xref: /freebsd/sys/kern/kern_sig.c (revision 7447ca0eb235974642312b9555caec00b57d8fc1)
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_gzio.h"
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/ctype.h>
46 #include <sys/systm.h>
47 #include <sys/signalvar.h>
48 #include <sys/vnode.h>
49 #include <sys/acct.h>
50 #include <sys/bus.h>
51 #include <sys/capsicum.h>
52 #include <sys/condvar.h>
53 #include <sys/event.h>
54 #include <sys/fcntl.h>
55 #include <sys/imgact.h>
56 #include <sys/kernel.h>
57 #include <sys/ktr.h>
58 #include <sys/ktrace.h>
59 #include <sys/lock.h>
60 #include <sys/malloc.h>
61 #include <sys/mutex.h>
62 #include <sys/refcount.h>
63 #include <sys/namei.h>
64 #include <sys/proc.h>
65 #include <sys/procdesc.h>
66 #include <sys/posix4.h>
67 #include <sys/pioctl.h>
68 #include <sys/racct.h>
69 #include <sys/resourcevar.h>
70 #include <sys/sdt.h>
71 #include <sys/sbuf.h>
72 #include <sys/sleepqueue.h>
73 #include <sys/smp.h>
74 #include <sys/stat.h>
75 #include <sys/sx.h>
76 #include <sys/syscallsubr.h>
77 #include <sys/sysctl.h>
78 #include <sys/sysent.h>
79 #include <sys/syslog.h>
80 #include <sys/sysproto.h>
81 #include <sys/timers.h>
82 #include <sys/unistd.h>
83 #include <sys/wait.h>
84 #include <vm/vm.h>
85 #include <vm/vm_extern.h>
86 #include <vm/uma.h>
87 
88 #include <sys/jail.h>
89 
90 #include <machine/cpu.h>
91 
92 #include <security/audit/audit.h>
93 
94 #define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
95 
96 SDT_PROVIDER_DECLARE(proc);
97 SDT_PROBE_DEFINE3(proc, , , signal__send,
98     "struct thread *", "struct proc *", "int");
99 SDT_PROBE_DEFINE2(proc, , , signal__clear,
100     "int", "ksiginfo_t *");
101 SDT_PROBE_DEFINE3(proc, , , signal__discard,
102     "struct thread *", "struct proc *", "int");
103 
104 static int	coredump(struct thread *);
105 static int	killpg1(struct thread *td, int sig, int pgid, int all,
106 		    ksiginfo_t *ksi);
107 static int	issignal(struct thread *td);
108 static int	sigprop(int sig);
109 static void	tdsigwakeup(struct thread *, int, sig_t, int);
110 static int	sig_suspend_threads(struct thread *, struct proc *, int);
111 static int	filt_sigattach(struct knote *kn);
112 static void	filt_sigdetach(struct knote *kn);
113 static int	filt_signal(struct knote *kn, long hint);
114 static struct thread *sigtd(struct proc *p, int sig, int prop);
115 static void	sigqueue_start(void);
116 
117 static uma_zone_t	ksiginfo_zone = NULL;
118 struct filterops sig_filtops = {
119 	.f_isfd = 0,
120 	.f_attach = filt_sigattach,
121 	.f_detach = filt_sigdetach,
122 	.f_event = filt_signal,
123 };
124 
125 static int	kern_logsigexit = 1;
126 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
127     &kern_logsigexit, 0,
128     "Log processes quitting on abnormal signals to syslog(3)");
129 
130 static int	kern_forcesigexit = 1;
131 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
132     &kern_forcesigexit, 0, "Force trap signal to be handled");
133 
134 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW, 0,
135     "POSIX real time signal");
136 
137 static int	max_pending_per_proc = 128;
138 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
139     &max_pending_per_proc, 0, "Max pending signals per proc");
140 
141 static int	preallocate_siginfo = 1024;
142 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
143     &preallocate_siginfo, 0, "Preallocated signal memory size");
144 
145 static int	signal_overflow = 0;
146 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
147     &signal_overflow, 0, "Number of signals overflew");
148 
149 static int	signal_alloc_fail = 0;
150 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
151     &signal_alloc_fail, 0, "signals failed to be allocated");
152 
153 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
154 
155 /*
156  * Policy -- Can ucred cr1 send SIGIO to process cr2?
157  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
158  * in the right situations.
159  */
160 #define CANSIGIO(cr1, cr2) \
161 	((cr1)->cr_uid == 0 || \
162 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
163 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
164 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
165 	    (cr1)->cr_uid == (cr2)->cr_uid)
166 
167 static int	sugid_coredump;
168 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
169     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
170 
171 static int	capmode_coredump;
172 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
173     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
174 
175 static int	do_coredump = 1;
176 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
177 	&do_coredump, 0, "Enable/Disable coredumps");
178 
179 static int	set_core_nodump_flag = 0;
180 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
181 	0, "Enable setting the NODUMP flag on coredump files");
182 
183 static int	coredump_devctl = 0;
184 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
185 	0, "Generate a devctl notification when processes coredump");
186 
187 /*
188  * Signal properties and actions.
189  * The array below categorizes the signals and their default actions
190  * according to the following properties:
191  */
192 #define	SA_KILL		0x01		/* terminates process by default */
193 #define	SA_CORE		0x02		/* ditto and coredumps */
194 #define	SA_STOP		0x04		/* suspend process */
195 #define	SA_TTYSTOP	0x08		/* ditto, from tty */
196 #define	SA_IGNORE	0x10		/* ignore by default */
197 #define	SA_CONT		0x20		/* continue if suspended */
198 #define	SA_CANTMASK	0x40		/* non-maskable, catchable */
199 
200 static int sigproptbl[NSIG] = {
201 	SA_KILL,			/* SIGHUP */
202 	SA_KILL,			/* SIGINT */
203 	SA_KILL|SA_CORE,		/* SIGQUIT */
204 	SA_KILL|SA_CORE,		/* SIGILL */
205 	SA_KILL|SA_CORE,		/* SIGTRAP */
206 	SA_KILL|SA_CORE,		/* SIGABRT */
207 	SA_KILL|SA_CORE,		/* SIGEMT */
208 	SA_KILL|SA_CORE,		/* SIGFPE */
209 	SA_KILL,			/* SIGKILL */
210 	SA_KILL|SA_CORE,		/* SIGBUS */
211 	SA_KILL|SA_CORE,		/* SIGSEGV */
212 	SA_KILL|SA_CORE,		/* SIGSYS */
213 	SA_KILL,			/* SIGPIPE */
214 	SA_KILL,			/* SIGALRM */
215 	SA_KILL,			/* SIGTERM */
216 	SA_IGNORE,			/* SIGURG */
217 	SA_STOP,			/* SIGSTOP */
218 	SA_STOP|SA_TTYSTOP,		/* SIGTSTP */
219 	SA_IGNORE|SA_CONT,		/* SIGCONT */
220 	SA_IGNORE,			/* SIGCHLD */
221 	SA_STOP|SA_TTYSTOP,		/* SIGTTIN */
222 	SA_STOP|SA_TTYSTOP,		/* SIGTTOU */
223 	SA_IGNORE,			/* SIGIO */
224 	SA_KILL,			/* SIGXCPU */
225 	SA_KILL,			/* SIGXFSZ */
226 	SA_KILL,			/* SIGVTALRM */
227 	SA_KILL,			/* SIGPROF */
228 	SA_IGNORE,			/* SIGWINCH  */
229 	SA_IGNORE,			/* SIGINFO */
230 	SA_KILL,			/* SIGUSR1 */
231 	SA_KILL,			/* SIGUSR2 */
232 };
233 
234 static void reschedule_signals(struct proc *p, sigset_t block, int flags);
235 
236 static void
237 sigqueue_start(void)
238 {
239 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
240 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
241 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
242 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
243 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
244 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
245 }
246 
247 ksiginfo_t *
248 ksiginfo_alloc(int wait)
249 {
250 	int flags;
251 
252 	flags = M_ZERO;
253 	if (! wait)
254 		flags |= M_NOWAIT;
255 	if (ksiginfo_zone != NULL)
256 		return ((ksiginfo_t *)uma_zalloc(ksiginfo_zone, flags));
257 	return (NULL);
258 }
259 
260 void
261 ksiginfo_free(ksiginfo_t *ksi)
262 {
263 	uma_zfree(ksiginfo_zone, ksi);
264 }
265 
266 static __inline int
267 ksiginfo_tryfree(ksiginfo_t *ksi)
268 {
269 	if (!(ksi->ksi_flags & KSI_EXT)) {
270 		uma_zfree(ksiginfo_zone, ksi);
271 		return (1);
272 	}
273 	return (0);
274 }
275 
276 void
277 sigqueue_init(sigqueue_t *list, struct proc *p)
278 {
279 	SIGEMPTYSET(list->sq_signals);
280 	SIGEMPTYSET(list->sq_kill);
281 	TAILQ_INIT(&list->sq_list);
282 	list->sq_proc = p;
283 	list->sq_flags = SQ_INIT;
284 }
285 
286 /*
287  * Get a signal's ksiginfo.
288  * Return:
289  *	0	-	signal not found
290  *	others	-	signal number
291  */
292 static int
293 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
294 {
295 	struct proc *p = sq->sq_proc;
296 	struct ksiginfo *ksi, *next;
297 	int count = 0;
298 
299 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
300 
301 	if (!SIGISMEMBER(sq->sq_signals, signo))
302 		return (0);
303 
304 	if (SIGISMEMBER(sq->sq_kill, signo)) {
305 		count++;
306 		SIGDELSET(sq->sq_kill, signo);
307 	}
308 
309 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
310 		if (ksi->ksi_signo == signo) {
311 			if (count == 0) {
312 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
313 				ksi->ksi_sigq = NULL;
314 				ksiginfo_copy(ksi, si);
315 				if (ksiginfo_tryfree(ksi) && p != NULL)
316 					p->p_pendingcnt--;
317 			}
318 			if (++count > 1)
319 				break;
320 		}
321 	}
322 
323 	if (count <= 1)
324 		SIGDELSET(sq->sq_signals, signo);
325 	si->ksi_signo = signo;
326 	return (signo);
327 }
328 
329 void
330 sigqueue_take(ksiginfo_t *ksi)
331 {
332 	struct ksiginfo *kp;
333 	struct proc	*p;
334 	sigqueue_t	*sq;
335 
336 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
337 		return;
338 
339 	p = sq->sq_proc;
340 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
341 	ksi->ksi_sigq = NULL;
342 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
343 		p->p_pendingcnt--;
344 
345 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
346 	     kp = TAILQ_NEXT(kp, ksi_link)) {
347 		if (kp->ksi_signo == ksi->ksi_signo)
348 			break;
349 	}
350 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo))
351 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
352 }
353 
354 static int
355 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
356 {
357 	struct proc *p = sq->sq_proc;
358 	struct ksiginfo *ksi;
359 	int ret = 0;
360 
361 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
362 
363 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
364 		SIGADDSET(sq->sq_kill, signo);
365 		goto out_set_bit;
366 	}
367 
368 	/* directly insert the ksi, don't copy it */
369 	if (si->ksi_flags & KSI_INS) {
370 		if (si->ksi_flags & KSI_HEAD)
371 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
372 		else
373 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
374 		si->ksi_sigq = sq;
375 		goto out_set_bit;
376 	}
377 
378 	if (__predict_false(ksiginfo_zone == NULL)) {
379 		SIGADDSET(sq->sq_kill, signo);
380 		goto out_set_bit;
381 	}
382 
383 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
384 		signal_overflow++;
385 		ret = EAGAIN;
386 	} else if ((ksi = ksiginfo_alloc(0)) == NULL) {
387 		signal_alloc_fail++;
388 		ret = EAGAIN;
389 	} else {
390 		if (p != NULL)
391 			p->p_pendingcnt++;
392 		ksiginfo_copy(si, ksi);
393 		ksi->ksi_signo = signo;
394 		if (si->ksi_flags & KSI_HEAD)
395 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
396 		else
397 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
398 		ksi->ksi_sigq = sq;
399 	}
400 
401 	if ((si->ksi_flags & KSI_TRAP) != 0 ||
402 	    (si->ksi_flags & KSI_SIGQ) == 0) {
403 		if (ret != 0)
404 			SIGADDSET(sq->sq_kill, signo);
405 		ret = 0;
406 		goto out_set_bit;
407 	}
408 
409 	if (ret != 0)
410 		return (ret);
411 
412 out_set_bit:
413 	SIGADDSET(sq->sq_signals, signo);
414 	return (ret);
415 }
416 
417 void
418 sigqueue_flush(sigqueue_t *sq)
419 {
420 	struct proc *p = sq->sq_proc;
421 	ksiginfo_t *ksi;
422 
423 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
424 
425 	if (p != NULL)
426 		PROC_LOCK_ASSERT(p, MA_OWNED);
427 
428 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
429 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
430 		ksi->ksi_sigq = NULL;
431 		if (ksiginfo_tryfree(ksi) && p != NULL)
432 			p->p_pendingcnt--;
433 	}
434 
435 	SIGEMPTYSET(sq->sq_signals);
436 	SIGEMPTYSET(sq->sq_kill);
437 }
438 
439 static void
440 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
441 {
442 	sigset_t tmp;
443 	struct proc *p1, *p2;
444 	ksiginfo_t *ksi, *next;
445 
446 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
447 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
448 	p1 = src->sq_proc;
449 	p2 = dst->sq_proc;
450 	/* Move siginfo to target list */
451 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
452 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
453 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
454 			if (p1 != NULL)
455 				p1->p_pendingcnt--;
456 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
457 			ksi->ksi_sigq = dst;
458 			if (p2 != NULL)
459 				p2->p_pendingcnt++;
460 		}
461 	}
462 
463 	/* Move pending bits to target list */
464 	tmp = src->sq_kill;
465 	SIGSETAND(tmp, *set);
466 	SIGSETOR(dst->sq_kill, tmp);
467 	SIGSETNAND(src->sq_kill, tmp);
468 
469 	tmp = src->sq_signals;
470 	SIGSETAND(tmp, *set);
471 	SIGSETOR(dst->sq_signals, tmp);
472 	SIGSETNAND(src->sq_signals, tmp);
473 }
474 
475 #if 0
476 static void
477 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
478 {
479 	sigset_t set;
480 
481 	SIGEMPTYSET(set);
482 	SIGADDSET(set, signo);
483 	sigqueue_move_set(src, dst, &set);
484 }
485 #endif
486 
487 static void
488 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
489 {
490 	struct proc *p = sq->sq_proc;
491 	ksiginfo_t *ksi, *next;
492 
493 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
494 
495 	/* Remove siginfo queue */
496 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
497 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
498 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
499 			ksi->ksi_sigq = NULL;
500 			if (ksiginfo_tryfree(ksi) && p != NULL)
501 				p->p_pendingcnt--;
502 		}
503 	}
504 	SIGSETNAND(sq->sq_kill, *set);
505 	SIGSETNAND(sq->sq_signals, *set);
506 }
507 
508 void
509 sigqueue_delete(sigqueue_t *sq, int signo)
510 {
511 	sigset_t set;
512 
513 	SIGEMPTYSET(set);
514 	SIGADDSET(set, signo);
515 	sigqueue_delete_set(sq, &set);
516 }
517 
518 /* Remove a set of signals for a process */
519 static void
520 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
521 {
522 	sigqueue_t worklist;
523 	struct thread *td0;
524 
525 	PROC_LOCK_ASSERT(p, MA_OWNED);
526 
527 	sigqueue_init(&worklist, NULL);
528 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
529 
530 	FOREACH_THREAD_IN_PROC(p, td0)
531 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
532 
533 	sigqueue_flush(&worklist);
534 }
535 
536 void
537 sigqueue_delete_proc(struct proc *p, int signo)
538 {
539 	sigset_t set;
540 
541 	SIGEMPTYSET(set);
542 	SIGADDSET(set, signo);
543 	sigqueue_delete_set_proc(p, &set);
544 }
545 
546 static void
547 sigqueue_delete_stopmask_proc(struct proc *p)
548 {
549 	sigset_t set;
550 
551 	SIGEMPTYSET(set);
552 	SIGADDSET(set, SIGSTOP);
553 	SIGADDSET(set, SIGTSTP);
554 	SIGADDSET(set, SIGTTIN);
555 	SIGADDSET(set, SIGTTOU);
556 	sigqueue_delete_set_proc(p, &set);
557 }
558 
559 /*
560  * Determine signal that should be delivered to thread td, the current
561  * thread, 0 if none.  If there is a pending stop signal with default
562  * action, the process stops in issignal().
563  */
564 int
565 cursig(struct thread *td)
566 {
567 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
568 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
569 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
570 	return (SIGPENDING(td) ? issignal(td) : 0);
571 }
572 
573 /*
574  * Arrange for ast() to handle unmasked pending signals on return to user
575  * mode.  This must be called whenever a signal is added to td_sigqueue or
576  * unmasked in td_sigmask.
577  */
578 void
579 signotify(struct thread *td)
580 {
581 	struct proc *p;
582 
583 	p = td->td_proc;
584 
585 	PROC_LOCK_ASSERT(p, MA_OWNED);
586 
587 	if (SIGPENDING(td)) {
588 		thread_lock(td);
589 		td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING;
590 		thread_unlock(td);
591 	}
592 }
593 
594 int
595 sigonstack(size_t sp)
596 {
597 	struct thread *td = curthread;
598 
599 	return ((td->td_pflags & TDP_ALTSTACK) ?
600 #if defined(COMPAT_43)
601 	    ((td->td_sigstk.ss_size == 0) ?
602 		(td->td_sigstk.ss_flags & SS_ONSTACK) :
603 		((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size))
604 #else
605 	    ((sp - (size_t)td->td_sigstk.ss_sp) < td->td_sigstk.ss_size)
606 #endif
607 	    : 0);
608 }
609 
610 static __inline int
611 sigprop(int sig)
612 {
613 
614 	if (sig > 0 && sig < NSIG)
615 		return (sigproptbl[_SIG_IDX(sig)]);
616 	return (0);
617 }
618 
619 int
620 sig_ffs(sigset_t *set)
621 {
622 	int i;
623 
624 	for (i = 0; i < _SIG_WORDS; i++)
625 		if (set->__bits[i])
626 			return (ffs(set->__bits[i]) + (i * 32));
627 	return (0);
628 }
629 
630 static bool
631 sigact_flag_test(const struct sigaction *act, int flag)
632 {
633 
634 	/*
635 	 * SA_SIGINFO is reset when signal disposition is set to
636 	 * ignore or default.  Other flags are kept according to user
637 	 * settings.
638 	 */
639 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
640 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
641 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
642 }
643 
644 /*
645  * kern_sigaction
646  * sigaction
647  * freebsd4_sigaction
648  * osigaction
649  */
650 int
651 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
652     struct sigaction *oact, int flags)
653 {
654 	struct sigacts *ps;
655 	struct proc *p = td->td_proc;
656 
657 	if (!_SIG_VALID(sig))
658 		return (EINVAL);
659 	if (act != NULL && act->sa_handler != SIG_DFL &&
660 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
661 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
662 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
663 		return (EINVAL);
664 
665 	PROC_LOCK(p);
666 	ps = p->p_sigacts;
667 	mtx_lock(&ps->ps_mtx);
668 	if (oact) {
669 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
670 		oact->sa_flags = 0;
671 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
672 			oact->sa_flags |= SA_ONSTACK;
673 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
674 			oact->sa_flags |= SA_RESTART;
675 		if (SIGISMEMBER(ps->ps_sigreset, sig))
676 			oact->sa_flags |= SA_RESETHAND;
677 		if (SIGISMEMBER(ps->ps_signodefer, sig))
678 			oact->sa_flags |= SA_NODEFER;
679 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
680 			oact->sa_flags |= SA_SIGINFO;
681 			oact->sa_sigaction =
682 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
683 		} else
684 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
685 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
686 			oact->sa_flags |= SA_NOCLDSTOP;
687 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
688 			oact->sa_flags |= SA_NOCLDWAIT;
689 	}
690 	if (act) {
691 		if ((sig == SIGKILL || sig == SIGSTOP) &&
692 		    act->sa_handler != SIG_DFL) {
693 			mtx_unlock(&ps->ps_mtx);
694 			PROC_UNLOCK(p);
695 			return (EINVAL);
696 		}
697 
698 		/*
699 		 * Change setting atomically.
700 		 */
701 
702 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
703 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
704 		if (sigact_flag_test(act, SA_SIGINFO)) {
705 			ps->ps_sigact[_SIG_IDX(sig)] =
706 			    (__sighandler_t *)act->sa_sigaction;
707 			SIGADDSET(ps->ps_siginfo, sig);
708 		} else {
709 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
710 			SIGDELSET(ps->ps_siginfo, sig);
711 		}
712 		if (!sigact_flag_test(act, SA_RESTART))
713 			SIGADDSET(ps->ps_sigintr, sig);
714 		else
715 			SIGDELSET(ps->ps_sigintr, sig);
716 		if (sigact_flag_test(act, SA_ONSTACK))
717 			SIGADDSET(ps->ps_sigonstack, sig);
718 		else
719 			SIGDELSET(ps->ps_sigonstack, sig);
720 		if (sigact_flag_test(act, SA_RESETHAND))
721 			SIGADDSET(ps->ps_sigreset, sig);
722 		else
723 			SIGDELSET(ps->ps_sigreset, sig);
724 		if (sigact_flag_test(act, SA_NODEFER))
725 			SIGADDSET(ps->ps_signodefer, sig);
726 		else
727 			SIGDELSET(ps->ps_signodefer, sig);
728 		if (sig == SIGCHLD) {
729 			if (act->sa_flags & SA_NOCLDSTOP)
730 				ps->ps_flag |= PS_NOCLDSTOP;
731 			else
732 				ps->ps_flag &= ~PS_NOCLDSTOP;
733 			if (act->sa_flags & SA_NOCLDWAIT) {
734 				/*
735 				 * Paranoia: since SA_NOCLDWAIT is implemented
736 				 * by reparenting the dying child to PID 1 (and
737 				 * trust it to reap the zombie), PID 1 itself
738 				 * is forbidden to set SA_NOCLDWAIT.
739 				 */
740 				if (p->p_pid == 1)
741 					ps->ps_flag &= ~PS_NOCLDWAIT;
742 				else
743 					ps->ps_flag |= PS_NOCLDWAIT;
744 			} else
745 				ps->ps_flag &= ~PS_NOCLDWAIT;
746 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
747 				ps->ps_flag |= PS_CLDSIGIGN;
748 			else
749 				ps->ps_flag &= ~PS_CLDSIGIGN;
750 		}
751 		/*
752 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
753 		 * and for signals set to SIG_DFL where the default is to
754 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
755 		 * have to restart the process.
756 		 */
757 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
758 		    (sigprop(sig) & SA_IGNORE &&
759 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
760 			/* never to be seen again */
761 			sigqueue_delete_proc(p, sig);
762 			if (sig != SIGCONT)
763 				/* easier in psignal */
764 				SIGADDSET(ps->ps_sigignore, sig);
765 			SIGDELSET(ps->ps_sigcatch, sig);
766 		} else {
767 			SIGDELSET(ps->ps_sigignore, sig);
768 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
769 				SIGDELSET(ps->ps_sigcatch, sig);
770 			else
771 				SIGADDSET(ps->ps_sigcatch, sig);
772 		}
773 #ifdef COMPAT_FREEBSD4
774 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
775 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
776 		    (flags & KSA_FREEBSD4) == 0)
777 			SIGDELSET(ps->ps_freebsd4, sig);
778 		else
779 			SIGADDSET(ps->ps_freebsd4, sig);
780 #endif
781 #ifdef COMPAT_43
782 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
783 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
784 		    (flags & KSA_OSIGSET) == 0)
785 			SIGDELSET(ps->ps_osigset, sig);
786 		else
787 			SIGADDSET(ps->ps_osigset, sig);
788 #endif
789 	}
790 	mtx_unlock(&ps->ps_mtx);
791 	PROC_UNLOCK(p);
792 	return (0);
793 }
794 
795 #ifndef _SYS_SYSPROTO_H_
796 struct sigaction_args {
797 	int	sig;
798 	struct	sigaction *act;
799 	struct	sigaction *oact;
800 };
801 #endif
802 int
803 sys_sigaction(td, uap)
804 	struct thread *td;
805 	register struct sigaction_args *uap;
806 {
807 	struct sigaction act, oact;
808 	register struct sigaction *actp, *oactp;
809 	int error;
810 
811 	actp = (uap->act != NULL) ? &act : NULL;
812 	oactp = (uap->oact != NULL) ? &oact : NULL;
813 	if (actp) {
814 		error = copyin(uap->act, actp, sizeof(act));
815 		if (error)
816 			return (error);
817 	}
818 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
819 	if (oactp && !error)
820 		error = copyout(oactp, uap->oact, sizeof(oact));
821 	return (error);
822 }
823 
824 #ifdef COMPAT_FREEBSD4
825 #ifndef _SYS_SYSPROTO_H_
826 struct freebsd4_sigaction_args {
827 	int	sig;
828 	struct	sigaction *act;
829 	struct	sigaction *oact;
830 };
831 #endif
832 int
833 freebsd4_sigaction(td, uap)
834 	struct thread *td;
835 	register struct freebsd4_sigaction_args *uap;
836 {
837 	struct sigaction act, oact;
838 	register struct sigaction *actp, *oactp;
839 	int error;
840 
841 
842 	actp = (uap->act != NULL) ? &act : NULL;
843 	oactp = (uap->oact != NULL) ? &oact : NULL;
844 	if (actp) {
845 		error = copyin(uap->act, actp, sizeof(act));
846 		if (error)
847 			return (error);
848 	}
849 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
850 	if (oactp && !error)
851 		error = copyout(oactp, uap->oact, sizeof(oact));
852 	return (error);
853 }
854 #endif	/* COMAPT_FREEBSD4 */
855 
856 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
857 #ifndef _SYS_SYSPROTO_H_
858 struct osigaction_args {
859 	int	signum;
860 	struct	osigaction *nsa;
861 	struct	osigaction *osa;
862 };
863 #endif
864 int
865 osigaction(td, uap)
866 	struct thread *td;
867 	register struct osigaction_args *uap;
868 {
869 	struct osigaction sa;
870 	struct sigaction nsa, osa;
871 	register struct sigaction *nsap, *osap;
872 	int error;
873 
874 	if (uap->signum <= 0 || uap->signum >= ONSIG)
875 		return (EINVAL);
876 
877 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
878 	osap = (uap->osa != NULL) ? &osa : NULL;
879 
880 	if (nsap) {
881 		error = copyin(uap->nsa, &sa, sizeof(sa));
882 		if (error)
883 			return (error);
884 		nsap->sa_handler = sa.sa_handler;
885 		nsap->sa_flags = sa.sa_flags;
886 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
887 	}
888 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
889 	if (osap && !error) {
890 		sa.sa_handler = osap->sa_handler;
891 		sa.sa_flags = osap->sa_flags;
892 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
893 		error = copyout(&sa, uap->osa, sizeof(sa));
894 	}
895 	return (error);
896 }
897 
898 #if !defined(__i386__)
899 /* Avoid replicating the same stub everywhere */
900 int
901 osigreturn(td, uap)
902 	struct thread *td;
903 	struct osigreturn_args *uap;
904 {
905 
906 	return (nosys(td, (struct nosys_args *)uap));
907 }
908 #endif
909 #endif /* COMPAT_43 */
910 
911 /*
912  * Initialize signal state for process 0;
913  * set to ignore signals that are ignored by default.
914  */
915 void
916 siginit(p)
917 	struct proc *p;
918 {
919 	register int i;
920 	struct sigacts *ps;
921 
922 	PROC_LOCK(p);
923 	ps = p->p_sigacts;
924 	mtx_lock(&ps->ps_mtx);
925 	for (i = 1; i <= NSIG; i++) {
926 		if (sigprop(i) & SA_IGNORE && i != SIGCONT) {
927 			SIGADDSET(ps->ps_sigignore, i);
928 		}
929 	}
930 	mtx_unlock(&ps->ps_mtx);
931 	PROC_UNLOCK(p);
932 }
933 
934 /*
935  * Reset specified signal to the default disposition.
936  */
937 static void
938 sigdflt(struct sigacts *ps, int sig)
939 {
940 
941 	mtx_assert(&ps->ps_mtx, MA_OWNED);
942 	SIGDELSET(ps->ps_sigcatch, sig);
943 	if ((sigprop(sig) & SA_IGNORE) != 0 && sig != SIGCONT)
944 		SIGADDSET(ps->ps_sigignore, sig);
945 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
946 	SIGDELSET(ps->ps_siginfo, sig);
947 }
948 
949 /*
950  * Reset signals for an exec of the specified process.
951  */
952 void
953 execsigs(struct proc *p)
954 {
955 	sigset_t osigignore;
956 	struct sigacts *ps;
957 	int sig;
958 	struct thread *td;
959 
960 	/*
961 	 * Reset caught signals.  Held signals remain held
962 	 * through td_sigmask (unless they were caught,
963 	 * and are now ignored by default).
964 	 */
965 	PROC_LOCK_ASSERT(p, MA_OWNED);
966 	td = FIRST_THREAD_IN_PROC(p);
967 	ps = p->p_sigacts;
968 	mtx_lock(&ps->ps_mtx);
969 	while (SIGNOTEMPTY(ps->ps_sigcatch)) {
970 		sig = sig_ffs(&ps->ps_sigcatch);
971 		sigdflt(ps, sig);
972 		if ((sigprop(sig) & SA_IGNORE) != 0)
973 			sigqueue_delete_proc(p, sig);
974 	}
975 
976 	/*
977 	 * As CloudABI processes cannot modify signal handlers, fully
978 	 * reset all signals to their default behavior. Do ignore
979 	 * SIGPIPE, as it would otherwise be impossible to recover from
980 	 * writes to broken pipes and sockets.
981 	 */
982 	if (SV_PROC_ABI(p) == SV_ABI_CLOUDABI) {
983 		osigignore = ps->ps_sigignore;
984 		while (SIGNOTEMPTY(osigignore)) {
985 			sig = sig_ffs(&osigignore);
986 			SIGDELSET(osigignore, sig);
987 			if (sig != SIGPIPE)
988 				sigdflt(ps, sig);
989 		}
990 		SIGADDSET(ps->ps_sigignore, SIGPIPE);
991 	}
992 
993 	/*
994 	 * Reset stack state to the user stack.
995 	 * Clear set of signals caught on the signal stack.
996 	 */
997 	td->td_sigstk.ss_flags = SS_DISABLE;
998 	td->td_sigstk.ss_size = 0;
999 	td->td_sigstk.ss_sp = 0;
1000 	td->td_pflags &= ~TDP_ALTSTACK;
1001 	/*
1002 	 * Reset no zombies if child dies flag as Solaris does.
1003 	 */
1004 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1005 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1006 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1007 	mtx_unlock(&ps->ps_mtx);
1008 }
1009 
1010 /*
1011  * kern_sigprocmask()
1012  *
1013  *	Manipulate signal mask.
1014  */
1015 int
1016 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1017     int flags)
1018 {
1019 	sigset_t new_block, oset1;
1020 	struct proc *p;
1021 	int error;
1022 
1023 	p = td->td_proc;
1024 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1025 		PROC_LOCK_ASSERT(p, MA_OWNED);
1026 	else
1027 		PROC_LOCK(p);
1028 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1029 	    ? MA_OWNED : MA_NOTOWNED);
1030 	if (oset != NULL)
1031 		*oset = td->td_sigmask;
1032 
1033 	error = 0;
1034 	if (set != NULL) {
1035 		switch (how) {
1036 		case SIG_BLOCK:
1037 			SIG_CANTMASK(*set);
1038 			oset1 = td->td_sigmask;
1039 			SIGSETOR(td->td_sigmask, *set);
1040 			new_block = td->td_sigmask;
1041 			SIGSETNAND(new_block, oset1);
1042 			break;
1043 		case SIG_UNBLOCK:
1044 			SIGSETNAND(td->td_sigmask, *set);
1045 			signotify(td);
1046 			goto out;
1047 		case SIG_SETMASK:
1048 			SIG_CANTMASK(*set);
1049 			oset1 = td->td_sigmask;
1050 			if (flags & SIGPROCMASK_OLD)
1051 				SIGSETLO(td->td_sigmask, *set);
1052 			else
1053 				td->td_sigmask = *set;
1054 			new_block = td->td_sigmask;
1055 			SIGSETNAND(new_block, oset1);
1056 			signotify(td);
1057 			break;
1058 		default:
1059 			error = EINVAL;
1060 			goto out;
1061 		}
1062 
1063 		/*
1064 		 * The new_block set contains signals that were not previously
1065 		 * blocked, but are blocked now.
1066 		 *
1067 		 * In case we block any signal that was not previously blocked
1068 		 * for td, and process has the signal pending, try to schedule
1069 		 * signal delivery to some thread that does not block the
1070 		 * signal, possibly waking it up.
1071 		 */
1072 		if (p->p_numthreads != 1)
1073 			reschedule_signals(p, new_block, flags);
1074 	}
1075 
1076 out:
1077 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1078 		PROC_UNLOCK(p);
1079 	return (error);
1080 }
1081 
1082 #ifndef _SYS_SYSPROTO_H_
1083 struct sigprocmask_args {
1084 	int	how;
1085 	const sigset_t *set;
1086 	sigset_t *oset;
1087 };
1088 #endif
1089 int
1090 sys_sigprocmask(td, uap)
1091 	register struct thread *td;
1092 	struct sigprocmask_args *uap;
1093 {
1094 	sigset_t set, oset;
1095 	sigset_t *setp, *osetp;
1096 	int error;
1097 
1098 	setp = (uap->set != NULL) ? &set : NULL;
1099 	osetp = (uap->oset != NULL) ? &oset : NULL;
1100 	if (setp) {
1101 		error = copyin(uap->set, setp, sizeof(set));
1102 		if (error)
1103 			return (error);
1104 	}
1105 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1106 	if (osetp && !error) {
1107 		error = copyout(osetp, uap->oset, sizeof(oset));
1108 	}
1109 	return (error);
1110 }
1111 
1112 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1113 #ifndef _SYS_SYSPROTO_H_
1114 struct osigprocmask_args {
1115 	int	how;
1116 	osigset_t mask;
1117 };
1118 #endif
1119 int
1120 osigprocmask(td, uap)
1121 	register struct thread *td;
1122 	struct osigprocmask_args *uap;
1123 {
1124 	sigset_t set, oset;
1125 	int error;
1126 
1127 	OSIG2SIG(uap->mask, set);
1128 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1129 	SIG2OSIG(oset, td->td_retval[0]);
1130 	return (error);
1131 }
1132 #endif /* COMPAT_43 */
1133 
1134 int
1135 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1136 {
1137 	ksiginfo_t ksi;
1138 	sigset_t set;
1139 	int error;
1140 
1141 	error = copyin(uap->set, &set, sizeof(set));
1142 	if (error) {
1143 		td->td_retval[0] = error;
1144 		return (0);
1145 	}
1146 
1147 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1148 	if (error) {
1149 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1150 			error = ERESTART;
1151 		if (error == ERESTART)
1152 			return (error);
1153 		td->td_retval[0] = error;
1154 		return (0);
1155 	}
1156 
1157 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1158 	td->td_retval[0] = error;
1159 	return (0);
1160 }
1161 
1162 int
1163 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1164 {
1165 	struct timespec ts;
1166 	struct timespec *timeout;
1167 	sigset_t set;
1168 	ksiginfo_t ksi;
1169 	int error;
1170 
1171 	if (uap->timeout) {
1172 		error = copyin(uap->timeout, &ts, sizeof(ts));
1173 		if (error)
1174 			return (error);
1175 
1176 		timeout = &ts;
1177 	} else
1178 		timeout = NULL;
1179 
1180 	error = copyin(uap->set, &set, sizeof(set));
1181 	if (error)
1182 		return (error);
1183 
1184 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1185 	if (error)
1186 		return (error);
1187 
1188 	if (uap->info)
1189 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1190 
1191 	if (error == 0)
1192 		td->td_retval[0] = ksi.ksi_signo;
1193 	return (error);
1194 }
1195 
1196 int
1197 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1198 {
1199 	ksiginfo_t ksi;
1200 	sigset_t set;
1201 	int error;
1202 
1203 	error = copyin(uap->set, &set, sizeof(set));
1204 	if (error)
1205 		return (error);
1206 
1207 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1208 	if (error)
1209 		return (error);
1210 
1211 	if (uap->info)
1212 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1213 
1214 	if (error == 0)
1215 		td->td_retval[0] = ksi.ksi_signo;
1216 	return (error);
1217 }
1218 
1219 int
1220 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1221 	struct timespec *timeout)
1222 {
1223 	struct sigacts *ps;
1224 	sigset_t saved_mask, new_block;
1225 	struct proc *p;
1226 	int error, sig, timo, timevalid = 0;
1227 	struct timespec rts, ets, ts;
1228 	struct timeval tv;
1229 
1230 	p = td->td_proc;
1231 	error = 0;
1232 	ets.tv_sec = 0;
1233 	ets.tv_nsec = 0;
1234 
1235 	if (timeout != NULL) {
1236 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1237 			timevalid = 1;
1238 			getnanouptime(&rts);
1239 			ets = rts;
1240 			timespecadd(&ets, timeout);
1241 		}
1242 	}
1243 	ksiginfo_init(ksi);
1244 	/* Some signals can not be waited for. */
1245 	SIG_CANTMASK(waitset);
1246 	ps = p->p_sigacts;
1247 	PROC_LOCK(p);
1248 	saved_mask = td->td_sigmask;
1249 	SIGSETNAND(td->td_sigmask, waitset);
1250 	for (;;) {
1251 		mtx_lock(&ps->ps_mtx);
1252 		sig = cursig(td);
1253 		mtx_unlock(&ps->ps_mtx);
1254 		KASSERT(sig >= 0, ("sig %d", sig));
1255 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1256 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1257 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1258 				error = 0;
1259 				break;
1260 			}
1261 		}
1262 
1263 		if (error != 0)
1264 			break;
1265 
1266 		/*
1267 		 * POSIX says this must be checked after looking for pending
1268 		 * signals.
1269 		 */
1270 		if (timeout != NULL) {
1271 			if (!timevalid) {
1272 				error = EINVAL;
1273 				break;
1274 			}
1275 			getnanouptime(&rts);
1276 			if (timespeccmp(&rts, &ets, >=)) {
1277 				error = EAGAIN;
1278 				break;
1279 			}
1280 			ts = ets;
1281 			timespecsub(&ts, &rts);
1282 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1283 			timo = tvtohz(&tv);
1284 		} else {
1285 			timo = 0;
1286 		}
1287 
1288 		error = msleep(ps, &p->p_mtx, PPAUSE|PCATCH, "sigwait", timo);
1289 
1290 		if (timeout != NULL) {
1291 			if (error == ERESTART) {
1292 				/* Timeout can not be restarted. */
1293 				error = EINTR;
1294 			} else if (error == EAGAIN) {
1295 				/* We will calculate timeout by ourself. */
1296 				error = 0;
1297 			}
1298 		}
1299 	}
1300 
1301 	new_block = saved_mask;
1302 	SIGSETNAND(new_block, td->td_sigmask);
1303 	td->td_sigmask = saved_mask;
1304 	/*
1305 	 * Fewer signals can be delivered to us, reschedule signal
1306 	 * notification.
1307 	 */
1308 	if (p->p_numthreads != 1)
1309 		reschedule_signals(p, new_block, 0);
1310 
1311 	if (error == 0) {
1312 		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1313 
1314 		if (ksi->ksi_code == SI_TIMER)
1315 			itimer_accept(p, ksi->ksi_timerid, ksi);
1316 
1317 #ifdef KTRACE
1318 		if (KTRPOINT(td, KTR_PSIG)) {
1319 			sig_t action;
1320 
1321 			mtx_lock(&ps->ps_mtx);
1322 			action = ps->ps_sigact[_SIG_IDX(sig)];
1323 			mtx_unlock(&ps->ps_mtx);
1324 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1325 		}
1326 #endif
1327 		if (sig == SIGKILL)
1328 			sigexit(td, sig);
1329 	}
1330 	PROC_UNLOCK(p);
1331 	return (error);
1332 }
1333 
1334 #ifndef _SYS_SYSPROTO_H_
1335 struct sigpending_args {
1336 	sigset_t	*set;
1337 };
1338 #endif
1339 int
1340 sys_sigpending(td, uap)
1341 	struct thread *td;
1342 	struct sigpending_args *uap;
1343 {
1344 	struct proc *p = td->td_proc;
1345 	sigset_t pending;
1346 
1347 	PROC_LOCK(p);
1348 	pending = p->p_sigqueue.sq_signals;
1349 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1350 	PROC_UNLOCK(p);
1351 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1352 }
1353 
1354 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1355 #ifndef _SYS_SYSPROTO_H_
1356 struct osigpending_args {
1357 	int	dummy;
1358 };
1359 #endif
1360 int
1361 osigpending(td, uap)
1362 	struct thread *td;
1363 	struct osigpending_args *uap;
1364 {
1365 	struct proc *p = td->td_proc;
1366 	sigset_t pending;
1367 
1368 	PROC_LOCK(p);
1369 	pending = p->p_sigqueue.sq_signals;
1370 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1371 	PROC_UNLOCK(p);
1372 	SIG2OSIG(pending, td->td_retval[0]);
1373 	return (0);
1374 }
1375 #endif /* COMPAT_43 */
1376 
1377 #if defined(COMPAT_43)
1378 /*
1379  * Generalized interface signal handler, 4.3-compatible.
1380  */
1381 #ifndef _SYS_SYSPROTO_H_
1382 struct osigvec_args {
1383 	int	signum;
1384 	struct	sigvec *nsv;
1385 	struct	sigvec *osv;
1386 };
1387 #endif
1388 /* ARGSUSED */
1389 int
1390 osigvec(td, uap)
1391 	struct thread *td;
1392 	register struct osigvec_args *uap;
1393 {
1394 	struct sigvec vec;
1395 	struct sigaction nsa, osa;
1396 	register struct sigaction *nsap, *osap;
1397 	int error;
1398 
1399 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1400 		return (EINVAL);
1401 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1402 	osap = (uap->osv != NULL) ? &osa : NULL;
1403 	if (nsap) {
1404 		error = copyin(uap->nsv, &vec, sizeof(vec));
1405 		if (error)
1406 			return (error);
1407 		nsap->sa_handler = vec.sv_handler;
1408 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1409 		nsap->sa_flags = vec.sv_flags;
1410 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1411 	}
1412 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1413 	if (osap && !error) {
1414 		vec.sv_handler = osap->sa_handler;
1415 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1416 		vec.sv_flags = osap->sa_flags;
1417 		vec.sv_flags &= ~SA_NOCLDWAIT;
1418 		vec.sv_flags ^= SA_RESTART;
1419 		error = copyout(&vec, uap->osv, sizeof(vec));
1420 	}
1421 	return (error);
1422 }
1423 
1424 #ifndef _SYS_SYSPROTO_H_
1425 struct osigblock_args {
1426 	int	mask;
1427 };
1428 #endif
1429 int
1430 osigblock(td, uap)
1431 	register struct thread *td;
1432 	struct osigblock_args *uap;
1433 {
1434 	sigset_t set, oset;
1435 
1436 	OSIG2SIG(uap->mask, set);
1437 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1438 	SIG2OSIG(oset, td->td_retval[0]);
1439 	return (0);
1440 }
1441 
1442 #ifndef _SYS_SYSPROTO_H_
1443 struct osigsetmask_args {
1444 	int	mask;
1445 };
1446 #endif
1447 int
1448 osigsetmask(td, uap)
1449 	struct thread *td;
1450 	struct osigsetmask_args *uap;
1451 {
1452 	sigset_t set, oset;
1453 
1454 	OSIG2SIG(uap->mask, set);
1455 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1456 	SIG2OSIG(oset, td->td_retval[0]);
1457 	return (0);
1458 }
1459 #endif /* COMPAT_43 */
1460 
1461 /*
1462  * Suspend calling thread until signal, providing mask to be set in the
1463  * meantime.
1464  */
1465 #ifndef _SYS_SYSPROTO_H_
1466 struct sigsuspend_args {
1467 	const sigset_t *sigmask;
1468 };
1469 #endif
1470 /* ARGSUSED */
1471 int
1472 sys_sigsuspend(td, uap)
1473 	struct thread *td;
1474 	struct sigsuspend_args *uap;
1475 {
1476 	sigset_t mask;
1477 	int error;
1478 
1479 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1480 	if (error)
1481 		return (error);
1482 	return (kern_sigsuspend(td, mask));
1483 }
1484 
1485 int
1486 kern_sigsuspend(struct thread *td, sigset_t mask)
1487 {
1488 	struct proc *p = td->td_proc;
1489 	int has_sig, sig;
1490 
1491 	/*
1492 	 * When returning from sigsuspend, we want
1493 	 * the old mask to be restored after the
1494 	 * signal handler has finished.  Thus, we
1495 	 * save it here and mark the sigacts structure
1496 	 * to indicate this.
1497 	 */
1498 	PROC_LOCK(p);
1499 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1500 	    SIGPROCMASK_PROC_LOCKED);
1501 	td->td_pflags |= TDP_OLDMASK;
1502 
1503 	/*
1504 	 * Process signals now. Otherwise, we can get spurious wakeup
1505 	 * due to signal entered process queue, but delivered to other
1506 	 * thread. But sigsuspend should return only on signal
1507 	 * delivery.
1508 	 */
1509 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1510 	for (has_sig = 0; !has_sig;) {
1511 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1512 			0) == 0)
1513 			/* void */;
1514 		thread_suspend_check(0);
1515 		mtx_lock(&p->p_sigacts->ps_mtx);
1516 		while ((sig = cursig(td)) != 0) {
1517 			KASSERT(sig >= 0, ("sig %d", sig));
1518 			has_sig += postsig(sig);
1519 		}
1520 		mtx_unlock(&p->p_sigacts->ps_mtx);
1521 	}
1522 	PROC_UNLOCK(p);
1523 	td->td_errno = EINTR;
1524 	td->td_pflags |= TDP_NERRNO;
1525 	return (EJUSTRETURN);
1526 }
1527 
1528 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1529 /*
1530  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1531  * convention: libc stub passes mask, not pointer, to save a copyin.
1532  */
1533 #ifndef _SYS_SYSPROTO_H_
1534 struct osigsuspend_args {
1535 	osigset_t mask;
1536 };
1537 #endif
1538 /* ARGSUSED */
1539 int
1540 osigsuspend(td, uap)
1541 	struct thread *td;
1542 	struct osigsuspend_args *uap;
1543 {
1544 	sigset_t mask;
1545 
1546 	OSIG2SIG(uap->mask, mask);
1547 	return (kern_sigsuspend(td, mask));
1548 }
1549 #endif /* COMPAT_43 */
1550 
1551 #if defined(COMPAT_43)
1552 #ifndef _SYS_SYSPROTO_H_
1553 struct osigstack_args {
1554 	struct	sigstack *nss;
1555 	struct	sigstack *oss;
1556 };
1557 #endif
1558 /* ARGSUSED */
1559 int
1560 osigstack(td, uap)
1561 	struct thread *td;
1562 	register struct osigstack_args *uap;
1563 {
1564 	struct sigstack nss, oss;
1565 	int error = 0;
1566 
1567 	if (uap->nss != NULL) {
1568 		error = copyin(uap->nss, &nss, sizeof(nss));
1569 		if (error)
1570 			return (error);
1571 	}
1572 	oss.ss_sp = td->td_sigstk.ss_sp;
1573 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1574 	if (uap->nss != NULL) {
1575 		td->td_sigstk.ss_sp = nss.ss_sp;
1576 		td->td_sigstk.ss_size = 0;
1577 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1578 		td->td_pflags |= TDP_ALTSTACK;
1579 	}
1580 	if (uap->oss != NULL)
1581 		error = copyout(&oss, uap->oss, sizeof(oss));
1582 
1583 	return (error);
1584 }
1585 #endif /* COMPAT_43 */
1586 
1587 #ifndef _SYS_SYSPROTO_H_
1588 struct sigaltstack_args {
1589 	stack_t	*ss;
1590 	stack_t	*oss;
1591 };
1592 #endif
1593 /* ARGSUSED */
1594 int
1595 sys_sigaltstack(td, uap)
1596 	struct thread *td;
1597 	register struct sigaltstack_args *uap;
1598 {
1599 	stack_t ss, oss;
1600 	int error;
1601 
1602 	if (uap->ss != NULL) {
1603 		error = copyin(uap->ss, &ss, sizeof(ss));
1604 		if (error)
1605 			return (error);
1606 	}
1607 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1608 	    (uap->oss != NULL) ? &oss : NULL);
1609 	if (error)
1610 		return (error);
1611 	if (uap->oss != NULL)
1612 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1613 	return (error);
1614 }
1615 
1616 int
1617 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1618 {
1619 	struct proc *p = td->td_proc;
1620 	int oonstack;
1621 
1622 	oonstack = sigonstack(cpu_getstack(td));
1623 
1624 	if (oss != NULL) {
1625 		*oss = td->td_sigstk;
1626 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1627 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1628 	}
1629 
1630 	if (ss != NULL) {
1631 		if (oonstack)
1632 			return (EPERM);
1633 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1634 			return (EINVAL);
1635 		if (!(ss->ss_flags & SS_DISABLE)) {
1636 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1637 				return (ENOMEM);
1638 
1639 			td->td_sigstk = *ss;
1640 			td->td_pflags |= TDP_ALTSTACK;
1641 		} else {
1642 			td->td_pflags &= ~TDP_ALTSTACK;
1643 		}
1644 	}
1645 	return (0);
1646 }
1647 
1648 /*
1649  * Common code for kill process group/broadcast kill.
1650  * cp is calling process.
1651  */
1652 static int
1653 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1654 {
1655 	struct proc *p;
1656 	struct pgrp *pgrp;
1657 	int err;
1658 	int ret;
1659 
1660 	ret = ESRCH;
1661 	if (all) {
1662 		/*
1663 		 * broadcast
1664 		 */
1665 		sx_slock(&allproc_lock);
1666 		FOREACH_PROC_IN_SYSTEM(p) {
1667 			PROC_LOCK(p);
1668 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1669 			    p == td->td_proc || p->p_state == PRS_NEW) {
1670 				PROC_UNLOCK(p);
1671 				continue;
1672 			}
1673 			err = p_cansignal(td, p, sig);
1674 			if (err == 0) {
1675 				if (sig)
1676 					pksignal(p, sig, ksi);
1677 				ret = err;
1678 			}
1679 			else if (ret == ESRCH)
1680 				ret = err;
1681 			PROC_UNLOCK(p);
1682 		}
1683 		sx_sunlock(&allproc_lock);
1684 	} else {
1685 		sx_slock(&proctree_lock);
1686 		if (pgid == 0) {
1687 			/*
1688 			 * zero pgid means send to my process group.
1689 			 */
1690 			pgrp = td->td_proc->p_pgrp;
1691 			PGRP_LOCK(pgrp);
1692 		} else {
1693 			pgrp = pgfind(pgid);
1694 			if (pgrp == NULL) {
1695 				sx_sunlock(&proctree_lock);
1696 				return (ESRCH);
1697 			}
1698 		}
1699 		sx_sunlock(&proctree_lock);
1700 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1701 			PROC_LOCK(p);
1702 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1703 			    p->p_state == PRS_NEW) {
1704 				PROC_UNLOCK(p);
1705 				continue;
1706 			}
1707 			err = p_cansignal(td, p, sig);
1708 			if (err == 0) {
1709 				if (sig)
1710 					pksignal(p, sig, ksi);
1711 				ret = err;
1712 			}
1713 			else if (ret == ESRCH)
1714 				ret = err;
1715 			PROC_UNLOCK(p);
1716 		}
1717 		PGRP_UNLOCK(pgrp);
1718 	}
1719 	return (ret);
1720 }
1721 
1722 #ifndef _SYS_SYSPROTO_H_
1723 struct kill_args {
1724 	int	pid;
1725 	int	signum;
1726 };
1727 #endif
1728 /* ARGSUSED */
1729 int
1730 sys_kill(struct thread *td, struct kill_args *uap)
1731 {
1732 	ksiginfo_t ksi;
1733 	struct proc *p;
1734 	int error;
1735 
1736 	/*
1737 	 * A process in capability mode can send signals only to himself.
1738 	 * The main rationale behind this is that abort(3) is implemented as
1739 	 * kill(getpid(), SIGABRT).
1740 	 */
1741 	if (IN_CAPABILITY_MODE(td) && uap->pid != td->td_proc->p_pid)
1742 		return (ECAPMODE);
1743 
1744 	AUDIT_ARG_SIGNUM(uap->signum);
1745 	AUDIT_ARG_PID(uap->pid);
1746 	if ((u_int)uap->signum > _SIG_MAXSIG)
1747 		return (EINVAL);
1748 
1749 	ksiginfo_init(&ksi);
1750 	ksi.ksi_signo = uap->signum;
1751 	ksi.ksi_code = SI_USER;
1752 	ksi.ksi_pid = td->td_proc->p_pid;
1753 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1754 
1755 	if (uap->pid > 0) {
1756 		/* kill single process */
1757 		if ((p = pfind(uap->pid)) == NULL) {
1758 			if ((p = zpfind(uap->pid)) == NULL)
1759 				return (ESRCH);
1760 		}
1761 		AUDIT_ARG_PROCESS(p);
1762 		error = p_cansignal(td, p, uap->signum);
1763 		if (error == 0 && uap->signum)
1764 			pksignal(p, uap->signum, &ksi);
1765 		PROC_UNLOCK(p);
1766 		return (error);
1767 	}
1768 	switch (uap->pid) {
1769 	case -1:		/* broadcast signal */
1770 		return (killpg1(td, uap->signum, 0, 1, &ksi));
1771 	case 0:			/* signal own process group */
1772 		return (killpg1(td, uap->signum, 0, 0, &ksi));
1773 	default:		/* negative explicit process group */
1774 		return (killpg1(td, uap->signum, -uap->pid, 0, &ksi));
1775 	}
1776 	/* NOTREACHED */
1777 }
1778 
1779 int
1780 sys_pdkill(td, uap)
1781 	struct thread *td;
1782 	struct pdkill_args *uap;
1783 {
1784 	struct proc *p;
1785 	cap_rights_t rights;
1786 	int error;
1787 
1788 	AUDIT_ARG_SIGNUM(uap->signum);
1789 	AUDIT_ARG_FD(uap->fd);
1790 	if ((u_int)uap->signum > _SIG_MAXSIG)
1791 		return (EINVAL);
1792 
1793 	error = procdesc_find(td, uap->fd,
1794 	    cap_rights_init(&rights, CAP_PDKILL), &p);
1795 	if (error)
1796 		return (error);
1797 	AUDIT_ARG_PROCESS(p);
1798 	error = p_cansignal(td, p, uap->signum);
1799 	if (error == 0 && uap->signum)
1800 		kern_psignal(p, uap->signum);
1801 	PROC_UNLOCK(p);
1802 	return (error);
1803 }
1804 
1805 #if defined(COMPAT_43)
1806 #ifndef _SYS_SYSPROTO_H_
1807 struct okillpg_args {
1808 	int	pgid;
1809 	int	signum;
1810 };
1811 #endif
1812 /* ARGSUSED */
1813 int
1814 okillpg(struct thread *td, struct okillpg_args *uap)
1815 {
1816 	ksiginfo_t ksi;
1817 
1818 	AUDIT_ARG_SIGNUM(uap->signum);
1819 	AUDIT_ARG_PID(uap->pgid);
1820 	if ((u_int)uap->signum > _SIG_MAXSIG)
1821 		return (EINVAL);
1822 
1823 	ksiginfo_init(&ksi);
1824 	ksi.ksi_signo = uap->signum;
1825 	ksi.ksi_code = SI_USER;
1826 	ksi.ksi_pid = td->td_proc->p_pid;
1827 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1828 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1829 }
1830 #endif /* COMPAT_43 */
1831 
1832 #ifndef _SYS_SYSPROTO_H_
1833 struct sigqueue_args {
1834 	pid_t pid;
1835 	int signum;
1836 	/* union sigval */ void *value;
1837 };
1838 #endif
1839 int
1840 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1841 {
1842 	ksiginfo_t ksi;
1843 	struct proc *p;
1844 	int error;
1845 
1846 	if ((u_int)uap->signum > _SIG_MAXSIG)
1847 		return (EINVAL);
1848 
1849 	/*
1850 	 * Specification says sigqueue can only send signal to
1851 	 * single process.
1852 	 */
1853 	if (uap->pid <= 0)
1854 		return (EINVAL);
1855 
1856 	if ((p = pfind(uap->pid)) == NULL) {
1857 		if ((p = zpfind(uap->pid)) == NULL)
1858 			return (ESRCH);
1859 	}
1860 	error = p_cansignal(td, p, uap->signum);
1861 	if (error == 0 && uap->signum != 0) {
1862 		ksiginfo_init(&ksi);
1863 		ksi.ksi_flags = KSI_SIGQ;
1864 		ksi.ksi_signo = uap->signum;
1865 		ksi.ksi_code = SI_QUEUE;
1866 		ksi.ksi_pid = td->td_proc->p_pid;
1867 		ksi.ksi_uid = td->td_ucred->cr_ruid;
1868 		ksi.ksi_value.sival_ptr = uap->value;
1869 		error = pksignal(p, ksi.ksi_signo, &ksi);
1870 	}
1871 	PROC_UNLOCK(p);
1872 	return (error);
1873 }
1874 
1875 /*
1876  * Send a signal to a process group.
1877  */
1878 void
1879 gsignal(int pgid, int sig, ksiginfo_t *ksi)
1880 {
1881 	struct pgrp *pgrp;
1882 
1883 	if (pgid != 0) {
1884 		sx_slock(&proctree_lock);
1885 		pgrp = pgfind(pgid);
1886 		sx_sunlock(&proctree_lock);
1887 		if (pgrp != NULL) {
1888 			pgsignal(pgrp, sig, 0, ksi);
1889 			PGRP_UNLOCK(pgrp);
1890 		}
1891 	}
1892 }
1893 
1894 /*
1895  * Send a signal to a process group.  If checktty is 1,
1896  * limit to members which have a controlling terminal.
1897  */
1898 void
1899 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
1900 {
1901 	struct proc *p;
1902 
1903 	if (pgrp) {
1904 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
1905 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1906 			PROC_LOCK(p);
1907 			if (p->p_state == PRS_NORMAL &&
1908 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
1909 				pksignal(p, sig, ksi);
1910 			PROC_UNLOCK(p);
1911 		}
1912 	}
1913 }
1914 
1915 
1916 /*
1917  * Recalculate the signal mask and reset the signal disposition after
1918  * usermode frame for delivery is formed.  Should be called after
1919  * mach-specific routine, because sysent->sv_sendsig() needs correct
1920  * ps_siginfo and signal mask.
1921  */
1922 static void
1923 postsig_done(int sig, struct thread *td, struct sigacts *ps)
1924 {
1925 	sigset_t mask;
1926 
1927 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1928 	td->td_ru.ru_nsignals++;
1929 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
1930 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
1931 		SIGADDSET(mask, sig);
1932 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
1933 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
1934 	if (SIGISMEMBER(ps->ps_sigreset, sig))
1935 		sigdflt(ps, sig);
1936 }
1937 
1938 
1939 /*
1940  * Send a signal caused by a trap to the current thread.  If it will be
1941  * caught immediately, deliver it with correct code.  Otherwise, post it
1942  * normally.
1943  */
1944 void
1945 trapsignal(struct thread *td, ksiginfo_t *ksi)
1946 {
1947 	struct sigacts *ps;
1948 	struct proc *p;
1949 	int sig;
1950 	int code;
1951 
1952 	p = td->td_proc;
1953 	sig = ksi->ksi_signo;
1954 	code = ksi->ksi_code;
1955 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
1956 
1957 	PROC_LOCK(p);
1958 	ps = p->p_sigacts;
1959 	mtx_lock(&ps->ps_mtx);
1960 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
1961 	    !SIGISMEMBER(td->td_sigmask, sig)) {
1962 #ifdef KTRACE
1963 		if (KTRPOINT(curthread, KTR_PSIG))
1964 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
1965 			    &td->td_sigmask, code);
1966 #endif
1967 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
1968 				ksi, &td->td_sigmask);
1969 		postsig_done(sig, td, ps);
1970 		mtx_unlock(&ps->ps_mtx);
1971 	} else {
1972 		/*
1973 		 * Avoid a possible infinite loop if the thread
1974 		 * masking the signal or process is ignoring the
1975 		 * signal.
1976 		 */
1977 		if (kern_forcesigexit &&
1978 		    (SIGISMEMBER(td->td_sigmask, sig) ||
1979 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
1980 			SIGDELSET(td->td_sigmask, sig);
1981 			SIGDELSET(ps->ps_sigcatch, sig);
1982 			SIGDELSET(ps->ps_sigignore, sig);
1983 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1984 		}
1985 		mtx_unlock(&ps->ps_mtx);
1986 		p->p_code = code;	/* XXX for core dump/debugger */
1987 		p->p_sig = sig;		/* XXX to verify code */
1988 		tdsendsignal(p, td, sig, ksi);
1989 	}
1990 	PROC_UNLOCK(p);
1991 }
1992 
1993 static struct thread *
1994 sigtd(struct proc *p, int sig, int prop)
1995 {
1996 	struct thread *td, *signal_td;
1997 
1998 	PROC_LOCK_ASSERT(p, MA_OWNED);
1999 
2000 	/*
2001 	 * Check if current thread can handle the signal without
2002 	 * switching context to another thread.
2003 	 */
2004 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig))
2005 		return (curthread);
2006 	signal_td = NULL;
2007 	FOREACH_THREAD_IN_PROC(p, td) {
2008 		if (!SIGISMEMBER(td->td_sigmask, sig)) {
2009 			signal_td = td;
2010 			break;
2011 		}
2012 	}
2013 	if (signal_td == NULL)
2014 		signal_td = FIRST_THREAD_IN_PROC(p);
2015 	return (signal_td);
2016 }
2017 
2018 /*
2019  * Send the signal to the process.  If the signal has an action, the action
2020  * is usually performed by the target process rather than the caller; we add
2021  * the signal to the set of pending signals for the process.
2022  *
2023  * Exceptions:
2024  *   o When a stop signal is sent to a sleeping process that takes the
2025  *     default action, the process is stopped without awakening it.
2026  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2027  *     regardless of the signal action (eg, blocked or ignored).
2028  *
2029  * Other ignored signals are discarded immediately.
2030  *
2031  * NB: This function may be entered from the debugger via the "kill" DDB
2032  * command.  There is little that can be done to mitigate the possibly messy
2033  * side effects of this unwise possibility.
2034  */
2035 void
2036 kern_psignal(struct proc *p, int sig)
2037 {
2038 	ksiginfo_t ksi;
2039 
2040 	ksiginfo_init(&ksi);
2041 	ksi.ksi_signo = sig;
2042 	ksi.ksi_code = SI_KERNEL;
2043 	(void) tdsendsignal(p, NULL, sig, &ksi);
2044 }
2045 
2046 int
2047 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2048 {
2049 
2050 	return (tdsendsignal(p, NULL, sig, ksi));
2051 }
2052 
2053 /* Utility function for finding a thread to send signal event to. */
2054 int
2055 sigev_findtd(struct proc *p ,struct sigevent *sigev, struct thread **ttd)
2056 {
2057 	struct thread *td;
2058 
2059 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2060 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2061 		if (td == NULL)
2062 			return (ESRCH);
2063 		*ttd = td;
2064 	} else {
2065 		*ttd = NULL;
2066 		PROC_LOCK(p);
2067 	}
2068 	return (0);
2069 }
2070 
2071 void
2072 tdsignal(struct thread *td, int sig)
2073 {
2074 	ksiginfo_t ksi;
2075 
2076 	ksiginfo_init(&ksi);
2077 	ksi.ksi_signo = sig;
2078 	ksi.ksi_code = SI_KERNEL;
2079 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2080 }
2081 
2082 void
2083 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2084 {
2085 
2086 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2087 }
2088 
2089 int
2090 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2091 {
2092 	sig_t action;
2093 	sigqueue_t *sigqueue;
2094 	int prop;
2095 	struct sigacts *ps;
2096 	int intrval;
2097 	int ret = 0;
2098 	int wakeup_swapper;
2099 
2100 	MPASS(td == NULL || p == td->td_proc);
2101 	PROC_LOCK_ASSERT(p, MA_OWNED);
2102 
2103 	if (!_SIG_VALID(sig))
2104 		panic("%s(): invalid signal %d", __func__, sig);
2105 
2106 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2107 
2108 	/*
2109 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2110 	 */
2111 	if (p->p_state == PRS_ZOMBIE) {
2112 		if (ksi && (ksi->ksi_flags & KSI_INS))
2113 			ksiginfo_tryfree(ksi);
2114 		return (ret);
2115 	}
2116 
2117 	ps = p->p_sigacts;
2118 	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2119 	prop = sigprop(sig);
2120 
2121 	if (td == NULL) {
2122 		td = sigtd(p, sig, prop);
2123 		sigqueue = &p->p_sigqueue;
2124 	} else
2125 		sigqueue = &td->td_sigqueue;
2126 
2127 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2128 
2129 	/*
2130 	 * If the signal is being ignored,
2131 	 * then we forget about it immediately.
2132 	 * (Note: we don't set SIGCONT in ps_sigignore,
2133 	 * and if it is set to SIG_IGN,
2134 	 * action will be SIG_DFL here.)
2135 	 */
2136 	mtx_lock(&ps->ps_mtx);
2137 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2138 		SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2139 
2140 		mtx_unlock(&ps->ps_mtx);
2141 		if (ksi && (ksi->ksi_flags & KSI_INS))
2142 			ksiginfo_tryfree(ksi);
2143 		return (ret);
2144 	}
2145 	if (SIGISMEMBER(td->td_sigmask, sig))
2146 		action = SIG_HOLD;
2147 	else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2148 		action = SIG_CATCH;
2149 	else
2150 		action = SIG_DFL;
2151 	if (SIGISMEMBER(ps->ps_sigintr, sig))
2152 		intrval = EINTR;
2153 	else
2154 		intrval = ERESTART;
2155 	mtx_unlock(&ps->ps_mtx);
2156 
2157 	if (prop & SA_CONT)
2158 		sigqueue_delete_stopmask_proc(p);
2159 	else if (prop & SA_STOP) {
2160 		/*
2161 		 * If sending a tty stop signal to a member of an orphaned
2162 		 * process group, discard the signal here if the action
2163 		 * is default; don't stop the process below if sleeping,
2164 		 * and don't clear any pending SIGCONT.
2165 		 */
2166 		if ((prop & SA_TTYSTOP) &&
2167 		    (p->p_pgrp->pg_jobc == 0) &&
2168 		    (action == SIG_DFL)) {
2169 			if (ksi && (ksi->ksi_flags & KSI_INS))
2170 				ksiginfo_tryfree(ksi);
2171 			return (ret);
2172 		}
2173 		sigqueue_delete_proc(p, SIGCONT);
2174 		if (p->p_flag & P_CONTINUED) {
2175 			p->p_flag &= ~P_CONTINUED;
2176 			PROC_LOCK(p->p_pptr);
2177 			sigqueue_take(p->p_ksi);
2178 			PROC_UNLOCK(p->p_pptr);
2179 		}
2180 	}
2181 
2182 	ret = sigqueue_add(sigqueue, sig, ksi);
2183 	if (ret != 0)
2184 		return (ret);
2185 	signotify(td);
2186 	/*
2187 	 * Defer further processing for signals which are held,
2188 	 * except that stopped processes must be continued by SIGCONT.
2189 	 */
2190 	if (action == SIG_HOLD &&
2191 	    !((prop & SA_CONT) && (p->p_flag & P_STOPPED_SIG)))
2192 		return (ret);
2193 	/*
2194 	 * SIGKILL: Remove procfs STOPEVENTs and ptrace events.
2195 	 */
2196 	if (sig == SIGKILL) {
2197 		p->p_ptevents = 0;
2198 		/* from procfs_ioctl.c: PIOCBIC */
2199 		p->p_stops = 0;
2200 		/* from procfs_ioctl.c: PIOCCONT */
2201 		p->p_step = 0;
2202 		wakeup(&p->p_step);
2203 	}
2204 	/*
2205 	 * Some signals have a process-wide effect and a per-thread
2206 	 * component.  Most processing occurs when the process next
2207 	 * tries to cross the user boundary, however there are some
2208 	 * times when processing needs to be done immediately, such as
2209 	 * waking up threads so that they can cross the user boundary.
2210 	 * We try to do the per-process part here.
2211 	 */
2212 	if (P_SHOULDSTOP(p)) {
2213 		KASSERT(!(p->p_flag & P_WEXIT),
2214 		    ("signal to stopped but exiting process"));
2215 		if (sig == SIGKILL) {
2216 			/*
2217 			 * If traced process is already stopped,
2218 			 * then no further action is necessary.
2219 			 */
2220 			if (p->p_flag & P_TRACED)
2221 				goto out;
2222 			/*
2223 			 * SIGKILL sets process running.
2224 			 * It will die elsewhere.
2225 			 * All threads must be restarted.
2226 			 */
2227 			p->p_flag &= ~P_STOPPED_SIG;
2228 			goto runfast;
2229 		}
2230 
2231 		if (prop & SA_CONT) {
2232 			/*
2233 			 * If traced process is already stopped,
2234 			 * then no further action is necessary.
2235 			 */
2236 			if (p->p_flag & P_TRACED)
2237 				goto out;
2238 			/*
2239 			 * If SIGCONT is default (or ignored), we continue the
2240 			 * process but don't leave the signal in sigqueue as
2241 			 * it has no further action.  If SIGCONT is held, we
2242 			 * continue the process and leave the signal in
2243 			 * sigqueue.  If the process catches SIGCONT, let it
2244 			 * handle the signal itself.  If it isn't waiting on
2245 			 * an event, it goes back to run state.
2246 			 * Otherwise, process goes back to sleep state.
2247 			 */
2248 			p->p_flag &= ~P_STOPPED_SIG;
2249 			PROC_SLOCK(p);
2250 			if (p->p_numthreads == p->p_suspcount) {
2251 				PROC_SUNLOCK(p);
2252 				p->p_flag |= P_CONTINUED;
2253 				p->p_xsig = SIGCONT;
2254 				PROC_LOCK(p->p_pptr);
2255 				childproc_continued(p);
2256 				PROC_UNLOCK(p->p_pptr);
2257 				PROC_SLOCK(p);
2258 			}
2259 			if (action == SIG_DFL) {
2260 				thread_unsuspend(p);
2261 				PROC_SUNLOCK(p);
2262 				sigqueue_delete(sigqueue, sig);
2263 				goto out;
2264 			}
2265 			if (action == SIG_CATCH) {
2266 				/*
2267 				 * The process wants to catch it so it needs
2268 				 * to run at least one thread, but which one?
2269 				 */
2270 				PROC_SUNLOCK(p);
2271 				goto runfast;
2272 			}
2273 			/*
2274 			 * The signal is not ignored or caught.
2275 			 */
2276 			thread_unsuspend(p);
2277 			PROC_SUNLOCK(p);
2278 			goto out;
2279 		}
2280 
2281 		if (prop & SA_STOP) {
2282 			/*
2283 			 * If traced process is already stopped,
2284 			 * then no further action is necessary.
2285 			 */
2286 			if (p->p_flag & P_TRACED)
2287 				goto out;
2288 			/*
2289 			 * Already stopped, don't need to stop again
2290 			 * (If we did the shell could get confused).
2291 			 * Just make sure the signal STOP bit set.
2292 			 */
2293 			p->p_flag |= P_STOPPED_SIG;
2294 			sigqueue_delete(sigqueue, sig);
2295 			goto out;
2296 		}
2297 
2298 		/*
2299 		 * All other kinds of signals:
2300 		 * If a thread is sleeping interruptibly, simulate a
2301 		 * wakeup so that when it is continued it will be made
2302 		 * runnable and can look at the signal.  However, don't make
2303 		 * the PROCESS runnable, leave it stopped.
2304 		 * It may run a bit until it hits a thread_suspend_check().
2305 		 */
2306 		wakeup_swapper = 0;
2307 		PROC_SLOCK(p);
2308 		thread_lock(td);
2309 		if (TD_ON_SLEEPQ(td) && (td->td_flags & TDF_SINTR))
2310 			wakeup_swapper = sleepq_abort(td, intrval);
2311 		thread_unlock(td);
2312 		PROC_SUNLOCK(p);
2313 		if (wakeup_swapper)
2314 			kick_proc0();
2315 		goto out;
2316 		/*
2317 		 * Mutexes are short lived. Threads waiting on them will
2318 		 * hit thread_suspend_check() soon.
2319 		 */
2320 	} else if (p->p_state == PRS_NORMAL) {
2321 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2322 			tdsigwakeup(td, sig, action, intrval);
2323 			goto out;
2324 		}
2325 
2326 		MPASS(action == SIG_DFL);
2327 
2328 		if (prop & SA_STOP) {
2329 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2330 				goto out;
2331 			p->p_flag |= P_STOPPED_SIG;
2332 			p->p_xsig = sig;
2333 			PROC_SLOCK(p);
2334 			wakeup_swapper = sig_suspend_threads(td, p, 1);
2335 			if (p->p_numthreads == p->p_suspcount) {
2336 				/*
2337 				 * only thread sending signal to another
2338 				 * process can reach here, if thread is sending
2339 				 * signal to its process, because thread does
2340 				 * not suspend itself here, p_numthreads
2341 				 * should never be equal to p_suspcount.
2342 				 */
2343 				thread_stopped(p);
2344 				PROC_SUNLOCK(p);
2345 				sigqueue_delete_proc(p, p->p_xsig);
2346 			} else
2347 				PROC_SUNLOCK(p);
2348 			if (wakeup_swapper)
2349 				kick_proc0();
2350 			goto out;
2351 		}
2352 	} else {
2353 		/* Not in "NORMAL" state. discard the signal. */
2354 		sigqueue_delete(sigqueue, sig);
2355 		goto out;
2356 	}
2357 
2358 	/*
2359 	 * The process is not stopped so we need to apply the signal to all the
2360 	 * running threads.
2361 	 */
2362 runfast:
2363 	tdsigwakeup(td, sig, action, intrval);
2364 	PROC_SLOCK(p);
2365 	thread_unsuspend(p);
2366 	PROC_SUNLOCK(p);
2367 out:
2368 	/* If we jump here, proc slock should not be owned. */
2369 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2370 	return (ret);
2371 }
2372 
2373 /*
2374  * The force of a signal has been directed against a single
2375  * thread.  We need to see what we can do about knocking it
2376  * out of any sleep it may be in etc.
2377  */
2378 static void
2379 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2380 {
2381 	struct proc *p = td->td_proc;
2382 	register int prop;
2383 	int wakeup_swapper;
2384 
2385 	wakeup_swapper = 0;
2386 	PROC_LOCK_ASSERT(p, MA_OWNED);
2387 	prop = sigprop(sig);
2388 
2389 	PROC_SLOCK(p);
2390 	thread_lock(td);
2391 	/*
2392 	 * Bring the priority of a thread up if we want it to get
2393 	 * killed in this lifetime.  Be careful to avoid bumping the
2394 	 * priority of the idle thread, since we still allow to signal
2395 	 * kernel processes.
2396 	 */
2397 	if (action == SIG_DFL && (prop & SA_KILL) != 0 &&
2398 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2399 		sched_prio(td, PUSER);
2400 	if (TD_ON_SLEEPQ(td)) {
2401 		/*
2402 		 * If thread is sleeping uninterruptibly
2403 		 * we can't interrupt the sleep... the signal will
2404 		 * be noticed when the process returns through
2405 		 * trap() or syscall().
2406 		 */
2407 		if ((td->td_flags & TDF_SINTR) == 0)
2408 			goto out;
2409 		/*
2410 		 * If SIGCONT is default (or ignored) and process is
2411 		 * asleep, we are finished; the process should not
2412 		 * be awakened.
2413 		 */
2414 		if ((prop & SA_CONT) && action == SIG_DFL) {
2415 			thread_unlock(td);
2416 			PROC_SUNLOCK(p);
2417 			sigqueue_delete(&p->p_sigqueue, sig);
2418 			/*
2419 			 * It may be on either list in this state.
2420 			 * Remove from both for now.
2421 			 */
2422 			sigqueue_delete(&td->td_sigqueue, sig);
2423 			return;
2424 		}
2425 
2426 		/*
2427 		 * Don't awaken a sleeping thread for SIGSTOP if the
2428 		 * STOP signal is deferred.
2429 		 */
2430 		if ((prop & SA_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2431 		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2432 			goto out;
2433 
2434 		/*
2435 		 * Give low priority threads a better chance to run.
2436 		 */
2437 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2438 			sched_prio(td, PUSER);
2439 
2440 		wakeup_swapper = sleepq_abort(td, intrval);
2441 	} else {
2442 		/*
2443 		 * Other states do nothing with the signal immediately,
2444 		 * other than kicking ourselves if we are running.
2445 		 * It will either never be noticed, or noticed very soon.
2446 		 */
2447 #ifdef SMP
2448 		if (TD_IS_RUNNING(td) && td != curthread)
2449 			forward_signal(td);
2450 #endif
2451 	}
2452 out:
2453 	PROC_SUNLOCK(p);
2454 	thread_unlock(td);
2455 	if (wakeup_swapper)
2456 		kick_proc0();
2457 }
2458 
2459 static int
2460 sig_suspend_threads(struct thread *td, struct proc *p, int sending)
2461 {
2462 	struct thread *td2;
2463 	int wakeup_swapper;
2464 
2465 	PROC_LOCK_ASSERT(p, MA_OWNED);
2466 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2467 
2468 	wakeup_swapper = 0;
2469 	FOREACH_THREAD_IN_PROC(p, td2) {
2470 		thread_lock(td2);
2471 		td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
2472 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2473 		    (td2->td_flags & TDF_SINTR)) {
2474 			if (td2->td_flags & TDF_SBDRY) {
2475 				/*
2476 				 * Once a thread is asleep with
2477 				 * TDF_SBDRY and without TDF_SERESTART
2478 				 * or TDF_SEINTR set, it should never
2479 				 * become suspended due to this check.
2480 				 */
2481 				KASSERT(!TD_IS_SUSPENDED(td2),
2482 				    ("thread with deferred stops suspended"));
2483 				if (TD_SBDRY_INTR(td2) && sending) {
2484 					wakeup_swapper |= sleepq_abort(td2,
2485 					    TD_SBDRY_ERRNO(td2));
2486 				}
2487 			} else if (!TD_IS_SUSPENDED(td2)) {
2488 				thread_suspend_one(td2);
2489 			}
2490 		} else if (!TD_IS_SUSPENDED(td2)) {
2491 			if (sending || td != td2)
2492 				td2->td_flags |= TDF_ASTPENDING;
2493 #ifdef SMP
2494 			if (TD_IS_RUNNING(td2) && td2 != td)
2495 				forward_signal(td2);
2496 #endif
2497 		}
2498 		thread_unlock(td2);
2499 	}
2500 	return (wakeup_swapper);
2501 }
2502 
2503 int
2504 ptracestop(struct thread *td, int sig)
2505 {
2506 	struct proc *p = td->td_proc;
2507 
2508 	PROC_LOCK_ASSERT(p, MA_OWNED);
2509 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2510 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2511 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2512 
2513 	td->td_dbgflags |= TDB_XSIG;
2514 	td->td_xsig = sig;
2515 	CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2516 	    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2517 	PROC_SLOCK(p);
2518 	while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2519 		if (p->p_flag & P_SINGLE_EXIT &&
2520 		    !(td->td_dbgflags & TDB_EXIT)) {
2521 			/*
2522 			 * Ignore ptrace stops except for thread exit
2523 			 * events when the process exits.
2524 			 */
2525 			td->td_dbgflags &= ~TDB_XSIG;
2526 			PROC_SUNLOCK(p);
2527 			return (sig);
2528 		}
2529 
2530 		/*
2531 		 * Make wait(2) work.  Ensure that right after the
2532 		 * attach, the thread which was decided to become the
2533 		 * leader of attach gets reported to the waiter.
2534 		 * Otherwise, just avoid overwriting another thread's
2535 		 * assignment to p_xthread.  If another thread has
2536 		 * already set p_xthread, the current thread will get
2537 		 * a chance to report itself upon the next iteration.
2538 		 */
2539 		if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2540 		    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2541 		    p->p_xthread == NULL)) {
2542 			p->p_xsig = sig;
2543 			p->p_xthread = td;
2544 			td->td_dbgflags &= ~TDB_FSTP;
2545 			p->p_flag2 &= ~P2_PTRACE_FSTP;
2546 			p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2547 			sig_suspend_threads(td, p, 0);
2548 		}
2549 		if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2550 			td->td_dbgflags &= ~TDB_STOPATFORK;
2551 			cv_broadcast(&p->p_dbgwait);
2552 		}
2553 stopme:
2554 		thread_suspend_switch(td, p);
2555 		if (p->p_xthread == td)
2556 			p->p_xthread = NULL;
2557 		if (!(p->p_flag & P_TRACED))
2558 			break;
2559 		if (td->td_dbgflags & TDB_SUSPEND) {
2560 			if (p->p_flag & P_SINGLE_EXIT)
2561 				break;
2562 			goto stopme;
2563 		}
2564 	}
2565 	PROC_SUNLOCK(p);
2566 	return (td->td_xsig);
2567 }
2568 
2569 static void
2570 reschedule_signals(struct proc *p, sigset_t block, int flags)
2571 {
2572 	struct sigacts *ps;
2573 	struct thread *td;
2574 	int sig;
2575 
2576 	PROC_LOCK_ASSERT(p, MA_OWNED);
2577 	ps = p->p_sigacts;
2578 	mtx_assert(&ps->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0 ?
2579 	    MA_OWNED : MA_NOTOWNED);
2580 	if (SIGISEMPTY(p->p_siglist))
2581 		return;
2582 	SIGSETAND(block, p->p_siglist);
2583 	while ((sig = sig_ffs(&block)) != 0) {
2584 		SIGDELSET(block, sig);
2585 		td = sigtd(p, sig, 0);
2586 		signotify(td);
2587 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2588 			mtx_lock(&ps->ps_mtx);
2589 		if (p->p_flag & P_TRACED || SIGISMEMBER(ps->ps_sigcatch, sig))
2590 			tdsigwakeup(td, sig, SIG_CATCH,
2591 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2592 			     ERESTART));
2593 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2594 			mtx_unlock(&ps->ps_mtx);
2595 	}
2596 }
2597 
2598 void
2599 tdsigcleanup(struct thread *td)
2600 {
2601 	struct proc *p;
2602 	sigset_t unblocked;
2603 
2604 	p = td->td_proc;
2605 	PROC_LOCK_ASSERT(p, MA_OWNED);
2606 
2607 	sigqueue_flush(&td->td_sigqueue);
2608 	if (p->p_numthreads == 1)
2609 		return;
2610 
2611 	/*
2612 	 * Since we cannot handle signals, notify signal post code
2613 	 * about this by filling the sigmask.
2614 	 *
2615 	 * Also, if needed, wake up thread(s) that do not block the
2616 	 * same signals as the exiting thread, since the thread might
2617 	 * have been selected for delivery and woken up.
2618 	 */
2619 	SIGFILLSET(unblocked);
2620 	SIGSETNAND(unblocked, td->td_sigmask);
2621 	SIGFILLSET(td->td_sigmask);
2622 	reschedule_signals(p, unblocked, 0);
2623 
2624 }
2625 
2626 static int
2627 sigdeferstop_curr_flags(int cflags)
2628 {
2629 
2630 	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
2631 	    (cflags & TDF_SBDRY) != 0);
2632 	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
2633 }
2634 
2635 /*
2636  * Defer the delivery of SIGSTOP for the current thread, according to
2637  * the requested mode.  Returns previous flags, which must be restored
2638  * by sigallowstop().
2639  *
2640  * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
2641  * cleared by the current thread, which allow the lock-less read-only
2642  * accesses below.
2643  */
2644 int
2645 sigdeferstop_impl(int mode)
2646 {
2647 	struct thread *td;
2648 	int cflags, nflags;
2649 
2650 	td = curthread;
2651 	cflags = sigdeferstop_curr_flags(td->td_flags);
2652 	switch (mode) {
2653 	case SIGDEFERSTOP_NOP:
2654 		nflags = cflags;
2655 		break;
2656 	case SIGDEFERSTOP_OFF:
2657 		nflags = 0;
2658 		break;
2659 	case SIGDEFERSTOP_SILENT:
2660 		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
2661 		break;
2662 	case SIGDEFERSTOP_EINTR:
2663 		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
2664 		break;
2665 	case SIGDEFERSTOP_ERESTART:
2666 		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
2667 		break;
2668 	default:
2669 		panic("sigdeferstop: invalid mode %x", mode);
2670 		break;
2671 	}
2672 	if (cflags == nflags)
2673 		return (SIGDEFERSTOP_VAL_NCHG);
2674 	thread_lock(td);
2675 	td->td_flags = (td->td_flags & ~cflags) | nflags;
2676 	thread_unlock(td);
2677 	return (cflags);
2678 }
2679 
2680 /*
2681  * Restores the STOP handling mode, typically permitting the delivery
2682  * of SIGSTOP for the current thread.  This does not immediately
2683  * suspend if a stop was posted.  Instead, the thread will suspend
2684  * either via ast() or a subsequent interruptible sleep.
2685  */
2686 void
2687 sigallowstop_impl(int prev)
2688 {
2689 	struct thread *td;
2690 	int cflags;
2691 
2692 	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
2693 	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
2694 	    ("sigallowstop: incorrect previous mode %x", prev));
2695 	td = curthread;
2696 	cflags = sigdeferstop_curr_flags(td->td_flags);
2697 	if (cflags != prev) {
2698 		thread_lock(td);
2699 		td->td_flags = (td->td_flags & ~cflags) | prev;
2700 		thread_unlock(td);
2701 	}
2702 }
2703 
2704 /*
2705  * If the current process has received a signal (should be caught or cause
2706  * termination, should interrupt current syscall), return the signal number.
2707  * Stop signals with default action are processed immediately, then cleared;
2708  * they aren't returned.  This is checked after each entry to the system for
2709  * a syscall or trap (though this can usually be done without calling issignal
2710  * by checking the pending signal masks in cursig.) The normal call
2711  * sequence is
2712  *
2713  *	while (sig = cursig(curthread))
2714  *		postsig(sig);
2715  */
2716 static int
2717 issignal(struct thread *td)
2718 {
2719 	struct proc *p;
2720 	struct sigacts *ps;
2721 	struct sigqueue *queue;
2722 	sigset_t sigpending;
2723 	int sig, prop, newsig;
2724 
2725 	p = td->td_proc;
2726 	ps = p->p_sigacts;
2727 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2728 	PROC_LOCK_ASSERT(p, MA_OWNED);
2729 	for (;;) {
2730 		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
2731 
2732 		sigpending = td->td_sigqueue.sq_signals;
2733 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
2734 		SIGSETNAND(sigpending, td->td_sigmask);
2735 
2736 		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
2737 		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2738 			SIG_STOPSIGMASK(sigpending);
2739 		if (SIGISEMPTY(sigpending))	/* no signal to send */
2740 			return (0);
2741 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
2742 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
2743 		    SIGISMEMBER(sigpending, SIGSTOP)) {
2744 			/*
2745 			 * If debugger just attached, always consume
2746 			 * SIGSTOP from ptrace(PT_ATTACH) first, to
2747 			 * execute the debugger attach ritual in
2748 			 * order.
2749 			 */
2750 			sig = SIGSTOP;
2751 			td->td_dbgflags |= TDB_FSTP;
2752 		} else {
2753 			sig = sig_ffs(&sigpending);
2754 		}
2755 
2756 		if (p->p_stops & S_SIG) {
2757 			mtx_unlock(&ps->ps_mtx);
2758 			stopevent(p, S_SIG, sig);
2759 			mtx_lock(&ps->ps_mtx);
2760 		}
2761 
2762 		/*
2763 		 * We should see pending but ignored signals
2764 		 * only if P_TRACED was on when they were posted.
2765 		 */
2766 		if (SIGISMEMBER(ps->ps_sigignore, sig) && (traced == 0)) {
2767 			sigqueue_delete(&td->td_sigqueue, sig);
2768 			sigqueue_delete(&p->p_sigqueue, sig);
2769 			continue;
2770 		}
2771 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
2772 			/*
2773 			 * If traced, always stop.
2774 			 * Remove old signal from queue before the stop.
2775 			 * XXX shrug off debugger, it causes siginfo to
2776 			 * be thrown away.
2777 			 */
2778 			queue = &td->td_sigqueue;
2779 			td->td_dbgksi.ksi_signo = 0;
2780 			if (sigqueue_get(queue, sig, &td->td_dbgksi) == 0) {
2781 				queue = &p->p_sigqueue;
2782 				sigqueue_get(queue, sig, &td->td_dbgksi);
2783 			}
2784 
2785 			mtx_unlock(&ps->ps_mtx);
2786 			newsig = ptracestop(td, sig);
2787 			mtx_lock(&ps->ps_mtx);
2788 
2789 			if (sig != newsig) {
2790 
2791 				/*
2792 				 * If parent wants us to take the signal,
2793 				 * then it will leave it in p->p_xsig;
2794 				 * otherwise we just look for signals again.
2795 				*/
2796 				if (newsig == 0)
2797 					continue;
2798 				sig = newsig;
2799 
2800 				/*
2801 				 * Put the new signal into td_sigqueue. If the
2802 				 * signal is being masked, look for other
2803 				 * signals.
2804 				 */
2805 				sigqueue_add(queue, sig, NULL);
2806 				if (SIGISMEMBER(td->td_sigmask, sig))
2807 					continue;
2808 				signotify(td);
2809 			} else {
2810 				if (td->td_dbgksi.ksi_signo != 0) {
2811 					td->td_dbgksi.ksi_flags |= KSI_HEAD;
2812 					if (sigqueue_add(&td->td_sigqueue, sig,
2813 					    &td->td_dbgksi) != 0)
2814 						td->td_dbgksi.ksi_signo = 0;
2815 				}
2816 				if (td->td_dbgksi.ksi_signo == 0)
2817 					sigqueue_add(&td->td_sigqueue, sig,
2818 					    NULL);
2819 			}
2820 
2821 			/*
2822 			 * If the traced bit got turned off, go back up
2823 			 * to the top to rescan signals.  This ensures
2824 			 * that p_sig* and p_sigact are consistent.
2825 			 */
2826 			if ((p->p_flag & P_TRACED) == 0)
2827 				continue;
2828 		}
2829 
2830 		prop = sigprop(sig);
2831 
2832 		/*
2833 		 * Decide whether the signal should be returned.
2834 		 * Return the signal's number, or fall through
2835 		 * to clear it from the pending mask.
2836 		 */
2837 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2838 
2839 		case (intptr_t)SIG_DFL:
2840 			/*
2841 			 * Don't take default actions on system processes.
2842 			 */
2843 			if (p->p_pid <= 1) {
2844 #ifdef DIAGNOSTIC
2845 				/*
2846 				 * Are you sure you want to ignore SIGSEGV
2847 				 * in init? XXX
2848 				 */
2849 				printf("Process (pid %lu) got signal %d\n",
2850 					(u_long)p->p_pid, sig);
2851 #endif
2852 				break;		/* == ignore */
2853 			}
2854 			/*
2855 			 * If there is a pending stop signal to process
2856 			 * with default action, stop here,
2857 			 * then clear the signal.  However,
2858 			 * if process is member of an orphaned
2859 			 * process group, ignore tty stop signals.
2860 			 */
2861 			if (prop & SA_STOP) {
2862 				if (p->p_flag & (P_TRACED|P_WEXIT) ||
2863 				    (p->p_pgrp->pg_jobc == 0 &&
2864 				     prop & SA_TTYSTOP))
2865 					break;	/* == ignore */
2866 				if (TD_SBDRY_INTR(td)) {
2867 					KASSERT((td->td_flags & TDF_SBDRY) != 0,
2868 					    ("lost TDF_SBDRY"));
2869 					return (-1);
2870 				}
2871 				mtx_unlock(&ps->ps_mtx);
2872 				WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2873 				    &p->p_mtx.lock_object, "Catching SIGSTOP");
2874 				sigqueue_delete(&td->td_sigqueue, sig);
2875 				sigqueue_delete(&p->p_sigqueue, sig);
2876 				p->p_flag |= P_STOPPED_SIG;
2877 				p->p_xsig = sig;
2878 				PROC_SLOCK(p);
2879 				sig_suspend_threads(td, p, 0);
2880 				thread_suspend_switch(td, p);
2881 				PROC_SUNLOCK(p);
2882 				mtx_lock(&ps->ps_mtx);
2883 				goto next;
2884 			} else if (prop & SA_IGNORE) {
2885 				/*
2886 				 * Except for SIGCONT, shouldn't get here.
2887 				 * Default action is to ignore; drop it.
2888 				 */
2889 				break;		/* == ignore */
2890 			} else
2891 				return (sig);
2892 			/*NOTREACHED*/
2893 
2894 		case (intptr_t)SIG_IGN:
2895 			/*
2896 			 * Masking above should prevent us ever trying
2897 			 * to take action on an ignored signal other
2898 			 * than SIGCONT, unless process is traced.
2899 			 */
2900 			if ((prop & SA_CONT) == 0 &&
2901 			    (p->p_flag & P_TRACED) == 0)
2902 				printf("issignal\n");
2903 			break;		/* == ignore */
2904 
2905 		default:
2906 			/*
2907 			 * This signal has an action, let
2908 			 * postsig() process it.
2909 			 */
2910 			return (sig);
2911 		}
2912 		sigqueue_delete(&td->td_sigqueue, sig);	/* take the signal! */
2913 		sigqueue_delete(&p->p_sigqueue, sig);
2914 next:;
2915 	}
2916 	/* NOTREACHED */
2917 }
2918 
2919 void
2920 thread_stopped(struct proc *p)
2921 {
2922 	int n;
2923 
2924 	PROC_LOCK_ASSERT(p, MA_OWNED);
2925 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2926 	n = p->p_suspcount;
2927 	if (p == curproc)
2928 		n++;
2929 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
2930 		PROC_SUNLOCK(p);
2931 		p->p_flag &= ~P_WAITED;
2932 		PROC_LOCK(p->p_pptr);
2933 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
2934 			CLD_TRAPPED : CLD_STOPPED);
2935 		PROC_UNLOCK(p->p_pptr);
2936 		PROC_SLOCK(p);
2937 	}
2938 }
2939 
2940 /*
2941  * Take the action for the specified signal
2942  * from the current set of pending signals.
2943  */
2944 int
2945 postsig(sig)
2946 	register int sig;
2947 {
2948 	struct thread *td = curthread;
2949 	register struct proc *p = td->td_proc;
2950 	struct sigacts *ps;
2951 	sig_t action;
2952 	ksiginfo_t ksi;
2953 	sigset_t returnmask;
2954 
2955 	KASSERT(sig != 0, ("postsig"));
2956 
2957 	PROC_LOCK_ASSERT(p, MA_OWNED);
2958 	ps = p->p_sigacts;
2959 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2960 	ksiginfo_init(&ksi);
2961 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
2962 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
2963 		return (0);
2964 	ksi.ksi_signo = sig;
2965 	if (ksi.ksi_code == SI_TIMER)
2966 		itimer_accept(p, ksi.ksi_timerid, &ksi);
2967 	action = ps->ps_sigact[_SIG_IDX(sig)];
2968 #ifdef KTRACE
2969 	if (KTRPOINT(td, KTR_PSIG))
2970 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
2971 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
2972 #endif
2973 	if (p->p_stops & S_SIG) {
2974 		mtx_unlock(&ps->ps_mtx);
2975 		stopevent(p, S_SIG, sig);
2976 		mtx_lock(&ps->ps_mtx);
2977 	}
2978 
2979 	if (action == SIG_DFL) {
2980 		/*
2981 		 * Default action, where the default is to kill
2982 		 * the process.  (Other cases were ignored above.)
2983 		 */
2984 		mtx_unlock(&ps->ps_mtx);
2985 		sigexit(td, sig);
2986 		/* NOTREACHED */
2987 	} else {
2988 		/*
2989 		 * If we get here, the signal must be caught.
2990 		 */
2991 		KASSERT(action != SIG_IGN && !SIGISMEMBER(td->td_sigmask, sig),
2992 		    ("postsig action"));
2993 		/*
2994 		 * Set the new mask value and also defer further
2995 		 * occurrences of this signal.
2996 		 *
2997 		 * Special case: user has done a sigsuspend.  Here the
2998 		 * current mask is not of interest, but rather the
2999 		 * mask from before the sigsuspend is what we want
3000 		 * restored after the signal processing is completed.
3001 		 */
3002 		if (td->td_pflags & TDP_OLDMASK) {
3003 			returnmask = td->td_oldsigmask;
3004 			td->td_pflags &= ~TDP_OLDMASK;
3005 		} else
3006 			returnmask = td->td_sigmask;
3007 
3008 		if (p->p_sig == sig) {
3009 			p->p_code = 0;
3010 			p->p_sig = 0;
3011 		}
3012 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3013 		postsig_done(sig, td, ps);
3014 	}
3015 	return (1);
3016 }
3017 
3018 /*
3019  * Kill the current process for stated reason.
3020  */
3021 void
3022 killproc(p, why)
3023 	struct proc *p;
3024 	char *why;
3025 {
3026 
3027 	PROC_LOCK_ASSERT(p, MA_OWNED);
3028 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3029 	    p->p_comm);
3030 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n", p->p_pid,
3031 	    p->p_comm, p->p_ucred ? p->p_ucred->cr_uid : -1, why);
3032 	p->p_flag |= P_WKILLED;
3033 	kern_psignal(p, SIGKILL);
3034 }
3035 
3036 /*
3037  * Force the current process to exit with the specified signal, dumping core
3038  * if appropriate.  We bypass the normal tests for masked and caught signals,
3039  * allowing unrecoverable failures to terminate the process without changing
3040  * signal state.  Mark the accounting record with the signal termination.
3041  * If dumping core, save the signal number for the debugger.  Calls exit and
3042  * does not return.
3043  */
3044 void
3045 sigexit(td, sig)
3046 	struct thread *td;
3047 	int sig;
3048 {
3049 	struct proc *p = td->td_proc;
3050 
3051 	PROC_LOCK_ASSERT(p, MA_OWNED);
3052 	p->p_acflag |= AXSIG;
3053 	/*
3054 	 * We must be single-threading to generate a core dump.  This
3055 	 * ensures that the registers in the core file are up-to-date.
3056 	 * Also, the ELF dump handler assumes that the thread list doesn't
3057 	 * change out from under it.
3058 	 *
3059 	 * XXX If another thread attempts to single-thread before us
3060 	 *     (e.g. via fork()), we won't get a dump at all.
3061 	 */
3062 	if ((sigprop(sig) & SA_CORE) && thread_single(p, SINGLE_NO_EXIT) == 0) {
3063 		p->p_sig = sig;
3064 		/*
3065 		 * Log signals which would cause core dumps
3066 		 * (Log as LOG_INFO to appease those who don't want
3067 		 * these messages.)
3068 		 * XXX : Todo, as well as euid, write out ruid too
3069 		 * Note that coredump() drops proc lock.
3070 		 */
3071 		if (coredump(td) == 0)
3072 			sig |= WCOREFLAG;
3073 		if (kern_logsigexit)
3074 			log(LOG_INFO,
3075 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
3076 			    p->p_pid, p->p_comm,
3077 			    td->td_ucred ? td->td_ucred->cr_uid : -1,
3078 			    sig &~ WCOREFLAG,
3079 			    sig & WCOREFLAG ? " (core dumped)" : "");
3080 	} else
3081 		PROC_UNLOCK(p);
3082 	exit1(td, 0, sig);
3083 	/* NOTREACHED */
3084 }
3085 
3086 /*
3087  * Send queued SIGCHLD to parent when child process's state
3088  * is changed.
3089  */
3090 static void
3091 sigparent(struct proc *p, int reason, int status)
3092 {
3093 	PROC_LOCK_ASSERT(p, MA_OWNED);
3094 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3095 
3096 	if (p->p_ksi != NULL) {
3097 		p->p_ksi->ksi_signo  = SIGCHLD;
3098 		p->p_ksi->ksi_code   = reason;
3099 		p->p_ksi->ksi_status = status;
3100 		p->p_ksi->ksi_pid    = p->p_pid;
3101 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3102 		if (KSI_ONQ(p->p_ksi))
3103 			return;
3104 	}
3105 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3106 }
3107 
3108 static void
3109 childproc_jobstate(struct proc *p, int reason, int sig)
3110 {
3111 	struct sigacts *ps;
3112 
3113 	PROC_LOCK_ASSERT(p, MA_OWNED);
3114 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3115 
3116 	/*
3117 	 * Wake up parent sleeping in kern_wait(), also send
3118 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3119 	 * that parent will awake, because parent may masked
3120 	 * the signal.
3121 	 */
3122 	p->p_pptr->p_flag |= P_STATCHILD;
3123 	wakeup(p->p_pptr);
3124 
3125 	ps = p->p_pptr->p_sigacts;
3126 	mtx_lock(&ps->ps_mtx);
3127 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3128 		mtx_unlock(&ps->ps_mtx);
3129 		sigparent(p, reason, sig);
3130 	} else
3131 		mtx_unlock(&ps->ps_mtx);
3132 }
3133 
3134 void
3135 childproc_stopped(struct proc *p, int reason)
3136 {
3137 
3138 	childproc_jobstate(p, reason, p->p_xsig);
3139 }
3140 
3141 void
3142 childproc_continued(struct proc *p)
3143 {
3144 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3145 }
3146 
3147 void
3148 childproc_exited(struct proc *p)
3149 {
3150 	int reason, status;
3151 
3152 	if (WCOREDUMP(p->p_xsig)) {
3153 		reason = CLD_DUMPED;
3154 		status = WTERMSIG(p->p_xsig);
3155 	} else if (WIFSIGNALED(p->p_xsig)) {
3156 		reason = CLD_KILLED;
3157 		status = WTERMSIG(p->p_xsig);
3158 	} else {
3159 		reason = CLD_EXITED;
3160 		status = p->p_xexit;
3161 	}
3162 	/*
3163 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3164 	 * done in exit1().
3165 	 */
3166 	sigparent(p, reason, status);
3167 }
3168 
3169 /*
3170  * We only have 1 character for the core count in the format
3171  * string, so the range will be 0-9
3172  */
3173 #define	MAX_NUM_CORE_FILES 10
3174 #ifndef NUM_CORE_FILES
3175 #define	NUM_CORE_FILES 5
3176 #endif
3177 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3178 static int num_cores = NUM_CORE_FILES;
3179 
3180 static int
3181 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3182 {
3183 	int error;
3184 	int new_val;
3185 
3186 	new_val = num_cores;
3187 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3188 	if (error != 0 || req->newptr == NULL)
3189 		return (error);
3190 	if (new_val > MAX_NUM_CORE_FILES)
3191 		new_val = MAX_NUM_CORE_FILES;
3192 	if (new_val < 0)
3193 		new_val = 0;
3194 	num_cores = new_val;
3195 	return (0);
3196 }
3197 SYSCTL_PROC(_debug, OID_AUTO, ncores, CTLTYPE_INT|CTLFLAG_RW,
3198 	    0, sizeof(int), sysctl_debug_num_cores_check, "I", "");
3199 
3200 #define	GZ_SUFFIX	".gz"
3201 
3202 #ifdef GZIO
3203 static int compress_user_cores = 1;
3204 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores, CTLFLAG_RWTUN,
3205     &compress_user_cores, 0, "Compression of user corefiles");
3206 
3207 int compress_user_cores_gzlevel = 6;
3208 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_gzlevel, CTLFLAG_RWTUN,
3209     &compress_user_cores_gzlevel, 0, "Corefile gzip compression level");
3210 #else
3211 static int compress_user_cores = 0;
3212 #endif
3213 
3214 /*
3215  * Protect the access to corefilename[] by allproc_lock.
3216  */
3217 #define	corefilename_lock	allproc_lock
3218 
3219 static char corefilename[MAXPATHLEN] = {"%N.core"};
3220 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3221 
3222 static int
3223 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3224 {
3225 	int error;
3226 
3227 	sx_xlock(&corefilename_lock);
3228 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3229 	    req);
3230 	sx_xunlock(&corefilename_lock);
3231 
3232 	return (error);
3233 }
3234 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3235     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3236     "Process corefile name format string");
3237 
3238 /*
3239  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3240  * Expand the name described in corefilename, using name, uid, and pid
3241  * and open/create core file.
3242  * corefilename is a printf-like string, with three format specifiers:
3243  *	%N	name of process ("name")
3244  *	%P	process id (pid)
3245  *	%U	user id (uid)
3246  * For example, "%N.core" is the default; they can be disabled completely
3247  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3248  * This is controlled by the sysctl variable kern.corefile (see above).
3249  */
3250 static int
3251 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3252     int compress, struct vnode **vpp, char **namep)
3253 {
3254 	struct nameidata nd;
3255 	struct sbuf sb;
3256 	const char *format;
3257 	char *hostname, *name;
3258 	int indexpos, i, error, cmode, flags, oflags;
3259 
3260 	hostname = NULL;
3261 	format = corefilename;
3262 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3263 	indexpos = -1;
3264 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3265 	sx_slock(&corefilename_lock);
3266 	for (i = 0; format[i] != '\0'; i++) {
3267 		switch (format[i]) {
3268 		case '%':	/* Format character */
3269 			i++;
3270 			switch (format[i]) {
3271 			case '%':
3272 				sbuf_putc(&sb, '%');
3273 				break;
3274 			case 'H':	/* hostname */
3275 				if (hostname == NULL) {
3276 					hostname = malloc(MAXHOSTNAMELEN,
3277 					    M_TEMP, M_WAITOK);
3278 				}
3279 				getcredhostname(td->td_ucred, hostname,
3280 				    MAXHOSTNAMELEN);
3281 				sbuf_printf(&sb, "%s", hostname);
3282 				break;
3283 			case 'I':	/* autoincrementing index */
3284 				sbuf_printf(&sb, "0");
3285 				indexpos = sbuf_len(&sb) - 1;
3286 				break;
3287 			case 'N':	/* process name */
3288 				sbuf_printf(&sb, "%s", comm);
3289 				break;
3290 			case 'P':	/* process id */
3291 				sbuf_printf(&sb, "%u", pid);
3292 				break;
3293 			case 'U':	/* user id */
3294 				sbuf_printf(&sb, "%u", uid);
3295 				break;
3296 			default:
3297 				log(LOG_ERR,
3298 				    "Unknown format character %c in "
3299 				    "corename `%s'\n", format[i], format);
3300 				break;
3301 			}
3302 			break;
3303 		default:
3304 			sbuf_putc(&sb, format[i]);
3305 			break;
3306 		}
3307 	}
3308 	sx_sunlock(&corefilename_lock);
3309 	free(hostname, M_TEMP);
3310 	if (compress)
3311 		sbuf_printf(&sb, GZ_SUFFIX);
3312 	if (sbuf_error(&sb) != 0) {
3313 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3314 		    "long\n", (long)pid, comm, (u_long)uid);
3315 		sbuf_delete(&sb);
3316 		free(name, M_TEMP);
3317 		return (ENOMEM);
3318 	}
3319 	sbuf_finish(&sb);
3320 	sbuf_delete(&sb);
3321 
3322 	cmode = S_IRUSR | S_IWUSR;
3323 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3324 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3325 
3326 	/*
3327 	 * If the core format has a %I in it, then we need to check
3328 	 * for existing corefiles before returning a name.
3329 	 * To do this we iterate over 0..num_cores to find a
3330 	 * non-existing core file name to use.
3331 	 */
3332 	if (indexpos != -1) {
3333 		for (i = 0; i < num_cores; i++) {
3334 			flags = O_CREAT | O_EXCL | FWRITE | O_NOFOLLOW;
3335 			name[indexpos] = '0' + i;
3336 			NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3337 			error = vn_open_cred(&nd, &flags, cmode, oflags,
3338 			    td->td_ucred, NULL);
3339 			if (error) {
3340 				if (error == EEXIST)
3341 					continue;
3342 				log(LOG_ERR,
3343 				    "pid %d (%s), uid (%u):  Path `%s' failed "
3344 				    "on initial open test, error = %d\n",
3345 				    pid, comm, uid, name, error);
3346 			}
3347 			goto out;
3348 		}
3349 	}
3350 
3351 	flags = O_CREAT | FWRITE | O_NOFOLLOW;
3352 	NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3353 	error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred, NULL);
3354 out:
3355 	if (error) {
3356 #ifdef AUDIT
3357 		audit_proc_coredump(td, name, error);
3358 #endif
3359 		free(name, M_TEMP);
3360 		return (error);
3361 	}
3362 	NDFREE(&nd, NDF_ONLY_PNBUF);
3363 	*vpp = nd.ni_vp;
3364 	*namep = name;
3365 	return (0);
3366 }
3367 
3368 static int
3369 coredump_sanitise_path(const char *path)
3370 {
3371 	size_t i;
3372 
3373 	/*
3374 	 * Only send a subset of ASCII to devd(8) because it
3375 	 * might pass these strings to sh -c.
3376 	 */
3377 	for (i = 0; path[i]; i++)
3378 		if (!(isalpha(path[i]) || isdigit(path[i])) &&
3379 		    path[i] != '/' && path[i] != '.' &&
3380 		    path[i] != '-')
3381 			return (0);
3382 
3383 	return (1);
3384 }
3385 
3386 /*
3387  * Dump a process' core.  The main routine does some
3388  * policy checking, and creates the name of the coredump;
3389  * then it passes on a vnode and a size limit to the process-specific
3390  * coredump routine if there is one; if there _is not_ one, it returns
3391  * ENOSYS; otherwise it returns the error from the process-specific routine.
3392  */
3393 
3394 static int
3395 coredump(struct thread *td)
3396 {
3397 	struct proc *p = td->td_proc;
3398 	struct ucred *cred = td->td_ucred;
3399 	struct vnode *vp;
3400 	struct flock lf;
3401 	struct vattr vattr;
3402 	int error, error1, locked;
3403 	char *name;			/* name of corefile */
3404 	void *rl_cookie;
3405 	off_t limit;
3406 	char *data = NULL;
3407 	char *fullpath, *freepath = NULL;
3408 	size_t len;
3409 	static const char comm_name[] = "comm=";
3410 	static const char core_name[] = "core=";
3411 
3412 	PROC_LOCK_ASSERT(p, MA_OWNED);
3413 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3414 	_STOPEVENT(p, S_CORE, 0);
3415 
3416 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
3417 	    (p->p_flag2 & P2_NOTRACE) != 0) {
3418 		PROC_UNLOCK(p);
3419 		return (EFAULT);
3420 	}
3421 
3422 	/*
3423 	 * Note that the bulk of limit checking is done after
3424 	 * the corefile is created.  The exception is if the limit
3425 	 * for corefiles is 0, in which case we don't bother
3426 	 * creating the corefile at all.  This layout means that
3427 	 * a corefile is truncated instead of not being created,
3428 	 * if it is larger than the limit.
3429 	 */
3430 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
3431 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
3432 		PROC_UNLOCK(p);
3433 		return (EFBIG);
3434 	}
3435 	PROC_UNLOCK(p);
3436 
3437 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
3438 	    compress_user_cores, &vp, &name);
3439 	if (error != 0)
3440 		return (error);
3441 
3442 	/*
3443 	 * Don't dump to non-regular files or files with links.
3444 	 * Do not dump into system files.
3445 	 */
3446 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
3447 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0) {
3448 		VOP_UNLOCK(vp, 0);
3449 		error = EFAULT;
3450 		goto out;
3451 	}
3452 
3453 	VOP_UNLOCK(vp, 0);
3454 
3455 	/* Postpone other writers, including core dumps of other processes. */
3456 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
3457 
3458 	lf.l_whence = SEEK_SET;
3459 	lf.l_start = 0;
3460 	lf.l_len = 0;
3461 	lf.l_type = F_WRLCK;
3462 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
3463 
3464 	VATTR_NULL(&vattr);
3465 	vattr.va_size = 0;
3466 	if (set_core_nodump_flag)
3467 		vattr.va_flags = UF_NODUMP;
3468 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3469 	VOP_SETATTR(vp, &vattr, cred);
3470 	VOP_UNLOCK(vp, 0);
3471 	PROC_LOCK(p);
3472 	p->p_acflag |= ACORE;
3473 	PROC_UNLOCK(p);
3474 
3475 	if (p->p_sysent->sv_coredump != NULL) {
3476 		error = p->p_sysent->sv_coredump(td, vp, limit,
3477 		    compress_user_cores ? IMGACT_CORE_COMPRESS : 0);
3478 	} else {
3479 		error = ENOSYS;
3480 	}
3481 
3482 	if (locked) {
3483 		lf.l_type = F_UNLCK;
3484 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3485 	}
3486 	vn_rangelock_unlock(vp, rl_cookie);
3487 
3488 	/*
3489 	 * Notify the userland helper that a process triggered a core dump.
3490 	 * This allows the helper to run an automated debugging session.
3491 	 */
3492 	if (error != 0 || coredump_devctl == 0)
3493 		goto out;
3494 	len = MAXPATHLEN * 2 + sizeof(comm_name) - 1 +
3495 	    sizeof(' ') + sizeof(core_name) - 1;
3496 	data = malloc(len, M_TEMP, M_WAITOK);
3497 	if (vn_fullpath_global(td, p->p_textvp, &fullpath, &freepath) != 0)
3498 		goto out;
3499 	if (!coredump_sanitise_path(fullpath))
3500 		goto out;
3501 	snprintf(data, len, "%s%s ", comm_name, fullpath);
3502 	free(freepath, M_TEMP);
3503 	freepath = NULL;
3504 	if (vn_fullpath_global(td, vp, &fullpath, &freepath) != 0)
3505 		goto out;
3506 	if (!coredump_sanitise_path(fullpath))
3507 		goto out;
3508 	strlcat(data, core_name, len);
3509 	strlcat(data, fullpath, len);
3510 	devctl_notify("kernel", "signal", "coredump", data);
3511 out:
3512 	error1 = vn_close(vp, FWRITE, cred, td);
3513 	if (error == 0)
3514 		error = error1;
3515 #ifdef AUDIT
3516 	audit_proc_coredump(td, name, error);
3517 #endif
3518 	free(freepath, M_TEMP);
3519 	free(data, M_TEMP);
3520 	free(name, M_TEMP);
3521 	return (error);
3522 }
3523 
3524 /*
3525  * Nonexistent system call-- signal process (may want to handle it).  Flag
3526  * error in case process won't see signal immediately (blocked or ignored).
3527  */
3528 #ifndef _SYS_SYSPROTO_H_
3529 struct nosys_args {
3530 	int	dummy;
3531 };
3532 #endif
3533 /* ARGSUSED */
3534 int
3535 nosys(td, args)
3536 	struct thread *td;
3537 	struct nosys_args *args;
3538 {
3539 	struct proc *p = td->td_proc;
3540 
3541 	PROC_LOCK(p);
3542 	tdsignal(td, SIGSYS);
3543 	PROC_UNLOCK(p);
3544 	return (ENOSYS);
3545 }
3546 
3547 /*
3548  * Send a SIGIO or SIGURG signal to a process or process group using stored
3549  * credentials rather than those of the current process.
3550  */
3551 void
3552 pgsigio(sigiop, sig, checkctty)
3553 	struct sigio **sigiop;
3554 	int sig, checkctty;
3555 {
3556 	ksiginfo_t ksi;
3557 	struct sigio *sigio;
3558 
3559 	ksiginfo_init(&ksi);
3560 	ksi.ksi_signo = sig;
3561 	ksi.ksi_code = SI_KERNEL;
3562 
3563 	SIGIO_LOCK();
3564 	sigio = *sigiop;
3565 	if (sigio == NULL) {
3566 		SIGIO_UNLOCK();
3567 		return;
3568 	}
3569 	if (sigio->sio_pgid > 0) {
3570 		PROC_LOCK(sigio->sio_proc);
3571 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
3572 			kern_psignal(sigio->sio_proc, sig);
3573 		PROC_UNLOCK(sigio->sio_proc);
3574 	} else if (sigio->sio_pgid < 0) {
3575 		struct proc *p;
3576 
3577 		PGRP_LOCK(sigio->sio_pgrp);
3578 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
3579 			PROC_LOCK(p);
3580 			if (p->p_state == PRS_NORMAL &&
3581 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
3582 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
3583 				kern_psignal(p, sig);
3584 			PROC_UNLOCK(p);
3585 		}
3586 		PGRP_UNLOCK(sigio->sio_pgrp);
3587 	}
3588 	SIGIO_UNLOCK();
3589 }
3590 
3591 static int
3592 filt_sigattach(struct knote *kn)
3593 {
3594 	struct proc *p = curproc;
3595 
3596 	kn->kn_ptr.p_proc = p;
3597 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
3598 
3599 	knlist_add(p->p_klist, kn, 0);
3600 
3601 	return (0);
3602 }
3603 
3604 static void
3605 filt_sigdetach(struct knote *kn)
3606 {
3607 	struct proc *p = kn->kn_ptr.p_proc;
3608 
3609 	knlist_remove(p->p_klist, kn, 0);
3610 }
3611 
3612 /*
3613  * signal knotes are shared with proc knotes, so we apply a mask to
3614  * the hint in order to differentiate them from process hints.  This
3615  * could be avoided by using a signal-specific knote list, but probably
3616  * isn't worth the trouble.
3617  */
3618 static int
3619 filt_signal(struct knote *kn, long hint)
3620 {
3621 
3622 	if (hint & NOTE_SIGNAL) {
3623 		hint &= ~NOTE_SIGNAL;
3624 
3625 		if (kn->kn_id == hint)
3626 			kn->kn_data++;
3627 	}
3628 	return (kn->kn_data != 0);
3629 }
3630 
3631 struct sigacts *
3632 sigacts_alloc(void)
3633 {
3634 	struct sigacts *ps;
3635 
3636 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
3637 	refcount_init(&ps->ps_refcnt, 1);
3638 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
3639 	return (ps);
3640 }
3641 
3642 void
3643 sigacts_free(struct sigacts *ps)
3644 {
3645 
3646 	if (refcount_release(&ps->ps_refcnt) == 0)
3647 		return;
3648 	mtx_destroy(&ps->ps_mtx);
3649 	free(ps, M_SUBPROC);
3650 }
3651 
3652 struct sigacts *
3653 sigacts_hold(struct sigacts *ps)
3654 {
3655 
3656 	refcount_acquire(&ps->ps_refcnt);
3657 	return (ps);
3658 }
3659 
3660 void
3661 sigacts_copy(struct sigacts *dest, struct sigacts *src)
3662 {
3663 
3664 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
3665 	mtx_lock(&src->ps_mtx);
3666 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
3667 	mtx_unlock(&src->ps_mtx);
3668 }
3669 
3670 int
3671 sigacts_shared(struct sigacts *ps)
3672 {
3673 
3674 	return (ps->ps_refcnt > 1);
3675 }
3676