xref: /freebsd/sys/kern/kern_sig.c (revision fcb560670601b2a4d87bb31d7531c8dcc37ee71b)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_compat.h"
41 #include "opt_ktrace.h"
42 #include "opt_core.h"
43 
44 #include <sys/param.h>
45 #include <sys/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, kernel, , signal__send, "struct thread *",
98     "struct proc *", "int");
99 SDT_PROBE_DEFINE2(proc, kernel, , signal__clear, "int",
100     "ksiginfo_t *");
101 SDT_PROBE_DEFINE3(proc, kernel, , 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 = 1;
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(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(td, sig, act, oact, flags)
652 	struct thread *td;
653 	register int sig;
654 	struct sigaction *act, *oact;
655 	int flags;
656 {
657 	struct sigacts *ps;
658 	struct proc *p = td->td_proc;
659 
660 	if (!_SIG_VALID(sig))
661 		return (EINVAL);
662 	if (act != NULL && act->sa_handler != SIG_DFL &&
663 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
664 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
665 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
666 		return (EINVAL);
667 
668 	PROC_LOCK(p);
669 	ps = p->p_sigacts;
670 	mtx_lock(&ps->ps_mtx);
671 	if (oact) {
672 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
673 		oact->sa_flags = 0;
674 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
675 			oact->sa_flags |= SA_ONSTACK;
676 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
677 			oact->sa_flags |= SA_RESTART;
678 		if (SIGISMEMBER(ps->ps_sigreset, sig))
679 			oact->sa_flags |= SA_RESETHAND;
680 		if (SIGISMEMBER(ps->ps_signodefer, sig))
681 			oact->sa_flags |= SA_NODEFER;
682 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
683 			oact->sa_flags |= SA_SIGINFO;
684 			oact->sa_sigaction =
685 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
686 		} else
687 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
688 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
689 			oact->sa_flags |= SA_NOCLDSTOP;
690 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
691 			oact->sa_flags |= SA_NOCLDWAIT;
692 	}
693 	if (act) {
694 		if ((sig == SIGKILL || sig == SIGSTOP) &&
695 		    act->sa_handler != SIG_DFL) {
696 			mtx_unlock(&ps->ps_mtx);
697 			PROC_UNLOCK(p);
698 			return (EINVAL);
699 		}
700 
701 		/*
702 		 * Change setting atomically.
703 		 */
704 
705 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
706 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
707 		if (sigact_flag_test(act, SA_SIGINFO)) {
708 			ps->ps_sigact[_SIG_IDX(sig)] =
709 			    (__sighandler_t *)act->sa_sigaction;
710 			SIGADDSET(ps->ps_siginfo, sig);
711 		} else {
712 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
713 			SIGDELSET(ps->ps_siginfo, sig);
714 		}
715 		if (!sigact_flag_test(act, SA_RESTART))
716 			SIGADDSET(ps->ps_sigintr, sig);
717 		else
718 			SIGDELSET(ps->ps_sigintr, sig);
719 		if (sigact_flag_test(act, SA_ONSTACK))
720 			SIGADDSET(ps->ps_sigonstack, sig);
721 		else
722 			SIGDELSET(ps->ps_sigonstack, sig);
723 		if (sigact_flag_test(act, SA_RESETHAND))
724 			SIGADDSET(ps->ps_sigreset, sig);
725 		else
726 			SIGDELSET(ps->ps_sigreset, sig);
727 		if (sigact_flag_test(act, SA_NODEFER))
728 			SIGADDSET(ps->ps_signodefer, sig);
729 		else
730 			SIGDELSET(ps->ps_signodefer, sig);
731 		if (sig == SIGCHLD) {
732 			if (act->sa_flags & SA_NOCLDSTOP)
733 				ps->ps_flag |= PS_NOCLDSTOP;
734 			else
735 				ps->ps_flag &= ~PS_NOCLDSTOP;
736 			if (act->sa_flags & SA_NOCLDWAIT) {
737 				/*
738 				 * Paranoia: since SA_NOCLDWAIT is implemented
739 				 * by reparenting the dying child to PID 1 (and
740 				 * trust it to reap the zombie), PID 1 itself
741 				 * is forbidden to set SA_NOCLDWAIT.
742 				 */
743 				if (p->p_pid == 1)
744 					ps->ps_flag &= ~PS_NOCLDWAIT;
745 				else
746 					ps->ps_flag |= PS_NOCLDWAIT;
747 			} else
748 				ps->ps_flag &= ~PS_NOCLDWAIT;
749 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
750 				ps->ps_flag |= PS_CLDSIGIGN;
751 			else
752 				ps->ps_flag &= ~PS_CLDSIGIGN;
753 		}
754 		/*
755 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
756 		 * and for signals set to SIG_DFL where the default is to
757 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
758 		 * have to restart the process.
759 		 */
760 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
761 		    (sigprop(sig) & SA_IGNORE &&
762 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
763 			/* never to be seen again */
764 			sigqueue_delete_proc(p, sig);
765 			if (sig != SIGCONT)
766 				/* easier in psignal */
767 				SIGADDSET(ps->ps_sigignore, sig);
768 			SIGDELSET(ps->ps_sigcatch, sig);
769 		} else {
770 			SIGDELSET(ps->ps_sigignore, sig);
771 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
772 				SIGDELSET(ps->ps_sigcatch, sig);
773 			else
774 				SIGADDSET(ps->ps_sigcatch, sig);
775 		}
776 #ifdef COMPAT_FREEBSD4
777 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
778 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
779 		    (flags & KSA_FREEBSD4) == 0)
780 			SIGDELSET(ps->ps_freebsd4, sig);
781 		else
782 			SIGADDSET(ps->ps_freebsd4, sig);
783 #endif
784 #ifdef COMPAT_43
785 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
786 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
787 		    (flags & KSA_OSIGSET) == 0)
788 			SIGDELSET(ps->ps_osigset, sig);
789 		else
790 			SIGADDSET(ps->ps_osigset, sig);
791 #endif
792 	}
793 	mtx_unlock(&ps->ps_mtx);
794 	PROC_UNLOCK(p);
795 	return (0);
796 }
797 
798 #ifndef _SYS_SYSPROTO_H_
799 struct sigaction_args {
800 	int	sig;
801 	struct	sigaction *act;
802 	struct	sigaction *oact;
803 };
804 #endif
805 int
806 sys_sigaction(td, uap)
807 	struct thread *td;
808 	register struct sigaction_args *uap;
809 {
810 	struct sigaction act, oact;
811 	register struct sigaction *actp, *oactp;
812 	int error;
813 
814 	actp = (uap->act != NULL) ? &act : NULL;
815 	oactp = (uap->oact != NULL) ? &oact : NULL;
816 	if (actp) {
817 		error = copyin(uap->act, actp, sizeof(act));
818 		if (error)
819 			return (error);
820 	}
821 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
822 	if (oactp && !error)
823 		error = copyout(oactp, uap->oact, sizeof(oact));
824 	return (error);
825 }
826 
827 #ifdef COMPAT_FREEBSD4
828 #ifndef _SYS_SYSPROTO_H_
829 struct freebsd4_sigaction_args {
830 	int	sig;
831 	struct	sigaction *act;
832 	struct	sigaction *oact;
833 };
834 #endif
835 int
836 freebsd4_sigaction(td, uap)
837 	struct thread *td;
838 	register struct freebsd4_sigaction_args *uap;
839 {
840 	struct sigaction act, oact;
841 	register struct sigaction *actp, *oactp;
842 	int error;
843 
844 
845 	actp = (uap->act != NULL) ? &act : NULL;
846 	oactp = (uap->oact != NULL) ? &oact : NULL;
847 	if (actp) {
848 		error = copyin(uap->act, actp, sizeof(act));
849 		if (error)
850 			return (error);
851 	}
852 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
853 	if (oactp && !error)
854 		error = copyout(oactp, uap->oact, sizeof(oact));
855 	return (error);
856 }
857 #endif	/* COMAPT_FREEBSD4 */
858 
859 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
860 #ifndef _SYS_SYSPROTO_H_
861 struct osigaction_args {
862 	int	signum;
863 	struct	osigaction *nsa;
864 	struct	osigaction *osa;
865 };
866 #endif
867 int
868 osigaction(td, uap)
869 	struct thread *td;
870 	register struct osigaction_args *uap;
871 {
872 	struct osigaction sa;
873 	struct sigaction nsa, osa;
874 	register struct sigaction *nsap, *osap;
875 	int error;
876 
877 	if (uap->signum <= 0 || uap->signum >= ONSIG)
878 		return (EINVAL);
879 
880 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
881 	osap = (uap->osa != NULL) ? &osa : NULL;
882 
883 	if (nsap) {
884 		error = copyin(uap->nsa, &sa, sizeof(sa));
885 		if (error)
886 			return (error);
887 		nsap->sa_handler = sa.sa_handler;
888 		nsap->sa_flags = sa.sa_flags;
889 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
890 	}
891 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
892 	if (osap && !error) {
893 		sa.sa_handler = osap->sa_handler;
894 		sa.sa_flags = osap->sa_flags;
895 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
896 		error = copyout(&sa, uap->osa, sizeof(sa));
897 	}
898 	return (error);
899 }
900 
901 #if !defined(__i386__)
902 /* Avoid replicating the same stub everywhere */
903 int
904 osigreturn(td, uap)
905 	struct thread *td;
906 	struct osigreturn_args *uap;
907 {
908 
909 	return (nosys(td, (struct nosys_args *)uap));
910 }
911 #endif
912 #endif /* COMPAT_43 */
913 
914 /*
915  * Initialize signal state for process 0;
916  * set to ignore signals that are ignored by default.
917  */
918 void
919 siginit(p)
920 	struct proc *p;
921 {
922 	register int i;
923 	struct sigacts *ps;
924 
925 	PROC_LOCK(p);
926 	ps = p->p_sigacts;
927 	mtx_lock(&ps->ps_mtx);
928 	for (i = 1; i <= NSIG; i++) {
929 		if (sigprop(i) & SA_IGNORE && i != SIGCONT) {
930 			SIGADDSET(ps->ps_sigignore, i);
931 		}
932 	}
933 	mtx_unlock(&ps->ps_mtx);
934 	PROC_UNLOCK(p);
935 }
936 
937 /*
938  * Reset specified signal to the default disposition.
939  */
940 static void
941 sigdflt(struct sigacts *ps, int sig)
942 {
943 
944 	mtx_assert(&ps->ps_mtx, MA_OWNED);
945 	SIGDELSET(ps->ps_sigcatch, sig);
946 	if ((sigprop(sig) & SA_IGNORE) != 0 && sig != SIGCONT)
947 		SIGADDSET(ps->ps_sigignore, sig);
948 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
949 	SIGDELSET(ps->ps_siginfo, sig);
950 }
951 
952 /*
953  * Reset signals for an exec of the specified process.
954  */
955 void
956 execsigs(struct proc *p)
957 {
958 	struct sigacts *ps;
959 	int sig;
960 	struct thread *td;
961 
962 	/*
963 	 * Reset caught signals.  Held signals remain held
964 	 * through td_sigmask (unless they were caught,
965 	 * and are now ignored by default).
966 	 */
967 	PROC_LOCK_ASSERT(p, MA_OWNED);
968 	td = FIRST_THREAD_IN_PROC(p);
969 	ps = p->p_sigacts;
970 	mtx_lock(&ps->ps_mtx);
971 	while (SIGNOTEMPTY(ps->ps_sigcatch)) {
972 		sig = sig_ffs(&ps->ps_sigcatch);
973 		sigdflt(ps, sig);
974 		if ((sigprop(sig) & SA_IGNORE) != 0)
975 			sigqueue_delete_proc(p, sig);
976 	}
977 	/*
978 	 * Reset stack state to the user stack.
979 	 * Clear set of signals caught on the signal stack.
980 	 */
981 	td->td_sigstk.ss_flags = SS_DISABLE;
982 	td->td_sigstk.ss_size = 0;
983 	td->td_sigstk.ss_sp = 0;
984 	td->td_pflags &= ~TDP_ALTSTACK;
985 	/*
986 	 * Reset no zombies if child dies flag as Solaris does.
987 	 */
988 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
989 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
990 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
991 	mtx_unlock(&ps->ps_mtx);
992 }
993 
994 /*
995  * kern_sigprocmask()
996  *
997  *	Manipulate signal mask.
998  */
999 int
1000 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1001     int flags)
1002 {
1003 	sigset_t new_block, oset1;
1004 	struct proc *p;
1005 	int error;
1006 
1007 	p = td->td_proc;
1008 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1009 		PROC_LOCK_ASSERT(p, MA_OWNED);
1010 	else
1011 		PROC_LOCK(p);
1012 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1013 	    ? MA_OWNED : MA_NOTOWNED);
1014 	if (oset != NULL)
1015 		*oset = td->td_sigmask;
1016 
1017 	error = 0;
1018 	if (set != NULL) {
1019 		switch (how) {
1020 		case SIG_BLOCK:
1021 			SIG_CANTMASK(*set);
1022 			oset1 = td->td_sigmask;
1023 			SIGSETOR(td->td_sigmask, *set);
1024 			new_block = td->td_sigmask;
1025 			SIGSETNAND(new_block, oset1);
1026 			break;
1027 		case SIG_UNBLOCK:
1028 			SIGSETNAND(td->td_sigmask, *set);
1029 			signotify(td);
1030 			goto out;
1031 		case SIG_SETMASK:
1032 			SIG_CANTMASK(*set);
1033 			oset1 = td->td_sigmask;
1034 			if (flags & SIGPROCMASK_OLD)
1035 				SIGSETLO(td->td_sigmask, *set);
1036 			else
1037 				td->td_sigmask = *set;
1038 			new_block = td->td_sigmask;
1039 			SIGSETNAND(new_block, oset1);
1040 			signotify(td);
1041 			break;
1042 		default:
1043 			error = EINVAL;
1044 			goto out;
1045 		}
1046 
1047 		/*
1048 		 * The new_block set contains signals that were not previously
1049 		 * blocked, but are blocked now.
1050 		 *
1051 		 * In case we block any signal that was not previously blocked
1052 		 * for td, and process has the signal pending, try to schedule
1053 		 * signal delivery to some thread that does not block the
1054 		 * signal, possibly waking it up.
1055 		 */
1056 		if (p->p_numthreads != 1)
1057 			reschedule_signals(p, new_block, flags);
1058 	}
1059 
1060 out:
1061 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1062 		PROC_UNLOCK(p);
1063 	return (error);
1064 }
1065 
1066 #ifndef _SYS_SYSPROTO_H_
1067 struct sigprocmask_args {
1068 	int	how;
1069 	const sigset_t *set;
1070 	sigset_t *oset;
1071 };
1072 #endif
1073 int
1074 sys_sigprocmask(td, uap)
1075 	register struct thread *td;
1076 	struct sigprocmask_args *uap;
1077 {
1078 	sigset_t set, oset;
1079 	sigset_t *setp, *osetp;
1080 	int error;
1081 
1082 	setp = (uap->set != NULL) ? &set : NULL;
1083 	osetp = (uap->oset != NULL) ? &oset : NULL;
1084 	if (setp) {
1085 		error = copyin(uap->set, setp, sizeof(set));
1086 		if (error)
1087 			return (error);
1088 	}
1089 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1090 	if (osetp && !error) {
1091 		error = copyout(osetp, uap->oset, sizeof(oset));
1092 	}
1093 	return (error);
1094 }
1095 
1096 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1097 #ifndef _SYS_SYSPROTO_H_
1098 struct osigprocmask_args {
1099 	int	how;
1100 	osigset_t mask;
1101 };
1102 #endif
1103 int
1104 osigprocmask(td, uap)
1105 	register struct thread *td;
1106 	struct osigprocmask_args *uap;
1107 {
1108 	sigset_t set, oset;
1109 	int error;
1110 
1111 	OSIG2SIG(uap->mask, set);
1112 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1113 	SIG2OSIG(oset, td->td_retval[0]);
1114 	return (error);
1115 }
1116 #endif /* COMPAT_43 */
1117 
1118 int
1119 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1120 {
1121 	ksiginfo_t ksi;
1122 	sigset_t set;
1123 	int error;
1124 
1125 	error = copyin(uap->set, &set, sizeof(set));
1126 	if (error) {
1127 		td->td_retval[0] = error;
1128 		return (0);
1129 	}
1130 
1131 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1132 	if (error) {
1133 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1134 			error = ERESTART;
1135 		if (error == ERESTART)
1136 			return (error);
1137 		td->td_retval[0] = error;
1138 		return (0);
1139 	}
1140 
1141 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1142 	td->td_retval[0] = error;
1143 	return (0);
1144 }
1145 
1146 int
1147 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1148 {
1149 	struct timespec ts;
1150 	struct timespec *timeout;
1151 	sigset_t set;
1152 	ksiginfo_t ksi;
1153 	int error;
1154 
1155 	if (uap->timeout) {
1156 		error = copyin(uap->timeout, &ts, sizeof(ts));
1157 		if (error)
1158 			return (error);
1159 
1160 		timeout = &ts;
1161 	} else
1162 		timeout = NULL;
1163 
1164 	error = copyin(uap->set, &set, sizeof(set));
1165 	if (error)
1166 		return (error);
1167 
1168 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1169 	if (error)
1170 		return (error);
1171 
1172 	if (uap->info)
1173 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1174 
1175 	if (error == 0)
1176 		td->td_retval[0] = ksi.ksi_signo;
1177 	return (error);
1178 }
1179 
1180 int
1181 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1182 {
1183 	ksiginfo_t ksi;
1184 	sigset_t set;
1185 	int error;
1186 
1187 	error = copyin(uap->set, &set, sizeof(set));
1188 	if (error)
1189 		return (error);
1190 
1191 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1192 	if (error)
1193 		return (error);
1194 
1195 	if (uap->info)
1196 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1197 
1198 	if (error == 0)
1199 		td->td_retval[0] = ksi.ksi_signo;
1200 	return (error);
1201 }
1202 
1203 int
1204 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1205 	struct timespec *timeout)
1206 {
1207 	struct sigacts *ps;
1208 	sigset_t saved_mask, new_block;
1209 	struct proc *p;
1210 	int error, sig, timo, timevalid = 0;
1211 	struct timespec rts, ets, ts;
1212 	struct timeval tv;
1213 
1214 	p = td->td_proc;
1215 	error = 0;
1216 	ets.tv_sec = 0;
1217 	ets.tv_nsec = 0;
1218 
1219 	if (timeout != NULL) {
1220 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1221 			timevalid = 1;
1222 			getnanouptime(&rts);
1223 			ets = rts;
1224 			timespecadd(&ets, timeout);
1225 		}
1226 	}
1227 	ksiginfo_init(ksi);
1228 	/* Some signals can not be waited for. */
1229 	SIG_CANTMASK(waitset);
1230 	ps = p->p_sigacts;
1231 	PROC_LOCK(p);
1232 	saved_mask = td->td_sigmask;
1233 	SIGSETNAND(td->td_sigmask, waitset);
1234 	for (;;) {
1235 		mtx_lock(&ps->ps_mtx);
1236 		sig = cursig(td);
1237 		mtx_unlock(&ps->ps_mtx);
1238 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1239 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1240 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1241 				error = 0;
1242 				break;
1243 			}
1244 		}
1245 
1246 		if (error != 0)
1247 			break;
1248 
1249 		/*
1250 		 * POSIX says this must be checked after looking for pending
1251 		 * signals.
1252 		 */
1253 		if (timeout != NULL) {
1254 			if (!timevalid) {
1255 				error = EINVAL;
1256 				break;
1257 			}
1258 			getnanouptime(&rts);
1259 			if (timespeccmp(&rts, &ets, >=)) {
1260 				error = EAGAIN;
1261 				break;
1262 			}
1263 			ts = ets;
1264 			timespecsub(&ts, &rts);
1265 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1266 			timo = tvtohz(&tv);
1267 		} else {
1268 			timo = 0;
1269 		}
1270 
1271 		error = msleep(ps, &p->p_mtx, PPAUSE|PCATCH, "sigwait", timo);
1272 
1273 		if (timeout != NULL) {
1274 			if (error == ERESTART) {
1275 				/* Timeout can not be restarted. */
1276 				error = EINTR;
1277 			} else if (error == EAGAIN) {
1278 				/* We will calculate timeout by ourself. */
1279 				error = 0;
1280 			}
1281 		}
1282 	}
1283 
1284 	new_block = saved_mask;
1285 	SIGSETNAND(new_block, td->td_sigmask);
1286 	td->td_sigmask = saved_mask;
1287 	/*
1288 	 * Fewer signals can be delivered to us, reschedule signal
1289 	 * notification.
1290 	 */
1291 	if (p->p_numthreads != 1)
1292 		reschedule_signals(p, new_block, 0);
1293 
1294 	if (error == 0) {
1295 		SDT_PROBE(proc, kernel, , signal__clear, sig, ksi, 0, 0, 0);
1296 
1297 		if (ksi->ksi_code == SI_TIMER)
1298 			itimer_accept(p, ksi->ksi_timerid, ksi);
1299 
1300 #ifdef KTRACE
1301 		if (KTRPOINT(td, KTR_PSIG)) {
1302 			sig_t action;
1303 
1304 			mtx_lock(&ps->ps_mtx);
1305 			action = ps->ps_sigact[_SIG_IDX(sig)];
1306 			mtx_unlock(&ps->ps_mtx);
1307 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1308 		}
1309 #endif
1310 		if (sig == SIGKILL)
1311 			sigexit(td, sig);
1312 	}
1313 	PROC_UNLOCK(p);
1314 	return (error);
1315 }
1316 
1317 #ifndef _SYS_SYSPROTO_H_
1318 struct sigpending_args {
1319 	sigset_t	*set;
1320 };
1321 #endif
1322 int
1323 sys_sigpending(td, uap)
1324 	struct thread *td;
1325 	struct sigpending_args *uap;
1326 {
1327 	struct proc *p = td->td_proc;
1328 	sigset_t pending;
1329 
1330 	PROC_LOCK(p);
1331 	pending = p->p_sigqueue.sq_signals;
1332 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1333 	PROC_UNLOCK(p);
1334 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1335 }
1336 
1337 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1338 #ifndef _SYS_SYSPROTO_H_
1339 struct osigpending_args {
1340 	int	dummy;
1341 };
1342 #endif
1343 int
1344 osigpending(td, uap)
1345 	struct thread *td;
1346 	struct osigpending_args *uap;
1347 {
1348 	struct proc *p = td->td_proc;
1349 	sigset_t pending;
1350 
1351 	PROC_LOCK(p);
1352 	pending = p->p_sigqueue.sq_signals;
1353 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1354 	PROC_UNLOCK(p);
1355 	SIG2OSIG(pending, td->td_retval[0]);
1356 	return (0);
1357 }
1358 #endif /* COMPAT_43 */
1359 
1360 #if defined(COMPAT_43)
1361 /*
1362  * Generalized interface signal handler, 4.3-compatible.
1363  */
1364 #ifndef _SYS_SYSPROTO_H_
1365 struct osigvec_args {
1366 	int	signum;
1367 	struct	sigvec *nsv;
1368 	struct	sigvec *osv;
1369 };
1370 #endif
1371 /* ARGSUSED */
1372 int
1373 osigvec(td, uap)
1374 	struct thread *td;
1375 	register struct osigvec_args *uap;
1376 {
1377 	struct sigvec vec;
1378 	struct sigaction nsa, osa;
1379 	register struct sigaction *nsap, *osap;
1380 	int error;
1381 
1382 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1383 		return (EINVAL);
1384 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1385 	osap = (uap->osv != NULL) ? &osa : NULL;
1386 	if (nsap) {
1387 		error = copyin(uap->nsv, &vec, sizeof(vec));
1388 		if (error)
1389 			return (error);
1390 		nsap->sa_handler = vec.sv_handler;
1391 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1392 		nsap->sa_flags = vec.sv_flags;
1393 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1394 	}
1395 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1396 	if (osap && !error) {
1397 		vec.sv_handler = osap->sa_handler;
1398 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1399 		vec.sv_flags = osap->sa_flags;
1400 		vec.sv_flags &= ~SA_NOCLDWAIT;
1401 		vec.sv_flags ^= SA_RESTART;
1402 		error = copyout(&vec, uap->osv, sizeof(vec));
1403 	}
1404 	return (error);
1405 }
1406 
1407 #ifndef _SYS_SYSPROTO_H_
1408 struct osigblock_args {
1409 	int	mask;
1410 };
1411 #endif
1412 int
1413 osigblock(td, uap)
1414 	register struct thread *td;
1415 	struct osigblock_args *uap;
1416 {
1417 	sigset_t set, oset;
1418 
1419 	OSIG2SIG(uap->mask, set);
1420 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1421 	SIG2OSIG(oset, td->td_retval[0]);
1422 	return (0);
1423 }
1424 
1425 #ifndef _SYS_SYSPROTO_H_
1426 struct osigsetmask_args {
1427 	int	mask;
1428 };
1429 #endif
1430 int
1431 osigsetmask(td, uap)
1432 	struct thread *td;
1433 	struct osigsetmask_args *uap;
1434 {
1435 	sigset_t set, oset;
1436 
1437 	OSIG2SIG(uap->mask, set);
1438 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1439 	SIG2OSIG(oset, td->td_retval[0]);
1440 	return (0);
1441 }
1442 #endif /* COMPAT_43 */
1443 
1444 /*
1445  * Suspend calling thread until signal, providing mask to be set in the
1446  * meantime.
1447  */
1448 #ifndef _SYS_SYSPROTO_H_
1449 struct sigsuspend_args {
1450 	const sigset_t *sigmask;
1451 };
1452 #endif
1453 /* ARGSUSED */
1454 int
1455 sys_sigsuspend(td, uap)
1456 	struct thread *td;
1457 	struct sigsuspend_args *uap;
1458 {
1459 	sigset_t mask;
1460 	int error;
1461 
1462 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1463 	if (error)
1464 		return (error);
1465 	return (kern_sigsuspend(td, mask));
1466 }
1467 
1468 int
1469 kern_sigsuspend(struct thread *td, sigset_t mask)
1470 {
1471 	struct proc *p = td->td_proc;
1472 	int has_sig, sig;
1473 
1474 	/*
1475 	 * When returning from sigsuspend, we want
1476 	 * the old mask to be restored after the
1477 	 * signal handler has finished.  Thus, we
1478 	 * save it here and mark the sigacts structure
1479 	 * to indicate this.
1480 	 */
1481 	PROC_LOCK(p);
1482 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1483 	    SIGPROCMASK_PROC_LOCKED);
1484 	td->td_pflags |= TDP_OLDMASK;
1485 
1486 	/*
1487 	 * Process signals now. Otherwise, we can get spurious wakeup
1488 	 * due to signal entered process queue, but delivered to other
1489 	 * thread. But sigsuspend should return only on signal
1490 	 * delivery.
1491 	 */
1492 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1493 	for (has_sig = 0; !has_sig;) {
1494 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1495 			0) == 0)
1496 			/* void */;
1497 		thread_suspend_check(0);
1498 		mtx_lock(&p->p_sigacts->ps_mtx);
1499 		while ((sig = cursig(td)) != 0)
1500 			has_sig += postsig(sig);
1501 		mtx_unlock(&p->p_sigacts->ps_mtx);
1502 	}
1503 	PROC_UNLOCK(p);
1504 	td->td_errno = EINTR;
1505 	td->td_pflags |= TDP_NERRNO;
1506 	return (EJUSTRETURN);
1507 }
1508 
1509 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1510 /*
1511  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1512  * convention: libc stub passes mask, not pointer, to save a copyin.
1513  */
1514 #ifndef _SYS_SYSPROTO_H_
1515 struct osigsuspend_args {
1516 	osigset_t mask;
1517 };
1518 #endif
1519 /* ARGSUSED */
1520 int
1521 osigsuspend(td, uap)
1522 	struct thread *td;
1523 	struct osigsuspend_args *uap;
1524 {
1525 	sigset_t mask;
1526 
1527 	OSIG2SIG(uap->mask, mask);
1528 	return (kern_sigsuspend(td, mask));
1529 }
1530 #endif /* COMPAT_43 */
1531 
1532 #if defined(COMPAT_43)
1533 #ifndef _SYS_SYSPROTO_H_
1534 struct osigstack_args {
1535 	struct	sigstack *nss;
1536 	struct	sigstack *oss;
1537 };
1538 #endif
1539 /* ARGSUSED */
1540 int
1541 osigstack(td, uap)
1542 	struct thread *td;
1543 	register struct osigstack_args *uap;
1544 {
1545 	struct sigstack nss, oss;
1546 	int error = 0;
1547 
1548 	if (uap->nss != NULL) {
1549 		error = copyin(uap->nss, &nss, sizeof(nss));
1550 		if (error)
1551 			return (error);
1552 	}
1553 	oss.ss_sp = td->td_sigstk.ss_sp;
1554 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1555 	if (uap->nss != NULL) {
1556 		td->td_sigstk.ss_sp = nss.ss_sp;
1557 		td->td_sigstk.ss_size = 0;
1558 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1559 		td->td_pflags |= TDP_ALTSTACK;
1560 	}
1561 	if (uap->oss != NULL)
1562 		error = copyout(&oss, uap->oss, sizeof(oss));
1563 
1564 	return (error);
1565 }
1566 #endif /* COMPAT_43 */
1567 
1568 #ifndef _SYS_SYSPROTO_H_
1569 struct sigaltstack_args {
1570 	stack_t	*ss;
1571 	stack_t	*oss;
1572 };
1573 #endif
1574 /* ARGSUSED */
1575 int
1576 sys_sigaltstack(td, uap)
1577 	struct thread *td;
1578 	register struct sigaltstack_args *uap;
1579 {
1580 	stack_t ss, oss;
1581 	int error;
1582 
1583 	if (uap->ss != NULL) {
1584 		error = copyin(uap->ss, &ss, sizeof(ss));
1585 		if (error)
1586 			return (error);
1587 	}
1588 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1589 	    (uap->oss != NULL) ? &oss : NULL);
1590 	if (error)
1591 		return (error);
1592 	if (uap->oss != NULL)
1593 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1594 	return (error);
1595 }
1596 
1597 int
1598 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1599 {
1600 	struct proc *p = td->td_proc;
1601 	int oonstack;
1602 
1603 	oonstack = sigonstack(cpu_getstack(td));
1604 
1605 	if (oss != NULL) {
1606 		*oss = td->td_sigstk;
1607 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1608 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1609 	}
1610 
1611 	if (ss != NULL) {
1612 		if (oonstack)
1613 			return (EPERM);
1614 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1615 			return (EINVAL);
1616 		if (!(ss->ss_flags & SS_DISABLE)) {
1617 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1618 				return (ENOMEM);
1619 
1620 			td->td_sigstk = *ss;
1621 			td->td_pflags |= TDP_ALTSTACK;
1622 		} else {
1623 			td->td_pflags &= ~TDP_ALTSTACK;
1624 		}
1625 	}
1626 	return (0);
1627 }
1628 
1629 /*
1630  * Common code for kill process group/broadcast kill.
1631  * cp is calling process.
1632  */
1633 static int
1634 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1635 {
1636 	struct proc *p;
1637 	struct pgrp *pgrp;
1638 	int err;
1639 	int ret;
1640 
1641 	ret = ESRCH;
1642 	if (all) {
1643 		/*
1644 		 * broadcast
1645 		 */
1646 		sx_slock(&allproc_lock);
1647 		FOREACH_PROC_IN_SYSTEM(p) {
1648 			PROC_LOCK(p);
1649 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1650 			    p == td->td_proc || p->p_state == PRS_NEW) {
1651 				PROC_UNLOCK(p);
1652 				continue;
1653 			}
1654 			err = p_cansignal(td, p, sig);
1655 			if (err == 0) {
1656 				if (sig)
1657 					pksignal(p, sig, ksi);
1658 				ret = err;
1659 			}
1660 			else if (ret == ESRCH)
1661 				ret = err;
1662 			PROC_UNLOCK(p);
1663 		}
1664 		sx_sunlock(&allproc_lock);
1665 	} else {
1666 		sx_slock(&proctree_lock);
1667 		if (pgid == 0) {
1668 			/*
1669 			 * zero pgid means send to my process group.
1670 			 */
1671 			pgrp = td->td_proc->p_pgrp;
1672 			PGRP_LOCK(pgrp);
1673 		} else {
1674 			pgrp = pgfind(pgid);
1675 			if (pgrp == NULL) {
1676 				sx_sunlock(&proctree_lock);
1677 				return (ESRCH);
1678 			}
1679 		}
1680 		sx_sunlock(&proctree_lock);
1681 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1682 			PROC_LOCK(p);
1683 			if (p->p_pid <= 1 || p->p_flag & P_SYSTEM ||
1684 			    p->p_state == PRS_NEW) {
1685 				PROC_UNLOCK(p);
1686 				continue;
1687 			}
1688 			err = p_cansignal(td, p, sig);
1689 			if (err == 0) {
1690 				if (sig)
1691 					pksignal(p, sig, ksi);
1692 				ret = err;
1693 			}
1694 			else if (ret == ESRCH)
1695 				ret = err;
1696 			PROC_UNLOCK(p);
1697 		}
1698 		PGRP_UNLOCK(pgrp);
1699 	}
1700 	return (ret);
1701 }
1702 
1703 #ifndef _SYS_SYSPROTO_H_
1704 struct kill_args {
1705 	int	pid;
1706 	int	signum;
1707 };
1708 #endif
1709 /* ARGSUSED */
1710 int
1711 sys_kill(struct thread *td, struct kill_args *uap)
1712 {
1713 	ksiginfo_t ksi;
1714 	struct proc *p;
1715 	int error;
1716 
1717 	/*
1718 	 * A process in capability mode can send signals only to himself.
1719 	 * The main rationale behind this is that abort(3) is implemented as
1720 	 * kill(getpid(), SIGABRT).
1721 	 */
1722 	if (IN_CAPABILITY_MODE(td) && uap->pid != td->td_proc->p_pid)
1723 		return (ECAPMODE);
1724 
1725 	AUDIT_ARG_SIGNUM(uap->signum);
1726 	AUDIT_ARG_PID(uap->pid);
1727 	if ((u_int)uap->signum > _SIG_MAXSIG)
1728 		return (EINVAL);
1729 
1730 	ksiginfo_init(&ksi);
1731 	ksi.ksi_signo = uap->signum;
1732 	ksi.ksi_code = SI_USER;
1733 	ksi.ksi_pid = td->td_proc->p_pid;
1734 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1735 
1736 	if (uap->pid > 0) {
1737 		/* kill single process */
1738 		if ((p = pfind(uap->pid)) == NULL) {
1739 			if ((p = zpfind(uap->pid)) == NULL)
1740 				return (ESRCH);
1741 		}
1742 		AUDIT_ARG_PROCESS(p);
1743 		error = p_cansignal(td, p, uap->signum);
1744 		if (error == 0 && uap->signum)
1745 			pksignal(p, uap->signum, &ksi);
1746 		PROC_UNLOCK(p);
1747 		return (error);
1748 	}
1749 	switch (uap->pid) {
1750 	case -1:		/* broadcast signal */
1751 		return (killpg1(td, uap->signum, 0, 1, &ksi));
1752 	case 0:			/* signal own process group */
1753 		return (killpg1(td, uap->signum, 0, 0, &ksi));
1754 	default:		/* negative explicit process group */
1755 		return (killpg1(td, uap->signum, -uap->pid, 0, &ksi));
1756 	}
1757 	/* NOTREACHED */
1758 }
1759 
1760 int
1761 sys_pdkill(td, uap)
1762 	struct thread *td;
1763 	struct pdkill_args *uap;
1764 {
1765 	struct proc *p;
1766 	cap_rights_t rights;
1767 	int error;
1768 
1769 	AUDIT_ARG_SIGNUM(uap->signum);
1770 	AUDIT_ARG_FD(uap->fd);
1771 	if ((u_int)uap->signum > _SIG_MAXSIG)
1772 		return (EINVAL);
1773 
1774 	error = procdesc_find(td, uap->fd,
1775 	    cap_rights_init(&rights, CAP_PDKILL), &p);
1776 	if (error)
1777 		return (error);
1778 	AUDIT_ARG_PROCESS(p);
1779 	error = p_cansignal(td, p, uap->signum);
1780 	if (error == 0 && uap->signum)
1781 		kern_psignal(p, uap->signum);
1782 	PROC_UNLOCK(p);
1783 	return (error);
1784 }
1785 
1786 #if defined(COMPAT_43)
1787 #ifndef _SYS_SYSPROTO_H_
1788 struct okillpg_args {
1789 	int	pgid;
1790 	int	signum;
1791 };
1792 #endif
1793 /* ARGSUSED */
1794 int
1795 okillpg(struct thread *td, struct okillpg_args *uap)
1796 {
1797 	ksiginfo_t ksi;
1798 
1799 	AUDIT_ARG_SIGNUM(uap->signum);
1800 	AUDIT_ARG_PID(uap->pgid);
1801 	if ((u_int)uap->signum > _SIG_MAXSIG)
1802 		return (EINVAL);
1803 
1804 	ksiginfo_init(&ksi);
1805 	ksi.ksi_signo = uap->signum;
1806 	ksi.ksi_code = SI_USER;
1807 	ksi.ksi_pid = td->td_proc->p_pid;
1808 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1809 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1810 }
1811 #endif /* COMPAT_43 */
1812 
1813 #ifndef _SYS_SYSPROTO_H_
1814 struct sigqueue_args {
1815 	pid_t pid;
1816 	int signum;
1817 	/* union sigval */ void *value;
1818 };
1819 #endif
1820 int
1821 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1822 {
1823 	ksiginfo_t ksi;
1824 	struct proc *p;
1825 	int error;
1826 
1827 	if ((u_int)uap->signum > _SIG_MAXSIG)
1828 		return (EINVAL);
1829 
1830 	/*
1831 	 * Specification says sigqueue can only send signal to
1832 	 * single process.
1833 	 */
1834 	if (uap->pid <= 0)
1835 		return (EINVAL);
1836 
1837 	if ((p = pfind(uap->pid)) == NULL) {
1838 		if ((p = zpfind(uap->pid)) == NULL)
1839 			return (ESRCH);
1840 	}
1841 	error = p_cansignal(td, p, uap->signum);
1842 	if (error == 0 && uap->signum != 0) {
1843 		ksiginfo_init(&ksi);
1844 		ksi.ksi_flags = KSI_SIGQ;
1845 		ksi.ksi_signo = uap->signum;
1846 		ksi.ksi_code = SI_QUEUE;
1847 		ksi.ksi_pid = td->td_proc->p_pid;
1848 		ksi.ksi_uid = td->td_ucred->cr_ruid;
1849 		ksi.ksi_value.sival_ptr = uap->value;
1850 		error = pksignal(p, ksi.ksi_signo, &ksi);
1851 	}
1852 	PROC_UNLOCK(p);
1853 	return (error);
1854 }
1855 
1856 /*
1857  * Send a signal to a process group.
1858  */
1859 void
1860 gsignal(int pgid, int sig, ksiginfo_t *ksi)
1861 {
1862 	struct pgrp *pgrp;
1863 
1864 	if (pgid != 0) {
1865 		sx_slock(&proctree_lock);
1866 		pgrp = pgfind(pgid);
1867 		sx_sunlock(&proctree_lock);
1868 		if (pgrp != NULL) {
1869 			pgsignal(pgrp, sig, 0, ksi);
1870 			PGRP_UNLOCK(pgrp);
1871 		}
1872 	}
1873 }
1874 
1875 /*
1876  * Send a signal to a process group.  If checktty is 1,
1877  * limit to members which have a controlling terminal.
1878  */
1879 void
1880 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
1881 {
1882 	struct proc *p;
1883 
1884 	if (pgrp) {
1885 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
1886 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1887 			PROC_LOCK(p);
1888 			if (p->p_state == PRS_NORMAL &&
1889 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
1890 				pksignal(p, sig, ksi);
1891 			PROC_UNLOCK(p);
1892 		}
1893 	}
1894 }
1895 
1896 
1897 /*
1898  * Recalculate the signal mask and reset the signal disposition after
1899  * usermode frame for delivery is formed.  Should be called after
1900  * mach-specific routine, because sysent->sv_sendsig() needs correct
1901  * ps_siginfo and signal mask.
1902  */
1903 static void
1904 postsig_done(int sig, struct thread *td, struct sigacts *ps)
1905 {
1906 	sigset_t mask;
1907 
1908 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1909 	td->td_ru.ru_nsignals++;
1910 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
1911 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
1912 		SIGADDSET(mask, sig);
1913 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
1914 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
1915 	if (SIGISMEMBER(ps->ps_sigreset, sig))
1916 		sigdflt(ps, sig);
1917 }
1918 
1919 
1920 /*
1921  * Send a signal caused by a trap to the current thread.  If it will be
1922  * caught immediately, deliver it with correct code.  Otherwise, post it
1923  * normally.
1924  */
1925 void
1926 trapsignal(struct thread *td, ksiginfo_t *ksi)
1927 {
1928 	struct sigacts *ps;
1929 	struct proc *p;
1930 	int sig;
1931 	int code;
1932 
1933 	p = td->td_proc;
1934 	sig = ksi->ksi_signo;
1935 	code = ksi->ksi_code;
1936 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
1937 
1938 	PROC_LOCK(p);
1939 	ps = p->p_sigacts;
1940 	mtx_lock(&ps->ps_mtx);
1941 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
1942 	    !SIGISMEMBER(td->td_sigmask, sig)) {
1943 #ifdef KTRACE
1944 		if (KTRPOINT(curthread, KTR_PSIG))
1945 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
1946 			    &td->td_sigmask, code);
1947 #endif
1948 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
1949 				ksi, &td->td_sigmask);
1950 		postsig_done(sig, td, ps);
1951 		mtx_unlock(&ps->ps_mtx);
1952 	} else {
1953 		/*
1954 		 * Avoid a possible infinite loop if the thread
1955 		 * masking the signal or process is ignoring the
1956 		 * signal.
1957 		 */
1958 		if (kern_forcesigexit &&
1959 		    (SIGISMEMBER(td->td_sigmask, sig) ||
1960 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
1961 			SIGDELSET(td->td_sigmask, sig);
1962 			SIGDELSET(ps->ps_sigcatch, sig);
1963 			SIGDELSET(ps->ps_sigignore, sig);
1964 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1965 		}
1966 		mtx_unlock(&ps->ps_mtx);
1967 		p->p_code = code;	/* XXX for core dump/debugger */
1968 		p->p_sig = sig;		/* XXX to verify code */
1969 		tdsendsignal(p, td, sig, ksi);
1970 	}
1971 	PROC_UNLOCK(p);
1972 }
1973 
1974 static struct thread *
1975 sigtd(struct proc *p, int sig, int prop)
1976 {
1977 	struct thread *td, *signal_td;
1978 
1979 	PROC_LOCK_ASSERT(p, MA_OWNED);
1980 
1981 	/*
1982 	 * Check if current thread can handle the signal without
1983 	 * switching context to another thread.
1984 	 */
1985 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig))
1986 		return (curthread);
1987 	signal_td = NULL;
1988 	FOREACH_THREAD_IN_PROC(p, td) {
1989 		if (!SIGISMEMBER(td->td_sigmask, sig)) {
1990 			signal_td = td;
1991 			break;
1992 		}
1993 	}
1994 	if (signal_td == NULL)
1995 		signal_td = FIRST_THREAD_IN_PROC(p);
1996 	return (signal_td);
1997 }
1998 
1999 /*
2000  * Send the signal to the process.  If the signal has an action, the action
2001  * is usually performed by the target process rather than the caller; we add
2002  * the signal to the set of pending signals for the process.
2003  *
2004  * Exceptions:
2005  *   o When a stop signal is sent to a sleeping process that takes the
2006  *     default action, the process is stopped without awakening it.
2007  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2008  *     regardless of the signal action (eg, blocked or ignored).
2009  *
2010  * Other ignored signals are discarded immediately.
2011  *
2012  * NB: This function may be entered from the debugger via the "kill" DDB
2013  * command.  There is little that can be done to mitigate the possibly messy
2014  * side effects of this unwise possibility.
2015  */
2016 void
2017 kern_psignal(struct proc *p, int sig)
2018 {
2019 	ksiginfo_t ksi;
2020 
2021 	ksiginfo_init(&ksi);
2022 	ksi.ksi_signo = sig;
2023 	ksi.ksi_code = SI_KERNEL;
2024 	(void) tdsendsignal(p, NULL, sig, &ksi);
2025 }
2026 
2027 int
2028 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2029 {
2030 
2031 	return (tdsendsignal(p, NULL, sig, ksi));
2032 }
2033 
2034 /* Utility function for finding a thread to send signal event to. */
2035 int
2036 sigev_findtd(struct proc *p ,struct sigevent *sigev, struct thread **ttd)
2037 {
2038 	struct thread *td;
2039 
2040 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2041 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2042 		if (td == NULL)
2043 			return (ESRCH);
2044 		*ttd = td;
2045 	} else {
2046 		*ttd = NULL;
2047 		PROC_LOCK(p);
2048 	}
2049 	return (0);
2050 }
2051 
2052 void
2053 tdsignal(struct thread *td, int sig)
2054 {
2055 	ksiginfo_t ksi;
2056 
2057 	ksiginfo_init(&ksi);
2058 	ksi.ksi_signo = sig;
2059 	ksi.ksi_code = SI_KERNEL;
2060 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2061 }
2062 
2063 void
2064 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2065 {
2066 
2067 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2068 }
2069 
2070 int
2071 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2072 {
2073 	sig_t action;
2074 	sigqueue_t *sigqueue;
2075 	int prop;
2076 	struct sigacts *ps;
2077 	int intrval;
2078 	int ret = 0;
2079 	int wakeup_swapper;
2080 
2081 	MPASS(td == NULL || p == td->td_proc);
2082 	PROC_LOCK_ASSERT(p, MA_OWNED);
2083 
2084 	if (!_SIG_VALID(sig))
2085 		panic("%s(): invalid signal %d", __func__, sig);
2086 
2087 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2088 
2089 	/*
2090 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2091 	 */
2092 	if (p->p_state == PRS_ZOMBIE) {
2093 		if (ksi && (ksi->ksi_flags & KSI_INS))
2094 			ksiginfo_tryfree(ksi);
2095 		return (ret);
2096 	}
2097 
2098 	ps = p->p_sigacts;
2099 	KNOTE_LOCKED(&p->p_klist, NOTE_SIGNAL | sig);
2100 	prop = sigprop(sig);
2101 
2102 	if (td == NULL) {
2103 		td = sigtd(p, sig, prop);
2104 		sigqueue = &p->p_sigqueue;
2105 	} else
2106 		sigqueue = &td->td_sigqueue;
2107 
2108 	SDT_PROBE(proc, kernel, , signal__send, td, p, sig, 0, 0 );
2109 
2110 	/*
2111 	 * If the signal is being ignored,
2112 	 * then we forget about it immediately.
2113 	 * (Note: we don't set SIGCONT in ps_sigignore,
2114 	 * and if it is set to SIG_IGN,
2115 	 * action will be SIG_DFL here.)
2116 	 */
2117 	mtx_lock(&ps->ps_mtx);
2118 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2119 		SDT_PROBE(proc, kernel, , signal__discard, td, p, sig, 0, 0 );
2120 
2121 		mtx_unlock(&ps->ps_mtx);
2122 		if (ksi && (ksi->ksi_flags & KSI_INS))
2123 			ksiginfo_tryfree(ksi);
2124 		return (ret);
2125 	}
2126 	if (SIGISMEMBER(td->td_sigmask, sig))
2127 		action = SIG_HOLD;
2128 	else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2129 		action = SIG_CATCH;
2130 	else
2131 		action = SIG_DFL;
2132 	if (SIGISMEMBER(ps->ps_sigintr, sig))
2133 		intrval = EINTR;
2134 	else
2135 		intrval = ERESTART;
2136 	mtx_unlock(&ps->ps_mtx);
2137 
2138 	if (prop & SA_CONT)
2139 		sigqueue_delete_stopmask_proc(p);
2140 	else if (prop & SA_STOP) {
2141 		/*
2142 		 * If sending a tty stop signal to a member of an orphaned
2143 		 * process group, discard the signal here if the action
2144 		 * is default; don't stop the process below if sleeping,
2145 		 * and don't clear any pending SIGCONT.
2146 		 */
2147 		if ((prop & SA_TTYSTOP) &&
2148 		    (p->p_pgrp->pg_jobc == 0) &&
2149 		    (action == SIG_DFL)) {
2150 			if (ksi && (ksi->ksi_flags & KSI_INS))
2151 				ksiginfo_tryfree(ksi);
2152 			return (ret);
2153 		}
2154 		sigqueue_delete_proc(p, SIGCONT);
2155 		if (p->p_flag & P_CONTINUED) {
2156 			p->p_flag &= ~P_CONTINUED;
2157 			PROC_LOCK(p->p_pptr);
2158 			sigqueue_take(p->p_ksi);
2159 			PROC_UNLOCK(p->p_pptr);
2160 		}
2161 	}
2162 
2163 	ret = sigqueue_add(sigqueue, sig, ksi);
2164 	if (ret != 0)
2165 		return (ret);
2166 	signotify(td);
2167 	/*
2168 	 * Defer further processing for signals which are held,
2169 	 * except that stopped processes must be continued by SIGCONT.
2170 	 */
2171 	if (action == SIG_HOLD &&
2172 	    !((prop & SA_CONT) && (p->p_flag & P_STOPPED_SIG)))
2173 		return (ret);
2174 	/*
2175 	 * SIGKILL: Remove procfs STOPEVENTs.
2176 	 */
2177 	if (sig == SIGKILL) {
2178 		/* from procfs_ioctl.c: PIOCBIC */
2179 		p->p_stops = 0;
2180 		/* from procfs_ioctl.c: PIOCCONT */
2181 		p->p_step = 0;
2182 		wakeup(&p->p_step);
2183 	}
2184 	/*
2185 	 * Some signals have a process-wide effect and a per-thread
2186 	 * component.  Most processing occurs when the process next
2187 	 * tries to cross the user boundary, however there are some
2188 	 * times when processing needs to be done immediately, such as
2189 	 * waking up threads so that they can cross the user boundary.
2190 	 * We try to do the per-process part here.
2191 	 */
2192 	if (P_SHOULDSTOP(p)) {
2193 		KASSERT(!(p->p_flag & P_WEXIT),
2194 		    ("signal to stopped but exiting process"));
2195 		if (sig == SIGKILL) {
2196 			/*
2197 			 * If traced process is already stopped,
2198 			 * then no further action is necessary.
2199 			 */
2200 			if (p->p_flag & P_TRACED)
2201 				goto out;
2202 			/*
2203 			 * SIGKILL sets process running.
2204 			 * It will die elsewhere.
2205 			 * All threads must be restarted.
2206 			 */
2207 			p->p_flag &= ~P_STOPPED_SIG;
2208 			goto runfast;
2209 		}
2210 
2211 		if (prop & SA_CONT) {
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 			 * If SIGCONT is default (or ignored), we continue the
2220 			 * process but don't leave the signal in sigqueue as
2221 			 * it has no further action.  If SIGCONT is held, we
2222 			 * continue the process and leave the signal in
2223 			 * sigqueue.  If the process catches SIGCONT, let it
2224 			 * handle the signal itself.  If it isn't waiting on
2225 			 * an event, it goes back to run state.
2226 			 * Otherwise, process goes back to sleep state.
2227 			 */
2228 			p->p_flag &= ~P_STOPPED_SIG;
2229 			PROC_SLOCK(p);
2230 			if (p->p_numthreads == p->p_suspcount) {
2231 				PROC_SUNLOCK(p);
2232 				p->p_flag |= P_CONTINUED;
2233 				p->p_xstat = SIGCONT;
2234 				PROC_LOCK(p->p_pptr);
2235 				childproc_continued(p);
2236 				PROC_UNLOCK(p->p_pptr);
2237 				PROC_SLOCK(p);
2238 			}
2239 			if (action == SIG_DFL) {
2240 				thread_unsuspend(p);
2241 				PROC_SUNLOCK(p);
2242 				sigqueue_delete(sigqueue, sig);
2243 				goto out;
2244 			}
2245 			if (action == SIG_CATCH) {
2246 				/*
2247 				 * The process wants to catch it so it needs
2248 				 * to run at least one thread, but which one?
2249 				 */
2250 				PROC_SUNLOCK(p);
2251 				goto runfast;
2252 			}
2253 			/*
2254 			 * The signal is not ignored or caught.
2255 			 */
2256 			thread_unsuspend(p);
2257 			PROC_SUNLOCK(p);
2258 			goto out;
2259 		}
2260 
2261 		if (prop & SA_STOP) {
2262 			/*
2263 			 * If traced process is already stopped,
2264 			 * then no further action is necessary.
2265 			 */
2266 			if (p->p_flag & P_TRACED)
2267 				goto out;
2268 			/*
2269 			 * Already stopped, don't need to stop again
2270 			 * (If we did the shell could get confused).
2271 			 * Just make sure the signal STOP bit set.
2272 			 */
2273 			p->p_flag |= P_STOPPED_SIG;
2274 			sigqueue_delete(sigqueue, sig);
2275 			goto out;
2276 		}
2277 
2278 		/*
2279 		 * All other kinds of signals:
2280 		 * If a thread is sleeping interruptibly, simulate a
2281 		 * wakeup so that when it is continued it will be made
2282 		 * runnable and can look at the signal.  However, don't make
2283 		 * the PROCESS runnable, leave it stopped.
2284 		 * It may run a bit until it hits a thread_suspend_check().
2285 		 */
2286 		wakeup_swapper = 0;
2287 		PROC_SLOCK(p);
2288 		thread_lock(td);
2289 		if (TD_ON_SLEEPQ(td) && (td->td_flags & TDF_SINTR))
2290 			wakeup_swapper = sleepq_abort(td, intrval);
2291 		thread_unlock(td);
2292 		PROC_SUNLOCK(p);
2293 		if (wakeup_swapper)
2294 			kick_proc0();
2295 		goto out;
2296 		/*
2297 		 * Mutexes are short lived. Threads waiting on them will
2298 		 * hit thread_suspend_check() soon.
2299 		 */
2300 	} else if (p->p_state == PRS_NORMAL) {
2301 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2302 			tdsigwakeup(td, sig, action, intrval);
2303 			goto out;
2304 		}
2305 
2306 		MPASS(action == SIG_DFL);
2307 
2308 		if (prop & SA_STOP) {
2309 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2310 				goto out;
2311 			p->p_flag |= P_STOPPED_SIG;
2312 			p->p_xstat = sig;
2313 			PROC_SLOCK(p);
2314 			sig_suspend_threads(td, p, 1);
2315 			if (p->p_numthreads == p->p_suspcount) {
2316 				/*
2317 				 * only thread sending signal to another
2318 				 * process can reach here, if thread is sending
2319 				 * signal to its process, because thread does
2320 				 * not suspend itself here, p_numthreads
2321 				 * should never be equal to p_suspcount.
2322 				 */
2323 				thread_stopped(p);
2324 				PROC_SUNLOCK(p);
2325 				sigqueue_delete_proc(p, p->p_xstat);
2326 			} else
2327 				PROC_SUNLOCK(p);
2328 			goto out;
2329 		}
2330 	} else {
2331 		/* Not in "NORMAL" state. discard the signal. */
2332 		sigqueue_delete(sigqueue, sig);
2333 		goto out;
2334 	}
2335 
2336 	/*
2337 	 * The process is not stopped so we need to apply the signal to all the
2338 	 * running threads.
2339 	 */
2340 runfast:
2341 	tdsigwakeup(td, sig, action, intrval);
2342 	PROC_SLOCK(p);
2343 	thread_unsuspend(p);
2344 	PROC_SUNLOCK(p);
2345 out:
2346 	/* If we jump here, proc slock should not be owned. */
2347 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2348 	return (ret);
2349 }
2350 
2351 /*
2352  * The force of a signal has been directed against a single
2353  * thread.  We need to see what we can do about knocking it
2354  * out of any sleep it may be in etc.
2355  */
2356 static void
2357 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2358 {
2359 	struct proc *p = td->td_proc;
2360 	register int prop;
2361 	int wakeup_swapper;
2362 
2363 	wakeup_swapper = 0;
2364 	PROC_LOCK_ASSERT(p, MA_OWNED);
2365 	prop = sigprop(sig);
2366 
2367 	PROC_SLOCK(p);
2368 	thread_lock(td);
2369 	/*
2370 	 * Bring the priority of a thread up if we want it to get
2371 	 * killed in this lifetime.
2372 	 */
2373 	if (action == SIG_DFL && (prop & SA_KILL) && td->td_priority > PUSER)
2374 		sched_prio(td, PUSER);
2375 	if (TD_ON_SLEEPQ(td)) {
2376 		/*
2377 		 * If thread is sleeping uninterruptibly
2378 		 * we can't interrupt the sleep... the signal will
2379 		 * be noticed when the process returns through
2380 		 * trap() or syscall().
2381 		 */
2382 		if ((td->td_flags & TDF_SINTR) == 0)
2383 			goto out;
2384 		/*
2385 		 * If SIGCONT is default (or ignored) and process is
2386 		 * asleep, we are finished; the process should not
2387 		 * be awakened.
2388 		 */
2389 		if ((prop & SA_CONT) && action == SIG_DFL) {
2390 			thread_unlock(td);
2391 			PROC_SUNLOCK(p);
2392 			sigqueue_delete(&p->p_sigqueue, sig);
2393 			/*
2394 			 * It may be on either list in this state.
2395 			 * Remove from both for now.
2396 			 */
2397 			sigqueue_delete(&td->td_sigqueue, sig);
2398 			return;
2399 		}
2400 
2401 		/*
2402 		 * Don't awaken a sleeping thread for SIGSTOP if the
2403 		 * STOP signal is deferred.
2404 		 */
2405 		if ((prop & SA_STOP) && (td->td_flags & TDF_SBDRY))
2406 			goto out;
2407 
2408 		/*
2409 		 * Give low priority threads a better chance to run.
2410 		 */
2411 		if (td->td_priority > PUSER)
2412 			sched_prio(td, PUSER);
2413 
2414 		wakeup_swapper = sleepq_abort(td, intrval);
2415 	} else {
2416 		/*
2417 		 * Other states do nothing with the signal immediately,
2418 		 * other than kicking ourselves if we are running.
2419 		 * It will either never be noticed, or noticed very soon.
2420 		 */
2421 #ifdef SMP
2422 		if (TD_IS_RUNNING(td) && td != curthread)
2423 			forward_signal(td);
2424 #endif
2425 	}
2426 out:
2427 	PROC_SUNLOCK(p);
2428 	thread_unlock(td);
2429 	if (wakeup_swapper)
2430 		kick_proc0();
2431 }
2432 
2433 static void
2434 sig_suspend_threads(struct thread *td, struct proc *p, int sending)
2435 {
2436 	struct thread *td2;
2437 
2438 	PROC_LOCK_ASSERT(p, MA_OWNED);
2439 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2440 
2441 	FOREACH_THREAD_IN_PROC(p, td2) {
2442 		thread_lock(td2);
2443 		td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
2444 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2445 		    (td2->td_flags & TDF_SINTR)) {
2446 			if (td2->td_flags & TDF_SBDRY) {
2447 				/*
2448 				 * Once a thread is asleep with
2449 				 * TDF_SBDRY set, it should never
2450 				 * become suspended due to this check.
2451 				 */
2452 				KASSERT(!TD_IS_SUSPENDED(td2),
2453 				    ("thread with deferred stops suspended"));
2454 			} else if (!TD_IS_SUSPENDED(td2)) {
2455 				thread_suspend_one(td2);
2456 			}
2457 		} else if (!TD_IS_SUSPENDED(td2)) {
2458 			if (sending || td != td2)
2459 				td2->td_flags |= TDF_ASTPENDING;
2460 #ifdef SMP
2461 			if (TD_IS_RUNNING(td2) && td2 != td)
2462 				forward_signal(td2);
2463 #endif
2464 		}
2465 		thread_unlock(td2);
2466 	}
2467 }
2468 
2469 int
2470 ptracestop(struct thread *td, int sig)
2471 {
2472 	struct proc *p = td->td_proc;
2473 
2474 	PROC_LOCK_ASSERT(p, MA_OWNED);
2475 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2476 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2477 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2478 
2479 	td->td_dbgflags |= TDB_XSIG;
2480 	td->td_xsig = sig;
2481 	PROC_SLOCK(p);
2482 	while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2483 		if (p->p_flag & P_SINGLE_EXIT) {
2484 			td->td_dbgflags &= ~TDB_XSIG;
2485 			PROC_SUNLOCK(p);
2486 			return (sig);
2487 		}
2488 		/*
2489 		 * Just make wait() to work, the last stopped thread
2490 		 * will win.
2491 		 */
2492 		p->p_xstat = sig;
2493 		p->p_xthread = td;
2494 		p->p_flag |= (P_STOPPED_SIG|P_STOPPED_TRACE);
2495 		sig_suspend_threads(td, p, 0);
2496 		if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2497 			td->td_dbgflags &= ~TDB_STOPATFORK;
2498 			cv_broadcast(&p->p_dbgwait);
2499 		}
2500 stopme:
2501 		thread_suspend_switch(td, p);
2502 		if (p->p_xthread == td)
2503 			p->p_xthread = NULL;
2504 		if (!(p->p_flag & P_TRACED))
2505 			break;
2506 		if (td->td_dbgflags & TDB_SUSPEND) {
2507 			if (p->p_flag & P_SINGLE_EXIT)
2508 				break;
2509 			goto stopme;
2510 		}
2511 	}
2512 	PROC_SUNLOCK(p);
2513 	return (td->td_xsig);
2514 }
2515 
2516 static void
2517 reschedule_signals(struct proc *p, sigset_t block, int flags)
2518 {
2519 	struct sigacts *ps;
2520 	struct thread *td;
2521 	int sig;
2522 
2523 	PROC_LOCK_ASSERT(p, MA_OWNED);
2524 	ps = p->p_sigacts;
2525 	mtx_assert(&ps->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0 ?
2526 	    MA_OWNED : MA_NOTOWNED);
2527 	if (SIGISEMPTY(p->p_siglist))
2528 		return;
2529 	SIGSETAND(block, p->p_siglist);
2530 	while ((sig = sig_ffs(&block)) != 0) {
2531 		SIGDELSET(block, sig);
2532 		td = sigtd(p, sig, 0);
2533 		signotify(td);
2534 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2535 			mtx_lock(&ps->ps_mtx);
2536 		if (p->p_flag & P_TRACED || SIGISMEMBER(ps->ps_sigcatch, sig))
2537 			tdsigwakeup(td, sig, SIG_CATCH,
2538 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2539 			     ERESTART));
2540 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2541 			mtx_unlock(&ps->ps_mtx);
2542 	}
2543 }
2544 
2545 void
2546 tdsigcleanup(struct thread *td)
2547 {
2548 	struct proc *p;
2549 	sigset_t unblocked;
2550 
2551 	p = td->td_proc;
2552 	PROC_LOCK_ASSERT(p, MA_OWNED);
2553 
2554 	sigqueue_flush(&td->td_sigqueue);
2555 	if (p->p_numthreads == 1)
2556 		return;
2557 
2558 	/*
2559 	 * Since we cannot handle signals, notify signal post code
2560 	 * about this by filling the sigmask.
2561 	 *
2562 	 * Also, if needed, wake up thread(s) that do not block the
2563 	 * same signals as the exiting thread, since the thread might
2564 	 * have been selected for delivery and woken up.
2565 	 */
2566 	SIGFILLSET(unblocked);
2567 	SIGSETNAND(unblocked, td->td_sigmask);
2568 	SIGFILLSET(td->td_sigmask);
2569 	reschedule_signals(p, unblocked, 0);
2570 
2571 }
2572 
2573 /*
2574  * Defer the delivery of SIGSTOP for the current thread.  Returns true
2575  * if stops were deferred and false if they were already deferred.
2576  */
2577 int
2578 sigdeferstop(void)
2579 {
2580 	struct thread *td;
2581 
2582 	td = curthread;
2583 	if (td->td_flags & TDF_SBDRY)
2584 		return (0);
2585 	thread_lock(td);
2586 	td->td_flags |= TDF_SBDRY;
2587 	thread_unlock(td);
2588 	return (1);
2589 }
2590 
2591 /*
2592  * Permit the delivery of SIGSTOP for the current thread.  This does
2593  * not immediately suspend if a stop was posted.  Instead, the thread
2594  * will suspend either via ast() or a subsequent interruptible sleep.
2595  */
2596 int
2597 sigallowstop(void)
2598 {
2599 	struct thread *td;
2600 	int prev;
2601 
2602 	td = curthread;
2603 	thread_lock(td);
2604 	prev = (td->td_flags & TDF_SBDRY) != 0;
2605 	td->td_flags &= ~TDF_SBDRY;
2606 	thread_unlock(td);
2607 	return (prev);
2608 }
2609 
2610 /*
2611  * If the current process has received a signal (should be caught or cause
2612  * termination, should interrupt current syscall), return the signal number.
2613  * Stop signals with default action are processed immediately, then cleared;
2614  * they aren't returned.  This is checked after each entry to the system for
2615  * a syscall or trap (though this can usually be done without calling issignal
2616  * by checking the pending signal masks in cursig.) The normal call
2617  * sequence is
2618  *
2619  *	while (sig = cursig(curthread))
2620  *		postsig(sig);
2621  */
2622 static int
2623 issignal(struct thread *td)
2624 {
2625 	struct proc *p;
2626 	struct sigacts *ps;
2627 	struct sigqueue *queue;
2628 	sigset_t sigpending;
2629 	int sig, prop, newsig;
2630 
2631 	p = td->td_proc;
2632 	ps = p->p_sigacts;
2633 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2634 	PROC_LOCK_ASSERT(p, MA_OWNED);
2635 	for (;;) {
2636 		int traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
2637 
2638 		sigpending = td->td_sigqueue.sq_signals;
2639 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
2640 		SIGSETNAND(sigpending, td->td_sigmask);
2641 
2642 		if (p->p_flag & P_PPWAIT || td->td_flags & TDF_SBDRY)
2643 			SIG_STOPSIGMASK(sigpending);
2644 		if (SIGISEMPTY(sigpending))	/* no signal to send */
2645 			return (0);
2646 		sig = sig_ffs(&sigpending);
2647 
2648 		if (p->p_stops & S_SIG) {
2649 			mtx_unlock(&ps->ps_mtx);
2650 			stopevent(p, S_SIG, sig);
2651 			mtx_lock(&ps->ps_mtx);
2652 		}
2653 
2654 		/*
2655 		 * We should see pending but ignored signals
2656 		 * only if P_TRACED was on when they were posted.
2657 		 */
2658 		if (SIGISMEMBER(ps->ps_sigignore, sig) && (traced == 0)) {
2659 			sigqueue_delete(&td->td_sigqueue, sig);
2660 			sigqueue_delete(&p->p_sigqueue, sig);
2661 			continue;
2662 		}
2663 		if (p->p_flag & P_TRACED && (p->p_flag & P_PPTRACE) == 0) {
2664 			/*
2665 			 * If traced, always stop.
2666 			 * Remove old signal from queue before the stop.
2667 			 * XXX shrug off debugger, it causes siginfo to
2668 			 * be thrown away.
2669 			 */
2670 			queue = &td->td_sigqueue;
2671 			td->td_dbgksi.ksi_signo = 0;
2672 			if (sigqueue_get(queue, sig, &td->td_dbgksi) == 0) {
2673 				queue = &p->p_sigqueue;
2674 				sigqueue_get(queue, sig, &td->td_dbgksi);
2675 			}
2676 
2677 			mtx_unlock(&ps->ps_mtx);
2678 			newsig = ptracestop(td, sig);
2679 			mtx_lock(&ps->ps_mtx);
2680 
2681 			if (sig != newsig) {
2682 
2683 				/*
2684 				 * If parent wants us to take the signal,
2685 				 * then it will leave it in p->p_xstat;
2686 				 * otherwise we just look for signals again.
2687 				*/
2688 				if (newsig == 0)
2689 					continue;
2690 				sig = newsig;
2691 
2692 				/*
2693 				 * Put the new signal into td_sigqueue. If the
2694 				 * signal is being masked, look for other
2695 				 * signals.
2696 				 */
2697 				sigqueue_add(queue, sig, NULL);
2698 				if (SIGISMEMBER(td->td_sigmask, sig))
2699 					continue;
2700 				signotify(td);
2701 			} else {
2702 				if (td->td_dbgksi.ksi_signo != 0) {
2703 					td->td_dbgksi.ksi_flags |= KSI_HEAD;
2704 					if (sigqueue_add(&td->td_sigqueue, sig,
2705 					    &td->td_dbgksi) != 0)
2706 						td->td_dbgksi.ksi_signo = 0;
2707 				}
2708 				if (td->td_dbgksi.ksi_signo == 0)
2709 					sigqueue_add(&td->td_sigqueue, sig,
2710 					    NULL);
2711 			}
2712 
2713 			/*
2714 			 * If the traced bit got turned off, go back up
2715 			 * to the top to rescan signals.  This ensures
2716 			 * that p_sig* and p_sigact are consistent.
2717 			 */
2718 			if ((p->p_flag & P_TRACED) == 0)
2719 				continue;
2720 		}
2721 
2722 		prop = sigprop(sig);
2723 
2724 		/*
2725 		 * Decide whether the signal should be returned.
2726 		 * Return the signal's number, or fall through
2727 		 * to clear it from the pending mask.
2728 		 */
2729 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2730 
2731 		case (intptr_t)SIG_DFL:
2732 			/*
2733 			 * Don't take default actions on system processes.
2734 			 */
2735 			if (p->p_pid <= 1) {
2736 #ifdef DIAGNOSTIC
2737 				/*
2738 				 * Are you sure you want to ignore SIGSEGV
2739 				 * in init? XXX
2740 				 */
2741 				printf("Process (pid %lu) got signal %d\n",
2742 					(u_long)p->p_pid, sig);
2743 #endif
2744 				break;		/* == ignore */
2745 			}
2746 			/*
2747 			 * If there is a pending stop signal to process
2748 			 * with default action, stop here,
2749 			 * then clear the signal.  However,
2750 			 * if process is member of an orphaned
2751 			 * process group, ignore tty stop signals.
2752 			 */
2753 			if (prop & SA_STOP) {
2754 				if (p->p_flag & (P_TRACED|P_WEXIT) ||
2755 				    (p->p_pgrp->pg_jobc == 0 &&
2756 				     prop & SA_TTYSTOP))
2757 					break;	/* == ignore */
2758 				mtx_unlock(&ps->ps_mtx);
2759 				WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2760 				    &p->p_mtx.lock_object, "Catching SIGSTOP");
2761 				p->p_flag |= P_STOPPED_SIG;
2762 				p->p_xstat = sig;
2763 				PROC_SLOCK(p);
2764 				sig_suspend_threads(td, p, 0);
2765 				thread_suspend_switch(td, p);
2766 				PROC_SUNLOCK(p);
2767 				mtx_lock(&ps->ps_mtx);
2768 				break;
2769 			} else if (prop & SA_IGNORE) {
2770 				/*
2771 				 * Except for SIGCONT, shouldn't get here.
2772 				 * Default action is to ignore; drop it.
2773 				 */
2774 				break;		/* == ignore */
2775 			} else
2776 				return (sig);
2777 			/*NOTREACHED*/
2778 
2779 		case (intptr_t)SIG_IGN:
2780 			/*
2781 			 * Masking above should prevent us ever trying
2782 			 * to take action on an ignored signal other
2783 			 * than SIGCONT, unless process is traced.
2784 			 */
2785 			if ((prop & SA_CONT) == 0 &&
2786 			    (p->p_flag & P_TRACED) == 0)
2787 				printf("issignal\n");
2788 			break;		/* == ignore */
2789 
2790 		default:
2791 			/*
2792 			 * This signal has an action, let
2793 			 * postsig() process it.
2794 			 */
2795 			return (sig);
2796 		}
2797 		sigqueue_delete(&td->td_sigqueue, sig);	/* take the signal! */
2798 		sigqueue_delete(&p->p_sigqueue, sig);
2799 	}
2800 	/* NOTREACHED */
2801 }
2802 
2803 void
2804 thread_stopped(struct proc *p)
2805 {
2806 	int n;
2807 
2808 	PROC_LOCK_ASSERT(p, MA_OWNED);
2809 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2810 	n = p->p_suspcount;
2811 	if (p == curproc)
2812 		n++;
2813 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
2814 		PROC_SUNLOCK(p);
2815 		p->p_flag &= ~P_WAITED;
2816 		PROC_LOCK(p->p_pptr);
2817 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
2818 			CLD_TRAPPED : CLD_STOPPED);
2819 		PROC_UNLOCK(p->p_pptr);
2820 		PROC_SLOCK(p);
2821 	}
2822 }
2823 
2824 /*
2825  * Take the action for the specified signal
2826  * from the current set of pending signals.
2827  */
2828 int
2829 postsig(sig)
2830 	register int sig;
2831 {
2832 	struct thread *td = curthread;
2833 	register struct proc *p = td->td_proc;
2834 	struct sigacts *ps;
2835 	sig_t action;
2836 	ksiginfo_t ksi;
2837 	sigset_t returnmask;
2838 
2839 	KASSERT(sig != 0, ("postsig"));
2840 
2841 	PROC_LOCK_ASSERT(p, MA_OWNED);
2842 	ps = p->p_sigacts;
2843 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2844 	ksiginfo_init(&ksi);
2845 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
2846 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
2847 		return (0);
2848 	ksi.ksi_signo = sig;
2849 	if (ksi.ksi_code == SI_TIMER)
2850 		itimer_accept(p, ksi.ksi_timerid, &ksi);
2851 	action = ps->ps_sigact[_SIG_IDX(sig)];
2852 #ifdef KTRACE
2853 	if (KTRPOINT(td, KTR_PSIG))
2854 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
2855 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
2856 #endif
2857 	if (p->p_stops & S_SIG) {
2858 		mtx_unlock(&ps->ps_mtx);
2859 		stopevent(p, S_SIG, sig);
2860 		mtx_lock(&ps->ps_mtx);
2861 	}
2862 
2863 	if (action == SIG_DFL) {
2864 		/*
2865 		 * Default action, where the default is to kill
2866 		 * the process.  (Other cases were ignored above.)
2867 		 */
2868 		mtx_unlock(&ps->ps_mtx);
2869 		sigexit(td, sig);
2870 		/* NOTREACHED */
2871 	} else {
2872 		/*
2873 		 * If we get here, the signal must be caught.
2874 		 */
2875 		KASSERT(action != SIG_IGN && !SIGISMEMBER(td->td_sigmask, sig),
2876 		    ("postsig action"));
2877 		/*
2878 		 * Set the new mask value and also defer further
2879 		 * occurrences of this signal.
2880 		 *
2881 		 * Special case: user has done a sigsuspend.  Here the
2882 		 * current mask is not of interest, but rather the
2883 		 * mask from before the sigsuspend is what we want
2884 		 * restored after the signal processing is completed.
2885 		 */
2886 		if (td->td_pflags & TDP_OLDMASK) {
2887 			returnmask = td->td_oldsigmask;
2888 			td->td_pflags &= ~TDP_OLDMASK;
2889 		} else
2890 			returnmask = td->td_sigmask;
2891 
2892 		if (p->p_sig == sig) {
2893 			p->p_code = 0;
2894 			p->p_sig = 0;
2895 		}
2896 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
2897 		postsig_done(sig, td, ps);
2898 	}
2899 	return (1);
2900 }
2901 
2902 /*
2903  * Kill the current process for stated reason.
2904  */
2905 void
2906 killproc(p, why)
2907 	struct proc *p;
2908 	char *why;
2909 {
2910 
2911 	PROC_LOCK_ASSERT(p, MA_OWNED);
2912 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
2913 	    p->p_comm);
2914 	log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n", p->p_pid,
2915 	    p->p_comm, p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2916 	p->p_flag |= P_WKILLED;
2917 	kern_psignal(p, SIGKILL);
2918 }
2919 
2920 /*
2921  * Force the current process to exit with the specified signal, dumping core
2922  * if appropriate.  We bypass the normal tests for masked and caught signals,
2923  * allowing unrecoverable failures to terminate the process without changing
2924  * signal state.  Mark the accounting record with the signal termination.
2925  * If dumping core, save the signal number for the debugger.  Calls exit and
2926  * does not return.
2927  */
2928 void
2929 sigexit(td, sig)
2930 	struct thread *td;
2931 	int sig;
2932 {
2933 	struct proc *p = td->td_proc;
2934 
2935 	PROC_LOCK_ASSERT(p, MA_OWNED);
2936 	p->p_acflag |= AXSIG;
2937 	/*
2938 	 * We must be single-threading to generate a core dump.  This
2939 	 * ensures that the registers in the core file are up-to-date.
2940 	 * Also, the ELF dump handler assumes that the thread list doesn't
2941 	 * change out from under it.
2942 	 *
2943 	 * XXX If another thread attempts to single-thread before us
2944 	 *     (e.g. via fork()), we won't get a dump at all.
2945 	 */
2946 	if ((sigprop(sig) & SA_CORE) && thread_single(p, SINGLE_NO_EXIT) == 0) {
2947 		p->p_sig = sig;
2948 		/*
2949 		 * Log signals which would cause core dumps
2950 		 * (Log as LOG_INFO to appease those who don't want
2951 		 * these messages.)
2952 		 * XXX : Todo, as well as euid, write out ruid too
2953 		 * Note that coredump() drops proc lock.
2954 		 */
2955 		if (coredump(td) == 0)
2956 			sig |= WCOREFLAG;
2957 		if (kern_logsigexit)
2958 			log(LOG_INFO,
2959 			    "pid %d (%s), uid %d: exited on signal %d%s\n",
2960 			    p->p_pid, p->p_comm,
2961 			    td->td_ucred ? td->td_ucred->cr_uid : -1,
2962 			    sig &~ WCOREFLAG,
2963 			    sig & WCOREFLAG ? " (core dumped)" : "");
2964 	} else
2965 		PROC_UNLOCK(p);
2966 	exit1(td, W_EXITCODE(0, sig));
2967 	/* NOTREACHED */
2968 }
2969 
2970 /*
2971  * Send queued SIGCHLD to parent when child process's state
2972  * is changed.
2973  */
2974 static void
2975 sigparent(struct proc *p, int reason, int status)
2976 {
2977 	PROC_LOCK_ASSERT(p, MA_OWNED);
2978 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
2979 
2980 	if (p->p_ksi != NULL) {
2981 		p->p_ksi->ksi_signo  = SIGCHLD;
2982 		p->p_ksi->ksi_code   = reason;
2983 		p->p_ksi->ksi_status = status;
2984 		p->p_ksi->ksi_pid    = p->p_pid;
2985 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
2986 		if (KSI_ONQ(p->p_ksi))
2987 			return;
2988 	}
2989 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
2990 }
2991 
2992 static void
2993 childproc_jobstate(struct proc *p, int reason, int sig)
2994 {
2995 	struct sigacts *ps;
2996 
2997 	PROC_LOCK_ASSERT(p, MA_OWNED);
2998 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
2999 
3000 	/*
3001 	 * Wake up parent sleeping in kern_wait(), also send
3002 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3003 	 * that parent will awake, because parent may masked
3004 	 * the signal.
3005 	 */
3006 	p->p_pptr->p_flag |= P_STATCHILD;
3007 	wakeup(p->p_pptr);
3008 
3009 	ps = p->p_pptr->p_sigacts;
3010 	mtx_lock(&ps->ps_mtx);
3011 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3012 		mtx_unlock(&ps->ps_mtx);
3013 		sigparent(p, reason, sig);
3014 	} else
3015 		mtx_unlock(&ps->ps_mtx);
3016 }
3017 
3018 void
3019 childproc_stopped(struct proc *p, int reason)
3020 {
3021 	/* p_xstat is a plain signal number, not a full wait() status here. */
3022 	childproc_jobstate(p, reason, p->p_xstat);
3023 }
3024 
3025 void
3026 childproc_continued(struct proc *p)
3027 {
3028 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3029 }
3030 
3031 void
3032 childproc_exited(struct proc *p)
3033 {
3034 	int reason;
3035 	int xstat = p->p_xstat; /* convert to int */
3036 	int status;
3037 
3038 	if (WCOREDUMP(xstat))
3039 		reason = CLD_DUMPED, status = WTERMSIG(xstat);
3040 	else if (WIFSIGNALED(xstat))
3041 		reason = CLD_KILLED, status = WTERMSIG(xstat);
3042 	else
3043 		reason = CLD_EXITED, status = WEXITSTATUS(xstat);
3044 	/*
3045 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3046 	 * done in exit1().
3047 	 */
3048 	sigparent(p, reason, status);
3049 }
3050 
3051 /*
3052  * We only have 1 character for the core count in the format
3053  * string, so the range will be 0-9
3054  */
3055 #define MAX_NUM_CORES 10
3056 static int num_cores = 5;
3057 
3058 static int
3059 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3060 {
3061 	int error;
3062 	int new_val;
3063 
3064 	new_val = num_cores;
3065 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3066 	if (error != 0 || req->newptr == NULL)
3067 		return (error);
3068 	if (new_val > MAX_NUM_CORES)
3069 		new_val = MAX_NUM_CORES;
3070 	if (new_val < 0)
3071 		new_val = 0;
3072 	num_cores = new_val;
3073 	return (0);
3074 }
3075 SYSCTL_PROC(_debug, OID_AUTO, ncores, CTLTYPE_INT|CTLFLAG_RW,
3076 	    0, sizeof(int), sysctl_debug_num_cores_check, "I", "");
3077 
3078 #if defined(COMPRESS_USER_CORES)
3079 int compress_user_cores = 1;
3080 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores, CTLFLAG_RW,
3081     &compress_user_cores, 0, "Compression of user corefiles");
3082 
3083 int compress_user_cores_gzlevel = -1; /* default level */
3084 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_gzlevel, CTLFLAG_RW,
3085     &compress_user_cores_gzlevel, -1, "Corefile gzip compression level");
3086 
3087 #define GZ_SUFFIX	".gz"
3088 #define GZ_SUFFIX_LEN	3
3089 #endif
3090 
3091 static char corefilename[MAXPATHLEN] = {"%N.core"};
3092 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RWTUN, corefilename,
3093     sizeof(corefilename), "Process corefile name format string");
3094 
3095 /*
3096  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3097  * Expand the name described in corefilename, using name, uid, and pid
3098  * and open/create core file.
3099  * corefilename is a printf-like string, with three format specifiers:
3100  *	%N	name of process ("name")
3101  *	%P	process id (pid)
3102  *	%U	user id (uid)
3103  * For example, "%N.core" is the default; they can be disabled completely
3104  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3105  * This is controlled by the sysctl variable kern.corefile (see above).
3106  */
3107 static int
3108 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3109     int compress, struct vnode **vpp, char **namep)
3110 {
3111 	struct nameidata nd;
3112 	struct sbuf sb;
3113 	const char *format;
3114 	char *hostname, *name;
3115 	int indexpos, i, error, cmode, flags, oflags;
3116 
3117 	hostname = NULL;
3118 	format = corefilename;
3119 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3120 	indexpos = -1;
3121 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3122 	for (i = 0; format[i] != '\0'; i++) {
3123 		switch (format[i]) {
3124 		case '%':	/* Format character */
3125 			i++;
3126 			switch (format[i]) {
3127 			case '%':
3128 				sbuf_putc(&sb, '%');
3129 				break;
3130 			case 'H':	/* hostname */
3131 				if (hostname == NULL) {
3132 					hostname = malloc(MAXHOSTNAMELEN,
3133 					    M_TEMP, M_WAITOK);
3134 				}
3135 				getcredhostname(td->td_ucred, hostname,
3136 				    MAXHOSTNAMELEN);
3137 				sbuf_printf(&sb, "%s", hostname);
3138 				break;
3139 			case 'I':	/* autoincrementing index */
3140 				sbuf_printf(&sb, "0");
3141 				indexpos = sbuf_len(&sb) - 1;
3142 				break;
3143 			case 'N':	/* process name */
3144 				sbuf_printf(&sb, "%s", comm);
3145 				break;
3146 			case 'P':	/* process id */
3147 				sbuf_printf(&sb, "%u", pid);
3148 				break;
3149 			case 'U':	/* user id */
3150 				sbuf_printf(&sb, "%u", uid);
3151 				break;
3152 			default:
3153 				log(LOG_ERR,
3154 				    "Unknown format character %c in "
3155 				    "corename `%s'\n", format[i], format);
3156 				break;
3157 			}
3158 			break;
3159 		default:
3160 			sbuf_putc(&sb, format[i]);
3161 			break;
3162 		}
3163 	}
3164 	free(hostname, M_TEMP);
3165 #ifdef COMPRESS_USER_CORES
3166 	if (compress)
3167 		sbuf_printf(&sb, GZ_SUFFIX);
3168 #endif
3169 	if (sbuf_error(&sb) != 0) {
3170 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3171 		    "long\n", (long)pid, comm, (u_long)uid);
3172 		sbuf_delete(&sb);
3173 		free(name, M_TEMP);
3174 		return (ENOMEM);
3175 	}
3176 	sbuf_finish(&sb);
3177 	sbuf_delete(&sb);
3178 
3179 	cmode = S_IRUSR | S_IWUSR;
3180 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3181 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3182 
3183 	/*
3184 	 * If the core format has a %I in it, then we need to check
3185 	 * for existing corefiles before returning a name.
3186 	 * To do this we iterate over 0..num_cores to find a
3187 	 * non-existing core file name to use.
3188 	 */
3189 	if (indexpos != -1) {
3190 		for (i = 0; i < num_cores; i++) {
3191 			flags = O_CREAT | O_EXCL | FWRITE | O_NOFOLLOW;
3192 			name[indexpos] = '0' + i;
3193 			NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3194 			error = vn_open_cred(&nd, &flags, cmode, oflags,
3195 			    td->td_ucred, NULL);
3196 			if (error) {
3197 				if (error == EEXIST)
3198 					continue;
3199 				log(LOG_ERR,
3200 				    "pid %d (%s), uid (%u):  Path `%s' failed "
3201 				    "on initial open test, error = %d\n",
3202 				    pid, comm, uid, name, error);
3203 			}
3204 			goto out;
3205 		}
3206 	}
3207 
3208 	flags = O_CREAT | FWRITE | O_NOFOLLOW;
3209 	NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3210 	error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred, NULL);
3211 out:
3212 	if (error) {
3213 #ifdef AUDIT
3214 		audit_proc_coredump(td, name, error);
3215 #endif
3216 		free(name, M_TEMP);
3217 		return (error);
3218 	}
3219 	NDFREE(&nd, NDF_ONLY_PNBUF);
3220 	*vpp = nd.ni_vp;
3221 	*namep = name;
3222 	return (0);
3223 }
3224 
3225 static int
3226 coredump_sanitise_path(const char *path)
3227 {
3228 	size_t i;
3229 
3230 	/*
3231 	 * Only send a subset of ASCII to devd(8) because it
3232 	 * might pass these strings to sh -c.
3233 	 */
3234 	for (i = 0; path[i]; i++)
3235 		if (!(isalpha(path[i]) || isdigit(path[i])) &&
3236 		    path[i] != '/' && path[i] != '.' &&
3237 		    path[i] != '-')
3238 			return (0);
3239 
3240 	return (1);
3241 }
3242 
3243 /*
3244  * Dump a process' core.  The main routine does some
3245  * policy checking, and creates the name of the coredump;
3246  * then it passes on a vnode and a size limit to the process-specific
3247  * coredump routine if there is one; if there _is not_ one, it returns
3248  * ENOSYS; otherwise it returns the error from the process-specific routine.
3249  */
3250 
3251 static int
3252 coredump(struct thread *td)
3253 {
3254 	struct proc *p = td->td_proc;
3255 	struct ucred *cred = td->td_ucred;
3256 	struct vnode *vp;
3257 	struct flock lf;
3258 	struct vattr vattr;
3259 	int error, error1, locked;
3260 	char *name;			/* name of corefile */
3261 	void *rl_cookie;
3262 	off_t limit;
3263 	int compress;
3264 	char data[MAXPATHLEN * 2 + 16]; /* space for devctl notification */
3265 	char *fullpath, *freepath = NULL;
3266 	size_t len;
3267 
3268 #ifdef COMPRESS_USER_CORES
3269 	compress = compress_user_cores;
3270 #else
3271 	compress = 0;
3272 #endif
3273 	PROC_LOCK_ASSERT(p, MA_OWNED);
3274 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3275 	_STOPEVENT(p, S_CORE, 0);
3276 
3277 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
3278 	    (p->p_flag2 & P2_NOTRACE) != 0) {
3279 		PROC_UNLOCK(p);
3280 		return (EFAULT);
3281 	}
3282 
3283 	/*
3284 	 * Note that the bulk of limit checking is done after
3285 	 * the corefile is created.  The exception is if the limit
3286 	 * for corefiles is 0, in which case we don't bother
3287 	 * creating the corefile at all.  This layout means that
3288 	 * a corefile is truncated instead of not being created,
3289 	 * if it is larger than the limit.
3290 	 */
3291 	limit = (off_t)lim_cur(p, RLIMIT_CORE);
3292 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
3293 		PROC_UNLOCK(p);
3294 		return (EFBIG);
3295 	}
3296 	PROC_UNLOCK(p);
3297 
3298 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td, compress,
3299 	    &vp, &name);
3300 	if (error != 0)
3301 		return (error);
3302 
3303 	/*
3304 	 * Don't dump to non-regular files or files with links.
3305 	 * Do not dump into system files.
3306 	 */
3307 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
3308 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0) {
3309 		VOP_UNLOCK(vp, 0);
3310 		error = EFAULT;
3311 		goto close;
3312 	}
3313 
3314 	VOP_UNLOCK(vp, 0);
3315 
3316 	/* Postpone other writers, including core dumps of other processes. */
3317 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
3318 
3319 	lf.l_whence = SEEK_SET;
3320 	lf.l_start = 0;
3321 	lf.l_len = 0;
3322 	lf.l_type = F_WRLCK;
3323 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
3324 
3325 	VATTR_NULL(&vattr);
3326 	vattr.va_size = 0;
3327 	if (set_core_nodump_flag)
3328 		vattr.va_flags = UF_NODUMP;
3329 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3330 	VOP_SETATTR(vp, &vattr, cred);
3331 	VOP_UNLOCK(vp, 0);
3332 	PROC_LOCK(p);
3333 	p->p_acflag |= ACORE;
3334 	PROC_UNLOCK(p);
3335 
3336 	if (p->p_sysent->sv_coredump != NULL) {
3337 		error = p->p_sysent->sv_coredump(td, vp, limit,
3338 		    compress ? IMGACT_CORE_COMPRESS : 0);
3339 	} else {
3340 		error = ENOSYS;
3341 	}
3342 
3343 	if (locked) {
3344 		lf.l_type = F_UNLCK;
3345 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3346 	}
3347 	vn_rangelock_unlock(vp, rl_cookie);
3348 close:
3349 	error1 = vn_close(vp, FWRITE, cred, td);
3350 	if (error == 0)
3351 		error = error1;
3352 	else
3353 		goto out;
3354 	/*
3355 	 * Notify the userland helper that a process triggered a core dump.
3356 	 * This allows the helper to run an automated debugging session.
3357 	 */
3358 	if (coredump_devctl == 0)
3359 		goto out;
3360 	if (vn_fullpath_global(td, p->p_textvp, &fullpath, &freepath) != 0)
3361 		goto out;
3362 	if (!coredump_sanitise_path(fullpath))
3363 		goto out;
3364 	snprintf(data, sizeof(data), "comm=%s ", fullpath);
3365 	free(freepath, M_TEMP);
3366 	freepath = NULL;
3367 	if (vn_fullpath_global(td, vp, &fullpath, &freepath) != 0)
3368 		goto out;
3369 	if (!coredump_sanitise_path(fullpath))
3370 		goto out;
3371 	strlcat(data, "core=", sizeof(data));
3372 	len = strlcat(data, fullpath, sizeof(data));
3373 	devctl_notify("kernel", "signal", "coredump", data);
3374 out:
3375 #ifdef AUDIT
3376 	audit_proc_coredump(td, name, error);
3377 #endif
3378 	free(freepath, M_TEMP);
3379 	free(name, M_TEMP);
3380 	return (error);
3381 }
3382 
3383 /*
3384  * Nonexistent system call-- signal process (may want to handle it).  Flag
3385  * error in case process won't see signal immediately (blocked or ignored).
3386  */
3387 #ifndef _SYS_SYSPROTO_H_
3388 struct nosys_args {
3389 	int	dummy;
3390 };
3391 #endif
3392 /* ARGSUSED */
3393 int
3394 nosys(td, args)
3395 	struct thread *td;
3396 	struct nosys_args *args;
3397 {
3398 	struct proc *p = td->td_proc;
3399 
3400 	PROC_LOCK(p);
3401 	tdsignal(td, SIGSYS);
3402 	PROC_UNLOCK(p);
3403 	return (ENOSYS);
3404 }
3405 
3406 /*
3407  * Send a SIGIO or SIGURG signal to a process or process group using stored
3408  * credentials rather than those of the current process.
3409  */
3410 void
3411 pgsigio(sigiop, sig, checkctty)
3412 	struct sigio **sigiop;
3413 	int sig, checkctty;
3414 {
3415 	ksiginfo_t ksi;
3416 	struct sigio *sigio;
3417 
3418 	ksiginfo_init(&ksi);
3419 	ksi.ksi_signo = sig;
3420 	ksi.ksi_code = SI_KERNEL;
3421 
3422 	SIGIO_LOCK();
3423 	sigio = *sigiop;
3424 	if (sigio == NULL) {
3425 		SIGIO_UNLOCK();
3426 		return;
3427 	}
3428 	if (sigio->sio_pgid > 0) {
3429 		PROC_LOCK(sigio->sio_proc);
3430 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
3431 			kern_psignal(sigio->sio_proc, sig);
3432 		PROC_UNLOCK(sigio->sio_proc);
3433 	} else if (sigio->sio_pgid < 0) {
3434 		struct proc *p;
3435 
3436 		PGRP_LOCK(sigio->sio_pgrp);
3437 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
3438 			PROC_LOCK(p);
3439 			if (p->p_state == PRS_NORMAL &&
3440 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
3441 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
3442 				kern_psignal(p, sig);
3443 			PROC_UNLOCK(p);
3444 		}
3445 		PGRP_UNLOCK(sigio->sio_pgrp);
3446 	}
3447 	SIGIO_UNLOCK();
3448 }
3449 
3450 static int
3451 filt_sigattach(struct knote *kn)
3452 {
3453 	struct proc *p = curproc;
3454 
3455 	kn->kn_ptr.p_proc = p;
3456 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
3457 
3458 	knlist_add(&p->p_klist, kn, 0);
3459 
3460 	return (0);
3461 }
3462 
3463 static void
3464 filt_sigdetach(struct knote *kn)
3465 {
3466 	struct proc *p = kn->kn_ptr.p_proc;
3467 
3468 	knlist_remove(&p->p_klist, kn, 0);
3469 }
3470 
3471 /*
3472  * signal knotes are shared with proc knotes, so we apply a mask to
3473  * the hint in order to differentiate them from process hints.  This
3474  * could be avoided by using a signal-specific knote list, but probably
3475  * isn't worth the trouble.
3476  */
3477 static int
3478 filt_signal(struct knote *kn, long hint)
3479 {
3480 
3481 	if (hint & NOTE_SIGNAL) {
3482 		hint &= ~NOTE_SIGNAL;
3483 
3484 		if (kn->kn_id == hint)
3485 			kn->kn_data++;
3486 	}
3487 	return (kn->kn_data != 0);
3488 }
3489 
3490 struct sigacts *
3491 sigacts_alloc(void)
3492 {
3493 	struct sigacts *ps;
3494 
3495 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
3496 	refcount_init(&ps->ps_refcnt, 1);
3497 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
3498 	return (ps);
3499 }
3500 
3501 void
3502 sigacts_free(struct sigacts *ps)
3503 {
3504 
3505 	if (refcount_release(&ps->ps_refcnt) == 0)
3506 		return;
3507 	mtx_destroy(&ps->ps_mtx);
3508 	free(ps, M_SUBPROC);
3509 }
3510 
3511 struct sigacts *
3512 sigacts_hold(struct sigacts *ps)
3513 {
3514 
3515 	refcount_acquire(&ps->ps_refcnt);
3516 	return (ps);
3517 }
3518 
3519 void
3520 sigacts_copy(struct sigacts *dest, struct sigacts *src)
3521 {
3522 
3523 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
3524 	mtx_lock(&src->ps_mtx);
3525 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
3526 	mtx_unlock(&src->ps_mtx);
3527 }
3528 
3529 int
3530 sigacts_shared(struct sigacts *ps)
3531 {
3532 
3533 	return (ps->ps_refcnt > 1);
3534 }
3535