xref: /freebsd/sys/kern/kern_sig.c (revision c697fb7f7cc9bedc5beee44d35b771c4e87b335a)
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 void	reschedule_signals(struct proc *p, sigset_t block, int flags);
112 static int	sigprop(int sig);
113 static void	tdsigwakeup(struct thread *, int, sig_t, int);
114 static int	sig_suspend_threads(struct thread *, struct proc *, int);
115 static int	filt_sigattach(struct knote *kn);
116 static void	filt_sigdetach(struct knote *kn);
117 static int	filt_signal(struct knote *kn, long hint);
118 static struct thread *sigtd(struct proc *p, int sig, bool fast_sigblock);
119 static void	sigqueue_start(void);
120 
121 static uma_zone_t	ksiginfo_zone = NULL;
122 struct filterops sig_filtops = {
123 	.f_isfd = 0,
124 	.f_attach = filt_sigattach,
125 	.f_detach = filt_sigdetach,
126 	.f_event = filt_signal,
127 };
128 
129 static int	kern_logsigexit = 1;
130 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
131     &kern_logsigexit, 0,
132     "Log processes quitting on abnormal signals to syslog(3)");
133 
134 static int	kern_forcesigexit = 1;
135 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
136     &kern_forcesigexit, 0, "Force trap signal to be handled");
137 
138 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
139     "POSIX real time signal");
140 
141 static int	max_pending_per_proc = 128;
142 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
143     &max_pending_per_proc, 0, "Max pending signals per proc");
144 
145 static int	preallocate_siginfo = 1024;
146 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
147     &preallocate_siginfo, 0, "Preallocated signal memory size");
148 
149 static int	signal_overflow = 0;
150 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
151     &signal_overflow, 0, "Number of signals overflew");
152 
153 static int	signal_alloc_fail = 0;
154 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
155     &signal_alloc_fail, 0, "signals failed to be allocated");
156 
157 static int	kern_lognosys = 0;
158 SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
159     "Log invalid syscalls");
160 
161 __read_frequently bool sigfastblock_fetch_always = false;
162 SYSCTL_BOOL(_kern, OID_AUTO, sigfastblock_fetch_always, CTLFLAG_RWTUN,
163     &sigfastblock_fetch_always, 0,
164     "Fetch sigfastblock word on each syscall entry for proper "
165     "blocking semantic");
166 
167 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
168 
169 /*
170  * Policy -- Can ucred cr1 send SIGIO to process cr2?
171  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
172  * in the right situations.
173  */
174 #define CANSIGIO(cr1, cr2) \
175 	((cr1)->cr_uid == 0 || \
176 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
177 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
178 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
179 	    (cr1)->cr_uid == (cr2)->cr_uid)
180 
181 static int	sugid_coredump;
182 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
183     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
184 
185 static int	capmode_coredump;
186 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
187     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
188 
189 static int	do_coredump = 1;
190 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
191 	&do_coredump, 0, "Enable/Disable coredumps");
192 
193 static int	set_core_nodump_flag = 0;
194 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
195 	0, "Enable setting the NODUMP flag on coredump files");
196 
197 static int	coredump_devctl = 0;
198 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
199 	0, "Generate a devctl notification when processes coredump");
200 
201 /*
202  * Signal properties and actions.
203  * The array below categorizes the signals and their default actions
204  * according to the following properties:
205  */
206 #define	SIGPROP_KILL		0x01	/* terminates process by default */
207 #define	SIGPROP_CORE		0x02	/* ditto and coredumps */
208 #define	SIGPROP_STOP		0x04	/* suspend process */
209 #define	SIGPROP_TTYSTOP		0x08	/* ditto, from tty */
210 #define	SIGPROP_IGNORE		0x10	/* ignore by default */
211 #define	SIGPROP_CONT		0x20	/* continue if suspended */
212 #define	SIGPROP_CANTMASK	0x40	/* non-maskable, catchable */
213 
214 static int sigproptbl[NSIG] = {
215 	[SIGHUP] =	SIGPROP_KILL,
216 	[SIGINT] =	SIGPROP_KILL,
217 	[SIGQUIT] =	SIGPROP_KILL | SIGPROP_CORE,
218 	[SIGILL] =	SIGPROP_KILL | SIGPROP_CORE,
219 	[SIGTRAP] =	SIGPROP_KILL | SIGPROP_CORE,
220 	[SIGABRT] =	SIGPROP_KILL | SIGPROP_CORE,
221 	[SIGEMT] =	SIGPROP_KILL | SIGPROP_CORE,
222 	[SIGFPE] =	SIGPROP_KILL | SIGPROP_CORE,
223 	[SIGKILL] =	SIGPROP_KILL,
224 	[SIGBUS] =	SIGPROP_KILL | SIGPROP_CORE,
225 	[SIGSEGV] =	SIGPROP_KILL | SIGPROP_CORE,
226 	[SIGSYS] =	SIGPROP_KILL | SIGPROP_CORE,
227 	[SIGPIPE] =	SIGPROP_KILL,
228 	[SIGALRM] =	SIGPROP_KILL,
229 	[SIGTERM] =	SIGPROP_KILL,
230 	[SIGURG] =	SIGPROP_IGNORE,
231 	[SIGSTOP] =	SIGPROP_STOP,
232 	[SIGTSTP] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
233 	[SIGCONT] =	SIGPROP_IGNORE | SIGPROP_CONT,
234 	[SIGCHLD] =	SIGPROP_IGNORE,
235 	[SIGTTIN] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
236 	[SIGTTOU] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
237 	[SIGIO] =	SIGPROP_IGNORE,
238 	[SIGXCPU] =	SIGPROP_KILL,
239 	[SIGXFSZ] =	SIGPROP_KILL,
240 	[SIGVTALRM] =	SIGPROP_KILL,
241 	[SIGPROF] =	SIGPROP_KILL,
242 	[SIGWINCH] =	SIGPROP_IGNORE,
243 	[SIGINFO] =	SIGPROP_IGNORE,
244 	[SIGUSR1] =	SIGPROP_KILL,
245 	[SIGUSR2] =	SIGPROP_KILL,
246 };
247 
248 sigset_t fastblock_mask;
249 
250 static void
251 sigqueue_start(void)
252 {
253 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
254 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
255 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
256 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
257 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
258 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
259 	SIGFILLSET(fastblock_mask);
260 	SIG_CANTMASK(fastblock_mask);
261 }
262 
263 ksiginfo_t *
264 ksiginfo_alloc(int wait)
265 {
266 	int flags;
267 
268 	flags = M_ZERO;
269 	if (! wait)
270 		flags |= M_NOWAIT;
271 	if (ksiginfo_zone != NULL)
272 		return ((ksiginfo_t *)uma_zalloc(ksiginfo_zone, flags));
273 	return (NULL);
274 }
275 
276 void
277 ksiginfo_free(ksiginfo_t *ksi)
278 {
279 	uma_zfree(ksiginfo_zone, ksi);
280 }
281 
282 static __inline int
283 ksiginfo_tryfree(ksiginfo_t *ksi)
284 {
285 	if (!(ksi->ksi_flags & KSI_EXT)) {
286 		uma_zfree(ksiginfo_zone, ksi);
287 		return (1);
288 	}
289 	return (0);
290 }
291 
292 void
293 sigqueue_init(sigqueue_t *list, struct proc *p)
294 {
295 	SIGEMPTYSET(list->sq_signals);
296 	SIGEMPTYSET(list->sq_kill);
297 	SIGEMPTYSET(list->sq_ptrace);
298 	TAILQ_INIT(&list->sq_list);
299 	list->sq_proc = p;
300 	list->sq_flags = SQ_INIT;
301 }
302 
303 /*
304  * Get a signal's ksiginfo.
305  * Return:
306  *	0	-	signal not found
307  *	others	-	signal number
308  */
309 static int
310 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
311 {
312 	struct proc *p = sq->sq_proc;
313 	struct ksiginfo *ksi, *next;
314 	int count = 0;
315 
316 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
317 
318 	if (!SIGISMEMBER(sq->sq_signals, signo))
319 		return (0);
320 
321 	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
322 		count++;
323 		SIGDELSET(sq->sq_ptrace, signo);
324 		si->ksi_flags |= KSI_PTRACE;
325 	}
326 	if (SIGISMEMBER(sq->sq_kill, signo)) {
327 		count++;
328 		if (count == 1)
329 			SIGDELSET(sq->sq_kill, signo);
330 	}
331 
332 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
333 		if (ksi->ksi_signo == signo) {
334 			if (count == 0) {
335 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
336 				ksi->ksi_sigq = NULL;
337 				ksiginfo_copy(ksi, si);
338 				if (ksiginfo_tryfree(ksi) && p != NULL)
339 					p->p_pendingcnt--;
340 			}
341 			if (++count > 1)
342 				break;
343 		}
344 	}
345 
346 	if (count <= 1)
347 		SIGDELSET(sq->sq_signals, signo);
348 	si->ksi_signo = signo;
349 	return (signo);
350 }
351 
352 void
353 sigqueue_take(ksiginfo_t *ksi)
354 {
355 	struct ksiginfo *kp;
356 	struct proc	*p;
357 	sigqueue_t	*sq;
358 
359 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
360 		return;
361 
362 	p = sq->sq_proc;
363 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
364 	ksi->ksi_sigq = NULL;
365 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
366 		p->p_pendingcnt--;
367 
368 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
369 	     kp = TAILQ_NEXT(kp, ksi_link)) {
370 		if (kp->ksi_signo == ksi->ksi_signo)
371 			break;
372 	}
373 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
374 	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
375 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
376 }
377 
378 static int
379 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
380 {
381 	struct proc *p = sq->sq_proc;
382 	struct ksiginfo *ksi;
383 	int ret = 0;
384 
385 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
386 
387 	/*
388 	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
389 	 * for these signals.
390 	 */
391 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
392 		SIGADDSET(sq->sq_kill, signo);
393 		goto out_set_bit;
394 	}
395 
396 	/* directly insert the ksi, don't copy it */
397 	if (si->ksi_flags & KSI_INS) {
398 		if (si->ksi_flags & KSI_HEAD)
399 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
400 		else
401 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
402 		si->ksi_sigq = sq;
403 		goto out_set_bit;
404 	}
405 
406 	if (__predict_false(ksiginfo_zone == NULL)) {
407 		SIGADDSET(sq->sq_kill, signo);
408 		goto out_set_bit;
409 	}
410 
411 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
412 		signal_overflow++;
413 		ret = EAGAIN;
414 	} else if ((ksi = ksiginfo_alloc(0)) == NULL) {
415 		signal_alloc_fail++;
416 		ret = EAGAIN;
417 	} else {
418 		if (p != NULL)
419 			p->p_pendingcnt++;
420 		ksiginfo_copy(si, ksi);
421 		ksi->ksi_signo = signo;
422 		if (si->ksi_flags & KSI_HEAD)
423 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
424 		else
425 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
426 		ksi->ksi_sigq = sq;
427 	}
428 
429 	if (ret != 0) {
430 		if ((si->ksi_flags & KSI_PTRACE) != 0) {
431 			SIGADDSET(sq->sq_ptrace, signo);
432 			ret = 0;
433 			goto out_set_bit;
434 		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
435 		    (si->ksi_flags & KSI_SIGQ) == 0) {
436 			SIGADDSET(sq->sq_kill, signo);
437 			ret = 0;
438 			goto out_set_bit;
439 		}
440 		return (ret);
441 	}
442 
443 out_set_bit:
444 	SIGADDSET(sq->sq_signals, signo);
445 	return (ret);
446 }
447 
448 void
449 sigqueue_flush(sigqueue_t *sq)
450 {
451 	struct proc *p = sq->sq_proc;
452 	ksiginfo_t *ksi;
453 
454 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
455 
456 	if (p != NULL)
457 		PROC_LOCK_ASSERT(p, MA_OWNED);
458 
459 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
460 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
461 		ksi->ksi_sigq = NULL;
462 		if (ksiginfo_tryfree(ksi) && p != NULL)
463 			p->p_pendingcnt--;
464 	}
465 
466 	SIGEMPTYSET(sq->sq_signals);
467 	SIGEMPTYSET(sq->sq_kill);
468 	SIGEMPTYSET(sq->sq_ptrace);
469 }
470 
471 static void
472 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
473 {
474 	sigset_t tmp;
475 	struct proc *p1, *p2;
476 	ksiginfo_t *ksi, *next;
477 
478 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
479 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
480 	p1 = src->sq_proc;
481 	p2 = dst->sq_proc;
482 	/* Move siginfo to target list */
483 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
484 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
485 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
486 			if (p1 != NULL)
487 				p1->p_pendingcnt--;
488 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
489 			ksi->ksi_sigq = dst;
490 			if (p2 != NULL)
491 				p2->p_pendingcnt++;
492 		}
493 	}
494 
495 	/* Move pending bits to target list */
496 	tmp = src->sq_kill;
497 	SIGSETAND(tmp, *set);
498 	SIGSETOR(dst->sq_kill, tmp);
499 	SIGSETNAND(src->sq_kill, tmp);
500 
501 	tmp = src->sq_ptrace;
502 	SIGSETAND(tmp, *set);
503 	SIGSETOR(dst->sq_ptrace, tmp);
504 	SIGSETNAND(src->sq_ptrace, tmp);
505 
506 	tmp = src->sq_signals;
507 	SIGSETAND(tmp, *set);
508 	SIGSETOR(dst->sq_signals, tmp);
509 	SIGSETNAND(src->sq_signals, tmp);
510 }
511 
512 #if 0
513 static void
514 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
515 {
516 	sigset_t set;
517 
518 	SIGEMPTYSET(set);
519 	SIGADDSET(set, signo);
520 	sigqueue_move_set(src, dst, &set);
521 }
522 #endif
523 
524 static void
525 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
526 {
527 	struct proc *p = sq->sq_proc;
528 	ksiginfo_t *ksi, *next;
529 
530 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
531 
532 	/* Remove siginfo queue */
533 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
534 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
535 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
536 			ksi->ksi_sigq = NULL;
537 			if (ksiginfo_tryfree(ksi) && p != NULL)
538 				p->p_pendingcnt--;
539 		}
540 	}
541 	SIGSETNAND(sq->sq_kill, *set);
542 	SIGSETNAND(sq->sq_ptrace, *set);
543 	SIGSETNAND(sq->sq_signals, *set);
544 }
545 
546 void
547 sigqueue_delete(sigqueue_t *sq, int signo)
548 {
549 	sigset_t set;
550 
551 	SIGEMPTYSET(set);
552 	SIGADDSET(set, signo);
553 	sigqueue_delete_set(sq, &set);
554 }
555 
556 /* Remove a set of signals for a process */
557 static void
558 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
559 {
560 	sigqueue_t worklist;
561 	struct thread *td0;
562 
563 	PROC_LOCK_ASSERT(p, MA_OWNED);
564 
565 	sigqueue_init(&worklist, NULL);
566 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
567 
568 	FOREACH_THREAD_IN_PROC(p, td0)
569 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
570 
571 	sigqueue_flush(&worklist);
572 }
573 
574 void
575 sigqueue_delete_proc(struct proc *p, int signo)
576 {
577 	sigset_t set;
578 
579 	SIGEMPTYSET(set);
580 	SIGADDSET(set, signo);
581 	sigqueue_delete_set_proc(p, &set);
582 }
583 
584 static void
585 sigqueue_delete_stopmask_proc(struct proc *p)
586 {
587 	sigset_t set;
588 
589 	SIGEMPTYSET(set);
590 	SIGADDSET(set, SIGSTOP);
591 	SIGADDSET(set, SIGTSTP);
592 	SIGADDSET(set, SIGTTIN);
593 	SIGADDSET(set, SIGTTOU);
594 	sigqueue_delete_set_proc(p, &set);
595 }
596 
597 /*
598  * Determine signal that should be delivered to thread td, the current
599  * thread, 0 if none.  If there is a pending stop signal with default
600  * action, the process stops in issignal().
601  */
602 int
603 cursig(struct thread *td)
604 {
605 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
606 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
607 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
608 	return (SIGPENDING(td) ? issignal(td) : 0);
609 }
610 
611 /*
612  * Arrange for ast() to handle unmasked pending signals on return to user
613  * mode.  This must be called whenever a signal is added to td_sigqueue or
614  * unmasked in td_sigmask.
615  */
616 void
617 signotify(struct thread *td)
618 {
619 
620 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
621 
622 	if (SIGPENDING(td)) {
623 		thread_lock(td);
624 		td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING;
625 		thread_unlock(td);
626 	}
627 }
628 
629 /*
630  * Returns 1 (true) if altstack is configured for the thread, and the
631  * passed stack bottom address falls into the altstack range.  Handles
632  * the 43 compat special case where the alt stack size is zero.
633  */
634 int
635 sigonstack(size_t sp)
636 {
637 	struct thread *td;
638 
639 	td = curthread;
640 	if ((td->td_pflags & TDP_ALTSTACK) == 0)
641 		return (0);
642 #if defined(COMPAT_43)
643 	if (SV_PROC_FLAG(td->td_proc, SV_AOUT) && td->td_sigstk.ss_size == 0)
644 		return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
645 #endif
646 	return (sp >= (size_t)td->td_sigstk.ss_sp &&
647 	    sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
648 }
649 
650 static __inline int
651 sigprop(int sig)
652 {
653 
654 	if (sig > 0 && sig < nitems(sigproptbl))
655 		return (sigproptbl[sig]);
656 	return (0);
657 }
658 
659 int
660 sig_ffs(sigset_t *set)
661 {
662 	int i;
663 
664 	for (i = 0; i < _SIG_WORDS; i++)
665 		if (set->__bits[i])
666 			return (ffs(set->__bits[i]) + (i * 32));
667 	return (0);
668 }
669 
670 static bool
671 sigact_flag_test(const struct sigaction *act, int flag)
672 {
673 
674 	/*
675 	 * SA_SIGINFO is reset when signal disposition is set to
676 	 * ignore or default.  Other flags are kept according to user
677 	 * settings.
678 	 */
679 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
680 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
681 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
682 }
683 
684 /*
685  * kern_sigaction
686  * sigaction
687  * freebsd4_sigaction
688  * osigaction
689  */
690 int
691 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
692     struct sigaction *oact, int flags)
693 {
694 	struct sigacts *ps;
695 	struct proc *p = td->td_proc;
696 
697 	if (!_SIG_VALID(sig))
698 		return (EINVAL);
699 	if (act != NULL && act->sa_handler != SIG_DFL &&
700 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
701 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
702 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
703 		return (EINVAL);
704 
705 	PROC_LOCK(p);
706 	ps = p->p_sigacts;
707 	mtx_lock(&ps->ps_mtx);
708 	if (oact) {
709 		memset(oact, 0, sizeof(*oact));
710 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
711 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
712 			oact->sa_flags |= SA_ONSTACK;
713 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
714 			oact->sa_flags |= SA_RESTART;
715 		if (SIGISMEMBER(ps->ps_sigreset, sig))
716 			oact->sa_flags |= SA_RESETHAND;
717 		if (SIGISMEMBER(ps->ps_signodefer, sig))
718 			oact->sa_flags |= SA_NODEFER;
719 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
720 			oact->sa_flags |= SA_SIGINFO;
721 			oact->sa_sigaction =
722 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
723 		} else
724 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
725 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
726 			oact->sa_flags |= SA_NOCLDSTOP;
727 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
728 			oact->sa_flags |= SA_NOCLDWAIT;
729 	}
730 	if (act) {
731 		if ((sig == SIGKILL || sig == SIGSTOP) &&
732 		    act->sa_handler != SIG_DFL) {
733 			mtx_unlock(&ps->ps_mtx);
734 			PROC_UNLOCK(p);
735 			return (EINVAL);
736 		}
737 
738 		/*
739 		 * Change setting atomically.
740 		 */
741 
742 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
743 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
744 		if (sigact_flag_test(act, SA_SIGINFO)) {
745 			ps->ps_sigact[_SIG_IDX(sig)] =
746 			    (__sighandler_t *)act->sa_sigaction;
747 			SIGADDSET(ps->ps_siginfo, sig);
748 		} else {
749 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
750 			SIGDELSET(ps->ps_siginfo, sig);
751 		}
752 		if (!sigact_flag_test(act, SA_RESTART))
753 			SIGADDSET(ps->ps_sigintr, sig);
754 		else
755 			SIGDELSET(ps->ps_sigintr, sig);
756 		if (sigact_flag_test(act, SA_ONSTACK))
757 			SIGADDSET(ps->ps_sigonstack, sig);
758 		else
759 			SIGDELSET(ps->ps_sigonstack, sig);
760 		if (sigact_flag_test(act, SA_RESETHAND))
761 			SIGADDSET(ps->ps_sigreset, sig);
762 		else
763 			SIGDELSET(ps->ps_sigreset, sig);
764 		if (sigact_flag_test(act, SA_NODEFER))
765 			SIGADDSET(ps->ps_signodefer, sig);
766 		else
767 			SIGDELSET(ps->ps_signodefer, sig);
768 		if (sig == SIGCHLD) {
769 			if (act->sa_flags & SA_NOCLDSTOP)
770 				ps->ps_flag |= PS_NOCLDSTOP;
771 			else
772 				ps->ps_flag &= ~PS_NOCLDSTOP;
773 			if (act->sa_flags & SA_NOCLDWAIT) {
774 				/*
775 				 * Paranoia: since SA_NOCLDWAIT is implemented
776 				 * by reparenting the dying child to PID 1 (and
777 				 * trust it to reap the zombie), PID 1 itself
778 				 * is forbidden to set SA_NOCLDWAIT.
779 				 */
780 				if (p->p_pid == 1)
781 					ps->ps_flag &= ~PS_NOCLDWAIT;
782 				else
783 					ps->ps_flag |= PS_NOCLDWAIT;
784 			} else
785 				ps->ps_flag &= ~PS_NOCLDWAIT;
786 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
787 				ps->ps_flag |= PS_CLDSIGIGN;
788 			else
789 				ps->ps_flag &= ~PS_CLDSIGIGN;
790 		}
791 		/*
792 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
793 		 * and for signals set to SIG_DFL where the default is to
794 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
795 		 * have to restart the process.
796 		 */
797 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
798 		    (sigprop(sig) & SIGPROP_IGNORE &&
799 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
800 			/* never to be seen again */
801 			sigqueue_delete_proc(p, sig);
802 			if (sig != SIGCONT)
803 				/* easier in psignal */
804 				SIGADDSET(ps->ps_sigignore, sig);
805 			SIGDELSET(ps->ps_sigcatch, sig);
806 		} else {
807 			SIGDELSET(ps->ps_sigignore, sig);
808 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
809 				SIGDELSET(ps->ps_sigcatch, sig);
810 			else
811 				SIGADDSET(ps->ps_sigcatch, sig);
812 		}
813 #ifdef COMPAT_FREEBSD4
814 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
815 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
816 		    (flags & KSA_FREEBSD4) == 0)
817 			SIGDELSET(ps->ps_freebsd4, sig);
818 		else
819 			SIGADDSET(ps->ps_freebsd4, sig);
820 #endif
821 #ifdef COMPAT_43
822 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
823 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
824 		    (flags & KSA_OSIGSET) == 0)
825 			SIGDELSET(ps->ps_osigset, sig);
826 		else
827 			SIGADDSET(ps->ps_osigset, sig);
828 #endif
829 	}
830 	mtx_unlock(&ps->ps_mtx);
831 	PROC_UNLOCK(p);
832 	return (0);
833 }
834 
835 #ifndef _SYS_SYSPROTO_H_
836 struct sigaction_args {
837 	int	sig;
838 	struct	sigaction *act;
839 	struct	sigaction *oact;
840 };
841 #endif
842 int
843 sys_sigaction(struct thread *td, struct sigaction_args *uap)
844 {
845 	struct sigaction act, oact;
846 	struct sigaction *actp, *oactp;
847 	int error;
848 
849 	actp = (uap->act != NULL) ? &act : NULL;
850 	oactp = (uap->oact != NULL) ? &oact : NULL;
851 	if (actp) {
852 		error = copyin(uap->act, actp, sizeof(act));
853 		if (error)
854 			return (error);
855 	}
856 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
857 	if (oactp && !error)
858 		error = copyout(oactp, uap->oact, sizeof(oact));
859 	return (error);
860 }
861 
862 #ifdef COMPAT_FREEBSD4
863 #ifndef _SYS_SYSPROTO_H_
864 struct freebsd4_sigaction_args {
865 	int	sig;
866 	struct	sigaction *act;
867 	struct	sigaction *oact;
868 };
869 #endif
870 int
871 freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
872 {
873 	struct sigaction act, oact;
874 	struct sigaction *actp, *oactp;
875 	int error;
876 
877 	actp = (uap->act != NULL) ? &act : NULL;
878 	oactp = (uap->oact != NULL) ? &oact : NULL;
879 	if (actp) {
880 		error = copyin(uap->act, actp, sizeof(act));
881 		if (error)
882 			return (error);
883 	}
884 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
885 	if (oactp && !error)
886 		error = copyout(oactp, uap->oact, sizeof(oact));
887 	return (error);
888 }
889 #endif	/* COMAPT_FREEBSD4 */
890 
891 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
892 #ifndef _SYS_SYSPROTO_H_
893 struct osigaction_args {
894 	int	signum;
895 	struct	osigaction *nsa;
896 	struct	osigaction *osa;
897 };
898 #endif
899 int
900 osigaction(struct thread *td, struct osigaction_args *uap)
901 {
902 	struct osigaction sa;
903 	struct sigaction nsa, osa;
904 	struct sigaction *nsap, *osap;
905 	int error;
906 
907 	if (uap->signum <= 0 || uap->signum >= ONSIG)
908 		return (EINVAL);
909 
910 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
911 	osap = (uap->osa != NULL) ? &osa : NULL;
912 
913 	if (nsap) {
914 		error = copyin(uap->nsa, &sa, sizeof(sa));
915 		if (error)
916 			return (error);
917 		nsap->sa_handler = sa.sa_handler;
918 		nsap->sa_flags = sa.sa_flags;
919 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
920 	}
921 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
922 	if (osap && !error) {
923 		sa.sa_handler = osap->sa_handler;
924 		sa.sa_flags = osap->sa_flags;
925 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
926 		error = copyout(&sa, uap->osa, sizeof(sa));
927 	}
928 	return (error);
929 }
930 
931 #if !defined(__i386__)
932 /* Avoid replicating the same stub everywhere */
933 int
934 osigreturn(struct thread *td, struct osigreturn_args *uap)
935 {
936 
937 	return (nosys(td, (struct nosys_args *)uap));
938 }
939 #endif
940 #endif /* COMPAT_43 */
941 
942 /*
943  * Initialize signal state for process 0;
944  * set to ignore signals that are ignored by default.
945  */
946 void
947 siginit(struct proc *p)
948 {
949 	int i;
950 	struct sigacts *ps;
951 
952 	PROC_LOCK(p);
953 	ps = p->p_sigacts;
954 	mtx_lock(&ps->ps_mtx);
955 	for (i = 1; i <= NSIG; i++) {
956 		if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) {
957 			SIGADDSET(ps->ps_sigignore, i);
958 		}
959 	}
960 	mtx_unlock(&ps->ps_mtx);
961 	PROC_UNLOCK(p);
962 }
963 
964 /*
965  * Reset specified signal to the default disposition.
966  */
967 static void
968 sigdflt(struct sigacts *ps, int sig)
969 {
970 
971 	mtx_assert(&ps->ps_mtx, MA_OWNED);
972 	SIGDELSET(ps->ps_sigcatch, sig);
973 	if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT)
974 		SIGADDSET(ps->ps_sigignore, sig);
975 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
976 	SIGDELSET(ps->ps_siginfo, sig);
977 }
978 
979 /*
980  * Reset signals for an exec of the specified process.
981  */
982 void
983 execsigs(struct proc *p)
984 {
985 	sigset_t osigignore;
986 	struct sigacts *ps;
987 	int sig;
988 	struct thread *td;
989 
990 	/*
991 	 * Reset caught signals.  Held signals remain held
992 	 * through td_sigmask (unless they were caught,
993 	 * and are now ignored by default).
994 	 */
995 	PROC_LOCK_ASSERT(p, MA_OWNED);
996 	ps = p->p_sigacts;
997 	mtx_lock(&ps->ps_mtx);
998 	sig_drop_caught(p);
999 
1000 	/*
1001 	 * As CloudABI processes cannot modify signal handlers, fully
1002 	 * reset all signals to their default behavior. Do ignore
1003 	 * SIGPIPE, as it would otherwise be impossible to recover from
1004 	 * writes to broken pipes and sockets.
1005 	 */
1006 	if (SV_PROC_ABI(p) == SV_ABI_CLOUDABI) {
1007 		osigignore = ps->ps_sigignore;
1008 		while (SIGNOTEMPTY(osigignore)) {
1009 			sig = sig_ffs(&osigignore);
1010 			SIGDELSET(osigignore, sig);
1011 			if (sig != SIGPIPE)
1012 				sigdflt(ps, sig);
1013 		}
1014 		SIGADDSET(ps->ps_sigignore, SIGPIPE);
1015 	}
1016 
1017 	/*
1018 	 * Reset stack state to the user stack.
1019 	 * Clear set of signals caught on the signal stack.
1020 	 */
1021 	td = curthread;
1022 	MPASS(td->td_proc == p);
1023 	td->td_sigstk.ss_flags = SS_DISABLE;
1024 	td->td_sigstk.ss_size = 0;
1025 	td->td_sigstk.ss_sp = 0;
1026 	td->td_pflags &= ~TDP_ALTSTACK;
1027 	/*
1028 	 * Reset no zombies if child dies flag as Solaris does.
1029 	 */
1030 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1031 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1032 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1033 	mtx_unlock(&ps->ps_mtx);
1034 }
1035 
1036 /*
1037  * kern_sigprocmask()
1038  *
1039  *	Manipulate signal mask.
1040  */
1041 int
1042 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1043     int flags)
1044 {
1045 	sigset_t new_block, oset1;
1046 	struct proc *p;
1047 	int error;
1048 
1049 	p = td->td_proc;
1050 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1051 		PROC_LOCK_ASSERT(p, MA_OWNED);
1052 	else
1053 		PROC_LOCK(p);
1054 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1055 	    ? MA_OWNED : MA_NOTOWNED);
1056 	if (oset != NULL)
1057 		*oset = td->td_sigmask;
1058 
1059 	error = 0;
1060 	if (set != NULL) {
1061 		switch (how) {
1062 		case SIG_BLOCK:
1063 			SIG_CANTMASK(*set);
1064 			oset1 = td->td_sigmask;
1065 			SIGSETOR(td->td_sigmask, *set);
1066 			new_block = td->td_sigmask;
1067 			SIGSETNAND(new_block, oset1);
1068 			break;
1069 		case SIG_UNBLOCK:
1070 			SIGSETNAND(td->td_sigmask, *set);
1071 			signotify(td);
1072 			goto out;
1073 		case SIG_SETMASK:
1074 			SIG_CANTMASK(*set);
1075 			oset1 = td->td_sigmask;
1076 			if (flags & SIGPROCMASK_OLD)
1077 				SIGSETLO(td->td_sigmask, *set);
1078 			else
1079 				td->td_sigmask = *set;
1080 			new_block = td->td_sigmask;
1081 			SIGSETNAND(new_block, oset1);
1082 			signotify(td);
1083 			break;
1084 		default:
1085 			error = EINVAL;
1086 			goto out;
1087 		}
1088 
1089 		/*
1090 		 * The new_block set contains signals that were not previously
1091 		 * blocked, but are blocked now.
1092 		 *
1093 		 * In case we block any signal that was not previously blocked
1094 		 * for td, and process has the signal pending, try to schedule
1095 		 * signal delivery to some thread that does not block the
1096 		 * signal, possibly waking it up.
1097 		 */
1098 		if (p->p_numthreads != 1)
1099 			reschedule_signals(p, new_block, flags);
1100 	}
1101 
1102 out:
1103 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1104 		PROC_UNLOCK(p);
1105 	return (error);
1106 }
1107 
1108 #ifndef _SYS_SYSPROTO_H_
1109 struct sigprocmask_args {
1110 	int	how;
1111 	const sigset_t *set;
1112 	sigset_t *oset;
1113 };
1114 #endif
1115 int
1116 sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1117 {
1118 	sigset_t set, oset;
1119 	sigset_t *setp, *osetp;
1120 	int error;
1121 
1122 	setp = (uap->set != NULL) ? &set : NULL;
1123 	osetp = (uap->oset != NULL) ? &oset : NULL;
1124 	if (setp) {
1125 		error = copyin(uap->set, setp, sizeof(set));
1126 		if (error)
1127 			return (error);
1128 	}
1129 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1130 	if (osetp && !error) {
1131 		error = copyout(osetp, uap->oset, sizeof(oset));
1132 	}
1133 	return (error);
1134 }
1135 
1136 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1137 #ifndef _SYS_SYSPROTO_H_
1138 struct osigprocmask_args {
1139 	int	how;
1140 	osigset_t mask;
1141 };
1142 #endif
1143 int
1144 osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1145 {
1146 	sigset_t set, oset;
1147 	int error;
1148 
1149 	OSIG2SIG(uap->mask, set);
1150 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1151 	SIG2OSIG(oset, td->td_retval[0]);
1152 	return (error);
1153 }
1154 #endif /* COMPAT_43 */
1155 
1156 int
1157 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1158 {
1159 	ksiginfo_t ksi;
1160 	sigset_t set;
1161 	int error;
1162 
1163 	error = copyin(uap->set, &set, sizeof(set));
1164 	if (error) {
1165 		td->td_retval[0] = error;
1166 		return (0);
1167 	}
1168 
1169 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1170 	if (error) {
1171 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1172 			error = ERESTART;
1173 		if (error == ERESTART)
1174 			return (error);
1175 		td->td_retval[0] = error;
1176 		return (0);
1177 	}
1178 
1179 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1180 	td->td_retval[0] = error;
1181 	return (0);
1182 }
1183 
1184 int
1185 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1186 {
1187 	struct timespec ts;
1188 	struct timespec *timeout;
1189 	sigset_t set;
1190 	ksiginfo_t ksi;
1191 	int error;
1192 
1193 	if (uap->timeout) {
1194 		error = copyin(uap->timeout, &ts, sizeof(ts));
1195 		if (error)
1196 			return (error);
1197 
1198 		timeout = &ts;
1199 	} else
1200 		timeout = NULL;
1201 
1202 	error = copyin(uap->set, &set, sizeof(set));
1203 	if (error)
1204 		return (error);
1205 
1206 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1207 	if (error)
1208 		return (error);
1209 
1210 	if (uap->info)
1211 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1212 
1213 	if (error == 0)
1214 		td->td_retval[0] = ksi.ksi_signo;
1215 	return (error);
1216 }
1217 
1218 int
1219 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1220 {
1221 	ksiginfo_t ksi;
1222 	sigset_t set;
1223 	int error;
1224 
1225 	error = copyin(uap->set, &set, sizeof(set));
1226 	if (error)
1227 		return (error);
1228 
1229 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1230 	if (error)
1231 		return (error);
1232 
1233 	if (uap->info)
1234 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1235 
1236 	if (error == 0)
1237 		td->td_retval[0] = ksi.ksi_signo;
1238 	return (error);
1239 }
1240 
1241 static void
1242 proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1243 {
1244 	struct thread *thr;
1245 
1246 	FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1247 		if (thr == td)
1248 			thr->td_si = *si;
1249 		else
1250 			thr->td_si.si_signo = 0;
1251 	}
1252 }
1253 
1254 int
1255 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1256 	struct timespec *timeout)
1257 {
1258 	struct sigacts *ps;
1259 	sigset_t saved_mask, new_block;
1260 	struct proc *p;
1261 	int error, sig, timo, timevalid = 0;
1262 	struct timespec rts, ets, ts;
1263 	struct timeval tv;
1264 	bool traced;
1265 
1266 	p = td->td_proc;
1267 	error = 0;
1268 	ets.tv_sec = 0;
1269 	ets.tv_nsec = 0;
1270 	traced = false;
1271 
1272 	if (timeout != NULL) {
1273 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1274 			timevalid = 1;
1275 			getnanouptime(&rts);
1276 			timespecadd(&rts, timeout, &ets);
1277 		}
1278 	}
1279 	ksiginfo_init(ksi);
1280 	/* Some signals can not be waited for. */
1281 	SIG_CANTMASK(waitset);
1282 	ps = p->p_sigacts;
1283 	PROC_LOCK(p);
1284 	saved_mask = td->td_sigmask;
1285 	SIGSETNAND(td->td_sigmask, waitset);
1286 	for (;;) {
1287 		mtx_lock(&ps->ps_mtx);
1288 		sig = cursig(td);
1289 		mtx_unlock(&ps->ps_mtx);
1290 		KASSERT(sig >= 0, ("sig %d", sig));
1291 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1292 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1293 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1294 				error = 0;
1295 				break;
1296 			}
1297 		}
1298 
1299 		if (error != 0)
1300 			break;
1301 
1302 		/*
1303 		 * POSIX says this must be checked after looking for pending
1304 		 * signals.
1305 		 */
1306 		if (timeout != NULL) {
1307 			if (!timevalid) {
1308 				error = EINVAL;
1309 				break;
1310 			}
1311 			getnanouptime(&rts);
1312 			if (timespeccmp(&rts, &ets, >=)) {
1313 				error = EAGAIN;
1314 				break;
1315 			}
1316 			timespecsub(&ets, &rts, &ts);
1317 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1318 			timo = tvtohz(&tv);
1319 		} else {
1320 			timo = 0;
1321 		}
1322 
1323 		if (traced) {
1324 			error = EINTR;
1325 			break;
1326 		}
1327 
1328 		error = msleep(ps, &p->p_mtx, PPAUSE|PCATCH, "sigwait", timo);
1329 
1330 		if (timeout != NULL) {
1331 			if (error == ERESTART) {
1332 				/* Timeout can not be restarted. */
1333 				error = EINTR;
1334 			} else if (error == EAGAIN) {
1335 				/* We will calculate timeout by ourself. */
1336 				error = 0;
1337 			}
1338 		}
1339 
1340 		/*
1341 		 * If PTRACE_SCE or PTRACE_SCX were set after
1342 		 * userspace entered the syscall, return spurious
1343 		 * EINTR after wait was done.  Only do this as last
1344 		 * resort after rechecking for possible queued signals
1345 		 * and expired timeouts.
1346 		 */
1347 		if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1348 			traced = true;
1349 	}
1350 
1351 	new_block = saved_mask;
1352 	SIGSETNAND(new_block, td->td_sigmask);
1353 	td->td_sigmask = saved_mask;
1354 	/*
1355 	 * Fewer signals can be delivered to us, reschedule signal
1356 	 * notification.
1357 	 */
1358 	if (p->p_numthreads != 1)
1359 		reschedule_signals(p, new_block, 0);
1360 
1361 	if (error == 0) {
1362 		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1363 
1364 		if (ksi->ksi_code == SI_TIMER)
1365 			itimer_accept(p, ksi->ksi_timerid, ksi);
1366 
1367 #ifdef KTRACE
1368 		if (KTRPOINT(td, KTR_PSIG)) {
1369 			sig_t action;
1370 
1371 			mtx_lock(&ps->ps_mtx);
1372 			action = ps->ps_sigact[_SIG_IDX(sig)];
1373 			mtx_unlock(&ps->ps_mtx);
1374 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1375 		}
1376 #endif
1377 		if (sig == SIGKILL) {
1378 			proc_td_siginfo_capture(td, &ksi->ksi_info);
1379 			sigexit(td, sig);
1380 		}
1381 	}
1382 	PROC_UNLOCK(p);
1383 	return (error);
1384 }
1385 
1386 #ifndef _SYS_SYSPROTO_H_
1387 struct sigpending_args {
1388 	sigset_t	*set;
1389 };
1390 #endif
1391 int
1392 sys_sigpending(struct thread *td, struct sigpending_args *uap)
1393 {
1394 	struct proc *p = td->td_proc;
1395 	sigset_t pending;
1396 
1397 	PROC_LOCK(p);
1398 	pending = p->p_sigqueue.sq_signals;
1399 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1400 	PROC_UNLOCK(p);
1401 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1402 }
1403 
1404 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1405 #ifndef _SYS_SYSPROTO_H_
1406 struct osigpending_args {
1407 	int	dummy;
1408 };
1409 #endif
1410 int
1411 osigpending(struct thread *td, struct osigpending_args *uap)
1412 {
1413 	struct proc *p = td->td_proc;
1414 	sigset_t pending;
1415 
1416 	PROC_LOCK(p);
1417 	pending = p->p_sigqueue.sq_signals;
1418 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1419 	PROC_UNLOCK(p);
1420 	SIG2OSIG(pending, td->td_retval[0]);
1421 	return (0);
1422 }
1423 #endif /* COMPAT_43 */
1424 
1425 #if defined(COMPAT_43)
1426 /*
1427  * Generalized interface signal handler, 4.3-compatible.
1428  */
1429 #ifndef _SYS_SYSPROTO_H_
1430 struct osigvec_args {
1431 	int	signum;
1432 	struct	sigvec *nsv;
1433 	struct	sigvec *osv;
1434 };
1435 #endif
1436 /* ARGSUSED */
1437 int
1438 osigvec(struct thread *td, struct osigvec_args *uap)
1439 {
1440 	struct sigvec vec;
1441 	struct sigaction nsa, osa;
1442 	struct sigaction *nsap, *osap;
1443 	int error;
1444 
1445 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1446 		return (EINVAL);
1447 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1448 	osap = (uap->osv != NULL) ? &osa : NULL;
1449 	if (nsap) {
1450 		error = copyin(uap->nsv, &vec, sizeof(vec));
1451 		if (error)
1452 			return (error);
1453 		nsap->sa_handler = vec.sv_handler;
1454 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1455 		nsap->sa_flags = vec.sv_flags;
1456 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1457 	}
1458 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1459 	if (osap && !error) {
1460 		vec.sv_handler = osap->sa_handler;
1461 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1462 		vec.sv_flags = osap->sa_flags;
1463 		vec.sv_flags &= ~SA_NOCLDWAIT;
1464 		vec.sv_flags ^= SA_RESTART;
1465 		error = copyout(&vec, uap->osv, sizeof(vec));
1466 	}
1467 	return (error);
1468 }
1469 
1470 #ifndef _SYS_SYSPROTO_H_
1471 struct osigblock_args {
1472 	int	mask;
1473 };
1474 #endif
1475 int
1476 osigblock(struct thread *td, struct osigblock_args *uap)
1477 {
1478 	sigset_t set, oset;
1479 
1480 	OSIG2SIG(uap->mask, set);
1481 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1482 	SIG2OSIG(oset, td->td_retval[0]);
1483 	return (0);
1484 }
1485 
1486 #ifndef _SYS_SYSPROTO_H_
1487 struct osigsetmask_args {
1488 	int	mask;
1489 };
1490 #endif
1491 int
1492 osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1493 {
1494 	sigset_t set, oset;
1495 
1496 	OSIG2SIG(uap->mask, set);
1497 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1498 	SIG2OSIG(oset, td->td_retval[0]);
1499 	return (0);
1500 }
1501 #endif /* COMPAT_43 */
1502 
1503 /*
1504  * Suspend calling thread until signal, providing mask to be set in the
1505  * meantime.
1506  */
1507 #ifndef _SYS_SYSPROTO_H_
1508 struct sigsuspend_args {
1509 	const sigset_t *sigmask;
1510 };
1511 #endif
1512 /* ARGSUSED */
1513 int
1514 sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1515 {
1516 	sigset_t mask;
1517 	int error;
1518 
1519 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1520 	if (error)
1521 		return (error);
1522 	return (kern_sigsuspend(td, mask));
1523 }
1524 
1525 int
1526 kern_sigsuspend(struct thread *td, sigset_t mask)
1527 {
1528 	struct proc *p = td->td_proc;
1529 	int has_sig, sig;
1530 
1531 	/*
1532 	 * When returning from sigsuspend, we want
1533 	 * the old mask to be restored after the
1534 	 * signal handler has finished.  Thus, we
1535 	 * save it here and mark the sigacts structure
1536 	 * to indicate this.
1537 	 */
1538 	PROC_LOCK(p);
1539 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1540 	    SIGPROCMASK_PROC_LOCKED);
1541 	td->td_pflags |= TDP_OLDMASK;
1542 
1543 	/*
1544 	 * Process signals now. Otherwise, we can get spurious wakeup
1545 	 * due to signal entered process queue, but delivered to other
1546 	 * thread. But sigsuspend should return only on signal
1547 	 * delivery.
1548 	 */
1549 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1550 	for (has_sig = 0; !has_sig;) {
1551 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1552 			0) == 0)
1553 			/* void */;
1554 		thread_suspend_check(0);
1555 		mtx_lock(&p->p_sigacts->ps_mtx);
1556 		while ((sig = cursig(td)) != 0) {
1557 			KASSERT(sig >= 0, ("sig %d", sig));
1558 			has_sig += postsig(sig);
1559 		}
1560 		mtx_unlock(&p->p_sigacts->ps_mtx);
1561 
1562 		/*
1563 		 * If PTRACE_SCE or PTRACE_SCX were set after
1564 		 * userspace entered the syscall, return spurious
1565 		 * EINTR.
1566 		 */
1567 		if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1568 			has_sig += 1;
1569 	}
1570 	PROC_UNLOCK(p);
1571 	td->td_errno = EINTR;
1572 	td->td_pflags |= TDP_NERRNO;
1573 	return (EJUSTRETURN);
1574 }
1575 
1576 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1577 /*
1578  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1579  * convention: libc stub passes mask, not pointer, to save a copyin.
1580  */
1581 #ifndef _SYS_SYSPROTO_H_
1582 struct osigsuspend_args {
1583 	osigset_t mask;
1584 };
1585 #endif
1586 /* ARGSUSED */
1587 int
1588 osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1589 {
1590 	sigset_t mask;
1591 
1592 	OSIG2SIG(uap->mask, mask);
1593 	return (kern_sigsuspend(td, mask));
1594 }
1595 #endif /* COMPAT_43 */
1596 
1597 #if defined(COMPAT_43)
1598 #ifndef _SYS_SYSPROTO_H_
1599 struct osigstack_args {
1600 	struct	sigstack *nss;
1601 	struct	sigstack *oss;
1602 };
1603 #endif
1604 /* ARGSUSED */
1605 int
1606 osigstack(struct thread *td, struct osigstack_args *uap)
1607 {
1608 	struct sigstack nss, oss;
1609 	int error = 0;
1610 
1611 	if (uap->nss != NULL) {
1612 		error = copyin(uap->nss, &nss, sizeof(nss));
1613 		if (error)
1614 			return (error);
1615 	}
1616 	oss.ss_sp = td->td_sigstk.ss_sp;
1617 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1618 	if (uap->nss != NULL) {
1619 		td->td_sigstk.ss_sp = nss.ss_sp;
1620 		td->td_sigstk.ss_size = 0;
1621 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1622 		td->td_pflags |= TDP_ALTSTACK;
1623 	}
1624 	if (uap->oss != NULL)
1625 		error = copyout(&oss, uap->oss, sizeof(oss));
1626 
1627 	return (error);
1628 }
1629 #endif /* COMPAT_43 */
1630 
1631 #ifndef _SYS_SYSPROTO_H_
1632 struct sigaltstack_args {
1633 	stack_t	*ss;
1634 	stack_t	*oss;
1635 };
1636 #endif
1637 /* ARGSUSED */
1638 int
1639 sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1640 {
1641 	stack_t ss, oss;
1642 	int error;
1643 
1644 	if (uap->ss != NULL) {
1645 		error = copyin(uap->ss, &ss, sizeof(ss));
1646 		if (error)
1647 			return (error);
1648 	}
1649 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1650 	    (uap->oss != NULL) ? &oss : NULL);
1651 	if (error)
1652 		return (error);
1653 	if (uap->oss != NULL)
1654 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1655 	return (error);
1656 }
1657 
1658 int
1659 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1660 {
1661 	struct proc *p = td->td_proc;
1662 	int oonstack;
1663 
1664 	oonstack = sigonstack(cpu_getstack(td));
1665 
1666 	if (oss != NULL) {
1667 		*oss = td->td_sigstk;
1668 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1669 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1670 	}
1671 
1672 	if (ss != NULL) {
1673 		if (oonstack)
1674 			return (EPERM);
1675 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1676 			return (EINVAL);
1677 		if (!(ss->ss_flags & SS_DISABLE)) {
1678 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1679 				return (ENOMEM);
1680 
1681 			td->td_sigstk = *ss;
1682 			td->td_pflags |= TDP_ALTSTACK;
1683 		} else {
1684 			td->td_pflags &= ~TDP_ALTSTACK;
1685 		}
1686 	}
1687 	return (0);
1688 }
1689 
1690 struct killpg1_ctx {
1691 	struct thread *td;
1692 	ksiginfo_t *ksi;
1693 	int sig;
1694 	bool sent;
1695 	bool found;
1696 	int ret;
1697 };
1698 
1699 static void
1700 killpg1_sendsig(struct proc *p, bool notself, struct killpg1_ctx *arg)
1701 {
1702 	int err;
1703 
1704 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1705 	    (notself && p == arg->td->td_proc) || p->p_state == PRS_NEW)
1706 		return;
1707 	PROC_LOCK(p);
1708 	err = p_cansignal(arg->td, p, arg->sig);
1709 	if (err == 0 && arg->sig != 0)
1710 		pksignal(p, arg->sig, arg->ksi);
1711 	PROC_UNLOCK(p);
1712 	if (err != ESRCH)
1713 		arg->found = true;
1714 	if (err == 0)
1715 		arg->sent = true;
1716 	else if (arg->ret == 0 && err != ESRCH && err != EPERM)
1717 		arg->ret = err;
1718 }
1719 
1720 /*
1721  * Common code for kill process group/broadcast kill.
1722  * cp is calling process.
1723  */
1724 static int
1725 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1726 {
1727 	struct proc *p;
1728 	struct pgrp *pgrp;
1729 	struct killpg1_ctx arg;
1730 
1731 	arg.td = td;
1732 	arg.ksi = ksi;
1733 	arg.sig = sig;
1734 	arg.sent = false;
1735 	arg.found = false;
1736 	arg.ret = 0;
1737 	if (all) {
1738 		/*
1739 		 * broadcast
1740 		 */
1741 		sx_slock(&allproc_lock);
1742 		FOREACH_PROC_IN_SYSTEM(p) {
1743 			killpg1_sendsig(p, true, &arg);
1744 		}
1745 		sx_sunlock(&allproc_lock);
1746 	} else {
1747 		sx_slock(&proctree_lock);
1748 		if (pgid == 0) {
1749 			/*
1750 			 * zero pgid means send to my process group.
1751 			 */
1752 			pgrp = td->td_proc->p_pgrp;
1753 			PGRP_LOCK(pgrp);
1754 		} else {
1755 			pgrp = pgfind(pgid);
1756 			if (pgrp == NULL) {
1757 				sx_sunlock(&proctree_lock);
1758 				return (ESRCH);
1759 			}
1760 		}
1761 		sx_sunlock(&proctree_lock);
1762 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1763 			killpg1_sendsig(p, false, &arg);
1764 		}
1765 		PGRP_UNLOCK(pgrp);
1766 	}
1767 	MPASS(arg.ret != 0 || arg.found || !arg.sent);
1768 	if (arg.ret == 0 && !arg.sent)
1769 		arg.ret = arg.found ? EPERM : ESRCH;
1770 	return (arg.ret);
1771 }
1772 
1773 #ifndef _SYS_SYSPROTO_H_
1774 struct kill_args {
1775 	int	pid;
1776 	int	signum;
1777 };
1778 #endif
1779 /* ARGSUSED */
1780 int
1781 sys_kill(struct thread *td, struct kill_args *uap)
1782 {
1783 
1784 	return (kern_kill(td, uap->pid, uap->signum));
1785 }
1786 
1787 int
1788 kern_kill(struct thread *td, pid_t pid, int signum)
1789 {
1790 	ksiginfo_t ksi;
1791 	struct proc *p;
1792 	int error;
1793 
1794 	/*
1795 	 * A process in capability mode can send signals only to himself.
1796 	 * The main rationale behind this is that abort(3) is implemented as
1797 	 * kill(getpid(), SIGABRT).
1798 	 */
1799 	if (IN_CAPABILITY_MODE(td) && pid != td->td_proc->p_pid)
1800 		return (ECAPMODE);
1801 
1802 	AUDIT_ARG_SIGNUM(signum);
1803 	AUDIT_ARG_PID(pid);
1804 	if ((u_int)signum > _SIG_MAXSIG)
1805 		return (EINVAL);
1806 
1807 	ksiginfo_init(&ksi);
1808 	ksi.ksi_signo = signum;
1809 	ksi.ksi_code = SI_USER;
1810 	ksi.ksi_pid = td->td_proc->p_pid;
1811 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1812 
1813 	if (pid > 0) {
1814 		/* kill single process */
1815 		if ((p = pfind_any(pid)) == NULL)
1816 			return (ESRCH);
1817 		AUDIT_ARG_PROCESS(p);
1818 		error = p_cansignal(td, p, signum);
1819 		if (error == 0 && signum)
1820 			pksignal(p, signum, &ksi);
1821 		PROC_UNLOCK(p);
1822 		return (error);
1823 	}
1824 	switch (pid) {
1825 	case -1:		/* broadcast signal */
1826 		return (killpg1(td, signum, 0, 1, &ksi));
1827 	case 0:			/* signal own process group */
1828 		return (killpg1(td, signum, 0, 0, &ksi));
1829 	default:		/* negative explicit process group */
1830 		return (killpg1(td, signum, -pid, 0, &ksi));
1831 	}
1832 	/* NOTREACHED */
1833 }
1834 
1835 int
1836 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1837 {
1838 	struct proc *p;
1839 	int error;
1840 
1841 	AUDIT_ARG_SIGNUM(uap->signum);
1842 	AUDIT_ARG_FD(uap->fd);
1843 	if ((u_int)uap->signum > _SIG_MAXSIG)
1844 		return (EINVAL);
1845 
1846 	error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1847 	if (error)
1848 		return (error);
1849 	AUDIT_ARG_PROCESS(p);
1850 	error = p_cansignal(td, p, uap->signum);
1851 	if (error == 0 && uap->signum)
1852 		kern_psignal(p, uap->signum);
1853 	PROC_UNLOCK(p);
1854 	return (error);
1855 }
1856 
1857 #if defined(COMPAT_43)
1858 #ifndef _SYS_SYSPROTO_H_
1859 struct okillpg_args {
1860 	int	pgid;
1861 	int	signum;
1862 };
1863 #endif
1864 /* ARGSUSED */
1865 int
1866 okillpg(struct thread *td, struct okillpg_args *uap)
1867 {
1868 	ksiginfo_t ksi;
1869 
1870 	AUDIT_ARG_SIGNUM(uap->signum);
1871 	AUDIT_ARG_PID(uap->pgid);
1872 	if ((u_int)uap->signum > _SIG_MAXSIG)
1873 		return (EINVAL);
1874 
1875 	ksiginfo_init(&ksi);
1876 	ksi.ksi_signo = uap->signum;
1877 	ksi.ksi_code = SI_USER;
1878 	ksi.ksi_pid = td->td_proc->p_pid;
1879 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1880 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1881 }
1882 #endif /* COMPAT_43 */
1883 
1884 #ifndef _SYS_SYSPROTO_H_
1885 struct sigqueue_args {
1886 	pid_t pid;
1887 	int signum;
1888 	/* union sigval */ void *value;
1889 };
1890 #endif
1891 int
1892 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1893 {
1894 	union sigval sv;
1895 
1896 	sv.sival_ptr = uap->value;
1897 
1898 	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
1899 }
1900 
1901 int
1902 kern_sigqueue(struct thread *td, pid_t pid, int signum, union sigval *value)
1903 {
1904 	ksiginfo_t ksi;
1905 	struct proc *p;
1906 	int error;
1907 
1908 	if ((u_int)signum > _SIG_MAXSIG)
1909 		return (EINVAL);
1910 
1911 	/*
1912 	 * Specification says sigqueue can only send signal to
1913 	 * single process.
1914 	 */
1915 	if (pid <= 0)
1916 		return (EINVAL);
1917 
1918 	if ((p = pfind_any(pid)) == NULL)
1919 		return (ESRCH);
1920 	error = p_cansignal(td, p, signum);
1921 	if (error == 0 && signum != 0) {
1922 		ksiginfo_init(&ksi);
1923 		ksi.ksi_flags = KSI_SIGQ;
1924 		ksi.ksi_signo = signum;
1925 		ksi.ksi_code = SI_QUEUE;
1926 		ksi.ksi_pid = td->td_proc->p_pid;
1927 		ksi.ksi_uid = td->td_ucred->cr_ruid;
1928 		ksi.ksi_value = *value;
1929 		error = pksignal(p, ksi.ksi_signo, &ksi);
1930 	}
1931 	PROC_UNLOCK(p);
1932 	return (error);
1933 }
1934 
1935 /*
1936  * Send a signal to a process group.
1937  */
1938 void
1939 gsignal(int pgid, int sig, ksiginfo_t *ksi)
1940 {
1941 	struct pgrp *pgrp;
1942 
1943 	if (pgid != 0) {
1944 		sx_slock(&proctree_lock);
1945 		pgrp = pgfind(pgid);
1946 		sx_sunlock(&proctree_lock);
1947 		if (pgrp != NULL) {
1948 			pgsignal(pgrp, sig, 0, ksi);
1949 			PGRP_UNLOCK(pgrp);
1950 		}
1951 	}
1952 }
1953 
1954 /*
1955  * Send a signal to a process group.  If checktty is 1,
1956  * limit to members which have a controlling terminal.
1957  */
1958 void
1959 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
1960 {
1961 	struct proc *p;
1962 
1963 	if (pgrp) {
1964 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
1965 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1966 			PROC_LOCK(p);
1967 			if (p->p_state == PRS_NORMAL &&
1968 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
1969 				pksignal(p, sig, ksi);
1970 			PROC_UNLOCK(p);
1971 		}
1972 	}
1973 }
1974 
1975 /*
1976  * Recalculate the signal mask and reset the signal disposition after
1977  * usermode frame for delivery is formed.  Should be called after
1978  * mach-specific routine, because sysent->sv_sendsig() needs correct
1979  * ps_siginfo and signal mask.
1980  */
1981 static void
1982 postsig_done(int sig, struct thread *td, struct sigacts *ps)
1983 {
1984 	sigset_t mask;
1985 
1986 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1987 	td->td_ru.ru_nsignals++;
1988 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
1989 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
1990 		SIGADDSET(mask, sig);
1991 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
1992 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
1993 	if (SIGISMEMBER(ps->ps_sigreset, sig))
1994 		sigdflt(ps, sig);
1995 }
1996 
1997 /*
1998  * Send a signal caused by a trap to the current thread.  If it will be
1999  * caught immediately, deliver it with correct code.  Otherwise, post it
2000  * normally.
2001  */
2002 void
2003 trapsignal(struct thread *td, ksiginfo_t *ksi)
2004 {
2005 	struct sigacts *ps;
2006 	struct proc *p;
2007 	sigset_t sigmask;
2008 	int code, sig;
2009 
2010 	p = td->td_proc;
2011 	sig = ksi->ksi_signo;
2012 	code = ksi->ksi_code;
2013 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
2014 
2015 	sigfastblock_fetch(td);
2016 	PROC_LOCK(p);
2017 	ps = p->p_sigacts;
2018 	mtx_lock(&ps->ps_mtx);
2019 	sigmask = td->td_sigmask;
2020 	if (td->td_sigblock_val != 0)
2021 		SIGSETOR(sigmask, fastblock_mask);
2022 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2023 	    !SIGISMEMBER(sigmask, sig)) {
2024 #ifdef KTRACE
2025 		if (KTRPOINT(curthread, KTR_PSIG))
2026 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2027 			    &td->td_sigmask, code);
2028 #endif
2029 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2030 				ksi, &td->td_sigmask);
2031 		postsig_done(sig, td, ps);
2032 		mtx_unlock(&ps->ps_mtx);
2033 	} else {
2034 		/*
2035 		 * Avoid a possible infinite loop if the thread
2036 		 * masking the signal or process is ignoring the
2037 		 * signal.
2038 		 */
2039 		if (kern_forcesigexit && (SIGISMEMBER(sigmask, sig) ||
2040 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2041 			SIGDELSET(td->td_sigmask, sig);
2042 			SIGDELSET(ps->ps_sigcatch, sig);
2043 			SIGDELSET(ps->ps_sigignore, sig);
2044 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2045 			td->td_pflags &= ~TDP_SIGFASTBLOCK;
2046 			td->td_sigblock_val = 0;
2047 		}
2048 		mtx_unlock(&ps->ps_mtx);
2049 		p->p_sig = sig;		/* XXX to verify code */
2050 		tdsendsignal(p, td, sig, ksi);
2051 	}
2052 	PROC_UNLOCK(p);
2053 }
2054 
2055 static struct thread *
2056 sigtd(struct proc *p, int sig, bool fast_sigblock)
2057 {
2058 	struct thread *td, *signal_td;
2059 
2060 	PROC_LOCK_ASSERT(p, MA_OWNED);
2061 	MPASS(!fast_sigblock || p == curproc);
2062 
2063 	/*
2064 	 * Check if current thread can handle the signal without
2065 	 * switching context to another thread.
2066 	 */
2067 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig) &&
2068 	    (!fast_sigblock || curthread->td_sigblock_val == 0))
2069 		return (curthread);
2070 	signal_td = NULL;
2071 	FOREACH_THREAD_IN_PROC(p, td) {
2072 		if (!SIGISMEMBER(td->td_sigmask, sig) && (!fast_sigblock ||
2073 		    td != curthread || td->td_sigblock_val == 0)) {
2074 			signal_td = td;
2075 			break;
2076 		}
2077 	}
2078 	if (signal_td == NULL)
2079 		signal_td = FIRST_THREAD_IN_PROC(p);
2080 	return (signal_td);
2081 }
2082 
2083 /*
2084  * Send the signal to the process.  If the signal has an action, the action
2085  * is usually performed by the target process rather than the caller; we add
2086  * the signal to the set of pending signals for the process.
2087  *
2088  * Exceptions:
2089  *   o When a stop signal is sent to a sleeping process that takes the
2090  *     default action, the process is stopped without awakening it.
2091  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2092  *     regardless of the signal action (eg, blocked or ignored).
2093  *
2094  * Other ignored signals are discarded immediately.
2095  *
2096  * NB: This function may be entered from the debugger via the "kill" DDB
2097  * command.  There is little that can be done to mitigate the possibly messy
2098  * side effects of this unwise possibility.
2099  */
2100 void
2101 kern_psignal(struct proc *p, int sig)
2102 {
2103 	ksiginfo_t ksi;
2104 
2105 	ksiginfo_init(&ksi);
2106 	ksi.ksi_signo = sig;
2107 	ksi.ksi_code = SI_KERNEL;
2108 	(void) tdsendsignal(p, NULL, sig, &ksi);
2109 }
2110 
2111 int
2112 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2113 {
2114 
2115 	return (tdsendsignal(p, NULL, sig, ksi));
2116 }
2117 
2118 /* Utility function for finding a thread to send signal event to. */
2119 int
2120 sigev_findtd(struct proc *p ,struct sigevent *sigev, struct thread **ttd)
2121 {
2122 	struct thread *td;
2123 
2124 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2125 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2126 		if (td == NULL)
2127 			return (ESRCH);
2128 		*ttd = td;
2129 	} else {
2130 		*ttd = NULL;
2131 		PROC_LOCK(p);
2132 	}
2133 	return (0);
2134 }
2135 
2136 void
2137 tdsignal(struct thread *td, int sig)
2138 {
2139 	ksiginfo_t ksi;
2140 
2141 	ksiginfo_init(&ksi);
2142 	ksi.ksi_signo = sig;
2143 	ksi.ksi_code = SI_KERNEL;
2144 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2145 }
2146 
2147 void
2148 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2149 {
2150 
2151 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2152 }
2153 
2154 int
2155 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2156 {
2157 	sig_t action;
2158 	sigqueue_t *sigqueue;
2159 	int prop;
2160 	struct sigacts *ps;
2161 	int intrval;
2162 	int ret = 0;
2163 	int wakeup_swapper;
2164 
2165 	MPASS(td == NULL || p == td->td_proc);
2166 	PROC_LOCK_ASSERT(p, MA_OWNED);
2167 
2168 	if (!_SIG_VALID(sig))
2169 		panic("%s(): invalid signal %d", __func__, sig);
2170 
2171 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2172 
2173 	/*
2174 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2175 	 */
2176 	if (p->p_state == PRS_ZOMBIE) {
2177 		if (ksi && (ksi->ksi_flags & KSI_INS))
2178 			ksiginfo_tryfree(ksi);
2179 		return (ret);
2180 	}
2181 
2182 	ps = p->p_sigacts;
2183 	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2184 	prop = sigprop(sig);
2185 
2186 	if (td == NULL) {
2187 		td = sigtd(p, sig, false);
2188 		sigqueue = &p->p_sigqueue;
2189 	} else
2190 		sigqueue = &td->td_sigqueue;
2191 
2192 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2193 
2194 	/*
2195 	 * If the signal is being ignored,
2196 	 * then we forget about it immediately.
2197 	 * (Note: we don't set SIGCONT in ps_sigignore,
2198 	 * and if it is set to SIG_IGN,
2199 	 * action will be SIG_DFL here.)
2200 	 */
2201 	mtx_lock(&ps->ps_mtx);
2202 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2203 		SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2204 
2205 		mtx_unlock(&ps->ps_mtx);
2206 		if (ksi && (ksi->ksi_flags & KSI_INS))
2207 			ksiginfo_tryfree(ksi);
2208 		return (ret);
2209 	}
2210 	if (SIGISMEMBER(td->td_sigmask, sig))
2211 		action = SIG_HOLD;
2212 	else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2213 		action = SIG_CATCH;
2214 	else
2215 		action = SIG_DFL;
2216 	if (SIGISMEMBER(ps->ps_sigintr, sig))
2217 		intrval = EINTR;
2218 	else
2219 		intrval = ERESTART;
2220 	mtx_unlock(&ps->ps_mtx);
2221 
2222 	if (prop & SIGPROP_CONT)
2223 		sigqueue_delete_stopmask_proc(p);
2224 	else if (prop & SIGPROP_STOP) {
2225 		/*
2226 		 * If sending a tty stop signal to a member of an orphaned
2227 		 * process group, discard the signal here if the action
2228 		 * is default; don't stop the process below if sleeping,
2229 		 * and don't clear any pending SIGCONT.
2230 		 */
2231 		if ((prop & SIGPROP_TTYSTOP) &&
2232 		    (p->p_pgrp->pg_jobc == 0) &&
2233 		    (action == SIG_DFL)) {
2234 			if (ksi && (ksi->ksi_flags & KSI_INS))
2235 				ksiginfo_tryfree(ksi);
2236 			return (ret);
2237 		}
2238 		sigqueue_delete_proc(p, SIGCONT);
2239 		if (p->p_flag & P_CONTINUED) {
2240 			p->p_flag &= ~P_CONTINUED;
2241 			PROC_LOCK(p->p_pptr);
2242 			sigqueue_take(p->p_ksi);
2243 			PROC_UNLOCK(p->p_pptr);
2244 		}
2245 	}
2246 
2247 	ret = sigqueue_add(sigqueue, sig, ksi);
2248 	if (ret != 0)
2249 		return (ret);
2250 	signotify(td);
2251 	/*
2252 	 * Defer further processing for signals which are held,
2253 	 * except that stopped processes must be continued by SIGCONT.
2254 	 */
2255 	if (action == SIG_HOLD &&
2256 	    !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2257 		return (ret);
2258 
2259 	/* SIGKILL: Remove procfs STOPEVENTs. */
2260 	if (sig == SIGKILL) {
2261 		/* from procfs_ioctl.c: PIOCBIC */
2262 		p->p_stops = 0;
2263 		/* from procfs_ioctl.c: PIOCCONT */
2264 		p->p_step = 0;
2265 		wakeup(&p->p_step);
2266 	}
2267 	wakeup_swapper = 0;
2268 
2269 	/*
2270 	 * Some signals have a process-wide effect and a per-thread
2271 	 * component.  Most processing occurs when the process next
2272 	 * tries to cross the user boundary, however there are some
2273 	 * times when processing needs to be done immediately, such as
2274 	 * waking up threads so that they can cross the user boundary.
2275 	 * We try to do the per-process part here.
2276 	 */
2277 	if (P_SHOULDSTOP(p)) {
2278 		KASSERT(!(p->p_flag & P_WEXIT),
2279 		    ("signal to stopped but exiting process"));
2280 		if (sig == SIGKILL) {
2281 			/*
2282 			 * If traced process is already stopped,
2283 			 * then no further action is necessary.
2284 			 */
2285 			if (p->p_flag & P_TRACED)
2286 				goto out;
2287 			/*
2288 			 * SIGKILL sets process running.
2289 			 * It will die elsewhere.
2290 			 * All threads must be restarted.
2291 			 */
2292 			p->p_flag &= ~P_STOPPED_SIG;
2293 			goto runfast;
2294 		}
2295 
2296 		if (prop & SIGPROP_CONT) {
2297 			/*
2298 			 * If traced process is already stopped,
2299 			 * then no further action is necessary.
2300 			 */
2301 			if (p->p_flag & P_TRACED)
2302 				goto out;
2303 			/*
2304 			 * If SIGCONT is default (or ignored), we continue the
2305 			 * process but don't leave the signal in sigqueue as
2306 			 * it has no further action.  If SIGCONT is held, we
2307 			 * continue the process and leave the signal in
2308 			 * sigqueue.  If the process catches SIGCONT, let it
2309 			 * handle the signal itself.  If it isn't waiting on
2310 			 * an event, it goes back to run state.
2311 			 * Otherwise, process goes back to sleep state.
2312 			 */
2313 			p->p_flag &= ~P_STOPPED_SIG;
2314 			PROC_SLOCK(p);
2315 			if (p->p_numthreads == p->p_suspcount) {
2316 				PROC_SUNLOCK(p);
2317 				p->p_flag |= P_CONTINUED;
2318 				p->p_xsig = SIGCONT;
2319 				PROC_LOCK(p->p_pptr);
2320 				childproc_continued(p);
2321 				PROC_UNLOCK(p->p_pptr);
2322 				PROC_SLOCK(p);
2323 			}
2324 			if (action == SIG_DFL) {
2325 				thread_unsuspend(p);
2326 				PROC_SUNLOCK(p);
2327 				sigqueue_delete(sigqueue, sig);
2328 				goto out;
2329 			}
2330 			if (action == SIG_CATCH) {
2331 				/*
2332 				 * The process wants to catch it so it needs
2333 				 * to run at least one thread, but which one?
2334 				 */
2335 				PROC_SUNLOCK(p);
2336 				goto runfast;
2337 			}
2338 			/*
2339 			 * The signal is not ignored or caught.
2340 			 */
2341 			thread_unsuspend(p);
2342 			PROC_SUNLOCK(p);
2343 			goto out;
2344 		}
2345 
2346 		if (prop & SIGPROP_STOP) {
2347 			/*
2348 			 * If traced process is already stopped,
2349 			 * then no further action is necessary.
2350 			 */
2351 			if (p->p_flag & P_TRACED)
2352 				goto out;
2353 			/*
2354 			 * Already stopped, don't need to stop again
2355 			 * (If we did the shell could get confused).
2356 			 * Just make sure the signal STOP bit set.
2357 			 */
2358 			p->p_flag |= P_STOPPED_SIG;
2359 			sigqueue_delete(sigqueue, sig);
2360 			goto out;
2361 		}
2362 
2363 		/*
2364 		 * All other kinds of signals:
2365 		 * If a thread is sleeping interruptibly, simulate a
2366 		 * wakeup so that when it is continued it will be made
2367 		 * runnable and can look at the signal.  However, don't make
2368 		 * the PROCESS runnable, leave it stopped.
2369 		 * It may run a bit until it hits a thread_suspend_check().
2370 		 */
2371 		PROC_SLOCK(p);
2372 		thread_lock(td);
2373 		if (TD_CAN_ABORT(td))
2374 			wakeup_swapper = sleepq_abort(td, intrval);
2375 		else
2376 			thread_unlock(td);
2377 		PROC_SUNLOCK(p);
2378 		goto out;
2379 		/*
2380 		 * Mutexes are short lived. Threads waiting on them will
2381 		 * hit thread_suspend_check() soon.
2382 		 */
2383 	} else if (p->p_state == PRS_NORMAL) {
2384 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2385 			tdsigwakeup(td, sig, action, intrval);
2386 			goto out;
2387 		}
2388 
2389 		MPASS(action == SIG_DFL);
2390 
2391 		if (prop & SIGPROP_STOP) {
2392 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2393 				goto out;
2394 			p->p_flag |= P_STOPPED_SIG;
2395 			p->p_xsig = sig;
2396 			PROC_SLOCK(p);
2397 			wakeup_swapper = sig_suspend_threads(td, p, 1);
2398 			if (p->p_numthreads == p->p_suspcount) {
2399 				/*
2400 				 * only thread sending signal to another
2401 				 * process can reach here, if thread is sending
2402 				 * signal to its process, because thread does
2403 				 * not suspend itself here, p_numthreads
2404 				 * should never be equal to p_suspcount.
2405 				 */
2406 				thread_stopped(p);
2407 				PROC_SUNLOCK(p);
2408 				sigqueue_delete_proc(p, p->p_xsig);
2409 			} else
2410 				PROC_SUNLOCK(p);
2411 			goto out;
2412 		}
2413 	} else {
2414 		/* Not in "NORMAL" state. discard the signal. */
2415 		sigqueue_delete(sigqueue, sig);
2416 		goto out;
2417 	}
2418 
2419 	/*
2420 	 * The process is not stopped so we need to apply the signal to all the
2421 	 * running threads.
2422 	 */
2423 runfast:
2424 	tdsigwakeup(td, sig, action, intrval);
2425 	PROC_SLOCK(p);
2426 	thread_unsuspend(p);
2427 	PROC_SUNLOCK(p);
2428 out:
2429 	/* If we jump here, proc slock should not be owned. */
2430 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2431 	if (wakeup_swapper)
2432 		kick_proc0();
2433 
2434 	return (ret);
2435 }
2436 
2437 /*
2438  * The force of a signal has been directed against a single
2439  * thread.  We need to see what we can do about knocking it
2440  * out of any sleep it may be in etc.
2441  */
2442 static void
2443 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2444 {
2445 	struct proc *p = td->td_proc;
2446 	int prop, wakeup_swapper;
2447 
2448 	PROC_LOCK_ASSERT(p, MA_OWNED);
2449 	prop = sigprop(sig);
2450 
2451 	PROC_SLOCK(p);
2452 	thread_lock(td);
2453 	/*
2454 	 * Bring the priority of a thread up if we want it to get
2455 	 * killed in this lifetime.  Be careful to avoid bumping the
2456 	 * priority of the idle thread, since we still allow to signal
2457 	 * kernel processes.
2458 	 */
2459 	if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2460 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2461 		sched_prio(td, PUSER);
2462 	if (TD_ON_SLEEPQ(td)) {
2463 		/*
2464 		 * If thread is sleeping uninterruptibly
2465 		 * we can't interrupt the sleep... the signal will
2466 		 * be noticed when the process returns through
2467 		 * trap() or syscall().
2468 		 */
2469 		if ((td->td_flags & TDF_SINTR) == 0)
2470 			goto out;
2471 		/*
2472 		 * If SIGCONT is default (or ignored) and process is
2473 		 * asleep, we are finished; the process should not
2474 		 * be awakened.
2475 		 */
2476 		if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2477 			thread_unlock(td);
2478 			PROC_SUNLOCK(p);
2479 			sigqueue_delete(&p->p_sigqueue, sig);
2480 			/*
2481 			 * It may be on either list in this state.
2482 			 * Remove from both for now.
2483 			 */
2484 			sigqueue_delete(&td->td_sigqueue, sig);
2485 			return;
2486 		}
2487 
2488 		/*
2489 		 * Don't awaken a sleeping thread for SIGSTOP if the
2490 		 * STOP signal is deferred.
2491 		 */
2492 		if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2493 		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2494 			goto out;
2495 
2496 		/*
2497 		 * Give low priority threads a better chance to run.
2498 		 */
2499 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2500 			sched_prio(td, PUSER);
2501 
2502 		wakeup_swapper = sleepq_abort(td, intrval);
2503 		PROC_SUNLOCK(p);
2504 		if (wakeup_swapper)
2505 			kick_proc0();
2506 		return;
2507 	}
2508 
2509 	/*
2510 	 * Other states do nothing with the signal immediately,
2511 	 * other than kicking ourselves if we are running.
2512 	 * It will either never be noticed, or noticed very soon.
2513 	 */
2514 #ifdef SMP
2515 	if (TD_IS_RUNNING(td) && td != curthread)
2516 		forward_signal(td);
2517 #endif
2518 
2519 out:
2520 	PROC_SUNLOCK(p);
2521 	thread_unlock(td);
2522 }
2523 
2524 static int
2525 sig_suspend_threads(struct thread *td, struct proc *p, int sending)
2526 {
2527 	struct thread *td2;
2528 	int wakeup_swapper;
2529 
2530 	PROC_LOCK_ASSERT(p, MA_OWNED);
2531 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2532 	MPASS(sending || td == curthread);
2533 
2534 	wakeup_swapper = 0;
2535 	FOREACH_THREAD_IN_PROC(p, td2) {
2536 		thread_lock(td2);
2537 		td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
2538 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2539 		    (td2->td_flags & TDF_SINTR)) {
2540 			if (td2->td_flags & TDF_SBDRY) {
2541 				/*
2542 				 * Once a thread is asleep with
2543 				 * TDF_SBDRY and without TDF_SERESTART
2544 				 * or TDF_SEINTR set, it should never
2545 				 * become suspended due to this check.
2546 				 */
2547 				KASSERT(!TD_IS_SUSPENDED(td2),
2548 				    ("thread with deferred stops suspended"));
2549 				if (TD_SBDRY_INTR(td2)) {
2550 					wakeup_swapper |= sleepq_abort(td2,
2551 					    TD_SBDRY_ERRNO(td2));
2552 					continue;
2553 				}
2554 			} else if (!TD_IS_SUSPENDED(td2))
2555 				thread_suspend_one(td2);
2556 		} else if (!TD_IS_SUSPENDED(td2)) {
2557 			if (sending || td != td2)
2558 				td2->td_flags |= TDF_ASTPENDING;
2559 #ifdef SMP
2560 			if (TD_IS_RUNNING(td2) && td2 != td)
2561 				forward_signal(td2);
2562 #endif
2563 		}
2564 		thread_unlock(td2);
2565 	}
2566 	return (wakeup_swapper);
2567 }
2568 
2569 /*
2570  * Stop the process for an event deemed interesting to the debugger. If si is
2571  * non-NULL, this is a signal exchange; the new signal requested by the
2572  * debugger will be returned for handling. If si is NULL, this is some other
2573  * type of interesting event. The debugger may request a signal be delivered in
2574  * that case as well, however it will be deferred until it can be handled.
2575  */
2576 int
2577 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2578 {
2579 	struct proc *p = td->td_proc;
2580 	struct thread *td2;
2581 	ksiginfo_t ksi;
2582 
2583 	PROC_LOCK_ASSERT(p, MA_OWNED);
2584 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2585 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2586 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2587 
2588 	td->td_xsig = sig;
2589 
2590 	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2591 		td->td_dbgflags |= TDB_XSIG;
2592 		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2593 		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2594 		PROC_SLOCK(p);
2595 		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2596 			if (P_KILLED(p)) {
2597 				/*
2598 				 * Ensure that, if we've been PT_KILLed, the
2599 				 * exit status reflects that. Another thread
2600 				 * may also be in ptracestop(), having just
2601 				 * received the SIGKILL, but this thread was
2602 				 * unsuspended first.
2603 				 */
2604 				td->td_dbgflags &= ~TDB_XSIG;
2605 				td->td_xsig = SIGKILL;
2606 				p->p_ptevents = 0;
2607 				break;
2608 			}
2609 			if (p->p_flag & P_SINGLE_EXIT &&
2610 			    !(td->td_dbgflags & TDB_EXIT)) {
2611 				/*
2612 				 * Ignore ptrace stops except for thread exit
2613 				 * events when the process exits.
2614 				 */
2615 				td->td_dbgflags &= ~TDB_XSIG;
2616 				PROC_SUNLOCK(p);
2617 				return (0);
2618 			}
2619 
2620 			/*
2621 			 * Make wait(2) work.  Ensure that right after the
2622 			 * attach, the thread which was decided to become the
2623 			 * leader of attach gets reported to the waiter.
2624 			 * Otherwise, just avoid overwriting another thread's
2625 			 * assignment to p_xthread.  If another thread has
2626 			 * already set p_xthread, the current thread will get
2627 			 * a chance to report itself upon the next iteration.
2628 			 */
2629 			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2630 			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2631 			    p->p_xthread == NULL)) {
2632 				p->p_xsig = sig;
2633 				p->p_xthread = td;
2634 
2635 				/*
2636 				 * If we are on sleepqueue already,
2637 				 * let sleepqueue code decide if it
2638 				 * needs to go sleep after attach.
2639 				 */
2640 				if (td->td_wchan == NULL)
2641 					td->td_dbgflags &= ~TDB_FSTP;
2642 
2643 				p->p_flag2 &= ~P2_PTRACE_FSTP;
2644 				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2645 				sig_suspend_threads(td, p, 0);
2646 			}
2647 			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2648 				td->td_dbgflags &= ~TDB_STOPATFORK;
2649 			}
2650 stopme:
2651 			thread_suspend_switch(td, p);
2652 			if (p->p_xthread == td)
2653 				p->p_xthread = NULL;
2654 			if (!(p->p_flag & P_TRACED))
2655 				break;
2656 			if (td->td_dbgflags & TDB_SUSPEND) {
2657 				if (p->p_flag & P_SINGLE_EXIT)
2658 					break;
2659 				goto stopme;
2660 			}
2661 		}
2662 		PROC_SUNLOCK(p);
2663 	}
2664 
2665 	if (si != NULL && sig == td->td_xsig) {
2666 		/* Parent wants us to take the original signal unchanged. */
2667 		si->ksi_flags |= KSI_HEAD;
2668 		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2669 			si->ksi_signo = 0;
2670 	} else if (td->td_xsig != 0) {
2671 		/*
2672 		 * If parent wants us to take a new signal, then it will leave
2673 		 * it in td->td_xsig; otherwise we just look for signals again.
2674 		 */
2675 		ksiginfo_init(&ksi);
2676 		ksi.ksi_signo = td->td_xsig;
2677 		ksi.ksi_flags |= KSI_PTRACE;
2678 		td2 = sigtd(p, td->td_xsig, false);
2679 		tdsendsignal(p, td2, td->td_xsig, &ksi);
2680 		if (td != td2)
2681 			return (0);
2682 	}
2683 
2684 	return (td->td_xsig);
2685 }
2686 
2687 static void
2688 reschedule_signals(struct proc *p, sigset_t block, int flags)
2689 {
2690 	struct sigacts *ps;
2691 	struct thread *td;
2692 	int sig;
2693 	bool fastblk, pslocked;
2694 
2695 	PROC_LOCK_ASSERT(p, MA_OWNED);
2696 	ps = p->p_sigacts;
2697 	pslocked = (flags & SIGPROCMASK_PS_LOCKED) != 0;
2698 	mtx_assert(&ps->ps_mtx, pslocked ? MA_OWNED : MA_NOTOWNED);
2699 	if (SIGISEMPTY(p->p_siglist))
2700 		return;
2701 	SIGSETAND(block, p->p_siglist);
2702 	fastblk = (flags & SIGPROCMASK_FASTBLK) != 0;
2703 	while ((sig = sig_ffs(&block)) != 0) {
2704 		SIGDELSET(block, sig);
2705 		td = sigtd(p, sig, fastblk);
2706 
2707 		/*
2708 		 * If sigtd() selected us despite sigfastblock is
2709 		 * blocking, do not activate AST or wake us, to avoid
2710 		 * loop in AST handler.
2711 		 */
2712 		if (fastblk && td == curthread)
2713 			continue;
2714 
2715 		signotify(td);
2716 		if (!pslocked)
2717 			mtx_lock(&ps->ps_mtx);
2718 		if (p->p_flag & P_TRACED ||
2719 		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2720 		    !SIGISMEMBER(td->td_sigmask, sig))) {
2721 			tdsigwakeup(td, sig, SIG_CATCH,
2722 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2723 			    ERESTART));
2724 		}
2725 		if (!pslocked)
2726 			mtx_unlock(&ps->ps_mtx);
2727 	}
2728 }
2729 
2730 void
2731 tdsigcleanup(struct thread *td)
2732 {
2733 	struct proc *p;
2734 	sigset_t unblocked;
2735 
2736 	p = td->td_proc;
2737 	PROC_LOCK_ASSERT(p, MA_OWNED);
2738 
2739 	sigqueue_flush(&td->td_sigqueue);
2740 	if (p->p_numthreads == 1)
2741 		return;
2742 
2743 	/*
2744 	 * Since we cannot handle signals, notify signal post code
2745 	 * about this by filling the sigmask.
2746 	 *
2747 	 * Also, if needed, wake up thread(s) that do not block the
2748 	 * same signals as the exiting thread, since the thread might
2749 	 * have been selected for delivery and woken up.
2750 	 */
2751 	SIGFILLSET(unblocked);
2752 	SIGSETNAND(unblocked, td->td_sigmask);
2753 	SIGFILLSET(td->td_sigmask);
2754 	reschedule_signals(p, unblocked, 0);
2755 
2756 }
2757 
2758 static int
2759 sigdeferstop_curr_flags(int cflags)
2760 {
2761 
2762 	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
2763 	    (cflags & TDF_SBDRY) != 0);
2764 	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
2765 }
2766 
2767 /*
2768  * Defer the delivery of SIGSTOP for the current thread, according to
2769  * the requested mode.  Returns previous flags, which must be restored
2770  * by sigallowstop().
2771  *
2772  * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
2773  * cleared by the current thread, which allow the lock-less read-only
2774  * accesses below.
2775  */
2776 int
2777 sigdeferstop_impl(int mode)
2778 {
2779 	struct thread *td;
2780 	int cflags, nflags;
2781 
2782 	td = curthread;
2783 	cflags = sigdeferstop_curr_flags(td->td_flags);
2784 	switch (mode) {
2785 	case SIGDEFERSTOP_NOP:
2786 		nflags = cflags;
2787 		break;
2788 	case SIGDEFERSTOP_OFF:
2789 		nflags = 0;
2790 		break;
2791 	case SIGDEFERSTOP_SILENT:
2792 		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
2793 		break;
2794 	case SIGDEFERSTOP_EINTR:
2795 		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
2796 		break;
2797 	case SIGDEFERSTOP_ERESTART:
2798 		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
2799 		break;
2800 	default:
2801 		panic("sigdeferstop: invalid mode %x", mode);
2802 		break;
2803 	}
2804 	if (cflags == nflags)
2805 		return (SIGDEFERSTOP_VAL_NCHG);
2806 	thread_lock(td);
2807 	td->td_flags = (td->td_flags & ~cflags) | nflags;
2808 	thread_unlock(td);
2809 	return (cflags);
2810 }
2811 
2812 /*
2813  * Restores the STOP handling mode, typically permitting the delivery
2814  * of SIGSTOP for the current thread.  This does not immediately
2815  * suspend if a stop was posted.  Instead, the thread will suspend
2816  * either via ast() or a subsequent interruptible sleep.
2817  */
2818 void
2819 sigallowstop_impl(int prev)
2820 {
2821 	struct thread *td;
2822 	int cflags;
2823 
2824 	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
2825 	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
2826 	    ("sigallowstop: incorrect previous mode %x", prev));
2827 	td = curthread;
2828 	cflags = sigdeferstop_curr_flags(td->td_flags);
2829 	if (cflags != prev) {
2830 		thread_lock(td);
2831 		td->td_flags = (td->td_flags & ~cflags) | prev;
2832 		thread_unlock(td);
2833 	}
2834 }
2835 
2836 /*
2837  * If the current process has received a signal (should be caught or cause
2838  * termination, should interrupt current syscall), return the signal number.
2839  * Stop signals with default action are processed immediately, then cleared;
2840  * they aren't returned.  This is checked after each entry to the system for
2841  * a syscall or trap (though this can usually be done without calling issignal
2842  * by checking the pending signal masks in cursig.) The normal call
2843  * sequence is
2844  *
2845  *	while (sig = cursig(curthread))
2846  *		postsig(sig);
2847  */
2848 static int
2849 issignal(struct thread *td)
2850 {
2851 	struct proc *p;
2852 	struct sigacts *ps;
2853 	struct sigqueue *queue;
2854 	sigset_t sigpending;
2855 	ksiginfo_t ksi;
2856 	int prop, sig, traced;
2857 
2858 	p = td->td_proc;
2859 	ps = p->p_sigacts;
2860 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2861 	PROC_LOCK_ASSERT(p, MA_OWNED);
2862 	for (;;) {
2863 		traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
2864 
2865 		sigpending = td->td_sigqueue.sq_signals;
2866 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
2867 		SIGSETNAND(sigpending, td->td_sigmask);
2868 
2869 		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
2870 		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2871 			SIG_STOPSIGMASK(sigpending);
2872 		if (SIGISEMPTY(sigpending))	/* no signal to send */
2873 			return (0);
2874 
2875 		/*
2876 		 * Do fast sigblock if requested by usermode.  Since
2877 		 * we do know that there was a signal pending at this
2878 		 * point, set the FAST_SIGBLOCK_PEND as indicator for
2879 		 * usermode to perform a dummy call to
2880 		 * FAST_SIGBLOCK_UNBLOCK, which causes immediate
2881 		 * delivery of postponed pending signal.
2882 		 */
2883 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
2884 			if (td->td_sigblock_val != 0)
2885 				SIGSETNAND(sigpending, fastblock_mask);
2886 			if (SIGISEMPTY(sigpending)) {
2887 				td->td_pflags |= TDP_SIGFASTPENDING;
2888 				return (0);
2889 			}
2890 		}
2891 
2892 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
2893 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
2894 		    SIGISMEMBER(sigpending, SIGSTOP)) {
2895 			/*
2896 			 * If debugger just attached, always consume
2897 			 * SIGSTOP from ptrace(PT_ATTACH) first, to
2898 			 * execute the debugger attach ritual in
2899 			 * order.
2900 			 */
2901 			sig = SIGSTOP;
2902 			td->td_dbgflags |= TDB_FSTP;
2903 		} else {
2904 			sig = sig_ffs(&sigpending);
2905 		}
2906 
2907 		if (p->p_stops & S_SIG) {
2908 			mtx_unlock(&ps->ps_mtx);
2909 			stopevent(p, S_SIG, sig);
2910 			mtx_lock(&ps->ps_mtx);
2911 		}
2912 
2913 		/*
2914 		 * We should see pending but ignored signals
2915 		 * only if P_TRACED was on when they were posted.
2916 		 */
2917 		if (SIGISMEMBER(ps->ps_sigignore, sig) && (traced == 0)) {
2918 			sigqueue_delete(&td->td_sigqueue, sig);
2919 			sigqueue_delete(&p->p_sigqueue, sig);
2920 			continue;
2921 		}
2922 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
2923 			/*
2924 			 * If traced, always stop.
2925 			 * Remove old signal from queue before the stop.
2926 			 * XXX shrug off debugger, it causes siginfo to
2927 			 * be thrown away.
2928 			 */
2929 			queue = &td->td_sigqueue;
2930 			ksiginfo_init(&ksi);
2931 			if (sigqueue_get(queue, sig, &ksi) == 0) {
2932 				queue = &p->p_sigqueue;
2933 				sigqueue_get(queue, sig, &ksi);
2934 			}
2935 			td->td_si = ksi.ksi_info;
2936 
2937 			mtx_unlock(&ps->ps_mtx);
2938 			sig = ptracestop(td, sig, &ksi);
2939 			mtx_lock(&ps->ps_mtx);
2940 
2941 			td->td_si.si_signo = 0;
2942 
2943 			/*
2944 			 * Keep looking if the debugger discarded or
2945 			 * replaced the signal.
2946 			 */
2947 			if (sig == 0)
2948 				continue;
2949 
2950 			/*
2951 			 * If the signal became masked, re-queue it.
2952 			 */
2953 			if (SIGISMEMBER(td->td_sigmask, sig)) {
2954 				ksi.ksi_flags |= KSI_HEAD;
2955 				sigqueue_add(&p->p_sigqueue, sig, &ksi);
2956 				continue;
2957 			}
2958 
2959 			/*
2960 			 * If the traced bit got turned off, requeue
2961 			 * the signal and go back up to the top to
2962 			 * rescan signals.  This ensures that p_sig*
2963 			 * and p_sigact are consistent.
2964 			 */
2965 			if ((p->p_flag & P_TRACED) == 0) {
2966 				ksi.ksi_flags |= KSI_HEAD;
2967 				sigqueue_add(queue, sig, &ksi);
2968 				continue;
2969 			}
2970 		}
2971 
2972 		prop = sigprop(sig);
2973 
2974 		/*
2975 		 * Decide whether the signal should be returned.
2976 		 * Return the signal's number, or fall through
2977 		 * to clear it from the pending mask.
2978 		 */
2979 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2980 
2981 		case (intptr_t)SIG_DFL:
2982 			/*
2983 			 * Don't take default actions on system processes.
2984 			 */
2985 			if (p->p_pid <= 1) {
2986 #ifdef DIAGNOSTIC
2987 				/*
2988 				 * Are you sure you want to ignore SIGSEGV
2989 				 * in init? XXX
2990 				 */
2991 				printf("Process (pid %lu) got signal %d\n",
2992 					(u_long)p->p_pid, sig);
2993 #endif
2994 				break;		/* == ignore */
2995 			}
2996 			/*
2997 			 * If there is a pending stop signal to process with
2998 			 * default action, stop here, then clear the signal.
2999 			 * Traced or exiting processes should ignore stops.
3000 			 * Additionally, a member of an orphaned process group
3001 			 * should ignore tty stops.
3002 			 */
3003 			if (prop & SIGPROP_STOP) {
3004 				if (p->p_flag &
3005 				    (P_TRACED | P_WEXIT | P_SINGLE_EXIT) ||
3006 				    (p->p_pgrp->pg_jobc == 0 &&
3007 				     prop & SIGPROP_TTYSTOP))
3008 					break;	/* == ignore */
3009 				if (TD_SBDRY_INTR(td)) {
3010 					KASSERT((td->td_flags & TDF_SBDRY) != 0,
3011 					    ("lost TDF_SBDRY"));
3012 					return (-1);
3013 				}
3014 				mtx_unlock(&ps->ps_mtx);
3015 				WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
3016 				    &p->p_mtx.lock_object, "Catching SIGSTOP");
3017 				sigqueue_delete(&td->td_sigqueue, sig);
3018 				sigqueue_delete(&p->p_sigqueue, sig);
3019 				p->p_flag |= P_STOPPED_SIG;
3020 				p->p_xsig = sig;
3021 				PROC_SLOCK(p);
3022 				sig_suspend_threads(td, p, 0);
3023 				thread_suspend_switch(td, p);
3024 				PROC_SUNLOCK(p);
3025 				mtx_lock(&ps->ps_mtx);
3026 				goto next;
3027 			} else if (prop & SIGPROP_IGNORE) {
3028 				/*
3029 				 * Except for SIGCONT, shouldn't get here.
3030 				 * Default action is to ignore; drop it.
3031 				 */
3032 				break;		/* == ignore */
3033 			} else
3034 				return (sig);
3035 			/*NOTREACHED*/
3036 
3037 		case (intptr_t)SIG_IGN:
3038 			/*
3039 			 * Masking above should prevent us ever trying
3040 			 * to take action on an ignored signal other
3041 			 * than SIGCONT, unless process is traced.
3042 			 */
3043 			if ((prop & SIGPROP_CONT) == 0 &&
3044 			    (p->p_flag & P_TRACED) == 0)
3045 				printf("issignal\n");
3046 			break;		/* == ignore */
3047 
3048 		default:
3049 			/*
3050 			 * This signal has an action, let
3051 			 * postsig() process it.
3052 			 */
3053 			return (sig);
3054 		}
3055 		sigqueue_delete(&td->td_sigqueue, sig);	/* take the signal! */
3056 		sigqueue_delete(&p->p_sigqueue, sig);
3057 next:;
3058 	}
3059 	/* NOTREACHED */
3060 }
3061 
3062 void
3063 thread_stopped(struct proc *p)
3064 {
3065 	int n;
3066 
3067 	PROC_LOCK_ASSERT(p, MA_OWNED);
3068 	PROC_SLOCK_ASSERT(p, MA_OWNED);
3069 	n = p->p_suspcount;
3070 	if (p == curproc)
3071 		n++;
3072 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3073 		PROC_SUNLOCK(p);
3074 		p->p_flag &= ~P_WAITED;
3075 		PROC_LOCK(p->p_pptr);
3076 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
3077 			CLD_TRAPPED : CLD_STOPPED);
3078 		PROC_UNLOCK(p->p_pptr);
3079 		PROC_SLOCK(p);
3080 	}
3081 }
3082 
3083 /*
3084  * Take the action for the specified signal
3085  * from the current set of pending signals.
3086  */
3087 int
3088 postsig(int sig)
3089 {
3090 	struct thread *td;
3091 	struct proc *p;
3092 	struct sigacts *ps;
3093 	sig_t action;
3094 	ksiginfo_t ksi;
3095 	sigset_t returnmask;
3096 
3097 	KASSERT(sig != 0, ("postsig"));
3098 
3099 	td = curthread;
3100 	p = td->td_proc;
3101 	PROC_LOCK_ASSERT(p, MA_OWNED);
3102 	ps = p->p_sigacts;
3103 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3104 	ksiginfo_init(&ksi);
3105 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3106 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3107 		return (0);
3108 	ksi.ksi_signo = sig;
3109 	if (ksi.ksi_code == SI_TIMER)
3110 		itimer_accept(p, ksi.ksi_timerid, &ksi);
3111 	action = ps->ps_sigact[_SIG_IDX(sig)];
3112 #ifdef KTRACE
3113 	if (KTRPOINT(td, KTR_PSIG))
3114 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3115 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3116 #endif
3117 	if ((p->p_stops & S_SIG) != 0) {
3118 		mtx_unlock(&ps->ps_mtx);
3119 		stopevent(p, S_SIG, sig);
3120 		mtx_lock(&ps->ps_mtx);
3121 	}
3122 
3123 	if (action == SIG_DFL) {
3124 		/*
3125 		 * Default action, where the default is to kill
3126 		 * the process.  (Other cases were ignored above.)
3127 		 */
3128 		mtx_unlock(&ps->ps_mtx);
3129 		proc_td_siginfo_capture(td, &ksi.ksi_info);
3130 		sigexit(td, sig);
3131 		/* NOTREACHED */
3132 	} else {
3133 		/*
3134 		 * If we get here, the signal must be caught.
3135 		 */
3136 		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3137 		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3138 		    ("postsig action: blocked sig %d", sig));
3139 
3140 		/*
3141 		 * Set the new mask value and also defer further
3142 		 * occurrences of this signal.
3143 		 *
3144 		 * Special case: user has done a sigsuspend.  Here the
3145 		 * current mask is not of interest, but rather the
3146 		 * mask from before the sigsuspend is what we want
3147 		 * restored after the signal processing is completed.
3148 		 */
3149 		if (td->td_pflags & TDP_OLDMASK) {
3150 			returnmask = td->td_oldsigmask;
3151 			td->td_pflags &= ~TDP_OLDMASK;
3152 		} else
3153 			returnmask = td->td_sigmask;
3154 
3155 		if (p->p_sig == sig) {
3156 			p->p_sig = 0;
3157 		}
3158 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3159 		postsig_done(sig, td, ps);
3160 	}
3161 	return (1);
3162 }
3163 
3164 void
3165 proc_wkilled(struct proc *p)
3166 {
3167 
3168 	PROC_LOCK_ASSERT(p, MA_OWNED);
3169 	if ((p->p_flag & P_WKILLED) == 0) {
3170 		p->p_flag |= P_WKILLED;
3171 		/*
3172 		 * Notify swapper that there is a process to swap in.
3173 		 * The notification is racy, at worst it would take 10
3174 		 * seconds for the swapper process to notice.
3175 		 */
3176 		if ((p->p_flag & (P_INMEM | P_SWAPPINGIN)) == 0)
3177 			wakeup(&proc0);
3178 	}
3179 }
3180 
3181 /*
3182  * Kill the current process for stated reason.
3183  */
3184 void
3185 killproc(struct proc *p, const char *why)
3186 {
3187 
3188 	PROC_LOCK_ASSERT(p, MA_OWNED);
3189 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3190 	    p->p_comm);
3191 	log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3192 	    p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3193 	    p->p_ucred->cr_uid, why);
3194 	proc_wkilled(p);
3195 	kern_psignal(p, SIGKILL);
3196 }
3197 
3198 /*
3199  * Force the current process to exit with the specified signal, dumping core
3200  * if appropriate.  We bypass the normal tests for masked and caught signals,
3201  * allowing unrecoverable failures to terminate the process without changing
3202  * signal state.  Mark the accounting record with the signal termination.
3203  * If dumping core, save the signal number for the debugger.  Calls exit and
3204  * does not return.
3205  */
3206 void
3207 sigexit(struct thread *td, int sig)
3208 {
3209 	struct proc *p = td->td_proc;
3210 
3211 	PROC_LOCK_ASSERT(p, MA_OWNED);
3212 	p->p_acflag |= AXSIG;
3213 	/*
3214 	 * We must be single-threading to generate a core dump.  This
3215 	 * ensures that the registers in the core file are up-to-date.
3216 	 * Also, the ELF dump handler assumes that the thread list doesn't
3217 	 * change out from under it.
3218 	 *
3219 	 * XXX If another thread attempts to single-thread before us
3220 	 *     (e.g. via fork()), we won't get a dump at all.
3221 	 */
3222 	if ((sigprop(sig) & SIGPROP_CORE) &&
3223 	    thread_single(p, SINGLE_NO_EXIT) == 0) {
3224 		p->p_sig = sig;
3225 		/*
3226 		 * Log signals which would cause core dumps
3227 		 * (Log as LOG_INFO to appease those who don't want
3228 		 * these messages.)
3229 		 * XXX : Todo, as well as euid, write out ruid too
3230 		 * Note that coredump() drops proc lock.
3231 		 */
3232 		if (coredump(td) == 0)
3233 			sig |= WCOREFLAG;
3234 		if (kern_logsigexit)
3235 			log(LOG_INFO,
3236 			    "pid %d (%s), jid %d, uid %d: exited on "
3237 			    "signal %d%s\n", p->p_pid, p->p_comm,
3238 			    p->p_ucred->cr_prison->pr_id,
3239 			    td->td_ucred->cr_uid,
3240 			    sig &~ WCOREFLAG,
3241 			    sig & WCOREFLAG ? " (core dumped)" : "");
3242 	} else
3243 		PROC_UNLOCK(p);
3244 	exit1(td, 0, sig);
3245 	/* NOTREACHED */
3246 }
3247 
3248 /*
3249  * Send queued SIGCHLD to parent when child process's state
3250  * is changed.
3251  */
3252 static void
3253 sigparent(struct proc *p, int reason, int status)
3254 {
3255 	PROC_LOCK_ASSERT(p, MA_OWNED);
3256 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3257 
3258 	if (p->p_ksi != NULL) {
3259 		p->p_ksi->ksi_signo  = SIGCHLD;
3260 		p->p_ksi->ksi_code   = reason;
3261 		p->p_ksi->ksi_status = status;
3262 		p->p_ksi->ksi_pid    = p->p_pid;
3263 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3264 		if (KSI_ONQ(p->p_ksi))
3265 			return;
3266 	}
3267 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3268 }
3269 
3270 static void
3271 childproc_jobstate(struct proc *p, int reason, int sig)
3272 {
3273 	struct sigacts *ps;
3274 
3275 	PROC_LOCK_ASSERT(p, MA_OWNED);
3276 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3277 
3278 	/*
3279 	 * Wake up parent sleeping in kern_wait(), also send
3280 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3281 	 * that parent will awake, because parent may masked
3282 	 * the signal.
3283 	 */
3284 	p->p_pptr->p_flag |= P_STATCHILD;
3285 	wakeup(p->p_pptr);
3286 
3287 	ps = p->p_pptr->p_sigacts;
3288 	mtx_lock(&ps->ps_mtx);
3289 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3290 		mtx_unlock(&ps->ps_mtx);
3291 		sigparent(p, reason, sig);
3292 	} else
3293 		mtx_unlock(&ps->ps_mtx);
3294 }
3295 
3296 void
3297 childproc_stopped(struct proc *p, int reason)
3298 {
3299 
3300 	childproc_jobstate(p, reason, p->p_xsig);
3301 }
3302 
3303 void
3304 childproc_continued(struct proc *p)
3305 {
3306 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3307 }
3308 
3309 void
3310 childproc_exited(struct proc *p)
3311 {
3312 	int reason, status;
3313 
3314 	if (WCOREDUMP(p->p_xsig)) {
3315 		reason = CLD_DUMPED;
3316 		status = WTERMSIG(p->p_xsig);
3317 	} else if (WIFSIGNALED(p->p_xsig)) {
3318 		reason = CLD_KILLED;
3319 		status = WTERMSIG(p->p_xsig);
3320 	} else {
3321 		reason = CLD_EXITED;
3322 		status = p->p_xexit;
3323 	}
3324 	/*
3325 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3326 	 * done in exit1().
3327 	 */
3328 	sigparent(p, reason, status);
3329 }
3330 
3331 #define	MAX_NUM_CORE_FILES 100000
3332 #ifndef NUM_CORE_FILES
3333 #define	NUM_CORE_FILES 5
3334 #endif
3335 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3336 static int num_cores = NUM_CORE_FILES;
3337 
3338 static int
3339 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3340 {
3341 	int error;
3342 	int new_val;
3343 
3344 	new_val = num_cores;
3345 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3346 	if (error != 0 || req->newptr == NULL)
3347 		return (error);
3348 	if (new_val > MAX_NUM_CORE_FILES)
3349 		new_val = MAX_NUM_CORE_FILES;
3350 	if (new_val < 0)
3351 		new_val = 0;
3352 	num_cores = new_val;
3353 	return (0);
3354 }
3355 SYSCTL_PROC(_debug, OID_AUTO, ncores,
3356     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, sizeof(int),
3357     sysctl_debug_num_cores_check, "I",
3358     "Maximum number of generated process corefiles while using index format");
3359 
3360 #define	GZIP_SUFFIX	".gz"
3361 #define	ZSTD_SUFFIX	".zst"
3362 
3363 int compress_user_cores = 0;
3364 
3365 static int
3366 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3367 {
3368 	int error, val;
3369 
3370 	val = compress_user_cores;
3371 	error = sysctl_handle_int(oidp, &val, 0, req);
3372 	if (error != 0 || req->newptr == NULL)
3373 		return (error);
3374 	if (val != 0 && !compressor_avail(val))
3375 		return (EINVAL);
3376 	compress_user_cores = val;
3377 	return (error);
3378 }
3379 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores,
3380     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, 0, sizeof(int),
3381     sysctl_compress_user_cores, "I",
3382     "Enable compression of user corefiles ("
3383     __XSTRING(COMPRESS_GZIP) " = gzip, "
3384     __XSTRING(COMPRESS_ZSTD) " = zstd)");
3385 
3386 int compress_user_cores_level = 6;
3387 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3388     &compress_user_cores_level, 0,
3389     "Corefile compression level");
3390 
3391 /*
3392  * Protect the access to corefilename[] by allproc_lock.
3393  */
3394 #define	corefilename_lock	allproc_lock
3395 
3396 static char corefilename[MAXPATHLEN] = {"%N.core"};
3397 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3398 
3399 static int
3400 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3401 {
3402 	int error;
3403 
3404 	sx_xlock(&corefilename_lock);
3405 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3406 	    req);
3407 	sx_xunlock(&corefilename_lock);
3408 
3409 	return (error);
3410 }
3411 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3412     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3413     "Process corefile name format string");
3414 
3415 static void
3416 vnode_close_locked(struct thread *td, struct vnode *vp)
3417 {
3418 
3419 	VOP_UNLOCK(vp);
3420 	vn_close(vp, FWRITE, td->td_ucred, td);
3421 }
3422 
3423 /*
3424  * If the core format has a %I in it, then we need to check
3425  * for existing corefiles before defining a name.
3426  * To do this we iterate over 0..ncores to find a
3427  * non-existing core file name to use. If all core files are
3428  * already used we choose the oldest one.
3429  */
3430 static int
3431 corefile_open_last(struct thread *td, char *name, int indexpos,
3432     int indexlen, int ncores, struct vnode **vpp)
3433 {
3434 	struct vnode *oldvp, *nextvp, *vp;
3435 	struct vattr vattr;
3436 	struct nameidata nd;
3437 	int error, i, flags, oflags, cmode;
3438 	char ch;
3439 	struct timespec lasttime;
3440 
3441 	nextvp = oldvp = NULL;
3442 	cmode = S_IRUSR | S_IWUSR;
3443 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3444 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3445 
3446 	for (i = 0; i < ncores; i++) {
3447 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3448 
3449 		ch = name[indexpos + indexlen];
3450 		(void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3451 		    i);
3452 		name[indexpos + indexlen] = ch;
3453 
3454 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3455 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3456 		    NULL);
3457 		if (error != 0)
3458 			break;
3459 
3460 		vp = nd.ni_vp;
3461 		NDFREE(&nd, NDF_ONLY_PNBUF);
3462 		if ((flags & O_CREAT) == O_CREAT) {
3463 			nextvp = vp;
3464 			break;
3465 		}
3466 
3467 		error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3468 		if (error != 0) {
3469 			vnode_close_locked(td, vp);
3470 			break;
3471 		}
3472 
3473 		if (oldvp == NULL ||
3474 		    lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3475 		    (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3476 		    lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3477 			if (oldvp != NULL)
3478 				vnode_close_locked(td, oldvp);
3479 			oldvp = vp;
3480 			lasttime = vattr.va_mtime;
3481 		} else {
3482 			vnode_close_locked(td, vp);
3483 		}
3484 	}
3485 
3486 	if (oldvp != NULL) {
3487 		if (nextvp == NULL) {
3488 			if ((td->td_proc->p_flag & P_SUGID) != 0) {
3489 				error = EFAULT;
3490 				vnode_close_locked(td, oldvp);
3491 			} else {
3492 				nextvp = oldvp;
3493 			}
3494 		} else {
3495 			vnode_close_locked(td, oldvp);
3496 		}
3497 	}
3498 	if (error != 0) {
3499 		if (nextvp != NULL)
3500 			vnode_close_locked(td, oldvp);
3501 	} else {
3502 		*vpp = nextvp;
3503 	}
3504 
3505 	return (error);
3506 }
3507 
3508 /*
3509  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3510  * Expand the name described in corefilename, using name, uid, and pid
3511  * and open/create core file.
3512  * corefilename is a printf-like string, with three format specifiers:
3513  *	%N	name of process ("name")
3514  *	%P	process id (pid)
3515  *	%U	user id (uid)
3516  * For example, "%N.core" is the default; they can be disabled completely
3517  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3518  * This is controlled by the sysctl variable kern.corefile (see above).
3519  */
3520 static int
3521 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3522     int compress, int signum, struct vnode **vpp, char **namep)
3523 {
3524 	struct sbuf sb;
3525 	struct nameidata nd;
3526 	const char *format;
3527 	char *hostname, *name;
3528 	int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
3529 
3530 	hostname = NULL;
3531 	format = corefilename;
3532 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3533 	indexlen = 0;
3534 	indexpos = -1;
3535 	ncores = num_cores;
3536 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3537 	sx_slock(&corefilename_lock);
3538 	for (i = 0; format[i] != '\0'; i++) {
3539 		switch (format[i]) {
3540 		case '%':	/* Format character */
3541 			i++;
3542 			switch (format[i]) {
3543 			case '%':
3544 				sbuf_putc(&sb, '%');
3545 				break;
3546 			case 'H':	/* hostname */
3547 				if (hostname == NULL) {
3548 					hostname = malloc(MAXHOSTNAMELEN,
3549 					    M_TEMP, M_WAITOK);
3550 				}
3551 				getcredhostname(td->td_ucred, hostname,
3552 				    MAXHOSTNAMELEN);
3553 				sbuf_printf(&sb, "%s", hostname);
3554 				break;
3555 			case 'I':	/* autoincrementing index */
3556 				if (indexpos != -1) {
3557 					sbuf_printf(&sb, "%%I");
3558 					break;
3559 				}
3560 
3561 				indexpos = sbuf_len(&sb);
3562 				sbuf_printf(&sb, "%u", ncores - 1);
3563 				indexlen = sbuf_len(&sb) - indexpos;
3564 				break;
3565 			case 'N':	/* process name */
3566 				sbuf_printf(&sb, "%s", comm);
3567 				break;
3568 			case 'P':	/* process id */
3569 				sbuf_printf(&sb, "%u", pid);
3570 				break;
3571 			case 'S':	/* signal number */
3572 				sbuf_printf(&sb, "%i", signum);
3573 				break;
3574 			case 'U':	/* user id */
3575 				sbuf_printf(&sb, "%u", uid);
3576 				break;
3577 			default:
3578 				log(LOG_ERR,
3579 				    "Unknown format character %c in "
3580 				    "corename `%s'\n", format[i], format);
3581 				break;
3582 			}
3583 			break;
3584 		default:
3585 			sbuf_putc(&sb, format[i]);
3586 			break;
3587 		}
3588 	}
3589 	sx_sunlock(&corefilename_lock);
3590 	free(hostname, M_TEMP);
3591 	if (compress == COMPRESS_GZIP)
3592 		sbuf_printf(&sb, GZIP_SUFFIX);
3593 	else if (compress == COMPRESS_ZSTD)
3594 		sbuf_printf(&sb, ZSTD_SUFFIX);
3595 	if (sbuf_error(&sb) != 0) {
3596 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3597 		    "long\n", (long)pid, comm, (u_long)uid);
3598 		sbuf_delete(&sb);
3599 		free(name, M_TEMP);
3600 		return (ENOMEM);
3601 	}
3602 	sbuf_finish(&sb);
3603 	sbuf_delete(&sb);
3604 
3605 	if (indexpos != -1) {
3606 		error = corefile_open_last(td, name, indexpos, indexlen, ncores,
3607 		    vpp);
3608 		if (error != 0) {
3609 			log(LOG_ERR,
3610 			    "pid %d (%s), uid (%u):  Path `%s' failed "
3611 			    "on initial open test, error = %d\n",
3612 			    pid, comm, uid, name, error);
3613 		}
3614 	} else {
3615 		cmode = S_IRUSR | S_IWUSR;
3616 		oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3617 		    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3618 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3619 		if ((td->td_proc->p_flag & P_SUGID) != 0)
3620 			flags |= O_EXCL;
3621 
3622 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3623 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3624 		    NULL);
3625 		if (error == 0) {
3626 			*vpp = nd.ni_vp;
3627 			NDFREE(&nd, NDF_ONLY_PNBUF);
3628 		}
3629 	}
3630 
3631 	if (error != 0) {
3632 #ifdef AUDIT
3633 		audit_proc_coredump(td, name, error);
3634 #endif
3635 		free(name, M_TEMP);
3636 		return (error);
3637 	}
3638 	*namep = name;
3639 	return (0);
3640 }
3641 
3642 /*
3643  * Dump a process' core.  The main routine does some
3644  * policy checking, and creates the name of the coredump;
3645  * then it passes on a vnode and a size limit to the process-specific
3646  * coredump routine if there is one; if there _is not_ one, it returns
3647  * ENOSYS; otherwise it returns the error from the process-specific routine.
3648  */
3649 
3650 static int
3651 coredump(struct thread *td)
3652 {
3653 	struct proc *p = td->td_proc;
3654 	struct ucred *cred = td->td_ucred;
3655 	struct vnode *vp;
3656 	struct flock lf;
3657 	struct vattr vattr;
3658 	size_t fullpathsize;
3659 	int error, error1, locked;
3660 	char *name;			/* name of corefile */
3661 	void *rl_cookie;
3662 	off_t limit;
3663 	char *fullpath, *freepath = NULL;
3664 	struct sbuf *sb;
3665 
3666 	PROC_LOCK_ASSERT(p, MA_OWNED);
3667 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3668 	_STOPEVENT(p, S_CORE, 0);
3669 
3670 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
3671 	    (p->p_flag2 & P2_NOTRACE) != 0) {
3672 		PROC_UNLOCK(p);
3673 		return (EFAULT);
3674 	}
3675 
3676 	/*
3677 	 * Note that the bulk of limit checking is done after
3678 	 * the corefile is created.  The exception is if the limit
3679 	 * for corefiles is 0, in which case we don't bother
3680 	 * creating the corefile at all.  This layout means that
3681 	 * a corefile is truncated instead of not being created,
3682 	 * if it is larger than the limit.
3683 	 */
3684 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
3685 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
3686 		PROC_UNLOCK(p);
3687 		return (EFBIG);
3688 	}
3689 	PROC_UNLOCK(p);
3690 
3691 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
3692 	    compress_user_cores, p->p_sig, &vp, &name);
3693 	if (error != 0)
3694 		return (error);
3695 
3696 	/*
3697 	 * Don't dump to non-regular files or files with links.
3698 	 * Do not dump into system files. Effective user must own the corefile.
3699 	 */
3700 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
3701 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
3702 	    vattr.va_uid != cred->cr_uid) {
3703 		VOP_UNLOCK(vp);
3704 		error = EFAULT;
3705 		goto out;
3706 	}
3707 
3708 	VOP_UNLOCK(vp);
3709 
3710 	/* Postpone other writers, including core dumps of other processes. */
3711 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
3712 
3713 	lf.l_whence = SEEK_SET;
3714 	lf.l_start = 0;
3715 	lf.l_len = 0;
3716 	lf.l_type = F_WRLCK;
3717 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
3718 
3719 	VATTR_NULL(&vattr);
3720 	vattr.va_size = 0;
3721 	if (set_core_nodump_flag)
3722 		vattr.va_flags = UF_NODUMP;
3723 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3724 	VOP_SETATTR(vp, &vattr, cred);
3725 	VOP_UNLOCK(vp);
3726 	PROC_LOCK(p);
3727 	p->p_acflag |= ACORE;
3728 	PROC_UNLOCK(p);
3729 
3730 	if (p->p_sysent->sv_coredump != NULL) {
3731 		error = p->p_sysent->sv_coredump(td, vp, limit, 0);
3732 	} else {
3733 		error = ENOSYS;
3734 	}
3735 
3736 	if (locked) {
3737 		lf.l_type = F_UNLCK;
3738 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3739 	}
3740 	vn_rangelock_unlock(vp, rl_cookie);
3741 
3742 	/*
3743 	 * Notify the userland helper that a process triggered a core dump.
3744 	 * This allows the helper to run an automated debugging session.
3745 	 */
3746 	if (error != 0 || coredump_devctl == 0)
3747 		goto out;
3748 	sb = sbuf_new_auto();
3749 	if (vn_fullpath_global(td, p->p_textvp, &fullpath, &freepath) != 0)
3750 		goto out2;
3751 	sbuf_printf(sb, "comm=\"");
3752 	devctl_safe_quote_sb(sb, fullpath);
3753 	free(freepath, M_TEMP);
3754 	sbuf_printf(sb, "\" core=\"");
3755 
3756 	/*
3757 	 * We can't lookup core file vp directly. When we're replacing a core, and
3758 	 * other random times, we flush the name cache, so it will fail. Instead,
3759 	 * if the path of the core is relative, add the current dir in front if it.
3760 	 */
3761 	if (name[0] != '/') {
3762 		fullpathsize = MAXPATHLEN;
3763 		freepath = malloc(fullpathsize, M_TEMP, M_WAITOK);
3764 		if (vn_getcwd(td, freepath, &fullpath, &fullpathsize) != 0) {
3765 			free(freepath, M_TEMP);
3766 			goto out2;
3767 		}
3768 		devctl_safe_quote_sb(sb, fullpath);
3769 		free(freepath, M_TEMP);
3770 		sbuf_putc(sb, '/');
3771 	}
3772 	devctl_safe_quote_sb(sb, name);
3773 	sbuf_printf(sb, "\"");
3774 	if (sbuf_finish(sb) == 0)
3775 		devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
3776 out2:
3777 	sbuf_delete(sb);
3778 out:
3779 	error1 = vn_close(vp, FWRITE, cred, td);
3780 	if (error == 0)
3781 		error = error1;
3782 #ifdef AUDIT
3783 	audit_proc_coredump(td, name, error);
3784 #endif
3785 	free(name, M_TEMP);
3786 	return (error);
3787 }
3788 
3789 /*
3790  * Nonexistent system call-- signal process (may want to handle it).  Flag
3791  * error in case process won't see signal immediately (blocked or ignored).
3792  */
3793 #ifndef _SYS_SYSPROTO_H_
3794 struct nosys_args {
3795 	int	dummy;
3796 };
3797 #endif
3798 /* ARGSUSED */
3799 int
3800 nosys(struct thread *td, struct nosys_args *args)
3801 {
3802 	struct proc *p;
3803 
3804 	p = td->td_proc;
3805 
3806 	PROC_LOCK(p);
3807 	tdsignal(td, SIGSYS);
3808 	PROC_UNLOCK(p);
3809 	if (kern_lognosys == 1 || kern_lognosys == 3) {
3810 		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
3811 		    td->td_sa.code);
3812 	}
3813 	if (kern_lognosys == 2 || kern_lognosys == 3) {
3814 		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
3815 		    td->td_sa.code);
3816 	}
3817 	return (ENOSYS);
3818 }
3819 
3820 /*
3821  * Send a SIGIO or SIGURG signal to a process or process group using stored
3822  * credentials rather than those of the current process.
3823  */
3824 void
3825 pgsigio(struct sigio **sigiop, int sig, int checkctty)
3826 {
3827 	ksiginfo_t ksi;
3828 	struct sigio *sigio;
3829 
3830 	ksiginfo_init(&ksi);
3831 	ksi.ksi_signo = sig;
3832 	ksi.ksi_code = SI_KERNEL;
3833 
3834 	SIGIO_LOCK();
3835 	sigio = *sigiop;
3836 	if (sigio == NULL) {
3837 		SIGIO_UNLOCK();
3838 		return;
3839 	}
3840 	if (sigio->sio_pgid > 0) {
3841 		PROC_LOCK(sigio->sio_proc);
3842 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
3843 			kern_psignal(sigio->sio_proc, sig);
3844 		PROC_UNLOCK(sigio->sio_proc);
3845 	} else if (sigio->sio_pgid < 0) {
3846 		struct proc *p;
3847 
3848 		PGRP_LOCK(sigio->sio_pgrp);
3849 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
3850 			PROC_LOCK(p);
3851 			if (p->p_state == PRS_NORMAL &&
3852 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
3853 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
3854 				kern_psignal(p, sig);
3855 			PROC_UNLOCK(p);
3856 		}
3857 		PGRP_UNLOCK(sigio->sio_pgrp);
3858 	}
3859 	SIGIO_UNLOCK();
3860 }
3861 
3862 static int
3863 filt_sigattach(struct knote *kn)
3864 {
3865 	struct proc *p = curproc;
3866 
3867 	kn->kn_ptr.p_proc = p;
3868 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
3869 
3870 	knlist_add(p->p_klist, kn, 0);
3871 
3872 	return (0);
3873 }
3874 
3875 static void
3876 filt_sigdetach(struct knote *kn)
3877 {
3878 	struct proc *p = kn->kn_ptr.p_proc;
3879 
3880 	knlist_remove(p->p_klist, kn, 0);
3881 }
3882 
3883 /*
3884  * signal knotes are shared with proc knotes, so we apply a mask to
3885  * the hint in order to differentiate them from process hints.  This
3886  * could be avoided by using a signal-specific knote list, but probably
3887  * isn't worth the trouble.
3888  */
3889 static int
3890 filt_signal(struct knote *kn, long hint)
3891 {
3892 
3893 	if (hint & NOTE_SIGNAL) {
3894 		hint &= ~NOTE_SIGNAL;
3895 
3896 		if (kn->kn_id == hint)
3897 			kn->kn_data++;
3898 	}
3899 	return (kn->kn_data != 0);
3900 }
3901 
3902 struct sigacts *
3903 sigacts_alloc(void)
3904 {
3905 	struct sigacts *ps;
3906 
3907 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
3908 	refcount_init(&ps->ps_refcnt, 1);
3909 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
3910 	return (ps);
3911 }
3912 
3913 void
3914 sigacts_free(struct sigacts *ps)
3915 {
3916 
3917 	if (refcount_release(&ps->ps_refcnt) == 0)
3918 		return;
3919 	mtx_destroy(&ps->ps_mtx);
3920 	free(ps, M_SUBPROC);
3921 }
3922 
3923 struct sigacts *
3924 sigacts_hold(struct sigacts *ps)
3925 {
3926 
3927 	refcount_acquire(&ps->ps_refcnt);
3928 	return (ps);
3929 }
3930 
3931 void
3932 sigacts_copy(struct sigacts *dest, struct sigacts *src)
3933 {
3934 
3935 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
3936 	mtx_lock(&src->ps_mtx);
3937 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
3938 	mtx_unlock(&src->ps_mtx);
3939 }
3940 
3941 int
3942 sigacts_shared(struct sigacts *ps)
3943 {
3944 
3945 	return (ps->ps_refcnt > 1);
3946 }
3947 
3948 void
3949 sig_drop_caught(struct proc *p)
3950 {
3951 	int sig;
3952 	struct sigacts *ps;
3953 
3954 	ps = p->p_sigacts;
3955 	PROC_LOCK_ASSERT(p, MA_OWNED);
3956 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3957 	while (SIGNOTEMPTY(ps->ps_sigcatch)) {
3958 		sig = sig_ffs(&ps->ps_sigcatch);
3959 		sigdflt(ps, sig);
3960 		if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
3961 			sigqueue_delete_proc(p, sig);
3962 	}
3963 }
3964 
3965 static void
3966 sigfastblock_failed(struct thread *td, bool sendsig, bool write)
3967 {
3968 	ksiginfo_t ksi;
3969 
3970 	/*
3971 	 * Prevent further fetches and SIGSEGVs, allowing thread to
3972 	 * issue syscalls despite corruption.
3973 	 */
3974 	sigfastblock_clear(td);
3975 
3976 	if (!sendsig)
3977 		return;
3978 	ksiginfo_init_trap(&ksi);
3979 	ksi.ksi_signo = SIGSEGV;
3980 	ksi.ksi_code = write ? SEGV_ACCERR : SEGV_MAPERR;
3981 	ksi.ksi_addr = td->td_sigblock_ptr;
3982 	trapsignal(td, &ksi);
3983 }
3984 
3985 static bool
3986 sigfastblock_fetch_sig(struct thread *td, bool sendsig, uint32_t *valp)
3987 {
3988 	uint32_t res;
3989 
3990 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
3991 		return (true);
3992 	if (fueword32((void *)td->td_sigblock_ptr, &res) == -1) {
3993 		sigfastblock_failed(td, sendsig, false);
3994 		return (false);
3995 	}
3996 	*valp = res;
3997 	td->td_sigblock_val = res & ~SIGFASTBLOCK_FLAGS;
3998 	return (true);
3999 }
4000 
4001 int
4002 sys_sigfastblock(struct thread *td, struct sigfastblock_args *uap)
4003 {
4004 	struct proc *p;
4005 	int error, res;
4006 	uint32_t oldval;
4007 
4008 	error = 0;
4009 	p = td->td_proc;
4010 	switch (uap->cmd) {
4011 	case SIGFASTBLOCK_SETPTR:
4012 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
4013 			error = EBUSY;
4014 			break;
4015 		}
4016 		if (((uintptr_t)(uap->ptr) & (sizeof(uint32_t) - 1)) != 0) {
4017 			error = EINVAL;
4018 			break;
4019 		}
4020 		td->td_pflags |= TDP_SIGFASTBLOCK;
4021 		td->td_sigblock_ptr = uap->ptr;
4022 		break;
4023 
4024 	case SIGFASTBLOCK_UNBLOCK:
4025 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4026 			error = EINVAL;
4027 			break;
4028 		}
4029 
4030 		for (;;) {
4031 			res = casueword32(td->td_sigblock_ptr,
4032 			    SIGFASTBLOCK_PEND, &oldval, 0);
4033 			if (res == -1) {
4034 				error = EFAULT;
4035 				sigfastblock_failed(td, false, true);
4036 				break;
4037 			}
4038 			if (res == 0)
4039 				break;
4040 			MPASS(res == 1);
4041 			if (oldval != SIGFASTBLOCK_PEND) {
4042 				error = EBUSY;
4043 				break;
4044 			}
4045 			error = thread_check_susp(td, false);
4046 			if (error != 0)
4047 				break;
4048 		}
4049 		if (error != 0)
4050 			break;
4051 
4052 		/*
4053 		 * td_sigblock_val is cleared there, but not on a
4054 		 * syscall exit.  The end effect is that a single
4055 		 * interruptible sleep, while user sigblock word is
4056 		 * set, might return EINTR or ERESTART to usermode
4057 		 * without delivering signal.  All further sleeps,
4058 		 * until userspace clears the word and does
4059 		 * sigfastblock(UNBLOCK), observe current word and no
4060 		 * longer get interrupted.  It is slight
4061 		 * non-conformance, with alternative to have read the
4062 		 * sigblock word on each syscall entry.
4063 		 */
4064 		td->td_sigblock_val = 0;
4065 
4066 		/*
4067 		 * Rely on normal ast mechanism to deliver pending
4068 		 * signals to current thread.  But notify others about
4069 		 * fake unblock.
4070 		 */
4071 		if (error == 0 && p->p_numthreads != 1) {
4072 			PROC_LOCK(p);
4073 			reschedule_signals(p, td->td_sigmask, 0);
4074 			PROC_UNLOCK(p);
4075 		}
4076 		break;
4077 
4078 	case SIGFASTBLOCK_UNSETPTR:
4079 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4080 			error = EINVAL;
4081 			break;
4082 		}
4083 		if (!sigfastblock_fetch_sig(td, false, &oldval)) {
4084 			error = EFAULT;
4085 			break;
4086 		}
4087 		if (oldval != 0 && oldval != SIGFASTBLOCK_PEND) {
4088 			error = EBUSY;
4089 			break;
4090 		}
4091 		sigfastblock_clear(td);
4092 		break;
4093 
4094 	default:
4095 		error = EINVAL;
4096 		break;
4097 	}
4098 	return (error);
4099 }
4100 
4101 void
4102 sigfastblock_clear(struct thread *td)
4103 {
4104 	struct proc *p;
4105 	bool resched;
4106 
4107 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4108 		return;
4109 	td->td_sigblock_val = 0;
4110 	resched = (td->td_pflags & TDP_SIGFASTPENDING) != 0 ||
4111 	    SIGPENDING(td);
4112 	td->td_pflags &= ~(TDP_SIGFASTBLOCK | TDP_SIGFASTPENDING);
4113 	if (resched) {
4114 		p = td->td_proc;
4115 		PROC_LOCK(p);
4116 		reschedule_signals(p, td->td_sigmask, 0);
4117 		PROC_UNLOCK(p);
4118 	}
4119 }
4120 
4121 void
4122 sigfastblock_fetch(struct thread *td)
4123 {
4124 	uint32_t val;
4125 
4126 	(void)sigfastblock_fetch_sig(td, true, &val);
4127 }
4128 
4129 static void
4130 sigfastblock_setpend1(struct thread *td)
4131 {
4132 	int res;
4133 	uint32_t oldval;
4134 
4135 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4136 		return;
4137 	res = fueword32((void *)td->td_sigblock_ptr, &oldval);
4138 	if (res == -1) {
4139 		sigfastblock_failed(td, true, false);
4140 		return;
4141 	}
4142 	for (;;) {
4143 		res = casueword32(td->td_sigblock_ptr, oldval, &oldval,
4144 		    oldval | SIGFASTBLOCK_PEND);
4145 		if (res == -1) {
4146 			sigfastblock_failed(td, true, true);
4147 			return;
4148 		}
4149 		if (res == 0) {
4150 			td->td_sigblock_val = oldval & ~SIGFASTBLOCK_FLAGS;
4151 			td->td_pflags &= ~TDP_SIGFASTPENDING;
4152 			break;
4153 		}
4154 		MPASS(res == 1);
4155 		if (thread_check_susp(td, false) != 0)
4156 			break;
4157 	}
4158 }
4159 
4160 void
4161 sigfastblock_setpend(struct thread *td, bool resched)
4162 {
4163 	struct proc *p;
4164 
4165 	sigfastblock_setpend1(td);
4166 	if (resched) {
4167 		p = td->td_proc;
4168 		PROC_LOCK(p);
4169 		reschedule_signals(p, fastblock_mask, SIGPROCMASK_FASTBLK);
4170 		PROC_UNLOCK(p);
4171 	}
4172 }
4173