xref: /freebsd/sys/kern/kern_sig.c (revision 38a52bd3)
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 *);
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 static void	sigfastblock_setpend(struct thread *td, bool resched);
120 
121 static uma_zone_t	ksiginfo_zone = NULL;
122 struct filterops sig_filtops = {
123 	.f_isfd = 0,
124 	.f_attach = filt_sigattach,
125 	.f_detach = filt_sigdetach,
126 	.f_event = filt_signal,
127 };
128 
129 static int	kern_logsigexit = 1;
130 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
131     &kern_logsigexit, 0,
132     "Log processes quitting on abnormal signals to syslog(3)");
133 
134 static int	kern_forcesigexit = 1;
135 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
136     &kern_forcesigexit, 0, "Force trap signal to be handled");
137 
138 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
139     "POSIX real time signal");
140 
141 static int	max_pending_per_proc = 128;
142 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
143     &max_pending_per_proc, 0, "Max pending signals per proc");
144 
145 static int	preallocate_siginfo = 1024;
146 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
147     &preallocate_siginfo, 0, "Preallocated signal memory size");
148 
149 static int	signal_overflow = 0;
150 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
151     &signal_overflow, 0, "Number of signals overflew");
152 
153 static int	signal_alloc_fail = 0;
154 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
155     &signal_alloc_fail, 0, "signals failed to be allocated");
156 
157 static int	kern_lognosys = 0;
158 SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
159     "Log invalid syscalls");
160 
161 __read_frequently bool sigfastblock_fetch_always = false;
162 SYSCTL_BOOL(_kern, OID_AUTO, sigfastblock_fetch_always, CTLFLAG_RWTUN,
163     &sigfastblock_fetch_always, 0,
164     "Fetch sigfastblock word on each syscall entry for proper "
165     "blocking semantic");
166 
167 static bool	kern_sig_discard_ign = true;
168 SYSCTL_BOOL(_kern, OID_AUTO, sig_discard_ign, CTLFLAG_RWTUN,
169     &kern_sig_discard_ign, 0,
170     "Discard ignored signals on delivery, otherwise queue them to "
171     "the target queue");
172 
173 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
174 
175 /*
176  * Policy -- Can ucred cr1 send SIGIO to process cr2?
177  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
178  * in the right situations.
179  */
180 #define CANSIGIO(cr1, cr2) \
181 	((cr1)->cr_uid == 0 || \
182 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
183 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
184 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
185 	    (cr1)->cr_uid == (cr2)->cr_uid)
186 
187 static int	sugid_coredump;
188 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
189     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
190 
191 static int	capmode_coredump;
192 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
193     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
194 
195 static int	do_coredump = 1;
196 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
197 	&do_coredump, 0, "Enable/Disable coredumps");
198 
199 static int	set_core_nodump_flag = 0;
200 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
201 	0, "Enable setting the NODUMP flag on coredump files");
202 
203 static int	coredump_devctl = 0;
204 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
205 	0, "Generate a devctl notification when processes coredump");
206 
207 /*
208  * Signal properties and actions.
209  * The array below categorizes the signals and their default actions
210  * according to the following properties:
211  */
212 #define	SIGPROP_KILL		0x01	/* terminates process by default */
213 #define	SIGPROP_CORE		0x02	/* ditto and coredumps */
214 #define	SIGPROP_STOP		0x04	/* suspend process */
215 #define	SIGPROP_TTYSTOP		0x08	/* ditto, from tty */
216 #define	SIGPROP_IGNORE		0x10	/* ignore by default */
217 #define	SIGPROP_CONT		0x20	/* continue if suspended */
218 
219 static int sigproptbl[NSIG] = {
220 	[SIGHUP] =	SIGPROP_KILL,
221 	[SIGINT] =	SIGPROP_KILL,
222 	[SIGQUIT] =	SIGPROP_KILL | SIGPROP_CORE,
223 	[SIGILL] =	SIGPROP_KILL | SIGPROP_CORE,
224 	[SIGTRAP] =	SIGPROP_KILL | SIGPROP_CORE,
225 	[SIGABRT] =	SIGPROP_KILL | SIGPROP_CORE,
226 	[SIGEMT] =	SIGPROP_KILL | SIGPROP_CORE,
227 	[SIGFPE] =	SIGPROP_KILL | SIGPROP_CORE,
228 	[SIGKILL] =	SIGPROP_KILL,
229 	[SIGBUS] =	SIGPROP_KILL | SIGPROP_CORE,
230 	[SIGSEGV] =	SIGPROP_KILL | SIGPROP_CORE,
231 	[SIGSYS] =	SIGPROP_KILL | SIGPROP_CORE,
232 	[SIGPIPE] =	SIGPROP_KILL,
233 	[SIGALRM] =	SIGPROP_KILL,
234 	[SIGTERM] =	SIGPROP_KILL,
235 	[SIGURG] =	SIGPROP_IGNORE,
236 	[SIGSTOP] =	SIGPROP_STOP,
237 	[SIGTSTP] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
238 	[SIGCONT] =	SIGPROP_IGNORE | SIGPROP_CONT,
239 	[SIGCHLD] =	SIGPROP_IGNORE,
240 	[SIGTTIN] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
241 	[SIGTTOU] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
242 	[SIGIO] =	SIGPROP_IGNORE,
243 	[SIGXCPU] =	SIGPROP_KILL,
244 	[SIGXFSZ] =	SIGPROP_KILL,
245 	[SIGVTALRM] =	SIGPROP_KILL,
246 	[SIGPROF] =	SIGPROP_KILL,
247 	[SIGWINCH] =	SIGPROP_IGNORE,
248 	[SIGINFO] =	SIGPROP_IGNORE,
249 	[SIGUSR1] =	SIGPROP_KILL,
250 	[SIGUSR2] =	SIGPROP_KILL,
251 };
252 
253 #define	_SIG_FOREACH_ADVANCE(i, set) ({					\
254 	int __found;							\
255 	for (;;) {							\
256 		if (__bits != 0) {					\
257 			int __sig = ffs(__bits);			\
258 			__bits &= ~(1u << (__sig - 1));			\
259 			sig = __i * sizeof((set)->__bits[0]) * NBBY + __sig; \
260 			__found = 1;					\
261 			break;						\
262 		}							\
263 		if (++__i == _SIG_WORDS) {				\
264 			__found = 0;					\
265 			break;						\
266 		}							\
267 		__bits = (set)->__bits[__i];				\
268 	}								\
269 	__found != 0;							\
270 })
271 
272 #define	SIG_FOREACH(i, set)						\
273 	for (int32_t __i = -1, __bits = 0;				\
274 	    _SIG_FOREACH_ADVANCE(i, set); )				\
275 
276 static sigset_t fastblock_mask;
277 
278 static void
279 ast_sig(struct thread *td, int tda)
280 {
281 	struct proc *p;
282 	int sig;
283 	bool resched_sigs;
284 
285 	p = td->td_proc;
286 
287 #ifdef DIAGNOSTIC
288 	if (p->p_numthreads == 1 && (tda & (TDAI(TDA_SIG) |
289 	    TDAI(TDA_AST))) == 0) {
290 		PROC_LOCK(p);
291 		thread_lock(td);
292 		/*
293 		 * Note that TDA_SIG should be re-read from
294 		 * td_ast, since signal might have been delivered
295 		 * after we cleared td_flags above.  This is one of
296 		 * the reason for looping check for AST condition.
297 		 * See comment in userret() about P_PPWAIT.
298 		 */
299 		if ((p->p_flag & P_PPWAIT) == 0 &&
300 		    (td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
301 			if (SIGPENDING(td) && ((tda | td->td_ast) &
302 			    (TDAI(TDA_SIG) | TDAI(TDA_AST))) == 0) {
303 				thread_unlock(td); /* fix dumps */
304 				panic(
305 				    "failed2 to set signal flags for ast p %p "
306 				    "td %p tda %#x td_ast %#x fl %#x",
307 				    p, td, tda, td->td_ast, td->td_flags);
308 			}
309 		}
310 		thread_unlock(td);
311 		PROC_UNLOCK(p);
312 	}
313 #endif
314 
315 	/*
316 	 * Check for signals. Unlocked reads of p_pendingcnt or
317 	 * p_siglist might cause process-directed signal to be handled
318 	 * later.
319 	 */
320 	if ((tda & TDAI(TDA_SIG)) != 0 || p->p_pendingcnt > 0 ||
321 	    !SIGISEMPTY(p->p_siglist)) {
322 		sigfastblock_fetch(td);
323 		PROC_LOCK(p);
324 		mtx_lock(&p->p_sigacts->ps_mtx);
325 		while ((sig = cursig(td)) != 0) {
326 			KASSERT(sig >= 0, ("sig %d", sig));
327 			postsig(sig);
328 		}
329 		mtx_unlock(&p->p_sigacts->ps_mtx);
330 		PROC_UNLOCK(p);
331 		resched_sigs = true;
332 	} else {
333 		resched_sigs = false;
334 	}
335 
336 	/*
337 	 * Handle deferred update of the fast sigblock value, after
338 	 * the postsig() loop was performed.
339 	 */
340 	sigfastblock_setpend(td, resched_sigs);
341 }
342 
343 static void
344 ast_sigsuspend(struct thread *td, int tda __unused)
345 {
346 	MPASS((td->td_pflags & TDP_OLDMASK) != 0);
347 	td->td_pflags &= ~TDP_OLDMASK;
348 	kern_sigprocmask(td, SIG_SETMASK, &td->td_oldsigmask, NULL, 0);
349 }
350 
351 static void
352 sigqueue_start(void)
353 {
354 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
355 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
356 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
357 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
358 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
359 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
360 	SIGFILLSET(fastblock_mask);
361 	SIG_CANTMASK(fastblock_mask);
362 	ast_register(TDA_SIG, ASTR_UNCOND, 0, ast_sig);
363 	ast_register(TDA_SIGSUSPEND, ASTR_ASTF_REQUIRED | ASTR_TDP,
364 	    TDP_OLDMASK, ast_sigsuspend);
365 }
366 
367 ksiginfo_t *
368 ksiginfo_alloc(int mwait)
369 {
370 	MPASS(mwait == M_WAITOK || mwait == M_NOWAIT);
371 
372 	if (ksiginfo_zone == NULL)
373 		return (NULL);
374 	return (uma_zalloc(ksiginfo_zone, mwait | M_ZERO));
375 }
376 
377 void
378 ksiginfo_free(ksiginfo_t *ksi)
379 {
380 	uma_zfree(ksiginfo_zone, ksi);
381 }
382 
383 static __inline bool
384 ksiginfo_tryfree(ksiginfo_t *ksi)
385 {
386 	if ((ksi->ksi_flags & KSI_EXT) == 0) {
387 		uma_zfree(ksiginfo_zone, ksi);
388 		return (true);
389 	}
390 	return (false);
391 }
392 
393 void
394 sigqueue_init(sigqueue_t *list, struct proc *p)
395 {
396 	SIGEMPTYSET(list->sq_signals);
397 	SIGEMPTYSET(list->sq_kill);
398 	SIGEMPTYSET(list->sq_ptrace);
399 	TAILQ_INIT(&list->sq_list);
400 	list->sq_proc = p;
401 	list->sq_flags = SQ_INIT;
402 }
403 
404 /*
405  * Get a signal's ksiginfo.
406  * Return:
407  *	0	-	signal not found
408  *	others	-	signal number
409  */
410 static int
411 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
412 {
413 	struct proc *p = sq->sq_proc;
414 	struct ksiginfo *ksi, *next;
415 	int count = 0;
416 
417 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
418 
419 	if (!SIGISMEMBER(sq->sq_signals, signo))
420 		return (0);
421 
422 	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
423 		count++;
424 		SIGDELSET(sq->sq_ptrace, signo);
425 		si->ksi_flags |= KSI_PTRACE;
426 	}
427 	if (SIGISMEMBER(sq->sq_kill, signo)) {
428 		count++;
429 		if (count == 1)
430 			SIGDELSET(sq->sq_kill, signo);
431 	}
432 
433 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
434 		if (ksi->ksi_signo == signo) {
435 			if (count == 0) {
436 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
437 				ksi->ksi_sigq = NULL;
438 				ksiginfo_copy(ksi, si);
439 				if (ksiginfo_tryfree(ksi) && p != NULL)
440 					p->p_pendingcnt--;
441 			}
442 			if (++count > 1)
443 				break;
444 		}
445 	}
446 
447 	if (count <= 1)
448 		SIGDELSET(sq->sq_signals, signo);
449 	si->ksi_signo = signo;
450 	return (signo);
451 }
452 
453 void
454 sigqueue_take(ksiginfo_t *ksi)
455 {
456 	struct ksiginfo *kp;
457 	struct proc	*p;
458 	sigqueue_t	*sq;
459 
460 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
461 		return;
462 
463 	p = sq->sq_proc;
464 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
465 	ksi->ksi_sigq = NULL;
466 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
467 		p->p_pendingcnt--;
468 
469 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
470 	     kp = TAILQ_NEXT(kp, ksi_link)) {
471 		if (kp->ksi_signo == ksi->ksi_signo)
472 			break;
473 	}
474 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
475 	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
476 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
477 }
478 
479 static int
480 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
481 {
482 	struct proc *p = sq->sq_proc;
483 	struct ksiginfo *ksi;
484 	int ret = 0;
485 
486 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
487 
488 	/*
489 	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
490 	 * for these signals.
491 	 */
492 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
493 		SIGADDSET(sq->sq_kill, signo);
494 		goto out_set_bit;
495 	}
496 
497 	/* directly insert the ksi, don't copy it */
498 	if (si->ksi_flags & KSI_INS) {
499 		if (si->ksi_flags & KSI_HEAD)
500 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
501 		else
502 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
503 		si->ksi_sigq = sq;
504 		goto out_set_bit;
505 	}
506 
507 	if (__predict_false(ksiginfo_zone == NULL)) {
508 		SIGADDSET(sq->sq_kill, signo);
509 		goto out_set_bit;
510 	}
511 
512 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
513 		signal_overflow++;
514 		ret = EAGAIN;
515 	} else if ((ksi = ksiginfo_alloc(M_NOWAIT)) == NULL) {
516 		signal_alloc_fail++;
517 		ret = EAGAIN;
518 	} else {
519 		if (p != NULL)
520 			p->p_pendingcnt++;
521 		ksiginfo_copy(si, ksi);
522 		ksi->ksi_signo = signo;
523 		if (si->ksi_flags & KSI_HEAD)
524 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
525 		else
526 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
527 		ksi->ksi_sigq = sq;
528 	}
529 
530 	if (ret != 0) {
531 		if ((si->ksi_flags & KSI_PTRACE) != 0) {
532 			SIGADDSET(sq->sq_ptrace, signo);
533 			ret = 0;
534 			goto out_set_bit;
535 		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
536 		    (si->ksi_flags & KSI_SIGQ) == 0) {
537 			SIGADDSET(sq->sq_kill, signo);
538 			ret = 0;
539 			goto out_set_bit;
540 		}
541 		return (ret);
542 	}
543 
544 out_set_bit:
545 	SIGADDSET(sq->sq_signals, signo);
546 	return (ret);
547 }
548 
549 void
550 sigqueue_flush(sigqueue_t *sq)
551 {
552 	struct proc *p = sq->sq_proc;
553 	ksiginfo_t *ksi;
554 
555 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
556 
557 	if (p != NULL)
558 		PROC_LOCK_ASSERT(p, MA_OWNED);
559 
560 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
561 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
562 		ksi->ksi_sigq = NULL;
563 		if (ksiginfo_tryfree(ksi) && p != NULL)
564 			p->p_pendingcnt--;
565 	}
566 
567 	SIGEMPTYSET(sq->sq_signals);
568 	SIGEMPTYSET(sq->sq_kill);
569 	SIGEMPTYSET(sq->sq_ptrace);
570 }
571 
572 static void
573 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
574 {
575 	sigset_t tmp;
576 	struct proc *p1, *p2;
577 	ksiginfo_t *ksi, *next;
578 
579 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
580 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
581 	p1 = src->sq_proc;
582 	p2 = dst->sq_proc;
583 	/* Move siginfo to target list */
584 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
585 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
586 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
587 			if (p1 != NULL)
588 				p1->p_pendingcnt--;
589 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
590 			ksi->ksi_sigq = dst;
591 			if (p2 != NULL)
592 				p2->p_pendingcnt++;
593 		}
594 	}
595 
596 	/* Move pending bits to target list */
597 	tmp = src->sq_kill;
598 	SIGSETAND(tmp, *set);
599 	SIGSETOR(dst->sq_kill, tmp);
600 	SIGSETNAND(src->sq_kill, tmp);
601 
602 	tmp = src->sq_ptrace;
603 	SIGSETAND(tmp, *set);
604 	SIGSETOR(dst->sq_ptrace, tmp);
605 	SIGSETNAND(src->sq_ptrace, tmp);
606 
607 	tmp = src->sq_signals;
608 	SIGSETAND(tmp, *set);
609 	SIGSETOR(dst->sq_signals, tmp);
610 	SIGSETNAND(src->sq_signals, tmp);
611 }
612 
613 #if 0
614 static void
615 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
616 {
617 	sigset_t set;
618 
619 	SIGEMPTYSET(set);
620 	SIGADDSET(set, signo);
621 	sigqueue_move_set(src, dst, &set);
622 }
623 #endif
624 
625 static void
626 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
627 {
628 	struct proc *p = sq->sq_proc;
629 	ksiginfo_t *ksi, *next;
630 
631 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
632 
633 	/* Remove siginfo queue */
634 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
635 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
636 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
637 			ksi->ksi_sigq = NULL;
638 			if (ksiginfo_tryfree(ksi) && p != NULL)
639 				p->p_pendingcnt--;
640 		}
641 	}
642 	SIGSETNAND(sq->sq_kill, *set);
643 	SIGSETNAND(sq->sq_ptrace, *set);
644 	SIGSETNAND(sq->sq_signals, *set);
645 }
646 
647 void
648 sigqueue_delete(sigqueue_t *sq, int signo)
649 {
650 	sigset_t set;
651 
652 	SIGEMPTYSET(set);
653 	SIGADDSET(set, signo);
654 	sigqueue_delete_set(sq, &set);
655 }
656 
657 /* Remove a set of signals for a process */
658 static void
659 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
660 {
661 	sigqueue_t worklist;
662 	struct thread *td0;
663 
664 	PROC_LOCK_ASSERT(p, MA_OWNED);
665 
666 	sigqueue_init(&worklist, NULL);
667 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
668 
669 	FOREACH_THREAD_IN_PROC(p, td0)
670 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
671 
672 	sigqueue_flush(&worklist);
673 }
674 
675 void
676 sigqueue_delete_proc(struct proc *p, int signo)
677 {
678 	sigset_t set;
679 
680 	SIGEMPTYSET(set);
681 	SIGADDSET(set, signo);
682 	sigqueue_delete_set_proc(p, &set);
683 }
684 
685 static void
686 sigqueue_delete_stopmask_proc(struct proc *p)
687 {
688 	sigset_t set;
689 
690 	SIGEMPTYSET(set);
691 	SIGADDSET(set, SIGSTOP);
692 	SIGADDSET(set, SIGTSTP);
693 	SIGADDSET(set, SIGTTIN);
694 	SIGADDSET(set, SIGTTOU);
695 	sigqueue_delete_set_proc(p, &set);
696 }
697 
698 /*
699  * Determine signal that should be delivered to thread td, the current
700  * thread, 0 if none.  If there is a pending stop signal with default
701  * action, the process stops in issignal().
702  */
703 int
704 cursig(struct thread *td)
705 {
706 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
707 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
708 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
709 	return (SIGPENDING(td) ? issignal(td) : 0);
710 }
711 
712 /*
713  * Arrange for ast() to handle unmasked pending signals on return to user
714  * mode.  This must be called whenever a signal is added to td_sigqueue or
715  * unmasked in td_sigmask.
716  */
717 void
718 signotify(struct thread *td)
719 {
720 
721 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
722 
723 	if (SIGPENDING(td))
724 		ast_sched(td, TDA_SIG);
725 }
726 
727 /*
728  * Returns 1 (true) if altstack is configured for the thread, and the
729  * passed stack bottom address falls into the altstack range.  Handles
730  * the 43 compat special case where the alt stack size is zero.
731  */
732 int
733 sigonstack(size_t sp)
734 {
735 	struct thread *td;
736 
737 	td = curthread;
738 	if ((td->td_pflags & TDP_ALTSTACK) == 0)
739 		return (0);
740 #if defined(COMPAT_43)
741 	if (SV_PROC_FLAG(td->td_proc, SV_AOUT) && td->td_sigstk.ss_size == 0)
742 		return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
743 #endif
744 	return (sp >= (size_t)td->td_sigstk.ss_sp &&
745 	    sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
746 }
747 
748 static __inline int
749 sigprop(int sig)
750 {
751 
752 	if (sig > 0 && sig < nitems(sigproptbl))
753 		return (sigproptbl[sig]);
754 	return (0);
755 }
756 
757 static bool
758 sigact_flag_test(const struct sigaction *act, int flag)
759 {
760 
761 	/*
762 	 * SA_SIGINFO is reset when signal disposition is set to
763 	 * ignore or default.  Other flags are kept according to user
764 	 * settings.
765 	 */
766 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
767 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
768 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
769 }
770 
771 /*
772  * kern_sigaction
773  * sigaction
774  * freebsd4_sigaction
775  * osigaction
776  */
777 int
778 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
779     struct sigaction *oact, int flags)
780 {
781 	struct sigacts *ps;
782 	struct proc *p = td->td_proc;
783 
784 	if (!_SIG_VALID(sig))
785 		return (EINVAL);
786 	if (act != NULL && act->sa_handler != SIG_DFL &&
787 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
788 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
789 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
790 		return (EINVAL);
791 
792 	PROC_LOCK(p);
793 	ps = p->p_sigacts;
794 	mtx_lock(&ps->ps_mtx);
795 	if (oact) {
796 		memset(oact, 0, sizeof(*oact));
797 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
798 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
799 			oact->sa_flags |= SA_ONSTACK;
800 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
801 			oact->sa_flags |= SA_RESTART;
802 		if (SIGISMEMBER(ps->ps_sigreset, sig))
803 			oact->sa_flags |= SA_RESETHAND;
804 		if (SIGISMEMBER(ps->ps_signodefer, sig))
805 			oact->sa_flags |= SA_NODEFER;
806 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
807 			oact->sa_flags |= SA_SIGINFO;
808 			oact->sa_sigaction =
809 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
810 		} else
811 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
812 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
813 			oact->sa_flags |= SA_NOCLDSTOP;
814 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
815 			oact->sa_flags |= SA_NOCLDWAIT;
816 	}
817 	if (act) {
818 		if ((sig == SIGKILL || sig == SIGSTOP) &&
819 		    act->sa_handler != SIG_DFL) {
820 			mtx_unlock(&ps->ps_mtx);
821 			PROC_UNLOCK(p);
822 			return (EINVAL);
823 		}
824 
825 		/*
826 		 * Change setting atomically.
827 		 */
828 
829 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
830 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
831 		if (sigact_flag_test(act, SA_SIGINFO)) {
832 			ps->ps_sigact[_SIG_IDX(sig)] =
833 			    (__sighandler_t *)act->sa_sigaction;
834 			SIGADDSET(ps->ps_siginfo, sig);
835 		} else {
836 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
837 			SIGDELSET(ps->ps_siginfo, sig);
838 		}
839 		if (!sigact_flag_test(act, SA_RESTART))
840 			SIGADDSET(ps->ps_sigintr, sig);
841 		else
842 			SIGDELSET(ps->ps_sigintr, sig);
843 		if (sigact_flag_test(act, SA_ONSTACK))
844 			SIGADDSET(ps->ps_sigonstack, sig);
845 		else
846 			SIGDELSET(ps->ps_sigonstack, sig);
847 		if (sigact_flag_test(act, SA_RESETHAND))
848 			SIGADDSET(ps->ps_sigreset, sig);
849 		else
850 			SIGDELSET(ps->ps_sigreset, sig);
851 		if (sigact_flag_test(act, SA_NODEFER))
852 			SIGADDSET(ps->ps_signodefer, sig);
853 		else
854 			SIGDELSET(ps->ps_signodefer, sig);
855 		if (sig == SIGCHLD) {
856 			if (act->sa_flags & SA_NOCLDSTOP)
857 				ps->ps_flag |= PS_NOCLDSTOP;
858 			else
859 				ps->ps_flag &= ~PS_NOCLDSTOP;
860 			if (act->sa_flags & SA_NOCLDWAIT) {
861 				/*
862 				 * Paranoia: since SA_NOCLDWAIT is implemented
863 				 * by reparenting the dying child to PID 1 (and
864 				 * trust it to reap the zombie), PID 1 itself
865 				 * is forbidden to set SA_NOCLDWAIT.
866 				 */
867 				if (p->p_pid == 1)
868 					ps->ps_flag &= ~PS_NOCLDWAIT;
869 				else
870 					ps->ps_flag |= PS_NOCLDWAIT;
871 			} else
872 				ps->ps_flag &= ~PS_NOCLDWAIT;
873 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
874 				ps->ps_flag |= PS_CLDSIGIGN;
875 			else
876 				ps->ps_flag &= ~PS_CLDSIGIGN;
877 		}
878 		/*
879 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
880 		 * and for signals set to SIG_DFL where the default is to
881 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
882 		 * have to restart the process.
883 		 */
884 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
885 		    (sigprop(sig) & SIGPROP_IGNORE &&
886 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
887 			/* never to be seen again */
888 			sigqueue_delete_proc(p, sig);
889 			if (sig != SIGCONT)
890 				/* easier in psignal */
891 				SIGADDSET(ps->ps_sigignore, sig);
892 			SIGDELSET(ps->ps_sigcatch, sig);
893 		} else {
894 			SIGDELSET(ps->ps_sigignore, sig);
895 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
896 				SIGDELSET(ps->ps_sigcatch, sig);
897 			else
898 				SIGADDSET(ps->ps_sigcatch, sig);
899 		}
900 #ifdef COMPAT_FREEBSD4
901 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
902 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
903 		    (flags & KSA_FREEBSD4) == 0)
904 			SIGDELSET(ps->ps_freebsd4, sig);
905 		else
906 			SIGADDSET(ps->ps_freebsd4, sig);
907 #endif
908 #ifdef COMPAT_43
909 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
910 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
911 		    (flags & KSA_OSIGSET) == 0)
912 			SIGDELSET(ps->ps_osigset, sig);
913 		else
914 			SIGADDSET(ps->ps_osigset, sig);
915 #endif
916 	}
917 	mtx_unlock(&ps->ps_mtx);
918 	PROC_UNLOCK(p);
919 	return (0);
920 }
921 
922 #ifndef _SYS_SYSPROTO_H_
923 struct sigaction_args {
924 	int	sig;
925 	struct	sigaction *act;
926 	struct	sigaction *oact;
927 };
928 #endif
929 int
930 sys_sigaction(struct thread *td, struct sigaction_args *uap)
931 {
932 	struct sigaction act, oact;
933 	struct sigaction *actp, *oactp;
934 	int error;
935 
936 	actp = (uap->act != NULL) ? &act : NULL;
937 	oactp = (uap->oact != NULL) ? &oact : NULL;
938 	if (actp) {
939 		error = copyin(uap->act, actp, sizeof(act));
940 		if (error)
941 			return (error);
942 	}
943 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
944 	if (oactp && !error)
945 		error = copyout(oactp, uap->oact, sizeof(oact));
946 	return (error);
947 }
948 
949 #ifdef COMPAT_FREEBSD4
950 #ifndef _SYS_SYSPROTO_H_
951 struct freebsd4_sigaction_args {
952 	int	sig;
953 	struct	sigaction *act;
954 	struct	sigaction *oact;
955 };
956 #endif
957 int
958 freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
959 {
960 	struct sigaction act, oact;
961 	struct sigaction *actp, *oactp;
962 	int error;
963 
964 	actp = (uap->act != NULL) ? &act : NULL;
965 	oactp = (uap->oact != NULL) ? &oact : NULL;
966 	if (actp) {
967 		error = copyin(uap->act, actp, sizeof(act));
968 		if (error)
969 			return (error);
970 	}
971 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
972 	if (oactp && !error)
973 		error = copyout(oactp, uap->oact, sizeof(oact));
974 	return (error);
975 }
976 #endif	/* COMAPT_FREEBSD4 */
977 
978 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
979 #ifndef _SYS_SYSPROTO_H_
980 struct osigaction_args {
981 	int	signum;
982 	struct	osigaction *nsa;
983 	struct	osigaction *osa;
984 };
985 #endif
986 int
987 osigaction(struct thread *td, struct osigaction_args *uap)
988 {
989 	struct osigaction sa;
990 	struct sigaction nsa, osa;
991 	struct sigaction *nsap, *osap;
992 	int error;
993 
994 	if (uap->signum <= 0 || uap->signum >= ONSIG)
995 		return (EINVAL);
996 
997 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
998 	osap = (uap->osa != NULL) ? &osa : NULL;
999 
1000 	if (nsap) {
1001 		error = copyin(uap->nsa, &sa, sizeof(sa));
1002 		if (error)
1003 			return (error);
1004 		nsap->sa_handler = sa.sa_handler;
1005 		nsap->sa_flags = sa.sa_flags;
1006 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
1007 	}
1008 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1009 	if (osap && !error) {
1010 		sa.sa_handler = osap->sa_handler;
1011 		sa.sa_flags = osap->sa_flags;
1012 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
1013 		error = copyout(&sa, uap->osa, sizeof(sa));
1014 	}
1015 	return (error);
1016 }
1017 
1018 #if !defined(__i386__)
1019 /* Avoid replicating the same stub everywhere */
1020 int
1021 osigreturn(struct thread *td, struct osigreturn_args *uap)
1022 {
1023 
1024 	return (nosys(td, (struct nosys_args *)uap));
1025 }
1026 #endif
1027 #endif /* COMPAT_43 */
1028 
1029 /*
1030  * Initialize signal state for process 0;
1031  * set to ignore signals that are ignored by default.
1032  */
1033 void
1034 siginit(struct proc *p)
1035 {
1036 	int i;
1037 	struct sigacts *ps;
1038 
1039 	PROC_LOCK(p);
1040 	ps = p->p_sigacts;
1041 	mtx_lock(&ps->ps_mtx);
1042 	for (i = 1; i <= NSIG; i++) {
1043 		if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) {
1044 			SIGADDSET(ps->ps_sigignore, i);
1045 		}
1046 	}
1047 	mtx_unlock(&ps->ps_mtx);
1048 	PROC_UNLOCK(p);
1049 }
1050 
1051 /*
1052  * Reset specified signal to the default disposition.
1053  */
1054 static void
1055 sigdflt(struct sigacts *ps, int sig)
1056 {
1057 
1058 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1059 	SIGDELSET(ps->ps_sigcatch, sig);
1060 	if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT)
1061 		SIGADDSET(ps->ps_sigignore, sig);
1062 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1063 	SIGDELSET(ps->ps_siginfo, sig);
1064 }
1065 
1066 /*
1067  * Reset signals for an exec of the specified process.
1068  */
1069 void
1070 execsigs(struct proc *p)
1071 {
1072 	struct sigacts *ps;
1073 	struct thread *td;
1074 
1075 	/*
1076 	 * Reset caught signals.  Held signals remain held
1077 	 * through td_sigmask (unless they were caught,
1078 	 * and are now ignored by default).
1079 	 */
1080 	PROC_LOCK_ASSERT(p, MA_OWNED);
1081 	ps = p->p_sigacts;
1082 	mtx_lock(&ps->ps_mtx);
1083 	sig_drop_caught(p);
1084 
1085 	/*
1086 	 * Reset stack state to the user stack.
1087 	 * Clear set of signals caught on the signal stack.
1088 	 */
1089 	td = curthread;
1090 	MPASS(td->td_proc == p);
1091 	td->td_sigstk.ss_flags = SS_DISABLE;
1092 	td->td_sigstk.ss_size = 0;
1093 	td->td_sigstk.ss_sp = 0;
1094 	td->td_pflags &= ~TDP_ALTSTACK;
1095 	/*
1096 	 * Reset no zombies if child dies flag as Solaris does.
1097 	 */
1098 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1099 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1100 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1101 	mtx_unlock(&ps->ps_mtx);
1102 }
1103 
1104 /*
1105  * kern_sigprocmask()
1106  *
1107  *	Manipulate signal mask.
1108  */
1109 int
1110 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1111     int flags)
1112 {
1113 	sigset_t new_block, oset1;
1114 	struct proc *p;
1115 	int error;
1116 
1117 	p = td->td_proc;
1118 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1119 		PROC_LOCK_ASSERT(p, MA_OWNED);
1120 	else
1121 		PROC_LOCK(p);
1122 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1123 	    ? MA_OWNED : MA_NOTOWNED);
1124 	if (oset != NULL)
1125 		*oset = td->td_sigmask;
1126 
1127 	error = 0;
1128 	if (set != NULL) {
1129 		switch (how) {
1130 		case SIG_BLOCK:
1131 			SIG_CANTMASK(*set);
1132 			oset1 = td->td_sigmask;
1133 			SIGSETOR(td->td_sigmask, *set);
1134 			new_block = td->td_sigmask;
1135 			SIGSETNAND(new_block, oset1);
1136 			break;
1137 		case SIG_UNBLOCK:
1138 			SIGSETNAND(td->td_sigmask, *set);
1139 			signotify(td);
1140 			goto out;
1141 		case SIG_SETMASK:
1142 			SIG_CANTMASK(*set);
1143 			oset1 = td->td_sigmask;
1144 			if (flags & SIGPROCMASK_OLD)
1145 				SIGSETLO(td->td_sigmask, *set);
1146 			else
1147 				td->td_sigmask = *set;
1148 			new_block = td->td_sigmask;
1149 			SIGSETNAND(new_block, oset1);
1150 			signotify(td);
1151 			break;
1152 		default:
1153 			error = EINVAL;
1154 			goto out;
1155 		}
1156 
1157 		/*
1158 		 * The new_block set contains signals that were not previously
1159 		 * blocked, but are blocked now.
1160 		 *
1161 		 * In case we block any signal that was not previously blocked
1162 		 * for td, and process has the signal pending, try to schedule
1163 		 * signal delivery to some thread that does not block the
1164 		 * signal, possibly waking it up.
1165 		 */
1166 		if (p->p_numthreads != 1)
1167 			reschedule_signals(p, new_block, flags);
1168 	}
1169 
1170 out:
1171 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1172 		PROC_UNLOCK(p);
1173 	return (error);
1174 }
1175 
1176 #ifndef _SYS_SYSPROTO_H_
1177 struct sigprocmask_args {
1178 	int	how;
1179 	const sigset_t *set;
1180 	sigset_t *oset;
1181 };
1182 #endif
1183 int
1184 sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1185 {
1186 	sigset_t set, oset;
1187 	sigset_t *setp, *osetp;
1188 	int error;
1189 
1190 	setp = (uap->set != NULL) ? &set : NULL;
1191 	osetp = (uap->oset != NULL) ? &oset : NULL;
1192 	if (setp) {
1193 		error = copyin(uap->set, setp, sizeof(set));
1194 		if (error)
1195 			return (error);
1196 	}
1197 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1198 	if (osetp && !error) {
1199 		error = copyout(osetp, uap->oset, sizeof(oset));
1200 	}
1201 	return (error);
1202 }
1203 
1204 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1205 #ifndef _SYS_SYSPROTO_H_
1206 struct osigprocmask_args {
1207 	int	how;
1208 	osigset_t mask;
1209 };
1210 #endif
1211 int
1212 osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1213 {
1214 	sigset_t set, oset;
1215 	int error;
1216 
1217 	OSIG2SIG(uap->mask, set);
1218 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1219 	SIG2OSIG(oset, td->td_retval[0]);
1220 	return (error);
1221 }
1222 #endif /* COMPAT_43 */
1223 
1224 int
1225 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1226 {
1227 	ksiginfo_t ksi;
1228 	sigset_t set;
1229 	int error;
1230 
1231 	error = copyin(uap->set, &set, sizeof(set));
1232 	if (error) {
1233 		td->td_retval[0] = error;
1234 		return (0);
1235 	}
1236 
1237 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1238 	if (error) {
1239 		/*
1240 		 * sigwait() function shall not return EINTR, but
1241 		 * the syscall does.  Non-ancient libc provides the
1242 		 * wrapper which hides EINTR.  Otherwise, EINTR return
1243 		 * is used by libthr to handle required cancellation
1244 		 * point in the sigwait().
1245 		 */
1246 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1247 			return (ERESTART);
1248 		td->td_retval[0] = error;
1249 		return (0);
1250 	}
1251 
1252 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1253 	td->td_retval[0] = error;
1254 	return (0);
1255 }
1256 
1257 int
1258 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1259 {
1260 	struct timespec ts;
1261 	struct timespec *timeout;
1262 	sigset_t set;
1263 	ksiginfo_t ksi;
1264 	int error;
1265 
1266 	if (uap->timeout) {
1267 		error = copyin(uap->timeout, &ts, sizeof(ts));
1268 		if (error)
1269 			return (error);
1270 
1271 		timeout = &ts;
1272 	} else
1273 		timeout = NULL;
1274 
1275 	error = copyin(uap->set, &set, sizeof(set));
1276 	if (error)
1277 		return (error);
1278 
1279 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1280 	if (error)
1281 		return (error);
1282 
1283 	if (uap->info)
1284 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1285 
1286 	if (error == 0)
1287 		td->td_retval[0] = ksi.ksi_signo;
1288 	return (error);
1289 }
1290 
1291 int
1292 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1293 {
1294 	ksiginfo_t ksi;
1295 	sigset_t set;
1296 	int error;
1297 
1298 	error = copyin(uap->set, &set, sizeof(set));
1299 	if (error)
1300 		return (error);
1301 
1302 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1303 	if (error)
1304 		return (error);
1305 
1306 	if (uap->info)
1307 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1308 
1309 	if (error == 0)
1310 		td->td_retval[0] = ksi.ksi_signo;
1311 	return (error);
1312 }
1313 
1314 static void
1315 proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1316 {
1317 	struct thread *thr;
1318 
1319 	FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1320 		if (thr == td)
1321 			thr->td_si = *si;
1322 		else
1323 			thr->td_si.si_signo = 0;
1324 	}
1325 }
1326 
1327 int
1328 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1329 	struct timespec *timeout)
1330 {
1331 	struct sigacts *ps;
1332 	sigset_t saved_mask, new_block;
1333 	struct proc *p;
1334 	int error, sig, timevalid = 0;
1335 	sbintime_t sbt, precision, tsbt;
1336 	struct timespec ts;
1337 	bool traced;
1338 
1339 	p = td->td_proc;
1340 	error = 0;
1341 	traced = false;
1342 
1343 	/* Ensure the sigfastblock value is up to date. */
1344 	sigfastblock_fetch(td);
1345 
1346 	if (timeout != NULL) {
1347 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1348 			timevalid = 1;
1349 			ts = *timeout;
1350 			if (ts.tv_sec < INT32_MAX / 2) {
1351 				tsbt = tstosbt(ts);
1352 				precision = tsbt;
1353 				precision >>= tc_precexp;
1354 				if (TIMESEL(&sbt, tsbt))
1355 					sbt += tc_tick_sbt;
1356 				sbt += tsbt;
1357 			} else
1358 				precision = sbt = 0;
1359 		}
1360 	} else
1361 		precision = sbt = 0;
1362 	ksiginfo_init(ksi);
1363 	/* Some signals can not be waited for. */
1364 	SIG_CANTMASK(waitset);
1365 	ps = p->p_sigacts;
1366 	PROC_LOCK(p);
1367 	saved_mask = td->td_sigmask;
1368 	SIGSETNAND(td->td_sigmask, waitset);
1369 	if ((p->p_sysent->sv_flags & SV_SIG_DISCIGN) != 0 ||
1370 	    !kern_sig_discard_ign) {
1371 		thread_lock(td);
1372 		td->td_flags |= TDF_SIGWAIT;
1373 		thread_unlock(td);
1374 	}
1375 	for (;;) {
1376 		mtx_lock(&ps->ps_mtx);
1377 		sig = cursig(td);
1378 		mtx_unlock(&ps->ps_mtx);
1379 		KASSERT(sig >= 0, ("sig %d", sig));
1380 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1381 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1382 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1383 				error = 0;
1384 				break;
1385 			}
1386 		}
1387 
1388 		if (error != 0)
1389 			break;
1390 
1391 		/*
1392 		 * POSIX says this must be checked after looking for pending
1393 		 * signals.
1394 		 */
1395 		if (timeout != NULL && !timevalid) {
1396 			error = EINVAL;
1397 			break;
1398 		}
1399 
1400 		if (traced) {
1401 			error = EINTR;
1402 			break;
1403 		}
1404 
1405 		error = msleep_sbt(&p->p_sigacts, &p->p_mtx, PPAUSE | PCATCH,
1406 		    "sigwait", sbt, precision, C_ABSOLUTE);
1407 
1408 		/* The syscalls can not be restarted. */
1409 		if (error == ERESTART)
1410 			error = EINTR;
1411 
1412 		/*
1413 		 * If PTRACE_SCE or PTRACE_SCX were set after
1414 		 * userspace entered the syscall, return spurious
1415 		 * EINTR after wait was done.  Only do this as last
1416 		 * resort after rechecking for possible queued signals
1417 		 * and expired timeouts.
1418 		 */
1419 		if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1420 			traced = true;
1421 	}
1422 	thread_lock(td);
1423 	td->td_flags &= ~TDF_SIGWAIT;
1424 	thread_unlock(td);
1425 
1426 	new_block = saved_mask;
1427 	SIGSETNAND(new_block, td->td_sigmask);
1428 	td->td_sigmask = saved_mask;
1429 	/*
1430 	 * Fewer signals can be delivered to us, reschedule signal
1431 	 * notification.
1432 	 */
1433 	if (p->p_numthreads != 1)
1434 		reschedule_signals(p, new_block, 0);
1435 
1436 	if (error == 0) {
1437 		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1438 
1439 		if (ksi->ksi_code == SI_TIMER)
1440 			itimer_accept(p, ksi->ksi_timerid, ksi);
1441 
1442 #ifdef KTRACE
1443 		if (KTRPOINT(td, KTR_PSIG)) {
1444 			sig_t action;
1445 
1446 			mtx_lock(&ps->ps_mtx);
1447 			action = ps->ps_sigact[_SIG_IDX(sig)];
1448 			mtx_unlock(&ps->ps_mtx);
1449 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1450 		}
1451 #endif
1452 		if (sig == SIGKILL) {
1453 			proc_td_siginfo_capture(td, &ksi->ksi_info);
1454 			sigexit(td, sig);
1455 		}
1456 	}
1457 	PROC_UNLOCK(p);
1458 	return (error);
1459 }
1460 
1461 #ifndef _SYS_SYSPROTO_H_
1462 struct sigpending_args {
1463 	sigset_t	*set;
1464 };
1465 #endif
1466 int
1467 sys_sigpending(struct thread *td, struct sigpending_args *uap)
1468 {
1469 	struct proc *p = td->td_proc;
1470 	sigset_t pending;
1471 
1472 	PROC_LOCK(p);
1473 	pending = p->p_sigqueue.sq_signals;
1474 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1475 	PROC_UNLOCK(p);
1476 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1477 }
1478 
1479 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1480 #ifndef _SYS_SYSPROTO_H_
1481 struct osigpending_args {
1482 	int	dummy;
1483 };
1484 #endif
1485 int
1486 osigpending(struct thread *td, struct osigpending_args *uap)
1487 {
1488 	struct proc *p = td->td_proc;
1489 	sigset_t pending;
1490 
1491 	PROC_LOCK(p);
1492 	pending = p->p_sigqueue.sq_signals;
1493 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1494 	PROC_UNLOCK(p);
1495 	SIG2OSIG(pending, td->td_retval[0]);
1496 	return (0);
1497 }
1498 #endif /* COMPAT_43 */
1499 
1500 #if defined(COMPAT_43)
1501 /*
1502  * Generalized interface signal handler, 4.3-compatible.
1503  */
1504 #ifndef _SYS_SYSPROTO_H_
1505 struct osigvec_args {
1506 	int	signum;
1507 	struct	sigvec *nsv;
1508 	struct	sigvec *osv;
1509 };
1510 #endif
1511 /* ARGSUSED */
1512 int
1513 osigvec(struct thread *td, struct osigvec_args *uap)
1514 {
1515 	struct sigvec vec;
1516 	struct sigaction nsa, osa;
1517 	struct sigaction *nsap, *osap;
1518 	int error;
1519 
1520 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1521 		return (EINVAL);
1522 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1523 	osap = (uap->osv != NULL) ? &osa : NULL;
1524 	if (nsap) {
1525 		error = copyin(uap->nsv, &vec, sizeof(vec));
1526 		if (error)
1527 			return (error);
1528 		nsap->sa_handler = vec.sv_handler;
1529 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1530 		nsap->sa_flags = vec.sv_flags;
1531 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1532 	}
1533 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1534 	if (osap && !error) {
1535 		vec.sv_handler = osap->sa_handler;
1536 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1537 		vec.sv_flags = osap->sa_flags;
1538 		vec.sv_flags &= ~SA_NOCLDWAIT;
1539 		vec.sv_flags ^= SA_RESTART;
1540 		error = copyout(&vec, uap->osv, sizeof(vec));
1541 	}
1542 	return (error);
1543 }
1544 
1545 #ifndef _SYS_SYSPROTO_H_
1546 struct osigblock_args {
1547 	int	mask;
1548 };
1549 #endif
1550 int
1551 osigblock(struct thread *td, struct osigblock_args *uap)
1552 {
1553 	sigset_t set, oset;
1554 
1555 	OSIG2SIG(uap->mask, set);
1556 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1557 	SIG2OSIG(oset, td->td_retval[0]);
1558 	return (0);
1559 }
1560 
1561 #ifndef _SYS_SYSPROTO_H_
1562 struct osigsetmask_args {
1563 	int	mask;
1564 };
1565 #endif
1566 int
1567 osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1568 {
1569 	sigset_t set, oset;
1570 
1571 	OSIG2SIG(uap->mask, set);
1572 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1573 	SIG2OSIG(oset, td->td_retval[0]);
1574 	return (0);
1575 }
1576 #endif /* COMPAT_43 */
1577 
1578 /*
1579  * Suspend calling thread until signal, providing mask to be set in the
1580  * meantime.
1581  */
1582 #ifndef _SYS_SYSPROTO_H_
1583 struct sigsuspend_args {
1584 	const sigset_t *sigmask;
1585 };
1586 #endif
1587 /* ARGSUSED */
1588 int
1589 sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1590 {
1591 	sigset_t mask;
1592 	int error;
1593 
1594 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1595 	if (error)
1596 		return (error);
1597 	return (kern_sigsuspend(td, mask));
1598 }
1599 
1600 int
1601 kern_sigsuspend(struct thread *td, sigset_t mask)
1602 {
1603 	struct proc *p = td->td_proc;
1604 	int has_sig, sig;
1605 
1606 	/* Ensure the sigfastblock value is up to date. */
1607 	sigfastblock_fetch(td);
1608 
1609 	/*
1610 	 * When returning from sigsuspend, we want
1611 	 * the old mask to be restored after the
1612 	 * signal handler has finished.  Thus, we
1613 	 * save it here and mark the sigacts structure
1614 	 * to indicate this.
1615 	 */
1616 	PROC_LOCK(p);
1617 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1618 	    SIGPROCMASK_PROC_LOCKED);
1619 	td->td_pflags |= TDP_OLDMASK;
1620 	ast_sched(td, TDA_SIGSUSPEND);
1621 
1622 	/*
1623 	 * Process signals now. Otherwise, we can get spurious wakeup
1624 	 * due to signal entered process queue, but delivered to other
1625 	 * thread. But sigsuspend should return only on signal
1626 	 * delivery.
1627 	 */
1628 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1629 	for (has_sig = 0; !has_sig;) {
1630 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1631 			0) == 0)
1632 			/* void */;
1633 		thread_suspend_check(0);
1634 		mtx_lock(&p->p_sigacts->ps_mtx);
1635 		while ((sig = cursig(td)) != 0) {
1636 			KASSERT(sig >= 0, ("sig %d", sig));
1637 			has_sig += postsig(sig);
1638 		}
1639 		mtx_unlock(&p->p_sigacts->ps_mtx);
1640 
1641 		/*
1642 		 * If PTRACE_SCE or PTRACE_SCX were set after
1643 		 * userspace entered the syscall, return spurious
1644 		 * EINTR.
1645 		 */
1646 		if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1647 			has_sig += 1;
1648 	}
1649 	PROC_UNLOCK(p);
1650 	td->td_errno = EINTR;
1651 	td->td_pflags |= TDP_NERRNO;
1652 	return (EJUSTRETURN);
1653 }
1654 
1655 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1656 /*
1657  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1658  * convention: libc stub passes mask, not pointer, to save a copyin.
1659  */
1660 #ifndef _SYS_SYSPROTO_H_
1661 struct osigsuspend_args {
1662 	osigset_t mask;
1663 };
1664 #endif
1665 /* ARGSUSED */
1666 int
1667 osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1668 {
1669 	sigset_t mask;
1670 
1671 	OSIG2SIG(uap->mask, mask);
1672 	return (kern_sigsuspend(td, mask));
1673 }
1674 #endif /* COMPAT_43 */
1675 
1676 #if defined(COMPAT_43)
1677 #ifndef _SYS_SYSPROTO_H_
1678 struct osigstack_args {
1679 	struct	sigstack *nss;
1680 	struct	sigstack *oss;
1681 };
1682 #endif
1683 /* ARGSUSED */
1684 int
1685 osigstack(struct thread *td, struct osigstack_args *uap)
1686 {
1687 	struct sigstack nss, oss;
1688 	int error = 0;
1689 
1690 	if (uap->nss != NULL) {
1691 		error = copyin(uap->nss, &nss, sizeof(nss));
1692 		if (error)
1693 			return (error);
1694 	}
1695 	oss.ss_sp = td->td_sigstk.ss_sp;
1696 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1697 	if (uap->nss != NULL) {
1698 		td->td_sigstk.ss_sp = nss.ss_sp;
1699 		td->td_sigstk.ss_size = 0;
1700 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1701 		td->td_pflags |= TDP_ALTSTACK;
1702 	}
1703 	if (uap->oss != NULL)
1704 		error = copyout(&oss, uap->oss, sizeof(oss));
1705 
1706 	return (error);
1707 }
1708 #endif /* COMPAT_43 */
1709 
1710 #ifndef _SYS_SYSPROTO_H_
1711 struct sigaltstack_args {
1712 	stack_t	*ss;
1713 	stack_t	*oss;
1714 };
1715 #endif
1716 /* ARGSUSED */
1717 int
1718 sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1719 {
1720 	stack_t ss, oss;
1721 	int error;
1722 
1723 	if (uap->ss != NULL) {
1724 		error = copyin(uap->ss, &ss, sizeof(ss));
1725 		if (error)
1726 			return (error);
1727 	}
1728 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1729 	    (uap->oss != NULL) ? &oss : NULL);
1730 	if (error)
1731 		return (error);
1732 	if (uap->oss != NULL)
1733 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1734 	return (error);
1735 }
1736 
1737 int
1738 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1739 {
1740 	struct proc *p = td->td_proc;
1741 	int oonstack;
1742 
1743 	oonstack = sigonstack(cpu_getstack(td));
1744 
1745 	if (oss != NULL) {
1746 		*oss = td->td_sigstk;
1747 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1748 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1749 	}
1750 
1751 	if (ss != NULL) {
1752 		if (oonstack)
1753 			return (EPERM);
1754 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1755 			return (EINVAL);
1756 		if (!(ss->ss_flags & SS_DISABLE)) {
1757 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1758 				return (ENOMEM);
1759 
1760 			td->td_sigstk = *ss;
1761 			td->td_pflags |= TDP_ALTSTACK;
1762 		} else {
1763 			td->td_pflags &= ~TDP_ALTSTACK;
1764 		}
1765 	}
1766 	return (0);
1767 }
1768 
1769 struct killpg1_ctx {
1770 	struct thread *td;
1771 	ksiginfo_t *ksi;
1772 	int sig;
1773 	bool sent;
1774 	bool found;
1775 	int ret;
1776 };
1777 
1778 static void
1779 killpg1_sendsig_locked(struct proc *p, struct killpg1_ctx *arg)
1780 {
1781 	int err;
1782 
1783 	err = p_cansignal(arg->td, p, arg->sig);
1784 	if (err == 0 && arg->sig != 0)
1785 		pksignal(p, arg->sig, arg->ksi);
1786 	if (err != ESRCH)
1787 		arg->found = true;
1788 	if (err == 0)
1789 		arg->sent = true;
1790 	else if (arg->ret == 0 && err != ESRCH && err != EPERM)
1791 		arg->ret = err;
1792 }
1793 
1794 static void
1795 killpg1_sendsig(struct proc *p, bool notself, struct killpg1_ctx *arg)
1796 {
1797 
1798 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1799 	    (notself && p == arg->td->td_proc) || p->p_state == PRS_NEW)
1800 		return;
1801 
1802 	PROC_LOCK(p);
1803 	killpg1_sendsig_locked(p, arg);
1804 	PROC_UNLOCK(p);
1805 }
1806 
1807 static void
1808 kill_processes_prison_cb(struct proc *p, void *arg)
1809 {
1810 	struct killpg1_ctx *ctx = arg;
1811 
1812 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1813 	    (p == ctx->td->td_proc) || p->p_state == PRS_NEW)
1814 		return;
1815 
1816 	killpg1_sendsig_locked(p, ctx);
1817 }
1818 
1819 /*
1820  * Common code for kill process group/broadcast kill.
1821  * cp is calling process.
1822  */
1823 static int
1824 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1825 {
1826 	struct proc *p;
1827 	struct pgrp *pgrp;
1828 	struct killpg1_ctx arg;
1829 
1830 	arg.td = td;
1831 	arg.ksi = ksi;
1832 	arg.sig = sig;
1833 	arg.sent = false;
1834 	arg.found = false;
1835 	arg.ret = 0;
1836 	if (all) {
1837 		/*
1838 		 * broadcast
1839 		 */
1840 		prison_proc_iterate(td->td_ucred->cr_prison,
1841 		    kill_processes_prison_cb, &arg);
1842 	} else {
1843 		sx_slock(&proctree_lock);
1844 		if (pgid == 0) {
1845 			/*
1846 			 * zero pgid means send to my process group.
1847 			 */
1848 			pgrp = td->td_proc->p_pgrp;
1849 			PGRP_LOCK(pgrp);
1850 		} else {
1851 			pgrp = pgfind(pgid);
1852 			if (pgrp == NULL) {
1853 				sx_sunlock(&proctree_lock);
1854 				return (ESRCH);
1855 			}
1856 		}
1857 		sx_sunlock(&proctree_lock);
1858 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1859 			killpg1_sendsig(p, false, &arg);
1860 		}
1861 		PGRP_UNLOCK(pgrp);
1862 	}
1863 	MPASS(arg.ret != 0 || arg.found || !arg.sent);
1864 	if (arg.ret == 0 && !arg.sent)
1865 		arg.ret = arg.found ? EPERM : ESRCH;
1866 	return (arg.ret);
1867 }
1868 
1869 #ifndef _SYS_SYSPROTO_H_
1870 struct kill_args {
1871 	int	pid;
1872 	int	signum;
1873 };
1874 #endif
1875 /* ARGSUSED */
1876 int
1877 sys_kill(struct thread *td, struct kill_args *uap)
1878 {
1879 
1880 	return (kern_kill(td, uap->pid, uap->signum));
1881 }
1882 
1883 int
1884 kern_kill(struct thread *td, pid_t pid, int signum)
1885 {
1886 	ksiginfo_t ksi;
1887 	struct proc *p;
1888 	int error;
1889 
1890 	/*
1891 	 * A process in capability mode can send signals only to himself.
1892 	 * The main rationale behind this is that abort(3) is implemented as
1893 	 * kill(getpid(), SIGABRT).
1894 	 */
1895 	if (IN_CAPABILITY_MODE(td) && pid != td->td_proc->p_pid)
1896 		return (ECAPMODE);
1897 
1898 	AUDIT_ARG_SIGNUM(signum);
1899 	AUDIT_ARG_PID(pid);
1900 	if ((u_int)signum > _SIG_MAXSIG)
1901 		return (EINVAL);
1902 
1903 	ksiginfo_init(&ksi);
1904 	ksi.ksi_signo = signum;
1905 	ksi.ksi_code = SI_USER;
1906 	ksi.ksi_pid = td->td_proc->p_pid;
1907 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1908 
1909 	if (pid > 0) {
1910 		/* kill single process */
1911 		if ((p = pfind_any(pid)) == NULL)
1912 			return (ESRCH);
1913 		AUDIT_ARG_PROCESS(p);
1914 		error = p_cansignal(td, p, signum);
1915 		if (error == 0 && signum)
1916 			pksignal(p, signum, &ksi);
1917 		PROC_UNLOCK(p);
1918 		return (error);
1919 	}
1920 	switch (pid) {
1921 	case -1:		/* broadcast signal */
1922 		return (killpg1(td, signum, 0, 1, &ksi));
1923 	case 0:			/* signal own process group */
1924 		return (killpg1(td, signum, 0, 0, &ksi));
1925 	default:		/* negative explicit process group */
1926 		return (killpg1(td, signum, -pid, 0, &ksi));
1927 	}
1928 	/* NOTREACHED */
1929 }
1930 
1931 int
1932 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1933 {
1934 	struct proc *p;
1935 	int error;
1936 
1937 	AUDIT_ARG_SIGNUM(uap->signum);
1938 	AUDIT_ARG_FD(uap->fd);
1939 	if ((u_int)uap->signum > _SIG_MAXSIG)
1940 		return (EINVAL);
1941 
1942 	error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1943 	if (error)
1944 		return (error);
1945 	AUDIT_ARG_PROCESS(p);
1946 	error = p_cansignal(td, p, uap->signum);
1947 	if (error == 0 && uap->signum)
1948 		kern_psignal(p, uap->signum);
1949 	PROC_UNLOCK(p);
1950 	return (error);
1951 }
1952 
1953 #if defined(COMPAT_43)
1954 #ifndef _SYS_SYSPROTO_H_
1955 struct okillpg_args {
1956 	int	pgid;
1957 	int	signum;
1958 };
1959 #endif
1960 /* ARGSUSED */
1961 int
1962 okillpg(struct thread *td, struct okillpg_args *uap)
1963 {
1964 	ksiginfo_t ksi;
1965 
1966 	AUDIT_ARG_SIGNUM(uap->signum);
1967 	AUDIT_ARG_PID(uap->pgid);
1968 	if ((u_int)uap->signum > _SIG_MAXSIG)
1969 		return (EINVAL);
1970 
1971 	ksiginfo_init(&ksi);
1972 	ksi.ksi_signo = uap->signum;
1973 	ksi.ksi_code = SI_USER;
1974 	ksi.ksi_pid = td->td_proc->p_pid;
1975 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1976 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1977 }
1978 #endif /* COMPAT_43 */
1979 
1980 #ifndef _SYS_SYSPROTO_H_
1981 struct sigqueue_args {
1982 	pid_t pid;
1983 	int signum;
1984 	/* union sigval */ void *value;
1985 };
1986 #endif
1987 int
1988 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1989 {
1990 	union sigval sv;
1991 
1992 	sv.sival_ptr = uap->value;
1993 
1994 	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
1995 }
1996 
1997 int
1998 kern_sigqueue(struct thread *td, pid_t pid, int signum, union sigval *value)
1999 {
2000 	ksiginfo_t ksi;
2001 	struct proc *p;
2002 	int error;
2003 
2004 	if ((u_int)signum > _SIG_MAXSIG)
2005 		return (EINVAL);
2006 
2007 	/*
2008 	 * Specification says sigqueue can only send signal to
2009 	 * single process.
2010 	 */
2011 	if (pid <= 0)
2012 		return (EINVAL);
2013 
2014 	if ((p = pfind_any(pid)) == NULL)
2015 		return (ESRCH);
2016 	error = p_cansignal(td, p, signum);
2017 	if (error == 0 && signum != 0) {
2018 		ksiginfo_init(&ksi);
2019 		ksi.ksi_flags = KSI_SIGQ;
2020 		ksi.ksi_signo = signum;
2021 		ksi.ksi_code = SI_QUEUE;
2022 		ksi.ksi_pid = td->td_proc->p_pid;
2023 		ksi.ksi_uid = td->td_ucred->cr_ruid;
2024 		ksi.ksi_value = *value;
2025 		error = pksignal(p, ksi.ksi_signo, &ksi);
2026 	}
2027 	PROC_UNLOCK(p);
2028 	return (error);
2029 }
2030 
2031 /*
2032  * Send a signal to a process group.
2033  */
2034 void
2035 gsignal(int pgid, int sig, ksiginfo_t *ksi)
2036 {
2037 	struct pgrp *pgrp;
2038 
2039 	if (pgid != 0) {
2040 		sx_slock(&proctree_lock);
2041 		pgrp = pgfind(pgid);
2042 		sx_sunlock(&proctree_lock);
2043 		if (pgrp != NULL) {
2044 			pgsignal(pgrp, sig, 0, ksi);
2045 			PGRP_UNLOCK(pgrp);
2046 		}
2047 	}
2048 }
2049 
2050 /*
2051  * Send a signal to a process group.  If checktty is 1,
2052  * limit to members which have a controlling terminal.
2053  */
2054 void
2055 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
2056 {
2057 	struct proc *p;
2058 
2059 	if (pgrp) {
2060 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
2061 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
2062 			PROC_LOCK(p);
2063 			if (p->p_state == PRS_NORMAL &&
2064 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
2065 				pksignal(p, sig, ksi);
2066 			PROC_UNLOCK(p);
2067 		}
2068 	}
2069 }
2070 
2071 /*
2072  * Recalculate the signal mask and reset the signal disposition after
2073  * usermode frame for delivery is formed.  Should be called after
2074  * mach-specific routine, because sysent->sv_sendsig() needs correct
2075  * ps_siginfo and signal mask.
2076  */
2077 static void
2078 postsig_done(int sig, struct thread *td, struct sigacts *ps)
2079 {
2080 	sigset_t mask;
2081 
2082 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2083 	td->td_ru.ru_nsignals++;
2084 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
2085 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
2086 		SIGADDSET(mask, sig);
2087 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
2088 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
2089 	if (SIGISMEMBER(ps->ps_sigreset, sig))
2090 		sigdflt(ps, sig);
2091 }
2092 
2093 /*
2094  * Send a signal caused by a trap to the current thread.  If it will be
2095  * caught immediately, deliver it with correct code.  Otherwise, post it
2096  * normally.
2097  */
2098 void
2099 trapsignal(struct thread *td, ksiginfo_t *ksi)
2100 {
2101 	struct sigacts *ps;
2102 	struct proc *p;
2103 	sigset_t sigmask;
2104 	int sig;
2105 
2106 	p = td->td_proc;
2107 	sig = ksi->ksi_signo;
2108 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
2109 
2110 	sigfastblock_fetch(td);
2111 	PROC_LOCK(p);
2112 	ps = p->p_sigacts;
2113 	mtx_lock(&ps->ps_mtx);
2114 	sigmask = td->td_sigmask;
2115 	if (td->td_sigblock_val != 0)
2116 		SIGSETOR(sigmask, fastblock_mask);
2117 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2118 	    !SIGISMEMBER(sigmask, sig)) {
2119 #ifdef KTRACE
2120 		if (KTRPOINT(curthread, KTR_PSIG))
2121 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2122 			    &td->td_sigmask, ksi->ksi_code);
2123 #endif
2124 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2125 		    ksi, &td->td_sigmask);
2126 		postsig_done(sig, td, ps);
2127 		mtx_unlock(&ps->ps_mtx);
2128 	} else {
2129 		/*
2130 		 * Avoid a possible infinite loop if the thread
2131 		 * masking the signal or process is ignoring the
2132 		 * signal.
2133 		 */
2134 		if (kern_forcesigexit && (SIGISMEMBER(sigmask, sig) ||
2135 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2136 			SIGDELSET(td->td_sigmask, sig);
2137 			SIGDELSET(ps->ps_sigcatch, sig);
2138 			SIGDELSET(ps->ps_sigignore, sig);
2139 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2140 			td->td_pflags &= ~TDP_SIGFASTBLOCK;
2141 			td->td_sigblock_val = 0;
2142 		}
2143 		mtx_unlock(&ps->ps_mtx);
2144 		p->p_sig = sig;		/* XXX to verify code */
2145 		tdsendsignal(p, td, sig, ksi);
2146 	}
2147 	PROC_UNLOCK(p);
2148 }
2149 
2150 static struct thread *
2151 sigtd(struct proc *p, int sig, bool fast_sigblock)
2152 {
2153 	struct thread *td, *signal_td;
2154 
2155 	PROC_LOCK_ASSERT(p, MA_OWNED);
2156 	MPASS(!fast_sigblock || p == curproc);
2157 
2158 	/*
2159 	 * Check if current thread can handle the signal without
2160 	 * switching context to another thread.
2161 	 */
2162 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig) &&
2163 	    (!fast_sigblock || curthread->td_sigblock_val == 0))
2164 		return (curthread);
2165 	signal_td = NULL;
2166 	FOREACH_THREAD_IN_PROC(p, td) {
2167 		if (!SIGISMEMBER(td->td_sigmask, sig) && (!fast_sigblock ||
2168 		    td != curthread || td->td_sigblock_val == 0)) {
2169 			signal_td = td;
2170 			break;
2171 		}
2172 	}
2173 	if (signal_td == NULL)
2174 		signal_td = FIRST_THREAD_IN_PROC(p);
2175 	return (signal_td);
2176 }
2177 
2178 /*
2179  * Send the signal to the process.  If the signal has an action, the action
2180  * is usually performed by the target process rather than the caller; we add
2181  * the signal to the set of pending signals for the process.
2182  *
2183  * Exceptions:
2184  *   o When a stop signal is sent to a sleeping process that takes the
2185  *     default action, the process is stopped without awakening it.
2186  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2187  *     regardless of the signal action (eg, blocked or ignored).
2188  *
2189  * Other ignored signals are discarded immediately.
2190  *
2191  * NB: This function may be entered from the debugger via the "kill" DDB
2192  * command.  There is little that can be done to mitigate the possibly messy
2193  * side effects of this unwise possibility.
2194  */
2195 void
2196 kern_psignal(struct proc *p, int sig)
2197 {
2198 	ksiginfo_t ksi;
2199 
2200 	ksiginfo_init(&ksi);
2201 	ksi.ksi_signo = sig;
2202 	ksi.ksi_code = SI_KERNEL;
2203 	(void) tdsendsignal(p, NULL, sig, &ksi);
2204 }
2205 
2206 int
2207 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2208 {
2209 
2210 	return (tdsendsignal(p, NULL, sig, ksi));
2211 }
2212 
2213 /* Utility function for finding a thread to send signal event to. */
2214 int
2215 sigev_findtd(struct proc *p, struct sigevent *sigev, struct thread **ttd)
2216 {
2217 	struct thread *td;
2218 
2219 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2220 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2221 		if (td == NULL)
2222 			return (ESRCH);
2223 		*ttd = td;
2224 	} else {
2225 		*ttd = NULL;
2226 		PROC_LOCK(p);
2227 	}
2228 	return (0);
2229 }
2230 
2231 void
2232 tdsignal(struct thread *td, int sig)
2233 {
2234 	ksiginfo_t ksi;
2235 
2236 	ksiginfo_init(&ksi);
2237 	ksi.ksi_signo = sig;
2238 	ksi.ksi_code = SI_KERNEL;
2239 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2240 }
2241 
2242 void
2243 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2244 {
2245 
2246 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2247 }
2248 
2249 static int
2250 sig_sleepq_abort(struct thread *td, int intrval)
2251 {
2252 	THREAD_LOCK_ASSERT(td, MA_OWNED);
2253 
2254 	if (intrval == 0 && (td->td_flags & TDF_SIGWAIT) == 0) {
2255 		thread_unlock(td);
2256 		return (0);
2257 	}
2258 	return (sleepq_abort(td, intrval));
2259 }
2260 
2261 int
2262 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2263 {
2264 	sig_t action;
2265 	sigqueue_t *sigqueue;
2266 	int prop;
2267 	struct sigacts *ps;
2268 	int intrval;
2269 	int ret = 0;
2270 	int wakeup_swapper;
2271 
2272 	MPASS(td == NULL || p == td->td_proc);
2273 	PROC_LOCK_ASSERT(p, MA_OWNED);
2274 
2275 	if (!_SIG_VALID(sig))
2276 		panic("%s(): invalid signal %d", __func__, sig);
2277 
2278 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2279 
2280 	/*
2281 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2282 	 */
2283 	if (p->p_state == PRS_ZOMBIE) {
2284 		if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2285 			ksiginfo_tryfree(ksi);
2286 		return (ret);
2287 	}
2288 
2289 	ps = p->p_sigacts;
2290 	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2291 	prop = sigprop(sig);
2292 
2293 	if (td == NULL) {
2294 		td = sigtd(p, sig, false);
2295 		sigqueue = &p->p_sigqueue;
2296 	} else
2297 		sigqueue = &td->td_sigqueue;
2298 
2299 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2300 
2301 	/*
2302 	 * If the signal is being ignored, then we forget about it
2303 	 * immediately, except when the target process executes
2304 	 * sigwait().  (Note: we don't set SIGCONT in ps_sigignore,
2305 	 * and if it is set to SIG_IGN, action will be SIG_DFL here.)
2306 	 */
2307 	mtx_lock(&ps->ps_mtx);
2308 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2309 		if (kern_sig_discard_ign &&
2310 		    (p->p_sysent->sv_flags & SV_SIG_DISCIGN) == 0) {
2311 			SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2312 
2313 			mtx_unlock(&ps->ps_mtx);
2314 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2315 				ksiginfo_tryfree(ksi);
2316 			return (ret);
2317 		} else {
2318 			action = SIG_CATCH;
2319 			intrval = 0;
2320 		}
2321 	} else {
2322 		if (SIGISMEMBER(td->td_sigmask, sig))
2323 			action = SIG_HOLD;
2324 		else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2325 			action = SIG_CATCH;
2326 		else
2327 			action = SIG_DFL;
2328 		if (SIGISMEMBER(ps->ps_sigintr, sig))
2329 			intrval = EINTR;
2330 		else
2331 			intrval = ERESTART;
2332 	}
2333 	mtx_unlock(&ps->ps_mtx);
2334 
2335 	if (prop & SIGPROP_CONT)
2336 		sigqueue_delete_stopmask_proc(p);
2337 	else if (prop & SIGPROP_STOP) {
2338 		/*
2339 		 * If sending a tty stop signal to a member of an orphaned
2340 		 * process group, discard the signal here if the action
2341 		 * is default; don't stop the process below if sleeping,
2342 		 * and don't clear any pending SIGCONT.
2343 		 */
2344 		if ((prop & SIGPROP_TTYSTOP) != 0 &&
2345 		    (p->p_pgrp->pg_flags & PGRP_ORPHANED) != 0 &&
2346 		    action == SIG_DFL) {
2347 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2348 				ksiginfo_tryfree(ksi);
2349 			return (ret);
2350 		}
2351 		sigqueue_delete_proc(p, SIGCONT);
2352 		if (p->p_flag & P_CONTINUED) {
2353 			p->p_flag &= ~P_CONTINUED;
2354 			PROC_LOCK(p->p_pptr);
2355 			sigqueue_take(p->p_ksi);
2356 			PROC_UNLOCK(p->p_pptr);
2357 		}
2358 	}
2359 
2360 	ret = sigqueue_add(sigqueue, sig, ksi);
2361 	if (ret != 0)
2362 		return (ret);
2363 	signotify(td);
2364 	/*
2365 	 * Defer further processing for signals which are held,
2366 	 * except that stopped processes must be continued by SIGCONT.
2367 	 */
2368 	if (action == SIG_HOLD &&
2369 	    !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2370 		return (ret);
2371 
2372 	wakeup_swapper = 0;
2373 
2374 	/*
2375 	 * Some signals have a process-wide effect and a per-thread
2376 	 * component.  Most processing occurs when the process next
2377 	 * tries to cross the user boundary, however there are some
2378 	 * times when processing needs to be done immediately, such as
2379 	 * waking up threads so that they can cross the user boundary.
2380 	 * We try to do the per-process part here.
2381 	 */
2382 	if (P_SHOULDSTOP(p)) {
2383 		KASSERT(!(p->p_flag & P_WEXIT),
2384 		    ("signal to stopped but exiting process"));
2385 		if (sig == SIGKILL) {
2386 			/*
2387 			 * If traced process is already stopped,
2388 			 * then no further action is necessary.
2389 			 */
2390 			if (p->p_flag & P_TRACED)
2391 				goto out;
2392 			/*
2393 			 * SIGKILL sets process running.
2394 			 * It will die elsewhere.
2395 			 * All threads must be restarted.
2396 			 */
2397 			p->p_flag &= ~P_STOPPED_SIG;
2398 			goto runfast;
2399 		}
2400 
2401 		if (prop & SIGPROP_CONT) {
2402 			/*
2403 			 * If traced process is already stopped,
2404 			 * then no further action is necessary.
2405 			 */
2406 			if (p->p_flag & P_TRACED)
2407 				goto out;
2408 			/*
2409 			 * If SIGCONT is default (or ignored), we continue the
2410 			 * process but don't leave the signal in sigqueue as
2411 			 * it has no further action.  If SIGCONT is held, we
2412 			 * continue the process and leave the signal in
2413 			 * sigqueue.  If the process catches SIGCONT, let it
2414 			 * handle the signal itself.  If it isn't waiting on
2415 			 * an event, it goes back to run state.
2416 			 * Otherwise, process goes back to sleep state.
2417 			 */
2418 			p->p_flag &= ~P_STOPPED_SIG;
2419 			PROC_SLOCK(p);
2420 			if (p->p_numthreads == p->p_suspcount) {
2421 				PROC_SUNLOCK(p);
2422 				p->p_flag |= P_CONTINUED;
2423 				p->p_xsig = SIGCONT;
2424 				PROC_LOCK(p->p_pptr);
2425 				childproc_continued(p);
2426 				PROC_UNLOCK(p->p_pptr);
2427 				PROC_SLOCK(p);
2428 			}
2429 			if (action == SIG_DFL) {
2430 				thread_unsuspend(p);
2431 				PROC_SUNLOCK(p);
2432 				sigqueue_delete(sigqueue, sig);
2433 				goto out_cont;
2434 			}
2435 			if (action == SIG_CATCH) {
2436 				/*
2437 				 * The process wants to catch it so it needs
2438 				 * to run at least one thread, but which one?
2439 				 */
2440 				PROC_SUNLOCK(p);
2441 				goto runfast;
2442 			}
2443 			/*
2444 			 * The signal is not ignored or caught.
2445 			 */
2446 			thread_unsuspend(p);
2447 			PROC_SUNLOCK(p);
2448 			goto out_cont;
2449 		}
2450 
2451 		if (prop & SIGPROP_STOP) {
2452 			/*
2453 			 * If traced process is already stopped,
2454 			 * then no further action is necessary.
2455 			 */
2456 			if (p->p_flag & P_TRACED)
2457 				goto out;
2458 			/*
2459 			 * Already stopped, don't need to stop again
2460 			 * (If we did the shell could get confused).
2461 			 * Just make sure the signal STOP bit set.
2462 			 */
2463 			p->p_flag |= P_STOPPED_SIG;
2464 			sigqueue_delete(sigqueue, sig);
2465 			goto out;
2466 		}
2467 
2468 		/*
2469 		 * All other kinds of signals:
2470 		 * If a thread is sleeping interruptibly, simulate a
2471 		 * wakeup so that when it is continued it will be made
2472 		 * runnable and can look at the signal.  However, don't make
2473 		 * the PROCESS runnable, leave it stopped.
2474 		 * It may run a bit until it hits a thread_suspend_check().
2475 		 */
2476 		PROC_SLOCK(p);
2477 		thread_lock(td);
2478 		if (TD_CAN_ABORT(td))
2479 			wakeup_swapper = sig_sleepq_abort(td, intrval);
2480 		else
2481 			thread_unlock(td);
2482 		PROC_SUNLOCK(p);
2483 		goto out;
2484 		/*
2485 		 * Mutexes are short lived. Threads waiting on them will
2486 		 * hit thread_suspend_check() soon.
2487 		 */
2488 	} else if (p->p_state == PRS_NORMAL) {
2489 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2490 			tdsigwakeup(td, sig, action, intrval);
2491 			goto out;
2492 		}
2493 
2494 		MPASS(action == SIG_DFL);
2495 
2496 		if (prop & SIGPROP_STOP) {
2497 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2498 				goto out;
2499 			p->p_flag |= P_STOPPED_SIG;
2500 			p->p_xsig = sig;
2501 			PROC_SLOCK(p);
2502 			wakeup_swapper = sig_suspend_threads(td, p);
2503 			if (p->p_numthreads == p->p_suspcount) {
2504 				/*
2505 				 * only thread sending signal to another
2506 				 * process can reach here, if thread is sending
2507 				 * signal to its process, because thread does
2508 				 * not suspend itself here, p_numthreads
2509 				 * should never be equal to p_suspcount.
2510 				 */
2511 				thread_stopped(p);
2512 				PROC_SUNLOCK(p);
2513 				sigqueue_delete_proc(p, p->p_xsig);
2514 			} else
2515 				PROC_SUNLOCK(p);
2516 			goto out;
2517 		}
2518 	} else {
2519 		/* Not in "NORMAL" state. discard the signal. */
2520 		sigqueue_delete(sigqueue, sig);
2521 		goto out;
2522 	}
2523 
2524 	/*
2525 	 * The process is not stopped so we need to apply the signal to all the
2526 	 * running threads.
2527 	 */
2528 runfast:
2529 	tdsigwakeup(td, sig, action, intrval);
2530 	PROC_SLOCK(p);
2531 	thread_unsuspend(p);
2532 	PROC_SUNLOCK(p);
2533 out_cont:
2534 	itimer_proc_continue(p);
2535 	kqtimer_proc_continue(p);
2536 out:
2537 	/* If we jump here, proc slock should not be owned. */
2538 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2539 	if (wakeup_swapper)
2540 		kick_proc0();
2541 
2542 	return (ret);
2543 }
2544 
2545 /*
2546  * The force of a signal has been directed against a single
2547  * thread.  We need to see what we can do about knocking it
2548  * out of any sleep it may be in etc.
2549  */
2550 static void
2551 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2552 {
2553 	struct proc *p = td->td_proc;
2554 	int prop, wakeup_swapper;
2555 
2556 	PROC_LOCK_ASSERT(p, MA_OWNED);
2557 	prop = sigprop(sig);
2558 
2559 	PROC_SLOCK(p);
2560 	thread_lock(td);
2561 	/*
2562 	 * Bring the priority of a thread up if we want it to get
2563 	 * killed in this lifetime.  Be careful to avoid bumping the
2564 	 * priority of the idle thread, since we still allow to signal
2565 	 * kernel processes.
2566 	 */
2567 	if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2568 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2569 		sched_prio(td, PUSER);
2570 	if (TD_ON_SLEEPQ(td)) {
2571 		/*
2572 		 * If thread is sleeping uninterruptibly
2573 		 * we can't interrupt the sleep... the signal will
2574 		 * be noticed when the process returns through
2575 		 * trap() or syscall().
2576 		 */
2577 		if ((td->td_flags & TDF_SINTR) == 0)
2578 			goto out;
2579 		/*
2580 		 * If SIGCONT is default (or ignored) and process is
2581 		 * asleep, we are finished; the process should not
2582 		 * be awakened.
2583 		 */
2584 		if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2585 			thread_unlock(td);
2586 			PROC_SUNLOCK(p);
2587 			sigqueue_delete(&p->p_sigqueue, sig);
2588 			/*
2589 			 * It may be on either list in this state.
2590 			 * Remove from both for now.
2591 			 */
2592 			sigqueue_delete(&td->td_sigqueue, sig);
2593 			return;
2594 		}
2595 
2596 		/*
2597 		 * Don't awaken a sleeping thread for SIGSTOP if the
2598 		 * STOP signal is deferred.
2599 		 */
2600 		if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2601 		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2602 			goto out;
2603 
2604 		/*
2605 		 * Give low priority threads a better chance to run.
2606 		 */
2607 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2608 			sched_prio(td, PUSER);
2609 
2610 		wakeup_swapper = sig_sleepq_abort(td, intrval);
2611 		PROC_SUNLOCK(p);
2612 		if (wakeup_swapper)
2613 			kick_proc0();
2614 		return;
2615 	}
2616 
2617 	/*
2618 	 * Other states do nothing with the signal immediately,
2619 	 * other than kicking ourselves if we are running.
2620 	 * It will either never be noticed, or noticed very soon.
2621 	 */
2622 #ifdef SMP
2623 	if (TD_IS_RUNNING(td) && td != curthread)
2624 		forward_signal(td);
2625 #endif
2626 
2627 out:
2628 	PROC_SUNLOCK(p);
2629 	thread_unlock(td);
2630 }
2631 
2632 static void
2633 ptrace_coredump(struct thread *td)
2634 {
2635 	struct proc *p;
2636 	struct thr_coredump_req *tcq;
2637 	void *rl_cookie;
2638 
2639 	MPASS(td == curthread);
2640 	p = td->td_proc;
2641 	PROC_LOCK_ASSERT(p, MA_OWNED);
2642 	if ((td->td_dbgflags & TDB_COREDUMPRQ) == 0)
2643 		return;
2644 	KASSERT((p->p_flag & P_STOPPED_TRACE) != 0, ("not stopped"));
2645 
2646 	tcq = td->td_coredump;
2647 	KASSERT(tcq != NULL, ("td_coredump is NULL"));
2648 
2649 	if (p->p_sysent->sv_coredump == NULL) {
2650 		tcq->tc_error = ENOSYS;
2651 		goto wake;
2652 	}
2653 
2654 	PROC_UNLOCK(p);
2655 	rl_cookie = vn_rangelock_wlock(tcq->tc_vp, 0, OFF_MAX);
2656 
2657 	tcq->tc_error = p->p_sysent->sv_coredump(td, tcq->tc_vp,
2658 	    tcq->tc_limit, tcq->tc_flags);
2659 
2660 	vn_rangelock_unlock(tcq->tc_vp, rl_cookie);
2661 	PROC_LOCK(p);
2662 wake:
2663 	td->td_dbgflags &= ~TDB_COREDUMPRQ;
2664 	td->td_coredump = NULL;
2665 	wakeup(p);
2666 }
2667 
2668 static int
2669 sig_suspend_threads(struct thread *td, struct proc *p)
2670 {
2671 	struct thread *td2;
2672 	int wakeup_swapper;
2673 
2674 	PROC_LOCK_ASSERT(p, MA_OWNED);
2675 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2676 
2677 	wakeup_swapper = 0;
2678 	FOREACH_THREAD_IN_PROC(p, td2) {
2679 		thread_lock(td2);
2680 		ast_sched_locked(td2, TDA_SUSPEND);
2681 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2682 		    (td2->td_flags & TDF_SINTR)) {
2683 			if (td2->td_flags & TDF_SBDRY) {
2684 				/*
2685 				 * Once a thread is asleep with
2686 				 * TDF_SBDRY and without TDF_SERESTART
2687 				 * or TDF_SEINTR set, it should never
2688 				 * become suspended due to this check.
2689 				 */
2690 				KASSERT(!TD_IS_SUSPENDED(td2),
2691 				    ("thread with deferred stops suspended"));
2692 				if (TD_SBDRY_INTR(td2)) {
2693 					wakeup_swapper |= sleepq_abort(td2,
2694 					    TD_SBDRY_ERRNO(td2));
2695 					continue;
2696 				}
2697 			} else if (!TD_IS_SUSPENDED(td2))
2698 				thread_suspend_one(td2);
2699 		} else if (!TD_IS_SUSPENDED(td2)) {
2700 #ifdef SMP
2701 			if (TD_IS_RUNNING(td2) && td2 != td)
2702 				forward_signal(td2);
2703 #endif
2704 		}
2705 		thread_unlock(td2);
2706 	}
2707 	return (wakeup_swapper);
2708 }
2709 
2710 /*
2711  * Stop the process for an event deemed interesting to the debugger. If si is
2712  * non-NULL, this is a signal exchange; the new signal requested by the
2713  * debugger will be returned for handling. If si is NULL, this is some other
2714  * type of interesting event. The debugger may request a signal be delivered in
2715  * that case as well, however it will be deferred until it can be handled.
2716  */
2717 int
2718 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2719 {
2720 	struct proc *p = td->td_proc;
2721 	struct thread *td2;
2722 	ksiginfo_t ksi;
2723 
2724 	PROC_LOCK_ASSERT(p, MA_OWNED);
2725 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2726 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2727 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2728 
2729 	td->td_xsig = sig;
2730 
2731 	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2732 		td->td_dbgflags |= TDB_XSIG;
2733 		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2734 		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2735 		PROC_SLOCK(p);
2736 		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2737 			if (P_KILLED(p)) {
2738 				/*
2739 				 * Ensure that, if we've been PT_KILLed, the
2740 				 * exit status reflects that. Another thread
2741 				 * may also be in ptracestop(), having just
2742 				 * received the SIGKILL, but this thread was
2743 				 * unsuspended first.
2744 				 */
2745 				td->td_dbgflags &= ~TDB_XSIG;
2746 				td->td_xsig = SIGKILL;
2747 				p->p_ptevents = 0;
2748 				break;
2749 			}
2750 			if (p->p_flag & P_SINGLE_EXIT &&
2751 			    !(td->td_dbgflags & TDB_EXIT)) {
2752 				/*
2753 				 * Ignore ptrace stops except for thread exit
2754 				 * events when the process exits.
2755 				 */
2756 				td->td_dbgflags &= ~TDB_XSIG;
2757 				PROC_SUNLOCK(p);
2758 				return (0);
2759 			}
2760 
2761 			/*
2762 			 * Make wait(2) work.  Ensure that right after the
2763 			 * attach, the thread which was decided to become the
2764 			 * leader of attach gets reported to the waiter.
2765 			 * Otherwise, just avoid overwriting another thread's
2766 			 * assignment to p_xthread.  If another thread has
2767 			 * already set p_xthread, the current thread will get
2768 			 * a chance to report itself upon the next iteration.
2769 			 */
2770 			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2771 			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2772 			    p->p_xthread == NULL)) {
2773 				p->p_xsig = sig;
2774 				p->p_xthread = td;
2775 
2776 				/*
2777 				 * If we are on sleepqueue already,
2778 				 * let sleepqueue code decide if it
2779 				 * needs to go sleep after attach.
2780 				 */
2781 				if (td->td_wchan == NULL)
2782 					td->td_dbgflags &= ~TDB_FSTP;
2783 
2784 				p->p_flag2 &= ~P2_PTRACE_FSTP;
2785 				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2786 				sig_suspend_threads(td, p);
2787 			}
2788 			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2789 				td->td_dbgflags &= ~TDB_STOPATFORK;
2790 			}
2791 stopme:
2792 			td->td_dbgflags |= TDB_SSWITCH;
2793 			thread_suspend_switch(td, p);
2794 			td->td_dbgflags &= ~TDB_SSWITCH;
2795 			if ((td->td_dbgflags & TDB_COREDUMPRQ) != 0) {
2796 				PROC_SUNLOCK(p);
2797 				ptrace_coredump(td);
2798 				PROC_SLOCK(p);
2799 				goto stopme;
2800 			}
2801 			if (p->p_xthread == td)
2802 				p->p_xthread = NULL;
2803 			if (!(p->p_flag & P_TRACED))
2804 				break;
2805 			if (td->td_dbgflags & TDB_SUSPEND) {
2806 				if (p->p_flag & P_SINGLE_EXIT)
2807 					break;
2808 				goto stopme;
2809 			}
2810 		}
2811 		PROC_SUNLOCK(p);
2812 	}
2813 
2814 	if (si != NULL && sig == td->td_xsig) {
2815 		/* Parent wants us to take the original signal unchanged. */
2816 		si->ksi_flags |= KSI_HEAD;
2817 		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2818 			si->ksi_signo = 0;
2819 	} else if (td->td_xsig != 0) {
2820 		/*
2821 		 * If parent wants us to take a new signal, then it will leave
2822 		 * it in td->td_xsig; otherwise we just look for signals again.
2823 		 */
2824 		ksiginfo_init(&ksi);
2825 		ksi.ksi_signo = td->td_xsig;
2826 		ksi.ksi_flags |= KSI_PTRACE;
2827 		td2 = sigtd(p, td->td_xsig, false);
2828 		tdsendsignal(p, td2, td->td_xsig, &ksi);
2829 		if (td != td2)
2830 			return (0);
2831 	}
2832 
2833 	return (td->td_xsig);
2834 }
2835 
2836 static void
2837 reschedule_signals(struct proc *p, sigset_t block, int flags)
2838 {
2839 	struct sigacts *ps;
2840 	struct thread *td;
2841 	int sig;
2842 	bool fastblk, pslocked;
2843 
2844 	PROC_LOCK_ASSERT(p, MA_OWNED);
2845 	ps = p->p_sigacts;
2846 	pslocked = (flags & SIGPROCMASK_PS_LOCKED) != 0;
2847 	mtx_assert(&ps->ps_mtx, pslocked ? MA_OWNED : MA_NOTOWNED);
2848 	if (SIGISEMPTY(p->p_siglist))
2849 		return;
2850 	SIGSETAND(block, p->p_siglist);
2851 	fastblk = (flags & SIGPROCMASK_FASTBLK) != 0;
2852 	SIG_FOREACH(sig, &block) {
2853 		td = sigtd(p, sig, fastblk);
2854 
2855 		/*
2856 		 * If sigtd() selected us despite sigfastblock is
2857 		 * blocking, do not activate AST or wake us, to avoid
2858 		 * loop in AST handler.
2859 		 */
2860 		if (fastblk && td == curthread)
2861 			continue;
2862 
2863 		signotify(td);
2864 		if (!pslocked)
2865 			mtx_lock(&ps->ps_mtx);
2866 		if (p->p_flag & P_TRACED ||
2867 		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2868 		    !SIGISMEMBER(td->td_sigmask, sig))) {
2869 			tdsigwakeup(td, sig, SIG_CATCH,
2870 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2871 			    ERESTART));
2872 		}
2873 		if (!pslocked)
2874 			mtx_unlock(&ps->ps_mtx);
2875 	}
2876 }
2877 
2878 void
2879 tdsigcleanup(struct thread *td)
2880 {
2881 	struct proc *p;
2882 	sigset_t unblocked;
2883 
2884 	p = td->td_proc;
2885 	PROC_LOCK_ASSERT(p, MA_OWNED);
2886 
2887 	sigqueue_flush(&td->td_sigqueue);
2888 	if (p->p_numthreads == 1)
2889 		return;
2890 
2891 	/*
2892 	 * Since we cannot handle signals, notify signal post code
2893 	 * about this by filling the sigmask.
2894 	 *
2895 	 * Also, if needed, wake up thread(s) that do not block the
2896 	 * same signals as the exiting thread, since the thread might
2897 	 * have been selected for delivery and woken up.
2898 	 */
2899 	SIGFILLSET(unblocked);
2900 	SIGSETNAND(unblocked, td->td_sigmask);
2901 	SIGFILLSET(td->td_sigmask);
2902 	reschedule_signals(p, unblocked, 0);
2903 
2904 }
2905 
2906 static int
2907 sigdeferstop_curr_flags(int cflags)
2908 {
2909 
2910 	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
2911 	    (cflags & TDF_SBDRY) != 0);
2912 	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
2913 }
2914 
2915 /*
2916  * Defer the delivery of SIGSTOP for the current thread, according to
2917  * the requested mode.  Returns previous flags, which must be restored
2918  * by sigallowstop().
2919  *
2920  * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
2921  * cleared by the current thread, which allow the lock-less read-only
2922  * accesses below.
2923  */
2924 int
2925 sigdeferstop_impl(int mode)
2926 {
2927 	struct thread *td;
2928 	int cflags, nflags;
2929 
2930 	td = curthread;
2931 	cflags = sigdeferstop_curr_flags(td->td_flags);
2932 	switch (mode) {
2933 	case SIGDEFERSTOP_NOP:
2934 		nflags = cflags;
2935 		break;
2936 	case SIGDEFERSTOP_OFF:
2937 		nflags = 0;
2938 		break;
2939 	case SIGDEFERSTOP_SILENT:
2940 		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
2941 		break;
2942 	case SIGDEFERSTOP_EINTR:
2943 		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
2944 		break;
2945 	case SIGDEFERSTOP_ERESTART:
2946 		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
2947 		break;
2948 	default:
2949 		panic("sigdeferstop: invalid mode %x", mode);
2950 		break;
2951 	}
2952 	if (cflags == nflags)
2953 		return (SIGDEFERSTOP_VAL_NCHG);
2954 	thread_lock(td);
2955 	td->td_flags = (td->td_flags & ~cflags) | nflags;
2956 	thread_unlock(td);
2957 	return (cflags);
2958 }
2959 
2960 /*
2961  * Restores the STOP handling mode, typically permitting the delivery
2962  * of SIGSTOP for the current thread.  This does not immediately
2963  * suspend if a stop was posted.  Instead, the thread will suspend
2964  * either via ast() or a subsequent interruptible sleep.
2965  */
2966 void
2967 sigallowstop_impl(int prev)
2968 {
2969 	struct thread *td;
2970 	int cflags;
2971 
2972 	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
2973 	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
2974 	    ("sigallowstop: incorrect previous mode %x", prev));
2975 	td = curthread;
2976 	cflags = sigdeferstop_curr_flags(td->td_flags);
2977 	if (cflags != prev) {
2978 		thread_lock(td);
2979 		td->td_flags = (td->td_flags & ~cflags) | prev;
2980 		thread_unlock(td);
2981 	}
2982 }
2983 
2984 enum sigstatus {
2985 	SIGSTATUS_HANDLE,
2986 	SIGSTATUS_HANDLED,
2987 	SIGSTATUS_IGNORE,
2988 	SIGSTATUS_SBDRY_STOP,
2989 };
2990 
2991 /*
2992  * The thread has signal "sig" pending.  Figure out what to do with it:
2993  *
2994  * _HANDLE     -> the caller should handle the signal
2995  * _HANDLED    -> handled internally, reload pending signal set
2996  * _IGNORE     -> ignored, remove from the set of pending signals and try the
2997  *                next pending signal
2998  * _SBDRY_STOP -> the signal should stop the thread but this is not
2999  *                permitted in the current context
3000  */
3001 static enum sigstatus
3002 sigprocess(struct thread *td, int sig)
3003 {
3004 	struct proc *p;
3005 	struct sigacts *ps;
3006 	struct sigqueue *queue;
3007 	ksiginfo_t ksi;
3008 	int prop;
3009 
3010 	KASSERT(_SIG_VALID(sig), ("%s: invalid signal %d", __func__, sig));
3011 
3012 	p = td->td_proc;
3013 	ps = p->p_sigacts;
3014 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3015 	PROC_LOCK_ASSERT(p, MA_OWNED);
3016 
3017 	/*
3018 	 * We should allow pending but ignored signals below
3019 	 * if there is sigwait() active, or P_TRACED was
3020 	 * on when they were posted.
3021 	 */
3022 	if (SIGISMEMBER(ps->ps_sigignore, sig) &&
3023 	    (p->p_flag & P_TRACED) == 0 &&
3024 	    (td->td_flags & TDF_SIGWAIT) == 0) {
3025 		return (SIGSTATUS_IGNORE);
3026 	}
3027 
3028 	/*
3029 	 * If the process is going to single-thread mode to prepare
3030 	 * for exit, there is no sense in delivering any signal
3031 	 * to usermode.  Another important consequence is that
3032 	 * msleep(..., PCATCH, ...) now is only interruptible by a
3033 	 * suspend request.
3034 	 */
3035 	if ((p->p_flag2 & P2_WEXIT) != 0)
3036 		return (SIGSTATUS_IGNORE);
3037 
3038 	if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
3039 		/*
3040 		 * If traced, always stop.
3041 		 * Remove old signal from queue before the stop.
3042 		 * XXX shrug off debugger, it causes siginfo to
3043 		 * be thrown away.
3044 		 */
3045 		queue = &td->td_sigqueue;
3046 		ksiginfo_init(&ksi);
3047 		if (sigqueue_get(queue, sig, &ksi) == 0) {
3048 			queue = &p->p_sigqueue;
3049 			sigqueue_get(queue, sig, &ksi);
3050 		}
3051 		td->td_si = ksi.ksi_info;
3052 
3053 		mtx_unlock(&ps->ps_mtx);
3054 		sig = ptracestop(td, sig, &ksi);
3055 		mtx_lock(&ps->ps_mtx);
3056 
3057 		td->td_si.si_signo = 0;
3058 
3059 		/*
3060 		 * Keep looking if the debugger discarded or
3061 		 * replaced the signal.
3062 		 */
3063 		if (sig == 0)
3064 			return (SIGSTATUS_HANDLED);
3065 
3066 		/*
3067 		 * If the signal became masked, re-queue it.
3068 		 */
3069 		if (SIGISMEMBER(td->td_sigmask, sig)) {
3070 			ksi.ksi_flags |= KSI_HEAD;
3071 			sigqueue_add(&p->p_sigqueue, sig, &ksi);
3072 			return (SIGSTATUS_HANDLED);
3073 		}
3074 
3075 		/*
3076 		 * If the traced bit got turned off, requeue the signal and
3077 		 * reload the set of pending signals.  This ensures that p_sig*
3078 		 * and p_sigact are consistent.
3079 		 */
3080 		if ((p->p_flag & P_TRACED) == 0) {
3081 			if ((ksi.ksi_flags & KSI_PTRACE) == 0) {
3082 				ksi.ksi_flags |= KSI_HEAD;
3083 				sigqueue_add(queue, sig, &ksi);
3084 			}
3085 			return (SIGSTATUS_HANDLED);
3086 		}
3087 	}
3088 
3089 	/*
3090 	 * Decide whether the signal should be returned.
3091 	 * Return the signal's number, or fall through
3092 	 * to clear it from the pending mask.
3093 	 */
3094 	switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
3095 	case (intptr_t)SIG_DFL:
3096 		/*
3097 		 * Don't take default actions on system processes.
3098 		 */
3099 		if (p->p_pid <= 1) {
3100 #ifdef DIAGNOSTIC
3101 			/*
3102 			 * Are you sure you want to ignore SIGSEGV
3103 			 * in init? XXX
3104 			 */
3105 			printf("Process (pid %lu) got signal %d\n",
3106 				(u_long)p->p_pid, sig);
3107 #endif
3108 			return (SIGSTATUS_IGNORE);
3109 		}
3110 
3111 		/*
3112 		 * If there is a pending stop signal to process with
3113 		 * default action, stop here, then clear the signal.
3114 		 * Traced or exiting processes should ignore stops.
3115 		 * Additionally, a member of an orphaned process group
3116 		 * should ignore tty stops.
3117 		 */
3118 		prop = sigprop(sig);
3119 		if (prop & SIGPROP_STOP) {
3120 			mtx_unlock(&ps->ps_mtx);
3121 			if ((p->p_flag & (P_TRACED | P_WEXIT |
3122 			    P_SINGLE_EXIT)) != 0 || ((p->p_pgrp->
3123 			    pg_flags & PGRP_ORPHANED) != 0 &&
3124 			    (prop & SIGPROP_TTYSTOP) != 0)) {
3125 				mtx_lock(&ps->ps_mtx);
3126 				return (SIGSTATUS_IGNORE);
3127 			}
3128 			if (TD_SBDRY_INTR(td)) {
3129 				KASSERT((td->td_flags & TDF_SBDRY) != 0,
3130 				    ("lost TDF_SBDRY"));
3131 				mtx_lock(&ps->ps_mtx);
3132 				return (SIGSTATUS_SBDRY_STOP);
3133 			}
3134 			WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
3135 			    &p->p_mtx.lock_object, "Catching SIGSTOP");
3136 			sigqueue_delete(&td->td_sigqueue, sig);
3137 			sigqueue_delete(&p->p_sigqueue, sig);
3138 			p->p_flag |= P_STOPPED_SIG;
3139 			p->p_xsig = sig;
3140 			PROC_SLOCK(p);
3141 			sig_suspend_threads(td, p);
3142 			thread_suspend_switch(td, p);
3143 			PROC_SUNLOCK(p);
3144 			mtx_lock(&ps->ps_mtx);
3145 			return (SIGSTATUS_HANDLED);
3146 		} else if ((prop & SIGPROP_IGNORE) != 0 &&
3147 		    (td->td_flags & TDF_SIGWAIT) == 0) {
3148 			/*
3149 			 * Default action is to ignore; drop it if
3150 			 * not in kern_sigtimedwait().
3151 			 */
3152 			return (SIGSTATUS_IGNORE);
3153 		} else {
3154 			return (SIGSTATUS_HANDLE);
3155 		}
3156 
3157 	case (intptr_t)SIG_IGN:
3158 		if ((td->td_flags & TDF_SIGWAIT) == 0)
3159 			return (SIGSTATUS_IGNORE);
3160 		else
3161 			return (SIGSTATUS_HANDLE);
3162 
3163 	default:
3164 		/*
3165 		 * This signal has an action, let postsig() process it.
3166 		 */
3167 		return (SIGSTATUS_HANDLE);
3168 	}
3169 }
3170 
3171 /*
3172  * If the current process has received a signal (should be caught or cause
3173  * termination, should interrupt current syscall), return the signal number.
3174  * Stop signals with default action are processed immediately, then cleared;
3175  * they aren't returned.  This is checked after each entry to the system for
3176  * a syscall or trap (though this can usually be done without calling
3177  * issignal by checking the pending signal masks in cursig.) The normal call
3178  * sequence is
3179  *
3180  *	while (sig = cursig(curthread))
3181  *		postsig(sig);
3182  */
3183 static int
3184 issignal(struct thread *td)
3185 {
3186 	struct proc *p;
3187 	sigset_t sigpending;
3188 	int sig;
3189 
3190 	p = td->td_proc;
3191 	PROC_LOCK_ASSERT(p, MA_OWNED);
3192 
3193 	for (;;) {
3194 		sigpending = td->td_sigqueue.sq_signals;
3195 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
3196 		SIGSETNAND(sigpending, td->td_sigmask);
3197 
3198 		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
3199 		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
3200 			SIG_STOPSIGMASK(sigpending);
3201 		if (SIGISEMPTY(sigpending))	/* no signal to send */
3202 			return (0);
3203 
3204 		/*
3205 		 * Do fast sigblock if requested by usermode.  Since
3206 		 * we do know that there was a signal pending at this
3207 		 * point, set the FAST_SIGBLOCK_PEND as indicator for
3208 		 * usermode to perform a dummy call to
3209 		 * FAST_SIGBLOCK_UNBLOCK, which causes immediate
3210 		 * delivery of postponed pending signal.
3211 		 */
3212 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
3213 			if (td->td_sigblock_val != 0)
3214 				SIGSETNAND(sigpending, fastblock_mask);
3215 			if (SIGISEMPTY(sigpending)) {
3216 				td->td_pflags |= TDP_SIGFASTPENDING;
3217 				return (0);
3218 			}
3219 		}
3220 
3221 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
3222 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
3223 		    SIGISMEMBER(sigpending, SIGSTOP)) {
3224 			/*
3225 			 * If debugger just attached, always consume
3226 			 * SIGSTOP from ptrace(PT_ATTACH) first, to
3227 			 * execute the debugger attach ritual in
3228 			 * order.
3229 			 */
3230 			td->td_dbgflags |= TDB_FSTP;
3231 			SIGEMPTYSET(sigpending);
3232 			SIGADDSET(sigpending, SIGSTOP);
3233 		}
3234 
3235 		SIG_FOREACH(sig, &sigpending) {
3236 			switch (sigprocess(td, sig)) {
3237 			case SIGSTATUS_HANDLE:
3238 				return (sig);
3239 			case SIGSTATUS_HANDLED:
3240 				goto next;
3241 			case SIGSTATUS_IGNORE:
3242 				sigqueue_delete(&td->td_sigqueue, sig);
3243 				sigqueue_delete(&p->p_sigqueue, sig);
3244 				break;
3245 			case SIGSTATUS_SBDRY_STOP:
3246 				return (-1);
3247 			}
3248 		}
3249 next:;
3250 	}
3251 }
3252 
3253 void
3254 thread_stopped(struct proc *p)
3255 {
3256 	int n;
3257 
3258 	PROC_LOCK_ASSERT(p, MA_OWNED);
3259 	PROC_SLOCK_ASSERT(p, MA_OWNED);
3260 	n = p->p_suspcount;
3261 	if (p == curproc)
3262 		n++;
3263 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3264 		PROC_SUNLOCK(p);
3265 		p->p_flag &= ~P_WAITED;
3266 		PROC_LOCK(p->p_pptr);
3267 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
3268 			CLD_TRAPPED : CLD_STOPPED);
3269 		PROC_UNLOCK(p->p_pptr);
3270 		PROC_SLOCK(p);
3271 	}
3272 }
3273 
3274 /*
3275  * Take the action for the specified signal
3276  * from the current set of pending signals.
3277  */
3278 int
3279 postsig(int sig)
3280 {
3281 	struct thread *td;
3282 	struct proc *p;
3283 	struct sigacts *ps;
3284 	sig_t action;
3285 	ksiginfo_t ksi;
3286 	sigset_t returnmask;
3287 
3288 	KASSERT(sig != 0, ("postsig"));
3289 
3290 	td = curthread;
3291 	p = td->td_proc;
3292 	PROC_LOCK_ASSERT(p, MA_OWNED);
3293 	ps = p->p_sigacts;
3294 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3295 	ksiginfo_init(&ksi);
3296 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3297 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3298 		return (0);
3299 	ksi.ksi_signo = sig;
3300 	if (ksi.ksi_code == SI_TIMER)
3301 		itimer_accept(p, ksi.ksi_timerid, &ksi);
3302 	action = ps->ps_sigact[_SIG_IDX(sig)];
3303 #ifdef KTRACE
3304 	if (KTRPOINT(td, KTR_PSIG))
3305 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3306 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3307 #endif
3308 
3309 	if (action == SIG_DFL) {
3310 		/*
3311 		 * Default action, where the default is to kill
3312 		 * the process.  (Other cases were ignored above.)
3313 		 */
3314 		mtx_unlock(&ps->ps_mtx);
3315 		proc_td_siginfo_capture(td, &ksi.ksi_info);
3316 		sigexit(td, sig);
3317 		/* NOTREACHED */
3318 	} else {
3319 		/*
3320 		 * If we get here, the signal must be caught.
3321 		 */
3322 		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3323 		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3324 		    ("postsig action: blocked sig %d", sig));
3325 
3326 		/*
3327 		 * Set the new mask value and also defer further
3328 		 * occurrences of this signal.
3329 		 *
3330 		 * Special case: user has done a sigsuspend.  Here the
3331 		 * current mask is not of interest, but rather the
3332 		 * mask from before the sigsuspend is what we want
3333 		 * restored after the signal processing is completed.
3334 		 */
3335 		if (td->td_pflags & TDP_OLDMASK) {
3336 			returnmask = td->td_oldsigmask;
3337 			td->td_pflags &= ~TDP_OLDMASK;
3338 		} else
3339 			returnmask = td->td_sigmask;
3340 
3341 		if (p->p_sig == sig) {
3342 			p->p_sig = 0;
3343 		}
3344 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3345 		postsig_done(sig, td, ps);
3346 	}
3347 	return (1);
3348 }
3349 
3350 int
3351 sig_ast_checksusp(struct thread *td)
3352 {
3353 	struct proc *p __diagused;
3354 	int ret;
3355 
3356 	p = td->td_proc;
3357 	PROC_LOCK_ASSERT(p, MA_OWNED);
3358 
3359 	if (!td_ast_pending(td, TDA_SUSPEND))
3360 		return (0);
3361 
3362 	ret = thread_suspend_check(1);
3363 	MPASS(ret == 0 || ret == EINTR || ret == ERESTART);
3364 	return (ret);
3365 }
3366 
3367 int
3368 sig_ast_needsigchk(struct thread *td)
3369 {
3370 	struct proc *p;
3371 	struct sigacts *ps;
3372 	int ret, sig;
3373 
3374 	p = td->td_proc;
3375 	PROC_LOCK_ASSERT(p, MA_OWNED);
3376 
3377 	if (!td_ast_pending(td, TDA_SIG))
3378 		return (0);
3379 
3380 	ps = p->p_sigacts;
3381 	mtx_lock(&ps->ps_mtx);
3382 	sig = cursig(td);
3383 	if (sig == -1) {
3384 		mtx_unlock(&ps->ps_mtx);
3385 		KASSERT((td->td_flags & TDF_SBDRY) != 0, ("lost TDF_SBDRY"));
3386 		KASSERT(TD_SBDRY_INTR(td),
3387 		    ("lost TDF_SERESTART of TDF_SEINTR"));
3388 		KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
3389 		    (TDF_SEINTR | TDF_SERESTART),
3390 		    ("both TDF_SEINTR and TDF_SERESTART"));
3391 		ret = TD_SBDRY_ERRNO(td);
3392 	} else if (sig != 0) {
3393 		ret = SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR : ERESTART;
3394 		mtx_unlock(&ps->ps_mtx);
3395 	} else {
3396 		mtx_unlock(&ps->ps_mtx);
3397 		ret = 0;
3398 	}
3399 
3400 	/*
3401 	 * Do not go into sleep if this thread was the ptrace(2)
3402 	 * attach leader.  cursig() consumed SIGSTOP from PT_ATTACH,
3403 	 * but we usually act on the signal by interrupting sleep, and
3404 	 * should do that here as well.
3405 	 */
3406 	if ((td->td_dbgflags & TDB_FSTP) != 0) {
3407 		if (ret == 0)
3408 			ret = EINTR;
3409 		td->td_dbgflags &= ~TDB_FSTP;
3410 	}
3411 
3412 	return (ret);
3413 }
3414 
3415 int
3416 sig_intr(void)
3417 {
3418 	struct thread *td;
3419 	struct proc *p;
3420 	int ret;
3421 
3422 	td = curthread;
3423 	if (!td_ast_pending(td, TDA_SIG) && !td_ast_pending(td, TDA_SUSPEND))
3424 		return (0);
3425 
3426 	p = td->td_proc;
3427 
3428 	PROC_LOCK(p);
3429 	ret = sig_ast_checksusp(td);
3430 	if (ret == 0)
3431 		ret = sig_ast_needsigchk(td);
3432 	PROC_UNLOCK(p);
3433 	return (ret);
3434 }
3435 
3436 bool
3437 curproc_sigkilled(void)
3438 {
3439 	struct thread *td;
3440 	struct proc *p;
3441 	struct sigacts *ps;
3442 	bool res;
3443 
3444 	td = curthread;
3445 	if (!td_ast_pending(td, TDA_SIG))
3446 		return (false);
3447 
3448 	p = td->td_proc;
3449 	PROC_LOCK(p);
3450 	ps = p->p_sigacts;
3451 	mtx_lock(&ps->ps_mtx);
3452 	res = SIGISMEMBER(td->td_sigqueue.sq_signals, SIGKILL) ||
3453 	    SIGISMEMBER(p->p_sigqueue.sq_signals, SIGKILL);
3454 	mtx_unlock(&ps->ps_mtx);
3455 	PROC_UNLOCK(p);
3456 	return (res);
3457 }
3458 
3459 void
3460 proc_wkilled(struct proc *p)
3461 {
3462 
3463 	PROC_LOCK_ASSERT(p, MA_OWNED);
3464 	if ((p->p_flag & P_WKILLED) == 0) {
3465 		p->p_flag |= P_WKILLED;
3466 		/*
3467 		 * Notify swapper that there is a process to swap in.
3468 		 * The notification is racy, at worst it would take 10
3469 		 * seconds for the swapper process to notice.
3470 		 */
3471 		if ((p->p_flag & (P_INMEM | P_SWAPPINGIN)) == 0)
3472 			wakeup(&proc0);
3473 	}
3474 }
3475 
3476 /*
3477  * Kill the current process for stated reason.
3478  */
3479 void
3480 killproc(struct proc *p, const char *why)
3481 {
3482 
3483 	PROC_LOCK_ASSERT(p, MA_OWNED);
3484 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3485 	    p->p_comm);
3486 	log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3487 	    p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3488 	    p->p_ucred->cr_uid, why);
3489 	proc_wkilled(p);
3490 	kern_psignal(p, SIGKILL);
3491 }
3492 
3493 /*
3494  * Force the current process to exit with the specified signal, dumping core
3495  * if appropriate.  We bypass the normal tests for masked and caught signals,
3496  * allowing unrecoverable failures to terminate the process without changing
3497  * signal state.  Mark the accounting record with the signal termination.
3498  * If dumping core, save the signal number for the debugger.  Calls exit and
3499  * does not return.
3500  */
3501 void
3502 sigexit(struct thread *td, int sig)
3503 {
3504 	struct proc *p = td->td_proc;
3505 
3506 	PROC_LOCK_ASSERT(p, MA_OWNED);
3507 	proc_set_p2_wexit(p);
3508 
3509 	p->p_acflag |= AXSIG;
3510 	/*
3511 	 * We must be single-threading to generate a core dump.  This
3512 	 * ensures that the registers in the core file are up-to-date.
3513 	 * Also, the ELF dump handler assumes that the thread list doesn't
3514 	 * change out from under it.
3515 	 *
3516 	 * XXX If another thread attempts to single-thread before us
3517 	 *     (e.g. via fork()), we won't get a dump at all.
3518 	 */
3519 	if ((sigprop(sig) & SIGPROP_CORE) &&
3520 	    thread_single(p, SINGLE_NO_EXIT) == 0) {
3521 		p->p_sig = sig;
3522 		/*
3523 		 * Log signals which would cause core dumps
3524 		 * (Log as LOG_INFO to appease those who don't want
3525 		 * these messages.)
3526 		 * XXX : Todo, as well as euid, write out ruid too
3527 		 * Note that coredump() drops proc lock.
3528 		 */
3529 		if (coredump(td) == 0)
3530 			sig |= WCOREFLAG;
3531 		if (kern_logsigexit)
3532 			log(LOG_INFO,
3533 			    "pid %d (%s), jid %d, uid %d: exited on "
3534 			    "signal %d%s\n", p->p_pid, p->p_comm,
3535 			    p->p_ucred->cr_prison->pr_id,
3536 			    td->td_ucred->cr_uid,
3537 			    sig &~ WCOREFLAG,
3538 			    sig & WCOREFLAG ? " (core dumped)" : "");
3539 	} else
3540 		PROC_UNLOCK(p);
3541 	exit1(td, 0, sig);
3542 	/* NOTREACHED */
3543 }
3544 
3545 /*
3546  * Send queued SIGCHLD to parent when child process's state
3547  * is changed.
3548  */
3549 static void
3550 sigparent(struct proc *p, int reason, int status)
3551 {
3552 	PROC_LOCK_ASSERT(p, MA_OWNED);
3553 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3554 
3555 	if (p->p_ksi != NULL) {
3556 		p->p_ksi->ksi_signo  = SIGCHLD;
3557 		p->p_ksi->ksi_code   = reason;
3558 		p->p_ksi->ksi_status = status;
3559 		p->p_ksi->ksi_pid    = p->p_pid;
3560 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3561 		if (KSI_ONQ(p->p_ksi))
3562 			return;
3563 	}
3564 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3565 }
3566 
3567 static void
3568 childproc_jobstate(struct proc *p, int reason, int sig)
3569 {
3570 	struct sigacts *ps;
3571 
3572 	PROC_LOCK_ASSERT(p, MA_OWNED);
3573 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3574 
3575 	/*
3576 	 * Wake up parent sleeping in kern_wait(), also send
3577 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3578 	 * that parent will awake, because parent may masked
3579 	 * the signal.
3580 	 */
3581 	p->p_pptr->p_flag |= P_STATCHILD;
3582 	wakeup(p->p_pptr);
3583 
3584 	ps = p->p_pptr->p_sigacts;
3585 	mtx_lock(&ps->ps_mtx);
3586 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3587 		mtx_unlock(&ps->ps_mtx);
3588 		sigparent(p, reason, sig);
3589 	} else
3590 		mtx_unlock(&ps->ps_mtx);
3591 }
3592 
3593 void
3594 childproc_stopped(struct proc *p, int reason)
3595 {
3596 
3597 	childproc_jobstate(p, reason, p->p_xsig);
3598 }
3599 
3600 void
3601 childproc_continued(struct proc *p)
3602 {
3603 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3604 }
3605 
3606 void
3607 childproc_exited(struct proc *p)
3608 {
3609 	int reason, status;
3610 
3611 	if (WCOREDUMP(p->p_xsig)) {
3612 		reason = CLD_DUMPED;
3613 		status = WTERMSIG(p->p_xsig);
3614 	} else if (WIFSIGNALED(p->p_xsig)) {
3615 		reason = CLD_KILLED;
3616 		status = WTERMSIG(p->p_xsig);
3617 	} else {
3618 		reason = CLD_EXITED;
3619 		status = p->p_xexit;
3620 	}
3621 	/*
3622 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3623 	 * done in exit1().
3624 	 */
3625 	sigparent(p, reason, status);
3626 }
3627 
3628 #define	MAX_NUM_CORE_FILES 100000
3629 #ifndef NUM_CORE_FILES
3630 #define	NUM_CORE_FILES 5
3631 #endif
3632 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3633 static int num_cores = NUM_CORE_FILES;
3634 
3635 static int
3636 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3637 {
3638 	int error;
3639 	int new_val;
3640 
3641 	new_val = num_cores;
3642 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3643 	if (error != 0 || req->newptr == NULL)
3644 		return (error);
3645 	if (new_val > MAX_NUM_CORE_FILES)
3646 		new_val = MAX_NUM_CORE_FILES;
3647 	if (new_val < 0)
3648 		new_val = 0;
3649 	num_cores = new_val;
3650 	return (0);
3651 }
3652 SYSCTL_PROC(_debug, OID_AUTO, ncores,
3653     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
3654     sysctl_debug_num_cores_check, "I",
3655     "Maximum number of generated process corefiles while using index format");
3656 
3657 #define	GZIP_SUFFIX	".gz"
3658 #define	ZSTD_SUFFIX	".zst"
3659 
3660 int compress_user_cores = 0;
3661 
3662 static int
3663 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3664 {
3665 	int error, val;
3666 
3667 	val = compress_user_cores;
3668 	error = sysctl_handle_int(oidp, &val, 0, req);
3669 	if (error != 0 || req->newptr == NULL)
3670 		return (error);
3671 	if (val != 0 && !compressor_avail(val))
3672 		return (EINVAL);
3673 	compress_user_cores = val;
3674 	return (error);
3675 }
3676 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores,
3677     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, 0, sizeof(int),
3678     sysctl_compress_user_cores, "I",
3679     "Enable compression of user corefiles ("
3680     __XSTRING(COMPRESS_GZIP) " = gzip, "
3681     __XSTRING(COMPRESS_ZSTD) " = zstd)");
3682 
3683 int compress_user_cores_level = 6;
3684 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3685     &compress_user_cores_level, 0,
3686     "Corefile compression level");
3687 
3688 /*
3689  * Protect the access to corefilename[] by allproc_lock.
3690  */
3691 #define	corefilename_lock	allproc_lock
3692 
3693 static char corefilename[MAXPATHLEN] = {"%N.core"};
3694 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3695 
3696 static int
3697 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3698 {
3699 	int error;
3700 
3701 	sx_xlock(&corefilename_lock);
3702 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3703 	    req);
3704 	sx_xunlock(&corefilename_lock);
3705 
3706 	return (error);
3707 }
3708 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3709     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3710     "Process corefile name format string");
3711 
3712 static void
3713 vnode_close_locked(struct thread *td, struct vnode *vp)
3714 {
3715 
3716 	VOP_UNLOCK(vp);
3717 	vn_close(vp, FWRITE, td->td_ucred, td);
3718 }
3719 
3720 /*
3721  * If the core format has a %I in it, then we need to check
3722  * for existing corefiles before defining a name.
3723  * To do this we iterate over 0..ncores to find a
3724  * non-existing core file name to use. If all core files are
3725  * already used we choose the oldest one.
3726  */
3727 static int
3728 corefile_open_last(struct thread *td, char *name, int indexpos,
3729     int indexlen, int ncores, struct vnode **vpp)
3730 {
3731 	struct vnode *oldvp, *nextvp, *vp;
3732 	struct vattr vattr;
3733 	struct nameidata nd;
3734 	int error, i, flags, oflags, cmode;
3735 	char ch;
3736 	struct timespec lasttime;
3737 
3738 	nextvp = oldvp = NULL;
3739 	cmode = S_IRUSR | S_IWUSR;
3740 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3741 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3742 
3743 	for (i = 0; i < ncores; i++) {
3744 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3745 
3746 		ch = name[indexpos + indexlen];
3747 		(void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3748 		    i);
3749 		name[indexpos + indexlen] = ch;
3750 
3751 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
3752 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3753 		    NULL);
3754 		if (error != 0)
3755 			break;
3756 
3757 		vp = nd.ni_vp;
3758 		NDFREE_PNBUF(&nd);
3759 		if ((flags & O_CREAT) == O_CREAT) {
3760 			nextvp = vp;
3761 			break;
3762 		}
3763 
3764 		error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3765 		if (error != 0) {
3766 			vnode_close_locked(td, vp);
3767 			break;
3768 		}
3769 
3770 		if (oldvp == NULL ||
3771 		    lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3772 		    (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3773 		    lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3774 			if (oldvp != NULL)
3775 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3776 			oldvp = vp;
3777 			VOP_UNLOCK(oldvp);
3778 			lasttime = vattr.va_mtime;
3779 		} else {
3780 			vnode_close_locked(td, vp);
3781 		}
3782 	}
3783 
3784 	if (oldvp != NULL) {
3785 		if (nextvp == NULL) {
3786 			if ((td->td_proc->p_flag & P_SUGID) != 0) {
3787 				error = EFAULT;
3788 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3789 			} else {
3790 				nextvp = oldvp;
3791 				error = vn_lock(nextvp, LK_EXCLUSIVE);
3792 				if (error != 0) {
3793 					vn_close(nextvp, FWRITE, td->td_ucred,
3794 					    td);
3795 					nextvp = NULL;
3796 				}
3797 			}
3798 		} else {
3799 			vn_close(oldvp, FWRITE, td->td_ucred, td);
3800 		}
3801 	}
3802 	if (error != 0) {
3803 		if (nextvp != NULL)
3804 			vnode_close_locked(td, oldvp);
3805 	} else {
3806 		*vpp = nextvp;
3807 	}
3808 
3809 	return (error);
3810 }
3811 
3812 /*
3813  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3814  * Expand the name described in corefilename, using name, uid, and pid
3815  * and open/create core file.
3816  * corefilename is a printf-like string, with three format specifiers:
3817  *	%N	name of process ("name")
3818  *	%P	process id (pid)
3819  *	%U	user id (uid)
3820  * For example, "%N.core" is the default; they can be disabled completely
3821  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3822  * This is controlled by the sysctl variable kern.corefile (see above).
3823  */
3824 static int
3825 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3826     int compress, int signum, struct vnode **vpp, char **namep)
3827 {
3828 	struct sbuf sb;
3829 	struct nameidata nd;
3830 	const char *format;
3831 	char *hostname, *name;
3832 	int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
3833 
3834 	hostname = NULL;
3835 	format = corefilename;
3836 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3837 	indexlen = 0;
3838 	indexpos = -1;
3839 	ncores = num_cores;
3840 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3841 	sx_slock(&corefilename_lock);
3842 	for (i = 0; format[i] != '\0'; i++) {
3843 		switch (format[i]) {
3844 		case '%':	/* Format character */
3845 			i++;
3846 			switch (format[i]) {
3847 			case '%':
3848 				sbuf_putc(&sb, '%');
3849 				break;
3850 			case 'H':	/* hostname */
3851 				if (hostname == NULL) {
3852 					hostname = malloc(MAXHOSTNAMELEN,
3853 					    M_TEMP, M_WAITOK);
3854 				}
3855 				getcredhostname(td->td_ucred, hostname,
3856 				    MAXHOSTNAMELEN);
3857 				sbuf_printf(&sb, "%s", hostname);
3858 				break;
3859 			case 'I':	/* autoincrementing index */
3860 				if (indexpos != -1) {
3861 					sbuf_printf(&sb, "%%I");
3862 					break;
3863 				}
3864 
3865 				indexpos = sbuf_len(&sb);
3866 				sbuf_printf(&sb, "%u", ncores - 1);
3867 				indexlen = sbuf_len(&sb) - indexpos;
3868 				break;
3869 			case 'N':	/* process name */
3870 				sbuf_printf(&sb, "%s", comm);
3871 				break;
3872 			case 'P':	/* process id */
3873 				sbuf_printf(&sb, "%u", pid);
3874 				break;
3875 			case 'S':	/* signal number */
3876 				sbuf_printf(&sb, "%i", signum);
3877 				break;
3878 			case 'U':	/* user id */
3879 				sbuf_printf(&sb, "%u", uid);
3880 				break;
3881 			default:
3882 				log(LOG_ERR,
3883 				    "Unknown format character %c in "
3884 				    "corename `%s'\n", format[i], format);
3885 				break;
3886 			}
3887 			break;
3888 		default:
3889 			sbuf_putc(&sb, format[i]);
3890 			break;
3891 		}
3892 	}
3893 	sx_sunlock(&corefilename_lock);
3894 	free(hostname, M_TEMP);
3895 	if (compress == COMPRESS_GZIP)
3896 		sbuf_printf(&sb, GZIP_SUFFIX);
3897 	else if (compress == COMPRESS_ZSTD)
3898 		sbuf_printf(&sb, ZSTD_SUFFIX);
3899 	if (sbuf_error(&sb) != 0) {
3900 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3901 		    "long\n", (long)pid, comm, (u_long)uid);
3902 		sbuf_delete(&sb);
3903 		free(name, M_TEMP);
3904 		return (ENOMEM);
3905 	}
3906 	sbuf_finish(&sb);
3907 	sbuf_delete(&sb);
3908 
3909 	if (indexpos != -1) {
3910 		error = corefile_open_last(td, name, indexpos, indexlen, ncores,
3911 		    vpp);
3912 		if (error != 0) {
3913 			log(LOG_ERR,
3914 			    "pid %d (%s), uid (%u):  Path `%s' failed "
3915 			    "on initial open test, error = %d\n",
3916 			    pid, comm, uid, name, error);
3917 		}
3918 	} else {
3919 		cmode = S_IRUSR | S_IWUSR;
3920 		oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3921 		    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3922 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3923 		if ((td->td_proc->p_flag & P_SUGID) != 0)
3924 			flags |= O_EXCL;
3925 
3926 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
3927 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3928 		    NULL);
3929 		if (error == 0) {
3930 			*vpp = nd.ni_vp;
3931 			NDFREE_PNBUF(&nd);
3932 		}
3933 	}
3934 
3935 	if (error != 0) {
3936 #ifdef AUDIT
3937 		audit_proc_coredump(td, name, error);
3938 #endif
3939 		free(name, M_TEMP);
3940 		return (error);
3941 	}
3942 	*namep = name;
3943 	return (0);
3944 }
3945 
3946 /*
3947  * Dump a process' core.  The main routine does some
3948  * policy checking, and creates the name of the coredump;
3949  * then it passes on a vnode and a size limit to the process-specific
3950  * coredump routine if there is one; if there _is not_ one, it returns
3951  * ENOSYS; otherwise it returns the error from the process-specific routine.
3952  */
3953 
3954 static int
3955 coredump(struct thread *td)
3956 {
3957 	struct proc *p = td->td_proc;
3958 	struct ucred *cred = td->td_ucred;
3959 	struct vnode *vp;
3960 	struct flock lf;
3961 	struct vattr vattr;
3962 	size_t fullpathsize;
3963 	int error, error1, locked;
3964 	char *name;			/* name of corefile */
3965 	void *rl_cookie;
3966 	off_t limit;
3967 	char *fullpath, *freepath = NULL;
3968 	struct sbuf *sb;
3969 
3970 	PROC_LOCK_ASSERT(p, MA_OWNED);
3971 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3972 
3973 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
3974 	    (p->p_flag2 & P2_NOTRACE) != 0) {
3975 		PROC_UNLOCK(p);
3976 		return (EFAULT);
3977 	}
3978 
3979 	/*
3980 	 * Note that the bulk of limit checking is done after
3981 	 * the corefile is created.  The exception is if the limit
3982 	 * for corefiles is 0, in which case we don't bother
3983 	 * creating the corefile at all.  This layout means that
3984 	 * a corefile is truncated instead of not being created,
3985 	 * if it is larger than the limit.
3986 	 */
3987 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
3988 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
3989 		PROC_UNLOCK(p);
3990 		return (EFBIG);
3991 	}
3992 	PROC_UNLOCK(p);
3993 
3994 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
3995 	    compress_user_cores, p->p_sig, &vp, &name);
3996 	if (error != 0)
3997 		return (error);
3998 
3999 	/*
4000 	 * Don't dump to non-regular files or files with links.
4001 	 * Do not dump into system files. Effective user must own the corefile.
4002 	 */
4003 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
4004 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
4005 	    vattr.va_uid != cred->cr_uid) {
4006 		VOP_UNLOCK(vp);
4007 		error = EFAULT;
4008 		goto out;
4009 	}
4010 
4011 	VOP_UNLOCK(vp);
4012 
4013 	/* Postpone other writers, including core dumps of other processes. */
4014 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
4015 
4016 	lf.l_whence = SEEK_SET;
4017 	lf.l_start = 0;
4018 	lf.l_len = 0;
4019 	lf.l_type = F_WRLCK;
4020 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
4021 
4022 	VATTR_NULL(&vattr);
4023 	vattr.va_size = 0;
4024 	if (set_core_nodump_flag)
4025 		vattr.va_flags = UF_NODUMP;
4026 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4027 	VOP_SETATTR(vp, &vattr, cred);
4028 	VOP_UNLOCK(vp);
4029 	PROC_LOCK(p);
4030 	p->p_acflag |= ACORE;
4031 	PROC_UNLOCK(p);
4032 
4033 	if (p->p_sysent->sv_coredump != NULL) {
4034 		error = p->p_sysent->sv_coredump(td, vp, limit, 0);
4035 	} else {
4036 		error = ENOSYS;
4037 	}
4038 
4039 	if (locked) {
4040 		lf.l_type = F_UNLCK;
4041 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
4042 	}
4043 	vn_rangelock_unlock(vp, rl_cookie);
4044 
4045 	/*
4046 	 * Notify the userland helper that a process triggered a core dump.
4047 	 * This allows the helper to run an automated debugging session.
4048 	 */
4049 	if (error != 0 || coredump_devctl == 0)
4050 		goto out;
4051 	sb = sbuf_new_auto();
4052 	if (vn_fullpath_global(p->p_textvp, &fullpath, &freepath) != 0)
4053 		goto out2;
4054 	sbuf_printf(sb, "comm=\"");
4055 	devctl_safe_quote_sb(sb, fullpath);
4056 	free(freepath, M_TEMP);
4057 	sbuf_printf(sb, "\" core=\"");
4058 
4059 	/*
4060 	 * We can't lookup core file vp directly. When we're replacing a core, and
4061 	 * other random times, we flush the name cache, so it will fail. Instead,
4062 	 * if the path of the core is relative, add the current dir in front if it.
4063 	 */
4064 	if (name[0] != '/') {
4065 		fullpathsize = MAXPATHLEN;
4066 		freepath = malloc(fullpathsize, M_TEMP, M_WAITOK);
4067 		if (vn_getcwd(freepath, &fullpath, &fullpathsize) != 0) {
4068 			free(freepath, M_TEMP);
4069 			goto out2;
4070 		}
4071 		devctl_safe_quote_sb(sb, fullpath);
4072 		free(freepath, M_TEMP);
4073 		sbuf_putc(sb, '/');
4074 	}
4075 	devctl_safe_quote_sb(sb, name);
4076 	sbuf_printf(sb, "\"");
4077 	if (sbuf_finish(sb) == 0)
4078 		devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
4079 out2:
4080 	sbuf_delete(sb);
4081 out:
4082 	error1 = vn_close(vp, FWRITE, cred, td);
4083 	if (error == 0)
4084 		error = error1;
4085 #ifdef AUDIT
4086 	audit_proc_coredump(td, name, error);
4087 #endif
4088 	free(name, M_TEMP);
4089 	return (error);
4090 }
4091 
4092 /*
4093  * Nonexistent system call-- signal process (may want to handle it).  Flag
4094  * error in case process won't see signal immediately (blocked or ignored).
4095  */
4096 #ifndef _SYS_SYSPROTO_H_
4097 struct nosys_args {
4098 	int	dummy;
4099 };
4100 #endif
4101 /* ARGSUSED */
4102 int
4103 nosys(struct thread *td, struct nosys_args *args)
4104 {
4105 	struct proc *p;
4106 
4107 	p = td->td_proc;
4108 
4109 	PROC_LOCK(p);
4110 	tdsignal(td, SIGSYS);
4111 	PROC_UNLOCK(p);
4112 	if (kern_lognosys == 1 || kern_lognosys == 3) {
4113 		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4114 		    td->td_sa.code);
4115 	}
4116 	if (kern_lognosys == 2 || kern_lognosys == 3 ||
4117 	    (p->p_pid == 1 && (kern_lognosys & 3) == 0)) {
4118 		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4119 		    td->td_sa.code);
4120 	}
4121 	return (ENOSYS);
4122 }
4123 
4124 /*
4125  * Send a SIGIO or SIGURG signal to a process or process group using stored
4126  * credentials rather than those of the current process.
4127  */
4128 void
4129 pgsigio(struct sigio **sigiop, int sig, int checkctty)
4130 {
4131 	ksiginfo_t ksi;
4132 	struct sigio *sigio;
4133 
4134 	ksiginfo_init(&ksi);
4135 	ksi.ksi_signo = sig;
4136 	ksi.ksi_code = SI_KERNEL;
4137 
4138 	SIGIO_LOCK();
4139 	sigio = *sigiop;
4140 	if (sigio == NULL) {
4141 		SIGIO_UNLOCK();
4142 		return;
4143 	}
4144 	if (sigio->sio_pgid > 0) {
4145 		PROC_LOCK(sigio->sio_proc);
4146 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
4147 			kern_psignal(sigio->sio_proc, sig);
4148 		PROC_UNLOCK(sigio->sio_proc);
4149 	} else if (sigio->sio_pgid < 0) {
4150 		struct proc *p;
4151 
4152 		PGRP_LOCK(sigio->sio_pgrp);
4153 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
4154 			PROC_LOCK(p);
4155 			if (p->p_state == PRS_NORMAL &&
4156 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
4157 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
4158 				kern_psignal(p, sig);
4159 			PROC_UNLOCK(p);
4160 		}
4161 		PGRP_UNLOCK(sigio->sio_pgrp);
4162 	}
4163 	SIGIO_UNLOCK();
4164 }
4165 
4166 static int
4167 filt_sigattach(struct knote *kn)
4168 {
4169 	struct proc *p = curproc;
4170 
4171 	kn->kn_ptr.p_proc = p;
4172 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
4173 
4174 	knlist_add(p->p_klist, kn, 0);
4175 
4176 	return (0);
4177 }
4178 
4179 static void
4180 filt_sigdetach(struct knote *kn)
4181 {
4182 	struct proc *p = kn->kn_ptr.p_proc;
4183 
4184 	knlist_remove(p->p_klist, kn, 0);
4185 }
4186 
4187 /*
4188  * signal knotes are shared with proc knotes, so we apply a mask to
4189  * the hint in order to differentiate them from process hints.  This
4190  * could be avoided by using a signal-specific knote list, but probably
4191  * isn't worth the trouble.
4192  */
4193 static int
4194 filt_signal(struct knote *kn, long hint)
4195 {
4196 
4197 	if (hint & NOTE_SIGNAL) {
4198 		hint &= ~NOTE_SIGNAL;
4199 
4200 		if (kn->kn_id == hint)
4201 			kn->kn_data++;
4202 	}
4203 	return (kn->kn_data != 0);
4204 }
4205 
4206 struct sigacts *
4207 sigacts_alloc(void)
4208 {
4209 	struct sigacts *ps;
4210 
4211 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
4212 	refcount_init(&ps->ps_refcnt, 1);
4213 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
4214 	return (ps);
4215 }
4216 
4217 void
4218 sigacts_free(struct sigacts *ps)
4219 {
4220 
4221 	if (refcount_release(&ps->ps_refcnt) == 0)
4222 		return;
4223 	mtx_destroy(&ps->ps_mtx);
4224 	free(ps, M_SUBPROC);
4225 }
4226 
4227 struct sigacts *
4228 sigacts_hold(struct sigacts *ps)
4229 {
4230 
4231 	refcount_acquire(&ps->ps_refcnt);
4232 	return (ps);
4233 }
4234 
4235 void
4236 sigacts_copy(struct sigacts *dest, struct sigacts *src)
4237 {
4238 
4239 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
4240 	mtx_lock(&src->ps_mtx);
4241 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
4242 	mtx_unlock(&src->ps_mtx);
4243 }
4244 
4245 int
4246 sigacts_shared(struct sigacts *ps)
4247 {
4248 
4249 	return (ps->ps_refcnt > 1);
4250 }
4251 
4252 void
4253 sig_drop_caught(struct proc *p)
4254 {
4255 	int sig;
4256 	struct sigacts *ps;
4257 
4258 	ps = p->p_sigacts;
4259 	PROC_LOCK_ASSERT(p, MA_OWNED);
4260 	mtx_assert(&ps->ps_mtx, MA_OWNED);
4261 	SIG_FOREACH(sig, &ps->ps_sigcatch) {
4262 		sigdflt(ps, sig);
4263 		if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
4264 			sigqueue_delete_proc(p, sig);
4265 	}
4266 }
4267 
4268 static void
4269 sigfastblock_failed(struct thread *td, bool sendsig, bool write)
4270 {
4271 	ksiginfo_t ksi;
4272 
4273 	/*
4274 	 * Prevent further fetches and SIGSEGVs, allowing thread to
4275 	 * issue syscalls despite corruption.
4276 	 */
4277 	sigfastblock_clear(td);
4278 
4279 	if (!sendsig)
4280 		return;
4281 	ksiginfo_init_trap(&ksi);
4282 	ksi.ksi_signo = SIGSEGV;
4283 	ksi.ksi_code = write ? SEGV_ACCERR : SEGV_MAPERR;
4284 	ksi.ksi_addr = td->td_sigblock_ptr;
4285 	trapsignal(td, &ksi);
4286 }
4287 
4288 static bool
4289 sigfastblock_fetch_sig(struct thread *td, bool sendsig, uint32_t *valp)
4290 {
4291 	uint32_t res;
4292 
4293 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4294 		return (true);
4295 	if (fueword32((void *)td->td_sigblock_ptr, &res) == -1) {
4296 		sigfastblock_failed(td, sendsig, false);
4297 		return (false);
4298 	}
4299 	*valp = res;
4300 	td->td_sigblock_val = res & ~SIGFASTBLOCK_FLAGS;
4301 	return (true);
4302 }
4303 
4304 static void
4305 sigfastblock_resched(struct thread *td, bool resched)
4306 {
4307 	struct proc *p;
4308 
4309 	if (resched) {
4310 		p = td->td_proc;
4311 		PROC_LOCK(p);
4312 		reschedule_signals(p, td->td_sigmask, 0);
4313 		PROC_UNLOCK(p);
4314 	}
4315 	ast_sched(td, TDA_SIG);
4316 }
4317 
4318 int
4319 sys_sigfastblock(struct thread *td, struct sigfastblock_args *uap)
4320 {
4321 	struct proc *p;
4322 	int error, res;
4323 	uint32_t oldval;
4324 
4325 	error = 0;
4326 	p = td->td_proc;
4327 	switch (uap->cmd) {
4328 	case SIGFASTBLOCK_SETPTR:
4329 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
4330 			error = EBUSY;
4331 			break;
4332 		}
4333 		if (((uintptr_t)(uap->ptr) & (sizeof(uint32_t) - 1)) != 0) {
4334 			error = EINVAL;
4335 			break;
4336 		}
4337 		td->td_pflags |= TDP_SIGFASTBLOCK;
4338 		td->td_sigblock_ptr = uap->ptr;
4339 		break;
4340 
4341 	case SIGFASTBLOCK_UNBLOCK:
4342 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4343 			error = EINVAL;
4344 			break;
4345 		}
4346 
4347 		for (;;) {
4348 			res = casueword32(td->td_sigblock_ptr,
4349 			    SIGFASTBLOCK_PEND, &oldval, 0);
4350 			if (res == -1) {
4351 				error = EFAULT;
4352 				sigfastblock_failed(td, false, true);
4353 				break;
4354 			}
4355 			if (res == 0)
4356 				break;
4357 			MPASS(res == 1);
4358 			if (oldval != SIGFASTBLOCK_PEND) {
4359 				error = EBUSY;
4360 				break;
4361 			}
4362 			error = thread_check_susp(td, false);
4363 			if (error != 0)
4364 				break;
4365 		}
4366 		if (error != 0)
4367 			break;
4368 
4369 		/*
4370 		 * td_sigblock_val is cleared there, but not on a
4371 		 * syscall exit.  The end effect is that a single
4372 		 * interruptible sleep, while user sigblock word is
4373 		 * set, might return EINTR or ERESTART to usermode
4374 		 * without delivering signal.  All further sleeps,
4375 		 * until userspace clears the word and does
4376 		 * sigfastblock(UNBLOCK), observe current word and no
4377 		 * longer get interrupted.  It is slight
4378 		 * non-conformance, with alternative to have read the
4379 		 * sigblock word on each syscall entry.
4380 		 */
4381 		td->td_sigblock_val = 0;
4382 
4383 		/*
4384 		 * Rely on normal ast mechanism to deliver pending
4385 		 * signals to current thread.  But notify others about
4386 		 * fake unblock.
4387 		 */
4388 		sigfastblock_resched(td, error == 0 && p->p_numthreads != 1);
4389 
4390 		break;
4391 
4392 	case SIGFASTBLOCK_UNSETPTR:
4393 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4394 			error = EINVAL;
4395 			break;
4396 		}
4397 		if (!sigfastblock_fetch_sig(td, false, &oldval)) {
4398 			error = EFAULT;
4399 			break;
4400 		}
4401 		if (oldval != 0 && oldval != SIGFASTBLOCK_PEND) {
4402 			error = EBUSY;
4403 			break;
4404 		}
4405 		sigfastblock_clear(td);
4406 		break;
4407 
4408 	default:
4409 		error = EINVAL;
4410 		break;
4411 	}
4412 	return (error);
4413 }
4414 
4415 void
4416 sigfastblock_clear(struct thread *td)
4417 {
4418 	bool resched;
4419 
4420 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4421 		return;
4422 	td->td_sigblock_val = 0;
4423 	resched = (td->td_pflags & TDP_SIGFASTPENDING) != 0 ||
4424 	    SIGPENDING(td);
4425 	td->td_pflags &= ~(TDP_SIGFASTBLOCK | TDP_SIGFASTPENDING);
4426 	sigfastblock_resched(td, resched);
4427 }
4428 
4429 void
4430 sigfastblock_fetch(struct thread *td)
4431 {
4432 	uint32_t val;
4433 
4434 	(void)sigfastblock_fetch_sig(td, true, &val);
4435 }
4436 
4437 static void
4438 sigfastblock_setpend1(struct thread *td)
4439 {
4440 	int res;
4441 	uint32_t oldval;
4442 
4443 	if ((td->td_pflags & TDP_SIGFASTPENDING) == 0)
4444 		return;
4445 	res = fueword32((void *)td->td_sigblock_ptr, &oldval);
4446 	if (res == -1) {
4447 		sigfastblock_failed(td, true, false);
4448 		return;
4449 	}
4450 	for (;;) {
4451 		res = casueword32(td->td_sigblock_ptr, oldval, &oldval,
4452 		    oldval | SIGFASTBLOCK_PEND);
4453 		if (res == -1) {
4454 			sigfastblock_failed(td, true, true);
4455 			return;
4456 		}
4457 		if (res == 0) {
4458 			td->td_sigblock_val = oldval & ~SIGFASTBLOCK_FLAGS;
4459 			td->td_pflags &= ~TDP_SIGFASTPENDING;
4460 			break;
4461 		}
4462 		MPASS(res == 1);
4463 		if (thread_check_susp(td, false) != 0)
4464 			break;
4465 	}
4466 }
4467 
4468 static void
4469 sigfastblock_setpend(struct thread *td, bool resched)
4470 {
4471 	struct proc *p;
4472 
4473 	sigfastblock_setpend1(td);
4474 	if (resched) {
4475 		p = td->td_proc;
4476 		PROC_LOCK(p);
4477 		reschedule_signals(p, fastblock_mask, SIGPROCMASK_FASTBLK);
4478 		PROC_UNLOCK(p);
4479 	}
4480 }
4481