xref: /dragonfly/sys/kern/kern_sig.c (revision 678e8cc6)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
39  * $FreeBSD: src/sys/kern/kern_sig.c,v 1.72.2.17 2003/05/16 16:34:34 obrien Exp $
40  */
41 
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/sysproto.h>
48 #include <sys/signalvar.h>
49 #include <sys/resourcevar.h>
50 #include <sys/vnode.h>
51 #include <sys/event.h>
52 #include <sys/proc.h>
53 #include <sys/nlookup.h>
54 #include <sys/pioctl.h>
55 #include <sys/acct.h>
56 #include <sys/fcntl.h>
57 #include <sys/lock.h>
58 #include <sys/wait.h>
59 #include <sys/ktrace.h>
60 #include <sys/syslog.h>
61 #include <sys/stat.h>
62 #include <sys/sysent.h>
63 #include <sys/sysctl.h>
64 #include <sys/malloc.h>
65 #include <sys/interrupt.h>
66 #include <sys/unistd.h>
67 #include <sys/kern_syscall.h>
68 #include <sys/vkernel.h>
69 
70 #include <sys/signal2.h>
71 #include <sys/thread2.h>
72 #include <sys/spinlock2.h>
73 
74 #include <machine/cpu.h>
75 #include <machine/smp.h>
76 
77 static int	coredump(struct lwp *, int);
78 static char	*expand_name(const char *, uid_t, pid_t);
79 static int	dokillpg(int sig, int pgid, int all);
80 static int	sig_ffs(sigset_t *set);
81 static int	sigprop(int sig);
82 static void	lwp_signotify(struct lwp *lp);
83 #ifdef SMP
84 static void	lwp_signotify_remote(void *arg);
85 #endif
86 static int	kern_sigtimedwait(sigset_t set, siginfo_t *info,
87 		    struct timespec *timeout);
88 
89 static int	filt_sigattach(struct knote *kn);
90 static void	filt_sigdetach(struct knote *kn);
91 static int	filt_signal(struct knote *kn, long hint);
92 
93 struct filterops sig_filtops =
94 	{ 0, filt_sigattach, filt_sigdetach, filt_signal };
95 
96 static int	kern_logsigexit = 1;
97 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
98     &kern_logsigexit, 0,
99     "Log processes quitting on abnormal signals to syslog(3)");
100 
101 /*
102  * Can process p, with pcred pc, send the signal sig to process q?
103  */
104 #define CANSIGNAL(q, sig) \
105 	(!p_trespass(curproc->p_ucred, (q)->p_ucred) || \
106 	((sig) == SIGCONT && (q)->p_session == curproc->p_session))
107 
108 /*
109  * Policy -- Can real uid ruid with ucred uc send a signal to process q?
110  */
111 #define CANSIGIO(ruid, uc, q) \
112 	((uc)->cr_uid == 0 || \
113 	    (ruid) == (q)->p_ucred->cr_ruid || \
114 	    (uc)->cr_uid == (q)->p_ucred->cr_ruid || \
115 	    (ruid) == (q)->p_ucred->cr_uid || \
116 	    (uc)->cr_uid == (q)->p_ucred->cr_uid)
117 
118 int sugid_coredump;
119 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
120 	&sugid_coredump, 0, "Enable coredumping set user/group ID processes");
121 
122 static int	do_coredump = 1;
123 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
124 	&do_coredump, 0, "Enable/Disable coredumps");
125 
126 /*
127  * Signal properties and actions.
128  * The array below categorizes the signals and their default actions
129  * according to the following properties:
130  */
131 #define	SA_KILL		0x01		/* terminates process by default */
132 #define	SA_CORE		0x02		/* ditto and coredumps */
133 #define	SA_STOP		0x04		/* suspend process */
134 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
135 #define	SA_IGNORE	0x10		/* ignore by default */
136 #define	SA_CONT		0x20		/* continue if suspended */
137 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
138 #define SA_CKPT         0x80            /* checkpoint process */
139 
140 
141 static int sigproptbl[NSIG] = {
142         SA_KILL,                /* SIGHUP */
143         SA_KILL,                /* SIGINT */
144         SA_KILL|SA_CORE,        /* SIGQUIT */
145         SA_KILL|SA_CORE,        /* SIGILL */
146         SA_KILL|SA_CORE,        /* SIGTRAP */
147         SA_KILL|SA_CORE,        /* SIGABRT */
148         SA_KILL|SA_CORE,        /* SIGEMT */
149         SA_KILL|SA_CORE,        /* SIGFPE */
150         SA_KILL,                /* SIGKILL */
151         SA_KILL|SA_CORE,        /* SIGBUS */
152         SA_KILL|SA_CORE,        /* SIGSEGV */
153         SA_KILL|SA_CORE,        /* SIGSYS */
154         SA_KILL,                /* SIGPIPE */
155         SA_KILL,                /* SIGALRM */
156         SA_KILL,                /* SIGTERM */
157         SA_IGNORE,              /* SIGURG */
158         SA_STOP,                /* SIGSTOP */
159         SA_STOP|SA_TTYSTOP,     /* SIGTSTP */
160         SA_IGNORE|SA_CONT,      /* SIGCONT */
161         SA_IGNORE,              /* SIGCHLD */
162         SA_STOP|SA_TTYSTOP,     /* SIGTTIN */
163         SA_STOP|SA_TTYSTOP,     /* SIGTTOU */
164         SA_IGNORE,              /* SIGIO */
165         SA_KILL,                /* SIGXCPU */
166         SA_KILL,                /* SIGXFSZ */
167         SA_KILL,                /* SIGVTALRM */
168         SA_KILL,                /* SIGPROF */
169         SA_IGNORE,              /* SIGWINCH  */
170         SA_IGNORE,              /* SIGINFO */
171         SA_KILL,                /* SIGUSR1 */
172         SA_KILL,                /* SIGUSR2 */
173 	SA_IGNORE,              /* SIGTHR */
174 	SA_CKPT,                /* SIGCKPT */
175 	SA_KILL|SA_CKPT,        /* SIGCKPTEXIT */
176 	SA_IGNORE,
177 	SA_IGNORE,
178 	SA_IGNORE,
179 	SA_IGNORE,
180 	SA_IGNORE,
181 	SA_IGNORE,
182 	SA_IGNORE,
183 	SA_IGNORE,
184 	SA_IGNORE,
185 	SA_IGNORE,
186 	SA_IGNORE,
187 	SA_IGNORE,
188 	SA_IGNORE,
189 	SA_IGNORE,
190 	SA_IGNORE,
191 	SA_IGNORE,
192 	SA_IGNORE,
193 	SA_IGNORE,
194 	SA_IGNORE,
195 	SA_IGNORE,
196 	SA_IGNORE,
197 	SA_IGNORE,
198 	SA_IGNORE,
199 	SA_IGNORE,
200 	SA_IGNORE,
201 	SA_IGNORE,
202 	SA_IGNORE,
203 	SA_IGNORE,
204 	SA_IGNORE,
205 	SA_IGNORE,
206 
207 };
208 
209 static __inline int
210 sigprop(int sig)
211 {
212 
213 	if (sig > 0 && sig < NSIG)
214 		return (sigproptbl[_SIG_IDX(sig)]);
215 	return (0);
216 }
217 
218 static __inline int
219 sig_ffs(sigset_t *set)
220 {
221 	int i;
222 
223 	for (i = 0; i < _SIG_WORDS; i++)
224 		if (set->__bits[i])
225 			return (ffs(set->__bits[i]) + (i * 32));
226 	return (0);
227 }
228 
229 /*
230  * No requirements.
231  */
232 int
233 kern_sigaction(int sig, struct sigaction *act, struct sigaction *oact)
234 {
235 	struct thread *td = curthread;
236 	struct proc *p = td->td_proc;
237 	struct lwp *lp;
238 	struct sigacts *ps = p->p_sigacts;
239 
240 	if (sig <= 0 || sig > _SIG_MAXSIG)
241 		return (EINVAL);
242 
243 	lwkt_gettoken(&p->p_token);
244 
245 	if (oact) {
246 		oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
247 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
248 		oact->sa_flags = 0;
249 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
250 			oact->sa_flags |= SA_ONSTACK;
251 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
252 			oact->sa_flags |= SA_RESTART;
253 		if (SIGISMEMBER(ps->ps_sigreset, sig))
254 			oact->sa_flags |= SA_RESETHAND;
255 		if (SIGISMEMBER(ps->ps_signodefer, sig))
256 			oact->sa_flags |= SA_NODEFER;
257 		if (SIGISMEMBER(ps->ps_siginfo, sig))
258 			oact->sa_flags |= SA_SIGINFO;
259 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDSTOP)
260 			oact->sa_flags |= SA_NOCLDSTOP;
261 		if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDWAIT)
262 			oact->sa_flags |= SA_NOCLDWAIT;
263 	}
264 	if (act) {
265 		/*
266 		 * Check for invalid requests.  KILL and STOP cannot be
267 		 * caught.
268 		 */
269 		if (sig == SIGKILL || sig == SIGSTOP) {
270 			if (act->sa_handler != SIG_DFL) {
271 				lwkt_reltoken(&p->p_token);
272 				return (EINVAL);
273 			}
274 		}
275 
276 		/*
277 		 * Change setting atomically.
278 		 */
279 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
280 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
281 		if (act->sa_flags & SA_SIGINFO) {
282 			ps->ps_sigact[_SIG_IDX(sig)] =
283 			    (__sighandler_t *)act->sa_sigaction;
284 			SIGADDSET(ps->ps_siginfo, sig);
285 		} else {
286 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
287 			SIGDELSET(ps->ps_siginfo, sig);
288 		}
289 		if (!(act->sa_flags & SA_RESTART))
290 			SIGADDSET(ps->ps_sigintr, sig);
291 		else
292 			SIGDELSET(ps->ps_sigintr, sig);
293 		if (act->sa_flags & SA_ONSTACK)
294 			SIGADDSET(ps->ps_sigonstack, sig);
295 		else
296 			SIGDELSET(ps->ps_sigonstack, sig);
297 		if (act->sa_flags & SA_RESETHAND)
298 			SIGADDSET(ps->ps_sigreset, sig);
299 		else
300 			SIGDELSET(ps->ps_sigreset, sig);
301 		if (act->sa_flags & SA_NODEFER)
302 			SIGADDSET(ps->ps_signodefer, sig);
303 		else
304 			SIGDELSET(ps->ps_signodefer, sig);
305 		if (sig == SIGCHLD) {
306 			if (act->sa_flags & SA_NOCLDSTOP)
307 				p->p_sigacts->ps_flag |= PS_NOCLDSTOP;
308 			else
309 				p->p_sigacts->ps_flag &= ~PS_NOCLDSTOP;
310 			if (act->sa_flags & SA_NOCLDWAIT) {
311 				/*
312 				 * Paranoia: since SA_NOCLDWAIT is implemented
313 				 * by reparenting the dying child to PID 1 (and
314 				 * trust it to reap the zombie), PID 1 itself
315 				 * is forbidden to set SA_NOCLDWAIT.
316 				 */
317 				if (p->p_pid == 1)
318 					p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
319 				else
320 					p->p_sigacts->ps_flag |= PS_NOCLDWAIT;
321 			} else {
322 				p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
323 			}
324 		}
325 		/*
326 		 * Set bit in p_sigignore for signals that are set to SIG_IGN,
327 		 * and for signals set to SIG_DFL where the default is to
328 		 * ignore. However, don't put SIGCONT in p_sigignore, as we
329 		 * have to restart the process.
330 		 *
331 		 * Also remove the signal from the process and lwp signal
332 		 * list.
333 		 */
334 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
335 		    (sigprop(sig) & SA_IGNORE &&
336 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
337 			SIGDELSET(p->p_siglist, sig);
338 			FOREACH_LWP_IN_PROC(lp, p) {
339 				spin_lock(&lp->lwp_spin);
340 				SIGDELSET(lp->lwp_siglist, sig);
341 				spin_unlock(&lp->lwp_spin);
342 			}
343 			if (sig != SIGCONT) {
344 				/* easier in ksignal */
345 				SIGADDSET(p->p_sigignore, sig);
346 			}
347 			SIGDELSET(p->p_sigcatch, sig);
348 		} else {
349 			SIGDELSET(p->p_sigignore, sig);
350 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
351 				SIGDELSET(p->p_sigcatch, sig);
352 			else
353 				SIGADDSET(p->p_sigcatch, sig);
354 		}
355 	}
356 	lwkt_reltoken(&p->p_token);
357 	return (0);
358 }
359 
360 int
361 sys_sigaction(struct sigaction_args *uap)
362 {
363 	struct sigaction act, oact;
364 	struct sigaction *actp, *oactp;
365 	int error;
366 
367 	actp = (uap->act != NULL) ? &act : NULL;
368 	oactp = (uap->oact != NULL) ? &oact : NULL;
369 	if (actp) {
370 		error = copyin(uap->act, actp, sizeof(act));
371 		if (error)
372 			return (error);
373 	}
374 	error = kern_sigaction(uap->sig, actp, oactp);
375 	if (oactp && !error) {
376 		error = copyout(oactp, uap->oact, sizeof(oact));
377 	}
378 	return (error);
379 }
380 
381 /*
382  * Initialize signal state for process 0;
383  * set to ignore signals that are ignored by default.
384  */
385 void
386 siginit(struct proc *p)
387 {
388 	int i;
389 
390 	for (i = 1; i <= NSIG; i++)
391 		if (sigprop(i) & SA_IGNORE && i != SIGCONT)
392 			SIGADDSET(p->p_sigignore, i);
393 }
394 
395 /*
396  * Reset signals for an exec of the specified process.
397  */
398 void
399 execsigs(struct proc *p)
400 {
401 	struct sigacts *ps = p->p_sigacts;
402 	struct lwp *lp;
403 	int sig;
404 
405 	lp = ONLY_LWP_IN_PROC(p);
406 
407 	/*
408 	 * Reset caught signals.  Held signals remain held
409 	 * through p_sigmask (unless they were caught,
410 	 * and are now ignored by default).
411 	 */
412 	while (SIGNOTEMPTY(p->p_sigcatch)) {
413 		sig = sig_ffs(&p->p_sigcatch);
414 		SIGDELSET(p->p_sigcatch, sig);
415 		if (sigprop(sig) & SA_IGNORE) {
416 			if (sig != SIGCONT)
417 				SIGADDSET(p->p_sigignore, sig);
418 			SIGDELSET(p->p_siglist, sig);
419 			/* don't need spinlock */
420 			SIGDELSET(lp->lwp_siglist, sig);
421 		}
422 		ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
423 	}
424 
425 	/*
426 	 * Reset stack state to the user stack.
427 	 * Clear set of signals caught on the signal stack.
428 	 */
429 	lp->lwp_sigstk.ss_flags = SS_DISABLE;
430 	lp->lwp_sigstk.ss_size = 0;
431 	lp->lwp_sigstk.ss_sp = 0;
432 	lp->lwp_flags &= ~LWP_ALTSTACK;
433 	/*
434 	 * Reset no zombies if child dies flag as Solaris does.
435 	 */
436 	p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
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 			if (lp->lwp_sigstk.ss_flags & SS_ONSTACK)
610 				return (EINVAL);
611 			lp->lwp_flags &= ~LWP_ALTSTACK;
612 			lp->lwp_sigstk.ss_flags = ss->ss_flags;
613 		} else {
614 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
615 				return (ENOMEM);
616 			lp->lwp_flags |= LWP_ALTSTACK;
617 			lp->lwp_sigstk = *ss;
618 		}
619 	}
620 
621 	return (0);
622 }
623 
624 /*
625  * MPSAFE
626  */
627 int
628 sys_sigaltstack(struct sigaltstack_args *uap)
629 {
630 	stack_t ss, oss;
631 	int error;
632 
633 	if (uap->ss) {
634 		error = copyin(uap->ss, &ss, sizeof(ss));
635 		if (error)
636 			return (error);
637 	}
638 
639 	error = kern_sigaltstack(uap->ss ? &ss : NULL,
640 	    uap->oss ? &oss : NULL);
641 
642 	if (error == 0 && uap->oss)
643 		error = copyout(&oss, uap->oss, sizeof(*uap->oss));
644 	return (error);
645 }
646 
647 /*
648  * Common code for kill process group/broadcast kill.
649  * cp is calling process.
650  */
651 struct killpg_info {
652 	int nfound;
653 	int sig;
654 };
655 
656 static int killpg_all_callback(struct proc *p, void *data);
657 
658 static int
659 dokillpg(int sig, int pgid, int all)
660 {
661 	struct killpg_info info;
662 	struct proc *cp = curproc;
663 	struct proc *p;
664 	struct pgrp *pgrp;
665 
666 	info.nfound = 0;
667 	info.sig = sig;
668 
669 	if (all) {
670 		/*
671 		 * broadcast
672 		 */
673 		allproc_scan(killpg_all_callback, &info);
674 	} else {
675 		if (pgid == 0) {
676 			/*
677 			 * zero pgid means send to my process group.
678 			 */
679 			pgrp = cp->p_pgrp;
680 			pgref(pgrp);
681 		} else {
682 			pgrp = pgfind(pgid);
683 			if (pgrp == NULL)
684 				return (ESRCH);
685 		}
686 
687 		/*
688 		 * Must interlock all signals against fork
689 		 */
690 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
691 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
692 			if (p->p_pid <= 1 ||
693 			    p->p_stat == SZOMB ||
694 			    (p->p_flags & P_SYSTEM) ||
695 			    !CANSIGNAL(p, sig)) {
696 				continue;
697 			}
698 			++info.nfound;
699 			if (sig)
700 				ksignal(p, sig);
701 		}
702 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
703 		pgrel(pgrp);
704 	}
705 	return (info.nfound ? 0 : ESRCH);
706 }
707 
708 static int
709 killpg_all_callback(struct proc *p, void *data)
710 {
711 	struct killpg_info *info = data;
712 
713 	if (p->p_pid <= 1 || (p->p_flags & P_SYSTEM) ||
714 	    p == curproc || !CANSIGNAL(p, info->sig)) {
715 		return (0);
716 	}
717 	++info->nfound;
718 	if (info->sig)
719 		ksignal(p, info->sig);
720 	return(0);
721 }
722 
723 /*
724  * Send a general signal to a process or LWPs within that process.
725  *
726  * Note that new signals cannot be sent if a process is exiting or already
727  * a zombie, but we return success anyway as userland is likely to not handle
728  * the race properly.
729  *
730  * No requirements.
731  */
732 int
733 kern_kill(int sig, pid_t pid, lwpid_t tid)
734 {
735 	int t;
736 
737 	if ((u_int)sig > _SIG_MAXSIG)
738 		return (EINVAL);
739 
740 	lwkt_gettoken(&proc_token);
741 
742 	if (pid > 0) {
743 		struct proc *p;
744 		struct lwp *lp = NULL;
745 
746 		/*
747 		 * Send a signal to a single process.  If the kill() is
748 		 * racing an exiting process which has not yet been reaped
749 		 * act as though the signal was delivered successfully but
750 		 * don't actually try to deliver the signal.
751 		 */
752 		if ((p = pfind(pid)) == NULL) {
753 			if ((p = zpfind(pid)) == NULL) {
754 				lwkt_reltoken(&proc_token);
755 				return (ESRCH);
756 			}
757 			lwkt_reltoken(&proc_token);
758 			PRELE(p);
759 			return (0);
760 		}
761 		lwkt_gettoken(&p->p_token);
762 		if (!CANSIGNAL(p, sig)) {
763 			lwkt_reltoken(&p->p_token);
764 			PRELE(p);
765 			lwkt_reltoken(&proc_token);
766 			return (EPERM);
767 		}
768 
769 		/*
770 		 * NOP if the process is exiting.  Note that lwpsignal() is
771 		 * called directly with P_WEXIT set to kill individual LWPs
772 		 * during exit, which is allowed.
773 		 */
774 		if (p->p_flags & P_WEXIT) {
775 			lwkt_reltoken(&p->p_token);
776 			PRELE(p);
777 			lwkt_reltoken(&proc_token);
778 			return (0);
779 		}
780 		if (tid != -1) {
781 			lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
782 			if (lp == NULL) {
783 				lwkt_reltoken(&p->p_token);
784 				PRELE(p);
785 				lwkt_reltoken(&proc_token);
786 				return (ESRCH);
787 			}
788 		}
789 		if (sig)
790 			lwpsignal(p, lp, sig);
791 		lwkt_reltoken(&p->p_token);
792 		PRELE(p);
793 		lwkt_reltoken(&proc_token);
794 		return (0);
795 	}
796 
797 	/*
798 	 * If we come here, pid is a special broadcast pid.
799 	 * This doesn't mix with a tid.
800 	 */
801 	if (tid != -1) {
802 		lwkt_reltoken(&proc_token);
803 		return (EINVAL);
804 	}
805 	switch (pid) {
806 	case -1:		/* broadcast signal */
807 		t = (dokillpg(sig, 0, 1));
808 		break;
809 	case 0:			/* signal own process group */
810 		t = (dokillpg(sig, 0, 0));
811 		break;
812 	default:		/* negative explicit process group */
813 		t = (dokillpg(sig, -pid, 0));
814 		break;
815 	}
816 	lwkt_reltoken(&proc_token);
817 	return t;
818 }
819 
820 int
821 sys_kill(struct kill_args *uap)
822 {
823 	int error;
824 
825 	error = kern_kill(uap->signum, uap->pid, -1);
826 	return (error);
827 }
828 
829 int
830 sys_lwp_kill(struct lwp_kill_args *uap)
831 {
832 	int error;
833 	pid_t pid = uap->pid;
834 
835 	/*
836 	 * A tid is mandatory for lwp_kill(), otherwise
837 	 * you could simply use kill().
838 	 */
839 	if (uap->tid == -1)
840 		return (EINVAL);
841 
842 	/*
843 	 * To save on a getpid() function call for intra-process
844 	 * signals, pid == -1 means current process.
845 	 */
846 	if (pid == -1)
847 		pid = curproc->p_pid;
848 
849 	error = kern_kill(uap->signum, pid, uap->tid);
850 	return (error);
851 }
852 
853 /*
854  * Send a signal to a process group.
855  */
856 void
857 gsignal(int pgid, int sig)
858 {
859 	struct pgrp *pgrp;
860 
861 	if (pgid && (pgrp = pgfind(pgid)))
862 		pgsignal(pgrp, sig, 0);
863 }
864 
865 /*
866  * Send a signal to a process group.  If checktty is 1,
867  * limit to members which have a controlling terminal.
868  *
869  * pg_lock interlocks against a fork that might be in progress, to
870  * ensure that the new child process picks up the signal.
871  */
872 void
873 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
874 {
875 	struct proc *p;
876 
877 	/*
878 	 * Must interlock all signals against fork
879 	 */
880 	if (pgrp) {
881 		pgref(pgrp);
882 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
883 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
884 			if (checkctty == 0 || p->p_flags & P_CONTROLT)
885 				ksignal(p, sig);
886 		}
887 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
888 		pgrel(pgrp);
889 	}
890 }
891 
892 /*
893  * Send a signal caused by a trap to the current lwp.  If it will be caught
894  * immediately, deliver it with correct code.  Otherwise, post it normally.
895  *
896  * These signals may ONLY be delivered to the specified lwp and may never
897  * be delivered to the process generically.
898  */
899 void
900 trapsignal(struct lwp *lp, int sig, u_long code)
901 {
902 	struct proc *p = lp->lwp_proc;
903 	struct sigacts *ps = p->p_sigacts;
904 
905 	/*
906 	 * If we are a virtual kernel running an emulated user process
907 	 * context, switch back to the virtual kernel context before
908 	 * trying to post the signal.
909 	 */
910 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
911 		struct trapframe *tf = lp->lwp_md.md_regs;
912 		tf->tf_trapno = 0;
913 		vkernel_trap(lp, tf);
914 	}
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 	lwkt_gettoken(&p->p_token);
1114 	if (lp) {
1115 		LWPHOLD(lp);
1116 		lwkt_gettoken(&lp->lwp_token);
1117 	} else if (p->p_flags & P_WEXIT) {
1118 		goto out;
1119 	}
1120 
1121 	prop = sigprop(sig);
1122 
1123 	/*
1124 	 * If proc is traced, always give parent a chance;
1125 	 * if signal event is tracked by procfs, give *that*
1126 	 * a chance, as well.
1127 	 */
1128 	if ((p->p_flags & P_TRACED) || (p->p_stops & S_SIG)) {
1129 		action = SIG_DFL;
1130 	} else {
1131 		/*
1132 		 * Do not try to deliver signals to an exiting lwp.  Note
1133 		 * that we must still deliver the signal if P_WEXIT is set
1134 		 * in the process flags.
1135 		 */
1136 		if (lp && (lp->lwp_mpflags & LWP_MP_WEXIT)) {
1137 			if (lp) {
1138 				lwkt_reltoken(&lp->lwp_token);
1139 				LWPRELE(lp);
1140 			}
1141 			lwkt_reltoken(&p->p_token);
1142 			PRELE(p);
1143 			return;
1144 		}
1145 
1146 		/*
1147 		 * If the signal is being ignored, then we forget about
1148 		 * it immediately.  NOTE: We don't set SIGCONT in p_sigignore,
1149 		 * and if it is set to SIG_IGN, action will be SIG_DFL here.
1150 		 */
1151 		if (SIGISMEMBER(p->p_sigignore, sig)) {
1152 			/*
1153 			 * Even if a signal is set SIG_IGN, it may still be
1154 			 * lurking in a kqueue.
1155 			 */
1156 			KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1157 			if (lp) {
1158 				lwkt_reltoken(&lp->lwp_token);
1159 				LWPRELE(lp);
1160 			}
1161 			lwkt_reltoken(&p->p_token);
1162 			PRELE(p);
1163 			return;
1164 		}
1165 		if (SIGISMEMBER(p->p_sigcatch, sig))
1166 			action = SIG_CATCH;
1167 		else
1168 			action = SIG_DFL;
1169 	}
1170 
1171 	/*
1172 	 * If continuing, clear any pending STOP signals.
1173 	 */
1174 	if (prop & SA_CONT)
1175 		SIG_STOPSIGMASK(p->p_siglist);
1176 
1177 	if (prop & SA_STOP) {
1178 		/*
1179 		 * If sending a tty stop signal to a member of an orphaned
1180 		 * process group, discard the signal here if the action
1181 		 * is default; don't stop the process below if sleeping,
1182 		 * and don't clear any pending SIGCONT.
1183 		 */
1184 		if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0 &&
1185 		    action == SIG_DFL) {
1186 			lwkt_reltoken(&p->p_token);
1187 			PRELE(p);
1188 		        return;
1189 		}
1190 		SIG_CONTSIGMASK(p->p_siglist);
1191 		p->p_flags &= ~P_CONTINUED;
1192 	}
1193 
1194 	if (p->p_stat == SSTOP) {
1195 		/*
1196 		 * Nobody can handle this signal, add it to the lwp or
1197 		 * process pending list
1198 		 */
1199 		if (lp) {
1200 			spin_lock(&lp->lwp_spin);
1201 			SIGADDSET(lp->lwp_siglist, sig);
1202 			spin_unlock(&lp->lwp_spin);
1203 		} else {
1204 			SIGADDSET(p->p_siglist, sig);
1205 		}
1206 
1207 		/*
1208 		 * If the process is stopped and is being traced, then no
1209 		 * further action is necessary.
1210 		 */
1211 		if (p->p_flags & P_TRACED)
1212 			goto out;
1213 
1214 		/*
1215 		 * If the process is stopped and receives a KILL signal,
1216 		 * make the process runnable.
1217 		 */
1218 		if (sig == SIGKILL) {
1219 			proc_unstop(p);
1220 			goto active_process;
1221 		}
1222 
1223 		/*
1224 		 * If the process is stopped and receives a CONT signal,
1225 		 * then try to make the process runnable again.
1226 		 */
1227 		if (prop & SA_CONT) {
1228 			/*
1229 			 * If SIGCONT is default (or ignored), we continue the
1230 			 * process but don't leave the signal in p_siglist, as
1231 			 * it has no further action.  If SIGCONT is held, we
1232 			 * continue the process and leave the signal in
1233 			 * p_siglist.  If the process catches SIGCONT, let it
1234 			 * handle the signal itself.
1235 			 *
1236 			 * XXX what if the signal is being held blocked?
1237 			 *
1238 			 * Token required to interlock kern_wait().
1239 			 * Reparenting can also cause a race so we have to
1240 			 * hold (q).
1241 			 */
1242 			q = p->p_pptr;
1243 			PHOLD(q);
1244 			lwkt_gettoken(&q->p_token);
1245 			p->p_flags |= P_CONTINUED;
1246 			wakeup(q);
1247 			if (action == SIG_DFL)
1248 				SIGDELSET(p->p_siglist, sig);
1249 			proc_unstop(p);
1250 			lwkt_reltoken(&q->p_token);
1251 			PRELE(q);
1252 			if (action == SIG_CATCH)
1253 				goto active_process;
1254 			goto out;
1255 		}
1256 
1257 		/*
1258 		 * If the process is stopped and receives another STOP
1259 		 * signal, we do not need to stop it again.  If we did
1260 		 * the shell could get confused.
1261 		 *
1262 		 * However, if the current/preempted lwp is part of the
1263 		 * process receiving the signal, we need to keep it,
1264 		 * so that this lwp can stop in issignal() later, as
1265 		 * we don't want to wait until it reaches userret!
1266 		 */
1267 		if (prop & SA_STOP) {
1268 			if (lwkt_preempted_proc() == NULL ||
1269 			    lwkt_preempted_proc()->lwp_proc != p)
1270 				SIGDELSET(p->p_siglist, sig);
1271 		}
1272 
1273 		/*
1274 		 * Otherwise the process is stopped and it received some
1275 		 * signal, which does not change its stopped state.  When
1276 		 * the process is continued a wakeup(p) will be issued which
1277 		 * will wakeup any threads sleeping in tstop().
1278 		 */
1279 		if (lp == NULL) {
1280 			/* NOTE: returns lp w/ token held */
1281 			lp = find_lwp_for_signal(p, sig);
1282 		}
1283 		goto out;
1284 
1285 		/* NOTREACHED */
1286 	}
1287 	/* else not stopped */
1288 active_process:
1289 
1290 	/*
1291 	 * Never deliver a lwp-specific signal to a random lwp.
1292 	 */
1293 	if (lp == NULL) {
1294 		/* NOTE: returns lp w/ token held */
1295 		lp = find_lwp_for_signal(p, sig);
1296 		if (lp) {
1297 			if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
1298 				lwkt_reltoken(&lp->lwp_token);
1299 				LWPRELE(lp);
1300 				lp = NULL;
1301 			}
1302 		}
1303 	}
1304 
1305 	/*
1306 	 * Deliver to the process generically if (1) the signal is being
1307 	 * sent to any thread or (2) we could not find a thread to deliver
1308 	 * it to.
1309 	 */
1310 	if (lp == NULL) {
1311 		SIGADDSET(p->p_siglist, sig);
1312 		goto out;
1313 	}
1314 
1315 	/*
1316 	 * Deliver to a specific LWP whether it masks it or not.  It will
1317 	 * not be dispatched if masked but we must still deliver it.
1318 	 */
1319 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1320 	    (p->p_flags & P_TRACED) == 0) {
1321 		p->p_nice = NZERO;
1322 	}
1323 
1324 	/*
1325 	 * If the process receives a STOP signal which indeed needs to
1326 	 * stop the process, do so.  If the process chose to catch the
1327 	 * signal, it will be treated like any other signal.
1328 	 */
1329 	if ((prop & SA_STOP) && action == SIG_DFL) {
1330 		/*
1331 		 * If a child holding parent blocked, stopping
1332 		 * could cause deadlock.  Take no action at this
1333 		 * time.
1334 		 */
1335 		if (p->p_flags & P_PPWAIT) {
1336 			SIGADDSET(p->p_siglist, sig);
1337 			goto out;
1338 		}
1339 
1340 		/*
1341 		 * Do not actually try to manipulate the process, but simply
1342 		 * stop it.  Lwps will stop as soon as they safely can.
1343 		 *
1344 		 * Ignore stop if the process is exiting.
1345 		 */
1346 		if ((p->p_flags & P_WEXIT) == 0) {
1347 			p->p_xstat = sig;
1348 			proc_stop(p);
1349 		}
1350 		goto out;
1351 	}
1352 
1353 	/*
1354 	 * If it is a CONT signal with default action, just ignore it.
1355 	 */
1356 	if ((prop & SA_CONT) && action == SIG_DFL)
1357 		goto out;
1358 
1359 	/*
1360 	 * Mark signal pending at this specific thread.
1361 	 */
1362 	spin_lock(&lp->lwp_spin);
1363 	SIGADDSET(lp->lwp_siglist, sig);
1364 	spin_unlock(&lp->lwp_spin);
1365 
1366 	lwp_signotify(lp);
1367 
1368 out:
1369 	if (lp) {
1370 		lwkt_reltoken(&lp->lwp_token);
1371 		LWPRELE(lp);
1372 	}
1373 	lwkt_reltoken(&p->p_token);
1374 	PRELE(p);
1375 }
1376 
1377 /*
1378  * Notify the LWP that a signal has arrived.  The LWP does not have to be
1379  * sleeping on the current cpu.
1380  *
1381  * p->p_token and lp->lwp_token must be held on call.
1382  *
1383  * We can only safely schedule the thread on its current cpu and only if
1384  * one of the SINTR flags is set.  If an SINTR flag is set AND we are on
1385  * the correct cpu we are properly interlocked, otherwise we could be
1386  * racing other thread transition states (or the lwp is on the user scheduler
1387  * runq but not scheduled) and must not do anything.
1388  *
1389  * Since we hold the lwp token we know the lwp cannot be ripped out from
1390  * under us so we can safely hold it to prevent it from being ripped out
1391  * from under us if we are forced to IPI another cpu to make the local
1392  * checks there.
1393  *
1394  * Adjustment of lp->lwp_stat can only occur when we hold the lwp_token,
1395  * which we won't in an IPI so any fixups have to be done here, effectively
1396  * replicating part of what setrunnable() does.
1397  */
1398 static void
1399 lwp_signotify(struct lwp *lp)
1400 {
1401 	ASSERT_LWKT_TOKEN_HELD(&lp->lwp_proc->p_token);
1402 
1403 	crit_enter();
1404 	if (lp == lwkt_preempted_proc()) {
1405 		/*
1406 		 * lwp is on the current cpu AND it is currently running
1407 		 * (we preempted it).
1408 		 */
1409 		signotify();
1410 	} else if (lp->lwp_flags & LWP_SINTR) {
1411 		/*
1412 		 * lwp is sitting in tsleep() with PCATCH set
1413 		 */
1414 #ifdef SMP
1415 		if (lp->lwp_thread->td_gd == mycpu) {
1416 			setrunnable(lp);
1417 		} else {
1418 			/*
1419 			 * We can only adjust lwp_stat while we hold the
1420 			 * lwp_token, and we won't in the IPI function.
1421 			 */
1422 			LWPHOLD(lp);
1423 			if (lp->lwp_stat == LSSTOP)
1424 				lp->lwp_stat = LSSLEEP;
1425 			lwkt_send_ipiq(lp->lwp_thread->td_gd,
1426 				       lwp_signotify_remote, lp);
1427 		}
1428 #else
1429 		setrunnable(lp);
1430 #endif
1431 	} else if (lp->lwp_thread->td_flags & TDF_SINTR) {
1432 		/*
1433 		 * lwp is sitting in lwkt_sleep() with PCATCH set.
1434 		 */
1435 #ifdef SMP
1436 		if (lp->lwp_thread->td_gd == mycpu) {
1437 			setrunnable(lp);
1438 		} else {
1439 			/*
1440 			 * We can only adjust lwp_stat while we hold the
1441 			 * lwp_token, and we won't in the IPI function.
1442 			 */
1443 			LWPHOLD(lp);
1444 			if (lp->lwp_stat == LSSTOP)
1445 				lp->lwp_stat = LSSLEEP;
1446 			lwkt_send_ipiq(lp->lwp_thread->td_gd,
1447 				       lwp_signotify_remote, lp);
1448 		}
1449 #else
1450 		setrunnable(lp);
1451 #endif
1452 	} else {
1453 		/*
1454 		 * Otherwise the lwp is either in some uninterruptable state
1455 		 * or it is on the userland scheduler's runqueue waiting to
1456 		 * be scheduled to a cpu.
1457 		 */
1458 	}
1459 	crit_exit();
1460 }
1461 
1462 #ifdef SMP
1463 
1464 /*
1465  * This function is called via an IPI so we cannot call setrunnable() here
1466  * (because while we hold the lp we don't own its token, and can't get it
1467  * from an IPI).
1468  *
1469  * We are interlocked by virtue of being on the same cpu as the target.  If
1470  * we still are and LWP_SINTR or TDF_SINTR is set we can safely schedule
1471  * the target thread.
1472  */
1473 static void
1474 lwp_signotify_remote(void *arg)
1475 {
1476 	struct lwp *lp = arg;
1477 	thread_t td = lp->lwp_thread;
1478 
1479 	if (lp == lwkt_preempted_proc()) {
1480 		signotify();
1481 		LWPRELE(lp);
1482 	} else if (td->td_gd == mycpu) {
1483 		if ((lp->lwp_flags & LWP_SINTR) ||
1484 		    (td->td_flags & TDF_SINTR)) {
1485 			lwkt_schedule(td);
1486 		}
1487 		LWPRELE(lp);
1488 	} else {
1489 		lwkt_send_ipiq(td->td_gd, lwp_signotify_remote, lp);
1490 		/* LWPHOLD() is forwarded to the target cpu */
1491 	}
1492 }
1493 
1494 #endif
1495 
1496 /*
1497  * Caller must hold p->p_token
1498  */
1499 void
1500 proc_stop(struct proc *p)
1501 {
1502 	struct proc *q;
1503 	struct lwp *lp;
1504 
1505 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1506 
1507 	/* If somebody raced us, be happy with it */
1508 	if (p->p_stat == SSTOP || p->p_stat == SZOMB) {
1509 		return;
1510 	}
1511 	p->p_stat = SSTOP;
1512 
1513 	FOREACH_LWP_IN_PROC(lp, p) {
1514 		LWPHOLD(lp);
1515 		lwkt_gettoken(&lp->lwp_token);
1516 
1517 		switch (lp->lwp_stat) {
1518 		case LSSTOP:
1519 			/*
1520 			 * Do nothing, we are already counted in
1521 			 * p_nstopped.
1522 			 */
1523 			break;
1524 
1525 		case LSSLEEP:
1526 			/*
1527 			 * We're sleeping, but we will stop before
1528 			 * returning to userspace, so count us
1529 			 * as stopped as well.  We set LWP_MP_WSTOP
1530 			 * to signal the lwp that it should not
1531 			 * increase p_nstopped when reaching tstop().
1532 			 *
1533 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1534 			 */
1535 			if ((lp->lwp_mpflags & LWP_MP_WSTOP) == 0) {
1536 				atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
1537 				++p->p_nstopped;
1538 			}
1539 			break;
1540 
1541 		case LSRUN:
1542 			/*
1543 			 * We might notify ourself, but that's not
1544 			 * a problem.
1545 			 */
1546 			lwp_signotify(lp);
1547 			break;
1548 		}
1549 		lwkt_reltoken(&lp->lwp_token);
1550 		LWPRELE(lp);
1551 	}
1552 
1553 	if (p->p_nstopped == p->p_nthreads) {
1554 		/*
1555 		 * Token required to interlock kern_wait().  Reparenting can
1556 		 * also cause a race so we have to hold (q).
1557 		 */
1558 		q = p->p_pptr;
1559 		PHOLD(q);
1560 		lwkt_gettoken(&q->p_token);
1561 		p->p_flags &= ~P_WAITED;
1562 		wakeup(q);
1563 		if ((q->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1564 			ksignal(p->p_pptr, SIGCHLD);
1565 		lwkt_reltoken(&q->p_token);
1566 		PRELE(q);
1567 	}
1568 }
1569 
1570 /*
1571  * Caller must hold proc_token
1572  */
1573 void
1574 proc_unstop(struct proc *p)
1575 {
1576 	struct lwp *lp;
1577 
1578 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1579 
1580 	if (p->p_stat != SSTOP)
1581 		return;
1582 
1583 	p->p_stat = SACTIVE;
1584 
1585 	FOREACH_LWP_IN_PROC(lp, p) {
1586 		LWPHOLD(lp);
1587 		lwkt_gettoken(&lp->lwp_token);
1588 
1589 		switch (lp->lwp_stat) {
1590 		case LSRUN:
1591 			/*
1592 			 * Uh?  Not stopped?  Well, I guess that's okay.
1593 			 */
1594 			if (bootverbose)
1595 				kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1596 					p->p_pid, lp->lwp_tid);
1597 			break;
1598 
1599 		case LSSLEEP:
1600 			/*
1601 			 * Still sleeping.  Don't bother waking it up.
1602 			 * However, if this thread was counted as
1603 			 * stopped, undo this.
1604 			 *
1605 			 * Nevertheless we call setrunnable() so that it
1606 			 * will wake up in case a signal or timeout arrived
1607 			 * in the meantime.
1608 			 *
1609 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1610 			 */
1611 			if (lp->lwp_mpflags & LWP_MP_WSTOP) {
1612 				atomic_clear_int(&lp->lwp_mpflags,
1613 						 LWP_MP_WSTOP);
1614 				--p->p_nstopped;
1615 			} else {
1616 				if (bootverbose)
1617 					kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1618 						p->p_pid, lp->lwp_tid);
1619 			}
1620 			/* FALLTHROUGH */
1621 
1622 		case LSSTOP:
1623 			/*
1624 			 * This handles any lwp's waiting in a tsleep with
1625 			 * SIGCATCH.
1626 			 */
1627 			lwp_signotify(lp);
1628 			break;
1629 
1630 		}
1631 		lwkt_reltoken(&lp->lwp_token);
1632 		LWPRELE(lp);
1633 	}
1634 
1635 	/*
1636 	 * This handles any lwp's waiting in tstop().  We have interlocked
1637 	 * the setting of p_stat by acquiring and releasing each lpw's
1638 	 * token.
1639 	 */
1640 	wakeup(p);
1641 }
1642 
1643 /*
1644  * No requirements.
1645  */
1646 static int
1647 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1648 {
1649 	sigset_t savedmask, set;
1650 	struct proc *p = curproc;
1651 	struct lwp *lp = curthread->td_lwp;
1652 	int error, sig, hz, timevalid = 0;
1653 	struct timespec rts, ets, ts;
1654 	struct timeval tv;
1655 
1656 	error = 0;
1657 	sig = 0;
1658 	ets.tv_sec = 0;		/* silence compiler warning */
1659 	ets.tv_nsec = 0;	/* silence compiler warning */
1660 	SIG_CANTMASK(waitset);
1661 	savedmask = lp->lwp_sigmask;
1662 
1663 	if (timeout) {
1664 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1665 		    timeout->tv_nsec < 1000000000) {
1666 			timevalid = 1;
1667 			getnanouptime(&rts);
1668 		 	ets = rts;
1669 			timespecadd(&ets, timeout);
1670 		}
1671 	}
1672 
1673 	for (;;) {
1674 		set = lwp_sigpend(lp);
1675 		SIGSETAND(set, waitset);
1676 		if ((sig = sig_ffs(&set)) != 0) {
1677 			SIGFILLSET(lp->lwp_sigmask);
1678 			SIGDELSET(lp->lwp_sigmask, sig);
1679 			SIG_CANTMASK(lp->lwp_sigmask);
1680 			sig = issignal(lp, 1);
1681 			/*
1682 			 * It may be a STOP signal, in the case, issignal
1683 			 * returns 0, because we may stop there, and new
1684 			 * signal can come in, we should restart if we got
1685 			 * nothing.
1686 			 */
1687 			if (sig == 0)
1688 				continue;
1689 			else
1690 				break;
1691 		}
1692 
1693 		/*
1694 		 * Previous checking got nothing, and we retried but still
1695 		 * got nothing, we should return the error status.
1696 		 */
1697 		if (error)
1698 			break;
1699 
1700 		/*
1701 		 * POSIX says this must be checked after looking for pending
1702 		 * signals.
1703 		 */
1704 		if (timeout) {
1705 			if (timevalid == 0) {
1706 				error = EINVAL;
1707 				break;
1708 			}
1709 			getnanouptime(&rts);
1710 			if (timespeccmp(&rts, &ets, >=)) {
1711 				error = EAGAIN;
1712 				break;
1713 			}
1714 			ts = ets;
1715 			timespecsub(&ts, &rts);
1716 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1717 			hz = tvtohz_high(&tv);
1718 		} else {
1719 			hz = 0;
1720 		}
1721 
1722 		lp->lwp_sigmask = savedmask;
1723 		SIGSETNAND(lp->lwp_sigmask, waitset);
1724 		/*
1725 		 * We won't ever be woken up.  Instead, our sleep will
1726 		 * be broken in lwpsignal().
1727 		 */
1728 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1729 		if (timeout) {
1730 			if (error == ERESTART) {
1731 				/* can not restart a timeout wait. */
1732 				error = EINTR;
1733 			} else if (error == EAGAIN) {
1734 				/* will calculate timeout by ourself. */
1735 				error = 0;
1736 			}
1737 		}
1738 		/* Retry ... */
1739 	}
1740 
1741 	lp->lwp_sigmask = savedmask;
1742 	if (sig) {
1743 		error = 0;
1744 		bzero(info, sizeof(*info));
1745 		info->si_signo = sig;
1746 		spin_lock(&lp->lwp_spin);
1747 		lwp_delsig(lp, sig);	/* take the signal! */
1748 		spin_unlock(&lp->lwp_spin);
1749 
1750 		if (sig == SIGKILL) {
1751 			sigexit(lp, sig);
1752 			/* NOT REACHED */
1753 		}
1754 	}
1755 
1756 	return (error);
1757 }
1758 
1759 /*
1760  * MPALMOSTSAFE
1761  */
1762 int
1763 sys_sigtimedwait(struct sigtimedwait_args *uap)
1764 {
1765 	struct timespec ts;
1766 	struct timespec *timeout;
1767 	sigset_t set;
1768 	siginfo_t info;
1769 	int error;
1770 
1771 	if (uap->timeout) {
1772 		error = copyin(uap->timeout, &ts, sizeof(ts));
1773 		if (error)
1774 			return (error);
1775 		timeout = &ts;
1776 	} else {
1777 		timeout = NULL;
1778 	}
1779 	error = copyin(uap->set, &set, sizeof(set));
1780 	if (error)
1781 		return (error);
1782 	error = kern_sigtimedwait(set, &info, timeout);
1783 	if (error)
1784 		return (error);
1785 	if (uap->info)
1786 		error = copyout(&info, uap->info, sizeof(info));
1787 	/* Repost if we got an error. */
1788 	/*
1789 	 * XXX lwp
1790 	 *
1791 	 * This could transform a thread-specific signal to another
1792 	 * thread / process pending signal.
1793 	 */
1794 	if (error) {
1795 		ksignal(curproc, info.si_signo);
1796 	} else {
1797 		uap->sysmsg_result = info.si_signo;
1798 	}
1799 	return (error);
1800 }
1801 
1802 /*
1803  * MPALMOSTSAFE
1804  */
1805 int
1806 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1807 {
1808 	siginfo_t info;
1809 	sigset_t set;
1810 	int error;
1811 
1812 	error = copyin(uap->set, &set, sizeof(set));
1813 	if (error)
1814 		return (error);
1815 	error = kern_sigtimedwait(set, &info, NULL);
1816 	if (error)
1817 		return (error);
1818 	if (uap->info)
1819 		error = copyout(&info, uap->info, sizeof(info));
1820 	/* Repost if we got an error. */
1821 	/*
1822 	 * XXX lwp
1823 	 *
1824 	 * This could transform a thread-specific signal to another
1825 	 * thread / process pending signal.
1826 	 */
1827 	if (error) {
1828 		ksignal(curproc, info.si_signo);
1829 	} else {
1830 		uap->sysmsg_result = info.si_signo;
1831 	}
1832 	return (error);
1833 }
1834 
1835 /*
1836  * If the current process has received a signal that would interrupt a
1837  * system call, return EINTR or ERESTART as appropriate.
1838  */
1839 int
1840 iscaught(struct lwp *lp)
1841 {
1842 	struct proc *p = lp->lwp_proc;
1843 	int sig;
1844 
1845 	if (p) {
1846 		if ((sig = CURSIG(lp)) != 0) {
1847 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1848 				return (EINTR);
1849 			return (ERESTART);
1850 		}
1851 	}
1852 	return(EWOULDBLOCK);
1853 }
1854 
1855 /*
1856  * If the current process has received a signal (should be caught or cause
1857  * termination, should interrupt current syscall), return the signal number.
1858  * Stop signals with default action are processed immediately, then cleared;
1859  * they aren't returned.  This is checked after each entry to the system for
1860  * a syscall or trap (though this can usually be done without calling issignal
1861  * by checking the pending signal masks in the CURSIG macro).
1862  *
1863  * This routine is called via CURSIG/__cursig.  We will acquire and release
1864  * p->p_token but if the caller needs to interlock the test the caller must
1865  * also hold p->p_token.
1866  *
1867  *	while (sig = CURSIG(curproc))
1868  *		postsig(sig);
1869  *
1870  * MPSAFE
1871  */
1872 int
1873 issignal(struct lwp *lp, int maytrace)
1874 {
1875 	struct proc *p = lp->lwp_proc;
1876 	sigset_t mask;
1877 	int sig, prop;
1878 
1879 	lwkt_gettoken(&p->p_token);
1880 
1881 	for (;;) {
1882 		int traced = (p->p_flags & P_TRACED) || (p->p_stops & S_SIG);
1883 
1884 		/*
1885 		 * If this process is supposed to stop, stop this thread.
1886 		 */
1887 		if (p->p_stat == SSTOP)
1888 			tstop();
1889 
1890 		mask = lwp_sigpend(lp);
1891 		SIGSETNAND(mask, lp->lwp_sigmask);
1892 		if (p->p_flags & P_PPWAIT)
1893 			SIG_STOPSIGMASK(mask);
1894 		if (SIGISEMPTY(mask)) {		/* no signal to send */
1895 			lwkt_reltoken(&p->p_token);
1896 			return (0);
1897 		}
1898 		sig = sig_ffs(&mask);
1899 
1900 		STOPEVENT(p, S_SIG, sig);
1901 
1902 		/*
1903 		 * We should see pending but ignored signals
1904 		 * only if P_TRACED was on when they were posted.
1905 		 */
1906 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1907 			spin_lock(&lp->lwp_spin);
1908 			lwp_delsig(lp, sig);
1909 			spin_unlock(&lp->lwp_spin);
1910 			continue;
1911 		}
1912 		if (maytrace &&
1913 		    (p->p_flags & P_TRACED) &&
1914 		    (p->p_flags & P_PPWAIT) == 0) {
1915 			/*
1916 			 * If traced, always stop, and stay stopped until
1917 			 * released by the parent.
1918 			 *
1919 			 * NOTE: SSTOP may get cleared during the loop,
1920 			 * but we do not re-notify the parent if we have
1921 			 * to loop several times waiting for the parent
1922 			 * to let us continue.
1923 			 *
1924 			 * XXX not sure if this is still true
1925 			 */
1926 			p->p_xstat = sig;
1927 			proc_stop(p);
1928 			do {
1929 				tstop();
1930 			} while (!trace_req(p) && (p->p_flags & P_TRACED));
1931 
1932 			/*
1933 			 * If parent wants us to take the signal,
1934 			 * then it will leave it in p->p_xstat;
1935 			 * otherwise we just look for signals again.
1936 			 */
1937 			spin_lock(&lp->lwp_spin);
1938 			lwp_delsig(lp, sig);	/* clear old signal */
1939 			spin_unlock(&lp->lwp_spin);
1940 			sig = p->p_xstat;
1941 			if (sig == 0)
1942 				continue;
1943 
1944 			/*
1945 			 * Put the new signal into p_siglist.  If the
1946 			 * signal is being masked, look for other signals.
1947 			 *
1948 			 * XXX lwp might need a call to ksignal()
1949 			 */
1950 			SIGADDSET(p->p_siglist, sig);
1951 			if (SIGISMEMBER(lp->lwp_sigmask, sig))
1952 				continue;
1953 
1954 			/*
1955 			 * If the traced bit got turned off, go back up
1956 			 * to the top to rescan signals.  This ensures
1957 			 * that p_sig* and ps_sigact are consistent.
1958 			 */
1959 			if ((p->p_flags & P_TRACED) == 0)
1960 				continue;
1961 		}
1962 
1963 		prop = sigprop(sig);
1964 
1965 		/*
1966 		 * Decide whether the signal should be returned.
1967 		 * Return the signal's number, or fall through
1968 		 * to clear it from the pending mask.
1969 		 */
1970 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
1971 		case (intptr_t)SIG_DFL:
1972 			/*
1973 			 * Don't take default actions on system processes.
1974 			 */
1975 			if (p->p_pid <= 1) {
1976 #ifdef DIAGNOSTIC
1977 				/*
1978 				 * Are you sure you want to ignore SIGSEGV
1979 				 * in init? XXX
1980 				 */
1981 				kprintf("Process (pid %lu) got signal %d\n",
1982 					(u_long)p->p_pid, sig);
1983 #endif
1984 				break;		/* == ignore */
1985 			}
1986 
1987 			/*
1988 			 * Handle the in-kernel checkpoint action
1989 			 */
1990 			if (prop & SA_CKPT) {
1991 				checkpoint_signal_handler(lp);
1992 				break;
1993 			}
1994 
1995 			/*
1996 			 * If there is a pending stop signal to process
1997 			 * with default action, stop here,
1998 			 * then clear the signal.  However,
1999 			 * if process is member of an orphaned
2000 			 * process group, ignore tty stop signals.
2001 			 */
2002 			if (prop & SA_STOP) {
2003 				if (p->p_flags & P_TRACED ||
2004 		    		    (p->p_pgrp->pg_jobc == 0 &&
2005 				    prop & SA_TTYSTOP))
2006 					break;	/* == ignore */
2007 				if ((p->p_flags & P_WEXIT) == 0) {
2008 					p->p_xstat = sig;
2009 					proc_stop(p);
2010 					tstop();
2011 				}
2012 				break;
2013 			} else if (prop & SA_IGNORE) {
2014 				/*
2015 				 * Except for SIGCONT, shouldn't get here.
2016 				 * Default action is to ignore; drop it.
2017 				 */
2018 				break;		/* == ignore */
2019 			} else {
2020 				lwkt_reltoken(&p->p_token);
2021 				return (sig);
2022 			}
2023 
2024 			/*NOTREACHED*/
2025 
2026 		case (intptr_t)SIG_IGN:
2027 			/*
2028 			 * Masking above should prevent us ever trying
2029 			 * to take action on an ignored signal other
2030 			 * than SIGCONT, unless process is traced.
2031 			 */
2032 			if ((prop & SA_CONT) == 0 &&
2033 			    (p->p_flags & P_TRACED) == 0)
2034 				kprintf("issignal\n");
2035 			break;		/* == ignore */
2036 
2037 		default:
2038 			/*
2039 			 * This signal has an action, let
2040 			 * postsig() process it.
2041 			 */
2042 			lwkt_reltoken(&p->p_token);
2043 			return (sig);
2044 		}
2045 		spin_lock(&lp->lwp_spin);
2046 		lwp_delsig(lp, sig);		/* take the signal! */
2047 		spin_unlock(&lp->lwp_spin);
2048 	}
2049 	/* NOTREACHED */
2050 }
2051 
2052 /*
2053  * Take the action for the specified signal
2054  * from the current set of pending signals.
2055  *
2056  * Caller must hold p->p_token
2057  */
2058 void
2059 postsig(int sig)
2060 {
2061 	struct lwp *lp = curthread->td_lwp;
2062 	struct proc *p = lp->lwp_proc;
2063 	struct sigacts *ps = p->p_sigacts;
2064 	sig_t action;
2065 	sigset_t returnmask;
2066 	int code;
2067 
2068 	KASSERT(sig != 0, ("postsig"));
2069 
2070 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
2071 
2072 	/*
2073 	 * If we are a virtual kernel running an emulated user process
2074 	 * context, switch back to the virtual kernel context before
2075 	 * trying to post the signal.
2076 	 */
2077 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
2078 		struct trapframe *tf = lp->lwp_md.md_regs;
2079 		tf->tf_trapno = 0;
2080 		vkernel_trap(lp, tf);
2081 	}
2082 
2083 	spin_lock(&lp->lwp_spin);
2084 	lwp_delsig(lp, sig);
2085 	spin_unlock(&lp->lwp_spin);
2086 	action = ps->ps_sigact[_SIG_IDX(sig)];
2087 #ifdef KTRACE
2088 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
2089 		ktrpsig(lp, sig, action, lp->lwp_flags & LWP_OLDMASK ?
2090 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
2091 #endif
2092 	STOPEVENT(p, S_SIG, sig);
2093 
2094 	if (action == SIG_DFL) {
2095 		/*
2096 		 * Default action, where the default is to kill
2097 		 * the process.  (Other cases were ignored above.)
2098 		 */
2099 		sigexit(lp, sig);
2100 		/* NOTREACHED */
2101 	} else {
2102 		/*
2103 		 * If we get here, the signal must be caught.
2104 		 */
2105 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
2106 		    ("postsig action"));
2107 
2108 		/*
2109 		 * Reset the signal handler if asked to
2110 		 */
2111 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
2112 			/*
2113 			 * See kern_sigaction() for origin of this code.
2114 			 */
2115 			SIGDELSET(p->p_sigcatch, sig);
2116 			if (sig != SIGCONT &&
2117 			    sigprop(sig) & SA_IGNORE)
2118 				SIGADDSET(p->p_sigignore, sig);
2119 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2120 		}
2121 
2122 		/*
2123 		 * Set the signal mask and calculate the mask to restore
2124 		 * when the signal function returns.
2125 		 *
2126 		 * Special case: user has done a sigsuspend.  Here the
2127 		 * current mask is not of interest, but rather the
2128 		 * mask from before the sigsuspend is what we want
2129 		 * restored after the signal processing is completed.
2130 		 */
2131 		if (lp->lwp_flags & LWP_OLDMASK) {
2132 			returnmask = lp->lwp_oldsigmask;
2133 			lp->lwp_flags &= ~LWP_OLDMASK;
2134 		} else {
2135 			returnmask = lp->lwp_sigmask;
2136 		}
2137 
2138 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
2139 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
2140 			SIGADDSET(lp->lwp_sigmask, sig);
2141 
2142 		lp->lwp_ru.ru_nsignals++;
2143 		if (lp->lwp_sig != sig) {
2144 			code = 0;
2145 		} else {
2146 			code = lp->lwp_code;
2147 			lp->lwp_code = 0;
2148 			lp->lwp_sig = 0;
2149 		}
2150 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
2151 	}
2152 }
2153 
2154 /*
2155  * Kill the current process for stated reason.
2156  */
2157 void
2158 killproc(struct proc *p, char *why)
2159 {
2160 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
2161 		p->p_pid, p->p_comm,
2162 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2163 	ksignal(p, SIGKILL);
2164 }
2165 
2166 /*
2167  * Force the current process to exit with the specified signal, dumping core
2168  * if appropriate.  We bypass the normal tests for masked and caught signals,
2169  * allowing unrecoverable failures to terminate the process without changing
2170  * signal state.  Mark the accounting record with the signal termination.
2171  * If dumping core, save the signal number for the debugger.  Calls exit and
2172  * does not return.
2173  *
2174  * This routine does not return.
2175  */
2176 void
2177 sigexit(struct lwp *lp, int sig)
2178 {
2179 	struct proc *p = lp->lwp_proc;
2180 
2181 	lwkt_gettoken(&p->p_token);
2182 	p->p_acflag |= AXSIG;
2183 	if (sigprop(sig) & SA_CORE) {
2184 		lp->lwp_sig = sig;
2185 		/*
2186 		 * Log signals which would cause core dumps
2187 		 * (Log as LOG_INFO to appease those who don't want
2188 		 * these messages.)
2189 		 * XXX : Todo, as well as euid, write out ruid too
2190 		 */
2191 		if (coredump(lp, sig) == 0)
2192 			sig |= WCOREFLAG;
2193 		if (kern_logsigexit)
2194 			log(LOG_INFO,
2195 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2196 			    p->p_pid, p->p_comm,
2197 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
2198 			    sig &~ WCOREFLAG,
2199 			    sig & WCOREFLAG ? " (core dumped)" : "");
2200 	}
2201 	lwkt_reltoken(&p->p_token);
2202 	exit1(W_EXITCODE(0, sig));
2203 	/* NOTREACHED */
2204 }
2205 
2206 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
2207 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2208 	      sizeof(corefilename), "process corefile name format string");
2209 
2210 /*
2211  * expand_name(name, uid, pid)
2212  * Expand the name described in corefilename, using name, uid, and pid.
2213  * corefilename is a kprintf-like string, with three format specifiers:
2214  *	%N	name of process ("name")
2215  *	%P	process id (pid)
2216  *	%U	user id (uid)
2217  * For example, "%N.core" is the default; they can be disabled completely
2218  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2219  * This is controlled by the sysctl variable kern.corefile (see above).
2220  */
2221 
2222 static char *
2223 expand_name(const char *name, uid_t uid, pid_t pid)
2224 {
2225 	char *temp;
2226 	char buf[11];		/* Buffer for pid/uid -- max 4B */
2227 	int i, n;
2228 	char *format = corefilename;
2229 	size_t namelen;
2230 
2231 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
2232 	if (temp == NULL)
2233 		return NULL;
2234 	namelen = strlen(name);
2235 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2236 		int l;
2237 		switch (format[i]) {
2238 		case '%':	/* Format character */
2239 			i++;
2240 			switch (format[i]) {
2241 			case '%':
2242 				temp[n++] = '%';
2243 				break;
2244 			case 'N':	/* process name */
2245 				if ((n + namelen) > MAXPATHLEN) {
2246 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2247 					    pid, name, uid, temp, name);
2248 					kfree(temp, M_TEMP);
2249 					return NULL;
2250 				}
2251 				memcpy(temp+n, name, namelen);
2252 				n += namelen;
2253 				break;
2254 			case 'P':	/* process id */
2255 				l = ksprintf(buf, "%u", pid);
2256 				if ((n + l) > MAXPATHLEN) {
2257 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2258 					    pid, name, uid, temp, name);
2259 					kfree(temp, M_TEMP);
2260 					return NULL;
2261 				}
2262 				memcpy(temp+n, buf, l);
2263 				n += l;
2264 				break;
2265 			case 'U':	/* user id */
2266 				l = ksprintf(buf, "%u", uid);
2267 				if ((n + l) > MAXPATHLEN) {
2268 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2269 					    pid, name, uid, temp, name);
2270 					kfree(temp, M_TEMP);
2271 					return NULL;
2272 				}
2273 				memcpy(temp+n, buf, l);
2274 				n += l;
2275 				break;
2276 			default:
2277 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2278 			}
2279 			break;
2280 		default:
2281 			temp[n++] = format[i];
2282 		}
2283 	}
2284 	temp[n] = '\0';
2285 	return temp;
2286 }
2287 
2288 /*
2289  * Dump a process' core.  The main routine does some
2290  * policy checking, and creates the name of the coredump;
2291  * then it passes on a vnode and a size limit to the process-specific
2292  * coredump routine if there is one; if there _is not_ one, it returns
2293  * ENOSYS; otherwise it returns the error from the process-specific routine.
2294  *
2295  * The parameter `lp' is the lwp which triggered the coredump.
2296  */
2297 
2298 static int
2299 coredump(struct lwp *lp, int sig)
2300 {
2301 	struct proc *p = lp->lwp_proc;
2302 	struct vnode *vp;
2303 	struct ucred *cred = p->p_ucred;
2304 	struct flock lf;
2305 	struct nlookupdata nd;
2306 	struct vattr vattr;
2307 	int error, error1;
2308 	char *name;			/* name of corefile */
2309 	off_t limit;
2310 
2311 	STOPEVENT(p, S_CORE, 0);
2312 
2313 	if (((sugid_coredump == 0) && p->p_flags & P_SUGID) || do_coredump == 0)
2314 		return (EFAULT);
2315 
2316 	/*
2317 	 * Note that the bulk of limit checking is done after
2318 	 * the corefile is created.  The exception is if the limit
2319 	 * for corefiles is 0, in which case we don't bother
2320 	 * creating the corefile at all.  This layout means that
2321 	 * a corefile is truncated instead of not being created,
2322 	 * if it is larger than the limit.
2323 	 */
2324 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2325 	if (limit == 0)
2326 		return EFBIG;
2327 
2328 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2329 	if (name == NULL)
2330 		return (EINVAL);
2331 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2332 	if (error == 0)
2333 		error = vn_open(&nd, NULL, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
2334 	kfree(name, M_TEMP);
2335 	if (error) {
2336 		nlookup_done(&nd);
2337 		return (error);
2338 	}
2339 	vp = nd.nl_open_vp;
2340 	nd.nl_open_vp = NULL;
2341 	nlookup_done(&nd);
2342 
2343 	vn_unlock(vp);
2344 	lf.l_whence = SEEK_SET;
2345 	lf.l_start = 0;
2346 	lf.l_len = 0;
2347 	lf.l_type = F_WRLCK;
2348 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2349 	if (error)
2350 		goto out2;
2351 
2352 	/* Don't dump to non-regular files or files with links. */
2353 	if (vp->v_type != VREG ||
2354 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2355 		error = EFAULT;
2356 		goto out1;
2357 	}
2358 
2359 	/* Don't dump to files current user does not own */
2360 	if (vattr.va_uid != p->p_ucred->cr_uid) {
2361 		error = EFAULT;
2362 		goto out1;
2363 	}
2364 
2365 	VATTR_NULL(&vattr);
2366 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2367 	vattr.va_size = 0;
2368 	VOP_SETATTR(vp, &vattr, cred);
2369 	p->p_acflag |= ACORE;
2370 	vn_unlock(vp);
2371 
2372 	error = p->p_sysent->sv_coredump ?
2373 		  p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2374 
2375 out1:
2376 	lf.l_type = F_UNLCK;
2377 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2378 out2:
2379 	error1 = vn_close(vp, FWRITE);
2380 	if (error == 0)
2381 		error = error1;
2382 	return (error);
2383 }
2384 
2385 /*
2386  * Nonexistent system call-- signal process (may want to handle it).
2387  * Flag error in case process won't see signal immediately (blocked or ignored).
2388  *
2389  * MPALMOSTSAFE
2390  */
2391 /* ARGSUSED */
2392 int
2393 sys_nosys(struct nosys_args *args)
2394 {
2395 	lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2396 	return (EINVAL);
2397 }
2398 
2399 /*
2400  * Send a SIGIO or SIGURG signal to a process or process group using
2401  * stored credentials rather than those of the current process.
2402  */
2403 void
2404 pgsigio(struct sigio *sigio, int sig, int checkctty)
2405 {
2406 	if (sigio == NULL)
2407 		return;
2408 
2409 	if (sigio->sio_pgid > 0) {
2410 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2411 		             sigio->sio_proc))
2412 			ksignal(sigio->sio_proc, sig);
2413 	} else if (sigio->sio_pgid < 0) {
2414 		struct proc *p;
2415 		struct pgrp *pg = sigio->sio_pgrp;
2416 
2417 		/*
2418 		 * Must interlock all signals against fork
2419 		 */
2420 		pgref(pg);
2421 		lockmgr(&pg->pg_lock, LK_EXCLUSIVE);
2422 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
2423 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2424 			    (checkctty == 0 || (p->p_flags & P_CONTROLT)))
2425 				ksignal(p, sig);
2426 		}
2427 		lockmgr(&pg->pg_lock, LK_RELEASE);
2428 		pgrel(pg);
2429 	}
2430 }
2431 
2432 static int
2433 filt_sigattach(struct knote *kn)
2434 {
2435 	struct proc *p = curproc;
2436 
2437 	kn->kn_ptr.p_proc = p;
2438 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
2439 
2440 	/* XXX lock the proc here while adding to the list? */
2441 	knote_insert(&p->p_klist, kn);
2442 
2443 	return (0);
2444 }
2445 
2446 static void
2447 filt_sigdetach(struct knote *kn)
2448 {
2449 	struct proc *p = kn->kn_ptr.p_proc;
2450 
2451 	knote_remove(&p->p_klist, kn);
2452 }
2453 
2454 /*
2455  * signal knotes are shared with proc knotes, so we apply a mask to
2456  * the hint in order to differentiate them from process hints.  This
2457  * could be avoided by using a signal-specific knote list, but probably
2458  * isn't worth the trouble.
2459  */
2460 static int
2461 filt_signal(struct knote *kn, long hint)
2462 {
2463 	if (hint & NOTE_SIGNAL) {
2464 		hint &= ~NOTE_SIGNAL;
2465 
2466 		if (kn->kn_id == hint)
2467 			kn->kn_data++;
2468 	}
2469 	return (kn->kn_data != 0);
2470 }
2471