kern_sig.c revision 1.411
1/*	$NetBSD: kern_sig.c,v 1.411 2025/12/10 21:33:01 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 1.411 2025/12/10 21:33:01 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	proc_stop(struct proc *, int);
127static void	proc_stop_done(struct proc *, int);
128static void	proc_stop_callout(void *);
129static int	sigchecktrace(void);
130static int	sigpost(struct lwp *, sig_t, int, int);
131static int	sigput(sigpend_t *, struct proc *, ksiginfo_t *);
132static int	sigunwait(struct proc *, const ksiginfo_t *);
133static void	sigswitch(int, int, bool);
134static void	sigswitch_unlock_and_switch_away(struct lwp *);
135
136static void	sigacts_poolpage_free(struct pool *, void *);
137static void	*sigacts_poolpage_alloc(struct pool *, int);
138
139/*
140 * DTrace SDT provider definitions
141 */
142SDT_PROVIDER_DECLARE(proc);
143SDT_PROBE_DEFINE3(proc, kernel, , signal__send,
144    "struct lwp *", 	/* target thread */
145    "struct proc *", 	/* target process */
146    "int");		/* signal */
147SDT_PROBE_DEFINE3(proc, kernel, , signal__discard,
148    "struct lwp *",	/* target thread */
149    "struct proc *",	/* target process */
150    "int");  		/* signal */
151SDT_PROBE_DEFINE3(proc, kernel, , signal__handle,
152    "int", 		/* signal */
153    "ksiginfo_t *", 	/* signal info */
154    "void (*)(void)");	/* handler address */
155
156
157static struct pool_allocator sigactspool_allocator = {
158	.pa_alloc = sigacts_poolpage_alloc,
159	.pa_free = sigacts_poolpage_free
160};
161
162#ifdef DEBUG
163int	kern_logsigexit = 1;
164#else
165int	kern_logsigexit = 0;
166#endif
167
168static const char logcoredump[] =
169    "pid %d (%s), uid %d: exited on signal %d (core dumped)\n";
170static const char lognocoredump[] =
171    "pid %d (%s), uid %d: exited on signal %d (core not dumped, err = %d)\n";
172
173static kauth_listener_t signal_listener;
174
175static int
176signal_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
177    void *arg0, void *arg1, void *arg2, void *arg3)
178{
179	struct proc *p;
180	int result, signum;
181
182	result = KAUTH_RESULT_DEFER;
183	p = arg0;
184	signum = (int)(unsigned long)arg1;
185
186	if (action != KAUTH_PROCESS_SIGNAL)
187		return result;
188
189	if (kauth_cred_uidmatch(cred, p->p_cred) ||
190	    (signum == SIGCONT && (curproc->p_session == p->p_session)))
191		result = KAUTH_RESULT_ALLOW;
192
193	return result;
194}
195
196static int
197sigacts_ctor(void *arg __unused, void *obj, int flags __unused)
198{
199	memset(obj, 0, sizeof(struct sigacts));
200	return 0;
201}
202
203/*
204 * signal_init:
205 *
206 *	Initialize global signal-related data structures.
207 */
208void
209signal_init(void)
210{
211
212	sigactspool_allocator.pa_pagesz = (PAGE_SIZE)*2;
213
214	sigacts_cache = pool_cache_init(sizeof(struct sigacts), 0, 0, 0,
215	    "sigacts", sizeof(struct sigacts) > PAGE_SIZE ?
216	    &sigactspool_allocator : NULL, IPL_NONE, sigacts_ctor, NULL, NULL);
217	ksiginfo_cache = pool_cache_init(sizeof(ksiginfo_t), 0, 0, 0,
218	    "ksiginfo", NULL, IPL_VM, NULL, NULL, NULL);
219
220	callout_init(&proc_stop_ch, CALLOUT_MPSAFE);
221	callout_setfunc(&proc_stop_ch, proc_stop_callout, NULL);
222
223	signal_listener = kauth_listen_scope(KAUTH_SCOPE_PROCESS,
224	    signal_listener_cb, NULL);
225}
226
227/*
228 * sigacts_poolpage_alloc:
229 *
230 *	Allocate a page for the sigacts memory pool.
231 */
232static void *
233sigacts_poolpage_alloc(struct pool *pp, int flags)
234{
235
236	return (void *)uvm_km_alloc(kernel_map,
237	    PAGE_SIZE * 2, PAGE_SIZE * 2,
238	    ((flags & PR_WAITOK) ? 0 : UVM_KMF_NOWAIT | UVM_KMF_TRYLOCK)
239	    | UVM_KMF_WIRED);
240}
241
242/*
243 * sigacts_poolpage_free:
244 *
245 *	Free a page on behalf of the sigacts memory pool.
246 */
247static void
248sigacts_poolpage_free(struct pool *pp, void *v)
249{
250
251	uvm_km_free(kernel_map, (vaddr_t)v, PAGE_SIZE * 2, UVM_KMF_WIRED);
252}
253
254/*
255 * sigactsinit:
256 *
257 *	Create an initial sigacts structure, using the same signal state
258 *	as of specified process.  If 'share' is set, share the sigacts by
259 *	holding a reference, otherwise just copy it from parent.
260 */
261struct sigacts *
262sigactsinit(struct proc *pp, int share)
263{
264	struct sigacts *ps = pp->p_sigacts, *ps2;
265
266	if (__predict_false(share)) {
267		atomic_inc_uint(&ps->sa_refcnt);
268		return ps;
269	}
270	ps2 = pool_cache_get(sigacts_cache, PR_WAITOK);
271	mutex_init(&ps2->sa_mutex, MUTEX_DEFAULT, IPL_SCHED);
272	ps2->sa_refcnt = 1;
273
274	mutex_enter(&ps->sa_mutex);
275	memcpy(ps2->sa_sigdesc, ps->sa_sigdesc, sizeof(ps2->sa_sigdesc));
276	mutex_exit(&ps->sa_mutex);
277	return ps2;
278}
279
280/*
281 * sigactsunshare:
282 *
283 *	Make this process not share its sigacts, maintaining all signal state.
284 */
285void
286sigactsunshare(struct proc *p)
287{
288	struct sigacts *ps, *oldps = p->p_sigacts;
289
290	if (__predict_true(oldps->sa_refcnt == 1))
291		return;
292
293	ps = pool_cache_get(sigacts_cache, PR_WAITOK);
294	mutex_init(&ps->sa_mutex, MUTEX_DEFAULT, IPL_SCHED);
295	memcpy(ps->sa_sigdesc, oldps->sa_sigdesc, sizeof(ps->sa_sigdesc));
296	ps->sa_refcnt = 1;
297
298	p->p_sigacts = ps;
299	sigactsfree(oldps);
300}
301
302/*
303 * sigactsfree;
304 *
305 *	Release a sigacts structure.
306 */
307void
308sigactsfree(struct sigacts *ps)
309{
310
311	membar_release();
312	if (atomic_dec_uint_nv(&ps->sa_refcnt) == 0) {
313		membar_acquire();
314		mutex_destroy(&ps->sa_mutex);
315		pool_cache_put(sigacts_cache, ps);
316	}
317}
318
319/*
320 * siginit:
321 *
322 *	Initialize signal state for process 0; set to ignore signals that
323 *	are ignored by default and disable the signal stack.  Locking not
324 *	required as the system is still cold.
325 */
326void
327siginit(struct proc *p)
328{
329	struct lwp *l;
330	struct sigacts *ps;
331	int signo, prop;
332
333	ps = p->p_sigacts;
334	sigemptyset(&contsigmask);
335	sigemptyset(&stopsigmask);
336	sigemptyset(&vforksigmask);
337	sigemptyset(&sigcantmask);
338	for (signo = 1; signo < NSIG; signo++) {
339		prop = sigprop[signo];
340		if (prop & SA_CONT)
341			sigaddset(&contsigmask, signo);
342		if (prop & SA_STOP)
343			sigaddset(&stopsigmask, signo);
344		if (prop & SA_STOP && signo != SIGSTOP)
345			sigaddset(&vforksigmask, signo);
346		if (prop & SA_CANTMASK)
347			sigaddset(&sigcantmask, signo);
348		if (prop & SA_IGNORE && signo != SIGCONT)
349			sigaddset(&p->p_sigctx.ps_sigignore, signo);
350		sigemptyset(&SIGACTION_PS(ps, signo).sa_mask);
351		SIGACTION_PS(ps, signo).sa_flags = SA_RESTART;
352	}
353	sigemptyset(&p->p_sigctx.ps_sigcatch);
354	p->p_sflag &= ~PS_NOCLDSTOP;
355
356	ksiginfo_queue_init(&p->p_sigpend.sp_info);
357	sigemptyset(&p->p_sigpend.sp_set);
358
359	/*
360	 * Reset per LWP state.
361	 */
362	l = LIST_FIRST(&p->p_lwps);
363	l->l_sigwaited = NULL;
364	l->l_sigstk = SS_INIT;
365	ksiginfo_queue_init(&l->l_sigpend.sp_info);
366	sigemptyset(&l->l_sigpend.sp_set);
367
368	/* One reference. */
369	ps->sa_refcnt = 1;
370}
371
372/*
373 * execsigs:
374 *
375 *	Reset signals for an exec of the specified process.
376 */
377void
378execsigs(struct proc *p)
379{
380	struct sigacts *ps;
381	struct lwp *l;
382	int signo, prop;
383	sigset_t tset;
384	ksiginfoq_t kq;
385
386	KASSERT(p->p_nlwps == 1);
387
388	sigactsunshare(p);
389	ps = p->p_sigacts;
390
391	/*
392	 * Reset caught signals.  Held signals remain held through
393	 * l->l_sigmask (unless they were caught, and are now ignored
394	 * by default).
395	 *
396	 * No need to lock yet, the process has only one LWP and
397	 * at this point the sigacts are private to the process.
398	 */
399	sigemptyset(&tset);
400	for (signo = 1; signo < NSIG; signo++) {
401		if (sigismember(&p->p_sigctx.ps_sigcatch, signo)) {
402			prop = sigprop[signo];
403			if (prop & SA_IGNORE) {
404				if ((prop & SA_CONT) == 0)
405					sigaddset(&p->p_sigctx.ps_sigignore,
406					    signo);
407				sigaddset(&tset, signo);
408			}
409			SIGACTION_PS(ps, signo).sa_handler = SIG_DFL;
410		}
411		sigemptyset(&SIGACTION_PS(ps, signo).sa_mask);
412		SIGACTION_PS(ps, signo).sa_flags = SA_RESTART;
413	}
414	ksiginfo_queue_init(&kq);
415
416	mutex_enter(p->p_lock);
417	sigclearall(p, &tset, &kq);
418	sigemptyset(&p->p_sigctx.ps_sigcatch);
419
420	/*
421	 * Reset no zombies if child dies flag as Solaris does.
422	 */
423	p->p_flag &= ~(PK_NOCLDWAIT | PK_CLDSIGIGN);
424	if (SIGACTION_PS(ps, SIGCHLD).sa_handler == SIG_IGN)
425		SIGACTION_PS(ps, SIGCHLD).sa_handler = SIG_DFL;
426
427	/*
428	 * Reset per-LWP state.
429	 */
430	l = LIST_FIRST(&p->p_lwps);
431	l->l_sigwaited = NULL;
432	l->l_sigstk = SS_INIT;
433	ksiginfo_queue_init(&l->l_sigpend.sp_info);
434	sigemptyset(&l->l_sigpend.sp_set);
435	mutex_exit(p->p_lock);
436
437	ksiginfo_queue_drain(&kq);
438}
439
440/*
441 * ksiginfo_alloc:
442 *
443 *	Allocate a new ksiginfo structure from the pool, and optionally copy
444 *	an existing one.  If the existing ksiginfo_t is from the pool, and
445 *	has not been queued somewhere, then just return it.  Additionally,
446 *	if the existing ksiginfo_t does not contain any information beyond
447 *	the signal number, then just return it.
448 */
449ksiginfo_t *
450ksiginfo_alloc(struct proc *p, ksiginfo_t *ok, int flags)
451{
452	ksiginfo_t *kp;
453
454	if (ok != NULL) {
455		if ((ok->ksi_flags & (KSI_QUEUED | KSI_FROMPOOL)) ==
456		    KSI_FROMPOOL)
457			return ok;
458		if (KSI_EMPTY_P(ok))
459			return ok;
460	}
461
462	kp = pool_cache_get(ksiginfo_cache, flags);
463	if (kp == NULL) {
464#ifdef DIAGNOSTIC
465		printf("Out of memory allocating ksiginfo for pid %d\n",
466		    p->p_pid);
467#endif
468		return NULL;
469	}
470
471	if (ok != NULL) {
472		memcpy(kp, ok, sizeof(*kp));
473		kp->ksi_flags &= ~KSI_QUEUED;
474	} else
475		KSI_INIT_EMPTY(kp);
476
477	kp->ksi_flags |= KSI_FROMPOOL;
478
479	return kp;
480}
481
482/*
483 * ksiginfo_free:
484 *
485 *	If the given ksiginfo_t is from the pool and has not been queued,
486 *	then free it.
487 */
488void
489ksiginfo_free(ksiginfo_t *kp)
490{
491
492	if ((kp->ksi_flags & (KSI_QUEUED | KSI_FROMPOOL)) != KSI_FROMPOOL)
493		return;
494	pool_cache_put(ksiginfo_cache, kp);
495}
496
497/*
498 * ksiginfo_queue_drain:
499 *
500 *	Drain a non-empty ksiginfo_t queue.
501 */
502void
503ksiginfo_queue_drain0(ksiginfoq_t *kq)
504{
505	ksiginfo_t *ksi;
506
507	KASSERT(!TAILQ_EMPTY(kq));
508
509	while (!TAILQ_EMPTY(kq)) {
510		ksi = TAILQ_FIRST(kq);
511		TAILQ_REMOVE(kq, ksi, ksi_list);
512		pool_cache_put(ksiginfo_cache, ksi);
513	}
514}
515
516static int
517siggetinfo(sigpend_t *sp, ksiginfo_t *out, int signo)
518{
519	ksiginfo_t *ksi, *nksi;
520
521	if (sp == NULL)
522		goto out;
523
524	/* Find siginfo and copy it out. */
525	int count = 0;
526	TAILQ_FOREACH_SAFE(ksi, &sp->sp_info, ksi_list, nksi) {
527		if (ksi->ksi_signo != signo)
528			continue;
529		if (count++ > 0) /* Only remove the first, count all of them */
530			continue;
531		TAILQ_REMOVE(&sp->sp_info, ksi, ksi_list);
532		KASSERT((ksi->ksi_flags & KSI_FROMPOOL) != 0);
533		KASSERT((ksi->ksi_flags & KSI_QUEUED) != 0);
534		ksi->ksi_flags &= ~KSI_QUEUED;
535		if (out != NULL) {
536			memcpy(out, ksi, sizeof(*out));
537			out->ksi_flags &= ~(KSI_FROMPOOL | KSI_QUEUED);
538		}
539		ksiginfo_free(ksi);
540	}
541	if (count)
542		return count;
543
544out:
545	/* If there is no siginfo, then manufacture it. */
546	if (out != NULL) {
547		KSI_INIT(out);
548		out->ksi_info._signo = signo;
549		out->ksi_info._code = SI_NOINFO;
550	}
551	return 0;
552}
553
554/*
555 * sigget:
556 *
557 *	Fetch the first pending signal from a set.  Optionally, also fetch
558 *	or manufacture a ksiginfo element.  Returns the number of the first
559 *	pending signal, or zero.
560 */
561int
562sigget(sigpend_t *sp, ksiginfo_t *out, int signo, const sigset_t *mask)
563{
564	sigset_t tset;
565	int count;
566
567	/* If there's no pending set, the signal is from the debugger. */
568	if (sp == NULL)
569		goto out;
570
571	/* Construct mask from signo, and 'mask'. */
572	if (signo == 0) {
573		if (mask != NULL) {
574			tset = *mask;
575			__sigandset(&sp->sp_set, &tset);
576		} else
577			tset = sp->sp_set;
578
579		/* If there are no signals pending - return. */
580		if ((signo = firstsig(&tset)) == 0)
581			goto out;
582	} else {
583		KASSERT(sigismember(&sp->sp_set, signo));
584	}
585
586	sigdelset(&sp->sp_set, signo);
587out:
588	count = siggetinfo(sp, out, signo);
589	if (count > 1)
590		sigaddset(&sp->sp_set, signo);
591	return signo;
592}
593
594/*
595 * sigput:
596 *
597 *	Append a new ksiginfo element to the list of pending ksiginfo's.
598 */
599static int
600sigput(sigpend_t *sp, struct proc *p, ksiginfo_t *ksi)
601{
602	ksiginfo_t *kp;
603
604	KASSERT(mutex_owned(p->p_lock));
605	KASSERT((ksi->ksi_flags & KSI_QUEUED) == 0);
606
607	sigaddset(&sp->sp_set, ksi->ksi_signo);
608
609	/*
610	 * If there is no siginfo, we are done.
611	 */
612	if (KSI_EMPTY_P(ksi))
613		return 0;
614
615	KASSERT((ksi->ksi_flags & KSI_FROMPOOL) != 0);
616
617	size_t count = 0;
618	TAILQ_FOREACH(kp, &sp->sp_info, ksi_list) {
619		count++;
620		if (ksi->ksi_signo >= SIGRTMIN && ksi->ksi_signo <= SIGRTMAX)
621			continue;
622		if (kp->ksi_signo == ksi->ksi_signo) {
623			KSI_COPY(ksi, kp);
624			kp->ksi_flags |= KSI_QUEUED;
625			return 0;
626		}
627	}
628
629	if (count >= SIGQUEUE_MAX) {
630#ifdef DIAGNOSTIC
631		printf("%s(%d): Signal queue is full signal=%d\n",
632		    p->p_comm, p->p_pid, ksi->ksi_signo);
633#endif
634		return EAGAIN;
635	}
636	ksi->ksi_flags |= KSI_QUEUED;
637	TAILQ_INSERT_TAIL(&sp->sp_info, ksi, ksi_list);
638
639	return 0;
640}
641
642/*
643 * sigclear:
644 *
645 *	Clear all pending signals in the specified set.
646 */
647void
648sigclear(sigpend_t *sp, const sigset_t *mask, ksiginfoq_t *kq)
649{
650	ksiginfo_t *ksi, *next;
651
652	if (mask == NULL)
653		sigemptyset(&sp->sp_set);
654	else
655		sigminusset(mask, &sp->sp_set);
656
657	TAILQ_FOREACH_SAFE(ksi, &sp->sp_info, ksi_list, next) {
658		if (mask == NULL || sigismember(mask, ksi->ksi_signo)) {
659			TAILQ_REMOVE(&sp->sp_info, ksi, ksi_list);
660			KASSERT((ksi->ksi_flags & KSI_FROMPOOL) != 0);
661			KASSERT((ksi->ksi_flags & KSI_QUEUED) != 0);
662			TAILQ_INSERT_TAIL(kq, ksi, ksi_list);
663		}
664	}
665}
666
667/*
668 * sigclearall:
669 *
670 *	Clear all pending signals in the specified set from a process and
671 *	its LWPs.
672 */
673void
674sigclearall(struct proc *p, const sigset_t *mask, ksiginfoq_t *kq)
675{
676	struct lwp *l;
677
678	KASSERT(mutex_owned(p->p_lock));
679
680	sigclear(&p->p_sigpend, mask, kq);
681
682	LIST_FOREACH(l, &p->p_lwps, l_sibling) {
683		sigclear(&l->l_sigpend, mask, kq);
684	}
685}
686
687/*
688 * sigispending:
689 *
690 *	Return the first signal number if there are pending signals for the
691 *	current LWP.  May be called unlocked provided that LW_PENDSIG is set,
692 *	and that the signal has been posted to the appropriate queue before
693 *	LW_PENDSIG is set.
694 *
695 *	This should only ever be called with (l == curlwp), unless the
696 *	result does not matter (procfs, sysctl).
697 */
698int
699sigispending(struct lwp *l, int signo)
700{
701	struct proc *p = l->l_proc;
702	sigset_t tset;
703
704	membar_consumer();
705
706	tset = l->l_sigpend.sp_set;
707	sigplusset(&p->p_sigpend.sp_set, &tset);
708	sigminusset(&p->p_sigctx.ps_sigignore, &tset);
709	sigminusset(&l->l_sigmask, &tset);
710
711	if (signo == 0) {
712		return firstsig(&tset);
713	}
714	return sigismember(&tset, signo) ? signo : 0;
715}
716
717void
718getucontext(struct lwp *l, ucontext_t *ucp)
719{
720	struct proc *p = l->l_proc;
721
722	KASSERT(mutex_owned(p->p_lock));
723
724	ucp->uc_flags = 0;
725	ucp->uc_link = l->l_ctxlink;
726	ucp->uc_sigmask = l->l_sigmask;
727	ucp->uc_flags |= _UC_SIGMASK;
728
729	/*
730	 * The (unsupplied) definition of the `current execution stack'
731	 * in the System V Interface Definition appears to allow returning
732	 * the main context stack.
733	 */
734	if ((l->l_sigstk.ss_flags & SS_ONSTACK) == 0) {
735		ucp->uc_stack.ss_sp = (void *)l->l_proc->p_stackbase;
736		ucp->uc_stack.ss_size = ctob(l->l_proc->p_vmspace->vm_ssize);
737		ucp->uc_stack.ss_flags = 0;	/* XXX, def. is Very Fishy */
738	} else {
739		/* Simply copy alternate signal execution stack. */
740		ucp->uc_stack = l->l_sigstk;
741	}
742	ucp->uc_flags |= _UC_STACK;
743	mutex_exit(p->p_lock);
744	cpu_getmcontext(l, &ucp->uc_mcontext, &ucp->uc_flags);
745	mutex_enter(p->p_lock);
746}
747
748int
749setucontext(struct lwp *l, const ucontext_t *ucp)
750{
751	struct proc *p = l->l_proc;
752	int error;
753
754	KASSERT(mutex_owned(p->p_lock));
755
756	if ((ucp->uc_flags & _UC_SIGMASK) != 0) {
757		error = sigprocmask1(l, SIG_SETMASK, &ucp->uc_sigmask, NULL);
758		if (error != 0)
759			return error;
760	}
761
762	mutex_exit(p->p_lock);
763	error = cpu_setmcontext(l, &ucp->uc_mcontext, ucp->uc_flags);
764	mutex_enter(p->p_lock);
765	if (error != 0)
766		return (error);
767
768	l->l_ctxlink = ucp->uc_link;
769
770	/*
771	 * If there was stack information, update whether or not we are
772	 * still running on an alternate signal stack.
773	 */
774	if ((ucp->uc_flags & _UC_STACK) != 0) {
775		if (ucp->uc_stack.ss_flags & SS_ONSTACK)
776			l->l_sigstk.ss_flags |= SS_ONSTACK;
777		else
778			l->l_sigstk.ss_flags &= ~SS_ONSTACK;
779	}
780
781	return 0;
782}
783
784/*
785 * killpg1: common code for kill process group/broadcast kill.
786 */
787int
788killpg1(struct lwp *l, ksiginfo_t *ksi, int pgid, int all)
789{
790	struct proc	*p, *cp;
791	kauth_cred_t	pc;
792	struct pgrp	*pgrp;
793	int		nfound;
794	int		signo = ksi->ksi_signo;
795
796	cp = l->l_proc;
797	pc = l->l_cred;
798	nfound = 0;
799
800	mutex_enter(&proc_lock);
801	if (all) {
802		/*
803		 * Broadcast.
804		 */
805		PROCLIST_FOREACH(p, &allproc) {
806			if (p->p_pid <= 1 || p == cp ||
807			    (p->p_flag & PK_SYSTEM) != 0)
808				continue;
809			mutex_enter(p->p_lock);
810			if (kauth_authorize_process(pc,
811			    KAUTH_PROCESS_SIGNAL, p, KAUTH_ARG(signo), NULL,
812			    NULL) == 0) {
813				nfound++;
814				if (signo)
815					kpsignal2(p, ksi);
816			}
817			mutex_exit(p->p_lock);
818		}
819	} else {
820		if (pgid == 0)
821			/* Zero pgid means send to my process group. */
822			pgrp = cp->p_pgrp;
823		else {
824			pgrp = pgrp_find(pgid);
825			if (pgrp == NULL)
826				goto out;
827		}
828		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
829			if (p->p_pid <= 1 || p->p_flag & PK_SYSTEM)
830				continue;
831			mutex_enter(p->p_lock);
832			if (kauth_authorize_process(pc, KAUTH_PROCESS_SIGNAL,
833			    p, KAUTH_ARG(signo), NULL, NULL) == 0) {
834				nfound++;
835				if (signo && P_ZOMBIE(p) == 0)
836					kpsignal2(p, ksi);
837			}
838			mutex_exit(p->p_lock);
839		}
840	}
841out:
842	mutex_exit(&proc_lock);
843	return nfound ? 0 : ESRCH;
844}
845
846/*
847 * Send a signal to a process group.  If checktty is set, limit to members
848 * which have a controlling terminal.
849 */
850void
851pgsignal(struct pgrp *pgrp, int sig, int checkctty)
852{
853	ksiginfo_t ksi;
854
855	KASSERT(!cpu_intr_p());
856	KASSERT(mutex_owned(&proc_lock));
857
858	KSI_INIT_EMPTY(&ksi);
859	ksi.ksi_signo = sig;
860	kpgsignal(pgrp, &ksi, NULL, checkctty);
861}
862
863void
864kpgsignal(struct pgrp *pgrp, ksiginfo_t *ksi, void *data, int checkctty)
865{
866	struct proc *p;
867
868	KASSERT(!cpu_intr_p());
869	KASSERT(mutex_owned(&proc_lock));
870	KASSERT(pgrp != NULL);
871
872	LIST_FOREACH(p, &pgrp->pg_members, p_pglist)
873		if (checkctty == 0 || p->p_lflag & PL_CONTROLT)
874			kpsignal(p, ksi, data);
875}
876
877/*
878 * Send a signal caused by a trap to the current LWP.  If it will be caught
879 * immediately, deliver it with correct code.  Otherwise, post it normally.
880 */
881void
882trapsignal(struct lwp *l, ksiginfo_t *ksi)
883{
884	struct proc	*p;
885	struct sigacts	*ps;
886	int signo = ksi->ksi_signo;
887	sigset_t *mask;
888	sig_t action;
889
890	KASSERT(KSI_TRAP_P(ksi));
891
892	ksi->ksi_lid = l->l_lid;
893	p = l->l_proc;
894
895	KASSERT(!cpu_intr_p());
896	mutex_enter(&proc_lock);
897	mutex_enter(p->p_lock);
898
899repeat:
900	/*
901	 * If we are exiting, demise now.
902	 *
903	 * This avoids notifying tracer and deadlocking.
904	 */
905	if (__predict_false(ISSET(p->p_sflag, PS_WEXIT))) {
906		mutex_exit(p->p_lock);
907		mutex_exit(&proc_lock);
908		lwp_exit(l);
909		panic("trapsignal");
910		/* NOTREACHED */
911	}
912
913	/*
914	 * The process is already stopping.
915	 */
916	if ((p->p_sflag & PS_STOPPING) != 0) {
917		mutex_exit(&proc_lock);
918		sigswitch_unlock_and_switch_away(l);
919		mutex_enter(&proc_lock);
920		mutex_enter(p->p_lock);
921		goto repeat;
922	}
923
924	mask = &l->l_sigmask;
925	ps = p->p_sigacts;
926	action = SIGACTION_PS(ps, signo).sa_handler;
927
928	if (ISSET(p->p_slflag, PSL_TRACED) &&
929	    !(p->p_pptr == p->p_opptr && ISSET(p->p_lflag, PL_PPWAIT)) &&
930	    p->p_xsig != SIGKILL &&
931	    !sigismember(&p->p_sigpend.sp_set, SIGKILL)) {
932		p->p_xsig = signo;
933		p->p_sigctx.ps_faked = true;
934		p->p_sigctx.ps_lwp = ksi->ksi_lid;
935		p->p_sigctx.ps_info = ksi->ksi_info;
936		sigswitch(0, signo, true);
937
938		if (ktrpoint(KTR_PSIG)) {
939			if (p->p_emul->e_ktrpsig)
940				p->p_emul->e_ktrpsig(signo, action, mask, ksi);
941			else
942				ktrpsig(signo, action, mask, ksi);
943		}
944		return;
945	}
946
947	const bool caught = sigismember(&p->p_sigctx.ps_sigcatch, signo);
948	const bool masked = sigismember(mask, signo);
949	if (caught && !masked) {
950		mutex_exit(&proc_lock);
951		l->l_ru.ru_nsignals++;
952		kpsendsig(l, ksi, mask);
953		mutex_exit(p->p_lock);
954
955		if (ktrpoint(KTR_PSIG)) {
956			if (p->p_emul->e_ktrpsig)
957				p->p_emul->e_ktrpsig(signo, action, mask, ksi);
958			else
959				ktrpsig(signo, action, mask, ksi);
960		}
961		return;
962	}
963
964	/*
965	 * If the signal is masked or ignored, then unmask it and
966	 * reset it to the default action so that the process or
967	 * its tracer will be notified.
968	 */
969	const bool ignored = action == SIG_IGN;
970	if (masked || ignored) {
971		mutex_enter(&ps->sa_mutex);
972		sigdelset(mask, signo);
973		sigdelset(&p->p_sigctx.ps_sigcatch, signo);
974		sigdelset(&p->p_sigctx.ps_sigignore, signo);
975		sigdelset(&SIGACTION_PS(ps, signo).sa_mask, signo);
976		SIGACTION_PS(ps, signo).sa_handler = SIG_DFL;
977		mutex_exit(&ps->sa_mutex);
978	}
979
980	kpsignal2(p, ksi);
981	mutex_exit(p->p_lock);
982	mutex_exit(&proc_lock);
983}
984
985/*
986 * Fill in signal information and signal the parent for a child status change.
987 */
988void
989child_psignal(struct proc *p, int mask)
990{
991	ksiginfo_t ksi;
992	struct proc *q;
993	int xsig;
994
995	KASSERT(mutex_owned(&proc_lock));
996	KASSERT(mutex_owned(p->p_lock));
997
998	xsig = p->p_xsig;
999
1000	KSI_INIT(&ksi);
1001	ksi.ksi_signo = SIGCHLD;
1002	ksi.ksi_code = (xsig == SIGCONT ? CLD_CONTINUED : CLD_STOPPED);
1003	ksi.ksi_pid = p->p_pid;
1004	ksi.ksi_uid = kauth_cred_geteuid(p->p_cred);
1005	ksi.ksi_status = xsig;
1006	ksi.ksi_utime = p->p_stats->p_ru.ru_utime.tv_sec;
1007	ksi.ksi_stime = p->p_stats->p_ru.ru_stime.tv_sec;
1008
1009	q = p->p_pptr;
1010
1011	mutex_exit(p->p_lock);
1012	mutex_enter(q->p_lock);
1013
1014	if ((q->p_sflag & mask) == 0)
1015		kpsignal2(q, &ksi);
1016
1017	mutex_exit(q->p_lock);
1018	mutex_enter(p->p_lock);
1019}
1020
1021void
1022psignal(struct proc *p, int signo)
1023{
1024	ksiginfo_t ksi;
1025
1026	KASSERT(!cpu_intr_p());
1027	KASSERT(mutex_owned(&proc_lock));
1028
1029	KSI_INIT_EMPTY(&ksi);
1030	ksi.ksi_signo = signo;
1031	mutex_enter(p->p_lock);
1032	kpsignal2(p, &ksi);
1033	mutex_exit(p->p_lock);
1034}
1035
1036void
1037kpsignal(struct proc *p, ksiginfo_t *ksi, void *data)
1038{
1039	fdfile_t *ff;
1040	file_t *fp;
1041	fdtab_t *dt;
1042
1043	KASSERT(!cpu_intr_p());
1044	KASSERT(mutex_owned(&proc_lock));
1045
1046	if ((p->p_sflag & PS_WEXIT) == 0 && data) {
1047		size_t fd;
1048		filedesc_t *fdp = p->p_fd;
1049
1050		/* XXXSMP locking */
1051		ksi->ksi_fd = -1;
1052		dt = atomic_load_consume(&fdp->fd_dt);
1053		for (fd = 0; fd < dt->dt_nfiles; fd++) {
1054			if ((ff = dt->dt_ff[fd]) == NULL)
1055				continue;
1056			if ((fp = atomic_load_consume(&ff->ff_file)) == NULL)
1057				continue;
1058			if (fp->f_data == data) {
1059				ksi->ksi_fd = fd;
1060				break;
1061			}
1062		}
1063	}
1064	mutex_enter(p->p_lock);
1065	kpsignal2(p, ksi);
1066	mutex_exit(p->p_lock);
1067}
1068
1069/*
1070 * sigismasked:
1071 *
1072 *	Returns true if signal is ignored or masked for the specified LWP.
1073 */
1074int
1075sigismasked(struct lwp *l, int sig)
1076{
1077	struct proc *p = l->l_proc;
1078
1079	return sigismember(&p->p_sigctx.ps_sigignore, sig) ||
1080	    sigismember(&l->l_sigmask, sig);
1081}
1082
1083/*
1084 * sigpost:
1085 *
1086 *	Post a pending signal to an LWP.  Returns non-zero if the LWP may
1087 *	be able to take the signal.
1088 */
1089static int
1090sigpost(struct lwp *l, sig_t action, int prop, int sig)
1091{
1092	int rv, masked;
1093	struct proc *p = l->l_proc;
1094
1095	KASSERT(mutex_owned(p->p_lock));
1096
1097	/*
1098	 * If the LWP is on the way out, sigclear() will be busy draining all
1099	 * pending signals.  Don't give it more.
1100	 */
1101	if (l->l_stat == LSZOMB)
1102		return 0;
1103
1104	SDT_PROBE(proc, kernel, , signal__send, l, p, sig, 0, 0);
1105
1106	lwp_lock(l);
1107	if (__predict_false((l->l_flag & LW_DBGSUSPEND) != 0)) {
1108		if ((prop & SA_KILL) != 0)
1109			l->l_flag &= ~LW_DBGSUSPEND;
1110		else {
1111			lwp_unlock(l);
1112			return 0;
1113		}
1114	}
1115
1116	/*
1117	 * Have the LWP check for signals.  This ensures that even if no LWP
1118	 * is found to take the signal immediately, it should be taken soon.
1119	 */
1120	signotify(l);
1121
1122	/*
1123	 * SIGCONT can be masked, but if LWP is stopped, it needs restart.
1124	 * Note: SIGKILL and SIGSTOP cannot be masked.
1125	 */
1126	masked = sigismember(&l->l_sigmask, sig);
1127	if (masked && ((prop & SA_CONT) == 0 || l->l_stat != LSSTOP)) {
1128		lwp_unlock(l);
1129		return 0;
1130	}
1131
1132	/*
1133	 * If killing the process, make it run fast.
1134	 */
1135	if (__predict_false((prop & SA_KILL) != 0) &&
1136	    action == SIG_DFL && l->l_priority < MAXPRI_USER) {
1137		KASSERT(l->l_class == SCHED_OTHER);
1138		lwp_changepri(l, MAXPRI_USER);
1139	}
1140
1141	/*
1142	 * If the LWP is running or on a run queue, then we win.  If it's
1143	 * sleeping interruptably, wake it and make it take the signal.  If
1144	 * the sleep isn't interruptable, then the chances are it will get
1145	 * to see the signal soon anyhow.  If suspended, it can't take the
1146	 * signal right now.  If it's LWP private or for all LWPs, save it
1147	 * for later; otherwise punt.
1148	 */
1149	rv = 0;
1150
1151	switch (l->l_stat) {
1152	case LSRUN:
1153	case LSONPROC:
1154		rv = 1;
1155		break;
1156
1157	case LSSLEEP:
1158		if ((l->l_flag & LW_SINTR) != 0) {
1159			/* setrunnable() will release the lock. */
1160			setrunnable(l);
1161			return 1;
1162		}
1163		break;
1164
1165	case LSSUSPENDED:
1166		if ((prop & SA_KILL) != 0 && (l->l_flag & LW_WCORE) != 0) {
1167			/* lwp_continue() will release the lock. */
1168			lwp_continue(l);
1169			return 1;
1170		}
1171		break;
1172
1173	case LSSTOP:
1174		if ((prop & SA_STOP) != 0)
1175			break;
1176
1177		/*
1178		 * If the LWP is stopped and we are sending a continue
1179		 * signal, then start it again.
1180		 */
1181		if ((prop & SA_CONT) != 0) {
1182			if (l->l_wchan != NULL) {
1183				l->l_stat = LSSLEEP;
1184				p->p_nrlwps++;
1185				rv = 1;
1186				break;
1187			}
1188			/* setrunnable() will release the lock. */
1189			setrunnable(l);
1190			return 1;
1191		} else if (l->l_wchan == NULL || (l->l_flag & LW_SINTR) != 0) {
1192			/* setrunnable() will release the lock. */
1193			setrunnable(l);
1194			return 1;
1195		}
1196		break;
1197
1198	default:
1199		break;
1200	}
1201
1202	lwp_unlock(l);
1203	return rv;
1204}
1205
1206/*
1207 * Notify an LWP that it has a pending signal.
1208 */
1209void
1210signotify(struct lwp *l)
1211{
1212	KASSERT(lwp_locked(l, NULL));
1213
1214	l->l_flag |= LW_PENDSIG;
1215	lwp_need_userret(l);
1216}
1217
1218/*
1219 * Find an LWP within process p that is waiting on signal ksi, and hand
1220 * it on.
1221 */
1222static int
1223sigunwait(struct proc *p, const ksiginfo_t *ksi)
1224{
1225	struct lwp *l;
1226	int signo;
1227
1228	KASSERT(mutex_owned(p->p_lock));
1229
1230	signo = ksi->ksi_signo;
1231
1232	if (ksi->ksi_lid != 0) {
1233		/*
1234		 * Signal came via _lwp_kill().  Find the LWP and see if
1235		 * it's interested.
1236		 */
1237		if ((l = lwp_find(p, ksi->ksi_lid)) == NULL)
1238			return 0;
1239		if (l->l_sigwaited == NULL ||
1240		    !sigismember(&l->l_sigwaitset, signo))
1241			return 0;
1242	} else {
1243		/*
1244		 * Look for any LWP that may be interested.
1245		 */
1246		LIST_FOREACH(l, &p->p_sigwaiters, l_sigwaiter) {
1247			KASSERT(l->l_sigwaited != NULL);
1248			if (sigismember(&l->l_sigwaitset, signo))
1249				break;
1250		}
1251	}
1252
1253	if (l != NULL) {
1254		l->l_sigwaited->ksi_info = ksi->ksi_info;
1255		l->l_sigwaited = NULL;
1256		LIST_REMOVE(l, l_sigwaiter);
1257		cv_signal(&l->l_sigcv);
1258		return 1;
1259	}
1260
1261	return 0;
1262}
1263
1264/*
1265 * Send the signal to the process.  If the signal has an action, the action
1266 * is usually performed by the target process rather than the caller; we add
1267 * the signal to the set of pending signals for the process.
1268 *
1269 * Exceptions:
1270 *   o When a stop signal is sent to a sleeping process that takes the
1271 *     default action, the process is stopped without awakening it.
1272 *   o SIGCONT restarts stopped processes (or puts them back to sleep)
1273 *     regardless of the signal action (eg, blocked or ignored).
1274 *
1275 * Other ignored signals are discarded immediately.
1276 */
1277int
1278kpsignal2(struct proc *p, ksiginfo_t *ksi)
1279{
1280	int prop, signo = ksi->ksi_signo;
1281	struct lwp *l = NULL;
1282	ksiginfo_t *kp;
1283	lwpid_t lid;
1284	sig_t action;
1285	bool toall;
1286	bool traced;
1287	int error = 0;
1288
1289	KASSERT(!cpu_intr_p());
1290	KASSERT(mutex_owned(&proc_lock));
1291	KASSERT(mutex_owned(p->p_lock));
1292	KASSERT((ksi->ksi_flags & KSI_QUEUED) == 0);
1293	KASSERT(signo > 0);
1294	KASSERT(signo < NSIG);
1295
1296	/*
1297	 * If the process is being created by fork, is a zombie or is
1298	 * exiting, then just drop the signal here and bail out.
1299	 */
1300	if (p->p_stat != SACTIVE && p->p_stat != SSTOP)
1301		return 0;
1302
1303	/*
1304	 * Notify any interested parties of the signal.
1305	 */
1306	KNOTE(&p->p_klist, NOTE_SIGNAL | signo);
1307
1308	/*
1309	 * Some signals including SIGKILL must act on the entire process.
1310	 */
1311	kp = NULL;
1312	prop = sigprop[signo];
1313	toall = ((prop & SA_TOALL) != 0);
1314	lid = toall ? 0 : ksi->ksi_lid;
1315	traced = ISSET(p->p_slflag, PSL_TRACED) &&
1316	    !sigismember(&p->p_sigctx.ps_sigpass, signo);
1317
1318	/*
1319	 * If proc is traced, always give parent a chance.
1320	 */
1321	if (traced) {
1322		action = SIG_DFL;
1323
1324		if (lid == 0) {
1325			/*
1326			 * If the process is being traced and the signal
1327			 * is being caught, make sure to save any ksiginfo.
1328			 */
1329			if ((kp = ksiginfo_alloc(p, ksi, PR_NOWAIT)) == NULL)
1330				goto discard;
1331			if ((error = sigput(&p->p_sigpend, p, kp)) != 0)
1332				goto out;
1333		}
1334	} else {
1335
1336		/*
1337		 * If the signal is being ignored, then drop it.  Note: we
1338		 * don't set SIGCONT in ps_sigignore, and if it is set to
1339		 * SIG_IGN, action will be SIG_DFL here.
1340		 */
1341		if (sigismember(&p->p_sigctx.ps_sigignore, signo))
1342			goto discard;
1343
1344		else if (sigismember(&p->p_sigctx.ps_sigcatch, signo))
1345			action = SIG_CATCH;
1346		else {
1347			action = SIG_DFL;
1348
1349			/*
1350			 * If sending a tty stop signal to a member of an
1351			 * orphaned process group, discard the signal here if
1352			 * the action is default; don't stop the process below
1353			 * if sleeping, and don't clear any pending SIGCONT.
1354			 */
1355			if (prop & SA_TTYSTOP && p->p_pgrp->pg_jobc == 0)
1356				goto discard;
1357
1358			if (prop & SA_KILL && p->p_nice > NZERO)
1359				p->p_nice = NZERO;
1360		}
1361	}
1362
1363	/*
1364	 * If stopping or continuing a process, discard any pending
1365	 * signals that would do the inverse.
1366	 */
1367	if ((prop & (SA_CONT | SA_STOP)) != 0) {
1368		ksiginfoq_t kq;
1369
1370		ksiginfo_queue_init(&kq);
1371		if ((prop & SA_CONT) != 0)
1372			sigclear(&p->p_sigpend, &stopsigmask, &kq);
1373		if ((prop & SA_STOP) != 0)
1374			sigclear(&p->p_sigpend, &contsigmask, &kq);
1375		ksiginfo_queue_drain(&kq);	/* XXXSMP */
1376	}
1377
1378	/*
1379	 * If the signal doesn't have SA_CANTMASK (no override for SIGKILL,
1380	 * please!), check if any LWPs are waiting on it.  If yes, pass on
1381	 * the signal info.  The signal won't be processed further here.
1382	 */
1383	if ((prop & SA_CANTMASK) == 0 && !LIST_EMPTY(&p->p_sigwaiters) &&
1384	    p->p_stat == SACTIVE && (p->p_sflag & PS_STOPPING) == 0 &&
1385	    sigunwait(p, ksi))
1386		goto discard;
1387
1388	/*
1389	 * XXXSMP Should be allocated by the caller, we're holding locks
1390	 * here.
1391	 */
1392	if (kp == NULL && (kp = ksiginfo_alloc(p, ksi, PR_NOWAIT)) == NULL)
1393		goto discard;
1394
1395	/*
1396	 * LWP private signals are easy - just find the LWP and post
1397	 * the signal to it.
1398	 */
1399	if (lid != 0) {
1400		l = lwp_find(p, lid);
1401		if (l != NULL) {
1402			if ((error = sigput(&l->l_sigpend, p, kp)) != 0)
1403				goto out;
1404			membar_producer();
1405			if (sigpost(l, action, prop, kp->ksi_signo) != 0)
1406				signo = -1;
1407		}
1408		goto out;
1409	}
1410
1411	/*
1412	 * Some signals go to all LWPs, even if posted with _lwp_kill()
1413	 * or for an SA process.
1414	 */
1415	if (p->p_stat == SACTIVE && (p->p_sflag & PS_STOPPING) == 0) {
1416		if (traced)
1417			goto deliver;
1418
1419		/*
1420		 * If SIGCONT is default (or ignored) and process is
1421		 * asleep, we are finished; the process should not
1422		 * be awakened.
1423		 */
1424		if ((prop & SA_CONT) != 0 && action == SIG_DFL)
1425			goto out;
1426	} else {
1427		/*
1428		 * Process is stopped or stopping.
1429		 * - If traced, then no action is needed, unless killing.
1430		 * - Run the process only if sending SIGCONT or SIGKILL.
1431		 */
1432		if (traced && signo != SIGKILL) {
1433			goto out;
1434		}
1435		if ((prop & SA_CONT) != 0 || signo == SIGKILL) {
1436			/*
1437			 * Re-adjust p_nstopchild if the process was
1438			 * stopped but not yet collected by its parent.
1439			 */
1440			if (p->p_stat == SSTOP && !p->p_waited)
1441				p->p_pptr->p_nstopchild--;
1442			p->p_stat = SACTIVE;
1443			p->p_sflag &= ~PS_STOPPING;
1444			if (traced) {
1445				KASSERT(signo == SIGKILL);
1446				goto deliver;
1447			}
1448			/*
1449			 * Do not make signal pending if SIGCONT is default.
1450			 *
1451			 * If the process catches SIGCONT, let it handle the
1452			 * signal itself (if waiting on event - process runs,
1453			 * otherwise continues sleeping).
1454			 */
1455			if ((prop & SA_CONT) != 0) {
1456				p->p_xsig = SIGCONT;
1457				p->p_sflag |= PS_CONTINUED;
1458				child_psignal(p, 0);
1459				if (action == SIG_DFL) {
1460					KASSERT(signo != SIGKILL);
1461					goto deliver;
1462				}
1463			}
1464		} else if ((prop & SA_STOP) != 0) {
1465			/*
1466			 * Already stopped, don't need to stop again.
1467			 * (If we did the shell could get confused.)
1468			 */
1469			goto out;
1470		}
1471	}
1472	/*
1473	 * Make signal pending.
1474	 */
1475	KASSERT(!traced);
1476	if ((error = sigput(&p->p_sigpend, p, kp)) != 0)
1477		goto out;
1478deliver:
1479	/*
1480	 * Before we set LW_PENDSIG on any LWP, ensure that the signal is
1481	 * visible on the per process list (for sigispending()).  This
1482	 * is unlikely to be needed in practice, but...
1483	 */
1484	membar_producer();
1485
1486	/*
1487	 * Try to find an LWP that can take the signal.
1488	 */
1489	LIST_FOREACH(l, &p->p_lwps, l_sibling) {
1490		if (sigpost(l, action, prop, kp->ksi_signo) && !toall)
1491			break;
1492	}
1493	signo = -1;
1494out:
1495	/*
1496	 * If the ksiginfo wasn't used, then bin it.  XXXSMP freeing memory
1497	 * with locks held.  The caller should take care of this.
1498	 */
1499	ksiginfo_free(kp);
1500	if (signo == -1)
1501		return error;
1502discard:
1503	SDT_PROBE(proc, kernel, , signal__discard, l, p, signo, 0, 0);
1504	return error;
1505}
1506
1507void
1508kpsendsig(struct lwp *l, const ksiginfo_t *ksi, const sigset_t *mask)
1509{
1510	struct proc *p = l->l_proc;
1511
1512	KASSERT(mutex_owned(p->p_lock));
1513	(*p->p_emul->e_sendsig)(ksi, mask);
1514}
1515
1516/*
1517 * Stop any LWPs sleeping interruptably.
1518 */
1519static void
1520proc_stop_lwps(struct proc *p)
1521{
1522	struct lwp *l;
1523
1524	KASSERT(mutex_owned(p->p_lock));
1525	KASSERT((p->p_sflag & PS_STOPPING) != 0);
1526
1527	LIST_FOREACH(l, &p->p_lwps, l_sibling) {
1528		lwp_lock(l);
1529		if (l->l_stat == LSSLEEP && (l->l_flag & LW_SINTR) != 0) {
1530			l->l_stat = LSSTOP;
1531			p->p_nrlwps--;
1532		}
1533		lwp_unlock(l);
1534	}
1535}
1536
1537/*
1538 * Finish stopping of a process.  Mark it stopped and notify the parent.
1539 *
1540 * Drop p_lock briefly if ppsig is true.
1541 */
1542static void
1543proc_stop_done(struct proc *p, int ppmask)
1544{
1545
1546	KASSERT(mutex_owned(&proc_lock));
1547	KASSERT(mutex_owned(p->p_lock));
1548	KASSERT((p->p_sflag & PS_STOPPING) != 0);
1549	KASSERT(p->p_nrlwps == 0 || p->p_nrlwps == 1);
1550	KASSERT(p->p_nrlwps == 0 || p == curproc);
1551
1552	p->p_sflag &= ~PS_STOPPING;
1553	p->p_stat = SSTOP;
1554	p->p_waited = 0;
1555	p->p_pptr->p_nstopchild++;
1556
1557	/* child_psignal drops p_lock briefly. */
1558	child_psignal(p, ppmask);
1559	cv_broadcast(&p->p_pptr->p_waitcv);
1560}
1561
1562/*
1563 * Stop the current process and switch away to the debugger notifying
1564 * an event specific to a traced process only.
1565 */
1566void
1567eventswitch(int code, int pe_report_event, int entity)
1568{
1569	struct lwp *l = curlwp;
1570	struct proc *p = l->l_proc;
1571	struct sigacts *ps;
1572	sigset_t *mask;
1573	sig_t action;
1574	ksiginfo_t ksi;
1575	const int signo = SIGTRAP;
1576
1577	KASSERT(mutex_owned(&proc_lock));
1578	KASSERT(mutex_owned(p->p_lock));
1579	KASSERT(p->p_pptr != initproc);
1580	KASSERT(l->l_stat == LSONPROC);
1581	KASSERT(ISSET(p->p_slflag, PSL_TRACED));
1582	KASSERT(!ISSET(l->l_flag, LW_SYSTEM));
1583	KASSERT(p->p_nrlwps > 0);
1584	KASSERT((code == TRAP_CHLD) || (code == TRAP_LWP) ||
1585	        (code == TRAP_EXEC));
1586	KASSERT((code != TRAP_CHLD) || (entity > 1)); /* prevent pid1 */
1587	KASSERT((code != TRAP_LWP) || (entity > 0));
1588
1589repeat:
1590	/*
1591	 * If we are exiting, demise now.
1592	 *
1593	 * This avoids notifying tracer and deadlocking.
1594	 */
1595	if (__predict_false(ISSET(p->p_sflag, PS_WEXIT))) {
1596		mutex_exit(p->p_lock);
1597		mutex_exit(&proc_lock);
1598
1599		if (pe_report_event == PTRACE_LWP_EXIT) {
1600			/* Avoid double lwp_exit() and panic. */
1601			return;
1602		}
1603
1604		lwp_exit(l);
1605		panic("eventswitch");
1606		/* NOTREACHED */
1607	}
1608
1609	/*
1610	 * If we are no longer traced, abandon this event signal.
1611	 *
1612	 * This avoids killing a process after detaching the debugger.
1613	 */
1614	if (__predict_false(!ISSET(p->p_slflag, PSL_TRACED))) {
1615		mutex_exit(p->p_lock);
1616		mutex_exit(&proc_lock);
1617		return;
1618	}
1619
1620	/*
1621	 * If there's a pending SIGKILL process it immediately.
1622	 */
1623	if (p->p_xsig == SIGKILL ||
1624	    sigismember(&p->p_sigpend.sp_set, SIGKILL)) {
1625		mutex_exit(p->p_lock);
1626		mutex_exit(&proc_lock);
1627		return;
1628	}
1629
1630	/*
1631	 * The process is already stopping.
1632	 */
1633	if ((p->p_sflag & PS_STOPPING) != 0) {
1634		mutex_exit(&proc_lock);
1635		sigswitch_unlock_and_switch_away(l);
1636		mutex_enter(&proc_lock);
1637		mutex_enter(p->p_lock);
1638		goto repeat;
1639	}
1640
1641	KSI_INIT_TRAP(&ksi);
1642	ksi.ksi_lid = l->l_lid;
1643	ksi.ksi_signo = signo;
1644	ksi.ksi_code = code;
1645	ksi.ksi_pe_report_event = pe_report_event;
1646
1647	CTASSERT(sizeof(ksi.ksi_pe_other_pid) == sizeof(ksi.ksi_pe_lwp));
1648	ksi.ksi_pe_other_pid = entity;
1649
1650	/* Needed for ktrace */
1651	ps = p->p_sigacts;
1652	action = SIGACTION_PS(ps, signo).sa_handler;
1653	mask = &l->l_sigmask;
1654
1655	p->p_xsig = signo;
1656	p->p_sigctx.ps_faked = true;
1657	p->p_sigctx.ps_lwp = ksi.ksi_lid;
1658	p->p_sigctx.ps_info = ksi.ksi_info;
1659
1660	sigswitch(0, signo, true);
1661
1662	if (code == TRAP_CHLD) {
1663		mutex_enter(&proc_lock);
1664		while (l->l_vforkwaiting)
1665			cv_wait(&l->l_waitcv, &proc_lock);
1666		mutex_exit(&proc_lock);
1667	}
1668
1669	if (ktrpoint(KTR_PSIG)) {
1670		if (p->p_emul->e_ktrpsig)
1671			p->p_emul->e_ktrpsig(signo, action, mask, &ksi);
1672		else
1673			ktrpsig(signo, action, mask, &ksi);
1674	}
1675}
1676
1677void
1678eventswitchchild(struct proc *p, int code, int pe_report_event)
1679{
1680	mutex_enter(&proc_lock);
1681	mutex_enter(p->p_lock);
1682	if ((p->p_slflag & (PSL_TRACED|PSL_TRACEDCHILD)) !=
1683	    (PSL_TRACED|PSL_TRACEDCHILD)) {
1684		mutex_exit(p->p_lock);
1685		mutex_exit(&proc_lock);
1686		return;
1687	}
1688	eventswitch(code, pe_report_event, p->p_oppid);
1689}
1690
1691/*
1692 * Stop the current process and switch away when being stopped or traced.
1693 */
1694static void
1695sigswitch(int ppmask, int signo, bool proc_lock_held)
1696{
1697	struct lwp *l = curlwp;
1698	struct proc *p = l->l_proc;
1699
1700	KASSERT(mutex_owned(p->p_lock));
1701	KASSERT(l->l_stat == LSONPROC);
1702	KASSERT(p->p_nrlwps > 0);
1703
1704	if (proc_lock_held) {
1705		KASSERT(mutex_owned(&proc_lock));
1706	} else {
1707		KASSERT(!mutex_owned(&proc_lock));
1708	}
1709
1710	/*
1711	 * On entry we know that the process needs to stop.  If it's
1712	 * the result of a 'sideways' stop signal that has been sourced
1713	 * through issignal(), then stop other LWPs in the process too.
1714	 */
1715	if (p->p_stat == SACTIVE && (p->p_sflag & PS_STOPPING) == 0) {
1716		KASSERT(signo != 0);
1717		proc_stop(p, signo);
1718		KASSERT(p->p_nrlwps > 0);
1719	}
1720
1721	/*
1722	 * If we are the last live LWP, and the stop was a result of
1723	 * a new signal, then signal the parent.
1724	 */
1725	if ((p->p_sflag & PS_STOPPING) != 0) {
1726		if (!proc_lock_held && !mutex_tryenter(&proc_lock)) {
1727			mutex_exit(p->p_lock);
1728			mutex_enter(&proc_lock);
1729			mutex_enter(p->p_lock);
1730		}
1731
1732		if (p->p_nrlwps == 1 && (p->p_sflag & PS_STOPPING) != 0) {
1733			/*
1734			 * Note that proc_stop_done() can drop
1735			 * p->p_lock briefly.
1736			 */
1737			proc_stop_done(p, ppmask);
1738		}
1739
1740		mutex_exit(&proc_lock);
1741	}
1742
1743	sigswitch_unlock_and_switch_away(l);
1744}
1745
1746/*
1747 * Unlock and switch away.
1748 */
1749static void
1750sigswitch_unlock_and_switch_away(struct lwp *l)
1751{
1752	struct proc *p;
1753
1754	p = l->l_proc;
1755
1756	KASSERT(mutex_owned(p->p_lock));
1757	KASSERT(!mutex_owned(&proc_lock));
1758
1759	KASSERT(l->l_stat == LSONPROC);
1760	KASSERT(p->p_nrlwps > 0);
1761	KASSERT(l->l_blcnt == 0);
1762
1763	if (p->p_stat == SSTOP || (p->p_sflag & PS_STOPPING) != 0) {
1764		p->p_nrlwps--;
1765		lwp_lock(l);
1766		KASSERT(l->l_stat == LSONPROC || l->l_stat == LSSLEEP);
1767		l->l_stat = LSSTOP;
1768		lwp_unlock(l);
1769	}
1770
1771	mutex_exit(p->p_lock);
1772	lwp_lock(l);
1773	spc_lock(l->l_cpu);
1774	mi_switch(l);
1775}
1776
1777/*
1778 * Check for a signal from the debugger.
1779 */
1780static int
1781sigchecktrace(void)
1782{
1783	struct lwp *l = curlwp;
1784	struct proc *p = l->l_proc;
1785	int signo;
1786
1787	KASSERT(mutex_owned(p->p_lock));
1788
1789	/* If there's a pending SIGKILL, process it immediately. */
1790	if (sigismember(&p->p_sigpend.sp_set, SIGKILL))
1791		return 0;
1792
1793	/*
1794	 * If we are no longer being traced, or the parent didn't
1795	 * give us a signal, or we're stopping, look for more signals.
1796	 */
1797	if ((p->p_slflag & PSL_TRACED) == 0 || p->p_xsig == 0 ||
1798	    (p->p_sflag & PS_STOPPING) != 0)
1799		return 0;
1800
1801	/*
1802	 * If the new signal is being masked, look for other signals.
1803	 * `p->p_sigctx.ps_siglist |= mask' is done in setrunnable().
1804	 */
1805	signo = p->p_xsig;
1806	p->p_xsig = 0;
1807	if (sigismember(&l->l_sigmask, signo)) {
1808		signo = 0;
1809	}
1810	return signo;
1811}
1812
1813/*
1814 * If the current process has received a signal (should be caught or cause
1815 * termination, should interrupt current syscall), return the signal number.
1816 *
1817 * Stop signals with default action are processed immediately, then cleared;
1818 * they aren't returned.  This is checked after each entry to the system for
1819 * a syscall or trap.
1820 *
1821 * We will also return -1 if the process is exiting and the current LWP must
1822 * follow suit.
1823 */
1824int
1825issignal(struct lwp *l)
1826{
1827	struct proc *p;
1828	int siglwp, signo, prop;
1829	sigpend_t *sp;
1830	sigset_t ss;
1831	bool traced;
1832
1833	p = l->l_proc;
1834	sp = NULL;
1835	signo = 0;
1836
1837	KASSERT(p == curproc);
1838	KASSERT(mutex_owned(p->p_lock));
1839
1840	for (;;) {
1841		/* Discard any signals that we have decided not to take. */
1842		if (signo != 0) {
1843			(void)sigget(sp, NULL, signo, NULL);
1844		}
1845
1846		/*
1847		 * If the process is stopped/stopping, then stop ourselves
1848		 * now that we're on the kernel/userspace boundary.  When
1849		 * we awaken, check for a signal from the debugger.
1850		 */
1851		if (p->p_stat == SSTOP || (p->p_sflag & PS_STOPPING) != 0) {
1852			sigswitch_unlock_and_switch_away(l);
1853			mutex_enter(p->p_lock);
1854			continue;
1855		} else if (p->p_stat == SACTIVE)
1856			signo = sigchecktrace();
1857		else
1858			signo = 0;
1859
1860		/* Signals from the debugger are "out of band". */
1861		sp = NULL;
1862
1863		/*
1864		 * If the debugger didn't provide a signal, find a pending
1865		 * signal from our set.  Check per-LWP signals first, and
1866		 * then per-process.
1867		 */
1868		if (signo == 0) {
1869			sp = &l->l_sigpend;
1870			ss = sp->sp_set;
1871			siglwp = l->l_lid;
1872			if ((p->p_lflag & PL_PPWAIT) != 0)
1873				sigminusset(&vforksigmask, &ss);
1874			sigminusset(&l->l_sigmask, &ss);
1875
1876			if ((signo = firstsig(&ss)) == 0) {
1877				sp = &p->p_sigpend;
1878				ss = sp->sp_set;
1879				siglwp = 0;
1880				if ((p->p_lflag & PL_PPWAIT) != 0)
1881					sigminusset(&vforksigmask, &ss);
1882				sigminusset(&l->l_sigmask, &ss);
1883
1884				if ((signo = firstsig(&ss)) == 0) {
1885					/*
1886					 * No signal pending - clear the
1887					 * indicator and bail out.
1888					 */
1889					lwp_lock(l);
1890					l->l_flag &= ~LW_PENDSIG;
1891					lwp_unlock(l);
1892					sp = NULL;
1893					break;
1894				}
1895			}
1896		}
1897
1898		traced = ISSET(p->p_slflag, PSL_TRACED) &&
1899		    !sigismember(&p->p_sigctx.ps_sigpass, signo);
1900
1901		if (sp) {
1902			/* Overwrite process' signal context to correspond
1903			 * to the currently reported LWP.  This is necessary
1904			 * for PT_GET_SIGINFO to report the correct signal when
1905			 * multiple LWPs have pending signals.  We do this only
1906			 * when the signal comes from the queue, for signals
1907			 * created by the debugger we assume it set correct
1908			 * siginfo.
1909			 */
1910			ksiginfo_t *ksi = TAILQ_FIRST(&sp->sp_info);
1911			if (ksi) {
1912				p->p_sigctx.ps_lwp = ksi->ksi_lid;
1913				p->p_sigctx.ps_info = ksi->ksi_info;
1914			} else {
1915				p->p_sigctx.ps_lwp = siglwp;
1916				memset(&p->p_sigctx.ps_info, 0,
1917				    sizeof(p->p_sigctx.ps_info));
1918				p->p_sigctx.ps_info._signo = signo;
1919				p->p_sigctx.ps_info._code = SI_NOINFO;
1920			}
1921		}
1922
1923		/*
1924		 * We should see pending but ignored signals only if
1925		 * we are being traced.
1926		 */
1927		if (sigismember(&p->p_sigctx.ps_sigignore, signo) &&
1928		    !traced) {
1929			/* Discard the signal. */
1930			continue;
1931		}
1932
1933		/*
1934		 * If traced, always stop, and stay stopped until released
1935		 * by the debugger.  If the our parent is our debugger waiting
1936		 * for us and we vforked, don't hang as we could deadlock.
1937		 */
1938		if (traced && signo != SIGKILL &&
1939		    !(ISSET(p->p_lflag, PL_PPWAIT) &&
1940		     (p->p_pptr == p->p_opptr))) {
1941			/*
1942			 * Take the signal, but don't remove it from the
1943			 * siginfo queue, because the debugger can send
1944			 * it later.
1945			 */
1946			if (sp)
1947				sigdelset(&sp->sp_set, signo);
1948			p->p_xsig = signo;
1949
1950			/* Handling of signal trace */
1951			sigswitch(0, signo, false);
1952			mutex_enter(p->p_lock);
1953
1954			/* Check for a signal from the debugger. */
1955			if ((signo = sigchecktrace()) == 0)
1956				continue;
1957
1958			/* Signals from the debugger are "out of band". */
1959			sp = NULL;
1960		}
1961
1962		prop = sigprop[signo];
1963
1964		/*
1965		 * Decide whether the signal should be returned.
1966		 */
1967		switch ((long)SIGACTION(p, signo).sa_handler) {
1968		case (long)SIG_DFL:
1969			/*
1970			 * Don't take default actions on system processes.
1971			 */
1972			if (p->p_pid <= 1) {
1973#ifdef DIAGNOSTIC
1974				/*
1975				 * Are you sure you want to ignore SIGSEGV
1976				 * in init? XXX
1977				 */
1978				printf_nolog("Process (pid %d) got sig %d\n",
1979				    p->p_pid, signo);
1980#endif
1981				continue;
1982			}
1983
1984			/*
1985			 * If there is a pending stop signal to process with
1986			 * default action, stop here, then clear the signal.
1987			 * However, if process is member of an orphaned
1988			 * process group, ignore tty stop signals.
1989			 */
1990			if (prop & SA_STOP) {
1991				/*
1992				 * XXX Don't hold proc_lock for p_lflag,
1993				 * but it's not a big deal.
1994				 */
1995				if ((traced &&
1996				     !(ISSET(p->p_lflag, PL_PPWAIT) &&
1997				     (p->p_pptr == p->p_opptr))) ||
1998				    ((p->p_lflag & PL_ORPHANPG) != 0 &&
1999				    prop & SA_TTYSTOP)) {
2000					/* Ignore the signal. */
2001					continue;
2002				}
2003				/* Take the signal. */
2004				(void)sigget(sp, NULL, signo, NULL);
2005				p->p_xsig = signo;
2006				p->p_sflag &= ~PS_CONTINUED;
2007				signo = 0;
2008				sigswitch(PS_NOCLDSTOP, p->p_xsig, false);
2009				mutex_enter(p->p_lock);
2010			} else if (prop & SA_IGNORE) {
2011				/*
2012				 * Except for SIGCONT, shouldn't get here.
2013				 * Default action is to ignore; drop it.
2014				 */
2015				continue;
2016			}
2017			break;
2018
2019		case (long)SIG_IGN:
2020#ifdef DEBUG_ISSIGNAL
2021			/*
2022			 * Masking above should prevent us ever trying
2023			 * to take action on an ignored signal other
2024			 * than SIGCONT, unless process is traced.
2025			 */
2026			if ((prop & SA_CONT) == 0 && !traced)
2027				printf_nolog("issignal\n");
2028#endif
2029			continue;
2030
2031		default:
2032			/*
2033			 * This signal has an action, let postsig() process
2034			 * it.
2035			 */
2036			break;
2037		}
2038
2039		break;
2040	}
2041
2042	l->l_sigpendset = sp;
2043	return signo;
2044}
2045
2046/*
2047 * Take the action for the specified signal
2048 * from the current set of pending signals.
2049 */
2050void
2051postsig(int signo)
2052{
2053	struct lwp	*l;
2054	struct proc	*p;
2055	struct sigacts	*ps;
2056	sig_t		action;
2057	sigset_t	*returnmask;
2058	ksiginfo_t	ksi;
2059
2060	l = curlwp;
2061	p = l->l_proc;
2062	ps = p->p_sigacts;
2063
2064	KASSERT(mutex_owned(p->p_lock));
2065	KASSERT(signo > 0);
2066
2067	/*
2068	 * Set the new mask value and also defer further occurrences of this
2069	 * signal.
2070	 *
2071	 * Special case: user has done a sigsuspend.  Here the current mask is
2072	 * not of interest, but rather the mask from before the sigsuspend is
2073	 * what we want restored after the signal processing is completed.
2074	 */
2075	if (l->l_sigrestore) {
2076		returnmask = &l->l_sigoldmask;
2077		l->l_sigrestore = 0;
2078	} else
2079		returnmask = &l->l_sigmask;
2080
2081	/*
2082	 * Commit to taking the signal before releasing the mutex.
2083	 */
2084	action = SIGACTION_PS(ps, signo).sa_handler;
2085	l->l_ru.ru_nsignals++;
2086	if (l->l_sigpendset == NULL) {
2087		/* From the debugger */
2088		if (p->p_sigctx.ps_faked &&
2089		    signo == p->p_sigctx.ps_info._signo) {
2090			KSI_INIT(&ksi);
2091			ksi.ksi_info = p->p_sigctx.ps_info;
2092			ksi.ksi_lid = p->p_sigctx.ps_lwp;
2093			p->p_sigctx.ps_faked = false;
2094		} else {
2095			if (!siggetinfo(&l->l_sigpend, &ksi, signo))
2096				(void)siggetinfo(&p->p_sigpend, &ksi, signo);
2097		}
2098	} else
2099		sigget(l->l_sigpendset, &ksi, signo, NULL);
2100
2101	if (ktrpoint(KTR_PSIG)) {
2102		mutex_exit(p->p_lock);
2103		if (p->p_emul->e_ktrpsig)
2104			p->p_emul->e_ktrpsig(signo, action,
2105			    returnmask, &ksi);
2106		else
2107			ktrpsig(signo, action, returnmask, &ksi);
2108		mutex_enter(p->p_lock);
2109	}
2110
2111	SDT_PROBE(proc, kernel, , signal__handle, signo, &ksi, action, 0, 0);
2112
2113	if (action == SIG_DFL) {
2114		/*
2115		 * Default action, where the default is to kill
2116		 * the process.  (Other cases were ignored above.)
2117		 */
2118		sigexit(l, signo);
2119		return;
2120	}
2121
2122	/*
2123	 * If we get here, the signal must be caught.
2124	 */
2125#ifdef DIAGNOSTIC
2126	if (action == SIG_IGN || sigismember(&l->l_sigmask, signo))
2127		panic("postsig action");
2128#endif
2129
2130	kpsendsig(l, &ksi, returnmask);
2131}
2132
2133/*
2134 * sendsig:
2135 *
2136 *	Default signal delivery method for NetBSD.
2137 */
2138void
2139sendsig(const struct ksiginfo *ksi, const sigset_t *mask)
2140{
2141	struct sigacts *sa;
2142	int sig;
2143
2144	sig = ksi->ksi_signo;
2145	sa = curproc->p_sigacts;
2146
2147	switch (sa->sa_sigdesc[sig].sd_vers)  {
2148	case __SIGTRAMP_SIGCODE_VERSION:
2149#ifdef __HAVE_STRUCT_SIGCONTEXT
2150	case __SIGTRAMP_SIGCONTEXT_VERSION_MIN ...
2151	     __SIGTRAMP_SIGCONTEXT_VERSION_MAX:
2152		/* Compat for 1.6 and earlier. */
2153		MODULE_HOOK_CALL_VOID(sendsig_sigcontext_16_hook, (ksi, mask),
2154		    break);
2155		return;
2156#endif /* __HAVE_STRUCT_SIGCONTEXT */
2157	case __SIGTRAMP_SIGINFO_VERSION_MIN ...
2158	     __SIGTRAMP_SIGINFO_VERSION_MAX:
2159		sendsig_siginfo(ksi, mask);
2160		return;
2161	default:
2162		break;
2163	}
2164
2165	printf("sendsig: bad version %d\n", sa->sa_sigdesc[sig].sd_vers);
2166	sigexit(curlwp, SIGILL);
2167}
2168
2169/*
2170 * sendsig_reset:
2171 *
2172 *	Reset the signal action.  Called from emulation specific sendsig()
2173 *	before unlocking to deliver the signal.
2174 */
2175void
2176sendsig_reset(struct lwp *l, int signo)
2177{
2178	struct proc *p = l->l_proc;
2179	struct sigacts *ps = p->p_sigacts;
2180
2181	KASSERT(mutex_owned(p->p_lock));
2182
2183	p->p_sigctx.ps_lwp = 0;
2184	memset(&p->p_sigctx.ps_info, 0, sizeof(p->p_sigctx.ps_info));
2185
2186	mutex_enter(&ps->sa_mutex);
2187	sigplusset(&SIGACTION_PS(ps, signo).sa_mask, &l->l_sigmask);
2188	if (SIGACTION_PS(ps, signo).sa_flags & SA_RESETHAND) {
2189		sigdelset(&p->p_sigctx.ps_sigcatch, signo);
2190		if (signo != SIGCONT && sigprop[signo] & SA_IGNORE)
2191			sigaddset(&p->p_sigctx.ps_sigignore, signo);
2192		SIGACTION_PS(ps, signo).sa_handler = SIG_DFL;
2193	}
2194	mutex_exit(&ps->sa_mutex);
2195}
2196
2197/*
2198 * Kill the current process for stated reason.
2199 */
2200void
2201killproc(struct proc *p, const char *why)
2202{
2203
2204	KASSERT(mutex_owned(&proc_lock));
2205
2206	log(LOG_ERR, "pid %d was killed: %s\n", p->p_pid, why);
2207	uprintf_locked("sorry, pid %d was killed: %s\n", p->p_pid, why);
2208	psignal(p, SIGKILL);
2209}
2210
2211/*
2212 * Force the current process to exit with the specified signal, dumping core
2213 * if appropriate.  We bypass the normal tests for masked and caught
2214 * signals, allowing unrecoverable failures to terminate the process without
2215 * changing signal state.  Mark the accounting record with the signal
2216 * termination.  If dumping core, save the signal number for the debugger.
2217 * Calls exit and does not return.
2218 */
2219void
2220sigexit(struct lwp *l, int signo)
2221{
2222	int exitsig, error, docore;
2223	struct proc *p;
2224	struct lwp *t;
2225
2226	p = l->l_proc;
2227
2228	KASSERT(mutex_owned(p->p_lock));
2229	KASSERT(l->l_blcnt == 0);
2230
2231	/*
2232	 * Don't permit coredump() multiple times in the same process.
2233	 * Call back into sigexit, where we will be suspended until
2234	 * the deed is done.  Note that this is a recursive call, but
2235	 * LW_WCORE will prevent us from coming back this way.
2236	 */
2237	if ((p->p_sflag & PS_WCORE) != 0) {
2238		lwp_lock(l);
2239		l->l_flag |= (LW_WCORE | LW_WEXIT | LW_WSUSPEND);
2240		lwp_need_userret(l);
2241		lwp_unlock(l);
2242		mutex_exit(p->p_lock);
2243		lwp_userret(l);
2244		panic("sigexit 1");
2245		/* NOTREACHED */
2246	}
2247
2248	/* If process is already on the way out, then bail now. */
2249	if ((p->p_sflag & PS_WEXIT) != 0) {
2250		mutex_exit(p->p_lock);
2251		lwp_exit(l);
2252		panic("sigexit 2");
2253		/* NOTREACHED */
2254	}
2255
2256	/*
2257	 * Prepare all other LWPs for exit.  If dumping core, suspend them
2258	 * so that their registers are available long enough to be dumped.
2259 	 */
2260	if ((docore = (sigprop[signo] & SA_CORE)) != 0) {
2261		p->p_sflag |= PS_WCORE;
2262		for (;;) {
2263			LIST_FOREACH(t, &p->p_lwps, l_sibling) {
2264				lwp_lock(t);
2265				if (t == l) {
2266					t->l_flag &=
2267					    ~(LW_WSUSPEND | LW_DBGSUSPEND);
2268					lwp_unlock(t);
2269					continue;
2270				}
2271				t->l_flag |= (LW_WCORE | LW_WEXIT);
2272				lwp_need_userret(t);
2273				lwp_suspend(l, t);
2274			}
2275
2276			if (p->p_nrlwps == 1)
2277				break;
2278
2279			/*
2280			 * Kick any LWPs sitting in lwp_wait1(), and wait
2281			 * for everyone else to stop before proceeding.
2282			 */
2283			p->p_nlwpwait++;
2284			cv_broadcast(&p->p_lwpcv);
2285			cv_wait(&p->p_lwpcv, p->p_lock);
2286			p->p_nlwpwait--;
2287		}
2288	}
2289
2290	exitsig = signo;
2291	p->p_acflag |= AXSIG;
2292	memset(&p->p_sigctx.ps_info, 0, sizeof(p->p_sigctx.ps_info));
2293	p->p_sigctx.ps_info._signo = signo;
2294	p->p_sigctx.ps_info._code = SI_NOINFO;
2295
2296	if (docore) {
2297		mutex_exit(p->p_lock);
2298		MODULE_HOOK_CALL(coredump_hook, (l, NULL), enosys(), error);
2299
2300		if (kern_logsigexit) {
2301			int uid = l->l_cred ?
2302			    (int)kauth_cred_geteuid(l->l_cred) : -1;
2303
2304			if (error)
2305				log(LOG_INFO, lognocoredump, p->p_pid,
2306				    p->p_comm, uid, signo, error);
2307			else
2308				log(LOG_INFO, logcoredump, p->p_pid,
2309				    p->p_comm, uid, signo);
2310		}
2311
2312#ifdef PAX_SEGVGUARD
2313		rw_enter(&exec_lock, RW_WRITER);
2314		pax_segvguard(l, p->p_textvp, p->p_comm, true);
2315		rw_exit(&exec_lock);
2316#endif /* PAX_SEGVGUARD */
2317
2318		/* Acquire the sched state mutex.  exit1() will release it. */
2319		mutex_enter(p->p_lock);
2320		if (error == 0)
2321			p->p_sflag |= PS_COREDUMP;
2322	}
2323
2324	/* No longer dumping core. */
2325	p->p_sflag &= ~PS_WCORE;
2326
2327	exit1(l, 0, exitsig);
2328	/* NOTREACHED */
2329}
2330
2331/*
2332 * Since the "real" code may (or may not) be present in loadable module,
2333 * we provide routines here which calls the module hooks.
2334 */
2335
2336int
2337coredump_netbsd(struct lwp *l, struct coredump_iostate *iocookie)
2338{
2339
2340	int retval;
2341
2342	MODULE_HOOK_CALL(coredump_netbsd_hook, (l, iocookie), ENOSYS, retval);
2343	return retval;
2344}
2345
2346int
2347coredump_netbsd32(struct lwp *l, struct coredump_iostate *iocookie)
2348{
2349
2350	int retval;
2351
2352	MODULE_HOOK_CALL(coredump_netbsd32_hook, (l, iocookie), ENOSYS, retval);
2353	return retval;
2354}
2355
2356int
2357coredump_elf32(struct lwp *l, struct coredump_iostate *iocookie)
2358{
2359	int retval;
2360
2361	MODULE_HOOK_CALL(coredump_elf32_hook, (l, iocookie), ENOSYS, retval);
2362	return retval;
2363}
2364
2365int
2366coredump_elf64(struct lwp *l, struct coredump_iostate *iocookie)
2367{
2368	int retval;
2369
2370	MODULE_HOOK_CALL(coredump_elf64_hook, (l, iocookie), ENOSYS, retval);
2371	return retval;
2372}
2373
2374/*
2375 * Put process 'p' into the stopped state and optionally, notify the parent.
2376 */
2377void
2378proc_stop(struct proc *p, int signo)
2379{
2380	struct lwp *l;
2381
2382	KASSERT(mutex_owned(p->p_lock));
2383
2384	/*
2385	 * First off, set the stopping indicator and bring all sleeping
2386	 * LWPs to a halt so they are included in p->p_nrlwps.  We mustn't
2387	 * unlock between here and the p->p_nrlwps check below.
2388	 */
2389	p->p_sflag |= PS_STOPPING;
2390	membar_producer();
2391
2392	proc_stop_lwps(p);
2393
2394	/*
2395	 * If there are no LWPs available to take the signal, then we
2396	 * signal the parent process immediately.  Otherwise, the last
2397	 * LWP to stop will take care of it.
2398	 */
2399
2400	if (p->p_nrlwps == 0) {
2401		proc_stop_done(p, PS_NOCLDSTOP);
2402	} else {
2403		/*
2404		 * Have the remaining LWPs come to a halt, and trigger
2405		 * proc_stop_callout() to ensure that they do.
2406		 */
2407		LIST_FOREACH(l, &p->p_lwps, l_sibling) {
2408			sigpost(l, SIG_DFL, SA_STOP, signo);
2409		}
2410		callout_schedule(&proc_stop_ch, 1);
2411	}
2412}
2413
2414/*
2415 * When stopping a process, we do not immediately set sleeping LWPs stopped,
2416 * but wait for them to come to a halt at the kernel-user boundary.  This is
2417 * to allow LWPs to release any locks that they may hold before stopping.
2418 *
2419 * Non-interruptable sleeps can be long, and there is the potential for an
2420 * LWP to begin sleeping interruptably soon after the process has been set
2421 * stopping (PS_STOPPING).  These LWPs will not notice that the process is
2422 * stopping, and so complete halt of the process and the return of status
2423 * information to the parent could be delayed indefinitely.
2424 *
2425 * To handle this race, proc_stop_callout() runs once per tick while there
2426 * are stopping processes in the system.  It sets LWPs that are sleeping
2427 * interruptably into the LSSTOP state.
2428 *
2429 * Note that we are not concerned about keeping all LWPs stopped while the
2430 * process is stopped: stopped LWPs can awaken briefly to handle signals.
2431 * What we do need to ensure is that all LWPs in a stopping process have
2432 * stopped at least once, so that notification can be sent to the parent
2433 * process.
2434 */
2435static void
2436proc_stop_callout(void *cookie)
2437{
2438	bool more, restart;
2439	struct proc *p;
2440
2441	(void)cookie;
2442
2443	do {
2444		restart = false;
2445		more = false;
2446
2447		mutex_enter(&proc_lock);
2448		PROCLIST_FOREACH(p, &allproc) {
2449			mutex_enter(p->p_lock);
2450
2451			if ((p->p_sflag & PS_STOPPING) == 0) {
2452				mutex_exit(p->p_lock);
2453				continue;
2454			}
2455
2456			/* Stop any LWPs sleeping interruptably. */
2457			proc_stop_lwps(p);
2458			if (p->p_nrlwps == 0) {
2459				/*
2460				 * We brought the process to a halt.
2461				 * Mark it as stopped and notify the
2462				 * parent.
2463				 *
2464				 * Note that proc_stop_done() will
2465				 * drop p->p_lock briefly.
2466				 * Arrange to restart and check
2467				 * all processes again.
2468				 */
2469				restart = true;
2470				proc_stop_done(p, PS_NOCLDSTOP);
2471			} else
2472				more = true;
2473
2474			mutex_exit(p->p_lock);
2475			if (restart)
2476				break;
2477		}
2478		mutex_exit(&proc_lock);
2479	} while (restart);
2480
2481	/*
2482	 * If we noted processes that are stopping but still have
2483	 * running LWPs, then arrange to check again in 1 tick.
2484	 */
2485	if (more)
2486		callout_schedule(&proc_stop_ch, 1);
2487}
2488
2489/*
2490 * Given a process in state SSTOP, set the state back to SACTIVE and
2491 * move LSSTOP'd LWPs to LSSLEEP or make them runnable.
2492 */
2493void
2494proc_unstop(struct proc *p)
2495{
2496	struct lwp *l;
2497	int sig;
2498
2499	KASSERT(mutex_owned(&proc_lock));
2500	KASSERT(mutex_owned(p->p_lock));
2501
2502	p->p_stat = SACTIVE;
2503	p->p_sflag &= ~PS_STOPPING;
2504	sig = p->p_xsig;
2505
2506	if (!p->p_waited)
2507		p->p_pptr->p_nstopchild--;
2508
2509	LIST_FOREACH(l, &p->p_lwps, l_sibling) {
2510		lwp_lock(l);
2511		if (l->l_stat != LSSTOP || (l->l_flag & LW_DBGSUSPEND) != 0) {
2512			lwp_unlock(l);
2513			continue;
2514		}
2515		if (l->l_wchan == NULL) {
2516			setrunnable(l);
2517			continue;
2518		}
2519		if (sig && (l->l_flag & LW_SINTR) != 0) {
2520			setrunnable(l);
2521			sig = 0;
2522		} else {
2523			l->l_stat = LSSLEEP;
2524			p->p_nrlwps++;
2525			lwp_unlock(l);
2526		}
2527	}
2528}
2529
2530void
2531proc_stoptrace(int trapno, int sysnum, const register_t args[],
2532               const register_t *ret, int error)
2533{
2534	struct lwp *l = curlwp;
2535	struct proc *p = l->l_proc;
2536	struct sigacts *ps;
2537	sigset_t *mask;
2538	sig_t action;
2539	ksiginfo_t ksi;
2540	size_t i, sy_narg;
2541	const int signo = SIGTRAP;
2542
2543	KASSERT((trapno == TRAP_SCE) || (trapno == TRAP_SCX));
2544	KASSERT(p->p_pptr != initproc);
2545	KASSERT(ISSET(p->p_slflag, PSL_TRACED));
2546	KASSERT(ISSET(p->p_slflag, PSL_SYSCALL));
2547
2548	sy_narg = p->p_emul->e_sysent[sysnum].sy_narg;
2549
2550	KSI_INIT_TRAP(&ksi);
2551	ksi.ksi_lid = l->l_lid;
2552	ksi.ksi_signo = signo;
2553	ksi.ksi_code = trapno;
2554
2555	ksi.ksi_sysnum = sysnum;
2556	if (trapno == TRAP_SCE) {
2557		ksi.ksi_retval[0] = 0;
2558		ksi.ksi_retval[1] = 0;
2559		ksi.ksi_error = 0;
2560	} else {
2561		ksi.ksi_retval[0] = ret[0];
2562		ksi.ksi_retval[1] = ret[1];
2563		ksi.ksi_error = error;
2564	}
2565
2566	memset(ksi.ksi_args, 0, sizeof(ksi.ksi_args));
2567
2568	for (i = 0; i < sy_narg; i++)
2569		ksi.ksi_args[i] = args[i];
2570
2571	mutex_enter(p->p_lock);
2572
2573repeat:
2574	/*
2575	 * If we are exiting, demise now.
2576	 *
2577	 * This avoids notifying tracer and deadlocking.
2578	 */
2579	if (__predict_false(ISSET(p->p_sflag, PS_WEXIT))) {
2580		mutex_exit(p->p_lock);
2581		lwp_exit(l);
2582		panic("proc_stoptrace");
2583		/* NOTREACHED */
2584	}
2585
2586	/*
2587	 * If there's a pending SIGKILL process it immediately.
2588	 */
2589	if (p->p_xsig == SIGKILL ||
2590	    sigismember(&p->p_sigpend.sp_set, SIGKILL)) {
2591		mutex_exit(p->p_lock);
2592		return;
2593	}
2594
2595	/*
2596	 * If we are no longer traced, abandon this event signal.
2597	 *
2598	 * This avoids killing a process after detaching the debugger.
2599	 */
2600	if (__predict_false(!ISSET(p->p_slflag, PSL_TRACED))) {
2601		mutex_exit(p->p_lock);
2602		return;
2603	}
2604
2605	/*
2606	 * The process is already stopping.
2607	 */
2608	if ((p->p_sflag & PS_STOPPING) != 0) {
2609		sigswitch_unlock_and_switch_away(l);
2610		mutex_enter(p->p_lock);
2611		goto repeat;
2612	}
2613
2614	/* Needed for ktrace */
2615	ps = p->p_sigacts;
2616	action = SIGACTION_PS(ps, signo).sa_handler;
2617	mask = &l->l_sigmask;
2618
2619	p->p_xsig = signo;
2620	p->p_sigctx.ps_lwp = ksi.ksi_lid;
2621	p->p_sigctx.ps_info = ksi.ksi_info;
2622	sigswitch(0, signo, false);
2623
2624	if (ktrpoint(KTR_PSIG)) {
2625		if (p->p_emul->e_ktrpsig)
2626			p->p_emul->e_ktrpsig(signo, action, mask, &ksi);
2627		else
2628			ktrpsig(signo, action, mask, &ksi);
2629	}
2630}
2631
2632static int
2633filt_sigattach(struct knote *kn)
2634{
2635	struct proc *p = curproc;
2636
2637	kn->kn_obj = p;
2638	kn->kn_flags |= EV_CLEAR;	/* automatically set */
2639
2640	mutex_enter(p->p_lock);
2641	klist_insert(&p->p_klist, kn);
2642	mutex_exit(p->p_lock);
2643
2644	return 0;
2645}
2646
2647static void
2648filt_sigdetach(struct knote *kn)
2649{
2650	struct proc *p = kn->kn_obj;
2651
2652	mutex_enter(p->p_lock);
2653	klist_remove(&p->p_klist, kn);
2654	mutex_exit(p->p_lock);
2655}
2656
2657/*
2658 * Signal knotes are shared with proc knotes, so we apply a mask to
2659 * the hint in order to differentiate them from process hints.  This
2660 * could be avoided by using a signal-specific knote list, but probably
2661 * isn't worth the trouble.
2662 */
2663static int
2664filt_signal(struct knote *kn, long hint)
2665{
2666
2667	if (hint & NOTE_SIGNAL) {
2668		hint &= ~NOTE_SIGNAL;
2669
2670		if (kn->kn_id == hint)
2671			kn->kn_data++;
2672	}
2673	return (kn->kn_data != 0);
2674}
2675
2676const struct filterops sig_filtops = {
2677	.f_flags = FILTEROP_MPSAFE,
2678	.f_attach = filt_sigattach,
2679	.f_detach = filt_sigdetach,
2680	.f_event = filt_signal,
2681};
2682