1/*	$NetBSD: kern_sig.c,v 1.409 2024/02/10 09:24:18 andvar Exp $	*/
2
3/*-
4 * Copyright (c) 2006, 2007, 2008, 2019, 2023 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Andrew Doran.
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 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32/*
33 * Copyright (c) 1982, 1986, 1989, 1991, 1993
34 *	The Regents of the University of California.  All rights reserved.
35 * (c) UNIX System Laboratories, Inc.
36 * All or some portions of this file are derived from material licensed
37 * to the University of California by American Telephone and Telegraph
38 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
39 * the permission of UNIX System Laboratories, Inc.
40 *
41 * Redistribution and use in source and binary forms, with or without
42 * modification, are permitted provided that the following conditions
43 * are met:
44 * 1. Redistributions of source code must retain the above copyright
45 *    notice, this list of conditions and the following disclaimer.
46 * 2. Redistributions in binary form must reproduce the above copyright
47 *    notice, this list of conditions and the following disclaimer in the
48 *    documentation and/or other materials provided with the distribution.
49 * 3. Neither the name of the University nor the names of its contributors
50 *    may be used to endorse or promote products derived from this software
51 *    without specific prior written permission.
52 *
53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63 * SUCH DAMAGE.
64 *
65 *	@(#)kern_sig.c	8.14 (Berkeley) 5/14/95
66 */
67
68/*
69 * Signal subsystem.
70 */
71
72#include <sys/cdefs.h>
73__KERNEL_RCSID(0, "$NetBSD: kern_sig.c,v 1.409 2024/02/10 09:24:18 andvar Exp $");
74
75#include "opt_execfmt.h"
76#include "opt_ptrace.h"
77#include "opt_dtrace.h"
78#include "opt_compat_sunos.h"
79#include "opt_compat_netbsd.h"
80#include "opt_compat_netbsd32.h"
81#include "opt_pax.h"
82
83#define	SIGPROP		/* include signal properties table */
84#include <sys/param.h>
85#include <sys/signalvar.h>
86#include <sys/proc.h>
87#include <sys/ptrace.h>
88#include <sys/systm.h>
89#include <sys/wait.h>
90#include <sys/ktrace.h>
91#include <sys/syslog.h>
92#include <sys/filedesc.h>
93#include <sys/file.h>
94#include <sys/pool.h>
95#include <sys/ucontext.h>
96#include <sys/exec.h>
97#include <sys/kauth.h>
98#include <sys/acct.h>
99#include <sys/callout.h>
100#include <sys/atomic.h>
101#include <sys/cpu.h>
102#include <sys/module.h>
103#include <sys/sdt.h>
104#include <sys/exec_elf.h>
105#include <sys/compat_stub.h>
106
107#ifdef PAX_SEGVGUARD
108#include <sys/pax.h>
109#endif /* PAX_SEGVGUARD */
110
111#include <uvm/uvm_extern.h>
112
113/* Many hard-coded assumptions that there are <= 4 x 32bit signal mask bits */
114__CTASSERT(NSIG <= 128);
115
116#define	SIGQUEUE_MAX	32
117static pool_cache_t	sigacts_cache	__read_mostly;
118static pool_cache_t	ksiginfo_cache	__read_mostly;
119static callout_t	proc_stop_ch	__cacheline_aligned;
120
121sigset_t		contsigmask	__cacheline_aligned;
122sigset_t		stopsigmask	__cacheline_aligned;
123static sigset_t		vforksigmask	__cacheline_aligned;
124sigset_t		sigcantmask	__cacheline_aligned;
125
126static void	ksiginfo_exechook(struct proc *, void *);
127static void	proc_stop(struct proc *, int);
128static void	proc_stop_done(struct proc *, int);
129static void	proc_stop_callout(void *);
130static int	sigchecktrace(void);
131static int	sigpost(struct lwp *, sig_t, int, int);
132static int	sigput(sigpend_t *, struct proc *, ksiginfo_t *);
133static int	sigunwait(struct proc *, const ksiginfo_t *);
134static void	sigswitch(int, int, bool);
135static void	sigswitch_unlock_and_switch_away(struct lwp *);
136
137static void	sigacts_poolpage_free(struct pool *, void *);
138static void	*sigacts_poolpage_alloc(struct pool *, int);
139
140/*
141 * DTrace SDT provider definitions
142 */
143SDT_PROVIDER_DECLARE(proc);
144SDT_PROBE_DEFINE3(proc, kernel, , signal__send,
145    "struct lwp *", 	/* target thread */
146    "struct proc *", 	/* target process */
147    "int");		/* signal */
148SDT_PROBE_DEFINE3(proc, kernel, , signal__discard,
149    "struct lwp *",	/* target thread */
150    "struct proc *",	/* target process */
151    "int");  		/* signal */
152SDT_PROBE_DEFINE3(proc, kernel, , signal__handle,
153    "int", 		/* signal */
154    "ksiginfo_t *", 	/* signal info */
155    "void (*)(void)");	/* handler address */
156
157
158static struct pool_allocator sigactspool_allocator = {
159	.pa_alloc = sigacts_poolpage_alloc,
160	.pa_free = sigacts_poolpage_free
161};
162
163#ifdef DEBUG
164int	kern_logsigexit = 1;
165#else
166int	kern_logsigexit = 0;
167#endif
168
169static const char logcoredump[] =
170    "pid %d (%s), uid %d: exited on signal %d (core dumped)\n";
171static const char lognocoredump[] =
172    "pid %d (%s), uid %d: exited on signal %d (core not dumped, err = %d)\n";
173
174static kauth_listener_t signal_listener;
175
176static int
177signal_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
178    void *arg0, void *arg1, void *arg2, void *arg3)
179{
180	struct proc *p;
181	int result, signum;
182
183	result = KAUTH_RESULT_DEFER;
184	p = arg0;
185	signum = (int)(unsigned long)arg1;
186
187	if (action != KAUTH_PROCESS_SIGNAL)
188		return result;
189
190	if (kauth_cred_uidmatch(cred, p->p_cred) ||
191	    (signum == SIGCONT && (curproc->p_session == p->p_session)))
192		result = KAUTH_RESULT_ALLOW;
193
194	return result;
195}
196
197static int
198sigacts_ctor(void *arg __unused, void *obj, int flags __unused)
199{
200	memset(obj, 0, sizeof(struct sigacts));
201	return 0;
202}
203
204/*
205 * signal_init:
206 *
207 *	Initialize global signal-related data structures.
208 */
209void
210signal_init(void)
211{
212
213	sigactspool_allocator.pa_pagesz = (PAGE_SIZE)*2;
214
215	sigacts_cache = pool_cache_init(sizeof(struct sigacts), 0, 0, 0,
216	    "sigacts", sizeof(struct sigacts) > PAGE_SIZE ?
217	    &sigactspool_allocator : NULL, IPL_NONE, sigacts_ctor, NULL, NULL);
218	ksiginfo_cache = pool_cache_init(sizeof(ksiginfo_t), 0, 0, 0,
219	    "ksiginfo", NULL, IPL_VM, NULL, NULL, NULL);
220
221	exechook_establish(ksiginfo_exechook, NULL);
222
223	callout_init(&proc_stop_ch, CALLOUT_MPSAFE);
224	callout_setfunc(&proc_stop_ch, proc_stop_callout, NULL);
225
226	signal_listener = kauth_listen_scope(KAUTH_SCOPE_PROCESS,
227	    signal_listener_cb, NULL);
228}
229
230/*
231 * sigacts_poolpage_alloc:
232 *
233 *	Allocate a page for the sigacts memory pool.
234 */
235static void *
236sigacts_poolpage_alloc(struct pool *pp, int flags)
237{
238
239	return (void *)uvm_km_alloc(kernel_map,
240	    PAGE_SIZE * 2, PAGE_SIZE * 2,
241	    ((flags & PR_WAITOK) ? 0 : UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK)
242	    | UVM_KMF_WIRED);
243}
244
245/*
246 * sigacts_poolpage_free:
247 *
248 *	Free a page on behalf of the sigacts memory pool.
249 */
250static void
251sigacts_poolpage_free(struct pool *pp, void *v)
252{
253
254	uvm_km_free(kernel_map, (vaddr_t)v, PAGE_SIZE * 2, UVM_KMF_WIRED);
255}
256
257/*
258 * sigactsinit:
259 *
260 *	Create an initial sigacts structure, using the same signal state
261 *	as of specified process.  If 'share' is set, share the sigacts by
262 *	holding a reference, otherwise just copy it from parent.
263 */
264struct sigacts *
265sigactsinit(struct proc *pp, int share)
266{
267	struct sigacts *ps = pp->p_sigacts, *ps2;
268
269	if (__predict_false(share)) {
270		atomic_inc_uint(&ps->sa_refcnt);
271		return ps;
272	}
273	ps2 = pool_cache_get(sigacts_cache, PR_WAITOK);
274	mutex_init(&ps2->sa_mutex, MUTEX_DEFAULT, IPL_SCHED);
275	ps2->sa_refcnt = 1;
276
277	mutex_enter(&ps->sa_mutex);
278	memcpy(ps2->sa_sigdesc, ps->sa_sigdesc, sizeof(ps2->sa_sigdesc));
279	mutex_exit(&ps->sa_mutex);
280	return ps2;
281}
282
283/*
284 * sigactsunshare:
285 *
286 *	Make this process not share its sigacts, maintaining all signal state.
287 */
288void
289sigactsunshare(struct proc *p)
290{
291	struct sigacts *ps, *oldps = p->p_sigacts;
292
293	if (__predict_true(oldps->sa_refcnt == 1))
294		return;
295
296	ps = pool_cache_get(sigacts_cache, PR_WAITOK);
297	mutex_init(&ps->sa_mutex, MUTEX_DEFAULT, IPL_SCHED);
298	memcpy(ps->sa_sigdesc, oldps->sa_sigdesc, sizeof(ps->sa_sigdesc));
299	ps->sa_refcnt = 1;
300
301	p->p_sigacts = ps;
302	sigactsfree(oldps);
303}
304
305/*
306 * sigactsfree;
307 *
308 *	Release a sigacts structure.
309 */
310void
311sigactsfree(struct sigacts *ps)
312{
313
314	membar_release();
315	if (atomic_dec_uint_nv(&ps->sa_refcnt) == 0) {
316		membar_acquire();
317		mutex_destroy(&ps->sa_mutex);
318		pool_cache_put(sigacts_cache, ps);
319	}
320}
321
322/*
323 * siginit:
324 *
325 *	Initialize signal state for process 0; set to ignore signals that
326 *	are ignored by default and disable the signal stack.  Locking not
327 *	required as the system is still cold.
328 */
329void
330siginit(struct proc *p)
331{
332	struct lwp *l;
333	struct sigacts *ps;
334	int signo, prop;
335
336	ps = p->p_sigacts;
337	sigemptyset(&contsigmask);
338	sigemptyset(&stopsigmask);
339	sigemptyset(&vforksigmask);
340	sigemptyset(&sigcantmask);
341	for (signo = 1; signo < NSIG; signo++) {
342		prop = sigprop[signo];
343		if (prop & SA_CONT)
344			sigaddset(&contsigmask, signo);
345		if (prop & SA_STOP)
346			sigaddset(&stopsigmask, signo);
347		if (prop & SA_STOP && signo != SIGSTOP)
348			sigaddset(&vforksigmask, signo);
349		if (prop & SA_CANTMASK)
350			sigaddset(&sigcantmask, signo);
351		if (prop & SA_IGNORE && signo != SIGCONT)
352			sigaddset(&p->p_sigctx.ps_sigignore, signo);
353		sigemptyset(&SIGACTION_PS(ps, signo).sa_mask);
354		SIGACTION_PS(ps, signo).sa_flags = SA_RESTART;
355	}
356	sigemptyset(&p->p_sigctx.ps_sigcatch);
357	p->p_sflag &= ~PS_NOCLDSTOP;
358
359	ksiginfo_queue_init(&p->p_sigpend.sp_info);
360	sigemptyset(&p->p_sigpend.sp_set);
361
362	/*
363	 * Reset per LWP state.
364	 */
365	l = LIST_FIRST(&p->p_lwps);
366	l->l_sigwaited = NULL;
367	l->l_sigstk = SS_INIT;
368	ksiginfo_queue_init(&l->l_sigpend.sp_info);
369	sigemptyset(&l->l_sigpend.sp_set);
370
371	/* One reference. */
372	ps->sa_refcnt = 1;
373}
374
375/*
376 * execsigs:
377 *
378 *	Reset signals for an exec of the specified process.
379 */
380void
381execsigs(struct proc *p)
382{
383	struct sigacts *ps;
384	struct lwp *l;
385	int signo, prop;
386	sigset_t tset;
387	ksiginfoq_t kq;
388
389	KASSERT(p->p_nlwps == 1);
390
391	sigactsunshare(p);
392	ps = p->p_sigacts;
393
394	/*
395	 * Reset caught signals.  Held signals remain held through
396	 * l->l_sigmask (unless they were caught, and are now ignored
397	 * by default).
398	 *
399	 * No need to lock yet, the process has only one LWP and
400	 * at this point the sigacts are private to the process.
401	 */
402	sigemptyset(&tset);
403	for (signo = 1; signo < NSIG; signo++) {
404		if (sigismember(&p->p_sigctx.ps_sigcatch, signo)) {
405			prop = sigprop[signo];
406			if (prop & SA_IGNORE) {
407				if ((prop & SA_CONT) == 0)
408					sigaddset(&p->p_sigctx.ps_sigignore,
409					    signo);
410				sigaddset(&tset, signo);
411			}
412			SIGACTION_PS(ps, signo).sa_handler = SIG_DFL;
413		}
414		sigemptyset(&SIGACTION_PS(ps, signo).sa_mask);
415		SIGACTION_PS(ps, signo).sa_flags = SA_RESTART;
416	}
417	ksiginfo_queue_init(&kq);
418
419	mutex_enter(p->p_lock);
420	sigclearall(p, &tset, &kq);
421	sigemptyset(&p->p_sigctx.ps_sigcatch);
422
423	/*
424	 * Reset no zombies if child dies flag as Solaris does.
425	 */
426	p->p_flag &= ~(PK_NOCLDWAIT | PK_CLDSIGIGN);
427	if (SIGACTION_PS(ps, SIGCHLD).sa_handler == SIG_IGN)
428		SIGACTION_PS(ps, SIGCHLD).sa_handler = SIG_DFL;
429
430	/*
431	 * Reset per-LWP state.
432	 */
433	l = LIST_FIRST(&p->p_lwps);
434	l->l_sigwaited = NULL;
435	l->l_sigstk = SS_INIT;
436	ksiginfo_queue_init(&l->l_sigpend.sp_info);
437	sigemptyset(&l->l_sigpend.sp_set);
438	mutex_exit(p->p_lock);
439
440	ksiginfo_queue_drain(&kq);
441}
442
443/*
444 * ksiginfo_exechook:
445 *
446 *	Free all pending ksiginfo entries from a process on exec.
447 *	Additionally, drain any unused ksiginfo structures in the
448 *	system back to the pool.
449 *
450 *	XXX This should not be a hook, every process has signals.
451 */
452static void
453ksiginfo_exechook(struct proc *p, void *v)
454{
455	ksiginfoq_t kq;
456
457	ksiginfo_queue_init(&kq);
458
459	mutex_enter(p->p_lock);
460	sigclearall(p, NULL, &kq);
461	mutex_exit(p->p_lock);
462
463	ksiginfo_queue_drain(&kq);
464}
465
466/*
467 * ksiginfo_alloc:
468 *
469 *	Allocate a new ksiginfo structure from the pool, and optionally copy
470 *	an existing one.  If the existing ksiginfo_t is from the pool, and
471 *	has not been queued somewhere, then just return it.  Additionally,
472 *	if the existing ksiginfo_t does not contain any information beyond
473 *	the signal number, then just return it.
474 */
475ksiginfo_t *
476ksiginfo_alloc(struct proc *p, ksiginfo_t *ok, int flags)
477{
478	ksiginfo_t *kp;
479
480	if (ok != NULL) {
481		if ((ok->ksi_flags & (KSI_QUEUED | KSI_FROMPOOL)) ==
482		    KSI_FROMPOOL)
483			return ok;
484		if (KSI_EMPTY_P(ok))
485			return ok;
486	}
487
488	kp = pool_cache_get(ksiginfo_cache, flags);
489	if (kp == NULL) {
490#ifdef DIAGNOSTIC
491		printf("Out of memory allocating ksiginfo for pid %d\n",
492		    p->p_pid);
493#endif
494		return NULL;
495	}
496
497	if (ok != NULL) {
498		memcpy(kp, ok, sizeof(*kp));
499		kp->ksi_flags &= ~KSI_QUEUED;
500	} else
501		KSI_INIT_EMPTY(kp);
502
503	kp->ksi_flags |= KSI_FROMPOOL;
504
505	return kp;
506}
507
508/*
509 * ksiginfo_free:
510 *
511 *	If the given ksiginfo_t is from the pool and has not been queued,
512 *	then free it.
513 */
514void
515ksiginfo_free(ksiginfo_t *kp)
516{
517
518	if ((kp->ksi_flags & (KSI_QUEUED | KSI_FROMPOOL)) != KSI_FROMPOOL)
519		return;
520	pool_cache_put(ksiginfo_cache, kp);
521}
522
523/*
524 * ksiginfo_queue_drain:
525 *
526 *	Drain a non-empty ksiginfo_t queue.
527 */
528void
529ksiginfo_queue_drain0(ksiginfoq_t *kq)
530{
531	ksiginfo_t *ksi;
532
533	KASSERT(!TAILQ_EMPTY(kq));
534
535	while (!TAILQ_EMPTY(kq)) {
536		ksi = TAILQ_FIRST(kq);
537		TAILQ_REMOVE(kq, ksi, ksi_list);
538		pool_cache_put(ksiginfo_cache, ksi);
539	}
540}
541
542static int
543siggetinfo(sigpend_t *sp, ksiginfo_t *out, int signo)
544{
545	ksiginfo_t *ksi, *nksi;
546
547	if (sp == NULL)
548		goto out;
549
550	/* Find siginfo and copy it out. */
551	int count = 0;
552	TAILQ_FOREACH_SAFE(ksi, &sp->sp_info, ksi_list, nksi) {
553		if (ksi->ksi_signo != signo)
554			continue;
555		if (count++ > 0) /* Only remove the first, count all of them */
556			continue;
557		TAILQ_REMOVE(&sp->sp_info, ksi, ksi_list);
558		KASSERT((ksi->ksi_flags & KSI_FROMPOOL) != 0);
559		KASSERT((ksi->ksi_flags & KSI_QUEUED) != 0);
560		ksi->ksi_flags &= ~KSI_QUEUED;
561		if (out != NULL) {
562			memcpy(out, ksi, sizeof(*out));
563			out->ksi_flags &= ~(KSI_FROMPOOL | KSI_QUEUED);
564		}
565		ksiginfo_free(ksi);
566	}
567	if (count)
568		return count;
569
570out:
571	/* If there is no siginfo, then manufacture it. */
572	if (out != NULL) {
573		KSI_INIT(out);
574		out->ksi_info._signo = signo;
575		out->ksi_info._code = SI_NOINFO;
576	}
577	return 0;
578}
579
580/*
581 * sigget:
582 *
583 *	Fetch the first pending signal from a set.  Optionally, also fetch
584 *	or manufacture a ksiginfo element.  Returns the number of the first
585 *	pending signal, or zero.
586 */
587int
588sigget(sigpend_t *sp, ksiginfo_t *out, int signo, const sigset_t *mask)
589{
590	sigset_t tset;
591	int count;
592
593	/* If there's no pending set, the signal is from the debugger. */
594	if (sp == NULL)
595		goto out;
596
597	/* Construct mask from signo, and 'mask'. */
598	if (signo == 0) {
599		if (mask != NULL) {
600			tset = *mask;
601			__sigandset(&sp->sp_set, &tset);
602		} else
603			tset = sp->sp_set;
604
605		/* If there are no signals pending - return. */
606		if ((signo = firstsig(&tset)) == 0)
607			goto out;
608	} else {
609		KASSERT(sigismember(&sp->sp_set, signo));
610	}
611
612	sigdelset(&sp->sp_set, signo);
613out:
614	count = siggetinfo(sp, out, signo);
615	if (count > 1)
616		sigaddset(&sp->sp_set, signo);
617	return signo;
618}
619
620/*
621 * sigput:
622 *
623 *	Append a new ksiginfo element to the list of pending ksiginfo's.
624 */
625static int
626sigput(sigpend_t *sp, struct proc *p, ksiginfo_t *ksi)
627{
628	ksiginfo_t *kp;
629
630	KASSERT(mutex_owned(p->p_lock));
631	KASSERT((ksi->ksi_flags & KSI_QUEUED) == 0);
632
633	sigaddset(&sp->sp_set, ksi->ksi_signo);
634
635	/*
636	 * If there is no siginfo, we are done.
637	 */
638	if (KSI_EMPTY_P(ksi))
639		return 0;
640
641	KASSERT((ksi->ksi_flags & KSI_FROMPOOL) != 0);
642
643	size_t count = 0;
644	TAILQ_FOREACH(kp, &sp->sp_info, ksi_list) {
645		count++;
646		if (ksi->ksi_signo >= SIGRTMIN && ksi->ksi_signo <= SIGRTMAX)
647			continue;
648		if (kp->ksi_signo == ksi->ksi_signo) {
649			KSI_COPY(ksi, kp);
650			kp->ksi_flags |= KSI_QUEUED;
651			return 0;
652		}
653	}
654
655	if (count >= SIGQUEUE_MAX) {
656#ifdef DIAGNOSTIC
657		printf("%s(%d): Signal queue is full signal=%d\n",
658		    p->p_comm, p->p_pid, ksi->ksi_signo);
659#endif
660		return EAGAIN;
661	}
662	ksi->ksi_flags |= KSI_QUEUED;
663	TAILQ_INSERT_TAIL(&sp->sp_info, ksi, ksi_list);
664
665	return 0;
666}
667
668/*
669 * sigclear:
670 *
671 *	Clear all pending signals in the specified set.
672 */
673void
674sigclear(sigpend_t *sp, const sigset_t *mask, ksiginfoq_t *kq)
675{
676	ksiginfo_t *ksi, *next;
677
678	if (mask == NULL)
679		sigemptyset(&sp->sp_set);
680	else
681		sigminusset(mask, &sp->sp_set);
682
683	TAILQ_FOREACH_SAFE(ksi, &sp->sp_info, ksi_list, next) {
684		if (mask == NULL || sigismember(mask, ksi->ksi_signo)) {
685			TAILQ_REMOVE(&sp->sp_info, ksi, ksi_list);
686			KASSERT((ksi->ksi_flags & KSI_FROMPOOL) != 0);
687			KASSERT((ksi->ksi_flags & KSI_QUEUED) != 0);
688			TAILQ_INSERT_TAIL(kq, ksi, ksi_list);
689		}
690	}
691}
692
693/*
694 * sigclearall:
695 *
696 *	Clear all pending signals in the specified set from a process and
697 *	its LWPs.
698 */
699void
700sigclearall(struct proc *p, const sigset_t *mask, ksiginfoq_t *kq)
701{
702	struct lwp *l;
703
704	KASSERT(mutex_owned(p->p_lock));
705
706	sigclear(&p->p_sigpend, mask, kq);
707
708	LIST_FOREACH(l, &p->p_lwps, l_sibling) {
709		sigclear(&l->l_sigpend, mask, kq);
710	}
711}
712
713/*
714 * sigispending:
715 *
716 *	Return the first signal number if there are pending signals for the
717 *	current LWP.  May be called unlocked provided that LW_PENDSIG is set,
718 *	and that the signal has been posted to the appopriate queue before
719 *	LW_PENDSIG is set.
720 *
721 *	This should only ever be called with (l == curlwp), unless the
722 *	result does not matter (procfs, sysctl).
723 */
724int
725sigispending(struct lwp *l, int signo)
726{
727	struct proc *p = l->l_proc;
728	sigset_t tset;
729
730	membar_consumer();
731
732	tset = l->l_sigpend.sp_set;
733	sigplusset(&p->p_sigpend.sp_set, &tset);
734	sigminusset(&p->p_sigctx.ps_sigignore, &tset);
735	sigminusset(&l->l_sigmask, &tset);
736
737	if (signo == 0) {
738		return firstsig(&tset);
739	}
740	return sigismember(&tset, signo) ? signo : 0;
741}
742
743void
744getucontext(struct lwp *l, ucontext_t *ucp)
745{
746	struct proc *p = l->l_proc;
747
748	KASSERT(mutex_owned(p->p_lock));
749
750	ucp->uc_flags = 0;
751	ucp->uc_link = l->l_ctxlink;
752	ucp->uc_sigmask = l->l_sigmask;
753	ucp->uc_flags |= _UC_SIGMASK;
754
755	/*
756	 * The (unsupplied) definition of the `current execution stack'
757	 * in the System V Interface Definition appears to allow returning
758	 * the main context stack.
759	 */
760	if ((l->l_sigstk.ss_flags & SS_ONSTACK) == 0) {
761		ucp->uc_stack.ss_sp = (void *)l->l_proc->p_stackbase;
762		ucp->uc_stack.ss_size = ctob(l->l_proc->p_vmspace->vm_ssize);
763		ucp->uc_stack.ss_flags = 0;	/* XXX, def. is Very Fishy */
764	} else {
765		/* Simply copy alternate signal execution stack. */
766		ucp->uc_stack = l->l_sigstk;
767	}
768	ucp->uc_flags |= _UC_STACK;
769	mutex_exit(p->p_lock);
770	cpu_getmcontext(l, &ucp->uc_mcontext, &ucp->uc_flags);
771	mutex_enter(p->p_lock);
772}
773
774int
775setucontext(struct lwp *l, const ucontext_t *ucp)
776{
777	struct proc *p = l->l_proc;
778	int error;
779
780	KASSERT(mutex_owned(p->p_lock));
781
782	if ((ucp->uc_flags & _UC_SIGMASK) != 0) {
783		error = sigprocmask1(l, SIG_SETMASK, &ucp->uc_sigmask, NULL);
784		if (error != 0)
785			return error;
786	}
787
788	mutex_exit(p->p_lock);
789	error = cpu_setmcontext(l, &ucp->uc_mcontext, ucp->uc_flags);
790	mutex_enter(p->p_lock);
791	if (error != 0)
792		return (error);
793
794	l->l_ctxlink = ucp->uc_link;
795
796	/*
797	 * If there was stack information, update whether or not we are
798	 * still running on an alternate signal stack.
799	 */
800	if ((ucp->uc_flags & _UC_STACK) != 0) {
801		if (ucp->uc_stack.ss_flags & SS_ONSTACK)
802			l->l_sigstk.ss_flags |= SS_ONSTACK;
803		else
804			l->l_sigstk.ss_flags &= ~SS_ONSTACK;
805	}
806
807	return 0;
808}
809
810/*
811 * killpg1: common code for kill process group/broadcast kill.
812 */
813int
814killpg1(struct lwp *l, ksiginfo_t *ksi, int pgid, int all)
815{
816	struct proc	*p, *cp;
817	kauth_cred_t	pc;
818	struct pgrp	*pgrp;
819	int		nfound;
820	int		signo = ksi->ksi_signo;
821
822	cp = l->l_proc;
823	pc = l->l_cred;
824	nfound = 0;
825
826	mutex_enter(&proc_lock);
827	if (all) {
828		/*
829		 * Broadcast.
830		 */
831		PROCLIST_FOREACH(p, &allproc) {
832			if (p->p_pid <= 1 || p == cp ||
833			    (p->p_flag & PK_SYSTEM) != 0)
834				continue;
835			mutex_enter(p->p_lock);
836			if (kauth_authorize_process(pc,
837			    KAUTH_PROCESS_SIGNAL, p, KAUTH_ARG(signo), NULL,
838			    NULL) == 0) {
839				nfound++;
840				if (signo)
841					kpsignal2(p, ksi);
842			}
843			mutex_exit(p->p_lock);
844		}
845	} else {
846		if (pgid == 0)
847			/* Zero pgid means send to my process group. */
848			pgrp = cp->p_pgrp;
849		else {
850			pgrp = pgrp_find(pgid);
851			if (pgrp == NULL)
852				goto out;
853		}
854		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
855			if (p->p_pid <= 1 || p->p_flag & PK_SYSTEM)
856				continue;
857			mutex_enter(p->p_lock);
858			if (kauth_authorize_process(pc, KAUTH_PROCESS_SIGNAL,
859			    p, KAUTH_ARG(signo), NULL, NULL) == 0) {
860				nfound++;
861				if (signo && P_ZOMBIE(p) == 0)
862					kpsignal2(p, ksi);
863			}
864			mutex_exit(p->p_lock);
865		}
866	}
867out:
868	mutex_exit(&proc_lock);
869	return nfound ? 0 : ESRCH;
870}
871
872/*
873 * Send a signal to a process group.  If checktty is set, limit to members
874 * which have a controlling terminal.
875 */
876void
877pgsignal(struct pgrp *pgrp, int sig, int checkctty)
878{
879	ksiginfo_t ksi;
880
881	KASSERT(!cpu_intr_p());
882	KASSERT(mutex_owned(&proc_lock));
883
884	KSI_INIT_EMPTY(&ksi);
885	ksi.ksi_signo = sig;
886	kpgsignal(pgrp, &ksi, NULL, checkctty);
887}
888
889void
890kpgsignal(struct pgrp *pgrp, ksiginfo_t *ksi, void *data, int checkctty)
891{
892	struct proc *p;
893
894	KASSERT(!cpu_intr_p());
895	KASSERT(mutex_owned(&proc_lock));
896	KASSERT(pgrp != NULL);
897
898	LIST_FOREACH(p, &pgrp->pg_members, p_pglist)
899		if (checkctty == 0 || p->p_lflag & PL_CONTROLT)
900			kpsignal(p, ksi, data);
901}
902
903/*
904 * Send a signal caused by a trap to the current LWP.  If it will be caught
905 * immediately, deliver it with correct code.  Otherwise, post it normally.
906 */
907void
908trapsignal(struct lwp *l, ksiginfo_t *ksi)
909{
910	struct proc	*p;
911	struct sigacts	*ps;
912	int signo = ksi->ksi_signo;
913	sigset_t *mask;
914	sig_t action;
915
916	KASSERT(KSI_TRAP_P(ksi));
917
918	ksi->ksi_lid = l->l_lid;
919	p = l->l_proc;
920
921	KASSERT(!cpu_intr_p());
922	mutex_enter(&proc_lock);
923	mutex_enter(p->p_lock);
924
925repeat:
926	/*
927	 * If we are exiting, demise now.
928	 *
929	 * This avoids notifying tracer and deadlocking.
930	 */
931	if (__predict_false(ISSET(p->p_sflag, PS_WEXIT))) {
932		mutex_exit(p->p_lock);
933		mutex_exit(&proc_lock);
934		lwp_exit(l);
935		panic("trapsignal");
936		/* NOTREACHED */
937	}
938
939	/*
940	 * The process is already stopping.
941	 */
942	if ((p->p_sflag & PS_STOPPING) != 0) {
943		mutex_exit(&proc_lock);
944		sigswitch_unlock_and_switch_away(l);
945		mutex_enter(&proc_lock);
946		mutex_enter(p->p_lock);
947		goto repeat;
948	}
949
950	mask = &l->l_sigmask;
951	ps = p->p_sigacts;
952	action = SIGACTION_PS(ps, signo).sa_handler;
953
954	if (ISSET(p->p_slflag, PSL_TRACED) &&
955	    !(p->p_pptr == p->p_opptr && ISSET(p->p_lflag, PL_PPWAIT)) &&
956	    p->p_xsig != SIGKILL &&
957	    !sigismember(&p->p_sigpend.sp_set, SIGKILL)) {
958		p->p_xsig = signo;
959		p->p_sigctx.ps_faked = true;
960		p->p_sigctx.ps_lwp = ksi->ksi_lid;
961		p->p_sigctx.ps_info = ksi->ksi_info;
962		sigswitch(0, signo, true);
963
964		if (ktrpoint(KTR_PSIG)) {
965			if (p->p_emul->e_ktrpsig)
966				p->p_emul->e_ktrpsig(signo, action, mask, ksi);
967			else
968				ktrpsig(signo, action, mask, ksi);
969		}
970		return;
971	}
972
973	const bool caught = sigismember(&p->p_sigctx.ps_sigcatch, signo);
974	const bool masked = sigismember(mask, signo);
975	if (caught && !masked) {
976		mutex_exit(&proc_lock);
977		l->l_ru.ru_nsignals++;
978		kpsendsig(l, ksi, mask);
979		mutex_exit(p->p_lock);
980
981		if (ktrpoint(KTR_PSIG)) {
982			if (p->p_emul->e_ktrpsig)
983				p->p_emul->e_ktrpsig(signo, action, mask, ksi);
984			else
985				ktrpsig(signo, action, mask, ksi);
986		}
987		return;
988	}
989
990	/*
991	 * If the signal is masked or ignored, then unmask it and
992	 * reset it to the default action so that the process or
993	 * its tracer will be notified.
994	 */
995	const bool ignored = action == SIG_IGN;
996	if (masked || ignored) {
997		mutex_enter(&ps->sa_mutex);
998		sigdelset(mask, signo);
999		sigdelset(&p->p_sigctx.ps_sigcatch, signo);
1000		sigdelset(&p->p_sigctx.ps_sigignore, signo);
1001		sigdelset(&SIGACTION_PS(ps, signo).sa_mask, signo);
1002		SIGACTION_PS(ps, signo).sa_handler = SIG_DFL;
1003		mutex_exit(&ps->sa_mutex);
1004	}
1005
1006	kpsignal2(p, ksi);
1007	mutex_exit(p->p_lock);
1008	mutex_exit(&proc_lock);
1009}
1010
1011/*
1012 * Fill in signal information and signal the parent for a child status change.
1013 */
1014void
1015child_psignal(struct proc *p, int mask)
1016{
1017	ksiginfo_t ksi;
1018	struct proc *q;
1019	int xsig;
1020
1021	KASSERT(mutex_owned(&proc_lock));
1022	KASSERT(mutex_owned(p->p_lock));
1023
1024	xsig = p->p_xsig;
1025
1026	KSI_INIT(&ksi);
1027	ksi.ksi_signo = SIGCHLD;
1028	ksi.ksi_code = (xsig == SIGCONT ? CLD_CONTINUED : CLD_STOPPED);
1029	ksi.ksi_pid = p->p_pid;
1030	ksi.ksi_uid = kauth_cred_geteuid(p->p_cred);
1031	ksi.ksi_status = xsig;
1032	ksi.ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
1033	ksi.ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
1034
1035	q = p->p_pptr;
1036
1037	mutex_exit(p->p_lock);
1038	mutex_enter(q->p_lock);
1039
1040	if ((q->p_sflag & mask) == 0)
1041		kpsignal2(q, &ksi);
1042
1043	mutex_exit(q->p_lock);
1044	mutex_enter(p->p_lock);
1045}
1046
1047void
1048psignal(struct proc *p, int signo)
1049{
1050	ksiginfo_t ksi;
1051
1052	KASSERT(!cpu_intr_p());
1053	KASSERT(mutex_owned(&proc_lock));
1054
1055	KSI_INIT_EMPTY(&ksi);
1056	ksi.ksi_signo = signo;
1057	mutex_enter(p->p_lock);
1058	kpsignal2(p, &ksi);
1059	mutex_exit(p->p_lock);
1060}
1061
1062void
1063kpsignal(struct proc *p, ksiginfo_t *ksi, void *data)
1064{
1065	fdfile_t *ff;
1066	file_t *fp;
1067	fdtab_t *dt;
1068
1069	KASSERT(!cpu_intr_p());
1070	KASSERT(mutex_owned(&proc_lock));
1071
1072	if ((p->p_sflag & PS_WEXIT) == 0 && data) {
1073		size_t fd;
1074		filedesc_t *fdp = p->p_fd;
1075
1076		/* XXXSMP locking */
1077		ksi->ksi_fd = -1;
1078		dt = atomic_load_consume(&fdp->fd_dt);
1079		for (fd = 0; fd < dt->dt_nfiles; fd++) {
1080			if ((ff = dt->dt_ff[fd]) == NULL)
1081				continue;
1082			if ((fp = atomic_load_consume(&ff->ff_file)) == NULL)
1083				continue;
1084			if (fp->f_data == data) {
1085				ksi->ksi_fd = fd;
1086				break;
1087			}
1088		}
1089	}
1090	mutex_enter(p->p_lock);
1091	kpsignal2(p, ksi);
1092	mutex_exit(p->p_lock);
1093}
1094
1095/*
1096 * sigismasked:
1097 *
1098 *	Returns true if signal is ignored or masked for the specified LWP.
1099 */
1100int
1101sigismasked(struct lwp *l, int sig)
1102{
1103	struct proc *p = l->l_proc;
1104
1105	return sigismember(&p->p_sigctx.ps_sigignore, sig) ||
1106	    sigismember(&l->l_sigmask, sig);
1107}
1108
1109/*
1110 * sigpost:
1111 *
1112 *	Post a pending signal to an LWP.  Returns non-zero if the LWP may
1113 *	be able to take the signal.
1114 */
1115static int
1116sigpost(struct lwp *l, sig_t action, int prop, int sig)
1117{
1118	int rv, masked;
1119	struct proc *p = l->l_proc;
1120
1121	KASSERT(mutex_owned(p->p_lock));
1122
1123	/*
1124	 * If the LWP is on the way out, sigclear() will be busy draining all
1125	 * pending signals.  Don't give it more.
1126	 */
1127	if (l->l_stat == LSZOMB)
1128		return 0;
1129
1130	SDT_PROBE(proc, kernel, , signal__send, l, p, sig, 0, 0);
1131
1132	lwp_lock(l);
1133	if (__predict_false((l->l_flag & LW_DBGSUSPEND) != 0)) {
1134		if ((prop & SA_KILL) != 0)
1135			l->l_flag &= ~LW_DBGSUSPEND;
1136		else {
1137			lwp_unlock(l);
1138			return 0;
1139		}
1140	}
1141
1142	/*
1143	 * Have the LWP check for signals.  This ensures that even if no LWP
1144	 * is found to take the signal immediately, it should be taken soon.
1145	 */
1146	signotify(l);
1147
1148	/*
1149	 * SIGCONT can be masked, but if LWP is stopped, it needs restart.
1150	 * Note: SIGKILL and SIGSTOP cannot be masked.
1151	 */
1152	masked = sigismember(&l->l_sigmask, sig);
1153	if (masked && ((prop & SA_CONT) == 0 || l->l_stat != LSSTOP)) {
1154		lwp_unlock(l);
1155		return 0;
1156	}
1157
1158	/*
1159	 * If killing the process, make it run fast.
1160	 */
1161	if (__predict_false((prop & SA_KILL) != 0) &&
1162	    action == SIG_DFL && l->l_priority < MAXPRI_USER) {
1163		KASSERT(l->l_class == SCHED_OTHER);
1164		lwp_changepri(l, MAXPRI_USER);
1165	}
1166
1167	/*
1168	 * If the LWP is running or on a run queue, then we win.  If it's
1169	 * sleeping interruptably, wake it and make it take the signal.  If
1170	 * the sleep isn't interruptable, then the chances are it will get
1171	 * to see the signal soon anyhow.  If suspended, it can't take the
1172	 * signal right now.  If it's LWP private or for all LWPs, save it
1173	 * for later; otherwise punt.
1174	 */
1175	rv = 0;
1176
1177	switch (l->l_stat) {
1178	case LSRUN:
1179	case LSONPROC:
1180		rv = 1;
1181		break;
1182
1183	case LSSLEEP:
1184		if ((l->l_flag & LW_SINTR) != 0) {
1185			/* setrunnable() will release the lock. */
1186			setrunnable(l);
1187			return 1;
1188		}
1189		break;
1190
1191	case LSSUSPENDED:
1192		if ((prop & SA_KILL) != 0 && (l->l_flag & LW_WCORE) != 0) {
1193			/* lwp_continue() will release the lock. */
1194			lwp_continue(l);
1195			return 1;
1196		}
1197		break;
1198
1199	case LSSTOP:
1200		if ((prop & SA_STOP) != 0)
1201			break;
1202
1203		/*
1204		 * If the LWP is stopped and we are sending a continue
1205		 * signal, then start it again.
1206		 */
1207		if ((prop & SA_CONT) != 0) {
1208			if (l->l_wchan != NULL) {
1209				l->l_stat = LSSLEEP;
1210				p->p_nrlwps++;
1211				rv = 1;
1212				break;
1213			}
1214			/* setrunnable() will release the lock. */
1215			setrunnable(l);
1216			return 1;
1217		} else if (l->l_wchan == NULL || (l->l_flag & LW_SINTR) != 0) {
1218			/* setrunnable() will release the lock. */
1219			setrunnable(l);
1220			return 1;
1221		}
1222		break;
1223
1224	default:
1225		break;
1226	}
1227
1228	lwp_unlock(l);
1229	return rv;
1230}
1231
1232/*
1233 * Notify an LWP that it has a pending signal.
1234 */
1235void
1236signotify(struct lwp *l)
1237{
1238	KASSERT(lwp_locked(l, NULL));
1239
1240	l->l_flag |= LW_PENDSIG;
1241	lwp_need_userret(l);
1242}
1243
1244/*
1245 * Find an LWP within process p that is waiting on signal ksi, and hand
1246 * it on.
1247 */
1248static int
1249sigunwait(struct proc *p, const ksiginfo_t *ksi)
1250{
1251	struct lwp *l;
1252	int signo;
1253
1254	KASSERT(mutex_owned(p->p_lock));
1255
1256	signo = ksi->ksi_signo;
1257
1258	if (ksi->ksi_lid != 0) {
1259		/*
1260		 * Signal came via _lwp_kill().  Find the LWP and see if
1261		 * it's interested.
1262		 */
1263		if ((l = lwp_find(p, ksi->ksi_lid)) == NULL)
1264			return 0;
1265		if (l->l_sigwaited == NULL ||
1266		    !sigismember(&l->l_sigwaitset, signo))
1267			return 0;
1268	} else {
1269		/*
1270		 * Look for any LWP that may be interested.
1271		 */
1272		LIST_FOREACH(l, &p->p_sigwaiters, l_sigwaiter) {
1273			KASSERT(l->l_sigwaited != NULL);
1274			if (sigismember(&l->l_sigwaitset, signo))
1275				break;
1276		}
1277	}
1278
1279	if (l != NULL) {
1280		l->l_sigwaited->ksi_info = ksi->ksi_info;
1281		l->l_sigwaited = NULL;
1282		LIST_REMOVE(l, l_sigwaiter);
1283		cv_signal(&l->l_sigcv);
1284		return 1;
1285	}
1286
1287	return 0;
1288}
1289
1290/*
1291 * Send the signal to the process.  If the signal has an action, the action
1292 * is usually performed by the target process rather than the caller; we add
1293 * the signal to the set of pending signals for the process.
1294 *
1295 * Exceptions:
1296 *   o When a stop signal is sent to a sleeping process that takes the
1297 *     default action, the process is stopped without awakening it.
1298 *   o SIGCONT restarts stopped processes (or puts them back to sleep)
1299 *     regardless of the signal action (eg, blocked or ignored).
1300 *
1301 * Other ignored signals are discarded immediately.
1302 */
1303int
1304kpsignal2(struct proc *p, ksiginfo_t *ksi)
1305{
1306	int prop, signo = ksi->ksi_signo;
1307	struct lwp *l = NULL;
1308	ksiginfo_t *kp;
1309	lwpid_t lid;
1310	sig_t action;
1311	bool toall;
1312	bool traced;
1313	int error = 0;
1314
1315	KASSERT(!cpu_intr_p());
1316	KASSERT(mutex_owned(&proc_lock));
1317	KASSERT(mutex_owned(p->p_lock));
1318	KASSERT((ksi->ksi_flags & KSI_QUEUED) == 0);
1319	KASSERT(signo > 0);
1320	KASSERT(signo < NSIG);
1321
1322	/*
1323	 * If the process is being created by fork, is a zombie or is
1324	 * exiting, then just drop the signal here and bail out.
1325	 */
1326	if (p->p_stat != SACTIVE && p->p_stat != SSTOP)
1327		return 0;
1328
1329	/*
1330	 * Notify any interested parties of the signal.
1331	 */
1332	KNOTE(&p->p_klist, NOTE_SIGNAL | signo);
1333
1334	/*
1335	 * Some signals including SIGKILL must act on the entire process.
1336	 */
1337	kp = NULL;
1338	prop = sigprop[signo];
1339	toall = ((prop & SA_TOALL) != 0);
1340	lid = toall ? 0 : ksi->ksi_lid;
1341	traced = ISSET(p->p_slflag, PSL_TRACED) &&
1342	    !sigismember(&p->p_sigctx.ps_sigpass, signo);
1343
1344	/*
1345	 * If proc is traced, always give parent a chance.
1346	 */
1347	if (traced) {
1348		action = SIG_DFL;
1349
1350		if (lid == 0) {
1351			/*
1352			 * If the process is being traced and the signal
1353			 * is being caught, make sure to save any ksiginfo.
1354			 */
1355			if ((kp = ksiginfo_alloc(p, ksi, PR_NOWAIT)) == NULL)
1356				goto discard;
1357			if ((error = sigput(&p->p_sigpend, p, kp)) != 0)
1358				goto out;
1359		}
1360	} else {
1361
1362		/*
1363		 * If the signal is being ignored, then drop it.  Note: we
1364		 * don't set SIGCONT in ps_sigignore, and if it is set to
1365		 * SIG_IGN, action will be SIG_DFL here.
1366		 */
1367		if (sigismember(&p->p_sigctx.ps_sigignore, signo))
1368			goto discard;
1369
1370		else if (sigismember(&p->p_sigctx.ps_sigcatch, signo))
1371			action = SIG_CATCH;
1372		else {
1373			action = SIG_DFL;
1374
1375			/*
1376			 * If sending a tty stop signal to a member of an
1377			 * orphaned process group, discard the signal here if
1378			 * the action is default; don't stop the process below
1379			 * if sleeping, and don't clear any pending SIGCONT.
1380			 */
1381			if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0)
1382				goto discard;
1383
1384			if (prop & SA_KILL && p->p_nice > NZERO)
1385				p->p_nice = NZERO;
1386		}
1387	}
1388
1389	/*
1390	 * If stopping or continuing a process, discard any pending
1391	 * signals that would do the inverse.
1392	 */
1393	if ((prop & (SA_CONT | SA_STOP)) != 0) {
1394		ksiginfoq_t kq;
1395
1396		ksiginfo_queue_init(&kq);
1397		if ((prop & SA_CONT) != 0)
1398			sigclear(&p->p_sigpend, &stopsigmask, &kq);
1399		if ((prop & SA_STOP) != 0)
1400			sigclear(&p->p_sigpend, &contsigmask, &kq);
1401		ksiginfo_queue_drain(&kq);	/* XXXSMP */
1402	}
1403
1404	/*
1405	 * If the signal doesn't have SA_CANTMASK (no override for SIGKILL,
1406	 * please!), check if any LWPs are waiting on it.  If yes, pass on
1407	 * the signal info.  The signal won't be processed further here.
1408	 */
1409	if ((prop & SA_CANTMASK) == 0 && !LIST_EMPTY(&p->p_sigwaiters) &&
1410	    p->p_stat == SACTIVE && (p->p_sflag & PS_STOPPING) == 0 &&
1411	    sigunwait(p, ksi))
1412		goto discard;
1413
1414	/*
1415	 * XXXSMP Should be allocated by the caller, we're holding locks
1416	 * here.
1417	 */
1418	if (kp == NULL && (kp = ksiginfo_alloc(p, ksi, PR_NOWAIT)) == NULL)
1419		goto discard;
1420
1421	/*
1422	 * LWP private signals are easy - just find the LWP and post
1423	 * the signal to it.
1424	 */
1425	if (lid != 0) {
1426		l = lwp_find(p, lid);
1427		if (l != NULL) {
1428			if ((error = sigput(&l->l_sigpend, p, kp)) != 0)
1429				goto out;
1430			membar_producer();
1431			if (sigpost(l, action, prop, kp->ksi_signo) != 0)
1432				signo = -1;
1433		}
1434		goto out;
1435	}
1436
1437	/*
1438	 * Some signals go to all LWPs, even if posted with _lwp_kill()
1439	 * or for an SA process.
1440	 */
1441	if (p->p_stat == SACTIVE && (p->p_sflag & PS_STOPPING) == 0) {
1442		if (traced)
1443			goto deliver;
1444
1445		/*
1446		 * If SIGCONT is default (or ignored) and process is
1447		 * asleep, we are finished; the process should not
1448		 * be awakened.
1449		 */
1450		if ((prop & SA_CONT) != 0 && action == SIG_DFL)
1451			goto out;
1452	} else {
1453		/*
1454		 * Process is stopped or stopping.
1455		 * - If traced, then no action is needed, unless killing.
1456		 * - Run the process only if sending SIGCONT or SIGKILL.
1457		 */
1458		if (traced && signo != SIGKILL) {
1459			goto out;
1460		}
1461		if ((prop & SA_CONT) != 0 || signo == SIGKILL) {
1462			/*
1463			 * Re-adjust p_nstopchild if the process was
1464			 * stopped but not yet collected by its parent.
1465			 */
1466			if (p->p_stat == SSTOP && !p->p_waited)
1467				p->p_pptr->p_nstopchild--;
1468			p->p_stat = SACTIVE;
1469			p->p_sflag &= ~PS_STOPPING;
1470			if (traced) {
1471				KASSERT(signo == SIGKILL);
1472				goto deliver;
1473			}
1474			/*
1475			 * Do not make signal pending if SIGCONT is default.
1476			 *
1477			 * If the process catches SIGCONT, let it handle the
1478			 * signal itself (if waiting on event - process runs,
1479			 * otherwise continues sleeping).
1480			 */
1481			if ((prop & SA_CONT) != 0) {
1482				p->p_xsig = SIGCONT;
1483				p->p_sflag |= PS_CONTINUED;
1484				child_psignal(p, 0);
1485				if (action == SIG_DFL) {
1486					KASSERT(signo != SIGKILL);
1487					goto deliver;
1488				}
1489			}
1490		} else if ((prop & SA_STOP) != 0) {
1491			/*
1492			 * Already stopped, don't need to stop again.
1493			 * (If we did the shell could get confused.)
1494			 */
1495			goto out;
1496		}
1497	}
1498	/*
1499	 * Make signal pending.
1500	 */
1501	KASSERT(!traced);
1502	if ((error = sigput(&p->p_sigpend, p, kp)) != 0)
1503		goto out;
1504deliver:
1505	/*
1506	 * Before we set LW_PENDSIG on any LWP, ensure that the signal is
1507	 * visible on the per process list (for sigispending()).  This
1508	 * is unlikely to be needed in practice, but...
1509	 */
1510	membar_producer();
1511
1512	/*
1513	 * Try to find an LWP that can take the signal.
1514	 */
1515	LIST_FOREACH(l, &p->p_lwps, l_sibling) {
1516		if (sigpost(l, action, prop, kp->ksi_signo) && !toall)
1517			break;
1518	}
1519	signo = -1;
1520out:
1521	/*
1522	 * If the ksiginfo wasn't used, then bin it.  XXXSMP freeing memory
1523	 * with locks held.  The caller should take care of this.
1524	 */
1525	ksiginfo_free(kp);
1526	if (signo == -1)
1527		return error;
1528discard:
1529	SDT_PROBE(proc, kernel, , signal__discard, l, p, signo, 0, 0);
1530	return error;
1531}
1532
1533void
1534kpsendsig(struct lwp *l, const ksiginfo_t *ksi, const sigset_t *mask)
1535{
1536	struct proc *p = l->l_proc;
1537
1538	KASSERT(mutex_owned(p->p_lock));
1539	(*p->p_emul->e_sendsig)(ksi, mask);
1540}
1541
1542/*
1543 * Stop any LWPs sleeping interruptably.
1544 */
1545static void
1546proc_stop_lwps(struct proc *p)
1547{
1548	struct lwp *l;
1549
1550	KASSERT(mutex_owned(p->p_lock));
1551	KASSERT((p->p_sflag & PS_STOPPING) != 0);
1552
1553	LIST_FOREACH(l, &p->p_lwps, l_sibling) {
1554		lwp_lock(l);
1555		if (l->l_stat == LSSLEEP && (l->l_flag & LW_SINTR) != 0) {
1556			l->l_stat = LSSTOP;
1557			p->p_nrlwps--;
1558		}
1559		lwp_unlock(l);
1560	}
1561}
1562
1563/*
1564 * Finish stopping of a process.  Mark it stopped and notify the parent.
1565 *
1566 * Drop p_lock briefly if ppsig is true.
1567 */
1568static void
1569proc_stop_done(struct proc *p, int ppmask)
1570{
1571
1572	KASSERT(mutex_owned(&proc_lock));
1573	KASSERT(mutex_owned(p->p_lock));
1574	KASSERT((p->p_sflag & PS_STOPPING) != 0);
1575	KASSERT(p->p_nrlwps == 0 || p->p_nrlwps == 1);
1576	KASSERT(p->p_nrlwps == 0 || p == curproc);
1577
1578	p->p_sflag &= ~PS_STOPPING;
1579	p->p_stat = SSTOP;
1580	p->p_waited = 0;
1581	p->p_pptr->p_nstopchild++;
1582
1583	/* child_psignal drops p_lock briefly. */
1584	child_psignal(p, ppmask);
1585	cv_broadcast(&p->p_pptr->p_waitcv);
1586}
1587
1588/*
1589 * Stop the current process and switch away to the debugger notifying
1590 * an event specific to a traced process only.
1591 */
1592void
1593eventswitch(int code, int pe_report_event, int entity)
1594{
1595	struct lwp *l = curlwp;
1596	struct proc *p = l->l_proc;
1597	struct sigacts *ps;
1598	sigset_t *mask;
1599	sig_t action;
1600	ksiginfo_t ksi;
1601	const int signo = SIGTRAP;
1602
1603	KASSERT(mutex_owned(&proc_lock));
1604	KASSERT(mutex_owned(p->p_lock));
1605	KASSERT(p->p_pptr != initproc);
1606	KASSERT(l->l_stat == LSONPROC);
1607	KASSERT(ISSET(p->p_slflag, PSL_TRACED));
1608	KASSERT(!ISSET(l->l_flag, LW_SYSTEM));
1609	KASSERT(p->p_nrlwps > 0);
1610	KASSERT((code == TRAP_CHLD) || (code == TRAP_LWP) ||
1611	        (code == TRAP_EXEC));
1612	KASSERT((code != TRAP_CHLD) || (entity > 1)); /* prevent pid1 */
1613	KASSERT((code != TRAP_LWP) || (entity > 0));
1614
1615repeat:
1616	/*
1617	 * If we are exiting, demise now.
1618	 *
1619	 * This avoids notifying tracer and deadlocking.
1620	 */
1621	if (__predict_false(ISSET(p->p_sflag, PS_WEXIT))) {
1622		mutex_exit(p->p_lock);
1623		mutex_exit(&proc_lock);
1624
1625		if (pe_report_event == PTRACE_LWP_EXIT) {
1626			/* Avoid double lwp_exit() and panic. */
1627			return;
1628		}
1629
1630		lwp_exit(l);
1631		panic("eventswitch");
1632		/* NOTREACHED */
1633	}
1634
1635	/*
1636	 * If we are no longer traced, abandon this event signal.
1637	 *
1638	 * This avoids killing a process after detaching the debugger.
1639	 */
1640	if (__predict_false(!ISSET(p->p_slflag, PSL_TRACED))) {
1641		mutex_exit(p->p_lock);
1642		mutex_exit(&proc_lock);
1643		return;
1644	}
1645
1646	/*
1647	 * If there's a pending SIGKILL process it immediately.
1648	 */
1649	if (p->p_xsig == SIGKILL ||
1650	    sigismember(&p->p_sigpend.sp_set, SIGKILL)) {
1651		mutex_exit(p->p_lock);
1652		mutex_exit(&proc_lock);
1653		return;
1654	}
1655
1656	/*
1657	 * The process is already stopping.
1658	 */
1659	if ((p->p_sflag & PS_STOPPING) != 0) {
1660		mutex_exit(&proc_lock);
1661		sigswitch_unlock_and_switch_away(l);
1662		mutex_enter(&proc_lock);
1663		mutex_enter(p->p_lock);
1664		goto repeat;
1665	}
1666
1667	KSI_INIT_TRAP(&ksi);
1668	ksi.ksi_lid = l->l_lid;
1669	ksi.ksi_signo = signo;
1670	ksi.ksi_code = code;
1671	ksi.ksi_pe_report_event = pe_report_event;
1672
1673	CTASSERT(sizeof(ksi.ksi_pe_other_pid) == sizeof(ksi.ksi_pe_lwp));
1674	ksi.ksi_pe_other_pid = entity;
1675
1676	/* Needed for ktrace */
1677	ps = p->p_sigacts;
1678	action = SIGACTION_PS(ps, signo).sa_handler;
1679	mask = &l->l_sigmask;
1680
1681	p->p_xsig = signo;
1682	p->p_sigctx.ps_faked = true;
1683	p->p_sigctx.ps_lwp = ksi.ksi_lid;
1684	p->p_sigctx.ps_info = ksi.ksi_info;
1685
1686	sigswitch(0, signo, true);
1687
1688	if (code == TRAP_CHLD) {
1689		mutex_enter(&proc_lock);
1690		while (l->l_vforkwaiting)
1691			cv_wait(&l->l_waitcv, &proc_lock);
1692		mutex_exit(&proc_lock);
1693	}
1694
1695	if (ktrpoint(KTR_PSIG)) {
1696		if (p->p_emul->e_ktrpsig)
1697			p->p_emul->e_ktrpsig(signo, action, mask, &ksi);
1698		else
1699			ktrpsig(signo, action, mask, &ksi);
1700	}
1701}
1702
1703void
1704eventswitchchild(struct proc *p, int code, int pe_report_event)
1705{
1706	mutex_enter(&proc_lock);
1707	mutex_enter(p->p_lock);
1708	if ((p->p_slflag & (PSL_TRACED|PSL_TRACEDCHILD)) !=
1709	    (PSL_TRACED|PSL_TRACEDCHILD)) {
1710		mutex_exit(p->p_lock);
1711		mutex_exit(&proc_lock);
1712		return;
1713	}
1714	eventswitch(code, pe_report_event, p->p_oppid);
1715}
1716
1717/*
1718 * Stop the current process and switch away when being stopped or traced.
1719 */
1720static void
1721sigswitch(int ppmask, int signo, bool proc_lock_held)
1722{
1723	struct lwp *l = curlwp;
1724	struct proc *p = l->l_proc;
1725
1726	KASSERT(mutex_owned(p->p_lock));
1727	KASSERT(l->l_stat == LSONPROC);
1728	KASSERT(p->p_nrlwps > 0);
1729
1730	if (proc_lock_held) {
1731		KASSERT(mutex_owned(&proc_lock));
1732	} else {
1733		KASSERT(!mutex_owned(&proc_lock));
1734	}
1735
1736	/*
1737	 * On entry we know that the process needs to stop.  If it's
1738	 * the result of a 'sideways' stop signal that has been sourced
1739	 * through issignal(), then stop other LWPs in the process too.
1740	 */
1741	if (p->p_stat == SACTIVE && (p->p_sflag & PS_STOPPING) == 0) {
1742		KASSERT(signo != 0);
1743		proc_stop(p, signo);
1744		KASSERT(p->p_nrlwps > 0);
1745	}
1746
1747	/*
1748	 * If we are the last live LWP, and the stop was a result of
1749	 * a new signal, then signal the parent.
1750	 */
1751	if ((p->p_sflag & PS_STOPPING) != 0) {
1752		if (!proc_lock_held && !mutex_tryenter(&proc_lock)) {
1753			mutex_exit(p->p_lock);
1754			mutex_enter(&proc_lock);
1755			mutex_enter(p->p_lock);
1756		}
1757
1758		if (p->p_nrlwps == 1 && (p->p_sflag & PS_STOPPING) != 0) {
1759			/*
1760			 * Note that proc_stop_done() can drop
1761			 * p->p_lock briefly.
1762			 */
1763			proc_stop_done(p, ppmask);
1764		}
1765
1766		mutex_exit(&proc_lock);
1767	}
1768
1769	sigswitch_unlock_and_switch_away(l);
1770}
1771
1772/*
1773 * Unlock and switch away.
1774 */
1775static void
1776sigswitch_unlock_and_switch_away(struct lwp *l)
1777{
1778	struct proc *p;
1779
1780	p = l->l_proc;
1781
1782	KASSERT(mutex_owned(p->p_lock));
1783	KASSERT(!mutex_owned(&proc_lock));
1784
1785	KASSERT(l->l_stat == LSONPROC);
1786	KASSERT(p->p_nrlwps > 0);
1787	KASSERT(l->l_blcnt == 0);
1788
1789	if (p->p_stat == SSTOP || (p->p_sflag & PS_STOPPING) != 0) {
1790		p->p_nrlwps--;
1791		lwp_lock(l);
1792		KASSERT(l->l_stat == LSONPROC || l->l_stat == LSSLEEP);
1793		l->l_stat = LSSTOP;
1794		lwp_unlock(l);
1795	}
1796
1797	mutex_exit(p->p_lock);
1798	lwp_lock(l);
1799	spc_lock(l->l_cpu);
1800	mi_switch(l);
1801}
1802
1803/*
1804 * Check for a signal from the debugger.
1805 */
1806static int
1807sigchecktrace(void)
1808{
1809	struct lwp *l = curlwp;
1810	struct proc *p = l->l_proc;
1811	int signo;
1812
1813	KASSERT(mutex_owned(p->p_lock));
1814
1815	/* If there's a pending SIGKILL, process it immediately. */
1816	if (sigismember(&p->p_sigpend.sp_set, SIGKILL))
1817		return 0;
1818
1819	/*
1820	 * If we are no longer being traced, or the parent didn't
1821	 * give us a signal, or we're stopping, look for more signals.
1822	 */
1823	if ((p->p_slflag & PSL_TRACED) == 0 || p->p_xsig == 0 ||
1824	    (p->p_sflag & PS_STOPPING) != 0)
1825		return 0;
1826
1827	/*
1828	 * If the new signal is being masked, look for other signals.
1829	 * `p->p_sigctx.ps_siglist |= mask' is done in setrunnable().
1830	 */
1831	signo = p->p_xsig;
1832	p->p_xsig = 0;
1833	if (sigismember(&l->l_sigmask, signo)) {
1834		signo = 0;
1835	}
1836	return signo;
1837}
1838
1839/*
1840 * If the current process has received a signal (should be caught or cause
1841 * termination, should interrupt current syscall), return the signal number.
1842 *
1843 * Stop signals with default action are processed immediately, then cleared;
1844 * they aren't returned.  This is checked after each entry to the system for
1845 * a syscall or trap.
1846 *
1847 * We will also return -1 if the process is exiting and the current LWP must
1848 * follow suit.
1849 */
1850int
1851issignal(struct lwp *l)
1852{
1853	struct proc *p;
1854	int siglwp, signo, prop;
1855	sigpend_t *sp;
1856	sigset_t ss;
1857	bool traced;
1858
1859	p = l->l_proc;
1860	sp = NULL;
1861	signo = 0;
1862
1863	KASSERT(p == curproc);
1864	KASSERT(mutex_owned(p->p_lock));
1865
1866	for (;;) {
1867		/* Discard any signals that we have decided not to take. */
1868		if (signo != 0) {
1869			(void)sigget(sp, NULL, signo, NULL);
1870		}
1871
1872		/*
1873		 * If the process is stopped/stopping, then stop ourselves
1874		 * now that we're on the kernel/userspace boundary.  When
1875		 * we awaken, check for a signal from the debugger.
1876		 */
1877		if (p->p_stat == SSTOP || (p->p_sflag & PS_STOPPING) != 0) {
1878			sigswitch_unlock_and_switch_away(l);
1879			mutex_enter(p->p_lock);
1880			continue;
1881		} else if (p->p_stat == SACTIVE)
1882			signo = sigchecktrace();
1883		else
1884			signo = 0;
1885
1886		/* Signals from the debugger are "out of band". */
1887		sp = NULL;
1888
1889		/*
1890		 * If the debugger didn't provide a signal, find a pending
1891		 * signal from our set.  Check per-LWP signals first, and
1892		 * then per-process.
1893		 */
1894		if (signo == 0) {
1895			sp = &l->l_sigpend;
1896			ss = sp->sp_set;
1897			siglwp = l->l_lid;
1898			if ((p->p_lflag & PL_PPWAIT) != 0)
1899				sigminusset(&vforksigmask, &ss);
1900			sigminusset(&l->l_sigmask, &ss);
1901
1902			if ((signo = firstsig(&ss)) == 0) {
1903				sp = &p->p_sigpend;
1904				ss = sp->sp_set;
1905				siglwp = 0;
1906				if ((p->p_lflag & PL_PPWAIT) != 0)
1907					sigminusset(&vforksigmask, &ss);
1908				sigminusset(&l->l_sigmask, &ss);
1909
1910				if ((signo = firstsig(&ss)) == 0) {
1911					/*
1912					 * No signal pending - clear the
1913					 * indicator and bail out.
1914					 */
1915					lwp_lock(l);
1916					l->l_flag &= ~LW_PENDSIG;
1917					lwp_unlock(l);
1918					sp = NULL;
1919					break;
1920				}
1921			}
1922		}
1923
1924		traced = ISSET(p->p_slflag, PSL_TRACED) &&
1925		    !sigismember(&p->p_sigctx.ps_sigpass, signo);
1926
1927		if (sp) {
1928			/* Overwrite process' signal context to correspond
1929			 * to the currently reported LWP.  This is necessary
1930			 * for PT_GET_SIGINFO to report the correct signal when
1931			 * multiple LWPs have pending signals.  We do this only
1932			 * when the signal comes from the queue, for signals
1933			 * created by the debugger we assume it set correct
1934			 * siginfo.
1935			 */
1936			ksiginfo_t *ksi = TAILQ_FIRST(&sp->sp_info);
1937			if (ksi) {
1938				p->p_sigctx.ps_lwp = ksi->ksi_lid;
1939				p->p_sigctx.ps_info = ksi->ksi_info;
1940			} else {
1941				p->p_sigctx.ps_lwp = siglwp;
1942				memset(&p->p_sigctx.ps_info, 0,
1943				    sizeof(p->p_sigctx.ps_info));
1944				p->p_sigctx.ps_info._signo = signo;
1945				p->p_sigctx.ps_info._code = SI_NOINFO;
1946			}
1947		}
1948
1949		/*
1950		 * We should see pending but ignored signals only if
1951		 * we are being traced.
1952		 */
1953		if (sigismember(&p->p_sigctx.ps_sigignore, signo) &&
1954		    !traced) {
1955			/* Discard the signal. */
1956			continue;
1957		}
1958
1959		/*
1960		 * If traced, always stop, and stay stopped until released
1961		 * by the debugger.  If the our parent is our debugger waiting
1962		 * for us and we vforked, don't hang as we could deadlock.
1963		 */
1964		if (traced && signo != SIGKILL &&
1965		    !(ISSET(p->p_lflag, PL_PPWAIT) &&
1966		     (p->p_pptr == p->p_opptr))) {
1967			/*
1968			 * Take the signal, but don't remove it from the
1969			 * siginfo queue, because the debugger can send
1970			 * it later.
1971			 */
1972			if (sp)
1973				sigdelset(&sp->sp_set, signo);
1974			p->p_xsig = signo;
1975
1976			/* Handling of signal trace */
1977			sigswitch(0, signo, false);
1978			mutex_enter(p->p_lock);
1979
1980			/* Check for a signal from the debugger. */
1981			if ((signo = sigchecktrace()) == 0)
1982				continue;
1983
1984			/* Signals from the debugger are "out of band". */
1985			sp = NULL;
1986		}
1987
1988		prop = sigprop[signo];
1989
1990		/*
1991		 * Decide whether the signal should be returned.
1992		 */
1993		switch ((long)SIGACTION(p, signo).sa_handler) {
1994		case (long)SIG_DFL:
1995			/*
1996			 * Don't take default actions on system processes.
1997			 */
1998			if (p->p_pid <= 1) {
1999#ifdef DIAGNOSTIC
2000				/*
2001				 * Are you sure you want to ignore SIGSEGV
2002				 * in init? XXX
2003				 */
2004				printf_nolog("Process (pid %d) got sig %d\n",
2005				    p->p_pid, signo);
2006#endif
2007				continue;
2008			}
2009
2010			/*
2011			 * If there is a pending stop signal to process with
2012			 * default action, stop here, then clear the signal.
2013			 * However, if process is member of an orphaned
2014			 * process group, ignore tty stop signals.
2015			 */
2016			if (prop & SA_STOP) {
2017				/*
2018				 * XXX Don't hold proc_lock for p_lflag,
2019				 * but it's not a big deal.
2020				 */
2021				if ((traced &&
2022				     !(ISSET(p->p_lflag, PL_PPWAIT) &&
2023				     (p->p_pptr == p->p_opptr))) ||
2024				    ((p->p_lflag & PL_ORPHANPG) != 0 &&
2025				    prop & SA_TTYSTOP)) {
2026					/* Ignore the signal. */
2027					continue;
2028				}
2029				/* Take the signal. */
2030				(void)sigget(sp, NULL, signo, NULL);
2031				p->p_xsig = signo;
2032				p->p_sflag &= ~PS_CONTINUED;
2033				signo = 0;
2034				sigswitch(PS_NOCLDSTOP, p->p_xsig, false);
2035				mutex_enter(p->p_lock);
2036			} else if (prop & SA_IGNORE) {
2037				/*
2038				 * Except for SIGCONT, shouldn't get here.
2039				 * Default action is to ignore; drop it.
2040				 */
2041				continue;
2042			}
2043			break;
2044
2045		case (long)SIG_IGN:
2046#ifdef DEBUG_ISSIGNAL
2047			/*
2048			 * Masking above should prevent us ever trying
2049			 * to take action on an ignored signal other
2050			 * than SIGCONT, unless process is traced.
2051			 */
2052			if ((prop & SA_CONT) == 0 && !traced)
2053				printf_nolog("issignal\n");
2054#endif
2055			continue;
2056
2057		default:
2058			/*
2059			 * This signal has an action, let postsig() process
2060			 * it.
2061			 */
2062			break;
2063		}
2064
2065		break;
2066	}
2067
2068	l->l_sigpendset = sp;
2069	return signo;
2070}
2071
2072/*
2073 * Take the action for the specified signal
2074 * from the current set of pending signals.
2075 */
2076void
2077postsig(int signo)
2078{
2079	struct lwp	*l;
2080	struct proc	*p;
2081	struct sigacts	*ps;
2082	sig_t		action;
2083	sigset_t	*returnmask;
2084	ksiginfo_t	ksi;
2085
2086	l = curlwp;
2087	p = l->l_proc;
2088	ps = p->p_sigacts;
2089
2090	KASSERT(mutex_owned(p->p_lock));
2091	KASSERT(signo > 0);
2092
2093	/*
2094	 * Set the new mask value and also defer further occurrences of this
2095	 * signal.
2096	 *
2097	 * Special case: user has done a sigsuspend.  Here the current mask is
2098	 * not of interest, but rather the mask from before the sigsuspend is
2099	 * what we want restored after the signal processing is completed.
2100	 */
2101	if (l->l_sigrestore) {
2102		returnmask = &l->l_sigoldmask;
2103		l->l_sigrestore = 0;
2104	} else
2105		returnmask = &l->l_sigmask;
2106
2107	/*
2108	 * Commit to taking the signal before releasing the mutex.
2109	 */
2110	action = SIGACTION_PS(ps, signo).sa_handler;
2111	l->l_ru.ru_nsignals++;
2112	if (l->l_sigpendset == NULL) {
2113		/* From the debugger */
2114		if (p->p_sigctx.ps_faked &&
2115		    signo == p->p_sigctx.ps_info._signo) {
2116			KSI_INIT(&ksi);
2117			ksi.ksi_info = p->p_sigctx.ps_info;
2118			ksi.ksi_lid = p->p_sigctx.ps_lwp;
2119			p->p_sigctx.ps_faked = false;
2120		} else {
2121			if (!siggetinfo(&l->l_sigpend, &ksi, signo))
2122				(void)siggetinfo(&p->p_sigpend, &ksi, signo);
2123		}
2124	} else
2125		sigget(l->l_sigpendset, &ksi, signo, NULL);
2126
2127	if (ktrpoint(KTR_PSIG)) {
2128		mutex_exit(p->p_lock);
2129		if (p->p_emul->e_ktrpsig)
2130			p->p_emul->e_ktrpsig(signo, action,
2131			    returnmask, &ksi);
2132		else
2133			ktrpsig(signo, action, returnmask, &ksi);
2134		mutex_enter(p->p_lock);
2135	}
2136
2137	SDT_PROBE(proc, kernel, , signal__handle, signo, &ksi, action, 0, 0);
2138
2139	if (action == SIG_DFL) {
2140		/*
2141		 * Default action, where the default is to kill
2142		 * the process.  (Other cases were ignored above.)
2143		 */
2144		sigexit(l, signo);
2145		return;
2146	}
2147
2148	/*
2149	 * If we get here, the signal must be caught.
2150	 */
2151#ifdef DIAGNOSTIC
2152	if (action == SIG_IGN || sigismember(&l->l_sigmask, signo))
2153		panic("postsig action");
2154#endif
2155
2156	kpsendsig(l, &ksi, returnmask);
2157}
2158
2159/*
2160 * sendsig:
2161 *
2162 *	Default signal delivery method for NetBSD.
2163 */
2164void
2165sendsig(const struct ksiginfo *ksi, const sigset_t *mask)
2166{
2167	struct sigacts *sa;
2168	int sig;
2169
2170	sig = ksi->ksi_signo;
2171	sa = curproc->p_sigacts;
2172
2173	switch (sa->sa_sigdesc[sig].sd_vers)  {
2174	case __SIGTRAMP_SIGCODE_VERSION:
2175#ifdef __HAVE_STRUCT_SIGCONTEXT
2176	case __SIGTRAMP_SIGCONTEXT_VERSION_MIN ...
2177	     __SIGTRAMP_SIGCONTEXT_VERSION_MAX:
2178		/* Compat for 1.6 and earlier. */
2179		MODULE_HOOK_CALL_VOID(sendsig_sigcontext_16_hook, (ksi, mask),
2180		    break);
2181		return;
2182#endif /* __HAVE_STRUCT_SIGCONTEXT */
2183	case __SIGTRAMP_SIGINFO_VERSION_MIN ...
2184	     __SIGTRAMP_SIGINFO_VERSION_MAX:
2185		sendsig_siginfo(ksi, mask);
2186		return;
2187	default:
2188		break;
2189	}
2190
2191	printf("sendsig: bad version %d\n", sa->sa_sigdesc[sig].sd_vers);
2192	sigexit(curlwp, SIGILL);
2193}
2194
2195/*
2196 * sendsig_reset:
2197 *
2198 *	Reset the signal action.  Called from emulation specific sendsig()
2199 *	before unlocking to deliver the signal.
2200 */
2201void
2202sendsig_reset(struct lwp *l, int signo)
2203{
2204	struct proc *p = l->l_proc;
2205	struct sigacts *ps = p->p_sigacts;
2206
2207	KASSERT(mutex_owned(p->p_lock));
2208
2209	p->p_sigctx.ps_lwp = 0;
2210	memset(&p->p_sigctx.ps_info, 0, sizeof(p->p_sigctx.ps_info));
2211
2212	mutex_enter(&ps->sa_mutex);
2213	sigplusset(&SIGACTION_PS(ps, signo).sa_mask, &l->l_sigmask);
2214	if (SIGACTION_PS(ps, signo).sa_flags & SA_RESETHAND) {
2215		sigdelset(&p->p_sigctx.ps_sigcatch, signo);
2216		if (signo != SIGCONT && sigprop[signo] & SA_IGNORE)
2217			sigaddset(&p->p_sigctx.ps_sigignore, signo);
2218		SIGACTION_PS(ps, signo).sa_handler = SIG_DFL;
2219	}
2220	mutex_exit(&ps->sa_mutex);
2221}
2222
2223/*
2224 * Kill the current process for stated reason.
2225 */
2226void
2227killproc(struct proc *p, const char *why)
2228{
2229
2230	KASSERT(mutex_owned(&proc_lock));
2231
2232	log(LOG_ERR, "pid %d was killed: %s\n", p->p_pid, why);
2233	uprintf_locked("sorry, pid %d was killed: %s\n", p->p_pid, why);
2234	psignal(p, SIGKILL);
2235}
2236
2237/*
2238 * Force the current process to exit with the specified signal, dumping core
2239 * if appropriate.  We bypass the normal tests for masked and caught
2240 * signals, allowing unrecoverable failures to terminate the process without
2241 * changing signal state.  Mark the accounting record with the signal
2242 * termination.  If dumping core, save the signal number for the debugger.
2243 * Calls exit and does not return.
2244 */
2245void
2246sigexit(struct lwp *l, int signo)
2247{
2248	int exitsig, error, docore;
2249	struct proc *p;
2250	struct lwp *t;
2251
2252	p = l->l_proc;
2253
2254	KASSERT(mutex_owned(p->p_lock));
2255	KASSERT(l->l_blcnt == 0);
2256
2257	/*
2258	 * Don't permit coredump() multiple times in the same process.
2259	 * Call back into sigexit, where we will be suspended until
2260	 * the deed is done.  Note that this is a recursive call, but
2261	 * LW_WCORE will prevent us from coming back this way.
2262	 */
2263	if ((p->p_sflag & PS_WCORE) != 0) {
2264		lwp_lock(l);
2265		l->l_flag |= (LW_WCORE | LW_WEXIT | LW_WSUSPEND);
2266		lwp_need_userret(l);
2267		lwp_unlock(l);
2268		mutex_exit(p->p_lock);
2269		lwp_userret(l);
2270		panic("sigexit 1");
2271		/* NOTREACHED */
2272	}
2273
2274	/* If process is already on the way out, then bail now. */
2275	if ((p->p_sflag & PS_WEXIT) != 0) {
2276		mutex_exit(p->p_lock);
2277		lwp_exit(l);
2278		panic("sigexit 2");
2279		/* NOTREACHED */
2280	}
2281
2282	/*
2283	 * Prepare all other LWPs for exit.  If dumping core, suspend them
2284	 * so that their registers are available long enough to be dumped.
2285 	 */
2286	if ((docore = (sigprop[signo] & SA_CORE)) != 0) {
2287		p->p_sflag |= PS_WCORE;
2288		for (;;) {
2289			LIST_FOREACH(t, &p->p_lwps, l_sibling) {
2290				lwp_lock(t);
2291				if (t == l) {
2292					t->l_flag &=
2293					    ~(LW_WSUSPEND | LW_DBGSUSPEND);
2294					lwp_unlock(t);
2295					continue;
2296				}
2297				t->l_flag |= (LW_WCORE | LW_WEXIT);
2298				lwp_need_userret(t);
2299				lwp_suspend(l, t);
2300			}
2301
2302			if (p->p_nrlwps == 1)
2303				break;
2304
2305			/*
2306			 * Kick any LWPs sitting in lwp_wait1(), and wait
2307			 * for everyone else to stop before proceeding.
2308			 */
2309			p->p_nlwpwait++;
2310			cv_broadcast(&p->p_lwpcv);
2311			cv_wait(&p->p_lwpcv, p->p_lock);
2312			p->p_nlwpwait--;
2313		}
2314	}
2315
2316	exitsig = signo;
2317	p->p_acflag |= AXSIG;
2318	memset(&p->p_sigctx.ps_info, 0, sizeof(p->p_sigctx.ps_info));
2319	p->p_sigctx.ps_info._signo = signo;
2320	p->p_sigctx.ps_info._code = SI_NOINFO;
2321
2322	if (docore) {
2323		mutex_exit(p->p_lock);
2324		MODULE_HOOK_CALL(coredump_hook, (l, NULL), enosys(), error);
2325
2326		if (kern_logsigexit) {
2327			int uid = l->l_cred ?
2328			    (int)kauth_cred_geteuid(l->l_cred) : -1;
2329
2330			if (error)
2331				log(LOG_INFO, lognocoredump, p->p_pid,
2332				    p->p_comm, uid, signo, error);
2333			else
2334				log(LOG_INFO, logcoredump, p->p_pid,
2335				    p->p_comm, uid, signo);
2336		}
2337
2338#ifdef PAX_SEGVGUARD
2339		rw_enter(&exec_lock, RW_WRITER);
2340		pax_segvguard(l, p->p_textvp, p->p_comm, true);
2341		rw_exit(&exec_lock);
2342#endif /* PAX_SEGVGUARD */
2343
2344		/* Acquire the sched state mutex.  exit1() will release it. */
2345		mutex_enter(p->p_lock);
2346		if (error == 0)
2347			p->p_sflag |= PS_COREDUMP;
2348	}
2349
2350	/* No longer dumping core. */
2351	p->p_sflag &= ~PS_WCORE;
2352
2353	exit1(l, 0, exitsig);
2354	/* NOTREACHED */
2355}
2356
2357/*
2358 * Since the "real" code may (or may not) be present in loadable module,
2359 * we provide routines here which calls the module hooks.
2360 */
2361
2362int
2363coredump_netbsd(struct lwp *l, struct coredump_iostate *iocookie)
2364{
2365
2366	int retval;
2367
2368	MODULE_HOOK_CALL(coredump_netbsd_hook, (l, iocookie), ENOSYS, retval);
2369	return retval;
2370}
2371
2372int
2373coredump_netbsd32(struct lwp *l, struct coredump_iostate *iocookie)
2374{
2375
2376	int retval;
2377
2378	MODULE_HOOK_CALL(coredump_netbsd32_hook, (l, iocookie), ENOSYS, retval);
2379	return retval;
2380}
2381
2382int
2383coredump_elf32(struct lwp *l, struct coredump_iostate *iocookie)
2384{
2385	int retval;
2386
2387	MODULE_HOOK_CALL(coredump_elf32_hook, (l, iocookie), ENOSYS, retval);
2388	return retval;
2389}
2390
2391int
2392coredump_elf64(struct lwp *l, struct coredump_iostate *iocookie)
2393{
2394	int retval;
2395
2396	MODULE_HOOK_CALL(coredump_elf64_hook, (l, iocookie), ENOSYS, retval);
2397	return retval;
2398}
2399
2400/*
2401 * Put process 'p' into the stopped state and optionally, notify the parent.
2402 */
2403void
2404proc_stop(struct proc *p, int signo)
2405{
2406	struct lwp *l;
2407
2408	KASSERT(mutex_owned(p->p_lock));
2409
2410	/*
2411	 * First off, set the stopping indicator and bring all sleeping
2412	 * LWPs to a halt so they are included in p->p_nrlwps.  We mustn't
2413	 * unlock between here and the p->p_nrlwps check below.
2414	 */
2415	p->p_sflag |= PS_STOPPING;
2416	membar_producer();
2417
2418	proc_stop_lwps(p);
2419
2420	/*
2421	 * If there are no LWPs available to take the signal, then we
2422	 * signal the parent process immediately.  Otherwise, the last
2423	 * LWP to stop will take care of it.
2424	 */
2425
2426	if (p->p_nrlwps == 0) {
2427		proc_stop_done(p, PS_NOCLDSTOP);
2428	} else {
2429		/*
2430		 * Have the remaining LWPs come to a halt, and trigger
2431		 * proc_stop_callout() to ensure that they do.
2432		 */
2433		LIST_FOREACH(l, &p->p_lwps, l_sibling) {
2434			sigpost(l, SIG_DFL, SA_STOP, signo);
2435		}
2436		callout_schedule(&proc_stop_ch, 1);
2437	}
2438}
2439
2440/*
2441 * When stopping a process, we do not immediately set sleeping LWPs stopped,
2442 * but wait for them to come to a halt at the kernel-user boundary.  This is
2443 * to allow LWPs to release any locks that they may hold before stopping.
2444 *
2445 * Non-interruptable sleeps can be long, and there is the potential for an
2446 * LWP to begin sleeping interruptably soon after the process has been set
2447 * stopping (PS_STOPPING).  These LWPs will not notice that the process is
2448 * stopping, and so complete halt of the process and the return of status
2449 * information to the parent could be delayed indefinitely.
2450 *
2451 * To handle this race, proc_stop_callout() runs once per tick while there
2452 * are stopping processes in the system.  It sets LWPs that are sleeping
2453 * interruptably into the LSSTOP state.
2454 *
2455 * Note that we are not concerned about keeping all LWPs stopped while the
2456 * process is stopped: stopped LWPs can awaken briefly to handle signals.
2457 * What we do need to ensure is that all LWPs in a stopping process have
2458 * stopped at least once, so that notification can be sent to the parent
2459 * process.
2460 */
2461static void
2462proc_stop_callout(void *cookie)
2463{
2464	bool more, restart;
2465	struct proc *p;
2466
2467	(void)cookie;
2468
2469	do {
2470		restart = false;
2471		more = false;
2472
2473		mutex_enter(&proc_lock);
2474		PROCLIST_FOREACH(p, &allproc) {
2475			mutex_enter(p->p_lock);
2476
2477			if ((p->p_sflag & PS_STOPPING) == 0) {
2478				mutex_exit(p->p_lock);
2479				continue;
2480			}
2481
2482			/* Stop any LWPs sleeping interruptably. */
2483			proc_stop_lwps(p);
2484			if (p->p_nrlwps == 0) {
2485				/*
2486				 * We brought the process to a halt.
2487				 * Mark it as stopped and notify the
2488				 * parent.
2489				 *
2490				 * Note that proc_stop_done() will
2491				 * drop p->p_lock briefly.
2492				 * Arrange to restart and check
2493				 * all processes again.
2494				 */
2495				restart = true;
2496				proc_stop_done(p, PS_NOCLDSTOP);
2497			} else
2498				more = true;
2499
2500			mutex_exit(p->p_lock);
2501			if (restart)
2502				break;
2503		}
2504		mutex_exit(&proc_lock);
2505	} while (restart);
2506
2507	/*
2508	 * If we noted processes that are stopping but still have
2509	 * running LWPs, then arrange to check again in 1 tick.
2510	 */
2511	if (more)
2512		callout_schedule(&proc_stop_ch, 1);
2513}
2514
2515/*
2516 * Given a process in state SSTOP, set the state back to SACTIVE and
2517 * move LSSTOP'd LWPs to LSSLEEP or make them runnable.
2518 */
2519void
2520proc_unstop(struct proc *p)
2521{
2522	struct lwp *l;
2523	int sig;
2524
2525	KASSERT(mutex_owned(&proc_lock));
2526	KASSERT(mutex_owned(p->p_lock));
2527
2528	p->p_stat = SACTIVE;
2529	p->p_sflag &= ~PS_STOPPING;
2530	sig = p->p_xsig;
2531
2532	if (!p->p_waited)
2533		p->p_pptr->p_nstopchild--;
2534
2535	LIST_FOREACH(l, &p->p_lwps, l_sibling) {
2536		lwp_lock(l);
2537		if (l->l_stat != LSSTOP || (l->l_flag & LW_DBGSUSPEND) != 0) {
2538			lwp_unlock(l);
2539			continue;
2540		}
2541		if (l->l_wchan == NULL) {
2542			setrunnable(l);
2543			continue;
2544		}
2545		if (sig && (l->l_flag & LW_SINTR) != 0) {
2546			setrunnable(l);
2547			sig = 0;
2548		} else {
2549			l->l_stat = LSSLEEP;
2550			p->p_nrlwps++;
2551			lwp_unlock(l);
2552		}
2553	}
2554}
2555
2556void
2557proc_stoptrace(int trapno, int sysnum, const register_t args[],
2558               const register_t *ret, int error)
2559{
2560	struct lwp *l = curlwp;
2561	struct proc *p = l->l_proc;
2562	struct sigacts *ps;
2563	sigset_t *mask;
2564	sig_t action;
2565	ksiginfo_t ksi;
2566	size_t i, sy_narg;
2567	const int signo = SIGTRAP;
2568
2569	KASSERT((trapno == TRAP_SCE) || (trapno == TRAP_SCX));
2570	KASSERT(p->p_pptr != initproc);
2571	KASSERT(ISSET(p->p_slflag, PSL_TRACED));
2572	KASSERT(ISSET(p->p_slflag, PSL_SYSCALL));
2573
2574	sy_narg = p->p_emul->e_sysent[sysnum].sy_narg;
2575
2576	KSI_INIT_TRAP(&ksi);
2577	ksi.ksi_lid = l->l_lid;
2578	ksi.ksi_signo = signo;
2579	ksi.ksi_code = trapno;
2580
2581	ksi.ksi_sysnum = sysnum;
2582	if (trapno == TRAP_SCE) {
2583		ksi.ksi_retval[0] = 0;
2584		ksi.ksi_retval[1] = 0;
2585		ksi.ksi_error = 0;
2586	} else {
2587		ksi.ksi_retval[0] = ret[0];
2588		ksi.ksi_retval[1] = ret[1];
2589		ksi.ksi_error = error;
2590	}
2591
2592	memset(ksi.ksi_args, 0, sizeof(ksi.ksi_args));
2593
2594	for (i = 0; i < sy_narg; i++)
2595		ksi.ksi_args[i] = args[i];
2596
2597	mutex_enter(p->p_lock);
2598
2599repeat:
2600	/*
2601	 * If we are exiting, demise now.
2602	 *
2603	 * This avoids notifying tracer and deadlocking.
2604	 */
2605	if (__predict_false(ISSET(p->p_sflag, PS_WEXIT))) {
2606		mutex_exit(p->p_lock);
2607		lwp_exit(l);
2608		panic("proc_stoptrace");
2609		/* NOTREACHED */
2610	}
2611
2612	/*
2613	 * If there's a pending SIGKILL process it immediately.
2614	 */
2615	if (p->p_xsig == SIGKILL ||
2616	    sigismember(&p->p_sigpend.sp_set, SIGKILL)) {
2617		mutex_exit(p->p_lock);
2618		return;
2619	}
2620
2621	/*
2622	 * If we are no longer traced, abandon this event signal.
2623	 *
2624	 * This avoids killing a process after detaching the debugger.
2625	 */
2626	if (__predict_false(!ISSET(p->p_slflag, PSL_TRACED))) {
2627		mutex_exit(p->p_lock);
2628		return;
2629	}
2630
2631	/*
2632	 * The process is already stopping.
2633	 */
2634	if ((p->p_sflag & PS_STOPPING) != 0) {
2635		sigswitch_unlock_and_switch_away(l);
2636		mutex_enter(p->p_lock);
2637		goto repeat;
2638	}
2639
2640	/* Needed for ktrace */
2641	ps = p->p_sigacts;
2642	action = SIGACTION_PS(ps, signo).sa_handler;
2643	mask = &l->l_sigmask;
2644
2645	p->p_xsig = signo;
2646	p->p_sigctx.ps_lwp = ksi.ksi_lid;
2647	p->p_sigctx.ps_info = ksi.ksi_info;
2648	sigswitch(0, signo, false);
2649
2650	if (ktrpoint(KTR_PSIG)) {
2651		if (p->p_emul->e_ktrpsig)
2652			p->p_emul->e_ktrpsig(signo, action, mask, &ksi);
2653		else
2654			ktrpsig(signo, action, mask, &ksi);
2655	}
2656}
2657
2658static int
2659filt_sigattach(struct knote *kn)
2660{
2661	struct proc *p = curproc;
2662
2663	kn->kn_obj = p;
2664	kn->kn_flags |= EV_CLEAR;	/* automatically set */
2665
2666	mutex_enter(p->p_lock);
2667	klist_insert(&p->p_klist, kn);
2668	mutex_exit(p->p_lock);
2669
2670	return 0;
2671}
2672
2673static void
2674filt_sigdetach(struct knote *kn)
2675{
2676	struct proc *p = kn->kn_obj;
2677
2678	mutex_enter(p->p_lock);
2679	klist_remove(&p->p_klist, kn);
2680	mutex_exit(p->p_lock);
2681}
2682
2683/*
2684 * Signal knotes are shared with proc knotes, so we apply a mask to
2685 * the hint in order to differentiate them from process hints.  This
2686 * could be avoided by using a signal-specific knote list, but probably
2687 * isn't worth the trouble.
2688 */
2689static int
2690filt_signal(struct knote *kn, long hint)
2691{
2692
2693	if (hint & NOTE_SIGNAL) {
2694		hint &= ~NOTE_SIGNAL;
2695
2696		if (kn->kn_id == hint)
2697			kn->kn_data++;
2698	}
2699	return (kn->kn_data != 0);
2700}
2701
2702const struct filterops sig_filtops = {
2703	.f_flags = FILTEROP_MPSAFE,
2704	.f_attach = filt_sigattach,
2705	.f_detach = filt_sigdetach,
2706	.f_event = filt_signal,
2707};
2708