1/*-
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 *    notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 *    notice, this list of conditions and the following disclaimer in the
17 *    documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 *    may be used to endorse or promote products derived from this software
20 *    without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
35 */
36
37#include <sys/cdefs.h>
38__FBSDID("$FreeBSD: stable/11/sys/kern/kern_sig.c 365864 2020-09-17 23:14:17Z brooks $");
39
40#include "opt_compat.h"
41#include "opt_gzio.h"
42#include "opt_ktrace.h"
43
44#include <sys/param.h>
45#include <sys/ctype.h>
46#include <sys/systm.h>
47#include <sys/signalvar.h>
48#include <sys/vnode.h>
49#include <sys/acct.h>
50#include <sys/bus.h>
51#include <sys/capsicum.h>
52#include <sys/condvar.h>
53#include <sys/event.h>
54#include <sys/fcntl.h>
55#include <sys/imgact.h>
56#include <sys/kernel.h>
57#include <sys/ktr.h>
58#include <sys/ktrace.h>
59#include <sys/lock.h>
60#include <sys/malloc.h>
61#include <sys/mutex.h>
62#include <sys/refcount.h>
63#include <sys/namei.h>
64#include <sys/proc.h>
65#include <sys/procdesc.h>
66#include <sys/ptrace.h>
67#include <sys/posix4.h>
68#include <sys/pioctl.h>
69#include <sys/racct.h>
70#include <sys/resourcevar.h>
71#include <sys/sdt.h>
72#include <sys/sbuf.h>
73#include <sys/sleepqueue.h>
74#include <sys/smp.h>
75#include <sys/stat.h>
76#include <sys/sx.h>
77#include <sys/syscallsubr.h>
78#include <sys/sysctl.h>
79#include <sys/sysent.h>
80#include <sys/syslog.h>
81#include <sys/sysproto.h>
82#include <sys/timers.h>
83#include <sys/unistd.h>
84#include <sys/wait.h>
85#include <vm/vm.h>
86#include <vm/vm_extern.h>
87#include <vm/uma.h>
88
89#include <sys/jail.h>
90
91#include <machine/cpu.h>
92
93#include <security/audit/audit.h>
94
95#define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
96
97SDT_PROVIDER_DECLARE(proc);
98SDT_PROBE_DEFINE3(proc, , , signal__send,
99    "struct thread *", "struct proc *", "int");
100SDT_PROBE_DEFINE2(proc, , , signal__clear,
101    "int", "ksiginfo_t *");
102SDT_PROBE_DEFINE3(proc, , , signal__discard,
103    "struct thread *", "struct proc *", "int");
104
105static int	coredump(struct thread *);
106static int	killpg1(struct thread *td, int sig, int pgid, int all,
107		    ksiginfo_t *ksi);
108static int	issignal(struct thread *td);
109static int	sigprop(int sig);
110static void	tdsigwakeup(struct thread *, int, sig_t, int);
111static int	sig_suspend_threads(struct thread *, struct proc *, int);
112static int	filt_sigattach(struct knote *kn);
113static void	filt_sigdetach(struct knote *kn);
114static int	filt_signal(struct knote *kn, long hint);
115static struct thread *sigtd(struct proc *p, int sig, int prop);
116static void	sigqueue_start(void);
117
118static uma_zone_t	ksiginfo_zone = NULL;
119struct filterops sig_filtops = {
120	.f_isfd = 0,
121	.f_attach = filt_sigattach,
122	.f_detach = filt_sigdetach,
123	.f_event = filt_signal,
124};
125
126static int	kern_logsigexit = 1;
127SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
128    &kern_logsigexit, 0,
129    "Log processes quitting on abnormal signals to syslog(3)");
130
131static int	kern_forcesigexit = 1;
132SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
133    &kern_forcesigexit, 0, "Force trap signal to be handled");
134
135static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW, 0,
136    "POSIX real time signal");
137
138static int	max_pending_per_proc = 128;
139SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
140    &max_pending_per_proc, 0, "Max pending signals per proc");
141
142static int	preallocate_siginfo = 1024;
143SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
144    &preallocate_siginfo, 0, "Preallocated signal memory size");
145
146static int	signal_overflow = 0;
147SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
148    &signal_overflow, 0, "Number of signals overflew");
149
150static int	signal_alloc_fail = 0;
151SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
152    &signal_alloc_fail, 0, "signals failed to be allocated");
153
154static int	kern_lognosys = 0;
155SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
156    "Log invalid syscalls");
157
158SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
159
160/*
161 * Policy -- Can ucred cr1 send SIGIO to process cr2?
162 * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
163 * in the right situations.
164 */
165#define CANSIGIO(cr1, cr2) \
166	((cr1)->cr_uid == 0 || \
167	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
168	    (cr1)->cr_uid == (cr2)->cr_ruid || \
169	    (cr1)->cr_ruid == (cr2)->cr_uid || \
170	    (cr1)->cr_uid == (cr2)->cr_uid)
171
172static int	sugid_coredump;
173SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
174    &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
175
176static int	capmode_coredump;
177SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
178    &capmode_coredump, 0, "Allow processes in capability mode to dump core");
179
180static int	do_coredump = 1;
181SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
182	&do_coredump, 0, "Enable/Disable coredumps");
183
184static int	set_core_nodump_flag = 0;
185SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
186	0, "Enable setting the NODUMP flag on coredump files");
187
188static int	coredump_devctl = 0;
189SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
190	0, "Generate a devctl notification when processes coredump");
191
192/*
193 * Signal properties and actions.
194 * The array below categorizes the signals and their default actions
195 * according to the following properties:
196 */
197#define	SA_KILL		0x01		/* terminates process by default */
198#define	SA_CORE		0x02		/* ditto and coredumps */
199#define	SA_STOP		0x04		/* suspend process */
200#define	SA_TTYSTOP	0x08		/* ditto, from tty */
201#define	SA_IGNORE	0x10		/* ignore by default */
202#define	SA_CONT		0x20		/* continue if suspended */
203#define	SA_CANTMASK	0x40		/* non-maskable, catchable */
204
205static int sigproptbl[NSIG] = {
206	SA_KILL,			/* SIGHUP */
207	SA_KILL,			/* SIGINT */
208	SA_KILL|SA_CORE,		/* SIGQUIT */
209	SA_KILL|SA_CORE,		/* SIGILL */
210	SA_KILL|SA_CORE,		/* SIGTRAP */
211	SA_KILL|SA_CORE,		/* SIGABRT */
212	SA_KILL|SA_CORE,		/* SIGEMT */
213	SA_KILL|SA_CORE,		/* SIGFPE */
214	SA_KILL,			/* SIGKILL */
215	SA_KILL|SA_CORE,		/* SIGBUS */
216	SA_KILL|SA_CORE,		/* SIGSEGV */
217	SA_KILL|SA_CORE,		/* SIGSYS */
218	SA_KILL,			/* SIGPIPE */
219	SA_KILL,			/* SIGALRM */
220	SA_KILL,			/* SIGTERM */
221	SA_IGNORE,			/* SIGURG */
222	SA_STOP,			/* SIGSTOP */
223	SA_STOP|SA_TTYSTOP,		/* SIGTSTP */
224	SA_IGNORE|SA_CONT,		/* SIGCONT */
225	SA_IGNORE,			/* SIGCHLD */
226	SA_STOP|SA_TTYSTOP,		/* SIGTTIN */
227	SA_STOP|SA_TTYSTOP,		/* SIGTTOU */
228	SA_IGNORE,			/* SIGIO */
229	SA_KILL,			/* SIGXCPU */
230	SA_KILL,			/* SIGXFSZ */
231	SA_KILL,			/* SIGVTALRM */
232	SA_KILL,			/* SIGPROF */
233	SA_IGNORE,			/* SIGWINCH  */
234	SA_IGNORE,			/* SIGINFO */
235	SA_KILL,			/* SIGUSR1 */
236	SA_KILL,			/* SIGUSR2 */
237};
238
239static void reschedule_signals(struct proc *p, sigset_t block, int flags);
240
241static void
242sigqueue_start(void)
243{
244	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
245		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
246	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
247	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
248	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
249	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
250}
251
252ksiginfo_t *
253ksiginfo_alloc(int wait)
254{
255	int flags;
256
257	flags = M_ZERO;
258	if (! wait)
259		flags |= M_NOWAIT;
260	if (ksiginfo_zone != NULL)
261		return ((ksiginfo_t *)uma_zalloc(ksiginfo_zone, flags));
262	return (NULL);
263}
264
265void
266ksiginfo_free(ksiginfo_t *ksi)
267{
268	uma_zfree(ksiginfo_zone, ksi);
269}
270
271static __inline int
272ksiginfo_tryfree(ksiginfo_t *ksi)
273{
274	if (!(ksi->ksi_flags & KSI_EXT)) {
275		uma_zfree(ksiginfo_zone, ksi);
276		return (1);
277	}
278	return (0);
279}
280
281void
282sigqueue_init(sigqueue_t *list, struct proc *p)
283{
284	SIGEMPTYSET(list->sq_signals);
285	SIGEMPTYSET(list->sq_kill);
286	SIGEMPTYSET(list->sq_ptrace);
287	TAILQ_INIT(&list->sq_list);
288	list->sq_proc = p;
289	list->sq_flags = SQ_INIT;
290}
291
292/*
293 * Get a signal's ksiginfo.
294 * Return:
295 *	0	-	signal not found
296 *	others	-	signal number
297 */
298static int
299sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
300{
301	struct proc *p = sq->sq_proc;
302	struct ksiginfo *ksi, *next;
303	int count = 0;
304
305	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
306
307	if (!SIGISMEMBER(sq->sq_signals, signo))
308		return (0);
309
310	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
311		count++;
312		SIGDELSET(sq->sq_ptrace, signo);
313		si->ksi_flags |= KSI_PTRACE;
314	}
315	if (SIGISMEMBER(sq->sq_kill, signo)) {
316		count++;
317		if (count == 1)
318			SIGDELSET(sq->sq_kill, signo);
319	}
320
321	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
322		if (ksi->ksi_signo == signo) {
323			if (count == 0) {
324				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
325				ksi->ksi_sigq = NULL;
326				ksiginfo_copy(ksi, si);
327				if (ksiginfo_tryfree(ksi) && p != NULL)
328					p->p_pendingcnt--;
329			}
330			if (++count > 1)
331				break;
332		}
333	}
334
335	if (count <= 1)
336		SIGDELSET(sq->sq_signals, signo);
337	si->ksi_signo = signo;
338	return (signo);
339}
340
341void
342sigqueue_take(ksiginfo_t *ksi)
343{
344	struct ksiginfo *kp;
345	struct proc	*p;
346	sigqueue_t	*sq;
347
348	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
349		return;
350
351	p = sq->sq_proc;
352	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
353	ksi->ksi_sigq = NULL;
354	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
355		p->p_pendingcnt--;
356
357	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
358	     kp = TAILQ_NEXT(kp, ksi_link)) {
359		if (kp->ksi_signo == ksi->ksi_signo)
360			break;
361	}
362	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
363	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
364		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
365}
366
367static int
368sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
369{
370	struct proc *p = sq->sq_proc;
371	struct ksiginfo *ksi;
372	int ret = 0;
373
374	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
375
376	/*
377	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
378	 * for these signals.
379	 */
380	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
381		SIGADDSET(sq->sq_kill, signo);
382		goto out_set_bit;
383	}
384
385	/* directly insert the ksi, don't copy it */
386	if (si->ksi_flags & KSI_INS) {
387		if (si->ksi_flags & KSI_HEAD)
388			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
389		else
390			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
391		si->ksi_sigq = sq;
392		goto out_set_bit;
393	}
394
395	if (__predict_false(ksiginfo_zone == NULL)) {
396		SIGADDSET(sq->sq_kill, signo);
397		goto out_set_bit;
398	}
399
400	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
401		signal_overflow++;
402		ret = EAGAIN;
403	} else if ((ksi = ksiginfo_alloc(0)) == NULL) {
404		signal_alloc_fail++;
405		ret = EAGAIN;
406	} else {
407		if (p != NULL)
408			p->p_pendingcnt++;
409		ksiginfo_copy(si, ksi);
410		ksi->ksi_signo = signo;
411		if (si->ksi_flags & KSI_HEAD)
412			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
413		else
414			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
415		ksi->ksi_sigq = sq;
416	}
417
418	if (ret != 0) {
419		if ((si->ksi_flags & KSI_PTRACE) != 0) {
420			SIGADDSET(sq->sq_ptrace, signo);
421			ret = 0;
422			goto out_set_bit;
423		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
424		    (si->ksi_flags & KSI_SIGQ) == 0) {
425			SIGADDSET(sq->sq_kill, signo);
426			ret = 0;
427			goto out_set_bit;
428		}
429		return (ret);
430	}
431
432out_set_bit:
433	SIGADDSET(sq->sq_signals, signo);
434	return (ret);
435}
436
437void
438sigqueue_flush(sigqueue_t *sq)
439{
440	struct proc *p = sq->sq_proc;
441	ksiginfo_t *ksi;
442
443	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
444
445	if (p != NULL)
446		PROC_LOCK_ASSERT(p, MA_OWNED);
447
448	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
449		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
450		ksi->ksi_sigq = NULL;
451		if (ksiginfo_tryfree(ksi) && p != NULL)
452			p->p_pendingcnt--;
453	}
454
455	SIGEMPTYSET(sq->sq_signals);
456	SIGEMPTYSET(sq->sq_kill);
457	SIGEMPTYSET(sq->sq_ptrace);
458}
459
460static void
461sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
462{
463	sigset_t tmp;
464	struct proc *p1, *p2;
465	ksiginfo_t *ksi, *next;
466
467	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
468	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
469	p1 = src->sq_proc;
470	p2 = dst->sq_proc;
471	/* Move siginfo to target list */
472	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
473		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
474			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
475			if (p1 != NULL)
476				p1->p_pendingcnt--;
477			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
478			ksi->ksi_sigq = dst;
479			if (p2 != NULL)
480				p2->p_pendingcnt++;
481		}
482	}
483
484	/* Move pending bits to target list */
485	tmp = src->sq_kill;
486	SIGSETAND(tmp, *set);
487	SIGSETOR(dst->sq_kill, tmp);
488	SIGSETNAND(src->sq_kill, tmp);
489
490	tmp = src->sq_ptrace;
491	SIGSETAND(tmp, *set);
492	SIGSETOR(dst->sq_ptrace, tmp);
493	SIGSETNAND(src->sq_ptrace, tmp);
494
495	tmp = src->sq_signals;
496	SIGSETAND(tmp, *set);
497	SIGSETOR(dst->sq_signals, tmp);
498	SIGSETNAND(src->sq_signals, tmp);
499}
500
501#if 0
502static void
503sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
504{
505	sigset_t set;
506
507	SIGEMPTYSET(set);
508	SIGADDSET(set, signo);
509	sigqueue_move_set(src, dst, &set);
510}
511#endif
512
513static void
514sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
515{
516	struct proc *p = sq->sq_proc;
517	ksiginfo_t *ksi, *next;
518
519	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
520
521	/* Remove siginfo queue */
522	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
523		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
524			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
525			ksi->ksi_sigq = NULL;
526			if (ksiginfo_tryfree(ksi) && p != NULL)
527				p->p_pendingcnt--;
528		}
529	}
530	SIGSETNAND(sq->sq_kill, *set);
531	SIGSETNAND(sq->sq_ptrace, *set);
532	SIGSETNAND(sq->sq_signals, *set);
533}
534
535void
536sigqueue_delete(sigqueue_t *sq, int signo)
537{
538	sigset_t set;
539
540	SIGEMPTYSET(set);
541	SIGADDSET(set, signo);
542	sigqueue_delete_set(sq, &set);
543}
544
545/* Remove a set of signals for a process */
546static void
547sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
548{
549	sigqueue_t worklist;
550	struct thread *td0;
551
552	PROC_LOCK_ASSERT(p, MA_OWNED);
553
554	sigqueue_init(&worklist, NULL);
555	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
556
557	FOREACH_THREAD_IN_PROC(p, td0)
558		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
559
560	sigqueue_flush(&worklist);
561}
562
563void
564sigqueue_delete_proc(struct proc *p, int signo)
565{
566	sigset_t set;
567
568	SIGEMPTYSET(set);
569	SIGADDSET(set, signo);
570	sigqueue_delete_set_proc(p, &set);
571}
572
573static void
574sigqueue_delete_stopmask_proc(struct proc *p)
575{
576	sigset_t set;
577
578	SIGEMPTYSET(set);
579	SIGADDSET(set, SIGSTOP);
580	SIGADDSET(set, SIGTSTP);
581	SIGADDSET(set, SIGTTIN);
582	SIGADDSET(set, SIGTTOU);
583	sigqueue_delete_set_proc(p, &set);
584}
585
586/*
587 * Determine signal that should be delivered to thread td, the current
588 * thread, 0 if none.  If there is a pending stop signal with default
589 * action, the process stops in issignal().
590 */
591int
592cursig(struct thread *td)
593{
594	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
595	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
596	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
597	return (SIGPENDING(td) ? issignal(td) : 0);
598}
599
600/*
601 * Arrange for ast() to handle unmasked pending signals on return to user
602 * mode.  This must be called whenever a signal is added to td_sigqueue or
603 * unmasked in td_sigmask.
604 */
605void
606signotify(struct thread *td)
607{
608	struct proc *p;
609
610	p = td->td_proc;
611
612	PROC_LOCK_ASSERT(p, MA_OWNED);
613
614	if (SIGPENDING(td)) {
615		thread_lock(td);
616		td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING;
617		thread_unlock(td);
618	}
619}
620
621/*
622 * Returns 1 (true) if altstack is configured for the thread, and the
623 * passed stack bottom address falls into the altstack range.  Handles
624 * the 43 compat special case where the alt stack size is zero.
625 */
626int
627sigonstack(size_t sp)
628{
629	struct thread *td;
630
631	td = curthread;
632	if ((td->td_pflags & TDP_ALTSTACK) == 0)
633		return (0);
634#if defined(COMPAT_43)
635	if (td->td_sigstk.ss_size == 0)
636		return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
637#endif
638	return (sp >= (size_t)td->td_sigstk.ss_sp &&
639	    sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
640}
641
642static __inline int
643sigprop(int sig)
644{
645
646	if (sig > 0 && sig < NSIG)
647		return (sigproptbl[_SIG_IDX(sig)]);
648	return (0);
649}
650
651int
652sig_ffs(sigset_t *set)
653{
654	int i;
655
656	for (i = 0; i < _SIG_WORDS; i++)
657		if (set->__bits[i])
658			return (ffs(set->__bits[i]) + (i * 32));
659	return (0);
660}
661
662static bool
663sigact_flag_test(const struct sigaction *act, int flag)
664{
665
666	/*
667	 * SA_SIGINFO is reset when signal disposition is set to
668	 * ignore or default.  Other flags are kept according to user
669	 * settings.
670	 */
671	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
672	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
673	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
674}
675
676/*
677 * kern_sigaction
678 * sigaction
679 * freebsd4_sigaction
680 * osigaction
681 */
682int
683kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
684    struct sigaction *oact, int flags)
685{
686	struct sigacts *ps;
687	struct proc *p = td->td_proc;
688
689	if (!_SIG_VALID(sig))
690		return (EINVAL);
691	if (act != NULL && act->sa_handler != SIG_DFL &&
692	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
693	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
694	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
695		return (EINVAL);
696
697	PROC_LOCK(p);
698	ps = p->p_sigacts;
699	mtx_lock(&ps->ps_mtx);
700	if (oact) {
701		memset(oact, 0, sizeof(*oact));
702		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
703		if (SIGISMEMBER(ps->ps_sigonstack, sig))
704			oact->sa_flags |= SA_ONSTACK;
705		if (!SIGISMEMBER(ps->ps_sigintr, sig))
706			oact->sa_flags |= SA_RESTART;
707		if (SIGISMEMBER(ps->ps_sigreset, sig))
708			oact->sa_flags |= SA_RESETHAND;
709		if (SIGISMEMBER(ps->ps_signodefer, sig))
710			oact->sa_flags |= SA_NODEFER;
711		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
712			oact->sa_flags |= SA_SIGINFO;
713			oact->sa_sigaction =
714			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
715		} else
716			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
717		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
718			oact->sa_flags |= SA_NOCLDSTOP;
719		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
720			oact->sa_flags |= SA_NOCLDWAIT;
721	}
722	if (act) {
723		if ((sig == SIGKILL || sig == SIGSTOP) &&
724		    act->sa_handler != SIG_DFL) {
725			mtx_unlock(&ps->ps_mtx);
726			PROC_UNLOCK(p);
727			return (EINVAL);
728		}
729
730		/*
731		 * Change setting atomically.
732		 */
733
734		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
735		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
736		if (sigact_flag_test(act, SA_SIGINFO)) {
737			ps->ps_sigact[_SIG_IDX(sig)] =
738			    (__sighandler_t *)act->sa_sigaction;
739			SIGADDSET(ps->ps_siginfo, sig);
740		} else {
741			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
742			SIGDELSET(ps->ps_siginfo, sig);
743		}
744		if (!sigact_flag_test(act, SA_RESTART))
745			SIGADDSET(ps->ps_sigintr, sig);
746		else
747			SIGDELSET(ps->ps_sigintr, sig);
748		if (sigact_flag_test(act, SA_ONSTACK))
749			SIGADDSET(ps->ps_sigonstack, sig);
750		else
751			SIGDELSET(ps->ps_sigonstack, sig);
752		if (sigact_flag_test(act, SA_RESETHAND))
753			SIGADDSET(ps->ps_sigreset, sig);
754		else
755			SIGDELSET(ps->ps_sigreset, sig);
756		if (sigact_flag_test(act, SA_NODEFER))
757			SIGADDSET(ps->ps_signodefer, sig);
758		else
759			SIGDELSET(ps->ps_signodefer, sig);
760		if (sig == SIGCHLD) {
761			if (act->sa_flags & SA_NOCLDSTOP)
762				ps->ps_flag |= PS_NOCLDSTOP;
763			else
764				ps->ps_flag &= ~PS_NOCLDSTOP;
765			if (act->sa_flags & SA_NOCLDWAIT) {
766				/*
767				 * Paranoia: since SA_NOCLDWAIT is implemented
768				 * by reparenting the dying child to PID 1 (and
769				 * trust it to reap the zombie), PID 1 itself
770				 * is forbidden to set SA_NOCLDWAIT.
771				 */
772				if (p->p_pid == 1)
773					ps->ps_flag &= ~PS_NOCLDWAIT;
774				else
775					ps->ps_flag |= PS_NOCLDWAIT;
776			} else
777				ps->ps_flag &= ~PS_NOCLDWAIT;
778			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
779				ps->ps_flag |= PS_CLDSIGIGN;
780			else
781				ps->ps_flag &= ~PS_CLDSIGIGN;
782		}
783		/*
784		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
785		 * and for signals set to SIG_DFL where the default is to
786		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
787		 * have to restart the process.
788		 */
789		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
790		    (sigprop(sig) & SA_IGNORE &&
791		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
792			/* never to be seen again */
793			sigqueue_delete_proc(p, sig);
794			if (sig != SIGCONT)
795				/* easier in psignal */
796				SIGADDSET(ps->ps_sigignore, sig);
797			SIGDELSET(ps->ps_sigcatch, sig);
798		} else {
799			SIGDELSET(ps->ps_sigignore, sig);
800			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
801				SIGDELSET(ps->ps_sigcatch, sig);
802			else
803				SIGADDSET(ps->ps_sigcatch, sig);
804		}
805#ifdef COMPAT_FREEBSD4
806		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
807		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
808		    (flags & KSA_FREEBSD4) == 0)
809			SIGDELSET(ps->ps_freebsd4, sig);
810		else
811			SIGADDSET(ps->ps_freebsd4, sig);
812#endif
813#ifdef COMPAT_43
814		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
815		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
816		    (flags & KSA_OSIGSET) == 0)
817			SIGDELSET(ps->ps_osigset, sig);
818		else
819			SIGADDSET(ps->ps_osigset, sig);
820#endif
821	}
822	mtx_unlock(&ps->ps_mtx);
823	PROC_UNLOCK(p);
824	return (0);
825}
826
827#ifndef _SYS_SYSPROTO_H_
828struct sigaction_args {
829	int	sig;
830	struct	sigaction *act;
831	struct	sigaction *oact;
832};
833#endif
834int
835sys_sigaction(struct thread *td, struct sigaction_args *uap)
836{
837	struct sigaction act, oact;
838	struct sigaction *actp, *oactp;
839	int error;
840
841	actp = (uap->act != NULL) ? &act : NULL;
842	oactp = (uap->oact != NULL) ? &oact : NULL;
843	if (actp) {
844		error = copyin(uap->act, actp, sizeof(act));
845		if (error)
846			return (error);
847	}
848	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
849	if (oactp && !error)
850		error = copyout(oactp, uap->oact, sizeof(oact));
851	return (error);
852}
853
854#ifdef COMPAT_FREEBSD4
855#ifndef _SYS_SYSPROTO_H_
856struct freebsd4_sigaction_args {
857	int	sig;
858	struct	sigaction *act;
859	struct	sigaction *oact;
860};
861#endif
862int
863freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
864{
865	struct sigaction act, oact;
866	struct sigaction *actp, *oactp;
867	int error;
868
869
870	actp = (uap->act != NULL) ? &act : NULL;
871	oactp = (uap->oact != NULL) ? &oact : NULL;
872	if (actp) {
873		error = copyin(uap->act, actp, sizeof(act));
874		if (error)
875			return (error);
876	}
877	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
878	if (oactp && !error)
879		error = copyout(oactp, uap->oact, sizeof(oact));
880	return (error);
881}
882#endif	/* COMAPT_FREEBSD4 */
883
884#ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
885#ifndef _SYS_SYSPROTO_H_
886struct osigaction_args {
887	int	signum;
888	struct	osigaction *nsa;
889	struct	osigaction *osa;
890};
891#endif
892int
893osigaction(struct thread *td, struct osigaction_args *uap)
894{
895	struct osigaction sa;
896	struct sigaction nsa, osa;
897	struct sigaction *nsap, *osap;
898	int error;
899
900	if (uap->signum <= 0 || uap->signum >= ONSIG)
901		return (EINVAL);
902
903	nsap = (uap->nsa != NULL) ? &nsa : NULL;
904	osap = (uap->osa != NULL) ? &osa : NULL;
905
906	if (nsap) {
907		error = copyin(uap->nsa, &sa, sizeof(sa));
908		if (error)
909			return (error);
910		nsap->sa_handler = sa.sa_handler;
911		nsap->sa_flags = sa.sa_flags;
912		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
913	}
914	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
915	if (osap && !error) {
916		sa.sa_handler = osap->sa_handler;
917		sa.sa_flags = osap->sa_flags;
918		SIG2OSIG(osap->sa_mask, sa.sa_mask);
919		error = copyout(&sa, uap->osa, sizeof(sa));
920	}
921	return (error);
922}
923
924#if !defined(__i386__)
925/* Avoid replicating the same stub everywhere */
926int
927osigreturn(struct thread *td, struct osigreturn_args *uap)
928{
929
930	return (nosys(td, (struct nosys_args *)uap));
931}
932#endif
933#endif /* COMPAT_43 */
934
935/*
936 * Initialize signal state for process 0;
937 * set to ignore signals that are ignored by default.
938 */
939void
940siginit(struct proc *p)
941{
942	int i;
943	struct sigacts *ps;
944
945	PROC_LOCK(p);
946	ps = p->p_sigacts;
947	mtx_lock(&ps->ps_mtx);
948	for (i = 1; i <= NSIG; i++) {
949		if (sigprop(i) & SA_IGNORE && i != SIGCONT) {
950			SIGADDSET(ps->ps_sigignore, i);
951		}
952	}
953	mtx_unlock(&ps->ps_mtx);
954	PROC_UNLOCK(p);
955}
956
957/*
958 * Reset specified signal to the default disposition.
959 */
960static void
961sigdflt(struct sigacts *ps, int sig)
962{
963
964	mtx_assert(&ps->ps_mtx, MA_OWNED);
965	SIGDELSET(ps->ps_sigcatch, sig);
966	if ((sigprop(sig) & SA_IGNORE) != 0 && sig != SIGCONT)
967		SIGADDSET(ps->ps_sigignore, sig);
968	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
969	SIGDELSET(ps->ps_siginfo, sig);
970}
971
972/*
973 * Reset signals for an exec of the specified process.
974 */
975void
976execsigs(struct proc *p)
977{
978	sigset_t osigignore;
979	struct sigacts *ps;
980	int sig;
981	struct thread *td;
982
983	/*
984	 * Reset caught signals.  Held signals remain held
985	 * through td_sigmask (unless they were caught,
986	 * and are now ignored by default).
987	 */
988	PROC_LOCK_ASSERT(p, MA_OWNED);
989	ps = p->p_sigacts;
990	mtx_lock(&ps->ps_mtx);
991	sig_drop_caught(p);
992
993	/*
994	 * As CloudABI processes cannot modify signal handlers, fully
995	 * reset all signals to their default behavior. Do ignore
996	 * SIGPIPE, as it would otherwise be impossible to recover from
997	 * writes to broken pipes and sockets.
998	 */
999	if (SV_PROC_ABI(p) == SV_ABI_CLOUDABI) {
1000		osigignore = ps->ps_sigignore;
1001		while (SIGNOTEMPTY(osigignore)) {
1002			sig = sig_ffs(&osigignore);
1003			SIGDELSET(osigignore, sig);
1004			if (sig != SIGPIPE)
1005				sigdflt(ps, sig);
1006		}
1007		SIGADDSET(ps->ps_sigignore, SIGPIPE);
1008	}
1009
1010	/*
1011	 * Reset stack state to the user stack.
1012	 * Clear set of signals caught on the signal stack.
1013	 */
1014	td = curthread;
1015	MPASS(td->td_proc == p);
1016	td->td_sigstk.ss_flags = SS_DISABLE;
1017	td->td_sigstk.ss_size = 0;
1018	td->td_sigstk.ss_sp = 0;
1019	td->td_pflags &= ~TDP_ALTSTACK;
1020	/*
1021	 * Reset no zombies if child dies flag as Solaris does.
1022	 */
1023	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1024	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1025		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1026	mtx_unlock(&ps->ps_mtx);
1027}
1028
1029/*
1030 * kern_sigprocmask()
1031 *
1032 *	Manipulate signal mask.
1033 */
1034int
1035kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1036    int flags)
1037{
1038	sigset_t new_block, oset1;
1039	struct proc *p;
1040	int error;
1041
1042	p = td->td_proc;
1043	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1044		PROC_LOCK_ASSERT(p, MA_OWNED);
1045	else
1046		PROC_LOCK(p);
1047	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1048	    ? MA_OWNED : MA_NOTOWNED);
1049	if (oset != NULL)
1050		*oset = td->td_sigmask;
1051
1052	error = 0;
1053	if (set != NULL) {
1054		switch (how) {
1055		case SIG_BLOCK:
1056			SIG_CANTMASK(*set);
1057			oset1 = td->td_sigmask;
1058			SIGSETOR(td->td_sigmask, *set);
1059			new_block = td->td_sigmask;
1060			SIGSETNAND(new_block, oset1);
1061			break;
1062		case SIG_UNBLOCK:
1063			SIGSETNAND(td->td_sigmask, *set);
1064			signotify(td);
1065			goto out;
1066		case SIG_SETMASK:
1067			SIG_CANTMASK(*set);
1068			oset1 = td->td_sigmask;
1069			if (flags & SIGPROCMASK_OLD)
1070				SIGSETLO(td->td_sigmask, *set);
1071			else
1072				td->td_sigmask = *set;
1073			new_block = td->td_sigmask;
1074			SIGSETNAND(new_block, oset1);
1075			signotify(td);
1076			break;
1077		default:
1078			error = EINVAL;
1079			goto out;
1080		}
1081
1082		/*
1083		 * The new_block set contains signals that were not previously
1084		 * blocked, but are blocked now.
1085		 *
1086		 * In case we block any signal that was not previously blocked
1087		 * for td, and process has the signal pending, try to schedule
1088		 * signal delivery to some thread that does not block the
1089		 * signal, possibly waking it up.
1090		 */
1091		if (p->p_numthreads != 1)
1092			reschedule_signals(p, new_block, flags);
1093	}
1094
1095out:
1096	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1097		PROC_UNLOCK(p);
1098	return (error);
1099}
1100
1101#ifndef _SYS_SYSPROTO_H_
1102struct sigprocmask_args {
1103	int	how;
1104	const sigset_t *set;
1105	sigset_t *oset;
1106};
1107#endif
1108int
1109sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1110{
1111	sigset_t set, oset;
1112	sigset_t *setp, *osetp;
1113	int error;
1114
1115	setp = (uap->set != NULL) ? &set : NULL;
1116	osetp = (uap->oset != NULL) ? &oset : NULL;
1117	if (setp) {
1118		error = copyin(uap->set, setp, sizeof(set));
1119		if (error)
1120			return (error);
1121	}
1122	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1123	if (osetp && !error) {
1124		error = copyout(osetp, uap->oset, sizeof(oset));
1125	}
1126	return (error);
1127}
1128
1129#ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1130#ifndef _SYS_SYSPROTO_H_
1131struct osigprocmask_args {
1132	int	how;
1133	osigset_t mask;
1134};
1135#endif
1136int
1137osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1138{
1139	sigset_t set, oset;
1140	int error;
1141
1142	OSIG2SIG(uap->mask, set);
1143	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1144	SIG2OSIG(oset, td->td_retval[0]);
1145	return (error);
1146}
1147#endif /* COMPAT_43 */
1148
1149int
1150sys_sigwait(struct thread *td, struct sigwait_args *uap)
1151{
1152	ksiginfo_t ksi;
1153	sigset_t set;
1154	int error;
1155
1156	error = copyin(uap->set, &set, sizeof(set));
1157	if (error) {
1158		td->td_retval[0] = error;
1159		return (0);
1160	}
1161
1162	error = kern_sigtimedwait(td, set, &ksi, NULL);
1163	if (error) {
1164		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1165			error = ERESTART;
1166		if (error == ERESTART)
1167			return (error);
1168		td->td_retval[0] = error;
1169		return (0);
1170	}
1171
1172	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1173	td->td_retval[0] = error;
1174	return (0);
1175}
1176
1177int
1178sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1179{
1180	struct timespec ts;
1181	struct timespec *timeout;
1182	sigset_t set;
1183	ksiginfo_t ksi;
1184	int error;
1185
1186	if (uap->timeout) {
1187		error = copyin(uap->timeout, &ts, sizeof(ts));
1188		if (error)
1189			return (error);
1190
1191		timeout = &ts;
1192	} else
1193		timeout = NULL;
1194
1195	error = copyin(uap->set, &set, sizeof(set));
1196	if (error)
1197		return (error);
1198
1199	error = kern_sigtimedwait(td, set, &ksi, timeout);
1200	if (error)
1201		return (error);
1202
1203	if (uap->info)
1204		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1205
1206	if (error == 0)
1207		td->td_retval[0] = ksi.ksi_signo;
1208	return (error);
1209}
1210
1211int
1212sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1213{
1214	ksiginfo_t ksi;
1215	sigset_t set;
1216	int error;
1217
1218	error = copyin(uap->set, &set, sizeof(set));
1219	if (error)
1220		return (error);
1221
1222	error = kern_sigtimedwait(td, set, &ksi, NULL);
1223	if (error)
1224		return (error);
1225
1226	if (uap->info)
1227		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1228
1229	if (error == 0)
1230		td->td_retval[0] = ksi.ksi_signo;
1231	return (error);
1232}
1233
1234static void
1235proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1236{
1237	struct thread *thr;
1238
1239	FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1240		if (thr == td)
1241			thr->td_si = *si;
1242		else
1243			thr->td_si.si_signo = 0;
1244	}
1245}
1246
1247int
1248kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1249	struct timespec *timeout)
1250{
1251	struct sigacts *ps;
1252	sigset_t saved_mask, new_block;
1253	struct proc *p;
1254	int error, sig, timo, timevalid = 0;
1255	struct timespec rts, ets, ts;
1256	struct timeval tv;
1257	bool traced;
1258
1259	p = td->td_proc;
1260	error = 0;
1261	ets.tv_sec = 0;
1262	ets.tv_nsec = 0;
1263	traced = false;
1264
1265	if (timeout != NULL) {
1266		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1267			timevalid = 1;
1268			getnanouptime(&rts);
1269			ets = rts;
1270			timespecadd(&ets, timeout);
1271		}
1272	}
1273	ksiginfo_init(ksi);
1274	/* Some signals can not be waited for. */
1275	SIG_CANTMASK(waitset);
1276	ps = p->p_sigacts;
1277	PROC_LOCK(p);
1278	saved_mask = td->td_sigmask;
1279	SIGSETNAND(td->td_sigmask, waitset);
1280	for (;;) {
1281		mtx_lock(&ps->ps_mtx);
1282		sig = cursig(td);
1283		mtx_unlock(&ps->ps_mtx);
1284		KASSERT(sig >= 0, ("sig %d", sig));
1285		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1286			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1287			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1288				error = 0;
1289				break;
1290			}
1291		}
1292
1293		if (error != 0)
1294			break;
1295
1296		/*
1297		 * POSIX says this must be checked after looking for pending
1298		 * signals.
1299		 */
1300		if (timeout != NULL) {
1301			if (!timevalid) {
1302				error = EINVAL;
1303				break;
1304			}
1305			getnanouptime(&rts);
1306			if (timespeccmp(&rts, &ets, >=)) {
1307				error = EAGAIN;
1308				break;
1309			}
1310			ts = ets;
1311			timespecsub(&ts, &rts);
1312			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1313			timo = tvtohz(&tv);
1314		} else {
1315			timo = 0;
1316		}
1317
1318		if (traced) {
1319			error = EINTR;
1320			break;
1321		}
1322
1323		error = msleep(ps, &p->p_mtx, PPAUSE|PCATCH, "sigwait", timo);
1324
1325		if (timeout != NULL) {
1326			if (error == ERESTART) {
1327				/* Timeout can not be restarted. */
1328				error = EINTR;
1329			} else if (error == EAGAIN) {
1330				/* We will calculate timeout by ourself. */
1331				error = 0;
1332			}
1333		}
1334
1335		/*
1336		 * If PTRACE_SCE or PTRACE_SCX were set after
1337		 * userspace entered the syscall, return spurious
1338		 * EINTR after wait was done.  Only do this as last
1339		 * resort after rechecking for possible queued signals
1340		 * and expired timeouts.
1341		 */
1342		if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1343			traced = true;
1344	}
1345
1346	new_block = saved_mask;
1347	SIGSETNAND(new_block, td->td_sigmask);
1348	td->td_sigmask = saved_mask;
1349	/*
1350	 * Fewer signals can be delivered to us, reschedule signal
1351	 * notification.
1352	 */
1353	if (p->p_numthreads != 1)
1354		reschedule_signals(p, new_block, 0);
1355
1356	if (error == 0) {
1357		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1358
1359		if (ksi->ksi_code == SI_TIMER)
1360			itimer_accept(p, ksi->ksi_timerid, ksi);
1361
1362#ifdef KTRACE
1363		if (KTRPOINT(td, KTR_PSIG)) {
1364			sig_t action;
1365
1366			mtx_lock(&ps->ps_mtx);
1367			action = ps->ps_sigact[_SIG_IDX(sig)];
1368			mtx_unlock(&ps->ps_mtx);
1369			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1370		}
1371#endif
1372		if (sig == SIGKILL) {
1373			proc_td_siginfo_capture(td, &ksi->ksi_info);
1374			sigexit(td, sig);
1375		}
1376	}
1377	PROC_UNLOCK(p);
1378	return (error);
1379}
1380
1381#ifndef _SYS_SYSPROTO_H_
1382struct sigpending_args {
1383	sigset_t	*set;
1384};
1385#endif
1386int
1387sys_sigpending(struct thread *td, struct sigpending_args *uap)
1388{
1389	struct proc *p = td->td_proc;
1390	sigset_t pending;
1391
1392	PROC_LOCK(p);
1393	pending = p->p_sigqueue.sq_signals;
1394	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1395	PROC_UNLOCK(p);
1396	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1397}
1398
1399#ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1400#ifndef _SYS_SYSPROTO_H_
1401struct osigpending_args {
1402	int	dummy;
1403};
1404#endif
1405int
1406osigpending(struct thread *td, struct osigpending_args *uap)
1407{
1408	struct proc *p = td->td_proc;
1409	sigset_t pending;
1410
1411	PROC_LOCK(p);
1412	pending = p->p_sigqueue.sq_signals;
1413	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1414	PROC_UNLOCK(p);
1415	SIG2OSIG(pending, td->td_retval[0]);
1416	return (0);
1417}
1418#endif /* COMPAT_43 */
1419
1420#if defined(COMPAT_43)
1421/*
1422 * Generalized interface signal handler, 4.3-compatible.
1423 */
1424#ifndef _SYS_SYSPROTO_H_
1425struct osigvec_args {
1426	int	signum;
1427	struct	sigvec *nsv;
1428	struct	sigvec *osv;
1429};
1430#endif
1431/* ARGSUSED */
1432int
1433osigvec(struct thread *td, struct osigvec_args *uap)
1434{
1435	struct sigvec vec;
1436	struct sigaction nsa, osa;
1437	struct sigaction *nsap, *osap;
1438	int error;
1439
1440	if (uap->signum <= 0 || uap->signum >= ONSIG)
1441		return (EINVAL);
1442	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1443	osap = (uap->osv != NULL) ? &osa : NULL;
1444	if (nsap) {
1445		error = copyin(uap->nsv, &vec, sizeof(vec));
1446		if (error)
1447			return (error);
1448		nsap->sa_handler = vec.sv_handler;
1449		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1450		nsap->sa_flags = vec.sv_flags;
1451		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1452	}
1453	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1454	if (osap && !error) {
1455		vec.sv_handler = osap->sa_handler;
1456		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1457		vec.sv_flags = osap->sa_flags;
1458		vec.sv_flags &= ~SA_NOCLDWAIT;
1459		vec.sv_flags ^= SA_RESTART;
1460		error = copyout(&vec, uap->osv, sizeof(vec));
1461	}
1462	return (error);
1463}
1464
1465#ifndef _SYS_SYSPROTO_H_
1466struct osigblock_args {
1467	int	mask;
1468};
1469#endif
1470int
1471osigblock(struct thread *td, struct osigblock_args *uap)
1472{
1473	sigset_t set, oset;
1474
1475	OSIG2SIG(uap->mask, set);
1476	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1477	SIG2OSIG(oset, td->td_retval[0]);
1478	return (0);
1479}
1480
1481#ifndef _SYS_SYSPROTO_H_
1482struct osigsetmask_args {
1483	int	mask;
1484};
1485#endif
1486int
1487osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1488{
1489	sigset_t set, oset;
1490
1491	OSIG2SIG(uap->mask, set);
1492	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1493	SIG2OSIG(oset, td->td_retval[0]);
1494	return (0);
1495}
1496#endif /* COMPAT_43 */
1497
1498/*
1499 * Suspend calling thread until signal, providing mask to be set in the
1500 * meantime.
1501 */
1502#ifndef _SYS_SYSPROTO_H_
1503struct sigsuspend_args {
1504	const sigset_t *sigmask;
1505};
1506#endif
1507/* ARGSUSED */
1508int
1509sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1510{
1511	sigset_t mask;
1512	int error;
1513
1514	error = copyin(uap->sigmask, &mask, sizeof(mask));
1515	if (error)
1516		return (error);
1517	return (kern_sigsuspend(td, mask));
1518}
1519
1520int
1521kern_sigsuspend(struct thread *td, sigset_t mask)
1522{
1523	struct proc *p = td->td_proc;
1524	int has_sig, sig;
1525
1526	/*
1527	 * When returning from sigsuspend, we want
1528	 * the old mask to be restored after the
1529	 * signal handler has finished.  Thus, we
1530	 * save it here and mark the sigacts structure
1531	 * to indicate this.
1532	 */
1533	PROC_LOCK(p);
1534	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1535	    SIGPROCMASK_PROC_LOCKED);
1536	td->td_pflags |= TDP_OLDMASK;
1537
1538	/*
1539	 * Process signals now. Otherwise, we can get spurious wakeup
1540	 * due to signal entered process queue, but delivered to other
1541	 * thread. But sigsuspend should return only on signal
1542	 * delivery.
1543	 */
1544	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1545	for (has_sig = 0; !has_sig;) {
1546		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1547			0) == 0)
1548			/* void */;
1549		thread_suspend_check(0);
1550		mtx_lock(&p->p_sigacts->ps_mtx);
1551		while ((sig = cursig(td)) != 0) {
1552			KASSERT(sig >= 0, ("sig %d", sig));
1553			has_sig += postsig(sig);
1554		}
1555		mtx_unlock(&p->p_sigacts->ps_mtx);
1556
1557		/*
1558		 * If PTRACE_SCE or PTRACE_SCX were set after
1559		 * userspace entered the syscall, return spurious
1560		 * EINTR.
1561		 */
1562		if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1563			has_sig += 1;
1564	}
1565	PROC_UNLOCK(p);
1566	td->td_errno = EINTR;
1567	td->td_pflags |= TDP_NERRNO;
1568	return (EJUSTRETURN);
1569}
1570
1571#ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1572/*
1573 * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1574 * convention: libc stub passes mask, not pointer, to save a copyin.
1575 */
1576#ifndef _SYS_SYSPROTO_H_
1577struct osigsuspend_args {
1578	osigset_t mask;
1579};
1580#endif
1581/* ARGSUSED */
1582int
1583osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1584{
1585	sigset_t mask;
1586
1587	OSIG2SIG(uap->mask, mask);
1588	return (kern_sigsuspend(td, mask));
1589}
1590#endif /* COMPAT_43 */
1591
1592#if defined(COMPAT_43)
1593#ifndef _SYS_SYSPROTO_H_
1594struct osigstack_args {
1595	struct	sigstack *nss;
1596	struct	sigstack *oss;
1597};
1598#endif
1599/* ARGSUSED */
1600int
1601osigstack(struct thread *td, struct osigstack_args *uap)
1602{
1603	struct sigstack nss, oss;
1604	int error = 0;
1605
1606	if (uap->nss != NULL) {
1607		error = copyin(uap->nss, &nss, sizeof(nss));
1608		if (error)
1609			return (error);
1610	}
1611	oss.ss_sp = td->td_sigstk.ss_sp;
1612	oss.ss_onstack = sigonstack(cpu_getstack(td));
1613	if (uap->nss != NULL) {
1614		td->td_sigstk.ss_sp = nss.ss_sp;
1615		td->td_sigstk.ss_size = 0;
1616		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1617		td->td_pflags |= TDP_ALTSTACK;
1618	}
1619	if (uap->oss != NULL)
1620		error = copyout(&oss, uap->oss, sizeof(oss));
1621
1622	return (error);
1623}
1624#endif /* COMPAT_43 */
1625
1626#ifndef _SYS_SYSPROTO_H_
1627struct sigaltstack_args {
1628	stack_t	*ss;
1629	stack_t	*oss;
1630};
1631#endif
1632/* ARGSUSED */
1633int
1634sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1635{
1636	stack_t ss, oss;
1637	int error;
1638
1639	if (uap->ss != NULL) {
1640		error = copyin(uap->ss, &ss, sizeof(ss));
1641		if (error)
1642			return (error);
1643	}
1644	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1645	    (uap->oss != NULL) ? &oss : NULL);
1646	if (error)
1647		return (error);
1648	if (uap->oss != NULL)
1649		error = copyout(&oss, uap->oss, sizeof(stack_t));
1650	return (error);
1651}
1652
1653int
1654kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1655{
1656	struct proc *p = td->td_proc;
1657	int oonstack;
1658
1659	oonstack = sigonstack(cpu_getstack(td));
1660
1661	if (oss != NULL) {
1662		*oss = td->td_sigstk;
1663		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1664		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1665	}
1666
1667	if (ss != NULL) {
1668		if (oonstack)
1669			return (EPERM);
1670		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1671			return (EINVAL);
1672		if (!(ss->ss_flags & SS_DISABLE)) {
1673			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1674				return (ENOMEM);
1675
1676			td->td_sigstk = *ss;
1677			td->td_pflags |= TDP_ALTSTACK;
1678		} else {
1679			td->td_pflags &= ~TDP_ALTSTACK;
1680		}
1681	}
1682	return (0);
1683}
1684
1685/*
1686 * Common code for kill process group/broadcast kill.
1687 * cp is calling process.
1688 */
1689static int
1690killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1691{
1692	struct proc *p;
1693	struct pgrp *pgrp;
1694	int err;
1695	int ret;
1696
1697	ret = ESRCH;
1698	if (all) {
1699		/*
1700		 * broadcast
1701		 */
1702		sx_slock(&allproc_lock);
1703		FOREACH_PROC_IN_SYSTEM(p) {
1704			PROC_LOCK(p);
1705			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1706			    p == td->td_proc || p->p_state == PRS_NEW) {
1707				PROC_UNLOCK(p);
1708				continue;
1709			}
1710			err = p_cansignal(td, p, sig);
1711			if (err == 0) {
1712				if (sig)
1713					pksignal(p, sig, ksi);
1714				ret = err;
1715			}
1716			else if (ret == ESRCH)
1717				ret = err;
1718			PROC_UNLOCK(p);
1719		}
1720		sx_sunlock(&allproc_lock);
1721	} else {
1722		sx_slock(&proctree_lock);
1723		if (pgid == 0) {
1724			/*
1725			 * zero pgid means send to my process group.
1726			 */
1727			pgrp = td->td_proc->p_pgrp;
1728			PGRP_LOCK(pgrp);
1729		} else {
1730			pgrp = pgfind(pgid);
1731			if (pgrp == NULL) {
1732				sx_sunlock(&proctree_lock);
1733				return (ESRCH);
1734			}
1735		}
1736		sx_sunlock(&proctree_lock);
1737		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1738			PROC_LOCK(p);
1739			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1740			    p->p_state == PRS_NEW) {
1741				PROC_UNLOCK(p);
1742				continue;
1743			}
1744			err = p_cansignal(td, p, sig);
1745			if (err == 0) {
1746				if (sig)
1747					pksignal(p, sig, ksi);
1748				ret = err;
1749			}
1750			else if (ret == ESRCH)
1751				ret = err;
1752			PROC_UNLOCK(p);
1753		}
1754		PGRP_UNLOCK(pgrp);
1755	}
1756	return (ret);
1757}
1758
1759#ifndef _SYS_SYSPROTO_H_
1760struct kill_args {
1761	int	pid;
1762	int	signum;
1763};
1764#endif
1765/* ARGSUSED */
1766int
1767sys_kill(struct thread *td, struct kill_args *uap)
1768{
1769	ksiginfo_t ksi;
1770	struct proc *p;
1771	int error;
1772
1773	/*
1774	 * A process in capability mode can send signals only to himself.
1775	 * The main rationale behind this is that abort(3) is implemented as
1776	 * kill(getpid(), SIGABRT).
1777	 */
1778	if (IN_CAPABILITY_MODE(td) && uap->pid != td->td_proc->p_pid)
1779		return (ECAPMODE);
1780
1781	AUDIT_ARG_SIGNUM(uap->signum);
1782	AUDIT_ARG_PID(uap->pid);
1783	if ((u_int)uap->signum > _SIG_MAXSIG)
1784		return (EINVAL);
1785
1786	ksiginfo_init(&ksi);
1787	ksi.ksi_signo = uap->signum;
1788	ksi.ksi_code = SI_USER;
1789	ksi.ksi_pid = td->td_proc->p_pid;
1790	ksi.ksi_uid = td->td_ucred->cr_ruid;
1791
1792	if (uap->pid > 0) {
1793		/* kill single process */
1794		if ((p = pfind(uap->pid)) == NULL) {
1795			if ((p = zpfind(uap->pid)) == NULL)
1796				return (ESRCH);
1797		}
1798		AUDIT_ARG_PROCESS(p);
1799		error = p_cansignal(td, p, uap->signum);
1800		if (error == 0 && uap->signum)
1801			pksignal(p, uap->signum, &ksi);
1802		PROC_UNLOCK(p);
1803		return (error);
1804	}
1805	switch (uap->pid) {
1806	case -1:		/* broadcast signal */
1807		return (killpg1(td, uap->signum, 0, 1, &ksi));
1808	case 0:			/* signal own process group */
1809		return (killpg1(td, uap->signum, 0, 0, &ksi));
1810	default:		/* negative explicit process group */
1811		return (killpg1(td, uap->signum, -uap->pid, 0, &ksi));
1812	}
1813	/* NOTREACHED */
1814}
1815
1816int
1817sys_pdkill(struct thread *td, struct pdkill_args *uap)
1818{
1819	struct proc *p;
1820	cap_rights_t rights;
1821	int error;
1822
1823	AUDIT_ARG_SIGNUM(uap->signum);
1824	AUDIT_ARG_FD(uap->fd);
1825	if ((u_int)uap->signum > _SIG_MAXSIG)
1826		return (EINVAL);
1827
1828	error = procdesc_find(td, uap->fd,
1829	    cap_rights_init(&rights, CAP_PDKILL), &p);
1830	if (error)
1831		return (error);
1832	AUDIT_ARG_PROCESS(p);
1833	error = p_cansignal(td, p, uap->signum);
1834	if (error == 0 && uap->signum)
1835		kern_psignal(p, uap->signum);
1836	PROC_UNLOCK(p);
1837	return (error);
1838}
1839
1840#if defined(COMPAT_43)
1841#ifndef _SYS_SYSPROTO_H_
1842struct okillpg_args {
1843	int	pgid;
1844	int	signum;
1845};
1846#endif
1847/* ARGSUSED */
1848int
1849okillpg(struct thread *td, struct okillpg_args *uap)
1850{
1851	ksiginfo_t ksi;
1852
1853	AUDIT_ARG_SIGNUM(uap->signum);
1854	AUDIT_ARG_PID(uap->pgid);
1855	if ((u_int)uap->signum > _SIG_MAXSIG)
1856		return (EINVAL);
1857
1858	ksiginfo_init(&ksi);
1859	ksi.ksi_signo = uap->signum;
1860	ksi.ksi_code = SI_USER;
1861	ksi.ksi_pid = td->td_proc->p_pid;
1862	ksi.ksi_uid = td->td_ucred->cr_ruid;
1863	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1864}
1865#endif /* COMPAT_43 */
1866
1867#ifndef _SYS_SYSPROTO_H_
1868struct sigqueue_args {
1869	pid_t pid;
1870	int signum;
1871	/* union sigval */ void *value;
1872};
1873#endif
1874int
1875sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1876{
1877	union sigval sv;
1878
1879	sv.sival_ptr = uap->value;
1880
1881	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
1882}
1883
1884int
1885kern_sigqueue(struct thread *td, pid_t pid, int signum, union sigval *value)
1886{
1887	ksiginfo_t ksi;
1888	struct proc *p;
1889	int error;
1890
1891	if ((u_int)signum > _SIG_MAXSIG)
1892		return (EINVAL);
1893
1894	/*
1895	 * Specification says sigqueue can only send signal to
1896	 * single process.
1897	 */
1898	if (pid <= 0)
1899		return (EINVAL);
1900
1901	if ((p = pfind(pid)) == NULL) {
1902		if ((p = zpfind(pid)) == NULL)
1903			return (ESRCH);
1904	}
1905	error = p_cansignal(td, p, signum);
1906	if (error == 0 && signum != 0) {
1907		ksiginfo_init(&ksi);
1908		ksi.ksi_flags = KSI_SIGQ;
1909		ksi.ksi_signo = signum;
1910		ksi.ksi_code = SI_QUEUE;
1911		ksi.ksi_pid = td->td_proc->p_pid;
1912		ksi.ksi_uid = td->td_ucred->cr_ruid;
1913		ksi.ksi_value = *value;
1914		error = pksignal(p, ksi.ksi_signo, &ksi);
1915	}
1916	PROC_UNLOCK(p);
1917	return (error);
1918}
1919
1920/*
1921 * Send a signal to a process group.
1922 */
1923void
1924gsignal(int pgid, int sig, ksiginfo_t *ksi)
1925{
1926	struct pgrp *pgrp;
1927
1928	if (pgid != 0) {
1929		sx_slock(&proctree_lock);
1930		pgrp = pgfind(pgid);
1931		sx_sunlock(&proctree_lock);
1932		if (pgrp != NULL) {
1933			pgsignal(pgrp, sig, 0, ksi);
1934			PGRP_UNLOCK(pgrp);
1935		}
1936	}
1937}
1938
1939/*
1940 * Send a signal to a process group.  If checktty is 1,
1941 * limit to members which have a controlling terminal.
1942 */
1943void
1944pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
1945{
1946	struct proc *p;
1947
1948	if (pgrp) {
1949		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
1950		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1951			PROC_LOCK(p);
1952			if (p->p_state == PRS_NORMAL &&
1953			    (checkctty == 0 || p->p_flag & P_CONTROLT))
1954				pksignal(p, sig, ksi);
1955			PROC_UNLOCK(p);
1956		}
1957	}
1958}
1959
1960
1961/*
1962 * Recalculate the signal mask and reset the signal disposition after
1963 * usermode frame for delivery is formed.  Should be called after
1964 * mach-specific routine, because sysent->sv_sendsig() needs correct
1965 * ps_siginfo and signal mask.
1966 */
1967static void
1968postsig_done(int sig, struct thread *td, struct sigacts *ps)
1969{
1970	sigset_t mask;
1971
1972	mtx_assert(&ps->ps_mtx, MA_OWNED);
1973	td->td_ru.ru_nsignals++;
1974	mask = ps->ps_catchmask[_SIG_IDX(sig)];
1975	if (!SIGISMEMBER(ps->ps_signodefer, sig))
1976		SIGADDSET(mask, sig);
1977	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
1978	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
1979	if (SIGISMEMBER(ps->ps_sigreset, sig))
1980		sigdflt(ps, sig);
1981}
1982
1983
1984/*
1985 * Send a signal caused by a trap to the current thread.  If it will be
1986 * caught immediately, deliver it with correct code.  Otherwise, post it
1987 * normally.
1988 */
1989void
1990trapsignal(struct thread *td, ksiginfo_t *ksi)
1991{
1992	struct sigacts *ps;
1993	struct proc *p;
1994	int sig;
1995	int code;
1996
1997	p = td->td_proc;
1998	sig = ksi->ksi_signo;
1999	code = ksi->ksi_code;
2000	KASSERT(_SIG_VALID(sig), ("invalid signal"));
2001
2002	PROC_LOCK(p);
2003	ps = p->p_sigacts;
2004	mtx_lock(&ps->ps_mtx);
2005	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2006	    !SIGISMEMBER(td->td_sigmask, sig)) {
2007#ifdef KTRACE
2008		if (KTRPOINT(curthread, KTR_PSIG))
2009			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2010			    &td->td_sigmask, code);
2011#endif
2012		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2013				ksi, &td->td_sigmask);
2014		postsig_done(sig, td, ps);
2015		mtx_unlock(&ps->ps_mtx);
2016	} else {
2017		/*
2018		 * Avoid a possible infinite loop if the thread
2019		 * masking the signal or process is ignoring the
2020		 * signal.
2021		 */
2022		if (kern_forcesigexit &&
2023		    (SIGISMEMBER(td->td_sigmask, sig) ||
2024		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2025			SIGDELSET(td->td_sigmask, sig);
2026			SIGDELSET(ps->ps_sigcatch, sig);
2027			SIGDELSET(ps->ps_sigignore, sig);
2028			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2029		}
2030		mtx_unlock(&ps->ps_mtx);
2031		p->p_code = code;	/* XXX for core dump/debugger */
2032		p->p_sig = sig;		/* XXX to verify code */
2033		tdsendsignal(p, td, sig, ksi);
2034	}
2035	PROC_UNLOCK(p);
2036}
2037
2038static struct thread *
2039sigtd(struct proc *p, int sig, int prop)
2040{
2041	struct thread *td, *signal_td;
2042
2043	PROC_LOCK_ASSERT(p, MA_OWNED);
2044
2045	/*
2046	 * Check if current thread can handle the signal without
2047	 * switching context to another thread.
2048	 */
2049	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig))
2050		return (curthread);
2051	signal_td = NULL;
2052	FOREACH_THREAD_IN_PROC(p, td) {
2053		if (!SIGISMEMBER(td->td_sigmask, sig)) {
2054			signal_td = td;
2055			break;
2056		}
2057	}
2058	if (signal_td == NULL)
2059		signal_td = FIRST_THREAD_IN_PROC(p);
2060	return (signal_td);
2061}
2062
2063/*
2064 * Send the signal to the process.  If the signal has an action, the action
2065 * is usually performed by the target process rather than the caller; we add
2066 * the signal to the set of pending signals for the process.
2067 *
2068 * Exceptions:
2069 *   o When a stop signal is sent to a sleeping process that takes the
2070 *     default action, the process is stopped without awakening it.
2071 *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2072 *     regardless of the signal action (eg, blocked or ignored).
2073 *
2074 * Other ignored signals are discarded immediately.
2075 *
2076 * NB: This function may be entered from the debugger via the "kill" DDB
2077 * command.  There is little that can be done to mitigate the possibly messy
2078 * side effects of this unwise possibility.
2079 */
2080void
2081kern_psignal(struct proc *p, int sig)
2082{
2083	ksiginfo_t ksi;
2084
2085	ksiginfo_init(&ksi);
2086	ksi.ksi_signo = sig;
2087	ksi.ksi_code = SI_KERNEL;
2088	(void) tdsendsignal(p, NULL, sig, &ksi);
2089}
2090
2091int
2092pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2093{
2094
2095	return (tdsendsignal(p, NULL, sig, ksi));
2096}
2097
2098/* Utility function for finding a thread to send signal event to. */
2099int
2100sigev_findtd(struct proc *p ,struct sigevent *sigev, struct thread **ttd)
2101{
2102	struct thread *td;
2103
2104	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2105		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2106		if (td == NULL)
2107			return (ESRCH);
2108		*ttd = td;
2109	} else {
2110		*ttd = NULL;
2111		PROC_LOCK(p);
2112	}
2113	return (0);
2114}
2115
2116void
2117tdsignal(struct thread *td, int sig)
2118{
2119	ksiginfo_t ksi;
2120
2121	ksiginfo_init(&ksi);
2122	ksi.ksi_signo = sig;
2123	ksi.ksi_code = SI_KERNEL;
2124	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2125}
2126
2127void
2128tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2129{
2130
2131	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2132}
2133
2134int
2135tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2136{
2137	sig_t action;
2138	sigqueue_t *sigqueue;
2139	int prop;
2140	struct sigacts *ps;
2141	int intrval;
2142	int ret = 0;
2143	int wakeup_swapper;
2144
2145	MPASS(td == NULL || p == td->td_proc);
2146	PROC_LOCK_ASSERT(p, MA_OWNED);
2147
2148	if (!_SIG_VALID(sig))
2149		panic("%s(): invalid signal %d", __func__, sig);
2150
2151	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2152
2153	/*
2154	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2155	 */
2156	if (p->p_state == PRS_ZOMBIE) {
2157		if (ksi && (ksi->ksi_flags & KSI_INS))
2158			ksiginfo_tryfree(ksi);
2159		return (ret);
2160	}
2161
2162	ps = p->p_sigacts;
2163	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2164	prop = sigprop(sig);
2165
2166	if (td == NULL) {
2167		td = sigtd(p, sig, prop);
2168		sigqueue = &p->p_sigqueue;
2169	} else
2170		sigqueue = &td->td_sigqueue;
2171
2172	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2173
2174	/*
2175	 * If the signal is being ignored,
2176	 * then we forget about it immediately.
2177	 * (Note: we don't set SIGCONT in ps_sigignore,
2178	 * and if it is set to SIG_IGN,
2179	 * action will be SIG_DFL here.)
2180	 */
2181	mtx_lock(&ps->ps_mtx);
2182	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2183		SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2184
2185		mtx_unlock(&ps->ps_mtx);
2186		if (ksi && (ksi->ksi_flags & KSI_INS))
2187			ksiginfo_tryfree(ksi);
2188		return (ret);
2189	}
2190	if (SIGISMEMBER(td->td_sigmask, sig))
2191		action = SIG_HOLD;
2192	else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2193		action = SIG_CATCH;
2194	else
2195		action = SIG_DFL;
2196	if (SIGISMEMBER(ps->ps_sigintr, sig))
2197		intrval = EINTR;
2198	else
2199		intrval = ERESTART;
2200	mtx_unlock(&ps->ps_mtx);
2201
2202	if (prop & SA_CONT)
2203		sigqueue_delete_stopmask_proc(p);
2204	else if (prop & SA_STOP) {
2205		/*
2206		 * If sending a tty stop signal to a member of an orphaned
2207		 * process group, discard the signal here if the action
2208		 * is default; don't stop the process below if sleeping,
2209		 * and don't clear any pending SIGCONT.
2210		 */
2211		if ((prop & SA_TTYSTOP) &&
2212		    (p->p_pgrp->pg_jobc == 0) &&
2213		    (action == SIG_DFL)) {
2214			if (ksi && (ksi->ksi_flags & KSI_INS))
2215				ksiginfo_tryfree(ksi);
2216			return (ret);
2217		}
2218		sigqueue_delete_proc(p, SIGCONT);
2219		if (p->p_flag & P_CONTINUED) {
2220			p->p_flag &= ~P_CONTINUED;
2221			PROC_LOCK(p->p_pptr);
2222			sigqueue_take(p->p_ksi);
2223			PROC_UNLOCK(p->p_pptr);
2224		}
2225	}
2226
2227	ret = sigqueue_add(sigqueue, sig, ksi);
2228	if (ret != 0)
2229		return (ret);
2230	signotify(td);
2231	/*
2232	 * Defer further processing for signals which are held,
2233	 * except that stopped processes must be continued by SIGCONT.
2234	 */
2235	if (action == SIG_HOLD &&
2236	    !((prop & SA_CONT) && (p->p_flag & P_STOPPED_SIG)))
2237		return (ret);
2238
2239	/* SIGKILL: Remove procfs STOPEVENTs. */
2240	if (sig == SIGKILL) {
2241		/* from procfs_ioctl.c: PIOCBIC */
2242		p->p_stops = 0;
2243		/* from procfs_ioctl.c: PIOCCONT */
2244		p->p_step = 0;
2245		wakeup(&p->p_step);
2246	}
2247	/*
2248	 * Some signals have a process-wide effect and a per-thread
2249	 * component.  Most processing occurs when the process next
2250	 * tries to cross the user boundary, however there are some
2251	 * times when processing needs to be done immediately, such as
2252	 * waking up threads so that they can cross the user boundary.
2253	 * We try to do the per-process part here.
2254	 */
2255	if (P_SHOULDSTOP(p)) {
2256		KASSERT(!(p->p_flag & P_WEXIT),
2257		    ("signal to stopped but exiting process"));
2258		if (sig == SIGKILL) {
2259			/*
2260			 * If traced process is already stopped,
2261			 * then no further action is necessary.
2262			 */
2263			if (p->p_flag & P_TRACED)
2264				goto out;
2265			/*
2266			 * SIGKILL sets process running.
2267			 * It will die elsewhere.
2268			 * All threads must be restarted.
2269			 */
2270			p->p_flag &= ~P_STOPPED_SIG;
2271			goto runfast;
2272		}
2273
2274		if (prop & SA_CONT) {
2275			/*
2276			 * If traced process is already stopped,
2277			 * then no further action is necessary.
2278			 */
2279			if (p->p_flag & P_TRACED)
2280				goto out;
2281			/*
2282			 * If SIGCONT is default (or ignored), we continue the
2283			 * process but don't leave the signal in sigqueue as
2284			 * it has no further action.  If SIGCONT is held, we
2285			 * continue the process and leave the signal in
2286			 * sigqueue.  If the process catches SIGCONT, let it
2287			 * handle the signal itself.  If it isn't waiting on
2288			 * an event, it goes back to run state.
2289			 * Otherwise, process goes back to sleep state.
2290			 */
2291			p->p_flag &= ~P_STOPPED_SIG;
2292			PROC_SLOCK(p);
2293			if (p->p_numthreads == p->p_suspcount) {
2294				PROC_SUNLOCK(p);
2295				p->p_flag |= P_CONTINUED;
2296				p->p_xsig = SIGCONT;
2297				PROC_LOCK(p->p_pptr);
2298				childproc_continued(p);
2299				PROC_UNLOCK(p->p_pptr);
2300				PROC_SLOCK(p);
2301			}
2302			if (action == SIG_DFL) {
2303				thread_unsuspend(p);
2304				PROC_SUNLOCK(p);
2305				sigqueue_delete(sigqueue, sig);
2306				goto out;
2307			}
2308			if (action == SIG_CATCH) {
2309				/*
2310				 * The process wants to catch it so it needs
2311				 * to run at least one thread, but which one?
2312				 */
2313				PROC_SUNLOCK(p);
2314				goto runfast;
2315			}
2316			/*
2317			 * The signal is not ignored or caught.
2318			 */
2319			thread_unsuspend(p);
2320			PROC_SUNLOCK(p);
2321			goto out;
2322		}
2323
2324		if (prop & SA_STOP) {
2325			/*
2326			 * If traced process is already stopped,
2327			 * then no further action is necessary.
2328			 */
2329			if (p->p_flag & P_TRACED)
2330				goto out;
2331			/*
2332			 * Already stopped, don't need to stop again
2333			 * (If we did the shell could get confused).
2334			 * Just make sure the signal STOP bit set.
2335			 */
2336			p->p_flag |= P_STOPPED_SIG;
2337			sigqueue_delete(sigqueue, sig);
2338			goto out;
2339		}
2340
2341		/*
2342		 * All other kinds of signals:
2343		 * If a thread is sleeping interruptibly, simulate a
2344		 * wakeup so that when it is continued it will be made
2345		 * runnable and can look at the signal.  However, don't make
2346		 * the PROCESS runnable, leave it stopped.
2347		 * It may run a bit until it hits a thread_suspend_check().
2348		 */
2349		wakeup_swapper = 0;
2350		PROC_SLOCK(p);
2351		thread_lock(td);
2352		if (TD_ON_SLEEPQ(td) && (td->td_flags & TDF_SINTR))
2353			wakeup_swapper = sleepq_abort(td, intrval);
2354		thread_unlock(td);
2355		PROC_SUNLOCK(p);
2356		if (wakeup_swapper)
2357			kick_proc0();
2358		goto out;
2359		/*
2360		 * Mutexes are short lived. Threads waiting on them will
2361		 * hit thread_suspend_check() soon.
2362		 */
2363	} else if (p->p_state == PRS_NORMAL) {
2364		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2365			tdsigwakeup(td, sig, action, intrval);
2366			goto out;
2367		}
2368
2369		MPASS(action == SIG_DFL);
2370
2371		if (prop & SA_STOP) {
2372			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2373				goto out;
2374			p->p_flag |= P_STOPPED_SIG;
2375			p->p_xsig = sig;
2376			PROC_SLOCK(p);
2377			wakeup_swapper = sig_suspend_threads(td, p, 1);
2378			if (p->p_numthreads == p->p_suspcount) {
2379				/*
2380				 * only thread sending signal to another
2381				 * process can reach here, if thread is sending
2382				 * signal to its process, because thread does
2383				 * not suspend itself here, p_numthreads
2384				 * should never be equal to p_suspcount.
2385				 */
2386				thread_stopped(p);
2387				PROC_SUNLOCK(p);
2388				sigqueue_delete_proc(p, p->p_xsig);
2389			} else
2390				PROC_SUNLOCK(p);
2391			if (wakeup_swapper)
2392				kick_proc0();
2393			goto out;
2394		}
2395	} else {
2396		/* Not in "NORMAL" state. discard the signal. */
2397		sigqueue_delete(sigqueue, sig);
2398		goto out;
2399	}
2400
2401	/*
2402	 * The process is not stopped so we need to apply the signal to all the
2403	 * running threads.
2404	 */
2405runfast:
2406	tdsigwakeup(td, sig, action, intrval);
2407	PROC_SLOCK(p);
2408	thread_unsuspend(p);
2409	PROC_SUNLOCK(p);
2410out:
2411	/* If we jump here, proc slock should not be owned. */
2412	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2413	return (ret);
2414}
2415
2416/*
2417 * The force of a signal has been directed against a single
2418 * thread.  We need to see what we can do about knocking it
2419 * out of any sleep it may be in etc.
2420 */
2421static void
2422tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2423{
2424	struct proc *p = td->td_proc;
2425	int prop;
2426	int wakeup_swapper;
2427
2428	wakeup_swapper = 0;
2429	PROC_LOCK_ASSERT(p, MA_OWNED);
2430	prop = sigprop(sig);
2431
2432	PROC_SLOCK(p);
2433	thread_lock(td);
2434	/*
2435	 * Bring the priority of a thread up if we want it to get
2436	 * killed in this lifetime.  Be careful to avoid bumping the
2437	 * priority of the idle thread, since we still allow to signal
2438	 * kernel processes.
2439	 */
2440	if (action == SIG_DFL && (prop & SA_KILL) != 0 &&
2441	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2442		sched_prio(td, PUSER);
2443	if (TD_ON_SLEEPQ(td)) {
2444		/*
2445		 * If thread is sleeping uninterruptibly
2446		 * we can't interrupt the sleep... the signal will
2447		 * be noticed when the process returns through
2448		 * trap() or syscall().
2449		 */
2450		if ((td->td_flags & TDF_SINTR) == 0)
2451			goto out;
2452		/*
2453		 * If SIGCONT is default (or ignored) and process is
2454		 * asleep, we are finished; the process should not
2455		 * be awakened.
2456		 */
2457		if ((prop & SA_CONT) && action == SIG_DFL) {
2458			thread_unlock(td);
2459			PROC_SUNLOCK(p);
2460			sigqueue_delete(&p->p_sigqueue, sig);
2461			/*
2462			 * It may be on either list in this state.
2463			 * Remove from both for now.
2464			 */
2465			sigqueue_delete(&td->td_sigqueue, sig);
2466			return;
2467		}
2468
2469		/*
2470		 * Don't awaken a sleeping thread for SIGSTOP if the
2471		 * STOP signal is deferred.
2472		 */
2473		if ((prop & SA_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2474		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2475			goto out;
2476
2477		/*
2478		 * Give low priority threads a better chance to run.
2479		 */
2480		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2481			sched_prio(td, PUSER);
2482
2483		wakeup_swapper = sleepq_abort(td, intrval);
2484	} else {
2485		/*
2486		 * Other states do nothing with the signal immediately,
2487		 * other than kicking ourselves if we are running.
2488		 * It will either never be noticed, or noticed very soon.
2489		 */
2490#ifdef SMP
2491		if (TD_IS_RUNNING(td) && td != curthread)
2492			forward_signal(td);
2493#endif
2494	}
2495out:
2496	PROC_SUNLOCK(p);
2497	thread_unlock(td);
2498	if (wakeup_swapper)
2499		kick_proc0();
2500}
2501
2502static int
2503sig_suspend_threads(struct thread *td, struct proc *p, int sending)
2504{
2505	struct thread *td2;
2506	int wakeup_swapper;
2507
2508	PROC_LOCK_ASSERT(p, MA_OWNED);
2509	PROC_SLOCK_ASSERT(p, MA_OWNED);
2510	MPASS(sending || td == curthread);
2511
2512	wakeup_swapper = 0;
2513	FOREACH_THREAD_IN_PROC(p, td2) {
2514		thread_lock(td2);
2515		td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
2516		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2517		    (td2->td_flags & TDF_SINTR)) {
2518			if (td2->td_flags & TDF_SBDRY) {
2519				/*
2520				 * Once a thread is asleep with
2521				 * TDF_SBDRY and without TDF_SERESTART
2522				 * or TDF_SEINTR set, it should never
2523				 * become suspended due to this check.
2524				 */
2525				KASSERT(!TD_IS_SUSPENDED(td2),
2526				    ("thread with deferred stops suspended"));
2527				if (TD_SBDRY_INTR(td2))
2528					wakeup_swapper |= sleepq_abort(td2,
2529					    TD_SBDRY_ERRNO(td2));
2530			} else if (!TD_IS_SUSPENDED(td2)) {
2531				thread_suspend_one(td2);
2532			}
2533		} else if (!TD_IS_SUSPENDED(td2)) {
2534			if (sending || td != td2)
2535				td2->td_flags |= TDF_ASTPENDING;
2536#ifdef SMP
2537			if (TD_IS_RUNNING(td2) && td2 != td)
2538				forward_signal(td2);
2539#endif
2540		}
2541		thread_unlock(td2);
2542	}
2543	return (wakeup_swapper);
2544}
2545
2546/*
2547 * Stop the process for an event deemed interesting to the debugger. If si is
2548 * non-NULL, this is a signal exchange; the new signal requested by the
2549 * debugger will be returned for handling. If si is NULL, this is some other
2550 * type of interesting event. The debugger may request a signal be delivered in
2551 * that case as well, however it will be deferred until it can be handled.
2552 */
2553int
2554ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2555{
2556	struct proc *p = td->td_proc;
2557	struct thread *td2;
2558	ksiginfo_t ksi;
2559	int prop;
2560
2561	PROC_LOCK_ASSERT(p, MA_OWNED);
2562	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2563	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2564	    &p->p_mtx.lock_object, "Stopping for traced signal");
2565
2566	td->td_xsig = sig;
2567
2568	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2569		td->td_dbgflags |= TDB_XSIG;
2570		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2571		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2572		PROC_SLOCK(p);
2573		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2574			if (P_KILLED(p)) {
2575				/*
2576				 * Ensure that, if we've been PT_KILLed, the
2577				 * exit status reflects that. Another thread
2578				 * may also be in ptracestop(), having just
2579				 * received the SIGKILL, but this thread was
2580				 * unsuspended first.
2581				 */
2582				td->td_dbgflags &= ~TDB_XSIG;
2583				td->td_xsig = SIGKILL;
2584				p->p_ptevents = 0;
2585				break;
2586			}
2587			if (p->p_flag & P_SINGLE_EXIT &&
2588			    !(td->td_dbgflags & TDB_EXIT)) {
2589				/*
2590				 * Ignore ptrace stops except for thread exit
2591				 * events when the process exits.
2592				 */
2593				td->td_dbgflags &= ~TDB_XSIG;
2594				PROC_SUNLOCK(p);
2595				return (0);
2596			}
2597
2598			/*
2599			 * Make wait(2) work.  Ensure that right after the
2600			 * attach, the thread which was decided to become the
2601			 * leader of attach gets reported to the waiter.
2602			 * Otherwise, just avoid overwriting another thread's
2603			 * assignment to p_xthread.  If another thread has
2604			 * already set p_xthread, the current thread will get
2605			 * a chance to report itself upon the next iteration.
2606			 */
2607			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2608			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2609			    p->p_xthread == NULL)) {
2610				p->p_xsig = sig;
2611				p->p_xthread = td;
2612
2613				/*
2614				 * If we are on sleepqueue already,
2615				 * let sleepqueue code decide if it
2616				 * needs to go sleep after attach.
2617				 */
2618				if (td->td_wchan == NULL)
2619					td->td_dbgflags &= ~TDB_FSTP;
2620
2621				p->p_flag2 &= ~P2_PTRACE_FSTP;
2622				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2623				sig_suspend_threads(td, p, 0);
2624			}
2625			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2626				td->td_dbgflags &= ~TDB_STOPATFORK;
2627			}
2628stopme:
2629			thread_suspend_switch(td, p);
2630			if (p->p_xthread == td)
2631				p->p_xthread = NULL;
2632			if (!(p->p_flag & P_TRACED))
2633				break;
2634			if (td->td_dbgflags & TDB_SUSPEND) {
2635				if (p->p_flag & P_SINGLE_EXIT)
2636					break;
2637				goto stopme;
2638			}
2639		}
2640		PROC_SUNLOCK(p);
2641	}
2642
2643	if (si != NULL && sig == td->td_xsig) {
2644		/* Parent wants us to take the original signal unchanged. */
2645		si->ksi_flags |= KSI_HEAD;
2646		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2647			si->ksi_signo = 0;
2648	} else if (td->td_xsig != 0) {
2649		/*
2650		 * If parent wants us to take a new signal, then it will leave
2651		 * it in td->td_xsig; otherwise we just look for signals again.
2652		 */
2653		ksiginfo_init(&ksi);
2654		ksi.ksi_signo = td->td_xsig;
2655		ksi.ksi_flags |= KSI_PTRACE;
2656		prop = sigprop(td->td_xsig);
2657		td2 = sigtd(p, td->td_xsig, prop);
2658		tdsendsignal(p, td2, td->td_xsig, &ksi);
2659		if (td != td2)
2660			return (0);
2661	}
2662
2663	return (td->td_xsig);
2664}
2665
2666static void
2667reschedule_signals(struct proc *p, sigset_t block, int flags)
2668{
2669	struct sigacts *ps;
2670	struct thread *td;
2671	int sig;
2672
2673	PROC_LOCK_ASSERT(p, MA_OWNED);
2674	ps = p->p_sigacts;
2675	mtx_assert(&ps->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0 ?
2676	    MA_OWNED : MA_NOTOWNED);
2677	if (SIGISEMPTY(p->p_siglist))
2678		return;
2679	SIGSETAND(block, p->p_siglist);
2680	while ((sig = sig_ffs(&block)) != 0) {
2681		SIGDELSET(block, sig);
2682		td = sigtd(p, sig, 0);
2683		signotify(td);
2684		if (!(flags & SIGPROCMASK_PS_LOCKED))
2685			mtx_lock(&ps->ps_mtx);
2686		if (p->p_flag & P_TRACED ||
2687		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2688		    !SIGISMEMBER(td->td_sigmask, sig)))
2689			tdsigwakeup(td, sig, SIG_CATCH,
2690			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2691			     ERESTART));
2692		if (!(flags & SIGPROCMASK_PS_LOCKED))
2693			mtx_unlock(&ps->ps_mtx);
2694	}
2695}
2696
2697void
2698tdsigcleanup(struct thread *td)
2699{
2700	struct proc *p;
2701	sigset_t unblocked;
2702
2703	p = td->td_proc;
2704	PROC_LOCK_ASSERT(p, MA_OWNED);
2705
2706	sigqueue_flush(&td->td_sigqueue);
2707	if (p->p_numthreads == 1)
2708		return;
2709
2710	/*
2711	 * Since we cannot handle signals, notify signal post code
2712	 * about this by filling the sigmask.
2713	 *
2714	 * Also, if needed, wake up thread(s) that do not block the
2715	 * same signals as the exiting thread, since the thread might
2716	 * have been selected for delivery and woken up.
2717	 */
2718	SIGFILLSET(unblocked);
2719	SIGSETNAND(unblocked, td->td_sigmask);
2720	SIGFILLSET(td->td_sigmask);
2721	reschedule_signals(p, unblocked, 0);
2722
2723}
2724
2725static int
2726sigdeferstop_curr_flags(int cflags)
2727{
2728
2729	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
2730	    (cflags & TDF_SBDRY) != 0);
2731	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
2732}
2733
2734/*
2735 * Defer the delivery of SIGSTOP for the current thread, according to
2736 * the requested mode.  Returns previous flags, which must be restored
2737 * by sigallowstop().
2738 *
2739 * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
2740 * cleared by the current thread, which allow the lock-less read-only
2741 * accesses below.
2742 */
2743int
2744sigdeferstop_impl(int mode)
2745{
2746	struct thread *td;
2747	int cflags, nflags;
2748
2749	td = curthread;
2750	cflags = sigdeferstop_curr_flags(td->td_flags);
2751	switch (mode) {
2752	case SIGDEFERSTOP_NOP:
2753		nflags = cflags;
2754		break;
2755	case SIGDEFERSTOP_OFF:
2756		nflags = 0;
2757		break;
2758	case SIGDEFERSTOP_SILENT:
2759		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
2760		break;
2761	case SIGDEFERSTOP_EINTR:
2762		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
2763		break;
2764	case SIGDEFERSTOP_ERESTART:
2765		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
2766		break;
2767	default:
2768		panic("sigdeferstop: invalid mode %x", mode);
2769		break;
2770	}
2771	if (cflags == nflags)
2772		return (SIGDEFERSTOP_VAL_NCHG);
2773	thread_lock(td);
2774	td->td_flags = (td->td_flags & ~cflags) | nflags;
2775	thread_unlock(td);
2776	return (cflags);
2777}
2778
2779/*
2780 * Restores the STOP handling mode, typically permitting the delivery
2781 * of SIGSTOP for the current thread.  This does not immediately
2782 * suspend if a stop was posted.  Instead, the thread will suspend
2783 * either via ast() or a subsequent interruptible sleep.
2784 */
2785void
2786sigallowstop_impl(int prev)
2787{
2788	struct thread *td;
2789	int cflags;
2790
2791	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
2792	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
2793	    ("sigallowstop: incorrect previous mode %x", prev));
2794	td = curthread;
2795	cflags = sigdeferstop_curr_flags(td->td_flags);
2796	if (cflags != prev) {
2797		thread_lock(td);
2798		td->td_flags = (td->td_flags & ~cflags) | prev;
2799		thread_unlock(td);
2800	}
2801}
2802
2803/*
2804 * If the current process has received a signal (should be caught or cause
2805 * termination, should interrupt current syscall), return the signal number.
2806 * Stop signals with default action are processed immediately, then cleared;
2807 * they aren't returned.  This is checked after each entry to the system for
2808 * a syscall or trap (though this can usually be done without calling issignal
2809 * by checking the pending signal masks in cursig.) The normal call
2810 * sequence is
2811 *
2812 *	while (sig = cursig(curthread))
2813 *		postsig(sig);
2814 */
2815static int
2816issignal(struct thread *td)
2817{
2818	struct proc *p;
2819	struct sigacts *ps;
2820	struct sigqueue *queue;
2821	sigset_t sigpending;
2822	int prop, sig, traced;
2823	ksiginfo_t ksi;
2824
2825	p = td->td_proc;
2826	ps = p->p_sigacts;
2827	mtx_assert(&ps->ps_mtx, MA_OWNED);
2828	PROC_LOCK_ASSERT(p, MA_OWNED);
2829	for (;;) {
2830		traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
2831
2832		sigpending = td->td_sigqueue.sq_signals;
2833		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
2834		SIGSETNAND(sigpending, td->td_sigmask);
2835
2836		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
2837		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2838			SIG_STOPSIGMASK(sigpending);
2839		if (SIGISEMPTY(sigpending))	/* no signal to send */
2840			return (0);
2841		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
2842		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
2843		    SIGISMEMBER(sigpending, SIGSTOP)) {
2844			/*
2845			 * If debugger just attached, always consume
2846			 * SIGSTOP from ptrace(PT_ATTACH) first, to
2847			 * execute the debugger attach ritual in
2848			 * order.
2849			 */
2850			sig = SIGSTOP;
2851			td->td_dbgflags |= TDB_FSTP;
2852		} else {
2853			sig = sig_ffs(&sigpending);
2854		}
2855
2856		if (p->p_stops & S_SIG) {
2857			mtx_unlock(&ps->ps_mtx);
2858			stopevent(p, S_SIG, sig);
2859			mtx_lock(&ps->ps_mtx);
2860		}
2861
2862		/*
2863		 * We should see pending but ignored signals
2864		 * only if P_TRACED was on when they were posted.
2865		 */
2866		if (SIGISMEMBER(ps->ps_sigignore, sig) && (traced == 0)) {
2867			sigqueue_delete(&td->td_sigqueue, sig);
2868			sigqueue_delete(&p->p_sigqueue, sig);
2869			continue;
2870		}
2871		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
2872			/*
2873			 * If traced, always stop.
2874			 * Remove old signal from queue before the stop.
2875			 * XXX shrug off debugger, it causes siginfo to
2876			 * be thrown away.
2877			 */
2878			queue = &td->td_sigqueue;
2879			ksiginfo_init(&ksi);
2880			if (sigqueue_get(queue, sig, &ksi) == 0) {
2881				queue = &p->p_sigqueue;
2882				sigqueue_get(queue, sig, &ksi);
2883			}
2884			td->td_si = ksi.ksi_info;
2885
2886			mtx_unlock(&ps->ps_mtx);
2887			sig = ptracestop(td, sig, &ksi);
2888			mtx_lock(&ps->ps_mtx);
2889
2890			td->td_si.si_signo = 0;
2891
2892			/*
2893			 * Keep looking if the debugger discarded or
2894			 * replaced the signal.
2895			 */
2896			if (sig == 0)
2897				continue;
2898
2899			/*
2900			 * If the signal became masked, re-queue it.
2901			 */
2902			if (SIGISMEMBER(td->td_sigmask, sig)) {
2903				ksi.ksi_flags |= KSI_HEAD;
2904				sigqueue_add(&p->p_sigqueue, sig, &ksi);
2905				continue;
2906			}
2907
2908			/*
2909			 * If the traced bit got turned off, requeue
2910			 * the signal and go back up to the top to
2911			 * rescan signals.  This ensures that p_sig*
2912			 * and p_sigact are consistent.
2913			 */
2914			if ((p->p_flag & P_TRACED) == 0) {
2915				ksi.ksi_flags |= KSI_HEAD;
2916				sigqueue_add(queue, sig, &ksi);
2917				continue;
2918			}
2919		}
2920
2921		prop = sigprop(sig);
2922
2923		/*
2924		 * Decide whether the signal should be returned.
2925		 * Return the signal's number, or fall through
2926		 * to clear it from the pending mask.
2927		 */
2928		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2929
2930		case (intptr_t)SIG_DFL:
2931			/*
2932			 * Don't take default actions on system processes.
2933			 */
2934			if (p->p_pid <= 1) {
2935#ifdef DIAGNOSTIC
2936				/*
2937				 * Are you sure you want to ignore SIGSEGV
2938				 * in init? XXX
2939				 */
2940				printf("Process (pid %lu) got signal %d\n",
2941					(u_long)p->p_pid, sig);
2942#endif
2943				break;		/* == ignore */
2944			}
2945			/*
2946			 * If there is a pending stop signal to process with
2947			 * default action, stop here, then clear the signal.
2948			 * Traced or exiting processes should ignore stops.
2949			 * Additionally, a member of an orphaned process group
2950			 * should ignore tty stops.
2951			 */
2952			if (prop & SA_STOP) {
2953				if (p->p_flag &
2954				    (P_TRACED | P_WEXIT | P_SINGLE_EXIT) ||
2955				    (p->p_pgrp->pg_jobc == 0 &&
2956				     prop & SA_TTYSTOP))
2957					break;	/* == ignore */
2958				if (TD_SBDRY_INTR(td)) {
2959					KASSERT((td->td_flags & TDF_SBDRY) != 0,
2960					    ("lost TDF_SBDRY"));
2961					return (-1);
2962				}
2963				mtx_unlock(&ps->ps_mtx);
2964				WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2965				    &p->p_mtx.lock_object, "Catching SIGSTOP");
2966				sigqueue_delete(&td->td_sigqueue, sig);
2967				sigqueue_delete(&p->p_sigqueue, sig);
2968				p->p_flag |= P_STOPPED_SIG;
2969				p->p_xsig = sig;
2970				PROC_SLOCK(p);
2971				sig_suspend_threads(td, p, 0);
2972				thread_suspend_switch(td, p);
2973				PROC_SUNLOCK(p);
2974				mtx_lock(&ps->ps_mtx);
2975				goto next;
2976			} else if (prop & SA_IGNORE) {
2977				/*
2978				 * Except for SIGCONT, shouldn't get here.
2979				 * Default action is to ignore; drop it.
2980				 */
2981				break;		/* == ignore */
2982			} else
2983				return (sig);
2984			/*NOTREACHED*/
2985
2986		case (intptr_t)SIG_IGN:
2987			/*
2988			 * Masking above should prevent us ever trying
2989			 * to take action on an ignored signal other
2990			 * than SIGCONT, unless process is traced.
2991			 */
2992			if ((prop & SA_CONT) == 0 &&
2993			    (p->p_flag & P_TRACED) == 0)
2994				printf("issignal\n");
2995			break;		/* == ignore */
2996
2997		default:
2998			/*
2999			 * This signal has an action, let
3000			 * postsig() process it.
3001			 */
3002			return (sig);
3003		}
3004		sigqueue_delete(&td->td_sigqueue, sig);	/* take the signal! */
3005		sigqueue_delete(&p->p_sigqueue, sig);
3006next:;
3007	}
3008	/* NOTREACHED */
3009}
3010
3011void
3012thread_stopped(struct proc *p)
3013{
3014	int n;
3015
3016	PROC_LOCK_ASSERT(p, MA_OWNED);
3017	PROC_SLOCK_ASSERT(p, MA_OWNED);
3018	n = p->p_suspcount;
3019	if (p == curproc)
3020		n++;
3021	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3022		PROC_SUNLOCK(p);
3023		p->p_flag &= ~P_WAITED;
3024		PROC_LOCK(p->p_pptr);
3025		childproc_stopped(p, (p->p_flag & P_TRACED) ?
3026			CLD_TRAPPED : CLD_STOPPED);
3027		PROC_UNLOCK(p->p_pptr);
3028		PROC_SLOCK(p);
3029	}
3030}
3031
3032/*
3033 * Take the action for the specified signal
3034 * from the current set of pending signals.
3035 */
3036int
3037postsig(int sig)
3038{
3039	struct thread *td;
3040	struct proc *p;
3041	struct sigacts *ps;
3042	sig_t action;
3043	ksiginfo_t ksi;
3044	sigset_t returnmask;
3045
3046	KASSERT(sig != 0, ("postsig"));
3047
3048	td = curthread;
3049	p = td->td_proc;
3050	PROC_LOCK_ASSERT(p, MA_OWNED);
3051	ps = p->p_sigacts;
3052	mtx_assert(&ps->ps_mtx, MA_OWNED);
3053	ksiginfo_init(&ksi);
3054	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3055	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3056		return (0);
3057	ksi.ksi_signo = sig;
3058	if (ksi.ksi_code == SI_TIMER)
3059		itimer_accept(p, ksi.ksi_timerid, &ksi);
3060	action = ps->ps_sigact[_SIG_IDX(sig)];
3061#ifdef KTRACE
3062	if (KTRPOINT(td, KTR_PSIG))
3063		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3064		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3065#endif
3066	if ((p->p_stops & S_SIG) != 0) {
3067		mtx_unlock(&ps->ps_mtx);
3068		stopevent(p, S_SIG, sig);
3069		mtx_lock(&ps->ps_mtx);
3070	}
3071
3072	if (action == SIG_DFL) {
3073		/*
3074		 * Default action, where the default is to kill
3075		 * the process.  (Other cases were ignored above.)
3076		 */
3077		mtx_unlock(&ps->ps_mtx);
3078		proc_td_siginfo_capture(td, &ksi.ksi_info);
3079		sigexit(td, sig);
3080		/* NOTREACHED */
3081	} else {
3082		/*
3083		 * If we get here, the signal must be caught.
3084		 */
3085		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3086		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3087		    ("postsig action: blocked sig %d", sig));
3088
3089		/*
3090		 * Set the new mask value and also defer further
3091		 * occurrences of this signal.
3092		 *
3093		 * Special case: user has done a sigsuspend.  Here the
3094		 * current mask is not of interest, but rather the
3095		 * mask from before the sigsuspend is what we want
3096		 * restored after the signal processing is completed.
3097		 */
3098		if (td->td_pflags & TDP_OLDMASK) {
3099			returnmask = td->td_oldsigmask;
3100			td->td_pflags &= ~TDP_OLDMASK;
3101		} else
3102			returnmask = td->td_sigmask;
3103
3104		if (p->p_sig == sig) {
3105			p->p_code = 0;
3106			p->p_sig = 0;
3107		}
3108		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3109		postsig_done(sig, td, ps);
3110	}
3111	return (1);
3112}
3113
3114void
3115proc_wkilled(struct proc *p)
3116{
3117
3118	PROC_LOCK_ASSERT(p, MA_OWNED);
3119	if ((p->p_flag & P_WKILLED) == 0) {
3120		p->p_flag |= P_WKILLED;
3121		/*
3122		 * Notify swapper that there is a process to swap in.
3123		 * The notification is racy, at worst it would take 10
3124		 * seconds for the swapper process to notice.
3125		 */
3126		if ((p->p_flag & (P_INMEM | P_SWAPPINGIN)) == 0)
3127			wakeup(&proc0);
3128	}
3129}
3130
3131/*
3132 * Kill the current process for stated reason.
3133 */
3134void
3135killproc(struct proc *p, char *why)
3136{
3137
3138	PROC_LOCK_ASSERT(p, MA_OWNED);
3139	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3140	    p->p_comm);
3141	log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3142	    p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3143	    p->p_ucred ? p->p_ucred->cr_uid : -1, why);
3144	proc_wkilled(p);
3145	kern_psignal(p, SIGKILL);
3146}
3147
3148/*
3149 * Force the current process to exit with the specified signal, dumping core
3150 * if appropriate.  We bypass the normal tests for masked and caught signals,
3151 * allowing unrecoverable failures to terminate the process without changing
3152 * signal state.  Mark the accounting record with the signal termination.
3153 * If dumping core, save the signal number for the debugger.  Calls exit and
3154 * does not return.
3155 */
3156void
3157sigexit(struct thread *td, int sig)
3158{
3159	struct proc *p = td->td_proc;
3160
3161	PROC_LOCK_ASSERT(p, MA_OWNED);
3162	p->p_acflag |= AXSIG;
3163	/*
3164	 * We must be single-threading to generate a core dump.  This
3165	 * ensures that the registers in the core file are up-to-date.
3166	 * Also, the ELF dump handler assumes that the thread list doesn't
3167	 * change out from under it.
3168	 *
3169	 * XXX If another thread attempts to single-thread before us
3170	 *     (e.g. via fork()), we won't get a dump at all.
3171	 */
3172	if ((sigprop(sig) & SA_CORE) && thread_single(p, SINGLE_NO_EXIT) == 0) {
3173		p->p_sig = sig;
3174		/*
3175		 * Log signals which would cause core dumps
3176		 * (Log as LOG_INFO to appease those who don't want
3177		 * these messages.)
3178		 * XXX : Todo, as well as euid, write out ruid too
3179		 * Note that coredump() drops proc lock.
3180		 */
3181		if (coredump(td) == 0)
3182			sig |= WCOREFLAG;
3183		if (kern_logsigexit)
3184			log(LOG_INFO,
3185			    "pid %d (%s), jid %d, uid %d: exited on "
3186			    "signal %d%s\n", p->p_pid, p->p_comm,
3187			    p->p_ucred->cr_prison->pr_id,
3188			    td->td_ucred ? td->td_ucred->cr_uid : -1,
3189			    sig &~ WCOREFLAG,
3190			    sig & WCOREFLAG ? " (core dumped)" : "");
3191	} else
3192		PROC_UNLOCK(p);
3193	exit1(td, 0, sig);
3194	/* NOTREACHED */
3195}
3196
3197/*
3198 * Send queued SIGCHLD to parent when child process's state
3199 * is changed.
3200 */
3201static void
3202sigparent(struct proc *p, int reason, int status)
3203{
3204	PROC_LOCK_ASSERT(p, MA_OWNED);
3205	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3206
3207	if (p->p_ksi != NULL) {
3208		p->p_ksi->ksi_signo  = SIGCHLD;
3209		p->p_ksi->ksi_code   = reason;
3210		p->p_ksi->ksi_status = status;
3211		p->p_ksi->ksi_pid    = p->p_pid;
3212		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3213		if (KSI_ONQ(p->p_ksi))
3214			return;
3215	}
3216	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3217}
3218
3219static void
3220childproc_jobstate(struct proc *p, int reason, int sig)
3221{
3222	struct sigacts *ps;
3223
3224	PROC_LOCK_ASSERT(p, MA_OWNED);
3225	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3226
3227	/*
3228	 * Wake up parent sleeping in kern_wait(), also send
3229	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3230	 * that parent will awake, because parent may masked
3231	 * the signal.
3232	 */
3233	p->p_pptr->p_flag |= P_STATCHILD;
3234	wakeup(p->p_pptr);
3235
3236	ps = p->p_pptr->p_sigacts;
3237	mtx_lock(&ps->ps_mtx);
3238	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3239		mtx_unlock(&ps->ps_mtx);
3240		sigparent(p, reason, sig);
3241	} else
3242		mtx_unlock(&ps->ps_mtx);
3243}
3244
3245void
3246childproc_stopped(struct proc *p, int reason)
3247{
3248
3249	childproc_jobstate(p, reason, p->p_xsig);
3250}
3251
3252void
3253childproc_continued(struct proc *p)
3254{
3255	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3256}
3257
3258void
3259childproc_exited(struct proc *p)
3260{
3261	int reason, status;
3262
3263	if (WCOREDUMP(p->p_xsig)) {
3264		reason = CLD_DUMPED;
3265		status = WTERMSIG(p->p_xsig);
3266	} else if (WIFSIGNALED(p->p_xsig)) {
3267		reason = CLD_KILLED;
3268		status = WTERMSIG(p->p_xsig);
3269	} else {
3270		reason = CLD_EXITED;
3271		status = p->p_xexit;
3272	}
3273	/*
3274	 * XXX avoid calling wakeup(p->p_pptr), the work is
3275	 * done in exit1().
3276	 */
3277	sigparent(p, reason, status);
3278}
3279
3280/*
3281 * We only have 1 character for the core count in the format
3282 * string, so the range will be 0-9
3283 */
3284#define MAX_NUM_CORES 10
3285static int num_cores = 5;
3286
3287static int
3288sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3289{
3290	int error;
3291	int new_val;
3292
3293	new_val = num_cores;
3294	error = sysctl_handle_int(oidp, &new_val, 0, req);
3295	if (error != 0 || req->newptr == NULL)
3296		return (error);
3297	if (new_val > MAX_NUM_CORES)
3298		new_val = MAX_NUM_CORES;
3299	if (new_val < 0)
3300		new_val = 0;
3301	num_cores = new_val;
3302	return (0);
3303}
3304SYSCTL_PROC(_debug, OID_AUTO, ncores, CTLTYPE_INT|CTLFLAG_RW,
3305	    0, sizeof(int), sysctl_debug_num_cores_check, "I", "");
3306
3307#define	GZ_SUFFIX	".gz"
3308
3309#ifdef GZIO
3310static int compress_user_cores = 1;
3311SYSCTL_INT(_kern, OID_AUTO, compress_user_cores, CTLFLAG_RWTUN,
3312    &compress_user_cores, 0, "Compression of user corefiles");
3313
3314int compress_user_cores_gzlevel = 6;
3315SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_gzlevel, CTLFLAG_RWTUN,
3316    &compress_user_cores_gzlevel, 0, "Corefile gzip compression level");
3317#else
3318static int compress_user_cores = 0;
3319#endif
3320
3321/*
3322 * Protect the access to corefilename[] by allproc_lock.
3323 */
3324#define	corefilename_lock	allproc_lock
3325
3326static char corefilename[MAXPATHLEN] = {"%N.core"};
3327TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3328
3329static int
3330sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3331{
3332	int error;
3333
3334	sx_xlock(&corefilename_lock);
3335	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3336	    req);
3337	sx_xunlock(&corefilename_lock);
3338
3339	return (error);
3340}
3341SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3342    CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3343    "Process corefile name format string");
3344
3345static void
3346vnode_close_locked(struct thread *td, struct vnode *vp)
3347{
3348
3349	VOP_UNLOCK(vp, 0);
3350	vn_close(vp, FWRITE, td->td_ucred, td);
3351}
3352
3353/*
3354 * If the core format has a %I in it, then we need to check
3355 * for existing corefiles before defining a name.
3356 * To do this we iterate over 0..num_cores to find a
3357 * non-existing core file name to use. If all core files are
3358 * already used we choose the oldest one.
3359 */
3360static int
3361corefile_open_last(struct thread *td, char *name, int indexpos,
3362    struct vnode **vpp)
3363{
3364	struct vnode *oldvp, *nextvp, *vp;
3365	struct vattr vattr;
3366	struct nameidata nd;
3367	int error, i, flags, oflags, cmode;
3368	struct timespec lasttime;
3369
3370	nextvp = oldvp = NULL;
3371	cmode = S_IRUSR | S_IWUSR;
3372	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3373	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3374
3375	for (i = 0; i < num_cores; i++) {
3376		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3377		name[indexpos] = '0' + i;
3378
3379		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3380		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3381		    NULL);
3382		if (error != 0)
3383			break;
3384
3385		vp = nd.ni_vp;
3386		NDFREE(&nd, NDF_ONLY_PNBUF);
3387		if ((flags & O_CREAT) == O_CREAT) {
3388			nextvp = vp;
3389			break;
3390		}
3391
3392		error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3393		if (error != 0) {
3394			vnode_close_locked(td, vp);
3395			break;
3396		}
3397
3398		if (oldvp == NULL ||
3399		    lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3400		    (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3401		    lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3402			if (oldvp != NULL)
3403				vnode_close_locked(td, oldvp);
3404			oldvp = vp;
3405			lasttime = vattr.va_mtime;
3406		} else {
3407			vnode_close_locked(td, vp);
3408		}
3409	}
3410
3411	if (oldvp != NULL) {
3412		if (nextvp == NULL) {
3413			if ((td->td_proc->p_flag & P_SUGID) != 0) {
3414				error = EFAULT;
3415				vnode_close_locked(td, oldvp);
3416			} else {
3417				nextvp = oldvp;
3418			}
3419		} else {
3420			vnode_close_locked(td, oldvp);
3421		}
3422	}
3423	if (error != 0) {
3424		if (nextvp != NULL)
3425			vnode_close_locked(td, oldvp);
3426	} else {
3427		*vpp = nextvp;
3428	}
3429
3430	return (error);
3431}
3432
3433/*
3434 * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3435 * Expand the name described in corefilename, using name, uid, and pid
3436 * and open/create core file.
3437 * corefilename is a printf-like string, with three format specifiers:
3438 *	%N	name of process ("name")
3439 *	%P	process id (pid)
3440 *	%U	user id (uid)
3441 * For example, "%N.core" is the default; they can be disabled completely
3442 * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3443 * This is controlled by the sysctl variable kern.corefile (see above).
3444 */
3445static int
3446corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3447    int compress, struct vnode **vpp, char **namep)
3448{
3449	struct sbuf sb;
3450	struct nameidata nd;
3451	const char *format;
3452	char *hostname, *name;
3453	int cmode, error, flags, i, indexpos, oflags;
3454
3455	hostname = NULL;
3456	format = corefilename;
3457	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3458	indexpos = -1;
3459	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3460	sx_slock(&corefilename_lock);
3461	for (i = 0; format[i] != '\0'; i++) {
3462		switch (format[i]) {
3463		case '%':	/* Format character */
3464			i++;
3465			switch (format[i]) {
3466			case '%':
3467				sbuf_putc(&sb, '%');
3468				break;
3469			case 'H':	/* hostname */
3470				if (hostname == NULL) {
3471					hostname = malloc(MAXHOSTNAMELEN,
3472					    M_TEMP, M_WAITOK);
3473				}
3474				getcredhostname(td->td_ucred, hostname,
3475				    MAXHOSTNAMELEN);
3476				sbuf_printf(&sb, "%s", hostname);
3477				break;
3478			case 'I':	/* autoincrementing index */
3479				sbuf_printf(&sb, "0");
3480				indexpos = sbuf_len(&sb) - 1;
3481				break;
3482			case 'N':	/* process name */
3483				sbuf_printf(&sb, "%s", comm);
3484				break;
3485			case 'P':	/* process id */
3486				sbuf_printf(&sb, "%u", pid);
3487				break;
3488			case 'U':	/* user id */
3489				sbuf_printf(&sb, "%u", uid);
3490				break;
3491			default:
3492				log(LOG_ERR,
3493				    "Unknown format character %c in "
3494				    "corename `%s'\n", format[i], format);
3495				break;
3496			}
3497			break;
3498		default:
3499			sbuf_putc(&sb, format[i]);
3500			break;
3501		}
3502	}
3503	sx_sunlock(&corefilename_lock);
3504	free(hostname, M_TEMP);
3505	if (compress)
3506		sbuf_printf(&sb, GZ_SUFFIX);
3507	if (sbuf_error(&sb) != 0) {
3508		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3509		    "long\n", (long)pid, comm, (u_long)uid);
3510		sbuf_delete(&sb);
3511		free(name, M_TEMP);
3512		return (ENOMEM);
3513	}
3514	sbuf_finish(&sb);
3515	sbuf_delete(&sb);
3516
3517	if (indexpos != -1) {
3518		error = corefile_open_last(td, name, indexpos, vpp);
3519		if (error != 0) {
3520			log(LOG_ERR,
3521			    "pid %d (%s), uid (%u):  Path `%s' failed "
3522			    "on initial open test, error = %d\n",
3523			    pid, comm, uid, name, error);
3524		}
3525	} else {
3526		cmode = S_IRUSR | S_IWUSR;
3527		oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3528		    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3529		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3530		if ((td->td_proc->p_flag & P_SUGID) != 0)
3531			flags |= O_EXCL;
3532
3533		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3534		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3535		    NULL);
3536		if (error == 0) {
3537			*vpp = nd.ni_vp;
3538			NDFREE(&nd, NDF_ONLY_PNBUF);
3539		}
3540	}
3541
3542	if (error != 0) {
3543#ifdef AUDIT
3544		audit_proc_coredump(td, name, error);
3545#endif
3546		free(name, M_TEMP);
3547		return (error);
3548	}
3549	*namep = name;
3550	return (0);
3551}
3552
3553static int
3554coredump_sanitise_path(const char *path)
3555{
3556	size_t i;
3557
3558	/*
3559	 * Only send a subset of ASCII to devd(8) because it
3560	 * might pass these strings to sh -c.
3561	 */
3562	for (i = 0; path[i]; i++)
3563		if (!(isalpha(path[i]) || isdigit(path[i])) &&
3564		    path[i] != '/' && path[i] != '.' &&
3565		    path[i] != '-')
3566			return (0);
3567
3568	return (1);
3569}
3570
3571/*
3572 * Dump a process' core.  The main routine does some
3573 * policy checking, and creates the name of the coredump;
3574 * then it passes on a vnode and a size limit to the process-specific
3575 * coredump routine if there is one; if there _is not_ one, it returns
3576 * ENOSYS; otherwise it returns the error from the process-specific routine.
3577 */
3578
3579static int
3580coredump(struct thread *td)
3581{
3582	struct proc *p = td->td_proc;
3583	struct ucred *cred = td->td_ucred;
3584	struct vnode *vp;
3585	struct flock lf;
3586	struct vattr vattr;
3587	int error, error1, locked;
3588	char *name;			/* name of corefile */
3589	void *rl_cookie;
3590	off_t limit;
3591	char *data = NULL;
3592	char *fullpath, *freepath = NULL;
3593	size_t len;
3594	static const char comm_name[] = "comm=";
3595	static const char core_name[] = "core=";
3596
3597	PROC_LOCK_ASSERT(p, MA_OWNED);
3598	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3599	_STOPEVENT(p, S_CORE, 0);
3600
3601	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
3602	    (p->p_flag2 & P2_NOTRACE) != 0) {
3603		PROC_UNLOCK(p);
3604		return (EFAULT);
3605	}
3606
3607	/*
3608	 * Note that the bulk of limit checking is done after
3609	 * the corefile is created.  The exception is if the limit
3610	 * for corefiles is 0, in which case we don't bother
3611	 * creating the corefile at all.  This layout means that
3612	 * a corefile is truncated instead of not being created,
3613	 * if it is larger than the limit.
3614	 */
3615	limit = (off_t)lim_cur(td, RLIMIT_CORE);
3616	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
3617		PROC_UNLOCK(p);
3618		return (EFBIG);
3619	}
3620	PROC_UNLOCK(p);
3621
3622	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
3623	    compress_user_cores, &vp, &name);
3624	if (error != 0)
3625		return (error);
3626
3627	/*
3628	 * Don't dump to non-regular files or files with links.
3629	 * Do not dump into system files. Effective user must own the corefile.
3630	 */
3631	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
3632	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
3633	    vattr.va_uid != cred->cr_uid) {
3634		VOP_UNLOCK(vp, 0);
3635		error = EFAULT;
3636		goto out;
3637	}
3638
3639	VOP_UNLOCK(vp, 0);
3640
3641	/* Postpone other writers, including core dumps of other processes. */
3642	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
3643
3644	lf.l_whence = SEEK_SET;
3645	lf.l_start = 0;
3646	lf.l_len = 0;
3647	lf.l_type = F_WRLCK;
3648	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
3649
3650	VATTR_NULL(&vattr);
3651	vattr.va_size = 0;
3652	if (set_core_nodump_flag)
3653		vattr.va_flags = UF_NODUMP;
3654	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3655	VOP_SETATTR(vp, &vattr, cred);
3656	VOP_UNLOCK(vp, 0);
3657	PROC_LOCK(p);
3658	p->p_acflag |= ACORE;
3659	PROC_UNLOCK(p);
3660
3661	if (p->p_sysent->sv_coredump != NULL) {
3662		error = p->p_sysent->sv_coredump(td, vp, limit,
3663		    compress_user_cores ? IMGACT_CORE_COMPRESS : 0);
3664	} else {
3665		error = ENOSYS;
3666	}
3667
3668	if (locked) {
3669		lf.l_type = F_UNLCK;
3670		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3671	}
3672	vn_rangelock_unlock(vp, rl_cookie);
3673
3674	/*
3675	 * Notify the userland helper that a process triggered a core dump.
3676	 * This allows the helper to run an automated debugging session.
3677	 */
3678	if (error != 0 || coredump_devctl == 0)
3679		goto out;
3680	len = MAXPATHLEN * 2 + sizeof(comm_name) - 1 +
3681	    sizeof(' ') + sizeof(core_name) - 1;
3682	data = malloc(len, M_TEMP, M_WAITOK);
3683	if (vn_fullpath_global(td, p->p_textvp, &fullpath, &freepath) != 0)
3684		goto out;
3685	if (!coredump_sanitise_path(fullpath))
3686		goto out;
3687	snprintf(data, len, "%s%s ", comm_name, fullpath);
3688	free(freepath, M_TEMP);
3689	freepath = NULL;
3690	if (vn_fullpath_global(td, vp, &fullpath, &freepath) != 0)
3691		goto out;
3692	if (!coredump_sanitise_path(fullpath))
3693		goto out;
3694	strlcat(data, core_name, len);
3695	strlcat(data, fullpath, len);
3696	devctl_notify("kernel", "signal", "coredump", data);
3697out:
3698	error1 = vn_close(vp, FWRITE, cred, td);
3699	if (error == 0)
3700		error = error1;
3701#ifdef AUDIT
3702	audit_proc_coredump(td, name, error);
3703#endif
3704	free(freepath, M_TEMP);
3705	free(data, M_TEMP);
3706	free(name, M_TEMP);
3707	return (error);
3708}
3709
3710/*
3711 * Nonexistent system call-- signal process (may want to handle it).  Flag
3712 * error in case process won't see signal immediately (blocked or ignored).
3713 */
3714#ifndef _SYS_SYSPROTO_H_
3715struct nosys_args {
3716	int	dummy;
3717};
3718#endif
3719/* ARGSUSED */
3720int
3721nosys(struct thread *td, struct nosys_args *args)
3722{
3723	struct proc *p;
3724
3725	p = td->td_proc;
3726
3727	PROC_LOCK(p);
3728	tdsignal(td, SIGSYS);
3729	PROC_UNLOCK(p);
3730	if (kern_lognosys == 1 || kern_lognosys == 3) {
3731		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
3732		    td->td_sa.code);
3733	}
3734	if (kern_lognosys == 2 || kern_lognosys == 3 ||
3735	    (p->p_pid == 1 && (kern_lognosys & 3) == 0)) {
3736		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
3737		    td->td_sa.code);
3738	}
3739	return (ENOSYS);
3740}
3741
3742/*
3743 * Send a SIGIO or SIGURG signal to a process or process group using stored
3744 * credentials rather than those of the current process.
3745 */
3746void
3747pgsigio(struct sigio **sigiop, int sig, int checkctty)
3748{
3749	ksiginfo_t ksi;
3750	struct sigio *sigio;
3751
3752	ksiginfo_init(&ksi);
3753	ksi.ksi_signo = sig;
3754	ksi.ksi_code = SI_KERNEL;
3755
3756	SIGIO_LOCK();
3757	sigio = *sigiop;
3758	if (sigio == NULL) {
3759		SIGIO_UNLOCK();
3760		return;
3761	}
3762	if (sigio->sio_pgid > 0) {
3763		PROC_LOCK(sigio->sio_proc);
3764		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
3765			kern_psignal(sigio->sio_proc, sig);
3766		PROC_UNLOCK(sigio->sio_proc);
3767	} else if (sigio->sio_pgid < 0) {
3768		struct proc *p;
3769
3770		PGRP_LOCK(sigio->sio_pgrp);
3771		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
3772			PROC_LOCK(p);
3773			if (p->p_state == PRS_NORMAL &&
3774			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
3775			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
3776				kern_psignal(p, sig);
3777			PROC_UNLOCK(p);
3778		}
3779		PGRP_UNLOCK(sigio->sio_pgrp);
3780	}
3781	SIGIO_UNLOCK();
3782}
3783
3784static int
3785filt_sigattach(struct knote *kn)
3786{
3787	struct proc *p = curproc;
3788
3789	kn->kn_ptr.p_proc = p;
3790	kn->kn_flags |= EV_CLEAR;		/* automatically set */
3791
3792	knlist_add(p->p_klist, kn, 0);
3793
3794	return (0);
3795}
3796
3797static void
3798filt_sigdetach(struct knote *kn)
3799{
3800	struct proc *p = kn->kn_ptr.p_proc;
3801
3802	knlist_remove(p->p_klist, kn, 0);
3803}
3804
3805/*
3806 * signal knotes are shared with proc knotes, so we apply a mask to
3807 * the hint in order to differentiate them from process hints.  This
3808 * could be avoided by using a signal-specific knote list, but probably
3809 * isn't worth the trouble.
3810 */
3811static int
3812filt_signal(struct knote *kn, long hint)
3813{
3814
3815	if (hint & NOTE_SIGNAL) {
3816		hint &= ~NOTE_SIGNAL;
3817
3818		if (kn->kn_id == hint)
3819			kn->kn_data++;
3820	}
3821	return (kn->kn_data != 0);
3822}
3823
3824struct sigacts *
3825sigacts_alloc(void)
3826{
3827	struct sigacts *ps;
3828
3829	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
3830	refcount_init(&ps->ps_refcnt, 1);
3831	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
3832	return (ps);
3833}
3834
3835void
3836sigacts_free(struct sigacts *ps)
3837{
3838
3839	if (refcount_release(&ps->ps_refcnt) == 0)
3840		return;
3841	mtx_destroy(&ps->ps_mtx);
3842	free(ps, M_SUBPROC);
3843}
3844
3845struct sigacts *
3846sigacts_hold(struct sigacts *ps)
3847{
3848
3849	refcount_acquire(&ps->ps_refcnt);
3850	return (ps);
3851}
3852
3853void
3854sigacts_copy(struct sigacts *dest, struct sigacts *src)
3855{
3856
3857	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
3858	mtx_lock(&src->ps_mtx);
3859	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
3860	mtx_unlock(&src->ps_mtx);
3861}
3862
3863int
3864sigacts_shared(struct sigacts *ps)
3865{
3866
3867	return (ps->ps_refcnt > 1);
3868}
3869
3870void
3871sig_drop_caught(struct proc *p)
3872{
3873	int sig;
3874	struct sigacts *ps;
3875
3876	ps = p->p_sigacts;
3877	PROC_LOCK_ASSERT(p, MA_OWNED);
3878	mtx_assert(&ps->ps_mtx, MA_OWNED);
3879	while (SIGNOTEMPTY(ps->ps_sigcatch)) {
3880		sig = sig_ffs(&ps->ps_sigcatch);
3881		sigdflt(ps, sig);
3882		if ((sigprop(sig) & SA_IGNORE) != 0)
3883			sigqueue_delete_proc(p, sig);
3884	}
3885}
3886