xref: /openbsd-src/sys/kern/kern_sig.c (revision dad4761d438f9efd07e22bccd576f478cf5eed13)
1 /*	$OpenBSD: kern_sig.c,v 1.226 2018/12/17 14:51:57 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.  If checktty is 1,
737  * limit to members which have a controlling terminal.
738  */
739 void
740 pgsignal(struct pgrp *pgrp, int signum, int checkctty)
741 {
742 	struct process *pr;
743 
744 	if (pgrp)
745 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist)
746 			if (checkctty == 0 || pr->ps_flags & PS_CONTROLT)
747 				prsignal(pr, signum);
748 }
749 
750 /*
751  * Send a SIGIO or SIGURG signal to a process or process group using stored
752  * credentials rather than those of the current process.
753  */
754 void
755 pgsigio(struct sigio_ref *sir, int sig, int checkctty)
756 {
757 	struct process *pr;
758 	struct sigio *sigio;
759 
760 	if (sir->sir_sigio == NULL)
761 		return;
762 
763 	mtx_enter(&sigio_lock);
764 	sigio = sir->sir_sigio;
765 	if (sigio == NULL)
766 		goto out;
767 	if (sigio->sio_pgid > 0) {
768 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc))
769 			prsignal(sigio->sio_proc, sig);
770 	} else if (sigio->sio_pgid < 0) {
771 		LIST_FOREACH(pr, &sigio->sio_pgrp->pg_members, ps_pglist) {
772 			if (CANSIGIO(sigio->sio_ucred, pr) &&
773 			    (checkctty == 0 || (pr->ps_flags & PS_CONTROLT)))
774 				prsignal(pr, sig);
775 		}
776 	}
777 out:
778 	mtx_leave(&sigio_lock);
779 }
780 
781 /*
782  * Recalculate the signal mask and reset the signal disposition after
783  * usermode frame for delivery is formed.
784  */
785 void
786 postsig_done(struct proc *p, int signum, struct sigacts *ps)
787 {
788 	int mask = sigmask(signum);
789 
790 	KERNEL_ASSERT_LOCKED();
791 
792 	p->p_ru.ru_nsignals++;
793 	atomic_setbits_int(&p->p_sigmask, ps->ps_catchmask[signum]);
794 	if ((ps->ps_sigreset & mask) != 0) {
795 		ps->ps_sigcatch &= ~mask;
796 		if (signum != SIGCONT && sigprop[signum] & SA_IGNORE)
797 			ps->ps_sigignore |= mask;
798 		ps->ps_sigact[signum] = SIG_DFL;
799 	}
800 }
801 
802 /*
803  * Send a signal caused by a trap to the current thread
804  * If it will be caught immediately, deliver it with correct code.
805  * Otherwise, post it normally.
806  */
807 void
808 trapsignal(struct proc *p, int signum, u_long trapno, int code,
809     union sigval sigval)
810 {
811 	struct process *pr = p->p_p;
812 	struct sigacts *ps = pr->ps_sigacts;
813 	int mask;
814 
815 	switch (signum) {
816 	case SIGILL:
817 	case SIGBUS:
818 	case SIGSEGV:
819 		pr->ps_acflag |= ATRAP;
820 		break;
821 	}
822 
823 	mask = sigmask(signum);
824 	if ((pr->ps_flags & PS_TRACED) == 0 &&
825 	    (ps->ps_sigcatch & mask) != 0 &&
826 	    (p->p_sigmask & mask) == 0) {
827 		siginfo_t si;
828 		initsiginfo(&si, signum, trapno, code, sigval);
829 #ifdef KTRACE
830 		if (KTRPOINT(p, KTR_PSIG)) {
831 			ktrpsig(p, signum, ps->ps_sigact[signum],
832 			    p->p_sigmask, code, &si);
833 		}
834 #endif
835 		sendsig(ps->ps_sigact[signum], signum, p->p_sigmask, &si);
836 		postsig_done(p, signum, ps);
837 	} else {
838 		p->p_sisig = signum;
839 		p->p_sitrapno = trapno;	/* XXX for core dump/debugger */
840 		p->p_sicode = code;
841 		p->p_sigval = sigval;
842 
843 		/*
844 		 * Signals like SIGBUS and SIGSEGV should not, when
845 		 * generated by the kernel, be ignorable or blockable.
846 		 * If it is and we're not being traced, then just kill
847 		 * the process.
848 		 */
849 		if ((pr->ps_flags & PS_TRACED) == 0 &&
850 		    (sigprop[signum] & SA_KILL) &&
851 		    ((p->p_sigmask & mask) || (ps->ps_sigignore & mask)))
852 			sigexit(p, signum);
853 		ptsignal(p, signum, STHREAD);
854 	}
855 }
856 
857 /*
858  * Send the signal to the process.  If the signal has an action, the action
859  * is usually performed by the target process rather than the caller; we add
860  * the signal to the set of pending signals for the process.
861  *
862  * Exceptions:
863  *   o When a stop signal is sent to a sleeping process that takes the
864  *     default action, the process is stopped without awakening it.
865  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
866  *     regardless of the signal action (eg, blocked or ignored).
867  *
868  * Other ignored signals are discarded immediately.
869  */
870 void
871 psignal(struct proc *p, int signum)
872 {
873 	ptsignal(p, signum, SPROCESS);
874 }
875 
876 /*
877  * type = SPROCESS	process signal, can be diverted (sigwait())
878  *	XXX if blocked in all threads, mark as pending in struct process
879  * type = STHREAD	thread signal, but should be propagated if unhandled
880  * type = SPROPAGATED	propagated to this thread, so don't propagate again
881  */
882 void
883 ptsignal(struct proc *p, int signum, enum signal_type type)
884 {
885 	int s, prop;
886 	sig_t action;
887 	int mask;
888 	struct process *pr = p->p_p;
889 	struct proc *q;
890 	int wakeparent = 0;
891 
892 #ifdef DIAGNOSTIC
893 	if ((u_int)signum >= NSIG || signum == 0)
894 		panic("psignal signal number");
895 #endif
896 
897 	/* Ignore signal if the target process is exiting */
898 	if (pr->ps_flags & PS_EXITING)
899 		return;
900 
901 	mask = sigmask(signum);
902 
903 	if (type == SPROCESS) {
904 		/* Accept SIGKILL to coredumping processes */
905 		if (pr->ps_flags & PS_COREDUMP && signum == SIGKILL) {
906 			if (pr->ps_single != NULL)
907 				p = pr->ps_single;
908 			atomic_setbits_int(&p->p_siglist, mask);
909 			return;
910 		}
911 
912 		/*
913 		 * If the current thread can process the signal
914 		 * immediately (it's unblocked) then have it take it.
915 		 */
916 		q = curproc;
917 		if (q != NULL && q->p_p == pr && (q->p_flag & P_WEXIT) == 0 &&
918 		    (q->p_sigmask & mask) == 0)
919 			p = q;
920 		else {
921 			/*
922 			 * A process-wide signal can be diverted to a
923 			 * different thread that's in sigwait() for this
924 			 * signal.  If there isn't such a thread, then
925 			 * pick a thread that doesn't have it blocked so
926 			 * that the stop/kill consideration isn't
927 			 * delayed.  Otherwise, mark it pending on the
928 			 * main thread.
929 			 */
930 			TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
931 				/* ignore exiting threads */
932 				if (q->p_flag & P_WEXIT)
933 					continue;
934 
935 				/* skip threads that have the signal blocked */
936 				if ((q->p_sigmask & mask) != 0)
937 					continue;
938 
939 				/* okay, could send to this thread */
940 				p = q;
941 
942 				/*
943 				 * sigsuspend, sigwait, ppoll/pselect, etc?
944 				 * Definitely go to this thread, as it's
945 				 * already blocked in the kernel.
946 				 */
947 				if (q->p_flag & P_SIGSUSPEND)
948 					break;
949 			}
950 		}
951 	}
952 
953 	if (type != SPROPAGATED)
954 		KNOTE(&pr->ps_klist, NOTE_SIGNAL | signum);
955 
956 	prop = sigprop[signum];
957 
958 	/*
959 	 * If proc is traced, always give parent a chance.
960 	 */
961 	if (pr->ps_flags & PS_TRACED) {
962 		action = SIG_DFL;
963 		atomic_setbits_int(&p->p_siglist, mask);
964 	} else {
965 		/*
966 		 * If the signal is being ignored,
967 		 * then we forget about it immediately.
968 		 * (Note: we don't set SIGCONT in ps_sigignore,
969 		 * and if it is set to SIG_IGN,
970 		 * action will be SIG_DFL here.)
971 		 */
972 		if (pr->ps_sigacts->ps_sigignore & mask)
973 			return;
974 		if (p->p_sigmask & mask) {
975 			action = SIG_HOLD;
976 		} else if (pr->ps_sigacts->ps_sigcatch & mask) {
977 			action = SIG_CATCH;
978 		} else {
979 			action = SIG_DFL;
980 
981 			if (prop & SA_KILL && pr->ps_nice > NZERO)
982 				 pr->ps_nice = NZERO;
983 
984 			/*
985 			 * If sending a tty stop signal to a member of an
986 			 * orphaned process group, discard the signal here if
987 			 * the action is default; don't stop the process below
988 			 * if sleeping, and don't clear any pending SIGCONT.
989 			 */
990 			if (prop & SA_TTYSTOP && pr->ps_pgrp->pg_jobc == 0)
991 				return;
992 		}
993 
994 		atomic_setbits_int(&p->p_siglist, mask);
995 	}
996 
997 	if (prop & SA_CONT)
998 		atomic_clearbits_int(&p->p_siglist, stopsigmask);
999 
1000 	if (prop & SA_STOP) {
1001 		atomic_clearbits_int(&p->p_siglist, contsigmask);
1002 		atomic_clearbits_int(&p->p_flag, P_CONTINUED);
1003 	}
1004 
1005 	/*
1006 	 * XXX delay processing of SA_STOP signals unless action == SIG_DFL?
1007 	 */
1008 	if (prop & (SA_CONT | SA_STOP) && type != SPROPAGATED)
1009 		TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link)
1010 			if (q != p)
1011 				ptsignal(q, signum, SPROPAGATED);
1012 
1013 	/*
1014 	 * Defer further processing for signals which are held,
1015 	 * except that stopped processes must be continued by SIGCONT.
1016 	 */
1017 	if (action == SIG_HOLD && ((prop & SA_CONT) == 0 || p->p_stat != SSTOP))
1018 		return;
1019 
1020 	SCHED_LOCK(s);
1021 
1022 	switch (p->p_stat) {
1023 
1024 	case SSLEEP:
1025 		/*
1026 		 * If process is sleeping uninterruptibly
1027 		 * we can't interrupt the sleep... the signal will
1028 		 * be noticed when the process returns through
1029 		 * trap() or syscall().
1030 		 */
1031 		if ((p->p_flag & P_SINTR) == 0)
1032 			goto out;
1033 		/*
1034 		 * Process is sleeping and traced... make it runnable
1035 		 * so it can discover the signal in issignal() and stop
1036 		 * for the parent.
1037 		 */
1038 		if (pr->ps_flags & PS_TRACED)
1039 			goto run;
1040 		/*
1041 		 * If SIGCONT is default (or ignored) and process is
1042 		 * asleep, we are finished; the process should not
1043 		 * be awakened.
1044 		 */
1045 		if ((prop & SA_CONT) && action == SIG_DFL) {
1046 			atomic_clearbits_int(&p->p_siglist, mask);
1047 			goto out;
1048 		}
1049 		/*
1050 		 * When a sleeping process receives a stop
1051 		 * signal, process immediately if possible.
1052 		 */
1053 		if ((prop & SA_STOP) && action == SIG_DFL) {
1054 			/*
1055 			 * If a child holding parent blocked,
1056 			 * stopping could cause deadlock.
1057 			 */
1058 			if (pr->ps_flags & PS_PPWAIT)
1059 				goto out;
1060 			atomic_clearbits_int(&p->p_siglist, mask);
1061 			p->p_xstat = signum;
1062 			proc_stop(p, 0);
1063 			goto out;
1064 		}
1065 		/*
1066 		 * All other (caught or default) signals
1067 		 * cause the process to run.
1068 		 */
1069 		goto runfast;
1070 		/*NOTREACHED*/
1071 
1072 	case SSTOP:
1073 		/*
1074 		 * If traced process is already stopped,
1075 		 * then no further action is necessary.
1076 		 */
1077 		if (pr->ps_flags & PS_TRACED)
1078 			goto out;
1079 
1080 		/*
1081 		 * Kill signal always sets processes running.
1082 		 */
1083 		if (signum == SIGKILL) {
1084 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1085 			goto runfast;
1086 		}
1087 
1088 		if (prop & SA_CONT) {
1089 			/*
1090 			 * If SIGCONT is default (or ignored), we continue the
1091 			 * process but don't leave the signal in p_siglist, as
1092 			 * it has no further action.  If SIGCONT is held, we
1093 			 * continue the process and leave the signal in
1094 			 * p_siglist.  If the process catches SIGCONT, let it
1095 			 * handle the signal itself.  If it isn't waiting on
1096 			 * an event, then it goes back to run state.
1097 			 * Otherwise, process goes back to sleep state.
1098 			 */
1099 			atomic_setbits_int(&p->p_flag, P_CONTINUED);
1100 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1101 			wakeparent = 1;
1102 			if (action == SIG_DFL)
1103 				atomic_clearbits_int(&p->p_siglist, mask);
1104 			if (action == SIG_CATCH)
1105 				goto runfast;
1106 			if (p->p_wchan == 0)
1107 				goto run;
1108 			p->p_stat = SSLEEP;
1109 			goto out;
1110 		}
1111 
1112 		if (prop & SA_STOP) {
1113 			/*
1114 			 * Already stopped, don't need to stop again.
1115 			 * (If we did the shell could get confused.)
1116 			 */
1117 			atomic_clearbits_int(&p->p_siglist, mask);
1118 			goto out;
1119 		}
1120 
1121 		/*
1122 		 * If process is sleeping interruptibly, then simulate a
1123 		 * wakeup so that when it is continued, it will be made
1124 		 * runnable and can look at the signal.  But don't make
1125 		 * the process runnable, leave it stopped.
1126 		 */
1127 		if (p->p_wchan && p->p_flag & P_SINTR)
1128 			unsleep(p);
1129 		goto out;
1130 
1131 	case SONPROC:
1132 		signotify(p);
1133 		/* FALLTHROUGH */
1134 	default:
1135 		/*
1136 		 * SRUN, SIDL, SDEAD do nothing with the signal,
1137 		 * other than kicking ourselves if we are running.
1138 		 * It will either never be noticed, or noticed very soon.
1139 		 */
1140 		goto out;
1141 	}
1142 	/*NOTREACHED*/
1143 
1144 runfast:
1145 	/*
1146 	 * Raise priority to at least PUSER.
1147 	 */
1148 	if (p->p_priority > PUSER)
1149 		p->p_priority = PUSER;
1150 run:
1151 	setrunnable(p);
1152 out:
1153 	SCHED_UNLOCK(s);
1154 	if (wakeparent)
1155 		wakeup(pr->ps_pptr);
1156 }
1157 
1158 /*
1159  * If the current process has received a signal (should be caught or cause
1160  * termination, should interrupt current syscall), return the signal number.
1161  * Stop signals with default action are processed immediately, then cleared;
1162  * they aren't returned.  This is checked after each entry to the system for
1163  * a syscall or trap (though this can usually be done without calling issignal
1164  * by checking the pending signal masks in the CURSIG macro.) The normal call
1165  * sequence is
1166  *
1167  *	while (signum = CURSIG(curproc))
1168  *		postsig(signum);
1169  *
1170  * Assumes that if the P_SINTR flag is set, we're holding both the
1171  * kernel and scheduler locks.
1172  */
1173 int
1174 issignal(struct proc *p)
1175 {
1176 	struct process *pr = p->p_p;
1177 	int signum, mask, prop;
1178 	int dolock = (p->p_flag & P_SINTR) == 0;
1179 	int s;
1180 
1181 	for (;;) {
1182 		mask = p->p_siglist & ~p->p_sigmask;
1183 		if (pr->ps_flags & PS_PPWAIT)
1184 			mask &= ~stopsigmask;
1185 		if (mask == 0)	 	/* no signal to send */
1186 			return (0);
1187 		signum = ffs((long)mask);
1188 		mask = sigmask(signum);
1189 		atomic_clearbits_int(&p->p_siglist, mask);
1190 
1191 		/*
1192 		 * We should see pending but ignored signals
1193 		 * only if PS_TRACED was on when they were posted.
1194 		 */
1195 		if (mask & pr->ps_sigacts->ps_sigignore &&
1196 		    (pr->ps_flags & PS_TRACED) == 0)
1197 			continue;
1198 
1199 		/*
1200 		 * If traced, always stop, and stay stopped until released
1201 		 * by the debugger.  If our parent process is waiting for
1202 		 * us, don't hang as we could deadlock.
1203 		 */
1204 		if (((pr->ps_flags & (PS_TRACED | PS_PPWAIT)) == PS_TRACED) &&
1205 		    signum != SIGKILL) {
1206 			p->p_xstat = signum;
1207 
1208 			if (dolock)
1209 				KERNEL_LOCK();
1210 			single_thread_set(p, SINGLE_PTRACE, 0);
1211 			if (dolock)
1212 				KERNEL_UNLOCK();
1213 
1214 			if (dolock)
1215 				SCHED_LOCK(s);
1216 			proc_stop(p, 1);
1217 			if (dolock)
1218 				SCHED_UNLOCK(s);
1219 
1220 			if (dolock)
1221 				KERNEL_LOCK();
1222 			single_thread_clear(p, 0);
1223 			if (dolock)
1224 				KERNEL_UNLOCK();
1225 
1226 			/*
1227 			 * If we are no longer being traced, or the parent
1228 			 * didn't give us a signal, look for more signals.
1229 			 */
1230 			if ((pr->ps_flags & PS_TRACED) == 0 || p->p_xstat == 0)
1231 				continue;
1232 
1233 			/*
1234 			 * If the new signal is being masked, look for other
1235 			 * signals.
1236 			 */
1237 			signum = p->p_xstat;
1238 			mask = sigmask(signum);
1239 			if ((p->p_sigmask & mask) != 0)
1240 				continue;
1241 
1242 			/* take the signal! */
1243 			atomic_clearbits_int(&p->p_siglist, mask);
1244 		}
1245 
1246 		prop = sigprop[signum];
1247 
1248 		/*
1249 		 * Decide whether the signal should be returned.
1250 		 * Return the signal's number, or fall through
1251 		 * to clear it from the pending mask.
1252 		 */
1253 		switch ((long)pr->ps_sigacts->ps_sigact[signum]) {
1254 		case (long)SIG_DFL:
1255 			/*
1256 			 * Don't take default actions on system processes.
1257 			 */
1258 			if (pr->ps_pid <= 1) {
1259 #ifdef DIAGNOSTIC
1260 				/*
1261 				 * Are you sure you want to ignore SIGSEGV
1262 				 * in init? XXX
1263 				 */
1264 				printf("Process (pid %d) got signal"
1265 				    " %d\n", pr->ps_pid, signum);
1266 #endif
1267 				break;		/* == ignore */
1268 			}
1269 			/*
1270 			 * If there is a pending stop signal to process
1271 			 * with default action, stop here,
1272 			 * then clear the signal.  However,
1273 			 * if process is member of an orphaned
1274 			 * process group, ignore tty stop signals.
1275 			 */
1276 			if (prop & SA_STOP) {
1277 				if (pr->ps_flags & PS_TRACED ||
1278 		    		    (pr->ps_pgrp->pg_jobc == 0 &&
1279 				    prop & SA_TTYSTOP))
1280 					break;	/* == ignore */
1281 				p->p_xstat = signum;
1282 				if (dolock)
1283 					SCHED_LOCK(s);
1284 				proc_stop(p, 1);
1285 				if (dolock)
1286 					SCHED_UNLOCK(s);
1287 				break;
1288 			} else if (prop & SA_IGNORE) {
1289 				/*
1290 				 * Except for SIGCONT, shouldn't get here.
1291 				 * Default action is to ignore; drop it.
1292 				 */
1293 				break;		/* == ignore */
1294 			} else
1295 				goto keep;
1296 			/*NOTREACHED*/
1297 		case (long)SIG_IGN:
1298 			/*
1299 			 * Masking above should prevent us ever trying
1300 			 * to take action on an ignored signal other
1301 			 * than SIGCONT, unless process is traced.
1302 			 */
1303 			if ((prop & SA_CONT) == 0 &&
1304 			    (pr->ps_flags & PS_TRACED) == 0)
1305 				printf("issignal\n");
1306 			break;		/* == ignore */
1307 		default:
1308 			/*
1309 			 * This signal has an action, let
1310 			 * postsig() process it.
1311 			 */
1312 			goto keep;
1313 		}
1314 	}
1315 	/* NOTREACHED */
1316 
1317 keep:
1318 	atomic_setbits_int(&p->p_siglist, mask); /*leave the signal for later */
1319 	return (signum);
1320 }
1321 
1322 /*
1323  * Put the argument process into the stopped state and notify the parent
1324  * via wakeup.  Signals are handled elsewhere.  The process must not be
1325  * on the run queue.
1326  */
1327 void
1328 proc_stop(struct proc *p, int sw)
1329 {
1330 	struct process *pr = p->p_p;
1331 	extern void *softclock_si;
1332 
1333 #ifdef MULTIPROCESSOR
1334 	SCHED_ASSERT_LOCKED();
1335 #endif
1336 
1337 	p->p_stat = SSTOP;
1338 	atomic_clearbits_int(&pr->ps_flags, PS_WAITED);
1339 	atomic_setbits_int(&pr->ps_flags, PS_STOPPED);
1340 	atomic_setbits_int(&p->p_flag, P_SUSPSIG);
1341 	if (!timeout_pending(&proc_stop_to)) {
1342 		timeout_add(&proc_stop_to, 0);
1343 		/*
1344 		 * We need this soft interrupt to be handled fast.
1345 		 * Extra calls to softclock don't hurt.
1346 		 */
1347                 softintr_schedule(softclock_si);
1348 	}
1349 	if (sw)
1350 		mi_switch();
1351 }
1352 
1353 /*
1354  * Called from a timeout to send signals to the parents of stopped processes.
1355  * We can't do this in proc_stop because it's called with nasty locks held
1356  * and we would need recursive scheduler lock to deal with that.
1357  */
1358 void
1359 proc_stop_sweep(void *v)
1360 {
1361 	struct process *pr;
1362 
1363 	LIST_FOREACH(pr, &allprocess, ps_list) {
1364 		if ((pr->ps_flags & PS_STOPPED) == 0)
1365 			continue;
1366 		atomic_clearbits_int(&pr->ps_flags, PS_STOPPED);
1367 
1368 		if ((pr->ps_pptr->ps_sigacts->ps_flags & SAS_NOCLDSTOP) == 0)
1369 			prsignal(pr->ps_pptr, SIGCHLD);
1370 		wakeup(pr->ps_pptr);
1371 	}
1372 }
1373 
1374 /*
1375  * Take the action for the specified signal
1376  * from the current set of pending signals.
1377  */
1378 void
1379 postsig(struct proc *p, int signum)
1380 {
1381 	struct process *pr = p->p_p;
1382 	struct sigacts *ps = pr->ps_sigacts;
1383 	sig_t action;
1384 	u_long trapno;
1385 	int mask, returnmask;
1386 	siginfo_t si;
1387 	union sigval sigval;
1388 	int s, code;
1389 
1390 	KASSERT(signum != 0);
1391 	KERNEL_ASSERT_LOCKED();
1392 
1393 	mask = sigmask(signum);
1394 	atomic_clearbits_int(&p->p_siglist, mask);
1395 	action = ps->ps_sigact[signum];
1396 	sigval.sival_ptr = 0;
1397 
1398 	if (p->p_sisig != signum) {
1399 		trapno = 0;
1400 		code = SI_USER;
1401 		sigval.sival_ptr = 0;
1402 	} else {
1403 		trapno = p->p_sitrapno;
1404 		code = p->p_sicode;
1405 		sigval = p->p_sigval;
1406 	}
1407 	initsiginfo(&si, signum, trapno, code, sigval);
1408 
1409 #ifdef KTRACE
1410 	if (KTRPOINT(p, KTR_PSIG)) {
1411 		ktrpsig(p, signum, action, p->p_flag & P_SIGSUSPEND ?
1412 		    p->p_oldmask : p->p_sigmask, code, &si);
1413 	}
1414 #endif
1415 	if (action == SIG_DFL) {
1416 		/*
1417 		 * Default action, where the default is to kill
1418 		 * the process.  (Other cases were ignored above.)
1419 		 */
1420 		sigexit(p, signum);
1421 		/* NOTREACHED */
1422 	} else {
1423 		/*
1424 		 * If we get here, the signal must be caught.
1425 		 */
1426 #ifdef DIAGNOSTIC
1427 		if (action == SIG_IGN || (p->p_sigmask & mask))
1428 			panic("postsig action");
1429 #endif
1430 		/*
1431 		 * Set the new mask value and also defer further
1432 		 * occurrences of this signal.
1433 		 *
1434 		 * Special case: user has done a sigpause.  Here the
1435 		 * current mask is not of interest, but rather the
1436 		 * mask from before the sigpause is what we want
1437 		 * restored after the signal processing is completed.
1438 		 */
1439 #ifdef MULTIPROCESSOR
1440 		s = splsched();
1441 #else
1442 		s = splhigh();
1443 #endif
1444 		if (p->p_flag & P_SIGSUSPEND) {
1445 			atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
1446 			returnmask = p->p_oldmask;
1447 		} else {
1448 			returnmask = p->p_sigmask;
1449 		}
1450 		if (p->p_sisig == signum) {
1451 			p->p_sisig = 0;
1452 			p->p_sitrapno = 0;
1453 			p->p_sicode = SI_USER;
1454 			p->p_sigval.sival_ptr = NULL;
1455 		}
1456 
1457 		sendsig(action, signum, returnmask, &si);
1458 		postsig_done(p, signum, ps);
1459 		splx(s);
1460 	}
1461 }
1462 
1463 /*
1464  * Force the current process to exit with the specified signal, dumping core
1465  * if appropriate.  We bypass the normal tests for masked and caught signals,
1466  * allowing unrecoverable failures to terminate the process without changing
1467  * signal state.  Mark the accounting record with the signal termination.
1468  * If dumping core, save the signal number for the debugger.  Calls exit and
1469  * does not return.
1470  */
1471 void
1472 sigexit(struct proc *p, int signum)
1473 {
1474 	/* Mark process as going away */
1475 	atomic_setbits_int(&p->p_flag, P_WEXIT);
1476 
1477 	p->p_p->ps_acflag |= AXSIG;
1478 	if (sigprop[signum] & SA_CORE) {
1479 		p->p_sisig = signum;
1480 
1481 		/* if there are other threads, pause them */
1482 		if (P_HASSIBLING(p))
1483 			single_thread_set(p, SINGLE_SUSPEND, 0);
1484 
1485 		if (coredump(p) == 0)
1486 			signum |= WCOREFLAG;
1487 	}
1488 	exit1(p, W_EXITCODE(0, signum), EXIT_NORMAL);
1489 	/* NOTREACHED */
1490 }
1491 
1492 int nosuidcoredump = 1;
1493 
1494 struct coredump_iostate {
1495 	struct proc *io_proc;
1496 	struct vnode *io_vp;
1497 	struct ucred *io_cred;
1498 	off_t io_offset;
1499 };
1500 
1501 /*
1502  * Dump core, into a file named "progname.core", unless the process was
1503  * setuid/setgid.
1504  */
1505 int
1506 coredump(struct proc *p)
1507 {
1508 #ifdef SMALL_KERNEL
1509 	return EPERM;
1510 #else
1511 	struct process *pr = p->p_p;
1512 	struct vnode *vp;
1513 	struct ucred *cred = p->p_ucred;
1514 	struct vmspace *vm = p->p_vmspace;
1515 	struct nameidata nd;
1516 	struct vattr vattr;
1517 	struct coredump_iostate	io;
1518 	int error, len, incrash = 0;
1519 	char name[MAXPATHLEN];
1520 	const char *dir = "/var/crash";
1521 
1522 	if (pr->ps_emul->e_coredump == NULL)
1523 		return (EINVAL);
1524 
1525 	pr->ps_flags |= PS_COREDUMP;
1526 
1527 	/*
1528 	 * If the process has inconsistent uids, nosuidcoredump
1529 	 * determines coredump placement policy.
1530 	 */
1531 	if (((pr->ps_flags & PS_SUGID) && (error = suser(p))) ||
1532 	   ((pr->ps_flags & PS_SUGID) && nosuidcoredump)) {
1533 		if (nosuidcoredump == 3 || nosuidcoredump == 2)
1534 			incrash = 1;
1535 		else
1536 			return (EPERM);
1537 	}
1538 
1539 	/* Don't dump if will exceed file size limit. */
1540 	if (USPACE + ptoa(vm->vm_dsize + vm->vm_ssize) >=
1541 	    p->p_rlimit[RLIMIT_CORE].rlim_cur)
1542 		return (EFBIG);
1543 
1544 	if (incrash && nosuidcoredump == 3) {
1545 		/*
1546 		 * If the program directory does not exist, dumps of
1547 		 * that core will silently fail.
1548 		 */
1549 		len = snprintf(name, sizeof(name), "%s/%s/%u.core",
1550 		    dir, pr->ps_comm, pr->ps_pid);
1551 	} else if (incrash && nosuidcoredump == 2)
1552 		len = snprintf(name, sizeof(name), "%s/%s.core",
1553 		    dir, pr->ps_comm);
1554 	else
1555 		len = snprintf(name, sizeof(name), "%s.core", pr->ps_comm);
1556 	if (len >= sizeof(name))
1557 		return (EACCES);
1558 
1559 	/*
1560 	 * Control the UID used to write out.  The normal case uses
1561 	 * the real UID.  If the sugid case is going to write into the
1562 	 * controlled directory, we do so as root.
1563 	 */
1564 	if (incrash == 0) {
1565 		cred = crdup(cred);
1566 		cred->cr_uid = cred->cr_ruid;
1567 		cred->cr_gid = cred->cr_rgid;
1568 	} else {
1569 		if (p->p_fd->fd_rdir) {
1570 			vrele(p->p_fd->fd_rdir);
1571 			p->p_fd->fd_rdir = NULL;
1572 		}
1573 		p->p_ucred = crdup(p->p_ucred);
1574 		crfree(cred);
1575 		cred = p->p_ucred;
1576 		crhold(cred);
1577 		cred->cr_uid = 0;
1578 		cred->cr_gid = 0;
1579 	}
1580 
1581 	NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, p);
1582 
1583 	error = vn_open(&nd, O_CREAT | FWRITE | O_NOFOLLOW, S_IRUSR | S_IWUSR);
1584 
1585 	if (error)
1586 		goto out;
1587 
1588 	/*
1589 	 * Don't dump to non-regular files, files with links, or files
1590 	 * owned by someone else.
1591 	 */
1592 	vp = nd.ni_vp;
1593 	if ((error = VOP_GETATTR(vp, &vattr, cred, p)) != 0) {
1594 		VOP_UNLOCK(vp);
1595 		vn_close(vp, FWRITE, cred, p);
1596 		goto out;
1597 	}
1598 	if (vp->v_type != VREG || vattr.va_nlink != 1 ||
1599 	    vattr.va_mode & ((VREAD | VWRITE) >> 3 | (VREAD | VWRITE) >> 6) ||
1600 	    vattr.va_uid != cred->cr_uid) {
1601 		error = EACCES;
1602 		VOP_UNLOCK(vp);
1603 		vn_close(vp, FWRITE, cred, p);
1604 		goto out;
1605 	}
1606 	VATTR_NULL(&vattr);
1607 	vattr.va_size = 0;
1608 	VOP_SETATTR(vp, &vattr, cred, p);
1609 	pr->ps_acflag |= ACORE;
1610 
1611 	io.io_proc = p;
1612 	io.io_vp = vp;
1613 	io.io_cred = cred;
1614 	io.io_offset = 0;
1615 	VOP_UNLOCK(vp);
1616 	vref(vp);
1617 	error = vn_close(vp, FWRITE, cred, p);
1618 	if (error == 0)
1619 		error = (*pr->ps_emul->e_coredump)(p, &io);
1620 	vrele(vp);
1621 out:
1622 	crfree(cred);
1623 	return (error);
1624 #endif
1625 }
1626 
1627 #ifndef SMALL_KERNEL
1628 int
1629 coredump_write(void *cookie, enum uio_seg segflg, const void *data, size_t len)
1630 {
1631 	struct coredump_iostate *io = cookie;
1632 	off_t coffset = 0;
1633 	size_t csize;
1634 	int chunk, error;
1635 
1636 	csize = len;
1637 	do {
1638 		if (io->io_proc->p_siglist & sigmask(SIGKILL))
1639 			return (EINTR);
1640 
1641 		/* Rest of the loop sleeps with lock held, so... */
1642 		yield();
1643 
1644 		chunk = MIN(csize, MAXPHYS);
1645 		error = vn_rdwr(UIO_WRITE, io->io_vp,
1646 		    (caddr_t)data + coffset, chunk,
1647 		    io->io_offset + coffset, segflg,
1648 		    IO_UNIT, io->io_cred, NULL, io->io_proc);
1649 		if (error) {
1650 			struct process *pr = io->io_proc->p_p;
1651 
1652 			if (error == ENOSPC)
1653 				log(LOG_ERR,
1654 				    "coredump of %s(%d) failed, filesystem full\n",
1655 				    pr->ps_comm, pr->ps_pid);
1656 			else
1657 				log(LOG_ERR,
1658 				    "coredump of %s(%d), write failed: errno %d\n",
1659 				    pr->ps_comm, pr->ps_pid, error);
1660 			return (error);
1661 		}
1662 
1663 		coffset += chunk;
1664 		csize -= chunk;
1665 	} while (csize > 0);
1666 
1667 	io->io_offset += len;
1668 	return (0);
1669 }
1670 
1671 void
1672 coredump_unmap(void *cookie, vaddr_t start, vaddr_t end)
1673 {
1674 	struct coredump_iostate *io = cookie;
1675 
1676 	uvm_unmap(&io->io_proc->p_vmspace->vm_map, start, end);
1677 }
1678 
1679 #endif	/* !SMALL_KERNEL */
1680 
1681 /*
1682  * Nonexistent system call-- signal process (may want to handle it).
1683  * Flag error in case process won't see signal immediately (blocked or ignored).
1684  */
1685 int
1686 sys_nosys(struct proc *p, void *v, register_t *retval)
1687 {
1688 
1689 	ptsignal(p, SIGSYS, STHREAD);
1690 	return (ENOSYS);
1691 }
1692 
1693 int
1694 sys___thrsigdivert(struct proc *p, void *v, register_t *retval)
1695 {
1696 	static int sigwaitsleep;
1697 	struct sys___thrsigdivert_args /* {
1698 		syscallarg(sigset_t) sigmask;
1699 		syscallarg(siginfo_t *) info;
1700 		syscallarg(const struct timespec *) timeout;
1701 	} */ *uap = v;
1702 	struct process *pr = p->p_p;
1703 	sigset_t *m;
1704 	sigset_t mask = SCARG(uap, sigmask) &~ sigcantmask;
1705 	siginfo_t si;
1706 	uint64_t to_ticks = 0;
1707 	int timeinvalid = 0;
1708 	int error = 0;
1709 
1710 	memset(&si, 0, sizeof(si));
1711 
1712 	if (SCARG(uap, timeout) != NULL) {
1713 		struct timespec ts;
1714 		if ((error = copyin(SCARG(uap, timeout), &ts, sizeof(ts))) != 0)
1715 			return (error);
1716 #ifdef KTRACE
1717 		if (KTRPOINT(p, KTR_STRUCT))
1718 			ktrreltimespec(p, &ts);
1719 #endif
1720 		if (ts.tv_nsec < 0 || ts.tv_nsec >= 1000000000)
1721 			timeinvalid = 1;
1722 		else {
1723 			to_ticks = (uint64_t)hz * ts.tv_sec +
1724 			    ts.tv_nsec / (tick * 1000);
1725 			if (to_ticks > INT_MAX)
1726 				to_ticks = INT_MAX;
1727 			if (to_ticks == 0 && ts.tv_nsec)
1728 				to_ticks = 1;
1729 		}
1730 	}
1731 
1732 	dosigsuspend(p, p->p_sigmask &~ mask);
1733 	for (;;) {
1734 		si.si_signo = CURSIG(p);
1735 		if (si.si_signo != 0) {
1736 			sigset_t smask = sigmask(si.si_signo);
1737 			if (smask & mask) {
1738 				if (p->p_siglist & smask)
1739 					m = &p->p_siglist;
1740 				else if (pr->ps_mainproc->p_siglist & smask)
1741 					m = &pr->ps_mainproc->p_siglist;
1742 				else {
1743 					/* signal got eaten by someone else? */
1744 					continue;
1745 				}
1746 				atomic_clearbits_int(m, smask);
1747 				error = 0;
1748 				break;
1749 			}
1750 		}
1751 
1752 		/* per-POSIX, delay this error until after the above */
1753 		if (timeinvalid)
1754 			error = EINVAL;
1755 
1756 		if (SCARG(uap, timeout) != NULL && to_ticks == 0)
1757 			error = EAGAIN;
1758 
1759 		if (error != 0)
1760 			break;
1761 
1762 		error = tsleep(&sigwaitsleep, PPAUSE|PCATCH, "sigwait",
1763 		    (int)to_ticks);
1764 	}
1765 
1766 	if (error == 0) {
1767 		*retval = si.si_signo;
1768 		if (SCARG(uap, info) != NULL)
1769 			error = copyout(&si, SCARG(uap, info), sizeof(si));
1770 	} else if (error == ERESTART && SCARG(uap, timeout) != NULL) {
1771 		/*
1772 		 * Restarting is wrong if there's a timeout, as it'll be
1773 		 * for the same interval again
1774 		 */
1775 		error = EINTR;
1776 	}
1777 
1778 	return (error);
1779 }
1780 
1781 void
1782 initsiginfo(siginfo_t *si, int sig, u_long trapno, int code, union sigval val)
1783 {
1784 	memset(si, 0, sizeof(*si));
1785 
1786 	si->si_signo = sig;
1787 	si->si_code = code;
1788 	if (code == SI_USER) {
1789 		si->si_value = val;
1790 	} else {
1791 		switch (sig) {
1792 		case SIGSEGV:
1793 		case SIGILL:
1794 		case SIGBUS:
1795 		case SIGFPE:
1796 			si->si_addr = val.sival_ptr;
1797 			si->si_trapno = trapno;
1798 			break;
1799 		case SIGXFSZ:
1800 			break;
1801 		}
1802 	}
1803 }
1804 
1805 int
1806 filt_sigattach(struct knote *kn)
1807 {
1808 	struct process *pr = curproc->p_p;
1809 
1810 	if (kn->kn_id >= NSIG)
1811 		return EINVAL;
1812 
1813 	kn->kn_ptr.p_process = pr;
1814 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
1815 
1816 	/* XXX lock the proc here while adding to the list? */
1817 	SLIST_INSERT_HEAD(&pr->ps_klist, kn, kn_selnext);
1818 
1819 	return (0);
1820 }
1821 
1822 void
1823 filt_sigdetach(struct knote *kn)
1824 {
1825 	struct process *pr = kn->kn_ptr.p_process;
1826 
1827 	SLIST_REMOVE(&pr->ps_klist, kn, knote, kn_selnext);
1828 }
1829 
1830 /*
1831  * signal knotes are shared with proc knotes, so we apply a mask to
1832  * the hint in order to differentiate them from process hints.  This
1833  * could be avoided by using a signal-specific knote list, but probably
1834  * isn't worth the trouble.
1835  */
1836 int
1837 filt_signal(struct knote *kn, long hint)
1838 {
1839 
1840 	if (hint & NOTE_SIGNAL) {
1841 		hint &= ~NOTE_SIGNAL;
1842 
1843 		if (kn->kn_id == hint)
1844 			kn->kn_data++;
1845 	}
1846 	return (kn->kn_data != 0);
1847 }
1848 
1849 void
1850 userret(struct proc *p)
1851 {
1852 	int signum;
1853 
1854 	/* send SIGPROF or SIGVTALRM if their timers interrupted this thread */
1855 	if (p->p_flag & P_PROFPEND) {
1856 		atomic_clearbits_int(&p->p_flag, P_PROFPEND);
1857 		KERNEL_LOCK();
1858 		psignal(p, SIGPROF);
1859 		KERNEL_UNLOCK();
1860 	}
1861 	if (p->p_flag & P_ALRMPEND) {
1862 		atomic_clearbits_int(&p->p_flag, P_ALRMPEND);
1863 		KERNEL_LOCK();
1864 		psignal(p, SIGVTALRM);
1865 		KERNEL_UNLOCK();
1866 	}
1867 
1868 	if (SIGPENDING(p)) {
1869 		KERNEL_LOCK();
1870 		while ((signum = CURSIG(p)) != 0)
1871 			postsig(p, signum);
1872 		KERNEL_UNLOCK();
1873 	}
1874 
1875 	/*
1876 	 * If P_SIGSUSPEND is still set here, then we still need to restore
1877 	 * the original sigmask before returning to userspace.  Also, this
1878 	 * might unmask some pending signals, so we need to check a second
1879 	 * time for signals to post.
1880 	 */
1881 	if (p->p_flag & P_SIGSUSPEND) {
1882 		atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
1883 		p->p_sigmask = p->p_oldmask;
1884 
1885 		KERNEL_LOCK();
1886 		while ((signum = CURSIG(p)) != 0)
1887 			postsig(p, signum);
1888 		KERNEL_UNLOCK();
1889 	}
1890 
1891 	if (p->p_flag & P_SUSPSINGLE) {
1892 		KERNEL_LOCK();
1893 		single_thread_check(p, 0);
1894 		KERNEL_UNLOCK();
1895 	}
1896 
1897 	WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
1898 
1899 	p->p_cpu->ci_schedstate.spc_curpriority = p->p_priority = p->p_usrpri;
1900 }
1901 
1902 int
1903 single_thread_check(struct proc *p, int deep)
1904 {
1905 	struct process *pr = p->p_p;
1906 
1907 	if (pr->ps_single != NULL && pr->ps_single != p) {
1908 		do {
1909 			int s;
1910 
1911 			/* if we're in deep, we need to unwind to the edge */
1912 			if (deep) {
1913 				if (pr->ps_flags & PS_SINGLEUNWIND)
1914 					return (ERESTART);
1915 				if (pr->ps_flags & PS_SINGLEEXIT)
1916 					return (EINTR);
1917 			}
1918 
1919 			if (--pr->ps_singlecount == 0)
1920 				wakeup(&pr->ps_singlecount);
1921 			if (pr->ps_flags & PS_SINGLEEXIT)
1922 				exit1(p, 0, EXIT_THREAD_NOCHECK);
1923 
1924 			/* not exiting and don't need to unwind, so suspend */
1925 			SCHED_LOCK(s);
1926 			p->p_stat = SSTOP;
1927 			mi_switch();
1928 			SCHED_UNLOCK(s);
1929 		} while (pr->ps_single != NULL);
1930 	}
1931 
1932 	return (0);
1933 }
1934 
1935 /*
1936  * Stop other threads in the process.  The mode controls how and
1937  * where the other threads should stop:
1938  *  - SINGLE_SUSPEND: stop wherever they are, will later either be told to exit
1939  *    (by setting to SINGLE_EXIT) or be released (via single_thread_clear())
1940  *  - SINGLE_PTRACE: stop wherever they are, will wait for them to stop
1941  *    later (via single_thread_wait()) and released as with SINGLE_SUSPEND
1942  *  - SINGLE_UNWIND: just unwind to kernel boundary, will be told to exit
1943  *    or released as with SINGLE_SUSPEND
1944  *  - SINGLE_EXIT: unwind to kernel boundary and exit
1945  */
1946 int
1947 single_thread_set(struct proc *p, enum single_thread_mode mode, int deep)
1948 {
1949 	struct process *pr = p->p_p;
1950 	struct proc *q;
1951 	int error;
1952 
1953 	KERNEL_ASSERT_LOCKED();
1954 
1955 	if ((error = single_thread_check(p, deep)))
1956 		return error;
1957 
1958 	switch (mode) {
1959 	case SINGLE_SUSPEND:
1960 	case SINGLE_PTRACE:
1961 		break;
1962 	case SINGLE_UNWIND:
1963 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
1964 		break;
1965 	case SINGLE_EXIT:
1966 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEEXIT);
1967 		atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
1968 		break;
1969 #ifdef DIAGNOSTIC
1970 	default:
1971 		panic("single_thread_mode = %d", mode);
1972 #endif
1973 	}
1974 	pr->ps_single = p;
1975 	pr->ps_singlecount = 0;
1976 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
1977 		int s;
1978 
1979 		if (q == p)
1980 			continue;
1981 		if (q->p_flag & P_WEXIT) {
1982 			if (mode == SINGLE_EXIT) {
1983 				SCHED_LOCK(s);
1984 				if (q->p_stat == SSTOP) {
1985 					setrunnable(q);
1986 					pr->ps_singlecount++;
1987 				}
1988 				SCHED_UNLOCK(s);
1989 			}
1990 			continue;
1991 		}
1992 		SCHED_LOCK(s);
1993 		atomic_setbits_int(&q->p_flag, P_SUSPSINGLE);
1994 		switch (q->p_stat) {
1995 		case SIDL:
1996 		case SRUN:
1997 			pr->ps_singlecount++;
1998 			break;
1999 		case SSLEEP:
2000 			/* if it's not interruptible, then just have to wait */
2001 			if (q->p_flag & P_SINTR) {
2002 				/* merely need to suspend?  just stop it */
2003 				if (mode == SINGLE_SUSPEND ||
2004 				    mode == SINGLE_PTRACE) {
2005 					q->p_stat = SSTOP;
2006 					break;
2007 				}
2008 				/* need to unwind or exit, so wake it */
2009 				setrunnable(q);
2010 			}
2011 			pr->ps_singlecount++;
2012 			break;
2013 		case SSTOP:
2014 			if (mode == SINGLE_EXIT) {
2015 				setrunnable(q);
2016 				pr->ps_singlecount++;
2017 			}
2018 			break;
2019 		case SDEAD:
2020 			break;
2021 		case SONPROC:
2022 			pr->ps_singlecount++;
2023 			signotify(q);
2024 			break;
2025 		}
2026 		SCHED_UNLOCK(s);
2027 	}
2028 
2029 	if (mode != SINGLE_PTRACE)
2030 		single_thread_wait(pr);
2031 
2032 	return 0;
2033 }
2034 
2035 void
2036 single_thread_wait(struct process *pr)
2037 {
2038 	/* wait until they're all suspended */
2039 	while (pr->ps_singlecount > 0)
2040 		tsleep(&pr->ps_singlecount, PUSER, "suspend", 0);
2041 }
2042 
2043 void
2044 single_thread_clear(struct proc *p, int flag)
2045 {
2046 	struct process *pr = p->p_p;
2047 	struct proc *q;
2048 
2049 	KASSERT(pr->ps_single == p);
2050 	KERNEL_ASSERT_LOCKED();
2051 
2052 	pr->ps_single = NULL;
2053 	atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND | PS_SINGLEEXIT);
2054 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
2055 		int s;
2056 
2057 		if (q == p || (q->p_flag & P_SUSPSINGLE) == 0)
2058 			continue;
2059 		atomic_clearbits_int(&q->p_flag, P_SUSPSINGLE);
2060 
2061 		/*
2062 		 * if the thread was only stopped for single threading
2063 		 * then clearing that either makes it runnable or puts
2064 		 * it back into some sleep queue
2065 		 */
2066 		SCHED_LOCK(s);
2067 		if (q->p_stat == SSTOP && (q->p_flag & flag) == 0) {
2068 			if (q->p_wchan == 0)
2069 				setrunnable(q);
2070 			else
2071 				q->p_stat = SSLEEP;
2072 		}
2073 		SCHED_UNLOCK(s);
2074 	}
2075 }
2076 
2077 void
2078 sigio_del(struct sigiolst *rmlist)
2079 {
2080 	struct sigio *sigio;
2081 
2082 	while ((sigio = LIST_FIRST(rmlist)) != NULL) {
2083 		LIST_REMOVE(sigio, sio_pgsigio);
2084 		crfree(sigio->sio_ucred);
2085 		free(sigio, M_SIGIO, sizeof(*sigio));
2086 	}
2087 }
2088 
2089 void
2090 sigio_unlink(struct sigio_ref *sir, struct sigiolst *rmlist)
2091 {
2092 	struct sigio *sigio;
2093 
2094 	MUTEX_ASSERT_LOCKED(&sigio_lock);
2095 
2096 	sigio = sir->sir_sigio;
2097 	if (sigio != NULL) {
2098 		KASSERT(sigio->sio_myref == sir);
2099 		sir->sir_sigio = NULL;
2100 
2101 		if (sigio->sio_pgid > 0)
2102 			sigio->sio_proc = NULL;
2103 		else
2104 			sigio->sio_pgrp = NULL;
2105 		LIST_REMOVE(sigio, sio_pgsigio);
2106 
2107 		LIST_INSERT_HEAD(rmlist, sigio, sio_pgsigio);
2108 	}
2109 }
2110 
2111 void
2112 sigio_free(struct sigio_ref *sir)
2113 {
2114 	struct sigiolst rmlist;
2115 
2116 	if (sir->sir_sigio == NULL)
2117 		return;
2118 
2119 	LIST_INIT(&rmlist);
2120 
2121 	mtx_enter(&sigio_lock);
2122 	sigio_unlink(sir, &rmlist);
2123 	mtx_leave(&sigio_lock);
2124 
2125 	sigio_del(&rmlist);
2126 }
2127 
2128 void
2129 sigio_freelist(struct sigiolst *sigiolst)
2130 {
2131 	struct sigiolst rmlist;
2132 	struct sigio *sigio;
2133 
2134 	if (LIST_EMPTY(sigiolst))
2135 		return;
2136 
2137 	LIST_INIT(&rmlist);
2138 
2139 	mtx_enter(&sigio_lock);
2140 	while ((sigio = LIST_FIRST(sigiolst)) != NULL)
2141 		sigio_unlink(sigio->sio_myref, &rmlist);
2142 	mtx_leave(&sigio_lock);
2143 
2144 	sigio_del(&rmlist);
2145 }
2146 
2147 int
2148 sigio_setown(struct sigio_ref *sir, pid_t pgid)
2149 {
2150 	struct sigiolst rmlist;
2151 	struct proc *p = curproc;
2152 	struct pgrp *pgrp = NULL;
2153 	struct process *pr = NULL;
2154 	struct sigio *sigio;
2155 	int error;
2156 
2157 	if (pgid == 0) {
2158 		sigio_free(sir);
2159 		return (0);
2160 	}
2161 
2162 	sigio = malloc(sizeof(*sigio), M_SIGIO, M_WAITOK);
2163 	sigio->sio_pgid = pgid;
2164 	sigio->sio_ucred = crhold(p->p_ucred);
2165 	sigio->sio_myref = sir;
2166 
2167 	LIST_INIT(&rmlist);
2168 
2169 	/*
2170 	 * The kernel lock, and not sleeping between prfind()/pgfind() and
2171 	 * linking of the sigio ensure that the process or process group does
2172 	 * not disappear unexpectedly.
2173 	 */
2174 	KERNEL_LOCK();
2175 	mtx_enter(&sigio_lock);
2176 
2177 	if (pgid > 0) {
2178 		pr = prfind(pgid);
2179 		if (pr == NULL) {
2180 			error = ESRCH;
2181 			goto fail;
2182 		}
2183 
2184 		/*
2185 		 * Policy - Don't allow a process to FSETOWN a process
2186 		 * in another session.
2187 		 *
2188 		 * Remove this test to allow maximum flexibility or
2189 		 * restrict FSETOWN to the current process or process
2190 		 * group for maximum safety.
2191 		 */
2192 		if (pr->ps_session != p->p_p->ps_session) {
2193 			error = EPERM;
2194 			goto fail;
2195 		}
2196 
2197 		if ((pr->ps_flags & PS_EXITING) != 0) {
2198 			error = ESRCH;
2199 			goto fail;
2200 		}
2201 	} else /* if (pgid < 0) */ {
2202 		pgrp = pgfind(-pgid);
2203 		if (pgrp == NULL) {
2204 			error = ESRCH;
2205 			goto fail;
2206 		}
2207 
2208 		/*
2209 		 * Policy - Don't allow a process to FSETOWN a process
2210 		 * in another session.
2211 		 *
2212 		 * Remove this test to allow maximum flexibility or
2213 		 * restrict FSETOWN to the current process or process
2214 		 * group for maximum safety.
2215 		 */
2216 		if (pgrp->pg_session != p->p_p->ps_session) {
2217 			error = EPERM;
2218 			goto fail;
2219 		}
2220 	}
2221 
2222 	if (pgid > 0) {
2223 		sigio->sio_proc = pr;
2224 		LIST_INSERT_HEAD(&pr->ps_sigiolst, sigio, sio_pgsigio);
2225 	} else {
2226 		sigio->sio_pgrp = pgrp;
2227 		LIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
2228 	}
2229 
2230 	sigio_unlink(sir, &rmlist);
2231 	sir->sir_sigio = sigio;
2232 
2233 	mtx_leave(&sigio_lock);
2234 	KERNEL_UNLOCK();
2235 
2236 	sigio_del(&rmlist);
2237 
2238 	return (0);
2239 
2240 fail:
2241 	mtx_leave(&sigio_lock);
2242 	KERNEL_UNLOCK();
2243 
2244 	crfree(sigio->sio_ucred);
2245 	free(sigio, M_SIGIO, sizeof(*sigio));
2246 
2247 	return (error);
2248 }
2249 
2250 pid_t
2251 sigio_getown(struct sigio_ref *sir)
2252 {
2253 	struct sigio *sigio;
2254 	pid_t pgid = 0;
2255 
2256 	mtx_enter(&sigio_lock);
2257 	sigio = sir->sir_sigio;
2258 	if (sigio != NULL)
2259 		pgid = sigio->sio_pgid;
2260 	mtx_leave(&sigio_lock);
2261 
2262 	return (pgid);
2263 }
2264 
2265 void
2266 sigio_copy(struct sigio_ref *dst, struct sigio_ref *src)
2267 {
2268 	struct sigiolst rmlist;
2269 	struct sigio *newsigio, *sigio;
2270 
2271 	sigio_free(dst);
2272 
2273 	if (src->sir_sigio == NULL)
2274 		return;
2275 
2276 	newsigio = malloc(sizeof(*newsigio), M_SIGIO, M_WAITOK);
2277 	LIST_INIT(&rmlist);
2278 
2279 	mtx_enter(&sigio_lock);
2280 
2281 	sigio = src->sir_sigio;
2282 	if (sigio == NULL) {
2283 		mtx_leave(&sigio_lock);
2284 		free(newsigio, M_SIGIO, sizeof(*newsigio));
2285 		return;
2286 	}
2287 
2288 	newsigio->sio_pgid = sigio->sio_pgid;
2289 	newsigio->sio_ucred = crhold(sigio->sio_ucred);
2290 	newsigio->sio_myref = dst;
2291 	if (newsigio->sio_pgid > 0) {
2292 		newsigio->sio_proc = sigio->sio_proc;
2293 		LIST_INSERT_HEAD(&newsigio->sio_proc->ps_sigiolst, newsigio,
2294 		    sio_pgsigio);
2295 	} else {
2296 		newsigio->sio_pgrp = sigio->sio_pgrp;
2297 		LIST_INSERT_HEAD(&newsigio->sio_pgrp->pg_sigiolst, newsigio,
2298 		    sio_pgsigio);
2299 	}
2300 
2301 	sigio_unlink(dst, &rmlist);
2302 	dst->sir_sigio = newsigio;
2303 
2304 	mtx_leave(&sigio_lock);
2305 
2306 	sigio_del(&rmlist);
2307 }
2308