xref: /freebsd/sys/kern/kern_sig.c (revision a90b9d0159070121c221b966469c3e36d912bf82)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 
37 #include "opt_capsicum.h"
38 #include "opt_ktrace.h"
39 
40 #include <sys/param.h>
41 #include <sys/capsicum.h>
42 #include <sys/ctype.h>
43 #include <sys/systm.h>
44 #include <sys/signalvar.h>
45 #include <sys/vnode.h>
46 #include <sys/acct.h>
47 #include <sys/capsicum.h>
48 #include <sys/compressor.h>
49 #include <sys/condvar.h>
50 #include <sys/devctl.h>
51 #include <sys/event.h>
52 #include <sys/fcntl.h>
53 #include <sys/imgact.h>
54 #include <sys/jail.h>
55 #include <sys/kernel.h>
56 #include <sys/ktr.h>
57 #include <sys/ktrace.h>
58 #include <sys/limits.h>
59 #include <sys/lock.h>
60 #include <sys/malloc.h>
61 #include <sys/mutex.h>
62 #include <sys/refcount.h>
63 #include <sys/namei.h>
64 #include <sys/proc.h>
65 #include <sys/procdesc.h>
66 #include <sys/ptrace.h>
67 #include <sys/posix4.h>
68 #include <sys/racct.h>
69 #include <sys/resourcevar.h>
70 #include <sys/sdt.h>
71 #include <sys/sbuf.h>
72 #include <sys/sleepqueue.h>
73 #include <sys/smp.h>
74 #include <sys/stat.h>
75 #include <sys/sx.h>
76 #include <sys/syscall.h>
77 #include <sys/syscallsubr.h>
78 #include <sys/sysctl.h>
79 #include <sys/sysent.h>
80 #include <sys/syslog.h>
81 #include <sys/sysproto.h>
82 #include <sys/timers.h>
83 #include <sys/unistd.h>
84 #include <sys/vmmeter.h>
85 #include <sys/wait.h>
86 #include <vm/vm.h>
87 #include <vm/vm_extern.h>
88 #include <vm/uma.h>
89 
90 #include <machine/cpu.h>
91 
92 #include <security/audit/audit.h>
93 
94 #define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
95 
96 SDT_PROVIDER_DECLARE(proc);
97 SDT_PROBE_DEFINE3(proc, , , signal__send,
98     "struct thread *", "struct proc *", "int");
99 SDT_PROBE_DEFINE2(proc, , , signal__clear,
100     "int", "ksiginfo_t *");
101 SDT_PROBE_DEFINE3(proc, , , signal__discard,
102     "struct thread *", "struct proc *", "int");
103 
104 static int	coredump(struct thread *);
105 static int	killpg1(struct thread *td, int sig, int pgid, int all,
106 		    ksiginfo_t *ksi);
107 static int	issignal(struct thread *td);
108 static void	reschedule_signals(struct proc *p, sigset_t block, int flags);
109 static int	sigprop(int sig);
110 static void	tdsigwakeup(struct thread *, int, sig_t, int);
111 static int	sig_suspend_threads(struct thread *, struct proc *);
112 static int	filt_sigattach(struct knote *kn);
113 static void	filt_sigdetach(struct knote *kn);
114 static int	filt_signal(struct knote *kn, long hint);
115 static struct thread *sigtd(struct proc *p, int sig, bool fast_sigblock);
116 static void	sigqueue_start(void);
117 static void	sigfastblock_setpend(struct thread *td, bool resched);
118 
119 static uma_zone_t	ksiginfo_zone = NULL;
120 struct filterops sig_filtops = {
121 	.f_isfd = 0,
122 	.f_attach = filt_sigattach,
123 	.f_detach = filt_sigdetach,
124 	.f_event = filt_signal,
125 };
126 
127 static int	kern_logsigexit = 1;
128 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
129     &kern_logsigexit, 0,
130     "Log processes quitting on abnormal signals to syslog(3)");
131 
132 static int	kern_forcesigexit = 1;
133 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
134     &kern_forcesigexit, 0, "Force trap signal to be handled");
135 
136 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
137     "POSIX real time signal");
138 
139 static int	max_pending_per_proc = 128;
140 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
141     &max_pending_per_proc, 0, "Max pending signals per proc");
142 
143 static int	preallocate_siginfo = 1024;
144 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
145     &preallocate_siginfo, 0, "Preallocated signal memory size");
146 
147 static int	signal_overflow = 0;
148 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
149     &signal_overflow, 0, "Number of signals overflew");
150 
151 static int	signal_alloc_fail = 0;
152 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
153     &signal_alloc_fail, 0, "signals failed to be allocated");
154 
155 static int	kern_lognosys = 0;
156 SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
157     "Log invalid syscalls");
158 
159 static int	kern_signosys = 1;
160 SYSCTL_INT(_kern, OID_AUTO, signosys, CTLFLAG_RWTUN, &kern_signosys, 0,
161     "Send SIGSYS on return from invalid syscall");
162 
163 __read_frequently bool sigfastblock_fetch_always = false;
164 SYSCTL_BOOL(_kern, OID_AUTO, sigfastblock_fetch_always, CTLFLAG_RWTUN,
165     &sigfastblock_fetch_always, 0,
166     "Fetch sigfastblock word on each syscall entry for proper "
167     "blocking semantic");
168 
169 static bool	kern_sig_discard_ign = true;
170 SYSCTL_BOOL(_kern, OID_AUTO, sig_discard_ign, CTLFLAG_RWTUN,
171     &kern_sig_discard_ign, 0,
172     "Discard ignored signals on delivery, otherwise queue them to "
173     "the target queue");
174 
175 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
176 
177 /*
178  * Policy -- Can ucred cr1 send SIGIO to process cr2?
179  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
180  * in the right situations.
181  */
182 #define CANSIGIO(cr1, cr2) \
183 	((cr1)->cr_uid == 0 || \
184 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
185 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
186 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
187 	    (cr1)->cr_uid == (cr2)->cr_uid)
188 
189 static int	sugid_coredump;
190 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
191     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
192 
193 static int	capmode_coredump;
194 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
195     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
196 
197 static int	do_coredump = 1;
198 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
199 	&do_coredump, 0, "Enable/Disable coredumps");
200 
201 static int	set_core_nodump_flag = 0;
202 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
203 	0, "Enable setting the NODUMP flag on coredump files");
204 
205 static int	coredump_devctl = 0;
206 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
207 	0, "Generate a devctl notification when processes coredump");
208 
209 /*
210  * Signal properties and actions.
211  * The array below categorizes the signals and their default actions
212  * according to the following properties:
213  */
214 #define	SIGPROP_KILL		0x01	/* terminates process by default */
215 #define	SIGPROP_CORE		0x02	/* ditto and coredumps */
216 #define	SIGPROP_STOP		0x04	/* suspend process */
217 #define	SIGPROP_TTYSTOP		0x08	/* ditto, from tty */
218 #define	SIGPROP_IGNORE		0x10	/* ignore by default */
219 #define	SIGPROP_CONT		0x20	/* continue if suspended */
220 
221 static const int sigproptbl[NSIG] = {
222 	[SIGHUP] =	SIGPROP_KILL,
223 	[SIGINT] =	SIGPROP_KILL,
224 	[SIGQUIT] =	SIGPROP_KILL | SIGPROP_CORE,
225 	[SIGILL] =	SIGPROP_KILL | SIGPROP_CORE,
226 	[SIGTRAP] =	SIGPROP_KILL | SIGPROP_CORE,
227 	[SIGABRT] =	SIGPROP_KILL | SIGPROP_CORE,
228 	[SIGEMT] =	SIGPROP_KILL | SIGPROP_CORE,
229 	[SIGFPE] =	SIGPROP_KILL | SIGPROP_CORE,
230 	[SIGKILL] =	SIGPROP_KILL,
231 	[SIGBUS] =	SIGPROP_KILL | SIGPROP_CORE,
232 	[SIGSEGV] =	SIGPROP_KILL | SIGPROP_CORE,
233 	[SIGSYS] =	SIGPROP_KILL | SIGPROP_CORE,
234 	[SIGPIPE] =	SIGPROP_KILL,
235 	[SIGALRM] =	SIGPROP_KILL,
236 	[SIGTERM] =	SIGPROP_KILL,
237 	[SIGURG] =	SIGPROP_IGNORE,
238 	[SIGSTOP] =	SIGPROP_STOP,
239 	[SIGTSTP] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
240 	[SIGCONT] =	SIGPROP_IGNORE | SIGPROP_CONT,
241 	[SIGCHLD] =	SIGPROP_IGNORE,
242 	[SIGTTIN] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
243 	[SIGTTOU] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
244 	[SIGIO] =	SIGPROP_IGNORE,
245 	[SIGXCPU] =	SIGPROP_KILL,
246 	[SIGXFSZ] =	SIGPROP_KILL,
247 	[SIGVTALRM] =	SIGPROP_KILL,
248 	[SIGPROF] =	SIGPROP_KILL,
249 	[SIGWINCH] =	SIGPROP_IGNORE,
250 	[SIGINFO] =	SIGPROP_IGNORE,
251 	[SIGUSR1] =	SIGPROP_KILL,
252 	[SIGUSR2] =	SIGPROP_KILL,
253 };
254 
255 #define	_SIG_FOREACH_ADVANCE(i, set) ({					\
256 	int __found;							\
257 	for (;;) {							\
258 		if (__bits != 0) {					\
259 			int __sig = ffs(__bits);			\
260 			__bits &= ~(1u << (__sig - 1));			\
261 			sig = __i * sizeof((set)->__bits[0]) * NBBY + __sig; \
262 			__found = 1;					\
263 			break;						\
264 		}							\
265 		if (++__i == _SIG_WORDS) {				\
266 			__found = 0;					\
267 			break;						\
268 		}							\
269 		__bits = (set)->__bits[__i];				\
270 	}								\
271 	__found != 0;							\
272 })
273 
274 #define	SIG_FOREACH(i, set)						\
275 	for (int32_t __i = -1, __bits = 0;				\
276 	    _SIG_FOREACH_ADVANCE(i, set); )				\
277 
278 static sigset_t fastblock_mask;
279 
280 static void
281 ast_sig(struct thread *td, int tda)
282 {
283 	struct proc *p;
284 	int old_boundary, sig;
285 	bool resched_sigs;
286 
287 	p = td->td_proc;
288 
289 #ifdef DIAGNOSTIC
290 	if (p->p_numthreads == 1 && (tda & (TDAI(TDA_SIG) |
291 	    TDAI(TDA_AST))) == 0) {
292 		PROC_LOCK(p);
293 		thread_lock(td);
294 		/*
295 		 * Note that TDA_SIG should be re-read from
296 		 * td_ast, since signal might have been delivered
297 		 * after we cleared td_flags above.  This is one of
298 		 * the reason for looping check for AST condition.
299 		 * See comment in userret() about P_PPWAIT.
300 		 */
301 		if ((p->p_flag & P_PPWAIT) == 0 &&
302 		    (td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
303 			if (SIGPENDING(td) && ((tda | td->td_ast) &
304 			    (TDAI(TDA_SIG) | TDAI(TDA_AST))) == 0) {
305 				thread_unlock(td); /* fix dumps */
306 				panic(
307 				    "failed2 to set signal flags for ast p %p "
308 				    "td %p tda %#x td_ast %#x fl %#x",
309 				    p, td, tda, td->td_ast, td->td_flags);
310 			}
311 		}
312 		thread_unlock(td);
313 		PROC_UNLOCK(p);
314 	}
315 #endif
316 
317 	/*
318 	 * Check for signals. Unlocked reads of p_pendingcnt or
319 	 * p_siglist might cause process-directed signal to be handled
320 	 * later.
321 	 */
322 	if ((tda & TDAI(TDA_SIG)) != 0 || p->p_pendingcnt > 0 ||
323 	    !SIGISEMPTY(p->p_siglist)) {
324 		sigfastblock_fetch(td);
325 		PROC_LOCK(p);
326 		old_boundary = ~TDB_BOUNDARY | (td->td_dbgflags & TDB_BOUNDARY);
327 		td->td_dbgflags |= TDB_BOUNDARY;
328 		mtx_lock(&p->p_sigacts->ps_mtx);
329 		while ((sig = cursig(td)) != 0) {
330 			KASSERT(sig >= 0, ("sig %d", sig));
331 			postsig(sig);
332 		}
333 		mtx_unlock(&p->p_sigacts->ps_mtx);
334 		td->td_dbgflags &= old_boundary;
335 		PROC_UNLOCK(p);
336 		resched_sigs = true;
337 	} else {
338 		resched_sigs = false;
339 	}
340 
341 	/*
342 	 * Handle deferred update of the fast sigblock value, after
343 	 * the postsig() loop was performed.
344 	 */
345 	sigfastblock_setpend(td, resched_sigs);
346 }
347 
348 static void
349 ast_sigsuspend(struct thread *td, int tda __unused)
350 {
351 	MPASS((td->td_pflags & TDP_OLDMASK) != 0);
352 	td->td_pflags &= ~TDP_OLDMASK;
353 	kern_sigprocmask(td, SIG_SETMASK, &td->td_oldsigmask, NULL, 0);
354 }
355 
356 static void
357 sigqueue_start(void)
358 {
359 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
360 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
361 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
362 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
363 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
364 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
365 	SIGFILLSET(fastblock_mask);
366 	SIG_CANTMASK(fastblock_mask);
367 	ast_register(TDA_SIG, ASTR_UNCOND, 0, ast_sig);
368 	ast_register(TDA_SIGSUSPEND, ASTR_ASTF_REQUIRED | ASTR_TDP,
369 	    TDP_OLDMASK, ast_sigsuspend);
370 }
371 
372 ksiginfo_t *
373 ksiginfo_alloc(int mwait)
374 {
375 	MPASS(mwait == M_WAITOK || mwait == M_NOWAIT);
376 
377 	if (ksiginfo_zone == NULL)
378 		return (NULL);
379 	return (uma_zalloc(ksiginfo_zone, mwait | M_ZERO));
380 }
381 
382 void
383 ksiginfo_free(ksiginfo_t *ksi)
384 {
385 	uma_zfree(ksiginfo_zone, ksi);
386 }
387 
388 static __inline bool
389 ksiginfo_tryfree(ksiginfo_t *ksi)
390 {
391 	if ((ksi->ksi_flags & KSI_EXT) == 0) {
392 		uma_zfree(ksiginfo_zone, ksi);
393 		return (true);
394 	}
395 	return (false);
396 }
397 
398 void
399 sigqueue_init(sigqueue_t *list, struct proc *p)
400 {
401 	SIGEMPTYSET(list->sq_signals);
402 	SIGEMPTYSET(list->sq_kill);
403 	SIGEMPTYSET(list->sq_ptrace);
404 	TAILQ_INIT(&list->sq_list);
405 	list->sq_proc = p;
406 	list->sq_flags = SQ_INIT;
407 }
408 
409 /*
410  * Get a signal's ksiginfo.
411  * Return:
412  *	0	-	signal not found
413  *	others	-	signal number
414  */
415 static int
416 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
417 {
418 	struct proc *p = sq->sq_proc;
419 	struct ksiginfo *ksi, *next;
420 	int count = 0;
421 
422 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
423 
424 	if (!SIGISMEMBER(sq->sq_signals, signo))
425 		return (0);
426 
427 	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
428 		count++;
429 		SIGDELSET(sq->sq_ptrace, signo);
430 		si->ksi_flags |= KSI_PTRACE;
431 	}
432 	if (SIGISMEMBER(sq->sq_kill, signo)) {
433 		count++;
434 		if (count == 1)
435 			SIGDELSET(sq->sq_kill, signo);
436 	}
437 
438 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
439 		if (ksi->ksi_signo == signo) {
440 			if (count == 0) {
441 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
442 				ksi->ksi_sigq = NULL;
443 				ksiginfo_copy(ksi, si);
444 				if (ksiginfo_tryfree(ksi) && p != NULL)
445 					p->p_pendingcnt--;
446 			}
447 			if (++count > 1)
448 				break;
449 		}
450 	}
451 
452 	if (count <= 1)
453 		SIGDELSET(sq->sq_signals, signo);
454 	si->ksi_signo = signo;
455 	return (signo);
456 }
457 
458 void
459 sigqueue_take(ksiginfo_t *ksi)
460 {
461 	struct ksiginfo *kp;
462 	struct proc	*p;
463 	sigqueue_t	*sq;
464 
465 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
466 		return;
467 
468 	p = sq->sq_proc;
469 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
470 	ksi->ksi_sigq = NULL;
471 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
472 		p->p_pendingcnt--;
473 
474 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
475 	     kp = TAILQ_NEXT(kp, ksi_link)) {
476 		if (kp->ksi_signo == ksi->ksi_signo)
477 			break;
478 	}
479 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
480 	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
481 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
482 }
483 
484 static int
485 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
486 {
487 	struct proc *p = sq->sq_proc;
488 	struct ksiginfo *ksi;
489 	int ret = 0;
490 
491 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
492 
493 	/*
494 	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
495 	 * for these signals.
496 	 */
497 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
498 		SIGADDSET(sq->sq_kill, signo);
499 		goto out_set_bit;
500 	}
501 
502 	/* directly insert the ksi, don't copy it */
503 	if (si->ksi_flags & KSI_INS) {
504 		if (si->ksi_flags & KSI_HEAD)
505 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
506 		else
507 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
508 		si->ksi_sigq = sq;
509 		goto out_set_bit;
510 	}
511 
512 	if (__predict_false(ksiginfo_zone == NULL)) {
513 		SIGADDSET(sq->sq_kill, signo);
514 		goto out_set_bit;
515 	}
516 
517 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
518 		signal_overflow++;
519 		ret = EAGAIN;
520 	} else if ((ksi = ksiginfo_alloc(M_NOWAIT)) == NULL) {
521 		signal_alloc_fail++;
522 		ret = EAGAIN;
523 	} else {
524 		if (p != NULL)
525 			p->p_pendingcnt++;
526 		ksiginfo_copy(si, ksi);
527 		ksi->ksi_signo = signo;
528 		if (si->ksi_flags & KSI_HEAD)
529 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
530 		else
531 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
532 		ksi->ksi_sigq = sq;
533 	}
534 
535 	if (ret != 0) {
536 		if ((si->ksi_flags & KSI_PTRACE) != 0) {
537 			SIGADDSET(sq->sq_ptrace, signo);
538 			ret = 0;
539 			goto out_set_bit;
540 		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
541 		    (si->ksi_flags & KSI_SIGQ) == 0) {
542 			SIGADDSET(sq->sq_kill, signo);
543 			ret = 0;
544 			goto out_set_bit;
545 		}
546 		return (ret);
547 	}
548 
549 out_set_bit:
550 	SIGADDSET(sq->sq_signals, signo);
551 	return (ret);
552 }
553 
554 void
555 sigqueue_flush(sigqueue_t *sq)
556 {
557 	struct proc *p = sq->sq_proc;
558 	ksiginfo_t *ksi;
559 
560 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
561 
562 	if (p != NULL)
563 		PROC_LOCK_ASSERT(p, MA_OWNED);
564 
565 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
566 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
567 		ksi->ksi_sigq = NULL;
568 		if (ksiginfo_tryfree(ksi) && p != NULL)
569 			p->p_pendingcnt--;
570 	}
571 
572 	SIGEMPTYSET(sq->sq_signals);
573 	SIGEMPTYSET(sq->sq_kill);
574 	SIGEMPTYSET(sq->sq_ptrace);
575 }
576 
577 static void
578 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
579 {
580 	sigset_t tmp;
581 	struct proc *p1, *p2;
582 	ksiginfo_t *ksi, *next;
583 
584 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
585 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
586 	p1 = src->sq_proc;
587 	p2 = dst->sq_proc;
588 	/* Move siginfo to target list */
589 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
590 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
591 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
592 			if (p1 != NULL)
593 				p1->p_pendingcnt--;
594 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
595 			ksi->ksi_sigq = dst;
596 			if (p2 != NULL)
597 				p2->p_pendingcnt++;
598 		}
599 	}
600 
601 	/* Move pending bits to target list */
602 	tmp = src->sq_kill;
603 	SIGSETAND(tmp, *set);
604 	SIGSETOR(dst->sq_kill, tmp);
605 	SIGSETNAND(src->sq_kill, tmp);
606 
607 	tmp = src->sq_ptrace;
608 	SIGSETAND(tmp, *set);
609 	SIGSETOR(dst->sq_ptrace, tmp);
610 	SIGSETNAND(src->sq_ptrace, tmp);
611 
612 	tmp = src->sq_signals;
613 	SIGSETAND(tmp, *set);
614 	SIGSETOR(dst->sq_signals, tmp);
615 	SIGSETNAND(src->sq_signals, tmp);
616 }
617 
618 #if 0
619 static void
620 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
621 {
622 	sigset_t set;
623 
624 	SIGEMPTYSET(set);
625 	SIGADDSET(set, signo);
626 	sigqueue_move_set(src, dst, &set);
627 }
628 #endif
629 
630 static void
631 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
632 {
633 	struct proc *p = sq->sq_proc;
634 	ksiginfo_t *ksi, *next;
635 
636 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
637 
638 	/* Remove siginfo queue */
639 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
640 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
641 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
642 			ksi->ksi_sigq = NULL;
643 			if (ksiginfo_tryfree(ksi) && p != NULL)
644 				p->p_pendingcnt--;
645 		}
646 	}
647 	SIGSETNAND(sq->sq_kill, *set);
648 	SIGSETNAND(sq->sq_ptrace, *set);
649 	SIGSETNAND(sq->sq_signals, *set);
650 }
651 
652 void
653 sigqueue_delete(sigqueue_t *sq, int signo)
654 {
655 	sigset_t set;
656 
657 	SIGEMPTYSET(set);
658 	SIGADDSET(set, signo);
659 	sigqueue_delete_set(sq, &set);
660 }
661 
662 /* Remove a set of signals for a process */
663 static void
664 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
665 {
666 	sigqueue_t worklist;
667 	struct thread *td0;
668 
669 	PROC_LOCK_ASSERT(p, MA_OWNED);
670 
671 	sigqueue_init(&worklist, NULL);
672 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
673 
674 	FOREACH_THREAD_IN_PROC(p, td0)
675 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
676 
677 	sigqueue_flush(&worklist);
678 }
679 
680 void
681 sigqueue_delete_proc(struct proc *p, int signo)
682 {
683 	sigset_t set;
684 
685 	SIGEMPTYSET(set);
686 	SIGADDSET(set, signo);
687 	sigqueue_delete_set_proc(p, &set);
688 }
689 
690 static void
691 sigqueue_delete_stopmask_proc(struct proc *p)
692 {
693 	sigset_t set;
694 
695 	SIGEMPTYSET(set);
696 	SIGADDSET(set, SIGSTOP);
697 	SIGADDSET(set, SIGTSTP);
698 	SIGADDSET(set, SIGTTIN);
699 	SIGADDSET(set, SIGTTOU);
700 	sigqueue_delete_set_proc(p, &set);
701 }
702 
703 /*
704  * Determine signal that should be delivered to thread td, the current
705  * thread, 0 if none.  If there is a pending stop signal with default
706  * action, the process stops in issignal().
707  */
708 int
709 cursig(struct thread *td)
710 {
711 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
712 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
713 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
714 	return (SIGPENDING(td) ? issignal(td) : 0);
715 }
716 
717 /*
718  * Arrange for ast() to handle unmasked pending signals on return to user
719  * mode.  This must be called whenever a signal is added to td_sigqueue or
720  * unmasked in td_sigmask.
721  */
722 void
723 signotify(struct thread *td)
724 {
725 
726 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
727 
728 	if (SIGPENDING(td))
729 		ast_sched(td, TDA_SIG);
730 }
731 
732 /*
733  * Returns 1 (true) if altstack is configured for the thread, and the
734  * passed stack bottom address falls into the altstack range.  Handles
735  * the 43 compat special case where the alt stack size is zero.
736  */
737 int
738 sigonstack(size_t sp)
739 {
740 	struct thread *td;
741 
742 	td = curthread;
743 	if ((td->td_pflags & TDP_ALTSTACK) == 0)
744 		return (0);
745 #if defined(COMPAT_43)
746 	if (SV_PROC_FLAG(td->td_proc, SV_AOUT) && td->td_sigstk.ss_size == 0)
747 		return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
748 #endif
749 	return (sp >= (size_t)td->td_sigstk.ss_sp &&
750 	    sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
751 }
752 
753 static __inline int
754 sigprop(int sig)
755 {
756 
757 	if (sig > 0 && sig < nitems(sigproptbl))
758 		return (sigproptbl[sig]);
759 	return (0);
760 }
761 
762 static bool
763 sigact_flag_test(const struct sigaction *act, int flag)
764 {
765 
766 	/*
767 	 * SA_SIGINFO is reset when signal disposition is set to
768 	 * ignore or default.  Other flags are kept according to user
769 	 * settings.
770 	 */
771 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
772 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
773 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
774 }
775 
776 /*
777  * kern_sigaction
778  * sigaction
779  * freebsd4_sigaction
780  * osigaction
781  */
782 int
783 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
784     struct sigaction *oact, int flags)
785 {
786 	struct sigacts *ps;
787 	struct proc *p = td->td_proc;
788 
789 	if (!_SIG_VALID(sig))
790 		return (EINVAL);
791 	if (act != NULL && act->sa_handler != SIG_DFL &&
792 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
793 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
794 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
795 		return (EINVAL);
796 
797 	PROC_LOCK(p);
798 	ps = p->p_sigacts;
799 	mtx_lock(&ps->ps_mtx);
800 	if (oact) {
801 		memset(oact, 0, sizeof(*oact));
802 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
803 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
804 			oact->sa_flags |= SA_ONSTACK;
805 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
806 			oact->sa_flags |= SA_RESTART;
807 		if (SIGISMEMBER(ps->ps_sigreset, sig))
808 			oact->sa_flags |= SA_RESETHAND;
809 		if (SIGISMEMBER(ps->ps_signodefer, sig))
810 			oact->sa_flags |= SA_NODEFER;
811 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
812 			oact->sa_flags |= SA_SIGINFO;
813 			oact->sa_sigaction =
814 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
815 		} else
816 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
817 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
818 			oact->sa_flags |= SA_NOCLDSTOP;
819 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
820 			oact->sa_flags |= SA_NOCLDWAIT;
821 	}
822 	if (act) {
823 		if ((sig == SIGKILL || sig == SIGSTOP) &&
824 		    act->sa_handler != SIG_DFL) {
825 			mtx_unlock(&ps->ps_mtx);
826 			PROC_UNLOCK(p);
827 			return (EINVAL);
828 		}
829 
830 		/*
831 		 * Change setting atomically.
832 		 */
833 
834 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
835 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
836 		if (sigact_flag_test(act, SA_SIGINFO)) {
837 			ps->ps_sigact[_SIG_IDX(sig)] =
838 			    (__sighandler_t *)act->sa_sigaction;
839 			SIGADDSET(ps->ps_siginfo, sig);
840 		} else {
841 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
842 			SIGDELSET(ps->ps_siginfo, sig);
843 		}
844 		if (!sigact_flag_test(act, SA_RESTART))
845 			SIGADDSET(ps->ps_sigintr, sig);
846 		else
847 			SIGDELSET(ps->ps_sigintr, sig);
848 		if (sigact_flag_test(act, SA_ONSTACK))
849 			SIGADDSET(ps->ps_sigonstack, sig);
850 		else
851 			SIGDELSET(ps->ps_sigonstack, sig);
852 		if (sigact_flag_test(act, SA_RESETHAND))
853 			SIGADDSET(ps->ps_sigreset, sig);
854 		else
855 			SIGDELSET(ps->ps_sigreset, sig);
856 		if (sigact_flag_test(act, SA_NODEFER))
857 			SIGADDSET(ps->ps_signodefer, sig);
858 		else
859 			SIGDELSET(ps->ps_signodefer, sig);
860 		if (sig == SIGCHLD) {
861 			if (act->sa_flags & SA_NOCLDSTOP)
862 				ps->ps_flag |= PS_NOCLDSTOP;
863 			else
864 				ps->ps_flag &= ~PS_NOCLDSTOP;
865 			if (act->sa_flags & SA_NOCLDWAIT) {
866 				/*
867 				 * Paranoia: since SA_NOCLDWAIT is implemented
868 				 * by reparenting the dying child to PID 1 (and
869 				 * trust it to reap the zombie), PID 1 itself
870 				 * is forbidden to set SA_NOCLDWAIT.
871 				 */
872 				if (p->p_pid == 1)
873 					ps->ps_flag &= ~PS_NOCLDWAIT;
874 				else
875 					ps->ps_flag |= PS_NOCLDWAIT;
876 			} else
877 				ps->ps_flag &= ~PS_NOCLDWAIT;
878 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
879 				ps->ps_flag |= PS_CLDSIGIGN;
880 			else
881 				ps->ps_flag &= ~PS_CLDSIGIGN;
882 		}
883 		/*
884 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
885 		 * and for signals set to SIG_DFL where the default is to
886 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
887 		 * have to restart the process.
888 		 */
889 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
890 		    (sigprop(sig) & SIGPROP_IGNORE &&
891 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
892 			/* never to be seen again */
893 			sigqueue_delete_proc(p, sig);
894 			if (sig != SIGCONT)
895 				/* easier in psignal */
896 				SIGADDSET(ps->ps_sigignore, sig);
897 			SIGDELSET(ps->ps_sigcatch, sig);
898 		} else {
899 			SIGDELSET(ps->ps_sigignore, sig);
900 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
901 				SIGDELSET(ps->ps_sigcatch, sig);
902 			else
903 				SIGADDSET(ps->ps_sigcatch, sig);
904 		}
905 #ifdef COMPAT_FREEBSD4
906 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
907 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
908 		    (flags & KSA_FREEBSD4) == 0)
909 			SIGDELSET(ps->ps_freebsd4, sig);
910 		else
911 			SIGADDSET(ps->ps_freebsd4, sig);
912 #endif
913 #ifdef COMPAT_43
914 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
915 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
916 		    (flags & KSA_OSIGSET) == 0)
917 			SIGDELSET(ps->ps_osigset, sig);
918 		else
919 			SIGADDSET(ps->ps_osigset, sig);
920 #endif
921 	}
922 	mtx_unlock(&ps->ps_mtx);
923 	PROC_UNLOCK(p);
924 	return (0);
925 }
926 
927 #ifndef _SYS_SYSPROTO_H_
928 struct sigaction_args {
929 	int	sig;
930 	struct	sigaction *act;
931 	struct	sigaction *oact;
932 };
933 #endif
934 int
935 sys_sigaction(struct thread *td, struct sigaction_args *uap)
936 {
937 	struct sigaction act, oact;
938 	struct sigaction *actp, *oactp;
939 	int error;
940 
941 	actp = (uap->act != NULL) ? &act : NULL;
942 	oactp = (uap->oact != NULL) ? &oact : NULL;
943 	if (actp) {
944 		error = copyin(uap->act, actp, sizeof(act));
945 		if (error)
946 			return (error);
947 	}
948 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
949 	if (oactp && !error)
950 		error = copyout(oactp, uap->oact, sizeof(oact));
951 	return (error);
952 }
953 
954 #ifdef COMPAT_FREEBSD4
955 #ifndef _SYS_SYSPROTO_H_
956 struct freebsd4_sigaction_args {
957 	int	sig;
958 	struct	sigaction *act;
959 	struct	sigaction *oact;
960 };
961 #endif
962 int
963 freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
964 {
965 	struct sigaction act, oact;
966 	struct sigaction *actp, *oactp;
967 	int error;
968 
969 	actp = (uap->act != NULL) ? &act : NULL;
970 	oactp = (uap->oact != NULL) ? &oact : NULL;
971 	if (actp) {
972 		error = copyin(uap->act, actp, sizeof(act));
973 		if (error)
974 			return (error);
975 	}
976 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
977 	if (oactp && !error)
978 		error = copyout(oactp, uap->oact, sizeof(oact));
979 	return (error);
980 }
981 #endif	/* COMAPT_FREEBSD4 */
982 
983 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
984 #ifndef _SYS_SYSPROTO_H_
985 struct osigaction_args {
986 	int	signum;
987 	struct	osigaction *nsa;
988 	struct	osigaction *osa;
989 };
990 #endif
991 int
992 osigaction(struct thread *td, struct osigaction_args *uap)
993 {
994 	struct osigaction sa;
995 	struct sigaction nsa, osa;
996 	struct sigaction *nsap, *osap;
997 	int error;
998 
999 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1000 		return (EINVAL);
1001 
1002 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
1003 	osap = (uap->osa != NULL) ? &osa : NULL;
1004 
1005 	if (nsap) {
1006 		error = copyin(uap->nsa, &sa, sizeof(sa));
1007 		if (error)
1008 			return (error);
1009 		nsap->sa_handler = sa.sa_handler;
1010 		nsap->sa_flags = sa.sa_flags;
1011 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
1012 	}
1013 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1014 	if (osap && !error) {
1015 		sa.sa_handler = osap->sa_handler;
1016 		sa.sa_flags = osap->sa_flags;
1017 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
1018 		error = copyout(&sa, uap->osa, sizeof(sa));
1019 	}
1020 	return (error);
1021 }
1022 
1023 #if !defined(__i386__)
1024 /* Avoid replicating the same stub everywhere */
1025 int
1026 osigreturn(struct thread *td, struct osigreturn_args *uap)
1027 {
1028 
1029 	return (nosys(td, (struct nosys_args *)uap));
1030 }
1031 #endif
1032 #endif /* COMPAT_43 */
1033 
1034 /*
1035  * Initialize signal state for process 0;
1036  * set to ignore signals that are ignored by default.
1037  */
1038 void
1039 siginit(struct proc *p)
1040 {
1041 	int i;
1042 	struct sigacts *ps;
1043 
1044 	PROC_LOCK(p);
1045 	ps = p->p_sigacts;
1046 	mtx_lock(&ps->ps_mtx);
1047 	for (i = 1; i <= NSIG; i++) {
1048 		if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) {
1049 			SIGADDSET(ps->ps_sigignore, i);
1050 		}
1051 	}
1052 	mtx_unlock(&ps->ps_mtx);
1053 	PROC_UNLOCK(p);
1054 }
1055 
1056 /*
1057  * Reset specified signal to the default disposition.
1058  */
1059 static void
1060 sigdflt(struct sigacts *ps, int sig)
1061 {
1062 
1063 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1064 	SIGDELSET(ps->ps_sigcatch, sig);
1065 	if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT)
1066 		SIGADDSET(ps->ps_sigignore, sig);
1067 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1068 	SIGDELSET(ps->ps_siginfo, sig);
1069 }
1070 
1071 /*
1072  * Reset signals for an exec of the specified process.
1073  */
1074 void
1075 execsigs(struct proc *p)
1076 {
1077 	struct sigacts *ps;
1078 	struct thread *td;
1079 
1080 	/*
1081 	 * Reset caught signals.  Held signals remain held
1082 	 * through td_sigmask (unless they were caught,
1083 	 * and are now ignored by default).
1084 	 */
1085 	PROC_LOCK_ASSERT(p, MA_OWNED);
1086 	ps = p->p_sigacts;
1087 	mtx_lock(&ps->ps_mtx);
1088 	sig_drop_caught(p);
1089 
1090 	/*
1091 	 * Reset stack state to the user stack.
1092 	 * Clear set of signals caught on the signal stack.
1093 	 */
1094 	td = curthread;
1095 	MPASS(td->td_proc == p);
1096 	td->td_sigstk.ss_flags = SS_DISABLE;
1097 	td->td_sigstk.ss_size = 0;
1098 	td->td_sigstk.ss_sp = 0;
1099 	td->td_pflags &= ~TDP_ALTSTACK;
1100 	/*
1101 	 * Reset no zombies if child dies flag as Solaris does.
1102 	 */
1103 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1104 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1105 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1106 	mtx_unlock(&ps->ps_mtx);
1107 }
1108 
1109 /*
1110  * kern_sigprocmask()
1111  *
1112  *	Manipulate signal mask.
1113  */
1114 int
1115 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1116     int flags)
1117 {
1118 	sigset_t new_block, oset1;
1119 	struct proc *p;
1120 	int error;
1121 
1122 	p = td->td_proc;
1123 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1124 		PROC_LOCK_ASSERT(p, MA_OWNED);
1125 	else
1126 		PROC_LOCK(p);
1127 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1128 	    ? MA_OWNED : MA_NOTOWNED);
1129 	if (oset != NULL)
1130 		*oset = td->td_sigmask;
1131 
1132 	error = 0;
1133 	if (set != NULL) {
1134 		switch (how) {
1135 		case SIG_BLOCK:
1136 			SIG_CANTMASK(*set);
1137 			oset1 = td->td_sigmask;
1138 			SIGSETOR(td->td_sigmask, *set);
1139 			new_block = td->td_sigmask;
1140 			SIGSETNAND(new_block, oset1);
1141 			break;
1142 		case SIG_UNBLOCK:
1143 			SIGSETNAND(td->td_sigmask, *set);
1144 			signotify(td);
1145 			goto out;
1146 		case SIG_SETMASK:
1147 			SIG_CANTMASK(*set);
1148 			oset1 = td->td_sigmask;
1149 			if (flags & SIGPROCMASK_OLD)
1150 				SIGSETLO(td->td_sigmask, *set);
1151 			else
1152 				td->td_sigmask = *set;
1153 			new_block = td->td_sigmask;
1154 			SIGSETNAND(new_block, oset1);
1155 			signotify(td);
1156 			break;
1157 		default:
1158 			error = EINVAL;
1159 			goto out;
1160 		}
1161 
1162 		/*
1163 		 * The new_block set contains signals that were not previously
1164 		 * blocked, but are blocked now.
1165 		 *
1166 		 * In case we block any signal that was not previously blocked
1167 		 * for td, and process has the signal pending, try to schedule
1168 		 * signal delivery to some thread that does not block the
1169 		 * signal, possibly waking it up.
1170 		 */
1171 		if (p->p_numthreads != 1)
1172 			reschedule_signals(p, new_block, flags);
1173 	}
1174 
1175 out:
1176 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1177 		PROC_UNLOCK(p);
1178 	return (error);
1179 }
1180 
1181 #ifndef _SYS_SYSPROTO_H_
1182 struct sigprocmask_args {
1183 	int	how;
1184 	const sigset_t *set;
1185 	sigset_t *oset;
1186 };
1187 #endif
1188 int
1189 sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1190 {
1191 	sigset_t set, oset;
1192 	sigset_t *setp, *osetp;
1193 	int error;
1194 
1195 	setp = (uap->set != NULL) ? &set : NULL;
1196 	osetp = (uap->oset != NULL) ? &oset : NULL;
1197 	if (setp) {
1198 		error = copyin(uap->set, setp, sizeof(set));
1199 		if (error)
1200 			return (error);
1201 	}
1202 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1203 	if (osetp && !error) {
1204 		error = copyout(osetp, uap->oset, sizeof(oset));
1205 	}
1206 	return (error);
1207 }
1208 
1209 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1210 #ifndef _SYS_SYSPROTO_H_
1211 struct osigprocmask_args {
1212 	int	how;
1213 	osigset_t mask;
1214 };
1215 #endif
1216 int
1217 osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1218 {
1219 	sigset_t set, oset;
1220 	int error;
1221 
1222 	OSIG2SIG(uap->mask, set);
1223 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1224 	SIG2OSIG(oset, td->td_retval[0]);
1225 	return (error);
1226 }
1227 #endif /* COMPAT_43 */
1228 
1229 int
1230 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1231 {
1232 	ksiginfo_t ksi;
1233 	sigset_t set;
1234 	int error;
1235 
1236 	error = copyin(uap->set, &set, sizeof(set));
1237 	if (error) {
1238 		td->td_retval[0] = error;
1239 		return (0);
1240 	}
1241 
1242 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1243 	if (error) {
1244 		/*
1245 		 * sigwait() function shall not return EINTR, but
1246 		 * the syscall does.  Non-ancient libc provides the
1247 		 * wrapper which hides EINTR.  Otherwise, EINTR return
1248 		 * is used by libthr to handle required cancellation
1249 		 * point in the sigwait().
1250 		 */
1251 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1252 			return (ERESTART);
1253 		td->td_retval[0] = error;
1254 		return (0);
1255 	}
1256 
1257 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1258 	td->td_retval[0] = error;
1259 	return (0);
1260 }
1261 
1262 int
1263 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1264 {
1265 	struct timespec ts;
1266 	struct timespec *timeout;
1267 	sigset_t set;
1268 	ksiginfo_t ksi;
1269 	int error;
1270 
1271 	if (uap->timeout) {
1272 		error = copyin(uap->timeout, &ts, sizeof(ts));
1273 		if (error)
1274 			return (error);
1275 
1276 		timeout = &ts;
1277 	} else
1278 		timeout = NULL;
1279 
1280 	error = copyin(uap->set, &set, sizeof(set));
1281 	if (error)
1282 		return (error);
1283 
1284 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1285 	if (error)
1286 		return (error);
1287 
1288 	if (uap->info)
1289 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1290 
1291 	if (error == 0)
1292 		td->td_retval[0] = ksi.ksi_signo;
1293 	return (error);
1294 }
1295 
1296 int
1297 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1298 {
1299 	ksiginfo_t ksi;
1300 	sigset_t set;
1301 	int error;
1302 
1303 	error = copyin(uap->set, &set, sizeof(set));
1304 	if (error)
1305 		return (error);
1306 
1307 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1308 	if (error)
1309 		return (error);
1310 
1311 	if (uap->info)
1312 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1313 
1314 	if (error == 0)
1315 		td->td_retval[0] = ksi.ksi_signo;
1316 	return (error);
1317 }
1318 
1319 static void
1320 proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1321 {
1322 	struct thread *thr;
1323 
1324 	FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1325 		if (thr == td)
1326 			thr->td_si = *si;
1327 		else
1328 			thr->td_si.si_signo = 0;
1329 	}
1330 }
1331 
1332 int
1333 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1334 	struct timespec *timeout)
1335 {
1336 	struct sigacts *ps;
1337 	sigset_t saved_mask, new_block;
1338 	struct proc *p;
1339 	int error, sig, timevalid = 0;
1340 	sbintime_t sbt, precision, tsbt;
1341 	struct timespec ts;
1342 	bool traced;
1343 
1344 	p = td->td_proc;
1345 	error = 0;
1346 	traced = false;
1347 
1348 	/* Ensure the sigfastblock value is up to date. */
1349 	sigfastblock_fetch(td);
1350 
1351 	if (timeout != NULL) {
1352 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1353 			timevalid = 1;
1354 			ts = *timeout;
1355 			if (ts.tv_sec < INT32_MAX / 2) {
1356 				tsbt = tstosbt(ts);
1357 				precision = tsbt;
1358 				precision >>= tc_precexp;
1359 				if (TIMESEL(&sbt, tsbt))
1360 					sbt += tc_tick_sbt;
1361 				sbt += tsbt;
1362 			} else
1363 				precision = sbt = 0;
1364 		}
1365 	} else
1366 		precision = sbt = 0;
1367 	ksiginfo_init(ksi);
1368 	/* Some signals can not be waited for. */
1369 	SIG_CANTMASK(waitset);
1370 	ps = p->p_sigacts;
1371 	PROC_LOCK(p);
1372 	saved_mask = td->td_sigmask;
1373 	SIGSETNAND(td->td_sigmask, waitset);
1374 	if ((p->p_sysent->sv_flags & SV_SIG_DISCIGN) != 0 ||
1375 	    !kern_sig_discard_ign) {
1376 		thread_lock(td);
1377 		td->td_flags |= TDF_SIGWAIT;
1378 		thread_unlock(td);
1379 	}
1380 	for (;;) {
1381 		mtx_lock(&ps->ps_mtx);
1382 		sig = cursig(td);
1383 		mtx_unlock(&ps->ps_mtx);
1384 		KASSERT(sig >= 0, ("sig %d", sig));
1385 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1386 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1387 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1388 				error = 0;
1389 				break;
1390 			}
1391 		}
1392 
1393 		if (error != 0)
1394 			break;
1395 
1396 		/*
1397 		 * POSIX says this must be checked after looking for pending
1398 		 * signals.
1399 		 */
1400 		if (timeout != NULL && !timevalid) {
1401 			error = EINVAL;
1402 			break;
1403 		}
1404 
1405 		if (traced) {
1406 			error = EINTR;
1407 			break;
1408 		}
1409 
1410 		error = msleep_sbt(&p->p_sigacts, &p->p_mtx, PPAUSE | PCATCH,
1411 		    "sigwait", sbt, precision, C_ABSOLUTE);
1412 
1413 		/* The syscalls can not be restarted. */
1414 		if (error == ERESTART)
1415 			error = EINTR;
1416 
1417 		/*
1418 		 * If PTRACE_SCE or PTRACE_SCX were set after
1419 		 * userspace entered the syscall, return spurious
1420 		 * EINTR after wait was done.  Only do this as last
1421 		 * resort after rechecking for possible queued signals
1422 		 * and expired timeouts.
1423 		 */
1424 		if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1425 			traced = true;
1426 	}
1427 	thread_lock(td);
1428 	td->td_flags &= ~TDF_SIGWAIT;
1429 	thread_unlock(td);
1430 
1431 	new_block = saved_mask;
1432 	SIGSETNAND(new_block, td->td_sigmask);
1433 	td->td_sigmask = saved_mask;
1434 	/*
1435 	 * Fewer signals can be delivered to us, reschedule signal
1436 	 * notification.
1437 	 */
1438 	if (p->p_numthreads != 1)
1439 		reschedule_signals(p, new_block, 0);
1440 
1441 	if (error == 0) {
1442 		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1443 
1444 		if (ksi->ksi_code == SI_TIMER)
1445 			itimer_accept(p, ksi->ksi_timerid, ksi);
1446 
1447 #ifdef KTRACE
1448 		if (KTRPOINT(td, KTR_PSIG)) {
1449 			sig_t action;
1450 
1451 			mtx_lock(&ps->ps_mtx);
1452 			action = ps->ps_sigact[_SIG_IDX(sig)];
1453 			mtx_unlock(&ps->ps_mtx);
1454 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1455 		}
1456 #endif
1457 		if (sig == SIGKILL) {
1458 			proc_td_siginfo_capture(td, &ksi->ksi_info);
1459 			sigexit(td, sig);
1460 		}
1461 	}
1462 	PROC_UNLOCK(p);
1463 	return (error);
1464 }
1465 
1466 #ifndef _SYS_SYSPROTO_H_
1467 struct sigpending_args {
1468 	sigset_t	*set;
1469 };
1470 #endif
1471 int
1472 sys_sigpending(struct thread *td, struct sigpending_args *uap)
1473 {
1474 	struct proc *p = td->td_proc;
1475 	sigset_t pending;
1476 
1477 	PROC_LOCK(p);
1478 	pending = p->p_sigqueue.sq_signals;
1479 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1480 	PROC_UNLOCK(p);
1481 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1482 }
1483 
1484 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1485 #ifndef _SYS_SYSPROTO_H_
1486 struct osigpending_args {
1487 	int	dummy;
1488 };
1489 #endif
1490 int
1491 osigpending(struct thread *td, struct osigpending_args *uap)
1492 {
1493 	struct proc *p = td->td_proc;
1494 	sigset_t pending;
1495 
1496 	PROC_LOCK(p);
1497 	pending = p->p_sigqueue.sq_signals;
1498 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1499 	PROC_UNLOCK(p);
1500 	SIG2OSIG(pending, td->td_retval[0]);
1501 	return (0);
1502 }
1503 #endif /* COMPAT_43 */
1504 
1505 #if defined(COMPAT_43)
1506 /*
1507  * Generalized interface signal handler, 4.3-compatible.
1508  */
1509 #ifndef _SYS_SYSPROTO_H_
1510 struct osigvec_args {
1511 	int	signum;
1512 	struct	sigvec *nsv;
1513 	struct	sigvec *osv;
1514 };
1515 #endif
1516 /* ARGSUSED */
1517 int
1518 osigvec(struct thread *td, struct osigvec_args *uap)
1519 {
1520 	struct sigvec vec;
1521 	struct sigaction nsa, osa;
1522 	struct sigaction *nsap, *osap;
1523 	int error;
1524 
1525 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1526 		return (EINVAL);
1527 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1528 	osap = (uap->osv != NULL) ? &osa : NULL;
1529 	if (nsap) {
1530 		error = copyin(uap->nsv, &vec, sizeof(vec));
1531 		if (error)
1532 			return (error);
1533 		nsap->sa_handler = vec.sv_handler;
1534 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1535 		nsap->sa_flags = vec.sv_flags;
1536 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1537 	}
1538 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1539 	if (osap && !error) {
1540 		vec.sv_handler = osap->sa_handler;
1541 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1542 		vec.sv_flags = osap->sa_flags;
1543 		vec.sv_flags &= ~SA_NOCLDWAIT;
1544 		vec.sv_flags ^= SA_RESTART;
1545 		error = copyout(&vec, uap->osv, sizeof(vec));
1546 	}
1547 	return (error);
1548 }
1549 
1550 #ifndef _SYS_SYSPROTO_H_
1551 struct osigblock_args {
1552 	int	mask;
1553 };
1554 #endif
1555 int
1556 osigblock(struct thread *td, struct osigblock_args *uap)
1557 {
1558 	sigset_t set, oset;
1559 
1560 	OSIG2SIG(uap->mask, set);
1561 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1562 	SIG2OSIG(oset, td->td_retval[0]);
1563 	return (0);
1564 }
1565 
1566 #ifndef _SYS_SYSPROTO_H_
1567 struct osigsetmask_args {
1568 	int	mask;
1569 };
1570 #endif
1571 int
1572 osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1573 {
1574 	sigset_t set, oset;
1575 
1576 	OSIG2SIG(uap->mask, set);
1577 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1578 	SIG2OSIG(oset, td->td_retval[0]);
1579 	return (0);
1580 }
1581 #endif /* COMPAT_43 */
1582 
1583 /*
1584  * Suspend calling thread until signal, providing mask to be set in the
1585  * meantime.
1586  */
1587 #ifndef _SYS_SYSPROTO_H_
1588 struct sigsuspend_args {
1589 	const sigset_t *sigmask;
1590 };
1591 #endif
1592 /* ARGSUSED */
1593 int
1594 sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1595 {
1596 	sigset_t mask;
1597 	int error;
1598 
1599 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1600 	if (error)
1601 		return (error);
1602 	return (kern_sigsuspend(td, mask));
1603 }
1604 
1605 int
1606 kern_sigsuspend(struct thread *td, sigset_t mask)
1607 {
1608 	struct proc *p = td->td_proc;
1609 	int has_sig, sig;
1610 
1611 	/* Ensure the sigfastblock value is up to date. */
1612 	sigfastblock_fetch(td);
1613 
1614 	/*
1615 	 * When returning from sigsuspend, we want
1616 	 * the old mask to be restored after the
1617 	 * signal handler has finished.  Thus, we
1618 	 * save it here and mark the sigacts structure
1619 	 * to indicate this.
1620 	 */
1621 	PROC_LOCK(p);
1622 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1623 	    SIGPROCMASK_PROC_LOCKED);
1624 	td->td_pflags |= TDP_OLDMASK;
1625 	ast_sched(td, TDA_SIGSUSPEND);
1626 
1627 	/*
1628 	 * Process signals now. Otherwise, we can get spurious wakeup
1629 	 * due to signal entered process queue, but delivered to other
1630 	 * thread. But sigsuspend should return only on signal
1631 	 * delivery.
1632 	 */
1633 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1634 	for (has_sig = 0; !has_sig;) {
1635 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1636 			0) == 0)
1637 			/* void */;
1638 		thread_suspend_check(0);
1639 		mtx_lock(&p->p_sigacts->ps_mtx);
1640 		while ((sig = cursig(td)) != 0) {
1641 			KASSERT(sig >= 0, ("sig %d", sig));
1642 			has_sig += postsig(sig);
1643 		}
1644 		mtx_unlock(&p->p_sigacts->ps_mtx);
1645 
1646 		/*
1647 		 * If PTRACE_SCE or PTRACE_SCX were set after
1648 		 * userspace entered the syscall, return spurious
1649 		 * EINTR.
1650 		 */
1651 		if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1652 			has_sig += 1;
1653 	}
1654 	PROC_UNLOCK(p);
1655 	td->td_errno = EINTR;
1656 	td->td_pflags |= TDP_NERRNO;
1657 	return (EJUSTRETURN);
1658 }
1659 
1660 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1661 /*
1662  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1663  * convention: libc stub passes mask, not pointer, to save a copyin.
1664  */
1665 #ifndef _SYS_SYSPROTO_H_
1666 struct osigsuspend_args {
1667 	osigset_t mask;
1668 };
1669 #endif
1670 /* ARGSUSED */
1671 int
1672 osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1673 {
1674 	sigset_t mask;
1675 
1676 	OSIG2SIG(uap->mask, mask);
1677 	return (kern_sigsuspend(td, mask));
1678 }
1679 #endif /* COMPAT_43 */
1680 
1681 #if defined(COMPAT_43)
1682 #ifndef _SYS_SYSPROTO_H_
1683 struct osigstack_args {
1684 	struct	sigstack *nss;
1685 	struct	sigstack *oss;
1686 };
1687 #endif
1688 /* ARGSUSED */
1689 int
1690 osigstack(struct thread *td, struct osigstack_args *uap)
1691 {
1692 	struct sigstack nss, oss;
1693 	int error = 0;
1694 
1695 	if (uap->nss != NULL) {
1696 		error = copyin(uap->nss, &nss, sizeof(nss));
1697 		if (error)
1698 			return (error);
1699 	}
1700 	oss.ss_sp = td->td_sigstk.ss_sp;
1701 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1702 	if (uap->nss != NULL) {
1703 		td->td_sigstk.ss_sp = nss.ss_sp;
1704 		td->td_sigstk.ss_size = 0;
1705 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1706 		td->td_pflags |= TDP_ALTSTACK;
1707 	}
1708 	if (uap->oss != NULL)
1709 		error = copyout(&oss, uap->oss, sizeof(oss));
1710 
1711 	return (error);
1712 }
1713 #endif /* COMPAT_43 */
1714 
1715 #ifndef _SYS_SYSPROTO_H_
1716 struct sigaltstack_args {
1717 	stack_t	*ss;
1718 	stack_t	*oss;
1719 };
1720 #endif
1721 /* ARGSUSED */
1722 int
1723 sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1724 {
1725 	stack_t ss, oss;
1726 	int error;
1727 
1728 	if (uap->ss != NULL) {
1729 		error = copyin(uap->ss, &ss, sizeof(ss));
1730 		if (error)
1731 			return (error);
1732 	}
1733 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1734 	    (uap->oss != NULL) ? &oss : NULL);
1735 	if (error)
1736 		return (error);
1737 	if (uap->oss != NULL)
1738 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1739 	return (error);
1740 }
1741 
1742 int
1743 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1744 {
1745 	struct proc *p = td->td_proc;
1746 	int oonstack;
1747 
1748 	oonstack = sigonstack(cpu_getstack(td));
1749 
1750 	if (oss != NULL) {
1751 		*oss = td->td_sigstk;
1752 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1753 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1754 	}
1755 
1756 	if (ss != NULL) {
1757 		if (oonstack)
1758 			return (EPERM);
1759 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1760 			return (EINVAL);
1761 		if (!(ss->ss_flags & SS_DISABLE)) {
1762 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1763 				return (ENOMEM);
1764 
1765 			td->td_sigstk = *ss;
1766 			td->td_pflags |= TDP_ALTSTACK;
1767 		} else {
1768 			td->td_pflags &= ~TDP_ALTSTACK;
1769 		}
1770 	}
1771 	return (0);
1772 }
1773 
1774 struct killpg1_ctx {
1775 	struct thread *td;
1776 	ksiginfo_t *ksi;
1777 	int sig;
1778 	bool sent;
1779 	bool found;
1780 	int ret;
1781 };
1782 
1783 static void
1784 killpg1_sendsig_locked(struct proc *p, struct killpg1_ctx *arg)
1785 {
1786 	int err;
1787 
1788 	err = p_cansignal(arg->td, p, arg->sig);
1789 	if (err == 0 && arg->sig != 0)
1790 		pksignal(p, arg->sig, arg->ksi);
1791 	if (err != ESRCH)
1792 		arg->found = true;
1793 	if (err == 0)
1794 		arg->sent = true;
1795 	else if (arg->ret == 0 && err != ESRCH && err != EPERM)
1796 		arg->ret = err;
1797 }
1798 
1799 static void
1800 killpg1_sendsig(struct proc *p, bool notself, struct killpg1_ctx *arg)
1801 {
1802 
1803 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1804 	    (notself && p == arg->td->td_proc) || p->p_state == PRS_NEW)
1805 		return;
1806 
1807 	PROC_LOCK(p);
1808 	killpg1_sendsig_locked(p, arg);
1809 	PROC_UNLOCK(p);
1810 }
1811 
1812 static void
1813 kill_processes_prison_cb(struct proc *p, void *arg)
1814 {
1815 	struct killpg1_ctx *ctx = arg;
1816 
1817 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1818 	    (p == ctx->td->td_proc) || p->p_state == PRS_NEW)
1819 		return;
1820 
1821 	killpg1_sendsig_locked(p, ctx);
1822 }
1823 
1824 /*
1825  * Common code for kill process group/broadcast kill.
1826  * td is the calling thread, as usual.
1827  */
1828 static int
1829 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1830 {
1831 	struct proc *p;
1832 	struct pgrp *pgrp;
1833 	struct killpg1_ctx arg;
1834 
1835 	arg.td = td;
1836 	arg.ksi = ksi;
1837 	arg.sig = sig;
1838 	arg.sent = false;
1839 	arg.found = false;
1840 	arg.ret = 0;
1841 	if (all) {
1842 		/*
1843 		 * broadcast
1844 		 */
1845 		prison_proc_iterate(td->td_ucred->cr_prison,
1846 		    kill_processes_prison_cb, &arg);
1847 	} else {
1848 again:
1849 		sx_slock(&proctree_lock);
1850 		if (pgid == 0) {
1851 			/*
1852 			 * zero pgid means send to my process group.
1853 			 */
1854 			pgrp = td->td_proc->p_pgrp;
1855 			PGRP_LOCK(pgrp);
1856 		} else {
1857 			pgrp = pgfind(pgid);
1858 			if (pgrp == NULL) {
1859 				sx_sunlock(&proctree_lock);
1860 				return (ESRCH);
1861 			}
1862 		}
1863 		sx_sunlock(&proctree_lock);
1864 		if (!sx_try_xlock(&pgrp->pg_killsx)) {
1865 			PGRP_UNLOCK(pgrp);
1866 			sx_xlock(&pgrp->pg_killsx);
1867 			sx_xunlock(&pgrp->pg_killsx);
1868 			goto again;
1869 		}
1870 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1871 			killpg1_sendsig(p, false, &arg);
1872 		}
1873 		PGRP_UNLOCK(pgrp);
1874 		sx_xunlock(&pgrp->pg_killsx);
1875 	}
1876 	MPASS(arg.ret != 0 || arg.found || !arg.sent);
1877 	if (arg.ret == 0 && !arg.sent)
1878 		arg.ret = arg.found ? EPERM : ESRCH;
1879 	return (arg.ret);
1880 }
1881 
1882 #ifndef _SYS_SYSPROTO_H_
1883 struct kill_args {
1884 	int	pid;
1885 	int	signum;
1886 };
1887 #endif
1888 /* ARGSUSED */
1889 int
1890 sys_kill(struct thread *td, struct kill_args *uap)
1891 {
1892 
1893 	return (kern_kill(td, uap->pid, uap->signum));
1894 }
1895 
1896 int
1897 kern_kill(struct thread *td, pid_t pid, int signum)
1898 {
1899 	ksiginfo_t ksi;
1900 	struct proc *p;
1901 	int error;
1902 
1903 	/*
1904 	 * A process in capability mode can send signals only to himself.
1905 	 * The main rationale behind this is that abort(3) is implemented as
1906 	 * kill(getpid(), SIGABRT).
1907 	 */
1908 	if (pid != td->td_proc->p_pid) {
1909 		if (CAP_TRACING(td))
1910 			ktrcapfail(CAPFAIL_SIGNAL, &signum);
1911 		if (IN_CAPABILITY_MODE(td))
1912 			return (ECAPMODE);
1913 	}
1914 
1915 	AUDIT_ARG_SIGNUM(signum);
1916 	AUDIT_ARG_PID(pid);
1917 	if ((u_int)signum > _SIG_MAXSIG)
1918 		return (EINVAL);
1919 
1920 	ksiginfo_init(&ksi);
1921 	ksi.ksi_signo = signum;
1922 	ksi.ksi_code = SI_USER;
1923 	ksi.ksi_pid = td->td_proc->p_pid;
1924 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1925 
1926 	if (pid > 0) {
1927 		/* kill single process */
1928 		if ((p = pfind_any(pid)) == NULL)
1929 			return (ESRCH);
1930 		AUDIT_ARG_PROCESS(p);
1931 		error = p_cansignal(td, p, signum);
1932 		if (error == 0 && signum)
1933 			pksignal(p, signum, &ksi);
1934 		PROC_UNLOCK(p);
1935 		return (error);
1936 	}
1937 	switch (pid) {
1938 	case -1:		/* broadcast signal */
1939 		return (killpg1(td, signum, 0, 1, &ksi));
1940 	case 0:			/* signal own process group */
1941 		return (killpg1(td, signum, 0, 0, &ksi));
1942 	default:		/* negative explicit process group */
1943 		return (killpg1(td, signum, -pid, 0, &ksi));
1944 	}
1945 	/* NOTREACHED */
1946 }
1947 
1948 int
1949 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1950 {
1951 	struct proc *p;
1952 	int error;
1953 
1954 	AUDIT_ARG_SIGNUM(uap->signum);
1955 	AUDIT_ARG_FD(uap->fd);
1956 	if ((u_int)uap->signum > _SIG_MAXSIG)
1957 		return (EINVAL);
1958 
1959 	error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1960 	if (error)
1961 		return (error);
1962 	AUDIT_ARG_PROCESS(p);
1963 	error = p_cansignal(td, p, uap->signum);
1964 	if (error == 0 && uap->signum)
1965 		kern_psignal(p, uap->signum);
1966 	PROC_UNLOCK(p);
1967 	return (error);
1968 }
1969 
1970 #if defined(COMPAT_43)
1971 #ifndef _SYS_SYSPROTO_H_
1972 struct okillpg_args {
1973 	int	pgid;
1974 	int	signum;
1975 };
1976 #endif
1977 /* ARGSUSED */
1978 int
1979 okillpg(struct thread *td, struct okillpg_args *uap)
1980 {
1981 	ksiginfo_t ksi;
1982 
1983 	AUDIT_ARG_SIGNUM(uap->signum);
1984 	AUDIT_ARG_PID(uap->pgid);
1985 	if ((u_int)uap->signum > _SIG_MAXSIG)
1986 		return (EINVAL);
1987 
1988 	ksiginfo_init(&ksi);
1989 	ksi.ksi_signo = uap->signum;
1990 	ksi.ksi_code = SI_USER;
1991 	ksi.ksi_pid = td->td_proc->p_pid;
1992 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1993 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1994 }
1995 #endif /* COMPAT_43 */
1996 
1997 #ifndef _SYS_SYSPROTO_H_
1998 struct sigqueue_args {
1999 	pid_t pid;
2000 	int signum;
2001 	/* union sigval */ void *value;
2002 };
2003 #endif
2004 int
2005 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
2006 {
2007 	union sigval sv;
2008 
2009 	sv.sival_ptr = uap->value;
2010 
2011 	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
2012 }
2013 
2014 int
2015 kern_sigqueue(struct thread *td, pid_t pid, int signumf, union sigval *value)
2016 {
2017 	ksiginfo_t ksi;
2018 	struct proc *p;
2019 	struct thread *td2;
2020 	u_int signum;
2021 	int error;
2022 
2023 	signum = signumf & ~__SIGQUEUE_TID;
2024 	if (signum > _SIG_MAXSIG)
2025 		return (EINVAL);
2026 
2027 	/*
2028 	 * Specification says sigqueue can only send signal to
2029 	 * single process.
2030 	 */
2031 	if (pid <= 0)
2032 		return (EINVAL);
2033 
2034 	if ((signumf & __SIGQUEUE_TID) == 0) {
2035 		if ((p = pfind_any(pid)) == NULL)
2036 			return (ESRCH);
2037 		td2 = NULL;
2038 	} else {
2039 		p = td->td_proc;
2040 		td2 = tdfind((lwpid_t)pid, p->p_pid);
2041 		if (td2 == NULL)
2042 			return (ESRCH);
2043 	}
2044 
2045 	error = p_cansignal(td, p, signum);
2046 	if (error == 0 && signum != 0) {
2047 		ksiginfo_init(&ksi);
2048 		ksi.ksi_flags = KSI_SIGQ;
2049 		ksi.ksi_signo = signum;
2050 		ksi.ksi_code = SI_QUEUE;
2051 		ksi.ksi_pid = td->td_proc->p_pid;
2052 		ksi.ksi_uid = td->td_ucred->cr_ruid;
2053 		ksi.ksi_value = *value;
2054 		error = tdsendsignal(p, td2, ksi.ksi_signo, &ksi);
2055 	}
2056 	PROC_UNLOCK(p);
2057 	return (error);
2058 }
2059 
2060 /*
2061  * Send a signal to a process group.  If checktty is 1,
2062  * limit to members which have a controlling terminal.
2063  */
2064 void
2065 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
2066 {
2067 	struct proc *p;
2068 
2069 	if (pgrp) {
2070 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
2071 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
2072 			PROC_LOCK(p);
2073 			if (p->p_state == PRS_NORMAL &&
2074 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
2075 				pksignal(p, sig, ksi);
2076 			PROC_UNLOCK(p);
2077 		}
2078 	}
2079 }
2080 
2081 /*
2082  * Recalculate the signal mask and reset the signal disposition after
2083  * usermode frame for delivery is formed.  Should be called after
2084  * mach-specific routine, because sysent->sv_sendsig() needs correct
2085  * ps_siginfo and signal mask.
2086  */
2087 static void
2088 postsig_done(int sig, struct thread *td, struct sigacts *ps)
2089 {
2090 	sigset_t mask;
2091 
2092 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2093 	td->td_ru.ru_nsignals++;
2094 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
2095 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
2096 		SIGADDSET(mask, sig);
2097 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
2098 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
2099 	if (SIGISMEMBER(ps->ps_sigreset, sig))
2100 		sigdflt(ps, sig);
2101 }
2102 
2103 /*
2104  * Send a signal caused by a trap to the current thread.  If it will be
2105  * caught immediately, deliver it with correct code.  Otherwise, post it
2106  * normally.
2107  */
2108 void
2109 trapsignal(struct thread *td, ksiginfo_t *ksi)
2110 {
2111 	struct sigacts *ps;
2112 	struct proc *p;
2113 	sigset_t sigmask;
2114 	int sig;
2115 
2116 	p = td->td_proc;
2117 	sig = ksi->ksi_signo;
2118 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
2119 
2120 	sigfastblock_fetch(td);
2121 	PROC_LOCK(p);
2122 	ps = p->p_sigacts;
2123 	mtx_lock(&ps->ps_mtx);
2124 	sigmask = td->td_sigmask;
2125 	if (td->td_sigblock_val != 0)
2126 		SIGSETOR(sigmask, fastblock_mask);
2127 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2128 	    !SIGISMEMBER(sigmask, sig)) {
2129 #ifdef KTRACE
2130 		if (KTRPOINT(curthread, KTR_PSIG))
2131 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2132 			    &td->td_sigmask, ksi->ksi_code);
2133 #endif
2134 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2135 		    ksi, &td->td_sigmask);
2136 		postsig_done(sig, td, ps);
2137 		mtx_unlock(&ps->ps_mtx);
2138 	} else {
2139 		/*
2140 		 * Avoid a possible infinite loop if the thread
2141 		 * masking the signal or process is ignoring the
2142 		 * signal.
2143 		 */
2144 		if (kern_forcesigexit && (SIGISMEMBER(sigmask, sig) ||
2145 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2146 			SIGDELSET(td->td_sigmask, sig);
2147 			SIGDELSET(ps->ps_sigcatch, sig);
2148 			SIGDELSET(ps->ps_sigignore, sig);
2149 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2150 			td->td_pflags &= ~TDP_SIGFASTBLOCK;
2151 			td->td_sigblock_val = 0;
2152 		}
2153 		mtx_unlock(&ps->ps_mtx);
2154 		p->p_sig = sig;		/* XXX to verify code */
2155 		tdsendsignal(p, td, sig, ksi);
2156 	}
2157 	PROC_UNLOCK(p);
2158 }
2159 
2160 static struct thread *
2161 sigtd(struct proc *p, int sig, bool fast_sigblock)
2162 {
2163 	struct thread *td, *signal_td;
2164 
2165 	PROC_LOCK_ASSERT(p, MA_OWNED);
2166 	MPASS(!fast_sigblock || p == curproc);
2167 
2168 	/*
2169 	 * Check if current thread can handle the signal without
2170 	 * switching context to another thread.
2171 	 */
2172 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig) &&
2173 	    (!fast_sigblock || curthread->td_sigblock_val == 0))
2174 		return (curthread);
2175 
2176 	/* Find a non-stopped thread that does not mask the signal. */
2177 	signal_td = NULL;
2178 	FOREACH_THREAD_IN_PROC(p, td) {
2179 		if (!SIGISMEMBER(td->td_sigmask, sig) && (!fast_sigblock ||
2180 		    td != curthread || td->td_sigblock_val == 0) &&
2181 		    (td->td_flags & TDF_BOUNDARY) == 0) {
2182 			signal_td = td;
2183 			break;
2184 		}
2185 	}
2186 	/* Select random (first) thread if no better match was found. */
2187 	if (signal_td == NULL)
2188 		signal_td = FIRST_THREAD_IN_PROC(p);
2189 	return (signal_td);
2190 }
2191 
2192 /*
2193  * Send the signal to the process.  If the signal has an action, the action
2194  * is usually performed by the target process rather than the caller; we add
2195  * the signal to the set of pending signals for the process.
2196  *
2197  * Exceptions:
2198  *   o When a stop signal is sent to a sleeping process that takes the
2199  *     default action, the process is stopped without awakening it.
2200  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2201  *     regardless of the signal action (eg, blocked or ignored).
2202  *
2203  * Other ignored signals are discarded immediately.
2204  *
2205  * NB: This function may be entered from the debugger via the "kill" DDB
2206  * command.  There is little that can be done to mitigate the possibly messy
2207  * side effects of this unwise possibility.
2208  */
2209 void
2210 kern_psignal(struct proc *p, int sig)
2211 {
2212 	ksiginfo_t ksi;
2213 
2214 	ksiginfo_init(&ksi);
2215 	ksi.ksi_signo = sig;
2216 	ksi.ksi_code = SI_KERNEL;
2217 	(void) tdsendsignal(p, NULL, sig, &ksi);
2218 }
2219 
2220 int
2221 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2222 {
2223 
2224 	return (tdsendsignal(p, NULL, sig, ksi));
2225 }
2226 
2227 /* Utility function for finding a thread to send signal event to. */
2228 int
2229 sigev_findtd(struct proc *p, struct sigevent *sigev, struct thread **ttd)
2230 {
2231 	struct thread *td;
2232 
2233 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2234 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2235 		if (td == NULL)
2236 			return (ESRCH);
2237 		*ttd = td;
2238 	} else {
2239 		*ttd = NULL;
2240 		PROC_LOCK(p);
2241 	}
2242 	return (0);
2243 }
2244 
2245 void
2246 tdsignal(struct thread *td, int sig)
2247 {
2248 	ksiginfo_t ksi;
2249 
2250 	ksiginfo_init(&ksi);
2251 	ksi.ksi_signo = sig;
2252 	ksi.ksi_code = SI_KERNEL;
2253 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2254 }
2255 
2256 void
2257 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2258 {
2259 
2260 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2261 }
2262 
2263 static int
2264 sig_sleepq_abort(struct thread *td, int intrval)
2265 {
2266 	THREAD_LOCK_ASSERT(td, MA_OWNED);
2267 
2268 	if (intrval == 0 && (td->td_flags & TDF_SIGWAIT) == 0) {
2269 		thread_unlock(td);
2270 		return (0);
2271 	}
2272 	return (sleepq_abort(td, intrval));
2273 }
2274 
2275 int
2276 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2277 {
2278 	sig_t action;
2279 	sigqueue_t *sigqueue;
2280 	int prop;
2281 	struct sigacts *ps;
2282 	int intrval;
2283 	int ret = 0;
2284 	int wakeup_swapper;
2285 
2286 	MPASS(td == NULL || p == td->td_proc);
2287 	PROC_LOCK_ASSERT(p, MA_OWNED);
2288 
2289 	if (!_SIG_VALID(sig))
2290 		panic("%s(): invalid signal %d", __func__, sig);
2291 
2292 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2293 
2294 	/*
2295 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2296 	 */
2297 	if (p->p_state == PRS_ZOMBIE) {
2298 		if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2299 			ksiginfo_tryfree(ksi);
2300 		return (ret);
2301 	}
2302 
2303 	ps = p->p_sigacts;
2304 	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2305 	prop = sigprop(sig);
2306 
2307 	if (td == NULL) {
2308 		td = sigtd(p, sig, false);
2309 		sigqueue = &p->p_sigqueue;
2310 	} else
2311 		sigqueue = &td->td_sigqueue;
2312 
2313 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2314 
2315 	/*
2316 	 * If the signal is being ignored, then we forget about it
2317 	 * immediately, except when the target process executes
2318 	 * sigwait().  (Note: we don't set SIGCONT in ps_sigignore,
2319 	 * and if it is set to SIG_IGN, action will be SIG_DFL here.)
2320 	 */
2321 	mtx_lock(&ps->ps_mtx);
2322 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2323 		if (kern_sig_discard_ign &&
2324 		    (p->p_sysent->sv_flags & SV_SIG_DISCIGN) == 0) {
2325 			SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2326 
2327 			mtx_unlock(&ps->ps_mtx);
2328 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2329 				ksiginfo_tryfree(ksi);
2330 			return (ret);
2331 		} else {
2332 			action = SIG_CATCH;
2333 			intrval = 0;
2334 		}
2335 	} else {
2336 		if (SIGISMEMBER(td->td_sigmask, sig))
2337 			action = SIG_HOLD;
2338 		else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2339 			action = SIG_CATCH;
2340 		else
2341 			action = SIG_DFL;
2342 		if (SIGISMEMBER(ps->ps_sigintr, sig))
2343 			intrval = EINTR;
2344 		else
2345 			intrval = ERESTART;
2346 	}
2347 	mtx_unlock(&ps->ps_mtx);
2348 
2349 	if (prop & SIGPROP_CONT)
2350 		sigqueue_delete_stopmask_proc(p);
2351 	else if (prop & SIGPROP_STOP) {
2352 		/*
2353 		 * If sending a tty stop signal to a member of an orphaned
2354 		 * process group, discard the signal here if the action
2355 		 * is default; don't stop the process below if sleeping,
2356 		 * and don't clear any pending SIGCONT.
2357 		 */
2358 		if ((prop & SIGPROP_TTYSTOP) != 0 &&
2359 		    (p->p_pgrp->pg_flags & PGRP_ORPHANED) != 0 &&
2360 		    action == SIG_DFL) {
2361 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2362 				ksiginfo_tryfree(ksi);
2363 			return (ret);
2364 		}
2365 		sigqueue_delete_proc(p, SIGCONT);
2366 		if (p->p_flag & P_CONTINUED) {
2367 			p->p_flag &= ~P_CONTINUED;
2368 			PROC_LOCK(p->p_pptr);
2369 			sigqueue_take(p->p_ksi);
2370 			PROC_UNLOCK(p->p_pptr);
2371 		}
2372 	}
2373 
2374 	ret = sigqueue_add(sigqueue, sig, ksi);
2375 	if (ret != 0)
2376 		return (ret);
2377 	signotify(td);
2378 	/*
2379 	 * Defer further processing for signals which are held,
2380 	 * except that stopped processes must be continued by SIGCONT.
2381 	 */
2382 	if (action == SIG_HOLD &&
2383 	    !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2384 		return (ret);
2385 
2386 	wakeup_swapper = 0;
2387 
2388 	/*
2389 	 * Some signals have a process-wide effect and a per-thread
2390 	 * component.  Most processing occurs when the process next
2391 	 * tries to cross the user boundary, however there are some
2392 	 * times when processing needs to be done immediately, such as
2393 	 * waking up threads so that they can cross the user boundary.
2394 	 * We try to do the per-process part here.
2395 	 */
2396 	if (P_SHOULDSTOP(p)) {
2397 		KASSERT(!(p->p_flag & P_WEXIT),
2398 		    ("signal to stopped but exiting process"));
2399 		if (sig == SIGKILL) {
2400 			/*
2401 			 * If traced process is already stopped,
2402 			 * then no further action is necessary.
2403 			 */
2404 			if (p->p_flag & P_TRACED)
2405 				goto out;
2406 			/*
2407 			 * SIGKILL sets process running.
2408 			 * It will die elsewhere.
2409 			 * All threads must be restarted.
2410 			 */
2411 			p->p_flag &= ~P_STOPPED_SIG;
2412 			goto runfast;
2413 		}
2414 
2415 		if (prop & SIGPROP_CONT) {
2416 			/*
2417 			 * If traced process is already stopped,
2418 			 * then no further action is necessary.
2419 			 */
2420 			if (p->p_flag & P_TRACED)
2421 				goto out;
2422 			/*
2423 			 * If SIGCONT is default (or ignored), we continue the
2424 			 * process but don't leave the signal in sigqueue as
2425 			 * it has no further action.  If SIGCONT is held, we
2426 			 * continue the process and leave the signal in
2427 			 * sigqueue.  If the process catches SIGCONT, let it
2428 			 * handle the signal itself.  If it isn't waiting on
2429 			 * an event, it goes back to run state.
2430 			 * Otherwise, process goes back to sleep state.
2431 			 */
2432 			p->p_flag &= ~P_STOPPED_SIG;
2433 			PROC_SLOCK(p);
2434 			if (p->p_numthreads == p->p_suspcount) {
2435 				PROC_SUNLOCK(p);
2436 				p->p_flag |= P_CONTINUED;
2437 				p->p_xsig = SIGCONT;
2438 				PROC_LOCK(p->p_pptr);
2439 				childproc_continued(p);
2440 				PROC_UNLOCK(p->p_pptr);
2441 				PROC_SLOCK(p);
2442 			}
2443 			if (action == SIG_DFL) {
2444 				thread_unsuspend(p);
2445 				PROC_SUNLOCK(p);
2446 				sigqueue_delete(sigqueue, sig);
2447 				goto out_cont;
2448 			}
2449 			if (action == SIG_CATCH) {
2450 				/*
2451 				 * The process wants to catch it so it needs
2452 				 * to run at least one thread, but which one?
2453 				 */
2454 				PROC_SUNLOCK(p);
2455 				goto runfast;
2456 			}
2457 			/*
2458 			 * The signal is not ignored or caught.
2459 			 */
2460 			thread_unsuspend(p);
2461 			PROC_SUNLOCK(p);
2462 			goto out_cont;
2463 		}
2464 
2465 		if (prop & SIGPROP_STOP) {
2466 			/*
2467 			 * If traced process is already stopped,
2468 			 * then no further action is necessary.
2469 			 */
2470 			if (p->p_flag & P_TRACED)
2471 				goto out;
2472 			/*
2473 			 * Already stopped, don't need to stop again
2474 			 * (If we did the shell could get confused).
2475 			 * Just make sure the signal STOP bit set.
2476 			 */
2477 			p->p_flag |= P_STOPPED_SIG;
2478 			sigqueue_delete(sigqueue, sig);
2479 			goto out;
2480 		}
2481 
2482 		/*
2483 		 * All other kinds of signals:
2484 		 * If a thread is sleeping interruptibly, simulate a
2485 		 * wakeup so that when it is continued it will be made
2486 		 * runnable and can look at the signal.  However, don't make
2487 		 * the PROCESS runnable, leave it stopped.
2488 		 * It may run a bit until it hits a thread_suspend_check().
2489 		 */
2490 		PROC_SLOCK(p);
2491 		thread_lock(td);
2492 		if (TD_CAN_ABORT(td))
2493 			wakeup_swapper = sig_sleepq_abort(td, intrval);
2494 		else
2495 			thread_unlock(td);
2496 		PROC_SUNLOCK(p);
2497 		goto out;
2498 		/*
2499 		 * Mutexes are short lived. Threads waiting on them will
2500 		 * hit thread_suspend_check() soon.
2501 		 */
2502 	} else if (p->p_state == PRS_NORMAL) {
2503 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2504 			tdsigwakeup(td, sig, action, intrval);
2505 			goto out;
2506 		}
2507 
2508 		MPASS(action == SIG_DFL);
2509 
2510 		if (prop & SIGPROP_STOP) {
2511 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2512 				goto out;
2513 			p->p_flag |= P_STOPPED_SIG;
2514 			p->p_xsig = sig;
2515 			PROC_SLOCK(p);
2516 			wakeup_swapper = sig_suspend_threads(td, p);
2517 			if (p->p_numthreads == p->p_suspcount) {
2518 				/*
2519 				 * only thread sending signal to another
2520 				 * process can reach here, if thread is sending
2521 				 * signal to its process, because thread does
2522 				 * not suspend itself here, p_numthreads
2523 				 * should never be equal to p_suspcount.
2524 				 */
2525 				thread_stopped(p);
2526 				PROC_SUNLOCK(p);
2527 				sigqueue_delete_proc(p, p->p_xsig);
2528 			} else
2529 				PROC_SUNLOCK(p);
2530 			goto out;
2531 		}
2532 	} else {
2533 		/* Not in "NORMAL" state. discard the signal. */
2534 		sigqueue_delete(sigqueue, sig);
2535 		goto out;
2536 	}
2537 
2538 	/*
2539 	 * The process is not stopped so we need to apply the signal to all the
2540 	 * running threads.
2541 	 */
2542 runfast:
2543 	tdsigwakeup(td, sig, action, intrval);
2544 	PROC_SLOCK(p);
2545 	thread_unsuspend(p);
2546 	PROC_SUNLOCK(p);
2547 out_cont:
2548 	itimer_proc_continue(p);
2549 	kqtimer_proc_continue(p);
2550 out:
2551 	/* If we jump here, proc slock should not be owned. */
2552 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2553 	if (wakeup_swapper)
2554 		kick_proc0();
2555 
2556 	return (ret);
2557 }
2558 
2559 /*
2560  * The force of a signal has been directed against a single
2561  * thread.  We need to see what we can do about knocking it
2562  * out of any sleep it may be in etc.
2563  */
2564 static void
2565 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2566 {
2567 	struct proc *p = td->td_proc;
2568 	int prop, wakeup_swapper;
2569 
2570 	PROC_LOCK_ASSERT(p, MA_OWNED);
2571 	prop = sigprop(sig);
2572 
2573 	PROC_SLOCK(p);
2574 	thread_lock(td);
2575 	/*
2576 	 * Bring the priority of a thread up if we want it to get
2577 	 * killed in this lifetime.  Be careful to avoid bumping the
2578 	 * priority of the idle thread, since we still allow to signal
2579 	 * kernel processes.
2580 	 */
2581 	if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2582 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2583 		sched_prio(td, PUSER);
2584 	if (TD_ON_SLEEPQ(td)) {
2585 		/*
2586 		 * If thread is sleeping uninterruptibly
2587 		 * we can't interrupt the sleep... the signal will
2588 		 * be noticed when the process returns through
2589 		 * trap() or syscall().
2590 		 */
2591 		if ((td->td_flags & TDF_SINTR) == 0)
2592 			goto out;
2593 		/*
2594 		 * If SIGCONT is default (or ignored) and process is
2595 		 * asleep, we are finished; the process should not
2596 		 * be awakened.
2597 		 */
2598 		if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2599 			thread_unlock(td);
2600 			PROC_SUNLOCK(p);
2601 			sigqueue_delete(&p->p_sigqueue, sig);
2602 			/*
2603 			 * It may be on either list in this state.
2604 			 * Remove from both for now.
2605 			 */
2606 			sigqueue_delete(&td->td_sigqueue, sig);
2607 			return;
2608 		}
2609 
2610 		/*
2611 		 * Don't awaken a sleeping thread for SIGSTOP if the
2612 		 * STOP signal is deferred.
2613 		 */
2614 		if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2615 		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2616 			goto out;
2617 
2618 		/*
2619 		 * Give low priority threads a better chance to run.
2620 		 */
2621 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2622 			sched_prio(td, PUSER);
2623 
2624 		wakeup_swapper = sig_sleepq_abort(td, intrval);
2625 		PROC_SUNLOCK(p);
2626 		if (wakeup_swapper)
2627 			kick_proc0();
2628 		return;
2629 	}
2630 
2631 	/*
2632 	 * Other states do nothing with the signal immediately,
2633 	 * other than kicking ourselves if we are running.
2634 	 * It will either never be noticed, or noticed very soon.
2635 	 */
2636 #ifdef SMP
2637 	if (TD_IS_RUNNING(td) && td != curthread)
2638 		forward_signal(td);
2639 #endif
2640 
2641 out:
2642 	PROC_SUNLOCK(p);
2643 	thread_unlock(td);
2644 }
2645 
2646 static void
2647 ptrace_coredumpreq(struct thread *td, struct proc *p,
2648     struct thr_coredump_req *tcq)
2649 {
2650 	void *rl_cookie;
2651 
2652 	if (p->p_sysent->sv_coredump == NULL) {
2653 		tcq->tc_error = ENOSYS;
2654 		return;
2655 	}
2656 
2657 	rl_cookie = vn_rangelock_wlock(tcq->tc_vp, 0, OFF_MAX);
2658 	tcq->tc_error = p->p_sysent->sv_coredump(td, tcq->tc_vp,
2659 	    tcq->tc_limit, tcq->tc_flags);
2660 	vn_rangelock_unlock(tcq->tc_vp, rl_cookie);
2661 }
2662 
2663 static void
2664 ptrace_syscallreq(struct thread *td, struct proc *p,
2665     struct thr_syscall_req *tsr)
2666 {
2667 	struct sysentvec *sv;
2668 	struct sysent *se;
2669 	register_t rv_saved[2];
2670 	int error, nerror;
2671 	int sc;
2672 	bool audited, sy_thr_static;
2673 
2674 	sv = p->p_sysent;
2675 	if (sv->sv_table == NULL || sv->sv_size < tsr->ts_sa.code) {
2676 		tsr->ts_ret.sr_error = ENOSYS;
2677 		return;
2678 	}
2679 
2680 	sc = tsr->ts_sa.code;
2681 	if (sc == SYS_syscall || sc == SYS___syscall) {
2682 		sc = tsr->ts_sa.args[0];
2683 		memmove(&tsr->ts_sa.args[0], &tsr->ts_sa.args[1],
2684 		    sizeof(register_t) * (tsr->ts_nargs - 1));
2685 	}
2686 
2687 	tsr->ts_sa.callp = se = &sv->sv_table[sc];
2688 
2689 	VM_CNT_INC(v_syscall);
2690 	td->td_pticks = 0;
2691 	if (__predict_false(td->td_cowgen != atomic_load_int(
2692 	    &td->td_proc->p_cowgen)))
2693 		thread_cow_update(td);
2694 
2695 	td->td_sa = tsr->ts_sa;
2696 
2697 #ifdef CAPABILITY_MODE
2698 	if ((se->sy_flags & SYF_CAPENABLED) == 0) {
2699 		if (CAP_TRACING(td))
2700 			ktrcapfail(CAPFAIL_SYSCALL, NULL);
2701 		if (IN_CAPABILITY_MODE(td)) {
2702 			tsr->ts_ret.sr_error = ECAPMODE;
2703 			return;
2704 		}
2705 	}
2706 #endif
2707 
2708 	sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2709 	audited = AUDIT_SYSCALL_ENTER(sc, td) != 0;
2710 
2711 	if (!sy_thr_static) {
2712 		error = syscall_thread_enter(td, &se);
2713 		sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2714 		if (error != 0) {
2715 			tsr->ts_ret.sr_error = error;
2716 			return;
2717 		}
2718 	}
2719 
2720 	rv_saved[0] = td->td_retval[0];
2721 	rv_saved[1] = td->td_retval[1];
2722 	nerror = td->td_errno;
2723 	td->td_retval[0] = 0;
2724 	td->td_retval[1] = 0;
2725 
2726 #ifdef KDTRACE_HOOKS
2727 	if (se->sy_entry != 0)
2728 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_ENTRY, 0);
2729 #endif
2730 	tsr->ts_ret.sr_error = se->sy_call(td, tsr->ts_sa.args);
2731 #ifdef KDTRACE_HOOKS
2732 	if (se->sy_return != 0)
2733 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_RETURN,
2734 		    tsr->ts_ret.sr_error != 0 ? -1 : td->td_retval[0]);
2735 #endif
2736 
2737 	tsr->ts_ret.sr_retval[0] = td->td_retval[0];
2738 	tsr->ts_ret.sr_retval[1] = td->td_retval[1];
2739 	td->td_retval[0] = rv_saved[0];
2740 	td->td_retval[1] = rv_saved[1];
2741 	td->td_errno = nerror;
2742 
2743 	if (audited)
2744 		AUDIT_SYSCALL_EXIT(error, td);
2745 	if (!sy_thr_static)
2746 		syscall_thread_exit(td, se);
2747 }
2748 
2749 static void
2750 ptrace_remotereq(struct thread *td, int flag)
2751 {
2752 	struct proc *p;
2753 
2754 	MPASS(td == curthread);
2755 	p = td->td_proc;
2756 	PROC_LOCK_ASSERT(p, MA_OWNED);
2757 	if ((td->td_dbgflags & flag) == 0)
2758 		return;
2759 	KASSERT((p->p_flag & P_STOPPED_TRACE) != 0, ("not stopped"));
2760 	KASSERT(td->td_remotereq != NULL, ("td_remotereq is NULL"));
2761 
2762 	PROC_UNLOCK(p);
2763 	switch (flag) {
2764 	case TDB_COREDUMPREQ:
2765 		ptrace_coredumpreq(td, p, td->td_remotereq);
2766 		break;
2767 	case TDB_SCREMOTEREQ:
2768 		ptrace_syscallreq(td, p, td->td_remotereq);
2769 		break;
2770 	default:
2771 		__unreachable();
2772 	}
2773 	PROC_LOCK(p);
2774 
2775 	MPASS((td->td_dbgflags & flag) != 0);
2776 	td->td_dbgflags &= ~flag;
2777 	td->td_remotereq = NULL;
2778 	wakeup(p);
2779 }
2780 
2781 static int
2782 sig_suspend_threads(struct thread *td, struct proc *p)
2783 {
2784 	struct thread *td2;
2785 	int wakeup_swapper;
2786 
2787 	PROC_LOCK_ASSERT(p, MA_OWNED);
2788 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2789 
2790 	wakeup_swapper = 0;
2791 	FOREACH_THREAD_IN_PROC(p, td2) {
2792 		thread_lock(td2);
2793 		ast_sched_locked(td2, TDA_SUSPEND);
2794 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2795 		    (td2->td_flags & TDF_SINTR)) {
2796 			if (td2->td_flags & TDF_SBDRY) {
2797 				/*
2798 				 * Once a thread is asleep with
2799 				 * TDF_SBDRY and without TDF_SERESTART
2800 				 * or TDF_SEINTR set, it should never
2801 				 * become suspended due to this check.
2802 				 */
2803 				KASSERT(!TD_IS_SUSPENDED(td2),
2804 				    ("thread with deferred stops suspended"));
2805 				if (TD_SBDRY_INTR(td2)) {
2806 					wakeup_swapper |= sleepq_abort(td2,
2807 					    TD_SBDRY_ERRNO(td2));
2808 					continue;
2809 				}
2810 			} else if (!TD_IS_SUSPENDED(td2))
2811 				thread_suspend_one(td2);
2812 		} else if (!TD_IS_SUSPENDED(td2)) {
2813 #ifdef SMP
2814 			if (TD_IS_RUNNING(td2) && td2 != td)
2815 				forward_signal(td2);
2816 #endif
2817 		}
2818 		thread_unlock(td2);
2819 	}
2820 	return (wakeup_swapper);
2821 }
2822 
2823 /*
2824  * Stop the process for an event deemed interesting to the debugger. If si is
2825  * non-NULL, this is a signal exchange; the new signal requested by the
2826  * debugger will be returned for handling. If si is NULL, this is some other
2827  * type of interesting event. The debugger may request a signal be delivered in
2828  * that case as well, however it will be deferred until it can be handled.
2829  */
2830 int
2831 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2832 {
2833 	struct proc *p = td->td_proc;
2834 	struct thread *td2;
2835 	ksiginfo_t ksi;
2836 
2837 	PROC_LOCK_ASSERT(p, MA_OWNED);
2838 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2839 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2840 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2841 
2842 	td->td_xsig = sig;
2843 
2844 	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2845 		td->td_dbgflags |= TDB_XSIG;
2846 		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2847 		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2848 		PROC_SLOCK(p);
2849 		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2850 			if (P_KILLED(p)) {
2851 				/*
2852 				 * Ensure that, if we've been PT_KILLed, the
2853 				 * exit status reflects that. Another thread
2854 				 * may also be in ptracestop(), having just
2855 				 * received the SIGKILL, but this thread was
2856 				 * unsuspended first.
2857 				 */
2858 				td->td_dbgflags &= ~TDB_XSIG;
2859 				td->td_xsig = SIGKILL;
2860 				p->p_ptevents = 0;
2861 				break;
2862 			}
2863 			if (p->p_flag & P_SINGLE_EXIT &&
2864 			    !(td->td_dbgflags & TDB_EXIT)) {
2865 				/*
2866 				 * Ignore ptrace stops except for thread exit
2867 				 * events when the process exits.
2868 				 */
2869 				td->td_dbgflags &= ~TDB_XSIG;
2870 				PROC_SUNLOCK(p);
2871 				return (0);
2872 			}
2873 
2874 			/*
2875 			 * Make wait(2) work.  Ensure that right after the
2876 			 * attach, the thread which was decided to become the
2877 			 * leader of attach gets reported to the waiter.
2878 			 * Otherwise, just avoid overwriting another thread's
2879 			 * assignment to p_xthread.  If another thread has
2880 			 * already set p_xthread, the current thread will get
2881 			 * a chance to report itself upon the next iteration.
2882 			 */
2883 			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2884 			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2885 			    p->p_xthread == NULL)) {
2886 				p->p_xsig = sig;
2887 				p->p_xthread = td;
2888 
2889 				/*
2890 				 * If we are on sleepqueue already,
2891 				 * let sleepqueue code decide if it
2892 				 * needs to go sleep after attach.
2893 				 */
2894 				if (td->td_wchan == NULL)
2895 					td->td_dbgflags &= ~TDB_FSTP;
2896 
2897 				p->p_flag2 &= ~P2_PTRACE_FSTP;
2898 				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2899 				sig_suspend_threads(td, p);
2900 			}
2901 			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2902 				td->td_dbgflags &= ~TDB_STOPATFORK;
2903 			}
2904 stopme:
2905 			td->td_dbgflags |= TDB_SSWITCH;
2906 			thread_suspend_switch(td, p);
2907 			td->td_dbgflags &= ~TDB_SSWITCH;
2908 			if ((td->td_dbgflags & (TDB_COREDUMPREQ |
2909 			    TDB_SCREMOTEREQ)) != 0) {
2910 				MPASS((td->td_dbgflags & (TDB_COREDUMPREQ |
2911 				    TDB_SCREMOTEREQ)) !=
2912 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2913 				PROC_SUNLOCK(p);
2914 				ptrace_remotereq(td, td->td_dbgflags &
2915 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2916 				PROC_SLOCK(p);
2917 				goto stopme;
2918 			}
2919 			if (p->p_xthread == td)
2920 				p->p_xthread = NULL;
2921 			if (!(p->p_flag & P_TRACED))
2922 				break;
2923 			if (td->td_dbgflags & TDB_SUSPEND) {
2924 				if (p->p_flag & P_SINGLE_EXIT)
2925 					break;
2926 				goto stopme;
2927 			}
2928 		}
2929 		PROC_SUNLOCK(p);
2930 	}
2931 
2932 	if (si != NULL && sig == td->td_xsig) {
2933 		/* Parent wants us to take the original signal unchanged. */
2934 		si->ksi_flags |= KSI_HEAD;
2935 		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2936 			si->ksi_signo = 0;
2937 	} else if (td->td_xsig != 0) {
2938 		/*
2939 		 * If parent wants us to take a new signal, then it will leave
2940 		 * it in td->td_xsig; otherwise we just look for signals again.
2941 		 */
2942 		ksiginfo_init(&ksi);
2943 		ksi.ksi_signo = td->td_xsig;
2944 		ksi.ksi_flags |= KSI_PTRACE;
2945 		td2 = sigtd(p, td->td_xsig, false);
2946 		tdsendsignal(p, td2, td->td_xsig, &ksi);
2947 		if (td != td2)
2948 			return (0);
2949 	}
2950 
2951 	return (td->td_xsig);
2952 }
2953 
2954 static void
2955 reschedule_signals(struct proc *p, sigset_t block, int flags)
2956 {
2957 	struct sigacts *ps;
2958 	struct thread *td;
2959 	int sig;
2960 	bool fastblk, pslocked;
2961 
2962 	PROC_LOCK_ASSERT(p, MA_OWNED);
2963 	ps = p->p_sigacts;
2964 	pslocked = (flags & SIGPROCMASK_PS_LOCKED) != 0;
2965 	mtx_assert(&ps->ps_mtx, pslocked ? MA_OWNED : MA_NOTOWNED);
2966 	if (SIGISEMPTY(p->p_siglist))
2967 		return;
2968 	SIGSETAND(block, p->p_siglist);
2969 	fastblk = (flags & SIGPROCMASK_FASTBLK) != 0;
2970 	SIG_FOREACH(sig, &block) {
2971 		td = sigtd(p, sig, fastblk);
2972 
2973 		/*
2974 		 * If sigtd() selected us despite sigfastblock is
2975 		 * blocking, do not activate AST or wake us, to avoid
2976 		 * loop in AST handler.
2977 		 */
2978 		if (fastblk && td == curthread)
2979 			continue;
2980 
2981 		signotify(td);
2982 		if (!pslocked)
2983 			mtx_lock(&ps->ps_mtx);
2984 		if (p->p_flag & P_TRACED ||
2985 		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2986 		    !SIGISMEMBER(td->td_sigmask, sig))) {
2987 			tdsigwakeup(td, sig, SIG_CATCH,
2988 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2989 			    ERESTART));
2990 		}
2991 		if (!pslocked)
2992 			mtx_unlock(&ps->ps_mtx);
2993 	}
2994 }
2995 
2996 void
2997 tdsigcleanup(struct thread *td)
2998 {
2999 	struct proc *p;
3000 	sigset_t unblocked;
3001 
3002 	p = td->td_proc;
3003 	PROC_LOCK_ASSERT(p, MA_OWNED);
3004 
3005 	sigqueue_flush(&td->td_sigqueue);
3006 	if (p->p_numthreads == 1)
3007 		return;
3008 
3009 	/*
3010 	 * Since we cannot handle signals, notify signal post code
3011 	 * about this by filling the sigmask.
3012 	 *
3013 	 * Also, if needed, wake up thread(s) that do not block the
3014 	 * same signals as the exiting thread, since the thread might
3015 	 * have been selected for delivery and woken up.
3016 	 */
3017 	SIGFILLSET(unblocked);
3018 	SIGSETNAND(unblocked, td->td_sigmask);
3019 	SIGFILLSET(td->td_sigmask);
3020 	reschedule_signals(p, unblocked, 0);
3021 
3022 }
3023 
3024 static int
3025 sigdeferstop_curr_flags(int cflags)
3026 {
3027 
3028 	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
3029 	    (cflags & TDF_SBDRY) != 0);
3030 	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
3031 }
3032 
3033 /*
3034  * Defer the delivery of SIGSTOP for the current thread, according to
3035  * the requested mode.  Returns previous flags, which must be restored
3036  * by sigallowstop().
3037  *
3038  * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
3039  * cleared by the current thread, which allow the lock-less read-only
3040  * accesses below.
3041  */
3042 int
3043 sigdeferstop_impl(int mode)
3044 {
3045 	struct thread *td;
3046 	int cflags, nflags;
3047 
3048 	td = curthread;
3049 	cflags = sigdeferstop_curr_flags(td->td_flags);
3050 	switch (mode) {
3051 	case SIGDEFERSTOP_NOP:
3052 		nflags = cflags;
3053 		break;
3054 	case SIGDEFERSTOP_OFF:
3055 		nflags = 0;
3056 		break;
3057 	case SIGDEFERSTOP_SILENT:
3058 		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
3059 		break;
3060 	case SIGDEFERSTOP_EINTR:
3061 		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
3062 		break;
3063 	case SIGDEFERSTOP_ERESTART:
3064 		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
3065 		break;
3066 	default:
3067 		panic("sigdeferstop: invalid mode %x", mode);
3068 		break;
3069 	}
3070 	if (cflags == nflags)
3071 		return (SIGDEFERSTOP_VAL_NCHG);
3072 	thread_lock(td);
3073 	td->td_flags = (td->td_flags & ~cflags) | nflags;
3074 	thread_unlock(td);
3075 	return (cflags);
3076 }
3077 
3078 /*
3079  * Restores the STOP handling mode, typically permitting the delivery
3080  * of SIGSTOP for the current thread.  This does not immediately
3081  * suspend if a stop was posted.  Instead, the thread will suspend
3082  * either via ast() or a subsequent interruptible sleep.
3083  */
3084 void
3085 sigallowstop_impl(int prev)
3086 {
3087 	struct thread *td;
3088 	int cflags;
3089 
3090 	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
3091 	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
3092 	    ("sigallowstop: incorrect previous mode %x", prev));
3093 	td = curthread;
3094 	cflags = sigdeferstop_curr_flags(td->td_flags);
3095 	if (cflags != prev) {
3096 		thread_lock(td);
3097 		td->td_flags = (td->td_flags & ~cflags) | prev;
3098 		thread_unlock(td);
3099 	}
3100 }
3101 
3102 enum sigstatus {
3103 	SIGSTATUS_HANDLE,
3104 	SIGSTATUS_HANDLED,
3105 	SIGSTATUS_IGNORE,
3106 	SIGSTATUS_SBDRY_STOP,
3107 };
3108 
3109 /*
3110  * The thread has signal "sig" pending.  Figure out what to do with it:
3111  *
3112  * _HANDLE     -> the caller should handle the signal
3113  * _HANDLED    -> handled internally, reload pending signal set
3114  * _IGNORE     -> ignored, remove from the set of pending signals and try the
3115  *                next pending signal
3116  * _SBDRY_STOP -> the signal should stop the thread but this is not
3117  *                permitted in the current context
3118  */
3119 static enum sigstatus
3120 sigprocess(struct thread *td, int sig)
3121 {
3122 	struct proc *p;
3123 	struct sigacts *ps;
3124 	struct sigqueue *queue;
3125 	ksiginfo_t ksi;
3126 	int prop;
3127 
3128 	KASSERT(_SIG_VALID(sig), ("%s: invalid signal %d", __func__, sig));
3129 
3130 	p = td->td_proc;
3131 	ps = p->p_sigacts;
3132 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3133 	PROC_LOCK_ASSERT(p, MA_OWNED);
3134 
3135 	/*
3136 	 * We should allow pending but ignored signals below
3137 	 * if there is sigwait() active, or P_TRACED was
3138 	 * on when they were posted.
3139 	 */
3140 	if (SIGISMEMBER(ps->ps_sigignore, sig) &&
3141 	    (p->p_flag & P_TRACED) == 0 &&
3142 	    (td->td_flags & TDF_SIGWAIT) == 0) {
3143 		return (SIGSTATUS_IGNORE);
3144 	}
3145 
3146 	/*
3147 	 * If the process is going to single-thread mode to prepare
3148 	 * for exit, there is no sense in delivering any signal
3149 	 * to usermode.  Another important consequence is that
3150 	 * msleep(..., PCATCH, ...) now is only interruptible by a
3151 	 * suspend request.
3152 	 */
3153 	if ((p->p_flag2 & P2_WEXIT) != 0)
3154 		return (SIGSTATUS_IGNORE);
3155 
3156 	if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
3157 		/*
3158 		 * If traced, always stop.
3159 		 * Remove old signal from queue before the stop.
3160 		 * XXX shrug off debugger, it causes siginfo to
3161 		 * be thrown away.
3162 		 */
3163 		queue = &td->td_sigqueue;
3164 		ksiginfo_init(&ksi);
3165 		if (sigqueue_get(queue, sig, &ksi) == 0) {
3166 			queue = &p->p_sigqueue;
3167 			sigqueue_get(queue, sig, &ksi);
3168 		}
3169 		td->td_si = ksi.ksi_info;
3170 
3171 		mtx_unlock(&ps->ps_mtx);
3172 		sig = ptracestop(td, sig, &ksi);
3173 		mtx_lock(&ps->ps_mtx);
3174 
3175 		td->td_si.si_signo = 0;
3176 
3177 		/*
3178 		 * Keep looking if the debugger discarded or
3179 		 * replaced the signal.
3180 		 */
3181 		if (sig == 0)
3182 			return (SIGSTATUS_HANDLED);
3183 
3184 		/*
3185 		 * If the signal became masked, re-queue it.
3186 		 */
3187 		if (SIGISMEMBER(td->td_sigmask, sig)) {
3188 			ksi.ksi_flags |= KSI_HEAD;
3189 			sigqueue_add(&p->p_sigqueue, sig, &ksi);
3190 			return (SIGSTATUS_HANDLED);
3191 		}
3192 
3193 		/*
3194 		 * If the traced bit got turned off, requeue the signal and
3195 		 * reload the set of pending signals.  This ensures that p_sig*
3196 		 * and p_sigact are consistent.
3197 		 */
3198 		if ((p->p_flag & P_TRACED) == 0) {
3199 			if ((ksi.ksi_flags & KSI_PTRACE) == 0) {
3200 				ksi.ksi_flags |= KSI_HEAD;
3201 				sigqueue_add(queue, sig, &ksi);
3202 			}
3203 			return (SIGSTATUS_HANDLED);
3204 		}
3205 	}
3206 
3207 	/*
3208 	 * Decide whether the signal should be returned.
3209 	 * Return the signal's number, or fall through
3210 	 * to clear it from the pending mask.
3211 	 */
3212 	switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
3213 	case (intptr_t)SIG_DFL:
3214 		/*
3215 		 * Don't take default actions on system processes.
3216 		 */
3217 		if (p->p_pid <= 1) {
3218 #ifdef DIAGNOSTIC
3219 			/*
3220 			 * Are you sure you want to ignore SIGSEGV
3221 			 * in init? XXX
3222 			 */
3223 			printf("Process (pid %lu) got signal %d\n",
3224 				(u_long)p->p_pid, sig);
3225 #endif
3226 			return (SIGSTATUS_IGNORE);
3227 		}
3228 
3229 		/*
3230 		 * If there is a pending stop signal to process with
3231 		 * default action, stop here, then clear the signal.
3232 		 * Traced or exiting processes should ignore stops.
3233 		 * Additionally, a member of an orphaned process group
3234 		 * should ignore tty stops.
3235 		 */
3236 		prop = sigprop(sig);
3237 		if (prop & SIGPROP_STOP) {
3238 			mtx_unlock(&ps->ps_mtx);
3239 			if ((p->p_flag & (P_TRACED | P_WEXIT |
3240 			    P_SINGLE_EXIT)) != 0 || ((p->p_pgrp->
3241 			    pg_flags & PGRP_ORPHANED) != 0 &&
3242 			    (prop & SIGPROP_TTYSTOP) != 0)) {
3243 				mtx_lock(&ps->ps_mtx);
3244 				return (SIGSTATUS_IGNORE);
3245 			}
3246 			if (TD_SBDRY_INTR(td)) {
3247 				KASSERT((td->td_flags & TDF_SBDRY) != 0,
3248 				    ("lost TDF_SBDRY"));
3249 				mtx_lock(&ps->ps_mtx);
3250 				return (SIGSTATUS_SBDRY_STOP);
3251 			}
3252 			WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
3253 			    &p->p_mtx.lock_object, "Catching SIGSTOP");
3254 			sigqueue_delete(&td->td_sigqueue, sig);
3255 			sigqueue_delete(&p->p_sigqueue, sig);
3256 			p->p_flag |= P_STOPPED_SIG;
3257 			p->p_xsig = sig;
3258 			PROC_SLOCK(p);
3259 			sig_suspend_threads(td, p);
3260 			thread_suspend_switch(td, p);
3261 			PROC_SUNLOCK(p);
3262 			mtx_lock(&ps->ps_mtx);
3263 			return (SIGSTATUS_HANDLED);
3264 		} else if ((prop & SIGPROP_IGNORE) != 0 &&
3265 		    (td->td_flags & TDF_SIGWAIT) == 0) {
3266 			/*
3267 			 * Default action is to ignore; drop it if
3268 			 * not in kern_sigtimedwait().
3269 			 */
3270 			return (SIGSTATUS_IGNORE);
3271 		} else {
3272 			return (SIGSTATUS_HANDLE);
3273 		}
3274 
3275 	case (intptr_t)SIG_IGN:
3276 		if ((td->td_flags & TDF_SIGWAIT) == 0)
3277 			return (SIGSTATUS_IGNORE);
3278 		else
3279 			return (SIGSTATUS_HANDLE);
3280 
3281 	default:
3282 		/*
3283 		 * This signal has an action, let postsig() process it.
3284 		 */
3285 		return (SIGSTATUS_HANDLE);
3286 	}
3287 }
3288 
3289 /*
3290  * If the current process has received a signal (should be caught or cause
3291  * termination, should interrupt current syscall), return the signal number.
3292  * Stop signals with default action are processed immediately, then cleared;
3293  * they aren't returned.  This is checked after each entry to the system for
3294  * a syscall or trap (though this can usually be done without calling
3295  * issignal by checking the pending signal masks in cursig.) The normal call
3296  * sequence is
3297  *
3298  *	while (sig = cursig(curthread))
3299  *		postsig(sig);
3300  */
3301 static int
3302 issignal(struct thread *td)
3303 {
3304 	struct proc *p;
3305 	sigset_t sigpending;
3306 	int sig;
3307 
3308 	p = td->td_proc;
3309 	PROC_LOCK_ASSERT(p, MA_OWNED);
3310 
3311 	for (;;) {
3312 		sigpending = td->td_sigqueue.sq_signals;
3313 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
3314 		SIGSETNAND(sigpending, td->td_sigmask);
3315 
3316 		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
3317 		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
3318 			SIG_STOPSIGMASK(sigpending);
3319 		if (SIGISEMPTY(sigpending))	/* no signal to send */
3320 			return (0);
3321 
3322 		/*
3323 		 * Do fast sigblock if requested by usermode.  Since
3324 		 * we do know that there was a signal pending at this
3325 		 * point, set the FAST_SIGBLOCK_PEND as indicator for
3326 		 * usermode to perform a dummy call to
3327 		 * FAST_SIGBLOCK_UNBLOCK, which causes immediate
3328 		 * delivery of postponed pending signal.
3329 		 */
3330 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
3331 			if (td->td_sigblock_val != 0)
3332 				SIGSETNAND(sigpending, fastblock_mask);
3333 			if (SIGISEMPTY(sigpending)) {
3334 				td->td_pflags |= TDP_SIGFASTPENDING;
3335 				return (0);
3336 			}
3337 		}
3338 
3339 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
3340 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
3341 		    SIGISMEMBER(sigpending, SIGSTOP)) {
3342 			/*
3343 			 * If debugger just attached, always consume
3344 			 * SIGSTOP from ptrace(PT_ATTACH) first, to
3345 			 * execute the debugger attach ritual in
3346 			 * order.
3347 			 */
3348 			td->td_dbgflags |= TDB_FSTP;
3349 			SIGEMPTYSET(sigpending);
3350 			SIGADDSET(sigpending, SIGSTOP);
3351 		}
3352 
3353 		SIG_FOREACH(sig, &sigpending) {
3354 			switch (sigprocess(td, sig)) {
3355 			case SIGSTATUS_HANDLE:
3356 				return (sig);
3357 			case SIGSTATUS_HANDLED:
3358 				goto next;
3359 			case SIGSTATUS_IGNORE:
3360 				sigqueue_delete(&td->td_sigqueue, sig);
3361 				sigqueue_delete(&p->p_sigqueue, sig);
3362 				break;
3363 			case SIGSTATUS_SBDRY_STOP:
3364 				return (-1);
3365 			}
3366 		}
3367 next:;
3368 	}
3369 }
3370 
3371 void
3372 thread_stopped(struct proc *p)
3373 {
3374 	int n;
3375 
3376 	PROC_LOCK_ASSERT(p, MA_OWNED);
3377 	PROC_SLOCK_ASSERT(p, MA_OWNED);
3378 	n = p->p_suspcount;
3379 	if (p == curproc)
3380 		n++;
3381 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3382 		PROC_SUNLOCK(p);
3383 		p->p_flag &= ~P_WAITED;
3384 		PROC_LOCK(p->p_pptr);
3385 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
3386 			CLD_TRAPPED : CLD_STOPPED);
3387 		PROC_UNLOCK(p->p_pptr);
3388 		PROC_SLOCK(p);
3389 	}
3390 }
3391 
3392 /*
3393  * Take the action for the specified signal
3394  * from the current set of pending signals.
3395  */
3396 int
3397 postsig(int sig)
3398 {
3399 	struct thread *td;
3400 	struct proc *p;
3401 	struct sigacts *ps;
3402 	sig_t action;
3403 	ksiginfo_t ksi;
3404 	sigset_t returnmask;
3405 
3406 	KASSERT(sig != 0, ("postsig"));
3407 
3408 	td = curthread;
3409 	p = td->td_proc;
3410 	PROC_LOCK_ASSERT(p, MA_OWNED);
3411 	ps = p->p_sigacts;
3412 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3413 	ksiginfo_init(&ksi);
3414 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3415 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3416 		return (0);
3417 	ksi.ksi_signo = sig;
3418 	if (ksi.ksi_code == SI_TIMER)
3419 		itimer_accept(p, ksi.ksi_timerid, &ksi);
3420 	action = ps->ps_sigact[_SIG_IDX(sig)];
3421 #ifdef KTRACE
3422 	if (KTRPOINT(td, KTR_PSIG))
3423 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3424 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3425 #endif
3426 
3427 	if (action == SIG_DFL) {
3428 		/*
3429 		 * Default action, where the default is to kill
3430 		 * the process.  (Other cases were ignored above.)
3431 		 */
3432 		mtx_unlock(&ps->ps_mtx);
3433 		proc_td_siginfo_capture(td, &ksi.ksi_info);
3434 		sigexit(td, sig);
3435 		/* NOTREACHED */
3436 	} else {
3437 		/*
3438 		 * If we get here, the signal must be caught.
3439 		 */
3440 		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3441 		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3442 		    ("postsig action: blocked sig %d", sig));
3443 
3444 		/*
3445 		 * Set the new mask value and also defer further
3446 		 * occurrences of this signal.
3447 		 *
3448 		 * Special case: user has done a sigsuspend.  Here the
3449 		 * current mask is not of interest, but rather the
3450 		 * mask from before the sigsuspend is what we want
3451 		 * restored after the signal processing is completed.
3452 		 */
3453 		if (td->td_pflags & TDP_OLDMASK) {
3454 			returnmask = td->td_oldsigmask;
3455 			td->td_pflags &= ~TDP_OLDMASK;
3456 		} else
3457 			returnmask = td->td_sigmask;
3458 
3459 		if (p->p_sig == sig) {
3460 			p->p_sig = 0;
3461 		}
3462 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3463 		postsig_done(sig, td, ps);
3464 	}
3465 	return (1);
3466 }
3467 
3468 int
3469 sig_ast_checksusp(struct thread *td)
3470 {
3471 	struct proc *p __diagused;
3472 	int ret;
3473 
3474 	p = td->td_proc;
3475 	PROC_LOCK_ASSERT(p, MA_OWNED);
3476 
3477 	if (!td_ast_pending(td, TDA_SUSPEND))
3478 		return (0);
3479 
3480 	ret = thread_suspend_check(1);
3481 	MPASS(ret == 0 || ret == EINTR || ret == ERESTART);
3482 	return (ret);
3483 }
3484 
3485 int
3486 sig_ast_needsigchk(struct thread *td)
3487 {
3488 	struct proc *p;
3489 	struct sigacts *ps;
3490 	int ret, sig;
3491 
3492 	p = td->td_proc;
3493 	PROC_LOCK_ASSERT(p, MA_OWNED);
3494 
3495 	if (!td_ast_pending(td, TDA_SIG))
3496 		return (0);
3497 
3498 	ps = p->p_sigacts;
3499 	mtx_lock(&ps->ps_mtx);
3500 	sig = cursig(td);
3501 	if (sig == -1) {
3502 		mtx_unlock(&ps->ps_mtx);
3503 		KASSERT((td->td_flags & TDF_SBDRY) != 0, ("lost TDF_SBDRY"));
3504 		KASSERT(TD_SBDRY_INTR(td),
3505 		    ("lost TDF_SERESTART of TDF_SEINTR"));
3506 		KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
3507 		    (TDF_SEINTR | TDF_SERESTART),
3508 		    ("both TDF_SEINTR and TDF_SERESTART"));
3509 		ret = TD_SBDRY_ERRNO(td);
3510 	} else if (sig != 0) {
3511 		ret = SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR : ERESTART;
3512 		mtx_unlock(&ps->ps_mtx);
3513 	} else {
3514 		mtx_unlock(&ps->ps_mtx);
3515 		ret = 0;
3516 	}
3517 
3518 	/*
3519 	 * Do not go into sleep if this thread was the ptrace(2)
3520 	 * attach leader.  cursig() consumed SIGSTOP from PT_ATTACH,
3521 	 * but we usually act on the signal by interrupting sleep, and
3522 	 * should do that here as well.
3523 	 */
3524 	if ((td->td_dbgflags & TDB_FSTP) != 0) {
3525 		if (ret == 0)
3526 			ret = EINTR;
3527 		td->td_dbgflags &= ~TDB_FSTP;
3528 	}
3529 
3530 	return (ret);
3531 }
3532 
3533 int
3534 sig_intr(void)
3535 {
3536 	struct thread *td;
3537 	struct proc *p;
3538 	int ret;
3539 
3540 	td = curthread;
3541 	if (!td_ast_pending(td, TDA_SIG) && !td_ast_pending(td, TDA_SUSPEND))
3542 		return (0);
3543 
3544 	p = td->td_proc;
3545 
3546 	PROC_LOCK(p);
3547 	ret = sig_ast_checksusp(td);
3548 	if (ret == 0)
3549 		ret = sig_ast_needsigchk(td);
3550 	PROC_UNLOCK(p);
3551 	return (ret);
3552 }
3553 
3554 bool
3555 curproc_sigkilled(void)
3556 {
3557 	struct thread *td;
3558 	struct proc *p;
3559 	struct sigacts *ps;
3560 	bool res;
3561 
3562 	td = curthread;
3563 	if (!td_ast_pending(td, TDA_SIG))
3564 		return (false);
3565 
3566 	p = td->td_proc;
3567 	PROC_LOCK(p);
3568 	ps = p->p_sigacts;
3569 	mtx_lock(&ps->ps_mtx);
3570 	res = SIGISMEMBER(td->td_sigqueue.sq_signals, SIGKILL) ||
3571 	    SIGISMEMBER(p->p_sigqueue.sq_signals, SIGKILL);
3572 	mtx_unlock(&ps->ps_mtx);
3573 	PROC_UNLOCK(p);
3574 	return (res);
3575 }
3576 
3577 void
3578 proc_wkilled(struct proc *p)
3579 {
3580 
3581 	PROC_LOCK_ASSERT(p, MA_OWNED);
3582 	if ((p->p_flag & P_WKILLED) == 0) {
3583 		p->p_flag |= P_WKILLED;
3584 		/*
3585 		 * Notify swapper that there is a process to swap in.
3586 		 * The notification is racy, at worst it would take 10
3587 		 * seconds for the swapper process to notice.
3588 		 */
3589 		if ((p->p_flag & (P_INMEM | P_SWAPPINGIN)) == 0)
3590 			wakeup(&proc0);
3591 	}
3592 }
3593 
3594 /*
3595  * Kill the current process for stated reason.
3596  */
3597 void
3598 killproc(struct proc *p, const char *why)
3599 {
3600 
3601 	PROC_LOCK_ASSERT(p, MA_OWNED);
3602 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3603 	    p->p_comm);
3604 	log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3605 	    p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3606 	    p->p_ucred->cr_uid, why);
3607 	proc_wkilled(p);
3608 	kern_psignal(p, SIGKILL);
3609 }
3610 
3611 /*
3612  * Force the current process to exit with the specified signal, dumping core
3613  * if appropriate.  We bypass the normal tests for masked and caught signals,
3614  * allowing unrecoverable failures to terminate the process without changing
3615  * signal state.  Mark the accounting record with the signal termination.
3616  * If dumping core, save the signal number for the debugger.  Calls exit and
3617  * does not return.
3618  */
3619 void
3620 sigexit(struct thread *td, int sig)
3621 {
3622 	struct proc *p = td->td_proc;
3623 	const char *coreinfo;
3624 	int rv;
3625 
3626 	PROC_LOCK_ASSERT(p, MA_OWNED);
3627 	proc_set_p2_wexit(p);
3628 
3629 	p->p_acflag |= AXSIG;
3630 	/*
3631 	 * We must be single-threading to generate a core dump.  This
3632 	 * ensures that the registers in the core file are up-to-date.
3633 	 * Also, the ELF dump handler assumes that the thread list doesn't
3634 	 * change out from under it.
3635 	 *
3636 	 * XXX If another thread attempts to single-thread before us
3637 	 *     (e.g. via fork()), we won't get a dump at all.
3638 	 */
3639 	if ((sigprop(sig) & SIGPROP_CORE) &&
3640 	    thread_single(p, SINGLE_NO_EXIT) == 0) {
3641 		p->p_sig = sig;
3642 		/*
3643 		 * Log signals which would cause core dumps
3644 		 * (Log as LOG_INFO to appease those who don't want
3645 		 * these messages.)
3646 		 * XXX : Todo, as well as euid, write out ruid too
3647 		 * Note that coredump() drops proc lock.
3648 		 */
3649 		rv = coredump(td);
3650 		switch (rv) {
3651 		case 0:
3652 			sig |= WCOREFLAG;
3653 			coreinfo = " (core dumped)";
3654 			break;
3655 		case EFAULT:
3656 			coreinfo = " (no core dump - bad address)";
3657 			break;
3658 		case EINVAL:
3659 			coreinfo = " (no core dump - invalid argument)";
3660 			break;
3661 		case EFBIG:
3662 			coreinfo = " (no core dump - too large)";
3663 			break;
3664 		default:
3665 			coreinfo = " (no core dump - other error)";
3666 			break;
3667 		}
3668 		if (kern_logsigexit)
3669 			log(LOG_INFO,
3670 			    "pid %d (%s), jid %d, uid %d: exited on "
3671 			    "signal %d%s\n", p->p_pid, p->p_comm,
3672 			    p->p_ucred->cr_prison->pr_id,
3673 			    td->td_ucred->cr_uid,
3674 			    sig &~ WCOREFLAG, coreinfo);
3675 	} else
3676 		PROC_UNLOCK(p);
3677 	exit1(td, 0, sig);
3678 	/* NOTREACHED */
3679 }
3680 
3681 /*
3682  * Send queued SIGCHLD to parent when child process's state
3683  * is changed.
3684  */
3685 static void
3686 sigparent(struct proc *p, int reason, int status)
3687 {
3688 	PROC_LOCK_ASSERT(p, MA_OWNED);
3689 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3690 
3691 	if (p->p_ksi != NULL) {
3692 		p->p_ksi->ksi_signo  = SIGCHLD;
3693 		p->p_ksi->ksi_code   = reason;
3694 		p->p_ksi->ksi_status = status;
3695 		p->p_ksi->ksi_pid    = p->p_pid;
3696 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3697 		if (KSI_ONQ(p->p_ksi))
3698 			return;
3699 	}
3700 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3701 }
3702 
3703 static void
3704 childproc_jobstate(struct proc *p, int reason, int sig)
3705 {
3706 	struct sigacts *ps;
3707 
3708 	PROC_LOCK_ASSERT(p, MA_OWNED);
3709 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3710 
3711 	/*
3712 	 * Wake up parent sleeping in kern_wait(), also send
3713 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3714 	 * that parent will awake, because parent may masked
3715 	 * the signal.
3716 	 */
3717 	p->p_pptr->p_flag |= P_STATCHILD;
3718 	wakeup(p->p_pptr);
3719 
3720 	ps = p->p_pptr->p_sigacts;
3721 	mtx_lock(&ps->ps_mtx);
3722 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3723 		mtx_unlock(&ps->ps_mtx);
3724 		sigparent(p, reason, sig);
3725 	} else
3726 		mtx_unlock(&ps->ps_mtx);
3727 }
3728 
3729 void
3730 childproc_stopped(struct proc *p, int reason)
3731 {
3732 
3733 	childproc_jobstate(p, reason, p->p_xsig);
3734 }
3735 
3736 void
3737 childproc_continued(struct proc *p)
3738 {
3739 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3740 }
3741 
3742 void
3743 childproc_exited(struct proc *p)
3744 {
3745 	int reason, status;
3746 
3747 	if (WCOREDUMP(p->p_xsig)) {
3748 		reason = CLD_DUMPED;
3749 		status = WTERMSIG(p->p_xsig);
3750 	} else if (WIFSIGNALED(p->p_xsig)) {
3751 		reason = CLD_KILLED;
3752 		status = WTERMSIG(p->p_xsig);
3753 	} else {
3754 		reason = CLD_EXITED;
3755 		status = p->p_xexit;
3756 	}
3757 	/*
3758 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3759 	 * done in exit1().
3760 	 */
3761 	sigparent(p, reason, status);
3762 }
3763 
3764 #define	MAX_NUM_CORE_FILES 100000
3765 #ifndef NUM_CORE_FILES
3766 #define	NUM_CORE_FILES 5
3767 #endif
3768 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3769 static int num_cores = NUM_CORE_FILES;
3770 
3771 static int
3772 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3773 {
3774 	int error;
3775 	int new_val;
3776 
3777 	new_val = num_cores;
3778 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3779 	if (error != 0 || req->newptr == NULL)
3780 		return (error);
3781 	if (new_val > MAX_NUM_CORE_FILES)
3782 		new_val = MAX_NUM_CORE_FILES;
3783 	if (new_val < 0)
3784 		new_val = 0;
3785 	num_cores = new_val;
3786 	return (0);
3787 }
3788 SYSCTL_PROC(_debug, OID_AUTO, ncores,
3789     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
3790     sysctl_debug_num_cores_check, "I",
3791     "Maximum number of generated process corefiles while using index format");
3792 
3793 #define	GZIP_SUFFIX	".gz"
3794 #define	ZSTD_SUFFIX	".zst"
3795 
3796 int compress_user_cores = 0;
3797 
3798 static int
3799 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3800 {
3801 	int error, val;
3802 
3803 	val = compress_user_cores;
3804 	error = sysctl_handle_int(oidp, &val, 0, req);
3805 	if (error != 0 || req->newptr == NULL)
3806 		return (error);
3807 	if (val != 0 && !compressor_avail(val))
3808 		return (EINVAL);
3809 	compress_user_cores = val;
3810 	return (error);
3811 }
3812 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores,
3813     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, 0, sizeof(int),
3814     sysctl_compress_user_cores, "I",
3815     "Enable compression of user corefiles ("
3816     __XSTRING(COMPRESS_GZIP) " = gzip, "
3817     __XSTRING(COMPRESS_ZSTD) " = zstd)");
3818 
3819 int compress_user_cores_level = 6;
3820 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3821     &compress_user_cores_level, 0,
3822     "Corefile compression level");
3823 
3824 /*
3825  * Protect the access to corefilename[] by allproc_lock.
3826  */
3827 #define	corefilename_lock	allproc_lock
3828 
3829 static char corefilename[MAXPATHLEN] = {"%N.core"};
3830 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3831 
3832 static int
3833 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3834 {
3835 	int error;
3836 
3837 	sx_xlock(&corefilename_lock);
3838 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3839 	    req);
3840 	sx_xunlock(&corefilename_lock);
3841 
3842 	return (error);
3843 }
3844 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3845     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3846     "Process corefile name format string");
3847 
3848 static void
3849 vnode_close_locked(struct thread *td, struct vnode *vp)
3850 {
3851 
3852 	VOP_UNLOCK(vp);
3853 	vn_close(vp, FWRITE, td->td_ucred, td);
3854 }
3855 
3856 /*
3857  * If the core format has a %I in it, then we need to check
3858  * for existing corefiles before defining a name.
3859  * To do this we iterate over 0..ncores to find a
3860  * non-existing core file name to use. If all core files are
3861  * already used we choose the oldest one.
3862  */
3863 static int
3864 corefile_open_last(struct thread *td, char *name, int indexpos,
3865     int indexlen, int ncores, struct vnode **vpp)
3866 {
3867 	struct vnode *oldvp, *nextvp, *vp;
3868 	struct vattr vattr;
3869 	struct nameidata nd;
3870 	int error, i, flags, oflags, cmode;
3871 	char ch;
3872 	struct timespec lasttime;
3873 
3874 	nextvp = oldvp = NULL;
3875 	cmode = S_IRUSR | S_IWUSR;
3876 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3877 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3878 
3879 	for (i = 0; i < ncores; i++) {
3880 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3881 
3882 		ch = name[indexpos + indexlen];
3883 		(void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3884 		    i);
3885 		name[indexpos + indexlen] = ch;
3886 
3887 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
3888 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3889 		    NULL);
3890 		if (error != 0)
3891 			break;
3892 
3893 		vp = nd.ni_vp;
3894 		NDFREE_PNBUF(&nd);
3895 		if ((flags & O_CREAT) == O_CREAT) {
3896 			nextvp = vp;
3897 			break;
3898 		}
3899 
3900 		error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3901 		if (error != 0) {
3902 			vnode_close_locked(td, vp);
3903 			break;
3904 		}
3905 
3906 		if (oldvp == NULL ||
3907 		    lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3908 		    (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3909 		    lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3910 			if (oldvp != NULL)
3911 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3912 			oldvp = vp;
3913 			VOP_UNLOCK(oldvp);
3914 			lasttime = vattr.va_mtime;
3915 		} else {
3916 			vnode_close_locked(td, vp);
3917 		}
3918 	}
3919 
3920 	if (oldvp != NULL) {
3921 		if (nextvp == NULL) {
3922 			if ((td->td_proc->p_flag & P_SUGID) != 0) {
3923 				error = EFAULT;
3924 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3925 			} else {
3926 				nextvp = oldvp;
3927 				error = vn_lock(nextvp, LK_EXCLUSIVE);
3928 				if (error != 0) {
3929 					vn_close(nextvp, FWRITE, td->td_ucred,
3930 					    td);
3931 					nextvp = NULL;
3932 				}
3933 			}
3934 		} else {
3935 			vn_close(oldvp, FWRITE, td->td_ucred, td);
3936 		}
3937 	}
3938 	if (error != 0) {
3939 		if (nextvp != NULL)
3940 			vnode_close_locked(td, oldvp);
3941 	} else {
3942 		*vpp = nextvp;
3943 	}
3944 
3945 	return (error);
3946 }
3947 
3948 /*
3949  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3950  * Expand the name described in corefilename, using name, uid, and pid
3951  * and open/create core file.
3952  * corefilename is a printf-like string, with three format specifiers:
3953  *	%N	name of process ("name")
3954  *	%P	process id (pid)
3955  *	%U	user id (uid)
3956  * For example, "%N.core" is the default; they can be disabled completely
3957  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3958  * This is controlled by the sysctl variable kern.corefile (see above).
3959  */
3960 static int
3961 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3962     int compress, int signum, struct vnode **vpp, char **namep)
3963 {
3964 	struct sbuf sb;
3965 	struct nameidata nd;
3966 	const char *format;
3967 	char *hostname, *name;
3968 	int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
3969 
3970 	hostname = NULL;
3971 	format = corefilename;
3972 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3973 	indexlen = 0;
3974 	indexpos = -1;
3975 	ncores = num_cores;
3976 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3977 	sx_slock(&corefilename_lock);
3978 	for (i = 0; format[i] != '\0'; i++) {
3979 		switch (format[i]) {
3980 		case '%':	/* Format character */
3981 			i++;
3982 			switch (format[i]) {
3983 			case '%':
3984 				sbuf_putc(&sb, '%');
3985 				break;
3986 			case 'H':	/* hostname */
3987 				if (hostname == NULL) {
3988 					hostname = malloc(MAXHOSTNAMELEN,
3989 					    M_TEMP, M_WAITOK);
3990 				}
3991 				getcredhostname(td->td_ucred, hostname,
3992 				    MAXHOSTNAMELEN);
3993 				sbuf_cat(&sb, hostname);
3994 				break;
3995 			case 'I':	/* autoincrementing index */
3996 				if (indexpos != -1) {
3997 					sbuf_printf(&sb, "%%I");
3998 					break;
3999 				}
4000 
4001 				indexpos = sbuf_len(&sb);
4002 				sbuf_printf(&sb, "%u", ncores - 1);
4003 				indexlen = sbuf_len(&sb) - indexpos;
4004 				break;
4005 			case 'N':	/* process name */
4006 				sbuf_printf(&sb, "%s", comm);
4007 				break;
4008 			case 'P':	/* process id */
4009 				sbuf_printf(&sb, "%u", pid);
4010 				break;
4011 			case 'S':	/* signal number */
4012 				sbuf_printf(&sb, "%i", signum);
4013 				break;
4014 			case 'U':	/* user id */
4015 				sbuf_printf(&sb, "%u", uid);
4016 				break;
4017 			default:
4018 				log(LOG_ERR,
4019 				    "Unknown format character %c in "
4020 				    "corename `%s'\n", format[i], format);
4021 				break;
4022 			}
4023 			break;
4024 		default:
4025 			sbuf_putc(&sb, format[i]);
4026 			break;
4027 		}
4028 	}
4029 	sx_sunlock(&corefilename_lock);
4030 	free(hostname, M_TEMP);
4031 	if (compress == COMPRESS_GZIP)
4032 		sbuf_cat(&sb, GZIP_SUFFIX);
4033 	else if (compress == COMPRESS_ZSTD)
4034 		sbuf_cat(&sb, ZSTD_SUFFIX);
4035 	if (sbuf_error(&sb) != 0) {
4036 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
4037 		    "long\n", (long)pid, comm, (u_long)uid);
4038 		sbuf_delete(&sb);
4039 		free(name, M_TEMP);
4040 		return (ENOMEM);
4041 	}
4042 	sbuf_finish(&sb);
4043 	sbuf_delete(&sb);
4044 
4045 	if (indexpos != -1) {
4046 		error = corefile_open_last(td, name, indexpos, indexlen, ncores,
4047 		    vpp);
4048 		if (error != 0) {
4049 			log(LOG_ERR,
4050 			    "pid %d (%s), uid (%u):  Path `%s' failed "
4051 			    "on initial open test, error = %d\n",
4052 			    pid, comm, uid, name, error);
4053 		}
4054 	} else {
4055 		cmode = S_IRUSR | S_IWUSR;
4056 		oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
4057 		    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
4058 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
4059 		if ((td->td_proc->p_flag & P_SUGID) != 0)
4060 			flags |= O_EXCL;
4061 
4062 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
4063 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
4064 		    NULL);
4065 		if (error == 0) {
4066 			*vpp = nd.ni_vp;
4067 			NDFREE_PNBUF(&nd);
4068 		}
4069 	}
4070 
4071 	if (error != 0) {
4072 #ifdef AUDIT
4073 		audit_proc_coredump(td, name, error);
4074 #endif
4075 		free(name, M_TEMP);
4076 		return (error);
4077 	}
4078 	*namep = name;
4079 	return (0);
4080 }
4081 
4082 /*
4083  * Dump a process' core.  The main routine does some
4084  * policy checking, and creates the name of the coredump;
4085  * then it passes on a vnode and a size limit to the process-specific
4086  * coredump routine if there is one; if there _is not_ one, it returns
4087  * ENOSYS; otherwise it returns the error from the process-specific routine.
4088  */
4089 
4090 static int
4091 coredump(struct thread *td)
4092 {
4093 	struct proc *p = td->td_proc;
4094 	struct ucred *cred = td->td_ucred;
4095 	struct vnode *vp;
4096 	struct flock lf;
4097 	struct vattr vattr;
4098 	size_t fullpathsize;
4099 	int error, error1, locked;
4100 	char *name;			/* name of corefile */
4101 	void *rl_cookie;
4102 	off_t limit;
4103 	char *fullpath, *freepath = NULL;
4104 	struct sbuf *sb;
4105 
4106 	PROC_LOCK_ASSERT(p, MA_OWNED);
4107 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
4108 
4109 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
4110 	    (p->p_flag2 & P2_NOTRACE) != 0) {
4111 		PROC_UNLOCK(p);
4112 		return (EFAULT);
4113 	}
4114 
4115 	/*
4116 	 * Note that the bulk of limit checking is done after
4117 	 * the corefile is created.  The exception is if the limit
4118 	 * for corefiles is 0, in which case we don't bother
4119 	 * creating the corefile at all.  This layout means that
4120 	 * a corefile is truncated instead of not being created,
4121 	 * if it is larger than the limit.
4122 	 */
4123 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
4124 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
4125 		PROC_UNLOCK(p);
4126 		return (EFBIG);
4127 	}
4128 	PROC_UNLOCK(p);
4129 
4130 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
4131 	    compress_user_cores, p->p_sig, &vp, &name);
4132 	if (error != 0)
4133 		return (error);
4134 
4135 	/*
4136 	 * Don't dump to non-regular files or files with links.
4137 	 * Do not dump into system files. Effective user must own the corefile.
4138 	 */
4139 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
4140 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
4141 	    vattr.va_uid != cred->cr_uid) {
4142 		VOP_UNLOCK(vp);
4143 		error = EFAULT;
4144 		goto out;
4145 	}
4146 
4147 	VOP_UNLOCK(vp);
4148 
4149 	/* Postpone other writers, including core dumps of other processes. */
4150 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
4151 
4152 	lf.l_whence = SEEK_SET;
4153 	lf.l_start = 0;
4154 	lf.l_len = 0;
4155 	lf.l_type = F_WRLCK;
4156 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
4157 
4158 	VATTR_NULL(&vattr);
4159 	vattr.va_size = 0;
4160 	if (set_core_nodump_flag)
4161 		vattr.va_flags = UF_NODUMP;
4162 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4163 	VOP_SETATTR(vp, &vattr, cred);
4164 	VOP_UNLOCK(vp);
4165 	PROC_LOCK(p);
4166 	p->p_acflag |= ACORE;
4167 	PROC_UNLOCK(p);
4168 
4169 	if (p->p_sysent->sv_coredump != NULL) {
4170 		error = p->p_sysent->sv_coredump(td, vp, limit, 0);
4171 	} else {
4172 		error = ENOSYS;
4173 	}
4174 
4175 	if (locked) {
4176 		lf.l_type = F_UNLCK;
4177 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
4178 	}
4179 	vn_rangelock_unlock(vp, rl_cookie);
4180 
4181 	/*
4182 	 * Notify the userland helper that a process triggered a core dump.
4183 	 * This allows the helper to run an automated debugging session.
4184 	 */
4185 	if (error != 0 || coredump_devctl == 0)
4186 		goto out;
4187 	sb = sbuf_new_auto();
4188 	if (vn_fullpath_global(p->p_textvp, &fullpath, &freepath) != 0)
4189 		goto out2;
4190 	sbuf_cat(sb, "comm=\"");
4191 	devctl_safe_quote_sb(sb, fullpath);
4192 	free(freepath, M_TEMP);
4193 	sbuf_cat(sb, "\" core=\"");
4194 
4195 	/*
4196 	 * We can't lookup core file vp directly. When we're replacing a core, and
4197 	 * other random times, we flush the name cache, so it will fail. Instead,
4198 	 * if the path of the core is relative, add the current dir in front if it.
4199 	 */
4200 	if (name[0] != '/') {
4201 		fullpathsize = MAXPATHLEN;
4202 		freepath = malloc(fullpathsize, M_TEMP, M_WAITOK);
4203 		if (vn_getcwd(freepath, &fullpath, &fullpathsize) != 0) {
4204 			free(freepath, M_TEMP);
4205 			goto out2;
4206 		}
4207 		devctl_safe_quote_sb(sb, fullpath);
4208 		free(freepath, M_TEMP);
4209 		sbuf_putc(sb, '/');
4210 	}
4211 	devctl_safe_quote_sb(sb, name);
4212 	sbuf_putc(sb, '"');
4213 	if (sbuf_finish(sb) == 0)
4214 		devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
4215 out2:
4216 	sbuf_delete(sb);
4217 out:
4218 	error1 = vn_close(vp, FWRITE, cred, td);
4219 	if (error == 0)
4220 		error = error1;
4221 #ifdef AUDIT
4222 	audit_proc_coredump(td, name, error);
4223 #endif
4224 	free(name, M_TEMP);
4225 	return (error);
4226 }
4227 
4228 /*
4229  * Nonexistent system call-- signal process (may want to handle it).  Flag
4230  * error in case process won't see signal immediately (blocked or ignored).
4231  */
4232 #ifndef _SYS_SYSPROTO_H_
4233 struct nosys_args {
4234 	int	dummy;
4235 };
4236 #endif
4237 /* ARGSUSED */
4238 int
4239 nosys(struct thread *td, struct nosys_args *args)
4240 {
4241 	struct proc *p;
4242 
4243 	p = td->td_proc;
4244 
4245 	if (SV_PROC_FLAG(p, SV_SIGSYS) != 0 && kern_signosys) {
4246 		PROC_LOCK(p);
4247 		tdsignal(td, SIGSYS);
4248 		PROC_UNLOCK(p);
4249 	}
4250 	if (kern_lognosys == 1 || kern_lognosys == 3) {
4251 		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4252 		    td->td_sa.code);
4253 	}
4254 	if (kern_lognosys == 2 || kern_lognosys == 3 ||
4255 	    (p->p_pid == 1 && (kern_lognosys & 3) == 0)) {
4256 		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4257 		    td->td_sa.code);
4258 	}
4259 	return (ENOSYS);
4260 }
4261 
4262 /*
4263  * Send a SIGIO or SIGURG signal to a process or process group using stored
4264  * credentials rather than those of the current process.
4265  */
4266 void
4267 pgsigio(struct sigio **sigiop, int sig, int checkctty)
4268 {
4269 	ksiginfo_t ksi;
4270 	struct sigio *sigio;
4271 
4272 	ksiginfo_init(&ksi);
4273 	ksi.ksi_signo = sig;
4274 	ksi.ksi_code = SI_KERNEL;
4275 
4276 	SIGIO_LOCK();
4277 	sigio = *sigiop;
4278 	if (sigio == NULL) {
4279 		SIGIO_UNLOCK();
4280 		return;
4281 	}
4282 	if (sigio->sio_pgid > 0) {
4283 		PROC_LOCK(sigio->sio_proc);
4284 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
4285 			kern_psignal(sigio->sio_proc, sig);
4286 		PROC_UNLOCK(sigio->sio_proc);
4287 	} else if (sigio->sio_pgid < 0) {
4288 		struct proc *p;
4289 
4290 		PGRP_LOCK(sigio->sio_pgrp);
4291 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
4292 			PROC_LOCK(p);
4293 			if (p->p_state == PRS_NORMAL &&
4294 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
4295 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
4296 				kern_psignal(p, sig);
4297 			PROC_UNLOCK(p);
4298 		}
4299 		PGRP_UNLOCK(sigio->sio_pgrp);
4300 	}
4301 	SIGIO_UNLOCK();
4302 }
4303 
4304 static int
4305 filt_sigattach(struct knote *kn)
4306 {
4307 	struct proc *p = curproc;
4308 
4309 	kn->kn_ptr.p_proc = p;
4310 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
4311 
4312 	knlist_add(p->p_klist, kn, 0);
4313 
4314 	return (0);
4315 }
4316 
4317 static void
4318 filt_sigdetach(struct knote *kn)
4319 {
4320 	knlist_remove(kn->kn_knlist, kn, 0);
4321 }
4322 
4323 /*
4324  * signal knotes are shared with proc knotes, so we apply a mask to
4325  * the hint in order to differentiate them from process hints.  This
4326  * could be avoided by using a signal-specific knote list, but probably
4327  * isn't worth the trouble.
4328  */
4329 static int
4330 filt_signal(struct knote *kn, long hint)
4331 {
4332 
4333 	if (hint & NOTE_SIGNAL) {
4334 		hint &= ~NOTE_SIGNAL;
4335 
4336 		if (kn->kn_id == hint)
4337 			kn->kn_data++;
4338 	}
4339 	return (kn->kn_data != 0);
4340 }
4341 
4342 struct sigacts *
4343 sigacts_alloc(void)
4344 {
4345 	struct sigacts *ps;
4346 
4347 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
4348 	refcount_init(&ps->ps_refcnt, 1);
4349 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
4350 	return (ps);
4351 }
4352 
4353 void
4354 sigacts_free(struct sigacts *ps)
4355 {
4356 
4357 	if (refcount_release(&ps->ps_refcnt) == 0)
4358 		return;
4359 	mtx_destroy(&ps->ps_mtx);
4360 	free(ps, M_SUBPROC);
4361 }
4362 
4363 struct sigacts *
4364 sigacts_hold(struct sigacts *ps)
4365 {
4366 
4367 	refcount_acquire(&ps->ps_refcnt);
4368 	return (ps);
4369 }
4370 
4371 void
4372 sigacts_copy(struct sigacts *dest, struct sigacts *src)
4373 {
4374 
4375 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
4376 	mtx_lock(&src->ps_mtx);
4377 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
4378 	mtx_unlock(&src->ps_mtx);
4379 }
4380 
4381 int
4382 sigacts_shared(struct sigacts *ps)
4383 {
4384 
4385 	return (ps->ps_refcnt > 1);
4386 }
4387 
4388 void
4389 sig_drop_caught(struct proc *p)
4390 {
4391 	int sig;
4392 	struct sigacts *ps;
4393 
4394 	ps = p->p_sigacts;
4395 	PROC_LOCK_ASSERT(p, MA_OWNED);
4396 	mtx_assert(&ps->ps_mtx, MA_OWNED);
4397 	SIG_FOREACH(sig, &ps->ps_sigcatch) {
4398 		sigdflt(ps, sig);
4399 		if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
4400 			sigqueue_delete_proc(p, sig);
4401 	}
4402 }
4403 
4404 static void
4405 sigfastblock_failed(struct thread *td, bool sendsig, bool write)
4406 {
4407 	ksiginfo_t ksi;
4408 
4409 	/*
4410 	 * Prevent further fetches and SIGSEGVs, allowing thread to
4411 	 * issue syscalls despite corruption.
4412 	 */
4413 	sigfastblock_clear(td);
4414 
4415 	if (!sendsig)
4416 		return;
4417 	ksiginfo_init_trap(&ksi);
4418 	ksi.ksi_signo = SIGSEGV;
4419 	ksi.ksi_code = write ? SEGV_ACCERR : SEGV_MAPERR;
4420 	ksi.ksi_addr = td->td_sigblock_ptr;
4421 	trapsignal(td, &ksi);
4422 }
4423 
4424 static bool
4425 sigfastblock_fetch_sig(struct thread *td, bool sendsig, uint32_t *valp)
4426 {
4427 	uint32_t res;
4428 
4429 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4430 		return (true);
4431 	if (fueword32((void *)td->td_sigblock_ptr, &res) == -1) {
4432 		sigfastblock_failed(td, sendsig, false);
4433 		return (false);
4434 	}
4435 	*valp = res;
4436 	td->td_sigblock_val = res & ~SIGFASTBLOCK_FLAGS;
4437 	return (true);
4438 }
4439 
4440 static void
4441 sigfastblock_resched(struct thread *td, bool resched)
4442 {
4443 	struct proc *p;
4444 
4445 	if (resched) {
4446 		p = td->td_proc;
4447 		PROC_LOCK(p);
4448 		reschedule_signals(p, td->td_sigmask, 0);
4449 		PROC_UNLOCK(p);
4450 	}
4451 	ast_sched(td, TDA_SIG);
4452 }
4453 
4454 int
4455 sys_sigfastblock(struct thread *td, struct sigfastblock_args *uap)
4456 {
4457 	struct proc *p;
4458 	int error, res;
4459 	uint32_t oldval;
4460 
4461 	error = 0;
4462 	p = td->td_proc;
4463 	switch (uap->cmd) {
4464 	case SIGFASTBLOCK_SETPTR:
4465 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
4466 			error = EBUSY;
4467 			break;
4468 		}
4469 		if (((uintptr_t)(uap->ptr) & (sizeof(uint32_t) - 1)) != 0) {
4470 			error = EINVAL;
4471 			break;
4472 		}
4473 		td->td_pflags |= TDP_SIGFASTBLOCK;
4474 		td->td_sigblock_ptr = uap->ptr;
4475 		break;
4476 
4477 	case SIGFASTBLOCK_UNBLOCK:
4478 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4479 			error = EINVAL;
4480 			break;
4481 		}
4482 
4483 		for (;;) {
4484 			res = casueword32(td->td_sigblock_ptr,
4485 			    SIGFASTBLOCK_PEND, &oldval, 0);
4486 			if (res == -1) {
4487 				error = EFAULT;
4488 				sigfastblock_failed(td, false, true);
4489 				break;
4490 			}
4491 			if (res == 0)
4492 				break;
4493 			MPASS(res == 1);
4494 			if (oldval != SIGFASTBLOCK_PEND) {
4495 				error = EBUSY;
4496 				break;
4497 			}
4498 			error = thread_check_susp(td, false);
4499 			if (error != 0)
4500 				break;
4501 		}
4502 		if (error != 0)
4503 			break;
4504 
4505 		/*
4506 		 * td_sigblock_val is cleared there, but not on a
4507 		 * syscall exit.  The end effect is that a single
4508 		 * interruptible sleep, while user sigblock word is
4509 		 * set, might return EINTR or ERESTART to usermode
4510 		 * without delivering signal.  All further sleeps,
4511 		 * until userspace clears the word and does
4512 		 * sigfastblock(UNBLOCK), observe current word and no
4513 		 * longer get interrupted.  It is slight
4514 		 * non-conformance, with alternative to have read the
4515 		 * sigblock word on each syscall entry.
4516 		 */
4517 		td->td_sigblock_val = 0;
4518 
4519 		/*
4520 		 * Rely on normal ast mechanism to deliver pending
4521 		 * signals to current thread.  But notify others about
4522 		 * fake unblock.
4523 		 */
4524 		sigfastblock_resched(td, error == 0 && p->p_numthreads != 1);
4525 
4526 		break;
4527 
4528 	case SIGFASTBLOCK_UNSETPTR:
4529 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4530 			error = EINVAL;
4531 			break;
4532 		}
4533 		if (!sigfastblock_fetch_sig(td, false, &oldval)) {
4534 			error = EFAULT;
4535 			break;
4536 		}
4537 		if (oldval != 0 && oldval != SIGFASTBLOCK_PEND) {
4538 			error = EBUSY;
4539 			break;
4540 		}
4541 		sigfastblock_clear(td);
4542 		break;
4543 
4544 	default:
4545 		error = EINVAL;
4546 		break;
4547 	}
4548 	return (error);
4549 }
4550 
4551 void
4552 sigfastblock_clear(struct thread *td)
4553 {
4554 	bool resched;
4555 
4556 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4557 		return;
4558 	td->td_sigblock_val = 0;
4559 	resched = (td->td_pflags & TDP_SIGFASTPENDING) != 0 ||
4560 	    SIGPENDING(td);
4561 	td->td_pflags &= ~(TDP_SIGFASTBLOCK | TDP_SIGFASTPENDING);
4562 	sigfastblock_resched(td, resched);
4563 }
4564 
4565 void
4566 sigfastblock_fetch(struct thread *td)
4567 {
4568 	uint32_t val;
4569 
4570 	(void)sigfastblock_fetch_sig(td, true, &val);
4571 }
4572 
4573 static void
4574 sigfastblock_setpend1(struct thread *td)
4575 {
4576 	int res;
4577 	uint32_t oldval;
4578 
4579 	if ((td->td_pflags & TDP_SIGFASTPENDING) == 0)
4580 		return;
4581 	res = fueword32((void *)td->td_sigblock_ptr, &oldval);
4582 	if (res == -1) {
4583 		sigfastblock_failed(td, true, false);
4584 		return;
4585 	}
4586 	for (;;) {
4587 		res = casueword32(td->td_sigblock_ptr, oldval, &oldval,
4588 		    oldval | SIGFASTBLOCK_PEND);
4589 		if (res == -1) {
4590 			sigfastblock_failed(td, true, true);
4591 			return;
4592 		}
4593 		if (res == 0) {
4594 			td->td_sigblock_val = oldval & ~SIGFASTBLOCK_FLAGS;
4595 			td->td_pflags &= ~TDP_SIGFASTPENDING;
4596 			break;
4597 		}
4598 		MPASS(res == 1);
4599 		if (thread_check_susp(td, false) != 0)
4600 			break;
4601 	}
4602 }
4603 
4604 static void
4605 sigfastblock_setpend(struct thread *td, bool resched)
4606 {
4607 	struct proc *p;
4608 
4609 	sigfastblock_setpend1(td);
4610 	if (resched) {
4611 		p = td->td_proc;
4612 		PROC_LOCK(p);
4613 		reschedule_signals(p, fastblock_mask, SIGPROCMASK_FASTBLK);
4614 		PROC_UNLOCK(p);
4615 	}
4616 }
4617