xref: /freebsd/sys/kern/kern_sig.c (revision 243e928310d073338c5ec089f0dce238a80b9866)
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 void	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 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1255 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1256 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1257 				error = 0;
1258 				break;
1259 			}
1260 		}
1261 
1262 		if (error != 0)
1263 			break;
1264 
1265 		/*
1266 		 * POSIX says this must be checked after looking for pending
1267 		 * signals.
1268 		 */
1269 		if (timeout != NULL) {
1270 			if (!timevalid) {
1271 				error = EINVAL;
1272 				break;
1273 			}
1274 			getnanouptime(&rts);
1275 			if (timespeccmp(&rts, &ets, >=)) {
1276 				error = EAGAIN;
1277 				break;
1278 			}
1279 			ts = ets;
1280 			timespecsub(&ts, &rts);
1281 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1282 			timo = tvtohz(&tv);
1283 		} else {
1284 			timo = 0;
1285 		}
1286 
1287 		error = msleep(ps, &p->p_mtx, PPAUSE|PCATCH, "sigwait", timo);
1288 
1289 		if (timeout != NULL) {
1290 			if (error == ERESTART) {
1291 				/* Timeout can not be restarted. */
1292 				error = EINTR;
1293 			} else if (error == EAGAIN) {
1294 				/* We will calculate timeout by ourself. */
1295 				error = 0;
1296 			}
1297 		}
1298 	}
1299 
1300 	new_block = saved_mask;
1301 	SIGSETNAND(new_block, td->td_sigmask);
1302 	td->td_sigmask = saved_mask;
1303 	/*
1304 	 * Fewer signals can be delivered to us, reschedule signal
1305 	 * notification.
1306 	 */
1307 	if (p->p_numthreads != 1)
1308 		reschedule_signals(p, new_block, 0);
1309 
1310 	if (error == 0) {
1311 		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1312 
1313 		if (ksi->ksi_code == SI_TIMER)
1314 			itimer_accept(p, ksi->ksi_timerid, ksi);
1315 
1316 #ifdef KTRACE
1317 		if (KTRPOINT(td, KTR_PSIG)) {
1318 			sig_t action;
1319 
1320 			mtx_lock(&ps->ps_mtx);
1321 			action = ps->ps_sigact[_SIG_IDX(sig)];
1322 			mtx_unlock(&ps->ps_mtx);
1323 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1324 		}
1325 #endif
1326 		if (sig == SIGKILL)
1327 			sigexit(td, sig);
1328 	}
1329 	PROC_UNLOCK(p);
1330 	return (error);
1331 }
1332 
1333 #ifndef _SYS_SYSPROTO_H_
1334 struct sigpending_args {
1335 	sigset_t	*set;
1336 };
1337 #endif
1338 int
1339 sys_sigpending(td, uap)
1340 	struct thread *td;
1341 	struct sigpending_args *uap;
1342 {
1343 	struct proc *p = td->td_proc;
1344 	sigset_t pending;
1345 
1346 	PROC_LOCK(p);
1347 	pending = p->p_sigqueue.sq_signals;
1348 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1349 	PROC_UNLOCK(p);
1350 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1351 }
1352 
1353 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1354 #ifndef _SYS_SYSPROTO_H_
1355 struct osigpending_args {
1356 	int	dummy;
1357 };
1358 #endif
1359 int
1360 osigpending(td, uap)
1361 	struct thread *td;
1362 	struct osigpending_args *uap;
1363 {
1364 	struct proc *p = td->td_proc;
1365 	sigset_t pending;
1366 
1367 	PROC_LOCK(p);
1368 	pending = p->p_sigqueue.sq_signals;
1369 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1370 	PROC_UNLOCK(p);
1371 	SIG2OSIG(pending, td->td_retval[0]);
1372 	return (0);
1373 }
1374 #endif /* COMPAT_43 */
1375 
1376 #if defined(COMPAT_43)
1377 /*
1378  * Generalized interface signal handler, 4.3-compatible.
1379  */
1380 #ifndef _SYS_SYSPROTO_H_
1381 struct osigvec_args {
1382 	int	signum;
1383 	struct	sigvec *nsv;
1384 	struct	sigvec *osv;
1385 };
1386 #endif
1387 /* ARGSUSED */
1388 int
1389 osigvec(td, uap)
1390 	struct thread *td;
1391 	register struct osigvec_args *uap;
1392 {
1393 	struct sigvec vec;
1394 	struct sigaction nsa, osa;
1395 	register struct sigaction *nsap, *osap;
1396 	int error;
1397 
1398 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1399 		return (EINVAL);
1400 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1401 	osap = (uap->osv != NULL) ? &osa : NULL;
1402 	if (nsap) {
1403 		error = copyin(uap->nsv, &vec, sizeof(vec));
1404 		if (error)
1405 			return (error);
1406 		nsap->sa_handler = vec.sv_handler;
1407 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1408 		nsap->sa_flags = vec.sv_flags;
1409 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1410 	}
1411 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1412 	if (osap && !error) {
1413 		vec.sv_handler = osap->sa_handler;
1414 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1415 		vec.sv_flags = osap->sa_flags;
1416 		vec.sv_flags &= ~SA_NOCLDWAIT;
1417 		vec.sv_flags ^= SA_RESTART;
1418 		error = copyout(&vec, uap->osv, sizeof(vec));
1419 	}
1420 	return (error);
1421 }
1422 
1423 #ifndef _SYS_SYSPROTO_H_
1424 struct osigblock_args {
1425 	int	mask;
1426 };
1427 #endif
1428 int
1429 osigblock(td, uap)
1430 	register struct thread *td;
1431 	struct osigblock_args *uap;
1432 {
1433 	sigset_t set, oset;
1434 
1435 	OSIG2SIG(uap->mask, set);
1436 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1437 	SIG2OSIG(oset, td->td_retval[0]);
1438 	return (0);
1439 }
1440 
1441 #ifndef _SYS_SYSPROTO_H_
1442 struct osigsetmask_args {
1443 	int	mask;
1444 };
1445 #endif
1446 int
1447 osigsetmask(td, uap)
1448 	struct thread *td;
1449 	struct osigsetmask_args *uap;
1450 {
1451 	sigset_t set, oset;
1452 
1453 	OSIG2SIG(uap->mask, set);
1454 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1455 	SIG2OSIG(oset, td->td_retval[0]);
1456 	return (0);
1457 }
1458 #endif /* COMPAT_43 */
1459 
1460 /*
1461  * Suspend calling thread until signal, providing mask to be set in the
1462  * meantime.
1463  */
1464 #ifndef _SYS_SYSPROTO_H_
1465 struct sigsuspend_args {
1466 	const sigset_t *sigmask;
1467 };
1468 #endif
1469 /* ARGSUSED */
1470 int
1471 sys_sigsuspend(td, uap)
1472 	struct thread *td;
1473 	struct sigsuspend_args *uap;
1474 {
1475 	sigset_t mask;
1476 	int error;
1477 
1478 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1479 	if (error)
1480 		return (error);
1481 	return (kern_sigsuspend(td, mask));
1482 }
1483 
1484 int
1485 kern_sigsuspend(struct thread *td, sigset_t mask)
1486 {
1487 	struct proc *p = td->td_proc;
1488 	int has_sig, sig;
1489 
1490 	/*
1491 	 * When returning from sigsuspend, we want
1492 	 * the old mask to be restored after the
1493 	 * signal handler has finished.  Thus, we
1494 	 * save it here and mark the sigacts structure
1495 	 * to indicate this.
1496 	 */
1497 	PROC_LOCK(p);
1498 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1499 	    SIGPROCMASK_PROC_LOCKED);
1500 	td->td_pflags |= TDP_OLDMASK;
1501 
1502 	/*
1503 	 * Process signals now. Otherwise, we can get spurious wakeup
1504 	 * due to signal entered process queue, but delivered to other
1505 	 * thread. But sigsuspend should return only on signal
1506 	 * delivery.
1507 	 */
1508 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1509 	for (has_sig = 0; !has_sig;) {
1510 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1511 			0) == 0)
1512 			/* void */;
1513 		thread_suspend_check(0);
1514 		mtx_lock(&p->p_sigacts->ps_mtx);
1515 		while ((sig = cursig(td)) != 0)
1516 			has_sig += postsig(sig);
1517 		mtx_unlock(&p->p_sigacts->ps_mtx);
1518 	}
1519 	PROC_UNLOCK(p);
1520 	td->td_errno = EINTR;
1521 	td->td_pflags |= TDP_NERRNO;
1522 	return (EJUSTRETURN);
1523 }
1524 
1525 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1526 /*
1527  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1528  * convention: libc stub passes mask, not pointer, to save a copyin.
1529  */
1530 #ifndef _SYS_SYSPROTO_H_
1531 struct osigsuspend_args {
1532 	osigset_t mask;
1533 };
1534 #endif
1535 /* ARGSUSED */
1536 int
1537 osigsuspend(td, uap)
1538 	struct thread *td;
1539 	struct osigsuspend_args *uap;
1540 {
1541 	sigset_t mask;
1542 
1543 	OSIG2SIG(uap->mask, mask);
1544 	return (kern_sigsuspend(td, mask));
1545 }
1546 #endif /* COMPAT_43 */
1547 
1548 #if defined(COMPAT_43)
1549 #ifndef _SYS_SYSPROTO_H_
1550 struct osigstack_args {
1551 	struct	sigstack *nss;
1552 	struct	sigstack *oss;
1553 };
1554 #endif
1555 /* ARGSUSED */
1556 int
1557 osigstack(td, uap)
1558 	struct thread *td;
1559 	register struct osigstack_args *uap;
1560 {
1561 	struct sigstack nss, oss;
1562 	int error = 0;
1563 
1564 	if (uap->nss != NULL) {
1565 		error = copyin(uap->nss, &nss, sizeof(nss));
1566 		if (error)
1567 			return (error);
1568 	}
1569 	oss.ss_sp = td->td_sigstk.ss_sp;
1570 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1571 	if (uap->nss != NULL) {
1572 		td->td_sigstk.ss_sp = nss.ss_sp;
1573 		td->td_sigstk.ss_size = 0;
1574 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1575 		td->td_pflags |= TDP_ALTSTACK;
1576 	}
1577 	if (uap->oss != NULL)
1578 		error = copyout(&oss, uap->oss, sizeof(oss));
1579 
1580 	return (error);
1581 }
1582 #endif /* COMPAT_43 */
1583 
1584 #ifndef _SYS_SYSPROTO_H_
1585 struct sigaltstack_args {
1586 	stack_t	*ss;
1587 	stack_t	*oss;
1588 };
1589 #endif
1590 /* ARGSUSED */
1591 int
1592 sys_sigaltstack(td, uap)
1593 	struct thread *td;
1594 	register struct sigaltstack_args *uap;
1595 {
1596 	stack_t ss, oss;
1597 	int error;
1598 
1599 	if (uap->ss != NULL) {
1600 		error = copyin(uap->ss, &ss, sizeof(ss));
1601 		if (error)
1602 			return (error);
1603 	}
1604 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1605 	    (uap->oss != NULL) ? &oss : NULL);
1606 	if (error)
1607 		return (error);
1608 	if (uap->oss != NULL)
1609 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1610 	return (error);
1611 }
1612 
1613 int
1614 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1615 {
1616 	struct proc *p = td->td_proc;
1617 	int oonstack;
1618 
1619 	oonstack = sigonstack(cpu_getstack(td));
1620 
1621 	if (oss != NULL) {
1622 		*oss = td->td_sigstk;
1623 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1624 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1625 	}
1626 
1627 	if (ss != NULL) {
1628 		if (oonstack)
1629 			return (EPERM);
1630 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1631 			return (EINVAL);
1632 		if (!(ss->ss_flags & SS_DISABLE)) {
1633 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1634 				return (ENOMEM);
1635 
1636 			td->td_sigstk = *ss;
1637 			td->td_pflags |= TDP_ALTSTACK;
1638 		} else {
1639 			td->td_pflags &= ~TDP_ALTSTACK;
1640 		}
1641 	}
1642 	return (0);
1643 }
1644 
1645 /*
1646  * Common code for kill process group/broadcast kill.
1647  * cp is calling process.
1648  */
1649 static int
1650 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1651 {
1652 	struct proc *p;
1653 	struct pgrp *pgrp;
1654 	int err;
1655 	int ret;
1656 
1657 	ret = ESRCH;
1658 	if (all) {
1659 		/*
1660 		 * broadcast
1661 		 */
1662 		sx_slock(&allproc_lock);
1663 		FOREACH_PROC_IN_SYSTEM(p) {
1664 			PROC_LOCK(p);
1665 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1666 			    p == td->td_proc || p->p_state == PRS_NEW) {
1667 				PROC_UNLOCK(p);
1668 				continue;
1669 			}
1670 			err = p_cansignal(td, p, sig);
1671 			if (err == 0) {
1672 				if (sig)
1673 					pksignal(p, sig, ksi);
1674 				ret = err;
1675 			}
1676 			else if (ret == ESRCH)
1677 				ret = err;
1678 			PROC_UNLOCK(p);
1679 		}
1680 		sx_sunlock(&allproc_lock);
1681 	} else {
1682 		sx_slock(&proctree_lock);
1683 		if (pgid == 0) {
1684 			/*
1685 			 * zero pgid means send to my process group.
1686 			 */
1687 			pgrp = td->td_proc->p_pgrp;
1688 			PGRP_LOCK(pgrp);
1689 		} else {
1690 			pgrp = pgfind(pgid);
1691 			if (pgrp == NULL) {
1692 				sx_sunlock(&proctree_lock);
1693 				return (ESRCH);
1694 			}
1695 		}
1696 		sx_sunlock(&proctree_lock);
1697 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1698 			PROC_LOCK(p);
1699 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1700 			    p->p_state == PRS_NEW) {
1701 				PROC_UNLOCK(p);
1702 				continue;
1703 			}
1704 			err = p_cansignal(td, p, sig);
1705 			if (err == 0) {
1706 				if (sig)
1707 					pksignal(p, sig, ksi);
1708 				ret = err;
1709 			}
1710 			else if (ret == ESRCH)
1711 				ret = err;
1712 			PROC_UNLOCK(p);
1713 		}
1714 		PGRP_UNLOCK(pgrp);
1715 	}
1716 	return (ret);
1717 }
1718 
1719 #ifndef _SYS_SYSPROTO_H_
1720 struct kill_args {
1721 	int	pid;
1722 	int	signum;
1723 };
1724 #endif
1725 /* ARGSUSED */
1726 int
1727 sys_kill(struct thread *td, struct kill_args *uap)
1728 {
1729 	ksiginfo_t ksi;
1730 	struct proc *p;
1731 	int error;
1732 
1733 	/*
1734 	 * A process in capability mode can send signals only to himself.
1735 	 * The main rationale behind this is that abort(3) is implemented as
1736 	 * kill(getpid(), SIGABRT).
1737 	 */
1738 	if (IN_CAPABILITY_MODE(td) && uap->pid != td->td_proc->p_pid)
1739 		return (ECAPMODE);
1740 
1741 	AUDIT_ARG_SIGNUM(uap->signum);
1742 	AUDIT_ARG_PID(uap->pid);
1743 	if ((u_int)uap->signum > _SIG_MAXSIG)
1744 		return (EINVAL);
1745 
1746 	ksiginfo_init(&ksi);
1747 	ksi.ksi_signo = uap->signum;
1748 	ksi.ksi_code = SI_USER;
1749 	ksi.ksi_pid = td->td_proc->p_pid;
1750 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1751 
1752 	if (uap->pid > 0) {
1753 		/* kill single process */
1754 		if ((p = pfind(uap->pid)) == NULL) {
1755 			if ((p = zpfind(uap->pid)) == NULL)
1756 				return (ESRCH);
1757 		}
1758 		AUDIT_ARG_PROCESS(p);
1759 		error = p_cansignal(td, p, uap->signum);
1760 		if (error == 0 && uap->signum)
1761 			pksignal(p, uap->signum, &ksi);
1762 		PROC_UNLOCK(p);
1763 		return (error);
1764 	}
1765 	switch (uap->pid) {
1766 	case -1:		/* broadcast signal */
1767 		return (killpg1(td, uap->signum, 0, 1, &ksi));
1768 	case 0:			/* signal own process group */
1769 		return (killpg1(td, uap->signum, 0, 0, &ksi));
1770 	default:		/* negative explicit process group */
1771 		return (killpg1(td, uap->signum, -uap->pid, 0, &ksi));
1772 	}
1773 	/* NOTREACHED */
1774 }
1775 
1776 int
1777 sys_pdkill(td, uap)
1778 	struct thread *td;
1779 	struct pdkill_args *uap;
1780 {
1781 	struct proc *p;
1782 	cap_rights_t rights;
1783 	int error;
1784 
1785 	AUDIT_ARG_SIGNUM(uap->signum);
1786 	AUDIT_ARG_FD(uap->fd);
1787 	if ((u_int)uap->signum > _SIG_MAXSIG)
1788 		return (EINVAL);
1789 
1790 	error = procdesc_find(td, uap->fd,
1791 	    cap_rights_init(&rights, CAP_PDKILL), &p);
1792 	if (error)
1793 		return (error);
1794 	AUDIT_ARG_PROCESS(p);
1795 	error = p_cansignal(td, p, uap->signum);
1796 	if (error == 0 && uap->signum)
1797 		kern_psignal(p, uap->signum);
1798 	PROC_UNLOCK(p);
1799 	return (error);
1800 }
1801 
1802 #if defined(COMPAT_43)
1803 #ifndef _SYS_SYSPROTO_H_
1804 struct okillpg_args {
1805 	int	pgid;
1806 	int	signum;
1807 };
1808 #endif
1809 /* ARGSUSED */
1810 int
1811 okillpg(struct thread *td, struct okillpg_args *uap)
1812 {
1813 	ksiginfo_t ksi;
1814 
1815 	AUDIT_ARG_SIGNUM(uap->signum);
1816 	AUDIT_ARG_PID(uap->pgid);
1817 	if ((u_int)uap->signum > _SIG_MAXSIG)
1818 		return (EINVAL);
1819 
1820 	ksiginfo_init(&ksi);
1821 	ksi.ksi_signo = uap->signum;
1822 	ksi.ksi_code = SI_USER;
1823 	ksi.ksi_pid = td->td_proc->p_pid;
1824 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1825 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1826 }
1827 #endif /* COMPAT_43 */
1828 
1829 #ifndef _SYS_SYSPROTO_H_
1830 struct sigqueue_args {
1831 	pid_t pid;
1832 	int signum;
1833 	/* union sigval */ void *value;
1834 };
1835 #endif
1836 int
1837 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1838 {
1839 	ksiginfo_t ksi;
1840 	struct proc *p;
1841 	int error;
1842 
1843 	if ((u_int)uap->signum > _SIG_MAXSIG)
1844 		return (EINVAL);
1845 
1846 	/*
1847 	 * Specification says sigqueue can only send signal to
1848 	 * single process.
1849 	 */
1850 	if (uap->pid <= 0)
1851 		return (EINVAL);
1852 
1853 	if ((p = pfind(uap->pid)) == NULL) {
1854 		if ((p = zpfind(uap->pid)) == NULL)
1855 			return (ESRCH);
1856 	}
1857 	error = p_cansignal(td, p, uap->signum);
1858 	if (error == 0 && uap->signum != 0) {
1859 		ksiginfo_init(&ksi);
1860 		ksi.ksi_flags = KSI_SIGQ;
1861 		ksi.ksi_signo = uap->signum;
1862 		ksi.ksi_code = SI_QUEUE;
1863 		ksi.ksi_pid = td->td_proc->p_pid;
1864 		ksi.ksi_uid = td->td_ucred->cr_ruid;
1865 		ksi.ksi_value.sival_ptr = uap->value;
1866 		error = pksignal(p, ksi.ksi_signo, &ksi);
1867 	}
1868 	PROC_UNLOCK(p);
1869 	return (error);
1870 }
1871 
1872 /*
1873  * Send a signal to a process group.
1874  */
1875 void
1876 gsignal(int pgid, int sig, ksiginfo_t *ksi)
1877 {
1878 	struct pgrp *pgrp;
1879 
1880 	if (pgid != 0) {
1881 		sx_slock(&proctree_lock);
1882 		pgrp = pgfind(pgid);
1883 		sx_sunlock(&proctree_lock);
1884 		if (pgrp != NULL) {
1885 			pgsignal(pgrp, sig, 0, ksi);
1886 			PGRP_UNLOCK(pgrp);
1887 		}
1888 	}
1889 }
1890 
1891 /*
1892  * Send a signal to a process group.  If checktty is 1,
1893  * limit to members which have a controlling terminal.
1894  */
1895 void
1896 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
1897 {
1898 	struct proc *p;
1899 
1900 	if (pgrp) {
1901 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
1902 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1903 			PROC_LOCK(p);
1904 			if (p->p_state == PRS_NORMAL &&
1905 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
1906 				pksignal(p, sig, ksi);
1907 			PROC_UNLOCK(p);
1908 		}
1909 	}
1910 }
1911 
1912 
1913 /*
1914  * Recalculate the signal mask and reset the signal disposition after
1915  * usermode frame for delivery is formed.  Should be called after
1916  * mach-specific routine, because sysent->sv_sendsig() needs correct
1917  * ps_siginfo and signal mask.
1918  */
1919 static void
1920 postsig_done(int sig, struct thread *td, struct sigacts *ps)
1921 {
1922 	sigset_t mask;
1923 
1924 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1925 	td->td_ru.ru_nsignals++;
1926 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
1927 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
1928 		SIGADDSET(mask, sig);
1929 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
1930 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
1931 	if (SIGISMEMBER(ps->ps_sigreset, sig))
1932 		sigdflt(ps, sig);
1933 }
1934 
1935 
1936 /*
1937  * Send a signal caused by a trap to the current thread.  If it will be
1938  * caught immediately, deliver it with correct code.  Otherwise, post it
1939  * normally.
1940  */
1941 void
1942 trapsignal(struct thread *td, ksiginfo_t *ksi)
1943 {
1944 	struct sigacts *ps;
1945 	struct proc *p;
1946 	int sig;
1947 	int code;
1948 
1949 	p = td->td_proc;
1950 	sig = ksi->ksi_signo;
1951 	code = ksi->ksi_code;
1952 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
1953 
1954 	PROC_LOCK(p);
1955 	ps = p->p_sigacts;
1956 	mtx_lock(&ps->ps_mtx);
1957 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
1958 	    !SIGISMEMBER(td->td_sigmask, sig)) {
1959 #ifdef KTRACE
1960 		if (KTRPOINT(curthread, KTR_PSIG))
1961 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
1962 			    &td->td_sigmask, code);
1963 #endif
1964 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
1965 				ksi, &td->td_sigmask);
1966 		postsig_done(sig, td, ps);
1967 		mtx_unlock(&ps->ps_mtx);
1968 	} else {
1969 		/*
1970 		 * Avoid a possible infinite loop if the thread
1971 		 * masking the signal or process is ignoring the
1972 		 * signal.
1973 		 */
1974 		if (kern_forcesigexit &&
1975 		    (SIGISMEMBER(td->td_sigmask, sig) ||
1976 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
1977 			SIGDELSET(td->td_sigmask, sig);
1978 			SIGDELSET(ps->ps_sigcatch, sig);
1979 			SIGDELSET(ps->ps_sigignore, sig);
1980 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1981 		}
1982 		mtx_unlock(&ps->ps_mtx);
1983 		p->p_code = code;	/* XXX for core dump/debugger */
1984 		p->p_sig = sig;		/* XXX to verify code */
1985 		tdsendsignal(p, td, sig, ksi);
1986 	}
1987 	PROC_UNLOCK(p);
1988 }
1989 
1990 static struct thread *
1991 sigtd(struct proc *p, int sig, int prop)
1992 {
1993 	struct thread *td, *signal_td;
1994 
1995 	PROC_LOCK_ASSERT(p, MA_OWNED);
1996 
1997 	/*
1998 	 * Check if current thread can handle the signal without
1999 	 * switching context to another thread.
2000 	 */
2001 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig))
2002 		return (curthread);
2003 	signal_td = NULL;
2004 	FOREACH_THREAD_IN_PROC(p, td) {
2005 		if (!SIGISMEMBER(td->td_sigmask, sig)) {
2006 			signal_td = td;
2007 			break;
2008 		}
2009 	}
2010 	if (signal_td == NULL)
2011 		signal_td = FIRST_THREAD_IN_PROC(p);
2012 	return (signal_td);
2013 }
2014 
2015 /*
2016  * Send the signal to the process.  If the signal has an action, the action
2017  * is usually performed by the target process rather than the caller; we add
2018  * the signal to the set of pending signals for the process.
2019  *
2020  * Exceptions:
2021  *   o When a stop signal is sent to a sleeping process that takes the
2022  *     default action, the process is stopped without awakening it.
2023  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2024  *     regardless of the signal action (eg, blocked or ignored).
2025  *
2026  * Other ignored signals are discarded immediately.
2027  *
2028  * NB: This function may be entered from the debugger via the "kill" DDB
2029  * command.  There is little that can be done to mitigate the possibly messy
2030  * side effects of this unwise possibility.
2031  */
2032 void
2033 kern_psignal(struct proc *p, int sig)
2034 {
2035 	ksiginfo_t ksi;
2036 
2037 	ksiginfo_init(&ksi);
2038 	ksi.ksi_signo = sig;
2039 	ksi.ksi_code = SI_KERNEL;
2040 	(void) tdsendsignal(p, NULL, sig, &ksi);
2041 }
2042 
2043 int
2044 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2045 {
2046 
2047 	return (tdsendsignal(p, NULL, sig, ksi));
2048 }
2049 
2050 /* Utility function for finding a thread to send signal event to. */
2051 int
2052 sigev_findtd(struct proc *p ,struct sigevent *sigev, struct thread **ttd)
2053 {
2054 	struct thread *td;
2055 
2056 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2057 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2058 		if (td == NULL)
2059 			return (ESRCH);
2060 		*ttd = td;
2061 	} else {
2062 		*ttd = NULL;
2063 		PROC_LOCK(p);
2064 	}
2065 	return (0);
2066 }
2067 
2068 void
2069 tdsignal(struct thread *td, int sig)
2070 {
2071 	ksiginfo_t ksi;
2072 
2073 	ksiginfo_init(&ksi);
2074 	ksi.ksi_signo = sig;
2075 	ksi.ksi_code = SI_KERNEL;
2076 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2077 }
2078 
2079 void
2080 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2081 {
2082 
2083 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2084 }
2085 
2086 int
2087 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2088 {
2089 	sig_t action;
2090 	sigqueue_t *sigqueue;
2091 	int prop;
2092 	struct sigacts *ps;
2093 	int intrval;
2094 	int ret = 0;
2095 	int wakeup_swapper;
2096 
2097 	MPASS(td == NULL || p == td->td_proc);
2098 	PROC_LOCK_ASSERT(p, MA_OWNED);
2099 
2100 	if (!_SIG_VALID(sig))
2101 		panic("%s(): invalid signal %d", __func__, sig);
2102 
2103 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2104 
2105 	/*
2106 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2107 	 */
2108 	if (p->p_state == PRS_ZOMBIE) {
2109 		if (ksi && (ksi->ksi_flags & KSI_INS))
2110 			ksiginfo_tryfree(ksi);
2111 		return (ret);
2112 	}
2113 
2114 	ps = p->p_sigacts;
2115 	KNOTE_LOCKED(&p->p_klist, NOTE_SIGNAL | sig);
2116 	prop = sigprop(sig);
2117 
2118 	if (td == NULL) {
2119 		td = sigtd(p, sig, prop);
2120 		sigqueue = &p->p_sigqueue;
2121 	} else
2122 		sigqueue = &td->td_sigqueue;
2123 
2124 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2125 
2126 	/*
2127 	 * If the signal is being ignored,
2128 	 * then we forget about it immediately.
2129 	 * (Note: we don't set SIGCONT in ps_sigignore,
2130 	 * and if it is set to SIG_IGN,
2131 	 * action will be SIG_DFL here.)
2132 	 */
2133 	mtx_lock(&ps->ps_mtx);
2134 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2135 		SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2136 
2137 		mtx_unlock(&ps->ps_mtx);
2138 		if (ksi && (ksi->ksi_flags & KSI_INS))
2139 			ksiginfo_tryfree(ksi);
2140 		return (ret);
2141 	}
2142 	if (SIGISMEMBER(td->td_sigmask, sig))
2143 		action = SIG_HOLD;
2144 	else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2145 		action = SIG_CATCH;
2146 	else
2147 		action = SIG_DFL;
2148 	if (SIGISMEMBER(ps->ps_sigintr, sig))
2149 		intrval = EINTR;
2150 	else
2151 		intrval = ERESTART;
2152 	mtx_unlock(&ps->ps_mtx);
2153 
2154 	if (prop & SA_CONT)
2155 		sigqueue_delete_stopmask_proc(p);
2156 	else if (prop & SA_STOP) {
2157 		/*
2158 		 * If sending a tty stop signal to a member of an orphaned
2159 		 * process group, discard the signal here if the action
2160 		 * is default; don't stop the process below if sleeping,
2161 		 * and don't clear any pending SIGCONT.
2162 		 */
2163 		if ((prop & SA_TTYSTOP) &&
2164 		    (p->p_pgrp->pg_jobc == 0) &&
2165 		    (action == SIG_DFL)) {
2166 			if (ksi && (ksi->ksi_flags & KSI_INS))
2167 				ksiginfo_tryfree(ksi);
2168 			return (ret);
2169 		}
2170 		sigqueue_delete_proc(p, SIGCONT);
2171 		if (p->p_flag & P_CONTINUED) {
2172 			p->p_flag &= ~P_CONTINUED;
2173 			PROC_LOCK(p->p_pptr);
2174 			sigqueue_take(p->p_ksi);
2175 			PROC_UNLOCK(p->p_pptr);
2176 		}
2177 	}
2178 
2179 	ret = sigqueue_add(sigqueue, sig, ksi);
2180 	if (ret != 0)
2181 		return (ret);
2182 	signotify(td);
2183 	/*
2184 	 * Defer further processing for signals which are held,
2185 	 * except that stopped processes must be continued by SIGCONT.
2186 	 */
2187 	if (action == SIG_HOLD &&
2188 	    !((prop & SA_CONT) && (p->p_flag & P_STOPPED_SIG)))
2189 		return (ret);
2190 	/*
2191 	 * SIGKILL: Remove procfs STOPEVENTs.
2192 	 */
2193 	if (sig == SIGKILL) {
2194 		/* from procfs_ioctl.c: PIOCBIC */
2195 		p->p_stops = 0;
2196 		/* from procfs_ioctl.c: PIOCCONT */
2197 		p->p_step = 0;
2198 		wakeup(&p->p_step);
2199 	}
2200 	/*
2201 	 * Some signals have a process-wide effect and a per-thread
2202 	 * component.  Most processing occurs when the process next
2203 	 * tries to cross the user boundary, however there are some
2204 	 * times when processing needs to be done immediately, such as
2205 	 * waking up threads so that they can cross the user boundary.
2206 	 * We try to do the per-process part here.
2207 	 */
2208 	if (P_SHOULDSTOP(p)) {
2209 		KASSERT(!(p->p_flag & P_WEXIT),
2210 		    ("signal to stopped but exiting process"));
2211 		if (sig == SIGKILL) {
2212 			/*
2213 			 * If traced process is already stopped,
2214 			 * then no further action is necessary.
2215 			 */
2216 			if (p->p_flag & P_TRACED)
2217 				goto out;
2218 			/*
2219 			 * SIGKILL sets process running.
2220 			 * It will die elsewhere.
2221 			 * All threads must be restarted.
2222 			 */
2223 			p->p_flag &= ~P_STOPPED_SIG;
2224 			goto runfast;
2225 		}
2226 
2227 		if (prop & SA_CONT) {
2228 			/*
2229 			 * If traced process is already stopped,
2230 			 * then no further action is necessary.
2231 			 */
2232 			if (p->p_flag & P_TRACED)
2233 				goto out;
2234 			/*
2235 			 * If SIGCONT is default (or ignored), we continue the
2236 			 * process but don't leave the signal in sigqueue as
2237 			 * it has no further action.  If SIGCONT is held, we
2238 			 * continue the process and leave the signal in
2239 			 * sigqueue.  If the process catches SIGCONT, let it
2240 			 * handle the signal itself.  If it isn't waiting on
2241 			 * an event, it goes back to run state.
2242 			 * Otherwise, process goes back to sleep state.
2243 			 */
2244 			p->p_flag &= ~P_STOPPED_SIG;
2245 			PROC_SLOCK(p);
2246 			if (p->p_numthreads == p->p_suspcount) {
2247 				PROC_SUNLOCK(p);
2248 				p->p_flag |= P_CONTINUED;
2249 				p->p_xsig = SIGCONT;
2250 				PROC_LOCK(p->p_pptr);
2251 				childproc_continued(p);
2252 				PROC_UNLOCK(p->p_pptr);
2253 				PROC_SLOCK(p);
2254 			}
2255 			if (action == SIG_DFL) {
2256 				thread_unsuspend(p);
2257 				PROC_SUNLOCK(p);
2258 				sigqueue_delete(sigqueue, sig);
2259 				goto out;
2260 			}
2261 			if (action == SIG_CATCH) {
2262 				/*
2263 				 * The process wants to catch it so it needs
2264 				 * to run at least one thread, but which one?
2265 				 */
2266 				PROC_SUNLOCK(p);
2267 				goto runfast;
2268 			}
2269 			/*
2270 			 * The signal is not ignored or caught.
2271 			 */
2272 			thread_unsuspend(p);
2273 			PROC_SUNLOCK(p);
2274 			goto out;
2275 		}
2276 
2277 		if (prop & SA_STOP) {
2278 			/*
2279 			 * If traced process is already stopped,
2280 			 * then no further action is necessary.
2281 			 */
2282 			if (p->p_flag & P_TRACED)
2283 				goto out;
2284 			/*
2285 			 * Already stopped, don't need to stop again
2286 			 * (If we did the shell could get confused).
2287 			 * Just make sure the signal STOP bit set.
2288 			 */
2289 			p->p_flag |= P_STOPPED_SIG;
2290 			sigqueue_delete(sigqueue, sig);
2291 			goto out;
2292 		}
2293 
2294 		/*
2295 		 * All other kinds of signals:
2296 		 * If a thread is sleeping interruptibly, simulate a
2297 		 * wakeup so that when it is continued it will be made
2298 		 * runnable and can look at the signal.  However, don't make
2299 		 * the PROCESS runnable, leave it stopped.
2300 		 * It may run a bit until it hits a thread_suspend_check().
2301 		 */
2302 		wakeup_swapper = 0;
2303 		PROC_SLOCK(p);
2304 		thread_lock(td);
2305 		if (TD_ON_SLEEPQ(td) && (td->td_flags & TDF_SINTR))
2306 			wakeup_swapper = sleepq_abort(td, intrval);
2307 		thread_unlock(td);
2308 		PROC_SUNLOCK(p);
2309 		if (wakeup_swapper)
2310 			kick_proc0();
2311 		goto out;
2312 		/*
2313 		 * Mutexes are short lived. Threads waiting on them will
2314 		 * hit thread_suspend_check() soon.
2315 		 */
2316 	} else if (p->p_state == PRS_NORMAL) {
2317 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2318 			tdsigwakeup(td, sig, action, intrval);
2319 			goto out;
2320 		}
2321 
2322 		MPASS(action == SIG_DFL);
2323 
2324 		if (prop & SA_STOP) {
2325 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2326 				goto out;
2327 			p->p_flag |= P_STOPPED_SIG;
2328 			p->p_xsig = sig;
2329 			PROC_SLOCK(p);
2330 			sig_suspend_threads(td, p, 1);
2331 			if (p->p_numthreads == p->p_suspcount) {
2332 				/*
2333 				 * only thread sending signal to another
2334 				 * process can reach here, if thread is sending
2335 				 * signal to its process, because thread does
2336 				 * not suspend itself here, p_numthreads
2337 				 * should never be equal to p_suspcount.
2338 				 */
2339 				thread_stopped(p);
2340 				PROC_SUNLOCK(p);
2341 				sigqueue_delete_proc(p, p->p_xsig);
2342 			} else
2343 				PROC_SUNLOCK(p);
2344 			goto out;
2345 		}
2346 	} else {
2347 		/* Not in "NORMAL" state. discard the signal. */
2348 		sigqueue_delete(sigqueue, sig);
2349 		goto out;
2350 	}
2351 
2352 	/*
2353 	 * The process is not stopped so we need to apply the signal to all the
2354 	 * running threads.
2355 	 */
2356 runfast:
2357 	tdsigwakeup(td, sig, action, intrval);
2358 	PROC_SLOCK(p);
2359 	thread_unsuspend(p);
2360 	PROC_SUNLOCK(p);
2361 out:
2362 	/* If we jump here, proc slock should not be owned. */
2363 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2364 	return (ret);
2365 }
2366 
2367 /*
2368  * The force of a signal has been directed against a single
2369  * thread.  We need to see what we can do about knocking it
2370  * out of any sleep it may be in etc.
2371  */
2372 static void
2373 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2374 {
2375 	struct proc *p = td->td_proc;
2376 	register int prop;
2377 	int wakeup_swapper;
2378 
2379 	wakeup_swapper = 0;
2380 	PROC_LOCK_ASSERT(p, MA_OWNED);
2381 	prop = sigprop(sig);
2382 
2383 	PROC_SLOCK(p);
2384 	thread_lock(td);
2385 	/*
2386 	 * Bring the priority of a thread up if we want it to get
2387 	 * killed in this lifetime.  Be careful to avoid bumping the
2388 	 * priority of the idle thread, since we still allow to signal
2389 	 * kernel processes.
2390 	 */
2391 	if (action == SIG_DFL && (prop & SA_KILL) != 0 &&
2392 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2393 		sched_prio(td, PUSER);
2394 	if (TD_ON_SLEEPQ(td)) {
2395 		/*
2396 		 * If thread is sleeping uninterruptibly
2397 		 * we can't interrupt the sleep... the signal will
2398 		 * be noticed when the process returns through
2399 		 * trap() or syscall().
2400 		 */
2401 		if ((td->td_flags & TDF_SINTR) == 0)
2402 			goto out;
2403 		/*
2404 		 * If SIGCONT is default (or ignored) and process is
2405 		 * asleep, we are finished; the process should not
2406 		 * be awakened.
2407 		 */
2408 		if ((prop & SA_CONT) && action == SIG_DFL) {
2409 			thread_unlock(td);
2410 			PROC_SUNLOCK(p);
2411 			sigqueue_delete(&p->p_sigqueue, sig);
2412 			/*
2413 			 * It may be on either list in this state.
2414 			 * Remove from both for now.
2415 			 */
2416 			sigqueue_delete(&td->td_sigqueue, sig);
2417 			return;
2418 		}
2419 
2420 		/*
2421 		 * Don't awaken a sleeping thread for SIGSTOP if the
2422 		 * STOP signal is deferred.
2423 		 */
2424 		if ((prop & SA_STOP) && (td->td_flags & TDF_SBDRY))
2425 			goto out;
2426 
2427 		/*
2428 		 * Give low priority threads a better chance to run.
2429 		 */
2430 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2431 			sched_prio(td, PUSER);
2432 
2433 		wakeup_swapper = sleepq_abort(td, intrval);
2434 	} else {
2435 		/*
2436 		 * Other states do nothing with the signal immediately,
2437 		 * other than kicking ourselves if we are running.
2438 		 * It will either never be noticed, or noticed very soon.
2439 		 */
2440 #ifdef SMP
2441 		if (TD_IS_RUNNING(td) && td != curthread)
2442 			forward_signal(td);
2443 #endif
2444 	}
2445 out:
2446 	PROC_SUNLOCK(p);
2447 	thread_unlock(td);
2448 	if (wakeup_swapper)
2449 		kick_proc0();
2450 }
2451 
2452 static void
2453 sig_suspend_threads(struct thread *td, struct proc *p, int sending)
2454 {
2455 	struct thread *td2;
2456 
2457 	PROC_LOCK_ASSERT(p, MA_OWNED);
2458 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2459 
2460 	FOREACH_THREAD_IN_PROC(p, td2) {
2461 		thread_lock(td2);
2462 		td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
2463 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2464 		    (td2->td_flags & TDF_SINTR)) {
2465 			if (td2->td_flags & TDF_SBDRY) {
2466 				/*
2467 				 * Once a thread is asleep with
2468 				 * TDF_SBDRY set, it should never
2469 				 * become suspended due to this check.
2470 				 */
2471 				KASSERT(!TD_IS_SUSPENDED(td2),
2472 				    ("thread with deferred stops suspended"));
2473 			} else if (!TD_IS_SUSPENDED(td2)) {
2474 				thread_suspend_one(td2);
2475 			}
2476 		} else if (!TD_IS_SUSPENDED(td2)) {
2477 			if (sending || td != td2)
2478 				td2->td_flags |= TDF_ASTPENDING;
2479 #ifdef SMP
2480 			if (TD_IS_RUNNING(td2) && td2 != td)
2481 				forward_signal(td2);
2482 #endif
2483 		}
2484 		thread_unlock(td2);
2485 	}
2486 }
2487 
2488 int
2489 ptracestop(struct thread *td, int sig)
2490 {
2491 	struct proc *p = td->td_proc;
2492 
2493 	PROC_LOCK_ASSERT(p, MA_OWNED);
2494 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2495 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2496 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2497 
2498 	td->td_dbgflags |= TDB_XSIG;
2499 	td->td_xsig = sig;
2500 	CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2501 	    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2502 	PROC_SLOCK(p);
2503 	while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2504 		if (p->p_flag & P_SINGLE_EXIT &&
2505 		    !(td->td_dbgflags & TDB_EXIT)) {
2506 			/*
2507 			 * Ignore ptrace stops except for thread exit
2508 			 * events when the process exits.
2509 			 */
2510 			td->td_dbgflags &= ~TDB_XSIG;
2511 			PROC_SUNLOCK(p);
2512 			return (sig);
2513 		}
2514 		/*
2515 		 * Just make wait() to work, the last stopped thread
2516 		 * will win.
2517 		 */
2518 		p->p_xsig = sig;
2519 		p->p_xthread = td;
2520 		p->p_flag |= (P_STOPPED_SIG|P_STOPPED_TRACE);
2521 		sig_suspend_threads(td, p, 0);
2522 		if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2523 			td->td_dbgflags &= ~TDB_STOPATFORK;
2524 			cv_broadcast(&p->p_dbgwait);
2525 		}
2526 stopme:
2527 		thread_suspend_switch(td, p);
2528 		if (p->p_xthread == td)
2529 			p->p_xthread = NULL;
2530 		if (!(p->p_flag & P_TRACED))
2531 			break;
2532 		if (td->td_dbgflags & TDB_SUSPEND) {
2533 			if (p->p_flag & P_SINGLE_EXIT)
2534 				break;
2535 			goto stopme;
2536 		}
2537 	}
2538 	PROC_SUNLOCK(p);
2539 	return (td->td_xsig);
2540 }
2541 
2542 static void
2543 reschedule_signals(struct proc *p, sigset_t block, int flags)
2544 {
2545 	struct sigacts *ps;
2546 	struct thread *td;
2547 	int sig;
2548 
2549 	PROC_LOCK_ASSERT(p, MA_OWNED);
2550 	ps = p->p_sigacts;
2551 	mtx_assert(&ps->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0 ?
2552 	    MA_OWNED : MA_NOTOWNED);
2553 	if (SIGISEMPTY(p->p_siglist))
2554 		return;
2555 	SIGSETAND(block, p->p_siglist);
2556 	while ((sig = sig_ffs(&block)) != 0) {
2557 		SIGDELSET(block, sig);
2558 		td = sigtd(p, sig, 0);
2559 		signotify(td);
2560 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2561 			mtx_lock(&ps->ps_mtx);
2562 		if (p->p_flag & P_TRACED || SIGISMEMBER(ps->ps_sigcatch, sig))
2563 			tdsigwakeup(td, sig, SIG_CATCH,
2564 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2565 			     ERESTART));
2566 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2567 			mtx_unlock(&ps->ps_mtx);
2568 	}
2569 }
2570 
2571 void
2572 tdsigcleanup(struct thread *td)
2573 {
2574 	struct proc *p;
2575 	sigset_t unblocked;
2576 
2577 	p = td->td_proc;
2578 	PROC_LOCK_ASSERT(p, MA_OWNED);
2579 
2580 	sigqueue_flush(&td->td_sigqueue);
2581 	if (p->p_numthreads == 1)
2582 		return;
2583 
2584 	/*
2585 	 * Since we cannot handle signals, notify signal post code
2586 	 * about this by filling the sigmask.
2587 	 *
2588 	 * Also, if needed, wake up thread(s) that do not block the
2589 	 * same signals as the exiting thread, since the thread might
2590 	 * have been selected for delivery and woken up.
2591 	 */
2592 	SIGFILLSET(unblocked);
2593 	SIGSETNAND(unblocked, td->td_sigmask);
2594 	SIGFILLSET(td->td_sigmask);
2595 	reschedule_signals(p, unblocked, 0);
2596 
2597 }
2598 
2599 /*
2600  * Defer the delivery of SIGSTOP for the current thread.  Returns true
2601  * if stops were deferred and false if they were already deferred.
2602  */
2603 int
2604 sigdeferstop(void)
2605 {
2606 	struct thread *td;
2607 
2608 	td = curthread;
2609 	if (td->td_flags & TDF_SBDRY)
2610 		return (0);
2611 	thread_lock(td);
2612 	td->td_flags |= TDF_SBDRY;
2613 	thread_unlock(td);
2614 	return (1);
2615 }
2616 
2617 /*
2618  * Permit the delivery of SIGSTOP for the current thread.  This does
2619  * not immediately suspend if a stop was posted.  Instead, the thread
2620  * will suspend either via ast() or a subsequent interruptible sleep.
2621  */
2622 int
2623 sigallowstop(void)
2624 {
2625 	struct thread *td;
2626 	int prev;
2627 
2628 	td = curthread;
2629 	thread_lock(td);
2630 	prev = (td->td_flags & TDF_SBDRY) != 0;
2631 	td->td_flags &= ~TDF_SBDRY;
2632 	thread_unlock(td);
2633 	return (prev);
2634 }
2635 
2636 /*
2637  * If the current process has received a signal (should be caught or cause
2638  * termination, should interrupt current syscall), return the signal number.
2639  * Stop signals with default action are processed immediately, then cleared;
2640  * they aren't returned.  This is checked after each entry to the system for
2641  * a syscall or trap (though this can usually be done without calling issignal
2642  * by checking the pending signal masks in cursig.) The normal call
2643  * sequence is
2644  *
2645  *	while (sig = cursig(curthread))
2646  *		postsig(sig);
2647  */
2648 static int
2649 issignal(struct thread *td)
2650 {
2651 	struct proc *p;
2652 	struct sigacts *ps;
2653 	struct sigqueue *queue;
2654 	sigset_t sigpending;
2655 	int sig, prop, newsig;
2656 
2657 	p = td->td_proc;
2658 	ps = p->p_sigacts;
2659 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2660 	PROC_LOCK_ASSERT(p, MA_OWNED);
2661 	for (;;) {
2662 		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
2663 
2664 		sigpending = td->td_sigqueue.sq_signals;
2665 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
2666 		SIGSETNAND(sigpending, td->td_sigmask);
2667 
2668 		if (p->p_flag & P_PPWAIT || td->td_flags & TDF_SBDRY)
2669 			SIG_STOPSIGMASK(sigpending);
2670 		if (SIGISEMPTY(sigpending))	/* no signal to send */
2671 			return (0);
2672 		sig = sig_ffs(&sigpending);
2673 
2674 		if (p->p_stops & S_SIG) {
2675 			mtx_unlock(&ps->ps_mtx);
2676 			stopevent(p, S_SIG, sig);
2677 			mtx_lock(&ps->ps_mtx);
2678 		}
2679 
2680 		/*
2681 		 * We should see pending but ignored signals
2682 		 * only if P_TRACED was on when they were posted.
2683 		 */
2684 		if (SIGISMEMBER(ps->ps_sigignore, sig) && (traced == 0)) {
2685 			sigqueue_delete(&td->td_sigqueue, sig);
2686 			sigqueue_delete(&p->p_sigqueue, sig);
2687 			continue;
2688 		}
2689 		if (p->p_flag & P_TRACED && (p->p_flag & P_PPTRACE) == 0) {
2690 			/*
2691 			 * If traced, always stop.
2692 			 * Remove old signal from queue before the stop.
2693 			 * XXX shrug off debugger, it causes siginfo to
2694 			 * be thrown away.
2695 			 */
2696 			queue = &td->td_sigqueue;
2697 			td->td_dbgksi.ksi_signo = 0;
2698 			if (sigqueue_get(queue, sig, &td->td_dbgksi) == 0) {
2699 				queue = &p->p_sigqueue;
2700 				sigqueue_get(queue, sig, &td->td_dbgksi);
2701 			}
2702 
2703 			mtx_unlock(&ps->ps_mtx);
2704 			newsig = ptracestop(td, sig);
2705 			mtx_lock(&ps->ps_mtx);
2706 
2707 			if (sig != newsig) {
2708 
2709 				/*
2710 				 * If parent wants us to take the signal,
2711 				 * then it will leave it in p->p_xsig;
2712 				 * otherwise we just look for signals again.
2713 				*/
2714 				if (newsig == 0)
2715 					continue;
2716 				sig = newsig;
2717 
2718 				/*
2719 				 * Put the new signal into td_sigqueue. If the
2720 				 * signal is being masked, look for other
2721 				 * signals.
2722 				 */
2723 				sigqueue_add(queue, sig, NULL);
2724 				if (SIGISMEMBER(td->td_sigmask, sig))
2725 					continue;
2726 				signotify(td);
2727 			} else {
2728 				if (td->td_dbgksi.ksi_signo != 0) {
2729 					td->td_dbgksi.ksi_flags |= KSI_HEAD;
2730 					if (sigqueue_add(&td->td_sigqueue, sig,
2731 					    &td->td_dbgksi) != 0)
2732 						td->td_dbgksi.ksi_signo = 0;
2733 				}
2734 				if (td->td_dbgksi.ksi_signo == 0)
2735 					sigqueue_add(&td->td_sigqueue, sig,
2736 					    NULL);
2737 			}
2738 
2739 			/*
2740 			 * If the traced bit got turned off, go back up
2741 			 * to the top to rescan signals.  This ensures
2742 			 * that p_sig* and p_sigact are consistent.
2743 			 */
2744 			if ((p->p_flag & P_TRACED) == 0)
2745 				continue;
2746 		}
2747 
2748 		prop = sigprop(sig);
2749 
2750 		/*
2751 		 * Decide whether the signal should be returned.
2752 		 * Return the signal's number, or fall through
2753 		 * to clear it from the pending mask.
2754 		 */
2755 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2756 
2757 		case (intptr_t)SIG_DFL:
2758 			/*
2759 			 * Don't take default actions on system processes.
2760 			 */
2761 			if (p->p_pid <= 1) {
2762 #ifdef DIAGNOSTIC
2763 				/*
2764 				 * Are you sure you want to ignore SIGSEGV
2765 				 * in init? XXX
2766 				 */
2767 				printf("Process (pid %lu) got signal %d\n",
2768 					(u_long)p->p_pid, sig);
2769 #endif
2770 				break;		/* == ignore */
2771 			}
2772 			/*
2773 			 * If there is a pending stop signal to process
2774 			 * with default action, stop here,
2775 			 * then clear the signal.  However,
2776 			 * if process is member of an orphaned
2777 			 * process group, ignore tty stop signals.
2778 			 */
2779 			if (prop & SA_STOP) {
2780 				if (p->p_flag & (P_TRACED|P_WEXIT) ||
2781 				    (p->p_pgrp->pg_jobc == 0 &&
2782 				     prop & SA_TTYSTOP))
2783 					break;	/* == ignore */
2784 				mtx_unlock(&ps->ps_mtx);
2785 				WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2786 				    &p->p_mtx.lock_object, "Catching SIGSTOP");
2787 				p->p_flag |= P_STOPPED_SIG;
2788 				p->p_xsig = sig;
2789 				PROC_SLOCK(p);
2790 				sig_suspend_threads(td, p, 0);
2791 				thread_suspend_switch(td, p);
2792 				PROC_SUNLOCK(p);
2793 				mtx_lock(&ps->ps_mtx);
2794 				break;
2795 			} else if (prop & SA_IGNORE) {
2796 				/*
2797 				 * Except for SIGCONT, shouldn't get here.
2798 				 * Default action is to ignore; drop it.
2799 				 */
2800 				break;		/* == ignore */
2801 			} else
2802 				return (sig);
2803 			/*NOTREACHED*/
2804 
2805 		case (intptr_t)SIG_IGN:
2806 			/*
2807 			 * Masking above should prevent us ever trying
2808 			 * to take action on an ignored signal other
2809 			 * than SIGCONT, unless process is traced.
2810 			 */
2811 			if ((prop & SA_CONT) == 0 &&
2812 			    (p->p_flag & P_TRACED) == 0)
2813 				printf("issignal\n");
2814 			break;		/* == ignore */
2815 
2816 		default:
2817 			/*
2818 			 * This signal has an action, let
2819 			 * postsig() process it.
2820 			 */
2821 			return (sig);
2822 		}
2823 		sigqueue_delete(&td->td_sigqueue, sig);	/* take the signal! */
2824 		sigqueue_delete(&p->p_sigqueue, sig);
2825 	}
2826 	/* NOTREACHED */
2827 }
2828 
2829 void
2830 thread_stopped(struct proc *p)
2831 {
2832 	int n;
2833 
2834 	PROC_LOCK_ASSERT(p, MA_OWNED);
2835 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2836 	n = p->p_suspcount;
2837 	if (p == curproc)
2838 		n++;
2839 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
2840 		PROC_SUNLOCK(p);
2841 		p->p_flag &= ~P_WAITED;
2842 		PROC_LOCK(p->p_pptr);
2843 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
2844 			CLD_TRAPPED : CLD_STOPPED);
2845 		PROC_UNLOCK(p->p_pptr);
2846 		PROC_SLOCK(p);
2847 	}
2848 }
2849 
2850 /*
2851  * Take the action for the specified signal
2852  * from the current set of pending signals.
2853  */
2854 int
2855 postsig(sig)
2856 	register int sig;
2857 {
2858 	struct thread *td = curthread;
2859 	register struct proc *p = td->td_proc;
2860 	struct sigacts *ps;
2861 	sig_t action;
2862 	ksiginfo_t ksi;
2863 	sigset_t returnmask;
2864 
2865 	KASSERT(sig != 0, ("postsig"));
2866 
2867 	PROC_LOCK_ASSERT(p, MA_OWNED);
2868 	ps = p->p_sigacts;
2869 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2870 	ksiginfo_init(&ksi);
2871 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
2872 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
2873 		return (0);
2874 	ksi.ksi_signo = sig;
2875 	if (ksi.ksi_code == SI_TIMER)
2876 		itimer_accept(p, ksi.ksi_timerid, &ksi);
2877 	action = ps->ps_sigact[_SIG_IDX(sig)];
2878 #ifdef KTRACE
2879 	if (KTRPOINT(td, KTR_PSIG))
2880 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
2881 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
2882 #endif
2883 	if (p->p_stops & S_SIG) {
2884 		mtx_unlock(&ps->ps_mtx);
2885 		stopevent(p, S_SIG, sig);
2886 		mtx_lock(&ps->ps_mtx);
2887 	}
2888 
2889 	if (action == SIG_DFL) {
2890 		/*
2891 		 * Default action, where the default is to kill
2892 		 * the process.  (Other cases were ignored above.)
2893 		 */
2894 		mtx_unlock(&ps->ps_mtx);
2895 		sigexit(td, sig);
2896 		/* NOTREACHED */
2897 	} else {
2898 		/*
2899 		 * If we get here, the signal must be caught.
2900 		 */
2901 		KASSERT(action != SIG_IGN && !SIGISMEMBER(td->td_sigmask, sig),
2902 		    ("postsig action"));
2903 		/*
2904 		 * Set the new mask value and also defer further
2905 		 * occurrences of this signal.
2906 		 *
2907 		 * Special case: user has done a sigsuspend.  Here the
2908 		 * current mask is not of interest, but rather the
2909 		 * mask from before the sigsuspend is what we want
2910 		 * restored after the signal processing is completed.
2911 		 */
2912 		if (td->td_pflags & TDP_OLDMASK) {
2913 			returnmask = td->td_oldsigmask;
2914 			td->td_pflags &= ~TDP_OLDMASK;
2915 		} else
2916 			returnmask = td->td_sigmask;
2917 
2918 		if (p->p_sig == sig) {
2919 			p->p_code = 0;
2920 			p->p_sig = 0;
2921 		}
2922 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
2923 		postsig_done(sig, td, ps);
2924 	}
2925 	return (1);
2926 }
2927 
2928 /*
2929  * Kill the current process for stated reason.
2930  */
2931 void
2932 killproc(p, why)
2933 	struct proc *p;
2934 	char *why;
2935 {
2936 
2937 	PROC_LOCK_ASSERT(p, MA_OWNED);
2938 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
2939 	    p->p_comm);
2940 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n", p->p_pid,
2941 	    p->p_comm, p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2942 	p->p_flag |= P_WKILLED;
2943 	kern_psignal(p, SIGKILL);
2944 }
2945 
2946 /*
2947  * Force the current process to exit with the specified signal, dumping core
2948  * if appropriate.  We bypass the normal tests for masked and caught signals,
2949  * allowing unrecoverable failures to terminate the process without changing
2950  * signal state.  Mark the accounting record with the signal termination.
2951  * If dumping core, save the signal number for the debugger.  Calls exit and
2952  * does not return.
2953  */
2954 void
2955 sigexit(td, sig)
2956 	struct thread *td;
2957 	int sig;
2958 {
2959 	struct proc *p = td->td_proc;
2960 
2961 	PROC_LOCK_ASSERT(p, MA_OWNED);
2962 	p->p_acflag |= AXSIG;
2963 	/*
2964 	 * We must be single-threading to generate a core dump.  This
2965 	 * ensures that the registers in the core file are up-to-date.
2966 	 * Also, the ELF dump handler assumes that the thread list doesn't
2967 	 * change out from under it.
2968 	 *
2969 	 * XXX If another thread attempts to single-thread before us
2970 	 *     (e.g. via fork()), we won't get a dump at all.
2971 	 */
2972 	if ((sigprop(sig) & SA_CORE) && thread_single(p, SINGLE_NO_EXIT) == 0) {
2973 		p->p_sig = sig;
2974 		/*
2975 		 * Log signals which would cause core dumps
2976 		 * (Log as LOG_INFO to appease those who don't want
2977 		 * these messages.)
2978 		 * XXX : Todo, as well as euid, write out ruid too
2979 		 * Note that coredump() drops proc lock.
2980 		 */
2981 		if (coredump(td) == 0)
2982 			sig |= WCOREFLAG;
2983 		if (kern_logsigexit)
2984 			log(LOG_INFO,
2985 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2986 			    p->p_pid, p->p_comm,
2987 			    td->td_ucred ? td->td_ucred->cr_uid : -1,
2988 			    sig &~ WCOREFLAG,
2989 			    sig & WCOREFLAG ? " (core dumped)" : "");
2990 	} else
2991 		PROC_UNLOCK(p);
2992 	exit1(td, 0, sig);
2993 	/* NOTREACHED */
2994 }
2995 
2996 /*
2997  * Send queued SIGCHLD to parent when child process's state
2998  * is changed.
2999  */
3000 static void
3001 sigparent(struct proc *p, int reason, int status)
3002 {
3003 	PROC_LOCK_ASSERT(p, MA_OWNED);
3004 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3005 
3006 	if (p->p_ksi != NULL) {
3007 		p->p_ksi->ksi_signo  = SIGCHLD;
3008 		p->p_ksi->ksi_code   = reason;
3009 		p->p_ksi->ksi_status = status;
3010 		p->p_ksi->ksi_pid    = p->p_pid;
3011 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3012 		if (KSI_ONQ(p->p_ksi))
3013 			return;
3014 	}
3015 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3016 }
3017 
3018 static void
3019 childproc_jobstate(struct proc *p, int reason, int sig)
3020 {
3021 	struct sigacts *ps;
3022 
3023 	PROC_LOCK_ASSERT(p, MA_OWNED);
3024 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3025 
3026 	/*
3027 	 * Wake up parent sleeping in kern_wait(), also send
3028 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3029 	 * that parent will awake, because parent may masked
3030 	 * the signal.
3031 	 */
3032 	p->p_pptr->p_flag |= P_STATCHILD;
3033 	wakeup(p->p_pptr);
3034 
3035 	ps = p->p_pptr->p_sigacts;
3036 	mtx_lock(&ps->ps_mtx);
3037 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3038 		mtx_unlock(&ps->ps_mtx);
3039 		sigparent(p, reason, sig);
3040 	} else
3041 		mtx_unlock(&ps->ps_mtx);
3042 }
3043 
3044 void
3045 childproc_stopped(struct proc *p, int reason)
3046 {
3047 
3048 	childproc_jobstate(p, reason, p->p_xsig);
3049 }
3050 
3051 void
3052 childproc_continued(struct proc *p)
3053 {
3054 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3055 }
3056 
3057 void
3058 childproc_exited(struct proc *p)
3059 {
3060 	int reason, status;
3061 
3062 	if (WCOREDUMP(p->p_xsig)) {
3063 		reason = CLD_DUMPED;
3064 		status = WTERMSIG(p->p_xsig);
3065 	} else if (WIFSIGNALED(p->p_xsig)) {
3066 		reason = CLD_KILLED;
3067 		status = WTERMSIG(p->p_xsig);
3068 	} else {
3069 		reason = CLD_EXITED;
3070 		status = p->p_xexit;
3071 	}
3072 	/*
3073 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3074 	 * done in exit1().
3075 	 */
3076 	sigparent(p, reason, status);
3077 }
3078 
3079 /*
3080  * We only have 1 character for the core count in the format
3081  * string, so the range will be 0-9
3082  */
3083 #define MAX_NUM_CORES 10
3084 static int num_cores = 5;
3085 
3086 static int
3087 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3088 {
3089 	int error;
3090 	int new_val;
3091 
3092 	new_val = num_cores;
3093 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3094 	if (error != 0 || req->newptr == NULL)
3095 		return (error);
3096 	if (new_val > MAX_NUM_CORES)
3097 		new_val = MAX_NUM_CORES;
3098 	if (new_val < 0)
3099 		new_val = 0;
3100 	num_cores = new_val;
3101 	return (0);
3102 }
3103 SYSCTL_PROC(_debug, OID_AUTO, ncores, CTLTYPE_INT|CTLFLAG_RW,
3104 	    0, sizeof(int), sysctl_debug_num_cores_check, "I", "");
3105 
3106 #define	GZ_SUFFIX	".gz"
3107 
3108 #ifdef GZIO
3109 static int compress_user_cores = 1;
3110 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores, CTLFLAG_RWTUN,
3111     &compress_user_cores, 0, "Compression of user corefiles");
3112 
3113 int compress_user_cores_gzlevel = 6;
3114 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_gzlevel, CTLFLAG_RWTUN,
3115     &compress_user_cores_gzlevel, 0, "Corefile gzip compression level");
3116 #else
3117 static int compress_user_cores = 0;
3118 #endif
3119 
3120 /*
3121  * Protect the access to corefilename[] by allproc_lock.
3122  */
3123 #define	corefilename_lock	allproc_lock
3124 
3125 static char corefilename[MAXPATHLEN] = {"%N.core"};
3126 
3127 static int
3128 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3129 {
3130 	int error;
3131 
3132 	sx_xlock(&corefilename_lock);
3133 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3134 	    req);
3135 	sx_xunlock(&corefilename_lock);
3136 
3137 	return (error);
3138 }
3139 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RWTUN |
3140     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3141     "Process corefile name format string");
3142 
3143 /*
3144  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3145  * Expand the name described in corefilename, using name, uid, and pid
3146  * and open/create core file.
3147  * corefilename is a printf-like string, with three format specifiers:
3148  *	%N	name of process ("name")
3149  *	%P	process id (pid)
3150  *	%U	user id (uid)
3151  * For example, "%N.core" is the default; they can be disabled completely
3152  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3153  * This is controlled by the sysctl variable kern.corefile (see above).
3154  */
3155 static int
3156 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3157     int compress, struct vnode **vpp, char **namep)
3158 {
3159 	struct nameidata nd;
3160 	struct sbuf sb;
3161 	const char *format;
3162 	char *hostname, *name;
3163 	int indexpos, i, error, cmode, flags, oflags;
3164 
3165 	hostname = NULL;
3166 	format = corefilename;
3167 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3168 	indexpos = -1;
3169 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3170 	sx_slock(&corefilename_lock);
3171 	for (i = 0; format[i] != '\0'; i++) {
3172 		switch (format[i]) {
3173 		case '%':	/* Format character */
3174 			i++;
3175 			switch (format[i]) {
3176 			case '%':
3177 				sbuf_putc(&sb, '%');
3178 				break;
3179 			case 'H':	/* hostname */
3180 				if (hostname == NULL) {
3181 					hostname = malloc(MAXHOSTNAMELEN,
3182 					    M_TEMP, M_WAITOK);
3183 				}
3184 				getcredhostname(td->td_ucred, hostname,
3185 				    MAXHOSTNAMELEN);
3186 				sbuf_printf(&sb, "%s", hostname);
3187 				break;
3188 			case 'I':	/* autoincrementing index */
3189 				sbuf_printf(&sb, "0");
3190 				indexpos = sbuf_len(&sb) - 1;
3191 				break;
3192 			case 'N':	/* process name */
3193 				sbuf_printf(&sb, "%s", comm);
3194 				break;
3195 			case 'P':	/* process id */
3196 				sbuf_printf(&sb, "%u", pid);
3197 				break;
3198 			case 'U':	/* user id */
3199 				sbuf_printf(&sb, "%u", uid);
3200 				break;
3201 			default:
3202 				log(LOG_ERR,
3203 				    "Unknown format character %c in "
3204 				    "corename `%s'\n", format[i], format);
3205 				break;
3206 			}
3207 			break;
3208 		default:
3209 			sbuf_putc(&sb, format[i]);
3210 			break;
3211 		}
3212 	}
3213 	sx_sunlock(&corefilename_lock);
3214 	free(hostname, M_TEMP);
3215 	if (compress)
3216 		sbuf_printf(&sb, GZ_SUFFIX);
3217 	if (sbuf_error(&sb) != 0) {
3218 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3219 		    "long\n", (long)pid, comm, (u_long)uid);
3220 		sbuf_delete(&sb);
3221 		free(name, M_TEMP);
3222 		return (ENOMEM);
3223 	}
3224 	sbuf_finish(&sb);
3225 	sbuf_delete(&sb);
3226 
3227 	cmode = S_IRUSR | S_IWUSR;
3228 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3229 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3230 
3231 	/*
3232 	 * If the core format has a %I in it, then we need to check
3233 	 * for existing corefiles before returning a name.
3234 	 * To do this we iterate over 0..num_cores to find a
3235 	 * non-existing core file name to use.
3236 	 */
3237 	if (indexpos != -1) {
3238 		for (i = 0; i < num_cores; i++) {
3239 			flags = O_CREAT | O_EXCL | FWRITE | O_NOFOLLOW;
3240 			name[indexpos] = '0' + i;
3241 			NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3242 			error = vn_open_cred(&nd, &flags, cmode, oflags,
3243 			    td->td_ucred, NULL);
3244 			if (error) {
3245 				if (error == EEXIST)
3246 					continue;
3247 				log(LOG_ERR,
3248 				    "pid %d (%s), uid (%u):  Path `%s' failed "
3249 				    "on initial open test, error = %d\n",
3250 				    pid, comm, uid, name, error);
3251 			}
3252 			goto out;
3253 		}
3254 	}
3255 
3256 	flags = O_CREAT | FWRITE | O_NOFOLLOW;
3257 	NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3258 	error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred, NULL);
3259 out:
3260 	if (error) {
3261 #ifdef AUDIT
3262 		audit_proc_coredump(td, name, error);
3263 #endif
3264 		free(name, M_TEMP);
3265 		return (error);
3266 	}
3267 	NDFREE(&nd, NDF_ONLY_PNBUF);
3268 	*vpp = nd.ni_vp;
3269 	*namep = name;
3270 	return (0);
3271 }
3272 
3273 static int
3274 coredump_sanitise_path(const char *path)
3275 {
3276 	size_t i;
3277 
3278 	/*
3279 	 * Only send a subset of ASCII to devd(8) because it
3280 	 * might pass these strings to sh -c.
3281 	 */
3282 	for (i = 0; path[i]; i++)
3283 		if (!(isalpha(path[i]) || isdigit(path[i])) &&
3284 		    path[i] != '/' && path[i] != '.' &&
3285 		    path[i] != '-')
3286 			return (0);
3287 
3288 	return (1);
3289 }
3290 
3291 /*
3292  * Dump a process' core.  The main routine does some
3293  * policy checking, and creates the name of the coredump;
3294  * then it passes on a vnode and a size limit to the process-specific
3295  * coredump routine if there is one; if there _is not_ one, it returns
3296  * ENOSYS; otherwise it returns the error from the process-specific routine.
3297  */
3298 
3299 static int
3300 coredump(struct thread *td)
3301 {
3302 	struct proc *p = td->td_proc;
3303 	struct ucred *cred = td->td_ucred;
3304 	struct vnode *vp;
3305 	struct flock lf;
3306 	struct vattr vattr;
3307 	int error, error1, locked;
3308 	char *name;			/* name of corefile */
3309 	void *rl_cookie;
3310 	off_t limit;
3311 	char *data = NULL;
3312 	char *fullpath, *freepath = NULL;
3313 	size_t len;
3314 	static const char comm_name[] = "comm=";
3315 	static const char core_name[] = "core=";
3316 
3317 	PROC_LOCK_ASSERT(p, MA_OWNED);
3318 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3319 	_STOPEVENT(p, S_CORE, 0);
3320 
3321 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
3322 	    (p->p_flag2 & P2_NOTRACE) != 0) {
3323 		PROC_UNLOCK(p);
3324 		return (EFAULT);
3325 	}
3326 
3327 	/*
3328 	 * Note that the bulk of limit checking is done after
3329 	 * the corefile is created.  The exception is if the limit
3330 	 * for corefiles is 0, in which case we don't bother
3331 	 * creating the corefile at all.  This layout means that
3332 	 * a corefile is truncated instead of not being created,
3333 	 * if it is larger than the limit.
3334 	 */
3335 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
3336 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
3337 		PROC_UNLOCK(p);
3338 		return (EFBIG);
3339 	}
3340 	PROC_UNLOCK(p);
3341 
3342 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
3343 	    compress_user_cores, &vp, &name);
3344 	if (error != 0)
3345 		return (error);
3346 
3347 	/*
3348 	 * Don't dump to non-regular files or files with links.
3349 	 * Do not dump into system files.
3350 	 */
3351 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
3352 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0) {
3353 		VOP_UNLOCK(vp, 0);
3354 		error = EFAULT;
3355 		goto out;
3356 	}
3357 
3358 	VOP_UNLOCK(vp, 0);
3359 
3360 	/* Postpone other writers, including core dumps of other processes. */
3361 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
3362 
3363 	lf.l_whence = SEEK_SET;
3364 	lf.l_start = 0;
3365 	lf.l_len = 0;
3366 	lf.l_type = F_WRLCK;
3367 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
3368 
3369 	VATTR_NULL(&vattr);
3370 	vattr.va_size = 0;
3371 	if (set_core_nodump_flag)
3372 		vattr.va_flags = UF_NODUMP;
3373 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3374 	VOP_SETATTR(vp, &vattr, cred);
3375 	VOP_UNLOCK(vp, 0);
3376 	PROC_LOCK(p);
3377 	p->p_acflag |= ACORE;
3378 	PROC_UNLOCK(p);
3379 
3380 	if (p->p_sysent->sv_coredump != NULL) {
3381 		error = p->p_sysent->sv_coredump(td, vp, limit,
3382 		    compress_user_cores ? IMGACT_CORE_COMPRESS : 0);
3383 	} else {
3384 		error = ENOSYS;
3385 	}
3386 
3387 	if (locked) {
3388 		lf.l_type = F_UNLCK;
3389 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3390 	}
3391 	vn_rangelock_unlock(vp, rl_cookie);
3392 
3393 	/*
3394 	 * Notify the userland helper that a process triggered a core dump.
3395 	 * This allows the helper to run an automated debugging session.
3396 	 */
3397 	if (error != 0 || coredump_devctl == 0)
3398 		goto out;
3399 	len = MAXPATHLEN * 2 + sizeof(comm_name) - 1 +
3400 	    sizeof(' ') + sizeof(core_name) - 1;
3401 	data = malloc(len, M_TEMP, M_WAITOK);
3402 	if (vn_fullpath_global(td, p->p_textvp, &fullpath, &freepath) != 0)
3403 		goto out;
3404 	if (!coredump_sanitise_path(fullpath))
3405 		goto out;
3406 	snprintf(data, len, "%s%s ", comm_name, fullpath);
3407 	free(freepath, M_TEMP);
3408 	freepath = NULL;
3409 	if (vn_fullpath_global(td, vp, &fullpath, &freepath) != 0)
3410 		goto out;
3411 	if (!coredump_sanitise_path(fullpath))
3412 		goto out;
3413 	strlcat(data, core_name, len);
3414 	strlcat(data, fullpath, len);
3415 	devctl_notify("kernel", "signal", "coredump", data);
3416 out:
3417 	error1 = vn_close(vp, FWRITE, cred, td);
3418 	if (error == 0)
3419 		error = error1;
3420 #ifdef AUDIT
3421 	audit_proc_coredump(td, name, error);
3422 #endif
3423 	free(freepath, M_TEMP);
3424 	free(data, M_TEMP);
3425 	free(name, M_TEMP);
3426 	return (error);
3427 }
3428 
3429 /*
3430  * Nonexistent system call-- signal process (may want to handle it).  Flag
3431  * error in case process won't see signal immediately (blocked or ignored).
3432  */
3433 #ifndef _SYS_SYSPROTO_H_
3434 struct nosys_args {
3435 	int	dummy;
3436 };
3437 #endif
3438 /* ARGSUSED */
3439 int
3440 nosys(td, args)
3441 	struct thread *td;
3442 	struct nosys_args *args;
3443 {
3444 	struct proc *p = td->td_proc;
3445 
3446 	PROC_LOCK(p);
3447 	tdsignal(td, SIGSYS);
3448 	PROC_UNLOCK(p);
3449 	return (ENOSYS);
3450 }
3451 
3452 /*
3453  * Send a SIGIO or SIGURG signal to a process or process group using stored
3454  * credentials rather than those of the current process.
3455  */
3456 void
3457 pgsigio(sigiop, sig, checkctty)
3458 	struct sigio **sigiop;
3459 	int sig, checkctty;
3460 {
3461 	ksiginfo_t ksi;
3462 	struct sigio *sigio;
3463 
3464 	ksiginfo_init(&ksi);
3465 	ksi.ksi_signo = sig;
3466 	ksi.ksi_code = SI_KERNEL;
3467 
3468 	SIGIO_LOCK();
3469 	sigio = *sigiop;
3470 	if (sigio == NULL) {
3471 		SIGIO_UNLOCK();
3472 		return;
3473 	}
3474 	if (sigio->sio_pgid > 0) {
3475 		PROC_LOCK(sigio->sio_proc);
3476 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
3477 			kern_psignal(sigio->sio_proc, sig);
3478 		PROC_UNLOCK(sigio->sio_proc);
3479 	} else if (sigio->sio_pgid < 0) {
3480 		struct proc *p;
3481 
3482 		PGRP_LOCK(sigio->sio_pgrp);
3483 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
3484 			PROC_LOCK(p);
3485 			if (p->p_state == PRS_NORMAL &&
3486 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
3487 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
3488 				kern_psignal(p, sig);
3489 			PROC_UNLOCK(p);
3490 		}
3491 		PGRP_UNLOCK(sigio->sio_pgrp);
3492 	}
3493 	SIGIO_UNLOCK();
3494 }
3495 
3496 static int
3497 filt_sigattach(struct knote *kn)
3498 {
3499 	struct proc *p = curproc;
3500 
3501 	kn->kn_ptr.p_proc = p;
3502 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
3503 
3504 	knlist_add(&p->p_klist, kn, 0);
3505 
3506 	return (0);
3507 }
3508 
3509 static void
3510 filt_sigdetach(struct knote *kn)
3511 {
3512 	struct proc *p = kn->kn_ptr.p_proc;
3513 
3514 	knlist_remove(&p->p_klist, kn, 0);
3515 }
3516 
3517 /*
3518  * signal knotes are shared with proc knotes, so we apply a mask to
3519  * the hint in order to differentiate them from process hints.  This
3520  * could be avoided by using a signal-specific knote list, but probably
3521  * isn't worth the trouble.
3522  */
3523 static int
3524 filt_signal(struct knote *kn, long hint)
3525 {
3526 
3527 	if (hint & NOTE_SIGNAL) {
3528 		hint &= ~NOTE_SIGNAL;
3529 
3530 		if (kn->kn_id == hint)
3531 			kn->kn_data++;
3532 	}
3533 	return (kn->kn_data != 0);
3534 }
3535 
3536 struct sigacts *
3537 sigacts_alloc(void)
3538 {
3539 	struct sigacts *ps;
3540 
3541 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
3542 	refcount_init(&ps->ps_refcnt, 1);
3543 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
3544 	return (ps);
3545 }
3546 
3547 void
3548 sigacts_free(struct sigacts *ps)
3549 {
3550 
3551 	if (refcount_release(&ps->ps_refcnt) == 0)
3552 		return;
3553 	mtx_destroy(&ps->ps_mtx);
3554 	free(ps, M_SUBPROC);
3555 }
3556 
3557 struct sigacts *
3558 sigacts_hold(struct sigacts *ps)
3559 {
3560 
3561 	refcount_acquire(&ps->ps_refcnt);
3562 	return (ps);
3563 }
3564 
3565 void
3566 sigacts_copy(struct sigacts *dest, struct sigacts *src)
3567 {
3568 
3569 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
3570 	mtx_lock(&src->ps_mtx);
3571 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
3572 	mtx_unlock(&src->ps_mtx);
3573 }
3574 
3575 int
3576 sigacts_shared(struct sigacts *ps)
3577 {
3578 
3579 	return (ps->ps_refcnt > 1);
3580 }
3581