xref: /openbsd-src/sys/kern/kern_sig.c (revision f90ef06a3045119dcc88b72d8b98ca60e3c00d5a)
1 /*	$OpenBSD: kern_sig.c,v 1.312 2023/08/13 15:53:31 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/queue.h>
44 #include <sys/namei.h>
45 #include <sys/vnode.h>
46 #include <sys/event.h>
47 #include <sys/proc.h>
48 #include <sys/systm.h>
49 #include <sys/acct.h>
50 #include <sys/fcntl.h>
51 #include <sys/filedesc.h>
52 #include <sys/wait.h>
53 #include <sys/ktrace.h>
54 #include <sys/stat.h>
55 #include <sys/malloc.h>
56 #include <sys/pool.h>
57 #include <sys/sched.h>
58 #include <sys/user.h>
59 #include <sys/syslog.h>
60 #include <sys/ttycom.h>
61 #include <sys/pledge.h>
62 #include <sys/witness.h>
63 #include <sys/exec_elf.h>
64 
65 #include <sys/mount.h>
66 #include <sys/syscallargs.h>
67 
68 #include <uvm/uvm_extern.h>
69 #include <machine/tcb.h>
70 
71 int nosuidcoredump = 1;
72 
73 int	filt_sigattach(struct knote *kn);
74 void	filt_sigdetach(struct knote *kn);
75 int	filt_signal(struct knote *kn, long hint);
76 
77 const struct filterops sig_filtops = {
78 	.f_flags	= 0,
79 	.f_attach	= filt_sigattach,
80 	.f_detach	= filt_sigdetach,
81 	.f_event	= filt_signal,
82 };
83 
84 /*
85  * The array below categorizes the signals and their default actions.
86  */
87 const int sigprop[NSIG] = {
88 	0,			/* unused */
89 	SA_KILL,		/* SIGHUP */
90 	SA_KILL,		/* SIGINT */
91 	SA_KILL|SA_CORE,	/* SIGQUIT */
92 	SA_KILL|SA_CORE,	/* SIGILL */
93 	SA_KILL|SA_CORE,	/* SIGTRAP */
94 	SA_KILL|SA_CORE,	/* SIGABRT */
95 	SA_KILL|SA_CORE,	/* SIGEMT */
96 	SA_KILL|SA_CORE,	/* SIGFPE */
97 	SA_KILL,		/* SIGKILL */
98 	SA_KILL|SA_CORE,	/* SIGBUS */
99 	SA_KILL|SA_CORE,	/* SIGSEGV */
100 	SA_KILL|SA_CORE,	/* SIGSYS */
101 	SA_KILL,		/* SIGPIPE */
102 	SA_KILL,		/* SIGALRM */
103 	SA_KILL,		/* SIGTERM */
104 	SA_IGNORE,		/* SIGURG */
105 	SA_STOP,		/* SIGSTOP */
106 	SA_STOP|SA_TTYSTOP,	/* SIGTSTP */
107 	SA_IGNORE|SA_CONT,	/* SIGCONT */
108 	SA_IGNORE,		/* SIGCHLD */
109 	SA_STOP|SA_TTYSTOP,	/* SIGTTIN */
110 	SA_STOP|SA_TTYSTOP,	/* SIGTTOU */
111 	SA_IGNORE,		/* SIGIO */
112 	SA_KILL,		/* SIGXCPU */
113 	SA_KILL,		/* SIGXFSZ */
114 	SA_KILL,		/* SIGVTALRM */
115 	SA_KILL,		/* SIGPROF */
116 	SA_IGNORE,		/* SIGWINCH  */
117 	SA_IGNORE,		/* SIGINFO */
118 	SA_KILL,		/* SIGUSR1 */
119 	SA_KILL,		/* SIGUSR2 */
120 	SA_IGNORE,		/* SIGTHR */
121 };
122 
123 #define	CONTSIGMASK	(sigmask(SIGCONT))
124 #define	STOPSIGMASK	(sigmask(SIGSTOP) | sigmask(SIGTSTP) | \
125 			    sigmask(SIGTTIN) | sigmask(SIGTTOU))
126 
127 void setsigvec(struct proc *, int, struct sigaction *);
128 
129 void proc_stop(struct proc *p, int);
130 void proc_stop_sweep(void *);
131 void *proc_stop_si;
132 
133 void setsigctx(struct proc *, int, struct sigctx *);
134 void postsig_done(struct proc *, int, sigset_t, int);
135 void postsig(struct proc *, int, struct sigctx *);
136 int cansignal(struct proc *, struct process *, int);
137 
138 struct pool sigacts_pool;	/* memory pool for sigacts structures */
139 
140 void sigio_del(struct sigiolst *);
141 void sigio_unlink(struct sigio_ref *, struct sigiolst *);
142 struct mutex sigio_lock = MUTEX_INITIALIZER(IPL_HIGH);
143 
144 /*
145  * Can thread p, send the signal signum to process qr?
146  */
147 int
148 cansignal(struct proc *p, struct process *qr, int signum)
149 {
150 	struct process *pr = p->p_p;
151 	struct ucred *uc = p->p_ucred;
152 	struct ucred *quc = qr->ps_ucred;
153 
154 	if (uc->cr_uid == 0)
155 		return (1);		/* root can always signal */
156 
157 	if (pr == qr)
158 		return (1);		/* process can always signal itself */
159 
160 	/* optimization: if the same creds then the tests below will pass */
161 	if (uc == quc)
162 		return (1);
163 
164 	if (signum == SIGCONT && qr->ps_session == pr->ps_session)
165 		return (1);		/* SIGCONT in session */
166 
167 	/*
168 	 * Using kill(), only certain signals can be sent to setugid
169 	 * child processes
170 	 */
171 	if (qr->ps_flags & PS_SUGID) {
172 		switch (signum) {
173 		case 0:
174 		case SIGKILL:
175 		case SIGINT:
176 		case SIGTERM:
177 		case SIGALRM:
178 		case SIGSTOP:
179 		case SIGTTIN:
180 		case SIGTTOU:
181 		case SIGTSTP:
182 		case SIGHUP:
183 		case SIGUSR1:
184 		case SIGUSR2:
185 			if (uc->cr_ruid == quc->cr_ruid ||
186 			    uc->cr_uid == quc->cr_ruid)
187 				return (1);
188 		}
189 		return (0);
190 	}
191 
192 	if (uc->cr_ruid == quc->cr_ruid ||
193 	    uc->cr_ruid == quc->cr_svuid ||
194 	    uc->cr_uid == quc->cr_ruid ||
195 	    uc->cr_uid == quc->cr_svuid)
196 		return (1);
197 	return (0);
198 }
199 
200 /*
201  * Initialize signal-related data structures.
202  */
203 void
204 signal_init(void)
205 {
206 	proc_stop_si = softintr_establish(IPL_SOFTCLOCK, proc_stop_sweep,
207 	    NULL);
208 	if (proc_stop_si == NULL)
209 		panic("signal_init failed to register softintr");
210 
211 	pool_init(&sigacts_pool, sizeof(struct sigacts), 0, IPL_NONE,
212 	    PR_WAITOK, "sigapl", NULL);
213 }
214 
215 /*
216  * Initialize a new sigaltstack structure.
217  */
218 void
219 sigstkinit(struct sigaltstack *ss)
220 {
221 	ss->ss_flags = SS_DISABLE;
222 	ss->ss_size = 0;
223 	ss->ss_sp = NULL;
224 }
225 
226 /*
227  * Create an initial sigacts structure, using the same signal state
228  * as pr.
229  */
230 struct sigacts *
231 sigactsinit(struct process *pr)
232 {
233 	struct sigacts *ps;
234 
235 	ps = pool_get(&sigacts_pool, PR_WAITOK);
236 	memcpy(ps, pr->ps_sigacts, sizeof(struct sigacts));
237 	return (ps);
238 }
239 
240 /*
241  * Release a sigacts structure.
242  */
243 void
244 sigactsfree(struct sigacts *ps)
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 		mtx_enter(&p->p_p->ps_mtx);
278 		sa->sa_handler = ps->ps_sigact[signum];
279 		sa->sa_mask = ps->ps_catchmask[signum];
280 		bit = sigmask(signum);
281 		sa->sa_flags = 0;
282 		if ((ps->ps_sigonstack & bit) != 0)
283 			sa->sa_flags |= SA_ONSTACK;
284 		if ((ps->ps_sigintr & bit) == 0)
285 			sa->sa_flags |= SA_RESTART;
286 		if ((ps->ps_sigreset & bit) != 0)
287 			sa->sa_flags |= SA_RESETHAND;
288 		if ((ps->ps_siginfo & bit) != 0)
289 			sa->sa_flags |= SA_SIGINFO;
290 		if (signum == SIGCHLD) {
291 			if ((ps->ps_sigflags & SAS_NOCLDSTOP) != 0)
292 				sa->sa_flags |= SA_NOCLDSTOP;
293 			if ((ps->ps_sigflags & SAS_NOCLDWAIT) != 0)
294 				sa->sa_flags |= SA_NOCLDWAIT;
295 		}
296 		mtx_leave(&p->p_p->ps_mtx);
297 		if ((sa->sa_mask & bit) == 0)
298 			sa->sa_flags |= SA_NODEFER;
299 		sa->sa_mask &= ~bit;
300 		error = copyout(sa, osa, sizeof (vec));
301 		if (error)
302 			return (error);
303 #ifdef KTRACE
304 		if (KTRPOINT(p, KTR_STRUCT))
305 			ovec = vec;
306 #endif
307 	}
308 	if (nsa) {
309 		error = copyin(nsa, sa, sizeof (vec));
310 		if (error)
311 			return (error);
312 #ifdef KTRACE
313 		if (KTRPOINT(p, KTR_STRUCT))
314 			ktrsigaction(p, sa);
315 #endif
316 		setsigvec(p, signum, sa);
317 	}
318 #ifdef KTRACE
319 	if (osa && KTRPOINT(p, KTR_STRUCT))
320 		ktrsigaction(p, &ovec);
321 #endif
322 	return (0);
323 }
324 
325 void
326 setsigvec(struct proc *p, int signum, struct sigaction *sa)
327 {
328 	struct sigacts *ps = p->p_p->ps_sigacts;
329 	int bit;
330 
331 	bit = sigmask(signum);
332 
333 	mtx_enter(&p->p_p->ps_mtx);
334 	ps->ps_sigact[signum] = sa->sa_handler;
335 	if ((sa->sa_flags & SA_NODEFER) == 0)
336 		sa->sa_mask |= sigmask(signum);
337 	ps->ps_catchmask[signum] = sa->sa_mask &~ sigcantmask;
338 	if (signum == SIGCHLD) {
339 		if (sa->sa_flags & SA_NOCLDSTOP)
340 			atomic_setbits_int(&ps->ps_sigflags, SAS_NOCLDSTOP);
341 		else
342 			atomic_clearbits_int(&ps->ps_sigflags, SAS_NOCLDSTOP);
343 		/*
344 		 * If the SA_NOCLDWAIT flag is set or the handler
345 		 * is SIG_IGN we reparent the dying child to PID 1
346 		 * (init) which will reap the zombie.  Because we use
347 		 * init to do our dirty work we never set SAS_NOCLDWAIT
348 		 * for PID 1.
349 		 * XXX exit1 rework means this is unnecessary?
350 		 */
351 		if (initprocess->ps_sigacts != ps &&
352 		    ((sa->sa_flags & SA_NOCLDWAIT) ||
353 		    sa->sa_handler == SIG_IGN))
354 			atomic_setbits_int(&ps->ps_sigflags, SAS_NOCLDWAIT);
355 		else
356 			atomic_clearbits_int(&ps->ps_sigflags, SAS_NOCLDWAIT);
357 	}
358 	if ((sa->sa_flags & SA_RESETHAND) != 0)
359 		ps->ps_sigreset |= bit;
360 	else
361 		ps->ps_sigreset &= ~bit;
362 	if ((sa->sa_flags & SA_SIGINFO) != 0)
363 		ps->ps_siginfo |= bit;
364 	else
365 		ps->ps_siginfo &= ~bit;
366 	if ((sa->sa_flags & SA_RESTART) == 0)
367 		ps->ps_sigintr |= bit;
368 	else
369 		ps->ps_sigintr &= ~bit;
370 	if ((sa->sa_flags & SA_ONSTACK) != 0)
371 		ps->ps_sigonstack |= bit;
372 	else
373 		ps->ps_sigonstack &= ~bit;
374 	/*
375 	 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
376 	 * and for signals set to SIG_DFL where the default is to ignore.
377 	 * However, don't put SIGCONT in ps_sigignore,
378 	 * as we have to restart the process.
379 	 */
380 	if (sa->sa_handler == SIG_IGN ||
381 	    (sigprop[signum] & SA_IGNORE && sa->sa_handler == SIG_DFL)) {
382 		atomic_clearbits_int(&p->p_siglist, bit);
383 		atomic_clearbits_int(&p->p_p->ps_siglist, bit);
384 		if (signum != SIGCONT)
385 			ps->ps_sigignore |= bit;	/* easier in psignal */
386 		ps->ps_sigcatch &= ~bit;
387 	} else {
388 		ps->ps_sigignore &= ~bit;
389 		if (sa->sa_handler == SIG_DFL)
390 			ps->ps_sigcatch &= ~bit;
391 		else
392 			ps->ps_sigcatch |= bit;
393 	}
394 	mtx_leave(&p->p_p->ps_mtx);
395 }
396 
397 /*
398  * Initialize signal state for process 0;
399  * set to ignore signals that are ignored by default.
400  */
401 void
402 siginit(struct sigacts *ps)
403 {
404 	int i;
405 
406 	for (i = 0; i < NSIG; i++)
407 		if (sigprop[i] & SA_IGNORE && i != SIGCONT)
408 			ps->ps_sigignore |= sigmask(i);
409 	ps->ps_sigflags = SAS_NOCLDWAIT | SAS_NOCLDSTOP;
410 }
411 
412 /*
413  * Reset signals for an exec by the specified thread.
414  */
415 void
416 execsigs(struct proc *p)
417 {
418 	struct sigacts *ps;
419 	int nc, mask;
420 
421 	ps = p->p_p->ps_sigacts;
422 	mtx_enter(&p->p_p->ps_mtx);
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 	mtx_leave(&p->p_p->ps_mtx);
450 }
451 
452 /*
453  * Manipulate signal mask.
454  * Note that we receive new mask, not pointer,
455  * and return old mask as return value;
456  * the library stub does the rest.
457  */
458 int
459 sys_sigprocmask(struct proc *p, void *v, register_t *retval)
460 {
461 	struct sys_sigprocmask_args /* {
462 		syscallarg(int) how;
463 		syscallarg(sigset_t) mask;
464 	} */ *uap = v;
465 	int error = 0;
466 	sigset_t mask;
467 
468 	KASSERT(p == curproc);
469 
470 	*retval = p->p_sigmask;
471 	mask = SCARG(uap, mask) &~ sigcantmask;
472 
473 	switch (SCARG(uap, how)) {
474 	case SIG_BLOCK:
475 		atomic_setbits_int(&p->p_sigmask, mask);
476 		break;
477 	case SIG_UNBLOCK:
478 		atomic_clearbits_int(&p->p_sigmask, mask);
479 		break;
480 	case SIG_SETMASK:
481 		p->p_sigmask = mask;
482 		break;
483 	default:
484 		error = EINVAL;
485 		break;
486 	}
487 	return (error);
488 }
489 
490 int
491 sys_sigpending(struct proc *p, void *v, register_t *retval)
492 {
493 	*retval = p->p_siglist | p->p_p->ps_siglist;
494 	return (0);
495 }
496 
497 /*
498  * Temporarily replace calling proc's signal mask for the duration of a
499  * system call.  Original signal mask will be restored by userret().
500  */
501 void
502 dosigsuspend(struct proc *p, sigset_t newmask)
503 {
504 	KASSERT(p == curproc);
505 
506 	p->p_oldmask = p->p_sigmask;
507 	atomic_setbits_int(&p->p_flag, P_SIGSUSPEND);
508 	p->p_sigmask = newmask;
509 }
510 
511 /*
512  * Suspend thread until signal, providing mask to be set
513  * in the meantime.  Note nonstandard calling convention:
514  * libc stub passes mask, not pointer, to save a copyin.
515  */
516 int
517 sys_sigsuspend(struct proc *p, void *v, register_t *retval)
518 {
519 	struct sys_sigsuspend_args /* {
520 		syscallarg(int) mask;
521 	} */ *uap = v;
522 
523 	dosigsuspend(p, SCARG(uap, mask) &~ sigcantmask);
524 	while (tsleep_nsec(&nowake, PPAUSE|PCATCH, "sigsusp", INFSLP) == 0)
525 		continue;
526 	/* always return EINTR rather than ERESTART... */
527 	return (EINTR);
528 }
529 
530 int
531 sigonstack(size_t stack)
532 {
533 	const struct sigaltstack *ss = &curproc->p_sigstk;
534 
535 	return (ss->ss_flags & SS_DISABLE ? 0 :
536 	    (stack - (size_t)ss->ss_sp < ss->ss_size));
537 }
538 
539 int
540 sys_sigaltstack(struct proc *p, void *v, register_t *retval)
541 {
542 	struct sys_sigaltstack_args /* {
543 		syscallarg(const struct sigaltstack *) nss;
544 		syscallarg(struct sigaltstack *) oss;
545 	} */ *uap = v;
546 	struct sigaltstack ss;
547 	const struct sigaltstack *nss;
548 	struct sigaltstack *oss;
549 	int onstack = sigonstack(PROC_STACK(p));
550 	int error;
551 
552 	nss = SCARG(uap, nss);
553 	oss = SCARG(uap, oss);
554 
555 	if (oss != NULL) {
556 		ss = p->p_sigstk;
557 		if (onstack)
558 			ss.ss_flags |= SS_ONSTACK;
559 		if ((error = copyout(&ss, oss, sizeof(ss))))
560 			return (error);
561 	}
562 	if (nss == NULL)
563 		return (0);
564 	error = copyin(nss, &ss, sizeof(ss));
565 	if (error)
566 		return (error);
567 	if (onstack)
568 		return (EPERM);
569 	if (ss.ss_flags & ~SS_DISABLE)
570 		return (EINVAL);
571 	if (ss.ss_flags & SS_DISABLE) {
572 		p->p_sigstk.ss_flags = ss.ss_flags;
573 		return (0);
574 	}
575 	if (ss.ss_size < MINSIGSTKSZ)
576 		return (ENOMEM);
577 
578 	error = uvm_map_remap_as_stack(p, (vaddr_t)ss.ss_sp, ss.ss_size);
579 	if (error)
580 		return (error);
581 
582 	p->p_sigstk = ss;
583 	return (0);
584 }
585 
586 int
587 sys_kill(struct proc *cp, void *v, register_t *retval)
588 {
589 	struct sys_kill_args /* {
590 		syscallarg(int) pid;
591 		syscallarg(int) signum;
592 	} */ *uap = v;
593 	struct process *pr;
594 	int pid = SCARG(uap, pid);
595 	int signum = SCARG(uap, signum);
596 	int error;
597 	int zombie = 0;
598 
599 	if ((error = pledge_kill(cp, pid)) != 0)
600 		return (error);
601 	if (((u_int)signum) >= NSIG)
602 		return (EINVAL);
603 	if (pid > 0) {
604 		if ((pr = prfind(pid)) == NULL) {
605 			if ((pr = zombiefind(pid)) == NULL)
606 				return (ESRCH);
607 			else
608 				zombie = 1;
609 		}
610 		if (!cansignal(cp, pr, signum))
611 			return (EPERM);
612 
613 		/* kill single process */
614 		if (signum && !zombie)
615 			prsignal(pr, signum);
616 		return (0);
617 	}
618 	switch (pid) {
619 	case -1:		/* broadcast signal */
620 		return (killpg1(cp, signum, 0, 1));
621 	case 0:			/* signal own process group */
622 		return (killpg1(cp, signum, 0, 0));
623 	default:		/* negative explicit process group */
624 		return (killpg1(cp, signum, -pid, 0));
625 	}
626 }
627 
628 int
629 sys_thrkill(struct proc *cp, void *v, register_t *retval)
630 {
631 	struct sys_thrkill_args /* {
632 		syscallarg(pid_t) tid;
633 		syscallarg(int) signum;
634 		syscallarg(void *) tcb;
635 	} */ *uap = v;
636 	struct proc *p;
637 	int tid = SCARG(uap, tid);
638 	int signum = SCARG(uap, signum);
639 	void *tcb;
640 
641 	if (((u_int)signum) >= NSIG)
642 		return (EINVAL);
643 
644 	p = tid ? tfind_user(tid, cp->p_p) : cp;
645 	if (p == NULL)
646 		return (ESRCH);
647 
648 	/* optionally require the target thread to have the given tcb addr */
649 	tcb = SCARG(uap, tcb);
650 	if (tcb != NULL && tcb != TCB_GET(p))
651 		return (ESRCH);
652 
653 	if (signum)
654 		ptsignal(p, signum, STHREAD);
655 	return (0);
656 }
657 
658 /*
659  * Common code for kill process group/broadcast kill.
660  * cp is calling process.
661  */
662 int
663 killpg1(struct proc *cp, int signum, int pgid, int all)
664 {
665 	struct process *pr;
666 	struct pgrp *pgrp;
667 	int nfound = 0;
668 
669 	if (all) {
670 		/*
671 		 * broadcast
672 		 */
673 		LIST_FOREACH(pr, &allprocess, ps_list) {
674 			if (pr->ps_pid <= 1 ||
675 			    pr->ps_flags & (PS_SYSTEM | PS_NOBROADCASTKILL) ||
676 			    pr == cp->p_p || !cansignal(cp, pr, signum))
677 				continue;
678 			nfound++;
679 			if (signum)
680 				prsignal(pr, signum);
681 		}
682 	} else {
683 		if (pgid == 0)
684 			/*
685 			 * zero pgid means send to my process group.
686 			 */
687 			pgrp = cp->p_p->ps_pgrp;
688 		else {
689 			pgrp = pgfind(pgid);
690 			if (pgrp == NULL)
691 				return (ESRCH);
692 		}
693 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist) {
694 			if (pr->ps_pid <= 1 || pr->ps_flags & PS_SYSTEM ||
695 			    !cansignal(cp, pr, signum))
696 				continue;
697 			nfound++;
698 			if (signum)
699 				prsignal(pr, signum);
700 		}
701 	}
702 	return (nfound ? 0 : ESRCH);
703 }
704 
705 #define CANDELIVER(uid, euid, pr) \
706 	(euid == 0 || \
707 	(uid) == (pr)->ps_ucred->cr_ruid || \
708 	(uid) == (pr)->ps_ucred->cr_svuid || \
709 	(uid) == (pr)->ps_ucred->cr_uid || \
710 	(euid) == (pr)->ps_ucred->cr_ruid || \
711 	(euid) == (pr)->ps_ucred->cr_svuid || \
712 	(euid) == (pr)->ps_ucred->cr_uid)
713 
714 #define CANSIGIO(cr, pr) \
715 	CANDELIVER((cr)->cr_ruid, (cr)->cr_uid, (pr))
716 
717 /*
718  * Send a signal to a process group.  If checktty is 1,
719  * limit to members which have a controlling terminal.
720  */
721 void
722 pgsignal(struct pgrp *pgrp, int signum, int checkctty)
723 {
724 	struct process *pr;
725 
726 	if (pgrp)
727 		LIST_FOREACH(pr, &pgrp->pg_members, ps_pglist)
728 			if (checkctty == 0 || pr->ps_flags & PS_CONTROLT)
729 				prsignal(pr, signum);
730 }
731 
732 /*
733  * Send a SIGIO or SIGURG signal to a process or process group using stored
734  * credentials rather than those of the current process.
735  */
736 void
737 pgsigio(struct sigio_ref *sir, int sig, int checkctty)
738 {
739 	struct process *pr;
740 	struct sigio *sigio;
741 
742 	if (sir->sir_sigio == NULL)
743 		return;
744 
745 	KERNEL_LOCK();
746 	mtx_enter(&sigio_lock);
747 	sigio = sir->sir_sigio;
748 	if (sigio == NULL)
749 		goto out;
750 	if (sigio->sio_pgid > 0) {
751 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc))
752 			prsignal(sigio->sio_proc, sig);
753 	} else if (sigio->sio_pgid < 0) {
754 		LIST_FOREACH(pr, &sigio->sio_pgrp->pg_members, ps_pglist) {
755 			if (CANSIGIO(sigio->sio_ucred, pr) &&
756 			    (checkctty == 0 || (pr->ps_flags & PS_CONTROLT)))
757 				prsignal(pr, sig);
758 		}
759 	}
760 out:
761 	mtx_leave(&sigio_lock);
762 	KERNEL_UNLOCK();
763 }
764 
765 /*
766  * Recalculate the signal mask and reset the signal disposition after
767  * usermode frame for delivery is formed.
768  */
769 void
770 postsig_done(struct proc *p, int signum, sigset_t catchmask, int reset)
771 {
772 	p->p_ru.ru_nsignals++;
773 	atomic_setbits_int(&p->p_sigmask, catchmask);
774 	if (reset != 0) {
775 		sigset_t mask = sigmask(signum);
776 		struct sigacts *ps = p->p_p->ps_sigacts;
777 
778 		mtx_enter(&p->p_p->ps_mtx);
779 		ps->ps_sigcatch &= ~mask;
780 		if (signum != SIGCONT && sigprop[signum] & SA_IGNORE)
781 			ps->ps_sigignore |= mask;
782 		ps->ps_sigact[signum] = SIG_DFL;
783 		mtx_leave(&p->p_p->ps_mtx);
784 	}
785 }
786 
787 /*
788  * Send a signal caused by a trap to the current thread
789  * If it will be caught immediately, deliver it with correct code.
790  * Otherwise, post it normally.
791  */
792 void
793 trapsignal(struct proc *p, int signum, u_long trapno, int code,
794     union sigval sigval)
795 {
796 	struct process *pr = p->p_p;
797 	struct sigctx ctx;
798 	int mask;
799 
800 	switch (signum) {
801 	case SIGILL:
802 	case SIGBUS:
803 	case SIGSEGV:
804 		pr->ps_acflag |= ATRAP;
805 		break;
806 	}
807 
808 	mask = sigmask(signum);
809 	setsigctx(p, signum, &ctx);
810 	if ((pr->ps_flags & PS_TRACED) == 0 && ctx.sig_catch != 0 &&
811 	    (p->p_sigmask & mask) == 0) {
812 		siginfo_t si;
813 
814 		initsiginfo(&si, signum, trapno, code, sigval);
815 #ifdef KTRACE
816 		if (KTRPOINT(p, KTR_PSIG)) {
817 			ktrpsig(p, signum, ctx.sig_action,
818 			    p->p_sigmask, code, &si);
819 		}
820 #endif
821 		if (sendsig(ctx.sig_action, signum, p->p_sigmask, &si,
822 		    ctx.sig_info, ctx.sig_onstack)) {
823 			KERNEL_LOCK();
824 			sigexit(p, SIGILL);
825 			/* NOTREACHED */
826 		}
827 		postsig_done(p, signum, ctx.sig_catchmask, ctx.sig_reset);
828 	} else {
829 		p->p_sisig = signum;
830 		p->p_sitrapno = trapno;	/* XXX for core dump/debugger */
831 		p->p_sicode = code;
832 		p->p_sigval = sigval;
833 
834 		/*
835 		 * If traced, stop if signal is masked, and stay stopped
836 		 * until released by the debugger.  If our parent process
837 		 * is waiting for us, don't hang as we could deadlock.
838 		 */
839 		if (((pr->ps_flags & (PS_TRACED | PS_PPWAIT)) == PS_TRACED) &&
840 		    signum != SIGKILL && (p->p_sigmask & mask) != 0) {
841 			int s;
842 
843 			single_thread_set(p, SINGLE_SUSPEND, 0);
844 			pr->ps_xsig = signum;
845 
846 			SCHED_LOCK(s);
847 			proc_stop(p, 1);
848 			SCHED_UNLOCK(s);
849 
850 			signum = pr->ps_xsig;
851 			single_thread_clear(p, 0);
852 
853 			/*
854 			 * If we are no longer being traced, or the parent
855 			 * didn't give us a signal, skip sending the signal.
856 			 */
857 			if ((pr->ps_flags & PS_TRACED) == 0 ||
858 			    signum == 0)
859 				return;
860 
861 			/* update signal info */
862 			p->p_sisig = signum;
863 			mask = sigmask(signum);
864 		}
865 
866 		/*
867 		 * Signals like SIGBUS and SIGSEGV should not, when
868 		 * generated by the kernel, be ignorable or blockable.
869 		 * If it is and we're not being traced, then just kill
870 		 * the process.
871 		 * After vfs_shutdown(9), init(8) cannot receive signals
872 		 * because new code pages of the signal handler cannot be
873 		 * mapped from halted storage.  init(8) may not die or the
874 		 * kernel panics.  Better loop between signal handler and
875 		 * page fault trap until the machine is halted.
876 		 */
877 		if ((pr->ps_flags & PS_TRACED) == 0 &&
878 		    (sigprop[signum] & SA_KILL) &&
879 		    ((p->p_sigmask & mask) || ctx.sig_ignore) &&
880 		    pr->ps_pid != 1) {
881 			KERNEL_LOCK();
882 			sigexit(p, signum);
883 			/* NOTREACHED */
884 		}
885 		KERNEL_LOCK();
886 		ptsignal(p, signum, STHREAD);
887 		KERNEL_UNLOCK();
888 	}
889 }
890 
891 /*
892  * Send the signal to the process.  If the signal has an action, the action
893  * is usually performed by the target process rather than the caller; we add
894  * the signal to the set of pending signals for the process.
895  *
896  * Exceptions:
897  *   o When a stop signal is sent to a sleeping process that takes the
898  *     default action, the process is stopped without awakening it.
899  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
900  *     regardless of the signal action (eg, blocked or ignored).
901  *
902  * Other ignored signals are discarded immediately.
903  */
904 void
905 psignal(struct proc *p, int signum)
906 {
907 	ptsignal(p, signum, SPROCESS);
908 }
909 
910 /*
911  * type = SPROCESS	process signal, can be diverted (sigwait())
912  * type = STHREAD	thread signal, but should be propagated if unhandled
913  * type = SPROPAGATED	propagated to this thread, so don't propagate again
914  */
915 void
916 ptsignal(struct proc *p, int signum, enum signal_type type)
917 {
918 	int s, prop;
919 	sig_t action;
920 	int mask;
921 	int *siglist;
922 	struct process *pr = p->p_p;
923 	struct proc *q;
924 	int wakeparent = 0;
925 
926 	KERNEL_ASSERT_LOCKED();
927 
928 #ifdef DIAGNOSTIC
929 	if ((u_int)signum >= NSIG || signum == 0)
930 		panic("psignal signal number");
931 #endif
932 
933 	/* Ignore signal if the target process is exiting */
934 	if (pr->ps_flags & PS_EXITING)
935 		return;
936 
937 	mask = sigmask(signum);
938 
939 	if (type == SPROCESS) {
940 		/* Accept SIGKILL to coredumping processes */
941 		if (pr->ps_flags & PS_COREDUMP && signum == SIGKILL) {
942 			atomic_setbits_int(&pr->ps_siglist, mask);
943 			return;
944 		}
945 
946 		/*
947 		 * If the current thread can process the signal
948 		 * immediately (it's unblocked) then have it take it.
949 		 */
950 		q = curproc;
951 		if (q != NULL && q->p_p == pr && (q->p_flag & P_WEXIT) == 0 &&
952 		    (q->p_sigmask & mask) == 0)
953 			p = q;
954 		else {
955 			/*
956 			 * A process-wide signal can be diverted to a
957 			 * different thread that's in sigwait() for this
958 			 * signal.  If there isn't such a thread, then
959 			 * pick a thread that doesn't have it blocked so
960 			 * that the stop/kill consideration isn't
961 			 * delayed.  Otherwise, mark it pending on the
962 			 * main thread.
963 			 */
964 			TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
965 				/* ignore exiting threads */
966 				if (q->p_flag & P_WEXIT)
967 					continue;
968 
969 				/* skip threads that have the signal blocked */
970 				if ((q->p_sigmask & mask) != 0)
971 					continue;
972 
973 				/* okay, could send to this thread */
974 				p = q;
975 
976 				/*
977 				 * sigsuspend, sigwait, ppoll/pselect, etc?
978 				 * Definitely go to this thread, as it's
979 				 * already blocked in the kernel.
980 				 */
981 				if (q->p_flag & P_SIGSUSPEND)
982 					break;
983 			}
984 		}
985 	}
986 
987 	if (type != SPROPAGATED)
988 		knote_locked(&pr->ps_klist, NOTE_SIGNAL | signum);
989 
990 	prop = sigprop[signum];
991 
992 	/*
993 	 * If proc is traced, always give parent a chance.
994 	 */
995 	if (pr->ps_flags & PS_TRACED) {
996 		action = SIG_DFL;
997 	} else {
998 		sigset_t sigcatch, sigignore;
999 
1000 		/*
1001 		 * If the signal is being ignored,
1002 		 * then we forget about it immediately.
1003 		 * (Note: we don't set SIGCONT in ps_sigignore,
1004 		 * and if it is set to SIG_IGN,
1005 		 * action will be SIG_DFL here.)
1006 		 */
1007 		mtx_enter(&pr->ps_mtx);
1008 		sigignore = pr->ps_sigacts->ps_sigignore;
1009 		sigcatch = pr->ps_sigacts->ps_sigcatch;
1010 		mtx_leave(&pr->ps_mtx);
1011 
1012 		if (sigignore & mask)
1013 			return;
1014 		if (p->p_sigmask & mask) {
1015 			action = SIG_HOLD;
1016 		} else if (sigcatch & mask) {
1017 			action = SIG_CATCH;
1018 		} else {
1019 			action = SIG_DFL;
1020 
1021 			if (prop & SA_KILL && pr->ps_nice > NZERO)
1022 				 pr->ps_nice = NZERO;
1023 
1024 			/*
1025 			 * If sending a tty stop signal to a member of an
1026 			 * orphaned process group, discard the signal here if
1027 			 * the action is default; don't stop the process below
1028 			 * if sleeping, and don't clear any pending SIGCONT.
1029 			 */
1030 			if (prop & SA_TTYSTOP && pr->ps_pgrp->pg_jobc == 0)
1031 				return;
1032 		}
1033 	}
1034 	/*
1035 	 * If delivered to process, mark as pending there.  Continue and stop
1036 	 * signals will be propagated to all threads.  So they are always
1037 	 * marked at thread level.
1038 	 */
1039 	siglist = (type == SPROCESS) ? &pr->ps_siglist : &p->p_siglist;
1040 	if (prop & SA_CONT) {
1041 		siglist = &p->p_siglist;
1042 		atomic_clearbits_int(siglist, STOPSIGMASK);
1043 	}
1044 	if (prop & SA_STOP) {
1045 		siglist = &p->p_siglist;
1046 		atomic_clearbits_int(siglist, CONTSIGMASK);
1047 		atomic_clearbits_int(&p->p_flag, P_CONTINUED);
1048 	}
1049 
1050 	/*
1051 	 * XXX delay processing of SA_STOP signals unless action == SIG_DFL?
1052 	 */
1053 	if (prop & (SA_CONT | SA_STOP) && type != SPROPAGATED)
1054 		TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link)
1055 			if (q != p)
1056 				ptsignal(q, signum, SPROPAGATED);
1057 
1058 	/*
1059 	 * Defer further processing for signals which are held,
1060 	 * except that stopped processes must be continued by SIGCONT.
1061 	 */
1062 	if (action == SIG_HOLD && ((prop & SA_CONT) == 0 ||
1063 	    p->p_stat != SSTOP)) {
1064 		atomic_setbits_int(siglist, mask);
1065 		return;
1066 	}
1067 
1068 	SCHED_LOCK(s);
1069 
1070 	switch (p->p_stat) {
1071 
1072 	case SSLEEP:
1073 		/*
1074 		 * If process is sleeping uninterruptibly
1075 		 * we can't interrupt the sleep... the signal will
1076 		 * be noticed when the process returns through
1077 		 * trap() or syscall().
1078 		 */
1079 		if ((p->p_flag & P_SINTR) == 0)
1080 			goto out;
1081 		/*
1082 		 * Process is sleeping and traced... make it runnable
1083 		 * so it can discover the signal in cursig() and stop
1084 		 * for the parent.
1085 		 */
1086 		if (pr->ps_flags & PS_TRACED)
1087 			goto run;
1088 		/*
1089 		 * If SIGCONT is default (or ignored) and process is
1090 		 * asleep, we are finished; the process should not
1091 		 * be awakened.
1092 		 */
1093 		if ((prop & SA_CONT) && action == SIG_DFL) {
1094 			mask = 0;
1095 			goto out;
1096 		}
1097 		/*
1098 		 * When a sleeping process receives a stop
1099 		 * signal, process immediately if possible.
1100 		 */
1101 		if ((prop & SA_STOP) && action == SIG_DFL) {
1102 			/*
1103 			 * If a child holding parent blocked,
1104 			 * stopping could cause deadlock.
1105 			 */
1106 			if (pr->ps_flags & PS_PPWAIT)
1107 				goto out;
1108 			mask = 0;
1109 			pr->ps_xsig = signum;
1110 			proc_stop(p, 0);
1111 			goto out;
1112 		}
1113 		/*
1114 		 * All other (caught or default) signals
1115 		 * cause the process to run.
1116 		 */
1117 		goto runfast;
1118 		/* NOTREACHED */
1119 
1120 	case SSTOP:
1121 		/*
1122 		 * If traced process is already stopped,
1123 		 * then no further action is necessary.
1124 		 */
1125 		if (pr->ps_flags & PS_TRACED)
1126 			goto out;
1127 
1128 		/*
1129 		 * Kill signal always sets processes running.
1130 		 */
1131 		if (signum == SIGKILL) {
1132 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1133 			goto runfast;
1134 		}
1135 
1136 		if (prop & SA_CONT) {
1137 			/*
1138 			 * If SIGCONT is default (or ignored), we continue the
1139 			 * process but don't leave the signal in p_siglist, as
1140 			 * it has no further action.  If SIGCONT is held, we
1141 			 * continue the process and leave the signal in
1142 			 * p_siglist.  If the process catches SIGCONT, let it
1143 			 * handle the signal itself.  If it isn't waiting on
1144 			 * an event, then it goes back to run state.
1145 			 * Otherwise, process goes back to sleep state.
1146 			 */
1147 			atomic_setbits_int(&p->p_flag, P_CONTINUED);
1148 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1149 			wakeparent = 1;
1150 			if (action == SIG_DFL)
1151 				atomic_clearbits_int(siglist, mask);
1152 			if (action == SIG_CATCH)
1153 				goto runfast;
1154 			if (p->p_wchan == NULL)
1155 				goto run;
1156 			atomic_clearbits_int(&p->p_flag, P_WSLEEP);
1157 			p->p_stat = SSLEEP;
1158 			goto out;
1159 		}
1160 
1161 		if (prop & SA_STOP) {
1162 			/*
1163 			 * Already stopped, don't need to stop again.
1164 			 * (If we did the shell could get confused.)
1165 			 */
1166 			mask = 0;
1167 			goto out;
1168 		}
1169 
1170 		/*
1171 		 * If process is sleeping interruptibly, then simulate a
1172 		 * wakeup so that when it is continued, it will be made
1173 		 * runnable and can look at the signal.  But don't make
1174 		 * the process runnable, leave it stopped.
1175 		 */
1176 		if (p->p_flag & P_SINTR)
1177 			unsleep(p);
1178 		goto out;
1179 
1180 	case SONPROC:
1181 		/* set siglist before issuing the ast */
1182 		atomic_setbits_int(siglist, mask);
1183 		mask = 0;
1184 		signotify(p);
1185 		/* FALLTHROUGH */
1186 	default:
1187 		/*
1188 		 * SRUN, SIDL, SDEAD do nothing with the signal,
1189 		 * other than kicking ourselves if we are running.
1190 		 * It will either never be noticed, or noticed very soon.
1191 		 */
1192 		goto out;
1193 	}
1194 	/* NOTREACHED */
1195 
1196 runfast:
1197 	/*
1198 	 * Raise priority to at least PUSER.
1199 	 */
1200 	if (p->p_usrpri > PUSER)
1201 		p->p_usrpri = PUSER;
1202 run:
1203 	setrunnable(p);
1204 out:
1205 	/* finally adjust siglist */
1206 	if (mask)
1207 		atomic_setbits_int(siglist, mask);
1208 	SCHED_UNLOCK(s);
1209 	if (wakeparent)
1210 		wakeup(pr->ps_pptr);
1211 }
1212 
1213 /* fill the signal context which should be used by postsig() and issignal() */
1214 void
1215 setsigctx(struct proc *p, int signum, struct sigctx *sctx)
1216 {
1217 	struct sigacts *ps = p->p_p->ps_sigacts;
1218 	sigset_t mask;
1219 
1220 	mtx_enter(&p->p_p->ps_mtx);
1221 	mask = sigmask(signum);
1222 	sctx->sig_action = ps->ps_sigact[signum];
1223 	sctx->sig_catchmask = ps->ps_catchmask[signum];
1224 	sctx->sig_reset = (ps->ps_sigreset & mask) != 0;
1225 	sctx->sig_info = (ps->ps_siginfo & mask) != 0;
1226 	sctx->sig_intr = (ps->ps_sigintr & mask) != 0;
1227 	sctx->sig_onstack = (ps->ps_sigonstack & mask) != 0;
1228 	sctx->sig_ignore = (ps->ps_sigignore & mask) != 0;
1229 	sctx->sig_catch = (ps->ps_sigcatch & mask) != 0;
1230 	mtx_leave(&p->p_p->ps_mtx);
1231 }
1232 
1233 /*
1234  * Determine signal that should be delivered to process p, the current
1235  * process, 0 if none.
1236  *
1237  * If the current process has received a signal (should be caught or cause
1238  * termination, should interrupt current syscall), return the signal number.
1239  * Stop signals with default action are processed immediately, then cleared;
1240  * they aren't returned.  This is checked after each entry to the system for
1241  * a syscall or trap. The normal call sequence is
1242  *
1243  *	while (signum = cursig(curproc, &ctx))
1244  *		postsig(signum, &ctx);
1245  *
1246  * Assumes that if the P_SINTR flag is set, we're holding both the
1247  * kernel and scheduler locks.
1248  */
1249 int
1250 cursig(struct proc *p, struct sigctx *sctx)
1251 {
1252 	struct process *pr = p->p_p;
1253 	int signum, mask, prop;
1254 	int dolock = (p->p_flag & P_SINTR) == 0;
1255 	sigset_t ps_siglist;
1256 	int s;
1257 
1258 	KASSERT(p == curproc);
1259 
1260 	for (;;) {
1261 		ps_siglist = READ_ONCE(pr->ps_siglist);
1262 		membar_consumer();
1263 		mask = SIGPENDING(p);
1264 		if (pr->ps_flags & PS_PPWAIT)
1265 			mask &= ~STOPSIGMASK;
1266 		if (mask == 0)	 	/* no signal to send */
1267 			return (0);
1268 		signum = ffs((long)mask);
1269 		mask = sigmask(signum);
1270 
1271 		/* take the signal! */
1272 		if (atomic_cas_uint(&pr->ps_siglist, ps_siglist,
1273 		    ps_siglist & ~mask) != ps_siglist) {
1274 			/* lost race taking the process signal, restart */
1275 			continue;
1276 		}
1277 		atomic_clearbits_int(&p->p_siglist, mask);
1278 		setsigctx(p, signum, sctx);
1279 
1280 		/*
1281 		 * We should see pending but ignored signals
1282 		 * only if PS_TRACED was on when they were posted.
1283 		 */
1284 		if (sctx->sig_ignore && (pr->ps_flags & PS_TRACED) == 0)
1285 			continue;
1286 
1287 		/*
1288 		 * If traced, always stop, and stay stopped until released
1289 		 * by the debugger.  If our parent process is waiting for
1290 		 * us, don't hang as we could deadlock.
1291 		 */
1292 		if (((pr->ps_flags & (PS_TRACED | PS_PPWAIT)) == PS_TRACED) &&
1293 		    signum != SIGKILL) {
1294 			single_thread_set(p, SINGLE_SUSPEND, 0);
1295 			pr->ps_xsig = signum;
1296 
1297 			if (dolock)
1298 				SCHED_LOCK(s);
1299 			proc_stop(p, 1);
1300 			if (dolock)
1301 				SCHED_UNLOCK(s);
1302 
1303 			/*
1304 			 * re-take the signal before releasing
1305 			 * the other threads. Must check the continue
1306 			 * conditions below and only take the signal if
1307 			 * those are not true.
1308 			 */
1309 			signum = pr->ps_xsig;
1310 			mask = sigmask(signum);
1311 			setsigctx(p, signum, sctx);
1312 			if (!((pr->ps_flags & PS_TRACED) == 0 ||
1313 			    signum == 0 ||
1314 			    (p->p_sigmask & mask) != 0)) {
1315 				atomic_clearbits_int(&p->p_siglist, mask);
1316 				atomic_clearbits_int(&pr->ps_siglist, mask);
1317 			}
1318 
1319 			single_thread_clear(p, 0);
1320 
1321 			/*
1322 			 * If we are no longer being traced, or the parent
1323 			 * didn't give us a signal, look for more signals.
1324 			 */
1325 			if ((pr->ps_flags & PS_TRACED) == 0 ||
1326 			    signum == 0)
1327 				continue;
1328 
1329 			/*
1330 			 * If the new signal is being masked, look for other
1331 			 * signals.
1332 			 */
1333 			if ((p->p_sigmask & mask) != 0)
1334 				continue;
1335 
1336 		}
1337 
1338 		prop = sigprop[signum];
1339 
1340 		/*
1341 		 * Decide whether the signal should be returned.
1342 		 * Return the signal's number, or fall through
1343 		 * to clear it from the pending mask.
1344 		 */
1345 		switch ((long)sctx->sig_action) {
1346 		case (long)SIG_DFL:
1347 			/*
1348 			 * Don't take default actions on system processes.
1349 			 */
1350 			if (pr->ps_pid <= 1) {
1351 #ifdef DIAGNOSTIC
1352 				/*
1353 				 * Are you sure you want to ignore SIGSEGV
1354 				 * in init? XXX
1355 				 */
1356 				printf("Process (pid %d) got signal"
1357 				    " %d\n", pr->ps_pid, signum);
1358 #endif
1359 				break;		/* == ignore */
1360 			}
1361 			/*
1362 			 * If there is a pending stop signal to process
1363 			 * with default action, stop here,
1364 			 * then clear the signal.  However,
1365 			 * if process is member of an orphaned
1366 			 * process group, ignore tty stop signals.
1367 			 */
1368 			if (prop & SA_STOP) {
1369 				if (pr->ps_flags & PS_TRACED ||
1370 		    		    (pr->ps_pgrp->pg_jobc == 0 &&
1371 				    prop & SA_TTYSTOP))
1372 					break;	/* == ignore */
1373 				pr->ps_xsig = signum;
1374 				if (dolock)
1375 					SCHED_LOCK(s);
1376 				proc_stop(p, 1);
1377 				if (dolock)
1378 					SCHED_UNLOCK(s);
1379 				break;
1380 			} else if (prop & SA_IGNORE) {
1381 				/*
1382 				 * Except for SIGCONT, shouldn't get here.
1383 				 * Default action is to ignore; drop it.
1384 				 */
1385 				break;		/* == ignore */
1386 			} else
1387 				goto keep;
1388 			/* NOTREACHED */
1389 		case (long)SIG_IGN:
1390 			/*
1391 			 * Masking above should prevent us ever trying
1392 			 * to take action on an ignored signal other
1393 			 * than SIGCONT, unless process is traced.
1394 			 */
1395 			if ((prop & SA_CONT) == 0 &&
1396 			    (pr->ps_flags & PS_TRACED) == 0)
1397 				printf("%s\n", __func__);
1398 			break;		/* == ignore */
1399 		default:
1400 			/*
1401 			 * This signal has an action, let
1402 			 * postsig() process it.
1403 			 */
1404 			goto keep;
1405 		}
1406 	}
1407 	/* NOTREACHED */
1408 
1409 keep:
1410 	atomic_setbits_int(&p->p_siglist, mask); /*leave the signal for later */
1411 	return (signum);
1412 }
1413 
1414 /*
1415  * Put the argument process into the stopped state and notify the parent
1416  * via wakeup.  Signals are handled elsewhere.  The process must not be
1417  * on the run queue.
1418  */
1419 void
1420 proc_stop(struct proc *p, int sw)
1421 {
1422 	struct process *pr = p->p_p;
1423 
1424 #ifdef MULTIPROCESSOR
1425 	SCHED_ASSERT_LOCKED();
1426 #endif
1427 
1428 	p->p_stat = SSTOP;
1429 	atomic_clearbits_int(&pr->ps_flags, PS_WAITED);
1430 	atomic_setbits_int(&pr->ps_flags, PS_STOPPED);
1431 	atomic_setbits_int(&p->p_flag, P_SUSPSIG);
1432 	/*
1433 	 * We need this soft interrupt to be handled fast.
1434 	 * Extra calls to softclock don't hurt.
1435 	 */
1436 	softintr_schedule(proc_stop_si);
1437 	if (sw)
1438 		mi_switch();
1439 }
1440 
1441 /*
1442  * Called from a soft interrupt to send signals to the parents of stopped
1443  * processes.
1444  * We can't do this in proc_stop because it's called with nasty locks held
1445  * and we would need recursive scheduler lock to deal with that.
1446  */
1447 void
1448 proc_stop_sweep(void *v)
1449 {
1450 	struct process *pr;
1451 
1452 	LIST_FOREACH(pr, &allprocess, ps_list) {
1453 		if ((pr->ps_flags & PS_STOPPED) == 0)
1454 			continue;
1455 		atomic_clearbits_int(&pr->ps_flags, PS_STOPPED);
1456 
1457 		if ((pr->ps_pptr->ps_sigacts->ps_sigflags & SAS_NOCLDSTOP) == 0)
1458 			prsignal(pr->ps_pptr, SIGCHLD);
1459 		wakeup(pr->ps_pptr);
1460 	}
1461 }
1462 
1463 /*
1464  * Take the action for the specified signal
1465  * from the current set of pending signals.
1466  */
1467 void
1468 postsig(struct proc *p, int signum, struct sigctx *sctx)
1469 {
1470 	u_long trapno;
1471 	int mask, returnmask;
1472 	siginfo_t si;
1473 	union sigval sigval;
1474 	int code;
1475 
1476 	KASSERT(signum != 0);
1477 
1478 	mask = sigmask(signum);
1479 	atomic_clearbits_int(&p->p_siglist, mask);
1480 	sigval.sival_ptr = NULL;
1481 
1482 	if (p->p_sisig != signum) {
1483 		trapno = 0;
1484 		code = SI_USER;
1485 		sigval.sival_ptr = NULL;
1486 	} else {
1487 		trapno = p->p_sitrapno;
1488 		code = p->p_sicode;
1489 		sigval = p->p_sigval;
1490 	}
1491 	initsiginfo(&si, signum, trapno, code, sigval);
1492 
1493 #ifdef KTRACE
1494 	if (KTRPOINT(p, KTR_PSIG)) {
1495 		ktrpsig(p, signum, sctx->sig_action, p->p_flag & P_SIGSUSPEND ?
1496 		    p->p_oldmask : p->p_sigmask, code, &si);
1497 	}
1498 #endif
1499 	if (sctx->sig_action == SIG_DFL) {
1500 		/*
1501 		 * Default action, where the default is to kill
1502 		 * the process.  (Other cases were ignored above.)
1503 		 */
1504 		KERNEL_LOCK();
1505 		sigexit(p, signum);
1506 		/* NOTREACHED */
1507 	} else {
1508 		/*
1509 		 * If we get here, the signal must be caught.
1510 		 */
1511 #ifdef DIAGNOSTIC
1512 		if (sctx->sig_action == SIG_IGN || (p->p_sigmask & mask))
1513 			panic("postsig action");
1514 #endif
1515 		/*
1516 		 * Set the new mask value and also defer further
1517 		 * occurrences of this signal.
1518 		 *
1519 		 * Special case: user has done a sigpause.  Here the
1520 		 * current mask is not of interest, but rather the
1521 		 * mask from before the sigpause is what we want
1522 		 * restored after the signal processing is completed.
1523 		 */
1524 		if (p->p_flag & P_SIGSUSPEND) {
1525 			atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
1526 			returnmask = p->p_oldmask;
1527 		} else {
1528 			returnmask = p->p_sigmask;
1529 		}
1530 		if (p->p_sisig == signum) {
1531 			p->p_sisig = 0;
1532 			p->p_sitrapno = 0;
1533 			p->p_sicode = SI_USER;
1534 			p->p_sigval.sival_ptr = NULL;
1535 		}
1536 
1537 		if (sendsig(sctx->sig_action, signum, returnmask, &si,
1538 		    sctx->sig_info, sctx->sig_onstack)) {
1539 			KERNEL_LOCK();
1540 			sigexit(p, SIGILL);
1541 			/* NOTREACHED */
1542 		}
1543 		postsig_done(p, signum, sctx->sig_catchmask, sctx->sig_reset);
1544 	}
1545 }
1546 
1547 /*
1548  * Force the current process to exit with the specified signal, dumping core
1549  * if appropriate.  We bypass the normal tests for masked and caught signals,
1550  * allowing unrecoverable failures to terminate the process without changing
1551  * signal state.  Mark the accounting record with the signal termination.
1552  * If dumping core, save the signal number for the debugger.  Calls exit and
1553  * does not return.
1554  */
1555 void
1556 sigexit(struct proc *p, int signum)
1557 {
1558 	/* Mark process as going away */
1559 	atomic_setbits_int(&p->p_flag, P_WEXIT);
1560 
1561 	p->p_p->ps_acflag |= AXSIG;
1562 	if (sigprop[signum] & SA_CORE) {
1563 		p->p_sisig = signum;
1564 
1565 		/* if there are other threads, pause them */
1566 		if (P_HASSIBLING(p))
1567 			single_thread_set(p, SINGLE_SUSPEND, 1);
1568 
1569 		if (coredump(p) == 0)
1570 			signum |= WCOREFLAG;
1571 	}
1572 	exit1(p, 0, signum, EXIT_NORMAL);
1573 	/* NOTREACHED */
1574 }
1575 
1576 /*
1577  * Send uncatchable SIGABRT for coredump.
1578  */
1579 void
1580 sigabort(struct proc *p)
1581 {
1582 	struct sigaction sa;
1583 
1584 	memset(&sa, 0, sizeof sa);
1585 	sa.sa_handler = SIG_DFL;
1586 	setsigvec(p, SIGABRT, &sa);
1587 	atomic_clearbits_int(&p->p_sigmask, sigmask(SIGABRT));
1588 	psignal(p, SIGABRT);
1589 }
1590 
1591 /*
1592  * Return 1 if `sig', a given signal, is ignored or masked for `p', a given
1593  * thread, and 0 otherwise.
1594  */
1595 int
1596 sigismasked(struct proc *p, int sig)
1597 {
1598 	struct process *pr = p->p_p;
1599 	int rv;
1600 
1601 	mtx_enter(&pr->ps_mtx);
1602 	rv = (pr->ps_sigacts->ps_sigignore & sigmask(sig)) ||
1603 	    (p->p_sigmask & sigmask(sig));
1604 	mtx_leave(&pr->ps_mtx);
1605 
1606 	return !!rv;
1607 }
1608 
1609 struct coredump_iostate {
1610 	struct proc *io_proc;
1611 	struct vnode *io_vp;
1612 	struct ucred *io_cred;
1613 	off_t io_offset;
1614 };
1615 
1616 /*
1617  * Dump core, into a file named "progname.core", unless the process was
1618  * setuid/setgid.
1619  */
1620 int
1621 coredump(struct proc *p)
1622 {
1623 #ifdef SMALL_KERNEL
1624 	return EPERM;
1625 #else
1626 	struct process *pr = p->p_p;
1627 	struct vnode *vp;
1628 	struct ucred *cred = p->p_ucred;
1629 	struct vmspace *vm = p->p_vmspace;
1630 	struct nameidata nd;
1631 	struct vattr vattr;
1632 	struct coredump_iostate	io;
1633 	int error, len, incrash = 0;
1634 	char *name;
1635 	const char *dir = "/var/crash";
1636 
1637 	atomic_setbits_int(&pr->ps_flags, PS_COREDUMP);
1638 
1639 #ifdef PMAP_CHECK_COPYIN
1640 	/* disable copyin checks, so we can write out text sections if needed */
1641 	p->p_vmspace->vm_map.check_copyin_count = 0;
1642 #endif
1643 
1644 	/* Don't dump if will exceed file size limit. */
1645 	if (USPACE + ptoa(vm->vm_dsize + vm->vm_ssize) >= lim_cur(RLIMIT_CORE))
1646 		return (EFBIG);
1647 
1648 	name = pool_get(&namei_pool, PR_WAITOK);
1649 
1650 	/*
1651 	 * If the process has inconsistent uids, nosuidcoredump
1652 	 * determines coredump placement policy.
1653 	 */
1654 	if (((pr->ps_flags & PS_SUGID) && (error = suser(p))) ||
1655 	   ((pr->ps_flags & PS_SUGID) && nosuidcoredump)) {
1656 		if (nosuidcoredump == 3) {
1657 			/*
1658 			 * If the program directory does not exist, dumps of
1659 			 * that core will silently fail.
1660 			 */
1661 			len = snprintf(name, MAXPATHLEN, "%s/%s/%u.core",
1662 			    dir, pr->ps_comm, pr->ps_pid);
1663 			incrash = KERNELPATH;
1664 		} else if (nosuidcoredump == 2) {
1665 			len = snprintf(name, MAXPATHLEN, "%s/%s.core",
1666 			    dir, pr->ps_comm);
1667 			incrash = KERNELPATH;
1668 		} else {
1669 			pool_put(&namei_pool, name);
1670 			return (EPERM);
1671 		}
1672 	} else
1673 		len = snprintf(name, MAXPATHLEN, "%s.core", pr->ps_comm);
1674 
1675 	if (len >= MAXPATHLEN) {
1676 		pool_put(&namei_pool, name);
1677 		return (EACCES);
1678 	}
1679 
1680 	/*
1681 	 * Control the UID used to write out.  The normal case uses
1682 	 * the real UID.  If the sugid case is going to write into the
1683 	 * controlled directory, we do so as root.
1684 	 */
1685 	if (incrash == 0) {
1686 		cred = crdup(cred);
1687 		cred->cr_uid = cred->cr_ruid;
1688 		cred->cr_gid = cred->cr_rgid;
1689 	} else {
1690 		if (p->p_fd->fd_rdir) {
1691 			vrele(p->p_fd->fd_rdir);
1692 			p->p_fd->fd_rdir = NULL;
1693 		}
1694 		p->p_ucred = crdup(p->p_ucred);
1695 		crfree(cred);
1696 		cred = p->p_ucred;
1697 		crhold(cred);
1698 		cred->cr_uid = 0;
1699 		cred->cr_gid = 0;
1700 	}
1701 
1702 	/* incrash should be 0 or KERNELPATH only */
1703 	NDINIT(&nd, 0, BYPASSUNVEIL | incrash, UIO_SYSSPACE, name, p);
1704 
1705 	error = vn_open(&nd, O_CREAT | FWRITE | O_NOFOLLOW | O_NONBLOCK,
1706 	    S_IRUSR | S_IWUSR);
1707 
1708 	if (error)
1709 		goto out;
1710 
1711 	/*
1712 	 * Don't dump to non-regular files, files with links, or files
1713 	 * owned by someone else.
1714 	 */
1715 	vp = nd.ni_vp;
1716 	if ((error = VOP_GETATTR(vp, &vattr, cred, p)) != 0) {
1717 		VOP_UNLOCK(vp);
1718 		vn_close(vp, FWRITE, cred, p);
1719 		goto out;
1720 	}
1721 	if (vp->v_type != VREG || vattr.va_nlink != 1 ||
1722 	    vattr.va_mode & ((VREAD | VWRITE) >> 3 | (VREAD | VWRITE) >> 6) ||
1723 	    vattr.va_uid != cred->cr_uid) {
1724 		error = EACCES;
1725 		VOP_UNLOCK(vp);
1726 		vn_close(vp, FWRITE, cred, p);
1727 		goto out;
1728 	}
1729 	VATTR_NULL(&vattr);
1730 	vattr.va_size = 0;
1731 	VOP_SETATTR(vp, &vattr, cred, p);
1732 	pr->ps_acflag |= ACORE;
1733 
1734 	io.io_proc = p;
1735 	io.io_vp = vp;
1736 	io.io_cred = cred;
1737 	io.io_offset = 0;
1738 	VOP_UNLOCK(vp);
1739 	vref(vp);
1740 	error = vn_close(vp, FWRITE, cred, p);
1741 	if (error == 0)
1742 		error = coredump_elf(p, &io);
1743 	vrele(vp);
1744 out:
1745 	crfree(cred);
1746 	pool_put(&namei_pool, name);
1747 	return (error);
1748 #endif
1749 }
1750 
1751 #ifndef SMALL_KERNEL
1752 int
1753 coredump_write(void *cookie, enum uio_seg segflg, const void *data, size_t len)
1754 {
1755 	struct coredump_iostate *io = cookie;
1756 	off_t coffset = 0;
1757 	size_t csize;
1758 	int chunk, error;
1759 
1760 	csize = len;
1761 	do {
1762 		if (sigmask(SIGKILL) &
1763 		    (io->io_proc->p_siglist | io->io_proc->p_p->ps_siglist))
1764 			return (EINTR);
1765 
1766 		/* Rest of the loop sleeps with lock held, so... */
1767 		yield();
1768 
1769 		chunk = MIN(csize, MAXPHYS);
1770 		error = vn_rdwr(UIO_WRITE, io->io_vp,
1771 		    (caddr_t)data + coffset, chunk,
1772 		    io->io_offset + coffset, segflg,
1773 		    IO_UNIT, io->io_cred, NULL, io->io_proc);
1774 		if (error) {
1775 			struct process *pr = io->io_proc->p_p;
1776 
1777 			if (error == ENOSPC)
1778 				log(LOG_ERR,
1779 				    "coredump of %s(%d) failed, filesystem full\n",
1780 				    pr->ps_comm, pr->ps_pid);
1781 			else
1782 				log(LOG_ERR,
1783 				    "coredump of %s(%d), write failed: errno %d\n",
1784 				    pr->ps_comm, pr->ps_pid, error);
1785 			return (error);
1786 		}
1787 
1788 		coffset += chunk;
1789 		csize -= chunk;
1790 	} while (csize > 0);
1791 
1792 	io->io_offset += len;
1793 	return (0);
1794 }
1795 
1796 void
1797 coredump_unmap(void *cookie, vaddr_t start, vaddr_t end)
1798 {
1799 	struct coredump_iostate *io = cookie;
1800 
1801 	uvm_unmap(&io->io_proc->p_vmspace->vm_map, start, end);
1802 }
1803 
1804 #endif	/* !SMALL_KERNEL */
1805 
1806 /*
1807  * Nonexistent system call-- signal process (may want to handle it).
1808  * Flag error in case process won't see signal immediately (blocked or ignored).
1809  */
1810 int
1811 sys_nosys(struct proc *p, void *v, register_t *retval)
1812 {
1813 	ptsignal(p, SIGSYS, STHREAD);
1814 	return (ENOSYS);
1815 }
1816 
1817 int
1818 sys___thrsigdivert(struct proc *p, void *v, register_t *retval)
1819 {
1820 	static int sigwaitsleep;
1821 	struct sys___thrsigdivert_args /* {
1822 		syscallarg(sigset_t) sigmask;
1823 		syscallarg(siginfo_t *) info;
1824 		syscallarg(const struct timespec *) timeout;
1825 	} */ *uap = v;
1826 	struct sigctx ctx;
1827 	sigset_t mask = SCARG(uap, sigmask) &~ sigcantmask;
1828 	siginfo_t si;
1829 	uint64_t nsecs = INFSLP;
1830 	int timeinvalid = 0;
1831 	int error = 0;
1832 
1833 	memset(&si, 0, sizeof(si));
1834 
1835 	if (SCARG(uap, timeout) != NULL) {
1836 		struct timespec ts;
1837 		if ((error = copyin(SCARG(uap, timeout), &ts, sizeof(ts))) != 0)
1838 			return (error);
1839 #ifdef KTRACE
1840 		if (KTRPOINT(p, KTR_STRUCT))
1841 			ktrreltimespec(p, &ts);
1842 #endif
1843 		if (!timespecisvalid(&ts))
1844 			timeinvalid = 1;
1845 		else
1846 			nsecs = TIMESPEC_TO_NSEC(&ts);
1847 	}
1848 
1849 	dosigsuspend(p, p->p_sigmask &~ mask);
1850 	for (;;) {
1851 		si.si_signo = cursig(p, &ctx);
1852 		if (si.si_signo != 0) {
1853 			sigset_t smask = sigmask(si.si_signo);
1854 			if (smask & mask) {
1855 				atomic_clearbits_int(&p->p_siglist, smask);
1856 				error = 0;
1857 				break;
1858 			}
1859 		}
1860 
1861 		/* per-POSIX, delay this error until after the above */
1862 		if (timeinvalid)
1863 			error = EINVAL;
1864 		/* per-POSIX, return immediately if timeout is zero-valued */
1865 		if (nsecs == 0)
1866 			error = EAGAIN;
1867 
1868 		if (error != 0)
1869 			break;
1870 
1871 		error = tsleep_nsec(&sigwaitsleep, PPAUSE|PCATCH, "sigwait",
1872 		    nsecs);
1873 	}
1874 
1875 	if (error == 0) {
1876 		*retval = si.si_signo;
1877 		if (SCARG(uap, info) != NULL) {
1878 			error = copyout(&si, SCARG(uap, info), sizeof(si));
1879 #ifdef KTRACE
1880 			if (error == 0 && KTRPOINT(p, KTR_STRUCT))
1881 				ktrsiginfo(p, &si);
1882 #endif
1883 		}
1884 	} else if (error == ERESTART && SCARG(uap, timeout) != NULL) {
1885 		/*
1886 		 * Restarting is wrong if there's a timeout, as it'll be
1887 		 * for the same interval again
1888 		 */
1889 		error = EINTR;
1890 	}
1891 
1892 	return (error);
1893 }
1894 
1895 void
1896 initsiginfo(siginfo_t *si, int sig, u_long trapno, int code, union sigval val)
1897 {
1898 	memset(si, 0, sizeof(*si));
1899 
1900 	si->si_signo = sig;
1901 	si->si_code = code;
1902 	if (code == SI_USER) {
1903 		si->si_value = val;
1904 	} else {
1905 		switch (sig) {
1906 		case SIGSEGV:
1907 		case SIGILL:
1908 		case SIGBUS:
1909 		case SIGFPE:
1910 			si->si_addr = val.sival_ptr;
1911 			si->si_trapno = trapno;
1912 			break;
1913 		case SIGXFSZ:
1914 			break;
1915 		}
1916 	}
1917 }
1918 
1919 int
1920 filt_sigattach(struct knote *kn)
1921 {
1922 	struct process *pr = curproc->p_p;
1923 	int s;
1924 
1925 	if (kn->kn_id >= NSIG)
1926 		return EINVAL;
1927 
1928 	kn->kn_ptr.p_process = pr;
1929 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
1930 
1931 	s = splhigh();
1932 	klist_insert_locked(&pr->ps_klist, kn);
1933 	splx(s);
1934 
1935 	return (0);
1936 }
1937 
1938 void
1939 filt_sigdetach(struct knote *kn)
1940 {
1941 	struct process *pr = kn->kn_ptr.p_process;
1942 	int s;
1943 
1944 	s = splhigh();
1945 	klist_remove_locked(&pr->ps_klist, kn);
1946 	splx(s);
1947 }
1948 
1949 /*
1950  * signal knotes are shared with proc knotes, so we apply a mask to
1951  * the hint in order to differentiate them from process hints.  This
1952  * could be avoided by using a signal-specific knote list, but probably
1953  * isn't worth the trouble.
1954  */
1955 int
1956 filt_signal(struct knote *kn, long hint)
1957 {
1958 
1959 	if (hint & NOTE_SIGNAL) {
1960 		hint &= ~NOTE_SIGNAL;
1961 
1962 		if (kn->kn_id == hint)
1963 			kn->kn_data++;
1964 	}
1965 	return (kn->kn_data != 0);
1966 }
1967 
1968 void
1969 userret(struct proc *p)
1970 {
1971 	struct sigctx ctx;
1972 	int signum;
1973 
1974 	if (p->p_flag & P_SUSPSINGLE)
1975 		single_thread_check(p, 0);
1976 
1977 	/* send SIGPROF or SIGVTALRM if their timers interrupted this thread */
1978 	if (p->p_flag & P_PROFPEND) {
1979 		atomic_clearbits_int(&p->p_flag, P_PROFPEND);
1980 		KERNEL_LOCK();
1981 		psignal(p, SIGPROF);
1982 		KERNEL_UNLOCK();
1983 	}
1984 	if (p->p_flag & P_ALRMPEND) {
1985 		atomic_clearbits_int(&p->p_flag, P_ALRMPEND);
1986 		KERNEL_LOCK();
1987 		psignal(p, SIGVTALRM);
1988 		KERNEL_UNLOCK();
1989 	}
1990 
1991 	if (SIGPENDING(p) != 0) {
1992 		while ((signum = cursig(p, &ctx)) != 0)
1993 			postsig(p, signum, &ctx);
1994 	}
1995 
1996 	/*
1997 	 * If P_SIGSUSPEND is still set here, then we still need to restore
1998 	 * the original sigmask before returning to userspace.  Also, this
1999 	 * might unmask some pending signals, so we need to check a second
2000 	 * time for signals to post.
2001 	 */
2002 	if (p->p_flag & P_SIGSUSPEND) {
2003 		atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
2004 		p->p_sigmask = p->p_oldmask;
2005 
2006 		while ((signum = cursig(p, &ctx)) != 0)
2007 			postsig(p, signum, &ctx);
2008 	}
2009 
2010 	WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
2011 
2012 	p->p_cpu->ci_schedstate.spc_curpriority = p->p_usrpri;
2013 }
2014 
2015 int
2016 single_thread_check_locked(struct proc *p, int deep, int s)
2017 {
2018 	struct process *pr = p->p_p;
2019 
2020 	SCHED_ASSERT_LOCKED();
2021 
2022 	if (pr->ps_single == NULL || pr->ps_single == p)
2023 		return (0);
2024 
2025 	do {
2026 		/* if we're in deep, we need to unwind to the edge */
2027 		if (deep) {
2028 			if (pr->ps_flags & PS_SINGLEUNWIND)
2029 				return (ERESTART);
2030 			if (pr->ps_flags & PS_SINGLEEXIT)
2031 				return (EINTR);
2032 		}
2033 
2034 		if (atomic_dec_int_nv(&pr->ps_singlecount) == 0)
2035 			wakeup(&pr->ps_singlecount);
2036 
2037 		if (pr->ps_flags & PS_SINGLEEXIT) {
2038 			SCHED_UNLOCK(s);
2039 			KERNEL_LOCK();
2040 			exit1(p, 0, 0, EXIT_THREAD_NOCHECK);
2041 			/* NOTREACHED */
2042 		}
2043 
2044 		/* not exiting and don't need to unwind, so suspend */
2045 		p->p_stat = SSTOP;
2046 		mi_switch();
2047 	} while (pr->ps_single != NULL);
2048 
2049 	return (0);
2050 }
2051 
2052 int
2053 single_thread_check(struct proc *p, int deep)
2054 {
2055 	int s, error;
2056 
2057 	SCHED_LOCK(s);
2058 	error = single_thread_check_locked(p, deep, s);
2059 	SCHED_UNLOCK(s);
2060 
2061 	return error;
2062 }
2063 
2064 /*
2065  * Stop other threads in the process.  The mode controls how and
2066  * where the other threads should stop:
2067  *  - SINGLE_SUSPEND: stop wherever they are, will later either be told to exit
2068  *    (by setting to SINGLE_EXIT) or be released (via single_thread_clear())
2069  *  - SINGLE_UNWIND: just unwind to kernel boundary, will be told to exit
2070  *    or released as with SINGLE_SUSPEND
2071  *  - SINGLE_EXIT: unwind to kernel boundary and exit
2072  */
2073 int
2074 single_thread_set(struct proc *p, enum single_thread_mode mode, int wait)
2075 {
2076 	struct process *pr = p->p_p;
2077 	struct proc *q;
2078 	int error, s;
2079 
2080 	KASSERT(curproc == p);
2081 
2082 	SCHED_LOCK(s);
2083 	error = single_thread_check_locked(p, (mode == SINGLE_UNWIND), s);
2084 	if (error) {
2085 		SCHED_UNLOCK(s);
2086 		return error;
2087 	}
2088 
2089 	switch (mode) {
2090 	case SINGLE_SUSPEND:
2091 		break;
2092 	case SINGLE_UNWIND:
2093 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
2094 		break;
2095 	case SINGLE_EXIT:
2096 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEEXIT);
2097 		atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
2098 		break;
2099 #ifdef DIAGNOSTIC
2100 	default:
2101 		panic("single_thread_mode = %d", mode);
2102 #endif
2103 	}
2104 	pr->ps_singlecount = 0;
2105 	membar_producer();
2106 	pr->ps_single = p;
2107 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
2108 		if (q == p)
2109 			continue;
2110 		if (q->p_flag & P_WEXIT) {
2111 			if (mode == SINGLE_EXIT) {
2112 				if (q->p_stat == SSTOP) {
2113 					setrunnable(q);
2114 					atomic_inc_int(&pr->ps_singlecount);
2115 				}
2116 			}
2117 			continue;
2118 		}
2119 		atomic_setbits_int(&q->p_flag, P_SUSPSINGLE);
2120 		switch (q->p_stat) {
2121 		case SIDL:
2122 		case SRUN:
2123 			atomic_inc_int(&pr->ps_singlecount);
2124 			break;
2125 		case SSLEEP:
2126 			/* if it's not interruptible, then just have to wait */
2127 			if (q->p_flag & P_SINTR) {
2128 				/* merely need to suspend?  just stop it */
2129 				if (mode == SINGLE_SUSPEND) {
2130 					q->p_stat = SSTOP;
2131 					break;
2132 				}
2133 				/* need to unwind or exit, so wake it */
2134 				setrunnable(q);
2135 			}
2136 			atomic_inc_int(&pr->ps_singlecount);
2137 			break;
2138 		case SSTOP:
2139 			if (mode == SINGLE_EXIT) {
2140 				setrunnable(q);
2141 				atomic_inc_int(&pr->ps_singlecount);
2142 			}
2143 			break;
2144 		case SDEAD:
2145 			break;
2146 		case SONPROC:
2147 			atomic_inc_int(&pr->ps_singlecount);
2148 			signotify(q);
2149 			break;
2150 		}
2151 	}
2152 	SCHED_UNLOCK(s);
2153 
2154 	if (wait)
2155 		single_thread_wait(pr, 1);
2156 
2157 	return 0;
2158 }
2159 
2160 /*
2161  * Wait for other threads to stop. If recheck is false then the function
2162  * returns non-zero if the caller needs to restart the check else 0 is
2163  * returned. If recheck is true the return value is always 0.
2164  */
2165 int
2166 single_thread_wait(struct process *pr, int recheck)
2167 {
2168 	int wait;
2169 
2170 	/* wait until they're all suspended */
2171 	wait = pr->ps_singlecount > 0;
2172 	while (wait) {
2173 		sleep_setup(&pr->ps_singlecount, PWAIT, "suspend");
2174 		wait = pr->ps_singlecount > 0;
2175 		sleep_finish(0, wait);
2176 		if (!recheck)
2177 			break;
2178 	}
2179 
2180 	return wait;
2181 }
2182 
2183 void
2184 single_thread_clear(struct proc *p, int flag)
2185 {
2186 	struct process *pr = p->p_p;
2187 	struct proc *q;
2188 	int s;
2189 
2190 	KASSERT(pr->ps_single == p);
2191 	KASSERT(curproc == p);
2192 
2193 	SCHED_LOCK(s);
2194 	pr->ps_single = NULL;
2195 	atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND | PS_SINGLEEXIT);
2196 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
2197 		if (q == p || (q->p_flag & P_SUSPSINGLE) == 0)
2198 			continue;
2199 		atomic_clearbits_int(&q->p_flag, P_SUSPSINGLE);
2200 
2201 		/*
2202 		 * if the thread was only stopped for single threading
2203 		 * then clearing that either makes it runnable or puts
2204 		 * it back into some sleep queue
2205 		 */
2206 		if (q->p_stat == SSTOP && (q->p_flag & flag) == 0) {
2207 			if (q->p_wchan == NULL)
2208 				setrunnable(q);
2209 			else {
2210 				atomic_clearbits_int(&q->p_flag, P_WSLEEP);
2211 				q->p_stat = SSLEEP;
2212 			}
2213 		}
2214 	}
2215 	SCHED_UNLOCK(s);
2216 }
2217 
2218 void
2219 sigio_del(struct sigiolst *rmlist)
2220 {
2221 	struct sigio *sigio;
2222 
2223 	while ((sigio = LIST_FIRST(rmlist)) != NULL) {
2224 		LIST_REMOVE(sigio, sio_pgsigio);
2225 		crfree(sigio->sio_ucred);
2226 		free(sigio, M_SIGIO, sizeof(*sigio));
2227 	}
2228 }
2229 
2230 void
2231 sigio_unlink(struct sigio_ref *sir, struct sigiolst *rmlist)
2232 {
2233 	struct sigio *sigio;
2234 
2235 	MUTEX_ASSERT_LOCKED(&sigio_lock);
2236 
2237 	sigio = sir->sir_sigio;
2238 	if (sigio != NULL) {
2239 		KASSERT(sigio->sio_myref == sir);
2240 		sir->sir_sigio = NULL;
2241 
2242 		if (sigio->sio_pgid > 0)
2243 			sigio->sio_proc = NULL;
2244 		else
2245 			sigio->sio_pgrp = NULL;
2246 		LIST_REMOVE(sigio, sio_pgsigio);
2247 
2248 		LIST_INSERT_HEAD(rmlist, sigio, sio_pgsigio);
2249 	}
2250 }
2251 
2252 void
2253 sigio_free(struct sigio_ref *sir)
2254 {
2255 	struct sigiolst rmlist;
2256 
2257 	if (sir->sir_sigio == NULL)
2258 		return;
2259 
2260 	LIST_INIT(&rmlist);
2261 
2262 	mtx_enter(&sigio_lock);
2263 	sigio_unlink(sir, &rmlist);
2264 	mtx_leave(&sigio_lock);
2265 
2266 	sigio_del(&rmlist);
2267 }
2268 
2269 void
2270 sigio_freelist(struct sigiolst *sigiolst)
2271 {
2272 	struct sigiolst rmlist;
2273 	struct sigio *sigio;
2274 
2275 	if (LIST_EMPTY(sigiolst))
2276 		return;
2277 
2278 	LIST_INIT(&rmlist);
2279 
2280 	mtx_enter(&sigio_lock);
2281 	while ((sigio = LIST_FIRST(sigiolst)) != NULL)
2282 		sigio_unlink(sigio->sio_myref, &rmlist);
2283 	mtx_leave(&sigio_lock);
2284 
2285 	sigio_del(&rmlist);
2286 }
2287 
2288 int
2289 sigio_setown(struct sigio_ref *sir, u_long cmd, caddr_t data)
2290 {
2291 	struct sigiolst rmlist;
2292 	struct proc *p = curproc;
2293 	struct pgrp *pgrp = NULL;
2294 	struct process *pr = NULL;
2295 	struct sigio *sigio;
2296 	int error;
2297 	pid_t pgid = *(int *)data;
2298 
2299 	if (pgid == 0) {
2300 		sigio_free(sir);
2301 		return (0);
2302 	}
2303 
2304 	if (cmd == TIOCSPGRP) {
2305 		if (pgid < 0)
2306 			return (EINVAL);
2307 		pgid = -pgid;
2308 	}
2309 
2310 	sigio = malloc(sizeof(*sigio), M_SIGIO, M_WAITOK);
2311 	sigio->sio_pgid = pgid;
2312 	sigio->sio_ucred = crhold(p->p_ucred);
2313 	sigio->sio_myref = sir;
2314 
2315 	LIST_INIT(&rmlist);
2316 
2317 	/*
2318 	 * The kernel lock, and not sleeping between prfind()/pgfind() and
2319 	 * linking of the sigio ensure that the process or process group does
2320 	 * not disappear unexpectedly.
2321 	 */
2322 	KERNEL_LOCK();
2323 	mtx_enter(&sigio_lock);
2324 
2325 	if (pgid > 0) {
2326 		pr = prfind(pgid);
2327 		if (pr == NULL) {
2328 			error = ESRCH;
2329 			goto fail;
2330 		}
2331 
2332 		/*
2333 		 * Policy - Don't allow a process to FSETOWN a process
2334 		 * in another session.
2335 		 *
2336 		 * Remove this test to allow maximum flexibility or
2337 		 * restrict FSETOWN to the current process or process
2338 		 * group for maximum safety.
2339 		 */
2340 		if (pr->ps_session != p->p_p->ps_session) {
2341 			error = EPERM;
2342 			goto fail;
2343 		}
2344 
2345 		if ((pr->ps_flags & PS_EXITING) != 0) {
2346 			error = ESRCH;
2347 			goto fail;
2348 		}
2349 	} else /* if (pgid < 0) */ {
2350 		pgrp = pgfind(-pgid);
2351 		if (pgrp == NULL) {
2352 			error = ESRCH;
2353 			goto fail;
2354 		}
2355 
2356 		/*
2357 		 * Policy - Don't allow a process to FSETOWN a process
2358 		 * in another session.
2359 		 *
2360 		 * Remove this test to allow maximum flexibility or
2361 		 * restrict FSETOWN to the current process or process
2362 		 * group for maximum safety.
2363 		 */
2364 		if (pgrp->pg_session != p->p_p->ps_session) {
2365 			error = EPERM;
2366 			goto fail;
2367 		}
2368 	}
2369 
2370 	if (pgid > 0) {
2371 		sigio->sio_proc = pr;
2372 		LIST_INSERT_HEAD(&pr->ps_sigiolst, sigio, sio_pgsigio);
2373 	} else {
2374 		sigio->sio_pgrp = pgrp;
2375 		LIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
2376 	}
2377 
2378 	sigio_unlink(sir, &rmlist);
2379 	sir->sir_sigio = sigio;
2380 
2381 	mtx_leave(&sigio_lock);
2382 	KERNEL_UNLOCK();
2383 
2384 	sigio_del(&rmlist);
2385 
2386 	return (0);
2387 
2388 fail:
2389 	mtx_leave(&sigio_lock);
2390 	KERNEL_UNLOCK();
2391 
2392 	crfree(sigio->sio_ucred);
2393 	free(sigio, M_SIGIO, sizeof(*sigio));
2394 
2395 	return (error);
2396 }
2397 
2398 void
2399 sigio_getown(struct sigio_ref *sir, u_long cmd, caddr_t data)
2400 {
2401 	struct sigio *sigio;
2402 	pid_t pgid = 0;
2403 
2404 	mtx_enter(&sigio_lock);
2405 	sigio = sir->sir_sigio;
2406 	if (sigio != NULL)
2407 		pgid = sigio->sio_pgid;
2408 	mtx_leave(&sigio_lock);
2409 
2410 	if (cmd == TIOCGPGRP)
2411 		pgid = -pgid;
2412 
2413 	*(int *)data = pgid;
2414 }
2415 
2416 void
2417 sigio_copy(struct sigio_ref *dst, struct sigio_ref *src)
2418 {
2419 	struct sigiolst rmlist;
2420 	struct sigio *newsigio, *sigio;
2421 
2422 	sigio_free(dst);
2423 
2424 	if (src->sir_sigio == NULL)
2425 		return;
2426 
2427 	newsigio = malloc(sizeof(*newsigio), M_SIGIO, M_WAITOK);
2428 	LIST_INIT(&rmlist);
2429 
2430 	mtx_enter(&sigio_lock);
2431 
2432 	sigio = src->sir_sigio;
2433 	if (sigio == NULL) {
2434 		mtx_leave(&sigio_lock);
2435 		free(newsigio, M_SIGIO, sizeof(*newsigio));
2436 		return;
2437 	}
2438 
2439 	newsigio->sio_pgid = sigio->sio_pgid;
2440 	newsigio->sio_ucred = crhold(sigio->sio_ucred);
2441 	newsigio->sio_myref = dst;
2442 	if (newsigio->sio_pgid > 0) {
2443 		newsigio->sio_proc = sigio->sio_proc;
2444 		LIST_INSERT_HEAD(&newsigio->sio_proc->ps_sigiolst, newsigio,
2445 		    sio_pgsigio);
2446 	} else {
2447 		newsigio->sio_pgrp = sigio->sio_pgrp;
2448 		LIST_INSERT_HEAD(&newsigio->sio_pgrp->pg_sigiolst, newsigio,
2449 		    sio_pgsigio);
2450 	}
2451 
2452 	sigio_unlink(dst, &rmlist);
2453 	dst->sir_sigio = newsigio;
2454 
2455 	mtx_leave(&sigio_lock);
2456 
2457 	sigio_del(&rmlist);
2458 }
2459