xref: /freebsd/sys/kern/kern_sig.c (revision 3157ba21)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_compat.h"
41 #include "opt_kdtrace.h"
42 #include "opt_ktrace.h"
43 #include "opt_core.h"
44 
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/signalvar.h>
48 #include <sys/vnode.h>
49 #include <sys/acct.h>
50 #include <sys/condvar.h>
51 #include <sys/event.h>
52 #include <sys/fcntl.h>
53 #include <sys/imgact.h>
54 #include <sys/kernel.h>
55 #include <sys/ktr.h>
56 #include <sys/ktrace.h>
57 #include <sys/lock.h>
58 #include <sys/malloc.h>
59 #include <sys/mutex.h>
60 #include <sys/namei.h>
61 #include <sys/proc.h>
62 #include <sys/posix4.h>
63 #include <sys/pioctl.h>
64 #include <sys/resourcevar.h>
65 #include <sys/sdt.h>
66 #include <sys/sbuf.h>
67 #include <sys/sleepqueue.h>
68 #include <sys/smp.h>
69 #include <sys/stat.h>
70 #include <sys/sx.h>
71 #include <sys/syscallsubr.h>
72 #include <sys/sysctl.h>
73 #include <sys/sysent.h>
74 #include <sys/syslog.h>
75 #include <sys/sysproto.h>
76 #include <sys/timers.h>
77 #include <sys/unistd.h>
78 #include <sys/wait.h>
79 #include <vm/vm.h>
80 #include <vm/vm_extern.h>
81 #include <vm/uma.h>
82 
83 #include <sys/jail.h>
84 
85 #include <machine/cpu.h>
86 
87 #include <security/audit/audit.h>
88 
89 #define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
90 
91 SDT_PROVIDER_DECLARE(proc);
92 SDT_PROBE_DEFINE(proc, kernel, , signal_send);
93 SDT_PROBE_ARGTYPE(proc, kernel, , signal_send, 0, "struct thread *");
94 SDT_PROBE_ARGTYPE(proc, kernel, , signal_send, 1, "struct proc *");
95 SDT_PROBE_ARGTYPE(proc, kernel, , signal_send, 2, "int");
96 SDT_PROBE_DEFINE(proc, kernel, , signal_clear);
97 SDT_PROBE_ARGTYPE(proc, kernel, , signal_clear, 0, "int");
98 SDT_PROBE_ARGTYPE(proc, kernel, , signal_clear, 1, "ksiginfo_t *");
99 SDT_PROBE_DEFINE(proc, kernel, , signal_discard);
100 SDT_PROBE_ARGTYPE(proc, kernel, , signal_discard, 0, "struct thread *");
101 SDT_PROBE_ARGTYPE(proc, kernel, , signal_discard, 1, "struct proc *");
102 SDT_PROBE_ARGTYPE(proc, kernel, , signal_discard, 2, "int");
103 
104 static int	coredump(struct thread *);
105 static char	*expand_name(const char *, uid_t, pid_t, struct thread *, int);
106 static int	killpg1(struct thread *td, int sig, int pgid, int all,
107 		    ksiginfo_t *ksi);
108 static int	issignal(struct thread *td, int stop_allowed);
109 static int	sigprop(int sig);
110 static void	tdsigwakeup(struct thread *, int, sig_t, int);
111 static void	sig_suspend_threads(struct thread *, struct proc *, int);
112 static int	filt_sigattach(struct knote *kn);
113 static void	filt_sigdetach(struct knote *kn);
114 static int	filt_signal(struct knote *kn, long hint);
115 static struct thread *sigtd(struct proc *p, int sig, int prop);
116 static void	sigqueue_start(void);
117 
118 static uma_zone_t	ksiginfo_zone = NULL;
119 struct filterops sig_filtops = {
120 	.f_isfd = 0,
121 	.f_attach = filt_sigattach,
122 	.f_detach = filt_sigdetach,
123 	.f_event = filt_signal,
124 };
125 
126 int	kern_logsigexit = 1;
127 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
128     &kern_logsigexit, 0,
129     "Log processes quitting on abnormal signals to syslog(3)");
130 
131 static int	kern_forcesigexit = 1;
132 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
133     &kern_forcesigexit, 0, "Force trap signal to be handled");
134 
135 SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW, 0, "POSIX real time signal");
136 
137 static int	max_pending_per_proc = 128;
138 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
139     &max_pending_per_proc, 0, "Max pending signals per proc");
140 
141 static int	preallocate_siginfo = 1024;
142 TUNABLE_INT("kern.sigqueue.preallocate", &preallocate_siginfo);
143 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RD,
144     &preallocate_siginfo, 0, "Preallocated signal memory size");
145 
146 static int	signal_overflow = 0;
147 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
148     &signal_overflow, 0, "Number of signals overflew");
149 
150 static int	signal_alloc_fail = 0;
151 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
152     &signal_alloc_fail, 0, "signals failed to be allocated");
153 
154 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
155 
156 /*
157  * Policy -- Can ucred cr1 send SIGIO to process cr2?
158  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
159  * in the right situations.
160  */
161 #define CANSIGIO(cr1, cr2) \
162 	((cr1)->cr_uid == 0 || \
163 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
164 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
165 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
166 	    (cr1)->cr_uid == (cr2)->cr_uid)
167 
168 int sugid_coredump;
169 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
170     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
171 
172 static int	do_coredump = 1;
173 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
174 	&do_coredump, 0, "Enable/Disable coredumps");
175 
176 static int	set_core_nodump_flag = 0;
177 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
178 	0, "Enable setting the NODUMP flag on coredump files");
179 
180 /*
181  * Signal properties and actions.
182  * The array below categorizes the signals and their default actions
183  * according to the following properties:
184  */
185 #define	SA_KILL		0x01		/* terminates process by default */
186 #define	SA_CORE		0x02		/* ditto and coredumps */
187 #define	SA_STOP		0x04		/* suspend process */
188 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
189 #define	SA_IGNORE	0x10		/* ignore by default */
190 #define	SA_CONT		0x20		/* continue if suspended */
191 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
192 #define	SA_PROC		0x80		/* deliverable to any thread */
193 
194 static int sigproptbl[NSIG] = {
195         SA_KILL|SA_PROC,		/* SIGHUP */
196         SA_KILL|SA_PROC,		/* SIGINT */
197         SA_KILL|SA_CORE|SA_PROC,	/* SIGQUIT */
198         SA_KILL|SA_CORE,		/* SIGILL */
199         SA_KILL|SA_CORE,		/* SIGTRAP */
200         SA_KILL|SA_CORE,		/* SIGABRT */
201         SA_KILL|SA_CORE|SA_PROC,	/* SIGEMT */
202         SA_KILL|SA_CORE,		/* SIGFPE */
203         SA_KILL|SA_PROC,		/* SIGKILL */
204         SA_KILL|SA_CORE,		/* SIGBUS */
205         SA_KILL|SA_CORE,		/* SIGSEGV */
206         SA_KILL|SA_CORE,		/* SIGSYS */
207         SA_KILL|SA_PROC,		/* SIGPIPE */
208         SA_KILL|SA_PROC,		/* SIGALRM */
209         SA_KILL|SA_PROC,		/* SIGTERM */
210         SA_IGNORE|SA_PROC,		/* SIGURG */
211         SA_STOP|SA_PROC,		/* SIGSTOP */
212         SA_STOP|SA_TTYSTOP|SA_PROC,	/* SIGTSTP */
213         SA_IGNORE|SA_CONT|SA_PROC,	/* SIGCONT */
214         SA_IGNORE|SA_PROC,		/* SIGCHLD */
215         SA_STOP|SA_TTYSTOP|SA_PROC,	/* SIGTTIN */
216         SA_STOP|SA_TTYSTOP|SA_PROC,	/* SIGTTOU */
217         SA_IGNORE|SA_PROC,		/* SIGIO */
218         SA_KILL,			/* SIGXCPU */
219         SA_KILL,			/* SIGXFSZ */
220         SA_KILL|SA_PROC,		/* SIGVTALRM */
221         SA_KILL|SA_PROC,		/* SIGPROF */
222         SA_IGNORE|SA_PROC,		/* SIGWINCH  */
223         SA_IGNORE|SA_PROC,		/* SIGINFO */
224         SA_KILL|SA_PROC,		/* SIGUSR1 */
225         SA_KILL|SA_PROC,		/* SIGUSR2 */
226 };
227 
228 static void reschedule_signals(struct proc *p, sigset_t block, int flags);
229 
230 static void
231 sigqueue_start(void)
232 {
233 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
234 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
235 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
236 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
237 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
238 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
239 }
240 
241 ksiginfo_t *
242 ksiginfo_alloc(int wait)
243 {
244 	int flags;
245 
246 	flags = M_ZERO;
247 	if (! wait)
248 		flags |= M_NOWAIT;
249 	if (ksiginfo_zone != NULL)
250 		return ((ksiginfo_t *)uma_zalloc(ksiginfo_zone, flags));
251 	return (NULL);
252 }
253 
254 void
255 ksiginfo_free(ksiginfo_t *ksi)
256 {
257 	uma_zfree(ksiginfo_zone, ksi);
258 }
259 
260 static __inline int
261 ksiginfo_tryfree(ksiginfo_t *ksi)
262 {
263 	if (!(ksi->ksi_flags & KSI_EXT)) {
264 		uma_zfree(ksiginfo_zone, ksi);
265 		return (1);
266 	}
267 	return (0);
268 }
269 
270 void
271 sigqueue_init(sigqueue_t *list, struct proc *p)
272 {
273 	SIGEMPTYSET(list->sq_signals);
274 	SIGEMPTYSET(list->sq_kill);
275 	TAILQ_INIT(&list->sq_list);
276 	list->sq_proc = p;
277 	list->sq_flags = SQ_INIT;
278 }
279 
280 /*
281  * Get a signal's ksiginfo.
282  * Return:
283  * 	0	-	signal not found
284  *	others	-	signal number
285  */
286 static int
287 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
288 {
289 	struct proc *p = sq->sq_proc;
290 	struct ksiginfo *ksi, *next;
291 	int count = 0;
292 
293 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
294 
295 	if (!SIGISMEMBER(sq->sq_signals, signo))
296 		return (0);
297 
298 	if (SIGISMEMBER(sq->sq_kill, signo)) {
299 		count++;
300 		SIGDELSET(sq->sq_kill, signo);
301 	}
302 
303 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
304 		if (ksi->ksi_signo == signo) {
305 			if (count == 0) {
306 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
307 				ksi->ksi_sigq = NULL;
308 				ksiginfo_copy(ksi, si);
309 				if (ksiginfo_tryfree(ksi) && p != NULL)
310 					p->p_pendingcnt--;
311 			}
312 			if (++count > 1)
313 				break;
314 		}
315 	}
316 
317 	if (count <= 1)
318 		SIGDELSET(sq->sq_signals, signo);
319 	si->ksi_signo = signo;
320 	return (signo);
321 }
322 
323 void
324 sigqueue_take(ksiginfo_t *ksi)
325 {
326 	struct ksiginfo *kp;
327 	struct proc	*p;
328 	sigqueue_t	*sq;
329 
330 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
331 		return;
332 
333 	p = sq->sq_proc;
334 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
335 	ksi->ksi_sigq = NULL;
336 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
337 		p->p_pendingcnt--;
338 
339 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
340 	     kp = TAILQ_NEXT(kp, ksi_link)) {
341 		if (kp->ksi_signo == ksi->ksi_signo)
342 			break;
343 	}
344 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo))
345 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
346 }
347 
348 static int
349 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
350 {
351 	struct proc *p = sq->sq_proc;
352 	struct ksiginfo *ksi;
353 	int ret = 0;
354 
355 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
356 
357 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
358 		SIGADDSET(sq->sq_kill, signo);
359 		goto out_set_bit;
360 	}
361 
362 	/* directly insert the ksi, don't copy it */
363 	if (si->ksi_flags & KSI_INS) {
364 		if (si->ksi_flags & KSI_HEAD)
365 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
366 		else
367 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
368 		si->ksi_sigq = sq;
369 		goto out_set_bit;
370 	}
371 
372 	if (__predict_false(ksiginfo_zone == NULL)) {
373 		SIGADDSET(sq->sq_kill, signo);
374 		goto out_set_bit;
375 	}
376 
377 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
378 		signal_overflow++;
379 		ret = EAGAIN;
380 	} else if ((ksi = ksiginfo_alloc(0)) == NULL) {
381 		signal_alloc_fail++;
382 		ret = EAGAIN;
383 	} else {
384 		if (p != NULL)
385 			p->p_pendingcnt++;
386 		ksiginfo_copy(si, ksi);
387 		ksi->ksi_signo = signo;
388 		if (si->ksi_flags & KSI_HEAD)
389 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
390 		else
391 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
392 		ksi->ksi_sigq = sq;
393 	}
394 
395 	if ((si->ksi_flags & KSI_TRAP) != 0 ||
396 	    (si->ksi_flags & KSI_SIGQ) == 0) {
397 		if (ret != 0)
398 			SIGADDSET(sq->sq_kill, signo);
399 		ret = 0;
400 		goto out_set_bit;
401 	}
402 
403 	if (ret != 0)
404 		return (ret);
405 
406 out_set_bit:
407 	SIGADDSET(sq->sq_signals, signo);
408 	return (ret);
409 }
410 
411 void
412 sigqueue_flush(sigqueue_t *sq)
413 {
414 	struct proc *p = sq->sq_proc;
415 	ksiginfo_t *ksi;
416 
417 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
418 
419 	if (p != NULL)
420 		PROC_LOCK_ASSERT(p, MA_OWNED);
421 
422 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
423 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
424 		ksi->ksi_sigq = NULL;
425 		if (ksiginfo_tryfree(ksi) && p != NULL)
426 			p->p_pendingcnt--;
427 	}
428 
429 	SIGEMPTYSET(sq->sq_signals);
430 	SIGEMPTYSET(sq->sq_kill);
431 }
432 
433 static void
434 sigqueue_collect_set(sigqueue_t *sq, sigset_t *set)
435 {
436 	ksiginfo_t *ksi;
437 
438 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
439 
440 	TAILQ_FOREACH(ksi, &sq->sq_list, ksi_link)
441 		SIGADDSET(*set, ksi->ksi_signo);
442 	SIGSETOR(*set, sq->sq_kill);
443 }
444 
445 static void
446 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, sigset_t *setp)
447 {
448 	sigset_t tmp, set;
449 	struct proc *p1, *p2;
450 	ksiginfo_t *ksi, *next;
451 
452 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
453 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
454 	/*
455 	 * make a copy, this allows setp to point to src or dst
456 	 * sq_signals without trouble.
457 	 */
458 	set = *setp;
459 	p1 = src->sq_proc;
460 	p2 = dst->sq_proc;
461 	/* Move siginfo to target list */
462 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
463 		if (SIGISMEMBER(set, ksi->ksi_signo)) {
464 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
465 			if (p1 != NULL)
466 				p1->p_pendingcnt--;
467 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
468 			ksi->ksi_sigq = dst;
469 			if (p2 != NULL)
470 				p2->p_pendingcnt++;
471 		}
472 	}
473 
474 	/* Move pending bits to target list */
475 	tmp = src->sq_kill;
476 	SIGSETAND(tmp, set);
477 	SIGSETOR(dst->sq_kill, tmp);
478 	SIGSETNAND(src->sq_kill, tmp);
479 
480 	tmp = src->sq_signals;
481 	SIGSETAND(tmp, set);
482 	SIGSETOR(dst->sq_signals, tmp);
483 	SIGSETNAND(src->sq_signals, tmp);
484 
485 	/* Finally, rescan src queue and set pending bits for it */
486 	sigqueue_collect_set(src, &src->sq_signals);
487 }
488 
489 static void
490 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
491 {
492 	sigset_t set;
493 
494 	SIGEMPTYSET(set);
495 	SIGADDSET(set, signo);
496 	sigqueue_move_set(src, dst, &set);
497 }
498 
499 static void
500 sigqueue_delete_set(sigqueue_t *sq, sigset_t *set)
501 {
502 	struct proc *p = sq->sq_proc;
503 	ksiginfo_t *ksi, *next;
504 
505 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
506 
507 	/* Remove siginfo queue */
508 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
509 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
510 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
511 			ksi->ksi_sigq = NULL;
512 			if (ksiginfo_tryfree(ksi) && p != NULL)
513 				p->p_pendingcnt--;
514 		}
515 	}
516 	SIGSETNAND(sq->sq_kill, *set);
517 	SIGSETNAND(sq->sq_signals, *set);
518 	/* Finally, rescan queue and set pending bits for it */
519 	sigqueue_collect_set(sq, &sq->sq_signals);
520 }
521 
522 void
523 sigqueue_delete(sigqueue_t *sq, int signo)
524 {
525 	sigset_t set;
526 
527 	SIGEMPTYSET(set);
528 	SIGADDSET(set, signo);
529 	sigqueue_delete_set(sq, &set);
530 }
531 
532 /* Remove a set of signals for a process */
533 static void
534 sigqueue_delete_set_proc(struct proc *p, sigset_t *set)
535 {
536 	sigqueue_t worklist;
537 	struct thread *td0;
538 
539 	PROC_LOCK_ASSERT(p, MA_OWNED);
540 
541 	sigqueue_init(&worklist, NULL);
542 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
543 
544 	FOREACH_THREAD_IN_PROC(p, td0)
545 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
546 
547 	sigqueue_flush(&worklist);
548 }
549 
550 void
551 sigqueue_delete_proc(struct proc *p, int signo)
552 {
553 	sigset_t set;
554 
555 	SIGEMPTYSET(set);
556 	SIGADDSET(set, signo);
557 	sigqueue_delete_set_proc(p, &set);
558 }
559 
560 static void
561 sigqueue_delete_stopmask_proc(struct proc *p)
562 {
563 	sigset_t set;
564 
565 	SIGEMPTYSET(set);
566 	SIGADDSET(set, SIGSTOP);
567 	SIGADDSET(set, SIGTSTP);
568 	SIGADDSET(set, SIGTTIN);
569 	SIGADDSET(set, SIGTTOU);
570 	sigqueue_delete_set_proc(p, &set);
571 }
572 
573 /*
574  * Determine signal that should be delivered to process p, the current
575  * process, 0 if none.  If there is a pending stop signal with default
576  * action, the process stops in issignal().
577  */
578 int
579 cursig(struct thread *td, int stop_allowed)
580 {
581 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
582 	KASSERT(stop_allowed == SIG_STOP_ALLOWED ||
583 	    stop_allowed == SIG_STOP_NOT_ALLOWED, ("cursig: stop_allowed"));
584 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
585 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
586 	return (SIGPENDING(td) ? issignal(td, stop_allowed) : 0);
587 }
588 
589 /*
590  * Arrange for ast() to handle unmasked pending signals on return to user
591  * mode.  This must be called whenever a signal is added to td_sigqueue or
592  * unmasked in td_sigmask.
593  */
594 void
595 signotify(struct thread *td)
596 {
597 	struct proc *p;
598 
599 	p = td->td_proc;
600 
601 	PROC_LOCK_ASSERT(p, MA_OWNED);
602 
603 	if (SIGPENDING(td)) {
604 		thread_lock(td);
605 		td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING;
606 		thread_unlock(td);
607 	}
608 }
609 
610 int
611 sigonstack(size_t sp)
612 {
613 	struct thread *td = curthread;
614 
615 	return ((td->td_pflags & TDP_ALTSTACK) ?
616 #if defined(COMPAT_43)
617 	    ((td->td_sigstk.ss_size == 0) ?
618 		(td->td_sigstk.ss_flags & SS_ONSTACK) :
619 		((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size))
620 #else
621 	    ((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size)
622 #endif
623 	    : 0);
624 }
625 
626 static __inline int
627 sigprop(int sig)
628 {
629 
630 	if (sig > 0 && sig < NSIG)
631 		return (sigproptbl[_SIG_IDX(sig)]);
632 	return (0);
633 }
634 
635 int
636 sig_ffs(sigset_t *set)
637 {
638 	int i;
639 
640 	for (i = 0; i < _SIG_WORDS; i++)
641 		if (set->__bits[i])
642 			return (ffs(set->__bits[i]) + (i * 32));
643 	return (0);
644 }
645 
646 /*
647  * kern_sigaction
648  * sigaction
649  * freebsd4_sigaction
650  * osigaction
651  */
652 int
653 kern_sigaction(td, sig, act, oact, flags)
654 	struct thread *td;
655 	register int sig;
656 	struct sigaction *act, *oact;
657 	int flags;
658 {
659 	struct sigacts *ps;
660 	struct proc *p = td->td_proc;
661 
662 	if (!_SIG_VALID(sig))
663 		return (EINVAL);
664 
665 	PROC_LOCK(p);
666 	ps = p->p_sigacts;
667 	mtx_lock(&ps->ps_mtx);
668 	if (oact) {
669 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
670 		oact->sa_flags = 0;
671 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
672 			oact->sa_flags |= SA_ONSTACK;
673 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
674 			oact->sa_flags |= SA_RESTART;
675 		if (SIGISMEMBER(ps->ps_sigreset, sig))
676 			oact->sa_flags |= SA_RESETHAND;
677 		if (SIGISMEMBER(ps->ps_signodefer, sig))
678 			oact->sa_flags |= SA_NODEFER;
679 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
680 			oact->sa_flags |= SA_SIGINFO;
681 			oact->sa_sigaction =
682 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
683 		} else
684 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
685 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
686 			oact->sa_flags |= SA_NOCLDSTOP;
687 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
688 			oact->sa_flags |= SA_NOCLDWAIT;
689 	}
690 	if (act) {
691 		if ((sig == SIGKILL || sig == SIGSTOP) &&
692 		    act->sa_handler != SIG_DFL) {
693 			mtx_unlock(&ps->ps_mtx);
694 			PROC_UNLOCK(p);
695 			return (EINVAL);
696 		}
697 
698 		/*
699 		 * Change setting atomically.
700 		 */
701 
702 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
703 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
704 		if (act->sa_flags & SA_SIGINFO) {
705 			ps->ps_sigact[_SIG_IDX(sig)] =
706 			    (__sighandler_t *)act->sa_sigaction;
707 			SIGADDSET(ps->ps_siginfo, sig);
708 		} else {
709 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
710 			SIGDELSET(ps->ps_siginfo, sig);
711 		}
712 		if (!(act->sa_flags & SA_RESTART))
713 			SIGADDSET(ps->ps_sigintr, sig);
714 		else
715 			SIGDELSET(ps->ps_sigintr, sig);
716 		if (act->sa_flags & SA_ONSTACK)
717 			SIGADDSET(ps->ps_sigonstack, sig);
718 		else
719 			SIGDELSET(ps->ps_sigonstack, sig);
720 		if (act->sa_flags & SA_RESETHAND)
721 			SIGADDSET(ps->ps_sigreset, sig);
722 		else
723 			SIGDELSET(ps->ps_sigreset, sig);
724 		if (act->sa_flags & SA_NODEFER)
725 			SIGADDSET(ps->ps_signodefer, sig);
726 		else
727 			SIGDELSET(ps->ps_signodefer, sig);
728 		if (sig == SIGCHLD) {
729 			if (act->sa_flags & SA_NOCLDSTOP)
730 				ps->ps_flag |= PS_NOCLDSTOP;
731 			else
732 				ps->ps_flag &= ~PS_NOCLDSTOP;
733 			if (act->sa_flags & SA_NOCLDWAIT) {
734 				/*
735 				 * Paranoia: since SA_NOCLDWAIT is implemented
736 				 * by reparenting the dying child to PID 1 (and
737 				 * trust it to reap the zombie), PID 1 itself
738 				 * is forbidden to set SA_NOCLDWAIT.
739 				 */
740 				if (p->p_pid == 1)
741 					ps->ps_flag &= ~PS_NOCLDWAIT;
742 				else
743 					ps->ps_flag |= PS_NOCLDWAIT;
744 			} else
745 				ps->ps_flag &= ~PS_NOCLDWAIT;
746 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
747 				ps->ps_flag |= PS_CLDSIGIGN;
748 			else
749 				ps->ps_flag &= ~PS_CLDSIGIGN;
750 		}
751 		/*
752 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
753 		 * and for signals set to SIG_DFL where the default is to
754 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
755 		 * have to restart the process.
756 		 */
757 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
758 		    (sigprop(sig) & SA_IGNORE &&
759 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
760 			/* never to be seen again */
761 			sigqueue_delete_proc(p, sig);
762 			if (sig != SIGCONT)
763 				/* easier in psignal */
764 				SIGADDSET(ps->ps_sigignore, sig);
765 			SIGDELSET(ps->ps_sigcatch, sig);
766 		} else {
767 			SIGDELSET(ps->ps_sigignore, sig);
768 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
769 				SIGDELSET(ps->ps_sigcatch, sig);
770 			else
771 				SIGADDSET(ps->ps_sigcatch, sig);
772 		}
773 #ifdef COMPAT_FREEBSD4
774 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
775 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
776 		    (flags & KSA_FREEBSD4) == 0)
777 			SIGDELSET(ps->ps_freebsd4, sig);
778 		else
779 			SIGADDSET(ps->ps_freebsd4, sig);
780 #endif
781 #ifdef COMPAT_43
782 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
783 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
784 		    (flags & KSA_OSIGSET) == 0)
785 			SIGDELSET(ps->ps_osigset, sig);
786 		else
787 			SIGADDSET(ps->ps_osigset, sig);
788 #endif
789 	}
790 	mtx_unlock(&ps->ps_mtx);
791 	PROC_UNLOCK(p);
792 	return (0);
793 }
794 
795 #ifndef _SYS_SYSPROTO_H_
796 struct sigaction_args {
797 	int	sig;
798 	struct	sigaction *act;
799 	struct	sigaction *oact;
800 };
801 #endif
802 int
803 sigaction(td, uap)
804 	struct thread *td;
805 	register struct sigaction_args *uap;
806 {
807 	struct sigaction act, oact;
808 	register struct sigaction *actp, *oactp;
809 	int error;
810 
811 	actp = (uap->act != NULL) ? &act : NULL;
812 	oactp = (uap->oact != NULL) ? &oact : NULL;
813 	if (actp) {
814 		error = copyin(uap->act, actp, sizeof(act));
815 		if (error)
816 			return (error);
817 	}
818 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
819 	if (oactp && !error)
820 		error = copyout(oactp, uap->oact, sizeof(oact));
821 	return (error);
822 }
823 
824 #ifdef COMPAT_FREEBSD4
825 #ifndef _SYS_SYSPROTO_H_
826 struct freebsd4_sigaction_args {
827 	int	sig;
828 	struct	sigaction *act;
829 	struct	sigaction *oact;
830 };
831 #endif
832 int
833 freebsd4_sigaction(td, uap)
834 	struct thread *td;
835 	register struct freebsd4_sigaction_args *uap;
836 {
837 	struct sigaction act, oact;
838 	register struct sigaction *actp, *oactp;
839 	int error;
840 
841 
842 	actp = (uap->act != NULL) ? &act : NULL;
843 	oactp = (uap->oact != NULL) ? &oact : NULL;
844 	if (actp) {
845 		error = copyin(uap->act, actp, sizeof(act));
846 		if (error)
847 			return (error);
848 	}
849 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
850 	if (oactp && !error)
851 		error = copyout(oactp, uap->oact, sizeof(oact));
852 	return (error);
853 }
854 #endif	/* COMAPT_FREEBSD4 */
855 
856 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
857 #ifndef _SYS_SYSPROTO_H_
858 struct osigaction_args {
859 	int	signum;
860 	struct	osigaction *nsa;
861 	struct	osigaction *osa;
862 };
863 #endif
864 int
865 osigaction(td, uap)
866 	struct thread *td;
867 	register struct osigaction_args *uap;
868 {
869 	struct osigaction sa;
870 	struct sigaction nsa, osa;
871 	register struct sigaction *nsap, *osap;
872 	int error;
873 
874 	if (uap->signum <= 0 || uap->signum >= ONSIG)
875 		return (EINVAL);
876 
877 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
878 	osap = (uap->osa != NULL) ? &osa : NULL;
879 
880 	if (nsap) {
881 		error = copyin(uap->nsa, &sa, sizeof(sa));
882 		if (error)
883 			return (error);
884 		nsap->sa_handler = sa.sa_handler;
885 		nsap->sa_flags = sa.sa_flags;
886 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
887 	}
888 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
889 	if (osap && !error) {
890 		sa.sa_handler = osap->sa_handler;
891 		sa.sa_flags = osap->sa_flags;
892 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
893 		error = copyout(&sa, uap->osa, sizeof(sa));
894 	}
895 	return (error);
896 }
897 
898 #if !defined(__i386__)
899 /* Avoid replicating the same stub everywhere */
900 int
901 osigreturn(td, uap)
902 	struct thread *td;
903 	struct osigreturn_args *uap;
904 {
905 
906 	return (nosys(td, (struct nosys_args *)uap));
907 }
908 #endif
909 #endif /* COMPAT_43 */
910 
911 /*
912  * Initialize signal state for process 0;
913  * set to ignore signals that are ignored by default.
914  */
915 void
916 siginit(p)
917 	struct proc *p;
918 {
919 	register int i;
920 	struct sigacts *ps;
921 
922 	PROC_LOCK(p);
923 	ps = p->p_sigacts;
924 	mtx_lock(&ps->ps_mtx);
925 	for (i = 1; i <= NSIG; i++)
926 		if (sigprop(i) & SA_IGNORE && i != SIGCONT)
927 			SIGADDSET(ps->ps_sigignore, i);
928 	mtx_unlock(&ps->ps_mtx);
929 	PROC_UNLOCK(p);
930 }
931 
932 /*
933  * Reset signals for an exec of the specified process.
934  */
935 void
936 execsigs(struct proc *p)
937 {
938 	struct sigacts *ps;
939 	int sig;
940 	struct thread *td;
941 
942 	/*
943 	 * Reset caught signals.  Held signals remain held
944 	 * through td_sigmask (unless they were caught,
945 	 * and are now ignored by default).
946 	 */
947 	PROC_LOCK_ASSERT(p, MA_OWNED);
948 	td = FIRST_THREAD_IN_PROC(p);
949 	ps = p->p_sigacts;
950 	mtx_lock(&ps->ps_mtx);
951 	while (SIGNOTEMPTY(ps->ps_sigcatch)) {
952 		sig = sig_ffs(&ps->ps_sigcatch);
953 		SIGDELSET(ps->ps_sigcatch, sig);
954 		if (sigprop(sig) & SA_IGNORE) {
955 			if (sig != SIGCONT)
956 				SIGADDSET(ps->ps_sigignore, sig);
957 			sigqueue_delete_proc(p, sig);
958 		}
959 		ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
960 	}
961 	/*
962 	 * Reset stack state to the user stack.
963 	 * Clear set of signals caught on the signal stack.
964 	 */
965 	td->td_sigstk.ss_flags = SS_DISABLE;
966 	td->td_sigstk.ss_size = 0;
967 	td->td_sigstk.ss_sp = 0;
968 	td->td_pflags &= ~TDP_ALTSTACK;
969 	/*
970 	 * Reset no zombies if child dies flag as Solaris does.
971 	 */
972 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
973 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
974 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
975 	mtx_unlock(&ps->ps_mtx);
976 }
977 
978 /*
979  * kern_sigprocmask()
980  *
981  *	Manipulate signal mask.
982  */
983 int
984 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
985     int flags)
986 {
987 	sigset_t new_block, oset1;
988 	struct proc *p;
989 	int error;
990 
991 	p = td->td_proc;
992 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
993 		PROC_LOCK(p);
994 	if (oset != NULL)
995 		*oset = td->td_sigmask;
996 
997 	error = 0;
998 	SIGEMPTYSET(new_block);
999 	if (set != NULL) {
1000 		switch (how) {
1001 		case SIG_BLOCK:
1002 			SIG_CANTMASK(*set);
1003 			oset1 = td->td_sigmask;
1004 			SIGSETOR(td->td_sigmask, *set);
1005 			new_block = td->td_sigmask;
1006 			SIGSETNAND(new_block, oset1);
1007 			break;
1008 		case SIG_UNBLOCK:
1009 			SIGSETNAND(td->td_sigmask, *set);
1010 			signotify(td);
1011 			break;
1012 		case SIG_SETMASK:
1013 			SIG_CANTMASK(*set);
1014 			oset1 = td->td_sigmask;
1015 			if (flags & SIGPROCMASK_OLD)
1016 				SIGSETLO(td->td_sigmask, *set);
1017 			else
1018 				td->td_sigmask = *set;
1019 			new_block = td->td_sigmask;
1020 			SIGSETNAND(new_block, oset1);
1021 			signotify(td);
1022 			break;
1023 		default:
1024 			error = EINVAL;
1025 			break;
1026 		}
1027 	}
1028 
1029 	/*
1030 	 * The new_block set contains signals that were not previously
1031 	 * blocked, but are blocked now.
1032 	 *
1033 	 * In case we block any signal that was not previously blocked
1034 	 * for td, and process has the signal pending, try to schedule
1035 	 * signal delivery to some thread that does not block the signal,
1036 	 * possibly waking it up.
1037 	 */
1038 	if (p->p_numthreads != 1)
1039 		reschedule_signals(p, new_block, flags);
1040 
1041 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1042 		PROC_UNLOCK(p);
1043 	return (error);
1044 }
1045 
1046 #ifndef _SYS_SYSPROTO_H_
1047 struct sigprocmask_args {
1048 	int	how;
1049 	const sigset_t *set;
1050 	sigset_t *oset;
1051 };
1052 #endif
1053 int
1054 sigprocmask(td, uap)
1055 	register struct thread *td;
1056 	struct sigprocmask_args *uap;
1057 {
1058 	sigset_t set, oset;
1059 	sigset_t *setp, *osetp;
1060 	int error;
1061 
1062 	setp = (uap->set != NULL) ? &set : NULL;
1063 	osetp = (uap->oset != NULL) ? &oset : NULL;
1064 	if (setp) {
1065 		error = copyin(uap->set, setp, sizeof(set));
1066 		if (error)
1067 			return (error);
1068 	}
1069 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1070 	if (osetp && !error) {
1071 		error = copyout(osetp, uap->oset, sizeof(oset));
1072 	}
1073 	return (error);
1074 }
1075 
1076 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1077 #ifndef _SYS_SYSPROTO_H_
1078 struct osigprocmask_args {
1079 	int	how;
1080 	osigset_t mask;
1081 };
1082 #endif
1083 int
1084 osigprocmask(td, uap)
1085 	register struct thread *td;
1086 	struct osigprocmask_args *uap;
1087 {
1088 	sigset_t set, oset;
1089 	int error;
1090 
1091 	OSIG2SIG(uap->mask, set);
1092 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1093 	SIG2OSIG(oset, td->td_retval[0]);
1094 	return (error);
1095 }
1096 #endif /* COMPAT_43 */
1097 
1098 int
1099 sigwait(struct thread *td, struct sigwait_args *uap)
1100 {
1101 	ksiginfo_t ksi;
1102 	sigset_t set;
1103 	int error;
1104 
1105 	error = copyin(uap->set, &set, sizeof(set));
1106 	if (error) {
1107 		td->td_retval[0] = error;
1108 		return (0);
1109 	}
1110 
1111 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1112 	if (error) {
1113 		if (error == ERESTART)
1114 			return (error);
1115 		td->td_retval[0] = error;
1116 		return (0);
1117 	}
1118 
1119 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1120 	td->td_retval[0] = error;
1121 	return (0);
1122 }
1123 
1124 int
1125 sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1126 {
1127 	struct timespec ts;
1128 	struct timespec *timeout;
1129 	sigset_t set;
1130 	ksiginfo_t ksi;
1131 	int error;
1132 
1133 	if (uap->timeout) {
1134 		error = copyin(uap->timeout, &ts, sizeof(ts));
1135 		if (error)
1136 			return (error);
1137 
1138 		timeout = &ts;
1139 	} else
1140 		timeout = NULL;
1141 
1142 	error = copyin(uap->set, &set, sizeof(set));
1143 	if (error)
1144 		return (error);
1145 
1146 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1147 	if (error)
1148 		return (error);
1149 
1150 	if (uap->info)
1151 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1152 
1153 	if (error == 0)
1154 		td->td_retval[0] = ksi.ksi_signo;
1155 	return (error);
1156 }
1157 
1158 int
1159 sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1160 {
1161 	ksiginfo_t ksi;
1162 	sigset_t set;
1163 	int error;
1164 
1165 	error = copyin(uap->set, &set, sizeof(set));
1166 	if (error)
1167 		return (error);
1168 
1169 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1170 	if (error)
1171 		return (error);
1172 
1173 	if (uap->info)
1174 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1175 
1176 	if (error == 0)
1177 		td->td_retval[0] = ksi.ksi_signo;
1178 	return (error);
1179 }
1180 
1181 int
1182 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1183 	struct timespec *timeout)
1184 {
1185 	struct sigacts *ps;
1186 	sigset_t savedmask;
1187 	struct proc *p;
1188 	int error, sig, hz, i, timevalid = 0;
1189 	struct timespec rts, ets, ts;
1190 	struct timeval tv;
1191 
1192 	p = td->td_proc;
1193 	error = 0;
1194 	sig = 0;
1195 	ets.tv_sec = 0;
1196 	ets.tv_nsec = 0;
1197 	SIG_CANTMASK(waitset);
1198 
1199 	PROC_LOCK(p);
1200 	ps = p->p_sigacts;
1201 	savedmask = td->td_sigmask;
1202 	if (timeout) {
1203 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1204 			timevalid = 1;
1205 			getnanouptime(&rts);
1206 		 	ets = rts;
1207 			timespecadd(&ets, timeout);
1208 		}
1209 	}
1210 
1211 restart:
1212 	for (i = 1; i <= _SIG_MAXSIG; ++i) {
1213 		if (!SIGISMEMBER(waitset, i))
1214 			continue;
1215 		if (!SIGISMEMBER(td->td_sigqueue.sq_signals, i)) {
1216 			if (SIGISMEMBER(p->p_sigqueue.sq_signals, i)) {
1217 				sigqueue_move(&p->p_sigqueue,
1218 					&td->td_sigqueue, i);
1219 			} else
1220 				continue;
1221 		}
1222 
1223 		SIGFILLSET(td->td_sigmask);
1224 		SIG_CANTMASK(td->td_sigmask);
1225 		SIGDELSET(td->td_sigmask, i);
1226 		mtx_lock(&ps->ps_mtx);
1227 		sig = cursig(td, SIG_STOP_ALLOWED);
1228 		mtx_unlock(&ps->ps_mtx);
1229 		if (sig)
1230 			goto out;
1231 		else {
1232 			/*
1233 			 * Because cursig() may have stopped current thread,
1234 			 * after it is resumed, things may have already been
1235 			 * changed, it should rescan any pending signals.
1236 			 */
1237 			goto restart;
1238 		}
1239 	}
1240 
1241 	if (error)
1242 		goto out;
1243 
1244 	/*
1245 	 * POSIX says this must be checked after looking for pending
1246 	 * signals.
1247 	 */
1248 	if (timeout) {
1249 		if (!timevalid) {
1250 			error = EINVAL;
1251 			goto out;
1252 		}
1253 		getnanouptime(&rts);
1254 		if (timespeccmp(&rts, &ets, >=)) {
1255 			error = EAGAIN;
1256 			goto out;
1257 		}
1258 		ts = ets;
1259 		timespecsub(&ts, &rts);
1260 		TIMESPEC_TO_TIMEVAL(&tv, &ts);
1261 		hz = tvtohz(&tv);
1262 	} else
1263 		hz = 0;
1264 
1265 	td->td_sigmask = savedmask;
1266 	SIGSETNAND(td->td_sigmask, waitset);
1267 	signotify(td);
1268 	error = msleep(&ps, &p->p_mtx, PPAUSE|PCATCH, "sigwait", hz);
1269 	if (timeout) {
1270 		if (error == ERESTART) {
1271 			/* timeout can not be restarted. */
1272 			error = EINTR;
1273 		} else if (error == EAGAIN) {
1274 			/* will calculate timeout by ourself. */
1275 			error = 0;
1276 		}
1277 	}
1278 	goto restart;
1279 
1280 out:
1281 	td->td_sigmask = savedmask;
1282 	signotify(td);
1283 	if (sig) {
1284 		ksiginfo_init(ksi);
1285 		sigqueue_get(&td->td_sigqueue, sig, ksi);
1286 		ksi->ksi_signo = sig;
1287 
1288 		SDT_PROBE(proc, kernel, , signal_clear, sig, ksi, 0, 0, 0);
1289 
1290 		if (ksi->ksi_code == SI_TIMER)
1291 			itimer_accept(p, ksi->ksi_timerid, ksi);
1292 		error = 0;
1293 
1294 #ifdef KTRACE
1295 		if (KTRPOINT(td, KTR_PSIG)) {
1296 			sig_t action;
1297 
1298 			mtx_lock(&ps->ps_mtx);
1299 			action = ps->ps_sigact[_SIG_IDX(sig)];
1300 			mtx_unlock(&ps->ps_mtx);
1301 			ktrpsig(sig, action, &td->td_sigmask, 0);
1302 		}
1303 #endif
1304 		if (sig == SIGKILL)
1305 			sigexit(td, sig);
1306 	}
1307 	PROC_UNLOCK(p);
1308 	return (error);
1309 }
1310 
1311 #ifndef _SYS_SYSPROTO_H_
1312 struct sigpending_args {
1313 	sigset_t	*set;
1314 };
1315 #endif
1316 int
1317 sigpending(td, uap)
1318 	struct thread *td;
1319 	struct sigpending_args *uap;
1320 {
1321 	struct proc *p = td->td_proc;
1322 	sigset_t pending;
1323 
1324 	PROC_LOCK(p);
1325 	pending = p->p_sigqueue.sq_signals;
1326 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1327 	PROC_UNLOCK(p);
1328 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1329 }
1330 
1331 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1332 #ifndef _SYS_SYSPROTO_H_
1333 struct osigpending_args {
1334 	int	dummy;
1335 };
1336 #endif
1337 int
1338 osigpending(td, uap)
1339 	struct thread *td;
1340 	struct osigpending_args *uap;
1341 {
1342 	struct proc *p = td->td_proc;
1343 	sigset_t pending;
1344 
1345 	PROC_LOCK(p);
1346 	pending = p->p_sigqueue.sq_signals;
1347 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1348 	PROC_UNLOCK(p);
1349 	SIG2OSIG(pending, td->td_retval[0]);
1350 	return (0);
1351 }
1352 #endif /* COMPAT_43 */
1353 
1354 #if defined(COMPAT_43)
1355 /*
1356  * Generalized interface signal handler, 4.3-compatible.
1357  */
1358 #ifndef _SYS_SYSPROTO_H_
1359 struct osigvec_args {
1360 	int	signum;
1361 	struct	sigvec *nsv;
1362 	struct	sigvec *osv;
1363 };
1364 #endif
1365 /* ARGSUSED */
1366 int
1367 osigvec(td, uap)
1368 	struct thread *td;
1369 	register struct osigvec_args *uap;
1370 {
1371 	struct sigvec vec;
1372 	struct sigaction nsa, osa;
1373 	register struct sigaction *nsap, *osap;
1374 	int error;
1375 
1376 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1377 		return (EINVAL);
1378 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1379 	osap = (uap->osv != NULL) ? &osa : NULL;
1380 	if (nsap) {
1381 		error = copyin(uap->nsv, &vec, sizeof(vec));
1382 		if (error)
1383 			return (error);
1384 		nsap->sa_handler = vec.sv_handler;
1385 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1386 		nsap->sa_flags = vec.sv_flags;
1387 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1388 	}
1389 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1390 	if (osap && !error) {
1391 		vec.sv_handler = osap->sa_handler;
1392 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1393 		vec.sv_flags = osap->sa_flags;
1394 		vec.sv_flags &= ~SA_NOCLDWAIT;
1395 		vec.sv_flags ^= SA_RESTART;
1396 		error = copyout(&vec, uap->osv, sizeof(vec));
1397 	}
1398 	return (error);
1399 }
1400 
1401 #ifndef _SYS_SYSPROTO_H_
1402 struct osigblock_args {
1403 	int	mask;
1404 };
1405 #endif
1406 int
1407 osigblock(td, uap)
1408 	register struct thread *td;
1409 	struct osigblock_args *uap;
1410 {
1411 	sigset_t set, oset;
1412 
1413 	OSIG2SIG(uap->mask, set);
1414 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1415 	SIG2OSIG(oset, td->td_retval[0]);
1416 	return (0);
1417 }
1418 
1419 #ifndef _SYS_SYSPROTO_H_
1420 struct osigsetmask_args {
1421 	int	mask;
1422 };
1423 #endif
1424 int
1425 osigsetmask(td, uap)
1426 	struct thread *td;
1427 	struct osigsetmask_args *uap;
1428 {
1429 	sigset_t set, oset;
1430 
1431 	OSIG2SIG(uap->mask, set);
1432 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1433 	SIG2OSIG(oset, td->td_retval[0]);
1434 	return (0);
1435 }
1436 #endif /* COMPAT_43 */
1437 
1438 /*
1439  * Suspend calling thread until signal, providing mask to be set in the
1440  * meantime.
1441  */
1442 #ifndef _SYS_SYSPROTO_H_
1443 struct sigsuspend_args {
1444 	const sigset_t *sigmask;
1445 };
1446 #endif
1447 /* ARGSUSED */
1448 int
1449 sigsuspend(td, uap)
1450 	struct thread *td;
1451 	struct sigsuspend_args *uap;
1452 {
1453 	sigset_t mask;
1454 	int error;
1455 
1456 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1457 	if (error)
1458 		return (error);
1459 	return (kern_sigsuspend(td, mask));
1460 }
1461 
1462 int
1463 kern_sigsuspend(struct thread *td, sigset_t mask)
1464 {
1465 	struct proc *p = td->td_proc;
1466 	int has_sig, sig;
1467 
1468 	/*
1469 	 * When returning from sigsuspend, we want
1470 	 * the old mask to be restored after the
1471 	 * signal handler has finished.  Thus, we
1472 	 * save it here and mark the sigacts structure
1473 	 * to indicate this.
1474 	 */
1475 	PROC_LOCK(p);
1476 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1477 	    SIGPROCMASK_PROC_LOCKED);
1478 	td->td_pflags |= TDP_OLDMASK;
1479 
1480 	/*
1481 	 * Process signals now. Otherwise, we can get spurious wakeup
1482 	 * due to signal entered process queue, but delivered to other
1483 	 * thread. But sigsuspend should return only on signal
1484 	 * delivery.
1485 	 */
1486 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1487 	for (has_sig = 0; !has_sig;) {
1488 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1489 			0) == 0)
1490 			/* void */;
1491 		thread_suspend_check(0);
1492 		mtx_lock(&p->p_sigacts->ps_mtx);
1493 		while ((sig = cursig(td, SIG_STOP_ALLOWED)) != 0)
1494 			has_sig += postsig(sig);
1495 		mtx_unlock(&p->p_sigacts->ps_mtx);
1496 	}
1497 	PROC_UNLOCK(p);
1498 	return (EJUSTRETURN);
1499 }
1500 
1501 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1502 /*
1503  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1504  * convention: libc stub passes mask, not pointer, to save a copyin.
1505  */
1506 #ifndef _SYS_SYSPROTO_H_
1507 struct osigsuspend_args {
1508 	osigset_t mask;
1509 };
1510 #endif
1511 /* ARGSUSED */
1512 int
1513 osigsuspend(td, uap)
1514 	struct thread *td;
1515 	struct osigsuspend_args *uap;
1516 {
1517 	sigset_t mask;
1518 
1519 	OSIG2SIG(uap->mask, mask);
1520 	return (kern_sigsuspend(td, mask));
1521 }
1522 #endif /* COMPAT_43 */
1523 
1524 #if defined(COMPAT_43)
1525 #ifndef _SYS_SYSPROTO_H_
1526 struct osigstack_args {
1527 	struct	sigstack *nss;
1528 	struct	sigstack *oss;
1529 };
1530 #endif
1531 /* ARGSUSED */
1532 int
1533 osigstack(td, uap)
1534 	struct thread *td;
1535 	register struct osigstack_args *uap;
1536 {
1537 	struct sigstack nss, oss;
1538 	int error = 0;
1539 
1540 	if (uap->nss != NULL) {
1541 		error = copyin(uap->nss, &nss, sizeof(nss));
1542 		if (error)
1543 			return (error);
1544 	}
1545 	oss.ss_sp = td->td_sigstk.ss_sp;
1546 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1547 	if (uap->nss != NULL) {
1548 		td->td_sigstk.ss_sp = nss.ss_sp;
1549 		td->td_sigstk.ss_size = 0;
1550 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1551 		td->td_pflags |= TDP_ALTSTACK;
1552 	}
1553 	if (uap->oss != NULL)
1554 		error = copyout(&oss, uap->oss, sizeof(oss));
1555 
1556 	return (error);
1557 }
1558 #endif /* COMPAT_43 */
1559 
1560 #ifndef _SYS_SYSPROTO_H_
1561 struct sigaltstack_args {
1562 	stack_t	*ss;
1563 	stack_t	*oss;
1564 };
1565 #endif
1566 /* ARGSUSED */
1567 int
1568 sigaltstack(td, uap)
1569 	struct thread *td;
1570 	register struct sigaltstack_args *uap;
1571 {
1572 	stack_t ss, oss;
1573 	int error;
1574 
1575 	if (uap->ss != NULL) {
1576 		error = copyin(uap->ss, &ss, sizeof(ss));
1577 		if (error)
1578 			return (error);
1579 	}
1580 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1581 	    (uap->oss != NULL) ? &oss : NULL);
1582 	if (error)
1583 		return (error);
1584 	if (uap->oss != NULL)
1585 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1586 	return (error);
1587 }
1588 
1589 int
1590 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1591 {
1592 	struct proc *p = td->td_proc;
1593 	int oonstack;
1594 
1595 	oonstack = sigonstack(cpu_getstack(td));
1596 
1597 	if (oss != NULL) {
1598 		*oss = td->td_sigstk;
1599 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1600 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1601 	}
1602 
1603 	if (ss != NULL) {
1604 		if (oonstack)
1605 			return (EPERM);
1606 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1607 			return (EINVAL);
1608 		if (!(ss->ss_flags & SS_DISABLE)) {
1609 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1610 				return (ENOMEM);
1611 
1612 			td->td_sigstk = *ss;
1613 			td->td_pflags |= TDP_ALTSTACK;
1614 		} else {
1615 			td->td_pflags &= ~TDP_ALTSTACK;
1616 		}
1617 	}
1618 	return (0);
1619 }
1620 
1621 /*
1622  * Common code for kill process group/broadcast kill.
1623  * cp is calling process.
1624  */
1625 static int
1626 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1627 {
1628 	struct proc *p;
1629 	struct pgrp *pgrp;
1630 	int nfound = 0;
1631 
1632 	if (all) {
1633 		/*
1634 		 * broadcast
1635 		 */
1636 		sx_slock(&allproc_lock);
1637 		FOREACH_PROC_IN_SYSTEM(p) {
1638 			PROC_LOCK(p);
1639 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1640 			    p == td->td_proc || p->p_state == PRS_NEW) {
1641 				PROC_UNLOCK(p);
1642 				continue;
1643 			}
1644 			if (p_cansignal(td, p, sig) == 0) {
1645 				nfound++;
1646 				if (sig)
1647 					pksignal(p, sig, ksi);
1648 			}
1649 			PROC_UNLOCK(p);
1650 		}
1651 		sx_sunlock(&allproc_lock);
1652 	} else {
1653 		sx_slock(&proctree_lock);
1654 		if (pgid == 0) {
1655 			/*
1656 			 * zero pgid means send to my process group.
1657 			 */
1658 			pgrp = td->td_proc->p_pgrp;
1659 			PGRP_LOCK(pgrp);
1660 		} else {
1661 			pgrp = pgfind(pgid);
1662 			if (pgrp == NULL) {
1663 				sx_sunlock(&proctree_lock);
1664 				return (ESRCH);
1665 			}
1666 		}
1667 		sx_sunlock(&proctree_lock);
1668 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1669 			PROC_LOCK(p);
1670 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1671 				p->p_state == PRS_NEW ) {
1672 				PROC_UNLOCK(p);
1673 				continue;
1674 			}
1675 			if (p_cansignal(td, p, sig) == 0) {
1676 				nfound++;
1677 				if (sig)
1678 					pksignal(p, sig, ksi);
1679 			}
1680 			PROC_UNLOCK(p);
1681 		}
1682 		PGRP_UNLOCK(pgrp);
1683 	}
1684 	return (nfound ? 0 : ESRCH);
1685 }
1686 
1687 #ifndef _SYS_SYSPROTO_H_
1688 struct kill_args {
1689 	int	pid;
1690 	int	signum;
1691 };
1692 #endif
1693 /* ARGSUSED */
1694 int
1695 kill(struct thread *td, struct kill_args *uap)
1696 {
1697 	ksiginfo_t ksi;
1698 	struct proc *p;
1699 	int error;
1700 
1701 	AUDIT_ARG_SIGNUM(uap->signum);
1702 	AUDIT_ARG_PID(uap->pid);
1703 	if ((u_int)uap->signum > _SIG_MAXSIG)
1704 		return (EINVAL);
1705 
1706 	ksiginfo_init(&ksi);
1707 	ksi.ksi_signo = uap->signum;
1708 	ksi.ksi_code = SI_USER;
1709 	ksi.ksi_pid = td->td_proc->p_pid;
1710 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1711 
1712 	if (uap->pid > 0) {
1713 		/* kill single process */
1714 		if ((p = pfind(uap->pid)) == NULL) {
1715 			if ((p = zpfind(uap->pid)) == NULL)
1716 				return (ESRCH);
1717 		}
1718 		AUDIT_ARG_PROCESS(p);
1719 		error = p_cansignal(td, p, uap->signum);
1720 		if (error == 0 && uap->signum)
1721 			pksignal(p, uap->signum, &ksi);
1722 		PROC_UNLOCK(p);
1723 		return (error);
1724 	}
1725 	switch (uap->pid) {
1726 	case -1:		/* broadcast signal */
1727 		return (killpg1(td, uap->signum, 0, 1, &ksi));
1728 	case 0:			/* signal own process group */
1729 		return (killpg1(td, uap->signum, 0, 0, &ksi));
1730 	default:		/* negative explicit process group */
1731 		return (killpg1(td, uap->signum, -uap->pid, 0, &ksi));
1732 	}
1733 	/* NOTREACHED */
1734 }
1735 
1736 #if defined(COMPAT_43)
1737 #ifndef _SYS_SYSPROTO_H_
1738 struct okillpg_args {
1739 	int	pgid;
1740 	int	signum;
1741 };
1742 #endif
1743 /* ARGSUSED */
1744 int
1745 okillpg(struct thread *td, struct okillpg_args *uap)
1746 {
1747 	ksiginfo_t ksi;
1748 
1749 	AUDIT_ARG_SIGNUM(uap->signum);
1750 	AUDIT_ARG_PID(uap->pgid);
1751 	if ((u_int)uap->signum > _SIG_MAXSIG)
1752 		return (EINVAL);
1753 
1754 	ksiginfo_init(&ksi);
1755 	ksi.ksi_signo = uap->signum;
1756 	ksi.ksi_code = SI_USER;
1757 	ksi.ksi_pid = td->td_proc->p_pid;
1758 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1759 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1760 }
1761 #endif /* COMPAT_43 */
1762 
1763 #ifndef _SYS_SYSPROTO_H_
1764 struct sigqueue_args {
1765 	pid_t pid;
1766 	int signum;
1767 	/* union sigval */ void *value;
1768 };
1769 #endif
1770 int
1771 sigqueue(struct thread *td, struct sigqueue_args *uap)
1772 {
1773 	ksiginfo_t ksi;
1774 	struct proc *p;
1775 	int error;
1776 
1777 	if ((u_int)uap->signum > _SIG_MAXSIG)
1778 		return (EINVAL);
1779 
1780 	/*
1781 	 * Specification says sigqueue can only send signal to
1782 	 * single process.
1783 	 */
1784 	if (uap->pid <= 0)
1785 		return (EINVAL);
1786 
1787 	if ((p = pfind(uap->pid)) == NULL) {
1788 		if ((p = zpfind(uap->pid)) == NULL)
1789 			return (ESRCH);
1790 	}
1791 	error = p_cansignal(td, p, uap->signum);
1792 	if (error == 0 && uap->signum != 0) {
1793 		ksiginfo_init(&ksi);
1794 		ksi.ksi_flags = KSI_SIGQ;
1795 		ksi.ksi_signo = uap->signum;
1796 		ksi.ksi_code = SI_QUEUE;
1797 		ksi.ksi_pid = td->td_proc->p_pid;
1798 		ksi.ksi_uid = td->td_ucred->cr_ruid;
1799 		ksi.ksi_value.sival_ptr = uap->value;
1800 		error = tdsignal(p, NULL, ksi.ksi_signo, &ksi);
1801 	}
1802 	PROC_UNLOCK(p);
1803 	return (error);
1804 }
1805 
1806 /*
1807  * Send a signal to a process group.
1808  */
1809 void
1810 gsignal(int pgid, int sig, ksiginfo_t *ksi)
1811 {
1812 	struct pgrp *pgrp;
1813 
1814 	if (pgid != 0) {
1815 		sx_slock(&proctree_lock);
1816 		pgrp = pgfind(pgid);
1817 		sx_sunlock(&proctree_lock);
1818 		if (pgrp != NULL) {
1819 			pgsignal(pgrp, sig, 0, ksi);
1820 			PGRP_UNLOCK(pgrp);
1821 		}
1822 	}
1823 }
1824 
1825 /*
1826  * Send a signal to a process group.  If checktty is 1,
1827  * limit to members which have a controlling terminal.
1828  */
1829 void
1830 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
1831 {
1832 	struct proc *p;
1833 
1834 	if (pgrp) {
1835 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
1836 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1837 			PROC_LOCK(p);
1838 			if (checkctty == 0 || p->p_flag & P_CONTROLT)
1839 				pksignal(p, sig, ksi);
1840 			PROC_UNLOCK(p);
1841 		}
1842 	}
1843 }
1844 
1845 /*
1846  * Send a signal caused by a trap to the current thread.  If it will be
1847  * caught immediately, deliver it with correct code.  Otherwise, post it
1848  * normally.
1849  */
1850 void
1851 trapsignal(struct thread *td, ksiginfo_t *ksi)
1852 {
1853 	struct sigacts *ps;
1854 	sigset_t mask;
1855 	struct proc *p;
1856 	int sig;
1857 	int code;
1858 
1859 	p = td->td_proc;
1860 	sig = ksi->ksi_signo;
1861 	code = ksi->ksi_code;
1862 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
1863 
1864 	PROC_LOCK(p);
1865 	ps = p->p_sigacts;
1866 	mtx_lock(&ps->ps_mtx);
1867 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
1868 	    !SIGISMEMBER(td->td_sigmask, sig)) {
1869 		td->td_ru.ru_nsignals++;
1870 #ifdef KTRACE
1871 		if (KTRPOINT(curthread, KTR_PSIG))
1872 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
1873 			    &td->td_sigmask, code);
1874 #endif
1875 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
1876 				ksi, &td->td_sigmask);
1877 		mask = ps->ps_catchmask[_SIG_IDX(sig)];
1878 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
1879 			SIGADDSET(mask, sig);
1880 		kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
1881 		    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
1882 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
1883 			/*
1884 			 * See kern_sigaction() for origin of this code.
1885 			 */
1886 			SIGDELSET(ps->ps_sigcatch, sig);
1887 			if (sig != SIGCONT &&
1888 			    sigprop(sig) & SA_IGNORE)
1889 				SIGADDSET(ps->ps_sigignore, sig);
1890 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1891 		}
1892 		mtx_unlock(&ps->ps_mtx);
1893 	} else {
1894 		/*
1895 		 * Avoid a possible infinite loop if the thread
1896 		 * masking the signal or process is ignoring the
1897 		 * signal.
1898 		 */
1899 		if (kern_forcesigexit &&
1900 		    (SIGISMEMBER(td->td_sigmask, sig) ||
1901 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
1902 			SIGDELSET(td->td_sigmask, sig);
1903 			SIGDELSET(ps->ps_sigcatch, sig);
1904 			SIGDELSET(ps->ps_sigignore, sig);
1905 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1906 		}
1907 		mtx_unlock(&ps->ps_mtx);
1908 		p->p_code = code;	/* XXX for core dump/debugger */
1909 		p->p_sig = sig;		/* XXX to verify code */
1910 		tdsignal(p, td, sig, ksi);
1911 	}
1912 	PROC_UNLOCK(p);
1913 }
1914 
1915 static struct thread *
1916 sigtd(struct proc *p, int sig, int prop)
1917 {
1918 	struct thread *td, *signal_td;
1919 
1920 	PROC_LOCK_ASSERT(p, MA_OWNED);
1921 
1922 	/*
1923 	 * Check if current thread can handle the signal without
1924 	 * switching context to another thread.
1925 	 */
1926 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig))
1927 		return (curthread);
1928 	signal_td = NULL;
1929 	FOREACH_THREAD_IN_PROC(p, td) {
1930 		if (!SIGISMEMBER(td->td_sigmask, sig)) {
1931 			signal_td = td;
1932 			break;
1933 		}
1934 	}
1935 	if (signal_td == NULL)
1936 		signal_td = FIRST_THREAD_IN_PROC(p);
1937 	return (signal_td);
1938 }
1939 
1940 /*
1941  * Send the signal to the process.  If the signal has an action, the action
1942  * is usually performed by the target process rather than the caller; we add
1943  * the signal to the set of pending signals for the process.
1944  *
1945  * Exceptions:
1946  *   o When a stop signal is sent to a sleeping process that takes the
1947  *     default action, the process is stopped without awakening it.
1948  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
1949  *     regardless of the signal action (eg, blocked or ignored).
1950  *
1951  * Other ignored signals are discarded immediately.
1952  *
1953  * NB: This function may be entered from the debugger via the "kill" DDB
1954  * command.  There is little that can be done to mitigate the possibly messy
1955  * side effects of this unwise possibility.
1956  */
1957 void
1958 psignal(struct proc *p, int sig)
1959 {
1960 	ksiginfo_t ksi;
1961 
1962 	ksiginfo_init(&ksi);
1963 	ksi.ksi_signo = sig;
1964 	ksi.ksi_code = SI_KERNEL;
1965 	(void) tdsignal(p, NULL, sig, &ksi);
1966 }
1967 
1968 void
1969 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
1970 {
1971 
1972 	(void) tdsignal(p, NULL, sig, ksi);
1973 }
1974 
1975 int
1976 psignal_event(struct proc *p, struct sigevent *sigev, ksiginfo_t *ksi)
1977 {
1978 	struct thread *td = NULL;
1979 
1980 	PROC_LOCK_ASSERT(p, MA_OWNED);
1981 
1982 	KASSERT(!KSI_ONQ(ksi), ("psignal_event: ksi on queue"));
1983 
1984 	/*
1985 	 * ksi_code and other fields should be set before
1986 	 * calling this function.
1987 	 */
1988 	ksi->ksi_signo = sigev->sigev_signo;
1989 	ksi->ksi_value = sigev->sigev_value;
1990 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
1991 		td = thread_find(p, sigev->sigev_notify_thread_id);
1992 		if (td == NULL)
1993 			return (ESRCH);
1994 	}
1995 	return (tdsignal(p, td, ksi->ksi_signo, ksi));
1996 }
1997 
1998 int
1999 tdsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2000 {
2001 	sig_t action;
2002 	sigqueue_t *sigqueue;
2003 	int prop;
2004 	struct sigacts *ps;
2005 	int intrval;
2006 	int ret = 0;
2007 	int wakeup_swapper;
2008 
2009 	PROC_LOCK_ASSERT(p, MA_OWNED);
2010 
2011 	if (!_SIG_VALID(sig))
2012 		panic("tdsignal(): invalid signal %d", sig);
2013 
2014 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("tdsignal: ksi on queue"));
2015 
2016 	/*
2017 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2018 	 */
2019 	if (p->p_state == PRS_ZOMBIE) {
2020 		if (ksi && (ksi->ksi_flags & KSI_INS))
2021 			ksiginfo_tryfree(ksi);
2022 		return (ret);
2023 	}
2024 
2025 	ps = p->p_sigacts;
2026 	KNOTE_LOCKED(&p->p_klist, NOTE_SIGNAL | sig);
2027 	prop = sigprop(sig);
2028 
2029 	if (td == NULL) {
2030 		td = sigtd(p, sig, prop);
2031 		sigqueue = &p->p_sigqueue;
2032 	} else {
2033 		KASSERT(td->td_proc == p, ("invalid thread"));
2034 		sigqueue = &td->td_sigqueue;
2035 	}
2036 
2037 	SDT_PROBE(proc, kernel, , signal_send, td, p, sig, 0, 0 );
2038 
2039 	/*
2040 	 * If the signal is being ignored,
2041 	 * then we forget about it immediately.
2042 	 * (Note: we don't set SIGCONT in ps_sigignore,
2043 	 * and if it is set to SIG_IGN,
2044 	 * action will be SIG_DFL here.)
2045 	 */
2046 	mtx_lock(&ps->ps_mtx);
2047 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2048 		SDT_PROBE(proc, kernel, , signal_discard, ps, td, sig, 0, 0 );
2049 
2050 		mtx_unlock(&ps->ps_mtx);
2051 		if (ksi && (ksi->ksi_flags & KSI_INS))
2052 			ksiginfo_tryfree(ksi);
2053 		return (ret);
2054 	}
2055 	if (SIGISMEMBER(td->td_sigmask, sig))
2056 		action = SIG_HOLD;
2057 	else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2058 		action = SIG_CATCH;
2059 	else
2060 		action = SIG_DFL;
2061 	if (SIGISMEMBER(ps->ps_sigintr, sig))
2062 		intrval = EINTR;
2063 	else
2064 		intrval = ERESTART;
2065 	mtx_unlock(&ps->ps_mtx);
2066 
2067 	if (prop & SA_CONT)
2068 		sigqueue_delete_stopmask_proc(p);
2069 	else if (prop & SA_STOP) {
2070 		/*
2071 		 * If sending a tty stop signal to a member of an orphaned
2072 		 * process group, discard the signal here if the action
2073 		 * is default; don't stop the process below if sleeping,
2074 		 * and don't clear any pending SIGCONT.
2075 		 */
2076 		if ((prop & SA_TTYSTOP) &&
2077 		    (p->p_pgrp->pg_jobc == 0) &&
2078 		    (action == SIG_DFL)) {
2079 			if (ksi && (ksi->ksi_flags & KSI_INS))
2080 				ksiginfo_tryfree(ksi);
2081 			return (ret);
2082 		}
2083 		sigqueue_delete_proc(p, SIGCONT);
2084 		if (p->p_flag & P_CONTINUED) {
2085 			p->p_flag &= ~P_CONTINUED;
2086 			PROC_LOCK(p->p_pptr);
2087 			sigqueue_take(p->p_ksi);
2088 			PROC_UNLOCK(p->p_pptr);
2089 		}
2090 	}
2091 
2092 	ret = sigqueue_add(sigqueue, sig, ksi);
2093 	if (ret != 0)
2094 		return (ret);
2095 	signotify(td);
2096 	/*
2097 	 * Defer further processing for signals which are held,
2098 	 * except that stopped processes must be continued by SIGCONT.
2099 	 */
2100 	if (action == SIG_HOLD &&
2101 	    !((prop & SA_CONT) && (p->p_flag & P_STOPPED_SIG)))
2102 		return (ret);
2103 	/*
2104 	 * SIGKILL: Remove procfs STOPEVENTs.
2105 	 */
2106 	if (sig == SIGKILL) {
2107 		/* from procfs_ioctl.c: PIOCBIC */
2108 		p->p_stops = 0;
2109 		/* from procfs_ioctl.c: PIOCCONT */
2110 		p->p_step = 0;
2111 		wakeup(&p->p_step);
2112 	}
2113 	/*
2114 	 * Some signals have a process-wide effect and a per-thread
2115 	 * component.  Most processing occurs when the process next
2116 	 * tries to cross the user boundary, however there are some
2117 	 * times when processing needs to be done immediatly, such as
2118 	 * waking up threads so that they can cross the user boundary.
2119 	 * We try do the per-process part here.
2120 	 */
2121 	if (P_SHOULDSTOP(p)) {
2122 		/*
2123 		 * The process is in stopped mode. All the threads should be
2124 		 * either winding down or already on the suspended queue.
2125 		 */
2126 		if (p->p_flag & P_TRACED) {
2127 			/*
2128 			 * The traced process is already stopped,
2129 			 * so no further action is necessary.
2130 			 * No signal can restart us.
2131 			 */
2132 			goto out;
2133 		}
2134 
2135 		if (sig == SIGKILL) {
2136 			/*
2137 			 * SIGKILL sets process running.
2138 			 * It will die elsewhere.
2139 			 * All threads must be restarted.
2140 			 */
2141 			p->p_flag &= ~P_STOPPED_SIG;
2142 			goto runfast;
2143 		}
2144 
2145 		if (prop & SA_CONT) {
2146 			/*
2147 			 * If SIGCONT is default (or ignored), we continue the
2148 			 * process but don't leave the signal in sigqueue as
2149 			 * it has no further action.  If SIGCONT is held, we
2150 			 * continue the process and leave the signal in
2151 			 * sigqueue.  If the process catches SIGCONT, let it
2152 			 * handle the signal itself.  If it isn't waiting on
2153 			 * an event, it goes back to run state.
2154 			 * Otherwise, process goes back to sleep state.
2155 			 */
2156 			p->p_flag &= ~P_STOPPED_SIG;
2157 			PROC_SLOCK(p);
2158 			if (p->p_numthreads == p->p_suspcount) {
2159 				PROC_SUNLOCK(p);
2160 				p->p_flag |= P_CONTINUED;
2161 				p->p_xstat = SIGCONT;
2162 				PROC_LOCK(p->p_pptr);
2163 				childproc_continued(p);
2164 				PROC_UNLOCK(p->p_pptr);
2165 				PROC_SLOCK(p);
2166 			}
2167 			if (action == SIG_DFL) {
2168 				thread_unsuspend(p);
2169 				PROC_SUNLOCK(p);
2170 				sigqueue_delete(sigqueue, sig);
2171 				goto out;
2172 			}
2173 			if (action == SIG_CATCH) {
2174 				/*
2175 				 * The process wants to catch it so it needs
2176 				 * to run at least one thread, but which one?
2177 				 */
2178 				PROC_SUNLOCK(p);
2179 				goto runfast;
2180 			}
2181 			/*
2182 			 * The signal is not ignored or caught.
2183 			 */
2184 			thread_unsuspend(p);
2185 			PROC_SUNLOCK(p);
2186 			goto out;
2187 		}
2188 
2189 		if (prop & SA_STOP) {
2190 			/*
2191 			 * Already stopped, don't need to stop again
2192 			 * (If we did the shell could get confused).
2193 			 * Just make sure the signal STOP bit set.
2194 			 */
2195 			p->p_flag |= P_STOPPED_SIG;
2196 			sigqueue_delete(sigqueue, sig);
2197 			goto out;
2198 		}
2199 
2200 		/*
2201 		 * All other kinds of signals:
2202 		 * If a thread is sleeping interruptibly, simulate a
2203 		 * wakeup so that when it is continued it will be made
2204 		 * runnable and can look at the signal.  However, don't make
2205 		 * the PROCESS runnable, leave it stopped.
2206 		 * It may run a bit until it hits a thread_suspend_check().
2207 		 */
2208 		wakeup_swapper = 0;
2209 		PROC_SLOCK(p);
2210 		thread_lock(td);
2211 		if (TD_ON_SLEEPQ(td) && (td->td_flags & TDF_SINTR))
2212 			wakeup_swapper = sleepq_abort(td, intrval);
2213 		thread_unlock(td);
2214 		PROC_SUNLOCK(p);
2215 		if (wakeup_swapper)
2216 			kick_proc0();
2217 		goto out;
2218 		/*
2219 		 * Mutexes are short lived. Threads waiting on them will
2220 		 * hit thread_suspend_check() soon.
2221 		 */
2222 	} else if (p->p_state == PRS_NORMAL) {
2223 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2224 			tdsigwakeup(td, sig, action, intrval);
2225 			goto out;
2226 		}
2227 
2228 		MPASS(action == SIG_DFL);
2229 
2230 		if (prop & SA_STOP) {
2231 			if (p->p_flag & P_PPWAIT)
2232 				goto out;
2233 			p->p_flag |= P_STOPPED_SIG;
2234 			p->p_xstat = sig;
2235 			PROC_SLOCK(p);
2236 			sig_suspend_threads(td, p, 1);
2237 			if (p->p_numthreads == p->p_suspcount) {
2238 				/*
2239 				 * only thread sending signal to another
2240 				 * process can reach here, if thread is sending
2241 				 * signal to its process, because thread does
2242 				 * not suspend itself here, p_numthreads
2243 				 * should never be equal to p_suspcount.
2244 				 */
2245 				thread_stopped(p);
2246 				PROC_SUNLOCK(p);
2247 				sigqueue_delete_proc(p, p->p_xstat);
2248 			} else
2249 				PROC_SUNLOCK(p);
2250 			goto out;
2251 		}
2252 	} else {
2253 		/* Not in "NORMAL" state. discard the signal. */
2254 		sigqueue_delete(sigqueue, sig);
2255 		goto out;
2256 	}
2257 
2258 	/*
2259 	 * The process is not stopped so we need to apply the signal to all the
2260 	 * running threads.
2261 	 */
2262 runfast:
2263 	tdsigwakeup(td, sig, action, intrval);
2264 	PROC_SLOCK(p);
2265 	thread_unsuspend(p);
2266 	PROC_SUNLOCK(p);
2267 out:
2268 	/* If we jump here, proc slock should not be owned. */
2269 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2270 	return (ret);
2271 }
2272 
2273 /*
2274  * The force of a signal has been directed against a single
2275  * thread.  We need to see what we can do about knocking it
2276  * out of any sleep it may be in etc.
2277  */
2278 static void
2279 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2280 {
2281 	struct proc *p = td->td_proc;
2282 	register int prop;
2283 	int wakeup_swapper;
2284 
2285 	wakeup_swapper = 0;
2286 	PROC_LOCK_ASSERT(p, MA_OWNED);
2287 	prop = sigprop(sig);
2288 
2289 	PROC_SLOCK(p);
2290 	thread_lock(td);
2291 	/*
2292 	 * Bring the priority of a thread up if we want it to get
2293 	 * killed in this lifetime.
2294 	 */
2295 	if (action == SIG_DFL && (prop & SA_KILL) && td->td_priority > PUSER)
2296 		sched_prio(td, PUSER);
2297 	if (TD_ON_SLEEPQ(td)) {
2298 		/*
2299 		 * If thread is sleeping uninterruptibly
2300 		 * we can't interrupt the sleep... the signal will
2301 		 * be noticed when the process returns through
2302 		 * trap() or syscall().
2303 		 */
2304 		if ((td->td_flags & TDF_SINTR) == 0)
2305 			goto out;
2306 		/*
2307 		 * If SIGCONT is default (or ignored) and process is
2308 		 * asleep, we are finished; the process should not
2309 		 * be awakened.
2310 		 */
2311 		if ((prop & SA_CONT) && action == SIG_DFL) {
2312 			thread_unlock(td);
2313 			PROC_SUNLOCK(p);
2314 			sigqueue_delete(&p->p_sigqueue, sig);
2315 			/*
2316 			 * It may be on either list in this state.
2317 			 * Remove from both for now.
2318 			 */
2319 			sigqueue_delete(&td->td_sigqueue, sig);
2320 			return;
2321 		}
2322 
2323 		/*
2324 		 * Give low priority threads a better chance to run.
2325 		 */
2326 		if (td->td_priority > PUSER)
2327 			sched_prio(td, PUSER);
2328 
2329 		wakeup_swapper = sleepq_abort(td, intrval);
2330 	} else {
2331 		/*
2332 		 * Other states do nothing with the signal immediately,
2333 		 * other than kicking ourselves if we are running.
2334 		 * It will either never be noticed, or noticed very soon.
2335 		 */
2336 #ifdef SMP
2337 		if (TD_IS_RUNNING(td) && td != curthread)
2338 			forward_signal(td);
2339 #endif
2340 	}
2341 out:
2342 	PROC_SUNLOCK(p);
2343 	thread_unlock(td);
2344 	if (wakeup_swapper)
2345 		kick_proc0();
2346 }
2347 
2348 static void
2349 sig_suspend_threads(struct thread *td, struct proc *p, int sending)
2350 {
2351 	struct thread *td2;
2352 	int wakeup_swapper;
2353 
2354 	PROC_LOCK_ASSERT(p, MA_OWNED);
2355 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2356 
2357 	wakeup_swapper = 0;
2358 	FOREACH_THREAD_IN_PROC(p, td2) {
2359 		thread_lock(td2);
2360 		td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
2361 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2362 		    (td2->td_flags & TDF_SINTR)) {
2363 			if (td2->td_flags & TDF_SBDRY) {
2364 				if (TD_IS_SUSPENDED(td2))
2365 					wakeup_swapper |=
2366 					    thread_unsuspend_one(td2);
2367 				if (TD_ON_SLEEPQ(td2))
2368 					wakeup_swapper |=
2369 					    sleepq_abort(td2, ERESTART);
2370 			} else if (!TD_IS_SUSPENDED(td2)) {
2371 				thread_suspend_one(td2);
2372 			}
2373 		} else if (!TD_IS_SUSPENDED(td2)) {
2374 			if (sending || td != td2)
2375 				td2->td_flags |= TDF_ASTPENDING;
2376 #ifdef SMP
2377 			if (TD_IS_RUNNING(td2) && td2 != td)
2378 				forward_signal(td2);
2379 #endif
2380 		}
2381 		thread_unlock(td2);
2382 	}
2383 	if (wakeup_swapper)
2384 		kick_proc0();
2385 }
2386 
2387 int
2388 ptracestop(struct thread *td, int sig)
2389 {
2390 	struct proc *p = td->td_proc;
2391 
2392 	PROC_LOCK_ASSERT(p, MA_OWNED);
2393 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2394 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2395 
2396 	td->td_dbgflags |= TDB_XSIG;
2397 	td->td_xsig = sig;
2398 	PROC_SLOCK(p);
2399 	while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2400 		if (p->p_flag & P_SINGLE_EXIT) {
2401 			td->td_dbgflags &= ~TDB_XSIG;
2402 			PROC_SUNLOCK(p);
2403 			return (sig);
2404 		}
2405 		/*
2406 		 * Just make wait() to work, the last stopped thread
2407 		 * will win.
2408 		 */
2409 		p->p_xstat = sig;
2410 		p->p_xthread = td;
2411 		p->p_flag |= (P_STOPPED_SIG|P_STOPPED_TRACE);
2412 		sig_suspend_threads(td, p, 0);
2413 stopme:
2414 		thread_suspend_switch(td);
2415 		if (!(p->p_flag & P_TRACED)) {
2416 			break;
2417 		}
2418 		if (td->td_dbgflags & TDB_SUSPEND) {
2419 			if (p->p_flag & P_SINGLE_EXIT)
2420 				break;
2421 			goto stopme;
2422 		}
2423 	}
2424 	PROC_SUNLOCK(p);
2425 	return (td->td_xsig);
2426 }
2427 
2428 static void
2429 reschedule_signals(struct proc *p, sigset_t block, int flags)
2430 {
2431 	struct sigacts *ps;
2432 	struct thread *td;
2433 	int i;
2434 
2435 	PROC_LOCK_ASSERT(p, MA_OWNED);
2436 
2437 	ps = p->p_sigacts;
2438 	for (i = 1; !SIGISEMPTY(block); i++) {
2439 		if (!SIGISMEMBER(block, i))
2440 			continue;
2441 		SIGDELSET(block, i);
2442 		if (!SIGISMEMBER(p->p_siglist, i))
2443 			continue;
2444 
2445 		td = sigtd(p, i, 0);
2446 		signotify(td);
2447 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2448 			mtx_lock(&ps->ps_mtx);
2449 		if (p->p_flag & P_TRACED || SIGISMEMBER(ps->ps_sigcatch, i))
2450 			tdsigwakeup(td, i, SIG_CATCH,
2451 			    (SIGISMEMBER(ps->ps_sigintr, i) ? EINTR :
2452 			     ERESTART));
2453 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2454 			mtx_unlock(&ps->ps_mtx);
2455 	}
2456 }
2457 
2458 void
2459 tdsigcleanup(struct thread *td)
2460 {
2461 	struct proc *p;
2462 	sigset_t unblocked;
2463 
2464 	p = td->td_proc;
2465 	PROC_LOCK_ASSERT(p, MA_OWNED);
2466 
2467 	sigqueue_flush(&td->td_sigqueue);
2468 	if (p->p_numthreads == 1)
2469 		return;
2470 
2471 	/*
2472 	 * Since we cannot handle signals, notify signal post code
2473 	 * about this by filling the sigmask.
2474 	 *
2475 	 * Also, if needed, wake up thread(s) that do not block the
2476 	 * same signals as the exiting thread, since the thread might
2477 	 * have been selected for delivery and woken up.
2478 	 */
2479 	SIGFILLSET(unblocked);
2480 	SIGSETNAND(unblocked, td->td_sigmask);
2481 	SIGFILLSET(td->td_sigmask);
2482 	reschedule_signals(p, unblocked, 0);
2483 
2484 }
2485 
2486 /*
2487  * If the current process has received a signal (should be caught or cause
2488  * termination, should interrupt current syscall), return the signal number.
2489  * Stop signals with default action are processed immediately, then cleared;
2490  * they aren't returned.  This is checked after each entry to the system for
2491  * a syscall or trap (though this can usually be done without calling issignal
2492  * by checking the pending signal masks in cursig.) The normal call
2493  * sequence is
2494  *
2495  *	while (sig = cursig(curthread))
2496  *		postsig(sig);
2497  */
2498 static int
2499 issignal(struct thread *td, int stop_allowed)
2500 {
2501 	struct proc *p;
2502 	struct sigacts *ps;
2503 	struct sigqueue *queue;
2504 	sigset_t sigpending;
2505 	ksiginfo_t ksi;
2506 	int sig, prop, newsig;
2507 
2508 	p = td->td_proc;
2509 	ps = p->p_sigacts;
2510 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2511 	PROC_LOCK_ASSERT(p, MA_OWNED);
2512 	for (;;) {
2513 		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
2514 
2515 		sigpending = td->td_sigqueue.sq_signals;
2516 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
2517 		SIGSETNAND(sigpending, td->td_sigmask);
2518 
2519 		if (p->p_flag & P_PPWAIT)
2520 			SIG_STOPSIGMASK(sigpending);
2521 		if (SIGISEMPTY(sigpending))	/* no signal to send */
2522 			return (0);
2523 		sig = sig_ffs(&sigpending);
2524 
2525 		if (p->p_stops & S_SIG) {
2526 			mtx_unlock(&ps->ps_mtx);
2527 			stopevent(p, S_SIG, sig);
2528 			mtx_lock(&ps->ps_mtx);
2529 		}
2530 
2531 		/*
2532 		 * We should see pending but ignored signals
2533 		 * only if P_TRACED was on when they were posted.
2534 		 */
2535 		if (SIGISMEMBER(ps->ps_sigignore, sig) && (traced == 0)) {
2536 			sigqueue_delete(&td->td_sigqueue, sig);
2537 			sigqueue_delete(&p->p_sigqueue, sig);
2538 			continue;
2539 		}
2540 		if (p->p_flag & P_TRACED && (p->p_flag & P_PPWAIT) == 0) {
2541 			/*
2542 			 * If traced, always stop.
2543 			 * Remove old signal from queue before the stop.
2544 			 * XXX shrug off debugger, it causes siginfo to
2545 			 * be thrown away.
2546 			 */
2547 			queue = &td->td_sigqueue;
2548 			ksi.ksi_signo = 0;
2549 			if (sigqueue_get(queue, sig, &ksi) == 0) {
2550 				queue = &p->p_sigqueue;
2551 				sigqueue_get(queue, sig, &ksi);
2552 			}
2553 
2554 			mtx_unlock(&ps->ps_mtx);
2555 			newsig = ptracestop(td, sig);
2556 			mtx_lock(&ps->ps_mtx);
2557 
2558 			if (sig != newsig) {
2559 
2560 				/*
2561 				 * If parent wants us to take the signal,
2562 				 * then it will leave it in p->p_xstat;
2563 				 * otherwise we just look for signals again.
2564 			 	*/
2565 				if (newsig == 0)
2566 					continue;
2567 				sig = newsig;
2568 
2569 				/*
2570 				 * Put the new signal into td_sigqueue. If the
2571 				 * signal is being masked, look for other signals.
2572 				 */
2573 				sigqueue_add(queue, sig, NULL);
2574 				if (SIGISMEMBER(td->td_sigmask, sig))
2575 					continue;
2576 				signotify(td);
2577 			} else {
2578 				if (ksi.ksi_signo != 0) {
2579 					ksi.ksi_flags |= KSI_HEAD;
2580 					if (sigqueue_add(&td->td_sigqueue, sig,
2581 					    &ksi) != 0)
2582 						ksi.ksi_signo = 0;
2583 				}
2584 				if (ksi.ksi_signo == 0)
2585 					sigqueue_add(&td->td_sigqueue, sig,
2586 					    NULL);
2587 			}
2588 
2589 			/*
2590 			 * If the traced bit got turned off, go back up
2591 			 * to the top to rescan signals.  This ensures
2592 			 * that p_sig* and p_sigact are consistent.
2593 			 */
2594 			if ((p->p_flag & P_TRACED) == 0)
2595 				continue;
2596 		}
2597 
2598 		prop = sigprop(sig);
2599 
2600 		/*
2601 		 * Decide whether the signal should be returned.
2602 		 * Return the signal's number, or fall through
2603 		 * to clear it from the pending mask.
2604 		 */
2605 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2606 
2607 		case (intptr_t)SIG_DFL:
2608 			/*
2609 			 * Don't take default actions on system processes.
2610 			 */
2611 			if (p->p_pid <= 1) {
2612 #ifdef DIAGNOSTIC
2613 				/*
2614 				 * Are you sure you want to ignore SIGSEGV
2615 				 * in init? XXX
2616 				 */
2617 				printf("Process (pid %lu) got signal %d\n",
2618 					(u_long)p->p_pid, sig);
2619 #endif
2620 				break;		/* == ignore */
2621 			}
2622 			/*
2623 			 * If there is a pending stop signal to process
2624 			 * with default action, stop here,
2625 			 * then clear the signal.  However,
2626 			 * if process is member of an orphaned
2627 			 * process group, ignore tty stop signals.
2628 			 */
2629 			if (prop & SA_STOP) {
2630 				if (p->p_flag & P_TRACED ||
2631 		    		    (p->p_pgrp->pg_jobc == 0 &&
2632 				     prop & SA_TTYSTOP))
2633 					break;	/* == ignore */
2634 
2635 				/* Ignore, but do not drop the stop signal. */
2636 				if (stop_allowed != SIG_STOP_ALLOWED)
2637 					return (sig);
2638 				mtx_unlock(&ps->ps_mtx);
2639 				WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2640 				    &p->p_mtx.lock_object, "Catching SIGSTOP");
2641 				p->p_flag |= P_STOPPED_SIG;
2642 				p->p_xstat = sig;
2643 				PROC_SLOCK(p);
2644 				sig_suspend_threads(td, p, 0);
2645 				thread_suspend_switch(td);
2646 				PROC_SUNLOCK(p);
2647 				mtx_lock(&ps->ps_mtx);
2648 				break;
2649 			} else if (prop & SA_IGNORE) {
2650 				/*
2651 				 * Except for SIGCONT, shouldn't get here.
2652 				 * Default action is to ignore; drop it.
2653 				 */
2654 				break;		/* == ignore */
2655 			} else
2656 				return (sig);
2657 			/*NOTREACHED*/
2658 
2659 		case (intptr_t)SIG_IGN:
2660 			/*
2661 			 * Masking above should prevent us ever trying
2662 			 * to take action on an ignored signal other
2663 			 * than SIGCONT, unless process is traced.
2664 			 */
2665 			if ((prop & SA_CONT) == 0 &&
2666 			    (p->p_flag & P_TRACED) == 0)
2667 				printf("issignal\n");
2668 			break;		/* == ignore */
2669 
2670 		default:
2671 			/*
2672 			 * This signal has an action, let
2673 			 * postsig() process it.
2674 			 */
2675 			return (sig);
2676 		}
2677 		sigqueue_delete(&td->td_sigqueue, sig);		/* take the signal! */
2678 		sigqueue_delete(&p->p_sigqueue, sig);
2679 	}
2680 	/* NOTREACHED */
2681 }
2682 
2683 void
2684 thread_stopped(struct proc *p)
2685 {
2686 	int n;
2687 
2688 	PROC_LOCK_ASSERT(p, MA_OWNED);
2689 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2690 	n = p->p_suspcount;
2691 	if (p == curproc)
2692 		n++;
2693 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
2694 		PROC_SUNLOCK(p);
2695 		p->p_flag &= ~P_WAITED;
2696 		PROC_LOCK(p->p_pptr);
2697 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
2698 			CLD_TRAPPED : CLD_STOPPED);
2699 		PROC_UNLOCK(p->p_pptr);
2700 		PROC_SLOCK(p);
2701 	}
2702 }
2703 
2704 /*
2705  * Take the action for the specified signal
2706  * from the current set of pending signals.
2707  */
2708 int
2709 postsig(sig)
2710 	register int sig;
2711 {
2712 	struct thread *td = curthread;
2713 	register struct proc *p = td->td_proc;
2714 	struct sigacts *ps;
2715 	sig_t action;
2716 	ksiginfo_t ksi;
2717 	sigset_t returnmask, mask;
2718 
2719 	KASSERT(sig != 0, ("postsig"));
2720 
2721 	PROC_LOCK_ASSERT(p, MA_OWNED);
2722 	ps = p->p_sigacts;
2723 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2724 	ksiginfo_init(&ksi);
2725 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
2726 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
2727 		return (0);
2728 	ksi.ksi_signo = sig;
2729 	if (ksi.ksi_code == SI_TIMER)
2730 		itimer_accept(p, ksi.ksi_timerid, &ksi);
2731 	action = ps->ps_sigact[_SIG_IDX(sig)];
2732 #ifdef KTRACE
2733 	if (KTRPOINT(td, KTR_PSIG))
2734 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
2735 		    &td->td_oldsigmask : &td->td_sigmask, 0);
2736 #endif
2737 	if (p->p_stops & S_SIG) {
2738 		mtx_unlock(&ps->ps_mtx);
2739 		stopevent(p, S_SIG, sig);
2740 		mtx_lock(&ps->ps_mtx);
2741 	}
2742 
2743 	if (action == SIG_DFL) {
2744 		/*
2745 		 * Default action, where the default is to kill
2746 		 * the process.  (Other cases were ignored above.)
2747 		 */
2748 		mtx_unlock(&ps->ps_mtx);
2749 		sigexit(td, sig);
2750 		/* NOTREACHED */
2751 	} else {
2752 		/*
2753 		 * If we get here, the signal must be caught.
2754 		 */
2755 		KASSERT(action != SIG_IGN && !SIGISMEMBER(td->td_sigmask, sig),
2756 		    ("postsig action"));
2757 		/*
2758 		 * Set the new mask value and also defer further
2759 		 * occurrences of this signal.
2760 		 *
2761 		 * Special case: user has done a sigsuspend.  Here the
2762 		 * current mask is not of interest, but rather the
2763 		 * mask from before the sigsuspend is what we want
2764 		 * restored after the signal processing is completed.
2765 		 */
2766 		if (td->td_pflags & TDP_OLDMASK) {
2767 			returnmask = td->td_oldsigmask;
2768 			td->td_pflags &= ~TDP_OLDMASK;
2769 		} else
2770 			returnmask = td->td_sigmask;
2771 
2772 		mask = ps->ps_catchmask[_SIG_IDX(sig)];
2773 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
2774 			SIGADDSET(mask, sig);
2775 		kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
2776 		    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
2777 
2778 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
2779 			/*
2780 			 * See kern_sigaction() for origin of this code.
2781 			 */
2782 			SIGDELSET(ps->ps_sigcatch, sig);
2783 			if (sig != SIGCONT &&
2784 			    sigprop(sig) & SA_IGNORE)
2785 				SIGADDSET(ps->ps_sigignore, sig);
2786 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2787 		}
2788 		td->td_ru.ru_nsignals++;
2789 		if (p->p_sig == sig) {
2790 			p->p_code = 0;
2791 			p->p_sig = 0;
2792 		}
2793 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
2794 	}
2795 	return (1);
2796 }
2797 
2798 /*
2799  * Kill the current process for stated reason.
2800  */
2801 void
2802 killproc(p, why)
2803 	struct proc *p;
2804 	char *why;
2805 {
2806 
2807 	PROC_LOCK_ASSERT(p, MA_OWNED);
2808 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)",
2809 		p, p->p_pid, p->p_comm);
2810 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n", p->p_pid, p->p_comm,
2811 		p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2812 	p->p_flag |= P_WKILLED;
2813 	psignal(p, SIGKILL);
2814 }
2815 
2816 /*
2817  * Force the current process to exit with the specified signal, dumping core
2818  * if appropriate.  We bypass the normal tests for masked and caught signals,
2819  * allowing unrecoverable failures to terminate the process without changing
2820  * signal state.  Mark the accounting record with the signal termination.
2821  * If dumping core, save the signal number for the debugger.  Calls exit and
2822  * does not return.
2823  */
2824 void
2825 sigexit(td, sig)
2826 	struct thread *td;
2827 	int sig;
2828 {
2829 	struct proc *p = td->td_proc;
2830 
2831 	PROC_LOCK_ASSERT(p, MA_OWNED);
2832 	p->p_acflag |= AXSIG;
2833 	/*
2834 	 * We must be single-threading to generate a core dump.  This
2835 	 * ensures that the registers in the core file are up-to-date.
2836 	 * Also, the ELF dump handler assumes that the thread list doesn't
2837 	 * change out from under it.
2838 	 *
2839 	 * XXX If another thread attempts to single-thread before us
2840 	 *     (e.g. via fork()), we won't get a dump at all.
2841 	 */
2842 	if ((sigprop(sig) & SA_CORE) && (thread_single(SINGLE_NO_EXIT) == 0)) {
2843 		p->p_sig = sig;
2844 		/*
2845 		 * Log signals which would cause core dumps
2846 		 * (Log as LOG_INFO to appease those who don't want
2847 		 * these messages.)
2848 		 * XXX : Todo, as well as euid, write out ruid too
2849 		 * Note that coredump() drops proc lock.
2850 		 */
2851 		if (coredump(td) == 0)
2852 			sig |= WCOREFLAG;
2853 		if (kern_logsigexit)
2854 			log(LOG_INFO,
2855 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2856 			    p->p_pid, p->p_comm,
2857 			    td->td_ucred ? td->td_ucred->cr_uid : -1,
2858 			    sig &~ WCOREFLAG,
2859 			    sig & WCOREFLAG ? " (core dumped)" : "");
2860 	} else
2861 		PROC_UNLOCK(p);
2862 	exit1(td, W_EXITCODE(0, sig));
2863 	/* NOTREACHED */
2864 }
2865 
2866 /*
2867  * Send queued SIGCHLD to parent when child process's state
2868  * is changed.
2869  */
2870 static void
2871 sigparent(struct proc *p, int reason, int status)
2872 {
2873 	PROC_LOCK_ASSERT(p, MA_OWNED);
2874 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
2875 
2876 	if (p->p_ksi != NULL) {
2877 		p->p_ksi->ksi_signo  = SIGCHLD;
2878 		p->p_ksi->ksi_code   = reason;
2879 		p->p_ksi->ksi_status = status;
2880 		p->p_ksi->ksi_pid    = p->p_pid;
2881 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
2882 		if (KSI_ONQ(p->p_ksi))
2883 			return;
2884 	}
2885 	tdsignal(p->p_pptr, NULL, SIGCHLD, p->p_ksi);
2886 }
2887 
2888 static void
2889 childproc_jobstate(struct proc *p, int reason, int status)
2890 {
2891 	struct sigacts *ps;
2892 
2893 	PROC_LOCK_ASSERT(p, MA_OWNED);
2894 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
2895 
2896 	/*
2897 	 * Wake up parent sleeping in kern_wait(), also send
2898 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
2899 	 * that parent will awake, because parent may masked
2900 	 * the signal.
2901 	 */
2902 	p->p_pptr->p_flag |= P_STATCHILD;
2903 	wakeup(p->p_pptr);
2904 
2905 	ps = p->p_pptr->p_sigacts;
2906 	mtx_lock(&ps->ps_mtx);
2907 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
2908 		mtx_unlock(&ps->ps_mtx);
2909 		sigparent(p, reason, status);
2910 	} else
2911 		mtx_unlock(&ps->ps_mtx);
2912 }
2913 
2914 void
2915 childproc_stopped(struct proc *p, int reason)
2916 {
2917 	childproc_jobstate(p, reason, p->p_xstat);
2918 }
2919 
2920 void
2921 childproc_continued(struct proc *p)
2922 {
2923 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
2924 }
2925 
2926 void
2927 childproc_exited(struct proc *p)
2928 {
2929 	int reason;
2930 	int status = p->p_xstat; /* convert to int */
2931 
2932 	reason = CLD_EXITED;
2933 	if (WCOREDUMP(status))
2934 		reason = CLD_DUMPED;
2935 	else if (WIFSIGNALED(status))
2936 		reason = CLD_KILLED;
2937 	/*
2938 	 * XXX avoid calling wakeup(p->p_pptr), the work is
2939 	 * done in exit1().
2940 	 */
2941 	sigparent(p, reason, status);
2942 }
2943 
2944 /*
2945  * We only have 1 character for the core count in the format
2946  * string, so the range will be 0-9
2947  */
2948 #define MAX_NUM_CORES 10
2949 static int num_cores = 5;
2950 
2951 static int
2952 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
2953 {
2954 	int error;
2955 	int new_val;
2956 
2957 	error = sysctl_handle_int(oidp, &new_val, 0, req);
2958 	if (error != 0 || req->newptr == NULL)
2959 		return (error);
2960 	if (new_val > MAX_NUM_CORES)
2961 		new_val = MAX_NUM_CORES;
2962 	if (new_val < 0)
2963 		new_val = 0;
2964 	num_cores = new_val;
2965 	return (0);
2966 }
2967 SYSCTL_PROC(_debug, OID_AUTO, ncores, CTLTYPE_INT|CTLFLAG_RW,
2968 	    0, sizeof(int), sysctl_debug_num_cores_check, "I", "");
2969 
2970 #if defined(COMPRESS_USER_CORES)
2971 int compress_user_cores = 1;
2972 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores, CTLFLAG_RW,
2973         &compress_user_cores, 0, "");
2974 
2975 int compress_user_cores_gzlevel = -1; /* default level */
2976 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_gzlevel, CTLFLAG_RW,
2977     &compress_user_cores_gzlevel, -1, "user core gz compression level");
2978 
2979 #define GZ_SUFFIX	".gz"
2980 #define GZ_SUFFIX_LEN	3
2981 #endif
2982 
2983 static char corefilename[MAXPATHLEN] = {"%N.core"};
2984 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2985 	      sizeof(corefilename), "process corefile name format string");
2986 
2987 /*
2988  * expand_name(name, uid, pid, td, compress)
2989  * Expand the name described in corefilename, using name, uid, and pid.
2990  * corefilename is a printf-like string, with three format specifiers:
2991  *	%N	name of process ("name")
2992  *	%P	process id (pid)
2993  *	%U	user id (uid)
2994  * For example, "%N.core" is the default; they can be disabled completely
2995  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2996  * This is controlled by the sysctl variable kern.corefile (see above).
2997  */
2998 static char *
2999 expand_name(const char *name, uid_t uid, pid_t pid, struct thread *td,
3000     int compress)
3001 {
3002 	struct sbuf sb;
3003 	const char *format;
3004 	char *temp;
3005 	size_t i;
3006 	int indexpos;
3007 	char *hostname;
3008 
3009 	hostname = NULL;
3010 	format = corefilename;
3011 	temp = malloc(MAXPATHLEN, M_TEMP, M_NOWAIT | M_ZERO);
3012 	if (temp == NULL)
3013 		return (NULL);
3014 	indexpos = -1;
3015 	(void)sbuf_new(&sb, temp, MAXPATHLEN, SBUF_FIXEDLEN);
3016 	for (i = 0; format[i]; i++) {
3017 		switch (format[i]) {
3018 		case '%':	/* Format character */
3019 			i++;
3020 			switch (format[i]) {
3021 			case '%':
3022 				sbuf_putc(&sb, '%');
3023 				break;
3024 			case 'H':	/* hostname */
3025 				if (hostname == NULL) {
3026 					hostname = malloc(MAXHOSTNAMELEN,
3027 					    M_TEMP, M_NOWAIT);
3028 					if (hostname == NULL) {
3029 						log(LOG_ERR,
3030 						    "pid %ld (%s), uid (%lu): "
3031 						    "unable to alloc memory "
3032 						    "for corefile hostname\n",
3033 						    (long)pid, name,
3034 						    (u_long)uid);
3035                                                 goto nomem;
3036                                         }
3037                                 }
3038 				getcredhostname(td->td_ucred, hostname,
3039 				    MAXHOSTNAMELEN);
3040 				sbuf_printf(&sb, "%s", hostname);
3041 				break;
3042 			case 'I':       /* autoincrementing index */
3043 				sbuf_printf(&sb, "0");
3044 				indexpos = sbuf_len(&sb) - 1;
3045 				break;
3046 			case 'N':	/* process name */
3047 				sbuf_printf(&sb, "%s", name);
3048 				break;
3049 			case 'P':	/* process id */
3050 				sbuf_printf(&sb, "%u", pid);
3051 				break;
3052 			case 'U':	/* user id */
3053 				sbuf_printf(&sb, "%u", uid);
3054 				break;
3055 			default:
3056 			  	log(LOG_ERR,
3057 				    "Unknown format character %c in "
3058 				    "corename `%s'\n", format[i], format);
3059 			}
3060 			break;
3061 		default:
3062 			sbuf_putc(&sb, format[i]);
3063 		}
3064 	}
3065 	free(hostname, M_TEMP);
3066 #ifdef COMPRESS_USER_CORES
3067 	if (compress) {
3068 		sbuf_printf(&sb, GZ_SUFFIX);
3069 	}
3070 #endif
3071 	if (sbuf_overflowed(&sb)) {
3072 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3073 		    "long\n", (long)pid, name, (u_long)uid);
3074 nomem:
3075 		sbuf_delete(&sb);
3076 		free(temp, M_TEMP);
3077 		return (NULL);
3078 	}
3079 	sbuf_finish(&sb);
3080 	sbuf_delete(&sb);
3081 
3082 	/*
3083 	 * If the core format has a %I in it, then we need to check
3084 	 * for existing corefiles before returning a name.
3085 	 * To do this we iterate over 0..num_cores to find a
3086 	 * non-existing core file name to use.
3087 	 */
3088 	if (indexpos != -1) {
3089 		struct nameidata nd;
3090 		int error, n;
3091 		int flags = O_CREAT | O_EXCL | FWRITE | O_NOFOLLOW;
3092 		int cmode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP;
3093 		int vfslocked;
3094 
3095 		for (n = 0; n < num_cores; n++) {
3096 			temp[indexpos] = '0' + n;
3097 			NDINIT(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE,
3098 			    temp, td);
3099 			error = vn_open(&nd, &flags, cmode, NULL);
3100 			if (error) {
3101 				if (error == EEXIST) {
3102 					continue;
3103 				}
3104 				log(LOG_ERR,
3105 				    "pid %d (%s), uid (%u):  Path `%s' failed "
3106                                     "on initial open test, error = %d\n",
3107 				    pid, name, uid, temp, error);
3108 				free(temp, M_TEMP);
3109 				return (NULL);
3110 			}
3111 			vfslocked = NDHASGIANT(&nd);
3112 			NDFREE(&nd, NDF_ONLY_PNBUF);
3113 			VOP_UNLOCK(nd.ni_vp, 0);
3114 			error = vn_close(nd.ni_vp, FWRITE, td->td_ucred, td);
3115 			VFS_UNLOCK_GIANT(vfslocked);
3116 			if (error) {
3117 				log(LOG_ERR,
3118 				    "pid %d (%s), uid (%u):  Path `%s' failed "
3119                                     "on close after initial open test, "
3120                                     "error = %d\n",
3121 				    pid, name, uid, temp, error);
3122 				free(temp, M_TEMP);
3123 				return (NULL);
3124 			}
3125 			break;
3126 		}
3127 	}
3128 	return (temp);
3129 }
3130 
3131 /*
3132  * Dump a process' core.  The main routine does some
3133  * policy checking, and creates the name of the coredump;
3134  * then it passes on a vnode and a size limit to the process-specific
3135  * coredump routine if there is one; if there _is not_ one, it returns
3136  * ENOSYS; otherwise it returns the error from the process-specific routine.
3137  */
3138 
3139 static int
3140 coredump(struct thread *td)
3141 {
3142 	struct proc *p = td->td_proc;
3143 	register struct vnode *vp;
3144 	register struct ucred *cred = td->td_ucred;
3145 	struct flock lf;
3146 	struct nameidata nd;
3147 	struct vattr vattr;
3148 	int error, error1, flags, locked;
3149 	struct mount *mp;
3150 	char *name;			/* name of corefile */
3151 	off_t limit;
3152 	int vfslocked;
3153 	int compress;
3154 
3155 #ifdef COMPRESS_USER_CORES
3156 	compress = compress_user_cores;
3157 #else
3158 	compress = 0;
3159 #endif
3160 	PROC_LOCK_ASSERT(p, MA_OWNED);
3161 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3162 	_STOPEVENT(p, S_CORE, 0);
3163 
3164 	name = expand_name(p->p_comm, td->td_ucred->cr_uid, p->p_pid, td,
3165 	    compress);
3166 	if (name == NULL) {
3167 		PROC_UNLOCK(p);
3168 #ifdef AUDIT
3169 		audit_proc_coredump(td, NULL, EINVAL);
3170 #endif
3171 		return (EINVAL);
3172 	}
3173 	if (((sugid_coredump == 0) && p->p_flag & P_SUGID) || do_coredump == 0) {
3174 		PROC_UNLOCK(p);
3175 #ifdef AUDIT
3176 		audit_proc_coredump(td, name, EFAULT);
3177 #endif
3178 		free(name, M_TEMP);
3179 		return (EFAULT);
3180 	}
3181 
3182 	/*
3183 	 * Note that the bulk of limit checking is done after
3184 	 * the corefile is created.  The exception is if the limit
3185 	 * for corefiles is 0, in which case we don't bother
3186 	 * creating the corefile at all.  This layout means that
3187 	 * a corefile is truncated instead of not being created,
3188 	 * if it is larger than the limit.
3189 	 */
3190 	limit = (off_t)lim_cur(p, RLIMIT_CORE);
3191 	PROC_UNLOCK(p);
3192 	if (limit == 0) {
3193 #ifdef AUDIT
3194 		audit_proc_coredump(td, name, EFBIG);
3195 #endif
3196 		free(name, M_TEMP);
3197 		return (EFBIG);
3198 	}
3199 
3200 restart:
3201 	NDINIT(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, name, td);
3202 	flags = O_CREAT | FWRITE | O_NOFOLLOW;
3203 	error = vn_open_cred(&nd, &flags, S_IRUSR | S_IWUSR, VN_OPEN_NOAUDIT,
3204 	    cred, NULL);
3205 	if (error) {
3206 #ifdef AUDIT
3207 		audit_proc_coredump(td, name, error);
3208 #endif
3209 		free(name, M_TEMP);
3210 		return (error);
3211 	}
3212 	vfslocked = NDHASGIANT(&nd);
3213 	NDFREE(&nd, NDF_ONLY_PNBUF);
3214 	vp = nd.ni_vp;
3215 
3216 	/* Don't dump to non-regular files or files with links. */
3217 	if (vp->v_type != VREG ||
3218 	    VOP_GETATTR(vp, &vattr, cred) || vattr.va_nlink != 1) {
3219 		VOP_UNLOCK(vp, 0);
3220 		error = EFAULT;
3221 		goto close;
3222 	}
3223 
3224 	VOP_UNLOCK(vp, 0);
3225 	lf.l_whence = SEEK_SET;
3226 	lf.l_start = 0;
3227 	lf.l_len = 0;
3228 	lf.l_type = F_WRLCK;
3229 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
3230 
3231 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
3232 		lf.l_type = F_UNLCK;
3233 		if (locked)
3234 			VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3235 		if ((error = vn_close(vp, FWRITE, cred, td)) != 0)
3236 			goto out;
3237 		if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0)
3238 			goto out;
3239 		VFS_UNLOCK_GIANT(vfslocked);
3240 		goto restart;
3241 	}
3242 
3243 	VATTR_NULL(&vattr);
3244 	vattr.va_size = 0;
3245 	if (set_core_nodump_flag)
3246 		vattr.va_flags = UF_NODUMP;
3247 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3248 	VOP_SETATTR(vp, &vattr, cred);
3249 	VOP_UNLOCK(vp, 0);
3250 	vn_finished_write(mp);
3251 	PROC_LOCK(p);
3252 	p->p_acflag |= ACORE;
3253 	PROC_UNLOCK(p);
3254 
3255 	error = p->p_sysent->sv_coredump ?
3256 	  p->p_sysent->sv_coredump(td, vp, limit, compress ? IMGACT_CORE_COMPRESS : 0) :
3257 	  ENOSYS;
3258 
3259 	if (locked) {
3260 		lf.l_type = F_UNLCK;
3261 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3262 	}
3263 close:
3264 	error1 = vn_close(vp, FWRITE, cred, td);
3265 	if (error == 0)
3266 		error = error1;
3267 out:
3268 #ifdef AUDIT
3269 	audit_proc_coredump(td, name, error);
3270 #endif
3271 	free(name, M_TEMP);
3272 	VFS_UNLOCK_GIANT(vfslocked);
3273 	return (error);
3274 }
3275 
3276 /*
3277  * Nonexistent system call-- signal process (may want to handle it).  Flag
3278  * error in case process won't see signal immediately (blocked or ignored).
3279  */
3280 #ifndef _SYS_SYSPROTO_H_
3281 struct nosys_args {
3282 	int	dummy;
3283 };
3284 #endif
3285 /* ARGSUSED */
3286 int
3287 nosys(td, args)
3288 	struct thread *td;
3289 	struct nosys_args *args;
3290 {
3291 	struct proc *p = td->td_proc;
3292 
3293 	PROC_LOCK(p);
3294 	psignal(p, SIGSYS);
3295 	PROC_UNLOCK(p);
3296 	return (ENOSYS);
3297 }
3298 
3299 /*
3300  * Send a SIGIO or SIGURG signal to a process or process group using stored
3301  * credentials rather than those of the current process.
3302  */
3303 void
3304 pgsigio(sigiop, sig, checkctty)
3305 	struct sigio **sigiop;
3306 	int sig, checkctty;
3307 {
3308 	ksiginfo_t ksi;
3309 	struct sigio *sigio;
3310 
3311 	ksiginfo_init(&ksi);
3312 	ksi.ksi_signo = sig;
3313 	ksi.ksi_code = SI_KERNEL;
3314 
3315 	SIGIO_LOCK();
3316 	sigio = *sigiop;
3317 	if (sigio == NULL) {
3318 		SIGIO_UNLOCK();
3319 		return;
3320 	}
3321 	if (sigio->sio_pgid > 0) {
3322 		PROC_LOCK(sigio->sio_proc);
3323 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
3324 			psignal(sigio->sio_proc, sig);
3325 		PROC_UNLOCK(sigio->sio_proc);
3326 	} else if (sigio->sio_pgid < 0) {
3327 		struct proc *p;
3328 
3329 		PGRP_LOCK(sigio->sio_pgrp);
3330 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
3331 			PROC_LOCK(p);
3332 			if (CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
3333 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
3334 				psignal(p, sig);
3335 			PROC_UNLOCK(p);
3336 		}
3337 		PGRP_UNLOCK(sigio->sio_pgrp);
3338 	}
3339 	SIGIO_UNLOCK();
3340 }
3341 
3342 static int
3343 filt_sigattach(struct knote *kn)
3344 {
3345 	struct proc *p = curproc;
3346 
3347 	kn->kn_ptr.p_proc = p;
3348 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
3349 
3350 	knlist_add(&p->p_klist, kn, 0);
3351 
3352 	return (0);
3353 }
3354 
3355 static void
3356 filt_sigdetach(struct knote *kn)
3357 {
3358 	struct proc *p = kn->kn_ptr.p_proc;
3359 
3360 	knlist_remove(&p->p_klist, kn, 0);
3361 }
3362 
3363 /*
3364  * signal knotes are shared with proc knotes, so we apply a mask to
3365  * the hint in order to differentiate them from process hints.  This
3366  * could be avoided by using a signal-specific knote list, but probably
3367  * isn't worth the trouble.
3368  */
3369 static int
3370 filt_signal(struct knote *kn, long hint)
3371 {
3372 
3373 	if (hint & NOTE_SIGNAL) {
3374 		hint &= ~NOTE_SIGNAL;
3375 
3376 		if (kn->kn_id == hint)
3377 			kn->kn_data++;
3378 	}
3379 	return (kn->kn_data != 0);
3380 }
3381 
3382 struct sigacts *
3383 sigacts_alloc(void)
3384 {
3385 	struct sigacts *ps;
3386 
3387 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
3388 	ps->ps_refcnt = 1;
3389 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
3390 	return (ps);
3391 }
3392 
3393 void
3394 sigacts_free(struct sigacts *ps)
3395 {
3396 
3397 	mtx_lock(&ps->ps_mtx);
3398 	ps->ps_refcnt--;
3399 	if (ps->ps_refcnt == 0) {
3400 		mtx_destroy(&ps->ps_mtx);
3401 		free(ps, M_SUBPROC);
3402 	} else
3403 		mtx_unlock(&ps->ps_mtx);
3404 }
3405 
3406 struct sigacts *
3407 sigacts_hold(struct sigacts *ps)
3408 {
3409 	mtx_lock(&ps->ps_mtx);
3410 	ps->ps_refcnt++;
3411 	mtx_unlock(&ps->ps_mtx);
3412 	return (ps);
3413 }
3414 
3415 void
3416 sigacts_copy(struct sigacts *dest, struct sigacts *src)
3417 {
3418 
3419 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
3420 	mtx_lock(&src->ps_mtx);
3421 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
3422 	mtx_unlock(&src->ps_mtx);
3423 }
3424 
3425 int
3426 sigacts_shared(struct sigacts *ps)
3427 {
3428 	int shared;
3429 
3430 	mtx_lock(&ps->ps_mtx);
3431 	shared = ps->ps_refcnt > 1;
3432 	mtx_unlock(&ps->ps_mtx);
3433 	return (shared);
3434 }
3435