xref: /dragonfly/sys/kern/kern_sig.c (revision cecb9aae)
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 static void	lwp_signotify_remote(void *arg);
84 static int	kern_sigtimedwait(sigset_t set, siginfo_t *info,
85 		    struct timespec *timeout);
86 
87 static int	filt_sigattach(struct knote *kn);
88 static void	filt_sigdetach(struct knote *kn);
89 static int	filt_signal(struct knote *kn, long hint);
90 
91 struct filterops sig_filtops =
92 	{ 0, filt_sigattach, filt_sigdetach, filt_signal };
93 
94 static int	kern_logsigexit = 1;
95 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
96     &kern_logsigexit, 0,
97     "Log processes quitting on abnormal signals to syslog(3)");
98 
99 /*
100  * Can process p, with pcred pc, send the signal sig to process q?
101  */
102 #define CANSIGNAL(q, sig) \
103 	(!p_trespass(curproc->p_ucred, (q)->p_ucred) || \
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(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(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 			if (lp->lwp_sigstk.ss_flags & SS_ONSTACK)
614 				return (EINVAL);
615 			lp->lwp_flags &= ~LWP_ALTSTACK;
616 			lp->lwp_sigstk.ss_flags = ss->ss_flags;
617 		} else {
618 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
619 				return (ENOMEM);
620 			lp->lwp_flags |= LWP_ALTSTACK;
621 			lp->lwp_sigstk = *ss;
622 		}
623 	}
624 
625 	return (0);
626 }
627 
628 /*
629  * MPSAFE
630  */
631 int
632 sys_sigaltstack(struct sigaltstack_args *uap)
633 {
634 	stack_t ss, oss;
635 	int error;
636 
637 	if (uap->ss) {
638 		error = copyin(uap->ss, &ss, sizeof(ss));
639 		if (error)
640 			return (error);
641 	}
642 
643 	error = kern_sigaltstack(uap->ss ? &ss : NULL,
644 	    uap->oss ? &oss : NULL);
645 
646 	if (error == 0 && uap->oss)
647 		error = copyout(&oss, uap->oss, sizeof(*uap->oss));
648 	return (error);
649 }
650 
651 /*
652  * Common code for kill process group/broadcast kill.
653  * cp is calling process.
654  */
655 struct killpg_info {
656 	int nfound;
657 	int sig;
658 };
659 
660 static int killpg_all_callback(struct proc *p, void *data);
661 
662 static int
663 dokillpg(int sig, int pgid, int all)
664 {
665 	struct killpg_info info;
666 	struct proc *cp = curproc;
667 	struct proc *p;
668 	struct pgrp *pgrp;
669 
670 	info.nfound = 0;
671 	info.sig = sig;
672 
673 	if (all) {
674 		/*
675 		 * broadcast
676 		 */
677 		allproc_scan(killpg_all_callback, &info);
678 	} else {
679 		if (pgid == 0) {
680 			/*
681 			 * zero pgid means send to my process group.
682 			 */
683 			pgrp = cp->p_pgrp;
684 			pgref(pgrp);
685 		} else {
686 			pgrp = pgfind(pgid);
687 			if (pgrp == NULL)
688 				return (ESRCH);
689 		}
690 
691 		/*
692 		 * Must interlock all signals against fork
693 		 */
694 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
695 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
696 			if (p->p_pid <= 1 ||
697 			    p->p_stat == SZOMB ||
698 			    (p->p_flags & P_SYSTEM) ||
699 			    !CANSIGNAL(p, sig)) {
700 				continue;
701 			}
702 			++info.nfound;
703 			if (sig)
704 				ksignal(p, sig);
705 		}
706 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
707 		pgrel(pgrp);
708 	}
709 	return (info.nfound ? 0 : ESRCH);
710 }
711 
712 static int
713 killpg_all_callback(struct proc *p, void *data)
714 {
715 	struct killpg_info *info = data;
716 
717 	if (p->p_pid <= 1 || (p->p_flags & P_SYSTEM) ||
718 	    p == curproc || !CANSIGNAL(p, info->sig)) {
719 		return (0);
720 	}
721 	++info->nfound;
722 	if (info->sig)
723 		ksignal(p, info->sig);
724 	return(0);
725 }
726 
727 /*
728  * Send a general signal to a process or LWPs within that process.
729  *
730  * Note that new signals cannot be sent if a process is exiting or already
731  * a zombie, but we return success anyway as userland is likely to not handle
732  * the race properly.
733  *
734  * No requirements.
735  */
736 int
737 kern_kill(int sig, pid_t pid, lwpid_t tid)
738 {
739 	int t;
740 
741 	if ((u_int)sig > _SIG_MAXSIG)
742 		return (EINVAL);
743 
744 	lwkt_gettoken(&proc_token);
745 
746 	if (pid > 0) {
747 		struct proc *p;
748 		struct lwp *lp = NULL;
749 
750 		/*
751 		 * Send a signal to a single process.  If the kill() is
752 		 * racing an exiting process which has not yet been reaped
753 		 * act as though the signal was delivered successfully but
754 		 * don't actually try to deliver the signal.
755 		 */
756 		if ((p = pfind(pid)) == NULL) {
757 			if ((p = zpfind(pid)) == NULL) {
758 				lwkt_reltoken(&proc_token);
759 				return (ESRCH);
760 			}
761 			lwkt_reltoken(&proc_token);
762 			PRELE(p);
763 			return (0);
764 		}
765 		lwkt_gettoken(&p->p_token);
766 		if (!CANSIGNAL(p, sig)) {
767 			lwkt_reltoken(&p->p_token);
768 			PRELE(p);
769 			lwkt_reltoken(&proc_token);
770 			return (EPERM);
771 		}
772 
773 		/*
774 		 * NOP if the process is exiting.  Note that lwpsignal() is
775 		 * called directly with P_WEXIT set to kill individual LWPs
776 		 * during exit, which is allowed.
777 		 */
778 		if (p->p_flags & P_WEXIT) {
779 			lwkt_reltoken(&p->p_token);
780 			PRELE(p);
781 			lwkt_reltoken(&proc_token);
782 			return (0);
783 		}
784 		if (tid != -1) {
785 			lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
786 			if (lp == NULL) {
787 				lwkt_reltoken(&p->p_token);
788 				PRELE(p);
789 				lwkt_reltoken(&proc_token);
790 				return (ESRCH);
791 			}
792 		}
793 		if (sig)
794 			lwpsignal(p, lp, sig);
795 		lwkt_reltoken(&p->p_token);
796 		PRELE(p);
797 		lwkt_reltoken(&proc_token);
798 		return (0);
799 	}
800 
801 	/*
802 	 * If we come here, pid is a special broadcast pid.
803 	 * This doesn't mix with a tid.
804 	 */
805 	if (tid != -1) {
806 		lwkt_reltoken(&proc_token);
807 		return (EINVAL);
808 	}
809 	switch (pid) {
810 	case -1:		/* broadcast signal */
811 		t = (dokillpg(sig, 0, 1));
812 		break;
813 	case 0:			/* signal own process group */
814 		t = (dokillpg(sig, 0, 0));
815 		break;
816 	default:		/* negative explicit process group */
817 		t = (dokillpg(sig, -pid, 0));
818 		break;
819 	}
820 	lwkt_reltoken(&proc_token);
821 	return t;
822 }
823 
824 int
825 sys_kill(struct kill_args *uap)
826 {
827 	int error;
828 
829 	error = kern_kill(uap->signum, uap->pid, -1);
830 	return (error);
831 }
832 
833 int
834 sys_lwp_kill(struct lwp_kill_args *uap)
835 {
836 	int error;
837 	pid_t pid = uap->pid;
838 
839 	/*
840 	 * A tid is mandatory for lwp_kill(), otherwise
841 	 * you could simply use kill().
842 	 */
843 	if (uap->tid == -1)
844 		return (EINVAL);
845 
846 	/*
847 	 * To save on a getpid() function call for intra-process
848 	 * signals, pid == -1 means current process.
849 	 */
850 	if (pid == -1)
851 		pid = curproc->p_pid;
852 
853 	error = kern_kill(uap->signum, pid, uap->tid);
854 	return (error);
855 }
856 
857 /*
858  * Send a signal to a process group.
859  */
860 void
861 gsignal(int pgid, int sig)
862 {
863 	struct pgrp *pgrp;
864 
865 	if (pgid && (pgrp = pgfind(pgid)))
866 		pgsignal(pgrp, sig, 0);
867 }
868 
869 /*
870  * Send a signal to a process group.  If checktty is 1,
871  * limit to members which have a controlling terminal.
872  *
873  * pg_lock interlocks against a fork that might be in progress, to
874  * ensure that the new child process picks up the signal.
875  */
876 void
877 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
878 {
879 	struct proc *p;
880 
881 	/*
882 	 * Must interlock all signals against fork
883 	 */
884 	if (pgrp) {
885 		pgref(pgrp);
886 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
887 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
888 			if (checkctty == 0 || p->p_flags & P_CONTROLT)
889 				ksignal(p, sig);
890 		}
891 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
892 		pgrel(pgrp);
893 	}
894 }
895 
896 /*
897  * Send a signal caused by a trap to the current lwp.  If it will be caught
898  * immediately, deliver it with correct code.  Otherwise, post it normally.
899  *
900  * These signals may ONLY be delivered to the specified lwp and may never
901  * be delivered to the process generically.
902  */
903 void
904 trapsignal(struct lwp *lp, int sig, u_long code)
905 {
906 	struct proc *p = lp->lwp_proc;
907 	struct sigacts *ps = p->p_sigacts;
908 
909 	/*
910 	 * If we are a virtual kernel running an emulated user process
911 	 * context, switch back to the virtual kernel context before
912 	 * trying to post the signal.
913 	 */
914 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
915 		struct trapframe *tf = lp->lwp_md.md_regs;
916 		tf->tf_trapno = 0;
917 		vkernel_trap(lp, tf);
918 	}
919 
920 
921 	if ((p->p_flags & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
922 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
923 		lp->lwp_ru.ru_nsignals++;
924 #ifdef KTRACE
925 		if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
926 			ktrpsig(lp, sig, ps->ps_sigact[_SIG_IDX(sig)],
927 				&lp->lwp_sigmask, code);
928 #endif
929 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
930 						&lp->lwp_sigmask, code);
931 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
932 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
933 			SIGADDSET(lp->lwp_sigmask, sig);
934 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
935 			/*
936 			 * See kern_sigaction() for origin of this code.
937 			 */
938 			SIGDELSET(p->p_sigcatch, sig);
939 			if (sig != SIGCONT &&
940 			    sigprop(sig) & SA_IGNORE)
941 				SIGADDSET(p->p_sigignore, sig);
942 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
943 		}
944 	} else {
945 		lp->lwp_code = code;	/* XXX for core dump/debugger */
946 		lp->lwp_sig = sig;	/* XXX to verify code */
947 		lwpsignal(p, lp, sig);
948 	}
949 }
950 
951 /*
952  * Find a suitable lwp to deliver the signal to.  Returns NULL if all
953  * lwps hold the signal blocked.
954  *
955  * Caller must hold p->p_token.
956  *
957  * Returns a lp or NULL.  If non-NULL the lp is held and its token is
958  * acquired.
959  */
960 static struct lwp *
961 find_lwp_for_signal(struct proc *p, int sig)
962 {
963 	struct lwp *lp;
964 	struct lwp *run, *sleep, *stop;
965 
966 	/*
967 	 * If the running/preempted thread belongs to the proc to which
968 	 * the signal is being delivered and this thread does not block
969 	 * the signal, then we can avoid a context switch by delivering
970 	 * the signal to this thread, because it will return to userland
971 	 * soon anyways.
972 	 */
973 	lp = lwkt_preempted_proc();
974 	if (lp != NULL && lp->lwp_proc == p) {
975 		LWPHOLD(lp);
976 		lwkt_gettoken(&lp->lwp_token);
977 		if (!SIGISMEMBER(lp->lwp_sigmask, sig)) {
978 			/* return w/ token held */
979 			return (lp);
980 		}
981 		lwkt_reltoken(&lp->lwp_token);
982 		LWPRELE(lp);
983 	}
984 
985 	run = sleep = stop = NULL;
986 	FOREACH_LWP_IN_PROC(lp, p) {
987 		/*
988 		 * If the signal is being blocked by the lwp, then this
989 		 * lwp is not eligible for receiving the signal.
990 		 */
991 		LWPHOLD(lp);
992 		lwkt_gettoken(&lp->lwp_token);
993 
994 		if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
995 			lwkt_reltoken(&lp->lwp_token);
996 			LWPRELE(lp);
997 			continue;
998 		}
999 
1000 		switch (lp->lwp_stat) {
1001 		case LSRUN:
1002 			if (sleep) {
1003 				lwkt_token_swap();
1004 				lwkt_reltoken(&sleep->lwp_token);
1005 				LWPRELE(sleep);
1006 				sleep = NULL;
1007 				run = lp;
1008 			} else if (stop) {
1009 				lwkt_token_swap();
1010 				lwkt_reltoken(&stop->lwp_token);
1011 				LWPRELE(stop);
1012 				stop = NULL;
1013 				run = lp;
1014 			} else {
1015 				run = lp;
1016 			}
1017 			break;
1018 		case LSSLEEP:
1019 			if (lp->lwp_flags & LWP_SINTR) {
1020 				if (sleep) {
1021 					lwkt_reltoken(&lp->lwp_token);
1022 					LWPRELE(lp);
1023 				} else if (stop) {
1024 					lwkt_token_swap();
1025 					lwkt_reltoken(&stop->lwp_token);
1026 					LWPRELE(stop);
1027 					stop = NULL;
1028 					sleep = lp;
1029 				} else {
1030 					sleep = lp;
1031 				}
1032 			} else {
1033 				lwkt_reltoken(&lp->lwp_token);
1034 				LWPRELE(lp);
1035 			}
1036 			break;
1037 		case LSSTOP:
1038 			if (sleep) {
1039 				lwkt_reltoken(&lp->lwp_token);
1040 				LWPRELE(lp);
1041 			} else if (stop) {
1042 				lwkt_reltoken(&lp->lwp_token);
1043 				LWPRELE(lp);
1044 			} else {
1045 				stop = lp;
1046 			}
1047 			break;
1048 		}
1049 		if (run)
1050 			break;
1051 	}
1052 
1053 	if (run != NULL)
1054 		return (run);
1055 	else if (sleep != NULL)
1056 		return (sleep);
1057 	else
1058 		return (stop);
1059 }
1060 
1061 /*
1062  * Send the signal to the process.  If the signal has an action, the action
1063  * is usually performed by the target process rather than the caller; we add
1064  * the signal to the set of pending signals for the process.
1065  *
1066  * Exceptions:
1067  *   o When a stop signal is sent to a sleeping process that takes the
1068  *     default action, the process is stopped without awakening it.
1069  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
1070  *     regardless of the signal action (eg, blocked or ignored).
1071  *
1072  * Other ignored signals are discarded immediately.
1073  *
1074  * If the caller wishes to call this function from a hard code section the
1075  * caller must already hold p->p_token (see kern_clock.c).
1076  *
1077  * No requirements.
1078  */
1079 void
1080 ksignal(struct proc *p, int sig)
1081 {
1082 	lwpsignal(p, NULL, sig);
1083 }
1084 
1085 /*
1086  * The core for ksignal.  lp may be NULL, then a suitable thread
1087  * will be chosen.  If not, lp MUST be a member of p.
1088  *
1089  * If the caller wishes to call this function from a hard code section the
1090  * caller must already hold p->p_token.
1091  *
1092  * No requirements.
1093  */
1094 void
1095 lwpsignal(struct proc *p, struct lwp *lp, int sig)
1096 {
1097 	struct proc *q;
1098 	sig_t action;
1099 	int prop;
1100 
1101 	if (sig > _SIG_MAXSIG || sig <= 0) {
1102 		kprintf("lwpsignal: signal %d\n", sig);
1103 		panic("lwpsignal signal number");
1104 	}
1105 
1106 	KKASSERT(lp == NULL || lp->lwp_proc == p);
1107 
1108 	/*
1109 	 * We don't want to race... well, all sorts of things.  Get appropriate
1110 	 * tokens.
1111 	 *
1112 	 * Don't try to deliver a generic signal to an exiting process,
1113 	 * the signal structures could be in flux.  We check the LWP later
1114 	 * on.
1115 	 */
1116 	PHOLD(p);
1117 	lwkt_gettoken(&p->p_token);
1118 	if (lp) {
1119 		LWPHOLD(lp);
1120 		lwkt_gettoken(&lp->lwp_token);
1121 	} else if (p->p_flags & P_WEXIT) {
1122 		goto out;
1123 	}
1124 
1125 	prop = sigprop(sig);
1126 
1127 	/*
1128 	 * If proc is traced, always give parent a chance;
1129 	 * if signal event is tracked by procfs, give *that*
1130 	 * a chance, as well.
1131 	 */
1132 	if ((p->p_flags & P_TRACED) || (p->p_stops & S_SIG)) {
1133 		action = SIG_DFL;
1134 	} else {
1135 		/*
1136 		 * Do not try to deliver signals to an exiting lwp.  Note
1137 		 * that we must still deliver the signal if P_WEXIT is set
1138 		 * in the process flags.
1139 		 */
1140 		if (lp && (lp->lwp_mpflags & LWP_MP_WEXIT)) {
1141 			if (lp) {
1142 				lwkt_reltoken(&lp->lwp_token);
1143 				LWPRELE(lp);
1144 			}
1145 			lwkt_reltoken(&p->p_token);
1146 			PRELE(p);
1147 			return;
1148 		}
1149 
1150 		/*
1151 		 * If the signal is being ignored, then we forget about
1152 		 * it immediately.  NOTE: We don't set SIGCONT in p_sigignore,
1153 		 * and if it is set to SIG_IGN, action will be SIG_DFL here.
1154 		 */
1155 		if (SIGISMEMBER(p->p_sigignore, sig)) {
1156 			/*
1157 			 * Even if a signal is set SIG_IGN, it may still be
1158 			 * lurking in a kqueue.
1159 			 */
1160 			KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1161 			if (lp) {
1162 				lwkt_reltoken(&lp->lwp_token);
1163 				LWPRELE(lp);
1164 			}
1165 			lwkt_reltoken(&p->p_token);
1166 			PRELE(p);
1167 			return;
1168 		}
1169 		if (SIGISMEMBER(p->p_sigcatch, sig))
1170 			action = SIG_CATCH;
1171 		else
1172 			action = SIG_DFL;
1173 	}
1174 
1175 	/*
1176 	 * If continuing, clear any pending STOP signals.
1177 	 */
1178 	if (prop & SA_CONT)
1179 		SIG_STOPSIGMASK(p->p_siglist);
1180 
1181 	if (prop & SA_STOP) {
1182 		/*
1183 		 * If sending a tty stop signal to a member of an orphaned
1184 		 * process group, discard the signal here if the action
1185 		 * is default; don't stop the process below if sleeping,
1186 		 * and don't clear any pending SIGCONT.
1187 		 */
1188 		if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0 &&
1189 		    action == SIG_DFL) {
1190 			lwkt_reltoken(&p->p_token);
1191 			PRELE(p);
1192 		        return;
1193 		}
1194 		SIG_CONTSIGMASK(p->p_siglist);
1195 		p->p_flags &= ~P_CONTINUED;
1196 	}
1197 
1198 	if (p->p_stat == SSTOP) {
1199 		/*
1200 		 * Nobody can handle this signal, add it to the lwp or
1201 		 * process pending list
1202 		 */
1203 		if (lp) {
1204 			spin_lock(&lp->lwp_spin);
1205 			SIGADDSET(lp->lwp_siglist, sig);
1206 			spin_unlock(&lp->lwp_spin);
1207 		} else {
1208 			SIGADDSET(p->p_siglist, sig);
1209 		}
1210 
1211 		/*
1212 		 * If the process is stopped and is being traced, then no
1213 		 * further action is necessary.
1214 		 */
1215 		if (p->p_flags & P_TRACED)
1216 			goto out;
1217 
1218 		/*
1219 		 * If the process is stopped and receives a KILL signal,
1220 		 * make the process runnable.
1221 		 */
1222 		if (sig == SIGKILL) {
1223 			proc_unstop(p);
1224 			goto active_process;
1225 		}
1226 
1227 		/*
1228 		 * If the process is stopped and receives a CONT signal,
1229 		 * then try to make the process runnable again.
1230 		 */
1231 		if (prop & SA_CONT) {
1232 			/*
1233 			 * If SIGCONT is default (or ignored), we continue the
1234 			 * process but don't leave the signal in p_siglist, as
1235 			 * it has no further action.  If SIGCONT is held, we
1236 			 * continue the process and leave the signal in
1237 			 * p_siglist.  If the process catches SIGCONT, let it
1238 			 * handle the signal itself.
1239 			 *
1240 			 * XXX what if the signal is being held blocked?
1241 			 *
1242 			 * Token required to interlock kern_wait().
1243 			 * Reparenting can also cause a race so we have to
1244 			 * hold (q).
1245 			 */
1246 			q = p->p_pptr;
1247 			PHOLD(q);
1248 			lwkt_gettoken(&q->p_token);
1249 			p->p_flags |= P_CONTINUED;
1250 			wakeup(q);
1251 			if (action == SIG_DFL)
1252 				SIGDELSET(p->p_siglist, sig);
1253 			proc_unstop(p);
1254 			lwkt_reltoken(&q->p_token);
1255 			PRELE(q);
1256 			if (action == SIG_CATCH)
1257 				goto active_process;
1258 			goto out;
1259 		}
1260 
1261 		/*
1262 		 * If the process is stopped and receives another STOP
1263 		 * signal, we do not need to stop it again.  If we did
1264 		 * the shell could get confused.
1265 		 *
1266 		 * However, if the current/preempted lwp is part of the
1267 		 * process receiving the signal, we need to keep it,
1268 		 * so that this lwp can stop in issignal() later, as
1269 		 * we don't want to wait until it reaches userret!
1270 		 */
1271 		if (prop & SA_STOP) {
1272 			if (lwkt_preempted_proc() == NULL ||
1273 			    lwkt_preempted_proc()->lwp_proc != p)
1274 				SIGDELSET(p->p_siglist, sig);
1275 		}
1276 
1277 		/*
1278 		 * Otherwise the process is stopped and it received some
1279 		 * signal, which does not change its stopped state.  When
1280 		 * the process is continued a wakeup(p) will be issued which
1281 		 * will wakeup any threads sleeping in tstop().
1282 		 */
1283 		if (lp == NULL) {
1284 			/* NOTE: returns lp w/ token held */
1285 			lp = find_lwp_for_signal(p, sig);
1286 		}
1287 		goto out;
1288 
1289 		/* NOTREACHED */
1290 	}
1291 	/* else not stopped */
1292 active_process:
1293 
1294 	/*
1295 	 * Never deliver a lwp-specific signal to a random lwp.
1296 	 */
1297 	if (lp == NULL) {
1298 		/* NOTE: returns lp w/ token held */
1299 		lp = find_lwp_for_signal(p, sig);
1300 		if (lp) {
1301 			if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
1302 				lwkt_reltoken(&lp->lwp_token);
1303 				LWPRELE(lp);
1304 				lp = NULL;
1305 			}
1306 		}
1307 	}
1308 
1309 	/*
1310 	 * Deliver to the process generically if (1) the signal is being
1311 	 * sent to any thread or (2) we could not find a thread to deliver
1312 	 * it to.
1313 	 */
1314 	if (lp == NULL) {
1315 		SIGADDSET(p->p_siglist, sig);
1316 		goto out;
1317 	}
1318 
1319 	/*
1320 	 * Deliver to a specific LWP whether it masks it or not.  It will
1321 	 * not be dispatched if masked but we must still deliver it.
1322 	 */
1323 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1324 	    (p->p_flags & P_TRACED) == 0) {
1325 		p->p_nice = NZERO;
1326 	}
1327 
1328 	/*
1329 	 * If the process receives a STOP signal which indeed needs to
1330 	 * stop the process, do so.  If the process chose to catch the
1331 	 * signal, it will be treated like any other signal.
1332 	 */
1333 	if ((prop & SA_STOP) && action == SIG_DFL) {
1334 		/*
1335 		 * If a child holding parent blocked, stopping
1336 		 * could cause deadlock.  Take no action at this
1337 		 * time.
1338 		 */
1339 		if (p->p_flags & P_PPWAIT) {
1340 			SIGADDSET(p->p_siglist, sig);
1341 			goto out;
1342 		}
1343 
1344 		/*
1345 		 * Do not actually try to manipulate the process, but simply
1346 		 * stop it.  Lwps will stop as soon as they safely can.
1347 		 *
1348 		 * Ignore stop if the process is exiting.
1349 		 */
1350 		if ((p->p_flags & P_WEXIT) == 0) {
1351 			p->p_xstat = sig;
1352 			proc_stop(p);
1353 		}
1354 		goto out;
1355 	}
1356 
1357 	/*
1358 	 * If it is a CONT signal with default action, just ignore it.
1359 	 */
1360 	if ((prop & SA_CONT) && action == SIG_DFL)
1361 		goto out;
1362 
1363 	/*
1364 	 * Mark signal pending at this specific thread.
1365 	 */
1366 	spin_lock(&lp->lwp_spin);
1367 	SIGADDSET(lp->lwp_siglist, sig);
1368 	spin_unlock(&lp->lwp_spin);
1369 
1370 	lwp_signotify(lp);
1371 
1372 out:
1373 	if (lp) {
1374 		lwkt_reltoken(&lp->lwp_token);
1375 		LWPRELE(lp);
1376 	}
1377 	lwkt_reltoken(&p->p_token);
1378 	PRELE(p);
1379 }
1380 
1381 /*
1382  * Notify the LWP that a signal has arrived.  The LWP does not have to be
1383  * sleeping on the current cpu.
1384  *
1385  * p->p_token and lp->lwp_token must be held on call.
1386  *
1387  * We can only safely schedule the thread on its current cpu and only if
1388  * one of the SINTR flags is set.  If an SINTR flag is set AND we are on
1389  * the correct cpu we are properly interlocked, otherwise we could be
1390  * racing other thread transition states (or the lwp is on the user scheduler
1391  * runq but not scheduled) and must not do anything.
1392  *
1393  * Since we hold the lwp token we know the lwp cannot be ripped out from
1394  * under us so we can safely hold it to prevent it from being ripped out
1395  * from under us if we are forced to IPI another cpu to make the local
1396  * checks there.
1397  *
1398  * Adjustment of lp->lwp_stat can only occur when we hold the lwp_token,
1399  * which we won't in an IPI so any fixups have to be done here, effectively
1400  * replicating part of what setrunnable() does.
1401  */
1402 static void
1403 lwp_signotify(struct lwp *lp)
1404 {
1405 	ASSERT_LWKT_TOKEN_HELD(&lp->lwp_proc->p_token);
1406 
1407 	crit_enter();
1408 	if (lp == lwkt_preempted_proc()) {
1409 		/*
1410 		 * lwp is on the current cpu AND it is currently running
1411 		 * (we preempted it).
1412 		 */
1413 		signotify();
1414 	} else if (lp->lwp_flags & LWP_SINTR) {
1415 		/*
1416 		 * lwp is sitting in tsleep() with PCATCH set
1417 		 */
1418 		if (lp->lwp_thread->td_gd == mycpu) {
1419 			setrunnable(lp);
1420 		} else {
1421 			/*
1422 			 * We can only adjust lwp_stat while we hold the
1423 			 * lwp_token, and we won't in the IPI function.
1424 			 */
1425 			LWPHOLD(lp);
1426 			if (lp->lwp_stat == LSSTOP)
1427 				lp->lwp_stat = LSSLEEP;
1428 			lwkt_send_ipiq(lp->lwp_thread->td_gd,
1429 				       lwp_signotify_remote, lp);
1430 		}
1431 	} else if (lp->lwp_thread->td_flags & TDF_SINTR) {
1432 		/*
1433 		 * lwp is sitting in lwkt_sleep() with PCATCH set.
1434 		 */
1435 		if (lp->lwp_thread->td_gd == mycpu) {
1436 			setrunnable(lp);
1437 		} else {
1438 			/*
1439 			 * We can only adjust lwp_stat while we hold the
1440 			 * lwp_token, and we won't in the IPI function.
1441 			 */
1442 			LWPHOLD(lp);
1443 			if (lp->lwp_stat == LSSTOP)
1444 				lp->lwp_stat = LSSLEEP;
1445 			lwkt_send_ipiq(lp->lwp_thread->td_gd,
1446 				       lwp_signotify_remote, lp);
1447 		}
1448 	} else {
1449 		/*
1450 		 * Otherwise the lwp is either in some uninterruptable state
1451 		 * or it is on the userland scheduler's runqueue waiting to
1452 		 * be scheduled to a cpu.
1453 		 */
1454 	}
1455 	crit_exit();
1456 }
1457 
1458 /*
1459  * This function is called via an IPI so we cannot call setrunnable() here
1460  * (because while we hold the lp we don't own its token, and can't get it
1461  * from an IPI).
1462  *
1463  * We are interlocked by virtue of being on the same cpu as the target.  If
1464  * we still are and LWP_SINTR or TDF_SINTR is set we can safely schedule
1465  * the target thread.
1466  */
1467 static void
1468 lwp_signotify_remote(void *arg)
1469 {
1470 	struct lwp *lp = arg;
1471 	thread_t td = lp->lwp_thread;
1472 
1473 	if (lp == lwkt_preempted_proc()) {
1474 		signotify();
1475 		LWPRELE(lp);
1476 	} else if (td->td_gd == mycpu) {
1477 		if ((lp->lwp_flags & LWP_SINTR) ||
1478 		    (td->td_flags & TDF_SINTR)) {
1479 			lwkt_schedule(td);
1480 		}
1481 		LWPRELE(lp);
1482 	} else {
1483 		lwkt_send_ipiq(td->td_gd, lwp_signotify_remote, lp);
1484 		/* LWPHOLD() is forwarded to the target cpu */
1485 	}
1486 }
1487 
1488 /*
1489  * Caller must hold p->p_token
1490  */
1491 void
1492 proc_stop(struct proc *p)
1493 {
1494 	struct proc *q;
1495 	struct lwp *lp;
1496 
1497 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1498 
1499 	/* If somebody raced us, be happy with it */
1500 	if (p->p_stat == SSTOP || p->p_stat == SZOMB) {
1501 		return;
1502 	}
1503 	p->p_stat = SSTOP;
1504 
1505 	FOREACH_LWP_IN_PROC(lp, p) {
1506 		LWPHOLD(lp);
1507 		lwkt_gettoken(&lp->lwp_token);
1508 
1509 		switch (lp->lwp_stat) {
1510 		case LSSTOP:
1511 			/*
1512 			 * Do nothing, we are already counted in
1513 			 * p_nstopped.
1514 			 */
1515 			break;
1516 
1517 		case LSSLEEP:
1518 			/*
1519 			 * We're sleeping, but we will stop before
1520 			 * returning to userspace, so count us
1521 			 * as stopped as well.  We set LWP_MP_WSTOP
1522 			 * to signal the lwp that it should not
1523 			 * increase p_nstopped when reaching tstop().
1524 			 *
1525 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1526 			 */
1527 			if ((lp->lwp_mpflags & LWP_MP_WSTOP) == 0) {
1528 				atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
1529 				++p->p_nstopped;
1530 			}
1531 			break;
1532 
1533 		case LSRUN:
1534 			/*
1535 			 * We might notify ourself, but that's not
1536 			 * a problem.
1537 			 */
1538 			lwp_signotify(lp);
1539 			break;
1540 		}
1541 		lwkt_reltoken(&lp->lwp_token);
1542 		LWPRELE(lp);
1543 	}
1544 
1545 	if (p->p_nstopped == p->p_nthreads) {
1546 		/*
1547 		 * Token required to interlock kern_wait().  Reparenting can
1548 		 * also cause a race so we have to hold (q).
1549 		 */
1550 		q = p->p_pptr;
1551 		PHOLD(q);
1552 		lwkt_gettoken(&q->p_token);
1553 		p->p_flags &= ~P_WAITED;
1554 		wakeup(q);
1555 		if ((q->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1556 			ksignal(p->p_pptr, SIGCHLD);
1557 		lwkt_reltoken(&q->p_token);
1558 		PRELE(q);
1559 	}
1560 }
1561 
1562 /*
1563  * Caller must hold proc_token
1564  */
1565 void
1566 proc_unstop(struct proc *p)
1567 {
1568 	struct lwp *lp;
1569 
1570 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1571 
1572 	if (p->p_stat != SSTOP)
1573 		return;
1574 
1575 	p->p_stat = SACTIVE;
1576 
1577 	FOREACH_LWP_IN_PROC(lp, p) {
1578 		LWPHOLD(lp);
1579 		lwkt_gettoken(&lp->lwp_token);
1580 
1581 		switch (lp->lwp_stat) {
1582 		case LSRUN:
1583 			/*
1584 			 * Uh?  Not stopped?  Well, I guess that's okay.
1585 			 */
1586 			if (bootverbose)
1587 				kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1588 					p->p_pid, lp->lwp_tid);
1589 			break;
1590 
1591 		case LSSLEEP:
1592 			/*
1593 			 * Still sleeping.  Don't bother waking it up.
1594 			 * However, if this thread was counted as
1595 			 * stopped, undo this.
1596 			 *
1597 			 * Nevertheless we call setrunnable() so that it
1598 			 * will wake up in case a signal or timeout arrived
1599 			 * in the meantime.
1600 			 *
1601 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1602 			 */
1603 			if (lp->lwp_mpflags & LWP_MP_WSTOP) {
1604 				atomic_clear_int(&lp->lwp_mpflags,
1605 						 LWP_MP_WSTOP);
1606 				--p->p_nstopped;
1607 			} else {
1608 				if (bootverbose)
1609 					kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1610 						p->p_pid, lp->lwp_tid);
1611 			}
1612 			/* FALLTHROUGH */
1613 
1614 		case LSSTOP:
1615 			/*
1616 			 * This handles any lwp's waiting in a tsleep with
1617 			 * SIGCATCH.
1618 			 */
1619 			lwp_signotify(lp);
1620 			break;
1621 
1622 		}
1623 		lwkt_reltoken(&lp->lwp_token);
1624 		LWPRELE(lp);
1625 	}
1626 
1627 	/*
1628 	 * This handles any lwp's waiting in tstop().  We have interlocked
1629 	 * the setting of p_stat by acquiring and releasing each lpw's
1630 	 * token.
1631 	 */
1632 	wakeup(p);
1633 }
1634 
1635 /*
1636  * No requirements.
1637  */
1638 static int
1639 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1640 {
1641 	sigset_t savedmask, set;
1642 	struct proc *p = curproc;
1643 	struct lwp *lp = curthread->td_lwp;
1644 	int error, sig, hz, timevalid = 0;
1645 	struct timespec rts, ets, ts;
1646 	struct timeval tv;
1647 
1648 	error = 0;
1649 	sig = 0;
1650 	ets.tv_sec = 0;		/* silence compiler warning */
1651 	ets.tv_nsec = 0;	/* silence compiler warning */
1652 	SIG_CANTMASK(waitset);
1653 	savedmask = lp->lwp_sigmask;
1654 
1655 	if (timeout) {
1656 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1657 		    timeout->tv_nsec < 1000000000) {
1658 			timevalid = 1;
1659 			getnanouptime(&rts);
1660 		 	ets = rts;
1661 			timespecadd(&ets, timeout);
1662 		}
1663 	}
1664 
1665 	for (;;) {
1666 		set = lwp_sigpend(lp);
1667 		SIGSETAND(set, waitset);
1668 		if ((sig = sig_ffs(&set)) != 0) {
1669 			SIGFILLSET(lp->lwp_sigmask);
1670 			SIGDELSET(lp->lwp_sigmask, sig);
1671 			SIG_CANTMASK(lp->lwp_sigmask);
1672 			sig = issignal(lp, 1);
1673 			/*
1674 			 * It may be a STOP signal, in the case, issignal
1675 			 * returns 0, because we may stop there, and new
1676 			 * signal can come in, we should restart if we got
1677 			 * nothing.
1678 			 */
1679 			if (sig == 0)
1680 				continue;
1681 			else
1682 				break;
1683 		}
1684 
1685 		/*
1686 		 * Previous checking got nothing, and we retried but still
1687 		 * got nothing, we should return the error status.
1688 		 */
1689 		if (error)
1690 			break;
1691 
1692 		/*
1693 		 * POSIX says this must be checked after looking for pending
1694 		 * signals.
1695 		 */
1696 		if (timeout) {
1697 			if (timevalid == 0) {
1698 				error = EINVAL;
1699 				break;
1700 			}
1701 			getnanouptime(&rts);
1702 			if (timespeccmp(&rts, &ets, >=)) {
1703 				error = EAGAIN;
1704 				break;
1705 			}
1706 			ts = ets;
1707 			timespecsub(&ts, &rts);
1708 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1709 			hz = tvtohz_high(&tv);
1710 		} else {
1711 			hz = 0;
1712 		}
1713 
1714 		lp->lwp_sigmask = savedmask;
1715 		SIGSETNAND(lp->lwp_sigmask, waitset);
1716 		/*
1717 		 * We won't ever be woken up.  Instead, our sleep will
1718 		 * be broken in lwpsignal().
1719 		 */
1720 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1721 		if (timeout) {
1722 			if (error == ERESTART) {
1723 				/* can not restart a timeout wait. */
1724 				error = EINTR;
1725 			} else if (error == EAGAIN) {
1726 				/* will calculate timeout by ourself. */
1727 				error = 0;
1728 			}
1729 		}
1730 		/* Retry ... */
1731 	}
1732 
1733 	lp->lwp_sigmask = savedmask;
1734 	if (sig) {
1735 		error = 0;
1736 		bzero(info, sizeof(*info));
1737 		info->si_signo = sig;
1738 		spin_lock(&lp->lwp_spin);
1739 		lwp_delsig(lp, sig);	/* take the signal! */
1740 		spin_unlock(&lp->lwp_spin);
1741 
1742 		if (sig == SIGKILL) {
1743 			sigexit(lp, sig);
1744 			/* NOT REACHED */
1745 		}
1746 	}
1747 
1748 	return (error);
1749 }
1750 
1751 /*
1752  * MPALMOSTSAFE
1753  */
1754 int
1755 sys_sigtimedwait(struct sigtimedwait_args *uap)
1756 {
1757 	struct timespec ts;
1758 	struct timespec *timeout;
1759 	sigset_t set;
1760 	siginfo_t info;
1761 	int error;
1762 
1763 	if (uap->timeout) {
1764 		error = copyin(uap->timeout, &ts, sizeof(ts));
1765 		if (error)
1766 			return (error);
1767 		timeout = &ts;
1768 	} else {
1769 		timeout = NULL;
1770 	}
1771 	error = copyin(uap->set, &set, sizeof(set));
1772 	if (error)
1773 		return (error);
1774 	error = kern_sigtimedwait(set, &info, timeout);
1775 	if (error)
1776 		return (error);
1777 	if (uap->info)
1778 		error = copyout(&info, uap->info, sizeof(info));
1779 	/* Repost if we got an error. */
1780 	/*
1781 	 * XXX lwp
1782 	 *
1783 	 * This could transform a thread-specific signal to another
1784 	 * thread / process pending signal.
1785 	 */
1786 	if (error) {
1787 		ksignal(curproc, info.si_signo);
1788 	} else {
1789 		uap->sysmsg_result = info.si_signo;
1790 	}
1791 	return (error);
1792 }
1793 
1794 /*
1795  * MPALMOSTSAFE
1796  */
1797 int
1798 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1799 {
1800 	siginfo_t info;
1801 	sigset_t set;
1802 	int error;
1803 
1804 	error = copyin(uap->set, &set, sizeof(set));
1805 	if (error)
1806 		return (error);
1807 	error = kern_sigtimedwait(set, &info, NULL);
1808 	if (error)
1809 		return (error);
1810 	if (uap->info)
1811 		error = copyout(&info, uap->info, sizeof(info));
1812 	/* Repost if we got an error. */
1813 	/*
1814 	 * XXX lwp
1815 	 *
1816 	 * This could transform a thread-specific signal to another
1817 	 * thread / process pending signal.
1818 	 */
1819 	if (error) {
1820 		ksignal(curproc, info.si_signo);
1821 	} else {
1822 		uap->sysmsg_result = info.si_signo;
1823 	}
1824 	return (error);
1825 }
1826 
1827 /*
1828  * If the current process has received a signal that would interrupt a
1829  * system call, return EINTR or ERESTART as appropriate.
1830  */
1831 int
1832 iscaught(struct lwp *lp)
1833 {
1834 	struct proc *p = lp->lwp_proc;
1835 	int sig;
1836 
1837 	if (p) {
1838 		if ((sig = CURSIG(lp)) != 0) {
1839 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1840 				return (EINTR);
1841 			return (ERESTART);
1842 		}
1843 	}
1844 	return(EWOULDBLOCK);
1845 }
1846 
1847 /*
1848  * If the current process has received a signal (should be caught or cause
1849  * termination, should interrupt current syscall), return the signal number.
1850  * Stop signals with default action are processed immediately, then cleared;
1851  * they aren't returned.  This is checked after each entry to the system for
1852  * a syscall or trap (though this can usually be done without calling issignal
1853  * by checking the pending signal masks in the CURSIG macro).
1854  *
1855  * This routine is called via CURSIG/__cursig.  We will acquire and release
1856  * p->p_token but if the caller needs to interlock the test the caller must
1857  * also hold p->p_token.
1858  *
1859  *	while (sig = CURSIG(curproc))
1860  *		postsig(sig);
1861  *
1862  * MPSAFE
1863  */
1864 int
1865 issignal(struct lwp *lp, int maytrace)
1866 {
1867 	struct proc *p = lp->lwp_proc;
1868 	sigset_t mask;
1869 	int sig, prop;
1870 
1871 	lwkt_gettoken(&p->p_token);
1872 
1873 	for (;;) {
1874 		int traced = (p->p_flags & P_TRACED) || (p->p_stops & S_SIG);
1875 
1876 		/*
1877 		 * If this process is supposed to stop, stop this thread.
1878 		 */
1879 		if (p->p_stat == SSTOP)
1880 			tstop();
1881 
1882 		mask = lwp_sigpend(lp);
1883 		SIGSETNAND(mask, lp->lwp_sigmask);
1884 		if (p->p_flags & P_PPWAIT)
1885 			SIG_STOPSIGMASK(mask);
1886 		if (SIGISEMPTY(mask)) {		/* no signal to send */
1887 			lwkt_reltoken(&p->p_token);
1888 			return (0);
1889 		}
1890 		sig = sig_ffs(&mask);
1891 
1892 		STOPEVENT(p, S_SIG, sig);
1893 
1894 		/*
1895 		 * We should see pending but ignored signals
1896 		 * only if P_TRACED was on when they were posted.
1897 		 */
1898 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1899 			spin_lock(&lp->lwp_spin);
1900 			lwp_delsig(lp, sig);
1901 			spin_unlock(&lp->lwp_spin);
1902 			continue;
1903 		}
1904 		if (maytrace &&
1905 		    (p->p_flags & P_TRACED) &&
1906 		    (p->p_flags & P_PPWAIT) == 0) {
1907 			/*
1908 			 * If traced, always stop, and stay stopped until
1909 			 * released by the parent.
1910 			 *
1911 			 * NOTE: SSTOP may get cleared during the loop,
1912 			 * but we do not re-notify the parent if we have
1913 			 * to loop several times waiting for the parent
1914 			 * to let us continue.
1915 			 *
1916 			 * XXX not sure if this is still true
1917 			 */
1918 			p->p_xstat = sig;
1919 			proc_stop(p);
1920 			do {
1921 				tstop();
1922 			} while (!trace_req(p) && (p->p_flags & P_TRACED));
1923 
1924 			/*
1925 			 * If parent wants us to take the signal,
1926 			 * then it will leave it in p->p_xstat;
1927 			 * otherwise we just look for signals again.
1928 			 */
1929 			spin_lock(&lp->lwp_spin);
1930 			lwp_delsig(lp, sig);	/* clear old signal */
1931 			spin_unlock(&lp->lwp_spin);
1932 			sig = p->p_xstat;
1933 			if (sig == 0)
1934 				continue;
1935 
1936 			/*
1937 			 * Put the new signal into p_siglist.  If the
1938 			 * signal is being masked, look for other signals.
1939 			 *
1940 			 * XXX lwp might need a call to ksignal()
1941 			 */
1942 			SIGADDSET(p->p_siglist, sig);
1943 			if (SIGISMEMBER(lp->lwp_sigmask, sig))
1944 				continue;
1945 
1946 			/*
1947 			 * If the traced bit got turned off, go back up
1948 			 * to the top to rescan signals.  This ensures
1949 			 * that p_sig* and ps_sigact are consistent.
1950 			 */
1951 			if ((p->p_flags & P_TRACED) == 0)
1952 				continue;
1953 		}
1954 
1955 		prop = sigprop(sig);
1956 
1957 		/*
1958 		 * Decide whether the signal should be returned.
1959 		 * Return the signal's number, or fall through
1960 		 * to clear it from the pending mask.
1961 		 */
1962 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
1963 		case (intptr_t)SIG_DFL:
1964 			/*
1965 			 * Don't take default actions on system processes.
1966 			 */
1967 			if (p->p_pid <= 1) {
1968 #ifdef DIAGNOSTIC
1969 				/*
1970 				 * Are you sure you want to ignore SIGSEGV
1971 				 * in init? XXX
1972 				 */
1973 				kprintf("Process (pid %lu) got signal %d\n",
1974 					(u_long)p->p_pid, sig);
1975 #endif
1976 				break;		/* == ignore */
1977 			}
1978 
1979 			/*
1980 			 * Handle the in-kernel checkpoint action
1981 			 */
1982 			if (prop & SA_CKPT) {
1983 				checkpoint_signal_handler(lp);
1984 				break;
1985 			}
1986 
1987 			/*
1988 			 * If there is a pending stop signal to process
1989 			 * with default action, stop here,
1990 			 * then clear the signal.  However,
1991 			 * if process is member of an orphaned
1992 			 * process group, ignore tty stop signals.
1993 			 */
1994 			if (prop & SA_STOP) {
1995 				if (p->p_flags & P_TRACED ||
1996 		    		    (p->p_pgrp->pg_jobc == 0 &&
1997 				    prop & SA_TTYSTOP))
1998 					break;	/* == ignore */
1999 				if ((p->p_flags & P_WEXIT) == 0) {
2000 					p->p_xstat = sig;
2001 					proc_stop(p);
2002 					tstop();
2003 				}
2004 				break;
2005 			} else if (prop & SA_IGNORE) {
2006 				/*
2007 				 * Except for SIGCONT, shouldn't get here.
2008 				 * Default action is to ignore; drop it.
2009 				 */
2010 				break;		/* == ignore */
2011 			} else {
2012 				lwkt_reltoken(&p->p_token);
2013 				return (sig);
2014 			}
2015 
2016 			/*NOTREACHED*/
2017 
2018 		case (intptr_t)SIG_IGN:
2019 			/*
2020 			 * Masking above should prevent us ever trying
2021 			 * to take action on an ignored signal other
2022 			 * than SIGCONT, unless process is traced.
2023 			 */
2024 			if ((prop & SA_CONT) == 0 &&
2025 			    (p->p_flags & P_TRACED) == 0)
2026 				kprintf("issignal\n");
2027 			break;		/* == ignore */
2028 
2029 		default:
2030 			/*
2031 			 * This signal has an action, let
2032 			 * postsig() process it.
2033 			 */
2034 			lwkt_reltoken(&p->p_token);
2035 			return (sig);
2036 		}
2037 		spin_lock(&lp->lwp_spin);
2038 		lwp_delsig(lp, sig);		/* take the signal! */
2039 		spin_unlock(&lp->lwp_spin);
2040 	}
2041 	/* NOTREACHED */
2042 }
2043 
2044 /*
2045  * Take the action for the specified signal
2046  * from the current set of pending signals.
2047  *
2048  * Caller must hold p->p_token
2049  */
2050 void
2051 postsig(int sig)
2052 {
2053 	struct lwp *lp = curthread->td_lwp;
2054 	struct proc *p = lp->lwp_proc;
2055 	struct sigacts *ps = p->p_sigacts;
2056 	sig_t action;
2057 	sigset_t returnmask;
2058 	int code;
2059 
2060 	KASSERT(sig != 0, ("postsig"));
2061 
2062 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
2063 
2064 	/*
2065 	 * If we are a virtual kernel running an emulated user process
2066 	 * context, switch back to the virtual kernel context before
2067 	 * trying to post the signal.
2068 	 */
2069 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
2070 		struct trapframe *tf = lp->lwp_md.md_regs;
2071 		tf->tf_trapno = 0;
2072 		vkernel_trap(lp, tf);
2073 	}
2074 
2075 	spin_lock(&lp->lwp_spin);
2076 	lwp_delsig(lp, sig);
2077 	spin_unlock(&lp->lwp_spin);
2078 	action = ps->ps_sigact[_SIG_IDX(sig)];
2079 #ifdef KTRACE
2080 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
2081 		ktrpsig(lp, sig, action, lp->lwp_flags & LWP_OLDMASK ?
2082 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
2083 #endif
2084 	STOPEVENT(p, S_SIG, sig);
2085 
2086 	if (action == SIG_DFL) {
2087 		/*
2088 		 * Default action, where the default is to kill
2089 		 * the process.  (Other cases were ignored above.)
2090 		 */
2091 		sigexit(lp, sig);
2092 		/* NOTREACHED */
2093 	} else {
2094 		/*
2095 		 * If we get here, the signal must be caught.
2096 		 */
2097 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
2098 		    ("postsig action"));
2099 
2100 		/*
2101 		 * Reset the signal handler if asked to
2102 		 */
2103 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
2104 			/*
2105 			 * See kern_sigaction() for origin of this code.
2106 			 */
2107 			SIGDELSET(p->p_sigcatch, sig);
2108 			if (sig != SIGCONT &&
2109 			    sigprop(sig) & SA_IGNORE)
2110 				SIGADDSET(p->p_sigignore, sig);
2111 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2112 		}
2113 
2114 		/*
2115 		 * Set the signal mask and calculate the mask to restore
2116 		 * when the signal function returns.
2117 		 *
2118 		 * Special case: user has done a sigsuspend.  Here the
2119 		 * current mask is not of interest, but rather the
2120 		 * mask from before the sigsuspend is what we want
2121 		 * restored after the signal processing is completed.
2122 		 */
2123 		if (lp->lwp_flags & LWP_OLDMASK) {
2124 			returnmask = lp->lwp_oldsigmask;
2125 			lp->lwp_flags &= ~LWP_OLDMASK;
2126 		} else {
2127 			returnmask = lp->lwp_sigmask;
2128 		}
2129 
2130 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
2131 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
2132 			SIGADDSET(lp->lwp_sigmask, sig);
2133 
2134 		lp->lwp_ru.ru_nsignals++;
2135 		if (lp->lwp_sig != sig) {
2136 			code = 0;
2137 		} else {
2138 			code = lp->lwp_code;
2139 			lp->lwp_code = 0;
2140 			lp->lwp_sig = 0;
2141 		}
2142 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
2143 	}
2144 }
2145 
2146 /*
2147  * Kill the current process for stated reason.
2148  */
2149 void
2150 killproc(struct proc *p, char *why)
2151 {
2152 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
2153 		p->p_pid, p->p_comm,
2154 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2155 	ksignal(p, SIGKILL);
2156 }
2157 
2158 /*
2159  * Force the current process to exit with the specified signal, dumping core
2160  * if appropriate.  We bypass the normal tests for masked and caught signals,
2161  * allowing unrecoverable failures to terminate the process without changing
2162  * signal state.  Mark the accounting record with the signal termination.
2163  * If dumping core, save the signal number for the debugger.  Calls exit and
2164  * does not return.
2165  *
2166  * This routine does not return.
2167  */
2168 void
2169 sigexit(struct lwp *lp, int sig)
2170 {
2171 	struct proc *p = lp->lwp_proc;
2172 
2173 	lwkt_gettoken(&p->p_token);
2174 	p->p_acflag |= AXSIG;
2175 	if (sigprop(sig) & SA_CORE) {
2176 		lp->lwp_sig = sig;
2177 		/*
2178 		 * Log signals which would cause core dumps
2179 		 * (Log as LOG_INFO to appease those who don't want
2180 		 * these messages.)
2181 		 * XXX : Todo, as well as euid, write out ruid too
2182 		 */
2183 		if (coredump(lp, sig) == 0)
2184 			sig |= WCOREFLAG;
2185 		if (kern_logsigexit)
2186 			log(LOG_INFO,
2187 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2188 			    p->p_pid, p->p_comm,
2189 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
2190 			    sig &~ WCOREFLAG,
2191 			    sig & WCOREFLAG ? " (core dumped)" : "");
2192 	}
2193 	lwkt_reltoken(&p->p_token);
2194 	exit1(W_EXITCODE(0, sig));
2195 	/* NOTREACHED */
2196 }
2197 
2198 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
2199 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2200 	      sizeof(corefilename), "process corefile name format string");
2201 
2202 /*
2203  * expand_name(name, uid, pid)
2204  * Expand the name described in corefilename, using name, uid, and pid.
2205  * corefilename is a kprintf-like string, with three format specifiers:
2206  *	%N	name of process ("name")
2207  *	%P	process id (pid)
2208  *	%U	user id (uid)
2209  * For example, "%N.core" is the default; they can be disabled completely
2210  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2211  * This is controlled by the sysctl variable kern.corefile (see above).
2212  */
2213 
2214 static char *
2215 expand_name(const char *name, uid_t uid, pid_t pid)
2216 {
2217 	char *temp;
2218 	char buf[11];		/* Buffer for pid/uid -- max 4B */
2219 	int i, n;
2220 	char *format = corefilename;
2221 	size_t namelen;
2222 
2223 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
2224 	if (temp == NULL)
2225 		return NULL;
2226 	namelen = strlen(name);
2227 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2228 		int l;
2229 		switch (format[i]) {
2230 		case '%':	/* Format character */
2231 			i++;
2232 			switch (format[i]) {
2233 			case '%':
2234 				temp[n++] = '%';
2235 				break;
2236 			case 'N':	/* process name */
2237 				if ((n + namelen) > MAXPATHLEN) {
2238 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2239 					    pid, name, uid, temp, name);
2240 					kfree(temp, M_TEMP);
2241 					return NULL;
2242 				}
2243 				memcpy(temp+n, name, namelen);
2244 				n += namelen;
2245 				break;
2246 			case 'P':	/* process id */
2247 				l = ksprintf(buf, "%u", pid);
2248 				if ((n + l) > MAXPATHLEN) {
2249 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2250 					    pid, name, uid, temp, name);
2251 					kfree(temp, M_TEMP);
2252 					return NULL;
2253 				}
2254 				memcpy(temp+n, buf, l);
2255 				n += l;
2256 				break;
2257 			case 'U':	/* user id */
2258 				l = ksprintf(buf, "%u", uid);
2259 				if ((n + l) > MAXPATHLEN) {
2260 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2261 					    pid, name, uid, temp, name);
2262 					kfree(temp, M_TEMP);
2263 					return NULL;
2264 				}
2265 				memcpy(temp+n, buf, l);
2266 				n += l;
2267 				break;
2268 			default:
2269 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2270 			}
2271 			break;
2272 		default:
2273 			temp[n++] = format[i];
2274 		}
2275 	}
2276 	temp[n] = '\0';
2277 	return temp;
2278 }
2279 
2280 /*
2281  * Dump a process' core.  The main routine does some
2282  * policy checking, and creates the name of the coredump;
2283  * then it passes on a vnode and a size limit to the process-specific
2284  * coredump routine if there is one; if there _is not_ one, it returns
2285  * ENOSYS; otherwise it returns the error from the process-specific routine.
2286  *
2287  * The parameter `lp' is the lwp which triggered the coredump.
2288  */
2289 
2290 static int
2291 coredump(struct lwp *lp, int sig)
2292 {
2293 	struct proc *p = lp->lwp_proc;
2294 	struct vnode *vp;
2295 	struct ucred *cred = p->p_ucred;
2296 	struct flock lf;
2297 	struct nlookupdata nd;
2298 	struct vattr vattr;
2299 	int error, error1;
2300 	char *name;			/* name of corefile */
2301 	off_t limit;
2302 
2303 	STOPEVENT(p, S_CORE, 0);
2304 
2305 	if (((sugid_coredump == 0) && p->p_flags & P_SUGID) || do_coredump == 0)
2306 		return (EFAULT);
2307 
2308 	/*
2309 	 * Note that the bulk of limit checking is done after
2310 	 * the corefile is created.  The exception is if the limit
2311 	 * for corefiles is 0, in which case we don't bother
2312 	 * creating the corefile at all.  This layout means that
2313 	 * a corefile is truncated instead of not being created,
2314 	 * if it is larger than the limit.
2315 	 */
2316 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2317 	if (limit == 0)
2318 		return EFBIG;
2319 
2320 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2321 	if (name == NULL)
2322 		return (EINVAL);
2323 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2324 	if (error == 0)
2325 		error = vn_open(&nd, NULL, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
2326 	kfree(name, M_TEMP);
2327 	if (error) {
2328 		nlookup_done(&nd);
2329 		return (error);
2330 	}
2331 	vp = nd.nl_open_vp;
2332 	nd.nl_open_vp = NULL;
2333 	nlookup_done(&nd);
2334 
2335 	vn_unlock(vp);
2336 	lf.l_whence = SEEK_SET;
2337 	lf.l_start = 0;
2338 	lf.l_len = 0;
2339 	lf.l_type = F_WRLCK;
2340 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2341 	if (error)
2342 		goto out2;
2343 
2344 	/* Don't dump to non-regular files or files with links. */
2345 	if (vp->v_type != VREG ||
2346 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2347 		error = EFAULT;
2348 		goto out1;
2349 	}
2350 
2351 	/* Don't dump to files current user does not own */
2352 	if (vattr.va_uid != p->p_ucred->cr_uid) {
2353 		error = EFAULT;
2354 		goto out1;
2355 	}
2356 
2357 	VATTR_NULL(&vattr);
2358 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2359 	vattr.va_size = 0;
2360 	VOP_SETATTR(vp, &vattr, cred);
2361 	p->p_acflag |= ACORE;
2362 	vn_unlock(vp);
2363 
2364 	error = p->p_sysent->sv_coredump ?
2365 		  p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2366 
2367 out1:
2368 	lf.l_type = F_UNLCK;
2369 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2370 out2:
2371 	error1 = vn_close(vp, FWRITE);
2372 	if (error == 0)
2373 		error = error1;
2374 	return (error);
2375 }
2376 
2377 /*
2378  * Nonexistent system call-- signal process (may want to handle it).
2379  * Flag error in case process won't see signal immediately (blocked or ignored).
2380  *
2381  * MPALMOSTSAFE
2382  */
2383 /* ARGSUSED */
2384 int
2385 sys_nosys(struct nosys_args *args)
2386 {
2387 	lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2388 	return (EINVAL);
2389 }
2390 
2391 /*
2392  * Send a SIGIO or SIGURG signal to a process or process group using
2393  * stored credentials rather than those of the current process.
2394  */
2395 void
2396 pgsigio(struct sigio *sigio, int sig, int checkctty)
2397 {
2398 	if (sigio == NULL)
2399 		return;
2400 
2401 	if (sigio->sio_pgid > 0) {
2402 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2403 		             sigio->sio_proc))
2404 			ksignal(sigio->sio_proc, sig);
2405 	} else if (sigio->sio_pgid < 0) {
2406 		struct proc *p;
2407 		struct pgrp *pg = sigio->sio_pgrp;
2408 
2409 		/*
2410 		 * Must interlock all signals against fork
2411 		 */
2412 		pgref(pg);
2413 		lockmgr(&pg->pg_lock, LK_EXCLUSIVE);
2414 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
2415 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2416 			    (checkctty == 0 || (p->p_flags & P_CONTROLT)))
2417 				ksignal(p, sig);
2418 		}
2419 		lockmgr(&pg->pg_lock, LK_RELEASE);
2420 		pgrel(pg);
2421 	}
2422 }
2423 
2424 static int
2425 filt_sigattach(struct knote *kn)
2426 {
2427 	struct proc *p = curproc;
2428 
2429 	kn->kn_ptr.p_proc = p;
2430 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
2431 
2432 	/* XXX lock the proc here while adding to the list? */
2433 	knote_insert(&p->p_klist, kn);
2434 
2435 	return (0);
2436 }
2437 
2438 static void
2439 filt_sigdetach(struct knote *kn)
2440 {
2441 	struct proc *p = kn->kn_ptr.p_proc;
2442 
2443 	knote_remove(&p->p_klist, kn);
2444 }
2445 
2446 /*
2447  * signal knotes are shared with proc knotes, so we apply a mask to
2448  * the hint in order to differentiate them from process hints.  This
2449  * could be avoided by using a signal-specific knote list, but probably
2450  * isn't worth the trouble.
2451  */
2452 static int
2453 filt_signal(struct knote *kn, long hint)
2454 {
2455 	if (hint & NOTE_SIGNAL) {
2456 		hint &= ~NOTE_SIGNAL;
2457 
2458 		if (kn->kn_id == hint)
2459 			kn->kn_data++;
2460 	}
2461 	return (kn->kn_data != 0);
2462 }
2463