xref: /openbsd-src/sys/kern/kern_sig.c (revision 4e1ee0786f11cc571bd0be17d38e46f635c719fc)
1 /*	$OpenBSD: kern_sig.c,v 1.285 2021/10/06 15:46:03 claudio Exp $	*/
2 /*	$NetBSD: kern_sig.c,v 1.54 1996/04/22 01:38:32 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1997 Theo de Raadt. All rights reserved.
6  * Copyright (c) 1982, 1986, 1989, 1991, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  * (c) UNIX System Laboratories, Inc.
9  * All or some portions of this file are derived from material licensed
10  * to the University of California by American Telephone and Telegraph
11  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12  * the permission of UNIX System Laboratories, Inc.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
39  */
40 
41 #include <sys/param.h>
42 #include <sys/signalvar.h>
43 #include <sys/resourcevar.h>
44 #include <sys/queue.h>
45 #include <sys/namei.h>
46 #include <sys/vnode.h>
47 #include <sys/event.h>
48 #include <sys/proc.h>
49 #include <sys/systm.h>
50 #include <sys/acct.h>
51 #include <sys/fcntl.h>
52 #include <sys/filedesc.h>
53 #include <sys/kernel.h>
54 #include <sys/wait.h>
55 #include <sys/ktrace.h>
56 #include <sys/stat.h>
57 #include <sys/core.h>
58 #include <sys/malloc.h>
59 #include <sys/pool.h>
60 #include <sys/ptrace.h>
61 #include <sys/sched.h>
62 #include <sys/user.h>
63 #include <sys/syslog.h>
64 #include <sys/ttycom.h>
65 #include <sys/pledge.h>
66 #include <sys/witness.h>
67 
68 #include <sys/mount.h>
69 #include <sys/syscallargs.h>
70 
71 #include <uvm/uvm_extern.h>
72 #include <machine/tcb.h>
73 
74 int	filt_sigattach(struct knote *kn);
75 void	filt_sigdetach(struct knote *kn);
76 int	filt_signal(struct knote *kn, long hint);
77 
78 const struct filterops sig_filtops = {
79 	.f_flags	= 0,
80 	.f_attach	= filt_sigattach,
81 	.f_detach	= filt_sigdetach,
82 	.f_event	= filt_signal,
83 };
84 
85 const int sigprop[NSIG + 1] = {
86 	0,			/* unused */
87 	SA_KILL,		/* SIGHUP */
88 	SA_KILL,		/* SIGINT */
89 	SA_KILL|SA_CORE,	/* SIGQUIT */
90 	SA_KILL|SA_CORE,	/* SIGILL */
91 	SA_KILL|SA_CORE,	/* SIGTRAP */
92 	SA_KILL|SA_CORE,	/* SIGABRT */
93 	SA_KILL|SA_CORE,	/* SIGEMT */
94 	SA_KILL|SA_CORE,	/* SIGFPE */
95 	SA_KILL,		/* SIGKILL */
96 	SA_KILL|SA_CORE,	/* SIGBUS */
97 	SA_KILL|SA_CORE,	/* SIGSEGV */
98 	SA_KILL|SA_CORE,	/* SIGSYS */
99 	SA_KILL,		/* SIGPIPE */
100 	SA_KILL,		/* SIGALRM */
101 	SA_KILL,		/* SIGTERM */
102 	SA_IGNORE,		/* SIGURG */
103 	SA_STOP,		/* SIGSTOP */
104 	SA_STOP|SA_TTYSTOP,	/* SIGTSTP */
105 	SA_IGNORE|SA_CONT,	/* SIGCONT */
106 	SA_IGNORE,		/* SIGCHLD */
107 	SA_STOP|SA_TTYSTOP,	/* SIGTTIN */
108 	SA_STOP|SA_TTYSTOP,	/* SIGTTOU */
109 	SA_IGNORE,		/* SIGIO */
110 	SA_KILL,		/* SIGXCPU */
111 	SA_KILL,		/* SIGXFSZ */
112 	SA_KILL,		/* SIGVTALRM */
113 	SA_KILL,		/* SIGPROF */
114 	SA_IGNORE,		/* SIGWINCH  */
115 	SA_IGNORE,		/* SIGINFO */
116 	SA_KILL,		/* SIGUSR1 */
117 	SA_KILL,		/* SIGUSR2 */
118 	SA_IGNORE,		/* SIGTHR */
119 };
120 
121 #define	CONTSIGMASK	(sigmask(SIGCONT))
122 #define	STOPSIGMASK	(sigmask(SIGSTOP) | sigmask(SIGTSTP) | \
123 			    sigmask(SIGTTIN) | sigmask(SIGTTOU))
124 
125 void setsigvec(struct proc *, int, struct sigaction *);
126 
127 void proc_stop(struct proc *p, int);
128 void proc_stop_sweep(void *);
129 void *proc_stop_si;
130 
131 void postsig(struct proc *, int);
132 int cansignal(struct proc *, struct process *, int);
133 
134 struct pool sigacts_pool;	/* memory pool for sigacts structures */
135 
136 void sigio_del(struct sigiolst *);
137 void sigio_unlink(struct sigio_ref *, struct sigiolst *);
138 struct mutex sigio_lock = MUTEX_INITIALIZER(IPL_HIGH);
139 
140 /*
141  * Can thread p, send the signal signum to process qr?
142  */
143 int
144 cansignal(struct proc *p, struct process *qr, int signum)
145 {
146 	struct process *pr = p->p_p;
147 	struct ucred *uc = p->p_ucred;
148 	struct ucred *quc = qr->ps_ucred;
149 
150 	if (uc->cr_uid == 0)
151 		return (1);		/* root can always signal */
152 
153 	if (pr == qr)
154 		return (1);		/* process can always signal itself */
155 
156 	/* optimization: if the same creds then the tests below will pass */
157 	if (uc == quc)
158 		return (1);
159 
160 	if (signum == SIGCONT && qr->ps_session == pr->ps_session)
161 		return (1);		/* SIGCONT in session */
162 
163 	/*
164 	 * Using kill(), only certain signals can be sent to setugid
165 	 * child processes
166 	 */
167 	if (qr->ps_flags & PS_SUGID) {
168 		switch (signum) {
169 		case 0:
170 		case SIGKILL:
171 		case SIGINT:
172 		case SIGTERM:
173 		case SIGALRM:
174 		case SIGSTOP:
175 		case SIGTTIN:
176 		case SIGTTOU:
177 		case SIGTSTP:
178 		case SIGHUP:
179 		case SIGUSR1:
180 		case SIGUSR2:
181 			if (uc->cr_ruid == quc->cr_ruid ||
182 			    uc->cr_uid == quc->cr_ruid)
183 				return (1);
184 		}
185 		return (0);
186 	}
187 
188 	if (uc->cr_ruid == quc->cr_ruid ||
189 	    uc->cr_ruid == quc->cr_svuid ||
190 	    uc->cr_uid == quc->cr_ruid ||
191 	    uc->cr_uid == quc->cr_svuid)
192 		return (1);
193 	return (0);
194 }
195 
196 /*
197  * Initialize signal-related data structures.
198  */
199 void
200 signal_init(void)
201 {
202 	proc_stop_si = softintr_establish(IPL_SOFTCLOCK, proc_stop_sweep,
203 	    NULL);
204 	if (proc_stop_si == NULL)
205 		panic("signal_init failed to register softintr");
206 
207 	pool_init(&sigacts_pool, sizeof(struct sigacts), 0, IPL_NONE,
208 	    PR_WAITOK, "sigapl", NULL);
209 }
210 
211 /*
212  * Create an initial sigacts structure, using the same signal state
213  * as pr.
214  */
215 struct sigacts *
216 sigactsinit(struct process *pr)
217 {
218 	struct sigacts *ps;
219 
220 	ps = pool_get(&sigacts_pool, PR_WAITOK);
221 	memcpy(ps, pr->ps_sigacts, sizeof(struct sigacts));
222 	return (ps);
223 }
224 
225 /*
226  * Initialize a new sigaltstack structure.
227  */
228 void
229 sigstkinit(struct sigaltstack *ss)
230 {
231 	ss->ss_flags = SS_DISABLE;
232 	ss->ss_size = 0;
233 	ss->ss_sp = 0;
234 }
235 
236 /*
237  * Release a sigacts structure.
238  */
239 void
240 sigactsfree(struct process *pr)
241 {
242 	struct sigacts *ps = pr->ps_sigacts;
243 
244 	pr->ps_sigacts = NULL;
245 
246 	pool_put(&sigacts_pool, ps);
247 }
248 
249 int
250 sys_sigaction(struct proc *p, void *v, register_t *retval)
251 {
252 	struct sys_sigaction_args /* {
253 		syscallarg(int) signum;
254 		syscallarg(const struct sigaction *) nsa;
255 		syscallarg(struct sigaction *) osa;
256 	} */ *uap = v;
257 	struct sigaction vec;
258 #ifdef KTRACE
259 	struct sigaction ovec;
260 #endif
261 	struct sigaction *sa;
262 	const struct sigaction *nsa;
263 	struct sigaction *osa;
264 	struct sigacts *ps = p->p_p->ps_sigacts;
265 	int signum;
266 	int bit, error;
267 
268 	signum = SCARG(uap, signum);
269 	nsa = SCARG(uap, nsa);
270 	osa = SCARG(uap, osa);
271 
272 	if (signum <= 0 || signum >= NSIG ||
273 	    (nsa && (signum == SIGKILL || signum == SIGSTOP)))
274 		return (EINVAL);
275 	sa = &vec;
276 	if (osa) {
277 		sa->sa_handler = ps->ps_sigact[signum];
278 		sa->sa_mask = ps->ps_catchmask[signum];
279 		bit = sigmask(signum);
280 		sa->sa_flags = 0;
281 		if ((ps->ps_sigonstack & bit) != 0)
282 			sa->sa_flags |= SA_ONSTACK;
283 		if ((ps->ps_sigintr & bit) == 0)
284 			sa->sa_flags |= SA_RESTART;
285 		if ((ps->ps_sigreset & bit) != 0)
286 			sa->sa_flags |= SA_RESETHAND;
287 		if ((ps->ps_siginfo & bit) != 0)
288 			sa->sa_flags |= SA_SIGINFO;
289 		if (signum == SIGCHLD) {
290 			if ((ps->ps_sigflags & SAS_NOCLDSTOP) != 0)
291 				sa->sa_flags |= SA_NOCLDSTOP;
292 			if ((ps->ps_sigflags & SAS_NOCLDWAIT) != 0)
293 				sa->sa_flags |= SA_NOCLDWAIT;
294 		}
295 		if ((sa->sa_mask & bit) == 0)
296 			sa->sa_flags |= SA_NODEFER;
297 		sa->sa_mask &= ~bit;
298 		error = copyout(sa, osa, sizeof (vec));
299 		if (error)
300 			return (error);
301 #ifdef KTRACE
302 		if (KTRPOINT(p, KTR_STRUCT))
303 			ovec = vec;
304 #endif
305 	}
306 	if (nsa) {
307 		error = copyin(nsa, sa, sizeof (vec));
308 		if (error)
309 			return (error);
310 #ifdef KTRACE
311 		if (KTRPOINT(p, KTR_STRUCT))
312 			ktrsigaction(p, sa);
313 #endif
314 		setsigvec(p, signum, sa);
315 	}
316 #ifdef KTRACE
317 	if (osa && KTRPOINT(p, KTR_STRUCT))
318 		ktrsigaction(p, &ovec);
319 #endif
320 	return (0);
321 }
322 
323 void
324 setsigvec(struct proc *p, int signum, struct sigaction *sa)
325 {
326 	struct sigacts *ps = p->p_p->ps_sigacts;
327 	int bit;
328 	int s;
329 
330 	bit = sigmask(signum);
331 	/*
332 	 * Change setting atomically.
333 	 */
334 	s = splhigh();
335 	ps->ps_sigact[signum] = sa->sa_handler;
336 	if ((sa->sa_flags & SA_NODEFER) == 0)
337 		sa->sa_mask |= sigmask(signum);
338 	ps->ps_catchmask[signum] = sa->sa_mask &~ sigcantmask;
339 	if (signum == SIGCHLD) {
340 		if (sa->sa_flags & SA_NOCLDSTOP)
341 			atomic_setbits_int(&ps->ps_sigflags, SAS_NOCLDSTOP);
342 		else
343 			atomic_clearbits_int(&ps->ps_sigflags, SAS_NOCLDSTOP);
344 		/*
345 		 * If the SA_NOCLDWAIT flag is set or the handler
346 		 * is SIG_IGN we reparent the dying child to PID 1
347 		 * (init) which will reap the zombie.  Because we use
348 		 * init to do our dirty work we never set SAS_NOCLDWAIT
349 		 * for PID 1.
350 		 * XXX exit1 rework means this is unnecessary?
351 		 */
352 		if (initprocess->ps_sigacts != ps &&
353 		    ((sa->sa_flags & SA_NOCLDWAIT) ||
354 		    sa->sa_handler == SIG_IGN))
355 			atomic_setbits_int(&ps->ps_sigflags, SAS_NOCLDWAIT);
356 		else
357 			atomic_clearbits_int(&ps->ps_sigflags, SAS_NOCLDWAIT);
358 	}
359 	if ((sa->sa_flags & SA_RESETHAND) != 0)
360 		ps->ps_sigreset |= bit;
361 	else
362 		ps->ps_sigreset &= ~bit;
363 	if ((sa->sa_flags & SA_SIGINFO) != 0)
364 		ps->ps_siginfo |= bit;
365 	else
366 		ps->ps_siginfo &= ~bit;
367 	if ((sa->sa_flags & SA_RESTART) == 0)
368 		ps->ps_sigintr |= bit;
369 	else
370 		ps->ps_sigintr &= ~bit;
371 	if ((sa->sa_flags & SA_ONSTACK) != 0)
372 		ps->ps_sigonstack |= bit;
373 	else
374 		ps->ps_sigonstack &= ~bit;
375 	/*
376 	 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
377 	 * and for signals set to SIG_DFL where the default is to ignore.
378 	 * However, don't put SIGCONT in ps_sigignore,
379 	 * as we have to restart the process.
380 	 */
381 	if (sa->sa_handler == SIG_IGN ||
382 	    (sigprop[signum] & SA_IGNORE && sa->sa_handler == SIG_DFL)) {
383 		atomic_clearbits_int(&p->p_siglist, bit);
384 		atomic_clearbits_int(&p->p_p->ps_siglist, bit);
385 		if (signum != SIGCONT)
386 			ps->ps_sigignore |= bit;	/* easier in psignal */
387 		ps->ps_sigcatch &= ~bit;
388 	} else {
389 		ps->ps_sigignore &= ~bit;
390 		if (sa->sa_handler == SIG_DFL)
391 			ps->ps_sigcatch &= ~bit;
392 		else
393 			ps->ps_sigcatch |= bit;
394 	}
395 	splx(s);
396 }
397 
398 /*
399  * Initialize signal state for process 0;
400  * set to ignore signals that are ignored by default.
401  */
402 void
403 siginit(struct sigacts *ps)
404 {
405 	int i;
406 
407 	for (i = 0; i < NSIG; i++)
408 		if (sigprop[i] & SA_IGNORE && i != SIGCONT)
409 			ps->ps_sigignore |= sigmask(i);
410 	ps->ps_sigflags = SAS_NOCLDWAIT | SAS_NOCLDSTOP;
411 }
412 
413 /*
414  * Reset signals for an exec by the specified thread.
415  */
416 void
417 execsigs(struct proc *p)
418 {
419 	struct sigacts *ps;
420 	int nc, mask;
421 
422 	ps = p->p_p->ps_sigacts;
423 
424 	/*
425 	 * Reset caught signals.  Held signals remain held
426 	 * through p_sigmask (unless they were caught,
427 	 * and are now ignored by default).
428 	 */
429 	while (ps->ps_sigcatch) {
430 		nc = ffs((long)ps->ps_sigcatch);
431 		mask = sigmask(nc);
432 		ps->ps_sigcatch &= ~mask;
433 		if (sigprop[nc] & SA_IGNORE) {
434 			if (nc != SIGCONT)
435 				ps->ps_sigignore |= mask;
436 			atomic_clearbits_int(&p->p_siglist, mask);
437 			atomic_clearbits_int(&p->p_p->ps_siglist, mask);
438 		}
439 		ps->ps_sigact[nc] = SIG_DFL;
440 	}
441 	/*
442 	 * Reset stack state to the user stack.
443 	 * Clear set of signals caught on the signal stack.
444 	 */
445 	sigstkinit(&p->p_sigstk);
446 	atomic_clearbits_int(&ps->ps_sigflags, SAS_NOCLDWAIT);
447 	if (ps->ps_sigact[SIGCHLD] == SIG_IGN)
448 		ps->ps_sigact[SIGCHLD] = SIG_DFL;
449 }
450 
451 /*
452  * Manipulate signal mask.
453  * Note that we receive new mask, not pointer,
454  * and return old mask as return value;
455  * the library stub does the rest.
456  */
457 int
458 sys_sigprocmask(struct proc *p, void *v, register_t *retval)
459 {
460 	struct sys_sigprocmask_args /* {
461 		syscallarg(int) how;
462 		syscallarg(sigset_t) mask;
463 	} */ *uap = v;
464 	int error = 0;
465 	sigset_t mask;
466 
467 	KASSERT(p == curproc);
468 
469 	*retval = p->p_sigmask;
470 	mask = SCARG(uap, mask) &~ sigcantmask;
471 
472 	switch (SCARG(uap, how)) {
473 	case SIG_BLOCK:
474 		atomic_setbits_int(&p->p_sigmask, mask);
475 		break;
476 	case SIG_UNBLOCK:
477 		atomic_clearbits_int(&p->p_sigmask, mask);
478 		break;
479 	case SIG_SETMASK:
480 		p->p_sigmask = mask;
481 		break;
482 	default:
483 		error = EINVAL;
484 		break;
485 	}
486 	return (error);
487 }
488 
489 int
490 sys_sigpending(struct proc *p, void *v, register_t *retval)
491 {
492 
493 	*retval = p->p_siglist | p->p_p->ps_siglist;
494 	return (0);
495 }
496 
497 /*
498  * Temporarily replace calling proc's signal mask for the duration of a
499  * system call.  Original signal mask will be restored by userret().
500  */
501 void
502 dosigsuspend(struct proc *p, sigset_t newmask)
503 {
504 	KASSERT(p == curproc);
505 
506 	p->p_oldmask = p->p_sigmask;
507 	atomic_setbits_int(&p->p_flag, P_SIGSUSPEND);
508 	p->p_sigmask = newmask;
509 }
510 
511 /*
512  * Suspend process until signal, providing mask to be set
513  * in the meantime.  Note nonstandard calling convention:
514  * libc stub passes mask, not pointer, to save a copyin.
515  */
516 int
517 sys_sigsuspend(struct proc *p, void *v, register_t *retval)
518 {
519 	struct sys_sigsuspend_args /* {
520 		syscallarg(int) mask;
521 	} */ *uap = v;
522 	struct process *pr = p->p_p;
523 	struct sigacts *ps = pr->ps_sigacts;
524 
525 	dosigsuspend(p, SCARG(uap, mask) &~ sigcantmask);
526 	while (tsleep_nsec(ps, PPAUSE|PCATCH, "sigsusp", INFSLP) == 0)
527 		/* void */;
528 	/* always return EINTR rather than ERESTART... */
529 	return (EINTR);
530 }
531 
532 int
533 sigonstack(size_t stack)
534 {
535 	const struct sigaltstack *ss = &curproc->p_sigstk;
536 
537 	return (ss->ss_flags & SS_DISABLE ? 0 :
538 	    (stack - (size_t)ss->ss_sp < ss->ss_size));
539 }
540 
541 int
542 sys_sigaltstack(struct proc *p, void *v, register_t *retval)
543 {
544 	struct sys_sigaltstack_args /* {
545 		syscallarg(const struct sigaltstack *) nss;
546 		syscallarg(struct sigaltstack *) oss;
547 	} */ *uap = v;
548 	struct sigaltstack ss;
549 	const struct sigaltstack *nss;
550 	struct sigaltstack *oss;
551 	int onstack = sigonstack(PROC_STACK(p));
552 	int error;
553 
554 	nss = SCARG(uap, nss);
555 	oss = SCARG(uap, oss);
556 
557 	if (oss != NULL) {
558 		ss = p->p_sigstk;
559 		if (onstack)
560 			ss.ss_flags |= SS_ONSTACK;
561 		if ((error = copyout(&ss, oss, sizeof(ss))))
562 			return (error);
563 	}
564 	if (nss == NULL)
565 		return (0);
566 	error = copyin(nss, &ss, sizeof(ss));
567 	if (error)
568 		return (error);
569 	if (onstack)
570 		return (EPERM);
571 	if (ss.ss_flags & ~SS_DISABLE)
572 		return (EINVAL);
573 	if (ss.ss_flags & SS_DISABLE) {
574 		p->p_sigstk.ss_flags = ss.ss_flags;
575 		return (0);
576 	}
577 	if (ss.ss_size < MINSIGSTKSZ)
578 		return (ENOMEM);
579 
580 	error = uvm_map_remap_as_stack(p, (vaddr_t)ss.ss_sp, ss.ss_size);
581 	if (error)
582 		return (error);
583 
584 	p->p_sigstk = ss;
585 	return (0);
586 }
587 
588 int
589 sys_kill(struct proc *cp, void *v, register_t *retval)
590 {
591 	struct sys_kill_args /* {
592 		syscallarg(int) pid;
593 		syscallarg(int) signum;
594 	} */ *uap = v;
595 	struct process *pr;
596 	int pid = SCARG(uap, pid);
597 	int signum = SCARG(uap, signum);
598 	int error;
599 	int zombie = 0;
600 
601 	if ((error = pledge_kill(cp, pid)) != 0)
602 		return (error);
603 	if (((u_int)signum) >= NSIG)
604 		return (EINVAL);
605 	if (pid > 0) {
606 		if ((pr = prfind(pid)) == NULL) {
607 			if ((pr = zombiefind(pid)) == NULL)
608 				return (ESRCH);
609 			else
610 				zombie = 1;
611 		}
612 		if (!cansignal(cp, pr, signum))
613 			return (EPERM);
614 
615 		/* kill single process */
616 		if (signum && !zombie)
617 			prsignal(pr, signum);
618 		return (0);
619 	}
620 	switch (pid) {
621 	case -1:		/* broadcast signal */
622 		return (killpg1(cp, signum, 0, 1));
623 	case 0:			/* signal own process group */
624 		return (killpg1(cp, signum, 0, 0));
625 	default:		/* negative explicit process group */
626 		return (killpg1(cp, signum, -pid, 0));
627 	}
628 }
629 
630 int
631 sys_thrkill(struct proc *cp, void *v, register_t *retval)
632 {
633 	struct sys_thrkill_args /* {
634 		syscallarg(pid_t) tid;
635 		syscallarg(int) signum;
636 		syscallarg(void *) tcb;
637 	} */ *uap = v;
638 	struct proc *p;
639 	int tid = SCARG(uap, tid);
640 	int signum = SCARG(uap, signum);
641 	void *tcb;
642 
643 	if (((u_int)signum) >= NSIG)
644 		return (EINVAL);
645 	if (tid > THREAD_PID_OFFSET) {
646 		if ((p = tfind(tid - THREAD_PID_OFFSET)) == NULL)
647 			return (ESRCH);
648 
649 		/* can only kill threads in the same process */
650 		if (p->p_p != cp->p_p)
651 			return (ESRCH);
652 	} else if (tid == 0)
653 		p = cp;
654 	else
655 		return (EINVAL);
656 
657 	/* optionally require the target thread to have the given tcb addr */
658 	tcb = SCARG(uap, tcb);
659 	if (tcb != NULL && tcb != TCB_GET(p))
660 		return (ESRCH);
661 
662 	if (signum)
663 		ptsignal(p, signum, STHREAD);
664 	return (0);
665 }
666 
667 /*
668  * Common code for kill process group/broadcast kill.
669  * cp is calling process.
670  */
671 int
672 killpg1(struct proc *cp, int signum, int pgid, int all)
673 {
674 	struct process *pr;
675 	struct pgrp *pgrp;
676 	int nfound = 0;
677 
678 	if (all) {
679 		/*
680 		 * broadcast
681 		 */
682 		LIST_FOREACH(pr, &allprocess, ps_list) {
683 			if (pr->ps_pid <= 1 ||
684 			    pr->ps_flags & (PS_SYSTEM | PS_NOBROADCASTKILL) ||
685 			    pr == cp->p_p || !cansignal(cp, pr, signum))
686 				continue;
687 			nfound++;
688 			if (signum)
689 				prsignal(pr, signum);
690 		}
691 	} else {
692 		if (pgid == 0)
693 			/*
694 			 * zero pgid means send to my process group.
695 			 */
696 			pgrp = cp->p_p->ps_pgrp;
697 		else {
698 			pgrp = pgfind(pgid);
699 			if (pgrp == NULL)
700 				return (ESRCH);
701 		}
702 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist) {
703 			if (pr->ps_pid <= 1 || pr->ps_flags & PS_SYSTEM ||
704 			    !cansignal(cp, pr, signum))
705 				continue;
706 			nfound++;
707 			if (signum)
708 				prsignal(pr, signum);
709 		}
710 	}
711 	return (nfound ? 0 : ESRCH);
712 }
713 
714 #define CANDELIVER(uid, euid, pr) \
715 	(euid == 0 || \
716 	(uid) == (pr)->ps_ucred->cr_ruid || \
717 	(uid) == (pr)->ps_ucred->cr_svuid || \
718 	(uid) == (pr)->ps_ucred->cr_uid || \
719 	(euid) == (pr)->ps_ucred->cr_ruid || \
720 	(euid) == (pr)->ps_ucred->cr_svuid || \
721 	(euid) == (pr)->ps_ucred->cr_uid)
722 
723 #define CANSIGIO(cr, pr) \
724 	CANDELIVER((cr)->cr_ruid, (cr)->cr_uid, (pr))
725 
726 /*
727  * Send a signal to a process group.  If checktty is 1,
728  * limit to members which have a controlling terminal.
729  */
730 void
731 pgsignal(struct pgrp *pgrp, int signum, int checkctty)
732 {
733 	struct process *pr;
734 
735 	if (pgrp)
736 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist)
737 			if (checkctty == 0 || pr->ps_flags & PS_CONTROLT)
738 				prsignal(pr, signum);
739 }
740 
741 /*
742  * Send a SIGIO or SIGURG signal to a process or process group using stored
743  * credentials rather than those of the current process.
744  */
745 void
746 pgsigio(struct sigio_ref *sir, int sig, int checkctty)
747 {
748 	struct process *pr;
749 	struct sigio *sigio;
750 
751 	if (sir->sir_sigio == NULL)
752 		return;
753 
754 	KERNEL_LOCK();
755 	mtx_enter(&sigio_lock);
756 	sigio = sir->sir_sigio;
757 	if (sigio == NULL)
758 		goto out;
759 	if (sigio->sio_pgid > 0) {
760 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc))
761 			prsignal(sigio->sio_proc, sig);
762 	} else if (sigio->sio_pgid < 0) {
763 		LIST_FOREACH(pr, &sigio->sio_pgrp->pg_members, ps_pglist) {
764 			if (CANSIGIO(sigio->sio_ucred, pr) &&
765 			    (checkctty == 0 || (pr->ps_flags & PS_CONTROLT)))
766 				prsignal(pr, sig);
767 		}
768 	}
769 out:
770 	mtx_leave(&sigio_lock);
771 	KERNEL_UNLOCK();
772 }
773 
774 /*
775  * Recalculate the signal mask and reset the signal disposition after
776  * usermode frame for delivery is formed.
777  */
778 void
779 postsig_done(struct proc *p, int signum, struct sigacts *ps)
780 {
781 	int mask = sigmask(signum);
782 
783 	KERNEL_ASSERT_LOCKED();
784 
785 	p->p_ru.ru_nsignals++;
786 	atomic_setbits_int(&p->p_sigmask, ps->ps_catchmask[signum]);
787 	if ((ps->ps_sigreset & mask) != 0) {
788 		ps->ps_sigcatch &= ~mask;
789 		if (signum != SIGCONT && sigprop[signum] & SA_IGNORE)
790 			ps->ps_sigignore |= mask;
791 		ps->ps_sigact[signum] = SIG_DFL;
792 	}
793 }
794 
795 /*
796  * Send a signal caused by a trap to the current thread
797  * If it will be caught immediately, deliver it with correct code.
798  * Otherwise, post it normally.
799  */
800 void
801 trapsignal(struct proc *p, int signum, u_long trapno, int code,
802     union sigval sigval)
803 {
804 	struct process *pr = p->p_p;
805 	struct sigacts *ps = pr->ps_sigacts;
806 	int mask;
807 
808 	KERNEL_LOCK();
809 	switch (signum) {
810 	case SIGILL:
811 	case SIGBUS:
812 	case SIGSEGV:
813 		pr->ps_acflag |= ATRAP;
814 		break;
815 	}
816 
817 	mask = sigmask(signum);
818 	if ((pr->ps_flags & PS_TRACED) == 0 &&
819 	    (ps->ps_sigcatch & mask) != 0 &&
820 	    (p->p_sigmask & mask) == 0) {
821 		siginfo_t si;
822 		int info = (ps->ps_siginfo & mask) != 0;
823 		int onstack = (ps->ps_sigonstack & mask) != 0;
824 
825 		initsiginfo(&si, signum, trapno, code, sigval);
826 #ifdef KTRACE
827 		if (KTRPOINT(p, KTR_PSIG)) {
828 			ktrpsig(p, signum, ps->ps_sigact[signum],
829 			    p->p_sigmask, code, &si);
830 		}
831 #endif
832 		if (sendsig(ps->ps_sigact[signum], signum, p->p_sigmask, &si,
833 		    info, onstack)) {
834 			sigexit(p, SIGILL);
835 			/* NOTREACHED */
836 		}
837 		postsig_done(p, signum, ps);
838 	} else {
839 		p->p_sisig = signum;
840 		p->p_sitrapno = trapno;	/* XXX for core dump/debugger */
841 		p->p_sicode = code;
842 		p->p_sigval = sigval;
843 
844 		/*
845 		 * Signals like SIGBUS and SIGSEGV should not, when
846 		 * generated by the kernel, be ignorable or blockable.
847 		 * If it is and we're not being traced, then just kill
848 		 * the process.
849 		 * After vfs_shutdown(9), init(8) cannot receive signals
850 		 * because new code pages of the signal handler cannot be
851 		 * mapped from halted storage.  init(8) may not die or the
852 		 * kernel panics.  Better loop between signal handler and
853 		 * page fault trap until the machine is halted.
854 		 */
855 		if ((pr->ps_flags & PS_TRACED) == 0 &&
856 		    (sigprop[signum] & SA_KILL) &&
857 		    ((p->p_sigmask & mask) || (ps->ps_sigignore & mask)) &&
858 		    pr->ps_pid != 1)
859 			sigexit(p, signum);
860 		ptsignal(p, signum, STHREAD);
861 	}
862 	KERNEL_UNLOCK();
863 }
864 
865 /*
866  * Send the signal to the process.  If the signal has an action, the action
867  * is usually performed by the target process rather than the caller; we add
868  * the signal to the set of pending signals for the process.
869  *
870  * Exceptions:
871  *   o When a stop signal is sent to a sleeping process that takes the
872  *     default action, the process is stopped without awakening it.
873  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
874  *     regardless of the signal action (eg, blocked or ignored).
875  *
876  * Other ignored signals are discarded immediately.
877  */
878 void
879 psignal(struct proc *p, int signum)
880 {
881 	ptsignal(p, signum, SPROCESS);
882 }
883 
884 /*
885  * type = SPROCESS	process signal, can be diverted (sigwait())
886  * type = STHREAD	thread signal, but should be propagated if unhandled
887  * type = SPROPAGATED	propagated to this thread, so don't propagate again
888  */
889 void
890 ptsignal(struct proc *p, int signum, enum signal_type type)
891 {
892 	int s, prop;
893 	sig_t action;
894 	int mask;
895 	int *siglist;
896 	struct process *pr = p->p_p;
897 	struct proc *q;
898 	int wakeparent = 0;
899 
900 	KERNEL_ASSERT_LOCKED();
901 
902 #ifdef DIAGNOSTIC
903 	if ((u_int)signum >= NSIG || signum == 0)
904 		panic("psignal signal number");
905 #endif
906 
907 	/* Ignore signal if the target process is exiting */
908 	if (pr->ps_flags & PS_EXITING)
909 		return;
910 
911 	mask = sigmask(signum);
912 
913 	if (type == SPROCESS) {
914 		/* Accept SIGKILL to coredumping processes */
915 		if (pr->ps_flags & PS_COREDUMP && signum == SIGKILL) {
916 			atomic_setbits_int(&pr->ps_siglist, mask);
917 			return;
918 		}
919 
920 		/*
921 		 * If the current thread can process the signal
922 		 * immediately (it's unblocked) then have it take it.
923 		 */
924 		q = curproc;
925 		if (q != NULL && q->p_p == pr && (q->p_flag & P_WEXIT) == 0 &&
926 		    (q->p_sigmask & mask) == 0)
927 			p = q;
928 		else {
929 			/*
930 			 * A process-wide signal can be diverted to a
931 			 * different thread that's in sigwait() for this
932 			 * signal.  If there isn't such a thread, then
933 			 * pick a thread that doesn't have it blocked so
934 			 * that the stop/kill consideration isn't
935 			 * delayed.  Otherwise, mark it pending on the
936 			 * main thread.
937 			 */
938 			TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
939 				/* ignore exiting threads */
940 				if (q->p_flag & P_WEXIT)
941 					continue;
942 
943 				/* skip threads that have the signal blocked */
944 				if ((q->p_sigmask & mask) != 0)
945 					continue;
946 
947 				/* okay, could send to this thread */
948 				p = q;
949 
950 				/*
951 				 * sigsuspend, sigwait, ppoll/pselect, etc?
952 				 * Definitely go to this thread, as it's
953 				 * already blocked in the kernel.
954 				 */
955 				if (q->p_flag & P_SIGSUSPEND)
956 					break;
957 			}
958 		}
959 	}
960 
961 	if (type != SPROPAGATED)
962 		KNOTE(&pr->ps_klist, NOTE_SIGNAL | signum);
963 
964 	prop = sigprop[signum];
965 
966 	/*
967 	 * If proc is traced, always give parent a chance.
968 	 */
969 	if (pr->ps_flags & PS_TRACED) {
970 		action = SIG_DFL;
971 	} else {
972 		/*
973 		 * If the signal is being ignored,
974 		 * then we forget about it immediately.
975 		 * (Note: we don't set SIGCONT in ps_sigignore,
976 		 * and if it is set to SIG_IGN,
977 		 * action will be SIG_DFL here.)
978 		 */
979 		if (pr->ps_sigacts->ps_sigignore & mask)
980 			return;
981 		if (p->p_sigmask & mask) {
982 			action = SIG_HOLD;
983 		} else if (pr->ps_sigacts->ps_sigcatch & mask) {
984 			action = SIG_CATCH;
985 		} else {
986 			action = SIG_DFL;
987 
988 			if (prop & SA_KILL && pr->ps_nice > NZERO)
989 				 pr->ps_nice = NZERO;
990 
991 			/*
992 			 * If sending a tty stop signal to a member of an
993 			 * orphaned process group, discard the signal here if
994 			 * the action is default; don't stop the process below
995 			 * if sleeping, and don't clear any pending SIGCONT.
996 			 */
997 			if (prop & SA_TTYSTOP && pr->ps_pgrp->pg_jobc == 0)
998 				return;
999 		}
1000 	}
1001 	/*
1002 	 * If delivered to process, mark as pending there.  Continue and stop
1003 	 * signals will be propagated to all threads.  So they are always
1004 	 * marked at thread level.
1005 	 */
1006 	siglist = (type == SPROCESS) ? &pr->ps_siglist : &p->p_siglist;
1007 	if (prop & SA_CONT) {
1008 		siglist = &p->p_siglist;
1009 		atomic_clearbits_int(siglist, STOPSIGMASK);
1010 	}
1011 	if (prop & SA_STOP) {
1012 		siglist = &p->p_siglist;
1013 		atomic_clearbits_int(siglist, CONTSIGMASK);
1014 		atomic_clearbits_int(&p->p_flag, P_CONTINUED);
1015 	}
1016 	atomic_setbits_int(siglist, mask);
1017 
1018 	/*
1019 	 * XXX delay processing of SA_STOP signals unless action == SIG_DFL?
1020 	 */
1021 	if (prop & (SA_CONT | SA_STOP) && type != SPROPAGATED)
1022 		TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link)
1023 			if (q != p)
1024 				ptsignal(q, signum, SPROPAGATED);
1025 
1026 	/*
1027 	 * Defer further processing for signals which are held,
1028 	 * except that stopped processes must be continued by SIGCONT.
1029 	 */
1030 	if (action == SIG_HOLD && ((prop & SA_CONT) == 0 || p->p_stat != SSTOP))
1031 		return;
1032 
1033 	SCHED_LOCK(s);
1034 
1035 	switch (p->p_stat) {
1036 
1037 	case SSLEEP:
1038 		/*
1039 		 * If process is sleeping uninterruptibly
1040 		 * we can't interrupt the sleep... the signal will
1041 		 * be noticed when the process returns through
1042 		 * trap() or syscall().
1043 		 */
1044 		if ((p->p_flag & P_SINTR) == 0)
1045 			goto out;
1046 		/*
1047 		 * Process is sleeping and traced... make it runnable
1048 		 * so it can discover the signal in cursig() and stop
1049 		 * for the parent.
1050 		 */
1051 		if (pr->ps_flags & PS_TRACED)
1052 			goto run;
1053 		/*
1054 		 * If SIGCONT is default (or ignored) and process is
1055 		 * asleep, we are finished; the process should not
1056 		 * be awakened.
1057 		 */
1058 		if ((prop & SA_CONT) && action == SIG_DFL) {
1059 			atomic_clearbits_int(siglist, mask);
1060 			goto out;
1061 		}
1062 		/*
1063 		 * When a sleeping process receives a stop
1064 		 * signal, process immediately if possible.
1065 		 */
1066 		if ((prop & SA_STOP) && action == SIG_DFL) {
1067 			/*
1068 			 * If a child holding parent blocked,
1069 			 * stopping could cause deadlock.
1070 			 */
1071 			if (pr->ps_flags & PS_PPWAIT)
1072 				goto out;
1073 			atomic_clearbits_int(siglist, mask);
1074 			pr->ps_xsig = signum;
1075 			proc_stop(p, 0);
1076 			goto out;
1077 		}
1078 		/*
1079 		 * All other (caught or default) signals
1080 		 * cause the process to run.
1081 		 */
1082 		goto runfast;
1083 		/*NOTREACHED*/
1084 
1085 	case SSTOP:
1086 		/*
1087 		 * If traced process is already stopped,
1088 		 * then no further action is necessary.
1089 		 */
1090 		if (pr->ps_flags & PS_TRACED)
1091 			goto out;
1092 
1093 		/*
1094 		 * Kill signal always sets processes running.
1095 		 */
1096 		if (signum == SIGKILL) {
1097 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1098 			goto runfast;
1099 		}
1100 
1101 		if (prop & SA_CONT) {
1102 			/*
1103 			 * If SIGCONT is default (or ignored), we continue the
1104 			 * process but don't leave the signal in p_siglist, as
1105 			 * it has no further action.  If SIGCONT is held, we
1106 			 * continue the process and leave the signal in
1107 			 * p_siglist.  If the process catches SIGCONT, let it
1108 			 * handle the signal itself.  If it isn't waiting on
1109 			 * an event, then it goes back to run state.
1110 			 * Otherwise, process goes back to sleep state.
1111 			 */
1112 			atomic_setbits_int(&p->p_flag, P_CONTINUED);
1113 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1114 			wakeparent = 1;
1115 			if (action == SIG_DFL)
1116 				atomic_clearbits_int(siglist, mask);
1117 			if (action == SIG_CATCH)
1118 				goto runfast;
1119 			if (p->p_wchan == 0)
1120 				goto run;
1121 			p->p_stat = SSLEEP;
1122 			goto out;
1123 		}
1124 
1125 		if (prop & SA_STOP) {
1126 			/*
1127 			 * Already stopped, don't need to stop again.
1128 			 * (If we did the shell could get confused.)
1129 			 */
1130 			atomic_clearbits_int(siglist, mask);
1131 			goto out;
1132 		}
1133 
1134 		/*
1135 		 * If process is sleeping interruptibly, then simulate a
1136 		 * wakeup so that when it is continued, it will be made
1137 		 * runnable and can look at the signal.  But don't make
1138 		 * the process runnable, leave it stopped.
1139 		 */
1140 		if (p->p_flag & P_SINTR)
1141 			unsleep(p);
1142 		goto out;
1143 
1144 	case SONPROC:
1145 		signotify(p);
1146 		/* FALLTHROUGH */
1147 	default:
1148 		/*
1149 		 * SRUN, SIDL, SDEAD do nothing with the signal,
1150 		 * other than kicking ourselves if we are running.
1151 		 * It will either never be noticed, or noticed very soon.
1152 		 */
1153 		goto out;
1154 	}
1155 	/*NOTREACHED*/
1156 
1157 runfast:
1158 	/*
1159 	 * Raise priority to at least PUSER.
1160 	 */
1161 	if (p->p_usrpri > PUSER)
1162 		p->p_usrpri = PUSER;
1163 run:
1164 	setrunnable(p);
1165 out:
1166 	SCHED_UNLOCK(s);
1167 	if (wakeparent)
1168 		wakeup(pr->ps_pptr);
1169 }
1170 
1171 /*
1172  * Determine signal that should be delivered to process p, the current
1173  * process, 0 if none.
1174  *
1175  * If the current process has received a signal (should be caught or cause
1176  * termination, should interrupt current syscall), return the signal number.
1177  * Stop signals with default action are processed immediately, then cleared;
1178  * they aren't returned.  This is checked after each entry to the system for
1179  * a syscall or trap. The normal call sequence is
1180  *
1181  *	while (signum = cursig(curproc))
1182  *		postsig(signum);
1183  *
1184  * Assumes that if the P_SINTR flag is set, we're holding both the
1185  * kernel and scheduler locks.
1186  */
1187 int
1188 cursig(struct proc *p)
1189 {
1190 	struct process *pr = p->p_p;
1191 	int sigpending, signum, mask, prop;
1192 	int dolock = (p->p_flag & P_SINTR) == 0;
1193 	int s;
1194 
1195 	KERNEL_ASSERT_LOCKED();
1196 
1197 	sigpending = (p->p_siglist | pr->ps_siglist);
1198 	if (sigpending == 0)
1199 		return 0;
1200 
1201 	if (!ISSET(pr->ps_flags, PS_TRACED) && SIGPENDING(p) == 0)
1202 		return 0;
1203 
1204 	for (;;) {
1205 		mask = SIGPENDING(p);
1206 		if (pr->ps_flags & PS_PPWAIT)
1207 			mask &= ~STOPSIGMASK;
1208 		if (mask == 0)	 	/* no signal to send */
1209 			return (0);
1210 		signum = ffs((long)mask);
1211 		mask = sigmask(signum);
1212 		atomic_clearbits_int(&p->p_siglist, mask);
1213 		atomic_clearbits_int(&pr->ps_siglist, mask);
1214 
1215 		/*
1216 		 * We should see pending but ignored signals
1217 		 * only if PS_TRACED was on when they were posted.
1218 		 */
1219 		if (mask & pr->ps_sigacts->ps_sigignore &&
1220 		    (pr->ps_flags & PS_TRACED) == 0)
1221 			continue;
1222 
1223 		/*
1224 		 * If traced, always stop, and stay stopped until released
1225 		 * by the debugger.  If our parent process is waiting for
1226 		 * us, don't hang as we could deadlock.
1227 		 */
1228 		if (((pr->ps_flags & (PS_TRACED | PS_PPWAIT)) == PS_TRACED) &&
1229 		    signum != SIGKILL) {
1230 			pr->ps_xsig = signum;
1231 
1232 			single_thread_set(p, SINGLE_SUSPEND, 0);
1233 
1234 			if (dolock)
1235 				SCHED_LOCK(s);
1236 			proc_stop(p, 1);
1237 			if (dolock)
1238 				SCHED_UNLOCK(s);
1239 
1240 			single_thread_clear(p, 0);
1241 
1242 			/*
1243 			 * If we are no longer being traced, or the parent
1244 			 * didn't give us a signal, look for more signals.
1245 			 */
1246 			if ((pr->ps_flags & PS_TRACED) == 0 ||
1247 			    pr->ps_xsig == 0)
1248 				continue;
1249 
1250 			/*
1251 			 * If the new signal is being masked, look for other
1252 			 * signals.
1253 			 */
1254 			signum = pr->ps_xsig;
1255 			mask = sigmask(signum);
1256 			if ((p->p_sigmask & mask) != 0)
1257 				continue;
1258 
1259 			/* take the signal! */
1260 			atomic_clearbits_int(&p->p_siglist, mask);
1261 			atomic_clearbits_int(&pr->ps_siglist, mask);
1262 		}
1263 
1264 		prop = sigprop[signum];
1265 
1266 		/*
1267 		 * Decide whether the signal should be returned.
1268 		 * Return the signal's number, or fall through
1269 		 * to clear it from the pending mask.
1270 		 */
1271 		switch ((long)pr->ps_sigacts->ps_sigact[signum]) {
1272 		case (long)SIG_DFL:
1273 			/*
1274 			 * Don't take default actions on system processes.
1275 			 */
1276 			if (pr->ps_pid <= 1) {
1277 #ifdef DIAGNOSTIC
1278 				/*
1279 				 * Are you sure you want to ignore SIGSEGV
1280 				 * in init? XXX
1281 				 */
1282 				printf("Process (pid %d) got signal"
1283 				    " %d\n", pr->ps_pid, signum);
1284 #endif
1285 				break;		/* == ignore */
1286 			}
1287 			/*
1288 			 * If there is a pending stop signal to process
1289 			 * with default action, stop here,
1290 			 * then clear the signal.  However,
1291 			 * if process is member of an orphaned
1292 			 * process group, ignore tty stop signals.
1293 			 */
1294 			if (prop & SA_STOP) {
1295 				if (pr->ps_flags & PS_TRACED ||
1296 		    		    (pr->ps_pgrp->pg_jobc == 0 &&
1297 				    prop & SA_TTYSTOP))
1298 					break;	/* == ignore */
1299 				pr->ps_xsig = signum;
1300 				if (dolock)
1301 					SCHED_LOCK(s);
1302 				proc_stop(p, 1);
1303 				if (dolock)
1304 					SCHED_UNLOCK(s);
1305 				break;
1306 			} else if (prop & SA_IGNORE) {
1307 				/*
1308 				 * Except for SIGCONT, shouldn't get here.
1309 				 * Default action is to ignore; drop it.
1310 				 */
1311 				break;		/* == ignore */
1312 			} else
1313 				goto keep;
1314 			/*NOTREACHED*/
1315 		case (long)SIG_IGN:
1316 			/*
1317 			 * Masking above should prevent us ever trying
1318 			 * to take action on an ignored signal other
1319 			 * than SIGCONT, unless process is traced.
1320 			 */
1321 			if ((prop & SA_CONT) == 0 &&
1322 			    (pr->ps_flags & PS_TRACED) == 0)
1323 				printf("%s\n", __func__);
1324 			break;		/* == ignore */
1325 		default:
1326 			/*
1327 			 * This signal has an action, let
1328 			 * postsig() process it.
1329 			 */
1330 			goto keep;
1331 		}
1332 	}
1333 	/* NOTREACHED */
1334 
1335 keep:
1336 	atomic_setbits_int(&p->p_siglist, mask); /*leave the signal for later */
1337 	return (signum);
1338 }
1339 
1340 /*
1341  * Put the argument process into the stopped state and notify the parent
1342  * via wakeup.  Signals are handled elsewhere.  The process must not be
1343  * on the run queue.
1344  */
1345 void
1346 proc_stop(struct proc *p, int sw)
1347 {
1348 	struct process *pr = p->p_p;
1349 
1350 #ifdef MULTIPROCESSOR
1351 	SCHED_ASSERT_LOCKED();
1352 #endif
1353 
1354 	p->p_stat = SSTOP;
1355 	atomic_clearbits_int(&pr->ps_flags, PS_WAITED);
1356 	atomic_setbits_int(&pr->ps_flags, PS_STOPPED);
1357 	atomic_setbits_int(&p->p_flag, P_SUSPSIG);
1358 	/*
1359 	 * We need this soft interrupt to be handled fast.
1360 	 * Extra calls to softclock don't hurt.
1361 	 */
1362 	softintr_schedule(proc_stop_si);
1363 	if (sw)
1364 		mi_switch();
1365 }
1366 
1367 /*
1368  * Called from a soft interrupt to send signals to the parents of stopped
1369  * processes.
1370  * We can't do this in proc_stop because it's called with nasty locks held
1371  * and we would need recursive scheduler lock to deal with that.
1372  */
1373 void
1374 proc_stop_sweep(void *v)
1375 {
1376 	struct process *pr;
1377 
1378 	LIST_FOREACH(pr, &allprocess, ps_list) {
1379 		if ((pr->ps_flags & PS_STOPPED) == 0)
1380 			continue;
1381 		atomic_clearbits_int(&pr->ps_flags, PS_STOPPED);
1382 
1383 		if ((pr->ps_pptr->ps_sigacts->ps_sigflags & SAS_NOCLDSTOP) == 0)
1384 			prsignal(pr->ps_pptr, SIGCHLD);
1385 		wakeup(pr->ps_pptr);
1386 	}
1387 }
1388 
1389 /*
1390  * Take the action for the specified signal
1391  * from the current set of pending signals.
1392  */
1393 void
1394 postsig(struct proc *p, int signum)
1395 {
1396 	struct process *pr = p->p_p;
1397 	struct sigacts *ps = pr->ps_sigacts;
1398 	sig_t action;
1399 	u_long trapno;
1400 	int mask, returnmask;
1401 	siginfo_t si;
1402 	union sigval sigval;
1403 	int s, code, info, onstack;
1404 
1405 	KASSERT(signum != 0);
1406 	KERNEL_ASSERT_LOCKED();
1407 
1408 	mask = sigmask(signum);
1409 	atomic_clearbits_int(&p->p_siglist, mask);
1410 	action = ps->ps_sigact[signum];
1411 	info = (ps->ps_siginfo & mask) != 0;
1412 	onstack = (ps->ps_sigonstack & mask) != 0;
1413 	sigval.sival_ptr = 0;
1414 
1415 	if (p->p_sisig != signum) {
1416 		trapno = 0;
1417 		code = SI_USER;
1418 		sigval.sival_ptr = 0;
1419 	} else {
1420 		trapno = p->p_sitrapno;
1421 		code = p->p_sicode;
1422 		sigval = p->p_sigval;
1423 	}
1424 	initsiginfo(&si, signum, trapno, code, sigval);
1425 
1426 #ifdef KTRACE
1427 	if (KTRPOINT(p, KTR_PSIG)) {
1428 		ktrpsig(p, signum, action, p->p_flag & P_SIGSUSPEND ?
1429 		    p->p_oldmask : p->p_sigmask, code, &si);
1430 	}
1431 #endif
1432 	if (action == SIG_DFL) {
1433 		/*
1434 		 * Default action, where the default is to kill
1435 		 * the process.  (Other cases were ignored above.)
1436 		 */
1437 		sigexit(p, signum);
1438 		/* NOTREACHED */
1439 	} else {
1440 		/*
1441 		 * If we get here, the signal must be caught.
1442 		 */
1443 #ifdef DIAGNOSTIC
1444 		if (action == SIG_IGN || (p->p_sigmask & mask))
1445 			panic("postsig action");
1446 #endif
1447 		/*
1448 		 * Set the new mask value and also defer further
1449 		 * occurrences of this signal.
1450 		 *
1451 		 * Special case: user has done a sigpause.  Here the
1452 		 * current mask is not of interest, but rather the
1453 		 * mask from before the sigpause is what we want
1454 		 * restored after the signal processing is completed.
1455 		 */
1456 #ifdef MULTIPROCESSOR
1457 		s = splsched();
1458 #else
1459 		s = splhigh();
1460 #endif
1461 		if (p->p_flag & P_SIGSUSPEND) {
1462 			atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
1463 			returnmask = p->p_oldmask;
1464 		} else {
1465 			returnmask = p->p_sigmask;
1466 		}
1467 		if (p->p_sisig == signum) {
1468 			p->p_sisig = 0;
1469 			p->p_sitrapno = 0;
1470 			p->p_sicode = SI_USER;
1471 			p->p_sigval.sival_ptr = NULL;
1472 		}
1473 
1474 		if (sendsig(action, signum, returnmask, &si, info, onstack)) {
1475 			sigexit(p, SIGILL);
1476 			/* NOTREACHED */
1477 		}
1478 		postsig_done(p, signum, ps);
1479 		splx(s);
1480 	}
1481 }
1482 
1483 /*
1484  * Force the current process to exit with the specified signal, dumping core
1485  * if appropriate.  We bypass the normal tests for masked and caught signals,
1486  * allowing unrecoverable failures to terminate the process without changing
1487  * signal state.  Mark the accounting record with the signal termination.
1488  * If dumping core, save the signal number for the debugger.  Calls exit and
1489  * does not return.
1490  */
1491 void
1492 sigexit(struct proc *p, int signum)
1493 {
1494 	/* Mark process as going away */
1495 	atomic_setbits_int(&p->p_flag, P_WEXIT);
1496 
1497 	p->p_p->ps_acflag |= AXSIG;
1498 	if (sigprop[signum] & SA_CORE) {
1499 		p->p_sisig = signum;
1500 
1501 		/* if there are other threads, pause them */
1502 		if (P_HASSIBLING(p))
1503 			single_thread_set(p, SINGLE_SUSPEND, 1);
1504 
1505 		if (coredump(p) == 0)
1506 			signum |= WCOREFLAG;
1507 	}
1508 	exit1(p, 0, signum, EXIT_NORMAL);
1509 	/* NOTREACHED */
1510 }
1511 
1512 /*
1513  * Send uncatchable SIGABRT for coredump.
1514  */
1515 void
1516 sigabort(struct proc *p)
1517 {
1518 	struct sigaction sa;
1519 
1520 	memset(&sa, 0, sizeof sa);
1521 	sa.sa_handler = SIG_DFL;
1522 	setsigvec(p, SIGABRT, &sa);
1523 	atomic_clearbits_int(&p->p_sigmask, sigmask(SIGABRT));
1524 	psignal(p, SIGABRT);
1525 }
1526 
1527 /*
1528  * Return 1 if `sig', a given signal, is ignored or masked for `p', a given
1529  * thread, and 0 otherwise.
1530  */
1531 int
1532 sigismasked(struct proc *p, int sig)
1533 {
1534 	struct process *pr = p->p_p;
1535 
1536 	if ((pr->ps_sigacts->ps_sigignore & sigmask(sig)) ||
1537 	    (p->p_sigmask & sigmask(sig)))
1538 	    	return 1;
1539 
1540 	return 0;
1541 }
1542 
1543 int nosuidcoredump = 1;
1544 
1545 struct coredump_iostate {
1546 	struct proc *io_proc;
1547 	struct vnode *io_vp;
1548 	struct ucred *io_cred;
1549 	off_t io_offset;
1550 };
1551 
1552 /*
1553  * Dump core, into a file named "progname.core", unless the process was
1554  * setuid/setgid.
1555  */
1556 int
1557 coredump(struct proc *p)
1558 {
1559 #ifdef SMALL_KERNEL
1560 	return EPERM;
1561 #else
1562 	struct process *pr = p->p_p;
1563 	struct vnode *vp;
1564 	struct ucred *cred = p->p_ucred;
1565 	struct vmspace *vm = p->p_vmspace;
1566 	struct nameidata nd;
1567 	struct vattr vattr;
1568 	struct coredump_iostate	io;
1569 	int error, len, incrash = 0;
1570 	char *name;
1571 	const char *dir = "/var/crash";
1572 
1573 	if (pr->ps_emul->e_coredump == NULL)
1574 		return (EINVAL);
1575 
1576 	atomic_setbits_int(&pr->ps_flags, PS_COREDUMP);
1577 
1578 	/* Don't dump if will exceed file size limit. */
1579 	if (USPACE + ptoa(vm->vm_dsize + vm->vm_ssize) >= lim_cur(RLIMIT_CORE))
1580 		return (EFBIG);
1581 
1582 	name = pool_get(&namei_pool, PR_WAITOK);
1583 
1584 	/*
1585 	 * If the process has inconsistent uids, nosuidcoredump
1586 	 * determines coredump placement policy.
1587 	 */
1588 	if (((pr->ps_flags & PS_SUGID) && (error = suser(p))) ||
1589 	   ((pr->ps_flags & PS_SUGID) && nosuidcoredump)) {
1590 		if (nosuidcoredump == 3) {
1591 			/*
1592 			 * If the program directory does not exist, dumps of
1593 			 * that core will silently fail.
1594 			 */
1595 			len = snprintf(name, MAXPATHLEN, "%s/%s/%u.core",
1596 			    dir, pr->ps_comm, pr->ps_pid);
1597 			incrash = KERNELPATH;
1598 		} else if (nosuidcoredump == 2) {
1599 			len = snprintf(name, MAXPATHLEN, "%s/%s.core",
1600 			    dir, pr->ps_comm);
1601 			incrash = KERNELPATH;
1602 		} else {
1603 			pool_put(&namei_pool, name);
1604 			return (EPERM);
1605 		}
1606 	} else
1607 		len = snprintf(name, MAXPATHLEN, "%s.core", pr->ps_comm);
1608 
1609 	if (len >= MAXPATHLEN) {
1610 		pool_put(&namei_pool, name);
1611 		return (EACCES);
1612 	}
1613 
1614 	/*
1615 	 * Control the UID used to write out.  The normal case uses
1616 	 * the real UID.  If the sugid case is going to write into the
1617 	 * controlled directory, we do so as root.
1618 	 */
1619 	if (incrash == 0) {
1620 		cred = crdup(cred);
1621 		cred->cr_uid = cred->cr_ruid;
1622 		cred->cr_gid = cred->cr_rgid;
1623 	} else {
1624 		if (p->p_fd->fd_rdir) {
1625 			vrele(p->p_fd->fd_rdir);
1626 			p->p_fd->fd_rdir = NULL;
1627 		}
1628 		p->p_ucred = crdup(p->p_ucred);
1629 		crfree(cred);
1630 		cred = p->p_ucred;
1631 		crhold(cred);
1632 		cred->cr_uid = 0;
1633 		cred->cr_gid = 0;
1634 	}
1635 
1636 	/* incrash should be 0 or KERNELPATH only */
1637 	NDINIT(&nd, 0, incrash, UIO_SYSSPACE, name, p);
1638 
1639 	error = vn_open(&nd, O_CREAT | FWRITE | O_NOFOLLOW | O_NONBLOCK,
1640 	    S_IRUSR | S_IWUSR);
1641 
1642 	if (error)
1643 		goto out;
1644 
1645 	/*
1646 	 * Don't dump to non-regular files, files with links, or files
1647 	 * owned by someone else.
1648 	 */
1649 	vp = nd.ni_vp;
1650 	if ((error = VOP_GETATTR(vp, &vattr, cred, p)) != 0) {
1651 		VOP_UNLOCK(vp);
1652 		vn_close(vp, FWRITE, cred, p);
1653 		goto out;
1654 	}
1655 	if (vp->v_type != VREG || vattr.va_nlink != 1 ||
1656 	    vattr.va_mode & ((VREAD | VWRITE) >> 3 | (VREAD | VWRITE) >> 6) ||
1657 	    vattr.va_uid != cred->cr_uid) {
1658 		error = EACCES;
1659 		VOP_UNLOCK(vp);
1660 		vn_close(vp, FWRITE, cred, p);
1661 		goto out;
1662 	}
1663 	VATTR_NULL(&vattr);
1664 	vattr.va_size = 0;
1665 	VOP_SETATTR(vp, &vattr, cred, p);
1666 	pr->ps_acflag |= ACORE;
1667 
1668 	io.io_proc = p;
1669 	io.io_vp = vp;
1670 	io.io_cred = cred;
1671 	io.io_offset = 0;
1672 	VOP_UNLOCK(vp);
1673 	vref(vp);
1674 	error = vn_close(vp, FWRITE, cred, p);
1675 	if (error == 0)
1676 		error = (*pr->ps_emul->e_coredump)(p, &io);
1677 	vrele(vp);
1678 out:
1679 	crfree(cred);
1680 	pool_put(&namei_pool, name);
1681 	return (error);
1682 #endif
1683 }
1684 
1685 #ifndef SMALL_KERNEL
1686 int
1687 coredump_write(void *cookie, enum uio_seg segflg, const void *data, size_t len)
1688 {
1689 	struct coredump_iostate *io = cookie;
1690 	off_t coffset = 0;
1691 	size_t csize;
1692 	int chunk, error;
1693 
1694 	csize = len;
1695 	do {
1696 		if (sigmask(SIGKILL) &
1697 		    (io->io_proc->p_siglist | io->io_proc->p_p->ps_siglist))
1698 			return (EINTR);
1699 
1700 		/* Rest of the loop sleeps with lock held, so... */
1701 		yield();
1702 
1703 		chunk = MIN(csize, MAXPHYS);
1704 		error = vn_rdwr(UIO_WRITE, io->io_vp,
1705 		    (caddr_t)data + coffset, chunk,
1706 		    io->io_offset + coffset, segflg,
1707 		    IO_UNIT, io->io_cred, NULL, io->io_proc);
1708 		if (error) {
1709 			struct process *pr = io->io_proc->p_p;
1710 
1711 			if (error == ENOSPC)
1712 				log(LOG_ERR,
1713 				    "coredump of %s(%d) failed, filesystem full\n",
1714 				    pr->ps_comm, pr->ps_pid);
1715 			else
1716 				log(LOG_ERR,
1717 				    "coredump of %s(%d), write failed: errno %d\n",
1718 				    pr->ps_comm, pr->ps_pid, error);
1719 			return (error);
1720 		}
1721 
1722 		coffset += chunk;
1723 		csize -= chunk;
1724 	} while (csize > 0);
1725 
1726 	io->io_offset += len;
1727 	return (0);
1728 }
1729 
1730 void
1731 coredump_unmap(void *cookie, vaddr_t start, vaddr_t end)
1732 {
1733 	struct coredump_iostate *io = cookie;
1734 
1735 	uvm_unmap(&io->io_proc->p_vmspace->vm_map, start, end);
1736 }
1737 
1738 #endif	/* !SMALL_KERNEL */
1739 
1740 /*
1741  * Nonexistent system call-- signal process (may want to handle it).
1742  * Flag error in case process won't see signal immediately (blocked or ignored).
1743  */
1744 int
1745 sys_nosys(struct proc *p, void *v, register_t *retval)
1746 {
1747 
1748 	ptsignal(p, SIGSYS, STHREAD);
1749 	return (ENOSYS);
1750 }
1751 
1752 int
1753 sys___thrsigdivert(struct proc *p, void *v, register_t *retval)
1754 {
1755 	static int sigwaitsleep;
1756 	struct sys___thrsigdivert_args /* {
1757 		syscallarg(sigset_t) sigmask;
1758 		syscallarg(siginfo_t *) info;
1759 		syscallarg(const struct timespec *) timeout;
1760 	} */ *uap = v;
1761 	sigset_t mask = SCARG(uap, sigmask) &~ sigcantmask;
1762 	siginfo_t si;
1763 	uint64_t nsecs = INFSLP;
1764 	int timeinvalid = 0;
1765 	int error = 0;
1766 
1767 	memset(&si, 0, sizeof(si));
1768 
1769 	if (SCARG(uap, timeout) != NULL) {
1770 		struct timespec ts;
1771 		if ((error = copyin(SCARG(uap, timeout), &ts, sizeof(ts))) != 0)
1772 			return (error);
1773 #ifdef KTRACE
1774 		if (KTRPOINT(p, KTR_STRUCT))
1775 			ktrreltimespec(p, &ts);
1776 #endif
1777 		if (!timespecisvalid(&ts))
1778 			timeinvalid = 1;
1779 		else
1780 			nsecs = TIMESPEC_TO_NSEC(&ts);
1781 	}
1782 
1783 	dosigsuspend(p, p->p_sigmask &~ mask);
1784 	for (;;) {
1785 		si.si_signo = cursig(p);
1786 		if (si.si_signo != 0) {
1787 			sigset_t smask = sigmask(si.si_signo);
1788 			if (smask & mask) {
1789 				atomic_clearbits_int(&p->p_siglist, smask);
1790 				error = 0;
1791 				break;
1792 			}
1793 		}
1794 
1795 		/* per-POSIX, delay this error until after the above */
1796 		if (timeinvalid)
1797 			error = EINVAL;
1798 		/* per-POSIX, return immediatly if timeout is zero-valued */
1799 		if (nsecs == 0)
1800 			error = EAGAIN;
1801 
1802 		if (error != 0)
1803 			break;
1804 
1805 		error = tsleep_nsec(&sigwaitsleep, PPAUSE|PCATCH, "sigwait",
1806 		    nsecs);
1807 	}
1808 
1809 	if (error == 0) {
1810 		*retval = si.si_signo;
1811 		if (SCARG(uap, info) != NULL)
1812 			error = copyout(&si, SCARG(uap, info), sizeof(si));
1813 	} else if (error == ERESTART && SCARG(uap, timeout) != NULL) {
1814 		/*
1815 		 * Restarting is wrong if there's a timeout, as it'll be
1816 		 * for the same interval again
1817 		 */
1818 		error = EINTR;
1819 	}
1820 
1821 	return (error);
1822 }
1823 
1824 void
1825 initsiginfo(siginfo_t *si, int sig, u_long trapno, int code, union sigval val)
1826 {
1827 	memset(si, 0, sizeof(*si));
1828 
1829 	si->si_signo = sig;
1830 	si->si_code = code;
1831 	if (code == SI_USER) {
1832 		si->si_value = val;
1833 	} else {
1834 		switch (sig) {
1835 		case SIGSEGV:
1836 		case SIGILL:
1837 		case SIGBUS:
1838 		case SIGFPE:
1839 			si->si_addr = val.sival_ptr;
1840 			si->si_trapno = trapno;
1841 			break;
1842 		case SIGXFSZ:
1843 			break;
1844 		}
1845 	}
1846 }
1847 
1848 int
1849 filt_sigattach(struct knote *kn)
1850 {
1851 	struct process *pr = curproc->p_p;
1852 	int s;
1853 
1854 	if (kn->kn_id >= NSIG)
1855 		return EINVAL;
1856 
1857 	kn->kn_ptr.p_process = pr;
1858 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
1859 
1860 	s = splhigh();
1861 	klist_insert_locked(&pr->ps_klist, kn);
1862 	splx(s);
1863 
1864 	return (0);
1865 }
1866 
1867 void
1868 filt_sigdetach(struct knote *kn)
1869 {
1870 	struct process *pr = kn->kn_ptr.p_process;
1871 	int s;
1872 
1873 	s = splhigh();
1874 	klist_remove_locked(&pr->ps_klist, kn);
1875 	splx(s);
1876 }
1877 
1878 /*
1879  * signal knotes are shared with proc knotes, so we apply a mask to
1880  * the hint in order to differentiate them from process hints.  This
1881  * could be avoided by using a signal-specific knote list, but probably
1882  * isn't worth the trouble.
1883  */
1884 int
1885 filt_signal(struct knote *kn, long hint)
1886 {
1887 
1888 	if (hint & NOTE_SIGNAL) {
1889 		hint &= ~NOTE_SIGNAL;
1890 
1891 		if (kn->kn_id == hint)
1892 			kn->kn_data++;
1893 	}
1894 	return (kn->kn_data != 0);
1895 }
1896 
1897 void
1898 userret(struct proc *p)
1899 {
1900 	int signum;
1901 
1902 	/* send SIGPROF or SIGVTALRM if their timers interrupted this thread */
1903 	if (p->p_flag & P_PROFPEND) {
1904 		atomic_clearbits_int(&p->p_flag, P_PROFPEND);
1905 		KERNEL_LOCK();
1906 		psignal(p, SIGPROF);
1907 		KERNEL_UNLOCK();
1908 	}
1909 	if (p->p_flag & P_ALRMPEND) {
1910 		atomic_clearbits_int(&p->p_flag, P_ALRMPEND);
1911 		KERNEL_LOCK();
1912 		psignal(p, SIGVTALRM);
1913 		KERNEL_UNLOCK();
1914 	}
1915 
1916 	if (SIGPENDING(p) != 0) {
1917 		KERNEL_LOCK();
1918 		while ((signum = cursig(p)) != 0)
1919 			postsig(p, signum);
1920 		KERNEL_UNLOCK();
1921 	}
1922 
1923 	/*
1924 	 * If P_SIGSUSPEND is still set here, then we still need to restore
1925 	 * the original sigmask before returning to userspace.  Also, this
1926 	 * might unmask some pending signals, so we need to check a second
1927 	 * time for signals to post.
1928 	 */
1929 	if (p->p_flag & P_SIGSUSPEND) {
1930 		atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
1931 		p->p_sigmask = p->p_oldmask;
1932 
1933 		KERNEL_LOCK();
1934 		while ((signum = cursig(p)) != 0)
1935 			postsig(p, signum);
1936 		KERNEL_UNLOCK();
1937 	}
1938 
1939 	if (p->p_flag & P_SUSPSINGLE)
1940 		single_thread_check(p, 0);
1941 
1942 	WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
1943 
1944 	p->p_cpu->ci_schedstate.spc_curpriority = p->p_usrpri;
1945 }
1946 
1947 int
1948 single_thread_check_locked(struct proc *p, int deep, int s)
1949 {
1950 	struct process *pr = p->p_p;
1951 
1952 	SCHED_ASSERT_LOCKED();
1953 
1954 	if (pr->ps_single != NULL && pr->ps_single != p) {
1955 		do {
1956 			/* if we're in deep, we need to unwind to the edge */
1957 			if (deep) {
1958 				if (pr->ps_flags & PS_SINGLEUNWIND)
1959 					return (ERESTART);
1960 				if (pr->ps_flags & PS_SINGLEEXIT)
1961 					return (EINTR);
1962 			}
1963 
1964 			if (pr->ps_single == NULL)
1965 				continue;
1966 
1967 			if (atomic_dec_int_nv(&pr->ps_singlecount) == 0)
1968 				wakeup(&pr->ps_singlecount);
1969 
1970 			if (pr->ps_flags & PS_SINGLEEXIT) {
1971 				SCHED_UNLOCK(s);
1972 				KERNEL_LOCK();
1973 				exit1(p, 0, 0, EXIT_THREAD_NOCHECK);
1974 				/* NOTREACHED */
1975 			}
1976 
1977 			/* not exiting and don't need to unwind, so suspend */
1978 			p->p_stat = SSTOP;
1979 			mi_switch();
1980 		} while (pr->ps_single != NULL);
1981 	}
1982 
1983 	return (0);
1984 }
1985 
1986 int
1987 single_thread_check(struct proc *p, int deep)
1988 {
1989 	int s, error;
1990 
1991 	SCHED_LOCK(s);
1992 	error = single_thread_check_locked(p, deep, s);
1993 	SCHED_UNLOCK(s);
1994 
1995 	return error;
1996 }
1997 
1998 /*
1999  * Stop other threads in the process.  The mode controls how and
2000  * where the other threads should stop:
2001  *  - SINGLE_SUSPEND: stop wherever they are, will later either be told to exit
2002  *    (by setting to SINGLE_EXIT) or be released (via single_thread_clear())
2003  *  - SINGLE_UNWIND: just unwind to kernel boundary, will be told to exit
2004  *    or released as with SINGLE_SUSPEND
2005  *  - SINGLE_EXIT: unwind to kernel boundary and exit
2006  */
2007 int
2008 single_thread_set(struct proc *p, enum single_thread_mode mode, int wait)
2009 {
2010 	struct process *pr = p->p_p;
2011 	struct proc *q;
2012 	int error, s;
2013 
2014 	KASSERT(curproc == p);
2015 
2016 	SCHED_LOCK(s);
2017 	error = single_thread_check_locked(p, (mode == SINGLE_UNWIND), s);
2018 	if (error) {
2019 		SCHED_UNLOCK(s);
2020 		return error;
2021 	}
2022 
2023 	switch (mode) {
2024 	case SINGLE_SUSPEND:
2025 		break;
2026 	case SINGLE_UNWIND:
2027 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
2028 		break;
2029 	case SINGLE_EXIT:
2030 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEEXIT);
2031 		atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
2032 		break;
2033 #ifdef DIAGNOSTIC
2034 	default:
2035 		panic("single_thread_mode = %d", mode);
2036 #endif
2037 	}
2038 	pr->ps_singlecount = 0;
2039 	membar_producer();
2040 	pr->ps_single = p;
2041 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
2042 		if (q == p)
2043 			continue;
2044 		if (q->p_flag & P_WEXIT) {
2045 			if (mode == SINGLE_EXIT) {
2046 				if (q->p_stat == SSTOP) {
2047 					setrunnable(q);
2048 					atomic_inc_int(&pr->ps_singlecount);
2049 				}
2050 			}
2051 			continue;
2052 		}
2053 		atomic_setbits_int(&q->p_flag, P_SUSPSINGLE);
2054 		switch (q->p_stat) {
2055 		case SIDL:
2056 		case SRUN:
2057 			atomic_inc_int(&pr->ps_singlecount);
2058 			break;
2059 		case SSLEEP:
2060 			/* if it's not interruptible, then just have to wait */
2061 			if (q->p_flag & P_SINTR) {
2062 				/* merely need to suspend?  just stop it */
2063 				if (mode == SINGLE_SUSPEND) {
2064 					q->p_stat = SSTOP;
2065 					break;
2066 				}
2067 				/* need to unwind or exit, so wake it */
2068 				setrunnable(q);
2069 			}
2070 			atomic_inc_int(&pr->ps_singlecount);
2071 			break;
2072 		case SSTOP:
2073 			if (mode == SINGLE_EXIT) {
2074 				setrunnable(q);
2075 				atomic_inc_int(&pr->ps_singlecount);
2076 			}
2077 			break;
2078 		case SDEAD:
2079 			break;
2080 		case SONPROC:
2081 			atomic_inc_int(&pr->ps_singlecount);
2082 			signotify(q);
2083 			break;
2084 		}
2085 	}
2086 	SCHED_UNLOCK(s);
2087 
2088 	if (wait)
2089 		single_thread_wait(pr, 1);
2090 
2091 	return 0;
2092 }
2093 
2094 /*
2095  * Wait for other threads to stop. If recheck is false then the function
2096  * returns non-zero if the caller needs to restart the check else 0 is
2097  * returned. If recheck is true the return value is always 0.
2098  */
2099 int
2100 single_thread_wait(struct process *pr, int recheck)
2101 {
2102 	struct sleep_state sls;
2103 	int wait;
2104 
2105 	/* wait until they're all suspended */
2106 	wait = pr->ps_singlecount > 0;
2107 	while (wait) {
2108 		sleep_setup(&sls, &pr->ps_singlecount, PWAIT, "suspend", 0);
2109 		wait = pr->ps_singlecount > 0;
2110 		sleep_finish(&sls, wait);
2111 		if (!recheck)
2112 			break;
2113 	}
2114 
2115 	return wait;
2116 }
2117 
2118 void
2119 single_thread_clear(struct proc *p, int flag)
2120 {
2121 	struct process *pr = p->p_p;
2122 	struct proc *q;
2123 	int s;
2124 
2125 	KASSERT(pr->ps_single == p);
2126 	KASSERT(curproc == p);
2127 
2128 	SCHED_LOCK(s);
2129 	pr->ps_single = NULL;
2130 	atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND | PS_SINGLEEXIT);
2131 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
2132 		if (q == p || (q->p_flag & P_SUSPSINGLE) == 0)
2133 			continue;
2134 		atomic_clearbits_int(&q->p_flag, P_SUSPSINGLE);
2135 
2136 		/*
2137 		 * if the thread was only stopped for single threading
2138 		 * then clearing that either makes it runnable or puts
2139 		 * it back into some sleep queue
2140 		 */
2141 		if (q->p_stat == SSTOP && (q->p_flag & flag) == 0) {
2142 			if (q->p_wchan == 0)
2143 				setrunnable(q);
2144 			else
2145 				q->p_stat = SSLEEP;
2146 		}
2147 	}
2148 	SCHED_UNLOCK(s);
2149 }
2150 
2151 void
2152 sigio_del(struct sigiolst *rmlist)
2153 {
2154 	struct sigio *sigio;
2155 
2156 	while ((sigio = LIST_FIRST(rmlist)) != NULL) {
2157 		LIST_REMOVE(sigio, sio_pgsigio);
2158 		crfree(sigio->sio_ucred);
2159 		free(sigio, M_SIGIO, sizeof(*sigio));
2160 	}
2161 }
2162 
2163 void
2164 sigio_unlink(struct sigio_ref *sir, struct sigiolst *rmlist)
2165 {
2166 	struct sigio *sigio;
2167 
2168 	MUTEX_ASSERT_LOCKED(&sigio_lock);
2169 
2170 	sigio = sir->sir_sigio;
2171 	if (sigio != NULL) {
2172 		KASSERT(sigio->sio_myref == sir);
2173 		sir->sir_sigio = NULL;
2174 
2175 		if (sigio->sio_pgid > 0)
2176 			sigio->sio_proc = NULL;
2177 		else
2178 			sigio->sio_pgrp = NULL;
2179 		LIST_REMOVE(sigio, sio_pgsigio);
2180 
2181 		LIST_INSERT_HEAD(rmlist, sigio, sio_pgsigio);
2182 	}
2183 }
2184 
2185 void
2186 sigio_free(struct sigio_ref *sir)
2187 {
2188 	struct sigiolst rmlist;
2189 
2190 	if (sir->sir_sigio == NULL)
2191 		return;
2192 
2193 	LIST_INIT(&rmlist);
2194 
2195 	mtx_enter(&sigio_lock);
2196 	sigio_unlink(sir, &rmlist);
2197 	mtx_leave(&sigio_lock);
2198 
2199 	sigio_del(&rmlist);
2200 }
2201 
2202 void
2203 sigio_freelist(struct sigiolst *sigiolst)
2204 {
2205 	struct sigiolst rmlist;
2206 	struct sigio *sigio;
2207 
2208 	if (LIST_EMPTY(sigiolst))
2209 		return;
2210 
2211 	LIST_INIT(&rmlist);
2212 
2213 	mtx_enter(&sigio_lock);
2214 	while ((sigio = LIST_FIRST(sigiolst)) != NULL)
2215 		sigio_unlink(sigio->sio_myref, &rmlist);
2216 	mtx_leave(&sigio_lock);
2217 
2218 	sigio_del(&rmlist);
2219 }
2220 
2221 int
2222 sigio_setown(struct sigio_ref *sir, u_long cmd, caddr_t data)
2223 {
2224 	struct sigiolst rmlist;
2225 	struct proc *p = curproc;
2226 	struct pgrp *pgrp = NULL;
2227 	struct process *pr = NULL;
2228 	struct sigio *sigio;
2229 	int error;
2230 	pid_t pgid = *(int *)data;
2231 
2232 	if (pgid == 0) {
2233 		sigio_free(sir);
2234 		return (0);
2235 	}
2236 
2237 	if (cmd == TIOCSPGRP) {
2238 		if (pgid < 0)
2239 			return (EINVAL);
2240 		pgid = -pgid;
2241 	}
2242 
2243 	sigio = malloc(sizeof(*sigio), M_SIGIO, M_WAITOK);
2244 	sigio->sio_pgid = pgid;
2245 	sigio->sio_ucred = crhold(p->p_ucred);
2246 	sigio->sio_myref = sir;
2247 
2248 	LIST_INIT(&rmlist);
2249 
2250 	/*
2251 	 * The kernel lock, and not sleeping between prfind()/pgfind() and
2252 	 * linking of the sigio ensure that the process or process group does
2253 	 * not disappear unexpectedly.
2254 	 */
2255 	KERNEL_LOCK();
2256 	mtx_enter(&sigio_lock);
2257 
2258 	if (pgid > 0) {
2259 		pr = prfind(pgid);
2260 		if (pr == NULL) {
2261 			error = ESRCH;
2262 			goto fail;
2263 		}
2264 
2265 		/*
2266 		 * Policy - Don't allow a process to FSETOWN a process
2267 		 * in another session.
2268 		 *
2269 		 * Remove this test to allow maximum flexibility or
2270 		 * restrict FSETOWN to the current process or process
2271 		 * group for maximum safety.
2272 		 */
2273 		if (pr->ps_session != p->p_p->ps_session) {
2274 			error = EPERM;
2275 			goto fail;
2276 		}
2277 
2278 		if ((pr->ps_flags & PS_EXITING) != 0) {
2279 			error = ESRCH;
2280 			goto fail;
2281 		}
2282 	} else /* if (pgid < 0) */ {
2283 		pgrp = pgfind(-pgid);
2284 		if (pgrp == NULL) {
2285 			error = ESRCH;
2286 			goto fail;
2287 		}
2288 
2289 		/*
2290 		 * Policy - Don't allow a process to FSETOWN a process
2291 		 * in another session.
2292 		 *
2293 		 * Remove this test to allow maximum flexibility or
2294 		 * restrict FSETOWN to the current process or process
2295 		 * group for maximum safety.
2296 		 */
2297 		if (pgrp->pg_session != p->p_p->ps_session) {
2298 			error = EPERM;
2299 			goto fail;
2300 		}
2301 	}
2302 
2303 	if (pgid > 0) {
2304 		sigio->sio_proc = pr;
2305 		LIST_INSERT_HEAD(&pr->ps_sigiolst, sigio, sio_pgsigio);
2306 	} else {
2307 		sigio->sio_pgrp = pgrp;
2308 		LIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
2309 	}
2310 
2311 	sigio_unlink(sir, &rmlist);
2312 	sir->sir_sigio = sigio;
2313 
2314 	mtx_leave(&sigio_lock);
2315 	KERNEL_UNLOCK();
2316 
2317 	sigio_del(&rmlist);
2318 
2319 	return (0);
2320 
2321 fail:
2322 	mtx_leave(&sigio_lock);
2323 	KERNEL_UNLOCK();
2324 
2325 	crfree(sigio->sio_ucred);
2326 	free(sigio, M_SIGIO, sizeof(*sigio));
2327 
2328 	return (error);
2329 }
2330 
2331 void
2332 sigio_getown(struct sigio_ref *sir, u_long cmd, caddr_t data)
2333 {
2334 	struct sigio *sigio;
2335 	pid_t pgid = 0;
2336 
2337 	mtx_enter(&sigio_lock);
2338 	sigio = sir->sir_sigio;
2339 	if (sigio != NULL)
2340 		pgid = sigio->sio_pgid;
2341 	mtx_leave(&sigio_lock);
2342 
2343 	if (cmd == TIOCGPGRP)
2344 		pgid = -pgid;
2345 
2346 	*(int *)data = pgid;
2347 }
2348 
2349 void
2350 sigio_copy(struct sigio_ref *dst, struct sigio_ref *src)
2351 {
2352 	struct sigiolst rmlist;
2353 	struct sigio *newsigio, *sigio;
2354 
2355 	sigio_free(dst);
2356 
2357 	if (src->sir_sigio == NULL)
2358 		return;
2359 
2360 	newsigio = malloc(sizeof(*newsigio), M_SIGIO, M_WAITOK);
2361 	LIST_INIT(&rmlist);
2362 
2363 	mtx_enter(&sigio_lock);
2364 
2365 	sigio = src->sir_sigio;
2366 	if (sigio == NULL) {
2367 		mtx_leave(&sigio_lock);
2368 		free(newsigio, M_SIGIO, sizeof(*newsigio));
2369 		return;
2370 	}
2371 
2372 	newsigio->sio_pgid = sigio->sio_pgid;
2373 	newsigio->sio_ucred = crhold(sigio->sio_ucred);
2374 	newsigio->sio_myref = dst;
2375 	if (newsigio->sio_pgid > 0) {
2376 		newsigio->sio_proc = sigio->sio_proc;
2377 		LIST_INSERT_HEAD(&newsigio->sio_proc->ps_sigiolst, newsigio,
2378 		    sio_pgsigio);
2379 	} else {
2380 		newsigio->sio_pgrp = sigio->sio_pgrp;
2381 		LIST_INSERT_HEAD(&newsigio->sio_pgrp->pg_sigiolst, newsigio,
2382 		    sio_pgsigio);
2383 	}
2384 
2385 	sigio_unlink(dst, &rmlist);
2386 	dst->sir_sigio = newsigio;
2387 
2388 	mtx_leave(&sigio_lock);
2389 
2390 	sigio_del(&rmlist);
2391 }
2392