xref: /freebsd/sys/kern/kern_sig.c (revision 59c8e88e72633afbc47a4ace0d2170d00d51f7dc)
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 
37 #include <sys/cdefs.h>
38 #include "opt_capsicum.h"
39 #include "opt_ktrace.h"
40 
41 #include <sys/param.h>
42 #include <sys/capsicum.h>
43 #include <sys/ctype.h>
44 #include <sys/systm.h>
45 #include <sys/signalvar.h>
46 #include <sys/vnode.h>
47 #include <sys/acct.h>
48 #include <sys/capsicum.h>
49 #include <sys/compressor.h>
50 #include <sys/condvar.h>
51 #include <sys/devctl.h>
52 #include <sys/event.h>
53 #include <sys/fcntl.h>
54 #include <sys/imgact.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/ktr.h>
58 #include <sys/ktrace.h>
59 #include <sys/limits.h>
60 #include <sys/lock.h>
61 #include <sys/malloc.h>
62 #include <sys/mutex.h>
63 #include <sys/refcount.h>
64 #include <sys/namei.h>
65 #include <sys/proc.h>
66 #include <sys/procdesc.h>
67 #include <sys/ptrace.h>
68 #include <sys/posix4.h>
69 #include <sys/racct.h>
70 #include <sys/resourcevar.h>
71 #include <sys/sdt.h>
72 #include <sys/sbuf.h>
73 #include <sys/sleepqueue.h>
74 #include <sys/smp.h>
75 #include <sys/stat.h>
76 #include <sys/sx.h>
77 #include <sys/syscall.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/vmmeter.h>
86 #include <sys/wait.h>
87 #include <vm/vm.h>
88 #include <vm/vm_extern.h>
89 #include <vm/uma.h>
90 
91 #include <machine/cpu.h>
92 
93 #include <security/audit/audit.h>
94 
95 #define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
96 
97 SDT_PROVIDER_DECLARE(proc);
98 SDT_PROBE_DEFINE3(proc, , , signal__send,
99     "struct thread *", "struct proc *", "int");
100 SDT_PROBE_DEFINE2(proc, , , signal__clear,
101     "int", "ksiginfo_t *");
102 SDT_PROBE_DEFINE3(proc, , , signal__discard,
103     "struct thread *", "struct proc *", "int");
104 
105 static int	coredump(struct thread *);
106 static int	killpg1(struct thread *td, int sig, int pgid, int all,
107 		    ksiginfo_t *ksi);
108 static int	issignal(struct thread *td);
109 static void	reschedule_signals(struct proc *p, sigset_t block, int flags);
110 static int	sigprop(int sig);
111 static void	tdsigwakeup(struct thread *, int, sig_t, int);
112 static int	sig_suspend_threads(struct thread *, struct proc *);
113 static int	filt_sigattach(struct knote *kn);
114 static void	filt_sigdetach(struct knote *kn);
115 static int	filt_signal(struct knote *kn, long hint);
116 static struct thread *sigtd(struct proc *p, int sig, bool fast_sigblock);
117 static void	sigqueue_start(void);
118 static void	sigfastblock_setpend(struct thread *td, bool resched);
119 
120 static uma_zone_t	ksiginfo_zone = NULL;
121 struct filterops sig_filtops = {
122 	.f_isfd = 0,
123 	.f_attach = filt_sigattach,
124 	.f_detach = filt_sigdetach,
125 	.f_event = filt_signal,
126 };
127 
128 static int	kern_logsigexit = 1;
129 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
130     &kern_logsigexit, 0,
131     "Log processes quitting on abnormal signals to syslog(3)");
132 
133 static int	kern_forcesigexit = 1;
134 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
135     &kern_forcesigexit, 0, "Force trap signal to be handled");
136 
137 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
138     "POSIX real time signal");
139 
140 static int	max_pending_per_proc = 128;
141 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
142     &max_pending_per_proc, 0, "Max pending signals per proc");
143 
144 static int	preallocate_siginfo = 1024;
145 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
146     &preallocate_siginfo, 0, "Preallocated signal memory size");
147 
148 static int	signal_overflow = 0;
149 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
150     &signal_overflow, 0, "Number of signals overflew");
151 
152 static int	signal_alloc_fail = 0;
153 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
154     &signal_alloc_fail, 0, "signals failed to be allocated");
155 
156 static int	kern_lognosys = 0;
157 SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
158     "Log invalid syscalls");
159 
160 static int	kern_signosys = 1;
161 SYSCTL_INT(_kern, OID_AUTO, signosys, CTLFLAG_RWTUN, &kern_signosys, 0,
162     "Send SIGSYS on return from invalid syscall");
163 
164 __read_frequently bool sigfastblock_fetch_always = false;
165 SYSCTL_BOOL(_kern, OID_AUTO, sigfastblock_fetch_always, CTLFLAG_RWTUN,
166     &sigfastblock_fetch_always, 0,
167     "Fetch sigfastblock word on each syscall entry for proper "
168     "blocking semantic");
169 
170 static bool	kern_sig_discard_ign = true;
171 SYSCTL_BOOL(_kern, OID_AUTO, sig_discard_ign, CTLFLAG_RWTUN,
172     &kern_sig_discard_ign, 0,
173     "Discard ignored signals on delivery, otherwise queue them to "
174     "the target queue");
175 
176 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
177 
178 /*
179  * Policy -- Can ucred cr1 send SIGIO to process cr2?
180  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
181  * in the right situations.
182  */
183 #define CANSIGIO(cr1, cr2) \
184 	((cr1)->cr_uid == 0 || \
185 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
186 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
187 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
188 	    (cr1)->cr_uid == (cr2)->cr_uid)
189 
190 static int	sugid_coredump;
191 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
192     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
193 
194 static int	capmode_coredump;
195 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
196     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
197 
198 static int	do_coredump = 1;
199 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
200 	&do_coredump, 0, "Enable/Disable coredumps");
201 
202 static int	set_core_nodump_flag = 0;
203 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
204 	0, "Enable setting the NODUMP flag on coredump files");
205 
206 static int	coredump_devctl = 0;
207 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
208 	0, "Generate a devctl notification when processes coredump");
209 
210 /*
211  * Signal properties and actions.
212  * The array below categorizes the signals and their default actions
213  * according to the following properties:
214  */
215 #define	SIGPROP_KILL		0x01	/* terminates process by default */
216 #define	SIGPROP_CORE		0x02	/* ditto and coredumps */
217 #define	SIGPROP_STOP		0x04	/* suspend process */
218 #define	SIGPROP_TTYSTOP		0x08	/* ditto, from tty */
219 #define	SIGPROP_IGNORE		0x10	/* ignore by default */
220 #define	SIGPROP_CONT		0x20	/* continue if suspended */
221 
222 static const int sigproptbl[NSIG] = {
223 	[SIGHUP] =	SIGPROP_KILL,
224 	[SIGINT] =	SIGPROP_KILL,
225 	[SIGQUIT] =	SIGPROP_KILL | SIGPROP_CORE,
226 	[SIGILL] =	SIGPROP_KILL | SIGPROP_CORE,
227 	[SIGTRAP] =	SIGPROP_KILL | SIGPROP_CORE,
228 	[SIGABRT] =	SIGPROP_KILL | SIGPROP_CORE,
229 	[SIGEMT] =	SIGPROP_KILL | SIGPROP_CORE,
230 	[SIGFPE] =	SIGPROP_KILL | SIGPROP_CORE,
231 	[SIGKILL] =	SIGPROP_KILL,
232 	[SIGBUS] =	SIGPROP_KILL | SIGPROP_CORE,
233 	[SIGSEGV] =	SIGPROP_KILL | SIGPROP_CORE,
234 	[SIGSYS] =	SIGPROP_KILL | SIGPROP_CORE,
235 	[SIGPIPE] =	SIGPROP_KILL,
236 	[SIGALRM] =	SIGPROP_KILL,
237 	[SIGTERM] =	SIGPROP_KILL,
238 	[SIGURG] =	SIGPROP_IGNORE,
239 	[SIGSTOP] =	SIGPROP_STOP,
240 	[SIGTSTP] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
241 	[SIGCONT] =	SIGPROP_IGNORE | SIGPROP_CONT,
242 	[SIGCHLD] =	SIGPROP_IGNORE,
243 	[SIGTTIN] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
244 	[SIGTTOU] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
245 	[SIGIO] =	SIGPROP_IGNORE,
246 	[SIGXCPU] =	SIGPROP_KILL,
247 	[SIGXFSZ] =	SIGPROP_KILL,
248 	[SIGVTALRM] =	SIGPROP_KILL,
249 	[SIGPROF] =	SIGPROP_KILL,
250 	[SIGWINCH] =	SIGPROP_IGNORE,
251 	[SIGINFO] =	SIGPROP_IGNORE,
252 	[SIGUSR1] =	SIGPROP_KILL,
253 	[SIGUSR2] =	SIGPROP_KILL,
254 };
255 
256 #define	_SIG_FOREACH_ADVANCE(i, set) ({					\
257 	int __found;							\
258 	for (;;) {							\
259 		if (__bits != 0) {					\
260 			int __sig = ffs(__bits);			\
261 			__bits &= ~(1u << (__sig - 1));			\
262 			sig = __i * sizeof((set)->__bits[0]) * NBBY + __sig; \
263 			__found = 1;					\
264 			break;						\
265 		}							\
266 		if (++__i == _SIG_WORDS) {				\
267 			__found = 0;					\
268 			break;						\
269 		}							\
270 		__bits = (set)->__bits[__i];				\
271 	}								\
272 	__found != 0;							\
273 })
274 
275 #define	SIG_FOREACH(i, set)						\
276 	for (int32_t __i = -1, __bits = 0;				\
277 	    _SIG_FOREACH_ADVANCE(i, set); )				\
278 
279 static sigset_t fastblock_mask;
280 
281 static void
282 ast_sig(struct thread *td, int tda)
283 {
284 	struct proc *p;
285 	int old_boundary, sig;
286 	bool resched_sigs;
287 
288 	p = td->td_proc;
289 
290 #ifdef DIAGNOSTIC
291 	if (p->p_numthreads == 1 && (tda & (TDAI(TDA_SIG) |
292 	    TDAI(TDA_AST))) == 0) {
293 		PROC_LOCK(p);
294 		thread_lock(td);
295 		/*
296 		 * Note that TDA_SIG should be re-read from
297 		 * td_ast, since signal might have been delivered
298 		 * after we cleared td_flags above.  This is one of
299 		 * the reason for looping check for AST condition.
300 		 * See comment in userret() about P_PPWAIT.
301 		 */
302 		if ((p->p_flag & P_PPWAIT) == 0 &&
303 		    (td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
304 			if (SIGPENDING(td) && ((tda | td->td_ast) &
305 			    (TDAI(TDA_SIG) | TDAI(TDA_AST))) == 0) {
306 				thread_unlock(td); /* fix dumps */
307 				panic(
308 				    "failed2 to set signal flags for ast p %p "
309 				    "td %p tda %#x td_ast %#x fl %#x",
310 				    p, td, tda, td->td_ast, td->td_flags);
311 			}
312 		}
313 		thread_unlock(td);
314 		PROC_UNLOCK(p);
315 	}
316 #endif
317 
318 	/*
319 	 * Check for signals. Unlocked reads of p_pendingcnt or
320 	 * p_siglist might cause process-directed signal to be handled
321 	 * later.
322 	 */
323 	if ((tda & TDAI(TDA_SIG)) != 0 || p->p_pendingcnt > 0 ||
324 	    !SIGISEMPTY(p->p_siglist)) {
325 		sigfastblock_fetch(td);
326 		PROC_LOCK(p);
327 		old_boundary = ~TDB_BOUNDARY | (td->td_dbgflags & TDB_BOUNDARY);
328 		td->td_dbgflags |= TDB_BOUNDARY;
329 		mtx_lock(&p->p_sigacts->ps_mtx);
330 		while ((sig = cursig(td)) != 0) {
331 			KASSERT(sig >= 0, ("sig %d", sig));
332 			postsig(sig);
333 		}
334 		mtx_unlock(&p->p_sigacts->ps_mtx);
335 		td->td_dbgflags &= old_boundary;
336 		PROC_UNLOCK(p);
337 		resched_sigs = true;
338 	} else {
339 		resched_sigs = false;
340 	}
341 
342 	/*
343 	 * Handle deferred update of the fast sigblock value, after
344 	 * the postsig() loop was performed.
345 	 */
346 	sigfastblock_setpend(td, resched_sigs);
347 }
348 
349 static void
350 ast_sigsuspend(struct thread *td, int tda __unused)
351 {
352 	MPASS((td->td_pflags & TDP_OLDMASK) != 0);
353 	td->td_pflags &= ~TDP_OLDMASK;
354 	kern_sigprocmask(td, SIG_SETMASK, &td->td_oldsigmask, NULL, 0);
355 }
356 
357 static void
358 sigqueue_start(void)
359 {
360 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
361 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
362 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
363 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
364 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
365 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
366 	SIGFILLSET(fastblock_mask);
367 	SIG_CANTMASK(fastblock_mask);
368 	ast_register(TDA_SIG, ASTR_UNCOND, 0, ast_sig);
369 	ast_register(TDA_SIGSUSPEND, ASTR_ASTF_REQUIRED | ASTR_TDP,
370 	    TDP_OLDMASK, ast_sigsuspend);
371 }
372 
373 ksiginfo_t *
374 ksiginfo_alloc(int mwait)
375 {
376 	MPASS(mwait == M_WAITOK || mwait == M_NOWAIT);
377 
378 	if (ksiginfo_zone == NULL)
379 		return (NULL);
380 	return (uma_zalloc(ksiginfo_zone, mwait | M_ZERO));
381 }
382 
383 void
384 ksiginfo_free(ksiginfo_t *ksi)
385 {
386 	uma_zfree(ksiginfo_zone, ksi);
387 }
388 
389 static __inline bool
390 ksiginfo_tryfree(ksiginfo_t *ksi)
391 {
392 	if ((ksi->ksi_flags & KSI_EXT) == 0) {
393 		uma_zfree(ksiginfo_zone, ksi);
394 		return (true);
395 	}
396 	return (false);
397 }
398 
399 void
400 sigqueue_init(sigqueue_t *list, struct proc *p)
401 {
402 	SIGEMPTYSET(list->sq_signals);
403 	SIGEMPTYSET(list->sq_kill);
404 	SIGEMPTYSET(list->sq_ptrace);
405 	TAILQ_INIT(&list->sq_list);
406 	list->sq_proc = p;
407 	list->sq_flags = SQ_INIT;
408 }
409 
410 /*
411  * Get a signal's ksiginfo.
412  * Return:
413  *	0	-	signal not found
414  *	others	-	signal number
415  */
416 static int
417 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
418 {
419 	struct proc *p = sq->sq_proc;
420 	struct ksiginfo *ksi, *next;
421 	int count = 0;
422 
423 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
424 
425 	if (!SIGISMEMBER(sq->sq_signals, signo))
426 		return (0);
427 
428 	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
429 		count++;
430 		SIGDELSET(sq->sq_ptrace, signo);
431 		si->ksi_flags |= KSI_PTRACE;
432 	}
433 	if (SIGISMEMBER(sq->sq_kill, signo)) {
434 		count++;
435 		if (count == 1)
436 			SIGDELSET(sq->sq_kill, signo);
437 	}
438 
439 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
440 		if (ksi->ksi_signo == signo) {
441 			if (count == 0) {
442 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
443 				ksi->ksi_sigq = NULL;
444 				ksiginfo_copy(ksi, si);
445 				if (ksiginfo_tryfree(ksi) && p != NULL)
446 					p->p_pendingcnt--;
447 			}
448 			if (++count > 1)
449 				break;
450 		}
451 	}
452 
453 	if (count <= 1)
454 		SIGDELSET(sq->sq_signals, signo);
455 	si->ksi_signo = signo;
456 	return (signo);
457 }
458 
459 void
460 sigqueue_take(ksiginfo_t *ksi)
461 {
462 	struct ksiginfo *kp;
463 	struct proc	*p;
464 	sigqueue_t	*sq;
465 
466 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
467 		return;
468 
469 	p = sq->sq_proc;
470 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
471 	ksi->ksi_sigq = NULL;
472 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
473 		p->p_pendingcnt--;
474 
475 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
476 	     kp = TAILQ_NEXT(kp, ksi_link)) {
477 		if (kp->ksi_signo == ksi->ksi_signo)
478 			break;
479 	}
480 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
481 	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
482 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
483 }
484 
485 static int
486 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
487 {
488 	struct proc *p = sq->sq_proc;
489 	struct ksiginfo *ksi;
490 	int ret = 0;
491 
492 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
493 
494 	/*
495 	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
496 	 * for these signals.
497 	 */
498 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
499 		SIGADDSET(sq->sq_kill, signo);
500 		goto out_set_bit;
501 	}
502 
503 	/* directly insert the ksi, don't copy it */
504 	if (si->ksi_flags & KSI_INS) {
505 		if (si->ksi_flags & KSI_HEAD)
506 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
507 		else
508 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
509 		si->ksi_sigq = sq;
510 		goto out_set_bit;
511 	}
512 
513 	if (__predict_false(ksiginfo_zone == NULL)) {
514 		SIGADDSET(sq->sq_kill, signo);
515 		goto out_set_bit;
516 	}
517 
518 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
519 		signal_overflow++;
520 		ret = EAGAIN;
521 	} else if ((ksi = ksiginfo_alloc(M_NOWAIT)) == NULL) {
522 		signal_alloc_fail++;
523 		ret = EAGAIN;
524 	} else {
525 		if (p != NULL)
526 			p->p_pendingcnt++;
527 		ksiginfo_copy(si, ksi);
528 		ksi->ksi_signo = signo;
529 		if (si->ksi_flags & KSI_HEAD)
530 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
531 		else
532 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
533 		ksi->ksi_sigq = sq;
534 	}
535 
536 	if (ret != 0) {
537 		if ((si->ksi_flags & KSI_PTRACE) != 0) {
538 			SIGADDSET(sq->sq_ptrace, signo);
539 			ret = 0;
540 			goto out_set_bit;
541 		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
542 		    (si->ksi_flags & KSI_SIGQ) == 0) {
543 			SIGADDSET(sq->sq_kill, signo);
544 			ret = 0;
545 			goto out_set_bit;
546 		}
547 		return (ret);
548 	}
549 
550 out_set_bit:
551 	SIGADDSET(sq->sq_signals, signo);
552 	return (ret);
553 }
554 
555 void
556 sigqueue_flush(sigqueue_t *sq)
557 {
558 	struct proc *p = sq->sq_proc;
559 	ksiginfo_t *ksi;
560 
561 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
562 
563 	if (p != NULL)
564 		PROC_LOCK_ASSERT(p, MA_OWNED);
565 
566 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
567 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
568 		ksi->ksi_sigq = NULL;
569 		if (ksiginfo_tryfree(ksi) && p != NULL)
570 			p->p_pendingcnt--;
571 	}
572 
573 	SIGEMPTYSET(sq->sq_signals);
574 	SIGEMPTYSET(sq->sq_kill);
575 	SIGEMPTYSET(sq->sq_ptrace);
576 }
577 
578 static void
579 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
580 {
581 	sigset_t tmp;
582 	struct proc *p1, *p2;
583 	ksiginfo_t *ksi, *next;
584 
585 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
586 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
587 	p1 = src->sq_proc;
588 	p2 = dst->sq_proc;
589 	/* Move siginfo to target list */
590 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
591 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
592 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
593 			if (p1 != NULL)
594 				p1->p_pendingcnt--;
595 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
596 			ksi->ksi_sigq = dst;
597 			if (p2 != NULL)
598 				p2->p_pendingcnt++;
599 		}
600 	}
601 
602 	/* Move pending bits to target list */
603 	tmp = src->sq_kill;
604 	SIGSETAND(tmp, *set);
605 	SIGSETOR(dst->sq_kill, tmp);
606 	SIGSETNAND(src->sq_kill, tmp);
607 
608 	tmp = src->sq_ptrace;
609 	SIGSETAND(tmp, *set);
610 	SIGSETOR(dst->sq_ptrace, tmp);
611 	SIGSETNAND(src->sq_ptrace, tmp);
612 
613 	tmp = src->sq_signals;
614 	SIGSETAND(tmp, *set);
615 	SIGSETOR(dst->sq_signals, tmp);
616 	SIGSETNAND(src->sq_signals, tmp);
617 }
618 
619 #if 0
620 static void
621 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
622 {
623 	sigset_t set;
624 
625 	SIGEMPTYSET(set);
626 	SIGADDSET(set, signo);
627 	sigqueue_move_set(src, dst, &set);
628 }
629 #endif
630 
631 static void
632 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
633 {
634 	struct proc *p = sq->sq_proc;
635 	ksiginfo_t *ksi, *next;
636 
637 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
638 
639 	/* Remove siginfo queue */
640 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
641 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
642 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
643 			ksi->ksi_sigq = NULL;
644 			if (ksiginfo_tryfree(ksi) && p != NULL)
645 				p->p_pendingcnt--;
646 		}
647 	}
648 	SIGSETNAND(sq->sq_kill, *set);
649 	SIGSETNAND(sq->sq_ptrace, *set);
650 	SIGSETNAND(sq->sq_signals, *set);
651 }
652 
653 void
654 sigqueue_delete(sigqueue_t *sq, int signo)
655 {
656 	sigset_t set;
657 
658 	SIGEMPTYSET(set);
659 	SIGADDSET(set, signo);
660 	sigqueue_delete_set(sq, &set);
661 }
662 
663 /* Remove a set of signals for a process */
664 static void
665 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
666 {
667 	sigqueue_t worklist;
668 	struct thread *td0;
669 
670 	PROC_LOCK_ASSERT(p, MA_OWNED);
671 
672 	sigqueue_init(&worklist, NULL);
673 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
674 
675 	FOREACH_THREAD_IN_PROC(p, td0)
676 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
677 
678 	sigqueue_flush(&worklist);
679 }
680 
681 void
682 sigqueue_delete_proc(struct proc *p, int signo)
683 {
684 	sigset_t set;
685 
686 	SIGEMPTYSET(set);
687 	SIGADDSET(set, signo);
688 	sigqueue_delete_set_proc(p, &set);
689 }
690 
691 static void
692 sigqueue_delete_stopmask_proc(struct proc *p)
693 {
694 	sigset_t set;
695 
696 	SIGEMPTYSET(set);
697 	SIGADDSET(set, SIGSTOP);
698 	SIGADDSET(set, SIGTSTP);
699 	SIGADDSET(set, SIGTTIN);
700 	SIGADDSET(set, SIGTTOU);
701 	sigqueue_delete_set_proc(p, &set);
702 }
703 
704 /*
705  * Determine signal that should be delivered to thread td, the current
706  * thread, 0 if none.  If there is a pending stop signal with default
707  * action, the process stops in issignal().
708  */
709 int
710 cursig(struct thread *td)
711 {
712 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
713 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
714 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
715 	return (SIGPENDING(td) ? issignal(td) : 0);
716 }
717 
718 /*
719  * Arrange for ast() to handle unmasked pending signals on return to user
720  * mode.  This must be called whenever a signal is added to td_sigqueue or
721  * unmasked in td_sigmask.
722  */
723 void
724 signotify(struct thread *td)
725 {
726 
727 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
728 
729 	if (SIGPENDING(td))
730 		ast_sched(td, TDA_SIG);
731 }
732 
733 /*
734  * Returns 1 (true) if altstack is configured for the thread, and the
735  * passed stack bottom address falls into the altstack range.  Handles
736  * the 43 compat special case where the alt stack size is zero.
737  */
738 int
739 sigonstack(size_t sp)
740 {
741 	struct thread *td;
742 
743 	td = curthread;
744 	if ((td->td_pflags & TDP_ALTSTACK) == 0)
745 		return (0);
746 #if defined(COMPAT_43)
747 	if (SV_PROC_FLAG(td->td_proc, SV_AOUT) && td->td_sigstk.ss_size == 0)
748 		return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
749 #endif
750 	return (sp >= (size_t)td->td_sigstk.ss_sp &&
751 	    sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
752 }
753 
754 static __inline int
755 sigprop(int sig)
756 {
757 
758 	if (sig > 0 && sig < nitems(sigproptbl))
759 		return (sigproptbl[sig]);
760 	return (0);
761 }
762 
763 static bool
764 sigact_flag_test(const struct sigaction *act, int flag)
765 {
766 
767 	/*
768 	 * SA_SIGINFO is reset when signal disposition is set to
769 	 * ignore or default.  Other flags are kept according to user
770 	 * settings.
771 	 */
772 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
773 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
774 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
775 }
776 
777 /*
778  * kern_sigaction
779  * sigaction
780  * freebsd4_sigaction
781  * osigaction
782  */
783 int
784 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
785     struct sigaction *oact, int flags)
786 {
787 	struct sigacts *ps;
788 	struct proc *p = td->td_proc;
789 
790 	if (!_SIG_VALID(sig))
791 		return (EINVAL);
792 	if (act != NULL && act->sa_handler != SIG_DFL &&
793 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
794 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
795 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
796 		return (EINVAL);
797 
798 	PROC_LOCK(p);
799 	ps = p->p_sigacts;
800 	mtx_lock(&ps->ps_mtx);
801 	if (oact) {
802 		memset(oact, 0, sizeof(*oact));
803 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
804 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
805 			oact->sa_flags |= SA_ONSTACK;
806 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
807 			oact->sa_flags |= SA_RESTART;
808 		if (SIGISMEMBER(ps->ps_sigreset, sig))
809 			oact->sa_flags |= SA_RESETHAND;
810 		if (SIGISMEMBER(ps->ps_signodefer, sig))
811 			oact->sa_flags |= SA_NODEFER;
812 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
813 			oact->sa_flags |= SA_SIGINFO;
814 			oact->sa_sigaction =
815 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
816 		} else
817 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
818 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
819 			oact->sa_flags |= SA_NOCLDSTOP;
820 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
821 			oact->sa_flags |= SA_NOCLDWAIT;
822 	}
823 	if (act) {
824 		if ((sig == SIGKILL || sig == SIGSTOP) &&
825 		    act->sa_handler != SIG_DFL) {
826 			mtx_unlock(&ps->ps_mtx);
827 			PROC_UNLOCK(p);
828 			return (EINVAL);
829 		}
830 
831 		/*
832 		 * Change setting atomically.
833 		 */
834 
835 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
836 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
837 		if (sigact_flag_test(act, SA_SIGINFO)) {
838 			ps->ps_sigact[_SIG_IDX(sig)] =
839 			    (__sighandler_t *)act->sa_sigaction;
840 			SIGADDSET(ps->ps_siginfo, sig);
841 		} else {
842 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
843 			SIGDELSET(ps->ps_siginfo, sig);
844 		}
845 		if (!sigact_flag_test(act, SA_RESTART))
846 			SIGADDSET(ps->ps_sigintr, sig);
847 		else
848 			SIGDELSET(ps->ps_sigintr, sig);
849 		if (sigact_flag_test(act, SA_ONSTACK))
850 			SIGADDSET(ps->ps_sigonstack, sig);
851 		else
852 			SIGDELSET(ps->ps_sigonstack, sig);
853 		if (sigact_flag_test(act, SA_RESETHAND))
854 			SIGADDSET(ps->ps_sigreset, sig);
855 		else
856 			SIGDELSET(ps->ps_sigreset, sig);
857 		if (sigact_flag_test(act, SA_NODEFER))
858 			SIGADDSET(ps->ps_signodefer, sig);
859 		else
860 			SIGDELSET(ps->ps_signodefer, sig);
861 		if (sig == SIGCHLD) {
862 			if (act->sa_flags & SA_NOCLDSTOP)
863 				ps->ps_flag |= PS_NOCLDSTOP;
864 			else
865 				ps->ps_flag &= ~PS_NOCLDSTOP;
866 			if (act->sa_flags & SA_NOCLDWAIT) {
867 				/*
868 				 * Paranoia: since SA_NOCLDWAIT is implemented
869 				 * by reparenting the dying child to PID 1 (and
870 				 * trust it to reap the zombie), PID 1 itself
871 				 * is forbidden to set SA_NOCLDWAIT.
872 				 */
873 				if (p->p_pid == 1)
874 					ps->ps_flag &= ~PS_NOCLDWAIT;
875 				else
876 					ps->ps_flag |= PS_NOCLDWAIT;
877 			} else
878 				ps->ps_flag &= ~PS_NOCLDWAIT;
879 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
880 				ps->ps_flag |= PS_CLDSIGIGN;
881 			else
882 				ps->ps_flag &= ~PS_CLDSIGIGN;
883 		}
884 		/*
885 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
886 		 * and for signals set to SIG_DFL where the default is to
887 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
888 		 * have to restart the process.
889 		 */
890 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
891 		    (sigprop(sig) & SIGPROP_IGNORE &&
892 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
893 			/* never to be seen again */
894 			sigqueue_delete_proc(p, sig);
895 			if (sig != SIGCONT)
896 				/* easier in psignal */
897 				SIGADDSET(ps->ps_sigignore, sig);
898 			SIGDELSET(ps->ps_sigcatch, sig);
899 		} else {
900 			SIGDELSET(ps->ps_sigignore, sig);
901 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
902 				SIGDELSET(ps->ps_sigcatch, sig);
903 			else
904 				SIGADDSET(ps->ps_sigcatch, sig);
905 		}
906 #ifdef COMPAT_FREEBSD4
907 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
908 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
909 		    (flags & KSA_FREEBSD4) == 0)
910 			SIGDELSET(ps->ps_freebsd4, sig);
911 		else
912 			SIGADDSET(ps->ps_freebsd4, sig);
913 #endif
914 #ifdef COMPAT_43
915 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
916 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
917 		    (flags & KSA_OSIGSET) == 0)
918 			SIGDELSET(ps->ps_osigset, sig);
919 		else
920 			SIGADDSET(ps->ps_osigset, sig);
921 #endif
922 	}
923 	mtx_unlock(&ps->ps_mtx);
924 	PROC_UNLOCK(p);
925 	return (0);
926 }
927 
928 #ifndef _SYS_SYSPROTO_H_
929 struct sigaction_args {
930 	int	sig;
931 	struct	sigaction *act;
932 	struct	sigaction *oact;
933 };
934 #endif
935 int
936 sys_sigaction(struct thread *td, struct sigaction_args *uap)
937 {
938 	struct sigaction act, oact;
939 	struct sigaction *actp, *oactp;
940 	int error;
941 
942 	actp = (uap->act != NULL) ? &act : NULL;
943 	oactp = (uap->oact != NULL) ? &oact : NULL;
944 	if (actp) {
945 		error = copyin(uap->act, actp, sizeof(act));
946 		if (error)
947 			return (error);
948 	}
949 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
950 	if (oactp && !error)
951 		error = copyout(oactp, uap->oact, sizeof(oact));
952 	return (error);
953 }
954 
955 #ifdef COMPAT_FREEBSD4
956 #ifndef _SYS_SYSPROTO_H_
957 struct freebsd4_sigaction_args {
958 	int	sig;
959 	struct	sigaction *act;
960 	struct	sigaction *oact;
961 };
962 #endif
963 int
964 freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
965 {
966 	struct sigaction act, oact;
967 	struct sigaction *actp, *oactp;
968 	int error;
969 
970 	actp = (uap->act != NULL) ? &act : NULL;
971 	oactp = (uap->oact != NULL) ? &oact : NULL;
972 	if (actp) {
973 		error = copyin(uap->act, actp, sizeof(act));
974 		if (error)
975 			return (error);
976 	}
977 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
978 	if (oactp && !error)
979 		error = copyout(oactp, uap->oact, sizeof(oact));
980 	return (error);
981 }
982 #endif	/* COMAPT_FREEBSD4 */
983 
984 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
985 #ifndef _SYS_SYSPROTO_H_
986 struct osigaction_args {
987 	int	signum;
988 	struct	osigaction *nsa;
989 	struct	osigaction *osa;
990 };
991 #endif
992 int
993 osigaction(struct thread *td, struct osigaction_args *uap)
994 {
995 	struct osigaction sa;
996 	struct sigaction nsa, osa;
997 	struct sigaction *nsap, *osap;
998 	int error;
999 
1000 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1001 		return (EINVAL);
1002 
1003 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
1004 	osap = (uap->osa != NULL) ? &osa : NULL;
1005 
1006 	if (nsap) {
1007 		error = copyin(uap->nsa, &sa, sizeof(sa));
1008 		if (error)
1009 			return (error);
1010 		nsap->sa_handler = sa.sa_handler;
1011 		nsap->sa_flags = sa.sa_flags;
1012 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
1013 	}
1014 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1015 	if (osap && !error) {
1016 		sa.sa_handler = osap->sa_handler;
1017 		sa.sa_flags = osap->sa_flags;
1018 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
1019 		error = copyout(&sa, uap->osa, sizeof(sa));
1020 	}
1021 	return (error);
1022 }
1023 
1024 #if !defined(__i386__)
1025 /* Avoid replicating the same stub everywhere */
1026 int
1027 osigreturn(struct thread *td, struct osigreturn_args *uap)
1028 {
1029 
1030 	return (nosys(td, (struct nosys_args *)uap));
1031 }
1032 #endif
1033 #endif /* COMPAT_43 */
1034 
1035 /*
1036  * Initialize signal state for process 0;
1037  * set to ignore signals that are ignored by default.
1038  */
1039 void
1040 siginit(struct proc *p)
1041 {
1042 	int i;
1043 	struct sigacts *ps;
1044 
1045 	PROC_LOCK(p);
1046 	ps = p->p_sigacts;
1047 	mtx_lock(&ps->ps_mtx);
1048 	for (i = 1; i <= NSIG; i++) {
1049 		if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) {
1050 			SIGADDSET(ps->ps_sigignore, i);
1051 		}
1052 	}
1053 	mtx_unlock(&ps->ps_mtx);
1054 	PROC_UNLOCK(p);
1055 }
1056 
1057 /*
1058  * Reset specified signal to the default disposition.
1059  */
1060 static void
1061 sigdflt(struct sigacts *ps, int sig)
1062 {
1063 
1064 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1065 	SIGDELSET(ps->ps_sigcatch, sig);
1066 	if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT)
1067 		SIGADDSET(ps->ps_sigignore, sig);
1068 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1069 	SIGDELSET(ps->ps_siginfo, sig);
1070 }
1071 
1072 /*
1073  * Reset signals for an exec of the specified process.
1074  */
1075 void
1076 execsigs(struct proc *p)
1077 {
1078 	struct sigacts *ps;
1079 	struct thread *td;
1080 
1081 	/*
1082 	 * Reset caught signals.  Held signals remain held
1083 	 * through td_sigmask (unless they were caught,
1084 	 * and are now ignored by default).
1085 	 */
1086 	PROC_LOCK_ASSERT(p, MA_OWNED);
1087 	ps = p->p_sigacts;
1088 	mtx_lock(&ps->ps_mtx);
1089 	sig_drop_caught(p);
1090 
1091 	/*
1092 	 * Reset stack state to the user stack.
1093 	 * Clear set of signals caught on the signal stack.
1094 	 */
1095 	td = curthread;
1096 	MPASS(td->td_proc == p);
1097 	td->td_sigstk.ss_flags = SS_DISABLE;
1098 	td->td_sigstk.ss_size = 0;
1099 	td->td_sigstk.ss_sp = 0;
1100 	td->td_pflags &= ~TDP_ALTSTACK;
1101 	/*
1102 	 * Reset no zombies if child dies flag as Solaris does.
1103 	 */
1104 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1105 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1106 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1107 	mtx_unlock(&ps->ps_mtx);
1108 }
1109 
1110 /*
1111  * kern_sigprocmask()
1112  *
1113  *	Manipulate signal mask.
1114  */
1115 int
1116 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1117     int flags)
1118 {
1119 	sigset_t new_block, oset1;
1120 	struct proc *p;
1121 	int error;
1122 
1123 	p = td->td_proc;
1124 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1125 		PROC_LOCK_ASSERT(p, MA_OWNED);
1126 	else
1127 		PROC_LOCK(p);
1128 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1129 	    ? MA_OWNED : MA_NOTOWNED);
1130 	if (oset != NULL)
1131 		*oset = td->td_sigmask;
1132 
1133 	error = 0;
1134 	if (set != NULL) {
1135 		switch (how) {
1136 		case SIG_BLOCK:
1137 			SIG_CANTMASK(*set);
1138 			oset1 = td->td_sigmask;
1139 			SIGSETOR(td->td_sigmask, *set);
1140 			new_block = td->td_sigmask;
1141 			SIGSETNAND(new_block, oset1);
1142 			break;
1143 		case SIG_UNBLOCK:
1144 			SIGSETNAND(td->td_sigmask, *set);
1145 			signotify(td);
1146 			goto out;
1147 		case SIG_SETMASK:
1148 			SIG_CANTMASK(*set);
1149 			oset1 = td->td_sigmask;
1150 			if (flags & SIGPROCMASK_OLD)
1151 				SIGSETLO(td->td_sigmask, *set);
1152 			else
1153 				td->td_sigmask = *set;
1154 			new_block = td->td_sigmask;
1155 			SIGSETNAND(new_block, oset1);
1156 			signotify(td);
1157 			break;
1158 		default:
1159 			error = EINVAL;
1160 			goto out;
1161 		}
1162 
1163 		/*
1164 		 * The new_block set contains signals that were not previously
1165 		 * blocked, but are blocked now.
1166 		 *
1167 		 * In case we block any signal that was not previously blocked
1168 		 * for td, and process has the signal pending, try to schedule
1169 		 * signal delivery to some thread that does not block the
1170 		 * signal, possibly waking it up.
1171 		 */
1172 		if (p->p_numthreads != 1)
1173 			reschedule_signals(p, new_block, flags);
1174 	}
1175 
1176 out:
1177 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1178 		PROC_UNLOCK(p);
1179 	return (error);
1180 }
1181 
1182 #ifndef _SYS_SYSPROTO_H_
1183 struct sigprocmask_args {
1184 	int	how;
1185 	const sigset_t *set;
1186 	sigset_t *oset;
1187 };
1188 #endif
1189 int
1190 sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1191 {
1192 	sigset_t set, oset;
1193 	sigset_t *setp, *osetp;
1194 	int error;
1195 
1196 	setp = (uap->set != NULL) ? &set : NULL;
1197 	osetp = (uap->oset != NULL) ? &oset : NULL;
1198 	if (setp) {
1199 		error = copyin(uap->set, setp, sizeof(set));
1200 		if (error)
1201 			return (error);
1202 	}
1203 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1204 	if (osetp && !error) {
1205 		error = copyout(osetp, uap->oset, sizeof(oset));
1206 	}
1207 	return (error);
1208 }
1209 
1210 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1211 #ifndef _SYS_SYSPROTO_H_
1212 struct osigprocmask_args {
1213 	int	how;
1214 	osigset_t mask;
1215 };
1216 #endif
1217 int
1218 osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1219 {
1220 	sigset_t set, oset;
1221 	int error;
1222 
1223 	OSIG2SIG(uap->mask, set);
1224 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1225 	SIG2OSIG(oset, td->td_retval[0]);
1226 	return (error);
1227 }
1228 #endif /* COMPAT_43 */
1229 
1230 int
1231 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1232 {
1233 	ksiginfo_t ksi;
1234 	sigset_t set;
1235 	int error;
1236 
1237 	error = copyin(uap->set, &set, sizeof(set));
1238 	if (error) {
1239 		td->td_retval[0] = error;
1240 		return (0);
1241 	}
1242 
1243 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1244 	if (error) {
1245 		/*
1246 		 * sigwait() function shall not return EINTR, but
1247 		 * the syscall does.  Non-ancient libc provides the
1248 		 * wrapper which hides EINTR.  Otherwise, EINTR return
1249 		 * is used by libthr to handle required cancellation
1250 		 * point in the sigwait().
1251 		 */
1252 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1253 			return (ERESTART);
1254 		td->td_retval[0] = error;
1255 		return (0);
1256 	}
1257 
1258 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1259 	td->td_retval[0] = error;
1260 	return (0);
1261 }
1262 
1263 int
1264 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1265 {
1266 	struct timespec ts;
1267 	struct timespec *timeout;
1268 	sigset_t set;
1269 	ksiginfo_t ksi;
1270 	int error;
1271 
1272 	if (uap->timeout) {
1273 		error = copyin(uap->timeout, &ts, sizeof(ts));
1274 		if (error)
1275 			return (error);
1276 
1277 		timeout = &ts;
1278 	} else
1279 		timeout = NULL;
1280 
1281 	error = copyin(uap->set, &set, sizeof(set));
1282 	if (error)
1283 		return (error);
1284 
1285 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1286 	if (error)
1287 		return (error);
1288 
1289 	if (uap->info)
1290 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1291 
1292 	if (error == 0)
1293 		td->td_retval[0] = ksi.ksi_signo;
1294 	return (error);
1295 }
1296 
1297 int
1298 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1299 {
1300 	ksiginfo_t ksi;
1301 	sigset_t set;
1302 	int error;
1303 
1304 	error = copyin(uap->set, &set, sizeof(set));
1305 	if (error)
1306 		return (error);
1307 
1308 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1309 	if (error)
1310 		return (error);
1311 
1312 	if (uap->info)
1313 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1314 
1315 	if (error == 0)
1316 		td->td_retval[0] = ksi.ksi_signo;
1317 	return (error);
1318 }
1319 
1320 static void
1321 proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1322 {
1323 	struct thread *thr;
1324 
1325 	FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1326 		if (thr == td)
1327 			thr->td_si = *si;
1328 		else
1329 			thr->td_si.si_signo = 0;
1330 	}
1331 }
1332 
1333 int
1334 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1335 	struct timespec *timeout)
1336 {
1337 	struct sigacts *ps;
1338 	sigset_t saved_mask, new_block;
1339 	struct proc *p;
1340 	int error, sig, timevalid = 0;
1341 	sbintime_t sbt, precision, tsbt;
1342 	struct timespec ts;
1343 	bool traced;
1344 
1345 	p = td->td_proc;
1346 	error = 0;
1347 	traced = false;
1348 
1349 	/* Ensure the sigfastblock value is up to date. */
1350 	sigfastblock_fetch(td);
1351 
1352 	if (timeout != NULL) {
1353 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1354 			timevalid = 1;
1355 			ts = *timeout;
1356 			if (ts.tv_sec < INT32_MAX / 2) {
1357 				tsbt = tstosbt(ts);
1358 				precision = tsbt;
1359 				precision >>= tc_precexp;
1360 				if (TIMESEL(&sbt, tsbt))
1361 					sbt += tc_tick_sbt;
1362 				sbt += tsbt;
1363 			} else
1364 				precision = sbt = 0;
1365 		}
1366 	} else
1367 		precision = sbt = 0;
1368 	ksiginfo_init(ksi);
1369 	/* Some signals can not be waited for. */
1370 	SIG_CANTMASK(waitset);
1371 	ps = p->p_sigacts;
1372 	PROC_LOCK(p);
1373 	saved_mask = td->td_sigmask;
1374 	SIGSETNAND(td->td_sigmask, waitset);
1375 	if ((p->p_sysent->sv_flags & SV_SIG_DISCIGN) != 0 ||
1376 	    !kern_sig_discard_ign) {
1377 		thread_lock(td);
1378 		td->td_flags |= TDF_SIGWAIT;
1379 		thread_unlock(td);
1380 	}
1381 	for (;;) {
1382 		mtx_lock(&ps->ps_mtx);
1383 		sig = cursig(td);
1384 		mtx_unlock(&ps->ps_mtx);
1385 		KASSERT(sig >= 0, ("sig %d", sig));
1386 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1387 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1388 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1389 				error = 0;
1390 				break;
1391 			}
1392 		}
1393 
1394 		if (error != 0)
1395 			break;
1396 
1397 		/*
1398 		 * POSIX says this must be checked after looking for pending
1399 		 * signals.
1400 		 */
1401 		if (timeout != NULL && !timevalid) {
1402 			error = EINVAL;
1403 			break;
1404 		}
1405 
1406 		if (traced) {
1407 			error = EINTR;
1408 			break;
1409 		}
1410 
1411 		error = msleep_sbt(&p->p_sigacts, &p->p_mtx, PPAUSE | PCATCH,
1412 		    "sigwait", sbt, precision, C_ABSOLUTE);
1413 
1414 		/* The syscalls can not be restarted. */
1415 		if (error == ERESTART)
1416 			error = EINTR;
1417 
1418 		/*
1419 		 * If PTRACE_SCE or PTRACE_SCX were set after
1420 		 * userspace entered the syscall, return spurious
1421 		 * EINTR after wait was done.  Only do this as last
1422 		 * resort after rechecking for possible queued signals
1423 		 * and expired timeouts.
1424 		 */
1425 		if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1426 			traced = true;
1427 	}
1428 	thread_lock(td);
1429 	td->td_flags &= ~TDF_SIGWAIT;
1430 	thread_unlock(td);
1431 
1432 	new_block = saved_mask;
1433 	SIGSETNAND(new_block, td->td_sigmask);
1434 	td->td_sigmask = saved_mask;
1435 	/*
1436 	 * Fewer signals can be delivered to us, reschedule signal
1437 	 * notification.
1438 	 */
1439 	if (p->p_numthreads != 1)
1440 		reschedule_signals(p, new_block, 0);
1441 
1442 	if (error == 0) {
1443 		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1444 
1445 		if (ksi->ksi_code == SI_TIMER)
1446 			itimer_accept(p, ksi->ksi_timerid, ksi);
1447 
1448 #ifdef KTRACE
1449 		if (KTRPOINT(td, KTR_PSIG)) {
1450 			sig_t action;
1451 
1452 			mtx_lock(&ps->ps_mtx);
1453 			action = ps->ps_sigact[_SIG_IDX(sig)];
1454 			mtx_unlock(&ps->ps_mtx);
1455 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1456 		}
1457 #endif
1458 		if (sig == SIGKILL) {
1459 			proc_td_siginfo_capture(td, &ksi->ksi_info);
1460 			sigexit(td, sig);
1461 		}
1462 	}
1463 	PROC_UNLOCK(p);
1464 	return (error);
1465 }
1466 
1467 #ifndef _SYS_SYSPROTO_H_
1468 struct sigpending_args {
1469 	sigset_t	*set;
1470 };
1471 #endif
1472 int
1473 sys_sigpending(struct thread *td, struct sigpending_args *uap)
1474 {
1475 	struct proc *p = td->td_proc;
1476 	sigset_t pending;
1477 
1478 	PROC_LOCK(p);
1479 	pending = p->p_sigqueue.sq_signals;
1480 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1481 	PROC_UNLOCK(p);
1482 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1483 }
1484 
1485 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1486 #ifndef _SYS_SYSPROTO_H_
1487 struct osigpending_args {
1488 	int	dummy;
1489 };
1490 #endif
1491 int
1492 osigpending(struct thread *td, struct osigpending_args *uap)
1493 {
1494 	struct proc *p = td->td_proc;
1495 	sigset_t pending;
1496 
1497 	PROC_LOCK(p);
1498 	pending = p->p_sigqueue.sq_signals;
1499 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1500 	PROC_UNLOCK(p);
1501 	SIG2OSIG(pending, td->td_retval[0]);
1502 	return (0);
1503 }
1504 #endif /* COMPAT_43 */
1505 
1506 #if defined(COMPAT_43)
1507 /*
1508  * Generalized interface signal handler, 4.3-compatible.
1509  */
1510 #ifndef _SYS_SYSPROTO_H_
1511 struct osigvec_args {
1512 	int	signum;
1513 	struct	sigvec *nsv;
1514 	struct	sigvec *osv;
1515 };
1516 #endif
1517 /* ARGSUSED */
1518 int
1519 osigvec(struct thread *td, struct osigvec_args *uap)
1520 {
1521 	struct sigvec vec;
1522 	struct sigaction nsa, osa;
1523 	struct sigaction *nsap, *osap;
1524 	int error;
1525 
1526 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1527 		return (EINVAL);
1528 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1529 	osap = (uap->osv != NULL) ? &osa : NULL;
1530 	if (nsap) {
1531 		error = copyin(uap->nsv, &vec, sizeof(vec));
1532 		if (error)
1533 			return (error);
1534 		nsap->sa_handler = vec.sv_handler;
1535 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1536 		nsap->sa_flags = vec.sv_flags;
1537 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1538 	}
1539 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1540 	if (osap && !error) {
1541 		vec.sv_handler = osap->sa_handler;
1542 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1543 		vec.sv_flags = osap->sa_flags;
1544 		vec.sv_flags &= ~SA_NOCLDWAIT;
1545 		vec.sv_flags ^= SA_RESTART;
1546 		error = copyout(&vec, uap->osv, sizeof(vec));
1547 	}
1548 	return (error);
1549 }
1550 
1551 #ifndef _SYS_SYSPROTO_H_
1552 struct osigblock_args {
1553 	int	mask;
1554 };
1555 #endif
1556 int
1557 osigblock(struct thread *td, struct osigblock_args *uap)
1558 {
1559 	sigset_t set, oset;
1560 
1561 	OSIG2SIG(uap->mask, set);
1562 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1563 	SIG2OSIG(oset, td->td_retval[0]);
1564 	return (0);
1565 }
1566 
1567 #ifndef _SYS_SYSPROTO_H_
1568 struct osigsetmask_args {
1569 	int	mask;
1570 };
1571 #endif
1572 int
1573 osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1574 {
1575 	sigset_t set, oset;
1576 
1577 	OSIG2SIG(uap->mask, set);
1578 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1579 	SIG2OSIG(oset, td->td_retval[0]);
1580 	return (0);
1581 }
1582 #endif /* COMPAT_43 */
1583 
1584 /*
1585  * Suspend calling thread until signal, providing mask to be set in the
1586  * meantime.
1587  */
1588 #ifndef _SYS_SYSPROTO_H_
1589 struct sigsuspend_args {
1590 	const sigset_t *sigmask;
1591 };
1592 #endif
1593 /* ARGSUSED */
1594 int
1595 sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1596 {
1597 	sigset_t mask;
1598 	int error;
1599 
1600 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1601 	if (error)
1602 		return (error);
1603 	return (kern_sigsuspend(td, mask));
1604 }
1605 
1606 int
1607 kern_sigsuspend(struct thread *td, sigset_t mask)
1608 {
1609 	struct proc *p = td->td_proc;
1610 	int has_sig, sig;
1611 
1612 	/* Ensure the sigfastblock value is up to date. */
1613 	sigfastblock_fetch(td);
1614 
1615 	/*
1616 	 * When returning from sigsuspend, we want
1617 	 * the old mask to be restored after the
1618 	 * signal handler has finished.  Thus, we
1619 	 * save it here and mark the sigacts structure
1620 	 * to indicate this.
1621 	 */
1622 	PROC_LOCK(p);
1623 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1624 	    SIGPROCMASK_PROC_LOCKED);
1625 	td->td_pflags |= TDP_OLDMASK;
1626 	ast_sched(td, TDA_SIGSUSPEND);
1627 
1628 	/*
1629 	 * Process signals now. Otherwise, we can get spurious wakeup
1630 	 * due to signal entered process queue, but delivered to other
1631 	 * thread. But sigsuspend should return only on signal
1632 	 * delivery.
1633 	 */
1634 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1635 	for (has_sig = 0; !has_sig;) {
1636 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1637 			0) == 0)
1638 			/* void */;
1639 		thread_suspend_check(0);
1640 		mtx_lock(&p->p_sigacts->ps_mtx);
1641 		while ((sig = cursig(td)) != 0) {
1642 			KASSERT(sig >= 0, ("sig %d", sig));
1643 			has_sig += postsig(sig);
1644 		}
1645 		mtx_unlock(&p->p_sigacts->ps_mtx);
1646 
1647 		/*
1648 		 * If PTRACE_SCE or PTRACE_SCX were set after
1649 		 * userspace entered the syscall, return spurious
1650 		 * EINTR.
1651 		 */
1652 		if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1653 			has_sig += 1;
1654 	}
1655 	PROC_UNLOCK(p);
1656 	td->td_errno = EINTR;
1657 	td->td_pflags |= TDP_NERRNO;
1658 	return (EJUSTRETURN);
1659 }
1660 
1661 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1662 /*
1663  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1664  * convention: libc stub passes mask, not pointer, to save a copyin.
1665  */
1666 #ifndef _SYS_SYSPROTO_H_
1667 struct osigsuspend_args {
1668 	osigset_t mask;
1669 };
1670 #endif
1671 /* ARGSUSED */
1672 int
1673 osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1674 {
1675 	sigset_t mask;
1676 
1677 	OSIG2SIG(uap->mask, mask);
1678 	return (kern_sigsuspend(td, mask));
1679 }
1680 #endif /* COMPAT_43 */
1681 
1682 #if defined(COMPAT_43)
1683 #ifndef _SYS_SYSPROTO_H_
1684 struct osigstack_args {
1685 	struct	sigstack *nss;
1686 	struct	sigstack *oss;
1687 };
1688 #endif
1689 /* ARGSUSED */
1690 int
1691 osigstack(struct thread *td, struct osigstack_args *uap)
1692 {
1693 	struct sigstack nss, oss;
1694 	int error = 0;
1695 
1696 	if (uap->nss != NULL) {
1697 		error = copyin(uap->nss, &nss, sizeof(nss));
1698 		if (error)
1699 			return (error);
1700 	}
1701 	oss.ss_sp = td->td_sigstk.ss_sp;
1702 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1703 	if (uap->nss != NULL) {
1704 		td->td_sigstk.ss_sp = nss.ss_sp;
1705 		td->td_sigstk.ss_size = 0;
1706 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1707 		td->td_pflags |= TDP_ALTSTACK;
1708 	}
1709 	if (uap->oss != NULL)
1710 		error = copyout(&oss, uap->oss, sizeof(oss));
1711 
1712 	return (error);
1713 }
1714 #endif /* COMPAT_43 */
1715 
1716 #ifndef _SYS_SYSPROTO_H_
1717 struct sigaltstack_args {
1718 	stack_t	*ss;
1719 	stack_t	*oss;
1720 };
1721 #endif
1722 /* ARGSUSED */
1723 int
1724 sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1725 {
1726 	stack_t ss, oss;
1727 	int error;
1728 
1729 	if (uap->ss != NULL) {
1730 		error = copyin(uap->ss, &ss, sizeof(ss));
1731 		if (error)
1732 			return (error);
1733 	}
1734 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1735 	    (uap->oss != NULL) ? &oss : NULL);
1736 	if (error)
1737 		return (error);
1738 	if (uap->oss != NULL)
1739 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1740 	return (error);
1741 }
1742 
1743 int
1744 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1745 {
1746 	struct proc *p = td->td_proc;
1747 	int oonstack;
1748 
1749 	oonstack = sigonstack(cpu_getstack(td));
1750 
1751 	if (oss != NULL) {
1752 		*oss = td->td_sigstk;
1753 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1754 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1755 	}
1756 
1757 	if (ss != NULL) {
1758 		if (oonstack)
1759 			return (EPERM);
1760 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1761 			return (EINVAL);
1762 		if (!(ss->ss_flags & SS_DISABLE)) {
1763 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1764 				return (ENOMEM);
1765 
1766 			td->td_sigstk = *ss;
1767 			td->td_pflags |= TDP_ALTSTACK;
1768 		} else {
1769 			td->td_pflags &= ~TDP_ALTSTACK;
1770 		}
1771 	}
1772 	return (0);
1773 }
1774 
1775 struct killpg1_ctx {
1776 	struct thread *td;
1777 	ksiginfo_t *ksi;
1778 	int sig;
1779 	bool sent;
1780 	bool found;
1781 	int ret;
1782 };
1783 
1784 static void
1785 killpg1_sendsig_locked(struct proc *p, struct killpg1_ctx *arg)
1786 {
1787 	int err;
1788 
1789 	err = p_cansignal(arg->td, p, arg->sig);
1790 	if (err == 0 && arg->sig != 0)
1791 		pksignal(p, arg->sig, arg->ksi);
1792 	if (err != ESRCH)
1793 		arg->found = true;
1794 	if (err == 0)
1795 		arg->sent = true;
1796 	else if (arg->ret == 0 && err != ESRCH && err != EPERM)
1797 		arg->ret = err;
1798 }
1799 
1800 static void
1801 killpg1_sendsig(struct proc *p, bool notself, struct killpg1_ctx *arg)
1802 {
1803 
1804 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1805 	    (notself && p == arg->td->td_proc) || p->p_state == PRS_NEW)
1806 		return;
1807 
1808 	PROC_LOCK(p);
1809 	killpg1_sendsig_locked(p, arg);
1810 	PROC_UNLOCK(p);
1811 }
1812 
1813 static void
1814 kill_processes_prison_cb(struct proc *p, void *arg)
1815 {
1816 	struct killpg1_ctx *ctx = arg;
1817 
1818 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1819 	    (p == ctx->td->td_proc) || p->p_state == PRS_NEW)
1820 		return;
1821 
1822 	killpg1_sendsig_locked(p, ctx);
1823 }
1824 
1825 /*
1826  * Common code for kill process group/broadcast kill.
1827  * td is the calling thread, as usual.
1828  */
1829 static int
1830 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1831 {
1832 	struct proc *p;
1833 	struct pgrp *pgrp;
1834 	struct killpg1_ctx arg;
1835 
1836 	arg.td = td;
1837 	arg.ksi = ksi;
1838 	arg.sig = sig;
1839 	arg.sent = false;
1840 	arg.found = false;
1841 	arg.ret = 0;
1842 	if (all) {
1843 		/*
1844 		 * broadcast
1845 		 */
1846 		prison_proc_iterate(td->td_ucred->cr_prison,
1847 		    kill_processes_prison_cb, &arg);
1848 	} else {
1849 again:
1850 		sx_slock(&proctree_lock);
1851 		if (pgid == 0) {
1852 			/*
1853 			 * zero pgid means send to my process group.
1854 			 */
1855 			pgrp = td->td_proc->p_pgrp;
1856 			PGRP_LOCK(pgrp);
1857 		} else {
1858 			pgrp = pgfind(pgid);
1859 			if (pgrp == NULL) {
1860 				sx_sunlock(&proctree_lock);
1861 				return (ESRCH);
1862 			}
1863 		}
1864 		sx_sunlock(&proctree_lock);
1865 		if (!sx_try_xlock(&pgrp->pg_killsx)) {
1866 			PGRP_UNLOCK(pgrp);
1867 			sx_xlock(&pgrp->pg_killsx);
1868 			sx_xunlock(&pgrp->pg_killsx);
1869 			goto again;
1870 		}
1871 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1872 			killpg1_sendsig(p, false, &arg);
1873 		}
1874 		PGRP_UNLOCK(pgrp);
1875 		sx_xunlock(&pgrp->pg_killsx);
1876 	}
1877 	MPASS(arg.ret != 0 || arg.found || !arg.sent);
1878 	if (arg.ret == 0 && !arg.sent)
1879 		arg.ret = arg.found ? EPERM : ESRCH;
1880 	return (arg.ret);
1881 }
1882 
1883 #ifndef _SYS_SYSPROTO_H_
1884 struct kill_args {
1885 	int	pid;
1886 	int	signum;
1887 };
1888 #endif
1889 /* ARGSUSED */
1890 int
1891 sys_kill(struct thread *td, struct kill_args *uap)
1892 {
1893 
1894 	return (kern_kill(td, uap->pid, uap->signum));
1895 }
1896 
1897 int
1898 kern_kill(struct thread *td, pid_t pid, int signum)
1899 {
1900 	ksiginfo_t ksi;
1901 	struct proc *p;
1902 	int error;
1903 
1904 	/*
1905 	 * A process in capability mode can send signals only to himself.
1906 	 * The main rationale behind this is that abort(3) is implemented as
1907 	 * kill(getpid(), SIGABRT).
1908 	 */
1909 	if (IN_CAPABILITY_MODE(td) && pid != td->td_proc->p_pid)
1910 		return (ECAPMODE);
1911 
1912 	AUDIT_ARG_SIGNUM(signum);
1913 	AUDIT_ARG_PID(pid);
1914 	if ((u_int)signum > _SIG_MAXSIG)
1915 		return (EINVAL);
1916 
1917 	ksiginfo_init(&ksi);
1918 	ksi.ksi_signo = signum;
1919 	ksi.ksi_code = SI_USER;
1920 	ksi.ksi_pid = td->td_proc->p_pid;
1921 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1922 
1923 	if (pid > 0) {
1924 		/* kill single process */
1925 		if ((p = pfind_any(pid)) == NULL)
1926 			return (ESRCH);
1927 		AUDIT_ARG_PROCESS(p);
1928 		error = p_cansignal(td, p, signum);
1929 		if (error == 0 && signum)
1930 			pksignal(p, signum, &ksi);
1931 		PROC_UNLOCK(p);
1932 		return (error);
1933 	}
1934 	switch (pid) {
1935 	case -1:		/* broadcast signal */
1936 		return (killpg1(td, signum, 0, 1, &ksi));
1937 	case 0:			/* signal own process group */
1938 		return (killpg1(td, signum, 0, 0, &ksi));
1939 	default:		/* negative explicit process group */
1940 		return (killpg1(td, signum, -pid, 0, &ksi));
1941 	}
1942 	/* NOTREACHED */
1943 }
1944 
1945 int
1946 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1947 {
1948 	struct proc *p;
1949 	int error;
1950 
1951 	AUDIT_ARG_SIGNUM(uap->signum);
1952 	AUDIT_ARG_FD(uap->fd);
1953 	if ((u_int)uap->signum > _SIG_MAXSIG)
1954 		return (EINVAL);
1955 
1956 	error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1957 	if (error)
1958 		return (error);
1959 	AUDIT_ARG_PROCESS(p);
1960 	error = p_cansignal(td, p, uap->signum);
1961 	if (error == 0 && uap->signum)
1962 		kern_psignal(p, uap->signum);
1963 	PROC_UNLOCK(p);
1964 	return (error);
1965 }
1966 
1967 #if defined(COMPAT_43)
1968 #ifndef _SYS_SYSPROTO_H_
1969 struct okillpg_args {
1970 	int	pgid;
1971 	int	signum;
1972 };
1973 #endif
1974 /* ARGSUSED */
1975 int
1976 okillpg(struct thread *td, struct okillpg_args *uap)
1977 {
1978 	ksiginfo_t ksi;
1979 
1980 	AUDIT_ARG_SIGNUM(uap->signum);
1981 	AUDIT_ARG_PID(uap->pgid);
1982 	if ((u_int)uap->signum > _SIG_MAXSIG)
1983 		return (EINVAL);
1984 
1985 	ksiginfo_init(&ksi);
1986 	ksi.ksi_signo = uap->signum;
1987 	ksi.ksi_code = SI_USER;
1988 	ksi.ksi_pid = td->td_proc->p_pid;
1989 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1990 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1991 }
1992 #endif /* COMPAT_43 */
1993 
1994 #ifndef _SYS_SYSPROTO_H_
1995 struct sigqueue_args {
1996 	pid_t pid;
1997 	int signum;
1998 	/* union sigval */ void *value;
1999 };
2000 #endif
2001 int
2002 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
2003 {
2004 	union sigval sv;
2005 
2006 	sv.sival_ptr = uap->value;
2007 
2008 	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
2009 }
2010 
2011 int
2012 kern_sigqueue(struct thread *td, pid_t pid, int signum, union sigval *value)
2013 {
2014 	ksiginfo_t ksi;
2015 	struct proc *p;
2016 	int error;
2017 
2018 	if ((u_int)signum > _SIG_MAXSIG)
2019 		return (EINVAL);
2020 
2021 	/*
2022 	 * Specification says sigqueue can only send signal to
2023 	 * single process.
2024 	 */
2025 	if (pid <= 0)
2026 		return (EINVAL);
2027 
2028 	if ((p = pfind_any(pid)) == NULL)
2029 		return (ESRCH);
2030 	error = p_cansignal(td, p, signum);
2031 	if (error == 0 && signum != 0) {
2032 		ksiginfo_init(&ksi);
2033 		ksi.ksi_flags = KSI_SIGQ;
2034 		ksi.ksi_signo = signum;
2035 		ksi.ksi_code = SI_QUEUE;
2036 		ksi.ksi_pid = td->td_proc->p_pid;
2037 		ksi.ksi_uid = td->td_ucred->cr_ruid;
2038 		ksi.ksi_value = *value;
2039 		error = pksignal(p, ksi.ksi_signo, &ksi);
2040 	}
2041 	PROC_UNLOCK(p);
2042 	return (error);
2043 }
2044 
2045 /*
2046  * Send a signal to a process group.  If checktty is 1,
2047  * limit to members which have a controlling terminal.
2048  */
2049 void
2050 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
2051 {
2052 	struct proc *p;
2053 
2054 	if (pgrp) {
2055 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
2056 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
2057 			PROC_LOCK(p);
2058 			if (p->p_state == PRS_NORMAL &&
2059 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
2060 				pksignal(p, sig, ksi);
2061 			PROC_UNLOCK(p);
2062 		}
2063 	}
2064 }
2065 
2066 /*
2067  * Recalculate the signal mask and reset the signal disposition after
2068  * usermode frame for delivery is formed.  Should be called after
2069  * mach-specific routine, because sysent->sv_sendsig() needs correct
2070  * ps_siginfo and signal mask.
2071  */
2072 static void
2073 postsig_done(int sig, struct thread *td, struct sigacts *ps)
2074 {
2075 	sigset_t mask;
2076 
2077 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2078 	td->td_ru.ru_nsignals++;
2079 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
2080 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
2081 		SIGADDSET(mask, sig);
2082 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
2083 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
2084 	if (SIGISMEMBER(ps->ps_sigreset, sig))
2085 		sigdflt(ps, sig);
2086 }
2087 
2088 /*
2089  * Send a signal caused by a trap to the current thread.  If it will be
2090  * caught immediately, deliver it with correct code.  Otherwise, post it
2091  * normally.
2092  */
2093 void
2094 trapsignal(struct thread *td, ksiginfo_t *ksi)
2095 {
2096 	struct sigacts *ps;
2097 	struct proc *p;
2098 	sigset_t sigmask;
2099 	int sig;
2100 
2101 	p = td->td_proc;
2102 	sig = ksi->ksi_signo;
2103 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
2104 
2105 	sigfastblock_fetch(td);
2106 	PROC_LOCK(p);
2107 	ps = p->p_sigacts;
2108 	mtx_lock(&ps->ps_mtx);
2109 	sigmask = td->td_sigmask;
2110 	if (td->td_sigblock_val != 0)
2111 		SIGSETOR(sigmask, fastblock_mask);
2112 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2113 	    !SIGISMEMBER(sigmask, sig)) {
2114 #ifdef KTRACE
2115 		if (KTRPOINT(curthread, KTR_PSIG))
2116 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2117 			    &td->td_sigmask, ksi->ksi_code);
2118 #endif
2119 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2120 		    ksi, &td->td_sigmask);
2121 		postsig_done(sig, td, ps);
2122 		mtx_unlock(&ps->ps_mtx);
2123 	} else {
2124 		/*
2125 		 * Avoid a possible infinite loop if the thread
2126 		 * masking the signal or process is ignoring the
2127 		 * signal.
2128 		 */
2129 		if (kern_forcesigexit && (SIGISMEMBER(sigmask, sig) ||
2130 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2131 			SIGDELSET(td->td_sigmask, sig);
2132 			SIGDELSET(ps->ps_sigcatch, sig);
2133 			SIGDELSET(ps->ps_sigignore, sig);
2134 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2135 			td->td_pflags &= ~TDP_SIGFASTBLOCK;
2136 			td->td_sigblock_val = 0;
2137 		}
2138 		mtx_unlock(&ps->ps_mtx);
2139 		p->p_sig = sig;		/* XXX to verify code */
2140 		tdsendsignal(p, td, sig, ksi);
2141 	}
2142 	PROC_UNLOCK(p);
2143 }
2144 
2145 static struct thread *
2146 sigtd(struct proc *p, int sig, bool fast_sigblock)
2147 {
2148 	struct thread *td, *signal_td;
2149 
2150 	PROC_LOCK_ASSERT(p, MA_OWNED);
2151 	MPASS(!fast_sigblock || p == curproc);
2152 
2153 	/*
2154 	 * Check if current thread can handle the signal without
2155 	 * switching context to another thread.
2156 	 */
2157 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig) &&
2158 	    (!fast_sigblock || curthread->td_sigblock_val == 0))
2159 		return (curthread);
2160 
2161 	/* Find a non-stopped thread that does not mask the signal. */
2162 	signal_td = NULL;
2163 	FOREACH_THREAD_IN_PROC(p, td) {
2164 		if (!SIGISMEMBER(td->td_sigmask, sig) && (!fast_sigblock ||
2165 		    td != curthread || td->td_sigblock_val == 0) &&
2166 		    (td->td_flags & TDF_BOUNDARY) == 0) {
2167 			signal_td = td;
2168 			break;
2169 		}
2170 	}
2171 	/* Select random (first) thread if no better match was found. */
2172 	if (signal_td == NULL)
2173 		signal_td = FIRST_THREAD_IN_PROC(p);
2174 	return (signal_td);
2175 }
2176 
2177 /*
2178  * Send the signal to the process.  If the signal has an action, the action
2179  * is usually performed by the target process rather than the caller; we add
2180  * the signal to the set of pending signals for the process.
2181  *
2182  * Exceptions:
2183  *   o When a stop signal is sent to a sleeping process that takes the
2184  *     default action, the process is stopped without awakening it.
2185  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2186  *     regardless of the signal action (eg, blocked or ignored).
2187  *
2188  * Other ignored signals are discarded immediately.
2189  *
2190  * NB: This function may be entered from the debugger via the "kill" DDB
2191  * command.  There is little that can be done to mitigate the possibly messy
2192  * side effects of this unwise possibility.
2193  */
2194 void
2195 kern_psignal(struct proc *p, int sig)
2196 {
2197 	ksiginfo_t ksi;
2198 
2199 	ksiginfo_init(&ksi);
2200 	ksi.ksi_signo = sig;
2201 	ksi.ksi_code = SI_KERNEL;
2202 	(void) tdsendsignal(p, NULL, sig, &ksi);
2203 }
2204 
2205 int
2206 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2207 {
2208 
2209 	return (tdsendsignal(p, NULL, sig, ksi));
2210 }
2211 
2212 /* Utility function for finding a thread to send signal event to. */
2213 int
2214 sigev_findtd(struct proc *p, struct sigevent *sigev, struct thread **ttd)
2215 {
2216 	struct thread *td;
2217 
2218 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2219 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2220 		if (td == NULL)
2221 			return (ESRCH);
2222 		*ttd = td;
2223 	} else {
2224 		*ttd = NULL;
2225 		PROC_LOCK(p);
2226 	}
2227 	return (0);
2228 }
2229 
2230 void
2231 tdsignal(struct thread *td, int sig)
2232 {
2233 	ksiginfo_t ksi;
2234 
2235 	ksiginfo_init(&ksi);
2236 	ksi.ksi_signo = sig;
2237 	ksi.ksi_code = SI_KERNEL;
2238 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2239 }
2240 
2241 void
2242 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2243 {
2244 
2245 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2246 }
2247 
2248 static int
2249 sig_sleepq_abort(struct thread *td, int intrval)
2250 {
2251 	THREAD_LOCK_ASSERT(td, MA_OWNED);
2252 
2253 	if (intrval == 0 && (td->td_flags & TDF_SIGWAIT) == 0) {
2254 		thread_unlock(td);
2255 		return (0);
2256 	}
2257 	return (sleepq_abort(td, intrval));
2258 }
2259 
2260 int
2261 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2262 {
2263 	sig_t action;
2264 	sigqueue_t *sigqueue;
2265 	int prop;
2266 	struct sigacts *ps;
2267 	int intrval;
2268 	int ret = 0;
2269 	int wakeup_swapper;
2270 
2271 	MPASS(td == NULL || p == td->td_proc);
2272 	PROC_LOCK_ASSERT(p, MA_OWNED);
2273 
2274 	if (!_SIG_VALID(sig))
2275 		panic("%s(): invalid signal %d", __func__, sig);
2276 
2277 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2278 
2279 	/*
2280 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2281 	 */
2282 	if (p->p_state == PRS_ZOMBIE) {
2283 		if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2284 			ksiginfo_tryfree(ksi);
2285 		return (ret);
2286 	}
2287 
2288 	ps = p->p_sigacts;
2289 	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2290 	prop = sigprop(sig);
2291 
2292 	if (td == NULL) {
2293 		td = sigtd(p, sig, false);
2294 		sigqueue = &p->p_sigqueue;
2295 	} else
2296 		sigqueue = &td->td_sigqueue;
2297 
2298 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2299 
2300 	/*
2301 	 * If the signal is being ignored, then we forget about it
2302 	 * immediately, except when the target process executes
2303 	 * sigwait().  (Note: we don't set SIGCONT in ps_sigignore,
2304 	 * and if it is set to SIG_IGN, action will be SIG_DFL here.)
2305 	 */
2306 	mtx_lock(&ps->ps_mtx);
2307 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2308 		if (kern_sig_discard_ign &&
2309 		    (p->p_sysent->sv_flags & SV_SIG_DISCIGN) == 0) {
2310 			SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2311 
2312 			mtx_unlock(&ps->ps_mtx);
2313 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2314 				ksiginfo_tryfree(ksi);
2315 			return (ret);
2316 		} else {
2317 			action = SIG_CATCH;
2318 			intrval = 0;
2319 		}
2320 	} else {
2321 		if (SIGISMEMBER(td->td_sigmask, sig))
2322 			action = SIG_HOLD;
2323 		else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2324 			action = SIG_CATCH;
2325 		else
2326 			action = SIG_DFL;
2327 		if (SIGISMEMBER(ps->ps_sigintr, sig))
2328 			intrval = EINTR;
2329 		else
2330 			intrval = ERESTART;
2331 	}
2332 	mtx_unlock(&ps->ps_mtx);
2333 
2334 	if (prop & SIGPROP_CONT)
2335 		sigqueue_delete_stopmask_proc(p);
2336 	else if (prop & SIGPROP_STOP) {
2337 		/*
2338 		 * If sending a tty stop signal to a member of an orphaned
2339 		 * process group, discard the signal here if the action
2340 		 * is default; don't stop the process below if sleeping,
2341 		 * and don't clear any pending SIGCONT.
2342 		 */
2343 		if ((prop & SIGPROP_TTYSTOP) != 0 &&
2344 		    (p->p_pgrp->pg_flags & PGRP_ORPHANED) != 0 &&
2345 		    action == SIG_DFL) {
2346 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2347 				ksiginfo_tryfree(ksi);
2348 			return (ret);
2349 		}
2350 		sigqueue_delete_proc(p, SIGCONT);
2351 		if (p->p_flag & P_CONTINUED) {
2352 			p->p_flag &= ~P_CONTINUED;
2353 			PROC_LOCK(p->p_pptr);
2354 			sigqueue_take(p->p_ksi);
2355 			PROC_UNLOCK(p->p_pptr);
2356 		}
2357 	}
2358 
2359 	ret = sigqueue_add(sigqueue, sig, ksi);
2360 	if (ret != 0)
2361 		return (ret);
2362 	signotify(td);
2363 	/*
2364 	 * Defer further processing for signals which are held,
2365 	 * except that stopped processes must be continued by SIGCONT.
2366 	 */
2367 	if (action == SIG_HOLD &&
2368 	    !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2369 		return (ret);
2370 
2371 	wakeup_swapper = 0;
2372 
2373 	/*
2374 	 * Some signals have a process-wide effect and a per-thread
2375 	 * component.  Most processing occurs when the process next
2376 	 * tries to cross the user boundary, however there are some
2377 	 * times when processing needs to be done immediately, such as
2378 	 * waking up threads so that they can cross the user boundary.
2379 	 * We try to do the per-process part here.
2380 	 */
2381 	if (P_SHOULDSTOP(p)) {
2382 		KASSERT(!(p->p_flag & P_WEXIT),
2383 		    ("signal to stopped but exiting process"));
2384 		if (sig == SIGKILL) {
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 			 * SIGKILL sets process running.
2393 			 * It will die elsewhere.
2394 			 * All threads must be restarted.
2395 			 */
2396 			p->p_flag &= ~P_STOPPED_SIG;
2397 			goto runfast;
2398 		}
2399 
2400 		if (prop & SIGPROP_CONT) {
2401 			/*
2402 			 * If traced process is already stopped,
2403 			 * then no further action is necessary.
2404 			 */
2405 			if (p->p_flag & P_TRACED)
2406 				goto out;
2407 			/*
2408 			 * If SIGCONT is default (or ignored), we continue the
2409 			 * process but don't leave the signal in sigqueue as
2410 			 * it has no further action.  If SIGCONT is held, we
2411 			 * continue the process and leave the signal in
2412 			 * sigqueue.  If the process catches SIGCONT, let it
2413 			 * handle the signal itself.  If it isn't waiting on
2414 			 * an event, it goes back to run state.
2415 			 * Otherwise, process goes back to sleep state.
2416 			 */
2417 			p->p_flag &= ~P_STOPPED_SIG;
2418 			PROC_SLOCK(p);
2419 			if (p->p_numthreads == p->p_suspcount) {
2420 				PROC_SUNLOCK(p);
2421 				p->p_flag |= P_CONTINUED;
2422 				p->p_xsig = SIGCONT;
2423 				PROC_LOCK(p->p_pptr);
2424 				childproc_continued(p);
2425 				PROC_UNLOCK(p->p_pptr);
2426 				PROC_SLOCK(p);
2427 			}
2428 			if (action == SIG_DFL) {
2429 				thread_unsuspend(p);
2430 				PROC_SUNLOCK(p);
2431 				sigqueue_delete(sigqueue, sig);
2432 				goto out_cont;
2433 			}
2434 			if (action == SIG_CATCH) {
2435 				/*
2436 				 * The process wants to catch it so it needs
2437 				 * to run at least one thread, but which one?
2438 				 */
2439 				PROC_SUNLOCK(p);
2440 				goto runfast;
2441 			}
2442 			/*
2443 			 * The signal is not ignored or caught.
2444 			 */
2445 			thread_unsuspend(p);
2446 			PROC_SUNLOCK(p);
2447 			goto out_cont;
2448 		}
2449 
2450 		if (prop & SIGPROP_STOP) {
2451 			/*
2452 			 * If traced process is already stopped,
2453 			 * then no further action is necessary.
2454 			 */
2455 			if (p->p_flag & P_TRACED)
2456 				goto out;
2457 			/*
2458 			 * Already stopped, don't need to stop again
2459 			 * (If we did the shell could get confused).
2460 			 * Just make sure the signal STOP bit set.
2461 			 */
2462 			p->p_flag |= P_STOPPED_SIG;
2463 			sigqueue_delete(sigqueue, sig);
2464 			goto out;
2465 		}
2466 
2467 		/*
2468 		 * All other kinds of signals:
2469 		 * If a thread is sleeping interruptibly, simulate a
2470 		 * wakeup so that when it is continued it will be made
2471 		 * runnable and can look at the signal.  However, don't make
2472 		 * the PROCESS runnable, leave it stopped.
2473 		 * It may run a bit until it hits a thread_suspend_check().
2474 		 */
2475 		PROC_SLOCK(p);
2476 		thread_lock(td);
2477 		if (TD_CAN_ABORT(td))
2478 			wakeup_swapper = sig_sleepq_abort(td, intrval);
2479 		else
2480 			thread_unlock(td);
2481 		PROC_SUNLOCK(p);
2482 		goto out;
2483 		/*
2484 		 * Mutexes are short lived. Threads waiting on them will
2485 		 * hit thread_suspend_check() soon.
2486 		 */
2487 	} else if (p->p_state == PRS_NORMAL) {
2488 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2489 			tdsigwakeup(td, sig, action, intrval);
2490 			goto out;
2491 		}
2492 
2493 		MPASS(action == SIG_DFL);
2494 
2495 		if (prop & SIGPROP_STOP) {
2496 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2497 				goto out;
2498 			p->p_flag |= P_STOPPED_SIG;
2499 			p->p_xsig = sig;
2500 			PROC_SLOCK(p);
2501 			wakeup_swapper = sig_suspend_threads(td, p);
2502 			if (p->p_numthreads == p->p_suspcount) {
2503 				/*
2504 				 * only thread sending signal to another
2505 				 * process can reach here, if thread is sending
2506 				 * signal to its process, because thread does
2507 				 * not suspend itself here, p_numthreads
2508 				 * should never be equal to p_suspcount.
2509 				 */
2510 				thread_stopped(p);
2511 				PROC_SUNLOCK(p);
2512 				sigqueue_delete_proc(p, p->p_xsig);
2513 			} else
2514 				PROC_SUNLOCK(p);
2515 			goto out;
2516 		}
2517 	} else {
2518 		/* Not in "NORMAL" state. discard the signal. */
2519 		sigqueue_delete(sigqueue, sig);
2520 		goto out;
2521 	}
2522 
2523 	/*
2524 	 * The process is not stopped so we need to apply the signal to all the
2525 	 * running threads.
2526 	 */
2527 runfast:
2528 	tdsigwakeup(td, sig, action, intrval);
2529 	PROC_SLOCK(p);
2530 	thread_unsuspend(p);
2531 	PROC_SUNLOCK(p);
2532 out_cont:
2533 	itimer_proc_continue(p);
2534 	kqtimer_proc_continue(p);
2535 out:
2536 	/* If we jump here, proc slock should not be owned. */
2537 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2538 	if (wakeup_swapper)
2539 		kick_proc0();
2540 
2541 	return (ret);
2542 }
2543 
2544 /*
2545  * The force of a signal has been directed against a single
2546  * thread.  We need to see what we can do about knocking it
2547  * out of any sleep it may be in etc.
2548  */
2549 static void
2550 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2551 {
2552 	struct proc *p = td->td_proc;
2553 	int prop, wakeup_swapper;
2554 
2555 	PROC_LOCK_ASSERT(p, MA_OWNED);
2556 	prop = sigprop(sig);
2557 
2558 	PROC_SLOCK(p);
2559 	thread_lock(td);
2560 	/*
2561 	 * Bring the priority of a thread up if we want it to get
2562 	 * killed in this lifetime.  Be careful to avoid bumping the
2563 	 * priority of the idle thread, since we still allow to signal
2564 	 * kernel processes.
2565 	 */
2566 	if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2567 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2568 		sched_prio(td, PUSER);
2569 	if (TD_ON_SLEEPQ(td)) {
2570 		/*
2571 		 * If thread is sleeping uninterruptibly
2572 		 * we can't interrupt the sleep... the signal will
2573 		 * be noticed when the process returns through
2574 		 * trap() or syscall().
2575 		 */
2576 		if ((td->td_flags & TDF_SINTR) == 0)
2577 			goto out;
2578 		/*
2579 		 * If SIGCONT is default (or ignored) and process is
2580 		 * asleep, we are finished; the process should not
2581 		 * be awakened.
2582 		 */
2583 		if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2584 			thread_unlock(td);
2585 			PROC_SUNLOCK(p);
2586 			sigqueue_delete(&p->p_sigqueue, sig);
2587 			/*
2588 			 * It may be on either list in this state.
2589 			 * Remove from both for now.
2590 			 */
2591 			sigqueue_delete(&td->td_sigqueue, sig);
2592 			return;
2593 		}
2594 
2595 		/*
2596 		 * Don't awaken a sleeping thread for SIGSTOP if the
2597 		 * STOP signal is deferred.
2598 		 */
2599 		if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2600 		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2601 			goto out;
2602 
2603 		/*
2604 		 * Give low priority threads a better chance to run.
2605 		 */
2606 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2607 			sched_prio(td, PUSER);
2608 
2609 		wakeup_swapper = sig_sleepq_abort(td, intrval);
2610 		PROC_SUNLOCK(p);
2611 		if (wakeup_swapper)
2612 			kick_proc0();
2613 		return;
2614 	}
2615 
2616 	/*
2617 	 * Other states do nothing with the signal immediately,
2618 	 * other than kicking ourselves if we are running.
2619 	 * It will either never be noticed, or noticed very soon.
2620 	 */
2621 #ifdef SMP
2622 	if (TD_IS_RUNNING(td) && td != curthread)
2623 		forward_signal(td);
2624 #endif
2625 
2626 out:
2627 	PROC_SUNLOCK(p);
2628 	thread_unlock(td);
2629 }
2630 
2631 static void
2632 ptrace_coredumpreq(struct thread *td, struct proc *p,
2633     struct thr_coredump_req *tcq)
2634 {
2635 	void *rl_cookie;
2636 
2637 	if (p->p_sysent->sv_coredump == NULL) {
2638 		tcq->tc_error = ENOSYS;
2639 		return;
2640 	}
2641 
2642 	rl_cookie = vn_rangelock_wlock(tcq->tc_vp, 0, OFF_MAX);
2643 	tcq->tc_error = p->p_sysent->sv_coredump(td, tcq->tc_vp,
2644 	    tcq->tc_limit, tcq->tc_flags);
2645 	vn_rangelock_unlock(tcq->tc_vp, rl_cookie);
2646 }
2647 
2648 static void
2649 ptrace_syscallreq(struct thread *td, struct proc *p,
2650     struct thr_syscall_req *tsr)
2651 {
2652 	struct sysentvec *sv;
2653 	struct sysent *se;
2654 	register_t rv_saved[2];
2655 	int error, nerror;
2656 	int sc;
2657 	bool audited, sy_thr_static;
2658 
2659 	sv = p->p_sysent;
2660 	if (sv->sv_table == NULL || sv->sv_size < tsr->ts_sa.code) {
2661 		tsr->ts_ret.sr_error = ENOSYS;
2662 		return;
2663 	}
2664 
2665 	sc = tsr->ts_sa.code;
2666 	if (sc == SYS_syscall || sc == SYS___syscall) {
2667 		sc = tsr->ts_sa.args[0];
2668 		memmove(&tsr->ts_sa.args[0], &tsr->ts_sa.args[1],
2669 		    sizeof(register_t) * (tsr->ts_nargs - 1));
2670 	}
2671 
2672 	tsr->ts_sa.callp = se = &sv->sv_table[sc];
2673 
2674 	VM_CNT_INC(v_syscall);
2675 	td->td_pticks = 0;
2676 	if (__predict_false(td->td_cowgen != atomic_load_int(
2677 	    &td->td_proc->p_cowgen)))
2678 		thread_cow_update(td);
2679 
2680 #ifdef CAPABILITY_MODE
2681 	if (IN_CAPABILITY_MODE(td) && (se->sy_flags & SYF_CAPENABLED) == 0) {
2682 		tsr->ts_ret.sr_error = ECAPMODE;
2683 		return;
2684 	}
2685 #endif
2686 
2687 	sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2688 	audited = AUDIT_SYSCALL_ENTER(sc, td) != 0;
2689 
2690 	if (!sy_thr_static) {
2691 		error = syscall_thread_enter(td, &se);
2692 		sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2693 		if (error != 0) {
2694 			tsr->ts_ret.sr_error = error;
2695 			return;
2696 		}
2697 	}
2698 
2699 	rv_saved[0] = td->td_retval[0];
2700 	rv_saved[1] = td->td_retval[1];
2701 	nerror = td->td_errno;
2702 	td->td_retval[0] = 0;
2703 	td->td_retval[1] = 0;
2704 
2705 #ifdef KDTRACE_HOOKS
2706 	if (se->sy_entry != 0)
2707 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_ENTRY, 0);
2708 #endif
2709 	tsr->ts_ret.sr_error = se->sy_call(td, tsr->ts_sa.args);
2710 #ifdef KDTRACE_HOOKS
2711 	if (se->sy_return != 0)
2712 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_RETURN,
2713 		    tsr->ts_ret.sr_error != 0 ? -1 : td->td_retval[0]);
2714 #endif
2715 
2716 	tsr->ts_ret.sr_retval[0] = td->td_retval[0];
2717 	tsr->ts_ret.sr_retval[1] = td->td_retval[1];
2718 	td->td_retval[0] = rv_saved[0];
2719 	td->td_retval[1] = rv_saved[1];
2720 	td->td_errno = nerror;
2721 
2722 	if (audited)
2723 		AUDIT_SYSCALL_EXIT(error, td);
2724 	if (!sy_thr_static)
2725 		syscall_thread_exit(td, se);
2726 }
2727 
2728 static void
2729 ptrace_remotereq(struct thread *td, int flag)
2730 {
2731 	struct proc *p;
2732 
2733 	MPASS(td == curthread);
2734 	p = td->td_proc;
2735 	PROC_LOCK_ASSERT(p, MA_OWNED);
2736 	if ((td->td_dbgflags & flag) == 0)
2737 		return;
2738 	KASSERT((p->p_flag & P_STOPPED_TRACE) != 0, ("not stopped"));
2739 	KASSERT(td->td_remotereq != NULL, ("td_remotereq is NULL"));
2740 
2741 	PROC_UNLOCK(p);
2742 	switch (flag) {
2743 	case TDB_COREDUMPREQ:
2744 		ptrace_coredumpreq(td, p, td->td_remotereq);
2745 		break;
2746 	case TDB_SCREMOTEREQ:
2747 		ptrace_syscallreq(td, p, td->td_remotereq);
2748 		break;
2749 	default:
2750 		__unreachable();
2751 	}
2752 	PROC_LOCK(p);
2753 
2754 	MPASS((td->td_dbgflags & flag) != 0);
2755 	td->td_dbgflags &= ~flag;
2756 	td->td_remotereq = NULL;
2757 	wakeup(p);
2758 }
2759 
2760 static int
2761 sig_suspend_threads(struct thread *td, struct proc *p)
2762 {
2763 	struct thread *td2;
2764 	int wakeup_swapper;
2765 
2766 	PROC_LOCK_ASSERT(p, MA_OWNED);
2767 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2768 
2769 	wakeup_swapper = 0;
2770 	FOREACH_THREAD_IN_PROC(p, td2) {
2771 		thread_lock(td2);
2772 		ast_sched_locked(td2, TDA_SUSPEND);
2773 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2774 		    (td2->td_flags & TDF_SINTR)) {
2775 			if (td2->td_flags & TDF_SBDRY) {
2776 				/*
2777 				 * Once a thread is asleep with
2778 				 * TDF_SBDRY and without TDF_SERESTART
2779 				 * or TDF_SEINTR set, it should never
2780 				 * become suspended due to this check.
2781 				 */
2782 				KASSERT(!TD_IS_SUSPENDED(td2),
2783 				    ("thread with deferred stops suspended"));
2784 				if (TD_SBDRY_INTR(td2)) {
2785 					wakeup_swapper |= sleepq_abort(td2,
2786 					    TD_SBDRY_ERRNO(td2));
2787 					continue;
2788 				}
2789 			} else if (!TD_IS_SUSPENDED(td2))
2790 				thread_suspend_one(td2);
2791 		} else if (!TD_IS_SUSPENDED(td2)) {
2792 #ifdef SMP
2793 			if (TD_IS_RUNNING(td2) && td2 != td)
2794 				forward_signal(td2);
2795 #endif
2796 		}
2797 		thread_unlock(td2);
2798 	}
2799 	return (wakeup_swapper);
2800 }
2801 
2802 /*
2803  * Stop the process for an event deemed interesting to the debugger. If si is
2804  * non-NULL, this is a signal exchange; the new signal requested by the
2805  * debugger will be returned for handling. If si is NULL, this is some other
2806  * type of interesting event. The debugger may request a signal be delivered in
2807  * that case as well, however it will be deferred until it can be handled.
2808  */
2809 int
2810 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2811 {
2812 	struct proc *p = td->td_proc;
2813 	struct thread *td2;
2814 	ksiginfo_t ksi;
2815 
2816 	PROC_LOCK_ASSERT(p, MA_OWNED);
2817 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2818 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2819 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2820 
2821 	td->td_xsig = sig;
2822 
2823 	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2824 		td->td_dbgflags |= TDB_XSIG;
2825 		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2826 		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2827 		PROC_SLOCK(p);
2828 		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2829 			if (P_KILLED(p)) {
2830 				/*
2831 				 * Ensure that, if we've been PT_KILLed, the
2832 				 * exit status reflects that. Another thread
2833 				 * may also be in ptracestop(), having just
2834 				 * received the SIGKILL, but this thread was
2835 				 * unsuspended first.
2836 				 */
2837 				td->td_dbgflags &= ~TDB_XSIG;
2838 				td->td_xsig = SIGKILL;
2839 				p->p_ptevents = 0;
2840 				break;
2841 			}
2842 			if (p->p_flag & P_SINGLE_EXIT &&
2843 			    !(td->td_dbgflags & TDB_EXIT)) {
2844 				/*
2845 				 * Ignore ptrace stops except for thread exit
2846 				 * events when the process exits.
2847 				 */
2848 				td->td_dbgflags &= ~TDB_XSIG;
2849 				PROC_SUNLOCK(p);
2850 				return (0);
2851 			}
2852 
2853 			/*
2854 			 * Make wait(2) work.  Ensure that right after the
2855 			 * attach, the thread which was decided to become the
2856 			 * leader of attach gets reported to the waiter.
2857 			 * Otherwise, just avoid overwriting another thread's
2858 			 * assignment to p_xthread.  If another thread has
2859 			 * already set p_xthread, the current thread will get
2860 			 * a chance to report itself upon the next iteration.
2861 			 */
2862 			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2863 			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2864 			    p->p_xthread == NULL)) {
2865 				p->p_xsig = sig;
2866 				p->p_xthread = td;
2867 
2868 				/*
2869 				 * If we are on sleepqueue already,
2870 				 * let sleepqueue code decide if it
2871 				 * needs to go sleep after attach.
2872 				 */
2873 				if (td->td_wchan == NULL)
2874 					td->td_dbgflags &= ~TDB_FSTP;
2875 
2876 				p->p_flag2 &= ~P2_PTRACE_FSTP;
2877 				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2878 				sig_suspend_threads(td, p);
2879 			}
2880 			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2881 				td->td_dbgflags &= ~TDB_STOPATFORK;
2882 			}
2883 stopme:
2884 			td->td_dbgflags |= TDB_SSWITCH;
2885 			thread_suspend_switch(td, p);
2886 			td->td_dbgflags &= ~TDB_SSWITCH;
2887 			if ((td->td_dbgflags & (TDB_COREDUMPREQ |
2888 			    TDB_SCREMOTEREQ)) != 0) {
2889 				MPASS((td->td_dbgflags & (TDB_COREDUMPREQ |
2890 				    TDB_SCREMOTEREQ)) !=
2891 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2892 				PROC_SUNLOCK(p);
2893 				ptrace_remotereq(td, td->td_dbgflags &
2894 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2895 				PROC_SLOCK(p);
2896 				goto stopme;
2897 			}
2898 			if (p->p_xthread == td)
2899 				p->p_xthread = NULL;
2900 			if (!(p->p_flag & P_TRACED))
2901 				break;
2902 			if (td->td_dbgflags & TDB_SUSPEND) {
2903 				if (p->p_flag & P_SINGLE_EXIT)
2904 					break;
2905 				goto stopme;
2906 			}
2907 		}
2908 		PROC_SUNLOCK(p);
2909 	}
2910 
2911 	if (si != NULL && sig == td->td_xsig) {
2912 		/* Parent wants us to take the original signal unchanged. */
2913 		si->ksi_flags |= KSI_HEAD;
2914 		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2915 			si->ksi_signo = 0;
2916 	} else if (td->td_xsig != 0) {
2917 		/*
2918 		 * If parent wants us to take a new signal, then it will leave
2919 		 * it in td->td_xsig; otherwise we just look for signals again.
2920 		 */
2921 		ksiginfo_init(&ksi);
2922 		ksi.ksi_signo = td->td_xsig;
2923 		ksi.ksi_flags |= KSI_PTRACE;
2924 		td2 = sigtd(p, td->td_xsig, false);
2925 		tdsendsignal(p, td2, td->td_xsig, &ksi);
2926 		if (td != td2)
2927 			return (0);
2928 	}
2929 
2930 	return (td->td_xsig);
2931 }
2932 
2933 static void
2934 reschedule_signals(struct proc *p, sigset_t block, int flags)
2935 {
2936 	struct sigacts *ps;
2937 	struct thread *td;
2938 	int sig;
2939 	bool fastblk, pslocked;
2940 
2941 	PROC_LOCK_ASSERT(p, MA_OWNED);
2942 	ps = p->p_sigacts;
2943 	pslocked = (flags & SIGPROCMASK_PS_LOCKED) != 0;
2944 	mtx_assert(&ps->ps_mtx, pslocked ? MA_OWNED : MA_NOTOWNED);
2945 	if (SIGISEMPTY(p->p_siglist))
2946 		return;
2947 	SIGSETAND(block, p->p_siglist);
2948 	fastblk = (flags & SIGPROCMASK_FASTBLK) != 0;
2949 	SIG_FOREACH(sig, &block) {
2950 		td = sigtd(p, sig, fastblk);
2951 
2952 		/*
2953 		 * If sigtd() selected us despite sigfastblock is
2954 		 * blocking, do not activate AST or wake us, to avoid
2955 		 * loop in AST handler.
2956 		 */
2957 		if (fastblk && td == curthread)
2958 			continue;
2959 
2960 		signotify(td);
2961 		if (!pslocked)
2962 			mtx_lock(&ps->ps_mtx);
2963 		if (p->p_flag & P_TRACED ||
2964 		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2965 		    !SIGISMEMBER(td->td_sigmask, sig))) {
2966 			tdsigwakeup(td, sig, SIG_CATCH,
2967 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2968 			    ERESTART));
2969 		}
2970 		if (!pslocked)
2971 			mtx_unlock(&ps->ps_mtx);
2972 	}
2973 }
2974 
2975 void
2976 tdsigcleanup(struct thread *td)
2977 {
2978 	struct proc *p;
2979 	sigset_t unblocked;
2980 
2981 	p = td->td_proc;
2982 	PROC_LOCK_ASSERT(p, MA_OWNED);
2983 
2984 	sigqueue_flush(&td->td_sigqueue);
2985 	if (p->p_numthreads == 1)
2986 		return;
2987 
2988 	/*
2989 	 * Since we cannot handle signals, notify signal post code
2990 	 * about this by filling the sigmask.
2991 	 *
2992 	 * Also, if needed, wake up thread(s) that do not block the
2993 	 * same signals as the exiting thread, since the thread might
2994 	 * have been selected for delivery and woken up.
2995 	 */
2996 	SIGFILLSET(unblocked);
2997 	SIGSETNAND(unblocked, td->td_sigmask);
2998 	SIGFILLSET(td->td_sigmask);
2999 	reschedule_signals(p, unblocked, 0);
3000 
3001 }
3002 
3003 static int
3004 sigdeferstop_curr_flags(int cflags)
3005 {
3006 
3007 	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
3008 	    (cflags & TDF_SBDRY) != 0);
3009 	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
3010 }
3011 
3012 /*
3013  * Defer the delivery of SIGSTOP for the current thread, according to
3014  * the requested mode.  Returns previous flags, which must be restored
3015  * by sigallowstop().
3016  *
3017  * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
3018  * cleared by the current thread, which allow the lock-less read-only
3019  * accesses below.
3020  */
3021 int
3022 sigdeferstop_impl(int mode)
3023 {
3024 	struct thread *td;
3025 	int cflags, nflags;
3026 
3027 	td = curthread;
3028 	cflags = sigdeferstop_curr_flags(td->td_flags);
3029 	switch (mode) {
3030 	case SIGDEFERSTOP_NOP:
3031 		nflags = cflags;
3032 		break;
3033 	case SIGDEFERSTOP_OFF:
3034 		nflags = 0;
3035 		break;
3036 	case SIGDEFERSTOP_SILENT:
3037 		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
3038 		break;
3039 	case SIGDEFERSTOP_EINTR:
3040 		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
3041 		break;
3042 	case SIGDEFERSTOP_ERESTART:
3043 		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
3044 		break;
3045 	default:
3046 		panic("sigdeferstop: invalid mode %x", mode);
3047 		break;
3048 	}
3049 	if (cflags == nflags)
3050 		return (SIGDEFERSTOP_VAL_NCHG);
3051 	thread_lock(td);
3052 	td->td_flags = (td->td_flags & ~cflags) | nflags;
3053 	thread_unlock(td);
3054 	return (cflags);
3055 }
3056 
3057 /*
3058  * Restores the STOP handling mode, typically permitting the delivery
3059  * of SIGSTOP for the current thread.  This does not immediately
3060  * suspend if a stop was posted.  Instead, the thread will suspend
3061  * either via ast() or a subsequent interruptible sleep.
3062  */
3063 void
3064 sigallowstop_impl(int prev)
3065 {
3066 	struct thread *td;
3067 	int cflags;
3068 
3069 	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
3070 	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
3071 	    ("sigallowstop: incorrect previous mode %x", prev));
3072 	td = curthread;
3073 	cflags = sigdeferstop_curr_flags(td->td_flags);
3074 	if (cflags != prev) {
3075 		thread_lock(td);
3076 		td->td_flags = (td->td_flags & ~cflags) | prev;
3077 		thread_unlock(td);
3078 	}
3079 }
3080 
3081 enum sigstatus {
3082 	SIGSTATUS_HANDLE,
3083 	SIGSTATUS_HANDLED,
3084 	SIGSTATUS_IGNORE,
3085 	SIGSTATUS_SBDRY_STOP,
3086 };
3087 
3088 /*
3089  * The thread has signal "sig" pending.  Figure out what to do with it:
3090  *
3091  * _HANDLE     -> the caller should handle the signal
3092  * _HANDLED    -> handled internally, reload pending signal set
3093  * _IGNORE     -> ignored, remove from the set of pending signals and try the
3094  *                next pending signal
3095  * _SBDRY_STOP -> the signal should stop the thread but this is not
3096  *                permitted in the current context
3097  */
3098 static enum sigstatus
3099 sigprocess(struct thread *td, int sig)
3100 {
3101 	struct proc *p;
3102 	struct sigacts *ps;
3103 	struct sigqueue *queue;
3104 	ksiginfo_t ksi;
3105 	int prop;
3106 
3107 	KASSERT(_SIG_VALID(sig), ("%s: invalid signal %d", __func__, sig));
3108 
3109 	p = td->td_proc;
3110 	ps = p->p_sigacts;
3111 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3112 	PROC_LOCK_ASSERT(p, MA_OWNED);
3113 
3114 	/*
3115 	 * We should allow pending but ignored signals below
3116 	 * if there is sigwait() active, or P_TRACED was
3117 	 * on when they were posted.
3118 	 */
3119 	if (SIGISMEMBER(ps->ps_sigignore, sig) &&
3120 	    (p->p_flag & P_TRACED) == 0 &&
3121 	    (td->td_flags & TDF_SIGWAIT) == 0) {
3122 		return (SIGSTATUS_IGNORE);
3123 	}
3124 
3125 	/*
3126 	 * If the process is going to single-thread mode to prepare
3127 	 * for exit, there is no sense in delivering any signal
3128 	 * to usermode.  Another important consequence is that
3129 	 * msleep(..., PCATCH, ...) now is only interruptible by a
3130 	 * suspend request.
3131 	 */
3132 	if ((p->p_flag2 & P2_WEXIT) != 0)
3133 		return (SIGSTATUS_IGNORE);
3134 
3135 	if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
3136 		/*
3137 		 * If traced, always stop.
3138 		 * Remove old signal from queue before the stop.
3139 		 * XXX shrug off debugger, it causes siginfo to
3140 		 * be thrown away.
3141 		 */
3142 		queue = &td->td_sigqueue;
3143 		ksiginfo_init(&ksi);
3144 		if (sigqueue_get(queue, sig, &ksi) == 0) {
3145 			queue = &p->p_sigqueue;
3146 			sigqueue_get(queue, sig, &ksi);
3147 		}
3148 		td->td_si = ksi.ksi_info;
3149 
3150 		mtx_unlock(&ps->ps_mtx);
3151 		sig = ptracestop(td, sig, &ksi);
3152 		mtx_lock(&ps->ps_mtx);
3153 
3154 		td->td_si.si_signo = 0;
3155 
3156 		/*
3157 		 * Keep looking if the debugger discarded or
3158 		 * replaced the signal.
3159 		 */
3160 		if (sig == 0)
3161 			return (SIGSTATUS_HANDLED);
3162 
3163 		/*
3164 		 * If the signal became masked, re-queue it.
3165 		 */
3166 		if (SIGISMEMBER(td->td_sigmask, sig)) {
3167 			ksi.ksi_flags |= KSI_HEAD;
3168 			sigqueue_add(&p->p_sigqueue, sig, &ksi);
3169 			return (SIGSTATUS_HANDLED);
3170 		}
3171 
3172 		/*
3173 		 * If the traced bit got turned off, requeue the signal and
3174 		 * reload the set of pending signals.  This ensures that p_sig*
3175 		 * and p_sigact are consistent.
3176 		 */
3177 		if ((p->p_flag & P_TRACED) == 0) {
3178 			if ((ksi.ksi_flags & KSI_PTRACE) == 0) {
3179 				ksi.ksi_flags |= KSI_HEAD;
3180 				sigqueue_add(queue, sig, &ksi);
3181 			}
3182 			return (SIGSTATUS_HANDLED);
3183 		}
3184 	}
3185 
3186 	/*
3187 	 * Decide whether the signal should be returned.
3188 	 * Return the signal's number, or fall through
3189 	 * to clear it from the pending mask.
3190 	 */
3191 	switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
3192 	case (intptr_t)SIG_DFL:
3193 		/*
3194 		 * Don't take default actions on system processes.
3195 		 */
3196 		if (p->p_pid <= 1) {
3197 #ifdef DIAGNOSTIC
3198 			/*
3199 			 * Are you sure you want to ignore SIGSEGV
3200 			 * in init? XXX
3201 			 */
3202 			printf("Process (pid %lu) got signal %d\n",
3203 				(u_long)p->p_pid, sig);
3204 #endif
3205 			return (SIGSTATUS_IGNORE);
3206 		}
3207 
3208 		/*
3209 		 * If there is a pending stop signal to process with
3210 		 * default action, stop here, then clear the signal.
3211 		 * Traced or exiting processes should ignore stops.
3212 		 * Additionally, a member of an orphaned process group
3213 		 * should ignore tty stops.
3214 		 */
3215 		prop = sigprop(sig);
3216 		if (prop & SIGPROP_STOP) {
3217 			mtx_unlock(&ps->ps_mtx);
3218 			if ((p->p_flag & (P_TRACED | P_WEXIT |
3219 			    P_SINGLE_EXIT)) != 0 || ((p->p_pgrp->
3220 			    pg_flags & PGRP_ORPHANED) != 0 &&
3221 			    (prop & SIGPROP_TTYSTOP) != 0)) {
3222 				mtx_lock(&ps->ps_mtx);
3223 				return (SIGSTATUS_IGNORE);
3224 			}
3225 			if (TD_SBDRY_INTR(td)) {
3226 				KASSERT((td->td_flags & TDF_SBDRY) != 0,
3227 				    ("lost TDF_SBDRY"));
3228 				mtx_lock(&ps->ps_mtx);
3229 				return (SIGSTATUS_SBDRY_STOP);
3230 			}
3231 			WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
3232 			    &p->p_mtx.lock_object, "Catching SIGSTOP");
3233 			sigqueue_delete(&td->td_sigqueue, sig);
3234 			sigqueue_delete(&p->p_sigqueue, sig);
3235 			p->p_flag |= P_STOPPED_SIG;
3236 			p->p_xsig = sig;
3237 			PROC_SLOCK(p);
3238 			sig_suspend_threads(td, p);
3239 			thread_suspend_switch(td, p);
3240 			PROC_SUNLOCK(p);
3241 			mtx_lock(&ps->ps_mtx);
3242 			return (SIGSTATUS_HANDLED);
3243 		} else if ((prop & SIGPROP_IGNORE) != 0 &&
3244 		    (td->td_flags & TDF_SIGWAIT) == 0) {
3245 			/*
3246 			 * Default action is to ignore; drop it if
3247 			 * not in kern_sigtimedwait().
3248 			 */
3249 			return (SIGSTATUS_IGNORE);
3250 		} else {
3251 			return (SIGSTATUS_HANDLE);
3252 		}
3253 
3254 	case (intptr_t)SIG_IGN:
3255 		if ((td->td_flags & TDF_SIGWAIT) == 0)
3256 			return (SIGSTATUS_IGNORE);
3257 		else
3258 			return (SIGSTATUS_HANDLE);
3259 
3260 	default:
3261 		/*
3262 		 * This signal has an action, let postsig() process it.
3263 		 */
3264 		return (SIGSTATUS_HANDLE);
3265 	}
3266 }
3267 
3268 /*
3269  * If the current process has received a signal (should be caught or cause
3270  * termination, should interrupt current syscall), return the signal number.
3271  * Stop signals with default action are processed immediately, then cleared;
3272  * they aren't returned.  This is checked after each entry to the system for
3273  * a syscall or trap (though this can usually be done without calling
3274  * issignal by checking the pending signal masks in cursig.) The normal call
3275  * sequence is
3276  *
3277  *	while (sig = cursig(curthread))
3278  *		postsig(sig);
3279  */
3280 static int
3281 issignal(struct thread *td)
3282 {
3283 	struct proc *p;
3284 	sigset_t sigpending;
3285 	int sig;
3286 
3287 	p = td->td_proc;
3288 	PROC_LOCK_ASSERT(p, MA_OWNED);
3289 
3290 	for (;;) {
3291 		sigpending = td->td_sigqueue.sq_signals;
3292 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
3293 		SIGSETNAND(sigpending, td->td_sigmask);
3294 
3295 		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
3296 		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
3297 			SIG_STOPSIGMASK(sigpending);
3298 		if (SIGISEMPTY(sigpending))	/* no signal to send */
3299 			return (0);
3300 
3301 		/*
3302 		 * Do fast sigblock if requested by usermode.  Since
3303 		 * we do know that there was a signal pending at this
3304 		 * point, set the FAST_SIGBLOCK_PEND as indicator for
3305 		 * usermode to perform a dummy call to
3306 		 * FAST_SIGBLOCK_UNBLOCK, which causes immediate
3307 		 * delivery of postponed pending signal.
3308 		 */
3309 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
3310 			if (td->td_sigblock_val != 0)
3311 				SIGSETNAND(sigpending, fastblock_mask);
3312 			if (SIGISEMPTY(sigpending)) {
3313 				td->td_pflags |= TDP_SIGFASTPENDING;
3314 				return (0);
3315 			}
3316 		}
3317 
3318 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
3319 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
3320 		    SIGISMEMBER(sigpending, SIGSTOP)) {
3321 			/*
3322 			 * If debugger just attached, always consume
3323 			 * SIGSTOP from ptrace(PT_ATTACH) first, to
3324 			 * execute the debugger attach ritual in
3325 			 * order.
3326 			 */
3327 			td->td_dbgflags |= TDB_FSTP;
3328 			SIGEMPTYSET(sigpending);
3329 			SIGADDSET(sigpending, SIGSTOP);
3330 		}
3331 
3332 		SIG_FOREACH(sig, &sigpending) {
3333 			switch (sigprocess(td, sig)) {
3334 			case SIGSTATUS_HANDLE:
3335 				return (sig);
3336 			case SIGSTATUS_HANDLED:
3337 				goto next;
3338 			case SIGSTATUS_IGNORE:
3339 				sigqueue_delete(&td->td_sigqueue, sig);
3340 				sigqueue_delete(&p->p_sigqueue, sig);
3341 				break;
3342 			case SIGSTATUS_SBDRY_STOP:
3343 				return (-1);
3344 			}
3345 		}
3346 next:;
3347 	}
3348 }
3349 
3350 void
3351 thread_stopped(struct proc *p)
3352 {
3353 	int n;
3354 
3355 	PROC_LOCK_ASSERT(p, MA_OWNED);
3356 	PROC_SLOCK_ASSERT(p, MA_OWNED);
3357 	n = p->p_suspcount;
3358 	if (p == curproc)
3359 		n++;
3360 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3361 		PROC_SUNLOCK(p);
3362 		p->p_flag &= ~P_WAITED;
3363 		PROC_LOCK(p->p_pptr);
3364 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
3365 			CLD_TRAPPED : CLD_STOPPED);
3366 		PROC_UNLOCK(p->p_pptr);
3367 		PROC_SLOCK(p);
3368 	}
3369 }
3370 
3371 /*
3372  * Take the action for the specified signal
3373  * from the current set of pending signals.
3374  */
3375 int
3376 postsig(int sig)
3377 {
3378 	struct thread *td;
3379 	struct proc *p;
3380 	struct sigacts *ps;
3381 	sig_t action;
3382 	ksiginfo_t ksi;
3383 	sigset_t returnmask;
3384 
3385 	KASSERT(sig != 0, ("postsig"));
3386 
3387 	td = curthread;
3388 	p = td->td_proc;
3389 	PROC_LOCK_ASSERT(p, MA_OWNED);
3390 	ps = p->p_sigacts;
3391 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3392 	ksiginfo_init(&ksi);
3393 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3394 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3395 		return (0);
3396 	ksi.ksi_signo = sig;
3397 	if (ksi.ksi_code == SI_TIMER)
3398 		itimer_accept(p, ksi.ksi_timerid, &ksi);
3399 	action = ps->ps_sigact[_SIG_IDX(sig)];
3400 #ifdef KTRACE
3401 	if (KTRPOINT(td, KTR_PSIG))
3402 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3403 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3404 #endif
3405 
3406 	if (action == SIG_DFL) {
3407 		/*
3408 		 * Default action, where the default is to kill
3409 		 * the process.  (Other cases were ignored above.)
3410 		 */
3411 		mtx_unlock(&ps->ps_mtx);
3412 		proc_td_siginfo_capture(td, &ksi.ksi_info);
3413 		sigexit(td, sig);
3414 		/* NOTREACHED */
3415 	} else {
3416 		/*
3417 		 * If we get here, the signal must be caught.
3418 		 */
3419 		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3420 		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3421 		    ("postsig action: blocked sig %d", sig));
3422 
3423 		/*
3424 		 * Set the new mask value and also defer further
3425 		 * occurrences of this signal.
3426 		 *
3427 		 * Special case: user has done a sigsuspend.  Here the
3428 		 * current mask is not of interest, but rather the
3429 		 * mask from before the sigsuspend is what we want
3430 		 * restored after the signal processing is completed.
3431 		 */
3432 		if (td->td_pflags & TDP_OLDMASK) {
3433 			returnmask = td->td_oldsigmask;
3434 			td->td_pflags &= ~TDP_OLDMASK;
3435 		} else
3436 			returnmask = td->td_sigmask;
3437 
3438 		if (p->p_sig == sig) {
3439 			p->p_sig = 0;
3440 		}
3441 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3442 		postsig_done(sig, td, ps);
3443 	}
3444 	return (1);
3445 }
3446 
3447 int
3448 sig_ast_checksusp(struct thread *td)
3449 {
3450 	struct proc *p __diagused;
3451 	int ret;
3452 
3453 	p = td->td_proc;
3454 	PROC_LOCK_ASSERT(p, MA_OWNED);
3455 
3456 	if (!td_ast_pending(td, TDA_SUSPEND))
3457 		return (0);
3458 
3459 	ret = thread_suspend_check(1);
3460 	MPASS(ret == 0 || ret == EINTR || ret == ERESTART);
3461 	return (ret);
3462 }
3463 
3464 int
3465 sig_ast_needsigchk(struct thread *td)
3466 {
3467 	struct proc *p;
3468 	struct sigacts *ps;
3469 	int ret, sig;
3470 
3471 	p = td->td_proc;
3472 	PROC_LOCK_ASSERT(p, MA_OWNED);
3473 
3474 	if (!td_ast_pending(td, TDA_SIG))
3475 		return (0);
3476 
3477 	ps = p->p_sigacts;
3478 	mtx_lock(&ps->ps_mtx);
3479 	sig = cursig(td);
3480 	if (sig == -1) {
3481 		mtx_unlock(&ps->ps_mtx);
3482 		KASSERT((td->td_flags & TDF_SBDRY) != 0, ("lost TDF_SBDRY"));
3483 		KASSERT(TD_SBDRY_INTR(td),
3484 		    ("lost TDF_SERESTART of TDF_SEINTR"));
3485 		KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
3486 		    (TDF_SEINTR | TDF_SERESTART),
3487 		    ("both TDF_SEINTR and TDF_SERESTART"));
3488 		ret = TD_SBDRY_ERRNO(td);
3489 	} else if (sig != 0) {
3490 		ret = SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR : ERESTART;
3491 		mtx_unlock(&ps->ps_mtx);
3492 	} else {
3493 		mtx_unlock(&ps->ps_mtx);
3494 		ret = 0;
3495 	}
3496 
3497 	/*
3498 	 * Do not go into sleep if this thread was the ptrace(2)
3499 	 * attach leader.  cursig() consumed SIGSTOP from PT_ATTACH,
3500 	 * but we usually act on the signal by interrupting sleep, and
3501 	 * should do that here as well.
3502 	 */
3503 	if ((td->td_dbgflags & TDB_FSTP) != 0) {
3504 		if (ret == 0)
3505 			ret = EINTR;
3506 		td->td_dbgflags &= ~TDB_FSTP;
3507 	}
3508 
3509 	return (ret);
3510 }
3511 
3512 int
3513 sig_intr(void)
3514 {
3515 	struct thread *td;
3516 	struct proc *p;
3517 	int ret;
3518 
3519 	td = curthread;
3520 	if (!td_ast_pending(td, TDA_SIG) && !td_ast_pending(td, TDA_SUSPEND))
3521 		return (0);
3522 
3523 	p = td->td_proc;
3524 
3525 	PROC_LOCK(p);
3526 	ret = sig_ast_checksusp(td);
3527 	if (ret == 0)
3528 		ret = sig_ast_needsigchk(td);
3529 	PROC_UNLOCK(p);
3530 	return (ret);
3531 }
3532 
3533 bool
3534 curproc_sigkilled(void)
3535 {
3536 	struct thread *td;
3537 	struct proc *p;
3538 	struct sigacts *ps;
3539 	bool res;
3540 
3541 	td = curthread;
3542 	if (!td_ast_pending(td, TDA_SIG))
3543 		return (false);
3544 
3545 	p = td->td_proc;
3546 	PROC_LOCK(p);
3547 	ps = p->p_sigacts;
3548 	mtx_lock(&ps->ps_mtx);
3549 	res = SIGISMEMBER(td->td_sigqueue.sq_signals, SIGKILL) ||
3550 	    SIGISMEMBER(p->p_sigqueue.sq_signals, SIGKILL);
3551 	mtx_unlock(&ps->ps_mtx);
3552 	PROC_UNLOCK(p);
3553 	return (res);
3554 }
3555 
3556 void
3557 proc_wkilled(struct proc *p)
3558 {
3559 
3560 	PROC_LOCK_ASSERT(p, MA_OWNED);
3561 	if ((p->p_flag & P_WKILLED) == 0) {
3562 		p->p_flag |= P_WKILLED;
3563 		/*
3564 		 * Notify swapper that there is a process to swap in.
3565 		 * The notification is racy, at worst it would take 10
3566 		 * seconds for the swapper process to notice.
3567 		 */
3568 		if ((p->p_flag & (P_INMEM | P_SWAPPINGIN)) == 0)
3569 			wakeup(&proc0);
3570 	}
3571 }
3572 
3573 /*
3574  * Kill the current process for stated reason.
3575  */
3576 void
3577 killproc(struct proc *p, const char *why)
3578 {
3579 
3580 	PROC_LOCK_ASSERT(p, MA_OWNED);
3581 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3582 	    p->p_comm);
3583 	log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3584 	    p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3585 	    p->p_ucred->cr_uid, why);
3586 	proc_wkilled(p);
3587 	kern_psignal(p, SIGKILL);
3588 }
3589 
3590 /*
3591  * Force the current process to exit with the specified signal, dumping core
3592  * if appropriate.  We bypass the normal tests for masked and caught signals,
3593  * allowing unrecoverable failures to terminate the process without changing
3594  * signal state.  Mark the accounting record with the signal termination.
3595  * If dumping core, save the signal number for the debugger.  Calls exit and
3596  * does not return.
3597  */
3598 void
3599 sigexit(struct thread *td, int sig)
3600 {
3601 	struct proc *p = td->td_proc;
3602 	const char *coreinfo;
3603 	int rv;
3604 
3605 	PROC_LOCK_ASSERT(p, MA_OWNED);
3606 	proc_set_p2_wexit(p);
3607 
3608 	p->p_acflag |= AXSIG;
3609 	/*
3610 	 * We must be single-threading to generate a core dump.  This
3611 	 * ensures that the registers in the core file are up-to-date.
3612 	 * Also, the ELF dump handler assumes that the thread list doesn't
3613 	 * change out from under it.
3614 	 *
3615 	 * XXX If another thread attempts to single-thread before us
3616 	 *     (e.g. via fork()), we won't get a dump at all.
3617 	 */
3618 	if ((sigprop(sig) & SIGPROP_CORE) &&
3619 	    thread_single(p, SINGLE_NO_EXIT) == 0) {
3620 		p->p_sig = sig;
3621 		/*
3622 		 * Log signals which would cause core dumps
3623 		 * (Log as LOG_INFO to appease those who don't want
3624 		 * these messages.)
3625 		 * XXX : Todo, as well as euid, write out ruid too
3626 		 * Note that coredump() drops proc lock.
3627 		 */
3628 		rv = coredump(td);
3629 		switch (rv) {
3630 		case 0:
3631 			sig |= WCOREFLAG;
3632 			coreinfo = " (core dumped)";
3633 			break;
3634 		case EFAULT:
3635 			coreinfo = " (no core dump - bad address)";
3636 			break;
3637 		case EINVAL:
3638 			coreinfo = " (no core dump - invalid argument)";
3639 			break;
3640 		case EFBIG:
3641 			coreinfo = " (no core dump - too large)";
3642 			break;
3643 		default:
3644 			coreinfo = " (no core dump - other error)";
3645 			break;
3646 		}
3647 		if (kern_logsigexit)
3648 			log(LOG_INFO,
3649 			    "pid %d (%s), jid %d, uid %d: exited on "
3650 			    "signal %d%s\n", p->p_pid, p->p_comm,
3651 			    p->p_ucred->cr_prison->pr_id,
3652 			    td->td_ucred->cr_uid,
3653 			    sig &~ WCOREFLAG, coreinfo);
3654 	} else
3655 		PROC_UNLOCK(p);
3656 	exit1(td, 0, sig);
3657 	/* NOTREACHED */
3658 }
3659 
3660 /*
3661  * Send queued SIGCHLD to parent when child process's state
3662  * is changed.
3663  */
3664 static void
3665 sigparent(struct proc *p, int reason, int status)
3666 {
3667 	PROC_LOCK_ASSERT(p, MA_OWNED);
3668 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3669 
3670 	if (p->p_ksi != NULL) {
3671 		p->p_ksi->ksi_signo  = SIGCHLD;
3672 		p->p_ksi->ksi_code   = reason;
3673 		p->p_ksi->ksi_status = status;
3674 		p->p_ksi->ksi_pid    = p->p_pid;
3675 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3676 		if (KSI_ONQ(p->p_ksi))
3677 			return;
3678 	}
3679 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3680 }
3681 
3682 static void
3683 childproc_jobstate(struct proc *p, int reason, int sig)
3684 {
3685 	struct sigacts *ps;
3686 
3687 	PROC_LOCK_ASSERT(p, MA_OWNED);
3688 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3689 
3690 	/*
3691 	 * Wake up parent sleeping in kern_wait(), also send
3692 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3693 	 * that parent will awake, because parent may masked
3694 	 * the signal.
3695 	 */
3696 	p->p_pptr->p_flag |= P_STATCHILD;
3697 	wakeup(p->p_pptr);
3698 
3699 	ps = p->p_pptr->p_sigacts;
3700 	mtx_lock(&ps->ps_mtx);
3701 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3702 		mtx_unlock(&ps->ps_mtx);
3703 		sigparent(p, reason, sig);
3704 	} else
3705 		mtx_unlock(&ps->ps_mtx);
3706 }
3707 
3708 void
3709 childproc_stopped(struct proc *p, int reason)
3710 {
3711 
3712 	childproc_jobstate(p, reason, p->p_xsig);
3713 }
3714 
3715 void
3716 childproc_continued(struct proc *p)
3717 {
3718 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3719 }
3720 
3721 void
3722 childproc_exited(struct proc *p)
3723 {
3724 	int reason, status;
3725 
3726 	if (WCOREDUMP(p->p_xsig)) {
3727 		reason = CLD_DUMPED;
3728 		status = WTERMSIG(p->p_xsig);
3729 	} else if (WIFSIGNALED(p->p_xsig)) {
3730 		reason = CLD_KILLED;
3731 		status = WTERMSIG(p->p_xsig);
3732 	} else {
3733 		reason = CLD_EXITED;
3734 		status = p->p_xexit;
3735 	}
3736 	/*
3737 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3738 	 * done in exit1().
3739 	 */
3740 	sigparent(p, reason, status);
3741 }
3742 
3743 #define	MAX_NUM_CORE_FILES 100000
3744 #ifndef NUM_CORE_FILES
3745 #define	NUM_CORE_FILES 5
3746 #endif
3747 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3748 static int num_cores = NUM_CORE_FILES;
3749 
3750 static int
3751 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3752 {
3753 	int error;
3754 	int new_val;
3755 
3756 	new_val = num_cores;
3757 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3758 	if (error != 0 || req->newptr == NULL)
3759 		return (error);
3760 	if (new_val > MAX_NUM_CORE_FILES)
3761 		new_val = MAX_NUM_CORE_FILES;
3762 	if (new_val < 0)
3763 		new_val = 0;
3764 	num_cores = new_val;
3765 	return (0);
3766 }
3767 SYSCTL_PROC(_debug, OID_AUTO, ncores,
3768     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
3769     sysctl_debug_num_cores_check, "I",
3770     "Maximum number of generated process corefiles while using index format");
3771 
3772 #define	GZIP_SUFFIX	".gz"
3773 #define	ZSTD_SUFFIX	".zst"
3774 
3775 int compress_user_cores = 0;
3776 
3777 static int
3778 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3779 {
3780 	int error, val;
3781 
3782 	val = compress_user_cores;
3783 	error = sysctl_handle_int(oidp, &val, 0, req);
3784 	if (error != 0 || req->newptr == NULL)
3785 		return (error);
3786 	if (val != 0 && !compressor_avail(val))
3787 		return (EINVAL);
3788 	compress_user_cores = val;
3789 	return (error);
3790 }
3791 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores,
3792     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, 0, sizeof(int),
3793     sysctl_compress_user_cores, "I",
3794     "Enable compression of user corefiles ("
3795     __XSTRING(COMPRESS_GZIP) " = gzip, "
3796     __XSTRING(COMPRESS_ZSTD) " = zstd)");
3797 
3798 int compress_user_cores_level = 6;
3799 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3800     &compress_user_cores_level, 0,
3801     "Corefile compression level");
3802 
3803 /*
3804  * Protect the access to corefilename[] by allproc_lock.
3805  */
3806 #define	corefilename_lock	allproc_lock
3807 
3808 static char corefilename[MAXPATHLEN] = {"%N.core"};
3809 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3810 
3811 static int
3812 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3813 {
3814 	int error;
3815 
3816 	sx_xlock(&corefilename_lock);
3817 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3818 	    req);
3819 	sx_xunlock(&corefilename_lock);
3820 
3821 	return (error);
3822 }
3823 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3824     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3825     "Process corefile name format string");
3826 
3827 static void
3828 vnode_close_locked(struct thread *td, struct vnode *vp)
3829 {
3830 
3831 	VOP_UNLOCK(vp);
3832 	vn_close(vp, FWRITE, td->td_ucred, td);
3833 }
3834 
3835 /*
3836  * If the core format has a %I in it, then we need to check
3837  * for existing corefiles before defining a name.
3838  * To do this we iterate over 0..ncores to find a
3839  * non-existing core file name to use. If all core files are
3840  * already used we choose the oldest one.
3841  */
3842 static int
3843 corefile_open_last(struct thread *td, char *name, int indexpos,
3844     int indexlen, int ncores, struct vnode **vpp)
3845 {
3846 	struct vnode *oldvp, *nextvp, *vp;
3847 	struct vattr vattr;
3848 	struct nameidata nd;
3849 	int error, i, flags, oflags, cmode;
3850 	char ch;
3851 	struct timespec lasttime;
3852 
3853 	nextvp = oldvp = NULL;
3854 	cmode = S_IRUSR | S_IWUSR;
3855 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3856 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3857 
3858 	for (i = 0; i < ncores; i++) {
3859 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3860 
3861 		ch = name[indexpos + indexlen];
3862 		(void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3863 		    i);
3864 		name[indexpos + indexlen] = ch;
3865 
3866 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
3867 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3868 		    NULL);
3869 		if (error != 0)
3870 			break;
3871 
3872 		vp = nd.ni_vp;
3873 		NDFREE_PNBUF(&nd);
3874 		if ((flags & O_CREAT) == O_CREAT) {
3875 			nextvp = vp;
3876 			break;
3877 		}
3878 
3879 		error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3880 		if (error != 0) {
3881 			vnode_close_locked(td, vp);
3882 			break;
3883 		}
3884 
3885 		if (oldvp == NULL ||
3886 		    lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3887 		    (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3888 		    lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3889 			if (oldvp != NULL)
3890 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3891 			oldvp = vp;
3892 			VOP_UNLOCK(oldvp);
3893 			lasttime = vattr.va_mtime;
3894 		} else {
3895 			vnode_close_locked(td, vp);
3896 		}
3897 	}
3898 
3899 	if (oldvp != NULL) {
3900 		if (nextvp == NULL) {
3901 			if ((td->td_proc->p_flag & P_SUGID) != 0) {
3902 				error = EFAULT;
3903 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3904 			} else {
3905 				nextvp = oldvp;
3906 				error = vn_lock(nextvp, LK_EXCLUSIVE);
3907 				if (error != 0) {
3908 					vn_close(nextvp, FWRITE, td->td_ucred,
3909 					    td);
3910 					nextvp = NULL;
3911 				}
3912 			}
3913 		} else {
3914 			vn_close(oldvp, FWRITE, td->td_ucred, td);
3915 		}
3916 	}
3917 	if (error != 0) {
3918 		if (nextvp != NULL)
3919 			vnode_close_locked(td, oldvp);
3920 	} else {
3921 		*vpp = nextvp;
3922 	}
3923 
3924 	return (error);
3925 }
3926 
3927 /*
3928  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3929  * Expand the name described in corefilename, using name, uid, and pid
3930  * and open/create core file.
3931  * corefilename is a printf-like string, with three format specifiers:
3932  *	%N	name of process ("name")
3933  *	%P	process id (pid)
3934  *	%U	user id (uid)
3935  * For example, "%N.core" is the default; they can be disabled completely
3936  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3937  * This is controlled by the sysctl variable kern.corefile (see above).
3938  */
3939 static int
3940 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3941     int compress, int signum, struct vnode **vpp, char **namep)
3942 {
3943 	struct sbuf sb;
3944 	struct nameidata nd;
3945 	const char *format;
3946 	char *hostname, *name;
3947 	int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
3948 
3949 	hostname = NULL;
3950 	format = corefilename;
3951 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3952 	indexlen = 0;
3953 	indexpos = -1;
3954 	ncores = num_cores;
3955 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3956 	sx_slock(&corefilename_lock);
3957 	for (i = 0; format[i] != '\0'; i++) {
3958 		switch (format[i]) {
3959 		case '%':	/* Format character */
3960 			i++;
3961 			switch (format[i]) {
3962 			case '%':
3963 				sbuf_putc(&sb, '%');
3964 				break;
3965 			case 'H':	/* hostname */
3966 				if (hostname == NULL) {
3967 					hostname = malloc(MAXHOSTNAMELEN,
3968 					    M_TEMP, M_WAITOK);
3969 				}
3970 				getcredhostname(td->td_ucred, hostname,
3971 				    MAXHOSTNAMELEN);
3972 				sbuf_cat(&sb, hostname);
3973 				break;
3974 			case 'I':	/* autoincrementing index */
3975 				if (indexpos != -1) {
3976 					sbuf_printf(&sb, "%%I");
3977 					break;
3978 				}
3979 
3980 				indexpos = sbuf_len(&sb);
3981 				sbuf_printf(&sb, "%u", ncores - 1);
3982 				indexlen = sbuf_len(&sb) - indexpos;
3983 				break;
3984 			case 'N':	/* process name */
3985 				sbuf_printf(&sb, "%s", comm);
3986 				break;
3987 			case 'P':	/* process id */
3988 				sbuf_printf(&sb, "%u", pid);
3989 				break;
3990 			case 'S':	/* signal number */
3991 				sbuf_printf(&sb, "%i", signum);
3992 				break;
3993 			case 'U':	/* user id */
3994 				sbuf_printf(&sb, "%u", uid);
3995 				break;
3996 			default:
3997 				log(LOG_ERR,
3998 				    "Unknown format character %c in "
3999 				    "corename `%s'\n", format[i], format);
4000 				break;
4001 			}
4002 			break;
4003 		default:
4004 			sbuf_putc(&sb, format[i]);
4005 			break;
4006 		}
4007 	}
4008 	sx_sunlock(&corefilename_lock);
4009 	free(hostname, M_TEMP);
4010 	if (compress == COMPRESS_GZIP)
4011 		sbuf_cat(&sb, GZIP_SUFFIX);
4012 	else if (compress == COMPRESS_ZSTD)
4013 		sbuf_cat(&sb, ZSTD_SUFFIX);
4014 	if (sbuf_error(&sb) != 0) {
4015 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
4016 		    "long\n", (long)pid, comm, (u_long)uid);
4017 		sbuf_delete(&sb);
4018 		free(name, M_TEMP);
4019 		return (ENOMEM);
4020 	}
4021 	sbuf_finish(&sb);
4022 	sbuf_delete(&sb);
4023 
4024 	if (indexpos != -1) {
4025 		error = corefile_open_last(td, name, indexpos, indexlen, ncores,
4026 		    vpp);
4027 		if (error != 0) {
4028 			log(LOG_ERR,
4029 			    "pid %d (%s), uid (%u):  Path `%s' failed "
4030 			    "on initial open test, error = %d\n",
4031 			    pid, comm, uid, name, error);
4032 		}
4033 	} else {
4034 		cmode = S_IRUSR | S_IWUSR;
4035 		oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
4036 		    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
4037 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
4038 		if ((td->td_proc->p_flag & P_SUGID) != 0)
4039 			flags |= O_EXCL;
4040 
4041 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
4042 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
4043 		    NULL);
4044 		if (error == 0) {
4045 			*vpp = nd.ni_vp;
4046 			NDFREE_PNBUF(&nd);
4047 		}
4048 	}
4049 
4050 	if (error != 0) {
4051 #ifdef AUDIT
4052 		audit_proc_coredump(td, name, error);
4053 #endif
4054 		free(name, M_TEMP);
4055 		return (error);
4056 	}
4057 	*namep = name;
4058 	return (0);
4059 }
4060 
4061 /*
4062  * Dump a process' core.  The main routine does some
4063  * policy checking, and creates the name of the coredump;
4064  * then it passes on a vnode and a size limit to the process-specific
4065  * coredump routine if there is one; if there _is not_ one, it returns
4066  * ENOSYS; otherwise it returns the error from the process-specific routine.
4067  */
4068 
4069 static int
4070 coredump(struct thread *td)
4071 {
4072 	struct proc *p = td->td_proc;
4073 	struct ucred *cred = td->td_ucred;
4074 	struct vnode *vp;
4075 	struct flock lf;
4076 	struct vattr vattr;
4077 	size_t fullpathsize;
4078 	int error, error1, locked;
4079 	char *name;			/* name of corefile */
4080 	void *rl_cookie;
4081 	off_t limit;
4082 	char *fullpath, *freepath = NULL;
4083 	struct sbuf *sb;
4084 
4085 	PROC_LOCK_ASSERT(p, MA_OWNED);
4086 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
4087 
4088 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
4089 	    (p->p_flag2 & P2_NOTRACE) != 0) {
4090 		PROC_UNLOCK(p);
4091 		return (EFAULT);
4092 	}
4093 
4094 	/*
4095 	 * Note that the bulk of limit checking is done after
4096 	 * the corefile is created.  The exception is if the limit
4097 	 * for corefiles is 0, in which case we don't bother
4098 	 * creating the corefile at all.  This layout means that
4099 	 * a corefile is truncated instead of not being created,
4100 	 * if it is larger than the limit.
4101 	 */
4102 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
4103 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
4104 		PROC_UNLOCK(p);
4105 		return (EFBIG);
4106 	}
4107 	PROC_UNLOCK(p);
4108 
4109 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
4110 	    compress_user_cores, p->p_sig, &vp, &name);
4111 	if (error != 0)
4112 		return (error);
4113 
4114 	/*
4115 	 * Don't dump to non-regular files or files with links.
4116 	 * Do not dump into system files. Effective user must own the corefile.
4117 	 */
4118 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
4119 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
4120 	    vattr.va_uid != cred->cr_uid) {
4121 		VOP_UNLOCK(vp);
4122 		error = EFAULT;
4123 		goto out;
4124 	}
4125 
4126 	VOP_UNLOCK(vp);
4127 
4128 	/* Postpone other writers, including core dumps of other processes. */
4129 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
4130 
4131 	lf.l_whence = SEEK_SET;
4132 	lf.l_start = 0;
4133 	lf.l_len = 0;
4134 	lf.l_type = F_WRLCK;
4135 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
4136 
4137 	VATTR_NULL(&vattr);
4138 	vattr.va_size = 0;
4139 	if (set_core_nodump_flag)
4140 		vattr.va_flags = UF_NODUMP;
4141 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4142 	VOP_SETATTR(vp, &vattr, cred);
4143 	VOP_UNLOCK(vp);
4144 	PROC_LOCK(p);
4145 	p->p_acflag |= ACORE;
4146 	PROC_UNLOCK(p);
4147 
4148 	if (p->p_sysent->sv_coredump != NULL) {
4149 		error = p->p_sysent->sv_coredump(td, vp, limit, 0);
4150 	} else {
4151 		error = ENOSYS;
4152 	}
4153 
4154 	if (locked) {
4155 		lf.l_type = F_UNLCK;
4156 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
4157 	}
4158 	vn_rangelock_unlock(vp, rl_cookie);
4159 
4160 	/*
4161 	 * Notify the userland helper that a process triggered a core dump.
4162 	 * This allows the helper to run an automated debugging session.
4163 	 */
4164 	if (error != 0 || coredump_devctl == 0)
4165 		goto out;
4166 	sb = sbuf_new_auto();
4167 	if (vn_fullpath_global(p->p_textvp, &fullpath, &freepath) != 0)
4168 		goto out2;
4169 	sbuf_cat(sb, "comm=\"");
4170 	devctl_safe_quote_sb(sb, fullpath);
4171 	free(freepath, M_TEMP);
4172 	sbuf_cat(sb, "\" core=\"");
4173 
4174 	/*
4175 	 * We can't lookup core file vp directly. When we're replacing a core, and
4176 	 * other random times, we flush the name cache, so it will fail. Instead,
4177 	 * if the path of the core is relative, add the current dir in front if it.
4178 	 */
4179 	if (name[0] != '/') {
4180 		fullpathsize = MAXPATHLEN;
4181 		freepath = malloc(fullpathsize, M_TEMP, M_WAITOK);
4182 		if (vn_getcwd(freepath, &fullpath, &fullpathsize) != 0) {
4183 			free(freepath, M_TEMP);
4184 			goto out2;
4185 		}
4186 		devctl_safe_quote_sb(sb, fullpath);
4187 		free(freepath, M_TEMP);
4188 		sbuf_putc(sb, '/');
4189 	}
4190 	devctl_safe_quote_sb(sb, name);
4191 	sbuf_putc(sb, '"');
4192 	if (sbuf_finish(sb) == 0)
4193 		devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
4194 out2:
4195 	sbuf_delete(sb);
4196 out:
4197 	error1 = vn_close(vp, FWRITE, cred, td);
4198 	if (error == 0)
4199 		error = error1;
4200 #ifdef AUDIT
4201 	audit_proc_coredump(td, name, error);
4202 #endif
4203 	free(name, M_TEMP);
4204 	return (error);
4205 }
4206 
4207 /*
4208  * Nonexistent system call-- signal process (may want to handle it).  Flag
4209  * error in case process won't see signal immediately (blocked or ignored).
4210  */
4211 #ifndef _SYS_SYSPROTO_H_
4212 struct nosys_args {
4213 	int	dummy;
4214 };
4215 #endif
4216 /* ARGSUSED */
4217 int
4218 nosys(struct thread *td, struct nosys_args *args)
4219 {
4220 	struct proc *p;
4221 
4222 	p = td->td_proc;
4223 
4224 	if (SV_PROC_FLAG(p, SV_SIGSYS) != 0 && kern_signosys) {
4225 		PROC_LOCK(p);
4226 		tdsignal(td, SIGSYS);
4227 		PROC_UNLOCK(p);
4228 	}
4229 	if (kern_lognosys == 1 || kern_lognosys == 3) {
4230 		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4231 		    td->td_sa.code);
4232 	}
4233 	if (kern_lognosys == 2 || kern_lognosys == 3 ||
4234 	    (p->p_pid == 1 && (kern_lognosys & 3) == 0)) {
4235 		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4236 		    td->td_sa.code);
4237 	}
4238 	return (ENOSYS);
4239 }
4240 
4241 /*
4242  * Send a SIGIO or SIGURG signal to a process or process group using stored
4243  * credentials rather than those of the current process.
4244  */
4245 void
4246 pgsigio(struct sigio **sigiop, int sig, int checkctty)
4247 {
4248 	ksiginfo_t ksi;
4249 	struct sigio *sigio;
4250 
4251 	ksiginfo_init(&ksi);
4252 	ksi.ksi_signo = sig;
4253 	ksi.ksi_code = SI_KERNEL;
4254 
4255 	SIGIO_LOCK();
4256 	sigio = *sigiop;
4257 	if (sigio == NULL) {
4258 		SIGIO_UNLOCK();
4259 		return;
4260 	}
4261 	if (sigio->sio_pgid > 0) {
4262 		PROC_LOCK(sigio->sio_proc);
4263 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
4264 			kern_psignal(sigio->sio_proc, sig);
4265 		PROC_UNLOCK(sigio->sio_proc);
4266 	} else if (sigio->sio_pgid < 0) {
4267 		struct proc *p;
4268 
4269 		PGRP_LOCK(sigio->sio_pgrp);
4270 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
4271 			PROC_LOCK(p);
4272 			if (p->p_state == PRS_NORMAL &&
4273 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
4274 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
4275 				kern_psignal(p, sig);
4276 			PROC_UNLOCK(p);
4277 		}
4278 		PGRP_UNLOCK(sigio->sio_pgrp);
4279 	}
4280 	SIGIO_UNLOCK();
4281 }
4282 
4283 static int
4284 filt_sigattach(struct knote *kn)
4285 {
4286 	struct proc *p = curproc;
4287 
4288 	kn->kn_ptr.p_proc = p;
4289 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
4290 
4291 	knlist_add(p->p_klist, kn, 0);
4292 
4293 	return (0);
4294 }
4295 
4296 static void
4297 filt_sigdetach(struct knote *kn)
4298 {
4299 	knlist_remove(kn->kn_knlist, kn, 0);
4300 }
4301 
4302 /*
4303  * signal knotes are shared with proc knotes, so we apply a mask to
4304  * the hint in order to differentiate them from process hints.  This
4305  * could be avoided by using a signal-specific knote list, but probably
4306  * isn't worth the trouble.
4307  */
4308 static int
4309 filt_signal(struct knote *kn, long hint)
4310 {
4311 
4312 	if (hint & NOTE_SIGNAL) {
4313 		hint &= ~NOTE_SIGNAL;
4314 
4315 		if (kn->kn_id == hint)
4316 			kn->kn_data++;
4317 	}
4318 	return (kn->kn_data != 0);
4319 }
4320 
4321 struct sigacts *
4322 sigacts_alloc(void)
4323 {
4324 	struct sigacts *ps;
4325 
4326 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
4327 	refcount_init(&ps->ps_refcnt, 1);
4328 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
4329 	return (ps);
4330 }
4331 
4332 void
4333 sigacts_free(struct sigacts *ps)
4334 {
4335 
4336 	if (refcount_release(&ps->ps_refcnt) == 0)
4337 		return;
4338 	mtx_destroy(&ps->ps_mtx);
4339 	free(ps, M_SUBPROC);
4340 }
4341 
4342 struct sigacts *
4343 sigacts_hold(struct sigacts *ps)
4344 {
4345 
4346 	refcount_acquire(&ps->ps_refcnt);
4347 	return (ps);
4348 }
4349 
4350 void
4351 sigacts_copy(struct sigacts *dest, struct sigacts *src)
4352 {
4353 
4354 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
4355 	mtx_lock(&src->ps_mtx);
4356 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
4357 	mtx_unlock(&src->ps_mtx);
4358 }
4359 
4360 int
4361 sigacts_shared(struct sigacts *ps)
4362 {
4363 
4364 	return (ps->ps_refcnt > 1);
4365 }
4366 
4367 void
4368 sig_drop_caught(struct proc *p)
4369 {
4370 	int sig;
4371 	struct sigacts *ps;
4372 
4373 	ps = p->p_sigacts;
4374 	PROC_LOCK_ASSERT(p, MA_OWNED);
4375 	mtx_assert(&ps->ps_mtx, MA_OWNED);
4376 	SIG_FOREACH(sig, &ps->ps_sigcatch) {
4377 		sigdflt(ps, sig);
4378 		if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
4379 			sigqueue_delete_proc(p, sig);
4380 	}
4381 }
4382 
4383 static void
4384 sigfastblock_failed(struct thread *td, bool sendsig, bool write)
4385 {
4386 	ksiginfo_t ksi;
4387 
4388 	/*
4389 	 * Prevent further fetches and SIGSEGVs, allowing thread to
4390 	 * issue syscalls despite corruption.
4391 	 */
4392 	sigfastblock_clear(td);
4393 
4394 	if (!sendsig)
4395 		return;
4396 	ksiginfo_init_trap(&ksi);
4397 	ksi.ksi_signo = SIGSEGV;
4398 	ksi.ksi_code = write ? SEGV_ACCERR : SEGV_MAPERR;
4399 	ksi.ksi_addr = td->td_sigblock_ptr;
4400 	trapsignal(td, &ksi);
4401 }
4402 
4403 static bool
4404 sigfastblock_fetch_sig(struct thread *td, bool sendsig, uint32_t *valp)
4405 {
4406 	uint32_t res;
4407 
4408 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4409 		return (true);
4410 	if (fueword32((void *)td->td_sigblock_ptr, &res) == -1) {
4411 		sigfastblock_failed(td, sendsig, false);
4412 		return (false);
4413 	}
4414 	*valp = res;
4415 	td->td_sigblock_val = res & ~SIGFASTBLOCK_FLAGS;
4416 	return (true);
4417 }
4418 
4419 static void
4420 sigfastblock_resched(struct thread *td, bool resched)
4421 {
4422 	struct proc *p;
4423 
4424 	if (resched) {
4425 		p = td->td_proc;
4426 		PROC_LOCK(p);
4427 		reschedule_signals(p, td->td_sigmask, 0);
4428 		PROC_UNLOCK(p);
4429 	}
4430 	ast_sched(td, TDA_SIG);
4431 }
4432 
4433 int
4434 sys_sigfastblock(struct thread *td, struct sigfastblock_args *uap)
4435 {
4436 	struct proc *p;
4437 	int error, res;
4438 	uint32_t oldval;
4439 
4440 	error = 0;
4441 	p = td->td_proc;
4442 	switch (uap->cmd) {
4443 	case SIGFASTBLOCK_SETPTR:
4444 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
4445 			error = EBUSY;
4446 			break;
4447 		}
4448 		if (((uintptr_t)(uap->ptr) & (sizeof(uint32_t) - 1)) != 0) {
4449 			error = EINVAL;
4450 			break;
4451 		}
4452 		td->td_pflags |= TDP_SIGFASTBLOCK;
4453 		td->td_sigblock_ptr = uap->ptr;
4454 		break;
4455 
4456 	case SIGFASTBLOCK_UNBLOCK:
4457 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4458 			error = EINVAL;
4459 			break;
4460 		}
4461 
4462 		for (;;) {
4463 			res = casueword32(td->td_sigblock_ptr,
4464 			    SIGFASTBLOCK_PEND, &oldval, 0);
4465 			if (res == -1) {
4466 				error = EFAULT;
4467 				sigfastblock_failed(td, false, true);
4468 				break;
4469 			}
4470 			if (res == 0)
4471 				break;
4472 			MPASS(res == 1);
4473 			if (oldval != SIGFASTBLOCK_PEND) {
4474 				error = EBUSY;
4475 				break;
4476 			}
4477 			error = thread_check_susp(td, false);
4478 			if (error != 0)
4479 				break;
4480 		}
4481 		if (error != 0)
4482 			break;
4483 
4484 		/*
4485 		 * td_sigblock_val is cleared there, but not on a
4486 		 * syscall exit.  The end effect is that a single
4487 		 * interruptible sleep, while user sigblock word is
4488 		 * set, might return EINTR or ERESTART to usermode
4489 		 * without delivering signal.  All further sleeps,
4490 		 * until userspace clears the word and does
4491 		 * sigfastblock(UNBLOCK), observe current word and no
4492 		 * longer get interrupted.  It is slight
4493 		 * non-conformance, with alternative to have read the
4494 		 * sigblock word on each syscall entry.
4495 		 */
4496 		td->td_sigblock_val = 0;
4497 
4498 		/*
4499 		 * Rely on normal ast mechanism to deliver pending
4500 		 * signals to current thread.  But notify others about
4501 		 * fake unblock.
4502 		 */
4503 		sigfastblock_resched(td, error == 0 && p->p_numthreads != 1);
4504 
4505 		break;
4506 
4507 	case SIGFASTBLOCK_UNSETPTR:
4508 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4509 			error = EINVAL;
4510 			break;
4511 		}
4512 		if (!sigfastblock_fetch_sig(td, false, &oldval)) {
4513 			error = EFAULT;
4514 			break;
4515 		}
4516 		if (oldval != 0 && oldval != SIGFASTBLOCK_PEND) {
4517 			error = EBUSY;
4518 			break;
4519 		}
4520 		sigfastblock_clear(td);
4521 		break;
4522 
4523 	default:
4524 		error = EINVAL;
4525 		break;
4526 	}
4527 	return (error);
4528 }
4529 
4530 void
4531 sigfastblock_clear(struct thread *td)
4532 {
4533 	bool resched;
4534 
4535 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4536 		return;
4537 	td->td_sigblock_val = 0;
4538 	resched = (td->td_pflags & TDP_SIGFASTPENDING) != 0 ||
4539 	    SIGPENDING(td);
4540 	td->td_pflags &= ~(TDP_SIGFASTBLOCK | TDP_SIGFASTPENDING);
4541 	sigfastblock_resched(td, resched);
4542 }
4543 
4544 void
4545 sigfastblock_fetch(struct thread *td)
4546 {
4547 	uint32_t val;
4548 
4549 	(void)sigfastblock_fetch_sig(td, true, &val);
4550 }
4551 
4552 static void
4553 sigfastblock_setpend1(struct thread *td)
4554 {
4555 	int res;
4556 	uint32_t oldval;
4557 
4558 	if ((td->td_pflags & TDP_SIGFASTPENDING) == 0)
4559 		return;
4560 	res = fueword32((void *)td->td_sigblock_ptr, &oldval);
4561 	if (res == -1) {
4562 		sigfastblock_failed(td, true, false);
4563 		return;
4564 	}
4565 	for (;;) {
4566 		res = casueword32(td->td_sigblock_ptr, oldval, &oldval,
4567 		    oldval | SIGFASTBLOCK_PEND);
4568 		if (res == -1) {
4569 			sigfastblock_failed(td, true, true);
4570 			return;
4571 		}
4572 		if (res == 0) {
4573 			td->td_sigblock_val = oldval & ~SIGFASTBLOCK_FLAGS;
4574 			td->td_pflags &= ~TDP_SIGFASTPENDING;
4575 			break;
4576 		}
4577 		MPASS(res == 1);
4578 		if (thread_check_susp(td, false) != 0)
4579 			break;
4580 	}
4581 }
4582 
4583 static void
4584 sigfastblock_setpend(struct thread *td, bool resched)
4585 {
4586 	struct proc *p;
4587 
4588 	sigfastblock_setpend1(td);
4589 	if (resched) {
4590 		p = td->td_proc;
4591 		PROC_LOCK(p);
4592 		reschedule_signals(p, fastblock_mask, SIGPROCMASK_FASTBLK);
4593 		PROC_UNLOCK(p);
4594 	}
4595 }
4596