xref: /freebsd/sys/kern/kern_sig.c (revision 4ef1c6f75d25ba0c264ab274a1b449467a6ee817)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "opt_capsicum.h"
43 #include "opt_ktrace.h"
44 
45 #include <sys/param.h>
46 #include <sys/capsicum.h>
47 #include <sys/ctype.h>
48 #include <sys/systm.h>
49 #include <sys/signalvar.h>
50 #include <sys/vnode.h>
51 #include <sys/acct.h>
52 #include <sys/capsicum.h>
53 #include <sys/compressor.h>
54 #include <sys/condvar.h>
55 #include <sys/devctl.h>
56 #include <sys/event.h>
57 #include <sys/fcntl.h>
58 #include <sys/imgact.h>
59 #include <sys/jail.h>
60 #include <sys/kernel.h>
61 #include <sys/ktr.h>
62 #include <sys/ktrace.h>
63 #include <sys/limits.h>
64 #include <sys/lock.h>
65 #include <sys/malloc.h>
66 #include <sys/mutex.h>
67 #include <sys/refcount.h>
68 #include <sys/namei.h>
69 #include <sys/proc.h>
70 #include <sys/procdesc.h>
71 #include <sys/ptrace.h>
72 #include <sys/posix4.h>
73 #include <sys/racct.h>
74 #include <sys/resourcevar.h>
75 #include <sys/sdt.h>
76 #include <sys/sbuf.h>
77 #include <sys/sleepqueue.h>
78 #include <sys/smp.h>
79 #include <sys/stat.h>
80 #include <sys/sx.h>
81 #include <sys/syscall.h>
82 #include <sys/syscallsubr.h>
83 #include <sys/sysctl.h>
84 #include <sys/sysent.h>
85 #include <sys/syslog.h>
86 #include <sys/sysproto.h>
87 #include <sys/timers.h>
88 #include <sys/unistd.h>
89 #include <sys/vmmeter.h>
90 #include <sys/wait.h>
91 #include <vm/vm.h>
92 #include <vm/vm_extern.h>
93 #include <vm/uma.h>
94 
95 #include <machine/cpu.h>
96 
97 #include <security/audit/audit.h>
98 
99 #define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
100 
101 SDT_PROVIDER_DECLARE(proc);
102 SDT_PROBE_DEFINE3(proc, , , signal__send,
103     "struct thread *", "struct proc *", "int");
104 SDT_PROBE_DEFINE2(proc, , , signal__clear,
105     "int", "ksiginfo_t *");
106 SDT_PROBE_DEFINE3(proc, , , signal__discard,
107     "struct thread *", "struct proc *", "int");
108 
109 static int	coredump(struct thread *);
110 static int	killpg1(struct thread *td, int sig, int pgid, int all,
111 		    ksiginfo_t *ksi);
112 static int	issignal(struct thread *td);
113 static void	reschedule_signals(struct proc *p, sigset_t block, int flags);
114 static int	sigprop(int sig);
115 static void	tdsigwakeup(struct thread *, int, sig_t, int);
116 static int	sig_suspend_threads(struct thread *, struct proc *);
117 static int	filt_sigattach(struct knote *kn);
118 static void	filt_sigdetach(struct knote *kn);
119 static int	filt_signal(struct knote *kn, long hint);
120 static struct thread *sigtd(struct proc *p, int sig, bool fast_sigblock);
121 static void	sigqueue_start(void);
122 static void	sigfastblock_setpend(struct thread *td, bool resched);
123 static void	sigexit1(struct thread *td, int sig, ksiginfo_t *ksi) __dead2;
124 
125 static uma_zone_t	ksiginfo_zone = NULL;
126 struct filterops sig_filtops = {
127 	.f_isfd = 0,
128 	.f_attach = filt_sigattach,
129 	.f_detach = filt_sigdetach,
130 	.f_event = filt_signal,
131 };
132 
133 static int	kern_logsigexit = 1;
134 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
135     &kern_logsigexit, 0,
136     "Log processes quitting on abnormal signals to syslog(3)");
137 
138 static int	kern_forcesigexit = 1;
139 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
140     &kern_forcesigexit, 0, "Force trap signal to be handled");
141 
142 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
143     "POSIX real time signal");
144 
145 static int	max_pending_per_proc = 128;
146 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
147     &max_pending_per_proc, 0, "Max pending signals per proc");
148 
149 static int	preallocate_siginfo = 1024;
150 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
151     &preallocate_siginfo, 0, "Preallocated signal memory size");
152 
153 static int	signal_overflow = 0;
154 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
155     &signal_overflow, 0, "Number of signals overflew");
156 
157 static int	signal_alloc_fail = 0;
158 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
159     &signal_alloc_fail, 0, "signals failed to be allocated");
160 
161 static int	kern_lognosys = 0;
162 SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
163     "Log invalid syscalls");
164 
165 __read_frequently bool sigfastblock_fetch_always = false;
166 SYSCTL_BOOL(_kern, OID_AUTO, sigfastblock_fetch_always, CTLFLAG_RWTUN,
167     &sigfastblock_fetch_always, 0,
168     "Fetch sigfastblock word on each syscall entry for proper "
169     "blocking semantic");
170 
171 static bool	kern_sig_discard_ign = true;
172 SYSCTL_BOOL(_kern, OID_AUTO, sig_discard_ign, CTLFLAG_RWTUN,
173     &kern_sig_discard_ign, 0,
174     "Discard ignored signals on delivery, otherwise queue them to "
175     "the target queue");
176 
177 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
178 
179 /*
180  * Policy -- Can ucred cr1 send SIGIO to process cr2?
181  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
182  * in the right situations.
183  */
184 #define CANSIGIO(cr1, cr2) \
185 	((cr1)->cr_uid == 0 || \
186 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
187 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
188 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
189 	    (cr1)->cr_uid == (cr2)->cr_uid)
190 
191 static int	sugid_coredump;
192 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
193     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
194 
195 static int	capmode_coredump;
196 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
197     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
198 
199 static int	do_coredump = 1;
200 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
201 	&do_coredump, 0, "Enable/Disable coredumps");
202 
203 static int	set_core_nodump_flag = 0;
204 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
205 	0, "Enable setting the NODUMP flag on coredump files");
206 
207 static int	coredump_devctl = 0;
208 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
209 	0, "Generate a devctl notification when processes coredump");
210 
211 /*
212  * Signal properties and actions.
213  * The array below categorizes the signals and their default actions
214  * according to the following properties:
215  */
216 #define	SIGPROP_KILL		0x01	/* terminates process by default */
217 #define	SIGPROP_CORE		0x02	/* ditto and coredumps */
218 #define	SIGPROP_STOP		0x04	/* suspend process */
219 #define	SIGPROP_TTYSTOP		0x08	/* ditto, from tty */
220 #define	SIGPROP_IGNORE		0x10	/* ignore by default */
221 #define	SIGPROP_CONT		0x20	/* continue if suspended */
222 
223 static const int sigproptbl[NSIG] = {
224 	[SIGHUP] =	SIGPROP_KILL,
225 	[SIGINT] =	SIGPROP_KILL,
226 	[SIGQUIT] =	SIGPROP_KILL | SIGPROP_CORE,
227 	[SIGILL] =	SIGPROP_KILL | SIGPROP_CORE,
228 	[SIGTRAP] =	SIGPROP_KILL | SIGPROP_CORE,
229 	[SIGABRT] =	SIGPROP_KILL | SIGPROP_CORE,
230 	[SIGEMT] =	SIGPROP_KILL | SIGPROP_CORE,
231 	[SIGFPE] =	SIGPROP_KILL | SIGPROP_CORE,
232 	[SIGKILL] =	SIGPROP_KILL,
233 	[SIGBUS] =	SIGPROP_KILL | SIGPROP_CORE,
234 	[SIGSEGV] =	SIGPROP_KILL | SIGPROP_CORE,
235 	[SIGSYS] =	SIGPROP_KILL | SIGPROP_CORE,
236 	[SIGPIPE] =	SIGPROP_KILL,
237 	[SIGALRM] =	SIGPROP_KILL,
238 	[SIGTERM] =	SIGPROP_KILL,
239 	[SIGURG] =	SIGPROP_IGNORE,
240 	[SIGSTOP] =	SIGPROP_STOP,
241 	[SIGTSTP] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
242 	[SIGCONT] =	SIGPROP_IGNORE | SIGPROP_CONT,
243 	[SIGCHLD] =	SIGPROP_IGNORE,
244 	[SIGTTIN] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
245 	[SIGTTOU] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
246 	[SIGIO] =	SIGPROP_IGNORE,
247 	[SIGXCPU] =	SIGPROP_KILL,
248 	[SIGXFSZ] =	SIGPROP_KILL,
249 	[SIGVTALRM] =	SIGPROP_KILL,
250 	[SIGPROF] =	SIGPROP_KILL,
251 	[SIGWINCH] =	SIGPROP_IGNORE,
252 	[SIGINFO] =	SIGPROP_IGNORE,
253 	[SIGUSR1] =	SIGPROP_KILL,
254 	[SIGUSR2] =	SIGPROP_KILL,
255 };
256 
257 #define	_SIG_FOREACH_ADVANCE(i, set) ({					\
258 	int __found;							\
259 	for (;;) {							\
260 		if (__bits != 0) {					\
261 			int __sig = ffs(__bits);			\
262 			__bits &= ~(1u << (__sig - 1));			\
263 			sig = __i * sizeof((set)->__bits[0]) * NBBY + __sig; \
264 			__found = 1;					\
265 			break;						\
266 		}							\
267 		if (++__i == _SIG_WORDS) {				\
268 			__found = 0;					\
269 			break;						\
270 		}							\
271 		__bits = (set)->__bits[__i];				\
272 	}								\
273 	__found != 0;							\
274 })
275 
276 #define	SIG_FOREACH(i, set)						\
277 	for (int32_t __i = -1, __bits = 0;				\
278 	    _SIG_FOREACH_ADVANCE(i, set); )				\
279 
280 static sigset_t fastblock_mask;
281 
282 static void
283 ast_sig(struct thread *td, int tda)
284 {
285 	struct proc *p;
286 	int old_boundary, sig;
287 	bool resched_sigs;
288 
289 	p = td->td_proc;
290 
291 #ifdef DIAGNOSTIC
292 	if (p->p_numthreads == 1 && (tda & (TDAI(TDA_SIG) |
293 	    TDAI(TDA_AST))) == 0) {
294 		PROC_LOCK(p);
295 		thread_lock(td);
296 		/*
297 		 * Note that TDA_SIG should be re-read from
298 		 * td_ast, since signal might have been delivered
299 		 * after we cleared td_flags above.  This is one of
300 		 * the reason for looping check for AST condition.
301 		 * See comment in userret() about P_PPWAIT.
302 		 */
303 		if ((p->p_flag & P_PPWAIT) == 0 &&
304 		    (td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
305 			if (SIGPENDING(td) && ((tda | td->td_ast) &
306 			    (TDAI(TDA_SIG) | TDAI(TDA_AST))) == 0) {
307 				thread_unlock(td); /* fix dumps */
308 				panic(
309 				    "failed2 to set signal flags for ast p %p "
310 				    "td %p tda %#x td_ast %#x fl %#x",
311 				    p, td, tda, td->td_ast, td->td_flags);
312 			}
313 		}
314 		thread_unlock(td);
315 		PROC_UNLOCK(p);
316 	}
317 #endif
318 
319 	/*
320 	 * Check for signals. Unlocked reads of p_pendingcnt or
321 	 * p_siglist might cause process-directed signal to be handled
322 	 * later.
323 	 */
324 	if ((tda & TDAI(TDA_SIG)) != 0 || p->p_pendingcnt > 0 ||
325 	    !SIGISEMPTY(p->p_siglist)) {
326 		sigfastblock_fetch(td);
327 		PROC_LOCK(p);
328 		old_boundary = ~TDB_BOUNDARY | (td->td_dbgflags & TDB_BOUNDARY);
329 		td->td_dbgflags |= TDB_BOUNDARY;
330 		mtx_lock(&p->p_sigacts->ps_mtx);
331 		while ((sig = cursig(td)) != 0) {
332 			KASSERT(sig >= 0, ("sig %d", sig));
333 			postsig(sig);
334 		}
335 		mtx_unlock(&p->p_sigacts->ps_mtx);
336 		td->td_dbgflags &= old_boundary;
337 		PROC_UNLOCK(p);
338 		resched_sigs = true;
339 	} else {
340 		resched_sigs = false;
341 	}
342 
343 	/*
344 	 * Handle deferred update of the fast sigblock value, after
345 	 * the postsig() loop was performed.
346 	 */
347 	sigfastblock_setpend(td, resched_sigs);
348 }
349 
350 static void
351 ast_sigsuspend(struct thread *td, int tda __unused)
352 {
353 	MPASS((td->td_pflags & TDP_OLDMASK) != 0);
354 	td->td_pflags &= ~TDP_OLDMASK;
355 	kern_sigprocmask(td, SIG_SETMASK, &td->td_oldsigmask, NULL, 0);
356 }
357 
358 static void
359 sigqueue_start(void)
360 {
361 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
362 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
363 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
364 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
365 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
366 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
367 	SIGFILLSET(fastblock_mask);
368 	SIG_CANTMASK(fastblock_mask);
369 	ast_register(TDA_SIG, ASTR_UNCOND, 0, ast_sig);
370 	ast_register(TDA_SIGSUSPEND, ASTR_ASTF_REQUIRED | ASTR_TDP,
371 	    TDP_OLDMASK, ast_sigsuspend);
372 }
373 
374 static void
375 sig_handle_killpg(struct proc *p, ksiginfo_t *ksi)
376 {
377 	if ((ksi->ksi_flags & KSI_KILLPG) != 0 && p != NULL) {
378 		MPASS(atomic_load_int(&p->p_killpg_cnt) > 0);
379 		atomic_add_int(&p->p_killpg_cnt, -1);
380 	}
381 }
382 
383 ksiginfo_t *
384 ksiginfo_alloc(int mwait)
385 {
386 	MPASS(mwait == M_WAITOK || mwait == M_NOWAIT);
387 
388 	if (ksiginfo_zone == NULL)
389 		return (NULL);
390 	return (uma_zalloc(ksiginfo_zone, mwait | M_ZERO));
391 }
392 
393 void
394 ksiginfo_free(ksiginfo_t *ksi)
395 {
396 	uma_zfree(ksiginfo_zone, ksi);
397 }
398 
399 static __inline bool
400 ksiginfo_tryfree(ksiginfo_t *ksi)
401 {
402 	if ((ksi->ksi_flags & KSI_EXT) == 0) {
403 		uma_zfree(ksiginfo_zone, ksi);
404 		return (true);
405 	}
406 	return (false);
407 }
408 
409 void
410 sigqueue_init(sigqueue_t *list, struct proc *p)
411 {
412 	SIGEMPTYSET(list->sq_signals);
413 	SIGEMPTYSET(list->sq_kill);
414 	SIGEMPTYSET(list->sq_ptrace);
415 	TAILQ_INIT(&list->sq_list);
416 	list->sq_proc = p;
417 	list->sq_flags = SQ_INIT;
418 }
419 
420 /*
421  * Get a signal's ksiginfo.
422  * Return:
423  *	0	-	signal not found
424  *	others	-	signal number
425  */
426 static int
427 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
428 {
429 	struct proc *p = sq->sq_proc;
430 	struct ksiginfo *ksi, *next;
431 	int count = 0;
432 
433 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
434 
435 	if (!SIGISMEMBER(sq->sq_signals, signo))
436 		return (0);
437 
438 	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
439 		count++;
440 		SIGDELSET(sq->sq_ptrace, signo);
441 		si->ksi_flags |= KSI_PTRACE;
442 	}
443 	if (SIGISMEMBER(sq->sq_kill, signo)) {
444 		count++;
445 		if (count == 1)
446 			SIGDELSET(sq->sq_kill, signo);
447 	}
448 
449 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
450 		if (ksi->ksi_signo == signo) {
451 			if (count == 0) {
452 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
453 				ksi->ksi_sigq = NULL;
454 				ksiginfo_copy(ksi, si);
455 				if (ksiginfo_tryfree(ksi) && p != NULL)
456 					p->p_pendingcnt--;
457 			}
458 			if (++count > 1)
459 				break;
460 		}
461 	}
462 
463 	if (count <= 1)
464 		SIGDELSET(sq->sq_signals, signo);
465 	si->ksi_signo = signo;
466 	return (signo);
467 }
468 
469 void
470 sigqueue_take(ksiginfo_t *ksi)
471 {
472 	struct ksiginfo *kp;
473 	struct proc	*p;
474 	sigqueue_t	*sq;
475 
476 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
477 		return;
478 
479 	p = sq->sq_proc;
480 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
481 	ksi->ksi_sigq = NULL;
482 	sig_handle_killpg(p, ksi);
483 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
484 		p->p_pendingcnt--;
485 
486 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
487 	     kp = TAILQ_NEXT(kp, ksi_link)) {
488 		if (kp->ksi_signo == ksi->ksi_signo)
489 			break;
490 	}
491 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
492 	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
493 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
494 }
495 
496 static int
497 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
498 {
499 	struct proc *p = sq->sq_proc;
500 	struct ksiginfo *ksi;
501 	int ret = 0;
502 
503 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
504 
505 	/*
506 	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
507 	 * for these signals.
508 	 */
509 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
510 		SIGADDSET(sq->sq_kill, signo);
511 		goto out_set_bit;
512 	}
513 
514 	/* directly insert the ksi, don't copy it */
515 	if (si->ksi_flags & KSI_INS) {
516 		if (si->ksi_flags & KSI_HEAD)
517 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
518 		else
519 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
520 		si->ksi_sigq = sq;
521 		goto out_set_bit;
522 	}
523 
524 	if (__predict_false(ksiginfo_zone == NULL)) {
525 		SIGADDSET(sq->sq_kill, signo);
526 		goto out_set_bit;
527 	}
528 
529 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
530 		signal_overflow++;
531 		ret = EAGAIN;
532 	} else if ((ksi = ksiginfo_alloc(M_NOWAIT)) == NULL) {
533 		signal_alloc_fail++;
534 		ret = EAGAIN;
535 	} else {
536 		if (p != NULL)
537 			p->p_pendingcnt++;
538 		ksiginfo_copy(si, ksi);
539 		ksi->ksi_signo = signo;
540 		if (si->ksi_flags & KSI_HEAD)
541 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
542 		else
543 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
544 		ksi->ksi_sigq = sq;
545 	}
546 
547 	if (ret != 0) {
548 		if ((si->ksi_flags & KSI_PTRACE) != 0) {
549 			SIGADDSET(sq->sq_ptrace, signo);
550 			ret = 0;
551 			goto out_set_bit;
552 		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
553 		    (si->ksi_flags & KSI_SIGQ) == 0) {
554 			SIGADDSET(sq->sq_kill, signo);
555 			ret = 0;
556 			goto out_set_bit;
557 		}
558 		return (ret);
559 	}
560 
561 out_set_bit:
562 	SIGADDSET(sq->sq_signals, signo);
563 	return (ret);
564 }
565 
566 void
567 sigqueue_flush(sigqueue_t *sq)
568 {
569 	struct proc *p = sq->sq_proc;
570 	ksiginfo_t *ksi;
571 
572 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
573 
574 	if (p != NULL)
575 		PROC_LOCK_ASSERT(p, MA_OWNED);
576 
577 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
578 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
579 		ksi->ksi_sigq = NULL;
580 		sig_handle_killpg(p, ksi);
581 		if (ksiginfo_tryfree(ksi) && p != NULL)
582 			p->p_pendingcnt--;
583 	}
584 
585 	SIGEMPTYSET(sq->sq_signals);
586 	SIGEMPTYSET(sq->sq_kill);
587 	SIGEMPTYSET(sq->sq_ptrace);
588 }
589 
590 static void
591 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
592 {
593 	sigset_t tmp;
594 	struct proc *p1, *p2;
595 	ksiginfo_t *ksi, *next;
596 
597 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
598 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
599 	p1 = src->sq_proc;
600 	p2 = dst->sq_proc;
601 	/* Move siginfo to target list */
602 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
603 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
604 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
605 			if (p1 != NULL)
606 				p1->p_pendingcnt--;
607 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
608 			ksi->ksi_sigq = dst;
609 			if (p2 != NULL)
610 				p2->p_pendingcnt++;
611 		}
612 	}
613 
614 	/* Move pending bits to target list */
615 	tmp = src->sq_kill;
616 	SIGSETAND(tmp, *set);
617 	SIGSETOR(dst->sq_kill, tmp);
618 	SIGSETNAND(src->sq_kill, tmp);
619 
620 	tmp = src->sq_ptrace;
621 	SIGSETAND(tmp, *set);
622 	SIGSETOR(dst->sq_ptrace, tmp);
623 	SIGSETNAND(src->sq_ptrace, tmp);
624 
625 	tmp = src->sq_signals;
626 	SIGSETAND(tmp, *set);
627 	SIGSETOR(dst->sq_signals, tmp);
628 	SIGSETNAND(src->sq_signals, tmp);
629 }
630 
631 #if 0
632 static void
633 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
634 {
635 	sigset_t set;
636 
637 	SIGEMPTYSET(set);
638 	SIGADDSET(set, signo);
639 	sigqueue_move_set(src, dst, &set);
640 }
641 #endif
642 
643 static void
644 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
645 {
646 	struct proc *p = sq->sq_proc;
647 	ksiginfo_t *ksi, *next;
648 
649 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
650 
651 	/* Remove siginfo queue */
652 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
653 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
654 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
655 			ksi->ksi_sigq = NULL;
656 			sig_handle_killpg(p, ksi);
657 			if (ksiginfo_tryfree(ksi) && p != NULL)
658 				p->p_pendingcnt--;
659 		}
660 	}
661 	SIGSETNAND(sq->sq_kill, *set);
662 	SIGSETNAND(sq->sq_ptrace, *set);
663 	SIGSETNAND(sq->sq_signals, *set);
664 }
665 
666 void
667 sigqueue_delete(sigqueue_t *sq, int signo)
668 {
669 	sigset_t set;
670 
671 	SIGEMPTYSET(set);
672 	SIGADDSET(set, signo);
673 	sigqueue_delete_set(sq, &set);
674 }
675 
676 /* Remove a set of signals for a process */
677 static void
678 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
679 {
680 	sigqueue_t worklist;
681 	struct thread *td0;
682 
683 	PROC_LOCK_ASSERT(p, MA_OWNED);
684 
685 	sigqueue_init(&worklist, p);
686 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
687 
688 	FOREACH_THREAD_IN_PROC(p, td0)
689 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
690 
691 	sigqueue_flush(&worklist);
692 }
693 
694 void
695 sigqueue_delete_proc(struct proc *p, int signo)
696 {
697 	sigset_t set;
698 
699 	SIGEMPTYSET(set);
700 	SIGADDSET(set, signo);
701 	sigqueue_delete_set_proc(p, &set);
702 }
703 
704 static void
705 sigqueue_delete_stopmask_proc(struct proc *p)
706 {
707 	sigset_t set;
708 
709 	SIGEMPTYSET(set);
710 	SIGADDSET(set, SIGSTOP);
711 	SIGADDSET(set, SIGTSTP);
712 	SIGADDSET(set, SIGTTIN);
713 	SIGADDSET(set, SIGTTOU);
714 	sigqueue_delete_set_proc(p, &set);
715 }
716 
717 /*
718  * Determine signal that should be delivered to thread td, the current
719  * thread, 0 if none.  If there is a pending stop signal with default
720  * action, the process stops in issignal().
721  */
722 int
723 cursig(struct thread *td)
724 {
725 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
726 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
727 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
728 	return (SIGPENDING(td) ? issignal(td) : 0);
729 }
730 
731 /*
732  * Arrange for ast() to handle unmasked pending signals on return to user
733  * mode.  This must be called whenever a signal is added to td_sigqueue or
734  * unmasked in td_sigmask.
735  */
736 void
737 signotify(struct thread *td)
738 {
739 
740 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
741 
742 	if (SIGPENDING(td))
743 		ast_sched(td, TDA_SIG);
744 }
745 
746 /*
747  * Returns 1 (true) if altstack is configured for the thread, and the
748  * passed stack bottom address falls into the altstack range.  Handles
749  * the 43 compat special case where the alt stack size is zero.
750  */
751 int
752 sigonstack(size_t sp)
753 {
754 	struct thread *td;
755 
756 	td = curthread;
757 	if ((td->td_pflags & TDP_ALTSTACK) == 0)
758 		return (0);
759 #if defined(COMPAT_43)
760 	if (SV_PROC_FLAG(td->td_proc, SV_AOUT) && td->td_sigstk.ss_size == 0)
761 		return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
762 #endif
763 	return (sp >= (size_t)td->td_sigstk.ss_sp &&
764 	    sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
765 }
766 
767 static __inline int
768 sigprop(int sig)
769 {
770 
771 	if (sig > 0 && sig < nitems(sigproptbl))
772 		return (sigproptbl[sig]);
773 	return (0);
774 }
775 
776 static bool
777 sigact_flag_test(const struct sigaction *act, int flag)
778 {
779 
780 	/*
781 	 * SA_SIGINFO is reset when signal disposition is set to
782 	 * ignore or default.  Other flags are kept according to user
783 	 * settings.
784 	 */
785 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
786 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
787 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
788 }
789 
790 /*
791  * kern_sigaction
792  * sigaction
793  * freebsd4_sigaction
794  * osigaction
795  */
796 int
797 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
798     struct sigaction *oact, int flags)
799 {
800 	struct sigacts *ps;
801 	struct proc *p = td->td_proc;
802 
803 	if (!_SIG_VALID(sig))
804 		return (EINVAL);
805 	if (act != NULL && act->sa_handler != SIG_DFL &&
806 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
807 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
808 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
809 		return (EINVAL);
810 
811 	PROC_LOCK(p);
812 	ps = p->p_sigacts;
813 	mtx_lock(&ps->ps_mtx);
814 	if (oact) {
815 		memset(oact, 0, sizeof(*oact));
816 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
817 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
818 			oact->sa_flags |= SA_ONSTACK;
819 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
820 			oact->sa_flags |= SA_RESTART;
821 		if (SIGISMEMBER(ps->ps_sigreset, sig))
822 			oact->sa_flags |= SA_RESETHAND;
823 		if (SIGISMEMBER(ps->ps_signodefer, sig))
824 			oact->sa_flags |= SA_NODEFER;
825 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
826 			oact->sa_flags |= SA_SIGINFO;
827 			oact->sa_sigaction =
828 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
829 		} else
830 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
831 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
832 			oact->sa_flags |= SA_NOCLDSTOP;
833 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
834 			oact->sa_flags |= SA_NOCLDWAIT;
835 	}
836 	if (act) {
837 		if ((sig == SIGKILL || sig == SIGSTOP) &&
838 		    act->sa_handler != SIG_DFL) {
839 			mtx_unlock(&ps->ps_mtx);
840 			PROC_UNLOCK(p);
841 			return (EINVAL);
842 		}
843 
844 		/*
845 		 * Change setting atomically.
846 		 */
847 
848 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
849 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
850 		if (sigact_flag_test(act, SA_SIGINFO)) {
851 			ps->ps_sigact[_SIG_IDX(sig)] =
852 			    (__sighandler_t *)act->sa_sigaction;
853 			SIGADDSET(ps->ps_siginfo, sig);
854 		} else {
855 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
856 			SIGDELSET(ps->ps_siginfo, sig);
857 		}
858 		if (!sigact_flag_test(act, SA_RESTART))
859 			SIGADDSET(ps->ps_sigintr, sig);
860 		else
861 			SIGDELSET(ps->ps_sigintr, sig);
862 		if (sigact_flag_test(act, SA_ONSTACK))
863 			SIGADDSET(ps->ps_sigonstack, sig);
864 		else
865 			SIGDELSET(ps->ps_sigonstack, sig);
866 		if (sigact_flag_test(act, SA_RESETHAND))
867 			SIGADDSET(ps->ps_sigreset, sig);
868 		else
869 			SIGDELSET(ps->ps_sigreset, sig);
870 		if (sigact_flag_test(act, SA_NODEFER))
871 			SIGADDSET(ps->ps_signodefer, sig);
872 		else
873 			SIGDELSET(ps->ps_signodefer, sig);
874 		if (sig == SIGCHLD) {
875 			if (act->sa_flags & SA_NOCLDSTOP)
876 				ps->ps_flag |= PS_NOCLDSTOP;
877 			else
878 				ps->ps_flag &= ~PS_NOCLDSTOP;
879 			if (act->sa_flags & SA_NOCLDWAIT) {
880 				/*
881 				 * Paranoia: since SA_NOCLDWAIT is implemented
882 				 * by reparenting the dying child to PID 1 (and
883 				 * trust it to reap the zombie), PID 1 itself
884 				 * is forbidden to set SA_NOCLDWAIT.
885 				 */
886 				if (p->p_pid == 1)
887 					ps->ps_flag &= ~PS_NOCLDWAIT;
888 				else
889 					ps->ps_flag |= PS_NOCLDWAIT;
890 			} else
891 				ps->ps_flag &= ~PS_NOCLDWAIT;
892 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
893 				ps->ps_flag |= PS_CLDSIGIGN;
894 			else
895 				ps->ps_flag &= ~PS_CLDSIGIGN;
896 		}
897 		/*
898 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
899 		 * and for signals set to SIG_DFL where the default is to
900 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
901 		 * have to restart the process.
902 		 */
903 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
904 		    (sigprop(sig) & SIGPROP_IGNORE &&
905 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
906 			/* never to be seen again */
907 			sigqueue_delete_proc(p, sig);
908 			if (sig != SIGCONT)
909 				/* easier in psignal */
910 				SIGADDSET(ps->ps_sigignore, sig);
911 			SIGDELSET(ps->ps_sigcatch, sig);
912 		} else {
913 			SIGDELSET(ps->ps_sigignore, sig);
914 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
915 				SIGDELSET(ps->ps_sigcatch, sig);
916 			else
917 				SIGADDSET(ps->ps_sigcatch, sig);
918 		}
919 #ifdef COMPAT_FREEBSD4
920 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
921 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
922 		    (flags & KSA_FREEBSD4) == 0)
923 			SIGDELSET(ps->ps_freebsd4, sig);
924 		else
925 			SIGADDSET(ps->ps_freebsd4, sig);
926 #endif
927 #ifdef COMPAT_43
928 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
929 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
930 		    (flags & KSA_OSIGSET) == 0)
931 			SIGDELSET(ps->ps_osigset, sig);
932 		else
933 			SIGADDSET(ps->ps_osigset, sig);
934 #endif
935 	}
936 	mtx_unlock(&ps->ps_mtx);
937 	PROC_UNLOCK(p);
938 	return (0);
939 }
940 
941 #ifndef _SYS_SYSPROTO_H_
942 struct sigaction_args {
943 	int	sig;
944 	struct	sigaction *act;
945 	struct	sigaction *oact;
946 };
947 #endif
948 int
949 sys_sigaction(struct thread *td, struct sigaction_args *uap)
950 {
951 	struct sigaction act, oact;
952 	struct sigaction *actp, *oactp;
953 	int error;
954 
955 	actp = (uap->act != NULL) ? &act : NULL;
956 	oactp = (uap->oact != NULL) ? &oact : NULL;
957 	if (actp) {
958 		error = copyin(uap->act, actp, sizeof(act));
959 		if (error)
960 			return (error);
961 	}
962 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
963 	if (oactp && !error)
964 		error = copyout(oactp, uap->oact, sizeof(oact));
965 	return (error);
966 }
967 
968 #ifdef COMPAT_FREEBSD4
969 #ifndef _SYS_SYSPROTO_H_
970 struct freebsd4_sigaction_args {
971 	int	sig;
972 	struct	sigaction *act;
973 	struct	sigaction *oact;
974 };
975 #endif
976 int
977 freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
978 {
979 	struct sigaction act, oact;
980 	struct sigaction *actp, *oactp;
981 	int error;
982 
983 	actp = (uap->act != NULL) ? &act : NULL;
984 	oactp = (uap->oact != NULL) ? &oact : NULL;
985 	if (actp) {
986 		error = copyin(uap->act, actp, sizeof(act));
987 		if (error)
988 			return (error);
989 	}
990 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
991 	if (oactp && !error)
992 		error = copyout(oactp, uap->oact, sizeof(oact));
993 	return (error);
994 }
995 #endif	/* COMAPT_FREEBSD4 */
996 
997 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
998 #ifndef _SYS_SYSPROTO_H_
999 struct osigaction_args {
1000 	int	signum;
1001 	struct	osigaction *nsa;
1002 	struct	osigaction *osa;
1003 };
1004 #endif
1005 int
1006 osigaction(struct thread *td, struct osigaction_args *uap)
1007 {
1008 	struct osigaction sa;
1009 	struct sigaction nsa, osa;
1010 	struct sigaction *nsap, *osap;
1011 	int error;
1012 
1013 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1014 		return (EINVAL);
1015 
1016 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
1017 	osap = (uap->osa != NULL) ? &osa : NULL;
1018 
1019 	if (nsap) {
1020 		error = copyin(uap->nsa, &sa, sizeof(sa));
1021 		if (error)
1022 			return (error);
1023 		nsap->sa_handler = sa.sa_handler;
1024 		nsap->sa_flags = sa.sa_flags;
1025 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
1026 	}
1027 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1028 	if (osap && !error) {
1029 		sa.sa_handler = osap->sa_handler;
1030 		sa.sa_flags = osap->sa_flags;
1031 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
1032 		error = copyout(&sa, uap->osa, sizeof(sa));
1033 	}
1034 	return (error);
1035 }
1036 
1037 #if !defined(__i386__)
1038 /* Avoid replicating the same stub everywhere */
1039 int
1040 osigreturn(struct thread *td, struct osigreturn_args *uap)
1041 {
1042 
1043 	return (nosys(td, (struct nosys_args *)uap));
1044 }
1045 #endif
1046 #endif /* COMPAT_43 */
1047 
1048 /*
1049  * Initialize signal state for process 0;
1050  * set to ignore signals that are ignored by default.
1051  */
1052 void
1053 siginit(struct proc *p)
1054 {
1055 	int i;
1056 	struct sigacts *ps;
1057 
1058 	PROC_LOCK(p);
1059 	ps = p->p_sigacts;
1060 	mtx_lock(&ps->ps_mtx);
1061 	for (i = 1; i <= NSIG; i++) {
1062 		if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) {
1063 			SIGADDSET(ps->ps_sigignore, i);
1064 		}
1065 	}
1066 	mtx_unlock(&ps->ps_mtx);
1067 	PROC_UNLOCK(p);
1068 }
1069 
1070 /*
1071  * Reset specified signal to the default disposition.
1072  */
1073 static void
1074 sigdflt(struct sigacts *ps, int sig)
1075 {
1076 
1077 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1078 	SIGDELSET(ps->ps_sigcatch, sig);
1079 	if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT)
1080 		SIGADDSET(ps->ps_sigignore, sig);
1081 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1082 	SIGDELSET(ps->ps_siginfo, sig);
1083 }
1084 
1085 /*
1086  * Reset signals for an exec of the specified process.
1087  */
1088 void
1089 execsigs(struct proc *p)
1090 {
1091 	struct sigacts *ps;
1092 	struct thread *td;
1093 
1094 	/*
1095 	 * Reset caught signals.  Held signals remain held
1096 	 * through td_sigmask (unless they were caught,
1097 	 * and are now ignored by default).
1098 	 */
1099 	PROC_LOCK_ASSERT(p, MA_OWNED);
1100 	ps = p->p_sigacts;
1101 	mtx_lock(&ps->ps_mtx);
1102 	sig_drop_caught(p);
1103 
1104 	/*
1105 	 * Reset stack state to the user stack.
1106 	 * Clear set of signals caught on the signal stack.
1107 	 */
1108 	td = curthread;
1109 	MPASS(td->td_proc == p);
1110 	td->td_sigstk.ss_flags = SS_DISABLE;
1111 	td->td_sigstk.ss_size = 0;
1112 	td->td_sigstk.ss_sp = 0;
1113 	td->td_pflags &= ~TDP_ALTSTACK;
1114 	/*
1115 	 * Reset no zombies if child dies flag as Solaris does.
1116 	 */
1117 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1118 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1119 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1120 	mtx_unlock(&ps->ps_mtx);
1121 }
1122 
1123 /*
1124  * kern_sigprocmask()
1125  *
1126  *	Manipulate signal mask.
1127  */
1128 int
1129 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1130     int flags)
1131 {
1132 	sigset_t new_block, oset1;
1133 	struct proc *p;
1134 	int error;
1135 
1136 	p = td->td_proc;
1137 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1138 		PROC_LOCK_ASSERT(p, MA_OWNED);
1139 	else
1140 		PROC_LOCK(p);
1141 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1142 	    ? MA_OWNED : MA_NOTOWNED);
1143 	if (oset != NULL)
1144 		*oset = td->td_sigmask;
1145 
1146 	error = 0;
1147 	if (set != NULL) {
1148 		switch (how) {
1149 		case SIG_BLOCK:
1150 			SIG_CANTMASK(*set);
1151 			oset1 = td->td_sigmask;
1152 			SIGSETOR(td->td_sigmask, *set);
1153 			new_block = td->td_sigmask;
1154 			SIGSETNAND(new_block, oset1);
1155 			break;
1156 		case SIG_UNBLOCK:
1157 			SIGSETNAND(td->td_sigmask, *set);
1158 			signotify(td);
1159 			goto out;
1160 		case SIG_SETMASK:
1161 			SIG_CANTMASK(*set);
1162 			oset1 = td->td_sigmask;
1163 			if (flags & SIGPROCMASK_OLD)
1164 				SIGSETLO(td->td_sigmask, *set);
1165 			else
1166 				td->td_sigmask = *set;
1167 			new_block = td->td_sigmask;
1168 			SIGSETNAND(new_block, oset1);
1169 			signotify(td);
1170 			break;
1171 		default:
1172 			error = EINVAL;
1173 			goto out;
1174 		}
1175 
1176 		/*
1177 		 * The new_block set contains signals that were not previously
1178 		 * blocked, but are blocked now.
1179 		 *
1180 		 * In case we block any signal that was not previously blocked
1181 		 * for td, and process has the signal pending, try to schedule
1182 		 * signal delivery to some thread that does not block the
1183 		 * signal, possibly waking it up.
1184 		 */
1185 		if (p->p_numthreads != 1)
1186 			reschedule_signals(p, new_block, flags);
1187 	}
1188 
1189 out:
1190 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1191 		PROC_UNLOCK(p);
1192 	return (error);
1193 }
1194 
1195 #ifndef _SYS_SYSPROTO_H_
1196 struct sigprocmask_args {
1197 	int	how;
1198 	const sigset_t *set;
1199 	sigset_t *oset;
1200 };
1201 #endif
1202 int
1203 sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1204 {
1205 	sigset_t set, oset;
1206 	sigset_t *setp, *osetp;
1207 	int error;
1208 
1209 	setp = (uap->set != NULL) ? &set : NULL;
1210 	osetp = (uap->oset != NULL) ? &oset : NULL;
1211 	if (setp) {
1212 		error = copyin(uap->set, setp, sizeof(set));
1213 		if (error)
1214 			return (error);
1215 	}
1216 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1217 	if (osetp && !error) {
1218 		error = copyout(osetp, uap->oset, sizeof(oset));
1219 	}
1220 	return (error);
1221 }
1222 
1223 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1224 #ifndef _SYS_SYSPROTO_H_
1225 struct osigprocmask_args {
1226 	int	how;
1227 	osigset_t mask;
1228 };
1229 #endif
1230 int
1231 osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1232 {
1233 	sigset_t set, oset;
1234 	int error;
1235 
1236 	OSIG2SIG(uap->mask, set);
1237 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1238 	SIG2OSIG(oset, td->td_retval[0]);
1239 	return (error);
1240 }
1241 #endif /* COMPAT_43 */
1242 
1243 int
1244 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1245 {
1246 	ksiginfo_t ksi;
1247 	sigset_t set;
1248 	int error;
1249 
1250 	error = copyin(uap->set, &set, sizeof(set));
1251 	if (error) {
1252 		td->td_retval[0] = error;
1253 		return (0);
1254 	}
1255 
1256 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1257 	if (error) {
1258 		/*
1259 		 * sigwait() function shall not return EINTR, but
1260 		 * the syscall does.  Non-ancient libc provides the
1261 		 * wrapper which hides EINTR.  Otherwise, EINTR return
1262 		 * is used by libthr to handle required cancellation
1263 		 * point in the sigwait().
1264 		 */
1265 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1266 			return (ERESTART);
1267 		td->td_retval[0] = error;
1268 		return (0);
1269 	}
1270 
1271 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1272 	td->td_retval[0] = error;
1273 	return (0);
1274 }
1275 
1276 int
1277 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1278 {
1279 	struct timespec ts;
1280 	struct timespec *timeout;
1281 	sigset_t set;
1282 	ksiginfo_t ksi;
1283 	int error;
1284 
1285 	if (uap->timeout) {
1286 		error = copyin(uap->timeout, &ts, sizeof(ts));
1287 		if (error)
1288 			return (error);
1289 
1290 		timeout = &ts;
1291 	} else
1292 		timeout = NULL;
1293 
1294 	error = copyin(uap->set, &set, sizeof(set));
1295 	if (error)
1296 		return (error);
1297 
1298 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1299 	if (error)
1300 		return (error);
1301 
1302 	if (uap->info)
1303 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1304 
1305 	if (error == 0)
1306 		td->td_retval[0] = ksi.ksi_signo;
1307 	return (error);
1308 }
1309 
1310 int
1311 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1312 {
1313 	ksiginfo_t ksi;
1314 	sigset_t set;
1315 	int error;
1316 
1317 	error = copyin(uap->set, &set, sizeof(set));
1318 	if (error)
1319 		return (error);
1320 
1321 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1322 	if (error)
1323 		return (error);
1324 
1325 	if (uap->info)
1326 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1327 
1328 	if (error == 0)
1329 		td->td_retval[0] = ksi.ksi_signo;
1330 	return (error);
1331 }
1332 
1333 static void
1334 proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1335 {
1336 	struct thread *thr;
1337 
1338 	FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1339 		if (thr == td)
1340 			thr->td_si = *si;
1341 		else
1342 			thr->td_si.si_signo = 0;
1343 	}
1344 }
1345 
1346 int
1347 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1348 	struct timespec *timeout)
1349 {
1350 	struct sigacts *ps;
1351 	sigset_t saved_mask, new_block;
1352 	struct proc *p;
1353 	int error, sig, timevalid = 0;
1354 	sbintime_t sbt, precision, tsbt;
1355 	struct timespec ts;
1356 	bool traced;
1357 
1358 	p = td->td_proc;
1359 	error = 0;
1360 	traced = false;
1361 
1362 	/* Ensure the sigfastblock value is up to date. */
1363 	sigfastblock_fetch(td);
1364 
1365 	if (timeout != NULL) {
1366 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1367 			timevalid = 1;
1368 			ts = *timeout;
1369 			if (ts.tv_sec < INT32_MAX / 2) {
1370 				tsbt = tstosbt(ts);
1371 				precision = tsbt;
1372 				precision >>= tc_precexp;
1373 				if (TIMESEL(&sbt, tsbt))
1374 					sbt += tc_tick_sbt;
1375 				sbt += tsbt;
1376 			} else
1377 				precision = sbt = 0;
1378 		}
1379 	} else
1380 		precision = sbt = 0;
1381 	ksiginfo_init(ksi);
1382 	/* Some signals can not be waited for. */
1383 	SIG_CANTMASK(waitset);
1384 	ps = p->p_sigacts;
1385 	PROC_LOCK(p);
1386 	saved_mask = td->td_sigmask;
1387 	SIGSETNAND(td->td_sigmask, waitset);
1388 	if ((p->p_sysent->sv_flags & SV_SIG_DISCIGN) != 0 ||
1389 	    !kern_sig_discard_ign) {
1390 		thread_lock(td);
1391 		td->td_flags |= TDF_SIGWAIT;
1392 		thread_unlock(td);
1393 	}
1394 	for (;;) {
1395 		mtx_lock(&ps->ps_mtx);
1396 		sig = cursig(td);
1397 		mtx_unlock(&ps->ps_mtx);
1398 		KASSERT(sig >= 0, ("sig %d", sig));
1399 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1400 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1401 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1402 				error = 0;
1403 				break;
1404 			}
1405 		}
1406 
1407 		if (error != 0)
1408 			break;
1409 
1410 		/*
1411 		 * POSIX says this must be checked after looking for pending
1412 		 * signals.
1413 		 */
1414 		if (timeout != NULL && !timevalid) {
1415 			error = EINVAL;
1416 			break;
1417 		}
1418 
1419 		if (traced) {
1420 			error = EINTR;
1421 			break;
1422 		}
1423 
1424 		error = msleep_sbt(&p->p_sigacts, &p->p_mtx, PPAUSE | PCATCH,
1425 		    "sigwait", sbt, precision, C_ABSOLUTE);
1426 
1427 		/* The syscalls can not be restarted. */
1428 		if (error == ERESTART)
1429 			error = EINTR;
1430 
1431 		/*
1432 		 * If PTRACE_SCE or PTRACE_SCX were set after
1433 		 * userspace entered the syscall, return spurious
1434 		 * EINTR after wait was done.  Only do this as last
1435 		 * resort after rechecking for possible queued signals
1436 		 * and expired timeouts.
1437 		 */
1438 		if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1439 			traced = true;
1440 	}
1441 	thread_lock(td);
1442 	td->td_flags &= ~TDF_SIGWAIT;
1443 	thread_unlock(td);
1444 
1445 	new_block = saved_mask;
1446 	SIGSETNAND(new_block, td->td_sigmask);
1447 	td->td_sigmask = saved_mask;
1448 	/*
1449 	 * Fewer signals can be delivered to us, reschedule signal
1450 	 * notification.
1451 	 */
1452 	if (p->p_numthreads != 1)
1453 		reschedule_signals(p, new_block, 0);
1454 
1455 	if (error == 0) {
1456 		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1457 
1458 		if (ksi->ksi_code == SI_TIMER)
1459 			itimer_accept(p, ksi->ksi_timerid, ksi);
1460 
1461 #ifdef KTRACE
1462 		if (KTRPOINT(td, KTR_PSIG)) {
1463 			sig_t action;
1464 
1465 			mtx_lock(&ps->ps_mtx);
1466 			action = ps->ps_sigact[_SIG_IDX(sig)];
1467 			mtx_unlock(&ps->ps_mtx);
1468 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1469 		}
1470 #endif
1471 		if (sig == SIGKILL) {
1472 			proc_td_siginfo_capture(td, &ksi->ksi_info);
1473 			sigexit1(td, sig, ksi);
1474 		}
1475 	}
1476 	PROC_UNLOCK(p);
1477 	return (error);
1478 }
1479 
1480 #ifndef _SYS_SYSPROTO_H_
1481 struct sigpending_args {
1482 	sigset_t	*set;
1483 };
1484 #endif
1485 int
1486 sys_sigpending(struct thread *td, struct sigpending_args *uap)
1487 {
1488 	struct proc *p = td->td_proc;
1489 	sigset_t pending;
1490 
1491 	PROC_LOCK(p);
1492 	pending = p->p_sigqueue.sq_signals;
1493 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1494 	PROC_UNLOCK(p);
1495 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1496 }
1497 
1498 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1499 #ifndef _SYS_SYSPROTO_H_
1500 struct osigpending_args {
1501 	int	dummy;
1502 };
1503 #endif
1504 int
1505 osigpending(struct thread *td, struct osigpending_args *uap)
1506 {
1507 	struct proc *p = td->td_proc;
1508 	sigset_t pending;
1509 
1510 	PROC_LOCK(p);
1511 	pending = p->p_sigqueue.sq_signals;
1512 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1513 	PROC_UNLOCK(p);
1514 	SIG2OSIG(pending, td->td_retval[0]);
1515 	return (0);
1516 }
1517 #endif /* COMPAT_43 */
1518 
1519 #if defined(COMPAT_43)
1520 /*
1521  * Generalized interface signal handler, 4.3-compatible.
1522  */
1523 #ifndef _SYS_SYSPROTO_H_
1524 struct osigvec_args {
1525 	int	signum;
1526 	struct	sigvec *nsv;
1527 	struct	sigvec *osv;
1528 };
1529 #endif
1530 /* ARGSUSED */
1531 int
1532 osigvec(struct thread *td, struct osigvec_args *uap)
1533 {
1534 	struct sigvec vec;
1535 	struct sigaction nsa, osa;
1536 	struct sigaction *nsap, *osap;
1537 	int error;
1538 
1539 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1540 		return (EINVAL);
1541 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1542 	osap = (uap->osv != NULL) ? &osa : NULL;
1543 	if (nsap) {
1544 		error = copyin(uap->nsv, &vec, sizeof(vec));
1545 		if (error)
1546 			return (error);
1547 		nsap->sa_handler = vec.sv_handler;
1548 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1549 		nsap->sa_flags = vec.sv_flags;
1550 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1551 	}
1552 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1553 	if (osap && !error) {
1554 		vec.sv_handler = osap->sa_handler;
1555 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1556 		vec.sv_flags = osap->sa_flags;
1557 		vec.sv_flags &= ~SA_NOCLDWAIT;
1558 		vec.sv_flags ^= SA_RESTART;
1559 		error = copyout(&vec, uap->osv, sizeof(vec));
1560 	}
1561 	return (error);
1562 }
1563 
1564 #ifndef _SYS_SYSPROTO_H_
1565 struct osigblock_args {
1566 	int	mask;
1567 };
1568 #endif
1569 int
1570 osigblock(struct thread *td, struct osigblock_args *uap)
1571 {
1572 	sigset_t set, oset;
1573 
1574 	OSIG2SIG(uap->mask, set);
1575 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1576 	SIG2OSIG(oset, td->td_retval[0]);
1577 	return (0);
1578 }
1579 
1580 #ifndef _SYS_SYSPROTO_H_
1581 struct osigsetmask_args {
1582 	int	mask;
1583 };
1584 #endif
1585 int
1586 osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1587 {
1588 	sigset_t set, oset;
1589 
1590 	OSIG2SIG(uap->mask, set);
1591 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1592 	SIG2OSIG(oset, td->td_retval[0]);
1593 	return (0);
1594 }
1595 #endif /* COMPAT_43 */
1596 
1597 /*
1598  * Suspend calling thread until signal, providing mask to be set in the
1599  * meantime.
1600  */
1601 #ifndef _SYS_SYSPROTO_H_
1602 struct sigsuspend_args {
1603 	const sigset_t *sigmask;
1604 };
1605 #endif
1606 /* ARGSUSED */
1607 int
1608 sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1609 {
1610 	sigset_t mask;
1611 	int error;
1612 
1613 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1614 	if (error)
1615 		return (error);
1616 	return (kern_sigsuspend(td, mask));
1617 }
1618 
1619 int
1620 kern_sigsuspend(struct thread *td, sigset_t mask)
1621 {
1622 	struct proc *p = td->td_proc;
1623 	int has_sig, sig;
1624 
1625 	/* Ensure the sigfastblock value is up to date. */
1626 	sigfastblock_fetch(td);
1627 
1628 	/*
1629 	 * When returning from sigsuspend, we want
1630 	 * the old mask to be restored after the
1631 	 * signal handler has finished.  Thus, we
1632 	 * save it here and mark the sigacts structure
1633 	 * to indicate this.
1634 	 */
1635 	PROC_LOCK(p);
1636 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1637 	    SIGPROCMASK_PROC_LOCKED);
1638 	td->td_pflags |= TDP_OLDMASK;
1639 	ast_sched(td, TDA_SIGSUSPEND);
1640 
1641 	/*
1642 	 * Process signals now. Otherwise, we can get spurious wakeup
1643 	 * due to signal entered process queue, but delivered to other
1644 	 * thread. But sigsuspend should return only on signal
1645 	 * delivery.
1646 	 */
1647 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1648 	for (has_sig = 0; !has_sig;) {
1649 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1650 			0) == 0)
1651 			/* void */;
1652 		thread_suspend_check(0);
1653 		mtx_lock(&p->p_sigacts->ps_mtx);
1654 		while ((sig = cursig(td)) != 0) {
1655 			KASSERT(sig >= 0, ("sig %d", sig));
1656 			has_sig += postsig(sig);
1657 		}
1658 		mtx_unlock(&p->p_sigacts->ps_mtx);
1659 
1660 		/*
1661 		 * If PTRACE_SCE or PTRACE_SCX were set after
1662 		 * userspace entered the syscall, return spurious
1663 		 * EINTR.
1664 		 */
1665 		if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1666 			has_sig += 1;
1667 	}
1668 	PROC_UNLOCK(p);
1669 	td->td_errno = EINTR;
1670 	td->td_pflags |= TDP_NERRNO;
1671 	return (EJUSTRETURN);
1672 }
1673 
1674 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1675 /*
1676  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1677  * convention: libc stub passes mask, not pointer, to save a copyin.
1678  */
1679 #ifndef _SYS_SYSPROTO_H_
1680 struct osigsuspend_args {
1681 	osigset_t mask;
1682 };
1683 #endif
1684 /* ARGSUSED */
1685 int
1686 osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1687 {
1688 	sigset_t mask;
1689 
1690 	OSIG2SIG(uap->mask, mask);
1691 	return (kern_sigsuspend(td, mask));
1692 }
1693 #endif /* COMPAT_43 */
1694 
1695 #if defined(COMPAT_43)
1696 #ifndef _SYS_SYSPROTO_H_
1697 struct osigstack_args {
1698 	struct	sigstack *nss;
1699 	struct	sigstack *oss;
1700 };
1701 #endif
1702 /* ARGSUSED */
1703 int
1704 osigstack(struct thread *td, struct osigstack_args *uap)
1705 {
1706 	struct sigstack nss, oss;
1707 	int error = 0;
1708 
1709 	if (uap->nss != NULL) {
1710 		error = copyin(uap->nss, &nss, sizeof(nss));
1711 		if (error)
1712 			return (error);
1713 	}
1714 	oss.ss_sp = td->td_sigstk.ss_sp;
1715 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1716 	if (uap->nss != NULL) {
1717 		td->td_sigstk.ss_sp = nss.ss_sp;
1718 		td->td_sigstk.ss_size = 0;
1719 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1720 		td->td_pflags |= TDP_ALTSTACK;
1721 	}
1722 	if (uap->oss != NULL)
1723 		error = copyout(&oss, uap->oss, sizeof(oss));
1724 
1725 	return (error);
1726 }
1727 #endif /* COMPAT_43 */
1728 
1729 #ifndef _SYS_SYSPROTO_H_
1730 struct sigaltstack_args {
1731 	stack_t	*ss;
1732 	stack_t	*oss;
1733 };
1734 #endif
1735 /* ARGSUSED */
1736 int
1737 sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1738 {
1739 	stack_t ss, oss;
1740 	int error;
1741 
1742 	if (uap->ss != NULL) {
1743 		error = copyin(uap->ss, &ss, sizeof(ss));
1744 		if (error)
1745 			return (error);
1746 	}
1747 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1748 	    (uap->oss != NULL) ? &oss : NULL);
1749 	if (error)
1750 		return (error);
1751 	if (uap->oss != NULL)
1752 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1753 	return (error);
1754 }
1755 
1756 int
1757 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1758 {
1759 	struct proc *p = td->td_proc;
1760 	int oonstack;
1761 
1762 	oonstack = sigonstack(cpu_getstack(td));
1763 
1764 	if (oss != NULL) {
1765 		*oss = td->td_sigstk;
1766 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1767 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1768 	}
1769 
1770 	if (ss != NULL) {
1771 		if (oonstack)
1772 			return (EPERM);
1773 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1774 			return (EINVAL);
1775 		if (!(ss->ss_flags & SS_DISABLE)) {
1776 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1777 				return (ENOMEM);
1778 
1779 			td->td_sigstk = *ss;
1780 			td->td_pflags |= TDP_ALTSTACK;
1781 		} else {
1782 			td->td_pflags &= ~TDP_ALTSTACK;
1783 		}
1784 	}
1785 	return (0);
1786 }
1787 
1788 struct killpg1_ctx {
1789 	struct thread *td;
1790 	ksiginfo_t *ksi;
1791 	int sig;
1792 	bool sent;
1793 	bool found;
1794 	int ret;
1795 };
1796 
1797 static void
1798 killpg1_sendsig_locked(struct proc *p, struct killpg1_ctx *arg)
1799 {
1800 	int err;
1801 
1802 	err = p_cansignal(arg->td, p, arg->sig);
1803 	if (err == 0 && arg->sig != 0)
1804 		pksignal(p, arg->sig, arg->ksi);
1805 	if (err != ESRCH)
1806 		arg->found = true;
1807 	if (err == 0)
1808 		arg->sent = true;
1809 	else if (arg->ret == 0 && err != ESRCH && err != EPERM)
1810 		arg->ret = err;
1811 }
1812 
1813 static void
1814 killpg1_sendsig(struct proc *p, bool notself, struct killpg1_ctx *arg)
1815 {
1816 
1817 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1818 	    (notself && p == arg->td->td_proc) || p->p_state == PRS_NEW)
1819 		return;
1820 
1821 	PROC_LOCK(p);
1822 	killpg1_sendsig_locked(p, arg);
1823 	PROC_UNLOCK(p);
1824 }
1825 
1826 static void
1827 kill_processes_prison_cb(struct proc *p, void *arg)
1828 {
1829 	struct killpg1_ctx *ctx = arg;
1830 
1831 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1832 	    (p == ctx->td->td_proc) || p->p_state == PRS_NEW)
1833 		return;
1834 
1835 	killpg1_sendsig_locked(p, ctx);
1836 }
1837 
1838 /*
1839  * Common code for kill process group/broadcast kill.
1840  * td is the calling thread, as usual.
1841  */
1842 static int
1843 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1844 {
1845 	struct proc *p;
1846 	struct pgrp *pgrp;
1847 	struct killpg1_ctx arg;
1848 
1849 	arg.td = td;
1850 	arg.ksi = ksi;
1851 	arg.sig = sig;
1852 	arg.sent = false;
1853 	arg.found = false;
1854 	arg.ret = 0;
1855 	if (all) {
1856 		/*
1857 		 * broadcast
1858 		 */
1859 		prison_proc_iterate(td->td_ucred->cr_prison,
1860 		    kill_processes_prison_cb, &arg);
1861 	} else {
1862 again:
1863 		sx_slock(&proctree_lock);
1864 		if (pgid == 0) {
1865 			/*
1866 			 * zero pgid means send to my process group.
1867 			 */
1868 			pgrp = td->td_proc->p_pgrp;
1869 			PGRP_LOCK(pgrp);
1870 		} else {
1871 			pgrp = pgfind(pgid);
1872 			if (pgrp == NULL) {
1873 				sx_sunlock(&proctree_lock);
1874 				return (ESRCH);
1875 			}
1876 		}
1877 		sx_sunlock(&proctree_lock);
1878 		if (!sx_try_xlock(&pgrp->pg_killsx)) {
1879 			PGRP_UNLOCK(pgrp);
1880 			sx_xlock(&pgrp->pg_killsx);
1881 			sx_xunlock(&pgrp->pg_killsx);
1882 			goto again;
1883 		}
1884 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1885 			killpg1_sendsig(p, false, &arg);
1886 		}
1887 		PGRP_UNLOCK(pgrp);
1888 		sx_xunlock(&pgrp->pg_killsx);
1889 	}
1890 	MPASS(arg.ret != 0 || arg.found || !arg.sent);
1891 	if (arg.ret == 0 && !arg.sent)
1892 		arg.ret = arg.found ? EPERM : ESRCH;
1893 	return (arg.ret);
1894 }
1895 
1896 #ifndef _SYS_SYSPROTO_H_
1897 struct kill_args {
1898 	int	pid;
1899 	int	signum;
1900 };
1901 #endif
1902 /* ARGSUSED */
1903 int
1904 sys_kill(struct thread *td, struct kill_args *uap)
1905 {
1906 
1907 	return (kern_kill(td, uap->pid, uap->signum));
1908 }
1909 
1910 int
1911 kern_kill(struct thread *td, pid_t pid, int signum)
1912 {
1913 	ksiginfo_t ksi;
1914 	struct proc *p;
1915 	int error;
1916 
1917 	/*
1918 	 * A process in capability mode can send signals only to himself.
1919 	 * The main rationale behind this is that abort(3) is implemented as
1920 	 * kill(getpid(), SIGABRT).
1921 	 */
1922 	if (IN_CAPABILITY_MODE(td) && pid != td->td_proc->p_pid)
1923 		return (ECAPMODE);
1924 
1925 	AUDIT_ARG_SIGNUM(signum);
1926 	AUDIT_ARG_PID(pid);
1927 	if ((u_int)signum > _SIG_MAXSIG)
1928 		return (EINVAL);
1929 
1930 	ksiginfo_init(&ksi);
1931 	ksi.ksi_signo = signum;
1932 	ksi.ksi_code = SI_USER;
1933 	ksi.ksi_pid = td->td_proc->p_pid;
1934 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1935 
1936 	if (pid > 0) {
1937 		/* kill single process */
1938 		if ((p = pfind_any(pid)) == NULL)
1939 			return (ESRCH);
1940 		AUDIT_ARG_PROCESS(p);
1941 		error = p_cansignal(td, p, signum);
1942 		if (error == 0 && signum)
1943 			pksignal(p, signum, &ksi);
1944 		PROC_UNLOCK(p);
1945 		return (error);
1946 	}
1947 	switch (pid) {
1948 	case -1:		/* broadcast signal */
1949 		return (killpg1(td, signum, 0, 1, &ksi));
1950 	case 0:			/* signal own process group */
1951 		ksi.ksi_flags |= KSI_KILLPG;
1952 		return (killpg1(td, signum, 0, 0, &ksi));
1953 	default:		/* negative explicit process group */
1954 		ksi.ksi_flags |= KSI_KILLPG;
1955 		return (killpg1(td, signum, -pid, 0, &ksi));
1956 	}
1957 	/* NOTREACHED */
1958 }
1959 
1960 int
1961 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1962 {
1963 	struct proc *p;
1964 	int error;
1965 
1966 	AUDIT_ARG_SIGNUM(uap->signum);
1967 	AUDIT_ARG_FD(uap->fd);
1968 	if ((u_int)uap->signum > _SIG_MAXSIG)
1969 		return (EINVAL);
1970 
1971 	error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1972 	if (error)
1973 		return (error);
1974 	AUDIT_ARG_PROCESS(p);
1975 	error = p_cansignal(td, p, uap->signum);
1976 	if (error == 0 && uap->signum)
1977 		kern_psignal(p, uap->signum);
1978 	PROC_UNLOCK(p);
1979 	return (error);
1980 }
1981 
1982 #if defined(COMPAT_43)
1983 #ifndef _SYS_SYSPROTO_H_
1984 struct okillpg_args {
1985 	int	pgid;
1986 	int	signum;
1987 };
1988 #endif
1989 /* ARGSUSED */
1990 int
1991 okillpg(struct thread *td, struct okillpg_args *uap)
1992 {
1993 	ksiginfo_t ksi;
1994 
1995 	AUDIT_ARG_SIGNUM(uap->signum);
1996 	AUDIT_ARG_PID(uap->pgid);
1997 	if ((u_int)uap->signum > _SIG_MAXSIG)
1998 		return (EINVAL);
1999 
2000 	ksiginfo_init(&ksi);
2001 	ksi.ksi_signo = uap->signum;
2002 	ksi.ksi_code = SI_USER;
2003 	ksi.ksi_pid = td->td_proc->p_pid;
2004 	ksi.ksi_uid = td->td_ucred->cr_ruid;
2005 	ksi.ksi_flags |= KSI_KILLPG;
2006 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
2007 }
2008 #endif /* COMPAT_43 */
2009 
2010 #ifndef _SYS_SYSPROTO_H_
2011 struct sigqueue_args {
2012 	pid_t pid;
2013 	int signum;
2014 	/* union sigval */ void *value;
2015 };
2016 #endif
2017 int
2018 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
2019 {
2020 	union sigval sv;
2021 
2022 	sv.sival_ptr = uap->value;
2023 
2024 	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
2025 }
2026 
2027 int
2028 kern_sigqueue(struct thread *td, pid_t pid, int signum, union sigval *value)
2029 {
2030 	ksiginfo_t ksi;
2031 	struct proc *p;
2032 	int error;
2033 
2034 	if ((u_int)signum > _SIG_MAXSIG)
2035 		return (EINVAL);
2036 
2037 	/*
2038 	 * Specification says sigqueue can only send signal to
2039 	 * single process.
2040 	 */
2041 	if (pid <= 0)
2042 		return (EINVAL);
2043 
2044 	if ((p = pfind_any(pid)) == NULL)
2045 		return (ESRCH);
2046 	error = p_cansignal(td, p, signum);
2047 	if (error == 0 && signum != 0) {
2048 		ksiginfo_init(&ksi);
2049 		ksi.ksi_flags = KSI_SIGQ;
2050 		ksi.ksi_signo = signum;
2051 		ksi.ksi_code = SI_QUEUE;
2052 		ksi.ksi_pid = td->td_proc->p_pid;
2053 		ksi.ksi_uid = td->td_ucred->cr_ruid;
2054 		ksi.ksi_value = *value;
2055 		error = pksignal(p, ksi.ksi_signo, &ksi);
2056 	}
2057 	PROC_UNLOCK(p);
2058 	return (error);
2059 }
2060 
2061 /*
2062  * Send a signal to a process group.  If checktty is 1,
2063  * limit to members which have a controlling terminal.
2064  */
2065 void
2066 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
2067 {
2068 	struct proc *p;
2069 
2070 	if (pgrp) {
2071 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
2072 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
2073 			PROC_LOCK(p);
2074 			if (p->p_state == PRS_NORMAL &&
2075 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
2076 				pksignal(p, sig, ksi);
2077 			PROC_UNLOCK(p);
2078 		}
2079 	}
2080 }
2081 
2082 /*
2083  * Recalculate the signal mask and reset the signal disposition after
2084  * usermode frame for delivery is formed.  Should be called after
2085  * mach-specific routine, because sysent->sv_sendsig() needs correct
2086  * ps_siginfo and signal mask.
2087  */
2088 static void
2089 postsig_done(int sig, struct thread *td, struct sigacts *ps)
2090 {
2091 	sigset_t mask;
2092 
2093 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2094 	td->td_ru.ru_nsignals++;
2095 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
2096 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
2097 		SIGADDSET(mask, sig);
2098 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
2099 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
2100 	if (SIGISMEMBER(ps->ps_sigreset, sig))
2101 		sigdflt(ps, sig);
2102 }
2103 
2104 /*
2105  * Send a signal caused by a trap to the current thread.  If it will be
2106  * caught immediately, deliver it with correct code.  Otherwise, post it
2107  * normally.
2108  */
2109 void
2110 trapsignal(struct thread *td, ksiginfo_t *ksi)
2111 {
2112 	struct sigacts *ps;
2113 	struct proc *p;
2114 	sigset_t sigmask;
2115 	int sig;
2116 
2117 	p = td->td_proc;
2118 	sig = ksi->ksi_signo;
2119 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
2120 
2121 	sigfastblock_fetch(td);
2122 	PROC_LOCK(p);
2123 	ps = p->p_sigacts;
2124 	mtx_lock(&ps->ps_mtx);
2125 	sigmask = td->td_sigmask;
2126 	if (td->td_sigblock_val != 0)
2127 		SIGSETOR(sigmask, fastblock_mask);
2128 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2129 	    !SIGISMEMBER(sigmask, sig)) {
2130 #ifdef KTRACE
2131 		if (KTRPOINT(curthread, KTR_PSIG))
2132 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2133 			    &td->td_sigmask, ksi->ksi_code);
2134 #endif
2135 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2136 		    ksi, &td->td_sigmask);
2137 		postsig_done(sig, td, ps);
2138 		mtx_unlock(&ps->ps_mtx);
2139 	} else {
2140 		/*
2141 		 * Avoid a possible infinite loop if the thread
2142 		 * masking the signal or process is ignoring the
2143 		 * signal.
2144 		 */
2145 		if (kern_forcesigexit && (SIGISMEMBER(sigmask, sig) ||
2146 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2147 			SIGDELSET(td->td_sigmask, sig);
2148 			SIGDELSET(ps->ps_sigcatch, sig);
2149 			SIGDELSET(ps->ps_sigignore, sig);
2150 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2151 			td->td_pflags &= ~TDP_SIGFASTBLOCK;
2152 			td->td_sigblock_val = 0;
2153 		}
2154 		mtx_unlock(&ps->ps_mtx);
2155 		p->p_sig = sig;		/* XXX to verify code */
2156 		tdsendsignal(p, td, sig, ksi);
2157 	}
2158 	PROC_UNLOCK(p);
2159 }
2160 
2161 static struct thread *
2162 sigtd(struct proc *p, int sig, bool fast_sigblock)
2163 {
2164 	struct thread *td, *signal_td;
2165 
2166 	PROC_LOCK_ASSERT(p, MA_OWNED);
2167 	MPASS(!fast_sigblock || p == curproc);
2168 
2169 	/*
2170 	 * Check if current thread can handle the signal without
2171 	 * switching context to another thread.
2172 	 */
2173 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig) &&
2174 	    (!fast_sigblock || curthread->td_sigblock_val == 0))
2175 		return (curthread);
2176 	signal_td = NULL;
2177 	FOREACH_THREAD_IN_PROC(p, td) {
2178 		if (!SIGISMEMBER(td->td_sigmask, sig) && (!fast_sigblock ||
2179 		    td != curthread || td->td_sigblock_val == 0)) {
2180 			signal_td = td;
2181 			break;
2182 		}
2183 	}
2184 	if (signal_td == NULL)
2185 		signal_td = FIRST_THREAD_IN_PROC(p);
2186 	return (signal_td);
2187 }
2188 
2189 /*
2190  * Send the signal to the process.  If the signal has an action, the action
2191  * is usually performed by the target process rather than the caller; we add
2192  * the signal to the set of pending signals for the process.
2193  *
2194  * Exceptions:
2195  *   o When a stop signal is sent to a sleeping process that takes the
2196  *     default action, the process is stopped without awakening it.
2197  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2198  *     regardless of the signal action (eg, blocked or ignored).
2199  *
2200  * Other ignored signals are discarded immediately.
2201  *
2202  * NB: This function may be entered from the debugger via the "kill" DDB
2203  * command.  There is little that can be done to mitigate the possibly messy
2204  * side effects of this unwise possibility.
2205  */
2206 void
2207 kern_psignal(struct proc *p, int sig)
2208 {
2209 	ksiginfo_t ksi;
2210 
2211 	ksiginfo_init(&ksi);
2212 	ksi.ksi_signo = sig;
2213 	ksi.ksi_code = SI_KERNEL;
2214 	(void) tdsendsignal(p, NULL, sig, &ksi);
2215 }
2216 
2217 int
2218 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2219 {
2220 
2221 	return (tdsendsignal(p, NULL, sig, ksi));
2222 }
2223 
2224 /* Utility function for finding a thread to send signal event to. */
2225 int
2226 sigev_findtd(struct proc *p, struct sigevent *sigev, struct thread **ttd)
2227 {
2228 	struct thread *td;
2229 
2230 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2231 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2232 		if (td == NULL)
2233 			return (ESRCH);
2234 		*ttd = td;
2235 	} else {
2236 		*ttd = NULL;
2237 		PROC_LOCK(p);
2238 	}
2239 	return (0);
2240 }
2241 
2242 void
2243 tdsignal(struct thread *td, int sig)
2244 {
2245 	ksiginfo_t ksi;
2246 
2247 	ksiginfo_init(&ksi);
2248 	ksi.ksi_signo = sig;
2249 	ksi.ksi_code = SI_KERNEL;
2250 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2251 }
2252 
2253 void
2254 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2255 {
2256 
2257 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2258 }
2259 
2260 static int
2261 sig_sleepq_abort(struct thread *td, int intrval)
2262 {
2263 	THREAD_LOCK_ASSERT(td, MA_OWNED);
2264 
2265 	if (intrval == 0 && (td->td_flags & TDF_SIGWAIT) == 0) {
2266 		thread_unlock(td);
2267 		return (0);
2268 	}
2269 	return (sleepq_abort(td, intrval));
2270 }
2271 
2272 int
2273 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2274 {
2275 	sig_t action;
2276 	sigqueue_t *sigqueue;
2277 	int prop;
2278 	struct sigacts *ps;
2279 	int intrval;
2280 	int ret = 0;
2281 	int wakeup_swapper;
2282 
2283 	MPASS(td == NULL || p == td->td_proc);
2284 	PROC_LOCK_ASSERT(p, MA_OWNED);
2285 
2286 	if (!_SIG_VALID(sig))
2287 		panic("%s(): invalid signal %d", __func__, sig);
2288 
2289 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2290 
2291 	/*
2292 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2293 	 */
2294 	if (p->p_state == PRS_ZOMBIE) {
2295 		if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2296 			ksiginfo_tryfree(ksi);
2297 		return (ret);
2298 	}
2299 
2300 	ps = p->p_sigacts;
2301 	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2302 	prop = sigprop(sig);
2303 
2304 	if (td == NULL) {
2305 		td = sigtd(p, sig, false);
2306 		sigqueue = &p->p_sigqueue;
2307 	} else
2308 		sigqueue = &td->td_sigqueue;
2309 
2310 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2311 
2312 	/*
2313 	 * If the signal is being ignored, then we forget about it
2314 	 * immediately, except when the target process executes
2315 	 * sigwait().  (Note: we don't set SIGCONT in ps_sigignore,
2316 	 * and if it is set to SIG_IGN, action will be SIG_DFL here.)
2317 	 */
2318 	mtx_lock(&ps->ps_mtx);
2319 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2320 		if (kern_sig_discard_ign &&
2321 		    (p->p_sysent->sv_flags & SV_SIG_DISCIGN) == 0) {
2322 			SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2323 
2324 			mtx_unlock(&ps->ps_mtx);
2325 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2326 				ksiginfo_tryfree(ksi);
2327 			return (ret);
2328 		} else {
2329 			action = SIG_CATCH;
2330 			intrval = 0;
2331 		}
2332 	} else {
2333 		if (SIGISMEMBER(td->td_sigmask, sig))
2334 			action = SIG_HOLD;
2335 		else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2336 			action = SIG_CATCH;
2337 		else
2338 			action = SIG_DFL;
2339 		if (SIGISMEMBER(ps->ps_sigintr, sig))
2340 			intrval = EINTR;
2341 		else
2342 			intrval = ERESTART;
2343 	}
2344 	mtx_unlock(&ps->ps_mtx);
2345 
2346 	if (prop & SIGPROP_CONT)
2347 		sigqueue_delete_stopmask_proc(p);
2348 	else if (prop & SIGPROP_STOP) {
2349 		/*
2350 		 * If sending a tty stop signal to a member of an orphaned
2351 		 * process group, discard the signal here if the action
2352 		 * is default; don't stop the process below if sleeping,
2353 		 * and don't clear any pending SIGCONT.
2354 		 */
2355 		if ((prop & SIGPROP_TTYSTOP) != 0 &&
2356 		    (p->p_pgrp->pg_flags & PGRP_ORPHANED) != 0 &&
2357 		    action == SIG_DFL) {
2358 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2359 				ksiginfo_tryfree(ksi);
2360 			return (ret);
2361 		}
2362 		sigqueue_delete_proc(p, SIGCONT);
2363 		if (p->p_flag & P_CONTINUED) {
2364 			p->p_flag &= ~P_CONTINUED;
2365 			PROC_LOCK(p->p_pptr);
2366 			sigqueue_take(p->p_ksi);
2367 			PROC_UNLOCK(p->p_pptr);
2368 		}
2369 	}
2370 
2371 	ret = sigqueue_add(sigqueue, sig, ksi);
2372 	if (ret != 0)
2373 		return (ret);
2374 	if ((ksi->ksi_flags & KSI_KILLPG) != 0) {
2375 		sx_assert(&p->p_pgrp->pg_killsx, SX_XLOCKED);
2376 		atomic_add_int(&p->p_killpg_cnt, 1);
2377 	}
2378 	signotify(td);
2379 	/*
2380 	 * Defer further processing for signals which are held,
2381 	 * except that stopped processes must be continued by SIGCONT.
2382 	 */
2383 	if (action == SIG_HOLD &&
2384 	    !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2385 		return (ret);
2386 
2387 	wakeup_swapper = 0;
2388 
2389 	/*
2390 	 * Some signals have a process-wide effect and a per-thread
2391 	 * component.  Most processing occurs when the process next
2392 	 * tries to cross the user boundary, however there are some
2393 	 * times when processing needs to be done immediately, such as
2394 	 * waking up threads so that they can cross the user boundary.
2395 	 * We try to do the per-process part here.
2396 	 */
2397 	if (P_SHOULDSTOP(p)) {
2398 		KASSERT(!(p->p_flag & P_WEXIT),
2399 		    ("signal to stopped but exiting process"));
2400 		if (sig == SIGKILL) {
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 			 * SIGKILL sets process running.
2409 			 * It will die elsewhere.
2410 			 * All threads must be restarted.
2411 			 */
2412 			p->p_flag &= ~P_STOPPED_SIG;
2413 			goto runfast;
2414 		}
2415 
2416 		if (prop & SIGPROP_CONT) {
2417 			/*
2418 			 * If traced process is already stopped,
2419 			 * then no further action is necessary.
2420 			 */
2421 			if (p->p_flag & P_TRACED)
2422 				goto out;
2423 			/*
2424 			 * If SIGCONT is default (or ignored), we continue the
2425 			 * process but don't leave the signal in sigqueue as
2426 			 * it has no further action.  If SIGCONT is held, we
2427 			 * continue the process and leave the signal in
2428 			 * sigqueue.  If the process catches SIGCONT, let it
2429 			 * handle the signal itself.  If it isn't waiting on
2430 			 * an event, it goes back to run state.
2431 			 * Otherwise, process goes back to sleep state.
2432 			 */
2433 			p->p_flag &= ~P_STOPPED_SIG;
2434 			PROC_SLOCK(p);
2435 			if (p->p_numthreads == p->p_suspcount) {
2436 				PROC_SUNLOCK(p);
2437 				p->p_flag |= P_CONTINUED;
2438 				p->p_xsig = SIGCONT;
2439 				PROC_LOCK(p->p_pptr);
2440 				childproc_continued(p);
2441 				PROC_UNLOCK(p->p_pptr);
2442 				PROC_SLOCK(p);
2443 			}
2444 			if (action == SIG_DFL) {
2445 				thread_unsuspend(p);
2446 				PROC_SUNLOCK(p);
2447 				sigqueue_delete(sigqueue, sig);
2448 				goto out_cont;
2449 			}
2450 			if (action == SIG_CATCH) {
2451 				/*
2452 				 * The process wants to catch it so it needs
2453 				 * to run at least one thread, but which one?
2454 				 */
2455 				PROC_SUNLOCK(p);
2456 				goto runfast;
2457 			}
2458 			/*
2459 			 * The signal is not ignored or caught.
2460 			 */
2461 			thread_unsuspend(p);
2462 			PROC_SUNLOCK(p);
2463 			goto out_cont;
2464 		}
2465 
2466 		if (prop & SIGPROP_STOP) {
2467 			/*
2468 			 * If traced process is already stopped,
2469 			 * then no further action is necessary.
2470 			 */
2471 			if (p->p_flag & P_TRACED)
2472 				goto out;
2473 			/*
2474 			 * Already stopped, don't need to stop again
2475 			 * (If we did the shell could get confused).
2476 			 * Just make sure the signal STOP bit set.
2477 			 */
2478 			p->p_flag |= P_STOPPED_SIG;
2479 			sigqueue_delete(sigqueue, sig);
2480 			goto out;
2481 		}
2482 
2483 		/*
2484 		 * All other kinds of signals:
2485 		 * If a thread is sleeping interruptibly, simulate a
2486 		 * wakeup so that when it is continued it will be made
2487 		 * runnable and can look at the signal.  However, don't make
2488 		 * the PROCESS runnable, leave it stopped.
2489 		 * It may run a bit until it hits a thread_suspend_check().
2490 		 */
2491 		PROC_SLOCK(p);
2492 		thread_lock(td);
2493 		if (TD_CAN_ABORT(td))
2494 			wakeup_swapper = sig_sleepq_abort(td, intrval);
2495 		else
2496 			thread_unlock(td);
2497 		PROC_SUNLOCK(p);
2498 		goto out;
2499 		/*
2500 		 * Mutexes are short lived. Threads waiting on them will
2501 		 * hit thread_suspend_check() soon.
2502 		 */
2503 	} else if (p->p_state == PRS_NORMAL) {
2504 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2505 			tdsigwakeup(td, sig, action, intrval);
2506 			goto out;
2507 		}
2508 
2509 		MPASS(action == SIG_DFL);
2510 
2511 		if (prop & SIGPROP_STOP) {
2512 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2513 				goto out;
2514 			p->p_flag |= P_STOPPED_SIG;
2515 			p->p_xsig = sig;
2516 			PROC_SLOCK(p);
2517 			wakeup_swapper = sig_suspend_threads(td, p);
2518 			if (p->p_numthreads == p->p_suspcount) {
2519 				/*
2520 				 * only thread sending signal to another
2521 				 * process can reach here, if thread is sending
2522 				 * signal to its process, because thread does
2523 				 * not suspend itself here, p_numthreads
2524 				 * should never be equal to p_suspcount.
2525 				 */
2526 				thread_stopped(p);
2527 				PROC_SUNLOCK(p);
2528 				sigqueue_delete_proc(p, p->p_xsig);
2529 			} else
2530 				PROC_SUNLOCK(p);
2531 			goto out;
2532 		}
2533 	} else {
2534 		/* Not in "NORMAL" state. discard the signal. */
2535 		sigqueue_delete(sigqueue, sig);
2536 		goto out;
2537 	}
2538 
2539 	/*
2540 	 * The process is not stopped so we need to apply the signal to all the
2541 	 * running threads.
2542 	 */
2543 runfast:
2544 	tdsigwakeup(td, sig, action, intrval);
2545 	PROC_SLOCK(p);
2546 	thread_unsuspend(p);
2547 	PROC_SUNLOCK(p);
2548 out_cont:
2549 	itimer_proc_continue(p);
2550 	kqtimer_proc_continue(p);
2551 out:
2552 	/* If we jump here, proc slock should not be owned. */
2553 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2554 	if (wakeup_swapper)
2555 		kick_proc0();
2556 
2557 	return (ret);
2558 }
2559 
2560 /*
2561  * The force of a signal has been directed against a single
2562  * thread.  We need to see what we can do about knocking it
2563  * out of any sleep it may be in etc.
2564  */
2565 static void
2566 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2567 {
2568 	struct proc *p = td->td_proc;
2569 	int prop, wakeup_swapper;
2570 
2571 	PROC_LOCK_ASSERT(p, MA_OWNED);
2572 	prop = sigprop(sig);
2573 
2574 	PROC_SLOCK(p);
2575 	thread_lock(td);
2576 	/*
2577 	 * Bring the priority of a thread up if we want it to get
2578 	 * killed in this lifetime.  Be careful to avoid bumping the
2579 	 * priority of the idle thread, since we still allow to signal
2580 	 * kernel processes.
2581 	 */
2582 	if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2583 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2584 		sched_prio(td, PUSER);
2585 	if (TD_ON_SLEEPQ(td)) {
2586 		/*
2587 		 * If thread is sleeping uninterruptibly
2588 		 * we can't interrupt the sleep... the signal will
2589 		 * be noticed when the process returns through
2590 		 * trap() or syscall().
2591 		 */
2592 		if ((td->td_flags & TDF_SINTR) == 0)
2593 			goto out;
2594 		/*
2595 		 * If SIGCONT is default (or ignored) and process is
2596 		 * asleep, we are finished; the process should not
2597 		 * be awakened.
2598 		 */
2599 		if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2600 			thread_unlock(td);
2601 			PROC_SUNLOCK(p);
2602 			sigqueue_delete(&p->p_sigqueue, sig);
2603 			/*
2604 			 * It may be on either list in this state.
2605 			 * Remove from both for now.
2606 			 */
2607 			sigqueue_delete(&td->td_sigqueue, sig);
2608 			return;
2609 		}
2610 
2611 		/*
2612 		 * Don't awaken a sleeping thread for SIGSTOP if the
2613 		 * STOP signal is deferred.
2614 		 */
2615 		if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2616 		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2617 			goto out;
2618 
2619 		/*
2620 		 * Give low priority threads a better chance to run.
2621 		 */
2622 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2623 			sched_prio(td, PUSER);
2624 
2625 		wakeup_swapper = sig_sleepq_abort(td, intrval);
2626 		PROC_SUNLOCK(p);
2627 		if (wakeup_swapper)
2628 			kick_proc0();
2629 		return;
2630 	}
2631 
2632 	/*
2633 	 * Other states do nothing with the signal immediately,
2634 	 * other than kicking ourselves if we are running.
2635 	 * It will either never be noticed, or noticed very soon.
2636 	 */
2637 #ifdef SMP
2638 	if (TD_IS_RUNNING(td) && td != curthread)
2639 		forward_signal(td);
2640 #endif
2641 
2642 out:
2643 	PROC_SUNLOCK(p);
2644 	thread_unlock(td);
2645 }
2646 
2647 static void
2648 ptrace_coredumpreq(struct thread *td, struct proc *p,
2649     struct thr_coredump_req *tcq)
2650 {
2651 	void *rl_cookie;
2652 
2653 	if (p->p_sysent->sv_coredump == NULL) {
2654 		tcq->tc_error = ENOSYS;
2655 		return;
2656 	}
2657 
2658 	rl_cookie = vn_rangelock_wlock(tcq->tc_vp, 0, OFF_MAX);
2659 	tcq->tc_error = p->p_sysent->sv_coredump(td, tcq->tc_vp,
2660 	    tcq->tc_limit, tcq->tc_flags);
2661 	vn_rangelock_unlock(tcq->tc_vp, rl_cookie);
2662 }
2663 
2664 static void
2665 ptrace_syscallreq(struct thread *td, struct proc *p,
2666     struct thr_syscall_req *tsr)
2667 {
2668 	struct sysentvec *sv;
2669 	struct sysent *se;
2670 	register_t rv_saved[2];
2671 	int error, nerror;
2672 	int sc;
2673 	bool audited, sy_thr_static;
2674 
2675 	sv = p->p_sysent;
2676 	if (sv->sv_table == NULL || sv->sv_size < tsr->ts_sa.code) {
2677 		tsr->ts_ret.sr_error = ENOSYS;
2678 		return;
2679 	}
2680 
2681 	sc = tsr->ts_sa.code;
2682 	if (sc == SYS_syscall || sc == SYS___syscall) {
2683 		sc = tsr->ts_sa.args[0];
2684 		memmove(&tsr->ts_sa.args[0], &tsr->ts_sa.args[1],
2685 		    sizeof(register_t) * (tsr->ts_nargs - 1));
2686 	}
2687 
2688 	tsr->ts_sa.callp = se = &sv->sv_table[sc];
2689 
2690 	VM_CNT_INC(v_syscall);
2691 	td->td_pticks = 0;
2692 	if (__predict_false(td->td_cowgen != atomic_load_int(
2693 	    &td->td_proc->p_cowgen)))
2694 		thread_cow_update(td);
2695 
2696 #ifdef CAPABILITY_MODE
2697 	if (IN_CAPABILITY_MODE(td) && (se->sy_flags & SYF_CAPENABLED) == 0) {
2698 		tsr->ts_ret.sr_error = ECAPMODE;
2699 		return;
2700 	}
2701 #endif
2702 
2703 	sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2704 	audited = AUDIT_SYSCALL_ENTER(sc, td) != 0;
2705 
2706 	if (!sy_thr_static) {
2707 		error = syscall_thread_enter(td, se);
2708 		if (error != 0) {
2709 			tsr->ts_ret.sr_error = error;
2710 			return;
2711 		}
2712 	}
2713 
2714 	rv_saved[0] = td->td_retval[0];
2715 	rv_saved[1] = td->td_retval[1];
2716 	nerror = td->td_errno;
2717 	td->td_retval[0] = 0;
2718 	td->td_retval[1] = 0;
2719 
2720 #ifdef KDTRACE_HOOKS
2721 	if (se->sy_entry != 0)
2722 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_ENTRY, 0);
2723 #endif
2724 	tsr->ts_ret.sr_error = se->sy_call(td, tsr->ts_sa.args);
2725 #ifdef KDTRACE_HOOKS
2726 	if (se->sy_return != 0)
2727 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_RETURN,
2728 		    tsr->ts_ret.sr_error != 0 ? -1 : td->td_retval[0]);
2729 #endif
2730 
2731 	tsr->ts_ret.sr_retval[0] = td->td_retval[0];
2732 	tsr->ts_ret.sr_retval[1] = td->td_retval[1];
2733 	td->td_retval[0] = rv_saved[0];
2734 	td->td_retval[1] = rv_saved[1];
2735 	td->td_errno = nerror;
2736 
2737 	if (audited)
2738 		AUDIT_SYSCALL_EXIT(error, td);
2739 	if (!sy_thr_static)
2740 		syscall_thread_exit(td, se);
2741 }
2742 
2743 static void
2744 ptrace_remotereq(struct thread *td, int flag)
2745 {
2746 	struct proc *p;
2747 
2748 	MPASS(td == curthread);
2749 	p = td->td_proc;
2750 	PROC_LOCK_ASSERT(p, MA_OWNED);
2751 	if ((td->td_dbgflags & flag) == 0)
2752 		return;
2753 	KASSERT((p->p_flag & P_STOPPED_TRACE) != 0, ("not stopped"));
2754 	KASSERT(td->td_remotereq != NULL, ("td_remotereq is NULL"));
2755 
2756 	PROC_UNLOCK(p);
2757 	switch (flag) {
2758 	case TDB_COREDUMPREQ:
2759 		ptrace_coredumpreq(td, p, td->td_remotereq);
2760 		break;
2761 	case TDB_SCREMOTEREQ:
2762 		ptrace_syscallreq(td, p, td->td_remotereq);
2763 		break;
2764 	default:
2765 		__unreachable();
2766 	}
2767 	PROC_LOCK(p);
2768 
2769 	MPASS((td->td_dbgflags & flag) != 0);
2770 	td->td_dbgflags &= ~flag;
2771 	td->td_remotereq = NULL;
2772 	wakeup(p);
2773 }
2774 
2775 static int
2776 sig_suspend_threads(struct thread *td, struct proc *p)
2777 {
2778 	struct thread *td2;
2779 	int wakeup_swapper;
2780 
2781 	PROC_LOCK_ASSERT(p, MA_OWNED);
2782 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2783 
2784 	wakeup_swapper = 0;
2785 	FOREACH_THREAD_IN_PROC(p, td2) {
2786 		thread_lock(td2);
2787 		ast_sched_locked(td2, TDA_SUSPEND);
2788 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2789 		    (td2->td_flags & TDF_SINTR)) {
2790 			if (td2->td_flags & TDF_SBDRY) {
2791 				/*
2792 				 * Once a thread is asleep with
2793 				 * TDF_SBDRY and without TDF_SERESTART
2794 				 * or TDF_SEINTR set, it should never
2795 				 * become suspended due to this check.
2796 				 */
2797 				KASSERT(!TD_IS_SUSPENDED(td2),
2798 				    ("thread with deferred stops suspended"));
2799 				if (TD_SBDRY_INTR(td2)) {
2800 					wakeup_swapper |= sleepq_abort(td2,
2801 					    TD_SBDRY_ERRNO(td2));
2802 					continue;
2803 				}
2804 			} else if (!TD_IS_SUSPENDED(td2))
2805 				thread_suspend_one(td2);
2806 		} else if (!TD_IS_SUSPENDED(td2)) {
2807 #ifdef SMP
2808 			if (TD_IS_RUNNING(td2) && td2 != td)
2809 				forward_signal(td2);
2810 #endif
2811 		}
2812 		thread_unlock(td2);
2813 	}
2814 	return (wakeup_swapper);
2815 }
2816 
2817 /*
2818  * Stop the process for an event deemed interesting to the debugger. If si is
2819  * non-NULL, this is a signal exchange; the new signal requested by the
2820  * debugger will be returned for handling. If si is NULL, this is some other
2821  * type of interesting event. The debugger may request a signal be delivered in
2822  * that case as well, however it will be deferred until it can be handled.
2823  */
2824 int
2825 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2826 {
2827 	struct proc *p = td->td_proc;
2828 	struct thread *td2;
2829 	ksiginfo_t ksi;
2830 
2831 	PROC_LOCK_ASSERT(p, MA_OWNED);
2832 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2833 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2834 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2835 
2836 	td->td_xsig = sig;
2837 
2838 	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2839 		td->td_dbgflags |= TDB_XSIG;
2840 		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2841 		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2842 		PROC_SLOCK(p);
2843 		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2844 			if (P_KILLED(p)) {
2845 				/*
2846 				 * Ensure that, if we've been PT_KILLed, the
2847 				 * exit status reflects that. Another thread
2848 				 * may also be in ptracestop(), having just
2849 				 * received the SIGKILL, but this thread was
2850 				 * unsuspended first.
2851 				 */
2852 				td->td_dbgflags &= ~TDB_XSIG;
2853 				td->td_xsig = SIGKILL;
2854 				p->p_ptevents = 0;
2855 				break;
2856 			}
2857 			if (p->p_flag & P_SINGLE_EXIT &&
2858 			    !(td->td_dbgflags & TDB_EXIT)) {
2859 				/*
2860 				 * Ignore ptrace stops except for thread exit
2861 				 * events when the process exits.
2862 				 */
2863 				td->td_dbgflags &= ~TDB_XSIG;
2864 				PROC_SUNLOCK(p);
2865 				return (0);
2866 			}
2867 
2868 			/*
2869 			 * Make wait(2) work.  Ensure that right after the
2870 			 * attach, the thread which was decided to become the
2871 			 * leader of attach gets reported to the waiter.
2872 			 * Otherwise, just avoid overwriting another thread's
2873 			 * assignment to p_xthread.  If another thread has
2874 			 * already set p_xthread, the current thread will get
2875 			 * a chance to report itself upon the next iteration.
2876 			 */
2877 			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2878 			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2879 			    p->p_xthread == NULL)) {
2880 				p->p_xsig = sig;
2881 				p->p_xthread = td;
2882 
2883 				/*
2884 				 * If we are on sleepqueue already,
2885 				 * let sleepqueue code decide if it
2886 				 * needs to go sleep after attach.
2887 				 */
2888 				if (td->td_wchan == NULL)
2889 					td->td_dbgflags &= ~TDB_FSTP;
2890 
2891 				p->p_flag2 &= ~P2_PTRACE_FSTP;
2892 				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2893 				sig_suspend_threads(td, p);
2894 			}
2895 			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2896 				td->td_dbgflags &= ~TDB_STOPATFORK;
2897 			}
2898 stopme:
2899 			td->td_dbgflags |= TDB_SSWITCH;
2900 			thread_suspend_switch(td, p);
2901 			td->td_dbgflags &= ~TDB_SSWITCH;
2902 			if ((td->td_dbgflags & (TDB_COREDUMPREQ |
2903 			    TDB_SCREMOTEREQ)) != 0) {
2904 				MPASS((td->td_dbgflags & (TDB_COREDUMPREQ |
2905 				    TDB_SCREMOTEREQ)) !=
2906 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2907 				PROC_SUNLOCK(p);
2908 				ptrace_remotereq(td, td->td_dbgflags &
2909 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2910 				PROC_SLOCK(p);
2911 				goto stopme;
2912 			}
2913 			if (p->p_xthread == td)
2914 				p->p_xthread = NULL;
2915 			if (!(p->p_flag & P_TRACED))
2916 				break;
2917 			if (td->td_dbgflags & TDB_SUSPEND) {
2918 				if (p->p_flag & P_SINGLE_EXIT)
2919 					break;
2920 				goto stopme;
2921 			}
2922 		}
2923 		PROC_SUNLOCK(p);
2924 	}
2925 
2926 	if (si != NULL && sig == td->td_xsig) {
2927 		/* Parent wants us to take the original signal unchanged. */
2928 		si->ksi_flags |= KSI_HEAD;
2929 		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2930 			si->ksi_signo = 0;
2931 	} else if (td->td_xsig != 0) {
2932 		/*
2933 		 * If parent wants us to take a new signal, then it will leave
2934 		 * it in td->td_xsig; otherwise we just look for signals again.
2935 		 */
2936 		ksiginfo_init(&ksi);
2937 		ksi.ksi_signo = td->td_xsig;
2938 		ksi.ksi_flags |= KSI_PTRACE;
2939 		td2 = sigtd(p, td->td_xsig, false);
2940 		tdsendsignal(p, td2, td->td_xsig, &ksi);
2941 		if (td != td2)
2942 			return (0);
2943 	}
2944 
2945 	return (td->td_xsig);
2946 }
2947 
2948 static void
2949 reschedule_signals(struct proc *p, sigset_t block, int flags)
2950 {
2951 	struct sigacts *ps;
2952 	struct thread *td;
2953 	int sig;
2954 	bool fastblk, pslocked;
2955 
2956 	PROC_LOCK_ASSERT(p, MA_OWNED);
2957 	ps = p->p_sigacts;
2958 	pslocked = (flags & SIGPROCMASK_PS_LOCKED) != 0;
2959 	mtx_assert(&ps->ps_mtx, pslocked ? MA_OWNED : MA_NOTOWNED);
2960 	if (SIGISEMPTY(p->p_siglist))
2961 		return;
2962 	SIGSETAND(block, p->p_siglist);
2963 	fastblk = (flags & SIGPROCMASK_FASTBLK) != 0;
2964 	SIG_FOREACH(sig, &block) {
2965 		td = sigtd(p, sig, fastblk);
2966 
2967 		/*
2968 		 * If sigtd() selected us despite sigfastblock is
2969 		 * blocking, do not activate AST or wake us, to avoid
2970 		 * loop in AST handler.
2971 		 */
2972 		if (fastblk && td == curthread)
2973 			continue;
2974 
2975 		signotify(td);
2976 		if (!pslocked)
2977 			mtx_lock(&ps->ps_mtx);
2978 		if (p->p_flag & P_TRACED ||
2979 		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2980 		    !SIGISMEMBER(td->td_sigmask, sig))) {
2981 			tdsigwakeup(td, sig, SIG_CATCH,
2982 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2983 			    ERESTART));
2984 		}
2985 		if (!pslocked)
2986 			mtx_unlock(&ps->ps_mtx);
2987 	}
2988 }
2989 
2990 void
2991 tdsigcleanup(struct thread *td)
2992 {
2993 	struct proc *p;
2994 	sigset_t unblocked;
2995 
2996 	p = td->td_proc;
2997 	PROC_LOCK_ASSERT(p, MA_OWNED);
2998 
2999 	sigqueue_flush(&td->td_sigqueue);
3000 	if (p->p_numthreads == 1)
3001 		return;
3002 
3003 	/*
3004 	 * Since we cannot handle signals, notify signal post code
3005 	 * about this by filling the sigmask.
3006 	 *
3007 	 * Also, if needed, wake up thread(s) that do not block the
3008 	 * same signals as the exiting thread, since the thread might
3009 	 * have been selected for delivery and woken up.
3010 	 */
3011 	SIGFILLSET(unblocked);
3012 	SIGSETNAND(unblocked, td->td_sigmask);
3013 	SIGFILLSET(td->td_sigmask);
3014 	reschedule_signals(p, unblocked, 0);
3015 
3016 }
3017 
3018 static int
3019 sigdeferstop_curr_flags(int cflags)
3020 {
3021 
3022 	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
3023 	    (cflags & TDF_SBDRY) != 0);
3024 	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
3025 }
3026 
3027 /*
3028  * Defer the delivery of SIGSTOP for the current thread, according to
3029  * the requested mode.  Returns previous flags, which must be restored
3030  * by sigallowstop().
3031  *
3032  * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
3033  * cleared by the current thread, which allow the lock-less read-only
3034  * accesses below.
3035  */
3036 int
3037 sigdeferstop_impl(int mode)
3038 {
3039 	struct thread *td;
3040 	int cflags, nflags;
3041 
3042 	td = curthread;
3043 	cflags = sigdeferstop_curr_flags(td->td_flags);
3044 	switch (mode) {
3045 	case SIGDEFERSTOP_NOP:
3046 		nflags = cflags;
3047 		break;
3048 	case SIGDEFERSTOP_OFF:
3049 		nflags = 0;
3050 		break;
3051 	case SIGDEFERSTOP_SILENT:
3052 		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
3053 		break;
3054 	case SIGDEFERSTOP_EINTR:
3055 		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
3056 		break;
3057 	case SIGDEFERSTOP_ERESTART:
3058 		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
3059 		break;
3060 	default:
3061 		panic("sigdeferstop: invalid mode %x", mode);
3062 		break;
3063 	}
3064 	if (cflags == nflags)
3065 		return (SIGDEFERSTOP_VAL_NCHG);
3066 	thread_lock(td);
3067 	td->td_flags = (td->td_flags & ~cflags) | nflags;
3068 	thread_unlock(td);
3069 	return (cflags);
3070 }
3071 
3072 /*
3073  * Restores the STOP handling mode, typically permitting the delivery
3074  * of SIGSTOP for the current thread.  This does not immediately
3075  * suspend if a stop was posted.  Instead, the thread will suspend
3076  * either via ast() or a subsequent interruptible sleep.
3077  */
3078 void
3079 sigallowstop_impl(int prev)
3080 {
3081 	struct thread *td;
3082 	int cflags;
3083 
3084 	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
3085 	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
3086 	    ("sigallowstop: incorrect previous mode %x", prev));
3087 	td = curthread;
3088 	cflags = sigdeferstop_curr_flags(td->td_flags);
3089 	if (cflags != prev) {
3090 		thread_lock(td);
3091 		td->td_flags = (td->td_flags & ~cflags) | prev;
3092 		thread_unlock(td);
3093 	}
3094 }
3095 
3096 enum sigstatus {
3097 	SIGSTATUS_HANDLE,
3098 	SIGSTATUS_HANDLED,
3099 	SIGSTATUS_IGNORE,
3100 	SIGSTATUS_SBDRY_STOP,
3101 };
3102 
3103 /*
3104  * The thread has signal "sig" pending.  Figure out what to do with it:
3105  *
3106  * _HANDLE     -> the caller should handle the signal
3107  * _HANDLED    -> handled internally, reload pending signal set
3108  * _IGNORE     -> ignored, remove from the set of pending signals and try the
3109  *                next pending signal
3110  * _SBDRY_STOP -> the signal should stop the thread but this is not
3111  *                permitted in the current context
3112  */
3113 static enum sigstatus
3114 sigprocess(struct thread *td, int sig)
3115 {
3116 	struct proc *p;
3117 	struct sigacts *ps;
3118 	struct sigqueue *queue;
3119 	ksiginfo_t ksi;
3120 	int prop;
3121 
3122 	KASSERT(_SIG_VALID(sig), ("%s: invalid signal %d", __func__, sig));
3123 
3124 	p = td->td_proc;
3125 	ps = p->p_sigacts;
3126 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3127 	PROC_LOCK_ASSERT(p, MA_OWNED);
3128 
3129 	/*
3130 	 * We should allow pending but ignored signals below
3131 	 * if there is sigwait() active, or P_TRACED was
3132 	 * on when they were posted.
3133 	 */
3134 	if (SIGISMEMBER(ps->ps_sigignore, sig) &&
3135 	    (p->p_flag & P_TRACED) == 0 &&
3136 	    (td->td_flags & TDF_SIGWAIT) == 0) {
3137 		return (SIGSTATUS_IGNORE);
3138 	}
3139 
3140 	/*
3141 	 * If the process is going to single-thread mode to prepare
3142 	 * for exit, there is no sense in delivering any signal
3143 	 * to usermode.  Another important consequence is that
3144 	 * msleep(..., PCATCH, ...) now is only interruptible by a
3145 	 * suspend request.
3146 	 */
3147 	if ((p->p_flag2 & P2_WEXIT) != 0)
3148 		return (SIGSTATUS_IGNORE);
3149 
3150 	if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
3151 		/*
3152 		 * If traced, always stop.
3153 		 * Remove old signal from queue before the stop.
3154 		 * XXX shrug off debugger, it causes siginfo to
3155 		 * be thrown away.
3156 		 */
3157 		queue = &td->td_sigqueue;
3158 		ksiginfo_init(&ksi);
3159 		if (sigqueue_get(queue, sig, &ksi) == 0) {
3160 			queue = &p->p_sigqueue;
3161 			sigqueue_get(queue, sig, &ksi);
3162 		}
3163 		td->td_si = ksi.ksi_info;
3164 
3165 		mtx_unlock(&ps->ps_mtx);
3166 		sig = ptracestop(td, sig, &ksi);
3167 		mtx_lock(&ps->ps_mtx);
3168 
3169 		td->td_si.si_signo = 0;
3170 
3171 		/*
3172 		 * Keep looking if the debugger discarded or
3173 		 * replaced the signal.
3174 		 */
3175 		if (sig == 0)
3176 			return (SIGSTATUS_HANDLED);
3177 
3178 		/*
3179 		 * If the signal became masked, re-queue it.
3180 		 */
3181 		if (SIGISMEMBER(td->td_sigmask, sig)) {
3182 			ksi.ksi_flags |= KSI_HEAD;
3183 			sigqueue_add(&p->p_sigqueue, sig, &ksi);
3184 			return (SIGSTATUS_HANDLED);
3185 		}
3186 
3187 		/*
3188 		 * If the traced bit got turned off, requeue the signal and
3189 		 * reload the set of pending signals.  This ensures that p_sig*
3190 		 * and p_sigact are consistent.
3191 		 */
3192 		if ((p->p_flag & P_TRACED) == 0) {
3193 			if ((ksi.ksi_flags & KSI_PTRACE) == 0) {
3194 				ksi.ksi_flags |= KSI_HEAD;
3195 				sigqueue_add(queue, sig, &ksi);
3196 			}
3197 			return (SIGSTATUS_HANDLED);
3198 		}
3199 	}
3200 
3201 	/*
3202 	 * Decide whether the signal should be returned.
3203 	 * Return the signal's number, or fall through
3204 	 * to clear it from the pending mask.
3205 	 */
3206 	switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
3207 	case (intptr_t)SIG_DFL:
3208 		/*
3209 		 * Don't take default actions on system processes.
3210 		 */
3211 		if (p->p_pid <= 1) {
3212 #ifdef DIAGNOSTIC
3213 			/*
3214 			 * Are you sure you want to ignore SIGSEGV
3215 			 * in init? XXX
3216 			 */
3217 			printf("Process (pid %lu) got signal %d\n",
3218 				(u_long)p->p_pid, sig);
3219 #endif
3220 			return (SIGSTATUS_IGNORE);
3221 		}
3222 
3223 		/*
3224 		 * If there is a pending stop signal to process with
3225 		 * default action, stop here, then clear the signal.
3226 		 * Traced or exiting processes should ignore stops.
3227 		 * Additionally, a member of an orphaned process group
3228 		 * should ignore tty stops.
3229 		 */
3230 		prop = sigprop(sig);
3231 		if (prop & SIGPROP_STOP) {
3232 			mtx_unlock(&ps->ps_mtx);
3233 			if ((p->p_flag & (P_TRACED | P_WEXIT |
3234 			    P_SINGLE_EXIT)) != 0 || ((p->p_pgrp->
3235 			    pg_flags & PGRP_ORPHANED) != 0 &&
3236 			    (prop & SIGPROP_TTYSTOP) != 0)) {
3237 				mtx_lock(&ps->ps_mtx);
3238 				return (SIGSTATUS_IGNORE);
3239 			}
3240 			if (TD_SBDRY_INTR(td)) {
3241 				KASSERT((td->td_flags & TDF_SBDRY) != 0,
3242 				    ("lost TDF_SBDRY"));
3243 				mtx_lock(&ps->ps_mtx);
3244 				return (SIGSTATUS_SBDRY_STOP);
3245 			}
3246 			WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
3247 			    &p->p_mtx.lock_object, "Catching SIGSTOP");
3248 			sigqueue_delete(&td->td_sigqueue, sig);
3249 			sigqueue_delete(&p->p_sigqueue, sig);
3250 			p->p_flag |= P_STOPPED_SIG;
3251 			p->p_xsig = sig;
3252 			PROC_SLOCK(p);
3253 			sig_suspend_threads(td, p);
3254 			thread_suspend_switch(td, p);
3255 			PROC_SUNLOCK(p);
3256 			mtx_lock(&ps->ps_mtx);
3257 			return (SIGSTATUS_HANDLED);
3258 		} else if ((prop & SIGPROP_IGNORE) != 0 &&
3259 		    (td->td_flags & TDF_SIGWAIT) == 0) {
3260 			/*
3261 			 * Default action is to ignore; drop it if
3262 			 * not in kern_sigtimedwait().
3263 			 */
3264 			return (SIGSTATUS_IGNORE);
3265 		} else {
3266 			return (SIGSTATUS_HANDLE);
3267 		}
3268 
3269 	case (intptr_t)SIG_IGN:
3270 		if ((td->td_flags & TDF_SIGWAIT) == 0)
3271 			return (SIGSTATUS_IGNORE);
3272 		else
3273 			return (SIGSTATUS_HANDLE);
3274 
3275 	default:
3276 		/*
3277 		 * This signal has an action, let postsig() process it.
3278 		 */
3279 		return (SIGSTATUS_HANDLE);
3280 	}
3281 }
3282 
3283 /*
3284  * If the current process has received a signal (should be caught or cause
3285  * termination, should interrupt current syscall), return the signal number.
3286  * Stop signals with default action are processed immediately, then cleared;
3287  * they aren't returned.  This is checked after each entry to the system for
3288  * a syscall or trap (though this can usually be done without calling
3289  * issignal by checking the pending signal masks in cursig.) The normal call
3290  * sequence is
3291  *
3292  *	while (sig = cursig(curthread))
3293  *		postsig(sig);
3294  */
3295 static int
3296 issignal(struct thread *td)
3297 {
3298 	struct proc *p;
3299 	sigset_t sigpending;
3300 	int sig;
3301 
3302 	p = td->td_proc;
3303 	PROC_LOCK_ASSERT(p, MA_OWNED);
3304 
3305 	for (;;) {
3306 		sigpending = td->td_sigqueue.sq_signals;
3307 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
3308 		SIGSETNAND(sigpending, td->td_sigmask);
3309 
3310 		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
3311 		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
3312 			SIG_STOPSIGMASK(sigpending);
3313 		if (SIGISEMPTY(sigpending))	/* no signal to send */
3314 			return (0);
3315 
3316 		/*
3317 		 * Do fast sigblock if requested by usermode.  Since
3318 		 * we do know that there was a signal pending at this
3319 		 * point, set the FAST_SIGBLOCK_PEND as indicator for
3320 		 * usermode to perform a dummy call to
3321 		 * FAST_SIGBLOCK_UNBLOCK, which causes immediate
3322 		 * delivery of postponed pending signal.
3323 		 */
3324 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
3325 			if (td->td_sigblock_val != 0)
3326 				SIGSETNAND(sigpending, fastblock_mask);
3327 			if (SIGISEMPTY(sigpending)) {
3328 				td->td_pflags |= TDP_SIGFASTPENDING;
3329 				return (0);
3330 			}
3331 		}
3332 
3333 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
3334 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
3335 		    SIGISMEMBER(sigpending, SIGSTOP)) {
3336 			/*
3337 			 * If debugger just attached, always consume
3338 			 * SIGSTOP from ptrace(PT_ATTACH) first, to
3339 			 * execute the debugger attach ritual in
3340 			 * order.
3341 			 */
3342 			td->td_dbgflags |= TDB_FSTP;
3343 			SIGEMPTYSET(sigpending);
3344 			SIGADDSET(sigpending, SIGSTOP);
3345 		}
3346 
3347 		SIG_FOREACH(sig, &sigpending) {
3348 			switch (sigprocess(td, sig)) {
3349 			case SIGSTATUS_HANDLE:
3350 				return (sig);
3351 			case SIGSTATUS_HANDLED:
3352 				goto next;
3353 			case SIGSTATUS_IGNORE:
3354 				sigqueue_delete(&td->td_sigqueue, sig);
3355 				sigqueue_delete(&p->p_sigqueue, sig);
3356 				break;
3357 			case SIGSTATUS_SBDRY_STOP:
3358 				return (-1);
3359 			}
3360 		}
3361 next:;
3362 	}
3363 }
3364 
3365 void
3366 thread_stopped(struct proc *p)
3367 {
3368 	int n;
3369 
3370 	PROC_LOCK_ASSERT(p, MA_OWNED);
3371 	PROC_SLOCK_ASSERT(p, MA_OWNED);
3372 	n = p->p_suspcount;
3373 	if (p == curproc)
3374 		n++;
3375 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3376 		PROC_SUNLOCK(p);
3377 		p->p_flag &= ~P_WAITED;
3378 		PROC_LOCK(p->p_pptr);
3379 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
3380 			CLD_TRAPPED : CLD_STOPPED);
3381 		PROC_UNLOCK(p->p_pptr);
3382 		PROC_SLOCK(p);
3383 	}
3384 }
3385 
3386 /*
3387  * Take the action for the specified signal
3388  * from the current set of pending signals.
3389  */
3390 int
3391 postsig(int sig)
3392 {
3393 	struct thread *td;
3394 	struct proc *p;
3395 	struct sigacts *ps;
3396 	sig_t action;
3397 	ksiginfo_t ksi;
3398 	sigset_t returnmask;
3399 
3400 	KASSERT(sig != 0, ("postsig"));
3401 
3402 	td = curthread;
3403 	p = td->td_proc;
3404 	PROC_LOCK_ASSERT(p, MA_OWNED);
3405 	ps = p->p_sigacts;
3406 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3407 	ksiginfo_init(&ksi);
3408 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3409 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3410 		return (0);
3411 	ksi.ksi_signo = sig;
3412 	if (ksi.ksi_code == SI_TIMER)
3413 		itimer_accept(p, ksi.ksi_timerid, &ksi);
3414 	action = ps->ps_sigact[_SIG_IDX(sig)];
3415 #ifdef KTRACE
3416 	if (KTRPOINT(td, KTR_PSIG))
3417 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3418 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3419 #endif
3420 
3421 	if (action == SIG_DFL) {
3422 		/*
3423 		 * Default action, where the default is to kill
3424 		 * the process.  (Other cases were ignored above.)
3425 		 */
3426 		mtx_unlock(&ps->ps_mtx);
3427 		proc_td_siginfo_capture(td, &ksi.ksi_info);
3428 		sigexit1(td, sig, &ksi);
3429 		/* NOTREACHED */
3430 	} else {
3431 		/*
3432 		 * If we get here, the signal must be caught.
3433 		 */
3434 		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3435 		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3436 		    ("postsig action: blocked sig %d", sig));
3437 
3438 		/*
3439 		 * Set the new mask value and also defer further
3440 		 * occurrences of this signal.
3441 		 *
3442 		 * Special case: user has done a sigsuspend.  Here the
3443 		 * current mask is not of interest, but rather the
3444 		 * mask from before the sigsuspend is what we want
3445 		 * restored after the signal processing is completed.
3446 		 */
3447 		if (td->td_pflags & TDP_OLDMASK) {
3448 			returnmask = td->td_oldsigmask;
3449 			td->td_pflags &= ~TDP_OLDMASK;
3450 		} else
3451 			returnmask = td->td_sigmask;
3452 
3453 		if (p->p_sig == sig) {
3454 			p->p_sig = 0;
3455 		}
3456 		sig_handle_killpg(p, &ksi);
3457 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3458 		postsig_done(sig, td, ps);
3459 	}
3460 	return (1);
3461 }
3462 
3463 int
3464 sig_ast_checksusp(struct thread *td)
3465 {
3466 	struct proc *p __diagused;
3467 	int ret;
3468 
3469 	p = td->td_proc;
3470 	PROC_LOCK_ASSERT(p, MA_OWNED);
3471 
3472 	if (!td_ast_pending(td, TDA_SUSPEND))
3473 		return (0);
3474 
3475 	ret = thread_suspend_check(1);
3476 	MPASS(ret == 0 || ret == EINTR || ret == ERESTART);
3477 	return (ret);
3478 }
3479 
3480 int
3481 sig_ast_needsigchk(struct thread *td)
3482 {
3483 	struct proc *p;
3484 	struct sigacts *ps;
3485 	int ret, sig;
3486 
3487 	p = td->td_proc;
3488 	PROC_LOCK_ASSERT(p, MA_OWNED);
3489 
3490 	if (!td_ast_pending(td, TDA_SIG))
3491 		return (0);
3492 
3493 	ps = p->p_sigacts;
3494 	mtx_lock(&ps->ps_mtx);
3495 	sig = cursig(td);
3496 	if (sig == -1) {
3497 		mtx_unlock(&ps->ps_mtx);
3498 		KASSERT((td->td_flags & TDF_SBDRY) != 0, ("lost TDF_SBDRY"));
3499 		KASSERT(TD_SBDRY_INTR(td),
3500 		    ("lost TDF_SERESTART of TDF_SEINTR"));
3501 		KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
3502 		    (TDF_SEINTR | TDF_SERESTART),
3503 		    ("both TDF_SEINTR and TDF_SERESTART"));
3504 		ret = TD_SBDRY_ERRNO(td);
3505 	} else if (sig != 0) {
3506 		ret = SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR : ERESTART;
3507 		mtx_unlock(&ps->ps_mtx);
3508 	} else {
3509 		mtx_unlock(&ps->ps_mtx);
3510 		ret = 0;
3511 	}
3512 
3513 	/*
3514 	 * Do not go into sleep if this thread was the ptrace(2)
3515 	 * attach leader.  cursig() consumed SIGSTOP from PT_ATTACH,
3516 	 * but we usually act on the signal by interrupting sleep, and
3517 	 * should do that here as well.
3518 	 */
3519 	if ((td->td_dbgflags & TDB_FSTP) != 0) {
3520 		if (ret == 0)
3521 			ret = EINTR;
3522 		td->td_dbgflags &= ~TDB_FSTP;
3523 	}
3524 
3525 	return (ret);
3526 }
3527 
3528 int
3529 sig_intr(void)
3530 {
3531 	struct thread *td;
3532 	struct proc *p;
3533 	int ret;
3534 
3535 	td = curthread;
3536 	if (!td_ast_pending(td, TDA_SIG) && !td_ast_pending(td, TDA_SUSPEND))
3537 		return (0);
3538 
3539 	p = td->td_proc;
3540 
3541 	PROC_LOCK(p);
3542 	ret = sig_ast_checksusp(td);
3543 	if (ret == 0)
3544 		ret = sig_ast_needsigchk(td);
3545 	PROC_UNLOCK(p);
3546 	return (ret);
3547 }
3548 
3549 bool
3550 curproc_sigkilled(void)
3551 {
3552 	struct thread *td;
3553 	struct proc *p;
3554 	struct sigacts *ps;
3555 	bool res;
3556 
3557 	td = curthread;
3558 	if (!td_ast_pending(td, TDA_SIG))
3559 		return (false);
3560 
3561 	p = td->td_proc;
3562 	PROC_LOCK(p);
3563 	ps = p->p_sigacts;
3564 	mtx_lock(&ps->ps_mtx);
3565 	res = SIGISMEMBER(td->td_sigqueue.sq_signals, SIGKILL) ||
3566 	    SIGISMEMBER(p->p_sigqueue.sq_signals, SIGKILL);
3567 	mtx_unlock(&ps->ps_mtx);
3568 	PROC_UNLOCK(p);
3569 	return (res);
3570 }
3571 
3572 void
3573 proc_wkilled(struct proc *p)
3574 {
3575 
3576 	PROC_LOCK_ASSERT(p, MA_OWNED);
3577 	if ((p->p_flag & P_WKILLED) == 0) {
3578 		p->p_flag |= P_WKILLED;
3579 		/*
3580 		 * Notify swapper that there is a process to swap in.
3581 		 * The notification is racy, at worst it would take 10
3582 		 * seconds for the swapper process to notice.
3583 		 */
3584 		if ((p->p_flag & (P_INMEM | P_SWAPPINGIN)) == 0)
3585 			wakeup(&proc0);
3586 	}
3587 }
3588 
3589 /*
3590  * Kill the current process for stated reason.
3591  */
3592 void
3593 killproc(struct proc *p, const char *why)
3594 {
3595 
3596 	PROC_LOCK_ASSERT(p, MA_OWNED);
3597 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3598 	    p->p_comm);
3599 	log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3600 	    p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3601 	    p->p_ucred->cr_uid, why);
3602 	proc_wkilled(p);
3603 	kern_psignal(p, SIGKILL);
3604 }
3605 
3606 /*
3607  * Force the current process to exit with the specified signal, dumping core
3608  * if appropriate.  We bypass the normal tests for masked and caught signals,
3609  * allowing unrecoverable failures to terminate the process without changing
3610  * signal state.  Mark the accounting record with the signal termination.
3611  * If dumping core, save the signal number for the debugger.  Calls exit and
3612  * does not return.
3613  */
3614 static void
3615 sigexit1(struct thread *td, int sig, ksiginfo_t *ksi)
3616 {
3617 	struct proc *p = td->td_proc;
3618 
3619 	PROC_LOCK_ASSERT(p, MA_OWNED);
3620 	proc_set_p2_wexit(p);
3621 
3622 	p->p_acflag |= AXSIG;
3623 	/*
3624 	 * We must be single-threading to generate a core dump.  This
3625 	 * ensures that the registers in the core file are up-to-date.
3626 	 * Also, the ELF dump handler assumes that the thread list doesn't
3627 	 * change out from under it.
3628 	 *
3629 	 * XXX If another thread attempts to single-thread before us
3630 	 *     (e.g. via fork()), we won't get a dump at all.
3631 	 */
3632 	if ((sigprop(sig) & SIGPROP_CORE) &&
3633 	    thread_single(p, SINGLE_NO_EXIT) == 0) {
3634 		p->p_sig = sig;
3635 		/*
3636 		 * Log signals which would cause core dumps
3637 		 * (Log as LOG_INFO to appease those who don't want
3638 		 * these messages.)
3639 		 * XXX : Todo, as well as euid, write out ruid too
3640 		 * Note that coredump() drops proc lock.
3641 		 */
3642 		if (coredump(td) == 0)
3643 			sig |= WCOREFLAG;
3644 		if (kern_logsigexit)
3645 			log(LOG_INFO,
3646 			    "pid %d (%s), jid %d, uid %d: exited on "
3647 			    "signal %d%s\n", p->p_pid, p->p_comm,
3648 			    p->p_ucred->cr_prison->pr_id,
3649 			    td->td_ucred->cr_uid,
3650 			    sig &~ WCOREFLAG,
3651 			    sig & WCOREFLAG ? " (core dumped)" : "");
3652 	} else
3653 		PROC_UNLOCK(p);
3654 	exit2(td, 0, sig, ksi != NULL && (ksi->ksi_flags & KSI_KILLPG) != 0);
3655 	/* NOTREACHED */
3656 }
3657 
3658 void
3659 sigexit(struct thread *td, int sig)
3660 {
3661 	sigexit1(td, sig, NULL);
3662 }
3663 
3664 /*
3665  * Send queued SIGCHLD to parent when child process's state
3666  * is changed.
3667  */
3668 static void
3669 sigparent(struct proc *p, int reason, int status)
3670 {
3671 	PROC_LOCK_ASSERT(p, MA_OWNED);
3672 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3673 
3674 	if (p->p_ksi != NULL) {
3675 		p->p_ksi->ksi_signo  = SIGCHLD;
3676 		p->p_ksi->ksi_code   = reason;
3677 		p->p_ksi->ksi_status = status;
3678 		p->p_ksi->ksi_pid    = p->p_pid;
3679 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3680 		if (KSI_ONQ(p->p_ksi))
3681 			return;
3682 	}
3683 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3684 }
3685 
3686 static void
3687 childproc_jobstate(struct proc *p, int reason, int sig)
3688 {
3689 	struct sigacts *ps;
3690 
3691 	PROC_LOCK_ASSERT(p, MA_OWNED);
3692 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3693 
3694 	/*
3695 	 * Wake up parent sleeping in kern_wait(), also send
3696 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3697 	 * that parent will awake, because parent may masked
3698 	 * the signal.
3699 	 */
3700 	p->p_pptr->p_flag |= P_STATCHILD;
3701 	wakeup(p->p_pptr);
3702 
3703 	ps = p->p_pptr->p_sigacts;
3704 	mtx_lock(&ps->ps_mtx);
3705 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3706 		mtx_unlock(&ps->ps_mtx);
3707 		sigparent(p, reason, sig);
3708 	} else
3709 		mtx_unlock(&ps->ps_mtx);
3710 }
3711 
3712 void
3713 childproc_stopped(struct proc *p, int reason)
3714 {
3715 
3716 	childproc_jobstate(p, reason, p->p_xsig);
3717 }
3718 
3719 void
3720 childproc_continued(struct proc *p)
3721 {
3722 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3723 }
3724 
3725 void
3726 childproc_exited(struct proc *p)
3727 {
3728 	int reason, status;
3729 
3730 	if (WCOREDUMP(p->p_xsig)) {
3731 		reason = CLD_DUMPED;
3732 		status = WTERMSIG(p->p_xsig);
3733 	} else if (WIFSIGNALED(p->p_xsig)) {
3734 		reason = CLD_KILLED;
3735 		status = WTERMSIG(p->p_xsig);
3736 	} else {
3737 		reason = CLD_EXITED;
3738 		status = p->p_xexit;
3739 	}
3740 	/*
3741 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3742 	 * done in exit1().
3743 	 */
3744 	sigparent(p, reason, status);
3745 }
3746 
3747 #define	MAX_NUM_CORE_FILES 100000
3748 #ifndef NUM_CORE_FILES
3749 #define	NUM_CORE_FILES 5
3750 #endif
3751 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3752 static int num_cores = NUM_CORE_FILES;
3753 
3754 static int
3755 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3756 {
3757 	int error;
3758 	int new_val;
3759 
3760 	new_val = num_cores;
3761 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3762 	if (error != 0 || req->newptr == NULL)
3763 		return (error);
3764 	if (new_val > MAX_NUM_CORE_FILES)
3765 		new_val = MAX_NUM_CORE_FILES;
3766 	if (new_val < 0)
3767 		new_val = 0;
3768 	num_cores = new_val;
3769 	return (0);
3770 }
3771 SYSCTL_PROC(_debug, OID_AUTO, ncores,
3772     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
3773     sysctl_debug_num_cores_check, "I",
3774     "Maximum number of generated process corefiles while using index format");
3775 
3776 #define	GZIP_SUFFIX	".gz"
3777 #define	ZSTD_SUFFIX	".zst"
3778 
3779 int compress_user_cores = 0;
3780 
3781 static int
3782 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3783 {
3784 	int error, val;
3785 
3786 	val = compress_user_cores;
3787 	error = sysctl_handle_int(oidp, &val, 0, req);
3788 	if (error != 0 || req->newptr == NULL)
3789 		return (error);
3790 	if (val != 0 && !compressor_avail(val))
3791 		return (EINVAL);
3792 	compress_user_cores = val;
3793 	return (error);
3794 }
3795 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores,
3796     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, 0, sizeof(int),
3797     sysctl_compress_user_cores, "I",
3798     "Enable compression of user corefiles ("
3799     __XSTRING(COMPRESS_GZIP) " = gzip, "
3800     __XSTRING(COMPRESS_ZSTD) " = zstd)");
3801 
3802 int compress_user_cores_level = 6;
3803 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3804     &compress_user_cores_level, 0,
3805     "Corefile compression level");
3806 
3807 /*
3808  * Protect the access to corefilename[] by allproc_lock.
3809  */
3810 #define	corefilename_lock	allproc_lock
3811 
3812 static char corefilename[MAXPATHLEN] = {"%N.core"};
3813 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3814 
3815 static int
3816 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3817 {
3818 	int error;
3819 
3820 	sx_xlock(&corefilename_lock);
3821 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3822 	    req);
3823 	sx_xunlock(&corefilename_lock);
3824 
3825 	return (error);
3826 }
3827 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3828     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3829     "Process corefile name format string");
3830 
3831 static void
3832 vnode_close_locked(struct thread *td, struct vnode *vp)
3833 {
3834 
3835 	VOP_UNLOCK(vp);
3836 	vn_close(vp, FWRITE, td->td_ucred, td);
3837 }
3838 
3839 /*
3840  * If the core format has a %I in it, then we need to check
3841  * for existing corefiles before defining a name.
3842  * To do this we iterate over 0..ncores to find a
3843  * non-existing core file name to use. If all core files are
3844  * already used we choose the oldest one.
3845  */
3846 static int
3847 corefile_open_last(struct thread *td, char *name, int indexpos,
3848     int indexlen, int ncores, struct vnode **vpp)
3849 {
3850 	struct vnode *oldvp, *nextvp, *vp;
3851 	struct vattr vattr;
3852 	struct nameidata nd;
3853 	int error, i, flags, oflags, cmode;
3854 	char ch;
3855 	struct timespec lasttime;
3856 
3857 	nextvp = oldvp = NULL;
3858 	cmode = S_IRUSR | S_IWUSR;
3859 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3860 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3861 
3862 	for (i = 0; i < ncores; i++) {
3863 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3864 
3865 		ch = name[indexpos + indexlen];
3866 		(void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3867 		    i);
3868 		name[indexpos + indexlen] = ch;
3869 
3870 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
3871 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3872 		    NULL);
3873 		if (error != 0)
3874 			break;
3875 
3876 		vp = nd.ni_vp;
3877 		NDFREE_PNBUF(&nd);
3878 		if ((flags & O_CREAT) == O_CREAT) {
3879 			nextvp = vp;
3880 			break;
3881 		}
3882 
3883 		error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3884 		if (error != 0) {
3885 			vnode_close_locked(td, vp);
3886 			break;
3887 		}
3888 
3889 		if (oldvp == NULL ||
3890 		    lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3891 		    (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3892 		    lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3893 			if (oldvp != NULL)
3894 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3895 			oldvp = vp;
3896 			VOP_UNLOCK(oldvp);
3897 			lasttime = vattr.va_mtime;
3898 		} else {
3899 			vnode_close_locked(td, vp);
3900 		}
3901 	}
3902 
3903 	if (oldvp != NULL) {
3904 		if (nextvp == NULL) {
3905 			if ((td->td_proc->p_flag & P_SUGID) != 0) {
3906 				error = EFAULT;
3907 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3908 			} else {
3909 				nextvp = oldvp;
3910 				error = vn_lock(nextvp, LK_EXCLUSIVE);
3911 				if (error != 0) {
3912 					vn_close(nextvp, FWRITE, td->td_ucred,
3913 					    td);
3914 					nextvp = NULL;
3915 				}
3916 			}
3917 		} else {
3918 			vn_close(oldvp, FWRITE, td->td_ucred, td);
3919 		}
3920 	}
3921 	if (error != 0) {
3922 		if (nextvp != NULL)
3923 			vnode_close_locked(td, oldvp);
3924 	} else {
3925 		*vpp = nextvp;
3926 	}
3927 
3928 	return (error);
3929 }
3930 
3931 /*
3932  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3933  * Expand the name described in corefilename, using name, uid, and pid
3934  * and open/create core file.
3935  * corefilename is a printf-like string, with three format specifiers:
3936  *	%N	name of process ("name")
3937  *	%P	process id (pid)
3938  *	%U	user id (uid)
3939  * For example, "%N.core" is the default; they can be disabled completely
3940  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3941  * This is controlled by the sysctl variable kern.corefile (see above).
3942  */
3943 static int
3944 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3945     int compress, int signum, struct vnode **vpp, char **namep)
3946 {
3947 	struct sbuf sb;
3948 	struct nameidata nd;
3949 	const char *format;
3950 	char *hostname, *name;
3951 	int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
3952 
3953 	hostname = NULL;
3954 	format = corefilename;
3955 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3956 	indexlen = 0;
3957 	indexpos = -1;
3958 	ncores = num_cores;
3959 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3960 	sx_slock(&corefilename_lock);
3961 	for (i = 0; format[i] != '\0'; i++) {
3962 		switch (format[i]) {
3963 		case '%':	/* Format character */
3964 			i++;
3965 			switch (format[i]) {
3966 			case '%':
3967 				sbuf_putc(&sb, '%');
3968 				break;
3969 			case 'H':	/* hostname */
3970 				if (hostname == NULL) {
3971 					hostname = malloc(MAXHOSTNAMELEN,
3972 					    M_TEMP, M_WAITOK);
3973 				}
3974 				getcredhostname(td->td_ucred, hostname,
3975 				    MAXHOSTNAMELEN);
3976 				sbuf_printf(&sb, "%s", hostname);
3977 				break;
3978 			case 'I':	/* autoincrementing index */
3979 				if (indexpos != -1) {
3980 					sbuf_printf(&sb, "%%I");
3981 					break;
3982 				}
3983 
3984 				indexpos = sbuf_len(&sb);
3985 				sbuf_printf(&sb, "%u", ncores - 1);
3986 				indexlen = sbuf_len(&sb) - indexpos;
3987 				break;
3988 			case 'N':	/* process name */
3989 				sbuf_printf(&sb, "%s", comm);
3990 				break;
3991 			case 'P':	/* process id */
3992 				sbuf_printf(&sb, "%u", pid);
3993 				break;
3994 			case 'S':	/* signal number */
3995 				sbuf_printf(&sb, "%i", signum);
3996 				break;
3997 			case 'U':	/* user id */
3998 				sbuf_printf(&sb, "%u", uid);
3999 				break;
4000 			default:
4001 				log(LOG_ERR,
4002 				    "Unknown format character %c in "
4003 				    "corename `%s'\n", format[i], format);
4004 				break;
4005 			}
4006 			break;
4007 		default:
4008 			sbuf_putc(&sb, format[i]);
4009 			break;
4010 		}
4011 	}
4012 	sx_sunlock(&corefilename_lock);
4013 	free(hostname, M_TEMP);
4014 	if (compress == COMPRESS_GZIP)
4015 		sbuf_printf(&sb, GZIP_SUFFIX);
4016 	else if (compress == COMPRESS_ZSTD)
4017 		sbuf_printf(&sb, ZSTD_SUFFIX);
4018 	if (sbuf_error(&sb) != 0) {
4019 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
4020 		    "long\n", (long)pid, comm, (u_long)uid);
4021 		sbuf_delete(&sb);
4022 		free(name, M_TEMP);
4023 		return (ENOMEM);
4024 	}
4025 	sbuf_finish(&sb);
4026 	sbuf_delete(&sb);
4027 
4028 	if (indexpos != -1) {
4029 		error = corefile_open_last(td, name, indexpos, indexlen, ncores,
4030 		    vpp);
4031 		if (error != 0) {
4032 			log(LOG_ERR,
4033 			    "pid %d (%s), uid (%u):  Path `%s' failed "
4034 			    "on initial open test, error = %d\n",
4035 			    pid, comm, uid, name, error);
4036 		}
4037 	} else {
4038 		cmode = S_IRUSR | S_IWUSR;
4039 		oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
4040 		    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
4041 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
4042 		if ((td->td_proc->p_flag & P_SUGID) != 0)
4043 			flags |= O_EXCL;
4044 
4045 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
4046 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
4047 		    NULL);
4048 		if (error == 0) {
4049 			*vpp = nd.ni_vp;
4050 			NDFREE_PNBUF(&nd);
4051 		}
4052 	}
4053 
4054 	if (error != 0) {
4055 #ifdef AUDIT
4056 		audit_proc_coredump(td, name, error);
4057 #endif
4058 		free(name, M_TEMP);
4059 		return (error);
4060 	}
4061 	*namep = name;
4062 	return (0);
4063 }
4064 
4065 /*
4066  * Dump a process' core.  The main routine does some
4067  * policy checking, and creates the name of the coredump;
4068  * then it passes on a vnode and a size limit to the process-specific
4069  * coredump routine if there is one; if there _is not_ one, it returns
4070  * ENOSYS; otherwise it returns the error from the process-specific routine.
4071  */
4072 
4073 static int
4074 coredump(struct thread *td)
4075 {
4076 	struct proc *p = td->td_proc;
4077 	struct ucred *cred = td->td_ucred;
4078 	struct vnode *vp;
4079 	struct flock lf;
4080 	struct vattr vattr;
4081 	size_t fullpathsize;
4082 	int error, error1, locked;
4083 	char *name;			/* name of corefile */
4084 	void *rl_cookie;
4085 	off_t limit;
4086 	char *fullpath, *freepath = NULL;
4087 	struct sbuf *sb;
4088 
4089 	PROC_LOCK_ASSERT(p, MA_OWNED);
4090 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
4091 
4092 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
4093 	    (p->p_flag2 & P2_NOTRACE) != 0) {
4094 		PROC_UNLOCK(p);
4095 		return (EFAULT);
4096 	}
4097 
4098 	/*
4099 	 * Note that the bulk of limit checking is done after
4100 	 * the corefile is created.  The exception is if the limit
4101 	 * for corefiles is 0, in which case we don't bother
4102 	 * creating the corefile at all.  This layout means that
4103 	 * a corefile is truncated instead of not being created,
4104 	 * if it is larger than the limit.
4105 	 */
4106 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
4107 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
4108 		PROC_UNLOCK(p);
4109 		return (EFBIG);
4110 	}
4111 	PROC_UNLOCK(p);
4112 
4113 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
4114 	    compress_user_cores, p->p_sig, &vp, &name);
4115 	if (error != 0)
4116 		return (error);
4117 
4118 	/*
4119 	 * Don't dump to non-regular files or files with links.
4120 	 * Do not dump into system files. Effective user must own the corefile.
4121 	 */
4122 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
4123 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
4124 	    vattr.va_uid != cred->cr_uid) {
4125 		VOP_UNLOCK(vp);
4126 		error = EFAULT;
4127 		goto out;
4128 	}
4129 
4130 	VOP_UNLOCK(vp);
4131 
4132 	/* Postpone other writers, including core dumps of other processes. */
4133 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
4134 
4135 	lf.l_whence = SEEK_SET;
4136 	lf.l_start = 0;
4137 	lf.l_len = 0;
4138 	lf.l_type = F_WRLCK;
4139 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
4140 
4141 	VATTR_NULL(&vattr);
4142 	vattr.va_size = 0;
4143 	if (set_core_nodump_flag)
4144 		vattr.va_flags = UF_NODUMP;
4145 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4146 	VOP_SETATTR(vp, &vattr, cred);
4147 	VOP_UNLOCK(vp);
4148 	PROC_LOCK(p);
4149 	p->p_acflag |= ACORE;
4150 	PROC_UNLOCK(p);
4151 
4152 	if (p->p_sysent->sv_coredump != NULL) {
4153 		error = p->p_sysent->sv_coredump(td, vp, limit, 0);
4154 	} else {
4155 		error = ENOSYS;
4156 	}
4157 
4158 	if (locked) {
4159 		lf.l_type = F_UNLCK;
4160 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
4161 	}
4162 	vn_rangelock_unlock(vp, rl_cookie);
4163 
4164 	/*
4165 	 * Notify the userland helper that a process triggered a core dump.
4166 	 * This allows the helper to run an automated debugging session.
4167 	 */
4168 	if (error != 0 || coredump_devctl == 0)
4169 		goto out;
4170 	sb = sbuf_new_auto();
4171 	if (vn_fullpath_global(p->p_textvp, &fullpath, &freepath) != 0)
4172 		goto out2;
4173 	sbuf_printf(sb, "comm=\"");
4174 	devctl_safe_quote_sb(sb, fullpath);
4175 	free(freepath, M_TEMP);
4176 	sbuf_printf(sb, "\" core=\"");
4177 
4178 	/*
4179 	 * We can't lookup core file vp directly. When we're replacing a core, and
4180 	 * other random times, we flush the name cache, so it will fail. Instead,
4181 	 * if the path of the core is relative, add the current dir in front if it.
4182 	 */
4183 	if (name[0] != '/') {
4184 		fullpathsize = MAXPATHLEN;
4185 		freepath = malloc(fullpathsize, M_TEMP, M_WAITOK);
4186 		if (vn_getcwd(freepath, &fullpath, &fullpathsize) != 0) {
4187 			free(freepath, M_TEMP);
4188 			goto out2;
4189 		}
4190 		devctl_safe_quote_sb(sb, fullpath);
4191 		free(freepath, M_TEMP);
4192 		sbuf_putc(sb, '/');
4193 	}
4194 	devctl_safe_quote_sb(sb, name);
4195 	sbuf_printf(sb, "\"");
4196 	if (sbuf_finish(sb) == 0)
4197 		devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
4198 out2:
4199 	sbuf_delete(sb);
4200 out:
4201 	error1 = vn_close(vp, FWRITE, cred, td);
4202 	if (error == 0)
4203 		error = error1;
4204 #ifdef AUDIT
4205 	audit_proc_coredump(td, name, error);
4206 #endif
4207 	free(name, M_TEMP);
4208 	return (error);
4209 }
4210 
4211 /*
4212  * Nonexistent system call-- signal process (may want to handle it).  Flag
4213  * error in case process won't see signal immediately (blocked or ignored).
4214  */
4215 #ifndef _SYS_SYSPROTO_H_
4216 struct nosys_args {
4217 	int	dummy;
4218 };
4219 #endif
4220 /* ARGSUSED */
4221 int
4222 nosys(struct thread *td, struct nosys_args *args)
4223 {
4224 	struct proc *p;
4225 
4226 	p = td->td_proc;
4227 
4228 	PROC_LOCK(p);
4229 	tdsignal(td, SIGSYS);
4230 	PROC_UNLOCK(p);
4231 	if (kern_lognosys == 1 || kern_lognosys == 3) {
4232 		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4233 		    td->td_sa.code);
4234 	}
4235 	if (kern_lognosys == 2 || kern_lognosys == 3 ||
4236 	    (p->p_pid == 1 && (kern_lognosys & 3) == 0)) {
4237 		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4238 		    td->td_sa.code);
4239 	}
4240 	return (ENOSYS);
4241 }
4242 
4243 /*
4244  * Send a SIGIO or SIGURG signal to a process or process group using stored
4245  * credentials rather than those of the current process.
4246  */
4247 void
4248 pgsigio(struct sigio **sigiop, int sig, int checkctty)
4249 {
4250 	ksiginfo_t ksi;
4251 	struct sigio *sigio;
4252 
4253 	ksiginfo_init(&ksi);
4254 	ksi.ksi_signo = sig;
4255 	ksi.ksi_code = SI_KERNEL;
4256 
4257 	SIGIO_LOCK();
4258 	sigio = *sigiop;
4259 	if (sigio == NULL) {
4260 		SIGIO_UNLOCK();
4261 		return;
4262 	}
4263 	if (sigio->sio_pgid > 0) {
4264 		PROC_LOCK(sigio->sio_proc);
4265 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
4266 			kern_psignal(sigio->sio_proc, sig);
4267 		PROC_UNLOCK(sigio->sio_proc);
4268 	} else if (sigio->sio_pgid < 0) {
4269 		struct proc *p;
4270 
4271 		PGRP_LOCK(sigio->sio_pgrp);
4272 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
4273 			PROC_LOCK(p);
4274 			if (p->p_state == PRS_NORMAL &&
4275 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
4276 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
4277 				kern_psignal(p, sig);
4278 			PROC_UNLOCK(p);
4279 		}
4280 		PGRP_UNLOCK(sigio->sio_pgrp);
4281 	}
4282 	SIGIO_UNLOCK();
4283 }
4284 
4285 static int
4286 filt_sigattach(struct knote *kn)
4287 {
4288 	struct proc *p = curproc;
4289 
4290 	kn->kn_ptr.p_proc = p;
4291 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
4292 
4293 	knlist_add(p->p_klist, kn, 0);
4294 
4295 	return (0);
4296 }
4297 
4298 static void
4299 filt_sigdetach(struct knote *kn)
4300 {
4301 	struct proc *p = kn->kn_ptr.p_proc;
4302 
4303 	knlist_remove(p->p_klist, kn, 0);
4304 }
4305 
4306 /*
4307  * signal knotes are shared with proc knotes, so we apply a mask to
4308  * the hint in order to differentiate them from process hints.  This
4309  * could be avoided by using a signal-specific knote list, but probably
4310  * isn't worth the trouble.
4311  */
4312 static int
4313 filt_signal(struct knote *kn, long hint)
4314 {
4315 
4316 	if (hint & NOTE_SIGNAL) {
4317 		hint &= ~NOTE_SIGNAL;
4318 
4319 		if (kn->kn_id == hint)
4320 			kn->kn_data++;
4321 	}
4322 	return (kn->kn_data != 0);
4323 }
4324 
4325 struct sigacts *
4326 sigacts_alloc(void)
4327 {
4328 	struct sigacts *ps;
4329 
4330 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
4331 	refcount_init(&ps->ps_refcnt, 1);
4332 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
4333 	return (ps);
4334 }
4335 
4336 void
4337 sigacts_free(struct sigacts *ps)
4338 {
4339 
4340 	if (refcount_release(&ps->ps_refcnt) == 0)
4341 		return;
4342 	mtx_destroy(&ps->ps_mtx);
4343 	free(ps, M_SUBPROC);
4344 }
4345 
4346 struct sigacts *
4347 sigacts_hold(struct sigacts *ps)
4348 {
4349 
4350 	refcount_acquire(&ps->ps_refcnt);
4351 	return (ps);
4352 }
4353 
4354 void
4355 sigacts_copy(struct sigacts *dest, struct sigacts *src)
4356 {
4357 
4358 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
4359 	mtx_lock(&src->ps_mtx);
4360 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
4361 	mtx_unlock(&src->ps_mtx);
4362 }
4363 
4364 int
4365 sigacts_shared(struct sigacts *ps)
4366 {
4367 
4368 	return (ps->ps_refcnt > 1);
4369 }
4370 
4371 void
4372 sig_drop_caught(struct proc *p)
4373 {
4374 	int sig;
4375 	struct sigacts *ps;
4376 
4377 	ps = p->p_sigacts;
4378 	PROC_LOCK_ASSERT(p, MA_OWNED);
4379 	mtx_assert(&ps->ps_mtx, MA_OWNED);
4380 	SIG_FOREACH(sig, &ps->ps_sigcatch) {
4381 		sigdflt(ps, sig);
4382 		if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
4383 			sigqueue_delete_proc(p, sig);
4384 	}
4385 }
4386 
4387 static void
4388 sigfastblock_failed(struct thread *td, bool sendsig, bool write)
4389 {
4390 	ksiginfo_t ksi;
4391 
4392 	/*
4393 	 * Prevent further fetches and SIGSEGVs, allowing thread to
4394 	 * issue syscalls despite corruption.
4395 	 */
4396 	sigfastblock_clear(td);
4397 
4398 	if (!sendsig)
4399 		return;
4400 	ksiginfo_init_trap(&ksi);
4401 	ksi.ksi_signo = SIGSEGV;
4402 	ksi.ksi_code = write ? SEGV_ACCERR : SEGV_MAPERR;
4403 	ksi.ksi_addr = td->td_sigblock_ptr;
4404 	trapsignal(td, &ksi);
4405 }
4406 
4407 static bool
4408 sigfastblock_fetch_sig(struct thread *td, bool sendsig, uint32_t *valp)
4409 {
4410 	uint32_t res;
4411 
4412 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4413 		return (true);
4414 	if (fueword32((void *)td->td_sigblock_ptr, &res) == -1) {
4415 		sigfastblock_failed(td, sendsig, false);
4416 		return (false);
4417 	}
4418 	*valp = res;
4419 	td->td_sigblock_val = res & ~SIGFASTBLOCK_FLAGS;
4420 	return (true);
4421 }
4422 
4423 static void
4424 sigfastblock_resched(struct thread *td, bool resched)
4425 {
4426 	struct proc *p;
4427 
4428 	if (resched) {
4429 		p = td->td_proc;
4430 		PROC_LOCK(p);
4431 		reschedule_signals(p, td->td_sigmask, 0);
4432 		PROC_UNLOCK(p);
4433 	}
4434 	ast_sched(td, TDA_SIG);
4435 }
4436 
4437 int
4438 sys_sigfastblock(struct thread *td, struct sigfastblock_args *uap)
4439 {
4440 	struct proc *p;
4441 	int error, res;
4442 	uint32_t oldval;
4443 
4444 	error = 0;
4445 	p = td->td_proc;
4446 	switch (uap->cmd) {
4447 	case SIGFASTBLOCK_SETPTR:
4448 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
4449 			error = EBUSY;
4450 			break;
4451 		}
4452 		if (((uintptr_t)(uap->ptr) & (sizeof(uint32_t) - 1)) != 0) {
4453 			error = EINVAL;
4454 			break;
4455 		}
4456 		td->td_pflags |= TDP_SIGFASTBLOCK;
4457 		td->td_sigblock_ptr = uap->ptr;
4458 		break;
4459 
4460 	case SIGFASTBLOCK_UNBLOCK:
4461 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4462 			error = EINVAL;
4463 			break;
4464 		}
4465 
4466 		for (;;) {
4467 			res = casueword32(td->td_sigblock_ptr,
4468 			    SIGFASTBLOCK_PEND, &oldval, 0);
4469 			if (res == -1) {
4470 				error = EFAULT;
4471 				sigfastblock_failed(td, false, true);
4472 				break;
4473 			}
4474 			if (res == 0)
4475 				break;
4476 			MPASS(res == 1);
4477 			if (oldval != SIGFASTBLOCK_PEND) {
4478 				error = EBUSY;
4479 				break;
4480 			}
4481 			error = thread_check_susp(td, false);
4482 			if (error != 0)
4483 				break;
4484 		}
4485 		if (error != 0)
4486 			break;
4487 
4488 		/*
4489 		 * td_sigblock_val is cleared there, but not on a
4490 		 * syscall exit.  The end effect is that a single
4491 		 * interruptible sleep, while user sigblock word is
4492 		 * set, might return EINTR or ERESTART to usermode
4493 		 * without delivering signal.  All further sleeps,
4494 		 * until userspace clears the word and does
4495 		 * sigfastblock(UNBLOCK), observe current word and no
4496 		 * longer get interrupted.  It is slight
4497 		 * non-conformance, with alternative to have read the
4498 		 * sigblock word on each syscall entry.
4499 		 */
4500 		td->td_sigblock_val = 0;
4501 
4502 		/*
4503 		 * Rely on normal ast mechanism to deliver pending
4504 		 * signals to current thread.  But notify others about
4505 		 * fake unblock.
4506 		 */
4507 		sigfastblock_resched(td, error == 0 && p->p_numthreads != 1);
4508 
4509 		break;
4510 
4511 	case SIGFASTBLOCK_UNSETPTR:
4512 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4513 			error = EINVAL;
4514 			break;
4515 		}
4516 		if (!sigfastblock_fetch_sig(td, false, &oldval)) {
4517 			error = EFAULT;
4518 			break;
4519 		}
4520 		if (oldval != 0 && oldval != SIGFASTBLOCK_PEND) {
4521 			error = EBUSY;
4522 			break;
4523 		}
4524 		sigfastblock_clear(td);
4525 		break;
4526 
4527 	default:
4528 		error = EINVAL;
4529 		break;
4530 	}
4531 	return (error);
4532 }
4533 
4534 void
4535 sigfastblock_clear(struct thread *td)
4536 {
4537 	bool resched;
4538 
4539 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4540 		return;
4541 	td->td_sigblock_val = 0;
4542 	resched = (td->td_pflags & TDP_SIGFASTPENDING) != 0 ||
4543 	    SIGPENDING(td);
4544 	td->td_pflags &= ~(TDP_SIGFASTBLOCK | TDP_SIGFASTPENDING);
4545 	sigfastblock_resched(td, resched);
4546 }
4547 
4548 void
4549 sigfastblock_fetch(struct thread *td)
4550 {
4551 	uint32_t val;
4552 
4553 	(void)sigfastblock_fetch_sig(td, true, &val);
4554 }
4555 
4556 static void
4557 sigfastblock_setpend1(struct thread *td)
4558 {
4559 	int res;
4560 	uint32_t oldval;
4561 
4562 	if ((td->td_pflags & TDP_SIGFASTPENDING) == 0)
4563 		return;
4564 	res = fueword32((void *)td->td_sigblock_ptr, &oldval);
4565 	if (res == -1) {
4566 		sigfastblock_failed(td, true, false);
4567 		return;
4568 	}
4569 	for (;;) {
4570 		res = casueword32(td->td_sigblock_ptr, oldval, &oldval,
4571 		    oldval | SIGFASTBLOCK_PEND);
4572 		if (res == -1) {
4573 			sigfastblock_failed(td, true, true);
4574 			return;
4575 		}
4576 		if (res == 0) {
4577 			td->td_sigblock_val = oldval & ~SIGFASTBLOCK_FLAGS;
4578 			td->td_pflags &= ~TDP_SIGFASTPENDING;
4579 			break;
4580 		}
4581 		MPASS(res == 1);
4582 		if (thread_check_susp(td, false) != 0)
4583 			break;
4584 	}
4585 }
4586 
4587 static void
4588 sigfastblock_setpend(struct thread *td, bool resched)
4589 {
4590 	struct proc *p;
4591 
4592 	sigfastblock_setpend1(td);
4593 	if (resched) {
4594 		p = td->td_proc;
4595 		PROC_LOCK(p);
4596 		reschedule_signals(p, fastblock_mask, SIGPROCMASK_FASTBLK);
4597 		PROC_UNLOCK(p);
4598 	}
4599 }
4600