xref: /freebsd/sys/kern/kern_sig.c (revision b197d4b893974c9eb4d7b38704c6d5c486235d6f)
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(struct proc *p, bool notself, struct killpg1_ctx *arg)
1780 {
1781 	int err;
1782 
1783 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1784 	    (notself && p == arg->td->td_proc) || p->p_state == PRS_NEW)
1785 		return;
1786 	PROC_LOCK(p);
1787 	err = p_cansignal(arg->td, p, arg->sig);
1788 	if (err == 0 && arg->sig != 0)
1789 		pksignal(p, arg->sig, arg->ksi);
1790 	PROC_UNLOCK(p);
1791 	if (err != ESRCH)
1792 		arg->found = true;
1793 	if (err == 0)
1794 		arg->sent = true;
1795 	else if (arg->ret == 0 && err != ESRCH && err != EPERM)
1796 		arg->ret = err;
1797 }
1798 
1799 /*
1800  * Common code for kill process group/broadcast kill.
1801  * cp is calling process.
1802  */
1803 static int
1804 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1805 {
1806 	struct proc *p;
1807 	struct pgrp *pgrp;
1808 	struct killpg1_ctx arg;
1809 
1810 	arg.td = td;
1811 	arg.ksi = ksi;
1812 	arg.sig = sig;
1813 	arg.sent = false;
1814 	arg.found = false;
1815 	arg.ret = 0;
1816 	if (all) {
1817 		/*
1818 		 * broadcast
1819 		 */
1820 		sx_slock(&allproc_lock);
1821 		FOREACH_PROC_IN_SYSTEM(p) {
1822 			killpg1_sendsig(p, true, &arg);
1823 		}
1824 		sx_sunlock(&allproc_lock);
1825 	} else {
1826 		sx_slock(&proctree_lock);
1827 		if (pgid == 0) {
1828 			/*
1829 			 * zero pgid means send to my process group.
1830 			 */
1831 			pgrp = td->td_proc->p_pgrp;
1832 			PGRP_LOCK(pgrp);
1833 		} else {
1834 			pgrp = pgfind(pgid);
1835 			if (pgrp == NULL) {
1836 				sx_sunlock(&proctree_lock);
1837 				return (ESRCH);
1838 			}
1839 		}
1840 		sx_sunlock(&proctree_lock);
1841 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1842 			killpg1_sendsig(p, false, &arg);
1843 		}
1844 		PGRP_UNLOCK(pgrp);
1845 	}
1846 	MPASS(arg.ret != 0 || arg.found || !arg.sent);
1847 	if (arg.ret == 0 && !arg.sent)
1848 		arg.ret = arg.found ? EPERM : ESRCH;
1849 	return (arg.ret);
1850 }
1851 
1852 #ifndef _SYS_SYSPROTO_H_
1853 struct kill_args {
1854 	int	pid;
1855 	int	signum;
1856 };
1857 #endif
1858 /* ARGSUSED */
1859 int
1860 sys_kill(struct thread *td, struct kill_args *uap)
1861 {
1862 
1863 	return (kern_kill(td, uap->pid, uap->signum));
1864 }
1865 
1866 int
1867 kern_kill(struct thread *td, pid_t pid, int signum)
1868 {
1869 	ksiginfo_t ksi;
1870 	struct proc *p;
1871 	int error;
1872 
1873 	/*
1874 	 * A process in capability mode can send signals only to himself.
1875 	 * The main rationale behind this is that abort(3) is implemented as
1876 	 * kill(getpid(), SIGABRT).
1877 	 */
1878 	if (IN_CAPABILITY_MODE(td) && pid != td->td_proc->p_pid)
1879 		return (ECAPMODE);
1880 
1881 	AUDIT_ARG_SIGNUM(signum);
1882 	AUDIT_ARG_PID(pid);
1883 	if ((u_int)signum > _SIG_MAXSIG)
1884 		return (EINVAL);
1885 
1886 	ksiginfo_init(&ksi);
1887 	ksi.ksi_signo = signum;
1888 	ksi.ksi_code = SI_USER;
1889 	ksi.ksi_pid = td->td_proc->p_pid;
1890 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1891 
1892 	if (pid > 0) {
1893 		/* kill single process */
1894 		if ((p = pfind_any(pid)) == NULL)
1895 			return (ESRCH);
1896 		AUDIT_ARG_PROCESS(p);
1897 		error = p_cansignal(td, p, signum);
1898 		if (error == 0 && signum)
1899 			pksignal(p, signum, &ksi);
1900 		PROC_UNLOCK(p);
1901 		return (error);
1902 	}
1903 	switch (pid) {
1904 	case -1:		/* broadcast signal */
1905 		return (killpg1(td, signum, 0, 1, &ksi));
1906 	case 0:			/* signal own process group */
1907 		return (killpg1(td, signum, 0, 0, &ksi));
1908 	default:		/* negative explicit process group */
1909 		return (killpg1(td, signum, -pid, 0, &ksi));
1910 	}
1911 	/* NOTREACHED */
1912 }
1913 
1914 int
1915 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1916 {
1917 	struct proc *p;
1918 	int error;
1919 
1920 	AUDIT_ARG_SIGNUM(uap->signum);
1921 	AUDIT_ARG_FD(uap->fd);
1922 	if ((u_int)uap->signum > _SIG_MAXSIG)
1923 		return (EINVAL);
1924 
1925 	error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1926 	if (error)
1927 		return (error);
1928 	AUDIT_ARG_PROCESS(p);
1929 	error = p_cansignal(td, p, uap->signum);
1930 	if (error == 0 && uap->signum)
1931 		kern_psignal(p, uap->signum);
1932 	PROC_UNLOCK(p);
1933 	return (error);
1934 }
1935 
1936 #if defined(COMPAT_43)
1937 #ifndef _SYS_SYSPROTO_H_
1938 struct okillpg_args {
1939 	int	pgid;
1940 	int	signum;
1941 };
1942 #endif
1943 /* ARGSUSED */
1944 int
1945 okillpg(struct thread *td, struct okillpg_args *uap)
1946 {
1947 	ksiginfo_t ksi;
1948 
1949 	AUDIT_ARG_SIGNUM(uap->signum);
1950 	AUDIT_ARG_PID(uap->pgid);
1951 	if ((u_int)uap->signum > _SIG_MAXSIG)
1952 		return (EINVAL);
1953 
1954 	ksiginfo_init(&ksi);
1955 	ksi.ksi_signo = uap->signum;
1956 	ksi.ksi_code = SI_USER;
1957 	ksi.ksi_pid = td->td_proc->p_pid;
1958 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1959 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1960 }
1961 #endif /* COMPAT_43 */
1962 
1963 #ifndef _SYS_SYSPROTO_H_
1964 struct sigqueue_args {
1965 	pid_t pid;
1966 	int signum;
1967 	/* union sigval */ void *value;
1968 };
1969 #endif
1970 int
1971 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1972 {
1973 	union sigval sv;
1974 
1975 	sv.sival_ptr = uap->value;
1976 
1977 	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
1978 }
1979 
1980 int
1981 kern_sigqueue(struct thread *td, pid_t pid, int signum, union sigval *value)
1982 {
1983 	ksiginfo_t ksi;
1984 	struct proc *p;
1985 	int error;
1986 
1987 	if ((u_int)signum > _SIG_MAXSIG)
1988 		return (EINVAL);
1989 
1990 	/*
1991 	 * Specification says sigqueue can only send signal to
1992 	 * single process.
1993 	 */
1994 	if (pid <= 0)
1995 		return (EINVAL);
1996 
1997 	if ((p = pfind_any(pid)) == NULL)
1998 		return (ESRCH);
1999 	error = p_cansignal(td, p, signum);
2000 	if (error == 0 && signum != 0) {
2001 		ksiginfo_init(&ksi);
2002 		ksi.ksi_flags = KSI_SIGQ;
2003 		ksi.ksi_signo = signum;
2004 		ksi.ksi_code = SI_QUEUE;
2005 		ksi.ksi_pid = td->td_proc->p_pid;
2006 		ksi.ksi_uid = td->td_ucred->cr_ruid;
2007 		ksi.ksi_value = *value;
2008 		error = pksignal(p, ksi.ksi_signo, &ksi);
2009 	}
2010 	PROC_UNLOCK(p);
2011 	return (error);
2012 }
2013 
2014 /*
2015  * Send a signal to a process group.
2016  */
2017 void
2018 gsignal(int pgid, int sig, ksiginfo_t *ksi)
2019 {
2020 	struct pgrp *pgrp;
2021 
2022 	if (pgid != 0) {
2023 		sx_slock(&proctree_lock);
2024 		pgrp = pgfind(pgid);
2025 		sx_sunlock(&proctree_lock);
2026 		if (pgrp != NULL) {
2027 			pgsignal(pgrp, sig, 0, ksi);
2028 			PGRP_UNLOCK(pgrp);
2029 		}
2030 	}
2031 }
2032 
2033 /*
2034  * Send a signal to a process group.  If checktty is 1,
2035  * limit to members which have a controlling terminal.
2036  */
2037 void
2038 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
2039 {
2040 	struct proc *p;
2041 
2042 	if (pgrp) {
2043 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
2044 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
2045 			PROC_LOCK(p);
2046 			if (p->p_state == PRS_NORMAL &&
2047 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
2048 				pksignal(p, sig, ksi);
2049 			PROC_UNLOCK(p);
2050 		}
2051 	}
2052 }
2053 
2054 /*
2055  * Recalculate the signal mask and reset the signal disposition after
2056  * usermode frame for delivery is formed.  Should be called after
2057  * mach-specific routine, because sysent->sv_sendsig() needs correct
2058  * ps_siginfo and signal mask.
2059  */
2060 static void
2061 postsig_done(int sig, struct thread *td, struct sigacts *ps)
2062 {
2063 	sigset_t mask;
2064 
2065 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2066 	td->td_ru.ru_nsignals++;
2067 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
2068 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
2069 		SIGADDSET(mask, sig);
2070 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
2071 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
2072 	if (SIGISMEMBER(ps->ps_sigreset, sig))
2073 		sigdflt(ps, sig);
2074 }
2075 
2076 /*
2077  * Send a signal caused by a trap to the current thread.  If it will be
2078  * caught immediately, deliver it with correct code.  Otherwise, post it
2079  * normally.
2080  */
2081 void
2082 trapsignal(struct thread *td, ksiginfo_t *ksi)
2083 {
2084 	struct sigacts *ps;
2085 	struct proc *p;
2086 	sigset_t sigmask;
2087 	int sig;
2088 
2089 	p = td->td_proc;
2090 	sig = ksi->ksi_signo;
2091 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
2092 
2093 	sigfastblock_fetch(td);
2094 	PROC_LOCK(p);
2095 	ps = p->p_sigacts;
2096 	mtx_lock(&ps->ps_mtx);
2097 	sigmask = td->td_sigmask;
2098 	if (td->td_sigblock_val != 0)
2099 		SIGSETOR(sigmask, fastblock_mask);
2100 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2101 	    !SIGISMEMBER(sigmask, sig)) {
2102 #ifdef KTRACE
2103 		if (KTRPOINT(curthread, KTR_PSIG))
2104 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2105 			    &td->td_sigmask, ksi->ksi_code);
2106 #endif
2107 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2108 		    ksi, &td->td_sigmask);
2109 		postsig_done(sig, td, ps);
2110 		mtx_unlock(&ps->ps_mtx);
2111 	} else {
2112 		/*
2113 		 * Avoid a possible infinite loop if the thread
2114 		 * masking the signal or process is ignoring the
2115 		 * signal.
2116 		 */
2117 		if (kern_forcesigexit && (SIGISMEMBER(sigmask, sig) ||
2118 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2119 			SIGDELSET(td->td_sigmask, sig);
2120 			SIGDELSET(ps->ps_sigcatch, sig);
2121 			SIGDELSET(ps->ps_sigignore, sig);
2122 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2123 			td->td_pflags &= ~TDP_SIGFASTBLOCK;
2124 			td->td_sigblock_val = 0;
2125 		}
2126 		mtx_unlock(&ps->ps_mtx);
2127 		p->p_sig = sig;		/* XXX to verify code */
2128 		tdsendsignal(p, td, sig, ksi);
2129 	}
2130 	PROC_UNLOCK(p);
2131 }
2132 
2133 static struct thread *
2134 sigtd(struct proc *p, int sig, bool fast_sigblock)
2135 {
2136 	struct thread *td, *signal_td;
2137 
2138 	PROC_LOCK_ASSERT(p, MA_OWNED);
2139 	MPASS(!fast_sigblock || p == curproc);
2140 
2141 	/*
2142 	 * Check if current thread can handle the signal without
2143 	 * switching context to another thread.
2144 	 */
2145 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig) &&
2146 	    (!fast_sigblock || curthread->td_sigblock_val == 0))
2147 		return (curthread);
2148 	signal_td = NULL;
2149 	FOREACH_THREAD_IN_PROC(p, td) {
2150 		if (!SIGISMEMBER(td->td_sigmask, sig) && (!fast_sigblock ||
2151 		    td != curthread || td->td_sigblock_val == 0)) {
2152 			signal_td = td;
2153 			break;
2154 		}
2155 	}
2156 	if (signal_td == NULL)
2157 		signal_td = FIRST_THREAD_IN_PROC(p);
2158 	return (signal_td);
2159 }
2160 
2161 /*
2162  * Send the signal to the process.  If the signal has an action, the action
2163  * is usually performed by the target process rather than the caller; we add
2164  * the signal to the set of pending signals for the process.
2165  *
2166  * Exceptions:
2167  *   o When a stop signal is sent to a sleeping process that takes the
2168  *     default action, the process is stopped without awakening it.
2169  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2170  *     regardless of the signal action (eg, blocked or ignored).
2171  *
2172  * Other ignored signals are discarded immediately.
2173  *
2174  * NB: This function may be entered from the debugger via the "kill" DDB
2175  * command.  There is little that can be done to mitigate the possibly messy
2176  * side effects of this unwise possibility.
2177  */
2178 void
2179 kern_psignal(struct proc *p, int sig)
2180 {
2181 	ksiginfo_t ksi;
2182 
2183 	ksiginfo_init(&ksi);
2184 	ksi.ksi_signo = sig;
2185 	ksi.ksi_code = SI_KERNEL;
2186 	(void) tdsendsignal(p, NULL, sig, &ksi);
2187 }
2188 
2189 int
2190 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2191 {
2192 
2193 	return (tdsendsignal(p, NULL, sig, ksi));
2194 }
2195 
2196 /* Utility function for finding a thread to send signal event to. */
2197 int
2198 sigev_findtd(struct proc *p, struct sigevent *sigev, struct thread **ttd)
2199 {
2200 	struct thread *td;
2201 
2202 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2203 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2204 		if (td == NULL)
2205 			return (ESRCH);
2206 		*ttd = td;
2207 	} else {
2208 		*ttd = NULL;
2209 		PROC_LOCK(p);
2210 	}
2211 	return (0);
2212 }
2213 
2214 void
2215 tdsignal(struct thread *td, int sig)
2216 {
2217 	ksiginfo_t ksi;
2218 
2219 	ksiginfo_init(&ksi);
2220 	ksi.ksi_signo = sig;
2221 	ksi.ksi_code = SI_KERNEL;
2222 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2223 }
2224 
2225 void
2226 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2227 {
2228 
2229 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2230 }
2231 
2232 static int
2233 sig_sleepq_abort(struct thread *td, int intrval)
2234 {
2235 	THREAD_LOCK_ASSERT(td, MA_OWNED);
2236 
2237 	if (intrval == 0 && (td->td_flags & TDF_SIGWAIT) == 0) {
2238 		thread_unlock(td);
2239 		return (0);
2240 	}
2241 	return (sleepq_abort(td, intrval));
2242 }
2243 
2244 int
2245 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2246 {
2247 	sig_t action;
2248 	sigqueue_t *sigqueue;
2249 	int prop;
2250 	struct sigacts *ps;
2251 	int intrval;
2252 	int ret = 0;
2253 	int wakeup_swapper;
2254 
2255 	MPASS(td == NULL || p == td->td_proc);
2256 	PROC_LOCK_ASSERT(p, MA_OWNED);
2257 
2258 	if (!_SIG_VALID(sig))
2259 		panic("%s(): invalid signal %d", __func__, sig);
2260 
2261 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2262 
2263 	/*
2264 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2265 	 */
2266 	if (p->p_state == PRS_ZOMBIE) {
2267 		if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2268 			ksiginfo_tryfree(ksi);
2269 		return (ret);
2270 	}
2271 
2272 	ps = p->p_sigacts;
2273 	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2274 	prop = sigprop(sig);
2275 
2276 	if (td == NULL) {
2277 		td = sigtd(p, sig, false);
2278 		sigqueue = &p->p_sigqueue;
2279 	} else
2280 		sigqueue = &td->td_sigqueue;
2281 
2282 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2283 
2284 	/*
2285 	 * If the signal is being ignored, then we forget about it
2286 	 * immediately, except when the target process executes
2287 	 * sigwait().  (Note: we don't set SIGCONT in ps_sigignore,
2288 	 * and if it is set to SIG_IGN, action will be SIG_DFL here.)
2289 	 */
2290 	mtx_lock(&ps->ps_mtx);
2291 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2292 		if (kern_sig_discard_ign &&
2293 		    (p->p_sysent->sv_flags & SV_SIG_DISCIGN) == 0) {
2294 			SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2295 
2296 			mtx_unlock(&ps->ps_mtx);
2297 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2298 				ksiginfo_tryfree(ksi);
2299 			return (ret);
2300 		} else {
2301 			action = SIG_CATCH;
2302 			intrval = 0;
2303 		}
2304 	} else {
2305 		if (SIGISMEMBER(td->td_sigmask, sig))
2306 			action = SIG_HOLD;
2307 		else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2308 			action = SIG_CATCH;
2309 		else
2310 			action = SIG_DFL;
2311 		if (SIGISMEMBER(ps->ps_sigintr, sig))
2312 			intrval = EINTR;
2313 		else
2314 			intrval = ERESTART;
2315 	}
2316 	mtx_unlock(&ps->ps_mtx);
2317 
2318 	if (prop & SIGPROP_CONT)
2319 		sigqueue_delete_stopmask_proc(p);
2320 	else if (prop & SIGPROP_STOP) {
2321 		/*
2322 		 * If sending a tty stop signal to a member of an orphaned
2323 		 * process group, discard the signal here if the action
2324 		 * is default; don't stop the process below if sleeping,
2325 		 * and don't clear any pending SIGCONT.
2326 		 */
2327 		if ((prop & SIGPROP_TTYSTOP) != 0 &&
2328 		    (p->p_pgrp->pg_flags & PGRP_ORPHANED) != 0 &&
2329 		    action == SIG_DFL) {
2330 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2331 				ksiginfo_tryfree(ksi);
2332 			return (ret);
2333 		}
2334 		sigqueue_delete_proc(p, SIGCONT);
2335 		if (p->p_flag & P_CONTINUED) {
2336 			p->p_flag &= ~P_CONTINUED;
2337 			PROC_LOCK(p->p_pptr);
2338 			sigqueue_take(p->p_ksi);
2339 			PROC_UNLOCK(p->p_pptr);
2340 		}
2341 	}
2342 
2343 	ret = sigqueue_add(sigqueue, sig, ksi);
2344 	if (ret != 0)
2345 		return (ret);
2346 	signotify(td);
2347 	/*
2348 	 * Defer further processing for signals which are held,
2349 	 * except that stopped processes must be continued by SIGCONT.
2350 	 */
2351 	if (action == SIG_HOLD &&
2352 	    !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2353 		return (ret);
2354 
2355 	wakeup_swapper = 0;
2356 
2357 	/*
2358 	 * Some signals have a process-wide effect and a per-thread
2359 	 * component.  Most processing occurs when the process next
2360 	 * tries to cross the user boundary, however there are some
2361 	 * times when processing needs to be done immediately, such as
2362 	 * waking up threads so that they can cross the user boundary.
2363 	 * We try to do the per-process part here.
2364 	 */
2365 	if (P_SHOULDSTOP(p)) {
2366 		KASSERT(!(p->p_flag & P_WEXIT),
2367 		    ("signal to stopped but exiting process"));
2368 		if (sig == SIGKILL) {
2369 			/*
2370 			 * If traced process is already stopped,
2371 			 * then no further action is necessary.
2372 			 */
2373 			if (p->p_flag & P_TRACED)
2374 				goto out;
2375 			/*
2376 			 * SIGKILL sets process running.
2377 			 * It will die elsewhere.
2378 			 * All threads must be restarted.
2379 			 */
2380 			p->p_flag &= ~P_STOPPED_SIG;
2381 			goto runfast;
2382 		}
2383 
2384 		if (prop & SIGPROP_CONT) {
2385 			/*
2386 			 * If traced process is already stopped,
2387 			 * then no further action is necessary.
2388 			 */
2389 			if (p->p_flag & P_TRACED)
2390 				goto out;
2391 			/*
2392 			 * If SIGCONT is default (or ignored), we continue the
2393 			 * process but don't leave the signal in sigqueue as
2394 			 * it has no further action.  If SIGCONT is held, we
2395 			 * continue the process and leave the signal in
2396 			 * sigqueue.  If the process catches SIGCONT, let it
2397 			 * handle the signal itself.  If it isn't waiting on
2398 			 * an event, it goes back to run state.
2399 			 * Otherwise, process goes back to sleep state.
2400 			 */
2401 			p->p_flag &= ~P_STOPPED_SIG;
2402 			PROC_SLOCK(p);
2403 			if (p->p_numthreads == p->p_suspcount) {
2404 				PROC_SUNLOCK(p);
2405 				p->p_flag |= P_CONTINUED;
2406 				p->p_xsig = SIGCONT;
2407 				PROC_LOCK(p->p_pptr);
2408 				childproc_continued(p);
2409 				PROC_UNLOCK(p->p_pptr);
2410 				PROC_SLOCK(p);
2411 			}
2412 			if (action == SIG_DFL) {
2413 				thread_unsuspend(p);
2414 				PROC_SUNLOCK(p);
2415 				sigqueue_delete(sigqueue, sig);
2416 				goto out_cont;
2417 			}
2418 			if (action == SIG_CATCH) {
2419 				/*
2420 				 * The process wants to catch it so it needs
2421 				 * to run at least one thread, but which one?
2422 				 */
2423 				PROC_SUNLOCK(p);
2424 				goto runfast;
2425 			}
2426 			/*
2427 			 * The signal is not ignored or caught.
2428 			 */
2429 			thread_unsuspend(p);
2430 			PROC_SUNLOCK(p);
2431 			goto out_cont;
2432 		}
2433 
2434 		if (prop & SIGPROP_STOP) {
2435 			/*
2436 			 * If traced process is already stopped,
2437 			 * then no further action is necessary.
2438 			 */
2439 			if (p->p_flag & P_TRACED)
2440 				goto out;
2441 			/*
2442 			 * Already stopped, don't need to stop again
2443 			 * (If we did the shell could get confused).
2444 			 * Just make sure the signal STOP bit set.
2445 			 */
2446 			p->p_flag |= P_STOPPED_SIG;
2447 			sigqueue_delete(sigqueue, sig);
2448 			goto out;
2449 		}
2450 
2451 		/*
2452 		 * All other kinds of signals:
2453 		 * If a thread is sleeping interruptibly, simulate a
2454 		 * wakeup so that when it is continued it will be made
2455 		 * runnable and can look at the signal.  However, don't make
2456 		 * the PROCESS runnable, leave it stopped.
2457 		 * It may run a bit until it hits a thread_suspend_check().
2458 		 */
2459 		PROC_SLOCK(p);
2460 		thread_lock(td);
2461 		if (TD_CAN_ABORT(td))
2462 			wakeup_swapper = sig_sleepq_abort(td, intrval);
2463 		else
2464 			thread_unlock(td);
2465 		PROC_SUNLOCK(p);
2466 		goto out;
2467 		/*
2468 		 * Mutexes are short lived. Threads waiting on them will
2469 		 * hit thread_suspend_check() soon.
2470 		 */
2471 	} else if (p->p_state == PRS_NORMAL) {
2472 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2473 			tdsigwakeup(td, sig, action, intrval);
2474 			goto out;
2475 		}
2476 
2477 		MPASS(action == SIG_DFL);
2478 
2479 		if (prop & SIGPROP_STOP) {
2480 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2481 				goto out;
2482 			p->p_flag |= P_STOPPED_SIG;
2483 			p->p_xsig = sig;
2484 			PROC_SLOCK(p);
2485 			wakeup_swapper = sig_suspend_threads(td, p);
2486 			if (p->p_numthreads == p->p_suspcount) {
2487 				/*
2488 				 * only thread sending signal to another
2489 				 * process can reach here, if thread is sending
2490 				 * signal to its process, because thread does
2491 				 * not suspend itself here, p_numthreads
2492 				 * should never be equal to p_suspcount.
2493 				 */
2494 				thread_stopped(p);
2495 				PROC_SUNLOCK(p);
2496 				sigqueue_delete_proc(p, p->p_xsig);
2497 			} else
2498 				PROC_SUNLOCK(p);
2499 			goto out;
2500 		}
2501 	} else {
2502 		/* Not in "NORMAL" state. discard the signal. */
2503 		sigqueue_delete(sigqueue, sig);
2504 		goto out;
2505 	}
2506 
2507 	/*
2508 	 * The process is not stopped so we need to apply the signal to all the
2509 	 * running threads.
2510 	 */
2511 runfast:
2512 	tdsigwakeup(td, sig, action, intrval);
2513 	PROC_SLOCK(p);
2514 	thread_unsuspend(p);
2515 	PROC_SUNLOCK(p);
2516 out_cont:
2517 	itimer_proc_continue(p);
2518 	kqtimer_proc_continue(p);
2519 out:
2520 	/* If we jump here, proc slock should not be owned. */
2521 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2522 	if (wakeup_swapper)
2523 		kick_proc0();
2524 
2525 	return (ret);
2526 }
2527 
2528 /*
2529  * The force of a signal has been directed against a single
2530  * thread.  We need to see what we can do about knocking it
2531  * out of any sleep it may be in etc.
2532  */
2533 static void
2534 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2535 {
2536 	struct proc *p = td->td_proc;
2537 	int prop, wakeup_swapper;
2538 
2539 	PROC_LOCK_ASSERT(p, MA_OWNED);
2540 	prop = sigprop(sig);
2541 
2542 	PROC_SLOCK(p);
2543 	thread_lock(td);
2544 	/*
2545 	 * Bring the priority of a thread up if we want it to get
2546 	 * killed in this lifetime.  Be careful to avoid bumping the
2547 	 * priority of the idle thread, since we still allow to signal
2548 	 * kernel processes.
2549 	 */
2550 	if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2551 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2552 		sched_prio(td, PUSER);
2553 	if (TD_ON_SLEEPQ(td)) {
2554 		/*
2555 		 * If thread is sleeping uninterruptibly
2556 		 * we can't interrupt the sleep... the signal will
2557 		 * be noticed when the process returns through
2558 		 * trap() or syscall().
2559 		 */
2560 		if ((td->td_flags & TDF_SINTR) == 0)
2561 			goto out;
2562 		/*
2563 		 * If SIGCONT is default (or ignored) and process is
2564 		 * asleep, we are finished; the process should not
2565 		 * be awakened.
2566 		 */
2567 		if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2568 			thread_unlock(td);
2569 			PROC_SUNLOCK(p);
2570 			sigqueue_delete(&p->p_sigqueue, sig);
2571 			/*
2572 			 * It may be on either list in this state.
2573 			 * Remove from both for now.
2574 			 */
2575 			sigqueue_delete(&td->td_sigqueue, sig);
2576 			return;
2577 		}
2578 
2579 		/*
2580 		 * Don't awaken a sleeping thread for SIGSTOP if the
2581 		 * STOP signal is deferred.
2582 		 */
2583 		if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2584 		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2585 			goto out;
2586 
2587 		/*
2588 		 * Give low priority threads a better chance to run.
2589 		 */
2590 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2591 			sched_prio(td, PUSER);
2592 
2593 		wakeup_swapper = sig_sleepq_abort(td, intrval);
2594 		PROC_SUNLOCK(p);
2595 		if (wakeup_swapper)
2596 			kick_proc0();
2597 		return;
2598 	}
2599 
2600 	/*
2601 	 * Other states do nothing with the signal immediately,
2602 	 * other than kicking ourselves if we are running.
2603 	 * It will either never be noticed, or noticed very soon.
2604 	 */
2605 #ifdef SMP
2606 	if (TD_IS_RUNNING(td) && td != curthread)
2607 		forward_signal(td);
2608 #endif
2609 
2610 out:
2611 	PROC_SUNLOCK(p);
2612 	thread_unlock(td);
2613 }
2614 
2615 static void
2616 ptrace_coredump(struct thread *td)
2617 {
2618 	struct proc *p;
2619 	struct thr_coredump_req *tcq;
2620 	void *rl_cookie;
2621 
2622 	MPASS(td == curthread);
2623 	p = td->td_proc;
2624 	PROC_LOCK_ASSERT(p, MA_OWNED);
2625 	if ((td->td_dbgflags & TDB_COREDUMPRQ) == 0)
2626 		return;
2627 	KASSERT((p->p_flag & P_STOPPED_TRACE) != 0, ("not stopped"));
2628 
2629 	tcq = td->td_coredump;
2630 	KASSERT(tcq != NULL, ("td_coredump is NULL"));
2631 
2632 	if (p->p_sysent->sv_coredump == NULL) {
2633 		tcq->tc_error = ENOSYS;
2634 		goto wake;
2635 	}
2636 
2637 	PROC_UNLOCK(p);
2638 	rl_cookie = vn_rangelock_wlock(tcq->tc_vp, 0, OFF_MAX);
2639 
2640 	tcq->tc_error = p->p_sysent->sv_coredump(td, tcq->tc_vp,
2641 	    tcq->tc_limit, tcq->tc_flags);
2642 
2643 	vn_rangelock_unlock(tcq->tc_vp, rl_cookie);
2644 	PROC_LOCK(p);
2645 wake:
2646 	td->td_dbgflags &= ~TDB_COREDUMPRQ;
2647 	td->td_coredump = NULL;
2648 	wakeup(p);
2649 }
2650 
2651 static int
2652 sig_suspend_threads(struct thread *td, struct proc *p)
2653 {
2654 	struct thread *td2;
2655 	int wakeup_swapper;
2656 
2657 	PROC_LOCK_ASSERT(p, MA_OWNED);
2658 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2659 
2660 	wakeup_swapper = 0;
2661 	FOREACH_THREAD_IN_PROC(p, td2) {
2662 		thread_lock(td2);
2663 		ast_sched_locked(td2, TDA_SUSPEND);
2664 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2665 		    (td2->td_flags & TDF_SINTR)) {
2666 			if (td2->td_flags & TDF_SBDRY) {
2667 				/*
2668 				 * Once a thread is asleep with
2669 				 * TDF_SBDRY and without TDF_SERESTART
2670 				 * or TDF_SEINTR set, it should never
2671 				 * become suspended due to this check.
2672 				 */
2673 				KASSERT(!TD_IS_SUSPENDED(td2),
2674 				    ("thread with deferred stops suspended"));
2675 				if (TD_SBDRY_INTR(td2)) {
2676 					wakeup_swapper |= sleepq_abort(td2,
2677 					    TD_SBDRY_ERRNO(td2));
2678 					continue;
2679 				}
2680 			} else if (!TD_IS_SUSPENDED(td2))
2681 				thread_suspend_one(td2);
2682 		} else if (!TD_IS_SUSPENDED(td2)) {
2683 #ifdef SMP
2684 			if (TD_IS_RUNNING(td2) && td2 != td)
2685 				forward_signal(td2);
2686 #endif
2687 		}
2688 		thread_unlock(td2);
2689 	}
2690 	return (wakeup_swapper);
2691 }
2692 
2693 /*
2694  * Stop the process for an event deemed interesting to the debugger. If si is
2695  * non-NULL, this is a signal exchange; the new signal requested by the
2696  * debugger will be returned for handling. If si is NULL, this is some other
2697  * type of interesting event. The debugger may request a signal be delivered in
2698  * that case as well, however it will be deferred until it can be handled.
2699  */
2700 int
2701 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2702 {
2703 	struct proc *p = td->td_proc;
2704 	struct thread *td2;
2705 	ksiginfo_t ksi;
2706 
2707 	PROC_LOCK_ASSERT(p, MA_OWNED);
2708 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2709 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2710 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2711 
2712 	td->td_xsig = sig;
2713 
2714 	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2715 		td->td_dbgflags |= TDB_XSIG;
2716 		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2717 		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2718 		PROC_SLOCK(p);
2719 		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2720 			if (P_KILLED(p)) {
2721 				/*
2722 				 * Ensure that, if we've been PT_KILLed, the
2723 				 * exit status reflects that. Another thread
2724 				 * may also be in ptracestop(), having just
2725 				 * received the SIGKILL, but this thread was
2726 				 * unsuspended first.
2727 				 */
2728 				td->td_dbgflags &= ~TDB_XSIG;
2729 				td->td_xsig = SIGKILL;
2730 				p->p_ptevents = 0;
2731 				break;
2732 			}
2733 			if (p->p_flag & P_SINGLE_EXIT &&
2734 			    !(td->td_dbgflags & TDB_EXIT)) {
2735 				/*
2736 				 * Ignore ptrace stops except for thread exit
2737 				 * events when the process exits.
2738 				 */
2739 				td->td_dbgflags &= ~TDB_XSIG;
2740 				PROC_SUNLOCK(p);
2741 				return (0);
2742 			}
2743 
2744 			/*
2745 			 * Make wait(2) work.  Ensure that right after the
2746 			 * attach, the thread which was decided to become the
2747 			 * leader of attach gets reported to the waiter.
2748 			 * Otherwise, just avoid overwriting another thread's
2749 			 * assignment to p_xthread.  If another thread has
2750 			 * already set p_xthread, the current thread will get
2751 			 * a chance to report itself upon the next iteration.
2752 			 */
2753 			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2754 			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2755 			    p->p_xthread == NULL)) {
2756 				p->p_xsig = sig;
2757 				p->p_xthread = td;
2758 
2759 				/*
2760 				 * If we are on sleepqueue already,
2761 				 * let sleepqueue code decide if it
2762 				 * needs to go sleep after attach.
2763 				 */
2764 				if (td->td_wchan == NULL)
2765 					td->td_dbgflags &= ~TDB_FSTP;
2766 
2767 				p->p_flag2 &= ~P2_PTRACE_FSTP;
2768 				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2769 				sig_suspend_threads(td, p);
2770 			}
2771 			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2772 				td->td_dbgflags &= ~TDB_STOPATFORK;
2773 			}
2774 stopme:
2775 			td->td_dbgflags |= TDB_SSWITCH;
2776 			thread_suspend_switch(td, p);
2777 			td->td_dbgflags &= ~TDB_SSWITCH;
2778 			if ((td->td_dbgflags & TDB_COREDUMPRQ) != 0) {
2779 				PROC_SUNLOCK(p);
2780 				ptrace_coredump(td);
2781 				PROC_SLOCK(p);
2782 				goto stopme;
2783 			}
2784 			if (p->p_xthread == td)
2785 				p->p_xthread = NULL;
2786 			if (!(p->p_flag & P_TRACED))
2787 				break;
2788 			if (td->td_dbgflags & TDB_SUSPEND) {
2789 				if (p->p_flag & P_SINGLE_EXIT)
2790 					break;
2791 				goto stopme;
2792 			}
2793 		}
2794 		PROC_SUNLOCK(p);
2795 	}
2796 
2797 	if (si != NULL && sig == td->td_xsig) {
2798 		/* Parent wants us to take the original signal unchanged. */
2799 		si->ksi_flags |= KSI_HEAD;
2800 		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2801 			si->ksi_signo = 0;
2802 	} else if (td->td_xsig != 0) {
2803 		/*
2804 		 * If parent wants us to take a new signal, then it will leave
2805 		 * it in td->td_xsig; otherwise we just look for signals again.
2806 		 */
2807 		ksiginfo_init(&ksi);
2808 		ksi.ksi_signo = td->td_xsig;
2809 		ksi.ksi_flags |= KSI_PTRACE;
2810 		td2 = sigtd(p, td->td_xsig, false);
2811 		tdsendsignal(p, td2, td->td_xsig, &ksi);
2812 		if (td != td2)
2813 			return (0);
2814 	}
2815 
2816 	return (td->td_xsig);
2817 }
2818 
2819 static void
2820 reschedule_signals(struct proc *p, sigset_t block, int flags)
2821 {
2822 	struct sigacts *ps;
2823 	struct thread *td;
2824 	int sig;
2825 	bool fastblk, pslocked;
2826 
2827 	PROC_LOCK_ASSERT(p, MA_OWNED);
2828 	ps = p->p_sigacts;
2829 	pslocked = (flags & SIGPROCMASK_PS_LOCKED) != 0;
2830 	mtx_assert(&ps->ps_mtx, pslocked ? MA_OWNED : MA_NOTOWNED);
2831 	if (SIGISEMPTY(p->p_siglist))
2832 		return;
2833 	SIGSETAND(block, p->p_siglist);
2834 	fastblk = (flags & SIGPROCMASK_FASTBLK) != 0;
2835 	SIG_FOREACH(sig, &block) {
2836 		td = sigtd(p, sig, fastblk);
2837 
2838 		/*
2839 		 * If sigtd() selected us despite sigfastblock is
2840 		 * blocking, do not activate AST or wake us, to avoid
2841 		 * loop in AST handler.
2842 		 */
2843 		if (fastblk && td == curthread)
2844 			continue;
2845 
2846 		signotify(td);
2847 		if (!pslocked)
2848 			mtx_lock(&ps->ps_mtx);
2849 		if (p->p_flag & P_TRACED ||
2850 		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2851 		    !SIGISMEMBER(td->td_sigmask, sig))) {
2852 			tdsigwakeup(td, sig, SIG_CATCH,
2853 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2854 			    ERESTART));
2855 		}
2856 		if (!pslocked)
2857 			mtx_unlock(&ps->ps_mtx);
2858 	}
2859 }
2860 
2861 void
2862 tdsigcleanup(struct thread *td)
2863 {
2864 	struct proc *p;
2865 	sigset_t unblocked;
2866 
2867 	p = td->td_proc;
2868 	PROC_LOCK_ASSERT(p, MA_OWNED);
2869 
2870 	sigqueue_flush(&td->td_sigqueue);
2871 	if (p->p_numthreads == 1)
2872 		return;
2873 
2874 	/*
2875 	 * Since we cannot handle signals, notify signal post code
2876 	 * about this by filling the sigmask.
2877 	 *
2878 	 * Also, if needed, wake up thread(s) that do not block the
2879 	 * same signals as the exiting thread, since the thread might
2880 	 * have been selected for delivery and woken up.
2881 	 */
2882 	SIGFILLSET(unblocked);
2883 	SIGSETNAND(unblocked, td->td_sigmask);
2884 	SIGFILLSET(td->td_sigmask);
2885 	reschedule_signals(p, unblocked, 0);
2886 
2887 }
2888 
2889 static int
2890 sigdeferstop_curr_flags(int cflags)
2891 {
2892 
2893 	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
2894 	    (cflags & TDF_SBDRY) != 0);
2895 	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
2896 }
2897 
2898 /*
2899  * Defer the delivery of SIGSTOP for the current thread, according to
2900  * the requested mode.  Returns previous flags, which must be restored
2901  * by sigallowstop().
2902  *
2903  * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
2904  * cleared by the current thread, which allow the lock-less read-only
2905  * accesses below.
2906  */
2907 int
2908 sigdeferstop_impl(int mode)
2909 {
2910 	struct thread *td;
2911 	int cflags, nflags;
2912 
2913 	td = curthread;
2914 	cflags = sigdeferstop_curr_flags(td->td_flags);
2915 	switch (mode) {
2916 	case SIGDEFERSTOP_NOP:
2917 		nflags = cflags;
2918 		break;
2919 	case SIGDEFERSTOP_OFF:
2920 		nflags = 0;
2921 		break;
2922 	case SIGDEFERSTOP_SILENT:
2923 		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
2924 		break;
2925 	case SIGDEFERSTOP_EINTR:
2926 		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
2927 		break;
2928 	case SIGDEFERSTOP_ERESTART:
2929 		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
2930 		break;
2931 	default:
2932 		panic("sigdeferstop: invalid mode %x", mode);
2933 		break;
2934 	}
2935 	if (cflags == nflags)
2936 		return (SIGDEFERSTOP_VAL_NCHG);
2937 	thread_lock(td);
2938 	td->td_flags = (td->td_flags & ~cflags) | nflags;
2939 	thread_unlock(td);
2940 	return (cflags);
2941 }
2942 
2943 /*
2944  * Restores the STOP handling mode, typically permitting the delivery
2945  * of SIGSTOP for the current thread.  This does not immediately
2946  * suspend if a stop was posted.  Instead, the thread will suspend
2947  * either via ast() or a subsequent interruptible sleep.
2948  */
2949 void
2950 sigallowstop_impl(int prev)
2951 {
2952 	struct thread *td;
2953 	int cflags;
2954 
2955 	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
2956 	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
2957 	    ("sigallowstop: incorrect previous mode %x", prev));
2958 	td = curthread;
2959 	cflags = sigdeferstop_curr_flags(td->td_flags);
2960 	if (cflags != prev) {
2961 		thread_lock(td);
2962 		td->td_flags = (td->td_flags & ~cflags) | prev;
2963 		thread_unlock(td);
2964 	}
2965 }
2966 
2967 enum sigstatus {
2968 	SIGSTATUS_HANDLE,
2969 	SIGSTATUS_HANDLED,
2970 	SIGSTATUS_IGNORE,
2971 	SIGSTATUS_SBDRY_STOP,
2972 };
2973 
2974 /*
2975  * The thread has signal "sig" pending.  Figure out what to do with it:
2976  *
2977  * _HANDLE     -> the caller should handle the signal
2978  * _HANDLED    -> handled internally, reload pending signal set
2979  * _IGNORE     -> ignored, remove from the set of pending signals and try the
2980  *                next pending signal
2981  * _SBDRY_STOP -> the signal should stop the thread but this is not
2982  *                permitted in the current context
2983  */
2984 static enum sigstatus
2985 sigprocess(struct thread *td, int sig)
2986 {
2987 	struct proc *p;
2988 	struct sigacts *ps;
2989 	struct sigqueue *queue;
2990 	ksiginfo_t ksi;
2991 	int prop;
2992 
2993 	KASSERT(_SIG_VALID(sig), ("%s: invalid signal %d", __func__, sig));
2994 
2995 	p = td->td_proc;
2996 	ps = p->p_sigacts;
2997 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2998 	PROC_LOCK_ASSERT(p, MA_OWNED);
2999 
3000 	/*
3001 	 * We should allow pending but ignored signals below
3002 	 * if there is sigwait() active, or P_TRACED was
3003 	 * on when they were posted.
3004 	 */
3005 	if (SIGISMEMBER(ps->ps_sigignore, sig) &&
3006 	    (p->p_flag & P_TRACED) == 0 &&
3007 	    (td->td_flags & TDF_SIGWAIT) == 0) {
3008 		return (SIGSTATUS_IGNORE);
3009 	}
3010 
3011 	/*
3012 	 * If the process is going to single-thread mode to prepare
3013 	 * for exit, there is no sense in delivering any signal
3014 	 * to usermode.  Another important consequence is that
3015 	 * msleep(..., PCATCH, ...) now is only interruptible by a
3016 	 * suspend request.
3017 	 */
3018 	if ((p->p_flag2 & P2_WEXIT) != 0)
3019 		return (SIGSTATUS_IGNORE);
3020 
3021 	if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
3022 		/*
3023 		 * If traced, always stop.
3024 		 * Remove old signal from queue before the stop.
3025 		 * XXX shrug off debugger, it causes siginfo to
3026 		 * be thrown away.
3027 		 */
3028 		queue = &td->td_sigqueue;
3029 		ksiginfo_init(&ksi);
3030 		if (sigqueue_get(queue, sig, &ksi) == 0) {
3031 			queue = &p->p_sigqueue;
3032 			sigqueue_get(queue, sig, &ksi);
3033 		}
3034 		td->td_si = ksi.ksi_info;
3035 
3036 		mtx_unlock(&ps->ps_mtx);
3037 		sig = ptracestop(td, sig, &ksi);
3038 		mtx_lock(&ps->ps_mtx);
3039 
3040 		td->td_si.si_signo = 0;
3041 
3042 		/*
3043 		 * Keep looking if the debugger discarded or
3044 		 * replaced the signal.
3045 		 */
3046 		if (sig == 0)
3047 			return (SIGSTATUS_HANDLED);
3048 
3049 		/*
3050 		 * If the signal became masked, re-queue it.
3051 		 */
3052 		if (SIGISMEMBER(td->td_sigmask, sig)) {
3053 			ksi.ksi_flags |= KSI_HEAD;
3054 			sigqueue_add(&p->p_sigqueue, sig, &ksi);
3055 			return (SIGSTATUS_HANDLED);
3056 		}
3057 
3058 		/*
3059 		 * If the traced bit got turned off, requeue the signal and
3060 		 * reload the set of pending signals.  This ensures that p_sig*
3061 		 * and p_sigact are consistent.
3062 		 */
3063 		if ((p->p_flag & P_TRACED) == 0) {
3064 			if ((ksi.ksi_flags & KSI_PTRACE) == 0) {
3065 				ksi.ksi_flags |= KSI_HEAD;
3066 				sigqueue_add(queue, sig, &ksi);
3067 			}
3068 			return (SIGSTATUS_HANDLED);
3069 		}
3070 	}
3071 
3072 	/*
3073 	 * Decide whether the signal should be returned.
3074 	 * Return the signal's number, or fall through
3075 	 * to clear it from the pending mask.
3076 	 */
3077 	switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
3078 	case (intptr_t)SIG_DFL:
3079 		/*
3080 		 * Don't take default actions on system processes.
3081 		 */
3082 		if (p->p_pid <= 1) {
3083 #ifdef DIAGNOSTIC
3084 			/*
3085 			 * Are you sure you want to ignore SIGSEGV
3086 			 * in init? XXX
3087 			 */
3088 			printf("Process (pid %lu) got signal %d\n",
3089 				(u_long)p->p_pid, sig);
3090 #endif
3091 			return (SIGSTATUS_IGNORE);
3092 		}
3093 
3094 		/*
3095 		 * If there is a pending stop signal to process with
3096 		 * default action, stop here, then clear the signal.
3097 		 * Traced or exiting processes should ignore stops.
3098 		 * Additionally, a member of an orphaned process group
3099 		 * should ignore tty stops.
3100 		 */
3101 		prop = sigprop(sig);
3102 		if (prop & SIGPROP_STOP) {
3103 			mtx_unlock(&ps->ps_mtx);
3104 			if ((p->p_flag & (P_TRACED | P_WEXIT |
3105 			    P_SINGLE_EXIT)) != 0 || ((p->p_pgrp->
3106 			    pg_flags & PGRP_ORPHANED) != 0 &&
3107 			    (prop & SIGPROP_TTYSTOP) != 0)) {
3108 				mtx_lock(&ps->ps_mtx);
3109 				return (SIGSTATUS_IGNORE);
3110 			}
3111 			if (TD_SBDRY_INTR(td)) {
3112 				KASSERT((td->td_flags & TDF_SBDRY) != 0,
3113 				    ("lost TDF_SBDRY"));
3114 				mtx_lock(&ps->ps_mtx);
3115 				return (SIGSTATUS_SBDRY_STOP);
3116 			}
3117 			WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
3118 			    &p->p_mtx.lock_object, "Catching SIGSTOP");
3119 			sigqueue_delete(&td->td_sigqueue, sig);
3120 			sigqueue_delete(&p->p_sigqueue, sig);
3121 			p->p_flag |= P_STOPPED_SIG;
3122 			p->p_xsig = sig;
3123 			PROC_SLOCK(p);
3124 			sig_suspend_threads(td, p);
3125 			thread_suspend_switch(td, p);
3126 			PROC_SUNLOCK(p);
3127 			mtx_lock(&ps->ps_mtx);
3128 			return (SIGSTATUS_HANDLED);
3129 		} else if ((prop & SIGPROP_IGNORE) != 0 &&
3130 		    (td->td_flags & TDF_SIGWAIT) == 0) {
3131 			/*
3132 			 * Default action is to ignore; drop it if
3133 			 * not in kern_sigtimedwait().
3134 			 */
3135 			return (SIGSTATUS_IGNORE);
3136 		} else {
3137 			return (SIGSTATUS_HANDLE);
3138 		}
3139 
3140 	case (intptr_t)SIG_IGN:
3141 		if ((td->td_flags & TDF_SIGWAIT) == 0)
3142 			return (SIGSTATUS_IGNORE);
3143 		else
3144 			return (SIGSTATUS_HANDLE);
3145 
3146 	default:
3147 		/*
3148 		 * This signal has an action, let postsig() process it.
3149 		 */
3150 		return (SIGSTATUS_HANDLE);
3151 	}
3152 }
3153 
3154 /*
3155  * If the current process has received a signal (should be caught or cause
3156  * termination, should interrupt current syscall), return the signal number.
3157  * Stop signals with default action are processed immediately, then cleared;
3158  * they aren't returned.  This is checked after each entry to the system for
3159  * a syscall or trap (though this can usually be done without calling
3160  * issignal by checking the pending signal masks in cursig.) The normal call
3161  * sequence is
3162  *
3163  *	while (sig = cursig(curthread))
3164  *		postsig(sig);
3165  */
3166 static int
3167 issignal(struct thread *td)
3168 {
3169 	struct proc *p;
3170 	sigset_t sigpending;
3171 	int sig;
3172 
3173 	p = td->td_proc;
3174 	PROC_LOCK_ASSERT(p, MA_OWNED);
3175 
3176 	for (;;) {
3177 		sigpending = td->td_sigqueue.sq_signals;
3178 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
3179 		SIGSETNAND(sigpending, td->td_sigmask);
3180 
3181 		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
3182 		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
3183 			SIG_STOPSIGMASK(sigpending);
3184 		if (SIGISEMPTY(sigpending))	/* no signal to send */
3185 			return (0);
3186 
3187 		/*
3188 		 * Do fast sigblock if requested by usermode.  Since
3189 		 * we do know that there was a signal pending at this
3190 		 * point, set the FAST_SIGBLOCK_PEND as indicator for
3191 		 * usermode to perform a dummy call to
3192 		 * FAST_SIGBLOCK_UNBLOCK, which causes immediate
3193 		 * delivery of postponed pending signal.
3194 		 */
3195 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
3196 			if (td->td_sigblock_val != 0)
3197 				SIGSETNAND(sigpending, fastblock_mask);
3198 			if (SIGISEMPTY(sigpending)) {
3199 				td->td_pflags |= TDP_SIGFASTPENDING;
3200 				return (0);
3201 			}
3202 		}
3203 
3204 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
3205 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
3206 		    SIGISMEMBER(sigpending, SIGSTOP)) {
3207 			/*
3208 			 * If debugger just attached, always consume
3209 			 * SIGSTOP from ptrace(PT_ATTACH) first, to
3210 			 * execute the debugger attach ritual in
3211 			 * order.
3212 			 */
3213 			td->td_dbgflags |= TDB_FSTP;
3214 			SIGEMPTYSET(sigpending);
3215 			SIGADDSET(sigpending, SIGSTOP);
3216 		}
3217 
3218 		SIG_FOREACH(sig, &sigpending) {
3219 			switch (sigprocess(td, sig)) {
3220 			case SIGSTATUS_HANDLE:
3221 				return (sig);
3222 			case SIGSTATUS_HANDLED:
3223 				goto next;
3224 			case SIGSTATUS_IGNORE:
3225 				sigqueue_delete(&td->td_sigqueue, sig);
3226 				sigqueue_delete(&p->p_sigqueue, sig);
3227 				break;
3228 			case SIGSTATUS_SBDRY_STOP:
3229 				return (-1);
3230 			}
3231 		}
3232 next:;
3233 	}
3234 }
3235 
3236 void
3237 thread_stopped(struct proc *p)
3238 {
3239 	int n;
3240 
3241 	PROC_LOCK_ASSERT(p, MA_OWNED);
3242 	PROC_SLOCK_ASSERT(p, MA_OWNED);
3243 	n = p->p_suspcount;
3244 	if (p == curproc)
3245 		n++;
3246 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3247 		PROC_SUNLOCK(p);
3248 		p->p_flag &= ~P_WAITED;
3249 		PROC_LOCK(p->p_pptr);
3250 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
3251 			CLD_TRAPPED : CLD_STOPPED);
3252 		PROC_UNLOCK(p->p_pptr);
3253 		PROC_SLOCK(p);
3254 	}
3255 }
3256 
3257 /*
3258  * Take the action for the specified signal
3259  * from the current set of pending signals.
3260  */
3261 int
3262 postsig(int sig)
3263 {
3264 	struct thread *td;
3265 	struct proc *p;
3266 	struct sigacts *ps;
3267 	sig_t action;
3268 	ksiginfo_t ksi;
3269 	sigset_t returnmask;
3270 
3271 	KASSERT(sig != 0, ("postsig"));
3272 
3273 	td = curthread;
3274 	p = td->td_proc;
3275 	PROC_LOCK_ASSERT(p, MA_OWNED);
3276 	ps = p->p_sigacts;
3277 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3278 	ksiginfo_init(&ksi);
3279 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3280 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3281 		return (0);
3282 	ksi.ksi_signo = sig;
3283 	if (ksi.ksi_code == SI_TIMER)
3284 		itimer_accept(p, ksi.ksi_timerid, &ksi);
3285 	action = ps->ps_sigact[_SIG_IDX(sig)];
3286 #ifdef KTRACE
3287 	if (KTRPOINT(td, KTR_PSIG))
3288 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3289 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3290 #endif
3291 
3292 	if (action == SIG_DFL) {
3293 		/*
3294 		 * Default action, where the default is to kill
3295 		 * the process.  (Other cases were ignored above.)
3296 		 */
3297 		mtx_unlock(&ps->ps_mtx);
3298 		proc_td_siginfo_capture(td, &ksi.ksi_info);
3299 		sigexit(td, sig);
3300 		/* NOTREACHED */
3301 	} else {
3302 		/*
3303 		 * If we get here, the signal must be caught.
3304 		 */
3305 		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3306 		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3307 		    ("postsig action: blocked sig %d", sig));
3308 
3309 		/*
3310 		 * Set the new mask value and also defer further
3311 		 * occurrences of this signal.
3312 		 *
3313 		 * Special case: user has done a sigsuspend.  Here the
3314 		 * current mask is not of interest, but rather the
3315 		 * mask from before the sigsuspend is what we want
3316 		 * restored after the signal processing is completed.
3317 		 */
3318 		if (td->td_pflags & TDP_OLDMASK) {
3319 			returnmask = td->td_oldsigmask;
3320 			td->td_pflags &= ~TDP_OLDMASK;
3321 		} else
3322 			returnmask = td->td_sigmask;
3323 
3324 		if (p->p_sig == sig) {
3325 			p->p_sig = 0;
3326 		}
3327 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3328 		postsig_done(sig, td, ps);
3329 	}
3330 	return (1);
3331 }
3332 
3333 int
3334 sig_ast_checksusp(struct thread *td)
3335 {
3336 	struct proc *p __diagused;
3337 	int ret;
3338 
3339 	p = td->td_proc;
3340 	PROC_LOCK_ASSERT(p, MA_OWNED);
3341 
3342 	if (!td_ast_pending(td, TDA_SUSPEND))
3343 		return (0);
3344 
3345 	ret = thread_suspend_check(1);
3346 	MPASS(ret == 0 || ret == EINTR || ret == ERESTART);
3347 	return (ret);
3348 }
3349 
3350 int
3351 sig_ast_needsigchk(struct thread *td)
3352 {
3353 	struct proc *p;
3354 	struct sigacts *ps;
3355 	int ret, sig;
3356 
3357 	p = td->td_proc;
3358 	PROC_LOCK_ASSERT(p, MA_OWNED);
3359 
3360 	if (!td_ast_pending(td, TDA_SIG))
3361 		return (0);
3362 
3363 	ps = p->p_sigacts;
3364 	mtx_lock(&ps->ps_mtx);
3365 	sig = cursig(td);
3366 	if (sig == -1) {
3367 		mtx_unlock(&ps->ps_mtx);
3368 		KASSERT((td->td_flags & TDF_SBDRY) != 0, ("lost TDF_SBDRY"));
3369 		KASSERT(TD_SBDRY_INTR(td),
3370 		    ("lost TDF_SERESTART of TDF_SEINTR"));
3371 		KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
3372 		    (TDF_SEINTR | TDF_SERESTART),
3373 		    ("both TDF_SEINTR and TDF_SERESTART"));
3374 		ret = TD_SBDRY_ERRNO(td);
3375 	} else if (sig != 0) {
3376 		ret = SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR : ERESTART;
3377 		mtx_unlock(&ps->ps_mtx);
3378 	} else {
3379 		mtx_unlock(&ps->ps_mtx);
3380 		ret = 0;
3381 	}
3382 
3383 	/*
3384 	 * Do not go into sleep if this thread was the ptrace(2)
3385 	 * attach leader.  cursig() consumed SIGSTOP from PT_ATTACH,
3386 	 * but we usually act on the signal by interrupting sleep, and
3387 	 * should do that here as well.
3388 	 */
3389 	if ((td->td_dbgflags & TDB_FSTP) != 0) {
3390 		if (ret == 0)
3391 			ret = EINTR;
3392 		td->td_dbgflags &= ~TDB_FSTP;
3393 	}
3394 
3395 	return (ret);
3396 }
3397 
3398 int
3399 sig_intr(void)
3400 {
3401 	struct thread *td;
3402 	struct proc *p;
3403 	int ret;
3404 
3405 	td = curthread;
3406 	if (!td_ast_pending(td, TDA_SIG) && !td_ast_pending(td, TDA_SUSPEND))
3407 		return (0);
3408 
3409 	p = td->td_proc;
3410 
3411 	PROC_LOCK(p);
3412 	ret = sig_ast_checksusp(td);
3413 	if (ret == 0)
3414 		ret = sig_ast_needsigchk(td);
3415 	PROC_UNLOCK(p);
3416 	return (ret);
3417 }
3418 
3419 bool
3420 curproc_sigkilled(void)
3421 {
3422 	struct thread *td;
3423 	struct proc *p;
3424 	struct sigacts *ps;
3425 	bool res;
3426 
3427 	td = curthread;
3428 	if (!td_ast_pending(td, TDA_SIG))
3429 		return (false);
3430 
3431 	p = td->td_proc;
3432 	PROC_LOCK(p);
3433 	ps = p->p_sigacts;
3434 	mtx_lock(&ps->ps_mtx);
3435 	res = SIGISMEMBER(td->td_sigqueue.sq_signals, SIGKILL) ||
3436 	    SIGISMEMBER(p->p_sigqueue.sq_signals, SIGKILL);
3437 	mtx_unlock(&ps->ps_mtx);
3438 	PROC_UNLOCK(p);
3439 	return (res);
3440 }
3441 
3442 void
3443 proc_wkilled(struct proc *p)
3444 {
3445 
3446 	PROC_LOCK_ASSERT(p, MA_OWNED);
3447 	if ((p->p_flag & P_WKILLED) == 0) {
3448 		p->p_flag |= P_WKILLED;
3449 		/*
3450 		 * Notify swapper that there is a process to swap in.
3451 		 * The notification is racy, at worst it would take 10
3452 		 * seconds for the swapper process to notice.
3453 		 */
3454 		if ((p->p_flag & (P_INMEM | P_SWAPPINGIN)) == 0)
3455 			wakeup(&proc0);
3456 	}
3457 }
3458 
3459 /*
3460  * Kill the current process for stated reason.
3461  */
3462 void
3463 killproc(struct proc *p, const char *why)
3464 {
3465 
3466 	PROC_LOCK_ASSERT(p, MA_OWNED);
3467 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3468 	    p->p_comm);
3469 	log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3470 	    p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3471 	    p->p_ucred->cr_uid, why);
3472 	proc_wkilled(p);
3473 	kern_psignal(p, SIGKILL);
3474 }
3475 
3476 /*
3477  * Force the current process to exit with the specified signal, dumping core
3478  * if appropriate.  We bypass the normal tests for masked and caught signals,
3479  * allowing unrecoverable failures to terminate the process without changing
3480  * signal state.  Mark the accounting record with the signal termination.
3481  * If dumping core, save the signal number for the debugger.  Calls exit and
3482  * does not return.
3483  */
3484 void
3485 sigexit(struct thread *td, int sig)
3486 {
3487 	struct proc *p = td->td_proc;
3488 
3489 	PROC_LOCK_ASSERT(p, MA_OWNED);
3490 	proc_set_p2_wexit(p);
3491 
3492 	p->p_acflag |= AXSIG;
3493 	/*
3494 	 * We must be single-threading to generate a core dump.  This
3495 	 * ensures that the registers in the core file are up-to-date.
3496 	 * Also, the ELF dump handler assumes that the thread list doesn't
3497 	 * change out from under it.
3498 	 *
3499 	 * XXX If another thread attempts to single-thread before us
3500 	 *     (e.g. via fork()), we won't get a dump at all.
3501 	 */
3502 	if ((sigprop(sig) & SIGPROP_CORE) &&
3503 	    thread_single(p, SINGLE_NO_EXIT) == 0) {
3504 		p->p_sig = sig;
3505 		/*
3506 		 * Log signals which would cause core dumps
3507 		 * (Log as LOG_INFO to appease those who don't want
3508 		 * these messages.)
3509 		 * XXX : Todo, as well as euid, write out ruid too
3510 		 * Note that coredump() drops proc lock.
3511 		 */
3512 		if (coredump(td) == 0)
3513 			sig |= WCOREFLAG;
3514 		if (kern_logsigexit)
3515 			log(LOG_INFO,
3516 			    "pid %d (%s), jid %d, uid %d: exited on "
3517 			    "signal %d%s\n", p->p_pid, p->p_comm,
3518 			    p->p_ucred->cr_prison->pr_id,
3519 			    td->td_ucred->cr_uid,
3520 			    sig &~ WCOREFLAG,
3521 			    sig & WCOREFLAG ? " (core dumped)" : "");
3522 	} else
3523 		PROC_UNLOCK(p);
3524 	exit1(td, 0, sig);
3525 	/* NOTREACHED */
3526 }
3527 
3528 /*
3529  * Send queued SIGCHLD to parent when child process's state
3530  * is changed.
3531  */
3532 static void
3533 sigparent(struct proc *p, int reason, int status)
3534 {
3535 	PROC_LOCK_ASSERT(p, MA_OWNED);
3536 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3537 
3538 	if (p->p_ksi != NULL) {
3539 		p->p_ksi->ksi_signo  = SIGCHLD;
3540 		p->p_ksi->ksi_code   = reason;
3541 		p->p_ksi->ksi_status = status;
3542 		p->p_ksi->ksi_pid    = p->p_pid;
3543 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3544 		if (KSI_ONQ(p->p_ksi))
3545 			return;
3546 	}
3547 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3548 }
3549 
3550 static void
3551 childproc_jobstate(struct proc *p, int reason, int sig)
3552 {
3553 	struct sigacts *ps;
3554 
3555 	PROC_LOCK_ASSERT(p, MA_OWNED);
3556 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3557 
3558 	/*
3559 	 * Wake up parent sleeping in kern_wait(), also send
3560 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3561 	 * that parent will awake, because parent may masked
3562 	 * the signal.
3563 	 */
3564 	p->p_pptr->p_flag |= P_STATCHILD;
3565 	wakeup(p->p_pptr);
3566 
3567 	ps = p->p_pptr->p_sigacts;
3568 	mtx_lock(&ps->ps_mtx);
3569 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3570 		mtx_unlock(&ps->ps_mtx);
3571 		sigparent(p, reason, sig);
3572 	} else
3573 		mtx_unlock(&ps->ps_mtx);
3574 }
3575 
3576 void
3577 childproc_stopped(struct proc *p, int reason)
3578 {
3579 
3580 	childproc_jobstate(p, reason, p->p_xsig);
3581 }
3582 
3583 void
3584 childproc_continued(struct proc *p)
3585 {
3586 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3587 }
3588 
3589 void
3590 childproc_exited(struct proc *p)
3591 {
3592 	int reason, status;
3593 
3594 	if (WCOREDUMP(p->p_xsig)) {
3595 		reason = CLD_DUMPED;
3596 		status = WTERMSIG(p->p_xsig);
3597 	} else if (WIFSIGNALED(p->p_xsig)) {
3598 		reason = CLD_KILLED;
3599 		status = WTERMSIG(p->p_xsig);
3600 	} else {
3601 		reason = CLD_EXITED;
3602 		status = p->p_xexit;
3603 	}
3604 	/*
3605 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3606 	 * done in exit1().
3607 	 */
3608 	sigparent(p, reason, status);
3609 }
3610 
3611 #define	MAX_NUM_CORE_FILES 100000
3612 #ifndef NUM_CORE_FILES
3613 #define	NUM_CORE_FILES 5
3614 #endif
3615 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3616 static int num_cores = NUM_CORE_FILES;
3617 
3618 static int
3619 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3620 {
3621 	int error;
3622 	int new_val;
3623 
3624 	new_val = num_cores;
3625 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3626 	if (error != 0 || req->newptr == NULL)
3627 		return (error);
3628 	if (new_val > MAX_NUM_CORE_FILES)
3629 		new_val = MAX_NUM_CORE_FILES;
3630 	if (new_val < 0)
3631 		new_val = 0;
3632 	num_cores = new_val;
3633 	return (0);
3634 }
3635 SYSCTL_PROC(_debug, OID_AUTO, ncores,
3636     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
3637     sysctl_debug_num_cores_check, "I",
3638     "Maximum number of generated process corefiles while using index format");
3639 
3640 #define	GZIP_SUFFIX	".gz"
3641 #define	ZSTD_SUFFIX	".zst"
3642 
3643 int compress_user_cores = 0;
3644 
3645 static int
3646 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3647 {
3648 	int error, val;
3649 
3650 	val = compress_user_cores;
3651 	error = sysctl_handle_int(oidp, &val, 0, req);
3652 	if (error != 0 || req->newptr == NULL)
3653 		return (error);
3654 	if (val != 0 && !compressor_avail(val))
3655 		return (EINVAL);
3656 	compress_user_cores = val;
3657 	return (error);
3658 }
3659 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores,
3660     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, 0, sizeof(int),
3661     sysctl_compress_user_cores, "I",
3662     "Enable compression of user corefiles ("
3663     __XSTRING(COMPRESS_GZIP) " = gzip, "
3664     __XSTRING(COMPRESS_ZSTD) " = zstd)");
3665 
3666 int compress_user_cores_level = 6;
3667 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3668     &compress_user_cores_level, 0,
3669     "Corefile compression level");
3670 
3671 /*
3672  * Protect the access to corefilename[] by allproc_lock.
3673  */
3674 #define	corefilename_lock	allproc_lock
3675 
3676 static char corefilename[MAXPATHLEN] = {"%N.core"};
3677 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3678 
3679 static int
3680 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3681 {
3682 	int error;
3683 
3684 	sx_xlock(&corefilename_lock);
3685 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3686 	    req);
3687 	sx_xunlock(&corefilename_lock);
3688 
3689 	return (error);
3690 }
3691 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3692     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3693     "Process corefile name format string");
3694 
3695 static void
3696 vnode_close_locked(struct thread *td, struct vnode *vp)
3697 {
3698 
3699 	VOP_UNLOCK(vp);
3700 	vn_close(vp, FWRITE, td->td_ucred, td);
3701 }
3702 
3703 /*
3704  * If the core format has a %I in it, then we need to check
3705  * for existing corefiles before defining a name.
3706  * To do this we iterate over 0..ncores to find a
3707  * non-existing core file name to use. If all core files are
3708  * already used we choose the oldest one.
3709  */
3710 static int
3711 corefile_open_last(struct thread *td, char *name, int indexpos,
3712     int indexlen, int ncores, struct vnode **vpp)
3713 {
3714 	struct vnode *oldvp, *nextvp, *vp;
3715 	struct vattr vattr;
3716 	struct nameidata nd;
3717 	int error, i, flags, oflags, cmode;
3718 	char ch;
3719 	struct timespec lasttime;
3720 
3721 	nextvp = oldvp = NULL;
3722 	cmode = S_IRUSR | S_IWUSR;
3723 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3724 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3725 
3726 	for (i = 0; i < ncores; i++) {
3727 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3728 
3729 		ch = name[indexpos + indexlen];
3730 		(void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3731 		    i);
3732 		name[indexpos + indexlen] = ch;
3733 
3734 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
3735 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3736 		    NULL);
3737 		if (error != 0)
3738 			break;
3739 
3740 		vp = nd.ni_vp;
3741 		NDFREE_PNBUF(&nd);
3742 		if ((flags & O_CREAT) == O_CREAT) {
3743 			nextvp = vp;
3744 			break;
3745 		}
3746 
3747 		error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3748 		if (error != 0) {
3749 			vnode_close_locked(td, vp);
3750 			break;
3751 		}
3752 
3753 		if (oldvp == NULL ||
3754 		    lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3755 		    (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3756 		    lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3757 			if (oldvp != NULL)
3758 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3759 			oldvp = vp;
3760 			VOP_UNLOCK(oldvp);
3761 			lasttime = vattr.va_mtime;
3762 		} else {
3763 			vnode_close_locked(td, vp);
3764 		}
3765 	}
3766 
3767 	if (oldvp != NULL) {
3768 		if (nextvp == NULL) {
3769 			if ((td->td_proc->p_flag & P_SUGID) != 0) {
3770 				error = EFAULT;
3771 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3772 			} else {
3773 				nextvp = oldvp;
3774 				error = vn_lock(nextvp, LK_EXCLUSIVE);
3775 				if (error != 0) {
3776 					vn_close(nextvp, FWRITE, td->td_ucred,
3777 					    td);
3778 					nextvp = NULL;
3779 				}
3780 			}
3781 		} else {
3782 			vn_close(oldvp, FWRITE, td->td_ucred, td);
3783 		}
3784 	}
3785 	if (error != 0) {
3786 		if (nextvp != NULL)
3787 			vnode_close_locked(td, oldvp);
3788 	} else {
3789 		*vpp = nextvp;
3790 	}
3791 
3792 	return (error);
3793 }
3794 
3795 /*
3796  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3797  * Expand the name described in corefilename, using name, uid, and pid
3798  * and open/create core file.
3799  * corefilename is a printf-like string, with three format specifiers:
3800  *	%N	name of process ("name")
3801  *	%P	process id (pid)
3802  *	%U	user id (uid)
3803  * For example, "%N.core" is the default; they can be disabled completely
3804  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3805  * This is controlled by the sysctl variable kern.corefile (see above).
3806  */
3807 static int
3808 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3809     int compress, int signum, struct vnode **vpp, char **namep)
3810 {
3811 	struct sbuf sb;
3812 	struct nameidata nd;
3813 	const char *format;
3814 	char *hostname, *name;
3815 	int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
3816 
3817 	hostname = NULL;
3818 	format = corefilename;
3819 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3820 	indexlen = 0;
3821 	indexpos = -1;
3822 	ncores = num_cores;
3823 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3824 	sx_slock(&corefilename_lock);
3825 	for (i = 0; format[i] != '\0'; i++) {
3826 		switch (format[i]) {
3827 		case '%':	/* Format character */
3828 			i++;
3829 			switch (format[i]) {
3830 			case '%':
3831 				sbuf_putc(&sb, '%');
3832 				break;
3833 			case 'H':	/* hostname */
3834 				if (hostname == NULL) {
3835 					hostname = malloc(MAXHOSTNAMELEN,
3836 					    M_TEMP, M_WAITOK);
3837 				}
3838 				getcredhostname(td->td_ucred, hostname,
3839 				    MAXHOSTNAMELEN);
3840 				sbuf_printf(&sb, "%s", hostname);
3841 				break;
3842 			case 'I':	/* autoincrementing index */
3843 				if (indexpos != -1) {
3844 					sbuf_printf(&sb, "%%I");
3845 					break;
3846 				}
3847 
3848 				indexpos = sbuf_len(&sb);
3849 				sbuf_printf(&sb, "%u", ncores - 1);
3850 				indexlen = sbuf_len(&sb) - indexpos;
3851 				break;
3852 			case 'N':	/* process name */
3853 				sbuf_printf(&sb, "%s", comm);
3854 				break;
3855 			case 'P':	/* process id */
3856 				sbuf_printf(&sb, "%u", pid);
3857 				break;
3858 			case 'S':	/* signal number */
3859 				sbuf_printf(&sb, "%i", signum);
3860 				break;
3861 			case 'U':	/* user id */
3862 				sbuf_printf(&sb, "%u", uid);
3863 				break;
3864 			default:
3865 				log(LOG_ERR,
3866 				    "Unknown format character %c in "
3867 				    "corename `%s'\n", format[i], format);
3868 				break;
3869 			}
3870 			break;
3871 		default:
3872 			sbuf_putc(&sb, format[i]);
3873 			break;
3874 		}
3875 	}
3876 	sx_sunlock(&corefilename_lock);
3877 	free(hostname, M_TEMP);
3878 	if (compress == COMPRESS_GZIP)
3879 		sbuf_printf(&sb, GZIP_SUFFIX);
3880 	else if (compress == COMPRESS_ZSTD)
3881 		sbuf_printf(&sb, ZSTD_SUFFIX);
3882 	if (sbuf_error(&sb) != 0) {
3883 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3884 		    "long\n", (long)pid, comm, (u_long)uid);
3885 		sbuf_delete(&sb);
3886 		free(name, M_TEMP);
3887 		return (ENOMEM);
3888 	}
3889 	sbuf_finish(&sb);
3890 	sbuf_delete(&sb);
3891 
3892 	if (indexpos != -1) {
3893 		error = corefile_open_last(td, name, indexpos, indexlen, ncores,
3894 		    vpp);
3895 		if (error != 0) {
3896 			log(LOG_ERR,
3897 			    "pid %d (%s), uid (%u):  Path `%s' failed "
3898 			    "on initial open test, error = %d\n",
3899 			    pid, comm, uid, name, error);
3900 		}
3901 	} else {
3902 		cmode = S_IRUSR | S_IWUSR;
3903 		oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3904 		    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3905 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3906 		if ((td->td_proc->p_flag & P_SUGID) != 0)
3907 			flags |= O_EXCL;
3908 
3909 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
3910 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3911 		    NULL);
3912 		if (error == 0) {
3913 			*vpp = nd.ni_vp;
3914 			NDFREE_PNBUF(&nd);
3915 		}
3916 	}
3917 
3918 	if (error != 0) {
3919 #ifdef AUDIT
3920 		audit_proc_coredump(td, name, error);
3921 #endif
3922 		free(name, M_TEMP);
3923 		return (error);
3924 	}
3925 	*namep = name;
3926 	return (0);
3927 }
3928 
3929 /*
3930  * Dump a process' core.  The main routine does some
3931  * policy checking, and creates the name of the coredump;
3932  * then it passes on a vnode and a size limit to the process-specific
3933  * coredump routine if there is one; if there _is not_ one, it returns
3934  * ENOSYS; otherwise it returns the error from the process-specific routine.
3935  */
3936 
3937 static int
3938 coredump(struct thread *td)
3939 {
3940 	struct proc *p = td->td_proc;
3941 	struct ucred *cred = td->td_ucred;
3942 	struct vnode *vp;
3943 	struct flock lf;
3944 	struct vattr vattr;
3945 	size_t fullpathsize;
3946 	int error, error1, locked;
3947 	char *name;			/* name of corefile */
3948 	void *rl_cookie;
3949 	off_t limit;
3950 	char *fullpath, *freepath = NULL;
3951 	struct sbuf *sb;
3952 
3953 	PROC_LOCK_ASSERT(p, MA_OWNED);
3954 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3955 
3956 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
3957 	    (p->p_flag2 & P2_NOTRACE) != 0) {
3958 		PROC_UNLOCK(p);
3959 		return (EFAULT);
3960 	}
3961 
3962 	/*
3963 	 * Note that the bulk of limit checking is done after
3964 	 * the corefile is created.  The exception is if the limit
3965 	 * for corefiles is 0, in which case we don't bother
3966 	 * creating the corefile at all.  This layout means that
3967 	 * a corefile is truncated instead of not being created,
3968 	 * if it is larger than the limit.
3969 	 */
3970 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
3971 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
3972 		PROC_UNLOCK(p);
3973 		return (EFBIG);
3974 	}
3975 	PROC_UNLOCK(p);
3976 
3977 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
3978 	    compress_user_cores, p->p_sig, &vp, &name);
3979 	if (error != 0)
3980 		return (error);
3981 
3982 	/*
3983 	 * Don't dump to non-regular files or files with links.
3984 	 * Do not dump into system files. Effective user must own the corefile.
3985 	 */
3986 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
3987 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
3988 	    vattr.va_uid != cred->cr_uid) {
3989 		VOP_UNLOCK(vp);
3990 		error = EFAULT;
3991 		goto out;
3992 	}
3993 
3994 	VOP_UNLOCK(vp);
3995 
3996 	/* Postpone other writers, including core dumps of other processes. */
3997 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
3998 
3999 	lf.l_whence = SEEK_SET;
4000 	lf.l_start = 0;
4001 	lf.l_len = 0;
4002 	lf.l_type = F_WRLCK;
4003 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
4004 
4005 	VATTR_NULL(&vattr);
4006 	vattr.va_size = 0;
4007 	if (set_core_nodump_flag)
4008 		vattr.va_flags = UF_NODUMP;
4009 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4010 	VOP_SETATTR(vp, &vattr, cred);
4011 	VOP_UNLOCK(vp);
4012 	PROC_LOCK(p);
4013 	p->p_acflag |= ACORE;
4014 	PROC_UNLOCK(p);
4015 
4016 	if (p->p_sysent->sv_coredump != NULL) {
4017 		error = p->p_sysent->sv_coredump(td, vp, limit, 0);
4018 	} else {
4019 		error = ENOSYS;
4020 	}
4021 
4022 	if (locked) {
4023 		lf.l_type = F_UNLCK;
4024 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
4025 	}
4026 	vn_rangelock_unlock(vp, rl_cookie);
4027 
4028 	/*
4029 	 * Notify the userland helper that a process triggered a core dump.
4030 	 * This allows the helper to run an automated debugging session.
4031 	 */
4032 	if (error != 0 || coredump_devctl == 0)
4033 		goto out;
4034 	sb = sbuf_new_auto();
4035 	if (vn_fullpath_global(p->p_textvp, &fullpath, &freepath) != 0)
4036 		goto out2;
4037 	sbuf_printf(sb, "comm=\"");
4038 	devctl_safe_quote_sb(sb, fullpath);
4039 	free(freepath, M_TEMP);
4040 	sbuf_printf(sb, "\" core=\"");
4041 
4042 	/*
4043 	 * We can't lookup core file vp directly. When we're replacing a core, and
4044 	 * other random times, we flush the name cache, so it will fail. Instead,
4045 	 * if the path of the core is relative, add the current dir in front if it.
4046 	 */
4047 	if (name[0] != '/') {
4048 		fullpathsize = MAXPATHLEN;
4049 		freepath = malloc(fullpathsize, M_TEMP, M_WAITOK);
4050 		if (vn_getcwd(freepath, &fullpath, &fullpathsize) != 0) {
4051 			free(freepath, M_TEMP);
4052 			goto out2;
4053 		}
4054 		devctl_safe_quote_sb(sb, fullpath);
4055 		free(freepath, M_TEMP);
4056 		sbuf_putc(sb, '/');
4057 	}
4058 	devctl_safe_quote_sb(sb, name);
4059 	sbuf_printf(sb, "\"");
4060 	if (sbuf_finish(sb) == 0)
4061 		devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
4062 out2:
4063 	sbuf_delete(sb);
4064 out:
4065 	error1 = vn_close(vp, FWRITE, cred, td);
4066 	if (error == 0)
4067 		error = error1;
4068 #ifdef AUDIT
4069 	audit_proc_coredump(td, name, error);
4070 #endif
4071 	free(name, M_TEMP);
4072 	return (error);
4073 }
4074 
4075 /*
4076  * Nonexistent system call-- signal process (may want to handle it).  Flag
4077  * error in case process won't see signal immediately (blocked or ignored).
4078  */
4079 #ifndef _SYS_SYSPROTO_H_
4080 struct nosys_args {
4081 	int	dummy;
4082 };
4083 #endif
4084 /* ARGSUSED */
4085 int
4086 nosys(struct thread *td, struct nosys_args *args)
4087 {
4088 	struct proc *p;
4089 
4090 	p = td->td_proc;
4091 
4092 	PROC_LOCK(p);
4093 	tdsignal(td, SIGSYS);
4094 	PROC_UNLOCK(p);
4095 	if (kern_lognosys == 1 || kern_lognosys == 3) {
4096 		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4097 		    td->td_sa.code);
4098 	}
4099 	if (kern_lognosys == 2 || kern_lognosys == 3 ||
4100 	    (p->p_pid == 1 && (kern_lognosys & 3) == 0)) {
4101 		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4102 		    td->td_sa.code);
4103 	}
4104 	return (ENOSYS);
4105 }
4106 
4107 /*
4108  * Send a SIGIO or SIGURG signal to a process or process group using stored
4109  * credentials rather than those of the current process.
4110  */
4111 void
4112 pgsigio(struct sigio **sigiop, int sig, int checkctty)
4113 {
4114 	ksiginfo_t ksi;
4115 	struct sigio *sigio;
4116 
4117 	ksiginfo_init(&ksi);
4118 	ksi.ksi_signo = sig;
4119 	ksi.ksi_code = SI_KERNEL;
4120 
4121 	SIGIO_LOCK();
4122 	sigio = *sigiop;
4123 	if (sigio == NULL) {
4124 		SIGIO_UNLOCK();
4125 		return;
4126 	}
4127 	if (sigio->sio_pgid > 0) {
4128 		PROC_LOCK(sigio->sio_proc);
4129 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
4130 			kern_psignal(sigio->sio_proc, sig);
4131 		PROC_UNLOCK(sigio->sio_proc);
4132 	} else if (sigio->sio_pgid < 0) {
4133 		struct proc *p;
4134 
4135 		PGRP_LOCK(sigio->sio_pgrp);
4136 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
4137 			PROC_LOCK(p);
4138 			if (p->p_state == PRS_NORMAL &&
4139 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
4140 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
4141 				kern_psignal(p, sig);
4142 			PROC_UNLOCK(p);
4143 		}
4144 		PGRP_UNLOCK(sigio->sio_pgrp);
4145 	}
4146 	SIGIO_UNLOCK();
4147 }
4148 
4149 static int
4150 filt_sigattach(struct knote *kn)
4151 {
4152 	struct proc *p = curproc;
4153 
4154 	kn->kn_ptr.p_proc = p;
4155 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
4156 
4157 	knlist_add(p->p_klist, kn, 0);
4158 
4159 	return (0);
4160 }
4161 
4162 static void
4163 filt_sigdetach(struct knote *kn)
4164 {
4165 	struct proc *p = kn->kn_ptr.p_proc;
4166 
4167 	knlist_remove(p->p_klist, kn, 0);
4168 }
4169 
4170 /*
4171  * signal knotes are shared with proc knotes, so we apply a mask to
4172  * the hint in order to differentiate them from process hints.  This
4173  * could be avoided by using a signal-specific knote list, but probably
4174  * isn't worth the trouble.
4175  */
4176 static int
4177 filt_signal(struct knote *kn, long hint)
4178 {
4179 
4180 	if (hint & NOTE_SIGNAL) {
4181 		hint &= ~NOTE_SIGNAL;
4182 
4183 		if (kn->kn_id == hint)
4184 			kn->kn_data++;
4185 	}
4186 	return (kn->kn_data != 0);
4187 }
4188 
4189 struct sigacts *
4190 sigacts_alloc(void)
4191 {
4192 	struct sigacts *ps;
4193 
4194 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
4195 	refcount_init(&ps->ps_refcnt, 1);
4196 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
4197 	return (ps);
4198 }
4199 
4200 void
4201 sigacts_free(struct sigacts *ps)
4202 {
4203 
4204 	if (refcount_release(&ps->ps_refcnt) == 0)
4205 		return;
4206 	mtx_destroy(&ps->ps_mtx);
4207 	free(ps, M_SUBPROC);
4208 }
4209 
4210 struct sigacts *
4211 sigacts_hold(struct sigacts *ps)
4212 {
4213 
4214 	refcount_acquire(&ps->ps_refcnt);
4215 	return (ps);
4216 }
4217 
4218 void
4219 sigacts_copy(struct sigacts *dest, struct sigacts *src)
4220 {
4221 
4222 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
4223 	mtx_lock(&src->ps_mtx);
4224 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
4225 	mtx_unlock(&src->ps_mtx);
4226 }
4227 
4228 int
4229 sigacts_shared(struct sigacts *ps)
4230 {
4231 
4232 	return (ps->ps_refcnt > 1);
4233 }
4234 
4235 void
4236 sig_drop_caught(struct proc *p)
4237 {
4238 	int sig;
4239 	struct sigacts *ps;
4240 
4241 	ps = p->p_sigacts;
4242 	PROC_LOCK_ASSERT(p, MA_OWNED);
4243 	mtx_assert(&ps->ps_mtx, MA_OWNED);
4244 	SIG_FOREACH(sig, &ps->ps_sigcatch) {
4245 		sigdflt(ps, sig);
4246 		if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
4247 			sigqueue_delete_proc(p, sig);
4248 	}
4249 }
4250 
4251 static void
4252 sigfastblock_failed(struct thread *td, bool sendsig, bool write)
4253 {
4254 	ksiginfo_t ksi;
4255 
4256 	/*
4257 	 * Prevent further fetches and SIGSEGVs, allowing thread to
4258 	 * issue syscalls despite corruption.
4259 	 */
4260 	sigfastblock_clear(td);
4261 
4262 	if (!sendsig)
4263 		return;
4264 	ksiginfo_init_trap(&ksi);
4265 	ksi.ksi_signo = SIGSEGV;
4266 	ksi.ksi_code = write ? SEGV_ACCERR : SEGV_MAPERR;
4267 	ksi.ksi_addr = td->td_sigblock_ptr;
4268 	trapsignal(td, &ksi);
4269 }
4270 
4271 static bool
4272 sigfastblock_fetch_sig(struct thread *td, bool sendsig, uint32_t *valp)
4273 {
4274 	uint32_t res;
4275 
4276 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4277 		return (true);
4278 	if (fueword32((void *)td->td_sigblock_ptr, &res) == -1) {
4279 		sigfastblock_failed(td, sendsig, false);
4280 		return (false);
4281 	}
4282 	*valp = res;
4283 	td->td_sigblock_val = res & ~SIGFASTBLOCK_FLAGS;
4284 	return (true);
4285 }
4286 
4287 static void
4288 sigfastblock_resched(struct thread *td, bool resched)
4289 {
4290 	struct proc *p;
4291 
4292 	if (resched) {
4293 		p = td->td_proc;
4294 		PROC_LOCK(p);
4295 		reschedule_signals(p, td->td_sigmask, 0);
4296 		PROC_UNLOCK(p);
4297 	}
4298 	ast_sched(td, TDA_SIG);
4299 }
4300 
4301 int
4302 sys_sigfastblock(struct thread *td, struct sigfastblock_args *uap)
4303 {
4304 	struct proc *p;
4305 	int error, res;
4306 	uint32_t oldval;
4307 
4308 	error = 0;
4309 	p = td->td_proc;
4310 	switch (uap->cmd) {
4311 	case SIGFASTBLOCK_SETPTR:
4312 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
4313 			error = EBUSY;
4314 			break;
4315 		}
4316 		if (((uintptr_t)(uap->ptr) & (sizeof(uint32_t) - 1)) != 0) {
4317 			error = EINVAL;
4318 			break;
4319 		}
4320 		td->td_pflags |= TDP_SIGFASTBLOCK;
4321 		td->td_sigblock_ptr = uap->ptr;
4322 		break;
4323 
4324 	case SIGFASTBLOCK_UNBLOCK:
4325 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4326 			error = EINVAL;
4327 			break;
4328 		}
4329 
4330 		for (;;) {
4331 			res = casueword32(td->td_sigblock_ptr,
4332 			    SIGFASTBLOCK_PEND, &oldval, 0);
4333 			if (res == -1) {
4334 				error = EFAULT;
4335 				sigfastblock_failed(td, false, true);
4336 				break;
4337 			}
4338 			if (res == 0)
4339 				break;
4340 			MPASS(res == 1);
4341 			if (oldval != SIGFASTBLOCK_PEND) {
4342 				error = EBUSY;
4343 				break;
4344 			}
4345 			error = thread_check_susp(td, false);
4346 			if (error != 0)
4347 				break;
4348 		}
4349 		if (error != 0)
4350 			break;
4351 
4352 		/*
4353 		 * td_sigblock_val is cleared there, but not on a
4354 		 * syscall exit.  The end effect is that a single
4355 		 * interruptible sleep, while user sigblock word is
4356 		 * set, might return EINTR or ERESTART to usermode
4357 		 * without delivering signal.  All further sleeps,
4358 		 * until userspace clears the word and does
4359 		 * sigfastblock(UNBLOCK), observe current word and no
4360 		 * longer get interrupted.  It is slight
4361 		 * non-conformance, with alternative to have read the
4362 		 * sigblock word on each syscall entry.
4363 		 */
4364 		td->td_sigblock_val = 0;
4365 
4366 		/*
4367 		 * Rely on normal ast mechanism to deliver pending
4368 		 * signals to current thread.  But notify others about
4369 		 * fake unblock.
4370 		 */
4371 		sigfastblock_resched(td, error == 0 && p->p_numthreads != 1);
4372 
4373 		break;
4374 
4375 	case SIGFASTBLOCK_UNSETPTR:
4376 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4377 			error = EINVAL;
4378 			break;
4379 		}
4380 		if (!sigfastblock_fetch_sig(td, false, &oldval)) {
4381 			error = EFAULT;
4382 			break;
4383 		}
4384 		if (oldval != 0 && oldval != SIGFASTBLOCK_PEND) {
4385 			error = EBUSY;
4386 			break;
4387 		}
4388 		sigfastblock_clear(td);
4389 		break;
4390 
4391 	default:
4392 		error = EINVAL;
4393 		break;
4394 	}
4395 	return (error);
4396 }
4397 
4398 void
4399 sigfastblock_clear(struct thread *td)
4400 {
4401 	bool resched;
4402 
4403 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4404 		return;
4405 	td->td_sigblock_val = 0;
4406 	resched = (td->td_pflags & TDP_SIGFASTPENDING) != 0 ||
4407 	    SIGPENDING(td);
4408 	td->td_pflags &= ~(TDP_SIGFASTBLOCK | TDP_SIGFASTPENDING);
4409 	sigfastblock_resched(td, resched);
4410 }
4411 
4412 void
4413 sigfastblock_fetch(struct thread *td)
4414 {
4415 	uint32_t val;
4416 
4417 	(void)sigfastblock_fetch_sig(td, true, &val);
4418 }
4419 
4420 static void
4421 sigfastblock_setpend1(struct thread *td)
4422 {
4423 	int res;
4424 	uint32_t oldval;
4425 
4426 	if ((td->td_pflags & TDP_SIGFASTPENDING) == 0)
4427 		return;
4428 	res = fueword32((void *)td->td_sigblock_ptr, &oldval);
4429 	if (res == -1) {
4430 		sigfastblock_failed(td, true, false);
4431 		return;
4432 	}
4433 	for (;;) {
4434 		res = casueword32(td->td_sigblock_ptr, oldval, &oldval,
4435 		    oldval | SIGFASTBLOCK_PEND);
4436 		if (res == -1) {
4437 			sigfastblock_failed(td, true, true);
4438 			return;
4439 		}
4440 		if (res == 0) {
4441 			td->td_sigblock_val = oldval & ~SIGFASTBLOCK_FLAGS;
4442 			td->td_pflags &= ~TDP_SIGFASTPENDING;
4443 			break;
4444 		}
4445 		MPASS(res == 1);
4446 		if (thread_check_susp(td, false) != 0)
4447 			break;
4448 	}
4449 }
4450 
4451 static void
4452 sigfastblock_setpend(struct thread *td, bool resched)
4453 {
4454 	struct proc *p;
4455 
4456 	sigfastblock_setpend1(td);
4457 	if (resched) {
4458 		p = td->td_proc;
4459 		PROC_LOCK(p);
4460 		reschedule_signals(p, fastblock_mask, SIGPROCMASK_FASTBLK);
4461 		PROC_UNLOCK(p);
4462 	}
4463 }
4464