xref: /freebsd/sys/kern/kern_sig.c (revision 058ac3e8063366dafa634d9107642e12b038bf09)
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/kernel.h>
60 #include <sys/ktr.h>
61 #include <sys/ktrace.h>
62 #include <sys/limits.h>
63 #include <sys/lock.h>
64 #include <sys/malloc.h>
65 #include <sys/mutex.h>
66 #include <sys/refcount.h>
67 #include <sys/namei.h>
68 #include <sys/proc.h>
69 #include <sys/procdesc.h>
70 #include <sys/ptrace.h>
71 #include <sys/posix4.h>
72 #include <sys/racct.h>
73 #include <sys/resourcevar.h>
74 #include <sys/sdt.h>
75 #include <sys/sbuf.h>
76 #include <sys/sleepqueue.h>
77 #include <sys/smp.h>
78 #include <sys/stat.h>
79 #include <sys/sx.h>
80 #include <sys/syscall.h>
81 #include <sys/syscallsubr.h>
82 #include <sys/sysctl.h>
83 #include <sys/sysent.h>
84 #include <sys/syslog.h>
85 #include <sys/sysproto.h>
86 #include <sys/timers.h>
87 #include <sys/unistd.h>
88 #include <sys/vmmeter.h>
89 #include <sys/wait.h>
90 #include <vm/vm.h>
91 #include <vm/vm_extern.h>
92 #include <vm/uma.h>
93 
94 #include <sys/jail.h>
95 
96 #include <machine/cpu.h>
97 
98 #include <security/audit/audit.h>
99 
100 #define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
101 
102 SDT_PROVIDER_DECLARE(proc);
103 SDT_PROBE_DEFINE3(proc, , , signal__send,
104     "struct thread *", "struct proc *", "int");
105 SDT_PROBE_DEFINE2(proc, , , signal__clear,
106     "int", "ksiginfo_t *");
107 SDT_PROBE_DEFINE3(proc, , , signal__discard,
108     "struct thread *", "struct proc *", "int");
109 
110 static int	coredump(struct thread *);
111 static int	killpg1(struct thread *td, int sig, int pgid, int all,
112 		    ksiginfo_t *ksi);
113 static int	issignal(struct thread *td);
114 static void	reschedule_signals(struct proc *p, sigset_t block, int flags);
115 static int	sigprop(int sig);
116 static void	tdsigwakeup(struct thread *, int, sig_t, int);
117 static int	sig_suspend_threads(struct thread *, struct proc *);
118 static int	filt_sigattach(struct knote *kn);
119 static void	filt_sigdetach(struct knote *kn);
120 static int	filt_signal(struct knote *kn, long hint);
121 static struct thread *sigtd(struct proc *p, int sig, bool fast_sigblock);
122 static void	sigqueue_start(void);
123 static void	sigfastblock_setpend(struct thread *td, bool resched);
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 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 ksiginfo_t *
375 ksiginfo_alloc(int mwait)
376 {
377 	MPASS(mwait == M_WAITOK || mwait == M_NOWAIT);
378 
379 	if (ksiginfo_zone == NULL)
380 		return (NULL);
381 	return (uma_zalloc(ksiginfo_zone, mwait | M_ZERO));
382 }
383 
384 void
385 ksiginfo_free(ksiginfo_t *ksi)
386 {
387 	uma_zfree(ksiginfo_zone, ksi);
388 }
389 
390 static __inline bool
391 ksiginfo_tryfree(ksiginfo_t *ksi)
392 {
393 	if ((ksi->ksi_flags & KSI_EXT) == 0) {
394 		uma_zfree(ksiginfo_zone, ksi);
395 		return (true);
396 	}
397 	return (false);
398 }
399 
400 void
401 sigqueue_init(sigqueue_t *list, struct proc *p)
402 {
403 	SIGEMPTYSET(list->sq_signals);
404 	SIGEMPTYSET(list->sq_kill);
405 	SIGEMPTYSET(list->sq_ptrace);
406 	TAILQ_INIT(&list->sq_list);
407 	list->sq_proc = p;
408 	list->sq_flags = SQ_INIT;
409 }
410 
411 /*
412  * Get a signal's ksiginfo.
413  * Return:
414  *	0	-	signal not found
415  *	others	-	signal number
416  */
417 static int
418 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
419 {
420 	struct proc *p = sq->sq_proc;
421 	struct ksiginfo *ksi, *next;
422 	int count = 0;
423 
424 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
425 
426 	if (!SIGISMEMBER(sq->sq_signals, signo))
427 		return (0);
428 
429 	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
430 		count++;
431 		SIGDELSET(sq->sq_ptrace, signo);
432 		si->ksi_flags |= KSI_PTRACE;
433 	}
434 	if (SIGISMEMBER(sq->sq_kill, signo)) {
435 		count++;
436 		if (count == 1)
437 			SIGDELSET(sq->sq_kill, signo);
438 	}
439 
440 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
441 		if (ksi->ksi_signo == signo) {
442 			if (count == 0) {
443 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
444 				ksi->ksi_sigq = NULL;
445 				ksiginfo_copy(ksi, si);
446 				if (ksiginfo_tryfree(ksi) && p != NULL)
447 					p->p_pendingcnt--;
448 			}
449 			if (++count > 1)
450 				break;
451 		}
452 	}
453 
454 	if (count <= 1)
455 		SIGDELSET(sq->sq_signals, signo);
456 	si->ksi_signo = signo;
457 	return (signo);
458 }
459 
460 void
461 sigqueue_take(ksiginfo_t *ksi)
462 {
463 	struct ksiginfo *kp;
464 	struct proc	*p;
465 	sigqueue_t	*sq;
466 
467 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
468 		return;
469 
470 	p = sq->sq_proc;
471 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
472 	ksi->ksi_sigq = NULL;
473 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
474 		p->p_pendingcnt--;
475 
476 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
477 	     kp = TAILQ_NEXT(kp, ksi_link)) {
478 		if (kp->ksi_signo == ksi->ksi_signo)
479 			break;
480 	}
481 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
482 	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
483 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
484 }
485 
486 static int
487 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
488 {
489 	struct proc *p = sq->sq_proc;
490 	struct ksiginfo *ksi;
491 	int ret = 0;
492 
493 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
494 
495 	/*
496 	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
497 	 * for these signals.
498 	 */
499 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
500 		SIGADDSET(sq->sq_kill, signo);
501 		goto out_set_bit;
502 	}
503 
504 	/* directly insert the ksi, don't copy it */
505 	if (si->ksi_flags & KSI_INS) {
506 		if (si->ksi_flags & KSI_HEAD)
507 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
508 		else
509 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
510 		si->ksi_sigq = sq;
511 		goto out_set_bit;
512 	}
513 
514 	if (__predict_false(ksiginfo_zone == NULL)) {
515 		SIGADDSET(sq->sq_kill, signo);
516 		goto out_set_bit;
517 	}
518 
519 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
520 		signal_overflow++;
521 		ret = EAGAIN;
522 	} else if ((ksi = ksiginfo_alloc(M_NOWAIT)) == NULL) {
523 		signal_alloc_fail++;
524 		ret = EAGAIN;
525 	} else {
526 		if (p != NULL)
527 			p->p_pendingcnt++;
528 		ksiginfo_copy(si, ksi);
529 		ksi->ksi_signo = signo;
530 		if (si->ksi_flags & KSI_HEAD)
531 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
532 		else
533 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
534 		ksi->ksi_sigq = sq;
535 	}
536 
537 	if (ret != 0) {
538 		if ((si->ksi_flags & KSI_PTRACE) != 0) {
539 			SIGADDSET(sq->sq_ptrace, signo);
540 			ret = 0;
541 			goto out_set_bit;
542 		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
543 		    (si->ksi_flags & KSI_SIGQ) == 0) {
544 			SIGADDSET(sq->sq_kill, signo);
545 			ret = 0;
546 			goto out_set_bit;
547 		}
548 		return (ret);
549 	}
550 
551 out_set_bit:
552 	SIGADDSET(sq->sq_signals, signo);
553 	return (ret);
554 }
555 
556 void
557 sigqueue_flush(sigqueue_t *sq)
558 {
559 	struct proc *p = sq->sq_proc;
560 	ksiginfo_t *ksi;
561 
562 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
563 
564 	if (p != NULL)
565 		PROC_LOCK_ASSERT(p, MA_OWNED);
566 
567 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
568 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
569 		ksi->ksi_sigq = NULL;
570 		if (ksiginfo_tryfree(ksi) && p != NULL)
571 			p->p_pendingcnt--;
572 	}
573 
574 	SIGEMPTYSET(sq->sq_signals);
575 	SIGEMPTYSET(sq->sq_kill);
576 	SIGEMPTYSET(sq->sq_ptrace);
577 }
578 
579 static void
580 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
581 {
582 	sigset_t tmp;
583 	struct proc *p1, *p2;
584 	ksiginfo_t *ksi, *next;
585 
586 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
587 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
588 	p1 = src->sq_proc;
589 	p2 = dst->sq_proc;
590 	/* Move siginfo to target list */
591 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
592 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
593 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
594 			if (p1 != NULL)
595 				p1->p_pendingcnt--;
596 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
597 			ksi->ksi_sigq = dst;
598 			if (p2 != NULL)
599 				p2->p_pendingcnt++;
600 		}
601 	}
602 
603 	/* Move pending bits to target list */
604 	tmp = src->sq_kill;
605 	SIGSETAND(tmp, *set);
606 	SIGSETOR(dst->sq_kill, tmp);
607 	SIGSETNAND(src->sq_kill, tmp);
608 
609 	tmp = src->sq_ptrace;
610 	SIGSETAND(tmp, *set);
611 	SIGSETOR(dst->sq_ptrace, tmp);
612 	SIGSETNAND(src->sq_ptrace, tmp);
613 
614 	tmp = src->sq_signals;
615 	SIGSETAND(tmp, *set);
616 	SIGSETOR(dst->sq_signals, tmp);
617 	SIGSETNAND(src->sq_signals, tmp);
618 }
619 
620 #if 0
621 static void
622 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
623 {
624 	sigset_t set;
625 
626 	SIGEMPTYSET(set);
627 	SIGADDSET(set, signo);
628 	sigqueue_move_set(src, dst, &set);
629 }
630 #endif
631 
632 static void
633 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
634 {
635 	struct proc *p = sq->sq_proc;
636 	ksiginfo_t *ksi, *next;
637 
638 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
639 
640 	/* Remove siginfo queue */
641 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
642 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
643 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
644 			ksi->ksi_sigq = NULL;
645 			if (ksiginfo_tryfree(ksi) && p != NULL)
646 				p->p_pendingcnt--;
647 		}
648 	}
649 	SIGSETNAND(sq->sq_kill, *set);
650 	SIGSETNAND(sq->sq_ptrace, *set);
651 	SIGSETNAND(sq->sq_signals, *set);
652 }
653 
654 void
655 sigqueue_delete(sigqueue_t *sq, int signo)
656 {
657 	sigset_t set;
658 
659 	SIGEMPTYSET(set);
660 	SIGADDSET(set, signo);
661 	sigqueue_delete_set(sq, &set);
662 }
663 
664 /* Remove a set of signals for a process */
665 static void
666 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
667 {
668 	sigqueue_t worklist;
669 	struct thread *td0;
670 
671 	PROC_LOCK_ASSERT(p, MA_OWNED);
672 
673 	sigqueue_init(&worklist, NULL);
674 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
675 
676 	FOREACH_THREAD_IN_PROC(p, td0)
677 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
678 
679 	sigqueue_flush(&worklist);
680 }
681 
682 void
683 sigqueue_delete_proc(struct proc *p, int signo)
684 {
685 	sigset_t set;
686 
687 	SIGEMPTYSET(set);
688 	SIGADDSET(set, signo);
689 	sigqueue_delete_set_proc(p, &set);
690 }
691 
692 static void
693 sigqueue_delete_stopmask_proc(struct proc *p)
694 {
695 	sigset_t set;
696 
697 	SIGEMPTYSET(set);
698 	SIGADDSET(set, SIGSTOP);
699 	SIGADDSET(set, SIGTSTP);
700 	SIGADDSET(set, SIGTTIN);
701 	SIGADDSET(set, SIGTTOU);
702 	sigqueue_delete_set_proc(p, &set);
703 }
704 
705 /*
706  * Determine signal that should be delivered to thread td, the current
707  * thread, 0 if none.  If there is a pending stop signal with default
708  * action, the process stops in issignal().
709  */
710 int
711 cursig(struct thread *td)
712 {
713 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
714 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
715 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
716 	return (SIGPENDING(td) ? issignal(td) : 0);
717 }
718 
719 /*
720  * Arrange for ast() to handle unmasked pending signals on return to user
721  * mode.  This must be called whenever a signal is added to td_sigqueue or
722  * unmasked in td_sigmask.
723  */
724 void
725 signotify(struct thread *td)
726 {
727 
728 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
729 
730 	if (SIGPENDING(td))
731 		ast_sched(td, TDA_SIG);
732 }
733 
734 /*
735  * Returns 1 (true) if altstack is configured for the thread, and the
736  * passed stack bottom address falls into the altstack range.  Handles
737  * the 43 compat special case where the alt stack size is zero.
738  */
739 int
740 sigonstack(size_t sp)
741 {
742 	struct thread *td;
743 
744 	td = curthread;
745 	if ((td->td_pflags & TDP_ALTSTACK) == 0)
746 		return (0);
747 #if defined(COMPAT_43)
748 	if (SV_PROC_FLAG(td->td_proc, SV_AOUT) && td->td_sigstk.ss_size == 0)
749 		return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
750 #endif
751 	return (sp >= (size_t)td->td_sigstk.ss_sp &&
752 	    sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
753 }
754 
755 static __inline int
756 sigprop(int sig)
757 {
758 
759 	if (sig > 0 && sig < nitems(sigproptbl))
760 		return (sigproptbl[sig]);
761 	return (0);
762 }
763 
764 static bool
765 sigact_flag_test(const struct sigaction *act, int flag)
766 {
767 
768 	/*
769 	 * SA_SIGINFO is reset when signal disposition is set to
770 	 * ignore or default.  Other flags are kept according to user
771 	 * settings.
772 	 */
773 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
774 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
775 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
776 }
777 
778 /*
779  * kern_sigaction
780  * sigaction
781  * freebsd4_sigaction
782  * osigaction
783  */
784 int
785 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
786     struct sigaction *oact, int flags)
787 {
788 	struct sigacts *ps;
789 	struct proc *p = td->td_proc;
790 
791 	if (!_SIG_VALID(sig))
792 		return (EINVAL);
793 	if (act != NULL && act->sa_handler != SIG_DFL &&
794 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
795 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
796 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
797 		return (EINVAL);
798 
799 	PROC_LOCK(p);
800 	ps = p->p_sigacts;
801 	mtx_lock(&ps->ps_mtx);
802 	if (oact) {
803 		memset(oact, 0, sizeof(*oact));
804 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
805 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
806 			oact->sa_flags |= SA_ONSTACK;
807 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
808 			oact->sa_flags |= SA_RESTART;
809 		if (SIGISMEMBER(ps->ps_sigreset, sig))
810 			oact->sa_flags |= SA_RESETHAND;
811 		if (SIGISMEMBER(ps->ps_signodefer, sig))
812 			oact->sa_flags |= SA_NODEFER;
813 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
814 			oact->sa_flags |= SA_SIGINFO;
815 			oact->sa_sigaction =
816 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
817 		} else
818 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
819 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
820 			oact->sa_flags |= SA_NOCLDSTOP;
821 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
822 			oact->sa_flags |= SA_NOCLDWAIT;
823 	}
824 	if (act) {
825 		if ((sig == SIGKILL || sig == SIGSTOP) &&
826 		    act->sa_handler != SIG_DFL) {
827 			mtx_unlock(&ps->ps_mtx);
828 			PROC_UNLOCK(p);
829 			return (EINVAL);
830 		}
831 
832 		/*
833 		 * Change setting atomically.
834 		 */
835 
836 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
837 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
838 		if (sigact_flag_test(act, SA_SIGINFO)) {
839 			ps->ps_sigact[_SIG_IDX(sig)] =
840 			    (__sighandler_t *)act->sa_sigaction;
841 			SIGADDSET(ps->ps_siginfo, sig);
842 		} else {
843 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
844 			SIGDELSET(ps->ps_siginfo, sig);
845 		}
846 		if (!sigact_flag_test(act, SA_RESTART))
847 			SIGADDSET(ps->ps_sigintr, sig);
848 		else
849 			SIGDELSET(ps->ps_sigintr, sig);
850 		if (sigact_flag_test(act, SA_ONSTACK))
851 			SIGADDSET(ps->ps_sigonstack, sig);
852 		else
853 			SIGDELSET(ps->ps_sigonstack, sig);
854 		if (sigact_flag_test(act, SA_RESETHAND))
855 			SIGADDSET(ps->ps_sigreset, sig);
856 		else
857 			SIGDELSET(ps->ps_sigreset, sig);
858 		if (sigact_flag_test(act, SA_NODEFER))
859 			SIGADDSET(ps->ps_signodefer, sig);
860 		else
861 			SIGDELSET(ps->ps_signodefer, sig);
862 		if (sig == SIGCHLD) {
863 			if (act->sa_flags & SA_NOCLDSTOP)
864 				ps->ps_flag |= PS_NOCLDSTOP;
865 			else
866 				ps->ps_flag &= ~PS_NOCLDSTOP;
867 			if (act->sa_flags & SA_NOCLDWAIT) {
868 				/*
869 				 * Paranoia: since SA_NOCLDWAIT is implemented
870 				 * by reparenting the dying child to PID 1 (and
871 				 * trust it to reap the zombie), PID 1 itself
872 				 * is forbidden to set SA_NOCLDWAIT.
873 				 */
874 				if (p->p_pid == 1)
875 					ps->ps_flag &= ~PS_NOCLDWAIT;
876 				else
877 					ps->ps_flag |= PS_NOCLDWAIT;
878 			} else
879 				ps->ps_flag &= ~PS_NOCLDWAIT;
880 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
881 				ps->ps_flag |= PS_CLDSIGIGN;
882 			else
883 				ps->ps_flag &= ~PS_CLDSIGIGN;
884 		}
885 		/*
886 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
887 		 * and for signals set to SIG_DFL where the default is to
888 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
889 		 * have to restart the process.
890 		 */
891 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
892 		    (sigprop(sig) & SIGPROP_IGNORE &&
893 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
894 			/* never to be seen again */
895 			sigqueue_delete_proc(p, sig);
896 			if (sig != SIGCONT)
897 				/* easier in psignal */
898 				SIGADDSET(ps->ps_sigignore, sig);
899 			SIGDELSET(ps->ps_sigcatch, sig);
900 		} else {
901 			SIGDELSET(ps->ps_sigignore, sig);
902 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
903 				SIGDELSET(ps->ps_sigcatch, sig);
904 			else
905 				SIGADDSET(ps->ps_sigcatch, sig);
906 		}
907 #ifdef COMPAT_FREEBSD4
908 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
909 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
910 		    (flags & KSA_FREEBSD4) == 0)
911 			SIGDELSET(ps->ps_freebsd4, sig);
912 		else
913 			SIGADDSET(ps->ps_freebsd4, sig);
914 #endif
915 #ifdef COMPAT_43
916 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
917 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
918 		    (flags & KSA_OSIGSET) == 0)
919 			SIGDELSET(ps->ps_osigset, sig);
920 		else
921 			SIGADDSET(ps->ps_osigset, sig);
922 #endif
923 	}
924 	mtx_unlock(&ps->ps_mtx);
925 	PROC_UNLOCK(p);
926 	return (0);
927 }
928 
929 #ifndef _SYS_SYSPROTO_H_
930 struct sigaction_args {
931 	int	sig;
932 	struct	sigaction *act;
933 	struct	sigaction *oact;
934 };
935 #endif
936 int
937 sys_sigaction(struct thread *td, struct sigaction_args *uap)
938 {
939 	struct sigaction act, oact;
940 	struct sigaction *actp, *oactp;
941 	int error;
942 
943 	actp = (uap->act != NULL) ? &act : NULL;
944 	oactp = (uap->oact != NULL) ? &oact : NULL;
945 	if (actp) {
946 		error = copyin(uap->act, actp, sizeof(act));
947 		if (error)
948 			return (error);
949 	}
950 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
951 	if (oactp && !error)
952 		error = copyout(oactp, uap->oact, sizeof(oact));
953 	return (error);
954 }
955 
956 #ifdef COMPAT_FREEBSD4
957 #ifndef _SYS_SYSPROTO_H_
958 struct freebsd4_sigaction_args {
959 	int	sig;
960 	struct	sigaction *act;
961 	struct	sigaction *oact;
962 };
963 #endif
964 int
965 freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
966 {
967 	struct sigaction act, oact;
968 	struct sigaction *actp, *oactp;
969 	int error;
970 
971 	actp = (uap->act != NULL) ? &act : NULL;
972 	oactp = (uap->oact != NULL) ? &oact : NULL;
973 	if (actp) {
974 		error = copyin(uap->act, actp, sizeof(act));
975 		if (error)
976 			return (error);
977 	}
978 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
979 	if (oactp && !error)
980 		error = copyout(oactp, uap->oact, sizeof(oact));
981 	return (error);
982 }
983 #endif	/* COMAPT_FREEBSD4 */
984 
985 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
986 #ifndef _SYS_SYSPROTO_H_
987 struct osigaction_args {
988 	int	signum;
989 	struct	osigaction *nsa;
990 	struct	osigaction *osa;
991 };
992 #endif
993 int
994 osigaction(struct thread *td, struct osigaction_args *uap)
995 {
996 	struct osigaction sa;
997 	struct sigaction nsa, osa;
998 	struct sigaction *nsap, *osap;
999 	int error;
1000 
1001 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1002 		return (EINVAL);
1003 
1004 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
1005 	osap = (uap->osa != NULL) ? &osa : NULL;
1006 
1007 	if (nsap) {
1008 		error = copyin(uap->nsa, &sa, sizeof(sa));
1009 		if (error)
1010 			return (error);
1011 		nsap->sa_handler = sa.sa_handler;
1012 		nsap->sa_flags = sa.sa_flags;
1013 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
1014 	}
1015 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1016 	if (osap && !error) {
1017 		sa.sa_handler = osap->sa_handler;
1018 		sa.sa_flags = osap->sa_flags;
1019 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
1020 		error = copyout(&sa, uap->osa, sizeof(sa));
1021 	}
1022 	return (error);
1023 }
1024 
1025 #if !defined(__i386__)
1026 /* Avoid replicating the same stub everywhere */
1027 int
1028 osigreturn(struct thread *td, struct osigreturn_args *uap)
1029 {
1030 
1031 	return (nosys(td, (struct nosys_args *)uap));
1032 }
1033 #endif
1034 #endif /* COMPAT_43 */
1035 
1036 /*
1037  * Initialize signal state for process 0;
1038  * set to ignore signals that are ignored by default.
1039  */
1040 void
1041 siginit(struct proc *p)
1042 {
1043 	int i;
1044 	struct sigacts *ps;
1045 
1046 	PROC_LOCK(p);
1047 	ps = p->p_sigacts;
1048 	mtx_lock(&ps->ps_mtx);
1049 	for (i = 1; i <= NSIG; i++) {
1050 		if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) {
1051 			SIGADDSET(ps->ps_sigignore, i);
1052 		}
1053 	}
1054 	mtx_unlock(&ps->ps_mtx);
1055 	PROC_UNLOCK(p);
1056 }
1057 
1058 /*
1059  * Reset specified signal to the default disposition.
1060  */
1061 static void
1062 sigdflt(struct sigacts *ps, int sig)
1063 {
1064 
1065 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1066 	SIGDELSET(ps->ps_sigcatch, sig);
1067 	if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT)
1068 		SIGADDSET(ps->ps_sigignore, sig);
1069 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1070 	SIGDELSET(ps->ps_siginfo, sig);
1071 }
1072 
1073 /*
1074  * Reset signals for an exec of the specified process.
1075  */
1076 void
1077 execsigs(struct proc *p)
1078 {
1079 	struct sigacts *ps;
1080 	struct thread *td;
1081 
1082 	/*
1083 	 * Reset caught signals.  Held signals remain held
1084 	 * through td_sigmask (unless they were caught,
1085 	 * and are now ignored by default).
1086 	 */
1087 	PROC_LOCK_ASSERT(p, MA_OWNED);
1088 	ps = p->p_sigacts;
1089 	mtx_lock(&ps->ps_mtx);
1090 	sig_drop_caught(p);
1091 
1092 	/*
1093 	 * Reset stack state to the user stack.
1094 	 * Clear set of signals caught on the signal stack.
1095 	 */
1096 	td = curthread;
1097 	MPASS(td->td_proc == p);
1098 	td->td_sigstk.ss_flags = SS_DISABLE;
1099 	td->td_sigstk.ss_size = 0;
1100 	td->td_sigstk.ss_sp = 0;
1101 	td->td_pflags &= ~TDP_ALTSTACK;
1102 	/*
1103 	 * Reset no zombies if child dies flag as Solaris does.
1104 	 */
1105 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1106 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1107 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1108 	mtx_unlock(&ps->ps_mtx);
1109 }
1110 
1111 /*
1112  * kern_sigprocmask()
1113  *
1114  *	Manipulate signal mask.
1115  */
1116 int
1117 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1118     int flags)
1119 {
1120 	sigset_t new_block, oset1;
1121 	struct proc *p;
1122 	int error;
1123 
1124 	p = td->td_proc;
1125 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1126 		PROC_LOCK_ASSERT(p, MA_OWNED);
1127 	else
1128 		PROC_LOCK(p);
1129 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1130 	    ? MA_OWNED : MA_NOTOWNED);
1131 	if (oset != NULL)
1132 		*oset = td->td_sigmask;
1133 
1134 	error = 0;
1135 	if (set != NULL) {
1136 		switch (how) {
1137 		case SIG_BLOCK:
1138 			SIG_CANTMASK(*set);
1139 			oset1 = td->td_sigmask;
1140 			SIGSETOR(td->td_sigmask, *set);
1141 			new_block = td->td_sigmask;
1142 			SIGSETNAND(new_block, oset1);
1143 			break;
1144 		case SIG_UNBLOCK:
1145 			SIGSETNAND(td->td_sigmask, *set);
1146 			signotify(td);
1147 			goto out;
1148 		case SIG_SETMASK:
1149 			SIG_CANTMASK(*set);
1150 			oset1 = td->td_sigmask;
1151 			if (flags & SIGPROCMASK_OLD)
1152 				SIGSETLO(td->td_sigmask, *set);
1153 			else
1154 				td->td_sigmask = *set;
1155 			new_block = td->td_sigmask;
1156 			SIGSETNAND(new_block, oset1);
1157 			signotify(td);
1158 			break;
1159 		default:
1160 			error = EINVAL;
1161 			goto out;
1162 		}
1163 
1164 		/*
1165 		 * The new_block set contains signals that were not previously
1166 		 * blocked, but are blocked now.
1167 		 *
1168 		 * In case we block any signal that was not previously blocked
1169 		 * for td, and process has the signal pending, try to schedule
1170 		 * signal delivery to some thread that does not block the
1171 		 * signal, possibly waking it up.
1172 		 */
1173 		if (p->p_numthreads != 1)
1174 			reschedule_signals(p, new_block, flags);
1175 	}
1176 
1177 out:
1178 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1179 		PROC_UNLOCK(p);
1180 	return (error);
1181 }
1182 
1183 #ifndef _SYS_SYSPROTO_H_
1184 struct sigprocmask_args {
1185 	int	how;
1186 	const sigset_t *set;
1187 	sigset_t *oset;
1188 };
1189 #endif
1190 int
1191 sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1192 {
1193 	sigset_t set, oset;
1194 	sigset_t *setp, *osetp;
1195 	int error;
1196 
1197 	setp = (uap->set != NULL) ? &set : NULL;
1198 	osetp = (uap->oset != NULL) ? &oset : NULL;
1199 	if (setp) {
1200 		error = copyin(uap->set, setp, sizeof(set));
1201 		if (error)
1202 			return (error);
1203 	}
1204 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1205 	if (osetp && !error) {
1206 		error = copyout(osetp, uap->oset, sizeof(oset));
1207 	}
1208 	return (error);
1209 }
1210 
1211 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1212 #ifndef _SYS_SYSPROTO_H_
1213 struct osigprocmask_args {
1214 	int	how;
1215 	osigset_t mask;
1216 };
1217 #endif
1218 int
1219 osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1220 {
1221 	sigset_t set, oset;
1222 	int error;
1223 
1224 	OSIG2SIG(uap->mask, set);
1225 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1226 	SIG2OSIG(oset, td->td_retval[0]);
1227 	return (error);
1228 }
1229 #endif /* COMPAT_43 */
1230 
1231 int
1232 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1233 {
1234 	ksiginfo_t ksi;
1235 	sigset_t set;
1236 	int error;
1237 
1238 	error = copyin(uap->set, &set, sizeof(set));
1239 	if (error) {
1240 		td->td_retval[0] = error;
1241 		return (0);
1242 	}
1243 
1244 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1245 	if (error) {
1246 		/*
1247 		 * sigwait() function shall not return EINTR, but
1248 		 * the syscall does.  Non-ancient libc provides the
1249 		 * wrapper which hides EINTR.  Otherwise, EINTR return
1250 		 * is used by libthr to handle required cancellation
1251 		 * point in the sigwait().
1252 		 */
1253 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1254 			return (ERESTART);
1255 		td->td_retval[0] = error;
1256 		return (0);
1257 	}
1258 
1259 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1260 	td->td_retval[0] = error;
1261 	return (0);
1262 }
1263 
1264 int
1265 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1266 {
1267 	struct timespec ts;
1268 	struct timespec *timeout;
1269 	sigset_t set;
1270 	ksiginfo_t ksi;
1271 	int error;
1272 
1273 	if (uap->timeout) {
1274 		error = copyin(uap->timeout, &ts, sizeof(ts));
1275 		if (error)
1276 			return (error);
1277 
1278 		timeout = &ts;
1279 	} else
1280 		timeout = NULL;
1281 
1282 	error = copyin(uap->set, &set, sizeof(set));
1283 	if (error)
1284 		return (error);
1285 
1286 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1287 	if (error)
1288 		return (error);
1289 
1290 	if (uap->info)
1291 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1292 
1293 	if (error == 0)
1294 		td->td_retval[0] = ksi.ksi_signo;
1295 	return (error);
1296 }
1297 
1298 int
1299 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1300 {
1301 	ksiginfo_t ksi;
1302 	sigset_t set;
1303 	int error;
1304 
1305 	error = copyin(uap->set, &set, sizeof(set));
1306 	if (error)
1307 		return (error);
1308 
1309 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1310 	if (error)
1311 		return (error);
1312 
1313 	if (uap->info)
1314 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1315 
1316 	if (error == 0)
1317 		td->td_retval[0] = ksi.ksi_signo;
1318 	return (error);
1319 }
1320 
1321 static void
1322 proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1323 {
1324 	struct thread *thr;
1325 
1326 	FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1327 		if (thr == td)
1328 			thr->td_si = *si;
1329 		else
1330 			thr->td_si.si_signo = 0;
1331 	}
1332 }
1333 
1334 int
1335 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1336 	struct timespec *timeout)
1337 {
1338 	struct sigacts *ps;
1339 	sigset_t saved_mask, new_block;
1340 	struct proc *p;
1341 	int error, sig, timevalid = 0;
1342 	sbintime_t sbt, precision, tsbt;
1343 	struct timespec ts;
1344 	bool traced;
1345 
1346 	p = td->td_proc;
1347 	error = 0;
1348 	traced = false;
1349 
1350 	/* Ensure the sigfastblock value is up to date. */
1351 	sigfastblock_fetch(td);
1352 
1353 	if (timeout != NULL) {
1354 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1355 			timevalid = 1;
1356 			ts = *timeout;
1357 			if (ts.tv_sec < INT32_MAX / 2) {
1358 				tsbt = tstosbt(ts);
1359 				precision = tsbt;
1360 				precision >>= tc_precexp;
1361 				if (TIMESEL(&sbt, tsbt))
1362 					sbt += tc_tick_sbt;
1363 				sbt += tsbt;
1364 			} else
1365 				precision = sbt = 0;
1366 		}
1367 	} else
1368 		precision = sbt = 0;
1369 	ksiginfo_init(ksi);
1370 	/* Some signals can not be waited for. */
1371 	SIG_CANTMASK(waitset);
1372 	ps = p->p_sigacts;
1373 	PROC_LOCK(p);
1374 	saved_mask = td->td_sigmask;
1375 	SIGSETNAND(td->td_sigmask, waitset);
1376 	if ((p->p_sysent->sv_flags & SV_SIG_DISCIGN) != 0 ||
1377 	    !kern_sig_discard_ign) {
1378 		thread_lock(td);
1379 		td->td_flags |= TDF_SIGWAIT;
1380 		thread_unlock(td);
1381 	}
1382 	for (;;) {
1383 		mtx_lock(&ps->ps_mtx);
1384 		sig = cursig(td);
1385 		mtx_unlock(&ps->ps_mtx);
1386 		KASSERT(sig >= 0, ("sig %d", sig));
1387 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1388 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1389 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1390 				error = 0;
1391 				break;
1392 			}
1393 		}
1394 
1395 		if (error != 0)
1396 			break;
1397 
1398 		/*
1399 		 * POSIX says this must be checked after looking for pending
1400 		 * signals.
1401 		 */
1402 		if (timeout != NULL && !timevalid) {
1403 			error = EINVAL;
1404 			break;
1405 		}
1406 
1407 		if (traced) {
1408 			error = EINTR;
1409 			break;
1410 		}
1411 
1412 		error = msleep_sbt(&p->p_sigacts, &p->p_mtx, PPAUSE | PCATCH,
1413 		    "sigwait", sbt, precision, C_ABSOLUTE);
1414 
1415 		/* The syscalls can not be restarted. */
1416 		if (error == ERESTART)
1417 			error = EINTR;
1418 
1419 		/*
1420 		 * If PTRACE_SCE or PTRACE_SCX were set after
1421 		 * userspace entered the syscall, return spurious
1422 		 * EINTR after wait was done.  Only do this as last
1423 		 * resort after rechecking for possible queued signals
1424 		 * and expired timeouts.
1425 		 */
1426 		if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1427 			traced = true;
1428 	}
1429 	thread_lock(td);
1430 	td->td_flags &= ~TDF_SIGWAIT;
1431 	thread_unlock(td);
1432 
1433 	new_block = saved_mask;
1434 	SIGSETNAND(new_block, td->td_sigmask);
1435 	td->td_sigmask = saved_mask;
1436 	/*
1437 	 * Fewer signals can be delivered to us, reschedule signal
1438 	 * notification.
1439 	 */
1440 	if (p->p_numthreads != 1)
1441 		reschedule_signals(p, new_block, 0);
1442 
1443 	if (error == 0) {
1444 		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1445 
1446 		if (ksi->ksi_code == SI_TIMER)
1447 			itimer_accept(p, ksi->ksi_timerid, ksi);
1448 
1449 #ifdef KTRACE
1450 		if (KTRPOINT(td, KTR_PSIG)) {
1451 			sig_t action;
1452 
1453 			mtx_lock(&ps->ps_mtx);
1454 			action = ps->ps_sigact[_SIG_IDX(sig)];
1455 			mtx_unlock(&ps->ps_mtx);
1456 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1457 		}
1458 #endif
1459 		if (sig == SIGKILL) {
1460 			proc_td_siginfo_capture(td, &ksi->ksi_info);
1461 			sigexit(td, sig);
1462 		}
1463 	}
1464 	PROC_UNLOCK(p);
1465 	return (error);
1466 }
1467 
1468 #ifndef _SYS_SYSPROTO_H_
1469 struct sigpending_args {
1470 	sigset_t	*set;
1471 };
1472 #endif
1473 int
1474 sys_sigpending(struct thread *td, struct sigpending_args *uap)
1475 {
1476 	struct proc *p = td->td_proc;
1477 	sigset_t pending;
1478 
1479 	PROC_LOCK(p);
1480 	pending = p->p_sigqueue.sq_signals;
1481 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1482 	PROC_UNLOCK(p);
1483 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1484 }
1485 
1486 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1487 #ifndef _SYS_SYSPROTO_H_
1488 struct osigpending_args {
1489 	int	dummy;
1490 };
1491 #endif
1492 int
1493 osigpending(struct thread *td, struct osigpending_args *uap)
1494 {
1495 	struct proc *p = td->td_proc;
1496 	sigset_t pending;
1497 
1498 	PROC_LOCK(p);
1499 	pending = p->p_sigqueue.sq_signals;
1500 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1501 	PROC_UNLOCK(p);
1502 	SIG2OSIG(pending, td->td_retval[0]);
1503 	return (0);
1504 }
1505 #endif /* COMPAT_43 */
1506 
1507 #if defined(COMPAT_43)
1508 /*
1509  * Generalized interface signal handler, 4.3-compatible.
1510  */
1511 #ifndef _SYS_SYSPROTO_H_
1512 struct osigvec_args {
1513 	int	signum;
1514 	struct	sigvec *nsv;
1515 	struct	sigvec *osv;
1516 };
1517 #endif
1518 /* ARGSUSED */
1519 int
1520 osigvec(struct thread *td, struct osigvec_args *uap)
1521 {
1522 	struct sigvec vec;
1523 	struct sigaction nsa, osa;
1524 	struct sigaction *nsap, *osap;
1525 	int error;
1526 
1527 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1528 		return (EINVAL);
1529 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1530 	osap = (uap->osv != NULL) ? &osa : NULL;
1531 	if (nsap) {
1532 		error = copyin(uap->nsv, &vec, sizeof(vec));
1533 		if (error)
1534 			return (error);
1535 		nsap->sa_handler = vec.sv_handler;
1536 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1537 		nsap->sa_flags = vec.sv_flags;
1538 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1539 	}
1540 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1541 	if (osap && !error) {
1542 		vec.sv_handler = osap->sa_handler;
1543 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1544 		vec.sv_flags = osap->sa_flags;
1545 		vec.sv_flags &= ~SA_NOCLDWAIT;
1546 		vec.sv_flags ^= SA_RESTART;
1547 		error = copyout(&vec, uap->osv, sizeof(vec));
1548 	}
1549 	return (error);
1550 }
1551 
1552 #ifndef _SYS_SYSPROTO_H_
1553 struct osigblock_args {
1554 	int	mask;
1555 };
1556 #endif
1557 int
1558 osigblock(struct thread *td, struct osigblock_args *uap)
1559 {
1560 	sigset_t set, oset;
1561 
1562 	OSIG2SIG(uap->mask, set);
1563 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1564 	SIG2OSIG(oset, td->td_retval[0]);
1565 	return (0);
1566 }
1567 
1568 #ifndef _SYS_SYSPROTO_H_
1569 struct osigsetmask_args {
1570 	int	mask;
1571 };
1572 #endif
1573 int
1574 osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1575 {
1576 	sigset_t set, oset;
1577 
1578 	OSIG2SIG(uap->mask, set);
1579 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1580 	SIG2OSIG(oset, td->td_retval[0]);
1581 	return (0);
1582 }
1583 #endif /* COMPAT_43 */
1584 
1585 /*
1586  * Suspend calling thread until signal, providing mask to be set in the
1587  * meantime.
1588  */
1589 #ifndef _SYS_SYSPROTO_H_
1590 struct sigsuspend_args {
1591 	const sigset_t *sigmask;
1592 };
1593 #endif
1594 /* ARGSUSED */
1595 int
1596 sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1597 {
1598 	sigset_t mask;
1599 	int error;
1600 
1601 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1602 	if (error)
1603 		return (error);
1604 	return (kern_sigsuspend(td, mask));
1605 }
1606 
1607 int
1608 kern_sigsuspend(struct thread *td, sigset_t mask)
1609 {
1610 	struct proc *p = td->td_proc;
1611 	int has_sig, sig;
1612 
1613 	/* Ensure the sigfastblock value is up to date. */
1614 	sigfastblock_fetch(td);
1615 
1616 	/*
1617 	 * When returning from sigsuspend, we want
1618 	 * the old mask to be restored after the
1619 	 * signal handler has finished.  Thus, we
1620 	 * save it here and mark the sigacts structure
1621 	 * to indicate this.
1622 	 */
1623 	PROC_LOCK(p);
1624 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1625 	    SIGPROCMASK_PROC_LOCKED);
1626 	td->td_pflags |= TDP_OLDMASK;
1627 	ast_sched(td, TDA_SIGSUSPEND);
1628 
1629 	/*
1630 	 * Process signals now. Otherwise, we can get spurious wakeup
1631 	 * due to signal entered process queue, but delivered to other
1632 	 * thread. But sigsuspend should return only on signal
1633 	 * delivery.
1634 	 */
1635 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1636 	for (has_sig = 0; !has_sig;) {
1637 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1638 			0) == 0)
1639 			/* void */;
1640 		thread_suspend_check(0);
1641 		mtx_lock(&p->p_sigacts->ps_mtx);
1642 		while ((sig = cursig(td)) != 0) {
1643 			KASSERT(sig >= 0, ("sig %d", sig));
1644 			has_sig += postsig(sig);
1645 		}
1646 		mtx_unlock(&p->p_sigacts->ps_mtx);
1647 
1648 		/*
1649 		 * If PTRACE_SCE or PTRACE_SCX were set after
1650 		 * userspace entered the syscall, return spurious
1651 		 * EINTR.
1652 		 */
1653 		if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1654 			has_sig += 1;
1655 	}
1656 	PROC_UNLOCK(p);
1657 	td->td_errno = EINTR;
1658 	td->td_pflags |= TDP_NERRNO;
1659 	return (EJUSTRETURN);
1660 }
1661 
1662 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1663 /*
1664  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1665  * convention: libc stub passes mask, not pointer, to save a copyin.
1666  */
1667 #ifndef _SYS_SYSPROTO_H_
1668 struct osigsuspend_args {
1669 	osigset_t mask;
1670 };
1671 #endif
1672 /* ARGSUSED */
1673 int
1674 osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1675 {
1676 	sigset_t mask;
1677 
1678 	OSIG2SIG(uap->mask, mask);
1679 	return (kern_sigsuspend(td, mask));
1680 }
1681 #endif /* COMPAT_43 */
1682 
1683 #if defined(COMPAT_43)
1684 #ifndef _SYS_SYSPROTO_H_
1685 struct osigstack_args {
1686 	struct	sigstack *nss;
1687 	struct	sigstack *oss;
1688 };
1689 #endif
1690 /* ARGSUSED */
1691 int
1692 osigstack(struct thread *td, struct osigstack_args *uap)
1693 {
1694 	struct sigstack nss, oss;
1695 	int error = 0;
1696 
1697 	if (uap->nss != NULL) {
1698 		error = copyin(uap->nss, &nss, sizeof(nss));
1699 		if (error)
1700 			return (error);
1701 	}
1702 	oss.ss_sp = td->td_sigstk.ss_sp;
1703 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1704 	if (uap->nss != NULL) {
1705 		td->td_sigstk.ss_sp = nss.ss_sp;
1706 		td->td_sigstk.ss_size = 0;
1707 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1708 		td->td_pflags |= TDP_ALTSTACK;
1709 	}
1710 	if (uap->oss != NULL)
1711 		error = copyout(&oss, uap->oss, sizeof(oss));
1712 
1713 	return (error);
1714 }
1715 #endif /* COMPAT_43 */
1716 
1717 #ifndef _SYS_SYSPROTO_H_
1718 struct sigaltstack_args {
1719 	stack_t	*ss;
1720 	stack_t	*oss;
1721 };
1722 #endif
1723 /* ARGSUSED */
1724 int
1725 sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1726 {
1727 	stack_t ss, oss;
1728 	int error;
1729 
1730 	if (uap->ss != NULL) {
1731 		error = copyin(uap->ss, &ss, sizeof(ss));
1732 		if (error)
1733 			return (error);
1734 	}
1735 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1736 	    (uap->oss != NULL) ? &oss : NULL);
1737 	if (error)
1738 		return (error);
1739 	if (uap->oss != NULL)
1740 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1741 	return (error);
1742 }
1743 
1744 int
1745 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1746 {
1747 	struct proc *p = td->td_proc;
1748 	int oonstack;
1749 
1750 	oonstack = sigonstack(cpu_getstack(td));
1751 
1752 	if (oss != NULL) {
1753 		*oss = td->td_sigstk;
1754 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1755 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1756 	}
1757 
1758 	if (ss != NULL) {
1759 		if (oonstack)
1760 			return (EPERM);
1761 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1762 			return (EINVAL);
1763 		if (!(ss->ss_flags & SS_DISABLE)) {
1764 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1765 				return (ENOMEM);
1766 
1767 			td->td_sigstk = *ss;
1768 			td->td_pflags |= TDP_ALTSTACK;
1769 		} else {
1770 			td->td_pflags &= ~TDP_ALTSTACK;
1771 		}
1772 	}
1773 	return (0);
1774 }
1775 
1776 struct killpg1_ctx {
1777 	struct thread *td;
1778 	ksiginfo_t *ksi;
1779 	int sig;
1780 	bool sent;
1781 	bool found;
1782 	int ret;
1783 };
1784 
1785 static void
1786 killpg1_sendsig_locked(struct proc *p, struct killpg1_ctx *arg)
1787 {
1788 	int err;
1789 
1790 	err = p_cansignal(arg->td, p, arg->sig);
1791 	if (err == 0 && arg->sig != 0)
1792 		pksignal(p, arg->sig, arg->ksi);
1793 	if (err != ESRCH)
1794 		arg->found = true;
1795 	if (err == 0)
1796 		arg->sent = true;
1797 	else if (arg->ret == 0 && err != ESRCH && err != EPERM)
1798 		arg->ret = err;
1799 }
1800 
1801 static void
1802 killpg1_sendsig(struct proc *p, bool notself, struct killpg1_ctx *arg)
1803 {
1804 
1805 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1806 	    (notself && p == arg->td->td_proc) || p->p_state == PRS_NEW)
1807 		return;
1808 
1809 	PROC_LOCK(p);
1810 	killpg1_sendsig_locked(p, arg);
1811 	PROC_UNLOCK(p);
1812 }
1813 
1814 static void
1815 kill_processes_prison_cb(struct proc *p, void *arg)
1816 {
1817 	struct killpg1_ctx *ctx = arg;
1818 
1819 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1820 	    (p == ctx->td->td_proc) || p->p_state == PRS_NEW)
1821 		return;
1822 
1823 	killpg1_sendsig_locked(p, ctx);
1824 }
1825 
1826 /*
1827  * Common code for kill process group/broadcast kill.
1828  * cp is calling process.
1829  */
1830 static int
1831 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1832 {
1833 	struct proc *p;
1834 	struct pgrp *pgrp;
1835 	struct killpg1_ctx arg;
1836 
1837 	arg.td = td;
1838 	arg.ksi = ksi;
1839 	arg.sig = sig;
1840 	arg.sent = false;
1841 	arg.found = false;
1842 	arg.ret = 0;
1843 	if (all) {
1844 		/*
1845 		 * broadcast
1846 		 */
1847 		prison_proc_iterate(td->td_ucred->cr_prison,
1848 		    kill_processes_prison_cb, &arg);
1849 	} else {
1850 		sx_slock(&proctree_lock);
1851 		if (pgid == 0) {
1852 			/*
1853 			 * zero pgid means send to my process group.
1854 			 */
1855 			pgrp = td->td_proc->p_pgrp;
1856 			PGRP_LOCK(pgrp);
1857 		} else {
1858 			pgrp = pgfind(pgid);
1859 			if (pgrp == NULL) {
1860 				sx_sunlock(&proctree_lock);
1861 				return (ESRCH);
1862 			}
1863 		}
1864 		sx_sunlock(&proctree_lock);
1865 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1866 			killpg1_sendsig(p, false, &arg);
1867 		}
1868 		PGRP_UNLOCK(pgrp);
1869 	}
1870 	MPASS(arg.ret != 0 || arg.found || !arg.sent);
1871 	if (arg.ret == 0 && !arg.sent)
1872 		arg.ret = arg.found ? EPERM : ESRCH;
1873 	return (arg.ret);
1874 }
1875 
1876 #ifndef _SYS_SYSPROTO_H_
1877 struct kill_args {
1878 	int	pid;
1879 	int	signum;
1880 };
1881 #endif
1882 /* ARGSUSED */
1883 int
1884 sys_kill(struct thread *td, struct kill_args *uap)
1885 {
1886 
1887 	return (kern_kill(td, uap->pid, uap->signum));
1888 }
1889 
1890 int
1891 kern_kill(struct thread *td, pid_t pid, int signum)
1892 {
1893 	ksiginfo_t ksi;
1894 	struct proc *p;
1895 	int error;
1896 
1897 	/*
1898 	 * A process in capability mode can send signals only to himself.
1899 	 * The main rationale behind this is that abort(3) is implemented as
1900 	 * kill(getpid(), SIGABRT).
1901 	 */
1902 	if (IN_CAPABILITY_MODE(td) && pid != td->td_proc->p_pid)
1903 		return (ECAPMODE);
1904 
1905 	AUDIT_ARG_SIGNUM(signum);
1906 	AUDIT_ARG_PID(pid);
1907 	if ((u_int)signum > _SIG_MAXSIG)
1908 		return (EINVAL);
1909 
1910 	ksiginfo_init(&ksi);
1911 	ksi.ksi_signo = signum;
1912 	ksi.ksi_code = SI_USER;
1913 	ksi.ksi_pid = td->td_proc->p_pid;
1914 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1915 
1916 	if (pid > 0) {
1917 		/* kill single process */
1918 		if ((p = pfind_any(pid)) == NULL)
1919 			return (ESRCH);
1920 		AUDIT_ARG_PROCESS(p);
1921 		error = p_cansignal(td, p, signum);
1922 		if (error == 0 && signum)
1923 			pksignal(p, signum, &ksi);
1924 		PROC_UNLOCK(p);
1925 		return (error);
1926 	}
1927 	switch (pid) {
1928 	case -1:		/* broadcast signal */
1929 		return (killpg1(td, signum, 0, 1, &ksi));
1930 	case 0:			/* signal own process group */
1931 		return (killpg1(td, signum, 0, 0, &ksi));
1932 	default:		/* negative explicit process group */
1933 		return (killpg1(td, signum, -pid, 0, &ksi));
1934 	}
1935 	/* NOTREACHED */
1936 }
1937 
1938 int
1939 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1940 {
1941 	struct proc *p;
1942 	int error;
1943 
1944 	AUDIT_ARG_SIGNUM(uap->signum);
1945 	AUDIT_ARG_FD(uap->fd);
1946 	if ((u_int)uap->signum > _SIG_MAXSIG)
1947 		return (EINVAL);
1948 
1949 	error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1950 	if (error)
1951 		return (error);
1952 	AUDIT_ARG_PROCESS(p);
1953 	error = p_cansignal(td, p, uap->signum);
1954 	if (error == 0 && uap->signum)
1955 		kern_psignal(p, uap->signum);
1956 	PROC_UNLOCK(p);
1957 	return (error);
1958 }
1959 
1960 #if defined(COMPAT_43)
1961 #ifndef _SYS_SYSPROTO_H_
1962 struct okillpg_args {
1963 	int	pgid;
1964 	int	signum;
1965 };
1966 #endif
1967 /* ARGSUSED */
1968 int
1969 okillpg(struct thread *td, struct okillpg_args *uap)
1970 {
1971 	ksiginfo_t ksi;
1972 
1973 	AUDIT_ARG_SIGNUM(uap->signum);
1974 	AUDIT_ARG_PID(uap->pgid);
1975 	if ((u_int)uap->signum > _SIG_MAXSIG)
1976 		return (EINVAL);
1977 
1978 	ksiginfo_init(&ksi);
1979 	ksi.ksi_signo = uap->signum;
1980 	ksi.ksi_code = SI_USER;
1981 	ksi.ksi_pid = td->td_proc->p_pid;
1982 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1983 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1984 }
1985 #endif /* COMPAT_43 */
1986 
1987 #ifndef _SYS_SYSPROTO_H_
1988 struct sigqueue_args {
1989 	pid_t pid;
1990 	int signum;
1991 	/* union sigval */ void *value;
1992 };
1993 #endif
1994 int
1995 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1996 {
1997 	union sigval sv;
1998 
1999 	sv.sival_ptr = uap->value;
2000 
2001 	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
2002 }
2003 
2004 int
2005 kern_sigqueue(struct thread *td, pid_t pid, int signum, union sigval *value)
2006 {
2007 	ksiginfo_t ksi;
2008 	struct proc *p;
2009 	int error;
2010 
2011 	if ((u_int)signum > _SIG_MAXSIG)
2012 		return (EINVAL);
2013 
2014 	/*
2015 	 * Specification says sigqueue can only send signal to
2016 	 * single process.
2017 	 */
2018 	if (pid <= 0)
2019 		return (EINVAL);
2020 
2021 	if ((p = pfind_any(pid)) == NULL)
2022 		return (ESRCH);
2023 	error = p_cansignal(td, p, signum);
2024 	if (error == 0 && signum != 0) {
2025 		ksiginfo_init(&ksi);
2026 		ksi.ksi_flags = KSI_SIGQ;
2027 		ksi.ksi_signo = signum;
2028 		ksi.ksi_code = SI_QUEUE;
2029 		ksi.ksi_pid = td->td_proc->p_pid;
2030 		ksi.ksi_uid = td->td_ucred->cr_ruid;
2031 		ksi.ksi_value = *value;
2032 		error = pksignal(p, ksi.ksi_signo, &ksi);
2033 	}
2034 	PROC_UNLOCK(p);
2035 	return (error);
2036 }
2037 
2038 /*
2039  * Send a signal to a process group.
2040  */
2041 void
2042 gsignal(int pgid, int sig, ksiginfo_t *ksi)
2043 {
2044 	struct pgrp *pgrp;
2045 
2046 	if (pgid != 0) {
2047 		sx_slock(&proctree_lock);
2048 		pgrp = pgfind(pgid);
2049 		sx_sunlock(&proctree_lock);
2050 		if (pgrp != NULL) {
2051 			pgsignal(pgrp, sig, 0, ksi);
2052 			PGRP_UNLOCK(pgrp);
2053 		}
2054 	}
2055 }
2056 
2057 /*
2058  * Send a signal to a process group.  If checktty is 1,
2059  * limit to members which have a controlling terminal.
2060  */
2061 void
2062 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
2063 {
2064 	struct proc *p;
2065 
2066 	if (pgrp) {
2067 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
2068 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
2069 			PROC_LOCK(p);
2070 			if (p->p_state == PRS_NORMAL &&
2071 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
2072 				pksignal(p, sig, ksi);
2073 			PROC_UNLOCK(p);
2074 		}
2075 	}
2076 }
2077 
2078 /*
2079  * Recalculate the signal mask and reset the signal disposition after
2080  * usermode frame for delivery is formed.  Should be called after
2081  * mach-specific routine, because sysent->sv_sendsig() needs correct
2082  * ps_siginfo and signal mask.
2083  */
2084 static void
2085 postsig_done(int sig, struct thread *td, struct sigacts *ps)
2086 {
2087 	sigset_t mask;
2088 
2089 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2090 	td->td_ru.ru_nsignals++;
2091 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
2092 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
2093 		SIGADDSET(mask, sig);
2094 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
2095 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
2096 	if (SIGISMEMBER(ps->ps_sigreset, sig))
2097 		sigdflt(ps, sig);
2098 }
2099 
2100 /*
2101  * Send a signal caused by a trap to the current thread.  If it will be
2102  * caught immediately, deliver it with correct code.  Otherwise, post it
2103  * normally.
2104  */
2105 void
2106 trapsignal(struct thread *td, ksiginfo_t *ksi)
2107 {
2108 	struct sigacts *ps;
2109 	struct proc *p;
2110 	sigset_t sigmask;
2111 	int sig;
2112 
2113 	p = td->td_proc;
2114 	sig = ksi->ksi_signo;
2115 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
2116 
2117 	sigfastblock_fetch(td);
2118 	PROC_LOCK(p);
2119 	ps = p->p_sigacts;
2120 	mtx_lock(&ps->ps_mtx);
2121 	sigmask = td->td_sigmask;
2122 	if (td->td_sigblock_val != 0)
2123 		SIGSETOR(sigmask, fastblock_mask);
2124 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2125 	    !SIGISMEMBER(sigmask, sig)) {
2126 #ifdef KTRACE
2127 		if (KTRPOINT(curthread, KTR_PSIG))
2128 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2129 			    &td->td_sigmask, ksi->ksi_code);
2130 #endif
2131 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2132 		    ksi, &td->td_sigmask);
2133 		postsig_done(sig, td, ps);
2134 		mtx_unlock(&ps->ps_mtx);
2135 	} else {
2136 		/*
2137 		 * Avoid a possible infinite loop if the thread
2138 		 * masking the signal or process is ignoring the
2139 		 * signal.
2140 		 */
2141 		if (kern_forcesigexit && (SIGISMEMBER(sigmask, sig) ||
2142 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2143 			SIGDELSET(td->td_sigmask, sig);
2144 			SIGDELSET(ps->ps_sigcatch, sig);
2145 			SIGDELSET(ps->ps_sigignore, sig);
2146 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2147 			td->td_pflags &= ~TDP_SIGFASTBLOCK;
2148 			td->td_sigblock_val = 0;
2149 		}
2150 		mtx_unlock(&ps->ps_mtx);
2151 		p->p_sig = sig;		/* XXX to verify code */
2152 		tdsendsignal(p, td, sig, ksi);
2153 	}
2154 	PROC_UNLOCK(p);
2155 }
2156 
2157 static struct thread *
2158 sigtd(struct proc *p, int sig, bool fast_sigblock)
2159 {
2160 	struct thread *td, *signal_td;
2161 
2162 	PROC_LOCK_ASSERT(p, MA_OWNED);
2163 	MPASS(!fast_sigblock || p == curproc);
2164 
2165 	/*
2166 	 * Check if current thread can handle the signal without
2167 	 * switching context to another thread.
2168 	 */
2169 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig) &&
2170 	    (!fast_sigblock || curthread->td_sigblock_val == 0))
2171 		return (curthread);
2172 	signal_td = NULL;
2173 	FOREACH_THREAD_IN_PROC(p, td) {
2174 		if (!SIGISMEMBER(td->td_sigmask, sig) && (!fast_sigblock ||
2175 		    td != curthread || td->td_sigblock_val == 0)) {
2176 			signal_td = td;
2177 			break;
2178 		}
2179 	}
2180 	if (signal_td == NULL)
2181 		signal_td = FIRST_THREAD_IN_PROC(p);
2182 	return (signal_td);
2183 }
2184 
2185 /*
2186  * Send the signal to the process.  If the signal has an action, the action
2187  * is usually performed by the target process rather than the caller; we add
2188  * the signal to the set of pending signals for the process.
2189  *
2190  * Exceptions:
2191  *   o When a stop signal is sent to a sleeping process that takes the
2192  *     default action, the process is stopped without awakening it.
2193  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2194  *     regardless of the signal action (eg, blocked or ignored).
2195  *
2196  * Other ignored signals are discarded immediately.
2197  *
2198  * NB: This function may be entered from the debugger via the "kill" DDB
2199  * command.  There is little that can be done to mitigate the possibly messy
2200  * side effects of this unwise possibility.
2201  */
2202 void
2203 kern_psignal(struct proc *p, int sig)
2204 {
2205 	ksiginfo_t ksi;
2206 
2207 	ksiginfo_init(&ksi);
2208 	ksi.ksi_signo = sig;
2209 	ksi.ksi_code = SI_KERNEL;
2210 	(void) tdsendsignal(p, NULL, sig, &ksi);
2211 }
2212 
2213 int
2214 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2215 {
2216 
2217 	return (tdsendsignal(p, NULL, sig, ksi));
2218 }
2219 
2220 /* Utility function for finding a thread to send signal event to. */
2221 int
2222 sigev_findtd(struct proc *p, struct sigevent *sigev, struct thread **ttd)
2223 {
2224 	struct thread *td;
2225 
2226 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2227 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2228 		if (td == NULL)
2229 			return (ESRCH);
2230 		*ttd = td;
2231 	} else {
2232 		*ttd = NULL;
2233 		PROC_LOCK(p);
2234 	}
2235 	return (0);
2236 }
2237 
2238 void
2239 tdsignal(struct thread *td, int sig)
2240 {
2241 	ksiginfo_t ksi;
2242 
2243 	ksiginfo_init(&ksi);
2244 	ksi.ksi_signo = sig;
2245 	ksi.ksi_code = SI_KERNEL;
2246 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2247 }
2248 
2249 void
2250 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2251 {
2252 
2253 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2254 }
2255 
2256 static int
2257 sig_sleepq_abort(struct thread *td, int intrval)
2258 {
2259 	THREAD_LOCK_ASSERT(td, MA_OWNED);
2260 
2261 	if (intrval == 0 && (td->td_flags & TDF_SIGWAIT) == 0) {
2262 		thread_unlock(td);
2263 		return (0);
2264 	}
2265 	return (sleepq_abort(td, intrval));
2266 }
2267 
2268 int
2269 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2270 {
2271 	sig_t action;
2272 	sigqueue_t *sigqueue;
2273 	int prop;
2274 	struct sigacts *ps;
2275 	int intrval;
2276 	int ret = 0;
2277 	int wakeup_swapper;
2278 
2279 	MPASS(td == NULL || p == td->td_proc);
2280 	PROC_LOCK_ASSERT(p, MA_OWNED);
2281 
2282 	if (!_SIG_VALID(sig))
2283 		panic("%s(): invalid signal %d", __func__, sig);
2284 
2285 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2286 
2287 	/*
2288 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2289 	 */
2290 	if (p->p_state == PRS_ZOMBIE) {
2291 		if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2292 			ksiginfo_tryfree(ksi);
2293 		return (ret);
2294 	}
2295 
2296 	ps = p->p_sigacts;
2297 	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2298 	prop = sigprop(sig);
2299 
2300 	if (td == NULL) {
2301 		td = sigtd(p, sig, false);
2302 		sigqueue = &p->p_sigqueue;
2303 	} else
2304 		sigqueue = &td->td_sigqueue;
2305 
2306 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2307 
2308 	/*
2309 	 * If the signal is being ignored, then we forget about it
2310 	 * immediately, except when the target process executes
2311 	 * sigwait().  (Note: we don't set SIGCONT in ps_sigignore,
2312 	 * and if it is set to SIG_IGN, action will be SIG_DFL here.)
2313 	 */
2314 	mtx_lock(&ps->ps_mtx);
2315 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2316 		if (kern_sig_discard_ign &&
2317 		    (p->p_sysent->sv_flags & SV_SIG_DISCIGN) == 0) {
2318 			SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2319 
2320 			mtx_unlock(&ps->ps_mtx);
2321 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2322 				ksiginfo_tryfree(ksi);
2323 			return (ret);
2324 		} else {
2325 			action = SIG_CATCH;
2326 			intrval = 0;
2327 		}
2328 	} else {
2329 		if (SIGISMEMBER(td->td_sigmask, sig))
2330 			action = SIG_HOLD;
2331 		else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2332 			action = SIG_CATCH;
2333 		else
2334 			action = SIG_DFL;
2335 		if (SIGISMEMBER(ps->ps_sigintr, sig))
2336 			intrval = EINTR;
2337 		else
2338 			intrval = ERESTART;
2339 	}
2340 	mtx_unlock(&ps->ps_mtx);
2341 
2342 	if (prop & SIGPROP_CONT)
2343 		sigqueue_delete_stopmask_proc(p);
2344 	else if (prop & SIGPROP_STOP) {
2345 		/*
2346 		 * If sending a tty stop signal to a member of an orphaned
2347 		 * process group, discard the signal here if the action
2348 		 * is default; don't stop the process below if sleeping,
2349 		 * and don't clear any pending SIGCONT.
2350 		 */
2351 		if ((prop & SIGPROP_TTYSTOP) != 0 &&
2352 		    (p->p_pgrp->pg_flags & PGRP_ORPHANED) != 0 &&
2353 		    action == SIG_DFL) {
2354 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2355 				ksiginfo_tryfree(ksi);
2356 			return (ret);
2357 		}
2358 		sigqueue_delete_proc(p, SIGCONT);
2359 		if (p->p_flag & P_CONTINUED) {
2360 			p->p_flag &= ~P_CONTINUED;
2361 			PROC_LOCK(p->p_pptr);
2362 			sigqueue_take(p->p_ksi);
2363 			PROC_UNLOCK(p->p_pptr);
2364 		}
2365 	}
2366 
2367 	ret = sigqueue_add(sigqueue, sig, ksi);
2368 	if (ret != 0)
2369 		return (ret);
2370 	signotify(td);
2371 	/*
2372 	 * Defer further processing for signals which are held,
2373 	 * except that stopped processes must be continued by SIGCONT.
2374 	 */
2375 	if (action == SIG_HOLD &&
2376 	    !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2377 		return (ret);
2378 
2379 	wakeup_swapper = 0;
2380 
2381 	/*
2382 	 * Some signals have a process-wide effect and a per-thread
2383 	 * component.  Most processing occurs when the process next
2384 	 * tries to cross the user boundary, however there are some
2385 	 * times when processing needs to be done immediately, such as
2386 	 * waking up threads so that they can cross the user boundary.
2387 	 * We try to do the per-process part here.
2388 	 */
2389 	if (P_SHOULDSTOP(p)) {
2390 		KASSERT(!(p->p_flag & P_WEXIT),
2391 		    ("signal to stopped but exiting process"));
2392 		if (sig == SIGKILL) {
2393 			/*
2394 			 * If traced process is already stopped,
2395 			 * then no further action is necessary.
2396 			 */
2397 			if (p->p_flag & P_TRACED)
2398 				goto out;
2399 			/*
2400 			 * SIGKILL sets process running.
2401 			 * It will die elsewhere.
2402 			 * All threads must be restarted.
2403 			 */
2404 			p->p_flag &= ~P_STOPPED_SIG;
2405 			goto runfast;
2406 		}
2407 
2408 		if (prop & SIGPROP_CONT) {
2409 			/*
2410 			 * If traced process is already stopped,
2411 			 * then no further action is necessary.
2412 			 */
2413 			if (p->p_flag & P_TRACED)
2414 				goto out;
2415 			/*
2416 			 * If SIGCONT is default (or ignored), we continue the
2417 			 * process but don't leave the signal in sigqueue as
2418 			 * it has no further action.  If SIGCONT is held, we
2419 			 * continue the process and leave the signal in
2420 			 * sigqueue.  If the process catches SIGCONT, let it
2421 			 * handle the signal itself.  If it isn't waiting on
2422 			 * an event, it goes back to run state.
2423 			 * Otherwise, process goes back to sleep state.
2424 			 */
2425 			p->p_flag &= ~P_STOPPED_SIG;
2426 			PROC_SLOCK(p);
2427 			if (p->p_numthreads == p->p_suspcount) {
2428 				PROC_SUNLOCK(p);
2429 				p->p_flag |= P_CONTINUED;
2430 				p->p_xsig = SIGCONT;
2431 				PROC_LOCK(p->p_pptr);
2432 				childproc_continued(p);
2433 				PROC_UNLOCK(p->p_pptr);
2434 				PROC_SLOCK(p);
2435 			}
2436 			if (action == SIG_DFL) {
2437 				thread_unsuspend(p);
2438 				PROC_SUNLOCK(p);
2439 				sigqueue_delete(sigqueue, sig);
2440 				goto out_cont;
2441 			}
2442 			if (action == SIG_CATCH) {
2443 				/*
2444 				 * The process wants to catch it so it needs
2445 				 * to run at least one thread, but which one?
2446 				 */
2447 				PROC_SUNLOCK(p);
2448 				goto runfast;
2449 			}
2450 			/*
2451 			 * The signal is not ignored or caught.
2452 			 */
2453 			thread_unsuspend(p);
2454 			PROC_SUNLOCK(p);
2455 			goto out_cont;
2456 		}
2457 
2458 		if (prop & SIGPROP_STOP) {
2459 			/*
2460 			 * If traced process is already stopped,
2461 			 * then no further action is necessary.
2462 			 */
2463 			if (p->p_flag & P_TRACED)
2464 				goto out;
2465 			/*
2466 			 * Already stopped, don't need to stop again
2467 			 * (If we did the shell could get confused).
2468 			 * Just make sure the signal STOP bit set.
2469 			 */
2470 			p->p_flag |= P_STOPPED_SIG;
2471 			sigqueue_delete(sigqueue, sig);
2472 			goto out;
2473 		}
2474 
2475 		/*
2476 		 * All other kinds of signals:
2477 		 * If a thread is sleeping interruptibly, simulate a
2478 		 * wakeup so that when it is continued it will be made
2479 		 * runnable and can look at the signal.  However, don't make
2480 		 * the PROCESS runnable, leave it stopped.
2481 		 * It may run a bit until it hits a thread_suspend_check().
2482 		 */
2483 		PROC_SLOCK(p);
2484 		thread_lock(td);
2485 		if (TD_CAN_ABORT(td))
2486 			wakeup_swapper = sig_sleepq_abort(td, intrval);
2487 		else
2488 			thread_unlock(td);
2489 		PROC_SUNLOCK(p);
2490 		goto out;
2491 		/*
2492 		 * Mutexes are short lived. Threads waiting on them will
2493 		 * hit thread_suspend_check() soon.
2494 		 */
2495 	} else if (p->p_state == PRS_NORMAL) {
2496 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2497 			tdsigwakeup(td, sig, action, intrval);
2498 			goto out;
2499 		}
2500 
2501 		MPASS(action == SIG_DFL);
2502 
2503 		if (prop & SIGPROP_STOP) {
2504 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2505 				goto out;
2506 			p->p_flag |= P_STOPPED_SIG;
2507 			p->p_xsig = sig;
2508 			PROC_SLOCK(p);
2509 			wakeup_swapper = sig_suspend_threads(td, p);
2510 			if (p->p_numthreads == p->p_suspcount) {
2511 				/*
2512 				 * only thread sending signal to another
2513 				 * process can reach here, if thread is sending
2514 				 * signal to its process, because thread does
2515 				 * not suspend itself here, p_numthreads
2516 				 * should never be equal to p_suspcount.
2517 				 */
2518 				thread_stopped(p);
2519 				PROC_SUNLOCK(p);
2520 				sigqueue_delete_proc(p, p->p_xsig);
2521 			} else
2522 				PROC_SUNLOCK(p);
2523 			goto out;
2524 		}
2525 	} else {
2526 		/* Not in "NORMAL" state. discard the signal. */
2527 		sigqueue_delete(sigqueue, sig);
2528 		goto out;
2529 	}
2530 
2531 	/*
2532 	 * The process is not stopped so we need to apply the signal to all the
2533 	 * running threads.
2534 	 */
2535 runfast:
2536 	tdsigwakeup(td, sig, action, intrval);
2537 	PROC_SLOCK(p);
2538 	thread_unsuspend(p);
2539 	PROC_SUNLOCK(p);
2540 out_cont:
2541 	itimer_proc_continue(p);
2542 	kqtimer_proc_continue(p);
2543 out:
2544 	/* If we jump here, proc slock should not be owned. */
2545 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2546 	if (wakeup_swapper)
2547 		kick_proc0();
2548 
2549 	return (ret);
2550 }
2551 
2552 /*
2553  * The force of a signal has been directed against a single
2554  * thread.  We need to see what we can do about knocking it
2555  * out of any sleep it may be in etc.
2556  */
2557 static void
2558 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2559 {
2560 	struct proc *p = td->td_proc;
2561 	int prop, wakeup_swapper;
2562 
2563 	PROC_LOCK_ASSERT(p, MA_OWNED);
2564 	prop = sigprop(sig);
2565 
2566 	PROC_SLOCK(p);
2567 	thread_lock(td);
2568 	/*
2569 	 * Bring the priority of a thread up if we want it to get
2570 	 * killed in this lifetime.  Be careful to avoid bumping the
2571 	 * priority of the idle thread, since we still allow to signal
2572 	 * kernel processes.
2573 	 */
2574 	if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2575 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2576 		sched_prio(td, PUSER);
2577 	if (TD_ON_SLEEPQ(td)) {
2578 		/*
2579 		 * If thread is sleeping uninterruptibly
2580 		 * we can't interrupt the sleep... the signal will
2581 		 * be noticed when the process returns through
2582 		 * trap() or syscall().
2583 		 */
2584 		if ((td->td_flags & TDF_SINTR) == 0)
2585 			goto out;
2586 		/*
2587 		 * If SIGCONT is default (or ignored) and process is
2588 		 * asleep, we are finished; the process should not
2589 		 * be awakened.
2590 		 */
2591 		if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2592 			thread_unlock(td);
2593 			PROC_SUNLOCK(p);
2594 			sigqueue_delete(&p->p_sigqueue, sig);
2595 			/*
2596 			 * It may be on either list in this state.
2597 			 * Remove from both for now.
2598 			 */
2599 			sigqueue_delete(&td->td_sigqueue, sig);
2600 			return;
2601 		}
2602 
2603 		/*
2604 		 * Don't awaken a sleeping thread for SIGSTOP if the
2605 		 * STOP signal is deferred.
2606 		 */
2607 		if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2608 		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2609 			goto out;
2610 
2611 		/*
2612 		 * Give low priority threads a better chance to run.
2613 		 */
2614 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2615 			sched_prio(td, PUSER);
2616 
2617 		wakeup_swapper = sig_sleepq_abort(td, intrval);
2618 		PROC_SUNLOCK(p);
2619 		if (wakeup_swapper)
2620 			kick_proc0();
2621 		return;
2622 	}
2623 
2624 	/*
2625 	 * Other states do nothing with the signal immediately,
2626 	 * other than kicking ourselves if we are running.
2627 	 * It will either never be noticed, or noticed very soon.
2628 	 */
2629 #ifdef SMP
2630 	if (TD_IS_RUNNING(td) && td != curthread)
2631 		forward_signal(td);
2632 #endif
2633 
2634 out:
2635 	PROC_SUNLOCK(p);
2636 	thread_unlock(td);
2637 }
2638 
2639 static void
2640 ptrace_coredumpreq(struct thread *td, struct proc *p,
2641     struct thr_coredump_req *tcq)
2642 {
2643 	void *rl_cookie;
2644 
2645 	if (p->p_sysent->sv_coredump == NULL) {
2646 		tcq->tc_error = ENOSYS;
2647 		return;
2648 	}
2649 
2650 	rl_cookie = vn_rangelock_wlock(tcq->tc_vp, 0, OFF_MAX);
2651 	tcq->tc_error = p->p_sysent->sv_coredump(td, tcq->tc_vp,
2652 	    tcq->tc_limit, tcq->tc_flags);
2653 	vn_rangelock_unlock(tcq->tc_vp, rl_cookie);
2654 }
2655 
2656 static void
2657 ptrace_syscallreq(struct thread *td, struct proc *p,
2658     struct thr_syscall_req *tsr)
2659 {
2660 	struct sysentvec *sv;
2661 	struct sysent *se;
2662 	register_t rv_saved[2];
2663 	int error, nerror;
2664 	int sc;
2665 	bool audited, sy_thr_static;
2666 
2667 	sv = p->p_sysent;
2668 	if (sv->sv_table == NULL || sv->sv_size < tsr->ts_sa.code) {
2669 		tsr->ts_ret.sr_error = ENOSYS;
2670 		return;
2671 	}
2672 
2673 	sc = tsr->ts_sa.code;
2674 	if (sc == SYS_syscall || sc == SYS___syscall) {
2675 		sc = tsr->ts_sa.args[0];
2676 		memmove(&tsr->ts_sa.args[0], &tsr->ts_sa.args[1],
2677 		    sizeof(register_t) * (tsr->ts_nargs - 1));
2678 	}
2679 
2680 	tsr->ts_sa.callp = se = &sv->sv_table[sc];
2681 
2682 	VM_CNT_INC(v_syscall);
2683 	td->td_pticks = 0;
2684 	if (__predict_false(td->td_cowgen != atomic_load_int(
2685 	    &td->td_proc->p_cowgen)))
2686 		thread_cow_update(td);
2687 
2688 #ifdef CAPABILITY_MODE
2689 	if (IN_CAPABILITY_MODE(td) && (se->sy_flags & SYF_CAPENABLED) == 0) {
2690 		tsr->ts_ret.sr_error = ECAPMODE;
2691 		return;
2692 	}
2693 #endif
2694 
2695 	sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2696 	audited = AUDIT_SYSCALL_ENTER(sc, td) != 0;
2697 
2698 	if (!sy_thr_static) {
2699 		error = syscall_thread_enter(td, se);
2700 		if (error != 0) {
2701 			tsr->ts_ret.sr_error = error;
2702 			return;
2703 		}
2704 	}
2705 
2706 	rv_saved[0] = td->td_retval[0];
2707 	rv_saved[1] = td->td_retval[1];
2708 	nerror = td->td_errno;
2709 	td->td_retval[0] = 0;
2710 	td->td_retval[1] = 0;
2711 
2712 #ifdef KDTRACE_HOOKS
2713 	if (se->sy_entry != 0)
2714 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_ENTRY, 0);
2715 #endif
2716 	tsr->ts_ret.sr_error = se->sy_call(td, tsr->ts_sa.args);
2717 #ifdef KDTRACE_HOOKS
2718 	if (se->sy_return != 0)
2719 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_RETURN,
2720 		    tsr->ts_ret.sr_error != 0 ? -1 : td->td_retval[0]);
2721 #endif
2722 
2723 	tsr->ts_ret.sr_retval[0] = td->td_retval[0];
2724 	tsr->ts_ret.sr_retval[1] = td->td_retval[1];
2725 	td->td_retval[0] = rv_saved[0];
2726 	td->td_retval[1] = rv_saved[1];
2727 	td->td_errno = nerror;
2728 
2729 	if (audited)
2730 		AUDIT_SYSCALL_EXIT(error, td);
2731 	if (!sy_thr_static)
2732 		syscall_thread_exit(td, se);
2733 }
2734 
2735 static void
2736 ptrace_remotereq(struct thread *td, int flag)
2737 {
2738 	struct proc *p;
2739 
2740 	MPASS(td == curthread);
2741 	p = td->td_proc;
2742 	PROC_LOCK_ASSERT(p, MA_OWNED);
2743 	if ((td->td_dbgflags & flag) == 0)
2744 		return;
2745 	KASSERT((p->p_flag & P_STOPPED_TRACE) != 0, ("not stopped"));
2746 	KASSERT(td->td_remotereq != NULL, ("td_remotereq is NULL"));
2747 
2748 	PROC_UNLOCK(p);
2749 	switch (flag) {
2750 	case TDB_COREDUMPREQ:
2751 		ptrace_coredumpreq(td, p, td->td_remotereq);
2752 		break;
2753 	case TDB_SCREMOTEREQ:
2754 		ptrace_syscallreq(td, p, td->td_remotereq);
2755 		break;
2756 	default:
2757 		__unreachable();
2758 	}
2759 	PROC_LOCK(p);
2760 
2761 	MPASS((td->td_dbgflags & flag) != 0);
2762 	td->td_dbgflags &= ~flag;
2763 	td->td_remotereq = NULL;
2764 	wakeup(p);
2765 }
2766 
2767 static int
2768 sig_suspend_threads(struct thread *td, struct proc *p)
2769 {
2770 	struct thread *td2;
2771 	int wakeup_swapper;
2772 
2773 	PROC_LOCK_ASSERT(p, MA_OWNED);
2774 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2775 
2776 	wakeup_swapper = 0;
2777 	FOREACH_THREAD_IN_PROC(p, td2) {
2778 		thread_lock(td2);
2779 		ast_sched_locked(td2, TDA_SUSPEND);
2780 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2781 		    (td2->td_flags & TDF_SINTR)) {
2782 			if (td2->td_flags & TDF_SBDRY) {
2783 				/*
2784 				 * Once a thread is asleep with
2785 				 * TDF_SBDRY and without TDF_SERESTART
2786 				 * or TDF_SEINTR set, it should never
2787 				 * become suspended due to this check.
2788 				 */
2789 				KASSERT(!TD_IS_SUSPENDED(td2),
2790 				    ("thread with deferred stops suspended"));
2791 				if (TD_SBDRY_INTR(td2)) {
2792 					wakeup_swapper |= sleepq_abort(td2,
2793 					    TD_SBDRY_ERRNO(td2));
2794 					continue;
2795 				}
2796 			} else if (!TD_IS_SUSPENDED(td2))
2797 				thread_suspend_one(td2);
2798 		} else if (!TD_IS_SUSPENDED(td2)) {
2799 #ifdef SMP
2800 			if (TD_IS_RUNNING(td2) && td2 != td)
2801 				forward_signal(td2);
2802 #endif
2803 		}
2804 		thread_unlock(td2);
2805 	}
2806 	return (wakeup_swapper);
2807 }
2808 
2809 /*
2810  * Stop the process for an event deemed interesting to the debugger. If si is
2811  * non-NULL, this is a signal exchange; the new signal requested by the
2812  * debugger will be returned for handling. If si is NULL, this is some other
2813  * type of interesting event. The debugger may request a signal be delivered in
2814  * that case as well, however it will be deferred until it can be handled.
2815  */
2816 int
2817 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2818 {
2819 	struct proc *p = td->td_proc;
2820 	struct thread *td2;
2821 	ksiginfo_t ksi;
2822 
2823 	PROC_LOCK_ASSERT(p, MA_OWNED);
2824 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2825 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2826 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2827 
2828 	td->td_xsig = sig;
2829 
2830 	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2831 		td->td_dbgflags |= TDB_XSIG;
2832 		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2833 		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2834 		PROC_SLOCK(p);
2835 		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2836 			if (P_KILLED(p)) {
2837 				/*
2838 				 * Ensure that, if we've been PT_KILLed, the
2839 				 * exit status reflects that. Another thread
2840 				 * may also be in ptracestop(), having just
2841 				 * received the SIGKILL, but this thread was
2842 				 * unsuspended first.
2843 				 */
2844 				td->td_dbgflags &= ~TDB_XSIG;
2845 				td->td_xsig = SIGKILL;
2846 				p->p_ptevents = 0;
2847 				break;
2848 			}
2849 			if (p->p_flag & P_SINGLE_EXIT &&
2850 			    !(td->td_dbgflags & TDB_EXIT)) {
2851 				/*
2852 				 * Ignore ptrace stops except for thread exit
2853 				 * events when the process exits.
2854 				 */
2855 				td->td_dbgflags &= ~TDB_XSIG;
2856 				PROC_SUNLOCK(p);
2857 				return (0);
2858 			}
2859 
2860 			/*
2861 			 * Make wait(2) work.  Ensure that right after the
2862 			 * attach, the thread which was decided to become the
2863 			 * leader of attach gets reported to the waiter.
2864 			 * Otherwise, just avoid overwriting another thread's
2865 			 * assignment to p_xthread.  If another thread has
2866 			 * already set p_xthread, the current thread will get
2867 			 * a chance to report itself upon the next iteration.
2868 			 */
2869 			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2870 			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2871 			    p->p_xthread == NULL)) {
2872 				p->p_xsig = sig;
2873 				p->p_xthread = td;
2874 
2875 				/*
2876 				 * If we are on sleepqueue already,
2877 				 * let sleepqueue code decide if it
2878 				 * needs to go sleep after attach.
2879 				 */
2880 				if (td->td_wchan == NULL)
2881 					td->td_dbgflags &= ~TDB_FSTP;
2882 
2883 				p->p_flag2 &= ~P2_PTRACE_FSTP;
2884 				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2885 				sig_suspend_threads(td, p);
2886 			}
2887 			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2888 				td->td_dbgflags &= ~TDB_STOPATFORK;
2889 			}
2890 stopme:
2891 			td->td_dbgflags |= TDB_SSWITCH;
2892 			thread_suspend_switch(td, p);
2893 			td->td_dbgflags &= ~TDB_SSWITCH;
2894 			if ((td->td_dbgflags & (TDB_COREDUMPREQ |
2895 			    TDB_SCREMOTEREQ)) != 0) {
2896 				MPASS((td->td_dbgflags & (TDB_COREDUMPREQ |
2897 				    TDB_SCREMOTEREQ)) !=
2898 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2899 				PROC_SUNLOCK(p);
2900 				ptrace_remotereq(td, td->td_dbgflags &
2901 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2902 				PROC_SLOCK(p);
2903 				goto stopme;
2904 			}
2905 			if (p->p_xthread == td)
2906 				p->p_xthread = NULL;
2907 			if (!(p->p_flag & P_TRACED))
2908 				break;
2909 			if (td->td_dbgflags & TDB_SUSPEND) {
2910 				if (p->p_flag & P_SINGLE_EXIT)
2911 					break;
2912 				goto stopme;
2913 			}
2914 		}
2915 		PROC_SUNLOCK(p);
2916 	}
2917 
2918 	if (si != NULL && sig == td->td_xsig) {
2919 		/* Parent wants us to take the original signal unchanged. */
2920 		si->ksi_flags |= KSI_HEAD;
2921 		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2922 			si->ksi_signo = 0;
2923 	} else if (td->td_xsig != 0) {
2924 		/*
2925 		 * If parent wants us to take a new signal, then it will leave
2926 		 * it in td->td_xsig; otherwise we just look for signals again.
2927 		 */
2928 		ksiginfo_init(&ksi);
2929 		ksi.ksi_signo = td->td_xsig;
2930 		ksi.ksi_flags |= KSI_PTRACE;
2931 		td2 = sigtd(p, td->td_xsig, false);
2932 		tdsendsignal(p, td2, td->td_xsig, &ksi);
2933 		if (td != td2)
2934 			return (0);
2935 	}
2936 
2937 	return (td->td_xsig);
2938 }
2939 
2940 static void
2941 reschedule_signals(struct proc *p, sigset_t block, int flags)
2942 {
2943 	struct sigacts *ps;
2944 	struct thread *td;
2945 	int sig;
2946 	bool fastblk, pslocked;
2947 
2948 	PROC_LOCK_ASSERT(p, MA_OWNED);
2949 	ps = p->p_sigacts;
2950 	pslocked = (flags & SIGPROCMASK_PS_LOCKED) != 0;
2951 	mtx_assert(&ps->ps_mtx, pslocked ? MA_OWNED : MA_NOTOWNED);
2952 	if (SIGISEMPTY(p->p_siglist))
2953 		return;
2954 	SIGSETAND(block, p->p_siglist);
2955 	fastblk = (flags & SIGPROCMASK_FASTBLK) != 0;
2956 	SIG_FOREACH(sig, &block) {
2957 		td = sigtd(p, sig, fastblk);
2958 
2959 		/*
2960 		 * If sigtd() selected us despite sigfastblock is
2961 		 * blocking, do not activate AST or wake us, to avoid
2962 		 * loop in AST handler.
2963 		 */
2964 		if (fastblk && td == curthread)
2965 			continue;
2966 
2967 		signotify(td);
2968 		if (!pslocked)
2969 			mtx_lock(&ps->ps_mtx);
2970 		if (p->p_flag & P_TRACED ||
2971 		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2972 		    !SIGISMEMBER(td->td_sigmask, sig))) {
2973 			tdsigwakeup(td, sig, SIG_CATCH,
2974 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2975 			    ERESTART));
2976 		}
2977 		if (!pslocked)
2978 			mtx_unlock(&ps->ps_mtx);
2979 	}
2980 }
2981 
2982 void
2983 tdsigcleanup(struct thread *td)
2984 {
2985 	struct proc *p;
2986 	sigset_t unblocked;
2987 
2988 	p = td->td_proc;
2989 	PROC_LOCK_ASSERT(p, MA_OWNED);
2990 
2991 	sigqueue_flush(&td->td_sigqueue);
2992 	if (p->p_numthreads == 1)
2993 		return;
2994 
2995 	/*
2996 	 * Since we cannot handle signals, notify signal post code
2997 	 * about this by filling the sigmask.
2998 	 *
2999 	 * Also, if needed, wake up thread(s) that do not block the
3000 	 * same signals as the exiting thread, since the thread might
3001 	 * have been selected for delivery and woken up.
3002 	 */
3003 	SIGFILLSET(unblocked);
3004 	SIGSETNAND(unblocked, td->td_sigmask);
3005 	SIGFILLSET(td->td_sigmask);
3006 	reschedule_signals(p, unblocked, 0);
3007 
3008 }
3009 
3010 static int
3011 sigdeferstop_curr_flags(int cflags)
3012 {
3013 
3014 	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
3015 	    (cflags & TDF_SBDRY) != 0);
3016 	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
3017 }
3018 
3019 /*
3020  * Defer the delivery of SIGSTOP for the current thread, according to
3021  * the requested mode.  Returns previous flags, which must be restored
3022  * by sigallowstop().
3023  *
3024  * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
3025  * cleared by the current thread, which allow the lock-less read-only
3026  * accesses below.
3027  */
3028 int
3029 sigdeferstop_impl(int mode)
3030 {
3031 	struct thread *td;
3032 	int cflags, nflags;
3033 
3034 	td = curthread;
3035 	cflags = sigdeferstop_curr_flags(td->td_flags);
3036 	switch (mode) {
3037 	case SIGDEFERSTOP_NOP:
3038 		nflags = cflags;
3039 		break;
3040 	case SIGDEFERSTOP_OFF:
3041 		nflags = 0;
3042 		break;
3043 	case SIGDEFERSTOP_SILENT:
3044 		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
3045 		break;
3046 	case SIGDEFERSTOP_EINTR:
3047 		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
3048 		break;
3049 	case SIGDEFERSTOP_ERESTART:
3050 		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
3051 		break;
3052 	default:
3053 		panic("sigdeferstop: invalid mode %x", mode);
3054 		break;
3055 	}
3056 	if (cflags == nflags)
3057 		return (SIGDEFERSTOP_VAL_NCHG);
3058 	thread_lock(td);
3059 	td->td_flags = (td->td_flags & ~cflags) | nflags;
3060 	thread_unlock(td);
3061 	return (cflags);
3062 }
3063 
3064 /*
3065  * Restores the STOP handling mode, typically permitting the delivery
3066  * of SIGSTOP for the current thread.  This does not immediately
3067  * suspend if a stop was posted.  Instead, the thread will suspend
3068  * either via ast() or a subsequent interruptible sleep.
3069  */
3070 void
3071 sigallowstop_impl(int prev)
3072 {
3073 	struct thread *td;
3074 	int cflags;
3075 
3076 	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
3077 	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
3078 	    ("sigallowstop: incorrect previous mode %x", prev));
3079 	td = curthread;
3080 	cflags = sigdeferstop_curr_flags(td->td_flags);
3081 	if (cflags != prev) {
3082 		thread_lock(td);
3083 		td->td_flags = (td->td_flags & ~cflags) | prev;
3084 		thread_unlock(td);
3085 	}
3086 }
3087 
3088 enum sigstatus {
3089 	SIGSTATUS_HANDLE,
3090 	SIGSTATUS_HANDLED,
3091 	SIGSTATUS_IGNORE,
3092 	SIGSTATUS_SBDRY_STOP,
3093 };
3094 
3095 /*
3096  * The thread has signal "sig" pending.  Figure out what to do with it:
3097  *
3098  * _HANDLE     -> the caller should handle the signal
3099  * _HANDLED    -> handled internally, reload pending signal set
3100  * _IGNORE     -> ignored, remove from the set of pending signals and try the
3101  *                next pending signal
3102  * _SBDRY_STOP -> the signal should stop the thread but this is not
3103  *                permitted in the current context
3104  */
3105 static enum sigstatus
3106 sigprocess(struct thread *td, int sig)
3107 {
3108 	struct proc *p;
3109 	struct sigacts *ps;
3110 	struct sigqueue *queue;
3111 	ksiginfo_t ksi;
3112 	int prop;
3113 
3114 	KASSERT(_SIG_VALID(sig), ("%s: invalid signal %d", __func__, sig));
3115 
3116 	p = td->td_proc;
3117 	ps = p->p_sigacts;
3118 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3119 	PROC_LOCK_ASSERT(p, MA_OWNED);
3120 
3121 	/*
3122 	 * We should allow pending but ignored signals below
3123 	 * if there is sigwait() active, or P_TRACED was
3124 	 * on when they were posted.
3125 	 */
3126 	if (SIGISMEMBER(ps->ps_sigignore, sig) &&
3127 	    (p->p_flag & P_TRACED) == 0 &&
3128 	    (td->td_flags & TDF_SIGWAIT) == 0) {
3129 		return (SIGSTATUS_IGNORE);
3130 	}
3131 
3132 	/*
3133 	 * If the process is going to single-thread mode to prepare
3134 	 * for exit, there is no sense in delivering any signal
3135 	 * to usermode.  Another important consequence is that
3136 	 * msleep(..., PCATCH, ...) now is only interruptible by a
3137 	 * suspend request.
3138 	 */
3139 	if ((p->p_flag2 & P2_WEXIT) != 0)
3140 		return (SIGSTATUS_IGNORE);
3141 
3142 	if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
3143 		/*
3144 		 * If traced, always stop.
3145 		 * Remove old signal from queue before the stop.
3146 		 * XXX shrug off debugger, it causes siginfo to
3147 		 * be thrown away.
3148 		 */
3149 		queue = &td->td_sigqueue;
3150 		ksiginfo_init(&ksi);
3151 		if (sigqueue_get(queue, sig, &ksi) == 0) {
3152 			queue = &p->p_sigqueue;
3153 			sigqueue_get(queue, sig, &ksi);
3154 		}
3155 		td->td_si = ksi.ksi_info;
3156 
3157 		mtx_unlock(&ps->ps_mtx);
3158 		sig = ptracestop(td, sig, &ksi);
3159 		mtx_lock(&ps->ps_mtx);
3160 
3161 		td->td_si.si_signo = 0;
3162 
3163 		/*
3164 		 * Keep looking if the debugger discarded or
3165 		 * replaced the signal.
3166 		 */
3167 		if (sig == 0)
3168 			return (SIGSTATUS_HANDLED);
3169 
3170 		/*
3171 		 * If the signal became masked, re-queue it.
3172 		 */
3173 		if (SIGISMEMBER(td->td_sigmask, sig)) {
3174 			ksi.ksi_flags |= KSI_HEAD;
3175 			sigqueue_add(&p->p_sigqueue, sig, &ksi);
3176 			return (SIGSTATUS_HANDLED);
3177 		}
3178 
3179 		/*
3180 		 * If the traced bit got turned off, requeue the signal and
3181 		 * reload the set of pending signals.  This ensures that p_sig*
3182 		 * and p_sigact are consistent.
3183 		 */
3184 		if ((p->p_flag & P_TRACED) == 0) {
3185 			if ((ksi.ksi_flags & KSI_PTRACE) == 0) {
3186 				ksi.ksi_flags |= KSI_HEAD;
3187 				sigqueue_add(queue, sig, &ksi);
3188 			}
3189 			return (SIGSTATUS_HANDLED);
3190 		}
3191 	}
3192 
3193 	/*
3194 	 * Decide whether the signal should be returned.
3195 	 * Return the signal's number, or fall through
3196 	 * to clear it from the pending mask.
3197 	 */
3198 	switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
3199 	case (intptr_t)SIG_DFL:
3200 		/*
3201 		 * Don't take default actions on system processes.
3202 		 */
3203 		if (p->p_pid <= 1) {
3204 #ifdef DIAGNOSTIC
3205 			/*
3206 			 * Are you sure you want to ignore SIGSEGV
3207 			 * in init? XXX
3208 			 */
3209 			printf("Process (pid %lu) got signal %d\n",
3210 				(u_long)p->p_pid, sig);
3211 #endif
3212 			return (SIGSTATUS_IGNORE);
3213 		}
3214 
3215 		/*
3216 		 * If there is a pending stop signal to process with
3217 		 * default action, stop here, then clear the signal.
3218 		 * Traced or exiting processes should ignore stops.
3219 		 * Additionally, a member of an orphaned process group
3220 		 * should ignore tty stops.
3221 		 */
3222 		prop = sigprop(sig);
3223 		if (prop & SIGPROP_STOP) {
3224 			mtx_unlock(&ps->ps_mtx);
3225 			if ((p->p_flag & (P_TRACED | P_WEXIT |
3226 			    P_SINGLE_EXIT)) != 0 || ((p->p_pgrp->
3227 			    pg_flags & PGRP_ORPHANED) != 0 &&
3228 			    (prop & SIGPROP_TTYSTOP) != 0)) {
3229 				mtx_lock(&ps->ps_mtx);
3230 				return (SIGSTATUS_IGNORE);
3231 			}
3232 			if (TD_SBDRY_INTR(td)) {
3233 				KASSERT((td->td_flags & TDF_SBDRY) != 0,
3234 				    ("lost TDF_SBDRY"));
3235 				mtx_lock(&ps->ps_mtx);
3236 				return (SIGSTATUS_SBDRY_STOP);
3237 			}
3238 			WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
3239 			    &p->p_mtx.lock_object, "Catching SIGSTOP");
3240 			sigqueue_delete(&td->td_sigqueue, sig);
3241 			sigqueue_delete(&p->p_sigqueue, sig);
3242 			p->p_flag |= P_STOPPED_SIG;
3243 			p->p_xsig = sig;
3244 			PROC_SLOCK(p);
3245 			sig_suspend_threads(td, p);
3246 			thread_suspend_switch(td, p);
3247 			PROC_SUNLOCK(p);
3248 			mtx_lock(&ps->ps_mtx);
3249 			return (SIGSTATUS_HANDLED);
3250 		} else if ((prop & SIGPROP_IGNORE) != 0 &&
3251 		    (td->td_flags & TDF_SIGWAIT) == 0) {
3252 			/*
3253 			 * Default action is to ignore; drop it if
3254 			 * not in kern_sigtimedwait().
3255 			 */
3256 			return (SIGSTATUS_IGNORE);
3257 		} else {
3258 			return (SIGSTATUS_HANDLE);
3259 		}
3260 
3261 	case (intptr_t)SIG_IGN:
3262 		if ((td->td_flags & TDF_SIGWAIT) == 0)
3263 			return (SIGSTATUS_IGNORE);
3264 		else
3265 			return (SIGSTATUS_HANDLE);
3266 
3267 	default:
3268 		/*
3269 		 * This signal has an action, let postsig() process it.
3270 		 */
3271 		return (SIGSTATUS_HANDLE);
3272 	}
3273 }
3274 
3275 /*
3276  * If the current process has received a signal (should be caught or cause
3277  * termination, should interrupt current syscall), return the signal number.
3278  * Stop signals with default action are processed immediately, then cleared;
3279  * they aren't returned.  This is checked after each entry to the system for
3280  * a syscall or trap (though this can usually be done without calling
3281  * issignal by checking the pending signal masks in cursig.) The normal call
3282  * sequence is
3283  *
3284  *	while (sig = cursig(curthread))
3285  *		postsig(sig);
3286  */
3287 static int
3288 issignal(struct thread *td)
3289 {
3290 	struct proc *p;
3291 	sigset_t sigpending;
3292 	int sig;
3293 
3294 	p = td->td_proc;
3295 	PROC_LOCK_ASSERT(p, MA_OWNED);
3296 
3297 	for (;;) {
3298 		sigpending = td->td_sigqueue.sq_signals;
3299 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
3300 		SIGSETNAND(sigpending, td->td_sigmask);
3301 
3302 		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
3303 		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
3304 			SIG_STOPSIGMASK(sigpending);
3305 		if (SIGISEMPTY(sigpending))	/* no signal to send */
3306 			return (0);
3307 
3308 		/*
3309 		 * Do fast sigblock if requested by usermode.  Since
3310 		 * we do know that there was a signal pending at this
3311 		 * point, set the FAST_SIGBLOCK_PEND as indicator for
3312 		 * usermode to perform a dummy call to
3313 		 * FAST_SIGBLOCK_UNBLOCK, which causes immediate
3314 		 * delivery of postponed pending signal.
3315 		 */
3316 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
3317 			if (td->td_sigblock_val != 0)
3318 				SIGSETNAND(sigpending, fastblock_mask);
3319 			if (SIGISEMPTY(sigpending)) {
3320 				td->td_pflags |= TDP_SIGFASTPENDING;
3321 				return (0);
3322 			}
3323 		}
3324 
3325 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
3326 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
3327 		    SIGISMEMBER(sigpending, SIGSTOP)) {
3328 			/*
3329 			 * If debugger just attached, always consume
3330 			 * SIGSTOP from ptrace(PT_ATTACH) first, to
3331 			 * execute the debugger attach ritual in
3332 			 * order.
3333 			 */
3334 			td->td_dbgflags |= TDB_FSTP;
3335 			SIGEMPTYSET(sigpending);
3336 			SIGADDSET(sigpending, SIGSTOP);
3337 		}
3338 
3339 		SIG_FOREACH(sig, &sigpending) {
3340 			switch (sigprocess(td, sig)) {
3341 			case SIGSTATUS_HANDLE:
3342 				return (sig);
3343 			case SIGSTATUS_HANDLED:
3344 				goto next;
3345 			case SIGSTATUS_IGNORE:
3346 				sigqueue_delete(&td->td_sigqueue, sig);
3347 				sigqueue_delete(&p->p_sigqueue, sig);
3348 				break;
3349 			case SIGSTATUS_SBDRY_STOP:
3350 				return (-1);
3351 			}
3352 		}
3353 next:;
3354 	}
3355 }
3356 
3357 void
3358 thread_stopped(struct proc *p)
3359 {
3360 	int n;
3361 
3362 	PROC_LOCK_ASSERT(p, MA_OWNED);
3363 	PROC_SLOCK_ASSERT(p, MA_OWNED);
3364 	n = p->p_suspcount;
3365 	if (p == curproc)
3366 		n++;
3367 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3368 		PROC_SUNLOCK(p);
3369 		p->p_flag &= ~P_WAITED;
3370 		PROC_LOCK(p->p_pptr);
3371 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
3372 			CLD_TRAPPED : CLD_STOPPED);
3373 		PROC_UNLOCK(p->p_pptr);
3374 		PROC_SLOCK(p);
3375 	}
3376 }
3377 
3378 /*
3379  * Take the action for the specified signal
3380  * from the current set of pending signals.
3381  */
3382 int
3383 postsig(int sig)
3384 {
3385 	struct thread *td;
3386 	struct proc *p;
3387 	struct sigacts *ps;
3388 	sig_t action;
3389 	ksiginfo_t ksi;
3390 	sigset_t returnmask;
3391 
3392 	KASSERT(sig != 0, ("postsig"));
3393 
3394 	td = curthread;
3395 	p = td->td_proc;
3396 	PROC_LOCK_ASSERT(p, MA_OWNED);
3397 	ps = p->p_sigacts;
3398 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3399 	ksiginfo_init(&ksi);
3400 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3401 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3402 		return (0);
3403 	ksi.ksi_signo = sig;
3404 	if (ksi.ksi_code == SI_TIMER)
3405 		itimer_accept(p, ksi.ksi_timerid, &ksi);
3406 	action = ps->ps_sigact[_SIG_IDX(sig)];
3407 #ifdef KTRACE
3408 	if (KTRPOINT(td, KTR_PSIG))
3409 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3410 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3411 #endif
3412 
3413 	if (action == SIG_DFL) {
3414 		/*
3415 		 * Default action, where the default is to kill
3416 		 * the process.  (Other cases were ignored above.)
3417 		 */
3418 		mtx_unlock(&ps->ps_mtx);
3419 		proc_td_siginfo_capture(td, &ksi.ksi_info);
3420 		sigexit(td, sig);
3421 		/* NOTREACHED */
3422 	} else {
3423 		/*
3424 		 * If we get here, the signal must be caught.
3425 		 */
3426 		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3427 		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3428 		    ("postsig action: blocked sig %d", sig));
3429 
3430 		/*
3431 		 * Set the new mask value and also defer further
3432 		 * occurrences of this signal.
3433 		 *
3434 		 * Special case: user has done a sigsuspend.  Here the
3435 		 * current mask is not of interest, but rather the
3436 		 * mask from before the sigsuspend is what we want
3437 		 * restored after the signal processing is completed.
3438 		 */
3439 		if (td->td_pflags & TDP_OLDMASK) {
3440 			returnmask = td->td_oldsigmask;
3441 			td->td_pflags &= ~TDP_OLDMASK;
3442 		} else
3443 			returnmask = td->td_sigmask;
3444 
3445 		if (p->p_sig == sig) {
3446 			p->p_sig = 0;
3447 		}
3448 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3449 		postsig_done(sig, td, ps);
3450 	}
3451 	return (1);
3452 }
3453 
3454 int
3455 sig_ast_checksusp(struct thread *td)
3456 {
3457 	struct proc *p __diagused;
3458 	int ret;
3459 
3460 	p = td->td_proc;
3461 	PROC_LOCK_ASSERT(p, MA_OWNED);
3462 
3463 	if (!td_ast_pending(td, TDA_SUSPEND))
3464 		return (0);
3465 
3466 	ret = thread_suspend_check(1);
3467 	MPASS(ret == 0 || ret == EINTR || ret == ERESTART);
3468 	return (ret);
3469 }
3470 
3471 int
3472 sig_ast_needsigchk(struct thread *td)
3473 {
3474 	struct proc *p;
3475 	struct sigacts *ps;
3476 	int ret, sig;
3477 
3478 	p = td->td_proc;
3479 	PROC_LOCK_ASSERT(p, MA_OWNED);
3480 
3481 	if (!td_ast_pending(td, TDA_SIG))
3482 		return (0);
3483 
3484 	ps = p->p_sigacts;
3485 	mtx_lock(&ps->ps_mtx);
3486 	sig = cursig(td);
3487 	if (sig == -1) {
3488 		mtx_unlock(&ps->ps_mtx);
3489 		KASSERT((td->td_flags & TDF_SBDRY) != 0, ("lost TDF_SBDRY"));
3490 		KASSERT(TD_SBDRY_INTR(td),
3491 		    ("lost TDF_SERESTART of TDF_SEINTR"));
3492 		KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
3493 		    (TDF_SEINTR | TDF_SERESTART),
3494 		    ("both TDF_SEINTR and TDF_SERESTART"));
3495 		ret = TD_SBDRY_ERRNO(td);
3496 	} else if (sig != 0) {
3497 		ret = SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR : ERESTART;
3498 		mtx_unlock(&ps->ps_mtx);
3499 	} else {
3500 		mtx_unlock(&ps->ps_mtx);
3501 		ret = 0;
3502 	}
3503 
3504 	/*
3505 	 * Do not go into sleep if this thread was the ptrace(2)
3506 	 * attach leader.  cursig() consumed SIGSTOP from PT_ATTACH,
3507 	 * but we usually act on the signal by interrupting sleep, and
3508 	 * should do that here as well.
3509 	 */
3510 	if ((td->td_dbgflags & TDB_FSTP) != 0) {
3511 		if (ret == 0)
3512 			ret = EINTR;
3513 		td->td_dbgflags &= ~TDB_FSTP;
3514 	}
3515 
3516 	return (ret);
3517 }
3518 
3519 int
3520 sig_intr(void)
3521 {
3522 	struct thread *td;
3523 	struct proc *p;
3524 	int ret;
3525 
3526 	td = curthread;
3527 	if (!td_ast_pending(td, TDA_SIG) && !td_ast_pending(td, TDA_SUSPEND))
3528 		return (0);
3529 
3530 	p = td->td_proc;
3531 
3532 	PROC_LOCK(p);
3533 	ret = sig_ast_checksusp(td);
3534 	if (ret == 0)
3535 		ret = sig_ast_needsigchk(td);
3536 	PROC_UNLOCK(p);
3537 	return (ret);
3538 }
3539 
3540 bool
3541 curproc_sigkilled(void)
3542 {
3543 	struct thread *td;
3544 	struct proc *p;
3545 	struct sigacts *ps;
3546 	bool res;
3547 
3548 	td = curthread;
3549 	if (!td_ast_pending(td, TDA_SIG))
3550 		return (false);
3551 
3552 	p = td->td_proc;
3553 	PROC_LOCK(p);
3554 	ps = p->p_sigacts;
3555 	mtx_lock(&ps->ps_mtx);
3556 	res = SIGISMEMBER(td->td_sigqueue.sq_signals, SIGKILL) ||
3557 	    SIGISMEMBER(p->p_sigqueue.sq_signals, SIGKILL);
3558 	mtx_unlock(&ps->ps_mtx);
3559 	PROC_UNLOCK(p);
3560 	return (res);
3561 }
3562 
3563 void
3564 proc_wkilled(struct proc *p)
3565 {
3566 
3567 	PROC_LOCK_ASSERT(p, MA_OWNED);
3568 	if ((p->p_flag & P_WKILLED) == 0) {
3569 		p->p_flag |= P_WKILLED;
3570 		/*
3571 		 * Notify swapper that there is a process to swap in.
3572 		 * The notification is racy, at worst it would take 10
3573 		 * seconds for the swapper process to notice.
3574 		 */
3575 		if ((p->p_flag & (P_INMEM | P_SWAPPINGIN)) == 0)
3576 			wakeup(&proc0);
3577 	}
3578 }
3579 
3580 /*
3581  * Kill the current process for stated reason.
3582  */
3583 void
3584 killproc(struct proc *p, const char *why)
3585 {
3586 
3587 	PROC_LOCK_ASSERT(p, MA_OWNED);
3588 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3589 	    p->p_comm);
3590 	log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3591 	    p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3592 	    p->p_ucred->cr_uid, why);
3593 	proc_wkilled(p);
3594 	kern_psignal(p, SIGKILL);
3595 }
3596 
3597 /*
3598  * Force the current process to exit with the specified signal, dumping core
3599  * if appropriate.  We bypass the normal tests for masked and caught signals,
3600  * allowing unrecoverable failures to terminate the process without changing
3601  * signal state.  Mark the accounting record with the signal termination.
3602  * If dumping core, save the signal number for the debugger.  Calls exit and
3603  * does not return.
3604  */
3605 void
3606 sigexit(struct thread *td, int sig)
3607 {
3608 	struct proc *p = td->td_proc;
3609 
3610 	PROC_LOCK_ASSERT(p, MA_OWNED);
3611 	proc_set_p2_wexit(p);
3612 
3613 	p->p_acflag |= AXSIG;
3614 	/*
3615 	 * We must be single-threading to generate a core dump.  This
3616 	 * ensures that the registers in the core file are up-to-date.
3617 	 * Also, the ELF dump handler assumes that the thread list doesn't
3618 	 * change out from under it.
3619 	 *
3620 	 * XXX If another thread attempts to single-thread before us
3621 	 *     (e.g. via fork()), we won't get a dump at all.
3622 	 */
3623 	if ((sigprop(sig) & SIGPROP_CORE) &&
3624 	    thread_single(p, SINGLE_NO_EXIT) == 0) {
3625 		p->p_sig = sig;
3626 		/*
3627 		 * Log signals which would cause core dumps
3628 		 * (Log as LOG_INFO to appease those who don't want
3629 		 * these messages.)
3630 		 * XXX : Todo, as well as euid, write out ruid too
3631 		 * Note that coredump() drops proc lock.
3632 		 */
3633 		if (coredump(td) == 0)
3634 			sig |= WCOREFLAG;
3635 		if (kern_logsigexit)
3636 			log(LOG_INFO,
3637 			    "pid %d (%s), jid %d, uid %d: exited on "
3638 			    "signal %d%s\n", p->p_pid, p->p_comm,
3639 			    p->p_ucred->cr_prison->pr_id,
3640 			    td->td_ucred->cr_uid,
3641 			    sig &~ WCOREFLAG,
3642 			    sig & WCOREFLAG ? " (core dumped)" : "");
3643 	} else
3644 		PROC_UNLOCK(p);
3645 	exit1(td, 0, sig);
3646 	/* NOTREACHED */
3647 }
3648 
3649 /*
3650  * Send queued SIGCHLD to parent when child process's state
3651  * is changed.
3652  */
3653 static void
3654 sigparent(struct proc *p, int reason, int status)
3655 {
3656 	PROC_LOCK_ASSERT(p, MA_OWNED);
3657 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3658 
3659 	if (p->p_ksi != NULL) {
3660 		p->p_ksi->ksi_signo  = SIGCHLD;
3661 		p->p_ksi->ksi_code   = reason;
3662 		p->p_ksi->ksi_status = status;
3663 		p->p_ksi->ksi_pid    = p->p_pid;
3664 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3665 		if (KSI_ONQ(p->p_ksi))
3666 			return;
3667 	}
3668 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3669 }
3670 
3671 static void
3672 childproc_jobstate(struct proc *p, int reason, int sig)
3673 {
3674 	struct sigacts *ps;
3675 
3676 	PROC_LOCK_ASSERT(p, MA_OWNED);
3677 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3678 
3679 	/*
3680 	 * Wake up parent sleeping in kern_wait(), also send
3681 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3682 	 * that parent will awake, because parent may masked
3683 	 * the signal.
3684 	 */
3685 	p->p_pptr->p_flag |= P_STATCHILD;
3686 	wakeup(p->p_pptr);
3687 
3688 	ps = p->p_pptr->p_sigacts;
3689 	mtx_lock(&ps->ps_mtx);
3690 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3691 		mtx_unlock(&ps->ps_mtx);
3692 		sigparent(p, reason, sig);
3693 	} else
3694 		mtx_unlock(&ps->ps_mtx);
3695 }
3696 
3697 void
3698 childproc_stopped(struct proc *p, int reason)
3699 {
3700 
3701 	childproc_jobstate(p, reason, p->p_xsig);
3702 }
3703 
3704 void
3705 childproc_continued(struct proc *p)
3706 {
3707 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3708 }
3709 
3710 void
3711 childproc_exited(struct proc *p)
3712 {
3713 	int reason, status;
3714 
3715 	if (WCOREDUMP(p->p_xsig)) {
3716 		reason = CLD_DUMPED;
3717 		status = WTERMSIG(p->p_xsig);
3718 	} else if (WIFSIGNALED(p->p_xsig)) {
3719 		reason = CLD_KILLED;
3720 		status = WTERMSIG(p->p_xsig);
3721 	} else {
3722 		reason = CLD_EXITED;
3723 		status = p->p_xexit;
3724 	}
3725 	/*
3726 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3727 	 * done in exit1().
3728 	 */
3729 	sigparent(p, reason, status);
3730 }
3731 
3732 #define	MAX_NUM_CORE_FILES 100000
3733 #ifndef NUM_CORE_FILES
3734 #define	NUM_CORE_FILES 5
3735 #endif
3736 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3737 static int num_cores = NUM_CORE_FILES;
3738 
3739 static int
3740 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3741 {
3742 	int error;
3743 	int new_val;
3744 
3745 	new_val = num_cores;
3746 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3747 	if (error != 0 || req->newptr == NULL)
3748 		return (error);
3749 	if (new_val > MAX_NUM_CORE_FILES)
3750 		new_val = MAX_NUM_CORE_FILES;
3751 	if (new_val < 0)
3752 		new_val = 0;
3753 	num_cores = new_val;
3754 	return (0);
3755 }
3756 SYSCTL_PROC(_debug, OID_AUTO, ncores,
3757     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
3758     sysctl_debug_num_cores_check, "I",
3759     "Maximum number of generated process corefiles while using index format");
3760 
3761 #define	GZIP_SUFFIX	".gz"
3762 #define	ZSTD_SUFFIX	".zst"
3763 
3764 int compress_user_cores = 0;
3765 
3766 static int
3767 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3768 {
3769 	int error, val;
3770 
3771 	val = compress_user_cores;
3772 	error = sysctl_handle_int(oidp, &val, 0, req);
3773 	if (error != 0 || req->newptr == NULL)
3774 		return (error);
3775 	if (val != 0 && !compressor_avail(val))
3776 		return (EINVAL);
3777 	compress_user_cores = val;
3778 	return (error);
3779 }
3780 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores,
3781     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, 0, sizeof(int),
3782     sysctl_compress_user_cores, "I",
3783     "Enable compression of user corefiles ("
3784     __XSTRING(COMPRESS_GZIP) " = gzip, "
3785     __XSTRING(COMPRESS_ZSTD) " = zstd)");
3786 
3787 int compress_user_cores_level = 6;
3788 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3789     &compress_user_cores_level, 0,
3790     "Corefile compression level");
3791 
3792 /*
3793  * Protect the access to corefilename[] by allproc_lock.
3794  */
3795 #define	corefilename_lock	allproc_lock
3796 
3797 static char corefilename[MAXPATHLEN] = {"%N.core"};
3798 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3799 
3800 static int
3801 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3802 {
3803 	int error;
3804 
3805 	sx_xlock(&corefilename_lock);
3806 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3807 	    req);
3808 	sx_xunlock(&corefilename_lock);
3809 
3810 	return (error);
3811 }
3812 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3813     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3814     "Process corefile name format string");
3815 
3816 static void
3817 vnode_close_locked(struct thread *td, struct vnode *vp)
3818 {
3819 
3820 	VOP_UNLOCK(vp);
3821 	vn_close(vp, FWRITE, td->td_ucred, td);
3822 }
3823 
3824 /*
3825  * If the core format has a %I in it, then we need to check
3826  * for existing corefiles before defining a name.
3827  * To do this we iterate over 0..ncores to find a
3828  * non-existing core file name to use. If all core files are
3829  * already used we choose the oldest one.
3830  */
3831 static int
3832 corefile_open_last(struct thread *td, char *name, int indexpos,
3833     int indexlen, int ncores, struct vnode **vpp)
3834 {
3835 	struct vnode *oldvp, *nextvp, *vp;
3836 	struct vattr vattr;
3837 	struct nameidata nd;
3838 	int error, i, flags, oflags, cmode;
3839 	char ch;
3840 	struct timespec lasttime;
3841 
3842 	nextvp = oldvp = NULL;
3843 	cmode = S_IRUSR | S_IWUSR;
3844 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3845 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3846 
3847 	for (i = 0; i < ncores; i++) {
3848 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3849 
3850 		ch = name[indexpos + indexlen];
3851 		(void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3852 		    i);
3853 		name[indexpos + indexlen] = ch;
3854 
3855 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
3856 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3857 		    NULL);
3858 		if (error != 0)
3859 			break;
3860 
3861 		vp = nd.ni_vp;
3862 		NDFREE_PNBUF(&nd);
3863 		if ((flags & O_CREAT) == O_CREAT) {
3864 			nextvp = vp;
3865 			break;
3866 		}
3867 
3868 		error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3869 		if (error != 0) {
3870 			vnode_close_locked(td, vp);
3871 			break;
3872 		}
3873 
3874 		if (oldvp == NULL ||
3875 		    lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3876 		    (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3877 		    lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3878 			if (oldvp != NULL)
3879 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3880 			oldvp = vp;
3881 			VOP_UNLOCK(oldvp);
3882 			lasttime = vattr.va_mtime;
3883 		} else {
3884 			vnode_close_locked(td, vp);
3885 		}
3886 	}
3887 
3888 	if (oldvp != NULL) {
3889 		if (nextvp == NULL) {
3890 			if ((td->td_proc->p_flag & P_SUGID) != 0) {
3891 				error = EFAULT;
3892 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3893 			} else {
3894 				nextvp = oldvp;
3895 				error = vn_lock(nextvp, LK_EXCLUSIVE);
3896 				if (error != 0) {
3897 					vn_close(nextvp, FWRITE, td->td_ucred,
3898 					    td);
3899 					nextvp = NULL;
3900 				}
3901 			}
3902 		} else {
3903 			vn_close(oldvp, FWRITE, td->td_ucred, td);
3904 		}
3905 	}
3906 	if (error != 0) {
3907 		if (nextvp != NULL)
3908 			vnode_close_locked(td, oldvp);
3909 	} else {
3910 		*vpp = nextvp;
3911 	}
3912 
3913 	return (error);
3914 }
3915 
3916 /*
3917  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3918  * Expand the name described in corefilename, using name, uid, and pid
3919  * and open/create core file.
3920  * corefilename is a printf-like string, with three format specifiers:
3921  *	%N	name of process ("name")
3922  *	%P	process id (pid)
3923  *	%U	user id (uid)
3924  * For example, "%N.core" is the default; they can be disabled completely
3925  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3926  * This is controlled by the sysctl variable kern.corefile (see above).
3927  */
3928 static int
3929 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3930     int compress, int signum, struct vnode **vpp, char **namep)
3931 {
3932 	struct sbuf sb;
3933 	struct nameidata nd;
3934 	const char *format;
3935 	char *hostname, *name;
3936 	int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
3937 
3938 	hostname = NULL;
3939 	format = corefilename;
3940 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3941 	indexlen = 0;
3942 	indexpos = -1;
3943 	ncores = num_cores;
3944 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3945 	sx_slock(&corefilename_lock);
3946 	for (i = 0; format[i] != '\0'; i++) {
3947 		switch (format[i]) {
3948 		case '%':	/* Format character */
3949 			i++;
3950 			switch (format[i]) {
3951 			case '%':
3952 				sbuf_putc(&sb, '%');
3953 				break;
3954 			case 'H':	/* hostname */
3955 				if (hostname == NULL) {
3956 					hostname = malloc(MAXHOSTNAMELEN,
3957 					    M_TEMP, M_WAITOK);
3958 				}
3959 				getcredhostname(td->td_ucred, hostname,
3960 				    MAXHOSTNAMELEN);
3961 				sbuf_printf(&sb, "%s", hostname);
3962 				break;
3963 			case 'I':	/* autoincrementing index */
3964 				if (indexpos != -1) {
3965 					sbuf_printf(&sb, "%%I");
3966 					break;
3967 				}
3968 
3969 				indexpos = sbuf_len(&sb);
3970 				sbuf_printf(&sb, "%u", ncores - 1);
3971 				indexlen = sbuf_len(&sb) - indexpos;
3972 				break;
3973 			case 'N':	/* process name */
3974 				sbuf_printf(&sb, "%s", comm);
3975 				break;
3976 			case 'P':	/* process id */
3977 				sbuf_printf(&sb, "%u", pid);
3978 				break;
3979 			case 'S':	/* signal number */
3980 				sbuf_printf(&sb, "%i", signum);
3981 				break;
3982 			case 'U':	/* user id */
3983 				sbuf_printf(&sb, "%u", uid);
3984 				break;
3985 			default:
3986 				log(LOG_ERR,
3987 				    "Unknown format character %c in "
3988 				    "corename `%s'\n", format[i], format);
3989 				break;
3990 			}
3991 			break;
3992 		default:
3993 			sbuf_putc(&sb, format[i]);
3994 			break;
3995 		}
3996 	}
3997 	sx_sunlock(&corefilename_lock);
3998 	free(hostname, M_TEMP);
3999 	if (compress == COMPRESS_GZIP)
4000 		sbuf_printf(&sb, GZIP_SUFFIX);
4001 	else if (compress == COMPRESS_ZSTD)
4002 		sbuf_printf(&sb, ZSTD_SUFFIX);
4003 	if (sbuf_error(&sb) != 0) {
4004 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
4005 		    "long\n", (long)pid, comm, (u_long)uid);
4006 		sbuf_delete(&sb);
4007 		free(name, M_TEMP);
4008 		return (ENOMEM);
4009 	}
4010 	sbuf_finish(&sb);
4011 	sbuf_delete(&sb);
4012 
4013 	if (indexpos != -1) {
4014 		error = corefile_open_last(td, name, indexpos, indexlen, ncores,
4015 		    vpp);
4016 		if (error != 0) {
4017 			log(LOG_ERR,
4018 			    "pid %d (%s), uid (%u):  Path `%s' failed "
4019 			    "on initial open test, error = %d\n",
4020 			    pid, comm, uid, name, error);
4021 		}
4022 	} else {
4023 		cmode = S_IRUSR | S_IWUSR;
4024 		oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
4025 		    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
4026 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
4027 		if ((td->td_proc->p_flag & P_SUGID) != 0)
4028 			flags |= O_EXCL;
4029 
4030 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
4031 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
4032 		    NULL);
4033 		if (error == 0) {
4034 			*vpp = nd.ni_vp;
4035 			NDFREE_PNBUF(&nd);
4036 		}
4037 	}
4038 
4039 	if (error != 0) {
4040 #ifdef AUDIT
4041 		audit_proc_coredump(td, name, error);
4042 #endif
4043 		free(name, M_TEMP);
4044 		return (error);
4045 	}
4046 	*namep = name;
4047 	return (0);
4048 }
4049 
4050 /*
4051  * Dump a process' core.  The main routine does some
4052  * policy checking, and creates the name of the coredump;
4053  * then it passes on a vnode and a size limit to the process-specific
4054  * coredump routine if there is one; if there _is not_ one, it returns
4055  * ENOSYS; otherwise it returns the error from the process-specific routine.
4056  */
4057 
4058 static int
4059 coredump(struct thread *td)
4060 {
4061 	struct proc *p = td->td_proc;
4062 	struct ucred *cred = td->td_ucred;
4063 	struct vnode *vp;
4064 	struct flock lf;
4065 	struct vattr vattr;
4066 	size_t fullpathsize;
4067 	int error, error1, locked;
4068 	char *name;			/* name of corefile */
4069 	void *rl_cookie;
4070 	off_t limit;
4071 	char *fullpath, *freepath = NULL;
4072 	struct sbuf *sb;
4073 
4074 	PROC_LOCK_ASSERT(p, MA_OWNED);
4075 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
4076 
4077 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
4078 	    (p->p_flag2 & P2_NOTRACE) != 0) {
4079 		PROC_UNLOCK(p);
4080 		return (EFAULT);
4081 	}
4082 
4083 	/*
4084 	 * Note that the bulk of limit checking is done after
4085 	 * the corefile is created.  The exception is if the limit
4086 	 * for corefiles is 0, in which case we don't bother
4087 	 * creating the corefile at all.  This layout means that
4088 	 * a corefile is truncated instead of not being created,
4089 	 * if it is larger than the limit.
4090 	 */
4091 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
4092 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
4093 		PROC_UNLOCK(p);
4094 		return (EFBIG);
4095 	}
4096 	PROC_UNLOCK(p);
4097 
4098 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
4099 	    compress_user_cores, p->p_sig, &vp, &name);
4100 	if (error != 0)
4101 		return (error);
4102 
4103 	/*
4104 	 * Don't dump to non-regular files or files with links.
4105 	 * Do not dump into system files. Effective user must own the corefile.
4106 	 */
4107 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
4108 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
4109 	    vattr.va_uid != cred->cr_uid) {
4110 		VOP_UNLOCK(vp);
4111 		error = EFAULT;
4112 		goto out;
4113 	}
4114 
4115 	VOP_UNLOCK(vp);
4116 
4117 	/* Postpone other writers, including core dumps of other processes. */
4118 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
4119 
4120 	lf.l_whence = SEEK_SET;
4121 	lf.l_start = 0;
4122 	lf.l_len = 0;
4123 	lf.l_type = F_WRLCK;
4124 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
4125 
4126 	VATTR_NULL(&vattr);
4127 	vattr.va_size = 0;
4128 	if (set_core_nodump_flag)
4129 		vattr.va_flags = UF_NODUMP;
4130 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4131 	VOP_SETATTR(vp, &vattr, cred);
4132 	VOP_UNLOCK(vp);
4133 	PROC_LOCK(p);
4134 	p->p_acflag |= ACORE;
4135 	PROC_UNLOCK(p);
4136 
4137 	if (p->p_sysent->sv_coredump != NULL) {
4138 		error = p->p_sysent->sv_coredump(td, vp, limit, 0);
4139 	} else {
4140 		error = ENOSYS;
4141 	}
4142 
4143 	if (locked) {
4144 		lf.l_type = F_UNLCK;
4145 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
4146 	}
4147 	vn_rangelock_unlock(vp, rl_cookie);
4148 
4149 	/*
4150 	 * Notify the userland helper that a process triggered a core dump.
4151 	 * This allows the helper to run an automated debugging session.
4152 	 */
4153 	if (error != 0 || coredump_devctl == 0)
4154 		goto out;
4155 	sb = sbuf_new_auto();
4156 	if (vn_fullpath_global(p->p_textvp, &fullpath, &freepath) != 0)
4157 		goto out2;
4158 	sbuf_printf(sb, "comm=\"");
4159 	devctl_safe_quote_sb(sb, fullpath);
4160 	free(freepath, M_TEMP);
4161 	sbuf_printf(sb, "\" core=\"");
4162 
4163 	/*
4164 	 * We can't lookup core file vp directly. When we're replacing a core, and
4165 	 * other random times, we flush the name cache, so it will fail. Instead,
4166 	 * if the path of the core is relative, add the current dir in front if it.
4167 	 */
4168 	if (name[0] != '/') {
4169 		fullpathsize = MAXPATHLEN;
4170 		freepath = malloc(fullpathsize, M_TEMP, M_WAITOK);
4171 		if (vn_getcwd(freepath, &fullpath, &fullpathsize) != 0) {
4172 			free(freepath, M_TEMP);
4173 			goto out2;
4174 		}
4175 		devctl_safe_quote_sb(sb, fullpath);
4176 		free(freepath, M_TEMP);
4177 		sbuf_putc(sb, '/');
4178 	}
4179 	devctl_safe_quote_sb(sb, name);
4180 	sbuf_printf(sb, "\"");
4181 	if (sbuf_finish(sb) == 0)
4182 		devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
4183 out2:
4184 	sbuf_delete(sb);
4185 out:
4186 	error1 = vn_close(vp, FWRITE, cred, td);
4187 	if (error == 0)
4188 		error = error1;
4189 #ifdef AUDIT
4190 	audit_proc_coredump(td, name, error);
4191 #endif
4192 	free(name, M_TEMP);
4193 	return (error);
4194 }
4195 
4196 /*
4197  * Nonexistent system call-- signal process (may want to handle it).  Flag
4198  * error in case process won't see signal immediately (blocked or ignored).
4199  */
4200 #ifndef _SYS_SYSPROTO_H_
4201 struct nosys_args {
4202 	int	dummy;
4203 };
4204 #endif
4205 /* ARGSUSED */
4206 int
4207 nosys(struct thread *td, struct nosys_args *args)
4208 {
4209 	struct proc *p;
4210 
4211 	p = td->td_proc;
4212 
4213 	PROC_LOCK(p);
4214 	tdsignal(td, SIGSYS);
4215 	PROC_UNLOCK(p);
4216 	if (kern_lognosys == 1 || kern_lognosys == 3) {
4217 		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4218 		    td->td_sa.code);
4219 	}
4220 	if (kern_lognosys == 2 || kern_lognosys == 3 ||
4221 	    (p->p_pid == 1 && (kern_lognosys & 3) == 0)) {
4222 		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4223 		    td->td_sa.code);
4224 	}
4225 	return (ENOSYS);
4226 }
4227 
4228 /*
4229  * Send a SIGIO or SIGURG signal to a process or process group using stored
4230  * credentials rather than those of the current process.
4231  */
4232 void
4233 pgsigio(struct sigio **sigiop, int sig, int checkctty)
4234 {
4235 	ksiginfo_t ksi;
4236 	struct sigio *sigio;
4237 
4238 	ksiginfo_init(&ksi);
4239 	ksi.ksi_signo = sig;
4240 	ksi.ksi_code = SI_KERNEL;
4241 
4242 	SIGIO_LOCK();
4243 	sigio = *sigiop;
4244 	if (sigio == NULL) {
4245 		SIGIO_UNLOCK();
4246 		return;
4247 	}
4248 	if (sigio->sio_pgid > 0) {
4249 		PROC_LOCK(sigio->sio_proc);
4250 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
4251 			kern_psignal(sigio->sio_proc, sig);
4252 		PROC_UNLOCK(sigio->sio_proc);
4253 	} else if (sigio->sio_pgid < 0) {
4254 		struct proc *p;
4255 
4256 		PGRP_LOCK(sigio->sio_pgrp);
4257 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
4258 			PROC_LOCK(p);
4259 			if (p->p_state == PRS_NORMAL &&
4260 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
4261 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
4262 				kern_psignal(p, sig);
4263 			PROC_UNLOCK(p);
4264 		}
4265 		PGRP_UNLOCK(sigio->sio_pgrp);
4266 	}
4267 	SIGIO_UNLOCK();
4268 }
4269 
4270 static int
4271 filt_sigattach(struct knote *kn)
4272 {
4273 	struct proc *p = curproc;
4274 
4275 	kn->kn_ptr.p_proc = p;
4276 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
4277 
4278 	knlist_add(p->p_klist, kn, 0);
4279 
4280 	return (0);
4281 }
4282 
4283 static void
4284 filt_sigdetach(struct knote *kn)
4285 {
4286 	struct proc *p = kn->kn_ptr.p_proc;
4287 
4288 	knlist_remove(p->p_klist, kn, 0);
4289 }
4290 
4291 /*
4292  * signal knotes are shared with proc knotes, so we apply a mask to
4293  * the hint in order to differentiate them from process hints.  This
4294  * could be avoided by using a signal-specific knote list, but probably
4295  * isn't worth the trouble.
4296  */
4297 static int
4298 filt_signal(struct knote *kn, long hint)
4299 {
4300 
4301 	if (hint & NOTE_SIGNAL) {
4302 		hint &= ~NOTE_SIGNAL;
4303 
4304 		if (kn->kn_id == hint)
4305 			kn->kn_data++;
4306 	}
4307 	return (kn->kn_data != 0);
4308 }
4309 
4310 struct sigacts *
4311 sigacts_alloc(void)
4312 {
4313 	struct sigacts *ps;
4314 
4315 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
4316 	refcount_init(&ps->ps_refcnt, 1);
4317 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
4318 	return (ps);
4319 }
4320 
4321 void
4322 sigacts_free(struct sigacts *ps)
4323 {
4324 
4325 	if (refcount_release(&ps->ps_refcnt) == 0)
4326 		return;
4327 	mtx_destroy(&ps->ps_mtx);
4328 	free(ps, M_SUBPROC);
4329 }
4330 
4331 struct sigacts *
4332 sigacts_hold(struct sigacts *ps)
4333 {
4334 
4335 	refcount_acquire(&ps->ps_refcnt);
4336 	return (ps);
4337 }
4338 
4339 void
4340 sigacts_copy(struct sigacts *dest, struct sigacts *src)
4341 {
4342 
4343 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
4344 	mtx_lock(&src->ps_mtx);
4345 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
4346 	mtx_unlock(&src->ps_mtx);
4347 }
4348 
4349 int
4350 sigacts_shared(struct sigacts *ps)
4351 {
4352 
4353 	return (ps->ps_refcnt > 1);
4354 }
4355 
4356 void
4357 sig_drop_caught(struct proc *p)
4358 {
4359 	int sig;
4360 	struct sigacts *ps;
4361 
4362 	ps = p->p_sigacts;
4363 	PROC_LOCK_ASSERT(p, MA_OWNED);
4364 	mtx_assert(&ps->ps_mtx, MA_OWNED);
4365 	SIG_FOREACH(sig, &ps->ps_sigcatch) {
4366 		sigdflt(ps, sig);
4367 		if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
4368 			sigqueue_delete_proc(p, sig);
4369 	}
4370 }
4371 
4372 static void
4373 sigfastblock_failed(struct thread *td, bool sendsig, bool write)
4374 {
4375 	ksiginfo_t ksi;
4376 
4377 	/*
4378 	 * Prevent further fetches and SIGSEGVs, allowing thread to
4379 	 * issue syscalls despite corruption.
4380 	 */
4381 	sigfastblock_clear(td);
4382 
4383 	if (!sendsig)
4384 		return;
4385 	ksiginfo_init_trap(&ksi);
4386 	ksi.ksi_signo = SIGSEGV;
4387 	ksi.ksi_code = write ? SEGV_ACCERR : SEGV_MAPERR;
4388 	ksi.ksi_addr = td->td_sigblock_ptr;
4389 	trapsignal(td, &ksi);
4390 }
4391 
4392 static bool
4393 sigfastblock_fetch_sig(struct thread *td, bool sendsig, uint32_t *valp)
4394 {
4395 	uint32_t res;
4396 
4397 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4398 		return (true);
4399 	if (fueword32((void *)td->td_sigblock_ptr, &res) == -1) {
4400 		sigfastblock_failed(td, sendsig, false);
4401 		return (false);
4402 	}
4403 	*valp = res;
4404 	td->td_sigblock_val = res & ~SIGFASTBLOCK_FLAGS;
4405 	return (true);
4406 }
4407 
4408 static void
4409 sigfastblock_resched(struct thread *td, bool resched)
4410 {
4411 	struct proc *p;
4412 
4413 	if (resched) {
4414 		p = td->td_proc;
4415 		PROC_LOCK(p);
4416 		reschedule_signals(p, td->td_sigmask, 0);
4417 		PROC_UNLOCK(p);
4418 	}
4419 	ast_sched(td, TDA_SIG);
4420 }
4421 
4422 int
4423 sys_sigfastblock(struct thread *td, struct sigfastblock_args *uap)
4424 {
4425 	struct proc *p;
4426 	int error, res;
4427 	uint32_t oldval;
4428 
4429 	error = 0;
4430 	p = td->td_proc;
4431 	switch (uap->cmd) {
4432 	case SIGFASTBLOCK_SETPTR:
4433 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
4434 			error = EBUSY;
4435 			break;
4436 		}
4437 		if (((uintptr_t)(uap->ptr) & (sizeof(uint32_t) - 1)) != 0) {
4438 			error = EINVAL;
4439 			break;
4440 		}
4441 		td->td_pflags |= TDP_SIGFASTBLOCK;
4442 		td->td_sigblock_ptr = uap->ptr;
4443 		break;
4444 
4445 	case SIGFASTBLOCK_UNBLOCK:
4446 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4447 			error = EINVAL;
4448 			break;
4449 		}
4450 
4451 		for (;;) {
4452 			res = casueword32(td->td_sigblock_ptr,
4453 			    SIGFASTBLOCK_PEND, &oldval, 0);
4454 			if (res == -1) {
4455 				error = EFAULT;
4456 				sigfastblock_failed(td, false, true);
4457 				break;
4458 			}
4459 			if (res == 0)
4460 				break;
4461 			MPASS(res == 1);
4462 			if (oldval != SIGFASTBLOCK_PEND) {
4463 				error = EBUSY;
4464 				break;
4465 			}
4466 			error = thread_check_susp(td, false);
4467 			if (error != 0)
4468 				break;
4469 		}
4470 		if (error != 0)
4471 			break;
4472 
4473 		/*
4474 		 * td_sigblock_val is cleared there, but not on a
4475 		 * syscall exit.  The end effect is that a single
4476 		 * interruptible sleep, while user sigblock word is
4477 		 * set, might return EINTR or ERESTART to usermode
4478 		 * without delivering signal.  All further sleeps,
4479 		 * until userspace clears the word and does
4480 		 * sigfastblock(UNBLOCK), observe current word and no
4481 		 * longer get interrupted.  It is slight
4482 		 * non-conformance, with alternative to have read the
4483 		 * sigblock word on each syscall entry.
4484 		 */
4485 		td->td_sigblock_val = 0;
4486 
4487 		/*
4488 		 * Rely on normal ast mechanism to deliver pending
4489 		 * signals to current thread.  But notify others about
4490 		 * fake unblock.
4491 		 */
4492 		sigfastblock_resched(td, error == 0 && p->p_numthreads != 1);
4493 
4494 		break;
4495 
4496 	case SIGFASTBLOCK_UNSETPTR:
4497 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4498 			error = EINVAL;
4499 			break;
4500 		}
4501 		if (!sigfastblock_fetch_sig(td, false, &oldval)) {
4502 			error = EFAULT;
4503 			break;
4504 		}
4505 		if (oldval != 0 && oldval != SIGFASTBLOCK_PEND) {
4506 			error = EBUSY;
4507 			break;
4508 		}
4509 		sigfastblock_clear(td);
4510 		break;
4511 
4512 	default:
4513 		error = EINVAL;
4514 		break;
4515 	}
4516 	return (error);
4517 }
4518 
4519 void
4520 sigfastblock_clear(struct thread *td)
4521 {
4522 	bool resched;
4523 
4524 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4525 		return;
4526 	td->td_sigblock_val = 0;
4527 	resched = (td->td_pflags & TDP_SIGFASTPENDING) != 0 ||
4528 	    SIGPENDING(td);
4529 	td->td_pflags &= ~(TDP_SIGFASTBLOCK | TDP_SIGFASTPENDING);
4530 	sigfastblock_resched(td, resched);
4531 }
4532 
4533 void
4534 sigfastblock_fetch(struct thread *td)
4535 {
4536 	uint32_t val;
4537 
4538 	(void)sigfastblock_fetch_sig(td, true, &val);
4539 }
4540 
4541 static void
4542 sigfastblock_setpend1(struct thread *td)
4543 {
4544 	int res;
4545 	uint32_t oldval;
4546 
4547 	if ((td->td_pflags & TDP_SIGFASTPENDING) == 0)
4548 		return;
4549 	res = fueword32((void *)td->td_sigblock_ptr, &oldval);
4550 	if (res == -1) {
4551 		sigfastblock_failed(td, true, false);
4552 		return;
4553 	}
4554 	for (;;) {
4555 		res = casueword32(td->td_sigblock_ptr, oldval, &oldval,
4556 		    oldval | SIGFASTBLOCK_PEND);
4557 		if (res == -1) {
4558 			sigfastblock_failed(td, true, true);
4559 			return;
4560 		}
4561 		if (res == 0) {
4562 			td->td_sigblock_val = oldval & ~SIGFASTBLOCK_FLAGS;
4563 			td->td_pflags &= ~TDP_SIGFASTPENDING;
4564 			break;
4565 		}
4566 		MPASS(res == 1);
4567 		if (thread_check_susp(td, false) != 0)
4568 			break;
4569 	}
4570 }
4571 
4572 static void
4573 sigfastblock_setpend(struct thread *td, bool resched)
4574 {
4575 	struct proc *p;
4576 
4577 	sigfastblock_setpend1(td);
4578 	if (resched) {
4579 		p = td->td_proc;
4580 		PROC_LOCK(p);
4581 		reschedule_signals(p, fastblock_mask, SIGPROCMASK_FASTBLK);
4582 		PROC_UNLOCK(p);
4583 	}
4584 }
4585