xref: /freebsd/sys/kern/kern_sig.c (revision 75855cb6c09a8ba100173ba7a124cfa5b83dce01)
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 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/signalvar.h>
47 #include <sys/vnode.h>
48 #include <sys/acct.h>
49 #include <sys/bus.h>
50 #include <sys/capsicum.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/refcount.h>
62 #include <sys/namei.h>
63 #include <sys/proc.h>
64 #include <sys/procdesc.h>
65 #include <sys/posix4.h>
66 #include <sys/pioctl.h>
67 #include <sys/racct.h>
68 #include <sys/resourcevar.h>
69 #include <sys/sdt.h>
70 #include <sys/sbuf.h>
71 #include <sys/sleepqueue.h>
72 #include <sys/smp.h>
73 #include <sys/stat.h>
74 #include <sys/sx.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysctl.h>
77 #include <sys/sysent.h>
78 #include <sys/syslog.h>
79 #include <sys/sysproto.h>
80 #include <sys/timers.h>
81 #include <sys/unistd.h>
82 #include <sys/wait.h>
83 #include <vm/vm.h>
84 #include <vm/vm_extern.h>
85 #include <vm/uma.h>
86 
87 #include <sys/jail.h>
88 
89 #include <machine/cpu.h>
90 
91 #include <security/audit/audit.h>
92 
93 #define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
94 
95 SDT_PROVIDER_DECLARE(proc);
96 SDT_PROBE_DEFINE3(proc, kernel, , signal__send, "struct thread *",
97     "struct proc *", "int");
98 SDT_PROBE_DEFINE2(proc, kernel, , signal__clear, "int",
99     "ksiginfo_t *");
100 SDT_PROBE_DEFINE3(proc, kernel, , signal__discard,
101     "struct thread *", "struct proc *", "int");
102 
103 static int	coredump(struct thread *);
104 static int	killpg1(struct thread *td, int sig, int pgid, int all,
105 		    ksiginfo_t *ksi);
106 static int	issignal(struct thread *td);
107 static int	sigprop(int sig);
108 static void	tdsigwakeup(struct thread *, int, sig_t, int);
109 static void	sig_suspend_threads(struct thread *, struct proc *, int);
110 static int	filt_sigattach(struct knote *kn);
111 static void	filt_sigdetach(struct knote *kn);
112 static int	filt_signal(struct knote *kn, long hint);
113 static struct thread *sigtd(struct proc *p, int sig, int prop);
114 static void	sigqueue_start(void);
115 
116 static uma_zone_t	ksiginfo_zone = NULL;
117 struct filterops sig_filtops = {
118 	.f_isfd = 0,
119 	.f_attach = filt_sigattach,
120 	.f_detach = filt_sigdetach,
121 	.f_event = filt_signal,
122 };
123 
124 static int	kern_logsigexit = 1;
125 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
126     &kern_logsigexit, 0,
127     "Log processes quitting on abnormal signals to syslog(3)");
128 
129 static int	kern_forcesigexit = 1;
130 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
131     &kern_forcesigexit, 0, "Force trap signal to be handled");
132 
133 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW, 0,
134     "POSIX real time signal");
135 
136 static int	max_pending_per_proc = 128;
137 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
138     &max_pending_per_proc, 0, "Max pending signals per proc");
139 
140 static int	preallocate_siginfo = 1024;
141 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
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 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
168     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
169 
170 static int	capmode_coredump;
171 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
172     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
173 
174 static int	do_coredump = 1;
175 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
176 	&do_coredump, 0, "Enable/Disable coredumps");
177 
178 static int	set_core_nodump_flag = 0;
179 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
180 	0, "Enable setting the NODUMP flag on coredump files");
181 
182 /*
183  * Signal properties and actions.
184  * The array below categorizes the signals and their default actions
185  * according to the following properties:
186  */
187 #define	SA_KILL		0x01		/* terminates process by default */
188 #define	SA_CORE		0x02		/* ditto and coredumps */
189 #define	SA_STOP		0x04		/* suspend process */
190 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
191 #define	SA_IGNORE	0x10		/* ignore by default */
192 #define	SA_CONT		0x20		/* continue if suspended */
193 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
194 
195 static int sigproptbl[NSIG] = {
196 	SA_KILL,			/* SIGHUP */
197 	SA_KILL,			/* SIGINT */
198 	SA_KILL|SA_CORE,		/* SIGQUIT */
199 	SA_KILL|SA_CORE,		/* SIGILL */
200 	SA_KILL|SA_CORE,		/* SIGTRAP */
201 	SA_KILL|SA_CORE,		/* SIGABRT */
202 	SA_KILL|SA_CORE,		/* SIGEMT */
203 	SA_KILL|SA_CORE,		/* SIGFPE */
204 	SA_KILL,			/* SIGKILL */
205 	SA_KILL|SA_CORE,		/* SIGBUS */
206 	SA_KILL|SA_CORE,		/* SIGSEGV */
207 	SA_KILL|SA_CORE,		/* SIGSYS */
208 	SA_KILL,			/* SIGPIPE */
209 	SA_KILL,			/* SIGALRM */
210 	SA_KILL,			/* SIGTERM */
211 	SA_IGNORE,			/* SIGURG */
212 	SA_STOP,			/* SIGSTOP */
213 	SA_STOP|SA_TTYSTOP,		/* SIGTSTP */
214 	SA_IGNORE|SA_CONT,		/* SIGCONT */
215 	SA_IGNORE,			/* SIGCHLD */
216 	SA_STOP|SA_TTYSTOP,		/* SIGTTIN */
217 	SA_STOP|SA_TTYSTOP,		/* SIGTTOU */
218 	SA_IGNORE,			/* SIGIO */
219 	SA_KILL,			/* SIGXCPU */
220 	SA_KILL,			/* SIGXFSZ */
221 	SA_KILL,			/* SIGVTALRM */
222 	SA_KILL,			/* SIGPROF */
223 	SA_IGNORE,			/* SIGWINCH  */
224 	SA_IGNORE,			/* SIGINFO */
225 	SA_KILL,			/* SIGUSR1 */
226 	SA_KILL,			/* SIGUSR2 */
227 };
228 
229 static void reschedule_signals(struct proc *p, sigset_t block, int flags);
230 
231 static void
232 sigqueue_start(void)
233 {
234 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
235 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
236 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
237 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
238 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
239 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
240 }
241 
242 ksiginfo_t *
243 ksiginfo_alloc(int wait)
244 {
245 	int flags;
246 
247 	flags = M_ZERO;
248 	if (! wait)
249 		flags |= M_NOWAIT;
250 	if (ksiginfo_zone != NULL)
251 		return ((ksiginfo_t *)uma_zalloc(ksiginfo_zone, flags));
252 	return (NULL);
253 }
254 
255 void
256 ksiginfo_free(ksiginfo_t *ksi)
257 {
258 	uma_zfree(ksiginfo_zone, ksi);
259 }
260 
261 static __inline int
262 ksiginfo_tryfree(ksiginfo_t *ksi)
263 {
264 	if (!(ksi->ksi_flags & KSI_EXT)) {
265 		uma_zfree(ksiginfo_zone, ksi);
266 		return (1);
267 	}
268 	return (0);
269 }
270 
271 void
272 sigqueue_init(sigqueue_t *list, struct proc *p)
273 {
274 	SIGEMPTYSET(list->sq_signals);
275 	SIGEMPTYSET(list->sq_kill);
276 	TAILQ_INIT(&list->sq_list);
277 	list->sq_proc = p;
278 	list->sq_flags = SQ_INIT;
279 }
280 
281 /*
282  * Get a signal's ksiginfo.
283  * Return:
284  *	0	-	signal not found
285  *	others	-	signal number
286  */
287 static int
288 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
289 {
290 	struct proc *p = sq->sq_proc;
291 	struct ksiginfo *ksi, *next;
292 	int count = 0;
293 
294 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
295 
296 	if (!SIGISMEMBER(sq->sq_signals, signo))
297 		return (0);
298 
299 	if (SIGISMEMBER(sq->sq_kill, signo)) {
300 		count++;
301 		SIGDELSET(sq->sq_kill, signo);
302 	}
303 
304 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
305 		if (ksi->ksi_signo == signo) {
306 			if (count == 0) {
307 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
308 				ksi->ksi_sigq = NULL;
309 				ksiginfo_copy(ksi, si);
310 				if (ksiginfo_tryfree(ksi) && p != NULL)
311 					p->p_pendingcnt--;
312 			}
313 			if (++count > 1)
314 				break;
315 		}
316 	}
317 
318 	if (count <= 1)
319 		SIGDELSET(sq->sq_signals, signo);
320 	si->ksi_signo = signo;
321 	return (signo);
322 }
323 
324 void
325 sigqueue_take(ksiginfo_t *ksi)
326 {
327 	struct ksiginfo *kp;
328 	struct proc	*p;
329 	sigqueue_t	*sq;
330 
331 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
332 		return;
333 
334 	p = sq->sq_proc;
335 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
336 	ksi->ksi_sigq = NULL;
337 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
338 		p->p_pendingcnt--;
339 
340 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
341 	     kp = TAILQ_NEXT(kp, ksi_link)) {
342 		if (kp->ksi_signo == ksi->ksi_signo)
343 			break;
344 	}
345 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo))
346 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
347 }
348 
349 static int
350 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
351 {
352 	struct proc *p = sq->sq_proc;
353 	struct ksiginfo *ksi;
354 	int ret = 0;
355 
356 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
357 
358 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
359 		SIGADDSET(sq->sq_kill, signo);
360 		goto out_set_bit;
361 	}
362 
363 	/* directly insert the ksi, don't copy it */
364 	if (si->ksi_flags & KSI_INS) {
365 		if (si->ksi_flags & KSI_HEAD)
366 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
367 		else
368 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
369 		si->ksi_sigq = sq;
370 		goto out_set_bit;
371 	}
372 
373 	if (__predict_false(ksiginfo_zone == NULL)) {
374 		SIGADDSET(sq->sq_kill, signo);
375 		goto out_set_bit;
376 	}
377 
378 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
379 		signal_overflow++;
380 		ret = EAGAIN;
381 	} else if ((ksi = ksiginfo_alloc(0)) == NULL) {
382 		signal_alloc_fail++;
383 		ret = EAGAIN;
384 	} else {
385 		if (p != NULL)
386 			p->p_pendingcnt++;
387 		ksiginfo_copy(si, ksi);
388 		ksi->ksi_signo = signo;
389 		if (si->ksi_flags & KSI_HEAD)
390 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
391 		else
392 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
393 		ksi->ksi_sigq = sq;
394 	}
395 
396 	if ((si->ksi_flags & KSI_TRAP) != 0 ||
397 	    (si->ksi_flags & KSI_SIGQ) == 0) {
398 		if (ret != 0)
399 			SIGADDSET(sq->sq_kill, signo);
400 		ret = 0;
401 		goto out_set_bit;
402 	}
403 
404 	if (ret != 0)
405 		return (ret);
406 
407 out_set_bit:
408 	SIGADDSET(sq->sq_signals, signo);
409 	return (ret);
410 }
411 
412 void
413 sigqueue_flush(sigqueue_t *sq)
414 {
415 	struct proc *p = sq->sq_proc;
416 	ksiginfo_t *ksi;
417 
418 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
419 
420 	if (p != NULL)
421 		PROC_LOCK_ASSERT(p, MA_OWNED);
422 
423 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
424 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
425 		ksi->ksi_sigq = NULL;
426 		if (ksiginfo_tryfree(ksi) && p != NULL)
427 			p->p_pendingcnt--;
428 	}
429 
430 	SIGEMPTYSET(sq->sq_signals);
431 	SIGEMPTYSET(sq->sq_kill);
432 }
433 
434 static void
435 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
436 {
437 	sigset_t tmp;
438 	struct proc *p1, *p2;
439 	ksiginfo_t *ksi, *next;
440 
441 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
442 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
443 	p1 = src->sq_proc;
444 	p2 = dst->sq_proc;
445 	/* Move siginfo to target list */
446 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
447 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
448 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
449 			if (p1 != NULL)
450 				p1->p_pendingcnt--;
451 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
452 			ksi->ksi_sigq = dst;
453 			if (p2 != NULL)
454 				p2->p_pendingcnt++;
455 		}
456 	}
457 
458 	/* Move pending bits to target list */
459 	tmp = src->sq_kill;
460 	SIGSETAND(tmp, *set);
461 	SIGSETOR(dst->sq_kill, tmp);
462 	SIGSETNAND(src->sq_kill, tmp);
463 
464 	tmp = src->sq_signals;
465 	SIGSETAND(tmp, *set);
466 	SIGSETOR(dst->sq_signals, tmp);
467 	SIGSETNAND(src->sq_signals, tmp);
468 }
469 
470 #if 0
471 static void
472 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
473 {
474 	sigset_t set;
475 
476 	SIGEMPTYSET(set);
477 	SIGADDSET(set, signo);
478 	sigqueue_move_set(src, dst, &set);
479 }
480 #endif
481 
482 static void
483 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
484 {
485 	struct proc *p = sq->sq_proc;
486 	ksiginfo_t *ksi, *next;
487 
488 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
489 
490 	/* Remove siginfo queue */
491 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
492 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
493 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
494 			ksi->ksi_sigq = NULL;
495 			if (ksiginfo_tryfree(ksi) && p != NULL)
496 				p->p_pendingcnt--;
497 		}
498 	}
499 	SIGSETNAND(sq->sq_kill, *set);
500 	SIGSETNAND(sq->sq_signals, *set);
501 }
502 
503 void
504 sigqueue_delete(sigqueue_t *sq, int signo)
505 {
506 	sigset_t set;
507 
508 	SIGEMPTYSET(set);
509 	SIGADDSET(set, signo);
510 	sigqueue_delete_set(sq, &set);
511 }
512 
513 /* Remove a set of signals for a process */
514 static void
515 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
516 {
517 	sigqueue_t worklist;
518 	struct thread *td0;
519 
520 	PROC_LOCK_ASSERT(p, MA_OWNED);
521 
522 	sigqueue_init(&worklist, NULL);
523 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
524 
525 	FOREACH_THREAD_IN_PROC(p, td0)
526 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
527 
528 	sigqueue_flush(&worklist);
529 }
530 
531 void
532 sigqueue_delete_proc(struct proc *p, int signo)
533 {
534 	sigset_t set;
535 
536 	SIGEMPTYSET(set);
537 	SIGADDSET(set, signo);
538 	sigqueue_delete_set_proc(p, &set);
539 }
540 
541 static void
542 sigqueue_delete_stopmask_proc(struct proc *p)
543 {
544 	sigset_t set;
545 
546 	SIGEMPTYSET(set);
547 	SIGADDSET(set, SIGSTOP);
548 	SIGADDSET(set, SIGTSTP);
549 	SIGADDSET(set, SIGTTIN);
550 	SIGADDSET(set, SIGTTOU);
551 	sigqueue_delete_set_proc(p, &set);
552 }
553 
554 /*
555  * Determine signal that should be delivered to thread td, the current
556  * thread, 0 if none.  If there is a pending stop signal with default
557  * action, the process stops in issignal().
558  */
559 int
560 cursig(struct thread *td)
561 {
562 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
563 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
564 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
565 	return (SIGPENDING(td) ? issignal(td) : 0);
566 }
567 
568 /*
569  * Arrange for ast() to handle unmasked pending signals on return to user
570  * mode.  This must be called whenever a signal is added to td_sigqueue or
571  * unmasked in td_sigmask.
572  */
573 void
574 signotify(struct thread *td)
575 {
576 	struct proc *p;
577 
578 	p = td->td_proc;
579 
580 	PROC_LOCK_ASSERT(p, MA_OWNED);
581 
582 	if (SIGPENDING(td)) {
583 		thread_lock(td);
584 		td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING;
585 		thread_unlock(td);
586 	}
587 }
588 
589 int
590 sigonstack(size_t sp)
591 {
592 	struct thread *td = curthread;
593 
594 	return ((td->td_pflags & TDP_ALTSTACK) ?
595 #if defined(COMPAT_43)
596 	    ((td->td_sigstk.ss_size == 0) ?
597 		(td->td_sigstk.ss_flags & SS_ONSTACK) :
598 		((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size))
599 #else
600 	    ((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size)
601 #endif
602 	    : 0);
603 }
604 
605 static __inline int
606 sigprop(int sig)
607 {
608 
609 	if (sig > 0 && sig < NSIG)
610 		return (sigproptbl[_SIG_IDX(sig)]);
611 	return (0);
612 }
613 
614 int
615 sig_ffs(sigset_t *set)
616 {
617 	int i;
618 
619 	for (i = 0; i < _SIG_WORDS; i++)
620 		if (set->__bits[i])
621 			return (ffs(set->__bits[i]) + (i * 32));
622 	return (0);
623 }
624 
625 static bool
626 sigact_flag_test(struct sigaction *act, int flag)
627 {
628 
629 	/*
630 	 * SA_SIGINFO is reset when signal disposition is set to
631 	 * ignore or default.  Other flags are kept according to user
632 	 * settings.
633 	 */
634 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
635 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
636 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
637 }
638 
639 /*
640  * kern_sigaction
641  * sigaction
642  * freebsd4_sigaction
643  * osigaction
644  */
645 int
646 kern_sigaction(td, sig, act, oact, flags)
647 	struct thread *td;
648 	register int sig;
649 	struct sigaction *act, *oact;
650 	int flags;
651 {
652 	struct sigacts *ps;
653 	struct proc *p = td->td_proc;
654 
655 	if (!_SIG_VALID(sig))
656 		return (EINVAL);
657 	if (act != NULL && act->sa_handler != SIG_DFL &&
658 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
659 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
660 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
661 		return (EINVAL);
662 
663 	PROC_LOCK(p);
664 	ps = p->p_sigacts;
665 	mtx_lock(&ps->ps_mtx);
666 	if (oact) {
667 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
668 		oact->sa_flags = 0;
669 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
670 			oact->sa_flags |= SA_ONSTACK;
671 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
672 			oact->sa_flags |= SA_RESTART;
673 		if (SIGISMEMBER(ps->ps_sigreset, sig))
674 			oact->sa_flags |= SA_RESETHAND;
675 		if (SIGISMEMBER(ps->ps_signodefer, sig))
676 			oact->sa_flags |= SA_NODEFER;
677 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
678 			oact->sa_flags |= SA_SIGINFO;
679 			oact->sa_sigaction =
680 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
681 		} else
682 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
683 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
684 			oact->sa_flags |= SA_NOCLDSTOP;
685 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
686 			oact->sa_flags |= SA_NOCLDWAIT;
687 	}
688 	if (act) {
689 		if ((sig == SIGKILL || sig == SIGSTOP) &&
690 		    act->sa_handler != SIG_DFL) {
691 			mtx_unlock(&ps->ps_mtx);
692 			PROC_UNLOCK(p);
693 			return (EINVAL);
694 		}
695 
696 		/*
697 		 * Change setting atomically.
698 		 */
699 
700 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
701 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
702 		if (sigact_flag_test(act, SA_SIGINFO)) {
703 			ps->ps_sigact[_SIG_IDX(sig)] =
704 			    (__sighandler_t *)act->sa_sigaction;
705 			SIGADDSET(ps->ps_siginfo, sig);
706 		} else {
707 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
708 			SIGDELSET(ps->ps_siginfo, sig);
709 		}
710 		if (!sigact_flag_test(act, SA_RESTART))
711 			SIGADDSET(ps->ps_sigintr, sig);
712 		else
713 			SIGDELSET(ps->ps_sigintr, sig);
714 		if (sigact_flag_test(act, SA_ONSTACK))
715 			SIGADDSET(ps->ps_sigonstack, sig);
716 		else
717 			SIGDELSET(ps->ps_sigonstack, sig);
718 		if (sigact_flag_test(act, SA_RESETHAND))
719 			SIGADDSET(ps->ps_sigreset, sig);
720 		else
721 			SIGDELSET(ps->ps_sigreset, sig);
722 		if (sigact_flag_test(act, SA_NODEFER))
723 			SIGADDSET(ps->ps_signodefer, sig);
724 		else
725 			SIGDELSET(ps->ps_signodefer, sig);
726 		if (sig == SIGCHLD) {
727 			if (act->sa_flags & SA_NOCLDSTOP)
728 				ps->ps_flag |= PS_NOCLDSTOP;
729 			else
730 				ps->ps_flag &= ~PS_NOCLDSTOP;
731 			if (act->sa_flags & SA_NOCLDWAIT) {
732 				/*
733 				 * Paranoia: since SA_NOCLDWAIT is implemented
734 				 * by reparenting the dying child to PID 1 (and
735 				 * trust it to reap the zombie), PID 1 itself
736 				 * is forbidden to set SA_NOCLDWAIT.
737 				 */
738 				if (p->p_pid == 1)
739 					ps->ps_flag &= ~PS_NOCLDWAIT;
740 				else
741 					ps->ps_flag |= PS_NOCLDWAIT;
742 			} else
743 				ps->ps_flag &= ~PS_NOCLDWAIT;
744 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
745 				ps->ps_flag |= PS_CLDSIGIGN;
746 			else
747 				ps->ps_flag &= ~PS_CLDSIGIGN;
748 		}
749 		/*
750 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
751 		 * and for signals set to SIG_DFL where the default is to
752 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
753 		 * have to restart the process.
754 		 */
755 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
756 		    (sigprop(sig) & SA_IGNORE &&
757 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
758 			/* never to be seen again */
759 			sigqueue_delete_proc(p, sig);
760 			if (sig != SIGCONT)
761 				/* easier in psignal */
762 				SIGADDSET(ps->ps_sigignore, sig);
763 			SIGDELSET(ps->ps_sigcatch, sig);
764 		} else {
765 			SIGDELSET(ps->ps_sigignore, sig);
766 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
767 				SIGDELSET(ps->ps_sigcatch, sig);
768 			else
769 				SIGADDSET(ps->ps_sigcatch, sig);
770 		}
771 #ifdef COMPAT_FREEBSD4
772 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
773 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
774 		    (flags & KSA_FREEBSD4) == 0)
775 			SIGDELSET(ps->ps_freebsd4, sig);
776 		else
777 			SIGADDSET(ps->ps_freebsd4, sig);
778 #endif
779 #ifdef COMPAT_43
780 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
781 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
782 		    (flags & KSA_OSIGSET) == 0)
783 			SIGDELSET(ps->ps_osigset, sig);
784 		else
785 			SIGADDSET(ps->ps_osigset, sig);
786 #endif
787 	}
788 	mtx_unlock(&ps->ps_mtx);
789 	PROC_UNLOCK(p);
790 	return (0);
791 }
792 
793 #ifndef _SYS_SYSPROTO_H_
794 struct sigaction_args {
795 	int	sig;
796 	struct	sigaction *act;
797 	struct	sigaction *oact;
798 };
799 #endif
800 int
801 sys_sigaction(td, uap)
802 	struct thread *td;
803 	register struct sigaction_args *uap;
804 {
805 	struct sigaction act, oact;
806 	register struct sigaction *actp, *oactp;
807 	int error;
808 
809 	actp = (uap->act != NULL) ? &act : NULL;
810 	oactp = (uap->oact != NULL) ? &oact : NULL;
811 	if (actp) {
812 		error = copyin(uap->act, actp, sizeof(act));
813 		if (error)
814 			return (error);
815 	}
816 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
817 	if (oactp && !error)
818 		error = copyout(oactp, uap->oact, sizeof(oact));
819 	return (error);
820 }
821 
822 #ifdef COMPAT_FREEBSD4
823 #ifndef _SYS_SYSPROTO_H_
824 struct freebsd4_sigaction_args {
825 	int	sig;
826 	struct	sigaction *act;
827 	struct	sigaction *oact;
828 };
829 #endif
830 int
831 freebsd4_sigaction(td, uap)
832 	struct thread *td;
833 	register struct freebsd4_sigaction_args *uap;
834 {
835 	struct sigaction act, oact;
836 	register struct sigaction *actp, *oactp;
837 	int error;
838 
839 
840 	actp = (uap->act != NULL) ? &act : NULL;
841 	oactp = (uap->oact != NULL) ? &oact : NULL;
842 	if (actp) {
843 		error = copyin(uap->act, actp, sizeof(act));
844 		if (error)
845 			return (error);
846 	}
847 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
848 	if (oactp && !error)
849 		error = copyout(oactp, uap->oact, sizeof(oact));
850 	return (error);
851 }
852 #endif	/* COMAPT_FREEBSD4 */
853 
854 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
855 #ifndef _SYS_SYSPROTO_H_
856 struct osigaction_args {
857 	int	signum;
858 	struct	osigaction *nsa;
859 	struct	osigaction *osa;
860 };
861 #endif
862 int
863 osigaction(td, uap)
864 	struct thread *td;
865 	register struct osigaction_args *uap;
866 {
867 	struct osigaction sa;
868 	struct sigaction nsa, osa;
869 	register struct sigaction *nsap, *osap;
870 	int error;
871 
872 	if (uap->signum <= 0 || uap->signum >= ONSIG)
873 		return (EINVAL);
874 
875 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
876 	osap = (uap->osa != NULL) ? &osa : NULL;
877 
878 	if (nsap) {
879 		error = copyin(uap->nsa, &sa, sizeof(sa));
880 		if (error)
881 			return (error);
882 		nsap->sa_handler = sa.sa_handler;
883 		nsap->sa_flags = sa.sa_flags;
884 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
885 	}
886 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
887 	if (osap && !error) {
888 		sa.sa_handler = osap->sa_handler;
889 		sa.sa_flags = osap->sa_flags;
890 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
891 		error = copyout(&sa, uap->osa, sizeof(sa));
892 	}
893 	return (error);
894 }
895 
896 #if !defined(__i386__)
897 /* Avoid replicating the same stub everywhere */
898 int
899 osigreturn(td, uap)
900 	struct thread *td;
901 	struct osigreturn_args *uap;
902 {
903 
904 	return (nosys(td, (struct nosys_args *)uap));
905 }
906 #endif
907 #endif /* COMPAT_43 */
908 
909 /*
910  * Initialize signal state for process 0;
911  * set to ignore signals that are ignored by default.
912  */
913 void
914 siginit(p)
915 	struct proc *p;
916 {
917 	register int i;
918 	struct sigacts *ps;
919 
920 	PROC_LOCK(p);
921 	ps = p->p_sigacts;
922 	mtx_lock(&ps->ps_mtx);
923 	for (i = 1; i <= NSIG; i++) {
924 		if (sigprop(i) & SA_IGNORE && i != SIGCONT) {
925 			SIGADDSET(ps->ps_sigignore, i);
926 		}
927 	}
928 	mtx_unlock(&ps->ps_mtx);
929 	PROC_UNLOCK(p);
930 }
931 
932 /*
933  * Reset specified signal to the default disposition.
934  */
935 static void
936 sigdflt(struct sigacts *ps, int sig)
937 {
938 
939 	mtx_assert(&ps->ps_mtx, MA_OWNED);
940 	SIGDELSET(ps->ps_sigcatch, sig);
941 	if ((sigprop(sig) & SA_IGNORE) != 0 && sig != SIGCONT)
942 		SIGADDSET(ps->ps_sigignore, sig);
943 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
944 	SIGDELSET(ps->ps_siginfo, sig);
945 }
946 
947 /*
948  * Reset signals for an exec of the specified process.
949  */
950 void
951 execsigs(struct proc *p)
952 {
953 	struct sigacts *ps;
954 	int sig;
955 	struct thread *td;
956 
957 	/*
958 	 * Reset caught signals.  Held signals remain held
959 	 * through td_sigmask (unless they were caught,
960 	 * and are now ignored by default).
961 	 */
962 	PROC_LOCK_ASSERT(p, MA_OWNED);
963 	td = FIRST_THREAD_IN_PROC(p);
964 	ps = p->p_sigacts;
965 	mtx_lock(&ps->ps_mtx);
966 	while (SIGNOTEMPTY(ps->ps_sigcatch)) {
967 		sig = sig_ffs(&ps->ps_sigcatch);
968 		sigdflt(ps, sig);
969 		if ((sigprop(sig) & SA_IGNORE) != 0)
970 			sigqueue_delete_proc(p, sig);
971 	}
972 	/*
973 	 * Reset stack state to the user stack.
974 	 * Clear set of signals caught on the signal stack.
975 	 */
976 	td->td_sigstk.ss_flags = SS_DISABLE;
977 	td->td_sigstk.ss_size = 0;
978 	td->td_sigstk.ss_sp = 0;
979 	td->td_pflags &= ~TDP_ALTSTACK;
980 	/*
981 	 * Reset no zombies if child dies flag as Solaris does.
982 	 */
983 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
984 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
985 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
986 	mtx_unlock(&ps->ps_mtx);
987 }
988 
989 /*
990  * kern_sigprocmask()
991  *
992  *	Manipulate signal mask.
993  */
994 int
995 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
996     int flags)
997 {
998 	sigset_t new_block, oset1;
999 	struct proc *p;
1000 	int error;
1001 
1002 	p = td->td_proc;
1003 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1004 		PROC_LOCK_ASSERT(p, MA_OWNED);
1005 	else
1006 		PROC_LOCK(p);
1007 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1008 	    ? MA_OWNED : MA_NOTOWNED);
1009 	if (oset != NULL)
1010 		*oset = td->td_sigmask;
1011 
1012 	error = 0;
1013 	if (set != NULL) {
1014 		switch (how) {
1015 		case SIG_BLOCK:
1016 			SIG_CANTMASK(*set);
1017 			oset1 = td->td_sigmask;
1018 			SIGSETOR(td->td_sigmask, *set);
1019 			new_block = td->td_sigmask;
1020 			SIGSETNAND(new_block, oset1);
1021 			break;
1022 		case SIG_UNBLOCK:
1023 			SIGSETNAND(td->td_sigmask, *set);
1024 			signotify(td);
1025 			goto out;
1026 		case SIG_SETMASK:
1027 			SIG_CANTMASK(*set);
1028 			oset1 = td->td_sigmask;
1029 			if (flags & SIGPROCMASK_OLD)
1030 				SIGSETLO(td->td_sigmask, *set);
1031 			else
1032 				td->td_sigmask = *set;
1033 			new_block = td->td_sigmask;
1034 			SIGSETNAND(new_block, oset1);
1035 			signotify(td);
1036 			break;
1037 		default:
1038 			error = EINVAL;
1039 			goto out;
1040 		}
1041 
1042 		/*
1043 		 * The new_block set contains signals that were not previously
1044 		 * blocked, but are blocked now.
1045 		 *
1046 		 * In case we block any signal that was not previously blocked
1047 		 * for td, and process has the signal pending, try to schedule
1048 		 * signal delivery to some thread that does not block the
1049 		 * signal, possibly waking it up.
1050 		 */
1051 		if (p->p_numthreads != 1)
1052 			reschedule_signals(p, new_block, flags);
1053 	}
1054 
1055 out:
1056 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1057 		PROC_UNLOCK(p);
1058 	return (error);
1059 }
1060 
1061 #ifndef _SYS_SYSPROTO_H_
1062 struct sigprocmask_args {
1063 	int	how;
1064 	const sigset_t *set;
1065 	sigset_t *oset;
1066 };
1067 #endif
1068 int
1069 sys_sigprocmask(td, uap)
1070 	register struct thread *td;
1071 	struct sigprocmask_args *uap;
1072 {
1073 	sigset_t set, oset;
1074 	sigset_t *setp, *osetp;
1075 	int error;
1076 
1077 	setp = (uap->set != NULL) ? &set : NULL;
1078 	osetp = (uap->oset != NULL) ? &oset : NULL;
1079 	if (setp) {
1080 		error = copyin(uap->set, setp, sizeof(set));
1081 		if (error)
1082 			return (error);
1083 	}
1084 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1085 	if (osetp && !error) {
1086 		error = copyout(osetp, uap->oset, sizeof(oset));
1087 	}
1088 	return (error);
1089 }
1090 
1091 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1092 #ifndef _SYS_SYSPROTO_H_
1093 struct osigprocmask_args {
1094 	int	how;
1095 	osigset_t mask;
1096 };
1097 #endif
1098 int
1099 osigprocmask(td, uap)
1100 	register struct thread *td;
1101 	struct osigprocmask_args *uap;
1102 {
1103 	sigset_t set, oset;
1104 	int error;
1105 
1106 	OSIG2SIG(uap->mask, set);
1107 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1108 	SIG2OSIG(oset, td->td_retval[0]);
1109 	return (error);
1110 }
1111 #endif /* COMPAT_43 */
1112 
1113 int
1114 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1115 {
1116 	ksiginfo_t ksi;
1117 	sigset_t set;
1118 	int error;
1119 
1120 	error = copyin(uap->set, &set, sizeof(set));
1121 	if (error) {
1122 		td->td_retval[0] = error;
1123 		return (0);
1124 	}
1125 
1126 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1127 	if (error) {
1128 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1129 			error = ERESTART;
1130 		if (error == ERESTART)
1131 			return (error);
1132 		td->td_retval[0] = error;
1133 		return (0);
1134 	}
1135 
1136 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1137 	td->td_retval[0] = error;
1138 	return (0);
1139 }
1140 
1141 int
1142 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1143 {
1144 	struct timespec ts;
1145 	struct timespec *timeout;
1146 	sigset_t set;
1147 	ksiginfo_t ksi;
1148 	int error;
1149 
1150 	if (uap->timeout) {
1151 		error = copyin(uap->timeout, &ts, sizeof(ts));
1152 		if (error)
1153 			return (error);
1154 
1155 		timeout = &ts;
1156 	} else
1157 		timeout = NULL;
1158 
1159 	error = copyin(uap->set, &set, sizeof(set));
1160 	if (error)
1161 		return (error);
1162 
1163 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1164 	if (error)
1165 		return (error);
1166 
1167 	if (uap->info)
1168 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1169 
1170 	if (error == 0)
1171 		td->td_retval[0] = ksi.ksi_signo;
1172 	return (error);
1173 }
1174 
1175 int
1176 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1177 {
1178 	ksiginfo_t ksi;
1179 	sigset_t set;
1180 	int error;
1181 
1182 	error = copyin(uap->set, &set, sizeof(set));
1183 	if (error)
1184 		return (error);
1185 
1186 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1187 	if (error)
1188 		return (error);
1189 
1190 	if (uap->info)
1191 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1192 
1193 	if (error == 0)
1194 		td->td_retval[0] = ksi.ksi_signo;
1195 	return (error);
1196 }
1197 
1198 int
1199 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1200 	struct timespec *timeout)
1201 {
1202 	struct sigacts *ps;
1203 	sigset_t saved_mask, new_block;
1204 	struct proc *p;
1205 	int error, sig, timo, timevalid = 0;
1206 	struct timespec rts, ets, ts;
1207 	struct timeval tv;
1208 
1209 	p = td->td_proc;
1210 	error = 0;
1211 	ets.tv_sec = 0;
1212 	ets.tv_nsec = 0;
1213 
1214 	if (timeout != NULL) {
1215 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1216 			timevalid = 1;
1217 			getnanouptime(&rts);
1218 			ets = rts;
1219 			timespecadd(&ets, timeout);
1220 		}
1221 	}
1222 	ksiginfo_init(ksi);
1223 	/* Some signals can not be waited for. */
1224 	SIG_CANTMASK(waitset);
1225 	ps = p->p_sigacts;
1226 	PROC_LOCK(p);
1227 	saved_mask = td->td_sigmask;
1228 	SIGSETNAND(td->td_sigmask, waitset);
1229 	for (;;) {
1230 		mtx_lock(&ps->ps_mtx);
1231 		sig = cursig(td);
1232 		mtx_unlock(&ps->ps_mtx);
1233 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1234 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1235 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1236 				error = 0;
1237 				break;
1238 			}
1239 		}
1240 
1241 		if (error != 0)
1242 			break;
1243 
1244 		/*
1245 		 * POSIX says this must be checked after looking for pending
1246 		 * signals.
1247 		 */
1248 		if (timeout != NULL) {
1249 			if (!timevalid) {
1250 				error = EINVAL;
1251 				break;
1252 			}
1253 			getnanouptime(&rts);
1254 			if (timespeccmp(&rts, &ets, >=)) {
1255 				error = EAGAIN;
1256 				break;
1257 			}
1258 			ts = ets;
1259 			timespecsub(&ts, &rts);
1260 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1261 			timo = tvtohz(&tv);
1262 		} else {
1263 			timo = 0;
1264 		}
1265 
1266 		error = msleep(ps, &p->p_mtx, PPAUSE|PCATCH, "sigwait", timo);
1267 
1268 		if (timeout != NULL) {
1269 			if (error == ERESTART) {
1270 				/* Timeout can not be restarted. */
1271 				error = EINTR;
1272 			} else if (error == EAGAIN) {
1273 				/* We will calculate timeout by ourself. */
1274 				error = 0;
1275 			}
1276 		}
1277 	}
1278 
1279 	new_block = saved_mask;
1280 	SIGSETNAND(new_block, td->td_sigmask);
1281 	td->td_sigmask = saved_mask;
1282 	/*
1283 	 * Fewer signals can be delivered to us, reschedule signal
1284 	 * notification.
1285 	 */
1286 	if (p->p_numthreads != 1)
1287 		reschedule_signals(p, new_block, 0);
1288 
1289 	if (error == 0) {
1290 		SDT_PROBE(proc, kernel, , signal__clear, sig, ksi, 0, 0, 0);
1291 
1292 		if (ksi->ksi_code == SI_TIMER)
1293 			itimer_accept(p, ksi->ksi_timerid, ksi);
1294 
1295 #ifdef KTRACE
1296 		if (KTRPOINT(td, KTR_PSIG)) {
1297 			sig_t action;
1298 
1299 			mtx_lock(&ps->ps_mtx);
1300 			action = ps->ps_sigact[_SIG_IDX(sig)];
1301 			mtx_unlock(&ps->ps_mtx);
1302 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1303 		}
1304 #endif
1305 		if (sig == SIGKILL)
1306 			sigexit(td, sig);
1307 	}
1308 	PROC_UNLOCK(p);
1309 	return (error);
1310 }
1311 
1312 #ifndef _SYS_SYSPROTO_H_
1313 struct sigpending_args {
1314 	sigset_t	*set;
1315 };
1316 #endif
1317 int
1318 sys_sigpending(td, uap)
1319 	struct thread *td;
1320 	struct sigpending_args *uap;
1321 {
1322 	struct proc *p = td->td_proc;
1323 	sigset_t pending;
1324 
1325 	PROC_LOCK(p);
1326 	pending = p->p_sigqueue.sq_signals;
1327 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1328 	PROC_UNLOCK(p);
1329 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1330 }
1331 
1332 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1333 #ifndef _SYS_SYSPROTO_H_
1334 struct osigpending_args {
1335 	int	dummy;
1336 };
1337 #endif
1338 int
1339 osigpending(td, uap)
1340 	struct thread *td;
1341 	struct osigpending_args *uap;
1342 {
1343 	struct proc *p = td->td_proc;
1344 	sigset_t pending;
1345 
1346 	PROC_LOCK(p);
1347 	pending = p->p_sigqueue.sq_signals;
1348 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1349 	PROC_UNLOCK(p);
1350 	SIG2OSIG(pending, td->td_retval[0]);
1351 	return (0);
1352 }
1353 #endif /* COMPAT_43 */
1354 
1355 #if defined(COMPAT_43)
1356 /*
1357  * Generalized interface signal handler, 4.3-compatible.
1358  */
1359 #ifndef _SYS_SYSPROTO_H_
1360 struct osigvec_args {
1361 	int	signum;
1362 	struct	sigvec *nsv;
1363 	struct	sigvec *osv;
1364 };
1365 #endif
1366 /* ARGSUSED */
1367 int
1368 osigvec(td, uap)
1369 	struct thread *td;
1370 	register struct osigvec_args *uap;
1371 {
1372 	struct sigvec vec;
1373 	struct sigaction nsa, osa;
1374 	register struct sigaction *nsap, *osap;
1375 	int error;
1376 
1377 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1378 		return (EINVAL);
1379 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1380 	osap = (uap->osv != NULL) ? &osa : NULL;
1381 	if (nsap) {
1382 		error = copyin(uap->nsv, &vec, sizeof(vec));
1383 		if (error)
1384 			return (error);
1385 		nsap->sa_handler = vec.sv_handler;
1386 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1387 		nsap->sa_flags = vec.sv_flags;
1388 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1389 	}
1390 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1391 	if (osap && !error) {
1392 		vec.sv_handler = osap->sa_handler;
1393 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1394 		vec.sv_flags = osap->sa_flags;
1395 		vec.sv_flags &= ~SA_NOCLDWAIT;
1396 		vec.sv_flags ^= SA_RESTART;
1397 		error = copyout(&vec, uap->osv, sizeof(vec));
1398 	}
1399 	return (error);
1400 }
1401 
1402 #ifndef _SYS_SYSPROTO_H_
1403 struct osigblock_args {
1404 	int	mask;
1405 };
1406 #endif
1407 int
1408 osigblock(td, uap)
1409 	register struct thread *td;
1410 	struct osigblock_args *uap;
1411 {
1412 	sigset_t set, oset;
1413 
1414 	OSIG2SIG(uap->mask, set);
1415 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1416 	SIG2OSIG(oset, td->td_retval[0]);
1417 	return (0);
1418 }
1419 
1420 #ifndef _SYS_SYSPROTO_H_
1421 struct osigsetmask_args {
1422 	int	mask;
1423 };
1424 #endif
1425 int
1426 osigsetmask(td, uap)
1427 	struct thread *td;
1428 	struct osigsetmask_args *uap;
1429 {
1430 	sigset_t set, oset;
1431 
1432 	OSIG2SIG(uap->mask, set);
1433 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1434 	SIG2OSIG(oset, td->td_retval[0]);
1435 	return (0);
1436 }
1437 #endif /* COMPAT_43 */
1438 
1439 /*
1440  * Suspend calling thread until signal, providing mask to be set in the
1441  * meantime.
1442  */
1443 #ifndef _SYS_SYSPROTO_H_
1444 struct sigsuspend_args {
1445 	const sigset_t *sigmask;
1446 };
1447 #endif
1448 /* ARGSUSED */
1449 int
1450 sys_sigsuspend(td, uap)
1451 	struct thread *td;
1452 	struct sigsuspend_args *uap;
1453 {
1454 	sigset_t mask;
1455 	int error;
1456 
1457 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1458 	if (error)
1459 		return (error);
1460 	return (kern_sigsuspend(td, mask));
1461 }
1462 
1463 int
1464 kern_sigsuspend(struct thread *td, sigset_t mask)
1465 {
1466 	struct proc *p = td->td_proc;
1467 	int has_sig, sig;
1468 
1469 	/*
1470 	 * When returning from sigsuspend, we want
1471 	 * the old mask to be restored after the
1472 	 * signal handler has finished.  Thus, we
1473 	 * save it here and mark the sigacts structure
1474 	 * to indicate this.
1475 	 */
1476 	PROC_LOCK(p);
1477 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1478 	    SIGPROCMASK_PROC_LOCKED);
1479 	td->td_pflags |= TDP_OLDMASK;
1480 
1481 	/*
1482 	 * Process signals now. Otherwise, we can get spurious wakeup
1483 	 * due to signal entered process queue, but delivered to other
1484 	 * thread. But sigsuspend should return only on signal
1485 	 * delivery.
1486 	 */
1487 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1488 	for (has_sig = 0; !has_sig;) {
1489 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1490 			0) == 0)
1491 			/* void */;
1492 		thread_suspend_check(0);
1493 		mtx_lock(&p->p_sigacts->ps_mtx);
1494 		while ((sig = cursig(td)) != 0)
1495 			has_sig += postsig(sig);
1496 		mtx_unlock(&p->p_sigacts->ps_mtx);
1497 	}
1498 	PROC_UNLOCK(p);
1499 	td->td_errno = EINTR;
1500 	td->td_pflags |= TDP_NERRNO;
1501 	return (EJUSTRETURN);
1502 }
1503 
1504 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1505 /*
1506  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1507  * convention: libc stub passes mask, not pointer, to save a copyin.
1508  */
1509 #ifndef _SYS_SYSPROTO_H_
1510 struct osigsuspend_args {
1511 	osigset_t mask;
1512 };
1513 #endif
1514 /* ARGSUSED */
1515 int
1516 osigsuspend(td, uap)
1517 	struct thread *td;
1518 	struct osigsuspend_args *uap;
1519 {
1520 	sigset_t mask;
1521 
1522 	OSIG2SIG(uap->mask, mask);
1523 	return (kern_sigsuspend(td, mask));
1524 }
1525 #endif /* COMPAT_43 */
1526 
1527 #if defined(COMPAT_43)
1528 #ifndef _SYS_SYSPROTO_H_
1529 struct osigstack_args {
1530 	struct	sigstack *nss;
1531 	struct	sigstack *oss;
1532 };
1533 #endif
1534 /* ARGSUSED */
1535 int
1536 osigstack(td, uap)
1537 	struct thread *td;
1538 	register struct osigstack_args *uap;
1539 {
1540 	struct sigstack nss, oss;
1541 	int error = 0;
1542 
1543 	if (uap->nss != NULL) {
1544 		error = copyin(uap->nss, &nss, sizeof(nss));
1545 		if (error)
1546 			return (error);
1547 	}
1548 	oss.ss_sp = td->td_sigstk.ss_sp;
1549 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1550 	if (uap->nss != NULL) {
1551 		td->td_sigstk.ss_sp = nss.ss_sp;
1552 		td->td_sigstk.ss_size = 0;
1553 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1554 		td->td_pflags |= TDP_ALTSTACK;
1555 	}
1556 	if (uap->oss != NULL)
1557 		error = copyout(&oss, uap->oss, sizeof(oss));
1558 
1559 	return (error);
1560 }
1561 #endif /* COMPAT_43 */
1562 
1563 #ifndef _SYS_SYSPROTO_H_
1564 struct sigaltstack_args {
1565 	stack_t	*ss;
1566 	stack_t	*oss;
1567 };
1568 #endif
1569 /* ARGSUSED */
1570 int
1571 sys_sigaltstack(td, uap)
1572 	struct thread *td;
1573 	register struct sigaltstack_args *uap;
1574 {
1575 	stack_t ss, oss;
1576 	int error;
1577 
1578 	if (uap->ss != NULL) {
1579 		error = copyin(uap->ss, &ss, sizeof(ss));
1580 		if (error)
1581 			return (error);
1582 	}
1583 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1584 	    (uap->oss != NULL) ? &oss : NULL);
1585 	if (error)
1586 		return (error);
1587 	if (uap->oss != NULL)
1588 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1589 	return (error);
1590 }
1591 
1592 int
1593 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1594 {
1595 	struct proc *p = td->td_proc;
1596 	int oonstack;
1597 
1598 	oonstack = sigonstack(cpu_getstack(td));
1599 
1600 	if (oss != NULL) {
1601 		*oss = td->td_sigstk;
1602 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1603 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1604 	}
1605 
1606 	if (ss != NULL) {
1607 		if (oonstack)
1608 			return (EPERM);
1609 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1610 			return (EINVAL);
1611 		if (!(ss->ss_flags & SS_DISABLE)) {
1612 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1613 				return (ENOMEM);
1614 
1615 			td->td_sigstk = *ss;
1616 			td->td_pflags |= TDP_ALTSTACK;
1617 		} else {
1618 			td->td_pflags &= ~TDP_ALTSTACK;
1619 		}
1620 	}
1621 	return (0);
1622 }
1623 
1624 /*
1625  * Common code for kill process group/broadcast kill.
1626  * cp is calling process.
1627  */
1628 static int
1629 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1630 {
1631 	struct proc *p;
1632 	struct pgrp *pgrp;
1633 	int err;
1634 	int ret;
1635 
1636 	ret = ESRCH;
1637 	if (all) {
1638 		/*
1639 		 * broadcast
1640 		 */
1641 		sx_slock(&allproc_lock);
1642 		FOREACH_PROC_IN_SYSTEM(p) {
1643 			PROC_LOCK(p);
1644 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1645 			    p == td->td_proc || 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 		sx_sunlock(&allproc_lock);
1660 	} else {
1661 		sx_slock(&proctree_lock);
1662 		if (pgid == 0) {
1663 			/*
1664 			 * zero pgid means send to my process group.
1665 			 */
1666 			pgrp = td->td_proc->p_pgrp;
1667 			PGRP_LOCK(pgrp);
1668 		} else {
1669 			pgrp = pgfind(pgid);
1670 			if (pgrp == NULL) {
1671 				sx_sunlock(&proctree_lock);
1672 				return (ESRCH);
1673 			}
1674 		}
1675 		sx_sunlock(&proctree_lock);
1676 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1677 			PROC_LOCK(p);
1678 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1679 			    p->p_state == PRS_NEW) {
1680 				PROC_UNLOCK(p);
1681 				continue;
1682 			}
1683 			err = p_cansignal(td, p, sig);
1684 			if (err == 0) {
1685 				if (sig)
1686 					pksignal(p, sig, ksi);
1687 				ret = err;
1688 			}
1689 			else if (ret == ESRCH)
1690 				ret = err;
1691 			PROC_UNLOCK(p);
1692 		}
1693 		PGRP_UNLOCK(pgrp);
1694 	}
1695 	return (ret);
1696 }
1697 
1698 #ifndef _SYS_SYSPROTO_H_
1699 struct kill_args {
1700 	int	pid;
1701 	int	signum;
1702 };
1703 #endif
1704 /* ARGSUSED */
1705 int
1706 sys_kill(struct thread *td, struct kill_args *uap)
1707 {
1708 	ksiginfo_t ksi;
1709 	struct proc *p;
1710 	int error;
1711 
1712 	/*
1713 	 * A process in capability mode can send signals only to himself.
1714 	 * The main rationale behind this is that abort(3) is implemented as
1715 	 * kill(getpid(), SIGABRT).
1716 	 */
1717 	if (IN_CAPABILITY_MODE(td) && uap->pid != td->td_proc->p_pid)
1718 		return (ECAPMODE);
1719 
1720 	AUDIT_ARG_SIGNUM(uap->signum);
1721 	AUDIT_ARG_PID(uap->pid);
1722 	if ((u_int)uap->signum > _SIG_MAXSIG)
1723 		return (EINVAL);
1724 
1725 	ksiginfo_init(&ksi);
1726 	ksi.ksi_signo = uap->signum;
1727 	ksi.ksi_code = SI_USER;
1728 	ksi.ksi_pid = td->td_proc->p_pid;
1729 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1730 
1731 	if (uap->pid > 0) {
1732 		/* kill single process */
1733 		if ((p = pfind(uap->pid)) == NULL) {
1734 			if ((p = zpfind(uap->pid)) == NULL)
1735 				return (ESRCH);
1736 		}
1737 		AUDIT_ARG_PROCESS(p);
1738 		error = p_cansignal(td, p, uap->signum);
1739 		if (error == 0 && uap->signum)
1740 			pksignal(p, uap->signum, &ksi);
1741 		PROC_UNLOCK(p);
1742 		return (error);
1743 	}
1744 	switch (uap->pid) {
1745 	case -1:		/* broadcast signal */
1746 		return (killpg1(td, uap->signum, 0, 1, &ksi));
1747 	case 0:			/* signal own process group */
1748 		return (killpg1(td, uap->signum, 0, 0, &ksi));
1749 	default:		/* negative explicit process group */
1750 		return (killpg1(td, uap->signum, -uap->pid, 0, &ksi));
1751 	}
1752 	/* NOTREACHED */
1753 }
1754 
1755 int
1756 sys_pdkill(td, uap)
1757 	struct thread *td;
1758 	struct pdkill_args *uap;
1759 {
1760 	struct proc *p;
1761 	cap_rights_t rights;
1762 	int error;
1763 
1764 	AUDIT_ARG_SIGNUM(uap->signum);
1765 	AUDIT_ARG_FD(uap->fd);
1766 	if ((u_int)uap->signum > _SIG_MAXSIG)
1767 		return (EINVAL);
1768 
1769 	error = procdesc_find(td, uap->fd,
1770 	    cap_rights_init(&rights, CAP_PDKILL), &p);
1771 	if (error)
1772 		return (error);
1773 	AUDIT_ARG_PROCESS(p);
1774 	error = p_cansignal(td, p, uap->signum);
1775 	if (error == 0 && uap->signum)
1776 		kern_psignal(p, uap->signum);
1777 	PROC_UNLOCK(p);
1778 	return (error);
1779 }
1780 
1781 #if defined(COMPAT_43)
1782 #ifndef _SYS_SYSPROTO_H_
1783 struct okillpg_args {
1784 	int	pgid;
1785 	int	signum;
1786 };
1787 #endif
1788 /* ARGSUSED */
1789 int
1790 okillpg(struct thread *td, struct okillpg_args *uap)
1791 {
1792 	ksiginfo_t ksi;
1793 
1794 	AUDIT_ARG_SIGNUM(uap->signum);
1795 	AUDIT_ARG_PID(uap->pgid);
1796 	if ((u_int)uap->signum > _SIG_MAXSIG)
1797 		return (EINVAL);
1798 
1799 	ksiginfo_init(&ksi);
1800 	ksi.ksi_signo = uap->signum;
1801 	ksi.ksi_code = SI_USER;
1802 	ksi.ksi_pid = td->td_proc->p_pid;
1803 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1804 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1805 }
1806 #endif /* COMPAT_43 */
1807 
1808 #ifndef _SYS_SYSPROTO_H_
1809 struct sigqueue_args {
1810 	pid_t pid;
1811 	int signum;
1812 	/* union sigval */ void *value;
1813 };
1814 #endif
1815 int
1816 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1817 {
1818 	ksiginfo_t ksi;
1819 	struct proc *p;
1820 	int error;
1821 
1822 	if ((u_int)uap->signum > _SIG_MAXSIG)
1823 		return (EINVAL);
1824 
1825 	/*
1826 	 * Specification says sigqueue can only send signal to
1827 	 * single process.
1828 	 */
1829 	if (uap->pid <= 0)
1830 		return (EINVAL);
1831 
1832 	if ((p = pfind(uap->pid)) == NULL) {
1833 		if ((p = zpfind(uap->pid)) == NULL)
1834 			return (ESRCH);
1835 	}
1836 	error = p_cansignal(td, p, uap->signum);
1837 	if (error == 0 && uap->signum != 0) {
1838 		ksiginfo_init(&ksi);
1839 		ksi.ksi_flags = KSI_SIGQ;
1840 		ksi.ksi_signo = uap->signum;
1841 		ksi.ksi_code = SI_QUEUE;
1842 		ksi.ksi_pid = td->td_proc->p_pid;
1843 		ksi.ksi_uid = td->td_ucred->cr_ruid;
1844 		ksi.ksi_value.sival_ptr = uap->value;
1845 		error = pksignal(p, ksi.ksi_signo, &ksi);
1846 	}
1847 	PROC_UNLOCK(p);
1848 	return (error);
1849 }
1850 
1851 /*
1852  * Send a signal to a process group.
1853  */
1854 void
1855 gsignal(int pgid, int sig, ksiginfo_t *ksi)
1856 {
1857 	struct pgrp *pgrp;
1858 
1859 	if (pgid != 0) {
1860 		sx_slock(&proctree_lock);
1861 		pgrp = pgfind(pgid);
1862 		sx_sunlock(&proctree_lock);
1863 		if (pgrp != NULL) {
1864 			pgsignal(pgrp, sig, 0, ksi);
1865 			PGRP_UNLOCK(pgrp);
1866 		}
1867 	}
1868 }
1869 
1870 /*
1871  * Send a signal to a process group.  If checktty is 1,
1872  * limit to members which have a controlling terminal.
1873  */
1874 void
1875 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
1876 {
1877 	struct proc *p;
1878 
1879 	if (pgrp) {
1880 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
1881 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1882 			PROC_LOCK(p);
1883 			if (p->p_state == PRS_NORMAL &&
1884 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
1885 				pksignal(p, sig, ksi);
1886 			PROC_UNLOCK(p);
1887 		}
1888 	}
1889 }
1890 
1891 
1892 /*
1893  * Recalculate the signal mask and reset the signal disposition after
1894  * usermode frame for delivery is formed.  Should be called after
1895  * mach-specific routine, because sysent->sv_sendsig() needs correct
1896  * ps_siginfo and signal mask.
1897  */
1898 static void
1899 postsig_done(int sig, struct thread *td, struct sigacts *ps)
1900 {
1901 	sigset_t mask;
1902 
1903 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1904 	td->td_ru.ru_nsignals++;
1905 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
1906 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
1907 		SIGADDSET(mask, sig);
1908 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
1909 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
1910 	if (SIGISMEMBER(ps->ps_sigreset, sig))
1911 		sigdflt(ps, sig);
1912 }
1913 
1914 
1915 /*
1916  * Send a signal caused by a trap to the current thread.  If it will be
1917  * caught immediately, deliver it with correct code.  Otherwise, post it
1918  * normally.
1919  */
1920 void
1921 trapsignal(struct thread *td, ksiginfo_t *ksi)
1922 {
1923 	struct sigacts *ps;
1924 	struct proc *p;
1925 	int sig;
1926 	int code;
1927 
1928 	p = td->td_proc;
1929 	sig = ksi->ksi_signo;
1930 	code = ksi->ksi_code;
1931 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
1932 
1933 	PROC_LOCK(p);
1934 	ps = p->p_sigacts;
1935 	mtx_lock(&ps->ps_mtx);
1936 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
1937 	    !SIGISMEMBER(td->td_sigmask, sig)) {
1938 #ifdef KTRACE
1939 		if (KTRPOINT(curthread, KTR_PSIG))
1940 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
1941 			    &td->td_sigmask, code);
1942 #endif
1943 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
1944 				ksi, &td->td_sigmask);
1945 		postsig_done(sig, td, ps);
1946 		mtx_unlock(&ps->ps_mtx);
1947 	} else {
1948 		/*
1949 		 * Avoid a possible infinite loop if the thread
1950 		 * masking the signal or process is ignoring the
1951 		 * signal.
1952 		 */
1953 		if (kern_forcesigexit &&
1954 		    (SIGISMEMBER(td->td_sigmask, sig) ||
1955 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
1956 			SIGDELSET(td->td_sigmask, sig);
1957 			SIGDELSET(ps->ps_sigcatch, sig);
1958 			SIGDELSET(ps->ps_sigignore, sig);
1959 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1960 		}
1961 		mtx_unlock(&ps->ps_mtx);
1962 		p->p_code = code;	/* XXX for core dump/debugger */
1963 		p->p_sig = sig;		/* XXX to verify code */
1964 		tdsendsignal(p, td, sig, ksi);
1965 	}
1966 	PROC_UNLOCK(p);
1967 }
1968 
1969 static struct thread *
1970 sigtd(struct proc *p, int sig, int prop)
1971 {
1972 	struct thread *td, *signal_td;
1973 
1974 	PROC_LOCK_ASSERT(p, MA_OWNED);
1975 
1976 	/*
1977 	 * Check if current thread can handle the signal without
1978 	 * switching context to another thread.
1979 	 */
1980 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig))
1981 		return (curthread);
1982 	signal_td = NULL;
1983 	FOREACH_THREAD_IN_PROC(p, td) {
1984 		if (!SIGISMEMBER(td->td_sigmask, sig)) {
1985 			signal_td = td;
1986 			break;
1987 		}
1988 	}
1989 	if (signal_td == NULL)
1990 		signal_td = FIRST_THREAD_IN_PROC(p);
1991 	return (signal_td);
1992 }
1993 
1994 /*
1995  * Send the signal to the process.  If the signal has an action, the action
1996  * is usually performed by the target process rather than the caller; we add
1997  * the signal to the set of pending signals for the process.
1998  *
1999  * Exceptions:
2000  *   o When a stop signal is sent to a sleeping process that takes the
2001  *     default action, the process is stopped without awakening it.
2002  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2003  *     regardless of the signal action (eg, blocked or ignored).
2004  *
2005  * Other ignored signals are discarded immediately.
2006  *
2007  * NB: This function may be entered from the debugger via the "kill" DDB
2008  * command.  There is little that can be done to mitigate the possibly messy
2009  * side effects of this unwise possibility.
2010  */
2011 void
2012 kern_psignal(struct proc *p, int sig)
2013 {
2014 	ksiginfo_t ksi;
2015 
2016 	ksiginfo_init(&ksi);
2017 	ksi.ksi_signo = sig;
2018 	ksi.ksi_code = SI_KERNEL;
2019 	(void) tdsendsignal(p, NULL, sig, &ksi);
2020 }
2021 
2022 int
2023 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2024 {
2025 
2026 	return (tdsendsignal(p, NULL, sig, ksi));
2027 }
2028 
2029 /* Utility function for finding a thread to send signal event to. */
2030 int
2031 sigev_findtd(struct proc *p ,struct sigevent *sigev, struct thread **ttd)
2032 {
2033 	struct thread *td;
2034 
2035 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2036 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2037 		if (td == NULL)
2038 			return (ESRCH);
2039 		*ttd = td;
2040 	} else {
2041 		*ttd = NULL;
2042 		PROC_LOCK(p);
2043 	}
2044 	return (0);
2045 }
2046 
2047 void
2048 tdsignal(struct thread *td, int sig)
2049 {
2050 	ksiginfo_t ksi;
2051 
2052 	ksiginfo_init(&ksi);
2053 	ksi.ksi_signo = sig;
2054 	ksi.ksi_code = SI_KERNEL;
2055 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2056 }
2057 
2058 void
2059 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2060 {
2061 
2062 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2063 }
2064 
2065 int
2066 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2067 {
2068 	sig_t action;
2069 	sigqueue_t *sigqueue;
2070 	int prop;
2071 	struct sigacts *ps;
2072 	int intrval;
2073 	int ret = 0;
2074 	int wakeup_swapper;
2075 
2076 	MPASS(td == NULL || p == td->td_proc);
2077 	PROC_LOCK_ASSERT(p, MA_OWNED);
2078 
2079 	if (!_SIG_VALID(sig))
2080 		panic("%s(): invalid signal %d", __func__, sig);
2081 
2082 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2083 
2084 	/*
2085 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2086 	 */
2087 	if (p->p_state == PRS_ZOMBIE) {
2088 		if (ksi && (ksi->ksi_flags & KSI_INS))
2089 			ksiginfo_tryfree(ksi);
2090 		return (ret);
2091 	}
2092 
2093 	ps = p->p_sigacts;
2094 	KNOTE_LOCKED(&p->p_klist, NOTE_SIGNAL | sig);
2095 	prop = sigprop(sig);
2096 
2097 	if (td == NULL) {
2098 		td = sigtd(p, sig, prop);
2099 		sigqueue = &p->p_sigqueue;
2100 	} else
2101 		sigqueue = &td->td_sigqueue;
2102 
2103 	SDT_PROBE(proc, kernel, , signal__send, td, p, sig, 0, 0 );
2104 
2105 	/*
2106 	 * If the signal is being ignored,
2107 	 * then we forget about it immediately.
2108 	 * (Note: we don't set SIGCONT in ps_sigignore,
2109 	 * and if it is set to SIG_IGN,
2110 	 * action will be SIG_DFL here.)
2111 	 */
2112 	mtx_lock(&ps->ps_mtx);
2113 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2114 		SDT_PROBE(proc, kernel, , signal__discard, td, p, sig, 0, 0 );
2115 
2116 		mtx_unlock(&ps->ps_mtx);
2117 		if (ksi && (ksi->ksi_flags & KSI_INS))
2118 			ksiginfo_tryfree(ksi);
2119 		return (ret);
2120 	}
2121 	if (SIGISMEMBER(td->td_sigmask, sig))
2122 		action = SIG_HOLD;
2123 	else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2124 		action = SIG_CATCH;
2125 	else
2126 		action = SIG_DFL;
2127 	if (SIGISMEMBER(ps->ps_sigintr, sig))
2128 		intrval = EINTR;
2129 	else
2130 		intrval = ERESTART;
2131 	mtx_unlock(&ps->ps_mtx);
2132 
2133 	if (prop & SA_CONT)
2134 		sigqueue_delete_stopmask_proc(p);
2135 	else if (prop & SA_STOP) {
2136 		/*
2137 		 * If sending a tty stop signal to a member of an orphaned
2138 		 * process group, discard the signal here if the action
2139 		 * is default; don't stop the process below if sleeping,
2140 		 * and don't clear any pending SIGCONT.
2141 		 */
2142 		if ((prop & SA_TTYSTOP) &&
2143 		    (p->p_pgrp->pg_jobc == 0) &&
2144 		    (action == SIG_DFL)) {
2145 			if (ksi && (ksi->ksi_flags & KSI_INS))
2146 				ksiginfo_tryfree(ksi);
2147 			return (ret);
2148 		}
2149 		sigqueue_delete_proc(p, SIGCONT);
2150 		if (p->p_flag & P_CONTINUED) {
2151 			p->p_flag &= ~P_CONTINUED;
2152 			PROC_LOCK(p->p_pptr);
2153 			sigqueue_take(p->p_ksi);
2154 			PROC_UNLOCK(p->p_pptr);
2155 		}
2156 	}
2157 
2158 	ret = sigqueue_add(sigqueue, sig, ksi);
2159 	if (ret != 0)
2160 		return (ret);
2161 	signotify(td);
2162 	/*
2163 	 * Defer further processing for signals which are held,
2164 	 * except that stopped processes must be continued by SIGCONT.
2165 	 */
2166 	if (action == SIG_HOLD &&
2167 	    !((prop & SA_CONT) && (p->p_flag & P_STOPPED_SIG)))
2168 		return (ret);
2169 	/*
2170 	 * SIGKILL: Remove procfs STOPEVENTs.
2171 	 */
2172 	if (sig == SIGKILL) {
2173 		/* from procfs_ioctl.c: PIOCBIC */
2174 		p->p_stops = 0;
2175 		/* from procfs_ioctl.c: PIOCCONT */
2176 		p->p_step = 0;
2177 		wakeup(&p->p_step);
2178 	}
2179 	/*
2180 	 * Some signals have a process-wide effect and a per-thread
2181 	 * component.  Most processing occurs when the process next
2182 	 * tries to cross the user boundary, however there are some
2183 	 * times when processing needs to be done immediately, such as
2184 	 * waking up threads so that they can cross the user boundary.
2185 	 * We try to do the per-process part here.
2186 	 */
2187 	if (P_SHOULDSTOP(p)) {
2188 		KASSERT(!(p->p_flag & P_WEXIT),
2189 		    ("signal to stopped but exiting process"));
2190 		if (sig == SIGKILL) {
2191 			/*
2192 			 * If traced process is already stopped,
2193 			 * then no further action is necessary.
2194 			 */
2195 			if (p->p_flag & P_TRACED)
2196 				goto out;
2197 			/*
2198 			 * SIGKILL sets process running.
2199 			 * It will die elsewhere.
2200 			 * All threads must be restarted.
2201 			 */
2202 			p->p_flag &= ~P_STOPPED_SIG;
2203 			goto runfast;
2204 		}
2205 
2206 		if (prop & SA_CONT) {
2207 			/*
2208 			 * If traced process is already stopped,
2209 			 * then no further action is necessary.
2210 			 */
2211 			if (p->p_flag & P_TRACED)
2212 				goto out;
2213 			/*
2214 			 * If SIGCONT is default (or ignored), we continue the
2215 			 * process but don't leave the signal in sigqueue as
2216 			 * it has no further action.  If SIGCONT is held, we
2217 			 * continue the process and leave the signal in
2218 			 * sigqueue.  If the process catches SIGCONT, let it
2219 			 * handle the signal itself.  If it isn't waiting on
2220 			 * an event, it goes back to run state.
2221 			 * Otherwise, process goes back to sleep state.
2222 			 */
2223 			p->p_flag &= ~P_STOPPED_SIG;
2224 			PROC_SLOCK(p);
2225 			if (p->p_numthreads == p->p_suspcount) {
2226 				PROC_SUNLOCK(p);
2227 				p->p_flag |= P_CONTINUED;
2228 				p->p_xstat = SIGCONT;
2229 				PROC_LOCK(p->p_pptr);
2230 				childproc_continued(p);
2231 				PROC_UNLOCK(p->p_pptr);
2232 				PROC_SLOCK(p);
2233 			}
2234 			if (action == SIG_DFL) {
2235 				thread_unsuspend(p);
2236 				PROC_SUNLOCK(p);
2237 				sigqueue_delete(sigqueue, sig);
2238 				goto out;
2239 			}
2240 			if (action == SIG_CATCH) {
2241 				/*
2242 				 * The process wants to catch it so it needs
2243 				 * to run at least one thread, but which one?
2244 				 */
2245 				PROC_SUNLOCK(p);
2246 				goto runfast;
2247 			}
2248 			/*
2249 			 * The signal is not ignored or caught.
2250 			 */
2251 			thread_unsuspend(p);
2252 			PROC_SUNLOCK(p);
2253 			goto out;
2254 		}
2255 
2256 		if (prop & SA_STOP) {
2257 			/*
2258 			 * If traced process is already stopped,
2259 			 * then no further action is necessary.
2260 			 */
2261 			if (p->p_flag & P_TRACED)
2262 				goto out;
2263 			/*
2264 			 * Already stopped, don't need to stop again
2265 			 * (If we did the shell could get confused).
2266 			 * Just make sure the signal STOP bit set.
2267 			 */
2268 			p->p_flag |= P_STOPPED_SIG;
2269 			sigqueue_delete(sigqueue, sig);
2270 			goto out;
2271 		}
2272 
2273 		/*
2274 		 * All other kinds of signals:
2275 		 * If a thread is sleeping interruptibly, simulate a
2276 		 * wakeup so that when it is continued it will be made
2277 		 * runnable and can look at the signal.  However, don't make
2278 		 * the PROCESS runnable, leave it stopped.
2279 		 * It may run a bit until it hits a thread_suspend_check().
2280 		 */
2281 		wakeup_swapper = 0;
2282 		PROC_SLOCK(p);
2283 		thread_lock(td);
2284 		if (TD_ON_SLEEPQ(td) && (td->td_flags & TDF_SINTR))
2285 			wakeup_swapper = sleepq_abort(td, intrval);
2286 		thread_unlock(td);
2287 		PROC_SUNLOCK(p);
2288 		if (wakeup_swapper)
2289 			kick_proc0();
2290 		goto out;
2291 		/*
2292 		 * Mutexes are short lived. Threads waiting on them will
2293 		 * hit thread_suspend_check() soon.
2294 		 */
2295 	} else if (p->p_state == PRS_NORMAL) {
2296 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2297 			tdsigwakeup(td, sig, action, intrval);
2298 			goto out;
2299 		}
2300 
2301 		MPASS(action == SIG_DFL);
2302 
2303 		if (prop & SA_STOP) {
2304 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2305 				goto out;
2306 			p->p_flag |= P_STOPPED_SIG;
2307 			p->p_xstat = sig;
2308 			PROC_SLOCK(p);
2309 			sig_suspend_threads(td, p, 1);
2310 			if (p->p_numthreads == p->p_suspcount) {
2311 				/*
2312 				 * only thread sending signal to another
2313 				 * process can reach here, if thread is sending
2314 				 * signal to its process, because thread does
2315 				 * not suspend itself here, p_numthreads
2316 				 * should never be equal to p_suspcount.
2317 				 */
2318 				thread_stopped(p);
2319 				PROC_SUNLOCK(p);
2320 				sigqueue_delete_proc(p, p->p_xstat);
2321 			} else
2322 				PROC_SUNLOCK(p);
2323 			goto out;
2324 		}
2325 	} else {
2326 		/* Not in "NORMAL" state. discard the signal. */
2327 		sigqueue_delete(sigqueue, sig);
2328 		goto out;
2329 	}
2330 
2331 	/*
2332 	 * The process is not stopped so we need to apply the signal to all the
2333 	 * running threads.
2334 	 */
2335 runfast:
2336 	tdsigwakeup(td, sig, action, intrval);
2337 	PROC_SLOCK(p);
2338 	thread_unsuspend(p);
2339 	PROC_SUNLOCK(p);
2340 out:
2341 	/* If we jump here, proc slock should not be owned. */
2342 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2343 	return (ret);
2344 }
2345 
2346 /*
2347  * The force of a signal has been directed against a single
2348  * thread.  We need to see what we can do about knocking it
2349  * out of any sleep it may be in etc.
2350  */
2351 static void
2352 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2353 {
2354 	struct proc *p = td->td_proc;
2355 	register int prop;
2356 	int wakeup_swapper;
2357 
2358 	wakeup_swapper = 0;
2359 	PROC_LOCK_ASSERT(p, MA_OWNED);
2360 	prop = sigprop(sig);
2361 
2362 	PROC_SLOCK(p);
2363 	thread_lock(td);
2364 	/*
2365 	 * Bring the priority of a thread up if we want it to get
2366 	 * killed in this lifetime.
2367 	 */
2368 	if (action == SIG_DFL && (prop & SA_KILL) && td->td_priority > PUSER)
2369 		sched_prio(td, PUSER);
2370 	if (TD_ON_SLEEPQ(td)) {
2371 		/*
2372 		 * If thread is sleeping uninterruptibly
2373 		 * we can't interrupt the sleep... the signal will
2374 		 * be noticed when the process returns through
2375 		 * trap() or syscall().
2376 		 */
2377 		if ((td->td_flags & TDF_SINTR) == 0)
2378 			goto out;
2379 		/*
2380 		 * If SIGCONT is default (or ignored) and process is
2381 		 * asleep, we are finished; the process should not
2382 		 * be awakened.
2383 		 */
2384 		if ((prop & SA_CONT) && action == SIG_DFL) {
2385 			thread_unlock(td);
2386 			PROC_SUNLOCK(p);
2387 			sigqueue_delete(&p->p_sigqueue, sig);
2388 			/*
2389 			 * It may be on either list in this state.
2390 			 * Remove from both for now.
2391 			 */
2392 			sigqueue_delete(&td->td_sigqueue, sig);
2393 			return;
2394 		}
2395 
2396 		/*
2397 		 * Don't awaken a sleeping thread for SIGSTOP if the
2398 		 * STOP signal is deferred.
2399 		 */
2400 		if ((prop & SA_STOP) && (td->td_flags & TDF_SBDRY))
2401 			goto out;
2402 
2403 		/*
2404 		 * Give low priority threads a better chance to run.
2405 		 */
2406 		if (td->td_priority > PUSER)
2407 			sched_prio(td, PUSER);
2408 
2409 		wakeup_swapper = sleepq_abort(td, intrval);
2410 	} else {
2411 		/*
2412 		 * Other states do nothing with the signal immediately,
2413 		 * other than kicking ourselves if we are running.
2414 		 * It will either never be noticed, or noticed very soon.
2415 		 */
2416 #ifdef SMP
2417 		if (TD_IS_RUNNING(td) && td != curthread)
2418 			forward_signal(td);
2419 #endif
2420 	}
2421 out:
2422 	PROC_SUNLOCK(p);
2423 	thread_unlock(td);
2424 	if (wakeup_swapper)
2425 		kick_proc0();
2426 }
2427 
2428 static void
2429 sig_suspend_threads(struct thread *td, struct proc *p, int sending)
2430 {
2431 	struct thread *td2;
2432 
2433 	PROC_LOCK_ASSERT(p, MA_OWNED);
2434 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2435 
2436 	FOREACH_THREAD_IN_PROC(p, td2) {
2437 		thread_lock(td2);
2438 		td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
2439 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2440 		    (td2->td_flags & TDF_SINTR)) {
2441 			if (td2->td_flags & TDF_SBDRY) {
2442 				/*
2443 				 * Once a thread is asleep with
2444 				 * TDF_SBDRY set, it should never
2445 				 * become suspended due to this check.
2446 				 */
2447 				KASSERT(!TD_IS_SUSPENDED(td2),
2448 				    ("thread with deferred stops suspended"));
2449 			} else if (!TD_IS_SUSPENDED(td2)) {
2450 				thread_suspend_one(td2);
2451 			}
2452 		} else if (!TD_IS_SUSPENDED(td2)) {
2453 			if (sending || td != td2)
2454 				td2->td_flags |= TDF_ASTPENDING;
2455 #ifdef SMP
2456 			if (TD_IS_RUNNING(td2) && td2 != td)
2457 				forward_signal(td2);
2458 #endif
2459 		}
2460 		thread_unlock(td2);
2461 	}
2462 }
2463 
2464 int
2465 ptracestop(struct thread *td, int sig)
2466 {
2467 	struct proc *p = td->td_proc;
2468 
2469 	PROC_LOCK_ASSERT(p, MA_OWNED);
2470 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2471 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2472 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2473 
2474 	td->td_dbgflags |= TDB_XSIG;
2475 	td->td_xsig = sig;
2476 	PROC_SLOCK(p);
2477 	while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2478 		if (p->p_flag & P_SINGLE_EXIT) {
2479 			td->td_dbgflags &= ~TDB_XSIG;
2480 			PROC_SUNLOCK(p);
2481 			return (sig);
2482 		}
2483 		/*
2484 		 * Just make wait() to work, the last stopped thread
2485 		 * will win.
2486 		 */
2487 		p->p_xstat = sig;
2488 		p->p_xthread = td;
2489 		p->p_flag |= (P_STOPPED_SIG|P_STOPPED_TRACE);
2490 		sig_suspend_threads(td, p, 0);
2491 		if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2492 			td->td_dbgflags &= ~TDB_STOPATFORK;
2493 			cv_broadcast(&p->p_dbgwait);
2494 		}
2495 stopme:
2496 		thread_suspend_switch(td, p);
2497 		if (p->p_xthread == td)
2498 			p->p_xthread = NULL;
2499 		if (!(p->p_flag & P_TRACED))
2500 			break;
2501 		if (td->td_dbgflags & TDB_SUSPEND) {
2502 			if (p->p_flag & P_SINGLE_EXIT)
2503 				break;
2504 			goto stopme;
2505 		}
2506 	}
2507 	PROC_SUNLOCK(p);
2508 	return (td->td_xsig);
2509 }
2510 
2511 static void
2512 reschedule_signals(struct proc *p, sigset_t block, int flags)
2513 {
2514 	struct sigacts *ps;
2515 	struct thread *td;
2516 	int sig;
2517 
2518 	PROC_LOCK_ASSERT(p, MA_OWNED);
2519 	ps = p->p_sigacts;
2520 	mtx_assert(&ps->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0 ?
2521 	    MA_OWNED : MA_NOTOWNED);
2522 	if (SIGISEMPTY(p->p_siglist))
2523 		return;
2524 	SIGSETAND(block, p->p_siglist);
2525 	while ((sig = sig_ffs(&block)) != 0) {
2526 		SIGDELSET(block, sig);
2527 		td = sigtd(p, sig, 0);
2528 		signotify(td);
2529 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2530 			mtx_lock(&ps->ps_mtx);
2531 		if (p->p_flag & P_TRACED || SIGISMEMBER(ps->ps_sigcatch, sig))
2532 			tdsigwakeup(td, sig, SIG_CATCH,
2533 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2534 			     ERESTART));
2535 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2536 			mtx_unlock(&ps->ps_mtx);
2537 	}
2538 }
2539 
2540 void
2541 tdsigcleanup(struct thread *td)
2542 {
2543 	struct proc *p;
2544 	sigset_t unblocked;
2545 
2546 	p = td->td_proc;
2547 	PROC_LOCK_ASSERT(p, MA_OWNED);
2548 
2549 	sigqueue_flush(&td->td_sigqueue);
2550 	if (p->p_numthreads == 1)
2551 		return;
2552 
2553 	/*
2554 	 * Since we cannot handle signals, notify signal post code
2555 	 * about this by filling the sigmask.
2556 	 *
2557 	 * Also, if needed, wake up thread(s) that do not block the
2558 	 * same signals as the exiting thread, since the thread might
2559 	 * have been selected for delivery and woken up.
2560 	 */
2561 	SIGFILLSET(unblocked);
2562 	SIGSETNAND(unblocked, td->td_sigmask);
2563 	SIGFILLSET(td->td_sigmask);
2564 	reschedule_signals(p, unblocked, 0);
2565 
2566 }
2567 
2568 /*
2569  * Defer the delivery of SIGSTOP for the current thread.  Returns true
2570  * if stops were deferred and false if they were already deferred.
2571  */
2572 int
2573 sigdeferstop(void)
2574 {
2575 	struct thread *td;
2576 
2577 	td = curthread;
2578 	if (td->td_flags & TDF_SBDRY)
2579 		return (0);
2580 	thread_lock(td);
2581 	td->td_flags |= TDF_SBDRY;
2582 	thread_unlock(td);
2583 	return (1);
2584 }
2585 
2586 /*
2587  * Permit the delivery of SIGSTOP for the current thread.  This does
2588  * not immediately suspend if a stop was posted.  Instead, the thread
2589  * will suspend either via ast() or a subsequent interruptible sleep.
2590  */
2591 int
2592 sigallowstop(void)
2593 {
2594 	struct thread *td;
2595 	int prev;
2596 
2597 	td = curthread;
2598 	thread_lock(td);
2599 	prev = (td->td_flags & TDF_SBDRY) != 0;
2600 	td->td_flags &= ~TDF_SBDRY;
2601 	thread_unlock(td);
2602 	return (prev);
2603 }
2604 
2605 /*
2606  * If the current process has received a signal (should be caught or cause
2607  * termination, should interrupt current syscall), return the signal number.
2608  * Stop signals with default action are processed immediately, then cleared;
2609  * they aren't returned.  This is checked after each entry to the system for
2610  * a syscall or trap (though this can usually be done without calling issignal
2611  * by checking the pending signal masks in cursig.) The normal call
2612  * sequence is
2613  *
2614  *	while (sig = cursig(curthread))
2615  *		postsig(sig);
2616  */
2617 static int
2618 issignal(struct thread *td)
2619 {
2620 	struct proc *p;
2621 	struct sigacts *ps;
2622 	struct sigqueue *queue;
2623 	sigset_t sigpending;
2624 	int sig, prop, newsig;
2625 
2626 	p = td->td_proc;
2627 	ps = p->p_sigacts;
2628 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2629 	PROC_LOCK_ASSERT(p, MA_OWNED);
2630 	for (;;) {
2631 		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
2632 
2633 		sigpending = td->td_sigqueue.sq_signals;
2634 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
2635 		SIGSETNAND(sigpending, td->td_sigmask);
2636 
2637 		if (p->p_flag & P_PPWAIT || td->td_flags & TDF_SBDRY)
2638 			SIG_STOPSIGMASK(sigpending);
2639 		if (SIGISEMPTY(sigpending))	/* no signal to send */
2640 			return (0);
2641 		sig = sig_ffs(&sigpending);
2642 
2643 		if (p->p_stops & S_SIG) {
2644 			mtx_unlock(&ps->ps_mtx);
2645 			stopevent(p, S_SIG, sig);
2646 			mtx_lock(&ps->ps_mtx);
2647 		}
2648 
2649 		/*
2650 		 * We should see pending but ignored signals
2651 		 * only if P_TRACED was on when they were posted.
2652 		 */
2653 		if (SIGISMEMBER(ps->ps_sigignore, sig) && (traced == 0)) {
2654 			sigqueue_delete(&td->td_sigqueue, sig);
2655 			sigqueue_delete(&p->p_sigqueue, sig);
2656 			continue;
2657 		}
2658 		if (p->p_flag & P_TRACED && (p->p_flag & P_PPTRACE) == 0) {
2659 			/*
2660 			 * If traced, always stop.
2661 			 * Remove old signal from queue before the stop.
2662 			 * XXX shrug off debugger, it causes siginfo to
2663 			 * be thrown away.
2664 			 */
2665 			queue = &td->td_sigqueue;
2666 			td->td_dbgksi.ksi_signo = 0;
2667 			if (sigqueue_get(queue, sig, &td->td_dbgksi) == 0) {
2668 				queue = &p->p_sigqueue;
2669 				sigqueue_get(queue, sig, &td->td_dbgksi);
2670 			}
2671 
2672 			mtx_unlock(&ps->ps_mtx);
2673 			newsig = ptracestop(td, sig);
2674 			mtx_lock(&ps->ps_mtx);
2675 
2676 			if (sig != newsig) {
2677 
2678 				/*
2679 				 * If parent wants us to take the signal,
2680 				 * then it will leave it in p->p_xstat;
2681 				 * otherwise we just look for signals again.
2682 				*/
2683 				if (newsig == 0)
2684 					continue;
2685 				sig = newsig;
2686 
2687 				/*
2688 				 * Put the new signal into td_sigqueue. If the
2689 				 * signal is being masked, look for other
2690 				 * signals.
2691 				 */
2692 				sigqueue_add(queue, sig, NULL);
2693 				if (SIGISMEMBER(td->td_sigmask, sig))
2694 					continue;
2695 				signotify(td);
2696 			} else {
2697 				if (td->td_dbgksi.ksi_signo != 0) {
2698 					td->td_dbgksi.ksi_flags |= KSI_HEAD;
2699 					if (sigqueue_add(&td->td_sigqueue, sig,
2700 					    &td->td_dbgksi) != 0)
2701 						td->td_dbgksi.ksi_signo = 0;
2702 				}
2703 				if (td->td_dbgksi.ksi_signo == 0)
2704 					sigqueue_add(&td->td_sigqueue, sig,
2705 					    NULL);
2706 			}
2707 
2708 			/*
2709 			 * If the traced bit got turned off, go back up
2710 			 * to the top to rescan signals.  This ensures
2711 			 * that p_sig* and p_sigact are consistent.
2712 			 */
2713 			if ((p->p_flag & P_TRACED) == 0)
2714 				continue;
2715 		}
2716 
2717 		prop = sigprop(sig);
2718 
2719 		/*
2720 		 * Decide whether the signal should be returned.
2721 		 * Return the signal's number, or fall through
2722 		 * to clear it from the pending mask.
2723 		 */
2724 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2725 
2726 		case (intptr_t)SIG_DFL:
2727 			/*
2728 			 * Don't take default actions on system processes.
2729 			 */
2730 			if (p->p_pid <= 1) {
2731 #ifdef DIAGNOSTIC
2732 				/*
2733 				 * Are you sure you want to ignore SIGSEGV
2734 				 * in init? XXX
2735 				 */
2736 				printf("Process (pid %lu) got signal %d\n",
2737 					(u_long)p->p_pid, sig);
2738 #endif
2739 				break;		/* == ignore */
2740 			}
2741 			/*
2742 			 * If there is a pending stop signal to process
2743 			 * with default action, stop here,
2744 			 * then clear the signal.  However,
2745 			 * if process is member of an orphaned
2746 			 * process group, ignore tty stop signals.
2747 			 */
2748 			if (prop & SA_STOP) {
2749 				if (p->p_flag & (P_TRACED|P_WEXIT) ||
2750 				    (p->p_pgrp->pg_jobc == 0 &&
2751 				     prop & SA_TTYSTOP))
2752 					break;	/* == ignore */
2753 				mtx_unlock(&ps->ps_mtx);
2754 				WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2755 				    &p->p_mtx.lock_object, "Catching SIGSTOP");
2756 				p->p_flag |= P_STOPPED_SIG;
2757 				p->p_xstat = sig;
2758 				PROC_SLOCK(p);
2759 				sig_suspend_threads(td, p, 0);
2760 				thread_suspend_switch(td, p);
2761 				PROC_SUNLOCK(p);
2762 				mtx_lock(&ps->ps_mtx);
2763 				break;
2764 			} else if (prop & SA_IGNORE) {
2765 				/*
2766 				 * Except for SIGCONT, shouldn't get here.
2767 				 * Default action is to ignore; drop it.
2768 				 */
2769 				break;		/* == ignore */
2770 			} else
2771 				return (sig);
2772 			/*NOTREACHED*/
2773 
2774 		case (intptr_t)SIG_IGN:
2775 			/*
2776 			 * Masking above should prevent us ever trying
2777 			 * to take action on an ignored signal other
2778 			 * than SIGCONT, unless process is traced.
2779 			 */
2780 			if ((prop & SA_CONT) == 0 &&
2781 			    (p->p_flag & P_TRACED) == 0)
2782 				printf("issignal\n");
2783 			break;		/* == ignore */
2784 
2785 		default:
2786 			/*
2787 			 * This signal has an action, let
2788 			 * postsig() process it.
2789 			 */
2790 			return (sig);
2791 		}
2792 		sigqueue_delete(&td->td_sigqueue, sig);	/* take the signal! */
2793 		sigqueue_delete(&p->p_sigqueue, sig);
2794 	}
2795 	/* NOTREACHED */
2796 }
2797 
2798 void
2799 thread_stopped(struct proc *p)
2800 {
2801 	int n;
2802 
2803 	PROC_LOCK_ASSERT(p, MA_OWNED);
2804 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2805 	n = p->p_suspcount;
2806 	if (p == curproc)
2807 		n++;
2808 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
2809 		PROC_SUNLOCK(p);
2810 		p->p_flag &= ~P_WAITED;
2811 		PROC_LOCK(p->p_pptr);
2812 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
2813 			CLD_TRAPPED : CLD_STOPPED);
2814 		PROC_UNLOCK(p->p_pptr);
2815 		PROC_SLOCK(p);
2816 	}
2817 }
2818 
2819 /*
2820  * Take the action for the specified signal
2821  * from the current set of pending signals.
2822  */
2823 int
2824 postsig(sig)
2825 	register int sig;
2826 {
2827 	struct thread *td = curthread;
2828 	register struct proc *p = td->td_proc;
2829 	struct sigacts *ps;
2830 	sig_t action;
2831 	ksiginfo_t ksi;
2832 	sigset_t returnmask;
2833 
2834 	KASSERT(sig != 0, ("postsig"));
2835 
2836 	PROC_LOCK_ASSERT(p, MA_OWNED);
2837 	ps = p->p_sigacts;
2838 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2839 	ksiginfo_init(&ksi);
2840 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
2841 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
2842 		return (0);
2843 	ksi.ksi_signo = sig;
2844 	if (ksi.ksi_code == SI_TIMER)
2845 		itimer_accept(p, ksi.ksi_timerid, &ksi);
2846 	action = ps->ps_sigact[_SIG_IDX(sig)];
2847 #ifdef KTRACE
2848 	if (KTRPOINT(td, KTR_PSIG))
2849 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
2850 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
2851 #endif
2852 	if (p->p_stops & S_SIG) {
2853 		mtx_unlock(&ps->ps_mtx);
2854 		stopevent(p, S_SIG, sig);
2855 		mtx_lock(&ps->ps_mtx);
2856 	}
2857 
2858 	if (action == SIG_DFL) {
2859 		/*
2860 		 * Default action, where the default is to kill
2861 		 * the process.  (Other cases were ignored above.)
2862 		 */
2863 		mtx_unlock(&ps->ps_mtx);
2864 		sigexit(td, sig);
2865 		/* NOTREACHED */
2866 	} else {
2867 		/*
2868 		 * If we get here, the signal must be caught.
2869 		 */
2870 		KASSERT(action != SIG_IGN && !SIGISMEMBER(td->td_sigmask, sig),
2871 		    ("postsig action"));
2872 		/*
2873 		 * Set the new mask value and also defer further
2874 		 * occurrences of this signal.
2875 		 *
2876 		 * Special case: user has done a sigsuspend.  Here the
2877 		 * current mask is not of interest, but rather the
2878 		 * mask from before the sigsuspend is what we want
2879 		 * restored after the signal processing is completed.
2880 		 */
2881 		if (td->td_pflags & TDP_OLDMASK) {
2882 			returnmask = td->td_oldsigmask;
2883 			td->td_pflags &= ~TDP_OLDMASK;
2884 		} else
2885 			returnmask = td->td_sigmask;
2886 
2887 		if (p->p_sig == sig) {
2888 			p->p_code = 0;
2889 			p->p_sig = 0;
2890 		}
2891 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
2892 		postsig_done(sig, td, ps);
2893 	}
2894 	return (1);
2895 }
2896 
2897 /*
2898  * Kill the current process for stated reason.
2899  */
2900 void
2901 killproc(p, why)
2902 	struct proc *p;
2903 	char *why;
2904 {
2905 
2906 	PROC_LOCK_ASSERT(p, MA_OWNED);
2907 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
2908 	    p->p_comm);
2909 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n", p->p_pid,
2910 	    p->p_comm, p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2911 	p->p_flag |= P_WKILLED;
2912 	kern_psignal(p, SIGKILL);
2913 }
2914 
2915 /*
2916  * Force the current process to exit with the specified signal, dumping core
2917  * if appropriate.  We bypass the normal tests for masked and caught signals,
2918  * allowing unrecoverable failures to terminate the process without changing
2919  * signal state.  Mark the accounting record with the signal termination.
2920  * If dumping core, save the signal number for the debugger.  Calls exit and
2921  * does not return.
2922  */
2923 void
2924 sigexit(td, sig)
2925 	struct thread *td;
2926 	int sig;
2927 {
2928 	struct proc *p = td->td_proc;
2929 
2930 	PROC_LOCK_ASSERT(p, MA_OWNED);
2931 	p->p_acflag |= AXSIG;
2932 	/*
2933 	 * We must be single-threading to generate a core dump.  This
2934 	 * ensures that the registers in the core file are up-to-date.
2935 	 * Also, the ELF dump handler assumes that the thread list doesn't
2936 	 * change out from under it.
2937 	 *
2938 	 * XXX If another thread attempts to single-thread before us
2939 	 *     (e.g. via fork()), we won't get a dump at all.
2940 	 */
2941 	if ((sigprop(sig) & SA_CORE) && thread_single(p, SINGLE_NO_EXIT) == 0) {
2942 		p->p_sig = sig;
2943 		/*
2944 		 * Log signals which would cause core dumps
2945 		 * (Log as LOG_INFO to appease those who don't want
2946 		 * these messages.)
2947 		 * XXX : Todo, as well as euid, write out ruid too
2948 		 * Note that coredump() drops proc lock.
2949 		 */
2950 		if (coredump(td) == 0)
2951 			sig |= WCOREFLAG;
2952 		if (kern_logsigexit)
2953 			log(LOG_INFO,
2954 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2955 			    p->p_pid, p->p_comm,
2956 			    td->td_ucred ? td->td_ucred->cr_uid : -1,
2957 			    sig &~ WCOREFLAG,
2958 			    sig & WCOREFLAG ? " (core dumped)" : "");
2959 	} else
2960 		PROC_UNLOCK(p);
2961 	exit1(td, W_EXITCODE(0, sig));
2962 	/* NOTREACHED */
2963 }
2964 
2965 /*
2966  * Send queued SIGCHLD to parent when child process's state
2967  * is changed.
2968  */
2969 static void
2970 sigparent(struct proc *p, int reason, int status)
2971 {
2972 	PROC_LOCK_ASSERT(p, MA_OWNED);
2973 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
2974 
2975 	if (p->p_ksi != NULL) {
2976 		p->p_ksi->ksi_signo  = SIGCHLD;
2977 		p->p_ksi->ksi_code   = reason;
2978 		p->p_ksi->ksi_status = status;
2979 		p->p_ksi->ksi_pid    = p->p_pid;
2980 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
2981 		if (KSI_ONQ(p->p_ksi))
2982 			return;
2983 	}
2984 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
2985 }
2986 
2987 static void
2988 childproc_jobstate(struct proc *p, int reason, int sig)
2989 {
2990 	struct sigacts *ps;
2991 
2992 	PROC_LOCK_ASSERT(p, MA_OWNED);
2993 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
2994 
2995 	/*
2996 	 * Wake up parent sleeping in kern_wait(), also send
2997 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
2998 	 * that parent will awake, because parent may masked
2999 	 * the signal.
3000 	 */
3001 	p->p_pptr->p_flag |= P_STATCHILD;
3002 	wakeup(p->p_pptr);
3003 
3004 	ps = p->p_pptr->p_sigacts;
3005 	mtx_lock(&ps->ps_mtx);
3006 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3007 		mtx_unlock(&ps->ps_mtx);
3008 		sigparent(p, reason, sig);
3009 	} else
3010 		mtx_unlock(&ps->ps_mtx);
3011 }
3012 
3013 void
3014 childproc_stopped(struct proc *p, int reason)
3015 {
3016 	/* p_xstat is a plain signal number, not a full wait() status here. */
3017 	childproc_jobstate(p, reason, p->p_xstat);
3018 }
3019 
3020 void
3021 childproc_continued(struct proc *p)
3022 {
3023 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3024 }
3025 
3026 void
3027 childproc_exited(struct proc *p)
3028 {
3029 	int reason;
3030 	int xstat = p->p_xstat; /* convert to int */
3031 	int status;
3032 
3033 	if (WCOREDUMP(xstat))
3034 		reason = CLD_DUMPED, status = WTERMSIG(xstat);
3035 	else if (WIFSIGNALED(xstat))
3036 		reason = CLD_KILLED, status = WTERMSIG(xstat);
3037 	else
3038 		reason = CLD_EXITED, status = WEXITSTATUS(xstat);
3039 	/*
3040 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3041 	 * done in exit1().
3042 	 */
3043 	sigparent(p, reason, status);
3044 }
3045 
3046 /*
3047  * We only have 1 character for the core count in the format
3048  * string, so the range will be 0-9
3049  */
3050 #define MAX_NUM_CORES 10
3051 static int num_cores = 5;
3052 
3053 static int
3054 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3055 {
3056 	int error;
3057 	int new_val;
3058 
3059 	new_val = num_cores;
3060 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3061 	if (error != 0 || req->newptr == NULL)
3062 		return (error);
3063 	if (new_val > MAX_NUM_CORES)
3064 		new_val = MAX_NUM_CORES;
3065 	if (new_val < 0)
3066 		new_val = 0;
3067 	num_cores = new_val;
3068 	return (0);
3069 }
3070 SYSCTL_PROC(_debug, OID_AUTO, ncores, CTLTYPE_INT|CTLFLAG_RW,
3071 	    0, sizeof(int), sysctl_debug_num_cores_check, "I", "");
3072 
3073 #if defined(COMPRESS_USER_CORES)
3074 int compress_user_cores = 1;
3075 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores, CTLFLAG_RW,
3076     &compress_user_cores, 0, "Compression of user corefiles");
3077 
3078 int compress_user_cores_gzlevel = -1; /* default level */
3079 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_gzlevel, CTLFLAG_RW,
3080     &compress_user_cores_gzlevel, -1, "Corefile gzip compression level");
3081 
3082 #define GZ_SUFFIX	".gz"
3083 #define GZ_SUFFIX_LEN	3
3084 #endif
3085 
3086 static char corefilename[MAXPATHLEN] = {"%N.core"};
3087 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RWTUN, corefilename,
3088     sizeof(corefilename), "Process corefile name format string");
3089 
3090 /*
3091  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3092  * Expand the name described in corefilename, using name, uid, and pid
3093  * and open/create core file.
3094  * corefilename is a printf-like string, with three format specifiers:
3095  *	%N	name of process ("name")
3096  *	%P	process id (pid)
3097  *	%U	user id (uid)
3098  * For example, "%N.core" is the default; they can be disabled completely
3099  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3100  * This is controlled by the sysctl variable kern.corefile (see above).
3101  */
3102 static int
3103 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3104     int compress, struct vnode **vpp, char **namep)
3105 {
3106 	struct nameidata nd;
3107 	struct sbuf sb;
3108 	const char *format;
3109 	char *hostname, *name;
3110 	int indexpos, i, error, cmode, flags, oflags;
3111 
3112 	hostname = NULL;
3113 	format = corefilename;
3114 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3115 	indexpos = -1;
3116 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3117 	for (i = 0; format[i] != '\0'; i++) {
3118 		switch (format[i]) {
3119 		case '%':	/* Format character */
3120 			i++;
3121 			switch (format[i]) {
3122 			case '%':
3123 				sbuf_putc(&sb, '%');
3124 				break;
3125 			case 'H':	/* hostname */
3126 				if (hostname == NULL) {
3127 					hostname = malloc(MAXHOSTNAMELEN,
3128 					    M_TEMP, M_WAITOK);
3129 				}
3130 				getcredhostname(td->td_ucred, hostname,
3131 				    MAXHOSTNAMELEN);
3132 				sbuf_printf(&sb, "%s", hostname);
3133 				break;
3134 			case 'I':	/* autoincrementing index */
3135 				sbuf_printf(&sb, "0");
3136 				indexpos = sbuf_len(&sb) - 1;
3137 				break;
3138 			case 'N':	/* process name */
3139 				sbuf_printf(&sb, "%s", comm);
3140 				break;
3141 			case 'P':	/* process id */
3142 				sbuf_printf(&sb, "%u", pid);
3143 				break;
3144 			case 'U':	/* user id */
3145 				sbuf_printf(&sb, "%u", uid);
3146 				break;
3147 			default:
3148 				log(LOG_ERR,
3149 				    "Unknown format character %c in "
3150 				    "corename `%s'\n", format[i], format);
3151 				break;
3152 			}
3153 			break;
3154 		default:
3155 			sbuf_putc(&sb, format[i]);
3156 			break;
3157 		}
3158 	}
3159 	free(hostname, M_TEMP);
3160 #ifdef COMPRESS_USER_CORES
3161 	if (compress)
3162 		sbuf_printf(&sb, GZ_SUFFIX);
3163 #endif
3164 	if (sbuf_error(&sb) != 0) {
3165 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3166 		    "long\n", (long)pid, comm, (u_long)uid);
3167 		sbuf_delete(&sb);
3168 		free(name, M_TEMP);
3169 		return (ENOMEM);
3170 	}
3171 	sbuf_finish(&sb);
3172 	sbuf_delete(&sb);
3173 
3174 	cmode = S_IRUSR | S_IWUSR;
3175 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3176 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3177 
3178 	/*
3179 	 * If the core format has a %I in it, then we need to check
3180 	 * for existing corefiles before returning a name.
3181 	 * To do this we iterate over 0..num_cores to find a
3182 	 * non-existing core file name to use.
3183 	 */
3184 	if (indexpos != -1) {
3185 		for (i = 0; i < num_cores; i++) {
3186 			flags = O_CREAT | O_EXCL | FWRITE | O_NOFOLLOW;
3187 			name[indexpos] = '0' + i;
3188 			NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3189 			error = vn_open_cred(&nd, &flags, cmode, oflags,
3190 			    td->td_ucred, NULL);
3191 			if (error) {
3192 				if (error == EEXIST)
3193 					continue;
3194 				log(LOG_ERR,
3195 				    "pid %d (%s), uid (%u):  Path `%s' failed "
3196 				    "on initial open test, error = %d\n",
3197 				    pid, comm, uid, name, error);
3198 			}
3199 			goto out;
3200 		}
3201 	}
3202 
3203 	flags = O_CREAT | FWRITE | O_NOFOLLOW;
3204 	NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3205 	error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred, NULL);
3206 out:
3207 	if (error) {
3208 #ifdef AUDIT
3209 		audit_proc_coredump(td, name, error);
3210 #endif
3211 		free(name, M_TEMP);
3212 		return (error);
3213 	}
3214 	NDFREE(&nd, NDF_ONLY_PNBUF);
3215 	*vpp = nd.ni_vp;
3216 	*namep = name;
3217 	return (0);
3218 }
3219 
3220 /*
3221  * Dump a process' core.  The main routine does some
3222  * policy checking, and creates the name of the coredump;
3223  * then it passes on a vnode and a size limit to the process-specific
3224  * coredump routine if there is one; if there _is not_ one, it returns
3225  * ENOSYS; otherwise it returns the error from the process-specific routine.
3226  */
3227 
3228 static int
3229 coredump(struct thread *td)
3230 {
3231 	struct proc *p = td->td_proc;
3232 	struct ucred *cred = td->td_ucred;
3233 	struct vnode *vp;
3234 	struct flock lf;
3235 	struct vattr vattr;
3236 	int error, error1, locked;
3237 	char *name;			/* name of corefile */
3238 	void *rl_cookie;
3239 	off_t limit;
3240 	int compress;
3241 	char *data = NULL;
3242 	size_t len;
3243 	char *fullpath, *freepath = NULL;
3244 
3245 #ifdef COMPRESS_USER_CORES
3246 	compress = compress_user_cores;
3247 #else
3248 	compress = 0;
3249 #endif
3250 	PROC_LOCK_ASSERT(p, MA_OWNED);
3251 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3252 	_STOPEVENT(p, S_CORE, 0);
3253 
3254 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
3255 	    (p->p_flag2 & P2_NOTRACE) != 0) {
3256 		PROC_UNLOCK(p);
3257 		return (EFAULT);
3258 	}
3259 
3260 	/*
3261 	 * Note that the bulk of limit checking is done after
3262 	 * the corefile is created.  The exception is if the limit
3263 	 * for corefiles is 0, in which case we don't bother
3264 	 * creating the corefile at all.  This layout means that
3265 	 * a corefile is truncated instead of not being created,
3266 	 * if it is larger than the limit.
3267 	 */
3268 	limit = (off_t)lim_cur(p, RLIMIT_CORE);
3269 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
3270 		PROC_UNLOCK(p);
3271 		return (EFBIG);
3272 	}
3273 	PROC_UNLOCK(p);
3274 
3275 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td, compress,
3276 	    &vp, &name);
3277 	if (error != 0)
3278 		return (error);
3279 
3280 	/*
3281 	 * Don't dump to non-regular files or files with links.
3282 	 * Do not dump into system files.
3283 	 */
3284 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
3285 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0) {
3286 		VOP_UNLOCK(vp, 0);
3287 		error = EFAULT;
3288 		goto close;
3289 	}
3290 
3291 	VOP_UNLOCK(vp, 0);
3292 
3293 	/* Postpone other writers, including core dumps of other processes. */
3294 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
3295 
3296 	lf.l_whence = SEEK_SET;
3297 	lf.l_start = 0;
3298 	lf.l_len = 0;
3299 	lf.l_type = F_WRLCK;
3300 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
3301 
3302 	VATTR_NULL(&vattr);
3303 	vattr.va_size = 0;
3304 	if (set_core_nodump_flag)
3305 		vattr.va_flags = UF_NODUMP;
3306 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3307 	VOP_SETATTR(vp, &vattr, cred);
3308 	VOP_UNLOCK(vp, 0);
3309 	PROC_LOCK(p);
3310 	p->p_acflag |= ACORE;
3311 	PROC_UNLOCK(p);
3312 
3313 	if (p->p_sysent->sv_coredump != NULL) {
3314 		error = p->p_sysent->sv_coredump(td, vp, limit,
3315 		    compress ? IMGACT_CORE_COMPRESS : 0);
3316 	} else {
3317 		error = ENOSYS;
3318 	}
3319 
3320 	if (locked) {
3321 		lf.l_type = F_UNLCK;
3322 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3323 	}
3324 	vn_rangelock_unlock(vp, rl_cookie);
3325 close:
3326 	error1 = vn_close(vp, FWRITE, cred, td);
3327 	if (error == 0)
3328 		error = error1;
3329 	else
3330 		goto out;
3331 	/*
3332 	 * Notify the userland helper that a process triggered a core dump.
3333 	 * This allows the helper to run an automated debugging session.
3334 	 */
3335 	len = MAXPATHLEN * 2 + 5 /* comm= */ + 5 /* core= */ + 1;
3336 	data = malloc(len, M_TEMP, M_NOWAIT);
3337 	if (data == NULL)
3338 		goto out;
3339 	if (vn_fullpath_global(td, p->p_textvp, &fullpath, &freepath) != 0)
3340 		goto out;
3341 	snprintf(data, len, "comm=%s", fullpath);
3342 	if (freepath != NULL) {
3343 		free(freepath, M_TEMP);
3344 		freepath = NULL;
3345 	}
3346 	if (vn_fullpath_global(td, vp, &fullpath, &freepath) != 0)
3347 		goto out;
3348 	snprintf(data, len, "%s core=%s", data, fullpath);
3349 	devctl_notify("kernel", "signal", "coredump", data);
3350 	free(name, M_TEMP);
3351 out:
3352 #ifdef AUDIT
3353 	audit_proc_coredump(td, name, error);
3354 #endif
3355 	if (freepath != NULL)
3356 		free(freepath, M_TEMP);
3357 	if (data != NULL)
3358 		free(data, M_TEMP);
3359 	free(name, M_TEMP);
3360 	return (error);
3361 }
3362 
3363 /*
3364  * Nonexistent system call-- signal process (may want to handle it).  Flag
3365  * error in case process won't see signal immediately (blocked or ignored).
3366  */
3367 #ifndef _SYS_SYSPROTO_H_
3368 struct nosys_args {
3369 	int	dummy;
3370 };
3371 #endif
3372 /* ARGSUSED */
3373 int
3374 nosys(td, args)
3375 	struct thread *td;
3376 	struct nosys_args *args;
3377 {
3378 	struct proc *p = td->td_proc;
3379 
3380 	PROC_LOCK(p);
3381 	tdsignal(td, SIGSYS);
3382 	PROC_UNLOCK(p);
3383 	return (ENOSYS);
3384 }
3385 
3386 /*
3387  * Send a SIGIO or SIGURG signal to a process or process group using stored
3388  * credentials rather than those of the current process.
3389  */
3390 void
3391 pgsigio(sigiop, sig, checkctty)
3392 	struct sigio **sigiop;
3393 	int sig, checkctty;
3394 {
3395 	ksiginfo_t ksi;
3396 	struct sigio *sigio;
3397 
3398 	ksiginfo_init(&ksi);
3399 	ksi.ksi_signo = sig;
3400 	ksi.ksi_code = SI_KERNEL;
3401 
3402 	SIGIO_LOCK();
3403 	sigio = *sigiop;
3404 	if (sigio == NULL) {
3405 		SIGIO_UNLOCK();
3406 		return;
3407 	}
3408 	if (sigio->sio_pgid > 0) {
3409 		PROC_LOCK(sigio->sio_proc);
3410 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
3411 			kern_psignal(sigio->sio_proc, sig);
3412 		PROC_UNLOCK(sigio->sio_proc);
3413 	} else if (sigio->sio_pgid < 0) {
3414 		struct proc *p;
3415 
3416 		PGRP_LOCK(sigio->sio_pgrp);
3417 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
3418 			PROC_LOCK(p);
3419 			if (p->p_state == PRS_NORMAL &&
3420 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
3421 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
3422 				kern_psignal(p, sig);
3423 			PROC_UNLOCK(p);
3424 		}
3425 		PGRP_UNLOCK(sigio->sio_pgrp);
3426 	}
3427 	SIGIO_UNLOCK();
3428 }
3429 
3430 static int
3431 filt_sigattach(struct knote *kn)
3432 {
3433 	struct proc *p = curproc;
3434 
3435 	kn->kn_ptr.p_proc = p;
3436 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
3437 
3438 	knlist_add(&p->p_klist, kn, 0);
3439 
3440 	return (0);
3441 }
3442 
3443 static void
3444 filt_sigdetach(struct knote *kn)
3445 {
3446 	struct proc *p = kn->kn_ptr.p_proc;
3447 
3448 	knlist_remove(&p->p_klist, kn, 0);
3449 }
3450 
3451 /*
3452  * signal knotes are shared with proc knotes, so we apply a mask to
3453  * the hint in order to differentiate them from process hints.  This
3454  * could be avoided by using a signal-specific knote list, but probably
3455  * isn't worth the trouble.
3456  */
3457 static int
3458 filt_signal(struct knote *kn, long hint)
3459 {
3460 
3461 	if (hint & NOTE_SIGNAL) {
3462 		hint &= ~NOTE_SIGNAL;
3463 
3464 		if (kn->kn_id == hint)
3465 			kn->kn_data++;
3466 	}
3467 	return (kn->kn_data != 0);
3468 }
3469 
3470 struct sigacts *
3471 sigacts_alloc(void)
3472 {
3473 	struct sigacts *ps;
3474 
3475 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
3476 	refcount_init(&ps->ps_refcnt, 1);
3477 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
3478 	return (ps);
3479 }
3480 
3481 void
3482 sigacts_free(struct sigacts *ps)
3483 {
3484 
3485 	if (refcount_release(&ps->ps_refcnt) == 0)
3486 		return;
3487 	mtx_destroy(&ps->ps_mtx);
3488 	free(ps, M_SUBPROC);
3489 }
3490 
3491 struct sigacts *
3492 sigacts_hold(struct sigacts *ps)
3493 {
3494 
3495 	refcount_acquire(&ps->ps_refcnt);
3496 	return (ps);
3497 }
3498 
3499 void
3500 sigacts_copy(struct sigacts *dest, struct sigacts *src)
3501 {
3502 
3503 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
3504 	mtx_lock(&src->ps_mtx);
3505 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
3506 	mtx_unlock(&src->ps_mtx);
3507 }
3508 
3509 int
3510 sigacts_shared(struct sigacts *ps)
3511 {
3512 
3513 	return (ps->ps_refcnt > 1);
3514 }
3515