kern_sig.c revision 315963
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/10/sys/kern/kern_sig.c 315963 2017-03-25 20:14:08Z badger $");
39
40#include "opt_compat.h"
41#include "opt_kdtrace.h"
42#include "opt_ktrace.h"
43#include "opt_core.h"
44#include "opt_procdesc.h"
45
46#include <sys/param.h>
47#include <sys/systm.h>
48#include <sys/signalvar.h>
49#include <sys/vnode.h>
50#include <sys/acct.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/posix4.h>
67#include <sys/pioctl.h>
68#include <sys/racct.h>
69#include <sys/resourcevar.h>
70#include <sys/sdt.h>
71#include <sys/sbuf.h>
72#include <sys/sleepqueue.h>
73#include <sys/smp.h>
74#include <sys/stat.h>
75#include <sys/sx.h>
76#include <sys/syscallsubr.h>
77#include <sys/sysctl.h>
78#include <sys/sysent.h>
79#include <sys/syslog.h>
80#include <sys/sysproto.h>
81#include <sys/timers.h>
82#include <sys/unistd.h>
83#include <sys/wait.h>
84#include <vm/vm.h>
85#include <vm/vm_extern.h>
86#include <vm/uma.h>
87
88#include <sys/jail.h>
89
90#include <machine/cpu.h>
91
92#include <security/audit/audit.h>
93
94#define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
95
96SDT_PROVIDER_DECLARE(proc);
97SDT_PROBE_DEFINE3(proc, , , signal__send,
98    "struct thread *", "struct proc *", "int");
99SDT_PROBE_DEFINE2(proc, , , signal__clear,
100    "int", "ksiginfo_t *");
101SDT_PROBE_DEFINE3(proc, , , signal__discard,
102    "struct thread *", "struct proc *", "int");
103
104static int	coredump(struct thread *);
105static int	killpg1(struct thread *td, int sig, int pgid, int all,
106		    ksiginfo_t *ksi);
107static int	issignal(struct thread *td);
108static int	sigprop(int sig);
109static void	tdsigwakeup(struct thread *, int, sig_t, int);
110static void	sig_suspend_threads(struct thread *, struct proc *, int);
111static int	filt_sigattach(struct knote *kn);
112static void	filt_sigdetach(struct knote *kn);
113static int	filt_signal(struct knote *kn, long hint);
114static struct thread *sigtd(struct proc *p, int sig, int prop);
115static void	sigqueue_start(void);
116
117static uma_zone_t	ksiginfo_zone = NULL;
118struct filterops sig_filtops = {
119	.f_isfd = 0,
120	.f_attach = filt_sigattach,
121	.f_detach = filt_sigdetach,
122	.f_event = filt_signal,
123};
124
125static int	kern_logsigexit = 1;
126SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
127    &kern_logsigexit, 0,
128    "Log processes quitting on abnormal signals to syslog(3)");
129
130static int	kern_forcesigexit = 1;
131SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
132    &kern_forcesigexit, 0, "Force trap signal to be handled");
133
134static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW, 0,
135    "POSIX real time signal");
136
137static int	max_pending_per_proc = 128;
138SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
139    &max_pending_per_proc, 0, "Max pending signals per proc");
140
141static int	preallocate_siginfo = 1024;
142TUNABLE_INT("kern.sigqueue.preallocate", &preallocate_siginfo);
143SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RD,
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
154SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
155
156/*
157 * Policy -- Can ucred cr1 send SIGIO to process cr2?
158 * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
159 * in the right situations.
160 */
161#define CANSIGIO(cr1, cr2) \
162	((cr1)->cr_uid == 0 || \
163	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
164	    (cr1)->cr_uid == (cr2)->cr_ruid || \
165	    (cr1)->cr_ruid == (cr2)->cr_uid || \
166	    (cr1)->cr_uid == (cr2)->cr_uid)
167
168static int	sugid_coredump;
169TUNABLE_INT("kern.sugid_coredump", &sugid_coredump);
170SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
171    &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
172
173static int	capmode_coredump;
174TUNABLE_INT("kern.capmode_coredump", &capmode_coredump);
175SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RW,
176    &capmode_coredump, 0, "Allow processes in capability mode to dump core");
177
178static int	do_coredump = 1;
179SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
180	&do_coredump, 0, "Enable/Disable coredumps");
181
182static int	set_core_nodump_flag = 0;
183SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
184	0, "Enable setting the NODUMP flag on coredump files");
185
186/*
187 * Signal properties and actions.
188 * The array below categorizes the signals and their default actions
189 * according to the following properties:
190 */
191#define	SA_KILL		0x01		/* terminates process by default */
192#define	SA_CORE		0x02		/* ditto and coredumps */
193#define	SA_STOP		0x04		/* suspend process */
194#define	SA_TTYSTOP	0x08		/* ditto, from tty */
195#define	SA_IGNORE	0x10		/* ignore by default */
196#define	SA_CONT		0x20		/* continue if suspended */
197#define	SA_CANTMASK	0x40		/* non-maskable, catchable */
198
199static int sigproptbl[NSIG] = {
200	SA_KILL,			/* SIGHUP */
201	SA_KILL,			/* SIGINT */
202	SA_KILL|SA_CORE,		/* SIGQUIT */
203	SA_KILL|SA_CORE,		/* SIGILL */
204	SA_KILL|SA_CORE,		/* SIGTRAP */
205	SA_KILL|SA_CORE,		/* SIGABRT */
206	SA_KILL|SA_CORE,		/* SIGEMT */
207	SA_KILL|SA_CORE,		/* SIGFPE */
208	SA_KILL,			/* SIGKILL */
209	SA_KILL|SA_CORE,		/* SIGBUS */
210	SA_KILL|SA_CORE,		/* SIGSEGV */
211	SA_KILL|SA_CORE,		/* SIGSYS */
212	SA_KILL,			/* SIGPIPE */
213	SA_KILL,			/* SIGALRM */
214	SA_KILL,			/* SIGTERM */
215	SA_IGNORE,			/* SIGURG */
216	SA_STOP,			/* SIGSTOP */
217	SA_STOP|SA_TTYSTOP,		/* SIGTSTP */
218	SA_IGNORE|SA_CONT,		/* SIGCONT */
219	SA_IGNORE,			/* SIGCHLD */
220	SA_STOP|SA_TTYSTOP,		/* SIGTTIN */
221	SA_STOP|SA_TTYSTOP,		/* SIGTTOU */
222	SA_IGNORE,			/* SIGIO */
223	SA_KILL,			/* SIGXCPU */
224	SA_KILL,			/* SIGXFSZ */
225	SA_KILL,			/* SIGVTALRM */
226	SA_KILL,			/* SIGPROF */
227	SA_IGNORE,			/* SIGWINCH  */
228	SA_IGNORE,			/* SIGINFO */
229	SA_KILL,			/* SIGUSR1 */
230	SA_KILL,			/* SIGUSR2 */
231};
232
233static void reschedule_signals(struct proc *p, sigset_t block, int flags);
234
235static void
236sigqueue_start(void)
237{
238	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
239		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
240	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
241	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
242	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
243	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
244}
245
246ksiginfo_t *
247ksiginfo_alloc(int wait)
248{
249	int flags;
250
251	flags = M_ZERO;
252	if (! wait)
253		flags |= M_NOWAIT;
254	if (ksiginfo_zone != NULL)
255		return ((ksiginfo_t *)uma_zalloc(ksiginfo_zone, flags));
256	return (NULL);
257}
258
259void
260ksiginfo_free(ksiginfo_t *ksi)
261{
262	uma_zfree(ksiginfo_zone, ksi);
263}
264
265static __inline int
266ksiginfo_tryfree(ksiginfo_t *ksi)
267{
268	if (!(ksi->ksi_flags & KSI_EXT)) {
269		uma_zfree(ksiginfo_zone, ksi);
270		return (1);
271	}
272	return (0);
273}
274
275void
276sigqueue_init(sigqueue_t *list, struct proc *p)
277{
278	SIGEMPTYSET(list->sq_signals);
279	SIGEMPTYSET(list->sq_kill);
280	SIGEMPTYSET(list->sq_ptrace);
281	TAILQ_INIT(&list->sq_list);
282	list->sq_proc = p;
283	list->sq_flags = SQ_INIT;
284}
285
286/*
287 * Get a signal's ksiginfo.
288 * Return:
289 *	0	-	signal not found
290 *	others	-	signal number
291 */
292static int
293sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
294{
295	struct proc *p = sq->sq_proc;
296	struct ksiginfo *ksi, *next;
297	int count = 0;
298
299	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
300
301	if (!SIGISMEMBER(sq->sq_signals, signo))
302		return (0);
303
304	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
305		count++;
306		SIGDELSET(sq->sq_ptrace, signo);
307		si->ksi_flags |= KSI_PTRACE;
308	}
309	if (SIGISMEMBER(sq->sq_kill, signo)) {
310		count++;
311		if (count == 1)
312			SIGDELSET(sq->sq_kill, signo);
313	}
314
315	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
316		if (ksi->ksi_signo == signo) {
317			if (count == 0) {
318				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
319				ksi->ksi_sigq = NULL;
320				ksiginfo_copy(ksi, si);
321				if (ksiginfo_tryfree(ksi) && p != NULL)
322					p->p_pendingcnt--;
323			}
324			if (++count > 1)
325				break;
326		}
327	}
328
329	if (count <= 1)
330		SIGDELSET(sq->sq_signals, signo);
331	si->ksi_signo = signo;
332	return (signo);
333}
334
335void
336sigqueue_take(ksiginfo_t *ksi)
337{
338	struct ksiginfo *kp;
339	struct proc	*p;
340	sigqueue_t	*sq;
341
342	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
343		return;
344
345	p = sq->sq_proc;
346	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
347	ksi->ksi_sigq = NULL;
348	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
349		p->p_pendingcnt--;
350
351	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
352	     kp = TAILQ_NEXT(kp, ksi_link)) {
353		if (kp->ksi_signo == ksi->ksi_signo)
354			break;
355	}
356	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
357	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
358		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
359}
360
361static int
362sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
363{
364	struct proc *p = sq->sq_proc;
365	struct ksiginfo *ksi;
366	int ret = 0;
367
368	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
369
370	/*
371	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
372	 * for these signals.
373	 */
374	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
375		SIGADDSET(sq->sq_kill, signo);
376		goto out_set_bit;
377	}
378
379	/* directly insert the ksi, don't copy it */
380	if (si->ksi_flags & KSI_INS) {
381		if (si->ksi_flags & KSI_HEAD)
382			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
383		else
384			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
385		si->ksi_sigq = sq;
386		goto out_set_bit;
387	}
388
389	if (__predict_false(ksiginfo_zone == NULL)) {
390		SIGADDSET(sq->sq_kill, signo);
391		goto out_set_bit;
392	}
393
394	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
395		signal_overflow++;
396		ret = EAGAIN;
397	} else if ((ksi = ksiginfo_alloc(0)) == NULL) {
398		signal_alloc_fail++;
399		ret = EAGAIN;
400	} else {
401		if (p != NULL)
402			p->p_pendingcnt++;
403		ksiginfo_copy(si, ksi);
404		ksi->ksi_signo = signo;
405		if (si->ksi_flags & KSI_HEAD)
406			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
407		else
408			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
409		ksi->ksi_sigq = sq;
410	}
411
412	if (ret != 0) {
413		if ((si->ksi_flags & KSI_PTRACE) != 0) {
414			SIGADDSET(sq->sq_ptrace, signo);
415			ret = 0;
416			goto out_set_bit;
417		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
418		    (si->ksi_flags & KSI_SIGQ) == 0) {
419			SIGADDSET(sq->sq_kill, signo);
420			ret = 0;
421			goto out_set_bit;
422		}
423		return (ret);
424	}
425
426out_set_bit:
427	SIGADDSET(sq->sq_signals, signo);
428	return (ret);
429}
430
431void
432sigqueue_flush(sigqueue_t *sq)
433{
434	struct proc *p = sq->sq_proc;
435	ksiginfo_t *ksi;
436
437	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
438
439	if (p != NULL)
440		PROC_LOCK_ASSERT(p, MA_OWNED);
441
442	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
443		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
444		ksi->ksi_sigq = NULL;
445		if (ksiginfo_tryfree(ksi) && p != NULL)
446			p->p_pendingcnt--;
447	}
448
449	SIGEMPTYSET(sq->sq_signals);
450	SIGEMPTYSET(sq->sq_kill);
451	SIGEMPTYSET(sq->sq_ptrace);
452}
453
454static void
455sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
456{
457	sigset_t tmp;
458	struct proc *p1, *p2;
459	ksiginfo_t *ksi, *next;
460
461	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
462	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
463	p1 = src->sq_proc;
464	p2 = dst->sq_proc;
465	/* Move siginfo to target list */
466	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
467		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
468			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
469			if (p1 != NULL)
470				p1->p_pendingcnt--;
471			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
472			ksi->ksi_sigq = dst;
473			if (p2 != NULL)
474				p2->p_pendingcnt++;
475		}
476	}
477
478	/* Move pending bits to target list */
479	tmp = src->sq_kill;
480	SIGSETAND(tmp, *set);
481	SIGSETOR(dst->sq_kill, tmp);
482	SIGSETNAND(src->sq_kill, tmp);
483
484	tmp = src->sq_ptrace;
485	SIGSETAND(tmp, *set);
486	SIGSETOR(dst->sq_ptrace, tmp);
487	SIGSETNAND(src->sq_ptrace, tmp);
488
489	tmp = src->sq_signals;
490	SIGSETAND(tmp, *set);
491	SIGSETOR(dst->sq_signals, tmp);
492	SIGSETNAND(src->sq_signals, tmp);
493}
494
495#if 0
496static void
497sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
498{
499	sigset_t set;
500
501	SIGEMPTYSET(set);
502	SIGADDSET(set, signo);
503	sigqueue_move_set(src, dst, &set);
504}
505#endif
506
507static void
508sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
509{
510	struct proc *p = sq->sq_proc;
511	ksiginfo_t *ksi, *next;
512
513	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
514
515	/* Remove siginfo queue */
516	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
517		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
518			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
519			ksi->ksi_sigq = NULL;
520			if (ksiginfo_tryfree(ksi) && p != NULL)
521				p->p_pendingcnt--;
522		}
523	}
524	SIGSETNAND(sq->sq_kill, *set);
525	SIGSETNAND(sq->sq_ptrace, *set);
526	SIGSETNAND(sq->sq_signals, *set);
527}
528
529void
530sigqueue_delete(sigqueue_t *sq, int signo)
531{
532	sigset_t set;
533
534	SIGEMPTYSET(set);
535	SIGADDSET(set, signo);
536	sigqueue_delete_set(sq, &set);
537}
538
539/* Remove a set of signals for a process */
540static void
541sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
542{
543	sigqueue_t worklist;
544	struct thread *td0;
545
546	PROC_LOCK_ASSERT(p, MA_OWNED);
547
548	sigqueue_init(&worklist, NULL);
549	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
550
551	FOREACH_THREAD_IN_PROC(p, td0)
552		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
553
554	sigqueue_flush(&worklist);
555}
556
557void
558sigqueue_delete_proc(struct proc *p, int signo)
559{
560	sigset_t set;
561
562	SIGEMPTYSET(set);
563	SIGADDSET(set, signo);
564	sigqueue_delete_set_proc(p, &set);
565}
566
567static void
568sigqueue_delete_stopmask_proc(struct proc *p)
569{
570	sigset_t set;
571
572	SIGEMPTYSET(set);
573	SIGADDSET(set, SIGSTOP);
574	SIGADDSET(set, SIGTSTP);
575	SIGADDSET(set, SIGTTIN);
576	SIGADDSET(set, SIGTTOU);
577	sigqueue_delete_set_proc(p, &set);
578}
579
580/*
581 * Determine signal that should be delivered to thread td, the current
582 * thread, 0 if none.  If there is a pending stop signal with default
583 * action, the process stops in issignal().
584 */
585int
586cursig(struct thread *td)
587{
588	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
589	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
590	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
591	return (SIGPENDING(td) ? issignal(td) : 0);
592}
593
594/*
595 * Arrange for ast() to handle unmasked pending signals on return to user
596 * mode.  This must be called whenever a signal is added to td_sigqueue or
597 * unmasked in td_sigmask.
598 */
599void
600signotify(struct thread *td)
601{
602	struct proc *p;
603
604	p = td->td_proc;
605
606	PROC_LOCK_ASSERT(p, MA_OWNED);
607
608	if (SIGPENDING(td)) {
609		thread_lock(td);
610		td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING;
611		thread_unlock(td);
612	}
613}
614
615int
616sigonstack(size_t sp)
617{
618	struct thread *td = curthread;
619
620	return ((td->td_pflags & TDP_ALTSTACK) ?
621#if defined(COMPAT_43)
622	    ((td->td_sigstk.ss_size == 0) ?
623		(td->td_sigstk.ss_flags & SS_ONSTACK) :
624		((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size))
625#else
626	    ((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size)
627#endif
628	    : 0);
629}
630
631static __inline int
632sigprop(int sig)
633{
634
635	if (sig > 0 && sig < NSIG)
636		return (sigproptbl[_SIG_IDX(sig)]);
637	return (0);
638}
639
640int
641sig_ffs(sigset_t *set)
642{
643	int i;
644
645	for (i = 0; i < _SIG_WORDS; i++)
646		if (set->__bits[i])
647			return (ffs(set->__bits[i]) + (i * 32));
648	return (0);
649}
650
651static bool
652sigact_flag_test(struct sigaction *act, int flag)
653{
654
655	/*
656	 * SA_SIGINFO is reset when signal disposition is set to
657	 * ignore or default.  Other flags are kept according to user
658	 * settings.
659	 */
660	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
661	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
662	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
663}
664
665/*
666 * kern_sigaction
667 * sigaction
668 * freebsd4_sigaction
669 * osigaction
670 */
671int
672kern_sigaction(td, sig, act, oact, flags)
673	struct thread *td;
674	register int sig;
675	struct sigaction *act, *oact;
676	int flags;
677{
678	struct sigacts *ps;
679	struct proc *p = td->td_proc;
680
681	if (!_SIG_VALID(sig))
682		return (EINVAL);
683	if (act != NULL && act->sa_handler != SIG_DFL &&
684	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
685	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
686	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
687		return (EINVAL);
688
689	PROC_LOCK(p);
690	ps = p->p_sigacts;
691	mtx_lock(&ps->ps_mtx);
692	if (oact) {
693		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
694		oact->sa_flags = 0;
695		if (SIGISMEMBER(ps->ps_sigonstack, sig))
696			oact->sa_flags |= SA_ONSTACK;
697		if (!SIGISMEMBER(ps->ps_sigintr, sig))
698			oact->sa_flags |= SA_RESTART;
699		if (SIGISMEMBER(ps->ps_sigreset, sig))
700			oact->sa_flags |= SA_RESETHAND;
701		if (SIGISMEMBER(ps->ps_signodefer, sig))
702			oact->sa_flags |= SA_NODEFER;
703		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
704			oact->sa_flags |= SA_SIGINFO;
705			oact->sa_sigaction =
706			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
707		} else
708			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
709		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
710			oact->sa_flags |= SA_NOCLDSTOP;
711		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
712			oact->sa_flags |= SA_NOCLDWAIT;
713	}
714	if (act) {
715		if ((sig == SIGKILL || sig == SIGSTOP) &&
716		    act->sa_handler != SIG_DFL) {
717			mtx_unlock(&ps->ps_mtx);
718			PROC_UNLOCK(p);
719			return (EINVAL);
720		}
721
722		/*
723		 * Change setting atomically.
724		 */
725
726		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
727		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
728		if (sigact_flag_test(act, SA_SIGINFO)) {
729			ps->ps_sigact[_SIG_IDX(sig)] =
730			    (__sighandler_t *)act->sa_sigaction;
731			SIGADDSET(ps->ps_siginfo, sig);
732		} else {
733			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
734			SIGDELSET(ps->ps_siginfo, sig);
735		}
736		if (!sigact_flag_test(act, SA_RESTART))
737			SIGADDSET(ps->ps_sigintr, sig);
738		else
739			SIGDELSET(ps->ps_sigintr, sig);
740		if (sigact_flag_test(act, SA_ONSTACK))
741			SIGADDSET(ps->ps_sigonstack, sig);
742		else
743			SIGDELSET(ps->ps_sigonstack, sig);
744		if (sigact_flag_test(act, SA_RESETHAND))
745			SIGADDSET(ps->ps_sigreset, sig);
746		else
747			SIGDELSET(ps->ps_sigreset, sig);
748		if (sigact_flag_test(act, SA_NODEFER))
749			SIGADDSET(ps->ps_signodefer, sig);
750		else
751			SIGDELSET(ps->ps_signodefer, sig);
752		if (sig == SIGCHLD) {
753			if (act->sa_flags & SA_NOCLDSTOP)
754				ps->ps_flag |= PS_NOCLDSTOP;
755			else
756				ps->ps_flag &= ~PS_NOCLDSTOP;
757			if (act->sa_flags & SA_NOCLDWAIT) {
758				/*
759				 * Paranoia: since SA_NOCLDWAIT is implemented
760				 * by reparenting the dying child to PID 1 (and
761				 * trust it to reap the zombie), PID 1 itself
762				 * is forbidden to set SA_NOCLDWAIT.
763				 */
764				if (p->p_pid == 1)
765					ps->ps_flag &= ~PS_NOCLDWAIT;
766				else
767					ps->ps_flag |= PS_NOCLDWAIT;
768			} else
769				ps->ps_flag &= ~PS_NOCLDWAIT;
770			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
771				ps->ps_flag |= PS_CLDSIGIGN;
772			else
773				ps->ps_flag &= ~PS_CLDSIGIGN;
774		}
775		/*
776		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
777		 * and for signals set to SIG_DFL where the default is to
778		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
779		 * have to restart the process.
780		 */
781		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
782		    (sigprop(sig) & SA_IGNORE &&
783		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
784			/* never to be seen again */
785			sigqueue_delete_proc(p, sig);
786			if (sig != SIGCONT)
787				/* easier in psignal */
788				SIGADDSET(ps->ps_sigignore, sig);
789			SIGDELSET(ps->ps_sigcatch, sig);
790		} else {
791			SIGDELSET(ps->ps_sigignore, sig);
792			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
793				SIGDELSET(ps->ps_sigcatch, sig);
794			else
795				SIGADDSET(ps->ps_sigcatch, sig);
796		}
797#ifdef COMPAT_FREEBSD4
798		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
799		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
800		    (flags & KSA_FREEBSD4) == 0)
801			SIGDELSET(ps->ps_freebsd4, sig);
802		else
803			SIGADDSET(ps->ps_freebsd4, sig);
804#endif
805#ifdef COMPAT_43
806		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
807		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
808		    (flags & KSA_OSIGSET) == 0)
809			SIGDELSET(ps->ps_osigset, sig);
810		else
811			SIGADDSET(ps->ps_osigset, sig);
812#endif
813	}
814	mtx_unlock(&ps->ps_mtx);
815	PROC_UNLOCK(p);
816	return (0);
817}
818
819#ifndef _SYS_SYSPROTO_H_
820struct sigaction_args {
821	int	sig;
822	struct	sigaction *act;
823	struct	sigaction *oact;
824};
825#endif
826int
827sys_sigaction(td, uap)
828	struct thread *td;
829	register struct sigaction_args *uap;
830{
831	struct sigaction act, oact;
832	register struct sigaction *actp, *oactp;
833	int error;
834
835	actp = (uap->act != NULL) ? &act : NULL;
836	oactp = (uap->oact != NULL) ? &oact : NULL;
837	if (actp) {
838		error = copyin(uap->act, actp, sizeof(act));
839		if (error)
840			return (error);
841	}
842	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
843	if (oactp && !error)
844		error = copyout(oactp, uap->oact, sizeof(oact));
845	return (error);
846}
847
848#ifdef COMPAT_FREEBSD4
849#ifndef _SYS_SYSPROTO_H_
850struct freebsd4_sigaction_args {
851	int	sig;
852	struct	sigaction *act;
853	struct	sigaction *oact;
854};
855#endif
856int
857freebsd4_sigaction(td, uap)
858	struct thread *td;
859	register struct freebsd4_sigaction_args *uap;
860{
861	struct sigaction act, oact;
862	register struct sigaction *actp, *oactp;
863	int error;
864
865
866	actp = (uap->act != NULL) ? &act : NULL;
867	oactp = (uap->oact != NULL) ? &oact : NULL;
868	if (actp) {
869		error = copyin(uap->act, actp, sizeof(act));
870		if (error)
871			return (error);
872	}
873	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
874	if (oactp && !error)
875		error = copyout(oactp, uap->oact, sizeof(oact));
876	return (error);
877}
878#endif	/* COMAPT_FREEBSD4 */
879
880#ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
881#ifndef _SYS_SYSPROTO_H_
882struct osigaction_args {
883	int	signum;
884	struct	osigaction *nsa;
885	struct	osigaction *osa;
886};
887#endif
888int
889osigaction(td, uap)
890	struct thread *td;
891	register struct osigaction_args *uap;
892{
893	struct osigaction sa;
894	struct sigaction nsa, osa;
895	register struct sigaction *nsap, *osap;
896	int error;
897
898	if (uap->signum <= 0 || uap->signum >= ONSIG)
899		return (EINVAL);
900
901	nsap = (uap->nsa != NULL) ? &nsa : NULL;
902	osap = (uap->osa != NULL) ? &osa : NULL;
903
904	if (nsap) {
905		error = copyin(uap->nsa, &sa, sizeof(sa));
906		if (error)
907			return (error);
908		nsap->sa_handler = sa.sa_handler;
909		nsap->sa_flags = sa.sa_flags;
910		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
911	}
912	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
913	if (osap && !error) {
914		sa.sa_handler = osap->sa_handler;
915		sa.sa_flags = osap->sa_flags;
916		SIG2OSIG(osap->sa_mask, sa.sa_mask);
917		error = copyout(&sa, uap->osa, sizeof(sa));
918	}
919	return (error);
920}
921
922#if !defined(__i386__)
923/* Avoid replicating the same stub everywhere */
924int
925osigreturn(td, uap)
926	struct thread *td;
927	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(p)
941	struct proc *p;
942{
943	register int i;
944	struct sigacts *ps;
945
946	PROC_LOCK(p);
947	ps = p->p_sigacts;
948	mtx_lock(&ps->ps_mtx);
949	for (i = 1; i <= NSIG; i++) {
950		if (sigprop(i) & SA_IGNORE && i != SIGCONT) {
951			SIGADDSET(ps->ps_sigignore, i);
952		}
953	}
954	mtx_unlock(&ps->ps_mtx);
955	PROC_UNLOCK(p);
956}
957
958/*
959 * Reset specified signal to the default disposition.
960 */
961static void
962sigdflt(struct sigacts *ps, int sig)
963{
964
965	mtx_assert(&ps->ps_mtx, MA_OWNED);
966	SIGDELSET(ps->ps_sigcatch, sig);
967	if ((sigprop(sig) & SA_IGNORE) != 0 && sig != SIGCONT)
968		SIGADDSET(ps->ps_sigignore, sig);
969	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
970	SIGDELSET(ps->ps_siginfo, sig);
971}
972
973/*
974 * Reset signals for an exec of the specified process.
975 */
976void
977execsigs(struct proc *p)
978{
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	while (SIGNOTEMPTY(ps->ps_sigcatch)) {
992		sig = sig_ffs(&ps->ps_sigcatch);
993		sigdflt(ps, sig);
994		if ((sigprop(sig) & SA_IGNORE) != 0)
995			sigqueue_delete_proc(p, sig);
996	}
997	/*
998	 * Reset stack state to the user stack.
999	 * Clear set of signals caught on the signal stack.
1000	 */
1001	td = curthread;
1002	MPASS(td->td_proc == p);
1003	td->td_sigstk.ss_flags = SS_DISABLE;
1004	td->td_sigstk.ss_size = 0;
1005	td->td_sigstk.ss_sp = 0;
1006	td->td_pflags &= ~TDP_ALTSTACK;
1007	/*
1008	 * Reset no zombies if child dies flag as Solaris does.
1009	 */
1010	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1011	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1012		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1013	mtx_unlock(&ps->ps_mtx);
1014}
1015
1016/*
1017 * kern_sigprocmask()
1018 *
1019 *	Manipulate signal mask.
1020 */
1021int
1022kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1023    int flags)
1024{
1025	sigset_t new_block, oset1;
1026	struct proc *p;
1027	int error;
1028
1029	p = td->td_proc;
1030	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1031		PROC_LOCK_ASSERT(p, MA_OWNED);
1032	else
1033		PROC_LOCK(p);
1034	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1035	    ? MA_OWNED : MA_NOTOWNED);
1036	if (oset != NULL)
1037		*oset = td->td_sigmask;
1038
1039	error = 0;
1040	if (set != NULL) {
1041		switch (how) {
1042		case SIG_BLOCK:
1043			SIG_CANTMASK(*set);
1044			oset1 = td->td_sigmask;
1045			SIGSETOR(td->td_sigmask, *set);
1046			new_block = td->td_sigmask;
1047			SIGSETNAND(new_block, oset1);
1048			break;
1049		case SIG_UNBLOCK:
1050			SIGSETNAND(td->td_sigmask, *set);
1051			signotify(td);
1052			goto out;
1053		case SIG_SETMASK:
1054			SIG_CANTMASK(*set);
1055			oset1 = td->td_sigmask;
1056			if (flags & SIGPROCMASK_OLD)
1057				SIGSETLO(td->td_sigmask, *set);
1058			else
1059				td->td_sigmask = *set;
1060			new_block = td->td_sigmask;
1061			SIGSETNAND(new_block, oset1);
1062			signotify(td);
1063			break;
1064		default:
1065			error = EINVAL;
1066			goto out;
1067		}
1068
1069		/*
1070		 * The new_block set contains signals that were not previously
1071		 * blocked, but are blocked now.
1072		 *
1073		 * In case we block any signal that was not previously blocked
1074		 * for td, and process has the signal pending, try to schedule
1075		 * signal delivery to some thread that does not block the
1076		 * signal, possibly waking it up.
1077		 */
1078		if (p->p_numthreads != 1)
1079			reschedule_signals(p, new_block, flags);
1080	}
1081
1082out:
1083	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1084		PROC_UNLOCK(p);
1085	return (error);
1086}
1087
1088#ifndef _SYS_SYSPROTO_H_
1089struct sigprocmask_args {
1090	int	how;
1091	const sigset_t *set;
1092	sigset_t *oset;
1093};
1094#endif
1095int
1096sys_sigprocmask(td, uap)
1097	register struct thread *td;
1098	struct sigprocmask_args *uap;
1099{
1100	sigset_t set, oset;
1101	sigset_t *setp, *osetp;
1102	int error;
1103
1104	setp = (uap->set != NULL) ? &set : NULL;
1105	osetp = (uap->oset != NULL) ? &oset : NULL;
1106	if (setp) {
1107		error = copyin(uap->set, setp, sizeof(set));
1108		if (error)
1109			return (error);
1110	}
1111	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1112	if (osetp && !error) {
1113		error = copyout(osetp, uap->oset, sizeof(oset));
1114	}
1115	return (error);
1116}
1117
1118#ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1119#ifndef _SYS_SYSPROTO_H_
1120struct osigprocmask_args {
1121	int	how;
1122	osigset_t mask;
1123};
1124#endif
1125int
1126osigprocmask(td, uap)
1127	register struct thread *td;
1128	struct osigprocmask_args *uap;
1129{
1130	sigset_t set, oset;
1131	int error;
1132
1133	OSIG2SIG(uap->mask, set);
1134	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1135	SIG2OSIG(oset, td->td_retval[0]);
1136	return (error);
1137}
1138#endif /* COMPAT_43 */
1139
1140int
1141sys_sigwait(struct thread *td, struct sigwait_args *uap)
1142{
1143	ksiginfo_t ksi;
1144	sigset_t set;
1145	int error;
1146
1147	error = copyin(uap->set, &set, sizeof(set));
1148	if (error) {
1149		td->td_retval[0] = error;
1150		return (0);
1151	}
1152
1153	error = kern_sigtimedwait(td, set, &ksi, NULL);
1154	if (error) {
1155		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1156			error = ERESTART;
1157		if (error == ERESTART)
1158			return (error);
1159		td->td_retval[0] = error;
1160		return (0);
1161	}
1162
1163	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1164	td->td_retval[0] = error;
1165	return (0);
1166}
1167
1168int
1169sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1170{
1171	struct timespec ts;
1172	struct timespec *timeout;
1173	sigset_t set;
1174	ksiginfo_t ksi;
1175	int error;
1176
1177	if (uap->timeout) {
1178		error = copyin(uap->timeout, &ts, sizeof(ts));
1179		if (error)
1180			return (error);
1181
1182		timeout = &ts;
1183	} else
1184		timeout = NULL;
1185
1186	error = copyin(uap->set, &set, sizeof(set));
1187	if (error)
1188		return (error);
1189
1190	error = kern_sigtimedwait(td, set, &ksi, timeout);
1191	if (error)
1192		return (error);
1193
1194	if (uap->info)
1195		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1196
1197	if (error == 0)
1198		td->td_retval[0] = ksi.ksi_signo;
1199	return (error);
1200}
1201
1202int
1203sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1204{
1205	ksiginfo_t ksi;
1206	sigset_t set;
1207	int error;
1208
1209	error = copyin(uap->set, &set, sizeof(set));
1210	if (error)
1211		return (error);
1212
1213	error = kern_sigtimedwait(td, set, &ksi, NULL);
1214	if (error)
1215		return (error);
1216
1217	if (uap->info)
1218		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1219
1220	if (error == 0)
1221		td->td_retval[0] = ksi.ksi_signo;
1222	return (error);
1223}
1224
1225int
1226kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1227	struct timespec *timeout)
1228{
1229	struct sigacts *ps;
1230	sigset_t saved_mask, new_block;
1231	struct proc *p;
1232	int error, sig, timo, timevalid = 0;
1233	struct timespec rts, ets, ts;
1234	struct timeval tv;
1235
1236	p = td->td_proc;
1237	error = 0;
1238	ets.tv_sec = 0;
1239	ets.tv_nsec = 0;
1240
1241	if (timeout != NULL) {
1242		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1243			timevalid = 1;
1244			getnanouptime(&rts);
1245			ets = rts;
1246			timespecadd(&ets, timeout);
1247		}
1248	}
1249	ksiginfo_init(ksi);
1250	/* Some signals can not be waited for. */
1251	SIG_CANTMASK(waitset);
1252	ps = p->p_sigacts;
1253	PROC_LOCK(p);
1254	saved_mask = td->td_sigmask;
1255	SIGSETNAND(td->td_sigmask, waitset);
1256	for (;;) {
1257		mtx_lock(&ps->ps_mtx);
1258		sig = cursig(td);
1259		mtx_unlock(&ps->ps_mtx);
1260		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1261			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1262			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1263				error = 0;
1264				break;
1265			}
1266		}
1267
1268		if (error != 0)
1269			break;
1270
1271		/*
1272		 * POSIX says this must be checked after looking for pending
1273		 * signals.
1274		 */
1275		if (timeout != NULL) {
1276			if (!timevalid) {
1277				error = EINVAL;
1278				break;
1279			}
1280			getnanouptime(&rts);
1281			if (timespeccmp(&rts, &ets, >=)) {
1282				error = EAGAIN;
1283				break;
1284			}
1285			ts = ets;
1286			timespecsub(&ts, &rts);
1287			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1288			timo = tvtohz(&tv);
1289		} else {
1290			timo = 0;
1291		}
1292
1293		error = msleep(ps, &p->p_mtx, PPAUSE|PCATCH, "sigwait", timo);
1294
1295		if (timeout != NULL) {
1296			if (error == ERESTART) {
1297				/* Timeout can not be restarted. */
1298				error = EINTR;
1299			} else if (error == EAGAIN) {
1300				/* We will calculate timeout by ourself. */
1301				error = 0;
1302			}
1303		}
1304	}
1305
1306	new_block = saved_mask;
1307	SIGSETNAND(new_block, td->td_sigmask);
1308	td->td_sigmask = saved_mask;
1309	/*
1310	 * Fewer signals can be delivered to us, reschedule signal
1311	 * notification.
1312	 */
1313	if (p->p_numthreads != 1)
1314		reschedule_signals(p, new_block, 0);
1315
1316	if (error == 0) {
1317		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1318
1319		if (ksi->ksi_code == SI_TIMER)
1320			itimer_accept(p, ksi->ksi_timerid, ksi);
1321
1322#ifdef KTRACE
1323		if (KTRPOINT(td, KTR_PSIG)) {
1324			sig_t action;
1325
1326			mtx_lock(&ps->ps_mtx);
1327			action = ps->ps_sigact[_SIG_IDX(sig)];
1328			mtx_unlock(&ps->ps_mtx);
1329			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1330		}
1331#endif
1332		if (sig == SIGKILL)
1333			sigexit(td, sig);
1334	}
1335	PROC_UNLOCK(p);
1336	return (error);
1337}
1338
1339#ifndef _SYS_SYSPROTO_H_
1340struct sigpending_args {
1341	sigset_t	*set;
1342};
1343#endif
1344int
1345sys_sigpending(td, uap)
1346	struct thread *td;
1347	struct sigpending_args *uap;
1348{
1349	struct proc *p = td->td_proc;
1350	sigset_t pending;
1351
1352	PROC_LOCK(p);
1353	pending = p->p_sigqueue.sq_signals;
1354	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1355	PROC_UNLOCK(p);
1356	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1357}
1358
1359#ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1360#ifndef _SYS_SYSPROTO_H_
1361struct osigpending_args {
1362	int	dummy;
1363};
1364#endif
1365int
1366osigpending(td, uap)
1367	struct thread *td;
1368	struct osigpending_args *uap;
1369{
1370	struct proc *p = td->td_proc;
1371	sigset_t pending;
1372
1373	PROC_LOCK(p);
1374	pending = p->p_sigqueue.sq_signals;
1375	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1376	PROC_UNLOCK(p);
1377	SIG2OSIG(pending, td->td_retval[0]);
1378	return (0);
1379}
1380#endif /* COMPAT_43 */
1381
1382#if defined(COMPAT_43)
1383/*
1384 * Generalized interface signal handler, 4.3-compatible.
1385 */
1386#ifndef _SYS_SYSPROTO_H_
1387struct osigvec_args {
1388	int	signum;
1389	struct	sigvec *nsv;
1390	struct	sigvec *osv;
1391};
1392#endif
1393/* ARGSUSED */
1394int
1395osigvec(td, uap)
1396	struct thread *td;
1397	register struct osigvec_args *uap;
1398{
1399	struct sigvec vec;
1400	struct sigaction nsa, osa;
1401	register struct sigaction *nsap, *osap;
1402	int error;
1403
1404	if (uap->signum <= 0 || uap->signum >= ONSIG)
1405		return (EINVAL);
1406	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1407	osap = (uap->osv != NULL) ? &osa : NULL;
1408	if (nsap) {
1409		error = copyin(uap->nsv, &vec, sizeof(vec));
1410		if (error)
1411			return (error);
1412		nsap->sa_handler = vec.sv_handler;
1413		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1414		nsap->sa_flags = vec.sv_flags;
1415		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1416	}
1417	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1418	if (osap && !error) {
1419		vec.sv_handler = osap->sa_handler;
1420		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1421		vec.sv_flags = osap->sa_flags;
1422		vec.sv_flags &= ~SA_NOCLDWAIT;
1423		vec.sv_flags ^= SA_RESTART;
1424		error = copyout(&vec, uap->osv, sizeof(vec));
1425	}
1426	return (error);
1427}
1428
1429#ifndef _SYS_SYSPROTO_H_
1430struct osigblock_args {
1431	int	mask;
1432};
1433#endif
1434int
1435osigblock(td, uap)
1436	register struct thread *td;
1437	struct osigblock_args *uap;
1438{
1439	sigset_t set, oset;
1440
1441	OSIG2SIG(uap->mask, set);
1442	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1443	SIG2OSIG(oset, td->td_retval[0]);
1444	return (0);
1445}
1446
1447#ifndef _SYS_SYSPROTO_H_
1448struct osigsetmask_args {
1449	int	mask;
1450};
1451#endif
1452int
1453osigsetmask(td, uap)
1454	struct thread *td;
1455	struct osigsetmask_args *uap;
1456{
1457	sigset_t set, oset;
1458
1459	OSIG2SIG(uap->mask, set);
1460	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1461	SIG2OSIG(oset, td->td_retval[0]);
1462	return (0);
1463}
1464#endif /* COMPAT_43 */
1465
1466/*
1467 * Suspend calling thread until signal, providing mask to be set in the
1468 * meantime.
1469 */
1470#ifndef _SYS_SYSPROTO_H_
1471struct sigsuspend_args {
1472	const sigset_t *sigmask;
1473};
1474#endif
1475/* ARGSUSED */
1476int
1477sys_sigsuspend(td, uap)
1478	struct thread *td;
1479	struct sigsuspend_args *uap;
1480{
1481	sigset_t mask;
1482	int error;
1483
1484	error = copyin(uap->sigmask, &mask, sizeof(mask));
1485	if (error)
1486		return (error);
1487	return (kern_sigsuspend(td, mask));
1488}
1489
1490int
1491kern_sigsuspend(struct thread *td, sigset_t mask)
1492{
1493	struct proc *p = td->td_proc;
1494	int has_sig, sig;
1495
1496	/*
1497	 * When returning from sigsuspend, we want
1498	 * the old mask to be restored after the
1499	 * signal handler has finished.  Thus, we
1500	 * save it here and mark the sigacts structure
1501	 * to indicate this.
1502	 */
1503	PROC_LOCK(p);
1504	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1505	    SIGPROCMASK_PROC_LOCKED);
1506	td->td_pflags |= TDP_OLDMASK;
1507
1508	/*
1509	 * Process signals now. Otherwise, we can get spurious wakeup
1510	 * due to signal entered process queue, but delivered to other
1511	 * thread. But sigsuspend should return only on signal
1512	 * delivery.
1513	 */
1514	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1515	for (has_sig = 0; !has_sig;) {
1516		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1517			0) == 0)
1518			/* void */;
1519		thread_suspend_check(0);
1520		mtx_lock(&p->p_sigacts->ps_mtx);
1521		while ((sig = cursig(td)) != 0)
1522			has_sig += postsig(sig);
1523		mtx_unlock(&p->p_sigacts->ps_mtx);
1524	}
1525	PROC_UNLOCK(p);
1526	td->td_errno = EINTR;
1527	td->td_pflags |= TDP_NERRNO;
1528	return (EJUSTRETURN);
1529}
1530
1531#ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1532/*
1533 * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1534 * convention: libc stub passes mask, not pointer, to save a copyin.
1535 */
1536#ifndef _SYS_SYSPROTO_H_
1537struct osigsuspend_args {
1538	osigset_t mask;
1539};
1540#endif
1541/* ARGSUSED */
1542int
1543osigsuspend(td, uap)
1544	struct thread *td;
1545	struct osigsuspend_args *uap;
1546{
1547	sigset_t mask;
1548
1549	OSIG2SIG(uap->mask, mask);
1550	return (kern_sigsuspend(td, mask));
1551}
1552#endif /* COMPAT_43 */
1553
1554#if defined(COMPAT_43)
1555#ifndef _SYS_SYSPROTO_H_
1556struct osigstack_args {
1557	struct	sigstack *nss;
1558	struct	sigstack *oss;
1559};
1560#endif
1561/* ARGSUSED */
1562int
1563osigstack(td, uap)
1564	struct thread *td;
1565	register struct osigstack_args *uap;
1566{
1567	struct sigstack nss, oss;
1568	int error = 0;
1569
1570	if (uap->nss != NULL) {
1571		error = copyin(uap->nss, &nss, sizeof(nss));
1572		if (error)
1573			return (error);
1574	}
1575	oss.ss_sp = td->td_sigstk.ss_sp;
1576	oss.ss_onstack = sigonstack(cpu_getstack(td));
1577	if (uap->nss != NULL) {
1578		td->td_sigstk.ss_sp = nss.ss_sp;
1579		td->td_sigstk.ss_size = 0;
1580		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1581		td->td_pflags |= TDP_ALTSTACK;
1582	}
1583	if (uap->oss != NULL)
1584		error = copyout(&oss, uap->oss, sizeof(oss));
1585
1586	return (error);
1587}
1588#endif /* COMPAT_43 */
1589
1590#ifndef _SYS_SYSPROTO_H_
1591struct sigaltstack_args {
1592	stack_t	*ss;
1593	stack_t	*oss;
1594};
1595#endif
1596/* ARGSUSED */
1597int
1598sys_sigaltstack(td, uap)
1599	struct thread *td;
1600	register struct sigaltstack_args *uap;
1601{
1602	stack_t ss, oss;
1603	int error;
1604
1605	if (uap->ss != NULL) {
1606		error = copyin(uap->ss, &ss, sizeof(ss));
1607		if (error)
1608			return (error);
1609	}
1610	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1611	    (uap->oss != NULL) ? &oss : NULL);
1612	if (error)
1613		return (error);
1614	if (uap->oss != NULL)
1615		error = copyout(&oss, uap->oss, sizeof(stack_t));
1616	return (error);
1617}
1618
1619int
1620kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1621{
1622	struct proc *p = td->td_proc;
1623	int oonstack;
1624
1625	oonstack = sigonstack(cpu_getstack(td));
1626
1627	if (oss != NULL) {
1628		*oss = td->td_sigstk;
1629		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1630		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1631	}
1632
1633	if (ss != NULL) {
1634		if (oonstack)
1635			return (EPERM);
1636		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1637			return (EINVAL);
1638		if (!(ss->ss_flags & SS_DISABLE)) {
1639			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1640				return (ENOMEM);
1641
1642			td->td_sigstk = *ss;
1643			td->td_pflags |= TDP_ALTSTACK;
1644		} else {
1645			td->td_pflags &= ~TDP_ALTSTACK;
1646		}
1647	}
1648	return (0);
1649}
1650
1651/*
1652 * Common code for kill process group/broadcast kill.
1653 * cp is calling process.
1654 */
1655static int
1656killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1657{
1658	struct proc *p;
1659	struct pgrp *pgrp;
1660	int err;
1661	int ret;
1662
1663	ret = ESRCH;
1664	if (all) {
1665		/*
1666		 * broadcast
1667		 */
1668		sx_slock(&allproc_lock);
1669		FOREACH_PROC_IN_SYSTEM(p) {
1670			PROC_LOCK(p);
1671			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1672			    p == td->td_proc || p->p_state == PRS_NEW) {
1673				PROC_UNLOCK(p);
1674				continue;
1675			}
1676			err = p_cansignal(td, p, sig);
1677			if (err == 0) {
1678				if (sig)
1679					pksignal(p, sig, ksi);
1680				ret = err;
1681			}
1682			else if (ret == ESRCH)
1683				ret = err;
1684			PROC_UNLOCK(p);
1685		}
1686		sx_sunlock(&allproc_lock);
1687	} else {
1688		sx_slock(&proctree_lock);
1689		if (pgid == 0) {
1690			/*
1691			 * zero pgid means send to my process group.
1692			 */
1693			pgrp = td->td_proc->p_pgrp;
1694			PGRP_LOCK(pgrp);
1695		} else {
1696			pgrp = pgfind(pgid);
1697			if (pgrp == NULL) {
1698				sx_sunlock(&proctree_lock);
1699				return (ESRCH);
1700			}
1701		}
1702		sx_sunlock(&proctree_lock);
1703		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1704			PROC_LOCK(p);
1705			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1706			    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		PGRP_UNLOCK(pgrp);
1721	}
1722	return (ret);
1723}
1724
1725#ifndef _SYS_SYSPROTO_H_
1726struct kill_args {
1727	int	pid;
1728	int	signum;
1729};
1730#endif
1731/* ARGSUSED */
1732int
1733sys_kill(struct thread *td, struct kill_args *uap)
1734{
1735	ksiginfo_t ksi;
1736	struct proc *p;
1737	int error;
1738
1739	/*
1740	 * A process in capability mode can send signals only to himself.
1741	 * The main rationale behind this is that abort(3) is implemented as
1742	 * kill(getpid(), SIGABRT).
1743	 */
1744	if (IN_CAPABILITY_MODE(td) && uap->pid != td->td_proc->p_pid)
1745		return (ECAPMODE);
1746
1747	AUDIT_ARG_SIGNUM(uap->signum);
1748	AUDIT_ARG_PID(uap->pid);
1749	if ((u_int)uap->signum > _SIG_MAXSIG)
1750		return (EINVAL);
1751
1752	ksiginfo_init(&ksi);
1753	ksi.ksi_signo = uap->signum;
1754	ksi.ksi_code = SI_USER;
1755	ksi.ksi_pid = td->td_proc->p_pid;
1756	ksi.ksi_uid = td->td_ucred->cr_ruid;
1757
1758	if (uap->pid > 0) {
1759		/* kill single process */
1760		if ((p = pfind(uap->pid)) == NULL) {
1761			if ((p = zpfind(uap->pid)) == NULL)
1762				return (ESRCH);
1763		}
1764		AUDIT_ARG_PROCESS(p);
1765		error = p_cansignal(td, p, uap->signum);
1766		if (error == 0 && uap->signum)
1767			pksignal(p, uap->signum, &ksi);
1768		PROC_UNLOCK(p);
1769		return (error);
1770	}
1771	switch (uap->pid) {
1772	case -1:		/* broadcast signal */
1773		return (killpg1(td, uap->signum, 0, 1, &ksi));
1774	case 0:			/* signal own process group */
1775		return (killpg1(td, uap->signum, 0, 0, &ksi));
1776	default:		/* negative explicit process group */
1777		return (killpg1(td, uap->signum, -uap->pid, 0, &ksi));
1778	}
1779	/* NOTREACHED */
1780}
1781
1782int
1783sys_pdkill(td, uap)
1784	struct thread *td;
1785	struct pdkill_args *uap;
1786{
1787#ifdef PROCDESC
1788	struct proc *p;
1789	cap_rights_t rights;
1790	int error;
1791
1792	AUDIT_ARG_SIGNUM(uap->signum);
1793	AUDIT_ARG_FD(uap->fd);
1794	if ((u_int)uap->signum > _SIG_MAXSIG)
1795		return (EINVAL);
1796
1797	error = procdesc_find(td, uap->fd,
1798	    cap_rights_init(&rights, CAP_PDKILL), &p);
1799	if (error)
1800		return (error);
1801	AUDIT_ARG_PROCESS(p);
1802	error = p_cansignal(td, p, uap->signum);
1803	if (error == 0 && uap->signum)
1804		kern_psignal(p, uap->signum);
1805	PROC_UNLOCK(p);
1806	return (error);
1807#else
1808	return (ENOSYS);
1809#endif
1810}
1811
1812#if defined(COMPAT_43)
1813#ifndef _SYS_SYSPROTO_H_
1814struct okillpg_args {
1815	int	pgid;
1816	int	signum;
1817};
1818#endif
1819/* ARGSUSED */
1820int
1821okillpg(struct thread *td, struct okillpg_args *uap)
1822{
1823	ksiginfo_t ksi;
1824
1825	AUDIT_ARG_SIGNUM(uap->signum);
1826	AUDIT_ARG_PID(uap->pgid);
1827	if ((u_int)uap->signum > _SIG_MAXSIG)
1828		return (EINVAL);
1829
1830	ksiginfo_init(&ksi);
1831	ksi.ksi_signo = uap->signum;
1832	ksi.ksi_code = SI_USER;
1833	ksi.ksi_pid = td->td_proc->p_pid;
1834	ksi.ksi_uid = td->td_ucred->cr_ruid;
1835	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1836}
1837#endif /* COMPAT_43 */
1838
1839#ifndef _SYS_SYSPROTO_H_
1840struct sigqueue_args {
1841	pid_t pid;
1842	int signum;
1843	/* union sigval */ void *value;
1844};
1845#endif
1846int
1847sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1848{
1849	ksiginfo_t ksi;
1850	struct proc *p;
1851	int error;
1852
1853	if ((u_int)uap->signum > _SIG_MAXSIG)
1854		return (EINVAL);
1855
1856	/*
1857	 * Specification says sigqueue can only send signal to
1858	 * single process.
1859	 */
1860	if (uap->pid <= 0)
1861		return (EINVAL);
1862
1863	if ((p = pfind(uap->pid)) == NULL) {
1864		if ((p = zpfind(uap->pid)) == NULL)
1865			return (ESRCH);
1866	}
1867	error = p_cansignal(td, p, uap->signum);
1868	if (error == 0 && uap->signum != 0) {
1869		ksiginfo_init(&ksi);
1870		ksi.ksi_flags = KSI_SIGQ;
1871		ksi.ksi_signo = uap->signum;
1872		ksi.ksi_code = SI_QUEUE;
1873		ksi.ksi_pid = td->td_proc->p_pid;
1874		ksi.ksi_uid = td->td_ucred->cr_ruid;
1875		ksi.ksi_value.sival_ptr = uap->value;
1876		error = pksignal(p, ksi.ksi_signo, &ksi);
1877	}
1878	PROC_UNLOCK(p);
1879	return (error);
1880}
1881
1882/*
1883 * Send a signal to a process group.
1884 */
1885void
1886gsignal(int pgid, int sig, ksiginfo_t *ksi)
1887{
1888	struct pgrp *pgrp;
1889
1890	if (pgid != 0) {
1891		sx_slock(&proctree_lock);
1892		pgrp = pgfind(pgid);
1893		sx_sunlock(&proctree_lock);
1894		if (pgrp != NULL) {
1895			pgsignal(pgrp, sig, 0, ksi);
1896			PGRP_UNLOCK(pgrp);
1897		}
1898	}
1899}
1900
1901/*
1902 * Send a signal to a process group.  If checktty is 1,
1903 * limit to members which have a controlling terminal.
1904 */
1905void
1906pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
1907{
1908	struct proc *p;
1909
1910	if (pgrp) {
1911		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
1912		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1913			PROC_LOCK(p);
1914			if (p->p_state == PRS_NORMAL &&
1915			    (checkctty == 0 || p->p_flag & P_CONTROLT))
1916				pksignal(p, sig, ksi);
1917			PROC_UNLOCK(p);
1918		}
1919	}
1920}
1921
1922
1923/*
1924 * Recalculate the signal mask and reset the signal disposition after
1925 * usermode frame for delivery is formed.  Should be called after
1926 * mach-specific routine, because sysent->sv_sendsig() needs correct
1927 * ps_siginfo and signal mask.
1928 */
1929static void
1930postsig_done(int sig, struct thread *td, struct sigacts *ps)
1931{
1932	sigset_t mask;
1933
1934	mtx_assert(&ps->ps_mtx, MA_OWNED);
1935	td->td_ru.ru_nsignals++;
1936	mask = ps->ps_catchmask[_SIG_IDX(sig)];
1937	if (!SIGISMEMBER(ps->ps_signodefer, sig))
1938		SIGADDSET(mask, sig);
1939	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
1940	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
1941	if (SIGISMEMBER(ps->ps_sigreset, sig))
1942		sigdflt(ps, sig);
1943}
1944
1945
1946/*
1947 * Send a signal caused by a trap to the current thread.  If it will be
1948 * caught immediately, deliver it with correct code.  Otherwise, post it
1949 * normally.
1950 */
1951void
1952trapsignal(struct thread *td, ksiginfo_t *ksi)
1953{
1954	struct sigacts *ps;
1955	struct proc *p;
1956	int sig;
1957	int code;
1958
1959	p = td->td_proc;
1960	sig = ksi->ksi_signo;
1961	code = ksi->ksi_code;
1962	KASSERT(_SIG_VALID(sig), ("invalid signal"));
1963
1964	PROC_LOCK(p);
1965	ps = p->p_sigacts;
1966	mtx_lock(&ps->ps_mtx);
1967	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
1968	    !SIGISMEMBER(td->td_sigmask, sig)) {
1969#ifdef KTRACE
1970		if (KTRPOINT(curthread, KTR_PSIG))
1971			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
1972			    &td->td_sigmask, code);
1973#endif
1974		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
1975				ksi, &td->td_sigmask);
1976		postsig_done(sig, td, ps);
1977		mtx_unlock(&ps->ps_mtx);
1978	} else {
1979		/*
1980		 * Avoid a possible infinite loop if the thread
1981		 * masking the signal or process is ignoring the
1982		 * signal.
1983		 */
1984		if (kern_forcesigexit &&
1985		    (SIGISMEMBER(td->td_sigmask, sig) ||
1986		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
1987			SIGDELSET(td->td_sigmask, sig);
1988			SIGDELSET(ps->ps_sigcatch, sig);
1989			SIGDELSET(ps->ps_sigignore, sig);
1990			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1991		}
1992		mtx_unlock(&ps->ps_mtx);
1993		p->p_code = code;	/* XXX for core dump/debugger */
1994		p->p_sig = sig;		/* XXX to verify code */
1995		tdsendsignal(p, td, sig, ksi);
1996	}
1997	PROC_UNLOCK(p);
1998}
1999
2000static struct thread *
2001sigtd(struct proc *p, int sig, int prop)
2002{
2003	struct thread *td, *signal_td;
2004
2005	PROC_LOCK_ASSERT(p, MA_OWNED);
2006
2007	/*
2008	 * Check if current thread can handle the signal without
2009	 * switching context to another thread.
2010	 */
2011	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig))
2012		return (curthread);
2013	signal_td = NULL;
2014	FOREACH_THREAD_IN_PROC(p, td) {
2015		if (!SIGISMEMBER(td->td_sigmask, sig)) {
2016			signal_td = td;
2017			break;
2018		}
2019	}
2020	if (signal_td == NULL)
2021		signal_td = FIRST_THREAD_IN_PROC(p);
2022	return (signal_td);
2023}
2024
2025/*
2026 * Send the signal to the process.  If the signal has an action, the action
2027 * is usually performed by the target process rather than the caller; we add
2028 * the signal to the set of pending signals for the process.
2029 *
2030 * Exceptions:
2031 *   o When a stop signal is sent to a sleeping process that takes the
2032 *     default action, the process is stopped without awakening it.
2033 *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2034 *     regardless of the signal action (eg, blocked or ignored).
2035 *
2036 * Other ignored signals are discarded immediately.
2037 *
2038 * NB: This function may be entered from the debugger via the "kill" DDB
2039 * command.  There is little that can be done to mitigate the possibly messy
2040 * side effects of this unwise possibility.
2041 */
2042void
2043kern_psignal(struct proc *p, int sig)
2044{
2045	ksiginfo_t ksi;
2046
2047	ksiginfo_init(&ksi);
2048	ksi.ksi_signo = sig;
2049	ksi.ksi_code = SI_KERNEL;
2050	(void) tdsendsignal(p, NULL, sig, &ksi);
2051}
2052
2053int
2054pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2055{
2056
2057	return (tdsendsignal(p, NULL, sig, ksi));
2058}
2059
2060/* Utility function for finding a thread to send signal event to. */
2061int
2062sigev_findtd(struct proc *p ,struct sigevent *sigev, struct thread **ttd)
2063{
2064	struct thread *td;
2065
2066	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2067		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2068		if (td == NULL)
2069			return (ESRCH);
2070		*ttd = td;
2071	} else {
2072		*ttd = NULL;
2073		PROC_LOCK(p);
2074	}
2075	return (0);
2076}
2077
2078void
2079tdsignal(struct thread *td, int sig)
2080{
2081	ksiginfo_t ksi;
2082
2083	ksiginfo_init(&ksi);
2084	ksi.ksi_signo = sig;
2085	ksi.ksi_code = SI_KERNEL;
2086	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2087}
2088
2089void
2090tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2091{
2092
2093	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2094}
2095
2096int
2097tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2098{
2099	sig_t action;
2100	sigqueue_t *sigqueue;
2101	int prop;
2102	struct sigacts *ps;
2103	int intrval;
2104	int ret = 0;
2105	int wakeup_swapper;
2106
2107	MPASS(td == NULL || p == td->td_proc);
2108	PROC_LOCK_ASSERT(p, MA_OWNED);
2109
2110	if (!_SIG_VALID(sig))
2111		panic("%s(): invalid signal %d", __func__, sig);
2112
2113	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2114
2115	/*
2116	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2117	 */
2118	if (p->p_state == PRS_ZOMBIE) {
2119		if (ksi && (ksi->ksi_flags & KSI_INS))
2120			ksiginfo_tryfree(ksi);
2121		return (ret);
2122	}
2123
2124	ps = p->p_sigacts;
2125	KNOTE_LOCKED(&p->p_klist, NOTE_SIGNAL | sig);
2126	prop = sigprop(sig);
2127
2128	if (td == NULL) {
2129		td = sigtd(p, sig, prop);
2130		sigqueue = &p->p_sigqueue;
2131	} else
2132		sigqueue = &td->td_sigqueue;
2133
2134	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2135
2136	/*
2137	 * If the signal is being ignored,
2138	 * then we forget about it immediately.
2139	 * (Note: we don't set SIGCONT in ps_sigignore,
2140	 * and if it is set to SIG_IGN,
2141	 * action will be SIG_DFL here.)
2142	 */
2143	mtx_lock(&ps->ps_mtx);
2144	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2145		SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2146
2147		mtx_unlock(&ps->ps_mtx);
2148		if (ksi && (ksi->ksi_flags & KSI_INS))
2149			ksiginfo_tryfree(ksi);
2150		return (ret);
2151	}
2152	if (SIGISMEMBER(td->td_sigmask, sig))
2153		action = SIG_HOLD;
2154	else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2155		action = SIG_CATCH;
2156	else
2157		action = SIG_DFL;
2158	if (SIGISMEMBER(ps->ps_sigintr, sig))
2159		intrval = EINTR;
2160	else
2161		intrval = ERESTART;
2162	mtx_unlock(&ps->ps_mtx);
2163
2164	if (prop & SA_CONT)
2165		sigqueue_delete_stopmask_proc(p);
2166	else if (prop & SA_STOP) {
2167		/*
2168		 * If sending a tty stop signal to a member of an orphaned
2169		 * process group, discard the signal here if the action
2170		 * is default; don't stop the process below if sleeping,
2171		 * and don't clear any pending SIGCONT.
2172		 */
2173		if ((prop & SA_TTYSTOP) &&
2174		    (p->p_pgrp->pg_jobc == 0) &&
2175		    (action == SIG_DFL)) {
2176			if (ksi && (ksi->ksi_flags & KSI_INS))
2177				ksiginfo_tryfree(ksi);
2178			return (ret);
2179		}
2180		sigqueue_delete_proc(p, SIGCONT);
2181		if (p->p_flag & P_CONTINUED) {
2182			p->p_flag &= ~P_CONTINUED;
2183			PROC_LOCK(p->p_pptr);
2184			sigqueue_take(p->p_ksi);
2185			PROC_UNLOCK(p->p_pptr);
2186		}
2187	}
2188
2189	ret = sigqueue_add(sigqueue, sig, ksi);
2190	if (ret != 0)
2191		return (ret);
2192	signotify(td);
2193	/*
2194	 * Defer further processing for signals which are held,
2195	 * except that stopped processes must be continued by SIGCONT.
2196	 */
2197	if (action == SIG_HOLD &&
2198	    !((prop & SA_CONT) && (p->p_flag & P_STOPPED_SIG)))
2199		return (ret);
2200
2201	/* SIGKILL: Remove procfs STOPEVENTs. */
2202	if (sig == SIGKILL) {
2203		/* from procfs_ioctl.c: PIOCBIC */
2204		p->p_stops = 0;
2205		/* from procfs_ioctl.c: PIOCCONT */
2206		p->p_step = 0;
2207		wakeup(&p->p_step);
2208	}
2209	/*
2210	 * Some signals have a process-wide effect and a per-thread
2211	 * component.  Most processing occurs when the process next
2212	 * tries to cross the user boundary, however there are some
2213	 * times when processing needs to be done immediately, such as
2214	 * waking up threads so that they can cross the user boundary.
2215	 * We try to do the per-process part here.
2216	 */
2217	if (P_SHOULDSTOP(p)) {
2218		KASSERT(!(p->p_flag & P_WEXIT),
2219		    ("signal to stopped but exiting process"));
2220		if (sig == SIGKILL) {
2221			/*
2222			 * If traced process is already stopped,
2223			 * then no further action is necessary.
2224			 */
2225			if (p->p_flag & P_TRACED)
2226				goto out;
2227			/*
2228			 * SIGKILL sets process running.
2229			 * It will die elsewhere.
2230			 * All threads must be restarted.
2231			 */
2232			p->p_flag &= ~P_STOPPED_SIG;
2233			goto runfast;
2234		}
2235
2236		if (prop & SA_CONT) {
2237			/*
2238			 * If traced process is already stopped,
2239			 * then no further action is necessary.
2240			 */
2241			if (p->p_flag & P_TRACED)
2242				goto out;
2243			/*
2244			 * If SIGCONT is default (or ignored), we continue the
2245			 * process but don't leave the signal in sigqueue as
2246			 * it has no further action.  If SIGCONT is held, we
2247			 * continue the process and leave the signal in
2248			 * sigqueue.  If the process catches SIGCONT, let it
2249			 * handle the signal itself.  If it isn't waiting on
2250			 * an event, it goes back to run state.
2251			 * Otherwise, process goes back to sleep state.
2252			 */
2253			p->p_flag &= ~P_STOPPED_SIG;
2254			PROC_SLOCK(p);
2255			if (p->p_numthreads == p->p_suspcount) {
2256				PROC_SUNLOCK(p);
2257				p->p_flag |= P_CONTINUED;
2258				p->p_xstat = SIGCONT;
2259				PROC_LOCK(p->p_pptr);
2260				childproc_continued(p);
2261				PROC_UNLOCK(p->p_pptr);
2262				PROC_SLOCK(p);
2263			}
2264			if (action == SIG_DFL) {
2265				thread_unsuspend(p);
2266				PROC_SUNLOCK(p);
2267				sigqueue_delete(sigqueue, sig);
2268				goto out;
2269			}
2270			if (action == SIG_CATCH) {
2271				/*
2272				 * The process wants to catch it so it needs
2273				 * to run at least one thread, but which one?
2274				 */
2275				PROC_SUNLOCK(p);
2276				goto runfast;
2277			}
2278			/*
2279			 * The signal is not ignored or caught.
2280			 */
2281			thread_unsuspend(p);
2282			PROC_SUNLOCK(p);
2283			goto out;
2284		}
2285
2286		if (prop & SA_STOP) {
2287			/*
2288			 * If traced process is already stopped,
2289			 * then no further action is necessary.
2290			 */
2291			if (p->p_flag & P_TRACED)
2292				goto out;
2293			/*
2294			 * Already stopped, don't need to stop again
2295			 * (If we did the shell could get confused).
2296			 * Just make sure the signal STOP bit set.
2297			 */
2298			p->p_flag |= P_STOPPED_SIG;
2299			sigqueue_delete(sigqueue, sig);
2300			goto out;
2301		}
2302
2303		/*
2304		 * All other kinds of signals:
2305		 * If a thread is sleeping interruptibly, simulate a
2306		 * wakeup so that when it is continued it will be made
2307		 * runnable and can look at the signal.  However, don't make
2308		 * the PROCESS runnable, leave it stopped.
2309		 * It may run a bit until it hits a thread_suspend_check().
2310		 */
2311		wakeup_swapper = 0;
2312		PROC_SLOCK(p);
2313		thread_lock(td);
2314		if (TD_ON_SLEEPQ(td) && (td->td_flags & TDF_SINTR))
2315			wakeup_swapper = sleepq_abort(td, intrval);
2316		thread_unlock(td);
2317		PROC_SUNLOCK(p);
2318		if (wakeup_swapper)
2319			kick_proc0();
2320		goto out;
2321		/*
2322		 * Mutexes are short lived. Threads waiting on them will
2323		 * hit thread_suspend_check() soon.
2324		 */
2325	} else if (p->p_state == PRS_NORMAL) {
2326		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2327			tdsigwakeup(td, sig, action, intrval);
2328			goto out;
2329		}
2330
2331		MPASS(action == SIG_DFL);
2332
2333		if (prop & SA_STOP) {
2334			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2335				goto out;
2336			p->p_flag |= P_STOPPED_SIG;
2337			p->p_xstat = sig;
2338			PROC_SLOCK(p);
2339			sig_suspend_threads(td, p, 1);
2340			if (p->p_numthreads == p->p_suspcount) {
2341				/*
2342				 * only thread sending signal to another
2343				 * process can reach here, if thread is sending
2344				 * signal to its process, because thread does
2345				 * not suspend itself here, p_numthreads
2346				 * should never be equal to p_suspcount.
2347				 */
2348				thread_stopped(p);
2349				PROC_SUNLOCK(p);
2350				sigqueue_delete_proc(p, p->p_xstat);
2351			} else
2352				PROC_SUNLOCK(p);
2353			goto out;
2354		}
2355	} else {
2356		/* Not in "NORMAL" state. discard the signal. */
2357		sigqueue_delete(sigqueue, sig);
2358		goto out;
2359	}
2360
2361	/*
2362	 * The process is not stopped so we need to apply the signal to all the
2363	 * running threads.
2364	 */
2365runfast:
2366	tdsigwakeup(td, sig, action, intrval);
2367	PROC_SLOCK(p);
2368	thread_unsuspend(p);
2369	PROC_SUNLOCK(p);
2370out:
2371	/* If we jump here, proc slock should not be owned. */
2372	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2373	return (ret);
2374}
2375
2376/*
2377 * The force of a signal has been directed against a single
2378 * thread.  We need to see what we can do about knocking it
2379 * out of any sleep it may be in etc.
2380 */
2381static void
2382tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2383{
2384	struct proc *p = td->td_proc;
2385	register int prop;
2386	int wakeup_swapper;
2387
2388	wakeup_swapper = 0;
2389	PROC_LOCK_ASSERT(p, MA_OWNED);
2390	prop = sigprop(sig);
2391
2392	PROC_SLOCK(p);
2393	thread_lock(td);
2394	/*
2395	 * Bring the priority of a thread up if we want it to get
2396	 * killed in this lifetime.  Be careful to avoid bumping the
2397	 * priority of the idle thread, since we still allow to signal
2398	 * kernel processes.
2399	 */
2400	if (action == SIG_DFL && (prop & SA_KILL) != 0 &&
2401	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2402		sched_prio(td, PUSER);
2403	if (TD_ON_SLEEPQ(td)) {
2404		/*
2405		 * If thread is sleeping uninterruptibly
2406		 * we can't interrupt the sleep... the signal will
2407		 * be noticed when the process returns through
2408		 * trap() or syscall().
2409		 */
2410		if ((td->td_flags & TDF_SINTR) == 0)
2411			goto out;
2412		/*
2413		 * If SIGCONT is default (or ignored) and process is
2414		 * asleep, we are finished; the process should not
2415		 * be awakened.
2416		 */
2417		if ((prop & SA_CONT) && action == SIG_DFL) {
2418			thread_unlock(td);
2419			PROC_SUNLOCK(p);
2420			sigqueue_delete(&p->p_sigqueue, sig);
2421			/*
2422			 * It may be on either list in this state.
2423			 * Remove from both for now.
2424			 */
2425			sigqueue_delete(&td->td_sigqueue, sig);
2426			return;
2427		}
2428
2429		/*
2430		 * Don't awaken a sleeping thread for SIGSTOP if the
2431		 * STOP signal is deferred.
2432		 */
2433		if ((prop & SA_STOP) && (td->td_flags & TDF_SBDRY))
2434			goto out;
2435
2436		/*
2437		 * Give low priority threads a better chance to run.
2438		 */
2439		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2440			sched_prio(td, PUSER);
2441
2442		wakeup_swapper = sleepq_abort(td, intrval);
2443	} else {
2444		/*
2445		 * Other states do nothing with the signal immediately,
2446		 * other than kicking ourselves if we are running.
2447		 * It will either never be noticed, or noticed very soon.
2448		 */
2449#ifdef SMP
2450		if (TD_IS_RUNNING(td) && td != curthread)
2451			forward_signal(td);
2452#endif
2453	}
2454out:
2455	PROC_SUNLOCK(p);
2456	thread_unlock(td);
2457	if (wakeup_swapper)
2458		kick_proc0();
2459}
2460
2461static void
2462sig_suspend_threads(struct thread *td, struct proc *p, int sending)
2463{
2464	struct thread *td2;
2465
2466	PROC_LOCK_ASSERT(p, MA_OWNED);
2467	PROC_SLOCK_ASSERT(p, MA_OWNED);
2468
2469	FOREACH_THREAD_IN_PROC(p, td2) {
2470		thread_lock(td2);
2471		td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
2472		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2473		    (td2->td_flags & TDF_SINTR)) {
2474			if (td2->td_flags & TDF_SBDRY) {
2475				/*
2476				 * Once a thread is asleep with
2477				 * TDF_SBDRY set, it should never
2478				 * become suspended due to this check.
2479				 */
2480				KASSERT(!TD_IS_SUSPENDED(td2),
2481				    ("thread with deferred stops suspended"));
2482			} else if (!TD_IS_SUSPENDED(td2)) {
2483				thread_suspend_one(td2);
2484			}
2485		} else if (!TD_IS_SUSPENDED(td2)) {
2486			if (sending || td != td2)
2487				td2->td_flags |= TDF_ASTPENDING;
2488#ifdef SMP
2489			if (TD_IS_RUNNING(td2) && td2 != td)
2490				forward_signal(td2);
2491#endif
2492		}
2493		thread_unlock(td2);
2494	}
2495}
2496
2497/*
2498 * Stop the process for an event deemed interesting to the debugger. If si is
2499 * non-NULL, this is a signal exchange; the new signal requested by the
2500 * debugger will be returned for handling. If si is NULL, this is some other
2501 * type of interesting event. The debugger may request a signal be delivered in
2502 * that case as well, however it will be deferred until it can be handled.
2503 */
2504int
2505ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2506{
2507	struct proc *p = td->td_proc;
2508	struct thread *td2;
2509	ksiginfo_t ksi;
2510	int prop;
2511
2512	PROC_LOCK_ASSERT(p, MA_OWNED);
2513	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2514	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2515	    &p->p_mtx.lock_object, "Stopping for traced signal");
2516
2517	td->td_xsig = sig;
2518
2519	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2520		td->td_dbgflags |= TDB_XSIG;
2521		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2522		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2523		PROC_SLOCK(p);
2524		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2525			if (P_KILLED(p)) {
2526				/*
2527				 * Ensure that, if we've been PT_KILLed, the
2528				 * exit status reflects that. Another thread
2529				 * may also be in ptracestop(), having just
2530				 * received the SIGKILL, but this thread was
2531				 * unsuspended first.
2532				 */
2533				td->td_dbgflags &= ~TDB_XSIG;
2534				td->td_xsig = SIGKILL;
2535				p->p_ptevents = 0;
2536				break;
2537			}
2538			if (p->p_flag & P_SINGLE_EXIT &&
2539			    !(td->td_dbgflags & TDB_EXIT)) {
2540				/*
2541				 * Ignore ptrace stops except for thread exit
2542				 * events when the process exits.
2543				 */
2544				td->td_dbgflags &= ~TDB_XSIG;
2545				PROC_SUNLOCK(p);
2546				return (0);
2547			}
2548
2549			/*
2550			 * Make wait(2) work.  Ensure that right after the
2551			 * attach, the thread which was decided to become the
2552			 * leader of attach gets reported to the waiter.
2553			 * Otherwise, just avoid overwriting another thread's
2554			 * assignment to p_xthread.  If another thread has
2555			 * already set p_xthread, the current thread will get
2556			 * a chance to report itself upon the next iteration.
2557			 */
2558			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2559			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2560			    p->p_xthread == NULL)) {
2561				p->p_xstat = sig;
2562				p->p_xthread = td;
2563				td->td_dbgflags &= ~TDB_FSTP;
2564				p->p_flag2 &= ~P2_PTRACE_FSTP;
2565				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2566				sig_suspend_threads(td, p, 0);
2567			}
2568			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2569				td->td_dbgflags &= ~TDB_STOPATFORK;
2570				cv_broadcast(&p->p_dbgwait);
2571			}
2572stopme:
2573			thread_suspend_switch(td, p);
2574			if (p->p_xthread == td)
2575				p->p_xthread = NULL;
2576			if (!(p->p_flag & P_TRACED))
2577				break;
2578			if (td->td_dbgflags & TDB_SUSPEND) {
2579				if (p->p_flag & P_SINGLE_EXIT)
2580					break;
2581				goto stopme;
2582			}
2583		}
2584		PROC_SUNLOCK(p);
2585	}
2586
2587	if (si != NULL && sig == td->td_xsig) {
2588		/* Parent wants us to take the original signal unchanged. */
2589		si->ksi_flags |= KSI_HEAD;
2590		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2591			si->ksi_signo = 0;
2592	} else if (td->td_xsig != 0) {
2593		/*
2594		 * If parent wants us to take a new signal, then it will leave
2595		 * it in td->td_xsig; otherwise we just look for signals again.
2596		 */
2597		ksiginfo_init(&ksi);
2598		ksi.ksi_signo = td->td_xsig;
2599		ksi.ksi_flags |= KSI_PTRACE;
2600		prop = sigprop(td->td_xsig);
2601		td2 = sigtd(p, td->td_xsig, prop);
2602		tdsendsignal(p, td2, td->td_xsig, &ksi);
2603		if (td != td2)
2604			return (0);
2605	}
2606
2607	return (td->td_xsig);
2608}
2609
2610static void
2611reschedule_signals(struct proc *p, sigset_t block, int flags)
2612{
2613	struct sigacts *ps;
2614	struct thread *td;
2615	int sig;
2616
2617	PROC_LOCK_ASSERT(p, MA_OWNED);
2618	ps = p->p_sigacts;
2619	mtx_assert(&ps->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0 ?
2620	    MA_OWNED : MA_NOTOWNED);
2621	if (SIGISEMPTY(p->p_siglist))
2622		return;
2623	SIGSETAND(block, p->p_siglist);
2624	while ((sig = sig_ffs(&block)) != 0) {
2625		SIGDELSET(block, sig);
2626		td = sigtd(p, sig, 0);
2627		signotify(td);
2628		if (!(flags & SIGPROCMASK_PS_LOCKED))
2629			mtx_lock(&ps->ps_mtx);
2630		if (p->p_flag & P_TRACED || SIGISMEMBER(ps->ps_sigcatch, sig))
2631			tdsigwakeup(td, sig, SIG_CATCH,
2632			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2633			     ERESTART));
2634		if (!(flags & SIGPROCMASK_PS_LOCKED))
2635			mtx_unlock(&ps->ps_mtx);
2636	}
2637}
2638
2639void
2640tdsigcleanup(struct thread *td)
2641{
2642	struct proc *p;
2643	sigset_t unblocked;
2644
2645	p = td->td_proc;
2646	PROC_LOCK_ASSERT(p, MA_OWNED);
2647
2648	sigqueue_flush(&td->td_sigqueue);
2649	if (p->p_numthreads == 1)
2650		return;
2651
2652	/*
2653	 * Since we cannot handle signals, notify signal post code
2654	 * about this by filling the sigmask.
2655	 *
2656	 * Also, if needed, wake up thread(s) that do not block the
2657	 * same signals as the exiting thread, since the thread might
2658	 * have been selected for delivery and woken up.
2659	 */
2660	SIGFILLSET(unblocked);
2661	SIGSETNAND(unblocked, td->td_sigmask);
2662	SIGFILLSET(td->td_sigmask);
2663	reschedule_signals(p, unblocked, 0);
2664
2665}
2666
2667/*
2668 * Defer the delivery of SIGSTOP for the current thread.  Returns true
2669 * if stops were deferred and false if they were already deferred.
2670 */
2671int
2672sigdeferstop(void)
2673{
2674	struct thread *td;
2675
2676	td = curthread;
2677	if (td->td_flags & TDF_SBDRY)
2678		return (0);
2679	thread_lock(td);
2680	td->td_flags |= TDF_SBDRY;
2681	thread_unlock(td);
2682	return (1);
2683}
2684
2685/*
2686 * Permit the delivery of SIGSTOP for the current thread.  This does
2687 * not immediately suspend if a stop was posted.  Instead, the thread
2688 * will suspend either via ast() or a subsequent interruptible sleep.
2689 */
2690int
2691sigallowstop(void)
2692{
2693	struct thread *td;
2694	int prev;
2695
2696	td = curthread;
2697	thread_lock(td);
2698	prev = (td->td_flags & TDF_SBDRY) != 0;
2699	td->td_flags &= ~TDF_SBDRY;
2700	thread_unlock(td);
2701	return (prev);
2702}
2703
2704/*
2705 * If the current process has received a signal (should be caught or cause
2706 * termination, should interrupt current syscall), return the signal number.
2707 * Stop signals with default action are processed immediately, then cleared;
2708 * they aren't returned.  This is checked after each entry to the system for
2709 * a syscall or trap (though this can usually be done without calling issignal
2710 * by checking the pending signal masks in cursig.) The normal call
2711 * sequence is
2712 *
2713 *	while (sig = cursig(curthread))
2714 *		postsig(sig);
2715 */
2716static int
2717issignal(struct thread *td)
2718{
2719	struct proc *p;
2720	struct sigacts *ps;
2721	struct sigqueue *queue;
2722	sigset_t sigpending;
2723	int sig, prop;
2724
2725	p = td->td_proc;
2726	ps = p->p_sigacts;
2727	mtx_assert(&ps->ps_mtx, MA_OWNED);
2728	PROC_LOCK_ASSERT(p, MA_OWNED);
2729	for (;;) {
2730		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
2731
2732		sigpending = td->td_sigqueue.sq_signals;
2733		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
2734		SIGSETNAND(sigpending, td->td_sigmask);
2735
2736		if (p->p_flag & P_PPWAIT || td->td_flags & TDF_SBDRY)
2737			SIG_STOPSIGMASK(sigpending);
2738		if (SIGISEMPTY(sigpending))	/* no signal to send */
2739			return (0);
2740		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
2741		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
2742		    SIGISMEMBER(sigpending, SIGSTOP)) {
2743			/*
2744			 * If debugger just attached, always consume
2745			 * SIGSTOP from ptrace(PT_ATTACH) first, to
2746			 * execute the debugger attach ritual in
2747			 * order.
2748			 */
2749			sig = SIGSTOP;
2750			td->td_dbgflags |= TDB_FSTP;
2751		} else {
2752			sig = sig_ffs(&sigpending);
2753		}
2754
2755		if (p->p_stops & S_SIG) {
2756			mtx_unlock(&ps->ps_mtx);
2757			stopevent(p, S_SIG, sig);
2758			mtx_lock(&ps->ps_mtx);
2759		}
2760
2761		/*
2762		 * We should see pending but ignored signals
2763		 * only if P_TRACED was on when they were posted.
2764		 */
2765		if (SIGISMEMBER(ps->ps_sigignore, sig) && (traced == 0)) {
2766			sigqueue_delete(&td->td_sigqueue, sig);
2767			sigqueue_delete(&p->p_sigqueue, sig);
2768			continue;
2769		}
2770		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
2771			/*
2772			 * If traced, always stop.
2773			 * Remove old signal from queue before the stop.
2774			 * XXX shrug off debugger, it causes siginfo to
2775			 * be thrown away.
2776			 */
2777			queue = &td->td_sigqueue;
2778			td->td_dbgksi.ksi_signo = 0;
2779			if (sigqueue_get(queue, sig, &td->td_dbgksi) == 0) {
2780				queue = &p->p_sigqueue;
2781				sigqueue_get(queue, sig, &td->td_dbgksi);
2782			}
2783
2784			mtx_unlock(&ps->ps_mtx);
2785			sig = ptracestop(td, sig, &td->td_dbgksi);
2786			mtx_lock(&ps->ps_mtx);
2787
2788			/*
2789			 * Keep looking if the debugger discarded the signal
2790			 * or replaced it with a masked signal.
2791			 *
2792			 * If the traced bit got turned off, go back up
2793			 * to the top to rescan signals.  This ensures
2794			 * that p_sig* and p_sigact are consistent.
2795			 */
2796			if (sig == 0 || (p->p_flag & P_TRACED) == 0)
2797				continue;
2798		}
2799
2800		prop = sigprop(sig);
2801
2802		/*
2803		 * Decide whether the signal should be returned.
2804		 * Return the signal's number, or fall through
2805		 * to clear it from the pending mask.
2806		 */
2807		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2808
2809		case (intptr_t)SIG_DFL:
2810			/*
2811			 * Don't take default actions on system processes.
2812			 */
2813			if (p->p_pid <= 1) {
2814#ifdef DIAGNOSTIC
2815				/*
2816				 * Are you sure you want to ignore SIGSEGV
2817				 * in init? XXX
2818				 */
2819				printf("Process (pid %lu) got signal %d\n",
2820					(u_long)p->p_pid, sig);
2821#endif
2822				break;		/* == ignore */
2823			}
2824			/*
2825			 * If there is a pending stop signal to process with
2826			 * default action, stop here, then clear the signal.
2827			 * Traced or exiting processes should ignore stops.
2828			 * Additionally, a member of an orphaned process group
2829			 * should ignore tty stops.
2830			 */
2831			if (prop & SA_STOP) {
2832				if (p->p_flag &
2833				    (P_TRACED | P_WEXIT | P_SINGLE_EXIT) ||
2834				    (p->p_pgrp->pg_jobc == 0 &&
2835				     prop & SA_TTYSTOP))
2836					break;	/* == ignore */
2837				mtx_unlock(&ps->ps_mtx);
2838				WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2839				    &p->p_mtx.lock_object, "Catching SIGSTOP");
2840				sigqueue_delete(&td->td_sigqueue, sig);
2841				sigqueue_delete(&p->p_sigqueue, sig);
2842				p->p_flag |= P_STOPPED_SIG;
2843				p->p_xstat = sig;
2844				PROC_SLOCK(p);
2845				sig_suspend_threads(td, p, 0);
2846				thread_suspend_switch(td, p);
2847				PROC_SUNLOCK(p);
2848				mtx_lock(&ps->ps_mtx);
2849				goto next;
2850			} else if (prop & SA_IGNORE) {
2851				/*
2852				 * Except for SIGCONT, shouldn't get here.
2853				 * Default action is to ignore; drop it.
2854				 */
2855				break;		/* == ignore */
2856			} else
2857				return (sig);
2858			/*NOTREACHED*/
2859
2860		case (intptr_t)SIG_IGN:
2861			/*
2862			 * Masking above should prevent us ever trying
2863			 * to take action on an ignored signal other
2864			 * than SIGCONT, unless process is traced.
2865			 */
2866			if ((prop & SA_CONT) == 0 &&
2867			    (p->p_flag & P_TRACED) == 0)
2868				printf("issignal\n");
2869			break;		/* == ignore */
2870
2871		default:
2872			/*
2873			 * This signal has an action, let
2874			 * postsig() process it.
2875			 */
2876			return (sig);
2877		}
2878		sigqueue_delete(&td->td_sigqueue, sig);	/* take the signal! */
2879		sigqueue_delete(&p->p_sigqueue, sig);
2880next:;
2881	}
2882	/* NOTREACHED */
2883}
2884
2885void
2886thread_stopped(struct proc *p)
2887{
2888	int n;
2889
2890	PROC_LOCK_ASSERT(p, MA_OWNED);
2891	PROC_SLOCK_ASSERT(p, MA_OWNED);
2892	n = p->p_suspcount;
2893	if (p == curproc)
2894		n++;
2895	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
2896		PROC_SUNLOCK(p);
2897		p->p_flag &= ~P_WAITED;
2898		PROC_LOCK(p->p_pptr);
2899		childproc_stopped(p, (p->p_flag & P_TRACED) ?
2900			CLD_TRAPPED : CLD_STOPPED);
2901		PROC_UNLOCK(p->p_pptr);
2902		PROC_SLOCK(p);
2903	}
2904}
2905
2906/*
2907 * Take the action for the specified signal
2908 * from the current set of pending signals.
2909 */
2910int
2911postsig(sig)
2912	register int sig;
2913{
2914	struct thread *td = curthread;
2915	register struct proc *p = td->td_proc;
2916	struct sigacts *ps;
2917	sig_t action;
2918	ksiginfo_t ksi;
2919	sigset_t returnmask;
2920
2921	KASSERT(sig != 0, ("postsig"));
2922
2923	PROC_LOCK_ASSERT(p, MA_OWNED);
2924	ps = p->p_sigacts;
2925	mtx_assert(&ps->ps_mtx, MA_OWNED);
2926	ksiginfo_init(&ksi);
2927	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
2928	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
2929		return (0);
2930	ksi.ksi_signo = sig;
2931	if (ksi.ksi_code == SI_TIMER)
2932		itimer_accept(p, ksi.ksi_timerid, &ksi);
2933	action = ps->ps_sigact[_SIG_IDX(sig)];
2934#ifdef KTRACE
2935	if (KTRPOINT(td, KTR_PSIG))
2936		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
2937		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
2938#endif
2939	if (p->p_stops & S_SIG) {
2940		mtx_unlock(&ps->ps_mtx);
2941		stopevent(p, S_SIG, sig);
2942		mtx_lock(&ps->ps_mtx);
2943	}
2944
2945	if (action == SIG_DFL) {
2946		/*
2947		 * Default action, where the default is to kill
2948		 * the process.  (Other cases were ignored above.)
2949		 */
2950		mtx_unlock(&ps->ps_mtx);
2951		sigexit(td, sig);
2952		/* NOTREACHED */
2953	} else {
2954		/*
2955		 * If we get here, the signal must be caught.
2956		 */
2957		KASSERT(action != SIG_IGN && !SIGISMEMBER(td->td_sigmask, sig),
2958		    ("postsig action"));
2959		/*
2960		 * Set the new mask value and also defer further
2961		 * occurrences of this signal.
2962		 *
2963		 * Special case: user has done a sigsuspend.  Here the
2964		 * current mask is not of interest, but rather the
2965		 * mask from before the sigsuspend is what we want
2966		 * restored after the signal processing is completed.
2967		 */
2968		if (td->td_pflags & TDP_OLDMASK) {
2969			returnmask = td->td_oldsigmask;
2970			td->td_pflags &= ~TDP_OLDMASK;
2971		} else
2972			returnmask = td->td_sigmask;
2973
2974		if (p->p_sig == sig) {
2975			p->p_code = 0;
2976			p->p_sig = 0;
2977		}
2978		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
2979		postsig_done(sig, td, ps);
2980	}
2981	return (1);
2982}
2983
2984/*
2985 * Kill the current process for stated reason.
2986 */
2987void
2988killproc(p, why)
2989	struct proc *p;
2990	char *why;
2991{
2992
2993	PROC_LOCK_ASSERT(p, MA_OWNED);
2994	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
2995	    p->p_comm);
2996	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n", p->p_pid,
2997	    p->p_comm, p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2998	p->p_flag |= P_WKILLED;
2999	kern_psignal(p, SIGKILL);
3000}
3001
3002/*
3003 * Force the current process to exit with the specified signal, dumping core
3004 * if appropriate.  We bypass the normal tests for masked and caught signals,
3005 * allowing unrecoverable failures to terminate the process without changing
3006 * signal state.  Mark the accounting record with the signal termination.
3007 * If dumping core, save the signal number for the debugger.  Calls exit and
3008 * does not return.
3009 */
3010void
3011sigexit(td, sig)
3012	struct thread *td;
3013	int sig;
3014{
3015	struct proc *p = td->td_proc;
3016
3017	PROC_LOCK_ASSERT(p, MA_OWNED);
3018	p->p_acflag |= AXSIG;
3019	/*
3020	 * We must be single-threading to generate a core dump.  This
3021	 * ensures that the registers in the core file are up-to-date.
3022	 * Also, the ELF dump handler assumes that the thread list doesn't
3023	 * change out from under it.
3024	 *
3025	 * XXX If another thread attempts to single-thread before us
3026	 *     (e.g. via fork()), we won't get a dump at all.
3027	 */
3028	if ((sigprop(sig) & SA_CORE) && thread_single(p, SINGLE_NO_EXIT) == 0) {
3029		p->p_sig = sig;
3030		/*
3031		 * Log signals which would cause core dumps
3032		 * (Log as LOG_INFO to appease those who don't want
3033		 * these messages.)
3034		 * XXX : Todo, as well as euid, write out ruid too
3035		 * Note that coredump() drops proc lock.
3036		 */
3037		if (coredump(td) == 0)
3038			sig |= WCOREFLAG;
3039		if (kern_logsigexit)
3040			log(LOG_INFO,
3041			    "pid %d (%s), uid %d: exited on signal %d%s\n",
3042			    p->p_pid, p->p_comm,
3043			    td->td_ucred ? td->td_ucred->cr_uid : -1,
3044			    sig &~ WCOREFLAG,
3045			    sig & WCOREFLAG ? " (core dumped)" : "");
3046	} else
3047		PROC_UNLOCK(p);
3048	exit1(td, W_EXITCODE(0, sig));
3049	/* NOTREACHED */
3050}
3051
3052/*
3053 * Send queued SIGCHLD to parent when child process's state
3054 * is changed.
3055 */
3056static void
3057sigparent(struct proc *p, int reason, int status)
3058{
3059	PROC_LOCK_ASSERT(p, MA_OWNED);
3060	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3061
3062	if (p->p_ksi != NULL) {
3063		p->p_ksi->ksi_signo  = SIGCHLD;
3064		p->p_ksi->ksi_code   = reason;
3065		p->p_ksi->ksi_status = status;
3066		p->p_ksi->ksi_pid    = p->p_pid;
3067		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3068		if (KSI_ONQ(p->p_ksi))
3069			return;
3070	}
3071	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3072}
3073
3074static void
3075childproc_jobstate(struct proc *p, int reason, int sig)
3076{
3077	struct sigacts *ps;
3078
3079	PROC_LOCK_ASSERT(p, MA_OWNED);
3080	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3081
3082	/*
3083	 * Wake up parent sleeping in kern_wait(), also send
3084	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3085	 * that parent will awake, because parent may masked
3086	 * the signal.
3087	 */
3088	p->p_pptr->p_flag |= P_STATCHILD;
3089	wakeup(p->p_pptr);
3090
3091	ps = p->p_pptr->p_sigacts;
3092	mtx_lock(&ps->ps_mtx);
3093	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3094		mtx_unlock(&ps->ps_mtx);
3095		sigparent(p, reason, sig);
3096	} else
3097		mtx_unlock(&ps->ps_mtx);
3098}
3099
3100void
3101childproc_stopped(struct proc *p, int reason)
3102{
3103	/* p_xstat is a plain signal number, not a full wait() status here. */
3104	childproc_jobstate(p, reason, p->p_xstat);
3105}
3106
3107void
3108childproc_continued(struct proc *p)
3109{
3110	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3111}
3112
3113void
3114childproc_exited(struct proc *p)
3115{
3116	int reason;
3117	int xstat = p->p_xstat; /* convert to int */
3118	int status;
3119
3120	if (WCOREDUMP(xstat))
3121		reason = CLD_DUMPED, status = WTERMSIG(xstat);
3122	else if (WIFSIGNALED(xstat))
3123		reason = CLD_KILLED, status = WTERMSIG(xstat);
3124	else
3125		reason = CLD_EXITED, status = WEXITSTATUS(xstat);
3126	/*
3127	 * XXX avoid calling wakeup(p->p_pptr), the work is
3128	 * done in exit1().
3129	 */
3130	sigparent(p, reason, status);
3131}
3132
3133/*
3134 * We only have 1 character for the core count in the format
3135 * string, so the range will be 0-9
3136 */
3137#define MAX_NUM_CORES 10
3138static int num_cores = 5;
3139
3140static int
3141sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3142{
3143	int error;
3144	int new_val;
3145
3146	new_val = num_cores;
3147	error = sysctl_handle_int(oidp, &new_val, 0, req);
3148	if (error != 0 || req->newptr == NULL)
3149		return (error);
3150	if (new_val > MAX_NUM_CORES)
3151		new_val = MAX_NUM_CORES;
3152	if (new_val < 0)
3153		new_val = 0;
3154	num_cores = new_val;
3155	return (0);
3156}
3157SYSCTL_PROC(_debug, OID_AUTO, ncores, CTLTYPE_INT|CTLFLAG_RW,
3158	    0, sizeof(int), sysctl_debug_num_cores_check, "I", "");
3159
3160#if defined(COMPRESS_USER_CORES)
3161int compress_user_cores = 1;
3162SYSCTL_INT(_kern, OID_AUTO, compress_user_cores, CTLFLAG_RW,
3163    &compress_user_cores, 0, "Compression of user corefiles");
3164
3165int compress_user_cores_gzlevel = -1; /* default level */
3166SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_gzlevel, CTLFLAG_RW,
3167    &compress_user_cores_gzlevel, -1, "Corefile gzip compression level");
3168
3169#define GZ_SUFFIX	".gz"
3170#define GZ_SUFFIX_LEN	3
3171#endif
3172
3173static char corefilename[MAXPATHLEN] = {"%N.core"};
3174TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3175SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
3176    sizeof(corefilename), "Process corefile name format string");
3177
3178/*
3179 * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3180 * Expand the name described in corefilename, using name, uid, and pid
3181 * and open/create core file.
3182 * corefilename is a printf-like string, with three format specifiers:
3183 *	%N	name of process ("name")
3184 *	%P	process id (pid)
3185 *	%U	user id (uid)
3186 * For example, "%N.core" is the default; they can be disabled completely
3187 * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3188 * This is controlled by the sysctl variable kern.corefile (see above).
3189 */
3190static int
3191corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3192    int compress, struct vnode **vpp, char **namep)
3193{
3194	struct nameidata nd;
3195	struct sbuf sb;
3196	const char *format;
3197	char *hostname, *name;
3198	int indexpos, i, error, cmode, flags, oflags;
3199
3200	hostname = NULL;
3201	format = corefilename;
3202	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3203	indexpos = -1;
3204	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3205	for (i = 0; format[i] != '\0'; i++) {
3206		switch (format[i]) {
3207		case '%':	/* Format character */
3208			i++;
3209			switch (format[i]) {
3210			case '%':
3211				sbuf_putc(&sb, '%');
3212				break;
3213			case 'H':	/* hostname */
3214				if (hostname == NULL) {
3215					hostname = malloc(MAXHOSTNAMELEN,
3216					    M_TEMP, M_WAITOK);
3217				}
3218				getcredhostname(td->td_ucred, hostname,
3219				    MAXHOSTNAMELEN);
3220				sbuf_printf(&sb, "%s", hostname);
3221				break;
3222			case 'I':	/* autoincrementing index */
3223				sbuf_printf(&sb, "0");
3224				indexpos = sbuf_len(&sb) - 1;
3225				break;
3226			case 'N':	/* process name */
3227				sbuf_printf(&sb, "%s", comm);
3228				break;
3229			case 'P':	/* process id */
3230				sbuf_printf(&sb, "%u", pid);
3231				break;
3232			case 'U':	/* user id */
3233				sbuf_printf(&sb, "%u", uid);
3234				break;
3235			default:
3236				log(LOG_ERR,
3237				    "Unknown format character %c in "
3238				    "corename `%s'\n", format[i], format);
3239				break;
3240			}
3241			break;
3242		default:
3243			sbuf_putc(&sb, format[i]);
3244			break;
3245		}
3246	}
3247	free(hostname, M_TEMP);
3248#ifdef COMPRESS_USER_CORES
3249	if (compress)
3250		sbuf_printf(&sb, GZ_SUFFIX);
3251#endif
3252	if (sbuf_error(&sb) != 0) {
3253		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3254		    "long\n", (long)pid, comm, (u_long)uid);
3255		sbuf_delete(&sb);
3256		free(name, M_TEMP);
3257		return (ENOMEM);
3258	}
3259	sbuf_finish(&sb);
3260	sbuf_delete(&sb);
3261
3262	cmode = S_IRUSR | S_IWUSR;
3263	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3264	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3265
3266	/*
3267	 * If the core format has a %I in it, then we need to check
3268	 * for existing corefiles before returning a name.
3269	 * To do this we iterate over 0..num_cores to find a
3270	 * non-existing core file name to use.
3271	 */
3272	if (indexpos != -1) {
3273		for (i = 0; i < num_cores; i++) {
3274			flags = O_CREAT | O_EXCL | FWRITE | O_NOFOLLOW;
3275			name[indexpos] = '0' + i;
3276			NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3277			error = vn_open_cred(&nd, &flags, cmode, oflags,
3278			    td->td_ucred, NULL);
3279			if (error) {
3280				if (error == EEXIST)
3281					continue;
3282				log(LOG_ERR,
3283				    "pid %d (%s), uid (%u):  Path `%s' failed "
3284				    "on initial open test, error = %d\n",
3285				    pid, comm, uid, name, error);
3286			}
3287			goto out;
3288		}
3289	}
3290
3291	flags = O_CREAT | FWRITE | O_NOFOLLOW;
3292	NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3293	error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred, NULL);
3294out:
3295	if (error) {
3296#ifdef AUDIT
3297		audit_proc_coredump(td, name, error);
3298#endif
3299		free(name, M_TEMP);
3300		return (error);
3301	}
3302	NDFREE(&nd, NDF_ONLY_PNBUF);
3303	*vpp = nd.ni_vp;
3304	*namep = name;
3305	return (0);
3306}
3307
3308/*
3309 * Dump a process' core.  The main routine does some
3310 * policy checking, and creates the name of the coredump;
3311 * then it passes on a vnode and a size limit to the process-specific
3312 * coredump routine if there is one; if there _is not_ one, it returns
3313 * ENOSYS; otherwise it returns the error from the process-specific routine.
3314 */
3315
3316static int
3317coredump(struct thread *td)
3318{
3319	struct proc *p = td->td_proc;
3320	struct ucred *cred = td->td_ucred;
3321	struct vnode *vp;
3322	struct flock lf;
3323	struct vattr vattr;
3324	int error, error1, locked;
3325	struct mount *mp;
3326	char *name;			/* name of corefile */
3327	off_t limit;
3328	int compress;
3329
3330#ifdef COMPRESS_USER_CORES
3331	compress = compress_user_cores;
3332#else
3333	compress = 0;
3334#endif
3335	PROC_LOCK_ASSERT(p, MA_OWNED);
3336	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3337	_STOPEVENT(p, S_CORE, 0);
3338
3339	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
3340	    (p->p_flag2 & P2_NOTRACE) != 0) {
3341		PROC_UNLOCK(p);
3342		return (EFAULT);
3343	}
3344
3345	/*
3346	 * Note that the bulk of limit checking is done after
3347	 * the corefile is created.  The exception is if the limit
3348	 * for corefiles is 0, in which case we don't bother
3349	 * creating the corefile at all.  This layout means that
3350	 * a corefile is truncated instead of not being created,
3351	 * if it is larger than the limit.
3352	 */
3353	limit = (off_t)lim_cur(p, RLIMIT_CORE);
3354	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
3355		PROC_UNLOCK(p);
3356		return (EFBIG);
3357	}
3358	PROC_UNLOCK(p);
3359
3360restart:
3361	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td, compress,
3362	    &vp, &name);
3363	if (error != 0)
3364		return (error);
3365
3366	/* Don't dump to non-regular files or files with links. */
3367	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
3368	    vattr.va_nlink != 1) {
3369		VOP_UNLOCK(vp, 0);
3370		error = EFAULT;
3371		goto close;
3372	}
3373
3374	VOP_UNLOCK(vp, 0);
3375	lf.l_whence = SEEK_SET;
3376	lf.l_start = 0;
3377	lf.l_len = 0;
3378	lf.l_type = F_WRLCK;
3379	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
3380
3381	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
3382		lf.l_type = F_UNLCK;
3383		if (locked)
3384			VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3385		if ((error = vn_close(vp, FWRITE, cred, td)) != 0)
3386			goto out;
3387		if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0)
3388			goto out;
3389		free(name, M_TEMP);
3390		goto restart;
3391	}
3392
3393	VATTR_NULL(&vattr);
3394	vattr.va_size = 0;
3395	if (set_core_nodump_flag)
3396		vattr.va_flags = UF_NODUMP;
3397	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3398	VOP_SETATTR(vp, &vattr, cred);
3399	VOP_UNLOCK(vp, 0);
3400	vn_finished_write(mp);
3401	PROC_LOCK(p);
3402	p->p_acflag |= ACORE;
3403	PROC_UNLOCK(p);
3404
3405	if (p->p_sysent->sv_coredump != NULL) {
3406		error = p->p_sysent->sv_coredump(td, vp, limit,
3407		    compress ? IMGACT_CORE_COMPRESS : 0);
3408	} else {
3409		error = ENOSYS;
3410	}
3411
3412	if (locked) {
3413		lf.l_type = F_UNLCK;
3414		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3415	}
3416close:
3417	error1 = vn_close(vp, FWRITE, cred, td);
3418	if (error == 0)
3419		error = error1;
3420out:
3421#ifdef AUDIT
3422	audit_proc_coredump(td, name, error);
3423#endif
3424	free(name, M_TEMP);
3425	return (error);
3426}
3427
3428/*
3429 * Nonexistent system call-- signal process (may want to handle it).  Flag
3430 * error in case process won't see signal immediately (blocked or ignored).
3431 */
3432#ifndef _SYS_SYSPROTO_H_
3433struct nosys_args {
3434	int	dummy;
3435};
3436#endif
3437/* ARGSUSED */
3438int
3439nosys(td, args)
3440	struct thread *td;
3441	struct nosys_args *args;
3442{
3443	struct proc *p = td->td_proc;
3444
3445	PROC_LOCK(p);
3446	tdsignal(td, SIGSYS);
3447	PROC_UNLOCK(p);
3448	return (ENOSYS);
3449}
3450
3451/*
3452 * Send a SIGIO or SIGURG signal to a process or process group using stored
3453 * credentials rather than those of the current process.
3454 */
3455void
3456pgsigio(sigiop, sig, checkctty)
3457	struct sigio **sigiop;
3458	int sig, checkctty;
3459{
3460	ksiginfo_t ksi;
3461	struct sigio *sigio;
3462
3463	ksiginfo_init(&ksi);
3464	ksi.ksi_signo = sig;
3465	ksi.ksi_code = SI_KERNEL;
3466
3467	SIGIO_LOCK();
3468	sigio = *sigiop;
3469	if (sigio == NULL) {
3470		SIGIO_UNLOCK();
3471		return;
3472	}
3473	if (sigio->sio_pgid > 0) {
3474		PROC_LOCK(sigio->sio_proc);
3475		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
3476			kern_psignal(sigio->sio_proc, sig);
3477		PROC_UNLOCK(sigio->sio_proc);
3478	} else if (sigio->sio_pgid < 0) {
3479		struct proc *p;
3480
3481		PGRP_LOCK(sigio->sio_pgrp);
3482		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
3483			PROC_LOCK(p);
3484			if (p->p_state == PRS_NORMAL &&
3485			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
3486			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
3487				kern_psignal(p, sig);
3488			PROC_UNLOCK(p);
3489		}
3490		PGRP_UNLOCK(sigio->sio_pgrp);
3491	}
3492	SIGIO_UNLOCK();
3493}
3494
3495static int
3496filt_sigattach(struct knote *kn)
3497{
3498	struct proc *p = curproc;
3499
3500	kn->kn_ptr.p_proc = p;
3501	kn->kn_flags |= EV_CLEAR;		/* automatically set */
3502
3503	knlist_add(&p->p_klist, kn, 0);
3504
3505	return (0);
3506}
3507
3508static void
3509filt_sigdetach(struct knote *kn)
3510{
3511	struct proc *p = kn->kn_ptr.p_proc;
3512
3513	knlist_remove(&p->p_klist, kn, 0);
3514}
3515
3516/*
3517 * signal knotes are shared with proc knotes, so we apply a mask to
3518 * the hint in order to differentiate them from process hints.  This
3519 * could be avoided by using a signal-specific knote list, but probably
3520 * isn't worth the trouble.
3521 */
3522static int
3523filt_signal(struct knote *kn, long hint)
3524{
3525
3526	if (hint & NOTE_SIGNAL) {
3527		hint &= ~NOTE_SIGNAL;
3528
3529		if (kn->kn_id == hint)
3530			kn->kn_data++;
3531	}
3532	return (kn->kn_data != 0);
3533}
3534
3535struct sigacts *
3536sigacts_alloc(void)
3537{
3538	struct sigacts *ps;
3539
3540	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
3541	ps->ps_refcnt = 1;
3542	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
3543	return (ps);
3544}
3545
3546void
3547sigacts_free(struct sigacts *ps)
3548{
3549
3550	if (refcount_release(&ps->ps_refcnt) == 0)
3551		return;
3552	mtx_destroy(&ps->ps_mtx);
3553	free(ps, M_SUBPROC);
3554}
3555
3556struct sigacts *
3557sigacts_hold(struct sigacts *ps)
3558{
3559
3560	refcount_acquire(&ps->ps_refcnt);
3561	return (ps);
3562}
3563
3564void
3565sigacts_copy(struct sigacts *dest, struct sigacts *src)
3566{
3567
3568	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
3569	mtx_lock(&src->ps_mtx);
3570	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
3571	mtx_unlock(&src->ps_mtx);
3572}
3573
3574int
3575sigacts_shared(struct sigacts *ps)
3576{
3577
3578	return (ps->ps_refcnt > 1);
3579}
3580