xref: /freebsd/sys/kern/kern_sig.c (revision 00698711dee1d990d3db9c41bf58394e589eecfe)
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/capsicum.h>
51 #include <sys/compressor.h>
52 #include <sys/condvar.h>
53 #include <sys/devctl.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/racct.h>
71 #include <sys/resourcevar.h>
72 #include <sys/sdt.h>
73 #include <sys/sbuf.h>
74 #include <sys/sleepqueue.h>
75 #include <sys/smp.h>
76 #include <sys/stat.h>
77 #include <sys/sx.h>
78 #include <sys/syscallsubr.h>
79 #include <sys/sysctl.h>
80 #include <sys/sysent.h>
81 #include <sys/syslog.h>
82 #include <sys/sysproto.h>
83 #include <sys/timers.h>
84 #include <sys/unistd.h>
85 #include <sys/wait.h>
86 #include <vm/vm.h>
87 #include <vm/vm_extern.h>
88 #include <vm/uma.h>
89 
90 #include <sys/jail.h>
91 
92 #include <machine/cpu.h>
93 
94 #include <security/audit/audit.h>
95 
96 #define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
97 
98 SDT_PROVIDER_DECLARE(proc);
99 SDT_PROBE_DEFINE3(proc, , , signal__send,
100     "struct thread *", "struct proc *", "int");
101 SDT_PROBE_DEFINE2(proc, , , signal__clear,
102     "int", "ksiginfo_t *");
103 SDT_PROBE_DEFINE3(proc, , , signal__discard,
104     "struct thread *", "struct proc *", "int");
105 
106 static int	coredump(struct thread *);
107 static int	killpg1(struct thread *td, int sig, int pgid, int all,
108 		    ksiginfo_t *ksi);
109 static int	issignal(struct thread *td);
110 static void	reschedule_signals(struct proc *p, sigset_t block, int flags);
111 static int	sigprop(int sig);
112 static void	tdsigwakeup(struct thread *, int, sig_t, int);
113 static int	sig_suspend_threads(struct thread *, struct proc *, int);
114 static int	filt_sigattach(struct knote *kn);
115 static void	filt_sigdetach(struct knote *kn);
116 static int	filt_signal(struct knote *kn, long hint);
117 static struct thread *sigtd(struct proc *p, int sig, bool fast_sigblock);
118 static void	sigqueue_start(void);
119 
120 static uma_zone_t	ksiginfo_zone = NULL;
121 struct filterops sig_filtops = {
122 	.f_isfd = 0,
123 	.f_attach = filt_sigattach,
124 	.f_detach = filt_sigdetach,
125 	.f_event = filt_signal,
126 };
127 
128 static int	kern_logsigexit = 1;
129 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
130     &kern_logsigexit, 0,
131     "Log processes quitting on abnormal signals to syslog(3)");
132 
133 static int	kern_forcesigexit = 1;
134 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
135     &kern_forcesigexit, 0, "Force trap signal to be handled");
136 
137 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
138     "POSIX real time signal");
139 
140 static int	max_pending_per_proc = 128;
141 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
142     &max_pending_per_proc, 0, "Max pending signals per proc");
143 
144 static int	preallocate_siginfo = 1024;
145 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
146     &preallocate_siginfo, 0, "Preallocated signal memory size");
147 
148 static int	signal_overflow = 0;
149 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
150     &signal_overflow, 0, "Number of signals overflew");
151 
152 static int	signal_alloc_fail = 0;
153 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
154     &signal_alloc_fail, 0, "signals failed to be allocated");
155 
156 static int	kern_lognosys = 0;
157 SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
158     "Log invalid syscalls");
159 
160 __read_frequently bool sigfastblock_fetch_always = false;
161 SYSCTL_BOOL(_kern, OID_AUTO, sigfastblock_fetch_always, CTLFLAG_RWTUN,
162     &sigfastblock_fetch_always, 0,
163     "Fetch sigfastblock word on each syscall entry for proper "
164     "blocking semantic");
165 
166 static bool	kern_sig_discard_ign = true;
167 SYSCTL_BOOL(_kern, OID_AUTO, sig_discard_ign, CTLFLAG_RWTUN,
168     &kern_sig_discard_ign, 0,
169     "Discard ignored signals on delivery, otherwise queue them to "
170     "the target queue");
171 
172 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
173 
174 /*
175  * Policy -- Can ucred cr1 send SIGIO to process cr2?
176  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
177  * in the right situations.
178  */
179 #define CANSIGIO(cr1, cr2) \
180 	((cr1)->cr_uid == 0 || \
181 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
182 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
183 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
184 	    (cr1)->cr_uid == (cr2)->cr_uid)
185 
186 static int	sugid_coredump;
187 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
188     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
189 
190 static int	capmode_coredump;
191 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
192     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
193 
194 static int	do_coredump = 1;
195 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
196 	&do_coredump, 0, "Enable/Disable coredumps");
197 
198 static int	set_core_nodump_flag = 0;
199 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
200 	0, "Enable setting the NODUMP flag on coredump files");
201 
202 static int	coredump_devctl = 0;
203 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
204 	0, "Generate a devctl notification when processes coredump");
205 
206 /*
207  * Signal properties and actions.
208  * The array below categorizes the signals and their default actions
209  * according to the following properties:
210  */
211 #define	SIGPROP_KILL		0x01	/* terminates process by default */
212 #define	SIGPROP_CORE		0x02	/* ditto and coredumps */
213 #define	SIGPROP_STOP		0x04	/* suspend process */
214 #define	SIGPROP_TTYSTOP		0x08	/* ditto, from tty */
215 #define	SIGPROP_IGNORE		0x10	/* ignore by default */
216 #define	SIGPROP_CONT		0x20	/* continue if suspended */
217 
218 static int sigproptbl[NSIG] = {
219 	[SIGHUP] =	SIGPROP_KILL,
220 	[SIGINT] =	SIGPROP_KILL,
221 	[SIGQUIT] =	SIGPROP_KILL | SIGPROP_CORE,
222 	[SIGILL] =	SIGPROP_KILL | SIGPROP_CORE,
223 	[SIGTRAP] =	SIGPROP_KILL | SIGPROP_CORE,
224 	[SIGABRT] =	SIGPROP_KILL | SIGPROP_CORE,
225 	[SIGEMT] =	SIGPROP_KILL | SIGPROP_CORE,
226 	[SIGFPE] =	SIGPROP_KILL | SIGPROP_CORE,
227 	[SIGKILL] =	SIGPROP_KILL,
228 	[SIGBUS] =	SIGPROP_KILL | SIGPROP_CORE,
229 	[SIGSEGV] =	SIGPROP_KILL | SIGPROP_CORE,
230 	[SIGSYS] =	SIGPROP_KILL | SIGPROP_CORE,
231 	[SIGPIPE] =	SIGPROP_KILL,
232 	[SIGALRM] =	SIGPROP_KILL,
233 	[SIGTERM] =	SIGPROP_KILL,
234 	[SIGURG] =	SIGPROP_IGNORE,
235 	[SIGSTOP] =	SIGPROP_STOP,
236 	[SIGTSTP] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
237 	[SIGCONT] =	SIGPROP_IGNORE | SIGPROP_CONT,
238 	[SIGCHLD] =	SIGPROP_IGNORE,
239 	[SIGTTIN] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
240 	[SIGTTOU] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
241 	[SIGIO] =	SIGPROP_IGNORE,
242 	[SIGXCPU] =	SIGPROP_KILL,
243 	[SIGXFSZ] =	SIGPROP_KILL,
244 	[SIGVTALRM] =	SIGPROP_KILL,
245 	[SIGPROF] =	SIGPROP_KILL,
246 	[SIGWINCH] =	SIGPROP_IGNORE,
247 	[SIGINFO] =	SIGPROP_IGNORE,
248 	[SIGUSR1] =	SIGPROP_KILL,
249 	[SIGUSR2] =	SIGPROP_KILL,
250 };
251 
252 #define	_SIG_FOREACH_ADVANCE(i, set) ({					\
253 	int __found;							\
254 	for (;;) {							\
255 		if (__bits != 0) {					\
256 			int __sig = ffs(__bits);			\
257 			__bits &= ~(1u << (__sig - 1));			\
258 			sig = __i * sizeof((set)->__bits[0]) * NBBY + __sig; \
259 			__found = 1;					\
260 			break;						\
261 		}							\
262 		if (++__i == _SIG_WORDS) {				\
263 			__found = 0;					\
264 			break;						\
265 		}							\
266 		__bits = (set)->__bits[__i];				\
267 	}								\
268 	__found != 0;							\
269 })
270 
271 #define	SIG_FOREACH(i, set)						\
272 	for (int32_t __i = -1, __bits = 0;				\
273 	    _SIG_FOREACH_ADVANCE(i, set); )				\
274 
275 sigset_t fastblock_mask;
276 
277 static void
278 sigqueue_start(void)
279 {
280 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
281 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
282 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
283 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
284 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
285 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
286 	SIGFILLSET(fastblock_mask);
287 	SIG_CANTMASK(fastblock_mask);
288 }
289 
290 ksiginfo_t *
291 ksiginfo_alloc(int wait)
292 {
293 	int flags;
294 
295 	flags = M_ZERO;
296 	if (! wait)
297 		flags |= M_NOWAIT;
298 	if (ksiginfo_zone != NULL)
299 		return ((ksiginfo_t *)uma_zalloc(ksiginfo_zone, flags));
300 	return (NULL);
301 }
302 
303 void
304 ksiginfo_free(ksiginfo_t *ksi)
305 {
306 	uma_zfree(ksiginfo_zone, ksi);
307 }
308 
309 static __inline int
310 ksiginfo_tryfree(ksiginfo_t *ksi)
311 {
312 	if (!(ksi->ksi_flags & KSI_EXT)) {
313 		uma_zfree(ksiginfo_zone, ksi);
314 		return (1);
315 	}
316 	return (0);
317 }
318 
319 void
320 sigqueue_init(sigqueue_t *list, struct proc *p)
321 {
322 	SIGEMPTYSET(list->sq_signals);
323 	SIGEMPTYSET(list->sq_kill);
324 	SIGEMPTYSET(list->sq_ptrace);
325 	TAILQ_INIT(&list->sq_list);
326 	list->sq_proc = p;
327 	list->sq_flags = SQ_INIT;
328 }
329 
330 /*
331  * Get a signal's ksiginfo.
332  * Return:
333  *	0	-	signal not found
334  *	others	-	signal number
335  */
336 static int
337 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
338 {
339 	struct proc *p = sq->sq_proc;
340 	struct ksiginfo *ksi, *next;
341 	int count = 0;
342 
343 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
344 
345 	if (!SIGISMEMBER(sq->sq_signals, signo))
346 		return (0);
347 
348 	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
349 		count++;
350 		SIGDELSET(sq->sq_ptrace, signo);
351 		si->ksi_flags |= KSI_PTRACE;
352 	}
353 	if (SIGISMEMBER(sq->sq_kill, signo)) {
354 		count++;
355 		if (count == 1)
356 			SIGDELSET(sq->sq_kill, signo);
357 	}
358 
359 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
360 		if (ksi->ksi_signo == signo) {
361 			if (count == 0) {
362 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
363 				ksi->ksi_sigq = NULL;
364 				ksiginfo_copy(ksi, si);
365 				if (ksiginfo_tryfree(ksi) && p != NULL)
366 					p->p_pendingcnt--;
367 			}
368 			if (++count > 1)
369 				break;
370 		}
371 	}
372 
373 	if (count <= 1)
374 		SIGDELSET(sq->sq_signals, signo);
375 	si->ksi_signo = signo;
376 	return (signo);
377 }
378 
379 void
380 sigqueue_take(ksiginfo_t *ksi)
381 {
382 	struct ksiginfo *kp;
383 	struct proc	*p;
384 	sigqueue_t	*sq;
385 
386 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
387 		return;
388 
389 	p = sq->sq_proc;
390 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
391 	ksi->ksi_sigq = NULL;
392 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
393 		p->p_pendingcnt--;
394 
395 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
396 	     kp = TAILQ_NEXT(kp, ksi_link)) {
397 		if (kp->ksi_signo == ksi->ksi_signo)
398 			break;
399 	}
400 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
401 	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
402 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
403 }
404 
405 static int
406 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
407 {
408 	struct proc *p = sq->sq_proc;
409 	struct ksiginfo *ksi;
410 	int ret = 0;
411 
412 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
413 
414 	/*
415 	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
416 	 * for these signals.
417 	 */
418 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
419 		SIGADDSET(sq->sq_kill, signo);
420 		goto out_set_bit;
421 	}
422 
423 	/* directly insert the ksi, don't copy it */
424 	if (si->ksi_flags & KSI_INS) {
425 		if (si->ksi_flags & KSI_HEAD)
426 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
427 		else
428 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
429 		si->ksi_sigq = sq;
430 		goto out_set_bit;
431 	}
432 
433 	if (__predict_false(ksiginfo_zone == NULL)) {
434 		SIGADDSET(sq->sq_kill, signo);
435 		goto out_set_bit;
436 	}
437 
438 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
439 		signal_overflow++;
440 		ret = EAGAIN;
441 	} else if ((ksi = ksiginfo_alloc(0)) == NULL) {
442 		signal_alloc_fail++;
443 		ret = EAGAIN;
444 	} else {
445 		if (p != NULL)
446 			p->p_pendingcnt++;
447 		ksiginfo_copy(si, ksi);
448 		ksi->ksi_signo = signo;
449 		if (si->ksi_flags & KSI_HEAD)
450 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
451 		else
452 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
453 		ksi->ksi_sigq = sq;
454 	}
455 
456 	if (ret != 0) {
457 		if ((si->ksi_flags & KSI_PTRACE) != 0) {
458 			SIGADDSET(sq->sq_ptrace, signo);
459 			ret = 0;
460 			goto out_set_bit;
461 		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
462 		    (si->ksi_flags & KSI_SIGQ) == 0) {
463 			SIGADDSET(sq->sq_kill, signo);
464 			ret = 0;
465 			goto out_set_bit;
466 		}
467 		return (ret);
468 	}
469 
470 out_set_bit:
471 	SIGADDSET(sq->sq_signals, signo);
472 	return (ret);
473 }
474 
475 void
476 sigqueue_flush(sigqueue_t *sq)
477 {
478 	struct proc *p = sq->sq_proc;
479 	ksiginfo_t *ksi;
480 
481 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
482 
483 	if (p != NULL)
484 		PROC_LOCK_ASSERT(p, MA_OWNED);
485 
486 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
487 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
488 		ksi->ksi_sigq = NULL;
489 		if (ksiginfo_tryfree(ksi) && p != NULL)
490 			p->p_pendingcnt--;
491 	}
492 
493 	SIGEMPTYSET(sq->sq_signals);
494 	SIGEMPTYSET(sq->sq_kill);
495 	SIGEMPTYSET(sq->sq_ptrace);
496 }
497 
498 static void
499 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
500 {
501 	sigset_t tmp;
502 	struct proc *p1, *p2;
503 	ksiginfo_t *ksi, *next;
504 
505 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
506 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
507 	p1 = src->sq_proc;
508 	p2 = dst->sq_proc;
509 	/* Move siginfo to target list */
510 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
511 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
512 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
513 			if (p1 != NULL)
514 				p1->p_pendingcnt--;
515 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
516 			ksi->ksi_sigq = dst;
517 			if (p2 != NULL)
518 				p2->p_pendingcnt++;
519 		}
520 	}
521 
522 	/* Move pending bits to target list */
523 	tmp = src->sq_kill;
524 	SIGSETAND(tmp, *set);
525 	SIGSETOR(dst->sq_kill, tmp);
526 	SIGSETNAND(src->sq_kill, tmp);
527 
528 	tmp = src->sq_ptrace;
529 	SIGSETAND(tmp, *set);
530 	SIGSETOR(dst->sq_ptrace, tmp);
531 	SIGSETNAND(src->sq_ptrace, tmp);
532 
533 	tmp = src->sq_signals;
534 	SIGSETAND(tmp, *set);
535 	SIGSETOR(dst->sq_signals, tmp);
536 	SIGSETNAND(src->sq_signals, tmp);
537 }
538 
539 #if 0
540 static void
541 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
542 {
543 	sigset_t set;
544 
545 	SIGEMPTYSET(set);
546 	SIGADDSET(set, signo);
547 	sigqueue_move_set(src, dst, &set);
548 }
549 #endif
550 
551 static void
552 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
553 {
554 	struct proc *p = sq->sq_proc;
555 	ksiginfo_t *ksi, *next;
556 
557 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
558 
559 	/* Remove siginfo queue */
560 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
561 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
562 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
563 			ksi->ksi_sigq = NULL;
564 			if (ksiginfo_tryfree(ksi) && p != NULL)
565 				p->p_pendingcnt--;
566 		}
567 	}
568 	SIGSETNAND(sq->sq_kill, *set);
569 	SIGSETNAND(sq->sq_ptrace, *set);
570 	SIGSETNAND(sq->sq_signals, *set);
571 }
572 
573 void
574 sigqueue_delete(sigqueue_t *sq, int signo)
575 {
576 	sigset_t set;
577 
578 	SIGEMPTYSET(set);
579 	SIGADDSET(set, signo);
580 	sigqueue_delete_set(sq, &set);
581 }
582 
583 /* Remove a set of signals for a process */
584 static void
585 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
586 {
587 	sigqueue_t worklist;
588 	struct thread *td0;
589 
590 	PROC_LOCK_ASSERT(p, MA_OWNED);
591 
592 	sigqueue_init(&worklist, NULL);
593 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
594 
595 	FOREACH_THREAD_IN_PROC(p, td0)
596 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
597 
598 	sigqueue_flush(&worklist);
599 }
600 
601 void
602 sigqueue_delete_proc(struct proc *p, int signo)
603 {
604 	sigset_t set;
605 
606 	SIGEMPTYSET(set);
607 	SIGADDSET(set, signo);
608 	sigqueue_delete_set_proc(p, &set);
609 }
610 
611 static void
612 sigqueue_delete_stopmask_proc(struct proc *p)
613 {
614 	sigset_t set;
615 
616 	SIGEMPTYSET(set);
617 	SIGADDSET(set, SIGSTOP);
618 	SIGADDSET(set, SIGTSTP);
619 	SIGADDSET(set, SIGTTIN);
620 	SIGADDSET(set, SIGTTOU);
621 	sigqueue_delete_set_proc(p, &set);
622 }
623 
624 /*
625  * Determine signal that should be delivered to thread td, the current
626  * thread, 0 if none.  If there is a pending stop signal with default
627  * action, the process stops in issignal().
628  */
629 int
630 cursig(struct thread *td)
631 {
632 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
633 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
634 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
635 	return (SIGPENDING(td) ? issignal(td) : 0);
636 }
637 
638 /*
639  * Arrange for ast() to handle unmasked pending signals on return to user
640  * mode.  This must be called whenever a signal is added to td_sigqueue or
641  * unmasked in td_sigmask.
642  */
643 void
644 signotify(struct thread *td)
645 {
646 
647 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
648 
649 	if (SIGPENDING(td)) {
650 		thread_lock(td);
651 		td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING;
652 		thread_unlock(td);
653 	}
654 }
655 
656 /*
657  * Returns 1 (true) if altstack is configured for the thread, and the
658  * passed stack bottom address falls into the altstack range.  Handles
659  * the 43 compat special case where the alt stack size is zero.
660  */
661 int
662 sigonstack(size_t sp)
663 {
664 	struct thread *td;
665 
666 	td = curthread;
667 	if ((td->td_pflags & TDP_ALTSTACK) == 0)
668 		return (0);
669 #if defined(COMPAT_43)
670 	if (SV_PROC_FLAG(td->td_proc, SV_AOUT) && td->td_sigstk.ss_size == 0)
671 		return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
672 #endif
673 	return (sp >= (size_t)td->td_sigstk.ss_sp &&
674 	    sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
675 }
676 
677 static __inline int
678 sigprop(int sig)
679 {
680 
681 	if (sig > 0 && sig < nitems(sigproptbl))
682 		return (sigproptbl[sig]);
683 	return (0);
684 }
685 
686 static bool
687 sigact_flag_test(const struct sigaction *act, int flag)
688 {
689 
690 	/*
691 	 * SA_SIGINFO is reset when signal disposition is set to
692 	 * ignore or default.  Other flags are kept according to user
693 	 * settings.
694 	 */
695 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
696 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
697 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
698 }
699 
700 /*
701  * kern_sigaction
702  * sigaction
703  * freebsd4_sigaction
704  * osigaction
705  */
706 int
707 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
708     struct sigaction *oact, int flags)
709 {
710 	struct sigacts *ps;
711 	struct proc *p = td->td_proc;
712 
713 	if (!_SIG_VALID(sig))
714 		return (EINVAL);
715 	if (act != NULL && act->sa_handler != SIG_DFL &&
716 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
717 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
718 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
719 		return (EINVAL);
720 
721 	PROC_LOCK(p);
722 	ps = p->p_sigacts;
723 	mtx_lock(&ps->ps_mtx);
724 	if (oact) {
725 		memset(oact, 0, sizeof(*oact));
726 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
727 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
728 			oact->sa_flags |= SA_ONSTACK;
729 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
730 			oact->sa_flags |= SA_RESTART;
731 		if (SIGISMEMBER(ps->ps_sigreset, sig))
732 			oact->sa_flags |= SA_RESETHAND;
733 		if (SIGISMEMBER(ps->ps_signodefer, sig))
734 			oact->sa_flags |= SA_NODEFER;
735 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
736 			oact->sa_flags |= SA_SIGINFO;
737 			oact->sa_sigaction =
738 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
739 		} else
740 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
741 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
742 			oact->sa_flags |= SA_NOCLDSTOP;
743 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
744 			oact->sa_flags |= SA_NOCLDWAIT;
745 	}
746 	if (act) {
747 		if ((sig == SIGKILL || sig == SIGSTOP) &&
748 		    act->sa_handler != SIG_DFL) {
749 			mtx_unlock(&ps->ps_mtx);
750 			PROC_UNLOCK(p);
751 			return (EINVAL);
752 		}
753 
754 		/*
755 		 * Change setting atomically.
756 		 */
757 
758 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
759 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
760 		if (sigact_flag_test(act, SA_SIGINFO)) {
761 			ps->ps_sigact[_SIG_IDX(sig)] =
762 			    (__sighandler_t *)act->sa_sigaction;
763 			SIGADDSET(ps->ps_siginfo, sig);
764 		} else {
765 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
766 			SIGDELSET(ps->ps_siginfo, sig);
767 		}
768 		if (!sigact_flag_test(act, SA_RESTART))
769 			SIGADDSET(ps->ps_sigintr, sig);
770 		else
771 			SIGDELSET(ps->ps_sigintr, sig);
772 		if (sigact_flag_test(act, SA_ONSTACK))
773 			SIGADDSET(ps->ps_sigonstack, sig);
774 		else
775 			SIGDELSET(ps->ps_sigonstack, sig);
776 		if (sigact_flag_test(act, SA_RESETHAND))
777 			SIGADDSET(ps->ps_sigreset, sig);
778 		else
779 			SIGDELSET(ps->ps_sigreset, sig);
780 		if (sigact_flag_test(act, SA_NODEFER))
781 			SIGADDSET(ps->ps_signodefer, sig);
782 		else
783 			SIGDELSET(ps->ps_signodefer, sig);
784 		if (sig == SIGCHLD) {
785 			if (act->sa_flags & SA_NOCLDSTOP)
786 				ps->ps_flag |= PS_NOCLDSTOP;
787 			else
788 				ps->ps_flag &= ~PS_NOCLDSTOP;
789 			if (act->sa_flags & SA_NOCLDWAIT) {
790 				/*
791 				 * Paranoia: since SA_NOCLDWAIT is implemented
792 				 * by reparenting the dying child to PID 1 (and
793 				 * trust it to reap the zombie), PID 1 itself
794 				 * is forbidden to set SA_NOCLDWAIT.
795 				 */
796 				if (p->p_pid == 1)
797 					ps->ps_flag &= ~PS_NOCLDWAIT;
798 				else
799 					ps->ps_flag |= PS_NOCLDWAIT;
800 			} else
801 				ps->ps_flag &= ~PS_NOCLDWAIT;
802 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
803 				ps->ps_flag |= PS_CLDSIGIGN;
804 			else
805 				ps->ps_flag &= ~PS_CLDSIGIGN;
806 		}
807 		/*
808 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
809 		 * and for signals set to SIG_DFL where the default is to
810 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
811 		 * have to restart the process.
812 		 */
813 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
814 		    (sigprop(sig) & SIGPROP_IGNORE &&
815 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
816 			/* never to be seen again */
817 			sigqueue_delete_proc(p, sig);
818 			if (sig != SIGCONT)
819 				/* easier in psignal */
820 				SIGADDSET(ps->ps_sigignore, sig);
821 			SIGDELSET(ps->ps_sigcatch, sig);
822 		} else {
823 			SIGDELSET(ps->ps_sigignore, sig);
824 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
825 				SIGDELSET(ps->ps_sigcatch, sig);
826 			else
827 				SIGADDSET(ps->ps_sigcatch, sig);
828 		}
829 #ifdef COMPAT_FREEBSD4
830 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
831 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
832 		    (flags & KSA_FREEBSD4) == 0)
833 			SIGDELSET(ps->ps_freebsd4, sig);
834 		else
835 			SIGADDSET(ps->ps_freebsd4, sig);
836 #endif
837 #ifdef COMPAT_43
838 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
839 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
840 		    (flags & KSA_OSIGSET) == 0)
841 			SIGDELSET(ps->ps_osigset, sig);
842 		else
843 			SIGADDSET(ps->ps_osigset, sig);
844 #endif
845 	}
846 	mtx_unlock(&ps->ps_mtx);
847 	PROC_UNLOCK(p);
848 	return (0);
849 }
850 
851 #ifndef _SYS_SYSPROTO_H_
852 struct sigaction_args {
853 	int	sig;
854 	struct	sigaction *act;
855 	struct	sigaction *oact;
856 };
857 #endif
858 int
859 sys_sigaction(struct thread *td, struct sigaction_args *uap)
860 {
861 	struct sigaction act, oact;
862 	struct sigaction *actp, *oactp;
863 	int error;
864 
865 	actp = (uap->act != NULL) ? &act : NULL;
866 	oactp = (uap->oact != NULL) ? &oact : NULL;
867 	if (actp) {
868 		error = copyin(uap->act, actp, sizeof(act));
869 		if (error)
870 			return (error);
871 	}
872 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
873 	if (oactp && !error)
874 		error = copyout(oactp, uap->oact, sizeof(oact));
875 	return (error);
876 }
877 
878 #ifdef COMPAT_FREEBSD4
879 #ifndef _SYS_SYSPROTO_H_
880 struct freebsd4_sigaction_args {
881 	int	sig;
882 	struct	sigaction *act;
883 	struct	sigaction *oact;
884 };
885 #endif
886 int
887 freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
888 {
889 	struct sigaction act, oact;
890 	struct sigaction *actp, *oactp;
891 	int error;
892 
893 	actp = (uap->act != NULL) ? &act : NULL;
894 	oactp = (uap->oact != NULL) ? &oact : NULL;
895 	if (actp) {
896 		error = copyin(uap->act, actp, sizeof(act));
897 		if (error)
898 			return (error);
899 	}
900 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
901 	if (oactp && !error)
902 		error = copyout(oactp, uap->oact, sizeof(oact));
903 	return (error);
904 }
905 #endif	/* COMAPT_FREEBSD4 */
906 
907 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
908 #ifndef _SYS_SYSPROTO_H_
909 struct osigaction_args {
910 	int	signum;
911 	struct	osigaction *nsa;
912 	struct	osigaction *osa;
913 };
914 #endif
915 int
916 osigaction(struct thread *td, struct osigaction_args *uap)
917 {
918 	struct osigaction sa;
919 	struct sigaction nsa, osa;
920 	struct sigaction *nsap, *osap;
921 	int error;
922 
923 	if (uap->signum <= 0 || uap->signum >= ONSIG)
924 		return (EINVAL);
925 
926 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
927 	osap = (uap->osa != NULL) ? &osa : NULL;
928 
929 	if (nsap) {
930 		error = copyin(uap->nsa, &sa, sizeof(sa));
931 		if (error)
932 			return (error);
933 		nsap->sa_handler = sa.sa_handler;
934 		nsap->sa_flags = sa.sa_flags;
935 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
936 	}
937 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
938 	if (osap && !error) {
939 		sa.sa_handler = osap->sa_handler;
940 		sa.sa_flags = osap->sa_flags;
941 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
942 		error = copyout(&sa, uap->osa, sizeof(sa));
943 	}
944 	return (error);
945 }
946 
947 #if !defined(__i386__)
948 /* Avoid replicating the same stub everywhere */
949 int
950 osigreturn(struct thread *td, struct osigreturn_args *uap)
951 {
952 
953 	return (nosys(td, (struct nosys_args *)uap));
954 }
955 #endif
956 #endif /* COMPAT_43 */
957 
958 /*
959  * Initialize signal state for process 0;
960  * set to ignore signals that are ignored by default.
961  */
962 void
963 siginit(struct proc *p)
964 {
965 	int i;
966 	struct sigacts *ps;
967 
968 	PROC_LOCK(p);
969 	ps = p->p_sigacts;
970 	mtx_lock(&ps->ps_mtx);
971 	for (i = 1; i <= NSIG; i++) {
972 		if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) {
973 			SIGADDSET(ps->ps_sigignore, i);
974 		}
975 	}
976 	mtx_unlock(&ps->ps_mtx);
977 	PROC_UNLOCK(p);
978 }
979 
980 /*
981  * Reset specified signal to the default disposition.
982  */
983 static void
984 sigdflt(struct sigacts *ps, int sig)
985 {
986 
987 	mtx_assert(&ps->ps_mtx, MA_OWNED);
988 	SIGDELSET(ps->ps_sigcatch, sig);
989 	if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT)
990 		SIGADDSET(ps->ps_sigignore, sig);
991 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
992 	SIGDELSET(ps->ps_siginfo, sig);
993 }
994 
995 /*
996  * Reset signals for an exec of the specified process.
997  */
998 void
999 execsigs(struct proc *p)
1000 {
1001 	struct sigacts *ps;
1002 	struct thread *td;
1003 
1004 	/*
1005 	 * Reset caught signals.  Held signals remain held
1006 	 * through td_sigmask (unless they were caught,
1007 	 * and are now ignored by default).
1008 	 */
1009 	PROC_LOCK_ASSERT(p, MA_OWNED);
1010 	ps = p->p_sigacts;
1011 	mtx_lock(&ps->ps_mtx);
1012 	sig_drop_caught(p);
1013 
1014 	/*
1015 	 * Reset stack state to the user stack.
1016 	 * Clear set of signals caught on the signal stack.
1017 	 */
1018 	td = curthread;
1019 	MPASS(td->td_proc == p);
1020 	td->td_sigstk.ss_flags = SS_DISABLE;
1021 	td->td_sigstk.ss_size = 0;
1022 	td->td_sigstk.ss_sp = 0;
1023 	td->td_pflags &= ~TDP_ALTSTACK;
1024 	/*
1025 	 * Reset no zombies if child dies flag as Solaris does.
1026 	 */
1027 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1028 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1029 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1030 	mtx_unlock(&ps->ps_mtx);
1031 }
1032 
1033 /*
1034  * kern_sigprocmask()
1035  *
1036  *	Manipulate signal mask.
1037  */
1038 int
1039 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1040     int flags)
1041 {
1042 	sigset_t new_block, oset1;
1043 	struct proc *p;
1044 	int error;
1045 
1046 	p = td->td_proc;
1047 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1048 		PROC_LOCK_ASSERT(p, MA_OWNED);
1049 	else
1050 		PROC_LOCK(p);
1051 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1052 	    ? MA_OWNED : MA_NOTOWNED);
1053 	if (oset != NULL)
1054 		*oset = td->td_sigmask;
1055 
1056 	error = 0;
1057 	if (set != NULL) {
1058 		switch (how) {
1059 		case SIG_BLOCK:
1060 			SIG_CANTMASK(*set);
1061 			oset1 = td->td_sigmask;
1062 			SIGSETOR(td->td_sigmask, *set);
1063 			new_block = td->td_sigmask;
1064 			SIGSETNAND(new_block, oset1);
1065 			break;
1066 		case SIG_UNBLOCK:
1067 			SIGSETNAND(td->td_sigmask, *set);
1068 			signotify(td);
1069 			goto out;
1070 		case SIG_SETMASK:
1071 			SIG_CANTMASK(*set);
1072 			oset1 = td->td_sigmask;
1073 			if (flags & SIGPROCMASK_OLD)
1074 				SIGSETLO(td->td_sigmask, *set);
1075 			else
1076 				td->td_sigmask = *set;
1077 			new_block = td->td_sigmask;
1078 			SIGSETNAND(new_block, oset1);
1079 			signotify(td);
1080 			break;
1081 		default:
1082 			error = EINVAL;
1083 			goto out;
1084 		}
1085 
1086 		/*
1087 		 * The new_block set contains signals that were not previously
1088 		 * blocked, but are blocked now.
1089 		 *
1090 		 * In case we block any signal that was not previously blocked
1091 		 * for td, and process has the signal pending, try to schedule
1092 		 * signal delivery to some thread that does not block the
1093 		 * signal, possibly waking it up.
1094 		 */
1095 		if (p->p_numthreads != 1)
1096 			reschedule_signals(p, new_block, flags);
1097 	}
1098 
1099 out:
1100 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1101 		PROC_UNLOCK(p);
1102 	return (error);
1103 }
1104 
1105 #ifndef _SYS_SYSPROTO_H_
1106 struct sigprocmask_args {
1107 	int	how;
1108 	const sigset_t *set;
1109 	sigset_t *oset;
1110 };
1111 #endif
1112 int
1113 sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1114 {
1115 	sigset_t set, oset;
1116 	sigset_t *setp, *osetp;
1117 	int error;
1118 
1119 	setp = (uap->set != NULL) ? &set : NULL;
1120 	osetp = (uap->oset != NULL) ? &oset : NULL;
1121 	if (setp) {
1122 		error = copyin(uap->set, setp, sizeof(set));
1123 		if (error)
1124 			return (error);
1125 	}
1126 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1127 	if (osetp && !error) {
1128 		error = copyout(osetp, uap->oset, sizeof(oset));
1129 	}
1130 	return (error);
1131 }
1132 
1133 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1134 #ifndef _SYS_SYSPROTO_H_
1135 struct osigprocmask_args {
1136 	int	how;
1137 	osigset_t mask;
1138 };
1139 #endif
1140 int
1141 osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1142 {
1143 	sigset_t set, oset;
1144 	int error;
1145 
1146 	OSIG2SIG(uap->mask, set);
1147 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1148 	SIG2OSIG(oset, td->td_retval[0]);
1149 	return (error);
1150 }
1151 #endif /* COMPAT_43 */
1152 
1153 int
1154 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1155 {
1156 	ksiginfo_t ksi;
1157 	sigset_t set;
1158 	int error;
1159 
1160 	error = copyin(uap->set, &set, sizeof(set));
1161 	if (error) {
1162 		td->td_retval[0] = error;
1163 		return (0);
1164 	}
1165 
1166 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1167 	if (error) {
1168 		/*
1169 		 * sigwait() function shall not return EINTR, but
1170 		 * the syscall does.  Non-ancient libc provides the
1171 		 * wrapper which hides EINTR.  Otherwise, EINTR return
1172 		 * is used by libthr to handle required cancellation
1173 		 * point in the sigwait().
1174 		 */
1175 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1176 			return (ERESTART);
1177 		td->td_retval[0] = error;
1178 		return (0);
1179 	}
1180 
1181 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1182 	td->td_retval[0] = error;
1183 	return (0);
1184 }
1185 
1186 int
1187 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1188 {
1189 	struct timespec ts;
1190 	struct timespec *timeout;
1191 	sigset_t set;
1192 	ksiginfo_t ksi;
1193 	int error;
1194 
1195 	if (uap->timeout) {
1196 		error = copyin(uap->timeout, &ts, sizeof(ts));
1197 		if (error)
1198 			return (error);
1199 
1200 		timeout = &ts;
1201 	} else
1202 		timeout = NULL;
1203 
1204 	error = copyin(uap->set, &set, sizeof(set));
1205 	if (error)
1206 		return (error);
1207 
1208 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1209 	if (error)
1210 		return (error);
1211 
1212 	if (uap->info)
1213 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1214 
1215 	if (error == 0)
1216 		td->td_retval[0] = ksi.ksi_signo;
1217 	return (error);
1218 }
1219 
1220 int
1221 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1222 {
1223 	ksiginfo_t ksi;
1224 	sigset_t set;
1225 	int error;
1226 
1227 	error = copyin(uap->set, &set, sizeof(set));
1228 	if (error)
1229 		return (error);
1230 
1231 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1232 	if (error)
1233 		return (error);
1234 
1235 	if (uap->info)
1236 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1237 
1238 	if (error == 0)
1239 		td->td_retval[0] = ksi.ksi_signo;
1240 	return (error);
1241 }
1242 
1243 static void
1244 proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1245 {
1246 	struct thread *thr;
1247 
1248 	FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1249 		if (thr == td)
1250 			thr->td_si = *si;
1251 		else
1252 			thr->td_si.si_signo = 0;
1253 	}
1254 }
1255 
1256 int
1257 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1258 	struct timespec *timeout)
1259 {
1260 	struct sigacts *ps;
1261 	sigset_t saved_mask, new_block;
1262 	struct proc *p;
1263 	int error, sig, timo, timevalid = 0;
1264 	struct timespec rts, ets, ts;
1265 	struct timeval tv;
1266 	bool traced;
1267 
1268 	p = td->td_proc;
1269 	error = 0;
1270 	ets.tv_sec = 0;
1271 	ets.tv_nsec = 0;
1272 	traced = false;
1273 
1274 	/* Ensure the sigfastblock value is up to date. */
1275 	sigfastblock_fetch(td);
1276 
1277 	if (timeout != NULL) {
1278 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1279 			timevalid = 1;
1280 			getnanouptime(&rts);
1281 			timespecadd(&rts, timeout, &ets);
1282 		}
1283 	}
1284 	ksiginfo_init(ksi);
1285 	/* Some signals can not be waited for. */
1286 	SIG_CANTMASK(waitset);
1287 	ps = p->p_sigacts;
1288 	PROC_LOCK(p);
1289 	saved_mask = td->td_sigmask;
1290 	SIGSETNAND(td->td_sigmask, waitset);
1291 	if ((p->p_sysent->sv_flags & SV_SIG_DISCIGN) != 0 ||
1292 	    !kern_sig_discard_ign) {
1293 		thread_lock(td);
1294 		td->td_flags |= TDF_SIGWAIT;
1295 		thread_unlock(td);
1296 	}
1297 	for (;;) {
1298 		mtx_lock(&ps->ps_mtx);
1299 		sig = cursig(td);
1300 		mtx_unlock(&ps->ps_mtx);
1301 		KASSERT(sig >= 0, ("sig %d", sig));
1302 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1303 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1304 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1305 				error = 0;
1306 				break;
1307 			}
1308 		}
1309 
1310 		if (error != 0)
1311 			break;
1312 
1313 		/*
1314 		 * POSIX says this must be checked after looking for pending
1315 		 * signals.
1316 		 */
1317 		if (timeout != NULL) {
1318 			if (!timevalid) {
1319 				error = EINVAL;
1320 				break;
1321 			}
1322 			getnanouptime(&rts);
1323 			if (timespeccmp(&rts, &ets, >=)) {
1324 				error = EAGAIN;
1325 				break;
1326 			}
1327 			timespecsub(&ets, &rts, &ts);
1328 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1329 			timo = tvtohz(&tv);
1330 		} else {
1331 			timo = 0;
1332 		}
1333 
1334 		if (traced) {
1335 			error = EINTR;
1336 			break;
1337 		}
1338 
1339 		error = msleep(&p->p_sigacts, &p->p_mtx, PPAUSE | PCATCH,
1340 		    "sigwait", timo);
1341 
1342 		/* The syscalls can not be restarted. */
1343 		if (error == ERESTART)
1344 			error = EINTR;
1345 
1346 		/* We will calculate timeout by ourself. */
1347 		if (timeout != NULL && error == EAGAIN)
1348 			error = 0;
1349 
1350 		/*
1351 		 * If PTRACE_SCE or PTRACE_SCX were set after
1352 		 * userspace entered the syscall, return spurious
1353 		 * EINTR after wait was done.  Only do this as last
1354 		 * resort after rechecking for possible queued signals
1355 		 * and expired timeouts.
1356 		 */
1357 		if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1358 			traced = true;
1359 	}
1360 	thread_lock(td);
1361 	td->td_flags &= ~TDF_SIGWAIT;
1362 	thread_unlock(td);
1363 
1364 	new_block = saved_mask;
1365 	SIGSETNAND(new_block, td->td_sigmask);
1366 	td->td_sigmask = saved_mask;
1367 	/*
1368 	 * Fewer signals can be delivered to us, reschedule signal
1369 	 * notification.
1370 	 */
1371 	if (p->p_numthreads != 1)
1372 		reschedule_signals(p, new_block, 0);
1373 
1374 	if (error == 0) {
1375 		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1376 
1377 		if (ksi->ksi_code == SI_TIMER)
1378 			itimer_accept(p, ksi->ksi_timerid, ksi);
1379 
1380 #ifdef KTRACE
1381 		if (KTRPOINT(td, KTR_PSIG)) {
1382 			sig_t action;
1383 
1384 			mtx_lock(&ps->ps_mtx);
1385 			action = ps->ps_sigact[_SIG_IDX(sig)];
1386 			mtx_unlock(&ps->ps_mtx);
1387 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1388 		}
1389 #endif
1390 		if (sig == SIGKILL) {
1391 			proc_td_siginfo_capture(td, &ksi->ksi_info);
1392 			sigexit(td, sig);
1393 		}
1394 	}
1395 	PROC_UNLOCK(p);
1396 	return (error);
1397 }
1398 
1399 #ifndef _SYS_SYSPROTO_H_
1400 struct sigpending_args {
1401 	sigset_t	*set;
1402 };
1403 #endif
1404 int
1405 sys_sigpending(struct thread *td, struct sigpending_args *uap)
1406 {
1407 	struct proc *p = td->td_proc;
1408 	sigset_t pending;
1409 
1410 	PROC_LOCK(p);
1411 	pending = p->p_sigqueue.sq_signals;
1412 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1413 	PROC_UNLOCK(p);
1414 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1415 }
1416 
1417 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1418 #ifndef _SYS_SYSPROTO_H_
1419 struct osigpending_args {
1420 	int	dummy;
1421 };
1422 #endif
1423 int
1424 osigpending(struct thread *td, struct osigpending_args *uap)
1425 {
1426 	struct proc *p = td->td_proc;
1427 	sigset_t pending;
1428 
1429 	PROC_LOCK(p);
1430 	pending = p->p_sigqueue.sq_signals;
1431 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1432 	PROC_UNLOCK(p);
1433 	SIG2OSIG(pending, td->td_retval[0]);
1434 	return (0);
1435 }
1436 #endif /* COMPAT_43 */
1437 
1438 #if defined(COMPAT_43)
1439 /*
1440  * Generalized interface signal handler, 4.3-compatible.
1441  */
1442 #ifndef _SYS_SYSPROTO_H_
1443 struct osigvec_args {
1444 	int	signum;
1445 	struct	sigvec *nsv;
1446 	struct	sigvec *osv;
1447 };
1448 #endif
1449 /* ARGSUSED */
1450 int
1451 osigvec(struct thread *td, struct osigvec_args *uap)
1452 {
1453 	struct sigvec vec;
1454 	struct sigaction nsa, osa;
1455 	struct sigaction *nsap, *osap;
1456 	int error;
1457 
1458 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1459 		return (EINVAL);
1460 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1461 	osap = (uap->osv != NULL) ? &osa : NULL;
1462 	if (nsap) {
1463 		error = copyin(uap->nsv, &vec, sizeof(vec));
1464 		if (error)
1465 			return (error);
1466 		nsap->sa_handler = vec.sv_handler;
1467 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1468 		nsap->sa_flags = vec.sv_flags;
1469 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1470 	}
1471 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1472 	if (osap && !error) {
1473 		vec.sv_handler = osap->sa_handler;
1474 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1475 		vec.sv_flags = osap->sa_flags;
1476 		vec.sv_flags &= ~SA_NOCLDWAIT;
1477 		vec.sv_flags ^= SA_RESTART;
1478 		error = copyout(&vec, uap->osv, sizeof(vec));
1479 	}
1480 	return (error);
1481 }
1482 
1483 #ifndef _SYS_SYSPROTO_H_
1484 struct osigblock_args {
1485 	int	mask;
1486 };
1487 #endif
1488 int
1489 osigblock(struct thread *td, struct osigblock_args *uap)
1490 {
1491 	sigset_t set, oset;
1492 
1493 	OSIG2SIG(uap->mask, set);
1494 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1495 	SIG2OSIG(oset, td->td_retval[0]);
1496 	return (0);
1497 }
1498 
1499 #ifndef _SYS_SYSPROTO_H_
1500 struct osigsetmask_args {
1501 	int	mask;
1502 };
1503 #endif
1504 int
1505 osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1506 {
1507 	sigset_t set, oset;
1508 
1509 	OSIG2SIG(uap->mask, set);
1510 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1511 	SIG2OSIG(oset, td->td_retval[0]);
1512 	return (0);
1513 }
1514 #endif /* COMPAT_43 */
1515 
1516 /*
1517  * Suspend calling thread until signal, providing mask to be set in the
1518  * meantime.
1519  */
1520 #ifndef _SYS_SYSPROTO_H_
1521 struct sigsuspend_args {
1522 	const sigset_t *sigmask;
1523 };
1524 #endif
1525 /* ARGSUSED */
1526 int
1527 sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1528 {
1529 	sigset_t mask;
1530 	int error;
1531 
1532 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1533 	if (error)
1534 		return (error);
1535 	return (kern_sigsuspend(td, mask));
1536 }
1537 
1538 int
1539 kern_sigsuspend(struct thread *td, sigset_t mask)
1540 {
1541 	struct proc *p = td->td_proc;
1542 	int has_sig, sig;
1543 
1544 	/* Ensure the sigfastblock value is up to date. */
1545 	sigfastblock_fetch(td);
1546 
1547 	/*
1548 	 * When returning from sigsuspend, we want
1549 	 * the old mask to be restored after the
1550 	 * signal handler has finished.  Thus, we
1551 	 * save it here and mark the sigacts structure
1552 	 * to indicate this.
1553 	 */
1554 	PROC_LOCK(p);
1555 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1556 	    SIGPROCMASK_PROC_LOCKED);
1557 	td->td_pflags |= TDP_OLDMASK;
1558 
1559 	/*
1560 	 * Process signals now. Otherwise, we can get spurious wakeup
1561 	 * due to signal entered process queue, but delivered to other
1562 	 * thread. But sigsuspend should return only on signal
1563 	 * delivery.
1564 	 */
1565 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1566 	for (has_sig = 0; !has_sig;) {
1567 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1568 			0) == 0)
1569 			/* void */;
1570 		thread_suspend_check(0);
1571 		mtx_lock(&p->p_sigacts->ps_mtx);
1572 		while ((sig = cursig(td)) != 0) {
1573 			KASSERT(sig >= 0, ("sig %d", sig));
1574 			has_sig += postsig(sig);
1575 		}
1576 		mtx_unlock(&p->p_sigacts->ps_mtx);
1577 
1578 		/*
1579 		 * If PTRACE_SCE or PTRACE_SCX were set after
1580 		 * userspace entered the syscall, return spurious
1581 		 * EINTR.
1582 		 */
1583 		if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1584 			has_sig += 1;
1585 	}
1586 	PROC_UNLOCK(p);
1587 	td->td_errno = EINTR;
1588 	td->td_pflags |= TDP_NERRNO;
1589 	return (EJUSTRETURN);
1590 }
1591 
1592 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1593 /*
1594  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1595  * convention: libc stub passes mask, not pointer, to save a copyin.
1596  */
1597 #ifndef _SYS_SYSPROTO_H_
1598 struct osigsuspend_args {
1599 	osigset_t mask;
1600 };
1601 #endif
1602 /* ARGSUSED */
1603 int
1604 osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1605 {
1606 	sigset_t mask;
1607 
1608 	OSIG2SIG(uap->mask, mask);
1609 	return (kern_sigsuspend(td, mask));
1610 }
1611 #endif /* COMPAT_43 */
1612 
1613 #if defined(COMPAT_43)
1614 #ifndef _SYS_SYSPROTO_H_
1615 struct osigstack_args {
1616 	struct	sigstack *nss;
1617 	struct	sigstack *oss;
1618 };
1619 #endif
1620 /* ARGSUSED */
1621 int
1622 osigstack(struct thread *td, struct osigstack_args *uap)
1623 {
1624 	struct sigstack nss, oss;
1625 	int error = 0;
1626 
1627 	if (uap->nss != NULL) {
1628 		error = copyin(uap->nss, &nss, sizeof(nss));
1629 		if (error)
1630 			return (error);
1631 	}
1632 	oss.ss_sp = td->td_sigstk.ss_sp;
1633 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1634 	if (uap->nss != NULL) {
1635 		td->td_sigstk.ss_sp = nss.ss_sp;
1636 		td->td_sigstk.ss_size = 0;
1637 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1638 		td->td_pflags |= TDP_ALTSTACK;
1639 	}
1640 	if (uap->oss != NULL)
1641 		error = copyout(&oss, uap->oss, sizeof(oss));
1642 
1643 	return (error);
1644 }
1645 #endif /* COMPAT_43 */
1646 
1647 #ifndef _SYS_SYSPROTO_H_
1648 struct sigaltstack_args {
1649 	stack_t	*ss;
1650 	stack_t	*oss;
1651 };
1652 #endif
1653 /* ARGSUSED */
1654 int
1655 sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1656 {
1657 	stack_t ss, oss;
1658 	int error;
1659 
1660 	if (uap->ss != NULL) {
1661 		error = copyin(uap->ss, &ss, sizeof(ss));
1662 		if (error)
1663 			return (error);
1664 	}
1665 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1666 	    (uap->oss != NULL) ? &oss : NULL);
1667 	if (error)
1668 		return (error);
1669 	if (uap->oss != NULL)
1670 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1671 	return (error);
1672 }
1673 
1674 int
1675 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1676 {
1677 	struct proc *p = td->td_proc;
1678 	int oonstack;
1679 
1680 	oonstack = sigonstack(cpu_getstack(td));
1681 
1682 	if (oss != NULL) {
1683 		*oss = td->td_sigstk;
1684 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1685 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1686 	}
1687 
1688 	if (ss != NULL) {
1689 		if (oonstack)
1690 			return (EPERM);
1691 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1692 			return (EINVAL);
1693 		if (!(ss->ss_flags & SS_DISABLE)) {
1694 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1695 				return (ENOMEM);
1696 
1697 			td->td_sigstk = *ss;
1698 			td->td_pflags |= TDP_ALTSTACK;
1699 		} else {
1700 			td->td_pflags &= ~TDP_ALTSTACK;
1701 		}
1702 	}
1703 	return (0);
1704 }
1705 
1706 struct killpg1_ctx {
1707 	struct thread *td;
1708 	ksiginfo_t *ksi;
1709 	int sig;
1710 	bool sent;
1711 	bool found;
1712 	int ret;
1713 };
1714 
1715 static void
1716 killpg1_sendsig(struct proc *p, bool notself, struct killpg1_ctx *arg)
1717 {
1718 	int err;
1719 
1720 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1721 	    (notself && p == arg->td->td_proc) || p->p_state == PRS_NEW)
1722 		return;
1723 	PROC_LOCK(p);
1724 	err = p_cansignal(arg->td, p, arg->sig);
1725 	if (err == 0 && arg->sig != 0)
1726 		pksignal(p, arg->sig, arg->ksi);
1727 	PROC_UNLOCK(p);
1728 	if (err != ESRCH)
1729 		arg->found = true;
1730 	if (err == 0)
1731 		arg->sent = true;
1732 	else if (arg->ret == 0 && err != ESRCH && err != EPERM)
1733 		arg->ret = err;
1734 }
1735 
1736 /*
1737  * Common code for kill process group/broadcast kill.
1738  * cp is calling process.
1739  */
1740 static int
1741 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1742 {
1743 	struct proc *p;
1744 	struct pgrp *pgrp;
1745 	struct killpg1_ctx arg;
1746 
1747 	arg.td = td;
1748 	arg.ksi = ksi;
1749 	arg.sig = sig;
1750 	arg.sent = false;
1751 	arg.found = false;
1752 	arg.ret = 0;
1753 	if (all) {
1754 		/*
1755 		 * broadcast
1756 		 */
1757 		sx_slock(&allproc_lock);
1758 		FOREACH_PROC_IN_SYSTEM(p) {
1759 			killpg1_sendsig(p, true, &arg);
1760 		}
1761 		sx_sunlock(&allproc_lock);
1762 	} else {
1763 		sx_slock(&proctree_lock);
1764 		if (pgid == 0) {
1765 			/*
1766 			 * zero pgid means send to my process group.
1767 			 */
1768 			pgrp = td->td_proc->p_pgrp;
1769 			PGRP_LOCK(pgrp);
1770 		} else {
1771 			pgrp = pgfind(pgid);
1772 			if (pgrp == NULL) {
1773 				sx_sunlock(&proctree_lock);
1774 				return (ESRCH);
1775 			}
1776 		}
1777 		sx_sunlock(&proctree_lock);
1778 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1779 			killpg1_sendsig(p, false, &arg);
1780 		}
1781 		PGRP_UNLOCK(pgrp);
1782 	}
1783 	MPASS(arg.ret != 0 || arg.found || !arg.sent);
1784 	if (arg.ret == 0 && !arg.sent)
1785 		arg.ret = arg.found ? EPERM : ESRCH;
1786 	return (arg.ret);
1787 }
1788 
1789 #ifndef _SYS_SYSPROTO_H_
1790 struct kill_args {
1791 	int	pid;
1792 	int	signum;
1793 };
1794 #endif
1795 /* ARGSUSED */
1796 int
1797 sys_kill(struct thread *td, struct kill_args *uap)
1798 {
1799 
1800 	return (kern_kill(td, uap->pid, uap->signum));
1801 }
1802 
1803 int
1804 kern_kill(struct thread *td, pid_t pid, int signum)
1805 {
1806 	ksiginfo_t ksi;
1807 	struct proc *p;
1808 	int error;
1809 
1810 	/*
1811 	 * A process in capability mode can send signals only to himself.
1812 	 * The main rationale behind this is that abort(3) is implemented as
1813 	 * kill(getpid(), SIGABRT).
1814 	 */
1815 	if (IN_CAPABILITY_MODE(td) && pid != td->td_proc->p_pid)
1816 		return (ECAPMODE);
1817 
1818 	AUDIT_ARG_SIGNUM(signum);
1819 	AUDIT_ARG_PID(pid);
1820 	if ((u_int)signum > _SIG_MAXSIG)
1821 		return (EINVAL);
1822 
1823 	ksiginfo_init(&ksi);
1824 	ksi.ksi_signo = signum;
1825 	ksi.ksi_code = SI_USER;
1826 	ksi.ksi_pid = td->td_proc->p_pid;
1827 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1828 
1829 	if (pid > 0) {
1830 		/* kill single process */
1831 		if ((p = pfind_any(pid)) == NULL)
1832 			return (ESRCH);
1833 		AUDIT_ARG_PROCESS(p);
1834 		error = p_cansignal(td, p, signum);
1835 		if (error == 0 && signum)
1836 			pksignal(p, signum, &ksi);
1837 		PROC_UNLOCK(p);
1838 		return (error);
1839 	}
1840 	switch (pid) {
1841 	case -1:		/* broadcast signal */
1842 		return (killpg1(td, signum, 0, 1, &ksi));
1843 	case 0:			/* signal own process group */
1844 		return (killpg1(td, signum, 0, 0, &ksi));
1845 	default:		/* negative explicit process group */
1846 		return (killpg1(td, signum, -pid, 0, &ksi));
1847 	}
1848 	/* NOTREACHED */
1849 }
1850 
1851 int
1852 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1853 {
1854 	struct proc *p;
1855 	int error;
1856 
1857 	AUDIT_ARG_SIGNUM(uap->signum);
1858 	AUDIT_ARG_FD(uap->fd);
1859 	if ((u_int)uap->signum > _SIG_MAXSIG)
1860 		return (EINVAL);
1861 
1862 	error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1863 	if (error)
1864 		return (error);
1865 	AUDIT_ARG_PROCESS(p);
1866 	error = p_cansignal(td, p, uap->signum);
1867 	if (error == 0 && uap->signum)
1868 		kern_psignal(p, uap->signum);
1869 	PROC_UNLOCK(p);
1870 	return (error);
1871 }
1872 
1873 #if defined(COMPAT_43)
1874 #ifndef _SYS_SYSPROTO_H_
1875 struct okillpg_args {
1876 	int	pgid;
1877 	int	signum;
1878 };
1879 #endif
1880 /* ARGSUSED */
1881 int
1882 okillpg(struct thread *td, struct okillpg_args *uap)
1883 {
1884 	ksiginfo_t ksi;
1885 
1886 	AUDIT_ARG_SIGNUM(uap->signum);
1887 	AUDIT_ARG_PID(uap->pgid);
1888 	if ((u_int)uap->signum > _SIG_MAXSIG)
1889 		return (EINVAL);
1890 
1891 	ksiginfo_init(&ksi);
1892 	ksi.ksi_signo = uap->signum;
1893 	ksi.ksi_code = SI_USER;
1894 	ksi.ksi_pid = td->td_proc->p_pid;
1895 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1896 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1897 }
1898 #endif /* COMPAT_43 */
1899 
1900 #ifndef _SYS_SYSPROTO_H_
1901 struct sigqueue_args {
1902 	pid_t pid;
1903 	int signum;
1904 	/* union sigval */ void *value;
1905 };
1906 #endif
1907 int
1908 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1909 {
1910 	union sigval sv;
1911 
1912 	sv.sival_ptr = uap->value;
1913 
1914 	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
1915 }
1916 
1917 int
1918 kern_sigqueue(struct thread *td, pid_t pid, int signum, union sigval *value)
1919 {
1920 	ksiginfo_t ksi;
1921 	struct proc *p;
1922 	int error;
1923 
1924 	if ((u_int)signum > _SIG_MAXSIG)
1925 		return (EINVAL);
1926 
1927 	/*
1928 	 * Specification says sigqueue can only send signal to
1929 	 * single process.
1930 	 */
1931 	if (pid <= 0)
1932 		return (EINVAL);
1933 
1934 	if ((p = pfind_any(pid)) == NULL)
1935 		return (ESRCH);
1936 	error = p_cansignal(td, p, signum);
1937 	if (error == 0 && signum != 0) {
1938 		ksiginfo_init(&ksi);
1939 		ksi.ksi_flags = KSI_SIGQ;
1940 		ksi.ksi_signo = signum;
1941 		ksi.ksi_code = SI_QUEUE;
1942 		ksi.ksi_pid = td->td_proc->p_pid;
1943 		ksi.ksi_uid = td->td_ucred->cr_ruid;
1944 		ksi.ksi_value = *value;
1945 		error = pksignal(p, ksi.ksi_signo, &ksi);
1946 	}
1947 	PROC_UNLOCK(p);
1948 	return (error);
1949 }
1950 
1951 /*
1952  * Send a signal to a process group.
1953  */
1954 void
1955 gsignal(int pgid, int sig, ksiginfo_t *ksi)
1956 {
1957 	struct pgrp *pgrp;
1958 
1959 	if (pgid != 0) {
1960 		sx_slock(&proctree_lock);
1961 		pgrp = pgfind(pgid);
1962 		sx_sunlock(&proctree_lock);
1963 		if (pgrp != NULL) {
1964 			pgsignal(pgrp, sig, 0, ksi);
1965 			PGRP_UNLOCK(pgrp);
1966 		}
1967 	}
1968 }
1969 
1970 /*
1971  * Send a signal to a process group.  If checktty is 1,
1972  * limit to members which have a controlling terminal.
1973  */
1974 void
1975 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
1976 {
1977 	struct proc *p;
1978 
1979 	if (pgrp) {
1980 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
1981 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1982 			PROC_LOCK(p);
1983 			if (p->p_state == PRS_NORMAL &&
1984 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
1985 				pksignal(p, sig, ksi);
1986 			PROC_UNLOCK(p);
1987 		}
1988 	}
1989 }
1990 
1991 /*
1992  * Recalculate the signal mask and reset the signal disposition after
1993  * usermode frame for delivery is formed.  Should be called after
1994  * mach-specific routine, because sysent->sv_sendsig() needs correct
1995  * ps_siginfo and signal mask.
1996  */
1997 static void
1998 postsig_done(int sig, struct thread *td, struct sigacts *ps)
1999 {
2000 	sigset_t mask;
2001 
2002 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2003 	td->td_ru.ru_nsignals++;
2004 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
2005 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
2006 		SIGADDSET(mask, sig);
2007 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
2008 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
2009 	if (SIGISMEMBER(ps->ps_sigreset, sig))
2010 		sigdflt(ps, sig);
2011 }
2012 
2013 /*
2014  * Send a signal caused by a trap to the current thread.  If it will be
2015  * caught immediately, deliver it with correct code.  Otherwise, post it
2016  * normally.
2017  */
2018 void
2019 trapsignal(struct thread *td, ksiginfo_t *ksi)
2020 {
2021 	struct sigacts *ps;
2022 	struct proc *p;
2023 	sigset_t sigmask;
2024 	int code, sig;
2025 
2026 	p = td->td_proc;
2027 	sig = ksi->ksi_signo;
2028 	code = ksi->ksi_code;
2029 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
2030 
2031 	sigfastblock_fetch(td);
2032 	PROC_LOCK(p);
2033 	ps = p->p_sigacts;
2034 	mtx_lock(&ps->ps_mtx);
2035 	sigmask = td->td_sigmask;
2036 	if (td->td_sigblock_val != 0)
2037 		SIGSETOR(sigmask, fastblock_mask);
2038 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2039 	    !SIGISMEMBER(sigmask, sig)) {
2040 #ifdef KTRACE
2041 		if (KTRPOINT(curthread, KTR_PSIG))
2042 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2043 			    &td->td_sigmask, code);
2044 #endif
2045 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2046 		    ksi, &td->td_sigmask);
2047 		postsig_done(sig, td, ps);
2048 		mtx_unlock(&ps->ps_mtx);
2049 	} else {
2050 		/*
2051 		 * Avoid a possible infinite loop if the thread
2052 		 * masking the signal or process is ignoring the
2053 		 * signal.
2054 		 */
2055 		if (kern_forcesigexit && (SIGISMEMBER(sigmask, sig) ||
2056 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2057 			SIGDELSET(td->td_sigmask, sig);
2058 			SIGDELSET(ps->ps_sigcatch, sig);
2059 			SIGDELSET(ps->ps_sigignore, sig);
2060 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2061 			td->td_pflags &= ~TDP_SIGFASTBLOCK;
2062 			td->td_sigblock_val = 0;
2063 		}
2064 		mtx_unlock(&ps->ps_mtx);
2065 		p->p_sig = sig;		/* XXX to verify code */
2066 		tdsendsignal(p, td, sig, ksi);
2067 	}
2068 	PROC_UNLOCK(p);
2069 }
2070 
2071 static struct thread *
2072 sigtd(struct proc *p, int sig, bool fast_sigblock)
2073 {
2074 	struct thread *td, *signal_td;
2075 
2076 	PROC_LOCK_ASSERT(p, MA_OWNED);
2077 	MPASS(!fast_sigblock || p == curproc);
2078 
2079 	/*
2080 	 * Check if current thread can handle the signal without
2081 	 * switching context to another thread.
2082 	 */
2083 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig) &&
2084 	    (!fast_sigblock || curthread->td_sigblock_val == 0))
2085 		return (curthread);
2086 	signal_td = NULL;
2087 	FOREACH_THREAD_IN_PROC(p, td) {
2088 		if (!SIGISMEMBER(td->td_sigmask, sig) && (!fast_sigblock ||
2089 		    td != curthread || td->td_sigblock_val == 0)) {
2090 			signal_td = td;
2091 			break;
2092 		}
2093 	}
2094 	if (signal_td == NULL)
2095 		signal_td = FIRST_THREAD_IN_PROC(p);
2096 	return (signal_td);
2097 }
2098 
2099 /*
2100  * Send the signal to the process.  If the signal has an action, the action
2101  * is usually performed by the target process rather than the caller; we add
2102  * the signal to the set of pending signals for the process.
2103  *
2104  * Exceptions:
2105  *   o When a stop signal is sent to a sleeping process that takes the
2106  *     default action, the process is stopped without awakening it.
2107  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2108  *     regardless of the signal action (eg, blocked or ignored).
2109  *
2110  * Other ignored signals are discarded immediately.
2111  *
2112  * NB: This function may be entered from the debugger via the "kill" DDB
2113  * command.  There is little that can be done to mitigate the possibly messy
2114  * side effects of this unwise possibility.
2115  */
2116 void
2117 kern_psignal(struct proc *p, int sig)
2118 {
2119 	ksiginfo_t ksi;
2120 
2121 	ksiginfo_init(&ksi);
2122 	ksi.ksi_signo = sig;
2123 	ksi.ksi_code = SI_KERNEL;
2124 	(void) tdsendsignal(p, NULL, sig, &ksi);
2125 }
2126 
2127 int
2128 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2129 {
2130 
2131 	return (tdsendsignal(p, NULL, sig, ksi));
2132 }
2133 
2134 /* Utility function for finding a thread to send signal event to. */
2135 int
2136 sigev_findtd(struct proc *p, struct sigevent *sigev, struct thread **ttd)
2137 {
2138 	struct thread *td;
2139 
2140 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2141 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2142 		if (td == NULL)
2143 			return (ESRCH);
2144 		*ttd = td;
2145 	} else {
2146 		*ttd = NULL;
2147 		PROC_LOCK(p);
2148 	}
2149 	return (0);
2150 }
2151 
2152 void
2153 tdsignal(struct thread *td, int sig)
2154 {
2155 	ksiginfo_t ksi;
2156 
2157 	ksiginfo_init(&ksi);
2158 	ksi.ksi_signo = sig;
2159 	ksi.ksi_code = SI_KERNEL;
2160 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2161 }
2162 
2163 void
2164 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2165 {
2166 
2167 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2168 }
2169 
2170 static int
2171 sig_sleepq_abort(struct thread *td, int intrval)
2172 {
2173 	THREAD_LOCK_ASSERT(td, MA_OWNED);
2174 
2175 	if (intrval == 0 && (td->td_flags & TDF_SIGWAIT) == 0) {
2176 		thread_unlock(td);
2177 		return (0);
2178 	}
2179 	return (sleepq_abort(td, intrval));
2180 }
2181 
2182 int
2183 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2184 {
2185 	sig_t action;
2186 	sigqueue_t *sigqueue;
2187 	int prop;
2188 	struct sigacts *ps;
2189 	int intrval;
2190 	int ret = 0;
2191 	int wakeup_swapper;
2192 
2193 	MPASS(td == NULL || p == td->td_proc);
2194 	PROC_LOCK_ASSERT(p, MA_OWNED);
2195 
2196 	if (!_SIG_VALID(sig))
2197 		panic("%s(): invalid signal %d", __func__, sig);
2198 
2199 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2200 
2201 	/*
2202 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2203 	 */
2204 	if (p->p_state == PRS_ZOMBIE) {
2205 		if (ksi && (ksi->ksi_flags & KSI_INS))
2206 			ksiginfo_tryfree(ksi);
2207 		return (ret);
2208 	}
2209 
2210 	ps = p->p_sigacts;
2211 	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2212 	prop = sigprop(sig);
2213 
2214 	if (td == NULL) {
2215 		td = sigtd(p, sig, false);
2216 		sigqueue = &p->p_sigqueue;
2217 	} else
2218 		sigqueue = &td->td_sigqueue;
2219 
2220 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2221 
2222 	/*
2223 	 * If the signal is being ignored, then we forget about it
2224 	 * immediately, except when the target process executes
2225 	 * sigwait().  (Note: we don't set SIGCONT in ps_sigignore,
2226 	 * and if it is set to SIG_IGN, action will be SIG_DFL here.)
2227 	 */
2228 	mtx_lock(&ps->ps_mtx);
2229 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2230 		if (kern_sig_discard_ign &&
2231 		    (p->p_sysent->sv_flags & SV_SIG_DISCIGN) == 0) {
2232 			SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2233 
2234 			mtx_unlock(&ps->ps_mtx);
2235 			if (ksi && (ksi->ksi_flags & KSI_INS))
2236 				ksiginfo_tryfree(ksi);
2237 			return (ret);
2238 		} else {
2239 			action = SIG_CATCH;
2240 			intrval = 0;
2241 		}
2242 	} else {
2243 		if (SIGISMEMBER(td->td_sigmask, sig))
2244 			action = SIG_HOLD;
2245 		else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2246 			action = SIG_CATCH;
2247 		else
2248 			action = SIG_DFL;
2249 		if (SIGISMEMBER(ps->ps_sigintr, sig))
2250 			intrval = EINTR;
2251 		else
2252 			intrval = ERESTART;
2253 	}
2254 	mtx_unlock(&ps->ps_mtx);
2255 
2256 	if (prop & SIGPROP_CONT)
2257 		sigqueue_delete_stopmask_proc(p);
2258 	else if (prop & SIGPROP_STOP) {
2259 		/*
2260 		 * If sending a tty stop signal to a member of an orphaned
2261 		 * process group, discard the signal here if the action
2262 		 * is default; don't stop the process below if sleeping,
2263 		 * and don't clear any pending SIGCONT.
2264 		 */
2265 		if ((prop & SIGPROP_TTYSTOP) != 0 &&
2266 		    (p->p_pgrp->pg_flags & PGRP_ORPHANED) != 0 &&
2267 		    action == SIG_DFL) {
2268 			if (ksi && (ksi->ksi_flags & KSI_INS))
2269 				ksiginfo_tryfree(ksi);
2270 			return (ret);
2271 		}
2272 		sigqueue_delete_proc(p, SIGCONT);
2273 		if (p->p_flag & P_CONTINUED) {
2274 			p->p_flag &= ~P_CONTINUED;
2275 			PROC_LOCK(p->p_pptr);
2276 			sigqueue_take(p->p_ksi);
2277 			PROC_UNLOCK(p->p_pptr);
2278 		}
2279 	}
2280 
2281 	ret = sigqueue_add(sigqueue, sig, ksi);
2282 	if (ret != 0)
2283 		return (ret);
2284 	signotify(td);
2285 	/*
2286 	 * Defer further processing for signals which are held,
2287 	 * except that stopped processes must be continued by SIGCONT.
2288 	 */
2289 	if (action == SIG_HOLD &&
2290 	    !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2291 		return (ret);
2292 
2293 	wakeup_swapper = 0;
2294 
2295 	/*
2296 	 * Some signals have a process-wide effect and a per-thread
2297 	 * component.  Most processing occurs when the process next
2298 	 * tries to cross the user boundary, however there are some
2299 	 * times when processing needs to be done immediately, such as
2300 	 * waking up threads so that they can cross the user boundary.
2301 	 * We try to do the per-process part here.
2302 	 */
2303 	if (P_SHOULDSTOP(p)) {
2304 		KASSERT(!(p->p_flag & P_WEXIT),
2305 		    ("signal to stopped but exiting process"));
2306 		if (sig == SIGKILL) {
2307 			/*
2308 			 * If traced process is already stopped,
2309 			 * then no further action is necessary.
2310 			 */
2311 			if (p->p_flag & P_TRACED)
2312 				goto out;
2313 			/*
2314 			 * SIGKILL sets process running.
2315 			 * It will die elsewhere.
2316 			 * All threads must be restarted.
2317 			 */
2318 			p->p_flag &= ~P_STOPPED_SIG;
2319 			goto runfast;
2320 		}
2321 
2322 		if (prop & SIGPROP_CONT) {
2323 			/*
2324 			 * If traced process is already stopped,
2325 			 * then no further action is necessary.
2326 			 */
2327 			if (p->p_flag & P_TRACED)
2328 				goto out;
2329 			/*
2330 			 * If SIGCONT is default (or ignored), we continue the
2331 			 * process but don't leave the signal in sigqueue as
2332 			 * it has no further action.  If SIGCONT is held, we
2333 			 * continue the process and leave the signal in
2334 			 * sigqueue.  If the process catches SIGCONT, let it
2335 			 * handle the signal itself.  If it isn't waiting on
2336 			 * an event, it goes back to run state.
2337 			 * Otherwise, process goes back to sleep state.
2338 			 */
2339 			p->p_flag &= ~P_STOPPED_SIG;
2340 			PROC_SLOCK(p);
2341 			if (p->p_numthreads == p->p_suspcount) {
2342 				PROC_SUNLOCK(p);
2343 				p->p_flag |= P_CONTINUED;
2344 				p->p_xsig = SIGCONT;
2345 				PROC_LOCK(p->p_pptr);
2346 				childproc_continued(p);
2347 				PROC_UNLOCK(p->p_pptr);
2348 				PROC_SLOCK(p);
2349 			}
2350 			if (action == SIG_DFL) {
2351 				thread_unsuspend(p);
2352 				PROC_SUNLOCK(p);
2353 				sigqueue_delete(sigqueue, sig);
2354 				goto out_cont;
2355 			}
2356 			if (action == SIG_CATCH) {
2357 				/*
2358 				 * The process wants to catch it so it needs
2359 				 * to run at least one thread, but which one?
2360 				 */
2361 				PROC_SUNLOCK(p);
2362 				goto runfast;
2363 			}
2364 			/*
2365 			 * The signal is not ignored or caught.
2366 			 */
2367 			thread_unsuspend(p);
2368 			PROC_SUNLOCK(p);
2369 			goto out_cont;
2370 		}
2371 
2372 		if (prop & SIGPROP_STOP) {
2373 			/*
2374 			 * If traced process is already stopped,
2375 			 * then no further action is necessary.
2376 			 */
2377 			if (p->p_flag & P_TRACED)
2378 				goto out;
2379 			/*
2380 			 * Already stopped, don't need to stop again
2381 			 * (If we did the shell could get confused).
2382 			 * Just make sure the signal STOP bit set.
2383 			 */
2384 			p->p_flag |= P_STOPPED_SIG;
2385 			sigqueue_delete(sigqueue, sig);
2386 			goto out;
2387 		}
2388 
2389 		/*
2390 		 * All other kinds of signals:
2391 		 * If a thread is sleeping interruptibly, simulate a
2392 		 * wakeup so that when it is continued it will be made
2393 		 * runnable and can look at the signal.  However, don't make
2394 		 * the PROCESS runnable, leave it stopped.
2395 		 * It may run a bit until it hits a thread_suspend_check().
2396 		 */
2397 		PROC_SLOCK(p);
2398 		thread_lock(td);
2399 		if (TD_CAN_ABORT(td))
2400 			wakeup_swapper = sig_sleepq_abort(td, intrval);
2401 		else
2402 			thread_unlock(td);
2403 		PROC_SUNLOCK(p);
2404 		goto out;
2405 		/*
2406 		 * Mutexes are short lived. Threads waiting on them will
2407 		 * hit thread_suspend_check() soon.
2408 		 */
2409 	} else if (p->p_state == PRS_NORMAL) {
2410 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2411 			tdsigwakeup(td, sig, action, intrval);
2412 			goto out;
2413 		}
2414 
2415 		MPASS(action == SIG_DFL);
2416 
2417 		if (prop & SIGPROP_STOP) {
2418 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2419 				goto out;
2420 			p->p_flag |= P_STOPPED_SIG;
2421 			p->p_xsig = sig;
2422 			PROC_SLOCK(p);
2423 			wakeup_swapper = sig_suspend_threads(td, p, 1);
2424 			if (p->p_numthreads == p->p_suspcount) {
2425 				/*
2426 				 * only thread sending signal to another
2427 				 * process can reach here, if thread is sending
2428 				 * signal to its process, because thread does
2429 				 * not suspend itself here, p_numthreads
2430 				 * should never be equal to p_suspcount.
2431 				 */
2432 				thread_stopped(p);
2433 				PROC_SUNLOCK(p);
2434 				sigqueue_delete_proc(p, p->p_xsig);
2435 			} else
2436 				PROC_SUNLOCK(p);
2437 			goto out;
2438 		}
2439 	} else {
2440 		/* Not in "NORMAL" state. discard the signal. */
2441 		sigqueue_delete(sigqueue, sig);
2442 		goto out;
2443 	}
2444 
2445 	/*
2446 	 * The process is not stopped so we need to apply the signal to all the
2447 	 * running threads.
2448 	 */
2449 runfast:
2450 	tdsigwakeup(td, sig, action, intrval);
2451 	PROC_SLOCK(p);
2452 	thread_unsuspend(p);
2453 	PROC_SUNLOCK(p);
2454 out_cont:
2455 	itimer_proc_continue(p);
2456 	kqtimer_proc_continue(p);
2457 out:
2458 	/* If we jump here, proc slock should not be owned. */
2459 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2460 	if (wakeup_swapper)
2461 		kick_proc0();
2462 
2463 	return (ret);
2464 }
2465 
2466 /*
2467  * The force of a signal has been directed against a single
2468  * thread.  We need to see what we can do about knocking it
2469  * out of any sleep it may be in etc.
2470  */
2471 static void
2472 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2473 {
2474 	struct proc *p = td->td_proc;
2475 	int prop, wakeup_swapper;
2476 
2477 	PROC_LOCK_ASSERT(p, MA_OWNED);
2478 	prop = sigprop(sig);
2479 
2480 	PROC_SLOCK(p);
2481 	thread_lock(td);
2482 	/*
2483 	 * Bring the priority of a thread up if we want it to get
2484 	 * killed in this lifetime.  Be careful to avoid bumping the
2485 	 * priority of the idle thread, since we still allow to signal
2486 	 * kernel processes.
2487 	 */
2488 	if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2489 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2490 		sched_prio(td, PUSER);
2491 	if (TD_ON_SLEEPQ(td)) {
2492 		/*
2493 		 * If thread is sleeping uninterruptibly
2494 		 * we can't interrupt the sleep... the signal will
2495 		 * be noticed when the process returns through
2496 		 * trap() or syscall().
2497 		 */
2498 		if ((td->td_flags & TDF_SINTR) == 0)
2499 			goto out;
2500 		/*
2501 		 * If SIGCONT is default (or ignored) and process is
2502 		 * asleep, we are finished; the process should not
2503 		 * be awakened.
2504 		 */
2505 		if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2506 			thread_unlock(td);
2507 			PROC_SUNLOCK(p);
2508 			sigqueue_delete(&p->p_sigqueue, sig);
2509 			/*
2510 			 * It may be on either list in this state.
2511 			 * Remove from both for now.
2512 			 */
2513 			sigqueue_delete(&td->td_sigqueue, sig);
2514 			return;
2515 		}
2516 
2517 		/*
2518 		 * Don't awaken a sleeping thread for SIGSTOP if the
2519 		 * STOP signal is deferred.
2520 		 */
2521 		if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2522 		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2523 			goto out;
2524 
2525 		/*
2526 		 * Give low priority threads a better chance to run.
2527 		 */
2528 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2529 			sched_prio(td, PUSER);
2530 
2531 		wakeup_swapper = sig_sleepq_abort(td, intrval);
2532 		PROC_SUNLOCK(p);
2533 		if (wakeup_swapper)
2534 			kick_proc0();
2535 		return;
2536 	}
2537 
2538 	/*
2539 	 * Other states do nothing with the signal immediately,
2540 	 * other than kicking ourselves if we are running.
2541 	 * It will either never be noticed, or noticed very soon.
2542 	 */
2543 #ifdef SMP
2544 	if (TD_IS_RUNNING(td) && td != curthread)
2545 		forward_signal(td);
2546 #endif
2547 
2548 out:
2549 	PROC_SUNLOCK(p);
2550 	thread_unlock(td);
2551 }
2552 
2553 static void
2554 ptrace_coredump(struct thread *td)
2555 {
2556 	struct proc *p;
2557 	struct thr_coredump_req *tcq;
2558 	void *rl_cookie;
2559 
2560 	MPASS(td == curthread);
2561 	p = td->td_proc;
2562 	PROC_LOCK_ASSERT(p, MA_OWNED);
2563 	if ((td->td_dbgflags & TDB_COREDUMPRQ) == 0)
2564 		return;
2565 	KASSERT((p->p_flag & P_STOPPED_TRACE) != 0, ("not stopped"));
2566 
2567 	tcq = td->td_coredump;
2568 	KASSERT(tcq != NULL, ("td_coredump is NULL"));
2569 
2570 	if (p->p_sysent->sv_coredump == NULL) {
2571 		tcq->tc_error = ENOSYS;
2572 		goto wake;
2573 	}
2574 
2575 	PROC_UNLOCK(p);
2576 	rl_cookie = vn_rangelock_wlock(tcq->tc_vp, 0, OFF_MAX);
2577 
2578 	tcq->tc_error = p->p_sysent->sv_coredump(td, tcq->tc_vp,
2579 	    tcq->tc_limit, tcq->tc_flags);
2580 
2581 	vn_rangelock_unlock(tcq->tc_vp, rl_cookie);
2582 	PROC_LOCK(p);
2583 wake:
2584 	td->td_dbgflags &= ~TDB_COREDUMPRQ;
2585 	td->td_coredump = NULL;
2586 	wakeup(p);
2587 }
2588 
2589 static int
2590 sig_suspend_threads(struct thread *td, struct proc *p, int sending)
2591 {
2592 	struct thread *td2;
2593 	int wakeup_swapper;
2594 
2595 	PROC_LOCK_ASSERT(p, MA_OWNED);
2596 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2597 	MPASS(sending || td == curthread);
2598 
2599 	wakeup_swapper = 0;
2600 	FOREACH_THREAD_IN_PROC(p, td2) {
2601 		thread_lock(td2);
2602 		td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
2603 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2604 		    (td2->td_flags & TDF_SINTR)) {
2605 			if (td2->td_flags & TDF_SBDRY) {
2606 				/*
2607 				 * Once a thread is asleep with
2608 				 * TDF_SBDRY and without TDF_SERESTART
2609 				 * or TDF_SEINTR set, it should never
2610 				 * become suspended due to this check.
2611 				 */
2612 				KASSERT(!TD_IS_SUSPENDED(td2),
2613 				    ("thread with deferred stops suspended"));
2614 				if (TD_SBDRY_INTR(td2)) {
2615 					wakeup_swapper |= sleepq_abort(td2,
2616 					    TD_SBDRY_ERRNO(td2));
2617 					continue;
2618 				}
2619 			} else if (!TD_IS_SUSPENDED(td2))
2620 				thread_suspend_one(td2);
2621 		} else if (!TD_IS_SUSPENDED(td2)) {
2622 			if (sending || td != td2)
2623 				td2->td_flags |= TDF_ASTPENDING;
2624 #ifdef SMP
2625 			if (TD_IS_RUNNING(td2) && td2 != td)
2626 				forward_signal(td2);
2627 #endif
2628 		}
2629 		thread_unlock(td2);
2630 	}
2631 	return (wakeup_swapper);
2632 }
2633 
2634 /*
2635  * Stop the process for an event deemed interesting to the debugger. If si is
2636  * non-NULL, this is a signal exchange; the new signal requested by the
2637  * debugger will be returned for handling. If si is NULL, this is some other
2638  * type of interesting event. The debugger may request a signal be delivered in
2639  * that case as well, however it will be deferred until it can be handled.
2640  */
2641 int
2642 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2643 {
2644 	struct proc *p = td->td_proc;
2645 	struct thread *td2;
2646 	ksiginfo_t ksi;
2647 
2648 	PROC_LOCK_ASSERT(p, MA_OWNED);
2649 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2650 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2651 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2652 
2653 	td->td_xsig = sig;
2654 
2655 	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2656 		td->td_dbgflags |= TDB_XSIG;
2657 		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2658 		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2659 		PROC_SLOCK(p);
2660 		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2661 			if (P_KILLED(p)) {
2662 				/*
2663 				 * Ensure that, if we've been PT_KILLed, the
2664 				 * exit status reflects that. Another thread
2665 				 * may also be in ptracestop(), having just
2666 				 * received the SIGKILL, but this thread was
2667 				 * unsuspended first.
2668 				 */
2669 				td->td_dbgflags &= ~TDB_XSIG;
2670 				td->td_xsig = SIGKILL;
2671 				p->p_ptevents = 0;
2672 				break;
2673 			}
2674 			if (p->p_flag & P_SINGLE_EXIT &&
2675 			    !(td->td_dbgflags & TDB_EXIT)) {
2676 				/*
2677 				 * Ignore ptrace stops except for thread exit
2678 				 * events when the process exits.
2679 				 */
2680 				td->td_dbgflags &= ~TDB_XSIG;
2681 				PROC_SUNLOCK(p);
2682 				return (0);
2683 			}
2684 
2685 			/*
2686 			 * Make wait(2) work.  Ensure that right after the
2687 			 * attach, the thread which was decided to become the
2688 			 * leader of attach gets reported to the waiter.
2689 			 * Otherwise, just avoid overwriting another thread's
2690 			 * assignment to p_xthread.  If another thread has
2691 			 * already set p_xthread, the current thread will get
2692 			 * a chance to report itself upon the next iteration.
2693 			 */
2694 			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2695 			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2696 			    p->p_xthread == NULL)) {
2697 				p->p_xsig = sig;
2698 				p->p_xthread = td;
2699 
2700 				/*
2701 				 * If we are on sleepqueue already,
2702 				 * let sleepqueue code decide if it
2703 				 * needs to go sleep after attach.
2704 				 */
2705 				if (td->td_wchan == NULL)
2706 					td->td_dbgflags &= ~TDB_FSTP;
2707 
2708 				p->p_flag2 &= ~P2_PTRACE_FSTP;
2709 				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2710 				sig_suspend_threads(td, p, 0);
2711 			}
2712 			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2713 				td->td_dbgflags &= ~TDB_STOPATFORK;
2714 			}
2715 stopme:
2716 			td->td_dbgflags |= TDB_SSWITCH;
2717 			thread_suspend_switch(td, p);
2718 			td->td_dbgflags &= ~TDB_SSWITCH;
2719 			if ((td->td_dbgflags & TDB_COREDUMPRQ) != 0) {
2720 				PROC_SUNLOCK(p);
2721 				ptrace_coredump(td);
2722 				PROC_SLOCK(p);
2723 				goto stopme;
2724 			}
2725 			if (p->p_xthread == td)
2726 				p->p_xthread = NULL;
2727 			if (!(p->p_flag & P_TRACED))
2728 				break;
2729 			if (td->td_dbgflags & TDB_SUSPEND) {
2730 				if (p->p_flag & P_SINGLE_EXIT)
2731 					break;
2732 				goto stopme;
2733 			}
2734 		}
2735 		PROC_SUNLOCK(p);
2736 	}
2737 
2738 	if (si != NULL && sig == td->td_xsig) {
2739 		/* Parent wants us to take the original signal unchanged. */
2740 		si->ksi_flags |= KSI_HEAD;
2741 		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2742 			si->ksi_signo = 0;
2743 	} else if (td->td_xsig != 0) {
2744 		/*
2745 		 * If parent wants us to take a new signal, then it will leave
2746 		 * it in td->td_xsig; otherwise we just look for signals again.
2747 		 */
2748 		ksiginfo_init(&ksi);
2749 		ksi.ksi_signo = td->td_xsig;
2750 		ksi.ksi_flags |= KSI_PTRACE;
2751 		td2 = sigtd(p, td->td_xsig, false);
2752 		tdsendsignal(p, td2, td->td_xsig, &ksi);
2753 		if (td != td2)
2754 			return (0);
2755 	}
2756 
2757 	return (td->td_xsig);
2758 }
2759 
2760 static void
2761 reschedule_signals(struct proc *p, sigset_t block, int flags)
2762 {
2763 	struct sigacts *ps;
2764 	struct thread *td;
2765 	int sig;
2766 	bool fastblk, pslocked;
2767 
2768 	PROC_LOCK_ASSERT(p, MA_OWNED);
2769 	ps = p->p_sigacts;
2770 	pslocked = (flags & SIGPROCMASK_PS_LOCKED) != 0;
2771 	mtx_assert(&ps->ps_mtx, pslocked ? MA_OWNED : MA_NOTOWNED);
2772 	if (SIGISEMPTY(p->p_siglist))
2773 		return;
2774 	SIGSETAND(block, p->p_siglist);
2775 	fastblk = (flags & SIGPROCMASK_FASTBLK) != 0;
2776 	SIG_FOREACH(sig, &block) {
2777 		td = sigtd(p, sig, fastblk);
2778 
2779 		/*
2780 		 * If sigtd() selected us despite sigfastblock is
2781 		 * blocking, do not activate AST or wake us, to avoid
2782 		 * loop in AST handler.
2783 		 */
2784 		if (fastblk && td == curthread)
2785 			continue;
2786 
2787 		signotify(td);
2788 		if (!pslocked)
2789 			mtx_lock(&ps->ps_mtx);
2790 		if (p->p_flag & P_TRACED ||
2791 		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2792 		    !SIGISMEMBER(td->td_sigmask, sig))) {
2793 			tdsigwakeup(td, sig, SIG_CATCH,
2794 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2795 			    ERESTART));
2796 		}
2797 		if (!pslocked)
2798 			mtx_unlock(&ps->ps_mtx);
2799 	}
2800 }
2801 
2802 void
2803 tdsigcleanup(struct thread *td)
2804 {
2805 	struct proc *p;
2806 	sigset_t unblocked;
2807 
2808 	p = td->td_proc;
2809 	PROC_LOCK_ASSERT(p, MA_OWNED);
2810 
2811 	sigqueue_flush(&td->td_sigqueue);
2812 	if (p->p_numthreads == 1)
2813 		return;
2814 
2815 	/*
2816 	 * Since we cannot handle signals, notify signal post code
2817 	 * about this by filling the sigmask.
2818 	 *
2819 	 * Also, if needed, wake up thread(s) that do not block the
2820 	 * same signals as the exiting thread, since the thread might
2821 	 * have been selected for delivery and woken up.
2822 	 */
2823 	SIGFILLSET(unblocked);
2824 	SIGSETNAND(unblocked, td->td_sigmask);
2825 	SIGFILLSET(td->td_sigmask);
2826 	reschedule_signals(p, unblocked, 0);
2827 
2828 }
2829 
2830 static int
2831 sigdeferstop_curr_flags(int cflags)
2832 {
2833 
2834 	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
2835 	    (cflags & TDF_SBDRY) != 0);
2836 	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
2837 }
2838 
2839 /*
2840  * Defer the delivery of SIGSTOP for the current thread, according to
2841  * the requested mode.  Returns previous flags, which must be restored
2842  * by sigallowstop().
2843  *
2844  * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
2845  * cleared by the current thread, which allow the lock-less read-only
2846  * accesses below.
2847  */
2848 int
2849 sigdeferstop_impl(int mode)
2850 {
2851 	struct thread *td;
2852 	int cflags, nflags;
2853 
2854 	td = curthread;
2855 	cflags = sigdeferstop_curr_flags(td->td_flags);
2856 	switch (mode) {
2857 	case SIGDEFERSTOP_NOP:
2858 		nflags = cflags;
2859 		break;
2860 	case SIGDEFERSTOP_OFF:
2861 		nflags = 0;
2862 		break;
2863 	case SIGDEFERSTOP_SILENT:
2864 		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
2865 		break;
2866 	case SIGDEFERSTOP_EINTR:
2867 		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
2868 		break;
2869 	case SIGDEFERSTOP_ERESTART:
2870 		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
2871 		break;
2872 	default:
2873 		panic("sigdeferstop: invalid mode %x", mode);
2874 		break;
2875 	}
2876 	if (cflags == nflags)
2877 		return (SIGDEFERSTOP_VAL_NCHG);
2878 	thread_lock(td);
2879 	td->td_flags = (td->td_flags & ~cflags) | nflags;
2880 	thread_unlock(td);
2881 	return (cflags);
2882 }
2883 
2884 /*
2885  * Restores the STOP handling mode, typically permitting the delivery
2886  * of SIGSTOP for the current thread.  This does not immediately
2887  * suspend if a stop was posted.  Instead, the thread will suspend
2888  * either via ast() or a subsequent interruptible sleep.
2889  */
2890 void
2891 sigallowstop_impl(int prev)
2892 {
2893 	struct thread *td;
2894 	int cflags;
2895 
2896 	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
2897 	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
2898 	    ("sigallowstop: incorrect previous mode %x", prev));
2899 	td = curthread;
2900 	cflags = sigdeferstop_curr_flags(td->td_flags);
2901 	if (cflags != prev) {
2902 		thread_lock(td);
2903 		td->td_flags = (td->td_flags & ~cflags) | prev;
2904 		thread_unlock(td);
2905 	}
2906 }
2907 
2908 enum sigstatus {
2909 	SIGSTATUS_HANDLE,
2910 	SIGSTATUS_HANDLED,
2911 	SIGSTATUS_IGNORE,
2912 	SIGSTATUS_SBDRY_STOP,
2913 };
2914 
2915 /*
2916  * The thread has signal "sig" pending.  Figure out what to do with it:
2917  *
2918  * _HANDLE     -> the caller should handle the signal
2919  * _HANDLED    -> handled internally, reload pending signal set
2920  * _IGNORE     -> ignored, remove from the set of pending signals and try the
2921  *                next pending signal
2922  * _SBDRY_STOP -> the signal should stop the thread but this is not
2923  *                permitted in the current context
2924  */
2925 static enum sigstatus
2926 sigprocess(struct thread *td, int sig)
2927 {
2928 	struct proc *p;
2929 	struct sigacts *ps;
2930 	struct sigqueue *queue;
2931 	ksiginfo_t ksi;
2932 	int prop;
2933 
2934 	KASSERT(_SIG_VALID(sig), ("%s: invalid signal %d", __func__, sig));
2935 
2936 	p = td->td_proc;
2937 	ps = p->p_sigacts;
2938 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2939 	PROC_LOCK_ASSERT(p, MA_OWNED);
2940 
2941 	/*
2942 	 * We should allow pending but ignored signals below
2943 	 * only if there is sigwait() active, or P_TRACED was
2944 	 * on when they were posted.
2945 	 */
2946 	if (SIGISMEMBER(ps->ps_sigignore, sig) &&
2947 	    (p->p_flag & P_TRACED) == 0 &&
2948 	    (td->td_flags & TDF_SIGWAIT) == 0) {
2949 		return (SIGSTATUS_IGNORE);
2950 	}
2951 
2952 	if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
2953 		/*
2954 		 * If traced, always stop.
2955 		 * Remove old signal from queue before the stop.
2956 		 * XXX shrug off debugger, it causes siginfo to
2957 		 * be thrown away.
2958 		 */
2959 		queue = &td->td_sigqueue;
2960 		ksiginfo_init(&ksi);
2961 		if (sigqueue_get(queue, sig, &ksi) == 0) {
2962 			queue = &p->p_sigqueue;
2963 			sigqueue_get(queue, sig, &ksi);
2964 		}
2965 		td->td_si = ksi.ksi_info;
2966 
2967 		mtx_unlock(&ps->ps_mtx);
2968 		sig = ptracestop(td, sig, &ksi);
2969 		mtx_lock(&ps->ps_mtx);
2970 
2971 		td->td_si.si_signo = 0;
2972 
2973 		/*
2974 		 * Keep looking if the debugger discarded or
2975 		 * replaced the signal.
2976 		 */
2977 		if (sig == 0)
2978 			return (SIGSTATUS_HANDLED);
2979 
2980 		/*
2981 		 * If the signal became masked, re-queue it.
2982 		 */
2983 		if (SIGISMEMBER(td->td_sigmask, sig)) {
2984 			ksi.ksi_flags |= KSI_HEAD;
2985 			sigqueue_add(&p->p_sigqueue, sig, &ksi);
2986 			return (SIGSTATUS_HANDLED);
2987 		}
2988 
2989 		/*
2990 		 * If the traced bit got turned off, requeue the signal and
2991 		 * reload the set of pending signals.  This ensures that p_sig*
2992 		 * and p_sigact are consistent.
2993 		 */
2994 		if ((p->p_flag & P_TRACED) == 0) {
2995 			if ((ksi.ksi_flags & KSI_PTRACE) == 0) {
2996 				ksi.ksi_flags |= KSI_HEAD;
2997 				sigqueue_add(queue, sig, &ksi);
2998 			}
2999 			return (SIGSTATUS_HANDLED);
3000 		}
3001 	}
3002 
3003 	/*
3004 	 * Decide whether the signal should be returned.
3005 	 * Return the signal's number, or fall through
3006 	 * to clear it from the pending mask.
3007 	 */
3008 	switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
3009 	case (intptr_t)SIG_DFL:
3010 		/*
3011 		 * Don't take default actions on system processes.
3012 		 */
3013 		if (p->p_pid <= 1) {
3014 #ifdef DIAGNOSTIC
3015 			/*
3016 			 * Are you sure you want to ignore SIGSEGV
3017 			 * in init? XXX
3018 			 */
3019 			printf("Process (pid %lu) got signal %d\n",
3020 				(u_long)p->p_pid, sig);
3021 #endif
3022 			return (SIGSTATUS_IGNORE);
3023 		}
3024 
3025 		/*
3026 		 * If there is a pending stop signal to process with
3027 		 * default action, stop here, then clear the signal.
3028 		 * Traced or exiting processes should ignore stops.
3029 		 * Additionally, a member of an orphaned process group
3030 		 * should ignore tty stops.
3031 		 */
3032 		prop = sigprop(sig);
3033 		if (prop & SIGPROP_STOP) {
3034 			mtx_unlock(&ps->ps_mtx);
3035 			if ((p->p_flag & (P_TRACED | P_WEXIT |
3036 			    P_SINGLE_EXIT)) != 0 || ((p->p_pgrp->
3037 			    pg_flags & PGRP_ORPHANED) != 0 &&
3038 			    (prop & SIGPROP_TTYSTOP) != 0)) {
3039 				mtx_lock(&ps->ps_mtx);
3040 				return (SIGSTATUS_IGNORE);
3041 			}
3042 			if (TD_SBDRY_INTR(td)) {
3043 				KASSERT((td->td_flags & TDF_SBDRY) != 0,
3044 				    ("lost TDF_SBDRY"));
3045 				mtx_lock(&ps->ps_mtx);
3046 				return (SIGSTATUS_SBDRY_STOP);
3047 			}
3048 			WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
3049 			    &p->p_mtx.lock_object, "Catching SIGSTOP");
3050 			sigqueue_delete(&td->td_sigqueue, sig);
3051 			sigqueue_delete(&p->p_sigqueue, sig);
3052 			p->p_flag |= P_STOPPED_SIG;
3053 			p->p_xsig = sig;
3054 			PROC_SLOCK(p);
3055 			sig_suspend_threads(td, p, 0);
3056 			thread_suspend_switch(td, p);
3057 			PROC_SUNLOCK(p);
3058 			mtx_lock(&ps->ps_mtx);
3059 			return (SIGSTATUS_HANDLED);
3060 		} else if ((prop & SIGPROP_IGNORE) != 0 &&
3061 		    (td->td_flags & TDF_SIGWAIT) == 0) {
3062 			/*
3063 			 * Default action is to ignore; drop it if
3064 			 * not in kern_sigtimedwait().
3065 			 */
3066 			return (SIGSTATUS_IGNORE);
3067 		} else {
3068 			return (SIGSTATUS_HANDLE);
3069 		}
3070 
3071 	case (intptr_t)SIG_IGN:
3072 		if ((td->td_flags & TDF_SIGWAIT) == 0)
3073 			return (SIGSTATUS_IGNORE);
3074 		else
3075 			return (SIGSTATUS_HANDLE);
3076 
3077 	default:
3078 		/*
3079 		 * This signal has an action, let postsig() process it.
3080 		 */
3081 		return (SIGSTATUS_HANDLE);
3082 	}
3083 }
3084 
3085 /*
3086  * If the current process has received a signal (should be caught or cause
3087  * termination, should interrupt current syscall), return the signal number.
3088  * Stop signals with default action are processed immediately, then cleared;
3089  * they aren't returned.  This is checked after each entry to the system for
3090  * a syscall or trap (though this can usually be done without calling
3091  * issignal by checking the pending signal masks in cursig.) The normal call
3092  * sequence is
3093  *
3094  *	while (sig = cursig(curthread))
3095  *		postsig(sig);
3096  */
3097 static int
3098 issignal(struct thread *td)
3099 {
3100 	struct proc *p;
3101 	sigset_t sigpending;
3102 	int sig;
3103 
3104 	p = td->td_proc;
3105 	PROC_LOCK_ASSERT(p, MA_OWNED);
3106 
3107 	for (;;) {
3108 		sigpending = td->td_sigqueue.sq_signals;
3109 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
3110 		SIGSETNAND(sigpending, td->td_sigmask);
3111 
3112 		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
3113 		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
3114 			SIG_STOPSIGMASK(sigpending);
3115 		if (SIGISEMPTY(sigpending))	/* no signal to send */
3116 			return (0);
3117 
3118 		/*
3119 		 * Do fast sigblock if requested by usermode.  Since
3120 		 * we do know that there was a signal pending at this
3121 		 * point, set the FAST_SIGBLOCK_PEND as indicator for
3122 		 * usermode to perform a dummy call to
3123 		 * FAST_SIGBLOCK_UNBLOCK, which causes immediate
3124 		 * delivery of postponed pending signal.
3125 		 */
3126 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
3127 			if (td->td_sigblock_val != 0)
3128 				SIGSETNAND(sigpending, fastblock_mask);
3129 			if (SIGISEMPTY(sigpending)) {
3130 				td->td_pflags |= TDP_SIGFASTPENDING;
3131 				return (0);
3132 			}
3133 		}
3134 
3135 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
3136 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
3137 		    SIGISMEMBER(sigpending, SIGSTOP)) {
3138 			/*
3139 			 * If debugger just attached, always consume
3140 			 * SIGSTOP from ptrace(PT_ATTACH) first, to
3141 			 * execute the debugger attach ritual in
3142 			 * order.
3143 			 */
3144 			td->td_dbgflags |= TDB_FSTP;
3145 			SIGEMPTYSET(sigpending);
3146 			SIGADDSET(sigpending, SIGSTOP);
3147 		}
3148 
3149 		SIG_FOREACH(sig, &sigpending) {
3150 			switch (sigprocess(td, sig)) {
3151 			case SIGSTATUS_HANDLE:
3152 				return (sig);
3153 			case SIGSTATUS_HANDLED:
3154 				goto next;
3155 			case SIGSTATUS_IGNORE:
3156 				sigqueue_delete(&td->td_sigqueue, sig);
3157 				sigqueue_delete(&p->p_sigqueue, sig);
3158 				break;
3159 			case SIGSTATUS_SBDRY_STOP:
3160 				return (-1);
3161 			}
3162 		}
3163 next:;
3164 	}
3165 }
3166 
3167 void
3168 thread_stopped(struct proc *p)
3169 {
3170 	int n;
3171 
3172 	PROC_LOCK_ASSERT(p, MA_OWNED);
3173 	PROC_SLOCK_ASSERT(p, MA_OWNED);
3174 	n = p->p_suspcount;
3175 	if (p == curproc)
3176 		n++;
3177 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3178 		PROC_SUNLOCK(p);
3179 		p->p_flag &= ~P_WAITED;
3180 		PROC_LOCK(p->p_pptr);
3181 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
3182 			CLD_TRAPPED : CLD_STOPPED);
3183 		PROC_UNLOCK(p->p_pptr);
3184 		PROC_SLOCK(p);
3185 	}
3186 }
3187 
3188 /*
3189  * Take the action for the specified signal
3190  * from the current set of pending signals.
3191  */
3192 int
3193 postsig(int sig)
3194 {
3195 	struct thread *td;
3196 	struct proc *p;
3197 	struct sigacts *ps;
3198 	sig_t action;
3199 	ksiginfo_t ksi;
3200 	sigset_t returnmask;
3201 
3202 	KASSERT(sig != 0, ("postsig"));
3203 
3204 	td = curthread;
3205 	p = td->td_proc;
3206 	PROC_LOCK_ASSERT(p, MA_OWNED);
3207 	ps = p->p_sigacts;
3208 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3209 	ksiginfo_init(&ksi);
3210 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3211 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3212 		return (0);
3213 	ksi.ksi_signo = sig;
3214 	if (ksi.ksi_code == SI_TIMER)
3215 		itimer_accept(p, ksi.ksi_timerid, &ksi);
3216 	action = ps->ps_sigact[_SIG_IDX(sig)];
3217 #ifdef KTRACE
3218 	if (KTRPOINT(td, KTR_PSIG))
3219 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3220 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3221 #endif
3222 
3223 	if (action == SIG_DFL) {
3224 		/*
3225 		 * Default action, where the default is to kill
3226 		 * the process.  (Other cases were ignored above.)
3227 		 */
3228 		mtx_unlock(&ps->ps_mtx);
3229 		proc_td_siginfo_capture(td, &ksi.ksi_info);
3230 		sigexit(td, sig);
3231 		/* NOTREACHED */
3232 	} else {
3233 		/*
3234 		 * If we get here, the signal must be caught.
3235 		 */
3236 		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3237 		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3238 		    ("postsig action: blocked sig %d", sig));
3239 
3240 		/*
3241 		 * Set the new mask value and also defer further
3242 		 * occurrences of this signal.
3243 		 *
3244 		 * Special case: user has done a sigsuspend.  Here the
3245 		 * current mask is not of interest, but rather the
3246 		 * mask from before the sigsuspend is what we want
3247 		 * restored after the signal processing is completed.
3248 		 */
3249 		if (td->td_pflags & TDP_OLDMASK) {
3250 			returnmask = td->td_oldsigmask;
3251 			td->td_pflags &= ~TDP_OLDMASK;
3252 		} else
3253 			returnmask = td->td_sigmask;
3254 
3255 		if (p->p_sig == sig) {
3256 			p->p_sig = 0;
3257 		}
3258 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3259 		postsig_done(sig, td, ps);
3260 	}
3261 	return (1);
3262 }
3263 
3264 int
3265 sig_ast_checksusp(struct thread *td)
3266 {
3267 	struct proc *p __diagused;
3268 	int ret;
3269 
3270 	p = td->td_proc;
3271 	PROC_LOCK_ASSERT(p, MA_OWNED);
3272 
3273 	if ((td->td_flags & TDF_NEEDSUSPCHK) == 0)
3274 		return (0);
3275 
3276 	ret = thread_suspend_check(1);
3277 	MPASS(ret == 0 || ret == EINTR || ret == ERESTART);
3278 	return (ret);
3279 }
3280 
3281 int
3282 sig_ast_needsigchk(struct thread *td)
3283 {
3284 	struct proc *p;
3285 	struct sigacts *ps;
3286 	int ret, sig;
3287 
3288 	p = td->td_proc;
3289 	PROC_LOCK_ASSERT(p, MA_OWNED);
3290 
3291 	if ((td->td_flags & TDF_NEEDSIGCHK) == 0)
3292 		return (0);
3293 
3294 	ps = p->p_sigacts;
3295 	mtx_lock(&ps->ps_mtx);
3296 	sig = cursig(td);
3297 	if (sig == -1) {
3298 		mtx_unlock(&ps->ps_mtx);
3299 		KASSERT((td->td_flags & TDF_SBDRY) != 0, ("lost TDF_SBDRY"));
3300 		KASSERT(TD_SBDRY_INTR(td),
3301 		    ("lost TDF_SERESTART of TDF_SEINTR"));
3302 		KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
3303 		    (TDF_SEINTR | TDF_SERESTART),
3304 		    ("both TDF_SEINTR and TDF_SERESTART"));
3305 		ret = TD_SBDRY_ERRNO(td);
3306 	} else if (sig != 0) {
3307 		ret = SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR : ERESTART;
3308 		mtx_unlock(&ps->ps_mtx);
3309 	} else {
3310 		mtx_unlock(&ps->ps_mtx);
3311 		ret = 0;
3312 	}
3313 
3314 	/*
3315 	 * Do not go into sleep if this thread was the ptrace(2)
3316 	 * attach leader.  cursig() consumed SIGSTOP from PT_ATTACH,
3317 	 * but we usually act on the signal by interrupting sleep, and
3318 	 * should do that here as well.
3319 	 */
3320 	if ((td->td_dbgflags & TDB_FSTP) != 0) {
3321 		if (ret == 0)
3322 			ret = EINTR;
3323 		td->td_dbgflags &= ~TDB_FSTP;
3324 	}
3325 
3326 	return (ret);
3327 }
3328 
3329 int
3330 sig_intr(void)
3331 {
3332 	struct thread *td;
3333 	struct proc *p;
3334 	int ret;
3335 
3336 	td = curthread;
3337 	if ((td->td_flags & (TDF_NEEDSIGCHK | TDF_NEEDSUSPCHK)) == 0)
3338 		return (0);
3339 
3340 	p = td->td_proc;
3341 
3342 	PROC_LOCK(p);
3343 	ret = sig_ast_checksusp(td);
3344 	if (ret == 0)
3345 		ret = sig_ast_needsigchk(td);
3346 	PROC_UNLOCK(p);
3347 	return (ret);
3348 }
3349 
3350 bool
3351 curproc_sigkilled(void)
3352 {
3353 	struct thread *td;
3354 	struct proc *p;
3355 	struct sigacts *ps;
3356 	bool res;
3357 
3358 	td = curthread;
3359 	if ((td->td_flags & TDF_NEEDSIGCHK) == 0)
3360 		return (false);
3361 
3362 	p = td->td_proc;
3363 	PROC_LOCK(p);
3364 	ps = p->p_sigacts;
3365 	mtx_lock(&ps->ps_mtx);
3366 	res = SIGISMEMBER(td->td_sigqueue.sq_signals, SIGKILL) ||
3367 	    SIGISMEMBER(p->p_sigqueue.sq_signals, SIGKILL);
3368 	mtx_unlock(&ps->ps_mtx);
3369 	PROC_UNLOCK(p);
3370 	return (res);
3371 }
3372 
3373 void
3374 proc_wkilled(struct proc *p)
3375 {
3376 
3377 	PROC_LOCK_ASSERT(p, MA_OWNED);
3378 	if ((p->p_flag & P_WKILLED) == 0) {
3379 		p->p_flag |= P_WKILLED;
3380 		/*
3381 		 * Notify swapper that there is a process to swap in.
3382 		 * The notification is racy, at worst it would take 10
3383 		 * seconds for the swapper process to notice.
3384 		 */
3385 		if ((p->p_flag & (P_INMEM | P_SWAPPINGIN)) == 0)
3386 			wakeup(&proc0);
3387 	}
3388 }
3389 
3390 /*
3391  * Kill the current process for stated reason.
3392  */
3393 void
3394 killproc(struct proc *p, const char *why)
3395 {
3396 
3397 	PROC_LOCK_ASSERT(p, MA_OWNED);
3398 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3399 	    p->p_comm);
3400 	log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3401 	    p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3402 	    p->p_ucred->cr_uid, why);
3403 	proc_wkilled(p);
3404 	kern_psignal(p, SIGKILL);
3405 }
3406 
3407 /*
3408  * Force the current process to exit with the specified signal, dumping core
3409  * if appropriate.  We bypass the normal tests for masked and caught signals,
3410  * allowing unrecoverable failures to terminate the process without changing
3411  * signal state.  Mark the accounting record with the signal termination.
3412  * If dumping core, save the signal number for the debugger.  Calls exit and
3413  * does not return.
3414  */
3415 void
3416 sigexit(struct thread *td, int sig)
3417 {
3418 	struct proc *p = td->td_proc;
3419 
3420 	PROC_LOCK_ASSERT(p, MA_OWNED);
3421 	p->p_acflag |= AXSIG;
3422 	/*
3423 	 * We must be single-threading to generate a core dump.  This
3424 	 * ensures that the registers in the core file are up-to-date.
3425 	 * Also, the ELF dump handler assumes that the thread list doesn't
3426 	 * change out from under it.
3427 	 *
3428 	 * XXX If another thread attempts to single-thread before us
3429 	 *     (e.g. via fork()), we won't get a dump at all.
3430 	 */
3431 	if ((sigprop(sig) & SIGPROP_CORE) &&
3432 	    thread_single(p, SINGLE_NO_EXIT) == 0) {
3433 		p->p_sig = sig;
3434 		/*
3435 		 * Log signals which would cause core dumps
3436 		 * (Log as LOG_INFO to appease those who don't want
3437 		 * these messages.)
3438 		 * XXX : Todo, as well as euid, write out ruid too
3439 		 * Note that coredump() drops proc lock.
3440 		 */
3441 		if (coredump(td) == 0)
3442 			sig |= WCOREFLAG;
3443 		if (kern_logsigexit)
3444 			log(LOG_INFO,
3445 			    "pid %d (%s), jid %d, uid %d: exited on "
3446 			    "signal %d%s\n", p->p_pid, p->p_comm,
3447 			    p->p_ucred->cr_prison->pr_id,
3448 			    td->td_ucred->cr_uid,
3449 			    sig &~ WCOREFLAG,
3450 			    sig & WCOREFLAG ? " (core dumped)" : "");
3451 	} else
3452 		PROC_UNLOCK(p);
3453 	exit1(td, 0, sig);
3454 	/* NOTREACHED */
3455 }
3456 
3457 /*
3458  * Send queued SIGCHLD to parent when child process's state
3459  * is changed.
3460  */
3461 static void
3462 sigparent(struct proc *p, int reason, int status)
3463 {
3464 	PROC_LOCK_ASSERT(p, MA_OWNED);
3465 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3466 
3467 	if (p->p_ksi != NULL) {
3468 		p->p_ksi->ksi_signo  = SIGCHLD;
3469 		p->p_ksi->ksi_code   = reason;
3470 		p->p_ksi->ksi_status = status;
3471 		p->p_ksi->ksi_pid    = p->p_pid;
3472 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3473 		if (KSI_ONQ(p->p_ksi))
3474 			return;
3475 	}
3476 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3477 }
3478 
3479 static void
3480 childproc_jobstate(struct proc *p, int reason, int sig)
3481 {
3482 	struct sigacts *ps;
3483 
3484 	PROC_LOCK_ASSERT(p, MA_OWNED);
3485 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3486 
3487 	/*
3488 	 * Wake up parent sleeping in kern_wait(), also send
3489 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3490 	 * that parent will awake, because parent may masked
3491 	 * the signal.
3492 	 */
3493 	p->p_pptr->p_flag |= P_STATCHILD;
3494 	wakeup(p->p_pptr);
3495 
3496 	ps = p->p_pptr->p_sigacts;
3497 	mtx_lock(&ps->ps_mtx);
3498 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3499 		mtx_unlock(&ps->ps_mtx);
3500 		sigparent(p, reason, sig);
3501 	} else
3502 		mtx_unlock(&ps->ps_mtx);
3503 }
3504 
3505 void
3506 childproc_stopped(struct proc *p, int reason)
3507 {
3508 
3509 	childproc_jobstate(p, reason, p->p_xsig);
3510 }
3511 
3512 void
3513 childproc_continued(struct proc *p)
3514 {
3515 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3516 }
3517 
3518 void
3519 childproc_exited(struct proc *p)
3520 {
3521 	int reason, status;
3522 
3523 	if (WCOREDUMP(p->p_xsig)) {
3524 		reason = CLD_DUMPED;
3525 		status = WTERMSIG(p->p_xsig);
3526 	} else if (WIFSIGNALED(p->p_xsig)) {
3527 		reason = CLD_KILLED;
3528 		status = WTERMSIG(p->p_xsig);
3529 	} else {
3530 		reason = CLD_EXITED;
3531 		status = p->p_xexit;
3532 	}
3533 	/*
3534 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3535 	 * done in exit1().
3536 	 */
3537 	sigparent(p, reason, status);
3538 }
3539 
3540 #define	MAX_NUM_CORE_FILES 100000
3541 #ifndef NUM_CORE_FILES
3542 #define	NUM_CORE_FILES 5
3543 #endif
3544 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3545 static int num_cores = NUM_CORE_FILES;
3546 
3547 static int
3548 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3549 {
3550 	int error;
3551 	int new_val;
3552 
3553 	new_val = num_cores;
3554 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3555 	if (error != 0 || req->newptr == NULL)
3556 		return (error);
3557 	if (new_val > MAX_NUM_CORE_FILES)
3558 		new_val = MAX_NUM_CORE_FILES;
3559 	if (new_val < 0)
3560 		new_val = 0;
3561 	num_cores = new_val;
3562 	return (0);
3563 }
3564 SYSCTL_PROC(_debug, OID_AUTO, ncores,
3565     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
3566     sysctl_debug_num_cores_check, "I",
3567     "Maximum number of generated process corefiles while using index format");
3568 
3569 #define	GZIP_SUFFIX	".gz"
3570 #define	ZSTD_SUFFIX	".zst"
3571 
3572 int compress_user_cores = 0;
3573 
3574 static int
3575 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3576 {
3577 	int error, val;
3578 
3579 	val = compress_user_cores;
3580 	error = sysctl_handle_int(oidp, &val, 0, req);
3581 	if (error != 0 || req->newptr == NULL)
3582 		return (error);
3583 	if (val != 0 && !compressor_avail(val))
3584 		return (EINVAL);
3585 	compress_user_cores = val;
3586 	return (error);
3587 }
3588 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores,
3589     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, 0, sizeof(int),
3590     sysctl_compress_user_cores, "I",
3591     "Enable compression of user corefiles ("
3592     __XSTRING(COMPRESS_GZIP) " = gzip, "
3593     __XSTRING(COMPRESS_ZSTD) " = zstd)");
3594 
3595 int compress_user_cores_level = 6;
3596 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3597     &compress_user_cores_level, 0,
3598     "Corefile compression level");
3599 
3600 /*
3601  * Protect the access to corefilename[] by allproc_lock.
3602  */
3603 #define	corefilename_lock	allproc_lock
3604 
3605 static char corefilename[MAXPATHLEN] = {"%N.core"};
3606 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3607 
3608 static int
3609 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3610 {
3611 	int error;
3612 
3613 	sx_xlock(&corefilename_lock);
3614 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3615 	    req);
3616 	sx_xunlock(&corefilename_lock);
3617 
3618 	return (error);
3619 }
3620 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3621     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3622     "Process corefile name format string");
3623 
3624 static void
3625 vnode_close_locked(struct thread *td, struct vnode *vp)
3626 {
3627 
3628 	VOP_UNLOCK(vp);
3629 	vn_close(vp, FWRITE, td->td_ucred, td);
3630 }
3631 
3632 /*
3633  * If the core format has a %I in it, then we need to check
3634  * for existing corefiles before defining a name.
3635  * To do this we iterate over 0..ncores to find a
3636  * non-existing core file name to use. If all core files are
3637  * already used we choose the oldest one.
3638  */
3639 static int
3640 corefile_open_last(struct thread *td, char *name, int indexpos,
3641     int indexlen, int ncores, struct vnode **vpp)
3642 {
3643 	struct vnode *oldvp, *nextvp, *vp;
3644 	struct vattr vattr;
3645 	struct nameidata nd;
3646 	int error, i, flags, oflags, cmode;
3647 	char ch;
3648 	struct timespec lasttime;
3649 
3650 	nextvp = oldvp = NULL;
3651 	cmode = S_IRUSR | S_IWUSR;
3652 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3653 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3654 
3655 	for (i = 0; i < ncores; i++) {
3656 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3657 
3658 		ch = name[indexpos + indexlen];
3659 		(void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3660 		    i);
3661 		name[indexpos + indexlen] = ch;
3662 
3663 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
3664 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3665 		    NULL);
3666 		if (error != 0)
3667 			break;
3668 
3669 		vp = nd.ni_vp;
3670 		NDFREE(&nd, NDF_ONLY_PNBUF);
3671 		if ((flags & O_CREAT) == O_CREAT) {
3672 			nextvp = vp;
3673 			break;
3674 		}
3675 
3676 		error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3677 		if (error != 0) {
3678 			vnode_close_locked(td, vp);
3679 			break;
3680 		}
3681 
3682 		if (oldvp == NULL ||
3683 		    lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3684 		    (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3685 		    lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3686 			if (oldvp != NULL)
3687 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3688 			oldvp = vp;
3689 			VOP_UNLOCK(oldvp);
3690 			lasttime = vattr.va_mtime;
3691 		} else {
3692 			vnode_close_locked(td, vp);
3693 		}
3694 	}
3695 
3696 	if (oldvp != NULL) {
3697 		if (nextvp == NULL) {
3698 			if ((td->td_proc->p_flag & P_SUGID) != 0) {
3699 				error = EFAULT;
3700 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3701 			} else {
3702 				nextvp = oldvp;
3703 				error = vn_lock(nextvp, LK_EXCLUSIVE);
3704 				if (error != 0) {
3705 					vn_close(nextvp, FWRITE, td->td_ucred,
3706 					    td);
3707 					nextvp = NULL;
3708 				}
3709 			}
3710 		} else {
3711 			vn_close(oldvp, FWRITE, td->td_ucred, td);
3712 		}
3713 	}
3714 	if (error != 0) {
3715 		if (nextvp != NULL)
3716 			vnode_close_locked(td, oldvp);
3717 	} else {
3718 		*vpp = nextvp;
3719 	}
3720 
3721 	return (error);
3722 }
3723 
3724 /*
3725  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3726  * Expand the name described in corefilename, using name, uid, and pid
3727  * and open/create core file.
3728  * corefilename is a printf-like string, with three format specifiers:
3729  *	%N	name of process ("name")
3730  *	%P	process id (pid)
3731  *	%U	user id (uid)
3732  * For example, "%N.core" is the default; they can be disabled completely
3733  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3734  * This is controlled by the sysctl variable kern.corefile (see above).
3735  */
3736 static int
3737 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3738     int compress, int signum, struct vnode **vpp, char **namep)
3739 {
3740 	struct sbuf sb;
3741 	struct nameidata nd;
3742 	const char *format;
3743 	char *hostname, *name;
3744 	int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
3745 
3746 	hostname = NULL;
3747 	format = corefilename;
3748 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3749 	indexlen = 0;
3750 	indexpos = -1;
3751 	ncores = num_cores;
3752 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3753 	sx_slock(&corefilename_lock);
3754 	for (i = 0; format[i] != '\0'; i++) {
3755 		switch (format[i]) {
3756 		case '%':	/* Format character */
3757 			i++;
3758 			switch (format[i]) {
3759 			case '%':
3760 				sbuf_putc(&sb, '%');
3761 				break;
3762 			case 'H':	/* hostname */
3763 				if (hostname == NULL) {
3764 					hostname = malloc(MAXHOSTNAMELEN,
3765 					    M_TEMP, M_WAITOK);
3766 				}
3767 				getcredhostname(td->td_ucred, hostname,
3768 				    MAXHOSTNAMELEN);
3769 				sbuf_printf(&sb, "%s", hostname);
3770 				break;
3771 			case 'I':	/* autoincrementing index */
3772 				if (indexpos != -1) {
3773 					sbuf_printf(&sb, "%%I");
3774 					break;
3775 				}
3776 
3777 				indexpos = sbuf_len(&sb);
3778 				sbuf_printf(&sb, "%u", ncores - 1);
3779 				indexlen = sbuf_len(&sb) - indexpos;
3780 				break;
3781 			case 'N':	/* process name */
3782 				sbuf_printf(&sb, "%s", comm);
3783 				break;
3784 			case 'P':	/* process id */
3785 				sbuf_printf(&sb, "%u", pid);
3786 				break;
3787 			case 'S':	/* signal number */
3788 				sbuf_printf(&sb, "%i", signum);
3789 				break;
3790 			case 'U':	/* user id */
3791 				sbuf_printf(&sb, "%u", uid);
3792 				break;
3793 			default:
3794 				log(LOG_ERR,
3795 				    "Unknown format character %c in "
3796 				    "corename `%s'\n", format[i], format);
3797 				break;
3798 			}
3799 			break;
3800 		default:
3801 			sbuf_putc(&sb, format[i]);
3802 			break;
3803 		}
3804 	}
3805 	sx_sunlock(&corefilename_lock);
3806 	free(hostname, M_TEMP);
3807 	if (compress == COMPRESS_GZIP)
3808 		sbuf_printf(&sb, GZIP_SUFFIX);
3809 	else if (compress == COMPRESS_ZSTD)
3810 		sbuf_printf(&sb, ZSTD_SUFFIX);
3811 	if (sbuf_error(&sb) != 0) {
3812 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3813 		    "long\n", (long)pid, comm, (u_long)uid);
3814 		sbuf_delete(&sb);
3815 		free(name, M_TEMP);
3816 		return (ENOMEM);
3817 	}
3818 	sbuf_finish(&sb);
3819 	sbuf_delete(&sb);
3820 
3821 	if (indexpos != -1) {
3822 		error = corefile_open_last(td, name, indexpos, indexlen, ncores,
3823 		    vpp);
3824 		if (error != 0) {
3825 			log(LOG_ERR,
3826 			    "pid %d (%s), uid (%u):  Path `%s' failed "
3827 			    "on initial open test, error = %d\n",
3828 			    pid, comm, uid, name, error);
3829 		}
3830 	} else {
3831 		cmode = S_IRUSR | S_IWUSR;
3832 		oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3833 		    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3834 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3835 		if ((td->td_proc->p_flag & P_SUGID) != 0)
3836 			flags |= O_EXCL;
3837 
3838 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
3839 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3840 		    NULL);
3841 		if (error == 0) {
3842 			*vpp = nd.ni_vp;
3843 			NDFREE(&nd, NDF_ONLY_PNBUF);
3844 		}
3845 	}
3846 
3847 	if (error != 0) {
3848 #ifdef AUDIT
3849 		audit_proc_coredump(td, name, error);
3850 #endif
3851 		free(name, M_TEMP);
3852 		return (error);
3853 	}
3854 	*namep = name;
3855 	return (0);
3856 }
3857 
3858 /*
3859  * Dump a process' core.  The main routine does some
3860  * policy checking, and creates the name of the coredump;
3861  * then it passes on a vnode and a size limit to the process-specific
3862  * coredump routine if there is one; if there _is not_ one, it returns
3863  * ENOSYS; otherwise it returns the error from the process-specific routine.
3864  */
3865 
3866 static int
3867 coredump(struct thread *td)
3868 {
3869 	struct proc *p = td->td_proc;
3870 	struct ucred *cred = td->td_ucred;
3871 	struct vnode *vp;
3872 	struct flock lf;
3873 	struct vattr vattr;
3874 	size_t fullpathsize;
3875 	int error, error1, locked;
3876 	char *name;			/* name of corefile */
3877 	void *rl_cookie;
3878 	off_t limit;
3879 	char *fullpath, *freepath = NULL;
3880 	struct sbuf *sb;
3881 
3882 	PROC_LOCK_ASSERT(p, MA_OWNED);
3883 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3884 
3885 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
3886 	    (p->p_flag2 & P2_NOTRACE) != 0) {
3887 		PROC_UNLOCK(p);
3888 		return (EFAULT);
3889 	}
3890 
3891 	/*
3892 	 * Note that the bulk of limit checking is done after
3893 	 * the corefile is created.  The exception is if the limit
3894 	 * for corefiles is 0, in which case we don't bother
3895 	 * creating the corefile at all.  This layout means that
3896 	 * a corefile is truncated instead of not being created,
3897 	 * if it is larger than the limit.
3898 	 */
3899 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
3900 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
3901 		PROC_UNLOCK(p);
3902 		return (EFBIG);
3903 	}
3904 	PROC_UNLOCK(p);
3905 
3906 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
3907 	    compress_user_cores, p->p_sig, &vp, &name);
3908 	if (error != 0)
3909 		return (error);
3910 
3911 	/*
3912 	 * Don't dump to non-regular files or files with links.
3913 	 * Do not dump into system files. Effective user must own the corefile.
3914 	 */
3915 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
3916 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
3917 	    vattr.va_uid != cred->cr_uid) {
3918 		VOP_UNLOCK(vp);
3919 		error = EFAULT;
3920 		goto out;
3921 	}
3922 
3923 	VOP_UNLOCK(vp);
3924 
3925 	/* Postpone other writers, including core dumps of other processes. */
3926 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
3927 
3928 	lf.l_whence = SEEK_SET;
3929 	lf.l_start = 0;
3930 	lf.l_len = 0;
3931 	lf.l_type = F_WRLCK;
3932 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
3933 
3934 	VATTR_NULL(&vattr);
3935 	vattr.va_size = 0;
3936 	if (set_core_nodump_flag)
3937 		vattr.va_flags = UF_NODUMP;
3938 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3939 	VOP_SETATTR(vp, &vattr, cred);
3940 	VOP_UNLOCK(vp);
3941 	PROC_LOCK(p);
3942 	p->p_acflag |= ACORE;
3943 	PROC_UNLOCK(p);
3944 
3945 	if (p->p_sysent->sv_coredump != NULL) {
3946 		error = p->p_sysent->sv_coredump(td, vp, limit, 0);
3947 	} else {
3948 		error = ENOSYS;
3949 	}
3950 
3951 	if (locked) {
3952 		lf.l_type = F_UNLCK;
3953 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3954 	}
3955 	vn_rangelock_unlock(vp, rl_cookie);
3956 
3957 	/*
3958 	 * Notify the userland helper that a process triggered a core dump.
3959 	 * This allows the helper to run an automated debugging session.
3960 	 */
3961 	if (error != 0 || coredump_devctl == 0)
3962 		goto out;
3963 	sb = sbuf_new_auto();
3964 	if (vn_fullpath_global(p->p_textvp, &fullpath, &freepath) != 0)
3965 		goto out2;
3966 	sbuf_printf(sb, "comm=\"");
3967 	devctl_safe_quote_sb(sb, fullpath);
3968 	free(freepath, M_TEMP);
3969 	sbuf_printf(sb, "\" core=\"");
3970 
3971 	/*
3972 	 * We can't lookup core file vp directly. When we're replacing a core, and
3973 	 * other random times, we flush the name cache, so it will fail. Instead,
3974 	 * if the path of the core is relative, add the current dir in front if it.
3975 	 */
3976 	if (name[0] != '/') {
3977 		fullpathsize = MAXPATHLEN;
3978 		freepath = malloc(fullpathsize, M_TEMP, M_WAITOK);
3979 		if (vn_getcwd(freepath, &fullpath, &fullpathsize) != 0) {
3980 			free(freepath, M_TEMP);
3981 			goto out2;
3982 		}
3983 		devctl_safe_quote_sb(sb, fullpath);
3984 		free(freepath, M_TEMP);
3985 		sbuf_putc(sb, '/');
3986 	}
3987 	devctl_safe_quote_sb(sb, name);
3988 	sbuf_printf(sb, "\"");
3989 	if (sbuf_finish(sb) == 0)
3990 		devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
3991 out2:
3992 	sbuf_delete(sb);
3993 out:
3994 	error1 = vn_close(vp, FWRITE, cred, td);
3995 	if (error == 0)
3996 		error = error1;
3997 #ifdef AUDIT
3998 	audit_proc_coredump(td, name, error);
3999 #endif
4000 	free(name, M_TEMP);
4001 	return (error);
4002 }
4003 
4004 /*
4005  * Nonexistent system call-- signal process (may want to handle it).  Flag
4006  * error in case process won't see signal immediately (blocked or ignored).
4007  */
4008 #ifndef _SYS_SYSPROTO_H_
4009 struct nosys_args {
4010 	int	dummy;
4011 };
4012 #endif
4013 /* ARGSUSED */
4014 int
4015 nosys(struct thread *td, struct nosys_args *args)
4016 {
4017 	struct proc *p;
4018 
4019 	p = td->td_proc;
4020 
4021 	PROC_LOCK(p);
4022 	tdsignal(td, SIGSYS);
4023 	PROC_UNLOCK(p);
4024 	if (kern_lognosys == 1 || kern_lognosys == 3) {
4025 		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4026 		    td->td_sa.code);
4027 	}
4028 	if (kern_lognosys == 2 || kern_lognosys == 3 ||
4029 	    (p->p_pid == 1 && (kern_lognosys & 3) == 0)) {
4030 		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4031 		    td->td_sa.code);
4032 	}
4033 	return (ENOSYS);
4034 }
4035 
4036 /*
4037  * Send a SIGIO or SIGURG signal to a process or process group using stored
4038  * credentials rather than those of the current process.
4039  */
4040 void
4041 pgsigio(struct sigio **sigiop, int sig, int checkctty)
4042 {
4043 	ksiginfo_t ksi;
4044 	struct sigio *sigio;
4045 
4046 	ksiginfo_init(&ksi);
4047 	ksi.ksi_signo = sig;
4048 	ksi.ksi_code = SI_KERNEL;
4049 
4050 	SIGIO_LOCK();
4051 	sigio = *sigiop;
4052 	if (sigio == NULL) {
4053 		SIGIO_UNLOCK();
4054 		return;
4055 	}
4056 	if (sigio->sio_pgid > 0) {
4057 		PROC_LOCK(sigio->sio_proc);
4058 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
4059 			kern_psignal(sigio->sio_proc, sig);
4060 		PROC_UNLOCK(sigio->sio_proc);
4061 	} else if (sigio->sio_pgid < 0) {
4062 		struct proc *p;
4063 
4064 		PGRP_LOCK(sigio->sio_pgrp);
4065 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
4066 			PROC_LOCK(p);
4067 			if (p->p_state == PRS_NORMAL &&
4068 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
4069 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
4070 				kern_psignal(p, sig);
4071 			PROC_UNLOCK(p);
4072 		}
4073 		PGRP_UNLOCK(sigio->sio_pgrp);
4074 	}
4075 	SIGIO_UNLOCK();
4076 }
4077 
4078 static int
4079 filt_sigattach(struct knote *kn)
4080 {
4081 	struct proc *p = curproc;
4082 
4083 	kn->kn_ptr.p_proc = p;
4084 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
4085 
4086 	knlist_add(p->p_klist, kn, 0);
4087 
4088 	return (0);
4089 }
4090 
4091 static void
4092 filt_sigdetach(struct knote *kn)
4093 {
4094 	struct proc *p = kn->kn_ptr.p_proc;
4095 
4096 	knlist_remove(p->p_klist, kn, 0);
4097 }
4098 
4099 /*
4100  * signal knotes are shared with proc knotes, so we apply a mask to
4101  * the hint in order to differentiate them from process hints.  This
4102  * could be avoided by using a signal-specific knote list, but probably
4103  * isn't worth the trouble.
4104  */
4105 static int
4106 filt_signal(struct knote *kn, long hint)
4107 {
4108 
4109 	if (hint & NOTE_SIGNAL) {
4110 		hint &= ~NOTE_SIGNAL;
4111 
4112 		if (kn->kn_id == hint)
4113 			kn->kn_data++;
4114 	}
4115 	return (kn->kn_data != 0);
4116 }
4117 
4118 struct sigacts *
4119 sigacts_alloc(void)
4120 {
4121 	struct sigacts *ps;
4122 
4123 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
4124 	refcount_init(&ps->ps_refcnt, 1);
4125 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
4126 	return (ps);
4127 }
4128 
4129 void
4130 sigacts_free(struct sigacts *ps)
4131 {
4132 
4133 	if (refcount_release(&ps->ps_refcnt) == 0)
4134 		return;
4135 	mtx_destroy(&ps->ps_mtx);
4136 	free(ps, M_SUBPROC);
4137 }
4138 
4139 struct sigacts *
4140 sigacts_hold(struct sigacts *ps)
4141 {
4142 
4143 	refcount_acquire(&ps->ps_refcnt);
4144 	return (ps);
4145 }
4146 
4147 void
4148 sigacts_copy(struct sigacts *dest, struct sigacts *src)
4149 {
4150 
4151 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
4152 	mtx_lock(&src->ps_mtx);
4153 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
4154 	mtx_unlock(&src->ps_mtx);
4155 }
4156 
4157 int
4158 sigacts_shared(struct sigacts *ps)
4159 {
4160 
4161 	return (ps->ps_refcnt > 1);
4162 }
4163 
4164 void
4165 sig_drop_caught(struct proc *p)
4166 {
4167 	int sig;
4168 	struct sigacts *ps;
4169 
4170 	ps = p->p_sigacts;
4171 	PROC_LOCK_ASSERT(p, MA_OWNED);
4172 	mtx_assert(&ps->ps_mtx, MA_OWNED);
4173 	SIG_FOREACH(sig, &ps->ps_sigcatch) {
4174 		sigdflt(ps, sig);
4175 		if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
4176 			sigqueue_delete_proc(p, sig);
4177 	}
4178 }
4179 
4180 static void
4181 sigfastblock_failed(struct thread *td, bool sendsig, bool write)
4182 {
4183 	ksiginfo_t ksi;
4184 
4185 	/*
4186 	 * Prevent further fetches and SIGSEGVs, allowing thread to
4187 	 * issue syscalls despite corruption.
4188 	 */
4189 	sigfastblock_clear(td);
4190 
4191 	if (!sendsig)
4192 		return;
4193 	ksiginfo_init_trap(&ksi);
4194 	ksi.ksi_signo = SIGSEGV;
4195 	ksi.ksi_code = write ? SEGV_ACCERR : SEGV_MAPERR;
4196 	ksi.ksi_addr = td->td_sigblock_ptr;
4197 	trapsignal(td, &ksi);
4198 }
4199 
4200 static bool
4201 sigfastblock_fetch_sig(struct thread *td, bool sendsig, uint32_t *valp)
4202 {
4203 	uint32_t res;
4204 
4205 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4206 		return (true);
4207 	if (fueword32((void *)td->td_sigblock_ptr, &res) == -1) {
4208 		sigfastblock_failed(td, sendsig, false);
4209 		return (false);
4210 	}
4211 	*valp = res;
4212 	td->td_sigblock_val = res & ~SIGFASTBLOCK_FLAGS;
4213 	return (true);
4214 }
4215 
4216 static void
4217 sigfastblock_resched(struct thread *td, bool resched)
4218 {
4219 	struct proc *p;
4220 
4221 	if (resched) {
4222 		p = td->td_proc;
4223 		PROC_LOCK(p);
4224 		reschedule_signals(p, td->td_sigmask, 0);
4225 		PROC_UNLOCK(p);
4226 	}
4227 	thread_lock(td);
4228 	td->td_flags |= TDF_ASTPENDING | TDF_NEEDSIGCHK;
4229 	thread_unlock(td);
4230 }
4231 
4232 int
4233 sys_sigfastblock(struct thread *td, struct sigfastblock_args *uap)
4234 {
4235 	struct proc *p;
4236 	int error, res;
4237 	uint32_t oldval;
4238 
4239 	error = 0;
4240 	p = td->td_proc;
4241 	switch (uap->cmd) {
4242 	case SIGFASTBLOCK_SETPTR:
4243 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
4244 			error = EBUSY;
4245 			break;
4246 		}
4247 		if (((uintptr_t)(uap->ptr) & (sizeof(uint32_t) - 1)) != 0) {
4248 			error = EINVAL;
4249 			break;
4250 		}
4251 		td->td_pflags |= TDP_SIGFASTBLOCK;
4252 		td->td_sigblock_ptr = uap->ptr;
4253 		break;
4254 
4255 	case SIGFASTBLOCK_UNBLOCK:
4256 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4257 			error = EINVAL;
4258 			break;
4259 		}
4260 
4261 		for (;;) {
4262 			res = casueword32(td->td_sigblock_ptr,
4263 			    SIGFASTBLOCK_PEND, &oldval, 0);
4264 			if (res == -1) {
4265 				error = EFAULT;
4266 				sigfastblock_failed(td, false, true);
4267 				break;
4268 			}
4269 			if (res == 0)
4270 				break;
4271 			MPASS(res == 1);
4272 			if (oldval != SIGFASTBLOCK_PEND) {
4273 				error = EBUSY;
4274 				break;
4275 			}
4276 			error = thread_check_susp(td, false);
4277 			if (error != 0)
4278 				break;
4279 		}
4280 		if (error != 0)
4281 			break;
4282 
4283 		/*
4284 		 * td_sigblock_val is cleared there, but not on a
4285 		 * syscall exit.  The end effect is that a single
4286 		 * interruptible sleep, while user sigblock word is
4287 		 * set, might return EINTR or ERESTART to usermode
4288 		 * without delivering signal.  All further sleeps,
4289 		 * until userspace clears the word and does
4290 		 * sigfastblock(UNBLOCK), observe current word and no
4291 		 * longer get interrupted.  It is slight
4292 		 * non-conformance, with alternative to have read the
4293 		 * sigblock word on each syscall entry.
4294 		 */
4295 		td->td_sigblock_val = 0;
4296 
4297 		/*
4298 		 * Rely on normal ast mechanism to deliver pending
4299 		 * signals to current thread.  But notify others about
4300 		 * fake unblock.
4301 		 */
4302 		sigfastblock_resched(td, error == 0 && p->p_numthreads != 1);
4303 
4304 		break;
4305 
4306 	case SIGFASTBLOCK_UNSETPTR:
4307 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4308 			error = EINVAL;
4309 			break;
4310 		}
4311 		if (!sigfastblock_fetch_sig(td, false, &oldval)) {
4312 			error = EFAULT;
4313 			break;
4314 		}
4315 		if (oldval != 0 && oldval != SIGFASTBLOCK_PEND) {
4316 			error = EBUSY;
4317 			break;
4318 		}
4319 		sigfastblock_clear(td);
4320 		break;
4321 
4322 	default:
4323 		error = EINVAL;
4324 		break;
4325 	}
4326 	return (error);
4327 }
4328 
4329 void
4330 sigfastblock_clear(struct thread *td)
4331 {
4332 	bool resched;
4333 
4334 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4335 		return;
4336 	td->td_sigblock_val = 0;
4337 	resched = (td->td_pflags & TDP_SIGFASTPENDING) != 0 ||
4338 	    SIGPENDING(td);
4339 	td->td_pflags &= ~(TDP_SIGFASTBLOCK | TDP_SIGFASTPENDING);
4340 	sigfastblock_resched(td, resched);
4341 }
4342 
4343 void
4344 sigfastblock_fetch(struct thread *td)
4345 {
4346 	uint32_t val;
4347 
4348 	(void)sigfastblock_fetch_sig(td, true, &val);
4349 }
4350 
4351 static void
4352 sigfastblock_setpend1(struct thread *td)
4353 {
4354 	int res;
4355 	uint32_t oldval;
4356 
4357 	if ((td->td_pflags & TDP_SIGFASTPENDING) == 0)
4358 		return;
4359 	res = fueword32((void *)td->td_sigblock_ptr, &oldval);
4360 	if (res == -1) {
4361 		sigfastblock_failed(td, true, false);
4362 		return;
4363 	}
4364 	for (;;) {
4365 		res = casueword32(td->td_sigblock_ptr, oldval, &oldval,
4366 		    oldval | SIGFASTBLOCK_PEND);
4367 		if (res == -1) {
4368 			sigfastblock_failed(td, true, true);
4369 			return;
4370 		}
4371 		if (res == 0) {
4372 			td->td_sigblock_val = oldval & ~SIGFASTBLOCK_FLAGS;
4373 			td->td_pflags &= ~TDP_SIGFASTPENDING;
4374 			break;
4375 		}
4376 		MPASS(res == 1);
4377 		if (thread_check_susp(td, false) != 0)
4378 			break;
4379 	}
4380 }
4381 
4382 void
4383 sigfastblock_setpend(struct thread *td, bool resched)
4384 {
4385 	struct proc *p;
4386 
4387 	sigfastblock_setpend1(td);
4388 	if (resched) {
4389 		p = td->td_proc;
4390 		PROC_LOCK(p);
4391 		reschedule_signals(p, fastblock_mask, SIGPROCMASK_FASTBLK);
4392 		PROC_UNLOCK(p);
4393 	}
4394 }
4395