xref: /dragonfly/sys/kern/kern_sig.c (revision 9d626b29)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
35  * $FreeBSD: src/sys/kern/kern_sig.c,v 1.72.2.17 2003/05/16 16:34:34 obrien Exp $
36  */
37 
38 #include "opt_ktrace.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/sysproto.h>
44 #include <sys/signalvar.h>
45 #include <sys/resourcevar.h>
46 #include <sys/vnode.h>
47 #include <sys/event.h>
48 #include <sys/proc.h>
49 #include <sys/nlookup.h>
50 #include <sys/pioctl.h>
51 #include <sys/acct.h>
52 #include <sys/fcntl.h>
53 #include <sys/lock.h>
54 #include <sys/wait.h>
55 #include <sys/ktrace.h>
56 #include <sys/syslog.h>
57 #include <sys/stat.h>
58 #include <sys/sysent.h>
59 #include <sys/sysctl.h>
60 #include <sys/malloc.h>
61 #include <sys/interrupt.h>
62 #include <sys/unistd.h>
63 #include <sys/kern_syscall.h>
64 #include <sys/vkernel.h>
65 
66 #include <sys/signal2.h>
67 #include <sys/thread2.h>
68 #include <sys/spinlock2.h>
69 
70 #include <machine/cpu.h>
71 #include <machine/smp.h>
72 
73 static int	coredump(struct lwp *, int);
74 static char	*expand_name(const char *, uid_t, pid_t);
75 static int	dokillpg(int sig, int pgid, int all);
76 static int	sig_ffs(sigset_t *set);
77 static int	sigprop(int sig);
78 static void	lwp_signotify(struct lwp *lp);
79 static void	lwp_signotify_remote(void *arg);
80 static int	kern_sigtimedwait(sigset_t set, siginfo_t *info,
81 		    struct timespec *timeout);
82 static void	proc_stopwait(struct proc *p);
83 
84 static int	filt_sigattach(struct knote *kn);
85 static void	filt_sigdetach(struct knote *kn);
86 static int	filt_signal(struct knote *kn, long hint);
87 
88 struct filterops sig_filtops =
89 	{ FILTEROP_MPSAFE, filt_sigattach, filt_sigdetach, filt_signal };
90 
91 static int	kern_logsigexit = 1;
92 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
93     &kern_logsigexit, 0,
94     "Log processes quitting on abnormal signals to syslog(3)");
95 
96 /*
97  * Can process p send the signal sig to process q?  Only processes within
98  * the current reaper or children of the current reaper can be signaled.
99  * Normally the reaper itself cannot be signalled, unless initok is set.
100  */
101 #define CANSIGNAL(q, sig, initok)				\
102 	((!p_trespass(curproc->p_ucred, (q)->p_ucred) &&	\
103 	reaper_sigtest(curproc, p, initok)) ||			\
104 	((sig) == SIGCONT && (q)->p_session == curproc->p_session))
105 
106 /*
107  * Policy -- Can real uid ruid with ucred uc send a signal to process q?
108  */
109 #define CANSIGIO(ruid, uc, q) \
110 	((uc)->cr_uid == 0 || \
111 	    (ruid) == (q)->p_ucred->cr_ruid || \
112 	    (uc)->cr_uid == (q)->p_ucred->cr_ruid || \
113 	    (ruid) == (q)->p_ucred->cr_uid || \
114 	    (uc)->cr_uid == (q)->p_ucred->cr_uid)
115 
116 int sugid_coredump;
117 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
118 	&sugid_coredump, 0, "Enable coredumping set user/group ID processes");
119 
120 static int	do_coredump = 1;
121 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
122 	&do_coredump, 0, "Enable/Disable coredumps");
123 
124 /*
125  * Signal properties and actions.
126  * The array below categorizes the signals and their default actions
127  * according to the following properties:
128  */
129 #define	SA_KILL		0x01		/* terminates process by default */
130 #define	SA_CORE		0x02		/* ditto and coredumps */
131 #define	SA_STOP		0x04		/* suspend process */
132 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
133 #define	SA_IGNORE	0x10		/* ignore by default */
134 #define	SA_CONT		0x20		/* continue if suspended */
135 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
136 #define SA_CKPT         0x80            /* checkpoint process */
137 
138 
139 static int sigproptbl[NSIG] = {
140         SA_KILL,                /* SIGHUP */
141         SA_KILL,                /* SIGINT */
142         SA_KILL|SA_CORE,        /* SIGQUIT */
143         SA_KILL|SA_CORE,        /* SIGILL */
144         SA_KILL|SA_CORE,        /* SIGTRAP */
145         SA_KILL|SA_CORE,        /* SIGABRT */
146         SA_KILL|SA_CORE,        /* SIGEMT */
147         SA_KILL|SA_CORE,        /* SIGFPE */
148         SA_KILL,                /* SIGKILL */
149         SA_KILL|SA_CORE,        /* SIGBUS */
150         SA_KILL|SA_CORE,        /* SIGSEGV */
151         SA_KILL|SA_CORE,        /* SIGSYS */
152         SA_KILL,                /* SIGPIPE */
153         SA_KILL,                /* SIGALRM */
154         SA_KILL,                /* SIGTERM */
155         SA_IGNORE,              /* SIGURG */
156         SA_STOP,                /* SIGSTOP */
157         SA_STOP|SA_TTYSTOP,     /* SIGTSTP */
158         SA_IGNORE|SA_CONT,      /* SIGCONT */
159         SA_IGNORE,              /* SIGCHLD */
160         SA_STOP|SA_TTYSTOP,     /* SIGTTIN */
161         SA_STOP|SA_TTYSTOP,     /* SIGTTOU */
162         SA_IGNORE,              /* SIGIO */
163         SA_KILL,                /* SIGXCPU */
164         SA_KILL,                /* SIGXFSZ */
165         SA_KILL,                /* SIGVTALRM */
166         SA_KILL,                /* SIGPROF */
167         SA_IGNORE,              /* SIGWINCH  */
168         SA_IGNORE,              /* SIGINFO */
169         SA_KILL,                /* SIGUSR1 */
170         SA_KILL,                /* SIGUSR2 */
171 	SA_IGNORE,              /* SIGTHR */
172 	SA_CKPT,                /* SIGCKPT */
173 	SA_KILL|SA_CKPT,        /* SIGCKPTEXIT */
174 	SA_IGNORE,
175 	SA_IGNORE,
176 	SA_IGNORE,
177 	SA_IGNORE,
178 	SA_IGNORE,
179 	SA_IGNORE,
180 	SA_IGNORE,
181 	SA_IGNORE,
182 	SA_IGNORE,
183 	SA_IGNORE,
184 	SA_IGNORE,
185 	SA_IGNORE,
186 	SA_IGNORE,
187 	SA_IGNORE,
188 	SA_IGNORE,
189 	SA_IGNORE,
190 	SA_IGNORE,
191 	SA_IGNORE,
192 	SA_IGNORE,
193 	SA_IGNORE,
194 	SA_IGNORE,
195 	SA_IGNORE,
196 	SA_IGNORE,
197 	SA_IGNORE,
198 	SA_IGNORE,
199 	SA_IGNORE,
200 	SA_IGNORE,
201 	SA_IGNORE,
202 	SA_IGNORE,
203 	SA_IGNORE,
204 
205 };
206 
207 static __inline int
208 sigprop(int sig)
209 {
210 
211 	if (sig > 0 && sig < NSIG)
212 		return (sigproptbl[_SIG_IDX(sig)]);
213 	return (0);
214 }
215 
216 static __inline int
217 sig_ffs(sigset_t *set)
218 {
219 	int i;
220 
221 	for (i = 0; i < _SIG_WORDS; i++)
222 		if (set->__bits[i])
223 			return (ffs(set->__bits[i]) + (i * 32));
224 	return (0);
225 }
226 
227 /*
228  * No requirements.
229  */
230 int
231 kern_sigaction(int sig, struct sigaction *act, struct sigaction *oact)
232 {
233 	struct thread *td = curthread;
234 	struct proc *p = td->td_proc;
235 	struct lwp *lp;
236 	struct sigacts *ps = p->p_sigacts;
237 
238 	if (sig <= 0 || sig > _SIG_MAXSIG)
239 		return (EINVAL);
240 
241 	lwkt_gettoken(&p->p_token);
242 
243 	if (oact) {
244 		oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
245 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
246 		oact->sa_flags = 0;
247 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
248 			oact->sa_flags |= SA_ONSTACK;
249 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
250 			oact->sa_flags |= SA_RESTART;
251 		if (SIGISMEMBER(ps->ps_sigreset, sig))
252 			oact->sa_flags |= SA_RESETHAND;
253 		if (SIGISMEMBER(ps->ps_signodefer, sig))
254 			oact->sa_flags |= SA_NODEFER;
255 		if (SIGISMEMBER(ps->ps_siginfo, sig))
256 			oact->sa_flags |= SA_SIGINFO;
257 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDSTOP)
258 			oact->sa_flags |= SA_NOCLDSTOP;
259 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDWAIT)
260 			oact->sa_flags |= SA_NOCLDWAIT;
261 	}
262 	if (act) {
263 		/*
264 		 * Check for invalid requests.  KILL and STOP cannot be
265 		 * caught.
266 		 */
267 		if (sig == SIGKILL || sig == SIGSTOP) {
268 			if (act->sa_handler != SIG_DFL) {
269 				lwkt_reltoken(&p->p_token);
270 				return (EINVAL);
271 			}
272 		}
273 
274 		/*
275 		 * Change setting atomically.
276 		 */
277 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
278 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
279 		if (act->sa_flags & SA_SIGINFO) {
280 			ps->ps_sigact[_SIG_IDX(sig)] =
281 			    (__sighandler_t *)act->sa_sigaction;
282 			SIGADDSET(ps->ps_siginfo, sig);
283 		} else {
284 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
285 			SIGDELSET(ps->ps_siginfo, sig);
286 		}
287 		if (!(act->sa_flags & SA_RESTART))
288 			SIGADDSET(ps->ps_sigintr, sig);
289 		else
290 			SIGDELSET(ps->ps_sigintr, sig);
291 		if (act->sa_flags & SA_ONSTACK)
292 			SIGADDSET(ps->ps_sigonstack, sig);
293 		else
294 			SIGDELSET(ps->ps_sigonstack, sig);
295 		if (act->sa_flags & SA_RESETHAND)
296 			SIGADDSET(ps->ps_sigreset, sig);
297 		else
298 			SIGDELSET(ps->ps_sigreset, sig);
299 		if (act->sa_flags & SA_NODEFER)
300 			SIGADDSET(ps->ps_signodefer, sig);
301 		else
302 			SIGDELSET(ps->ps_signodefer, sig);
303 		if (sig == SIGCHLD) {
304 			if (act->sa_flags & SA_NOCLDSTOP)
305 				p->p_sigacts->ps_flag |= PS_NOCLDSTOP;
306 			else
307 				p->p_sigacts->ps_flag &= ~PS_NOCLDSTOP;
308 			if (act->sa_flags & SA_NOCLDWAIT) {
309 				/*
310 				 * Paranoia: since SA_NOCLDWAIT is implemented
311 				 * by reparenting the dying child to PID 1 (and
312 				 * trust it to reap the zombie), PID 1 itself
313 				 * is forbidden to set SA_NOCLDWAIT.
314 				 */
315 				if (p->p_pid == 1)
316 					p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
317 				else
318 					p->p_sigacts->ps_flag |= PS_NOCLDWAIT;
319 			} else {
320 				p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
321 			}
322 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
323 				ps->ps_flag |= PS_CLDSIGIGN;
324 			else
325 				ps->ps_flag &= ~PS_CLDSIGIGN;
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_ATOMIC(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 	lwkt_reltoken(&p->p_token);
359 	return (0);
360 }
361 
362 int
363 sys_sigaction(struct sigaction_args *uap)
364 {
365 	struct sigaction act, oact;
366 	struct sigaction *actp, *oactp;
367 	int error;
368 
369 	actp = (uap->act != NULL) ? &act : NULL;
370 	oactp = (uap->oact != NULL) ? &oact : NULL;
371 	if (actp) {
372 		error = copyin(uap->act, actp, sizeof(act));
373 		if (error)
374 			return (error);
375 	}
376 	error = kern_sigaction(uap->sig, actp, oactp);
377 	if (oactp && !error) {
378 		error = copyout(oactp, uap->oact, sizeof(oact));
379 	}
380 	return (error);
381 }
382 
383 /*
384  * Initialize signal state for process 0;
385  * set to ignore signals that are ignored by default.
386  */
387 void
388 siginit(struct proc *p)
389 {
390 	int i;
391 
392 	for (i = 1; i <= NSIG; i++)
393 		if (sigprop(i) & SA_IGNORE && i != SIGCONT)
394 			SIGADDSET(p->p_sigignore, i);
395 }
396 
397 /*
398  * Reset signals for an exec of the specified process.
399  */
400 void
401 execsigs(struct proc *p)
402 {
403 	struct sigacts *ps = p->p_sigacts;
404 	struct lwp *lp;
405 	int sig;
406 
407 	lp = ONLY_LWP_IN_PROC(p);
408 
409 	/*
410 	 * Reset caught signals.  Held signals remain held
411 	 * through p_sigmask (unless they were caught,
412 	 * and are now ignored by default).
413 	 */
414 	while (SIGNOTEMPTY(p->p_sigcatch)) {
415 		sig = sig_ffs(&p->p_sigcatch);
416 		SIGDELSET(p->p_sigcatch, sig);
417 		if (sigprop(sig) & SA_IGNORE) {
418 			if (sig != SIGCONT)
419 				SIGADDSET(p->p_sigignore, sig);
420 			SIGDELSET_ATOMIC(p->p_siglist, sig);
421 			/* don't need spinlock */
422 			SIGDELSET(lp->lwp_siglist, sig);
423 		}
424 		ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
425 	}
426 
427 	/*
428 	 * Reset stack state to the user stack.
429 	 * Clear set of signals caught on the signal stack.
430 	 */
431 	lp->lwp_sigstk.ss_flags = SS_DISABLE;
432 	lp->lwp_sigstk.ss_size = 0;
433 	lp->lwp_sigstk.ss_sp = NULL;
434 	lp->lwp_flags &= ~LWP_ALTSTACK;
435 	/*
436 	 * Reset no zombies if child dies flag as Solaris does.
437 	 */
438 	p->p_sigacts->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
439 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
440 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
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 			return (EINVAL);
614 		if (ss->ss_flags & SS_DISABLE) {
615 			if (lp->lwp_sigstk.ss_flags & SS_ONSTACK)
616 				return (EPERM);
617 			lp->lwp_flags &= ~LWP_ALTSTACK;
618 			lp->lwp_sigstk.ss_flags = ss->ss_flags;
619 		} else {
620 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
621 				return (ENOMEM);
622 			lp->lwp_flags |= LWP_ALTSTACK;
623 			lp->lwp_sigstk = *ss;
624 		}
625 	}
626 
627 	return (0);
628 }
629 
630 /*
631  * MPSAFE
632  */
633 int
634 sys_sigaltstack(struct sigaltstack_args *uap)
635 {
636 	stack_t ss, oss;
637 	int error;
638 
639 	if (uap->ss) {
640 		error = copyin(uap->ss, &ss, sizeof(ss));
641 		if (error)
642 			return (error);
643 	}
644 
645 	error = kern_sigaltstack(uap->ss ? &ss : NULL, uap->oss ? &oss : NULL);
646 
647 	if (error == 0 && uap->oss)
648 		error = copyout(&oss, uap->oss, sizeof(*uap->oss));
649 	return (error);
650 }
651 
652 /*
653  * Common code for kill process group/broadcast kill.
654  * cp is calling process.
655  */
656 struct killpg_info {
657 	int nfound;
658 	int sig;
659 };
660 
661 static int killpg_all_callback(struct proc *p, void *data);
662 
663 static int
664 dokillpg(int sig, int pgid, int all)
665 {
666 	struct killpg_info info;
667 	struct proc *cp = curproc;
668 	struct proc *p;
669 	struct pgrp *pgrp;
670 
671 	info.nfound = 0;
672 	info.sig = sig;
673 
674 	if (all) {
675 		/*
676 		 * broadcast
677 		 */
678 		allproc_scan(killpg_all_callback, &info, 0);
679 	} else {
680 		if (pgid == 0) {
681 			/*
682 			 * zero pgid means send to my process group.
683 			 */
684 			pgrp = cp->p_pgrp;
685 			pgref(pgrp);
686 		} else {
687 			pgrp = pgfind(pgid);
688 			if (pgrp == NULL)
689 				return (ESRCH);
690 		}
691 
692 		/*
693 		 * Must interlock all signals against fork
694 		 */
695 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
696 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
697 			if (p->p_pid <= 1 ||
698 			    p->p_stat == SZOMB ||
699 			    (p->p_flags & P_SYSTEM) ||
700 			    !CANSIGNAL(p, sig, 0)) {
701 				continue;
702 			}
703 			++info.nfound;
704 			if (sig)
705 				ksignal(p, sig);
706 		}
707 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
708 		pgrel(pgrp);
709 	}
710 	return (info.nfound ? 0 : ESRCH);
711 }
712 
713 static int
714 killpg_all_callback(struct proc *p, void *data)
715 {
716 	struct killpg_info *info = data;
717 
718 	if (p->p_pid <= 1 || (p->p_flags & P_SYSTEM) ||
719 	    p == curproc || !CANSIGNAL(p, info->sig, 0)) {
720 		return (0);
721 	}
722 	++info->nfound;
723 	if (info->sig)
724 		ksignal(p, info->sig);
725 	return(0);
726 }
727 
728 /*
729  * Send a general signal to a process or LWPs within that process.
730  *
731  * Note that new signals cannot be sent if a process is exiting or already
732  * a zombie, but we return success anyway as userland is likely to not handle
733  * the race properly.
734  *
735  * No requirements.
736  */
737 int
738 kern_kill(int sig, pid_t pid, lwpid_t tid)
739 {
740 	int t;
741 
742 	if ((u_int)sig > _SIG_MAXSIG)
743 		return (EINVAL);
744 
745 	if (pid > 0) {
746 		struct proc *p;
747 		struct lwp *lp = NULL;
748 
749 		/*
750 		 * Send a signal to a single process.  If the kill() is
751 		 * racing an exiting process which has not yet been reaped
752 		 * act as though the signal was delivered successfully but
753 		 * don't actually try to deliver the signal.
754 		 */
755 		if ((p = pfind(pid)) == NULL) {
756 			if ((p = zpfind(pid)) == NULL)
757 				return (ESRCH);
758 			PRELE(p);
759 			return (0);
760 		}
761 		if (p != curproc) {
762 			lwkt_gettoken_shared(&p->p_token);
763 			if (!CANSIGNAL(p, sig, 1)) {
764 				lwkt_reltoken(&p->p_token);
765 				PRELE(p);
766 				return (EPERM);
767 			}
768 			lwkt_reltoken(&p->p_token);
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 			PRELE(p);
778 			return (0);
779 		}
780 		if (tid != -1) {
781 			lwkt_gettoken_shared(&p->p_token);
782 			lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
783 			if (lp == NULL) {
784 				lwkt_reltoken(&p->p_token);
785 				PRELE(p);
786 				return (ESRCH);
787 			}
788 			LWPHOLD(lp);
789 			lwkt_reltoken(&p->p_token);
790 		}
791 		if (sig)
792 			lwpsignal(p, lp, sig);
793 		if (lp)
794 			LWPRELE(lp);
795 		PRELE(p);
796 
797 		return (0);
798 	}
799 
800 	/*
801 	 * If we come here, pid is a special broadcast pid.
802 	 * This doesn't mix with a tid.
803 	 */
804 	if (tid != -1)
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 	return t;
819 }
820 
821 int
822 sys_kill(struct kill_args *uap)
823 {
824 	int error;
825 
826 	error = kern_kill(uap->signum, uap->pid, -1);
827 	return (error);
828 }
829 
830 int
831 sys_lwp_kill(struct lwp_kill_args *uap)
832 {
833 	int error;
834 	pid_t pid = uap->pid;
835 
836 	/*
837 	 * A tid is mandatory for lwp_kill(), otherwise
838 	 * you could simply use kill().
839 	 */
840 	if (uap->tid == -1)
841 		return (EINVAL);
842 
843 	/*
844 	 * To save on a getpid() function call for intra-process
845 	 * signals, pid == -1 means current process.
846 	 */
847 	if (pid == -1)
848 		pid = curproc->p_pid;
849 
850 	error = kern_kill(uap->signum, pid, uap->tid);
851 	return (error);
852 }
853 
854 /*
855  * Send a signal to a process group.
856  */
857 void
858 gsignal(int pgid, int sig)
859 {
860 	struct pgrp *pgrp;
861 
862 	if (pgid && (pgrp = pgfind(pgid)))
863 		pgsignal(pgrp, sig, 0);
864 }
865 
866 /*
867  * Send a signal to a process group.  If checktty is 1,
868  * limit to members which have a controlling terminal.
869  *
870  * pg_lock interlocks against a fork that might be in progress, to
871  * ensure that the new child process picks up the signal.
872  */
873 void
874 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
875 {
876 	struct proc *p;
877 
878 	/*
879 	 * Must interlock all signals against fork
880 	 */
881 	if (pgrp) {
882 		pgref(pgrp);
883 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
884 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
885 			if (checkctty == 0 || p->p_flags & P_CONTROLT)
886 				ksignal(p, sig);
887 		}
888 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
889 		pgrel(pgrp);
890 	}
891 }
892 
893 /*
894  * Send a signal caused by a trap to the current lwp.  If it will be caught
895  * immediately, deliver it with correct code.  Otherwise, post it normally.
896  *
897  * These signals may ONLY be delivered to the specified lwp and may never
898  * be delivered to the process generically.
899  */
900 void
901 trapsignal(struct lwp *lp, int sig, u_long code)
902 {
903 	struct proc *p = lp->lwp_proc;
904 	struct sigacts *ps = p->p_sigacts;
905 
906 	/*
907 	 * If we are a virtual kernel running an emulated user process
908 	 * context, switch back to the virtual kernel context before
909 	 * trying to post the signal.
910 	 */
911 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
912 		struct trapframe *tf = lp->lwp_md.md_regs;
913 		tf->tf_trapno = 0;
914 		vkernel_trap(lp, tf);
915 	}
916 
917 	if ((p->p_flags & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
918 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
919 		lp->lwp_ru.ru_nsignals++;
920 #ifdef KTRACE
921 		if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
922 			ktrpsig(lp, sig, ps->ps_sigact[_SIG_IDX(sig)],
923 				&lp->lwp_sigmask, code);
924 #endif
925 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
926 						&lp->lwp_sigmask, code);
927 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
928 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
929 			SIGADDSET(lp->lwp_sigmask, sig);
930 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
931 			/*
932 			 * See kern_sigaction() for origin of this code.
933 			 */
934 			SIGDELSET(p->p_sigcatch, sig);
935 			if (sig != SIGCONT &&
936 			    sigprop(sig) & SA_IGNORE)
937 				SIGADDSET(p->p_sigignore, sig);
938 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
939 		}
940 	} else {
941 		lp->lwp_code = code;	/* XXX for core dump/debugger */
942 		lp->lwp_sig = sig;	/* XXX to verify code */
943 		lwpsignal(p, lp, sig);
944 	}
945 }
946 
947 /*
948  * Find a suitable lwp to deliver the signal to.  Returns NULL if all
949  * lwps hold the signal blocked.
950  *
951  * Caller must hold p->p_token.
952  *
953  * Returns a lp or NULL.  If non-NULL the lp is held and its token is
954  * acquired.
955  */
956 static struct lwp *
957 find_lwp_for_signal(struct proc *p, int sig)
958 {
959 	struct lwp *lp;
960 	struct lwp *run, *sleep, *stop;
961 
962 	/*
963 	 * If the running/preempted thread belongs to the proc to which
964 	 * the signal is being delivered and this thread does not block
965 	 * the signal, then we can avoid a context switch by delivering
966 	 * the signal to this thread, because it will return to userland
967 	 * soon anyways.
968 	 */
969 	lp = lwkt_preempted_proc();
970 	if (lp != NULL && lp->lwp_proc == p) {
971 		LWPHOLD(lp);
972 		lwkt_gettoken(&lp->lwp_token);
973 		if (!SIGISMEMBER(lp->lwp_sigmask, sig)) {
974 			/* return w/ token held */
975 			return (lp);
976 		}
977 		lwkt_reltoken(&lp->lwp_token);
978 		LWPRELE(lp);
979 	}
980 
981 	run = sleep = stop = NULL;
982 	FOREACH_LWP_IN_PROC(lp, p) {
983 		/*
984 		 * If the signal is being blocked by the lwp, then this
985 		 * lwp is not eligible for receiving the signal.
986 		 */
987 		LWPHOLD(lp);
988 		lwkt_gettoken(&lp->lwp_token);
989 
990 		if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
991 			lwkt_reltoken(&lp->lwp_token);
992 			LWPRELE(lp);
993 			continue;
994 		}
995 
996 		switch (lp->lwp_stat) {
997 		case LSRUN:
998 			if (sleep) {
999 				lwkt_token_swap();
1000 				lwkt_reltoken(&sleep->lwp_token);
1001 				LWPRELE(sleep);
1002 				sleep = NULL;
1003 				run = lp;
1004 			} else if (stop) {
1005 				lwkt_token_swap();
1006 				lwkt_reltoken(&stop->lwp_token);
1007 				LWPRELE(stop);
1008 				stop = NULL;
1009 				run = lp;
1010 			} else {
1011 				run = lp;
1012 			}
1013 			break;
1014 		case LSSLEEP:
1015 			if (lp->lwp_flags & LWP_SINTR) {
1016 				if (sleep) {
1017 					lwkt_reltoken(&lp->lwp_token);
1018 					LWPRELE(lp);
1019 				} else if (stop) {
1020 					lwkt_token_swap();
1021 					lwkt_reltoken(&stop->lwp_token);
1022 					LWPRELE(stop);
1023 					stop = NULL;
1024 					sleep = lp;
1025 				} else {
1026 					sleep = lp;
1027 				}
1028 			} else {
1029 				lwkt_reltoken(&lp->lwp_token);
1030 				LWPRELE(lp);
1031 			}
1032 			break;
1033 		case LSSTOP:
1034 			if (sleep) {
1035 				lwkt_reltoken(&lp->lwp_token);
1036 				LWPRELE(lp);
1037 			} else if (stop) {
1038 				lwkt_reltoken(&lp->lwp_token);
1039 				LWPRELE(lp);
1040 			} else {
1041 				stop = lp;
1042 			}
1043 			break;
1044 		}
1045 		if (run)
1046 			break;
1047 	}
1048 
1049 	if (run != NULL)
1050 		return (run);
1051 	else if (sleep != NULL)
1052 		return (sleep);
1053 	else
1054 		return (stop);
1055 }
1056 
1057 /*
1058  * Send the signal to the process.  If the signal has an action, the action
1059  * is usually performed by the target process rather than the caller; we add
1060  * the signal to the set of pending signals for the process.
1061  *
1062  * Exceptions:
1063  *   o When a stop signal is sent to a sleeping process that takes the
1064  *     default action, the process is stopped without awakening it.
1065  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
1066  *     regardless of the signal action (eg, blocked or ignored).
1067  *
1068  * Other ignored signals are discarded immediately.
1069  *
1070  * If the caller wishes to call this function from a hard code section the
1071  * caller must already hold p->p_token (see kern_clock.c).
1072  *
1073  * No requirements.
1074  */
1075 void
1076 ksignal(struct proc *p, int sig)
1077 {
1078 	lwpsignal(p, NULL, sig);
1079 }
1080 
1081 /*
1082  * The core for ksignal.  lp may be NULL, then a suitable thread
1083  * will be chosen.  If not, lp MUST be a member of p.
1084  *
1085  * If the caller wishes to call this function from a hard code section the
1086  * caller must already hold p->p_token.
1087  *
1088  * No requirements.
1089  */
1090 void
1091 lwpsignal(struct proc *p, struct lwp *lp, int sig)
1092 {
1093 	struct proc *q;
1094 	sig_t action;
1095 	int prop;
1096 
1097 	if (sig > _SIG_MAXSIG || sig <= 0) {
1098 		kprintf("lwpsignal: signal %d\n", sig);
1099 		panic("lwpsignal signal number");
1100 	}
1101 
1102 	KKASSERT(lp == NULL || lp->lwp_proc == p);
1103 
1104 	/*
1105 	 * We don't want to race... well, all sorts of things.  Get appropriate
1106 	 * tokens.
1107 	 *
1108 	 * Don't try to deliver a generic signal to an exiting process,
1109 	 * the signal structures could be in flux.  We check the LWP later
1110 	 * on.
1111 	 */
1112 	PHOLD(p);
1113 	if (lp) {
1114 		LWPHOLD(lp);
1115 		lwkt_gettoken(&lp->lwp_token);
1116 	} else {
1117 		lwkt_gettoken(&p->p_token);
1118 		if (p->p_flags & P_WEXIT)
1119 			goto out;
1120 	}
1121 
1122 	prop = sigprop(sig);
1123 
1124 	/*
1125 	 * If proc is traced, always give parent a chance;
1126 	 * if signal event is tracked by procfs, give *that*
1127 	 * a chance, as well.
1128 	 */
1129 	if ((p->p_flags & P_TRACED) || (p->p_stops & S_SIG)) {
1130 		action = SIG_DFL;
1131 	} else {
1132 		/*
1133 		 * Do not try to deliver signals to an exiting lwp other
1134 		 * than SIGKILL.  Note that we must still deliver the signal
1135 		 * if P_WEXIT is set in the process flags.
1136 		 */
1137 		if (lp && (lp->lwp_mpflags & LWP_MP_WEXIT) && sig != SIGKILL) {
1138 			lwkt_reltoken(&lp->lwp_token);
1139 			LWPRELE(lp);
1140 			PRELE(p);
1141 			return;
1142 		}
1143 
1144 		/*
1145 		 * If the signal is being ignored, then we forget about
1146 		 * it immediately.  NOTE: We don't set SIGCONT in p_sigignore,
1147 		 * and if it is set to SIG_IGN, action will be SIG_DFL here.
1148 		 */
1149 		if (SIGISMEMBER(p->p_sigignore, sig)) {
1150 			/*
1151 			 * Even if a signal is set SIG_IGN, it may still be
1152 			 * lurking in a kqueue.
1153 			 */
1154 			KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1155 			if (lp) {
1156 				lwkt_reltoken(&lp->lwp_token);
1157 				LWPRELE(lp);
1158 			} else {
1159 				lwkt_reltoken(&p->p_token);
1160 			}
1161 			PRELE(p);
1162 			return;
1163 		}
1164 		if (SIGISMEMBER(p->p_sigcatch, sig))
1165 			action = SIG_CATCH;
1166 		else
1167 			action = SIG_DFL;
1168 	}
1169 
1170 	/*
1171 	 * If continuing, clear any pending STOP signals for the whole
1172 	 * process.
1173 	 */
1174 	if (prop & SA_CONT) {
1175 		lwkt_gettoken(&p->p_token);
1176 		SIG_STOPSIGMASK_ATOMIC(p->p_siglist);
1177 		lwkt_reltoken(&p->p_token);
1178 	}
1179 
1180 	if (prop & SA_STOP) {
1181 		/*
1182 		 * If sending a tty stop signal to a member of an orphaned
1183 		 * process group, discard the signal here if the action
1184 		 * is default; don't stop the process below if sleeping,
1185 		 * and don't clear any pending SIGCONT.
1186 		 */
1187 		if ((prop & SA_TTYSTOP) && p->p_pgrp->pg_jobc == 0 &&
1188 		    action == SIG_DFL) {
1189 			if (lp) {
1190 				lwkt_reltoken(&lp->lwp_token);
1191 				LWPRELE(lp);
1192 			} else {
1193 				lwkt_reltoken(&p->p_token);
1194 			}
1195 			PRELE(p);
1196 		        return;
1197 		}
1198 		lwkt_gettoken(&p->p_token);
1199 		SIG_CONTSIGMASK_ATOMIC(p->p_siglist);
1200 		p->p_flags &= ~P_CONTINUED;
1201 		lwkt_reltoken(&p->p_token);
1202 	}
1203 
1204 	if (p->p_stat == SSTOP) {
1205 		/*
1206 		 * Nobody can handle this signal, add it to the lwp or
1207 		 * process pending list
1208 		 */
1209 		lwkt_gettoken(&p->p_token);
1210 		if (p->p_stat != SSTOP) {
1211 			lwkt_reltoken(&p->p_token);
1212 			goto not_stopped;
1213 		}
1214 		if (lp) {
1215 			spin_lock(&lp->lwp_spin);
1216 			SIGADDSET(lp->lwp_siglist, sig);
1217 			spin_unlock(&lp->lwp_spin);
1218 		} else {
1219 			SIGADDSET_ATOMIC(p->p_siglist, sig);
1220 		}
1221 
1222 		/*
1223 		 * If the process is stopped and is being traced, then no
1224 		 * further action is necessary.
1225 		 */
1226 		if (p->p_flags & P_TRACED) {
1227 			lwkt_reltoken(&p->p_token);
1228 			goto out;
1229 		}
1230 
1231 		/*
1232 		 * If the process is stopped and receives a KILL signal,
1233 		 * make the process runnable.
1234 		 */
1235 		if (sig == SIGKILL) {
1236 			proc_unstop(p, SSTOP);
1237 			lwkt_reltoken(&p->p_token);
1238 			goto active_process;
1239 		}
1240 
1241 		/*
1242 		 * If the process is stopped and receives a CONT signal,
1243 		 * then try to make the process runnable again.
1244 		 */
1245 		if (prop & SA_CONT) {
1246 			/*
1247 			 * If SIGCONT is default (or ignored), we continue the
1248 			 * process but don't leave the signal in p_siglist, as
1249 			 * it has no further action.  If SIGCONT is held, we
1250 			 * continue the process and leave the signal in
1251 			 * p_siglist.  If the process catches SIGCONT, let it
1252 			 * handle the signal itself.
1253 			 *
1254 			 * XXX what if the signal is being held blocked?
1255 			 *
1256 			 * Token required to interlock kern_wait().
1257 			 * Reparenting can also cause a race so we have to
1258 			 * hold (q).
1259 			 */
1260 			q = p->p_pptr;
1261 			PHOLD(q);
1262 			lwkt_gettoken(&q->p_token);
1263 			p->p_flags |= P_CONTINUED;
1264 			wakeup(q);
1265 			if (action == SIG_DFL)
1266 				SIGDELSET_ATOMIC(p->p_siglist, sig);
1267 			proc_unstop(p, SSTOP);
1268 			lwkt_reltoken(&q->p_token);
1269 			PRELE(q);
1270 			lwkt_reltoken(&p->p_token);
1271 			if (action == SIG_CATCH)
1272 				goto active_process;
1273 			goto out;
1274 		}
1275 
1276 		/*
1277 		 * If the process is stopped and receives another STOP
1278 		 * signal, we do not need to stop it again.  If we did
1279 		 * the shell could get confused.
1280 		 *
1281 		 * However, if the current/preempted lwp is part of the
1282 		 * process receiving the signal, we need to keep it,
1283 		 * so that this lwp can stop in issignal() later, as
1284 		 * we don't want to wait until it reaches userret!
1285 		 */
1286 		if (prop & SA_STOP) {
1287 			if (lwkt_preempted_proc() == NULL ||
1288 			    lwkt_preempted_proc()->lwp_proc != p) {
1289 				SIGDELSET_ATOMIC(p->p_siglist, sig);
1290 			}
1291 		}
1292 
1293 		/*
1294 		 * Otherwise the process is stopped and it received some
1295 		 * signal, which does not change its stopped state.  When
1296 		 * the process is continued a wakeup(p) will be issued which
1297 		 * will wakeup any threads sleeping in tstop().
1298 		 */
1299 		lwkt_reltoken(&p->p_token);
1300 		goto out;
1301 		/* NOTREACHED */
1302 	}
1303 not_stopped:
1304 	;
1305 	/* else not stopped */
1306 active_process:
1307 
1308 	/*
1309 	 * Never deliver a lwp-specific signal to a random lwp.
1310 	 */
1311 	if (lp == NULL) {
1312 		/* NOTE: returns lp w/ token held */
1313 		lp = find_lwp_for_signal(p, sig);
1314 		if (lp) {
1315 			if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
1316 				lwkt_reltoken(&lp->lwp_token);
1317 				LWPRELE(lp);
1318 				lp = NULL;
1319 				/* maintain proc token */
1320 			} else {
1321 				lwkt_token_swap();
1322 				lwkt_reltoken(&p->p_token);
1323 				/* maintain lp token */
1324 			}
1325 		}
1326 	}
1327 
1328 	/*
1329 	 * Deliver to the process generically if (1) the signal is being
1330 	 * sent to any thread or (2) we could not find a thread to deliver
1331 	 * it to.
1332 	 */
1333 	if (lp == NULL) {
1334 		KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1335 		SIGADDSET_ATOMIC(p->p_siglist, sig);
1336 		goto out;
1337 	}
1338 
1339 	/*
1340 	 * Deliver to a specific LWP whether it masks it or not.  It will
1341 	 * not be dispatched if masked but we must still deliver it.
1342 	 */
1343 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1344 	    (p->p_flags & P_TRACED) == 0) {
1345 		lwkt_gettoken(&p->p_token);
1346 		p->p_nice = NZERO;
1347 		lwkt_reltoken(&p->p_token);
1348 	}
1349 
1350 	/*
1351 	 * If the process receives a STOP signal which indeed needs to
1352 	 * stop the process, do so.  If the process chose to catch the
1353 	 * signal, it will be treated like any other signal.
1354 	 */
1355 	if ((prop & SA_STOP) && action == SIG_DFL) {
1356 		/*
1357 		 * If a child holding parent blocked, stopping
1358 		 * could cause deadlock.  Take no action at this
1359 		 * time.
1360 		 */
1361 		lwkt_gettoken(&p->p_token);
1362 		if (p->p_flags & P_PPWAIT) {
1363 			SIGADDSET_ATOMIC(p->p_siglist, sig);
1364 			lwkt_reltoken(&p->p_token);
1365 			goto out;
1366 		}
1367 
1368 		/*
1369 		 * Do not actually try to manipulate the process, but simply
1370 		 * stop it.  Lwps will stop as soon as they safely can.
1371 		 *
1372 		 * Ignore stop if the process is exiting.
1373 		 */
1374 		if ((p->p_flags & P_WEXIT) == 0) {
1375 			p->p_xstat = sig;
1376 			proc_stop(p, SSTOP);
1377 		}
1378 		lwkt_reltoken(&p->p_token);
1379 		goto out;
1380 	}
1381 
1382 	/*
1383 	 * If it is a CONT signal with default action, just ignore it.
1384 	 */
1385 	if ((prop & SA_CONT) && action == SIG_DFL)
1386 		goto out;
1387 
1388 	/*
1389 	 * Mark signal pending at this specific thread.
1390 	 */
1391 	spin_lock(&lp->lwp_spin);
1392 	SIGADDSET(lp->lwp_siglist, sig);
1393 	spin_unlock(&lp->lwp_spin);
1394 
1395 	lwp_signotify(lp);
1396 
1397 out:
1398 	if (lp) {
1399 		lwkt_reltoken(&lp->lwp_token);
1400 		LWPRELE(lp);
1401 	} else {
1402 		lwkt_reltoken(&p->p_token);
1403 	}
1404 	PRELE(p);
1405 }
1406 
1407 /*
1408  * Notify the LWP that a signal has arrived.  The LWP does not have to be
1409  * sleeping on the current cpu.
1410  *
1411  * p->p_token and lp->lwp_token must be held on call.
1412  *
1413  * We can only safely schedule the thread on its current cpu and only if
1414  * one of the SINTR flags is set.  If an SINTR flag is set AND we are on
1415  * the correct cpu we are properly interlocked, otherwise we could be
1416  * racing other thread transition states (or the lwp is on the user scheduler
1417  * runq but not scheduled) and must not do anything.
1418  *
1419  * Since we hold the lwp token we know the lwp cannot be ripped out from
1420  * under us so we can safely hold it to prevent it from being ripped out
1421  * from under us if we are forced to IPI another cpu to make the local
1422  * checks there.
1423  *
1424  * Adjustment of lp->lwp_stat can only occur when we hold the lwp_token,
1425  * which we won't in an IPI so any fixups have to be done here, effectively
1426  * replicating part of what setrunnable() does.
1427  */
1428 static void
1429 lwp_signotify(struct lwp *lp)
1430 {
1431 	thread_t dtd;
1432 
1433 	ASSERT_LWKT_TOKEN_HELD(&lp->lwp_token);
1434 	dtd = lp->lwp_thread;
1435 
1436 	crit_enter();
1437 	if (lp == lwkt_preempted_proc()) {
1438 		/*
1439 		 * lwp is on the current cpu AND it is currently running
1440 		 * (we preempted it).
1441 		 */
1442 		signotify();
1443 	} else if (lp->lwp_flags & LWP_SINTR) {
1444 		/*
1445 		 * lwp is sitting in tsleep() with PCATCH set
1446 		 */
1447 		if (dtd->td_gd == mycpu) {
1448 			setrunnable(lp);
1449 		} else {
1450 			/*
1451 			 * We can only adjust lwp_stat while we hold the
1452 			 * lwp_token, and we won't in the IPI function.
1453 			 */
1454 			LWPHOLD(lp);
1455 			if (lp->lwp_stat == LSSTOP)
1456 				lp->lwp_stat = LSSLEEP;
1457 			lwkt_send_ipiq(dtd->td_gd, lwp_signotify_remote, lp);
1458 		}
1459 	} else if (dtd->td_flags & TDF_SINTR) {
1460 		/*
1461 		 * lwp is sitting in lwkt_sleep() with PCATCH set.
1462 		 */
1463 		if (dtd->td_gd == mycpu) {
1464 			setrunnable(lp);
1465 		} else {
1466 			/*
1467 			 * We can only adjust lwp_stat while we hold the
1468 			 * lwp_token, and we won't in the IPI function.
1469 			 */
1470 			LWPHOLD(lp);
1471 			if (lp->lwp_stat == LSSTOP)
1472 				lp->lwp_stat = LSSLEEP;
1473 			lwkt_send_ipiq(dtd->td_gd, lwp_signotify_remote, lp);
1474 		}
1475 	} else {
1476 		/*
1477 		 * Otherwise the lwp is either in some uninterruptible state
1478 		 * or it is on the userland scheduler's runqueue waiting to
1479 		 * be scheduled to a cpu, or it is running in userland.  We
1480 		 * generally want to send an IPI so a running target gets the
1481 		 * signal ASAP, otherwise a scheduler-tick worth of latency
1482 		 * will occur.
1483 		 *
1484 		 * Issue an IPI to the remote cpu to knock it into the kernel,
1485 		 * remote cpu will issue the cpu-local signotify() if the IPI
1486 		 * preempts the desired thread.
1487 		 */
1488 		if (dtd->td_gd != mycpu) {
1489 			LWPHOLD(lp);
1490 			lwkt_send_ipiq(dtd->td_gd, lwp_signotify_remote, lp);
1491 		}
1492 	}
1493 	crit_exit();
1494 }
1495 
1496 /*
1497  * This function is called via an IPI so we cannot call setrunnable() here
1498  * (because while we hold the lp we don't own its token, and can't get it
1499  * from an IPI).
1500  *
1501  * We are interlocked by virtue of being on the same cpu as the target.  If
1502  * we still are and LWP_SINTR or TDF_SINTR is set we can safely schedule
1503  * the target thread.
1504  */
1505 static void
1506 lwp_signotify_remote(void *arg)
1507 {
1508 	struct lwp *lp = arg;
1509 	thread_t td = lp->lwp_thread;
1510 
1511 	if (lp == lwkt_preempted_proc()) {
1512 		signotify();
1513 		LWPRELE(lp);
1514 	} else if (td->td_gd == mycpu) {
1515 		if ((lp->lwp_flags & LWP_SINTR) ||
1516 		    (td->td_flags & TDF_SINTR)) {
1517 			lwkt_schedule(td);
1518 		}
1519 		LWPRELE(lp);
1520 	} else {
1521 		lwkt_send_ipiq(td->td_gd, lwp_signotify_remote, lp);
1522 		/* LWPHOLD() is forwarded to the target cpu */
1523 	}
1524 }
1525 
1526 /*
1527  * Caller must hold p->p_token
1528  */
1529 void
1530 proc_stop(struct proc *p, int sig)
1531 {
1532 	struct proc *q;
1533 	struct lwp *lp;
1534 
1535 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1536 
1537 	/*
1538 	 * If somebody raced us, be happy with it.  SCORE overrides SSTOP.
1539 	 */
1540 	if (sig == SCORE) {
1541 		if (p->p_stat == SCORE || p->p_stat == SZOMB)
1542 			return;
1543 	} else {
1544 		if (p->p_stat == SSTOP || p->p_stat == SCORE ||
1545 		    p->p_stat == SZOMB) {
1546 			return;
1547 		}
1548 	}
1549 	p->p_stat = sig;
1550 
1551 	FOREACH_LWP_IN_PROC(lp, p) {
1552 		LWPHOLD(lp);
1553 		lwkt_gettoken(&lp->lwp_token);
1554 
1555 		switch (lp->lwp_stat) {
1556 		case LSSTOP:
1557 			/*
1558 			 * Do nothing, we are already counted in
1559 			 * p_nstopped.
1560 			 */
1561 			break;
1562 
1563 		case LSSLEEP:
1564 			/*
1565 			 * We're sleeping, but we will stop before
1566 			 * returning to userspace, so count us
1567 			 * as stopped as well.  We set LWP_MP_WSTOP
1568 			 * to signal the lwp that it should not
1569 			 * increase p_nstopped when reaching tstop().
1570 			 *
1571 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1572 			 */
1573 			if ((lp->lwp_mpflags & LWP_MP_WSTOP) == 0) {
1574 				atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
1575 				++p->p_nstopped;
1576 			}
1577 			break;
1578 
1579 		case LSRUN:
1580 			/*
1581 			 * We might notify ourself, but that's not
1582 			 * a problem.
1583 			 */
1584 			lwp_signotify(lp);
1585 			break;
1586 		}
1587 		lwkt_reltoken(&lp->lwp_token);
1588 		LWPRELE(lp);
1589 	}
1590 
1591 	if (p->p_nstopped == p->p_nthreads) {
1592 		/*
1593 		 * Token required to interlock kern_wait().  Reparenting can
1594 		 * also cause a race so we have to hold (q).
1595 		 */
1596 		q = p->p_pptr;
1597 		PHOLD(q);
1598 		lwkt_gettoken(&q->p_token);
1599 		p->p_flags &= ~P_WAITED;
1600 		wakeup(q);
1601 		if ((q->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1602 			ksignal(p->p_pptr, SIGCHLD);
1603 		lwkt_reltoken(&q->p_token);
1604 		PRELE(q);
1605 	}
1606 }
1607 
1608 /*
1609  * Caller must hold p_token
1610  */
1611 void
1612 proc_unstop(struct proc *p, int sig)
1613 {
1614 	struct lwp *lp;
1615 
1616 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1617 
1618 	if (p->p_stat != sig)
1619 		return;
1620 
1621 	p->p_stat = SACTIVE;
1622 
1623 	FOREACH_LWP_IN_PROC(lp, p) {
1624 		LWPHOLD(lp);
1625 		lwkt_gettoken(&lp->lwp_token);
1626 
1627 		switch (lp->lwp_stat) {
1628 		case LSRUN:
1629 			/*
1630 			 * Uh?  Not stopped?  Well, I guess that's okay.
1631 			 */
1632 			if (bootverbose)
1633 				kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1634 					p->p_pid, lp->lwp_tid);
1635 			break;
1636 
1637 		case LSSLEEP:
1638 			/*
1639 			 * Still sleeping.  Don't bother waking it up.
1640 			 * However, if this thread was counted as
1641 			 * stopped, undo this.
1642 			 *
1643 			 * Nevertheless we call setrunnable() so that it
1644 			 * will wake up in case a signal or timeout arrived
1645 			 * in the meantime.
1646 			 *
1647 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1648 			 */
1649 			if (lp->lwp_mpflags & LWP_MP_WSTOP) {
1650 				atomic_clear_int(&lp->lwp_mpflags,
1651 						 LWP_MP_WSTOP);
1652 				--p->p_nstopped;
1653 			} else {
1654 				if (bootverbose)
1655 					kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1656 						p->p_pid, lp->lwp_tid);
1657 			}
1658 			/* FALLTHROUGH */
1659 
1660 		case LSSTOP:
1661 			/*
1662 			 * This handles any lwp's waiting in a tsleep with
1663 			 * SIGCATCH.
1664 			 */
1665 			lwp_signotify(lp);
1666 			break;
1667 
1668 		}
1669 		lwkt_reltoken(&lp->lwp_token);
1670 		LWPRELE(lp);
1671 	}
1672 
1673 	/*
1674 	 * This handles any lwp's waiting in tstop().  We have interlocked
1675 	 * the setting of p_stat by acquiring and releasing each lpw's
1676 	 * token.
1677 	 */
1678 	wakeup(p);
1679 }
1680 
1681 /*
1682  * Wait for all threads except the current thread to stop.
1683  */
1684 static void
1685 proc_stopwait(struct proc *p)
1686 {
1687 	while ((p->p_stat == SSTOP || p->p_stat == SCORE) &&
1688 	       p->p_nstopped < p->p_nthreads - 1) {
1689 		tsleep_interlock(&p->p_nstopped, 0);
1690 		if (p->p_nstopped < p->p_nthreads - 1) {
1691 			tsleep(&p->p_nstopped, PINTERLOCKED, "stopwt", hz);
1692 		}
1693 	}
1694 }
1695 
1696 /*
1697  * No requirements.
1698  */
1699 static int
1700 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1701 {
1702 	sigset_t savedmask, set;
1703 	struct proc *p = curproc;
1704 	struct lwp *lp = curthread->td_lwp;
1705 	int error, sig, hz, timevalid = 0;
1706 	struct timespec rts, ets, ts;
1707 	struct timeval tv;
1708 
1709 	error = 0;
1710 	sig = 0;
1711 	ets.tv_sec = 0;		/* silence compiler warning */
1712 	ets.tv_nsec = 0;	/* silence compiler warning */
1713 	SIG_CANTMASK(waitset);
1714 	savedmask = lp->lwp_sigmask;
1715 
1716 	if (timeout) {
1717 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1718 		    timeout->tv_nsec < 1000000000) {
1719 			timevalid = 1;
1720 			getnanouptime(&rts);
1721 		 	ets = rts;
1722 			timespecadd(&ets, timeout);
1723 		}
1724 	}
1725 
1726 	for (;;) {
1727 		set = lwp_sigpend(lp);
1728 		SIGSETAND(set, waitset);
1729 		if ((sig = sig_ffs(&set)) != 0) {
1730 			SIGFILLSET(lp->lwp_sigmask);
1731 			SIGDELSET(lp->lwp_sigmask, sig);
1732 			SIG_CANTMASK(lp->lwp_sigmask);
1733 			sig = issignal(lp, 1, 0);
1734 			/*
1735 			 * It may be a STOP signal, in the case, issignal
1736 			 * returns 0, because we may stop there, and new
1737 			 * signal can come in, we should restart if we got
1738 			 * nothing.
1739 			 */
1740 			if (sig == 0)
1741 				continue;
1742 			else
1743 				break;
1744 		}
1745 
1746 		/*
1747 		 * Previous checking got nothing, and we retried but still
1748 		 * got nothing, we should return the error status.
1749 		 */
1750 		if (error)
1751 			break;
1752 
1753 		/*
1754 		 * POSIX says this must be checked after looking for pending
1755 		 * signals.
1756 		 */
1757 		if (timeout) {
1758 			if (timevalid == 0) {
1759 				error = EINVAL;
1760 				break;
1761 			}
1762 			getnanouptime(&rts);
1763 			if (timespeccmp(&rts, &ets, >=)) {
1764 				error = EAGAIN;
1765 				break;
1766 			}
1767 			ts = ets;
1768 			timespecsub(&ts, &rts);
1769 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1770 			hz = tvtohz_high(&tv);
1771 		} else {
1772 			hz = 0;
1773 		}
1774 
1775 		lp->lwp_sigmask = savedmask;
1776 		SIGSETNAND(lp->lwp_sigmask, waitset);
1777 		/*
1778 		 * We won't ever be woken up.  Instead, our sleep will
1779 		 * be broken in lwpsignal().
1780 		 */
1781 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1782 		if (timeout) {
1783 			if (error == ERESTART) {
1784 				/* can not restart a timeout wait. */
1785 				error = EINTR;
1786 			} else if (error == EAGAIN) {
1787 				/* will calculate timeout by ourself. */
1788 				error = 0;
1789 			}
1790 		}
1791 		/* Retry ... */
1792 	}
1793 
1794 	lp->lwp_sigmask = savedmask;
1795 	if (sig) {
1796 		error = 0;
1797 		bzero(info, sizeof(*info));
1798 		info->si_signo = sig;
1799 		spin_lock(&lp->lwp_spin);
1800 		lwp_delsig(lp, sig, 1);	/* take the signal! */
1801 		spin_unlock(&lp->lwp_spin);
1802 
1803 		if (sig == SIGKILL) {
1804 			sigexit(lp, sig);
1805 			/* NOT REACHED */
1806 		}
1807 	}
1808 
1809 	return (error);
1810 }
1811 
1812 /*
1813  * MPALMOSTSAFE
1814  */
1815 int
1816 sys_sigtimedwait(struct sigtimedwait_args *uap)
1817 {
1818 	struct timespec ts;
1819 	struct timespec *timeout;
1820 	sigset_t set;
1821 	siginfo_t info;
1822 	int error;
1823 
1824 	if (uap->timeout) {
1825 		error = copyin(uap->timeout, &ts, sizeof(ts));
1826 		if (error)
1827 			return (error);
1828 		timeout = &ts;
1829 	} else {
1830 		timeout = NULL;
1831 	}
1832 	error = copyin(uap->set, &set, sizeof(set));
1833 	if (error)
1834 		return (error);
1835 	error = kern_sigtimedwait(set, &info, timeout);
1836 	if (error)
1837 		return (error);
1838 	if (uap->info)
1839 		error = copyout(&info, uap->info, sizeof(info));
1840 	/* Repost if we got an error. */
1841 	/*
1842 	 * XXX lwp
1843 	 *
1844 	 * This could transform a thread-specific signal to another
1845 	 * thread / process pending signal.
1846 	 */
1847 	if (error) {
1848 		ksignal(curproc, info.si_signo);
1849 	} else {
1850 		uap->sysmsg_result = info.si_signo;
1851 	}
1852 	return (error);
1853 }
1854 
1855 /*
1856  * MPALMOSTSAFE
1857  */
1858 int
1859 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1860 {
1861 	siginfo_t info;
1862 	sigset_t set;
1863 	int error;
1864 
1865 	error = copyin(uap->set, &set, sizeof(set));
1866 	if (error)
1867 		return (error);
1868 	error = kern_sigtimedwait(set, &info, NULL);
1869 	if (error)
1870 		return (error);
1871 	if (uap->info)
1872 		error = copyout(&info, uap->info, sizeof(info));
1873 	/* Repost if we got an error. */
1874 	/*
1875 	 * XXX lwp
1876 	 *
1877 	 * This could transform a thread-specific signal to another
1878 	 * thread / process pending signal.
1879 	 */
1880 	if (error) {
1881 		ksignal(curproc, info.si_signo);
1882 	} else {
1883 		uap->sysmsg_result = info.si_signo;
1884 	}
1885 	return (error);
1886 }
1887 
1888 /*
1889  * If the current process has received a signal that would interrupt a
1890  * system call, return EINTR or ERESTART as appropriate.
1891  */
1892 int
1893 iscaught(struct lwp *lp)
1894 {
1895 	struct proc *p = lp->lwp_proc;
1896 	int sig;
1897 
1898 	if (p) {
1899 		if ((sig = CURSIG(lp)) != 0) {
1900 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1901 				return (EINTR);
1902 			return (ERESTART);
1903 		}
1904 	}
1905 	return(EWOULDBLOCK);
1906 }
1907 
1908 /*
1909  * If the current lwp/proc has received a signal (should be caught or cause
1910  * termination, should interrupt current syscall), return the signal number.
1911  * Stop signals with default action are processed immediately, then cleared;
1912  * they aren't returned.  This is checked after each entry to the system for
1913  * a syscall or trap (though this can usually be done without calling issignal
1914  * by checking the pending signal masks in the CURSIG macro).
1915  *
1916  * This routine is called via CURSIG/__cursig.  We will acquire and release
1917  * p->p_token but if the caller needs to interlock the test the caller must
1918  * also hold p->p_token.
1919  *
1920  *	while (sig = CURSIG(curproc))
1921  *		postsig(sig);
1922  */
1923 int
1924 issignal(struct lwp *lp, int maytrace, int *ptokp)
1925 {
1926 	struct proc *p = lp->lwp_proc;
1927 	sigset_t mask;
1928 	int sig, prop;
1929 	int haveptok;
1930 
1931 	for (;;) {
1932 		int traced = (p->p_flags & P_TRACED) || (p->p_stops & S_SIG);
1933 
1934 		haveptok = 0;
1935 
1936 		/*
1937 		 * If this process is supposed to stop, stop this thread.
1938 		 */
1939 		if (STOPLWP(p, lp)) {
1940 			lwkt_gettoken(&p->p_token);
1941 			tstop();
1942 			lwkt_reltoken(&p->p_token);
1943 		}
1944 
1945 		/*
1946 		 * Quick check without token
1947 		 */
1948 		mask = lwp_sigpend(lp);
1949 		SIGSETNAND(mask, lp->lwp_sigmask);
1950 		if (p->p_flags & P_PPWAIT)
1951 			SIG_STOPSIGMASK(mask);
1952 		if (SIGISEMPTY(mask)) 		/* no signal to send */
1953 			return (0);
1954 
1955 		/*
1956 		 * If the signal is a member of the process signal set
1957 		 * we need p_token (even if it is also a member of the
1958 		 * lwp signal set).
1959 		 */
1960 		sig = sig_ffs(&mask);
1961 		if (SIGISMEMBER(p->p_siglist, sig)) {
1962 			/*
1963 			 * Recheck with token
1964 			 */
1965 			haveptok = 1;
1966 			lwkt_gettoken(&p->p_token);
1967 
1968 			mask = lwp_sigpend(lp);
1969 			SIGSETNAND(mask, lp->lwp_sigmask);
1970 			if (p->p_flags & P_PPWAIT)
1971 				SIG_STOPSIGMASK(mask);
1972 			if (SIGISEMPTY(mask)) {		/* no signal to send */
1973 				/* haveptok is TRUE */
1974 				lwkt_reltoken(&p->p_token);
1975 				return (0);
1976 			}
1977 			sig = sig_ffs(&mask);
1978 		}
1979 
1980 		STOPEVENT(p, S_SIG, sig);
1981 
1982 		/*
1983 		 * We should see pending but ignored signals
1984 		 * only if P_TRACED was on when they were posted.
1985 		 */
1986 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1987 			spin_lock(&lp->lwp_spin);
1988 			lwp_delsig(lp, sig, haveptok);
1989 			spin_unlock(&lp->lwp_spin);
1990 			if (haveptok)
1991 				lwkt_reltoken(&p->p_token);
1992 			continue;
1993 		}
1994 		if (maytrace &&
1995 		    (p->p_flags & P_TRACED) &&
1996 		    (p->p_flags & P_PPWAIT) == 0) {
1997 			/*
1998 			 * If traced, always stop, and stay stopped until
1999 			 * released by the parent.
2000 			 *
2001 			 * NOTE: SSTOP may get cleared during the loop,
2002 			 * but we do not re-notify the parent if we have
2003 			 * to loop several times waiting for the parent
2004 			 * to let us continue.
2005 			 *
2006 			 * XXX not sure if this is still true
2007 			 */
2008 			if (haveptok == 0) {
2009 				lwkt_gettoken(&p->p_token);
2010 				haveptok = 1;
2011 			}
2012 			p->p_xstat = sig;
2013 			proc_stop(p, SSTOP);
2014 			do {
2015 				tstop();
2016 			} while (!trace_req(p) && (p->p_flags & P_TRACED));
2017 
2018 			/*
2019 			 * If parent wants us to take the signal,
2020 			 * then it will leave it in p->p_xstat;
2021 			 * otherwise we just look for signals again.
2022 			 */
2023 			spin_lock(&lp->lwp_spin);
2024 			lwp_delsig(lp, sig, 1);	/* clear old signal */
2025 			spin_unlock(&lp->lwp_spin);
2026 			sig = p->p_xstat;
2027 			if (sig == 0) {
2028 				/* haveptok is TRUE */
2029 				lwkt_reltoken(&p->p_token);
2030 				continue;
2031 			}
2032 
2033 			/*
2034 			 * Put the new signal into p_siglist.  If the
2035 			 * signal is being masked, look for other signals.
2036 			 *
2037 			 * XXX lwp might need a call to ksignal()
2038 			 */
2039 			SIGADDSET_ATOMIC(p->p_siglist, sig);
2040 			if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
2041 				/* haveptok is TRUE */
2042 				lwkt_reltoken(&p->p_token);
2043 				continue;
2044 			}
2045 
2046 			/*
2047 			 * If the traced bit got turned off, go back up
2048 			 * to the top to rescan signals.  This ensures
2049 			 * that p_sig* and ps_sigact are consistent.
2050 			 */
2051 			if ((p->p_flags & P_TRACED) == 0) {
2052 				/* haveptok is TRUE */
2053 				lwkt_reltoken(&p->p_token);
2054 				continue;
2055 			}
2056 		}
2057 
2058 		/*
2059 		 * p_token may be held here
2060 		 */
2061 		prop = sigprop(sig);
2062 
2063 		/*
2064 		 * Decide whether the signal should be returned.
2065 		 * Return the signal's number, or fall through
2066 		 * to clear it from the pending mask.
2067 		 */
2068 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2069 		case (intptr_t)SIG_DFL:
2070 			/*
2071 			 * Don't take default actions on system processes.
2072 			 */
2073 			if (p->p_pid <= 1) {
2074 #ifdef DIAGNOSTIC
2075 				/*
2076 				 * Are you sure you want to ignore SIGSEGV
2077 				 * in init? XXX
2078 				 */
2079 				kprintf("Process (pid %lu) got signal %d\n",
2080 					(u_long)p->p_pid, sig);
2081 #endif
2082 				break;		/* == ignore */
2083 			}
2084 
2085 			/*
2086 			 * Handle the in-kernel checkpoint action
2087 			 */
2088 			if (prop & SA_CKPT) {
2089 				if (haveptok == 0) {
2090 					lwkt_gettoken(&p->p_token);
2091 					haveptok = 1;
2092 				}
2093 				checkpoint_signal_handler(lp);
2094 				break;
2095 			}
2096 
2097 			/*
2098 			 * If there is a pending stop signal to process
2099 			 * with default action, stop here,
2100 			 * then clear the signal.  However,
2101 			 * if process is member of an orphaned
2102 			 * process group, ignore tty stop signals.
2103 			 */
2104 			if (prop & SA_STOP) {
2105 				if (haveptok == 0) {
2106 					lwkt_gettoken(&p->p_token);
2107 					haveptok = 1;
2108 				}
2109 				if (p->p_flags & P_TRACED ||
2110 		    		    (p->p_pgrp->pg_jobc == 0 &&
2111 				    prop & SA_TTYSTOP))
2112 					break;	/* == ignore */
2113 				if ((p->p_flags & P_WEXIT) == 0) {
2114 					p->p_xstat = sig;
2115 					proc_stop(p, SSTOP);
2116 					tstop();
2117 				}
2118 				break;
2119 			} else if (prop & SA_IGNORE) {
2120 				/*
2121 				 * Except for SIGCONT, shouldn't get here.
2122 				 * Default action is to ignore; drop it.
2123 				 */
2124 				break;		/* == ignore */
2125 			} else {
2126 				if (ptokp)
2127 					*ptokp = haveptok;
2128 				else if (haveptok)
2129 					lwkt_reltoken(&p->p_token);
2130 				return (sig);
2131 			}
2132 
2133 			/*NOTREACHED*/
2134 
2135 		case (intptr_t)SIG_IGN:
2136 			/*
2137 			 * Masking above should prevent us ever trying
2138 			 * to take action on an ignored signal other
2139 			 * than SIGCONT, unless process is traced.
2140 			 */
2141 			if ((prop & SA_CONT) == 0 &&
2142 			    (p->p_flags & P_TRACED) == 0)
2143 				kprintf("issignal\n");
2144 			break;		/* == ignore */
2145 
2146 		default:
2147 			/*
2148 			 * This signal has an action, let
2149 			 * postsig() process it.
2150 			 */
2151 			if (ptokp)
2152 				*ptokp = haveptok;
2153 			else if (haveptok)
2154 				lwkt_reltoken(&p->p_token);
2155 			return (sig);
2156 		}
2157 		spin_lock(&lp->lwp_spin);
2158 		lwp_delsig(lp, sig, haveptok);		/* take the signal! */
2159 		spin_unlock(&lp->lwp_spin);
2160 
2161 		if (haveptok)
2162 			lwkt_reltoken(&p->p_token);
2163 	}
2164 	/* NOTREACHED */
2165 }
2166 
2167 /*
2168  * Take the action for the specified signal from the current set of
2169  * pending signals.
2170  *
2171  * haveptok indicates whether the caller is holding p->p_token.  If the
2172  * caller is, we are responsible for releasing it.
2173  *
2174  * This routine can only be called from the top-level trap from usermode.
2175  * It is expecting to be able to modify the top-level stack frame.
2176  */
2177 void
2178 postsig(int sig, int haveptok)
2179 {
2180 	struct lwp *lp = curthread->td_lwp;
2181 	struct proc *p = lp->lwp_proc;
2182 	struct sigacts *ps = p->p_sigacts;
2183 	sig_t action;
2184 	sigset_t returnmask;
2185 	int code;
2186 
2187 	KASSERT(sig != 0, ("postsig"));
2188 
2189 	/*
2190 	 * If we are a virtual kernel running an emulated user process
2191 	 * context, switch back to the virtual kernel context before
2192 	 * trying to post the signal.
2193 	 */
2194 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
2195 		struct trapframe *tf = lp->lwp_md.md_regs;
2196 		tf->tf_trapno = 0;
2197 		vkernel_trap(lp, tf);
2198 	}
2199 
2200 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
2201 
2202 	spin_lock(&lp->lwp_spin);
2203 	lwp_delsig(lp, sig, haveptok);
2204 	spin_unlock(&lp->lwp_spin);
2205 	action = ps->ps_sigact[_SIG_IDX(sig)];
2206 #ifdef KTRACE
2207 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
2208 		ktrpsig(lp, sig, action, lp->lwp_flags & LWP_OLDMASK ?
2209 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
2210 #endif
2211 	/*
2212 	 * We don't need p_token after this point.
2213 	 */
2214 	if (haveptok)
2215 		lwkt_reltoken(&p->p_token);
2216 
2217 	STOPEVENT(p, S_SIG, sig);
2218 
2219 	if (action == SIG_DFL) {
2220 		/*
2221 		 * Default action, where the default is to kill
2222 		 * the process.  (Other cases were ignored above.)
2223 		 */
2224 		sigexit(lp, sig);
2225 		/* NOTREACHED */
2226 	} else {
2227 		/*
2228 		 * If we get here, the signal must be caught.
2229 		 */
2230 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
2231 		    ("postsig action"));
2232 
2233 		/*
2234 		 * Reset the signal handler if asked to
2235 		 */
2236 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
2237 			/*
2238 			 * See kern_sigaction() for origin of this code.
2239 			 */
2240 			SIGDELSET(p->p_sigcatch, sig);
2241 			if (sig != SIGCONT &&
2242 			    sigprop(sig) & SA_IGNORE)
2243 				SIGADDSET(p->p_sigignore, sig);
2244 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2245 		}
2246 
2247 		/*
2248 		 * Set the signal mask and calculate the mask to restore
2249 		 * when the signal function returns.
2250 		 *
2251 		 * Special case: user has done a sigsuspend.  Here the
2252 		 * current mask is not of interest, but rather the
2253 		 * mask from before the sigsuspend is what we want
2254 		 * restored after the signal processing is completed.
2255 		 */
2256 		if (lp->lwp_flags & LWP_OLDMASK) {
2257 			returnmask = lp->lwp_oldsigmask;
2258 			lp->lwp_flags &= ~LWP_OLDMASK;
2259 		} else {
2260 			returnmask = lp->lwp_sigmask;
2261 		}
2262 
2263 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
2264 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
2265 			SIGADDSET(lp->lwp_sigmask, sig);
2266 
2267 		lp->lwp_ru.ru_nsignals++;
2268 		if (lp->lwp_sig != sig) {
2269 			code = 0;
2270 		} else {
2271 			code = lp->lwp_code;
2272 			lp->lwp_code = 0;
2273 			lp->lwp_sig = 0;
2274 		}
2275 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
2276 	}
2277 }
2278 
2279 /*
2280  * Kill the current process for stated reason.
2281  */
2282 void
2283 killproc(struct proc *p, char *why)
2284 {
2285 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
2286 		p->p_pid, p->p_comm,
2287 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2288 	ksignal(p, SIGKILL);
2289 }
2290 
2291 /*
2292  * Force the current process to exit with the specified signal, dumping core
2293  * if appropriate.  We bypass the normal tests for masked and caught signals,
2294  * allowing unrecoverable failures to terminate the process without changing
2295  * signal state.  Mark the accounting record with the signal termination.
2296  * If dumping core, save the signal number for the debugger.  Calls exit and
2297  * does not return.
2298  *
2299  * This routine does not return.
2300  */
2301 void
2302 sigexit(struct lwp *lp, int sig)
2303 {
2304 	struct proc *p = lp->lwp_proc;
2305 
2306 	lwkt_gettoken(&p->p_token);
2307 	p->p_acflag |= AXSIG;
2308 	if (sigprop(sig) & SA_CORE) {
2309 		lp->lwp_sig = sig;
2310 
2311 		/*
2312 		 * All threads must be stopped before we can safely coredump.
2313 		 * Stop threads using SCORE, which cannot be overridden.
2314 		 */
2315 		if (p->p_stat != SCORE) {
2316 			proc_stop(p, SCORE);
2317 			proc_stopwait(p);
2318 
2319 			if (coredump(lp, sig) == 0)
2320 				sig |= WCOREFLAG;
2321 			p->p_stat = SSTOP;
2322 		}
2323 
2324 		/*
2325 		 * Log signals which would cause core dumps
2326 		 * (Log as LOG_INFO to appease those who don't want
2327 		 * these messages.)
2328 		 * XXX : Todo, as well as euid, write out ruid too
2329 		 */
2330 		if (kern_logsigexit) {
2331 			log(LOG_INFO,
2332 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2333 			    p->p_pid, p->p_comm,
2334 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
2335 			    sig &~ WCOREFLAG,
2336 			    sig & WCOREFLAG ? " (core dumped)" : "");
2337 		}
2338 	}
2339 	lwkt_reltoken(&p->p_token);
2340 	exit1(W_EXITCODE(0, sig));
2341 	/* NOTREACHED */
2342 }
2343 
2344 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
2345 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2346 	      sizeof(corefilename), "process corefile name format string");
2347 
2348 /*
2349  * expand_name(name, uid, pid)
2350  * Expand the name described in corefilename, using name, uid, and pid.
2351  * corefilename is a kprintf-like string, with three format specifiers:
2352  *	%N	name of process ("name")
2353  *	%P	process id (pid)
2354  *	%U	user id (uid)
2355  * For example, "%N.core" is the default; they can be disabled completely
2356  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2357  * This is controlled by the sysctl variable kern.corefile (see above).
2358  */
2359 
2360 static char *
2361 expand_name(const char *name, uid_t uid, pid_t pid)
2362 {
2363 	char *temp;
2364 	char buf[11];		/* Buffer for pid/uid -- max 4B */
2365 	int i, n;
2366 	char *format = corefilename;
2367 	size_t namelen;
2368 
2369 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
2370 	if (temp == NULL)
2371 		return NULL;
2372 	namelen = strlen(name);
2373 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2374 		int l;
2375 		switch (format[i]) {
2376 		case '%':	/* Format character */
2377 			i++;
2378 			switch (format[i]) {
2379 			case '%':
2380 				temp[n++] = '%';
2381 				break;
2382 			case 'N':	/* process name */
2383 				if ((n + namelen) > MAXPATHLEN) {
2384 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2385 					    pid, name, uid, temp, name);
2386 					kfree(temp, M_TEMP);
2387 					return NULL;
2388 				}
2389 				memcpy(temp+n, name, namelen);
2390 				n += namelen;
2391 				break;
2392 			case 'P':	/* process id */
2393 				l = ksprintf(buf, "%u", pid);
2394 				if ((n + l) > MAXPATHLEN) {
2395 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2396 					    pid, name, uid, temp, name);
2397 					kfree(temp, M_TEMP);
2398 					return NULL;
2399 				}
2400 				memcpy(temp+n, buf, l);
2401 				n += l;
2402 				break;
2403 			case 'U':	/* user id */
2404 				l = ksprintf(buf, "%u", uid);
2405 				if ((n + l) > MAXPATHLEN) {
2406 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2407 					    pid, name, uid, temp, name);
2408 					kfree(temp, M_TEMP);
2409 					return NULL;
2410 				}
2411 				memcpy(temp+n, buf, l);
2412 				n += l;
2413 				break;
2414 			default:
2415 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2416 			}
2417 			break;
2418 		default:
2419 			temp[n++] = format[i];
2420 		}
2421 	}
2422 	temp[n] = '\0';
2423 	return temp;
2424 }
2425 
2426 /*
2427  * Dump a process' core.  The main routine does some
2428  * policy checking, and creates the name of the coredump;
2429  * then it passes on a vnode and a size limit to the process-specific
2430  * coredump routine if there is one; if there _is not_ one, it returns
2431  * ENOSYS; otherwise it returns the error from the process-specific routine.
2432  *
2433  * The parameter `lp' is the lwp which triggered the coredump.
2434  */
2435 
2436 static int
2437 coredump(struct lwp *lp, int sig)
2438 {
2439 	struct proc *p = lp->lwp_proc;
2440 	struct vnode *vp;
2441 	struct ucred *cred = p->p_ucred;
2442 	struct flock lf;
2443 	struct nlookupdata nd;
2444 	struct vattr vattr;
2445 	int error, error1;
2446 	char *name;			/* name of corefile */
2447 	off_t limit;
2448 
2449 	STOPEVENT(p, S_CORE, 0);
2450 
2451 	if (((sugid_coredump == 0) && p->p_flags & P_SUGID) || do_coredump == 0)
2452 		return (EFAULT);
2453 
2454 	/*
2455 	 * Note that the bulk of limit checking is done after
2456 	 * the corefile is created.  The exception is if the limit
2457 	 * for corefiles is 0, in which case we don't bother
2458 	 * creating the corefile at all.  This layout means that
2459 	 * a corefile is truncated instead of not being created,
2460 	 * if it is larger than the limit.
2461 	 */
2462 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2463 	if (limit == 0)
2464 		return EFBIG;
2465 
2466 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2467 	if (name == NULL)
2468 		return (EINVAL);
2469 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2470 	if (error == 0)
2471 		error = vn_open(&nd, NULL,
2472 				O_CREAT | FWRITE | O_NOFOLLOW,
2473 				S_IRUSR | S_IWUSR);
2474 	kfree(name, M_TEMP);
2475 	if (error) {
2476 		nlookup_done(&nd);
2477 		return (error);
2478 	}
2479 	vp = nd.nl_open_vp;
2480 	nd.nl_open_vp = NULL;
2481 	nlookup_done(&nd);
2482 
2483 	vn_unlock(vp);
2484 	lf.l_whence = SEEK_SET;
2485 	lf.l_start = 0;
2486 	lf.l_len = 0;
2487 	lf.l_type = F_WRLCK;
2488 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2489 	if (error)
2490 		goto out2;
2491 
2492 	/* Don't dump to non-regular files or files with links. */
2493 	if (vp->v_type != VREG ||
2494 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2495 		error = EFAULT;
2496 		goto out1;
2497 	}
2498 
2499 	/* Don't dump to files current user does not own */
2500 	if (vattr.va_uid != p->p_ucred->cr_uid) {
2501 		error = EFAULT;
2502 		goto out1;
2503 	}
2504 
2505 	VATTR_NULL(&vattr);
2506 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2507 	vattr.va_size = 0;
2508 	VOP_SETATTR(vp, &vattr, cred);
2509 	p->p_acflag |= ACORE;
2510 	vn_unlock(vp);
2511 
2512 	error = p->p_sysent->sv_coredump ?
2513 		  p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2514 
2515 out1:
2516 	lf.l_type = F_UNLCK;
2517 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2518 out2:
2519 	error1 = vn_close(vp, FWRITE, NULL);
2520 	if (error == 0)
2521 		error = error1;
2522 	return (error);
2523 }
2524 
2525 /*
2526  * Nonexistent system call-- signal process (may want to handle it).
2527  * Flag error in case process won't see signal immediately (blocked or ignored).
2528  *
2529  * MPALMOSTSAFE
2530  */
2531 /* ARGSUSED */
2532 int
2533 sys_nosys(struct nosys_args *args)
2534 {
2535 	lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2536 	return (EINVAL);
2537 }
2538 
2539 /*
2540  * Send a SIGIO or SIGURG signal to a process or process group using
2541  * stored credentials rather than those of the current process.
2542  */
2543 void
2544 pgsigio(struct sigio *sigio, int sig, int checkctty)
2545 {
2546 	if (sigio == NULL)
2547 		return;
2548 
2549 	if (sigio->sio_pgid > 0) {
2550 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2551 		             sigio->sio_proc))
2552 			ksignal(sigio->sio_proc, sig);
2553 	} else if (sigio->sio_pgid < 0) {
2554 		struct proc *p;
2555 		struct pgrp *pg = sigio->sio_pgrp;
2556 
2557 		/*
2558 		 * Must interlock all signals against fork
2559 		 */
2560 		pgref(pg);
2561 		lockmgr(&pg->pg_lock, LK_EXCLUSIVE);
2562 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
2563 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2564 			    (checkctty == 0 || (p->p_flags & P_CONTROLT)))
2565 				ksignal(p, sig);
2566 		}
2567 		lockmgr(&pg->pg_lock, LK_RELEASE);
2568 		pgrel(pg);
2569 	}
2570 }
2571 
2572 static int
2573 filt_sigattach(struct knote *kn)
2574 {
2575 	struct proc *p = curproc;
2576 
2577 	kn->kn_ptr.p_proc = p;
2578 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
2579 
2580 	/* XXX lock the proc here while adding to the list? */
2581 	knote_insert(&p->p_klist, kn);
2582 
2583 	return (0);
2584 }
2585 
2586 static void
2587 filt_sigdetach(struct knote *kn)
2588 {
2589 	struct proc *p = kn->kn_ptr.p_proc;
2590 
2591 	knote_remove(&p->p_klist, kn);
2592 }
2593 
2594 /*
2595  * signal knotes are shared with proc knotes, so we apply a mask to
2596  * the hint in order to differentiate them from process hints.  This
2597  * could be avoided by using a signal-specific knote list, but probably
2598  * isn't worth the trouble.
2599  */
2600 static int
2601 filt_signal(struct knote *kn, long hint)
2602 {
2603 	if (hint & NOTE_SIGNAL) {
2604 		hint &= ~NOTE_SIGNAL;
2605 
2606 		if (kn->kn_id == hint)
2607 			kn->kn_data++;
2608 	}
2609 	return (kn->kn_data != 0);
2610 }
2611