xref: /dragonfly/sys/kern/kern_sig.c (revision a32bc35d)
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 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
325 				ps->ps_flag |= PS_CLDSIGIGN;
326 			else
327 				ps->ps_flag &= ~PS_CLDSIGIGN;
328 		}
329 		/*
330 		 * Set bit in p_sigignore for signals that are set to SIG_IGN,
331 		 * and for signals set to SIG_DFL where the default is to
332 		 * ignore. However, don't put SIGCONT in p_sigignore, as we
333 		 * have to restart the process.
334 		 *
335 		 * Also remove the signal from the process and lwp signal
336 		 * list.
337 		 */
338 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
339 		    (sigprop(sig) & SA_IGNORE &&
340 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
341 			SIGDELSET(p->p_siglist, sig);
342 			FOREACH_LWP_IN_PROC(lp, p) {
343 				spin_lock(&lp->lwp_spin);
344 				SIGDELSET(lp->lwp_siglist, sig);
345 				spin_unlock(&lp->lwp_spin);
346 			}
347 			if (sig != SIGCONT) {
348 				/* easier in ksignal */
349 				SIGADDSET(p->p_sigignore, sig);
350 			}
351 			SIGDELSET(p->p_sigcatch, sig);
352 		} else {
353 			SIGDELSET(p->p_sigignore, sig);
354 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
355 				SIGDELSET(p->p_sigcatch, sig);
356 			else
357 				SIGADDSET(p->p_sigcatch, sig);
358 		}
359 	}
360 	lwkt_reltoken(&p->p_token);
361 	return (0);
362 }
363 
364 int
365 sys_sigaction(struct sigaction_args *uap)
366 {
367 	struct sigaction act, oact;
368 	struct sigaction *actp, *oactp;
369 	int error;
370 
371 	actp = (uap->act != NULL) ? &act : NULL;
372 	oactp = (uap->oact != NULL) ? &oact : NULL;
373 	if (actp) {
374 		error = copyin(uap->act, actp, sizeof(act));
375 		if (error)
376 			return (error);
377 	}
378 	error = kern_sigaction(uap->sig, actp, oactp);
379 	if (oactp && !error) {
380 		error = copyout(oactp, uap->oact, sizeof(oact));
381 	}
382 	return (error);
383 }
384 
385 /*
386  * Initialize signal state for process 0;
387  * set to ignore signals that are ignored by default.
388  */
389 void
390 siginit(struct proc *p)
391 {
392 	int i;
393 
394 	for (i = 1; i <= NSIG; i++)
395 		if (sigprop(i) & SA_IGNORE && i != SIGCONT)
396 			SIGADDSET(p->p_sigignore, i);
397 }
398 
399 /*
400  * Reset signals for an exec of the specified process.
401  */
402 void
403 execsigs(struct proc *p)
404 {
405 	struct sigacts *ps = p->p_sigacts;
406 	struct lwp *lp;
407 	int sig;
408 
409 	lp = ONLY_LWP_IN_PROC(p);
410 
411 	/*
412 	 * Reset caught signals.  Held signals remain held
413 	 * through p_sigmask (unless they were caught,
414 	 * and are now ignored by default).
415 	 */
416 	while (SIGNOTEMPTY(p->p_sigcatch)) {
417 		sig = sig_ffs(&p->p_sigcatch);
418 		SIGDELSET(p->p_sigcatch, sig);
419 		if (sigprop(sig) & SA_IGNORE) {
420 			if (sig != SIGCONT)
421 				SIGADDSET(p->p_sigignore, sig);
422 			SIGDELSET(p->p_siglist, sig);
423 			/* don't need spinlock */
424 			SIGDELSET(lp->lwp_siglist, sig);
425 		}
426 		ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
427 	}
428 
429 	/*
430 	 * Reset stack state to the user stack.
431 	 * Clear set of signals caught on the signal stack.
432 	 */
433 	lp->lwp_sigstk.ss_flags = SS_DISABLE;
434 	lp->lwp_sigstk.ss_size = 0;
435 	lp->lwp_sigstk.ss_sp = 0;
436 	lp->lwp_flags &= ~LWP_ALTSTACK;
437 	/*
438 	 * Reset no zombies if child dies flag as Solaris does.
439 	 */
440 	p->p_sigacts->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
441 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
442 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
443 }
444 
445 /*
446  * kern_sigprocmask() - MP SAFE ONLY IF p == curproc
447  *
448  *	Manipulate signal mask.  This routine is MP SAFE *ONLY* if
449  *	p == curproc.
450  */
451 int
452 kern_sigprocmask(int how, sigset_t *set, sigset_t *oset)
453 {
454 	struct thread *td = curthread;
455 	struct lwp *lp = td->td_lwp;
456 	struct proc *p = td->td_proc;
457 	int error;
458 
459 	lwkt_gettoken(&p->p_token);
460 
461 	if (oset != NULL)
462 		*oset = lp->lwp_sigmask;
463 
464 	error = 0;
465 	if (set != NULL) {
466 		switch (how) {
467 		case SIG_BLOCK:
468 			SIG_CANTMASK(*set);
469 			SIGSETOR(lp->lwp_sigmask, *set);
470 			break;
471 		case SIG_UNBLOCK:
472 			SIGSETNAND(lp->lwp_sigmask, *set);
473 			break;
474 		case SIG_SETMASK:
475 			SIG_CANTMASK(*set);
476 			lp->lwp_sigmask = *set;
477 			break;
478 		default:
479 			error = EINVAL;
480 			break;
481 		}
482 	}
483 
484 	lwkt_reltoken(&p->p_token);
485 
486 	return (error);
487 }
488 
489 /*
490  * sigprocmask()
491  *
492  * MPSAFE
493  */
494 int
495 sys_sigprocmask(struct sigprocmask_args *uap)
496 {
497 	sigset_t set, oset;
498 	sigset_t *setp, *osetp;
499 	int error;
500 
501 	setp = (uap->set != NULL) ? &set : NULL;
502 	osetp = (uap->oset != NULL) ? &oset : NULL;
503 	if (setp) {
504 		error = copyin(uap->set, setp, sizeof(set));
505 		if (error)
506 			return (error);
507 	}
508 	error = kern_sigprocmask(uap->how, setp, osetp);
509 	if (osetp && !error) {
510 		error = copyout(osetp, uap->oset, sizeof(oset));
511 	}
512 	return (error);
513 }
514 
515 /*
516  * MPSAFE
517  */
518 int
519 kern_sigpending(struct __sigset *set)
520 {
521 	struct lwp *lp = curthread->td_lwp;
522 
523 	*set = lwp_sigpend(lp);
524 
525 	return (0);
526 }
527 
528 /*
529  * MPSAFE
530  */
531 int
532 sys_sigpending(struct sigpending_args *uap)
533 {
534 	sigset_t set;
535 	int error;
536 
537 	error = kern_sigpending(&set);
538 
539 	if (error == 0)
540 		error = copyout(&set, uap->set, sizeof(set));
541 	return (error);
542 }
543 
544 /*
545  * Suspend process until signal, providing mask to be set
546  * in the meantime.
547  *
548  * MPSAFE
549  */
550 int
551 kern_sigsuspend(struct __sigset *set)
552 {
553 	struct thread *td = curthread;
554 	struct lwp *lp = td->td_lwp;
555 	struct proc *p = td->td_proc;
556 	struct sigacts *ps = p->p_sigacts;
557 
558 	/*
559 	 * When returning from sigsuspend, we want
560 	 * the old mask to be restored after the
561 	 * signal handler has finished.  Thus, we
562 	 * save it here and mark the sigacts structure
563 	 * to indicate this.
564 	 */
565 	lp->lwp_oldsigmask = lp->lwp_sigmask;
566 	lp->lwp_flags |= LWP_OLDMASK;
567 
568 	SIG_CANTMASK(*set);
569 	lp->lwp_sigmask = *set;
570 	while (tsleep(ps, PCATCH, "pause", 0) == 0)
571 		/* void */;
572 	/* always return EINTR rather than ERESTART... */
573 	return (EINTR);
574 }
575 
576 /*
577  * Note nonstandard calling convention: libc stub passes mask, not
578  * pointer, to save a copyin.
579  *
580  * MPSAFE
581  */
582 int
583 sys_sigsuspend(struct sigsuspend_args *uap)
584 {
585 	sigset_t mask;
586 	int error;
587 
588 	error = copyin(uap->sigmask, &mask, sizeof(mask));
589 	if (error)
590 		return (error);
591 
592 	error = kern_sigsuspend(&mask);
593 
594 	return (error);
595 }
596 
597 /*
598  * MPSAFE
599  */
600 int
601 kern_sigaltstack(struct sigaltstack *ss, struct sigaltstack *oss)
602 {
603 	struct thread *td = curthread;
604 	struct lwp *lp = td->td_lwp;
605 	struct proc *p = td->td_proc;
606 
607 	if ((lp->lwp_flags & LWP_ALTSTACK) == 0)
608 		lp->lwp_sigstk.ss_flags |= SS_DISABLE;
609 
610 	if (oss)
611 		*oss = lp->lwp_sigstk;
612 
613 	if (ss) {
614 		if (ss->ss_flags & SS_DISABLE) {
615 			if (lp->lwp_sigstk.ss_flags & SS_ONSTACK)
616 				return (EINVAL);
617 			lp->lwp_flags &= ~LWP_ALTSTACK;
618 			lp->lwp_sigstk.ss_flags = ss->ss_flags;
619 		} else {
620 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
621 				return (ENOMEM);
622 			lp->lwp_flags |= LWP_ALTSTACK;
623 			lp->lwp_sigstk = *ss;
624 		}
625 	}
626 
627 	return (0);
628 }
629 
630 /*
631  * MPSAFE
632  */
633 int
634 sys_sigaltstack(struct sigaltstack_args *uap)
635 {
636 	stack_t ss, oss;
637 	int error;
638 
639 	if (uap->ss) {
640 		error = copyin(uap->ss, &ss, sizeof(ss));
641 		if (error)
642 			return (error);
643 	}
644 
645 	error = kern_sigaltstack(uap->ss ? &ss : NULL,
646 	    uap->oss ? &oss : NULL);
647 
648 	if (error == 0 && uap->oss)
649 		error = copyout(&oss, uap->oss, sizeof(*uap->oss));
650 	return (error);
651 }
652 
653 /*
654  * Common code for kill process group/broadcast kill.
655  * cp is calling process.
656  */
657 struct killpg_info {
658 	int nfound;
659 	int sig;
660 };
661 
662 static int killpg_all_callback(struct proc *p, void *data);
663 
664 static int
665 dokillpg(int sig, int pgid, int all)
666 {
667 	struct killpg_info info;
668 	struct proc *cp = curproc;
669 	struct proc *p;
670 	struct pgrp *pgrp;
671 
672 	info.nfound = 0;
673 	info.sig = sig;
674 
675 	if (all) {
676 		/*
677 		 * broadcast
678 		 */
679 		allproc_scan(killpg_all_callback, &info);
680 	} else {
681 		if (pgid == 0) {
682 			/*
683 			 * zero pgid means send to my process group.
684 			 */
685 			pgrp = cp->p_pgrp;
686 			pgref(pgrp);
687 		} else {
688 			pgrp = pgfind(pgid);
689 			if (pgrp == NULL)
690 				return (ESRCH);
691 		}
692 
693 		/*
694 		 * Must interlock all signals against fork
695 		 */
696 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
697 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
698 			if (p->p_pid <= 1 ||
699 			    p->p_stat == SZOMB ||
700 			    (p->p_flags & P_SYSTEM) ||
701 			    !CANSIGNAL(p, sig)) {
702 				continue;
703 			}
704 			++info.nfound;
705 			if (sig)
706 				ksignal(p, sig);
707 		}
708 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
709 		pgrel(pgrp);
710 	}
711 	return (info.nfound ? 0 : ESRCH);
712 }
713 
714 static int
715 killpg_all_callback(struct proc *p, void *data)
716 {
717 	struct killpg_info *info = data;
718 
719 	if (p->p_pid <= 1 || (p->p_flags & P_SYSTEM) ||
720 	    p == curproc || !CANSIGNAL(p, info->sig)) {
721 		return (0);
722 	}
723 	++info->nfound;
724 	if (info->sig)
725 		ksignal(p, info->sig);
726 	return(0);
727 }
728 
729 /*
730  * Send a general signal to a process or LWPs within that process.
731  *
732  * Note that new signals cannot be sent if a process is exiting or already
733  * a zombie, but we return success anyway as userland is likely to not handle
734  * the race properly.
735  *
736  * No requirements.
737  */
738 int
739 kern_kill(int sig, pid_t pid, lwpid_t tid)
740 {
741 	int t;
742 
743 	if ((u_int)sig > _SIG_MAXSIG)
744 		return (EINVAL);
745 
746 	lwkt_gettoken(&proc_token);
747 
748 	if (pid > 0) {
749 		struct proc *p;
750 		struct lwp *lp = NULL;
751 
752 		/*
753 		 * Send a signal to a single process.  If the kill() is
754 		 * racing an exiting process which has not yet been reaped
755 		 * act as though the signal was delivered successfully but
756 		 * don't actually try to deliver the signal.
757 		 */
758 		if ((p = pfind(pid)) == NULL) {
759 			if ((p = zpfind(pid)) == NULL) {
760 				lwkt_reltoken(&proc_token);
761 				return (ESRCH);
762 			}
763 			lwkt_reltoken(&proc_token);
764 			PRELE(p);
765 			return (0);
766 		}
767 		lwkt_gettoken(&p->p_token);
768 		if (!CANSIGNAL(p, sig)) {
769 			lwkt_reltoken(&p->p_token);
770 			PRELE(p);
771 			lwkt_reltoken(&proc_token);
772 			return (EPERM);
773 		}
774 
775 		/*
776 		 * NOP if the process is exiting.  Note that lwpsignal() is
777 		 * called directly with P_WEXIT set to kill individual LWPs
778 		 * during exit, which is allowed.
779 		 */
780 		if (p->p_flags & P_WEXIT) {
781 			lwkt_reltoken(&p->p_token);
782 			PRELE(p);
783 			lwkt_reltoken(&proc_token);
784 			return (0);
785 		}
786 		if (tid != -1) {
787 			lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
788 			if (lp == NULL) {
789 				lwkt_reltoken(&p->p_token);
790 				PRELE(p);
791 				lwkt_reltoken(&proc_token);
792 				return (ESRCH);
793 			}
794 		}
795 		if (sig)
796 			lwpsignal(p, lp, sig);
797 		lwkt_reltoken(&p->p_token);
798 		PRELE(p);
799 		lwkt_reltoken(&proc_token);
800 		return (0);
801 	}
802 
803 	/*
804 	 * If we come here, pid is a special broadcast pid.
805 	 * This doesn't mix with a tid.
806 	 */
807 	if (tid != -1) {
808 		lwkt_reltoken(&proc_token);
809 		return (EINVAL);
810 	}
811 	switch (pid) {
812 	case -1:		/* broadcast signal */
813 		t = (dokillpg(sig, 0, 1));
814 		break;
815 	case 0:			/* signal own process group */
816 		t = (dokillpg(sig, 0, 0));
817 		break;
818 	default:		/* negative explicit process group */
819 		t = (dokillpg(sig, -pid, 0));
820 		break;
821 	}
822 	lwkt_reltoken(&proc_token);
823 	return t;
824 }
825 
826 int
827 sys_kill(struct kill_args *uap)
828 {
829 	int error;
830 
831 	error = kern_kill(uap->signum, uap->pid, -1);
832 	return (error);
833 }
834 
835 int
836 sys_lwp_kill(struct lwp_kill_args *uap)
837 {
838 	int error;
839 	pid_t pid = uap->pid;
840 
841 	/*
842 	 * A tid is mandatory for lwp_kill(), otherwise
843 	 * you could simply use kill().
844 	 */
845 	if (uap->tid == -1)
846 		return (EINVAL);
847 
848 	/*
849 	 * To save on a getpid() function call for intra-process
850 	 * signals, pid == -1 means current process.
851 	 */
852 	if (pid == -1)
853 		pid = curproc->p_pid;
854 
855 	error = kern_kill(uap->signum, pid, uap->tid);
856 	return (error);
857 }
858 
859 /*
860  * Send a signal to a process group.
861  */
862 void
863 gsignal(int pgid, int sig)
864 {
865 	struct pgrp *pgrp;
866 
867 	if (pgid && (pgrp = pgfind(pgid)))
868 		pgsignal(pgrp, sig, 0);
869 }
870 
871 /*
872  * Send a signal to a process group.  If checktty is 1,
873  * limit to members which have a controlling terminal.
874  *
875  * pg_lock interlocks against a fork that might be in progress, to
876  * ensure that the new child process picks up the signal.
877  */
878 void
879 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
880 {
881 	struct proc *p;
882 
883 	/*
884 	 * Must interlock all signals against fork
885 	 */
886 	if (pgrp) {
887 		pgref(pgrp);
888 		lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
889 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
890 			if (checkctty == 0 || p->p_flags & P_CONTROLT)
891 				ksignal(p, sig);
892 		}
893 		lockmgr(&pgrp->pg_lock, LK_RELEASE);
894 		pgrel(pgrp);
895 	}
896 }
897 
898 /*
899  * Send a signal caused by a trap to the current lwp.  If it will be caught
900  * immediately, deliver it with correct code.  Otherwise, post it normally.
901  *
902  * These signals may ONLY be delivered to the specified lwp and may never
903  * be delivered to the process generically.
904  */
905 void
906 trapsignal(struct lwp *lp, int sig, u_long code)
907 {
908 	struct proc *p = lp->lwp_proc;
909 	struct sigacts *ps = p->p_sigacts;
910 
911 	/*
912 	 * If we are a virtual kernel running an emulated user process
913 	 * context, switch back to the virtual kernel context before
914 	 * trying to post the signal.
915 	 */
916 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
917 		struct trapframe *tf = lp->lwp_md.md_regs;
918 		tf->tf_trapno = 0;
919 		vkernel_trap(lp, tf);
920 	}
921 
922 
923 	if ((p->p_flags & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
924 	    !SIGISMEMBER(lp->lwp_sigmask, sig)) {
925 		lp->lwp_ru.ru_nsignals++;
926 #ifdef KTRACE
927 		if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
928 			ktrpsig(lp, sig, ps->ps_sigact[_SIG_IDX(sig)],
929 				&lp->lwp_sigmask, code);
930 #endif
931 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
932 						&lp->lwp_sigmask, code);
933 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
934 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
935 			SIGADDSET(lp->lwp_sigmask, sig);
936 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
937 			/*
938 			 * See kern_sigaction() for origin of this code.
939 			 */
940 			SIGDELSET(p->p_sigcatch, sig);
941 			if (sig != SIGCONT &&
942 			    sigprop(sig) & SA_IGNORE)
943 				SIGADDSET(p->p_sigignore, sig);
944 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
945 		}
946 	} else {
947 		lp->lwp_code = code;	/* XXX for core dump/debugger */
948 		lp->lwp_sig = sig;	/* XXX to verify code */
949 		lwpsignal(p, lp, sig);
950 	}
951 }
952 
953 /*
954  * Find a suitable lwp to deliver the signal to.  Returns NULL if all
955  * lwps hold the signal blocked.
956  *
957  * Caller must hold p->p_token.
958  *
959  * Returns a lp or NULL.  If non-NULL the lp is held and its token is
960  * acquired.
961  */
962 static struct lwp *
963 find_lwp_for_signal(struct proc *p, int sig)
964 {
965 	struct lwp *lp;
966 	struct lwp *run, *sleep, *stop;
967 
968 	/*
969 	 * If the running/preempted thread belongs to the proc to which
970 	 * the signal is being delivered and this thread does not block
971 	 * the signal, then we can avoid a context switch by delivering
972 	 * the signal to this thread, because it will return to userland
973 	 * soon anyways.
974 	 */
975 	lp = lwkt_preempted_proc();
976 	if (lp != NULL && lp->lwp_proc == p) {
977 		LWPHOLD(lp);
978 		lwkt_gettoken(&lp->lwp_token);
979 		if (!SIGISMEMBER(lp->lwp_sigmask, sig)) {
980 			/* return w/ token held */
981 			return (lp);
982 		}
983 		lwkt_reltoken(&lp->lwp_token);
984 		LWPRELE(lp);
985 	}
986 
987 	run = sleep = stop = NULL;
988 	FOREACH_LWP_IN_PROC(lp, p) {
989 		/*
990 		 * If the signal is being blocked by the lwp, then this
991 		 * lwp is not eligible for receiving the signal.
992 		 */
993 		LWPHOLD(lp);
994 		lwkt_gettoken(&lp->lwp_token);
995 
996 		if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
997 			lwkt_reltoken(&lp->lwp_token);
998 			LWPRELE(lp);
999 			continue;
1000 		}
1001 
1002 		switch (lp->lwp_stat) {
1003 		case LSRUN:
1004 			if (sleep) {
1005 				lwkt_token_swap();
1006 				lwkt_reltoken(&sleep->lwp_token);
1007 				LWPRELE(sleep);
1008 				sleep = NULL;
1009 				run = lp;
1010 			} else if (stop) {
1011 				lwkt_token_swap();
1012 				lwkt_reltoken(&stop->lwp_token);
1013 				LWPRELE(stop);
1014 				stop = NULL;
1015 				run = lp;
1016 			} else {
1017 				run = lp;
1018 			}
1019 			break;
1020 		case LSSLEEP:
1021 			if (lp->lwp_flags & LWP_SINTR) {
1022 				if (sleep) {
1023 					lwkt_reltoken(&lp->lwp_token);
1024 					LWPRELE(lp);
1025 				} else if (stop) {
1026 					lwkt_token_swap();
1027 					lwkt_reltoken(&stop->lwp_token);
1028 					LWPRELE(stop);
1029 					stop = NULL;
1030 					sleep = lp;
1031 				} else {
1032 					sleep = lp;
1033 				}
1034 			} else {
1035 				lwkt_reltoken(&lp->lwp_token);
1036 				LWPRELE(lp);
1037 			}
1038 			break;
1039 		case LSSTOP:
1040 			if (sleep) {
1041 				lwkt_reltoken(&lp->lwp_token);
1042 				LWPRELE(lp);
1043 			} else if (stop) {
1044 				lwkt_reltoken(&lp->lwp_token);
1045 				LWPRELE(lp);
1046 			} else {
1047 				stop = lp;
1048 			}
1049 			break;
1050 		}
1051 		if (run)
1052 			break;
1053 	}
1054 
1055 	if (run != NULL)
1056 		return (run);
1057 	else if (sleep != NULL)
1058 		return (sleep);
1059 	else
1060 		return (stop);
1061 }
1062 
1063 /*
1064  * Send the signal to the process.  If the signal has an action, the action
1065  * is usually performed by the target process rather than the caller; we add
1066  * the signal to the set of pending signals for the process.
1067  *
1068  * Exceptions:
1069  *   o When a stop signal is sent to a sleeping process that takes the
1070  *     default action, the process is stopped without awakening it.
1071  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
1072  *     regardless of the signal action (eg, blocked or ignored).
1073  *
1074  * Other ignored signals are discarded immediately.
1075  *
1076  * If the caller wishes to call this function from a hard code section the
1077  * caller must already hold p->p_token (see kern_clock.c).
1078  *
1079  * No requirements.
1080  */
1081 void
1082 ksignal(struct proc *p, int sig)
1083 {
1084 	lwpsignal(p, NULL, sig);
1085 }
1086 
1087 /*
1088  * The core for ksignal.  lp may be NULL, then a suitable thread
1089  * will be chosen.  If not, lp MUST be a member of p.
1090  *
1091  * If the caller wishes to call this function from a hard code section the
1092  * caller must already hold p->p_token.
1093  *
1094  * No requirements.
1095  */
1096 void
1097 lwpsignal(struct proc *p, struct lwp *lp, int sig)
1098 {
1099 	struct proc *q;
1100 	sig_t action;
1101 	int prop;
1102 
1103 	if (sig > _SIG_MAXSIG || sig <= 0) {
1104 		kprintf("lwpsignal: signal %d\n", sig);
1105 		panic("lwpsignal signal number");
1106 	}
1107 
1108 	KKASSERT(lp == NULL || lp->lwp_proc == p);
1109 
1110 	/*
1111 	 * We don't want to race... well, all sorts of things.  Get appropriate
1112 	 * tokens.
1113 	 *
1114 	 * Don't try to deliver a generic signal to an exiting process,
1115 	 * the signal structures could be in flux.  We check the LWP later
1116 	 * on.
1117 	 */
1118 	PHOLD(p);
1119 	lwkt_gettoken(&p->p_token);
1120 	if (lp) {
1121 		LWPHOLD(lp);
1122 		lwkt_gettoken(&lp->lwp_token);
1123 	} else if (p->p_flags & P_WEXIT) {
1124 		goto out;
1125 	}
1126 
1127 	prop = sigprop(sig);
1128 
1129 	/*
1130 	 * If proc is traced, always give parent a chance;
1131 	 * if signal event is tracked by procfs, give *that*
1132 	 * a chance, as well.
1133 	 */
1134 	if ((p->p_flags & P_TRACED) || (p->p_stops & S_SIG)) {
1135 		action = SIG_DFL;
1136 	} else {
1137 		/*
1138 		 * Do not try to deliver signals to an exiting lwp.  Note
1139 		 * that we must still deliver the signal if P_WEXIT is set
1140 		 * in the process flags.
1141 		 */
1142 		if (lp && (lp->lwp_mpflags & LWP_MP_WEXIT)) {
1143 			if (lp) {
1144 				lwkt_reltoken(&lp->lwp_token);
1145 				LWPRELE(lp);
1146 			}
1147 			lwkt_reltoken(&p->p_token);
1148 			PRELE(p);
1149 			return;
1150 		}
1151 
1152 		/*
1153 		 * If the signal is being ignored, then we forget about
1154 		 * it immediately.  NOTE: We don't set SIGCONT in p_sigignore,
1155 		 * and if it is set to SIG_IGN, action will be SIG_DFL here.
1156 		 */
1157 		if (SIGISMEMBER(p->p_sigignore, sig)) {
1158 			/*
1159 			 * Even if a signal is set SIG_IGN, it may still be
1160 			 * lurking in a kqueue.
1161 			 */
1162 			KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1163 			if (lp) {
1164 				lwkt_reltoken(&lp->lwp_token);
1165 				LWPRELE(lp);
1166 			}
1167 			lwkt_reltoken(&p->p_token);
1168 			PRELE(p);
1169 			return;
1170 		}
1171 		if (SIGISMEMBER(p->p_sigcatch, sig))
1172 			action = SIG_CATCH;
1173 		else
1174 			action = SIG_DFL;
1175 	}
1176 
1177 	/*
1178 	 * If continuing, clear any pending STOP signals.
1179 	 */
1180 	if (prop & SA_CONT)
1181 		SIG_STOPSIGMASK(p->p_siglist);
1182 
1183 	if (prop & SA_STOP) {
1184 		/*
1185 		 * If sending a tty stop signal to a member of an orphaned
1186 		 * process group, discard the signal here if the action
1187 		 * is default; don't stop the process below if sleeping,
1188 		 * and don't clear any pending SIGCONT.
1189 		 */
1190 		if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0 &&
1191 		    action == SIG_DFL) {
1192 			lwkt_reltoken(&p->p_token);
1193 			PRELE(p);
1194 		        return;
1195 		}
1196 		SIG_CONTSIGMASK(p->p_siglist);
1197 		p->p_flags &= ~P_CONTINUED;
1198 	}
1199 
1200 	if (p->p_stat == SSTOP) {
1201 		/*
1202 		 * Nobody can handle this signal, add it to the lwp or
1203 		 * process pending list
1204 		 */
1205 		if (lp) {
1206 			spin_lock(&lp->lwp_spin);
1207 			SIGADDSET(lp->lwp_siglist, sig);
1208 			spin_unlock(&lp->lwp_spin);
1209 		} else {
1210 			SIGADDSET(p->p_siglist, sig);
1211 		}
1212 
1213 		/*
1214 		 * If the process is stopped and is being traced, then no
1215 		 * further action is necessary.
1216 		 */
1217 		if (p->p_flags & P_TRACED)
1218 			goto out;
1219 
1220 		/*
1221 		 * If the process is stopped and receives a KILL signal,
1222 		 * make the process runnable.
1223 		 */
1224 		if (sig == SIGKILL) {
1225 			proc_unstop(p);
1226 			goto active_process;
1227 		}
1228 
1229 		/*
1230 		 * If the process is stopped and receives a CONT signal,
1231 		 * then try to make the process runnable again.
1232 		 */
1233 		if (prop & SA_CONT) {
1234 			/*
1235 			 * If SIGCONT is default (or ignored), we continue the
1236 			 * process but don't leave the signal in p_siglist, as
1237 			 * it has no further action.  If SIGCONT is held, we
1238 			 * continue the process and leave the signal in
1239 			 * p_siglist.  If the process catches SIGCONT, let it
1240 			 * handle the signal itself.
1241 			 *
1242 			 * XXX what if the signal is being held blocked?
1243 			 *
1244 			 * Token required to interlock kern_wait().
1245 			 * Reparenting can also cause a race so we have to
1246 			 * hold (q).
1247 			 */
1248 			q = p->p_pptr;
1249 			PHOLD(q);
1250 			lwkt_gettoken(&q->p_token);
1251 			p->p_flags |= P_CONTINUED;
1252 			wakeup(q);
1253 			if (action == SIG_DFL)
1254 				SIGDELSET(p->p_siglist, sig);
1255 			proc_unstop(p);
1256 			lwkt_reltoken(&q->p_token);
1257 			PRELE(q);
1258 			if (action == SIG_CATCH)
1259 				goto active_process;
1260 			goto out;
1261 		}
1262 
1263 		/*
1264 		 * If the process is stopped and receives another STOP
1265 		 * signal, we do not need to stop it again.  If we did
1266 		 * the shell could get confused.
1267 		 *
1268 		 * However, if the current/preempted lwp is part of the
1269 		 * process receiving the signal, we need to keep it,
1270 		 * so that this lwp can stop in issignal() later, as
1271 		 * we don't want to wait until it reaches userret!
1272 		 */
1273 		if (prop & SA_STOP) {
1274 			if (lwkt_preempted_proc() == NULL ||
1275 			    lwkt_preempted_proc()->lwp_proc != p)
1276 				SIGDELSET(p->p_siglist, sig);
1277 		}
1278 
1279 		/*
1280 		 * Otherwise the process is stopped and it received some
1281 		 * signal, which does not change its stopped state.  When
1282 		 * the process is continued a wakeup(p) will be issued which
1283 		 * will wakeup any threads sleeping in tstop().
1284 		 */
1285 		if (lp == NULL) {
1286 			/* NOTE: returns lp w/ token held */
1287 			lp = find_lwp_for_signal(p, sig);
1288 		}
1289 		goto out;
1290 
1291 		/* NOTREACHED */
1292 	}
1293 	/* else not stopped */
1294 active_process:
1295 
1296 	/*
1297 	 * Never deliver a lwp-specific signal to a random lwp.
1298 	 */
1299 	if (lp == NULL) {
1300 		/* NOTE: returns lp w/ token held */
1301 		lp = find_lwp_for_signal(p, sig);
1302 		if (lp) {
1303 			if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
1304 				lwkt_reltoken(&lp->lwp_token);
1305 				LWPRELE(lp);
1306 				lp = NULL;
1307 			}
1308 		}
1309 	}
1310 
1311 	/*
1312 	 * Deliver to the process generically if (1) the signal is being
1313 	 * sent to any thread or (2) we could not find a thread to deliver
1314 	 * it to.
1315 	 */
1316 	if (lp == NULL) {
1317 		SIGADDSET(p->p_siglist, sig);
1318 		goto out;
1319 	}
1320 
1321 	/*
1322 	 * Deliver to a specific LWP whether it masks it or not.  It will
1323 	 * not be dispatched if masked but we must still deliver it.
1324 	 */
1325 	if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1326 	    (p->p_flags & P_TRACED) == 0) {
1327 		p->p_nice = NZERO;
1328 	}
1329 
1330 	/*
1331 	 * If the process receives a STOP signal which indeed needs to
1332 	 * stop the process, do so.  If the process chose to catch the
1333 	 * signal, it will be treated like any other signal.
1334 	 */
1335 	if ((prop & SA_STOP) && action == SIG_DFL) {
1336 		/*
1337 		 * If a child holding parent blocked, stopping
1338 		 * could cause deadlock.  Take no action at this
1339 		 * time.
1340 		 */
1341 		if (p->p_flags & P_PPWAIT) {
1342 			SIGADDSET(p->p_siglist, sig);
1343 			goto out;
1344 		}
1345 
1346 		/*
1347 		 * Do not actually try to manipulate the process, but simply
1348 		 * stop it.  Lwps will stop as soon as they safely can.
1349 		 *
1350 		 * Ignore stop if the process is exiting.
1351 		 */
1352 		if ((p->p_flags & P_WEXIT) == 0) {
1353 			p->p_xstat = sig;
1354 			proc_stop(p);
1355 		}
1356 		goto out;
1357 	}
1358 
1359 	/*
1360 	 * If it is a CONT signal with default action, just ignore it.
1361 	 */
1362 	if ((prop & SA_CONT) && action == SIG_DFL)
1363 		goto out;
1364 
1365 	/*
1366 	 * Mark signal pending at this specific thread.
1367 	 */
1368 	spin_lock(&lp->lwp_spin);
1369 	SIGADDSET(lp->lwp_siglist, sig);
1370 	spin_unlock(&lp->lwp_spin);
1371 
1372 	lwp_signotify(lp);
1373 
1374 out:
1375 	if (lp) {
1376 		lwkt_reltoken(&lp->lwp_token);
1377 		LWPRELE(lp);
1378 	}
1379 	lwkt_reltoken(&p->p_token);
1380 	PRELE(p);
1381 }
1382 
1383 /*
1384  * Notify the LWP that a signal has arrived.  The LWP does not have to be
1385  * sleeping on the current cpu.
1386  *
1387  * p->p_token and lp->lwp_token must be held on call.
1388  *
1389  * We can only safely schedule the thread on its current cpu and only if
1390  * one of the SINTR flags is set.  If an SINTR flag is set AND we are on
1391  * the correct cpu we are properly interlocked, otherwise we could be
1392  * racing other thread transition states (or the lwp is on the user scheduler
1393  * runq but not scheduled) and must not do anything.
1394  *
1395  * Since we hold the lwp token we know the lwp cannot be ripped out from
1396  * under us so we can safely hold it to prevent it from being ripped out
1397  * from under us if we are forced to IPI another cpu to make the local
1398  * checks there.
1399  *
1400  * Adjustment of lp->lwp_stat can only occur when we hold the lwp_token,
1401  * which we won't in an IPI so any fixups have to be done here, effectively
1402  * replicating part of what setrunnable() does.
1403  */
1404 static void
1405 lwp_signotify(struct lwp *lp)
1406 {
1407 	ASSERT_LWKT_TOKEN_HELD(&lp->lwp_proc->p_token);
1408 
1409 	crit_enter();
1410 	if (lp == lwkt_preempted_proc()) {
1411 		/*
1412 		 * lwp is on the current cpu AND it is currently running
1413 		 * (we preempted it).
1414 		 */
1415 		signotify();
1416 	} else if (lp->lwp_flags & LWP_SINTR) {
1417 		/*
1418 		 * lwp is sitting in tsleep() with PCATCH set
1419 		 */
1420 #ifdef SMP
1421 		if (lp->lwp_thread->td_gd == mycpu) {
1422 			setrunnable(lp);
1423 		} else {
1424 			/*
1425 			 * We can only adjust lwp_stat while we hold the
1426 			 * lwp_token, and we won't in the IPI function.
1427 			 */
1428 			LWPHOLD(lp);
1429 			if (lp->lwp_stat == LSSTOP)
1430 				lp->lwp_stat = LSSLEEP;
1431 			lwkt_send_ipiq(lp->lwp_thread->td_gd,
1432 				       lwp_signotify_remote, lp);
1433 		}
1434 #else
1435 		setrunnable(lp);
1436 #endif
1437 	} else if (lp->lwp_thread->td_flags & TDF_SINTR) {
1438 		/*
1439 		 * lwp is sitting in lwkt_sleep() with PCATCH set.
1440 		 */
1441 #ifdef SMP
1442 		if (lp->lwp_thread->td_gd == mycpu) {
1443 			setrunnable(lp);
1444 		} else {
1445 			/*
1446 			 * We can only adjust lwp_stat while we hold the
1447 			 * lwp_token, and we won't in the IPI function.
1448 			 */
1449 			LWPHOLD(lp);
1450 			if (lp->lwp_stat == LSSTOP)
1451 				lp->lwp_stat = LSSLEEP;
1452 			lwkt_send_ipiq(lp->lwp_thread->td_gd,
1453 				       lwp_signotify_remote, lp);
1454 		}
1455 #else
1456 		setrunnable(lp);
1457 #endif
1458 	} else {
1459 		/*
1460 		 * Otherwise the lwp is either in some uninterruptable state
1461 		 * or it is on the userland scheduler's runqueue waiting to
1462 		 * be scheduled to a cpu.
1463 		 */
1464 	}
1465 	crit_exit();
1466 }
1467 
1468 #ifdef SMP
1469 
1470 /*
1471  * This function is called via an IPI so we cannot call setrunnable() here
1472  * (because while we hold the lp we don't own its token, and can't get it
1473  * from an IPI).
1474  *
1475  * We are interlocked by virtue of being on the same cpu as the target.  If
1476  * we still are and LWP_SINTR or TDF_SINTR is set we can safely schedule
1477  * the target thread.
1478  */
1479 static void
1480 lwp_signotify_remote(void *arg)
1481 {
1482 	struct lwp *lp = arg;
1483 	thread_t td = lp->lwp_thread;
1484 
1485 	if (lp == lwkt_preempted_proc()) {
1486 		signotify();
1487 		LWPRELE(lp);
1488 	} else if (td->td_gd == mycpu) {
1489 		if ((lp->lwp_flags & LWP_SINTR) ||
1490 		    (td->td_flags & TDF_SINTR)) {
1491 			lwkt_schedule(td);
1492 		}
1493 		LWPRELE(lp);
1494 	} else {
1495 		lwkt_send_ipiq(td->td_gd, lwp_signotify_remote, lp);
1496 		/* LWPHOLD() is forwarded to the target cpu */
1497 	}
1498 }
1499 
1500 #endif
1501 
1502 /*
1503  * Caller must hold p->p_token
1504  */
1505 void
1506 proc_stop(struct proc *p)
1507 {
1508 	struct proc *q;
1509 	struct lwp *lp;
1510 
1511 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1512 
1513 	/* If somebody raced us, be happy with it */
1514 	if (p->p_stat == SSTOP || p->p_stat == SZOMB) {
1515 		return;
1516 	}
1517 	p->p_stat = SSTOP;
1518 
1519 	FOREACH_LWP_IN_PROC(lp, p) {
1520 		LWPHOLD(lp);
1521 		lwkt_gettoken(&lp->lwp_token);
1522 
1523 		switch (lp->lwp_stat) {
1524 		case LSSTOP:
1525 			/*
1526 			 * Do nothing, we are already counted in
1527 			 * p_nstopped.
1528 			 */
1529 			break;
1530 
1531 		case LSSLEEP:
1532 			/*
1533 			 * We're sleeping, but we will stop before
1534 			 * returning to userspace, so count us
1535 			 * as stopped as well.  We set LWP_MP_WSTOP
1536 			 * to signal the lwp that it should not
1537 			 * increase p_nstopped when reaching tstop().
1538 			 *
1539 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1540 			 */
1541 			if ((lp->lwp_mpflags & LWP_MP_WSTOP) == 0) {
1542 				atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
1543 				++p->p_nstopped;
1544 			}
1545 			break;
1546 
1547 		case LSRUN:
1548 			/*
1549 			 * We might notify ourself, but that's not
1550 			 * a problem.
1551 			 */
1552 			lwp_signotify(lp);
1553 			break;
1554 		}
1555 		lwkt_reltoken(&lp->lwp_token);
1556 		LWPRELE(lp);
1557 	}
1558 
1559 	if (p->p_nstopped == p->p_nthreads) {
1560 		/*
1561 		 * Token required to interlock kern_wait().  Reparenting can
1562 		 * also cause a race so we have to hold (q).
1563 		 */
1564 		q = p->p_pptr;
1565 		PHOLD(q);
1566 		lwkt_gettoken(&q->p_token);
1567 		p->p_flags &= ~P_WAITED;
1568 		wakeup(q);
1569 		if ((q->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1570 			ksignal(p->p_pptr, SIGCHLD);
1571 		lwkt_reltoken(&q->p_token);
1572 		PRELE(q);
1573 	}
1574 }
1575 
1576 /*
1577  * Caller must hold proc_token
1578  */
1579 void
1580 proc_unstop(struct proc *p)
1581 {
1582 	struct lwp *lp;
1583 
1584 	ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1585 
1586 	if (p->p_stat != SSTOP)
1587 		return;
1588 
1589 	p->p_stat = SACTIVE;
1590 
1591 	FOREACH_LWP_IN_PROC(lp, p) {
1592 		LWPHOLD(lp);
1593 		lwkt_gettoken(&lp->lwp_token);
1594 
1595 		switch (lp->lwp_stat) {
1596 		case LSRUN:
1597 			/*
1598 			 * Uh?  Not stopped?  Well, I guess that's okay.
1599 			 */
1600 			if (bootverbose)
1601 				kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1602 					p->p_pid, lp->lwp_tid);
1603 			break;
1604 
1605 		case LSSLEEP:
1606 			/*
1607 			 * Still sleeping.  Don't bother waking it up.
1608 			 * However, if this thread was counted as
1609 			 * stopped, undo this.
1610 			 *
1611 			 * Nevertheless we call setrunnable() so that it
1612 			 * will wake up in case a signal or timeout arrived
1613 			 * in the meantime.
1614 			 *
1615 			 * LWP_MP_WSTOP is protected by lp->lwp_token.
1616 			 */
1617 			if (lp->lwp_mpflags & LWP_MP_WSTOP) {
1618 				atomic_clear_int(&lp->lwp_mpflags,
1619 						 LWP_MP_WSTOP);
1620 				--p->p_nstopped;
1621 			} else {
1622 				if (bootverbose)
1623 					kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1624 						p->p_pid, lp->lwp_tid);
1625 			}
1626 			/* FALLTHROUGH */
1627 
1628 		case LSSTOP:
1629 			/*
1630 			 * This handles any lwp's waiting in a tsleep with
1631 			 * SIGCATCH.
1632 			 */
1633 			lwp_signotify(lp);
1634 			break;
1635 
1636 		}
1637 		lwkt_reltoken(&lp->lwp_token);
1638 		LWPRELE(lp);
1639 	}
1640 
1641 	/*
1642 	 * This handles any lwp's waiting in tstop().  We have interlocked
1643 	 * the setting of p_stat by acquiring and releasing each lpw's
1644 	 * token.
1645 	 */
1646 	wakeup(p);
1647 }
1648 
1649 /*
1650  * No requirements.
1651  */
1652 static int
1653 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1654 {
1655 	sigset_t savedmask, set;
1656 	struct proc *p = curproc;
1657 	struct lwp *lp = curthread->td_lwp;
1658 	int error, sig, hz, timevalid = 0;
1659 	struct timespec rts, ets, ts;
1660 	struct timeval tv;
1661 
1662 	error = 0;
1663 	sig = 0;
1664 	ets.tv_sec = 0;		/* silence compiler warning */
1665 	ets.tv_nsec = 0;	/* silence compiler warning */
1666 	SIG_CANTMASK(waitset);
1667 	savedmask = lp->lwp_sigmask;
1668 
1669 	if (timeout) {
1670 		if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1671 		    timeout->tv_nsec < 1000000000) {
1672 			timevalid = 1;
1673 			getnanouptime(&rts);
1674 		 	ets = rts;
1675 			timespecadd(&ets, timeout);
1676 		}
1677 	}
1678 
1679 	for (;;) {
1680 		set = lwp_sigpend(lp);
1681 		SIGSETAND(set, waitset);
1682 		if ((sig = sig_ffs(&set)) != 0) {
1683 			SIGFILLSET(lp->lwp_sigmask);
1684 			SIGDELSET(lp->lwp_sigmask, sig);
1685 			SIG_CANTMASK(lp->lwp_sigmask);
1686 			sig = issignal(lp, 1);
1687 			/*
1688 			 * It may be a STOP signal, in the case, issignal
1689 			 * returns 0, because we may stop there, and new
1690 			 * signal can come in, we should restart if we got
1691 			 * nothing.
1692 			 */
1693 			if (sig == 0)
1694 				continue;
1695 			else
1696 				break;
1697 		}
1698 
1699 		/*
1700 		 * Previous checking got nothing, and we retried but still
1701 		 * got nothing, we should return the error status.
1702 		 */
1703 		if (error)
1704 			break;
1705 
1706 		/*
1707 		 * POSIX says this must be checked after looking for pending
1708 		 * signals.
1709 		 */
1710 		if (timeout) {
1711 			if (timevalid == 0) {
1712 				error = EINVAL;
1713 				break;
1714 			}
1715 			getnanouptime(&rts);
1716 			if (timespeccmp(&rts, &ets, >=)) {
1717 				error = EAGAIN;
1718 				break;
1719 			}
1720 			ts = ets;
1721 			timespecsub(&ts, &rts);
1722 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1723 			hz = tvtohz_high(&tv);
1724 		} else {
1725 			hz = 0;
1726 		}
1727 
1728 		lp->lwp_sigmask = savedmask;
1729 		SIGSETNAND(lp->lwp_sigmask, waitset);
1730 		/*
1731 		 * We won't ever be woken up.  Instead, our sleep will
1732 		 * be broken in lwpsignal().
1733 		 */
1734 		error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1735 		if (timeout) {
1736 			if (error == ERESTART) {
1737 				/* can not restart a timeout wait. */
1738 				error = EINTR;
1739 			} else if (error == EAGAIN) {
1740 				/* will calculate timeout by ourself. */
1741 				error = 0;
1742 			}
1743 		}
1744 		/* Retry ... */
1745 	}
1746 
1747 	lp->lwp_sigmask = savedmask;
1748 	if (sig) {
1749 		error = 0;
1750 		bzero(info, sizeof(*info));
1751 		info->si_signo = sig;
1752 		spin_lock(&lp->lwp_spin);
1753 		lwp_delsig(lp, sig);	/* take the signal! */
1754 		spin_unlock(&lp->lwp_spin);
1755 
1756 		if (sig == SIGKILL) {
1757 			sigexit(lp, sig);
1758 			/* NOT REACHED */
1759 		}
1760 	}
1761 
1762 	return (error);
1763 }
1764 
1765 /*
1766  * MPALMOSTSAFE
1767  */
1768 int
1769 sys_sigtimedwait(struct sigtimedwait_args *uap)
1770 {
1771 	struct timespec ts;
1772 	struct timespec *timeout;
1773 	sigset_t set;
1774 	siginfo_t info;
1775 	int error;
1776 
1777 	if (uap->timeout) {
1778 		error = copyin(uap->timeout, &ts, sizeof(ts));
1779 		if (error)
1780 			return (error);
1781 		timeout = &ts;
1782 	} else {
1783 		timeout = NULL;
1784 	}
1785 	error = copyin(uap->set, &set, sizeof(set));
1786 	if (error)
1787 		return (error);
1788 	error = kern_sigtimedwait(set, &info, timeout);
1789 	if (error)
1790 		return (error);
1791 	if (uap->info)
1792 		error = copyout(&info, uap->info, sizeof(info));
1793 	/* Repost if we got an error. */
1794 	/*
1795 	 * XXX lwp
1796 	 *
1797 	 * This could transform a thread-specific signal to another
1798 	 * thread / process pending signal.
1799 	 */
1800 	if (error) {
1801 		ksignal(curproc, info.si_signo);
1802 	} else {
1803 		uap->sysmsg_result = info.si_signo;
1804 	}
1805 	return (error);
1806 }
1807 
1808 /*
1809  * MPALMOSTSAFE
1810  */
1811 int
1812 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1813 {
1814 	siginfo_t info;
1815 	sigset_t set;
1816 	int error;
1817 
1818 	error = copyin(uap->set, &set, sizeof(set));
1819 	if (error)
1820 		return (error);
1821 	error = kern_sigtimedwait(set, &info, NULL);
1822 	if (error)
1823 		return (error);
1824 	if (uap->info)
1825 		error = copyout(&info, uap->info, sizeof(info));
1826 	/* Repost if we got an error. */
1827 	/*
1828 	 * XXX lwp
1829 	 *
1830 	 * This could transform a thread-specific signal to another
1831 	 * thread / process pending signal.
1832 	 */
1833 	if (error) {
1834 		ksignal(curproc, info.si_signo);
1835 	} else {
1836 		uap->sysmsg_result = info.si_signo;
1837 	}
1838 	return (error);
1839 }
1840 
1841 /*
1842  * If the current process has received a signal that would interrupt a
1843  * system call, return EINTR or ERESTART as appropriate.
1844  */
1845 int
1846 iscaught(struct lwp *lp)
1847 {
1848 	struct proc *p = lp->lwp_proc;
1849 	int sig;
1850 
1851 	if (p) {
1852 		if ((sig = CURSIG(lp)) != 0) {
1853 			if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1854 				return (EINTR);
1855 			return (ERESTART);
1856 		}
1857 	}
1858 	return(EWOULDBLOCK);
1859 }
1860 
1861 /*
1862  * If the current process has received a signal (should be caught or cause
1863  * termination, should interrupt current syscall), return the signal number.
1864  * Stop signals with default action are processed immediately, then cleared;
1865  * they aren't returned.  This is checked after each entry to the system for
1866  * a syscall or trap (though this can usually be done without calling issignal
1867  * by checking the pending signal masks in the CURSIG macro).
1868  *
1869  * This routine is called via CURSIG/__cursig.  We will acquire and release
1870  * p->p_token but if the caller needs to interlock the test the caller must
1871  * also hold p->p_token.
1872  *
1873  *	while (sig = CURSIG(curproc))
1874  *		postsig(sig);
1875  *
1876  * MPSAFE
1877  */
1878 int
1879 issignal(struct lwp *lp, int maytrace)
1880 {
1881 	struct proc *p = lp->lwp_proc;
1882 	sigset_t mask;
1883 	int sig, prop;
1884 
1885 	lwkt_gettoken(&p->p_token);
1886 
1887 	for (;;) {
1888 		int traced = (p->p_flags & P_TRACED) || (p->p_stops & S_SIG);
1889 
1890 		/*
1891 		 * If this process is supposed to stop, stop this thread.
1892 		 */
1893 		if (p->p_stat == SSTOP)
1894 			tstop();
1895 
1896 		mask = lwp_sigpend(lp);
1897 		SIGSETNAND(mask, lp->lwp_sigmask);
1898 		if (p->p_flags & P_PPWAIT)
1899 			SIG_STOPSIGMASK(mask);
1900 		if (SIGISEMPTY(mask)) {		/* no signal to send */
1901 			lwkt_reltoken(&p->p_token);
1902 			return (0);
1903 		}
1904 		sig = sig_ffs(&mask);
1905 
1906 		STOPEVENT(p, S_SIG, sig);
1907 
1908 		/*
1909 		 * We should see pending but ignored signals
1910 		 * only if P_TRACED was on when they were posted.
1911 		 */
1912 		if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1913 			spin_lock(&lp->lwp_spin);
1914 			lwp_delsig(lp, sig);
1915 			spin_unlock(&lp->lwp_spin);
1916 			continue;
1917 		}
1918 		if (maytrace &&
1919 		    (p->p_flags & P_TRACED) &&
1920 		    (p->p_flags & P_PPWAIT) == 0) {
1921 			/*
1922 			 * If traced, always stop, and stay stopped until
1923 			 * released by the parent.
1924 			 *
1925 			 * NOTE: SSTOP may get cleared during the loop,
1926 			 * but we do not re-notify the parent if we have
1927 			 * to loop several times waiting for the parent
1928 			 * to let us continue.
1929 			 *
1930 			 * XXX not sure if this is still true
1931 			 */
1932 			p->p_xstat = sig;
1933 			proc_stop(p);
1934 			do {
1935 				tstop();
1936 			} while (!trace_req(p) && (p->p_flags & P_TRACED));
1937 
1938 			/*
1939 			 * If parent wants us to take the signal,
1940 			 * then it will leave it in p->p_xstat;
1941 			 * otherwise we just look for signals again.
1942 			 */
1943 			spin_lock(&lp->lwp_spin);
1944 			lwp_delsig(lp, sig);	/* clear old signal */
1945 			spin_unlock(&lp->lwp_spin);
1946 			sig = p->p_xstat;
1947 			if (sig == 0)
1948 				continue;
1949 
1950 			/*
1951 			 * Put the new signal into p_siglist.  If the
1952 			 * signal is being masked, look for other signals.
1953 			 *
1954 			 * XXX lwp might need a call to ksignal()
1955 			 */
1956 			SIGADDSET(p->p_siglist, sig);
1957 			if (SIGISMEMBER(lp->lwp_sigmask, sig))
1958 				continue;
1959 
1960 			/*
1961 			 * If the traced bit got turned off, go back up
1962 			 * to the top to rescan signals.  This ensures
1963 			 * that p_sig* and ps_sigact are consistent.
1964 			 */
1965 			if ((p->p_flags & P_TRACED) == 0)
1966 				continue;
1967 		}
1968 
1969 		prop = sigprop(sig);
1970 
1971 		/*
1972 		 * Decide whether the signal should be returned.
1973 		 * Return the signal's number, or fall through
1974 		 * to clear it from the pending mask.
1975 		 */
1976 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
1977 		case (intptr_t)SIG_DFL:
1978 			/*
1979 			 * Don't take default actions on system processes.
1980 			 */
1981 			if (p->p_pid <= 1) {
1982 #ifdef DIAGNOSTIC
1983 				/*
1984 				 * Are you sure you want to ignore SIGSEGV
1985 				 * in init? XXX
1986 				 */
1987 				kprintf("Process (pid %lu) got signal %d\n",
1988 					(u_long)p->p_pid, sig);
1989 #endif
1990 				break;		/* == ignore */
1991 			}
1992 
1993 			/*
1994 			 * Handle the in-kernel checkpoint action
1995 			 */
1996 			if (prop & SA_CKPT) {
1997 				checkpoint_signal_handler(lp);
1998 				break;
1999 			}
2000 
2001 			/*
2002 			 * If there is a pending stop signal to process
2003 			 * with default action, stop here,
2004 			 * then clear the signal.  However,
2005 			 * if process is member of an orphaned
2006 			 * process group, ignore tty stop signals.
2007 			 */
2008 			if (prop & SA_STOP) {
2009 				if (p->p_flags & P_TRACED ||
2010 		    		    (p->p_pgrp->pg_jobc == 0 &&
2011 				    prop & SA_TTYSTOP))
2012 					break;	/* == ignore */
2013 				if ((p->p_flags & P_WEXIT) == 0) {
2014 					p->p_xstat = sig;
2015 					proc_stop(p);
2016 					tstop();
2017 				}
2018 				break;
2019 			} else if (prop & SA_IGNORE) {
2020 				/*
2021 				 * Except for SIGCONT, shouldn't get here.
2022 				 * Default action is to ignore; drop it.
2023 				 */
2024 				break;		/* == ignore */
2025 			} else {
2026 				lwkt_reltoken(&p->p_token);
2027 				return (sig);
2028 			}
2029 
2030 			/*NOTREACHED*/
2031 
2032 		case (intptr_t)SIG_IGN:
2033 			/*
2034 			 * Masking above should prevent us ever trying
2035 			 * to take action on an ignored signal other
2036 			 * than SIGCONT, unless process is traced.
2037 			 */
2038 			if ((prop & SA_CONT) == 0 &&
2039 			    (p->p_flags & P_TRACED) == 0)
2040 				kprintf("issignal\n");
2041 			break;		/* == ignore */
2042 
2043 		default:
2044 			/*
2045 			 * This signal has an action, let
2046 			 * postsig() process it.
2047 			 */
2048 			lwkt_reltoken(&p->p_token);
2049 			return (sig);
2050 		}
2051 		spin_lock(&lp->lwp_spin);
2052 		lwp_delsig(lp, sig);		/* take the signal! */
2053 		spin_unlock(&lp->lwp_spin);
2054 	}
2055 	/* NOTREACHED */
2056 }
2057 
2058 /*
2059  * Take the action for the specified signal
2060  * from the current set of pending signals.
2061  *
2062  * Caller must hold p->p_token
2063  */
2064 void
2065 postsig(int sig)
2066 {
2067 	struct lwp *lp = curthread->td_lwp;
2068 	struct proc *p = lp->lwp_proc;
2069 	struct sigacts *ps = p->p_sigacts;
2070 	sig_t action;
2071 	sigset_t returnmask;
2072 	int code;
2073 
2074 	KASSERT(sig != 0, ("postsig"));
2075 
2076 	KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
2077 
2078 	/*
2079 	 * If we are a virtual kernel running an emulated user process
2080 	 * context, switch back to the virtual kernel context before
2081 	 * trying to post the signal.
2082 	 */
2083 	if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
2084 		struct trapframe *tf = lp->lwp_md.md_regs;
2085 		tf->tf_trapno = 0;
2086 		vkernel_trap(lp, tf);
2087 	}
2088 
2089 	spin_lock(&lp->lwp_spin);
2090 	lwp_delsig(lp, sig);
2091 	spin_unlock(&lp->lwp_spin);
2092 	action = ps->ps_sigact[_SIG_IDX(sig)];
2093 #ifdef KTRACE
2094 	if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
2095 		ktrpsig(lp, sig, action, lp->lwp_flags & LWP_OLDMASK ?
2096 			&lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
2097 #endif
2098 	STOPEVENT(p, S_SIG, sig);
2099 
2100 	if (action == SIG_DFL) {
2101 		/*
2102 		 * Default action, where the default is to kill
2103 		 * the process.  (Other cases were ignored above.)
2104 		 */
2105 		sigexit(lp, sig);
2106 		/* NOTREACHED */
2107 	} else {
2108 		/*
2109 		 * If we get here, the signal must be caught.
2110 		 */
2111 		KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
2112 		    ("postsig action"));
2113 
2114 		/*
2115 		 * Reset the signal handler if asked to
2116 		 */
2117 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
2118 			/*
2119 			 * See kern_sigaction() for origin of this code.
2120 			 */
2121 			SIGDELSET(p->p_sigcatch, sig);
2122 			if (sig != SIGCONT &&
2123 			    sigprop(sig) & SA_IGNORE)
2124 				SIGADDSET(p->p_sigignore, sig);
2125 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2126 		}
2127 
2128 		/*
2129 		 * Set the signal mask and calculate the mask to restore
2130 		 * when the signal function returns.
2131 		 *
2132 		 * Special case: user has done a sigsuspend.  Here the
2133 		 * current mask is not of interest, but rather the
2134 		 * mask from before the sigsuspend is what we want
2135 		 * restored after the signal processing is completed.
2136 		 */
2137 		if (lp->lwp_flags & LWP_OLDMASK) {
2138 			returnmask = lp->lwp_oldsigmask;
2139 			lp->lwp_flags &= ~LWP_OLDMASK;
2140 		} else {
2141 			returnmask = lp->lwp_sigmask;
2142 		}
2143 
2144 		SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
2145 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
2146 			SIGADDSET(lp->lwp_sigmask, sig);
2147 
2148 		lp->lwp_ru.ru_nsignals++;
2149 		if (lp->lwp_sig != sig) {
2150 			code = 0;
2151 		} else {
2152 			code = lp->lwp_code;
2153 			lp->lwp_code = 0;
2154 			lp->lwp_sig = 0;
2155 		}
2156 		(*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
2157 	}
2158 }
2159 
2160 /*
2161  * Kill the current process for stated reason.
2162  */
2163 void
2164 killproc(struct proc *p, char *why)
2165 {
2166 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
2167 		p->p_pid, p->p_comm,
2168 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2169 	ksignal(p, SIGKILL);
2170 }
2171 
2172 /*
2173  * Force the current process to exit with the specified signal, dumping core
2174  * if appropriate.  We bypass the normal tests for masked and caught signals,
2175  * allowing unrecoverable failures to terminate the process without changing
2176  * signal state.  Mark the accounting record with the signal termination.
2177  * If dumping core, save the signal number for the debugger.  Calls exit and
2178  * does not return.
2179  *
2180  * This routine does not return.
2181  */
2182 void
2183 sigexit(struct lwp *lp, int sig)
2184 {
2185 	struct proc *p = lp->lwp_proc;
2186 
2187 	lwkt_gettoken(&p->p_token);
2188 	p->p_acflag |= AXSIG;
2189 	if (sigprop(sig) & SA_CORE) {
2190 		lp->lwp_sig = sig;
2191 		/*
2192 		 * Log signals which would cause core dumps
2193 		 * (Log as LOG_INFO to appease those who don't want
2194 		 * these messages.)
2195 		 * XXX : Todo, as well as euid, write out ruid too
2196 		 */
2197 		if (coredump(lp, sig) == 0)
2198 			sig |= WCOREFLAG;
2199 		if (kern_logsigexit)
2200 			log(LOG_INFO,
2201 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2202 			    p->p_pid, p->p_comm,
2203 			    p->p_ucred ? p->p_ucred->cr_uid : -1,
2204 			    sig &~ WCOREFLAG,
2205 			    sig & WCOREFLAG ? " (core dumped)" : "");
2206 	}
2207 	lwkt_reltoken(&p->p_token);
2208 	exit1(W_EXITCODE(0, sig));
2209 	/* NOTREACHED */
2210 }
2211 
2212 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
2213 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2214 	      sizeof(corefilename), "process corefile name format string");
2215 
2216 /*
2217  * expand_name(name, uid, pid)
2218  * Expand the name described in corefilename, using name, uid, and pid.
2219  * corefilename is a kprintf-like string, with three format specifiers:
2220  *	%N	name of process ("name")
2221  *	%P	process id (pid)
2222  *	%U	user id (uid)
2223  * For example, "%N.core" is the default; they can be disabled completely
2224  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2225  * This is controlled by the sysctl variable kern.corefile (see above).
2226  */
2227 
2228 static char *
2229 expand_name(const char *name, uid_t uid, pid_t pid)
2230 {
2231 	char *temp;
2232 	char buf[11];		/* Buffer for pid/uid -- max 4B */
2233 	int i, n;
2234 	char *format = corefilename;
2235 	size_t namelen;
2236 
2237 	temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
2238 	if (temp == NULL)
2239 		return NULL;
2240 	namelen = strlen(name);
2241 	for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2242 		int l;
2243 		switch (format[i]) {
2244 		case '%':	/* Format character */
2245 			i++;
2246 			switch (format[i]) {
2247 			case '%':
2248 				temp[n++] = '%';
2249 				break;
2250 			case 'N':	/* process name */
2251 				if ((n + namelen) > MAXPATHLEN) {
2252 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2253 					    pid, name, uid, temp, name);
2254 					kfree(temp, M_TEMP);
2255 					return NULL;
2256 				}
2257 				memcpy(temp+n, name, namelen);
2258 				n += namelen;
2259 				break;
2260 			case 'P':	/* process id */
2261 				l = ksprintf(buf, "%u", pid);
2262 				if ((n + l) > MAXPATHLEN) {
2263 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2264 					    pid, name, uid, temp, name);
2265 					kfree(temp, M_TEMP);
2266 					return NULL;
2267 				}
2268 				memcpy(temp+n, buf, l);
2269 				n += l;
2270 				break;
2271 			case 'U':	/* user id */
2272 				l = ksprintf(buf, "%u", uid);
2273 				if ((n + l) > MAXPATHLEN) {
2274 					log(LOG_ERR, "pid %d (%s), uid (%u):  Path `%s%s' is too long\n",
2275 					    pid, name, uid, temp, name);
2276 					kfree(temp, M_TEMP);
2277 					return NULL;
2278 				}
2279 				memcpy(temp+n, buf, l);
2280 				n += l;
2281 				break;
2282 			default:
2283 			  	log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2284 			}
2285 			break;
2286 		default:
2287 			temp[n++] = format[i];
2288 		}
2289 	}
2290 	temp[n] = '\0';
2291 	return temp;
2292 }
2293 
2294 /*
2295  * Dump a process' core.  The main routine does some
2296  * policy checking, and creates the name of the coredump;
2297  * then it passes on a vnode and a size limit to the process-specific
2298  * coredump routine if there is one; if there _is not_ one, it returns
2299  * ENOSYS; otherwise it returns the error from the process-specific routine.
2300  *
2301  * The parameter `lp' is the lwp which triggered the coredump.
2302  */
2303 
2304 static int
2305 coredump(struct lwp *lp, int sig)
2306 {
2307 	struct proc *p = lp->lwp_proc;
2308 	struct vnode *vp;
2309 	struct ucred *cred = p->p_ucred;
2310 	struct flock lf;
2311 	struct nlookupdata nd;
2312 	struct vattr vattr;
2313 	int error, error1;
2314 	char *name;			/* name of corefile */
2315 	off_t limit;
2316 
2317 	STOPEVENT(p, S_CORE, 0);
2318 
2319 	if (((sugid_coredump == 0) && p->p_flags & P_SUGID) || do_coredump == 0)
2320 		return (EFAULT);
2321 
2322 	/*
2323 	 * Note that the bulk of limit checking is done after
2324 	 * the corefile is created.  The exception is if the limit
2325 	 * for corefiles is 0, in which case we don't bother
2326 	 * creating the corefile at all.  This layout means that
2327 	 * a corefile is truncated instead of not being created,
2328 	 * if it is larger than the limit.
2329 	 */
2330 	limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2331 	if (limit == 0)
2332 		return EFBIG;
2333 
2334 	name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2335 	if (name == NULL)
2336 		return (EINVAL);
2337 	error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2338 	if (error == 0)
2339 		error = vn_open(&nd, NULL, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
2340 	kfree(name, M_TEMP);
2341 	if (error) {
2342 		nlookup_done(&nd);
2343 		return (error);
2344 	}
2345 	vp = nd.nl_open_vp;
2346 	nd.nl_open_vp = NULL;
2347 	nlookup_done(&nd);
2348 
2349 	vn_unlock(vp);
2350 	lf.l_whence = SEEK_SET;
2351 	lf.l_start = 0;
2352 	lf.l_len = 0;
2353 	lf.l_type = F_WRLCK;
2354 	error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2355 	if (error)
2356 		goto out2;
2357 
2358 	/* Don't dump to non-regular files or files with links. */
2359 	if (vp->v_type != VREG ||
2360 	    VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2361 		error = EFAULT;
2362 		goto out1;
2363 	}
2364 
2365 	/* Don't dump to files current user does not own */
2366 	if (vattr.va_uid != p->p_ucred->cr_uid) {
2367 		error = EFAULT;
2368 		goto out1;
2369 	}
2370 
2371 	VATTR_NULL(&vattr);
2372 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2373 	vattr.va_size = 0;
2374 	VOP_SETATTR(vp, &vattr, cred);
2375 	p->p_acflag |= ACORE;
2376 	vn_unlock(vp);
2377 
2378 	error = p->p_sysent->sv_coredump ?
2379 		  p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2380 
2381 out1:
2382 	lf.l_type = F_UNLCK;
2383 	VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2384 out2:
2385 	error1 = vn_close(vp, FWRITE);
2386 	if (error == 0)
2387 		error = error1;
2388 	return (error);
2389 }
2390 
2391 /*
2392  * Nonexistent system call-- signal process (may want to handle it).
2393  * Flag error in case process won't see signal immediately (blocked or ignored).
2394  *
2395  * MPALMOSTSAFE
2396  */
2397 /* ARGSUSED */
2398 int
2399 sys_nosys(struct nosys_args *args)
2400 {
2401 	lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2402 	return (EINVAL);
2403 }
2404 
2405 /*
2406  * Send a SIGIO or SIGURG signal to a process or process group using
2407  * stored credentials rather than those of the current process.
2408  */
2409 void
2410 pgsigio(struct sigio *sigio, int sig, int checkctty)
2411 {
2412 	if (sigio == NULL)
2413 		return;
2414 
2415 	if (sigio->sio_pgid > 0) {
2416 		if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2417 		             sigio->sio_proc))
2418 			ksignal(sigio->sio_proc, sig);
2419 	} else if (sigio->sio_pgid < 0) {
2420 		struct proc *p;
2421 		struct pgrp *pg = sigio->sio_pgrp;
2422 
2423 		/*
2424 		 * Must interlock all signals against fork
2425 		 */
2426 		pgref(pg);
2427 		lockmgr(&pg->pg_lock, LK_EXCLUSIVE);
2428 		LIST_FOREACH(p, &pg->pg_members, p_pglist) {
2429 			if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2430 			    (checkctty == 0 || (p->p_flags & P_CONTROLT)))
2431 				ksignal(p, sig);
2432 		}
2433 		lockmgr(&pg->pg_lock, LK_RELEASE);
2434 		pgrel(pg);
2435 	}
2436 }
2437 
2438 static int
2439 filt_sigattach(struct knote *kn)
2440 {
2441 	struct proc *p = curproc;
2442 
2443 	kn->kn_ptr.p_proc = p;
2444 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
2445 
2446 	/* XXX lock the proc here while adding to the list? */
2447 	knote_insert(&p->p_klist, kn);
2448 
2449 	return (0);
2450 }
2451 
2452 static void
2453 filt_sigdetach(struct knote *kn)
2454 {
2455 	struct proc *p = kn->kn_ptr.p_proc;
2456 
2457 	knote_remove(&p->p_klist, kn);
2458 }
2459 
2460 /*
2461  * signal knotes are shared with proc knotes, so we apply a mask to
2462  * the hint in order to differentiate them from process hints.  This
2463  * could be avoided by using a signal-specific knote list, but probably
2464  * isn't worth the trouble.
2465  */
2466 static int
2467 filt_signal(struct knote *kn, long hint)
2468 {
2469 	if (hint & NOTE_SIGNAL) {
2470 		hint &= ~NOTE_SIGNAL;
2471 
2472 		if (kn->kn_id == hint)
2473 			kn->kn_data++;
2474 	}
2475 	return (kn->kn_data != 0);
2476 }
2477