xref: /freebsd/sys/kern/kern_sig.c (revision 61ba55bcf70f2340f9c943c9571113b3fd8eda69)
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 #include "opt_capsicum.h"
41 #include "opt_ktrace.h"
42 
43 #include <sys/param.h>
44 #include <sys/capsicum.h>
45 #include <sys/ctype.h>
46 #include <sys/systm.h>
47 #include <sys/signalvar.h>
48 #include <sys/vnode.h>
49 #include <sys/acct.h>
50 #include <sys/capsicum.h>
51 #include <sys/compressor.h>
52 #include <sys/condvar.h>
53 #include <sys/devctl.h>
54 #include <sys/event.h>
55 #include <sys/fcntl.h>
56 #include <sys/imgact.h>
57 #include <sys/jail.h>
58 #include <sys/kernel.h>
59 #include <sys/ktr.h>
60 #include <sys/ktrace.h>
61 #include <sys/limits.h>
62 #include <sys/lock.h>
63 #include <sys/malloc.h>
64 #include <sys/mutex.h>
65 #include <sys/refcount.h>
66 #include <sys/namei.h>
67 #include <sys/proc.h>
68 #include <sys/procdesc.h>
69 #include <sys/ptrace.h>
70 #include <sys/posix4.h>
71 #include <sys/racct.h>
72 #include <sys/resourcevar.h>
73 #include <sys/sdt.h>
74 #include <sys/sbuf.h>
75 #include <sys/sleepqueue.h>
76 #include <sys/smp.h>
77 #include <sys/stat.h>
78 #include <sys/sx.h>
79 #include <sys/syscall.h>
80 #include <sys/syscallsubr.h>
81 #include <sys/sysctl.h>
82 #include <sys/sysent.h>
83 #include <sys/syslog.h>
84 #include <sys/sysproto.h>
85 #include <sys/timers.h>
86 #include <sys/unistd.h>
87 #include <sys/vmmeter.h>
88 #include <sys/wait.h>
89 #include <vm/vm.h>
90 #include <vm/vm_extern.h>
91 #include <vm/uma.h>
92 
93 #include <machine/cpu.h>
94 
95 #include <security/audit/audit.h>
96 
97 #define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
98 
99 SDT_PROVIDER_DECLARE(proc);
100 SDT_PROBE_DEFINE3(proc, , , signal__send,
101     "struct thread *", "struct proc *", "int");
102 SDT_PROBE_DEFINE2(proc, , , signal__clear,
103     "int", "ksiginfo_t *");
104 SDT_PROBE_DEFINE3(proc, , , signal__discard,
105     "struct thread *", "struct proc *", "int");
106 
107 static int	coredump(struct thread *);
108 static int	killpg1(struct thread *td, int sig, int pgid, int all,
109 		    ksiginfo_t *ksi);
110 static int	issignal(struct thread *td);
111 static void	reschedule_signals(struct proc *p, sigset_t block, int flags);
112 static int	sigprop(int sig);
113 static void	tdsigwakeup(struct thread *, int, sig_t, int);
114 static int	sig_suspend_threads(struct thread *, struct proc *);
115 static int	filt_sigattach(struct knote *kn);
116 static void	filt_sigdetach(struct knote *kn);
117 static int	filt_signal(struct knote *kn, long hint);
118 static struct thread *sigtd(struct proc *p, int sig, bool fast_sigblock);
119 static void	sigqueue_start(void);
120 static void	sigfastblock_setpend(struct thread *td, bool resched);
121 
122 static uma_zone_t	ksiginfo_zone = NULL;
123 struct filterops sig_filtops = {
124 	.f_isfd = 0,
125 	.f_attach = filt_sigattach,
126 	.f_detach = filt_sigdetach,
127 	.f_event = filt_signal,
128 };
129 
130 static int	kern_logsigexit = 1;
131 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
132     &kern_logsigexit, 0,
133     "Log processes quitting on abnormal signals to syslog(3)");
134 
135 static int	kern_forcesigexit = 1;
136 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
137     &kern_forcesigexit, 0, "Force trap signal to be handled");
138 
139 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
140     "POSIX real time signal");
141 
142 static int	max_pending_per_proc = 128;
143 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
144     &max_pending_per_proc, 0, "Max pending signals per proc");
145 
146 static int	preallocate_siginfo = 1024;
147 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
148     &preallocate_siginfo, 0, "Preallocated signal memory size");
149 
150 static int	signal_overflow = 0;
151 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
152     &signal_overflow, 0, "Number of signals overflew");
153 
154 static int	signal_alloc_fail = 0;
155 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
156     &signal_alloc_fail, 0, "signals failed to be allocated");
157 
158 static int	kern_lognosys = 0;
159 SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
160     "Log invalid syscalls");
161 
162 static int	kern_signosys = 1;
163 SYSCTL_INT(_kern, OID_AUTO, signosys, CTLFLAG_RWTUN, &kern_signosys, 0,
164     "Send SIGSYS on return from invalid syscall");
165 
166 __read_frequently bool sigfastblock_fetch_always = false;
167 SYSCTL_BOOL(_kern, OID_AUTO, sigfastblock_fetch_always, CTLFLAG_RWTUN,
168     &sigfastblock_fetch_always, 0,
169     "Fetch sigfastblock word on each syscall entry for proper "
170     "blocking semantic");
171 
172 static bool	kern_sig_discard_ign = true;
173 SYSCTL_BOOL(_kern, OID_AUTO, sig_discard_ign, CTLFLAG_RWTUN,
174     &kern_sig_discard_ign, 0,
175     "Discard ignored signals on delivery, otherwise queue them to "
176     "the target queue");
177 
178 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
179 
180 /*
181  * Policy -- Can ucred cr1 send SIGIO to process cr2?
182  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
183  * in the right situations.
184  */
185 #define CANSIGIO(cr1, cr2) \
186 	((cr1)->cr_uid == 0 || \
187 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
188 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
189 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
190 	    (cr1)->cr_uid == (cr2)->cr_uid)
191 
192 static int	sugid_coredump;
193 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
194     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
195 
196 static int	capmode_coredump;
197 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
198     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
199 
200 static int	do_coredump = 1;
201 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
202 	&do_coredump, 0, "Enable/Disable coredumps");
203 
204 static int	set_core_nodump_flag = 0;
205 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
206 	0, "Enable setting the NODUMP flag on coredump files");
207 
208 static int	coredump_devctl = 0;
209 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
210 	0, "Generate a devctl notification when processes coredump");
211 
212 /*
213  * Signal properties and actions.
214  * The array below categorizes the signals and their default actions
215  * according to the following properties:
216  */
217 #define	SIGPROP_KILL		0x01	/* terminates process by default */
218 #define	SIGPROP_CORE		0x02	/* ditto and coredumps */
219 #define	SIGPROP_STOP		0x04	/* suspend process */
220 #define	SIGPROP_TTYSTOP		0x08	/* ditto, from tty */
221 #define	SIGPROP_IGNORE		0x10	/* ignore by default */
222 #define	SIGPROP_CONT		0x20	/* continue if suspended */
223 
224 static const int sigproptbl[NSIG] = {
225 	[SIGHUP] =	SIGPROP_KILL,
226 	[SIGINT] =	SIGPROP_KILL,
227 	[SIGQUIT] =	SIGPROP_KILL | SIGPROP_CORE,
228 	[SIGILL] =	SIGPROP_KILL | SIGPROP_CORE,
229 	[SIGTRAP] =	SIGPROP_KILL | SIGPROP_CORE,
230 	[SIGABRT] =	SIGPROP_KILL | SIGPROP_CORE,
231 	[SIGEMT] =	SIGPROP_KILL | SIGPROP_CORE,
232 	[SIGFPE] =	SIGPROP_KILL | SIGPROP_CORE,
233 	[SIGKILL] =	SIGPROP_KILL,
234 	[SIGBUS] =	SIGPROP_KILL | SIGPROP_CORE,
235 	[SIGSEGV] =	SIGPROP_KILL | SIGPROP_CORE,
236 	[SIGSYS] =	SIGPROP_KILL | SIGPROP_CORE,
237 	[SIGPIPE] =	SIGPROP_KILL,
238 	[SIGALRM] =	SIGPROP_KILL,
239 	[SIGTERM] =	SIGPROP_KILL,
240 	[SIGURG] =	SIGPROP_IGNORE,
241 	[SIGSTOP] =	SIGPROP_STOP,
242 	[SIGTSTP] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
243 	[SIGCONT] =	SIGPROP_IGNORE | SIGPROP_CONT,
244 	[SIGCHLD] =	SIGPROP_IGNORE,
245 	[SIGTTIN] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
246 	[SIGTTOU] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
247 	[SIGIO] =	SIGPROP_IGNORE,
248 	[SIGXCPU] =	SIGPROP_KILL,
249 	[SIGXFSZ] =	SIGPROP_KILL,
250 	[SIGVTALRM] =	SIGPROP_KILL,
251 	[SIGPROF] =	SIGPROP_KILL,
252 	[SIGWINCH] =	SIGPROP_IGNORE,
253 	[SIGINFO] =	SIGPROP_IGNORE,
254 	[SIGUSR1] =	SIGPROP_KILL,
255 	[SIGUSR2] =	SIGPROP_KILL,
256 };
257 
258 #define	_SIG_FOREACH_ADVANCE(i, set) ({					\
259 	int __found;							\
260 	for (;;) {							\
261 		if (__bits != 0) {					\
262 			int __sig = ffs(__bits);			\
263 			__bits &= ~(1u << (__sig - 1));			\
264 			sig = __i * sizeof((set)->__bits[0]) * NBBY + __sig; \
265 			__found = 1;					\
266 			break;						\
267 		}							\
268 		if (++__i == _SIG_WORDS) {				\
269 			__found = 0;					\
270 			break;						\
271 		}							\
272 		__bits = (set)->__bits[__i];				\
273 	}								\
274 	__found != 0;							\
275 })
276 
277 #define	SIG_FOREACH(i, set)						\
278 	for (int32_t __i = -1, __bits = 0;				\
279 	    _SIG_FOREACH_ADVANCE(i, set); )				\
280 
281 static sigset_t fastblock_mask;
282 
283 static void
284 ast_sig(struct thread *td, int tda)
285 {
286 	struct proc *p;
287 	int old_boundary, sig;
288 	bool resched_sigs;
289 
290 	p = td->td_proc;
291 
292 #ifdef DIAGNOSTIC
293 	if (p->p_numthreads == 1 && (tda & (TDAI(TDA_SIG) |
294 	    TDAI(TDA_AST))) == 0) {
295 		PROC_LOCK(p);
296 		thread_lock(td);
297 		/*
298 		 * Note that TDA_SIG should be re-read from
299 		 * td_ast, since signal might have been delivered
300 		 * after we cleared td_flags above.  This is one of
301 		 * the reason for looping check for AST condition.
302 		 * See comment in userret() about P_PPWAIT.
303 		 */
304 		if ((p->p_flag & P_PPWAIT) == 0 &&
305 		    (td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
306 			if (SIGPENDING(td) && ((tda | td->td_ast) &
307 			    (TDAI(TDA_SIG) | TDAI(TDA_AST))) == 0) {
308 				thread_unlock(td); /* fix dumps */
309 				panic(
310 				    "failed2 to set signal flags for ast p %p "
311 				    "td %p tda %#x td_ast %#x fl %#x",
312 				    p, td, tda, td->td_ast, td->td_flags);
313 			}
314 		}
315 		thread_unlock(td);
316 		PROC_UNLOCK(p);
317 	}
318 #endif
319 
320 	/*
321 	 * Check for signals. Unlocked reads of p_pendingcnt or
322 	 * p_siglist might cause process-directed signal to be handled
323 	 * later.
324 	 */
325 	if ((tda & TDAI(TDA_SIG)) != 0 || p->p_pendingcnt > 0 ||
326 	    !SIGISEMPTY(p->p_siglist)) {
327 		sigfastblock_fetch(td);
328 		PROC_LOCK(p);
329 		old_boundary = ~TDB_BOUNDARY | (td->td_dbgflags & TDB_BOUNDARY);
330 		td->td_dbgflags |= TDB_BOUNDARY;
331 		mtx_lock(&p->p_sigacts->ps_mtx);
332 		while ((sig = cursig(td)) != 0) {
333 			KASSERT(sig >= 0, ("sig %d", sig));
334 			postsig(sig);
335 		}
336 		mtx_unlock(&p->p_sigacts->ps_mtx);
337 		td->td_dbgflags &= old_boundary;
338 		PROC_UNLOCK(p);
339 		resched_sigs = true;
340 	} else {
341 		resched_sigs = false;
342 	}
343 
344 	/*
345 	 * Handle deferred update of the fast sigblock value, after
346 	 * the postsig() loop was performed.
347 	 */
348 	sigfastblock_setpend(td, resched_sigs);
349 }
350 
351 static void
352 ast_sigsuspend(struct thread *td, int tda __unused)
353 {
354 	MPASS((td->td_pflags & TDP_OLDMASK) != 0);
355 	td->td_pflags &= ~TDP_OLDMASK;
356 	kern_sigprocmask(td, SIG_SETMASK, &td->td_oldsigmask, NULL, 0);
357 }
358 
359 static void
360 sigqueue_start(void)
361 {
362 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
363 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
364 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
365 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
366 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
367 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
368 	SIGFILLSET(fastblock_mask);
369 	SIG_CANTMASK(fastblock_mask);
370 	ast_register(TDA_SIG, ASTR_UNCOND, 0, ast_sig);
371 	ast_register(TDA_SIGSUSPEND, ASTR_ASTF_REQUIRED | ASTR_TDP,
372 	    TDP_OLDMASK, ast_sigsuspend);
373 }
374 
375 ksiginfo_t *
376 ksiginfo_alloc(int mwait)
377 {
378 	MPASS(mwait == M_WAITOK || mwait == M_NOWAIT);
379 
380 	if (ksiginfo_zone == NULL)
381 		return (NULL);
382 	return (uma_zalloc(ksiginfo_zone, mwait | M_ZERO));
383 }
384 
385 void
386 ksiginfo_free(ksiginfo_t *ksi)
387 {
388 	uma_zfree(ksiginfo_zone, ksi);
389 }
390 
391 static __inline bool
392 ksiginfo_tryfree(ksiginfo_t *ksi)
393 {
394 	if ((ksi->ksi_flags & KSI_EXT) == 0) {
395 		uma_zfree(ksiginfo_zone, ksi);
396 		return (true);
397 	}
398 	return (false);
399 }
400 
401 void
402 sigqueue_init(sigqueue_t *list, struct proc *p)
403 {
404 	SIGEMPTYSET(list->sq_signals);
405 	SIGEMPTYSET(list->sq_kill);
406 	SIGEMPTYSET(list->sq_ptrace);
407 	TAILQ_INIT(&list->sq_list);
408 	list->sq_proc = p;
409 	list->sq_flags = SQ_INIT;
410 }
411 
412 /*
413  * Get a signal's ksiginfo.
414  * Return:
415  *	0	-	signal not found
416  *	others	-	signal number
417  */
418 static int
419 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
420 {
421 	struct proc *p = sq->sq_proc;
422 	struct ksiginfo *ksi, *next;
423 	int count = 0;
424 
425 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
426 
427 	if (!SIGISMEMBER(sq->sq_signals, signo))
428 		return (0);
429 
430 	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
431 		count++;
432 		SIGDELSET(sq->sq_ptrace, signo);
433 		si->ksi_flags |= KSI_PTRACE;
434 	}
435 	if (SIGISMEMBER(sq->sq_kill, signo)) {
436 		count++;
437 		if (count == 1)
438 			SIGDELSET(sq->sq_kill, signo);
439 	}
440 
441 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
442 		if (ksi->ksi_signo == signo) {
443 			if (count == 0) {
444 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
445 				ksi->ksi_sigq = NULL;
446 				ksiginfo_copy(ksi, si);
447 				if (ksiginfo_tryfree(ksi) && p != NULL)
448 					p->p_pendingcnt--;
449 			}
450 			if (++count > 1)
451 				break;
452 		}
453 	}
454 
455 	if (count <= 1)
456 		SIGDELSET(sq->sq_signals, signo);
457 	si->ksi_signo = signo;
458 	return (signo);
459 }
460 
461 void
462 sigqueue_take(ksiginfo_t *ksi)
463 {
464 	struct ksiginfo *kp;
465 	struct proc	*p;
466 	sigqueue_t	*sq;
467 
468 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
469 		return;
470 
471 	p = sq->sq_proc;
472 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
473 	ksi->ksi_sigq = NULL;
474 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
475 		p->p_pendingcnt--;
476 
477 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
478 	     kp = TAILQ_NEXT(kp, ksi_link)) {
479 		if (kp->ksi_signo == ksi->ksi_signo)
480 			break;
481 	}
482 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
483 	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
484 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
485 }
486 
487 static int
488 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
489 {
490 	struct proc *p = sq->sq_proc;
491 	struct ksiginfo *ksi;
492 	int ret = 0;
493 
494 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
495 
496 	/*
497 	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
498 	 * for these signals.
499 	 */
500 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
501 		SIGADDSET(sq->sq_kill, signo);
502 		goto out_set_bit;
503 	}
504 
505 	/* directly insert the ksi, don't copy it */
506 	if (si->ksi_flags & KSI_INS) {
507 		if (si->ksi_flags & KSI_HEAD)
508 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
509 		else
510 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
511 		si->ksi_sigq = sq;
512 		goto out_set_bit;
513 	}
514 
515 	if (__predict_false(ksiginfo_zone == NULL)) {
516 		SIGADDSET(sq->sq_kill, signo);
517 		goto out_set_bit;
518 	}
519 
520 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
521 		signal_overflow++;
522 		ret = EAGAIN;
523 	} else if ((ksi = ksiginfo_alloc(M_NOWAIT)) == NULL) {
524 		signal_alloc_fail++;
525 		ret = EAGAIN;
526 	} else {
527 		if (p != NULL)
528 			p->p_pendingcnt++;
529 		ksiginfo_copy(si, ksi);
530 		ksi->ksi_signo = signo;
531 		if (si->ksi_flags & KSI_HEAD)
532 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
533 		else
534 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
535 		ksi->ksi_sigq = sq;
536 	}
537 
538 	if (ret != 0) {
539 		if ((si->ksi_flags & KSI_PTRACE) != 0) {
540 			SIGADDSET(sq->sq_ptrace, signo);
541 			ret = 0;
542 			goto out_set_bit;
543 		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
544 		    (si->ksi_flags & KSI_SIGQ) == 0) {
545 			SIGADDSET(sq->sq_kill, signo);
546 			ret = 0;
547 			goto out_set_bit;
548 		}
549 		return (ret);
550 	}
551 
552 out_set_bit:
553 	SIGADDSET(sq->sq_signals, signo);
554 	return (ret);
555 }
556 
557 void
558 sigqueue_flush(sigqueue_t *sq)
559 {
560 	struct proc *p = sq->sq_proc;
561 	ksiginfo_t *ksi;
562 
563 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
564 
565 	if (p != NULL)
566 		PROC_LOCK_ASSERT(p, MA_OWNED);
567 
568 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
569 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
570 		ksi->ksi_sigq = NULL;
571 		if (ksiginfo_tryfree(ksi) && p != NULL)
572 			p->p_pendingcnt--;
573 	}
574 
575 	SIGEMPTYSET(sq->sq_signals);
576 	SIGEMPTYSET(sq->sq_kill);
577 	SIGEMPTYSET(sq->sq_ptrace);
578 }
579 
580 static void
581 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
582 {
583 	sigset_t tmp;
584 	struct proc *p1, *p2;
585 	ksiginfo_t *ksi, *next;
586 
587 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
588 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
589 	p1 = src->sq_proc;
590 	p2 = dst->sq_proc;
591 	/* Move siginfo to target list */
592 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
593 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
594 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
595 			if (p1 != NULL)
596 				p1->p_pendingcnt--;
597 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
598 			ksi->ksi_sigq = dst;
599 			if (p2 != NULL)
600 				p2->p_pendingcnt++;
601 		}
602 	}
603 
604 	/* Move pending bits to target list */
605 	tmp = src->sq_kill;
606 	SIGSETAND(tmp, *set);
607 	SIGSETOR(dst->sq_kill, tmp);
608 	SIGSETNAND(src->sq_kill, tmp);
609 
610 	tmp = src->sq_ptrace;
611 	SIGSETAND(tmp, *set);
612 	SIGSETOR(dst->sq_ptrace, tmp);
613 	SIGSETNAND(src->sq_ptrace, tmp);
614 
615 	tmp = src->sq_signals;
616 	SIGSETAND(tmp, *set);
617 	SIGSETOR(dst->sq_signals, tmp);
618 	SIGSETNAND(src->sq_signals, tmp);
619 }
620 
621 #if 0
622 static void
623 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
624 {
625 	sigset_t set;
626 
627 	SIGEMPTYSET(set);
628 	SIGADDSET(set, signo);
629 	sigqueue_move_set(src, dst, &set);
630 }
631 #endif
632 
633 static void
634 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
635 {
636 	struct proc *p = sq->sq_proc;
637 	ksiginfo_t *ksi, *next;
638 
639 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
640 
641 	/* Remove siginfo queue */
642 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
643 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
644 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
645 			ksi->ksi_sigq = NULL;
646 			if (ksiginfo_tryfree(ksi) && p != NULL)
647 				p->p_pendingcnt--;
648 		}
649 	}
650 	SIGSETNAND(sq->sq_kill, *set);
651 	SIGSETNAND(sq->sq_ptrace, *set);
652 	SIGSETNAND(sq->sq_signals, *set);
653 }
654 
655 void
656 sigqueue_delete(sigqueue_t *sq, int signo)
657 {
658 	sigset_t set;
659 
660 	SIGEMPTYSET(set);
661 	SIGADDSET(set, signo);
662 	sigqueue_delete_set(sq, &set);
663 }
664 
665 /* Remove a set of signals for a process */
666 static void
667 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
668 {
669 	sigqueue_t worklist;
670 	struct thread *td0;
671 
672 	PROC_LOCK_ASSERT(p, MA_OWNED);
673 
674 	sigqueue_init(&worklist, NULL);
675 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
676 
677 	FOREACH_THREAD_IN_PROC(p, td0)
678 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
679 
680 	sigqueue_flush(&worklist);
681 }
682 
683 void
684 sigqueue_delete_proc(struct proc *p, int signo)
685 {
686 	sigset_t set;
687 
688 	SIGEMPTYSET(set);
689 	SIGADDSET(set, signo);
690 	sigqueue_delete_set_proc(p, &set);
691 }
692 
693 static void
694 sigqueue_delete_stopmask_proc(struct proc *p)
695 {
696 	sigset_t set;
697 
698 	SIGEMPTYSET(set);
699 	SIGADDSET(set, SIGSTOP);
700 	SIGADDSET(set, SIGTSTP);
701 	SIGADDSET(set, SIGTTIN);
702 	SIGADDSET(set, SIGTTOU);
703 	sigqueue_delete_set_proc(p, &set);
704 }
705 
706 /*
707  * Determine signal that should be delivered to thread td, the current
708  * thread, 0 if none.  If there is a pending stop signal with default
709  * action, the process stops in issignal().
710  */
711 int
712 cursig(struct thread *td)
713 {
714 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
715 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
716 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
717 	return (SIGPENDING(td) ? issignal(td) : 0);
718 }
719 
720 /*
721  * Arrange for ast() to handle unmasked pending signals on return to user
722  * mode.  This must be called whenever a signal is added to td_sigqueue or
723  * unmasked in td_sigmask.
724  */
725 void
726 signotify(struct thread *td)
727 {
728 
729 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
730 
731 	if (SIGPENDING(td))
732 		ast_sched(td, TDA_SIG);
733 }
734 
735 /*
736  * Returns 1 (true) if altstack is configured for the thread, and the
737  * passed stack bottom address falls into the altstack range.  Handles
738  * the 43 compat special case where the alt stack size is zero.
739  */
740 int
741 sigonstack(size_t sp)
742 {
743 	struct thread *td;
744 
745 	td = curthread;
746 	if ((td->td_pflags & TDP_ALTSTACK) == 0)
747 		return (0);
748 #if defined(COMPAT_43)
749 	if (SV_PROC_FLAG(td->td_proc, SV_AOUT) && td->td_sigstk.ss_size == 0)
750 		return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
751 #endif
752 	return (sp >= (size_t)td->td_sigstk.ss_sp &&
753 	    sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
754 }
755 
756 static __inline int
757 sigprop(int sig)
758 {
759 
760 	if (sig > 0 && sig < nitems(sigproptbl))
761 		return (sigproptbl[sig]);
762 	return (0);
763 }
764 
765 static bool
766 sigact_flag_test(const struct sigaction *act, int flag)
767 {
768 
769 	/*
770 	 * SA_SIGINFO is reset when signal disposition is set to
771 	 * ignore or default.  Other flags are kept according to user
772 	 * settings.
773 	 */
774 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
775 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
776 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
777 }
778 
779 /*
780  * kern_sigaction
781  * sigaction
782  * freebsd4_sigaction
783  * osigaction
784  */
785 int
786 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
787     struct sigaction *oact, int flags)
788 {
789 	struct sigacts *ps;
790 	struct proc *p = td->td_proc;
791 
792 	if (!_SIG_VALID(sig))
793 		return (EINVAL);
794 	if (act != NULL && act->sa_handler != SIG_DFL &&
795 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
796 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
797 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
798 		return (EINVAL);
799 
800 	PROC_LOCK(p);
801 	ps = p->p_sigacts;
802 	mtx_lock(&ps->ps_mtx);
803 	if (oact) {
804 		memset(oact, 0, sizeof(*oact));
805 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
806 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
807 			oact->sa_flags |= SA_ONSTACK;
808 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
809 			oact->sa_flags |= SA_RESTART;
810 		if (SIGISMEMBER(ps->ps_sigreset, sig))
811 			oact->sa_flags |= SA_RESETHAND;
812 		if (SIGISMEMBER(ps->ps_signodefer, sig))
813 			oact->sa_flags |= SA_NODEFER;
814 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
815 			oact->sa_flags |= SA_SIGINFO;
816 			oact->sa_sigaction =
817 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
818 		} else
819 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
820 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
821 			oact->sa_flags |= SA_NOCLDSTOP;
822 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
823 			oact->sa_flags |= SA_NOCLDWAIT;
824 	}
825 	if (act) {
826 		if ((sig == SIGKILL || sig == SIGSTOP) &&
827 		    act->sa_handler != SIG_DFL) {
828 			mtx_unlock(&ps->ps_mtx);
829 			PROC_UNLOCK(p);
830 			return (EINVAL);
831 		}
832 
833 		/*
834 		 * Change setting atomically.
835 		 */
836 
837 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
838 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
839 		if (sigact_flag_test(act, SA_SIGINFO)) {
840 			ps->ps_sigact[_SIG_IDX(sig)] =
841 			    (__sighandler_t *)act->sa_sigaction;
842 			SIGADDSET(ps->ps_siginfo, sig);
843 		} else {
844 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
845 			SIGDELSET(ps->ps_siginfo, sig);
846 		}
847 		if (!sigact_flag_test(act, SA_RESTART))
848 			SIGADDSET(ps->ps_sigintr, sig);
849 		else
850 			SIGDELSET(ps->ps_sigintr, sig);
851 		if (sigact_flag_test(act, SA_ONSTACK))
852 			SIGADDSET(ps->ps_sigonstack, sig);
853 		else
854 			SIGDELSET(ps->ps_sigonstack, sig);
855 		if (sigact_flag_test(act, SA_RESETHAND))
856 			SIGADDSET(ps->ps_sigreset, sig);
857 		else
858 			SIGDELSET(ps->ps_sigreset, sig);
859 		if (sigact_flag_test(act, SA_NODEFER))
860 			SIGADDSET(ps->ps_signodefer, sig);
861 		else
862 			SIGDELSET(ps->ps_signodefer, sig);
863 		if (sig == SIGCHLD) {
864 			if (act->sa_flags & SA_NOCLDSTOP)
865 				ps->ps_flag |= PS_NOCLDSTOP;
866 			else
867 				ps->ps_flag &= ~PS_NOCLDSTOP;
868 			if (act->sa_flags & SA_NOCLDWAIT) {
869 				/*
870 				 * Paranoia: since SA_NOCLDWAIT is implemented
871 				 * by reparenting the dying child to PID 1 (and
872 				 * trust it to reap the zombie), PID 1 itself
873 				 * is forbidden to set SA_NOCLDWAIT.
874 				 */
875 				if (p->p_pid == 1)
876 					ps->ps_flag &= ~PS_NOCLDWAIT;
877 				else
878 					ps->ps_flag |= PS_NOCLDWAIT;
879 			} else
880 				ps->ps_flag &= ~PS_NOCLDWAIT;
881 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
882 				ps->ps_flag |= PS_CLDSIGIGN;
883 			else
884 				ps->ps_flag &= ~PS_CLDSIGIGN;
885 		}
886 		/*
887 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
888 		 * and for signals set to SIG_DFL where the default is to
889 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
890 		 * have to restart the process.
891 		 */
892 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
893 		    (sigprop(sig) & SIGPROP_IGNORE &&
894 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
895 			/* never to be seen again */
896 			sigqueue_delete_proc(p, sig);
897 			if (sig != SIGCONT)
898 				/* easier in psignal */
899 				SIGADDSET(ps->ps_sigignore, sig);
900 			SIGDELSET(ps->ps_sigcatch, sig);
901 		} else {
902 			SIGDELSET(ps->ps_sigignore, sig);
903 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
904 				SIGDELSET(ps->ps_sigcatch, sig);
905 			else
906 				SIGADDSET(ps->ps_sigcatch, sig);
907 		}
908 #ifdef COMPAT_FREEBSD4
909 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
910 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
911 		    (flags & KSA_FREEBSD4) == 0)
912 			SIGDELSET(ps->ps_freebsd4, sig);
913 		else
914 			SIGADDSET(ps->ps_freebsd4, sig);
915 #endif
916 #ifdef COMPAT_43
917 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
918 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
919 		    (flags & KSA_OSIGSET) == 0)
920 			SIGDELSET(ps->ps_osigset, sig);
921 		else
922 			SIGADDSET(ps->ps_osigset, sig);
923 #endif
924 	}
925 	mtx_unlock(&ps->ps_mtx);
926 	PROC_UNLOCK(p);
927 	return (0);
928 }
929 
930 #ifndef _SYS_SYSPROTO_H_
931 struct sigaction_args {
932 	int	sig;
933 	struct	sigaction *act;
934 	struct	sigaction *oact;
935 };
936 #endif
937 int
938 sys_sigaction(struct thread *td, struct sigaction_args *uap)
939 {
940 	struct sigaction act, oact;
941 	struct sigaction *actp, *oactp;
942 	int error;
943 
944 	actp = (uap->act != NULL) ? &act : NULL;
945 	oactp = (uap->oact != NULL) ? &oact : NULL;
946 	if (actp) {
947 		error = copyin(uap->act, actp, sizeof(act));
948 		if (error)
949 			return (error);
950 	}
951 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
952 	if (oactp && !error)
953 		error = copyout(oactp, uap->oact, sizeof(oact));
954 	return (error);
955 }
956 
957 #ifdef COMPAT_FREEBSD4
958 #ifndef _SYS_SYSPROTO_H_
959 struct freebsd4_sigaction_args {
960 	int	sig;
961 	struct	sigaction *act;
962 	struct	sigaction *oact;
963 };
964 #endif
965 int
966 freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
967 {
968 	struct sigaction act, oact;
969 	struct sigaction *actp, *oactp;
970 	int error;
971 
972 	actp = (uap->act != NULL) ? &act : NULL;
973 	oactp = (uap->oact != NULL) ? &oact : NULL;
974 	if (actp) {
975 		error = copyin(uap->act, actp, sizeof(act));
976 		if (error)
977 			return (error);
978 	}
979 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
980 	if (oactp && !error)
981 		error = copyout(oactp, uap->oact, sizeof(oact));
982 	return (error);
983 }
984 #endif	/* COMAPT_FREEBSD4 */
985 
986 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
987 #ifndef _SYS_SYSPROTO_H_
988 struct osigaction_args {
989 	int	signum;
990 	struct	osigaction *nsa;
991 	struct	osigaction *osa;
992 };
993 #endif
994 int
995 osigaction(struct thread *td, struct osigaction_args *uap)
996 {
997 	struct osigaction sa;
998 	struct sigaction nsa, osa;
999 	struct sigaction *nsap, *osap;
1000 	int error;
1001 
1002 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1003 		return (EINVAL);
1004 
1005 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
1006 	osap = (uap->osa != NULL) ? &osa : NULL;
1007 
1008 	if (nsap) {
1009 		error = copyin(uap->nsa, &sa, sizeof(sa));
1010 		if (error)
1011 			return (error);
1012 		nsap->sa_handler = sa.sa_handler;
1013 		nsap->sa_flags = sa.sa_flags;
1014 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
1015 	}
1016 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1017 	if (osap && !error) {
1018 		sa.sa_handler = osap->sa_handler;
1019 		sa.sa_flags = osap->sa_flags;
1020 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
1021 		error = copyout(&sa, uap->osa, sizeof(sa));
1022 	}
1023 	return (error);
1024 }
1025 
1026 #if !defined(__i386__)
1027 /* Avoid replicating the same stub everywhere */
1028 int
1029 osigreturn(struct thread *td, struct osigreturn_args *uap)
1030 {
1031 
1032 	return (nosys(td, (struct nosys_args *)uap));
1033 }
1034 #endif
1035 #endif /* COMPAT_43 */
1036 
1037 /*
1038  * Initialize signal state for process 0;
1039  * set to ignore signals that are ignored by default.
1040  */
1041 void
1042 siginit(struct proc *p)
1043 {
1044 	int i;
1045 	struct sigacts *ps;
1046 
1047 	PROC_LOCK(p);
1048 	ps = p->p_sigacts;
1049 	mtx_lock(&ps->ps_mtx);
1050 	for (i = 1; i <= NSIG; i++) {
1051 		if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) {
1052 			SIGADDSET(ps->ps_sigignore, i);
1053 		}
1054 	}
1055 	mtx_unlock(&ps->ps_mtx);
1056 	PROC_UNLOCK(p);
1057 }
1058 
1059 /*
1060  * Reset specified signal to the default disposition.
1061  */
1062 static void
1063 sigdflt(struct sigacts *ps, int sig)
1064 {
1065 
1066 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1067 	SIGDELSET(ps->ps_sigcatch, sig);
1068 	if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT)
1069 		SIGADDSET(ps->ps_sigignore, sig);
1070 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1071 	SIGDELSET(ps->ps_siginfo, sig);
1072 }
1073 
1074 /*
1075  * Reset signals for an exec of the specified process.
1076  */
1077 void
1078 execsigs(struct proc *p)
1079 {
1080 	struct sigacts *ps;
1081 	struct thread *td;
1082 
1083 	/*
1084 	 * Reset caught signals.  Held signals remain held
1085 	 * through td_sigmask (unless they were caught,
1086 	 * and are now ignored by default).
1087 	 */
1088 	PROC_LOCK_ASSERT(p, MA_OWNED);
1089 	ps = p->p_sigacts;
1090 	mtx_lock(&ps->ps_mtx);
1091 	sig_drop_caught(p);
1092 
1093 	/*
1094 	 * Reset stack state to the user stack.
1095 	 * Clear set of signals caught on the signal stack.
1096 	 */
1097 	td = curthread;
1098 	MPASS(td->td_proc == p);
1099 	td->td_sigstk.ss_flags = SS_DISABLE;
1100 	td->td_sigstk.ss_size = 0;
1101 	td->td_sigstk.ss_sp = 0;
1102 	td->td_pflags &= ~TDP_ALTSTACK;
1103 	/*
1104 	 * Reset no zombies if child dies flag as Solaris does.
1105 	 */
1106 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1107 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1108 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1109 	mtx_unlock(&ps->ps_mtx);
1110 }
1111 
1112 /*
1113  * kern_sigprocmask()
1114  *
1115  *	Manipulate signal mask.
1116  */
1117 int
1118 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1119     int flags)
1120 {
1121 	sigset_t new_block, oset1;
1122 	struct proc *p;
1123 	int error;
1124 
1125 	p = td->td_proc;
1126 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1127 		PROC_LOCK_ASSERT(p, MA_OWNED);
1128 	else
1129 		PROC_LOCK(p);
1130 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1131 	    ? MA_OWNED : MA_NOTOWNED);
1132 	if (oset != NULL)
1133 		*oset = td->td_sigmask;
1134 
1135 	error = 0;
1136 	if (set != NULL) {
1137 		switch (how) {
1138 		case SIG_BLOCK:
1139 			SIG_CANTMASK(*set);
1140 			oset1 = td->td_sigmask;
1141 			SIGSETOR(td->td_sigmask, *set);
1142 			new_block = td->td_sigmask;
1143 			SIGSETNAND(new_block, oset1);
1144 			break;
1145 		case SIG_UNBLOCK:
1146 			SIGSETNAND(td->td_sigmask, *set);
1147 			signotify(td);
1148 			goto out;
1149 		case SIG_SETMASK:
1150 			SIG_CANTMASK(*set);
1151 			oset1 = td->td_sigmask;
1152 			if (flags & SIGPROCMASK_OLD)
1153 				SIGSETLO(td->td_sigmask, *set);
1154 			else
1155 				td->td_sigmask = *set;
1156 			new_block = td->td_sigmask;
1157 			SIGSETNAND(new_block, oset1);
1158 			signotify(td);
1159 			break;
1160 		default:
1161 			error = EINVAL;
1162 			goto out;
1163 		}
1164 
1165 		/*
1166 		 * The new_block set contains signals that were not previously
1167 		 * blocked, but are blocked now.
1168 		 *
1169 		 * In case we block any signal that was not previously blocked
1170 		 * for td, and process has the signal pending, try to schedule
1171 		 * signal delivery to some thread that does not block the
1172 		 * signal, possibly waking it up.
1173 		 */
1174 		if (p->p_numthreads != 1)
1175 			reschedule_signals(p, new_block, flags);
1176 	}
1177 
1178 out:
1179 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1180 		PROC_UNLOCK(p);
1181 	return (error);
1182 }
1183 
1184 #ifndef _SYS_SYSPROTO_H_
1185 struct sigprocmask_args {
1186 	int	how;
1187 	const sigset_t *set;
1188 	sigset_t *oset;
1189 };
1190 #endif
1191 int
1192 sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1193 {
1194 	sigset_t set, oset;
1195 	sigset_t *setp, *osetp;
1196 	int error;
1197 
1198 	setp = (uap->set != NULL) ? &set : NULL;
1199 	osetp = (uap->oset != NULL) ? &oset : NULL;
1200 	if (setp) {
1201 		error = copyin(uap->set, setp, sizeof(set));
1202 		if (error)
1203 			return (error);
1204 	}
1205 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1206 	if (osetp && !error) {
1207 		error = copyout(osetp, uap->oset, sizeof(oset));
1208 	}
1209 	return (error);
1210 }
1211 
1212 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1213 #ifndef _SYS_SYSPROTO_H_
1214 struct osigprocmask_args {
1215 	int	how;
1216 	osigset_t mask;
1217 };
1218 #endif
1219 int
1220 osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1221 {
1222 	sigset_t set, oset;
1223 	int error;
1224 
1225 	OSIG2SIG(uap->mask, set);
1226 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1227 	SIG2OSIG(oset, td->td_retval[0]);
1228 	return (error);
1229 }
1230 #endif /* COMPAT_43 */
1231 
1232 int
1233 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1234 {
1235 	ksiginfo_t ksi;
1236 	sigset_t set;
1237 	int error;
1238 
1239 	error = copyin(uap->set, &set, sizeof(set));
1240 	if (error) {
1241 		td->td_retval[0] = error;
1242 		return (0);
1243 	}
1244 
1245 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1246 	if (error) {
1247 		/*
1248 		 * sigwait() function shall not return EINTR, but
1249 		 * the syscall does.  Non-ancient libc provides the
1250 		 * wrapper which hides EINTR.  Otherwise, EINTR return
1251 		 * is used by libthr to handle required cancellation
1252 		 * point in the sigwait().
1253 		 */
1254 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1255 			return (ERESTART);
1256 		td->td_retval[0] = error;
1257 		return (0);
1258 	}
1259 
1260 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1261 	td->td_retval[0] = error;
1262 	return (0);
1263 }
1264 
1265 int
1266 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1267 {
1268 	struct timespec ts;
1269 	struct timespec *timeout;
1270 	sigset_t set;
1271 	ksiginfo_t ksi;
1272 	int error;
1273 
1274 	if (uap->timeout) {
1275 		error = copyin(uap->timeout, &ts, sizeof(ts));
1276 		if (error)
1277 			return (error);
1278 
1279 		timeout = &ts;
1280 	} else
1281 		timeout = NULL;
1282 
1283 	error = copyin(uap->set, &set, sizeof(set));
1284 	if (error)
1285 		return (error);
1286 
1287 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1288 	if (error)
1289 		return (error);
1290 
1291 	if (uap->info)
1292 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1293 
1294 	if (error == 0)
1295 		td->td_retval[0] = ksi.ksi_signo;
1296 	return (error);
1297 }
1298 
1299 int
1300 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1301 {
1302 	ksiginfo_t ksi;
1303 	sigset_t set;
1304 	int error;
1305 
1306 	error = copyin(uap->set, &set, sizeof(set));
1307 	if (error)
1308 		return (error);
1309 
1310 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1311 	if (error)
1312 		return (error);
1313 
1314 	if (uap->info)
1315 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1316 
1317 	if (error == 0)
1318 		td->td_retval[0] = ksi.ksi_signo;
1319 	return (error);
1320 }
1321 
1322 static void
1323 proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1324 {
1325 	struct thread *thr;
1326 
1327 	FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1328 		if (thr == td)
1329 			thr->td_si = *si;
1330 		else
1331 			thr->td_si.si_signo = 0;
1332 	}
1333 }
1334 
1335 int
1336 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1337 	struct timespec *timeout)
1338 {
1339 	struct sigacts *ps;
1340 	sigset_t saved_mask, new_block;
1341 	struct proc *p;
1342 	int error, sig, timevalid = 0;
1343 	sbintime_t sbt, precision, tsbt;
1344 	struct timespec ts;
1345 	bool traced;
1346 
1347 	p = td->td_proc;
1348 	error = 0;
1349 	traced = false;
1350 
1351 	/* Ensure the sigfastblock value is up to date. */
1352 	sigfastblock_fetch(td);
1353 
1354 	if (timeout != NULL) {
1355 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1356 			timevalid = 1;
1357 			ts = *timeout;
1358 			if (ts.tv_sec < INT32_MAX / 2) {
1359 				tsbt = tstosbt(ts);
1360 				precision = tsbt;
1361 				precision >>= tc_precexp;
1362 				if (TIMESEL(&sbt, tsbt))
1363 					sbt += tc_tick_sbt;
1364 				sbt += tsbt;
1365 			} else
1366 				precision = sbt = 0;
1367 		}
1368 	} else
1369 		precision = sbt = 0;
1370 	ksiginfo_init(ksi);
1371 	/* Some signals can not be waited for. */
1372 	SIG_CANTMASK(waitset);
1373 	ps = p->p_sigacts;
1374 	PROC_LOCK(p);
1375 	saved_mask = td->td_sigmask;
1376 	SIGSETNAND(td->td_sigmask, waitset);
1377 	if ((p->p_sysent->sv_flags & SV_SIG_DISCIGN) != 0 ||
1378 	    !kern_sig_discard_ign) {
1379 		thread_lock(td);
1380 		td->td_flags |= TDF_SIGWAIT;
1381 		thread_unlock(td);
1382 	}
1383 	for (;;) {
1384 		mtx_lock(&ps->ps_mtx);
1385 		sig = cursig(td);
1386 		mtx_unlock(&ps->ps_mtx);
1387 		KASSERT(sig >= 0, ("sig %d", sig));
1388 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1389 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1390 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1391 				error = 0;
1392 				break;
1393 			}
1394 		}
1395 
1396 		if (error != 0)
1397 			break;
1398 
1399 		/*
1400 		 * POSIX says this must be checked after looking for pending
1401 		 * signals.
1402 		 */
1403 		if (timeout != NULL && !timevalid) {
1404 			error = EINVAL;
1405 			break;
1406 		}
1407 
1408 		if (traced) {
1409 			error = EINTR;
1410 			break;
1411 		}
1412 
1413 		error = msleep_sbt(&p->p_sigacts, &p->p_mtx, PPAUSE | PCATCH,
1414 		    "sigwait", sbt, precision, C_ABSOLUTE);
1415 
1416 		/* The syscalls can not be restarted. */
1417 		if (error == ERESTART)
1418 			error = EINTR;
1419 
1420 		/*
1421 		 * If PTRACE_SCE or PTRACE_SCX were set after
1422 		 * userspace entered the syscall, return spurious
1423 		 * EINTR after wait was done.  Only do this as last
1424 		 * resort after rechecking for possible queued signals
1425 		 * and expired timeouts.
1426 		 */
1427 		if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1428 			traced = true;
1429 	}
1430 	thread_lock(td);
1431 	td->td_flags &= ~TDF_SIGWAIT;
1432 	thread_unlock(td);
1433 
1434 	new_block = saved_mask;
1435 	SIGSETNAND(new_block, td->td_sigmask);
1436 	td->td_sigmask = saved_mask;
1437 	/*
1438 	 * Fewer signals can be delivered to us, reschedule signal
1439 	 * notification.
1440 	 */
1441 	if (p->p_numthreads != 1)
1442 		reschedule_signals(p, new_block, 0);
1443 
1444 	if (error == 0) {
1445 		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1446 
1447 		if (ksi->ksi_code == SI_TIMER)
1448 			itimer_accept(p, ksi->ksi_timerid, ksi);
1449 
1450 #ifdef KTRACE
1451 		if (KTRPOINT(td, KTR_PSIG)) {
1452 			sig_t action;
1453 
1454 			mtx_lock(&ps->ps_mtx);
1455 			action = ps->ps_sigact[_SIG_IDX(sig)];
1456 			mtx_unlock(&ps->ps_mtx);
1457 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1458 		}
1459 #endif
1460 		if (sig == SIGKILL) {
1461 			proc_td_siginfo_capture(td, &ksi->ksi_info);
1462 			sigexit(td, sig);
1463 		}
1464 	}
1465 	PROC_UNLOCK(p);
1466 	return (error);
1467 }
1468 
1469 #ifndef _SYS_SYSPROTO_H_
1470 struct sigpending_args {
1471 	sigset_t	*set;
1472 };
1473 #endif
1474 int
1475 sys_sigpending(struct thread *td, struct sigpending_args *uap)
1476 {
1477 	struct proc *p = td->td_proc;
1478 	sigset_t pending;
1479 
1480 	PROC_LOCK(p);
1481 	pending = p->p_sigqueue.sq_signals;
1482 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1483 	PROC_UNLOCK(p);
1484 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1485 }
1486 
1487 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1488 #ifndef _SYS_SYSPROTO_H_
1489 struct osigpending_args {
1490 	int	dummy;
1491 };
1492 #endif
1493 int
1494 osigpending(struct thread *td, struct osigpending_args *uap)
1495 {
1496 	struct proc *p = td->td_proc;
1497 	sigset_t pending;
1498 
1499 	PROC_LOCK(p);
1500 	pending = p->p_sigqueue.sq_signals;
1501 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1502 	PROC_UNLOCK(p);
1503 	SIG2OSIG(pending, td->td_retval[0]);
1504 	return (0);
1505 }
1506 #endif /* COMPAT_43 */
1507 
1508 #if defined(COMPAT_43)
1509 /*
1510  * Generalized interface signal handler, 4.3-compatible.
1511  */
1512 #ifndef _SYS_SYSPROTO_H_
1513 struct osigvec_args {
1514 	int	signum;
1515 	struct	sigvec *nsv;
1516 	struct	sigvec *osv;
1517 };
1518 #endif
1519 /* ARGSUSED */
1520 int
1521 osigvec(struct thread *td, struct osigvec_args *uap)
1522 {
1523 	struct sigvec vec;
1524 	struct sigaction nsa, osa;
1525 	struct sigaction *nsap, *osap;
1526 	int error;
1527 
1528 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1529 		return (EINVAL);
1530 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1531 	osap = (uap->osv != NULL) ? &osa : NULL;
1532 	if (nsap) {
1533 		error = copyin(uap->nsv, &vec, sizeof(vec));
1534 		if (error)
1535 			return (error);
1536 		nsap->sa_handler = vec.sv_handler;
1537 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1538 		nsap->sa_flags = vec.sv_flags;
1539 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1540 	}
1541 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1542 	if (osap && !error) {
1543 		vec.sv_handler = osap->sa_handler;
1544 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1545 		vec.sv_flags = osap->sa_flags;
1546 		vec.sv_flags &= ~SA_NOCLDWAIT;
1547 		vec.sv_flags ^= SA_RESTART;
1548 		error = copyout(&vec, uap->osv, sizeof(vec));
1549 	}
1550 	return (error);
1551 }
1552 
1553 #ifndef _SYS_SYSPROTO_H_
1554 struct osigblock_args {
1555 	int	mask;
1556 };
1557 #endif
1558 int
1559 osigblock(struct thread *td, struct osigblock_args *uap)
1560 {
1561 	sigset_t set, oset;
1562 
1563 	OSIG2SIG(uap->mask, set);
1564 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1565 	SIG2OSIG(oset, td->td_retval[0]);
1566 	return (0);
1567 }
1568 
1569 #ifndef _SYS_SYSPROTO_H_
1570 struct osigsetmask_args {
1571 	int	mask;
1572 };
1573 #endif
1574 int
1575 osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1576 {
1577 	sigset_t set, oset;
1578 
1579 	OSIG2SIG(uap->mask, set);
1580 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1581 	SIG2OSIG(oset, td->td_retval[0]);
1582 	return (0);
1583 }
1584 #endif /* COMPAT_43 */
1585 
1586 /*
1587  * Suspend calling thread until signal, providing mask to be set in the
1588  * meantime.
1589  */
1590 #ifndef _SYS_SYSPROTO_H_
1591 struct sigsuspend_args {
1592 	const sigset_t *sigmask;
1593 };
1594 #endif
1595 /* ARGSUSED */
1596 int
1597 sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1598 {
1599 	sigset_t mask;
1600 	int error;
1601 
1602 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1603 	if (error)
1604 		return (error);
1605 	return (kern_sigsuspend(td, mask));
1606 }
1607 
1608 int
1609 kern_sigsuspend(struct thread *td, sigset_t mask)
1610 {
1611 	struct proc *p = td->td_proc;
1612 	int has_sig, sig;
1613 
1614 	/* Ensure the sigfastblock value is up to date. */
1615 	sigfastblock_fetch(td);
1616 
1617 	/*
1618 	 * When returning from sigsuspend, we want
1619 	 * the old mask to be restored after the
1620 	 * signal handler has finished.  Thus, we
1621 	 * save it here and mark the sigacts structure
1622 	 * to indicate this.
1623 	 */
1624 	PROC_LOCK(p);
1625 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1626 	    SIGPROCMASK_PROC_LOCKED);
1627 	td->td_pflags |= TDP_OLDMASK;
1628 	ast_sched(td, TDA_SIGSUSPEND);
1629 
1630 	/*
1631 	 * Process signals now. Otherwise, we can get spurious wakeup
1632 	 * due to signal entered process queue, but delivered to other
1633 	 * thread. But sigsuspend should return only on signal
1634 	 * delivery.
1635 	 */
1636 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1637 	for (has_sig = 0; !has_sig;) {
1638 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1639 			0) == 0)
1640 			/* void */;
1641 		thread_suspend_check(0);
1642 		mtx_lock(&p->p_sigacts->ps_mtx);
1643 		while ((sig = cursig(td)) != 0) {
1644 			KASSERT(sig >= 0, ("sig %d", sig));
1645 			has_sig += postsig(sig);
1646 		}
1647 		mtx_unlock(&p->p_sigacts->ps_mtx);
1648 
1649 		/*
1650 		 * If PTRACE_SCE or PTRACE_SCX were set after
1651 		 * userspace entered the syscall, return spurious
1652 		 * EINTR.
1653 		 */
1654 		if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1655 			has_sig += 1;
1656 	}
1657 	PROC_UNLOCK(p);
1658 	td->td_errno = EINTR;
1659 	td->td_pflags |= TDP_NERRNO;
1660 	return (EJUSTRETURN);
1661 }
1662 
1663 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1664 /*
1665  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1666  * convention: libc stub passes mask, not pointer, to save a copyin.
1667  */
1668 #ifndef _SYS_SYSPROTO_H_
1669 struct osigsuspend_args {
1670 	osigset_t mask;
1671 };
1672 #endif
1673 /* ARGSUSED */
1674 int
1675 osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1676 {
1677 	sigset_t mask;
1678 
1679 	OSIG2SIG(uap->mask, mask);
1680 	return (kern_sigsuspend(td, mask));
1681 }
1682 #endif /* COMPAT_43 */
1683 
1684 #if defined(COMPAT_43)
1685 #ifndef _SYS_SYSPROTO_H_
1686 struct osigstack_args {
1687 	struct	sigstack *nss;
1688 	struct	sigstack *oss;
1689 };
1690 #endif
1691 /* ARGSUSED */
1692 int
1693 osigstack(struct thread *td, struct osigstack_args *uap)
1694 {
1695 	struct sigstack nss, oss;
1696 	int error = 0;
1697 
1698 	if (uap->nss != NULL) {
1699 		error = copyin(uap->nss, &nss, sizeof(nss));
1700 		if (error)
1701 			return (error);
1702 	}
1703 	oss.ss_sp = td->td_sigstk.ss_sp;
1704 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1705 	if (uap->nss != NULL) {
1706 		td->td_sigstk.ss_sp = nss.ss_sp;
1707 		td->td_sigstk.ss_size = 0;
1708 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1709 		td->td_pflags |= TDP_ALTSTACK;
1710 	}
1711 	if (uap->oss != NULL)
1712 		error = copyout(&oss, uap->oss, sizeof(oss));
1713 
1714 	return (error);
1715 }
1716 #endif /* COMPAT_43 */
1717 
1718 #ifndef _SYS_SYSPROTO_H_
1719 struct sigaltstack_args {
1720 	stack_t	*ss;
1721 	stack_t	*oss;
1722 };
1723 #endif
1724 /* ARGSUSED */
1725 int
1726 sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1727 {
1728 	stack_t ss, oss;
1729 	int error;
1730 
1731 	if (uap->ss != NULL) {
1732 		error = copyin(uap->ss, &ss, sizeof(ss));
1733 		if (error)
1734 			return (error);
1735 	}
1736 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1737 	    (uap->oss != NULL) ? &oss : NULL);
1738 	if (error)
1739 		return (error);
1740 	if (uap->oss != NULL)
1741 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1742 	return (error);
1743 }
1744 
1745 int
1746 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1747 {
1748 	struct proc *p = td->td_proc;
1749 	int oonstack;
1750 
1751 	oonstack = sigonstack(cpu_getstack(td));
1752 
1753 	if (oss != NULL) {
1754 		*oss = td->td_sigstk;
1755 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1756 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1757 	}
1758 
1759 	if (ss != NULL) {
1760 		if (oonstack)
1761 			return (EPERM);
1762 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1763 			return (EINVAL);
1764 		if (!(ss->ss_flags & SS_DISABLE)) {
1765 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1766 				return (ENOMEM);
1767 
1768 			td->td_sigstk = *ss;
1769 			td->td_pflags |= TDP_ALTSTACK;
1770 		} else {
1771 			td->td_pflags &= ~TDP_ALTSTACK;
1772 		}
1773 	}
1774 	return (0);
1775 }
1776 
1777 struct killpg1_ctx {
1778 	struct thread *td;
1779 	ksiginfo_t *ksi;
1780 	int sig;
1781 	bool sent;
1782 	bool found;
1783 	int ret;
1784 };
1785 
1786 static void
1787 killpg1_sendsig_locked(struct proc *p, struct killpg1_ctx *arg)
1788 {
1789 	int err;
1790 
1791 	err = p_cansignal(arg->td, p, arg->sig);
1792 	if (err == 0 && arg->sig != 0)
1793 		pksignal(p, arg->sig, arg->ksi);
1794 	if (err != ESRCH)
1795 		arg->found = true;
1796 	if (err == 0)
1797 		arg->sent = true;
1798 	else if (arg->ret == 0 && err != ESRCH && err != EPERM)
1799 		arg->ret = err;
1800 }
1801 
1802 static void
1803 killpg1_sendsig(struct proc *p, bool notself, struct killpg1_ctx *arg)
1804 {
1805 
1806 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1807 	    (notself && p == arg->td->td_proc) || p->p_state == PRS_NEW)
1808 		return;
1809 
1810 	PROC_LOCK(p);
1811 	killpg1_sendsig_locked(p, arg);
1812 	PROC_UNLOCK(p);
1813 }
1814 
1815 static void
1816 kill_processes_prison_cb(struct proc *p, void *arg)
1817 {
1818 	struct killpg1_ctx *ctx = arg;
1819 
1820 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1821 	    (p == ctx->td->td_proc) || p->p_state == PRS_NEW)
1822 		return;
1823 
1824 	killpg1_sendsig_locked(p, ctx);
1825 }
1826 
1827 /*
1828  * Common code for kill process group/broadcast kill.
1829  * td is the calling thread, as usual.
1830  */
1831 static int
1832 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1833 {
1834 	struct proc *p;
1835 	struct pgrp *pgrp;
1836 	struct killpg1_ctx arg;
1837 
1838 	arg.td = td;
1839 	arg.ksi = ksi;
1840 	arg.sig = sig;
1841 	arg.sent = false;
1842 	arg.found = false;
1843 	arg.ret = 0;
1844 	if (all) {
1845 		/*
1846 		 * broadcast
1847 		 */
1848 		prison_proc_iterate(td->td_ucred->cr_prison,
1849 		    kill_processes_prison_cb, &arg);
1850 	} else {
1851 again:
1852 		sx_slock(&proctree_lock);
1853 		if (pgid == 0) {
1854 			/*
1855 			 * zero pgid means send to my process group.
1856 			 */
1857 			pgrp = td->td_proc->p_pgrp;
1858 			PGRP_LOCK(pgrp);
1859 		} else {
1860 			pgrp = pgfind(pgid);
1861 			if (pgrp == NULL) {
1862 				sx_sunlock(&proctree_lock);
1863 				return (ESRCH);
1864 			}
1865 		}
1866 		sx_sunlock(&proctree_lock);
1867 		if (!sx_try_xlock(&pgrp->pg_killsx)) {
1868 			PGRP_UNLOCK(pgrp);
1869 			sx_xlock(&pgrp->pg_killsx);
1870 			sx_xunlock(&pgrp->pg_killsx);
1871 			goto again;
1872 		}
1873 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1874 			killpg1_sendsig(p, false, &arg);
1875 		}
1876 		PGRP_UNLOCK(pgrp);
1877 		sx_xunlock(&pgrp->pg_killsx);
1878 	}
1879 	MPASS(arg.ret != 0 || arg.found || !arg.sent);
1880 	if (arg.ret == 0 && !arg.sent)
1881 		arg.ret = arg.found ? EPERM : ESRCH;
1882 	return (arg.ret);
1883 }
1884 
1885 #ifndef _SYS_SYSPROTO_H_
1886 struct kill_args {
1887 	int	pid;
1888 	int	signum;
1889 };
1890 #endif
1891 /* ARGSUSED */
1892 int
1893 sys_kill(struct thread *td, struct kill_args *uap)
1894 {
1895 
1896 	return (kern_kill(td, uap->pid, uap->signum));
1897 }
1898 
1899 int
1900 kern_kill(struct thread *td, pid_t pid, int signum)
1901 {
1902 	ksiginfo_t ksi;
1903 	struct proc *p;
1904 	int error;
1905 
1906 	/*
1907 	 * A process in capability mode can send signals only to himself.
1908 	 * The main rationale behind this is that abort(3) is implemented as
1909 	 * kill(getpid(), SIGABRT).
1910 	 */
1911 	if (IN_CAPABILITY_MODE(td) && pid != td->td_proc->p_pid)
1912 		return (ECAPMODE);
1913 
1914 	AUDIT_ARG_SIGNUM(signum);
1915 	AUDIT_ARG_PID(pid);
1916 	if ((u_int)signum > _SIG_MAXSIG)
1917 		return (EINVAL);
1918 
1919 	ksiginfo_init(&ksi);
1920 	ksi.ksi_signo = signum;
1921 	ksi.ksi_code = SI_USER;
1922 	ksi.ksi_pid = td->td_proc->p_pid;
1923 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1924 
1925 	if (pid > 0) {
1926 		/* kill single process */
1927 		if ((p = pfind_any(pid)) == NULL)
1928 			return (ESRCH);
1929 		AUDIT_ARG_PROCESS(p);
1930 		error = p_cansignal(td, p, signum);
1931 		if (error == 0 && signum)
1932 			pksignal(p, signum, &ksi);
1933 		PROC_UNLOCK(p);
1934 		return (error);
1935 	}
1936 	switch (pid) {
1937 	case -1:		/* broadcast signal */
1938 		return (killpg1(td, signum, 0, 1, &ksi));
1939 	case 0:			/* signal own process group */
1940 		return (killpg1(td, signum, 0, 0, &ksi));
1941 	default:		/* negative explicit process group */
1942 		return (killpg1(td, signum, -pid, 0, &ksi));
1943 	}
1944 	/* NOTREACHED */
1945 }
1946 
1947 int
1948 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1949 {
1950 	struct proc *p;
1951 	int error;
1952 
1953 	AUDIT_ARG_SIGNUM(uap->signum);
1954 	AUDIT_ARG_FD(uap->fd);
1955 	if ((u_int)uap->signum > _SIG_MAXSIG)
1956 		return (EINVAL);
1957 
1958 	error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1959 	if (error)
1960 		return (error);
1961 	AUDIT_ARG_PROCESS(p);
1962 	error = p_cansignal(td, p, uap->signum);
1963 	if (error == 0 && uap->signum)
1964 		kern_psignal(p, uap->signum);
1965 	PROC_UNLOCK(p);
1966 	return (error);
1967 }
1968 
1969 #if defined(COMPAT_43)
1970 #ifndef _SYS_SYSPROTO_H_
1971 struct okillpg_args {
1972 	int	pgid;
1973 	int	signum;
1974 };
1975 #endif
1976 /* ARGSUSED */
1977 int
1978 okillpg(struct thread *td, struct okillpg_args *uap)
1979 {
1980 	ksiginfo_t ksi;
1981 
1982 	AUDIT_ARG_SIGNUM(uap->signum);
1983 	AUDIT_ARG_PID(uap->pgid);
1984 	if ((u_int)uap->signum > _SIG_MAXSIG)
1985 		return (EINVAL);
1986 
1987 	ksiginfo_init(&ksi);
1988 	ksi.ksi_signo = uap->signum;
1989 	ksi.ksi_code = SI_USER;
1990 	ksi.ksi_pid = td->td_proc->p_pid;
1991 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1992 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1993 }
1994 #endif /* COMPAT_43 */
1995 
1996 #ifndef _SYS_SYSPROTO_H_
1997 struct sigqueue_args {
1998 	pid_t pid;
1999 	int signum;
2000 	/* union sigval */ void *value;
2001 };
2002 #endif
2003 int
2004 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
2005 {
2006 	union sigval sv;
2007 
2008 	sv.sival_ptr = uap->value;
2009 
2010 	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
2011 }
2012 
2013 int
2014 kern_sigqueue(struct thread *td, pid_t pid, int signum, union sigval *value)
2015 {
2016 	ksiginfo_t ksi;
2017 	struct proc *p;
2018 	int error;
2019 
2020 	if ((u_int)signum > _SIG_MAXSIG)
2021 		return (EINVAL);
2022 
2023 	/*
2024 	 * Specification says sigqueue can only send signal to
2025 	 * single process.
2026 	 */
2027 	if (pid <= 0)
2028 		return (EINVAL);
2029 
2030 	if ((p = pfind_any(pid)) == NULL)
2031 		return (ESRCH);
2032 	error = p_cansignal(td, p, signum);
2033 	if (error == 0 && signum != 0) {
2034 		ksiginfo_init(&ksi);
2035 		ksi.ksi_flags = KSI_SIGQ;
2036 		ksi.ksi_signo = signum;
2037 		ksi.ksi_code = SI_QUEUE;
2038 		ksi.ksi_pid = td->td_proc->p_pid;
2039 		ksi.ksi_uid = td->td_ucred->cr_ruid;
2040 		ksi.ksi_value = *value;
2041 		error = pksignal(p, ksi.ksi_signo, &ksi);
2042 	}
2043 	PROC_UNLOCK(p);
2044 	return (error);
2045 }
2046 
2047 /*
2048  * Send a signal to a process group.  If checktty is 1,
2049  * limit to members which have a controlling terminal.
2050  */
2051 void
2052 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
2053 {
2054 	struct proc *p;
2055 
2056 	if (pgrp) {
2057 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
2058 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
2059 			PROC_LOCK(p);
2060 			if (p->p_state == PRS_NORMAL &&
2061 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
2062 				pksignal(p, sig, ksi);
2063 			PROC_UNLOCK(p);
2064 		}
2065 	}
2066 }
2067 
2068 /*
2069  * Recalculate the signal mask and reset the signal disposition after
2070  * usermode frame for delivery is formed.  Should be called after
2071  * mach-specific routine, because sysent->sv_sendsig() needs correct
2072  * ps_siginfo and signal mask.
2073  */
2074 static void
2075 postsig_done(int sig, struct thread *td, struct sigacts *ps)
2076 {
2077 	sigset_t mask;
2078 
2079 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2080 	td->td_ru.ru_nsignals++;
2081 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
2082 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
2083 		SIGADDSET(mask, sig);
2084 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
2085 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
2086 	if (SIGISMEMBER(ps->ps_sigreset, sig))
2087 		sigdflt(ps, sig);
2088 }
2089 
2090 /*
2091  * Send a signal caused by a trap to the current thread.  If it will be
2092  * caught immediately, deliver it with correct code.  Otherwise, post it
2093  * normally.
2094  */
2095 void
2096 trapsignal(struct thread *td, ksiginfo_t *ksi)
2097 {
2098 	struct sigacts *ps;
2099 	struct proc *p;
2100 	sigset_t sigmask;
2101 	int sig;
2102 
2103 	p = td->td_proc;
2104 	sig = ksi->ksi_signo;
2105 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
2106 
2107 	sigfastblock_fetch(td);
2108 	PROC_LOCK(p);
2109 	ps = p->p_sigacts;
2110 	mtx_lock(&ps->ps_mtx);
2111 	sigmask = td->td_sigmask;
2112 	if (td->td_sigblock_val != 0)
2113 		SIGSETOR(sigmask, fastblock_mask);
2114 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2115 	    !SIGISMEMBER(sigmask, sig)) {
2116 #ifdef KTRACE
2117 		if (KTRPOINT(curthread, KTR_PSIG))
2118 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2119 			    &td->td_sigmask, ksi->ksi_code);
2120 #endif
2121 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2122 		    ksi, &td->td_sigmask);
2123 		postsig_done(sig, td, ps);
2124 		mtx_unlock(&ps->ps_mtx);
2125 	} else {
2126 		/*
2127 		 * Avoid a possible infinite loop if the thread
2128 		 * masking the signal or process is ignoring the
2129 		 * signal.
2130 		 */
2131 		if (kern_forcesigexit && (SIGISMEMBER(sigmask, sig) ||
2132 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2133 			SIGDELSET(td->td_sigmask, sig);
2134 			SIGDELSET(ps->ps_sigcatch, sig);
2135 			SIGDELSET(ps->ps_sigignore, sig);
2136 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2137 			td->td_pflags &= ~TDP_SIGFASTBLOCK;
2138 			td->td_sigblock_val = 0;
2139 		}
2140 		mtx_unlock(&ps->ps_mtx);
2141 		p->p_sig = sig;		/* XXX to verify code */
2142 		tdsendsignal(p, td, sig, ksi);
2143 	}
2144 	PROC_UNLOCK(p);
2145 }
2146 
2147 static struct thread *
2148 sigtd(struct proc *p, int sig, bool fast_sigblock)
2149 {
2150 	struct thread *td, *signal_td;
2151 
2152 	PROC_LOCK_ASSERT(p, MA_OWNED);
2153 	MPASS(!fast_sigblock || p == curproc);
2154 
2155 	/*
2156 	 * Check if current thread can handle the signal without
2157 	 * switching context to another thread.
2158 	 */
2159 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig) &&
2160 	    (!fast_sigblock || curthread->td_sigblock_val == 0))
2161 		return (curthread);
2162 
2163 	/* Find a non-stopped thread that does not mask the signal. */
2164 	signal_td = NULL;
2165 	FOREACH_THREAD_IN_PROC(p, td) {
2166 		if (!SIGISMEMBER(td->td_sigmask, sig) && (!fast_sigblock ||
2167 		    td != curthread || td->td_sigblock_val == 0) &&
2168 		    (td->td_flags & TDF_BOUNDARY) == 0) {
2169 			signal_td = td;
2170 			break;
2171 		}
2172 	}
2173 	/* Select random (first) thread if no better match was found. */
2174 	if (signal_td == NULL)
2175 		signal_td = FIRST_THREAD_IN_PROC(p);
2176 	return (signal_td);
2177 }
2178 
2179 /*
2180  * Send the signal to the process.  If the signal has an action, the action
2181  * is usually performed by the target process rather than the caller; we add
2182  * the signal to the set of pending signals for the process.
2183  *
2184  * Exceptions:
2185  *   o When a stop signal is sent to a sleeping process that takes the
2186  *     default action, the process is stopped without awakening it.
2187  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2188  *     regardless of the signal action (eg, blocked or ignored).
2189  *
2190  * Other ignored signals are discarded immediately.
2191  *
2192  * NB: This function may be entered from the debugger via the "kill" DDB
2193  * command.  There is little that can be done to mitigate the possibly messy
2194  * side effects of this unwise possibility.
2195  */
2196 void
2197 kern_psignal(struct proc *p, int sig)
2198 {
2199 	ksiginfo_t ksi;
2200 
2201 	ksiginfo_init(&ksi);
2202 	ksi.ksi_signo = sig;
2203 	ksi.ksi_code = SI_KERNEL;
2204 	(void) tdsendsignal(p, NULL, sig, &ksi);
2205 }
2206 
2207 int
2208 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2209 {
2210 
2211 	return (tdsendsignal(p, NULL, sig, ksi));
2212 }
2213 
2214 /* Utility function for finding a thread to send signal event to. */
2215 int
2216 sigev_findtd(struct proc *p, struct sigevent *sigev, struct thread **ttd)
2217 {
2218 	struct thread *td;
2219 
2220 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2221 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2222 		if (td == NULL)
2223 			return (ESRCH);
2224 		*ttd = td;
2225 	} else {
2226 		*ttd = NULL;
2227 		PROC_LOCK(p);
2228 	}
2229 	return (0);
2230 }
2231 
2232 void
2233 tdsignal(struct thread *td, int sig)
2234 {
2235 	ksiginfo_t ksi;
2236 
2237 	ksiginfo_init(&ksi);
2238 	ksi.ksi_signo = sig;
2239 	ksi.ksi_code = SI_KERNEL;
2240 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2241 }
2242 
2243 void
2244 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2245 {
2246 
2247 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2248 }
2249 
2250 static int
2251 sig_sleepq_abort(struct thread *td, int intrval)
2252 {
2253 	THREAD_LOCK_ASSERT(td, MA_OWNED);
2254 
2255 	if (intrval == 0 && (td->td_flags & TDF_SIGWAIT) == 0) {
2256 		thread_unlock(td);
2257 		return (0);
2258 	}
2259 	return (sleepq_abort(td, intrval));
2260 }
2261 
2262 int
2263 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2264 {
2265 	sig_t action;
2266 	sigqueue_t *sigqueue;
2267 	int prop;
2268 	struct sigacts *ps;
2269 	int intrval;
2270 	int ret = 0;
2271 	int wakeup_swapper;
2272 
2273 	MPASS(td == NULL || p == td->td_proc);
2274 	PROC_LOCK_ASSERT(p, MA_OWNED);
2275 
2276 	if (!_SIG_VALID(sig))
2277 		panic("%s(): invalid signal %d", __func__, sig);
2278 
2279 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2280 
2281 	/*
2282 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2283 	 */
2284 	if (p->p_state == PRS_ZOMBIE) {
2285 		if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2286 			ksiginfo_tryfree(ksi);
2287 		return (ret);
2288 	}
2289 
2290 	ps = p->p_sigacts;
2291 	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2292 	prop = sigprop(sig);
2293 
2294 	if (td == NULL) {
2295 		td = sigtd(p, sig, false);
2296 		sigqueue = &p->p_sigqueue;
2297 	} else
2298 		sigqueue = &td->td_sigqueue;
2299 
2300 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2301 
2302 	/*
2303 	 * If the signal is being ignored, then we forget about it
2304 	 * immediately, except when the target process executes
2305 	 * sigwait().  (Note: we don't set SIGCONT in ps_sigignore,
2306 	 * and if it is set to SIG_IGN, action will be SIG_DFL here.)
2307 	 */
2308 	mtx_lock(&ps->ps_mtx);
2309 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2310 		if (kern_sig_discard_ign &&
2311 		    (p->p_sysent->sv_flags & SV_SIG_DISCIGN) == 0) {
2312 			SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2313 
2314 			mtx_unlock(&ps->ps_mtx);
2315 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2316 				ksiginfo_tryfree(ksi);
2317 			return (ret);
2318 		} else {
2319 			action = SIG_CATCH;
2320 			intrval = 0;
2321 		}
2322 	} else {
2323 		if (SIGISMEMBER(td->td_sigmask, sig))
2324 			action = SIG_HOLD;
2325 		else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2326 			action = SIG_CATCH;
2327 		else
2328 			action = SIG_DFL;
2329 		if (SIGISMEMBER(ps->ps_sigintr, sig))
2330 			intrval = EINTR;
2331 		else
2332 			intrval = ERESTART;
2333 	}
2334 	mtx_unlock(&ps->ps_mtx);
2335 
2336 	if (prop & SIGPROP_CONT)
2337 		sigqueue_delete_stopmask_proc(p);
2338 	else if (prop & SIGPROP_STOP) {
2339 		/*
2340 		 * If sending a tty stop signal to a member of an orphaned
2341 		 * process group, discard the signal here if the action
2342 		 * is default; don't stop the process below if sleeping,
2343 		 * and don't clear any pending SIGCONT.
2344 		 */
2345 		if ((prop & SIGPROP_TTYSTOP) != 0 &&
2346 		    (p->p_pgrp->pg_flags & PGRP_ORPHANED) != 0 &&
2347 		    action == SIG_DFL) {
2348 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2349 				ksiginfo_tryfree(ksi);
2350 			return (ret);
2351 		}
2352 		sigqueue_delete_proc(p, SIGCONT);
2353 		if (p->p_flag & P_CONTINUED) {
2354 			p->p_flag &= ~P_CONTINUED;
2355 			PROC_LOCK(p->p_pptr);
2356 			sigqueue_take(p->p_ksi);
2357 			PROC_UNLOCK(p->p_pptr);
2358 		}
2359 	}
2360 
2361 	ret = sigqueue_add(sigqueue, sig, ksi);
2362 	if (ret != 0)
2363 		return (ret);
2364 	signotify(td);
2365 	/*
2366 	 * Defer further processing for signals which are held,
2367 	 * except that stopped processes must be continued by SIGCONT.
2368 	 */
2369 	if (action == SIG_HOLD &&
2370 	    !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2371 		return (ret);
2372 
2373 	wakeup_swapper = 0;
2374 
2375 	/*
2376 	 * Some signals have a process-wide effect and a per-thread
2377 	 * component.  Most processing occurs when the process next
2378 	 * tries to cross the user boundary, however there are some
2379 	 * times when processing needs to be done immediately, such as
2380 	 * waking up threads so that they can cross the user boundary.
2381 	 * We try to do the per-process part here.
2382 	 */
2383 	if (P_SHOULDSTOP(p)) {
2384 		KASSERT(!(p->p_flag & P_WEXIT),
2385 		    ("signal to stopped but exiting process"));
2386 		if (sig == SIGKILL) {
2387 			/*
2388 			 * If traced process is already stopped,
2389 			 * then no further action is necessary.
2390 			 */
2391 			if (p->p_flag & P_TRACED)
2392 				goto out;
2393 			/*
2394 			 * SIGKILL sets process running.
2395 			 * It will die elsewhere.
2396 			 * All threads must be restarted.
2397 			 */
2398 			p->p_flag &= ~P_STOPPED_SIG;
2399 			goto runfast;
2400 		}
2401 
2402 		if (prop & SIGPROP_CONT) {
2403 			/*
2404 			 * If traced process is already stopped,
2405 			 * then no further action is necessary.
2406 			 */
2407 			if (p->p_flag & P_TRACED)
2408 				goto out;
2409 			/*
2410 			 * If SIGCONT is default (or ignored), we continue the
2411 			 * process but don't leave the signal in sigqueue as
2412 			 * it has no further action.  If SIGCONT is held, we
2413 			 * continue the process and leave the signal in
2414 			 * sigqueue.  If the process catches SIGCONT, let it
2415 			 * handle the signal itself.  If it isn't waiting on
2416 			 * an event, it goes back to run state.
2417 			 * Otherwise, process goes back to sleep state.
2418 			 */
2419 			p->p_flag &= ~P_STOPPED_SIG;
2420 			PROC_SLOCK(p);
2421 			if (p->p_numthreads == p->p_suspcount) {
2422 				PROC_SUNLOCK(p);
2423 				p->p_flag |= P_CONTINUED;
2424 				p->p_xsig = SIGCONT;
2425 				PROC_LOCK(p->p_pptr);
2426 				childproc_continued(p);
2427 				PROC_UNLOCK(p->p_pptr);
2428 				PROC_SLOCK(p);
2429 			}
2430 			if (action == SIG_DFL) {
2431 				thread_unsuspend(p);
2432 				PROC_SUNLOCK(p);
2433 				sigqueue_delete(sigqueue, sig);
2434 				goto out_cont;
2435 			}
2436 			if (action == SIG_CATCH) {
2437 				/*
2438 				 * The process wants to catch it so it needs
2439 				 * to run at least one thread, but which one?
2440 				 */
2441 				PROC_SUNLOCK(p);
2442 				goto runfast;
2443 			}
2444 			/*
2445 			 * The signal is not ignored or caught.
2446 			 */
2447 			thread_unsuspend(p);
2448 			PROC_SUNLOCK(p);
2449 			goto out_cont;
2450 		}
2451 
2452 		if (prop & SIGPROP_STOP) {
2453 			/*
2454 			 * If traced process is already stopped,
2455 			 * then no further action is necessary.
2456 			 */
2457 			if (p->p_flag & P_TRACED)
2458 				goto out;
2459 			/*
2460 			 * Already stopped, don't need to stop again
2461 			 * (If we did the shell could get confused).
2462 			 * Just make sure the signal STOP bit set.
2463 			 */
2464 			p->p_flag |= P_STOPPED_SIG;
2465 			sigqueue_delete(sigqueue, sig);
2466 			goto out;
2467 		}
2468 
2469 		/*
2470 		 * All other kinds of signals:
2471 		 * If a thread is sleeping interruptibly, simulate a
2472 		 * wakeup so that when it is continued it will be made
2473 		 * runnable and can look at the signal.  However, don't make
2474 		 * the PROCESS runnable, leave it stopped.
2475 		 * It may run a bit until it hits a thread_suspend_check().
2476 		 */
2477 		PROC_SLOCK(p);
2478 		thread_lock(td);
2479 		if (TD_CAN_ABORT(td))
2480 			wakeup_swapper = sig_sleepq_abort(td, intrval);
2481 		else
2482 			thread_unlock(td);
2483 		PROC_SUNLOCK(p);
2484 		goto out;
2485 		/*
2486 		 * Mutexes are short lived. Threads waiting on them will
2487 		 * hit thread_suspend_check() soon.
2488 		 */
2489 	} else if (p->p_state == PRS_NORMAL) {
2490 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2491 			tdsigwakeup(td, sig, action, intrval);
2492 			goto out;
2493 		}
2494 
2495 		MPASS(action == SIG_DFL);
2496 
2497 		if (prop & SIGPROP_STOP) {
2498 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2499 				goto out;
2500 			p->p_flag |= P_STOPPED_SIG;
2501 			p->p_xsig = sig;
2502 			PROC_SLOCK(p);
2503 			wakeup_swapper = sig_suspend_threads(td, p);
2504 			if (p->p_numthreads == p->p_suspcount) {
2505 				/*
2506 				 * only thread sending signal to another
2507 				 * process can reach here, if thread is sending
2508 				 * signal to its process, because thread does
2509 				 * not suspend itself here, p_numthreads
2510 				 * should never be equal to p_suspcount.
2511 				 */
2512 				thread_stopped(p);
2513 				PROC_SUNLOCK(p);
2514 				sigqueue_delete_proc(p, p->p_xsig);
2515 			} else
2516 				PROC_SUNLOCK(p);
2517 			goto out;
2518 		}
2519 	} else {
2520 		/* Not in "NORMAL" state. discard the signal. */
2521 		sigqueue_delete(sigqueue, sig);
2522 		goto out;
2523 	}
2524 
2525 	/*
2526 	 * The process is not stopped so we need to apply the signal to all the
2527 	 * running threads.
2528 	 */
2529 runfast:
2530 	tdsigwakeup(td, sig, action, intrval);
2531 	PROC_SLOCK(p);
2532 	thread_unsuspend(p);
2533 	PROC_SUNLOCK(p);
2534 out_cont:
2535 	itimer_proc_continue(p);
2536 	kqtimer_proc_continue(p);
2537 out:
2538 	/* If we jump here, proc slock should not be owned. */
2539 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2540 	if (wakeup_swapper)
2541 		kick_proc0();
2542 
2543 	return (ret);
2544 }
2545 
2546 /*
2547  * The force of a signal has been directed against a single
2548  * thread.  We need to see what we can do about knocking it
2549  * out of any sleep it may be in etc.
2550  */
2551 static void
2552 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2553 {
2554 	struct proc *p = td->td_proc;
2555 	int prop, wakeup_swapper;
2556 
2557 	PROC_LOCK_ASSERT(p, MA_OWNED);
2558 	prop = sigprop(sig);
2559 
2560 	PROC_SLOCK(p);
2561 	thread_lock(td);
2562 	/*
2563 	 * Bring the priority of a thread up if we want it to get
2564 	 * killed in this lifetime.  Be careful to avoid bumping the
2565 	 * priority of the idle thread, since we still allow to signal
2566 	 * kernel processes.
2567 	 */
2568 	if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2569 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2570 		sched_prio(td, PUSER);
2571 	if (TD_ON_SLEEPQ(td)) {
2572 		/*
2573 		 * If thread is sleeping uninterruptibly
2574 		 * we can't interrupt the sleep... the signal will
2575 		 * be noticed when the process returns through
2576 		 * trap() or syscall().
2577 		 */
2578 		if ((td->td_flags & TDF_SINTR) == 0)
2579 			goto out;
2580 		/*
2581 		 * If SIGCONT is default (or ignored) and process is
2582 		 * asleep, we are finished; the process should not
2583 		 * be awakened.
2584 		 */
2585 		if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2586 			thread_unlock(td);
2587 			PROC_SUNLOCK(p);
2588 			sigqueue_delete(&p->p_sigqueue, sig);
2589 			/*
2590 			 * It may be on either list in this state.
2591 			 * Remove from both for now.
2592 			 */
2593 			sigqueue_delete(&td->td_sigqueue, sig);
2594 			return;
2595 		}
2596 
2597 		/*
2598 		 * Don't awaken a sleeping thread for SIGSTOP if the
2599 		 * STOP signal is deferred.
2600 		 */
2601 		if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2602 		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2603 			goto out;
2604 
2605 		/*
2606 		 * Give low priority threads a better chance to run.
2607 		 */
2608 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2609 			sched_prio(td, PUSER);
2610 
2611 		wakeup_swapper = sig_sleepq_abort(td, intrval);
2612 		PROC_SUNLOCK(p);
2613 		if (wakeup_swapper)
2614 			kick_proc0();
2615 		return;
2616 	}
2617 
2618 	/*
2619 	 * Other states do nothing with the signal immediately,
2620 	 * other than kicking ourselves if we are running.
2621 	 * It will either never be noticed, or noticed very soon.
2622 	 */
2623 #ifdef SMP
2624 	if (TD_IS_RUNNING(td) && td != curthread)
2625 		forward_signal(td);
2626 #endif
2627 
2628 out:
2629 	PROC_SUNLOCK(p);
2630 	thread_unlock(td);
2631 }
2632 
2633 static void
2634 ptrace_coredumpreq(struct thread *td, struct proc *p,
2635     struct thr_coredump_req *tcq)
2636 {
2637 	void *rl_cookie;
2638 
2639 	if (p->p_sysent->sv_coredump == NULL) {
2640 		tcq->tc_error = ENOSYS;
2641 		return;
2642 	}
2643 
2644 	rl_cookie = vn_rangelock_wlock(tcq->tc_vp, 0, OFF_MAX);
2645 	tcq->tc_error = p->p_sysent->sv_coredump(td, tcq->tc_vp,
2646 	    tcq->tc_limit, tcq->tc_flags);
2647 	vn_rangelock_unlock(tcq->tc_vp, rl_cookie);
2648 }
2649 
2650 static void
2651 ptrace_syscallreq(struct thread *td, struct proc *p,
2652     struct thr_syscall_req *tsr)
2653 {
2654 	struct sysentvec *sv;
2655 	struct sysent *se;
2656 	register_t rv_saved[2];
2657 	int error, nerror;
2658 	int sc;
2659 	bool audited, sy_thr_static;
2660 
2661 	sv = p->p_sysent;
2662 	if (sv->sv_table == NULL || sv->sv_size < tsr->ts_sa.code) {
2663 		tsr->ts_ret.sr_error = ENOSYS;
2664 		return;
2665 	}
2666 
2667 	sc = tsr->ts_sa.code;
2668 	if (sc == SYS_syscall || sc == SYS___syscall) {
2669 		sc = tsr->ts_sa.args[0];
2670 		memmove(&tsr->ts_sa.args[0], &tsr->ts_sa.args[1],
2671 		    sizeof(register_t) * (tsr->ts_nargs - 1));
2672 	}
2673 
2674 	tsr->ts_sa.callp = se = &sv->sv_table[sc];
2675 
2676 	VM_CNT_INC(v_syscall);
2677 	td->td_pticks = 0;
2678 	if (__predict_false(td->td_cowgen != atomic_load_int(
2679 	    &td->td_proc->p_cowgen)))
2680 		thread_cow_update(td);
2681 
2682 #ifdef CAPABILITY_MODE
2683 	if (IN_CAPABILITY_MODE(td) && (se->sy_flags & SYF_CAPENABLED) == 0) {
2684 		tsr->ts_ret.sr_error = ECAPMODE;
2685 		return;
2686 	}
2687 #endif
2688 
2689 	sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2690 	audited = AUDIT_SYSCALL_ENTER(sc, td) != 0;
2691 
2692 	if (!sy_thr_static) {
2693 		error = syscall_thread_enter(td, &se);
2694 		sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2695 		if (error != 0) {
2696 			tsr->ts_ret.sr_error = error;
2697 			return;
2698 		}
2699 	}
2700 
2701 	rv_saved[0] = td->td_retval[0];
2702 	rv_saved[1] = td->td_retval[1];
2703 	nerror = td->td_errno;
2704 	td->td_retval[0] = 0;
2705 	td->td_retval[1] = 0;
2706 
2707 #ifdef KDTRACE_HOOKS
2708 	if (se->sy_entry != 0)
2709 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_ENTRY, 0);
2710 #endif
2711 	tsr->ts_ret.sr_error = se->sy_call(td, tsr->ts_sa.args);
2712 #ifdef KDTRACE_HOOKS
2713 	if (se->sy_return != 0)
2714 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_RETURN,
2715 		    tsr->ts_ret.sr_error != 0 ? -1 : td->td_retval[0]);
2716 #endif
2717 
2718 	tsr->ts_ret.sr_retval[0] = td->td_retval[0];
2719 	tsr->ts_ret.sr_retval[1] = td->td_retval[1];
2720 	td->td_retval[0] = rv_saved[0];
2721 	td->td_retval[1] = rv_saved[1];
2722 	td->td_errno = nerror;
2723 
2724 	if (audited)
2725 		AUDIT_SYSCALL_EXIT(error, td);
2726 	if (!sy_thr_static)
2727 		syscall_thread_exit(td, se);
2728 }
2729 
2730 static void
2731 ptrace_remotereq(struct thread *td, int flag)
2732 {
2733 	struct proc *p;
2734 
2735 	MPASS(td == curthread);
2736 	p = td->td_proc;
2737 	PROC_LOCK_ASSERT(p, MA_OWNED);
2738 	if ((td->td_dbgflags & flag) == 0)
2739 		return;
2740 	KASSERT((p->p_flag & P_STOPPED_TRACE) != 0, ("not stopped"));
2741 	KASSERT(td->td_remotereq != NULL, ("td_remotereq is NULL"));
2742 
2743 	PROC_UNLOCK(p);
2744 	switch (flag) {
2745 	case TDB_COREDUMPREQ:
2746 		ptrace_coredumpreq(td, p, td->td_remotereq);
2747 		break;
2748 	case TDB_SCREMOTEREQ:
2749 		ptrace_syscallreq(td, p, td->td_remotereq);
2750 		break;
2751 	default:
2752 		__unreachable();
2753 	}
2754 	PROC_LOCK(p);
2755 
2756 	MPASS((td->td_dbgflags & flag) != 0);
2757 	td->td_dbgflags &= ~flag;
2758 	td->td_remotereq = NULL;
2759 	wakeup(p);
2760 }
2761 
2762 static int
2763 sig_suspend_threads(struct thread *td, struct proc *p)
2764 {
2765 	struct thread *td2;
2766 	int wakeup_swapper;
2767 
2768 	PROC_LOCK_ASSERT(p, MA_OWNED);
2769 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2770 
2771 	wakeup_swapper = 0;
2772 	FOREACH_THREAD_IN_PROC(p, td2) {
2773 		thread_lock(td2);
2774 		ast_sched_locked(td2, TDA_SUSPEND);
2775 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2776 		    (td2->td_flags & TDF_SINTR)) {
2777 			if (td2->td_flags & TDF_SBDRY) {
2778 				/*
2779 				 * Once a thread is asleep with
2780 				 * TDF_SBDRY and without TDF_SERESTART
2781 				 * or TDF_SEINTR set, it should never
2782 				 * become suspended due to this check.
2783 				 */
2784 				KASSERT(!TD_IS_SUSPENDED(td2),
2785 				    ("thread with deferred stops suspended"));
2786 				if (TD_SBDRY_INTR(td2)) {
2787 					wakeup_swapper |= sleepq_abort(td2,
2788 					    TD_SBDRY_ERRNO(td2));
2789 					continue;
2790 				}
2791 			} else if (!TD_IS_SUSPENDED(td2))
2792 				thread_suspend_one(td2);
2793 		} else if (!TD_IS_SUSPENDED(td2)) {
2794 #ifdef SMP
2795 			if (TD_IS_RUNNING(td2) && td2 != td)
2796 				forward_signal(td2);
2797 #endif
2798 		}
2799 		thread_unlock(td2);
2800 	}
2801 	return (wakeup_swapper);
2802 }
2803 
2804 /*
2805  * Stop the process for an event deemed interesting to the debugger. If si is
2806  * non-NULL, this is a signal exchange; the new signal requested by the
2807  * debugger will be returned for handling. If si is NULL, this is some other
2808  * type of interesting event. The debugger may request a signal be delivered in
2809  * that case as well, however it will be deferred until it can be handled.
2810  */
2811 int
2812 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2813 {
2814 	struct proc *p = td->td_proc;
2815 	struct thread *td2;
2816 	ksiginfo_t ksi;
2817 
2818 	PROC_LOCK_ASSERT(p, MA_OWNED);
2819 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2820 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2821 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2822 
2823 	td->td_xsig = sig;
2824 
2825 	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2826 		td->td_dbgflags |= TDB_XSIG;
2827 		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2828 		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2829 		PROC_SLOCK(p);
2830 		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2831 			if (P_KILLED(p)) {
2832 				/*
2833 				 * Ensure that, if we've been PT_KILLed, the
2834 				 * exit status reflects that. Another thread
2835 				 * may also be in ptracestop(), having just
2836 				 * received the SIGKILL, but this thread was
2837 				 * unsuspended first.
2838 				 */
2839 				td->td_dbgflags &= ~TDB_XSIG;
2840 				td->td_xsig = SIGKILL;
2841 				p->p_ptevents = 0;
2842 				break;
2843 			}
2844 			if (p->p_flag & P_SINGLE_EXIT &&
2845 			    !(td->td_dbgflags & TDB_EXIT)) {
2846 				/*
2847 				 * Ignore ptrace stops except for thread exit
2848 				 * events when the process exits.
2849 				 */
2850 				td->td_dbgflags &= ~TDB_XSIG;
2851 				PROC_SUNLOCK(p);
2852 				return (0);
2853 			}
2854 
2855 			/*
2856 			 * Make wait(2) work.  Ensure that right after the
2857 			 * attach, the thread which was decided to become the
2858 			 * leader of attach gets reported to the waiter.
2859 			 * Otherwise, just avoid overwriting another thread's
2860 			 * assignment to p_xthread.  If another thread has
2861 			 * already set p_xthread, the current thread will get
2862 			 * a chance to report itself upon the next iteration.
2863 			 */
2864 			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2865 			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2866 			    p->p_xthread == NULL)) {
2867 				p->p_xsig = sig;
2868 				p->p_xthread = td;
2869 
2870 				/*
2871 				 * If we are on sleepqueue already,
2872 				 * let sleepqueue code decide if it
2873 				 * needs to go sleep after attach.
2874 				 */
2875 				if (td->td_wchan == NULL)
2876 					td->td_dbgflags &= ~TDB_FSTP;
2877 
2878 				p->p_flag2 &= ~P2_PTRACE_FSTP;
2879 				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2880 				sig_suspend_threads(td, p);
2881 			}
2882 			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2883 				td->td_dbgflags &= ~TDB_STOPATFORK;
2884 			}
2885 stopme:
2886 			td->td_dbgflags |= TDB_SSWITCH;
2887 			thread_suspend_switch(td, p);
2888 			td->td_dbgflags &= ~TDB_SSWITCH;
2889 			if ((td->td_dbgflags & (TDB_COREDUMPREQ |
2890 			    TDB_SCREMOTEREQ)) != 0) {
2891 				MPASS((td->td_dbgflags & (TDB_COREDUMPREQ |
2892 				    TDB_SCREMOTEREQ)) !=
2893 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2894 				PROC_SUNLOCK(p);
2895 				ptrace_remotereq(td, td->td_dbgflags &
2896 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2897 				PROC_SLOCK(p);
2898 				goto stopme;
2899 			}
2900 			if (p->p_xthread == td)
2901 				p->p_xthread = NULL;
2902 			if (!(p->p_flag & P_TRACED))
2903 				break;
2904 			if (td->td_dbgflags & TDB_SUSPEND) {
2905 				if (p->p_flag & P_SINGLE_EXIT)
2906 					break;
2907 				goto stopme;
2908 			}
2909 		}
2910 		PROC_SUNLOCK(p);
2911 	}
2912 
2913 	if (si != NULL && sig == td->td_xsig) {
2914 		/* Parent wants us to take the original signal unchanged. */
2915 		si->ksi_flags |= KSI_HEAD;
2916 		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2917 			si->ksi_signo = 0;
2918 	} else if (td->td_xsig != 0) {
2919 		/*
2920 		 * If parent wants us to take a new signal, then it will leave
2921 		 * it in td->td_xsig; otherwise we just look for signals again.
2922 		 */
2923 		ksiginfo_init(&ksi);
2924 		ksi.ksi_signo = td->td_xsig;
2925 		ksi.ksi_flags |= KSI_PTRACE;
2926 		td2 = sigtd(p, td->td_xsig, false);
2927 		tdsendsignal(p, td2, td->td_xsig, &ksi);
2928 		if (td != td2)
2929 			return (0);
2930 	}
2931 
2932 	return (td->td_xsig);
2933 }
2934 
2935 static void
2936 reschedule_signals(struct proc *p, sigset_t block, int flags)
2937 {
2938 	struct sigacts *ps;
2939 	struct thread *td;
2940 	int sig;
2941 	bool fastblk, pslocked;
2942 
2943 	PROC_LOCK_ASSERT(p, MA_OWNED);
2944 	ps = p->p_sigacts;
2945 	pslocked = (flags & SIGPROCMASK_PS_LOCKED) != 0;
2946 	mtx_assert(&ps->ps_mtx, pslocked ? MA_OWNED : MA_NOTOWNED);
2947 	if (SIGISEMPTY(p->p_siglist))
2948 		return;
2949 	SIGSETAND(block, p->p_siglist);
2950 	fastblk = (flags & SIGPROCMASK_FASTBLK) != 0;
2951 	SIG_FOREACH(sig, &block) {
2952 		td = sigtd(p, sig, fastblk);
2953 
2954 		/*
2955 		 * If sigtd() selected us despite sigfastblock is
2956 		 * blocking, do not activate AST or wake us, to avoid
2957 		 * loop in AST handler.
2958 		 */
2959 		if (fastblk && td == curthread)
2960 			continue;
2961 
2962 		signotify(td);
2963 		if (!pslocked)
2964 			mtx_lock(&ps->ps_mtx);
2965 		if (p->p_flag & P_TRACED ||
2966 		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2967 		    !SIGISMEMBER(td->td_sigmask, sig))) {
2968 			tdsigwakeup(td, sig, SIG_CATCH,
2969 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2970 			    ERESTART));
2971 		}
2972 		if (!pslocked)
2973 			mtx_unlock(&ps->ps_mtx);
2974 	}
2975 }
2976 
2977 void
2978 tdsigcleanup(struct thread *td)
2979 {
2980 	struct proc *p;
2981 	sigset_t unblocked;
2982 
2983 	p = td->td_proc;
2984 	PROC_LOCK_ASSERT(p, MA_OWNED);
2985 
2986 	sigqueue_flush(&td->td_sigqueue);
2987 	if (p->p_numthreads == 1)
2988 		return;
2989 
2990 	/*
2991 	 * Since we cannot handle signals, notify signal post code
2992 	 * about this by filling the sigmask.
2993 	 *
2994 	 * Also, if needed, wake up thread(s) that do not block the
2995 	 * same signals as the exiting thread, since the thread might
2996 	 * have been selected for delivery and woken up.
2997 	 */
2998 	SIGFILLSET(unblocked);
2999 	SIGSETNAND(unblocked, td->td_sigmask);
3000 	SIGFILLSET(td->td_sigmask);
3001 	reschedule_signals(p, unblocked, 0);
3002 
3003 }
3004 
3005 static int
3006 sigdeferstop_curr_flags(int cflags)
3007 {
3008 
3009 	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
3010 	    (cflags & TDF_SBDRY) != 0);
3011 	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
3012 }
3013 
3014 /*
3015  * Defer the delivery of SIGSTOP for the current thread, according to
3016  * the requested mode.  Returns previous flags, which must be restored
3017  * by sigallowstop().
3018  *
3019  * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
3020  * cleared by the current thread, which allow the lock-less read-only
3021  * accesses below.
3022  */
3023 int
3024 sigdeferstop_impl(int mode)
3025 {
3026 	struct thread *td;
3027 	int cflags, nflags;
3028 
3029 	td = curthread;
3030 	cflags = sigdeferstop_curr_flags(td->td_flags);
3031 	switch (mode) {
3032 	case SIGDEFERSTOP_NOP:
3033 		nflags = cflags;
3034 		break;
3035 	case SIGDEFERSTOP_OFF:
3036 		nflags = 0;
3037 		break;
3038 	case SIGDEFERSTOP_SILENT:
3039 		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
3040 		break;
3041 	case SIGDEFERSTOP_EINTR:
3042 		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
3043 		break;
3044 	case SIGDEFERSTOP_ERESTART:
3045 		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
3046 		break;
3047 	default:
3048 		panic("sigdeferstop: invalid mode %x", mode);
3049 		break;
3050 	}
3051 	if (cflags == nflags)
3052 		return (SIGDEFERSTOP_VAL_NCHG);
3053 	thread_lock(td);
3054 	td->td_flags = (td->td_flags & ~cflags) | nflags;
3055 	thread_unlock(td);
3056 	return (cflags);
3057 }
3058 
3059 /*
3060  * Restores the STOP handling mode, typically permitting the delivery
3061  * of SIGSTOP for the current thread.  This does not immediately
3062  * suspend if a stop was posted.  Instead, the thread will suspend
3063  * either via ast() or a subsequent interruptible sleep.
3064  */
3065 void
3066 sigallowstop_impl(int prev)
3067 {
3068 	struct thread *td;
3069 	int cflags;
3070 
3071 	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
3072 	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
3073 	    ("sigallowstop: incorrect previous mode %x", prev));
3074 	td = curthread;
3075 	cflags = sigdeferstop_curr_flags(td->td_flags);
3076 	if (cflags != prev) {
3077 		thread_lock(td);
3078 		td->td_flags = (td->td_flags & ~cflags) | prev;
3079 		thread_unlock(td);
3080 	}
3081 }
3082 
3083 enum sigstatus {
3084 	SIGSTATUS_HANDLE,
3085 	SIGSTATUS_HANDLED,
3086 	SIGSTATUS_IGNORE,
3087 	SIGSTATUS_SBDRY_STOP,
3088 };
3089 
3090 /*
3091  * The thread has signal "sig" pending.  Figure out what to do with it:
3092  *
3093  * _HANDLE     -> the caller should handle the signal
3094  * _HANDLED    -> handled internally, reload pending signal set
3095  * _IGNORE     -> ignored, remove from the set of pending signals and try the
3096  *                next pending signal
3097  * _SBDRY_STOP -> the signal should stop the thread but this is not
3098  *                permitted in the current context
3099  */
3100 static enum sigstatus
3101 sigprocess(struct thread *td, int sig)
3102 {
3103 	struct proc *p;
3104 	struct sigacts *ps;
3105 	struct sigqueue *queue;
3106 	ksiginfo_t ksi;
3107 	int prop;
3108 
3109 	KASSERT(_SIG_VALID(sig), ("%s: invalid signal %d", __func__, sig));
3110 
3111 	p = td->td_proc;
3112 	ps = p->p_sigacts;
3113 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3114 	PROC_LOCK_ASSERT(p, MA_OWNED);
3115 
3116 	/*
3117 	 * We should allow pending but ignored signals below
3118 	 * if there is sigwait() active, or P_TRACED was
3119 	 * on when they were posted.
3120 	 */
3121 	if (SIGISMEMBER(ps->ps_sigignore, sig) &&
3122 	    (p->p_flag & P_TRACED) == 0 &&
3123 	    (td->td_flags & TDF_SIGWAIT) == 0) {
3124 		return (SIGSTATUS_IGNORE);
3125 	}
3126 
3127 	/*
3128 	 * If the process is going to single-thread mode to prepare
3129 	 * for exit, there is no sense in delivering any signal
3130 	 * to usermode.  Another important consequence is that
3131 	 * msleep(..., PCATCH, ...) now is only interruptible by a
3132 	 * suspend request.
3133 	 */
3134 	if ((p->p_flag2 & P2_WEXIT) != 0)
3135 		return (SIGSTATUS_IGNORE);
3136 
3137 	if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
3138 		/*
3139 		 * If traced, always stop.
3140 		 * Remove old signal from queue before the stop.
3141 		 * XXX shrug off debugger, it causes siginfo to
3142 		 * be thrown away.
3143 		 */
3144 		queue = &td->td_sigqueue;
3145 		ksiginfo_init(&ksi);
3146 		if (sigqueue_get(queue, sig, &ksi) == 0) {
3147 			queue = &p->p_sigqueue;
3148 			sigqueue_get(queue, sig, &ksi);
3149 		}
3150 		td->td_si = ksi.ksi_info;
3151 
3152 		mtx_unlock(&ps->ps_mtx);
3153 		sig = ptracestop(td, sig, &ksi);
3154 		mtx_lock(&ps->ps_mtx);
3155 
3156 		td->td_si.si_signo = 0;
3157 
3158 		/*
3159 		 * Keep looking if the debugger discarded or
3160 		 * replaced the signal.
3161 		 */
3162 		if (sig == 0)
3163 			return (SIGSTATUS_HANDLED);
3164 
3165 		/*
3166 		 * If the signal became masked, re-queue it.
3167 		 */
3168 		if (SIGISMEMBER(td->td_sigmask, sig)) {
3169 			ksi.ksi_flags |= KSI_HEAD;
3170 			sigqueue_add(&p->p_sigqueue, sig, &ksi);
3171 			return (SIGSTATUS_HANDLED);
3172 		}
3173 
3174 		/*
3175 		 * If the traced bit got turned off, requeue the signal and
3176 		 * reload the set of pending signals.  This ensures that p_sig*
3177 		 * and p_sigact are consistent.
3178 		 */
3179 		if ((p->p_flag & P_TRACED) == 0) {
3180 			if ((ksi.ksi_flags & KSI_PTRACE) == 0) {
3181 				ksi.ksi_flags |= KSI_HEAD;
3182 				sigqueue_add(queue, sig, &ksi);
3183 			}
3184 			return (SIGSTATUS_HANDLED);
3185 		}
3186 	}
3187 
3188 	/*
3189 	 * Decide whether the signal should be returned.
3190 	 * Return the signal's number, or fall through
3191 	 * to clear it from the pending mask.
3192 	 */
3193 	switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
3194 	case (intptr_t)SIG_DFL:
3195 		/*
3196 		 * Don't take default actions on system processes.
3197 		 */
3198 		if (p->p_pid <= 1) {
3199 #ifdef DIAGNOSTIC
3200 			/*
3201 			 * Are you sure you want to ignore SIGSEGV
3202 			 * in init? XXX
3203 			 */
3204 			printf("Process (pid %lu) got signal %d\n",
3205 				(u_long)p->p_pid, sig);
3206 #endif
3207 			return (SIGSTATUS_IGNORE);
3208 		}
3209 
3210 		/*
3211 		 * If there is a pending stop signal to process with
3212 		 * default action, stop here, then clear the signal.
3213 		 * Traced or exiting processes should ignore stops.
3214 		 * Additionally, a member of an orphaned process group
3215 		 * should ignore tty stops.
3216 		 */
3217 		prop = sigprop(sig);
3218 		if (prop & SIGPROP_STOP) {
3219 			mtx_unlock(&ps->ps_mtx);
3220 			if ((p->p_flag & (P_TRACED | P_WEXIT |
3221 			    P_SINGLE_EXIT)) != 0 || ((p->p_pgrp->
3222 			    pg_flags & PGRP_ORPHANED) != 0 &&
3223 			    (prop & SIGPROP_TTYSTOP) != 0)) {
3224 				mtx_lock(&ps->ps_mtx);
3225 				return (SIGSTATUS_IGNORE);
3226 			}
3227 			if (TD_SBDRY_INTR(td)) {
3228 				KASSERT((td->td_flags & TDF_SBDRY) != 0,
3229 				    ("lost TDF_SBDRY"));
3230 				mtx_lock(&ps->ps_mtx);
3231 				return (SIGSTATUS_SBDRY_STOP);
3232 			}
3233 			WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
3234 			    &p->p_mtx.lock_object, "Catching SIGSTOP");
3235 			sigqueue_delete(&td->td_sigqueue, sig);
3236 			sigqueue_delete(&p->p_sigqueue, sig);
3237 			p->p_flag |= P_STOPPED_SIG;
3238 			p->p_xsig = sig;
3239 			PROC_SLOCK(p);
3240 			sig_suspend_threads(td, p);
3241 			thread_suspend_switch(td, p);
3242 			PROC_SUNLOCK(p);
3243 			mtx_lock(&ps->ps_mtx);
3244 			return (SIGSTATUS_HANDLED);
3245 		} else if ((prop & SIGPROP_IGNORE) != 0 &&
3246 		    (td->td_flags & TDF_SIGWAIT) == 0) {
3247 			/*
3248 			 * Default action is to ignore; drop it if
3249 			 * not in kern_sigtimedwait().
3250 			 */
3251 			return (SIGSTATUS_IGNORE);
3252 		} else {
3253 			return (SIGSTATUS_HANDLE);
3254 		}
3255 
3256 	case (intptr_t)SIG_IGN:
3257 		if ((td->td_flags & TDF_SIGWAIT) == 0)
3258 			return (SIGSTATUS_IGNORE);
3259 		else
3260 			return (SIGSTATUS_HANDLE);
3261 
3262 	default:
3263 		/*
3264 		 * This signal has an action, let postsig() process it.
3265 		 */
3266 		return (SIGSTATUS_HANDLE);
3267 	}
3268 }
3269 
3270 /*
3271  * If the current process has received a signal (should be caught or cause
3272  * termination, should interrupt current syscall), return the signal number.
3273  * Stop signals with default action are processed immediately, then cleared;
3274  * they aren't returned.  This is checked after each entry to the system for
3275  * a syscall or trap (though this can usually be done without calling
3276  * issignal by checking the pending signal masks in cursig.) The normal call
3277  * sequence is
3278  *
3279  *	while (sig = cursig(curthread))
3280  *		postsig(sig);
3281  */
3282 static int
3283 issignal(struct thread *td)
3284 {
3285 	struct proc *p;
3286 	sigset_t sigpending;
3287 	int sig;
3288 
3289 	p = td->td_proc;
3290 	PROC_LOCK_ASSERT(p, MA_OWNED);
3291 
3292 	for (;;) {
3293 		sigpending = td->td_sigqueue.sq_signals;
3294 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
3295 		SIGSETNAND(sigpending, td->td_sigmask);
3296 
3297 		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
3298 		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
3299 			SIG_STOPSIGMASK(sigpending);
3300 		if (SIGISEMPTY(sigpending))	/* no signal to send */
3301 			return (0);
3302 
3303 		/*
3304 		 * Do fast sigblock if requested by usermode.  Since
3305 		 * we do know that there was a signal pending at this
3306 		 * point, set the FAST_SIGBLOCK_PEND as indicator for
3307 		 * usermode to perform a dummy call to
3308 		 * FAST_SIGBLOCK_UNBLOCK, which causes immediate
3309 		 * delivery of postponed pending signal.
3310 		 */
3311 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
3312 			if (td->td_sigblock_val != 0)
3313 				SIGSETNAND(sigpending, fastblock_mask);
3314 			if (SIGISEMPTY(sigpending)) {
3315 				td->td_pflags |= TDP_SIGFASTPENDING;
3316 				return (0);
3317 			}
3318 		}
3319 
3320 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
3321 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
3322 		    SIGISMEMBER(sigpending, SIGSTOP)) {
3323 			/*
3324 			 * If debugger just attached, always consume
3325 			 * SIGSTOP from ptrace(PT_ATTACH) first, to
3326 			 * execute the debugger attach ritual in
3327 			 * order.
3328 			 */
3329 			td->td_dbgflags |= TDB_FSTP;
3330 			SIGEMPTYSET(sigpending);
3331 			SIGADDSET(sigpending, SIGSTOP);
3332 		}
3333 
3334 		SIG_FOREACH(sig, &sigpending) {
3335 			switch (sigprocess(td, sig)) {
3336 			case SIGSTATUS_HANDLE:
3337 				return (sig);
3338 			case SIGSTATUS_HANDLED:
3339 				goto next;
3340 			case SIGSTATUS_IGNORE:
3341 				sigqueue_delete(&td->td_sigqueue, sig);
3342 				sigqueue_delete(&p->p_sigqueue, sig);
3343 				break;
3344 			case SIGSTATUS_SBDRY_STOP:
3345 				return (-1);
3346 			}
3347 		}
3348 next:;
3349 	}
3350 }
3351 
3352 void
3353 thread_stopped(struct proc *p)
3354 {
3355 	int n;
3356 
3357 	PROC_LOCK_ASSERT(p, MA_OWNED);
3358 	PROC_SLOCK_ASSERT(p, MA_OWNED);
3359 	n = p->p_suspcount;
3360 	if (p == curproc)
3361 		n++;
3362 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3363 		PROC_SUNLOCK(p);
3364 		p->p_flag &= ~P_WAITED;
3365 		PROC_LOCK(p->p_pptr);
3366 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
3367 			CLD_TRAPPED : CLD_STOPPED);
3368 		PROC_UNLOCK(p->p_pptr);
3369 		PROC_SLOCK(p);
3370 	}
3371 }
3372 
3373 /*
3374  * Take the action for the specified signal
3375  * from the current set of pending signals.
3376  */
3377 int
3378 postsig(int sig)
3379 {
3380 	struct thread *td;
3381 	struct proc *p;
3382 	struct sigacts *ps;
3383 	sig_t action;
3384 	ksiginfo_t ksi;
3385 	sigset_t returnmask;
3386 
3387 	KASSERT(sig != 0, ("postsig"));
3388 
3389 	td = curthread;
3390 	p = td->td_proc;
3391 	PROC_LOCK_ASSERT(p, MA_OWNED);
3392 	ps = p->p_sigacts;
3393 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3394 	ksiginfo_init(&ksi);
3395 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3396 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3397 		return (0);
3398 	ksi.ksi_signo = sig;
3399 	if (ksi.ksi_code == SI_TIMER)
3400 		itimer_accept(p, ksi.ksi_timerid, &ksi);
3401 	action = ps->ps_sigact[_SIG_IDX(sig)];
3402 #ifdef KTRACE
3403 	if (KTRPOINT(td, KTR_PSIG))
3404 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3405 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3406 #endif
3407 
3408 	if (action == SIG_DFL) {
3409 		/*
3410 		 * Default action, where the default is to kill
3411 		 * the process.  (Other cases were ignored above.)
3412 		 */
3413 		mtx_unlock(&ps->ps_mtx);
3414 		proc_td_siginfo_capture(td, &ksi.ksi_info);
3415 		sigexit(td, sig);
3416 		/* NOTREACHED */
3417 	} else {
3418 		/*
3419 		 * If we get here, the signal must be caught.
3420 		 */
3421 		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3422 		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3423 		    ("postsig action: blocked sig %d", sig));
3424 
3425 		/*
3426 		 * Set the new mask value and also defer further
3427 		 * occurrences of this signal.
3428 		 *
3429 		 * Special case: user has done a sigsuspend.  Here the
3430 		 * current mask is not of interest, but rather the
3431 		 * mask from before the sigsuspend is what we want
3432 		 * restored after the signal processing is completed.
3433 		 */
3434 		if (td->td_pflags & TDP_OLDMASK) {
3435 			returnmask = td->td_oldsigmask;
3436 			td->td_pflags &= ~TDP_OLDMASK;
3437 		} else
3438 			returnmask = td->td_sigmask;
3439 
3440 		if (p->p_sig == sig) {
3441 			p->p_sig = 0;
3442 		}
3443 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3444 		postsig_done(sig, td, ps);
3445 	}
3446 	return (1);
3447 }
3448 
3449 int
3450 sig_ast_checksusp(struct thread *td)
3451 {
3452 	struct proc *p __diagused;
3453 	int ret;
3454 
3455 	p = td->td_proc;
3456 	PROC_LOCK_ASSERT(p, MA_OWNED);
3457 
3458 	if (!td_ast_pending(td, TDA_SUSPEND))
3459 		return (0);
3460 
3461 	ret = thread_suspend_check(1);
3462 	MPASS(ret == 0 || ret == EINTR || ret == ERESTART);
3463 	return (ret);
3464 }
3465 
3466 int
3467 sig_ast_needsigchk(struct thread *td)
3468 {
3469 	struct proc *p;
3470 	struct sigacts *ps;
3471 	int ret, sig;
3472 
3473 	p = td->td_proc;
3474 	PROC_LOCK_ASSERT(p, MA_OWNED);
3475 
3476 	if (!td_ast_pending(td, TDA_SIG))
3477 		return (0);
3478 
3479 	ps = p->p_sigacts;
3480 	mtx_lock(&ps->ps_mtx);
3481 	sig = cursig(td);
3482 	if (sig == -1) {
3483 		mtx_unlock(&ps->ps_mtx);
3484 		KASSERT((td->td_flags & TDF_SBDRY) != 0, ("lost TDF_SBDRY"));
3485 		KASSERT(TD_SBDRY_INTR(td),
3486 		    ("lost TDF_SERESTART of TDF_SEINTR"));
3487 		KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
3488 		    (TDF_SEINTR | TDF_SERESTART),
3489 		    ("both TDF_SEINTR and TDF_SERESTART"));
3490 		ret = TD_SBDRY_ERRNO(td);
3491 	} else if (sig != 0) {
3492 		ret = SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR : ERESTART;
3493 		mtx_unlock(&ps->ps_mtx);
3494 	} else {
3495 		mtx_unlock(&ps->ps_mtx);
3496 		ret = 0;
3497 	}
3498 
3499 	/*
3500 	 * Do not go into sleep if this thread was the ptrace(2)
3501 	 * attach leader.  cursig() consumed SIGSTOP from PT_ATTACH,
3502 	 * but we usually act on the signal by interrupting sleep, and
3503 	 * should do that here as well.
3504 	 */
3505 	if ((td->td_dbgflags & TDB_FSTP) != 0) {
3506 		if (ret == 0)
3507 			ret = EINTR;
3508 		td->td_dbgflags &= ~TDB_FSTP;
3509 	}
3510 
3511 	return (ret);
3512 }
3513 
3514 int
3515 sig_intr(void)
3516 {
3517 	struct thread *td;
3518 	struct proc *p;
3519 	int ret;
3520 
3521 	td = curthread;
3522 	if (!td_ast_pending(td, TDA_SIG) && !td_ast_pending(td, TDA_SUSPEND))
3523 		return (0);
3524 
3525 	p = td->td_proc;
3526 
3527 	PROC_LOCK(p);
3528 	ret = sig_ast_checksusp(td);
3529 	if (ret == 0)
3530 		ret = sig_ast_needsigchk(td);
3531 	PROC_UNLOCK(p);
3532 	return (ret);
3533 }
3534 
3535 bool
3536 curproc_sigkilled(void)
3537 {
3538 	struct thread *td;
3539 	struct proc *p;
3540 	struct sigacts *ps;
3541 	bool res;
3542 
3543 	td = curthread;
3544 	if (!td_ast_pending(td, TDA_SIG))
3545 		return (false);
3546 
3547 	p = td->td_proc;
3548 	PROC_LOCK(p);
3549 	ps = p->p_sigacts;
3550 	mtx_lock(&ps->ps_mtx);
3551 	res = SIGISMEMBER(td->td_sigqueue.sq_signals, SIGKILL) ||
3552 	    SIGISMEMBER(p->p_sigqueue.sq_signals, SIGKILL);
3553 	mtx_unlock(&ps->ps_mtx);
3554 	PROC_UNLOCK(p);
3555 	return (res);
3556 }
3557 
3558 void
3559 proc_wkilled(struct proc *p)
3560 {
3561 
3562 	PROC_LOCK_ASSERT(p, MA_OWNED);
3563 	if ((p->p_flag & P_WKILLED) == 0) {
3564 		p->p_flag |= P_WKILLED;
3565 		/*
3566 		 * Notify swapper that there is a process to swap in.
3567 		 * The notification is racy, at worst it would take 10
3568 		 * seconds for the swapper process to notice.
3569 		 */
3570 		if ((p->p_flag & (P_INMEM | P_SWAPPINGIN)) == 0)
3571 			wakeup(&proc0);
3572 	}
3573 }
3574 
3575 /*
3576  * Kill the current process for stated reason.
3577  */
3578 void
3579 killproc(struct proc *p, const char *why)
3580 {
3581 
3582 	PROC_LOCK_ASSERT(p, MA_OWNED);
3583 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3584 	    p->p_comm);
3585 	log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3586 	    p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3587 	    p->p_ucred->cr_uid, why);
3588 	proc_wkilled(p);
3589 	kern_psignal(p, SIGKILL);
3590 }
3591 
3592 /*
3593  * Force the current process to exit with the specified signal, dumping core
3594  * if appropriate.  We bypass the normal tests for masked and caught signals,
3595  * allowing unrecoverable failures to terminate the process without changing
3596  * signal state.  Mark the accounting record with the signal termination.
3597  * If dumping core, save the signal number for the debugger.  Calls exit and
3598  * does not return.
3599  */
3600 void
3601 sigexit(struct thread *td, int sig)
3602 {
3603 	struct proc *p = td->td_proc;
3604 	const char *coreinfo;
3605 	int rv;
3606 
3607 	PROC_LOCK_ASSERT(p, MA_OWNED);
3608 	proc_set_p2_wexit(p);
3609 
3610 	p->p_acflag |= AXSIG;
3611 	/*
3612 	 * We must be single-threading to generate a core dump.  This
3613 	 * ensures that the registers in the core file are up-to-date.
3614 	 * Also, the ELF dump handler assumes that the thread list doesn't
3615 	 * change out from under it.
3616 	 *
3617 	 * XXX If another thread attempts to single-thread before us
3618 	 *     (e.g. via fork()), we won't get a dump at all.
3619 	 */
3620 	if ((sigprop(sig) & SIGPROP_CORE) &&
3621 	    thread_single(p, SINGLE_NO_EXIT) == 0) {
3622 		p->p_sig = sig;
3623 		/*
3624 		 * Log signals which would cause core dumps
3625 		 * (Log as LOG_INFO to appease those who don't want
3626 		 * these messages.)
3627 		 * XXX : Todo, as well as euid, write out ruid too
3628 		 * Note that coredump() drops proc lock.
3629 		 */
3630 		rv = coredump(td);
3631 		switch (rv) {
3632 		case 0:
3633 			sig |= WCOREFLAG;
3634 			coreinfo = " (core dumped)";
3635 			break;
3636 		case EFAULT:
3637 			coreinfo = " (no core dump - bad address)";
3638 			break;
3639 		case EINVAL:
3640 			coreinfo = " (no core dump - invalid argument)";
3641 			break;
3642 		case EFBIG:
3643 			coreinfo = " (no core dump - too large)";
3644 			break;
3645 		default:
3646 			coreinfo = " (no core dump - other error)";
3647 			break;
3648 		}
3649 		if (kern_logsigexit)
3650 			log(LOG_INFO,
3651 			    "pid %d (%s), jid %d, uid %d: exited on "
3652 			    "signal %d%s\n", p->p_pid, p->p_comm,
3653 			    p->p_ucred->cr_prison->pr_id,
3654 			    td->td_ucred->cr_uid,
3655 			    sig &~ WCOREFLAG, coreinfo);
3656 	} else
3657 		PROC_UNLOCK(p);
3658 	exit1(td, 0, sig);
3659 	/* NOTREACHED */
3660 }
3661 
3662 /*
3663  * Send queued SIGCHLD to parent when child process's state
3664  * is changed.
3665  */
3666 static void
3667 sigparent(struct proc *p, int reason, int status)
3668 {
3669 	PROC_LOCK_ASSERT(p, MA_OWNED);
3670 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3671 
3672 	if (p->p_ksi != NULL) {
3673 		p->p_ksi->ksi_signo  = SIGCHLD;
3674 		p->p_ksi->ksi_code   = reason;
3675 		p->p_ksi->ksi_status = status;
3676 		p->p_ksi->ksi_pid    = p->p_pid;
3677 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3678 		if (KSI_ONQ(p->p_ksi))
3679 			return;
3680 	}
3681 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3682 }
3683 
3684 static void
3685 childproc_jobstate(struct proc *p, int reason, int sig)
3686 {
3687 	struct sigacts *ps;
3688 
3689 	PROC_LOCK_ASSERT(p, MA_OWNED);
3690 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3691 
3692 	/*
3693 	 * Wake up parent sleeping in kern_wait(), also send
3694 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3695 	 * that parent will awake, because parent may masked
3696 	 * the signal.
3697 	 */
3698 	p->p_pptr->p_flag |= P_STATCHILD;
3699 	wakeup(p->p_pptr);
3700 
3701 	ps = p->p_pptr->p_sigacts;
3702 	mtx_lock(&ps->ps_mtx);
3703 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3704 		mtx_unlock(&ps->ps_mtx);
3705 		sigparent(p, reason, sig);
3706 	} else
3707 		mtx_unlock(&ps->ps_mtx);
3708 }
3709 
3710 void
3711 childproc_stopped(struct proc *p, int reason)
3712 {
3713 
3714 	childproc_jobstate(p, reason, p->p_xsig);
3715 }
3716 
3717 void
3718 childproc_continued(struct proc *p)
3719 {
3720 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3721 }
3722 
3723 void
3724 childproc_exited(struct proc *p)
3725 {
3726 	int reason, status;
3727 
3728 	if (WCOREDUMP(p->p_xsig)) {
3729 		reason = CLD_DUMPED;
3730 		status = WTERMSIG(p->p_xsig);
3731 	} else if (WIFSIGNALED(p->p_xsig)) {
3732 		reason = CLD_KILLED;
3733 		status = WTERMSIG(p->p_xsig);
3734 	} else {
3735 		reason = CLD_EXITED;
3736 		status = p->p_xexit;
3737 	}
3738 	/*
3739 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3740 	 * done in exit1().
3741 	 */
3742 	sigparent(p, reason, status);
3743 }
3744 
3745 #define	MAX_NUM_CORE_FILES 100000
3746 #ifndef NUM_CORE_FILES
3747 #define	NUM_CORE_FILES 5
3748 #endif
3749 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3750 static int num_cores = NUM_CORE_FILES;
3751 
3752 static int
3753 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3754 {
3755 	int error;
3756 	int new_val;
3757 
3758 	new_val = num_cores;
3759 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3760 	if (error != 0 || req->newptr == NULL)
3761 		return (error);
3762 	if (new_val > MAX_NUM_CORE_FILES)
3763 		new_val = MAX_NUM_CORE_FILES;
3764 	if (new_val < 0)
3765 		new_val = 0;
3766 	num_cores = new_val;
3767 	return (0);
3768 }
3769 SYSCTL_PROC(_debug, OID_AUTO, ncores,
3770     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
3771     sysctl_debug_num_cores_check, "I",
3772     "Maximum number of generated process corefiles while using index format");
3773 
3774 #define	GZIP_SUFFIX	".gz"
3775 #define	ZSTD_SUFFIX	".zst"
3776 
3777 int compress_user_cores = 0;
3778 
3779 static int
3780 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3781 {
3782 	int error, val;
3783 
3784 	val = compress_user_cores;
3785 	error = sysctl_handle_int(oidp, &val, 0, req);
3786 	if (error != 0 || req->newptr == NULL)
3787 		return (error);
3788 	if (val != 0 && !compressor_avail(val))
3789 		return (EINVAL);
3790 	compress_user_cores = val;
3791 	return (error);
3792 }
3793 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores,
3794     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, 0, sizeof(int),
3795     sysctl_compress_user_cores, "I",
3796     "Enable compression of user corefiles ("
3797     __XSTRING(COMPRESS_GZIP) " = gzip, "
3798     __XSTRING(COMPRESS_ZSTD) " = zstd)");
3799 
3800 int compress_user_cores_level = 6;
3801 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3802     &compress_user_cores_level, 0,
3803     "Corefile compression level");
3804 
3805 /*
3806  * Protect the access to corefilename[] by allproc_lock.
3807  */
3808 #define	corefilename_lock	allproc_lock
3809 
3810 static char corefilename[MAXPATHLEN] = {"%N.core"};
3811 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3812 
3813 static int
3814 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3815 {
3816 	int error;
3817 
3818 	sx_xlock(&corefilename_lock);
3819 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3820 	    req);
3821 	sx_xunlock(&corefilename_lock);
3822 
3823 	return (error);
3824 }
3825 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3826     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3827     "Process corefile name format string");
3828 
3829 static void
3830 vnode_close_locked(struct thread *td, struct vnode *vp)
3831 {
3832 
3833 	VOP_UNLOCK(vp);
3834 	vn_close(vp, FWRITE, td->td_ucred, td);
3835 }
3836 
3837 /*
3838  * If the core format has a %I in it, then we need to check
3839  * for existing corefiles before defining a name.
3840  * To do this we iterate over 0..ncores to find a
3841  * non-existing core file name to use. If all core files are
3842  * already used we choose the oldest one.
3843  */
3844 static int
3845 corefile_open_last(struct thread *td, char *name, int indexpos,
3846     int indexlen, int ncores, struct vnode **vpp)
3847 {
3848 	struct vnode *oldvp, *nextvp, *vp;
3849 	struct vattr vattr;
3850 	struct nameidata nd;
3851 	int error, i, flags, oflags, cmode;
3852 	char ch;
3853 	struct timespec lasttime;
3854 
3855 	nextvp = oldvp = NULL;
3856 	cmode = S_IRUSR | S_IWUSR;
3857 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3858 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3859 
3860 	for (i = 0; i < ncores; i++) {
3861 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3862 
3863 		ch = name[indexpos + indexlen];
3864 		(void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3865 		    i);
3866 		name[indexpos + indexlen] = ch;
3867 
3868 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
3869 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3870 		    NULL);
3871 		if (error != 0)
3872 			break;
3873 
3874 		vp = nd.ni_vp;
3875 		NDFREE_PNBUF(&nd);
3876 		if ((flags & O_CREAT) == O_CREAT) {
3877 			nextvp = vp;
3878 			break;
3879 		}
3880 
3881 		error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3882 		if (error != 0) {
3883 			vnode_close_locked(td, vp);
3884 			break;
3885 		}
3886 
3887 		if (oldvp == NULL ||
3888 		    lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3889 		    (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3890 		    lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3891 			if (oldvp != NULL)
3892 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3893 			oldvp = vp;
3894 			VOP_UNLOCK(oldvp);
3895 			lasttime = vattr.va_mtime;
3896 		} else {
3897 			vnode_close_locked(td, vp);
3898 		}
3899 	}
3900 
3901 	if (oldvp != NULL) {
3902 		if (nextvp == NULL) {
3903 			if ((td->td_proc->p_flag & P_SUGID) != 0) {
3904 				error = EFAULT;
3905 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3906 			} else {
3907 				nextvp = oldvp;
3908 				error = vn_lock(nextvp, LK_EXCLUSIVE);
3909 				if (error != 0) {
3910 					vn_close(nextvp, FWRITE, td->td_ucred,
3911 					    td);
3912 					nextvp = NULL;
3913 				}
3914 			}
3915 		} else {
3916 			vn_close(oldvp, FWRITE, td->td_ucred, td);
3917 		}
3918 	}
3919 	if (error != 0) {
3920 		if (nextvp != NULL)
3921 			vnode_close_locked(td, oldvp);
3922 	} else {
3923 		*vpp = nextvp;
3924 	}
3925 
3926 	return (error);
3927 }
3928 
3929 /*
3930  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3931  * Expand the name described in corefilename, using name, uid, and pid
3932  * and open/create core file.
3933  * corefilename is a printf-like string, with three format specifiers:
3934  *	%N	name of process ("name")
3935  *	%P	process id (pid)
3936  *	%U	user id (uid)
3937  * For example, "%N.core" is the default; they can be disabled completely
3938  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3939  * This is controlled by the sysctl variable kern.corefile (see above).
3940  */
3941 static int
3942 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3943     int compress, int signum, struct vnode **vpp, char **namep)
3944 {
3945 	struct sbuf sb;
3946 	struct nameidata nd;
3947 	const char *format;
3948 	char *hostname, *name;
3949 	int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
3950 
3951 	hostname = NULL;
3952 	format = corefilename;
3953 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3954 	indexlen = 0;
3955 	indexpos = -1;
3956 	ncores = num_cores;
3957 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3958 	sx_slock(&corefilename_lock);
3959 	for (i = 0; format[i] != '\0'; i++) {
3960 		switch (format[i]) {
3961 		case '%':	/* Format character */
3962 			i++;
3963 			switch (format[i]) {
3964 			case '%':
3965 				sbuf_putc(&sb, '%');
3966 				break;
3967 			case 'H':	/* hostname */
3968 				if (hostname == NULL) {
3969 					hostname = malloc(MAXHOSTNAMELEN,
3970 					    M_TEMP, M_WAITOK);
3971 				}
3972 				getcredhostname(td->td_ucred, hostname,
3973 				    MAXHOSTNAMELEN);
3974 				sbuf_printf(&sb, "%s", hostname);
3975 				break;
3976 			case 'I':	/* autoincrementing index */
3977 				if (indexpos != -1) {
3978 					sbuf_printf(&sb, "%%I");
3979 					break;
3980 				}
3981 
3982 				indexpos = sbuf_len(&sb);
3983 				sbuf_printf(&sb, "%u", ncores - 1);
3984 				indexlen = sbuf_len(&sb) - indexpos;
3985 				break;
3986 			case 'N':	/* process name */
3987 				sbuf_printf(&sb, "%s", comm);
3988 				break;
3989 			case 'P':	/* process id */
3990 				sbuf_printf(&sb, "%u", pid);
3991 				break;
3992 			case 'S':	/* signal number */
3993 				sbuf_printf(&sb, "%i", signum);
3994 				break;
3995 			case 'U':	/* user id */
3996 				sbuf_printf(&sb, "%u", uid);
3997 				break;
3998 			default:
3999 				log(LOG_ERR,
4000 				    "Unknown format character %c in "
4001 				    "corename `%s'\n", format[i], format);
4002 				break;
4003 			}
4004 			break;
4005 		default:
4006 			sbuf_putc(&sb, format[i]);
4007 			break;
4008 		}
4009 	}
4010 	sx_sunlock(&corefilename_lock);
4011 	free(hostname, M_TEMP);
4012 	if (compress == COMPRESS_GZIP)
4013 		sbuf_printf(&sb, GZIP_SUFFIX);
4014 	else if (compress == COMPRESS_ZSTD)
4015 		sbuf_printf(&sb, ZSTD_SUFFIX);
4016 	if (sbuf_error(&sb) != 0) {
4017 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
4018 		    "long\n", (long)pid, comm, (u_long)uid);
4019 		sbuf_delete(&sb);
4020 		free(name, M_TEMP);
4021 		return (ENOMEM);
4022 	}
4023 	sbuf_finish(&sb);
4024 	sbuf_delete(&sb);
4025 
4026 	if (indexpos != -1) {
4027 		error = corefile_open_last(td, name, indexpos, indexlen, ncores,
4028 		    vpp);
4029 		if (error != 0) {
4030 			log(LOG_ERR,
4031 			    "pid %d (%s), uid (%u):  Path `%s' failed "
4032 			    "on initial open test, error = %d\n",
4033 			    pid, comm, uid, name, error);
4034 		}
4035 	} else {
4036 		cmode = S_IRUSR | S_IWUSR;
4037 		oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
4038 		    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
4039 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
4040 		if ((td->td_proc->p_flag & P_SUGID) != 0)
4041 			flags |= O_EXCL;
4042 
4043 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
4044 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
4045 		    NULL);
4046 		if (error == 0) {
4047 			*vpp = nd.ni_vp;
4048 			NDFREE_PNBUF(&nd);
4049 		}
4050 	}
4051 
4052 	if (error != 0) {
4053 #ifdef AUDIT
4054 		audit_proc_coredump(td, name, error);
4055 #endif
4056 		free(name, M_TEMP);
4057 		return (error);
4058 	}
4059 	*namep = name;
4060 	return (0);
4061 }
4062 
4063 /*
4064  * Dump a process' core.  The main routine does some
4065  * policy checking, and creates the name of the coredump;
4066  * then it passes on a vnode and a size limit to the process-specific
4067  * coredump routine if there is one; if there _is not_ one, it returns
4068  * ENOSYS; otherwise it returns the error from the process-specific routine.
4069  */
4070 
4071 static int
4072 coredump(struct thread *td)
4073 {
4074 	struct proc *p = td->td_proc;
4075 	struct ucred *cred = td->td_ucred;
4076 	struct vnode *vp;
4077 	struct flock lf;
4078 	struct vattr vattr;
4079 	size_t fullpathsize;
4080 	int error, error1, locked;
4081 	char *name;			/* name of corefile */
4082 	void *rl_cookie;
4083 	off_t limit;
4084 	char *fullpath, *freepath = NULL;
4085 	struct sbuf *sb;
4086 
4087 	PROC_LOCK_ASSERT(p, MA_OWNED);
4088 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
4089 
4090 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
4091 	    (p->p_flag2 & P2_NOTRACE) != 0) {
4092 		PROC_UNLOCK(p);
4093 		return (EFAULT);
4094 	}
4095 
4096 	/*
4097 	 * Note that the bulk of limit checking is done after
4098 	 * the corefile is created.  The exception is if the limit
4099 	 * for corefiles is 0, in which case we don't bother
4100 	 * creating the corefile at all.  This layout means that
4101 	 * a corefile is truncated instead of not being created,
4102 	 * if it is larger than the limit.
4103 	 */
4104 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
4105 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
4106 		PROC_UNLOCK(p);
4107 		return (EFBIG);
4108 	}
4109 	PROC_UNLOCK(p);
4110 
4111 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
4112 	    compress_user_cores, p->p_sig, &vp, &name);
4113 	if (error != 0)
4114 		return (error);
4115 
4116 	/*
4117 	 * Don't dump to non-regular files or files with links.
4118 	 * Do not dump into system files. Effective user must own the corefile.
4119 	 */
4120 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
4121 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
4122 	    vattr.va_uid != cred->cr_uid) {
4123 		VOP_UNLOCK(vp);
4124 		error = EFAULT;
4125 		goto out;
4126 	}
4127 
4128 	VOP_UNLOCK(vp);
4129 
4130 	/* Postpone other writers, including core dumps of other processes. */
4131 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
4132 
4133 	lf.l_whence = SEEK_SET;
4134 	lf.l_start = 0;
4135 	lf.l_len = 0;
4136 	lf.l_type = F_WRLCK;
4137 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
4138 
4139 	VATTR_NULL(&vattr);
4140 	vattr.va_size = 0;
4141 	if (set_core_nodump_flag)
4142 		vattr.va_flags = UF_NODUMP;
4143 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4144 	VOP_SETATTR(vp, &vattr, cred);
4145 	VOP_UNLOCK(vp);
4146 	PROC_LOCK(p);
4147 	p->p_acflag |= ACORE;
4148 	PROC_UNLOCK(p);
4149 
4150 	if (p->p_sysent->sv_coredump != NULL) {
4151 		error = p->p_sysent->sv_coredump(td, vp, limit, 0);
4152 	} else {
4153 		error = ENOSYS;
4154 	}
4155 
4156 	if (locked) {
4157 		lf.l_type = F_UNLCK;
4158 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
4159 	}
4160 	vn_rangelock_unlock(vp, rl_cookie);
4161 
4162 	/*
4163 	 * Notify the userland helper that a process triggered a core dump.
4164 	 * This allows the helper to run an automated debugging session.
4165 	 */
4166 	if (error != 0 || coredump_devctl == 0)
4167 		goto out;
4168 	sb = sbuf_new_auto();
4169 	if (vn_fullpath_global(p->p_textvp, &fullpath, &freepath) != 0)
4170 		goto out2;
4171 	sbuf_printf(sb, "comm=\"");
4172 	devctl_safe_quote_sb(sb, fullpath);
4173 	free(freepath, M_TEMP);
4174 	sbuf_printf(sb, "\" core=\"");
4175 
4176 	/*
4177 	 * We can't lookup core file vp directly. When we're replacing a core, and
4178 	 * other random times, we flush the name cache, so it will fail. Instead,
4179 	 * if the path of the core is relative, add the current dir in front if it.
4180 	 */
4181 	if (name[0] != '/') {
4182 		fullpathsize = MAXPATHLEN;
4183 		freepath = malloc(fullpathsize, M_TEMP, M_WAITOK);
4184 		if (vn_getcwd(freepath, &fullpath, &fullpathsize) != 0) {
4185 			free(freepath, M_TEMP);
4186 			goto out2;
4187 		}
4188 		devctl_safe_quote_sb(sb, fullpath);
4189 		free(freepath, M_TEMP);
4190 		sbuf_putc(sb, '/');
4191 	}
4192 	devctl_safe_quote_sb(sb, name);
4193 	sbuf_printf(sb, "\"");
4194 	if (sbuf_finish(sb) == 0)
4195 		devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
4196 out2:
4197 	sbuf_delete(sb);
4198 out:
4199 	error1 = vn_close(vp, FWRITE, cred, td);
4200 	if (error == 0)
4201 		error = error1;
4202 #ifdef AUDIT
4203 	audit_proc_coredump(td, name, error);
4204 #endif
4205 	free(name, M_TEMP);
4206 	return (error);
4207 }
4208 
4209 /*
4210  * Nonexistent system call-- signal process (may want to handle it).  Flag
4211  * error in case process won't see signal immediately (blocked or ignored).
4212  */
4213 #ifndef _SYS_SYSPROTO_H_
4214 struct nosys_args {
4215 	int	dummy;
4216 };
4217 #endif
4218 /* ARGSUSED */
4219 int
4220 nosys(struct thread *td, struct nosys_args *args)
4221 {
4222 	struct proc *p;
4223 
4224 	p = td->td_proc;
4225 
4226 	if (SV_PROC_FLAG(p, SV_SIGSYS) != 0 && kern_signosys) {
4227 		PROC_LOCK(p);
4228 		tdsignal(td, SIGSYS);
4229 		PROC_UNLOCK(p);
4230 	}
4231 	if (kern_lognosys == 1 || kern_lognosys == 3) {
4232 		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4233 		    td->td_sa.code);
4234 	}
4235 	if (kern_lognosys == 2 || kern_lognosys == 3 ||
4236 	    (p->p_pid == 1 && (kern_lognosys & 3) == 0)) {
4237 		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4238 		    td->td_sa.code);
4239 	}
4240 	return (ENOSYS);
4241 }
4242 
4243 /*
4244  * Send a SIGIO or SIGURG signal to a process or process group using stored
4245  * credentials rather than those of the current process.
4246  */
4247 void
4248 pgsigio(struct sigio **sigiop, int sig, int checkctty)
4249 {
4250 	ksiginfo_t ksi;
4251 	struct sigio *sigio;
4252 
4253 	ksiginfo_init(&ksi);
4254 	ksi.ksi_signo = sig;
4255 	ksi.ksi_code = SI_KERNEL;
4256 
4257 	SIGIO_LOCK();
4258 	sigio = *sigiop;
4259 	if (sigio == NULL) {
4260 		SIGIO_UNLOCK();
4261 		return;
4262 	}
4263 	if (sigio->sio_pgid > 0) {
4264 		PROC_LOCK(sigio->sio_proc);
4265 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
4266 			kern_psignal(sigio->sio_proc, sig);
4267 		PROC_UNLOCK(sigio->sio_proc);
4268 	} else if (sigio->sio_pgid < 0) {
4269 		struct proc *p;
4270 
4271 		PGRP_LOCK(sigio->sio_pgrp);
4272 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
4273 			PROC_LOCK(p);
4274 			if (p->p_state == PRS_NORMAL &&
4275 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
4276 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
4277 				kern_psignal(p, sig);
4278 			PROC_UNLOCK(p);
4279 		}
4280 		PGRP_UNLOCK(sigio->sio_pgrp);
4281 	}
4282 	SIGIO_UNLOCK();
4283 }
4284 
4285 static int
4286 filt_sigattach(struct knote *kn)
4287 {
4288 	struct proc *p = curproc;
4289 
4290 	kn->kn_ptr.p_proc = p;
4291 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
4292 
4293 	knlist_add(p->p_klist, kn, 0);
4294 
4295 	return (0);
4296 }
4297 
4298 static void
4299 filt_sigdetach(struct knote *kn)
4300 {
4301 	struct proc *p = kn->kn_ptr.p_proc;
4302 
4303 	knlist_remove(p->p_klist, kn, 0);
4304 }
4305 
4306 /*
4307  * signal knotes are shared with proc knotes, so we apply a mask to
4308  * the hint in order to differentiate them from process hints.  This
4309  * could be avoided by using a signal-specific knote list, but probably
4310  * isn't worth the trouble.
4311  */
4312 static int
4313 filt_signal(struct knote *kn, long hint)
4314 {
4315 
4316 	if (hint & NOTE_SIGNAL) {
4317 		hint &= ~NOTE_SIGNAL;
4318 
4319 		if (kn->kn_id == hint)
4320 			kn->kn_data++;
4321 	}
4322 	return (kn->kn_data != 0);
4323 }
4324 
4325 struct sigacts *
4326 sigacts_alloc(void)
4327 {
4328 	struct sigacts *ps;
4329 
4330 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
4331 	refcount_init(&ps->ps_refcnt, 1);
4332 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
4333 	return (ps);
4334 }
4335 
4336 void
4337 sigacts_free(struct sigacts *ps)
4338 {
4339 
4340 	if (refcount_release(&ps->ps_refcnt) == 0)
4341 		return;
4342 	mtx_destroy(&ps->ps_mtx);
4343 	free(ps, M_SUBPROC);
4344 }
4345 
4346 struct sigacts *
4347 sigacts_hold(struct sigacts *ps)
4348 {
4349 
4350 	refcount_acquire(&ps->ps_refcnt);
4351 	return (ps);
4352 }
4353 
4354 void
4355 sigacts_copy(struct sigacts *dest, struct sigacts *src)
4356 {
4357 
4358 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
4359 	mtx_lock(&src->ps_mtx);
4360 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
4361 	mtx_unlock(&src->ps_mtx);
4362 }
4363 
4364 int
4365 sigacts_shared(struct sigacts *ps)
4366 {
4367 
4368 	return (ps->ps_refcnt > 1);
4369 }
4370 
4371 void
4372 sig_drop_caught(struct proc *p)
4373 {
4374 	int sig;
4375 	struct sigacts *ps;
4376 
4377 	ps = p->p_sigacts;
4378 	PROC_LOCK_ASSERT(p, MA_OWNED);
4379 	mtx_assert(&ps->ps_mtx, MA_OWNED);
4380 	SIG_FOREACH(sig, &ps->ps_sigcatch) {
4381 		sigdflt(ps, sig);
4382 		if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
4383 			sigqueue_delete_proc(p, sig);
4384 	}
4385 }
4386 
4387 static void
4388 sigfastblock_failed(struct thread *td, bool sendsig, bool write)
4389 {
4390 	ksiginfo_t ksi;
4391 
4392 	/*
4393 	 * Prevent further fetches and SIGSEGVs, allowing thread to
4394 	 * issue syscalls despite corruption.
4395 	 */
4396 	sigfastblock_clear(td);
4397 
4398 	if (!sendsig)
4399 		return;
4400 	ksiginfo_init_trap(&ksi);
4401 	ksi.ksi_signo = SIGSEGV;
4402 	ksi.ksi_code = write ? SEGV_ACCERR : SEGV_MAPERR;
4403 	ksi.ksi_addr = td->td_sigblock_ptr;
4404 	trapsignal(td, &ksi);
4405 }
4406 
4407 static bool
4408 sigfastblock_fetch_sig(struct thread *td, bool sendsig, uint32_t *valp)
4409 {
4410 	uint32_t res;
4411 
4412 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4413 		return (true);
4414 	if (fueword32((void *)td->td_sigblock_ptr, &res) == -1) {
4415 		sigfastblock_failed(td, sendsig, false);
4416 		return (false);
4417 	}
4418 	*valp = res;
4419 	td->td_sigblock_val = res & ~SIGFASTBLOCK_FLAGS;
4420 	return (true);
4421 }
4422 
4423 static void
4424 sigfastblock_resched(struct thread *td, bool resched)
4425 {
4426 	struct proc *p;
4427 
4428 	if (resched) {
4429 		p = td->td_proc;
4430 		PROC_LOCK(p);
4431 		reschedule_signals(p, td->td_sigmask, 0);
4432 		PROC_UNLOCK(p);
4433 	}
4434 	ast_sched(td, TDA_SIG);
4435 }
4436 
4437 int
4438 sys_sigfastblock(struct thread *td, struct sigfastblock_args *uap)
4439 {
4440 	struct proc *p;
4441 	int error, res;
4442 	uint32_t oldval;
4443 
4444 	error = 0;
4445 	p = td->td_proc;
4446 	switch (uap->cmd) {
4447 	case SIGFASTBLOCK_SETPTR:
4448 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
4449 			error = EBUSY;
4450 			break;
4451 		}
4452 		if (((uintptr_t)(uap->ptr) & (sizeof(uint32_t) - 1)) != 0) {
4453 			error = EINVAL;
4454 			break;
4455 		}
4456 		td->td_pflags |= TDP_SIGFASTBLOCK;
4457 		td->td_sigblock_ptr = uap->ptr;
4458 		break;
4459 
4460 	case SIGFASTBLOCK_UNBLOCK:
4461 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4462 			error = EINVAL;
4463 			break;
4464 		}
4465 
4466 		for (;;) {
4467 			res = casueword32(td->td_sigblock_ptr,
4468 			    SIGFASTBLOCK_PEND, &oldval, 0);
4469 			if (res == -1) {
4470 				error = EFAULT;
4471 				sigfastblock_failed(td, false, true);
4472 				break;
4473 			}
4474 			if (res == 0)
4475 				break;
4476 			MPASS(res == 1);
4477 			if (oldval != SIGFASTBLOCK_PEND) {
4478 				error = EBUSY;
4479 				break;
4480 			}
4481 			error = thread_check_susp(td, false);
4482 			if (error != 0)
4483 				break;
4484 		}
4485 		if (error != 0)
4486 			break;
4487 
4488 		/*
4489 		 * td_sigblock_val is cleared there, but not on a
4490 		 * syscall exit.  The end effect is that a single
4491 		 * interruptible sleep, while user sigblock word is
4492 		 * set, might return EINTR or ERESTART to usermode
4493 		 * without delivering signal.  All further sleeps,
4494 		 * until userspace clears the word and does
4495 		 * sigfastblock(UNBLOCK), observe current word and no
4496 		 * longer get interrupted.  It is slight
4497 		 * non-conformance, with alternative to have read the
4498 		 * sigblock word on each syscall entry.
4499 		 */
4500 		td->td_sigblock_val = 0;
4501 
4502 		/*
4503 		 * Rely on normal ast mechanism to deliver pending
4504 		 * signals to current thread.  But notify others about
4505 		 * fake unblock.
4506 		 */
4507 		sigfastblock_resched(td, error == 0 && p->p_numthreads != 1);
4508 
4509 		break;
4510 
4511 	case SIGFASTBLOCK_UNSETPTR:
4512 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4513 			error = EINVAL;
4514 			break;
4515 		}
4516 		if (!sigfastblock_fetch_sig(td, false, &oldval)) {
4517 			error = EFAULT;
4518 			break;
4519 		}
4520 		if (oldval != 0 && oldval != SIGFASTBLOCK_PEND) {
4521 			error = EBUSY;
4522 			break;
4523 		}
4524 		sigfastblock_clear(td);
4525 		break;
4526 
4527 	default:
4528 		error = EINVAL;
4529 		break;
4530 	}
4531 	return (error);
4532 }
4533 
4534 void
4535 sigfastblock_clear(struct thread *td)
4536 {
4537 	bool resched;
4538 
4539 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4540 		return;
4541 	td->td_sigblock_val = 0;
4542 	resched = (td->td_pflags & TDP_SIGFASTPENDING) != 0 ||
4543 	    SIGPENDING(td);
4544 	td->td_pflags &= ~(TDP_SIGFASTBLOCK | TDP_SIGFASTPENDING);
4545 	sigfastblock_resched(td, resched);
4546 }
4547 
4548 void
4549 sigfastblock_fetch(struct thread *td)
4550 {
4551 	uint32_t val;
4552 
4553 	(void)sigfastblock_fetch_sig(td, true, &val);
4554 }
4555 
4556 static void
4557 sigfastblock_setpend1(struct thread *td)
4558 {
4559 	int res;
4560 	uint32_t oldval;
4561 
4562 	if ((td->td_pflags & TDP_SIGFASTPENDING) == 0)
4563 		return;
4564 	res = fueword32((void *)td->td_sigblock_ptr, &oldval);
4565 	if (res == -1) {
4566 		sigfastblock_failed(td, true, false);
4567 		return;
4568 	}
4569 	for (;;) {
4570 		res = casueword32(td->td_sigblock_ptr, oldval, &oldval,
4571 		    oldval | SIGFASTBLOCK_PEND);
4572 		if (res == -1) {
4573 			sigfastblock_failed(td, true, true);
4574 			return;
4575 		}
4576 		if (res == 0) {
4577 			td->td_sigblock_val = oldval & ~SIGFASTBLOCK_FLAGS;
4578 			td->td_pflags &= ~TDP_SIGFASTPENDING;
4579 			break;
4580 		}
4581 		MPASS(res == 1);
4582 		if (thread_check_susp(td, false) != 0)
4583 			break;
4584 	}
4585 }
4586 
4587 static void
4588 sigfastblock_setpend(struct thread *td, bool resched)
4589 {
4590 	struct proc *p;
4591 
4592 	sigfastblock_setpend1(td);
4593 	if (resched) {
4594 		p = td->td_proc;
4595 		PROC_LOCK(p);
4596 		reschedule_signals(p, fastblock_mask, SIGPROCMASK_FASTBLK);
4597 		PROC_UNLOCK(p);
4598 	}
4599 }
4600