kern_sig.c revision 331643
1/*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 *	The Regents of the University of California.  All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 *    notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 *    notice, this list of conditions and the following disclaimer in the
19 *    documentation and/or other materials provided with the distribution.
20 * 4. Neither the name of the University nor the names of its contributors
21 *    may be used to endorse or promote products derived from this software
22 *    without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
37 */
38
39#include <sys/cdefs.h>
40__FBSDID("$FreeBSD: stable/11/sys/kern/kern_sig.c 331643 2018-03-27 18:52:27Z dim $");
41
42#include "opt_compat.h"
43#include "opt_gzio.h"
44#include "opt_ktrace.h"
45
46#include <sys/param.h>
47#include <sys/ctype.h>
48#include <sys/systm.h>
49#include <sys/signalvar.h>
50#include <sys/vnode.h>
51#include <sys/acct.h>
52#include <sys/bus.h>
53#include <sys/capsicum.h>
54#include <sys/condvar.h>
55#include <sys/event.h>
56#include <sys/fcntl.h>
57#include <sys/imgact.h>
58#include <sys/kernel.h>
59#include <sys/ktr.h>
60#include <sys/ktrace.h>
61#include <sys/lock.h>
62#include <sys/malloc.h>
63#include <sys/mutex.h>
64#include <sys/refcount.h>
65#include <sys/namei.h>
66#include <sys/proc.h>
67#include <sys/procdesc.h>
68#include <sys/posix4.h>
69#include <sys/pioctl.h>
70#include <sys/racct.h>
71#include <sys/resourcevar.h>
72#include <sys/sdt.h>
73#include <sys/sbuf.h>
74#include <sys/sleepqueue.h>
75#include <sys/smp.h>
76#include <sys/stat.h>
77#include <sys/sx.h>
78#include <sys/syscallsubr.h>
79#include <sys/sysctl.h>
80#include <sys/sysent.h>
81#include <sys/syslog.h>
82#include <sys/sysproto.h>
83#include <sys/timers.h>
84#include <sys/unistd.h>
85#include <sys/wait.h>
86#include <vm/vm.h>
87#include <vm/vm_extern.h>
88#include <vm/uma.h>
89
90#include <sys/jail.h>
91
92#include <machine/cpu.h>
93
94#include <security/audit/audit.h>
95
96#define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
97
98SDT_PROVIDER_DECLARE(proc);
99SDT_PROBE_DEFINE3(proc, , , signal__send,
100    "struct thread *", "struct proc *", "int");
101SDT_PROBE_DEFINE2(proc, , , signal__clear,
102    "int", "ksiginfo_t *");
103SDT_PROBE_DEFINE3(proc, , , signal__discard,
104    "struct thread *", "struct proc *", "int");
105
106static int	coredump(struct thread *);
107static int	killpg1(struct thread *td, int sig, int pgid, int all,
108		    ksiginfo_t *ksi);
109static int	issignal(struct thread *td);
110static int	sigprop(int sig);
111static void	tdsigwakeup(struct thread *, int, sig_t, int);
112static int	sig_suspend_threads(struct thread *, struct proc *, int);
113static int	filt_sigattach(struct knote *kn);
114static void	filt_sigdetach(struct knote *kn);
115static int	filt_signal(struct knote *kn, long hint);
116static struct thread *sigtd(struct proc *p, int sig, int prop);
117static void	sigqueue_start(void);
118
119static uma_zone_t	ksiginfo_zone = NULL;
120struct filterops sig_filtops = {
121	.f_isfd = 0,
122	.f_attach = filt_sigattach,
123	.f_detach = filt_sigdetach,
124	.f_event = filt_signal,
125};
126
127static int	kern_logsigexit = 1;
128SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
129    &kern_logsigexit, 0,
130    "Log processes quitting on abnormal signals to syslog(3)");
131
132static int	kern_forcesigexit = 1;
133SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
134    &kern_forcesigexit, 0, "Force trap signal to be handled");
135
136static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW, 0,
137    "POSIX real time signal");
138
139static int	max_pending_per_proc = 128;
140SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
141    &max_pending_per_proc, 0, "Max pending signals per proc");
142
143static int	preallocate_siginfo = 1024;
144SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
145    &preallocate_siginfo, 0, "Preallocated signal memory size");
146
147static int	signal_overflow = 0;
148SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
149    &signal_overflow, 0, "Number of signals overflew");
150
151static int	signal_alloc_fail = 0;
152SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
153    &signal_alloc_fail, 0, "signals failed to be allocated");
154
155static int	kern_lognosys = 0;
156SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
157    "Log invalid syscalls");
158
159SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
160
161/*
162 * Policy -- Can ucred cr1 send SIGIO to process cr2?
163 * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
164 * in the right situations.
165 */
166#define CANSIGIO(cr1, cr2) \
167	((cr1)->cr_uid == 0 || \
168	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
169	    (cr1)->cr_uid == (cr2)->cr_ruid || \
170	    (cr1)->cr_ruid == (cr2)->cr_uid || \
171	    (cr1)->cr_uid == (cr2)->cr_uid)
172
173static int	sugid_coredump;
174SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
175    &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
176
177static int	capmode_coredump;
178SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
179    &capmode_coredump, 0, "Allow processes in capability mode to dump core");
180
181static int	do_coredump = 1;
182SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
183	&do_coredump, 0, "Enable/Disable coredumps");
184
185static int	set_core_nodump_flag = 0;
186SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
187	0, "Enable setting the NODUMP flag on coredump files");
188
189static int	coredump_devctl = 0;
190SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
191	0, "Generate a devctl notification when processes coredump");
192
193/*
194 * Signal properties and actions.
195 * The array below categorizes the signals and their default actions
196 * according to the following properties:
197 */
198#define	SA_KILL		0x01		/* terminates process by default */
199#define	SA_CORE		0x02		/* ditto and coredumps */
200#define	SA_STOP		0x04		/* suspend process */
201#define	SA_TTYSTOP	0x08		/* ditto, from tty */
202#define	SA_IGNORE	0x10		/* ignore by default */
203#define	SA_CONT		0x20		/* continue if suspended */
204#define	SA_CANTMASK	0x40		/* non-maskable, catchable */
205
206static int sigproptbl[NSIG] = {
207	SA_KILL,			/* SIGHUP */
208	SA_KILL,			/* SIGINT */
209	SA_KILL|SA_CORE,		/* SIGQUIT */
210	SA_KILL|SA_CORE,		/* SIGILL */
211	SA_KILL|SA_CORE,		/* SIGTRAP */
212	SA_KILL|SA_CORE,		/* SIGABRT */
213	SA_KILL|SA_CORE,		/* SIGEMT */
214	SA_KILL|SA_CORE,		/* SIGFPE */
215	SA_KILL,			/* SIGKILL */
216	SA_KILL|SA_CORE,		/* SIGBUS */
217	SA_KILL|SA_CORE,		/* SIGSEGV */
218	SA_KILL|SA_CORE,		/* SIGSYS */
219	SA_KILL,			/* SIGPIPE */
220	SA_KILL,			/* SIGALRM */
221	SA_KILL,			/* SIGTERM */
222	SA_IGNORE,			/* SIGURG */
223	SA_STOP,			/* SIGSTOP */
224	SA_STOP|SA_TTYSTOP,		/* SIGTSTP */
225	SA_IGNORE|SA_CONT,		/* SIGCONT */
226	SA_IGNORE,			/* SIGCHLD */
227	SA_STOP|SA_TTYSTOP,		/* SIGTTIN */
228	SA_STOP|SA_TTYSTOP,		/* SIGTTOU */
229	SA_IGNORE,			/* SIGIO */
230	SA_KILL,			/* SIGXCPU */
231	SA_KILL,			/* SIGXFSZ */
232	SA_KILL,			/* SIGVTALRM */
233	SA_KILL,			/* SIGPROF */
234	SA_IGNORE,			/* SIGWINCH  */
235	SA_IGNORE,			/* SIGINFO */
236	SA_KILL,			/* SIGUSR1 */
237	SA_KILL,			/* SIGUSR2 */
238};
239
240static void reschedule_signals(struct proc *p, sigset_t block, int flags);
241
242static void
243sigqueue_start(void)
244{
245	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
246		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
247	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
248	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
249	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
250	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
251}
252
253ksiginfo_t *
254ksiginfo_alloc(int wait)
255{
256	int flags;
257
258	flags = M_ZERO;
259	if (! wait)
260		flags |= M_NOWAIT;
261	if (ksiginfo_zone != NULL)
262		return ((ksiginfo_t *)uma_zalloc(ksiginfo_zone, flags));
263	return (NULL);
264}
265
266void
267ksiginfo_free(ksiginfo_t *ksi)
268{
269	uma_zfree(ksiginfo_zone, ksi);
270}
271
272static __inline int
273ksiginfo_tryfree(ksiginfo_t *ksi)
274{
275	if (!(ksi->ksi_flags & KSI_EXT)) {
276		uma_zfree(ksiginfo_zone, ksi);
277		return (1);
278	}
279	return (0);
280}
281
282void
283sigqueue_init(sigqueue_t *list, struct proc *p)
284{
285	SIGEMPTYSET(list->sq_signals);
286	SIGEMPTYSET(list->sq_kill);
287	SIGEMPTYSET(list->sq_ptrace);
288	TAILQ_INIT(&list->sq_list);
289	list->sq_proc = p;
290	list->sq_flags = SQ_INIT;
291}
292
293/*
294 * Get a signal's ksiginfo.
295 * Return:
296 *	0	-	signal not found
297 *	others	-	signal number
298 */
299static int
300sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
301{
302	struct proc *p = sq->sq_proc;
303	struct ksiginfo *ksi, *next;
304	int count = 0;
305
306	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
307
308	if (!SIGISMEMBER(sq->sq_signals, signo))
309		return (0);
310
311	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
312		count++;
313		SIGDELSET(sq->sq_ptrace, signo);
314		si->ksi_flags |= KSI_PTRACE;
315	}
316	if (SIGISMEMBER(sq->sq_kill, signo)) {
317		count++;
318		if (count == 1)
319			SIGDELSET(sq->sq_kill, signo);
320	}
321
322	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
323		if (ksi->ksi_signo == signo) {
324			if (count == 0) {
325				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
326				ksi->ksi_sigq = NULL;
327				ksiginfo_copy(ksi, si);
328				if (ksiginfo_tryfree(ksi) && p != NULL)
329					p->p_pendingcnt--;
330			}
331			if (++count > 1)
332				break;
333		}
334	}
335
336	if (count <= 1)
337		SIGDELSET(sq->sq_signals, signo);
338	si->ksi_signo = signo;
339	return (signo);
340}
341
342void
343sigqueue_take(ksiginfo_t *ksi)
344{
345	struct ksiginfo *kp;
346	struct proc	*p;
347	sigqueue_t	*sq;
348
349	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
350		return;
351
352	p = sq->sq_proc;
353	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
354	ksi->ksi_sigq = NULL;
355	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
356		p->p_pendingcnt--;
357
358	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
359	     kp = TAILQ_NEXT(kp, ksi_link)) {
360		if (kp->ksi_signo == ksi->ksi_signo)
361			break;
362	}
363	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
364	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
365		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
366}
367
368static int
369sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
370{
371	struct proc *p = sq->sq_proc;
372	struct ksiginfo *ksi;
373	int ret = 0;
374
375	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
376
377	/*
378	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
379	 * for these signals.
380	 */
381	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
382		SIGADDSET(sq->sq_kill, signo);
383		goto out_set_bit;
384	}
385
386	/* directly insert the ksi, don't copy it */
387	if (si->ksi_flags & KSI_INS) {
388		if (si->ksi_flags & KSI_HEAD)
389			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
390		else
391			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
392		si->ksi_sigq = sq;
393		goto out_set_bit;
394	}
395
396	if (__predict_false(ksiginfo_zone == NULL)) {
397		SIGADDSET(sq->sq_kill, signo);
398		goto out_set_bit;
399	}
400
401	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
402		signal_overflow++;
403		ret = EAGAIN;
404	} else if ((ksi = ksiginfo_alloc(0)) == NULL) {
405		signal_alloc_fail++;
406		ret = EAGAIN;
407	} else {
408		if (p != NULL)
409			p->p_pendingcnt++;
410		ksiginfo_copy(si, ksi);
411		ksi->ksi_signo = signo;
412		if (si->ksi_flags & KSI_HEAD)
413			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
414		else
415			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
416		ksi->ksi_sigq = sq;
417	}
418
419	if (ret != 0) {
420		if ((si->ksi_flags & KSI_PTRACE) != 0) {
421			SIGADDSET(sq->sq_ptrace, signo);
422			ret = 0;
423			goto out_set_bit;
424		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
425		    (si->ksi_flags & KSI_SIGQ) == 0) {
426			SIGADDSET(sq->sq_kill, signo);
427			ret = 0;
428			goto out_set_bit;
429		}
430		return (ret);
431	}
432
433out_set_bit:
434	SIGADDSET(sq->sq_signals, signo);
435	return (ret);
436}
437
438void
439sigqueue_flush(sigqueue_t *sq)
440{
441	struct proc *p = sq->sq_proc;
442	ksiginfo_t *ksi;
443
444	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
445
446	if (p != NULL)
447		PROC_LOCK_ASSERT(p, MA_OWNED);
448
449	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
450		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
451		ksi->ksi_sigq = NULL;
452		if (ksiginfo_tryfree(ksi) && p != NULL)
453			p->p_pendingcnt--;
454	}
455
456	SIGEMPTYSET(sq->sq_signals);
457	SIGEMPTYSET(sq->sq_kill);
458	SIGEMPTYSET(sq->sq_ptrace);
459}
460
461static void
462sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
463{
464	sigset_t tmp;
465	struct proc *p1, *p2;
466	ksiginfo_t *ksi, *next;
467
468	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
469	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
470	p1 = src->sq_proc;
471	p2 = dst->sq_proc;
472	/* Move siginfo to target list */
473	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
474		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
475			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
476			if (p1 != NULL)
477				p1->p_pendingcnt--;
478			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
479			ksi->ksi_sigq = dst;
480			if (p2 != NULL)
481				p2->p_pendingcnt++;
482		}
483	}
484
485	/* Move pending bits to target list */
486	tmp = src->sq_kill;
487	SIGSETAND(tmp, *set);
488	SIGSETOR(dst->sq_kill, tmp);
489	SIGSETNAND(src->sq_kill, tmp);
490
491	tmp = src->sq_ptrace;
492	SIGSETAND(tmp, *set);
493	SIGSETOR(dst->sq_ptrace, tmp);
494	SIGSETNAND(src->sq_ptrace, tmp);
495
496	tmp = src->sq_signals;
497	SIGSETAND(tmp, *set);
498	SIGSETOR(dst->sq_signals, tmp);
499	SIGSETNAND(src->sq_signals, tmp);
500}
501
502#if 0
503static void
504sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
505{
506	sigset_t set;
507
508	SIGEMPTYSET(set);
509	SIGADDSET(set, signo);
510	sigqueue_move_set(src, dst, &set);
511}
512#endif
513
514static void
515sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
516{
517	struct proc *p = sq->sq_proc;
518	ksiginfo_t *ksi, *next;
519
520	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
521
522	/* Remove siginfo queue */
523	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
524		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
525			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
526			ksi->ksi_sigq = NULL;
527			if (ksiginfo_tryfree(ksi) && p != NULL)
528				p->p_pendingcnt--;
529		}
530	}
531	SIGSETNAND(sq->sq_kill, *set);
532	SIGSETNAND(sq->sq_ptrace, *set);
533	SIGSETNAND(sq->sq_signals, *set);
534}
535
536void
537sigqueue_delete(sigqueue_t *sq, int signo)
538{
539	sigset_t set;
540
541	SIGEMPTYSET(set);
542	SIGADDSET(set, signo);
543	sigqueue_delete_set(sq, &set);
544}
545
546/* Remove a set of signals for a process */
547static void
548sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
549{
550	sigqueue_t worklist;
551	struct thread *td0;
552
553	PROC_LOCK_ASSERT(p, MA_OWNED);
554
555	sigqueue_init(&worklist, NULL);
556	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
557
558	FOREACH_THREAD_IN_PROC(p, td0)
559		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
560
561	sigqueue_flush(&worklist);
562}
563
564void
565sigqueue_delete_proc(struct proc *p, int signo)
566{
567	sigset_t set;
568
569	SIGEMPTYSET(set);
570	SIGADDSET(set, signo);
571	sigqueue_delete_set_proc(p, &set);
572}
573
574static void
575sigqueue_delete_stopmask_proc(struct proc *p)
576{
577	sigset_t set;
578
579	SIGEMPTYSET(set);
580	SIGADDSET(set, SIGSTOP);
581	SIGADDSET(set, SIGTSTP);
582	SIGADDSET(set, SIGTTIN);
583	SIGADDSET(set, SIGTTOU);
584	sigqueue_delete_set_proc(p, &set);
585}
586
587/*
588 * Determine signal that should be delivered to thread td, the current
589 * thread, 0 if none.  If there is a pending stop signal with default
590 * action, the process stops in issignal().
591 */
592int
593cursig(struct thread *td)
594{
595	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
596	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
597	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
598	return (SIGPENDING(td) ? issignal(td) : 0);
599}
600
601/*
602 * Arrange for ast() to handle unmasked pending signals on return to user
603 * mode.  This must be called whenever a signal is added to td_sigqueue or
604 * unmasked in td_sigmask.
605 */
606void
607signotify(struct thread *td)
608{
609	struct proc *p;
610
611	p = td->td_proc;
612
613	PROC_LOCK_ASSERT(p, MA_OWNED);
614
615	if (SIGPENDING(td)) {
616		thread_lock(td);
617		td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING;
618		thread_unlock(td);
619	}
620}
621
622int
623sigonstack(size_t sp)
624{
625	struct thread *td = curthread;
626
627	return ((td->td_pflags & TDP_ALTSTACK) ?
628#if defined(COMPAT_43)
629	    ((td->td_sigstk.ss_size == 0) ?
630		(td->td_sigstk.ss_flags & SS_ONSTACK) :
631		((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size))
632#else
633	    ((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size)
634#endif
635	    : 0);
636}
637
638static __inline int
639sigprop(int sig)
640{
641
642	if (sig > 0 && sig < NSIG)
643		return (sigproptbl[_SIG_IDX(sig)]);
644	return (0);
645}
646
647int
648sig_ffs(sigset_t *set)
649{
650	int i;
651
652	for (i = 0; i < _SIG_WORDS; i++)
653		if (set->__bits[i])
654			return (ffs(set->__bits[i]) + (i * 32));
655	return (0);
656}
657
658static bool
659sigact_flag_test(const struct sigaction *act, int flag)
660{
661
662	/*
663	 * SA_SIGINFO is reset when signal disposition is set to
664	 * ignore or default.  Other flags are kept according to user
665	 * settings.
666	 */
667	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
668	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
669	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
670}
671
672/*
673 * kern_sigaction
674 * sigaction
675 * freebsd4_sigaction
676 * osigaction
677 */
678int
679kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
680    struct sigaction *oact, int flags)
681{
682	struct sigacts *ps;
683	struct proc *p = td->td_proc;
684
685	if (!_SIG_VALID(sig))
686		return (EINVAL);
687	if (act != NULL && act->sa_handler != SIG_DFL &&
688	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
689	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
690	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
691		return (EINVAL);
692
693	PROC_LOCK(p);
694	ps = p->p_sigacts;
695	mtx_lock(&ps->ps_mtx);
696	if (oact) {
697		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
698		oact->sa_flags = 0;
699		if (SIGISMEMBER(ps->ps_sigonstack, sig))
700			oact->sa_flags |= SA_ONSTACK;
701		if (!SIGISMEMBER(ps->ps_sigintr, sig))
702			oact->sa_flags |= SA_RESTART;
703		if (SIGISMEMBER(ps->ps_sigreset, sig))
704			oact->sa_flags |= SA_RESETHAND;
705		if (SIGISMEMBER(ps->ps_signodefer, sig))
706			oact->sa_flags |= SA_NODEFER;
707		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
708			oact->sa_flags |= SA_SIGINFO;
709			oact->sa_sigaction =
710			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
711		} else
712			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
713		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
714			oact->sa_flags |= SA_NOCLDSTOP;
715		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
716			oact->sa_flags |= SA_NOCLDWAIT;
717	}
718	if (act) {
719		if ((sig == SIGKILL || sig == SIGSTOP) &&
720		    act->sa_handler != SIG_DFL) {
721			mtx_unlock(&ps->ps_mtx);
722			PROC_UNLOCK(p);
723			return (EINVAL);
724		}
725
726		/*
727		 * Change setting atomically.
728		 */
729
730		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
731		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
732		if (sigact_flag_test(act, SA_SIGINFO)) {
733			ps->ps_sigact[_SIG_IDX(sig)] =
734			    (__sighandler_t *)act->sa_sigaction;
735			SIGADDSET(ps->ps_siginfo, sig);
736		} else {
737			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
738			SIGDELSET(ps->ps_siginfo, sig);
739		}
740		if (!sigact_flag_test(act, SA_RESTART))
741			SIGADDSET(ps->ps_sigintr, sig);
742		else
743			SIGDELSET(ps->ps_sigintr, sig);
744		if (sigact_flag_test(act, SA_ONSTACK))
745			SIGADDSET(ps->ps_sigonstack, sig);
746		else
747			SIGDELSET(ps->ps_sigonstack, sig);
748		if (sigact_flag_test(act, SA_RESETHAND))
749			SIGADDSET(ps->ps_sigreset, sig);
750		else
751			SIGDELSET(ps->ps_sigreset, sig);
752		if (sigact_flag_test(act, SA_NODEFER))
753			SIGADDSET(ps->ps_signodefer, sig);
754		else
755			SIGDELSET(ps->ps_signodefer, sig);
756		if (sig == SIGCHLD) {
757			if (act->sa_flags & SA_NOCLDSTOP)
758				ps->ps_flag |= PS_NOCLDSTOP;
759			else
760				ps->ps_flag &= ~PS_NOCLDSTOP;
761			if (act->sa_flags & SA_NOCLDWAIT) {
762				/*
763				 * Paranoia: since SA_NOCLDWAIT is implemented
764				 * by reparenting the dying child to PID 1 (and
765				 * trust it to reap the zombie), PID 1 itself
766				 * is forbidden to set SA_NOCLDWAIT.
767				 */
768				if (p->p_pid == 1)
769					ps->ps_flag &= ~PS_NOCLDWAIT;
770				else
771					ps->ps_flag |= PS_NOCLDWAIT;
772			} else
773				ps->ps_flag &= ~PS_NOCLDWAIT;
774			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
775				ps->ps_flag |= PS_CLDSIGIGN;
776			else
777				ps->ps_flag &= ~PS_CLDSIGIGN;
778		}
779		/*
780		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
781		 * and for signals set to SIG_DFL where the default is to
782		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
783		 * have to restart the process.
784		 */
785		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
786		    (sigprop(sig) & SA_IGNORE &&
787		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
788			/* never to be seen again */
789			sigqueue_delete_proc(p, sig);
790			if (sig != SIGCONT)
791				/* easier in psignal */
792				SIGADDSET(ps->ps_sigignore, sig);
793			SIGDELSET(ps->ps_sigcatch, sig);
794		} else {
795			SIGDELSET(ps->ps_sigignore, sig);
796			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
797				SIGDELSET(ps->ps_sigcatch, sig);
798			else
799				SIGADDSET(ps->ps_sigcatch, sig);
800		}
801#ifdef COMPAT_FREEBSD4
802		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
803		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
804		    (flags & KSA_FREEBSD4) == 0)
805			SIGDELSET(ps->ps_freebsd4, sig);
806		else
807			SIGADDSET(ps->ps_freebsd4, sig);
808#endif
809#ifdef COMPAT_43
810		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
811		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
812		    (flags & KSA_OSIGSET) == 0)
813			SIGDELSET(ps->ps_osigset, sig);
814		else
815			SIGADDSET(ps->ps_osigset, sig);
816#endif
817	}
818	mtx_unlock(&ps->ps_mtx);
819	PROC_UNLOCK(p);
820	return (0);
821}
822
823#ifndef _SYS_SYSPROTO_H_
824struct sigaction_args {
825	int	sig;
826	struct	sigaction *act;
827	struct	sigaction *oact;
828};
829#endif
830int
831sys_sigaction(struct thread *td, struct sigaction_args *uap)
832{
833	struct sigaction act, oact;
834	struct sigaction *actp, *oactp;
835	int error;
836
837	actp = (uap->act != NULL) ? &act : NULL;
838	oactp = (uap->oact != NULL) ? &oact : NULL;
839	if (actp) {
840		error = copyin(uap->act, actp, sizeof(act));
841		if (error)
842			return (error);
843	}
844	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
845	if (oactp && !error)
846		error = copyout(oactp, uap->oact, sizeof(oact));
847	return (error);
848}
849
850#ifdef COMPAT_FREEBSD4
851#ifndef _SYS_SYSPROTO_H_
852struct freebsd4_sigaction_args {
853	int	sig;
854	struct	sigaction *act;
855	struct	sigaction *oact;
856};
857#endif
858int
859freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
860{
861	struct sigaction act, oact;
862	struct sigaction *actp, *oactp;
863	int error;
864
865
866	actp = (uap->act != NULL) ? &act : NULL;
867	oactp = (uap->oact != NULL) ? &oact : NULL;
868	if (actp) {
869		error = copyin(uap->act, actp, sizeof(act));
870		if (error)
871			return (error);
872	}
873	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
874	if (oactp && !error)
875		error = copyout(oactp, uap->oact, sizeof(oact));
876	return (error);
877}
878#endif	/* COMAPT_FREEBSD4 */
879
880#ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
881#ifndef _SYS_SYSPROTO_H_
882struct osigaction_args {
883	int	signum;
884	struct	osigaction *nsa;
885	struct	osigaction *osa;
886};
887#endif
888int
889osigaction(struct thread *td, struct osigaction_args *uap)
890{
891	struct osigaction sa;
892	struct sigaction nsa, osa;
893	struct sigaction *nsap, *osap;
894	int error;
895
896	if (uap->signum <= 0 || uap->signum >= ONSIG)
897		return (EINVAL);
898
899	nsap = (uap->nsa != NULL) ? &nsa : NULL;
900	osap = (uap->osa != NULL) ? &osa : NULL;
901
902	if (nsap) {
903		error = copyin(uap->nsa, &sa, sizeof(sa));
904		if (error)
905			return (error);
906		nsap->sa_handler = sa.sa_handler;
907		nsap->sa_flags = sa.sa_flags;
908		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
909	}
910	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
911	if (osap && !error) {
912		sa.sa_handler = osap->sa_handler;
913		sa.sa_flags = osap->sa_flags;
914		SIG2OSIG(osap->sa_mask, sa.sa_mask);
915		error = copyout(&sa, uap->osa, sizeof(sa));
916	}
917	return (error);
918}
919
920#if !defined(__i386__)
921/* Avoid replicating the same stub everywhere */
922int
923osigreturn(struct thread *td, struct osigreturn_args *uap)
924{
925
926	return (nosys(td, (struct nosys_args *)uap));
927}
928#endif
929#endif /* COMPAT_43 */
930
931/*
932 * Initialize signal state for process 0;
933 * set to ignore signals that are ignored by default.
934 */
935void
936siginit(struct proc *p)
937{
938	int i;
939	struct sigacts *ps;
940
941	PROC_LOCK(p);
942	ps = p->p_sigacts;
943	mtx_lock(&ps->ps_mtx);
944	for (i = 1; i <= NSIG; i++) {
945		if (sigprop(i) & SA_IGNORE && i != SIGCONT) {
946			SIGADDSET(ps->ps_sigignore, i);
947		}
948	}
949	mtx_unlock(&ps->ps_mtx);
950	PROC_UNLOCK(p);
951}
952
953/*
954 * Reset specified signal to the default disposition.
955 */
956static void
957sigdflt(struct sigacts *ps, int sig)
958{
959
960	mtx_assert(&ps->ps_mtx, MA_OWNED);
961	SIGDELSET(ps->ps_sigcatch, sig);
962	if ((sigprop(sig) & SA_IGNORE) != 0 && sig != SIGCONT)
963		SIGADDSET(ps->ps_sigignore, sig);
964	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
965	SIGDELSET(ps->ps_siginfo, sig);
966}
967
968/*
969 * Reset signals for an exec of the specified process.
970 */
971void
972execsigs(struct proc *p)
973{
974	sigset_t osigignore;
975	struct sigacts *ps;
976	int sig;
977	struct thread *td;
978
979	/*
980	 * Reset caught signals.  Held signals remain held
981	 * through td_sigmask (unless they were caught,
982	 * and are now ignored by default).
983	 */
984	PROC_LOCK_ASSERT(p, MA_OWNED);
985	ps = p->p_sigacts;
986	mtx_lock(&ps->ps_mtx);
987	while (SIGNOTEMPTY(ps->ps_sigcatch)) {
988		sig = sig_ffs(&ps->ps_sigcatch);
989		sigdflt(ps, sig);
990		if ((sigprop(sig) & SA_IGNORE) != 0)
991			sigqueue_delete_proc(p, sig);
992	}
993
994	/*
995	 * As CloudABI processes cannot modify signal handlers, fully
996	 * reset all signals to their default behavior. Do ignore
997	 * SIGPIPE, as it would otherwise be impossible to recover from
998	 * writes to broken pipes and sockets.
999	 */
1000	if (SV_PROC_ABI(p) == SV_ABI_CLOUDABI) {
1001		osigignore = ps->ps_sigignore;
1002		while (SIGNOTEMPTY(osigignore)) {
1003			sig = sig_ffs(&osigignore);
1004			SIGDELSET(osigignore, sig);
1005			if (sig != SIGPIPE)
1006				sigdflt(ps, sig);
1007		}
1008		SIGADDSET(ps->ps_sigignore, SIGPIPE);
1009	}
1010
1011	/*
1012	 * Reset stack state to the user stack.
1013	 * Clear set of signals caught on the signal stack.
1014	 */
1015	td = curthread;
1016	MPASS(td->td_proc == p);
1017	td->td_sigstk.ss_flags = SS_DISABLE;
1018	td->td_sigstk.ss_size = 0;
1019	td->td_sigstk.ss_sp = 0;
1020	td->td_pflags &= ~TDP_ALTSTACK;
1021	/*
1022	 * Reset no zombies if child dies flag as Solaris does.
1023	 */
1024	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1025	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1026		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1027	mtx_unlock(&ps->ps_mtx);
1028}
1029
1030/*
1031 * kern_sigprocmask()
1032 *
1033 *	Manipulate signal mask.
1034 */
1035int
1036kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1037    int flags)
1038{
1039	sigset_t new_block, oset1;
1040	struct proc *p;
1041	int error;
1042
1043	p = td->td_proc;
1044	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1045		PROC_LOCK_ASSERT(p, MA_OWNED);
1046	else
1047		PROC_LOCK(p);
1048	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1049	    ? MA_OWNED : MA_NOTOWNED);
1050	if (oset != NULL)
1051		*oset = td->td_sigmask;
1052
1053	error = 0;
1054	if (set != NULL) {
1055		switch (how) {
1056		case SIG_BLOCK:
1057			SIG_CANTMASK(*set);
1058			oset1 = td->td_sigmask;
1059			SIGSETOR(td->td_sigmask, *set);
1060			new_block = td->td_sigmask;
1061			SIGSETNAND(new_block, oset1);
1062			break;
1063		case SIG_UNBLOCK:
1064			SIGSETNAND(td->td_sigmask, *set);
1065			signotify(td);
1066			goto out;
1067		case SIG_SETMASK:
1068			SIG_CANTMASK(*set);
1069			oset1 = td->td_sigmask;
1070			if (flags & SIGPROCMASK_OLD)
1071				SIGSETLO(td->td_sigmask, *set);
1072			else
1073				td->td_sigmask = *set;
1074			new_block = td->td_sigmask;
1075			SIGSETNAND(new_block, oset1);
1076			signotify(td);
1077			break;
1078		default:
1079			error = EINVAL;
1080			goto out;
1081		}
1082
1083		/*
1084		 * The new_block set contains signals that were not previously
1085		 * blocked, but are blocked now.
1086		 *
1087		 * In case we block any signal that was not previously blocked
1088		 * for td, and process has the signal pending, try to schedule
1089		 * signal delivery to some thread that does not block the
1090		 * signal, possibly waking it up.
1091		 */
1092		if (p->p_numthreads != 1)
1093			reschedule_signals(p, new_block, flags);
1094	}
1095
1096out:
1097	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1098		PROC_UNLOCK(p);
1099	return (error);
1100}
1101
1102#ifndef _SYS_SYSPROTO_H_
1103struct sigprocmask_args {
1104	int	how;
1105	const sigset_t *set;
1106	sigset_t *oset;
1107};
1108#endif
1109int
1110sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1111{
1112	sigset_t set, oset;
1113	sigset_t *setp, *osetp;
1114	int error;
1115
1116	setp = (uap->set != NULL) ? &set : NULL;
1117	osetp = (uap->oset != NULL) ? &oset : NULL;
1118	if (setp) {
1119		error = copyin(uap->set, setp, sizeof(set));
1120		if (error)
1121			return (error);
1122	}
1123	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1124	if (osetp && !error) {
1125		error = copyout(osetp, uap->oset, sizeof(oset));
1126	}
1127	return (error);
1128}
1129
1130#ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1131#ifndef _SYS_SYSPROTO_H_
1132struct osigprocmask_args {
1133	int	how;
1134	osigset_t mask;
1135};
1136#endif
1137int
1138osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1139{
1140	sigset_t set, oset;
1141	int error;
1142
1143	OSIG2SIG(uap->mask, set);
1144	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1145	SIG2OSIG(oset, td->td_retval[0]);
1146	return (error);
1147}
1148#endif /* COMPAT_43 */
1149
1150int
1151sys_sigwait(struct thread *td, struct sigwait_args *uap)
1152{
1153	ksiginfo_t ksi;
1154	sigset_t set;
1155	int error;
1156
1157	error = copyin(uap->set, &set, sizeof(set));
1158	if (error) {
1159		td->td_retval[0] = error;
1160		return (0);
1161	}
1162
1163	error = kern_sigtimedwait(td, set, &ksi, NULL);
1164	if (error) {
1165		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1166			error = ERESTART;
1167		if (error == ERESTART)
1168			return (error);
1169		td->td_retval[0] = error;
1170		return (0);
1171	}
1172
1173	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1174	td->td_retval[0] = error;
1175	return (0);
1176}
1177
1178int
1179sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1180{
1181	struct timespec ts;
1182	struct timespec *timeout;
1183	sigset_t set;
1184	ksiginfo_t ksi;
1185	int error;
1186
1187	if (uap->timeout) {
1188		error = copyin(uap->timeout, &ts, sizeof(ts));
1189		if (error)
1190			return (error);
1191
1192		timeout = &ts;
1193	} else
1194		timeout = NULL;
1195
1196	error = copyin(uap->set, &set, sizeof(set));
1197	if (error)
1198		return (error);
1199
1200	error = kern_sigtimedwait(td, set, &ksi, timeout);
1201	if (error)
1202		return (error);
1203
1204	if (uap->info)
1205		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1206
1207	if (error == 0)
1208		td->td_retval[0] = ksi.ksi_signo;
1209	return (error);
1210}
1211
1212int
1213sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1214{
1215	ksiginfo_t ksi;
1216	sigset_t set;
1217	int error;
1218
1219	error = copyin(uap->set, &set, sizeof(set));
1220	if (error)
1221		return (error);
1222
1223	error = kern_sigtimedwait(td, set, &ksi, NULL);
1224	if (error)
1225		return (error);
1226
1227	if (uap->info)
1228		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1229
1230	if (error == 0)
1231		td->td_retval[0] = ksi.ksi_signo;
1232	return (error);
1233}
1234
1235static void
1236proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1237{
1238	struct thread *thr;
1239
1240	FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1241		if (thr == td)
1242			thr->td_si = *si;
1243		else
1244			thr->td_si.si_signo = 0;
1245	}
1246}
1247
1248int
1249kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1250	struct timespec *timeout)
1251{
1252	struct sigacts *ps;
1253	sigset_t saved_mask, new_block;
1254	struct proc *p;
1255	int error, sig, timo, timevalid = 0;
1256	struct timespec rts, ets, ts;
1257	struct timeval tv;
1258
1259	p = td->td_proc;
1260	error = 0;
1261	ets.tv_sec = 0;
1262	ets.tv_nsec = 0;
1263
1264	if (timeout != NULL) {
1265		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1266			timevalid = 1;
1267			getnanouptime(&rts);
1268			ets = rts;
1269			timespecadd(&ets, timeout);
1270		}
1271	}
1272	ksiginfo_init(ksi);
1273	/* Some signals can not be waited for. */
1274	SIG_CANTMASK(waitset);
1275	ps = p->p_sigacts;
1276	PROC_LOCK(p);
1277	saved_mask = td->td_sigmask;
1278	SIGSETNAND(td->td_sigmask, waitset);
1279	for (;;) {
1280		mtx_lock(&ps->ps_mtx);
1281		sig = cursig(td);
1282		mtx_unlock(&ps->ps_mtx);
1283		KASSERT(sig >= 0, ("sig %d", sig));
1284		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1285			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1286			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1287				error = 0;
1288				break;
1289			}
1290		}
1291
1292		if (error != 0)
1293			break;
1294
1295		/*
1296		 * POSIX says this must be checked after looking for pending
1297		 * signals.
1298		 */
1299		if (timeout != NULL) {
1300			if (!timevalid) {
1301				error = EINVAL;
1302				break;
1303			}
1304			getnanouptime(&rts);
1305			if (timespeccmp(&rts, &ets, >=)) {
1306				error = EAGAIN;
1307				break;
1308			}
1309			ts = ets;
1310			timespecsub(&ts, &rts);
1311			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1312			timo = tvtohz(&tv);
1313		} else {
1314			timo = 0;
1315		}
1316
1317		error = msleep(ps, &p->p_mtx, PPAUSE|PCATCH, "sigwait", timo);
1318
1319		if (timeout != NULL) {
1320			if (error == ERESTART) {
1321				/* Timeout can not be restarted. */
1322				error = EINTR;
1323			} else if (error == EAGAIN) {
1324				/* We will calculate timeout by ourself. */
1325				error = 0;
1326			}
1327		}
1328	}
1329
1330	new_block = saved_mask;
1331	SIGSETNAND(new_block, td->td_sigmask);
1332	td->td_sigmask = saved_mask;
1333	/*
1334	 * Fewer signals can be delivered to us, reschedule signal
1335	 * notification.
1336	 */
1337	if (p->p_numthreads != 1)
1338		reschedule_signals(p, new_block, 0);
1339
1340	if (error == 0) {
1341		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1342
1343		if (ksi->ksi_code == SI_TIMER)
1344			itimer_accept(p, ksi->ksi_timerid, ksi);
1345
1346#ifdef KTRACE
1347		if (KTRPOINT(td, KTR_PSIG)) {
1348			sig_t action;
1349
1350			mtx_lock(&ps->ps_mtx);
1351			action = ps->ps_sigact[_SIG_IDX(sig)];
1352			mtx_unlock(&ps->ps_mtx);
1353			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1354		}
1355#endif
1356		if (sig == SIGKILL) {
1357			proc_td_siginfo_capture(td, &ksi->ksi_info);
1358			sigexit(td, sig);
1359		}
1360	}
1361	PROC_UNLOCK(p);
1362	return (error);
1363}
1364
1365#ifndef _SYS_SYSPROTO_H_
1366struct sigpending_args {
1367	sigset_t	*set;
1368};
1369#endif
1370int
1371sys_sigpending(struct thread *td, struct sigpending_args *uap)
1372{
1373	struct proc *p = td->td_proc;
1374	sigset_t pending;
1375
1376	PROC_LOCK(p);
1377	pending = p->p_sigqueue.sq_signals;
1378	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1379	PROC_UNLOCK(p);
1380	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1381}
1382
1383#ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1384#ifndef _SYS_SYSPROTO_H_
1385struct osigpending_args {
1386	int	dummy;
1387};
1388#endif
1389int
1390osigpending(struct thread *td, struct osigpending_args *uap)
1391{
1392	struct proc *p = td->td_proc;
1393	sigset_t pending;
1394
1395	PROC_LOCK(p);
1396	pending = p->p_sigqueue.sq_signals;
1397	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1398	PROC_UNLOCK(p);
1399	SIG2OSIG(pending, td->td_retval[0]);
1400	return (0);
1401}
1402#endif /* COMPAT_43 */
1403
1404#if defined(COMPAT_43)
1405/*
1406 * Generalized interface signal handler, 4.3-compatible.
1407 */
1408#ifndef _SYS_SYSPROTO_H_
1409struct osigvec_args {
1410	int	signum;
1411	struct	sigvec *nsv;
1412	struct	sigvec *osv;
1413};
1414#endif
1415/* ARGSUSED */
1416int
1417osigvec(struct thread *td, struct osigvec_args *uap)
1418{
1419	struct sigvec vec;
1420	struct sigaction nsa, osa;
1421	struct sigaction *nsap, *osap;
1422	int error;
1423
1424	if (uap->signum <= 0 || uap->signum >= ONSIG)
1425		return (EINVAL);
1426	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1427	osap = (uap->osv != NULL) ? &osa : NULL;
1428	if (nsap) {
1429		error = copyin(uap->nsv, &vec, sizeof(vec));
1430		if (error)
1431			return (error);
1432		nsap->sa_handler = vec.sv_handler;
1433		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1434		nsap->sa_flags = vec.sv_flags;
1435		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1436	}
1437	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1438	if (osap && !error) {
1439		vec.sv_handler = osap->sa_handler;
1440		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1441		vec.sv_flags = osap->sa_flags;
1442		vec.sv_flags &= ~SA_NOCLDWAIT;
1443		vec.sv_flags ^= SA_RESTART;
1444		error = copyout(&vec, uap->osv, sizeof(vec));
1445	}
1446	return (error);
1447}
1448
1449#ifndef _SYS_SYSPROTO_H_
1450struct osigblock_args {
1451	int	mask;
1452};
1453#endif
1454int
1455osigblock(struct thread *td, struct osigblock_args *uap)
1456{
1457	sigset_t set, oset;
1458
1459	OSIG2SIG(uap->mask, set);
1460	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1461	SIG2OSIG(oset, td->td_retval[0]);
1462	return (0);
1463}
1464
1465#ifndef _SYS_SYSPROTO_H_
1466struct osigsetmask_args {
1467	int	mask;
1468};
1469#endif
1470int
1471osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1472{
1473	sigset_t set, oset;
1474
1475	OSIG2SIG(uap->mask, set);
1476	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1477	SIG2OSIG(oset, td->td_retval[0]);
1478	return (0);
1479}
1480#endif /* COMPAT_43 */
1481
1482/*
1483 * Suspend calling thread until signal, providing mask to be set in the
1484 * meantime.
1485 */
1486#ifndef _SYS_SYSPROTO_H_
1487struct sigsuspend_args {
1488	const sigset_t *sigmask;
1489};
1490#endif
1491/* ARGSUSED */
1492int
1493sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1494{
1495	sigset_t mask;
1496	int error;
1497
1498	error = copyin(uap->sigmask, &mask, sizeof(mask));
1499	if (error)
1500		return (error);
1501	return (kern_sigsuspend(td, mask));
1502}
1503
1504int
1505kern_sigsuspend(struct thread *td, sigset_t mask)
1506{
1507	struct proc *p = td->td_proc;
1508	int has_sig, sig;
1509
1510	/*
1511	 * When returning from sigsuspend, we want
1512	 * the old mask to be restored after the
1513	 * signal handler has finished.  Thus, we
1514	 * save it here and mark the sigacts structure
1515	 * to indicate this.
1516	 */
1517	PROC_LOCK(p);
1518	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1519	    SIGPROCMASK_PROC_LOCKED);
1520	td->td_pflags |= TDP_OLDMASK;
1521
1522	/*
1523	 * Process signals now. Otherwise, we can get spurious wakeup
1524	 * due to signal entered process queue, but delivered to other
1525	 * thread. But sigsuspend should return only on signal
1526	 * delivery.
1527	 */
1528	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1529	for (has_sig = 0; !has_sig;) {
1530		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1531			0) == 0)
1532			/* void */;
1533		thread_suspend_check(0);
1534		mtx_lock(&p->p_sigacts->ps_mtx);
1535		while ((sig = cursig(td)) != 0) {
1536			KASSERT(sig >= 0, ("sig %d", sig));
1537			has_sig += postsig(sig);
1538		}
1539		mtx_unlock(&p->p_sigacts->ps_mtx);
1540	}
1541	PROC_UNLOCK(p);
1542	td->td_errno = EINTR;
1543	td->td_pflags |= TDP_NERRNO;
1544	return (EJUSTRETURN);
1545}
1546
1547#ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1548/*
1549 * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1550 * convention: libc stub passes mask, not pointer, to save a copyin.
1551 */
1552#ifndef _SYS_SYSPROTO_H_
1553struct osigsuspend_args {
1554	osigset_t mask;
1555};
1556#endif
1557/* ARGSUSED */
1558int
1559osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1560{
1561	sigset_t mask;
1562
1563	OSIG2SIG(uap->mask, mask);
1564	return (kern_sigsuspend(td, mask));
1565}
1566#endif /* COMPAT_43 */
1567
1568#if defined(COMPAT_43)
1569#ifndef _SYS_SYSPROTO_H_
1570struct osigstack_args {
1571	struct	sigstack *nss;
1572	struct	sigstack *oss;
1573};
1574#endif
1575/* ARGSUSED */
1576int
1577osigstack(struct thread *td, struct osigstack_args *uap)
1578{
1579	struct sigstack nss, oss;
1580	int error = 0;
1581
1582	if (uap->nss != NULL) {
1583		error = copyin(uap->nss, &nss, sizeof(nss));
1584		if (error)
1585			return (error);
1586	}
1587	oss.ss_sp = td->td_sigstk.ss_sp;
1588	oss.ss_onstack = sigonstack(cpu_getstack(td));
1589	if (uap->nss != NULL) {
1590		td->td_sigstk.ss_sp = nss.ss_sp;
1591		td->td_sigstk.ss_size = 0;
1592		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1593		td->td_pflags |= TDP_ALTSTACK;
1594	}
1595	if (uap->oss != NULL)
1596		error = copyout(&oss, uap->oss, sizeof(oss));
1597
1598	return (error);
1599}
1600#endif /* COMPAT_43 */
1601
1602#ifndef _SYS_SYSPROTO_H_
1603struct sigaltstack_args {
1604	stack_t	*ss;
1605	stack_t	*oss;
1606};
1607#endif
1608/* ARGSUSED */
1609int
1610sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1611{
1612	stack_t ss, oss;
1613	int error;
1614
1615	if (uap->ss != NULL) {
1616		error = copyin(uap->ss, &ss, sizeof(ss));
1617		if (error)
1618			return (error);
1619	}
1620	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1621	    (uap->oss != NULL) ? &oss : NULL);
1622	if (error)
1623		return (error);
1624	if (uap->oss != NULL)
1625		error = copyout(&oss, uap->oss, sizeof(stack_t));
1626	return (error);
1627}
1628
1629int
1630kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1631{
1632	struct proc *p = td->td_proc;
1633	int oonstack;
1634
1635	oonstack = sigonstack(cpu_getstack(td));
1636
1637	if (oss != NULL) {
1638		*oss = td->td_sigstk;
1639		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1640		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1641	}
1642
1643	if (ss != NULL) {
1644		if (oonstack)
1645			return (EPERM);
1646		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1647			return (EINVAL);
1648		if (!(ss->ss_flags & SS_DISABLE)) {
1649			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1650				return (ENOMEM);
1651
1652			td->td_sigstk = *ss;
1653			td->td_pflags |= TDP_ALTSTACK;
1654		} else {
1655			td->td_pflags &= ~TDP_ALTSTACK;
1656		}
1657	}
1658	return (0);
1659}
1660
1661/*
1662 * Common code for kill process group/broadcast kill.
1663 * cp is calling process.
1664 */
1665static int
1666killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1667{
1668	struct proc *p;
1669	struct pgrp *pgrp;
1670	int err;
1671	int ret;
1672
1673	ret = ESRCH;
1674	if (all) {
1675		/*
1676		 * broadcast
1677		 */
1678		sx_slock(&allproc_lock);
1679		FOREACH_PROC_IN_SYSTEM(p) {
1680			PROC_LOCK(p);
1681			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1682			    p == td->td_proc || p->p_state == PRS_NEW) {
1683				PROC_UNLOCK(p);
1684				continue;
1685			}
1686			err = p_cansignal(td, p, sig);
1687			if (err == 0) {
1688				if (sig)
1689					pksignal(p, sig, ksi);
1690				ret = err;
1691			}
1692			else if (ret == ESRCH)
1693				ret = err;
1694			PROC_UNLOCK(p);
1695		}
1696		sx_sunlock(&allproc_lock);
1697	} else {
1698		sx_slock(&proctree_lock);
1699		if (pgid == 0) {
1700			/*
1701			 * zero pgid means send to my process group.
1702			 */
1703			pgrp = td->td_proc->p_pgrp;
1704			PGRP_LOCK(pgrp);
1705		} else {
1706			pgrp = pgfind(pgid);
1707			if (pgrp == NULL) {
1708				sx_sunlock(&proctree_lock);
1709				return (ESRCH);
1710			}
1711		}
1712		sx_sunlock(&proctree_lock);
1713		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1714			PROC_LOCK(p);
1715			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1716			    p->p_state == PRS_NEW) {
1717				PROC_UNLOCK(p);
1718				continue;
1719			}
1720			err = p_cansignal(td, p, sig);
1721			if (err == 0) {
1722				if (sig)
1723					pksignal(p, sig, ksi);
1724				ret = err;
1725			}
1726			else if (ret == ESRCH)
1727				ret = err;
1728			PROC_UNLOCK(p);
1729		}
1730		PGRP_UNLOCK(pgrp);
1731	}
1732	return (ret);
1733}
1734
1735#ifndef _SYS_SYSPROTO_H_
1736struct kill_args {
1737	int	pid;
1738	int	signum;
1739};
1740#endif
1741/* ARGSUSED */
1742int
1743sys_kill(struct thread *td, struct kill_args *uap)
1744{
1745	ksiginfo_t ksi;
1746	struct proc *p;
1747	int error;
1748
1749	/*
1750	 * A process in capability mode can send signals only to himself.
1751	 * The main rationale behind this is that abort(3) is implemented as
1752	 * kill(getpid(), SIGABRT).
1753	 */
1754	if (IN_CAPABILITY_MODE(td) && uap->pid != td->td_proc->p_pid)
1755		return (ECAPMODE);
1756
1757	AUDIT_ARG_SIGNUM(uap->signum);
1758	AUDIT_ARG_PID(uap->pid);
1759	if ((u_int)uap->signum > _SIG_MAXSIG)
1760		return (EINVAL);
1761
1762	ksiginfo_init(&ksi);
1763	ksi.ksi_signo = uap->signum;
1764	ksi.ksi_code = SI_USER;
1765	ksi.ksi_pid = td->td_proc->p_pid;
1766	ksi.ksi_uid = td->td_ucred->cr_ruid;
1767
1768	if (uap->pid > 0) {
1769		/* kill single process */
1770		if ((p = pfind(uap->pid)) == NULL) {
1771			if ((p = zpfind(uap->pid)) == NULL)
1772				return (ESRCH);
1773		}
1774		AUDIT_ARG_PROCESS(p);
1775		error = p_cansignal(td, p, uap->signum);
1776		if (error == 0 && uap->signum)
1777			pksignal(p, uap->signum, &ksi);
1778		PROC_UNLOCK(p);
1779		return (error);
1780	}
1781	switch (uap->pid) {
1782	case -1:		/* broadcast signal */
1783		return (killpg1(td, uap->signum, 0, 1, &ksi));
1784	case 0:			/* signal own process group */
1785		return (killpg1(td, uap->signum, 0, 0, &ksi));
1786	default:		/* negative explicit process group */
1787		return (killpg1(td, uap->signum, -uap->pid, 0, &ksi));
1788	}
1789	/* NOTREACHED */
1790}
1791
1792int
1793sys_pdkill(struct thread *td, struct pdkill_args *uap)
1794{
1795	struct proc *p;
1796	cap_rights_t rights;
1797	int error;
1798
1799	AUDIT_ARG_SIGNUM(uap->signum);
1800	AUDIT_ARG_FD(uap->fd);
1801	if ((u_int)uap->signum > _SIG_MAXSIG)
1802		return (EINVAL);
1803
1804	error = procdesc_find(td, uap->fd,
1805	    cap_rights_init(&rights, CAP_PDKILL), &p);
1806	if (error)
1807		return (error);
1808	AUDIT_ARG_PROCESS(p);
1809	error = p_cansignal(td, p, uap->signum);
1810	if (error == 0 && uap->signum)
1811		kern_psignal(p, uap->signum);
1812	PROC_UNLOCK(p);
1813	return (error);
1814}
1815
1816#if defined(COMPAT_43)
1817#ifndef _SYS_SYSPROTO_H_
1818struct okillpg_args {
1819	int	pgid;
1820	int	signum;
1821};
1822#endif
1823/* ARGSUSED */
1824int
1825okillpg(struct thread *td, struct okillpg_args *uap)
1826{
1827	ksiginfo_t ksi;
1828
1829	AUDIT_ARG_SIGNUM(uap->signum);
1830	AUDIT_ARG_PID(uap->pgid);
1831	if ((u_int)uap->signum > _SIG_MAXSIG)
1832		return (EINVAL);
1833
1834	ksiginfo_init(&ksi);
1835	ksi.ksi_signo = uap->signum;
1836	ksi.ksi_code = SI_USER;
1837	ksi.ksi_pid = td->td_proc->p_pid;
1838	ksi.ksi_uid = td->td_ucred->cr_ruid;
1839	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1840}
1841#endif /* COMPAT_43 */
1842
1843#ifndef _SYS_SYSPROTO_H_
1844struct sigqueue_args {
1845	pid_t pid;
1846	int signum;
1847	/* union sigval */ void *value;
1848};
1849#endif
1850int
1851sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1852{
1853	union sigval sv;
1854
1855	sv.sival_ptr = uap->value;
1856
1857	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
1858}
1859
1860int
1861kern_sigqueue(struct thread *td, pid_t pid, int signum, union sigval *value)
1862{
1863	ksiginfo_t ksi;
1864	struct proc *p;
1865	int error;
1866
1867	if ((u_int)signum > _SIG_MAXSIG)
1868		return (EINVAL);
1869
1870	/*
1871	 * Specification says sigqueue can only send signal to
1872	 * single process.
1873	 */
1874	if (pid <= 0)
1875		return (EINVAL);
1876
1877	if ((p = pfind(pid)) == NULL) {
1878		if ((p = zpfind(pid)) == NULL)
1879			return (ESRCH);
1880	}
1881	error = p_cansignal(td, p, signum);
1882	if (error == 0 && signum != 0) {
1883		ksiginfo_init(&ksi);
1884		ksi.ksi_flags = KSI_SIGQ;
1885		ksi.ksi_signo = signum;
1886		ksi.ksi_code = SI_QUEUE;
1887		ksi.ksi_pid = td->td_proc->p_pid;
1888		ksi.ksi_uid = td->td_ucred->cr_ruid;
1889		ksi.ksi_value = *value;
1890		error = pksignal(p, ksi.ksi_signo, &ksi);
1891	}
1892	PROC_UNLOCK(p);
1893	return (error);
1894}
1895
1896/*
1897 * Send a signal to a process group.
1898 */
1899void
1900gsignal(int pgid, int sig, ksiginfo_t *ksi)
1901{
1902	struct pgrp *pgrp;
1903
1904	if (pgid != 0) {
1905		sx_slock(&proctree_lock);
1906		pgrp = pgfind(pgid);
1907		sx_sunlock(&proctree_lock);
1908		if (pgrp != NULL) {
1909			pgsignal(pgrp, sig, 0, ksi);
1910			PGRP_UNLOCK(pgrp);
1911		}
1912	}
1913}
1914
1915/*
1916 * Send a signal to a process group.  If checktty is 1,
1917 * limit to members which have a controlling terminal.
1918 */
1919void
1920pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
1921{
1922	struct proc *p;
1923
1924	if (pgrp) {
1925		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
1926		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1927			PROC_LOCK(p);
1928			if (p->p_state == PRS_NORMAL &&
1929			    (checkctty == 0 || p->p_flag & P_CONTROLT))
1930				pksignal(p, sig, ksi);
1931			PROC_UNLOCK(p);
1932		}
1933	}
1934}
1935
1936
1937/*
1938 * Recalculate the signal mask and reset the signal disposition after
1939 * usermode frame for delivery is formed.  Should be called after
1940 * mach-specific routine, because sysent->sv_sendsig() needs correct
1941 * ps_siginfo and signal mask.
1942 */
1943static void
1944postsig_done(int sig, struct thread *td, struct sigacts *ps)
1945{
1946	sigset_t mask;
1947
1948	mtx_assert(&ps->ps_mtx, MA_OWNED);
1949	td->td_ru.ru_nsignals++;
1950	mask = ps->ps_catchmask[_SIG_IDX(sig)];
1951	if (!SIGISMEMBER(ps->ps_signodefer, sig))
1952		SIGADDSET(mask, sig);
1953	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
1954	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
1955	if (SIGISMEMBER(ps->ps_sigreset, sig))
1956		sigdflt(ps, sig);
1957}
1958
1959
1960/*
1961 * Send a signal caused by a trap to the current thread.  If it will be
1962 * caught immediately, deliver it with correct code.  Otherwise, post it
1963 * normally.
1964 */
1965void
1966trapsignal(struct thread *td, ksiginfo_t *ksi)
1967{
1968	struct sigacts *ps;
1969	struct proc *p;
1970	int sig;
1971	int code;
1972
1973	p = td->td_proc;
1974	sig = ksi->ksi_signo;
1975	code = ksi->ksi_code;
1976	KASSERT(_SIG_VALID(sig), ("invalid signal"));
1977
1978	PROC_LOCK(p);
1979	ps = p->p_sigacts;
1980	mtx_lock(&ps->ps_mtx);
1981	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
1982	    !SIGISMEMBER(td->td_sigmask, sig)) {
1983#ifdef KTRACE
1984		if (KTRPOINT(curthread, KTR_PSIG))
1985			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
1986			    &td->td_sigmask, code);
1987#endif
1988		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
1989				ksi, &td->td_sigmask);
1990		postsig_done(sig, td, ps);
1991		mtx_unlock(&ps->ps_mtx);
1992	} else {
1993		/*
1994		 * Avoid a possible infinite loop if the thread
1995		 * masking the signal or process is ignoring the
1996		 * signal.
1997		 */
1998		if (kern_forcesigexit &&
1999		    (SIGISMEMBER(td->td_sigmask, sig) ||
2000		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2001			SIGDELSET(td->td_sigmask, sig);
2002			SIGDELSET(ps->ps_sigcatch, sig);
2003			SIGDELSET(ps->ps_sigignore, sig);
2004			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2005		}
2006		mtx_unlock(&ps->ps_mtx);
2007		p->p_code = code;	/* XXX for core dump/debugger */
2008		p->p_sig = sig;		/* XXX to verify code */
2009		tdsendsignal(p, td, sig, ksi);
2010	}
2011	PROC_UNLOCK(p);
2012}
2013
2014static struct thread *
2015sigtd(struct proc *p, int sig, int prop)
2016{
2017	struct thread *td, *signal_td;
2018
2019	PROC_LOCK_ASSERT(p, MA_OWNED);
2020
2021	/*
2022	 * Check if current thread can handle the signal without
2023	 * switching context to another thread.
2024	 */
2025	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig))
2026		return (curthread);
2027	signal_td = NULL;
2028	FOREACH_THREAD_IN_PROC(p, td) {
2029		if (!SIGISMEMBER(td->td_sigmask, sig)) {
2030			signal_td = td;
2031			break;
2032		}
2033	}
2034	if (signal_td == NULL)
2035		signal_td = FIRST_THREAD_IN_PROC(p);
2036	return (signal_td);
2037}
2038
2039/*
2040 * Send the signal to the process.  If the signal has an action, the action
2041 * is usually performed by the target process rather than the caller; we add
2042 * the signal to the set of pending signals for the process.
2043 *
2044 * Exceptions:
2045 *   o When a stop signal is sent to a sleeping process that takes the
2046 *     default action, the process is stopped without awakening it.
2047 *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2048 *     regardless of the signal action (eg, blocked or ignored).
2049 *
2050 * Other ignored signals are discarded immediately.
2051 *
2052 * NB: This function may be entered from the debugger via the "kill" DDB
2053 * command.  There is little that can be done to mitigate the possibly messy
2054 * side effects of this unwise possibility.
2055 */
2056void
2057kern_psignal(struct proc *p, int sig)
2058{
2059	ksiginfo_t ksi;
2060
2061	ksiginfo_init(&ksi);
2062	ksi.ksi_signo = sig;
2063	ksi.ksi_code = SI_KERNEL;
2064	(void) tdsendsignal(p, NULL, sig, &ksi);
2065}
2066
2067int
2068pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2069{
2070
2071	return (tdsendsignal(p, NULL, sig, ksi));
2072}
2073
2074/* Utility function for finding a thread to send signal event to. */
2075int
2076sigev_findtd(struct proc *p ,struct sigevent *sigev, struct thread **ttd)
2077{
2078	struct thread *td;
2079
2080	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2081		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2082		if (td == NULL)
2083			return (ESRCH);
2084		*ttd = td;
2085	} else {
2086		*ttd = NULL;
2087		PROC_LOCK(p);
2088	}
2089	return (0);
2090}
2091
2092void
2093tdsignal(struct thread *td, int sig)
2094{
2095	ksiginfo_t ksi;
2096
2097	ksiginfo_init(&ksi);
2098	ksi.ksi_signo = sig;
2099	ksi.ksi_code = SI_KERNEL;
2100	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2101}
2102
2103void
2104tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2105{
2106
2107	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2108}
2109
2110int
2111tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2112{
2113	sig_t action;
2114	sigqueue_t *sigqueue;
2115	int prop;
2116	struct sigacts *ps;
2117	int intrval;
2118	int ret = 0;
2119	int wakeup_swapper;
2120
2121	MPASS(td == NULL || p == td->td_proc);
2122	PROC_LOCK_ASSERT(p, MA_OWNED);
2123
2124	if (!_SIG_VALID(sig))
2125		panic("%s(): invalid signal %d", __func__, sig);
2126
2127	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2128
2129	/*
2130	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2131	 */
2132	if (p->p_state == PRS_ZOMBIE) {
2133		if (ksi && (ksi->ksi_flags & KSI_INS))
2134			ksiginfo_tryfree(ksi);
2135		return (ret);
2136	}
2137
2138	ps = p->p_sigacts;
2139	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2140	prop = sigprop(sig);
2141
2142	if (td == NULL) {
2143		td = sigtd(p, sig, prop);
2144		sigqueue = &p->p_sigqueue;
2145	} else
2146		sigqueue = &td->td_sigqueue;
2147
2148	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2149
2150	/*
2151	 * If the signal is being ignored,
2152	 * then we forget about it immediately.
2153	 * (Note: we don't set SIGCONT in ps_sigignore,
2154	 * and if it is set to SIG_IGN,
2155	 * action will be SIG_DFL here.)
2156	 */
2157	mtx_lock(&ps->ps_mtx);
2158	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2159		SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2160
2161		mtx_unlock(&ps->ps_mtx);
2162		if (ksi && (ksi->ksi_flags & KSI_INS))
2163			ksiginfo_tryfree(ksi);
2164		return (ret);
2165	}
2166	if (SIGISMEMBER(td->td_sigmask, sig))
2167		action = SIG_HOLD;
2168	else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2169		action = SIG_CATCH;
2170	else
2171		action = SIG_DFL;
2172	if (SIGISMEMBER(ps->ps_sigintr, sig))
2173		intrval = EINTR;
2174	else
2175		intrval = ERESTART;
2176	mtx_unlock(&ps->ps_mtx);
2177
2178	if (prop & SA_CONT)
2179		sigqueue_delete_stopmask_proc(p);
2180	else if (prop & SA_STOP) {
2181		/*
2182		 * If sending a tty stop signal to a member of an orphaned
2183		 * process group, discard the signal here if the action
2184		 * is default; don't stop the process below if sleeping,
2185		 * and don't clear any pending SIGCONT.
2186		 */
2187		if ((prop & SA_TTYSTOP) &&
2188		    (p->p_pgrp->pg_jobc == 0) &&
2189		    (action == SIG_DFL)) {
2190			if (ksi && (ksi->ksi_flags & KSI_INS))
2191				ksiginfo_tryfree(ksi);
2192			return (ret);
2193		}
2194		sigqueue_delete_proc(p, SIGCONT);
2195		if (p->p_flag & P_CONTINUED) {
2196			p->p_flag &= ~P_CONTINUED;
2197			PROC_LOCK(p->p_pptr);
2198			sigqueue_take(p->p_ksi);
2199			PROC_UNLOCK(p->p_pptr);
2200		}
2201	}
2202
2203	ret = sigqueue_add(sigqueue, sig, ksi);
2204	if (ret != 0)
2205		return (ret);
2206	signotify(td);
2207	/*
2208	 * Defer further processing for signals which are held,
2209	 * except that stopped processes must be continued by SIGCONT.
2210	 */
2211	if (action == SIG_HOLD &&
2212	    !((prop & SA_CONT) && (p->p_flag & P_STOPPED_SIG)))
2213		return (ret);
2214
2215	/* SIGKILL: Remove procfs STOPEVENTs. */
2216	if (sig == SIGKILL) {
2217		/* from procfs_ioctl.c: PIOCBIC */
2218		p->p_stops = 0;
2219		/* from procfs_ioctl.c: PIOCCONT */
2220		p->p_step = 0;
2221		wakeup(&p->p_step);
2222	}
2223	/*
2224	 * Some signals have a process-wide effect and a per-thread
2225	 * component.  Most processing occurs when the process next
2226	 * tries to cross the user boundary, however there are some
2227	 * times when processing needs to be done immediately, such as
2228	 * waking up threads so that they can cross the user boundary.
2229	 * We try to do the per-process part here.
2230	 */
2231	if (P_SHOULDSTOP(p)) {
2232		KASSERT(!(p->p_flag & P_WEXIT),
2233		    ("signal to stopped but exiting process"));
2234		if (sig == SIGKILL) {
2235			/*
2236			 * If traced process is already stopped,
2237			 * then no further action is necessary.
2238			 */
2239			if (p->p_flag & P_TRACED)
2240				goto out;
2241			/*
2242			 * SIGKILL sets process running.
2243			 * It will die elsewhere.
2244			 * All threads must be restarted.
2245			 */
2246			p->p_flag &= ~P_STOPPED_SIG;
2247			goto runfast;
2248		}
2249
2250		if (prop & SA_CONT) {
2251			/*
2252			 * If traced process is already stopped,
2253			 * then no further action is necessary.
2254			 */
2255			if (p->p_flag & P_TRACED)
2256				goto out;
2257			/*
2258			 * If SIGCONT is default (or ignored), we continue the
2259			 * process but don't leave the signal in sigqueue as
2260			 * it has no further action.  If SIGCONT is held, we
2261			 * continue the process and leave the signal in
2262			 * sigqueue.  If the process catches SIGCONT, let it
2263			 * handle the signal itself.  If it isn't waiting on
2264			 * an event, it goes back to run state.
2265			 * Otherwise, process goes back to sleep state.
2266			 */
2267			p->p_flag &= ~P_STOPPED_SIG;
2268			PROC_SLOCK(p);
2269			if (p->p_numthreads == p->p_suspcount) {
2270				PROC_SUNLOCK(p);
2271				p->p_flag |= P_CONTINUED;
2272				p->p_xsig = SIGCONT;
2273				PROC_LOCK(p->p_pptr);
2274				childproc_continued(p);
2275				PROC_UNLOCK(p->p_pptr);
2276				PROC_SLOCK(p);
2277			}
2278			if (action == SIG_DFL) {
2279				thread_unsuspend(p);
2280				PROC_SUNLOCK(p);
2281				sigqueue_delete(sigqueue, sig);
2282				goto out;
2283			}
2284			if (action == SIG_CATCH) {
2285				/*
2286				 * The process wants to catch it so it needs
2287				 * to run at least one thread, but which one?
2288				 */
2289				PROC_SUNLOCK(p);
2290				goto runfast;
2291			}
2292			/*
2293			 * The signal is not ignored or caught.
2294			 */
2295			thread_unsuspend(p);
2296			PROC_SUNLOCK(p);
2297			goto out;
2298		}
2299
2300		if (prop & SA_STOP) {
2301			/*
2302			 * If traced process is already stopped,
2303			 * then no further action is necessary.
2304			 */
2305			if (p->p_flag & P_TRACED)
2306				goto out;
2307			/*
2308			 * Already stopped, don't need to stop again
2309			 * (If we did the shell could get confused).
2310			 * Just make sure the signal STOP bit set.
2311			 */
2312			p->p_flag |= P_STOPPED_SIG;
2313			sigqueue_delete(sigqueue, sig);
2314			goto out;
2315		}
2316
2317		/*
2318		 * All other kinds of signals:
2319		 * If a thread is sleeping interruptibly, simulate a
2320		 * wakeup so that when it is continued it will be made
2321		 * runnable and can look at the signal.  However, don't make
2322		 * the PROCESS runnable, leave it stopped.
2323		 * It may run a bit until it hits a thread_suspend_check().
2324		 */
2325		wakeup_swapper = 0;
2326		PROC_SLOCK(p);
2327		thread_lock(td);
2328		if (TD_ON_SLEEPQ(td) && (td->td_flags & TDF_SINTR))
2329			wakeup_swapper = sleepq_abort(td, intrval);
2330		thread_unlock(td);
2331		PROC_SUNLOCK(p);
2332		if (wakeup_swapper)
2333			kick_proc0();
2334		goto out;
2335		/*
2336		 * Mutexes are short lived. Threads waiting on them will
2337		 * hit thread_suspend_check() soon.
2338		 */
2339	} else if (p->p_state == PRS_NORMAL) {
2340		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2341			tdsigwakeup(td, sig, action, intrval);
2342			goto out;
2343		}
2344
2345		MPASS(action == SIG_DFL);
2346
2347		if (prop & SA_STOP) {
2348			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2349				goto out;
2350			p->p_flag |= P_STOPPED_SIG;
2351			p->p_xsig = sig;
2352			PROC_SLOCK(p);
2353			wakeup_swapper = sig_suspend_threads(td, p, 1);
2354			if (p->p_numthreads == p->p_suspcount) {
2355				/*
2356				 * only thread sending signal to another
2357				 * process can reach here, if thread is sending
2358				 * signal to its process, because thread does
2359				 * not suspend itself here, p_numthreads
2360				 * should never be equal to p_suspcount.
2361				 */
2362				thread_stopped(p);
2363				PROC_SUNLOCK(p);
2364				sigqueue_delete_proc(p, p->p_xsig);
2365			} else
2366				PROC_SUNLOCK(p);
2367			if (wakeup_swapper)
2368				kick_proc0();
2369			goto out;
2370		}
2371	} else {
2372		/* Not in "NORMAL" state. discard the signal. */
2373		sigqueue_delete(sigqueue, sig);
2374		goto out;
2375	}
2376
2377	/*
2378	 * The process is not stopped so we need to apply the signal to all the
2379	 * running threads.
2380	 */
2381runfast:
2382	tdsigwakeup(td, sig, action, intrval);
2383	PROC_SLOCK(p);
2384	thread_unsuspend(p);
2385	PROC_SUNLOCK(p);
2386out:
2387	/* If we jump here, proc slock should not be owned. */
2388	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2389	return (ret);
2390}
2391
2392/*
2393 * The force of a signal has been directed against a single
2394 * thread.  We need to see what we can do about knocking it
2395 * out of any sleep it may be in etc.
2396 */
2397static void
2398tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2399{
2400	struct proc *p = td->td_proc;
2401	int prop;
2402	int wakeup_swapper;
2403
2404	wakeup_swapper = 0;
2405	PROC_LOCK_ASSERT(p, MA_OWNED);
2406	prop = sigprop(sig);
2407
2408	PROC_SLOCK(p);
2409	thread_lock(td);
2410	/*
2411	 * Bring the priority of a thread up if we want it to get
2412	 * killed in this lifetime.  Be careful to avoid bumping the
2413	 * priority of the idle thread, since we still allow to signal
2414	 * kernel processes.
2415	 */
2416	if (action == SIG_DFL && (prop & SA_KILL) != 0 &&
2417	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2418		sched_prio(td, PUSER);
2419	if (TD_ON_SLEEPQ(td)) {
2420		/*
2421		 * If thread is sleeping uninterruptibly
2422		 * we can't interrupt the sleep... the signal will
2423		 * be noticed when the process returns through
2424		 * trap() or syscall().
2425		 */
2426		if ((td->td_flags & TDF_SINTR) == 0)
2427			goto out;
2428		/*
2429		 * If SIGCONT is default (or ignored) and process is
2430		 * asleep, we are finished; the process should not
2431		 * be awakened.
2432		 */
2433		if ((prop & SA_CONT) && action == SIG_DFL) {
2434			thread_unlock(td);
2435			PROC_SUNLOCK(p);
2436			sigqueue_delete(&p->p_sigqueue, sig);
2437			/*
2438			 * It may be on either list in this state.
2439			 * Remove from both for now.
2440			 */
2441			sigqueue_delete(&td->td_sigqueue, sig);
2442			return;
2443		}
2444
2445		/*
2446		 * Don't awaken a sleeping thread for SIGSTOP if the
2447		 * STOP signal is deferred.
2448		 */
2449		if ((prop & SA_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2450		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2451			goto out;
2452
2453		/*
2454		 * Give low priority threads a better chance to run.
2455		 */
2456		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2457			sched_prio(td, PUSER);
2458
2459		wakeup_swapper = sleepq_abort(td, intrval);
2460	} else {
2461		/*
2462		 * Other states do nothing with the signal immediately,
2463		 * other than kicking ourselves if we are running.
2464		 * It will either never be noticed, or noticed very soon.
2465		 */
2466#ifdef SMP
2467		if (TD_IS_RUNNING(td) && td != curthread)
2468			forward_signal(td);
2469#endif
2470	}
2471out:
2472	PROC_SUNLOCK(p);
2473	thread_unlock(td);
2474	if (wakeup_swapper)
2475		kick_proc0();
2476}
2477
2478static int
2479sig_suspend_threads(struct thread *td, struct proc *p, int sending)
2480{
2481	struct thread *td2;
2482	int wakeup_swapper;
2483
2484	PROC_LOCK_ASSERT(p, MA_OWNED);
2485	PROC_SLOCK_ASSERT(p, MA_OWNED);
2486	MPASS(sending || td == curthread);
2487
2488	wakeup_swapper = 0;
2489	FOREACH_THREAD_IN_PROC(p, td2) {
2490		thread_lock(td2);
2491		td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
2492		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2493		    (td2->td_flags & TDF_SINTR)) {
2494			if (td2->td_flags & TDF_SBDRY) {
2495				/*
2496				 * Once a thread is asleep with
2497				 * TDF_SBDRY and without TDF_SERESTART
2498				 * or TDF_SEINTR set, it should never
2499				 * become suspended due to this check.
2500				 */
2501				KASSERT(!TD_IS_SUSPENDED(td2),
2502				    ("thread with deferred stops suspended"));
2503				if (TD_SBDRY_INTR(td2))
2504					wakeup_swapper |= sleepq_abort(td2,
2505					    TD_SBDRY_ERRNO(td2));
2506			} else if (!TD_IS_SUSPENDED(td2)) {
2507				thread_suspend_one(td2);
2508			}
2509		} else if (!TD_IS_SUSPENDED(td2)) {
2510			if (sending || td != td2)
2511				td2->td_flags |= TDF_ASTPENDING;
2512#ifdef SMP
2513			if (TD_IS_RUNNING(td2) && td2 != td)
2514				forward_signal(td2);
2515#endif
2516		}
2517		thread_unlock(td2);
2518	}
2519	return (wakeup_swapper);
2520}
2521
2522/*
2523 * Stop the process for an event deemed interesting to the debugger. If si is
2524 * non-NULL, this is a signal exchange; the new signal requested by the
2525 * debugger will be returned for handling. If si is NULL, this is some other
2526 * type of interesting event. The debugger may request a signal be delivered in
2527 * that case as well, however it will be deferred until it can be handled.
2528 */
2529int
2530ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2531{
2532	struct proc *p = td->td_proc;
2533	struct thread *td2;
2534	ksiginfo_t ksi;
2535	int prop;
2536
2537	PROC_LOCK_ASSERT(p, MA_OWNED);
2538	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2539	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2540	    &p->p_mtx.lock_object, "Stopping for traced signal");
2541
2542	td->td_xsig = sig;
2543
2544	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2545		td->td_dbgflags |= TDB_XSIG;
2546		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2547		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2548		PROC_SLOCK(p);
2549		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2550			if (P_KILLED(p)) {
2551				/*
2552				 * Ensure that, if we've been PT_KILLed, the
2553				 * exit status reflects that. Another thread
2554				 * may also be in ptracestop(), having just
2555				 * received the SIGKILL, but this thread was
2556				 * unsuspended first.
2557				 */
2558				td->td_dbgflags &= ~TDB_XSIG;
2559				td->td_xsig = SIGKILL;
2560				p->p_ptevents = 0;
2561				break;
2562			}
2563			if (p->p_flag & P_SINGLE_EXIT &&
2564			    !(td->td_dbgflags & TDB_EXIT)) {
2565				/*
2566				 * Ignore ptrace stops except for thread exit
2567				 * events when the process exits.
2568				 */
2569				td->td_dbgflags &= ~TDB_XSIG;
2570				PROC_SUNLOCK(p);
2571				return (0);
2572			}
2573
2574			/*
2575			 * Make wait(2) work.  Ensure that right after the
2576			 * attach, the thread which was decided to become the
2577			 * leader of attach gets reported to the waiter.
2578			 * Otherwise, just avoid overwriting another thread's
2579			 * assignment to p_xthread.  If another thread has
2580			 * already set p_xthread, the current thread will get
2581			 * a chance to report itself upon the next iteration.
2582			 */
2583			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2584			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2585			    p->p_xthread == NULL)) {
2586				p->p_xsig = sig;
2587				p->p_xthread = td;
2588				td->td_dbgflags &= ~TDB_FSTP;
2589				p->p_flag2 &= ~P2_PTRACE_FSTP;
2590				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2591				sig_suspend_threads(td, p, 0);
2592			}
2593			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2594				td->td_dbgflags &= ~TDB_STOPATFORK;
2595				cv_broadcast(&p->p_dbgwait);
2596			}
2597stopme:
2598			thread_suspend_switch(td, p);
2599			if (p->p_xthread == td)
2600				p->p_xthread = NULL;
2601			if (!(p->p_flag & P_TRACED))
2602				break;
2603			if (td->td_dbgflags & TDB_SUSPEND) {
2604				if (p->p_flag & P_SINGLE_EXIT)
2605					break;
2606				goto stopme;
2607			}
2608		}
2609		PROC_SUNLOCK(p);
2610	}
2611
2612	if (si != NULL && sig == td->td_xsig) {
2613		/* Parent wants us to take the original signal unchanged. */
2614		si->ksi_flags |= KSI_HEAD;
2615		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2616			si->ksi_signo = 0;
2617	} else if (td->td_xsig != 0) {
2618		/*
2619		 * If parent wants us to take a new signal, then it will leave
2620		 * it in td->td_xsig; otherwise we just look for signals again.
2621		 */
2622		ksiginfo_init(&ksi);
2623		ksi.ksi_signo = td->td_xsig;
2624		ksi.ksi_flags |= KSI_PTRACE;
2625		prop = sigprop(td->td_xsig);
2626		td2 = sigtd(p, td->td_xsig, prop);
2627		tdsendsignal(p, td2, td->td_xsig, &ksi);
2628		if (td != td2)
2629			return (0);
2630	}
2631
2632	return (td->td_xsig);
2633}
2634
2635static void
2636reschedule_signals(struct proc *p, sigset_t block, int flags)
2637{
2638	struct sigacts *ps;
2639	struct thread *td;
2640	int sig;
2641
2642	PROC_LOCK_ASSERT(p, MA_OWNED);
2643	ps = p->p_sigacts;
2644	mtx_assert(&ps->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0 ?
2645	    MA_OWNED : MA_NOTOWNED);
2646	if (SIGISEMPTY(p->p_siglist))
2647		return;
2648	SIGSETAND(block, p->p_siglist);
2649	while ((sig = sig_ffs(&block)) != 0) {
2650		SIGDELSET(block, sig);
2651		td = sigtd(p, sig, 0);
2652		signotify(td);
2653		if (!(flags & SIGPROCMASK_PS_LOCKED))
2654			mtx_lock(&ps->ps_mtx);
2655		if (p->p_flag & P_TRACED ||
2656		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2657		    !SIGISMEMBER(td->td_sigmask, sig)))
2658			tdsigwakeup(td, sig, SIG_CATCH,
2659			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2660			     ERESTART));
2661		if (!(flags & SIGPROCMASK_PS_LOCKED))
2662			mtx_unlock(&ps->ps_mtx);
2663	}
2664}
2665
2666void
2667tdsigcleanup(struct thread *td)
2668{
2669	struct proc *p;
2670	sigset_t unblocked;
2671
2672	p = td->td_proc;
2673	PROC_LOCK_ASSERT(p, MA_OWNED);
2674
2675	sigqueue_flush(&td->td_sigqueue);
2676	if (p->p_numthreads == 1)
2677		return;
2678
2679	/*
2680	 * Since we cannot handle signals, notify signal post code
2681	 * about this by filling the sigmask.
2682	 *
2683	 * Also, if needed, wake up thread(s) that do not block the
2684	 * same signals as the exiting thread, since the thread might
2685	 * have been selected for delivery and woken up.
2686	 */
2687	SIGFILLSET(unblocked);
2688	SIGSETNAND(unblocked, td->td_sigmask);
2689	SIGFILLSET(td->td_sigmask);
2690	reschedule_signals(p, unblocked, 0);
2691
2692}
2693
2694static int
2695sigdeferstop_curr_flags(int cflags)
2696{
2697
2698	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
2699	    (cflags & TDF_SBDRY) != 0);
2700	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
2701}
2702
2703/*
2704 * Defer the delivery of SIGSTOP for the current thread, according to
2705 * the requested mode.  Returns previous flags, which must be restored
2706 * by sigallowstop().
2707 *
2708 * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
2709 * cleared by the current thread, which allow the lock-less read-only
2710 * accesses below.
2711 */
2712int
2713sigdeferstop_impl(int mode)
2714{
2715	struct thread *td;
2716	int cflags, nflags;
2717
2718	td = curthread;
2719	cflags = sigdeferstop_curr_flags(td->td_flags);
2720	switch (mode) {
2721	case SIGDEFERSTOP_NOP:
2722		nflags = cflags;
2723		break;
2724	case SIGDEFERSTOP_OFF:
2725		nflags = 0;
2726		break;
2727	case SIGDEFERSTOP_SILENT:
2728		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
2729		break;
2730	case SIGDEFERSTOP_EINTR:
2731		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
2732		break;
2733	case SIGDEFERSTOP_ERESTART:
2734		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
2735		break;
2736	default:
2737		panic("sigdeferstop: invalid mode %x", mode);
2738		break;
2739	}
2740	if (cflags == nflags)
2741		return (SIGDEFERSTOP_VAL_NCHG);
2742	thread_lock(td);
2743	td->td_flags = (td->td_flags & ~cflags) | nflags;
2744	thread_unlock(td);
2745	return (cflags);
2746}
2747
2748/*
2749 * Restores the STOP handling mode, typically permitting the delivery
2750 * of SIGSTOP for the current thread.  This does not immediately
2751 * suspend if a stop was posted.  Instead, the thread will suspend
2752 * either via ast() or a subsequent interruptible sleep.
2753 */
2754void
2755sigallowstop_impl(int prev)
2756{
2757	struct thread *td;
2758	int cflags;
2759
2760	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
2761	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
2762	    ("sigallowstop: incorrect previous mode %x", prev));
2763	td = curthread;
2764	cflags = sigdeferstop_curr_flags(td->td_flags);
2765	if (cflags != prev) {
2766		thread_lock(td);
2767		td->td_flags = (td->td_flags & ~cflags) | prev;
2768		thread_unlock(td);
2769	}
2770}
2771
2772/*
2773 * If the current process has received a signal (should be caught or cause
2774 * termination, should interrupt current syscall), return the signal number.
2775 * Stop signals with default action are processed immediately, then cleared;
2776 * they aren't returned.  This is checked after each entry to the system for
2777 * a syscall or trap (though this can usually be done without calling issignal
2778 * by checking the pending signal masks in cursig.) The normal call
2779 * sequence is
2780 *
2781 *	while (sig = cursig(curthread))
2782 *		postsig(sig);
2783 */
2784static int
2785issignal(struct thread *td)
2786{
2787	struct proc *p;
2788	struct sigacts *ps;
2789	struct sigqueue *queue;
2790	sigset_t sigpending;
2791	int prop, sig, traced;
2792	ksiginfo_t ksi;
2793
2794	p = td->td_proc;
2795	ps = p->p_sigacts;
2796	mtx_assert(&ps->ps_mtx, MA_OWNED);
2797	PROC_LOCK_ASSERT(p, MA_OWNED);
2798	for (;;) {
2799		traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
2800
2801		sigpending = td->td_sigqueue.sq_signals;
2802		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
2803		SIGSETNAND(sigpending, td->td_sigmask);
2804
2805		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
2806		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2807			SIG_STOPSIGMASK(sigpending);
2808		if (SIGISEMPTY(sigpending))	/* no signal to send */
2809			return (0);
2810		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
2811		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
2812		    SIGISMEMBER(sigpending, SIGSTOP)) {
2813			/*
2814			 * If debugger just attached, always consume
2815			 * SIGSTOP from ptrace(PT_ATTACH) first, to
2816			 * execute the debugger attach ritual in
2817			 * order.
2818			 */
2819			sig = SIGSTOP;
2820			td->td_dbgflags |= TDB_FSTP;
2821		} else {
2822			sig = sig_ffs(&sigpending);
2823		}
2824
2825		if (p->p_stops & S_SIG) {
2826			mtx_unlock(&ps->ps_mtx);
2827			stopevent(p, S_SIG, sig);
2828			mtx_lock(&ps->ps_mtx);
2829		}
2830
2831		/*
2832		 * We should see pending but ignored signals
2833		 * only if P_TRACED was on when they were posted.
2834		 */
2835		if (SIGISMEMBER(ps->ps_sigignore, sig) && (traced == 0)) {
2836			sigqueue_delete(&td->td_sigqueue, sig);
2837			sigqueue_delete(&p->p_sigqueue, sig);
2838			continue;
2839		}
2840		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
2841			/*
2842			 * If traced, always stop.
2843			 * Remove old signal from queue before the stop.
2844			 * XXX shrug off debugger, it causes siginfo to
2845			 * be thrown away.
2846			 */
2847			queue = &td->td_sigqueue;
2848			ksiginfo_init(&ksi);
2849			if (sigqueue_get(queue, sig, &ksi) == 0) {
2850				queue = &p->p_sigqueue;
2851				sigqueue_get(queue, sig, &ksi);
2852			}
2853			td->td_si = ksi.ksi_info;
2854
2855			mtx_unlock(&ps->ps_mtx);
2856			sig = ptracestop(td, sig, &ksi);
2857			mtx_lock(&ps->ps_mtx);
2858
2859			/*
2860			 * Keep looking if the debugger discarded or
2861			 * replaced the signal.
2862			 */
2863			if (sig == 0)
2864				continue;
2865
2866			/*
2867			 * If the signal became masked, re-queue it.
2868			 */
2869			if (SIGISMEMBER(td->td_sigmask, sig)) {
2870				ksi.ksi_flags |= KSI_HEAD;
2871				sigqueue_add(&p->p_sigqueue, sig, &ksi);
2872				continue;
2873			}
2874
2875			/*
2876			 * If the traced bit got turned off, requeue
2877			 * the signal and go back up to the top to
2878			 * rescan signals.  This ensures that p_sig*
2879			 * and p_sigact are consistent.
2880			 */
2881			if ((p->p_flag & P_TRACED) == 0) {
2882				ksi.ksi_flags |= KSI_HEAD;
2883				sigqueue_add(queue, sig, &ksi);
2884				continue;
2885			}
2886		}
2887
2888		prop = sigprop(sig);
2889
2890		/*
2891		 * Decide whether the signal should be returned.
2892		 * Return the signal's number, or fall through
2893		 * to clear it from the pending mask.
2894		 */
2895		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2896
2897		case (intptr_t)SIG_DFL:
2898			/*
2899			 * Don't take default actions on system processes.
2900			 */
2901			if (p->p_pid <= 1) {
2902#ifdef DIAGNOSTIC
2903				/*
2904				 * Are you sure you want to ignore SIGSEGV
2905				 * in init? XXX
2906				 */
2907				printf("Process (pid %lu) got signal %d\n",
2908					(u_long)p->p_pid, sig);
2909#endif
2910				break;		/* == ignore */
2911			}
2912			/*
2913			 * If there is a pending stop signal to process with
2914			 * default action, stop here, then clear the signal.
2915			 * Traced or exiting processes should ignore stops.
2916			 * Additionally, a member of an orphaned process group
2917			 * should ignore tty stops.
2918			 */
2919			if (prop & SA_STOP) {
2920				if (p->p_flag &
2921				    (P_TRACED | P_WEXIT | P_SINGLE_EXIT) ||
2922				    (p->p_pgrp->pg_jobc == 0 &&
2923				     prop & SA_TTYSTOP))
2924					break;	/* == ignore */
2925				if (TD_SBDRY_INTR(td)) {
2926					KASSERT((td->td_flags & TDF_SBDRY) != 0,
2927					    ("lost TDF_SBDRY"));
2928					return (-1);
2929				}
2930				mtx_unlock(&ps->ps_mtx);
2931				WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2932				    &p->p_mtx.lock_object, "Catching SIGSTOP");
2933				sigqueue_delete(&td->td_sigqueue, sig);
2934				sigqueue_delete(&p->p_sigqueue, sig);
2935				p->p_flag |= P_STOPPED_SIG;
2936				p->p_xsig = sig;
2937				PROC_SLOCK(p);
2938				sig_suspend_threads(td, p, 0);
2939				thread_suspend_switch(td, p);
2940				PROC_SUNLOCK(p);
2941				mtx_lock(&ps->ps_mtx);
2942				goto next;
2943			} else if (prop & SA_IGNORE) {
2944				/*
2945				 * Except for SIGCONT, shouldn't get here.
2946				 * Default action is to ignore; drop it.
2947				 */
2948				break;		/* == ignore */
2949			} else
2950				return (sig);
2951			/*NOTREACHED*/
2952
2953		case (intptr_t)SIG_IGN:
2954			/*
2955			 * Masking above should prevent us ever trying
2956			 * to take action on an ignored signal other
2957			 * than SIGCONT, unless process is traced.
2958			 */
2959			if ((prop & SA_CONT) == 0 &&
2960			    (p->p_flag & P_TRACED) == 0)
2961				printf("issignal\n");
2962			break;		/* == ignore */
2963
2964		default:
2965			/*
2966			 * This signal has an action, let
2967			 * postsig() process it.
2968			 */
2969			return (sig);
2970		}
2971		sigqueue_delete(&td->td_sigqueue, sig);	/* take the signal! */
2972		sigqueue_delete(&p->p_sigqueue, sig);
2973next:;
2974	}
2975	/* NOTREACHED */
2976}
2977
2978void
2979thread_stopped(struct proc *p)
2980{
2981	int n;
2982
2983	PROC_LOCK_ASSERT(p, MA_OWNED);
2984	PROC_SLOCK_ASSERT(p, MA_OWNED);
2985	n = p->p_suspcount;
2986	if (p == curproc)
2987		n++;
2988	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
2989		PROC_SUNLOCK(p);
2990		p->p_flag &= ~P_WAITED;
2991		PROC_LOCK(p->p_pptr);
2992		childproc_stopped(p, (p->p_flag & P_TRACED) ?
2993			CLD_TRAPPED : CLD_STOPPED);
2994		PROC_UNLOCK(p->p_pptr);
2995		PROC_SLOCK(p);
2996	}
2997}
2998
2999/*
3000 * Take the action for the specified signal
3001 * from the current set of pending signals.
3002 */
3003int
3004postsig(int sig)
3005{
3006	struct thread *td;
3007	struct proc *p;
3008	struct sigacts *ps;
3009	sig_t action;
3010	ksiginfo_t ksi;
3011	sigset_t returnmask;
3012
3013	KASSERT(sig != 0, ("postsig"));
3014
3015	td = curthread;
3016	p = td->td_proc;
3017	PROC_LOCK_ASSERT(p, MA_OWNED);
3018	ps = p->p_sigacts;
3019	mtx_assert(&ps->ps_mtx, MA_OWNED);
3020	ksiginfo_init(&ksi);
3021	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3022	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3023		return (0);
3024	ksi.ksi_signo = sig;
3025	if (ksi.ksi_code == SI_TIMER)
3026		itimer_accept(p, ksi.ksi_timerid, &ksi);
3027	action = ps->ps_sigact[_SIG_IDX(sig)];
3028#ifdef KTRACE
3029	if (KTRPOINT(td, KTR_PSIG))
3030		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3031		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3032#endif
3033	if ((p->p_stops & S_SIG) != 0) {
3034		mtx_unlock(&ps->ps_mtx);
3035		stopevent(p, S_SIG, sig);
3036		mtx_lock(&ps->ps_mtx);
3037	}
3038
3039	if (action == SIG_DFL) {
3040		/*
3041		 * Default action, where the default is to kill
3042		 * the process.  (Other cases were ignored above.)
3043		 */
3044		mtx_unlock(&ps->ps_mtx);
3045		proc_td_siginfo_capture(td, &ksi.ksi_info);
3046		sigexit(td, sig);
3047		/* NOTREACHED */
3048	} else {
3049		/*
3050		 * If we get here, the signal must be caught.
3051		 */
3052		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3053		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3054		    ("postsig action: blocked sig %d", sig));
3055
3056		/*
3057		 * Set the new mask value and also defer further
3058		 * occurrences of this signal.
3059		 *
3060		 * Special case: user has done a sigsuspend.  Here the
3061		 * current mask is not of interest, but rather the
3062		 * mask from before the sigsuspend is what we want
3063		 * restored after the signal processing is completed.
3064		 */
3065		if (td->td_pflags & TDP_OLDMASK) {
3066			returnmask = td->td_oldsigmask;
3067			td->td_pflags &= ~TDP_OLDMASK;
3068		} else
3069			returnmask = td->td_sigmask;
3070
3071		if (p->p_sig == sig) {
3072			p->p_code = 0;
3073			p->p_sig = 0;
3074		}
3075		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3076		postsig_done(sig, td, ps);
3077	}
3078	return (1);
3079}
3080
3081/*
3082 * Kill the current process for stated reason.
3083 */
3084void
3085killproc(struct proc *p, char *why)
3086{
3087
3088	PROC_LOCK_ASSERT(p, MA_OWNED);
3089	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3090	    p->p_comm);
3091	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n", p->p_pid,
3092	    p->p_comm, p->p_ucred ? p->p_ucred->cr_uid : -1, why);
3093	p->p_flag |= P_WKILLED;
3094	kern_psignal(p, SIGKILL);
3095}
3096
3097/*
3098 * Force the current process to exit with the specified signal, dumping core
3099 * if appropriate.  We bypass the normal tests for masked and caught signals,
3100 * allowing unrecoverable failures to terminate the process without changing
3101 * signal state.  Mark the accounting record with the signal termination.
3102 * If dumping core, save the signal number for the debugger.  Calls exit and
3103 * does not return.
3104 */
3105void
3106sigexit(struct thread *td, int sig)
3107{
3108	struct proc *p = td->td_proc;
3109
3110	PROC_LOCK_ASSERT(p, MA_OWNED);
3111	p->p_acflag |= AXSIG;
3112	/*
3113	 * We must be single-threading to generate a core dump.  This
3114	 * ensures that the registers in the core file are up-to-date.
3115	 * Also, the ELF dump handler assumes that the thread list doesn't
3116	 * change out from under it.
3117	 *
3118	 * XXX If another thread attempts to single-thread before us
3119	 *     (e.g. via fork()), we won't get a dump at all.
3120	 */
3121	if ((sigprop(sig) & SA_CORE) && thread_single(p, SINGLE_NO_EXIT) == 0) {
3122		p->p_sig = sig;
3123		/*
3124		 * Log signals which would cause core dumps
3125		 * (Log as LOG_INFO to appease those who don't want
3126		 * these messages.)
3127		 * XXX : Todo, as well as euid, write out ruid too
3128		 * Note that coredump() drops proc lock.
3129		 */
3130		if (coredump(td) == 0)
3131			sig |= WCOREFLAG;
3132		if (kern_logsigexit)
3133			log(LOG_INFO,
3134			    "pid %d (%s), uid %d: exited on signal %d%s\n",
3135			    p->p_pid, p->p_comm,
3136			    td->td_ucred ? td->td_ucred->cr_uid : -1,
3137			    sig &~ WCOREFLAG,
3138			    sig & WCOREFLAG ? " (core dumped)" : "");
3139	} else
3140		PROC_UNLOCK(p);
3141	exit1(td, 0, sig);
3142	/* NOTREACHED */
3143}
3144
3145/*
3146 * Send queued SIGCHLD to parent when child process's state
3147 * is changed.
3148 */
3149static void
3150sigparent(struct proc *p, int reason, int status)
3151{
3152	PROC_LOCK_ASSERT(p, MA_OWNED);
3153	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3154
3155	if (p->p_ksi != NULL) {
3156		p->p_ksi->ksi_signo  = SIGCHLD;
3157		p->p_ksi->ksi_code   = reason;
3158		p->p_ksi->ksi_status = status;
3159		p->p_ksi->ksi_pid    = p->p_pid;
3160		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3161		if (KSI_ONQ(p->p_ksi))
3162			return;
3163	}
3164	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3165}
3166
3167static void
3168childproc_jobstate(struct proc *p, int reason, int sig)
3169{
3170	struct sigacts *ps;
3171
3172	PROC_LOCK_ASSERT(p, MA_OWNED);
3173	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3174
3175	/*
3176	 * Wake up parent sleeping in kern_wait(), also send
3177	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3178	 * that parent will awake, because parent may masked
3179	 * the signal.
3180	 */
3181	p->p_pptr->p_flag |= P_STATCHILD;
3182	wakeup(p->p_pptr);
3183
3184	ps = p->p_pptr->p_sigacts;
3185	mtx_lock(&ps->ps_mtx);
3186	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3187		mtx_unlock(&ps->ps_mtx);
3188		sigparent(p, reason, sig);
3189	} else
3190		mtx_unlock(&ps->ps_mtx);
3191}
3192
3193void
3194childproc_stopped(struct proc *p, int reason)
3195{
3196
3197	childproc_jobstate(p, reason, p->p_xsig);
3198}
3199
3200void
3201childproc_continued(struct proc *p)
3202{
3203	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3204}
3205
3206void
3207childproc_exited(struct proc *p)
3208{
3209	int reason, status;
3210
3211	if (WCOREDUMP(p->p_xsig)) {
3212		reason = CLD_DUMPED;
3213		status = WTERMSIG(p->p_xsig);
3214	} else if (WIFSIGNALED(p->p_xsig)) {
3215		reason = CLD_KILLED;
3216		status = WTERMSIG(p->p_xsig);
3217	} else {
3218		reason = CLD_EXITED;
3219		status = p->p_xexit;
3220	}
3221	/*
3222	 * XXX avoid calling wakeup(p->p_pptr), the work is
3223	 * done in exit1().
3224	 */
3225	sigparent(p, reason, status);
3226}
3227
3228/*
3229 * We only have 1 character for the core count in the format
3230 * string, so the range will be 0-9
3231 */
3232#define MAX_NUM_CORES 10
3233static int num_cores = 5;
3234
3235static int
3236sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3237{
3238	int error;
3239	int new_val;
3240
3241	new_val = num_cores;
3242	error = sysctl_handle_int(oidp, &new_val, 0, req);
3243	if (error != 0 || req->newptr == NULL)
3244		return (error);
3245	if (new_val > MAX_NUM_CORES)
3246		new_val = MAX_NUM_CORES;
3247	if (new_val < 0)
3248		new_val = 0;
3249	num_cores = new_val;
3250	return (0);
3251}
3252SYSCTL_PROC(_debug, OID_AUTO, ncores, CTLTYPE_INT|CTLFLAG_RW,
3253	    0, sizeof(int), sysctl_debug_num_cores_check, "I", "");
3254
3255#define	GZ_SUFFIX	".gz"
3256
3257#ifdef GZIO
3258static int compress_user_cores = 1;
3259SYSCTL_INT(_kern, OID_AUTO, compress_user_cores, CTLFLAG_RWTUN,
3260    &compress_user_cores, 0, "Compression of user corefiles");
3261
3262int compress_user_cores_gzlevel = 6;
3263SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_gzlevel, CTLFLAG_RWTUN,
3264    &compress_user_cores_gzlevel, 0, "Corefile gzip compression level");
3265#else
3266static int compress_user_cores = 0;
3267#endif
3268
3269/*
3270 * Protect the access to corefilename[] by allproc_lock.
3271 */
3272#define	corefilename_lock	allproc_lock
3273
3274static char corefilename[MAXPATHLEN] = {"%N.core"};
3275TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3276
3277static int
3278sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3279{
3280	int error;
3281
3282	sx_xlock(&corefilename_lock);
3283	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3284	    req);
3285	sx_xunlock(&corefilename_lock);
3286
3287	return (error);
3288}
3289SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3290    CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3291    "Process corefile name format string");
3292
3293/*
3294 * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3295 * Expand the name described in corefilename, using name, uid, and pid
3296 * and open/create core file.
3297 * corefilename is a printf-like string, with three format specifiers:
3298 *	%N	name of process ("name")
3299 *	%P	process id (pid)
3300 *	%U	user id (uid)
3301 * For example, "%N.core" is the default; they can be disabled completely
3302 * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3303 * This is controlled by the sysctl variable kern.corefile (see above).
3304 */
3305static int
3306corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3307    int compress, struct vnode **vpp, char **namep)
3308{
3309	struct nameidata nd;
3310	struct sbuf sb;
3311	const char *format;
3312	char *hostname, *name;
3313	int indexpos, i, error, cmode, flags, oflags;
3314
3315	hostname = NULL;
3316	format = corefilename;
3317	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3318	indexpos = -1;
3319	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3320	sx_slock(&corefilename_lock);
3321	for (i = 0; format[i] != '\0'; i++) {
3322		switch (format[i]) {
3323		case '%':	/* Format character */
3324			i++;
3325			switch (format[i]) {
3326			case '%':
3327				sbuf_putc(&sb, '%');
3328				break;
3329			case 'H':	/* hostname */
3330				if (hostname == NULL) {
3331					hostname = malloc(MAXHOSTNAMELEN,
3332					    M_TEMP, M_WAITOK);
3333				}
3334				getcredhostname(td->td_ucred, hostname,
3335				    MAXHOSTNAMELEN);
3336				sbuf_printf(&sb, "%s", hostname);
3337				break;
3338			case 'I':	/* autoincrementing index */
3339				sbuf_printf(&sb, "0");
3340				indexpos = sbuf_len(&sb) - 1;
3341				break;
3342			case 'N':	/* process name */
3343				sbuf_printf(&sb, "%s", comm);
3344				break;
3345			case 'P':	/* process id */
3346				sbuf_printf(&sb, "%u", pid);
3347				break;
3348			case 'U':	/* user id */
3349				sbuf_printf(&sb, "%u", uid);
3350				break;
3351			default:
3352				log(LOG_ERR,
3353				    "Unknown format character %c in "
3354				    "corename `%s'\n", format[i], format);
3355				break;
3356			}
3357			break;
3358		default:
3359			sbuf_putc(&sb, format[i]);
3360			break;
3361		}
3362	}
3363	sx_sunlock(&corefilename_lock);
3364	free(hostname, M_TEMP);
3365	if (compress)
3366		sbuf_printf(&sb, GZ_SUFFIX);
3367	if (sbuf_error(&sb) != 0) {
3368		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3369		    "long\n", (long)pid, comm, (u_long)uid);
3370		sbuf_delete(&sb);
3371		free(name, M_TEMP);
3372		return (ENOMEM);
3373	}
3374	sbuf_finish(&sb);
3375	sbuf_delete(&sb);
3376
3377	cmode = S_IRUSR | S_IWUSR;
3378	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3379	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3380
3381	/*
3382	 * If the core format has a %I in it, then we need to check
3383	 * for existing corefiles before returning a name.
3384	 * To do this we iterate over 0..num_cores to find a
3385	 * non-existing core file name to use.
3386	 */
3387	if (indexpos != -1) {
3388		for (i = 0; i < num_cores; i++) {
3389			flags = O_CREAT | O_EXCL | FWRITE | O_NOFOLLOW;
3390			name[indexpos] = '0' + i;
3391			NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3392			error = vn_open_cred(&nd, &flags, cmode, oflags,
3393			    td->td_ucred, NULL);
3394			if (error) {
3395				if (error == EEXIST)
3396					continue;
3397				log(LOG_ERR,
3398				    "pid %d (%s), uid (%u):  Path `%s' failed "
3399				    "on initial open test, error = %d\n",
3400				    pid, comm, uid, name, error);
3401			}
3402			goto out;
3403		}
3404	}
3405
3406	flags = O_CREAT | FWRITE | O_NOFOLLOW;
3407	NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3408	error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred, NULL);
3409out:
3410	if (error) {
3411#ifdef AUDIT
3412		audit_proc_coredump(td, name, error);
3413#endif
3414		free(name, M_TEMP);
3415		return (error);
3416	}
3417	NDFREE(&nd, NDF_ONLY_PNBUF);
3418	*vpp = nd.ni_vp;
3419	*namep = name;
3420	return (0);
3421}
3422
3423static int
3424coredump_sanitise_path(const char *path)
3425{
3426	size_t i;
3427
3428	/*
3429	 * Only send a subset of ASCII to devd(8) because it
3430	 * might pass these strings to sh -c.
3431	 */
3432	for (i = 0; path[i]; i++)
3433		if (!(isalpha(path[i]) || isdigit(path[i])) &&
3434		    path[i] != '/' && path[i] != '.' &&
3435		    path[i] != '-')
3436			return (0);
3437
3438	return (1);
3439}
3440
3441/*
3442 * Dump a process' core.  The main routine does some
3443 * policy checking, and creates the name of the coredump;
3444 * then it passes on a vnode and a size limit to the process-specific
3445 * coredump routine if there is one; if there _is not_ one, it returns
3446 * ENOSYS; otherwise it returns the error from the process-specific routine.
3447 */
3448
3449static int
3450coredump(struct thread *td)
3451{
3452	struct proc *p = td->td_proc;
3453	struct ucred *cred = td->td_ucred;
3454	struct vnode *vp;
3455	struct flock lf;
3456	struct vattr vattr;
3457	int error, error1, locked;
3458	char *name;			/* name of corefile */
3459	void *rl_cookie;
3460	off_t limit;
3461	char *data = NULL;
3462	char *fullpath, *freepath = NULL;
3463	size_t len;
3464	static const char comm_name[] = "comm=";
3465	static const char core_name[] = "core=";
3466
3467	PROC_LOCK_ASSERT(p, MA_OWNED);
3468	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3469	_STOPEVENT(p, S_CORE, 0);
3470
3471	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
3472	    (p->p_flag2 & P2_NOTRACE) != 0) {
3473		PROC_UNLOCK(p);
3474		return (EFAULT);
3475	}
3476
3477	/*
3478	 * Note that the bulk of limit checking is done after
3479	 * the corefile is created.  The exception is if the limit
3480	 * for corefiles is 0, in which case we don't bother
3481	 * creating the corefile at all.  This layout means that
3482	 * a corefile is truncated instead of not being created,
3483	 * if it is larger than the limit.
3484	 */
3485	limit = (off_t)lim_cur(td, RLIMIT_CORE);
3486	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
3487		PROC_UNLOCK(p);
3488		return (EFBIG);
3489	}
3490	PROC_UNLOCK(p);
3491
3492	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
3493	    compress_user_cores, &vp, &name);
3494	if (error != 0)
3495		return (error);
3496
3497	/*
3498	 * Don't dump to non-regular files or files with links.
3499	 * Do not dump into system files.
3500	 */
3501	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
3502	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0) {
3503		VOP_UNLOCK(vp, 0);
3504		error = EFAULT;
3505		goto out;
3506	}
3507
3508	VOP_UNLOCK(vp, 0);
3509
3510	/* Postpone other writers, including core dumps of other processes. */
3511	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
3512
3513	lf.l_whence = SEEK_SET;
3514	lf.l_start = 0;
3515	lf.l_len = 0;
3516	lf.l_type = F_WRLCK;
3517	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
3518
3519	VATTR_NULL(&vattr);
3520	vattr.va_size = 0;
3521	if (set_core_nodump_flag)
3522		vattr.va_flags = UF_NODUMP;
3523	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3524	VOP_SETATTR(vp, &vattr, cred);
3525	VOP_UNLOCK(vp, 0);
3526	PROC_LOCK(p);
3527	p->p_acflag |= ACORE;
3528	PROC_UNLOCK(p);
3529
3530	if (p->p_sysent->sv_coredump != NULL) {
3531		error = p->p_sysent->sv_coredump(td, vp, limit,
3532		    compress_user_cores ? IMGACT_CORE_COMPRESS : 0);
3533	} else {
3534		error = ENOSYS;
3535	}
3536
3537	if (locked) {
3538		lf.l_type = F_UNLCK;
3539		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3540	}
3541	vn_rangelock_unlock(vp, rl_cookie);
3542
3543	/*
3544	 * Notify the userland helper that a process triggered a core dump.
3545	 * This allows the helper to run an automated debugging session.
3546	 */
3547	if (error != 0 || coredump_devctl == 0)
3548		goto out;
3549	len = MAXPATHLEN * 2 + sizeof(comm_name) - 1 +
3550	    sizeof(' ') + sizeof(core_name) - 1;
3551	data = malloc(len, M_TEMP, M_WAITOK);
3552	if (vn_fullpath_global(td, p->p_textvp, &fullpath, &freepath) != 0)
3553		goto out;
3554	if (!coredump_sanitise_path(fullpath))
3555		goto out;
3556	snprintf(data, len, "%s%s ", comm_name, fullpath);
3557	free(freepath, M_TEMP);
3558	freepath = NULL;
3559	if (vn_fullpath_global(td, vp, &fullpath, &freepath) != 0)
3560		goto out;
3561	if (!coredump_sanitise_path(fullpath))
3562		goto out;
3563	strlcat(data, core_name, len);
3564	strlcat(data, fullpath, len);
3565	devctl_notify("kernel", "signal", "coredump", data);
3566out:
3567	error1 = vn_close(vp, FWRITE, cred, td);
3568	if (error == 0)
3569		error = error1;
3570#ifdef AUDIT
3571	audit_proc_coredump(td, name, error);
3572#endif
3573	free(freepath, M_TEMP);
3574	free(data, M_TEMP);
3575	free(name, M_TEMP);
3576	return (error);
3577}
3578
3579/*
3580 * Nonexistent system call-- signal process (may want to handle it).  Flag
3581 * error in case process won't see signal immediately (blocked or ignored).
3582 */
3583#ifndef _SYS_SYSPROTO_H_
3584struct nosys_args {
3585	int	dummy;
3586};
3587#endif
3588/* ARGSUSED */
3589int
3590nosys(struct thread *td, struct nosys_args *args)
3591{
3592	struct proc *p;
3593
3594	p = td->td_proc;
3595
3596	PROC_LOCK(p);
3597	tdsignal(td, SIGSYS);
3598	PROC_UNLOCK(p);
3599	if (kern_lognosys == 1 || kern_lognosys == 3) {
3600		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
3601		    td->td_sa.code);
3602	}
3603	if (kern_lognosys == 2 || kern_lognosys == 3) {
3604		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
3605		    td->td_sa.code);
3606	}
3607	return (ENOSYS);
3608}
3609
3610/*
3611 * Send a SIGIO or SIGURG signal to a process or process group using stored
3612 * credentials rather than those of the current process.
3613 */
3614void
3615pgsigio(struct sigio **sigiop, int sig, int checkctty)
3616{
3617	ksiginfo_t ksi;
3618	struct sigio *sigio;
3619
3620	ksiginfo_init(&ksi);
3621	ksi.ksi_signo = sig;
3622	ksi.ksi_code = SI_KERNEL;
3623
3624	SIGIO_LOCK();
3625	sigio = *sigiop;
3626	if (sigio == NULL) {
3627		SIGIO_UNLOCK();
3628		return;
3629	}
3630	if (sigio->sio_pgid > 0) {
3631		PROC_LOCK(sigio->sio_proc);
3632		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
3633			kern_psignal(sigio->sio_proc, sig);
3634		PROC_UNLOCK(sigio->sio_proc);
3635	} else if (sigio->sio_pgid < 0) {
3636		struct proc *p;
3637
3638		PGRP_LOCK(sigio->sio_pgrp);
3639		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
3640			PROC_LOCK(p);
3641			if (p->p_state == PRS_NORMAL &&
3642			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
3643			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
3644				kern_psignal(p, sig);
3645			PROC_UNLOCK(p);
3646		}
3647		PGRP_UNLOCK(sigio->sio_pgrp);
3648	}
3649	SIGIO_UNLOCK();
3650}
3651
3652static int
3653filt_sigattach(struct knote *kn)
3654{
3655	struct proc *p = curproc;
3656
3657	kn->kn_ptr.p_proc = p;
3658	kn->kn_flags |= EV_CLEAR;		/* automatically set */
3659
3660	knlist_add(p->p_klist, kn, 0);
3661
3662	return (0);
3663}
3664
3665static void
3666filt_sigdetach(struct knote *kn)
3667{
3668	struct proc *p = kn->kn_ptr.p_proc;
3669
3670	knlist_remove(p->p_klist, kn, 0);
3671}
3672
3673/*
3674 * signal knotes are shared with proc knotes, so we apply a mask to
3675 * the hint in order to differentiate them from process hints.  This
3676 * could be avoided by using a signal-specific knote list, but probably
3677 * isn't worth the trouble.
3678 */
3679static int
3680filt_signal(struct knote *kn, long hint)
3681{
3682
3683	if (hint & NOTE_SIGNAL) {
3684		hint &= ~NOTE_SIGNAL;
3685
3686		if (kn->kn_id == hint)
3687			kn->kn_data++;
3688	}
3689	return (kn->kn_data != 0);
3690}
3691
3692struct sigacts *
3693sigacts_alloc(void)
3694{
3695	struct sigacts *ps;
3696
3697	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
3698	refcount_init(&ps->ps_refcnt, 1);
3699	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
3700	return (ps);
3701}
3702
3703void
3704sigacts_free(struct sigacts *ps)
3705{
3706
3707	if (refcount_release(&ps->ps_refcnt) == 0)
3708		return;
3709	mtx_destroy(&ps->ps_mtx);
3710	free(ps, M_SUBPROC);
3711}
3712
3713struct sigacts *
3714sigacts_hold(struct sigacts *ps)
3715{
3716
3717	refcount_acquire(&ps->ps_refcnt);
3718	return (ps);
3719}
3720
3721void
3722sigacts_copy(struct sigacts *dest, struct sigacts *src)
3723{
3724
3725	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
3726	mtx_lock(&src->ps_mtx);
3727	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
3728	mtx_unlock(&src->ps_mtx);
3729}
3730
3731int
3732sigacts_shared(struct sigacts *ps)
3733{
3734
3735	return (ps->ps_refcnt > 1);
3736}
3737