xref: /openbsd/sys/kern/kern_sig.c (revision 350ab3c3)
1 /*	$OpenBSD: kern_sig.c,v 1.334 2024/07/24 15:31:08 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 		SET(p->p_sigmask, mask);
476 		break;
477 	case SIG_UNBLOCK:
478 		CLR(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 	p->p_sigmask = newmask;
508 	atomic_setbits_int(&p->p_flag, P_SIGSUSPEND);
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 	SET(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 		if (code == ILL_BTCFI) {
803 			pr->ps_acflag |= ABTCFI;
804 			break;
805 		}
806 		/* FALLTHROUGH */
807 	case SIGBUS:
808 	case SIGSEGV:
809 		pr->ps_acflag |= ATRAP;
810 		break;
811 	}
812 
813 	mask = sigmask(signum);
814 	setsigctx(p, signum, &ctx);
815 	if ((pr->ps_flags & PS_TRACED) == 0 && ctx.sig_catch != 0 &&
816 	    (p->p_sigmask & mask) == 0) {
817 		siginfo_t si;
818 
819 		initsiginfo(&si, signum, trapno, code, sigval);
820 #ifdef KTRACE
821 		if (KTRPOINT(p, KTR_PSIG)) {
822 			ktrpsig(p, signum, ctx.sig_action,
823 			    p->p_sigmask, code, &si);
824 		}
825 #endif
826 		if (sendsig(ctx.sig_action, signum, p->p_sigmask, &si,
827 		    ctx.sig_info, ctx.sig_onstack)) {
828 			KERNEL_LOCK();
829 			sigexit(p, SIGILL);
830 			/* NOTREACHED */
831 		}
832 		postsig_done(p, signum, ctx.sig_catchmask, ctx.sig_reset);
833 	} else {
834 		p->p_sisig = signum;
835 		p->p_sitrapno = trapno;	/* XXX for core dump/debugger */
836 		p->p_sicode = code;
837 		p->p_sigval = sigval;
838 
839 		/*
840 		 * If traced, stop if signal is masked, and stay stopped
841 		 * until released by the debugger.  If our parent process
842 		 * is waiting for us, don't hang as we could deadlock.
843 		 */
844 		if (((pr->ps_flags & (PS_TRACED | PS_PPWAIT)) == PS_TRACED) &&
845 		    signum != SIGKILL && (p->p_sigmask & mask) != 0) {
846 			single_thread_set(p, SINGLE_SUSPEND | SINGLE_NOWAIT);
847 			pr->ps_xsig = signum;
848 
849 			SCHED_LOCK();
850 			proc_stop(p, 1);
851 			SCHED_UNLOCK();
852 
853 			signum = pr->ps_xsig;
854 			single_thread_clear(p, 0);
855 
856 			/*
857 			 * If we are no longer being traced, or the parent
858 			 * didn't give us a signal, skip sending the signal.
859 			 */
860 			if ((pr->ps_flags & PS_TRACED) == 0 ||
861 			    signum == 0)
862 				return;
863 
864 			/* update signal info */
865 			p->p_sisig = signum;
866 			mask = sigmask(signum);
867 		}
868 
869 		/*
870 		 * Signals like SIGBUS and SIGSEGV should not, when
871 		 * generated by the kernel, be ignorable or blockable.
872 		 * If it is and we're not being traced, then just kill
873 		 * the process.
874 		 * After vfs_shutdown(9), init(8) cannot receive signals
875 		 * because new code pages of the signal handler cannot be
876 		 * mapped from halted storage.  init(8) may not die or the
877 		 * kernel panics.  Better loop between signal handler and
878 		 * page fault trap until the machine is halted.
879 		 */
880 		if ((pr->ps_flags & PS_TRACED) == 0 &&
881 		    (sigprop[signum] & SA_KILL) &&
882 		    ((p->p_sigmask & mask) || ctx.sig_ignore) &&
883 		    pr->ps_pid != 1) {
884 			KERNEL_LOCK();
885 			sigexit(p, signum);
886 			/* NOTREACHED */
887 		}
888 		KERNEL_LOCK();
889 		ptsignal(p, signum, STHREAD);
890 		KERNEL_UNLOCK();
891 	}
892 }
893 
894 /*
895  * Send the signal to the process.  If the signal has an action, the action
896  * is usually performed by the target process rather than the caller; we add
897  * the signal to the set of pending signals for the process.
898  *
899  * Exceptions:
900  *   o When a stop signal is sent to a sleeping process that takes the
901  *     default action, the process is stopped without awakening it.
902  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
903  *     regardless of the signal action (eg, blocked or ignored).
904  *
905  * Other ignored signals are discarded immediately.
906  */
907 void
908 psignal(struct proc *p, int signum)
909 {
910 	ptsignal(p, signum, SPROCESS);
911 }
912 
913 /*
914  * type = SPROCESS	process signal, can be diverted (sigwait())
915  * type = STHREAD	thread signal, but should be propagated if unhandled
916  * type = SPROPAGATED	propagated to this thread, so don't propagate again
917  */
918 void
919 ptsignal(struct proc *p, int signum, enum signal_type type)
920 {
921 	int prop;
922 	sig_t action, altaction = SIG_DFL;
923 	sigset_t mask, sigmask;
924 	int *siglist;
925 	struct process *pr = p->p_p;
926 	struct proc *q;
927 	int wakeparent = 0;
928 
929 	KERNEL_ASSERT_LOCKED();
930 
931 #ifdef DIAGNOSTIC
932 	if ((u_int)signum >= NSIG || signum == 0)
933 		panic("psignal signal number");
934 #endif
935 
936 	/* Ignore signal if the target process is exiting */
937 	if (pr->ps_flags & PS_EXITING)
938 		return;
939 
940 	mask = sigmask(signum);
941 	sigmask = READ_ONCE(p->p_sigmask);
942 
943 	if (type == SPROCESS) {
944 		sigset_t tmpmask;
945 
946 		/* Accept SIGKILL to coredumping processes */
947 		if (pr->ps_flags & PS_COREDUMP && signum == SIGKILL) {
948 			atomic_setbits_int(&pr->ps_siglist, mask);
949 			return;
950 		}
951 
952 		/*
953 		 * If the current thread can process the signal
954 		 * immediately (it's unblocked) then have it take it.
955 		 */
956 		q = curproc;
957 		tmpmask = READ_ONCE(q->p_sigmask);
958 		if (q->p_p == pr && (q->p_flag & P_WEXIT) == 0 &&
959 		    (tmpmask & mask) == 0) {
960 			p = q;
961 			sigmask = tmpmask;
962 		} else {
963 			/*
964 			 * A process-wide signal can be diverted to a
965 			 * different thread that's in sigwait() for this
966 			 * signal.  If there isn't such a thread, then
967 			 * pick a thread that doesn't have it blocked so
968 			 * that the stop/kill consideration isn't
969 			 * delayed.  Otherwise, mark it pending on the
970 			 * main thread.
971 			 */
972 			TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
973 
974 				/* ignore exiting threads */
975 				if (q->p_flag & P_WEXIT)
976 					continue;
977 
978 				/* skip threads that have the signal blocked */
979 				tmpmask = READ_ONCE(q->p_sigmask);
980 				if ((tmpmask & mask) != 0)
981 					continue;
982 
983 				/* okay, could send to this thread */
984 				p = q;
985 				sigmask = tmpmask;
986 
987 				/*
988 				 * sigsuspend, sigwait, ppoll/pselect, etc?
989 				 * Definitely go to this thread, as it's
990 				 * already blocked in the kernel.
991 				 */
992 				if (q->p_flag & P_SIGSUSPEND)
993 					break;
994 			}
995 		}
996 	}
997 
998 	if (type != SPROPAGATED)
999 		knote_locked(&pr->ps_klist, NOTE_SIGNAL | signum);
1000 
1001 	prop = sigprop[signum];
1002 
1003 	/*
1004 	 * If proc is traced, always give parent a chance.
1005 	 */
1006 	if (pr->ps_flags & PS_TRACED) {
1007 		action = SIG_DFL;
1008 	} else {
1009 		sigset_t sigcatch, sigignore;
1010 
1011 		/*
1012 		 * If the signal is being ignored,
1013 		 * then we forget about it immediately.
1014 		 * (Note: we don't set SIGCONT in ps_sigignore,
1015 		 * and if it is set to SIG_IGN,
1016 		 * action will be SIG_DFL here.)
1017 		 */
1018 		mtx_enter(&pr->ps_mtx);
1019 		sigignore = pr->ps_sigacts->ps_sigignore;
1020 		sigcatch = pr->ps_sigacts->ps_sigcatch;
1021 		mtx_leave(&pr->ps_mtx);
1022 
1023 		if (sigignore & mask)
1024 			return;
1025 		if (sigmask & mask) {
1026 			action = SIG_HOLD;
1027 			if (sigcatch & mask)
1028 				altaction = SIG_CATCH;
1029 		} else if (sigcatch & mask) {
1030 			action = SIG_CATCH;
1031 		} else {
1032 			action = SIG_DFL;
1033 
1034 			if (prop & SA_KILL && pr->ps_nice > NZERO)
1035 				 pr->ps_nice = NZERO;
1036 
1037 			/*
1038 			 * If sending a tty stop signal to a member of an
1039 			 * orphaned process group, discard the signal here if
1040 			 * the action is default; don't stop the process below
1041 			 * if sleeping, and don't clear any pending SIGCONT.
1042 			 */
1043 			if (prop & SA_TTYSTOP && pr->ps_pgrp->pg_jobc == 0)
1044 				return;
1045 		}
1046 	}
1047 	/*
1048 	 * If delivered to process, mark as pending there.  Continue and stop
1049 	 * signals will be propagated to all threads.  So they are always
1050 	 * marked at thread level.
1051 	 */
1052 	siglist = (type == SPROCESS) ? &pr->ps_siglist : &p->p_siglist;
1053 	if (prop & (SA_CONT | SA_STOP))
1054 		siglist = &p->p_siglist;
1055 
1056 	/*
1057 	 * XXX delay processing of SA_STOP signals unless action == SIG_DFL?
1058 	 */
1059 	if (prop & (SA_CONT | SA_STOP) && type != SPROPAGATED)
1060 		TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link)
1061 			if (q != p)
1062 				ptsignal(q, signum, SPROPAGATED);
1063 
1064 	SCHED_LOCK();
1065 
1066 	switch (p->p_stat) {
1067 
1068 	case SSTOP:
1069 		/*
1070 		 * If traced process is already stopped,
1071 		 * then no further action is necessary.
1072 		 */
1073 		if (pr->ps_flags & PS_TRACED)
1074 			goto out;
1075 
1076 		/*
1077 		 * Kill signal always sets processes running.
1078 		 */
1079 		if (signum == SIGKILL) {
1080 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1081 			/* Raise priority to at least PUSER. */
1082 			if (p->p_usrpri > PUSER)
1083 				p->p_usrpri = PUSER;
1084 			unsleep(p);
1085 			setrunnable(p);
1086 			goto out;
1087 		}
1088 
1089 		if (prop & SA_CONT) {
1090 			/*
1091 			 * If SIGCONT is default (or ignored), we continue the
1092 			 * process but don't leave the signal in p_siglist, as
1093 			 * it has no further action.  If SIGCONT is held, we
1094 			 * continue the process and leave the signal in
1095 			 * p_siglist.  If the process catches SIGCONT, let it
1096 			 * handle the signal itself.  If it isn't waiting on
1097 			 * an event, then it goes back to run state.
1098 			 * Otherwise, process goes back to sleep state.
1099 			 */
1100 			atomic_setbits_int(&p->p_flag, P_CONTINUED);
1101 			atomic_clearbits_int(&p->p_flag, P_SUSPSIG);
1102 			wakeparent = 1;
1103 			if (action == SIG_DFL)
1104 				mask = 0;
1105 			if (action == SIG_CATCH) {
1106 				/* Raise priority to at least PUSER. */
1107 				if (p->p_usrpri > PUSER)
1108 					p->p_usrpri = PUSER;
1109 				unsleep(p);
1110 				setrunnable(p);
1111 				goto out;
1112 			}
1113 			if (p->p_wchan == NULL) {
1114 				unsleep(p);
1115 				setrunnable(p);
1116 				goto out;
1117 			}
1118 			atomic_clearbits_int(&p->p_flag, P_WSLEEP);
1119 			p->p_stat = SSLEEP;
1120 			goto out;
1121 		}
1122 
1123 		/*
1124 		 * Defer further processing for signals which are held,
1125 		 * except that stopped processes must be continued by SIGCONT.
1126 		 */
1127 		if (action == SIG_HOLD)
1128 			goto out;
1129 
1130 		if (prop & SA_STOP) {
1131 			/*
1132 			 * Already stopped, don't need to stop again.
1133 			 * (If we did the shell could get confused.)
1134 			 */
1135 			mask = 0;
1136 			goto out;
1137 		}
1138 
1139 		/*
1140 		 * If process is sleeping interruptibly, then simulate a
1141 		 * wakeup so that when it is continued, it will be made
1142 		 * runnable and can look at the signal.  But don't make
1143 		 * the process runnable, leave it stopped.
1144 		 */
1145 		if (p->p_flag & P_SINTR)
1146 			unsleep(p);
1147 		goto out;
1148 
1149 	case SSLEEP:
1150 		/*
1151 		 * If process is sleeping uninterruptibly
1152 		 * we can't interrupt the sleep... the signal will
1153 		 * be noticed when the process returns through
1154 		 * trap() or syscall().
1155 		 */
1156 		if ((p->p_flag & P_SINTR) == 0)
1157 			goto out;
1158 		/*
1159 		 * Process is sleeping and traced... make it runnable
1160 		 * so it can discover the signal in cursig() and stop
1161 		 * for the parent.
1162 		 */
1163 		if (pr->ps_flags & PS_TRACED) {
1164 			unsleep(p);
1165 			setrunnable(p);
1166 			goto out;
1167 		}
1168 
1169 		/*
1170 		 * Recheck sigmask before waking up the process,
1171 		 * there is a chance that while sending the signal
1172 		 * the process changed sigmask and went to sleep.
1173 		 */
1174 		sigmask = READ_ONCE(p->p_sigmask);
1175 		if (sigmask & mask)
1176 			goto out;
1177 		else if (action == SIG_HOLD) {
1178 			/* signal got unmasked, get proper action */
1179 			action = altaction;
1180 
1181 			if (action == SIG_DFL) {
1182 				if (prop & SA_KILL && pr->ps_nice > NZERO)
1183 					 pr->ps_nice = NZERO;
1184 
1185 				/*
1186 				 * Discard tty stop signals sent to an
1187 				 * orphaned process group, see above.
1188 				 */
1189 				if (prop & SA_TTYSTOP &&
1190 				    pr->ps_pgrp->pg_jobc == 0) {
1191 					mask = 0;
1192 					prop = 0;
1193 					goto out;
1194 				}
1195 			}
1196 		}
1197 
1198 		/*
1199 		 * If SIGCONT is default (or ignored) and process is
1200 		 * asleep, we are finished; the process should not
1201 		 * be awakened.
1202 		 */
1203 		if ((prop & SA_CONT) && action == SIG_DFL) {
1204 			mask = 0;
1205 			goto out;
1206 		}
1207 		/*
1208 		 * When a sleeping process receives a stop
1209 		 * signal, process immediately if possible.
1210 		 */
1211 		if ((prop & SA_STOP) && action == SIG_DFL) {
1212 			/*
1213 			 * If a child holding parent blocked,
1214 			 * stopping could cause deadlock.
1215 			 */
1216 			if (pr->ps_flags & PS_PPWAIT)
1217 				goto out;
1218 			mask = 0;
1219 			pr->ps_xsig = signum;
1220 			proc_stop(p, 0);
1221 			goto out;
1222 		}
1223 		/*
1224 		 * All other (caught or default) signals
1225 		 * cause the process to run.
1226 		 * Raise priority to at least PUSER.
1227 		 */
1228 		if (p->p_usrpri > PUSER)
1229 			p->p_usrpri = PUSER;
1230 		unsleep(p);
1231 		setrunnable(p);
1232 		goto out;
1233 		/* NOTREACHED */
1234 
1235 	case SONPROC:
1236 		if (action == SIG_HOLD)
1237 			goto out;
1238 
1239 		/* set siglist before issuing the ast */
1240 		atomic_setbits_int(siglist, mask);
1241 		mask = 0;
1242 		signotify(p);
1243 		/* FALLTHROUGH */
1244 	default:
1245 		/*
1246 		 * SRUN, SIDL, SDEAD do nothing with the signal,
1247 		 * other than kicking ourselves if we are running.
1248 		 * It will either never be noticed, or noticed very soon.
1249 		 */
1250 		goto out;
1251 	}
1252 	/* NOTREACHED */
1253 
1254 out:
1255 	/* finally adjust siglist */
1256 	if (mask)
1257 		atomic_setbits_int(siglist, mask);
1258 	if (prop & SA_CONT) {
1259 		atomic_clearbits_int(siglist, STOPSIGMASK);
1260 	}
1261 	if (prop & SA_STOP) {
1262 		atomic_clearbits_int(siglist, CONTSIGMASK);
1263 		atomic_clearbits_int(&p->p_flag, P_CONTINUED);
1264 	}
1265 
1266 	SCHED_UNLOCK();
1267 	if (wakeparent)
1268 		wakeup(pr->ps_pptr);
1269 }
1270 
1271 /* fill the signal context which should be used by postsig() and issignal() */
1272 void
1273 setsigctx(struct proc *p, int signum, struct sigctx *sctx)
1274 {
1275 	struct sigacts *ps = p->p_p->ps_sigacts;
1276 	sigset_t mask;
1277 
1278 	mtx_enter(&p->p_p->ps_mtx);
1279 	mask = sigmask(signum);
1280 	sctx->sig_action = ps->ps_sigact[signum];
1281 	sctx->sig_catchmask = ps->ps_catchmask[signum];
1282 	sctx->sig_reset = (ps->ps_sigreset & mask) != 0;
1283 	sctx->sig_info = (ps->ps_siginfo & mask) != 0;
1284 	sctx->sig_intr = (ps->ps_sigintr & mask) != 0;
1285 	sctx->sig_onstack = (ps->ps_sigonstack & mask) != 0;
1286 	sctx->sig_ignore = (ps->ps_sigignore & mask) != 0;
1287 	sctx->sig_catch = (ps->ps_sigcatch & mask) != 0;
1288 	mtx_leave(&p->p_p->ps_mtx);
1289 }
1290 
1291 /*
1292  * Determine signal that should be delivered to process p, the current
1293  * process, 0 if none.
1294  *
1295  * If the current process has received a signal (should be caught or cause
1296  * termination, should interrupt current syscall), return the signal number.
1297  * Stop signals with default action are processed immediately, then cleared;
1298  * they aren't returned.  This is checked after each entry to the system for
1299  * a syscall or trap. The normal call sequence is
1300  *
1301  *	while (signum = cursig(curproc, &ctx))
1302  *		postsig(signum, &ctx);
1303  *
1304  * Assumes that if the P_SINTR flag is set, we're holding both the
1305  * kernel and scheduler locks.
1306  */
1307 int
1308 cursig(struct proc *p, struct sigctx *sctx)
1309 {
1310 	struct process *pr = p->p_p;
1311 	int signum, mask, prop;
1312 	sigset_t ps_siglist;
1313 
1314 	KASSERT(p == curproc);
1315 
1316 	for (;;) {
1317 		ps_siglist = READ_ONCE(pr->ps_siglist);
1318 		membar_consumer();
1319 		mask = SIGPENDING(p);
1320 		if (pr->ps_flags & PS_PPWAIT)
1321 			mask &= ~STOPSIGMASK;
1322 		if (mask == 0)	 	/* no signal to send */
1323 			return (0);
1324 		signum = ffs((long)mask);
1325 		mask = sigmask(signum);
1326 
1327 		/* take the signal! */
1328 		if (atomic_cas_uint(&pr->ps_siglist, ps_siglist,
1329 		    ps_siglist & ~mask) != ps_siglist) {
1330 			/* lost race taking the process signal, restart */
1331 			continue;
1332 		}
1333 		atomic_clearbits_int(&p->p_siglist, mask);
1334 		setsigctx(p, signum, sctx);
1335 
1336 		/*
1337 		 * We should see pending but ignored signals
1338 		 * only if PS_TRACED was on when they were posted.
1339 		 */
1340 		if (sctx->sig_ignore && (pr->ps_flags & PS_TRACED) == 0)
1341 			continue;
1342 
1343 		/*
1344 		 * If traced, always stop, and stay stopped until released
1345 		 * by the debugger.  If our parent process is waiting for
1346 		 * us, don't hang as we could deadlock.
1347 		 */
1348 		if (((pr->ps_flags & (PS_TRACED | PS_PPWAIT)) == PS_TRACED) &&
1349 		    signum != SIGKILL) {
1350 			single_thread_set(p, SINGLE_SUSPEND | SINGLE_NOWAIT);
1351 			pr->ps_xsig = signum;
1352 
1353 			SCHED_LOCK();
1354 			proc_stop(p, 1);
1355 			SCHED_UNLOCK();
1356 
1357 			/*
1358 			 * re-take the signal before releasing
1359 			 * the other threads. Must check the continue
1360 			 * conditions below and only take the signal if
1361 			 * those are not true.
1362 			 */
1363 			signum = pr->ps_xsig;
1364 			mask = sigmask(signum);
1365 			setsigctx(p, signum, sctx);
1366 			if (!((pr->ps_flags & PS_TRACED) == 0 ||
1367 			    signum == 0 ||
1368 			    (p->p_sigmask & mask) != 0)) {
1369 				atomic_clearbits_int(&p->p_siglist, mask);
1370 				atomic_clearbits_int(&pr->ps_siglist, mask);
1371 			}
1372 
1373 			single_thread_clear(p, 0);
1374 
1375 			/*
1376 			 * If we are no longer being traced, or the parent
1377 			 * didn't give us a signal, look for more signals.
1378 			 */
1379 			if ((pr->ps_flags & PS_TRACED) == 0 ||
1380 			    signum == 0)
1381 				continue;
1382 
1383 			/*
1384 			 * If the new signal is being masked, look for other
1385 			 * signals.
1386 			 */
1387 			if ((p->p_sigmask & mask) != 0)
1388 				continue;
1389 
1390 		}
1391 
1392 		prop = sigprop[signum];
1393 
1394 		/*
1395 		 * Decide whether the signal should be returned.
1396 		 * Return the signal's number, or fall through
1397 		 * to clear it from the pending mask.
1398 		 */
1399 		switch ((long)sctx->sig_action) {
1400 		case (long)SIG_DFL:
1401 			/*
1402 			 * Don't take default actions on system processes.
1403 			 */
1404 			if (pr->ps_pid <= 1) {
1405 #ifdef DIAGNOSTIC
1406 				/*
1407 				 * Are you sure you want to ignore SIGSEGV
1408 				 * in init? XXX
1409 				 */
1410 				printf("Process (pid %d) got signal"
1411 				    " %d\n", pr->ps_pid, signum);
1412 #endif
1413 				break;		/* == ignore */
1414 			}
1415 			/*
1416 			 * If there is a pending stop signal to process
1417 			 * with default action, stop here,
1418 			 * then clear the signal.  However,
1419 			 * if process is member of an orphaned
1420 			 * process group, ignore tty stop signals.
1421 			 */
1422 			if (prop & SA_STOP) {
1423 				if (pr->ps_flags & PS_TRACED ||
1424 		    		    (pr->ps_pgrp->pg_jobc == 0 &&
1425 				    prop & SA_TTYSTOP))
1426 					break;	/* == ignore */
1427 				pr->ps_xsig = signum;
1428 				SCHED_LOCK();
1429 				proc_stop(p, 1);
1430 				SCHED_UNLOCK();
1431 				break;
1432 			} else if (prop & SA_IGNORE) {
1433 				/*
1434 				 * Except for SIGCONT, shouldn't get here.
1435 				 * Default action is to ignore; drop it.
1436 				 */
1437 				break;		/* == ignore */
1438 			} else
1439 				goto keep;
1440 			/* NOTREACHED */
1441 		case (long)SIG_IGN:
1442 			/*
1443 			 * Masking above should prevent us ever trying
1444 			 * to take action on an ignored signal other
1445 			 * than SIGCONT, unless process is traced.
1446 			 */
1447 			if ((prop & SA_CONT) == 0 &&
1448 			    (pr->ps_flags & PS_TRACED) == 0)
1449 				printf("%s\n", __func__);
1450 			break;		/* == ignore */
1451 		default:
1452 			/*
1453 			 * This signal has an action, let
1454 			 * postsig() process it.
1455 			 */
1456 			goto keep;
1457 		}
1458 	}
1459 	/* NOTREACHED */
1460 
1461 keep:
1462 	atomic_setbits_int(&p->p_siglist, mask); /*leave the signal for later */
1463 	return (signum);
1464 }
1465 
1466 /*
1467  * Put the argument process into the stopped state and notify the parent
1468  * via wakeup.  Signals are handled elsewhere.  The process must not be
1469  * on the run queue.
1470  */
1471 void
1472 proc_stop(struct proc *p, int sw)
1473 {
1474 	struct process *pr = p->p_p;
1475 
1476 #ifdef MULTIPROCESSOR
1477 	SCHED_ASSERT_LOCKED();
1478 #endif
1479 	/* do not stop exiting procs */
1480 	if (ISSET(p->p_flag, P_WEXIT))
1481 		return;
1482 
1483 	p->p_stat = SSTOP;
1484 	atomic_clearbits_int(&pr->ps_flags, PS_WAITED);
1485 	atomic_setbits_int(&pr->ps_flags, PS_STOPPING);
1486 	atomic_setbits_int(&p->p_flag, P_SUSPSIG);
1487 	/*
1488 	 * We need this soft interrupt to be handled fast.
1489 	 * Extra calls to softclock don't hurt.
1490 	 */
1491 	softintr_schedule(proc_stop_si);
1492 	if (sw)
1493 		mi_switch();
1494 }
1495 
1496 /*
1497  * Called from a soft interrupt to send signals to the parents of stopped
1498  * processes.
1499  * We can't do this in proc_stop because it's called with nasty locks held
1500  * and we would need recursive scheduler lock to deal with that.
1501  */
1502 void
1503 proc_stop_sweep(void *v)
1504 {
1505 	struct process *pr;
1506 
1507 	LIST_FOREACH(pr, &allprocess, ps_list) {
1508 		if ((pr->ps_flags & PS_STOPPING) == 0)
1509 			continue;
1510 		atomic_clearbits_int(&pr->ps_flags, PS_STOPPING);
1511 
1512 		if ((pr->ps_pptr->ps_sigacts->ps_sigflags & SAS_NOCLDSTOP) == 0)
1513 			prsignal(pr->ps_pptr, SIGCHLD);
1514 		wakeup(pr->ps_pptr);
1515 	}
1516 }
1517 
1518 /*
1519  * Take the action for the specified signal
1520  * from the current set of pending signals.
1521  */
1522 void
1523 postsig(struct proc *p, int signum, struct sigctx *sctx)
1524 {
1525 	u_long trapno;
1526 	int mask, returnmask;
1527 	siginfo_t si;
1528 	union sigval sigval;
1529 	int code;
1530 
1531 	KASSERT(signum != 0);
1532 
1533 	mask = sigmask(signum);
1534 	atomic_clearbits_int(&p->p_siglist, mask);
1535 	sigval.sival_ptr = NULL;
1536 
1537 	if (p->p_sisig != signum) {
1538 		trapno = 0;
1539 		code = SI_USER;
1540 		sigval.sival_ptr = NULL;
1541 	} else {
1542 		trapno = p->p_sitrapno;
1543 		code = p->p_sicode;
1544 		sigval = p->p_sigval;
1545 	}
1546 	initsiginfo(&si, signum, trapno, code, sigval);
1547 
1548 #ifdef KTRACE
1549 	if (KTRPOINT(p, KTR_PSIG)) {
1550 		ktrpsig(p, signum, sctx->sig_action, p->p_flag & P_SIGSUSPEND ?
1551 		    p->p_oldmask : p->p_sigmask, code, &si);
1552 	}
1553 #endif
1554 	if (sctx->sig_action == SIG_DFL) {
1555 		/*
1556 		 * Default action, where the default is to kill
1557 		 * the process.  (Other cases were ignored above.)
1558 		 */
1559 		KERNEL_LOCK();
1560 		sigexit(p, signum);
1561 		/* NOTREACHED */
1562 	} else {
1563 		/*
1564 		 * If we get here, the signal must be caught.
1565 		 */
1566 #ifdef DIAGNOSTIC
1567 		if (sctx->sig_action == SIG_IGN || (p->p_sigmask & mask))
1568 			panic("postsig action");
1569 #endif
1570 		/*
1571 		 * Set the new mask value and also defer further
1572 		 * occurrences of this signal.
1573 		 *
1574 		 * Special case: user has done a sigpause.  Here the
1575 		 * current mask is not of interest, but rather the
1576 		 * mask from before the sigpause is what we want
1577 		 * restored after the signal processing is completed.
1578 		 */
1579 		if (p->p_flag & P_SIGSUSPEND) {
1580 			atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
1581 			returnmask = p->p_oldmask;
1582 		} else {
1583 			returnmask = p->p_sigmask;
1584 		}
1585 		if (p->p_sisig == signum) {
1586 			p->p_sisig = 0;
1587 			p->p_sitrapno = 0;
1588 			p->p_sicode = SI_USER;
1589 			p->p_sigval.sival_ptr = NULL;
1590 		}
1591 
1592 		if (sendsig(sctx->sig_action, signum, returnmask, &si,
1593 		    sctx->sig_info, sctx->sig_onstack)) {
1594 			KERNEL_LOCK();
1595 			sigexit(p, SIGILL);
1596 			/* NOTREACHED */
1597 		}
1598 		postsig_done(p, signum, sctx->sig_catchmask, sctx->sig_reset);
1599 	}
1600 }
1601 
1602 /*
1603  * Force the current process to exit with the specified signal, dumping core
1604  * if appropriate.  We bypass the normal tests for masked and caught signals,
1605  * allowing unrecoverable failures to terminate the process without changing
1606  * signal state.  Mark the accounting record with the signal termination.
1607  * If dumping core, save the signal number for the debugger.  Calls exit and
1608  * does not return.
1609  */
1610 void
1611 sigexit(struct proc *p, int signum)
1612 {
1613 	/* Mark process as going away */
1614 	atomic_setbits_int(&p->p_flag, P_WEXIT);
1615 
1616 	p->p_p->ps_acflag |= AXSIG;
1617 	if (sigprop[signum] & SA_CORE) {
1618 		p->p_sisig = signum;
1619 
1620 		/* if there are other threads, pause them */
1621 		if (P_HASSIBLING(p))
1622 			single_thread_set(p, SINGLE_UNWIND);
1623 
1624 		if (coredump(p) == 0)
1625 			signum |= WCOREFLAG;
1626 	}
1627 	exit1(p, 0, signum, EXIT_NORMAL);
1628 	/* NOTREACHED */
1629 }
1630 
1631 /*
1632  * Send uncatchable SIGABRT for coredump.
1633  */
1634 void
1635 sigabort(struct proc *p)
1636 {
1637 	struct sigaction sa;
1638 
1639 	KASSERT(p == curproc || panicstr || db_active);
1640 
1641 	memset(&sa, 0, sizeof sa);
1642 	sa.sa_handler = SIG_DFL;
1643 	setsigvec(p, SIGABRT, &sa);
1644 	CLR(p->p_sigmask, sigmask(SIGABRT));
1645 	psignal(p, SIGABRT);
1646 }
1647 
1648 /*
1649  * Return 1 if `sig', a given signal, is ignored or masked for `p', a given
1650  * thread, and 0 otherwise.
1651  */
1652 int
1653 sigismasked(struct proc *p, int sig)
1654 {
1655 	struct process *pr = p->p_p;
1656 	int rv;
1657 
1658 	KASSERT(p == curproc);
1659 
1660 	mtx_enter(&pr->ps_mtx);
1661 	rv = (pr->ps_sigacts->ps_sigignore & sigmask(sig)) ||
1662 	    (p->p_sigmask & sigmask(sig));
1663 	mtx_leave(&pr->ps_mtx);
1664 
1665 	return !!rv;
1666 }
1667 
1668 struct coredump_iostate {
1669 	struct proc *io_proc;
1670 	struct vnode *io_vp;
1671 	struct ucred *io_cred;
1672 	off_t io_offset;
1673 };
1674 
1675 /*
1676  * Dump core, into a file named "progname.core", unless the process was
1677  * setuid/setgid.
1678  */
1679 int
1680 coredump(struct proc *p)
1681 {
1682 #ifdef SMALL_KERNEL
1683 	return EPERM;
1684 #else
1685 	struct process *pr = p->p_p;
1686 	struct vnode *vp;
1687 	struct ucred *cred = p->p_ucred;
1688 	struct vmspace *vm = p->p_vmspace;
1689 	struct nameidata nd;
1690 	struct vattr vattr;
1691 	struct coredump_iostate	io;
1692 	int error, len, incrash = 0;
1693 	char *name;
1694 	const char *dir = "/var/crash";
1695 
1696 	atomic_setbits_int(&pr->ps_flags, PS_COREDUMP);
1697 
1698 #ifdef PMAP_CHECK_COPYIN
1699 	/* disable copyin checks, so we can write out text sections if needed */
1700 	p->p_vmspace->vm_map.check_copyin_count = 0;
1701 #endif
1702 
1703 	/* Don't dump if will exceed file size limit. */
1704 	if (USPACE + ptoa(vm->vm_dsize + vm->vm_ssize) >= lim_cur(RLIMIT_CORE))
1705 		return (EFBIG);
1706 
1707 	name = pool_get(&namei_pool, PR_WAITOK);
1708 
1709 	/*
1710 	 * If the process has inconsistent uids, nosuidcoredump
1711 	 * determines coredump placement policy.
1712 	 */
1713 	if (((pr->ps_flags & PS_SUGID) && (error = suser(p))) ||
1714 	   ((pr->ps_flags & PS_SUGID) && nosuidcoredump)) {
1715 		if (nosuidcoredump == 3) {
1716 			/*
1717 			 * If the program directory does not exist, dumps of
1718 			 * that core will silently fail.
1719 			 */
1720 			len = snprintf(name, MAXPATHLEN, "%s/%s/%u.core",
1721 			    dir, pr->ps_comm, pr->ps_pid);
1722 			incrash = KERNELPATH;
1723 		} else if (nosuidcoredump == 2) {
1724 			len = snprintf(name, MAXPATHLEN, "%s/%s.core",
1725 			    dir, pr->ps_comm);
1726 			incrash = KERNELPATH;
1727 		} else {
1728 			pool_put(&namei_pool, name);
1729 			return (EPERM);
1730 		}
1731 	} else
1732 		len = snprintf(name, MAXPATHLEN, "%s.core", pr->ps_comm);
1733 
1734 	if (len >= MAXPATHLEN) {
1735 		pool_put(&namei_pool, name);
1736 		return (EACCES);
1737 	}
1738 
1739 	/*
1740 	 * Control the UID used to write out.  The normal case uses
1741 	 * the real UID.  If the sugid case is going to write into the
1742 	 * controlled directory, we do so as root.
1743 	 */
1744 	if (incrash == 0) {
1745 		cred = crdup(cred);
1746 		cred->cr_uid = cred->cr_ruid;
1747 		cred->cr_gid = cred->cr_rgid;
1748 	} else {
1749 		if (p->p_fd->fd_rdir) {
1750 			vrele(p->p_fd->fd_rdir);
1751 			p->p_fd->fd_rdir = NULL;
1752 		}
1753 		p->p_ucred = crdup(p->p_ucred);
1754 		crfree(cred);
1755 		cred = p->p_ucred;
1756 		crhold(cred);
1757 		cred->cr_uid = 0;
1758 		cred->cr_gid = 0;
1759 	}
1760 
1761 	/* incrash should be 0 or KERNELPATH only */
1762 	NDINIT(&nd, 0, BYPASSUNVEIL | incrash, UIO_SYSSPACE, name, p);
1763 
1764 	error = vn_open(&nd, O_CREAT | FWRITE | O_NOFOLLOW | O_NONBLOCK,
1765 	    S_IRUSR | S_IWUSR);
1766 
1767 	if (error)
1768 		goto out;
1769 
1770 	/*
1771 	 * Don't dump to non-regular files, files with links, or files
1772 	 * owned by someone else.
1773 	 */
1774 	vp = nd.ni_vp;
1775 	if ((error = VOP_GETATTR(vp, &vattr, cred, p)) != 0) {
1776 		VOP_UNLOCK(vp);
1777 		vn_close(vp, FWRITE, cred, p);
1778 		goto out;
1779 	}
1780 	if (vp->v_type != VREG || vattr.va_nlink != 1 ||
1781 	    vattr.va_mode & ((VREAD | VWRITE) >> 3 | (VREAD | VWRITE) >> 6) ||
1782 	    vattr.va_uid != cred->cr_uid) {
1783 		error = EACCES;
1784 		VOP_UNLOCK(vp);
1785 		vn_close(vp, FWRITE, cred, p);
1786 		goto out;
1787 	}
1788 	VATTR_NULL(&vattr);
1789 	vattr.va_size = 0;
1790 	VOP_SETATTR(vp, &vattr, cred, p);
1791 	pr->ps_acflag |= ACORE;
1792 
1793 	io.io_proc = p;
1794 	io.io_vp = vp;
1795 	io.io_cred = cred;
1796 	io.io_offset = 0;
1797 	VOP_UNLOCK(vp);
1798 	vref(vp);
1799 	error = vn_close(vp, FWRITE, cred, p);
1800 	if (error == 0)
1801 		error = coredump_elf(p, &io);
1802 	vrele(vp);
1803 out:
1804 	crfree(cred);
1805 	pool_put(&namei_pool, name);
1806 	return (error);
1807 #endif
1808 }
1809 
1810 #ifndef SMALL_KERNEL
1811 int
1812 coredump_write(void *cookie, enum uio_seg segflg, const void *data, size_t len,
1813     int isvnode)
1814 {
1815 	struct coredump_iostate *io = cookie;
1816 	off_t coffset = 0;
1817 	size_t csize;
1818 	int chunk, error;
1819 
1820 	csize = len;
1821 	do {
1822 		if (sigmask(SIGKILL) &
1823 		    (io->io_proc->p_siglist | io->io_proc->p_p->ps_siglist))
1824 			return (EINTR);
1825 
1826 		/* Rest of the loop sleeps with lock held, so... */
1827 		yield();
1828 
1829 		chunk = MIN(csize, MAXPHYS);
1830 		error = vn_rdwr(UIO_WRITE, io->io_vp,
1831 		    (caddr_t)data + coffset, chunk,
1832 		    io->io_offset + coffset, segflg,
1833 		    IO_UNIT, io->io_cred, NULL, io->io_proc);
1834 		if (error && (error != EFAULT || !isvnode)) {
1835 			struct process *pr = io->io_proc->p_p;
1836 
1837 			if (error == ENOSPC)
1838 				log(LOG_ERR,
1839 				    "coredump of %s(%d) failed, filesystem full\n",
1840 				    pr->ps_comm, pr->ps_pid);
1841 			else
1842 				log(LOG_ERR,
1843 				    "coredump of %s(%d), write failed: errno %d\n",
1844 				    pr->ps_comm, pr->ps_pid, error);
1845 			return (error);
1846 		}
1847 
1848 		coffset += chunk;
1849 		csize -= chunk;
1850 	} while (csize > 0);
1851 
1852 	io->io_offset += len;
1853 	return (0);
1854 }
1855 
1856 void
1857 coredump_unmap(void *cookie, vaddr_t start, vaddr_t end)
1858 {
1859 	struct coredump_iostate *io = cookie;
1860 
1861 	uvm_unmap(&io->io_proc->p_vmspace->vm_map, start, end);
1862 }
1863 
1864 #endif	/* !SMALL_KERNEL */
1865 
1866 /*
1867  * Nonexistent system call-- signal process (may want to handle it).
1868  * Flag error in case process won't see signal immediately (blocked or ignored).
1869  */
1870 int
1871 sys_nosys(struct proc *p, void *v, register_t *retval)
1872 {
1873 	ptsignal(p, SIGSYS, STHREAD);
1874 	return (ENOSYS);
1875 }
1876 
1877 int
1878 sys___thrsigdivert(struct proc *p, void *v, register_t *retval)
1879 {
1880 	struct sys___thrsigdivert_args /* {
1881 		syscallarg(sigset_t) sigmask;
1882 		syscallarg(siginfo_t *) info;
1883 		syscallarg(const struct timespec *) timeout;
1884 	} */ *uap = v;
1885 	struct sigctx ctx;
1886 	sigset_t mask = SCARG(uap, sigmask) &~ sigcantmask;
1887 	siginfo_t si;
1888 	uint64_t nsecs = INFSLP;
1889 	int timeinvalid = 0;
1890 	int error = 0;
1891 
1892 	memset(&si, 0, sizeof(si));
1893 
1894 	if (SCARG(uap, timeout) != NULL) {
1895 		struct timespec ts;
1896 		if ((error = copyin(SCARG(uap, timeout), &ts, sizeof(ts))) != 0)
1897 			return (error);
1898 #ifdef KTRACE
1899 		if (KTRPOINT(p, KTR_STRUCT))
1900 			ktrreltimespec(p, &ts);
1901 #endif
1902 		if (!timespecisvalid(&ts))
1903 			timeinvalid = 1;
1904 		else
1905 			nsecs = TIMESPEC_TO_NSEC(&ts);
1906 	}
1907 
1908 	dosigsuspend(p, p->p_sigmask &~ mask);
1909 	for (;;) {
1910 		si.si_signo = cursig(p, &ctx);
1911 		if (si.si_signo != 0) {
1912 			sigset_t smask = sigmask(si.si_signo);
1913 			if (smask & mask) {
1914 				atomic_clearbits_int(&p->p_siglist, smask);
1915 				error = 0;
1916 				break;
1917 			}
1918 		}
1919 
1920 		/* per-POSIX, delay this error until after the above */
1921 		if (timeinvalid)
1922 			error = EINVAL;
1923 		/* per-POSIX, return immediately if timeout is zero-valued */
1924 		if (nsecs == 0)
1925 			error = EAGAIN;
1926 
1927 		if (error != 0)
1928 			break;
1929 
1930 		error = tsleep_nsec(&nowake, PPAUSE|PCATCH, "sigwait", nsecs);
1931 	}
1932 
1933 	if (error == 0) {
1934 		*retval = si.si_signo;
1935 		if (SCARG(uap, info) != NULL) {
1936 			error = copyout(&si, SCARG(uap, info), sizeof(si));
1937 #ifdef KTRACE
1938 			if (error == 0 && KTRPOINT(p, KTR_STRUCT))
1939 				ktrsiginfo(p, &si);
1940 #endif
1941 		}
1942 	} else if (error == ERESTART && SCARG(uap, timeout) != NULL) {
1943 		/*
1944 		 * Restarting is wrong if there's a timeout, as it'll be
1945 		 * for the same interval again
1946 		 */
1947 		error = EINTR;
1948 	}
1949 
1950 	return (error);
1951 }
1952 
1953 void
1954 initsiginfo(siginfo_t *si, int sig, u_long trapno, int code, union sigval val)
1955 {
1956 	memset(si, 0, sizeof(*si));
1957 
1958 	si->si_signo = sig;
1959 	si->si_code = code;
1960 	if (code == SI_USER) {
1961 		si->si_value = val;
1962 	} else {
1963 		switch (sig) {
1964 		case SIGSEGV:
1965 		case SIGILL:
1966 		case SIGBUS:
1967 		case SIGFPE:
1968 			si->si_addr = val.sival_ptr;
1969 			si->si_trapno = trapno;
1970 			break;
1971 		case SIGXFSZ:
1972 			break;
1973 		}
1974 	}
1975 }
1976 
1977 int
1978 filt_sigattach(struct knote *kn)
1979 {
1980 	struct process *pr = curproc->p_p;
1981 	int s;
1982 
1983 	if (kn->kn_id >= NSIG)
1984 		return EINVAL;
1985 
1986 	kn->kn_ptr.p_process = pr;
1987 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
1988 
1989 	s = splhigh();
1990 	klist_insert_locked(&pr->ps_klist, kn);
1991 	splx(s);
1992 
1993 	return (0);
1994 }
1995 
1996 void
1997 filt_sigdetach(struct knote *kn)
1998 {
1999 	struct process *pr = kn->kn_ptr.p_process;
2000 	int s;
2001 
2002 	s = splhigh();
2003 	klist_remove_locked(&pr->ps_klist, kn);
2004 	splx(s);
2005 }
2006 
2007 /*
2008  * signal knotes are shared with proc knotes, so we apply a mask to
2009  * the hint in order to differentiate them from process hints.  This
2010  * could be avoided by using a signal-specific knote list, but probably
2011  * isn't worth the trouble.
2012  */
2013 int
2014 filt_signal(struct knote *kn, long hint)
2015 {
2016 
2017 	if (hint & NOTE_SIGNAL) {
2018 		hint &= ~NOTE_SIGNAL;
2019 
2020 		if (kn->kn_id == hint)
2021 			kn->kn_data++;
2022 	}
2023 	return (kn->kn_data != 0);
2024 }
2025 
2026 void
2027 userret(struct proc *p)
2028 {
2029 	struct sigctx ctx;
2030 	int signum;
2031 
2032 	if (p->p_flag & P_SUSPSINGLE)
2033 		single_thread_check(p, 0);
2034 
2035 	/* send SIGPROF or SIGVTALRM if their timers interrupted this thread */
2036 	if (p->p_flag & P_PROFPEND) {
2037 		atomic_clearbits_int(&p->p_flag, P_PROFPEND);
2038 		KERNEL_LOCK();
2039 		psignal(p, SIGPROF);
2040 		KERNEL_UNLOCK();
2041 	}
2042 	if (p->p_flag & P_ALRMPEND) {
2043 		atomic_clearbits_int(&p->p_flag, P_ALRMPEND);
2044 		KERNEL_LOCK();
2045 		psignal(p, SIGVTALRM);
2046 		KERNEL_UNLOCK();
2047 	}
2048 
2049 	if (SIGPENDING(p) != 0) {
2050 		while ((signum = cursig(p, &ctx)) != 0)
2051 			postsig(p, signum, &ctx);
2052 	}
2053 
2054 	/*
2055 	 * If P_SIGSUSPEND is still set here, then we still need to restore
2056 	 * the original sigmask before returning to userspace.  Also, this
2057 	 * might unmask some pending signals, so we need to check a second
2058 	 * time for signals to post.
2059 	 */
2060 	if (p->p_flag & P_SIGSUSPEND) {
2061 		p->p_sigmask = p->p_oldmask;
2062 		atomic_clearbits_int(&p->p_flag, P_SIGSUSPEND);
2063 
2064 		while ((signum = cursig(p, &ctx)) != 0)
2065 			postsig(p, signum, &ctx);
2066 	}
2067 
2068 	WITNESS_WARN(WARN_PANIC, NULL, "userret: returning");
2069 
2070 	p->p_cpu->ci_schedstate.spc_curpriority = p->p_usrpri;
2071 }
2072 
2073 int
2074 single_thread_check_locked(struct proc *p, int deep)
2075 {
2076 	struct process *pr = p->p_p;
2077 
2078 	MUTEX_ASSERT_LOCKED(&pr->ps_mtx);
2079 
2080 	if (pr->ps_single == NULL || pr->ps_single == p)
2081 		return (0);
2082 
2083 	do {
2084 		/* if we're in deep, we need to unwind to the edge */
2085 		if (deep) {
2086 			if (pr->ps_flags & PS_SINGLEUNWIND)
2087 				return (ERESTART);
2088 			if (pr->ps_flags & PS_SINGLEEXIT)
2089 				return (EINTR);
2090 		}
2091 
2092 		if (pr->ps_flags & PS_SINGLEEXIT) {
2093 			mtx_leave(&pr->ps_mtx);
2094 			KERNEL_LOCK();
2095 			exit1(p, 0, 0, EXIT_THREAD_NOCHECK);
2096 			/* NOTREACHED */
2097 		}
2098 
2099 		if (--pr->ps_singlecnt == 0)
2100 			wakeup(&pr->ps_singlecnt);
2101 
2102 		/* not exiting and don't need to unwind, so suspend */
2103 		mtx_leave(&pr->ps_mtx);
2104 
2105 		SCHED_LOCK();
2106 		p->p_stat = SSTOP;
2107 		mi_switch();
2108 		SCHED_UNLOCK();
2109 		mtx_enter(&pr->ps_mtx);
2110 	} while (pr->ps_single != NULL);
2111 
2112 	return (0);
2113 }
2114 
2115 int
2116 single_thread_check(struct proc *p, int deep)
2117 {
2118 	int error;
2119 
2120 	mtx_enter(&p->p_p->ps_mtx);
2121 	error = single_thread_check_locked(p, deep);
2122 	mtx_leave(&p->p_p->ps_mtx);
2123 
2124 	return error;
2125 }
2126 
2127 /*
2128  * Stop other threads in the process.  The mode controls how and
2129  * where the other threads should stop:
2130  *  - SINGLE_SUSPEND: stop wherever they are, will later be released (via
2131  *    single_thread_clear())
2132  *  - SINGLE_UNWIND: just unwind to kernel boundary, will be told to exit
2133  *    (by setting to SINGLE_EXIT) or released as with SINGLE_SUSPEND
2134  *  - SINGLE_EXIT: unwind to kernel boundary and exit
2135  */
2136 int
2137 single_thread_set(struct proc *p, int flags)
2138 {
2139 	struct process *pr = p->p_p;
2140 	struct proc *q;
2141 	int error, mode = flags & SINGLE_MASK;
2142 
2143 	KASSERT(curproc == p);
2144 
2145 	mtx_enter(&pr->ps_mtx);
2146 	error = single_thread_check_locked(p, flags & SINGLE_DEEP);
2147 	if (error) {
2148 		mtx_leave(&pr->ps_mtx);
2149 		return error;
2150 	}
2151 
2152 	switch (mode) {
2153 	case SINGLE_SUSPEND:
2154 		break;
2155 	case SINGLE_UNWIND:
2156 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
2157 		break;
2158 	case SINGLE_EXIT:
2159 		atomic_setbits_int(&pr->ps_flags, PS_SINGLEEXIT);
2160 		atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND);
2161 		break;
2162 #ifdef DIAGNOSTIC
2163 	default:
2164 		panic("single_thread_mode = %d", mode);
2165 #endif
2166 	}
2167 	KASSERT((p->p_flag & P_SUSPSINGLE) == 0);
2168 	pr->ps_single = p;
2169 	pr->ps_singlecnt = pr->ps_threadcnt;
2170 
2171 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
2172 		if (q == p)
2173 			continue;
2174 		SCHED_LOCK();
2175 		atomic_setbits_int(&q->p_flag, P_SUSPSINGLE);
2176 		switch (q->p_stat) {
2177 		case SSTOP:
2178 			if (mode == SINGLE_EXIT) {
2179 				unsleep(q);
2180 				setrunnable(q);
2181 			} else
2182 				--pr->ps_singlecnt;
2183 			break;
2184 		case SSLEEP:
2185 			/* if it's not interruptible, then just have to wait */
2186 			if (q->p_flag & P_SINTR) {
2187 				/* merely need to suspend?  just stop it */
2188 				if (mode == SINGLE_SUSPEND) {
2189 					q->p_stat = SSTOP;
2190 					--pr->ps_singlecnt;
2191 					break;
2192 				}
2193 				/* need to unwind or exit, so wake it */
2194 				unsleep(q);
2195 				setrunnable(q);
2196 			}
2197 			break;
2198 		case SONPROC:
2199 			signotify(q);
2200 			break;
2201 		case SRUN:
2202 		case SIDL:
2203 		case SDEAD:
2204 			break;
2205 		}
2206 		SCHED_UNLOCK();
2207 	}
2208 
2209 	/* count ourselfs out */
2210 	--pr->ps_singlecnt;
2211 	mtx_leave(&pr->ps_mtx);
2212 
2213 	if ((flags & SINGLE_NOWAIT) == 0)
2214 		single_thread_wait(pr, 1);
2215 
2216 	return 0;
2217 }
2218 
2219 /*
2220  * Wait for other threads to stop. If recheck is false then the function
2221  * returns non-zero if the caller needs to restart the check else 0 is
2222  * returned. If recheck is true the return value is always 0.
2223  */
2224 int
2225 single_thread_wait(struct process *pr, int recheck)
2226 {
2227 	int wait;
2228 
2229 	/* wait until they're all suspended */
2230 	mtx_enter(&pr->ps_mtx);
2231 	while ((wait = pr->ps_singlecnt > 0)) {
2232 		msleep_nsec(&pr->ps_singlecnt, &pr->ps_mtx, PWAIT, "suspend",
2233 		    INFSLP);
2234 		if (!recheck)
2235 			break;
2236 	}
2237 	KASSERT((pr->ps_single->p_flag & P_SUSPSINGLE) == 0);
2238 	mtx_leave(&pr->ps_mtx);
2239 
2240 	return wait;
2241 }
2242 
2243 void
2244 single_thread_clear(struct proc *p, int flag)
2245 {
2246 	struct process *pr = p->p_p;
2247 	struct proc *q;
2248 
2249 	KASSERT(pr->ps_single == p);
2250 	KASSERT(curproc == p);
2251 
2252 	mtx_enter(&pr->ps_mtx);
2253 	pr->ps_single = NULL;
2254 	atomic_clearbits_int(&pr->ps_flags, PS_SINGLEUNWIND | PS_SINGLEEXIT);
2255 
2256 	TAILQ_FOREACH(q, &pr->ps_threads, p_thr_link) {
2257 		if (q == p || (q->p_flag & P_SUSPSINGLE) == 0)
2258 			continue;
2259 		atomic_clearbits_int(&q->p_flag, P_SUSPSINGLE);
2260 
2261 		/*
2262 		 * if the thread was only stopped for single threading
2263 		 * then clearing that either makes it runnable or puts
2264 		 * it back into some sleep queue
2265 		 */
2266 		SCHED_LOCK();
2267 		if (q->p_stat == SSTOP && (q->p_flag & flag) == 0) {
2268 			if (q->p_wchan == NULL)
2269 				setrunnable(q);
2270 			else {
2271 				atomic_clearbits_int(&q->p_flag, P_WSLEEP);
2272 				q->p_stat = SSLEEP;
2273 			}
2274 		}
2275 		SCHED_UNLOCK();
2276 	}
2277 	mtx_leave(&pr->ps_mtx);
2278 }
2279 
2280 void
2281 sigio_del(struct sigiolst *rmlist)
2282 {
2283 	struct sigio *sigio;
2284 
2285 	while ((sigio = LIST_FIRST(rmlist)) != NULL) {
2286 		LIST_REMOVE(sigio, sio_pgsigio);
2287 		crfree(sigio->sio_ucred);
2288 		free(sigio, M_SIGIO, sizeof(*sigio));
2289 	}
2290 }
2291 
2292 void
2293 sigio_unlink(struct sigio_ref *sir, struct sigiolst *rmlist)
2294 {
2295 	struct sigio *sigio;
2296 
2297 	MUTEX_ASSERT_LOCKED(&sigio_lock);
2298 
2299 	sigio = sir->sir_sigio;
2300 	if (sigio != NULL) {
2301 		KASSERT(sigio->sio_myref == sir);
2302 		sir->sir_sigio = NULL;
2303 
2304 		if (sigio->sio_pgid > 0)
2305 			sigio->sio_proc = NULL;
2306 		else
2307 			sigio->sio_pgrp = NULL;
2308 		LIST_REMOVE(sigio, sio_pgsigio);
2309 
2310 		LIST_INSERT_HEAD(rmlist, sigio, sio_pgsigio);
2311 	}
2312 }
2313 
2314 void
2315 sigio_free(struct sigio_ref *sir)
2316 {
2317 	struct sigiolst rmlist;
2318 
2319 	if (sir->sir_sigio == NULL)
2320 		return;
2321 
2322 	LIST_INIT(&rmlist);
2323 
2324 	mtx_enter(&sigio_lock);
2325 	sigio_unlink(sir, &rmlist);
2326 	mtx_leave(&sigio_lock);
2327 
2328 	sigio_del(&rmlist);
2329 }
2330 
2331 void
2332 sigio_freelist(struct sigiolst *sigiolst)
2333 {
2334 	struct sigiolst rmlist;
2335 	struct sigio *sigio;
2336 
2337 	if (LIST_EMPTY(sigiolst))
2338 		return;
2339 
2340 	LIST_INIT(&rmlist);
2341 
2342 	mtx_enter(&sigio_lock);
2343 	while ((sigio = LIST_FIRST(sigiolst)) != NULL)
2344 		sigio_unlink(sigio->sio_myref, &rmlist);
2345 	mtx_leave(&sigio_lock);
2346 
2347 	sigio_del(&rmlist);
2348 }
2349 
2350 int
2351 sigio_setown(struct sigio_ref *sir, u_long cmd, caddr_t data)
2352 {
2353 	struct sigiolst rmlist;
2354 	struct proc *p = curproc;
2355 	struct pgrp *pgrp = NULL;
2356 	struct process *pr = NULL;
2357 	struct sigio *sigio;
2358 	int error;
2359 	pid_t pgid = *(int *)data;
2360 
2361 	if (pgid == 0) {
2362 		sigio_free(sir);
2363 		return (0);
2364 	}
2365 
2366 	if (cmd == TIOCSPGRP) {
2367 		if (pgid < 0)
2368 			return (EINVAL);
2369 		pgid = -pgid;
2370 	}
2371 
2372 	sigio = malloc(sizeof(*sigio), M_SIGIO, M_WAITOK);
2373 	sigio->sio_pgid = pgid;
2374 	sigio->sio_ucred = crhold(p->p_ucred);
2375 	sigio->sio_myref = sir;
2376 
2377 	LIST_INIT(&rmlist);
2378 
2379 	/*
2380 	 * The kernel lock, and not sleeping between prfind()/pgfind() and
2381 	 * linking of the sigio ensure that the process or process group does
2382 	 * not disappear unexpectedly.
2383 	 */
2384 	KERNEL_LOCK();
2385 	mtx_enter(&sigio_lock);
2386 
2387 	if (pgid > 0) {
2388 		pr = prfind(pgid);
2389 		if (pr == NULL) {
2390 			error = ESRCH;
2391 			goto fail;
2392 		}
2393 
2394 		/*
2395 		 * Policy - Don't allow a process to FSETOWN a process
2396 		 * in another session.
2397 		 *
2398 		 * Remove this test to allow maximum flexibility or
2399 		 * restrict FSETOWN to the current process or process
2400 		 * group for maximum safety.
2401 		 */
2402 		if (pr->ps_session != p->p_p->ps_session) {
2403 			error = EPERM;
2404 			goto fail;
2405 		}
2406 
2407 		if ((pr->ps_flags & PS_EXITING) != 0) {
2408 			error = ESRCH;
2409 			goto fail;
2410 		}
2411 	} else /* if (pgid < 0) */ {
2412 		pgrp = pgfind(-pgid);
2413 		if (pgrp == NULL) {
2414 			error = ESRCH;
2415 			goto fail;
2416 		}
2417 
2418 		/*
2419 		 * Policy - Don't allow a process to FSETOWN a process
2420 		 * in another session.
2421 		 *
2422 		 * Remove this test to allow maximum flexibility or
2423 		 * restrict FSETOWN to the current process or process
2424 		 * group for maximum safety.
2425 		 */
2426 		if (pgrp->pg_session != p->p_p->ps_session) {
2427 			error = EPERM;
2428 			goto fail;
2429 		}
2430 	}
2431 
2432 	if (pgid > 0) {
2433 		sigio->sio_proc = pr;
2434 		LIST_INSERT_HEAD(&pr->ps_sigiolst, sigio, sio_pgsigio);
2435 	} else {
2436 		sigio->sio_pgrp = pgrp;
2437 		LIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio);
2438 	}
2439 
2440 	sigio_unlink(sir, &rmlist);
2441 	sir->sir_sigio = sigio;
2442 
2443 	mtx_leave(&sigio_lock);
2444 	KERNEL_UNLOCK();
2445 
2446 	sigio_del(&rmlist);
2447 
2448 	return (0);
2449 
2450 fail:
2451 	mtx_leave(&sigio_lock);
2452 	KERNEL_UNLOCK();
2453 
2454 	crfree(sigio->sio_ucred);
2455 	free(sigio, M_SIGIO, sizeof(*sigio));
2456 
2457 	return (error);
2458 }
2459 
2460 void
2461 sigio_getown(struct sigio_ref *sir, u_long cmd, caddr_t data)
2462 {
2463 	struct sigio *sigio;
2464 	pid_t pgid = 0;
2465 
2466 	mtx_enter(&sigio_lock);
2467 	sigio = sir->sir_sigio;
2468 	if (sigio != NULL)
2469 		pgid = sigio->sio_pgid;
2470 	mtx_leave(&sigio_lock);
2471 
2472 	if (cmd == TIOCGPGRP)
2473 		pgid = -pgid;
2474 
2475 	*(int *)data = pgid;
2476 }
2477 
2478 void
2479 sigio_copy(struct sigio_ref *dst, struct sigio_ref *src)
2480 {
2481 	struct sigiolst rmlist;
2482 	struct sigio *newsigio, *sigio;
2483 
2484 	sigio_free(dst);
2485 
2486 	if (src->sir_sigio == NULL)
2487 		return;
2488 
2489 	newsigio = malloc(sizeof(*newsigio), M_SIGIO, M_WAITOK);
2490 	LIST_INIT(&rmlist);
2491 
2492 	mtx_enter(&sigio_lock);
2493 
2494 	sigio = src->sir_sigio;
2495 	if (sigio == NULL) {
2496 		mtx_leave(&sigio_lock);
2497 		free(newsigio, M_SIGIO, sizeof(*newsigio));
2498 		return;
2499 	}
2500 
2501 	newsigio->sio_pgid = sigio->sio_pgid;
2502 	newsigio->sio_ucred = crhold(sigio->sio_ucred);
2503 	newsigio->sio_myref = dst;
2504 	if (newsigio->sio_pgid > 0) {
2505 		newsigio->sio_proc = sigio->sio_proc;
2506 		LIST_INSERT_HEAD(&newsigio->sio_proc->ps_sigiolst, newsigio,
2507 		    sio_pgsigio);
2508 	} else {
2509 		newsigio->sio_pgrp = sigio->sio_pgrp;
2510 		LIST_INSERT_HEAD(&newsigio->sio_pgrp->pg_sigiolst, newsigio,
2511 		    sio_pgsigio);
2512 	}
2513 
2514 	sigio_unlink(dst, &rmlist);
2515 	dst->sir_sigio = newsigio;
2516 
2517 	mtx_leave(&sigio_lock);
2518 
2519 	sigio_del(&rmlist);
2520 }
2521