xref: /openbsd-src/sys/kern/kern_sig.c (revision e0e8a7c4892bc434a32dba76ca02883f9b1071d1)
1 /*	$OpenBSD: kern_sig.c,v 1.225 2018/11/12 15:09:17 visa 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 #define	SIGPROP		/* include signal properties table */
42 #include <sys/param.h>
43 #include <sys/signalvar.h>
44 #include <sys/resourcevar.h>
45 #include <sys/queue.h>
46 #include <sys/namei.h>
47 #include <sys/vnode.h>
48 #include <sys/event.h>
49 #include <sys/proc.h>
50 #include <sys/systm.h>
51 #include <sys/acct.h>
52 #include <sys/fcntl.h>
53 #include <sys/filedesc.h>
54 #include <sys/kernel.h>
55 #include <sys/wait.h>
56 #include <sys/ktrace.h>
57 #include <sys/stat.h>
58 #include <sys/core.h>
59 #include <sys/malloc.h>
60 #include <sys/pool.h>
61 #include <sys/ptrace.h>
62 #include <sys/sched.h>
63 #include <sys/user.h>
64 #include <sys/syslog.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 struct filterops sig_filtops =
79 	{ 0, filt_sigattach, filt_sigdetach, filt_signal };
80 
81 void proc_stop(struct proc *p, int);
82 void proc_stop_sweep(void *);
83 struct timeout proc_stop_to;
84 
85 void postsig(struct proc *, int);
86 int cansignal(struct proc *, struct process *, int);
87 
88 struct pool sigacts_pool;	/* memory pool for sigacts structures */
89 
90 void sigio_del(struct sigiolst *);
91 void sigio_unlink(struct sigio_ref *, struct sigiolst *);
92 struct mutex sigio_lock = MUTEX_INITIALIZER(IPL_HIGH);
93 
94 /*
95  * Can thread p, send the signal signum to process qr?
96  */
97 int
98 cansignal(struct proc *p, struct process *qr, int signum)
99 {
100 	struct process *pr = p->p_p;
101 	struct ucred *uc = p->p_ucred;
102 	struct ucred *quc = qr->ps_ucred;
103 
104 	if (uc->cr_uid == 0)
105 		return (1);		/* root can always signal */
106 
107 	if (pr == qr)
108 		return (1);		/* process can always signal itself */
109 
110 	/* optimization: if the same creds then the tests below will pass */
111 	if (uc == quc)
112 		return (1);
113 
114 	if (signum == SIGCONT && qr->ps_session == pr->ps_session)
115 		return (1);		/* SIGCONT in session */
116 
117 	/*
118 	 * Using kill(), only certain signals can be sent to setugid
119 	 * child processes
120 	 */
121 	if (qr->ps_flags & PS_SUGID) {
122 		switch (signum) {
123 		case 0:
124 		case SIGKILL:
125 		case SIGINT:
126 		case SIGTERM:
127 		case SIGALRM:
128 		case SIGSTOP:
129 		case SIGTTIN:
130 		case SIGTTOU:
131 		case SIGTSTP:
132 		case SIGHUP:
133 		case SIGUSR1:
134 		case SIGUSR2:
135 			if (uc->cr_ruid == quc->cr_ruid ||
136 			    uc->cr_uid == quc->cr_ruid)
137 				return (1);
138 		}
139 		return (0);
140 	}
141 
142 	if (uc->cr_ruid == quc->cr_ruid ||
143 	    uc->cr_ruid == quc->cr_svuid ||
144 	    uc->cr_uid == quc->cr_ruid ||
145 	    uc->cr_uid == quc->cr_svuid)
146 		return (1);
147 	return (0);
148 }
149 
150 /*
151  * Initialize signal-related data structures.
152  */
153 void
154 signal_init(void)
155 {
156 	timeout_set(&proc_stop_to, proc_stop_sweep, NULL);
157 
158 	pool_init(&sigacts_pool, sizeof(struct sigacts), 0, IPL_NONE,
159 	    PR_WAITOK, "sigapl", NULL);
160 }
161 
162 /*
163  * Create an initial sigacts structure, using the same signal state
164  * as p.
165  */
166 struct sigacts *
167 sigactsinit(struct process *pr)
168 {
169 	struct sigacts *ps;
170 
171 	ps = pool_get(&sigacts_pool, PR_WAITOK);
172 	memcpy(ps, pr->ps_sigacts, sizeof(struct sigacts));
173 	ps->ps_refcnt = 1;
174 	return (ps);
175 }
176 
177 /*
178  * Share a sigacts structure.
179  */
180 struct sigacts *
181 sigactsshare(struct process *pr)
182 {
183 	struct sigacts *ps = pr->ps_sigacts;
184 
185 	ps->ps_refcnt++;
186 	return ps;
187 }
188 
189 /*
190  * Initialize a new sigaltstack structure.
191  */
192 void
193 sigstkinit(struct sigaltstack *ss)
194 {
195 	ss->ss_flags = SS_DISABLE;
196 	ss->ss_size = 0;
197 	ss->ss_sp = 0;
198 }
199 
200 /*
201  * Make this process not share its sigacts, maintaining all
202  * signal state.
203  */
204 void
205 sigactsunshare(struct process *pr)
206 {
207 	struct sigacts *newps;
208 
209 	if (pr->ps_sigacts->ps_refcnt == 1)
210 		return;
211 
212 	newps = sigactsinit(pr);
213 	sigactsfree(pr);
214 	pr->ps_sigacts = newps;
215 }
216 
217 /*
218  * Release a sigacts structure.
219  */
220 void
221 sigactsfree(struct process *pr)
222 {
223 	struct sigacts *ps = pr->ps_sigacts;
224 
225 	if (--ps->ps_refcnt > 0)
226 		return;
227 
228 	pr->ps_sigacts = NULL;
229 
230 	pool_put(&sigacts_pool, ps);
231 }
232 
233 int
234 sys_sigaction(struct proc *p, void *v, register_t *retval)
235 {
236 	struct sys_sigaction_args /* {
237 		syscallarg(int) signum;
238 		syscallarg(const struct sigaction *) nsa;
239 		syscallarg(struct sigaction *) osa;
240 	} */ *uap = v;
241 	struct sigaction vec;
242 #ifdef KTRACE
243 	struct sigaction ovec;
244 #endif
245 	struct sigaction *sa;
246 	const struct sigaction *nsa;
247 	struct sigaction *osa;
248 	struct sigacts *ps = p->p_p->ps_sigacts;
249 	int signum;
250 	int bit, error;
251 
252 	signum = SCARG(uap, signum);
253 	nsa = SCARG(uap, nsa);
254 	osa = SCARG(uap, osa);
255 
256 	if (signum <= 0 || signum >= NSIG ||
257 	    (nsa && (signum == SIGKILL || signum == SIGSTOP)))
258 		return (EINVAL);
259 	sa = &vec;
260 	if (osa) {
261 		sa->sa_handler = ps->ps_sigact[signum];
262 		sa->sa_mask = ps->ps_catchmask[signum];
263 		bit = sigmask(signum);
264 		sa->sa_flags = 0;
265 		if ((ps->ps_sigonstack & bit) != 0)
266 			sa->sa_flags |= SA_ONSTACK;
267 		if ((ps->ps_sigintr & bit) == 0)
268 			sa->sa_flags |= SA_RESTART;
269 		if ((ps->ps_sigreset & bit) != 0)
270 			sa->sa_flags |= SA_RESETHAND;
271 		if ((ps->ps_siginfo & bit) != 0)
272 			sa->sa_flags |= SA_SIGINFO;
273 		if (signum == SIGCHLD) {
274 			if ((ps->ps_flags & SAS_NOCLDSTOP) != 0)
275 				sa->sa_flags |= SA_NOCLDSTOP;
276 			if ((ps->ps_flags & SAS_NOCLDWAIT) != 0)
277 				sa->sa_flags |= SA_NOCLDWAIT;
278 		}
279 		if ((sa->sa_mask & bit) == 0)
280 			sa->sa_flags |= SA_NODEFER;
281 		sa->sa_mask &= ~bit;
282 		error = copyout(sa, osa, sizeof (vec));
283 		if (error)
284 			return (error);
285 #ifdef KTRACE
286 		if (KTRPOINT(p, KTR_STRUCT))
287 			ovec = vec;
288 #endif
289 	}
290 	if (nsa) {
291 		error = copyin(nsa, sa, sizeof (vec));
292 		if (error)
293 			return (error);
294 #ifdef KTRACE
295 		if (KTRPOINT(p, KTR_STRUCT))
296 			ktrsigaction(p, sa);
297 #endif
298 		setsigvec(p, signum, sa);
299 	}
300 #ifdef KTRACE
301 	if (osa && KTRPOINT(p, KTR_STRUCT))
302 		ktrsigaction(p, &ovec);
303 #endif
304 	return (0);
305 }
306 
307 void
308 setsigvec(struct proc *p, int signum, struct sigaction *sa)
309 {
310 	struct sigacts *ps = p->p_p->ps_sigacts;
311 	int bit;
312 	int s;
313 
314 	bit = sigmask(signum);
315 	/*
316 	 * Change setting atomically.
317 	 */
318 	s = splhigh();
319 	ps->ps_sigact[signum] = sa->sa_handler;
320 	if ((sa->sa_flags & SA_NODEFER) == 0)
321 		sa->sa_mask |= sigmask(signum);
322 	ps->ps_catchmask[signum] = sa->sa_mask &~ sigcantmask;
323 	if (signum == SIGCHLD) {
324 		if (sa->sa_flags & SA_NOCLDSTOP)
325 			atomic_setbits_int(&ps->ps_flags, SAS_NOCLDSTOP);
326 		else
327 			atomic_clearbits_int(&ps->ps_flags, SAS_NOCLDSTOP);
328 		/*
329 		 * If the SA_NOCLDWAIT flag is set or the handler
330 		 * is SIG_IGN we reparent the dying child to PID 1
331 		 * (init) which will reap the zombie.  Because we use
332 		 * init to do our dirty work we never set SAS_NOCLDWAIT
333 		 * for PID 1.
334 		 * XXX exit1 rework means this is unnecessary?
335 		 */
336 		if (initprocess->ps_sigacts != ps &&
337 		    ((sa->sa_flags & SA_NOCLDWAIT) ||
338 		    sa->sa_handler == SIG_IGN))
339 			atomic_setbits_int(&ps->ps_flags, SAS_NOCLDWAIT);
340 		else
341 			atomic_clearbits_int(&ps->ps_flags, SAS_NOCLDWAIT);
342 	}
343 	if ((sa->sa_flags & SA_RESETHAND) != 0)
344 		ps->ps_sigreset |= bit;
345 	else
346 		ps->ps_sigreset &= ~bit;
347 	if ((sa->sa_flags & SA_SIGINFO) != 0)
348 		ps->ps_siginfo |= bit;
349 	else
350 		ps->ps_siginfo &= ~bit;
351 	if ((sa->sa_flags & SA_RESTART) == 0)
352 		ps->ps_sigintr |= bit;
353 	else
354 		ps->ps_sigintr &= ~bit;
355 	if ((sa->sa_flags & SA_ONSTACK) != 0)
356 		ps->ps_sigonstack |= bit;
357 	else
358 		ps->ps_sigonstack &= ~bit;
359 	/*
360 	 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
361 	 * and for signals set to SIG_DFL where the default is to ignore.
362 	 * However, don't put SIGCONT in ps_sigignore,
363 	 * as we have to restart the process.
364 	 */
365 	if (sa->sa_handler == SIG_IGN ||
366 	    (sigprop[signum] & SA_IGNORE && sa->sa_handler == SIG_DFL)) {
367 		atomic_clearbits_int(&p->p_siglist, bit);
368 		if (signum != SIGCONT)
369 			ps->ps_sigignore |= bit;	/* easier in psignal */
370 		ps->ps_sigcatch &= ~bit;
371 	} else {
372 		ps->ps_sigignore &= ~bit;
373 		if (sa->sa_handler == SIG_DFL)
374 			ps->ps_sigcatch &= ~bit;
375 		else
376 			ps->ps_sigcatch |= bit;
377 	}
378 	splx(s);
379 }
380 
381 /*
382  * Initialize signal state for process 0;
383  * set to ignore signals that are ignored by default.
384  */
385 void
386 siginit(struct process *pr)
387 {
388 	struct sigacts *ps = pr->ps_sigacts;
389 	int i;
390 
391 	for (i = 0; i < NSIG; i++)
392 		if (sigprop[i] & SA_IGNORE && i != SIGCONT)
393 			ps->ps_sigignore |= sigmask(i);
394 	ps->ps_flags = SAS_NOCLDWAIT | SAS_NOCLDSTOP;
395 }
396 
397 /*
398  * Reset signals for an exec by the specified thread.
399  */
400 void
401 execsigs(struct proc *p)
402 {
403 	struct sigacts *ps;
404 	int nc, mask;
405 
406 	sigactsunshare(p->p_p);
407 	ps = p->p_p->ps_sigacts;
408 
409 	/*
410 	 * Reset caught signals.  Held signals remain held
411 	 * through p_sigmask (unless they were caught,
412 	 * and are now ignored by default).
413 	 */
414 	while (ps->ps_sigcatch) {
415 		nc = ffs((long)ps->ps_sigcatch);
416 		mask = sigmask(nc);
417 		ps->ps_sigcatch &= ~mask;
418 		if (sigprop[nc] & SA_IGNORE) {
419 			if (nc != SIGCONT)
420 				ps->ps_sigignore |= mask;
421 			atomic_clearbits_int(&p->p_siglist, mask);
422 		}
423 		ps->ps_sigact[nc] = SIG_DFL;
424 	}
425 	/*
426 	 * Reset stack state to the user stack.
427 	 * Clear set of signals caught on the signal stack.
428 	 */
429 	sigstkinit(&p->p_sigstk);
430 	ps->ps_flags &= ~SAS_NOCLDWAIT;
431 	if (ps->ps_sigact[SIGCHLD] == SIG_IGN)
432 		ps->ps_sigact[SIGCHLD] = SIG_DFL;
433 }
434 
435 /*
436  * Manipulate signal mask.
437  * Note that we receive new mask, not pointer,
438  * and return old mask as return value;
439  * the library stub does the rest.
440  */
441 int
442 sys_sigprocmask(struct proc *p, void *v, register_t *retval)
443 {
444 	struct sys_sigprocmask_args /* {
445 		syscallarg(int) how;
446 		syscallarg(sigset_t) mask;
447 	} */ *uap = v;
448 	int error = 0;
449 	sigset_t mask;
450 
451 	*retval = p->p_sigmask;
452 	mask = SCARG(uap, mask) &~ sigcantmask;
453 
454 	switch (SCARG(uap, how)) {
455 	case SIG_BLOCK:
456 		atomic_setbits_int(&p->p_sigmask, mask);
457 		break;
458 	case SIG_UNBLOCK:
459 		atomic_clearbits_int(&p->p_sigmask, mask);
460 		break;
461 	case SIG_SETMASK:
462 		p->p_sigmask = mask;
463 		break;
464 	default:
465 		error = EINVAL;
466 		break;
467 	}
468 	return (error);
469 }
470 
471 int
472 sys_sigpending(struct proc *p, void *v, register_t *retval)
473 {
474 
475 	*retval = p->p_siglist;
476 	return (0);
477 }
478 
479 /*
480  * Temporarily replace calling proc's signal mask for the duration of a
481  * system call.  Original signal mask will be restored by userret().
482  */
483 void
484 dosigsuspend(struct proc *p, sigset_t newmask)
485 {
486 	KASSERT(p == curproc);
487 
488 	p->p_oldmask = p->p_sigmask;
489 	atomic_setbits_int(&p->p_flag, P_SIGSUSPEND);
490 	p->p_sigmask = newmask;
491 }
492 
493 /*
494  * Suspend process until signal, providing mask to be set
495  * in the meantime.  Note nonstandard calling convention:
496  * libc stub passes mask, not pointer, to save a copyin.
497  */
498 int
499 sys_sigsuspend(struct proc *p, void *v, register_t *retval)
500 {
501 	struct sys_sigsuspend_args /* {
502 		syscallarg(int) mask;
503 	} */ *uap = v;
504 	struct process *pr = p->p_p;
505 	struct sigacts *ps = pr->ps_sigacts;
506 
507 	dosigsuspend(p, SCARG(uap, mask) &~ sigcantmask);
508 	while (tsleep(ps, PPAUSE|PCATCH, "pause", 0) == 0)
509 		/* void */;
510 	/* always return EINTR rather than ERESTART... */
511 	return (EINTR);
512 }
513 
514 int
515 sigonstack(size_t stack)
516 {
517 	const struct sigaltstack *ss = &curproc->p_sigstk;
518 
519 	return (ss->ss_flags & SS_DISABLE ? 0 :
520 	    (stack - (size_t)ss->ss_sp < ss->ss_size));
521 }
522 
523 int
524 sys_sigaltstack(struct proc *p, void *v, register_t *retval)
525 {
526 	struct sys_sigaltstack_args /* {
527 		syscallarg(const struct sigaltstack *) nss;
528 		syscallarg(struct sigaltstack *) oss;
529 	} */ *uap = v;
530 	struct sigaltstack ss;
531 	const struct sigaltstack *nss;
532 	struct sigaltstack *oss;
533 	int onstack = sigonstack(PROC_STACK(p));
534 	int error;
535 
536 	nss = SCARG(uap, nss);
537 	oss = SCARG(uap, oss);
538 
539 	if (oss != NULL) {
540 		ss = p->p_sigstk;
541 		if (onstack)
542 			ss.ss_flags |= SS_ONSTACK;
543 		if ((error = copyout(&ss, oss, sizeof(ss))))
544 			return (error);
545 	}
546 	if (nss == NULL)
547 		return (0);
548 	error = copyin(nss, &ss, sizeof(ss));
549 	if (error)
550 		return (error);
551 	if (onstack)
552 		return (EPERM);
553 	if (ss.ss_flags & ~SS_DISABLE)
554 		return (EINVAL);
555 	if (ss.ss_flags & SS_DISABLE) {
556 		p->p_sigstk.ss_flags = ss.ss_flags;
557 		return (0);
558 	}
559 	if (ss.ss_size < MINSIGSTKSZ)
560 		return (ENOMEM);
561 
562 	error = uvm_map_remap_as_stack(p, (vaddr_t)ss.ss_sp, ss.ss_size);
563 	if (error)
564 		return (error);
565 
566 	p->p_sigstk = ss;
567 	return (0);
568 }
569 
570 int
571 sys_kill(struct proc *cp, void *v, register_t *retval)
572 {
573 	struct sys_kill_args /* {
574 		syscallarg(int) pid;
575 		syscallarg(int) signum;
576 	} */ *uap = v;
577 	struct process *pr;
578 	int pid = SCARG(uap, pid);
579 	int signum = SCARG(uap, signum);
580 	int error;
581 	int zombie = 0;
582 
583 	if ((error = pledge_kill(cp, pid)) != 0)
584 		return (error);
585 	if (((u_int)signum) >= NSIG)
586 		return (EINVAL);
587 	if (pid > 0) {
588 		if ((pr = prfind(pid)) == NULL) {
589 			if ((pr = zombiefind(pid)) == NULL)
590 				return (ESRCH);
591 			else
592 				zombie = 1;
593 		}
594 		if (!cansignal(cp, pr, signum))
595 			return (EPERM);
596 
597 		/* kill single process */
598 		if (signum && !zombie)
599 			prsignal(pr, signum);
600 		return (0);
601 	}
602 	switch (pid) {
603 	case -1:		/* broadcast signal */
604 		return (killpg1(cp, signum, 0, 1));
605 	case 0:			/* signal own process group */
606 		return (killpg1(cp, signum, 0, 0));
607 	default:		/* negative explicit process group */
608 		return (killpg1(cp, signum, -pid, 0));
609 	}
610 }
611 
612 int
613 sys_thrkill(struct proc *cp, void *v, register_t *retval)
614 {
615 	struct sys_thrkill_args /* {
616 		syscallarg(pid_t) tid;
617 		syscallarg(int) signum;
618 		syscallarg(void *) tcb;
619 	} */ *uap = v;
620 	struct proc *p;
621 	int tid = SCARG(uap, tid);
622 	int signum = SCARG(uap, signum);
623 	void *tcb;
624 
625 	if (((u_int)signum) >= NSIG)
626 		return (EINVAL);
627 	if (tid > THREAD_PID_OFFSET) {
628 		if ((p = tfind(tid - THREAD_PID_OFFSET)) == NULL)
629 			return (ESRCH);
630 
631 		/* can only kill threads in the same process */
632 		if (p->p_p != cp->p_p)
633 			return (ESRCH);
634 	} else if (tid == 0)
635 		p = cp;
636 	else
637 		return (EINVAL);
638 
639 	/* optionally require the target thread to have the given tcb addr */
640 	tcb = SCARG(uap, tcb);
641 	if (tcb != NULL && tcb != TCB_GET(p))
642 		return (ESRCH);
643 
644 	if (signum)
645 		ptsignal(p, signum, STHREAD);
646 	return (0);
647 }
648 
649 /*
650  * Common code for kill process group/broadcast kill.
651  * cp is calling process.
652  */
653 int
654 killpg1(struct proc *cp, int signum, int pgid, int all)
655 {
656 	struct process *pr;
657 	struct pgrp *pgrp;
658 	int nfound = 0;
659 
660 	if (all) {
661 		/*
662 		 * broadcast
663 		 */
664 		LIST_FOREACH(pr, &allprocess, ps_list) {
665 			if (pr->ps_pid <= 1 ||
666 			    pr->ps_flags & (PS_SYSTEM | PS_NOBROADCASTKILL) ||
667 			    pr == cp->p_p || !cansignal(cp, pr, signum))
668 				continue;
669 			nfound++;
670 			if (signum)
671 				prsignal(pr, signum);
672 		}
673 	} else {
674 		if (pgid == 0)
675 			/*
676 			 * zero pgid means send to my process group.
677 			 */
678 			pgrp = cp->p_p->ps_pgrp;
679 		else {
680 			pgrp = pgfind(pgid);
681 			if (pgrp == NULL)
682 				return (ESRCH);
683 		}
684 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist) {
685 			if (pr->ps_pid <= 1 || pr->ps_flags & PS_SYSTEM ||
686 			    !cansignal(cp, pr, signum))
687 				continue;
688 			nfound++;
689 			if (signum)
690 				prsignal(pr, signum);
691 		}
692 	}
693 	return (nfound ? 0 : ESRCH);
694 }
695 
696 #define CANDELIVER(uid, euid, pr) \
697 	(euid == 0 || \
698 	(uid) == (pr)->ps_ucred->cr_ruid || \
699 	(uid) == (pr)->ps_ucred->cr_svuid || \
700 	(uid) == (pr)->ps_ucred->cr_uid || \
701 	(euid) == (pr)->ps_ucred->cr_ruid || \
702 	(euid) == (pr)->ps_ucred->cr_svuid || \
703 	(euid) == (pr)->ps_ucred->cr_uid)
704 
705 #define CANSIGIO(cr, pr) \
706 	CANDELIVER((cr)->cr_ruid, (cr)->cr_uid, (pr))
707 
708 /*
709  * Deliver signum to pgid, but first check uid/euid against each
710  * process and see if it is permitted.
711  */
712 void
713 csignal(pid_t pgid, int signum, uid_t uid, uid_t euid)
714 {
715 	struct pgrp *pgrp;
716 	struct process *pr;
717 
718 	if (pgid == 0)
719 		return;
720 	if (pgid < 0) {
721 		pgid = -pgid;
722 		if ((pgrp = pgfind(pgid)) == NULL)
723 			return;
724 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist)
725 			if (CANDELIVER(uid, euid, pr))
726 				prsignal(pr, signum);
727 	} else {
728 		if ((pr = prfind(pgid)) == NULL)
729 			return;
730 		if (CANDELIVER(uid, euid, pr))
731 			prsignal(pr, signum);
732 	}
733 }
734 
735 /*
736  * Send a signal to a process group.
737  */
738 void
739 gsignal(int pgid, int signum)
740 {
741 	struct pgrp *pgrp;
742 
743 	if (pgid && (pgrp = pgfind(pgid)))
744 		pgsignal(pgrp, signum, 0);
745 }
746 
747 /*
748  * Send a signal to a process group.  If checktty is 1,
749  * limit to members which have a controlling terminal.
750  */
751 void
752 pgsignal(struct pgrp *pgrp, int signum, int checkctty)
753 {
754 	struct process *pr;
755 
756 	if (pgrp)
757 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist)
758 			if (checkctty == 0 || pr->ps_flags & PS_CONTROLT)
759 				prsignal(pr, signum);
760 }
761 
762 /*
763  * Send a SIGIO or SIGURG signal to a process or process group using stored
764  * credentials rather than those of the current process.
765  */
766 void
767 pgsigio(struct sigio_ref *sir, int sig, int checkctty)
768 {
769 	struct process *pr;
770 	struct sigio *sigio;
771 
772 	if (sir->sir_sigio == NULL)
773 		return;
774 
775 	mtx_enter(&sigio_lock);
776 	sigio = sir->sir_sigio;
777 	if (sigio == NULL)
778 		goto out;
779 	if (sigio->sio_pgid > 0) {
780 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc))
781 			prsignal(sigio->sio_proc, sig);
782 	} else if (sigio->sio_pgid < 0) {
783 		LIST_FOREACH(pr, &sigio->sio_pgrp->pg_members, ps_pglist) {
784 			if (CANSIGIO(sigio->sio_ucred, pr) &&
785 			    (checkctty == 0 || (pr->ps_flags & PS_CONTROLT)))
786 				prsignal(pr, sig);
787 		}
788 	}
789 out:
790 	mtx_leave(&sigio_lock);
791 }
792 
793 /*
794  * Recalculate the signal mask and reset the signal disposition after
795  * usermode frame for delivery is formed.
796  */
797 void
798 postsig_done(struct proc *p, int signum, struct sigacts *ps)
799 {
800 	int mask = sigmask(signum);
801 
802 	KERNEL_ASSERT_LOCKED();
803 
804 	p->p_ru.ru_nsignals++;
805 	atomic_setbits_int(&p->p_sigmask, ps->ps_catchmask[signum]);
806 	if ((ps->ps_sigreset & mask) != 0) {
807 		ps->ps_sigcatch &= ~mask;
808 		if (signum != SIGCONT && sigprop[signum] & SA_IGNORE)
809 			ps->ps_sigignore |= mask;
810 		ps->ps_sigact[signum] = SIG_DFL;
811 	}
812 }
813 
814 /*
815  * Send a signal caused by a trap to the current thread
816  * If it will be caught immediately, deliver it with correct code.
817  * Otherwise, post it normally.
818  */
819 void
820 trapsignal(struct proc *p, int signum, u_long trapno, int code,
821     union sigval sigval)
822 {
823 	struct process *pr = p->p_p;
824 	struct sigacts *ps = pr->ps_sigacts;
825 	int mask;
826 
827 	switch (signum) {
828 	case SIGILL:
829 	case SIGBUS:
830 	case SIGSEGV:
831 		pr->ps_acflag |= ATRAP;
832 		break;
833 	}
834 
835 	mask = sigmask(signum);
836 	if ((pr->ps_flags & PS_TRACED) == 0 &&
837 	    (ps->ps_sigcatch & mask) != 0 &&
838 	    (p->p_sigmask & mask) == 0) {
839 		siginfo_t si;
840 		initsiginfo(&si, signum, trapno, code, sigval);
841 #ifdef KTRACE
842 		if (KTRPOINT(p, KTR_PSIG)) {
843 			ktrpsig(p, signum, ps->ps_sigact[signum],
844 			    p->p_sigmask, code, &si);
845 		}
846 #endif
847 		sendsig(ps->ps_sigact[signum], signum, p->p_sigmask, &si);
848 		postsig_done(p, signum, ps);
849 	} else {
850 		p->p_sisig = signum;
851 		p->p_sitrapno = trapno;	/* XXX for core dump/debugger */
852 		p->p_sicode = code;
853 		p->p_sigval = sigval;
854 
855 		/*
856 		 * Signals like SIGBUS and SIGSEGV should not, when
857 		 * generated by the kernel, be ignorable or blockable.
858 		 * If it is and we're not being traced, then just kill
859 		 * the process.
860 		 */
861 		if ((pr->ps_flags & PS_TRACED) == 0 &&
862 		    (sigprop[signum] & SA_KILL) &&
863 		    ((p->p_sigmask & mask) || (ps->ps_sigignore & mask)))
864 			sigexit(p, signum);
865 		ptsignal(p, signum, STHREAD);
866 	}
867 }
868 
869 /*
870  * Send the signal to the process.  If the signal has an action, the action
871  * is usually performed by the target process rather than the caller; we add
872  * the signal to the set of pending signals for the process.
873  *
874  * Exceptions:
875  *   o When a stop signal is sent to a sleeping process that takes the
876  *     default action, the process is stopped without awakening it.
877  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
878  *     regardless of the signal action (eg, blocked or ignored).
879  *
880  * Other ignored signals are discarded immediately.
881  */
882 void
883 psignal(struct proc *p, int signum)
884 {
885 	ptsignal(p, signum, SPROCESS);
886 }
887 
888 /*
889  * type = SPROCESS	process signal, can be diverted (sigwait())
890  *	XXX if blocked in all threads, mark as pending in struct process
891  * type = STHREAD	thread signal, but should be propagated if unhandled
892  * type = SPROPAGATED	propagated to this thread, so don't propagate again
893  */
894 void
895 ptsignal(struct proc *p, int signum, enum signal_type type)
896 {
897 	int s, prop;
898 	sig_t action;
899 	int mask;
900 	struct process *pr = p->p_p;
901 	struct proc *q;
902 	int wakeparent = 0;
903 
904 #ifdef DIAGNOSTIC
905 	if ((u_int)signum >= NSIG || signum == 0)
906 		panic("psignal signal number");
907 #endif
908 
909 	/* Ignore signal if the target process is exiting */
910 	if (pr->ps_flags & PS_EXITING)
911 		return;
912 
913 	mask = sigmask(signum);
914 
915 	if (type == SPROCESS) {
916 		/* Accept SIGKILL to coredumping processes */
917 		if (pr->ps_flags & PS_COREDUMP && signum == SIGKILL) {
918 			if (pr->ps_single != NULL)
919 				p = pr->ps_single;
920 			atomic_setbits_int(&p->p_siglist, mask);
921 			return;
922 		}
923 
924 		/*
925 		 * If the current thread can process the signal
926 		 * immediately (it's unblocked) then have it take it.
927 		 */
928 		q = curproc;
929 		if (q != NULL && q->p_p == pr && (q->p_flag & P_WEXIT) == 0 &&
930 		    (q->p_sigmask & mask) == 0)
931 			p = q;
932 		else {
933 			/*
934 			 * A process-wide signal can be diverted to a
935 			 * different thread that's in sigwait() for this
936 			 * signal.  If there isn't such a thread, then
937 			 * pick a thread that doesn't have it blocked so
938 			 * that the stop/kill consideration isn't
939 			 * delayed.  Otherwise, mark it pending on the
940 			 * main thread.
941 			 */
942 			TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
943 				/* ignore exiting threads */
944 				if (q->p_flag & P_WEXIT)
945 					continue;
946 
947 				/* skip threads that have the signal blocked */
948 				if ((q->p_sigmask & mask) != 0)
949 					continue;
950 
951 				/* okay, could send to this thread */
952 				p = q;
953 
954 				/*
955 				 * sigsuspend, sigwait, ppoll/pselect, etc?
956 				 * Definitely go to this thread, as it's
957 				 * already blocked in the kernel.
958 				 */
959 				if (q->p_flag & P_SIGSUSPEND)
960 					break;
961 			}
962 		}
963 	}
964 
965 	if (type != SPROPAGATED)
966 		KNOTE(&pr->ps_klist, NOTE_SIGNAL | signum);
967 
968 	prop = sigprop[signum];
969 
970 	/*
971 	 * If proc is traced, always give parent a chance.
972 	 */
973 	if (pr->ps_flags & PS_TRACED) {
974 		action = SIG_DFL;
975 		atomic_setbits_int(&p->p_siglist, mask);
976 	} else {
977 		/*
978 		 * If the signal is being ignored,
979 		 * then we forget about it immediately.
980 		 * (Note: we don't set SIGCONT in ps_sigignore,
981 		 * and if it is set to SIG_IGN,
982 		 * action will be SIG_DFL here.)
983 		 */
984 		if (pr->ps_sigacts->ps_sigignore & mask)
985 			return;
986 		if (p->p_sigmask & mask) {
987 			action = SIG_HOLD;
988 		} else if (pr->ps_sigacts->ps_sigcatch & mask) {
989 			action = SIG_CATCH;
990 		} else {
991 			action = SIG_DFL;
992 
993 			if (prop & SA_KILL && pr->ps_nice > NZERO)
994 				 pr->ps_nice = NZERO;
995 
996 			/*
997 			 * If sending a tty stop signal to a member of an
998 			 * orphaned process group, discard the signal here if
999 			 * the action is default; don't stop the process below
1000 			 * if sleeping, and don't clear any pending SIGCONT.
1001 			 */
1002 			if (prop & SA_TTYSTOP && pr->ps_pgrp->pg_jobc == 0)
1003 				return;
1004 		}
1005 
1006 		atomic_setbits_int(&p->p_siglist, mask);
1007 	}
1008 
1009 	if (prop & SA_CONT)
1010 		atomic_clearbits_int(&p->p_siglist, stopsigmask);
1011 
1012 	if (prop & SA_STOP) {
1013 		atomic_clearbits_int(&p->p_siglist, contsigmask);
1014 		atomic_clearbits_int(&p->p_flag, P_CONTINUED);
1015 	}
1016 
1017 	/*
1018 	 * XXX delay processing of SA_STOP signals unless action == SIG_DFL?
1019 	 */
1020 	if (prop & (SA_CONT | SA_STOP) && type != SPROPAGATED)
1021 		TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link)
1022 			if (q != p)
1023 				ptsignal(q, signum, SPROPAGATED);
1024 
1025 	/*
1026 	 * Defer further processing for signals which are held,
1027 	 * except that stopped processes must be continued by SIGCONT.
1028 	 */
1029 	if (action == SIG_HOLD && ((prop & SA_CONT) == 0 || p->p_stat != SSTOP))
1030 		return;
1031 
1032 	SCHED_LOCK(s);
1033 
1034 	switch (p->p_stat) {
1035 
1036 	case SSLEEP:
1037 		/*
1038 		 * If process is sleeping uninterruptibly
1039 		 * we can't interrupt the sleep... the signal will
1040 		 * be noticed when the process returns through
1041 		 * trap() or syscall().
1042 		 */
1043 		if ((p->p_flag & P_SINTR) == 0)
1044 			goto out;
1045 		/*
1046 		 * Process is sleeping and traced... make it runnable
1047 		 * so it can discover the signal in issignal() and stop
1048 		 * for the parent.
1049 		 */
1050 		if (pr->ps_flags & PS_TRACED)
1051 			goto run;
1052 		/*
1053 		 * If SIGCONT is default (or ignored) and process is
1054 		 * asleep, we are finished; the process should not
1055 		 * be awakened.
1056 		 */
1057 		if ((prop & SA_CONT) && action == SIG_DFL) {
1058 			atomic_clearbits_int(&p->p_siglist, mask);
1059 			goto out;
1060 		}
1061 		/*
1062 		 * When a sleeping process receives a stop
1063 		 * signal, process immediately if possible.
1064 		 */
1065 		if ((prop & SA_STOP) && action == SIG_DFL) {
1066 			/*
1067 			 * If a child holding parent blocked,
1068 			 * stopping could cause deadlock.
1069 			 */
1070 			if (pr->ps_flags & PS_PPWAIT)
1071 				goto out;
1072 			atomic_clearbits_int(&p->p_siglist, mask);
1073 			p->p_xstat = signum;
1074 			proc_stop(p, 0);
1075 			goto out;
1076 		}
1077 		/*
1078 		 * All other (caught or default) signals
1079 		 * cause the process to run.
1080 		 */
1081 		goto runfast;
1082 		/*NOTREACHED*/
1083 
1084 	case SSTOP:
1085 		/*
1086 		 * If traced process is already stopped,
1087 		 * then no further action is necessary.
1088 		 */
1089 		if (pr->ps_flags & PS_TRACED)
1090 			goto out;
1091 
1092 		/*
1093 		 * Kill signal always sets processes running.
1094 		 */
1095 		if (signum == SIGKILL) {
1096 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1097 			goto runfast;
1098 		}
1099 
1100 		if (prop & SA_CONT) {
1101 			/*
1102 			 * If SIGCONT is default (or ignored), we continue the
1103 			 * process but don't leave the signal in p_siglist, as
1104 			 * it has no further action.  If SIGCONT is held, we
1105 			 * continue the process and leave the signal in
1106 			 * p_siglist.  If the process catches SIGCONT, let it
1107 			 * handle the signal itself.  If it isn't waiting on
1108 			 * an event, then it goes back to run state.
1109 			 * Otherwise, process goes back to sleep state.
1110 			 */
1111 			atomic_setbits_int(&p->p_flag, P_CONTINUED);
1112 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1113 			wakeparent = 1;
1114 			if (action == SIG_DFL)
1115 				atomic_clearbits_int(&p->p_siglist, mask);
1116 			if (action == SIG_CATCH)
1117 				goto runfast;
1118 			if (p->p_wchan == 0)
1119 				goto run;
1120 			p->p_stat = SSLEEP;
1121 			goto out;
1122 		}
1123 
1124 		if (prop & SA_STOP) {
1125 			/*
1126 			 * Already stopped, don't need to stop again.
1127 			 * (If we did the shell could get confused.)
1128 			 */
1129 			atomic_clearbits_int(&p->p_siglist, mask);
1130 			goto out;
1131 		}
1132 
1133 		/*
1134 		 * If process is sleeping interruptibly, then simulate a
1135 		 * wakeup so that when it is continued, it will be made
1136 		 * runnable and can look at the signal.  But don't make
1137 		 * the process runnable, leave it stopped.
1138 		 */
1139 		if (p->p_wchan && p->p_flag & P_SINTR)
1140 			unsleep(p);
1141 		goto out;
1142 
1143 	case SONPROC:
1144 		signotify(p);
1145 		/* FALLTHROUGH */
1146 	default:
1147 		/*
1148 		 * SRUN, SIDL, SDEAD do nothing with the signal,
1149 		 * other than kicking ourselves if we are running.
1150 		 * It will either never be noticed, or noticed very soon.
1151 		 */
1152 		goto out;
1153 	}
1154 	/*NOTREACHED*/
1155 
1156 runfast:
1157 	/*
1158 	 * Raise priority to at least PUSER.
1159 	 */
1160 	if (p->p_priority > PUSER)
1161 		p->p_priority = PUSER;
1162 run:
1163 	setrunnable(p);
1164 out:
1165 	SCHED_UNLOCK(s);
1166 	if (wakeparent)
1167 		wakeup(pr->ps_pptr);
1168 }
1169 
1170 /*
1171  * If the current process has received a signal (should be caught or cause
1172  * termination, should interrupt current syscall), return the signal number.
1173  * Stop signals with default action are processed immediately, then cleared;
1174  * they aren't returned.  This is checked after each entry to the system for
1175  * a syscall or trap (though this can usually be done without calling issignal
1176  * by checking the pending signal masks in the CURSIG macro.) The normal call
1177  * 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 issignal(struct proc *p)
1187 {
1188 	struct process *pr = p->p_p;
1189 	int signum, mask, prop;
1190 	int dolock = (p->p_flag & P_SINTR) == 0;
1191 	int s;
1192 
1193 	for (;;) {
1194 		mask = p->p_siglist & ~p->p_sigmask;
1195 		if (pr->ps_flags & PS_PPWAIT)
1196 			mask &= ~stopsigmask;
1197 		if (mask == 0)	 	/* no signal to send */
1198 			return (0);
1199 		signum = ffs((long)mask);
1200 		mask = sigmask(signum);
1201 		atomic_clearbits_int(&p->p_siglist, mask);
1202 
1203 		/*
1204 		 * We should see pending but ignored signals
1205 		 * only if PS_TRACED was on when they were posted.
1206 		 */
1207 		if (mask & pr->ps_sigacts->ps_sigignore &&
1208 		    (pr->ps_flags & PS_TRACED) == 0)
1209 			continue;
1210 
1211 		/*
1212 		 * If traced, always stop, and stay stopped until released
1213 		 * by the debugger.  If our parent process is waiting for
1214 		 * us, don't hang as we could deadlock.
1215 		 */
1216 		if (((pr->ps_flags & (PS_TRACED | PS_PPWAIT)) == PS_TRACED) &&
1217 		    signum != SIGKILL) {
1218 			p->p_xstat = signum;
1219 
1220 			if (dolock)
1221 				KERNEL_LOCK();
1222 			single_thread_set(p, SINGLE_PTRACE, 0);
1223 			if (dolock)
1224 				KERNEL_UNLOCK();
1225 
1226 			if (dolock)
1227 				SCHED_LOCK(s);
1228 			proc_stop(p, 1);
1229 			if (dolock)
1230 				SCHED_UNLOCK(s);
1231 
1232 			if (dolock)
1233 				KERNEL_LOCK();
1234 			single_thread_clear(p, 0);
1235 			if (dolock)
1236 				KERNEL_UNLOCK();
1237 
1238 			/*
1239 			 * If we are no longer being traced, or the parent
1240 			 * didn't give us a signal, look for more signals.
1241 			 */
1242 			if ((pr->ps_flags & PS_TRACED) == 0 || p->p_xstat == 0)
1243 				continue;
1244 
1245 			/*
1246 			 * If the new signal is being masked, look for other
1247 			 * signals.
1248 			 */
1249 			signum = p->p_xstat;
1250 			mask = sigmask(signum);
1251 			if ((p->p_sigmask & mask) != 0)
1252 				continue;
1253 
1254 			/* take the signal! */
1255 			atomic_clearbits_int(&p->p_siglist, mask);
1256 		}
1257 
1258 		prop = sigprop[signum];
1259 
1260 		/*
1261 		 * Decide whether the signal should be returned.
1262 		 * Return the signal's number, or fall through
1263 		 * to clear it from the pending mask.
1264 		 */
1265 		switch ((long)pr->ps_sigacts->ps_sigact[signum]) {
1266 		case (long)SIG_DFL:
1267 			/*
1268 			 * Don't take default actions on system processes.
1269 			 */
1270 			if (pr->ps_pid <= 1) {
1271 #ifdef DIAGNOSTIC
1272 				/*
1273 				 * Are you sure you want to ignore SIGSEGV
1274 				 * in init? XXX
1275 				 */
1276 				printf("Process (pid %d) got signal"
1277 				    " %d\n", pr->ps_pid, signum);
1278 #endif
1279 				break;		/* == ignore */
1280 			}
1281 			/*
1282 			 * If there is a pending stop signal to process
1283 			 * with default action, stop here,
1284 			 * then clear the signal.  However,
1285 			 * if process is member of an orphaned
1286 			 * process group, ignore tty stop signals.
1287 			 */
1288 			if (prop & SA_STOP) {
1289 				if (pr->ps_flags & PS_TRACED ||
1290 		    		    (pr->ps_pgrp->pg_jobc == 0 &&
1291 				    prop & SA_TTYSTOP))
1292 					break;	/* == ignore */
1293 				p->p_xstat = signum;
1294 				if (dolock)
1295 					SCHED_LOCK(s);
1296 				proc_stop(p, 1);
1297 				if (dolock)
1298 					SCHED_UNLOCK(s);
1299 				break;
1300 			} else if (prop & SA_IGNORE) {
1301 				/*
1302 				 * Except for SIGCONT, shouldn't get here.
1303 				 * Default action is to ignore; drop it.
1304 				 */
1305 				break;		/* == ignore */
1306 			} else
1307 				goto keep;
1308 			/*NOTREACHED*/
1309 		case (long)SIG_IGN:
1310 			/*
1311 			 * Masking above should prevent us ever trying
1312 			 * to take action on an ignored signal other
1313 			 * than SIGCONT, unless process is traced.
1314 			 */
1315 			if ((prop & SA_CONT) == 0 &&
1316 			    (pr->ps_flags & PS_TRACED) == 0)
1317 				printf("issignal\n");
1318 			break;		/* == ignore */
1319 		default:
1320 			/*
1321 			 * This signal has an action, let
1322 			 * postsig() process it.
1323 			 */
1324 			goto keep;
1325 		}
1326 	}
1327 	/* NOTREACHED */
1328 
1329 keep:
1330 	atomic_setbits_int(&p->p_siglist, mask); /*leave the signal for later */
1331 	return (signum);
1332 }
1333 
1334 /*
1335  * Put the argument process into the stopped state and notify the parent
1336  * via wakeup.  Signals are handled elsewhere.  The process must not be
1337  * on the run queue.
1338  */
1339 void
1340 proc_stop(struct proc *p, int sw)
1341 {
1342 	struct process *pr = p->p_p;
1343 	extern void *softclock_si;
1344 
1345 #ifdef MULTIPROCESSOR
1346 	SCHED_ASSERT_LOCKED();
1347 #endif
1348 
1349 	p->p_stat = SSTOP;
1350 	atomic_clearbits_int(&pr->ps_flags, PS_WAITED);
1351 	atomic_setbits_int(&pr->ps_flags, PS_STOPPED);
1352 	atomic_setbits_int(&p->p_flag, P_SUSPSIG);
1353 	if (!timeout_pending(&proc_stop_to)) {
1354 		timeout_add(&proc_stop_to, 0);
1355 		/*
1356 		 * We need this soft interrupt to be handled fast.
1357 		 * Extra calls to softclock don't hurt.
1358 		 */
1359                 softintr_schedule(softclock_si);
1360 	}
1361 	if (sw)
1362 		mi_switch();
1363 }
1364 
1365 /*
1366  * Called from a timeout to send signals to the parents of stopped processes.
1367  * We can't do this in proc_stop because it's called with nasty locks held
1368  * and we would need recursive scheduler lock to deal with that.
1369  */
1370 void
1371 proc_stop_sweep(void *v)
1372 {
1373 	struct process *pr;
1374 
1375 	LIST_FOREACH(pr, &allprocess, ps_list) {
1376 		if ((pr->ps_flags & PS_STOPPED) == 0)
1377 			continue;
1378 		atomic_clearbits_int(&pr->ps_flags, PS_STOPPED);
1379 
1380 		if ((pr->ps_pptr->ps_sigacts->ps_flags & SAS_NOCLDSTOP) == 0)
1381 			prsignal(pr->ps_pptr, SIGCHLD);
1382 		wakeup(pr->ps_pptr);
1383 	}
1384 }
1385 
1386 /*
1387  * Take the action for the specified signal
1388  * from the current set of pending signals.
1389  */
1390 void
1391 postsig(struct proc *p, int signum)
1392 {
1393 	struct process *pr = p->p_p;
1394 	struct sigacts *ps = pr->ps_sigacts;
1395 	sig_t action;
1396 	u_long trapno;
1397 	int mask, returnmask;
1398 	siginfo_t si;
1399 	union sigval sigval;
1400 	int s, code;
1401 
1402 	KASSERT(signum != 0);
1403 	KERNEL_ASSERT_LOCKED();
1404 
1405 	mask = sigmask(signum);
1406 	atomic_clearbits_int(&p->p_siglist, mask);
1407 	action = ps->ps_sigact[signum];
1408 	sigval.sival_ptr = 0;
1409 
1410 	if (p->p_sisig != signum) {
1411 		trapno = 0;
1412 		code = SI_USER;
1413 		sigval.sival_ptr = 0;
1414 	} else {
1415 		trapno = p->p_sitrapno;
1416 		code = p->p_sicode;
1417 		sigval = p->p_sigval;
1418 	}
1419 	initsiginfo(&si, signum, trapno, code, sigval);
1420 
1421 #ifdef KTRACE
1422 	if (KTRPOINT(p, KTR_PSIG)) {
1423 		ktrpsig(p, signum, action, p->p_flag & P_SIGSUSPEND ?
1424 		    p->p_oldmask : p->p_sigmask, code, &si);
1425 	}
1426 #endif
1427 	if (action == SIG_DFL) {
1428 		/*
1429 		 * Default action, where the default is to kill
1430 		 * the process.  (Other cases were ignored above.)
1431 		 */
1432 		sigexit(p, signum);
1433 		/* NOTREACHED */
1434 	} else {
1435 		/*
1436 		 * If we get here, the signal must be caught.
1437 		 */
1438 #ifdef DIAGNOSTIC
1439 		if (action == SIG_IGN || (p->p_sigmask & mask))
1440 			panic("postsig action");
1441 #endif
1442 		/*
1443 		 * Set the new mask value and also defer further
1444 		 * occurrences of this signal.
1445 		 *
1446 		 * Special case: user has done a sigpause.  Here the
1447 		 * current mask is not of interest, but rather the
1448 		 * mask from before the sigpause is what we want
1449 		 * restored after the signal processing is completed.
1450 		 */
1451 #ifdef MULTIPROCESSOR
1452 		s = splsched();
1453 #else
1454 		s = splhigh();
1455 #endif
1456 		if (p->p_flag & P_SIGSUSPEND) {
1457 			atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
1458 			returnmask = p->p_oldmask;
1459 		} else {
1460 			returnmask = p->p_sigmask;
1461 		}
1462 		if (p->p_sisig == signum) {
1463 			p->p_sisig = 0;
1464 			p->p_sitrapno = 0;
1465 			p->p_sicode = SI_USER;
1466 			p->p_sigval.sival_ptr = NULL;
1467 		}
1468 
1469 		sendsig(action, signum, returnmask, &si);
1470 		postsig_done(p, signum, ps);
1471 		splx(s);
1472 	}
1473 }
1474 
1475 /*
1476  * Force the current process to exit with the specified signal, dumping core
1477  * if appropriate.  We bypass the normal tests for masked and caught signals,
1478  * allowing unrecoverable failures to terminate the process without changing
1479  * signal state.  Mark the accounting record with the signal termination.
1480  * If dumping core, save the signal number for the debugger.  Calls exit and
1481  * does not return.
1482  */
1483 void
1484 sigexit(struct proc *p, int signum)
1485 {
1486 	/* Mark process as going away */
1487 	atomic_setbits_int(&p->p_flag, P_WEXIT);
1488 
1489 	p->p_p->ps_acflag |= AXSIG;
1490 	if (sigprop[signum] & SA_CORE) {
1491 		p->p_sisig = signum;
1492 
1493 		/* if there are other threads, pause them */
1494 		if (P_HASSIBLING(p))
1495 			single_thread_set(p, SINGLE_SUSPEND, 0);
1496 
1497 		if (coredump(p) == 0)
1498 			signum |= WCOREFLAG;
1499 	}
1500 	exit1(p, W_EXITCODE(0, signum), EXIT_NORMAL);
1501 	/* NOTREACHED */
1502 }
1503 
1504 int nosuidcoredump = 1;
1505 
1506 struct coredump_iostate {
1507 	struct proc *io_proc;
1508 	struct vnode *io_vp;
1509 	struct ucred *io_cred;
1510 	off_t io_offset;
1511 };
1512 
1513 /*
1514  * Dump core, into a file named "progname.core", unless the process was
1515  * setuid/setgid.
1516  */
1517 int
1518 coredump(struct proc *p)
1519 {
1520 #ifdef SMALL_KERNEL
1521 	return EPERM;
1522 #else
1523 	struct process *pr = p->p_p;
1524 	struct vnode *vp;
1525 	struct ucred *cred = p->p_ucred;
1526 	struct vmspace *vm = p->p_vmspace;
1527 	struct nameidata nd;
1528 	struct vattr vattr;
1529 	struct coredump_iostate	io;
1530 	int error, len, incrash = 0;
1531 	char name[MAXPATHLEN];
1532 	const char *dir = "/var/crash";
1533 
1534 	if (pr->ps_emul->e_coredump == NULL)
1535 		return (EINVAL);
1536 
1537 	pr->ps_flags |= PS_COREDUMP;
1538 
1539 	/*
1540 	 * If the process has inconsistent uids, nosuidcoredump
1541 	 * determines coredump placement policy.
1542 	 */
1543 	if (((pr->ps_flags & PS_SUGID) && (error = suser(p))) ||
1544 	   ((pr->ps_flags & PS_SUGID) && nosuidcoredump)) {
1545 		if (nosuidcoredump == 3 || nosuidcoredump == 2)
1546 			incrash = 1;
1547 		else
1548 			return (EPERM);
1549 	}
1550 
1551 	/* Don't dump if will exceed file size limit. */
1552 	if (USPACE + ptoa(vm->vm_dsize + vm->vm_ssize) >=
1553 	    p->p_rlimit[RLIMIT_CORE].rlim_cur)
1554 		return (EFBIG);
1555 
1556 	if (incrash && nosuidcoredump == 3) {
1557 		/*
1558 		 * If the program directory does not exist, dumps of
1559 		 * that core will silently fail.
1560 		 */
1561 		len = snprintf(name, sizeof(name), "%s/%s/%u.core",
1562 		    dir, pr->ps_comm, pr->ps_pid);
1563 	} else if (incrash && nosuidcoredump == 2)
1564 		len = snprintf(name, sizeof(name), "%s/%s.core",
1565 		    dir, pr->ps_comm);
1566 	else
1567 		len = snprintf(name, sizeof(name), "%s.core", pr->ps_comm);
1568 	if (len >= sizeof(name))
1569 		return (EACCES);
1570 
1571 	/*
1572 	 * Control the UID used to write out.  The normal case uses
1573 	 * the real UID.  If the sugid case is going to write into the
1574 	 * controlled directory, we do so as root.
1575 	 */
1576 	if (incrash == 0) {
1577 		cred = crdup(cred);
1578 		cred->cr_uid = cred->cr_ruid;
1579 		cred->cr_gid = cred->cr_rgid;
1580 	} else {
1581 		if (p->p_fd->fd_rdir) {
1582 			vrele(p->p_fd->fd_rdir);
1583 			p->p_fd->fd_rdir = NULL;
1584 		}
1585 		p->p_ucred = crdup(p->p_ucred);
1586 		crfree(cred);
1587 		cred = p->p_ucred;
1588 		crhold(cred);
1589 		cred->cr_uid = 0;
1590 		cred->cr_gid = 0;
1591 	}
1592 
1593 	NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, p);
1594 
1595 	error = vn_open(&nd, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
1596 
1597 	if (error)
1598 		goto out;
1599 
1600 	/*
1601 	 * Don't dump to non-regular files, files with links, or files
1602 	 * owned by someone else.
1603 	 */
1604 	vp = nd.ni_vp;
1605 	if ((error = VOP_GETATTR(vp, &vattr, cred, p)) != 0) {
1606 		VOP_UNLOCK(vp);
1607 		vn_close(vp, FWRITE, cred, p);
1608 		goto out;
1609 	}
1610 	if (vp->v_type != VREG || vattr.va_nlink != 1 ||
1611 	    vattr.va_mode & ((VREAD | VWRITE) >> 3 | (VREAD | VWRITE) >> 6) ||
1612 	    vattr.va_uid != cred->cr_uid) {
1613 		error = EACCES;
1614 		VOP_UNLOCK(vp);
1615 		vn_close(vp, FWRITE, cred, p);
1616 		goto out;
1617 	}
1618 	VATTR_NULL(&vattr);
1619 	vattr.va_size = 0;
1620 	VOP_SETATTR(vp, &vattr, cred, p);
1621 	pr->ps_acflag |= ACORE;
1622 
1623 	io.io_proc = p;
1624 	io.io_vp = vp;
1625 	io.io_cred = cred;
1626 	io.io_offset = 0;
1627 	VOP_UNLOCK(vp);
1628 	vref(vp);
1629 	error = vn_close(vp, FWRITE, cred, p);
1630 	if (error == 0)
1631 		error = (*pr->ps_emul->e_coredump)(p, &io);
1632 	vrele(vp);
1633 out:
1634 	crfree(cred);
1635 	return (error);
1636 #endif
1637 }
1638 
1639 #ifndef SMALL_KERNEL
1640 int
1641 coredump_write(void *cookie, enum uio_seg segflg, const void *data, size_t len)
1642 {
1643 	struct coredump_iostate *io = cookie;
1644 	off_t coffset = 0;
1645 	size_t csize;
1646 	int chunk, error;
1647 
1648 	csize = len;
1649 	do {
1650 		if (io->io_proc->p_siglist & sigmask(SIGKILL))
1651 			return (EINTR);
1652 
1653 		/* Rest of the loop sleeps with lock held, so... */
1654 		yield();
1655 
1656 		chunk = MIN(csize, MAXPHYS);
1657 		error = vn_rdwr(UIO_WRITE, io->io_vp,
1658 		    (caddr_t)data + coffset, chunk,
1659 		    io->io_offset + coffset, segflg,
1660 		    IO_UNIT, io->io_cred, NULL, io->io_proc);
1661 		if (error) {
1662 			struct process *pr = io->io_proc->p_p;
1663 
1664 			if (error == ENOSPC)
1665 				log(LOG_ERR,
1666 				    "coredump of %s(%d) failed, filesystem full\n",
1667 				    pr->ps_comm, pr->ps_pid);
1668 			else
1669 				log(LOG_ERR,
1670 				    "coredump of %s(%d), write failed: errno %d\n",
1671 				    pr->ps_comm, pr->ps_pid, error);
1672 			return (error);
1673 		}
1674 
1675 		coffset += chunk;
1676 		csize -= chunk;
1677 	} while (csize > 0);
1678 
1679 	io->io_offset += len;
1680 	return (0);
1681 }
1682 
1683 void
1684 coredump_unmap(void *cookie, vaddr_t start, vaddr_t end)
1685 {
1686 	struct coredump_iostate *io = cookie;
1687 
1688 	uvm_unmap(&io->io_proc->p_vmspace->vm_map, start, end);
1689 }
1690 
1691 #endif	/* !SMALL_KERNEL */
1692 
1693 /*
1694  * Nonexistent system call-- signal process (may want to handle it).
1695  * Flag error in case process won't see signal immediately (blocked or ignored).
1696  */
1697 int
1698 sys_nosys(struct proc *p, void *v, register_t *retval)
1699 {
1700 
1701 	ptsignal(p, SIGSYS, STHREAD);
1702 	return (ENOSYS);
1703 }
1704 
1705 int
1706 sys___thrsigdivert(struct proc *p, void *v, register_t *retval)
1707 {
1708 	static int sigwaitsleep;
1709 	struct sys___thrsigdivert_args /* {
1710 		syscallarg(sigset_t) sigmask;
1711 		syscallarg(siginfo_t *) info;
1712 		syscallarg(const struct timespec *) timeout;
1713 	} */ *uap = v;
1714 	struct process *pr = p->p_p;
1715 	sigset_t *m;
1716 	sigset_t mask = SCARG(uap, sigmask) &~ sigcantmask;
1717 	siginfo_t si;
1718 	uint64_t to_ticks = 0;
1719 	int timeinvalid = 0;
1720 	int error = 0;
1721 
1722 	memset(&si, 0, sizeof(si));
1723 
1724 	if (SCARG(uap, timeout) != NULL) {
1725 		struct timespec ts;
1726 		if ((error = copyin(SCARG(uap, timeout), &ts, sizeof(ts))) != 0)
1727 			return (error);
1728 #ifdef KTRACE
1729 		if (KTRPOINT(p, KTR_STRUCT))
1730 			ktrreltimespec(p, &ts);
1731 #endif
1732 		if (ts.tv_nsec < 0 || ts.tv_nsec >= 1000000000)
1733 			timeinvalid = 1;
1734 		else {
1735 			to_ticks = (uint64_t)hz * ts.tv_sec +
1736 			    ts.tv_nsec / (tick * 1000);
1737 			if (to_ticks > INT_MAX)
1738 				to_ticks = INT_MAX;
1739 			if (to_ticks == 0 && ts.tv_nsec)
1740 				to_ticks = 1;
1741 		}
1742 	}
1743 
1744 	dosigsuspend(p, p->p_sigmask &~ mask);
1745 	for (;;) {
1746 		si.si_signo = CURSIG(p);
1747 		if (si.si_signo != 0) {
1748 			sigset_t smask = sigmask(si.si_signo);
1749 			if (smask & mask) {
1750 				if (p->p_siglist & smask)
1751 					m = &p->p_siglist;
1752 				else if (pr->ps_mainproc->p_siglist & smask)
1753 					m = &pr->ps_mainproc->p_siglist;
1754 				else {
1755 					/* signal got eaten by someone else? */
1756 					continue;
1757 				}
1758 				atomic_clearbits_int(m, smask);
1759 				error = 0;
1760 				break;
1761 			}
1762 		}
1763 
1764 		/* per-POSIX, delay this error until after the above */
1765 		if (timeinvalid)
1766 			error = EINVAL;
1767 
1768 		if (SCARG(uap, timeout) != NULL && to_ticks == 0)
1769 			error = EAGAIN;
1770 
1771 		if (error != 0)
1772 			break;
1773 
1774 		error = tsleep(&sigwaitsleep, PPAUSE|PCATCH, "sigwait",
1775 		    (int)to_ticks);
1776 	}
1777 
1778 	if (error == 0) {
1779 		*retval = si.si_signo;
1780 		if (SCARG(uap, info) != NULL)
1781 			error = copyout(&si, SCARG(uap, info), sizeof(si));
1782 	} else if (error == ERESTART && SCARG(uap, timeout) != NULL) {
1783 		/*
1784 		 * Restarting is wrong if there's a timeout, as it'll be
1785 		 * for the same interval again
1786 		 */
1787 		error = EINTR;
1788 	}
1789 
1790 	return (error);
1791 }
1792 
1793 void
1794 initsiginfo(siginfo_t *si, int sig, u_long trapno, int code, union sigval val)
1795 {
1796 	memset(si, 0, sizeof(*si));
1797 
1798 	si->si_signo = sig;
1799 	si->si_code = code;
1800 	if (code == SI_USER) {
1801 		si->si_value = val;
1802 	} else {
1803 		switch (sig) {
1804 		case SIGSEGV:
1805 		case SIGILL:
1806 		case SIGBUS:
1807 		case SIGFPE:
1808 			si->si_addr = val.sival_ptr;
1809 			si->si_trapno = trapno;
1810 			break;
1811 		case SIGXFSZ:
1812 			break;
1813 		}
1814 	}
1815 }
1816 
1817 int
1818 filt_sigattach(struct knote *kn)
1819 {
1820 	struct process *pr = curproc->p_p;
1821 
1822 	if (kn->kn_id >= NSIG)
1823 		return EINVAL;
1824 
1825 	kn->kn_ptr.p_process = pr;
1826 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
1827 
1828 	/* XXX lock the proc here while adding to the list? */
1829 	SLIST_INSERT_HEAD(&pr->ps_klist, kn, kn_selnext);
1830 
1831 	return (0);
1832 }
1833 
1834 void
1835 filt_sigdetach(struct knote *kn)
1836 {
1837 	struct process *pr = kn->kn_ptr.p_process;
1838 
1839 	SLIST_REMOVE(&pr->ps_klist, kn, knote, kn_selnext);
1840 }
1841 
1842 /*
1843  * signal knotes are shared with proc knotes, so we apply a mask to
1844  * the hint in order to differentiate them from process hints.  This
1845  * could be avoided by using a signal-specific knote list, but probably
1846  * isn't worth the trouble.
1847  */
1848 int
1849 filt_signal(struct knote *kn, long hint)
1850 {
1851 
1852 	if (hint & NOTE_SIGNAL) {
1853 		hint &= ~NOTE_SIGNAL;
1854 
1855 		if (kn->kn_id == hint)
1856 			kn->kn_data++;
1857 	}
1858 	return (kn->kn_data != 0);
1859 }
1860 
1861 void
1862 userret(struct proc *p)
1863 {
1864 	int signum;
1865 
1866 	/* send SIGPROF or SIGVTALRM if their timers interrupted this thread */
1867 	if (p->p_flag & P_PROFPEND) {
1868 		atomic_clearbits_int(&p->p_flag, P_PROFPEND);
1869 		KERNEL_LOCK();
1870 		psignal(p, SIGPROF);
1871 		KERNEL_UNLOCK();
1872 	}
1873 	if (p->p_flag & P_ALRMPEND) {
1874 		atomic_clearbits_int(&p->p_flag, P_ALRMPEND);
1875 		KERNEL_LOCK();
1876 		psignal(p, SIGVTALRM);
1877 		KERNEL_UNLOCK();
1878 	}
1879 
1880 	if (SIGPENDING(p)) {
1881 		KERNEL_LOCK();
1882 		while ((signum = CURSIG(p)) != 0)
1883 			postsig(p, signum);
1884 		KERNEL_UNLOCK();
1885 	}
1886 
1887 	/*
1888 	 * If P_SIGSUSPEND is still set here, then we still need to restore
1889 	 * the original sigmask before returning to userspace.  Also, this
1890 	 * might unmask some pending signals, so we need to check a second
1891 	 * time for signals to post.
1892 	 */
1893 	if (p->p_flag & P_SIGSUSPEND) {
1894 		atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
1895 		p->p_sigmask = p->p_oldmask;
1896 
1897 		KERNEL_LOCK();
1898 		while ((signum = CURSIG(p)) != 0)
1899 			postsig(p, signum);
1900 		KERNEL_UNLOCK();
1901 	}
1902 
1903 	if (p->p_flag & P_SUSPSINGLE) {
1904 		KERNEL_LOCK();
1905 		single_thread_check(p, 0);
1906 		KERNEL_UNLOCK();
1907 	}
1908 
1909 	WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
1910 
1911 	p->p_cpu->ci_schedstate.spc_curpriority = p->p_priority = p->p_usrpri;
1912 }
1913 
1914 int
1915 single_thread_check(struct proc *p, int deep)
1916 {
1917 	struct process *pr = p->p_p;
1918 
1919 	if (pr->ps_single != NULL && pr->ps_single != p) {
1920 		do {
1921 			int s;
1922 
1923 			/* if we're in deep, we need to unwind to the edge */
1924 			if (deep) {
1925 				if (pr->ps_flags & PS_SINGLEUNWIND)
1926 					return (ERESTART);
1927 				if (pr->ps_flags & PS_SINGLEEXIT)
1928 					return (EINTR);
1929 			}
1930 
1931 			if (--pr->ps_singlecount == 0)
1932 				wakeup(&pr->ps_singlecount);
1933 			if (pr->ps_flags & PS_SINGLEEXIT)
1934 				exit1(p, 0, EXIT_THREAD_NOCHECK);
1935 
1936 			/* not exiting and don't need to unwind, so suspend */
1937 			SCHED_LOCK(s);
1938 			p->p_stat = SSTOP;
1939 			mi_switch();
1940 			SCHED_UNLOCK(s);
1941 		} while (pr->ps_single != NULL);
1942 	}
1943 
1944 	return (0);
1945 }
1946 
1947 /*
1948  * Stop other threads in the process.  The mode controls how and
1949  * where the other threads should stop:
1950  *  - SINGLE_SUSPEND: stop wherever they are, will later either be told to exit
1951  *    (by setting to SINGLE_EXIT) or be released (via single_thread_clear())
1952  *  - SINGLE_PTRACE: stop wherever they are, will wait for them to stop
1953  *    later (via single_thread_wait()) and released as with SINGLE_SUSPEND
1954  *  - SINGLE_UNWIND: just unwind to kernel boundary, will be told to exit
1955  *    or released as with SINGLE_SUSPEND
1956  *  - SINGLE_EXIT: unwind to kernel boundary and exit
1957  */
1958 int
1959 single_thread_set(struct proc *p, enum single_thread_mode mode, int deep)
1960 {
1961 	struct process *pr = p->p_p;
1962 	struct proc *q;
1963 	int error;
1964 
1965 	KERNEL_ASSERT_LOCKED();
1966 
1967 	if ((error = single_thread_check(p, deep)))
1968 		return error;
1969 
1970 	switch (mode) {
1971 	case SINGLE_SUSPEND:
1972 	case SINGLE_PTRACE:
1973 		break;
1974 	case SINGLE_UNWIND:
1975 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
1976 		break;
1977 	case SINGLE_EXIT:
1978 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEEXIT);
1979 		atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
1980 		break;
1981 #ifdef DIAGNOSTIC
1982 	default:
1983 		panic("single_thread_mode = %d", mode);
1984 #endif
1985 	}
1986 	pr->ps_single = p;
1987 	pr->ps_singlecount = 0;
1988 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
1989 		int s;
1990 
1991 		if (q == p)
1992 			continue;
1993 		if (q->p_flag & P_WEXIT) {
1994 			if (mode == SINGLE_EXIT) {
1995 				SCHED_LOCK(s);
1996 				if (q->p_stat == SSTOP) {
1997 					setrunnable(q);
1998 					pr->ps_singlecount++;
1999 				}
2000 				SCHED_UNLOCK(s);
2001 			}
2002 			continue;
2003 		}
2004 		SCHED_LOCK(s);
2005 		atomic_setbits_int(&q->p_flag, P_SUSPSINGLE);
2006 		switch (q->p_stat) {
2007 		case SIDL:
2008 		case SRUN:
2009 			pr->ps_singlecount++;
2010 			break;
2011 		case SSLEEP:
2012 			/* if it's not interruptible, then just have to wait */
2013 			if (q->p_flag & P_SINTR) {
2014 				/* merely need to suspend?  just stop it */
2015 				if (mode == SINGLE_SUSPEND ||
2016 				    mode == SINGLE_PTRACE) {
2017 					q->p_stat = SSTOP;
2018 					break;
2019 				}
2020 				/* need to unwind or exit, so wake it */
2021 				setrunnable(q);
2022 			}
2023 			pr->ps_singlecount++;
2024 			break;
2025 		case SSTOP:
2026 			if (mode == SINGLE_EXIT) {
2027 				setrunnable(q);
2028 				pr->ps_singlecount++;
2029 			}
2030 			break;
2031 		case SDEAD:
2032 			break;
2033 		case SONPROC:
2034 			pr->ps_singlecount++;
2035 			signotify(q);
2036 			break;
2037 		}
2038 		SCHED_UNLOCK(s);
2039 	}
2040 
2041 	if (mode != SINGLE_PTRACE)
2042 		single_thread_wait(pr);
2043 
2044 	return 0;
2045 }
2046 
2047 void
2048 single_thread_wait(struct process *pr)
2049 {
2050 	/* wait until they're all suspended */
2051 	while (pr->ps_singlecount > 0)
2052 		tsleep(&pr->ps_singlecount, PUSER, "suspend", 0);
2053 }
2054 
2055 void
2056 single_thread_clear(struct proc *p, int flag)
2057 {
2058 	struct process *pr = p->p_p;
2059 	struct proc *q;
2060 
2061 	KASSERT(pr->ps_single == p);
2062 	KERNEL_ASSERT_LOCKED();
2063 
2064 	pr->ps_single = NULL;
2065 	atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND | PS_SINGLEEXIT);
2066 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
2067 		int s;
2068 
2069 		if (q == p || (q->p_flag & P_SUSPSINGLE) == 0)
2070 			continue;
2071 		atomic_clearbits_int(&q->p_flag, P_SUSPSINGLE);
2072 
2073 		/*
2074 		 * if the thread was only stopped for single threading
2075 		 * then clearing that either makes it runnable or puts
2076 		 * it back into some sleep queue
2077 		 */
2078 		SCHED_LOCK(s);
2079 		if (q->p_stat == SSTOP && (q->p_flag & flag) == 0) {
2080 			if (q->p_wchan == 0)
2081 				setrunnable(q);
2082 			else
2083 				q->p_stat = SSLEEP;
2084 		}
2085 		SCHED_UNLOCK(s);
2086 	}
2087 }
2088 
2089 void
2090 sigio_del(struct sigiolst *rmlist)
2091 {
2092 	struct sigio *sigio;
2093 
2094 	while ((sigio = LIST_FIRST(rmlist)) != NULL) {
2095 		LIST_REMOVE(sigio, sio_pgsigio);
2096 		crfree(sigio->sio_ucred);
2097 		free(sigio, M_SIGIO, sizeof(*sigio));
2098 	}
2099 }
2100 
2101 void
2102 sigio_unlink(struct sigio_ref *sir, struct sigiolst *rmlist)
2103 {
2104 	struct sigio *sigio;
2105 
2106 	MUTEX_ASSERT_LOCKED(&sigio_lock);
2107 
2108 	sigio = sir->sir_sigio;
2109 	if (sigio != NULL) {
2110 		KASSERT(sigio->sio_myref == sir);
2111 		sir->sir_sigio = NULL;
2112 
2113 		if (sigio->sio_pgid > 0)
2114 			sigio->sio_proc = NULL;
2115 		else
2116 			sigio->sio_pgrp = NULL;
2117 		LIST_REMOVE(sigio, sio_pgsigio);
2118 
2119 		LIST_INSERT_HEAD(rmlist, sigio, sio_pgsigio);
2120 	}
2121 }
2122 
2123 void
2124 sigio_free(struct sigio_ref *sir)
2125 {
2126 	struct sigiolst rmlist;
2127 
2128 	if (sir->sir_sigio == NULL)
2129 		return;
2130 
2131 	LIST_INIT(&rmlist);
2132 
2133 	mtx_enter(&sigio_lock);
2134 	sigio_unlink(sir, &rmlist);
2135 	mtx_leave(&sigio_lock);
2136 
2137 	sigio_del(&rmlist);
2138 }
2139 
2140 void
2141 sigio_freelist(struct sigiolst *sigiolst)
2142 {
2143 	struct sigiolst rmlist;
2144 	struct sigio *sigio;
2145 
2146 	if (LIST_EMPTY(sigiolst))
2147 		return;
2148 
2149 	LIST_INIT(&rmlist);
2150 
2151 	mtx_enter(&sigio_lock);
2152 	while ((sigio = LIST_FIRST(sigiolst)) != NULL)
2153 		sigio_unlink(sigio->sio_myref, &rmlist);
2154 	mtx_leave(&sigio_lock);
2155 
2156 	sigio_del(&rmlist);
2157 }
2158 
2159 int
2160 sigio_setown(struct sigio_ref *sir, pid_t pgid)
2161 {
2162 	struct sigiolst rmlist;
2163 	struct proc *p = curproc;
2164 	struct pgrp *pgrp = NULL;
2165 	struct process *pr = NULL;
2166 	struct sigio *sigio;
2167 	int error;
2168 
2169 	if (pgid == 0) {
2170 		sigio_free(sir);
2171 		return (0);
2172 	}
2173 
2174 	sigio = malloc(sizeof(*sigio), M_SIGIO, M_WAITOK);
2175 	sigio->sio_pgid = pgid;
2176 	sigio->sio_ucred = crhold(p->p_ucred);
2177 	sigio->sio_myref = sir;
2178 
2179 	LIST_INIT(&rmlist);
2180 
2181 	/*
2182 	 * The kernel lock, and not sleeping between prfind()/pgfind() and
2183 	 * linking of the sigio ensure that the process or process group does
2184 	 * not disappear unexpectedly.
2185 	 */
2186 	KERNEL_LOCK();
2187 	mtx_enter(&sigio_lock);
2188 
2189 	if (pgid > 0) {
2190 		pr = prfind(pgid);
2191 		if (pr == NULL) {
2192 			error = ESRCH;
2193 			goto fail;
2194 		}
2195 
2196 		/*
2197 		 * Policy - Don't allow a process to FSETOWN a process
2198 		 * in another session.
2199 		 *
2200 		 * Remove this test to allow maximum flexibility or
2201 		 * restrict FSETOWN to the current process or process
2202 		 * group for maximum safety.
2203 		 */
2204 		if (pr->ps_session != p->p_p->ps_session) {
2205 			error = EPERM;
2206 			goto fail;
2207 		}
2208 
2209 		if ((pr->ps_flags & PS_EXITING) != 0) {
2210 			error = ESRCH;
2211 			goto fail;
2212 		}
2213 	} else /* if (pgid < 0) */ {
2214 		pgrp = pgfind(-pgid);
2215 		if (pgrp == NULL) {
2216 			error = ESRCH;
2217 			goto fail;
2218 		}
2219 
2220 		/*
2221 		 * Policy - Don't allow a process to FSETOWN a process
2222 		 * in another session.
2223 		 *
2224 		 * Remove this test to allow maximum flexibility or
2225 		 * restrict FSETOWN to the current process or process
2226 		 * group for maximum safety.
2227 		 */
2228 		if (pgrp->pg_session != p->p_p->ps_session) {
2229 			error = EPERM;
2230 			goto fail;
2231 		}
2232 	}
2233 
2234 	if (pgid > 0) {
2235 		sigio->sio_proc = pr;
2236 		LIST_INSERT_HEAD(&pr->ps_sigiolst, sigio, sio_pgsigio);
2237 	} else {
2238 		sigio->sio_pgrp = pgrp;
2239 		LIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
2240 	}
2241 
2242 	sigio_unlink(sir, &rmlist);
2243 	sir->sir_sigio = sigio;
2244 
2245 	mtx_leave(&sigio_lock);
2246 	KERNEL_UNLOCK();
2247 
2248 	sigio_del(&rmlist);
2249 
2250 	return (0);
2251 
2252 fail:
2253 	mtx_leave(&sigio_lock);
2254 	KERNEL_UNLOCK();
2255 
2256 	crfree(sigio->sio_ucred);
2257 	free(sigio, M_SIGIO, sizeof(*sigio));
2258 
2259 	return (error);
2260 }
2261 
2262 pid_t
2263 sigio_getown(struct sigio_ref *sir)
2264 {
2265 	struct sigio *sigio;
2266 	pid_t pgid = 0;
2267 
2268 	mtx_enter(&sigio_lock);
2269 	sigio = sir->sir_sigio;
2270 	if (sigio != NULL)
2271 		pgid = sigio->sio_pgid;
2272 	mtx_leave(&sigio_lock);
2273 
2274 	return (pgid);
2275 }
2276 
2277 void
2278 sigio_copy(struct sigio_ref *dst, struct sigio_ref *src)
2279 {
2280 	struct sigiolst rmlist;
2281 	struct sigio *newsigio, *sigio;
2282 
2283 	sigio_free(dst);
2284 
2285 	if (src->sir_sigio == NULL)
2286 		return;
2287 
2288 	newsigio = malloc(sizeof(*newsigio), M_SIGIO, M_WAITOK);
2289 	LIST_INIT(&rmlist);
2290 
2291 	mtx_enter(&sigio_lock);
2292 
2293 	sigio = src->sir_sigio;
2294 	if (sigio == NULL) {
2295 		mtx_leave(&sigio_lock);
2296 		free(newsigio, M_SIGIO, sizeof(*newsigio));
2297 		return;
2298 	}
2299 
2300 	newsigio->sio_pgid = sigio->sio_pgid;
2301 	newsigio->sio_ucred = crhold(sigio->sio_ucred);
2302 	newsigio->sio_myref = dst;
2303 	if (newsigio->sio_pgid > 0) {
2304 		newsigio->sio_proc = sigio->sio_proc;
2305 		LIST_INSERT_HEAD(&newsigio->sio_proc->ps_sigiolst, newsigio,
2306 		    sio_pgsigio);
2307 	} else {
2308 		newsigio->sio_pgrp = sigio->sio_pgrp;
2309 		LIST_INSERT_HEAD(&newsigio->sio_pgrp->pg_sigiolst, newsigio,
2310 		    sio_pgsigio);
2311 	}
2312 
2313 	sigio_unlink(dst, &rmlist);
2314 	dst->sir_sigio = newsigio;
2315 
2316 	mtx_leave(&sigio_lock);
2317 
2318 	sigio_del(&rmlist);
2319 }
2320