xref: /freebsd/sys/kern/kern_sig.c (revision f18d3c411697ff46d85e579a72be54ca0cc67dd0)
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_ktrace.h"
42 #include "opt_core.h"
43 #include "opt_procdesc.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/capability.h>
51 #include <sys/condvar.h>
52 #include <sys/event.h>
53 #include <sys/fcntl.h>
54 #include <sys/imgact.h>
55 #include <sys/kernel.h>
56 #include <sys/ktr.h>
57 #include <sys/ktrace.h>
58 #include <sys/lock.h>
59 #include <sys/malloc.h>
60 #include <sys/mutex.h>
61 #include <sys/namei.h>
62 #include <sys/proc.h>
63 #include <sys/procdesc.h>
64 #include <sys/posix4.h>
65 #include <sys/pioctl.h>
66 #include <sys/racct.h>
67 #include <sys/resourcevar.h>
68 #include <sys/sdt.h>
69 #include <sys/sbuf.h>
70 #include <sys/sleepqueue.h>
71 #include <sys/smp.h>
72 #include <sys/stat.h>
73 #include <sys/sx.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysctl.h>
76 #include <sys/sysent.h>
77 #include <sys/syslog.h>
78 #include <sys/sysproto.h>
79 #include <sys/timers.h>
80 #include <sys/unistd.h>
81 #include <sys/wait.h>
82 #include <vm/vm.h>
83 #include <vm/vm_extern.h>
84 #include <vm/uma.h>
85 
86 #include <sys/jail.h>
87 
88 #include <machine/cpu.h>
89 
90 #include <security/audit/audit.h>
91 
92 #define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
93 
94 SDT_PROVIDER_DECLARE(proc);
95 SDT_PROBE_DEFINE3(proc, kernel, , signal_send, signal-send, "struct thread *",
96     "struct proc *", "int");
97 SDT_PROBE_DEFINE2(proc, kernel, , signal_clear, signal-clear, "int",
98     "ksiginfo_t *");
99 SDT_PROBE_DEFINE3(proc, kernel, , signal_discard, signal-discard,
100     "struct thread *", "struct proc *", "int");
101 
102 static int	coredump(struct thread *);
103 static int	killpg1(struct thread *td, int sig, int pgid, int all,
104 		    ksiginfo_t *ksi);
105 static int	issignal(struct thread *td);
106 static int	sigprop(int sig);
107 static void	tdsigwakeup(struct thread *, int, sig_t, int);
108 static void	sig_suspend_threads(struct thread *, struct proc *, int);
109 static int	filt_sigattach(struct knote *kn);
110 static void	filt_sigdetach(struct knote *kn);
111 static int	filt_signal(struct knote *kn, long hint);
112 static struct thread *sigtd(struct proc *p, int sig, int prop);
113 static void	sigqueue_start(void);
114 
115 static uma_zone_t	ksiginfo_zone = NULL;
116 struct filterops sig_filtops = {
117 	.f_isfd = 0,
118 	.f_attach = filt_sigattach,
119 	.f_detach = filt_sigdetach,
120 	.f_event = filt_signal,
121 };
122 
123 static int	kern_logsigexit = 1;
124 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
125     &kern_logsigexit, 0,
126     "Log processes quitting on abnormal signals to syslog(3)");
127 
128 static int	kern_forcesigexit = 1;
129 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
130     &kern_forcesigexit, 0, "Force trap signal to be handled");
131 
132 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW, 0,
133     "POSIX real time signal");
134 
135 static int	max_pending_per_proc = 128;
136 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
137     &max_pending_per_proc, 0, "Max pending signals per proc");
138 
139 static int	preallocate_siginfo = 1024;
140 TUNABLE_INT("kern.sigqueue.preallocate", &preallocate_siginfo);
141 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RD,
142     &preallocate_siginfo, 0, "Preallocated signal memory size");
143 
144 static int	signal_overflow = 0;
145 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
146     &signal_overflow, 0, "Number of signals overflew");
147 
148 static int	signal_alloc_fail = 0;
149 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
150     &signal_alloc_fail, 0, "signals failed to be allocated");
151 
152 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
153 
154 /*
155  * Policy -- Can ucred cr1 send SIGIO to process cr2?
156  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
157  * in the right situations.
158  */
159 #define CANSIGIO(cr1, cr2) \
160 	((cr1)->cr_uid == 0 || \
161 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
162 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
163 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
164 	    (cr1)->cr_uid == (cr2)->cr_uid)
165 
166 static int	sugid_coredump;
167 TUNABLE_INT("kern.sugid_coredump", &sugid_coredump);
168 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
169     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
170 
171 static int	capmode_coredump;
172 TUNABLE_INT("kern.capmode_coredump", &capmode_coredump);
173 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RW,
174     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
175 
176 static int	do_coredump = 1;
177 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
178 	&do_coredump, 0, "Enable/Disable coredumps");
179 
180 static int	set_core_nodump_flag = 0;
181 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
182 	0, "Enable setting the NODUMP flag on coredump files");
183 
184 /*
185  * Signal properties and actions.
186  * The array below categorizes the signals and their default actions
187  * according to the following properties:
188  */
189 #define	SA_KILL		0x01		/* terminates process by default */
190 #define	SA_CORE		0x02		/* ditto and coredumps */
191 #define	SA_STOP		0x04		/* suspend process */
192 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
193 #define	SA_IGNORE	0x10		/* ignore by default */
194 #define	SA_CONT		0x20		/* continue if suspended */
195 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
196 
197 static int sigproptbl[NSIG] = {
198 	SA_KILL,			/* SIGHUP */
199 	SA_KILL,			/* SIGINT */
200 	SA_KILL|SA_CORE,		/* SIGQUIT */
201 	SA_KILL|SA_CORE,		/* SIGILL */
202 	SA_KILL|SA_CORE,		/* SIGTRAP */
203 	SA_KILL|SA_CORE,		/* SIGABRT */
204 	SA_KILL|SA_CORE,		/* SIGEMT */
205 	SA_KILL|SA_CORE,		/* SIGFPE */
206 	SA_KILL,			/* SIGKILL */
207 	SA_KILL|SA_CORE,		/* SIGBUS */
208 	SA_KILL|SA_CORE,		/* SIGSEGV */
209 	SA_KILL|SA_CORE,		/* SIGSYS */
210 	SA_KILL,			/* SIGPIPE */
211 	SA_KILL,			/* SIGALRM */
212 	SA_KILL,			/* SIGTERM */
213 	SA_IGNORE,			/* SIGURG */
214 	SA_STOP,			/* SIGSTOP */
215 	SA_STOP|SA_TTYSTOP,		/* SIGTSTP */
216 	SA_IGNORE|SA_CONT,		/* SIGCONT */
217 	SA_IGNORE,			/* SIGCHLD */
218 	SA_STOP|SA_TTYSTOP,		/* SIGTTIN */
219 	SA_STOP|SA_TTYSTOP,		/* SIGTTOU */
220 	SA_IGNORE,			/* SIGIO */
221 	SA_KILL,			/* SIGXCPU */
222 	SA_KILL,			/* SIGXFSZ */
223 	SA_KILL,			/* SIGVTALRM */
224 	SA_KILL,			/* SIGPROF */
225 	SA_IGNORE,			/* SIGWINCH  */
226 	SA_IGNORE,			/* SIGINFO */
227 	SA_KILL,			/* SIGUSR1 */
228 	SA_KILL,			/* SIGUSR2 */
229 };
230 
231 static void reschedule_signals(struct proc *p, sigset_t block, int flags);
232 
233 static void
234 sigqueue_start(void)
235 {
236 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
237 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
238 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
239 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
240 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
241 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
242 }
243 
244 ksiginfo_t *
245 ksiginfo_alloc(int wait)
246 {
247 	int flags;
248 
249 	flags = M_ZERO;
250 	if (! wait)
251 		flags |= M_NOWAIT;
252 	if (ksiginfo_zone != NULL)
253 		return ((ksiginfo_t *)uma_zalloc(ksiginfo_zone, flags));
254 	return (NULL);
255 }
256 
257 void
258 ksiginfo_free(ksiginfo_t *ksi)
259 {
260 	uma_zfree(ksiginfo_zone, ksi);
261 }
262 
263 static __inline int
264 ksiginfo_tryfree(ksiginfo_t *ksi)
265 {
266 	if (!(ksi->ksi_flags & KSI_EXT)) {
267 		uma_zfree(ksiginfo_zone, ksi);
268 		return (1);
269 	}
270 	return (0);
271 }
272 
273 void
274 sigqueue_init(sigqueue_t *list, struct proc *p)
275 {
276 	SIGEMPTYSET(list->sq_signals);
277 	SIGEMPTYSET(list->sq_kill);
278 	TAILQ_INIT(&list->sq_list);
279 	list->sq_proc = p;
280 	list->sq_flags = SQ_INIT;
281 }
282 
283 /*
284  * Get a signal's ksiginfo.
285  * Return:
286  *	0	-	signal not found
287  *	others	-	signal number
288  */
289 static int
290 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
291 {
292 	struct proc *p = sq->sq_proc;
293 	struct ksiginfo *ksi, *next;
294 	int count = 0;
295 
296 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
297 
298 	if (!SIGISMEMBER(sq->sq_signals, signo))
299 		return (0);
300 
301 	if (SIGISMEMBER(sq->sq_kill, signo)) {
302 		count++;
303 		SIGDELSET(sq->sq_kill, signo);
304 	}
305 
306 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
307 		if (ksi->ksi_signo == signo) {
308 			if (count == 0) {
309 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
310 				ksi->ksi_sigq = NULL;
311 				ksiginfo_copy(ksi, si);
312 				if (ksiginfo_tryfree(ksi) && p != NULL)
313 					p->p_pendingcnt--;
314 			}
315 			if (++count > 1)
316 				break;
317 		}
318 	}
319 
320 	if (count <= 1)
321 		SIGDELSET(sq->sq_signals, signo);
322 	si->ksi_signo = signo;
323 	return (signo);
324 }
325 
326 void
327 sigqueue_take(ksiginfo_t *ksi)
328 {
329 	struct ksiginfo *kp;
330 	struct proc	*p;
331 	sigqueue_t	*sq;
332 
333 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
334 		return;
335 
336 	p = sq->sq_proc;
337 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
338 	ksi->ksi_sigq = NULL;
339 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
340 		p->p_pendingcnt--;
341 
342 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
343 	     kp = TAILQ_NEXT(kp, ksi_link)) {
344 		if (kp->ksi_signo == ksi->ksi_signo)
345 			break;
346 	}
347 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo))
348 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
349 }
350 
351 static int
352 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
353 {
354 	struct proc *p = sq->sq_proc;
355 	struct ksiginfo *ksi;
356 	int ret = 0;
357 
358 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
359 
360 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
361 		SIGADDSET(sq->sq_kill, signo);
362 		goto out_set_bit;
363 	}
364 
365 	/* directly insert the ksi, don't copy it */
366 	if (si->ksi_flags & KSI_INS) {
367 		if (si->ksi_flags & KSI_HEAD)
368 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
369 		else
370 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
371 		si->ksi_sigq = sq;
372 		goto out_set_bit;
373 	}
374 
375 	if (__predict_false(ksiginfo_zone == NULL)) {
376 		SIGADDSET(sq->sq_kill, signo);
377 		goto out_set_bit;
378 	}
379 
380 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
381 		signal_overflow++;
382 		ret = EAGAIN;
383 	} else if ((ksi = ksiginfo_alloc(0)) == NULL) {
384 		signal_alloc_fail++;
385 		ret = EAGAIN;
386 	} else {
387 		if (p != NULL)
388 			p->p_pendingcnt++;
389 		ksiginfo_copy(si, ksi);
390 		ksi->ksi_signo = signo;
391 		if (si->ksi_flags & KSI_HEAD)
392 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
393 		else
394 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
395 		ksi->ksi_sigq = sq;
396 	}
397 
398 	if ((si->ksi_flags & KSI_TRAP) != 0 ||
399 	    (si->ksi_flags & KSI_SIGQ) == 0) {
400 		if (ret != 0)
401 			SIGADDSET(sq->sq_kill, signo);
402 		ret = 0;
403 		goto out_set_bit;
404 	}
405 
406 	if (ret != 0)
407 		return (ret);
408 
409 out_set_bit:
410 	SIGADDSET(sq->sq_signals, signo);
411 	return (ret);
412 }
413 
414 void
415 sigqueue_flush(sigqueue_t *sq)
416 {
417 	struct proc *p = sq->sq_proc;
418 	ksiginfo_t *ksi;
419 
420 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
421 
422 	if (p != NULL)
423 		PROC_LOCK_ASSERT(p, MA_OWNED);
424 
425 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
426 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
427 		ksi->ksi_sigq = NULL;
428 		if (ksiginfo_tryfree(ksi) && p != NULL)
429 			p->p_pendingcnt--;
430 	}
431 
432 	SIGEMPTYSET(sq->sq_signals);
433 	SIGEMPTYSET(sq->sq_kill);
434 }
435 
436 static void
437 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
438 {
439 	sigset_t tmp;
440 	struct proc *p1, *p2;
441 	ksiginfo_t *ksi, *next;
442 
443 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
444 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
445 	p1 = src->sq_proc;
446 	p2 = dst->sq_proc;
447 	/* Move siginfo to target list */
448 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
449 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
450 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
451 			if (p1 != NULL)
452 				p1->p_pendingcnt--;
453 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
454 			ksi->ksi_sigq = dst;
455 			if (p2 != NULL)
456 				p2->p_pendingcnt++;
457 		}
458 	}
459 
460 	/* Move pending bits to target list */
461 	tmp = src->sq_kill;
462 	SIGSETAND(tmp, *set);
463 	SIGSETOR(dst->sq_kill, tmp);
464 	SIGSETNAND(src->sq_kill, tmp);
465 
466 	tmp = src->sq_signals;
467 	SIGSETAND(tmp, *set);
468 	SIGSETOR(dst->sq_signals, tmp);
469 	SIGSETNAND(src->sq_signals, tmp);
470 }
471 
472 #if 0
473 static void
474 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
475 {
476 	sigset_t set;
477 
478 	SIGEMPTYSET(set);
479 	SIGADDSET(set, signo);
480 	sigqueue_move_set(src, dst, &set);
481 }
482 #endif
483 
484 static void
485 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
486 {
487 	struct proc *p = sq->sq_proc;
488 	ksiginfo_t *ksi, *next;
489 
490 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
491 
492 	/* Remove siginfo queue */
493 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
494 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
495 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
496 			ksi->ksi_sigq = NULL;
497 			if (ksiginfo_tryfree(ksi) && p != NULL)
498 				p->p_pendingcnt--;
499 		}
500 	}
501 	SIGSETNAND(sq->sq_kill, *set);
502 	SIGSETNAND(sq->sq_signals, *set);
503 }
504 
505 void
506 sigqueue_delete(sigqueue_t *sq, int signo)
507 {
508 	sigset_t set;
509 
510 	SIGEMPTYSET(set);
511 	SIGADDSET(set, signo);
512 	sigqueue_delete_set(sq, &set);
513 }
514 
515 /* Remove a set of signals for a process */
516 static void
517 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
518 {
519 	sigqueue_t worklist;
520 	struct thread *td0;
521 
522 	PROC_LOCK_ASSERT(p, MA_OWNED);
523 
524 	sigqueue_init(&worklist, NULL);
525 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
526 
527 	FOREACH_THREAD_IN_PROC(p, td0)
528 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
529 
530 	sigqueue_flush(&worklist);
531 }
532 
533 void
534 sigqueue_delete_proc(struct proc *p, int signo)
535 {
536 	sigset_t set;
537 
538 	SIGEMPTYSET(set);
539 	SIGADDSET(set, signo);
540 	sigqueue_delete_set_proc(p, &set);
541 }
542 
543 static void
544 sigqueue_delete_stopmask_proc(struct proc *p)
545 {
546 	sigset_t set;
547 
548 	SIGEMPTYSET(set);
549 	SIGADDSET(set, SIGSTOP);
550 	SIGADDSET(set, SIGTSTP);
551 	SIGADDSET(set, SIGTTIN);
552 	SIGADDSET(set, SIGTTOU);
553 	sigqueue_delete_set_proc(p, &set);
554 }
555 
556 /*
557  * Determine signal that should be delivered to thread td, the current
558  * thread, 0 if none.  If there is a pending stop signal with default
559  * action, the process stops in issignal().
560  */
561 int
562 cursig(struct thread *td)
563 {
564 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
565 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
566 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
567 	return (SIGPENDING(td) ? issignal(td) : 0);
568 }
569 
570 /*
571  * Arrange for ast() to handle unmasked pending signals on return to user
572  * mode.  This must be called whenever a signal is added to td_sigqueue or
573  * unmasked in td_sigmask.
574  */
575 void
576 signotify(struct thread *td)
577 {
578 	struct proc *p;
579 
580 	p = td->td_proc;
581 
582 	PROC_LOCK_ASSERT(p, MA_OWNED);
583 
584 	if (SIGPENDING(td)) {
585 		thread_lock(td);
586 		td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING;
587 		thread_unlock(td);
588 	}
589 }
590 
591 int
592 sigonstack(size_t sp)
593 {
594 	struct thread *td = curthread;
595 
596 	return ((td->td_pflags & TDP_ALTSTACK) ?
597 #if defined(COMPAT_43)
598 	    ((td->td_sigstk.ss_size == 0) ?
599 		(td->td_sigstk.ss_flags & SS_ONSTACK) :
600 		((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size))
601 #else
602 	    ((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size)
603 #endif
604 	    : 0);
605 }
606 
607 static __inline int
608 sigprop(int sig)
609 {
610 
611 	if (sig > 0 && sig < NSIG)
612 		return (sigproptbl[_SIG_IDX(sig)]);
613 	return (0);
614 }
615 
616 int
617 sig_ffs(sigset_t *set)
618 {
619 	int i;
620 
621 	for (i = 0; i < _SIG_WORDS; i++)
622 		if (set->__bits[i])
623 			return (ffs(set->__bits[i]) + (i * 32));
624 	return (0);
625 }
626 
627 /*
628  * kern_sigaction
629  * sigaction
630  * freebsd4_sigaction
631  * osigaction
632  */
633 int
634 kern_sigaction(td, sig, act, oact, flags)
635 	struct thread *td;
636 	register int sig;
637 	struct sigaction *act, *oact;
638 	int flags;
639 {
640 	struct sigacts *ps;
641 	struct proc *p = td->td_proc;
642 
643 	if (!_SIG_VALID(sig))
644 		return (EINVAL);
645 
646 	PROC_LOCK(p);
647 	ps = p->p_sigacts;
648 	mtx_lock(&ps->ps_mtx);
649 	if (oact) {
650 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
651 		oact->sa_flags = 0;
652 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
653 			oact->sa_flags |= SA_ONSTACK;
654 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
655 			oact->sa_flags |= SA_RESTART;
656 		if (SIGISMEMBER(ps->ps_sigreset, sig))
657 			oact->sa_flags |= SA_RESETHAND;
658 		if (SIGISMEMBER(ps->ps_signodefer, sig))
659 			oact->sa_flags |= SA_NODEFER;
660 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
661 			oact->sa_flags |= SA_SIGINFO;
662 			oact->sa_sigaction =
663 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
664 		} else
665 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
666 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
667 			oact->sa_flags |= SA_NOCLDSTOP;
668 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
669 			oact->sa_flags |= SA_NOCLDWAIT;
670 	}
671 	if (act) {
672 		if ((sig == SIGKILL || sig == SIGSTOP) &&
673 		    act->sa_handler != SIG_DFL) {
674 			mtx_unlock(&ps->ps_mtx);
675 			PROC_UNLOCK(p);
676 			return (EINVAL);
677 		}
678 
679 		/*
680 		 * Change setting atomically.
681 		 */
682 
683 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
684 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
685 		if (act->sa_flags & SA_SIGINFO) {
686 			ps->ps_sigact[_SIG_IDX(sig)] =
687 			    (__sighandler_t *)act->sa_sigaction;
688 			SIGADDSET(ps->ps_siginfo, sig);
689 		} else {
690 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
691 			SIGDELSET(ps->ps_siginfo, sig);
692 		}
693 		if (!(act->sa_flags & SA_RESTART))
694 			SIGADDSET(ps->ps_sigintr, sig);
695 		else
696 			SIGDELSET(ps->ps_sigintr, sig);
697 		if (act->sa_flags & SA_ONSTACK)
698 			SIGADDSET(ps->ps_sigonstack, sig);
699 		else
700 			SIGDELSET(ps->ps_sigonstack, sig);
701 		if (act->sa_flags & SA_RESETHAND)
702 			SIGADDSET(ps->ps_sigreset, sig);
703 		else
704 			SIGDELSET(ps->ps_sigreset, sig);
705 		if (act->sa_flags & SA_NODEFER)
706 			SIGADDSET(ps->ps_signodefer, sig);
707 		else
708 			SIGDELSET(ps->ps_signodefer, sig);
709 		if (sig == SIGCHLD) {
710 			if (act->sa_flags & SA_NOCLDSTOP)
711 				ps->ps_flag |= PS_NOCLDSTOP;
712 			else
713 				ps->ps_flag &= ~PS_NOCLDSTOP;
714 			if (act->sa_flags & SA_NOCLDWAIT) {
715 				/*
716 				 * Paranoia: since SA_NOCLDWAIT is implemented
717 				 * by reparenting the dying child to PID 1 (and
718 				 * trust it to reap the zombie), PID 1 itself
719 				 * is forbidden to set SA_NOCLDWAIT.
720 				 */
721 				if (p->p_pid == 1)
722 					ps->ps_flag &= ~PS_NOCLDWAIT;
723 				else
724 					ps->ps_flag |= PS_NOCLDWAIT;
725 			} else
726 				ps->ps_flag &= ~PS_NOCLDWAIT;
727 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
728 				ps->ps_flag |= PS_CLDSIGIGN;
729 			else
730 				ps->ps_flag &= ~PS_CLDSIGIGN;
731 		}
732 		/*
733 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
734 		 * and for signals set to SIG_DFL where the default is to
735 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
736 		 * have to restart the process.
737 		 */
738 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
739 		    (sigprop(sig) & SA_IGNORE &&
740 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
741 			/* never to be seen again */
742 			sigqueue_delete_proc(p, sig);
743 			if (sig != SIGCONT)
744 				/* easier in psignal */
745 				SIGADDSET(ps->ps_sigignore, sig);
746 			SIGDELSET(ps->ps_sigcatch, sig);
747 		} else {
748 			SIGDELSET(ps->ps_sigignore, sig);
749 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
750 				SIGDELSET(ps->ps_sigcatch, sig);
751 			else
752 				SIGADDSET(ps->ps_sigcatch, sig);
753 		}
754 #ifdef COMPAT_FREEBSD4
755 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
756 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
757 		    (flags & KSA_FREEBSD4) == 0)
758 			SIGDELSET(ps->ps_freebsd4, sig);
759 		else
760 			SIGADDSET(ps->ps_freebsd4, sig);
761 #endif
762 #ifdef COMPAT_43
763 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
764 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
765 		    (flags & KSA_OSIGSET) == 0)
766 			SIGDELSET(ps->ps_osigset, sig);
767 		else
768 			SIGADDSET(ps->ps_osigset, sig);
769 #endif
770 	}
771 	mtx_unlock(&ps->ps_mtx);
772 	PROC_UNLOCK(p);
773 	return (0);
774 }
775 
776 #ifndef _SYS_SYSPROTO_H_
777 struct sigaction_args {
778 	int	sig;
779 	struct	sigaction *act;
780 	struct	sigaction *oact;
781 };
782 #endif
783 int
784 sys_sigaction(td, uap)
785 	struct thread *td;
786 	register struct sigaction_args *uap;
787 {
788 	struct sigaction act, oact;
789 	register struct sigaction *actp, *oactp;
790 	int error;
791 
792 	actp = (uap->act != NULL) ? &act : NULL;
793 	oactp = (uap->oact != NULL) ? &oact : NULL;
794 	if (actp) {
795 		error = copyin(uap->act, actp, sizeof(act));
796 		if (error)
797 			return (error);
798 	}
799 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
800 	if (oactp && !error)
801 		error = copyout(oactp, uap->oact, sizeof(oact));
802 	return (error);
803 }
804 
805 #ifdef COMPAT_FREEBSD4
806 #ifndef _SYS_SYSPROTO_H_
807 struct freebsd4_sigaction_args {
808 	int	sig;
809 	struct	sigaction *act;
810 	struct	sigaction *oact;
811 };
812 #endif
813 int
814 freebsd4_sigaction(td, uap)
815 	struct thread *td;
816 	register struct freebsd4_sigaction_args *uap;
817 {
818 	struct sigaction act, oact;
819 	register struct sigaction *actp, *oactp;
820 	int error;
821 
822 
823 	actp = (uap->act != NULL) ? &act : NULL;
824 	oactp = (uap->oact != NULL) ? &oact : NULL;
825 	if (actp) {
826 		error = copyin(uap->act, actp, sizeof(act));
827 		if (error)
828 			return (error);
829 	}
830 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
831 	if (oactp && !error)
832 		error = copyout(oactp, uap->oact, sizeof(oact));
833 	return (error);
834 }
835 #endif	/* COMAPT_FREEBSD4 */
836 
837 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
838 #ifndef _SYS_SYSPROTO_H_
839 struct osigaction_args {
840 	int	signum;
841 	struct	osigaction *nsa;
842 	struct	osigaction *osa;
843 };
844 #endif
845 int
846 osigaction(td, uap)
847 	struct thread *td;
848 	register struct osigaction_args *uap;
849 {
850 	struct osigaction sa;
851 	struct sigaction nsa, osa;
852 	register struct sigaction *nsap, *osap;
853 	int error;
854 
855 	if (uap->signum <= 0 || uap->signum >= ONSIG)
856 		return (EINVAL);
857 
858 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
859 	osap = (uap->osa != NULL) ? &osa : NULL;
860 
861 	if (nsap) {
862 		error = copyin(uap->nsa, &sa, sizeof(sa));
863 		if (error)
864 			return (error);
865 		nsap->sa_handler = sa.sa_handler;
866 		nsap->sa_flags = sa.sa_flags;
867 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
868 	}
869 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
870 	if (osap && !error) {
871 		sa.sa_handler = osap->sa_handler;
872 		sa.sa_flags = osap->sa_flags;
873 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
874 		error = copyout(&sa, uap->osa, sizeof(sa));
875 	}
876 	return (error);
877 }
878 
879 #if !defined(__i386__)
880 /* Avoid replicating the same stub everywhere */
881 int
882 osigreturn(td, uap)
883 	struct thread *td;
884 	struct osigreturn_args *uap;
885 {
886 
887 	return (nosys(td, (struct nosys_args *)uap));
888 }
889 #endif
890 #endif /* COMPAT_43 */
891 
892 /*
893  * Initialize signal state for process 0;
894  * set to ignore signals that are ignored by default.
895  */
896 void
897 siginit(p)
898 	struct proc *p;
899 {
900 	register int i;
901 	struct sigacts *ps;
902 
903 	PROC_LOCK(p);
904 	ps = p->p_sigacts;
905 	mtx_lock(&ps->ps_mtx);
906 	for (i = 1; i <= NSIG; i++)
907 		if (sigprop(i) & SA_IGNORE && i != SIGCONT)
908 			SIGADDSET(ps->ps_sigignore, i);
909 	mtx_unlock(&ps->ps_mtx);
910 	PROC_UNLOCK(p);
911 }
912 
913 /*
914  * Reset signals for an exec of the specified process.
915  */
916 void
917 execsigs(struct proc *p)
918 {
919 	struct sigacts *ps;
920 	int sig;
921 	struct thread *td;
922 
923 	/*
924 	 * Reset caught signals.  Held signals remain held
925 	 * through td_sigmask (unless they were caught,
926 	 * and are now ignored by default).
927 	 */
928 	PROC_LOCK_ASSERT(p, MA_OWNED);
929 	td = FIRST_THREAD_IN_PROC(p);
930 	ps = p->p_sigacts;
931 	mtx_lock(&ps->ps_mtx);
932 	while (SIGNOTEMPTY(ps->ps_sigcatch)) {
933 		sig = sig_ffs(&ps->ps_sigcatch);
934 		SIGDELSET(ps->ps_sigcatch, sig);
935 		if (sigprop(sig) & SA_IGNORE) {
936 			if (sig != SIGCONT)
937 				SIGADDSET(ps->ps_sigignore, sig);
938 			sigqueue_delete_proc(p, sig);
939 		}
940 		ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
941 	}
942 	/*
943 	 * Reset stack state to the user stack.
944 	 * Clear set of signals caught on the signal stack.
945 	 */
946 	td->td_sigstk.ss_flags = SS_DISABLE;
947 	td->td_sigstk.ss_size = 0;
948 	td->td_sigstk.ss_sp = 0;
949 	td->td_pflags &= ~TDP_ALTSTACK;
950 	/*
951 	 * Reset no zombies if child dies flag as Solaris does.
952 	 */
953 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
954 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
955 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
956 	mtx_unlock(&ps->ps_mtx);
957 }
958 
959 /*
960  * kern_sigprocmask()
961  *
962  *	Manipulate signal mask.
963  */
964 int
965 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
966     int flags)
967 {
968 	sigset_t new_block, oset1;
969 	struct proc *p;
970 	int error;
971 
972 	p = td->td_proc;
973 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
974 		PROC_LOCK(p);
975 	if (oset != NULL)
976 		*oset = td->td_sigmask;
977 
978 	error = 0;
979 	if (set != NULL) {
980 		switch (how) {
981 		case SIG_BLOCK:
982 			SIG_CANTMASK(*set);
983 			oset1 = td->td_sigmask;
984 			SIGSETOR(td->td_sigmask, *set);
985 			new_block = td->td_sigmask;
986 			SIGSETNAND(new_block, oset1);
987 			break;
988 		case SIG_UNBLOCK:
989 			SIGSETNAND(td->td_sigmask, *set);
990 			signotify(td);
991 			goto out;
992 		case SIG_SETMASK:
993 			SIG_CANTMASK(*set);
994 			oset1 = td->td_sigmask;
995 			if (flags & SIGPROCMASK_OLD)
996 				SIGSETLO(td->td_sigmask, *set);
997 			else
998 				td->td_sigmask = *set;
999 			new_block = td->td_sigmask;
1000 			SIGSETNAND(new_block, oset1);
1001 			signotify(td);
1002 			break;
1003 		default:
1004 			error = EINVAL;
1005 			goto out;
1006 		}
1007 
1008 		/*
1009 		 * The new_block set contains signals that were not previously
1010 		 * blocked, but are blocked now.
1011 		 *
1012 		 * In case we block any signal that was not previously blocked
1013 		 * for td, and process has the signal pending, try to schedule
1014 		 * signal delivery to some thread that does not block the
1015 		 * signal, possibly waking it up.
1016 		 */
1017 		if (p->p_numthreads != 1)
1018 			reschedule_signals(p, new_block, flags);
1019 	}
1020 
1021 out:
1022 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1023 		PROC_UNLOCK(p);
1024 	return (error);
1025 }
1026 
1027 #ifndef _SYS_SYSPROTO_H_
1028 struct sigprocmask_args {
1029 	int	how;
1030 	const sigset_t *set;
1031 	sigset_t *oset;
1032 };
1033 #endif
1034 int
1035 sys_sigprocmask(td, uap)
1036 	register struct thread *td;
1037 	struct sigprocmask_args *uap;
1038 {
1039 	sigset_t set, oset;
1040 	sigset_t *setp, *osetp;
1041 	int error;
1042 
1043 	setp = (uap->set != NULL) ? &set : NULL;
1044 	osetp = (uap->oset != NULL) ? &oset : NULL;
1045 	if (setp) {
1046 		error = copyin(uap->set, setp, sizeof(set));
1047 		if (error)
1048 			return (error);
1049 	}
1050 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1051 	if (osetp && !error) {
1052 		error = copyout(osetp, uap->oset, sizeof(oset));
1053 	}
1054 	return (error);
1055 }
1056 
1057 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1058 #ifndef _SYS_SYSPROTO_H_
1059 struct osigprocmask_args {
1060 	int	how;
1061 	osigset_t mask;
1062 };
1063 #endif
1064 int
1065 osigprocmask(td, uap)
1066 	register struct thread *td;
1067 	struct osigprocmask_args *uap;
1068 {
1069 	sigset_t set, oset;
1070 	int error;
1071 
1072 	OSIG2SIG(uap->mask, set);
1073 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1074 	SIG2OSIG(oset, td->td_retval[0]);
1075 	return (error);
1076 }
1077 #endif /* COMPAT_43 */
1078 
1079 int
1080 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1081 {
1082 	ksiginfo_t ksi;
1083 	sigset_t set;
1084 	int error;
1085 
1086 	error = copyin(uap->set, &set, sizeof(set));
1087 	if (error) {
1088 		td->td_retval[0] = error;
1089 		return (0);
1090 	}
1091 
1092 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1093 	if (error) {
1094 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1095 			error = ERESTART;
1096 		if (error == ERESTART)
1097 			return (error);
1098 		td->td_retval[0] = error;
1099 		return (0);
1100 	}
1101 
1102 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1103 	td->td_retval[0] = error;
1104 	return (0);
1105 }
1106 
1107 int
1108 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1109 {
1110 	struct timespec ts;
1111 	struct timespec *timeout;
1112 	sigset_t set;
1113 	ksiginfo_t ksi;
1114 	int error;
1115 
1116 	if (uap->timeout) {
1117 		error = copyin(uap->timeout, &ts, sizeof(ts));
1118 		if (error)
1119 			return (error);
1120 
1121 		timeout = &ts;
1122 	} else
1123 		timeout = NULL;
1124 
1125 	error = copyin(uap->set, &set, sizeof(set));
1126 	if (error)
1127 		return (error);
1128 
1129 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1130 	if (error)
1131 		return (error);
1132 
1133 	if (uap->info)
1134 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1135 
1136 	if (error == 0)
1137 		td->td_retval[0] = ksi.ksi_signo;
1138 	return (error);
1139 }
1140 
1141 int
1142 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1143 {
1144 	ksiginfo_t ksi;
1145 	sigset_t set;
1146 	int error;
1147 
1148 	error = copyin(uap->set, &set, sizeof(set));
1149 	if (error)
1150 		return (error);
1151 
1152 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1153 	if (error)
1154 		return (error);
1155 
1156 	if (uap->info)
1157 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1158 
1159 	if (error == 0)
1160 		td->td_retval[0] = ksi.ksi_signo;
1161 	return (error);
1162 }
1163 
1164 int
1165 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1166 	struct timespec *timeout)
1167 {
1168 	struct sigacts *ps;
1169 	sigset_t saved_mask, new_block;
1170 	struct proc *p;
1171 	int error, sig, timo, timevalid = 0;
1172 	struct timespec rts, ets, ts;
1173 	struct timeval tv;
1174 
1175 	p = td->td_proc;
1176 	error = 0;
1177 	ets.tv_sec = 0;
1178 	ets.tv_nsec = 0;
1179 
1180 	if (timeout != NULL) {
1181 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1182 			timevalid = 1;
1183 			getnanouptime(&rts);
1184 			ets = rts;
1185 			timespecadd(&ets, timeout);
1186 		}
1187 	}
1188 	ksiginfo_init(ksi);
1189 	/* Some signals can not be waited for. */
1190 	SIG_CANTMASK(waitset);
1191 	ps = p->p_sigacts;
1192 	PROC_LOCK(p);
1193 	saved_mask = td->td_sigmask;
1194 	SIGSETNAND(td->td_sigmask, waitset);
1195 	for (;;) {
1196 		mtx_lock(&ps->ps_mtx);
1197 		sig = cursig(td);
1198 		mtx_unlock(&ps->ps_mtx);
1199 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1200 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1201 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1202 				error = 0;
1203 				break;
1204 			}
1205 		}
1206 
1207 		if (error != 0)
1208 			break;
1209 
1210 		/*
1211 		 * POSIX says this must be checked after looking for pending
1212 		 * signals.
1213 		 */
1214 		if (timeout != NULL) {
1215 			if (!timevalid) {
1216 				error = EINVAL;
1217 				break;
1218 			}
1219 			getnanouptime(&rts);
1220 			if (timespeccmp(&rts, &ets, >=)) {
1221 				error = EAGAIN;
1222 				break;
1223 			}
1224 			ts = ets;
1225 			timespecsub(&ts, &rts);
1226 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1227 			timo = tvtohz(&tv);
1228 		} else {
1229 			timo = 0;
1230 		}
1231 
1232 		error = msleep(ps, &p->p_mtx, PPAUSE|PCATCH, "sigwait", timo);
1233 
1234 		if (timeout != NULL) {
1235 			if (error == ERESTART) {
1236 				/* Timeout can not be restarted. */
1237 				error = EINTR;
1238 			} else if (error == EAGAIN) {
1239 				/* We will calculate timeout by ourself. */
1240 				error = 0;
1241 			}
1242 		}
1243 	}
1244 
1245 	new_block = saved_mask;
1246 	SIGSETNAND(new_block, td->td_sigmask);
1247 	td->td_sigmask = saved_mask;
1248 	/*
1249 	 * Fewer signals can be delivered to us, reschedule signal
1250 	 * notification.
1251 	 */
1252 	if (p->p_numthreads != 1)
1253 		reschedule_signals(p, new_block, 0);
1254 
1255 	if (error == 0) {
1256 		SDT_PROBE(proc, kernel, , signal_clear, sig, ksi, 0, 0, 0);
1257 
1258 		if (ksi->ksi_code == SI_TIMER)
1259 			itimer_accept(p, ksi->ksi_timerid, ksi);
1260 
1261 #ifdef KTRACE
1262 		if (KTRPOINT(td, KTR_PSIG)) {
1263 			sig_t action;
1264 
1265 			mtx_lock(&ps->ps_mtx);
1266 			action = ps->ps_sigact[_SIG_IDX(sig)];
1267 			mtx_unlock(&ps->ps_mtx);
1268 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1269 		}
1270 #endif
1271 		if (sig == SIGKILL)
1272 			sigexit(td, sig);
1273 	}
1274 	PROC_UNLOCK(p);
1275 	return (error);
1276 }
1277 
1278 #ifndef _SYS_SYSPROTO_H_
1279 struct sigpending_args {
1280 	sigset_t	*set;
1281 };
1282 #endif
1283 int
1284 sys_sigpending(td, uap)
1285 	struct thread *td;
1286 	struct sigpending_args *uap;
1287 {
1288 	struct proc *p = td->td_proc;
1289 	sigset_t pending;
1290 
1291 	PROC_LOCK(p);
1292 	pending = p->p_sigqueue.sq_signals;
1293 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1294 	PROC_UNLOCK(p);
1295 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1296 }
1297 
1298 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1299 #ifndef _SYS_SYSPROTO_H_
1300 struct osigpending_args {
1301 	int	dummy;
1302 };
1303 #endif
1304 int
1305 osigpending(td, uap)
1306 	struct thread *td;
1307 	struct osigpending_args *uap;
1308 {
1309 	struct proc *p = td->td_proc;
1310 	sigset_t pending;
1311 
1312 	PROC_LOCK(p);
1313 	pending = p->p_sigqueue.sq_signals;
1314 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1315 	PROC_UNLOCK(p);
1316 	SIG2OSIG(pending, td->td_retval[0]);
1317 	return (0);
1318 }
1319 #endif /* COMPAT_43 */
1320 
1321 #if defined(COMPAT_43)
1322 /*
1323  * Generalized interface signal handler, 4.3-compatible.
1324  */
1325 #ifndef _SYS_SYSPROTO_H_
1326 struct osigvec_args {
1327 	int	signum;
1328 	struct	sigvec *nsv;
1329 	struct	sigvec *osv;
1330 };
1331 #endif
1332 /* ARGSUSED */
1333 int
1334 osigvec(td, uap)
1335 	struct thread *td;
1336 	register struct osigvec_args *uap;
1337 {
1338 	struct sigvec vec;
1339 	struct sigaction nsa, osa;
1340 	register struct sigaction *nsap, *osap;
1341 	int error;
1342 
1343 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1344 		return (EINVAL);
1345 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1346 	osap = (uap->osv != NULL) ? &osa : NULL;
1347 	if (nsap) {
1348 		error = copyin(uap->nsv, &vec, sizeof(vec));
1349 		if (error)
1350 			return (error);
1351 		nsap->sa_handler = vec.sv_handler;
1352 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1353 		nsap->sa_flags = vec.sv_flags;
1354 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1355 	}
1356 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1357 	if (osap && !error) {
1358 		vec.sv_handler = osap->sa_handler;
1359 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1360 		vec.sv_flags = osap->sa_flags;
1361 		vec.sv_flags &= ~SA_NOCLDWAIT;
1362 		vec.sv_flags ^= SA_RESTART;
1363 		error = copyout(&vec, uap->osv, sizeof(vec));
1364 	}
1365 	return (error);
1366 }
1367 
1368 #ifndef _SYS_SYSPROTO_H_
1369 struct osigblock_args {
1370 	int	mask;
1371 };
1372 #endif
1373 int
1374 osigblock(td, uap)
1375 	register struct thread *td;
1376 	struct osigblock_args *uap;
1377 {
1378 	sigset_t set, oset;
1379 
1380 	OSIG2SIG(uap->mask, set);
1381 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1382 	SIG2OSIG(oset, td->td_retval[0]);
1383 	return (0);
1384 }
1385 
1386 #ifndef _SYS_SYSPROTO_H_
1387 struct osigsetmask_args {
1388 	int	mask;
1389 };
1390 #endif
1391 int
1392 osigsetmask(td, uap)
1393 	struct thread *td;
1394 	struct osigsetmask_args *uap;
1395 {
1396 	sigset_t set, oset;
1397 
1398 	OSIG2SIG(uap->mask, set);
1399 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1400 	SIG2OSIG(oset, td->td_retval[0]);
1401 	return (0);
1402 }
1403 #endif /* COMPAT_43 */
1404 
1405 /*
1406  * Suspend calling thread until signal, providing mask to be set in the
1407  * meantime.
1408  */
1409 #ifndef _SYS_SYSPROTO_H_
1410 struct sigsuspend_args {
1411 	const sigset_t *sigmask;
1412 };
1413 #endif
1414 /* ARGSUSED */
1415 int
1416 sys_sigsuspend(td, uap)
1417 	struct thread *td;
1418 	struct sigsuspend_args *uap;
1419 {
1420 	sigset_t mask;
1421 	int error;
1422 
1423 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1424 	if (error)
1425 		return (error);
1426 	return (kern_sigsuspend(td, mask));
1427 }
1428 
1429 int
1430 kern_sigsuspend(struct thread *td, sigset_t mask)
1431 {
1432 	struct proc *p = td->td_proc;
1433 	int has_sig, sig;
1434 
1435 	/*
1436 	 * When returning from sigsuspend, we want
1437 	 * the old mask to be restored after the
1438 	 * signal handler has finished.  Thus, we
1439 	 * save it here and mark the sigacts structure
1440 	 * to indicate this.
1441 	 */
1442 	PROC_LOCK(p);
1443 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1444 	    SIGPROCMASK_PROC_LOCKED);
1445 	td->td_pflags |= TDP_OLDMASK;
1446 
1447 	/*
1448 	 * Process signals now. Otherwise, we can get spurious wakeup
1449 	 * due to signal entered process queue, but delivered to other
1450 	 * thread. But sigsuspend should return only on signal
1451 	 * delivery.
1452 	 */
1453 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1454 	for (has_sig = 0; !has_sig;) {
1455 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1456 			0) == 0)
1457 			/* void */;
1458 		thread_suspend_check(0);
1459 		mtx_lock(&p->p_sigacts->ps_mtx);
1460 		while ((sig = cursig(td)) != 0)
1461 			has_sig += postsig(sig);
1462 		mtx_unlock(&p->p_sigacts->ps_mtx);
1463 	}
1464 	PROC_UNLOCK(p);
1465 	td->td_errno = EINTR;
1466 	td->td_pflags |= TDP_NERRNO;
1467 	return (EJUSTRETURN);
1468 }
1469 
1470 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1471 /*
1472  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1473  * convention: libc stub passes mask, not pointer, to save a copyin.
1474  */
1475 #ifndef _SYS_SYSPROTO_H_
1476 struct osigsuspend_args {
1477 	osigset_t mask;
1478 };
1479 #endif
1480 /* ARGSUSED */
1481 int
1482 osigsuspend(td, uap)
1483 	struct thread *td;
1484 	struct osigsuspend_args *uap;
1485 {
1486 	sigset_t mask;
1487 
1488 	OSIG2SIG(uap->mask, mask);
1489 	return (kern_sigsuspend(td, mask));
1490 }
1491 #endif /* COMPAT_43 */
1492 
1493 #if defined(COMPAT_43)
1494 #ifndef _SYS_SYSPROTO_H_
1495 struct osigstack_args {
1496 	struct	sigstack *nss;
1497 	struct	sigstack *oss;
1498 };
1499 #endif
1500 /* ARGSUSED */
1501 int
1502 osigstack(td, uap)
1503 	struct thread *td;
1504 	register struct osigstack_args *uap;
1505 {
1506 	struct sigstack nss, oss;
1507 	int error = 0;
1508 
1509 	if (uap->nss != NULL) {
1510 		error = copyin(uap->nss, &nss, sizeof(nss));
1511 		if (error)
1512 			return (error);
1513 	}
1514 	oss.ss_sp = td->td_sigstk.ss_sp;
1515 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1516 	if (uap->nss != NULL) {
1517 		td->td_sigstk.ss_sp = nss.ss_sp;
1518 		td->td_sigstk.ss_size = 0;
1519 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1520 		td->td_pflags |= TDP_ALTSTACK;
1521 	}
1522 	if (uap->oss != NULL)
1523 		error = copyout(&oss, uap->oss, sizeof(oss));
1524 
1525 	return (error);
1526 }
1527 #endif /* COMPAT_43 */
1528 
1529 #ifndef _SYS_SYSPROTO_H_
1530 struct sigaltstack_args {
1531 	stack_t	*ss;
1532 	stack_t	*oss;
1533 };
1534 #endif
1535 /* ARGSUSED */
1536 int
1537 sys_sigaltstack(td, uap)
1538 	struct thread *td;
1539 	register struct sigaltstack_args *uap;
1540 {
1541 	stack_t ss, oss;
1542 	int error;
1543 
1544 	if (uap->ss != NULL) {
1545 		error = copyin(uap->ss, &ss, sizeof(ss));
1546 		if (error)
1547 			return (error);
1548 	}
1549 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1550 	    (uap->oss != NULL) ? &oss : NULL);
1551 	if (error)
1552 		return (error);
1553 	if (uap->oss != NULL)
1554 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1555 	return (error);
1556 }
1557 
1558 int
1559 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1560 {
1561 	struct proc *p = td->td_proc;
1562 	int oonstack;
1563 
1564 	oonstack = sigonstack(cpu_getstack(td));
1565 
1566 	if (oss != NULL) {
1567 		*oss = td->td_sigstk;
1568 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1569 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1570 	}
1571 
1572 	if (ss != NULL) {
1573 		if (oonstack)
1574 			return (EPERM);
1575 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1576 			return (EINVAL);
1577 		if (!(ss->ss_flags & SS_DISABLE)) {
1578 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1579 				return (ENOMEM);
1580 
1581 			td->td_sigstk = *ss;
1582 			td->td_pflags |= TDP_ALTSTACK;
1583 		} else {
1584 			td->td_pflags &= ~TDP_ALTSTACK;
1585 		}
1586 	}
1587 	return (0);
1588 }
1589 
1590 /*
1591  * Common code for kill process group/broadcast kill.
1592  * cp is calling process.
1593  */
1594 static int
1595 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1596 {
1597 	struct proc *p;
1598 	struct pgrp *pgrp;
1599 	int err;
1600 	int ret;
1601 
1602 	ret = ESRCH;
1603 	if (all) {
1604 		/*
1605 		 * broadcast
1606 		 */
1607 		sx_slock(&allproc_lock);
1608 		FOREACH_PROC_IN_SYSTEM(p) {
1609 			PROC_LOCK(p);
1610 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1611 			    p == td->td_proc || p->p_state == PRS_NEW) {
1612 				PROC_UNLOCK(p);
1613 				continue;
1614 			}
1615 			err = p_cansignal(td, p, sig);
1616 			if (err == 0) {
1617 				if (sig)
1618 					pksignal(p, sig, ksi);
1619 				ret = err;
1620 			}
1621 			else if (ret == ESRCH)
1622 				ret = err;
1623 			PROC_UNLOCK(p);
1624 		}
1625 		sx_sunlock(&allproc_lock);
1626 	} else {
1627 		sx_slock(&proctree_lock);
1628 		if (pgid == 0) {
1629 			/*
1630 			 * zero pgid means send to my process group.
1631 			 */
1632 			pgrp = td->td_proc->p_pgrp;
1633 			PGRP_LOCK(pgrp);
1634 		} else {
1635 			pgrp = pgfind(pgid);
1636 			if (pgrp == NULL) {
1637 				sx_sunlock(&proctree_lock);
1638 				return (ESRCH);
1639 			}
1640 		}
1641 		sx_sunlock(&proctree_lock);
1642 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1643 			PROC_LOCK(p);
1644 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1645 			    p->p_state == PRS_NEW) {
1646 				PROC_UNLOCK(p);
1647 				continue;
1648 			}
1649 			err = p_cansignal(td, p, sig);
1650 			if (err == 0) {
1651 				if (sig)
1652 					pksignal(p, sig, ksi);
1653 				ret = err;
1654 			}
1655 			else if (ret == ESRCH)
1656 				ret = err;
1657 			PROC_UNLOCK(p);
1658 		}
1659 		PGRP_UNLOCK(pgrp);
1660 	}
1661 	return (ret);
1662 }
1663 
1664 #ifndef _SYS_SYSPROTO_H_
1665 struct kill_args {
1666 	int	pid;
1667 	int	signum;
1668 };
1669 #endif
1670 /* ARGSUSED */
1671 int
1672 sys_kill(struct thread *td, struct kill_args *uap)
1673 {
1674 	ksiginfo_t ksi;
1675 	struct proc *p;
1676 	int error;
1677 
1678 	/*
1679 	 * A process in capability mode can send signals only to himself.
1680 	 * The main rationale behind this is that abort(3) is implemented as
1681 	 * kill(getpid(), SIGABRT).
1682 	 */
1683 	if (IN_CAPABILITY_MODE(td) && uap->pid != td->td_proc->p_pid)
1684 		return (ECAPMODE);
1685 
1686 	AUDIT_ARG_SIGNUM(uap->signum);
1687 	AUDIT_ARG_PID(uap->pid);
1688 	if ((u_int)uap->signum > _SIG_MAXSIG)
1689 		return (EINVAL);
1690 
1691 	ksiginfo_init(&ksi);
1692 	ksi.ksi_signo = uap->signum;
1693 	ksi.ksi_code = SI_USER;
1694 	ksi.ksi_pid = td->td_proc->p_pid;
1695 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1696 
1697 	if (uap->pid > 0) {
1698 		/* kill single process */
1699 		if ((p = pfind(uap->pid)) == NULL) {
1700 			if ((p = zpfind(uap->pid)) == NULL)
1701 				return (ESRCH);
1702 		}
1703 		AUDIT_ARG_PROCESS(p);
1704 		error = p_cansignal(td, p, uap->signum);
1705 		if (error == 0 && uap->signum)
1706 			pksignal(p, uap->signum, &ksi);
1707 		PROC_UNLOCK(p);
1708 		return (error);
1709 	}
1710 	switch (uap->pid) {
1711 	case -1:		/* broadcast signal */
1712 		return (killpg1(td, uap->signum, 0, 1, &ksi));
1713 	case 0:			/* signal own process group */
1714 		return (killpg1(td, uap->signum, 0, 0, &ksi));
1715 	default:		/* negative explicit process group */
1716 		return (killpg1(td, uap->signum, -uap->pid, 0, &ksi));
1717 	}
1718 	/* NOTREACHED */
1719 }
1720 
1721 int
1722 sys_pdkill(td, uap)
1723 	struct thread *td;
1724 	struct pdkill_args *uap;
1725 {
1726 #ifdef PROCDESC
1727 	struct proc *p;
1728 	cap_rights_t rights;
1729 	int error;
1730 
1731 	AUDIT_ARG_SIGNUM(uap->signum);
1732 	AUDIT_ARG_FD(uap->fd);
1733 	if ((u_int)uap->signum > _SIG_MAXSIG)
1734 		return (EINVAL);
1735 
1736 	error = procdesc_find(td, uap->fd,
1737 	    cap_rights_init(&rights, CAP_PDKILL), &p);
1738 	if (error)
1739 		return (error);
1740 	AUDIT_ARG_PROCESS(p);
1741 	error = p_cansignal(td, p, uap->signum);
1742 	if (error == 0 && uap->signum)
1743 		kern_psignal(p, uap->signum);
1744 	PROC_UNLOCK(p);
1745 	return (error);
1746 #else
1747 	return (ENOSYS);
1748 #endif
1749 }
1750 
1751 #if defined(COMPAT_43)
1752 #ifndef _SYS_SYSPROTO_H_
1753 struct okillpg_args {
1754 	int	pgid;
1755 	int	signum;
1756 };
1757 #endif
1758 /* ARGSUSED */
1759 int
1760 okillpg(struct thread *td, struct okillpg_args *uap)
1761 {
1762 	ksiginfo_t ksi;
1763 
1764 	AUDIT_ARG_SIGNUM(uap->signum);
1765 	AUDIT_ARG_PID(uap->pgid);
1766 	if ((u_int)uap->signum > _SIG_MAXSIG)
1767 		return (EINVAL);
1768 
1769 	ksiginfo_init(&ksi);
1770 	ksi.ksi_signo = uap->signum;
1771 	ksi.ksi_code = SI_USER;
1772 	ksi.ksi_pid = td->td_proc->p_pid;
1773 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1774 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1775 }
1776 #endif /* COMPAT_43 */
1777 
1778 #ifndef _SYS_SYSPROTO_H_
1779 struct sigqueue_args {
1780 	pid_t pid;
1781 	int signum;
1782 	/* union sigval */ void *value;
1783 };
1784 #endif
1785 int
1786 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1787 {
1788 	ksiginfo_t ksi;
1789 	struct proc *p;
1790 	int error;
1791 
1792 	if ((u_int)uap->signum > _SIG_MAXSIG)
1793 		return (EINVAL);
1794 
1795 	/*
1796 	 * Specification says sigqueue can only send signal to
1797 	 * single process.
1798 	 */
1799 	if (uap->pid <= 0)
1800 		return (EINVAL);
1801 
1802 	if ((p = pfind(uap->pid)) == NULL) {
1803 		if ((p = zpfind(uap->pid)) == NULL)
1804 			return (ESRCH);
1805 	}
1806 	error = p_cansignal(td, p, uap->signum);
1807 	if (error == 0 && uap->signum != 0) {
1808 		ksiginfo_init(&ksi);
1809 		ksi.ksi_flags = KSI_SIGQ;
1810 		ksi.ksi_signo = uap->signum;
1811 		ksi.ksi_code = SI_QUEUE;
1812 		ksi.ksi_pid = td->td_proc->p_pid;
1813 		ksi.ksi_uid = td->td_ucred->cr_ruid;
1814 		ksi.ksi_value.sival_ptr = uap->value;
1815 		error = pksignal(p, ksi.ksi_signo, &ksi);
1816 	}
1817 	PROC_UNLOCK(p);
1818 	return (error);
1819 }
1820 
1821 /*
1822  * Send a signal to a process group.
1823  */
1824 void
1825 gsignal(int pgid, int sig, ksiginfo_t *ksi)
1826 {
1827 	struct pgrp *pgrp;
1828 
1829 	if (pgid != 0) {
1830 		sx_slock(&proctree_lock);
1831 		pgrp = pgfind(pgid);
1832 		sx_sunlock(&proctree_lock);
1833 		if (pgrp != NULL) {
1834 			pgsignal(pgrp, sig, 0, ksi);
1835 			PGRP_UNLOCK(pgrp);
1836 		}
1837 	}
1838 }
1839 
1840 /*
1841  * Send a signal to a process group.  If checktty is 1,
1842  * limit to members which have a controlling terminal.
1843  */
1844 void
1845 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
1846 {
1847 	struct proc *p;
1848 
1849 	if (pgrp) {
1850 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
1851 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1852 			PROC_LOCK(p);
1853 			if (p->p_state == PRS_NORMAL &&
1854 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
1855 				pksignal(p, sig, ksi);
1856 			PROC_UNLOCK(p);
1857 		}
1858 	}
1859 }
1860 
1861 /*
1862  * Send a signal caused by a trap to the current thread.  If it will be
1863  * caught immediately, deliver it with correct code.  Otherwise, post it
1864  * normally.
1865  */
1866 void
1867 trapsignal(struct thread *td, ksiginfo_t *ksi)
1868 {
1869 	struct sigacts *ps;
1870 	sigset_t mask;
1871 	struct proc *p;
1872 	int sig;
1873 	int code;
1874 
1875 	p = td->td_proc;
1876 	sig = ksi->ksi_signo;
1877 	code = ksi->ksi_code;
1878 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
1879 
1880 	PROC_LOCK(p);
1881 	ps = p->p_sigacts;
1882 	mtx_lock(&ps->ps_mtx);
1883 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
1884 	    !SIGISMEMBER(td->td_sigmask, sig)) {
1885 		td->td_ru.ru_nsignals++;
1886 #ifdef KTRACE
1887 		if (KTRPOINT(curthread, KTR_PSIG))
1888 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
1889 			    &td->td_sigmask, code);
1890 #endif
1891 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
1892 				ksi, &td->td_sigmask);
1893 		mask = ps->ps_catchmask[_SIG_IDX(sig)];
1894 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
1895 			SIGADDSET(mask, sig);
1896 		kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
1897 		    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
1898 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
1899 			/*
1900 			 * See kern_sigaction() for origin of this code.
1901 			 */
1902 			SIGDELSET(ps->ps_sigcatch, sig);
1903 			if (sig != SIGCONT &&
1904 			    sigprop(sig) & SA_IGNORE)
1905 				SIGADDSET(ps->ps_sigignore, sig);
1906 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1907 		}
1908 		mtx_unlock(&ps->ps_mtx);
1909 	} else {
1910 		/*
1911 		 * Avoid a possible infinite loop if the thread
1912 		 * masking the signal or process is ignoring the
1913 		 * signal.
1914 		 */
1915 		if (kern_forcesigexit &&
1916 		    (SIGISMEMBER(td->td_sigmask, sig) ||
1917 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
1918 			SIGDELSET(td->td_sigmask, sig);
1919 			SIGDELSET(ps->ps_sigcatch, sig);
1920 			SIGDELSET(ps->ps_sigignore, sig);
1921 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1922 		}
1923 		mtx_unlock(&ps->ps_mtx);
1924 		p->p_code = code;	/* XXX for core dump/debugger */
1925 		p->p_sig = sig;		/* XXX to verify code */
1926 		tdsendsignal(p, td, sig, ksi);
1927 	}
1928 	PROC_UNLOCK(p);
1929 }
1930 
1931 static struct thread *
1932 sigtd(struct proc *p, int sig, int prop)
1933 {
1934 	struct thread *td, *signal_td;
1935 
1936 	PROC_LOCK_ASSERT(p, MA_OWNED);
1937 
1938 	/*
1939 	 * Check if current thread can handle the signal without
1940 	 * switching context to another thread.
1941 	 */
1942 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig))
1943 		return (curthread);
1944 	signal_td = NULL;
1945 	FOREACH_THREAD_IN_PROC(p, td) {
1946 		if (!SIGISMEMBER(td->td_sigmask, sig)) {
1947 			signal_td = td;
1948 			break;
1949 		}
1950 	}
1951 	if (signal_td == NULL)
1952 		signal_td = FIRST_THREAD_IN_PROC(p);
1953 	return (signal_td);
1954 }
1955 
1956 /*
1957  * Send the signal to the process.  If the signal has an action, the action
1958  * is usually performed by the target process rather than the caller; we add
1959  * the signal to the set of pending signals for the process.
1960  *
1961  * Exceptions:
1962  *   o When a stop signal is sent to a sleeping process that takes the
1963  *     default action, the process is stopped without awakening it.
1964  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
1965  *     regardless of the signal action (eg, blocked or ignored).
1966  *
1967  * Other ignored signals are discarded immediately.
1968  *
1969  * NB: This function may be entered from the debugger via the "kill" DDB
1970  * command.  There is little that can be done to mitigate the possibly messy
1971  * side effects of this unwise possibility.
1972  */
1973 void
1974 kern_psignal(struct proc *p, int sig)
1975 {
1976 	ksiginfo_t ksi;
1977 
1978 	ksiginfo_init(&ksi);
1979 	ksi.ksi_signo = sig;
1980 	ksi.ksi_code = SI_KERNEL;
1981 	(void) tdsendsignal(p, NULL, sig, &ksi);
1982 }
1983 
1984 int
1985 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
1986 {
1987 
1988 	return (tdsendsignal(p, NULL, sig, ksi));
1989 }
1990 
1991 /* Utility function for finding a thread to send signal event to. */
1992 int
1993 sigev_findtd(struct proc *p ,struct sigevent *sigev, struct thread **ttd)
1994 {
1995 	struct thread *td;
1996 
1997 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
1998 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
1999 		if (td == NULL)
2000 			return (ESRCH);
2001 		*ttd = td;
2002 	} else {
2003 		*ttd = NULL;
2004 		PROC_LOCK(p);
2005 	}
2006 	return (0);
2007 }
2008 
2009 void
2010 tdsignal(struct thread *td, int sig)
2011 {
2012 	ksiginfo_t ksi;
2013 
2014 	ksiginfo_init(&ksi);
2015 	ksi.ksi_signo = sig;
2016 	ksi.ksi_code = SI_KERNEL;
2017 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2018 }
2019 
2020 void
2021 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2022 {
2023 
2024 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2025 }
2026 
2027 int
2028 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2029 {
2030 	sig_t action;
2031 	sigqueue_t *sigqueue;
2032 	int prop;
2033 	struct sigacts *ps;
2034 	int intrval;
2035 	int ret = 0;
2036 	int wakeup_swapper;
2037 
2038 	MPASS(td == NULL || p == td->td_proc);
2039 	PROC_LOCK_ASSERT(p, MA_OWNED);
2040 
2041 	if (!_SIG_VALID(sig))
2042 		panic("%s(): invalid signal %d", __func__, sig);
2043 
2044 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2045 
2046 	/*
2047 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2048 	 */
2049 	if (p->p_state == PRS_ZOMBIE) {
2050 		if (ksi && (ksi->ksi_flags & KSI_INS))
2051 			ksiginfo_tryfree(ksi);
2052 		return (ret);
2053 	}
2054 
2055 	ps = p->p_sigacts;
2056 	KNOTE_LOCKED(&p->p_klist, NOTE_SIGNAL | sig);
2057 	prop = sigprop(sig);
2058 
2059 	if (td == NULL) {
2060 		td = sigtd(p, sig, prop);
2061 		sigqueue = &p->p_sigqueue;
2062 	} else
2063 		sigqueue = &td->td_sigqueue;
2064 
2065 	SDT_PROBE(proc, kernel, , signal_send, td, p, sig, 0, 0 );
2066 
2067 	/*
2068 	 * If the signal is being ignored,
2069 	 * then we forget about it immediately.
2070 	 * (Note: we don't set SIGCONT in ps_sigignore,
2071 	 * and if it is set to SIG_IGN,
2072 	 * action will be SIG_DFL here.)
2073 	 */
2074 	mtx_lock(&ps->ps_mtx);
2075 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2076 		SDT_PROBE(proc, kernel, , signal_discard, td, p, sig, 0, 0 );
2077 
2078 		mtx_unlock(&ps->ps_mtx);
2079 		if (ksi && (ksi->ksi_flags & KSI_INS))
2080 			ksiginfo_tryfree(ksi);
2081 		return (ret);
2082 	}
2083 	if (SIGISMEMBER(td->td_sigmask, sig))
2084 		action = SIG_HOLD;
2085 	else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2086 		action = SIG_CATCH;
2087 	else
2088 		action = SIG_DFL;
2089 	if (SIGISMEMBER(ps->ps_sigintr, sig))
2090 		intrval = EINTR;
2091 	else
2092 		intrval = ERESTART;
2093 	mtx_unlock(&ps->ps_mtx);
2094 
2095 	if (prop & SA_CONT)
2096 		sigqueue_delete_stopmask_proc(p);
2097 	else if (prop & SA_STOP) {
2098 		/*
2099 		 * If sending a tty stop signal to a member of an orphaned
2100 		 * process group, discard the signal here if the action
2101 		 * is default; don't stop the process below if sleeping,
2102 		 * and don't clear any pending SIGCONT.
2103 		 */
2104 		if ((prop & SA_TTYSTOP) &&
2105 		    (p->p_pgrp->pg_jobc == 0) &&
2106 		    (action == SIG_DFL)) {
2107 			if (ksi && (ksi->ksi_flags & KSI_INS))
2108 				ksiginfo_tryfree(ksi);
2109 			return (ret);
2110 		}
2111 		sigqueue_delete_proc(p, SIGCONT);
2112 		if (p->p_flag & P_CONTINUED) {
2113 			p->p_flag &= ~P_CONTINUED;
2114 			PROC_LOCK(p->p_pptr);
2115 			sigqueue_take(p->p_ksi);
2116 			PROC_UNLOCK(p->p_pptr);
2117 		}
2118 	}
2119 
2120 	ret = sigqueue_add(sigqueue, sig, ksi);
2121 	if (ret != 0)
2122 		return (ret);
2123 	signotify(td);
2124 	/*
2125 	 * Defer further processing for signals which are held,
2126 	 * except that stopped processes must be continued by SIGCONT.
2127 	 */
2128 	if (action == SIG_HOLD &&
2129 	    !((prop & SA_CONT) && (p->p_flag & P_STOPPED_SIG)))
2130 		return (ret);
2131 	/*
2132 	 * SIGKILL: Remove procfs STOPEVENTs.
2133 	 */
2134 	if (sig == SIGKILL) {
2135 		/* from procfs_ioctl.c: PIOCBIC */
2136 		p->p_stops = 0;
2137 		/* from procfs_ioctl.c: PIOCCONT */
2138 		p->p_step = 0;
2139 		wakeup(&p->p_step);
2140 	}
2141 	/*
2142 	 * Some signals have a process-wide effect and a per-thread
2143 	 * component.  Most processing occurs when the process next
2144 	 * tries to cross the user boundary, however there are some
2145 	 * times when processing needs to be done immediately, such as
2146 	 * waking up threads so that they can cross the user boundary.
2147 	 * We try to do the per-process part here.
2148 	 */
2149 	if (P_SHOULDSTOP(p)) {
2150 		KASSERT(!(p->p_flag & P_WEXIT),
2151 		    ("signal to stopped but exiting process"));
2152 		if (sig == SIGKILL) {
2153 			/*
2154 			 * If traced process is already stopped,
2155 			 * then no further action is necessary.
2156 			 */
2157 			if (p->p_flag & P_TRACED)
2158 				goto out;
2159 			/*
2160 			 * SIGKILL sets process running.
2161 			 * It will die elsewhere.
2162 			 * All threads must be restarted.
2163 			 */
2164 			p->p_flag &= ~P_STOPPED_SIG;
2165 			goto runfast;
2166 		}
2167 
2168 		if (prop & SA_CONT) {
2169 			/*
2170 			 * If traced process is already stopped,
2171 			 * then no further action is necessary.
2172 			 */
2173 			if (p->p_flag & P_TRACED)
2174 				goto out;
2175 			/*
2176 			 * If SIGCONT is default (or ignored), we continue the
2177 			 * process but don't leave the signal in sigqueue as
2178 			 * it has no further action.  If SIGCONT is held, we
2179 			 * continue the process and leave the signal in
2180 			 * sigqueue.  If the process catches SIGCONT, let it
2181 			 * handle the signal itself.  If it isn't waiting on
2182 			 * an event, it goes back to run state.
2183 			 * Otherwise, process goes back to sleep state.
2184 			 */
2185 			p->p_flag &= ~P_STOPPED_SIG;
2186 			PROC_SLOCK(p);
2187 			if (p->p_numthreads == p->p_suspcount) {
2188 				PROC_SUNLOCK(p);
2189 				p->p_flag |= P_CONTINUED;
2190 				p->p_xstat = SIGCONT;
2191 				PROC_LOCK(p->p_pptr);
2192 				childproc_continued(p);
2193 				PROC_UNLOCK(p->p_pptr);
2194 				PROC_SLOCK(p);
2195 			}
2196 			if (action == SIG_DFL) {
2197 				thread_unsuspend(p);
2198 				PROC_SUNLOCK(p);
2199 				sigqueue_delete(sigqueue, sig);
2200 				goto out;
2201 			}
2202 			if (action == SIG_CATCH) {
2203 				/*
2204 				 * The process wants to catch it so it needs
2205 				 * to run at least one thread, but which one?
2206 				 */
2207 				PROC_SUNLOCK(p);
2208 				goto runfast;
2209 			}
2210 			/*
2211 			 * The signal is not ignored or caught.
2212 			 */
2213 			thread_unsuspend(p);
2214 			PROC_SUNLOCK(p);
2215 			goto out;
2216 		}
2217 
2218 		if (prop & SA_STOP) {
2219 			/*
2220 			 * If traced process is already stopped,
2221 			 * then no further action is necessary.
2222 			 */
2223 			if (p->p_flag & P_TRACED)
2224 				goto out;
2225 			/*
2226 			 * Already stopped, don't need to stop again
2227 			 * (If we did the shell could get confused).
2228 			 * Just make sure the signal STOP bit set.
2229 			 */
2230 			p->p_flag |= P_STOPPED_SIG;
2231 			sigqueue_delete(sigqueue, sig);
2232 			goto out;
2233 		}
2234 
2235 		/*
2236 		 * All other kinds of signals:
2237 		 * If a thread is sleeping interruptibly, simulate a
2238 		 * wakeup so that when it is continued it will be made
2239 		 * runnable and can look at the signal.  However, don't make
2240 		 * the PROCESS runnable, leave it stopped.
2241 		 * It may run a bit until it hits a thread_suspend_check().
2242 		 */
2243 		wakeup_swapper = 0;
2244 		PROC_SLOCK(p);
2245 		thread_lock(td);
2246 		if (TD_ON_SLEEPQ(td) && (td->td_flags & TDF_SINTR))
2247 			wakeup_swapper = sleepq_abort(td, intrval);
2248 		thread_unlock(td);
2249 		PROC_SUNLOCK(p);
2250 		if (wakeup_swapper)
2251 			kick_proc0();
2252 		goto out;
2253 		/*
2254 		 * Mutexes are short lived. Threads waiting on them will
2255 		 * hit thread_suspend_check() soon.
2256 		 */
2257 	} else if (p->p_state == PRS_NORMAL) {
2258 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2259 			tdsigwakeup(td, sig, action, intrval);
2260 			goto out;
2261 		}
2262 
2263 		MPASS(action == SIG_DFL);
2264 
2265 		if (prop & SA_STOP) {
2266 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2267 				goto out;
2268 			p->p_flag |= P_STOPPED_SIG;
2269 			p->p_xstat = sig;
2270 			PROC_SLOCK(p);
2271 			sig_suspend_threads(td, p, 1);
2272 			if (p->p_numthreads == p->p_suspcount) {
2273 				/*
2274 				 * only thread sending signal to another
2275 				 * process can reach here, if thread is sending
2276 				 * signal to its process, because thread does
2277 				 * not suspend itself here, p_numthreads
2278 				 * should never be equal to p_suspcount.
2279 				 */
2280 				thread_stopped(p);
2281 				PROC_SUNLOCK(p);
2282 				sigqueue_delete_proc(p, p->p_xstat);
2283 			} else
2284 				PROC_SUNLOCK(p);
2285 			goto out;
2286 		}
2287 	} else {
2288 		/* Not in "NORMAL" state. discard the signal. */
2289 		sigqueue_delete(sigqueue, sig);
2290 		goto out;
2291 	}
2292 
2293 	/*
2294 	 * The process is not stopped so we need to apply the signal to all the
2295 	 * running threads.
2296 	 */
2297 runfast:
2298 	tdsigwakeup(td, sig, action, intrval);
2299 	PROC_SLOCK(p);
2300 	thread_unsuspend(p);
2301 	PROC_SUNLOCK(p);
2302 out:
2303 	/* If we jump here, proc slock should not be owned. */
2304 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2305 	return (ret);
2306 }
2307 
2308 /*
2309  * The force of a signal has been directed against a single
2310  * thread.  We need to see what we can do about knocking it
2311  * out of any sleep it may be in etc.
2312  */
2313 static void
2314 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2315 {
2316 	struct proc *p = td->td_proc;
2317 	register int prop;
2318 	int wakeup_swapper;
2319 
2320 	wakeup_swapper = 0;
2321 	PROC_LOCK_ASSERT(p, MA_OWNED);
2322 	prop = sigprop(sig);
2323 
2324 	PROC_SLOCK(p);
2325 	thread_lock(td);
2326 	/*
2327 	 * Bring the priority of a thread up if we want it to get
2328 	 * killed in this lifetime.
2329 	 */
2330 	if (action == SIG_DFL && (prop & SA_KILL) && td->td_priority > PUSER)
2331 		sched_prio(td, PUSER);
2332 	if (TD_ON_SLEEPQ(td)) {
2333 		/*
2334 		 * If thread is sleeping uninterruptibly
2335 		 * we can't interrupt the sleep... the signal will
2336 		 * be noticed when the process returns through
2337 		 * trap() or syscall().
2338 		 */
2339 		if ((td->td_flags & TDF_SINTR) == 0)
2340 			goto out;
2341 		/*
2342 		 * If SIGCONT is default (or ignored) and process is
2343 		 * asleep, we are finished; the process should not
2344 		 * be awakened.
2345 		 */
2346 		if ((prop & SA_CONT) && action == SIG_DFL) {
2347 			thread_unlock(td);
2348 			PROC_SUNLOCK(p);
2349 			sigqueue_delete(&p->p_sigqueue, sig);
2350 			/*
2351 			 * It may be on either list in this state.
2352 			 * Remove from both for now.
2353 			 */
2354 			sigqueue_delete(&td->td_sigqueue, sig);
2355 			return;
2356 		}
2357 
2358 		/*
2359 		 * Don't awaken a sleeping thread for SIGSTOP if the
2360 		 * STOP signal is deferred.
2361 		 */
2362 		if ((prop & SA_STOP) && (td->td_flags & TDF_SBDRY))
2363 			goto out;
2364 
2365 		/*
2366 		 * Give low priority threads a better chance to run.
2367 		 */
2368 		if (td->td_priority > PUSER)
2369 			sched_prio(td, PUSER);
2370 
2371 		wakeup_swapper = sleepq_abort(td, intrval);
2372 	} else {
2373 		/*
2374 		 * Other states do nothing with the signal immediately,
2375 		 * other than kicking ourselves if we are running.
2376 		 * It will either never be noticed, or noticed very soon.
2377 		 */
2378 #ifdef SMP
2379 		if (TD_IS_RUNNING(td) && td != curthread)
2380 			forward_signal(td);
2381 #endif
2382 	}
2383 out:
2384 	PROC_SUNLOCK(p);
2385 	thread_unlock(td);
2386 	if (wakeup_swapper)
2387 		kick_proc0();
2388 }
2389 
2390 static void
2391 sig_suspend_threads(struct thread *td, struct proc *p, int sending)
2392 {
2393 	struct thread *td2;
2394 
2395 	PROC_LOCK_ASSERT(p, MA_OWNED);
2396 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2397 
2398 	FOREACH_THREAD_IN_PROC(p, td2) {
2399 		thread_lock(td2);
2400 		td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
2401 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2402 		    (td2->td_flags & TDF_SINTR)) {
2403 			if (td2->td_flags & TDF_SBDRY) {
2404 				/*
2405 				 * Once a thread is asleep with
2406 				 * TDF_SBDRY set, it should never
2407 				 * become suspended due to this check.
2408 				 */
2409 				KASSERT(!TD_IS_SUSPENDED(td2),
2410 				    ("thread with deferred stops suspended"));
2411 			} else if (!TD_IS_SUSPENDED(td2)) {
2412 				thread_suspend_one(td2);
2413 			}
2414 		} else if (!TD_IS_SUSPENDED(td2)) {
2415 			if (sending || td != td2)
2416 				td2->td_flags |= TDF_ASTPENDING;
2417 #ifdef SMP
2418 			if (TD_IS_RUNNING(td2) && td2 != td)
2419 				forward_signal(td2);
2420 #endif
2421 		}
2422 		thread_unlock(td2);
2423 	}
2424 }
2425 
2426 int
2427 ptracestop(struct thread *td, int sig)
2428 {
2429 	struct proc *p = td->td_proc;
2430 
2431 	PROC_LOCK_ASSERT(p, MA_OWNED);
2432 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2433 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2434 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2435 
2436 	td->td_dbgflags |= TDB_XSIG;
2437 	td->td_xsig = sig;
2438 	PROC_SLOCK(p);
2439 	while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2440 		if (p->p_flag & P_SINGLE_EXIT) {
2441 			td->td_dbgflags &= ~TDB_XSIG;
2442 			PROC_SUNLOCK(p);
2443 			return (sig);
2444 		}
2445 		/*
2446 		 * Just make wait() to work, the last stopped thread
2447 		 * will win.
2448 		 */
2449 		p->p_xstat = sig;
2450 		p->p_xthread = td;
2451 		p->p_flag |= (P_STOPPED_SIG|P_STOPPED_TRACE);
2452 		sig_suspend_threads(td, p, 0);
2453 		if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2454 			td->td_dbgflags &= ~TDB_STOPATFORK;
2455 			cv_broadcast(&p->p_dbgwait);
2456 		}
2457 stopme:
2458 		thread_suspend_switch(td);
2459 		if (p->p_xthread == td)
2460 			p->p_xthread = NULL;
2461 		if (!(p->p_flag & P_TRACED))
2462 			break;
2463 		if (td->td_dbgflags & TDB_SUSPEND) {
2464 			if (p->p_flag & P_SINGLE_EXIT)
2465 				break;
2466 			goto stopme;
2467 		}
2468 	}
2469 	PROC_SUNLOCK(p);
2470 	return (td->td_xsig);
2471 }
2472 
2473 static void
2474 reschedule_signals(struct proc *p, sigset_t block, int flags)
2475 {
2476 	struct sigacts *ps;
2477 	struct thread *td;
2478 	int sig;
2479 
2480 	PROC_LOCK_ASSERT(p, MA_OWNED);
2481 	if (SIGISEMPTY(p->p_siglist))
2482 		return;
2483 	ps = p->p_sigacts;
2484 	SIGSETAND(block, p->p_siglist);
2485 	while ((sig = sig_ffs(&block)) != 0) {
2486 		SIGDELSET(block, sig);
2487 		td = sigtd(p, sig, 0);
2488 		signotify(td);
2489 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2490 			mtx_lock(&ps->ps_mtx);
2491 		if (p->p_flag & P_TRACED || SIGISMEMBER(ps->ps_sigcatch, sig))
2492 			tdsigwakeup(td, sig, SIG_CATCH,
2493 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2494 			     ERESTART));
2495 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2496 			mtx_unlock(&ps->ps_mtx);
2497 	}
2498 }
2499 
2500 void
2501 tdsigcleanup(struct thread *td)
2502 {
2503 	struct proc *p;
2504 	sigset_t unblocked;
2505 
2506 	p = td->td_proc;
2507 	PROC_LOCK_ASSERT(p, MA_OWNED);
2508 
2509 	sigqueue_flush(&td->td_sigqueue);
2510 	if (p->p_numthreads == 1)
2511 		return;
2512 
2513 	/*
2514 	 * Since we cannot handle signals, notify signal post code
2515 	 * about this by filling the sigmask.
2516 	 *
2517 	 * Also, if needed, wake up thread(s) that do not block the
2518 	 * same signals as the exiting thread, since the thread might
2519 	 * have been selected for delivery and woken up.
2520 	 */
2521 	SIGFILLSET(unblocked);
2522 	SIGSETNAND(unblocked, td->td_sigmask);
2523 	SIGFILLSET(td->td_sigmask);
2524 	reschedule_signals(p, unblocked, 0);
2525 
2526 }
2527 
2528 /*
2529  * Defer the delivery of SIGSTOP for the current thread.  Returns true
2530  * if stops were deferred and false if they were already deferred.
2531  */
2532 int
2533 sigdeferstop(void)
2534 {
2535 	struct thread *td;
2536 
2537 	td = curthread;
2538 	if (td->td_flags & TDF_SBDRY)
2539 		return (0);
2540 	thread_lock(td);
2541 	td->td_flags |= TDF_SBDRY;
2542 	thread_unlock(td);
2543 	return (1);
2544 }
2545 
2546 /*
2547  * Permit the delivery of SIGSTOP for the current thread.  This does
2548  * not immediately suspend if a stop was posted.  Instead, the thread
2549  * will suspend either via ast() or a subsequent interruptible sleep.
2550  */
2551 void
2552 sigallowstop()
2553 {
2554 	struct thread *td;
2555 
2556 	td = curthread;
2557 	thread_lock(td);
2558 	td->td_flags &= ~TDF_SBDRY;
2559 	thread_unlock(td);
2560 }
2561 
2562 /*
2563  * If the current process has received a signal (should be caught or cause
2564  * termination, should interrupt current syscall), return the signal number.
2565  * Stop signals with default action are processed immediately, then cleared;
2566  * they aren't returned.  This is checked after each entry to the system for
2567  * a syscall or trap (though this can usually be done without calling issignal
2568  * by checking the pending signal masks in cursig.) The normal call
2569  * sequence is
2570  *
2571  *	while (sig = cursig(curthread))
2572  *		postsig(sig);
2573  */
2574 static int
2575 issignal(struct thread *td)
2576 {
2577 	struct proc *p;
2578 	struct sigacts *ps;
2579 	struct sigqueue *queue;
2580 	sigset_t sigpending;
2581 	int sig, prop, newsig;
2582 
2583 	p = td->td_proc;
2584 	ps = p->p_sigacts;
2585 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2586 	PROC_LOCK_ASSERT(p, MA_OWNED);
2587 	for (;;) {
2588 		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
2589 
2590 		sigpending = td->td_sigqueue.sq_signals;
2591 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
2592 		SIGSETNAND(sigpending, td->td_sigmask);
2593 
2594 		if (p->p_flag & P_PPWAIT || td->td_flags & TDF_SBDRY)
2595 			SIG_STOPSIGMASK(sigpending);
2596 		if (SIGISEMPTY(sigpending))	/* no signal to send */
2597 			return (0);
2598 		sig = sig_ffs(&sigpending);
2599 
2600 		if (p->p_stops & S_SIG) {
2601 			mtx_unlock(&ps->ps_mtx);
2602 			stopevent(p, S_SIG, sig);
2603 			mtx_lock(&ps->ps_mtx);
2604 		}
2605 
2606 		/*
2607 		 * We should see pending but ignored signals
2608 		 * only if P_TRACED was on when they were posted.
2609 		 */
2610 		if (SIGISMEMBER(ps->ps_sigignore, sig) && (traced == 0)) {
2611 			sigqueue_delete(&td->td_sigqueue, sig);
2612 			sigqueue_delete(&p->p_sigqueue, sig);
2613 			continue;
2614 		}
2615 		if (p->p_flag & P_TRACED && (p->p_flag & P_PPTRACE) == 0) {
2616 			/*
2617 			 * If traced, always stop.
2618 			 * Remove old signal from queue before the stop.
2619 			 * XXX shrug off debugger, it causes siginfo to
2620 			 * be thrown away.
2621 			 */
2622 			queue = &td->td_sigqueue;
2623 			td->td_dbgksi.ksi_signo = 0;
2624 			if (sigqueue_get(queue, sig, &td->td_dbgksi) == 0) {
2625 				queue = &p->p_sigqueue;
2626 				sigqueue_get(queue, sig, &td->td_dbgksi);
2627 			}
2628 
2629 			mtx_unlock(&ps->ps_mtx);
2630 			newsig = ptracestop(td, sig);
2631 			mtx_lock(&ps->ps_mtx);
2632 
2633 			if (sig != newsig) {
2634 
2635 				/*
2636 				 * If parent wants us to take the signal,
2637 				 * then it will leave it in p->p_xstat;
2638 				 * otherwise we just look for signals again.
2639 				*/
2640 				if (newsig == 0)
2641 					continue;
2642 				sig = newsig;
2643 
2644 				/*
2645 				 * Put the new signal into td_sigqueue. If the
2646 				 * signal is being masked, look for other
2647 				 * signals.
2648 				 */
2649 				sigqueue_add(queue, sig, NULL);
2650 				if (SIGISMEMBER(td->td_sigmask, sig))
2651 					continue;
2652 				signotify(td);
2653 			} else {
2654 				if (td->td_dbgksi.ksi_signo != 0) {
2655 					td->td_dbgksi.ksi_flags |= KSI_HEAD;
2656 					if (sigqueue_add(&td->td_sigqueue, sig,
2657 					    &td->td_dbgksi) != 0)
2658 						td->td_dbgksi.ksi_signo = 0;
2659 				}
2660 				if (td->td_dbgksi.ksi_signo == 0)
2661 					sigqueue_add(&td->td_sigqueue, sig,
2662 					    NULL);
2663 			}
2664 
2665 			/*
2666 			 * If the traced bit got turned off, go back up
2667 			 * to the top to rescan signals.  This ensures
2668 			 * that p_sig* and p_sigact are consistent.
2669 			 */
2670 			if ((p->p_flag & P_TRACED) == 0)
2671 				continue;
2672 		}
2673 
2674 		prop = sigprop(sig);
2675 
2676 		/*
2677 		 * Decide whether the signal should be returned.
2678 		 * Return the signal's number, or fall through
2679 		 * to clear it from the pending mask.
2680 		 */
2681 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2682 
2683 		case (intptr_t)SIG_DFL:
2684 			/*
2685 			 * Don't take default actions on system processes.
2686 			 */
2687 			if (p->p_pid <= 1) {
2688 #ifdef DIAGNOSTIC
2689 				/*
2690 				 * Are you sure you want to ignore SIGSEGV
2691 				 * in init? XXX
2692 				 */
2693 				printf("Process (pid %lu) got signal %d\n",
2694 					(u_long)p->p_pid, sig);
2695 #endif
2696 				break;		/* == ignore */
2697 			}
2698 			/*
2699 			 * If there is a pending stop signal to process
2700 			 * with default action, stop here,
2701 			 * then clear the signal.  However,
2702 			 * if process is member of an orphaned
2703 			 * process group, ignore tty stop signals.
2704 			 */
2705 			if (prop & SA_STOP) {
2706 				if (p->p_flag & (P_TRACED|P_WEXIT) ||
2707 				    (p->p_pgrp->pg_jobc == 0 &&
2708 				     prop & SA_TTYSTOP))
2709 					break;	/* == ignore */
2710 				mtx_unlock(&ps->ps_mtx);
2711 				WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2712 				    &p->p_mtx.lock_object, "Catching SIGSTOP");
2713 				p->p_flag |= P_STOPPED_SIG;
2714 				p->p_xstat = sig;
2715 				PROC_SLOCK(p);
2716 				sig_suspend_threads(td, p, 0);
2717 				thread_suspend_switch(td);
2718 				PROC_SUNLOCK(p);
2719 				mtx_lock(&ps->ps_mtx);
2720 				break;
2721 			} else if (prop & SA_IGNORE) {
2722 				/*
2723 				 * Except for SIGCONT, shouldn't get here.
2724 				 * Default action is to ignore; drop it.
2725 				 */
2726 				break;		/* == ignore */
2727 			} else
2728 				return (sig);
2729 			/*NOTREACHED*/
2730 
2731 		case (intptr_t)SIG_IGN:
2732 			/*
2733 			 * Masking above should prevent us ever trying
2734 			 * to take action on an ignored signal other
2735 			 * than SIGCONT, unless process is traced.
2736 			 */
2737 			if ((prop & SA_CONT) == 0 &&
2738 			    (p->p_flag & P_TRACED) == 0)
2739 				printf("issignal\n");
2740 			break;		/* == ignore */
2741 
2742 		default:
2743 			/*
2744 			 * This signal has an action, let
2745 			 * postsig() process it.
2746 			 */
2747 			return (sig);
2748 		}
2749 		sigqueue_delete(&td->td_sigqueue, sig);	/* take the signal! */
2750 		sigqueue_delete(&p->p_sigqueue, sig);
2751 	}
2752 	/* NOTREACHED */
2753 }
2754 
2755 void
2756 thread_stopped(struct proc *p)
2757 {
2758 	int n;
2759 
2760 	PROC_LOCK_ASSERT(p, MA_OWNED);
2761 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2762 	n = p->p_suspcount;
2763 	if (p == curproc)
2764 		n++;
2765 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
2766 		PROC_SUNLOCK(p);
2767 		p->p_flag &= ~P_WAITED;
2768 		PROC_LOCK(p->p_pptr);
2769 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
2770 			CLD_TRAPPED : CLD_STOPPED);
2771 		PROC_UNLOCK(p->p_pptr);
2772 		PROC_SLOCK(p);
2773 	}
2774 }
2775 
2776 /*
2777  * Take the action for the specified signal
2778  * from the current set of pending signals.
2779  */
2780 int
2781 postsig(sig)
2782 	register int sig;
2783 {
2784 	struct thread *td = curthread;
2785 	register struct proc *p = td->td_proc;
2786 	struct sigacts *ps;
2787 	sig_t action;
2788 	ksiginfo_t ksi;
2789 	sigset_t returnmask, mask;
2790 
2791 	KASSERT(sig != 0, ("postsig"));
2792 
2793 	PROC_LOCK_ASSERT(p, MA_OWNED);
2794 	ps = p->p_sigacts;
2795 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2796 	ksiginfo_init(&ksi);
2797 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
2798 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
2799 		return (0);
2800 	ksi.ksi_signo = sig;
2801 	if (ksi.ksi_code == SI_TIMER)
2802 		itimer_accept(p, ksi.ksi_timerid, &ksi);
2803 	action = ps->ps_sigact[_SIG_IDX(sig)];
2804 #ifdef KTRACE
2805 	if (KTRPOINT(td, KTR_PSIG))
2806 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
2807 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
2808 #endif
2809 	if (p->p_stops & S_SIG) {
2810 		mtx_unlock(&ps->ps_mtx);
2811 		stopevent(p, S_SIG, sig);
2812 		mtx_lock(&ps->ps_mtx);
2813 	}
2814 
2815 	if (action == SIG_DFL) {
2816 		/*
2817 		 * Default action, where the default is to kill
2818 		 * the process.  (Other cases were ignored above.)
2819 		 */
2820 		mtx_unlock(&ps->ps_mtx);
2821 		sigexit(td, sig);
2822 		/* NOTREACHED */
2823 	} else {
2824 		/*
2825 		 * If we get here, the signal must be caught.
2826 		 */
2827 		KASSERT(action != SIG_IGN && !SIGISMEMBER(td->td_sigmask, sig),
2828 		    ("postsig action"));
2829 		/*
2830 		 * Set the new mask value and also defer further
2831 		 * occurrences of this signal.
2832 		 *
2833 		 * Special case: user has done a sigsuspend.  Here the
2834 		 * current mask is not of interest, but rather the
2835 		 * mask from before the sigsuspend is what we want
2836 		 * restored after the signal processing is completed.
2837 		 */
2838 		if (td->td_pflags & TDP_OLDMASK) {
2839 			returnmask = td->td_oldsigmask;
2840 			td->td_pflags &= ~TDP_OLDMASK;
2841 		} else
2842 			returnmask = td->td_sigmask;
2843 
2844 		mask = ps->ps_catchmask[_SIG_IDX(sig)];
2845 		if (!SIGISMEMBER(ps->ps_signodefer, sig))
2846 			SIGADDSET(mask, sig);
2847 		kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
2848 		    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
2849 
2850 		if (SIGISMEMBER(ps->ps_sigreset, sig)) {
2851 			/*
2852 			 * See kern_sigaction() for origin of this code.
2853 			 */
2854 			SIGDELSET(ps->ps_sigcatch, sig);
2855 			if (sig != SIGCONT &&
2856 			    sigprop(sig) & SA_IGNORE)
2857 				SIGADDSET(ps->ps_sigignore, sig);
2858 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2859 		}
2860 		td->td_ru.ru_nsignals++;
2861 		if (p->p_sig == sig) {
2862 			p->p_code = 0;
2863 			p->p_sig = 0;
2864 		}
2865 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
2866 	}
2867 	return (1);
2868 }
2869 
2870 /*
2871  * Kill the current process for stated reason.
2872  */
2873 void
2874 killproc(p, why)
2875 	struct proc *p;
2876 	char *why;
2877 {
2878 
2879 	PROC_LOCK_ASSERT(p, MA_OWNED);
2880 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
2881 	    p->p_comm);
2882 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n", p->p_pid,
2883 	    p->p_comm, p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2884 	p->p_flag |= P_WKILLED;
2885 	kern_psignal(p, SIGKILL);
2886 }
2887 
2888 /*
2889  * Force the current process to exit with the specified signal, dumping core
2890  * if appropriate.  We bypass the normal tests for masked and caught signals,
2891  * allowing unrecoverable failures to terminate the process without changing
2892  * signal state.  Mark the accounting record with the signal termination.
2893  * If dumping core, save the signal number for the debugger.  Calls exit and
2894  * does not return.
2895  */
2896 void
2897 sigexit(td, sig)
2898 	struct thread *td;
2899 	int sig;
2900 {
2901 	struct proc *p = td->td_proc;
2902 
2903 	PROC_LOCK_ASSERT(p, MA_OWNED);
2904 	p->p_acflag |= AXSIG;
2905 	/*
2906 	 * We must be single-threading to generate a core dump.  This
2907 	 * ensures that the registers in the core file are up-to-date.
2908 	 * Also, the ELF dump handler assumes that the thread list doesn't
2909 	 * change out from under it.
2910 	 *
2911 	 * XXX If another thread attempts to single-thread before us
2912 	 *     (e.g. via fork()), we won't get a dump at all.
2913 	 */
2914 	if ((sigprop(sig) & SA_CORE) && (thread_single(SINGLE_NO_EXIT) == 0)) {
2915 		p->p_sig = sig;
2916 		/*
2917 		 * Log signals which would cause core dumps
2918 		 * (Log as LOG_INFO to appease those who don't want
2919 		 * these messages.)
2920 		 * XXX : Todo, as well as euid, write out ruid too
2921 		 * Note that coredump() drops proc lock.
2922 		 */
2923 		if (coredump(td) == 0)
2924 			sig |= WCOREFLAG;
2925 		if (kern_logsigexit)
2926 			log(LOG_INFO,
2927 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2928 			    p->p_pid, p->p_comm,
2929 			    td->td_ucred ? td->td_ucred->cr_uid : -1,
2930 			    sig &~ WCOREFLAG,
2931 			    sig & WCOREFLAG ? " (core dumped)" : "");
2932 	} else
2933 		PROC_UNLOCK(p);
2934 	exit1(td, W_EXITCODE(0, sig));
2935 	/* NOTREACHED */
2936 }
2937 
2938 /*
2939  * Send queued SIGCHLD to parent when child process's state
2940  * is changed.
2941  */
2942 static void
2943 sigparent(struct proc *p, int reason, int status)
2944 {
2945 	PROC_LOCK_ASSERT(p, MA_OWNED);
2946 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
2947 
2948 	if (p->p_ksi != NULL) {
2949 		p->p_ksi->ksi_signo  = SIGCHLD;
2950 		p->p_ksi->ksi_code   = reason;
2951 		p->p_ksi->ksi_status = status;
2952 		p->p_ksi->ksi_pid    = p->p_pid;
2953 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
2954 		if (KSI_ONQ(p->p_ksi))
2955 			return;
2956 	}
2957 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
2958 }
2959 
2960 static void
2961 childproc_jobstate(struct proc *p, int reason, int sig)
2962 {
2963 	struct sigacts *ps;
2964 
2965 	PROC_LOCK_ASSERT(p, MA_OWNED);
2966 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
2967 
2968 	/*
2969 	 * Wake up parent sleeping in kern_wait(), also send
2970 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
2971 	 * that parent will awake, because parent may masked
2972 	 * the signal.
2973 	 */
2974 	p->p_pptr->p_flag |= P_STATCHILD;
2975 	wakeup(p->p_pptr);
2976 
2977 	ps = p->p_pptr->p_sigacts;
2978 	mtx_lock(&ps->ps_mtx);
2979 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
2980 		mtx_unlock(&ps->ps_mtx);
2981 		sigparent(p, reason, sig);
2982 	} else
2983 		mtx_unlock(&ps->ps_mtx);
2984 }
2985 
2986 void
2987 childproc_stopped(struct proc *p, int reason)
2988 {
2989 	/* p_xstat is a plain signal number, not a full wait() status here. */
2990 	childproc_jobstate(p, reason, p->p_xstat);
2991 }
2992 
2993 void
2994 childproc_continued(struct proc *p)
2995 {
2996 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
2997 }
2998 
2999 void
3000 childproc_exited(struct proc *p)
3001 {
3002 	int reason;
3003 	int xstat = p->p_xstat; /* convert to int */
3004 	int status;
3005 
3006 	if (WCOREDUMP(xstat))
3007 		reason = CLD_DUMPED, status = WTERMSIG(xstat);
3008 	else if (WIFSIGNALED(xstat))
3009 		reason = CLD_KILLED, status = WTERMSIG(xstat);
3010 	else
3011 		reason = CLD_EXITED, status = WEXITSTATUS(xstat);
3012 	/*
3013 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3014 	 * done in exit1().
3015 	 */
3016 	sigparent(p, reason, status);
3017 }
3018 
3019 /*
3020  * We only have 1 character for the core count in the format
3021  * string, so the range will be 0-9
3022  */
3023 #define MAX_NUM_CORES 10
3024 static int num_cores = 5;
3025 
3026 static int
3027 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3028 {
3029 	int error;
3030 	int new_val;
3031 
3032 	new_val = num_cores;
3033 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3034 	if (error != 0 || req->newptr == NULL)
3035 		return (error);
3036 	if (new_val > MAX_NUM_CORES)
3037 		new_val = MAX_NUM_CORES;
3038 	if (new_val < 0)
3039 		new_val = 0;
3040 	num_cores = new_val;
3041 	return (0);
3042 }
3043 SYSCTL_PROC(_debug, OID_AUTO, ncores, CTLTYPE_INT|CTLFLAG_RW,
3044 	    0, sizeof(int), sysctl_debug_num_cores_check, "I", "");
3045 
3046 #if defined(COMPRESS_USER_CORES)
3047 int compress_user_cores = 1;
3048 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores, CTLFLAG_RW,
3049     &compress_user_cores, 0, "Compression of user corefiles");
3050 
3051 int compress_user_cores_gzlevel = -1; /* default level */
3052 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_gzlevel, CTLFLAG_RW,
3053     &compress_user_cores_gzlevel, -1, "Corefile gzip compression level");
3054 
3055 #define GZ_SUFFIX	".gz"
3056 #define GZ_SUFFIX_LEN	3
3057 #endif
3058 
3059 static char corefilename[MAXPATHLEN] = {"%N.core"};
3060 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3061 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
3062     sizeof(corefilename), "Process corefile name format string");
3063 
3064 /*
3065  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3066  * Expand the name described in corefilename, using name, uid, and pid
3067  * and open/create core file.
3068  * corefilename is a printf-like string, with three format specifiers:
3069  *	%N	name of process ("name")
3070  *	%P	process id (pid)
3071  *	%U	user id (uid)
3072  * For example, "%N.core" is the default; they can be disabled completely
3073  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3074  * This is controlled by the sysctl variable kern.corefile (see above).
3075  */
3076 static int
3077 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3078     int compress, struct vnode **vpp, char **namep)
3079 {
3080 	struct nameidata nd;
3081 	struct sbuf sb;
3082 	const char *format;
3083 	char *hostname, *name;
3084 	int indexpos, i, error, cmode, flags, oflags;
3085 
3086 	hostname = NULL;
3087 	format = corefilename;
3088 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3089 	indexpos = -1;
3090 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3091 	for (i = 0; format[i] != '\0'; i++) {
3092 		switch (format[i]) {
3093 		case '%':	/* Format character */
3094 			i++;
3095 			switch (format[i]) {
3096 			case '%':
3097 				sbuf_putc(&sb, '%');
3098 				break;
3099 			case 'H':	/* hostname */
3100 				if (hostname == NULL) {
3101 					hostname = malloc(MAXHOSTNAMELEN,
3102 					    M_TEMP, M_WAITOK);
3103 				}
3104 				getcredhostname(td->td_ucred, hostname,
3105 				    MAXHOSTNAMELEN);
3106 				sbuf_printf(&sb, "%s", hostname);
3107 				break;
3108 			case 'I':	/* autoincrementing index */
3109 				sbuf_printf(&sb, "0");
3110 				indexpos = sbuf_len(&sb) - 1;
3111 				break;
3112 			case 'N':	/* process name */
3113 				sbuf_printf(&sb, "%s", comm);
3114 				break;
3115 			case 'P':	/* process id */
3116 				sbuf_printf(&sb, "%u", pid);
3117 				break;
3118 			case 'U':	/* user id */
3119 				sbuf_printf(&sb, "%u", uid);
3120 				break;
3121 			default:
3122 				log(LOG_ERR,
3123 				    "Unknown format character %c in "
3124 				    "corename `%s'\n", format[i], format);
3125 				break;
3126 			}
3127 			break;
3128 		default:
3129 			sbuf_putc(&sb, format[i]);
3130 			break;
3131 		}
3132 	}
3133 	free(hostname, M_TEMP);
3134 #ifdef COMPRESS_USER_CORES
3135 	if (compress)
3136 		sbuf_printf(&sb, GZ_SUFFIX);
3137 #endif
3138 	if (sbuf_error(&sb) != 0) {
3139 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3140 		    "long\n", (long)pid, comm, (u_long)uid);
3141 		sbuf_delete(&sb);
3142 		free(name, M_TEMP);
3143 		return (ENOMEM);
3144 	}
3145 	sbuf_finish(&sb);
3146 	sbuf_delete(&sb);
3147 
3148 	cmode = S_IRUSR | S_IWUSR;
3149 	oflags = VN_OPEN_NOAUDIT | (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3150 
3151 	/*
3152 	 * If the core format has a %I in it, then we need to check
3153 	 * for existing corefiles before returning a name.
3154 	 * To do this we iterate over 0..num_cores to find a
3155 	 * non-existing core file name to use.
3156 	 */
3157 	if (indexpos != -1) {
3158 		for (i = 0; i < num_cores; i++) {
3159 			flags = O_CREAT | O_EXCL | FWRITE | O_NOFOLLOW;
3160 			name[indexpos] = '0' + i;
3161 			NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3162 			error = vn_open_cred(&nd, &flags, cmode, oflags,
3163 			    td->td_ucred, NULL);
3164 			if (error) {
3165 				if (error == EEXIST)
3166 					continue;
3167 				log(LOG_ERR,
3168 				    "pid %d (%s), uid (%u):  Path `%s' failed "
3169 				    "on initial open test, error = %d\n",
3170 				    pid, comm, uid, name, error);
3171 			}
3172 			goto out;
3173 		}
3174 	}
3175 
3176 	flags = O_CREAT | FWRITE | O_NOFOLLOW;
3177 	NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3178 	error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred, NULL);
3179 out:
3180 	if (error) {
3181 #ifdef AUDIT
3182 		audit_proc_coredump(td, name, error);
3183 #endif
3184 		free(name, M_TEMP);
3185 		return (error);
3186 	}
3187 	NDFREE(&nd, NDF_ONLY_PNBUF);
3188 	*vpp = nd.ni_vp;
3189 	*namep = name;
3190 	return (0);
3191 }
3192 
3193 /*
3194  * Dump a process' core.  The main routine does some
3195  * policy checking, and creates the name of the coredump;
3196  * then it passes on a vnode and a size limit to the process-specific
3197  * coredump routine if there is one; if there _is not_ one, it returns
3198  * ENOSYS; otherwise it returns the error from the process-specific routine.
3199  */
3200 
3201 static int
3202 coredump(struct thread *td)
3203 {
3204 	struct proc *p = td->td_proc;
3205 	struct ucred *cred = td->td_ucred;
3206 	struct vnode *vp;
3207 	struct flock lf;
3208 	struct vattr vattr;
3209 	int error, error1, locked;
3210 	struct mount *mp;
3211 	char *name;			/* name of corefile */
3212 	off_t limit;
3213 	int compress;
3214 
3215 #ifdef COMPRESS_USER_CORES
3216 	compress = compress_user_cores;
3217 #else
3218 	compress = 0;
3219 #endif
3220 	PROC_LOCK_ASSERT(p, MA_OWNED);
3221 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3222 	_STOPEVENT(p, S_CORE, 0);
3223 
3224 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0)) {
3225 		PROC_UNLOCK(p);
3226 		return (EFAULT);
3227 	}
3228 
3229 	/*
3230 	 * Note that the bulk of limit checking is done after
3231 	 * the corefile is created.  The exception is if the limit
3232 	 * for corefiles is 0, in which case we don't bother
3233 	 * creating the corefile at all.  This layout means that
3234 	 * a corefile is truncated instead of not being created,
3235 	 * if it is larger than the limit.
3236 	 */
3237 	limit = (off_t)lim_cur(p, RLIMIT_CORE);
3238 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
3239 		PROC_UNLOCK(p);
3240 		return (EFBIG);
3241 	}
3242 	PROC_UNLOCK(p);
3243 
3244 restart:
3245 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td, compress,
3246 	    &vp, &name);
3247 	if (error != 0)
3248 		return (error);
3249 
3250 	/* Don't dump to non-regular files or files with links. */
3251 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
3252 	    vattr.va_nlink != 1) {
3253 		VOP_UNLOCK(vp, 0);
3254 		error = EFAULT;
3255 		goto close;
3256 	}
3257 
3258 	VOP_UNLOCK(vp, 0);
3259 	lf.l_whence = SEEK_SET;
3260 	lf.l_start = 0;
3261 	lf.l_len = 0;
3262 	lf.l_type = F_WRLCK;
3263 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
3264 
3265 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
3266 		lf.l_type = F_UNLCK;
3267 		if (locked)
3268 			VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3269 		if ((error = vn_close(vp, FWRITE, cred, td)) != 0)
3270 			goto out;
3271 		if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0)
3272 			goto out;
3273 		free(name, M_TEMP);
3274 		goto restart;
3275 	}
3276 
3277 	VATTR_NULL(&vattr);
3278 	vattr.va_size = 0;
3279 	if (set_core_nodump_flag)
3280 		vattr.va_flags = UF_NODUMP;
3281 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3282 	VOP_SETATTR(vp, &vattr, cred);
3283 	VOP_UNLOCK(vp, 0);
3284 	vn_finished_write(mp);
3285 	PROC_LOCK(p);
3286 	p->p_acflag |= ACORE;
3287 	PROC_UNLOCK(p);
3288 
3289 	if (p->p_sysent->sv_coredump != NULL) {
3290 		error = p->p_sysent->sv_coredump(td, vp, limit,
3291 		    compress ? IMGACT_CORE_COMPRESS : 0);
3292 	} else {
3293 		error = ENOSYS;
3294 	}
3295 
3296 	if (locked) {
3297 		lf.l_type = F_UNLCK;
3298 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3299 	}
3300 close:
3301 	error1 = vn_close(vp, FWRITE, cred, td);
3302 	if (error == 0)
3303 		error = error1;
3304 out:
3305 #ifdef AUDIT
3306 	audit_proc_coredump(td, name, error);
3307 #endif
3308 	free(name, M_TEMP);
3309 	return (error);
3310 }
3311 
3312 /*
3313  * Nonexistent system call-- signal process (may want to handle it).  Flag
3314  * error in case process won't see signal immediately (blocked or ignored).
3315  */
3316 #ifndef _SYS_SYSPROTO_H_
3317 struct nosys_args {
3318 	int	dummy;
3319 };
3320 #endif
3321 /* ARGSUSED */
3322 int
3323 nosys(td, args)
3324 	struct thread *td;
3325 	struct nosys_args *args;
3326 {
3327 	struct proc *p = td->td_proc;
3328 
3329 	PROC_LOCK(p);
3330 	tdsignal(td, SIGSYS);
3331 	PROC_UNLOCK(p);
3332 	return (ENOSYS);
3333 }
3334 
3335 /*
3336  * Send a SIGIO or SIGURG signal to a process or process group using stored
3337  * credentials rather than those of the current process.
3338  */
3339 void
3340 pgsigio(sigiop, sig, checkctty)
3341 	struct sigio **sigiop;
3342 	int sig, checkctty;
3343 {
3344 	ksiginfo_t ksi;
3345 	struct sigio *sigio;
3346 
3347 	ksiginfo_init(&ksi);
3348 	ksi.ksi_signo = sig;
3349 	ksi.ksi_code = SI_KERNEL;
3350 
3351 	SIGIO_LOCK();
3352 	sigio = *sigiop;
3353 	if (sigio == NULL) {
3354 		SIGIO_UNLOCK();
3355 		return;
3356 	}
3357 	if (sigio->sio_pgid > 0) {
3358 		PROC_LOCK(sigio->sio_proc);
3359 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
3360 			kern_psignal(sigio->sio_proc, sig);
3361 		PROC_UNLOCK(sigio->sio_proc);
3362 	} else if (sigio->sio_pgid < 0) {
3363 		struct proc *p;
3364 
3365 		PGRP_LOCK(sigio->sio_pgrp);
3366 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
3367 			PROC_LOCK(p);
3368 			if (p->p_state == PRS_NORMAL &&
3369 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
3370 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
3371 				kern_psignal(p, sig);
3372 			PROC_UNLOCK(p);
3373 		}
3374 		PGRP_UNLOCK(sigio->sio_pgrp);
3375 	}
3376 	SIGIO_UNLOCK();
3377 }
3378 
3379 static int
3380 filt_sigattach(struct knote *kn)
3381 {
3382 	struct proc *p = curproc;
3383 
3384 	kn->kn_ptr.p_proc = p;
3385 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
3386 
3387 	knlist_add(&p->p_klist, kn, 0);
3388 
3389 	return (0);
3390 }
3391 
3392 static void
3393 filt_sigdetach(struct knote *kn)
3394 {
3395 	struct proc *p = kn->kn_ptr.p_proc;
3396 
3397 	knlist_remove(&p->p_klist, kn, 0);
3398 }
3399 
3400 /*
3401  * signal knotes are shared with proc knotes, so we apply a mask to
3402  * the hint in order to differentiate them from process hints.  This
3403  * could be avoided by using a signal-specific knote list, but probably
3404  * isn't worth the trouble.
3405  */
3406 static int
3407 filt_signal(struct knote *kn, long hint)
3408 {
3409 
3410 	if (hint & NOTE_SIGNAL) {
3411 		hint &= ~NOTE_SIGNAL;
3412 
3413 		if (kn->kn_id == hint)
3414 			kn->kn_data++;
3415 	}
3416 	return (kn->kn_data != 0);
3417 }
3418 
3419 struct sigacts *
3420 sigacts_alloc(void)
3421 {
3422 	struct sigacts *ps;
3423 
3424 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
3425 	ps->ps_refcnt = 1;
3426 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
3427 	return (ps);
3428 }
3429 
3430 void
3431 sigacts_free(struct sigacts *ps)
3432 {
3433 
3434 	mtx_lock(&ps->ps_mtx);
3435 	ps->ps_refcnt--;
3436 	if (ps->ps_refcnt == 0) {
3437 		mtx_destroy(&ps->ps_mtx);
3438 		free(ps, M_SUBPROC);
3439 	} else
3440 		mtx_unlock(&ps->ps_mtx);
3441 }
3442 
3443 struct sigacts *
3444 sigacts_hold(struct sigacts *ps)
3445 {
3446 	mtx_lock(&ps->ps_mtx);
3447 	ps->ps_refcnt++;
3448 	mtx_unlock(&ps->ps_mtx);
3449 	return (ps);
3450 }
3451 
3452 void
3453 sigacts_copy(struct sigacts *dest, struct sigacts *src)
3454 {
3455 
3456 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
3457 	mtx_lock(&src->ps_mtx);
3458 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
3459 	mtx_unlock(&src->ps_mtx);
3460 }
3461 
3462 int
3463 sigacts_shared(struct sigacts *ps)
3464 {
3465 	int shared;
3466 
3467 	mtx_lock(&ps->ps_mtx);
3468 	shared = ps->ps_refcnt > 1;
3469 	mtx_unlock(&ps->ps_mtx);
3470 	return (shared);
3471 }
3472