xref: /freebsd/sys/kern/kern_sig.c (revision adc56f5a383771f594829b7db9c263b6f0dcf1bd)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/ctype.h>
46 #include <sys/systm.h>
47 #include <sys/signalvar.h>
48 #include <sys/vnode.h>
49 #include <sys/acct.h>
50 #include <sys/bus.h>
51 #include <sys/capsicum.h>
52 #include <sys/compressor.h>
53 #include <sys/condvar.h>
54 #include <sys/event.h>
55 #include <sys/fcntl.h>
56 #include <sys/imgact.h>
57 #include <sys/kernel.h>
58 #include <sys/ktr.h>
59 #include <sys/ktrace.h>
60 #include <sys/limits.h>
61 #include <sys/lock.h>
62 #include <sys/malloc.h>
63 #include <sys/mutex.h>
64 #include <sys/refcount.h>
65 #include <sys/namei.h>
66 #include <sys/proc.h>
67 #include <sys/procdesc.h>
68 #include <sys/ptrace.h>
69 #include <sys/posix4.h>
70 #include <sys/pioctl.h>
71 #include <sys/racct.h>
72 #include <sys/resourcevar.h>
73 #include <sys/sdt.h>
74 #include <sys/sbuf.h>
75 #include <sys/sleepqueue.h>
76 #include <sys/smp.h>
77 #include <sys/stat.h>
78 #include <sys/sx.h>
79 #include <sys/syscallsubr.h>
80 #include <sys/sysctl.h>
81 #include <sys/sysent.h>
82 #include <sys/syslog.h>
83 #include <sys/sysproto.h>
84 #include <sys/timers.h>
85 #include <sys/unistd.h>
86 #include <sys/wait.h>
87 #include <vm/vm.h>
88 #include <vm/vm_extern.h>
89 #include <vm/uma.h>
90 
91 #include <sys/jail.h>
92 
93 #include <machine/cpu.h>
94 
95 #include <security/audit/audit.h>
96 
97 #define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
98 
99 SDT_PROVIDER_DECLARE(proc);
100 SDT_PROBE_DEFINE3(proc, , , signal__send,
101     "struct thread *", "struct proc *", "int");
102 SDT_PROBE_DEFINE2(proc, , , signal__clear,
103     "int", "ksiginfo_t *");
104 SDT_PROBE_DEFINE3(proc, , , signal__discard,
105     "struct thread *", "struct proc *", "int");
106 
107 static int	coredump(struct thread *);
108 static int	killpg1(struct thread *td, int sig, int pgid, int all,
109 		    ksiginfo_t *ksi);
110 static int	issignal(struct thread *td);
111 static int	sigprop(int sig);
112 static void	tdsigwakeup(struct thread *, int, sig_t, int);
113 static int	sig_suspend_threads(struct thread *, struct proc *, int);
114 static int	filt_sigattach(struct knote *kn);
115 static void	filt_sigdetach(struct knote *kn);
116 static int	filt_signal(struct knote *kn, long hint);
117 static struct thread *sigtd(struct proc *p, int sig, int prop);
118 static void	sigqueue_start(void);
119 
120 static uma_zone_t	ksiginfo_zone = NULL;
121 struct filterops sig_filtops = {
122 	.f_isfd = 0,
123 	.f_attach = filt_sigattach,
124 	.f_detach = filt_sigdetach,
125 	.f_event = filt_signal,
126 };
127 
128 static int	kern_logsigexit = 1;
129 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
130     &kern_logsigexit, 0,
131     "Log processes quitting on abnormal signals to syslog(3)");
132 
133 static int	kern_forcesigexit = 1;
134 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
135     &kern_forcesigexit, 0, "Force trap signal to be handled");
136 
137 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW, 0,
138     "POSIX real time signal");
139 
140 static int	max_pending_per_proc = 128;
141 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
142     &max_pending_per_proc, 0, "Max pending signals per proc");
143 
144 static int	preallocate_siginfo = 1024;
145 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
146     &preallocate_siginfo, 0, "Preallocated signal memory size");
147 
148 static int	signal_overflow = 0;
149 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
150     &signal_overflow, 0, "Number of signals overflew");
151 
152 static int	signal_alloc_fail = 0;
153 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
154     &signal_alloc_fail, 0, "signals failed to be allocated");
155 
156 static int	kern_lognosys = 0;
157 SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
158     "Log invalid syscalls");
159 
160 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
161 
162 /*
163  * Policy -- Can ucred cr1 send SIGIO to process cr2?
164  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
165  * in the right situations.
166  */
167 #define CANSIGIO(cr1, cr2) \
168 	((cr1)->cr_uid == 0 || \
169 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
170 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
171 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
172 	    (cr1)->cr_uid == (cr2)->cr_uid)
173 
174 static int	sugid_coredump;
175 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
176     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
177 
178 static int	capmode_coredump;
179 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
180     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
181 
182 static int	do_coredump = 1;
183 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
184 	&do_coredump, 0, "Enable/Disable coredumps");
185 
186 static int	set_core_nodump_flag = 0;
187 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
188 	0, "Enable setting the NODUMP flag on coredump files");
189 
190 static int	coredump_devctl = 0;
191 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
192 	0, "Generate a devctl notification when processes coredump");
193 
194 /*
195  * Signal properties and actions.
196  * The array below categorizes the signals and their default actions
197  * according to the following properties:
198  */
199 #define	SIGPROP_KILL		0x01	/* terminates process by default */
200 #define	SIGPROP_CORE		0x02	/* ditto and coredumps */
201 #define	SIGPROP_STOP		0x04	/* suspend process */
202 #define	SIGPROP_TTYSTOP		0x08	/* ditto, from tty */
203 #define	SIGPROP_IGNORE		0x10	/* ignore by default */
204 #define	SIGPROP_CONT		0x20	/* continue if suspended */
205 #define	SIGPROP_CANTMASK	0x40	/* non-maskable, catchable */
206 
207 static int sigproptbl[NSIG] = {
208 	[SIGHUP] =	SIGPROP_KILL,
209 	[SIGINT] =	SIGPROP_KILL,
210 	[SIGQUIT] =	SIGPROP_KILL | SIGPROP_CORE,
211 	[SIGILL] =	SIGPROP_KILL | SIGPROP_CORE,
212 	[SIGTRAP] =	SIGPROP_KILL | SIGPROP_CORE,
213 	[SIGABRT] =	SIGPROP_KILL | SIGPROP_CORE,
214 	[SIGEMT] =	SIGPROP_KILL | SIGPROP_CORE,
215 	[SIGFPE] =	SIGPROP_KILL | SIGPROP_CORE,
216 	[SIGKILL] =	SIGPROP_KILL,
217 	[SIGBUS] =	SIGPROP_KILL | SIGPROP_CORE,
218 	[SIGSEGV] =	SIGPROP_KILL | SIGPROP_CORE,
219 	[SIGSYS] =	SIGPROP_KILL | SIGPROP_CORE,
220 	[SIGPIPE] =	SIGPROP_KILL,
221 	[SIGALRM] =	SIGPROP_KILL,
222 	[SIGTERM] =	SIGPROP_KILL,
223 	[SIGURG] =	SIGPROP_IGNORE,
224 	[SIGSTOP] =	SIGPROP_STOP,
225 	[SIGTSTP] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
226 	[SIGCONT] =	SIGPROP_IGNORE | SIGPROP_CONT,
227 	[SIGCHLD] =	SIGPROP_IGNORE,
228 	[SIGTTIN] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
229 	[SIGTTOU] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
230 	[SIGIO] =	SIGPROP_IGNORE,
231 	[SIGXCPU] =	SIGPROP_KILL,
232 	[SIGXFSZ] =	SIGPROP_KILL,
233 	[SIGVTALRM] =	SIGPROP_KILL,
234 	[SIGPROF] =	SIGPROP_KILL,
235 	[SIGWINCH] =	SIGPROP_IGNORE,
236 	[SIGINFO] =	SIGPROP_IGNORE,
237 	[SIGUSR1] =	SIGPROP_KILL,
238 	[SIGUSR2] =	SIGPROP_KILL,
239 };
240 
241 static void reschedule_signals(struct proc *p, sigset_t block, int flags);
242 
243 static void
244 sigqueue_start(void)
245 {
246 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
247 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
248 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
249 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
250 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
251 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
252 }
253 
254 ksiginfo_t *
255 ksiginfo_alloc(int wait)
256 {
257 	int flags;
258 
259 	flags = M_ZERO;
260 	if (! wait)
261 		flags |= M_NOWAIT;
262 	if (ksiginfo_zone != NULL)
263 		return ((ksiginfo_t *)uma_zalloc(ksiginfo_zone, flags));
264 	return (NULL);
265 }
266 
267 void
268 ksiginfo_free(ksiginfo_t *ksi)
269 {
270 	uma_zfree(ksiginfo_zone, ksi);
271 }
272 
273 static __inline int
274 ksiginfo_tryfree(ksiginfo_t *ksi)
275 {
276 	if (!(ksi->ksi_flags & KSI_EXT)) {
277 		uma_zfree(ksiginfo_zone, ksi);
278 		return (1);
279 	}
280 	return (0);
281 }
282 
283 void
284 sigqueue_init(sigqueue_t *list, struct proc *p)
285 {
286 	SIGEMPTYSET(list->sq_signals);
287 	SIGEMPTYSET(list->sq_kill);
288 	SIGEMPTYSET(list->sq_ptrace);
289 	TAILQ_INIT(&list->sq_list);
290 	list->sq_proc = p;
291 	list->sq_flags = SQ_INIT;
292 }
293 
294 /*
295  * Get a signal's ksiginfo.
296  * Return:
297  *	0	-	signal not found
298  *	others	-	signal number
299  */
300 static int
301 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
302 {
303 	struct proc *p = sq->sq_proc;
304 	struct ksiginfo *ksi, *next;
305 	int count = 0;
306 
307 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
308 
309 	if (!SIGISMEMBER(sq->sq_signals, signo))
310 		return (0);
311 
312 	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
313 		count++;
314 		SIGDELSET(sq->sq_ptrace, signo);
315 		si->ksi_flags |= KSI_PTRACE;
316 	}
317 	if (SIGISMEMBER(sq->sq_kill, signo)) {
318 		count++;
319 		if (count == 1)
320 			SIGDELSET(sq->sq_kill, signo);
321 	}
322 
323 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
324 		if (ksi->ksi_signo == signo) {
325 			if (count == 0) {
326 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
327 				ksi->ksi_sigq = NULL;
328 				ksiginfo_copy(ksi, si);
329 				if (ksiginfo_tryfree(ksi) && p != NULL)
330 					p->p_pendingcnt--;
331 			}
332 			if (++count > 1)
333 				break;
334 		}
335 	}
336 
337 	if (count <= 1)
338 		SIGDELSET(sq->sq_signals, signo);
339 	si->ksi_signo = signo;
340 	return (signo);
341 }
342 
343 void
344 sigqueue_take(ksiginfo_t *ksi)
345 {
346 	struct ksiginfo *kp;
347 	struct proc	*p;
348 	sigqueue_t	*sq;
349 
350 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
351 		return;
352 
353 	p = sq->sq_proc;
354 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
355 	ksi->ksi_sigq = NULL;
356 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
357 		p->p_pendingcnt--;
358 
359 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
360 	     kp = TAILQ_NEXT(kp, ksi_link)) {
361 		if (kp->ksi_signo == ksi->ksi_signo)
362 			break;
363 	}
364 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
365 	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
366 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
367 }
368 
369 static int
370 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
371 {
372 	struct proc *p = sq->sq_proc;
373 	struct ksiginfo *ksi;
374 	int ret = 0;
375 
376 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
377 
378 	/*
379 	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
380 	 * for these signals.
381 	 */
382 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
383 		SIGADDSET(sq->sq_kill, signo);
384 		goto out_set_bit;
385 	}
386 
387 	/* directly insert the ksi, don't copy it */
388 	if (si->ksi_flags & KSI_INS) {
389 		if (si->ksi_flags & KSI_HEAD)
390 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
391 		else
392 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
393 		si->ksi_sigq = sq;
394 		goto out_set_bit;
395 	}
396 
397 	if (__predict_false(ksiginfo_zone == NULL)) {
398 		SIGADDSET(sq->sq_kill, signo);
399 		goto out_set_bit;
400 	}
401 
402 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
403 		signal_overflow++;
404 		ret = EAGAIN;
405 	} else if ((ksi = ksiginfo_alloc(0)) == NULL) {
406 		signal_alloc_fail++;
407 		ret = EAGAIN;
408 	} else {
409 		if (p != NULL)
410 			p->p_pendingcnt++;
411 		ksiginfo_copy(si, ksi);
412 		ksi->ksi_signo = signo;
413 		if (si->ksi_flags & KSI_HEAD)
414 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
415 		else
416 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
417 		ksi->ksi_sigq = sq;
418 	}
419 
420 	if (ret != 0) {
421 		if ((si->ksi_flags & KSI_PTRACE) != 0) {
422 			SIGADDSET(sq->sq_ptrace, signo);
423 			ret = 0;
424 			goto out_set_bit;
425 		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
426 		    (si->ksi_flags & KSI_SIGQ) == 0) {
427 			SIGADDSET(sq->sq_kill, signo);
428 			ret = 0;
429 			goto out_set_bit;
430 		}
431 		return (ret);
432 	}
433 
434 out_set_bit:
435 	SIGADDSET(sq->sq_signals, signo);
436 	return (ret);
437 }
438 
439 void
440 sigqueue_flush(sigqueue_t *sq)
441 {
442 	struct proc *p = sq->sq_proc;
443 	ksiginfo_t *ksi;
444 
445 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
446 
447 	if (p != NULL)
448 		PROC_LOCK_ASSERT(p, MA_OWNED);
449 
450 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
451 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
452 		ksi->ksi_sigq = NULL;
453 		if (ksiginfo_tryfree(ksi) && p != NULL)
454 			p->p_pendingcnt--;
455 	}
456 
457 	SIGEMPTYSET(sq->sq_signals);
458 	SIGEMPTYSET(sq->sq_kill);
459 	SIGEMPTYSET(sq->sq_ptrace);
460 }
461 
462 static void
463 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
464 {
465 	sigset_t tmp;
466 	struct proc *p1, *p2;
467 	ksiginfo_t *ksi, *next;
468 
469 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
470 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
471 	p1 = src->sq_proc;
472 	p2 = dst->sq_proc;
473 	/* Move siginfo to target list */
474 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
475 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
476 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
477 			if (p1 != NULL)
478 				p1->p_pendingcnt--;
479 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
480 			ksi->ksi_sigq = dst;
481 			if (p2 != NULL)
482 				p2->p_pendingcnt++;
483 		}
484 	}
485 
486 	/* Move pending bits to target list */
487 	tmp = src->sq_kill;
488 	SIGSETAND(tmp, *set);
489 	SIGSETOR(dst->sq_kill, tmp);
490 	SIGSETNAND(src->sq_kill, tmp);
491 
492 	tmp = src->sq_ptrace;
493 	SIGSETAND(tmp, *set);
494 	SIGSETOR(dst->sq_ptrace, tmp);
495 	SIGSETNAND(src->sq_ptrace, tmp);
496 
497 	tmp = src->sq_signals;
498 	SIGSETAND(tmp, *set);
499 	SIGSETOR(dst->sq_signals, tmp);
500 	SIGSETNAND(src->sq_signals, tmp);
501 }
502 
503 #if 0
504 static void
505 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
506 {
507 	sigset_t set;
508 
509 	SIGEMPTYSET(set);
510 	SIGADDSET(set, signo);
511 	sigqueue_move_set(src, dst, &set);
512 }
513 #endif
514 
515 static void
516 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
517 {
518 	struct proc *p = sq->sq_proc;
519 	ksiginfo_t *ksi, *next;
520 
521 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
522 
523 	/* Remove siginfo queue */
524 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
525 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
526 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
527 			ksi->ksi_sigq = NULL;
528 			if (ksiginfo_tryfree(ksi) && p != NULL)
529 				p->p_pendingcnt--;
530 		}
531 	}
532 	SIGSETNAND(sq->sq_kill, *set);
533 	SIGSETNAND(sq->sq_ptrace, *set);
534 	SIGSETNAND(sq->sq_signals, *set);
535 }
536 
537 void
538 sigqueue_delete(sigqueue_t *sq, int signo)
539 {
540 	sigset_t set;
541 
542 	SIGEMPTYSET(set);
543 	SIGADDSET(set, signo);
544 	sigqueue_delete_set(sq, &set);
545 }
546 
547 /* Remove a set of signals for a process */
548 static void
549 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
550 {
551 	sigqueue_t worklist;
552 	struct thread *td0;
553 
554 	PROC_LOCK_ASSERT(p, MA_OWNED);
555 
556 	sigqueue_init(&worklist, NULL);
557 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
558 
559 	FOREACH_THREAD_IN_PROC(p, td0)
560 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
561 
562 	sigqueue_flush(&worklist);
563 }
564 
565 void
566 sigqueue_delete_proc(struct proc *p, int signo)
567 {
568 	sigset_t set;
569 
570 	SIGEMPTYSET(set);
571 	SIGADDSET(set, signo);
572 	sigqueue_delete_set_proc(p, &set);
573 }
574 
575 static void
576 sigqueue_delete_stopmask_proc(struct proc *p)
577 {
578 	sigset_t set;
579 
580 	SIGEMPTYSET(set);
581 	SIGADDSET(set, SIGSTOP);
582 	SIGADDSET(set, SIGTSTP);
583 	SIGADDSET(set, SIGTTIN);
584 	SIGADDSET(set, SIGTTOU);
585 	sigqueue_delete_set_proc(p, &set);
586 }
587 
588 /*
589  * Determine signal that should be delivered to thread td, the current
590  * thread, 0 if none.  If there is a pending stop signal with default
591  * action, the process stops in issignal().
592  */
593 int
594 cursig(struct thread *td)
595 {
596 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
597 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
598 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
599 	return (SIGPENDING(td) ? issignal(td) : 0);
600 }
601 
602 /*
603  * Arrange for ast() to handle unmasked pending signals on return to user
604  * mode.  This must be called whenever a signal is added to td_sigqueue or
605  * unmasked in td_sigmask.
606  */
607 void
608 signotify(struct thread *td)
609 {
610 
611 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
612 
613 	if (SIGPENDING(td)) {
614 		thread_lock(td);
615 		td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING;
616 		thread_unlock(td);
617 	}
618 }
619 
620 /*
621  * Returns 1 (true) if altstack is configured for the thread, and the
622  * passed stack bottom address falls into the altstack range.  Handles
623  * the 43 compat special case where the alt stack size is zero.
624  */
625 int
626 sigonstack(size_t sp)
627 {
628 	struct thread *td;
629 
630 	td = curthread;
631 	if ((td->td_pflags & TDP_ALTSTACK) == 0)
632 		return (0);
633 #if defined(COMPAT_43)
634 	if (SV_PROC_FLAG(td->td_proc, SV_AOUT) && td->td_sigstk.ss_size == 0)
635 		return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
636 #endif
637 	return (sp >= (size_t)td->td_sigstk.ss_sp &&
638 	    sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
639 }
640 
641 static __inline int
642 sigprop(int sig)
643 {
644 
645 	if (sig > 0 && sig < nitems(sigproptbl))
646 		return (sigproptbl[sig]);
647 	return (0);
648 }
649 
650 int
651 sig_ffs(sigset_t *set)
652 {
653 	int i;
654 
655 	for (i = 0; i < _SIG_WORDS; i++)
656 		if (set->__bits[i])
657 			return (ffs(set->__bits[i]) + (i * 32));
658 	return (0);
659 }
660 
661 static bool
662 sigact_flag_test(const struct sigaction *act, int flag)
663 {
664 
665 	/*
666 	 * SA_SIGINFO is reset when signal disposition is set to
667 	 * ignore or default.  Other flags are kept according to user
668 	 * settings.
669 	 */
670 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
671 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
672 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
673 }
674 
675 /*
676  * kern_sigaction
677  * sigaction
678  * freebsd4_sigaction
679  * osigaction
680  */
681 int
682 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
683     struct sigaction *oact, int flags)
684 {
685 	struct sigacts *ps;
686 	struct proc *p = td->td_proc;
687 
688 	if (!_SIG_VALID(sig))
689 		return (EINVAL);
690 	if (act != NULL && act->sa_handler != SIG_DFL &&
691 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
692 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
693 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
694 		return (EINVAL);
695 
696 	PROC_LOCK(p);
697 	ps = p->p_sigacts;
698 	mtx_lock(&ps->ps_mtx);
699 	if (oact) {
700 		memset(oact, 0, sizeof(*oact));
701 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
702 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
703 			oact->sa_flags |= SA_ONSTACK;
704 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
705 			oact->sa_flags |= SA_RESTART;
706 		if (SIGISMEMBER(ps->ps_sigreset, sig))
707 			oact->sa_flags |= SA_RESETHAND;
708 		if (SIGISMEMBER(ps->ps_signodefer, sig))
709 			oact->sa_flags |= SA_NODEFER;
710 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
711 			oact->sa_flags |= SA_SIGINFO;
712 			oact->sa_sigaction =
713 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
714 		} else
715 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
716 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
717 			oact->sa_flags |= SA_NOCLDSTOP;
718 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
719 			oact->sa_flags |= SA_NOCLDWAIT;
720 	}
721 	if (act) {
722 		if ((sig == SIGKILL || sig == SIGSTOP) &&
723 		    act->sa_handler != SIG_DFL) {
724 			mtx_unlock(&ps->ps_mtx);
725 			PROC_UNLOCK(p);
726 			return (EINVAL);
727 		}
728 
729 		/*
730 		 * Change setting atomically.
731 		 */
732 
733 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
734 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
735 		if (sigact_flag_test(act, SA_SIGINFO)) {
736 			ps->ps_sigact[_SIG_IDX(sig)] =
737 			    (__sighandler_t *)act->sa_sigaction;
738 			SIGADDSET(ps->ps_siginfo, sig);
739 		} else {
740 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
741 			SIGDELSET(ps->ps_siginfo, sig);
742 		}
743 		if (!sigact_flag_test(act, SA_RESTART))
744 			SIGADDSET(ps->ps_sigintr, sig);
745 		else
746 			SIGDELSET(ps->ps_sigintr, sig);
747 		if (sigact_flag_test(act, SA_ONSTACK))
748 			SIGADDSET(ps->ps_sigonstack, sig);
749 		else
750 			SIGDELSET(ps->ps_sigonstack, sig);
751 		if (sigact_flag_test(act, SA_RESETHAND))
752 			SIGADDSET(ps->ps_sigreset, sig);
753 		else
754 			SIGDELSET(ps->ps_sigreset, sig);
755 		if (sigact_flag_test(act, SA_NODEFER))
756 			SIGADDSET(ps->ps_signodefer, sig);
757 		else
758 			SIGDELSET(ps->ps_signodefer, sig);
759 		if (sig == SIGCHLD) {
760 			if (act->sa_flags & SA_NOCLDSTOP)
761 				ps->ps_flag |= PS_NOCLDSTOP;
762 			else
763 				ps->ps_flag &= ~PS_NOCLDSTOP;
764 			if (act->sa_flags & SA_NOCLDWAIT) {
765 				/*
766 				 * Paranoia: since SA_NOCLDWAIT is implemented
767 				 * by reparenting the dying child to PID 1 (and
768 				 * trust it to reap the zombie), PID 1 itself
769 				 * is forbidden to set SA_NOCLDWAIT.
770 				 */
771 				if (p->p_pid == 1)
772 					ps->ps_flag &= ~PS_NOCLDWAIT;
773 				else
774 					ps->ps_flag |= PS_NOCLDWAIT;
775 			} else
776 				ps->ps_flag &= ~PS_NOCLDWAIT;
777 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
778 				ps->ps_flag |= PS_CLDSIGIGN;
779 			else
780 				ps->ps_flag &= ~PS_CLDSIGIGN;
781 		}
782 		/*
783 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
784 		 * and for signals set to SIG_DFL where the default is to
785 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
786 		 * have to restart the process.
787 		 */
788 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
789 		    (sigprop(sig) & SIGPROP_IGNORE &&
790 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
791 			/* never to be seen again */
792 			sigqueue_delete_proc(p, sig);
793 			if (sig != SIGCONT)
794 				/* easier in psignal */
795 				SIGADDSET(ps->ps_sigignore, sig);
796 			SIGDELSET(ps->ps_sigcatch, sig);
797 		} else {
798 			SIGDELSET(ps->ps_sigignore, sig);
799 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
800 				SIGDELSET(ps->ps_sigcatch, sig);
801 			else
802 				SIGADDSET(ps->ps_sigcatch, sig);
803 		}
804 #ifdef COMPAT_FREEBSD4
805 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
806 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
807 		    (flags & KSA_FREEBSD4) == 0)
808 			SIGDELSET(ps->ps_freebsd4, sig);
809 		else
810 			SIGADDSET(ps->ps_freebsd4, sig);
811 #endif
812 #ifdef COMPAT_43
813 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
814 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
815 		    (flags & KSA_OSIGSET) == 0)
816 			SIGDELSET(ps->ps_osigset, sig);
817 		else
818 			SIGADDSET(ps->ps_osigset, sig);
819 #endif
820 	}
821 	mtx_unlock(&ps->ps_mtx);
822 	PROC_UNLOCK(p);
823 	return (0);
824 }
825 
826 #ifndef _SYS_SYSPROTO_H_
827 struct sigaction_args {
828 	int	sig;
829 	struct	sigaction *act;
830 	struct	sigaction *oact;
831 };
832 #endif
833 int
834 sys_sigaction(struct thread *td, struct sigaction_args *uap)
835 {
836 	struct sigaction act, oact;
837 	struct sigaction *actp, *oactp;
838 	int error;
839 
840 	actp = (uap->act != NULL) ? &act : NULL;
841 	oactp = (uap->oact != NULL) ? &oact : NULL;
842 	if (actp) {
843 		error = copyin(uap->act, actp, sizeof(act));
844 		if (error)
845 			return (error);
846 	}
847 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
848 	if (oactp && !error)
849 		error = copyout(oactp, uap->oact, sizeof(oact));
850 	return (error);
851 }
852 
853 #ifdef COMPAT_FREEBSD4
854 #ifndef _SYS_SYSPROTO_H_
855 struct freebsd4_sigaction_args {
856 	int	sig;
857 	struct	sigaction *act;
858 	struct	sigaction *oact;
859 };
860 #endif
861 int
862 freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
863 {
864 	struct sigaction act, oact;
865 	struct sigaction *actp, *oactp;
866 	int error;
867 
868 
869 	actp = (uap->act != NULL) ? &act : NULL;
870 	oactp = (uap->oact != NULL) ? &oact : NULL;
871 	if (actp) {
872 		error = copyin(uap->act, actp, sizeof(act));
873 		if (error)
874 			return (error);
875 	}
876 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
877 	if (oactp && !error)
878 		error = copyout(oactp, uap->oact, sizeof(oact));
879 	return (error);
880 }
881 #endif	/* COMAPT_FREEBSD4 */
882 
883 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
884 #ifndef _SYS_SYSPROTO_H_
885 struct osigaction_args {
886 	int	signum;
887 	struct	osigaction *nsa;
888 	struct	osigaction *osa;
889 };
890 #endif
891 int
892 osigaction(struct thread *td, struct osigaction_args *uap)
893 {
894 	struct osigaction sa;
895 	struct sigaction nsa, osa;
896 	struct sigaction *nsap, *osap;
897 	int error;
898 
899 	if (uap->signum <= 0 || uap->signum >= ONSIG)
900 		return (EINVAL);
901 
902 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
903 	osap = (uap->osa != NULL) ? &osa : NULL;
904 
905 	if (nsap) {
906 		error = copyin(uap->nsa, &sa, sizeof(sa));
907 		if (error)
908 			return (error);
909 		nsap->sa_handler = sa.sa_handler;
910 		nsap->sa_flags = sa.sa_flags;
911 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
912 	}
913 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
914 	if (osap && !error) {
915 		sa.sa_handler = osap->sa_handler;
916 		sa.sa_flags = osap->sa_flags;
917 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
918 		error = copyout(&sa, uap->osa, sizeof(sa));
919 	}
920 	return (error);
921 }
922 
923 #if !defined(__i386__)
924 /* Avoid replicating the same stub everywhere */
925 int
926 osigreturn(struct thread *td, struct osigreturn_args *uap)
927 {
928 
929 	return (nosys(td, (struct nosys_args *)uap));
930 }
931 #endif
932 #endif /* COMPAT_43 */
933 
934 /*
935  * Initialize signal state for process 0;
936  * set to ignore signals that are ignored by default.
937  */
938 void
939 siginit(struct proc *p)
940 {
941 	int i;
942 	struct sigacts *ps;
943 
944 	PROC_LOCK(p);
945 	ps = p->p_sigacts;
946 	mtx_lock(&ps->ps_mtx);
947 	for (i = 1; i <= NSIG; i++) {
948 		if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) {
949 			SIGADDSET(ps->ps_sigignore, i);
950 		}
951 	}
952 	mtx_unlock(&ps->ps_mtx);
953 	PROC_UNLOCK(p);
954 }
955 
956 /*
957  * Reset specified signal to the default disposition.
958  */
959 static void
960 sigdflt(struct sigacts *ps, int sig)
961 {
962 
963 	mtx_assert(&ps->ps_mtx, MA_OWNED);
964 	SIGDELSET(ps->ps_sigcatch, sig);
965 	if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT)
966 		SIGADDSET(ps->ps_sigignore, sig);
967 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
968 	SIGDELSET(ps->ps_siginfo, sig);
969 }
970 
971 /*
972  * Reset signals for an exec of the specified process.
973  */
974 void
975 execsigs(struct proc *p)
976 {
977 	sigset_t osigignore;
978 	struct sigacts *ps;
979 	int sig;
980 	struct thread *td;
981 
982 	/*
983 	 * Reset caught signals.  Held signals remain held
984 	 * through td_sigmask (unless they were caught,
985 	 * and are now ignored by default).
986 	 */
987 	PROC_LOCK_ASSERT(p, MA_OWNED);
988 	ps = p->p_sigacts;
989 	mtx_lock(&ps->ps_mtx);
990 	sig_drop_caught(p);
991 
992 	/*
993 	 * As CloudABI processes cannot modify signal handlers, fully
994 	 * reset all signals to their default behavior. Do ignore
995 	 * SIGPIPE, as it would otherwise be impossible to recover from
996 	 * writes to broken pipes and sockets.
997 	 */
998 	if (SV_PROC_ABI(p) == SV_ABI_CLOUDABI) {
999 		osigignore = ps->ps_sigignore;
1000 		while (SIGNOTEMPTY(osigignore)) {
1001 			sig = sig_ffs(&osigignore);
1002 			SIGDELSET(osigignore, sig);
1003 			if (sig != SIGPIPE)
1004 				sigdflt(ps, sig);
1005 		}
1006 		SIGADDSET(ps->ps_sigignore, SIGPIPE);
1007 	}
1008 
1009 	/*
1010 	 * Reset stack state to the user stack.
1011 	 * Clear set of signals caught on the signal stack.
1012 	 */
1013 	td = curthread;
1014 	MPASS(td->td_proc == p);
1015 	td->td_sigstk.ss_flags = SS_DISABLE;
1016 	td->td_sigstk.ss_size = 0;
1017 	td->td_sigstk.ss_sp = 0;
1018 	td->td_pflags &= ~TDP_ALTSTACK;
1019 	/*
1020 	 * Reset no zombies if child dies flag as Solaris does.
1021 	 */
1022 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1023 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1024 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1025 	mtx_unlock(&ps->ps_mtx);
1026 }
1027 
1028 /*
1029  * kern_sigprocmask()
1030  *
1031  *	Manipulate signal mask.
1032  */
1033 int
1034 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1035     int flags)
1036 {
1037 	sigset_t new_block, oset1;
1038 	struct proc *p;
1039 	int error;
1040 
1041 	p = td->td_proc;
1042 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1043 		PROC_LOCK_ASSERT(p, MA_OWNED);
1044 	else
1045 		PROC_LOCK(p);
1046 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1047 	    ? MA_OWNED : MA_NOTOWNED);
1048 	if (oset != NULL)
1049 		*oset = td->td_sigmask;
1050 
1051 	error = 0;
1052 	if (set != NULL) {
1053 		switch (how) {
1054 		case SIG_BLOCK:
1055 			SIG_CANTMASK(*set);
1056 			oset1 = td->td_sigmask;
1057 			SIGSETOR(td->td_sigmask, *set);
1058 			new_block = td->td_sigmask;
1059 			SIGSETNAND(new_block, oset1);
1060 			break;
1061 		case SIG_UNBLOCK:
1062 			SIGSETNAND(td->td_sigmask, *set);
1063 			signotify(td);
1064 			goto out;
1065 		case SIG_SETMASK:
1066 			SIG_CANTMASK(*set);
1067 			oset1 = td->td_sigmask;
1068 			if (flags & SIGPROCMASK_OLD)
1069 				SIGSETLO(td->td_sigmask, *set);
1070 			else
1071 				td->td_sigmask = *set;
1072 			new_block = td->td_sigmask;
1073 			SIGSETNAND(new_block, oset1);
1074 			signotify(td);
1075 			break;
1076 		default:
1077 			error = EINVAL;
1078 			goto out;
1079 		}
1080 
1081 		/*
1082 		 * The new_block set contains signals that were not previously
1083 		 * blocked, but are blocked now.
1084 		 *
1085 		 * In case we block any signal that was not previously blocked
1086 		 * for td, and process has the signal pending, try to schedule
1087 		 * signal delivery to some thread that does not block the
1088 		 * signal, possibly waking it up.
1089 		 */
1090 		if (p->p_numthreads != 1)
1091 			reschedule_signals(p, new_block, flags);
1092 	}
1093 
1094 out:
1095 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1096 		PROC_UNLOCK(p);
1097 	return (error);
1098 }
1099 
1100 #ifndef _SYS_SYSPROTO_H_
1101 struct sigprocmask_args {
1102 	int	how;
1103 	const sigset_t *set;
1104 	sigset_t *oset;
1105 };
1106 #endif
1107 int
1108 sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1109 {
1110 	sigset_t set, oset;
1111 	sigset_t *setp, *osetp;
1112 	int error;
1113 
1114 	setp = (uap->set != NULL) ? &set : NULL;
1115 	osetp = (uap->oset != NULL) ? &oset : NULL;
1116 	if (setp) {
1117 		error = copyin(uap->set, setp, sizeof(set));
1118 		if (error)
1119 			return (error);
1120 	}
1121 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1122 	if (osetp && !error) {
1123 		error = copyout(osetp, uap->oset, sizeof(oset));
1124 	}
1125 	return (error);
1126 }
1127 
1128 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1129 #ifndef _SYS_SYSPROTO_H_
1130 struct osigprocmask_args {
1131 	int	how;
1132 	osigset_t mask;
1133 };
1134 #endif
1135 int
1136 osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1137 {
1138 	sigset_t set, oset;
1139 	int error;
1140 
1141 	OSIG2SIG(uap->mask, set);
1142 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1143 	SIG2OSIG(oset, td->td_retval[0]);
1144 	return (error);
1145 }
1146 #endif /* COMPAT_43 */
1147 
1148 int
1149 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1150 {
1151 	ksiginfo_t ksi;
1152 	sigset_t set;
1153 	int error;
1154 
1155 	error = copyin(uap->set, &set, sizeof(set));
1156 	if (error) {
1157 		td->td_retval[0] = error;
1158 		return (0);
1159 	}
1160 
1161 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1162 	if (error) {
1163 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1164 			error = ERESTART;
1165 		if (error == ERESTART)
1166 			return (error);
1167 		td->td_retval[0] = error;
1168 		return (0);
1169 	}
1170 
1171 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1172 	td->td_retval[0] = error;
1173 	return (0);
1174 }
1175 
1176 int
1177 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1178 {
1179 	struct timespec ts;
1180 	struct timespec *timeout;
1181 	sigset_t set;
1182 	ksiginfo_t ksi;
1183 	int error;
1184 
1185 	if (uap->timeout) {
1186 		error = copyin(uap->timeout, &ts, sizeof(ts));
1187 		if (error)
1188 			return (error);
1189 
1190 		timeout = &ts;
1191 	} else
1192 		timeout = NULL;
1193 
1194 	error = copyin(uap->set, &set, sizeof(set));
1195 	if (error)
1196 		return (error);
1197 
1198 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1199 	if (error)
1200 		return (error);
1201 
1202 	if (uap->info)
1203 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1204 
1205 	if (error == 0)
1206 		td->td_retval[0] = ksi.ksi_signo;
1207 	return (error);
1208 }
1209 
1210 int
1211 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1212 {
1213 	ksiginfo_t ksi;
1214 	sigset_t set;
1215 	int error;
1216 
1217 	error = copyin(uap->set, &set, sizeof(set));
1218 	if (error)
1219 		return (error);
1220 
1221 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1222 	if (error)
1223 		return (error);
1224 
1225 	if (uap->info)
1226 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1227 
1228 	if (error == 0)
1229 		td->td_retval[0] = ksi.ksi_signo;
1230 	return (error);
1231 }
1232 
1233 static void
1234 proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1235 {
1236 	struct thread *thr;
1237 
1238 	FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1239 		if (thr == td)
1240 			thr->td_si = *si;
1241 		else
1242 			thr->td_si.si_signo = 0;
1243 	}
1244 }
1245 
1246 int
1247 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1248 	struct timespec *timeout)
1249 {
1250 	struct sigacts *ps;
1251 	sigset_t saved_mask, new_block;
1252 	struct proc *p;
1253 	int error, sig, timo, timevalid = 0;
1254 	struct timespec rts, ets, ts;
1255 	struct timeval tv;
1256 	bool traced;
1257 
1258 	p = td->td_proc;
1259 	error = 0;
1260 	ets.tv_sec = 0;
1261 	ets.tv_nsec = 0;
1262 	traced = false;
1263 
1264 	if (timeout != NULL) {
1265 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1266 			timevalid = 1;
1267 			getnanouptime(&rts);
1268 			timespecadd(&rts, timeout, &ets);
1269 		}
1270 	}
1271 	ksiginfo_init(ksi);
1272 	/* Some signals can not be waited for. */
1273 	SIG_CANTMASK(waitset);
1274 	ps = p->p_sigacts;
1275 	PROC_LOCK(p);
1276 	saved_mask = td->td_sigmask;
1277 	SIGSETNAND(td->td_sigmask, waitset);
1278 	for (;;) {
1279 		mtx_lock(&ps->ps_mtx);
1280 		sig = cursig(td);
1281 		mtx_unlock(&ps->ps_mtx);
1282 		KASSERT(sig >= 0, ("sig %d", sig));
1283 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1284 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1285 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1286 				error = 0;
1287 				break;
1288 			}
1289 		}
1290 
1291 		if (error != 0)
1292 			break;
1293 
1294 		/*
1295 		 * POSIX says this must be checked after looking for pending
1296 		 * signals.
1297 		 */
1298 		if (timeout != NULL) {
1299 			if (!timevalid) {
1300 				error = EINVAL;
1301 				break;
1302 			}
1303 			getnanouptime(&rts);
1304 			if (timespeccmp(&rts, &ets, >=)) {
1305 				error = EAGAIN;
1306 				break;
1307 			}
1308 			timespecsub(&ets, &rts, &ts);
1309 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1310 			timo = tvtohz(&tv);
1311 		} else {
1312 			timo = 0;
1313 		}
1314 
1315 		if (traced) {
1316 			error = EINTR;
1317 			break;
1318 		}
1319 
1320 		error = msleep(ps, &p->p_mtx, PPAUSE|PCATCH, "sigwait", timo);
1321 
1322 		if (timeout != NULL) {
1323 			if (error == ERESTART) {
1324 				/* Timeout can not be restarted. */
1325 				error = EINTR;
1326 			} else if (error == EAGAIN) {
1327 				/* We will calculate timeout by ourself. */
1328 				error = 0;
1329 			}
1330 		}
1331 
1332 		/*
1333 		 * If PTRACE_SCE or PTRACE_SCX were set after
1334 		 * userspace entered the syscall, return spurious
1335 		 * EINTR after wait was done.  Only do this as last
1336 		 * resort after rechecking for possible queued signals
1337 		 * and expired timeouts.
1338 		 */
1339 		if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1340 			traced = true;
1341 	}
1342 
1343 	new_block = saved_mask;
1344 	SIGSETNAND(new_block, td->td_sigmask);
1345 	td->td_sigmask = saved_mask;
1346 	/*
1347 	 * Fewer signals can be delivered to us, reschedule signal
1348 	 * notification.
1349 	 */
1350 	if (p->p_numthreads != 1)
1351 		reschedule_signals(p, new_block, 0);
1352 
1353 	if (error == 0) {
1354 		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1355 
1356 		if (ksi->ksi_code == SI_TIMER)
1357 			itimer_accept(p, ksi->ksi_timerid, ksi);
1358 
1359 #ifdef KTRACE
1360 		if (KTRPOINT(td, KTR_PSIG)) {
1361 			sig_t action;
1362 
1363 			mtx_lock(&ps->ps_mtx);
1364 			action = ps->ps_sigact[_SIG_IDX(sig)];
1365 			mtx_unlock(&ps->ps_mtx);
1366 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1367 		}
1368 #endif
1369 		if (sig == SIGKILL) {
1370 			proc_td_siginfo_capture(td, &ksi->ksi_info);
1371 			sigexit(td, sig);
1372 		}
1373 	}
1374 	PROC_UNLOCK(p);
1375 	return (error);
1376 }
1377 
1378 #ifndef _SYS_SYSPROTO_H_
1379 struct sigpending_args {
1380 	sigset_t	*set;
1381 };
1382 #endif
1383 int
1384 sys_sigpending(struct thread *td, struct sigpending_args *uap)
1385 {
1386 	struct proc *p = td->td_proc;
1387 	sigset_t pending;
1388 
1389 	PROC_LOCK(p);
1390 	pending = p->p_sigqueue.sq_signals;
1391 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1392 	PROC_UNLOCK(p);
1393 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1394 }
1395 
1396 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1397 #ifndef _SYS_SYSPROTO_H_
1398 struct osigpending_args {
1399 	int	dummy;
1400 };
1401 #endif
1402 int
1403 osigpending(struct thread *td, struct osigpending_args *uap)
1404 {
1405 	struct proc *p = td->td_proc;
1406 	sigset_t pending;
1407 
1408 	PROC_LOCK(p);
1409 	pending = p->p_sigqueue.sq_signals;
1410 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1411 	PROC_UNLOCK(p);
1412 	SIG2OSIG(pending, td->td_retval[0]);
1413 	return (0);
1414 }
1415 #endif /* COMPAT_43 */
1416 
1417 #if defined(COMPAT_43)
1418 /*
1419  * Generalized interface signal handler, 4.3-compatible.
1420  */
1421 #ifndef _SYS_SYSPROTO_H_
1422 struct osigvec_args {
1423 	int	signum;
1424 	struct	sigvec *nsv;
1425 	struct	sigvec *osv;
1426 };
1427 #endif
1428 /* ARGSUSED */
1429 int
1430 osigvec(struct thread *td, struct osigvec_args *uap)
1431 {
1432 	struct sigvec vec;
1433 	struct sigaction nsa, osa;
1434 	struct sigaction *nsap, *osap;
1435 	int error;
1436 
1437 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1438 		return (EINVAL);
1439 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1440 	osap = (uap->osv != NULL) ? &osa : NULL;
1441 	if (nsap) {
1442 		error = copyin(uap->nsv, &vec, sizeof(vec));
1443 		if (error)
1444 			return (error);
1445 		nsap->sa_handler = vec.sv_handler;
1446 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1447 		nsap->sa_flags = vec.sv_flags;
1448 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1449 	}
1450 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1451 	if (osap && !error) {
1452 		vec.sv_handler = osap->sa_handler;
1453 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1454 		vec.sv_flags = osap->sa_flags;
1455 		vec.sv_flags &= ~SA_NOCLDWAIT;
1456 		vec.sv_flags ^= SA_RESTART;
1457 		error = copyout(&vec, uap->osv, sizeof(vec));
1458 	}
1459 	return (error);
1460 }
1461 
1462 #ifndef _SYS_SYSPROTO_H_
1463 struct osigblock_args {
1464 	int	mask;
1465 };
1466 #endif
1467 int
1468 osigblock(struct thread *td, struct osigblock_args *uap)
1469 {
1470 	sigset_t set, oset;
1471 
1472 	OSIG2SIG(uap->mask, set);
1473 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1474 	SIG2OSIG(oset, td->td_retval[0]);
1475 	return (0);
1476 }
1477 
1478 #ifndef _SYS_SYSPROTO_H_
1479 struct osigsetmask_args {
1480 	int	mask;
1481 };
1482 #endif
1483 int
1484 osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1485 {
1486 	sigset_t set, oset;
1487 
1488 	OSIG2SIG(uap->mask, set);
1489 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1490 	SIG2OSIG(oset, td->td_retval[0]);
1491 	return (0);
1492 }
1493 #endif /* COMPAT_43 */
1494 
1495 /*
1496  * Suspend calling thread until signal, providing mask to be set in the
1497  * meantime.
1498  */
1499 #ifndef _SYS_SYSPROTO_H_
1500 struct sigsuspend_args {
1501 	const sigset_t *sigmask;
1502 };
1503 #endif
1504 /* ARGSUSED */
1505 int
1506 sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1507 {
1508 	sigset_t mask;
1509 	int error;
1510 
1511 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1512 	if (error)
1513 		return (error);
1514 	return (kern_sigsuspend(td, mask));
1515 }
1516 
1517 int
1518 kern_sigsuspend(struct thread *td, sigset_t mask)
1519 {
1520 	struct proc *p = td->td_proc;
1521 	int has_sig, sig;
1522 
1523 	/*
1524 	 * When returning from sigsuspend, we want
1525 	 * the old mask to be restored after the
1526 	 * signal handler has finished.  Thus, we
1527 	 * save it here and mark the sigacts structure
1528 	 * to indicate this.
1529 	 */
1530 	PROC_LOCK(p);
1531 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1532 	    SIGPROCMASK_PROC_LOCKED);
1533 	td->td_pflags |= TDP_OLDMASK;
1534 
1535 	/*
1536 	 * Process signals now. Otherwise, we can get spurious wakeup
1537 	 * due to signal entered process queue, but delivered to other
1538 	 * thread. But sigsuspend should return only on signal
1539 	 * delivery.
1540 	 */
1541 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1542 	for (has_sig = 0; !has_sig;) {
1543 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1544 			0) == 0)
1545 			/* void */;
1546 		thread_suspend_check(0);
1547 		mtx_lock(&p->p_sigacts->ps_mtx);
1548 		while ((sig = cursig(td)) != 0) {
1549 			KASSERT(sig >= 0, ("sig %d", sig));
1550 			has_sig += postsig(sig);
1551 		}
1552 		mtx_unlock(&p->p_sigacts->ps_mtx);
1553 
1554 		/*
1555 		 * If PTRACE_SCE or PTRACE_SCX were set after
1556 		 * userspace entered the syscall, return spurious
1557 		 * EINTR.
1558 		 */
1559 		if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1560 			has_sig += 1;
1561 	}
1562 	PROC_UNLOCK(p);
1563 	td->td_errno = EINTR;
1564 	td->td_pflags |= TDP_NERRNO;
1565 	return (EJUSTRETURN);
1566 }
1567 
1568 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1569 /*
1570  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1571  * convention: libc stub passes mask, not pointer, to save a copyin.
1572  */
1573 #ifndef _SYS_SYSPROTO_H_
1574 struct osigsuspend_args {
1575 	osigset_t mask;
1576 };
1577 #endif
1578 /* ARGSUSED */
1579 int
1580 osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1581 {
1582 	sigset_t mask;
1583 
1584 	OSIG2SIG(uap->mask, mask);
1585 	return (kern_sigsuspend(td, mask));
1586 }
1587 #endif /* COMPAT_43 */
1588 
1589 #if defined(COMPAT_43)
1590 #ifndef _SYS_SYSPROTO_H_
1591 struct osigstack_args {
1592 	struct	sigstack *nss;
1593 	struct	sigstack *oss;
1594 };
1595 #endif
1596 /* ARGSUSED */
1597 int
1598 osigstack(struct thread *td, struct osigstack_args *uap)
1599 {
1600 	struct sigstack nss, oss;
1601 	int error = 0;
1602 
1603 	if (uap->nss != NULL) {
1604 		error = copyin(uap->nss, &nss, sizeof(nss));
1605 		if (error)
1606 			return (error);
1607 	}
1608 	oss.ss_sp = td->td_sigstk.ss_sp;
1609 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1610 	if (uap->nss != NULL) {
1611 		td->td_sigstk.ss_sp = nss.ss_sp;
1612 		td->td_sigstk.ss_size = 0;
1613 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1614 		td->td_pflags |= TDP_ALTSTACK;
1615 	}
1616 	if (uap->oss != NULL)
1617 		error = copyout(&oss, uap->oss, sizeof(oss));
1618 
1619 	return (error);
1620 }
1621 #endif /* COMPAT_43 */
1622 
1623 #ifndef _SYS_SYSPROTO_H_
1624 struct sigaltstack_args {
1625 	stack_t	*ss;
1626 	stack_t	*oss;
1627 };
1628 #endif
1629 /* ARGSUSED */
1630 int
1631 sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1632 {
1633 	stack_t ss, oss;
1634 	int error;
1635 
1636 	if (uap->ss != NULL) {
1637 		error = copyin(uap->ss, &ss, sizeof(ss));
1638 		if (error)
1639 			return (error);
1640 	}
1641 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1642 	    (uap->oss != NULL) ? &oss : NULL);
1643 	if (error)
1644 		return (error);
1645 	if (uap->oss != NULL)
1646 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1647 	return (error);
1648 }
1649 
1650 int
1651 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1652 {
1653 	struct proc *p = td->td_proc;
1654 	int oonstack;
1655 
1656 	oonstack = sigonstack(cpu_getstack(td));
1657 
1658 	if (oss != NULL) {
1659 		*oss = td->td_sigstk;
1660 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1661 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1662 	}
1663 
1664 	if (ss != NULL) {
1665 		if (oonstack)
1666 			return (EPERM);
1667 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1668 			return (EINVAL);
1669 		if (!(ss->ss_flags & SS_DISABLE)) {
1670 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1671 				return (ENOMEM);
1672 
1673 			td->td_sigstk = *ss;
1674 			td->td_pflags |= TDP_ALTSTACK;
1675 		} else {
1676 			td->td_pflags &= ~TDP_ALTSTACK;
1677 		}
1678 	}
1679 	return (0);
1680 }
1681 
1682 /*
1683  * Common code for kill process group/broadcast kill.
1684  * cp is calling process.
1685  */
1686 static int
1687 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1688 {
1689 	struct proc *p;
1690 	struct pgrp *pgrp;
1691 	int err;
1692 	int ret;
1693 
1694 	ret = ESRCH;
1695 	if (all) {
1696 		/*
1697 		 * broadcast
1698 		 */
1699 		sx_slock(&allproc_lock);
1700 		FOREACH_PROC_IN_SYSTEM(p) {
1701 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1702 			    p == td->td_proc || p->p_state == PRS_NEW) {
1703 				continue;
1704 			}
1705 			PROC_LOCK(p);
1706 			err = p_cansignal(td, p, sig);
1707 			if (err == 0) {
1708 				if (sig)
1709 					pksignal(p, sig, ksi);
1710 				ret = err;
1711 			}
1712 			else if (ret == ESRCH)
1713 				ret = err;
1714 			PROC_UNLOCK(p);
1715 		}
1716 		sx_sunlock(&allproc_lock);
1717 	} else {
1718 		sx_slock(&proctree_lock);
1719 		if (pgid == 0) {
1720 			/*
1721 			 * zero pgid means send to my process group.
1722 			 */
1723 			pgrp = td->td_proc->p_pgrp;
1724 			PGRP_LOCK(pgrp);
1725 		} else {
1726 			pgrp = pgfind(pgid);
1727 			if (pgrp == NULL) {
1728 				sx_sunlock(&proctree_lock);
1729 				return (ESRCH);
1730 			}
1731 		}
1732 		sx_sunlock(&proctree_lock);
1733 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1734 			PROC_LOCK(p);
1735 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1736 			    p->p_state == PRS_NEW) {
1737 				PROC_UNLOCK(p);
1738 				continue;
1739 			}
1740 			err = p_cansignal(td, p, sig);
1741 			if (err == 0) {
1742 				if (sig)
1743 					pksignal(p, sig, ksi);
1744 				ret = err;
1745 			}
1746 			else if (ret == ESRCH)
1747 				ret = err;
1748 			PROC_UNLOCK(p);
1749 		}
1750 		PGRP_UNLOCK(pgrp);
1751 	}
1752 	return (ret);
1753 }
1754 
1755 #ifndef _SYS_SYSPROTO_H_
1756 struct kill_args {
1757 	int	pid;
1758 	int	signum;
1759 };
1760 #endif
1761 /* ARGSUSED */
1762 int
1763 sys_kill(struct thread *td, struct kill_args *uap)
1764 {
1765 	ksiginfo_t ksi;
1766 	struct proc *p;
1767 	int error;
1768 
1769 	/*
1770 	 * A process in capability mode can send signals only to himself.
1771 	 * The main rationale behind this is that abort(3) is implemented as
1772 	 * kill(getpid(), SIGABRT).
1773 	 */
1774 	if (IN_CAPABILITY_MODE(td) && uap->pid != td->td_proc->p_pid)
1775 		return (ECAPMODE);
1776 
1777 	AUDIT_ARG_SIGNUM(uap->signum);
1778 	AUDIT_ARG_PID(uap->pid);
1779 	if ((u_int)uap->signum > _SIG_MAXSIG)
1780 		return (EINVAL);
1781 
1782 	ksiginfo_init(&ksi);
1783 	ksi.ksi_signo = uap->signum;
1784 	ksi.ksi_code = SI_USER;
1785 	ksi.ksi_pid = td->td_proc->p_pid;
1786 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1787 
1788 	if (uap->pid > 0) {
1789 		/* kill single process */
1790 		if ((p = pfind_any(uap->pid)) == NULL)
1791 			return (ESRCH);
1792 		AUDIT_ARG_PROCESS(p);
1793 		error = p_cansignal(td, p, uap->signum);
1794 		if (error == 0 && uap->signum)
1795 			pksignal(p, uap->signum, &ksi);
1796 		PROC_UNLOCK(p);
1797 		return (error);
1798 	}
1799 	switch (uap->pid) {
1800 	case -1:		/* broadcast signal */
1801 		return (killpg1(td, uap->signum, 0, 1, &ksi));
1802 	case 0:			/* signal own process group */
1803 		return (killpg1(td, uap->signum, 0, 0, &ksi));
1804 	default:		/* negative explicit process group */
1805 		return (killpg1(td, uap->signum, -uap->pid, 0, &ksi));
1806 	}
1807 	/* NOTREACHED */
1808 }
1809 
1810 int
1811 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1812 {
1813 	struct proc *p;
1814 	int error;
1815 
1816 	AUDIT_ARG_SIGNUM(uap->signum);
1817 	AUDIT_ARG_FD(uap->fd);
1818 	if ((u_int)uap->signum > _SIG_MAXSIG)
1819 		return (EINVAL);
1820 
1821 	error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1822 	if (error)
1823 		return (error);
1824 	AUDIT_ARG_PROCESS(p);
1825 	error = p_cansignal(td, p, uap->signum);
1826 	if (error == 0 && uap->signum)
1827 		kern_psignal(p, uap->signum);
1828 	PROC_UNLOCK(p);
1829 	return (error);
1830 }
1831 
1832 #if defined(COMPAT_43)
1833 #ifndef _SYS_SYSPROTO_H_
1834 struct okillpg_args {
1835 	int	pgid;
1836 	int	signum;
1837 };
1838 #endif
1839 /* ARGSUSED */
1840 int
1841 okillpg(struct thread *td, struct okillpg_args *uap)
1842 {
1843 	ksiginfo_t ksi;
1844 
1845 	AUDIT_ARG_SIGNUM(uap->signum);
1846 	AUDIT_ARG_PID(uap->pgid);
1847 	if ((u_int)uap->signum > _SIG_MAXSIG)
1848 		return (EINVAL);
1849 
1850 	ksiginfo_init(&ksi);
1851 	ksi.ksi_signo = uap->signum;
1852 	ksi.ksi_code = SI_USER;
1853 	ksi.ksi_pid = td->td_proc->p_pid;
1854 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1855 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1856 }
1857 #endif /* COMPAT_43 */
1858 
1859 #ifndef _SYS_SYSPROTO_H_
1860 struct sigqueue_args {
1861 	pid_t pid;
1862 	int signum;
1863 	/* union sigval */ void *value;
1864 };
1865 #endif
1866 int
1867 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1868 {
1869 	union sigval sv;
1870 
1871 	sv.sival_ptr = uap->value;
1872 
1873 	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
1874 }
1875 
1876 int
1877 kern_sigqueue(struct thread *td, pid_t pid, int signum, union sigval *value)
1878 {
1879 	ksiginfo_t ksi;
1880 	struct proc *p;
1881 	int error;
1882 
1883 	if ((u_int)signum > _SIG_MAXSIG)
1884 		return (EINVAL);
1885 
1886 	/*
1887 	 * Specification says sigqueue can only send signal to
1888 	 * single process.
1889 	 */
1890 	if (pid <= 0)
1891 		return (EINVAL);
1892 
1893 	if ((p = pfind_any(pid)) == NULL)
1894 		return (ESRCH);
1895 	error = p_cansignal(td, p, signum);
1896 	if (error == 0 && signum != 0) {
1897 		ksiginfo_init(&ksi);
1898 		ksi.ksi_flags = KSI_SIGQ;
1899 		ksi.ksi_signo = signum;
1900 		ksi.ksi_code = SI_QUEUE;
1901 		ksi.ksi_pid = td->td_proc->p_pid;
1902 		ksi.ksi_uid = td->td_ucred->cr_ruid;
1903 		ksi.ksi_value = *value;
1904 		error = pksignal(p, ksi.ksi_signo, &ksi);
1905 	}
1906 	PROC_UNLOCK(p);
1907 	return (error);
1908 }
1909 
1910 /*
1911  * Send a signal to a process group.
1912  */
1913 void
1914 gsignal(int pgid, int sig, ksiginfo_t *ksi)
1915 {
1916 	struct pgrp *pgrp;
1917 
1918 	if (pgid != 0) {
1919 		sx_slock(&proctree_lock);
1920 		pgrp = pgfind(pgid);
1921 		sx_sunlock(&proctree_lock);
1922 		if (pgrp != NULL) {
1923 			pgsignal(pgrp, sig, 0, ksi);
1924 			PGRP_UNLOCK(pgrp);
1925 		}
1926 	}
1927 }
1928 
1929 /*
1930  * Send a signal to a process group.  If checktty is 1,
1931  * limit to members which have a controlling terminal.
1932  */
1933 void
1934 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
1935 {
1936 	struct proc *p;
1937 
1938 	if (pgrp) {
1939 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
1940 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1941 			PROC_LOCK(p);
1942 			if (p->p_state == PRS_NORMAL &&
1943 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
1944 				pksignal(p, sig, ksi);
1945 			PROC_UNLOCK(p);
1946 		}
1947 	}
1948 }
1949 
1950 
1951 /*
1952  * Recalculate the signal mask and reset the signal disposition after
1953  * usermode frame for delivery is formed.  Should be called after
1954  * mach-specific routine, because sysent->sv_sendsig() needs correct
1955  * ps_siginfo and signal mask.
1956  */
1957 static void
1958 postsig_done(int sig, struct thread *td, struct sigacts *ps)
1959 {
1960 	sigset_t mask;
1961 
1962 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1963 	td->td_ru.ru_nsignals++;
1964 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
1965 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
1966 		SIGADDSET(mask, sig);
1967 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
1968 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
1969 	if (SIGISMEMBER(ps->ps_sigreset, sig))
1970 		sigdflt(ps, sig);
1971 }
1972 
1973 
1974 /*
1975  * Send a signal caused by a trap to the current thread.  If it will be
1976  * caught immediately, deliver it with correct code.  Otherwise, post it
1977  * normally.
1978  */
1979 void
1980 trapsignal(struct thread *td, ksiginfo_t *ksi)
1981 {
1982 	struct sigacts *ps;
1983 	struct proc *p;
1984 	int sig;
1985 	int code;
1986 
1987 	p = td->td_proc;
1988 	sig = ksi->ksi_signo;
1989 	code = ksi->ksi_code;
1990 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
1991 
1992 	PROC_LOCK(p);
1993 	ps = p->p_sigacts;
1994 	mtx_lock(&ps->ps_mtx);
1995 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
1996 	    !SIGISMEMBER(td->td_sigmask, sig)) {
1997 #ifdef KTRACE
1998 		if (KTRPOINT(curthread, KTR_PSIG))
1999 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2000 			    &td->td_sigmask, code);
2001 #endif
2002 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2003 				ksi, &td->td_sigmask);
2004 		postsig_done(sig, td, ps);
2005 		mtx_unlock(&ps->ps_mtx);
2006 	} else {
2007 		/*
2008 		 * Avoid a possible infinite loop if the thread
2009 		 * masking the signal or process is ignoring the
2010 		 * signal.
2011 		 */
2012 		if (kern_forcesigexit &&
2013 		    (SIGISMEMBER(td->td_sigmask, sig) ||
2014 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2015 			SIGDELSET(td->td_sigmask, sig);
2016 			SIGDELSET(ps->ps_sigcatch, sig);
2017 			SIGDELSET(ps->ps_sigignore, sig);
2018 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2019 		}
2020 		mtx_unlock(&ps->ps_mtx);
2021 		p->p_sig = sig;		/* XXX to verify code */
2022 		tdsendsignal(p, td, sig, ksi);
2023 	}
2024 	PROC_UNLOCK(p);
2025 }
2026 
2027 static struct thread *
2028 sigtd(struct proc *p, int sig, int prop)
2029 {
2030 	struct thread *td, *signal_td;
2031 
2032 	PROC_LOCK_ASSERT(p, MA_OWNED);
2033 
2034 	/*
2035 	 * Check if current thread can handle the signal without
2036 	 * switching context to another thread.
2037 	 */
2038 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig))
2039 		return (curthread);
2040 	signal_td = NULL;
2041 	FOREACH_THREAD_IN_PROC(p, td) {
2042 		if (!SIGISMEMBER(td->td_sigmask, sig)) {
2043 			signal_td = td;
2044 			break;
2045 		}
2046 	}
2047 	if (signal_td == NULL)
2048 		signal_td = FIRST_THREAD_IN_PROC(p);
2049 	return (signal_td);
2050 }
2051 
2052 /*
2053  * Send the signal to the process.  If the signal has an action, the action
2054  * is usually performed by the target process rather than the caller; we add
2055  * the signal to the set of pending signals for the process.
2056  *
2057  * Exceptions:
2058  *   o When a stop signal is sent to a sleeping process that takes the
2059  *     default action, the process is stopped without awakening it.
2060  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2061  *     regardless of the signal action (eg, blocked or ignored).
2062  *
2063  * Other ignored signals are discarded immediately.
2064  *
2065  * NB: This function may be entered from the debugger via the "kill" DDB
2066  * command.  There is little that can be done to mitigate the possibly messy
2067  * side effects of this unwise possibility.
2068  */
2069 void
2070 kern_psignal(struct proc *p, int sig)
2071 {
2072 	ksiginfo_t ksi;
2073 
2074 	ksiginfo_init(&ksi);
2075 	ksi.ksi_signo = sig;
2076 	ksi.ksi_code = SI_KERNEL;
2077 	(void) tdsendsignal(p, NULL, sig, &ksi);
2078 }
2079 
2080 int
2081 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2082 {
2083 
2084 	return (tdsendsignal(p, NULL, sig, ksi));
2085 }
2086 
2087 /* Utility function for finding a thread to send signal event to. */
2088 int
2089 sigev_findtd(struct proc *p ,struct sigevent *sigev, struct thread **ttd)
2090 {
2091 	struct thread *td;
2092 
2093 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2094 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2095 		if (td == NULL)
2096 			return (ESRCH);
2097 		*ttd = td;
2098 	} else {
2099 		*ttd = NULL;
2100 		PROC_LOCK(p);
2101 	}
2102 	return (0);
2103 }
2104 
2105 void
2106 tdsignal(struct thread *td, int sig)
2107 {
2108 	ksiginfo_t ksi;
2109 
2110 	ksiginfo_init(&ksi);
2111 	ksi.ksi_signo = sig;
2112 	ksi.ksi_code = SI_KERNEL;
2113 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2114 }
2115 
2116 void
2117 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2118 {
2119 
2120 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2121 }
2122 
2123 int
2124 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2125 {
2126 	sig_t action;
2127 	sigqueue_t *sigqueue;
2128 	int prop;
2129 	struct sigacts *ps;
2130 	int intrval;
2131 	int ret = 0;
2132 	int wakeup_swapper;
2133 
2134 	MPASS(td == NULL || p == td->td_proc);
2135 	PROC_LOCK_ASSERT(p, MA_OWNED);
2136 
2137 	if (!_SIG_VALID(sig))
2138 		panic("%s(): invalid signal %d", __func__, sig);
2139 
2140 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2141 
2142 	/*
2143 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2144 	 */
2145 	if (p->p_state == PRS_ZOMBIE) {
2146 		if (ksi && (ksi->ksi_flags & KSI_INS))
2147 			ksiginfo_tryfree(ksi);
2148 		return (ret);
2149 	}
2150 
2151 	ps = p->p_sigacts;
2152 	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2153 	prop = sigprop(sig);
2154 
2155 	if (td == NULL) {
2156 		td = sigtd(p, sig, prop);
2157 		sigqueue = &p->p_sigqueue;
2158 	} else
2159 		sigqueue = &td->td_sigqueue;
2160 
2161 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2162 
2163 	/*
2164 	 * If the signal is being ignored,
2165 	 * then we forget about it immediately.
2166 	 * (Note: we don't set SIGCONT in ps_sigignore,
2167 	 * and if it is set to SIG_IGN,
2168 	 * action will be SIG_DFL here.)
2169 	 */
2170 	mtx_lock(&ps->ps_mtx);
2171 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2172 		SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2173 
2174 		mtx_unlock(&ps->ps_mtx);
2175 		if (ksi && (ksi->ksi_flags & KSI_INS))
2176 			ksiginfo_tryfree(ksi);
2177 		return (ret);
2178 	}
2179 	if (SIGISMEMBER(td->td_sigmask, sig))
2180 		action = SIG_HOLD;
2181 	else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2182 		action = SIG_CATCH;
2183 	else
2184 		action = SIG_DFL;
2185 	if (SIGISMEMBER(ps->ps_sigintr, sig))
2186 		intrval = EINTR;
2187 	else
2188 		intrval = ERESTART;
2189 	mtx_unlock(&ps->ps_mtx);
2190 
2191 	if (prop & SIGPROP_CONT)
2192 		sigqueue_delete_stopmask_proc(p);
2193 	else if (prop & SIGPROP_STOP) {
2194 		/*
2195 		 * If sending a tty stop signal to a member of an orphaned
2196 		 * process group, discard the signal here if the action
2197 		 * is default; don't stop the process below if sleeping,
2198 		 * and don't clear any pending SIGCONT.
2199 		 */
2200 		if ((prop & SIGPROP_TTYSTOP) &&
2201 		    (p->p_pgrp->pg_jobc == 0) &&
2202 		    (action == SIG_DFL)) {
2203 			if (ksi && (ksi->ksi_flags & KSI_INS))
2204 				ksiginfo_tryfree(ksi);
2205 			return (ret);
2206 		}
2207 		sigqueue_delete_proc(p, SIGCONT);
2208 		if (p->p_flag & P_CONTINUED) {
2209 			p->p_flag &= ~P_CONTINUED;
2210 			PROC_LOCK(p->p_pptr);
2211 			sigqueue_take(p->p_ksi);
2212 			PROC_UNLOCK(p->p_pptr);
2213 		}
2214 	}
2215 
2216 	ret = sigqueue_add(sigqueue, sig, ksi);
2217 	if (ret != 0)
2218 		return (ret);
2219 	signotify(td);
2220 	/*
2221 	 * Defer further processing for signals which are held,
2222 	 * except that stopped processes must be continued by SIGCONT.
2223 	 */
2224 	if (action == SIG_HOLD &&
2225 	    !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2226 		return (ret);
2227 
2228 	/* SIGKILL: Remove procfs STOPEVENTs. */
2229 	if (sig == SIGKILL) {
2230 		/* from procfs_ioctl.c: PIOCBIC */
2231 		p->p_stops = 0;
2232 		/* from procfs_ioctl.c: PIOCCONT */
2233 		p->p_step = 0;
2234 		wakeup(&p->p_step);
2235 	}
2236 	/*
2237 	 * Some signals have a process-wide effect and a per-thread
2238 	 * component.  Most processing occurs when the process next
2239 	 * tries to cross the user boundary, however there are some
2240 	 * times when processing needs to be done immediately, such as
2241 	 * waking up threads so that they can cross the user boundary.
2242 	 * We try to do the per-process part here.
2243 	 */
2244 	if (P_SHOULDSTOP(p)) {
2245 		KASSERT(!(p->p_flag & P_WEXIT),
2246 		    ("signal to stopped but exiting process"));
2247 		if (sig == SIGKILL) {
2248 			/*
2249 			 * If traced process is already stopped,
2250 			 * then no further action is necessary.
2251 			 */
2252 			if (p->p_flag & P_TRACED)
2253 				goto out;
2254 			/*
2255 			 * SIGKILL sets process running.
2256 			 * It will die elsewhere.
2257 			 * All threads must be restarted.
2258 			 */
2259 			p->p_flag &= ~P_STOPPED_SIG;
2260 			goto runfast;
2261 		}
2262 
2263 		if (prop & SIGPROP_CONT) {
2264 			/*
2265 			 * If traced process is already stopped,
2266 			 * then no further action is necessary.
2267 			 */
2268 			if (p->p_flag & P_TRACED)
2269 				goto out;
2270 			/*
2271 			 * If SIGCONT is default (or ignored), we continue the
2272 			 * process but don't leave the signal in sigqueue as
2273 			 * it has no further action.  If SIGCONT is held, we
2274 			 * continue the process and leave the signal in
2275 			 * sigqueue.  If the process catches SIGCONT, let it
2276 			 * handle the signal itself.  If it isn't waiting on
2277 			 * an event, it goes back to run state.
2278 			 * Otherwise, process goes back to sleep state.
2279 			 */
2280 			p->p_flag &= ~P_STOPPED_SIG;
2281 			PROC_SLOCK(p);
2282 			if (p->p_numthreads == p->p_suspcount) {
2283 				PROC_SUNLOCK(p);
2284 				p->p_flag |= P_CONTINUED;
2285 				p->p_xsig = SIGCONT;
2286 				PROC_LOCK(p->p_pptr);
2287 				childproc_continued(p);
2288 				PROC_UNLOCK(p->p_pptr);
2289 				PROC_SLOCK(p);
2290 			}
2291 			if (action == SIG_DFL) {
2292 				thread_unsuspend(p);
2293 				PROC_SUNLOCK(p);
2294 				sigqueue_delete(sigqueue, sig);
2295 				goto out;
2296 			}
2297 			if (action == SIG_CATCH) {
2298 				/*
2299 				 * The process wants to catch it so it needs
2300 				 * to run at least one thread, but which one?
2301 				 */
2302 				PROC_SUNLOCK(p);
2303 				goto runfast;
2304 			}
2305 			/*
2306 			 * The signal is not ignored or caught.
2307 			 */
2308 			thread_unsuspend(p);
2309 			PROC_SUNLOCK(p);
2310 			goto out;
2311 		}
2312 
2313 		if (prop & SIGPROP_STOP) {
2314 			/*
2315 			 * If traced process is already stopped,
2316 			 * then no further action is necessary.
2317 			 */
2318 			if (p->p_flag & P_TRACED)
2319 				goto out;
2320 			/*
2321 			 * Already stopped, don't need to stop again
2322 			 * (If we did the shell could get confused).
2323 			 * Just make sure the signal STOP bit set.
2324 			 */
2325 			p->p_flag |= P_STOPPED_SIG;
2326 			sigqueue_delete(sigqueue, sig);
2327 			goto out;
2328 		}
2329 
2330 		/*
2331 		 * All other kinds of signals:
2332 		 * If a thread is sleeping interruptibly, simulate a
2333 		 * wakeup so that when it is continued it will be made
2334 		 * runnable and can look at the signal.  However, don't make
2335 		 * the PROCESS runnable, leave it stopped.
2336 		 * It may run a bit until it hits a thread_suspend_check().
2337 		 */
2338 		wakeup_swapper = 0;
2339 		PROC_SLOCK(p);
2340 		thread_lock(td);
2341 		if (TD_ON_SLEEPQ(td) && (td->td_flags & TDF_SINTR))
2342 			wakeup_swapper = sleepq_abort(td, intrval);
2343 		thread_unlock(td);
2344 		PROC_SUNLOCK(p);
2345 		if (wakeup_swapper)
2346 			kick_proc0();
2347 		goto out;
2348 		/*
2349 		 * Mutexes are short lived. Threads waiting on them will
2350 		 * hit thread_suspend_check() soon.
2351 		 */
2352 	} else if (p->p_state == PRS_NORMAL) {
2353 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2354 			tdsigwakeup(td, sig, action, intrval);
2355 			goto out;
2356 		}
2357 
2358 		MPASS(action == SIG_DFL);
2359 
2360 		if (prop & SIGPROP_STOP) {
2361 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2362 				goto out;
2363 			p->p_flag |= P_STOPPED_SIG;
2364 			p->p_xsig = sig;
2365 			PROC_SLOCK(p);
2366 			wakeup_swapper = sig_suspend_threads(td, p, 1);
2367 			if (p->p_numthreads == p->p_suspcount) {
2368 				/*
2369 				 * only thread sending signal to another
2370 				 * process can reach here, if thread is sending
2371 				 * signal to its process, because thread does
2372 				 * not suspend itself here, p_numthreads
2373 				 * should never be equal to p_suspcount.
2374 				 */
2375 				thread_stopped(p);
2376 				PROC_SUNLOCK(p);
2377 				sigqueue_delete_proc(p, p->p_xsig);
2378 			} else
2379 				PROC_SUNLOCK(p);
2380 			if (wakeup_swapper)
2381 				kick_proc0();
2382 			goto out;
2383 		}
2384 	} else {
2385 		/* Not in "NORMAL" state. discard the signal. */
2386 		sigqueue_delete(sigqueue, sig);
2387 		goto out;
2388 	}
2389 
2390 	/*
2391 	 * The process is not stopped so we need to apply the signal to all the
2392 	 * running threads.
2393 	 */
2394 runfast:
2395 	tdsigwakeup(td, sig, action, intrval);
2396 	PROC_SLOCK(p);
2397 	thread_unsuspend(p);
2398 	PROC_SUNLOCK(p);
2399 out:
2400 	/* If we jump here, proc slock should not be owned. */
2401 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2402 	return (ret);
2403 }
2404 
2405 /*
2406  * The force of a signal has been directed against a single
2407  * thread.  We need to see what we can do about knocking it
2408  * out of any sleep it may be in etc.
2409  */
2410 static void
2411 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2412 {
2413 	struct proc *p = td->td_proc;
2414 	int prop;
2415 	int wakeup_swapper;
2416 
2417 	wakeup_swapper = 0;
2418 	PROC_LOCK_ASSERT(p, MA_OWNED);
2419 	prop = sigprop(sig);
2420 
2421 	PROC_SLOCK(p);
2422 	thread_lock(td);
2423 	/*
2424 	 * Bring the priority of a thread up if we want it to get
2425 	 * killed in this lifetime.  Be careful to avoid bumping the
2426 	 * priority of the idle thread, since we still allow to signal
2427 	 * kernel processes.
2428 	 */
2429 	if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2430 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2431 		sched_prio(td, PUSER);
2432 	if (TD_ON_SLEEPQ(td)) {
2433 		/*
2434 		 * If thread is sleeping uninterruptibly
2435 		 * we can't interrupt the sleep... the signal will
2436 		 * be noticed when the process returns through
2437 		 * trap() or syscall().
2438 		 */
2439 		if ((td->td_flags & TDF_SINTR) == 0)
2440 			goto out;
2441 		/*
2442 		 * If SIGCONT is default (or ignored) and process is
2443 		 * asleep, we are finished; the process should not
2444 		 * be awakened.
2445 		 */
2446 		if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2447 			thread_unlock(td);
2448 			PROC_SUNLOCK(p);
2449 			sigqueue_delete(&p->p_sigqueue, sig);
2450 			/*
2451 			 * It may be on either list in this state.
2452 			 * Remove from both for now.
2453 			 */
2454 			sigqueue_delete(&td->td_sigqueue, sig);
2455 			return;
2456 		}
2457 
2458 		/*
2459 		 * Don't awaken a sleeping thread for SIGSTOP if the
2460 		 * STOP signal is deferred.
2461 		 */
2462 		if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2463 		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2464 			goto out;
2465 
2466 		/*
2467 		 * Give low priority threads a better chance to run.
2468 		 */
2469 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2470 			sched_prio(td, PUSER);
2471 
2472 		wakeup_swapper = sleepq_abort(td, intrval);
2473 	} else {
2474 		/*
2475 		 * Other states do nothing with the signal immediately,
2476 		 * other than kicking ourselves if we are running.
2477 		 * It will either never be noticed, or noticed very soon.
2478 		 */
2479 #ifdef SMP
2480 		if (TD_IS_RUNNING(td) && td != curthread)
2481 			forward_signal(td);
2482 #endif
2483 	}
2484 out:
2485 	PROC_SUNLOCK(p);
2486 	thread_unlock(td);
2487 	if (wakeup_swapper)
2488 		kick_proc0();
2489 }
2490 
2491 static int
2492 sig_suspend_threads(struct thread *td, struct proc *p, int sending)
2493 {
2494 	struct thread *td2;
2495 	int wakeup_swapper;
2496 
2497 	PROC_LOCK_ASSERT(p, MA_OWNED);
2498 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2499 	MPASS(sending || td == curthread);
2500 
2501 	wakeup_swapper = 0;
2502 	FOREACH_THREAD_IN_PROC(p, td2) {
2503 		thread_lock(td2);
2504 		td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
2505 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2506 		    (td2->td_flags & TDF_SINTR)) {
2507 			if (td2->td_flags & TDF_SBDRY) {
2508 				/*
2509 				 * Once a thread is asleep with
2510 				 * TDF_SBDRY and without TDF_SERESTART
2511 				 * or TDF_SEINTR set, it should never
2512 				 * become suspended due to this check.
2513 				 */
2514 				KASSERT(!TD_IS_SUSPENDED(td2),
2515 				    ("thread with deferred stops suspended"));
2516 				if (TD_SBDRY_INTR(td2))
2517 					wakeup_swapper |= sleepq_abort(td2,
2518 					    TD_SBDRY_ERRNO(td2));
2519 			} else if (!TD_IS_SUSPENDED(td2)) {
2520 				thread_suspend_one(td2);
2521 			}
2522 		} else if (!TD_IS_SUSPENDED(td2)) {
2523 			if (sending || td != td2)
2524 				td2->td_flags |= TDF_ASTPENDING;
2525 #ifdef SMP
2526 			if (TD_IS_RUNNING(td2) && td2 != td)
2527 				forward_signal(td2);
2528 #endif
2529 		}
2530 		thread_unlock(td2);
2531 	}
2532 	return (wakeup_swapper);
2533 }
2534 
2535 /*
2536  * Stop the process for an event deemed interesting to the debugger. If si is
2537  * non-NULL, this is a signal exchange; the new signal requested by the
2538  * debugger will be returned for handling. If si is NULL, this is some other
2539  * type of interesting event. The debugger may request a signal be delivered in
2540  * that case as well, however it will be deferred until it can be handled.
2541  */
2542 int
2543 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2544 {
2545 	struct proc *p = td->td_proc;
2546 	struct thread *td2;
2547 	ksiginfo_t ksi;
2548 	int prop;
2549 
2550 	PROC_LOCK_ASSERT(p, MA_OWNED);
2551 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2552 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2553 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2554 
2555 	td->td_xsig = sig;
2556 
2557 	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2558 		td->td_dbgflags |= TDB_XSIG;
2559 		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2560 		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2561 		PROC_SLOCK(p);
2562 		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2563 			if (P_KILLED(p)) {
2564 				/*
2565 				 * Ensure that, if we've been PT_KILLed, the
2566 				 * exit status reflects that. Another thread
2567 				 * may also be in ptracestop(), having just
2568 				 * received the SIGKILL, but this thread was
2569 				 * unsuspended first.
2570 				 */
2571 				td->td_dbgflags &= ~TDB_XSIG;
2572 				td->td_xsig = SIGKILL;
2573 				p->p_ptevents = 0;
2574 				break;
2575 			}
2576 			if (p->p_flag & P_SINGLE_EXIT &&
2577 			    !(td->td_dbgflags & TDB_EXIT)) {
2578 				/*
2579 				 * Ignore ptrace stops except for thread exit
2580 				 * events when the process exits.
2581 				 */
2582 				td->td_dbgflags &= ~TDB_XSIG;
2583 				PROC_SUNLOCK(p);
2584 				return (0);
2585 			}
2586 
2587 			/*
2588 			 * Make wait(2) work.  Ensure that right after the
2589 			 * attach, the thread which was decided to become the
2590 			 * leader of attach gets reported to the waiter.
2591 			 * Otherwise, just avoid overwriting another thread's
2592 			 * assignment to p_xthread.  If another thread has
2593 			 * already set p_xthread, the current thread will get
2594 			 * a chance to report itself upon the next iteration.
2595 			 */
2596 			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2597 			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2598 			    p->p_xthread == NULL)) {
2599 				p->p_xsig = sig;
2600 				p->p_xthread = td;
2601 
2602 				/*
2603 				 * If we are on sleepqueue already,
2604 				 * let sleepqueue code decide if it
2605 				 * needs to go sleep after attach.
2606 				 */
2607 				if (td->td_wchan == NULL)
2608 					td->td_dbgflags &= ~TDB_FSTP;
2609 
2610 				p->p_flag2 &= ~P2_PTRACE_FSTP;
2611 				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2612 				sig_suspend_threads(td, p, 0);
2613 			}
2614 			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2615 				td->td_dbgflags &= ~TDB_STOPATFORK;
2616 			}
2617 stopme:
2618 			thread_suspend_switch(td, p);
2619 			if (p->p_xthread == td)
2620 				p->p_xthread = NULL;
2621 			if (!(p->p_flag & P_TRACED))
2622 				break;
2623 			if (td->td_dbgflags & TDB_SUSPEND) {
2624 				if (p->p_flag & P_SINGLE_EXIT)
2625 					break;
2626 				goto stopme;
2627 			}
2628 		}
2629 		PROC_SUNLOCK(p);
2630 	}
2631 
2632 	if (si != NULL && sig == td->td_xsig) {
2633 		/* Parent wants us to take the original signal unchanged. */
2634 		si->ksi_flags |= KSI_HEAD;
2635 		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2636 			si->ksi_signo = 0;
2637 	} else if (td->td_xsig != 0) {
2638 		/*
2639 		 * If parent wants us to take a new signal, then it will leave
2640 		 * it in td->td_xsig; otherwise we just look for signals again.
2641 		 */
2642 		ksiginfo_init(&ksi);
2643 		ksi.ksi_signo = td->td_xsig;
2644 		ksi.ksi_flags |= KSI_PTRACE;
2645 		prop = sigprop(td->td_xsig);
2646 		td2 = sigtd(p, td->td_xsig, prop);
2647 		tdsendsignal(p, td2, td->td_xsig, &ksi);
2648 		if (td != td2)
2649 			return (0);
2650 	}
2651 
2652 	return (td->td_xsig);
2653 }
2654 
2655 static void
2656 reschedule_signals(struct proc *p, sigset_t block, int flags)
2657 {
2658 	struct sigacts *ps;
2659 	struct thread *td;
2660 	int sig;
2661 
2662 	PROC_LOCK_ASSERT(p, MA_OWNED);
2663 	ps = p->p_sigacts;
2664 	mtx_assert(&ps->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0 ?
2665 	    MA_OWNED : MA_NOTOWNED);
2666 	if (SIGISEMPTY(p->p_siglist))
2667 		return;
2668 	SIGSETAND(block, p->p_siglist);
2669 	while ((sig = sig_ffs(&block)) != 0) {
2670 		SIGDELSET(block, sig);
2671 		td = sigtd(p, sig, 0);
2672 		signotify(td);
2673 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2674 			mtx_lock(&ps->ps_mtx);
2675 		if (p->p_flag & P_TRACED ||
2676 		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2677 		    !SIGISMEMBER(td->td_sigmask, sig)))
2678 			tdsigwakeup(td, sig, SIG_CATCH,
2679 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2680 			     ERESTART));
2681 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2682 			mtx_unlock(&ps->ps_mtx);
2683 	}
2684 }
2685 
2686 void
2687 tdsigcleanup(struct thread *td)
2688 {
2689 	struct proc *p;
2690 	sigset_t unblocked;
2691 
2692 	p = td->td_proc;
2693 	PROC_LOCK_ASSERT(p, MA_OWNED);
2694 
2695 	sigqueue_flush(&td->td_sigqueue);
2696 	if (p->p_numthreads == 1)
2697 		return;
2698 
2699 	/*
2700 	 * Since we cannot handle signals, notify signal post code
2701 	 * about this by filling the sigmask.
2702 	 *
2703 	 * Also, if needed, wake up thread(s) that do not block the
2704 	 * same signals as the exiting thread, since the thread might
2705 	 * have been selected for delivery and woken up.
2706 	 */
2707 	SIGFILLSET(unblocked);
2708 	SIGSETNAND(unblocked, td->td_sigmask);
2709 	SIGFILLSET(td->td_sigmask);
2710 	reschedule_signals(p, unblocked, 0);
2711 
2712 }
2713 
2714 static int
2715 sigdeferstop_curr_flags(int cflags)
2716 {
2717 
2718 	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
2719 	    (cflags & TDF_SBDRY) != 0);
2720 	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
2721 }
2722 
2723 /*
2724  * Defer the delivery of SIGSTOP for the current thread, according to
2725  * the requested mode.  Returns previous flags, which must be restored
2726  * by sigallowstop().
2727  *
2728  * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
2729  * cleared by the current thread, which allow the lock-less read-only
2730  * accesses below.
2731  */
2732 int
2733 sigdeferstop_impl(int mode)
2734 {
2735 	struct thread *td;
2736 	int cflags, nflags;
2737 
2738 	td = curthread;
2739 	cflags = sigdeferstop_curr_flags(td->td_flags);
2740 	switch (mode) {
2741 	case SIGDEFERSTOP_NOP:
2742 		nflags = cflags;
2743 		break;
2744 	case SIGDEFERSTOP_OFF:
2745 		nflags = 0;
2746 		break;
2747 	case SIGDEFERSTOP_SILENT:
2748 		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
2749 		break;
2750 	case SIGDEFERSTOP_EINTR:
2751 		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
2752 		break;
2753 	case SIGDEFERSTOP_ERESTART:
2754 		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
2755 		break;
2756 	default:
2757 		panic("sigdeferstop: invalid mode %x", mode);
2758 		break;
2759 	}
2760 	if (cflags == nflags)
2761 		return (SIGDEFERSTOP_VAL_NCHG);
2762 	thread_lock(td);
2763 	td->td_flags = (td->td_flags & ~cflags) | nflags;
2764 	thread_unlock(td);
2765 	return (cflags);
2766 }
2767 
2768 /*
2769  * Restores the STOP handling mode, typically permitting the delivery
2770  * of SIGSTOP for the current thread.  This does not immediately
2771  * suspend if a stop was posted.  Instead, the thread will suspend
2772  * either via ast() or a subsequent interruptible sleep.
2773  */
2774 void
2775 sigallowstop_impl(int prev)
2776 {
2777 	struct thread *td;
2778 	int cflags;
2779 
2780 	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
2781 	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
2782 	    ("sigallowstop: incorrect previous mode %x", prev));
2783 	td = curthread;
2784 	cflags = sigdeferstop_curr_flags(td->td_flags);
2785 	if (cflags != prev) {
2786 		thread_lock(td);
2787 		td->td_flags = (td->td_flags & ~cflags) | prev;
2788 		thread_unlock(td);
2789 	}
2790 }
2791 
2792 /*
2793  * If the current process has received a signal (should be caught or cause
2794  * termination, should interrupt current syscall), return the signal number.
2795  * Stop signals with default action are processed immediately, then cleared;
2796  * they aren't returned.  This is checked after each entry to the system for
2797  * a syscall or trap (though this can usually be done without calling issignal
2798  * by checking the pending signal masks in cursig.) The normal call
2799  * sequence is
2800  *
2801  *	while (sig = cursig(curthread))
2802  *		postsig(sig);
2803  */
2804 static int
2805 issignal(struct thread *td)
2806 {
2807 	struct proc *p;
2808 	struct sigacts *ps;
2809 	struct sigqueue *queue;
2810 	sigset_t sigpending;
2811 	ksiginfo_t ksi;
2812 	int prop, sig, traced;
2813 
2814 	p = td->td_proc;
2815 	ps = p->p_sigacts;
2816 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2817 	PROC_LOCK_ASSERT(p, MA_OWNED);
2818 	for (;;) {
2819 		traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
2820 
2821 		sigpending = td->td_sigqueue.sq_signals;
2822 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
2823 		SIGSETNAND(sigpending, td->td_sigmask);
2824 
2825 		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
2826 		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2827 			SIG_STOPSIGMASK(sigpending);
2828 		if (SIGISEMPTY(sigpending))	/* no signal to send */
2829 			return (0);
2830 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
2831 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
2832 		    SIGISMEMBER(sigpending, SIGSTOP)) {
2833 			/*
2834 			 * If debugger just attached, always consume
2835 			 * SIGSTOP from ptrace(PT_ATTACH) first, to
2836 			 * execute the debugger attach ritual in
2837 			 * order.
2838 			 */
2839 			sig = SIGSTOP;
2840 			td->td_dbgflags |= TDB_FSTP;
2841 		} else {
2842 			sig = sig_ffs(&sigpending);
2843 		}
2844 
2845 		if (p->p_stops & S_SIG) {
2846 			mtx_unlock(&ps->ps_mtx);
2847 			stopevent(p, S_SIG, sig);
2848 			mtx_lock(&ps->ps_mtx);
2849 		}
2850 
2851 		/*
2852 		 * We should see pending but ignored signals
2853 		 * only if P_TRACED was on when they were posted.
2854 		 */
2855 		if (SIGISMEMBER(ps->ps_sigignore, sig) && (traced == 0)) {
2856 			sigqueue_delete(&td->td_sigqueue, sig);
2857 			sigqueue_delete(&p->p_sigqueue, sig);
2858 			continue;
2859 		}
2860 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
2861 			/*
2862 			 * If traced, always stop.
2863 			 * Remove old signal from queue before the stop.
2864 			 * XXX shrug off debugger, it causes siginfo to
2865 			 * be thrown away.
2866 			 */
2867 			queue = &td->td_sigqueue;
2868 			ksiginfo_init(&ksi);
2869 			if (sigqueue_get(queue, sig, &ksi) == 0) {
2870 				queue = &p->p_sigqueue;
2871 				sigqueue_get(queue, sig, &ksi);
2872 			}
2873 			td->td_si = ksi.ksi_info;
2874 
2875 			mtx_unlock(&ps->ps_mtx);
2876 			sig = ptracestop(td, sig, &ksi);
2877 			mtx_lock(&ps->ps_mtx);
2878 
2879 			td->td_si.si_signo = 0;
2880 
2881 			/*
2882 			 * Keep looking if the debugger discarded or
2883 			 * replaced the signal.
2884 			 */
2885 			if (sig == 0)
2886 				continue;
2887 
2888 			/*
2889 			 * If the signal became masked, re-queue it.
2890 			 */
2891 			if (SIGISMEMBER(td->td_sigmask, sig)) {
2892 				ksi.ksi_flags |= KSI_HEAD;
2893 				sigqueue_add(&p->p_sigqueue, sig, &ksi);
2894 				continue;
2895 			}
2896 
2897 			/*
2898 			 * If the traced bit got turned off, requeue
2899 			 * the signal and go back up to the top to
2900 			 * rescan signals.  This ensures that p_sig*
2901 			 * and p_sigact are consistent.
2902 			 */
2903 			if ((p->p_flag & P_TRACED) == 0) {
2904 				ksi.ksi_flags |= KSI_HEAD;
2905 				sigqueue_add(queue, sig, &ksi);
2906 				continue;
2907 			}
2908 		}
2909 
2910 		prop = sigprop(sig);
2911 
2912 		/*
2913 		 * Decide whether the signal should be returned.
2914 		 * Return the signal's number, or fall through
2915 		 * to clear it from the pending mask.
2916 		 */
2917 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2918 
2919 		case (intptr_t)SIG_DFL:
2920 			/*
2921 			 * Don't take default actions on system processes.
2922 			 */
2923 			if (p->p_pid <= 1) {
2924 #ifdef DIAGNOSTIC
2925 				/*
2926 				 * Are you sure you want to ignore SIGSEGV
2927 				 * in init? XXX
2928 				 */
2929 				printf("Process (pid %lu) got signal %d\n",
2930 					(u_long)p->p_pid, sig);
2931 #endif
2932 				break;		/* == ignore */
2933 			}
2934 			/*
2935 			 * If there is a pending stop signal to process with
2936 			 * default action, stop here, then clear the signal.
2937 			 * Traced or exiting processes should ignore stops.
2938 			 * Additionally, a member of an orphaned process group
2939 			 * should ignore tty stops.
2940 			 */
2941 			if (prop & SIGPROP_STOP) {
2942 				if (p->p_flag &
2943 				    (P_TRACED | P_WEXIT | P_SINGLE_EXIT) ||
2944 				    (p->p_pgrp->pg_jobc == 0 &&
2945 				     prop & SIGPROP_TTYSTOP))
2946 					break;	/* == ignore */
2947 				if (TD_SBDRY_INTR(td)) {
2948 					KASSERT((td->td_flags & TDF_SBDRY) != 0,
2949 					    ("lost TDF_SBDRY"));
2950 					return (-1);
2951 				}
2952 				mtx_unlock(&ps->ps_mtx);
2953 				WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2954 				    &p->p_mtx.lock_object, "Catching SIGSTOP");
2955 				sigqueue_delete(&td->td_sigqueue, sig);
2956 				sigqueue_delete(&p->p_sigqueue, sig);
2957 				p->p_flag |= P_STOPPED_SIG;
2958 				p->p_xsig = sig;
2959 				PROC_SLOCK(p);
2960 				sig_suspend_threads(td, p, 0);
2961 				thread_suspend_switch(td, p);
2962 				PROC_SUNLOCK(p);
2963 				mtx_lock(&ps->ps_mtx);
2964 				goto next;
2965 			} else if (prop & SIGPROP_IGNORE) {
2966 				/*
2967 				 * Except for SIGCONT, shouldn't get here.
2968 				 * Default action is to ignore; drop it.
2969 				 */
2970 				break;		/* == ignore */
2971 			} else
2972 				return (sig);
2973 			/*NOTREACHED*/
2974 
2975 		case (intptr_t)SIG_IGN:
2976 			/*
2977 			 * Masking above should prevent us ever trying
2978 			 * to take action on an ignored signal other
2979 			 * than SIGCONT, unless process is traced.
2980 			 */
2981 			if ((prop & SIGPROP_CONT) == 0 &&
2982 			    (p->p_flag & P_TRACED) == 0)
2983 				printf("issignal\n");
2984 			break;		/* == ignore */
2985 
2986 		default:
2987 			/*
2988 			 * This signal has an action, let
2989 			 * postsig() process it.
2990 			 */
2991 			return (sig);
2992 		}
2993 		sigqueue_delete(&td->td_sigqueue, sig);	/* take the signal! */
2994 		sigqueue_delete(&p->p_sigqueue, sig);
2995 next:;
2996 	}
2997 	/* NOTREACHED */
2998 }
2999 
3000 void
3001 thread_stopped(struct proc *p)
3002 {
3003 	int n;
3004 
3005 	PROC_LOCK_ASSERT(p, MA_OWNED);
3006 	PROC_SLOCK_ASSERT(p, MA_OWNED);
3007 	n = p->p_suspcount;
3008 	if (p == curproc)
3009 		n++;
3010 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3011 		PROC_SUNLOCK(p);
3012 		p->p_flag &= ~P_WAITED;
3013 		PROC_LOCK(p->p_pptr);
3014 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
3015 			CLD_TRAPPED : CLD_STOPPED);
3016 		PROC_UNLOCK(p->p_pptr);
3017 		PROC_SLOCK(p);
3018 	}
3019 }
3020 
3021 /*
3022  * Take the action for the specified signal
3023  * from the current set of pending signals.
3024  */
3025 int
3026 postsig(int sig)
3027 {
3028 	struct thread *td;
3029 	struct proc *p;
3030 	struct sigacts *ps;
3031 	sig_t action;
3032 	ksiginfo_t ksi;
3033 	sigset_t returnmask;
3034 
3035 	KASSERT(sig != 0, ("postsig"));
3036 
3037 	td = curthread;
3038 	p = td->td_proc;
3039 	PROC_LOCK_ASSERT(p, MA_OWNED);
3040 	ps = p->p_sigacts;
3041 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3042 	ksiginfo_init(&ksi);
3043 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3044 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3045 		return (0);
3046 	ksi.ksi_signo = sig;
3047 	if (ksi.ksi_code == SI_TIMER)
3048 		itimer_accept(p, ksi.ksi_timerid, &ksi);
3049 	action = ps->ps_sigact[_SIG_IDX(sig)];
3050 #ifdef KTRACE
3051 	if (KTRPOINT(td, KTR_PSIG))
3052 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3053 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3054 #endif
3055 	if ((p->p_stops & S_SIG) != 0) {
3056 		mtx_unlock(&ps->ps_mtx);
3057 		stopevent(p, S_SIG, sig);
3058 		mtx_lock(&ps->ps_mtx);
3059 	}
3060 
3061 	if (action == SIG_DFL) {
3062 		/*
3063 		 * Default action, where the default is to kill
3064 		 * the process.  (Other cases were ignored above.)
3065 		 */
3066 		mtx_unlock(&ps->ps_mtx);
3067 		proc_td_siginfo_capture(td, &ksi.ksi_info);
3068 		sigexit(td, sig);
3069 		/* NOTREACHED */
3070 	} else {
3071 		/*
3072 		 * If we get here, the signal must be caught.
3073 		 */
3074 		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3075 		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3076 		    ("postsig action: blocked sig %d", sig));
3077 
3078 		/*
3079 		 * Set the new mask value and also defer further
3080 		 * occurrences of this signal.
3081 		 *
3082 		 * Special case: user has done a sigsuspend.  Here the
3083 		 * current mask is not of interest, but rather the
3084 		 * mask from before the sigsuspend is what we want
3085 		 * restored after the signal processing is completed.
3086 		 */
3087 		if (td->td_pflags & TDP_OLDMASK) {
3088 			returnmask = td->td_oldsigmask;
3089 			td->td_pflags &= ~TDP_OLDMASK;
3090 		} else
3091 			returnmask = td->td_sigmask;
3092 
3093 		if (p->p_sig == sig) {
3094 			p->p_sig = 0;
3095 		}
3096 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3097 		postsig_done(sig, td, ps);
3098 	}
3099 	return (1);
3100 }
3101 
3102 void
3103 proc_wkilled(struct proc *p)
3104 {
3105 
3106 	PROC_LOCK_ASSERT(p, MA_OWNED);
3107 	if ((p->p_flag & P_WKILLED) == 0) {
3108 		p->p_flag |= P_WKILLED;
3109 		/*
3110 		 * Notify swapper that there is a process to swap in.
3111 		 * The notification is racy, at worst it would take 10
3112 		 * seconds for the swapper process to notice.
3113 		 */
3114 		if ((p->p_flag & (P_INMEM | P_SWAPPINGIN)) == 0)
3115 			wakeup(&proc0);
3116 	}
3117 }
3118 
3119 /*
3120  * Kill the current process for stated reason.
3121  */
3122 void
3123 killproc(struct proc *p, char *why)
3124 {
3125 
3126 	PROC_LOCK_ASSERT(p, MA_OWNED);
3127 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3128 	    p->p_comm);
3129 	log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3130 	    p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3131 	    p->p_ucred->cr_uid, why);
3132 	proc_wkilled(p);
3133 	kern_psignal(p, SIGKILL);
3134 }
3135 
3136 /*
3137  * Force the current process to exit with the specified signal, dumping core
3138  * if appropriate.  We bypass the normal tests for masked and caught signals,
3139  * allowing unrecoverable failures to terminate the process without changing
3140  * signal state.  Mark the accounting record with the signal termination.
3141  * If dumping core, save the signal number for the debugger.  Calls exit and
3142  * does not return.
3143  */
3144 void
3145 sigexit(struct thread *td, int sig)
3146 {
3147 	struct proc *p = td->td_proc;
3148 
3149 	PROC_LOCK_ASSERT(p, MA_OWNED);
3150 	p->p_acflag |= AXSIG;
3151 	/*
3152 	 * We must be single-threading to generate a core dump.  This
3153 	 * ensures that the registers in the core file are up-to-date.
3154 	 * Also, the ELF dump handler assumes that the thread list doesn't
3155 	 * change out from under it.
3156 	 *
3157 	 * XXX If another thread attempts to single-thread before us
3158 	 *     (e.g. via fork()), we won't get a dump at all.
3159 	 */
3160 	if ((sigprop(sig) & SIGPROP_CORE) &&
3161 	    thread_single(p, SINGLE_NO_EXIT) == 0) {
3162 		p->p_sig = sig;
3163 		/*
3164 		 * Log signals which would cause core dumps
3165 		 * (Log as LOG_INFO to appease those who don't want
3166 		 * these messages.)
3167 		 * XXX : Todo, as well as euid, write out ruid too
3168 		 * Note that coredump() drops proc lock.
3169 		 */
3170 		if (coredump(td) == 0)
3171 			sig |= WCOREFLAG;
3172 		if (kern_logsigexit)
3173 			log(LOG_INFO,
3174 			    "pid %d (%s), jid %d, uid %d: exited on "
3175 			    "signal %d%s\n", p->p_pid, p->p_comm,
3176 			    p->p_ucred->cr_prison->pr_id,
3177 			    td->td_ucred->cr_uid,
3178 			    sig &~ WCOREFLAG,
3179 			    sig & WCOREFLAG ? " (core dumped)" : "");
3180 	} else
3181 		PROC_UNLOCK(p);
3182 	exit1(td, 0, sig);
3183 	/* NOTREACHED */
3184 }
3185 
3186 /*
3187  * Send queued SIGCHLD to parent when child process's state
3188  * is changed.
3189  */
3190 static void
3191 sigparent(struct proc *p, int reason, int status)
3192 {
3193 	PROC_LOCK_ASSERT(p, MA_OWNED);
3194 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3195 
3196 	if (p->p_ksi != NULL) {
3197 		p->p_ksi->ksi_signo  = SIGCHLD;
3198 		p->p_ksi->ksi_code   = reason;
3199 		p->p_ksi->ksi_status = status;
3200 		p->p_ksi->ksi_pid    = p->p_pid;
3201 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3202 		if (KSI_ONQ(p->p_ksi))
3203 			return;
3204 	}
3205 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3206 }
3207 
3208 static void
3209 childproc_jobstate(struct proc *p, int reason, int sig)
3210 {
3211 	struct sigacts *ps;
3212 
3213 	PROC_LOCK_ASSERT(p, MA_OWNED);
3214 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3215 
3216 	/*
3217 	 * Wake up parent sleeping in kern_wait(), also send
3218 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3219 	 * that parent will awake, because parent may masked
3220 	 * the signal.
3221 	 */
3222 	p->p_pptr->p_flag |= P_STATCHILD;
3223 	wakeup(p->p_pptr);
3224 
3225 	ps = p->p_pptr->p_sigacts;
3226 	mtx_lock(&ps->ps_mtx);
3227 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3228 		mtx_unlock(&ps->ps_mtx);
3229 		sigparent(p, reason, sig);
3230 	} else
3231 		mtx_unlock(&ps->ps_mtx);
3232 }
3233 
3234 void
3235 childproc_stopped(struct proc *p, int reason)
3236 {
3237 
3238 	childproc_jobstate(p, reason, p->p_xsig);
3239 }
3240 
3241 void
3242 childproc_continued(struct proc *p)
3243 {
3244 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3245 }
3246 
3247 void
3248 childproc_exited(struct proc *p)
3249 {
3250 	int reason, status;
3251 
3252 	if (WCOREDUMP(p->p_xsig)) {
3253 		reason = CLD_DUMPED;
3254 		status = WTERMSIG(p->p_xsig);
3255 	} else if (WIFSIGNALED(p->p_xsig)) {
3256 		reason = CLD_KILLED;
3257 		status = WTERMSIG(p->p_xsig);
3258 	} else {
3259 		reason = CLD_EXITED;
3260 		status = p->p_xexit;
3261 	}
3262 	/*
3263 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3264 	 * done in exit1().
3265 	 */
3266 	sigparent(p, reason, status);
3267 }
3268 
3269 #define	MAX_NUM_CORE_FILES 100000
3270 #ifndef NUM_CORE_FILES
3271 #define	NUM_CORE_FILES 5
3272 #endif
3273 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3274 static int num_cores = NUM_CORE_FILES;
3275 
3276 static int
3277 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3278 {
3279 	int error;
3280 	int new_val;
3281 
3282 	new_val = num_cores;
3283 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3284 	if (error != 0 || req->newptr == NULL)
3285 		return (error);
3286 	if (new_val > MAX_NUM_CORE_FILES)
3287 		new_val = MAX_NUM_CORE_FILES;
3288 	if (new_val < 0)
3289 		new_val = 0;
3290 	num_cores = new_val;
3291 	return (0);
3292 }
3293 SYSCTL_PROC(_debug, OID_AUTO, ncores, CTLTYPE_INT|CTLFLAG_RW,
3294 	    0, sizeof(int), sysctl_debug_num_cores_check, "I",
3295 	    "Maximum number of generated process corefiles while using index format");
3296 
3297 #define	GZIP_SUFFIX	".gz"
3298 #define	ZSTD_SUFFIX	".zst"
3299 
3300 int compress_user_cores = 0;
3301 
3302 static int
3303 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3304 {
3305 	int error, val;
3306 
3307 	val = compress_user_cores;
3308 	error = sysctl_handle_int(oidp, &val, 0, req);
3309 	if (error != 0 || req->newptr == NULL)
3310 		return (error);
3311 	if (val != 0 && !compressor_avail(val))
3312 		return (EINVAL);
3313 	compress_user_cores = val;
3314 	return (error);
3315 }
3316 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores, CTLTYPE_INT | CTLFLAG_RWTUN,
3317     0, sizeof(int), sysctl_compress_user_cores, "I",
3318     "Enable compression of user corefiles ("
3319     __XSTRING(COMPRESS_GZIP) " = gzip, "
3320     __XSTRING(COMPRESS_ZSTD) " = zstd)");
3321 
3322 int compress_user_cores_level = 6;
3323 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3324     &compress_user_cores_level, 0,
3325     "Corefile compression level");
3326 
3327 /*
3328  * Protect the access to corefilename[] by allproc_lock.
3329  */
3330 #define	corefilename_lock	allproc_lock
3331 
3332 static char corefilename[MAXPATHLEN] = {"%N.core"};
3333 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3334 
3335 static int
3336 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3337 {
3338 	int error;
3339 
3340 	sx_xlock(&corefilename_lock);
3341 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3342 	    req);
3343 	sx_xunlock(&corefilename_lock);
3344 
3345 	return (error);
3346 }
3347 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3348     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3349     "Process corefile name format string");
3350 
3351 static void
3352 vnode_close_locked(struct thread *td, struct vnode *vp)
3353 {
3354 
3355 	VOP_UNLOCK(vp, 0);
3356 	vn_close(vp, FWRITE, td->td_ucred, td);
3357 }
3358 
3359 /*
3360  * If the core format has a %I in it, then we need to check
3361  * for existing corefiles before defining a name.
3362  * To do this we iterate over 0..ncores to find a
3363  * non-existing core file name to use. If all core files are
3364  * already used we choose the oldest one.
3365  */
3366 static int
3367 corefile_open_last(struct thread *td, char *name, int indexpos,
3368     int indexlen, int ncores, struct vnode **vpp)
3369 {
3370 	struct vnode *oldvp, *nextvp, *vp;
3371 	struct vattr vattr;
3372 	struct nameidata nd;
3373 	int error, i, flags, oflags, cmode;
3374 	char ch;
3375 	struct timespec lasttime;
3376 
3377 	nextvp = oldvp = NULL;
3378 	cmode = S_IRUSR | S_IWUSR;
3379 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3380 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3381 
3382 	for (i = 0; i < ncores; i++) {
3383 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3384 
3385 		ch = name[indexpos + indexlen];
3386 		(void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3387 		    i);
3388 		name[indexpos + indexlen] = ch;
3389 
3390 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3391 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3392 		    NULL);
3393 		if (error != 0)
3394 			break;
3395 
3396 		vp = nd.ni_vp;
3397 		NDFREE(&nd, NDF_ONLY_PNBUF);
3398 		if ((flags & O_CREAT) == O_CREAT) {
3399 			nextvp = vp;
3400 			break;
3401 		}
3402 
3403 		error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3404 		if (error != 0) {
3405 			vnode_close_locked(td, vp);
3406 			break;
3407 		}
3408 
3409 		if (oldvp == NULL ||
3410 		    lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3411 		    (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3412 		    lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3413 			if (oldvp != NULL)
3414 				vnode_close_locked(td, oldvp);
3415 			oldvp = vp;
3416 			lasttime = vattr.va_mtime;
3417 		} else {
3418 			vnode_close_locked(td, vp);
3419 		}
3420 	}
3421 
3422 	if (oldvp != NULL) {
3423 		if (nextvp == NULL) {
3424 			if ((td->td_proc->p_flag & P_SUGID) != 0) {
3425 				error = EFAULT;
3426 				vnode_close_locked(td, oldvp);
3427 			} else {
3428 				nextvp = oldvp;
3429 			}
3430 		} else {
3431 			vnode_close_locked(td, oldvp);
3432 		}
3433 	}
3434 	if (error != 0) {
3435 		if (nextvp != NULL)
3436 			vnode_close_locked(td, oldvp);
3437 	} else {
3438 		*vpp = nextvp;
3439 	}
3440 
3441 	return (error);
3442 }
3443 
3444 /*
3445  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3446  * Expand the name described in corefilename, using name, uid, and pid
3447  * and open/create core file.
3448  * corefilename is a printf-like string, with three format specifiers:
3449  *	%N	name of process ("name")
3450  *	%P	process id (pid)
3451  *	%U	user id (uid)
3452  * For example, "%N.core" is the default; they can be disabled completely
3453  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3454  * This is controlled by the sysctl variable kern.corefile (see above).
3455  */
3456 static int
3457 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3458     int compress, int signum, struct vnode **vpp, char **namep)
3459 {
3460 	struct sbuf sb;
3461 	struct nameidata nd;
3462 	const char *format;
3463 	char *hostname, *name;
3464 	int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
3465 
3466 	hostname = NULL;
3467 	format = corefilename;
3468 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3469 	indexlen = 0;
3470 	indexpos = -1;
3471 	ncores = num_cores;
3472 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3473 	sx_slock(&corefilename_lock);
3474 	for (i = 0; format[i] != '\0'; i++) {
3475 		switch (format[i]) {
3476 		case '%':	/* Format character */
3477 			i++;
3478 			switch (format[i]) {
3479 			case '%':
3480 				sbuf_putc(&sb, '%');
3481 				break;
3482 			case 'H':	/* hostname */
3483 				if (hostname == NULL) {
3484 					hostname = malloc(MAXHOSTNAMELEN,
3485 					    M_TEMP, M_WAITOK);
3486 				}
3487 				getcredhostname(td->td_ucred, hostname,
3488 				    MAXHOSTNAMELEN);
3489 				sbuf_printf(&sb, "%s", hostname);
3490 				break;
3491 			case 'I':	/* autoincrementing index */
3492 				if (indexpos != -1) {
3493 					sbuf_printf(&sb, "%%I");
3494 					break;
3495 				}
3496 
3497 				indexpos = sbuf_len(&sb);
3498 				sbuf_printf(&sb, "%u", ncores - 1);
3499 				indexlen = sbuf_len(&sb) - indexpos;
3500 				break;
3501 			case 'N':	/* process name */
3502 				sbuf_printf(&sb, "%s", comm);
3503 				break;
3504 			case 'P':	/* process id */
3505 				sbuf_printf(&sb, "%u", pid);
3506 				break;
3507 			case 'S':	/* signal number */
3508 				sbuf_printf(&sb, "%i", signum);
3509 				break;
3510 			case 'U':	/* user id */
3511 				sbuf_printf(&sb, "%u", uid);
3512 				break;
3513 			default:
3514 				log(LOG_ERR,
3515 				    "Unknown format character %c in "
3516 				    "corename `%s'\n", format[i], format);
3517 				break;
3518 			}
3519 			break;
3520 		default:
3521 			sbuf_putc(&sb, format[i]);
3522 			break;
3523 		}
3524 	}
3525 	sx_sunlock(&corefilename_lock);
3526 	free(hostname, M_TEMP);
3527 	if (compress == COMPRESS_GZIP)
3528 		sbuf_printf(&sb, GZIP_SUFFIX);
3529 	else if (compress == COMPRESS_ZSTD)
3530 		sbuf_printf(&sb, ZSTD_SUFFIX);
3531 	if (sbuf_error(&sb) != 0) {
3532 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3533 		    "long\n", (long)pid, comm, (u_long)uid);
3534 		sbuf_delete(&sb);
3535 		free(name, M_TEMP);
3536 		return (ENOMEM);
3537 	}
3538 	sbuf_finish(&sb);
3539 	sbuf_delete(&sb);
3540 
3541 	if (indexpos != -1) {
3542 		error = corefile_open_last(td, name, indexpos, indexlen, ncores,
3543 		    vpp);
3544 		if (error != 0) {
3545 			log(LOG_ERR,
3546 			    "pid %d (%s), uid (%u):  Path `%s' failed "
3547 			    "on initial open test, error = %d\n",
3548 			    pid, comm, uid, name, error);
3549 		}
3550 	} else {
3551 		cmode = S_IRUSR | S_IWUSR;
3552 		oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3553 		    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3554 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3555 		if ((td->td_proc->p_flag & P_SUGID) != 0)
3556 			flags |= O_EXCL;
3557 
3558 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3559 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3560 		    NULL);
3561 		if (error == 0) {
3562 			*vpp = nd.ni_vp;
3563 			NDFREE(&nd, NDF_ONLY_PNBUF);
3564 		}
3565 	}
3566 
3567 	if (error != 0) {
3568 #ifdef AUDIT
3569 		audit_proc_coredump(td, name, error);
3570 #endif
3571 		free(name, M_TEMP);
3572 		return (error);
3573 	}
3574 	*namep = name;
3575 	return (0);
3576 }
3577 
3578 /*
3579  * Dump a process' core.  The main routine does some
3580  * policy checking, and creates the name of the coredump;
3581  * then it passes on a vnode and a size limit to the process-specific
3582  * coredump routine if there is one; if there _is not_ one, it returns
3583  * ENOSYS; otherwise it returns the error from the process-specific routine.
3584  */
3585 
3586 static int
3587 coredump(struct thread *td)
3588 {
3589 	struct proc *p = td->td_proc;
3590 	struct ucred *cred = td->td_ucred;
3591 	struct vnode *vp;
3592 	struct flock lf;
3593 	struct vattr vattr;
3594 	int error, error1, locked;
3595 	char *name;			/* name of corefile */
3596 	void *rl_cookie;
3597 	off_t limit;
3598 	char *fullpath, *freepath = NULL;
3599 	struct sbuf *sb;
3600 
3601 	PROC_LOCK_ASSERT(p, MA_OWNED);
3602 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3603 	_STOPEVENT(p, S_CORE, 0);
3604 
3605 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
3606 	    (p->p_flag2 & P2_NOTRACE) != 0) {
3607 		PROC_UNLOCK(p);
3608 		return (EFAULT);
3609 	}
3610 
3611 	/*
3612 	 * Note that the bulk of limit checking is done after
3613 	 * the corefile is created.  The exception is if the limit
3614 	 * for corefiles is 0, in which case we don't bother
3615 	 * creating the corefile at all.  This layout means that
3616 	 * a corefile is truncated instead of not being created,
3617 	 * if it is larger than the limit.
3618 	 */
3619 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
3620 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
3621 		PROC_UNLOCK(p);
3622 		return (EFBIG);
3623 	}
3624 	PROC_UNLOCK(p);
3625 
3626 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
3627 	    compress_user_cores, p->p_sig, &vp, &name);
3628 	if (error != 0)
3629 		return (error);
3630 
3631 	/*
3632 	 * Don't dump to non-regular files or files with links.
3633 	 * Do not dump into system files. Effective user must own the corefile.
3634 	 */
3635 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
3636 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
3637 	    vattr.va_uid != cred->cr_uid) {
3638 		VOP_UNLOCK(vp, 0);
3639 		error = EFAULT;
3640 		goto out;
3641 	}
3642 
3643 	VOP_UNLOCK(vp, 0);
3644 
3645 	/* Postpone other writers, including core dumps of other processes. */
3646 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
3647 
3648 	lf.l_whence = SEEK_SET;
3649 	lf.l_start = 0;
3650 	lf.l_len = 0;
3651 	lf.l_type = F_WRLCK;
3652 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
3653 
3654 	VATTR_NULL(&vattr);
3655 	vattr.va_size = 0;
3656 	if (set_core_nodump_flag)
3657 		vattr.va_flags = UF_NODUMP;
3658 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3659 	VOP_SETATTR(vp, &vattr, cred);
3660 	VOP_UNLOCK(vp, 0);
3661 	PROC_LOCK(p);
3662 	p->p_acflag |= ACORE;
3663 	PROC_UNLOCK(p);
3664 
3665 	if (p->p_sysent->sv_coredump != NULL) {
3666 		error = p->p_sysent->sv_coredump(td, vp, limit, 0);
3667 	} else {
3668 		error = ENOSYS;
3669 	}
3670 
3671 	if (locked) {
3672 		lf.l_type = F_UNLCK;
3673 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3674 	}
3675 	vn_rangelock_unlock(vp, rl_cookie);
3676 
3677 	/*
3678 	 * Notify the userland helper that a process triggered a core dump.
3679 	 * This allows the helper to run an automated debugging session.
3680 	 */
3681 	if (error != 0 || coredump_devctl == 0)
3682 		goto out;
3683 	sb = sbuf_new_auto();
3684 	if (vn_fullpath_global(td, p->p_textvp, &fullpath, &freepath) != 0)
3685 		goto out2;
3686 	sbuf_printf(sb, "comm=\"");
3687 	devctl_safe_quote_sb(sb, fullpath);
3688 	free(freepath, M_TEMP);
3689 	sbuf_printf(sb, "\" core=\"");
3690 
3691 	/*
3692 	 * We can't lookup core file vp directly. When we're replacing a core, and
3693 	 * other random times, we flush the name cache, so it will fail. Instead,
3694 	 * if the path of the core is relative, add the current dir in front if it.
3695 	 */
3696 	if (name[0] != '/') {
3697 		fullpath = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
3698 		if (kern___getcwd(td, fullpath, UIO_SYSSPACE, MAXPATHLEN, MAXPATHLEN) != 0) {
3699 			free(fullpath, M_TEMP);
3700 			goto out2;
3701 		}
3702 		devctl_safe_quote_sb(sb, fullpath);
3703 		free(fullpath, M_TEMP);
3704 		sbuf_putc(sb, '/');
3705 	}
3706 	devctl_safe_quote_sb(sb, name);
3707 	sbuf_printf(sb, "\"");
3708 	if (sbuf_finish(sb) == 0)
3709 		devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
3710 out2:
3711 	sbuf_delete(sb);
3712 out:
3713 	error1 = vn_close(vp, FWRITE, cred, td);
3714 	if (error == 0)
3715 		error = error1;
3716 #ifdef AUDIT
3717 	audit_proc_coredump(td, name, error);
3718 #endif
3719 	free(name, M_TEMP);
3720 	return (error);
3721 }
3722 
3723 /*
3724  * Nonexistent system call-- signal process (may want to handle it).  Flag
3725  * error in case process won't see signal immediately (blocked or ignored).
3726  */
3727 #ifndef _SYS_SYSPROTO_H_
3728 struct nosys_args {
3729 	int	dummy;
3730 };
3731 #endif
3732 /* ARGSUSED */
3733 int
3734 nosys(struct thread *td, struct nosys_args *args)
3735 {
3736 	struct proc *p;
3737 
3738 	p = td->td_proc;
3739 
3740 	PROC_LOCK(p);
3741 	tdsignal(td, SIGSYS);
3742 	PROC_UNLOCK(p);
3743 	if (kern_lognosys == 1 || kern_lognosys == 3) {
3744 		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
3745 		    td->td_sa.code);
3746 	}
3747 	if (kern_lognosys == 2 || kern_lognosys == 3) {
3748 		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
3749 		    td->td_sa.code);
3750 	}
3751 	return (ENOSYS);
3752 }
3753 
3754 /*
3755  * Send a SIGIO or SIGURG signal to a process or process group using stored
3756  * credentials rather than those of the current process.
3757  */
3758 void
3759 pgsigio(struct sigio **sigiop, int sig, int checkctty)
3760 {
3761 	ksiginfo_t ksi;
3762 	struct sigio *sigio;
3763 
3764 	ksiginfo_init(&ksi);
3765 	ksi.ksi_signo = sig;
3766 	ksi.ksi_code = SI_KERNEL;
3767 
3768 	SIGIO_LOCK();
3769 	sigio = *sigiop;
3770 	if (sigio == NULL) {
3771 		SIGIO_UNLOCK();
3772 		return;
3773 	}
3774 	if (sigio->sio_pgid > 0) {
3775 		PROC_LOCK(sigio->sio_proc);
3776 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
3777 			kern_psignal(sigio->sio_proc, sig);
3778 		PROC_UNLOCK(sigio->sio_proc);
3779 	} else if (sigio->sio_pgid < 0) {
3780 		struct proc *p;
3781 
3782 		PGRP_LOCK(sigio->sio_pgrp);
3783 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
3784 			PROC_LOCK(p);
3785 			if (p->p_state == PRS_NORMAL &&
3786 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
3787 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
3788 				kern_psignal(p, sig);
3789 			PROC_UNLOCK(p);
3790 		}
3791 		PGRP_UNLOCK(sigio->sio_pgrp);
3792 	}
3793 	SIGIO_UNLOCK();
3794 }
3795 
3796 static int
3797 filt_sigattach(struct knote *kn)
3798 {
3799 	struct proc *p = curproc;
3800 
3801 	kn->kn_ptr.p_proc = p;
3802 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
3803 
3804 	knlist_add(p->p_klist, kn, 0);
3805 
3806 	return (0);
3807 }
3808 
3809 static void
3810 filt_sigdetach(struct knote *kn)
3811 {
3812 	struct proc *p = kn->kn_ptr.p_proc;
3813 
3814 	knlist_remove(p->p_klist, kn, 0);
3815 }
3816 
3817 /*
3818  * signal knotes are shared with proc knotes, so we apply a mask to
3819  * the hint in order to differentiate them from process hints.  This
3820  * could be avoided by using a signal-specific knote list, but probably
3821  * isn't worth the trouble.
3822  */
3823 static int
3824 filt_signal(struct knote *kn, long hint)
3825 {
3826 
3827 	if (hint & NOTE_SIGNAL) {
3828 		hint &= ~NOTE_SIGNAL;
3829 
3830 		if (kn->kn_id == hint)
3831 			kn->kn_data++;
3832 	}
3833 	return (kn->kn_data != 0);
3834 }
3835 
3836 struct sigacts *
3837 sigacts_alloc(void)
3838 {
3839 	struct sigacts *ps;
3840 
3841 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
3842 	refcount_init(&ps->ps_refcnt, 1);
3843 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
3844 	return (ps);
3845 }
3846 
3847 void
3848 sigacts_free(struct sigacts *ps)
3849 {
3850 
3851 	if (refcount_release(&ps->ps_refcnt) == 0)
3852 		return;
3853 	mtx_destroy(&ps->ps_mtx);
3854 	free(ps, M_SUBPROC);
3855 }
3856 
3857 struct sigacts *
3858 sigacts_hold(struct sigacts *ps)
3859 {
3860 
3861 	refcount_acquire(&ps->ps_refcnt);
3862 	return (ps);
3863 }
3864 
3865 void
3866 sigacts_copy(struct sigacts *dest, struct sigacts *src)
3867 {
3868 
3869 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
3870 	mtx_lock(&src->ps_mtx);
3871 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
3872 	mtx_unlock(&src->ps_mtx);
3873 }
3874 
3875 int
3876 sigacts_shared(struct sigacts *ps)
3877 {
3878 
3879 	return (ps->ps_refcnt > 1);
3880 }
3881 
3882 void
3883 sig_drop_caught(struct proc *p)
3884 {
3885 	int sig;
3886 	struct sigacts *ps;
3887 
3888 	ps = p->p_sigacts;
3889 	PROC_LOCK_ASSERT(p, MA_OWNED);
3890 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3891 	while (SIGNOTEMPTY(ps->ps_sigcatch)) {
3892 		sig = sig_ffs(&ps->ps_sigcatch);
3893 		sigdflt(ps, sig);
3894 		if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
3895 			sigqueue_delete_proc(p, sig);
3896 	}
3897 }
3898