1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)kern_sig.c	8.7 (Berkeley) 4/18/94
37  */
38 
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD: stable/12/sys/kern/kern_sig.c 370365 2021-08-23 00:44:32Z markj $");
41 
42 #include "opt_ktrace.h"
43 
44 #include <sys/param.h>
45 #include <sys/ctype.h>
46 #include <sys/systm.h>
47 #include <sys/signalvar.h>
48 #include <sys/vnode.h>
49 #include <sys/acct.h>
50 #include <sys/bus.h>
51 #include <sys/capsicum.h>
52 #include <sys/compressor.h>
53 #include <sys/condvar.h>
54 #include <sys/event.h>
55 #include <sys/fcntl.h>
56 #include <sys/imgact.h>
57 #include <sys/kernel.h>
58 #include <sys/ktr.h>
59 #include <sys/ktrace.h>
60 #include <sys/limits.h>
61 #include <sys/lock.h>
62 #include <sys/malloc.h>
63 #include <sys/mutex.h>
64 #include <sys/refcount.h>
65 #include <sys/namei.h>
66 #include <sys/proc.h>
67 #include <sys/procdesc.h>
68 #include <sys/ptrace.h>
69 #include <sys/posix4.h>
70 #include <sys/pioctl.h>
71 #include <sys/racct.h>
72 #include <sys/resourcevar.h>
73 #include <sys/sdt.h>
74 #include <sys/sbuf.h>
75 #include <sys/sleepqueue.h>
76 #include <sys/smp.h>
77 #include <sys/stat.h>
78 #include <sys/sx.h>
79 #include <sys/syscallsubr.h>
80 #include <sys/sysctl.h>
81 #include <sys/sysent.h>
82 #include <sys/syslog.h>
83 #include <sys/sysproto.h>
84 #include <sys/timers.h>
85 #include <sys/unistd.h>
86 #include <sys/wait.h>
87 #include <vm/vm.h>
88 #include <vm/vm_extern.h>
89 #include <vm/uma.h>
90 
91 #include <sys/jail.h>
92 
93 #include <machine/cpu.h>
94 
95 #include <security/audit/audit.h>
96 
97 #define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
98 
99 SDT_PROVIDER_DECLARE(proc);
100 SDT_PROBE_DEFINE3(proc, , , signal__send,
101     "struct thread *", "struct proc *", "int");
102 SDT_PROBE_DEFINE2(proc, , , signal__clear,
103     "int", "ksiginfo_t *");
104 SDT_PROBE_DEFINE3(proc, , , signal__discard,
105     "struct thread *", "struct proc *", "int");
106 
107 static int	coredump(struct thread *);
108 static int	killpg1(struct thread *td, int sig, int pgid, int all,
109 		    ksiginfo_t *ksi);
110 static int	issignal(struct thread *td);
111 static int	sigprop(int sig);
112 static void	tdsigwakeup(struct thread *, int, sig_t, int);
113 static int	sig_suspend_threads(struct thread *, struct proc *, int);
114 static int	filt_sigattach(struct knote *kn);
115 static void	filt_sigdetach(struct knote *kn);
116 static int	filt_signal(struct knote *kn, long hint);
117 static struct thread *sigtd(struct proc *p, int sig, int prop);
118 static void	sigqueue_start(void);
119 
120 static uma_zone_t	ksiginfo_zone = NULL;
121 struct filterops sig_filtops = {
122 	.f_isfd = 0,
123 	.f_attach = filt_sigattach,
124 	.f_detach = filt_sigdetach,
125 	.f_event = filt_signal,
126 };
127 
128 static int	kern_logsigexit = 1;
129 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
130     &kern_logsigexit, 0,
131     "Log processes quitting on abnormal signals to syslog(3)");
132 
133 static int	kern_forcesigexit = 1;
134 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
135     &kern_forcesigexit, 0, "Force trap signal to be handled");
136 
137 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW, 0,
138     "POSIX real time signal");
139 
140 static int	max_pending_per_proc = 128;
141 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
142     &max_pending_per_proc, 0, "Max pending signals per proc");
143 
144 static int	preallocate_siginfo = 1024;
145 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
146     &preallocate_siginfo, 0, "Preallocated signal memory size");
147 
148 static int	signal_overflow = 0;
149 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
150     &signal_overflow, 0, "Number of signals overflew");
151 
152 static int	signal_alloc_fail = 0;
153 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
154     &signal_alloc_fail, 0, "signals failed to be allocated");
155 
156 static int	kern_lognosys = 0;
157 SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
158     "Log invalid syscalls");
159 
160 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
161 
162 /*
163  * Policy -- Can ucred cr1 send SIGIO to process cr2?
164  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
165  * in the right situations.
166  */
167 #define CANSIGIO(cr1, cr2) \
168 	((cr1)->cr_uid == 0 || \
169 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
170 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
171 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
172 	    (cr1)->cr_uid == (cr2)->cr_uid)
173 
174 static int	sugid_coredump;
175 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
176     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
177 
178 static int	capmode_coredump;
179 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
180     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
181 
182 static int	do_coredump = 1;
183 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
184 	&do_coredump, 0, "Enable/Disable coredumps");
185 
186 static int	set_core_nodump_flag = 0;
187 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
188 	0, "Enable setting the NODUMP flag on coredump files");
189 
190 static int	coredump_devctl = 0;
191 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
192 	0, "Generate a devctl notification when processes coredump");
193 
194 /*
195  * Signal properties and actions.
196  * The array below categorizes the signals and their default actions
197  * according to the following properties:
198  */
199 #define	SIGPROP_KILL		0x01	/* terminates process by default */
200 #define	SIGPROP_CORE		0x02	/* ditto and coredumps */
201 #define	SIGPROP_STOP		0x04	/* suspend process */
202 #define	SIGPROP_TTYSTOP		0x08	/* ditto, from tty */
203 #define	SIGPROP_IGNORE		0x10	/* ignore by default */
204 #define	SIGPROP_CONT		0x20	/* continue if suspended */
205 #define	SIGPROP_CANTMASK	0x40	/* non-maskable, catchable */
206 
207 static int sigproptbl[NSIG] = {
208 	[SIGHUP] =	SIGPROP_KILL,
209 	[SIGINT] =	SIGPROP_KILL,
210 	[SIGQUIT] =	SIGPROP_KILL | SIGPROP_CORE,
211 	[SIGILL] =	SIGPROP_KILL | SIGPROP_CORE,
212 	[SIGTRAP] =	SIGPROP_KILL | SIGPROP_CORE,
213 	[SIGABRT] =	SIGPROP_KILL | SIGPROP_CORE,
214 	[SIGEMT] =	SIGPROP_KILL | SIGPROP_CORE,
215 	[SIGFPE] =	SIGPROP_KILL | SIGPROP_CORE,
216 	[SIGKILL] =	SIGPROP_KILL,
217 	[SIGBUS] =	SIGPROP_KILL | SIGPROP_CORE,
218 	[SIGSEGV] =	SIGPROP_KILL | SIGPROP_CORE,
219 	[SIGSYS] =	SIGPROP_KILL | SIGPROP_CORE,
220 	[SIGPIPE] =	SIGPROP_KILL,
221 	[SIGALRM] =	SIGPROP_KILL,
222 	[SIGTERM] =	SIGPROP_KILL,
223 	[SIGURG] =	SIGPROP_IGNORE,
224 	[SIGSTOP] =	SIGPROP_STOP,
225 	[SIGTSTP] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
226 	[SIGCONT] =	SIGPROP_IGNORE | SIGPROP_CONT,
227 	[SIGCHLD] =	SIGPROP_IGNORE,
228 	[SIGTTIN] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
229 	[SIGTTOU] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
230 	[SIGIO] =	SIGPROP_IGNORE,
231 	[SIGXCPU] =	SIGPROP_KILL,
232 	[SIGXFSZ] =	SIGPROP_KILL,
233 	[SIGVTALRM] =	SIGPROP_KILL,
234 	[SIGPROF] =	SIGPROP_KILL,
235 	[SIGWINCH] =	SIGPROP_IGNORE,
236 	[SIGINFO] =	SIGPROP_IGNORE,
237 	[SIGUSR1] =	SIGPROP_KILL,
238 	[SIGUSR2] =	SIGPROP_KILL,
239 };
240 
241 static void reschedule_signals(struct proc *p, sigset_t block, int flags);
242 
243 static void
sigqueue_start(void)244 sigqueue_start(void)
245 {
246 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
247 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
248 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
249 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
250 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
251 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
252 }
253 
254 ksiginfo_t *
ksiginfo_alloc(int wait)255 ksiginfo_alloc(int wait)
256 {
257 	int flags;
258 
259 	flags = M_ZERO;
260 	if (! wait)
261 		flags |= M_NOWAIT;
262 	if (ksiginfo_zone != NULL)
263 		return ((ksiginfo_t *)uma_zalloc(ksiginfo_zone, flags));
264 	return (NULL);
265 }
266 
267 void
ksiginfo_free(ksiginfo_t * ksi)268 ksiginfo_free(ksiginfo_t *ksi)
269 {
270 	uma_zfree(ksiginfo_zone, ksi);
271 }
272 
273 static __inline int
ksiginfo_tryfree(ksiginfo_t * ksi)274 ksiginfo_tryfree(ksiginfo_t *ksi)
275 {
276 	if (!(ksi->ksi_flags & KSI_EXT)) {
277 		uma_zfree(ksiginfo_zone, ksi);
278 		return (1);
279 	}
280 	return (0);
281 }
282 
283 void
sigqueue_init(sigqueue_t * list,struct proc * p)284 sigqueue_init(sigqueue_t *list, struct proc *p)
285 {
286 	SIGEMPTYSET(list->sq_signals);
287 	SIGEMPTYSET(list->sq_kill);
288 	SIGEMPTYSET(list->sq_ptrace);
289 	TAILQ_INIT(&list->sq_list);
290 	list->sq_proc = p;
291 	list->sq_flags = SQ_INIT;
292 }
293 
294 /*
295  * Get a signal's ksiginfo.
296  * Return:
297  *	0	-	signal not found
298  *	others	-	signal number
299  */
300 static int
sigqueue_get(sigqueue_t * sq,int signo,ksiginfo_t * si)301 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
302 {
303 	struct proc *p = sq->sq_proc;
304 	struct ksiginfo *ksi, *next;
305 	int count = 0;
306 
307 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
308 
309 	if (!SIGISMEMBER(sq->sq_signals, signo))
310 		return (0);
311 
312 	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
313 		count++;
314 		SIGDELSET(sq->sq_ptrace, signo);
315 		si->ksi_flags |= KSI_PTRACE;
316 	}
317 	if (SIGISMEMBER(sq->sq_kill, signo)) {
318 		count++;
319 		if (count == 1)
320 			SIGDELSET(sq->sq_kill, signo);
321 	}
322 
323 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
324 		if (ksi->ksi_signo == signo) {
325 			if (count == 0) {
326 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
327 				ksi->ksi_sigq = NULL;
328 				ksiginfo_copy(ksi, si);
329 				if (ksiginfo_tryfree(ksi) && p != NULL)
330 					p->p_pendingcnt--;
331 			}
332 			if (++count > 1)
333 				break;
334 		}
335 	}
336 
337 	if (count <= 1)
338 		SIGDELSET(sq->sq_signals, signo);
339 	si->ksi_signo = signo;
340 	return (signo);
341 }
342 
343 void
sigqueue_take(ksiginfo_t * ksi)344 sigqueue_take(ksiginfo_t *ksi)
345 {
346 	struct ksiginfo *kp;
347 	struct proc	*p;
348 	sigqueue_t	*sq;
349 
350 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
351 		return;
352 
353 	p = sq->sq_proc;
354 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
355 	ksi->ksi_sigq = NULL;
356 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
357 		p->p_pendingcnt--;
358 
359 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
360 	     kp = TAILQ_NEXT(kp, ksi_link)) {
361 		if (kp->ksi_signo == ksi->ksi_signo)
362 			break;
363 	}
364 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
365 	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
366 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
367 }
368 
369 static int
sigqueue_add(sigqueue_t * sq,int signo,ksiginfo_t * si)370 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
371 {
372 	struct proc *p = sq->sq_proc;
373 	struct ksiginfo *ksi;
374 	int ret = 0;
375 
376 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
377 
378 	/*
379 	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
380 	 * for these signals.
381 	 */
382 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
383 		SIGADDSET(sq->sq_kill, signo);
384 		goto out_set_bit;
385 	}
386 
387 	/* directly insert the ksi, don't copy it */
388 	if (si->ksi_flags & KSI_INS) {
389 		if (si->ksi_flags & KSI_HEAD)
390 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
391 		else
392 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
393 		si->ksi_sigq = sq;
394 		goto out_set_bit;
395 	}
396 
397 	if (__predict_false(ksiginfo_zone == NULL)) {
398 		SIGADDSET(sq->sq_kill, signo);
399 		goto out_set_bit;
400 	}
401 
402 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
403 		signal_overflow++;
404 		ret = EAGAIN;
405 	} else if ((ksi = ksiginfo_alloc(0)) == NULL) {
406 		signal_alloc_fail++;
407 		ret = EAGAIN;
408 	} else {
409 		if (p != NULL)
410 			p->p_pendingcnt++;
411 		ksiginfo_copy(si, ksi);
412 		ksi->ksi_signo = signo;
413 		if (si->ksi_flags & KSI_HEAD)
414 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
415 		else
416 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
417 		ksi->ksi_sigq = sq;
418 	}
419 
420 	if (ret != 0) {
421 		if ((si->ksi_flags & KSI_PTRACE) != 0) {
422 			SIGADDSET(sq->sq_ptrace, signo);
423 			ret = 0;
424 			goto out_set_bit;
425 		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
426 		    (si->ksi_flags & KSI_SIGQ) == 0) {
427 			SIGADDSET(sq->sq_kill, signo);
428 			ret = 0;
429 			goto out_set_bit;
430 		}
431 		return (ret);
432 	}
433 
434 out_set_bit:
435 	SIGADDSET(sq->sq_signals, signo);
436 	return (ret);
437 }
438 
439 void
sigqueue_flush(sigqueue_t * sq)440 sigqueue_flush(sigqueue_t *sq)
441 {
442 	struct proc *p = sq->sq_proc;
443 	ksiginfo_t *ksi;
444 
445 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
446 
447 	if (p != NULL)
448 		PROC_LOCK_ASSERT(p, MA_OWNED);
449 
450 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
451 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
452 		ksi->ksi_sigq = NULL;
453 		if (ksiginfo_tryfree(ksi) && p != NULL)
454 			p->p_pendingcnt--;
455 	}
456 
457 	SIGEMPTYSET(sq->sq_signals);
458 	SIGEMPTYSET(sq->sq_kill);
459 	SIGEMPTYSET(sq->sq_ptrace);
460 }
461 
462 static void
sigqueue_move_set(sigqueue_t * src,sigqueue_t * dst,const sigset_t * set)463 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
464 {
465 	sigset_t tmp;
466 	struct proc *p1, *p2;
467 	ksiginfo_t *ksi, *next;
468 
469 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
470 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
471 	p1 = src->sq_proc;
472 	p2 = dst->sq_proc;
473 	/* Move siginfo to target list */
474 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
475 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
476 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
477 			if (p1 != NULL)
478 				p1->p_pendingcnt--;
479 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
480 			ksi->ksi_sigq = dst;
481 			if (p2 != NULL)
482 				p2->p_pendingcnt++;
483 		}
484 	}
485 
486 	/* Move pending bits to target list */
487 	tmp = src->sq_kill;
488 	SIGSETAND(tmp, *set);
489 	SIGSETOR(dst->sq_kill, tmp);
490 	SIGSETNAND(src->sq_kill, tmp);
491 
492 	tmp = src->sq_ptrace;
493 	SIGSETAND(tmp, *set);
494 	SIGSETOR(dst->sq_ptrace, tmp);
495 	SIGSETNAND(src->sq_ptrace, tmp);
496 
497 	tmp = src->sq_signals;
498 	SIGSETAND(tmp, *set);
499 	SIGSETOR(dst->sq_signals, tmp);
500 	SIGSETNAND(src->sq_signals, tmp);
501 }
502 
503 #if 0
504 static void
505 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
506 {
507 	sigset_t set;
508 
509 	SIGEMPTYSET(set);
510 	SIGADDSET(set, signo);
511 	sigqueue_move_set(src, dst, &set);
512 }
513 #endif
514 
515 static void
sigqueue_delete_set(sigqueue_t * sq,const sigset_t * set)516 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
517 {
518 	struct proc *p = sq->sq_proc;
519 	ksiginfo_t *ksi, *next;
520 
521 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
522 
523 	/* Remove siginfo queue */
524 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
525 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
526 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
527 			ksi->ksi_sigq = NULL;
528 			if (ksiginfo_tryfree(ksi) && p != NULL)
529 				p->p_pendingcnt--;
530 		}
531 	}
532 	SIGSETNAND(sq->sq_kill, *set);
533 	SIGSETNAND(sq->sq_ptrace, *set);
534 	SIGSETNAND(sq->sq_signals, *set);
535 }
536 
537 void
sigqueue_delete(sigqueue_t * sq,int signo)538 sigqueue_delete(sigqueue_t *sq, int signo)
539 {
540 	sigset_t set;
541 
542 	SIGEMPTYSET(set);
543 	SIGADDSET(set, signo);
544 	sigqueue_delete_set(sq, &set);
545 }
546 
547 /* Remove a set of signals for a process */
548 static void
sigqueue_delete_set_proc(struct proc * p,const sigset_t * set)549 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
550 {
551 	sigqueue_t worklist;
552 	struct thread *td0;
553 
554 	PROC_LOCK_ASSERT(p, MA_OWNED);
555 
556 	sigqueue_init(&worklist, NULL);
557 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
558 
559 	FOREACH_THREAD_IN_PROC(p, td0)
560 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
561 
562 	sigqueue_flush(&worklist);
563 }
564 
565 void
sigqueue_delete_proc(struct proc * p,int signo)566 sigqueue_delete_proc(struct proc *p, int signo)
567 {
568 	sigset_t set;
569 
570 	SIGEMPTYSET(set);
571 	SIGADDSET(set, signo);
572 	sigqueue_delete_set_proc(p, &set);
573 }
574 
575 static void
sigqueue_delete_stopmask_proc(struct proc * p)576 sigqueue_delete_stopmask_proc(struct proc *p)
577 {
578 	sigset_t set;
579 
580 	SIGEMPTYSET(set);
581 	SIGADDSET(set, SIGSTOP);
582 	SIGADDSET(set, SIGTSTP);
583 	SIGADDSET(set, SIGTTIN);
584 	SIGADDSET(set, SIGTTOU);
585 	sigqueue_delete_set_proc(p, &set);
586 }
587 
588 /*
589  * Determine signal that should be delivered to thread td, the current
590  * thread, 0 if none.  If there is a pending stop signal with default
591  * action, the process stops in issignal().
592  */
593 int
cursig(struct thread * td)594 cursig(struct thread *td)
595 {
596 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
597 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
598 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
599 	return (SIGPENDING(td) ? issignal(td) : 0);
600 }
601 
602 /*
603  * Arrange for ast() to handle unmasked pending signals on return to user
604  * mode.  This must be called whenever a signal is added to td_sigqueue or
605  * unmasked in td_sigmask.
606  */
607 void
signotify(struct thread * td)608 signotify(struct thread *td)
609 {
610 
611 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
612 
613 	if (SIGPENDING(td)) {
614 		thread_lock(td);
615 		td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING;
616 		thread_unlock(td);
617 	}
618 }
619 
620 /*
621  * Returns 1 (true) if altstack is configured for the thread, and the
622  * passed stack bottom address falls into the altstack range.  Handles
623  * the 43 compat special case where the alt stack size is zero.
624  */
625 int
sigonstack(size_t sp)626 sigonstack(size_t sp)
627 {
628 	struct thread *td;
629 
630 	td = curthread;
631 	if ((td->td_pflags & TDP_ALTSTACK) == 0)
632 		return (0);
633 #if defined(COMPAT_43)
634 	if (SV_PROC_FLAG(td->td_proc, SV_AOUT) && td->td_sigstk.ss_size == 0)
635 		return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
636 #endif
637 	return (sp >= (size_t)td->td_sigstk.ss_sp &&
638 	    sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
639 }
640 
641 static __inline int
sigprop(int sig)642 sigprop(int sig)
643 {
644 
645 	if (sig > 0 && sig < nitems(sigproptbl))
646 		return (sigproptbl[sig]);
647 	return (0);
648 }
649 
650 int
sig_ffs(sigset_t * set)651 sig_ffs(sigset_t *set)
652 {
653 	int i;
654 
655 	for (i = 0; i < _SIG_WORDS; i++)
656 		if (set->__bits[i])
657 			return (ffs(set->__bits[i]) + (i * 32));
658 	return (0);
659 }
660 
661 static bool
sigact_flag_test(const struct sigaction * act,int flag)662 sigact_flag_test(const struct sigaction *act, int flag)
663 {
664 
665 	/*
666 	 * SA_SIGINFO is reset when signal disposition is set to
667 	 * ignore or default.  Other flags are kept according to user
668 	 * settings.
669 	 */
670 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
671 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
672 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
673 }
674 
675 /*
676  * kern_sigaction
677  * sigaction
678  * freebsd4_sigaction
679  * osigaction
680  */
681 int
kern_sigaction(struct thread * td,int sig,const struct sigaction * act,struct sigaction * oact,int flags)682 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
683     struct sigaction *oact, int flags)
684 {
685 	struct sigacts *ps;
686 	struct proc *p = td->td_proc;
687 
688 	if (!_SIG_VALID(sig))
689 		return (EINVAL);
690 	if (act != NULL && act->sa_handler != SIG_DFL &&
691 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
692 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
693 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
694 		return (EINVAL);
695 
696 	PROC_LOCK(p);
697 	ps = p->p_sigacts;
698 	mtx_lock(&ps->ps_mtx);
699 	if (oact) {
700 		memset(oact, 0, sizeof(*oact));
701 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
702 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
703 			oact->sa_flags |= SA_ONSTACK;
704 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
705 			oact->sa_flags |= SA_RESTART;
706 		if (SIGISMEMBER(ps->ps_sigreset, sig))
707 			oact->sa_flags |= SA_RESETHAND;
708 		if (SIGISMEMBER(ps->ps_signodefer, sig))
709 			oact->sa_flags |= SA_NODEFER;
710 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
711 			oact->sa_flags |= SA_SIGINFO;
712 			oact->sa_sigaction =
713 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
714 		} else
715 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
716 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
717 			oact->sa_flags |= SA_NOCLDSTOP;
718 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
719 			oact->sa_flags |= SA_NOCLDWAIT;
720 	}
721 	if (act) {
722 		if ((sig == SIGKILL || sig == SIGSTOP) &&
723 		    act->sa_handler != SIG_DFL) {
724 			mtx_unlock(&ps->ps_mtx);
725 			PROC_UNLOCK(p);
726 			return (EINVAL);
727 		}
728 
729 		/*
730 		 * Change setting atomically.
731 		 */
732 
733 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
734 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
735 		if (sigact_flag_test(act, SA_SIGINFO)) {
736 			ps->ps_sigact[_SIG_IDX(sig)] =
737 			    (__sighandler_t *)act->sa_sigaction;
738 			SIGADDSET(ps->ps_siginfo, sig);
739 		} else {
740 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
741 			SIGDELSET(ps->ps_siginfo, sig);
742 		}
743 		if (!sigact_flag_test(act, SA_RESTART))
744 			SIGADDSET(ps->ps_sigintr, sig);
745 		else
746 			SIGDELSET(ps->ps_sigintr, sig);
747 		if (sigact_flag_test(act, SA_ONSTACK))
748 			SIGADDSET(ps->ps_sigonstack, sig);
749 		else
750 			SIGDELSET(ps->ps_sigonstack, sig);
751 		if (sigact_flag_test(act, SA_RESETHAND))
752 			SIGADDSET(ps->ps_sigreset, sig);
753 		else
754 			SIGDELSET(ps->ps_sigreset, sig);
755 		if (sigact_flag_test(act, SA_NODEFER))
756 			SIGADDSET(ps->ps_signodefer, sig);
757 		else
758 			SIGDELSET(ps->ps_signodefer, sig);
759 		if (sig == SIGCHLD) {
760 			if (act->sa_flags & SA_NOCLDSTOP)
761 				ps->ps_flag |= PS_NOCLDSTOP;
762 			else
763 				ps->ps_flag &= ~PS_NOCLDSTOP;
764 			if (act->sa_flags & SA_NOCLDWAIT) {
765 				/*
766 				 * Paranoia: since SA_NOCLDWAIT is implemented
767 				 * by reparenting the dying child to PID 1 (and
768 				 * trust it to reap the zombie), PID 1 itself
769 				 * is forbidden to set SA_NOCLDWAIT.
770 				 */
771 				if (p->p_pid == 1)
772 					ps->ps_flag &= ~PS_NOCLDWAIT;
773 				else
774 					ps->ps_flag |= PS_NOCLDWAIT;
775 			} else
776 				ps->ps_flag &= ~PS_NOCLDWAIT;
777 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
778 				ps->ps_flag |= PS_CLDSIGIGN;
779 			else
780 				ps->ps_flag &= ~PS_CLDSIGIGN;
781 		}
782 		/*
783 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
784 		 * and for signals set to SIG_DFL where the default is to
785 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
786 		 * have to restart the process.
787 		 */
788 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
789 		    (sigprop(sig) & SIGPROP_IGNORE &&
790 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
791 			/* never to be seen again */
792 			sigqueue_delete_proc(p, sig);
793 			if (sig != SIGCONT)
794 				/* easier in psignal */
795 				SIGADDSET(ps->ps_sigignore, sig);
796 			SIGDELSET(ps->ps_sigcatch, sig);
797 		} else {
798 			SIGDELSET(ps->ps_sigignore, sig);
799 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
800 				SIGDELSET(ps->ps_sigcatch, sig);
801 			else
802 				SIGADDSET(ps->ps_sigcatch, sig);
803 		}
804 #ifdef COMPAT_FREEBSD4
805 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
806 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
807 		    (flags & KSA_FREEBSD4) == 0)
808 			SIGDELSET(ps->ps_freebsd4, sig);
809 		else
810 			SIGADDSET(ps->ps_freebsd4, sig);
811 #endif
812 #ifdef COMPAT_43
813 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
814 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
815 		    (flags & KSA_OSIGSET) == 0)
816 			SIGDELSET(ps->ps_osigset, sig);
817 		else
818 			SIGADDSET(ps->ps_osigset, sig);
819 #endif
820 	}
821 	mtx_unlock(&ps->ps_mtx);
822 	PROC_UNLOCK(p);
823 	return (0);
824 }
825 
826 #ifndef _SYS_SYSPROTO_H_
827 struct sigaction_args {
828 	int	sig;
829 	struct	sigaction *act;
830 	struct	sigaction *oact;
831 };
832 #endif
833 int
sys_sigaction(struct thread * td,struct sigaction_args * uap)834 sys_sigaction(struct thread *td, struct sigaction_args *uap)
835 {
836 	struct sigaction act, oact;
837 	struct sigaction *actp, *oactp;
838 	int error;
839 
840 	actp = (uap->act != NULL) ? &act : NULL;
841 	oactp = (uap->oact != NULL) ? &oact : NULL;
842 	if (actp) {
843 		error = copyin(uap->act, actp, sizeof(act));
844 		if (error)
845 			return (error);
846 	}
847 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
848 	if (oactp && !error)
849 		error = copyout(oactp, uap->oact, sizeof(oact));
850 	return (error);
851 }
852 
853 #ifdef COMPAT_FREEBSD4
854 #ifndef _SYS_SYSPROTO_H_
855 struct freebsd4_sigaction_args {
856 	int	sig;
857 	struct	sigaction *act;
858 	struct	sigaction *oact;
859 };
860 #endif
861 int
freebsd4_sigaction(struct thread * td,struct freebsd4_sigaction_args * uap)862 freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
863 {
864 	struct sigaction act, oact;
865 	struct sigaction *actp, *oactp;
866 	int error;
867 
868 
869 	actp = (uap->act != NULL) ? &act : NULL;
870 	oactp = (uap->oact != NULL) ? &oact : NULL;
871 	if (actp) {
872 		error = copyin(uap->act, actp, sizeof(act));
873 		if (error)
874 			return (error);
875 	}
876 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
877 	if (oactp && !error)
878 		error = copyout(oactp, uap->oact, sizeof(oact));
879 	return (error);
880 }
881 #endif	/* COMAPT_FREEBSD4 */
882 
883 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
884 #ifndef _SYS_SYSPROTO_H_
885 struct osigaction_args {
886 	int	signum;
887 	struct	osigaction *nsa;
888 	struct	osigaction *osa;
889 };
890 #endif
891 int
osigaction(struct thread * td,struct osigaction_args * uap)892 osigaction(struct thread *td, struct osigaction_args *uap)
893 {
894 	struct osigaction sa;
895 	struct sigaction nsa, osa;
896 	struct sigaction *nsap, *osap;
897 	int error;
898 
899 	if (uap->signum <= 0 || uap->signum >= ONSIG)
900 		return (EINVAL);
901 
902 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
903 	osap = (uap->osa != NULL) ? &osa : NULL;
904 
905 	if (nsap) {
906 		error = copyin(uap->nsa, &sa, sizeof(sa));
907 		if (error)
908 			return (error);
909 		nsap->sa_handler = sa.sa_handler;
910 		nsap->sa_flags = sa.sa_flags;
911 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
912 	}
913 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
914 	if (osap && !error) {
915 		sa.sa_handler = osap->sa_handler;
916 		sa.sa_flags = osap->sa_flags;
917 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
918 		error = copyout(&sa, uap->osa, sizeof(sa));
919 	}
920 	return (error);
921 }
922 
923 #if !defined(__i386__)
924 /* Avoid replicating the same stub everywhere */
925 int
osigreturn(struct thread * td,struct osigreturn_args * uap)926 osigreturn(struct thread *td, struct osigreturn_args *uap)
927 {
928 
929 	return (nosys(td, (struct nosys_args *)uap));
930 }
931 #endif
932 #endif /* COMPAT_43 */
933 
934 /*
935  * Initialize signal state for process 0;
936  * set to ignore signals that are ignored by default.
937  */
938 void
siginit(struct proc * p)939 siginit(struct proc *p)
940 {
941 	int i;
942 	struct sigacts *ps;
943 
944 	PROC_LOCK(p);
945 	ps = p->p_sigacts;
946 	mtx_lock(&ps->ps_mtx);
947 	for (i = 1; i <= NSIG; i++) {
948 		if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) {
949 			SIGADDSET(ps->ps_sigignore, i);
950 		}
951 	}
952 	mtx_unlock(&ps->ps_mtx);
953 	PROC_UNLOCK(p);
954 }
955 
956 /*
957  * Reset specified signal to the default disposition.
958  */
959 static void
sigdflt(struct sigacts * ps,int sig)960 sigdflt(struct sigacts *ps, int sig)
961 {
962 
963 	mtx_assert(&ps->ps_mtx, MA_OWNED);
964 	SIGDELSET(ps->ps_sigcatch, sig);
965 	if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT)
966 		SIGADDSET(ps->ps_sigignore, sig);
967 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
968 	SIGDELSET(ps->ps_siginfo, sig);
969 }
970 
971 /*
972  * Reset signals for an exec of the specified process.
973  */
974 void
execsigs(struct proc * p)975 execsigs(struct proc *p)
976 {
977 	sigset_t osigignore;
978 	struct sigacts *ps;
979 	int sig;
980 	struct thread *td;
981 
982 	/*
983 	 * Reset caught signals.  Held signals remain held
984 	 * through td_sigmask (unless they were caught,
985 	 * and are now ignored by default).
986 	 */
987 	PROC_LOCK_ASSERT(p, MA_OWNED);
988 	ps = p->p_sigacts;
989 	mtx_lock(&ps->ps_mtx);
990 	sig_drop_caught(p);
991 
992 	/*
993 	 * As CloudABI processes cannot modify signal handlers, fully
994 	 * reset all signals to their default behavior. Do ignore
995 	 * SIGPIPE, as it would otherwise be impossible to recover from
996 	 * writes to broken pipes and sockets.
997 	 */
998 	if (SV_PROC_ABI(p) == SV_ABI_CLOUDABI) {
999 		osigignore = ps->ps_sigignore;
1000 		while (SIGNOTEMPTY(osigignore)) {
1001 			sig = sig_ffs(&osigignore);
1002 			SIGDELSET(osigignore, sig);
1003 			if (sig != SIGPIPE)
1004 				sigdflt(ps, sig);
1005 		}
1006 		SIGADDSET(ps->ps_sigignore, SIGPIPE);
1007 	}
1008 
1009 	/*
1010 	 * Reset stack state to the user stack.
1011 	 * Clear set of signals caught on the signal stack.
1012 	 */
1013 	td = curthread;
1014 	MPASS(td->td_proc == p);
1015 	td->td_sigstk.ss_flags = SS_DISABLE;
1016 	td->td_sigstk.ss_size = 0;
1017 	td->td_sigstk.ss_sp = 0;
1018 	td->td_pflags &= ~TDP_ALTSTACK;
1019 	/*
1020 	 * Reset no zombies if child dies flag as Solaris does.
1021 	 */
1022 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1023 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1024 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1025 	mtx_unlock(&ps->ps_mtx);
1026 }
1027 
1028 /*
1029  * kern_sigprocmask()
1030  *
1031  *	Manipulate signal mask.
1032  */
1033 int
kern_sigprocmask(struct thread * td,int how,sigset_t * set,sigset_t * oset,int flags)1034 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1035     int flags)
1036 {
1037 	sigset_t new_block, oset1;
1038 	struct proc *p;
1039 	int error;
1040 
1041 	p = td->td_proc;
1042 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1043 		PROC_LOCK_ASSERT(p, MA_OWNED);
1044 	else
1045 		PROC_LOCK(p);
1046 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1047 	    ? MA_OWNED : MA_NOTOWNED);
1048 	if (oset != NULL)
1049 		*oset = td->td_sigmask;
1050 
1051 	error = 0;
1052 	if (set != NULL) {
1053 		switch (how) {
1054 		case SIG_BLOCK:
1055 			SIG_CANTMASK(*set);
1056 			oset1 = td->td_sigmask;
1057 			SIGSETOR(td->td_sigmask, *set);
1058 			new_block = td->td_sigmask;
1059 			SIGSETNAND(new_block, oset1);
1060 			break;
1061 		case SIG_UNBLOCK:
1062 			SIGSETNAND(td->td_sigmask, *set);
1063 			signotify(td);
1064 			goto out;
1065 		case SIG_SETMASK:
1066 			SIG_CANTMASK(*set);
1067 			oset1 = td->td_sigmask;
1068 			if (flags & SIGPROCMASK_OLD)
1069 				SIGSETLO(td->td_sigmask, *set);
1070 			else
1071 				td->td_sigmask = *set;
1072 			new_block = td->td_sigmask;
1073 			SIGSETNAND(new_block, oset1);
1074 			signotify(td);
1075 			break;
1076 		default:
1077 			error = EINVAL;
1078 			goto out;
1079 		}
1080 
1081 		/*
1082 		 * The new_block set contains signals that were not previously
1083 		 * blocked, but are blocked now.
1084 		 *
1085 		 * In case we block any signal that was not previously blocked
1086 		 * for td, and process has the signal pending, try to schedule
1087 		 * signal delivery to some thread that does not block the
1088 		 * signal, possibly waking it up.
1089 		 */
1090 		if (p->p_numthreads != 1)
1091 			reschedule_signals(p, new_block, flags);
1092 	}
1093 
1094 out:
1095 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1096 		PROC_UNLOCK(p);
1097 	return (error);
1098 }
1099 
1100 #ifndef _SYS_SYSPROTO_H_
1101 struct sigprocmask_args {
1102 	int	how;
1103 	const sigset_t *set;
1104 	sigset_t *oset;
1105 };
1106 #endif
1107 int
sys_sigprocmask(struct thread * td,struct sigprocmask_args * uap)1108 sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1109 {
1110 	sigset_t set, oset;
1111 	sigset_t *setp, *osetp;
1112 	int error;
1113 
1114 	setp = (uap->set != NULL) ? &set : NULL;
1115 	osetp = (uap->oset != NULL) ? &oset : NULL;
1116 	if (setp) {
1117 		error = copyin(uap->set, setp, sizeof(set));
1118 		if (error)
1119 			return (error);
1120 	}
1121 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1122 	if (osetp && !error) {
1123 		error = copyout(osetp, uap->oset, sizeof(oset));
1124 	}
1125 	return (error);
1126 }
1127 
1128 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1129 #ifndef _SYS_SYSPROTO_H_
1130 struct osigprocmask_args {
1131 	int	how;
1132 	osigset_t mask;
1133 };
1134 #endif
1135 int
osigprocmask(struct thread * td,struct osigprocmask_args * uap)1136 osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1137 {
1138 	sigset_t set, oset;
1139 	int error;
1140 
1141 	OSIG2SIG(uap->mask, set);
1142 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1143 	SIG2OSIG(oset, td->td_retval[0]);
1144 	return (error);
1145 }
1146 #endif /* COMPAT_43 */
1147 
1148 int
sys_sigwait(struct thread * td,struct sigwait_args * uap)1149 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1150 {
1151 	ksiginfo_t ksi;
1152 	sigset_t set;
1153 	int error;
1154 
1155 	error = copyin(uap->set, &set, sizeof(set));
1156 	if (error) {
1157 		td->td_retval[0] = error;
1158 		return (0);
1159 	}
1160 
1161 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1162 	if (error) {
1163 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1164 			error = ERESTART;
1165 		if (error == ERESTART)
1166 			return (error);
1167 		td->td_retval[0] = error;
1168 		return (0);
1169 	}
1170 
1171 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1172 	td->td_retval[0] = error;
1173 	return (0);
1174 }
1175 
1176 int
sys_sigtimedwait(struct thread * td,struct sigtimedwait_args * uap)1177 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1178 {
1179 	struct timespec ts;
1180 	struct timespec *timeout;
1181 	sigset_t set;
1182 	ksiginfo_t ksi;
1183 	int error;
1184 
1185 	if (uap->timeout) {
1186 		error = copyin(uap->timeout, &ts, sizeof(ts));
1187 		if (error)
1188 			return (error);
1189 
1190 		timeout = &ts;
1191 	} else
1192 		timeout = NULL;
1193 
1194 	error = copyin(uap->set, &set, sizeof(set));
1195 	if (error)
1196 		return (error);
1197 
1198 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1199 	if (error)
1200 		return (error);
1201 
1202 	if (uap->info)
1203 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1204 
1205 	if (error == 0)
1206 		td->td_retval[0] = ksi.ksi_signo;
1207 	return (error);
1208 }
1209 
1210 int
sys_sigwaitinfo(struct thread * td,struct sigwaitinfo_args * uap)1211 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1212 {
1213 	ksiginfo_t ksi;
1214 	sigset_t set;
1215 	int error;
1216 
1217 	error = copyin(uap->set, &set, sizeof(set));
1218 	if (error)
1219 		return (error);
1220 
1221 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1222 	if (error)
1223 		return (error);
1224 
1225 	if (uap->info)
1226 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1227 
1228 	if (error == 0)
1229 		td->td_retval[0] = ksi.ksi_signo;
1230 	return (error);
1231 }
1232 
1233 static void
proc_td_siginfo_capture(struct thread * td,siginfo_t * si)1234 proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1235 {
1236 	struct thread *thr;
1237 
1238 	FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1239 		if (thr == td)
1240 			thr->td_si = *si;
1241 		else
1242 			thr->td_si.si_signo = 0;
1243 	}
1244 }
1245 
1246 int
kern_sigtimedwait(struct thread * td,sigset_t waitset,ksiginfo_t * ksi,struct timespec * timeout)1247 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1248 	struct timespec *timeout)
1249 {
1250 	struct sigacts *ps;
1251 	sigset_t saved_mask, new_block;
1252 	struct proc *p;
1253 	int error, sig, timo, timevalid = 0;
1254 	struct timespec rts, ets, ts;
1255 	struct timeval tv;
1256 	bool traced;
1257 
1258 	p = td->td_proc;
1259 	error = 0;
1260 	ets.tv_sec = 0;
1261 	ets.tv_nsec = 0;
1262 	traced = false;
1263 
1264 	if (timeout != NULL) {
1265 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1266 			timevalid = 1;
1267 			getnanouptime(&rts);
1268 			timespecadd(&rts, timeout, &ets);
1269 		}
1270 	}
1271 	ksiginfo_init(ksi);
1272 	/* Some signals can not be waited for. */
1273 	SIG_CANTMASK(waitset);
1274 	ps = p->p_sigacts;
1275 	PROC_LOCK(p);
1276 	saved_mask = td->td_sigmask;
1277 	SIGSETNAND(td->td_sigmask, waitset);
1278 	for (;;) {
1279 		mtx_lock(&ps->ps_mtx);
1280 		sig = cursig(td);
1281 		mtx_unlock(&ps->ps_mtx);
1282 		KASSERT(sig >= 0, ("sig %d", sig));
1283 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1284 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1285 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1286 				error = 0;
1287 				break;
1288 			}
1289 		}
1290 
1291 		if (error != 0)
1292 			break;
1293 
1294 		/*
1295 		 * POSIX says this must be checked after looking for pending
1296 		 * signals.
1297 		 */
1298 		if (timeout != NULL) {
1299 			if (!timevalid) {
1300 				error = EINVAL;
1301 				break;
1302 			}
1303 			getnanouptime(&rts);
1304 			if (timespeccmp(&rts, &ets, >=)) {
1305 				error = EAGAIN;
1306 				break;
1307 			}
1308 			timespecsub(&ets, &rts, &ts);
1309 			TIMESPEC_TO_TIMEVAL(&tv, &ts);
1310 			timo = tvtohz(&tv);
1311 		} else {
1312 			timo = 0;
1313 		}
1314 
1315 		if (traced) {
1316 			error = EINTR;
1317 			break;
1318 		}
1319 
1320 		error = msleep(&p->p_sigacts, &p->p_mtx, PPAUSE | PCATCH,
1321 		    "sigwait", timo);
1322 
1323 		if (timeout != NULL) {
1324 			if (error == ERESTART) {
1325 				/* Timeout can not be restarted. */
1326 				error = EINTR;
1327 			} else if (error == EAGAIN) {
1328 				/* We will calculate timeout by ourself. */
1329 				error = 0;
1330 			}
1331 		}
1332 
1333 		/*
1334 		 * If PTRACE_SCE or PTRACE_SCX were set after
1335 		 * userspace entered the syscall, return spurious
1336 		 * EINTR after wait was done.  Only do this as last
1337 		 * resort after rechecking for possible queued signals
1338 		 * and expired timeouts.
1339 		 */
1340 		if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1341 			traced = true;
1342 	}
1343 
1344 	new_block = saved_mask;
1345 	SIGSETNAND(new_block, td->td_sigmask);
1346 	td->td_sigmask = saved_mask;
1347 	/*
1348 	 * Fewer signals can be delivered to us, reschedule signal
1349 	 * notification.
1350 	 */
1351 	if (p->p_numthreads != 1)
1352 		reschedule_signals(p, new_block, 0);
1353 
1354 	if (error == 0) {
1355 		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1356 
1357 		if (ksi->ksi_code == SI_TIMER)
1358 			itimer_accept(p, ksi->ksi_timerid, ksi);
1359 
1360 #ifdef KTRACE
1361 		if (KTRPOINT(td, KTR_PSIG)) {
1362 			sig_t action;
1363 
1364 			mtx_lock(&ps->ps_mtx);
1365 			action = ps->ps_sigact[_SIG_IDX(sig)];
1366 			mtx_unlock(&ps->ps_mtx);
1367 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1368 		}
1369 #endif
1370 		if (sig == SIGKILL) {
1371 			proc_td_siginfo_capture(td, &ksi->ksi_info);
1372 			sigexit(td, sig);
1373 		}
1374 	}
1375 	PROC_UNLOCK(p);
1376 	return (error);
1377 }
1378 
1379 #ifndef _SYS_SYSPROTO_H_
1380 struct sigpending_args {
1381 	sigset_t	*set;
1382 };
1383 #endif
1384 int
sys_sigpending(struct thread * td,struct sigpending_args * uap)1385 sys_sigpending(struct thread *td, struct sigpending_args *uap)
1386 {
1387 	struct proc *p = td->td_proc;
1388 	sigset_t pending;
1389 
1390 	PROC_LOCK(p);
1391 	pending = p->p_sigqueue.sq_signals;
1392 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1393 	PROC_UNLOCK(p);
1394 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1395 }
1396 
1397 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1398 #ifndef _SYS_SYSPROTO_H_
1399 struct osigpending_args {
1400 	int	dummy;
1401 };
1402 #endif
1403 int
osigpending(struct thread * td,struct osigpending_args * uap)1404 osigpending(struct thread *td, struct osigpending_args *uap)
1405 {
1406 	struct proc *p = td->td_proc;
1407 	sigset_t pending;
1408 
1409 	PROC_LOCK(p);
1410 	pending = p->p_sigqueue.sq_signals;
1411 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1412 	PROC_UNLOCK(p);
1413 	SIG2OSIG(pending, td->td_retval[0]);
1414 	return (0);
1415 }
1416 #endif /* COMPAT_43 */
1417 
1418 #if defined(COMPAT_43)
1419 /*
1420  * Generalized interface signal handler, 4.3-compatible.
1421  */
1422 #ifndef _SYS_SYSPROTO_H_
1423 struct osigvec_args {
1424 	int	signum;
1425 	struct	sigvec *nsv;
1426 	struct	sigvec *osv;
1427 };
1428 #endif
1429 /* ARGSUSED */
1430 int
osigvec(struct thread * td,struct osigvec_args * uap)1431 osigvec(struct thread *td, struct osigvec_args *uap)
1432 {
1433 	struct sigvec vec;
1434 	struct sigaction nsa, osa;
1435 	struct sigaction *nsap, *osap;
1436 	int error;
1437 
1438 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1439 		return (EINVAL);
1440 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1441 	osap = (uap->osv != NULL) ? &osa : NULL;
1442 	if (nsap) {
1443 		error = copyin(uap->nsv, &vec, sizeof(vec));
1444 		if (error)
1445 			return (error);
1446 		nsap->sa_handler = vec.sv_handler;
1447 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1448 		nsap->sa_flags = vec.sv_flags;
1449 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1450 	}
1451 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1452 	if (osap && !error) {
1453 		vec.sv_handler = osap->sa_handler;
1454 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1455 		vec.sv_flags = osap->sa_flags;
1456 		vec.sv_flags &= ~SA_NOCLDWAIT;
1457 		vec.sv_flags ^= SA_RESTART;
1458 		error = copyout(&vec, uap->osv, sizeof(vec));
1459 	}
1460 	return (error);
1461 }
1462 
1463 #ifndef _SYS_SYSPROTO_H_
1464 struct osigblock_args {
1465 	int	mask;
1466 };
1467 #endif
1468 int
osigblock(struct thread * td,struct osigblock_args * uap)1469 osigblock(struct thread *td, struct osigblock_args *uap)
1470 {
1471 	sigset_t set, oset;
1472 
1473 	OSIG2SIG(uap->mask, set);
1474 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1475 	SIG2OSIG(oset, td->td_retval[0]);
1476 	return (0);
1477 }
1478 
1479 #ifndef _SYS_SYSPROTO_H_
1480 struct osigsetmask_args {
1481 	int	mask;
1482 };
1483 #endif
1484 int
osigsetmask(struct thread * td,struct osigsetmask_args * uap)1485 osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1486 {
1487 	sigset_t set, oset;
1488 
1489 	OSIG2SIG(uap->mask, set);
1490 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1491 	SIG2OSIG(oset, td->td_retval[0]);
1492 	return (0);
1493 }
1494 #endif /* COMPAT_43 */
1495 
1496 /*
1497  * Suspend calling thread until signal, providing mask to be set in the
1498  * meantime.
1499  */
1500 #ifndef _SYS_SYSPROTO_H_
1501 struct sigsuspend_args {
1502 	const sigset_t *sigmask;
1503 };
1504 #endif
1505 /* ARGSUSED */
1506 int
sys_sigsuspend(struct thread * td,struct sigsuspend_args * uap)1507 sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1508 {
1509 	sigset_t mask;
1510 	int error;
1511 
1512 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1513 	if (error)
1514 		return (error);
1515 	return (kern_sigsuspend(td, mask));
1516 }
1517 
1518 int
kern_sigsuspend(struct thread * td,sigset_t mask)1519 kern_sigsuspend(struct thread *td, sigset_t mask)
1520 {
1521 	struct proc *p = td->td_proc;
1522 	int has_sig, sig;
1523 
1524 	/*
1525 	 * When returning from sigsuspend, we want
1526 	 * the old mask to be restored after the
1527 	 * signal handler has finished.  Thus, we
1528 	 * save it here and mark the sigacts structure
1529 	 * to indicate this.
1530 	 */
1531 	PROC_LOCK(p);
1532 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1533 	    SIGPROCMASK_PROC_LOCKED);
1534 	td->td_pflags |= TDP_OLDMASK;
1535 
1536 	/*
1537 	 * Process signals now. Otherwise, we can get spurious wakeup
1538 	 * due to signal entered process queue, but delivered to other
1539 	 * thread. But sigsuspend should return only on signal
1540 	 * delivery.
1541 	 */
1542 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1543 	for (has_sig = 0; !has_sig;) {
1544 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1545 			0) == 0)
1546 			/* void */;
1547 		thread_suspend_check(0);
1548 		mtx_lock(&p->p_sigacts->ps_mtx);
1549 		while ((sig = cursig(td)) != 0) {
1550 			KASSERT(sig >= 0, ("sig %d", sig));
1551 			has_sig += postsig(sig);
1552 		}
1553 		mtx_unlock(&p->p_sigacts->ps_mtx);
1554 
1555 		/*
1556 		 * If PTRACE_SCE or PTRACE_SCX were set after
1557 		 * userspace entered the syscall, return spurious
1558 		 * EINTR.
1559 		 */
1560 		if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1561 			has_sig += 1;
1562 	}
1563 	PROC_UNLOCK(p);
1564 	td->td_errno = EINTR;
1565 	td->td_pflags |= TDP_NERRNO;
1566 	return (EJUSTRETURN);
1567 }
1568 
1569 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1570 /*
1571  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1572  * convention: libc stub passes mask, not pointer, to save a copyin.
1573  */
1574 #ifndef _SYS_SYSPROTO_H_
1575 struct osigsuspend_args {
1576 	osigset_t mask;
1577 };
1578 #endif
1579 /* ARGSUSED */
1580 int
osigsuspend(struct thread * td,struct osigsuspend_args * uap)1581 osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1582 {
1583 	sigset_t mask;
1584 
1585 	OSIG2SIG(uap->mask, mask);
1586 	return (kern_sigsuspend(td, mask));
1587 }
1588 #endif /* COMPAT_43 */
1589 
1590 #if defined(COMPAT_43)
1591 #ifndef _SYS_SYSPROTO_H_
1592 struct osigstack_args {
1593 	struct	sigstack *nss;
1594 	struct	sigstack *oss;
1595 };
1596 #endif
1597 /* ARGSUSED */
1598 int
osigstack(struct thread * td,struct osigstack_args * uap)1599 osigstack(struct thread *td, struct osigstack_args *uap)
1600 {
1601 	struct sigstack nss, oss;
1602 	int error = 0;
1603 
1604 	if (uap->nss != NULL) {
1605 		error = copyin(uap->nss, &nss, sizeof(nss));
1606 		if (error)
1607 			return (error);
1608 	}
1609 	oss.ss_sp = td->td_sigstk.ss_sp;
1610 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1611 	if (uap->nss != NULL) {
1612 		td->td_sigstk.ss_sp = nss.ss_sp;
1613 		td->td_sigstk.ss_size = 0;
1614 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1615 		td->td_pflags |= TDP_ALTSTACK;
1616 	}
1617 	if (uap->oss != NULL)
1618 		error = copyout(&oss, uap->oss, sizeof(oss));
1619 
1620 	return (error);
1621 }
1622 #endif /* COMPAT_43 */
1623 
1624 #ifndef _SYS_SYSPROTO_H_
1625 struct sigaltstack_args {
1626 	stack_t	*ss;
1627 	stack_t	*oss;
1628 };
1629 #endif
1630 /* ARGSUSED */
1631 int
sys_sigaltstack(struct thread * td,struct sigaltstack_args * uap)1632 sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1633 {
1634 	stack_t ss, oss;
1635 	int error;
1636 
1637 	if (uap->ss != NULL) {
1638 		error = copyin(uap->ss, &ss, sizeof(ss));
1639 		if (error)
1640 			return (error);
1641 	}
1642 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1643 	    (uap->oss != NULL) ? &oss : NULL);
1644 	if (error)
1645 		return (error);
1646 	if (uap->oss != NULL)
1647 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1648 	return (error);
1649 }
1650 
1651 int
kern_sigaltstack(struct thread * td,stack_t * ss,stack_t * oss)1652 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1653 {
1654 	struct proc *p = td->td_proc;
1655 	int oonstack;
1656 
1657 	oonstack = sigonstack(cpu_getstack(td));
1658 
1659 	if (oss != NULL) {
1660 		*oss = td->td_sigstk;
1661 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1662 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1663 	}
1664 
1665 	if (ss != NULL) {
1666 		if (oonstack)
1667 			return (EPERM);
1668 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1669 			return (EINVAL);
1670 		if (!(ss->ss_flags & SS_DISABLE)) {
1671 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1672 				return (ENOMEM);
1673 
1674 			td->td_sigstk = *ss;
1675 			td->td_pflags |= TDP_ALTSTACK;
1676 		} else {
1677 			td->td_pflags &= ~TDP_ALTSTACK;
1678 		}
1679 	}
1680 	return (0);
1681 }
1682 
1683 struct killpg1_ctx {
1684 	struct thread *td;
1685 	ksiginfo_t *ksi;
1686 	int sig;
1687 	bool sent;
1688 	bool found;
1689 	int ret;
1690 };
1691 
1692 static void
killpg1_sendsig(struct proc * p,bool notself,struct killpg1_ctx * arg)1693 killpg1_sendsig(struct proc *p, bool notself, struct killpg1_ctx *arg)
1694 {
1695 	int err;
1696 
1697 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1698 	    (notself && p == arg->td->td_proc) || p->p_state == PRS_NEW)
1699 		return;
1700 	PROC_LOCK(p);
1701 	err = p_cansignal(arg->td, p, arg->sig);
1702 	if (err == 0 && arg->sig != 0)
1703 		pksignal(p, arg->sig, arg->ksi);
1704 	PROC_UNLOCK(p);
1705 	if (err != ESRCH)
1706 		arg->found = true;
1707 	if (err == 0)
1708 		arg->sent = true;
1709 	else if (arg->ret == 0 && err != ESRCH && err != EPERM)
1710 		arg->ret = err;
1711 }
1712 
1713 /*
1714  * Common code for kill process group/broadcast kill.
1715  * cp is calling process.
1716  */
1717 static int
killpg1(struct thread * td,int sig,int pgid,int all,ksiginfo_t * ksi)1718 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1719 {
1720 	struct proc *p;
1721 	struct pgrp *pgrp;
1722 	struct killpg1_ctx arg;
1723 
1724 	arg.td = td;
1725 	arg.ksi = ksi;
1726 	arg.sig = sig;
1727 	arg.sent = false;
1728 	arg.found = false;
1729 	arg.ret = 0;
1730 	if (all) {
1731 		/*
1732 		 * broadcast
1733 		 */
1734 		sx_slock(&allproc_lock);
1735 		FOREACH_PROC_IN_SYSTEM(p) {
1736 			killpg1_sendsig(p, true, &arg);
1737 		}
1738 		sx_sunlock(&allproc_lock);
1739 	} else {
1740 		sx_slock(&proctree_lock);
1741 		if (pgid == 0) {
1742 			/*
1743 			 * zero pgid means send to my process group.
1744 			 */
1745 			pgrp = td->td_proc->p_pgrp;
1746 			PGRP_LOCK(pgrp);
1747 		} else {
1748 			pgrp = pgfind(pgid);
1749 			if (pgrp == NULL) {
1750 				sx_sunlock(&proctree_lock);
1751 				return (ESRCH);
1752 			}
1753 		}
1754 		sx_sunlock(&proctree_lock);
1755 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1756 			killpg1_sendsig(p, false, &arg);
1757 		}
1758 		PGRP_UNLOCK(pgrp);
1759 	}
1760 	MPASS(arg.ret != 0 || arg.found || !arg.sent);
1761 	if (arg.ret == 0 && !arg.sent)
1762 		arg.ret = arg.found ? EPERM : ESRCH;
1763 	return (arg.ret);
1764 }
1765 
1766 #ifndef _SYS_SYSPROTO_H_
1767 struct kill_args {
1768 	int	pid;
1769 	int	signum;
1770 };
1771 #endif
1772 /* ARGSUSED */
1773 int
sys_kill(struct thread * td,struct kill_args * uap)1774 sys_kill(struct thread *td, struct kill_args *uap)
1775 {
1776 	ksiginfo_t ksi;
1777 	struct proc *p;
1778 	int error;
1779 
1780 	/*
1781 	 * A process in capability mode can send signals only to himself.
1782 	 * The main rationale behind this is that abort(3) is implemented as
1783 	 * kill(getpid(), SIGABRT).
1784 	 */
1785 	if (IN_CAPABILITY_MODE(td) && uap->pid != td->td_proc->p_pid)
1786 		return (ECAPMODE);
1787 
1788 	AUDIT_ARG_SIGNUM(uap->signum);
1789 	AUDIT_ARG_PID(uap->pid);
1790 	if ((u_int)uap->signum > _SIG_MAXSIG)
1791 		return (EINVAL);
1792 
1793 	ksiginfo_init(&ksi);
1794 	ksi.ksi_signo = uap->signum;
1795 	ksi.ksi_code = SI_USER;
1796 	ksi.ksi_pid = td->td_proc->p_pid;
1797 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1798 
1799 	if (uap->pid > 0) {
1800 		/* kill single process */
1801 		if ((p = pfind_any(uap->pid)) == NULL)
1802 			return (ESRCH);
1803 		AUDIT_ARG_PROCESS(p);
1804 		error = p_cansignal(td, p, uap->signum);
1805 		if (error == 0 && uap->signum)
1806 			pksignal(p, uap->signum, &ksi);
1807 		PROC_UNLOCK(p);
1808 		return (error);
1809 	}
1810 	switch (uap->pid) {
1811 	case -1:		/* broadcast signal */
1812 		return (killpg1(td, uap->signum, 0, 1, &ksi));
1813 	case 0:			/* signal own process group */
1814 		return (killpg1(td, uap->signum, 0, 0, &ksi));
1815 	default:		/* negative explicit process group */
1816 		return (killpg1(td, uap->signum, -uap->pid, 0, &ksi));
1817 	}
1818 	/* NOTREACHED */
1819 }
1820 
1821 int
sys_pdkill(struct thread * td,struct pdkill_args * uap)1822 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1823 {
1824 	struct proc *p;
1825 	int error;
1826 
1827 	AUDIT_ARG_SIGNUM(uap->signum);
1828 	AUDIT_ARG_FD(uap->fd);
1829 	if ((u_int)uap->signum > _SIG_MAXSIG)
1830 		return (EINVAL);
1831 
1832 	error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1833 	if (error)
1834 		return (error);
1835 	AUDIT_ARG_PROCESS(p);
1836 	error = p_cansignal(td, p, uap->signum);
1837 	if (error == 0 && uap->signum)
1838 		kern_psignal(p, uap->signum);
1839 	PROC_UNLOCK(p);
1840 	return (error);
1841 }
1842 
1843 #if defined(COMPAT_43)
1844 #ifndef _SYS_SYSPROTO_H_
1845 struct okillpg_args {
1846 	int	pgid;
1847 	int	signum;
1848 };
1849 #endif
1850 /* ARGSUSED */
1851 int
okillpg(struct thread * td,struct okillpg_args * uap)1852 okillpg(struct thread *td, struct okillpg_args *uap)
1853 {
1854 	ksiginfo_t ksi;
1855 
1856 	AUDIT_ARG_SIGNUM(uap->signum);
1857 	AUDIT_ARG_PID(uap->pgid);
1858 	if ((u_int)uap->signum > _SIG_MAXSIG)
1859 		return (EINVAL);
1860 
1861 	ksiginfo_init(&ksi);
1862 	ksi.ksi_signo = uap->signum;
1863 	ksi.ksi_code = SI_USER;
1864 	ksi.ksi_pid = td->td_proc->p_pid;
1865 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1866 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1867 }
1868 #endif /* COMPAT_43 */
1869 
1870 #ifndef _SYS_SYSPROTO_H_
1871 struct sigqueue_args {
1872 	pid_t pid;
1873 	int signum;
1874 	/* union sigval */ void *value;
1875 };
1876 #endif
1877 int
sys_sigqueue(struct thread * td,struct sigqueue_args * uap)1878 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1879 {
1880 	union sigval sv;
1881 
1882 	sv.sival_ptr = uap->value;
1883 
1884 	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
1885 }
1886 
1887 int
kern_sigqueue(struct thread * td,pid_t pid,int signum,union sigval * value)1888 kern_sigqueue(struct thread *td, pid_t pid, int signum, union sigval *value)
1889 {
1890 	ksiginfo_t ksi;
1891 	struct proc *p;
1892 	int error;
1893 
1894 	if ((u_int)signum > _SIG_MAXSIG)
1895 		return (EINVAL);
1896 
1897 	/*
1898 	 * Specification says sigqueue can only send signal to
1899 	 * single process.
1900 	 */
1901 	if (pid <= 0)
1902 		return (EINVAL);
1903 
1904 	if ((p = pfind_any(pid)) == NULL)
1905 		return (ESRCH);
1906 	error = p_cansignal(td, p, signum);
1907 	if (error == 0 && signum != 0) {
1908 		ksiginfo_init(&ksi);
1909 		ksi.ksi_flags = KSI_SIGQ;
1910 		ksi.ksi_signo = signum;
1911 		ksi.ksi_code = SI_QUEUE;
1912 		ksi.ksi_pid = td->td_proc->p_pid;
1913 		ksi.ksi_uid = td->td_ucred->cr_ruid;
1914 		ksi.ksi_value = *value;
1915 		error = pksignal(p, ksi.ksi_signo, &ksi);
1916 	}
1917 	PROC_UNLOCK(p);
1918 	return (error);
1919 }
1920 
1921 /*
1922  * Send a signal to a process group.
1923  */
1924 void
gsignal(int pgid,int sig,ksiginfo_t * ksi)1925 gsignal(int pgid, int sig, ksiginfo_t *ksi)
1926 {
1927 	struct pgrp *pgrp;
1928 
1929 	if (pgid != 0) {
1930 		sx_slock(&proctree_lock);
1931 		pgrp = pgfind(pgid);
1932 		sx_sunlock(&proctree_lock);
1933 		if (pgrp != NULL) {
1934 			pgsignal(pgrp, sig, 0, ksi);
1935 			PGRP_UNLOCK(pgrp);
1936 		}
1937 	}
1938 }
1939 
1940 /*
1941  * Send a signal to a process group.  If checktty is 1,
1942  * limit to members which have a controlling terminal.
1943  */
1944 void
pgsignal(struct pgrp * pgrp,int sig,int checkctty,ksiginfo_t * ksi)1945 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
1946 {
1947 	struct proc *p;
1948 
1949 	if (pgrp) {
1950 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
1951 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1952 			PROC_LOCK(p);
1953 			if (p->p_state == PRS_NORMAL &&
1954 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
1955 				pksignal(p, sig, ksi);
1956 			PROC_UNLOCK(p);
1957 		}
1958 	}
1959 }
1960 
1961 
1962 /*
1963  * Recalculate the signal mask and reset the signal disposition after
1964  * usermode frame for delivery is formed.  Should be called after
1965  * mach-specific routine, because sysent->sv_sendsig() needs correct
1966  * ps_siginfo and signal mask.
1967  */
1968 static void
postsig_done(int sig,struct thread * td,struct sigacts * ps)1969 postsig_done(int sig, struct thread *td, struct sigacts *ps)
1970 {
1971 	sigset_t mask;
1972 
1973 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1974 	td->td_ru.ru_nsignals++;
1975 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
1976 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
1977 		SIGADDSET(mask, sig);
1978 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
1979 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
1980 	if (SIGISMEMBER(ps->ps_sigreset, sig))
1981 		sigdflt(ps, sig);
1982 }
1983 
1984 
1985 /*
1986  * Send a signal caused by a trap to the current thread.  If it will be
1987  * caught immediately, deliver it with correct code.  Otherwise, post it
1988  * normally.
1989  */
1990 void
trapsignal(struct thread * td,ksiginfo_t * ksi)1991 trapsignal(struct thread *td, ksiginfo_t *ksi)
1992 {
1993 	struct sigacts *ps;
1994 	struct proc *p;
1995 	int sig;
1996 	int code;
1997 
1998 	p = td->td_proc;
1999 	sig = ksi->ksi_signo;
2000 	code = ksi->ksi_code;
2001 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
2002 
2003 	PROC_LOCK(p);
2004 	ps = p->p_sigacts;
2005 	mtx_lock(&ps->ps_mtx);
2006 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2007 	    !SIGISMEMBER(td->td_sigmask, sig)) {
2008 #ifdef KTRACE
2009 		if (KTRPOINT(curthread, KTR_PSIG))
2010 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2011 			    &td->td_sigmask, code);
2012 #endif
2013 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2014 				ksi, &td->td_sigmask);
2015 		postsig_done(sig, td, ps);
2016 		mtx_unlock(&ps->ps_mtx);
2017 	} else {
2018 		/*
2019 		 * Avoid a possible infinite loop if the thread
2020 		 * masking the signal or process is ignoring the
2021 		 * signal.
2022 		 */
2023 		if (kern_forcesigexit &&
2024 		    (SIGISMEMBER(td->td_sigmask, sig) ||
2025 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2026 			SIGDELSET(td->td_sigmask, sig);
2027 			SIGDELSET(ps->ps_sigcatch, sig);
2028 			SIGDELSET(ps->ps_sigignore, sig);
2029 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2030 		}
2031 		mtx_unlock(&ps->ps_mtx);
2032 		p->p_code = code;	/* XXX for core dump/debugger */
2033 		p->p_sig = sig;		/* XXX to verify code */
2034 		tdsendsignal(p, td, sig, ksi);
2035 	}
2036 	PROC_UNLOCK(p);
2037 }
2038 
2039 static struct thread *
sigtd(struct proc * p,int sig,int prop)2040 sigtd(struct proc *p, int sig, int prop)
2041 {
2042 	struct thread *td, *signal_td;
2043 
2044 	PROC_LOCK_ASSERT(p, MA_OWNED);
2045 
2046 	/*
2047 	 * Check if current thread can handle the signal without
2048 	 * switching context to another thread.
2049 	 */
2050 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig))
2051 		return (curthread);
2052 	signal_td = NULL;
2053 	FOREACH_THREAD_IN_PROC(p, td) {
2054 		if (!SIGISMEMBER(td->td_sigmask, sig)) {
2055 			signal_td = td;
2056 			break;
2057 		}
2058 	}
2059 	if (signal_td == NULL)
2060 		signal_td = FIRST_THREAD_IN_PROC(p);
2061 	return (signal_td);
2062 }
2063 
2064 /*
2065  * Send the signal to the process.  If the signal has an action, the action
2066  * is usually performed by the target process rather than the caller; we add
2067  * the signal to the set of pending signals for the process.
2068  *
2069  * Exceptions:
2070  *   o When a stop signal is sent to a sleeping process that takes the
2071  *     default action, the process is stopped without awakening it.
2072  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2073  *     regardless of the signal action (eg, blocked or ignored).
2074  *
2075  * Other ignored signals are discarded immediately.
2076  *
2077  * NB: This function may be entered from the debugger via the "kill" DDB
2078  * command.  There is little that can be done to mitigate the possibly messy
2079  * side effects of this unwise possibility.
2080  */
2081 void
kern_psignal(struct proc * p,int sig)2082 kern_psignal(struct proc *p, int sig)
2083 {
2084 	ksiginfo_t ksi;
2085 
2086 	ksiginfo_init(&ksi);
2087 	ksi.ksi_signo = sig;
2088 	ksi.ksi_code = SI_KERNEL;
2089 	(void) tdsendsignal(p, NULL, sig, &ksi);
2090 }
2091 
2092 int
pksignal(struct proc * p,int sig,ksiginfo_t * ksi)2093 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2094 {
2095 
2096 	return (tdsendsignal(p, NULL, sig, ksi));
2097 }
2098 
2099 /* Utility function for finding a thread to send signal event to. */
2100 int
sigev_findtd(struct proc * p,struct sigevent * sigev,struct thread ** ttd)2101 sigev_findtd(struct proc *p ,struct sigevent *sigev, struct thread **ttd)
2102 {
2103 	struct thread *td;
2104 
2105 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2106 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2107 		if (td == NULL)
2108 			return (ESRCH);
2109 		*ttd = td;
2110 	} else {
2111 		*ttd = NULL;
2112 		PROC_LOCK(p);
2113 	}
2114 	return (0);
2115 }
2116 
2117 void
tdsignal(struct thread * td,int sig)2118 tdsignal(struct thread *td, int sig)
2119 {
2120 	ksiginfo_t ksi;
2121 
2122 	ksiginfo_init(&ksi);
2123 	ksi.ksi_signo = sig;
2124 	ksi.ksi_code = SI_KERNEL;
2125 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2126 }
2127 
2128 void
tdksignal(struct thread * td,int sig,ksiginfo_t * ksi)2129 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2130 {
2131 
2132 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2133 }
2134 
2135 int
tdsendsignal(struct proc * p,struct thread * td,int sig,ksiginfo_t * ksi)2136 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2137 {
2138 	sig_t action;
2139 	sigqueue_t *sigqueue;
2140 	int prop;
2141 	struct sigacts *ps;
2142 	int intrval;
2143 	int ret = 0;
2144 	int wakeup_swapper;
2145 
2146 	MPASS(td == NULL || p == td->td_proc);
2147 	PROC_LOCK_ASSERT(p, MA_OWNED);
2148 
2149 	if (!_SIG_VALID(sig))
2150 		panic("%s(): invalid signal %d", __func__, sig);
2151 
2152 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2153 
2154 	/*
2155 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2156 	 */
2157 	if (p->p_state == PRS_ZOMBIE) {
2158 		if (ksi && (ksi->ksi_flags & KSI_INS))
2159 			ksiginfo_tryfree(ksi);
2160 		return (ret);
2161 	}
2162 
2163 	ps = p->p_sigacts;
2164 	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2165 	prop = sigprop(sig);
2166 
2167 	if (td == NULL) {
2168 		td = sigtd(p, sig, prop);
2169 		sigqueue = &p->p_sigqueue;
2170 	} else
2171 		sigqueue = &td->td_sigqueue;
2172 
2173 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2174 
2175 	/*
2176 	 * If the signal is being ignored,
2177 	 * then we forget about it immediately.
2178 	 * (Note: we don't set SIGCONT in ps_sigignore,
2179 	 * and if it is set to SIG_IGN,
2180 	 * action will be SIG_DFL here.)
2181 	 */
2182 	mtx_lock(&ps->ps_mtx);
2183 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2184 		SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2185 
2186 		mtx_unlock(&ps->ps_mtx);
2187 		if (ksi && (ksi->ksi_flags & KSI_INS))
2188 			ksiginfo_tryfree(ksi);
2189 		return (ret);
2190 	}
2191 	if (SIGISMEMBER(td->td_sigmask, sig))
2192 		action = SIG_HOLD;
2193 	else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2194 		action = SIG_CATCH;
2195 	else
2196 		action = SIG_DFL;
2197 	if (SIGISMEMBER(ps->ps_sigintr, sig))
2198 		intrval = EINTR;
2199 	else
2200 		intrval = ERESTART;
2201 	mtx_unlock(&ps->ps_mtx);
2202 
2203 	if (prop & SIGPROP_CONT)
2204 		sigqueue_delete_stopmask_proc(p);
2205 	else if (prop & SIGPROP_STOP) {
2206 		/*
2207 		 * If sending a tty stop signal to a member of an orphaned
2208 		 * process group, discard the signal here if the action
2209 		 * is default; don't stop the process below if sleeping,
2210 		 * and don't clear any pending SIGCONT.
2211 		 */
2212 		if ((prop & SIGPROP_TTYSTOP) != 0 &&
2213 		    (p->p_pgrp->pg_flags & PGRP_ORPHANED) != 0 &&
2214 		    action == SIG_DFL) {
2215 			if (ksi && (ksi->ksi_flags & KSI_INS))
2216 				ksiginfo_tryfree(ksi);
2217 			return (ret);
2218 		}
2219 		sigqueue_delete_proc(p, SIGCONT);
2220 		if (p->p_flag & P_CONTINUED) {
2221 			p->p_flag &= ~P_CONTINUED;
2222 			PROC_LOCK(p->p_pptr);
2223 			sigqueue_take(p->p_ksi);
2224 			PROC_UNLOCK(p->p_pptr);
2225 		}
2226 	}
2227 
2228 	ret = sigqueue_add(sigqueue, sig, ksi);
2229 	if (ret != 0)
2230 		return (ret);
2231 	signotify(td);
2232 	/*
2233 	 * Defer further processing for signals which are held,
2234 	 * except that stopped processes must be continued by SIGCONT.
2235 	 */
2236 	if (action == SIG_HOLD &&
2237 	    !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2238 		return (ret);
2239 
2240 	/* SIGKILL: Remove procfs STOPEVENTs. */
2241 	if (sig == SIGKILL) {
2242 		/* from procfs_ioctl.c: PIOCBIC */
2243 		p->p_stops = 0;
2244 		/* from procfs_ioctl.c: PIOCCONT */
2245 		p->p_step = 0;
2246 		wakeup(&p->p_step);
2247 	}
2248 	/*
2249 	 * Some signals have a process-wide effect and a per-thread
2250 	 * component.  Most processing occurs when the process next
2251 	 * tries to cross the user boundary, however there are some
2252 	 * times when processing needs to be done immediately, such as
2253 	 * waking up threads so that they can cross the user boundary.
2254 	 * We try to do the per-process part here.
2255 	 */
2256 	if (P_SHOULDSTOP(p)) {
2257 		KASSERT(!(p->p_flag & P_WEXIT),
2258 		    ("signal to stopped but exiting process"));
2259 		if (sig == SIGKILL) {
2260 			/*
2261 			 * If traced process is already stopped,
2262 			 * then no further action is necessary.
2263 			 */
2264 			if (p->p_flag & P_TRACED)
2265 				goto out;
2266 			/*
2267 			 * SIGKILL sets process running.
2268 			 * It will die elsewhere.
2269 			 * All threads must be restarted.
2270 			 */
2271 			p->p_flag &= ~P_STOPPED_SIG;
2272 			goto runfast;
2273 		}
2274 
2275 		if (prop & SIGPROP_CONT) {
2276 			/*
2277 			 * If traced process is already stopped,
2278 			 * then no further action is necessary.
2279 			 */
2280 			if (p->p_flag & P_TRACED)
2281 				goto out;
2282 			/*
2283 			 * If SIGCONT is default (or ignored), we continue the
2284 			 * process but don't leave the signal in sigqueue as
2285 			 * it has no further action.  If SIGCONT is held, we
2286 			 * continue the process and leave the signal in
2287 			 * sigqueue.  If the process catches SIGCONT, let it
2288 			 * handle the signal itself.  If it isn't waiting on
2289 			 * an event, it goes back to run state.
2290 			 * Otherwise, process goes back to sleep state.
2291 			 */
2292 			p->p_flag &= ~P_STOPPED_SIG;
2293 			PROC_SLOCK(p);
2294 			if (p->p_numthreads == p->p_suspcount) {
2295 				PROC_SUNLOCK(p);
2296 				p->p_flag |= P_CONTINUED;
2297 				p->p_xsig = SIGCONT;
2298 				PROC_LOCK(p->p_pptr);
2299 				childproc_continued(p);
2300 				PROC_UNLOCK(p->p_pptr);
2301 				PROC_SLOCK(p);
2302 			}
2303 			if (action == SIG_DFL) {
2304 				thread_unsuspend(p);
2305 				PROC_SUNLOCK(p);
2306 				sigqueue_delete(sigqueue, sig);
2307 				goto out;
2308 			}
2309 			if (action == SIG_CATCH) {
2310 				/*
2311 				 * The process wants to catch it so it needs
2312 				 * to run at least one thread, but which one?
2313 				 */
2314 				PROC_SUNLOCK(p);
2315 				goto runfast;
2316 			}
2317 			/*
2318 			 * The signal is not ignored or caught.
2319 			 */
2320 			thread_unsuspend(p);
2321 			PROC_SUNLOCK(p);
2322 			goto out;
2323 		}
2324 
2325 		if (prop & SIGPROP_STOP) {
2326 			/*
2327 			 * If traced process is already stopped,
2328 			 * then no further action is necessary.
2329 			 */
2330 			if (p->p_flag & P_TRACED)
2331 				goto out;
2332 			/*
2333 			 * Already stopped, don't need to stop again
2334 			 * (If we did the shell could get confused).
2335 			 * Just make sure the signal STOP bit set.
2336 			 */
2337 			p->p_flag |= P_STOPPED_SIG;
2338 			sigqueue_delete(sigqueue, sig);
2339 			goto out;
2340 		}
2341 
2342 		/*
2343 		 * All other kinds of signals:
2344 		 * If a thread is sleeping interruptibly, simulate a
2345 		 * wakeup so that when it is continued it will be made
2346 		 * runnable and can look at the signal.  However, don't make
2347 		 * the PROCESS runnable, leave it stopped.
2348 		 * It may run a bit until it hits a thread_suspend_check().
2349 		 */
2350 		wakeup_swapper = 0;
2351 		PROC_SLOCK(p);
2352 		thread_lock(td);
2353 		if (TD_ON_SLEEPQ(td) && (td->td_flags & TDF_SINTR))
2354 			wakeup_swapper = sleepq_abort(td, intrval);
2355 		thread_unlock(td);
2356 		PROC_SUNLOCK(p);
2357 		if (wakeup_swapper)
2358 			kick_proc0();
2359 		goto out;
2360 		/*
2361 		 * Mutexes are short lived. Threads waiting on them will
2362 		 * hit thread_suspend_check() soon.
2363 		 */
2364 	} else if (p->p_state == PRS_NORMAL) {
2365 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2366 			tdsigwakeup(td, sig, action, intrval);
2367 			goto out;
2368 		}
2369 
2370 		MPASS(action == SIG_DFL);
2371 
2372 		if (prop & SIGPROP_STOP) {
2373 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2374 				goto out;
2375 			p->p_flag |= P_STOPPED_SIG;
2376 			p->p_xsig = sig;
2377 			PROC_SLOCK(p);
2378 			wakeup_swapper = sig_suspend_threads(td, p, 1);
2379 			if (p->p_numthreads == p->p_suspcount) {
2380 				/*
2381 				 * only thread sending signal to another
2382 				 * process can reach here, if thread is sending
2383 				 * signal to its process, because thread does
2384 				 * not suspend itself here, p_numthreads
2385 				 * should never be equal to p_suspcount.
2386 				 */
2387 				thread_stopped(p);
2388 				PROC_SUNLOCK(p);
2389 				sigqueue_delete_proc(p, p->p_xsig);
2390 			} else
2391 				PROC_SUNLOCK(p);
2392 			if (wakeup_swapper)
2393 				kick_proc0();
2394 			goto out;
2395 		}
2396 	} else {
2397 		/* Not in "NORMAL" state. discard the signal. */
2398 		sigqueue_delete(sigqueue, sig);
2399 		goto out;
2400 	}
2401 
2402 	/*
2403 	 * The process is not stopped so we need to apply the signal to all the
2404 	 * running threads.
2405 	 */
2406 runfast:
2407 	tdsigwakeup(td, sig, action, intrval);
2408 	PROC_SLOCK(p);
2409 	thread_unsuspend(p);
2410 	PROC_SUNLOCK(p);
2411 out:
2412 	/* If we jump here, proc slock should not be owned. */
2413 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2414 	return (ret);
2415 }
2416 
2417 /*
2418  * The force of a signal has been directed against a single
2419  * thread.  We need to see what we can do about knocking it
2420  * out of any sleep it may be in etc.
2421  */
2422 static void
tdsigwakeup(struct thread * td,int sig,sig_t action,int intrval)2423 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2424 {
2425 	struct proc *p = td->td_proc;
2426 	int prop;
2427 	int wakeup_swapper;
2428 
2429 	wakeup_swapper = 0;
2430 	PROC_LOCK_ASSERT(p, MA_OWNED);
2431 	prop = sigprop(sig);
2432 
2433 	PROC_SLOCK(p);
2434 	thread_lock(td);
2435 	/*
2436 	 * Bring the priority of a thread up if we want it to get
2437 	 * killed in this lifetime.  Be careful to avoid bumping the
2438 	 * priority of the idle thread, since we still allow to signal
2439 	 * kernel processes.
2440 	 */
2441 	if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2442 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2443 		sched_prio(td, PUSER);
2444 	if (TD_ON_SLEEPQ(td)) {
2445 		/*
2446 		 * If thread is sleeping uninterruptibly
2447 		 * we can't interrupt the sleep... the signal will
2448 		 * be noticed when the process returns through
2449 		 * trap() or syscall().
2450 		 */
2451 		if ((td->td_flags & TDF_SINTR) == 0)
2452 			goto out;
2453 		/*
2454 		 * If SIGCONT is default (or ignored) and process is
2455 		 * asleep, we are finished; the process should not
2456 		 * be awakened.
2457 		 */
2458 		if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2459 			thread_unlock(td);
2460 			PROC_SUNLOCK(p);
2461 			sigqueue_delete(&p->p_sigqueue, sig);
2462 			/*
2463 			 * It may be on either list in this state.
2464 			 * Remove from both for now.
2465 			 */
2466 			sigqueue_delete(&td->td_sigqueue, sig);
2467 			return;
2468 		}
2469 
2470 		/*
2471 		 * Don't awaken a sleeping thread for SIGSTOP if the
2472 		 * STOP signal is deferred.
2473 		 */
2474 		if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2475 		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2476 			goto out;
2477 
2478 		/*
2479 		 * Give low priority threads a better chance to run.
2480 		 */
2481 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2482 			sched_prio(td, PUSER);
2483 
2484 		wakeup_swapper = sleepq_abort(td, intrval);
2485 	} else {
2486 		/*
2487 		 * Other states do nothing with the signal immediately,
2488 		 * other than kicking ourselves if we are running.
2489 		 * It will either never be noticed, or noticed very soon.
2490 		 */
2491 #ifdef SMP
2492 		if (TD_IS_RUNNING(td) && td != curthread)
2493 			forward_signal(td);
2494 #endif
2495 	}
2496 out:
2497 	PROC_SUNLOCK(p);
2498 	thread_unlock(td);
2499 	if (wakeup_swapper)
2500 		kick_proc0();
2501 }
2502 
2503 static int
sig_suspend_threads(struct thread * td,struct proc * p,int sending)2504 sig_suspend_threads(struct thread *td, struct proc *p, int sending)
2505 {
2506 	struct thread *td2;
2507 	int wakeup_swapper;
2508 
2509 	PROC_LOCK_ASSERT(p, MA_OWNED);
2510 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2511 	MPASS(sending || td == curthread);
2512 
2513 	wakeup_swapper = 0;
2514 	FOREACH_THREAD_IN_PROC(p, td2) {
2515 		thread_lock(td2);
2516 		td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
2517 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2518 		    (td2->td_flags & TDF_SINTR)) {
2519 			if (td2->td_flags & TDF_SBDRY) {
2520 				/*
2521 				 * Once a thread is asleep with
2522 				 * TDF_SBDRY and without TDF_SERESTART
2523 				 * or TDF_SEINTR set, it should never
2524 				 * become suspended due to this check.
2525 				 */
2526 				KASSERT(!TD_IS_SUSPENDED(td2),
2527 				    ("thread with deferred stops suspended"));
2528 				if (TD_SBDRY_INTR(td2))
2529 					wakeup_swapper |= sleepq_abort(td2,
2530 					    TD_SBDRY_ERRNO(td2));
2531 			} else if (!TD_IS_SUSPENDED(td2)) {
2532 				thread_suspend_one(td2);
2533 			}
2534 		} else if (!TD_IS_SUSPENDED(td2)) {
2535 			if (sending || td != td2)
2536 				td2->td_flags |= TDF_ASTPENDING;
2537 #ifdef SMP
2538 			if (TD_IS_RUNNING(td2) && td2 != td)
2539 				forward_signal(td2);
2540 #endif
2541 		}
2542 		thread_unlock(td2);
2543 	}
2544 	return (wakeup_swapper);
2545 }
2546 
2547 /*
2548  * Stop the process for an event deemed interesting to the debugger. If si is
2549  * non-NULL, this is a signal exchange; the new signal requested by the
2550  * debugger will be returned for handling. If si is NULL, this is some other
2551  * type of interesting event. The debugger may request a signal be delivered in
2552  * that case as well, however it will be deferred until it can be handled.
2553  */
2554 int
ptracestop(struct thread * td,int sig,ksiginfo_t * si)2555 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2556 {
2557 	struct proc *p = td->td_proc;
2558 	struct thread *td2;
2559 	ksiginfo_t ksi;
2560 	int prop;
2561 
2562 	PROC_LOCK_ASSERT(p, MA_OWNED);
2563 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2564 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2565 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2566 
2567 	td->td_xsig = sig;
2568 
2569 	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2570 		td->td_dbgflags |= TDB_XSIG;
2571 		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2572 		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2573 		PROC_SLOCK(p);
2574 		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2575 			if (P_KILLED(p)) {
2576 				/*
2577 				 * Ensure that, if we've been PT_KILLed, the
2578 				 * exit status reflects that. Another thread
2579 				 * may also be in ptracestop(), having just
2580 				 * received the SIGKILL, but this thread was
2581 				 * unsuspended first.
2582 				 */
2583 				td->td_dbgflags &= ~TDB_XSIG;
2584 				td->td_xsig = SIGKILL;
2585 				p->p_ptevents = 0;
2586 				break;
2587 			}
2588 			if (p->p_flag & P_SINGLE_EXIT &&
2589 			    !(td->td_dbgflags & TDB_EXIT)) {
2590 				/*
2591 				 * Ignore ptrace stops except for thread exit
2592 				 * events when the process exits.
2593 				 */
2594 				td->td_dbgflags &= ~TDB_XSIG;
2595 				PROC_SUNLOCK(p);
2596 				return (0);
2597 			}
2598 
2599 			/*
2600 			 * Make wait(2) work.  Ensure that right after the
2601 			 * attach, the thread which was decided to become the
2602 			 * leader of attach gets reported to the waiter.
2603 			 * Otherwise, just avoid overwriting another thread's
2604 			 * assignment to p_xthread.  If another thread has
2605 			 * already set p_xthread, the current thread will get
2606 			 * a chance to report itself upon the next iteration.
2607 			 */
2608 			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2609 			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2610 			    p->p_xthread == NULL)) {
2611 				p->p_xsig = sig;
2612 				p->p_xthread = td;
2613 
2614 				/*
2615 				 * If we are on sleepqueue already,
2616 				 * let sleepqueue code decide if it
2617 				 * needs to go sleep after attach.
2618 				 */
2619 				if (td->td_wchan == NULL)
2620 					td->td_dbgflags &= ~TDB_FSTP;
2621 
2622 				p->p_flag2 &= ~P2_PTRACE_FSTP;
2623 				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2624 				sig_suspend_threads(td, p, 0);
2625 			}
2626 			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2627 				td->td_dbgflags &= ~TDB_STOPATFORK;
2628 			}
2629 stopme:
2630 			thread_suspend_switch(td, p);
2631 			if (p->p_xthread == td)
2632 				p->p_xthread = NULL;
2633 			if (!(p->p_flag & P_TRACED))
2634 				break;
2635 			if (td->td_dbgflags & TDB_SUSPEND) {
2636 				if (p->p_flag & P_SINGLE_EXIT)
2637 					break;
2638 				goto stopme;
2639 			}
2640 		}
2641 		PROC_SUNLOCK(p);
2642 	}
2643 
2644 	if (si != NULL && sig == td->td_xsig) {
2645 		/* Parent wants us to take the original signal unchanged. */
2646 		si->ksi_flags |= KSI_HEAD;
2647 		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2648 			si->ksi_signo = 0;
2649 	} else if (td->td_xsig != 0) {
2650 		/*
2651 		 * If parent wants us to take a new signal, then it will leave
2652 		 * it in td->td_xsig; otherwise we just look for signals again.
2653 		 */
2654 		ksiginfo_init(&ksi);
2655 		ksi.ksi_signo = td->td_xsig;
2656 		ksi.ksi_flags |= KSI_PTRACE;
2657 		prop = sigprop(td->td_xsig);
2658 		td2 = sigtd(p, td->td_xsig, prop);
2659 		tdsendsignal(p, td2, td->td_xsig, &ksi);
2660 		if (td != td2)
2661 			return (0);
2662 	}
2663 
2664 	return (td->td_xsig);
2665 }
2666 
2667 static void
reschedule_signals(struct proc * p,sigset_t block,int flags)2668 reschedule_signals(struct proc *p, sigset_t block, int flags)
2669 {
2670 	struct sigacts *ps;
2671 	struct thread *td;
2672 	int sig;
2673 
2674 	PROC_LOCK_ASSERT(p, MA_OWNED);
2675 	ps = p->p_sigacts;
2676 	mtx_assert(&ps->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0 ?
2677 	    MA_OWNED : MA_NOTOWNED);
2678 	if (SIGISEMPTY(p->p_siglist))
2679 		return;
2680 	SIGSETAND(block, p->p_siglist);
2681 	while ((sig = sig_ffs(&block)) != 0) {
2682 		SIGDELSET(block, sig);
2683 		td = sigtd(p, sig, 0);
2684 		signotify(td);
2685 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2686 			mtx_lock(&ps->ps_mtx);
2687 		if (p->p_flag & P_TRACED ||
2688 		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2689 		    !SIGISMEMBER(td->td_sigmask, sig)))
2690 			tdsigwakeup(td, sig, SIG_CATCH,
2691 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2692 			     ERESTART));
2693 		if (!(flags & SIGPROCMASK_PS_LOCKED))
2694 			mtx_unlock(&ps->ps_mtx);
2695 	}
2696 }
2697 
2698 void
tdsigcleanup(struct thread * td)2699 tdsigcleanup(struct thread *td)
2700 {
2701 	struct proc *p;
2702 	sigset_t unblocked;
2703 
2704 	p = td->td_proc;
2705 	PROC_LOCK_ASSERT(p, MA_OWNED);
2706 
2707 	sigqueue_flush(&td->td_sigqueue);
2708 	if (p->p_numthreads == 1)
2709 		return;
2710 
2711 	/*
2712 	 * Since we cannot handle signals, notify signal post code
2713 	 * about this by filling the sigmask.
2714 	 *
2715 	 * Also, if needed, wake up thread(s) that do not block the
2716 	 * same signals as the exiting thread, since the thread might
2717 	 * have been selected for delivery and woken up.
2718 	 */
2719 	SIGFILLSET(unblocked);
2720 	SIGSETNAND(unblocked, td->td_sigmask);
2721 	SIGFILLSET(td->td_sigmask);
2722 	reschedule_signals(p, unblocked, 0);
2723 
2724 }
2725 
2726 static int
sigdeferstop_curr_flags(int cflags)2727 sigdeferstop_curr_flags(int cflags)
2728 {
2729 
2730 	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
2731 	    (cflags & TDF_SBDRY) != 0);
2732 	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
2733 }
2734 
2735 /*
2736  * Defer the delivery of SIGSTOP for the current thread, according to
2737  * the requested mode.  Returns previous flags, which must be restored
2738  * by sigallowstop().
2739  *
2740  * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
2741  * cleared by the current thread, which allow the lock-less read-only
2742  * accesses below.
2743  */
2744 int
sigdeferstop_impl(int mode)2745 sigdeferstop_impl(int mode)
2746 {
2747 	struct thread *td;
2748 	int cflags, nflags;
2749 
2750 	td = curthread;
2751 	cflags = sigdeferstop_curr_flags(td->td_flags);
2752 	switch (mode) {
2753 	case SIGDEFERSTOP_NOP:
2754 		nflags = cflags;
2755 		break;
2756 	case SIGDEFERSTOP_OFF:
2757 		nflags = 0;
2758 		break;
2759 	case SIGDEFERSTOP_SILENT:
2760 		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
2761 		break;
2762 	case SIGDEFERSTOP_EINTR:
2763 		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
2764 		break;
2765 	case SIGDEFERSTOP_ERESTART:
2766 		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
2767 		break;
2768 	default:
2769 		panic("sigdeferstop: invalid mode %x", mode);
2770 		break;
2771 	}
2772 	if (cflags == nflags)
2773 		return (SIGDEFERSTOP_VAL_NCHG);
2774 	thread_lock(td);
2775 	td->td_flags = (td->td_flags & ~cflags) | nflags;
2776 	thread_unlock(td);
2777 	return (cflags);
2778 }
2779 
2780 /*
2781  * Restores the STOP handling mode, typically permitting the delivery
2782  * of SIGSTOP for the current thread.  This does not immediately
2783  * suspend if a stop was posted.  Instead, the thread will suspend
2784  * either via ast() or a subsequent interruptible sleep.
2785  */
2786 void
sigallowstop_impl(int prev)2787 sigallowstop_impl(int prev)
2788 {
2789 	struct thread *td;
2790 	int cflags;
2791 
2792 	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
2793 	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
2794 	    ("sigallowstop: incorrect previous mode %x", prev));
2795 	td = curthread;
2796 	cflags = sigdeferstop_curr_flags(td->td_flags);
2797 	if (cflags != prev) {
2798 		thread_lock(td);
2799 		td->td_flags = (td->td_flags & ~cflags) | prev;
2800 		thread_unlock(td);
2801 	}
2802 }
2803 
2804 /*
2805  * If the current process has received a signal (should be caught or cause
2806  * termination, should interrupt current syscall), return the signal number.
2807  * Stop signals with default action are processed immediately, then cleared;
2808  * they aren't returned.  This is checked after each entry to the system for
2809  * a syscall or trap (though this can usually be done without calling issignal
2810  * by checking the pending signal masks in cursig.) The normal call
2811  * sequence is
2812  *
2813  *	while (sig = cursig(curthread))
2814  *		postsig(sig);
2815  */
2816 static int
issignal(struct thread * td)2817 issignal(struct thread *td)
2818 {
2819 	struct proc *p;
2820 	struct sigacts *ps;
2821 	struct sigqueue *queue;
2822 	sigset_t sigpending;
2823 	ksiginfo_t ksi;
2824 	int prop, sig, traced;
2825 
2826 	p = td->td_proc;
2827 	ps = p->p_sigacts;
2828 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2829 	PROC_LOCK_ASSERT(p, MA_OWNED);
2830 	for (;;) {
2831 		traced = (p->p_flag & P_TRACED) || (p->p_stops & S_SIG);
2832 
2833 		sigpending = td->td_sigqueue.sq_signals;
2834 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
2835 		SIGSETNAND(sigpending, td->td_sigmask);
2836 
2837 		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
2838 		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2839 			SIG_STOPSIGMASK(sigpending);
2840 		if (SIGISEMPTY(sigpending))	/* no signal to send */
2841 			return (0);
2842 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
2843 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
2844 		    SIGISMEMBER(sigpending, SIGSTOP)) {
2845 			/*
2846 			 * If debugger just attached, always consume
2847 			 * SIGSTOP from ptrace(PT_ATTACH) first, to
2848 			 * execute the debugger attach ritual in
2849 			 * order.
2850 			 */
2851 			sig = SIGSTOP;
2852 			td->td_dbgflags |= TDB_FSTP;
2853 		} else {
2854 			sig = sig_ffs(&sigpending);
2855 		}
2856 
2857 		if (p->p_stops & S_SIG) {
2858 			mtx_unlock(&ps->ps_mtx);
2859 			stopevent(p, S_SIG, sig);
2860 			mtx_lock(&ps->ps_mtx);
2861 		}
2862 
2863 		/*
2864 		 * We should see pending but ignored signals
2865 		 * only if P_TRACED was on when they were posted.
2866 		 */
2867 		if (SIGISMEMBER(ps->ps_sigignore, sig) && (traced == 0)) {
2868 			sigqueue_delete(&td->td_sigqueue, sig);
2869 			sigqueue_delete(&p->p_sigqueue, sig);
2870 			continue;
2871 		}
2872 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
2873 			/*
2874 			 * If traced, always stop.
2875 			 * Remove old signal from queue before the stop.
2876 			 * XXX shrug off debugger, it causes siginfo to
2877 			 * be thrown away.
2878 			 */
2879 			queue = &td->td_sigqueue;
2880 			ksiginfo_init(&ksi);
2881 			if (sigqueue_get(queue, sig, &ksi) == 0) {
2882 				queue = &p->p_sigqueue;
2883 				sigqueue_get(queue, sig, &ksi);
2884 			}
2885 			td->td_si = ksi.ksi_info;
2886 
2887 			mtx_unlock(&ps->ps_mtx);
2888 			sig = ptracestop(td, sig, &ksi);
2889 			mtx_lock(&ps->ps_mtx);
2890 
2891 			td->td_si.si_signo = 0;
2892 
2893 			/*
2894 			 * Keep looking if the debugger discarded or
2895 			 * replaced the signal.
2896 			 */
2897 			if (sig == 0)
2898 				continue;
2899 
2900 			/*
2901 			 * If the signal became masked, re-queue it.
2902 			 */
2903 			if (SIGISMEMBER(td->td_sigmask, sig)) {
2904 				ksi.ksi_flags |= KSI_HEAD;
2905 				sigqueue_add(&p->p_sigqueue, sig, &ksi);
2906 				continue;
2907 			}
2908 
2909 			/*
2910 			 * If the traced bit got turned off, requeue
2911 			 * the signal and go back up to the top to
2912 			 * rescan signals.  This ensures that p_sig*
2913 			 * and p_sigact are consistent.
2914 			 */
2915 			if ((p->p_flag & P_TRACED) == 0) {
2916 				ksi.ksi_flags |= KSI_HEAD;
2917 				sigqueue_add(queue, sig, &ksi);
2918 				continue;
2919 			}
2920 		}
2921 
2922 		prop = sigprop(sig);
2923 
2924 		/*
2925 		 * Decide whether the signal should be returned.
2926 		 * Return the signal's number, or fall through
2927 		 * to clear it from the pending mask.
2928 		 */
2929 		switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2930 
2931 		case (intptr_t)SIG_DFL:
2932 			/*
2933 			 * Don't take default actions on system processes.
2934 			 */
2935 			if (p->p_pid <= 1) {
2936 #ifdef DIAGNOSTIC
2937 				/*
2938 				 * Are you sure you want to ignore SIGSEGV
2939 				 * in init? XXX
2940 				 */
2941 				printf("Process (pid %lu) got signal %d\n",
2942 					(u_long)p->p_pid, sig);
2943 #endif
2944 				break;		/* == ignore */
2945 			}
2946 			/*
2947 			 * If there is a pending stop signal to process with
2948 			 * default action, stop here, then clear the signal.
2949 			 * Traced or exiting processes should ignore stops.
2950 			 * Additionally, a member of an orphaned process group
2951 			 * should ignore tty stops.
2952 			 */
2953 			if (prop & SIGPROP_STOP) {
2954 				mtx_unlock(&ps->ps_mtx);
2955 				if ((p->p_flag & (P_TRACED | P_WEXIT |
2956 				    P_SINGLE_EXIT)) != 0 || ((p->p_pgrp->
2957 				    pg_flags & PGRP_ORPHANED) != 0 &&
2958 				    (prop & SIGPROP_TTYSTOP) != 0)) {
2959 					mtx_lock(&ps->ps_mtx);
2960 					break;	/* == ignore */
2961 				}
2962 				if (TD_SBDRY_INTR(td)) {
2963 					KASSERT((td->td_flags & TDF_SBDRY) != 0,
2964 					    ("lost TDF_SBDRY"));
2965 					mtx_lock(&ps->ps_mtx);
2966 					return (-1);
2967 				}
2968 				WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2969 				    &p->p_mtx.lock_object, "Catching SIGSTOP");
2970 				sigqueue_delete(&td->td_sigqueue, sig);
2971 				sigqueue_delete(&p->p_sigqueue, sig);
2972 				p->p_flag |= P_STOPPED_SIG;
2973 				p->p_xsig = sig;
2974 				PROC_SLOCK(p);
2975 				sig_suspend_threads(td, p, 0);
2976 				thread_suspend_switch(td, p);
2977 				PROC_SUNLOCK(p);
2978 				mtx_lock(&ps->ps_mtx);
2979 				goto next;
2980 			} else if (prop & SIGPROP_IGNORE) {
2981 				/*
2982 				 * Except for SIGCONT, shouldn't get here.
2983 				 * Default action is to ignore; drop it.
2984 				 */
2985 				break;		/* == ignore */
2986 			} else
2987 				return (sig);
2988 			/*NOTREACHED*/
2989 
2990 		case (intptr_t)SIG_IGN:
2991 			/*
2992 			 * Masking above should prevent us ever trying
2993 			 * to take action on an ignored signal other
2994 			 * than SIGCONT, unless process is traced.
2995 			 */
2996 			if ((prop & SIGPROP_CONT) == 0 &&
2997 			    (p->p_flag & P_TRACED) == 0)
2998 				printf("issignal\n");
2999 			break;		/* == ignore */
3000 
3001 		default:
3002 			/*
3003 			 * This signal has an action, let
3004 			 * postsig() process it.
3005 			 */
3006 			return (sig);
3007 		}
3008 		sigqueue_delete(&td->td_sigqueue, sig);	/* take the signal! */
3009 		sigqueue_delete(&p->p_sigqueue, sig);
3010 next:;
3011 	}
3012 	/* NOTREACHED */
3013 }
3014 
3015 void
thread_stopped(struct proc * p)3016 thread_stopped(struct proc *p)
3017 {
3018 	int n;
3019 
3020 	PROC_LOCK_ASSERT(p, MA_OWNED);
3021 	PROC_SLOCK_ASSERT(p, MA_OWNED);
3022 	n = p->p_suspcount;
3023 	if (p == curproc)
3024 		n++;
3025 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3026 		PROC_SUNLOCK(p);
3027 		p->p_flag &= ~P_WAITED;
3028 		PROC_LOCK(p->p_pptr);
3029 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
3030 			CLD_TRAPPED : CLD_STOPPED);
3031 		PROC_UNLOCK(p->p_pptr);
3032 		PROC_SLOCK(p);
3033 	}
3034 }
3035 
3036 /*
3037  * Take the action for the specified signal
3038  * from the current set of pending signals.
3039  */
3040 int
postsig(int sig)3041 postsig(int sig)
3042 {
3043 	struct thread *td;
3044 	struct proc *p;
3045 	struct sigacts *ps;
3046 	sig_t action;
3047 	ksiginfo_t ksi;
3048 	sigset_t returnmask;
3049 
3050 	KASSERT(sig != 0, ("postsig"));
3051 
3052 	td = curthread;
3053 	p = td->td_proc;
3054 	PROC_LOCK_ASSERT(p, MA_OWNED);
3055 	ps = p->p_sigacts;
3056 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3057 	ksiginfo_init(&ksi);
3058 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3059 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3060 		return (0);
3061 	ksi.ksi_signo = sig;
3062 	if (ksi.ksi_code == SI_TIMER)
3063 		itimer_accept(p, ksi.ksi_timerid, &ksi);
3064 	action = ps->ps_sigact[_SIG_IDX(sig)];
3065 #ifdef KTRACE
3066 	if (KTRPOINT(td, KTR_PSIG))
3067 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3068 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3069 #endif
3070 	if ((p->p_stops & S_SIG) != 0) {
3071 		mtx_unlock(&ps->ps_mtx);
3072 		stopevent(p, S_SIG, sig);
3073 		mtx_lock(&ps->ps_mtx);
3074 	}
3075 
3076 	if (action == SIG_DFL) {
3077 		/*
3078 		 * Default action, where the default is to kill
3079 		 * the process.  (Other cases were ignored above.)
3080 		 */
3081 		mtx_unlock(&ps->ps_mtx);
3082 		proc_td_siginfo_capture(td, &ksi.ksi_info);
3083 		sigexit(td, sig);
3084 		/* NOTREACHED */
3085 	} else {
3086 		/*
3087 		 * If we get here, the signal must be caught.
3088 		 */
3089 		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3090 		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3091 		    ("postsig action: blocked sig %d", sig));
3092 
3093 		/*
3094 		 * Set the new mask value and also defer further
3095 		 * occurrences of this signal.
3096 		 *
3097 		 * Special case: user has done a sigsuspend.  Here the
3098 		 * current mask is not of interest, but rather the
3099 		 * mask from before the sigsuspend is what we want
3100 		 * restored after the signal processing is completed.
3101 		 */
3102 		if (td->td_pflags & TDP_OLDMASK) {
3103 			returnmask = td->td_oldsigmask;
3104 			td->td_pflags &= ~TDP_OLDMASK;
3105 		} else
3106 			returnmask = td->td_sigmask;
3107 
3108 		if (p->p_sig == sig) {
3109 			p->p_code = 0;
3110 			p->p_sig = 0;
3111 		}
3112 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3113 		postsig_done(sig, td, ps);
3114 	}
3115 	return (1);
3116 }
3117 
3118 void
proc_wkilled(struct proc * p)3119 proc_wkilled(struct proc *p)
3120 {
3121 
3122 	PROC_LOCK_ASSERT(p, MA_OWNED);
3123 	if ((p->p_flag & P_WKILLED) == 0) {
3124 		p->p_flag |= P_WKILLED;
3125 		/*
3126 		 * Notify swapper that there is a process to swap in.
3127 		 * The notification is racy, at worst it would take 10
3128 		 * seconds for the swapper process to notice.
3129 		 */
3130 		if ((p->p_flag & (P_INMEM | P_SWAPPINGIN)) == 0)
3131 			wakeup(&proc0);
3132 	}
3133 }
3134 
3135 /*
3136  * Kill the current process for stated reason.
3137  */
3138 void
killproc(struct proc * p,char * why)3139 killproc(struct proc *p, char *why)
3140 {
3141 
3142 	PROC_LOCK_ASSERT(p, MA_OWNED);
3143 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3144 	    p->p_comm);
3145 	log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3146 	    p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3147 	    p->p_ucred ? p->p_ucred->cr_uid : -1, why);
3148 	proc_wkilled(p);
3149 	kern_psignal(p, SIGKILL);
3150 }
3151 
3152 /*
3153  * Force the current process to exit with the specified signal, dumping core
3154  * if appropriate.  We bypass the normal tests for masked and caught signals,
3155  * allowing unrecoverable failures to terminate the process without changing
3156  * signal state.  Mark the accounting record with the signal termination.
3157  * If dumping core, save the signal number for the debugger.  Calls exit and
3158  * does not return.
3159  */
3160 void
sigexit(struct thread * td,int sig)3161 sigexit(struct thread *td, int sig)
3162 {
3163 	struct proc *p = td->td_proc;
3164 
3165 	PROC_LOCK_ASSERT(p, MA_OWNED);
3166 	p->p_acflag |= AXSIG;
3167 	/*
3168 	 * We must be single-threading to generate a core dump.  This
3169 	 * ensures that the registers in the core file are up-to-date.
3170 	 * Also, the ELF dump handler assumes that the thread list doesn't
3171 	 * change out from under it.
3172 	 *
3173 	 * XXX If another thread attempts to single-thread before us
3174 	 *     (e.g. via fork()), we won't get a dump at all.
3175 	 */
3176 	if ((sigprop(sig) & SIGPROP_CORE) &&
3177 	    thread_single(p, SINGLE_NO_EXIT) == 0) {
3178 		p->p_sig = sig;
3179 		/*
3180 		 * Log signals which would cause core dumps
3181 		 * (Log as LOG_INFO to appease those who don't want
3182 		 * these messages.)
3183 		 * XXX : Todo, as well as euid, write out ruid too
3184 		 * Note that coredump() drops proc lock.
3185 		 */
3186 		if (coredump(td) == 0)
3187 			sig |= WCOREFLAG;
3188 		if (kern_logsigexit)
3189 			log(LOG_INFO,
3190 			    "pid %d (%s), jid %d, uid %d: exited on "
3191 			    "signal %d%s\n", p->p_pid, p->p_comm,
3192 			    p->p_ucred->cr_prison->pr_id,
3193 			    td->td_ucred ? td->td_ucred->cr_uid : -1,
3194 			    sig &~ WCOREFLAG,
3195 			    sig & WCOREFLAG ? " (core dumped)" : "");
3196 	} else
3197 		PROC_UNLOCK(p);
3198 	exit1(td, 0, sig);
3199 	/* NOTREACHED */
3200 }
3201 
3202 /*
3203  * Send queued SIGCHLD to parent when child process's state
3204  * is changed.
3205  */
3206 static void
sigparent(struct proc * p,int reason,int status)3207 sigparent(struct proc *p, int reason, int status)
3208 {
3209 	PROC_LOCK_ASSERT(p, MA_OWNED);
3210 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3211 
3212 	if (p->p_ksi != NULL) {
3213 		p->p_ksi->ksi_signo  = SIGCHLD;
3214 		p->p_ksi->ksi_code   = reason;
3215 		p->p_ksi->ksi_status = status;
3216 		p->p_ksi->ksi_pid    = p->p_pid;
3217 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3218 		if (KSI_ONQ(p->p_ksi))
3219 			return;
3220 	}
3221 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3222 }
3223 
3224 static void
childproc_jobstate(struct proc * p,int reason,int sig)3225 childproc_jobstate(struct proc *p, int reason, int sig)
3226 {
3227 	struct sigacts *ps;
3228 
3229 	PROC_LOCK_ASSERT(p, MA_OWNED);
3230 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3231 
3232 	/*
3233 	 * Wake up parent sleeping in kern_wait(), also send
3234 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3235 	 * that parent will awake, because parent may masked
3236 	 * the signal.
3237 	 */
3238 	p->p_pptr->p_flag |= P_STATCHILD;
3239 	wakeup(p->p_pptr);
3240 
3241 	ps = p->p_pptr->p_sigacts;
3242 	mtx_lock(&ps->ps_mtx);
3243 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3244 		mtx_unlock(&ps->ps_mtx);
3245 		sigparent(p, reason, sig);
3246 	} else
3247 		mtx_unlock(&ps->ps_mtx);
3248 }
3249 
3250 void
childproc_stopped(struct proc * p,int reason)3251 childproc_stopped(struct proc *p, int reason)
3252 {
3253 
3254 	childproc_jobstate(p, reason, p->p_xsig);
3255 }
3256 
3257 void
childproc_continued(struct proc * p)3258 childproc_continued(struct proc *p)
3259 {
3260 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3261 }
3262 
3263 void
childproc_exited(struct proc * p)3264 childproc_exited(struct proc *p)
3265 {
3266 	int reason, status;
3267 
3268 	if (WCOREDUMP(p->p_xsig)) {
3269 		reason = CLD_DUMPED;
3270 		status = WTERMSIG(p->p_xsig);
3271 	} else if (WIFSIGNALED(p->p_xsig)) {
3272 		reason = CLD_KILLED;
3273 		status = WTERMSIG(p->p_xsig);
3274 	} else {
3275 		reason = CLD_EXITED;
3276 		status = p->p_xexit;
3277 	}
3278 	/*
3279 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3280 	 * done in exit1().
3281 	 */
3282 	sigparent(p, reason, status);
3283 }
3284 
3285 #define	MAX_NUM_CORE_FILES 100000
3286 #ifndef NUM_CORE_FILES
3287 #define	NUM_CORE_FILES 5
3288 #endif
3289 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3290 static int num_cores = NUM_CORE_FILES;
3291 
3292 static int
sysctl_debug_num_cores_check(SYSCTL_HANDLER_ARGS)3293 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3294 {
3295 	int error;
3296 	int new_val;
3297 
3298 	new_val = num_cores;
3299 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3300 	if (error != 0 || req->newptr == NULL)
3301 		return (error);
3302 	if (new_val > MAX_NUM_CORE_FILES)
3303 		new_val = MAX_NUM_CORE_FILES;
3304 	if (new_val < 0)
3305 		new_val = 0;
3306 	num_cores = new_val;
3307 	return (0);
3308 }
3309 SYSCTL_PROC(_debug, OID_AUTO, ncores, CTLTYPE_INT|CTLFLAG_RW,
3310 	    0, sizeof(int), sysctl_debug_num_cores_check, "I",
3311 	    "Maximum number of generated process corefiles while using index format");
3312 
3313 #define	GZIP_SUFFIX	".gz"
3314 #define	ZSTD_SUFFIX	".zst"
3315 
3316 int compress_user_cores = 0;
3317 
3318 static int
sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)3319 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3320 {
3321 	int error, val;
3322 
3323 	val = compress_user_cores;
3324 	error = sysctl_handle_int(oidp, &val, 0, req);
3325 	if (error != 0 || req->newptr == NULL)
3326 		return (error);
3327 	if (val != 0 && !compressor_avail(val))
3328 		return (EINVAL);
3329 	compress_user_cores = val;
3330 	return (error);
3331 }
3332 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores, CTLTYPE_INT | CTLFLAG_RWTUN,
3333     0, sizeof(int), sysctl_compress_user_cores, "I",
3334     "Enable compression of user corefiles ("
3335     __XSTRING(COMPRESS_GZIP) " = gzip, "
3336     __XSTRING(COMPRESS_ZSTD) " = zstd)");
3337 
3338 int compress_user_cores_level = 6;
3339 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3340     &compress_user_cores_level, 0,
3341     "Corefile compression level");
3342 
3343 /*
3344  * Protect the access to corefilename[] by allproc_lock.
3345  */
3346 #define	corefilename_lock	allproc_lock
3347 
3348 static char corefilename[MAXPATHLEN] = {"%N.core"};
3349 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3350 
3351 static int
sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)3352 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3353 {
3354 	int error;
3355 
3356 	sx_xlock(&corefilename_lock);
3357 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3358 	    req);
3359 	sx_xunlock(&corefilename_lock);
3360 
3361 	return (error);
3362 }
3363 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3364     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3365     "Process corefile name format string");
3366 
3367 static void
vnode_close_locked(struct thread * td,struct vnode * vp)3368 vnode_close_locked(struct thread *td, struct vnode *vp)
3369 {
3370 
3371 	VOP_UNLOCK(vp, 0);
3372 	vn_close(vp, FWRITE, td->td_ucred, td);
3373 }
3374 
3375 /*
3376  * If the core format has a %I in it, then we need to check
3377  * for existing corefiles before defining a name.
3378  * To do this we iterate over 0..ncores to find a
3379  * non-existing core file name to use. If all core files are
3380  * already used we choose the oldest one.
3381  */
3382 static int
corefile_open_last(struct thread * td,char * name,int indexpos,int indexlen,int ncores,struct vnode ** vpp)3383 corefile_open_last(struct thread *td, char *name, int indexpos,
3384     int indexlen, int ncores, struct vnode **vpp)
3385 {
3386 	struct vnode *oldvp, *nextvp, *vp;
3387 	struct vattr vattr;
3388 	struct nameidata nd;
3389 	int error, i, flags, oflags, cmode;
3390 	char ch;
3391 	struct timespec lasttime;
3392 
3393 	nextvp = oldvp = NULL;
3394 	cmode = S_IRUSR | S_IWUSR;
3395 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3396 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3397 
3398 	for (i = 0; i < ncores; i++) {
3399 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3400 
3401 		ch = name[indexpos + indexlen];
3402 		(void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3403 		    i);
3404 		name[indexpos + indexlen] = ch;
3405 
3406 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3407 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3408 		    NULL);
3409 		if (error != 0)
3410 			break;
3411 
3412 		vp = nd.ni_vp;
3413 		NDFREE(&nd, NDF_ONLY_PNBUF);
3414 		if ((flags & O_CREAT) == O_CREAT) {
3415 			nextvp = vp;
3416 			break;
3417 		}
3418 
3419 		error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3420 		if (error != 0) {
3421 			vnode_close_locked(td, vp);
3422 			break;
3423 		}
3424 
3425 		if (oldvp == NULL ||
3426 		    lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3427 		    (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3428 		    lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3429 			if (oldvp != NULL)
3430 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3431 			oldvp = vp;
3432 			VOP_UNLOCK(oldvp, 0);
3433 			lasttime = vattr.va_mtime;
3434 		} else {
3435 			vnode_close_locked(td, vp);
3436 		}
3437 	}
3438 
3439 	if (oldvp != NULL) {
3440 		if (nextvp == NULL) {
3441 			if ((td->td_proc->p_flag & P_SUGID) != 0) {
3442 				error = EFAULT;
3443 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3444 			} else {
3445 				nextvp = oldvp;
3446 				error = vn_lock(nextvp, LK_EXCLUSIVE);
3447 				if (error != 0) {
3448 					vn_close(nextvp, FWRITE, td->td_ucred,
3449 					    td);
3450 					nextvp = NULL;
3451 				}
3452 			}
3453 		} else {
3454 			vn_close(oldvp, FWRITE, td->td_ucred, td);
3455 		}
3456 	}
3457 	if (error != 0) {
3458 		if (nextvp != NULL)
3459 			vnode_close_locked(td, oldvp);
3460 	} else {
3461 		*vpp = nextvp;
3462 	}
3463 
3464 	return (error);
3465 }
3466 
3467 /*
3468  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3469  * Expand the name described in corefilename, using name, uid, and pid
3470  * and open/create core file.
3471  * corefilename is a printf-like string, with three format specifiers:
3472  *	%N	name of process ("name")
3473  *	%P	process id (pid)
3474  *	%U	user id (uid)
3475  * For example, "%N.core" is the default; they can be disabled completely
3476  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3477  * This is controlled by the sysctl variable kern.corefile (see above).
3478  */
3479 static int
corefile_open(const char * comm,uid_t uid,pid_t pid,struct thread * td,int compress,struct vnode ** vpp,char ** namep)3480 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3481     int compress, struct vnode **vpp, char **namep)
3482 {
3483 	struct sbuf sb;
3484 	struct nameidata nd;
3485 	const char *format;
3486 	char *hostname, *name;
3487 	int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
3488 
3489 	hostname = NULL;
3490 	format = corefilename;
3491 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3492 	indexlen = 0;
3493 	indexpos = -1;
3494 	ncores = num_cores;
3495 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3496 	sx_slock(&corefilename_lock);
3497 	for (i = 0; format[i] != '\0'; i++) {
3498 		switch (format[i]) {
3499 		case '%':	/* Format character */
3500 			i++;
3501 			switch (format[i]) {
3502 			case '%':
3503 				sbuf_putc(&sb, '%');
3504 				break;
3505 			case 'H':	/* hostname */
3506 				if (hostname == NULL) {
3507 					hostname = malloc(MAXHOSTNAMELEN,
3508 					    M_TEMP, M_WAITOK);
3509 				}
3510 				getcredhostname(td->td_ucred, hostname,
3511 				    MAXHOSTNAMELEN);
3512 				sbuf_printf(&sb, "%s", hostname);
3513 				break;
3514 			case 'I':	/* autoincrementing index */
3515 				if (indexpos != -1) {
3516 					sbuf_printf(&sb, "%%I");
3517 					break;
3518 				}
3519 
3520 				indexpos = sbuf_len(&sb);
3521 				sbuf_printf(&sb, "%u", ncores - 1);
3522 				indexlen = sbuf_len(&sb) - indexpos;
3523 				break;
3524 			case 'N':	/* process name */
3525 				sbuf_printf(&sb, "%s", comm);
3526 				break;
3527 			case 'P':	/* process id */
3528 				sbuf_printf(&sb, "%u", pid);
3529 				break;
3530 			case 'U':	/* user id */
3531 				sbuf_printf(&sb, "%u", uid);
3532 				break;
3533 			default:
3534 				log(LOG_ERR,
3535 				    "Unknown format character %c in "
3536 				    "corename `%s'\n", format[i], format);
3537 				break;
3538 			}
3539 			break;
3540 		default:
3541 			sbuf_putc(&sb, format[i]);
3542 			break;
3543 		}
3544 	}
3545 	sx_sunlock(&corefilename_lock);
3546 	free(hostname, M_TEMP);
3547 	if (compress == COMPRESS_GZIP)
3548 		sbuf_printf(&sb, GZIP_SUFFIX);
3549 	else if (compress == COMPRESS_ZSTD)
3550 		sbuf_printf(&sb, ZSTD_SUFFIX);
3551 	if (sbuf_error(&sb) != 0) {
3552 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3553 		    "long\n", (long)pid, comm, (u_long)uid);
3554 		sbuf_delete(&sb);
3555 		free(name, M_TEMP);
3556 		return (ENOMEM);
3557 	}
3558 	sbuf_finish(&sb);
3559 	sbuf_delete(&sb);
3560 
3561 	if (indexpos != -1) {
3562 		error = corefile_open_last(td, name, indexpos, indexlen, ncores,
3563 		    vpp);
3564 		if (error != 0) {
3565 			log(LOG_ERR,
3566 			    "pid %d (%s), uid (%u):  Path `%s' failed "
3567 			    "on initial open test, error = %d\n",
3568 			    pid, comm, uid, name, error);
3569 		}
3570 	} else {
3571 		cmode = S_IRUSR | S_IWUSR;
3572 		oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3573 		    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3574 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3575 		if ((td->td_proc->p_flag & P_SUGID) != 0)
3576 			flags |= O_EXCL;
3577 
3578 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3579 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3580 		    NULL);
3581 		if (error == 0) {
3582 			*vpp = nd.ni_vp;
3583 			NDFREE(&nd, NDF_ONLY_PNBUF);
3584 		}
3585 	}
3586 
3587 	if (error != 0) {
3588 #ifdef AUDIT
3589 		audit_proc_coredump(td, name, error);
3590 #endif
3591 		free(name, M_TEMP);
3592 		return (error);
3593 	}
3594 	*namep = name;
3595 	return (0);
3596 }
3597 
3598 /*
3599  * Dump a process' core.  The main routine does some
3600  * policy checking, and creates the name of the coredump;
3601  * then it passes on a vnode and a size limit to the process-specific
3602  * coredump routine if there is one; if there _is not_ one, it returns
3603  * ENOSYS; otherwise it returns the error from the process-specific routine.
3604  */
3605 
3606 static int
coredump(struct thread * td)3607 coredump(struct thread *td)
3608 {
3609 	struct proc *p = td->td_proc;
3610 	struct ucred *cred = td->td_ucred;
3611 	struct vnode *vp;
3612 	struct flock lf;
3613 	struct vattr vattr;
3614 	int error, error1, locked;
3615 	char *name;			/* name of corefile */
3616 	void *rl_cookie;
3617 	off_t limit;
3618 	char *fullpath, *freepath = NULL;
3619 	struct sbuf *sb;
3620 
3621 	PROC_LOCK_ASSERT(p, MA_OWNED);
3622 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3623 	_STOPEVENT(p, S_CORE, 0);
3624 
3625 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
3626 	    (p->p_flag2 & P2_NOTRACE) != 0) {
3627 		PROC_UNLOCK(p);
3628 		return (EFAULT);
3629 	}
3630 
3631 	/*
3632 	 * Note that the bulk of limit checking is done after
3633 	 * the corefile is created.  The exception is if the limit
3634 	 * for corefiles is 0, in which case we don't bother
3635 	 * creating the corefile at all.  This layout means that
3636 	 * a corefile is truncated instead of not being created,
3637 	 * if it is larger than the limit.
3638 	 */
3639 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
3640 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
3641 		PROC_UNLOCK(p);
3642 		return (EFBIG);
3643 	}
3644 	PROC_UNLOCK(p);
3645 
3646 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
3647 	    compress_user_cores, &vp, &name);
3648 	if (error != 0)
3649 		return (error);
3650 
3651 	/*
3652 	 * Don't dump to non-regular files or files with links.
3653 	 * Do not dump into system files. Effective user must own the corefile.
3654 	 */
3655 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
3656 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
3657 	    vattr.va_uid != cred->cr_uid) {
3658 		VOP_UNLOCK(vp, 0);
3659 		error = EFAULT;
3660 		goto out;
3661 	}
3662 
3663 	VOP_UNLOCK(vp, 0);
3664 
3665 	/* Postpone other writers, including core dumps of other processes. */
3666 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
3667 
3668 	lf.l_whence = SEEK_SET;
3669 	lf.l_start = 0;
3670 	lf.l_len = 0;
3671 	lf.l_type = F_WRLCK;
3672 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
3673 
3674 	VATTR_NULL(&vattr);
3675 	vattr.va_size = 0;
3676 	if (set_core_nodump_flag)
3677 		vattr.va_flags = UF_NODUMP;
3678 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3679 	VOP_SETATTR(vp, &vattr, cred);
3680 	VOP_UNLOCK(vp, 0);
3681 	PROC_LOCK(p);
3682 	p->p_acflag |= ACORE;
3683 	PROC_UNLOCK(p);
3684 
3685 	if (p->p_sysent->sv_coredump != NULL) {
3686 		error = p->p_sysent->sv_coredump(td, vp, limit, 0);
3687 	} else {
3688 		error = ENOSYS;
3689 	}
3690 
3691 	if (locked) {
3692 		lf.l_type = F_UNLCK;
3693 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
3694 	}
3695 	vn_rangelock_unlock(vp, rl_cookie);
3696 
3697 	/*
3698 	 * Notify the userland helper that a process triggered a core dump.
3699 	 * This allows the helper to run an automated debugging session.
3700 	 */
3701 	if (error != 0 || coredump_devctl == 0)
3702 		goto out;
3703 	sb = sbuf_new_auto();
3704 	if (vn_fullpath_global(td, p->p_textvp, &fullpath, &freepath) != 0)
3705 		goto out2;
3706 	sbuf_printf(sb, "comm=\"");
3707 	devctl_safe_quote_sb(sb, fullpath);
3708 	free(freepath, M_TEMP);
3709 	sbuf_printf(sb, "\" core=\"");
3710 
3711 	/*
3712 	 * We can't lookup core file vp directly. When we're replacing a core, and
3713 	 * other random times, we flush the name cache, so it will fail. Instead,
3714 	 * if the path of the core is relative, add the current dir in front if it.
3715 	 */
3716 	if (name[0] != '/') {
3717 		fullpath = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
3718 		if (kern___getcwd(td, fullpath, UIO_SYSSPACE, MAXPATHLEN, MAXPATHLEN) != 0) {
3719 			free(fullpath, M_TEMP);
3720 			goto out2;
3721 		}
3722 		devctl_safe_quote_sb(sb, fullpath);
3723 		free(fullpath, M_TEMP);
3724 		sbuf_putc(sb, '/');
3725 	}
3726 	devctl_safe_quote_sb(sb, name);
3727 	sbuf_printf(sb, "\"");
3728 	if (sbuf_finish(sb) == 0)
3729 		devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
3730 out2:
3731 	sbuf_delete(sb);
3732 out:
3733 	error1 = vn_close(vp, FWRITE, cred, td);
3734 	if (error == 0)
3735 		error = error1;
3736 #ifdef AUDIT
3737 	audit_proc_coredump(td, name, error);
3738 #endif
3739 	free(name, M_TEMP);
3740 	return (error);
3741 }
3742 
3743 /*
3744  * Nonexistent system call-- signal process (may want to handle it).  Flag
3745  * error in case process won't see signal immediately (blocked or ignored).
3746  */
3747 #ifndef _SYS_SYSPROTO_H_
3748 struct nosys_args {
3749 	int	dummy;
3750 };
3751 #endif
3752 /* ARGSUSED */
3753 int
nosys(struct thread * td,struct nosys_args * args)3754 nosys(struct thread *td, struct nosys_args *args)
3755 {
3756 	struct proc *p;
3757 
3758 	p = td->td_proc;
3759 
3760 	PROC_LOCK(p);
3761 	tdsignal(td, SIGSYS);
3762 	PROC_UNLOCK(p);
3763 	if (kern_lognosys == 1 || kern_lognosys == 3) {
3764 		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
3765 		    td->td_sa.code);
3766 	}
3767 	if (kern_lognosys == 2 || kern_lognosys == 3 ||
3768 	    (p->p_pid == 1 && (kern_lognosys & 3) == 0)) {
3769 		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
3770 		    td->td_sa.code);
3771 	}
3772 	return (ENOSYS);
3773 }
3774 
3775 /*
3776  * Send a SIGIO or SIGURG signal to a process or process group using stored
3777  * credentials rather than those of the current process.
3778  */
3779 void
pgsigio(struct sigio ** sigiop,int sig,int checkctty)3780 pgsigio(struct sigio **sigiop, int sig, int checkctty)
3781 {
3782 	ksiginfo_t ksi;
3783 	struct sigio *sigio;
3784 
3785 	ksiginfo_init(&ksi);
3786 	ksi.ksi_signo = sig;
3787 	ksi.ksi_code = SI_KERNEL;
3788 
3789 	SIGIO_LOCK();
3790 	sigio = *sigiop;
3791 	if (sigio == NULL) {
3792 		SIGIO_UNLOCK();
3793 		return;
3794 	}
3795 	if (sigio->sio_pgid > 0) {
3796 		PROC_LOCK(sigio->sio_proc);
3797 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
3798 			kern_psignal(sigio->sio_proc, sig);
3799 		PROC_UNLOCK(sigio->sio_proc);
3800 	} else if (sigio->sio_pgid < 0) {
3801 		struct proc *p;
3802 
3803 		PGRP_LOCK(sigio->sio_pgrp);
3804 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
3805 			PROC_LOCK(p);
3806 			if (p->p_state == PRS_NORMAL &&
3807 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
3808 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
3809 				kern_psignal(p, sig);
3810 			PROC_UNLOCK(p);
3811 		}
3812 		PGRP_UNLOCK(sigio->sio_pgrp);
3813 	}
3814 	SIGIO_UNLOCK();
3815 }
3816 
3817 static int
filt_sigattach(struct knote * kn)3818 filt_sigattach(struct knote *kn)
3819 {
3820 	struct proc *p = curproc;
3821 
3822 	kn->kn_ptr.p_proc = p;
3823 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
3824 
3825 	knlist_add(p->p_klist, kn, 0);
3826 
3827 	return (0);
3828 }
3829 
3830 static void
filt_sigdetach(struct knote * kn)3831 filt_sigdetach(struct knote *kn)
3832 {
3833 	struct proc *p = kn->kn_ptr.p_proc;
3834 
3835 	knlist_remove(p->p_klist, kn, 0);
3836 }
3837 
3838 /*
3839  * signal knotes are shared with proc knotes, so we apply a mask to
3840  * the hint in order to differentiate them from process hints.  This
3841  * could be avoided by using a signal-specific knote list, but probably
3842  * isn't worth the trouble.
3843  */
3844 static int
filt_signal(struct knote * kn,long hint)3845 filt_signal(struct knote *kn, long hint)
3846 {
3847 
3848 	if (hint & NOTE_SIGNAL) {
3849 		hint &= ~NOTE_SIGNAL;
3850 
3851 		if (kn->kn_id == hint)
3852 			kn->kn_data++;
3853 	}
3854 	return (kn->kn_data != 0);
3855 }
3856 
3857 struct sigacts *
sigacts_alloc(void)3858 sigacts_alloc(void)
3859 {
3860 	struct sigacts *ps;
3861 
3862 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
3863 	refcount_init(&ps->ps_refcnt, 1);
3864 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
3865 	return (ps);
3866 }
3867 
3868 void
sigacts_free(struct sigacts * ps)3869 sigacts_free(struct sigacts *ps)
3870 {
3871 
3872 	if (refcount_release(&ps->ps_refcnt) == 0)
3873 		return;
3874 	mtx_destroy(&ps->ps_mtx);
3875 	free(ps, M_SUBPROC);
3876 }
3877 
3878 struct sigacts *
sigacts_hold(struct sigacts * ps)3879 sigacts_hold(struct sigacts *ps)
3880 {
3881 
3882 	refcount_acquire(&ps->ps_refcnt);
3883 	return (ps);
3884 }
3885 
3886 void
sigacts_copy(struct sigacts * dest,struct sigacts * src)3887 sigacts_copy(struct sigacts *dest, struct sigacts *src)
3888 {
3889 
3890 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
3891 	mtx_lock(&src->ps_mtx);
3892 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
3893 	mtx_unlock(&src->ps_mtx);
3894 }
3895 
3896 int
sigacts_shared(struct sigacts * ps)3897 sigacts_shared(struct sigacts *ps)
3898 {
3899 
3900 	return (ps->ps_refcnt > 1);
3901 }
3902 
3903 void
sig_drop_caught(struct proc * p)3904 sig_drop_caught(struct proc *p)
3905 {
3906 	int sig;
3907 	struct sigacts *ps;
3908 
3909 	ps = p->p_sigacts;
3910 	PROC_LOCK_ASSERT(p, MA_OWNED);
3911 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3912 	while (SIGNOTEMPTY(ps->ps_sigcatch)) {
3913 		sig = sig_ffs(&ps->ps_sigcatch);
3914 		sigdflt(ps, sig);
3915 		if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
3916 			sigqueue_delete_proc(p, sig);
3917 	}
3918 }
3919