xref: /netbsd-src/sys/kern/kern_sig.c (revision b7ae68fde0d8ef1c03714e8bbb1ee7c6118ea93b)
1 /*	$NetBSD: kern_sig.c,v 1.228 2006/09/02 06:29:13 christos Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_sig.c	8.14 (Berkeley) 5/14/95
37  */
38 
39 #include <sys/cdefs.h>
40 __KERNEL_RCSID(0, "$NetBSD: kern_sig.c,v 1.228 2006/09/02 06:29:13 christos Exp $");
41 
42 #include "opt_coredump.h"
43 #include "opt_ktrace.h"
44 #include "opt_ptrace.h"
45 #include "opt_multiprocessor.h"
46 #include "opt_compat_sunos.h"
47 #include "opt_compat_netbsd.h"
48 #include "opt_compat_netbsd32.h"
49 
50 #define	SIGPROP		/* include signal properties table */
51 #include <sys/param.h>
52 #include <sys/signalvar.h>
53 #include <sys/resourcevar.h>
54 #include <sys/namei.h>
55 #include <sys/vnode.h>
56 #include <sys/proc.h>
57 #include <sys/systm.h>
58 #include <sys/timeb.h>
59 #include <sys/times.h>
60 #include <sys/buf.h>
61 #include <sys/acct.h>
62 #include <sys/file.h>
63 #include <sys/kernel.h>
64 #include <sys/wait.h>
65 #include <sys/ktrace.h>
66 #include <sys/syslog.h>
67 #include <sys/stat.h>
68 #include <sys/core.h>
69 #include <sys/filedesc.h>
70 #include <sys/malloc.h>
71 #include <sys/pool.h>
72 #include <sys/ucontext.h>
73 #include <sys/sa.h>
74 #include <sys/savar.h>
75 #include <sys/exec.h>
76 #include <sys/sysctl.h>
77 #include <sys/kauth.h>
78 
79 #include <sys/mount.h>
80 #include <sys/syscallargs.h>
81 
82 #include <machine/cpu.h>
83 
84 #include <sys/user.h>		/* for coredump */
85 
86 #include <uvm/uvm.h>
87 #include <uvm/uvm_extern.h>
88 
89 #ifdef COREDUMP
90 static int	build_corename(struct proc *, char *, const char *, size_t);
91 #endif
92 static void	ksiginfo_exithook(struct proc *, void *);
93 static void	ksiginfo_put(struct proc *, const ksiginfo_t *);
94 static ksiginfo_t *ksiginfo_get(struct proc *, int);
95 static void	kpsignal2(struct proc *, const ksiginfo_t *, int);
96 
97 sigset_t	contsigmask, stopsigmask, sigcantmask;
98 
99 struct pool	sigacts_pool;	/* memory pool for sigacts structures */
100 
101 /*
102  * struct sigacts memory pool allocator.
103  */
104 
105 static void *
106 sigacts_poolpage_alloc(struct pool *pp, int flags)
107 {
108 
109 	return (void *)uvm_km_alloc(kernel_map,
110 	    (PAGE_SIZE)*2, (PAGE_SIZE)*2,
111 	    ((flags & PR_WAITOK) ? 0 : UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK)
112 	    | UVM_KMF_WIRED);
113 }
114 
115 static void
116 sigacts_poolpage_free(struct pool *pp, void *v)
117 {
118         uvm_km_free(kernel_map, (vaddr_t)v, (PAGE_SIZE)*2, UVM_KMF_WIRED);
119 }
120 
121 static struct pool_allocator sigactspool_allocator = {
122         .pa_alloc = sigacts_poolpage_alloc,
123 	.pa_free = sigacts_poolpage_free,
124 };
125 
126 POOL_INIT(siginfo_pool, sizeof(siginfo_t), 0, 0, 0, "siginfo",
127     &pool_allocator_nointr);
128 POOL_INIT(ksiginfo_pool, sizeof(ksiginfo_t), 0, 0, 0, "ksiginfo", NULL);
129 
130 /*
131  * Remove and return the first ksiginfo element that matches our requested
132  * signal, or return NULL if one not found.
133  */
134 static ksiginfo_t *
135 ksiginfo_get(struct proc *p, int signo)
136 {
137 	ksiginfo_t *ksi;
138 	int s;
139 
140 	s = splsoftclock();
141 	simple_lock(&p->p_sigctx.ps_silock);
142 	CIRCLEQ_FOREACH(ksi, &p->p_sigctx.ps_siginfo, ksi_list) {
143 		if (ksi->ksi_signo == signo) {
144 			CIRCLEQ_REMOVE(&p->p_sigctx.ps_siginfo, ksi, ksi_list);
145 			goto out;
146 		}
147 	}
148 	ksi = NULL;
149 out:
150 	simple_unlock(&p->p_sigctx.ps_silock);
151 	splx(s);
152 	return ksi;
153 }
154 
155 /*
156  * Append a new ksiginfo element to the list of pending ksiginfo's, if
157  * we need to (SA_SIGINFO was requested). We replace non RT signals if
158  * they already existed in the queue and we add new entries for RT signals,
159  * or for non RT signals with non-existing entries.
160  */
161 static void
162 ksiginfo_put(struct proc *p, const ksiginfo_t *ksi)
163 {
164 	ksiginfo_t *kp;
165 	struct sigaction *sa = &SIGACTION_PS(p->p_sigacts, ksi->ksi_signo);
166 	int s;
167 
168 	if ((sa->sa_flags & SA_SIGINFO) == 0)
169 		return;
170 	/*
171 	 * If there's no info, don't save it.
172 	 */
173 	if (KSI_EMPTY_P(ksi))
174 		return;
175 
176 	s = splsoftclock();
177 	simple_lock(&p->p_sigctx.ps_silock);
178 #ifdef notyet	/* XXX: QUEUING */
179 	if (ksi->ksi_signo < SIGRTMIN)
180 #endif
181 	{
182 		CIRCLEQ_FOREACH(kp, &p->p_sigctx.ps_siginfo, ksi_list) {
183 			if (kp->ksi_signo == ksi->ksi_signo) {
184 				KSI_COPY(ksi, kp);
185 				goto out;
186 			}
187 		}
188 	}
189 	kp = pool_get(&ksiginfo_pool, PR_NOWAIT);
190 	if (kp == NULL) {
191 #ifdef DIAGNOSTIC
192 		printf("Out of memory allocating siginfo for pid %d\n",
193 		    p->p_pid);
194 #endif
195 		goto out;
196 	}
197 	*kp = *ksi;
198 	CIRCLEQ_INSERT_TAIL(&p->p_sigctx.ps_siginfo, kp, ksi_list);
199 out:
200 	simple_unlock(&p->p_sigctx.ps_silock);
201 	splx(s);
202 }
203 
204 /*
205  * free all pending ksiginfo on exit
206  */
207 static void
208 ksiginfo_exithook(struct proc *p, void *v)
209 {
210 	int s;
211 
212 	s = splsoftclock();
213 	simple_lock(&p->p_sigctx.ps_silock);
214 	while (!CIRCLEQ_EMPTY(&p->p_sigctx.ps_siginfo)) {
215 		ksiginfo_t *ksi = CIRCLEQ_FIRST(&p->p_sigctx.ps_siginfo);
216 		CIRCLEQ_REMOVE(&p->p_sigctx.ps_siginfo, ksi, ksi_list);
217 		pool_put(&ksiginfo_pool, ksi);
218 	}
219 	simple_unlock(&p->p_sigctx.ps_silock);
220 	splx(s);
221 }
222 
223 /*
224  * Initialize signal-related data structures.
225  */
226 void
227 signal_init(void)
228 {
229 
230 	sigactspool_allocator.pa_pagesz = (PAGE_SIZE)*2;
231 
232 	pool_init(&sigacts_pool, sizeof(struct sigacts), 0, 0, 0, "sigapl",
233 	    sizeof(struct sigacts) > PAGE_SIZE ?
234 	    &sigactspool_allocator : &pool_allocator_nointr);
235 
236 	exithook_establish(ksiginfo_exithook, NULL);
237 	exechook_establish(ksiginfo_exithook, NULL);
238 }
239 
240 /*
241  * Create an initial sigctx structure, using the same signal state
242  * as p. If 'share' is set, share the sigctx_proc part, otherwise just
243  * copy it from parent.
244  */
245 void
246 sigactsinit(struct proc *np, struct proc *pp, int share)
247 {
248 	struct sigacts *ps;
249 
250 	if (share) {
251 		np->p_sigacts = pp->p_sigacts;
252 		pp->p_sigacts->sa_refcnt++;
253 	} else {
254 		ps = pool_get(&sigacts_pool, PR_WAITOK);
255 		if (pp)
256 			memcpy(ps, pp->p_sigacts, sizeof(struct sigacts));
257 		else
258 			memset(ps, '\0', sizeof(struct sigacts));
259 		ps->sa_refcnt = 1;
260 		np->p_sigacts = ps;
261 	}
262 }
263 
264 /*
265  * Make this process not share its sigctx, maintaining all
266  * signal state.
267  */
268 void
269 sigactsunshare(struct proc *p)
270 {
271 	struct sigacts *oldps;
272 
273 	if (p->p_sigacts->sa_refcnt == 1)
274 		return;
275 
276 	oldps = p->p_sigacts;
277 	sigactsinit(p, NULL, 0);
278 
279 	if (--oldps->sa_refcnt == 0)
280 		pool_put(&sigacts_pool, oldps);
281 }
282 
283 /*
284  * Release a sigctx structure.
285  */
286 void
287 sigactsfree(struct sigacts *ps)
288 {
289 
290 	if (--ps->sa_refcnt > 0)
291 		return;
292 
293 	pool_put(&sigacts_pool, ps);
294 }
295 
296 int
297 sigaction1(struct proc *p, int signum, const struct sigaction *nsa,
298 	struct sigaction *osa, const void *tramp, int vers)
299 {
300 	struct sigacts	*ps;
301 	int		prop;
302 
303 	ps = p->p_sigacts;
304 	if (signum <= 0 || signum >= NSIG)
305 		return (EINVAL);
306 
307 	/*
308 	 * Trampoline ABI version 0 is reserved for the legacy
309 	 * kernel-provided on-stack trampoline.  Conversely, if we are
310 	 * using a non-0 ABI version, we must have a trampoline.  Only
311 	 * validate the vers if a new sigaction was supplied. Emulations
312 	 * use legacy kernel trampolines with version 0, alternatively
313 	 * check for that too.
314 	 */
315 	if ((vers != 0 && tramp == NULL) ||
316 #ifdef SIGTRAMP_VALID
317 	    (nsa != NULL &&
318 	    ((vers == 0) ?
319 		(p->p_emul->e_sigcode == NULL) :
320 		!SIGTRAMP_VALID(vers))) ||
321 #endif
322 	    (vers == 0 && tramp != NULL))
323 		return (EINVAL);
324 
325 	if (osa)
326 		*osa = SIGACTION_PS(ps, signum);
327 
328 	if (nsa) {
329 		if (nsa->sa_flags & ~SA_ALLBITS)
330 			return (EINVAL);
331 
332 		prop = sigprop[signum];
333 		if (prop & SA_CANTMASK)
334 			return (EINVAL);
335 
336 		(void) splsched();	/* XXXSMP */
337 		SIGACTION_PS(ps, signum) = *nsa;
338 		ps->sa_sigdesc[signum].sd_tramp = tramp;
339 		ps->sa_sigdesc[signum].sd_vers = vers;
340 		sigminusset(&sigcantmask, &SIGACTION_PS(ps, signum).sa_mask);
341 		if ((prop & SA_NORESET) != 0)
342 			SIGACTION_PS(ps, signum).sa_flags &= ~SA_RESETHAND;
343 		if (signum == SIGCHLD) {
344 			if (nsa->sa_flags & SA_NOCLDSTOP)
345 				p->p_flag |= P_NOCLDSTOP;
346 			else
347 				p->p_flag &= ~P_NOCLDSTOP;
348 			if (nsa->sa_flags & SA_NOCLDWAIT) {
349 				/*
350 				 * Paranoia: since SA_NOCLDWAIT is implemented
351 				 * by reparenting the dying child to PID 1 (and
352 				 * trust it to reap the zombie), PID 1 itself
353 				 * is forbidden to set SA_NOCLDWAIT.
354 				 */
355 				if (p->p_pid == 1)
356 					p->p_flag &= ~P_NOCLDWAIT;
357 				else
358 					p->p_flag |= P_NOCLDWAIT;
359 			} else
360 				p->p_flag &= ~P_NOCLDWAIT;
361 
362 			if (nsa->sa_handler == SIG_IGN) {
363 				/*
364 				 * Paranoia: same as above.
365 				 */
366 				if (p->p_pid == 1)
367 					p->p_flag &= ~P_CLDSIGIGN;
368 				else
369 					p->p_flag |= P_CLDSIGIGN;
370 			} else
371 				p->p_flag &= ~P_CLDSIGIGN;
372 
373 		}
374 		if ((nsa->sa_flags & SA_NODEFER) == 0)
375 			sigaddset(&SIGACTION_PS(ps, signum).sa_mask, signum);
376 		else
377 			sigdelset(&SIGACTION_PS(ps, signum).sa_mask, signum);
378 		/*
379 	 	 * Set bit in p_sigctx.ps_sigignore for signals that are set to
380 		 * SIG_IGN, and for signals set to SIG_DFL where the default is
381 		 * to ignore. However, don't put SIGCONT in
382 		 * p_sigctx.ps_sigignore, as we have to restart the process.
383 	 	 */
384 		if (nsa->sa_handler == SIG_IGN ||
385 		    (nsa->sa_handler == SIG_DFL && (prop & SA_IGNORE) != 0)) {
386 						/* never to be seen again */
387 			sigdelset(&p->p_sigctx.ps_siglist, signum);
388 			if (signum != SIGCONT) {
389 						/* easier in psignal */
390 				sigaddset(&p->p_sigctx.ps_sigignore, signum);
391 			}
392 			sigdelset(&p->p_sigctx.ps_sigcatch, signum);
393 		} else {
394 			sigdelset(&p->p_sigctx.ps_sigignore, signum);
395 			if (nsa->sa_handler == SIG_DFL)
396 				sigdelset(&p->p_sigctx.ps_sigcatch, signum);
397 			else
398 				sigaddset(&p->p_sigctx.ps_sigcatch, signum);
399 		}
400 		(void) spl0();
401 	}
402 
403 	return (0);
404 }
405 
406 #ifdef COMPAT_16
407 /* ARGSUSED */
408 int
409 compat_16_sys___sigaction14(struct lwp *l, void *v, register_t *retval)
410 {
411 	struct compat_16_sys___sigaction14_args /* {
412 		syscallarg(int)				signum;
413 		syscallarg(const struct sigaction *)	nsa;
414 		syscallarg(struct sigaction *)		osa;
415 	} */ *uap = v;
416 	struct proc		*p;
417 	struct sigaction	nsa, osa;
418 	int			error;
419 
420 	if (SCARG(uap, nsa)) {
421 		error = copyin(SCARG(uap, nsa), &nsa, sizeof(nsa));
422 		if (error)
423 			return (error);
424 	}
425 	p = l->l_proc;
426 	error = sigaction1(p, SCARG(uap, signum),
427 	    SCARG(uap, nsa) ? &nsa : 0, SCARG(uap, osa) ? &osa : 0,
428 	    NULL, 0);
429 	if (error)
430 		return (error);
431 	if (SCARG(uap, osa)) {
432 		error = copyout(&osa, SCARG(uap, osa), sizeof(osa));
433 		if (error)
434 			return (error);
435 	}
436 	return (0);
437 }
438 #endif
439 
440 /* ARGSUSED */
441 int
442 sys___sigaction_sigtramp(struct lwp *l, void *v, register_t *retval)
443 {
444 	struct sys___sigaction_sigtramp_args /* {
445 		syscallarg(int)				signum;
446 		syscallarg(const struct sigaction *)	nsa;
447 		syscallarg(struct sigaction *)		osa;
448 		syscallarg(void *)			tramp;
449 		syscallarg(int)				vers;
450 	} */ *uap = v;
451 	struct proc *p = l->l_proc;
452 	struct sigaction nsa, osa;
453 	int error;
454 
455 	if (SCARG(uap, nsa)) {
456 		error = copyin(SCARG(uap, nsa), &nsa, sizeof(nsa));
457 		if (error)
458 			return (error);
459 	}
460 	error = sigaction1(p, SCARG(uap, signum),
461 	    SCARG(uap, nsa) ? &nsa : 0, SCARG(uap, osa) ? &osa : 0,
462 	    SCARG(uap, tramp), SCARG(uap, vers));
463 	if (error)
464 		return (error);
465 	if (SCARG(uap, osa)) {
466 		error = copyout(&osa, SCARG(uap, osa), sizeof(osa));
467 		if (error)
468 			return (error);
469 	}
470 	return (0);
471 }
472 
473 /*
474  * Initialize signal state for process 0;
475  * set to ignore signals that are ignored by default and disable the signal
476  * stack.
477  */
478 void
479 siginit(struct proc *p)
480 {
481 	struct sigacts	*ps;
482 	int		signum, prop;
483 
484 	ps = p->p_sigacts;
485 	sigemptyset(&contsigmask);
486 	sigemptyset(&stopsigmask);
487 	sigemptyset(&sigcantmask);
488 	for (signum = 1; signum < NSIG; signum++) {
489 		prop = sigprop[signum];
490 		if (prop & SA_CONT)
491 			sigaddset(&contsigmask, signum);
492 		if (prop & SA_STOP)
493 			sigaddset(&stopsigmask, signum);
494 		if (prop & SA_CANTMASK)
495 			sigaddset(&sigcantmask, signum);
496 		if (prop & SA_IGNORE && signum != SIGCONT)
497 			sigaddset(&p->p_sigctx.ps_sigignore, signum);
498 		sigemptyset(&SIGACTION_PS(ps, signum).sa_mask);
499 		SIGACTION_PS(ps, signum).sa_flags = SA_RESTART;
500 	}
501 	sigemptyset(&p->p_sigctx.ps_sigcatch);
502 	p->p_sigctx.ps_sigwaited = NULL;
503 	p->p_flag &= ~P_NOCLDSTOP;
504 
505 	/*
506 	 * Reset stack state to the user stack.
507 	 */
508 	p->p_sigctx.ps_sigstk.ss_flags = SS_DISABLE;
509 	p->p_sigctx.ps_sigstk.ss_size = 0;
510 	p->p_sigctx.ps_sigstk.ss_sp = 0;
511 
512 	/* One reference. */
513 	ps->sa_refcnt = 1;
514 }
515 
516 /*
517  * Reset signals for an exec of the specified process.
518  */
519 void
520 execsigs(struct proc *p)
521 {
522 	struct sigacts	*ps;
523 	int		signum, prop;
524 
525 	sigactsunshare(p);
526 
527 	ps = p->p_sigacts;
528 
529 	/*
530 	 * Reset caught signals.  Held signals remain held
531 	 * through p_sigctx.ps_sigmask (unless they were caught,
532 	 * and are now ignored by default).
533 	 */
534 	for (signum = 1; signum < NSIG; signum++) {
535 		if (sigismember(&p->p_sigctx.ps_sigcatch, signum)) {
536 			prop = sigprop[signum];
537 			if (prop & SA_IGNORE) {
538 				if ((prop & SA_CONT) == 0)
539 					sigaddset(&p->p_sigctx.ps_sigignore,
540 					    signum);
541 				sigdelset(&p->p_sigctx.ps_siglist, signum);
542 			}
543 			SIGACTION_PS(ps, signum).sa_handler = SIG_DFL;
544 		}
545 		sigemptyset(&SIGACTION_PS(ps, signum).sa_mask);
546 		SIGACTION_PS(ps, signum).sa_flags = SA_RESTART;
547 	}
548 	sigemptyset(&p->p_sigctx.ps_sigcatch);
549 	p->p_sigctx.ps_sigwaited = NULL;
550 
551 	/*
552 	 * Reset no zombies if child dies flag as Solaris does.
553 	 */
554 	p->p_flag &= ~(P_NOCLDWAIT | P_CLDSIGIGN);
555 	if (SIGACTION_PS(ps, SIGCHLD).sa_handler == SIG_IGN)
556 		SIGACTION_PS(ps, SIGCHLD).sa_handler = SIG_DFL;
557 
558 	/*
559 	 * Reset stack state to the user stack.
560 	 */
561 	p->p_sigctx.ps_sigstk.ss_flags = SS_DISABLE;
562 	p->p_sigctx.ps_sigstk.ss_size = 0;
563 	p->p_sigctx.ps_sigstk.ss_sp = 0;
564 }
565 
566 int
567 sigprocmask1(struct proc *p, int how, const sigset_t *nss, sigset_t *oss)
568 {
569 
570 	if (oss)
571 		*oss = p->p_sigctx.ps_sigmask;
572 
573 	if (nss) {
574 		(void)splsched();	/* XXXSMP */
575 		switch (how) {
576 		case SIG_BLOCK:
577 			sigplusset(nss, &p->p_sigctx.ps_sigmask);
578 			break;
579 		case SIG_UNBLOCK:
580 			sigminusset(nss, &p->p_sigctx.ps_sigmask);
581 			CHECKSIGS(p);
582 			break;
583 		case SIG_SETMASK:
584 			p->p_sigctx.ps_sigmask = *nss;
585 			CHECKSIGS(p);
586 			break;
587 		default:
588 			(void)spl0();	/* XXXSMP */
589 			return (EINVAL);
590 		}
591 		sigminusset(&sigcantmask, &p->p_sigctx.ps_sigmask);
592 		(void)spl0();		/* XXXSMP */
593 	}
594 
595 	return (0);
596 }
597 
598 /*
599  * Manipulate signal mask.
600  * Note that we receive new mask, not pointer,
601  * and return old mask as return value;
602  * the library stub does the rest.
603  */
604 int
605 sys___sigprocmask14(struct lwp *l, void *v, register_t *retval)
606 {
607 	struct sys___sigprocmask14_args /* {
608 		syscallarg(int)			how;
609 		syscallarg(const sigset_t *)	set;
610 		syscallarg(sigset_t *)		oset;
611 	} */ *uap = v;
612 	struct proc	*p;
613 	sigset_t	nss, oss;
614 	int		error;
615 
616 	if (SCARG(uap, set)) {
617 		error = copyin(SCARG(uap, set), &nss, sizeof(nss));
618 		if (error)
619 			return (error);
620 	}
621 	p = l->l_proc;
622 	error = sigprocmask1(p, SCARG(uap, how),
623 	    SCARG(uap, set) ? &nss : 0, SCARG(uap, oset) ? &oss : 0);
624 	if (error)
625 		return (error);
626 	if (SCARG(uap, oset)) {
627 		error = copyout(&oss, SCARG(uap, oset), sizeof(oss));
628 		if (error)
629 			return (error);
630 	}
631 	return (0);
632 }
633 
634 void
635 sigpending1(struct proc *p, sigset_t *ss)
636 {
637 
638 	*ss = p->p_sigctx.ps_siglist;
639 	sigminusset(&p->p_sigctx.ps_sigmask, ss);
640 }
641 
642 /* ARGSUSED */
643 int
644 sys___sigpending14(struct lwp *l, void *v, register_t *retval)
645 {
646 	struct sys___sigpending14_args /* {
647 		syscallarg(sigset_t *)	set;
648 	} */ *uap = v;
649 	struct proc	*p;
650 	sigset_t	ss;
651 
652 	p = l->l_proc;
653 	sigpending1(p, &ss);
654 	return (copyout(&ss, SCARG(uap, set), sizeof(ss)));
655 }
656 
657 int
658 sigsuspend1(struct proc *p, const sigset_t *ss)
659 {
660 	struct sigacts *ps;
661 
662 	ps = p->p_sigacts;
663 	if (ss) {
664 		/*
665 		 * When returning from sigpause, we want
666 		 * the old mask to be restored after the
667 		 * signal handler has finished.  Thus, we
668 		 * save it here and mark the sigctx structure
669 		 * to indicate this.
670 		 */
671 		p->p_sigctx.ps_oldmask = p->p_sigctx.ps_sigmask;
672 		p->p_sigctx.ps_flags |= SAS_OLDMASK;
673 		(void) splsched();	/* XXXSMP */
674 		p->p_sigctx.ps_sigmask = *ss;
675 		CHECKSIGS(p);
676 		sigminusset(&sigcantmask, &p->p_sigctx.ps_sigmask);
677 		(void) spl0();		/* XXXSMP */
678 	}
679 
680 	while (tsleep((caddr_t) ps, PPAUSE|PCATCH, "pause", 0) == 0)
681 		/* void */;
682 
683 	/* always return EINTR rather than ERESTART... */
684 	return (EINTR);
685 }
686 
687 /*
688  * Suspend process until signal, providing mask to be set
689  * in the meantime.  Note nonstandard calling convention:
690  * libc stub passes mask, not pointer, to save a copyin.
691  */
692 /* ARGSUSED */
693 int
694 sys___sigsuspend14(struct lwp *l, void *v, register_t *retval)
695 {
696 	struct sys___sigsuspend14_args /* {
697 		syscallarg(const sigset_t *)	set;
698 	} */ *uap = v;
699 	struct proc	*p;
700 	sigset_t	ss;
701 	int		error;
702 
703 	if (SCARG(uap, set)) {
704 		error = copyin(SCARG(uap, set), &ss, sizeof(ss));
705 		if (error)
706 			return (error);
707 	}
708 
709 	p = l->l_proc;
710 	return (sigsuspend1(p, SCARG(uap, set) ? &ss : 0));
711 }
712 
713 int
714 sigaltstack1(struct proc *p, const struct sigaltstack *nss,
715 	struct sigaltstack *oss)
716 {
717 
718 	if (oss)
719 		*oss = p->p_sigctx.ps_sigstk;
720 
721 	if (nss) {
722 		if (nss->ss_flags & ~SS_ALLBITS)
723 			return (EINVAL);
724 
725 		if (nss->ss_flags & SS_DISABLE) {
726 			if (p->p_sigctx.ps_sigstk.ss_flags & SS_ONSTACK)
727 				return (EINVAL);
728 		} else {
729 			if (nss->ss_size < MINSIGSTKSZ)
730 				return (ENOMEM);
731 		}
732 		p->p_sigctx.ps_sigstk = *nss;
733 	}
734 
735 	return (0);
736 }
737 
738 /* ARGSUSED */
739 int
740 sys___sigaltstack14(struct lwp *l, void *v, register_t *retval)
741 {
742 	struct sys___sigaltstack14_args /* {
743 		syscallarg(const struct sigaltstack *)	nss;
744 		syscallarg(struct sigaltstack *)	oss;
745 	} */ *uap = v;
746 	struct proc		*p;
747 	struct sigaltstack	nss, oss;
748 	int			error;
749 
750 	if (SCARG(uap, nss)) {
751 		error = copyin(SCARG(uap, nss), &nss, sizeof(nss));
752 		if (error)
753 			return (error);
754 	}
755 	p = l->l_proc;
756 	error = sigaltstack1(p,
757 	    SCARG(uap, nss) ? &nss : 0, SCARG(uap, oss) ? &oss : 0);
758 	if (error)
759 		return (error);
760 	if (SCARG(uap, oss)) {
761 		error = copyout(&oss, SCARG(uap, oss), sizeof(oss));
762 		if (error)
763 			return (error);
764 	}
765 	return (0);
766 }
767 
768 /* ARGSUSED */
769 int
770 sys_kill(struct lwp *l, void *v, register_t *retval)
771 {
772 	struct sys_kill_args /* {
773 		syscallarg(int)	pid;
774 		syscallarg(int)	signum;
775 	} */ *uap = v;
776 	struct proc	*p;
777 	ksiginfo_t	ksi;
778 	int signum = SCARG(uap, signum);
779 	int error;
780 
781 	if ((u_int)signum >= NSIG)
782 		return (EINVAL);
783 	KSI_INIT(&ksi);
784 	ksi.ksi_signo = signum;
785 	ksi.ksi_code = SI_USER;
786 	ksi.ksi_pid = l->l_proc->p_pid;
787 	ksi.ksi_uid = kauth_cred_geteuid(l->l_cred);
788 	if (SCARG(uap, pid) > 0) {
789 		/* kill single process */
790 		if ((p = pfind(SCARG(uap, pid))) == NULL)
791 			return (ESRCH);
792 		error = kauth_authorize_process(l->l_cred,
793 		    KAUTH_PROCESS_CANSIGNAL, p, (void *)(uintptr_t)signum,
794 		    NULL, NULL);
795 		if (error)
796 			return error;
797 		if (signum)
798 			kpsignal2(p, &ksi, 1);
799 		return (0);
800 	}
801 	switch (SCARG(uap, pid)) {
802 	case -1:		/* broadcast signal */
803 		return (killpg1(l, &ksi, 0, 1));
804 	case 0:			/* signal own process group */
805 		return (killpg1(l, &ksi, 0, 0));
806 	default:		/* negative explicit process group */
807 		return (killpg1(l, &ksi, -SCARG(uap, pid), 0));
808 	}
809 	/* NOTREACHED */
810 }
811 
812 /*
813  * Common code for kill process group/broadcast kill.
814  * cp is calling process.
815  */
816 int
817 killpg1(struct lwp *l, ksiginfo_t *ksi, int pgid, int all)
818 {
819 	struct proc	*p, *cp;
820 	kauth_cred_t	pc;
821 	struct pgrp	*pgrp;
822 	int		nfound;
823 	int		signum = ksi->ksi_signo;
824 
825 	cp = l->l_proc;
826 	pc = l->l_cred;
827 	nfound = 0;
828 	if (all) {
829 		/*
830 		 * broadcast
831 		 */
832 		proclist_lock_read();
833 		PROCLIST_FOREACH(p, &allproc) {
834 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM || p == cp ||
835 			    kauth_authorize_process(pc, KAUTH_PROCESS_CANSIGNAL,
836 			    p, (void *)(uintptr_t)signum, NULL, NULL) != 0)
837 				continue;
838 			nfound++;
839 			if (signum)
840 				kpsignal2(p, ksi, 1);
841 		}
842 		proclist_unlock_read();
843 	} else {
844 		if (pgid == 0)
845 			/*
846 			 * zero pgid means send to my process group.
847 			 */
848 			pgrp = cp->p_pgrp;
849 		else {
850 			pgrp = pgfind(pgid);
851 			if (pgrp == NULL)
852 				return (ESRCH);
853 		}
854 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
855 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
856 			    kauth_authorize_process(pc, KAUTH_PROCESS_CANSIGNAL,
857 			    p, (void *)(uintptr_t)signum, NULL, NULL) != 0)
858 				continue;
859 			nfound++;
860 			if (signum && P_ZOMBIE(p) == 0)
861 				kpsignal2(p, ksi, 1);
862 		}
863 	}
864 	return (nfound ? 0 : ESRCH);
865 }
866 
867 /*
868  * Send a signal to a process group.
869  */
870 void
871 gsignal(int pgid, int signum)
872 {
873 	ksiginfo_t ksi;
874 	KSI_INIT_EMPTY(&ksi);
875 	ksi.ksi_signo = signum;
876 	kgsignal(pgid, &ksi, NULL);
877 }
878 
879 void
880 kgsignal(int pgid, ksiginfo_t *ksi, void *data)
881 {
882 	struct pgrp *pgrp;
883 
884 	if (pgid && (pgrp = pgfind(pgid)))
885 		kpgsignal(pgrp, ksi, data, 0);
886 }
887 
888 /*
889  * Send a signal to a process group. If checktty is 1,
890  * limit to members which have a controlling terminal.
891  */
892 void
893 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
894 {
895 	ksiginfo_t ksi;
896 	KSI_INIT_EMPTY(&ksi);
897 	ksi.ksi_signo = sig;
898 	kpgsignal(pgrp, &ksi, NULL, checkctty);
899 }
900 
901 void
902 kpgsignal(struct pgrp *pgrp, ksiginfo_t *ksi, void *data, int checkctty)
903 {
904 	struct proc *p;
905 
906 	if (pgrp)
907 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist)
908 			if (checkctty == 0 || p->p_flag & P_CONTROLT)
909 				kpsignal(p, ksi, data);
910 }
911 
912 /*
913  * Send a signal caused by a trap to the current process.
914  * If it will be caught immediately, deliver it with correct code.
915  * Otherwise, post it normally.
916  */
917 void
918 trapsignal(struct lwp *l, const ksiginfo_t *ksi)
919 {
920 	struct proc	*p;
921 	struct sigacts	*ps;
922 	int signum = ksi->ksi_signo;
923 
924 	KASSERT(KSI_TRAP_P(ksi));
925 
926 	p = l->l_proc;
927 	ps = p->p_sigacts;
928 	if ((p->p_flag & P_TRACED) == 0 &&
929 	    sigismember(&p->p_sigctx.ps_sigcatch, signum) &&
930 	    !sigismember(&p->p_sigctx.ps_sigmask, signum)) {
931 		p->p_stats->p_ru.ru_nsignals++;
932 #ifdef KTRACE
933 		if (KTRPOINT(p, KTR_PSIG))
934 			ktrpsig(l, signum, SIGACTION_PS(ps, signum).sa_handler,
935 			    &p->p_sigctx.ps_sigmask, ksi);
936 #endif
937 		kpsendsig(l, ksi, &p->p_sigctx.ps_sigmask);
938 		(void) splsched();	/* XXXSMP */
939 		sigplusset(&SIGACTION_PS(ps, signum).sa_mask,
940 		    &p->p_sigctx.ps_sigmask);
941 		if (SIGACTION_PS(ps, signum).sa_flags & SA_RESETHAND) {
942 			sigdelset(&p->p_sigctx.ps_sigcatch, signum);
943 			if (signum != SIGCONT && sigprop[signum] & SA_IGNORE)
944 				sigaddset(&p->p_sigctx.ps_sigignore, signum);
945 			SIGACTION_PS(ps, signum).sa_handler = SIG_DFL;
946 		}
947 		(void) spl0();		/* XXXSMP */
948 	} else {
949 		p->p_sigctx.ps_lwp = l->l_lid;
950 		/* XXX for core dump/debugger */
951 		p->p_sigctx.ps_signo = ksi->ksi_signo;
952 		p->p_sigctx.ps_code = ksi->ksi_trap;
953 		kpsignal2(p, ksi, 1);
954 	}
955 }
956 
957 /*
958  * Fill in signal information and signal the parent for a child status change.
959  */
960 void
961 child_psignal(struct proc *p, int dolock)
962 {
963 	ksiginfo_t ksi;
964 
965 	KSI_INIT(&ksi);
966 	ksi.ksi_signo = SIGCHLD;
967 	ksi.ksi_code = p->p_xstat == SIGCONT ? CLD_CONTINUED : CLD_STOPPED;
968 	ksi.ksi_pid = p->p_pid;
969 	ksi.ksi_uid = kauth_cred_geteuid(p->p_cred);
970 	ksi.ksi_status = p->p_xstat;
971 	ksi.ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
972 	ksi.ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
973 	kpsignal2(p->p_pptr, &ksi, dolock);
974 }
975 
976 /*
977  * Send the signal to the process.  If the signal has an action, the action
978  * is usually performed by the target process rather than the caller; we add
979  * the signal to the set of pending signals for the process.
980  *
981  * Exceptions:
982  *   o When a stop signal is sent to a sleeping process that takes the
983  *     default action, the process is stopped without awakening it.
984  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
985  *     regardless of the signal action (eg, blocked or ignored).
986  *
987  * Other ignored signals are discarded immediately.
988  *
989  * XXXSMP: Invoked as psignal() or sched_psignal().
990  */
991 void
992 psignal1(struct proc *p, int signum, int dolock)
993 {
994 	ksiginfo_t ksi;
995 
996 	KSI_INIT_EMPTY(&ksi);
997 	ksi.ksi_signo = signum;
998 	kpsignal2(p, &ksi, dolock);
999 }
1000 
1001 void
1002 kpsignal1(struct proc *p, ksiginfo_t *ksi, void *data, int dolock)
1003 {
1004 
1005 	if ((p->p_flag & P_WEXIT) == 0 && data) {
1006 		size_t fd;
1007 		struct filedesc *fdp = p->p_fd;
1008 
1009 		ksi->ksi_fd = -1;
1010 		for (fd = 0; fd < fdp->fd_nfiles; fd++) {
1011 			struct file *fp = fdp->fd_ofiles[fd];
1012 			/* XXX: lock? */
1013 			if (fp && fp->f_data == data) {
1014 				ksi->ksi_fd = fd;
1015 				break;
1016 			}
1017 		}
1018 	}
1019 	kpsignal2(p, ksi, dolock);
1020 }
1021 
1022 static void
1023 kpsignal2(struct proc *p, const ksiginfo_t *ksi, int dolock)
1024 {
1025 	struct lwp *l, *suspended = NULL;
1026 	struct sadata_vp *vp;
1027 	int	s = 0, prop, allsusp;
1028 	sig_t	action;
1029 	int	signum = ksi->ksi_signo;
1030 
1031 #ifdef DIAGNOSTIC
1032 	if (signum <= 0 || signum >= NSIG)
1033 		panic("psignal signal number %d", signum);
1034 
1035 	/* XXXSMP: works, but icky */
1036 	if (dolock)
1037 		SCHED_ASSERT_UNLOCKED();
1038 	else
1039 		SCHED_ASSERT_LOCKED();
1040 #endif
1041 
1042 	/*
1043 	 * Notify any interested parties in the signal.
1044 	 */
1045 	KNOTE(&p->p_klist, NOTE_SIGNAL | signum);
1046 
1047 	prop = sigprop[signum];
1048 
1049 	/*
1050 	 * If proc is traced, always give parent a chance.
1051 	 */
1052 	if (p->p_flag & P_TRACED) {
1053 		action = SIG_DFL;
1054 
1055 		/*
1056 		 * If the process is being traced and the signal is being
1057 		 * caught, make sure to save any ksiginfo.
1058 		 */
1059 		if (sigismember(&p->p_sigctx.ps_sigcatch, signum))
1060 			ksiginfo_put(p, ksi);
1061 	} else {
1062 		/*
1063 		 * If the signal was the result of a trap, reset it
1064 		 * to default action if it's currently masked, so that it would
1065 		 * coredump immediatelly instead of spinning repeatedly
1066 		 * taking the signal.
1067 		 */
1068 		if (KSI_TRAP_P(ksi)
1069 		    && sigismember(&p->p_sigctx.ps_sigmask, signum)
1070 		    && !sigismember(&p->p_sigctx.ps_sigcatch, signum)) {
1071 			sigdelset(&p->p_sigctx.ps_sigignore, signum);
1072 			sigdelset(&p->p_sigctx.ps_sigcatch, signum);
1073 			sigdelset(&p->p_sigctx.ps_sigmask, signum);
1074 			SIGACTION(p, signum).sa_handler = SIG_DFL;
1075 		}
1076 
1077 		/*
1078 		 * If the signal is being ignored,
1079 		 * then we forget about it immediately.
1080 		 * (Note: we don't set SIGCONT in p_sigctx.ps_sigignore,
1081 		 * and if it is set to SIG_IGN,
1082 		 * action will be SIG_DFL here.)
1083 		 */
1084 		if (sigismember(&p->p_sigctx.ps_sigignore, signum))
1085 			return;
1086 		if (sigismember(&p->p_sigctx.ps_sigmask, signum))
1087 			action = SIG_HOLD;
1088 		else if (sigismember(&p->p_sigctx.ps_sigcatch, signum))
1089 			action = SIG_CATCH;
1090 		else {
1091 			action = SIG_DFL;
1092 
1093 			if (prop & SA_KILL && p->p_nice > NZERO)
1094 				p->p_nice = NZERO;
1095 
1096 			/*
1097 			 * If sending a tty stop signal to a member of an
1098 			 * orphaned process group, discard the signal here if
1099 			 * the action is default; don't stop the process below
1100 			 * if sleeping, and don't clear any pending SIGCONT.
1101 			 */
1102 			if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0)
1103 				return;
1104 		}
1105 	}
1106 
1107 	if (prop & SA_CONT)
1108 		sigminusset(&stopsigmask, &p->p_sigctx.ps_siglist);
1109 
1110 	if (prop & SA_STOP)
1111 		sigminusset(&contsigmask, &p->p_sigctx.ps_siglist);
1112 
1113 	/*
1114 	 * If the signal doesn't have SA_CANTMASK (no override for SIGKILL,
1115 	 * please!), check if anything waits on it. If yes, save the
1116 	 * info into provided ps_sigwaited, and wake-up the waiter.
1117 	 * The signal won't be processed further here.
1118 	 */
1119 	if ((prop & SA_CANTMASK) == 0
1120 	    && p->p_sigctx.ps_sigwaited
1121 	    && sigismember(p->p_sigctx.ps_sigwait, signum)
1122 	    && p->p_stat != SSTOP) {
1123 		p->p_sigctx.ps_sigwaited->ksi_info = ksi->ksi_info;
1124 		p->p_sigctx.ps_sigwaited = NULL;
1125 		if (dolock)
1126 			wakeup_one(&p->p_sigctx.ps_sigwait);
1127 		else
1128 			sched_wakeup(&p->p_sigctx.ps_sigwait);
1129 		return;
1130 	}
1131 
1132 	sigaddset(&p->p_sigctx.ps_siglist, signum);
1133 
1134 	/* CHECKSIGS() is "inlined" here. */
1135 	p->p_sigctx.ps_sigcheck = 1;
1136 
1137 	/*
1138 	 * Defer further processing for signals which are held,
1139 	 * except that stopped processes must be continued by SIGCONT.
1140 	 */
1141 	if (action == SIG_HOLD &&
1142 	    ((prop & SA_CONT) == 0 || p->p_stat != SSTOP)) {
1143 		ksiginfo_put(p, ksi);
1144 		return;
1145 	}
1146 	/* XXXSMP: works, but icky */
1147 	if (dolock)
1148 		SCHED_LOCK(s);
1149 
1150 	if (p->p_flag & P_SA) {
1151 		allsusp = 0;
1152 		l = NULL;
1153 		if (p->p_stat == SACTIVE) {
1154 			SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) {
1155 				l = vp->savp_lwp;
1156 				KDASSERT(l != NULL);
1157 				if (l->l_flag & L_SA_IDLE) {
1158 					/* wakeup idle LWP */
1159 					goto found;
1160 					/*NOTREACHED*/
1161 				} else if (l->l_flag & L_SA_YIELD) {
1162 					/* idle LWP is already waking up */
1163 					goto out;
1164 					/*NOTREACHED*/
1165 				}
1166 			}
1167 			SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) {
1168 				l = vp->savp_lwp;
1169 				if (l->l_stat == LSRUN ||
1170 				    l->l_stat == LSONPROC) {
1171 					signotify(p);
1172 					goto out;
1173 					/*NOTREACHED*/
1174 				}
1175 				if (l->l_stat == LSSLEEP &&
1176 				    l->l_flag & L_SINTR) {
1177 					/* ok to signal vp lwp */
1178 					break;
1179 				} else
1180 					l = NULL;
1181 			}
1182 		} else if (p->p_stat == SSTOP) {
1183 			SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) {
1184 				l = vp->savp_lwp;
1185 				if (l->l_stat == LSSLEEP && (l->l_flag & L_SINTR) != 0)
1186 					break;
1187 				l = NULL;
1188 			}
1189 		}
1190 	} else if (p->p_nrlwps > 0 && (p->p_stat != SSTOP)) {
1191 		/*
1192 		 * At least one LWP is running or on a run queue.
1193 		 * The signal will be noticed when one of them returns
1194 		 * to userspace.
1195 		 */
1196 		signotify(p);
1197 		/*
1198 		 * The signal will be noticed very soon.
1199 		 */
1200 		goto out;
1201 		/*NOTREACHED*/
1202 	} else {
1203 		/*
1204 		 * Find out if any of the sleeps are interruptable,
1205 		 * and if all the live LWPs remaining are suspended.
1206 		 */
1207 		allsusp = 1;
1208 		LIST_FOREACH(l, &p->p_lwps, l_sibling) {
1209 			if (l->l_stat == LSSLEEP &&
1210 			    l->l_flag & L_SINTR)
1211 				break;
1212 			if (l->l_stat == LSSUSPENDED)
1213 				suspended = l;
1214 			else if ((l->l_stat != LSZOMB) &&
1215 			    (l->l_stat != LSDEAD))
1216 				allsusp = 0;
1217 		}
1218 	}
1219 
1220  found:
1221 	switch (p->p_stat) {
1222 	case SACTIVE:
1223 
1224 		if (l != NULL && (p->p_flag & P_TRACED))
1225 			goto run;
1226 
1227 		/*
1228 		 * If SIGCONT is default (or ignored) and process is
1229 		 * asleep, we are finished; the process should not
1230 		 * be awakened.
1231 		 */
1232 		if ((prop & SA_CONT) && action == SIG_DFL) {
1233 			sigdelset(&p->p_sigctx.ps_siglist, signum);
1234 			goto done;
1235 		}
1236 
1237 		/*
1238 		 * When a sleeping process receives a stop
1239 		 * signal, process immediately if possible.
1240 		 */
1241 		if ((prop & SA_STOP) && action == SIG_DFL) {
1242 			/*
1243 			 * If a child holding parent blocked,
1244 			 * stopping could cause deadlock.
1245 			 */
1246 			if (p->p_flag & P_PPWAIT) {
1247 				goto out;
1248 			}
1249 			sigdelset(&p->p_sigctx.ps_siglist, signum);
1250 			p->p_xstat = signum;
1251 			if ((p->p_pptr->p_flag & P_NOCLDSTOP) == 0) {
1252 				/*
1253 				 * XXXSMP: recursive call; don't lock
1254 				 * the second time around.
1255 				 */
1256 				child_psignal(p, 0);
1257 			}
1258 			proc_stop(p, 1);	/* XXXSMP: recurse? */
1259 			goto done;
1260 		}
1261 
1262 		if (l == NULL) {
1263 			/*
1264 			 * Special case: SIGKILL of a process
1265 			 * which is entirely composed of
1266 			 * suspended LWPs should succeed. We
1267 			 * make this happen by unsuspending one of
1268 			 * them.
1269 			 */
1270 			if (allsusp && (signum == SIGKILL)) {
1271 				lwp_continue(suspended);
1272 			}
1273 			goto done;
1274 		}
1275 		/*
1276 		 * All other (caught or default) signals
1277 		 * cause the process to run.
1278 		 */
1279 		goto runfast;
1280 		/*NOTREACHED*/
1281 	case SSTOP:
1282 		/* Process is stopped */
1283 		/*
1284 		 * If traced process is already stopped,
1285 		 * then no further action is necessary.
1286 		 */
1287 		if (p->p_flag & P_TRACED)
1288 			goto done;
1289 
1290 		/*
1291 		 * Kill signal always sets processes running,
1292 		 * if possible.
1293 		 */
1294 		if (signum == SIGKILL) {
1295 			l = proc_unstop(p);
1296 			if (l)
1297 				goto runfast;
1298 			goto done;
1299 		}
1300 
1301 		if (prop & SA_CONT) {
1302 			/*
1303 			 * If SIGCONT is default (or ignored),
1304 			 * we continue the process but don't
1305 			 * leave the signal in ps_siglist, as
1306 			 * it has no further action.  If
1307 			 * SIGCONT is held, we continue the
1308 			 * process and leave the signal in
1309 			 * ps_siglist.  If the process catches
1310 			 * SIGCONT, let it handle the signal
1311 			 * itself.  If it isn't waiting on an
1312 			 * event, then it goes back to run
1313 			 * state.  Otherwise, process goes
1314 			 * back to sleep state.
1315 			 */
1316 			if (action == SIG_DFL)
1317 				sigdelset(&p->p_sigctx.ps_siglist,
1318 				    signum);
1319 			l = proc_unstop(p);
1320 			if (l && (action == SIG_CATCH))
1321 				goto runfast;
1322 			goto out;
1323 		}
1324 
1325 		if (prop & SA_STOP) {
1326 			/*
1327 			 * Already stopped, don't need to stop again.
1328 			 * (If we did the shell could get confused.)
1329 			 */
1330 			sigdelset(&p->p_sigctx.ps_siglist, signum);
1331 			goto done;
1332 		}
1333 
1334 		/*
1335 		 * If a lwp is sleeping interruptibly, then
1336 		 * wake it up; it will run until the kernel
1337 		 * boundary, where it will stop in issignal(),
1338 		 * since p->p_stat is still SSTOP. When the
1339 		 * process is continued, it will be made
1340 		 * runnable and can look at the signal.
1341 		 */
1342 		if (l)
1343 			goto run;
1344 		goto out;
1345 	case SIDL:
1346 		/* Process is being created by fork */
1347 		/* XXX: We are not ready to receive signals yet */
1348 		goto done;
1349 	default:
1350 		/* Else what? */
1351 		panic("psignal: Invalid process state %d.", p->p_stat);
1352 	}
1353 	/*NOTREACHED*/
1354 
1355  runfast:
1356 	if (action == SIG_CATCH) {
1357 		ksiginfo_put(p, ksi);
1358 		action = SIG_HOLD;
1359 	}
1360 	/*
1361 	 * Raise priority to at least PUSER.
1362 	 */
1363 	if (l->l_priority > PUSER)
1364 		l->l_priority = PUSER;
1365  run:
1366 	if (action == SIG_CATCH) {
1367 		ksiginfo_put(p, ksi);
1368 		action = SIG_HOLD;
1369 	}
1370 
1371 	setrunnable(l);		/* XXXSMP: recurse? */
1372  out:
1373 	if (action == SIG_CATCH)
1374 		ksiginfo_put(p, ksi);
1375  done:
1376 	/* XXXSMP: works, but icky */
1377 	if (dolock)
1378 		SCHED_UNLOCK(s);
1379 }
1380 
1381 siginfo_t *
1382 siginfo_alloc(int flags)
1383 {
1384 
1385 	return pool_get(&siginfo_pool, flags);
1386 }
1387 
1388 void
1389 siginfo_free(void *arg)
1390 {
1391 
1392 	pool_put(&siginfo_pool, arg);
1393 }
1394 
1395 void
1396 kpsendsig(struct lwp *l, const ksiginfo_t *ksi, const sigset_t *mask)
1397 {
1398 	struct proc *p = l->l_proc;
1399 	struct lwp *le, *li;
1400 	siginfo_t *si;
1401 	int f;
1402 
1403 	if (p->p_flag & P_SA) {
1404 
1405 		/* XXXUPSXXX What if not on sa_vp ? */
1406 
1407 		f = l->l_flag & L_SA;
1408 		l->l_flag &= ~L_SA;
1409 		si = siginfo_alloc(PR_WAITOK);
1410 		si->_info = ksi->ksi_info;
1411 		le = li = NULL;
1412 		if (KSI_TRAP_P(ksi))
1413 			le = l;
1414 		else
1415 			li = l;
1416 		if (sa_upcall(l, SA_UPCALL_SIGNAL | SA_UPCALL_DEFER, le, li,
1417 		    sizeof(*si), si, siginfo_free) != 0) {
1418 			siginfo_free(si);
1419 #if 0
1420 			if (KSI_TRAP_P(ksi))
1421 				/* XXX What do we do here?? */;
1422 #endif
1423 		}
1424 		l->l_flag |= f;
1425 		return;
1426 	}
1427 
1428 	(*p->p_emul->e_sendsig)(ksi, mask);
1429 }
1430 
1431 static inline int firstsig(const sigset_t *);
1432 
1433 static inline int
1434 firstsig(const sigset_t *ss)
1435 {
1436 	int sig;
1437 
1438 	sig = ffs(ss->__bits[0]);
1439 	if (sig != 0)
1440 		return (sig);
1441 #if NSIG > 33
1442 	sig = ffs(ss->__bits[1]);
1443 	if (sig != 0)
1444 		return (sig + 32);
1445 #endif
1446 #if NSIG > 65
1447 	sig = ffs(ss->__bits[2]);
1448 	if (sig != 0)
1449 		return (sig + 64);
1450 #endif
1451 #if NSIG > 97
1452 	sig = ffs(ss->__bits[3]);
1453 	if (sig != 0)
1454 		return (sig + 96);
1455 #endif
1456 	return (0);
1457 }
1458 
1459 /*
1460  * If the current process has received a signal (should be caught or cause
1461  * termination, should interrupt current syscall), return the signal number.
1462  * Stop signals with default action are processed immediately, then cleared;
1463  * they aren't returned.  This is checked after each entry to the system for
1464  * a syscall or trap (though this can usually be done without calling issignal
1465  * by checking the pending signal masks in the CURSIG macro.) The normal call
1466  * sequence is
1467  *
1468  *	while (signum = CURSIG(curlwp))
1469  *		postsig(signum);
1470  */
1471 int
1472 issignal(struct lwp *l)
1473 {
1474 	struct proc	*p = l->l_proc;
1475 	int		s = 0, signum, prop;
1476 	int		dolock = (l->l_flag & L_SINTR) == 0, locked = !dolock;
1477 	sigset_t	ss;
1478 
1479 	/* Bail out if we do not own the virtual processor */
1480 	if (l->l_flag & L_SA && l->l_savp->savp_lwp != l)
1481 		return 0;
1482 
1483 	if (p->p_stat == SSTOP) {
1484 		/*
1485 		 * The process is stopped/stopping. Stop ourselves now that
1486 		 * we're on the kernel/userspace boundary.
1487 		 */
1488 		if (dolock)
1489 			SCHED_LOCK(s);
1490 		l->l_stat = LSSTOP;
1491 		p->p_nrlwps--;
1492 		if (p->p_flag & P_TRACED)
1493 			goto sigtraceswitch;
1494 		else
1495 			goto sigswitch;
1496 	}
1497 	for (;;) {
1498 		sigpending1(p, &ss);
1499 		if (p->p_flag & P_PPWAIT)
1500 			sigminusset(&stopsigmask, &ss);
1501 		signum = firstsig(&ss);
1502 		if (signum == 0) {		 	/* no signal to send */
1503 			p->p_sigctx.ps_sigcheck = 0;
1504 			if (locked && dolock)
1505 				SCHED_LOCK(s);
1506 			return (0);
1507 		}
1508 							/* take the signal! */
1509 		sigdelset(&p->p_sigctx.ps_siglist, signum);
1510 
1511 		/*
1512 		 * We should see pending but ignored signals
1513 		 * only if P_TRACED was on when they were posted.
1514 		 */
1515 		if (sigismember(&p->p_sigctx.ps_sigignore, signum) &&
1516 		    (p->p_flag & P_TRACED) == 0)
1517 			continue;
1518 
1519 		if (p->p_flag & P_TRACED && (p->p_flag & P_PPWAIT) == 0) {
1520 			/*
1521 			 * If traced, always stop, and stay
1522 			 * stopped until released by the debugger.
1523 			 */
1524 			p->p_xstat = signum;
1525 
1526 			/* Emulation-specific handling of signal trace */
1527 			if ((p->p_emul->e_tracesig != NULL) &&
1528 			    ((*p->p_emul->e_tracesig)(p, signum) != 0))
1529 				goto childresumed;
1530 
1531 			if ((p->p_flag & P_FSTRACE) == 0)
1532 				child_psignal(p, dolock);
1533 			if (dolock)
1534 				SCHED_LOCK(s);
1535 			proc_stop(p, 1);
1536 		sigtraceswitch:
1537 			mi_switch(l, NULL);
1538 			SCHED_ASSERT_UNLOCKED();
1539 			if (dolock)
1540 				splx(s);
1541 			else
1542 				dolock = 1;
1543 
1544 		childresumed:
1545 			/*
1546 			 * If we are no longer being traced, or the parent
1547 			 * didn't give us a signal, look for more signals.
1548 			 */
1549 			if ((p->p_flag & P_TRACED) == 0 || p->p_xstat == 0)
1550 				continue;
1551 
1552 			/*
1553 			 * If the new signal is being masked, look for other
1554 			 * signals.
1555 			 */
1556 			signum = p->p_xstat;
1557 			p->p_xstat = 0;
1558 			/*
1559 			 * `p->p_sigctx.ps_siglist |= mask' is done
1560 			 * in setrunnable().
1561 			 */
1562 			if (sigismember(&p->p_sigctx.ps_sigmask, signum))
1563 				continue;
1564 							/* take the signal! */
1565 			sigdelset(&p->p_sigctx.ps_siglist, signum);
1566 		}
1567 
1568 		prop = sigprop[signum];
1569 
1570 		/*
1571 		 * Decide whether the signal should be returned.
1572 		 * Return the signal's number, or fall through
1573 		 * to clear it from the pending mask.
1574 		 */
1575 		switch ((long)SIGACTION(p, signum).sa_handler) {
1576 
1577 		case (long)SIG_DFL:
1578 			/*
1579 			 * Don't take default actions on system processes.
1580 			 */
1581 			if (p->p_pid <= 1) {
1582 #ifdef DIAGNOSTIC
1583 				/*
1584 				 * Are you sure you want to ignore SIGSEGV
1585 				 * in init? XXX
1586 				 */
1587 				printf("Process (pid %d) got signal %d\n",
1588 				    p->p_pid, signum);
1589 #endif
1590 				break;		/* == ignore */
1591 			}
1592 			/*
1593 			 * If there is a pending stop signal to process
1594 			 * with default action, stop here,
1595 			 * then clear the signal.  However,
1596 			 * if process is member of an orphaned
1597 			 * process group, ignore tty stop signals.
1598 			 */
1599 			if (prop & SA_STOP) {
1600 				if (p->p_flag & P_TRACED ||
1601 		    		    (p->p_pgrp->pg_jobc == 0 &&
1602 				    prop & SA_TTYSTOP))
1603 					break;	/* == ignore */
1604 				p->p_xstat = signum;
1605 				if ((p->p_pptr->p_flag & P_NOCLDSTOP) == 0)
1606 					child_psignal(p, dolock);
1607 				if (dolock)
1608 					SCHED_LOCK(s);
1609 				proc_stop(p, 1);
1610 			sigswitch:
1611 				mi_switch(l, NULL);
1612 				SCHED_ASSERT_UNLOCKED();
1613 				if (dolock)
1614 					splx(s);
1615 				else
1616 					dolock = 1;
1617 				break;
1618 			} else if (prop & SA_IGNORE) {
1619 				/*
1620 				 * Except for SIGCONT, shouldn't get here.
1621 				 * Default action is to ignore; drop it.
1622 				 */
1623 				break;		/* == ignore */
1624 			} else
1625 				goto keep;
1626 			/*NOTREACHED*/
1627 
1628 		case (long)SIG_IGN:
1629 			/*
1630 			 * Masking above should prevent us ever trying
1631 			 * to take action on an ignored signal other
1632 			 * than SIGCONT, unless process is traced.
1633 			 */
1634 #ifdef DEBUG_ISSIGNAL
1635 			if ((prop & SA_CONT) == 0 &&
1636 			    (p->p_flag & P_TRACED) == 0)
1637 				printf("issignal\n");
1638 #endif
1639 			break;		/* == ignore */
1640 
1641 		default:
1642 			/*
1643 			 * This signal has an action, let
1644 			 * postsig() process it.
1645 			 */
1646 			goto keep;
1647 		}
1648 	}
1649 	/* NOTREACHED */
1650 
1651  keep:
1652 						/* leave the signal for later */
1653 	sigaddset(&p->p_sigctx.ps_siglist, signum);
1654 	CHECKSIGS(p);
1655 	if (locked && dolock)
1656 		SCHED_LOCK(s);
1657 	return (signum);
1658 }
1659 
1660 /*
1661  * Put the argument process into the stopped state and notify the parent
1662  * via wakeup.  Signals are handled elsewhere.  The process must not be
1663  * on the run queue.
1664  */
1665 void
1666 proc_stop(struct proc *p, int dowakeup)
1667 {
1668 	struct lwp *l;
1669 	struct proc *parent;
1670 	struct sadata_vp *vp;
1671 
1672 	SCHED_ASSERT_LOCKED();
1673 
1674 	/* XXX lock process LWP state */
1675 	p->p_flag &= ~P_WAITED;
1676 	p->p_stat = SSTOP;
1677 	parent = p->p_pptr;
1678 	parent->p_nstopchild++;
1679 
1680 	if (p->p_flag & P_SA) {
1681 		/*
1682 		 * Only (try to) put the LWP on the VP in stopped
1683 		 * state.
1684 		 * All other LWPs will suspend in sa_setwoken()
1685 		 * because the VP-LWP in stopped state cannot be
1686 		 * repossessed.
1687 		 */
1688 		SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) {
1689 			l = vp->savp_lwp;
1690 			if (l->l_stat == LSONPROC && l->l_cpu == curcpu()) {
1691 				l->l_stat = LSSTOP;
1692 				p->p_nrlwps--;
1693 			} else if (l->l_stat == LSRUN) {
1694 				/* Remove LWP from the run queue */
1695 				remrunqueue(l);
1696 				l->l_stat = LSSTOP;
1697 				p->p_nrlwps--;
1698 			} else if (l->l_stat == LSSLEEP &&
1699 			    l->l_flag & L_SA_IDLE) {
1700 				l->l_flag &= ~L_SA_IDLE;
1701 				l->l_stat = LSSTOP;
1702 			}
1703 		}
1704 		goto out;
1705 	}
1706 
1707 	/*
1708 	 * Put as many LWP's as possible in stopped state.
1709 	 * Sleeping ones will notice the stopped state as they try to
1710 	 * return to userspace.
1711 	 */
1712 
1713 	LIST_FOREACH(l, &p->p_lwps, l_sibling) {
1714 		if (l->l_stat == LSONPROC) {
1715 			/* XXX SMP this assumes that a LWP that is LSONPROC
1716 			 * is curlwp and hence is about to be mi_switched
1717 			 * away; the only callers of proc_stop() are:
1718 			 * - psignal
1719 			 * - issignal()
1720 			 * For the former, proc_stop() is only called when
1721 			 * no processes are running, so we don't worry.
1722 			 * For the latter, proc_stop() is called right
1723 			 * before mi_switch().
1724 			 */
1725 			l->l_stat = LSSTOP;
1726 			p->p_nrlwps--;
1727 		} else if (l->l_stat == LSRUN) {
1728 			/* Remove LWP from the run queue */
1729 			remrunqueue(l);
1730 			l->l_stat = LSSTOP;
1731 			p->p_nrlwps--;
1732 		} else if ((l->l_stat == LSSLEEP) ||
1733 		    (l->l_stat == LSSUSPENDED) ||
1734 		    (l->l_stat == LSZOMB) ||
1735 		    (l->l_stat == LSDEAD)) {
1736 			/*
1737 			 * Don't do anything; let sleeping LWPs
1738 			 * discover the stopped state of the process
1739 			 * on their way out of the kernel; otherwise,
1740 			 * things like NFS threads that sleep with
1741 			 * locks will block the rest of the system
1742 			 * from getting any work done.
1743 			 *
1744 			 * Suspended/dead/zombie LWPs aren't going
1745 			 * anywhere, so we don't need to touch them.
1746 			 */
1747 		}
1748 #ifdef DIAGNOSTIC
1749 		else {
1750 			panic("proc_stop: process %d lwp %d "
1751 			      "in unstoppable state %d.\n",
1752 			    p->p_pid, l->l_lid, l->l_stat);
1753 		}
1754 #endif
1755 	}
1756 
1757  out:
1758 	/* XXX unlock process LWP state */
1759 
1760 	if (dowakeup)
1761 		sched_wakeup((caddr_t)p->p_pptr);
1762 }
1763 
1764 /*
1765  * Given a process in state SSTOP, set the state back to SACTIVE and
1766  * move LSSTOP'd LWPs to LSSLEEP or make them runnable.
1767  *
1768  * If no LWPs ended up runnable (and therefore able to take a signal),
1769  * return a LWP that is sleeping interruptably. The caller can wake
1770  * that LWP up to take a signal.
1771  */
1772 struct lwp *
1773 proc_unstop(struct proc *p)
1774 {
1775 	struct lwp *l, *lr = NULL;
1776 	struct sadata_vp *vp;
1777 	int cantake = 0;
1778 
1779 	SCHED_ASSERT_LOCKED();
1780 
1781 	/*
1782 	 * Our caller wants to be informed if there are only sleeping
1783 	 * and interruptable LWPs left after we have run so that it
1784 	 * can invoke setrunnable() if required - return one of the
1785 	 * interruptable LWPs if this is the case.
1786 	 */
1787 
1788 	if (!(p->p_flag & P_WAITED))
1789 		p->p_pptr->p_nstopchild--;
1790 	p->p_stat = SACTIVE;
1791 	LIST_FOREACH(l, &p->p_lwps, l_sibling) {
1792 		if (l->l_stat == LSRUN) {
1793 			lr = NULL;
1794 			cantake = 1;
1795 		}
1796 		if (l->l_stat != LSSTOP)
1797 			continue;
1798 
1799 		if (l->l_wchan != NULL) {
1800 			l->l_stat = LSSLEEP;
1801 			if ((cantake == 0) && (l->l_flag & L_SINTR)) {
1802 				lr = l;
1803 				cantake = 1;
1804 			}
1805 		} else {
1806 			setrunnable(l);
1807 			lr = NULL;
1808 			cantake = 1;
1809 		}
1810 	}
1811 	if (p->p_flag & P_SA) {
1812 		/* Only consider returning the LWP on the VP. */
1813 		SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) {
1814 			lr = vp->savp_lwp;
1815 			if (lr->l_stat == LSSLEEP) {
1816 				if (lr->l_flag & L_SA_YIELD) {
1817 					setrunnable(lr);
1818 					break;
1819 				} else if (lr->l_flag & L_SINTR)
1820 					return lr;
1821 			}
1822 		}
1823 		return NULL;
1824 	}
1825 	return lr;
1826 }
1827 
1828 /*
1829  * Take the action for the specified signal
1830  * from the current set of pending signals.
1831  */
1832 void
1833 postsig(int signum)
1834 {
1835 	struct lwp *l;
1836 	struct proc	*p;
1837 	struct sigacts	*ps;
1838 	sig_t		action;
1839 	sigset_t	*returnmask;
1840 
1841 	l = curlwp;
1842 	p = l->l_proc;
1843 	ps = p->p_sigacts;
1844 #ifdef DIAGNOSTIC
1845 	if (signum == 0)
1846 		panic("postsig");
1847 #endif
1848 
1849 	KERNEL_PROC_LOCK(l);
1850 
1851 #ifdef MULTIPROCESSOR
1852 	/*
1853 	 * On MP, issignal() can return the same signal to multiple
1854 	 * LWPs.  The LWPs will block above waiting for the kernel
1855 	 * lock and the first LWP which gets through will then remove
1856 	 * the signal from ps_siglist.  All other LWPs exit here.
1857 	 */
1858 	if (!sigismember(&p->p_sigctx.ps_siglist, signum)) {
1859 		KERNEL_PROC_UNLOCK(l);
1860 		return;
1861 	}
1862 #endif
1863 	sigdelset(&p->p_sigctx.ps_siglist, signum);
1864 	action = SIGACTION_PS(ps, signum).sa_handler;
1865 	if (action == SIG_DFL) {
1866 #ifdef KTRACE
1867 		if (KTRPOINT(p, KTR_PSIG))
1868 			ktrpsig(l, signum, action,
1869 			    p->p_sigctx.ps_flags & SAS_OLDMASK ?
1870 			    &p->p_sigctx.ps_oldmask : &p->p_sigctx.ps_sigmask,
1871 			    NULL);
1872 #endif
1873 		/*
1874 		 * Default action, where the default is to kill
1875 		 * the process.  (Other cases were ignored above.)
1876 		 */
1877 		sigexit(l, signum);
1878 		/* NOTREACHED */
1879 	} else {
1880 		ksiginfo_t *ksi;
1881 		/*
1882 		 * If we get here, the signal must be caught.
1883 		 */
1884 #ifdef DIAGNOSTIC
1885 		if (action == SIG_IGN ||
1886 		    sigismember(&p->p_sigctx.ps_sigmask, signum))
1887 			panic("postsig action");
1888 #endif
1889 		/*
1890 		 * Set the new mask value and also defer further
1891 		 * occurrences of this signal.
1892 		 *
1893 		 * Special case: user has done a sigpause.  Here the
1894 		 * current mask is not of interest, but rather the
1895 		 * mask from before the sigpause is what we want
1896 		 * restored after the signal processing is completed.
1897 		 */
1898 		if (p->p_sigctx.ps_flags & SAS_OLDMASK) {
1899 			returnmask = &p->p_sigctx.ps_oldmask;
1900 			p->p_sigctx.ps_flags &= ~SAS_OLDMASK;
1901 		} else
1902 			returnmask = &p->p_sigctx.ps_sigmask;
1903 		p->p_stats->p_ru.ru_nsignals++;
1904 		ksi = ksiginfo_get(p, signum);
1905 #ifdef KTRACE
1906 		if (KTRPOINT(p, KTR_PSIG))
1907 			ktrpsig(l, signum, action,
1908 			    p->p_sigctx.ps_flags & SAS_OLDMASK ?
1909 			    &p->p_sigctx.ps_oldmask : &p->p_sigctx.ps_sigmask,
1910 			    ksi);
1911 #endif
1912 		if (ksi == NULL) {
1913 			ksiginfo_t ksi1;
1914 			/*
1915 			 * we did not save any siginfo for this, either
1916 			 * because the signal was not caught, or because the
1917 			 * user did not request SA_SIGINFO
1918 			 */
1919 			KSI_INIT_EMPTY(&ksi1);
1920 			ksi1.ksi_signo = signum;
1921 			kpsendsig(l, &ksi1, returnmask);
1922 		} else {
1923 			kpsendsig(l, ksi, returnmask);
1924 			pool_put(&ksiginfo_pool, ksi);
1925 		}
1926 		p->p_sigctx.ps_lwp = 0;
1927 		p->p_sigctx.ps_code = 0;
1928 		p->p_sigctx.ps_signo = 0;
1929 		(void) splsched();	/* XXXSMP */
1930 		sigplusset(&SIGACTION_PS(ps, signum).sa_mask,
1931 		    &p->p_sigctx.ps_sigmask);
1932 		if (SIGACTION_PS(ps, signum).sa_flags & SA_RESETHAND) {
1933 			sigdelset(&p->p_sigctx.ps_sigcatch, signum);
1934 			if (signum != SIGCONT && sigprop[signum] & SA_IGNORE)
1935 				sigaddset(&p->p_sigctx.ps_sigignore, signum);
1936 			SIGACTION_PS(ps, signum).sa_handler = SIG_DFL;
1937 		}
1938 		(void) spl0();		/* XXXSMP */
1939 	}
1940 
1941 	KERNEL_PROC_UNLOCK(l);
1942 }
1943 
1944 /*
1945  * Kill the current process for stated reason.
1946  */
1947 void
1948 killproc(struct proc *p, const char *why)
1949 {
1950 	log(LOG_ERR, "pid %d was killed: %s\n", p->p_pid, why);
1951 	uprintf("sorry, pid %d was killed: %s\n", p->p_pid, why);
1952 	psignal(p, SIGKILL);
1953 }
1954 
1955 /*
1956  * Force the current process to exit with the specified signal, dumping core
1957  * if appropriate.  We bypass the normal tests for masked and caught signals,
1958  * allowing unrecoverable failures to terminate the process without changing
1959  * signal state.  Mark the accounting record with the signal termination.
1960  * If dumping core, save the signal number for the debugger.  Calls exit and
1961  * does not return.
1962  */
1963 
1964 #if defined(DEBUG)
1965 int	kern_logsigexit = 1;	/* not static to make public for sysctl */
1966 #else
1967 int	kern_logsigexit = 0;	/* not static to make public for sysctl */
1968 #endif
1969 
1970 static	const char logcoredump[] =
1971 	"pid %d (%s), uid %d: exited on signal %d (core dumped)\n";
1972 static	const char lognocoredump[] =
1973 	"pid %d (%s), uid %d: exited on signal %d (core not dumped, err = %d)\n";
1974 
1975 /* Wrapper function for use in p_userret */
1976 static void
1977 lwp_coredump_hook(struct lwp *l, void *arg)
1978 {
1979 	int s;
1980 
1981 	/*
1982 	 * Suspend ourselves, so that the kernel stack and therefore
1983 	 * the userland registers saved in the trapframe are around
1984 	 * for coredump() to write them out.
1985 	 */
1986 	KERNEL_PROC_LOCK(l);
1987 	l->l_flag &= ~L_DETACHED;
1988 	SCHED_LOCK(s);
1989 	l->l_stat = LSSUSPENDED;
1990 	l->l_proc->p_nrlwps--;
1991 	/* XXX NJWLWP check if this makes sense here: */
1992 	l->l_proc->p_stats->p_ru.ru_nvcsw++;
1993 	mi_switch(l, NULL);
1994 	SCHED_ASSERT_UNLOCKED();
1995 	splx(s);
1996 
1997 	lwp_exit(l);
1998 }
1999 
2000 void
2001 sigexit(struct lwp *l, int signum)
2002 {
2003 	struct proc	*p;
2004 #if 0
2005 	struct lwp	*l2;
2006 #endif
2007 	int		exitsig;
2008 #ifdef COREDUMP
2009 	int		error;
2010 #endif
2011 
2012 	p = l->l_proc;
2013 
2014 	/*
2015 	 * Don't permit coredump() or exit1() multiple times
2016 	 * in the same process.
2017 	 */
2018 	if (p->p_flag & P_WEXIT) {
2019 		KERNEL_PROC_UNLOCK(l);
2020 		(*p->p_userret)(l, p->p_userret_arg);
2021 	}
2022 	p->p_flag |= P_WEXIT;
2023 	/* We don't want to switch away from exiting. */
2024 	/* XXX multiprocessor: stop LWPs on other processors. */
2025 #if 0
2026 	if (p->p_flag & P_SA) {
2027 		LIST_FOREACH(l2, &p->p_lwps, l_sibling)
2028 		    l2->l_flag &= ~L_SA;
2029 		p->p_flag &= ~P_SA;
2030 	}
2031 #endif
2032 
2033 	/* Make other LWPs stick around long enough to be dumped */
2034 	p->p_userret = lwp_coredump_hook;
2035 	p->p_userret_arg = NULL;
2036 
2037 	exitsig = signum;
2038 	p->p_acflag |= AXSIG;
2039 	if (sigprop[signum] & SA_CORE) {
2040 		p->p_sigctx.ps_signo = signum;
2041 #ifdef COREDUMP
2042 		if ((error = coredump(l, NULL)) == 0)
2043 			exitsig |= WCOREFLAG;
2044 #endif
2045 
2046 		if (kern_logsigexit) {
2047 			/* XXX What if we ever have really large UIDs? */
2048 			int uid = l->l_cred ?
2049 			    (int)kauth_cred_geteuid(l->l_cred) : -1;
2050 
2051 #ifdef COREDUMP
2052 			if (error)
2053 				log(LOG_INFO, lognocoredump, p->p_pid,
2054 				    p->p_comm, uid, signum, error);
2055 			else
2056 #endif
2057 				log(LOG_INFO, logcoredump, p->p_pid,
2058 				    p->p_comm, uid, signum);
2059 		}
2060 
2061 	}
2062 
2063 	exit1(l, W_EXITCODE(0, exitsig));
2064 	/* NOTREACHED */
2065 }
2066 
2067 #ifdef COREDUMP
2068 struct coredump_iostate {
2069 	struct lwp *io_lwp;
2070 	struct vnode *io_vp;
2071 	kauth_cred_t io_cred;
2072 	off_t io_offset;
2073 };
2074 
2075 int
2076 coredump_write(void *cookie, enum uio_seg segflg, const void *data, size_t len)
2077 {
2078 	struct coredump_iostate *io = cookie;
2079 	int error;
2080 
2081 	error = vn_rdwr(UIO_WRITE, io->io_vp, __UNCONST(data), len,
2082 	    io->io_offset, segflg,
2083 	    IO_NODELOCKED|IO_UNIT, io->io_cred, NULL,
2084 	    segflg == UIO_USERSPACE ? io->io_lwp : NULL);
2085 	if (error) {
2086 		printf("pid %d (%s): %s write of %zu@%p at %lld failed: %d\n",
2087 		    io->io_lwp->l_proc->p_pid, io->io_lwp->l_proc->p_comm,
2088 		    segflg == UIO_USERSPACE ? "user" : "system",
2089 		    len, data, (long long) io->io_offset, error);
2090 		return (error);
2091 	}
2092 
2093 	io->io_offset += len;
2094 	return (0);
2095 }
2096 
2097 /*
2098  * Dump core, into a file named "progname.core" or "core" (depending on the
2099  * value of shortcorename), unless the process was setuid/setgid.
2100  */
2101 int
2102 coredump(struct lwp *l, const char *pattern)
2103 {
2104 	struct vnode		*vp;
2105 	struct proc		*p;
2106 	struct vmspace		*vm;
2107 	kauth_cred_t		cred;
2108 	struct nameidata	nd;
2109 	struct vattr		vattr;
2110 	struct mount		*mp;
2111 	struct coredump_iostate	io;
2112 	int			error, error1;
2113 	char			*name = NULL;
2114 
2115 	p = l->l_proc;
2116 	vm = p->p_vmspace;
2117 	cred = l->l_cred;
2118 
2119 	/*
2120 	 * Make sure the process has not set-id, to prevent data leaks,
2121 	 * unless it was specifically requested to allow set-id coredumps.
2122 	 */
2123 	if ((p->p_flag & P_SUGID) && !security_setidcore_dump)
2124 		return EPERM;
2125 
2126 	/*
2127 	 * Refuse to core if the data + stack + user size is larger than
2128 	 * the core dump limit.  XXX THIS IS WRONG, because of mapped
2129 	 * data.
2130 	 */
2131 	if (USPACE + ctob(vm->vm_dsize + vm->vm_ssize) >=
2132 	    p->p_rlimit[RLIMIT_CORE].rlim_cur)
2133 		return EFBIG;		/* better error code? */
2134 
2135 restart:
2136 	/*
2137 	 * The core dump will go in the current working directory.  Make
2138 	 * sure that the directory is still there and that the mount flags
2139 	 * allow us to write core dumps there.
2140 	 */
2141 	vp = p->p_cwdi->cwdi_cdir;
2142 	if (vp->v_mount == NULL ||
2143 	    (vp->v_mount->mnt_flag & MNT_NOCOREDUMP) != 0) {
2144 		error = EPERM;
2145 		goto done;
2146 	}
2147 
2148 	if ((p->p_flag & P_SUGID) && security_setidcore_dump)
2149 		pattern = security_setidcore_path;
2150 
2151 	if (pattern == NULL)
2152 		pattern = p->p_limit->pl_corename;
2153 	if (name == NULL) {
2154 		name = PNBUF_GET();
2155 	}
2156 	if ((error = build_corename(p, name, pattern, MAXPATHLEN)) != 0)
2157 		goto done;
2158 	NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, l);
2159 	if ((error = vn_open(&nd, O_CREAT | O_NOFOLLOW | FWRITE,
2160 	    S_IRUSR | S_IWUSR)) != 0)
2161 		goto done;
2162 	vp = nd.ni_vp;
2163 
2164 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
2165 		VOP_UNLOCK(vp, 0);
2166 		if ((error = vn_close(vp, FWRITE, cred, l)) != 0)
2167 			goto done;
2168 		if ((error = vn_start_write(NULL, &mp,
2169 		    V_WAIT | V_SLEEPONLY | V_PCATCH)) != 0)
2170 			goto done;
2171 		goto restart;
2172 	}
2173 
2174 	/* Don't dump to non-regular files or files with links. */
2175 	if (vp->v_type != VREG ||
2176 	    VOP_GETATTR(vp, &vattr, cred, l) || vattr.va_nlink != 1) {
2177 		error = EINVAL;
2178 		goto out;
2179 	}
2180 	VATTR_NULL(&vattr);
2181 	vattr.va_size = 0;
2182 
2183 	if ((p->p_flag & P_SUGID) && security_setidcore_dump) {
2184 		vattr.va_uid = security_setidcore_owner;
2185 		vattr.va_gid = security_setidcore_group;
2186 		vattr.va_mode = security_setidcore_mode;
2187 	}
2188 
2189 	VOP_LEASE(vp, l, cred, LEASE_WRITE);
2190 	VOP_SETATTR(vp, &vattr, cred, l);
2191 	p->p_acflag |= ACORE;
2192 
2193 	io.io_lwp = l;
2194 	io.io_vp = vp;
2195 	io.io_cred = cred;
2196 	io.io_offset = 0;
2197 
2198 	/* Now dump the actual core file. */
2199 	error = (*p->p_execsw->es_coredump)(l, &io);
2200  out:
2201 	VOP_UNLOCK(vp, 0);
2202 	vn_finished_write(mp, 0);
2203 	error1 = vn_close(vp, FWRITE, cred, l);
2204 	if (error == 0)
2205 		error = error1;
2206 done:
2207 	if (name != NULL)
2208 		PNBUF_PUT(name);
2209 	return error;
2210 }
2211 #endif /* COREDUMP */
2212 
2213 /*
2214  * Nonexistent system call-- signal process (may want to handle it).
2215  * Flag error in case process won't see signal immediately (blocked or ignored).
2216  */
2217 #ifndef PTRACE
2218 __weak_alias(sys_ptrace, sys_nosys);
2219 #endif
2220 
2221 /* ARGSUSED */
2222 int
2223 sys_nosys(struct lwp *l, void *v, register_t *retval)
2224 {
2225 	struct proc 	*p;
2226 
2227 	p = l->l_proc;
2228 	psignal(p, SIGSYS);
2229 	return (ENOSYS);
2230 }
2231 
2232 #ifdef COREDUMP
2233 static int
2234 build_corename(struct proc *p, char *dst, const char *src, size_t len)
2235 {
2236 	const char	*s;
2237 	char		*d, *end;
2238 	int		i;
2239 
2240 	for (s = src, d = dst, end = d + len; *s != '\0'; s++) {
2241 		if (*s == '%') {
2242 			switch (*(s + 1)) {
2243 			case 'n':
2244 				i = snprintf(d, end - d, "%s", p->p_comm);
2245 				break;
2246 			case 'p':
2247 				i = snprintf(d, end - d, "%d", p->p_pid);
2248 				break;
2249 			case 'u':
2250 				i = snprintf(d, end - d, "%.*s",
2251 				    (int)sizeof p->p_pgrp->pg_session->s_login,
2252 				    p->p_pgrp->pg_session->s_login);
2253 				break;
2254 			case 't':
2255 				i = snprintf(d, end - d, "%ld",
2256 				    p->p_stats->p_start.tv_sec);
2257 				break;
2258 			default:
2259 				goto copy;
2260 			}
2261 			d += i;
2262 			s++;
2263 		} else {
2264  copy:			*d = *s;
2265 			d++;
2266 		}
2267 		if (d >= end)
2268 			return (ENAMETOOLONG);
2269 	}
2270 	*d = '\0';
2271 	return 0;
2272 }
2273 #endif /* COREDUMP */
2274 
2275 void
2276 getucontext(struct lwp *l, ucontext_t *ucp)
2277 {
2278 	struct proc	*p;
2279 
2280 	p = l->l_proc;
2281 
2282 	ucp->uc_flags = 0;
2283 	ucp->uc_link = l->l_ctxlink;
2284 
2285 	(void)sigprocmask1(p, 0, NULL, &ucp->uc_sigmask);
2286 	ucp->uc_flags |= _UC_SIGMASK;
2287 
2288 	/*
2289 	 * The (unsupplied) definition of the `current execution stack'
2290 	 * in the System V Interface Definition appears to allow returning
2291 	 * the main context stack.
2292 	 */
2293 	if ((p->p_sigctx.ps_sigstk.ss_flags & SS_ONSTACK) == 0) {
2294 		ucp->uc_stack.ss_sp = (void *)USRSTACK;
2295 		ucp->uc_stack.ss_size = ctob(p->p_vmspace->vm_ssize);
2296 		ucp->uc_stack.ss_flags = 0;	/* XXX, def. is Very Fishy */
2297 	} else {
2298 		/* Simply copy alternate signal execution stack. */
2299 		ucp->uc_stack = p->p_sigctx.ps_sigstk;
2300 	}
2301 	ucp->uc_flags |= _UC_STACK;
2302 
2303 	cpu_getmcontext(l, &ucp->uc_mcontext, &ucp->uc_flags);
2304 }
2305 
2306 /* ARGSUSED */
2307 int
2308 sys_getcontext(struct lwp *l, void *v, register_t *retval)
2309 {
2310 	struct sys_getcontext_args /* {
2311 		syscallarg(struct __ucontext *) ucp;
2312 	} */ *uap = v;
2313 	ucontext_t uc;
2314 
2315 	getucontext(l, &uc);
2316 
2317 	return (copyout(&uc, SCARG(uap, ucp), sizeof (*SCARG(uap, ucp))));
2318 }
2319 
2320 int
2321 setucontext(struct lwp *l, const ucontext_t *ucp)
2322 {
2323 	struct proc	*p;
2324 	int		error;
2325 
2326 	p = l->l_proc;
2327 	if ((error = cpu_setmcontext(l, &ucp->uc_mcontext, ucp->uc_flags)) != 0)
2328 		return (error);
2329 	l->l_ctxlink = ucp->uc_link;
2330 
2331 	if ((ucp->uc_flags & _UC_SIGMASK) != 0)
2332 		sigprocmask1(p, SIG_SETMASK, &ucp->uc_sigmask, NULL);
2333 
2334 	/*
2335 	 * If there was stack information, update whether or not we are
2336 	 * still running on an alternate signal stack.
2337 	 */
2338 	if ((ucp->uc_flags & _UC_STACK) != 0) {
2339 		if (ucp->uc_stack.ss_flags & SS_ONSTACK)
2340 			p->p_sigctx.ps_sigstk.ss_flags |= SS_ONSTACK;
2341 		else
2342 			p->p_sigctx.ps_sigstk.ss_flags &= ~SS_ONSTACK;
2343 	}
2344 
2345 	return 0;
2346 }
2347 
2348 /* ARGSUSED */
2349 int
2350 sys_setcontext(struct lwp *l, void *v, register_t *retval)
2351 {
2352 	struct sys_setcontext_args /* {
2353 		syscallarg(const ucontext_t *) ucp;
2354 	} */ *uap = v;
2355 	ucontext_t uc;
2356 	int error;
2357 
2358 	if (SCARG(uap, ucp) == NULL)	/* i.e. end of uc_link chain */
2359 		exit1(l, W_EXITCODE(0, 0));
2360 	else if ((error = copyin(SCARG(uap, ucp), &uc, sizeof (uc))) != 0 ||
2361 	    (error = setucontext(l, &uc)) != 0)
2362 		return (error);
2363 
2364 	return (EJUSTRETURN);
2365 }
2366 
2367 /*
2368  * sigtimedwait(2) system call, used also for implementation
2369  * of sigwaitinfo() and sigwait().
2370  *
2371  * This only handles single LWP in signal wait. libpthread provides
2372  * it's own sigtimedwait() wrapper to DTRT WRT individual threads.
2373  */
2374 int
2375 sys___sigtimedwait(struct lwp *l, void *v, register_t *retval)
2376 {
2377 	return __sigtimedwait1(l, v, retval, copyout, copyin, copyout);
2378 }
2379 
2380 int
2381 __sigtimedwait1(struct lwp *l, void *v, register_t *retval,
2382     copyout_t put_info, copyin_t fetch_timeout, copyout_t put_timeout)
2383 {
2384 	struct sys___sigtimedwait_args /* {
2385 		syscallarg(const sigset_t *) set;
2386 		syscallarg(siginfo_t *) info;
2387 		syscallarg(struct timespec *) timeout;
2388 	} */ *uap = v;
2389 	sigset_t *waitset, twaitset;
2390 	struct proc *p = l->l_proc;
2391 	int error, signum;
2392 	int timo = 0;
2393 	struct timespec ts, tsstart;
2394 	ksiginfo_t *ksi;
2395 
2396 	memset(&tsstart, 0, sizeof tsstart);	 /* XXX gcc */
2397 
2398 	MALLOC(waitset, sigset_t *, sizeof(sigset_t), M_TEMP, M_WAITOK);
2399 
2400 	if ((error = copyin(SCARG(uap, set), waitset, sizeof(sigset_t)))) {
2401 		FREE(waitset, M_TEMP);
2402 		return (error);
2403 	}
2404 
2405 	/*
2406 	 * Silently ignore SA_CANTMASK signals. psignal1() would
2407 	 * ignore SA_CANTMASK signals in waitset, we do this
2408 	 * only for the below siglist check.
2409 	 */
2410 	sigminusset(&sigcantmask, waitset);
2411 
2412 	/*
2413 	 * First scan siglist and check if there is signal from
2414 	 * our waitset already pending.
2415 	 */
2416 	twaitset = *waitset;
2417 	__sigandset(&p->p_sigctx.ps_siglist, &twaitset);
2418 	if ((signum = firstsig(&twaitset))) {
2419 		/* found pending signal */
2420 		sigdelset(&p->p_sigctx.ps_siglist, signum);
2421 		ksi = ksiginfo_get(p, signum);
2422 		if (!ksi) {
2423 			/* No queued siginfo, manufacture one */
2424 			ksi = pool_get(&ksiginfo_pool, PR_WAITOK);
2425 			KSI_INIT(ksi);
2426 			ksi->ksi_info._signo = signum;
2427 			ksi->ksi_info._code = SI_USER;
2428 		}
2429 
2430 		goto sig;
2431 	}
2432 
2433 	/*
2434 	 * Calculate timeout, if it was specified.
2435 	 */
2436 	if (SCARG(uap, timeout)) {
2437 		uint64_t ms;
2438 
2439 		if ((error = (*fetch_timeout)(SCARG(uap, timeout), &ts, sizeof(ts))))
2440 			return (error);
2441 
2442 		ms = (ts.tv_sec * 1000) + (ts.tv_nsec / 1000000);
2443 		timo = mstohz(ms);
2444 		if (timo == 0 && ts.tv_sec == 0 && ts.tv_nsec > 0)
2445 			timo = 1;
2446 		if (timo <= 0)
2447 			return (EAGAIN);
2448 
2449 		/*
2450 		 * Remember current uptime, it would be used in
2451 		 * ECANCELED/ERESTART case.
2452 		 */
2453 		getnanouptime(&tsstart);
2454 	}
2455 
2456 	/*
2457 	 * Setup ps_sigwait list. Pass pointer to malloced memory
2458 	 * here; it's not possible to pass pointer to a structure
2459 	 * on current process's stack, the current process might
2460 	 * be swapped out at the time the signal would get delivered.
2461 	 */
2462 	ksi = pool_get(&ksiginfo_pool, PR_WAITOK);
2463 	p->p_sigctx.ps_sigwaited = ksi;
2464 	p->p_sigctx.ps_sigwait = waitset;
2465 
2466 	/*
2467 	 * Wait for signal to arrive. We can either be woken up or
2468 	 * time out.
2469 	 */
2470 	error = tsleep(&p->p_sigctx.ps_sigwait, PPAUSE|PCATCH, "sigwait", timo);
2471 
2472 	/*
2473 	 * Need to find out if we woke as a result of lwp_wakeup()
2474 	 * or a signal outside our wait set.
2475 	 */
2476 	if (error == EINTR && p->p_sigctx.ps_sigwaited
2477 	    && !firstsig(&p->p_sigctx.ps_siglist)) {
2478 		/* wakeup via _lwp_wakeup() */
2479 		error = ECANCELED;
2480 	} else if (!error && p->p_sigctx.ps_sigwaited) {
2481 		/* spurious wakeup - arrange for syscall restart */
2482 		error = ERESTART;
2483 		goto fail;
2484 	}
2485 
2486 	/*
2487 	 * On error, clear sigwait indication. psignal1() clears it
2488 	 * in !error case.
2489 	 */
2490 	if (error) {
2491 		p->p_sigctx.ps_sigwaited = NULL;
2492 
2493 		/*
2494 		 * If the sleep was interrupted (either by signal or wakeup),
2495 		 * update the timeout and copyout new value back.
2496 		 * It would be used when the syscall would be restarted
2497 		 * or called again.
2498 		 */
2499 		if (timo && (error == ERESTART || error == ECANCELED)) {
2500 			struct timespec tsnow;
2501 			int err;
2502 
2503 /* XXX double check the following change */
2504 			getnanouptime(&tsnow);
2505 
2506 			/* compute how much time has passed since start */
2507 			timespecsub(&tsnow, &tsstart, &tsnow);
2508 			/* substract passed time from timeout */
2509 			timespecsub(&ts, &tsnow, &ts);
2510 
2511 			if (ts.tv_sec < 0) {
2512 				error = EAGAIN;
2513 				goto fail;
2514 			}
2515 /* XXX double check the previous change */
2516 
2517 			/* copy updated timeout to userland */
2518 			if ((err = (*put_timeout)(&ts, SCARG(uap, timeout),
2519 			    sizeof(ts)))) {
2520 				error = err;
2521 				goto fail;
2522 			}
2523 		}
2524 
2525 		goto fail;
2526 	}
2527 
2528 	/*
2529 	 * If a signal from the wait set arrived, copy it to userland.
2530 	 * Copy only the used part of siginfo, the padding part is
2531 	 * left unchanged (userland is not supposed to touch it anyway).
2532 	 */
2533  sig:
2534 	return (*put_info)(&ksi->ksi_info, SCARG(uap, info), sizeof(ksi->ksi_info));
2535 
2536  fail:
2537 	FREE(waitset, M_TEMP);
2538 	pool_put(&ksiginfo_pool, ksi);
2539 	p->p_sigctx.ps_sigwait = NULL;
2540 
2541 	return (error);
2542 }
2543 
2544 /*
2545  * Returns true if signal is ignored or masked for passed process.
2546  */
2547 int
2548 sigismasked(struct proc *p, int sig)
2549 {
2550 
2551 	return (sigismember(&p->p_sigctx.ps_sigignore, sig) ||
2552 	    sigismember(&p->p_sigctx.ps_sigmask, sig));
2553 }
2554 
2555 static int
2556 filt_sigattach(struct knote *kn)
2557 {
2558 	struct proc *p = curproc;
2559 
2560 	kn->kn_ptr.p_proc = p;
2561 	kn->kn_flags |= EV_CLEAR;               /* automatically set */
2562 
2563 	SLIST_INSERT_HEAD(&p->p_klist, kn, kn_selnext);
2564 
2565 	return (0);
2566 }
2567 
2568 static void
2569 filt_sigdetach(struct knote *kn)
2570 {
2571 	struct proc *p = kn->kn_ptr.p_proc;
2572 
2573 	SLIST_REMOVE(&p->p_klist, kn, knote, kn_selnext);
2574 }
2575 
2576 /*
2577  * signal knotes are shared with proc knotes, so we apply a mask to
2578  * the hint in order to differentiate them from process hints.  This
2579  * could be avoided by using a signal-specific knote list, but probably
2580  * isn't worth the trouble.
2581  */
2582 static int
2583 filt_signal(struct knote *kn, long hint)
2584 {
2585 
2586 	if (hint & NOTE_SIGNAL) {
2587 		hint &= ~NOTE_SIGNAL;
2588 
2589 		if (kn->kn_id == hint)
2590 			kn->kn_data++;
2591 	}
2592 	return (kn->kn_data != 0);
2593 }
2594 
2595 const struct filterops sig_filtops = {
2596 	0, filt_sigattach, filt_sigdetach, filt_signal
2597 };
2598