xref: /freebsd/sys/kern/kern_sig.c (revision da5432eda807c4b7232d030d5157d5b417ea4f52)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
37  */
38 
39 #include <sys/cdefs.h>
40 #include "opt_capsicum.h"
41 #include "opt_ktrace.h"
42 
43 #include <sys/param.h>
44 #include <sys/capsicum.h>
45 #include <sys/ctype.h>
46 #include <sys/systm.h>
47 #include <sys/signalvar.h>
48 #include <sys/vnode.h>
49 #include <sys/acct.h>
50 #include <sys/capsicum.h>
51 #include <sys/compressor.h>
52 #include <sys/condvar.h>
53 #include <sys/devctl.h>
54 #include <sys/event.h>
55 #include <sys/fcntl.h>
56 #include <sys/imgact.h>
57 #include <sys/jail.h>
58 #include <sys/kernel.h>
59 #include <sys/ktr.h>
60 #include <sys/ktrace.h>
61 #include <sys/limits.h>
62 #include <sys/lock.h>
63 #include <sys/malloc.h>
64 #include <sys/mutex.h>
65 #include <sys/refcount.h>
66 #include <sys/namei.h>
67 #include <sys/proc.h>
68 #include <sys/procdesc.h>
69 #include <sys/ptrace.h>
70 #include <sys/posix4.h>
71 #include <sys/racct.h>
72 #include <sys/resourcevar.h>
73 #include <sys/sdt.h>
74 #include <sys/sbuf.h>
75 #include <sys/sleepqueue.h>
76 #include <sys/smp.h>
77 #include <sys/stat.h>
78 #include <sys/sx.h>
79 #include <sys/syscall.h>
80 #include <sys/syscallsubr.h>
81 #include <sys/sysctl.h>
82 #include <sys/sysent.h>
83 #include <sys/syslog.h>
84 #include <sys/sysproto.h>
85 #include <sys/timers.h>
86 #include <sys/unistd.h>
87 #include <sys/vmmeter.h>
88 #include <sys/wait.h>
89 #include <vm/vm.h>
90 #include <vm/vm_extern.h>
91 #include <vm/uma.h>
92 
93 #include <machine/cpu.h>
94 
95 #include <security/audit/audit.h>
96 
97 #define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
98 
99 SDT_PROVIDER_DECLARE(proc);
100 SDT_PROBE_DEFINE3(proc, , , signal__send,
101     "struct thread *", "struct proc *", "int");
102 SDT_PROBE_DEFINE2(proc, , , signal__clear,
103     "int", "ksiginfo_t *");
104 SDT_PROBE_DEFINE3(proc, , , signal__discard,
105     "struct thread *", "struct proc *", "int");
106 
107 static int	coredump(struct thread *);
108 static int	killpg1(struct thread *td, int sig, int pgid, int all,
109 		    ksiginfo_t *ksi);
110 static int	issignal(struct thread *td);
111 static void	reschedule_signals(struct proc *p, sigset_t block, int flags);
112 static int	sigprop(int sig);
113 static void	tdsigwakeup(struct thread *, int, sig_t, int);
114 static int	sig_suspend_threads(struct thread *, struct proc *);
115 static int	filt_sigattach(struct knote *kn);
116 static void	filt_sigdetach(struct knote *kn);
117 static int	filt_signal(struct knote *kn, long hint);
118 static struct thread *sigtd(struct proc *p, int sig, bool fast_sigblock);
119 static void	sigqueue_start(void);
120 static void	sigfastblock_setpend(struct thread *td, bool resched);
121 
122 static uma_zone_t	ksiginfo_zone = NULL;
123 struct filterops sig_filtops = {
124 	.f_isfd = 0,
125 	.f_attach = filt_sigattach,
126 	.f_detach = filt_sigdetach,
127 	.f_event = filt_signal,
128 };
129 
130 static int	kern_logsigexit = 1;
131 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
132     &kern_logsigexit, 0,
133     "Log processes quitting on abnormal signals to syslog(3)");
134 
135 static int	kern_forcesigexit = 1;
136 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
137     &kern_forcesigexit, 0, "Force trap signal to be handled");
138 
139 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
140     "POSIX real time signal");
141 
142 static int	max_pending_per_proc = 128;
143 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
144     &max_pending_per_proc, 0, "Max pending signals per proc");
145 
146 static int	preallocate_siginfo = 1024;
147 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
148     &preallocate_siginfo, 0, "Preallocated signal memory size");
149 
150 static int	signal_overflow = 0;
151 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
152     &signal_overflow, 0, "Number of signals overflew");
153 
154 static int	signal_alloc_fail = 0;
155 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
156     &signal_alloc_fail, 0, "signals failed to be allocated");
157 
158 static int	kern_lognosys = 0;
159 SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
160     "Log invalid syscalls");
161 
162 __read_frequently bool sigfastblock_fetch_always = false;
163 SYSCTL_BOOL(_kern, OID_AUTO, sigfastblock_fetch_always, CTLFLAG_RWTUN,
164     &sigfastblock_fetch_always, 0,
165     "Fetch sigfastblock word on each syscall entry for proper "
166     "blocking semantic");
167 
168 static bool	kern_sig_discard_ign = true;
169 SYSCTL_BOOL(_kern, OID_AUTO, sig_discard_ign, CTLFLAG_RWTUN,
170     &kern_sig_discard_ign, 0,
171     "Discard ignored signals on delivery, otherwise queue them to "
172     "the target queue");
173 
174 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
175 
176 /*
177  * Policy -- Can ucred cr1 send SIGIO to process cr2?
178  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
179  * in the right situations.
180  */
181 #define CANSIGIO(cr1, cr2) \
182 	((cr1)->cr_uid == 0 || \
183 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
184 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
185 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
186 	    (cr1)->cr_uid == (cr2)->cr_uid)
187 
188 static int	sugid_coredump;
189 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
190     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
191 
192 static int	capmode_coredump;
193 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
194     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
195 
196 static int	do_coredump = 1;
197 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
198 	&do_coredump, 0, "Enable/Disable coredumps");
199 
200 static int	set_core_nodump_flag = 0;
201 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
202 	0, "Enable setting the NODUMP flag on coredump files");
203 
204 static int	coredump_devctl = 0;
205 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
206 	0, "Generate a devctl notification when processes coredump");
207 
208 /*
209  * Signal properties and actions.
210  * The array below categorizes the signals and their default actions
211  * according to the following properties:
212  */
213 #define	SIGPROP_KILL		0x01	/* terminates process by default */
214 #define	SIGPROP_CORE		0x02	/* ditto and coredumps */
215 #define	SIGPROP_STOP		0x04	/* suspend process */
216 #define	SIGPROP_TTYSTOP		0x08	/* ditto, from tty */
217 #define	SIGPROP_IGNORE		0x10	/* ignore by default */
218 #define	SIGPROP_CONT		0x20	/* continue if suspended */
219 
220 static const int sigproptbl[NSIG] = {
221 	[SIGHUP] =	SIGPROP_KILL,
222 	[SIGINT] =	SIGPROP_KILL,
223 	[SIGQUIT] =	SIGPROP_KILL | SIGPROP_CORE,
224 	[SIGILL] =	SIGPROP_KILL | SIGPROP_CORE,
225 	[SIGTRAP] =	SIGPROP_KILL | SIGPROP_CORE,
226 	[SIGABRT] =	SIGPROP_KILL | SIGPROP_CORE,
227 	[SIGEMT] =	SIGPROP_KILL | SIGPROP_CORE,
228 	[SIGFPE] =	SIGPROP_KILL | SIGPROP_CORE,
229 	[SIGKILL] =	SIGPROP_KILL,
230 	[SIGBUS] =	SIGPROP_KILL | SIGPROP_CORE,
231 	[SIGSEGV] =	SIGPROP_KILL | SIGPROP_CORE,
232 	[SIGSYS] =	SIGPROP_KILL | SIGPROP_CORE,
233 	[SIGPIPE] =	SIGPROP_KILL,
234 	[SIGALRM] =	SIGPROP_KILL,
235 	[SIGTERM] =	SIGPROP_KILL,
236 	[SIGURG] =	SIGPROP_IGNORE,
237 	[SIGSTOP] =	SIGPROP_STOP,
238 	[SIGTSTP] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
239 	[SIGCONT] =	SIGPROP_IGNORE | SIGPROP_CONT,
240 	[SIGCHLD] =	SIGPROP_IGNORE,
241 	[SIGTTIN] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
242 	[SIGTTOU] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
243 	[SIGIO] =	SIGPROP_IGNORE,
244 	[SIGXCPU] =	SIGPROP_KILL,
245 	[SIGXFSZ] =	SIGPROP_KILL,
246 	[SIGVTALRM] =	SIGPROP_KILL,
247 	[SIGPROF] =	SIGPROP_KILL,
248 	[SIGWINCH] =	SIGPROP_IGNORE,
249 	[SIGINFO] =	SIGPROP_IGNORE,
250 	[SIGUSR1] =	SIGPROP_KILL,
251 	[SIGUSR2] =	SIGPROP_KILL,
252 };
253 
254 #define	_SIG_FOREACH_ADVANCE(i, set) ({					\
255 	int __found;							\
256 	for (;;) {							\
257 		if (__bits != 0) {					\
258 			int __sig = ffs(__bits);			\
259 			__bits &= ~(1u << (__sig - 1));			\
260 			sig = __i * sizeof((set)->__bits[0]) * NBBY + __sig; \
261 			__found = 1;					\
262 			break;						\
263 		}							\
264 		if (++__i == _SIG_WORDS) {				\
265 			__found = 0;					\
266 			break;						\
267 		}							\
268 		__bits = (set)->__bits[__i];				\
269 	}								\
270 	__found != 0;							\
271 })
272 
273 #define	SIG_FOREACH(i, set)						\
274 	for (int32_t __i = -1, __bits = 0;				\
275 	    _SIG_FOREACH_ADVANCE(i, set); )				\
276 
277 static sigset_t fastblock_mask;
278 
279 static void
280 ast_sig(struct thread *td, int tda)
281 {
282 	struct proc *p;
283 	int old_boundary, sig;
284 	bool resched_sigs;
285 
286 	p = td->td_proc;
287 
288 #ifdef DIAGNOSTIC
289 	if (p->p_numthreads == 1 && (tda & (TDAI(TDA_SIG) |
290 	    TDAI(TDA_AST))) == 0) {
291 		PROC_LOCK(p);
292 		thread_lock(td);
293 		/*
294 		 * Note that TDA_SIG should be re-read from
295 		 * td_ast, since signal might have been delivered
296 		 * after we cleared td_flags above.  This is one of
297 		 * the reason for looping check for AST condition.
298 		 * See comment in userret() about P_PPWAIT.
299 		 */
300 		if ((p->p_flag & P_PPWAIT) == 0 &&
301 		    (td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
302 			if (SIGPENDING(td) && ((tda | td->td_ast) &
303 			    (TDAI(TDA_SIG) | TDAI(TDA_AST))) == 0) {
304 				thread_unlock(td); /* fix dumps */
305 				panic(
306 				    "failed2 to set signal flags for ast p %p "
307 				    "td %p tda %#x td_ast %#x fl %#x",
308 				    p, td, tda, td->td_ast, td->td_flags);
309 			}
310 		}
311 		thread_unlock(td);
312 		PROC_UNLOCK(p);
313 	}
314 #endif
315 
316 	/*
317 	 * Check for signals. Unlocked reads of p_pendingcnt or
318 	 * p_siglist might cause process-directed signal to be handled
319 	 * later.
320 	 */
321 	if ((tda & TDAI(TDA_SIG)) != 0 || p->p_pendingcnt > 0 ||
322 	    !SIGISEMPTY(p->p_siglist)) {
323 		sigfastblock_fetch(td);
324 		PROC_LOCK(p);
325 		old_boundary = ~TDB_BOUNDARY | (td->td_dbgflags & TDB_BOUNDARY);
326 		td->td_dbgflags |= TDB_BOUNDARY;
327 		mtx_lock(&p->p_sigacts->ps_mtx);
328 		while ((sig = cursig(td)) != 0) {
329 			KASSERT(sig >= 0, ("sig %d", sig));
330 			postsig(sig);
331 		}
332 		mtx_unlock(&p->p_sigacts->ps_mtx);
333 		td->td_dbgflags &= old_boundary;
334 		PROC_UNLOCK(p);
335 		resched_sigs = true;
336 	} else {
337 		resched_sigs = false;
338 	}
339 
340 	/*
341 	 * Handle deferred update of the fast sigblock value, after
342 	 * the postsig() loop was performed.
343 	 */
344 	sigfastblock_setpend(td, resched_sigs);
345 }
346 
347 static void
348 ast_sigsuspend(struct thread *td, int tda __unused)
349 {
350 	MPASS((td->td_pflags & TDP_OLDMASK) != 0);
351 	td->td_pflags &= ~TDP_OLDMASK;
352 	kern_sigprocmask(td, SIG_SETMASK, &td->td_oldsigmask, NULL, 0);
353 }
354 
355 static void
356 sigqueue_start(void)
357 {
358 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
359 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
360 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
361 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
362 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
363 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
364 	SIGFILLSET(fastblock_mask);
365 	SIG_CANTMASK(fastblock_mask);
366 	ast_register(TDA_SIG, ASTR_UNCOND, 0, ast_sig);
367 	ast_register(TDA_SIGSUSPEND, ASTR_ASTF_REQUIRED | ASTR_TDP,
368 	    TDP_OLDMASK, ast_sigsuspend);
369 }
370 
371 ksiginfo_t *
372 ksiginfo_alloc(int mwait)
373 {
374 	MPASS(mwait == M_WAITOK || mwait == M_NOWAIT);
375 
376 	if (ksiginfo_zone == NULL)
377 		return (NULL);
378 	return (uma_zalloc(ksiginfo_zone, mwait | M_ZERO));
379 }
380 
381 void
382 ksiginfo_free(ksiginfo_t *ksi)
383 {
384 	uma_zfree(ksiginfo_zone, ksi);
385 }
386 
387 static __inline bool
388 ksiginfo_tryfree(ksiginfo_t *ksi)
389 {
390 	if ((ksi->ksi_flags & KSI_EXT) == 0) {
391 		uma_zfree(ksiginfo_zone, ksi);
392 		return (true);
393 	}
394 	return (false);
395 }
396 
397 void
398 sigqueue_init(sigqueue_t *list, struct proc *p)
399 {
400 	SIGEMPTYSET(list->sq_signals);
401 	SIGEMPTYSET(list->sq_kill);
402 	SIGEMPTYSET(list->sq_ptrace);
403 	TAILQ_INIT(&list->sq_list);
404 	list->sq_proc = p;
405 	list->sq_flags = SQ_INIT;
406 }
407 
408 /*
409  * Get a signal's ksiginfo.
410  * Return:
411  *	0	-	signal not found
412  *	others	-	signal number
413  */
414 static int
415 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
416 {
417 	struct proc *p = sq->sq_proc;
418 	struct ksiginfo *ksi, *next;
419 	int count = 0;
420 
421 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
422 
423 	if (!SIGISMEMBER(sq->sq_signals, signo))
424 		return (0);
425 
426 	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
427 		count++;
428 		SIGDELSET(sq->sq_ptrace, signo);
429 		si->ksi_flags |= KSI_PTRACE;
430 	}
431 	if (SIGISMEMBER(sq->sq_kill, signo)) {
432 		count++;
433 		if (count == 1)
434 			SIGDELSET(sq->sq_kill, signo);
435 	}
436 
437 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
438 		if (ksi->ksi_signo == signo) {
439 			if (count == 0) {
440 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
441 				ksi->ksi_sigq = NULL;
442 				ksiginfo_copy(ksi, si);
443 				if (ksiginfo_tryfree(ksi) && p != NULL)
444 					p->p_pendingcnt--;
445 			}
446 			if (++count > 1)
447 				break;
448 		}
449 	}
450 
451 	if (count <= 1)
452 		SIGDELSET(sq->sq_signals, signo);
453 	si->ksi_signo = signo;
454 	return (signo);
455 }
456 
457 void
458 sigqueue_take(ksiginfo_t *ksi)
459 {
460 	struct ksiginfo *kp;
461 	struct proc	*p;
462 	sigqueue_t	*sq;
463 
464 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
465 		return;
466 
467 	p = sq->sq_proc;
468 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
469 	ksi->ksi_sigq = NULL;
470 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
471 		p->p_pendingcnt--;
472 
473 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
474 	     kp = TAILQ_NEXT(kp, ksi_link)) {
475 		if (kp->ksi_signo == ksi->ksi_signo)
476 			break;
477 	}
478 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
479 	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
480 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
481 }
482 
483 static int
484 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
485 {
486 	struct proc *p = sq->sq_proc;
487 	struct ksiginfo *ksi;
488 	int ret = 0;
489 
490 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
491 
492 	/*
493 	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
494 	 * for these signals.
495 	 */
496 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
497 		SIGADDSET(sq->sq_kill, signo);
498 		goto out_set_bit;
499 	}
500 
501 	/* directly insert the ksi, don't copy it */
502 	if (si->ksi_flags & KSI_INS) {
503 		if (si->ksi_flags & KSI_HEAD)
504 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
505 		else
506 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
507 		si->ksi_sigq = sq;
508 		goto out_set_bit;
509 	}
510 
511 	if (__predict_false(ksiginfo_zone == NULL)) {
512 		SIGADDSET(sq->sq_kill, signo);
513 		goto out_set_bit;
514 	}
515 
516 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
517 		signal_overflow++;
518 		ret = EAGAIN;
519 	} else if ((ksi = ksiginfo_alloc(M_NOWAIT)) == NULL) {
520 		signal_alloc_fail++;
521 		ret = EAGAIN;
522 	} else {
523 		if (p != NULL)
524 			p->p_pendingcnt++;
525 		ksiginfo_copy(si, ksi);
526 		ksi->ksi_signo = signo;
527 		if (si->ksi_flags & KSI_HEAD)
528 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
529 		else
530 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
531 		ksi->ksi_sigq = sq;
532 	}
533 
534 	if (ret != 0) {
535 		if ((si->ksi_flags & KSI_PTRACE) != 0) {
536 			SIGADDSET(sq->sq_ptrace, signo);
537 			ret = 0;
538 			goto out_set_bit;
539 		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
540 		    (si->ksi_flags & KSI_SIGQ) == 0) {
541 			SIGADDSET(sq->sq_kill, signo);
542 			ret = 0;
543 			goto out_set_bit;
544 		}
545 		return (ret);
546 	}
547 
548 out_set_bit:
549 	SIGADDSET(sq->sq_signals, signo);
550 	return (ret);
551 }
552 
553 void
554 sigqueue_flush(sigqueue_t *sq)
555 {
556 	struct proc *p = sq->sq_proc;
557 	ksiginfo_t *ksi;
558 
559 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
560 
561 	if (p != NULL)
562 		PROC_LOCK_ASSERT(p, MA_OWNED);
563 
564 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
565 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
566 		ksi->ksi_sigq = NULL;
567 		if (ksiginfo_tryfree(ksi) && p != NULL)
568 			p->p_pendingcnt--;
569 	}
570 
571 	SIGEMPTYSET(sq->sq_signals);
572 	SIGEMPTYSET(sq->sq_kill);
573 	SIGEMPTYSET(sq->sq_ptrace);
574 }
575 
576 static void
577 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
578 {
579 	sigset_t tmp;
580 	struct proc *p1, *p2;
581 	ksiginfo_t *ksi, *next;
582 
583 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
584 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
585 	p1 = src->sq_proc;
586 	p2 = dst->sq_proc;
587 	/* Move siginfo to target list */
588 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
589 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
590 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
591 			if (p1 != NULL)
592 				p1->p_pendingcnt--;
593 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
594 			ksi->ksi_sigq = dst;
595 			if (p2 != NULL)
596 				p2->p_pendingcnt++;
597 		}
598 	}
599 
600 	/* Move pending bits to target list */
601 	tmp = src->sq_kill;
602 	SIGSETAND(tmp, *set);
603 	SIGSETOR(dst->sq_kill, tmp);
604 	SIGSETNAND(src->sq_kill, tmp);
605 
606 	tmp = src->sq_ptrace;
607 	SIGSETAND(tmp, *set);
608 	SIGSETOR(dst->sq_ptrace, tmp);
609 	SIGSETNAND(src->sq_ptrace, tmp);
610 
611 	tmp = src->sq_signals;
612 	SIGSETAND(tmp, *set);
613 	SIGSETOR(dst->sq_signals, tmp);
614 	SIGSETNAND(src->sq_signals, tmp);
615 }
616 
617 #if 0
618 static void
619 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
620 {
621 	sigset_t set;
622 
623 	SIGEMPTYSET(set);
624 	SIGADDSET(set, signo);
625 	sigqueue_move_set(src, dst, &set);
626 }
627 #endif
628 
629 static void
630 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
631 {
632 	struct proc *p = sq->sq_proc;
633 	ksiginfo_t *ksi, *next;
634 
635 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
636 
637 	/* Remove siginfo queue */
638 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
639 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
640 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
641 			ksi->ksi_sigq = NULL;
642 			if (ksiginfo_tryfree(ksi) && p != NULL)
643 				p->p_pendingcnt--;
644 		}
645 	}
646 	SIGSETNAND(sq->sq_kill, *set);
647 	SIGSETNAND(sq->sq_ptrace, *set);
648 	SIGSETNAND(sq->sq_signals, *set);
649 }
650 
651 void
652 sigqueue_delete(sigqueue_t *sq, int signo)
653 {
654 	sigset_t set;
655 
656 	SIGEMPTYSET(set);
657 	SIGADDSET(set, signo);
658 	sigqueue_delete_set(sq, &set);
659 }
660 
661 /* Remove a set of signals for a process */
662 static void
663 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
664 {
665 	sigqueue_t worklist;
666 	struct thread *td0;
667 
668 	PROC_LOCK_ASSERT(p, MA_OWNED);
669 
670 	sigqueue_init(&worklist, NULL);
671 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
672 
673 	FOREACH_THREAD_IN_PROC(p, td0)
674 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
675 
676 	sigqueue_flush(&worklist);
677 }
678 
679 void
680 sigqueue_delete_proc(struct proc *p, int signo)
681 {
682 	sigset_t set;
683 
684 	SIGEMPTYSET(set);
685 	SIGADDSET(set, signo);
686 	sigqueue_delete_set_proc(p, &set);
687 }
688 
689 static void
690 sigqueue_delete_stopmask_proc(struct proc *p)
691 {
692 	sigset_t set;
693 
694 	SIGEMPTYSET(set);
695 	SIGADDSET(set, SIGSTOP);
696 	SIGADDSET(set, SIGTSTP);
697 	SIGADDSET(set, SIGTTIN);
698 	SIGADDSET(set, SIGTTOU);
699 	sigqueue_delete_set_proc(p, &set);
700 }
701 
702 /*
703  * Determine signal that should be delivered to thread td, the current
704  * thread, 0 if none.  If there is a pending stop signal with default
705  * action, the process stops in issignal().
706  */
707 int
708 cursig(struct thread *td)
709 {
710 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
711 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
712 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
713 	return (SIGPENDING(td) ? issignal(td) : 0);
714 }
715 
716 /*
717  * Arrange for ast() to handle unmasked pending signals on return to user
718  * mode.  This must be called whenever a signal is added to td_sigqueue or
719  * unmasked in td_sigmask.
720  */
721 void
722 signotify(struct thread *td)
723 {
724 
725 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
726 
727 	if (SIGPENDING(td))
728 		ast_sched(td, TDA_SIG);
729 }
730 
731 /*
732  * Returns 1 (true) if altstack is configured for the thread, and the
733  * passed stack bottom address falls into the altstack range.  Handles
734  * the 43 compat special case where the alt stack size is zero.
735  */
736 int
737 sigonstack(size_t sp)
738 {
739 	struct thread *td;
740 
741 	td = curthread;
742 	if ((td->td_pflags & TDP_ALTSTACK) == 0)
743 		return (0);
744 #if defined(COMPAT_43)
745 	if (SV_PROC_FLAG(td->td_proc, SV_AOUT) && td->td_sigstk.ss_size == 0)
746 		return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
747 #endif
748 	return (sp >= (size_t)td->td_sigstk.ss_sp &&
749 	    sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
750 }
751 
752 static __inline int
753 sigprop(int sig)
754 {
755 
756 	if (sig > 0 && sig < nitems(sigproptbl))
757 		return (sigproptbl[sig]);
758 	return (0);
759 }
760 
761 static bool
762 sigact_flag_test(const struct sigaction *act, int flag)
763 {
764 
765 	/*
766 	 * SA_SIGINFO is reset when signal disposition is set to
767 	 * ignore or default.  Other flags are kept according to user
768 	 * settings.
769 	 */
770 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
771 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
772 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
773 }
774 
775 /*
776  * kern_sigaction
777  * sigaction
778  * freebsd4_sigaction
779  * osigaction
780  */
781 int
782 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
783     struct sigaction *oact, int flags)
784 {
785 	struct sigacts *ps;
786 	struct proc *p = td->td_proc;
787 
788 	if (!_SIG_VALID(sig))
789 		return (EINVAL);
790 	if (act != NULL && act->sa_handler != SIG_DFL &&
791 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
792 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
793 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
794 		return (EINVAL);
795 
796 	PROC_LOCK(p);
797 	ps = p->p_sigacts;
798 	mtx_lock(&ps->ps_mtx);
799 	if (oact) {
800 		memset(oact, 0, sizeof(*oact));
801 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
802 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
803 			oact->sa_flags |= SA_ONSTACK;
804 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
805 			oact->sa_flags |= SA_RESTART;
806 		if (SIGISMEMBER(ps->ps_sigreset, sig))
807 			oact->sa_flags |= SA_RESETHAND;
808 		if (SIGISMEMBER(ps->ps_signodefer, sig))
809 			oact->sa_flags |= SA_NODEFER;
810 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
811 			oact->sa_flags |= SA_SIGINFO;
812 			oact->sa_sigaction =
813 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
814 		} else
815 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
816 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
817 			oact->sa_flags |= SA_NOCLDSTOP;
818 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
819 			oact->sa_flags |= SA_NOCLDWAIT;
820 	}
821 	if (act) {
822 		if ((sig == SIGKILL || sig == SIGSTOP) &&
823 		    act->sa_handler != SIG_DFL) {
824 			mtx_unlock(&ps->ps_mtx);
825 			PROC_UNLOCK(p);
826 			return (EINVAL);
827 		}
828 
829 		/*
830 		 * Change setting atomically.
831 		 */
832 
833 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
834 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
835 		if (sigact_flag_test(act, SA_SIGINFO)) {
836 			ps->ps_sigact[_SIG_IDX(sig)] =
837 			    (__sighandler_t *)act->sa_sigaction;
838 			SIGADDSET(ps->ps_siginfo, sig);
839 		} else {
840 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
841 			SIGDELSET(ps->ps_siginfo, sig);
842 		}
843 		if (!sigact_flag_test(act, SA_RESTART))
844 			SIGADDSET(ps->ps_sigintr, sig);
845 		else
846 			SIGDELSET(ps->ps_sigintr, sig);
847 		if (sigact_flag_test(act, SA_ONSTACK))
848 			SIGADDSET(ps->ps_sigonstack, sig);
849 		else
850 			SIGDELSET(ps->ps_sigonstack, sig);
851 		if (sigact_flag_test(act, SA_RESETHAND))
852 			SIGADDSET(ps->ps_sigreset, sig);
853 		else
854 			SIGDELSET(ps->ps_sigreset, sig);
855 		if (sigact_flag_test(act, SA_NODEFER))
856 			SIGADDSET(ps->ps_signodefer, sig);
857 		else
858 			SIGDELSET(ps->ps_signodefer, sig);
859 		if (sig == SIGCHLD) {
860 			if (act->sa_flags & SA_NOCLDSTOP)
861 				ps->ps_flag |= PS_NOCLDSTOP;
862 			else
863 				ps->ps_flag &= ~PS_NOCLDSTOP;
864 			if (act->sa_flags & SA_NOCLDWAIT) {
865 				/*
866 				 * Paranoia: since SA_NOCLDWAIT is implemented
867 				 * by reparenting the dying child to PID 1 (and
868 				 * trust it to reap the zombie), PID 1 itself
869 				 * is forbidden to set SA_NOCLDWAIT.
870 				 */
871 				if (p->p_pid == 1)
872 					ps->ps_flag &= ~PS_NOCLDWAIT;
873 				else
874 					ps->ps_flag |= PS_NOCLDWAIT;
875 			} else
876 				ps->ps_flag &= ~PS_NOCLDWAIT;
877 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
878 				ps->ps_flag |= PS_CLDSIGIGN;
879 			else
880 				ps->ps_flag &= ~PS_CLDSIGIGN;
881 		}
882 		/*
883 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
884 		 * and for signals set to SIG_DFL where the default is to
885 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
886 		 * have to restart the process.
887 		 */
888 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
889 		    (sigprop(sig) & SIGPROP_IGNORE &&
890 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
891 			/* never to be seen again */
892 			sigqueue_delete_proc(p, sig);
893 			if (sig != SIGCONT)
894 				/* easier in psignal */
895 				SIGADDSET(ps->ps_sigignore, sig);
896 			SIGDELSET(ps->ps_sigcatch, sig);
897 		} else {
898 			SIGDELSET(ps->ps_sigignore, sig);
899 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
900 				SIGDELSET(ps->ps_sigcatch, sig);
901 			else
902 				SIGADDSET(ps->ps_sigcatch, sig);
903 		}
904 #ifdef COMPAT_FREEBSD4
905 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
906 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
907 		    (flags & KSA_FREEBSD4) == 0)
908 			SIGDELSET(ps->ps_freebsd4, sig);
909 		else
910 			SIGADDSET(ps->ps_freebsd4, sig);
911 #endif
912 #ifdef COMPAT_43
913 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
914 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
915 		    (flags & KSA_OSIGSET) == 0)
916 			SIGDELSET(ps->ps_osigset, sig);
917 		else
918 			SIGADDSET(ps->ps_osigset, sig);
919 #endif
920 	}
921 	mtx_unlock(&ps->ps_mtx);
922 	PROC_UNLOCK(p);
923 	return (0);
924 }
925 
926 #ifndef _SYS_SYSPROTO_H_
927 struct sigaction_args {
928 	int	sig;
929 	struct	sigaction *act;
930 	struct	sigaction *oact;
931 };
932 #endif
933 int
934 sys_sigaction(struct thread *td, struct sigaction_args *uap)
935 {
936 	struct sigaction act, oact;
937 	struct sigaction *actp, *oactp;
938 	int error;
939 
940 	actp = (uap->act != NULL) ? &act : NULL;
941 	oactp = (uap->oact != NULL) ? &oact : NULL;
942 	if (actp) {
943 		error = copyin(uap->act, actp, sizeof(act));
944 		if (error)
945 			return (error);
946 	}
947 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
948 	if (oactp && !error)
949 		error = copyout(oactp, uap->oact, sizeof(oact));
950 	return (error);
951 }
952 
953 #ifdef COMPAT_FREEBSD4
954 #ifndef _SYS_SYSPROTO_H_
955 struct freebsd4_sigaction_args {
956 	int	sig;
957 	struct	sigaction *act;
958 	struct	sigaction *oact;
959 };
960 #endif
961 int
962 freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
963 {
964 	struct sigaction act, oact;
965 	struct sigaction *actp, *oactp;
966 	int error;
967 
968 	actp = (uap->act != NULL) ? &act : NULL;
969 	oactp = (uap->oact != NULL) ? &oact : NULL;
970 	if (actp) {
971 		error = copyin(uap->act, actp, sizeof(act));
972 		if (error)
973 			return (error);
974 	}
975 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
976 	if (oactp && !error)
977 		error = copyout(oactp, uap->oact, sizeof(oact));
978 	return (error);
979 }
980 #endif	/* COMAPT_FREEBSD4 */
981 
982 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
983 #ifndef _SYS_SYSPROTO_H_
984 struct osigaction_args {
985 	int	signum;
986 	struct	osigaction *nsa;
987 	struct	osigaction *osa;
988 };
989 #endif
990 int
991 osigaction(struct thread *td, struct osigaction_args *uap)
992 {
993 	struct osigaction sa;
994 	struct sigaction nsa, osa;
995 	struct sigaction *nsap, *osap;
996 	int error;
997 
998 	if (uap->signum <= 0 || uap->signum >= ONSIG)
999 		return (EINVAL);
1000 
1001 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
1002 	osap = (uap->osa != NULL) ? &osa : NULL;
1003 
1004 	if (nsap) {
1005 		error = copyin(uap->nsa, &sa, sizeof(sa));
1006 		if (error)
1007 			return (error);
1008 		nsap->sa_handler = sa.sa_handler;
1009 		nsap->sa_flags = sa.sa_flags;
1010 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
1011 	}
1012 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1013 	if (osap && !error) {
1014 		sa.sa_handler = osap->sa_handler;
1015 		sa.sa_flags = osap->sa_flags;
1016 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
1017 		error = copyout(&sa, uap->osa, sizeof(sa));
1018 	}
1019 	return (error);
1020 }
1021 
1022 #if !defined(__i386__)
1023 /* Avoid replicating the same stub everywhere */
1024 int
1025 osigreturn(struct thread *td, struct osigreturn_args *uap)
1026 {
1027 
1028 	return (nosys(td, (struct nosys_args *)uap));
1029 }
1030 #endif
1031 #endif /* COMPAT_43 */
1032 
1033 /*
1034  * Initialize signal state for process 0;
1035  * set to ignore signals that are ignored by default.
1036  */
1037 void
1038 siginit(struct proc *p)
1039 {
1040 	int i;
1041 	struct sigacts *ps;
1042 
1043 	PROC_LOCK(p);
1044 	ps = p->p_sigacts;
1045 	mtx_lock(&ps->ps_mtx);
1046 	for (i = 1; i <= NSIG; i++) {
1047 		if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) {
1048 			SIGADDSET(ps->ps_sigignore, i);
1049 		}
1050 	}
1051 	mtx_unlock(&ps->ps_mtx);
1052 	PROC_UNLOCK(p);
1053 }
1054 
1055 /*
1056  * Reset specified signal to the default disposition.
1057  */
1058 static void
1059 sigdflt(struct sigacts *ps, int sig)
1060 {
1061 
1062 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1063 	SIGDELSET(ps->ps_sigcatch, sig);
1064 	if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT)
1065 		SIGADDSET(ps->ps_sigignore, sig);
1066 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1067 	SIGDELSET(ps->ps_siginfo, sig);
1068 }
1069 
1070 /*
1071  * Reset signals for an exec of the specified process.
1072  */
1073 void
1074 execsigs(struct proc *p)
1075 {
1076 	struct sigacts *ps;
1077 	struct thread *td;
1078 
1079 	/*
1080 	 * Reset caught signals.  Held signals remain held
1081 	 * through td_sigmask (unless they were caught,
1082 	 * and are now ignored by default).
1083 	 */
1084 	PROC_LOCK_ASSERT(p, MA_OWNED);
1085 	ps = p->p_sigacts;
1086 	mtx_lock(&ps->ps_mtx);
1087 	sig_drop_caught(p);
1088 
1089 	/*
1090 	 * Reset stack state to the user stack.
1091 	 * Clear set of signals caught on the signal stack.
1092 	 */
1093 	td = curthread;
1094 	MPASS(td->td_proc == p);
1095 	td->td_sigstk.ss_flags = SS_DISABLE;
1096 	td->td_sigstk.ss_size = 0;
1097 	td->td_sigstk.ss_sp = 0;
1098 	td->td_pflags &= ~TDP_ALTSTACK;
1099 	/*
1100 	 * Reset no zombies if child dies flag as Solaris does.
1101 	 */
1102 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1103 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1104 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1105 	mtx_unlock(&ps->ps_mtx);
1106 }
1107 
1108 /*
1109  * kern_sigprocmask()
1110  *
1111  *	Manipulate signal mask.
1112  */
1113 int
1114 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1115     int flags)
1116 {
1117 	sigset_t new_block, oset1;
1118 	struct proc *p;
1119 	int error;
1120 
1121 	p = td->td_proc;
1122 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1123 		PROC_LOCK_ASSERT(p, MA_OWNED);
1124 	else
1125 		PROC_LOCK(p);
1126 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1127 	    ? MA_OWNED : MA_NOTOWNED);
1128 	if (oset != NULL)
1129 		*oset = td->td_sigmask;
1130 
1131 	error = 0;
1132 	if (set != NULL) {
1133 		switch (how) {
1134 		case SIG_BLOCK:
1135 			SIG_CANTMASK(*set);
1136 			oset1 = td->td_sigmask;
1137 			SIGSETOR(td->td_sigmask, *set);
1138 			new_block = td->td_sigmask;
1139 			SIGSETNAND(new_block, oset1);
1140 			break;
1141 		case SIG_UNBLOCK:
1142 			SIGSETNAND(td->td_sigmask, *set);
1143 			signotify(td);
1144 			goto out;
1145 		case SIG_SETMASK:
1146 			SIG_CANTMASK(*set);
1147 			oset1 = td->td_sigmask;
1148 			if (flags & SIGPROCMASK_OLD)
1149 				SIGSETLO(td->td_sigmask, *set);
1150 			else
1151 				td->td_sigmask = *set;
1152 			new_block = td->td_sigmask;
1153 			SIGSETNAND(new_block, oset1);
1154 			signotify(td);
1155 			break;
1156 		default:
1157 			error = EINVAL;
1158 			goto out;
1159 		}
1160 
1161 		/*
1162 		 * The new_block set contains signals that were not previously
1163 		 * blocked, but are blocked now.
1164 		 *
1165 		 * In case we block any signal that was not previously blocked
1166 		 * for td, and process has the signal pending, try to schedule
1167 		 * signal delivery to some thread that does not block the
1168 		 * signal, possibly waking it up.
1169 		 */
1170 		if (p->p_numthreads != 1)
1171 			reschedule_signals(p, new_block, flags);
1172 	}
1173 
1174 out:
1175 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1176 		PROC_UNLOCK(p);
1177 	return (error);
1178 }
1179 
1180 #ifndef _SYS_SYSPROTO_H_
1181 struct sigprocmask_args {
1182 	int	how;
1183 	const sigset_t *set;
1184 	sigset_t *oset;
1185 };
1186 #endif
1187 int
1188 sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1189 {
1190 	sigset_t set, oset;
1191 	sigset_t *setp, *osetp;
1192 	int error;
1193 
1194 	setp = (uap->set != NULL) ? &set : NULL;
1195 	osetp = (uap->oset != NULL) ? &oset : NULL;
1196 	if (setp) {
1197 		error = copyin(uap->set, setp, sizeof(set));
1198 		if (error)
1199 			return (error);
1200 	}
1201 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1202 	if (osetp && !error) {
1203 		error = copyout(osetp, uap->oset, sizeof(oset));
1204 	}
1205 	return (error);
1206 }
1207 
1208 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1209 #ifndef _SYS_SYSPROTO_H_
1210 struct osigprocmask_args {
1211 	int	how;
1212 	osigset_t mask;
1213 };
1214 #endif
1215 int
1216 osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1217 {
1218 	sigset_t set, oset;
1219 	int error;
1220 
1221 	OSIG2SIG(uap->mask, set);
1222 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1223 	SIG2OSIG(oset, td->td_retval[0]);
1224 	return (error);
1225 }
1226 #endif /* COMPAT_43 */
1227 
1228 int
1229 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1230 {
1231 	ksiginfo_t ksi;
1232 	sigset_t set;
1233 	int error;
1234 
1235 	error = copyin(uap->set, &set, sizeof(set));
1236 	if (error) {
1237 		td->td_retval[0] = error;
1238 		return (0);
1239 	}
1240 
1241 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1242 	if (error) {
1243 		/*
1244 		 * sigwait() function shall not return EINTR, but
1245 		 * the syscall does.  Non-ancient libc provides the
1246 		 * wrapper which hides EINTR.  Otherwise, EINTR return
1247 		 * is used by libthr to handle required cancellation
1248 		 * point in the sigwait().
1249 		 */
1250 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1251 			return (ERESTART);
1252 		td->td_retval[0] = error;
1253 		return (0);
1254 	}
1255 
1256 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1257 	td->td_retval[0] = error;
1258 	return (0);
1259 }
1260 
1261 int
1262 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1263 {
1264 	struct timespec ts;
1265 	struct timespec *timeout;
1266 	sigset_t set;
1267 	ksiginfo_t ksi;
1268 	int error;
1269 
1270 	if (uap->timeout) {
1271 		error = copyin(uap->timeout, &ts, sizeof(ts));
1272 		if (error)
1273 			return (error);
1274 
1275 		timeout = &ts;
1276 	} else
1277 		timeout = NULL;
1278 
1279 	error = copyin(uap->set, &set, sizeof(set));
1280 	if (error)
1281 		return (error);
1282 
1283 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1284 	if (error)
1285 		return (error);
1286 
1287 	if (uap->info)
1288 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1289 
1290 	if (error == 0)
1291 		td->td_retval[0] = ksi.ksi_signo;
1292 	return (error);
1293 }
1294 
1295 int
1296 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1297 {
1298 	ksiginfo_t ksi;
1299 	sigset_t set;
1300 	int error;
1301 
1302 	error = copyin(uap->set, &set, sizeof(set));
1303 	if (error)
1304 		return (error);
1305 
1306 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1307 	if (error)
1308 		return (error);
1309 
1310 	if (uap->info)
1311 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1312 
1313 	if (error == 0)
1314 		td->td_retval[0] = ksi.ksi_signo;
1315 	return (error);
1316 }
1317 
1318 static void
1319 proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1320 {
1321 	struct thread *thr;
1322 
1323 	FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1324 		if (thr == td)
1325 			thr->td_si = *si;
1326 		else
1327 			thr->td_si.si_signo = 0;
1328 	}
1329 }
1330 
1331 int
1332 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1333 	struct timespec *timeout)
1334 {
1335 	struct sigacts *ps;
1336 	sigset_t saved_mask, new_block;
1337 	struct proc *p;
1338 	int error, sig, timevalid = 0;
1339 	sbintime_t sbt, precision, tsbt;
1340 	struct timespec ts;
1341 	bool traced;
1342 
1343 	p = td->td_proc;
1344 	error = 0;
1345 	traced = false;
1346 
1347 	/* Ensure the sigfastblock value is up to date. */
1348 	sigfastblock_fetch(td);
1349 
1350 	if (timeout != NULL) {
1351 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1352 			timevalid = 1;
1353 			ts = *timeout;
1354 			if (ts.tv_sec < INT32_MAX / 2) {
1355 				tsbt = tstosbt(ts);
1356 				precision = tsbt;
1357 				precision >>= tc_precexp;
1358 				if (TIMESEL(&sbt, tsbt))
1359 					sbt += tc_tick_sbt;
1360 				sbt += tsbt;
1361 			} else
1362 				precision = sbt = 0;
1363 		}
1364 	} else
1365 		precision = sbt = 0;
1366 	ksiginfo_init(ksi);
1367 	/* Some signals can not be waited for. */
1368 	SIG_CANTMASK(waitset);
1369 	ps = p->p_sigacts;
1370 	PROC_LOCK(p);
1371 	saved_mask = td->td_sigmask;
1372 	SIGSETNAND(td->td_sigmask, waitset);
1373 	if ((p->p_sysent->sv_flags & SV_SIG_DISCIGN) != 0 ||
1374 	    !kern_sig_discard_ign) {
1375 		thread_lock(td);
1376 		td->td_flags |= TDF_SIGWAIT;
1377 		thread_unlock(td);
1378 	}
1379 	for (;;) {
1380 		mtx_lock(&ps->ps_mtx);
1381 		sig = cursig(td);
1382 		mtx_unlock(&ps->ps_mtx);
1383 		KASSERT(sig >= 0, ("sig %d", sig));
1384 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1385 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1386 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1387 				error = 0;
1388 				break;
1389 			}
1390 		}
1391 
1392 		if (error != 0)
1393 			break;
1394 
1395 		/*
1396 		 * POSIX says this must be checked after looking for pending
1397 		 * signals.
1398 		 */
1399 		if (timeout != NULL && !timevalid) {
1400 			error = EINVAL;
1401 			break;
1402 		}
1403 
1404 		if (traced) {
1405 			error = EINTR;
1406 			break;
1407 		}
1408 
1409 		error = msleep_sbt(&p->p_sigacts, &p->p_mtx, PPAUSE | PCATCH,
1410 		    "sigwait", sbt, precision, C_ABSOLUTE);
1411 
1412 		/* The syscalls can not be restarted. */
1413 		if (error == ERESTART)
1414 			error = EINTR;
1415 
1416 		/*
1417 		 * If PTRACE_SCE or PTRACE_SCX were set after
1418 		 * userspace entered the syscall, return spurious
1419 		 * EINTR after wait was done.  Only do this as last
1420 		 * resort after rechecking for possible queued signals
1421 		 * and expired timeouts.
1422 		 */
1423 		if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1424 			traced = true;
1425 	}
1426 	thread_lock(td);
1427 	td->td_flags &= ~TDF_SIGWAIT;
1428 	thread_unlock(td);
1429 
1430 	new_block = saved_mask;
1431 	SIGSETNAND(new_block, td->td_sigmask);
1432 	td->td_sigmask = saved_mask;
1433 	/*
1434 	 * Fewer signals can be delivered to us, reschedule signal
1435 	 * notification.
1436 	 */
1437 	if (p->p_numthreads != 1)
1438 		reschedule_signals(p, new_block, 0);
1439 
1440 	if (error == 0) {
1441 		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1442 
1443 		if (ksi->ksi_code == SI_TIMER)
1444 			itimer_accept(p, ksi->ksi_timerid, ksi);
1445 
1446 #ifdef KTRACE
1447 		if (KTRPOINT(td, KTR_PSIG)) {
1448 			sig_t action;
1449 
1450 			mtx_lock(&ps->ps_mtx);
1451 			action = ps->ps_sigact[_SIG_IDX(sig)];
1452 			mtx_unlock(&ps->ps_mtx);
1453 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1454 		}
1455 #endif
1456 		if (sig == SIGKILL) {
1457 			proc_td_siginfo_capture(td, &ksi->ksi_info);
1458 			sigexit(td, sig);
1459 		}
1460 	}
1461 	PROC_UNLOCK(p);
1462 	return (error);
1463 }
1464 
1465 #ifndef _SYS_SYSPROTO_H_
1466 struct sigpending_args {
1467 	sigset_t	*set;
1468 };
1469 #endif
1470 int
1471 sys_sigpending(struct thread *td, struct sigpending_args *uap)
1472 {
1473 	struct proc *p = td->td_proc;
1474 	sigset_t pending;
1475 
1476 	PROC_LOCK(p);
1477 	pending = p->p_sigqueue.sq_signals;
1478 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1479 	PROC_UNLOCK(p);
1480 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1481 }
1482 
1483 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1484 #ifndef _SYS_SYSPROTO_H_
1485 struct osigpending_args {
1486 	int	dummy;
1487 };
1488 #endif
1489 int
1490 osigpending(struct thread *td, struct osigpending_args *uap)
1491 {
1492 	struct proc *p = td->td_proc;
1493 	sigset_t pending;
1494 
1495 	PROC_LOCK(p);
1496 	pending = p->p_sigqueue.sq_signals;
1497 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1498 	PROC_UNLOCK(p);
1499 	SIG2OSIG(pending, td->td_retval[0]);
1500 	return (0);
1501 }
1502 #endif /* COMPAT_43 */
1503 
1504 #if defined(COMPAT_43)
1505 /*
1506  * Generalized interface signal handler, 4.3-compatible.
1507  */
1508 #ifndef _SYS_SYSPROTO_H_
1509 struct osigvec_args {
1510 	int	signum;
1511 	struct	sigvec *nsv;
1512 	struct	sigvec *osv;
1513 };
1514 #endif
1515 /* ARGSUSED */
1516 int
1517 osigvec(struct thread *td, struct osigvec_args *uap)
1518 {
1519 	struct sigvec vec;
1520 	struct sigaction nsa, osa;
1521 	struct sigaction *nsap, *osap;
1522 	int error;
1523 
1524 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1525 		return (EINVAL);
1526 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1527 	osap = (uap->osv != NULL) ? &osa : NULL;
1528 	if (nsap) {
1529 		error = copyin(uap->nsv, &vec, sizeof(vec));
1530 		if (error)
1531 			return (error);
1532 		nsap->sa_handler = vec.sv_handler;
1533 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1534 		nsap->sa_flags = vec.sv_flags;
1535 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1536 	}
1537 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1538 	if (osap && !error) {
1539 		vec.sv_handler = osap->sa_handler;
1540 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1541 		vec.sv_flags = osap->sa_flags;
1542 		vec.sv_flags &= ~SA_NOCLDWAIT;
1543 		vec.sv_flags ^= SA_RESTART;
1544 		error = copyout(&vec, uap->osv, sizeof(vec));
1545 	}
1546 	return (error);
1547 }
1548 
1549 #ifndef _SYS_SYSPROTO_H_
1550 struct osigblock_args {
1551 	int	mask;
1552 };
1553 #endif
1554 int
1555 osigblock(struct thread *td, struct osigblock_args *uap)
1556 {
1557 	sigset_t set, oset;
1558 
1559 	OSIG2SIG(uap->mask, set);
1560 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1561 	SIG2OSIG(oset, td->td_retval[0]);
1562 	return (0);
1563 }
1564 
1565 #ifndef _SYS_SYSPROTO_H_
1566 struct osigsetmask_args {
1567 	int	mask;
1568 };
1569 #endif
1570 int
1571 osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1572 {
1573 	sigset_t set, oset;
1574 
1575 	OSIG2SIG(uap->mask, set);
1576 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1577 	SIG2OSIG(oset, td->td_retval[0]);
1578 	return (0);
1579 }
1580 #endif /* COMPAT_43 */
1581 
1582 /*
1583  * Suspend calling thread until signal, providing mask to be set in the
1584  * meantime.
1585  */
1586 #ifndef _SYS_SYSPROTO_H_
1587 struct sigsuspend_args {
1588 	const sigset_t *sigmask;
1589 };
1590 #endif
1591 /* ARGSUSED */
1592 int
1593 sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1594 {
1595 	sigset_t mask;
1596 	int error;
1597 
1598 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1599 	if (error)
1600 		return (error);
1601 	return (kern_sigsuspend(td, mask));
1602 }
1603 
1604 int
1605 kern_sigsuspend(struct thread *td, sigset_t mask)
1606 {
1607 	struct proc *p = td->td_proc;
1608 	int has_sig, sig;
1609 
1610 	/* Ensure the sigfastblock value is up to date. */
1611 	sigfastblock_fetch(td);
1612 
1613 	/*
1614 	 * When returning from sigsuspend, we want
1615 	 * the old mask to be restored after the
1616 	 * signal handler has finished.  Thus, we
1617 	 * save it here and mark the sigacts structure
1618 	 * to indicate this.
1619 	 */
1620 	PROC_LOCK(p);
1621 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1622 	    SIGPROCMASK_PROC_LOCKED);
1623 	td->td_pflags |= TDP_OLDMASK;
1624 	ast_sched(td, TDA_SIGSUSPEND);
1625 
1626 	/*
1627 	 * Process signals now. Otherwise, we can get spurious wakeup
1628 	 * due to signal entered process queue, but delivered to other
1629 	 * thread. But sigsuspend should return only on signal
1630 	 * delivery.
1631 	 */
1632 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1633 	for (has_sig = 0; !has_sig;) {
1634 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1635 			0) == 0)
1636 			/* void */;
1637 		thread_suspend_check(0);
1638 		mtx_lock(&p->p_sigacts->ps_mtx);
1639 		while ((sig = cursig(td)) != 0) {
1640 			KASSERT(sig >= 0, ("sig %d", sig));
1641 			has_sig += postsig(sig);
1642 		}
1643 		mtx_unlock(&p->p_sigacts->ps_mtx);
1644 
1645 		/*
1646 		 * If PTRACE_SCE or PTRACE_SCX were set after
1647 		 * userspace entered the syscall, return spurious
1648 		 * EINTR.
1649 		 */
1650 		if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1651 			has_sig += 1;
1652 	}
1653 	PROC_UNLOCK(p);
1654 	td->td_errno = EINTR;
1655 	td->td_pflags |= TDP_NERRNO;
1656 	return (EJUSTRETURN);
1657 }
1658 
1659 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1660 /*
1661  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1662  * convention: libc stub passes mask, not pointer, to save a copyin.
1663  */
1664 #ifndef _SYS_SYSPROTO_H_
1665 struct osigsuspend_args {
1666 	osigset_t mask;
1667 };
1668 #endif
1669 /* ARGSUSED */
1670 int
1671 osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1672 {
1673 	sigset_t mask;
1674 
1675 	OSIG2SIG(uap->mask, mask);
1676 	return (kern_sigsuspend(td, mask));
1677 }
1678 #endif /* COMPAT_43 */
1679 
1680 #if defined(COMPAT_43)
1681 #ifndef _SYS_SYSPROTO_H_
1682 struct osigstack_args {
1683 	struct	sigstack *nss;
1684 	struct	sigstack *oss;
1685 };
1686 #endif
1687 /* ARGSUSED */
1688 int
1689 osigstack(struct thread *td, struct osigstack_args *uap)
1690 {
1691 	struct sigstack nss, oss;
1692 	int error = 0;
1693 
1694 	if (uap->nss != NULL) {
1695 		error = copyin(uap->nss, &nss, sizeof(nss));
1696 		if (error)
1697 			return (error);
1698 	}
1699 	oss.ss_sp = td->td_sigstk.ss_sp;
1700 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1701 	if (uap->nss != NULL) {
1702 		td->td_sigstk.ss_sp = nss.ss_sp;
1703 		td->td_sigstk.ss_size = 0;
1704 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1705 		td->td_pflags |= TDP_ALTSTACK;
1706 	}
1707 	if (uap->oss != NULL)
1708 		error = copyout(&oss, uap->oss, sizeof(oss));
1709 
1710 	return (error);
1711 }
1712 #endif /* COMPAT_43 */
1713 
1714 #ifndef _SYS_SYSPROTO_H_
1715 struct sigaltstack_args {
1716 	stack_t	*ss;
1717 	stack_t	*oss;
1718 };
1719 #endif
1720 /* ARGSUSED */
1721 int
1722 sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1723 {
1724 	stack_t ss, oss;
1725 	int error;
1726 
1727 	if (uap->ss != NULL) {
1728 		error = copyin(uap->ss, &ss, sizeof(ss));
1729 		if (error)
1730 			return (error);
1731 	}
1732 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1733 	    (uap->oss != NULL) ? &oss : NULL);
1734 	if (error)
1735 		return (error);
1736 	if (uap->oss != NULL)
1737 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1738 	return (error);
1739 }
1740 
1741 int
1742 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1743 {
1744 	struct proc *p = td->td_proc;
1745 	int oonstack;
1746 
1747 	oonstack = sigonstack(cpu_getstack(td));
1748 
1749 	if (oss != NULL) {
1750 		*oss = td->td_sigstk;
1751 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1752 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1753 	}
1754 
1755 	if (ss != NULL) {
1756 		if (oonstack)
1757 			return (EPERM);
1758 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1759 			return (EINVAL);
1760 		if (!(ss->ss_flags & SS_DISABLE)) {
1761 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1762 				return (ENOMEM);
1763 
1764 			td->td_sigstk = *ss;
1765 			td->td_pflags |= TDP_ALTSTACK;
1766 		} else {
1767 			td->td_pflags &= ~TDP_ALTSTACK;
1768 		}
1769 	}
1770 	return (0);
1771 }
1772 
1773 struct killpg1_ctx {
1774 	struct thread *td;
1775 	ksiginfo_t *ksi;
1776 	int sig;
1777 	bool sent;
1778 	bool found;
1779 	int ret;
1780 };
1781 
1782 static void
1783 killpg1_sendsig_locked(struct proc *p, struct killpg1_ctx *arg)
1784 {
1785 	int err;
1786 
1787 	err = p_cansignal(arg->td, p, arg->sig);
1788 	if (err == 0 && arg->sig != 0)
1789 		pksignal(p, arg->sig, arg->ksi);
1790 	if (err != ESRCH)
1791 		arg->found = true;
1792 	if (err == 0)
1793 		arg->sent = true;
1794 	else if (arg->ret == 0 && err != ESRCH && err != EPERM)
1795 		arg->ret = err;
1796 }
1797 
1798 static void
1799 killpg1_sendsig(struct proc *p, bool notself, struct killpg1_ctx *arg)
1800 {
1801 
1802 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1803 	    (notself && p == arg->td->td_proc) || p->p_state == PRS_NEW)
1804 		return;
1805 
1806 	PROC_LOCK(p);
1807 	killpg1_sendsig_locked(p, arg);
1808 	PROC_UNLOCK(p);
1809 }
1810 
1811 static void
1812 kill_processes_prison_cb(struct proc *p, void *arg)
1813 {
1814 	struct killpg1_ctx *ctx = arg;
1815 
1816 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1817 	    (p == ctx->td->td_proc) || p->p_state == PRS_NEW)
1818 		return;
1819 
1820 	killpg1_sendsig_locked(p, ctx);
1821 }
1822 
1823 /*
1824  * Common code for kill process group/broadcast kill.
1825  * td is the calling thread, as usual.
1826  */
1827 static int
1828 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1829 {
1830 	struct proc *p;
1831 	struct pgrp *pgrp;
1832 	struct killpg1_ctx arg;
1833 
1834 	arg.td = td;
1835 	arg.ksi = ksi;
1836 	arg.sig = sig;
1837 	arg.sent = false;
1838 	arg.found = false;
1839 	arg.ret = 0;
1840 	if (all) {
1841 		/*
1842 		 * broadcast
1843 		 */
1844 		prison_proc_iterate(td->td_ucred->cr_prison,
1845 		    kill_processes_prison_cb, &arg);
1846 	} else {
1847 again:
1848 		sx_slock(&proctree_lock);
1849 		if (pgid == 0) {
1850 			/*
1851 			 * zero pgid means send to my process group.
1852 			 */
1853 			pgrp = td->td_proc->p_pgrp;
1854 			PGRP_LOCK(pgrp);
1855 		} else {
1856 			pgrp = pgfind(pgid);
1857 			if (pgrp == NULL) {
1858 				sx_sunlock(&proctree_lock);
1859 				return (ESRCH);
1860 			}
1861 		}
1862 		sx_sunlock(&proctree_lock);
1863 		if (!sx_try_xlock(&pgrp->pg_killsx)) {
1864 			PGRP_UNLOCK(pgrp);
1865 			sx_xlock(&pgrp->pg_killsx);
1866 			sx_xunlock(&pgrp->pg_killsx);
1867 			goto again;
1868 		}
1869 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1870 			killpg1_sendsig(p, false, &arg);
1871 		}
1872 		PGRP_UNLOCK(pgrp);
1873 		sx_xunlock(&pgrp->pg_killsx);
1874 	}
1875 	MPASS(arg.ret != 0 || arg.found || !arg.sent);
1876 	if (arg.ret == 0 && !arg.sent)
1877 		arg.ret = arg.found ? EPERM : ESRCH;
1878 	return (arg.ret);
1879 }
1880 
1881 #ifndef _SYS_SYSPROTO_H_
1882 struct kill_args {
1883 	int	pid;
1884 	int	signum;
1885 };
1886 #endif
1887 /* ARGSUSED */
1888 int
1889 sys_kill(struct thread *td, struct kill_args *uap)
1890 {
1891 
1892 	return (kern_kill(td, uap->pid, uap->signum));
1893 }
1894 
1895 int
1896 kern_kill(struct thread *td, pid_t pid, int signum)
1897 {
1898 	ksiginfo_t ksi;
1899 	struct proc *p;
1900 	int error;
1901 
1902 	/*
1903 	 * A process in capability mode can send signals only to himself.
1904 	 * The main rationale behind this is that abort(3) is implemented as
1905 	 * kill(getpid(), SIGABRT).
1906 	 */
1907 	if (IN_CAPABILITY_MODE(td) && pid != td->td_proc->p_pid)
1908 		return (ECAPMODE);
1909 
1910 	AUDIT_ARG_SIGNUM(signum);
1911 	AUDIT_ARG_PID(pid);
1912 	if ((u_int)signum > _SIG_MAXSIG)
1913 		return (EINVAL);
1914 
1915 	ksiginfo_init(&ksi);
1916 	ksi.ksi_signo = signum;
1917 	ksi.ksi_code = SI_USER;
1918 	ksi.ksi_pid = td->td_proc->p_pid;
1919 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1920 
1921 	if (pid > 0) {
1922 		/* kill single process */
1923 		if ((p = pfind_any(pid)) == NULL)
1924 			return (ESRCH);
1925 		AUDIT_ARG_PROCESS(p);
1926 		error = p_cansignal(td, p, signum);
1927 		if (error == 0 && signum)
1928 			pksignal(p, signum, &ksi);
1929 		PROC_UNLOCK(p);
1930 		return (error);
1931 	}
1932 	switch (pid) {
1933 	case -1:		/* broadcast signal */
1934 		return (killpg1(td, signum, 0, 1, &ksi));
1935 	case 0:			/* signal own process group */
1936 		return (killpg1(td, signum, 0, 0, &ksi));
1937 	default:		/* negative explicit process group */
1938 		return (killpg1(td, signum, -pid, 0, &ksi));
1939 	}
1940 	/* NOTREACHED */
1941 }
1942 
1943 int
1944 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1945 {
1946 	struct proc *p;
1947 	int error;
1948 
1949 	AUDIT_ARG_SIGNUM(uap->signum);
1950 	AUDIT_ARG_FD(uap->fd);
1951 	if ((u_int)uap->signum > _SIG_MAXSIG)
1952 		return (EINVAL);
1953 
1954 	error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1955 	if (error)
1956 		return (error);
1957 	AUDIT_ARG_PROCESS(p);
1958 	error = p_cansignal(td, p, uap->signum);
1959 	if (error == 0 && uap->signum)
1960 		kern_psignal(p, uap->signum);
1961 	PROC_UNLOCK(p);
1962 	return (error);
1963 }
1964 
1965 #if defined(COMPAT_43)
1966 #ifndef _SYS_SYSPROTO_H_
1967 struct okillpg_args {
1968 	int	pgid;
1969 	int	signum;
1970 };
1971 #endif
1972 /* ARGSUSED */
1973 int
1974 okillpg(struct thread *td, struct okillpg_args *uap)
1975 {
1976 	ksiginfo_t ksi;
1977 
1978 	AUDIT_ARG_SIGNUM(uap->signum);
1979 	AUDIT_ARG_PID(uap->pgid);
1980 	if ((u_int)uap->signum > _SIG_MAXSIG)
1981 		return (EINVAL);
1982 
1983 	ksiginfo_init(&ksi);
1984 	ksi.ksi_signo = uap->signum;
1985 	ksi.ksi_code = SI_USER;
1986 	ksi.ksi_pid = td->td_proc->p_pid;
1987 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1988 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1989 }
1990 #endif /* COMPAT_43 */
1991 
1992 #ifndef _SYS_SYSPROTO_H_
1993 struct sigqueue_args {
1994 	pid_t pid;
1995 	int signum;
1996 	/* union sigval */ void *value;
1997 };
1998 #endif
1999 int
2000 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
2001 {
2002 	union sigval sv;
2003 
2004 	sv.sival_ptr = uap->value;
2005 
2006 	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
2007 }
2008 
2009 int
2010 kern_sigqueue(struct thread *td, pid_t pid, int signum, union sigval *value)
2011 {
2012 	ksiginfo_t ksi;
2013 	struct proc *p;
2014 	int error;
2015 
2016 	if ((u_int)signum > _SIG_MAXSIG)
2017 		return (EINVAL);
2018 
2019 	/*
2020 	 * Specification says sigqueue can only send signal to
2021 	 * single process.
2022 	 */
2023 	if (pid <= 0)
2024 		return (EINVAL);
2025 
2026 	if ((p = pfind_any(pid)) == NULL)
2027 		return (ESRCH);
2028 	error = p_cansignal(td, p, signum);
2029 	if (error == 0 && signum != 0) {
2030 		ksiginfo_init(&ksi);
2031 		ksi.ksi_flags = KSI_SIGQ;
2032 		ksi.ksi_signo = signum;
2033 		ksi.ksi_code = SI_QUEUE;
2034 		ksi.ksi_pid = td->td_proc->p_pid;
2035 		ksi.ksi_uid = td->td_ucred->cr_ruid;
2036 		ksi.ksi_value = *value;
2037 		error = pksignal(p, ksi.ksi_signo, &ksi);
2038 	}
2039 	PROC_UNLOCK(p);
2040 	return (error);
2041 }
2042 
2043 /*
2044  * Send a signal to a process group.  If checktty is 1,
2045  * limit to members which have a controlling terminal.
2046  */
2047 void
2048 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
2049 {
2050 	struct proc *p;
2051 
2052 	if (pgrp) {
2053 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
2054 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
2055 			PROC_LOCK(p);
2056 			if (p->p_state == PRS_NORMAL &&
2057 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
2058 				pksignal(p, sig, ksi);
2059 			PROC_UNLOCK(p);
2060 		}
2061 	}
2062 }
2063 
2064 /*
2065  * Recalculate the signal mask and reset the signal disposition after
2066  * usermode frame for delivery is formed.  Should be called after
2067  * mach-specific routine, because sysent->sv_sendsig() needs correct
2068  * ps_siginfo and signal mask.
2069  */
2070 static void
2071 postsig_done(int sig, struct thread *td, struct sigacts *ps)
2072 {
2073 	sigset_t mask;
2074 
2075 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2076 	td->td_ru.ru_nsignals++;
2077 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
2078 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
2079 		SIGADDSET(mask, sig);
2080 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
2081 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
2082 	if (SIGISMEMBER(ps->ps_sigreset, sig))
2083 		sigdflt(ps, sig);
2084 }
2085 
2086 /*
2087  * Send a signal caused by a trap to the current thread.  If it will be
2088  * caught immediately, deliver it with correct code.  Otherwise, post it
2089  * normally.
2090  */
2091 void
2092 trapsignal(struct thread *td, ksiginfo_t *ksi)
2093 {
2094 	struct sigacts *ps;
2095 	struct proc *p;
2096 	sigset_t sigmask;
2097 	int sig;
2098 
2099 	p = td->td_proc;
2100 	sig = ksi->ksi_signo;
2101 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
2102 
2103 	sigfastblock_fetch(td);
2104 	PROC_LOCK(p);
2105 	ps = p->p_sigacts;
2106 	mtx_lock(&ps->ps_mtx);
2107 	sigmask = td->td_sigmask;
2108 	if (td->td_sigblock_val != 0)
2109 		SIGSETOR(sigmask, fastblock_mask);
2110 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2111 	    !SIGISMEMBER(sigmask, sig)) {
2112 #ifdef KTRACE
2113 		if (KTRPOINT(curthread, KTR_PSIG))
2114 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2115 			    &td->td_sigmask, ksi->ksi_code);
2116 #endif
2117 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2118 		    ksi, &td->td_sigmask);
2119 		postsig_done(sig, td, ps);
2120 		mtx_unlock(&ps->ps_mtx);
2121 	} else {
2122 		/*
2123 		 * Avoid a possible infinite loop if the thread
2124 		 * masking the signal or process is ignoring the
2125 		 * signal.
2126 		 */
2127 		if (kern_forcesigexit && (SIGISMEMBER(sigmask, sig) ||
2128 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2129 			SIGDELSET(td->td_sigmask, sig);
2130 			SIGDELSET(ps->ps_sigcatch, sig);
2131 			SIGDELSET(ps->ps_sigignore, sig);
2132 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2133 			td->td_pflags &= ~TDP_SIGFASTBLOCK;
2134 			td->td_sigblock_val = 0;
2135 		}
2136 		mtx_unlock(&ps->ps_mtx);
2137 		p->p_sig = sig;		/* XXX to verify code */
2138 		tdsendsignal(p, td, sig, ksi);
2139 	}
2140 	PROC_UNLOCK(p);
2141 }
2142 
2143 static struct thread *
2144 sigtd(struct proc *p, int sig, bool fast_sigblock)
2145 {
2146 	struct thread *td, *signal_td;
2147 
2148 	PROC_LOCK_ASSERT(p, MA_OWNED);
2149 	MPASS(!fast_sigblock || p == curproc);
2150 
2151 	/*
2152 	 * Check if current thread can handle the signal without
2153 	 * switching context to another thread.
2154 	 */
2155 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig) &&
2156 	    (!fast_sigblock || curthread->td_sigblock_val == 0))
2157 		return (curthread);
2158 
2159 	/* Find a non-stopped thread that does not mask the signal. */
2160 	signal_td = NULL;
2161 	FOREACH_THREAD_IN_PROC(p, td) {
2162 		if (!SIGISMEMBER(td->td_sigmask, sig) && (!fast_sigblock ||
2163 		    td != curthread || td->td_sigblock_val == 0) &&
2164 		    (td->td_flags & TDF_BOUNDARY) == 0) {
2165 			signal_td = td;
2166 			break;
2167 		}
2168 	}
2169 	/* Select random (first) thread if no better match was found. */
2170 	if (signal_td == NULL)
2171 		signal_td = FIRST_THREAD_IN_PROC(p);
2172 	return (signal_td);
2173 }
2174 
2175 /*
2176  * Send the signal to the process.  If the signal has an action, the action
2177  * is usually performed by the target process rather than the caller; we add
2178  * the signal to the set of pending signals for the process.
2179  *
2180  * Exceptions:
2181  *   o When a stop signal is sent to a sleeping process that takes the
2182  *     default action, the process is stopped without awakening it.
2183  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2184  *     regardless of the signal action (eg, blocked or ignored).
2185  *
2186  * Other ignored signals are discarded immediately.
2187  *
2188  * NB: This function may be entered from the debugger via the "kill" DDB
2189  * command.  There is little that can be done to mitigate the possibly messy
2190  * side effects of this unwise possibility.
2191  */
2192 void
2193 kern_psignal(struct proc *p, int sig)
2194 {
2195 	ksiginfo_t ksi;
2196 
2197 	ksiginfo_init(&ksi);
2198 	ksi.ksi_signo = sig;
2199 	ksi.ksi_code = SI_KERNEL;
2200 	(void) tdsendsignal(p, NULL, sig, &ksi);
2201 }
2202 
2203 int
2204 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2205 {
2206 
2207 	return (tdsendsignal(p, NULL, sig, ksi));
2208 }
2209 
2210 /* Utility function for finding a thread to send signal event to. */
2211 int
2212 sigev_findtd(struct proc *p, struct sigevent *sigev, struct thread **ttd)
2213 {
2214 	struct thread *td;
2215 
2216 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2217 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2218 		if (td == NULL)
2219 			return (ESRCH);
2220 		*ttd = td;
2221 	} else {
2222 		*ttd = NULL;
2223 		PROC_LOCK(p);
2224 	}
2225 	return (0);
2226 }
2227 
2228 void
2229 tdsignal(struct thread *td, int sig)
2230 {
2231 	ksiginfo_t ksi;
2232 
2233 	ksiginfo_init(&ksi);
2234 	ksi.ksi_signo = sig;
2235 	ksi.ksi_code = SI_KERNEL;
2236 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2237 }
2238 
2239 void
2240 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2241 {
2242 
2243 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2244 }
2245 
2246 static int
2247 sig_sleepq_abort(struct thread *td, int intrval)
2248 {
2249 	THREAD_LOCK_ASSERT(td, MA_OWNED);
2250 
2251 	if (intrval == 0 && (td->td_flags & TDF_SIGWAIT) == 0) {
2252 		thread_unlock(td);
2253 		return (0);
2254 	}
2255 	return (sleepq_abort(td, intrval));
2256 }
2257 
2258 int
2259 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2260 {
2261 	sig_t action;
2262 	sigqueue_t *sigqueue;
2263 	int prop;
2264 	struct sigacts *ps;
2265 	int intrval;
2266 	int ret = 0;
2267 	int wakeup_swapper;
2268 
2269 	MPASS(td == NULL || p == td->td_proc);
2270 	PROC_LOCK_ASSERT(p, MA_OWNED);
2271 
2272 	if (!_SIG_VALID(sig))
2273 		panic("%s(): invalid signal %d", __func__, sig);
2274 
2275 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2276 
2277 	/*
2278 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2279 	 */
2280 	if (p->p_state == PRS_ZOMBIE) {
2281 		if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2282 			ksiginfo_tryfree(ksi);
2283 		return (ret);
2284 	}
2285 
2286 	ps = p->p_sigacts;
2287 	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2288 	prop = sigprop(sig);
2289 
2290 	if (td == NULL) {
2291 		td = sigtd(p, sig, false);
2292 		sigqueue = &p->p_sigqueue;
2293 	} else
2294 		sigqueue = &td->td_sigqueue;
2295 
2296 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2297 
2298 	/*
2299 	 * If the signal is being ignored, then we forget about it
2300 	 * immediately, except when the target process executes
2301 	 * sigwait().  (Note: we don't set SIGCONT in ps_sigignore,
2302 	 * and if it is set to SIG_IGN, action will be SIG_DFL here.)
2303 	 */
2304 	mtx_lock(&ps->ps_mtx);
2305 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2306 		if (kern_sig_discard_ign &&
2307 		    (p->p_sysent->sv_flags & SV_SIG_DISCIGN) == 0) {
2308 			SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2309 
2310 			mtx_unlock(&ps->ps_mtx);
2311 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2312 				ksiginfo_tryfree(ksi);
2313 			return (ret);
2314 		} else {
2315 			action = SIG_CATCH;
2316 			intrval = 0;
2317 		}
2318 	} else {
2319 		if (SIGISMEMBER(td->td_sigmask, sig))
2320 			action = SIG_HOLD;
2321 		else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2322 			action = SIG_CATCH;
2323 		else
2324 			action = SIG_DFL;
2325 		if (SIGISMEMBER(ps->ps_sigintr, sig))
2326 			intrval = EINTR;
2327 		else
2328 			intrval = ERESTART;
2329 	}
2330 	mtx_unlock(&ps->ps_mtx);
2331 
2332 	if (prop & SIGPROP_CONT)
2333 		sigqueue_delete_stopmask_proc(p);
2334 	else if (prop & SIGPROP_STOP) {
2335 		/*
2336 		 * If sending a tty stop signal to a member of an orphaned
2337 		 * process group, discard the signal here if the action
2338 		 * is default; don't stop the process below if sleeping,
2339 		 * and don't clear any pending SIGCONT.
2340 		 */
2341 		if ((prop & SIGPROP_TTYSTOP) != 0 &&
2342 		    (p->p_pgrp->pg_flags & PGRP_ORPHANED) != 0 &&
2343 		    action == SIG_DFL) {
2344 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2345 				ksiginfo_tryfree(ksi);
2346 			return (ret);
2347 		}
2348 		sigqueue_delete_proc(p, SIGCONT);
2349 		if (p->p_flag & P_CONTINUED) {
2350 			p->p_flag &= ~P_CONTINUED;
2351 			PROC_LOCK(p->p_pptr);
2352 			sigqueue_take(p->p_ksi);
2353 			PROC_UNLOCK(p->p_pptr);
2354 		}
2355 	}
2356 
2357 	ret = sigqueue_add(sigqueue, sig, ksi);
2358 	if (ret != 0)
2359 		return (ret);
2360 	signotify(td);
2361 	/*
2362 	 * Defer further processing for signals which are held,
2363 	 * except that stopped processes must be continued by SIGCONT.
2364 	 */
2365 	if (action == SIG_HOLD &&
2366 	    !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2367 		return (ret);
2368 
2369 	wakeup_swapper = 0;
2370 
2371 	/*
2372 	 * Some signals have a process-wide effect and a per-thread
2373 	 * component.  Most processing occurs when the process next
2374 	 * tries to cross the user boundary, however there are some
2375 	 * times when processing needs to be done immediately, such as
2376 	 * waking up threads so that they can cross the user boundary.
2377 	 * We try to do the per-process part here.
2378 	 */
2379 	if (P_SHOULDSTOP(p)) {
2380 		KASSERT(!(p->p_flag & P_WEXIT),
2381 		    ("signal to stopped but exiting process"));
2382 		if (sig == SIGKILL) {
2383 			/*
2384 			 * If traced process is already stopped,
2385 			 * then no further action is necessary.
2386 			 */
2387 			if (p->p_flag & P_TRACED)
2388 				goto out;
2389 			/*
2390 			 * SIGKILL sets process running.
2391 			 * It will die elsewhere.
2392 			 * All threads must be restarted.
2393 			 */
2394 			p->p_flag &= ~P_STOPPED_SIG;
2395 			goto runfast;
2396 		}
2397 
2398 		if (prop & SIGPROP_CONT) {
2399 			/*
2400 			 * If traced process is already stopped,
2401 			 * then no further action is necessary.
2402 			 */
2403 			if (p->p_flag & P_TRACED)
2404 				goto out;
2405 			/*
2406 			 * If SIGCONT is default (or ignored), we continue the
2407 			 * process but don't leave the signal in sigqueue as
2408 			 * it has no further action.  If SIGCONT is held, we
2409 			 * continue the process and leave the signal in
2410 			 * sigqueue.  If the process catches SIGCONT, let it
2411 			 * handle the signal itself.  If it isn't waiting on
2412 			 * an event, it goes back to run state.
2413 			 * Otherwise, process goes back to sleep state.
2414 			 */
2415 			p->p_flag &= ~P_STOPPED_SIG;
2416 			PROC_SLOCK(p);
2417 			if (p->p_numthreads == p->p_suspcount) {
2418 				PROC_SUNLOCK(p);
2419 				p->p_flag |= P_CONTINUED;
2420 				p->p_xsig = SIGCONT;
2421 				PROC_LOCK(p->p_pptr);
2422 				childproc_continued(p);
2423 				PROC_UNLOCK(p->p_pptr);
2424 				PROC_SLOCK(p);
2425 			}
2426 			if (action == SIG_DFL) {
2427 				thread_unsuspend(p);
2428 				PROC_SUNLOCK(p);
2429 				sigqueue_delete(sigqueue, sig);
2430 				goto out_cont;
2431 			}
2432 			if (action == SIG_CATCH) {
2433 				/*
2434 				 * The process wants to catch it so it needs
2435 				 * to run at least one thread, but which one?
2436 				 */
2437 				PROC_SUNLOCK(p);
2438 				goto runfast;
2439 			}
2440 			/*
2441 			 * The signal is not ignored or caught.
2442 			 */
2443 			thread_unsuspend(p);
2444 			PROC_SUNLOCK(p);
2445 			goto out_cont;
2446 		}
2447 
2448 		if (prop & SIGPROP_STOP) {
2449 			/*
2450 			 * If traced process is already stopped,
2451 			 * then no further action is necessary.
2452 			 */
2453 			if (p->p_flag & P_TRACED)
2454 				goto out;
2455 			/*
2456 			 * Already stopped, don't need to stop again
2457 			 * (If we did the shell could get confused).
2458 			 * Just make sure the signal STOP bit set.
2459 			 */
2460 			p->p_flag |= P_STOPPED_SIG;
2461 			sigqueue_delete(sigqueue, sig);
2462 			goto out;
2463 		}
2464 
2465 		/*
2466 		 * All other kinds of signals:
2467 		 * If a thread is sleeping interruptibly, simulate a
2468 		 * wakeup so that when it is continued it will be made
2469 		 * runnable and can look at the signal.  However, don't make
2470 		 * the PROCESS runnable, leave it stopped.
2471 		 * It may run a bit until it hits a thread_suspend_check().
2472 		 */
2473 		PROC_SLOCK(p);
2474 		thread_lock(td);
2475 		if (TD_CAN_ABORT(td))
2476 			wakeup_swapper = sig_sleepq_abort(td, intrval);
2477 		else
2478 			thread_unlock(td);
2479 		PROC_SUNLOCK(p);
2480 		goto out;
2481 		/*
2482 		 * Mutexes are short lived. Threads waiting on them will
2483 		 * hit thread_suspend_check() soon.
2484 		 */
2485 	} else if (p->p_state == PRS_NORMAL) {
2486 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2487 			tdsigwakeup(td, sig, action, intrval);
2488 			goto out;
2489 		}
2490 
2491 		MPASS(action == SIG_DFL);
2492 
2493 		if (prop & SIGPROP_STOP) {
2494 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2495 				goto out;
2496 			p->p_flag |= P_STOPPED_SIG;
2497 			p->p_xsig = sig;
2498 			PROC_SLOCK(p);
2499 			wakeup_swapper = sig_suspend_threads(td, p);
2500 			if (p->p_numthreads == p->p_suspcount) {
2501 				/*
2502 				 * only thread sending signal to another
2503 				 * process can reach here, if thread is sending
2504 				 * signal to its process, because thread does
2505 				 * not suspend itself here, p_numthreads
2506 				 * should never be equal to p_suspcount.
2507 				 */
2508 				thread_stopped(p);
2509 				PROC_SUNLOCK(p);
2510 				sigqueue_delete_proc(p, p->p_xsig);
2511 			} else
2512 				PROC_SUNLOCK(p);
2513 			goto out;
2514 		}
2515 	} else {
2516 		/* Not in "NORMAL" state. discard the signal. */
2517 		sigqueue_delete(sigqueue, sig);
2518 		goto out;
2519 	}
2520 
2521 	/*
2522 	 * The process is not stopped so we need to apply the signal to all the
2523 	 * running threads.
2524 	 */
2525 runfast:
2526 	tdsigwakeup(td, sig, action, intrval);
2527 	PROC_SLOCK(p);
2528 	thread_unsuspend(p);
2529 	PROC_SUNLOCK(p);
2530 out_cont:
2531 	itimer_proc_continue(p);
2532 	kqtimer_proc_continue(p);
2533 out:
2534 	/* If we jump here, proc slock should not be owned. */
2535 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2536 	if (wakeup_swapper)
2537 		kick_proc0();
2538 
2539 	return (ret);
2540 }
2541 
2542 /*
2543  * The force of a signal has been directed against a single
2544  * thread.  We need to see what we can do about knocking it
2545  * out of any sleep it may be in etc.
2546  */
2547 static void
2548 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2549 {
2550 	struct proc *p = td->td_proc;
2551 	int prop, wakeup_swapper;
2552 
2553 	PROC_LOCK_ASSERT(p, MA_OWNED);
2554 	prop = sigprop(sig);
2555 
2556 	PROC_SLOCK(p);
2557 	thread_lock(td);
2558 	/*
2559 	 * Bring the priority of a thread up if we want it to get
2560 	 * killed in this lifetime.  Be careful to avoid bumping the
2561 	 * priority of the idle thread, since we still allow to signal
2562 	 * kernel processes.
2563 	 */
2564 	if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2565 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2566 		sched_prio(td, PUSER);
2567 	if (TD_ON_SLEEPQ(td)) {
2568 		/*
2569 		 * If thread is sleeping uninterruptibly
2570 		 * we can't interrupt the sleep... the signal will
2571 		 * be noticed when the process returns through
2572 		 * trap() or syscall().
2573 		 */
2574 		if ((td->td_flags & TDF_SINTR) == 0)
2575 			goto out;
2576 		/*
2577 		 * If SIGCONT is default (or ignored) and process is
2578 		 * asleep, we are finished; the process should not
2579 		 * be awakened.
2580 		 */
2581 		if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2582 			thread_unlock(td);
2583 			PROC_SUNLOCK(p);
2584 			sigqueue_delete(&p->p_sigqueue, sig);
2585 			/*
2586 			 * It may be on either list in this state.
2587 			 * Remove from both for now.
2588 			 */
2589 			sigqueue_delete(&td->td_sigqueue, sig);
2590 			return;
2591 		}
2592 
2593 		/*
2594 		 * Don't awaken a sleeping thread for SIGSTOP if the
2595 		 * STOP signal is deferred.
2596 		 */
2597 		if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2598 		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2599 			goto out;
2600 
2601 		/*
2602 		 * Give low priority threads a better chance to run.
2603 		 */
2604 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2605 			sched_prio(td, PUSER);
2606 
2607 		wakeup_swapper = sig_sleepq_abort(td, intrval);
2608 		PROC_SUNLOCK(p);
2609 		if (wakeup_swapper)
2610 			kick_proc0();
2611 		return;
2612 	}
2613 
2614 	/*
2615 	 * Other states do nothing with the signal immediately,
2616 	 * other than kicking ourselves if we are running.
2617 	 * It will either never be noticed, or noticed very soon.
2618 	 */
2619 #ifdef SMP
2620 	if (TD_IS_RUNNING(td) && td != curthread)
2621 		forward_signal(td);
2622 #endif
2623 
2624 out:
2625 	PROC_SUNLOCK(p);
2626 	thread_unlock(td);
2627 }
2628 
2629 static void
2630 ptrace_coredumpreq(struct thread *td, struct proc *p,
2631     struct thr_coredump_req *tcq)
2632 {
2633 	void *rl_cookie;
2634 
2635 	if (p->p_sysent->sv_coredump == NULL) {
2636 		tcq->tc_error = ENOSYS;
2637 		return;
2638 	}
2639 
2640 	rl_cookie = vn_rangelock_wlock(tcq->tc_vp, 0, OFF_MAX);
2641 	tcq->tc_error = p->p_sysent->sv_coredump(td, tcq->tc_vp,
2642 	    tcq->tc_limit, tcq->tc_flags);
2643 	vn_rangelock_unlock(tcq->tc_vp, rl_cookie);
2644 }
2645 
2646 static void
2647 ptrace_syscallreq(struct thread *td, struct proc *p,
2648     struct thr_syscall_req *tsr)
2649 {
2650 	struct sysentvec *sv;
2651 	struct sysent *se;
2652 	register_t rv_saved[2];
2653 	int error, nerror;
2654 	int sc;
2655 	bool audited, sy_thr_static;
2656 
2657 	sv = p->p_sysent;
2658 	if (sv->sv_table == NULL || sv->sv_size < tsr->ts_sa.code) {
2659 		tsr->ts_ret.sr_error = ENOSYS;
2660 		return;
2661 	}
2662 
2663 	sc = tsr->ts_sa.code;
2664 	if (sc == SYS_syscall || sc == SYS___syscall) {
2665 		sc = tsr->ts_sa.args[0];
2666 		memmove(&tsr->ts_sa.args[0], &tsr->ts_sa.args[1],
2667 		    sizeof(register_t) * (tsr->ts_nargs - 1));
2668 	}
2669 
2670 	tsr->ts_sa.callp = se = &sv->sv_table[sc];
2671 
2672 	VM_CNT_INC(v_syscall);
2673 	td->td_pticks = 0;
2674 	if (__predict_false(td->td_cowgen != atomic_load_int(
2675 	    &td->td_proc->p_cowgen)))
2676 		thread_cow_update(td);
2677 
2678 #ifdef CAPABILITY_MODE
2679 	if (IN_CAPABILITY_MODE(td) && (se->sy_flags & SYF_CAPENABLED) == 0) {
2680 		tsr->ts_ret.sr_error = ECAPMODE;
2681 		return;
2682 	}
2683 #endif
2684 
2685 	sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2686 	audited = AUDIT_SYSCALL_ENTER(sc, td) != 0;
2687 
2688 	if (!sy_thr_static) {
2689 		error = syscall_thread_enter(td, se);
2690 		if (error != 0) {
2691 			tsr->ts_ret.sr_error = error;
2692 			return;
2693 		}
2694 	}
2695 
2696 	rv_saved[0] = td->td_retval[0];
2697 	rv_saved[1] = td->td_retval[1];
2698 	nerror = td->td_errno;
2699 	td->td_retval[0] = 0;
2700 	td->td_retval[1] = 0;
2701 
2702 #ifdef KDTRACE_HOOKS
2703 	if (se->sy_entry != 0)
2704 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_ENTRY, 0);
2705 #endif
2706 	tsr->ts_ret.sr_error = se->sy_call(td, tsr->ts_sa.args);
2707 #ifdef KDTRACE_HOOKS
2708 	if (se->sy_return != 0)
2709 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_RETURN,
2710 		    tsr->ts_ret.sr_error != 0 ? -1 : td->td_retval[0]);
2711 #endif
2712 
2713 	tsr->ts_ret.sr_retval[0] = td->td_retval[0];
2714 	tsr->ts_ret.sr_retval[1] = td->td_retval[1];
2715 	td->td_retval[0] = rv_saved[0];
2716 	td->td_retval[1] = rv_saved[1];
2717 	td->td_errno = nerror;
2718 
2719 	if (audited)
2720 		AUDIT_SYSCALL_EXIT(error, td);
2721 	if (!sy_thr_static)
2722 		syscall_thread_exit(td, se);
2723 }
2724 
2725 static void
2726 ptrace_remotereq(struct thread *td, int flag)
2727 {
2728 	struct proc *p;
2729 
2730 	MPASS(td == curthread);
2731 	p = td->td_proc;
2732 	PROC_LOCK_ASSERT(p, MA_OWNED);
2733 	if ((td->td_dbgflags & flag) == 0)
2734 		return;
2735 	KASSERT((p->p_flag & P_STOPPED_TRACE) != 0, ("not stopped"));
2736 	KASSERT(td->td_remotereq != NULL, ("td_remotereq is NULL"));
2737 
2738 	PROC_UNLOCK(p);
2739 	switch (flag) {
2740 	case TDB_COREDUMPREQ:
2741 		ptrace_coredumpreq(td, p, td->td_remotereq);
2742 		break;
2743 	case TDB_SCREMOTEREQ:
2744 		ptrace_syscallreq(td, p, td->td_remotereq);
2745 		break;
2746 	default:
2747 		__unreachable();
2748 	}
2749 	PROC_LOCK(p);
2750 
2751 	MPASS((td->td_dbgflags & flag) != 0);
2752 	td->td_dbgflags &= ~flag;
2753 	td->td_remotereq = NULL;
2754 	wakeup(p);
2755 }
2756 
2757 static int
2758 sig_suspend_threads(struct thread *td, struct proc *p)
2759 {
2760 	struct thread *td2;
2761 	int wakeup_swapper;
2762 
2763 	PROC_LOCK_ASSERT(p, MA_OWNED);
2764 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2765 
2766 	wakeup_swapper = 0;
2767 	FOREACH_THREAD_IN_PROC(p, td2) {
2768 		thread_lock(td2);
2769 		ast_sched_locked(td2, TDA_SUSPEND);
2770 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2771 		    (td2->td_flags & TDF_SINTR)) {
2772 			if (td2->td_flags & TDF_SBDRY) {
2773 				/*
2774 				 * Once a thread is asleep with
2775 				 * TDF_SBDRY and without TDF_SERESTART
2776 				 * or TDF_SEINTR set, it should never
2777 				 * become suspended due to this check.
2778 				 */
2779 				KASSERT(!TD_IS_SUSPENDED(td2),
2780 				    ("thread with deferred stops suspended"));
2781 				if (TD_SBDRY_INTR(td2)) {
2782 					wakeup_swapper |= sleepq_abort(td2,
2783 					    TD_SBDRY_ERRNO(td2));
2784 					continue;
2785 				}
2786 			} else if (!TD_IS_SUSPENDED(td2))
2787 				thread_suspend_one(td2);
2788 		} else if (!TD_IS_SUSPENDED(td2)) {
2789 #ifdef SMP
2790 			if (TD_IS_RUNNING(td2) && td2 != td)
2791 				forward_signal(td2);
2792 #endif
2793 		}
2794 		thread_unlock(td2);
2795 	}
2796 	return (wakeup_swapper);
2797 }
2798 
2799 /*
2800  * Stop the process for an event deemed interesting to the debugger. If si is
2801  * non-NULL, this is a signal exchange; the new signal requested by the
2802  * debugger will be returned for handling. If si is NULL, this is some other
2803  * type of interesting event. The debugger may request a signal be delivered in
2804  * that case as well, however it will be deferred until it can be handled.
2805  */
2806 int
2807 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2808 {
2809 	struct proc *p = td->td_proc;
2810 	struct thread *td2;
2811 	ksiginfo_t ksi;
2812 
2813 	PROC_LOCK_ASSERT(p, MA_OWNED);
2814 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2815 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2816 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2817 
2818 	td->td_xsig = sig;
2819 
2820 	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2821 		td->td_dbgflags |= TDB_XSIG;
2822 		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2823 		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2824 		PROC_SLOCK(p);
2825 		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2826 			if (P_KILLED(p)) {
2827 				/*
2828 				 * Ensure that, if we've been PT_KILLed, the
2829 				 * exit status reflects that. Another thread
2830 				 * may also be in ptracestop(), having just
2831 				 * received the SIGKILL, but this thread was
2832 				 * unsuspended first.
2833 				 */
2834 				td->td_dbgflags &= ~TDB_XSIG;
2835 				td->td_xsig = SIGKILL;
2836 				p->p_ptevents = 0;
2837 				break;
2838 			}
2839 			if (p->p_flag & P_SINGLE_EXIT &&
2840 			    !(td->td_dbgflags & TDB_EXIT)) {
2841 				/*
2842 				 * Ignore ptrace stops except for thread exit
2843 				 * events when the process exits.
2844 				 */
2845 				td->td_dbgflags &= ~TDB_XSIG;
2846 				PROC_SUNLOCK(p);
2847 				return (0);
2848 			}
2849 
2850 			/*
2851 			 * Make wait(2) work.  Ensure that right after the
2852 			 * attach, the thread which was decided to become the
2853 			 * leader of attach gets reported to the waiter.
2854 			 * Otherwise, just avoid overwriting another thread's
2855 			 * assignment to p_xthread.  If another thread has
2856 			 * already set p_xthread, the current thread will get
2857 			 * a chance to report itself upon the next iteration.
2858 			 */
2859 			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2860 			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2861 			    p->p_xthread == NULL)) {
2862 				p->p_xsig = sig;
2863 				p->p_xthread = td;
2864 
2865 				/*
2866 				 * If we are on sleepqueue already,
2867 				 * let sleepqueue code decide if it
2868 				 * needs to go sleep after attach.
2869 				 */
2870 				if (td->td_wchan == NULL)
2871 					td->td_dbgflags &= ~TDB_FSTP;
2872 
2873 				p->p_flag2 &= ~P2_PTRACE_FSTP;
2874 				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2875 				sig_suspend_threads(td, p);
2876 			}
2877 			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2878 				td->td_dbgflags &= ~TDB_STOPATFORK;
2879 			}
2880 stopme:
2881 			td->td_dbgflags |= TDB_SSWITCH;
2882 			thread_suspend_switch(td, p);
2883 			td->td_dbgflags &= ~TDB_SSWITCH;
2884 			if ((td->td_dbgflags & (TDB_COREDUMPREQ |
2885 			    TDB_SCREMOTEREQ)) != 0) {
2886 				MPASS((td->td_dbgflags & (TDB_COREDUMPREQ |
2887 				    TDB_SCREMOTEREQ)) !=
2888 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2889 				PROC_SUNLOCK(p);
2890 				ptrace_remotereq(td, td->td_dbgflags &
2891 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2892 				PROC_SLOCK(p);
2893 				goto stopme;
2894 			}
2895 			if (p->p_xthread == td)
2896 				p->p_xthread = NULL;
2897 			if (!(p->p_flag & P_TRACED))
2898 				break;
2899 			if (td->td_dbgflags & TDB_SUSPEND) {
2900 				if (p->p_flag & P_SINGLE_EXIT)
2901 					break;
2902 				goto stopme;
2903 			}
2904 		}
2905 		PROC_SUNLOCK(p);
2906 	}
2907 
2908 	if (si != NULL && sig == td->td_xsig) {
2909 		/* Parent wants us to take the original signal unchanged. */
2910 		si->ksi_flags |= KSI_HEAD;
2911 		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2912 			si->ksi_signo = 0;
2913 	} else if (td->td_xsig != 0) {
2914 		/*
2915 		 * If parent wants us to take a new signal, then it will leave
2916 		 * it in td->td_xsig; otherwise we just look for signals again.
2917 		 */
2918 		ksiginfo_init(&ksi);
2919 		ksi.ksi_signo = td->td_xsig;
2920 		ksi.ksi_flags |= KSI_PTRACE;
2921 		td2 = sigtd(p, td->td_xsig, false);
2922 		tdsendsignal(p, td2, td->td_xsig, &ksi);
2923 		if (td != td2)
2924 			return (0);
2925 	}
2926 
2927 	return (td->td_xsig);
2928 }
2929 
2930 static void
2931 reschedule_signals(struct proc *p, sigset_t block, int flags)
2932 {
2933 	struct sigacts *ps;
2934 	struct thread *td;
2935 	int sig;
2936 	bool fastblk, pslocked;
2937 
2938 	PROC_LOCK_ASSERT(p, MA_OWNED);
2939 	ps = p->p_sigacts;
2940 	pslocked = (flags & SIGPROCMASK_PS_LOCKED) != 0;
2941 	mtx_assert(&ps->ps_mtx, pslocked ? MA_OWNED : MA_NOTOWNED);
2942 	if (SIGISEMPTY(p->p_siglist))
2943 		return;
2944 	SIGSETAND(block, p->p_siglist);
2945 	fastblk = (flags & SIGPROCMASK_FASTBLK) != 0;
2946 	SIG_FOREACH(sig, &block) {
2947 		td = sigtd(p, sig, fastblk);
2948 
2949 		/*
2950 		 * If sigtd() selected us despite sigfastblock is
2951 		 * blocking, do not activate AST or wake us, to avoid
2952 		 * loop in AST handler.
2953 		 */
2954 		if (fastblk && td == curthread)
2955 			continue;
2956 
2957 		signotify(td);
2958 		if (!pslocked)
2959 			mtx_lock(&ps->ps_mtx);
2960 		if (p->p_flag & P_TRACED ||
2961 		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2962 		    !SIGISMEMBER(td->td_sigmask, sig))) {
2963 			tdsigwakeup(td, sig, SIG_CATCH,
2964 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2965 			    ERESTART));
2966 		}
2967 		if (!pslocked)
2968 			mtx_unlock(&ps->ps_mtx);
2969 	}
2970 }
2971 
2972 void
2973 tdsigcleanup(struct thread *td)
2974 {
2975 	struct proc *p;
2976 	sigset_t unblocked;
2977 
2978 	p = td->td_proc;
2979 	PROC_LOCK_ASSERT(p, MA_OWNED);
2980 
2981 	sigqueue_flush(&td->td_sigqueue);
2982 	if (p->p_numthreads == 1)
2983 		return;
2984 
2985 	/*
2986 	 * Since we cannot handle signals, notify signal post code
2987 	 * about this by filling the sigmask.
2988 	 *
2989 	 * Also, if needed, wake up thread(s) that do not block the
2990 	 * same signals as the exiting thread, since the thread might
2991 	 * have been selected for delivery and woken up.
2992 	 */
2993 	SIGFILLSET(unblocked);
2994 	SIGSETNAND(unblocked, td->td_sigmask);
2995 	SIGFILLSET(td->td_sigmask);
2996 	reschedule_signals(p, unblocked, 0);
2997 
2998 }
2999 
3000 static int
3001 sigdeferstop_curr_flags(int cflags)
3002 {
3003 
3004 	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
3005 	    (cflags & TDF_SBDRY) != 0);
3006 	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
3007 }
3008 
3009 /*
3010  * Defer the delivery of SIGSTOP for the current thread, according to
3011  * the requested mode.  Returns previous flags, which must be restored
3012  * by sigallowstop().
3013  *
3014  * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
3015  * cleared by the current thread, which allow the lock-less read-only
3016  * accesses below.
3017  */
3018 int
3019 sigdeferstop_impl(int mode)
3020 {
3021 	struct thread *td;
3022 	int cflags, nflags;
3023 
3024 	td = curthread;
3025 	cflags = sigdeferstop_curr_flags(td->td_flags);
3026 	switch (mode) {
3027 	case SIGDEFERSTOP_NOP:
3028 		nflags = cflags;
3029 		break;
3030 	case SIGDEFERSTOP_OFF:
3031 		nflags = 0;
3032 		break;
3033 	case SIGDEFERSTOP_SILENT:
3034 		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
3035 		break;
3036 	case SIGDEFERSTOP_EINTR:
3037 		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
3038 		break;
3039 	case SIGDEFERSTOP_ERESTART:
3040 		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
3041 		break;
3042 	default:
3043 		panic("sigdeferstop: invalid mode %x", mode);
3044 		break;
3045 	}
3046 	if (cflags == nflags)
3047 		return (SIGDEFERSTOP_VAL_NCHG);
3048 	thread_lock(td);
3049 	td->td_flags = (td->td_flags & ~cflags) | nflags;
3050 	thread_unlock(td);
3051 	return (cflags);
3052 }
3053 
3054 /*
3055  * Restores the STOP handling mode, typically permitting the delivery
3056  * of SIGSTOP for the current thread.  This does not immediately
3057  * suspend if a stop was posted.  Instead, the thread will suspend
3058  * either via ast() or a subsequent interruptible sleep.
3059  */
3060 void
3061 sigallowstop_impl(int prev)
3062 {
3063 	struct thread *td;
3064 	int cflags;
3065 
3066 	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
3067 	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
3068 	    ("sigallowstop: incorrect previous mode %x", prev));
3069 	td = curthread;
3070 	cflags = sigdeferstop_curr_flags(td->td_flags);
3071 	if (cflags != prev) {
3072 		thread_lock(td);
3073 		td->td_flags = (td->td_flags & ~cflags) | prev;
3074 		thread_unlock(td);
3075 	}
3076 }
3077 
3078 enum sigstatus {
3079 	SIGSTATUS_HANDLE,
3080 	SIGSTATUS_HANDLED,
3081 	SIGSTATUS_IGNORE,
3082 	SIGSTATUS_SBDRY_STOP,
3083 };
3084 
3085 /*
3086  * The thread has signal "sig" pending.  Figure out what to do with it:
3087  *
3088  * _HANDLE     -> the caller should handle the signal
3089  * _HANDLED    -> handled internally, reload pending signal set
3090  * _IGNORE     -> ignored, remove from the set of pending signals and try the
3091  *                next pending signal
3092  * _SBDRY_STOP -> the signal should stop the thread but this is not
3093  *                permitted in the current context
3094  */
3095 static enum sigstatus
3096 sigprocess(struct thread *td, int sig)
3097 {
3098 	struct proc *p;
3099 	struct sigacts *ps;
3100 	struct sigqueue *queue;
3101 	ksiginfo_t ksi;
3102 	int prop;
3103 
3104 	KASSERT(_SIG_VALID(sig), ("%s: invalid signal %d", __func__, sig));
3105 
3106 	p = td->td_proc;
3107 	ps = p->p_sigacts;
3108 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3109 	PROC_LOCK_ASSERT(p, MA_OWNED);
3110 
3111 	/*
3112 	 * We should allow pending but ignored signals below
3113 	 * if there is sigwait() active, or P_TRACED was
3114 	 * on when they were posted.
3115 	 */
3116 	if (SIGISMEMBER(ps->ps_sigignore, sig) &&
3117 	    (p->p_flag & P_TRACED) == 0 &&
3118 	    (td->td_flags & TDF_SIGWAIT) == 0) {
3119 		return (SIGSTATUS_IGNORE);
3120 	}
3121 
3122 	/*
3123 	 * If the process is going to single-thread mode to prepare
3124 	 * for exit, there is no sense in delivering any signal
3125 	 * to usermode.  Another important consequence is that
3126 	 * msleep(..., PCATCH, ...) now is only interruptible by a
3127 	 * suspend request.
3128 	 */
3129 	if ((p->p_flag2 & P2_WEXIT) != 0)
3130 		return (SIGSTATUS_IGNORE);
3131 
3132 	if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
3133 		/*
3134 		 * If traced, always stop.
3135 		 * Remove old signal from queue before the stop.
3136 		 * XXX shrug off debugger, it causes siginfo to
3137 		 * be thrown away.
3138 		 */
3139 		queue = &td->td_sigqueue;
3140 		ksiginfo_init(&ksi);
3141 		if (sigqueue_get(queue, sig, &ksi) == 0) {
3142 			queue = &p->p_sigqueue;
3143 			sigqueue_get(queue, sig, &ksi);
3144 		}
3145 		td->td_si = ksi.ksi_info;
3146 
3147 		mtx_unlock(&ps->ps_mtx);
3148 		sig = ptracestop(td, sig, &ksi);
3149 		mtx_lock(&ps->ps_mtx);
3150 
3151 		td->td_si.si_signo = 0;
3152 
3153 		/*
3154 		 * Keep looking if the debugger discarded or
3155 		 * replaced the signal.
3156 		 */
3157 		if (sig == 0)
3158 			return (SIGSTATUS_HANDLED);
3159 
3160 		/*
3161 		 * If the signal became masked, re-queue it.
3162 		 */
3163 		if (SIGISMEMBER(td->td_sigmask, sig)) {
3164 			ksi.ksi_flags |= KSI_HEAD;
3165 			sigqueue_add(&p->p_sigqueue, sig, &ksi);
3166 			return (SIGSTATUS_HANDLED);
3167 		}
3168 
3169 		/*
3170 		 * If the traced bit got turned off, requeue the signal and
3171 		 * reload the set of pending signals.  This ensures that p_sig*
3172 		 * and p_sigact are consistent.
3173 		 */
3174 		if ((p->p_flag & P_TRACED) == 0) {
3175 			if ((ksi.ksi_flags & KSI_PTRACE) == 0) {
3176 				ksi.ksi_flags |= KSI_HEAD;
3177 				sigqueue_add(queue, sig, &ksi);
3178 			}
3179 			return (SIGSTATUS_HANDLED);
3180 		}
3181 	}
3182 
3183 	/*
3184 	 * Decide whether the signal should be returned.
3185 	 * Return the signal's number, or fall through
3186 	 * to clear it from the pending mask.
3187 	 */
3188 	switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
3189 	case (intptr_t)SIG_DFL:
3190 		/*
3191 		 * Don't take default actions on system processes.
3192 		 */
3193 		if (p->p_pid <= 1) {
3194 #ifdef DIAGNOSTIC
3195 			/*
3196 			 * Are you sure you want to ignore SIGSEGV
3197 			 * in init? XXX
3198 			 */
3199 			printf("Process (pid %lu) got signal %d\n",
3200 				(u_long)p->p_pid, sig);
3201 #endif
3202 			return (SIGSTATUS_IGNORE);
3203 		}
3204 
3205 		/*
3206 		 * If there is a pending stop signal to process with
3207 		 * default action, stop here, then clear the signal.
3208 		 * Traced or exiting processes should ignore stops.
3209 		 * Additionally, a member of an orphaned process group
3210 		 * should ignore tty stops.
3211 		 */
3212 		prop = sigprop(sig);
3213 		if (prop & SIGPROP_STOP) {
3214 			mtx_unlock(&ps->ps_mtx);
3215 			if ((p->p_flag & (P_TRACED | P_WEXIT |
3216 			    P_SINGLE_EXIT)) != 0 || ((p->p_pgrp->
3217 			    pg_flags & PGRP_ORPHANED) != 0 &&
3218 			    (prop & SIGPROP_TTYSTOP) != 0)) {
3219 				mtx_lock(&ps->ps_mtx);
3220 				return (SIGSTATUS_IGNORE);
3221 			}
3222 			if (TD_SBDRY_INTR(td)) {
3223 				KASSERT((td->td_flags & TDF_SBDRY) != 0,
3224 				    ("lost TDF_SBDRY"));
3225 				mtx_lock(&ps->ps_mtx);
3226 				return (SIGSTATUS_SBDRY_STOP);
3227 			}
3228 			WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
3229 			    &p->p_mtx.lock_object, "Catching SIGSTOP");
3230 			sigqueue_delete(&td->td_sigqueue, sig);
3231 			sigqueue_delete(&p->p_sigqueue, sig);
3232 			p->p_flag |= P_STOPPED_SIG;
3233 			p->p_xsig = sig;
3234 			PROC_SLOCK(p);
3235 			sig_suspend_threads(td, p);
3236 			thread_suspend_switch(td, p);
3237 			PROC_SUNLOCK(p);
3238 			mtx_lock(&ps->ps_mtx);
3239 			return (SIGSTATUS_HANDLED);
3240 		} else if ((prop & SIGPROP_IGNORE) != 0 &&
3241 		    (td->td_flags & TDF_SIGWAIT) == 0) {
3242 			/*
3243 			 * Default action is to ignore; drop it if
3244 			 * not in kern_sigtimedwait().
3245 			 */
3246 			return (SIGSTATUS_IGNORE);
3247 		} else {
3248 			return (SIGSTATUS_HANDLE);
3249 		}
3250 
3251 	case (intptr_t)SIG_IGN:
3252 		if ((td->td_flags & TDF_SIGWAIT) == 0)
3253 			return (SIGSTATUS_IGNORE);
3254 		else
3255 			return (SIGSTATUS_HANDLE);
3256 
3257 	default:
3258 		/*
3259 		 * This signal has an action, let postsig() process it.
3260 		 */
3261 		return (SIGSTATUS_HANDLE);
3262 	}
3263 }
3264 
3265 /*
3266  * If the current process has received a signal (should be caught or cause
3267  * termination, should interrupt current syscall), return the signal number.
3268  * Stop signals with default action are processed immediately, then cleared;
3269  * they aren't returned.  This is checked after each entry to the system for
3270  * a syscall or trap (though this can usually be done without calling
3271  * issignal by checking the pending signal masks in cursig.) The normal call
3272  * sequence is
3273  *
3274  *	while (sig = cursig(curthread))
3275  *		postsig(sig);
3276  */
3277 static int
3278 issignal(struct thread *td)
3279 {
3280 	struct proc *p;
3281 	sigset_t sigpending;
3282 	int sig;
3283 
3284 	p = td->td_proc;
3285 	PROC_LOCK_ASSERT(p, MA_OWNED);
3286 
3287 	for (;;) {
3288 		sigpending = td->td_sigqueue.sq_signals;
3289 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
3290 		SIGSETNAND(sigpending, td->td_sigmask);
3291 
3292 		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
3293 		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
3294 			SIG_STOPSIGMASK(sigpending);
3295 		if (SIGISEMPTY(sigpending))	/* no signal to send */
3296 			return (0);
3297 
3298 		/*
3299 		 * Do fast sigblock if requested by usermode.  Since
3300 		 * we do know that there was a signal pending at this
3301 		 * point, set the FAST_SIGBLOCK_PEND as indicator for
3302 		 * usermode to perform a dummy call to
3303 		 * FAST_SIGBLOCK_UNBLOCK, which causes immediate
3304 		 * delivery of postponed pending signal.
3305 		 */
3306 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
3307 			if (td->td_sigblock_val != 0)
3308 				SIGSETNAND(sigpending, fastblock_mask);
3309 			if (SIGISEMPTY(sigpending)) {
3310 				td->td_pflags |= TDP_SIGFASTPENDING;
3311 				return (0);
3312 			}
3313 		}
3314 
3315 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
3316 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
3317 		    SIGISMEMBER(sigpending, SIGSTOP)) {
3318 			/*
3319 			 * If debugger just attached, always consume
3320 			 * SIGSTOP from ptrace(PT_ATTACH) first, to
3321 			 * execute the debugger attach ritual in
3322 			 * order.
3323 			 */
3324 			td->td_dbgflags |= TDB_FSTP;
3325 			SIGEMPTYSET(sigpending);
3326 			SIGADDSET(sigpending, SIGSTOP);
3327 		}
3328 
3329 		SIG_FOREACH(sig, &sigpending) {
3330 			switch (sigprocess(td, sig)) {
3331 			case SIGSTATUS_HANDLE:
3332 				return (sig);
3333 			case SIGSTATUS_HANDLED:
3334 				goto next;
3335 			case SIGSTATUS_IGNORE:
3336 				sigqueue_delete(&td->td_sigqueue, sig);
3337 				sigqueue_delete(&p->p_sigqueue, sig);
3338 				break;
3339 			case SIGSTATUS_SBDRY_STOP:
3340 				return (-1);
3341 			}
3342 		}
3343 next:;
3344 	}
3345 }
3346 
3347 void
3348 thread_stopped(struct proc *p)
3349 {
3350 	int n;
3351 
3352 	PROC_LOCK_ASSERT(p, MA_OWNED);
3353 	PROC_SLOCK_ASSERT(p, MA_OWNED);
3354 	n = p->p_suspcount;
3355 	if (p == curproc)
3356 		n++;
3357 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3358 		PROC_SUNLOCK(p);
3359 		p->p_flag &= ~P_WAITED;
3360 		PROC_LOCK(p->p_pptr);
3361 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
3362 			CLD_TRAPPED : CLD_STOPPED);
3363 		PROC_UNLOCK(p->p_pptr);
3364 		PROC_SLOCK(p);
3365 	}
3366 }
3367 
3368 /*
3369  * Take the action for the specified signal
3370  * from the current set of pending signals.
3371  */
3372 int
3373 postsig(int sig)
3374 {
3375 	struct thread *td;
3376 	struct proc *p;
3377 	struct sigacts *ps;
3378 	sig_t action;
3379 	ksiginfo_t ksi;
3380 	sigset_t returnmask;
3381 
3382 	KASSERT(sig != 0, ("postsig"));
3383 
3384 	td = curthread;
3385 	p = td->td_proc;
3386 	PROC_LOCK_ASSERT(p, MA_OWNED);
3387 	ps = p->p_sigacts;
3388 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3389 	ksiginfo_init(&ksi);
3390 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3391 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3392 		return (0);
3393 	ksi.ksi_signo = sig;
3394 	if (ksi.ksi_code == SI_TIMER)
3395 		itimer_accept(p, ksi.ksi_timerid, &ksi);
3396 	action = ps->ps_sigact[_SIG_IDX(sig)];
3397 #ifdef KTRACE
3398 	if (KTRPOINT(td, KTR_PSIG))
3399 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3400 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3401 #endif
3402 
3403 	if (action == SIG_DFL) {
3404 		/*
3405 		 * Default action, where the default is to kill
3406 		 * the process.  (Other cases were ignored above.)
3407 		 */
3408 		mtx_unlock(&ps->ps_mtx);
3409 		proc_td_siginfo_capture(td, &ksi.ksi_info);
3410 		sigexit(td, sig);
3411 		/* NOTREACHED */
3412 	} else {
3413 		/*
3414 		 * If we get here, the signal must be caught.
3415 		 */
3416 		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3417 		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3418 		    ("postsig action: blocked sig %d", sig));
3419 
3420 		/*
3421 		 * Set the new mask value and also defer further
3422 		 * occurrences of this signal.
3423 		 *
3424 		 * Special case: user has done a sigsuspend.  Here the
3425 		 * current mask is not of interest, but rather the
3426 		 * mask from before the sigsuspend is what we want
3427 		 * restored after the signal processing is completed.
3428 		 */
3429 		if (td->td_pflags & TDP_OLDMASK) {
3430 			returnmask = td->td_oldsigmask;
3431 			td->td_pflags &= ~TDP_OLDMASK;
3432 		} else
3433 			returnmask = td->td_sigmask;
3434 
3435 		if (p->p_sig == sig) {
3436 			p->p_sig = 0;
3437 		}
3438 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3439 		postsig_done(sig, td, ps);
3440 	}
3441 	return (1);
3442 }
3443 
3444 int
3445 sig_ast_checksusp(struct thread *td)
3446 {
3447 	struct proc *p __diagused;
3448 	int ret;
3449 
3450 	p = td->td_proc;
3451 	PROC_LOCK_ASSERT(p, MA_OWNED);
3452 
3453 	if (!td_ast_pending(td, TDA_SUSPEND))
3454 		return (0);
3455 
3456 	ret = thread_suspend_check(1);
3457 	MPASS(ret == 0 || ret == EINTR || ret == ERESTART);
3458 	return (ret);
3459 }
3460 
3461 int
3462 sig_ast_needsigchk(struct thread *td)
3463 {
3464 	struct proc *p;
3465 	struct sigacts *ps;
3466 	int ret, sig;
3467 
3468 	p = td->td_proc;
3469 	PROC_LOCK_ASSERT(p, MA_OWNED);
3470 
3471 	if (!td_ast_pending(td, TDA_SIG))
3472 		return (0);
3473 
3474 	ps = p->p_sigacts;
3475 	mtx_lock(&ps->ps_mtx);
3476 	sig = cursig(td);
3477 	if (sig == -1) {
3478 		mtx_unlock(&ps->ps_mtx);
3479 		KASSERT((td->td_flags & TDF_SBDRY) != 0, ("lost TDF_SBDRY"));
3480 		KASSERT(TD_SBDRY_INTR(td),
3481 		    ("lost TDF_SERESTART of TDF_SEINTR"));
3482 		KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
3483 		    (TDF_SEINTR | TDF_SERESTART),
3484 		    ("both TDF_SEINTR and TDF_SERESTART"));
3485 		ret = TD_SBDRY_ERRNO(td);
3486 	} else if (sig != 0) {
3487 		ret = SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR : ERESTART;
3488 		mtx_unlock(&ps->ps_mtx);
3489 	} else {
3490 		mtx_unlock(&ps->ps_mtx);
3491 		ret = 0;
3492 	}
3493 
3494 	/*
3495 	 * Do not go into sleep if this thread was the ptrace(2)
3496 	 * attach leader.  cursig() consumed SIGSTOP from PT_ATTACH,
3497 	 * but we usually act on the signal by interrupting sleep, and
3498 	 * should do that here as well.
3499 	 */
3500 	if ((td->td_dbgflags & TDB_FSTP) != 0) {
3501 		if (ret == 0)
3502 			ret = EINTR;
3503 		td->td_dbgflags &= ~TDB_FSTP;
3504 	}
3505 
3506 	return (ret);
3507 }
3508 
3509 int
3510 sig_intr(void)
3511 {
3512 	struct thread *td;
3513 	struct proc *p;
3514 	int ret;
3515 
3516 	td = curthread;
3517 	if (!td_ast_pending(td, TDA_SIG) && !td_ast_pending(td, TDA_SUSPEND))
3518 		return (0);
3519 
3520 	p = td->td_proc;
3521 
3522 	PROC_LOCK(p);
3523 	ret = sig_ast_checksusp(td);
3524 	if (ret == 0)
3525 		ret = sig_ast_needsigchk(td);
3526 	PROC_UNLOCK(p);
3527 	return (ret);
3528 }
3529 
3530 bool
3531 curproc_sigkilled(void)
3532 {
3533 	struct thread *td;
3534 	struct proc *p;
3535 	struct sigacts *ps;
3536 	bool res;
3537 
3538 	td = curthread;
3539 	if (!td_ast_pending(td, TDA_SIG))
3540 		return (false);
3541 
3542 	p = td->td_proc;
3543 	PROC_LOCK(p);
3544 	ps = p->p_sigacts;
3545 	mtx_lock(&ps->ps_mtx);
3546 	res = SIGISMEMBER(td->td_sigqueue.sq_signals, SIGKILL) ||
3547 	    SIGISMEMBER(p->p_sigqueue.sq_signals, SIGKILL);
3548 	mtx_unlock(&ps->ps_mtx);
3549 	PROC_UNLOCK(p);
3550 	return (res);
3551 }
3552 
3553 void
3554 proc_wkilled(struct proc *p)
3555 {
3556 
3557 	PROC_LOCK_ASSERT(p, MA_OWNED);
3558 	if ((p->p_flag & P_WKILLED) == 0) {
3559 		p->p_flag |= P_WKILLED;
3560 		/*
3561 		 * Notify swapper that there is a process to swap in.
3562 		 * The notification is racy, at worst it would take 10
3563 		 * seconds for the swapper process to notice.
3564 		 */
3565 		if ((p->p_flag & (P_INMEM | P_SWAPPINGIN)) == 0)
3566 			wakeup(&proc0);
3567 	}
3568 }
3569 
3570 /*
3571  * Kill the current process for stated reason.
3572  */
3573 void
3574 killproc(struct proc *p, const char *why)
3575 {
3576 
3577 	PROC_LOCK_ASSERT(p, MA_OWNED);
3578 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3579 	    p->p_comm);
3580 	log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3581 	    p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3582 	    p->p_ucred->cr_uid, why);
3583 	proc_wkilled(p);
3584 	kern_psignal(p, SIGKILL);
3585 }
3586 
3587 /*
3588  * Force the current process to exit with the specified signal, dumping core
3589  * if appropriate.  We bypass the normal tests for masked and caught signals,
3590  * allowing unrecoverable failures to terminate the process without changing
3591  * signal state.  Mark the accounting record with the signal termination.
3592  * If dumping core, save the signal number for the debugger.  Calls exit and
3593  * does not return.
3594  */
3595 void
3596 sigexit(struct thread *td, int sig)
3597 {
3598 	struct proc *p = td->td_proc;
3599 	const char *coreinfo;
3600 	int rv;
3601 
3602 	PROC_LOCK_ASSERT(p, MA_OWNED);
3603 	proc_set_p2_wexit(p);
3604 
3605 	p->p_acflag |= AXSIG;
3606 	/*
3607 	 * We must be single-threading to generate a core dump.  This
3608 	 * ensures that the registers in the core file are up-to-date.
3609 	 * Also, the ELF dump handler assumes that the thread list doesn't
3610 	 * change out from under it.
3611 	 *
3612 	 * XXX If another thread attempts to single-thread before us
3613 	 *     (e.g. via fork()), we won't get a dump at all.
3614 	 */
3615 	if ((sigprop(sig) & SIGPROP_CORE) &&
3616 	    thread_single(p, SINGLE_NO_EXIT) == 0) {
3617 		p->p_sig = sig;
3618 		/*
3619 		 * Log signals which would cause core dumps
3620 		 * (Log as LOG_INFO to appease those who don't want
3621 		 * these messages.)
3622 		 * XXX : Todo, as well as euid, write out ruid too
3623 		 * Note that coredump() drops proc lock.
3624 		 */
3625 		rv = coredump(td);
3626 		switch (rv) {
3627 		case 0:
3628 			sig |= WCOREFLAG;
3629 			coreinfo = " (core dumped)";
3630 			break;
3631 		case EFAULT:
3632 			coreinfo = " (no core dump - bad address)";
3633 			break;
3634 		case EINVAL:
3635 			coreinfo = " (no core dump - invalid argument)";
3636 			break;
3637 		case EFBIG:
3638 			coreinfo = " (no core dump - too large)";
3639 			break;
3640 		default:
3641 			coreinfo = " (no core dump - other error)";
3642 			break;
3643 		}
3644 		if (kern_logsigexit)
3645 			log(LOG_INFO,
3646 			    "pid %d (%s), jid %d, uid %d: exited on "
3647 			    "signal %d%s\n", p->p_pid, p->p_comm,
3648 			    p->p_ucred->cr_prison->pr_id,
3649 			    td->td_ucred->cr_uid,
3650 			    sig &~ WCOREFLAG, coreinfo);
3651 	} else
3652 		PROC_UNLOCK(p);
3653 	exit1(td, 0, sig);
3654 	/* NOTREACHED */
3655 }
3656 
3657 /*
3658  * Send queued SIGCHLD to parent when child process's state
3659  * is changed.
3660  */
3661 static void
3662 sigparent(struct proc *p, int reason, int status)
3663 {
3664 	PROC_LOCK_ASSERT(p, MA_OWNED);
3665 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3666 
3667 	if (p->p_ksi != NULL) {
3668 		p->p_ksi->ksi_signo  = SIGCHLD;
3669 		p->p_ksi->ksi_code   = reason;
3670 		p->p_ksi->ksi_status = status;
3671 		p->p_ksi->ksi_pid    = p->p_pid;
3672 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3673 		if (KSI_ONQ(p->p_ksi))
3674 			return;
3675 	}
3676 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3677 }
3678 
3679 static void
3680 childproc_jobstate(struct proc *p, int reason, int sig)
3681 {
3682 	struct sigacts *ps;
3683 
3684 	PROC_LOCK_ASSERT(p, MA_OWNED);
3685 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3686 
3687 	/*
3688 	 * Wake up parent sleeping in kern_wait(), also send
3689 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3690 	 * that parent will awake, because parent may masked
3691 	 * the signal.
3692 	 */
3693 	p->p_pptr->p_flag |= P_STATCHILD;
3694 	wakeup(p->p_pptr);
3695 
3696 	ps = p->p_pptr->p_sigacts;
3697 	mtx_lock(&ps->ps_mtx);
3698 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3699 		mtx_unlock(&ps->ps_mtx);
3700 		sigparent(p, reason, sig);
3701 	} else
3702 		mtx_unlock(&ps->ps_mtx);
3703 }
3704 
3705 void
3706 childproc_stopped(struct proc *p, int reason)
3707 {
3708 
3709 	childproc_jobstate(p, reason, p->p_xsig);
3710 }
3711 
3712 void
3713 childproc_continued(struct proc *p)
3714 {
3715 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3716 }
3717 
3718 void
3719 childproc_exited(struct proc *p)
3720 {
3721 	int reason, status;
3722 
3723 	if (WCOREDUMP(p->p_xsig)) {
3724 		reason = CLD_DUMPED;
3725 		status = WTERMSIG(p->p_xsig);
3726 	} else if (WIFSIGNALED(p->p_xsig)) {
3727 		reason = CLD_KILLED;
3728 		status = WTERMSIG(p->p_xsig);
3729 	} else {
3730 		reason = CLD_EXITED;
3731 		status = p->p_xexit;
3732 	}
3733 	/*
3734 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3735 	 * done in exit1().
3736 	 */
3737 	sigparent(p, reason, status);
3738 }
3739 
3740 #define	MAX_NUM_CORE_FILES 100000
3741 #ifndef NUM_CORE_FILES
3742 #define	NUM_CORE_FILES 5
3743 #endif
3744 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3745 static int num_cores = NUM_CORE_FILES;
3746 
3747 static int
3748 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3749 {
3750 	int error;
3751 	int new_val;
3752 
3753 	new_val = num_cores;
3754 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3755 	if (error != 0 || req->newptr == NULL)
3756 		return (error);
3757 	if (new_val > MAX_NUM_CORE_FILES)
3758 		new_val = MAX_NUM_CORE_FILES;
3759 	if (new_val < 0)
3760 		new_val = 0;
3761 	num_cores = new_val;
3762 	return (0);
3763 }
3764 SYSCTL_PROC(_debug, OID_AUTO, ncores,
3765     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
3766     sysctl_debug_num_cores_check, "I",
3767     "Maximum number of generated process corefiles while using index format");
3768 
3769 #define	GZIP_SUFFIX	".gz"
3770 #define	ZSTD_SUFFIX	".zst"
3771 
3772 int compress_user_cores = 0;
3773 
3774 static int
3775 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3776 {
3777 	int error, val;
3778 
3779 	val = compress_user_cores;
3780 	error = sysctl_handle_int(oidp, &val, 0, req);
3781 	if (error != 0 || req->newptr == NULL)
3782 		return (error);
3783 	if (val != 0 && !compressor_avail(val))
3784 		return (EINVAL);
3785 	compress_user_cores = val;
3786 	return (error);
3787 }
3788 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores,
3789     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, 0, sizeof(int),
3790     sysctl_compress_user_cores, "I",
3791     "Enable compression of user corefiles ("
3792     __XSTRING(COMPRESS_GZIP) " = gzip, "
3793     __XSTRING(COMPRESS_ZSTD) " = zstd)");
3794 
3795 int compress_user_cores_level = 6;
3796 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3797     &compress_user_cores_level, 0,
3798     "Corefile compression level");
3799 
3800 /*
3801  * Protect the access to corefilename[] by allproc_lock.
3802  */
3803 #define	corefilename_lock	allproc_lock
3804 
3805 static char corefilename[MAXPATHLEN] = {"%N.core"};
3806 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3807 
3808 static int
3809 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3810 {
3811 	int error;
3812 
3813 	sx_xlock(&corefilename_lock);
3814 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3815 	    req);
3816 	sx_xunlock(&corefilename_lock);
3817 
3818 	return (error);
3819 }
3820 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3821     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3822     "Process corefile name format string");
3823 
3824 static void
3825 vnode_close_locked(struct thread *td, struct vnode *vp)
3826 {
3827 
3828 	VOP_UNLOCK(vp);
3829 	vn_close(vp, FWRITE, td->td_ucred, td);
3830 }
3831 
3832 /*
3833  * If the core format has a %I in it, then we need to check
3834  * for existing corefiles before defining a name.
3835  * To do this we iterate over 0..ncores to find a
3836  * non-existing core file name to use. If all core files are
3837  * already used we choose the oldest one.
3838  */
3839 static int
3840 corefile_open_last(struct thread *td, char *name, int indexpos,
3841     int indexlen, int ncores, struct vnode **vpp)
3842 {
3843 	struct vnode *oldvp, *nextvp, *vp;
3844 	struct vattr vattr;
3845 	struct nameidata nd;
3846 	int error, i, flags, oflags, cmode;
3847 	char ch;
3848 	struct timespec lasttime;
3849 
3850 	nextvp = oldvp = NULL;
3851 	cmode = S_IRUSR | S_IWUSR;
3852 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3853 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3854 
3855 	for (i = 0; i < ncores; i++) {
3856 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3857 
3858 		ch = name[indexpos + indexlen];
3859 		(void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3860 		    i);
3861 		name[indexpos + indexlen] = ch;
3862 
3863 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
3864 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3865 		    NULL);
3866 		if (error != 0)
3867 			break;
3868 
3869 		vp = nd.ni_vp;
3870 		NDFREE_PNBUF(&nd);
3871 		if ((flags & O_CREAT) == O_CREAT) {
3872 			nextvp = vp;
3873 			break;
3874 		}
3875 
3876 		error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3877 		if (error != 0) {
3878 			vnode_close_locked(td, vp);
3879 			break;
3880 		}
3881 
3882 		if (oldvp == NULL ||
3883 		    lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3884 		    (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3885 		    lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3886 			if (oldvp != NULL)
3887 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3888 			oldvp = vp;
3889 			VOP_UNLOCK(oldvp);
3890 			lasttime = vattr.va_mtime;
3891 		} else {
3892 			vnode_close_locked(td, vp);
3893 		}
3894 	}
3895 
3896 	if (oldvp != NULL) {
3897 		if (nextvp == NULL) {
3898 			if ((td->td_proc->p_flag & P_SUGID) != 0) {
3899 				error = EFAULT;
3900 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3901 			} else {
3902 				nextvp = oldvp;
3903 				error = vn_lock(nextvp, LK_EXCLUSIVE);
3904 				if (error != 0) {
3905 					vn_close(nextvp, FWRITE, td->td_ucred,
3906 					    td);
3907 					nextvp = NULL;
3908 				}
3909 			}
3910 		} else {
3911 			vn_close(oldvp, FWRITE, td->td_ucred, td);
3912 		}
3913 	}
3914 	if (error != 0) {
3915 		if (nextvp != NULL)
3916 			vnode_close_locked(td, oldvp);
3917 	} else {
3918 		*vpp = nextvp;
3919 	}
3920 
3921 	return (error);
3922 }
3923 
3924 /*
3925  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3926  * Expand the name described in corefilename, using name, uid, and pid
3927  * and open/create core file.
3928  * corefilename is a printf-like string, with three format specifiers:
3929  *	%N	name of process ("name")
3930  *	%P	process id (pid)
3931  *	%U	user id (uid)
3932  * For example, "%N.core" is the default; they can be disabled completely
3933  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3934  * This is controlled by the sysctl variable kern.corefile (see above).
3935  */
3936 static int
3937 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3938     int compress, int signum, struct vnode **vpp, char **namep)
3939 {
3940 	struct sbuf sb;
3941 	struct nameidata nd;
3942 	const char *format;
3943 	char *hostname, *name;
3944 	int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
3945 
3946 	hostname = NULL;
3947 	format = corefilename;
3948 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3949 	indexlen = 0;
3950 	indexpos = -1;
3951 	ncores = num_cores;
3952 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3953 	sx_slock(&corefilename_lock);
3954 	for (i = 0; format[i] != '\0'; i++) {
3955 		switch (format[i]) {
3956 		case '%':	/* Format character */
3957 			i++;
3958 			switch (format[i]) {
3959 			case '%':
3960 				sbuf_putc(&sb, '%');
3961 				break;
3962 			case 'H':	/* hostname */
3963 				if (hostname == NULL) {
3964 					hostname = malloc(MAXHOSTNAMELEN,
3965 					    M_TEMP, M_WAITOK);
3966 				}
3967 				getcredhostname(td->td_ucred, hostname,
3968 				    MAXHOSTNAMELEN);
3969 				sbuf_printf(&sb, "%s", hostname);
3970 				break;
3971 			case 'I':	/* autoincrementing index */
3972 				if (indexpos != -1) {
3973 					sbuf_printf(&sb, "%%I");
3974 					break;
3975 				}
3976 
3977 				indexpos = sbuf_len(&sb);
3978 				sbuf_printf(&sb, "%u", ncores - 1);
3979 				indexlen = sbuf_len(&sb) - indexpos;
3980 				break;
3981 			case 'N':	/* process name */
3982 				sbuf_printf(&sb, "%s", comm);
3983 				break;
3984 			case 'P':	/* process id */
3985 				sbuf_printf(&sb, "%u", pid);
3986 				break;
3987 			case 'S':	/* signal number */
3988 				sbuf_printf(&sb, "%i", signum);
3989 				break;
3990 			case 'U':	/* user id */
3991 				sbuf_printf(&sb, "%u", uid);
3992 				break;
3993 			default:
3994 				log(LOG_ERR,
3995 				    "Unknown format character %c in "
3996 				    "corename `%s'\n", format[i], format);
3997 				break;
3998 			}
3999 			break;
4000 		default:
4001 			sbuf_putc(&sb, format[i]);
4002 			break;
4003 		}
4004 	}
4005 	sx_sunlock(&corefilename_lock);
4006 	free(hostname, M_TEMP);
4007 	if (compress == COMPRESS_GZIP)
4008 		sbuf_printf(&sb, GZIP_SUFFIX);
4009 	else if (compress == COMPRESS_ZSTD)
4010 		sbuf_printf(&sb, ZSTD_SUFFIX);
4011 	if (sbuf_error(&sb) != 0) {
4012 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
4013 		    "long\n", (long)pid, comm, (u_long)uid);
4014 		sbuf_delete(&sb);
4015 		free(name, M_TEMP);
4016 		return (ENOMEM);
4017 	}
4018 	sbuf_finish(&sb);
4019 	sbuf_delete(&sb);
4020 
4021 	if (indexpos != -1) {
4022 		error = corefile_open_last(td, name, indexpos, indexlen, ncores,
4023 		    vpp);
4024 		if (error != 0) {
4025 			log(LOG_ERR,
4026 			    "pid %d (%s), uid (%u):  Path `%s' failed "
4027 			    "on initial open test, error = %d\n",
4028 			    pid, comm, uid, name, error);
4029 		}
4030 	} else {
4031 		cmode = S_IRUSR | S_IWUSR;
4032 		oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
4033 		    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
4034 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
4035 		if ((td->td_proc->p_flag & P_SUGID) != 0)
4036 			flags |= O_EXCL;
4037 
4038 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
4039 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
4040 		    NULL);
4041 		if (error == 0) {
4042 			*vpp = nd.ni_vp;
4043 			NDFREE_PNBUF(&nd);
4044 		}
4045 	}
4046 
4047 	if (error != 0) {
4048 #ifdef AUDIT
4049 		audit_proc_coredump(td, name, error);
4050 #endif
4051 		free(name, M_TEMP);
4052 		return (error);
4053 	}
4054 	*namep = name;
4055 	return (0);
4056 }
4057 
4058 /*
4059  * Dump a process' core.  The main routine does some
4060  * policy checking, and creates the name of the coredump;
4061  * then it passes on a vnode and a size limit to the process-specific
4062  * coredump routine if there is one; if there _is not_ one, it returns
4063  * ENOSYS; otherwise it returns the error from the process-specific routine.
4064  */
4065 
4066 static int
4067 coredump(struct thread *td)
4068 {
4069 	struct proc *p = td->td_proc;
4070 	struct ucred *cred = td->td_ucred;
4071 	struct vnode *vp;
4072 	struct flock lf;
4073 	struct vattr vattr;
4074 	size_t fullpathsize;
4075 	int error, error1, locked;
4076 	char *name;			/* name of corefile */
4077 	void *rl_cookie;
4078 	off_t limit;
4079 	char *fullpath, *freepath = NULL;
4080 	struct sbuf *sb;
4081 
4082 	PROC_LOCK_ASSERT(p, MA_OWNED);
4083 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
4084 
4085 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
4086 	    (p->p_flag2 & P2_NOTRACE) != 0) {
4087 		PROC_UNLOCK(p);
4088 		return (EFAULT);
4089 	}
4090 
4091 	/*
4092 	 * Note that the bulk of limit checking is done after
4093 	 * the corefile is created.  The exception is if the limit
4094 	 * for corefiles is 0, in which case we don't bother
4095 	 * creating the corefile at all.  This layout means that
4096 	 * a corefile is truncated instead of not being created,
4097 	 * if it is larger than the limit.
4098 	 */
4099 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
4100 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
4101 		PROC_UNLOCK(p);
4102 		return (EFBIG);
4103 	}
4104 	PROC_UNLOCK(p);
4105 
4106 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
4107 	    compress_user_cores, p->p_sig, &vp, &name);
4108 	if (error != 0)
4109 		return (error);
4110 
4111 	/*
4112 	 * Don't dump to non-regular files or files with links.
4113 	 * Do not dump into system files. Effective user must own the corefile.
4114 	 */
4115 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
4116 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
4117 	    vattr.va_uid != cred->cr_uid) {
4118 		VOP_UNLOCK(vp);
4119 		error = EFAULT;
4120 		goto out;
4121 	}
4122 
4123 	VOP_UNLOCK(vp);
4124 
4125 	/* Postpone other writers, including core dumps of other processes. */
4126 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
4127 
4128 	lf.l_whence = SEEK_SET;
4129 	lf.l_start = 0;
4130 	lf.l_len = 0;
4131 	lf.l_type = F_WRLCK;
4132 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
4133 
4134 	VATTR_NULL(&vattr);
4135 	vattr.va_size = 0;
4136 	if (set_core_nodump_flag)
4137 		vattr.va_flags = UF_NODUMP;
4138 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4139 	VOP_SETATTR(vp, &vattr, cred);
4140 	VOP_UNLOCK(vp);
4141 	PROC_LOCK(p);
4142 	p->p_acflag |= ACORE;
4143 	PROC_UNLOCK(p);
4144 
4145 	if (p->p_sysent->sv_coredump != NULL) {
4146 		error = p->p_sysent->sv_coredump(td, vp, limit, 0);
4147 	} else {
4148 		error = ENOSYS;
4149 	}
4150 
4151 	if (locked) {
4152 		lf.l_type = F_UNLCK;
4153 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
4154 	}
4155 	vn_rangelock_unlock(vp, rl_cookie);
4156 
4157 	/*
4158 	 * Notify the userland helper that a process triggered a core dump.
4159 	 * This allows the helper to run an automated debugging session.
4160 	 */
4161 	if (error != 0 || coredump_devctl == 0)
4162 		goto out;
4163 	sb = sbuf_new_auto();
4164 	if (vn_fullpath_global(p->p_textvp, &fullpath, &freepath) != 0)
4165 		goto out2;
4166 	sbuf_printf(sb, "comm=\"");
4167 	devctl_safe_quote_sb(sb, fullpath);
4168 	free(freepath, M_TEMP);
4169 	sbuf_printf(sb, "\" core=\"");
4170 
4171 	/*
4172 	 * We can't lookup core file vp directly. When we're replacing a core, and
4173 	 * other random times, we flush the name cache, so it will fail. Instead,
4174 	 * if the path of the core is relative, add the current dir in front if it.
4175 	 */
4176 	if (name[0] != '/') {
4177 		fullpathsize = MAXPATHLEN;
4178 		freepath = malloc(fullpathsize, M_TEMP, M_WAITOK);
4179 		if (vn_getcwd(freepath, &fullpath, &fullpathsize) != 0) {
4180 			free(freepath, M_TEMP);
4181 			goto out2;
4182 		}
4183 		devctl_safe_quote_sb(sb, fullpath);
4184 		free(freepath, M_TEMP);
4185 		sbuf_putc(sb, '/');
4186 	}
4187 	devctl_safe_quote_sb(sb, name);
4188 	sbuf_printf(sb, "\"");
4189 	if (sbuf_finish(sb) == 0)
4190 		devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
4191 out2:
4192 	sbuf_delete(sb);
4193 out:
4194 	error1 = vn_close(vp, FWRITE, cred, td);
4195 	if (error == 0)
4196 		error = error1;
4197 #ifdef AUDIT
4198 	audit_proc_coredump(td, name, error);
4199 #endif
4200 	free(name, M_TEMP);
4201 	return (error);
4202 }
4203 
4204 /*
4205  * Nonexistent system call-- signal process (may want to handle it).  Flag
4206  * error in case process won't see signal immediately (blocked or ignored).
4207  */
4208 #ifndef _SYS_SYSPROTO_H_
4209 struct nosys_args {
4210 	int	dummy;
4211 };
4212 #endif
4213 /* ARGSUSED */
4214 int
4215 nosys(struct thread *td, struct nosys_args *args)
4216 {
4217 	struct proc *p;
4218 
4219 	p = td->td_proc;
4220 
4221 	PROC_LOCK(p);
4222 	tdsignal(td, SIGSYS);
4223 	PROC_UNLOCK(p);
4224 	if (kern_lognosys == 1 || kern_lognosys == 3) {
4225 		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4226 		    td->td_sa.code);
4227 	}
4228 	if (kern_lognosys == 2 || kern_lognosys == 3 ||
4229 	    (p->p_pid == 1 && (kern_lognosys & 3) == 0)) {
4230 		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4231 		    td->td_sa.code);
4232 	}
4233 	return (ENOSYS);
4234 }
4235 
4236 /*
4237  * Send a SIGIO or SIGURG signal to a process or process group using stored
4238  * credentials rather than those of the current process.
4239  */
4240 void
4241 pgsigio(struct sigio **sigiop, int sig, int checkctty)
4242 {
4243 	ksiginfo_t ksi;
4244 	struct sigio *sigio;
4245 
4246 	ksiginfo_init(&ksi);
4247 	ksi.ksi_signo = sig;
4248 	ksi.ksi_code = SI_KERNEL;
4249 
4250 	SIGIO_LOCK();
4251 	sigio = *sigiop;
4252 	if (sigio == NULL) {
4253 		SIGIO_UNLOCK();
4254 		return;
4255 	}
4256 	if (sigio->sio_pgid > 0) {
4257 		PROC_LOCK(sigio->sio_proc);
4258 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
4259 			kern_psignal(sigio->sio_proc, sig);
4260 		PROC_UNLOCK(sigio->sio_proc);
4261 	} else if (sigio->sio_pgid < 0) {
4262 		struct proc *p;
4263 
4264 		PGRP_LOCK(sigio->sio_pgrp);
4265 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
4266 			PROC_LOCK(p);
4267 			if (p->p_state == PRS_NORMAL &&
4268 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
4269 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
4270 				kern_psignal(p, sig);
4271 			PROC_UNLOCK(p);
4272 		}
4273 		PGRP_UNLOCK(sigio->sio_pgrp);
4274 	}
4275 	SIGIO_UNLOCK();
4276 }
4277 
4278 static int
4279 filt_sigattach(struct knote *kn)
4280 {
4281 	struct proc *p = curproc;
4282 
4283 	kn->kn_ptr.p_proc = p;
4284 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
4285 
4286 	knlist_add(p->p_klist, kn, 0);
4287 
4288 	return (0);
4289 }
4290 
4291 static void
4292 filt_sigdetach(struct knote *kn)
4293 {
4294 	struct proc *p = kn->kn_ptr.p_proc;
4295 
4296 	knlist_remove(p->p_klist, kn, 0);
4297 }
4298 
4299 /*
4300  * signal knotes are shared with proc knotes, so we apply a mask to
4301  * the hint in order to differentiate them from process hints.  This
4302  * could be avoided by using a signal-specific knote list, but probably
4303  * isn't worth the trouble.
4304  */
4305 static int
4306 filt_signal(struct knote *kn, long hint)
4307 {
4308 
4309 	if (hint & NOTE_SIGNAL) {
4310 		hint &= ~NOTE_SIGNAL;
4311 
4312 		if (kn->kn_id == hint)
4313 			kn->kn_data++;
4314 	}
4315 	return (kn->kn_data != 0);
4316 }
4317 
4318 struct sigacts *
4319 sigacts_alloc(void)
4320 {
4321 	struct sigacts *ps;
4322 
4323 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
4324 	refcount_init(&ps->ps_refcnt, 1);
4325 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
4326 	return (ps);
4327 }
4328 
4329 void
4330 sigacts_free(struct sigacts *ps)
4331 {
4332 
4333 	if (refcount_release(&ps->ps_refcnt) == 0)
4334 		return;
4335 	mtx_destroy(&ps->ps_mtx);
4336 	free(ps, M_SUBPROC);
4337 }
4338 
4339 struct sigacts *
4340 sigacts_hold(struct sigacts *ps)
4341 {
4342 
4343 	refcount_acquire(&ps->ps_refcnt);
4344 	return (ps);
4345 }
4346 
4347 void
4348 sigacts_copy(struct sigacts *dest, struct sigacts *src)
4349 {
4350 
4351 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
4352 	mtx_lock(&src->ps_mtx);
4353 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
4354 	mtx_unlock(&src->ps_mtx);
4355 }
4356 
4357 int
4358 sigacts_shared(struct sigacts *ps)
4359 {
4360 
4361 	return (ps->ps_refcnt > 1);
4362 }
4363 
4364 void
4365 sig_drop_caught(struct proc *p)
4366 {
4367 	int sig;
4368 	struct sigacts *ps;
4369 
4370 	ps = p->p_sigacts;
4371 	PROC_LOCK_ASSERT(p, MA_OWNED);
4372 	mtx_assert(&ps->ps_mtx, MA_OWNED);
4373 	SIG_FOREACH(sig, &ps->ps_sigcatch) {
4374 		sigdflt(ps, sig);
4375 		if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
4376 			sigqueue_delete_proc(p, sig);
4377 	}
4378 }
4379 
4380 static void
4381 sigfastblock_failed(struct thread *td, bool sendsig, bool write)
4382 {
4383 	ksiginfo_t ksi;
4384 
4385 	/*
4386 	 * Prevent further fetches and SIGSEGVs, allowing thread to
4387 	 * issue syscalls despite corruption.
4388 	 */
4389 	sigfastblock_clear(td);
4390 
4391 	if (!sendsig)
4392 		return;
4393 	ksiginfo_init_trap(&ksi);
4394 	ksi.ksi_signo = SIGSEGV;
4395 	ksi.ksi_code = write ? SEGV_ACCERR : SEGV_MAPERR;
4396 	ksi.ksi_addr = td->td_sigblock_ptr;
4397 	trapsignal(td, &ksi);
4398 }
4399 
4400 static bool
4401 sigfastblock_fetch_sig(struct thread *td, bool sendsig, uint32_t *valp)
4402 {
4403 	uint32_t res;
4404 
4405 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4406 		return (true);
4407 	if (fueword32((void *)td->td_sigblock_ptr, &res) == -1) {
4408 		sigfastblock_failed(td, sendsig, false);
4409 		return (false);
4410 	}
4411 	*valp = res;
4412 	td->td_sigblock_val = res & ~SIGFASTBLOCK_FLAGS;
4413 	return (true);
4414 }
4415 
4416 static void
4417 sigfastblock_resched(struct thread *td, bool resched)
4418 {
4419 	struct proc *p;
4420 
4421 	if (resched) {
4422 		p = td->td_proc;
4423 		PROC_LOCK(p);
4424 		reschedule_signals(p, td->td_sigmask, 0);
4425 		PROC_UNLOCK(p);
4426 	}
4427 	ast_sched(td, TDA_SIG);
4428 }
4429 
4430 int
4431 sys_sigfastblock(struct thread *td, struct sigfastblock_args *uap)
4432 {
4433 	struct proc *p;
4434 	int error, res;
4435 	uint32_t oldval;
4436 
4437 	error = 0;
4438 	p = td->td_proc;
4439 	switch (uap->cmd) {
4440 	case SIGFASTBLOCK_SETPTR:
4441 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
4442 			error = EBUSY;
4443 			break;
4444 		}
4445 		if (((uintptr_t)(uap->ptr) & (sizeof(uint32_t) - 1)) != 0) {
4446 			error = EINVAL;
4447 			break;
4448 		}
4449 		td->td_pflags |= TDP_SIGFASTBLOCK;
4450 		td->td_sigblock_ptr = uap->ptr;
4451 		break;
4452 
4453 	case SIGFASTBLOCK_UNBLOCK:
4454 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4455 			error = EINVAL;
4456 			break;
4457 		}
4458 
4459 		for (;;) {
4460 			res = casueword32(td->td_sigblock_ptr,
4461 			    SIGFASTBLOCK_PEND, &oldval, 0);
4462 			if (res == -1) {
4463 				error = EFAULT;
4464 				sigfastblock_failed(td, false, true);
4465 				break;
4466 			}
4467 			if (res == 0)
4468 				break;
4469 			MPASS(res == 1);
4470 			if (oldval != SIGFASTBLOCK_PEND) {
4471 				error = EBUSY;
4472 				break;
4473 			}
4474 			error = thread_check_susp(td, false);
4475 			if (error != 0)
4476 				break;
4477 		}
4478 		if (error != 0)
4479 			break;
4480 
4481 		/*
4482 		 * td_sigblock_val is cleared there, but not on a
4483 		 * syscall exit.  The end effect is that a single
4484 		 * interruptible sleep, while user sigblock word is
4485 		 * set, might return EINTR or ERESTART to usermode
4486 		 * without delivering signal.  All further sleeps,
4487 		 * until userspace clears the word and does
4488 		 * sigfastblock(UNBLOCK), observe current word and no
4489 		 * longer get interrupted.  It is slight
4490 		 * non-conformance, with alternative to have read the
4491 		 * sigblock word on each syscall entry.
4492 		 */
4493 		td->td_sigblock_val = 0;
4494 
4495 		/*
4496 		 * Rely on normal ast mechanism to deliver pending
4497 		 * signals to current thread.  But notify others about
4498 		 * fake unblock.
4499 		 */
4500 		sigfastblock_resched(td, error == 0 && p->p_numthreads != 1);
4501 
4502 		break;
4503 
4504 	case SIGFASTBLOCK_UNSETPTR:
4505 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4506 			error = EINVAL;
4507 			break;
4508 		}
4509 		if (!sigfastblock_fetch_sig(td, false, &oldval)) {
4510 			error = EFAULT;
4511 			break;
4512 		}
4513 		if (oldval != 0 && oldval != SIGFASTBLOCK_PEND) {
4514 			error = EBUSY;
4515 			break;
4516 		}
4517 		sigfastblock_clear(td);
4518 		break;
4519 
4520 	default:
4521 		error = EINVAL;
4522 		break;
4523 	}
4524 	return (error);
4525 }
4526 
4527 void
4528 sigfastblock_clear(struct thread *td)
4529 {
4530 	bool resched;
4531 
4532 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4533 		return;
4534 	td->td_sigblock_val = 0;
4535 	resched = (td->td_pflags & TDP_SIGFASTPENDING) != 0 ||
4536 	    SIGPENDING(td);
4537 	td->td_pflags &= ~(TDP_SIGFASTBLOCK | TDP_SIGFASTPENDING);
4538 	sigfastblock_resched(td, resched);
4539 }
4540 
4541 void
4542 sigfastblock_fetch(struct thread *td)
4543 {
4544 	uint32_t val;
4545 
4546 	(void)sigfastblock_fetch_sig(td, true, &val);
4547 }
4548 
4549 static void
4550 sigfastblock_setpend1(struct thread *td)
4551 {
4552 	int res;
4553 	uint32_t oldval;
4554 
4555 	if ((td->td_pflags & TDP_SIGFASTPENDING) == 0)
4556 		return;
4557 	res = fueword32((void *)td->td_sigblock_ptr, &oldval);
4558 	if (res == -1) {
4559 		sigfastblock_failed(td, true, false);
4560 		return;
4561 	}
4562 	for (;;) {
4563 		res = casueword32(td->td_sigblock_ptr, oldval, &oldval,
4564 		    oldval | SIGFASTBLOCK_PEND);
4565 		if (res == -1) {
4566 			sigfastblock_failed(td, true, true);
4567 			return;
4568 		}
4569 		if (res == 0) {
4570 			td->td_sigblock_val = oldval & ~SIGFASTBLOCK_FLAGS;
4571 			td->td_pflags &= ~TDP_SIGFASTPENDING;
4572 			break;
4573 		}
4574 		MPASS(res == 1);
4575 		if (thread_check_susp(td, false) != 0)
4576 			break;
4577 	}
4578 }
4579 
4580 static void
4581 sigfastblock_setpend(struct thread *td, bool resched)
4582 {
4583 	struct proc *p;
4584 
4585 	sigfastblock_setpend1(td);
4586 	if (resched) {
4587 		p = td->td_proc;
4588 		PROC_LOCK(p);
4589 		reschedule_signals(p, fastblock_mask, SIGPROCMASK_FASTBLK);
4590 		PROC_UNLOCK(p);
4591 	}
4592 }
4593