
1228 		 */
1229 		if (prop & SA_CONT) {
1230 			/*
1231 			 * If SIGCONT is default (or ignored), we continue the
1232 			 * process but don't leave the signal in p_siglist, as
1233 			 * it has no further action.  If SIGCONT is held, we
1234 			 * continue the process and leave the signal in
1235 			 * p_siglist.  If the process catches SIGCONT, let it
1236 			 * handle the signal itself.
1237 			 *
1238 			 * XXX what if the signal is being held blocked?
1239 			 *
1240 			 * Token required to interlock kern_wait().
1241 			 * Reparenting can also cause a race so we have to
1242 			 * hold (q).
1243 			 */
1244 			q = p->p_pptr;
1245 			PHOLD(q);
1246 			lwkt_gettoken(&q->p_token);
1247 			p->p_flags |= P_CONTINUED;
1248 			wakeup(q);
1249 			if (action == SIG_DFL)
1250 				SIGDELSET(p->p_siglist, sig);
1251 			proc_unstop(p);
1252 			lwkt_reltoken(&q->p_token);
1253 			PRELE(q);
1254 			if (action == SIG_CATCH)
1255 				goto active_process;
1256 			goto out;
1257 		}
1258 
1259 		/*
1260 		 * If the process is stopped and receives another STOP
1261 		 * signal, we do not need to stop it again.  If we did
1262 		 * the shell could get confused.
1263 		 *
1264 		 * However, if the current/preempted lwp is part of the
1265 		 * process receiving the signal, we need to keep it,
1266 		 * so that this lwp can stop in issignal() later, as
1267 		 * we don't want to wait until it reaches userret!
1268 		 */
1269 		if (prop & SA_STOP) {
1270 			if (lwkt_preempted_proc() == NULL ||
1271 			    lwkt_preempted_proc()->lwp_proc != p)
1272 				SIGDELSET(p->p_siglist, sig);
1273 		}
1274 
1275 		/*
1276 		 * Otherwise the process is stopped and it received some
1277 		 * signal, which does not change its stopped state.  When
1278 		 * the process is continued a wakeup(p) will be issued which
1279 		 * will wakeup any threads sleeping in tstop().
1280 		 */
1281 		if (lp == NULL) {
1282 			/* NOTE: returns lp w/ token held */
1283 			lp = find_lwp_for_signal(p, sig);
1284 		}
1285 		goto out;
1286 
1287 		/* NOTREACHED */
1288 	}
1289 	/* else not stopped */
1290 active_process:
1291 
1292 	/*
1293 	 * Never deliver a lwp-specific signal to a random lwp.
1294 	 */
1295 	if (lp == NULL) {
1296 		/* NOTE: returns lp w/ token held */
1297 		lp = find_lwp_for_signal(p, sig);
1298 		if (lp) {
1299 			if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
1300 				lwkt_reltoken(&lp->lwp_token);
1301 				LWPRELE(lp);
1302 				lp = NULL;
1303 			}
1304 		}
1305 	}
1306 
1307 	/*
1308 	 * Deliver to the process generically if (1) the signal is being
1309 	 * sent to any thread or (2) we could not find a thread to deliver
1310 	 * it to.
1311 	 */
1312 	if (lp == NULL) {
1313 		SIGADDSET(p->p_siglist, sig);
1314 		goto out;
1315 	}
1316 
1317 	/*
1318 	 * Deliver to a specific LWP whether it masks it or not.  It will
1319 	 * not be dispatched if masked but we must still deliver it.
1320 	 */
1321 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1322 	    (p->p_flags & P_TRACED) == 0) {
1323 		p->p_nice = NZERO;
1324 	}
1325 
1326 	/*
1327 	 * If the process receives a STOP signal which indeed needs to
1328 	 * stop the process, do so.  If the process chose to catch the
1329 	 * signal, it will be treated like any other signal.
1330 	 */
1331 	if ((prop & SA_STOP) && action == SIG_DFL) {
1332 		/*
1333 		 * If a child holding parent blocked, stopping
1334 		 * could cause deadlock.  Take no action at this
1335 		 * time.
1336 		 */
1337 		if (p->p_flags & P_PPWAIT) {
1338 			SIGADDSET(p->p_siglist, sig);
1339 			goto out;
1340 		}
1341 
1342 		/*
1343 		 * Do not actually try to manipulate the process, but simply
1344 		 * stop it.  Lwps will stop as soon as they safely can.
1345 		 *
1346 		 * Ignore stop if the process is exiting.
1347 		 */
1348 		if ((p->p_flags & P_WEXIT) == 0) {
1349 			p->p_xstat = sig;
1350 			proc_stop(p);
1351 		}
1352 		goto out;
1353 	}
1354 
1355 	/*
1356 	 * If it is a CONT signal with default action, just ignore it.
1357 	 */
1358 	if ((prop & SA_CONT) && action == SIG_DFL)
1359 		goto out;
1360 
1361 	/*
1362 	 * Mark signal pending at this specific thread.
1363 	 */
1364 	spin_lock(&lp->lwp_spin);
1365 	SIGADDSET(lp->lwp_siglist, sig);
1366 	spin_unlock(&lp->lwp_spin);
1367 
1368 	lwp_signotify(lp);
1369 
1370 out:
1371 	if (lp) {
1372 		lwkt_reltoken(&lp->lwp_token);
1373 		LWPRELE(lp);
1374 	}
1375 	lwkt_reltoken(&p->p_token);
1376 	PRELE(p);
1377 }
1378 
1379 /*
1380  * Notify the LWP that a signal has arrived.  The LWP does not have to be
1381  * sleeping on the current cpu.
1382  *
1383  * p->p_token and lp->lwp_token must be held on call.
1384  *
1385  * We can only safely schedule the thread on its current cpu and only if
1386  * one of the SINTR flags is set.  If an SINTR flag is set AND we are on
1387  * the correct cpu we are properly interlocked, otherwise we could be
1388  * racing other thread transition states (or the lwp is on the user scheduler
1389  * runq but not scheduled) and must not do anything.
1390  *
1391  * Since we hold the lwp token we know the lwp cannot be ripped out from
1392  * under us so we can safely hold it to prevent it from being ripped out
1393  * from under us if we are forced to IPI another cpu to make the local
1394  * checks there.
1395  *
1396  * Adjustment of lp->lwp_stat can only occur when we hold the lwp_token,
1397  * which we won't in an IPI so any fixups have to be done here, effectively
1398  * replicating part of what setrunnable() does.
1399  */
1400 static void
1401 lwp_signotify(struct lwp *lp)
1402 {
1403 	ASSERT_LWKT_TOKEN_HELD(&lp->lwp_proc->p_token);
1404 
1405 	crit_enter();
1406 	if (lp == lwkt_preempted_proc()) {
1407 		/*
1408 		 * lwp is on the current cpu AND it is currently running
1409 		 * (we preempted it).
1410 		 */
1411 		signotify();
1412 	} else if (lp->lwp_flags & LWP_SINTR) {
1413 		/*
1414 		 * lwp is sitting in tsleep() with PCATCH set
1415 		 */
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 if (lp->lwp_thread->td_flags & TDF_SINTR) {
1430 		/*
1431 		 * lwp is sitting in lwkt_sleep() with PCATCH set.
1432 		 */
1433 		if (lp->lwp_thread->td_gd == mycpu) {
1434 			setrunnable(lp);
1435 		} else {
1436 			/*
1437 			 * We can only adjust lwp_stat while we hold the
1438 			 * lwp_token, and we won't in the IPI function.
1439 			 */
1440 			LWPHOLD(lp);
1441 			if (lp->lwp_stat == LSSTOP)
1442 				lp->lwp_stat = LSSLEEP;
1443 			lwkt_send_ipiq(lp->lwp_thread->td_gd,
1444 				       lwp_signotify_remote, lp);
1445 		}
1446 	} else {
1447 		/*
1448 		 * Otherwise the lwp is either in some uninterruptable state
1449 		 * or it is on the userland scheduler's runqueue waiting to
1450 		 * be scheduled to a cpu.
1451 		 */
1452 	}
1453 	crit_exit();
1454 }
1455 
1456 /*
1457  * This function is called via an IPI so we cannot call setrunnable() here
1458  * (because while we hold the lp we don't own its token, and can't get it
1459  * from an IPI).
1460  *
1461  * We are interlocked by virtue of being on the same cpu as the target.  If
1462  * we still are and LWP_SINTR or TDF_SINTR is set we can safely schedule
1463  * the target thread.
1464  */
1465 static void
1466 lwp_signotify_remote(void *arg)
1467 {
1468 	struct lwp *lp = arg;
1469 	thread_t td = lp->lwp_thread;
1470 
1471 	if (lp == lwkt_preempted_proc()) {
1472 		signotify();
1473 		LWPRELE(lp);
1474 	} else if (td->td_gd == mycpu) {
1475 		if ((lp->lwp_flags & LWP_SINTR) ||
1476 		    (td->td_flags & TDF_SINTR)) {
1477 			lwkt_schedule(td);
1478 		}
1479 		LWPRELE(lp);
1480 	} else {
1481 		lwkt_send_ipiq(td->td_gd, lwp_signotify_remote, lp);
1482 		/* LWPHOLD() is forwarded to the target cpu */
1483 	}
1484 }
1485 
1486 /*
1487  * Caller must hold p->p_token
1488  */
1489 void
1490 proc_stop(struct proc *p)
1491 {
1492 	struct proc *q;
1493 	struct lwp *lp;
1494 
1495 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1496 
1497 	/* If somebody raced us, be happy with it */
1498 	if (p->p_stat == SSTOP || p->p_stat == SZOMB) {
1499 		return;
1500 	}
1501 	p->p_stat = SSTOP;
1502 
1503 	FOREACH_LWP_IN_PROC(lp, p) {
1504 		LWPHOLD(lp);
1505 		lwkt_gettoken(&lp->lwp_token);
1506 
1507 		switch (lp->lwp_stat) {
1508 		case LSSTOP:
1509 			/*
1510 			 * Do nothing, we are already counted in
1511 			 * p_nstopped.
1512 			 */
1513 			break;
1514 
1515 		case LSSLEEP:
1516 			/*
1517 			 * We're sleeping, but we will stop before
1518 			 * returning to userspace, so count us
1519 			 * as stopped as well.  We set LWP_MP_WSTOP
1520 			 * to signal the lwp that it should not
1521 			 * increase p_nstopped when reaching tstop().
1522 			 *
1523 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1524 			 */
1525 			if ((lp->lwp_mpflags & LWP_MP_WSTOP) == 0) {
1526 				atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
1527 				++p->p_nstopped;
1528 			}
1529 			break;
1530 
1531 		case LSRUN:
1532 			/*
1533 			 * We might notify ourself, but that's not
1534 			 * a problem.
1535 			 */
1536 			lwp_signotify(lp);
1537 			break;
1538 		}
1539 		lwkt_reltoken(&lp->lwp_token);
1540 		LWPRELE(lp);
1541 	}
1542 
1543 	if (p->p_nstopped == p->p_nthreads) {
1544 		/*
1545 		 * Token required to interlock kern_wait().  Reparenting can
1546 		 * also cause a race so we have to hold (q).
1547 		 */
1548 		q = p->p_pptr;
1549 		PHOLD(q);
1550 		lwkt_gettoken(&q->p_token);
1551 		p->p_flags &= ~P_WAITED;
1552 		wakeup(q);
1553 		if ((q->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1554 			ksignal(p->p_pptr, SIGCHLD);
1555 		lwkt_reltoken(&q->p_token);
1556 		PRELE(q);
1557 	}
1558 }
1559 
1560 /*
1561  * Caller must hold proc_token
1562  */
1563 void
1564 proc_unstop(struct proc *p)
1565 {
1566 	struct lwp *lp;
1567 
1568 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1569 
1570 	if (p->p_stat != SSTOP)
1571 		return;
1572 
1573 	p->p_stat = SACTIVE;
1574 
1575 	FOREACH_LWP_IN_PROC(lp, p) {
1576 		LWPHOLD(lp);
1577 		lwkt_gettoken(&lp->lwp_token);
1578 
1579 		switch (lp->lwp_stat) {
1580 		case LSRUN:
1581 			/*
1582 			 * Uh?  Not stopped?  Well, I guess that's okay.
1583 			 */
1584 			if (bootverbose)
1585 				kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1586 					p->p_pid, lp->lwp_tid);
1587 			break;
1588 
1589 		case LSSLEEP:
1590 			/*
1591 			 * Still sleeping.  Don't bother waking it up.
1592 			 * However, if this thread was counted as
1593 			 * stopped, undo this.
1594 			 *
1595 			 * Nevertheless we call setrunnable() so that it
1596 			 * will wake up in case a signal or timeout arrived
1597 			 * in the meantime.
1598 			 *
1599 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1600 			 */
1601 			if (lp->lwp_mpflags & LWP_MP_WSTOP) {
1602 				atomic_clear_int(&lp->lwp_mpflags,
1603 						 LWP_MP_WSTOP);
1604 				--p->p_nstopped;
1605 			} else {
1606 				if (bootverbose)
1607 					kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1608 						p->p_pid, lp->lwp_tid);
1609 			}
1610 			/* FALLTHROUGH */
1611 
1612 		case LSSTOP:
1613 			/*
1614 			 * This handles any lwp's waiting in a tsleep with
1615 			 * SIGCATCH.
1616 			 */
1617 			lwp_signotify(lp);
1618 			break;
1619 
1620 		}
1621 		lwkt_reltoken(&lp->lwp_token);
1622 		LWPRELE(lp);
1623 	}
1624 
1625 	/*
1626 	 * This handles any lwp's waiting in tstop().  We have interlocked
1627 	 * the setting of p_stat by acquiring and releasing each lpw's
1628 	 * token.
1629 	 */
1630 	wakeup(p);
1631 }
1632 
1633 /*
1634  * No requirements.
1635  */
1636 static int
1637 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1638 {
1639 	sigset_t savedmask, set;
1640 	struct proc *p = curproc;
1641 	struct lwp *lp = curthread->td_lwp;
1642 	int error, sig, hz, timevalid = 0;
1643 	struct timespec rts, ets, ts;
1644 	struct timeval tv;
1645 
1646 	error = 0;
1647 	sig = 0;
1648 	ets.tv_sec = 0;		/* silence compiler warning */
1649 	ets.tv_nsec = 0;	/* silence compiler warning */
1650 	SIG_CANTMASK(waitset);
1651 	savedmask = lp->lwp_sigmask;
1652 
1653 	if (timeout) {
1654 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1655 		    timeout->tv_nsec < 1000000000) {
1656 			timevalid = 1;
1657 			getnanouptime(&rts);
1658 		 	ets = rts;
1659 			timespecadd(&ets, timeout);
1660 		}
1661 	}
1662 
1663 	for (;;) {
1664 		set = lwp_sigpend(lp);
1665 		SIGSETAND(set, waitset);
1666 		if ((sig = sig_ffs(&set)) != 0) {
1667 			SIGFILLSET(lp->lwp_sigmask);
1668 			SIGDELSET(lp->lwp_sigmask, sig);
1669 			SIG_CANTMASK(lp->lwp_sigmask);
1670 			sig = issignal(lp, 1);
1671 			/*
1672 			 * It may be a STOP signal, in the case, issignal
1673 			 * returns 0, because we may stop there, and new
1674 			 * signal can come in, we should restart if we got
1675 			 * nothing.
1676 			 */
1677 			if (sig == 0)
1678 				continue;
1679 			else
1680 				break;
1681 		}
1682 
1683 		/*
1684 		 * Previous checking got nothing, and we retried but still
1685 		 * got nothing, we should return the error status.
1686 		 */
1687 		if (error)
1688 			break;
1689 
1690 		/*
1691 		 * POSIX says this must be checked after looking for pending
1692 		 * signals.
1693 		 */
1694 		if (timeout) {
1695 			if (timevalid == 0) {
1696 				error = EINVAL;
1697 				break;
1698 			}
1699 			getnanouptime(&rts);
1700 			if (timespeccmp(&rts, &ets, >=)) {
1701 				error = EAGAIN;
1702 				break;
1703 			}
1704 			ts = ets;
1705 			timespecsub(&ts, &rts);
1706 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1707 			hz = tvtohz_high(&tv);
1708 		} else {
1709 			hz = 0;
1710 		}
1711 
1712 		lp->lwp_sigmask = savedmask;
1713 		SIGSETNAND(lp->lwp_sigmask, waitset);
1714 		/*
1715 		 * We won't ever be woken up.  Instead, our sleep will
1716 		 * be broken in lwpsignal().
1717 		 */
1718 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1719 		if (timeout) {
1720 			if (error == ERESTART) {
1721 				/* can not restart a timeout wait. */
1722 				error = EINTR;
1723 			} else if (error == EAGAIN) {
1724 				/* will calculate timeout by ourself. */
1725 				error = 0;
1726 			}
1727 		}
1728 		/* Retry ... */
1729 	}
1730 
1731 	lp->lwp_sigmask = savedmask;
1732 	if (sig) {
1733 		error = 0;
1734 		bzero(info, sizeof(*info));
1735 		info->si_signo = sig;
1736 		spin_lock(&lp->lwp_spin);
1737 		lwp_delsig(lp, sig);	/* take the signal! */
1738 		spin_unlock(&lp->lwp_spin);
1739 
1740 		if (sig == SIGKILL) {
1741 			sigexit(lp, sig);
1742 			/* NOT REACHED */
1743 		}
1744 	}
1745 
1746 	return (error);
1747 }
1748 
1749 /*
1750  * MPALMOSTSAFE
1751  */
1752 int
1753 sys_sigtimedwait(struct sigtimedwait_args *uap)
1754 {
1755 	struct timespec ts;
1756 	struct timespec *timeout;
1757 	sigset_t set;
1758 	siginfo_t info;
1759 	int error;
1760 
1761 	if (uap->timeout) {
1762 		error = copyin(uap->timeout, &ts, sizeof(ts));
1763 		if (error)
1764 			return (error);
1765 		timeout = &ts;
1766 	} else {
1767 		timeout = NULL;
1768 	}
1769 	error = copyin(uap->set, &set, sizeof(set));
1770 	if (error)
1771 		return (error);
1772 	error = kern_sigtimedwait(set, &info, timeout);
1773 	if (error)
1774 		return (error);
1775 	if (uap->info)
1776 		error = copyout(&info, uap->info, sizeof(info));
1777 	/* Repost if we got an error. */
1778 	/*
1779 	 * XXX lwp
1780 	 *
1781 	 * This could transform a thread-specific signal to another
1782 	 * thread / process pending signal.
1783 	 */
1784 	if (error) {
1785 		ksignal(curproc, info.si_signo);
1786 	} else {
1787 		uap->sysmsg_result = info.si_signo;
1788 	}
1789 	return (error);
1790 }
1791 
1792 /*
1793  * MPALMOSTSAFE
1794  */
1795 int
1796 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1797 {
1798 	siginfo_t info;
1799 	sigset_t set;
1800 	int error;
1801 
1802 	error = copyin(uap->set, &set, sizeof(set));
1803 	if (error)
1804 		return (error);
1805 	error = kern_sigtimedwait(set, &info, NULL);
1806 	if (error)
1807 		return (error);
1808 	if (uap->info)
1809 		error = copyout(&info, uap->info, sizeof(info));
1810 	/* Repost if we got an error. */
1811 	/*
1812 	 * XXX lwp
1813 	 *
1814 	 * This could transform a thread-specific signal to another
1815 	 * thread / process pending signal.
1816 	 */
1817 	if (error) {
1818 		ksignal(curproc, info.si_signo);
1819 	} else {
1820 		uap->sysmsg_result = info.si_signo;
1821 	}
1822 	return (error);
1823 }
1824 
1825 /*
1826  * If the current process has received a signal that would interrupt a
1827  * system call, return EINTR or ERESTART as appropriate.
1828  */
1829 int
1830 iscaught(struct lwp *lp)
1831 {
1832 	struct proc *p = lp->lwp_proc;
1833 	int sig;
1834 
1835 	if (p) {
1836 		if ((sig = CURSIG(lp)) != 0) {
1837 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1838 				return (EINTR);
1839 			return (ERESTART);
1840 		}
1841 	}
1842 	return(EWOULDBLOCK);
1843 }
1844 
1845 /*
1846  * If the current process has received a signal (should be caught or cause
1847  * termination, should interrupt current syscall), return the signal number.
1848  * Stop signals with default action are processed immediately, then cleared;
1849  * they aren't returned.  This is checked after each entry to the system for
1850  * a syscall or trap (though this can usually be done without calling issignal
1851  * by checking the pending signal masks in the CURSIG macro).
1852  *
1853  * This routine is called via CURSIG/__cursig.  We will acquire and release
1854  * p->p_token but if the caller needs to interlock the test the caller must
1855  * also hold p->p_token.
1856  *
1857  *	while (sig = CURSIG(curproc))
1858  *		postsig(sig);
1859  *
1860  * MPSAFE
1861  */
1862 int
1863 issignal(struct lwp *lp, int maytrace)
1864 {
1865 	struct proc *p = lp->lwp_proc;
1866 	sigset_t mask;
1867 	int sig, prop;
1868 
1869 	lwkt_gettoken(&p->p_token);
1870 
1871 	for (;;) {
1872 		int traced = (p->p_flags & P_TRACED) || (p->p_stops & S_SIG);
1873 
1874 		/*
1875 		 * If this process is supposed to stop, stop this thread.
1876 		 */
1877 		if (p->p_stat == SSTOP)
1878 			tstop();
1879 
1880 		mask = lwp_sigpend(lp);
1881 		SIGSETNAND(mask, lp->lwp_sigmask);
1882 		if (p->p_flags & P_PPWAIT)
1883 			SIG_STOPSIGMASK(mask);
1884 		if (SIGISEMPTY(mask)) {		/* no signal to send */
1885 			lwkt_reltoken(&p->p_token);
1886 			return (0);
1887 		}
1888 		sig = sig_ffs(&mask);
1889 
1890 		STOPEVENT(p, S_SIG, sig);
1891 
1892 		/*
1893 		 * We should see pending but ignored signals
1894 		 * only if P_TRACED was on when they were posted.
1895 		 */
1896 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1897 			spin_lock(&lp->lwp_spin);
1898 			lwp_delsig(lp, sig);
1899 			spin_unlock(&lp->lwp_spin);
1900 			continue;
1901 		}
1902 		if (maytrace &&
1903 		    (p->p_flags & P_TRACED) &&
1904 		    (p->p_flags & P_PPWAIT) == 0) {
1905 			/*
1906 			 * If traced, always stop, and stay stopped until
1907 			 * released by the parent.
1908 			 *
1909 			 * NOTE: SSTOP may get cleared during the loop,
1910 			 * but we do not re-notify the parent if we have
1911 			 * to loop several times waiting for the parent
1912 			 * to let us continue.
1913 			 *
1914 			 * XXX not sure if this is still true
1915 			 */
1916 			p->p_xstat = sig;
1917 			proc_stop(p);
1918 			do {
1919 				tstop();
1920 			} while (!trace_req(p) && (p->p_flags & P_TRACED));
1921 
1922 			/*
1923 			 * If parent wants us to take the signal,
1924 			 * then it will leave it in p->p_xstat;
1925 			 * otherwise we just look for signals again.
1926 			 */
1927 			spin_lock(&lp->lwp_spin);
1928 			lwp_delsig(lp, sig);	/* clear old signal */
1929 			spin_unlock(&lp->lwp_spin);
1930 			sig = p->p_xstat;
1931 			if (sig == 0)
1932 				continue;
1933 
1934 			/*
1935 			 * Put the new signal into p_siglist.  If the
1936 			 * signal is being masked, look for other signals.
1937 			 *
1938 			 * XXX lwp might need a call to ksignal()
1939 			 */
1940 			SIGADDSET(p->p_siglist, sig);
1941 			if (SIGISMEMBER(lp->lwp_sigmask, sig))
1942 				continue;
1943 
1944 			/*
1945 			 * If the traced bit got turned off, go back up
1946 			 * to the top to rescan signals.  This ensures
1947 			 * that p_sig* and ps_sigact are consistent.
1948 			 */
1949 			if ((p->p_flags & P_TRACED) == 0)
1950 				continue;
1951 		}
1952 
1953 		prop = sigprop(sig);
1954 
1955 		/*
1956 		 * Decide whether the signal should be returned.
1957 		 * Return the signal's number, or fall through
1958 		 * to clear it from the pending mask.
1959 		 */
1960 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
1961 		case (intptr_t)SIG_DFL:
1962 			/*
1963 			 * Don't take default actions on system processes.
1964 			 */
1965 			if (p->p_pid <= 1) {
1966 #ifdef DIAGNOSTIC
1967 				/*
1968 				 * Are you sure you want to ignore SIGSEGV
1969 				 * in init? XXX
1970 				 */
1971 				kprintf("Process (pid %lu) got signal %d\n",
1972 					(u_long)p->p_pid, sig);
1973 #endif
1974 				break;		/* == ignore */
1975 			}
1976 
1977 			/*
1978 			 * Handle the in-kernel checkpoint action
1979 			 */
1980 			if (prop & SA_CKPT) {
1981 				checkpoint_signal_handler(lp);
1982 				break;
1983 			}
1984 
1985 			/*
1986 			 * If there is a pending stop signal to process
1987 			 * with default action, stop here,
1988 			 * then clear the signal.  However,
1989 			 * if process is member of an orphaned
1990 			 * process group, ignore tty stop signals.
1991 			 */
1992 			if (prop & SA_STOP) {
1993 				if (p->p_flags & P_TRACED ||
1994 		    		    (p->p_pgrp->pg_jobc == 0 &&
1995 				    prop & SA_TTYSTOP))
1996 					break;	/* == ignore */
1997 				if ((p->p_flags & P_WEXIT) == 0) {
1998 					p->p_xstat = sig;
1999 					proc_stop(p);
2000 					tstop();
2001 				}
2002 				break;
2003 			} else if (prop & SA_IGNORE) {
2004 				/*
2005 				 * Except for SIGCONT, shouldn't get here.
2006 				 * Default action is to ignore; drop it.
2007 				 */
2008 				break;		/* == ignore */
2009 			} else {
2010 				lwkt_reltoken(&p->p_token);
2011 				return (sig);
2012 			}
2013 
2014 			/*NOTREACHED*/
2015 
2016 		case (intptr_t)SIG_IGN:
2017 			/*
2018 			 * Masking above should prevent us ever trying
2019 			 * to take action on an ignored signal other
2020 			 * than SIGCONT, unless process is traced.
2021 			 */
2022 			if ((prop & SA_CONT) == 0 &&
2023 			    (p->p_flags & P_TRACED) == 0)
2024 				kprintf("issignal\n");
2025 			break;		/* == ignore */
2026 
2027 		default:
2028 			/*
2029 			 * This signal has an action, let
2030 			 * postsig() process it.
2031 			 */
2032 			lwkt_reltoken(&p->p_token);
2033 			return (sig);
2034 		}
2035 		spin_lock(&lp->lwp_spin);
2036 		lwp_delsig(lp, sig);		/* take the signal! */
2037 		spin_unlock(&lp->lwp_spin);
2038 	}
2039 	/* NOTREACHED */
2040 }
2041 
2042 /*
2043  * Take the action for the specified signal
2044  * from the current set of pending signals.
2045  *
2046  * Caller must hold p->p_token
2047  */
2048 void
2049 postsig(int sig)
2050 {
2051 	struct lwp *lp = curthread->td_lwp;
2052 	struct proc *p = lp->lwp_proc;
2053 	struct sigacts *ps = p->p_sigacts;
2054 	sig_t action;
2055 	sigset_t returnmask;
2056 	int code;
2057 
2058 	KASSERT(sig != 0, ("postsig"));
2059 
2060 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
2061 
2062 	/*
2063 	 * If we are a virtual kernel running an emulated user process
2064 	 * context, switch back to the virtual kernel context before
2065 	 * trying to post the signal.
2066 	 */
2067 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
2068 		struct trapframe *tf = lp->lwp_md.md_regs;
2069 		tf->tf_trapno = 0;
2070 		vkernel_trap(lp, tf);
2071 	}
2072 
2073 	spin_lock(&lp->lwp_spin);
2074 	lwp_delsig(lp, sig);
2075 	spin_unlock(&lp->lwp_spin);
2076 	action = ps->ps_sigact[_SIG_IDX(sig)];
2077 #ifdef KTRACE
2078 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
2079 		ktrpsig(lp, sig, action, lp->lwp_flags & LWP_OLDMASK ?
2080 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
2081 #endif
2082 	STOPEVENT(p, S_SIG, sig);
2083 
2084 	if (action == SIG_DFL) {
2085 		/*
2086 		 * Default action, where the default is to kill
2087 		 * the process.  (Other cases were ignored above.)
2088 		 */
2089 		sigexit(lp, sig);
2090 		/* NOTREACHED */
2091 	} else {
2092 		/*
2093 		 * If we get here, the signal must be caught.
2094 		 */
2095 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
2096 		    ("postsig action"));
2097 
2098 		/*
2099 		 * Reset the signal handler if asked to
2100 		 */
2101 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
2102 			/*
2103 			 * See kern_sigaction() for origin of this code.
2104 			 */
2105 			SIGDELSET(p->p_sigcatch, sig);
2106 			if (sig != SIGCONT &&
2107 			    sigprop(sig) & SA_IGNORE)
2108 				SIGADDSET(p->p_sigignore, sig);
2109 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2110 		}
2111 
2112 		/*
2113 		 * Set the signal mask and calculate the mask to restore
2114 		 * when the signal function returns.
2115 		 *
2116 		 * Special case: user has done a sigsuspend.  Here the
2117 		 * current mask is not of interest, but rather the
2118 		 * mask from before the sigsuspend is what we want
2119 		 * restored after the signal processing is completed.
2120 		 */
2121 		if (lp->lwp_flags & LWP_OLDMASK) {
2122 			returnmask = lp->lwp_oldsigmask;
2123 			lp->lwp_flags &= ~LWP_OLDMASK;
2124 		} else {
2125 			returnmask = lp->lwp_sigmask;
2126 		}
2127 
2128 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
2129 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
2130 			SIGADDSET(lp->lwp_sigmask, sig);
2131 
2132 		lp->lwp_ru.ru_nsignals++;
2133 		if (lp->lwp_sig != sig) {
2134 			code = 0;
2135 		} else {
2136 			code = lp->lwp_code;
2137 			lp->lwp_code = 0;
2138 			lp->lwp_sig = 0;
2139 		}
2140 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
2141 	}
2142 }
2143 
2144 /*
2145  * Kill the current process for stated reason.
2146  */
2147 void
2148 killproc(struct proc *p, char *why)
2149 {
2150 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
2151 		p->p_pid, p->p_comm,
2152 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2153 	ksignal(p, SIGKILL);
2154 }
2155 
2156 /*
2157  * Force the current process to exit with the specified signal, dumping core
2158  * if appropriate.  We bypass the normal tests for masked and caught signals,
2159  * allowing unrecoverable failures to terminate the process without changing
2160  * signal state.  Mark the accounting record with the signal termination.
2161  * If dumping core, save the signal number for the debugger.  Calls exit and
2162  * does not return.
2163  *
2164  * This routine does not return.
2165  */
2166 void
2167 sigexit(struct lwp *lp, int sig)
2168 {
2169 	struct proc *p = lp->lwp_proc;
2170 
2171 	lwkt_gettoken(&p->p_token);
2172 	p->p_acflag |= AXSIG;
2173 	if (sigprop(sig) & SA_CORE) {
2174 		lp->lwp_sig = sig;
2175 		/*
2176 		 * Log signals which would cause core dumps
2177 		 * (Log as LOG_INFO to appease those who don't want
2178 		 * these messages.)
2179 		 * XXX : Todo, as well as euid, write out ruid too
2180 		 */
2181 		if (coredump(lp, sig) == 0)
2182 			sig |= WCOREFLAG;
2183 		if (kern_logsigexit)
2184 			log(LOG_INFO,
2185 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2186 			    p->p_pid, p->p_comm,
2187 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
2188 			    sig &~ WCOREFLAG,
2189 			    sig & WCOREFLAG ? " (core dumped)" : "");
2190 	}
2191 	lwkt_reltoken(&p->p_token);
2192 	exit1(W_EXITCODE(0, sig));
2193 	/* NOTREACHED */
2194 }
2195 
2196 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
2197 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2198 	      sizeof(corefilename), "process corefile name format string");
2199 
2200 /*
2201  * expand_name(name, uid, pid)
2202  * Expand the name described in corefilename, using name, uid, and pid.
2203  * corefilename is a kprintf-like string, with three format specifiers:
2204  *	%N	name of process ("name")
2205  *	%P	process id (pid)
2206  *	%U	user id (uid)
2207  * For example, "%N.core" is the default; they can be disabled completely
2208  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2209  * This is controlled by the sysctl variable kern.corefile (see above).
2210  */
2211 
2212 static char *
2213 expand_name(const char *name, uid_t uid, pid_t pid)
2214 {
2215 	char *temp;
2216 	char buf[11];		/* Buffer for pid/uid -- max 4B */
2217 	int i, n;
2218 	char *format = corefilename;
2219 	size_t namelen;
2220 
2221 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
2222 	if (temp == NULL)
2223 		return NULL;
2224 	namelen = strlen(name);
2225 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2226 		int l;
2227 		switch (format[i]) {
2228 		case '%':	/* Format character */
2229 			i++;
2230 			switch (format[i]) {
2231 			case '%':
2232 				temp[n++] = '%';
2233 				break;
2234 			case 'N':	/* process name */
2235 				if ((n + namelen) > MAXPATHLEN) {
2236 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2237 					    pid, name, uid, temp, name);
2238 					kfree(temp, M_TEMP);
2239 					return NULL;
2240 				}
2241 				memcpy(temp+n, name, namelen);
2242 				n += namelen;
2243 				break;
2244 			case 'P':	/* process id */
2245 				l = ksprintf(buf, "%u", pid);
2246 				if ((n + l) > MAXPATHLEN) {
2247 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2248 					    pid, name, uid, temp, name);
2249 					kfree(temp, M_TEMP);
2250 					return NULL;
2251 				}
2252 				memcpy(temp+n, buf, l);
2253 				n += l;
2254 				break;
2255 			case 'U':	/* user id */
2256 				l = ksprintf(buf, "%u", uid);
2257 				if ((n + l) > MAXPATHLEN) {
2258 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2259 					    pid, name, uid, temp, name);
2260 					kfree(temp, M_TEMP);
2261 					return NULL;
2262 				}
2263 				memcpy(temp+n, buf, l);
2264 				n += l;
2265 				break;
2266 			default:
2267 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2268 			}
2269 			break;
2270 		default:
2271 			temp[n++] = format[i];
2272 		}
2273 	}
2274 	temp[n] = '\0';
2275 	return temp;
2276 }
2277 
2278 /*
2279  * Dump a process' core.  The main routine does some
2280  * policy checking, and creates the name of the coredump;
2281  * then it passes on a vnode and a size limit to the process-specific
2282  * coredump routine if there is one; if there _is not_ one, it returns
2283  * ENOSYS; otherwise it returns the error from the process-specific routine.
2284  *
2285  * The parameter `lp' is the lwp which triggered the coredump.
2286  */
2287 
2288 static int
2289 coredump(struct lwp *lp, int sig)
2290 {
2291 	struct proc *p = lp->lwp_proc;
2292 	struct vnode *vp;
2293 	struct ucred *cred = p->p_ucred;
2294 	struct flock lf;
2295 	struct nlookupdata nd;
2296 	struct vattr vattr;
2297 	int error, error1;
2298 	char *name;			/* name of corefile */
2299 	off_t limit;
2300 
2301 	STOPEVENT(p, S_CORE, 0);
2302 
2303 	if (((sugid_coredump == 0) && p->p_flags & P_SUGID) || do_coredump == 0)
2304 		return (EFAULT);
2305 
2306 	/*
2307 	 * Note that the bulk of limit checking is done after
2308 	 * the corefile is created.  The exception is if the limit
2309 	 * for corefiles is 0, in which case we don't bother
2310 	 * creating the corefile at all.  This layout means that
2311 	 * a corefile is truncated instead of not being created,
2312 	 * if it is larger than the limit.
2313 	 */
2314 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2315 	if (limit == 0)
2316 		return EFBIG;
2317 
2318 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2319 	if (name == NULL)
2320 		return (EINVAL);
2321 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2322 	if (error == 0)
2323 		error = vn_open(&nd, NULL, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
2324 	kfree(name, M_TEMP);
2325 	if (error) {
2326 		nlookup_done(&nd);
2327 		return (error);
2328 	}
2329 	vp = nd.nl_open_vp;
2330 	nd.nl_open_vp = NULL;
2331 	nlookup_done(&nd);
2332 
2333 	vn_unlock(vp);
2334 	lf.l_whence = SEEK_SET;
2335 	lf.l_start = 0;
2336 	lf.l_len = 0;
2337 	lf.l_type = F_WRLCK;
2338 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2339 	if (error)
2340 		goto out2;
2341 
2342 	/* Don't dump to non-regular files or files with links. */
2343 	if (vp->v_type != VREG ||
2344 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2345 		error = EFAULT;
2346 		goto out1;
2347 	}
2348 
2349 	/* Don't dump to files current user does not own */
2350 	if (vattr.va_uid != p->p_ucred->cr_uid) {
2351 		error = EFAULT;
2352 		goto out1;
2353 	}
2354 
2355 	VATTR_NULL(&vattr);
2356 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2357 	vattr.va_size = 0;
2358 	VOP_SETATTR(vp, &vattr, cred);
2359 	p->p_acflag |= ACORE;
2360 	vn_unlock(vp);
2361 
2362 	error = p->p_sysent->sv_coredump ?
2363 		  p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2364 
2365 out1:
2366 	lf.l_type = F_UNLCK;
2367 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2368 out2:
2369 	error1 = vn_close(vp, FWRITE);
2370 	if (error == 0)
2371 		error = error1;
2372 	return (error);
2373 }
2374 
2375 /*
2376  * Nonexistent system call-- signal process (may want to handle it).
2377  * Flag error in case process won't see signal immediately (blocked or ignored).
2378  *
2379  * MPALMOSTSAFE
2380  */
2381 /* ARGSUSED */
2382 int
2383 sys_nosys(struct nosys_args *args)
2384 {
2385 	lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2386 	return (EINVAL);
2387 }
2388 
2389 /*
2390  * Send a SIGIO or SIGURG signal to a process or process group using
2391  * stored credentials rather than those of the current process.
2392  */
2393 void
2394 pgsigio(struct sigio *sigio, int sig, int checkctty)
2395 {
2396 	if (sigio == NULL)
2397 		return;
2398 
2399 	if (sigio->sio_pgid > 0) {
2400 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2401 		             sigio->sio_proc))
2402 			ksignal(sigio->sio_proc, sig);
2403 	} else if (sigio->sio_pgid < 0) {
2404 		struct proc *p;
2405 		struct pgrp *pg = sigio->sio_pgrp;
2406 
2407 		/*
2408 		 * Must interlock all signals against fork
2409 		 */
2410 		pgref(pg);
2411 		lockmgr(&pg->pg_lock, LK_EXCLUSIVE);
2412 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
2413 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2414 			    (checkctty == 0 || (p->p_flags & P_CONTROLT)))
2415 				ksignal(p, sig);
2416 		}
2417 		lockmgr(&pg->pg_lock, LK_RELEASE);
2418 		pgrel(pg);
2419 	}
2420 }
2421 
2422 static int
2423 filt_sigattach(struct knote *kn)
2424 {
2425 	struct proc *p = curproc;
2426 
2427 	kn->kn_ptr.p_proc = p;
2428 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
2429 
2430 	/* XXX lock the proc here while adding to the list? */
2431 	knote_insert(&p->p_klist, kn);
2432 
2433 	return (0);
2434 }
2435 
2436 static void
2437 filt_sigdetach(struct knote *kn)
2438 {
2439 	struct proc *p = kn->kn_ptr.p_proc;
2440 
2441 	knote_remove(&p->p_klist, kn);
2442 }
2443 
2444 /*
2445  * signal knotes are shared with proc knotes, so we apply a mask to
2446  * the hint in order to differentiate them from process hints.  This
2447  * could be avoided by using a signal-specific knote list, but probably
2448  * isn't worth the trouble.
2449  */
2450 static int
2451 filt_signal(struct knote *kn, long hint)
2452 {
2453 	if (hint & NOTE_SIGNAL) {
2454 		hint &= ~NOTE_SIGNAL;
2455 
2456 		if (kn->kn_id == hint)
2457 			kn->kn_data++;
2458 	}
2459 	return (kn->kn_data != 0);
2460 }
2461