xref: /openbsd-src/sys/kern/kern_sig.c (revision ef70a379631ec7e97481e1d0e500e21496e6c4aa)
1 /*	$OpenBSD: kern_sig.c,v 1.288 2021/11/12 17:57:13 cheloha Exp $	*/
2 /*	$NetBSD: kern_sig.c,v 1.54 1996/04/22 01:38:32 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1997 Theo de Raadt. All rights reserved.
6  * Copyright (c) 1982, 1986, 1989, 1991, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  * (c) UNIX System Laboratories, Inc.
9  * All or some portions of this file are derived from material licensed
10  * to the University of California by American Telephone and Telegraph
11  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12  * the permission of UNIX System Laboratories, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
39  */
40 
41 #include <sys/param.h>
42 #include <sys/signalvar.h>
43 #include <sys/resourcevar.h>
44 #include <sys/queue.h>
45 #include <sys/namei.h>
46 #include <sys/vnode.h>
47 #include <sys/event.h>
48 #include <sys/proc.h>
49 #include <sys/systm.h>
50 #include <sys/acct.h>
51 #include <sys/fcntl.h>
52 #include <sys/filedesc.h>
53 #include <sys/kernel.h>
54 #include <sys/wait.h>
55 #include <sys/ktrace.h>
56 #include <sys/stat.h>
57 #include <sys/core.h>
58 #include <sys/malloc.h>
59 #include <sys/pool.h>
60 #include <sys/ptrace.h>
61 #include <sys/sched.h>
62 #include <sys/user.h>
63 #include <sys/syslog.h>
64 #include <sys/ttycom.h>
65 #include <sys/pledge.h>
66 #include <sys/witness.h>
67 
68 #include <sys/mount.h>
69 #include <sys/syscallargs.h>
70 
71 #include <uvm/uvm_extern.h>
72 #include <machine/tcb.h>
73 
74 int	filt_sigattach(struct knote *kn);
75 void	filt_sigdetach(struct knote *kn);
76 int	filt_signal(struct knote *kn, long hint);
77 
78 const struct filterops sig_filtops = {
79 	.f_flags	= 0,
80 	.f_attach	= filt_sigattach,
81 	.f_detach	= filt_sigdetach,
82 	.f_event	= filt_signal,
83 };
84 
85 const int sigprop[NSIG + 1] = {
86 	0,			/* unused */
87 	SA_KILL,		/* SIGHUP */
88 	SA_KILL,		/* SIGINT */
89 	SA_KILL|SA_CORE,	/* SIGQUIT */
90 	SA_KILL|SA_CORE,	/* SIGILL */
91 	SA_KILL|SA_CORE,	/* SIGTRAP */
92 	SA_KILL|SA_CORE,	/* SIGABRT */
93 	SA_KILL|SA_CORE,	/* SIGEMT */
94 	SA_KILL|SA_CORE,	/* SIGFPE */
95 	SA_KILL,		/* SIGKILL */
96 	SA_KILL|SA_CORE,	/* SIGBUS */
97 	SA_KILL|SA_CORE,	/* SIGSEGV */
98 	SA_KILL|SA_CORE,	/* SIGSYS */
99 	SA_KILL,		/* SIGPIPE */
100 	SA_KILL,		/* SIGALRM */
101 	SA_KILL,		/* SIGTERM */
102 	SA_IGNORE,		/* SIGURG */
103 	SA_STOP,		/* SIGSTOP */
104 	SA_STOP|SA_TTYSTOP,	/* SIGTSTP */
105 	SA_IGNORE|SA_CONT,	/* SIGCONT */
106 	SA_IGNORE,		/* SIGCHLD */
107 	SA_STOP|SA_TTYSTOP,	/* SIGTTIN */
108 	SA_STOP|SA_TTYSTOP,	/* SIGTTOU */
109 	SA_IGNORE,		/* SIGIO */
110 	SA_KILL,		/* SIGXCPU */
111 	SA_KILL,		/* SIGXFSZ */
112 	SA_KILL,		/* SIGVTALRM */
113 	SA_KILL,		/* SIGPROF */
114 	SA_IGNORE,		/* SIGWINCH  */
115 	SA_IGNORE,		/* SIGINFO */
116 	SA_KILL,		/* SIGUSR1 */
117 	SA_KILL,		/* SIGUSR2 */
118 	SA_IGNORE,		/* SIGTHR */
119 };
120 
121 #define	CONTSIGMASK	(sigmask(SIGCONT))
122 #define	STOPSIGMASK	(sigmask(SIGSTOP) | sigmask(SIGTSTP) | \
123 			    sigmask(SIGTTIN) | sigmask(SIGTTOU))
124 
125 void setsigvec(struct proc *, int, struct sigaction *);
126 
127 void proc_stop(struct proc *p, int);
128 void proc_stop_sweep(void *);
129 void *proc_stop_si;
130 
131 void postsig(struct proc *, int);
132 int cansignal(struct proc *, struct process *, int);
133 
134 struct pool sigacts_pool;	/* memory pool for sigacts structures */
135 
136 void sigio_del(struct sigiolst *);
137 void sigio_unlink(struct sigio_ref *, struct sigiolst *);
138 struct mutex sigio_lock = MUTEX_INITIALIZER(IPL_HIGH);
139 
140 /*
141  * Can thread p, send the signal signum to process qr?
142  */
143 int
144 cansignal(struct proc *p, struct process *qr, int signum)
145 {
146 	struct process *pr = p->p_p;
147 	struct ucred *uc = p->p_ucred;
148 	struct ucred *quc = qr->ps_ucred;
149 
150 	if (uc->cr_uid == 0)
151 		return (1);		/* root can always signal */
152 
153 	if (pr == qr)
154 		return (1);		/* process can always signal itself */
155 
156 	/* optimization: if the same creds then the tests below will pass */
157 	if (uc == quc)
158 		return (1);
159 
160 	if (signum == SIGCONT && qr->ps_session == pr->ps_session)
161 		return (1);		/* SIGCONT in session */
162 
163 	/*
164 	 * Using kill(), only certain signals can be sent to setugid
165 	 * child processes
166 	 */
167 	if (qr->ps_flags & PS_SUGID) {
168 		switch (signum) {
169 		case 0:
170 		case SIGKILL:
171 		case SIGINT:
172 		case SIGTERM:
173 		case SIGALRM:
174 		case SIGSTOP:
175 		case SIGTTIN:
176 		case SIGTTOU:
177 		case SIGTSTP:
178 		case SIGHUP:
179 		case SIGUSR1:
180 		case SIGUSR2:
181 			if (uc->cr_ruid == quc->cr_ruid ||
182 			    uc->cr_uid == quc->cr_ruid)
183 				return (1);
184 		}
185 		return (0);
186 	}
187 
188 	if (uc->cr_ruid == quc->cr_ruid ||
189 	    uc->cr_ruid == quc->cr_svuid ||
190 	    uc->cr_uid == quc->cr_ruid ||
191 	    uc->cr_uid == quc->cr_svuid)
192 		return (1);
193 	return (0);
194 }
195 
196 /*
197  * Initialize signal-related data structures.
198  */
199 void
200 signal_init(void)
201 {
202 	proc_stop_si = softintr_establish(IPL_SOFTCLOCK, proc_stop_sweep,
203 	    NULL);
204 	if (proc_stop_si == NULL)
205 		panic("signal_init failed to register softintr");
206 
207 	pool_init(&sigacts_pool, sizeof(struct sigacts), 0, IPL_NONE,
208 	    PR_WAITOK, "sigapl", NULL);
209 }
210 
211 /*
212  * Create an initial sigacts structure, using the same signal state
213  * as pr.
214  */
215 struct sigacts *
216 sigactsinit(struct process *pr)
217 {
218 	struct sigacts *ps;
219 
220 	ps = pool_get(&sigacts_pool, PR_WAITOK);
221 	memcpy(ps, pr->ps_sigacts, sizeof(struct sigacts));
222 	return (ps);
223 }
224 
225 /*
226  * Initialize a new sigaltstack structure.
227  */
228 void
229 sigstkinit(struct sigaltstack *ss)
230 {
231 	ss->ss_flags = SS_DISABLE;
232 	ss->ss_size = 0;
233 	ss->ss_sp = NULL;
234 }
235 
236 /*
237  * Release a sigacts structure.
238  */
239 void
240 sigactsfree(struct process *pr)
241 {
242 	struct sigacts *ps = pr->ps_sigacts;
243 
244 	pr->ps_sigacts = NULL;
245 
246 	pool_put(&sigacts_pool, ps);
247 }
248 
249 int
250 sys_sigaction(struct proc *p, void *v, register_t *retval)
251 {
252 	struct sys_sigaction_args /* {
253 		syscallarg(int) signum;
254 		syscallarg(const struct sigaction *) nsa;
255 		syscallarg(struct sigaction *) osa;
256 	} */ *uap = v;
257 	struct sigaction vec;
258 #ifdef KTRACE
259 	struct sigaction ovec;
260 #endif
261 	struct sigaction *sa;
262 	const struct sigaction *nsa;
263 	struct sigaction *osa;
264 	struct sigacts *ps = p->p_p->ps_sigacts;
265 	int signum;
266 	int bit, error;
267 
268 	signum = SCARG(uap, signum);
269 	nsa = SCARG(uap, nsa);
270 	osa = SCARG(uap, osa);
271 
272 	if (signum <= 0 || signum >= NSIG ||
273 	    (nsa && (signum == SIGKILL || signum == SIGSTOP)))
274 		return (EINVAL);
275 	sa = &vec;
276 	if (osa) {
277 		sa->sa_handler = ps->ps_sigact[signum];
278 		sa->sa_mask = ps->ps_catchmask[signum];
279 		bit = sigmask(signum);
280 		sa->sa_flags = 0;
281 		if ((ps->ps_sigonstack & bit) != 0)
282 			sa->sa_flags |= SA_ONSTACK;
283 		if ((ps->ps_sigintr & bit) == 0)
284 			sa->sa_flags |= SA_RESTART;
285 		if ((ps->ps_sigreset & bit) != 0)
286 			sa->sa_flags |= SA_RESETHAND;
287 		if ((ps->ps_siginfo & bit) != 0)
288 			sa->sa_flags |= SA_SIGINFO;
289 		if (signum == SIGCHLD) {
290 			if ((ps->ps_sigflags & SAS_NOCLDSTOP) != 0)
291 				sa->sa_flags |= SA_NOCLDSTOP;
292 			if ((ps->ps_sigflags & SAS_NOCLDWAIT) != 0)
293 				sa->sa_flags |= SA_NOCLDWAIT;
294 		}
295 		if ((sa->sa_mask & bit) == 0)
296 			sa->sa_flags |= SA_NODEFER;
297 		sa->sa_mask &= ~bit;
298 		error = copyout(sa, osa, sizeof (vec));
299 		if (error)
300 			return (error);
301 #ifdef KTRACE
302 		if (KTRPOINT(p, KTR_STRUCT))
303 			ovec = vec;
304 #endif
305 	}
306 	if (nsa) {
307 		error = copyin(nsa, sa, sizeof (vec));
308 		if (error)
309 			return (error);
310 #ifdef KTRACE
311 		if (KTRPOINT(p, KTR_STRUCT))
312 			ktrsigaction(p, sa);
313 #endif
314 		setsigvec(p, signum, sa);
315 	}
316 #ifdef KTRACE
317 	if (osa && KTRPOINT(p, KTR_STRUCT))
318 		ktrsigaction(p, &ovec);
319 #endif
320 	return (0);
321 }
322 
323 void
324 setsigvec(struct proc *p, int signum, struct sigaction *sa)
325 {
326 	struct sigacts *ps = p->p_p->ps_sigacts;
327 	int bit;
328 	int s;
329 
330 	bit = sigmask(signum);
331 	/*
332 	 * Change setting atomically.
333 	 */
334 	s = splhigh();
335 	ps->ps_sigact[signum] = sa->sa_handler;
336 	if ((sa->sa_flags & SA_NODEFER) == 0)
337 		sa->sa_mask |= sigmask(signum);
338 	ps->ps_catchmask[signum] = sa->sa_mask &~ sigcantmask;
339 	if (signum == SIGCHLD) {
340 		if (sa->sa_flags & SA_NOCLDSTOP)
341 			atomic_setbits_int(&ps->ps_sigflags, SAS_NOCLDSTOP);
342 		else
343 			atomic_clearbits_int(&ps->ps_sigflags, SAS_NOCLDSTOP);
344 		/*
345 		 * If the SA_NOCLDWAIT flag is set or the handler
346 		 * is SIG_IGN we reparent the dying child to PID 1
347 		 * (init) which will reap the zombie.  Because we use
348 		 * init to do our dirty work we never set SAS_NOCLDWAIT
349 		 * for PID 1.
350 		 * XXX exit1 rework means this is unnecessary?
351 		 */
352 		if (initprocess->ps_sigacts != ps &&
353 		    ((sa->sa_flags & SA_NOCLDWAIT) ||
354 		    sa->sa_handler == SIG_IGN))
355 			atomic_setbits_int(&ps->ps_sigflags, SAS_NOCLDWAIT);
356 		else
357 			atomic_clearbits_int(&ps->ps_sigflags, SAS_NOCLDWAIT);
358 	}
359 	if ((sa->sa_flags & SA_RESETHAND) != 0)
360 		ps->ps_sigreset |= bit;
361 	else
362 		ps->ps_sigreset &= ~bit;
363 	if ((sa->sa_flags & SA_SIGINFO) != 0)
364 		ps->ps_siginfo |= bit;
365 	else
366 		ps->ps_siginfo &= ~bit;
367 	if ((sa->sa_flags & SA_RESTART) == 0)
368 		ps->ps_sigintr |= bit;
369 	else
370 		ps->ps_sigintr &= ~bit;
371 	if ((sa->sa_flags & SA_ONSTACK) != 0)
372 		ps->ps_sigonstack |= bit;
373 	else
374 		ps->ps_sigonstack &= ~bit;
375 	/*
376 	 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
377 	 * and for signals set to SIG_DFL where the default is to ignore.
378 	 * However, don't put SIGCONT in ps_sigignore,
379 	 * as we have to restart the process.
380 	 */
381 	if (sa->sa_handler == SIG_IGN ||
382 	    (sigprop[signum] & SA_IGNORE && sa->sa_handler == SIG_DFL)) {
383 		atomic_clearbits_int(&p->p_siglist, bit);
384 		atomic_clearbits_int(&p->p_p->ps_siglist, bit);
385 		if (signum != SIGCONT)
386 			ps->ps_sigignore |= bit;	/* easier in psignal */
387 		ps->ps_sigcatch &= ~bit;
388 	} else {
389 		ps->ps_sigignore &= ~bit;
390 		if (sa->sa_handler == SIG_DFL)
391 			ps->ps_sigcatch &= ~bit;
392 		else
393 			ps->ps_sigcatch |= bit;
394 	}
395 	splx(s);
396 }
397 
398 /*
399  * Initialize signal state for process 0;
400  * set to ignore signals that are ignored by default.
401  */
402 void
403 siginit(struct sigacts *ps)
404 {
405 	int i;
406 
407 	for (i = 0; i < NSIG; i++)
408 		if (sigprop[i] & SA_IGNORE && i != SIGCONT)
409 			ps->ps_sigignore |= sigmask(i);
410 	ps->ps_sigflags = SAS_NOCLDWAIT | SAS_NOCLDSTOP;
411 }
412 
413 /*
414  * Reset signals for an exec by the specified thread.
415  */
416 void
417 execsigs(struct proc *p)
418 {
419 	struct sigacts *ps;
420 	int nc, mask;
421 
422 	ps = p->p_p->ps_sigacts;
423 
424 	/*
425 	 * Reset caught signals.  Held signals remain held
426 	 * through p_sigmask (unless they were caught,
427 	 * and are now ignored by default).
428 	 */
429 	while (ps->ps_sigcatch) {
430 		nc = ffs((long)ps->ps_sigcatch);
431 		mask = sigmask(nc);
432 		ps->ps_sigcatch &= ~mask;
433 		if (sigprop[nc] & SA_IGNORE) {
434 			if (nc != SIGCONT)
435 				ps->ps_sigignore |= mask;
436 			atomic_clearbits_int(&p->p_siglist, mask);
437 			atomic_clearbits_int(&p->p_p->ps_siglist, mask);
438 		}
439 		ps->ps_sigact[nc] = SIG_DFL;
440 	}
441 	/*
442 	 * Reset stack state to the user stack.
443 	 * Clear set of signals caught on the signal stack.
444 	 */
445 	sigstkinit(&p->p_sigstk);
446 	atomic_clearbits_int(&ps->ps_sigflags, SAS_NOCLDWAIT);
447 	if (ps->ps_sigact[SIGCHLD] == SIG_IGN)
448 		ps->ps_sigact[SIGCHLD] = SIG_DFL;
449 }
450 
451 /*
452  * Manipulate signal mask.
453  * Note that we receive new mask, not pointer,
454  * and return old mask as return value;
455  * the library stub does the rest.
456  */
457 int
458 sys_sigprocmask(struct proc *p, void *v, register_t *retval)
459 {
460 	struct sys_sigprocmask_args /* {
461 		syscallarg(int) how;
462 		syscallarg(sigset_t) mask;
463 	} */ *uap = v;
464 	int error = 0;
465 	sigset_t mask;
466 
467 	KASSERT(p == curproc);
468 
469 	*retval = p->p_sigmask;
470 	mask = SCARG(uap, mask) &~ sigcantmask;
471 
472 	switch (SCARG(uap, how)) {
473 	case SIG_BLOCK:
474 		atomic_setbits_int(&p->p_sigmask, mask);
475 		break;
476 	case SIG_UNBLOCK:
477 		atomic_clearbits_int(&p->p_sigmask, mask);
478 		break;
479 	case SIG_SETMASK:
480 		p->p_sigmask = mask;
481 		break;
482 	default:
483 		error = EINVAL;
484 		break;
485 	}
486 	return (error);
487 }
488 
489 int
490 sys_sigpending(struct proc *p, void *v, register_t *retval)
491 {
492 
493 	*retval = p->p_siglist | p->p_p->ps_siglist;
494 	return (0);
495 }
496 
497 /*
498  * Temporarily replace calling proc's signal mask for the duration of a
499  * system call.  Original signal mask will be restored by userret().
500  */
501 void
502 dosigsuspend(struct proc *p, sigset_t newmask)
503 {
504 	KASSERT(p == curproc);
505 
506 	p->p_oldmask = p->p_sigmask;
507 	atomic_setbits_int(&p->p_flag, P_SIGSUSPEND);
508 	p->p_sigmask = newmask;
509 }
510 
511 /*
512  * Suspend thread until signal, providing mask to be set
513  * in the meantime.  Note nonstandard calling convention:
514  * libc stub passes mask, not pointer, to save a copyin.
515  */
516 int
517 sys_sigsuspend(struct proc *p, void *v, register_t *retval)
518 {
519 	struct sys_sigsuspend_args /* {
520 		syscallarg(int) mask;
521 	} */ *uap = v;
522 
523 	dosigsuspend(p, SCARG(uap, mask) &~ sigcantmask);
524 	while (tsleep_nsec(&nowake, PPAUSE|PCATCH, "sigsusp", INFSLP) == 0)
525 		continue;
526 	/* always return EINTR rather than ERESTART... */
527 	return (EINTR);
528 }
529 
530 int
531 sigonstack(size_t stack)
532 {
533 	const struct sigaltstack *ss = &curproc->p_sigstk;
534 
535 	return (ss->ss_flags & SS_DISABLE ? 0 :
536 	    (stack - (size_t)ss->ss_sp < ss->ss_size));
537 }
538 
539 int
540 sys_sigaltstack(struct proc *p, void *v, register_t *retval)
541 {
542 	struct sys_sigaltstack_args /* {
543 		syscallarg(const struct sigaltstack *) nss;
544 		syscallarg(struct sigaltstack *) oss;
545 	} */ *uap = v;
546 	struct sigaltstack ss;
547 	const struct sigaltstack *nss;
548 	struct sigaltstack *oss;
549 	int onstack = sigonstack(PROC_STACK(p));
550 	int error;
551 
552 	nss = SCARG(uap, nss);
553 	oss = SCARG(uap, oss);
554 
555 	if (oss != NULL) {
556 		ss = p->p_sigstk;
557 		if (onstack)
558 			ss.ss_flags |= SS_ONSTACK;
559 		if ((error = copyout(&ss, oss, sizeof(ss))))
560 			return (error);
561 	}
562 	if (nss == NULL)
563 		return (0);
564 	error = copyin(nss, &ss, sizeof(ss));
565 	if (error)
566 		return (error);
567 	if (onstack)
568 		return (EPERM);
569 	if (ss.ss_flags & ~SS_DISABLE)
570 		return (EINVAL);
571 	if (ss.ss_flags & SS_DISABLE) {
572 		p->p_sigstk.ss_flags = ss.ss_flags;
573 		return (0);
574 	}
575 	if (ss.ss_size < MINSIGSTKSZ)
576 		return (ENOMEM);
577 
578 	error = uvm_map_remap_as_stack(p, (vaddr_t)ss.ss_sp, ss.ss_size);
579 	if (error)
580 		return (error);
581 
582 	p->p_sigstk = ss;
583 	return (0);
584 }
585 
586 int
587 sys_kill(struct proc *cp, void *v, register_t *retval)
588 {
589 	struct sys_kill_args /* {
590 		syscallarg(int) pid;
591 		syscallarg(int) signum;
592 	} */ *uap = v;
593 	struct process *pr;
594 	int pid = SCARG(uap, pid);
595 	int signum = SCARG(uap, signum);
596 	int error;
597 	int zombie = 0;
598 
599 	if ((error = pledge_kill(cp, pid)) != 0)
600 		return (error);
601 	if (((u_int)signum) >= NSIG)
602 		return (EINVAL);
603 	if (pid > 0) {
604 		if ((pr = prfind(pid)) == NULL) {
605 			if ((pr = zombiefind(pid)) == NULL)
606 				return (ESRCH);
607 			else
608 				zombie = 1;
609 		}
610 		if (!cansignal(cp, pr, signum))
611 			return (EPERM);
612 
613 		/* kill single process */
614 		if (signum && !zombie)
615 			prsignal(pr, signum);
616 		return (0);
617 	}
618 	switch (pid) {
619 	case -1:		/* broadcast signal */
620 		return (killpg1(cp, signum, 0, 1));
621 	case 0:			/* signal own process group */
622 		return (killpg1(cp, signum, 0, 0));
623 	default:		/* negative explicit process group */
624 		return (killpg1(cp, signum, -pid, 0));
625 	}
626 }
627 
628 int
629 sys_thrkill(struct proc *cp, void *v, register_t *retval)
630 {
631 	struct sys_thrkill_args /* {
632 		syscallarg(pid_t) tid;
633 		syscallarg(int) signum;
634 		syscallarg(void *) tcb;
635 	} */ *uap = v;
636 	struct proc *p;
637 	int tid = SCARG(uap, tid);
638 	int signum = SCARG(uap, signum);
639 	void *tcb;
640 
641 	if (((u_int)signum) >= NSIG)
642 		return (EINVAL);
643 	if (tid > THREAD_PID_OFFSET) {
644 		if ((p = tfind(tid - THREAD_PID_OFFSET)) == NULL)
645 			return (ESRCH);
646 
647 		/* can only kill threads in the same process */
648 		if (p->p_p != cp->p_p)
649 			return (ESRCH);
650 	} else if (tid == 0)
651 		p = cp;
652 	else
653 		return (EINVAL);
654 
655 	/* optionally require the target thread to have the given tcb addr */
656 	tcb = SCARG(uap, tcb);
657 	if (tcb != NULL && tcb != TCB_GET(p))
658 		return (ESRCH);
659 
660 	if (signum)
661 		ptsignal(p, signum, STHREAD);
662 	return (0);
663 }
664 
665 /*
666  * Common code for kill process group/broadcast kill.
667  * cp is calling process.
668  */
669 int
670 killpg1(struct proc *cp, int signum, int pgid, int all)
671 {
672 	struct process *pr;
673 	struct pgrp *pgrp;
674 	int nfound = 0;
675 
676 	if (all) {
677 		/*
678 		 * broadcast
679 		 */
680 		LIST_FOREACH(pr, &allprocess, ps_list) {
681 			if (pr->ps_pid <= 1 ||
682 			    pr->ps_flags & (PS_SYSTEM | PS_NOBROADCASTKILL) ||
683 			    pr == cp->p_p || !cansignal(cp, pr, signum))
684 				continue;
685 			nfound++;
686 			if (signum)
687 				prsignal(pr, signum);
688 		}
689 	} else {
690 		if (pgid == 0)
691 			/*
692 			 * zero pgid means send to my process group.
693 			 */
694 			pgrp = cp->p_p->ps_pgrp;
695 		else {
696 			pgrp = pgfind(pgid);
697 			if (pgrp == NULL)
698 				return (ESRCH);
699 		}
700 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist) {
701 			if (pr->ps_pid <= 1 || pr->ps_flags & PS_SYSTEM ||
702 			    !cansignal(cp, pr, signum))
703 				continue;
704 			nfound++;
705 			if (signum)
706 				prsignal(pr, signum);
707 		}
708 	}
709 	return (nfound ? 0 : ESRCH);
710 }
711 
712 #define CANDELIVER(uid, euid, pr) \
713 	(euid == 0 || \
714 	(uid) == (pr)->ps_ucred->cr_ruid || \
715 	(uid) == (pr)->ps_ucred->cr_svuid || \
716 	(uid) == (pr)->ps_ucred->cr_uid || \
717 	(euid) == (pr)->ps_ucred->cr_ruid || \
718 	(euid) == (pr)->ps_ucred->cr_svuid || \
719 	(euid) == (pr)->ps_ucred->cr_uid)
720 
721 #define CANSIGIO(cr, pr) \
722 	CANDELIVER((cr)->cr_ruid, (cr)->cr_uid, (pr))
723 
724 /*
725  * Send a signal to a process group.  If checktty is 1,
726  * limit to members which have a controlling terminal.
727  */
728 void
729 pgsignal(struct pgrp *pgrp, int signum, int checkctty)
730 {
731 	struct process *pr;
732 
733 	if (pgrp)
734 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist)
735 			if (checkctty == 0 || pr->ps_flags & PS_CONTROLT)
736 				prsignal(pr, signum);
737 }
738 
739 /*
740  * Send a SIGIO or SIGURG signal to a process or process group using stored
741  * credentials rather than those of the current process.
742  */
743 void
744 pgsigio(struct sigio_ref *sir, int sig, int checkctty)
745 {
746 	struct process *pr;
747 	struct sigio *sigio;
748 
749 	if (sir->sir_sigio == NULL)
750 		return;
751 
752 	KERNEL_LOCK();
753 	mtx_enter(&sigio_lock);
754 	sigio = sir->sir_sigio;
755 	if (sigio == NULL)
756 		goto out;
757 	if (sigio->sio_pgid > 0) {
758 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc))
759 			prsignal(sigio->sio_proc, sig);
760 	} else if (sigio->sio_pgid < 0) {
761 		LIST_FOREACH(pr, &sigio->sio_pgrp->pg_members, ps_pglist) {
762 			if (CANSIGIO(sigio->sio_ucred, pr) &&
763 			    (checkctty == 0 || (pr->ps_flags & PS_CONTROLT)))
764 				prsignal(pr, sig);
765 		}
766 	}
767 out:
768 	mtx_leave(&sigio_lock);
769 	KERNEL_UNLOCK();
770 }
771 
772 /*
773  * Recalculate the signal mask and reset the signal disposition after
774  * usermode frame for delivery is formed.
775  */
776 void
777 postsig_done(struct proc *p, int signum, struct sigacts *ps)
778 {
779 	int mask = sigmask(signum);
780 
781 	KERNEL_ASSERT_LOCKED();
782 
783 	p->p_ru.ru_nsignals++;
784 	atomic_setbits_int(&p->p_sigmask, ps->ps_catchmask[signum]);
785 	if ((ps->ps_sigreset & mask) != 0) {
786 		ps->ps_sigcatch &= ~mask;
787 		if (signum != SIGCONT && sigprop[signum] & SA_IGNORE)
788 			ps->ps_sigignore |= mask;
789 		ps->ps_sigact[signum] = SIG_DFL;
790 	}
791 }
792 
793 /*
794  * Send a signal caused by a trap to the current thread
795  * If it will be caught immediately, deliver it with correct code.
796  * Otherwise, post it normally.
797  */
798 void
799 trapsignal(struct proc *p, int signum, u_long trapno, int code,
800     union sigval sigval)
801 {
802 	struct process *pr = p->p_p;
803 	struct sigacts *ps = pr->ps_sigacts;
804 	int mask;
805 
806 	KERNEL_LOCK();
807 	switch (signum) {
808 	case SIGILL:
809 	case SIGBUS:
810 	case SIGSEGV:
811 		pr->ps_acflag |= ATRAP;
812 		break;
813 	}
814 
815 	mask = sigmask(signum);
816 	if ((pr->ps_flags & PS_TRACED) == 0 &&
817 	    (ps->ps_sigcatch & mask) != 0 &&
818 	    (p->p_sigmask & mask) == 0) {
819 		siginfo_t si;
820 		int info = (ps->ps_siginfo & mask) != 0;
821 		int onstack = (ps->ps_sigonstack & mask) != 0;
822 
823 		initsiginfo(&si, signum, trapno, code, sigval);
824 #ifdef KTRACE
825 		if (KTRPOINT(p, KTR_PSIG)) {
826 			ktrpsig(p, signum, ps->ps_sigact[signum],
827 			    p->p_sigmask, code, &si);
828 		}
829 #endif
830 		if (sendsig(ps->ps_sigact[signum], signum, p->p_sigmask, &si,
831 		    info, onstack)) {
832 			sigexit(p, SIGILL);
833 			/* NOTREACHED */
834 		}
835 		postsig_done(p, signum, ps);
836 	} else {
837 		p->p_sisig = signum;
838 		p->p_sitrapno = trapno;	/* XXX for core dump/debugger */
839 		p->p_sicode = code;
840 		p->p_sigval = sigval;
841 
842 		/*
843 		 * Signals like SIGBUS and SIGSEGV should not, when
844 		 * generated by the kernel, be ignorable or blockable.
845 		 * If it is and we're not being traced, then just kill
846 		 * the process.
847 		 * After vfs_shutdown(9), init(8) cannot receive signals
848 		 * because new code pages of the signal handler cannot be
849 		 * mapped from halted storage.  init(8) may not die or the
850 		 * kernel panics.  Better loop between signal handler and
851 		 * page fault trap until the machine is halted.
852 		 */
853 		if ((pr->ps_flags & PS_TRACED) == 0 &&
854 		    (sigprop[signum] & SA_KILL) &&
855 		    ((p->p_sigmask & mask) || (ps->ps_sigignore & mask)) &&
856 		    pr->ps_pid != 1)
857 			sigexit(p, signum);
858 		ptsignal(p, signum, STHREAD);
859 	}
860 	KERNEL_UNLOCK();
861 }
862 
863 /*
864  * Send the signal to the process.  If the signal has an action, the action
865  * is usually performed by the target process rather than the caller; we add
866  * the signal to the set of pending signals for the process.
867  *
868  * Exceptions:
869  *   o When a stop signal is sent to a sleeping process that takes the
870  *     default action, the process is stopped without awakening it.
871  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
872  *     regardless of the signal action (eg, blocked or ignored).
873  *
874  * Other ignored signals are discarded immediately.
875  */
876 void
877 psignal(struct proc *p, int signum)
878 {
879 	ptsignal(p, signum, SPROCESS);
880 }
881 
882 /*
883  * type = SPROCESS	process signal, can be diverted (sigwait())
884  * type = STHREAD	thread signal, but should be propagated if unhandled
885  * type = SPROPAGATED	propagated to this thread, so don't propagate again
886  */
887 void
888 ptsignal(struct proc *p, int signum, enum signal_type type)
889 {
890 	int s, prop;
891 	sig_t action;
892 	int mask;
893 	int *siglist;
894 	struct process *pr = p->p_p;
895 	struct proc *q;
896 	int wakeparent = 0;
897 
898 	KERNEL_ASSERT_LOCKED();
899 
900 #ifdef DIAGNOSTIC
901 	if ((u_int)signum >= NSIG || signum == 0)
902 		panic("psignal signal number");
903 #endif
904 
905 	/* Ignore signal if the target process is exiting */
906 	if (pr->ps_flags & PS_EXITING)
907 		return;
908 
909 	mask = sigmask(signum);
910 
911 	if (type == SPROCESS) {
912 		/* Accept SIGKILL to coredumping processes */
913 		if (pr->ps_flags & PS_COREDUMP && signum == SIGKILL) {
914 			atomic_setbits_int(&pr->ps_siglist, mask);
915 			return;
916 		}
917 
918 		/*
919 		 * If the current thread can process the signal
920 		 * immediately (it's unblocked) then have it take it.
921 		 */
922 		q = curproc;
923 		if (q != NULL && q->p_p == pr && (q->p_flag & P_WEXIT) == 0 &&
924 		    (q->p_sigmask & mask) == 0)
925 			p = q;
926 		else {
927 			/*
928 			 * A process-wide signal can be diverted to a
929 			 * different thread that's in sigwait() for this
930 			 * signal.  If there isn't such a thread, then
931 			 * pick a thread that doesn't have it blocked so
932 			 * that the stop/kill consideration isn't
933 			 * delayed.  Otherwise, mark it pending on the
934 			 * main thread.
935 			 */
936 			TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
937 				/* ignore exiting threads */
938 				if (q->p_flag & P_WEXIT)
939 					continue;
940 
941 				/* skip threads that have the signal blocked */
942 				if ((q->p_sigmask & mask) != 0)
943 					continue;
944 
945 				/* okay, could send to this thread */
946 				p = q;
947 
948 				/*
949 				 * sigsuspend, sigwait, ppoll/pselect, etc?
950 				 * Definitely go to this thread, as it's
951 				 * already blocked in the kernel.
952 				 */
953 				if (q->p_flag & P_SIGSUSPEND)
954 					break;
955 			}
956 		}
957 	}
958 
959 	if (type != SPROPAGATED)
960 		KNOTE(&pr->ps_klist, NOTE_SIGNAL | signum);
961 
962 	prop = sigprop[signum];
963 
964 	/*
965 	 * If proc is traced, always give parent a chance.
966 	 */
967 	if (pr->ps_flags & PS_TRACED) {
968 		action = SIG_DFL;
969 	} else {
970 		/*
971 		 * If the signal is being ignored,
972 		 * then we forget about it immediately.
973 		 * (Note: we don't set SIGCONT in ps_sigignore,
974 		 * and if it is set to SIG_IGN,
975 		 * action will be SIG_DFL here.)
976 		 */
977 		if (pr->ps_sigacts->ps_sigignore & mask)
978 			return;
979 		if (p->p_sigmask & mask) {
980 			action = SIG_HOLD;
981 		} else if (pr->ps_sigacts->ps_sigcatch & mask) {
982 			action = SIG_CATCH;
983 		} else {
984 			action = SIG_DFL;
985 
986 			if (prop & SA_KILL && pr->ps_nice > NZERO)
987 				 pr->ps_nice = NZERO;
988 
989 			/*
990 			 * If sending a tty stop signal to a member of an
991 			 * orphaned process group, discard the signal here if
992 			 * the action is default; don't stop the process below
993 			 * if sleeping, and don't clear any pending SIGCONT.
994 			 */
995 			if (prop & SA_TTYSTOP && pr->ps_pgrp->pg_jobc == 0)
996 				return;
997 		}
998 	}
999 	/*
1000 	 * If delivered to process, mark as pending there.  Continue and stop
1001 	 * signals will be propagated to all threads.  So they are always
1002 	 * marked at thread level.
1003 	 */
1004 	siglist = (type == SPROCESS) ? &pr->ps_siglist : &p->p_siglist;
1005 	if (prop & SA_CONT) {
1006 		siglist = &p->p_siglist;
1007 		atomic_clearbits_int(siglist, STOPSIGMASK);
1008 	}
1009 	if (prop & SA_STOP) {
1010 		siglist = &p->p_siglist;
1011 		atomic_clearbits_int(siglist, CONTSIGMASK);
1012 		atomic_clearbits_int(&p->p_flag, P_CONTINUED);
1013 	}
1014 	atomic_setbits_int(siglist, mask);
1015 
1016 	/*
1017 	 * XXX delay processing of SA_STOP signals unless action == SIG_DFL?
1018 	 */
1019 	if (prop & (SA_CONT | SA_STOP) && type != SPROPAGATED)
1020 		TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link)
1021 			if (q != p)
1022 				ptsignal(q, signum, SPROPAGATED);
1023 
1024 	/*
1025 	 * Defer further processing for signals which are held,
1026 	 * except that stopped processes must be continued by SIGCONT.
1027 	 */
1028 	if (action == SIG_HOLD && ((prop & SA_CONT) == 0 || p->p_stat != SSTOP))
1029 		return;
1030 
1031 	SCHED_LOCK(s);
1032 
1033 	switch (p->p_stat) {
1034 
1035 	case SSLEEP:
1036 		/*
1037 		 * If process is sleeping uninterruptibly
1038 		 * we can't interrupt the sleep... the signal will
1039 		 * be noticed when the process returns through
1040 		 * trap() or syscall().
1041 		 */
1042 		if ((p->p_flag & P_SINTR) == 0)
1043 			goto out;
1044 		/*
1045 		 * Process is sleeping and traced... make it runnable
1046 		 * so it can discover the signal in cursig() and stop
1047 		 * for the parent.
1048 		 */
1049 		if (pr->ps_flags & PS_TRACED)
1050 			goto run;
1051 		/*
1052 		 * If SIGCONT is default (or ignored) and process is
1053 		 * asleep, we are finished; the process should not
1054 		 * be awakened.
1055 		 */
1056 		if ((prop & SA_CONT) && action == SIG_DFL) {
1057 			atomic_clearbits_int(siglist, mask);
1058 			goto out;
1059 		}
1060 		/*
1061 		 * When a sleeping process receives a stop
1062 		 * signal, process immediately if possible.
1063 		 */
1064 		if ((prop & SA_STOP) && action == SIG_DFL) {
1065 			/*
1066 			 * If a child holding parent blocked,
1067 			 * stopping could cause deadlock.
1068 			 */
1069 			if (pr->ps_flags & PS_PPWAIT)
1070 				goto out;
1071 			atomic_clearbits_int(siglist, mask);
1072 			pr->ps_xsig = signum;
1073 			proc_stop(p, 0);
1074 			goto out;
1075 		}
1076 		/*
1077 		 * All other (caught or default) signals
1078 		 * cause the process to run.
1079 		 */
1080 		goto runfast;
1081 		/* NOTREACHED */
1082 
1083 	case SSTOP:
1084 		/*
1085 		 * If traced process is already stopped,
1086 		 * then no further action is necessary.
1087 		 */
1088 		if (pr->ps_flags & PS_TRACED)
1089 			goto out;
1090 
1091 		/*
1092 		 * Kill signal always sets processes running.
1093 		 */
1094 		if (signum == SIGKILL) {
1095 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1096 			goto runfast;
1097 		}
1098 
1099 		if (prop & SA_CONT) {
1100 			/*
1101 			 * If SIGCONT is default (or ignored), we continue the
1102 			 * process but don't leave the signal in p_siglist, as
1103 			 * it has no further action.  If SIGCONT is held, we
1104 			 * continue the process and leave the signal in
1105 			 * p_siglist.  If the process catches SIGCONT, let it
1106 			 * handle the signal itself.  If it isn't waiting on
1107 			 * an event, then it goes back to run state.
1108 			 * Otherwise, process goes back to sleep state.
1109 			 */
1110 			atomic_setbits_int(&p->p_flag, P_CONTINUED);
1111 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1112 			wakeparent = 1;
1113 			if (action == SIG_DFL)
1114 				atomic_clearbits_int(siglist, mask);
1115 			if (action == SIG_CATCH)
1116 				goto runfast;
1117 			if (p->p_wchan == NULL)
1118 				goto run;
1119 			p->p_stat = SSLEEP;
1120 			goto out;
1121 		}
1122 
1123 		if (prop & SA_STOP) {
1124 			/*
1125 			 * Already stopped, don't need to stop again.
1126 			 * (If we did the shell could get confused.)
1127 			 */
1128 			atomic_clearbits_int(siglist, mask);
1129 			goto out;
1130 		}
1131 
1132 		/*
1133 		 * If process is sleeping interruptibly, then simulate a
1134 		 * wakeup so that when it is continued, it will be made
1135 		 * runnable and can look at the signal.  But don't make
1136 		 * the process runnable, leave it stopped.
1137 		 */
1138 		if (p->p_flag & P_SINTR)
1139 			unsleep(p);
1140 		goto out;
1141 
1142 	case SONPROC:
1143 		signotify(p);
1144 		/* FALLTHROUGH */
1145 	default:
1146 		/*
1147 		 * SRUN, SIDL, SDEAD do nothing with the signal,
1148 		 * other than kicking ourselves if we are running.
1149 		 * It will either never be noticed, or noticed very soon.
1150 		 */
1151 		goto out;
1152 	}
1153 	/* NOTREACHED */
1154 
1155 runfast:
1156 	/*
1157 	 * Raise priority to at least PUSER.
1158 	 */
1159 	if (p->p_usrpri > PUSER)
1160 		p->p_usrpri = PUSER;
1161 run:
1162 	setrunnable(p);
1163 out:
1164 	SCHED_UNLOCK(s);
1165 	if (wakeparent)
1166 		wakeup(pr->ps_pptr);
1167 }
1168 
1169 /*
1170  * Determine signal that should be delivered to process p, the current
1171  * process, 0 if none.
1172  *
1173  * If the current process has received a signal (should be caught or cause
1174  * termination, should interrupt current syscall), return the signal number.
1175  * Stop signals with default action are processed immediately, then cleared;
1176  * they aren't returned.  This is checked after each entry to the system for
1177  * a syscall or trap. The normal call sequence is
1178  *
1179  *	while (signum = cursig(curproc))
1180  *		postsig(signum);
1181  *
1182  * Assumes that if the P_SINTR flag is set, we're holding both the
1183  * kernel and scheduler locks.
1184  */
1185 int
1186 cursig(struct proc *p)
1187 {
1188 	struct process *pr = p->p_p;
1189 	int sigpending, signum, mask, prop;
1190 	int dolock = (p->p_flag & P_SINTR) == 0;
1191 	int s;
1192 
1193 	KERNEL_ASSERT_LOCKED();
1194 
1195 	sigpending = (p->p_siglist | pr->ps_siglist);
1196 	if (sigpending == 0)
1197 		return 0;
1198 
1199 	if (!ISSET(pr->ps_flags, PS_TRACED) && SIGPENDING(p) == 0)
1200 		return 0;
1201 
1202 	for (;;) {
1203 		mask = SIGPENDING(p);
1204 		if (pr->ps_flags & PS_PPWAIT)
1205 			mask &= ~STOPSIGMASK;
1206 		if (mask == 0)	 	/* no signal to send */
1207 			return (0);
1208 		signum = ffs((long)mask);
1209 		mask = sigmask(signum);
1210 		atomic_clearbits_int(&p->p_siglist, mask);
1211 		atomic_clearbits_int(&pr->ps_siglist, mask);
1212 
1213 		/*
1214 		 * We should see pending but ignored signals
1215 		 * only if PS_TRACED was on when they were posted.
1216 		 */
1217 		if (mask & pr->ps_sigacts->ps_sigignore &&
1218 		    (pr->ps_flags & PS_TRACED) == 0)
1219 			continue;
1220 
1221 		/*
1222 		 * If traced, always stop, and stay stopped until released
1223 		 * by the debugger.  If our parent process is waiting for
1224 		 * us, don't hang as we could deadlock.
1225 		 */
1226 		if (((pr->ps_flags & (PS_TRACED | PS_PPWAIT)) == PS_TRACED) &&
1227 		    signum != SIGKILL) {
1228 			pr->ps_xsig = signum;
1229 
1230 			single_thread_set(p, SINGLE_SUSPEND, 0);
1231 
1232 			if (dolock)
1233 				SCHED_LOCK(s);
1234 			proc_stop(p, 1);
1235 			if (dolock)
1236 				SCHED_UNLOCK(s);
1237 
1238 			single_thread_clear(p, 0);
1239 
1240 			/*
1241 			 * If we are no longer being traced, or the parent
1242 			 * didn't give us a signal, look for more signals.
1243 			 */
1244 			if ((pr->ps_flags & PS_TRACED) == 0 ||
1245 			    pr->ps_xsig == 0)
1246 				continue;
1247 
1248 			/*
1249 			 * If the new signal is being masked, look for other
1250 			 * signals.
1251 			 */
1252 			signum = pr->ps_xsig;
1253 			mask = sigmask(signum);
1254 			if ((p->p_sigmask & mask) != 0)
1255 				continue;
1256 
1257 			/* take the signal! */
1258 			atomic_clearbits_int(&p->p_siglist, mask);
1259 			atomic_clearbits_int(&pr->ps_siglist, mask);
1260 		}
1261 
1262 		prop = sigprop[signum];
1263 
1264 		/*
1265 		 * Decide whether the signal should be returned.
1266 		 * Return the signal's number, or fall through
1267 		 * to clear it from the pending mask.
1268 		 */
1269 		switch ((long)pr->ps_sigacts->ps_sigact[signum]) {
1270 		case (long)SIG_DFL:
1271 			/*
1272 			 * Don't take default actions on system processes.
1273 			 */
1274 			if (pr->ps_pid <= 1) {
1275 #ifdef DIAGNOSTIC
1276 				/*
1277 				 * Are you sure you want to ignore SIGSEGV
1278 				 * in init? XXX
1279 				 */
1280 				printf("Process (pid %d) got signal"
1281 				    " %d\n", pr->ps_pid, signum);
1282 #endif
1283 				break;		/* == ignore */
1284 			}
1285 			/*
1286 			 * If there is a pending stop signal to process
1287 			 * with default action, stop here,
1288 			 * then clear the signal.  However,
1289 			 * if process is member of an orphaned
1290 			 * process group, ignore tty stop signals.
1291 			 */
1292 			if (prop & SA_STOP) {
1293 				if (pr->ps_flags & PS_TRACED ||
1294 		    		    (pr->ps_pgrp->pg_jobc == 0 &&
1295 				    prop & SA_TTYSTOP))
1296 					break;	/* == ignore */
1297 				pr->ps_xsig = signum;
1298 				if (dolock)
1299 					SCHED_LOCK(s);
1300 				proc_stop(p, 1);
1301 				if (dolock)
1302 					SCHED_UNLOCK(s);
1303 				break;
1304 			} else if (prop & SA_IGNORE) {
1305 				/*
1306 				 * Except for SIGCONT, shouldn't get here.
1307 				 * Default action is to ignore; drop it.
1308 				 */
1309 				break;		/* == ignore */
1310 			} else
1311 				goto keep;
1312 			/* NOTREACHED */
1313 		case (long)SIG_IGN:
1314 			/*
1315 			 * Masking above should prevent us ever trying
1316 			 * to take action on an ignored signal other
1317 			 * than SIGCONT, unless process is traced.
1318 			 */
1319 			if ((prop & SA_CONT) == 0 &&
1320 			    (pr->ps_flags & PS_TRACED) == 0)
1321 				printf("%s\n", __func__);
1322 			break;		/* == ignore */
1323 		default:
1324 			/*
1325 			 * This signal has an action, let
1326 			 * postsig() process it.
1327 			 */
1328 			goto keep;
1329 		}
1330 	}
1331 	/* NOTREACHED */
1332 
1333 keep:
1334 	atomic_setbits_int(&p->p_siglist, mask); /*leave the signal for later */
1335 	return (signum);
1336 }
1337 
1338 /*
1339  * Put the argument process into the stopped state and notify the parent
1340  * via wakeup.  Signals are handled elsewhere.  The process must not be
1341  * on the run queue.
1342  */
1343 void
1344 proc_stop(struct proc *p, int sw)
1345 {
1346 	struct process *pr = p->p_p;
1347 
1348 #ifdef MULTIPROCESSOR
1349 	SCHED_ASSERT_LOCKED();
1350 #endif
1351 
1352 	p->p_stat = SSTOP;
1353 	atomic_clearbits_int(&pr->ps_flags, PS_WAITED);
1354 	atomic_setbits_int(&pr->ps_flags, PS_STOPPED);
1355 	atomic_setbits_int(&p->p_flag, P_SUSPSIG);
1356 	/*
1357 	 * We need this soft interrupt to be handled fast.
1358 	 * Extra calls to softclock don't hurt.
1359 	 */
1360 	softintr_schedule(proc_stop_si);
1361 	if (sw)
1362 		mi_switch();
1363 }
1364 
1365 /*
1366  * Called from a soft interrupt to send signals to the parents of stopped
1367  * processes.
1368  * We can't do this in proc_stop because it's called with nasty locks held
1369  * and we would need recursive scheduler lock to deal with that.
1370  */
1371 void
1372 proc_stop_sweep(void *v)
1373 {
1374 	struct process *pr;
1375 
1376 	LIST_FOREACH(pr, &allprocess, ps_list) {
1377 		if ((pr->ps_flags & PS_STOPPED) == 0)
1378 			continue;
1379 		atomic_clearbits_int(&pr->ps_flags, PS_STOPPED);
1380 
1381 		if ((pr->ps_pptr->ps_sigacts->ps_sigflags & SAS_NOCLDSTOP) == 0)
1382 			prsignal(pr->ps_pptr, SIGCHLD);
1383 		wakeup(pr->ps_pptr);
1384 	}
1385 }
1386 
1387 /*
1388  * Take the action for the specified signal
1389  * from the current set of pending signals.
1390  */
1391 void
1392 postsig(struct proc *p, int signum)
1393 {
1394 	struct process *pr = p->p_p;
1395 	struct sigacts *ps = pr->ps_sigacts;
1396 	sig_t action;
1397 	u_long trapno;
1398 	int mask, returnmask;
1399 	siginfo_t si;
1400 	union sigval sigval;
1401 	int s, code, info, onstack;
1402 
1403 	KASSERT(signum != 0);
1404 	KERNEL_ASSERT_LOCKED();
1405 
1406 	mask = sigmask(signum);
1407 	atomic_clearbits_int(&p->p_siglist, mask);
1408 	action = ps->ps_sigact[signum];
1409 	info = (ps->ps_siginfo & mask) != 0;
1410 	onstack = (ps->ps_sigonstack & mask) != 0;
1411 	sigval.sival_ptr = NULL;
1412 
1413 	if (p->p_sisig != signum) {
1414 		trapno = 0;
1415 		code = SI_USER;
1416 		sigval.sival_ptr = NULL;
1417 	} else {
1418 		trapno = p->p_sitrapno;
1419 		code = p->p_sicode;
1420 		sigval = p->p_sigval;
1421 	}
1422 	initsiginfo(&si, signum, trapno, code, sigval);
1423 
1424 #ifdef KTRACE
1425 	if (KTRPOINT(p, KTR_PSIG)) {
1426 		ktrpsig(p, signum, action, p->p_flag & P_SIGSUSPEND ?
1427 		    p->p_oldmask : p->p_sigmask, code, &si);
1428 	}
1429 #endif
1430 	if (action == SIG_DFL) {
1431 		/*
1432 		 * Default action, where the default is to kill
1433 		 * the process.  (Other cases were ignored above.)
1434 		 */
1435 		sigexit(p, signum);
1436 		/* NOTREACHED */
1437 	} else {
1438 		/*
1439 		 * If we get here, the signal must be caught.
1440 		 */
1441 #ifdef DIAGNOSTIC
1442 		if (action == SIG_IGN || (p->p_sigmask & mask))
1443 			panic("postsig action");
1444 #endif
1445 		/*
1446 		 * Set the new mask value and also defer further
1447 		 * occurrences of this signal.
1448 		 *
1449 		 * Special case: user has done a sigpause.  Here the
1450 		 * current mask is not of interest, but rather the
1451 		 * mask from before the sigpause is what we want
1452 		 * restored after the signal processing is completed.
1453 		 */
1454 #ifdef MULTIPROCESSOR
1455 		s = splsched();
1456 #else
1457 		s = splhigh();
1458 #endif
1459 		if (p->p_flag & P_SIGSUSPEND) {
1460 			atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
1461 			returnmask = p->p_oldmask;
1462 		} else {
1463 			returnmask = p->p_sigmask;
1464 		}
1465 		if (p->p_sisig == signum) {
1466 			p->p_sisig = 0;
1467 			p->p_sitrapno = 0;
1468 			p->p_sicode = SI_USER;
1469 			p->p_sigval.sival_ptr = NULL;
1470 		}
1471 
1472 		if (sendsig(action, signum, returnmask, &si, info, onstack)) {
1473 			sigexit(p, SIGILL);
1474 			/* NOTREACHED */
1475 		}
1476 		postsig_done(p, signum, ps);
1477 		splx(s);
1478 	}
1479 }
1480 
1481 /*
1482  * Force the current process to exit with the specified signal, dumping core
1483  * if appropriate.  We bypass the normal tests for masked and caught signals,
1484  * allowing unrecoverable failures to terminate the process without changing
1485  * signal state.  Mark the accounting record with the signal termination.
1486  * If dumping core, save the signal number for the debugger.  Calls exit and
1487  * does not return.
1488  */
1489 void
1490 sigexit(struct proc *p, int signum)
1491 {
1492 	/* Mark process as going away */
1493 	atomic_setbits_int(&p->p_flag, P_WEXIT);
1494 
1495 	p->p_p->ps_acflag |= AXSIG;
1496 	if (sigprop[signum] & SA_CORE) {
1497 		p->p_sisig = signum;
1498 
1499 		/* if there are other threads, pause them */
1500 		if (P_HASSIBLING(p))
1501 			single_thread_set(p, SINGLE_SUSPEND, 1);
1502 
1503 		if (coredump(p) == 0)
1504 			signum |= WCOREFLAG;
1505 	}
1506 	exit1(p, 0, signum, EXIT_NORMAL);
1507 	/* NOTREACHED */
1508 }
1509 
1510 /*
1511  * Send uncatchable SIGABRT for coredump.
1512  */
1513 void
1514 sigabort(struct proc *p)
1515 {
1516 	struct sigaction sa;
1517 
1518 	memset(&sa, 0, sizeof sa);
1519 	sa.sa_handler = SIG_DFL;
1520 	setsigvec(p, SIGABRT, &sa);
1521 	atomic_clearbits_int(&p->p_sigmask, sigmask(SIGABRT));
1522 	psignal(p, SIGABRT);
1523 }
1524 
1525 /*
1526  * Return 1 if `sig', a given signal, is ignored or masked for `p', a given
1527  * thread, and 0 otherwise.
1528  */
1529 int
1530 sigismasked(struct proc *p, int sig)
1531 {
1532 	struct process *pr = p->p_p;
1533 
1534 	if ((pr->ps_sigacts->ps_sigignore & sigmask(sig)) ||
1535 	    (p->p_sigmask & sigmask(sig)))
1536 	    	return 1;
1537 
1538 	return 0;
1539 }
1540 
1541 int nosuidcoredump = 1;
1542 
1543 struct coredump_iostate {
1544 	struct proc *io_proc;
1545 	struct vnode *io_vp;
1546 	struct ucred *io_cred;
1547 	off_t io_offset;
1548 };
1549 
1550 /*
1551  * Dump core, into a file named "progname.core", unless the process was
1552  * setuid/setgid.
1553  */
1554 int
1555 coredump(struct proc *p)
1556 {
1557 #ifdef SMALL_KERNEL
1558 	return EPERM;
1559 #else
1560 	struct process *pr = p->p_p;
1561 	struct vnode *vp;
1562 	struct ucred *cred = p->p_ucred;
1563 	struct vmspace *vm = p->p_vmspace;
1564 	struct nameidata nd;
1565 	struct vattr vattr;
1566 	struct coredump_iostate	io;
1567 	int error, len, incrash = 0;
1568 	char *name;
1569 	const char *dir = "/var/crash";
1570 
1571 	if (pr->ps_emul->e_coredump == NULL)
1572 		return (EINVAL);
1573 
1574 	atomic_setbits_int(&pr->ps_flags, PS_COREDUMP);
1575 
1576 	/* Don't dump if will exceed file size limit. */
1577 	if (USPACE + ptoa(vm->vm_dsize + vm->vm_ssize) >= lim_cur(RLIMIT_CORE))
1578 		return (EFBIG);
1579 
1580 	name = pool_get(&namei_pool, PR_WAITOK);
1581 
1582 	/*
1583 	 * If the process has inconsistent uids, nosuidcoredump
1584 	 * determines coredump placement policy.
1585 	 */
1586 	if (((pr->ps_flags & PS_SUGID) && (error = suser(p))) ||
1587 	   ((pr->ps_flags & PS_SUGID) && nosuidcoredump)) {
1588 		if (nosuidcoredump == 3) {
1589 			/*
1590 			 * If the program directory does not exist, dumps of
1591 			 * that core will silently fail.
1592 			 */
1593 			len = snprintf(name, MAXPATHLEN, "%s/%s/%u.core",
1594 			    dir, pr->ps_comm, pr->ps_pid);
1595 			incrash = KERNELPATH;
1596 		} else if (nosuidcoredump == 2) {
1597 			len = snprintf(name, MAXPATHLEN, "%s/%s.core",
1598 			    dir, pr->ps_comm);
1599 			incrash = KERNELPATH;
1600 		} else {
1601 			pool_put(&namei_pool, name);
1602 			return (EPERM);
1603 		}
1604 	} else
1605 		len = snprintf(name, MAXPATHLEN, "%s.core", pr->ps_comm);
1606 
1607 	if (len >= MAXPATHLEN) {
1608 		pool_put(&namei_pool, name);
1609 		return (EACCES);
1610 	}
1611 
1612 	/*
1613 	 * Control the UID used to write out.  The normal case uses
1614 	 * the real UID.  If the sugid case is going to write into the
1615 	 * controlled directory, we do so as root.
1616 	 */
1617 	if (incrash == 0) {
1618 		cred = crdup(cred);
1619 		cred->cr_uid = cred->cr_ruid;
1620 		cred->cr_gid = cred->cr_rgid;
1621 	} else {
1622 		if (p->p_fd->fd_rdir) {
1623 			vrele(p->p_fd->fd_rdir);
1624 			p->p_fd->fd_rdir = NULL;
1625 		}
1626 		p->p_ucred = crdup(p->p_ucred);
1627 		crfree(cred);
1628 		cred = p->p_ucred;
1629 		crhold(cred);
1630 		cred->cr_uid = 0;
1631 		cred->cr_gid = 0;
1632 	}
1633 
1634 	/* incrash should be 0 or KERNELPATH only */
1635 	NDINIT(&nd, 0, incrash, UIO_SYSSPACE, name, p);
1636 
1637 	error = vn_open(&nd, O_CREAT | FWRITE | O_NOFOLLOW | O_NONBLOCK,
1638 	    S_IRUSR | S_IWUSR);
1639 
1640 	if (error)
1641 		goto out;
1642 
1643 	/*
1644 	 * Don't dump to non-regular files, files with links, or files
1645 	 * owned by someone else.
1646 	 */
1647 	vp = nd.ni_vp;
1648 	if ((error = VOP_GETATTR(vp, &vattr, cred, p)) != 0) {
1649 		VOP_UNLOCK(vp);
1650 		vn_close(vp, FWRITE, cred, p);
1651 		goto out;
1652 	}
1653 	if (vp->v_type != VREG || vattr.va_nlink != 1 ||
1654 	    vattr.va_mode & ((VREAD | VWRITE) >> 3 | (VREAD | VWRITE) >> 6) ||
1655 	    vattr.va_uid != cred->cr_uid) {
1656 		error = EACCES;
1657 		VOP_UNLOCK(vp);
1658 		vn_close(vp, FWRITE, cred, p);
1659 		goto out;
1660 	}
1661 	VATTR_NULL(&vattr);
1662 	vattr.va_size = 0;
1663 	VOP_SETATTR(vp, &vattr, cred, p);
1664 	pr->ps_acflag |= ACORE;
1665 
1666 	io.io_proc = p;
1667 	io.io_vp = vp;
1668 	io.io_cred = cred;
1669 	io.io_offset = 0;
1670 	VOP_UNLOCK(vp);
1671 	vref(vp);
1672 	error = vn_close(vp, FWRITE, cred, p);
1673 	if (error == 0)
1674 		error = (*pr->ps_emul->e_coredump)(p, &io);
1675 	vrele(vp);
1676 out:
1677 	crfree(cred);
1678 	pool_put(&namei_pool, name);
1679 	return (error);
1680 #endif
1681 }
1682 
1683 #ifndef SMALL_KERNEL
1684 int
1685 coredump_write(void *cookie, enum uio_seg segflg, const void *data, size_t len)
1686 {
1687 	struct coredump_iostate *io = cookie;
1688 	off_t coffset = 0;
1689 	size_t csize;
1690 	int chunk, error;
1691 
1692 	csize = len;
1693 	do {
1694 		if (sigmask(SIGKILL) &
1695 		    (io->io_proc->p_siglist | io->io_proc->p_p->ps_siglist))
1696 			return (EINTR);
1697 
1698 		/* Rest of the loop sleeps with lock held, so... */
1699 		yield();
1700 
1701 		chunk = MIN(csize, MAXPHYS);
1702 		error = vn_rdwr(UIO_WRITE, io->io_vp,
1703 		    (caddr_t)data + coffset, chunk,
1704 		    io->io_offset + coffset, segflg,
1705 		    IO_UNIT, io->io_cred, NULL, io->io_proc);
1706 		if (error) {
1707 			struct process *pr = io->io_proc->p_p;
1708 
1709 			if (error == ENOSPC)
1710 				log(LOG_ERR,
1711 				    "coredump of %s(%d) failed, filesystem full\n",
1712 				    pr->ps_comm, pr->ps_pid);
1713 			else
1714 				log(LOG_ERR,
1715 				    "coredump of %s(%d), write failed: errno %d\n",
1716 				    pr->ps_comm, pr->ps_pid, error);
1717 			return (error);
1718 		}
1719 
1720 		coffset += chunk;
1721 		csize -= chunk;
1722 	} while (csize > 0);
1723 
1724 	io->io_offset += len;
1725 	return (0);
1726 }
1727 
1728 void
1729 coredump_unmap(void *cookie, vaddr_t start, vaddr_t end)
1730 {
1731 	struct coredump_iostate *io = cookie;
1732 
1733 	uvm_unmap(&io->io_proc->p_vmspace->vm_map, start, end);
1734 }
1735 
1736 #endif	/* !SMALL_KERNEL */
1737 
1738 /*
1739  * Nonexistent system call-- signal process (may want to handle it).
1740  * Flag error in case process won't see signal immediately (blocked or ignored).
1741  */
1742 int
1743 sys_nosys(struct proc *p, void *v, register_t *retval)
1744 {
1745 
1746 	ptsignal(p, SIGSYS, STHREAD);
1747 	return (ENOSYS);
1748 }
1749 
1750 int
1751 sys___thrsigdivert(struct proc *p, void *v, register_t *retval)
1752 {
1753 	static int sigwaitsleep;
1754 	struct sys___thrsigdivert_args /* {
1755 		syscallarg(sigset_t) sigmask;
1756 		syscallarg(siginfo_t *) info;
1757 		syscallarg(const struct timespec *) timeout;
1758 	} */ *uap = v;
1759 	sigset_t mask = SCARG(uap, sigmask) &~ sigcantmask;
1760 	siginfo_t si;
1761 	uint64_t nsecs = INFSLP;
1762 	int timeinvalid = 0;
1763 	int error = 0;
1764 
1765 	memset(&si, 0, sizeof(si));
1766 
1767 	if (SCARG(uap, timeout) != NULL) {
1768 		struct timespec ts;
1769 		if ((error = copyin(SCARG(uap, timeout), &ts, sizeof(ts))) != 0)
1770 			return (error);
1771 #ifdef KTRACE
1772 		if (KTRPOINT(p, KTR_STRUCT))
1773 			ktrreltimespec(p, &ts);
1774 #endif
1775 		if (!timespecisvalid(&ts))
1776 			timeinvalid = 1;
1777 		else
1778 			nsecs = TIMESPEC_TO_NSEC(&ts);
1779 	}
1780 
1781 	dosigsuspend(p, p->p_sigmask &~ mask);
1782 	for (;;) {
1783 		si.si_signo = cursig(p);
1784 		if (si.si_signo != 0) {
1785 			sigset_t smask = sigmask(si.si_signo);
1786 			if (smask & mask) {
1787 				atomic_clearbits_int(&p->p_siglist, smask);
1788 				error = 0;
1789 				break;
1790 			}
1791 		}
1792 
1793 		/* per-POSIX, delay this error until after the above */
1794 		if (timeinvalid)
1795 			error = EINVAL;
1796 		/* per-POSIX, return immediatly if timeout is zero-valued */
1797 		if (nsecs == 0)
1798 			error = EAGAIN;
1799 
1800 		if (error != 0)
1801 			break;
1802 
1803 		error = tsleep_nsec(&sigwaitsleep, PPAUSE|PCATCH, "sigwait",
1804 		    nsecs);
1805 	}
1806 
1807 	if (error == 0) {
1808 		*retval = si.si_signo;
1809 		if (SCARG(uap, info) != NULL)
1810 			error = copyout(&si, SCARG(uap, info), sizeof(si));
1811 	} else if (error == ERESTART && SCARG(uap, timeout) != NULL) {
1812 		/*
1813 		 * Restarting is wrong if there's a timeout, as it'll be
1814 		 * for the same interval again
1815 		 */
1816 		error = EINTR;
1817 	}
1818 
1819 	return (error);
1820 }
1821 
1822 void
1823 initsiginfo(siginfo_t *si, int sig, u_long trapno, int code, union sigval val)
1824 {
1825 	memset(si, 0, sizeof(*si));
1826 
1827 	si->si_signo = sig;
1828 	si->si_code = code;
1829 	if (code == SI_USER) {
1830 		si->si_value = val;
1831 	} else {
1832 		switch (sig) {
1833 		case SIGSEGV:
1834 		case SIGILL:
1835 		case SIGBUS:
1836 		case SIGFPE:
1837 			si->si_addr = val.sival_ptr;
1838 			si->si_trapno = trapno;
1839 			break;
1840 		case SIGXFSZ:
1841 			break;
1842 		}
1843 	}
1844 }
1845 
1846 int
1847 filt_sigattach(struct knote *kn)
1848 {
1849 	struct process *pr = curproc->p_p;
1850 	int s;
1851 
1852 	if (kn->kn_id >= NSIG)
1853 		return EINVAL;
1854 
1855 	kn->kn_ptr.p_process = pr;
1856 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
1857 
1858 	s = splhigh();
1859 	klist_insert_locked(&pr->ps_klist, kn);
1860 	splx(s);
1861 
1862 	return (0);
1863 }
1864 
1865 void
1866 filt_sigdetach(struct knote *kn)
1867 {
1868 	struct process *pr = kn->kn_ptr.p_process;
1869 	int s;
1870 
1871 	s = splhigh();
1872 	klist_remove_locked(&pr->ps_klist, kn);
1873 	splx(s);
1874 }
1875 
1876 /*
1877  * signal knotes are shared with proc knotes, so we apply a mask to
1878  * the hint in order to differentiate them from process hints.  This
1879  * could be avoided by using a signal-specific knote list, but probably
1880  * isn't worth the trouble.
1881  */
1882 int
1883 filt_signal(struct knote *kn, long hint)
1884 {
1885 
1886 	if (hint & NOTE_SIGNAL) {
1887 		hint &= ~NOTE_SIGNAL;
1888 
1889 		if (kn->kn_id == hint)
1890 			kn->kn_data++;
1891 	}
1892 	return (kn->kn_data != 0);
1893 }
1894 
1895 void
1896 userret(struct proc *p)
1897 {
1898 	int signum;
1899 
1900 	/* send SIGPROF or SIGVTALRM if their timers interrupted this thread */
1901 	if (p->p_flag & P_PROFPEND) {
1902 		atomic_clearbits_int(&p->p_flag, P_PROFPEND);
1903 		KERNEL_LOCK();
1904 		psignal(p, SIGPROF);
1905 		KERNEL_UNLOCK();
1906 	}
1907 	if (p->p_flag & P_ALRMPEND) {
1908 		atomic_clearbits_int(&p->p_flag, P_ALRMPEND);
1909 		KERNEL_LOCK();
1910 		psignal(p, SIGVTALRM);
1911 		KERNEL_UNLOCK();
1912 	}
1913 
1914 	if (SIGPENDING(p) != 0) {
1915 		KERNEL_LOCK();
1916 		while ((signum = cursig(p)) != 0)
1917 			postsig(p, signum);
1918 		KERNEL_UNLOCK();
1919 	}
1920 
1921 	/*
1922 	 * If P_SIGSUSPEND is still set here, then we still need to restore
1923 	 * the original sigmask before returning to userspace.  Also, this
1924 	 * might unmask some pending signals, so we need to check a second
1925 	 * time for signals to post.
1926 	 */
1927 	if (p->p_flag & P_SIGSUSPEND) {
1928 		atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
1929 		p->p_sigmask = p->p_oldmask;
1930 
1931 		KERNEL_LOCK();
1932 		while ((signum = cursig(p)) != 0)
1933 			postsig(p, signum);
1934 		KERNEL_UNLOCK();
1935 	}
1936 
1937 	if (p->p_flag & P_SUSPSINGLE)
1938 		single_thread_check(p, 0);
1939 
1940 	WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
1941 
1942 	p->p_cpu->ci_schedstate.spc_curpriority = p->p_usrpri;
1943 }
1944 
1945 int
1946 single_thread_check_locked(struct proc *p, int deep, int s)
1947 {
1948 	struct process *pr = p->p_p;
1949 
1950 	SCHED_ASSERT_LOCKED();
1951 
1952 	if (pr->ps_single != NULL && pr->ps_single != p) {
1953 		do {
1954 			/* if we're in deep, we need to unwind to the edge */
1955 			if (deep) {
1956 				if (pr->ps_flags & PS_SINGLEUNWIND)
1957 					return (ERESTART);
1958 				if (pr->ps_flags & PS_SINGLEEXIT)
1959 					return (EINTR);
1960 			}
1961 
1962 			if (pr->ps_single == NULL)
1963 				continue;
1964 
1965 			if (atomic_dec_int_nv(&pr->ps_singlecount) == 0)
1966 				wakeup(&pr->ps_singlecount);
1967 
1968 			if (pr->ps_flags & PS_SINGLEEXIT) {
1969 				SCHED_UNLOCK(s);
1970 				KERNEL_LOCK();
1971 				exit1(p, 0, 0, EXIT_THREAD_NOCHECK);
1972 				/* NOTREACHED */
1973 			}
1974 
1975 			/* not exiting and don't need to unwind, so suspend */
1976 			p->p_stat = SSTOP;
1977 			mi_switch();
1978 		} while (pr->ps_single != NULL);
1979 	}
1980 
1981 	return (0);
1982 }
1983 
1984 int
1985 single_thread_check(struct proc *p, int deep)
1986 {
1987 	int s, error;
1988 
1989 	SCHED_LOCK(s);
1990 	error = single_thread_check_locked(p, deep, s);
1991 	SCHED_UNLOCK(s);
1992 
1993 	return error;
1994 }
1995 
1996 /*
1997  * Stop other threads in the process.  The mode controls how and
1998  * where the other threads should stop:
1999  *  - SINGLE_SUSPEND: stop wherever they are, will later either be told to exit
2000  *    (by setting to SINGLE_EXIT) or be released (via single_thread_clear())
2001  *  - SINGLE_UNWIND: just unwind to kernel boundary, will be told to exit
2002  *    or released as with SINGLE_SUSPEND
2003  *  - SINGLE_EXIT: unwind to kernel boundary and exit
2004  */
2005 int
2006 single_thread_set(struct proc *p, enum single_thread_mode mode, int wait)
2007 {
2008 	struct process *pr = p->p_p;
2009 	struct proc *q;
2010 	int error, s;
2011 
2012 	KASSERT(curproc == p);
2013 
2014 	SCHED_LOCK(s);
2015 	error = single_thread_check_locked(p, (mode == SINGLE_UNWIND), s);
2016 	if (error) {
2017 		SCHED_UNLOCK(s);
2018 		return error;
2019 	}
2020 
2021 	switch (mode) {
2022 	case SINGLE_SUSPEND:
2023 		break;
2024 	case SINGLE_UNWIND:
2025 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
2026 		break;
2027 	case SINGLE_EXIT:
2028 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEEXIT);
2029 		atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
2030 		break;
2031 #ifdef DIAGNOSTIC
2032 	default:
2033 		panic("single_thread_mode = %d", mode);
2034 #endif
2035 	}
2036 	pr->ps_singlecount = 0;
2037 	membar_producer();
2038 	pr->ps_single = p;
2039 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
2040 		if (q == p)
2041 			continue;
2042 		if (q->p_flag & P_WEXIT) {
2043 			if (mode == SINGLE_EXIT) {
2044 				if (q->p_stat == SSTOP) {
2045 					setrunnable(q);
2046 					atomic_inc_int(&pr->ps_singlecount);
2047 				}
2048 			}
2049 			continue;
2050 		}
2051 		atomic_setbits_int(&q->p_flag, P_SUSPSINGLE);
2052 		switch (q->p_stat) {
2053 		case SIDL:
2054 		case SRUN:
2055 			atomic_inc_int(&pr->ps_singlecount);
2056 			break;
2057 		case SSLEEP:
2058 			/* if it's not interruptible, then just have to wait */
2059 			if (q->p_flag & P_SINTR) {
2060 				/* merely need to suspend?  just stop it */
2061 				if (mode == SINGLE_SUSPEND) {
2062 					q->p_stat = SSTOP;
2063 					break;
2064 				}
2065 				/* need to unwind or exit, so wake it */
2066 				setrunnable(q);
2067 			}
2068 			atomic_inc_int(&pr->ps_singlecount);
2069 			break;
2070 		case SSTOP:
2071 			if (mode == SINGLE_EXIT) {
2072 				setrunnable(q);
2073 				atomic_inc_int(&pr->ps_singlecount);
2074 			}
2075 			break;
2076 		case SDEAD:
2077 			break;
2078 		case SONPROC:
2079 			atomic_inc_int(&pr->ps_singlecount);
2080 			signotify(q);
2081 			break;
2082 		}
2083 	}
2084 	SCHED_UNLOCK(s);
2085 
2086 	if (wait)
2087 		single_thread_wait(pr, 1);
2088 
2089 	return 0;
2090 }
2091 
2092 /*
2093  * Wait for other threads to stop. If recheck is false then the function
2094  * returns non-zero if the caller needs to restart the check else 0 is
2095  * returned. If recheck is true the return value is always 0.
2096  */
2097 int
2098 single_thread_wait(struct process *pr, int recheck)
2099 {
2100 	struct sleep_state sls;
2101 	int wait;
2102 
2103 	/* wait until they're all suspended */
2104 	wait = pr->ps_singlecount > 0;
2105 	while (wait) {
2106 		sleep_setup(&sls, &pr->ps_singlecount, PWAIT, "suspend", 0);
2107 		wait = pr->ps_singlecount > 0;
2108 		sleep_finish(&sls, wait);
2109 		if (!recheck)
2110 			break;
2111 	}
2112 
2113 	return wait;
2114 }
2115 
2116 void
2117 single_thread_clear(struct proc *p, int flag)
2118 {
2119 	struct process *pr = p->p_p;
2120 	struct proc *q;
2121 	int s;
2122 
2123 	KASSERT(pr->ps_single == p);
2124 	KASSERT(curproc == p);
2125 
2126 	SCHED_LOCK(s);
2127 	pr->ps_single = NULL;
2128 	atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND | PS_SINGLEEXIT);
2129 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
2130 		if (q == p || (q->p_flag & P_SUSPSINGLE) == 0)
2131 			continue;
2132 		atomic_clearbits_int(&q->p_flag, P_SUSPSINGLE);
2133 
2134 		/*
2135 		 * if the thread was only stopped for single threading
2136 		 * then clearing that either makes it runnable or puts
2137 		 * it back into some sleep queue
2138 		 */
2139 		if (q->p_stat == SSTOP && (q->p_flag & flag) == 0) {
2140 			if (q->p_wchan == NULL)
2141 				setrunnable(q);
2142 			else
2143 				q->p_stat = SSLEEP;
2144 		}
2145 	}
2146 	SCHED_UNLOCK(s);
2147 }
2148 
2149 void
2150 sigio_del(struct sigiolst *rmlist)
2151 {
2152 	struct sigio *sigio;
2153 
2154 	while ((sigio = LIST_FIRST(rmlist)) != NULL) {
2155 		LIST_REMOVE(sigio, sio_pgsigio);
2156 		crfree(sigio->sio_ucred);
2157 		free(sigio, M_SIGIO, sizeof(*sigio));
2158 	}
2159 }
2160 
2161 void
2162 sigio_unlink(struct sigio_ref *sir, struct sigiolst *rmlist)
2163 {
2164 	struct sigio *sigio;
2165 
2166 	MUTEX_ASSERT_LOCKED(&sigio_lock);
2167 
2168 	sigio = sir->sir_sigio;
2169 	if (sigio != NULL) {
2170 		KASSERT(sigio->sio_myref == sir);
2171 		sir->sir_sigio = NULL;
2172 
2173 		if (sigio->sio_pgid > 0)
2174 			sigio->sio_proc = NULL;
2175 		else
2176 			sigio->sio_pgrp = NULL;
2177 		LIST_REMOVE(sigio, sio_pgsigio);
2178 
2179 		LIST_INSERT_HEAD(rmlist, sigio, sio_pgsigio);
2180 	}
2181 }
2182 
2183 void
2184 sigio_free(struct sigio_ref *sir)
2185 {
2186 	struct sigiolst rmlist;
2187 
2188 	if (sir->sir_sigio == NULL)
2189 		return;
2190 
2191 	LIST_INIT(&rmlist);
2192 
2193 	mtx_enter(&sigio_lock);
2194 	sigio_unlink(sir, &rmlist);
2195 	mtx_leave(&sigio_lock);
2196 
2197 	sigio_del(&rmlist);
2198 }
2199 
2200 void
2201 sigio_freelist(struct sigiolst *sigiolst)
2202 {
2203 	struct sigiolst rmlist;
2204 	struct sigio *sigio;
2205 
2206 	if (LIST_EMPTY(sigiolst))
2207 		return;
2208 
2209 	LIST_INIT(&rmlist);
2210 
2211 	mtx_enter(&sigio_lock);
2212 	while ((sigio = LIST_FIRST(sigiolst)) != NULL)
2213 		sigio_unlink(sigio->sio_myref, &rmlist);
2214 	mtx_leave(&sigio_lock);
2215 
2216 	sigio_del(&rmlist);
2217 }
2218 
2219 int
2220 sigio_setown(struct sigio_ref *sir, u_long cmd, caddr_t data)
2221 {
2222 	struct sigiolst rmlist;
2223 	struct proc *p = curproc;
2224 	struct pgrp *pgrp = NULL;
2225 	struct process *pr = NULL;
2226 	struct sigio *sigio;
2227 	int error;
2228 	pid_t pgid = *(int *)data;
2229 
2230 	if (pgid == 0) {
2231 		sigio_free(sir);
2232 		return (0);
2233 	}
2234 
2235 	if (cmd == TIOCSPGRP) {
2236 		if (pgid < 0)
2237 			return (EINVAL);
2238 		pgid = -pgid;
2239 	}
2240 
2241 	sigio = malloc(sizeof(*sigio), M_SIGIO, M_WAITOK);
2242 	sigio->sio_pgid = pgid;
2243 	sigio->sio_ucred = crhold(p->p_ucred);
2244 	sigio->sio_myref = sir;
2245 
2246 	LIST_INIT(&rmlist);
2247 
2248 	/*
2249 	 * The kernel lock, and not sleeping between prfind()/pgfind() and
2250 	 * linking of the sigio ensure that the process or process group does
2251 	 * not disappear unexpectedly.
2252 	 */
2253 	KERNEL_LOCK();
2254 	mtx_enter(&sigio_lock);
2255 
2256 	if (pgid > 0) {
2257 		pr = prfind(pgid);
2258 		if (pr == NULL) {
2259 			error = ESRCH;
2260 			goto fail;
2261 		}
2262 
2263 		/*
2264 		 * Policy - Don't allow a process to FSETOWN a process
2265 		 * in another session.
2266 		 *
2267 		 * Remove this test to allow maximum flexibility or
2268 		 * restrict FSETOWN to the current process or process
2269 		 * group for maximum safety.
2270 		 */
2271 		if (pr->ps_session != p->p_p->ps_session) {
2272 			error = EPERM;
2273 			goto fail;
2274 		}
2275 
2276 		if ((pr->ps_flags & PS_EXITING) != 0) {
2277 			error = ESRCH;
2278 			goto fail;
2279 		}
2280 	} else /* if (pgid < 0) */ {
2281 		pgrp = pgfind(-pgid);
2282 		if (pgrp == NULL) {
2283 			error = ESRCH;
2284 			goto fail;
2285 		}
2286 
2287 		/*
2288 		 * Policy - Don't allow a process to FSETOWN a process
2289 		 * in another session.
2290 		 *
2291 		 * Remove this test to allow maximum flexibility or
2292 		 * restrict FSETOWN to the current process or process
2293 		 * group for maximum safety.
2294 		 */
2295 		if (pgrp->pg_session != p->p_p->ps_session) {
2296 			error = EPERM;
2297 			goto fail;
2298 		}
2299 	}
2300 
2301 	if (pgid > 0) {
2302 		sigio->sio_proc = pr;
2303 		LIST_INSERT_HEAD(&pr->ps_sigiolst, sigio, sio_pgsigio);
2304 	} else {
2305 		sigio->sio_pgrp = pgrp;
2306 		LIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
2307 	}
2308 
2309 	sigio_unlink(sir, &rmlist);
2310 	sir->sir_sigio = sigio;
2311 
2312 	mtx_leave(&sigio_lock);
2313 	KERNEL_UNLOCK();
2314 
2315 	sigio_del(&rmlist);
2316 
2317 	return (0);
2318 
2319 fail:
2320 	mtx_leave(&sigio_lock);
2321 	KERNEL_UNLOCK();
2322 
2323 	crfree(sigio->sio_ucred);
2324 	free(sigio, M_SIGIO, sizeof(*sigio));
2325 
2326 	return (error);
2327 }
2328 
2329 void
2330 sigio_getown(struct sigio_ref *sir, u_long cmd, caddr_t data)
2331 {
2332 	struct sigio *sigio;
2333 	pid_t pgid = 0;
2334 
2335 	mtx_enter(&sigio_lock);
2336 	sigio = sir->sir_sigio;
2337 	if (sigio != NULL)
2338 		pgid = sigio->sio_pgid;
2339 	mtx_leave(&sigio_lock);
2340 
2341 	if (cmd == TIOCGPGRP)
2342 		pgid = -pgid;
2343 
2344 	*(int *)data = pgid;
2345 }
2346 
2347 void
2348 sigio_copy(struct sigio_ref *dst, struct sigio_ref *src)
2349 {
2350 	struct sigiolst rmlist;
2351 	struct sigio *newsigio, *sigio;
2352 
2353 	sigio_free(dst);
2354 
2355 	if (src->sir_sigio == NULL)
2356 		return;
2357 
2358 	newsigio = malloc(sizeof(*newsigio), M_SIGIO, M_WAITOK);
2359 	LIST_INIT(&rmlist);
2360 
2361 	mtx_enter(&sigio_lock);
2362 
2363 	sigio = src->sir_sigio;
2364 	if (sigio == NULL) {
2365 		mtx_leave(&sigio_lock);
2366 		free(newsigio, M_SIGIO, sizeof(*newsigio));
2367 		return;
2368 	}
2369 
2370 	newsigio->sio_pgid = sigio->sio_pgid;
2371 	newsigio->sio_ucred = crhold(sigio->sio_ucred);
2372 	newsigio->sio_myref = dst;
2373 	if (newsigio->sio_pgid > 0) {
2374 		newsigio->sio_proc = sigio->sio_proc;
2375 		LIST_INSERT_HEAD(&newsigio->sio_proc->ps_sigiolst, newsigio,
2376 		    sio_pgsigio);
2377 	} else {
2378 		newsigio->sio_pgrp = sigio->sio_pgrp;
2379 		LIST_INSERT_HEAD(&newsigio->sio_pgrp->pg_sigiolst, newsigio,
2380 		    sio_pgsigio);
2381 	}
2382 
2383 	sigio_unlink(dst, &rmlist);
2384 	dst->sir_sigio = newsigio;
2385 
2386 	mtx_leave(&sigio_lock);
2387 
2388 	sigio_del(&rmlist);
2389 }
2390