xref: /freebsd-14.2/sys/kern/kern_sig.c (revision 109b7712)
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 "opt_capsicum.h"
40 #include "opt_ktrace.h"
41 
42 #include <sys/param.h>
43 #include <sys/capsicum.h>
44 #include <sys/ctype.h>
45 #include <sys/systm.h>
46 #include <sys/signalvar.h>
47 #include <sys/vnode.h>
48 #include <sys/acct.h>
49 #include <sys/capsicum.h>
50 #include <sys/compressor.h>
51 #include <sys/condvar.h>
52 #include <sys/devctl.h>
53 #include <sys/event.h>
54 #include <sys/fcntl.h>
55 #include <sys/imgact.h>
56 #include <sys/jail.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/racct.h>
71 #include <sys/resourcevar.h>
72 #include <sys/sdt.h>
73 #include <sys/sbuf.h>
74 #include <sys/sleepqueue.h>
75 #include <sys/smp.h>
76 #include <sys/stat.h>
77 #include <sys/sx.h>
78 #include <sys/syscall.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/vmmeter.h>
87 #include <sys/wait.h>
88 #include <vm/vm.h>
89 #include <vm/vm_extern.h>
90 #include <vm/uma.h>
91 
92 #include <machine/cpu.h>
93 
94 #include <security/audit/audit.h>
95 
96 #define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
97 
98 SDT_PROVIDER_DECLARE(proc);
99 SDT_PROBE_DEFINE3(proc, , , signal__send,
100     "struct thread *", "struct proc *", "int");
101 SDT_PROBE_DEFINE2(proc, , , signal__clear,
102     "int", "ksiginfo_t *");
103 SDT_PROBE_DEFINE3(proc, , , signal__discard,
104     "struct thread *", "struct proc *", "int");
105 
106 static int	coredump(struct thread *);
107 static int	killpg1(struct thread *td, int sig, int pgid, int all,
108 		    ksiginfo_t *ksi);
109 static int	issignal(struct thread *td);
110 static void	reschedule_signals(struct proc *p, sigset_t block, int flags);
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 *);
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, bool fast_sigblock);
118 static void	sigqueue_start(void);
119 static void	sigfastblock_setpend(struct thread *td, bool resched);
120 
121 static uma_zone_t	ksiginfo_zone = NULL;
122 struct filterops sig_filtops = {
123 	.f_isfd = 0,
124 	.f_attach = filt_sigattach,
125 	.f_detach = filt_sigdetach,
126 	.f_event = filt_signal,
127 };
128 
129 static int	kern_logsigexit = 1;
130 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
131     &kern_logsigexit, 0,
132     "Log processes quitting on abnormal signals to syslog(3)");
133 
134 static int	kern_forcesigexit = 1;
135 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
136     &kern_forcesigexit, 0, "Force trap signal to be handled");
137 
138 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
139     "POSIX real time signal");
140 
141 static int	max_pending_per_proc = 128;
142 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
143     &max_pending_per_proc, 0, "Max pending signals per proc");
144 
145 static int	preallocate_siginfo = 1024;
146 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
147     &preallocate_siginfo, 0, "Preallocated signal memory size");
148 
149 static int	signal_overflow = 0;
150 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
151     &signal_overflow, 0, "Number of signals overflew");
152 
153 static int	signal_alloc_fail = 0;
154 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
155     &signal_alloc_fail, 0, "signals failed to be allocated");
156 
157 static int	kern_lognosys = 0;
158 SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
159     "Log invalid syscalls");
160 
161 static int	kern_signosys = 1;
162 SYSCTL_INT(_kern, OID_AUTO, signosys, CTLFLAG_RWTUN, &kern_signosys, 0,
163     "Send SIGSYS on return from invalid syscall");
164 
165 __read_frequently bool sigfastblock_fetch_always = false;
166 SYSCTL_BOOL(_kern, OID_AUTO, sigfastblock_fetch_always, CTLFLAG_RWTUN,
167     &sigfastblock_fetch_always, 0,
168     "Fetch sigfastblock word on each syscall entry for proper "
169     "blocking semantic");
170 
171 static bool	kern_sig_discard_ign = true;
172 SYSCTL_BOOL(_kern, OID_AUTO, sig_discard_ign, CTLFLAG_RWTUN,
173     &kern_sig_discard_ign, 0,
174     "Discard ignored signals on delivery, otherwise queue them to "
175     "the target queue");
176 
177 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
178 
179 /*
180  * Policy -- Can ucred cr1 send SIGIO to process cr2?
181  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
182  * in the right situations.
183  */
184 #define CANSIGIO(cr1, cr2) \
185 	((cr1)->cr_uid == 0 || \
186 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
187 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
188 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
189 	    (cr1)->cr_uid == (cr2)->cr_uid)
190 
191 static int	sugid_coredump;
192 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
193     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
194 
195 static int	capmode_coredump;
196 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
197     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
198 
199 static int	do_coredump = 1;
200 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
201 	&do_coredump, 0, "Enable/Disable coredumps");
202 
203 static int	set_core_nodump_flag = 0;
204 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
205 	0, "Enable setting the NODUMP flag on coredump files");
206 
207 static int	coredump_devctl = 0;
208 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
209 	0, "Generate a devctl notification when processes coredump");
210 
211 /*
212  * Signal properties and actions.
213  * The array below categorizes the signals and their default actions
214  * according to the following properties:
215  */
216 #define	SIGPROP_KILL		0x01	/* terminates process by default */
217 #define	SIGPROP_CORE		0x02	/* ditto and coredumps */
218 #define	SIGPROP_STOP		0x04	/* suspend process */
219 #define	SIGPROP_TTYSTOP		0x08	/* ditto, from tty */
220 #define	SIGPROP_IGNORE		0x10	/* ignore by default */
221 #define	SIGPROP_CONT		0x20	/* continue if suspended */
222 
223 static const int sigproptbl[NSIG] = {
224 	[SIGHUP] =	SIGPROP_KILL,
225 	[SIGINT] =	SIGPROP_KILL,
226 	[SIGQUIT] =	SIGPROP_KILL | SIGPROP_CORE,
227 	[SIGILL] =	SIGPROP_KILL | SIGPROP_CORE,
228 	[SIGTRAP] =	SIGPROP_KILL | SIGPROP_CORE,
229 	[SIGABRT] =	SIGPROP_KILL | SIGPROP_CORE,
230 	[SIGEMT] =	SIGPROP_KILL | SIGPROP_CORE,
231 	[SIGFPE] =	SIGPROP_KILL | SIGPROP_CORE,
232 	[SIGKILL] =	SIGPROP_KILL,
233 	[SIGBUS] =	SIGPROP_KILL | SIGPROP_CORE,
234 	[SIGSEGV] =	SIGPROP_KILL | SIGPROP_CORE,
235 	[SIGSYS] =	SIGPROP_KILL | SIGPROP_CORE,
236 	[SIGPIPE] =	SIGPROP_KILL,
237 	[SIGALRM] =	SIGPROP_KILL,
238 	[SIGTERM] =	SIGPROP_KILL,
239 	[SIGURG] =	SIGPROP_IGNORE,
240 	[SIGSTOP] =	SIGPROP_STOP,
241 	[SIGTSTP] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
242 	[SIGCONT] =	SIGPROP_IGNORE | SIGPROP_CONT,
243 	[SIGCHLD] =	SIGPROP_IGNORE,
244 	[SIGTTIN] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
245 	[SIGTTOU] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
246 	[SIGIO] =	SIGPROP_IGNORE,
247 	[SIGXCPU] =	SIGPROP_KILL,
248 	[SIGXFSZ] =	SIGPROP_KILL,
249 	[SIGVTALRM] =	SIGPROP_KILL,
250 	[SIGPROF] =	SIGPROP_KILL,
251 	[SIGWINCH] =	SIGPROP_IGNORE,
252 	[SIGINFO] =	SIGPROP_IGNORE,
253 	[SIGUSR1] =	SIGPROP_KILL,
254 	[SIGUSR2] =	SIGPROP_KILL,
255 };
256 
257 #define	_SIG_FOREACH_ADVANCE(i, set) ({					\
258 	int __found;							\
259 	for (;;) {							\
260 		if (__bits != 0) {					\
261 			int __sig = ffs(__bits);			\
262 			__bits &= ~(1u << (__sig - 1));			\
263 			sig = __i * sizeof((set)->__bits[0]) * NBBY + __sig; \
264 			__found = 1;					\
265 			break;						\
266 		}							\
267 		if (++__i == _SIG_WORDS) {				\
268 			__found = 0;					\
269 			break;						\
270 		}							\
271 		__bits = (set)->__bits[__i];				\
272 	}								\
273 	__found != 0;							\
274 })
275 
276 #define	SIG_FOREACH(i, set)						\
277 	for (int32_t __i = -1, __bits = 0;				\
278 	    _SIG_FOREACH_ADVANCE(i, set); )				\
279 
280 static sigset_t fastblock_mask;
281 
282 static void
ast_sig(struct thread * td,int tda)283 ast_sig(struct thread *td, int tda)
284 {
285 	struct proc *p;
286 	int old_boundary, sig;
287 	bool resched_sigs;
288 
289 	p = td->td_proc;
290 
291 #ifdef DIAGNOSTIC
292 	if (p->p_numthreads == 1 && (tda & (TDAI(TDA_SIG) |
293 	    TDAI(TDA_AST))) == 0) {
294 		PROC_LOCK(p);
295 		thread_lock(td);
296 		/*
297 		 * Note that TDA_SIG should be re-read from
298 		 * td_ast, since signal might have been delivered
299 		 * after we cleared td_flags above.  This is one of
300 		 * the reason for looping check for AST condition.
301 		 * See comment in userret() about P_PPWAIT.
302 		 */
303 		if ((p->p_flag & P_PPWAIT) == 0 &&
304 		    (td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
305 			if (SIGPENDING(td) && ((tda | td->td_ast) &
306 			    (TDAI(TDA_SIG) | TDAI(TDA_AST))) == 0) {
307 				thread_unlock(td); /* fix dumps */
308 				panic(
309 				    "failed2 to set signal flags for ast p %p "
310 				    "td %p tda %#x td_ast %#x fl %#x",
311 				    p, td, tda, td->td_ast, td->td_flags);
312 			}
313 		}
314 		thread_unlock(td);
315 		PROC_UNLOCK(p);
316 	}
317 #endif
318 
319 	/*
320 	 * Check for signals. Unlocked reads of p_pendingcnt or
321 	 * p_siglist might cause process-directed signal to be handled
322 	 * later.
323 	 */
324 	if ((tda & TDAI(TDA_SIG)) != 0 || p->p_pendingcnt > 0 ||
325 	    !SIGISEMPTY(p->p_siglist)) {
326 		sigfastblock_fetch(td);
327 		PROC_LOCK(p);
328 		old_boundary = ~TDB_BOUNDARY | (td->td_dbgflags & TDB_BOUNDARY);
329 		td->td_dbgflags |= TDB_BOUNDARY;
330 		mtx_lock(&p->p_sigacts->ps_mtx);
331 		while ((sig = cursig(td)) != 0) {
332 			KASSERT(sig >= 0, ("sig %d", sig));
333 			postsig(sig);
334 		}
335 		mtx_unlock(&p->p_sigacts->ps_mtx);
336 		td->td_dbgflags &= old_boundary;
337 		PROC_UNLOCK(p);
338 		resched_sigs = true;
339 	} else {
340 		resched_sigs = false;
341 	}
342 
343 	/*
344 	 * Handle deferred update of the fast sigblock value, after
345 	 * the postsig() loop was performed.
346 	 */
347 	sigfastblock_setpend(td, resched_sigs);
348 }
349 
350 static void
ast_sigsuspend(struct thread * td,int tda __unused)351 ast_sigsuspend(struct thread *td, int tda __unused)
352 {
353 	MPASS((td->td_pflags & TDP_OLDMASK) != 0);
354 	td->td_pflags &= ~TDP_OLDMASK;
355 	kern_sigprocmask(td, SIG_SETMASK, &td->td_oldsigmask, NULL, 0);
356 }
357 
358 static void
sigqueue_start(void)359 sigqueue_start(void)
360 {
361 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
362 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
363 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
364 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
365 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
366 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
367 	SIGFILLSET(fastblock_mask);
368 	SIG_CANTMASK(fastblock_mask);
369 	ast_register(TDA_SIG, ASTR_UNCOND, 0, ast_sig);
370 	ast_register(TDA_SIGSUSPEND, ASTR_ASTF_REQUIRED | ASTR_TDP,
371 	    TDP_OLDMASK, ast_sigsuspend);
372 }
373 
374 ksiginfo_t *
ksiginfo_alloc(int mwait)375 ksiginfo_alloc(int mwait)
376 {
377 	MPASS(mwait == M_WAITOK || mwait == M_NOWAIT);
378 
379 	if (ksiginfo_zone == NULL)
380 		return (NULL);
381 	return (uma_zalloc(ksiginfo_zone, mwait | M_ZERO));
382 }
383 
384 void
ksiginfo_free(ksiginfo_t * ksi)385 ksiginfo_free(ksiginfo_t *ksi)
386 {
387 	uma_zfree(ksiginfo_zone, ksi);
388 }
389 
390 static __inline bool
ksiginfo_tryfree(ksiginfo_t * ksi)391 ksiginfo_tryfree(ksiginfo_t *ksi)
392 {
393 	if ((ksi->ksi_flags & KSI_EXT) == 0) {
394 		uma_zfree(ksiginfo_zone, ksi);
395 		return (true);
396 	}
397 	return (false);
398 }
399 
400 void
sigqueue_init(sigqueue_t * list,struct proc * p)401 sigqueue_init(sigqueue_t *list, struct proc *p)
402 {
403 	SIGEMPTYSET(list->sq_signals);
404 	SIGEMPTYSET(list->sq_kill);
405 	SIGEMPTYSET(list->sq_ptrace);
406 	TAILQ_INIT(&list->sq_list);
407 	list->sq_proc = p;
408 	list->sq_flags = SQ_INIT;
409 }
410 
411 /*
412  * Get a signal's ksiginfo.
413  * Return:
414  *	0	-	signal not found
415  *	others	-	signal number
416  */
417 static int
sigqueue_get(sigqueue_t * sq,int signo,ksiginfo_t * si)418 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
419 {
420 	struct proc *p = sq->sq_proc;
421 	struct ksiginfo *ksi, *next;
422 	int count = 0;
423 
424 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
425 
426 	if (!SIGISMEMBER(sq->sq_signals, signo))
427 		return (0);
428 
429 	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
430 		count++;
431 		SIGDELSET(sq->sq_ptrace, signo);
432 		si->ksi_flags |= KSI_PTRACE;
433 	}
434 	if (SIGISMEMBER(sq->sq_kill, signo)) {
435 		count++;
436 		if (count == 1)
437 			SIGDELSET(sq->sq_kill, signo);
438 	}
439 
440 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
441 		if (ksi->ksi_signo == signo) {
442 			if (count == 0) {
443 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
444 				ksi->ksi_sigq = NULL;
445 				ksiginfo_copy(ksi, si);
446 				if (ksiginfo_tryfree(ksi) && p != NULL)
447 					p->p_pendingcnt--;
448 			}
449 			if (++count > 1)
450 				break;
451 		}
452 	}
453 
454 	if (count <= 1)
455 		SIGDELSET(sq->sq_signals, signo);
456 	si->ksi_signo = signo;
457 	return (signo);
458 }
459 
460 void
sigqueue_take(ksiginfo_t * ksi)461 sigqueue_take(ksiginfo_t *ksi)
462 {
463 	struct ksiginfo *kp;
464 	struct proc	*p;
465 	sigqueue_t	*sq;
466 
467 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
468 		return;
469 
470 	p = sq->sq_proc;
471 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
472 	ksi->ksi_sigq = NULL;
473 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
474 		p->p_pendingcnt--;
475 
476 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
477 	     kp = TAILQ_NEXT(kp, ksi_link)) {
478 		if (kp->ksi_signo == ksi->ksi_signo)
479 			break;
480 	}
481 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
482 	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
483 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
484 }
485 
486 static int
sigqueue_add(sigqueue_t * sq,int signo,ksiginfo_t * si)487 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
488 {
489 	struct proc *p = sq->sq_proc;
490 	struct ksiginfo *ksi;
491 	int ret = 0;
492 
493 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
494 
495 	/*
496 	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
497 	 * for these signals.
498 	 */
499 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
500 		SIGADDSET(sq->sq_kill, signo);
501 		goto out_set_bit;
502 	}
503 
504 	/* directly insert the ksi, don't copy it */
505 	if (si->ksi_flags & KSI_INS) {
506 		if (si->ksi_flags & KSI_HEAD)
507 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
508 		else
509 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
510 		si->ksi_sigq = sq;
511 		goto out_set_bit;
512 	}
513 
514 	if (__predict_false(ksiginfo_zone == NULL)) {
515 		SIGADDSET(sq->sq_kill, signo);
516 		goto out_set_bit;
517 	}
518 
519 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
520 		signal_overflow++;
521 		ret = EAGAIN;
522 	} else if ((ksi = ksiginfo_alloc(M_NOWAIT)) == NULL) {
523 		signal_alloc_fail++;
524 		ret = EAGAIN;
525 	} else {
526 		if (p != NULL)
527 			p->p_pendingcnt++;
528 		ksiginfo_copy(si, ksi);
529 		ksi->ksi_signo = signo;
530 		if (si->ksi_flags & KSI_HEAD)
531 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
532 		else
533 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
534 		ksi->ksi_sigq = sq;
535 	}
536 
537 	if (ret != 0) {
538 		if ((si->ksi_flags & KSI_PTRACE) != 0) {
539 			SIGADDSET(sq->sq_ptrace, signo);
540 			ret = 0;
541 			goto out_set_bit;
542 		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
543 		    (si->ksi_flags & KSI_SIGQ) == 0) {
544 			SIGADDSET(sq->sq_kill, signo);
545 			ret = 0;
546 			goto out_set_bit;
547 		}
548 		return (ret);
549 	}
550 
551 out_set_bit:
552 	SIGADDSET(sq->sq_signals, signo);
553 	return (ret);
554 }
555 
556 void
sigqueue_flush(sigqueue_t * sq)557 sigqueue_flush(sigqueue_t *sq)
558 {
559 	struct proc *p = sq->sq_proc;
560 	ksiginfo_t *ksi;
561 
562 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
563 
564 	if (p != NULL)
565 		PROC_LOCK_ASSERT(p, MA_OWNED);
566 
567 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
568 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
569 		ksi->ksi_sigq = NULL;
570 		if (ksiginfo_tryfree(ksi) && p != NULL)
571 			p->p_pendingcnt--;
572 	}
573 
574 	SIGEMPTYSET(sq->sq_signals);
575 	SIGEMPTYSET(sq->sq_kill);
576 	SIGEMPTYSET(sq->sq_ptrace);
577 }
578 
579 static void
sigqueue_move_set(sigqueue_t * src,sigqueue_t * dst,const sigset_t * set)580 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
581 {
582 	sigset_t tmp;
583 	struct proc *p1, *p2;
584 	ksiginfo_t *ksi, *next;
585 
586 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
587 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
588 	p1 = src->sq_proc;
589 	p2 = dst->sq_proc;
590 	/* Move siginfo to target list */
591 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
592 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
593 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
594 			if (p1 != NULL)
595 				p1->p_pendingcnt--;
596 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
597 			ksi->ksi_sigq = dst;
598 			if (p2 != NULL)
599 				p2->p_pendingcnt++;
600 		}
601 	}
602 
603 	/* Move pending bits to target list */
604 	tmp = src->sq_kill;
605 	SIGSETAND(tmp, *set);
606 	SIGSETOR(dst->sq_kill, tmp);
607 	SIGSETNAND(src->sq_kill, tmp);
608 
609 	tmp = src->sq_ptrace;
610 	SIGSETAND(tmp, *set);
611 	SIGSETOR(dst->sq_ptrace, tmp);
612 	SIGSETNAND(src->sq_ptrace, tmp);
613 
614 	tmp = src->sq_signals;
615 	SIGSETAND(tmp, *set);
616 	SIGSETOR(dst->sq_signals, tmp);
617 	SIGSETNAND(src->sq_signals, tmp);
618 }
619 
620 #if 0
621 static void
622 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
623 {
624 	sigset_t set;
625 
626 	SIGEMPTYSET(set);
627 	SIGADDSET(set, signo);
628 	sigqueue_move_set(src, dst, &set);
629 }
630 #endif
631 
632 static void
sigqueue_delete_set(sigqueue_t * sq,const sigset_t * set)633 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
634 {
635 	struct proc *p = sq->sq_proc;
636 	ksiginfo_t *ksi, *next;
637 
638 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
639 
640 	/* Remove siginfo queue */
641 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
642 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
643 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
644 			ksi->ksi_sigq = NULL;
645 			if (ksiginfo_tryfree(ksi) && p != NULL)
646 				p->p_pendingcnt--;
647 		}
648 	}
649 	SIGSETNAND(sq->sq_kill, *set);
650 	SIGSETNAND(sq->sq_ptrace, *set);
651 	SIGSETNAND(sq->sq_signals, *set);
652 }
653 
654 void
sigqueue_delete(sigqueue_t * sq,int signo)655 sigqueue_delete(sigqueue_t *sq, int signo)
656 {
657 	sigset_t set;
658 
659 	SIGEMPTYSET(set);
660 	SIGADDSET(set, signo);
661 	sigqueue_delete_set(sq, &set);
662 }
663 
664 /* Remove a set of signals for a process */
665 static void
sigqueue_delete_set_proc(struct proc * p,const sigset_t * set)666 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
667 {
668 	sigqueue_t worklist;
669 	struct thread *td0;
670 
671 	PROC_LOCK_ASSERT(p, MA_OWNED);
672 
673 	sigqueue_init(&worklist, NULL);
674 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
675 
676 	FOREACH_THREAD_IN_PROC(p, td0)
677 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
678 
679 	sigqueue_flush(&worklist);
680 }
681 
682 void
sigqueue_delete_proc(struct proc * p,int signo)683 sigqueue_delete_proc(struct proc *p, int signo)
684 {
685 	sigset_t set;
686 
687 	SIGEMPTYSET(set);
688 	SIGADDSET(set, signo);
689 	sigqueue_delete_set_proc(p, &set);
690 }
691 
692 static void
sigqueue_delete_stopmask_proc(struct proc * p)693 sigqueue_delete_stopmask_proc(struct proc *p)
694 {
695 	sigset_t set;
696 
697 	SIGEMPTYSET(set);
698 	SIGADDSET(set, SIGSTOP);
699 	SIGADDSET(set, SIGTSTP);
700 	SIGADDSET(set, SIGTTIN);
701 	SIGADDSET(set, SIGTTOU);
702 	sigqueue_delete_set_proc(p, &set);
703 }
704 
705 /*
706  * Determine signal that should be delivered to thread td, the current
707  * thread, 0 if none.  If there is a pending stop signal with default
708  * action, the process stops in issignal().
709  */
710 int
cursig(struct thread * td)711 cursig(struct thread *td)
712 {
713 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
714 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
715 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
716 	return (SIGPENDING(td) ? issignal(td) : 0);
717 }
718 
719 /*
720  * Arrange for ast() to handle unmasked pending signals on return to user
721  * mode.  This must be called whenever a signal is added to td_sigqueue or
722  * unmasked in td_sigmask.
723  */
724 void
signotify(struct thread * td)725 signotify(struct thread *td)
726 {
727 
728 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
729 
730 	if (SIGPENDING(td))
731 		ast_sched(td, TDA_SIG);
732 }
733 
734 /*
735  * Returns 1 (true) if altstack is configured for the thread, and the
736  * passed stack bottom address falls into the altstack range.  Handles
737  * the 43 compat special case where the alt stack size is zero.
738  */
739 int
sigonstack(size_t sp)740 sigonstack(size_t sp)
741 {
742 	struct thread *td;
743 
744 	td = curthread;
745 	if ((td->td_pflags & TDP_ALTSTACK) == 0)
746 		return (0);
747 #if defined(COMPAT_43)
748 	if (SV_PROC_FLAG(td->td_proc, SV_AOUT) && td->td_sigstk.ss_size == 0)
749 		return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
750 #endif
751 	return (sp >= (size_t)td->td_sigstk.ss_sp &&
752 	    sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
753 }
754 
755 static __inline int
sigprop(int sig)756 sigprop(int sig)
757 {
758 
759 	if (sig > 0 && sig < nitems(sigproptbl))
760 		return (sigproptbl[sig]);
761 	return (0);
762 }
763 
764 static bool
sigact_flag_test(const struct sigaction * act,int flag)765 sigact_flag_test(const struct sigaction *act, int flag)
766 {
767 
768 	/*
769 	 * SA_SIGINFO is reset when signal disposition is set to
770 	 * ignore or default.  Other flags are kept according to user
771 	 * settings.
772 	 */
773 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
774 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
775 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
776 }
777 
778 /*
779  * kern_sigaction
780  * sigaction
781  * freebsd4_sigaction
782  * osigaction
783  */
784 int
kern_sigaction(struct thread * td,int sig,const struct sigaction * act,struct sigaction * oact,int flags)785 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
786     struct sigaction *oact, int flags)
787 {
788 	struct sigacts *ps;
789 	struct proc *p = td->td_proc;
790 
791 	if (!_SIG_VALID(sig))
792 		return (EINVAL);
793 	if (act != NULL && act->sa_handler != SIG_DFL &&
794 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
795 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
796 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
797 		return (EINVAL);
798 
799 	PROC_LOCK(p);
800 	ps = p->p_sigacts;
801 	mtx_lock(&ps->ps_mtx);
802 	if (oact) {
803 		memset(oact, 0, sizeof(*oact));
804 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
805 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
806 			oact->sa_flags |= SA_ONSTACK;
807 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
808 			oact->sa_flags |= SA_RESTART;
809 		if (SIGISMEMBER(ps->ps_sigreset, sig))
810 			oact->sa_flags |= SA_RESETHAND;
811 		if (SIGISMEMBER(ps->ps_signodefer, sig))
812 			oact->sa_flags |= SA_NODEFER;
813 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
814 			oact->sa_flags |= SA_SIGINFO;
815 			oact->sa_sigaction =
816 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
817 		} else
818 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
819 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
820 			oact->sa_flags |= SA_NOCLDSTOP;
821 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
822 			oact->sa_flags |= SA_NOCLDWAIT;
823 	}
824 	if (act) {
825 		if ((sig == SIGKILL || sig == SIGSTOP) &&
826 		    act->sa_handler != SIG_DFL) {
827 			mtx_unlock(&ps->ps_mtx);
828 			PROC_UNLOCK(p);
829 			return (EINVAL);
830 		}
831 
832 		/*
833 		 * Change setting atomically.
834 		 */
835 
836 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
837 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
838 		if (sigact_flag_test(act, SA_SIGINFO)) {
839 			ps->ps_sigact[_SIG_IDX(sig)] =
840 			    (__sighandler_t *)act->sa_sigaction;
841 			SIGADDSET(ps->ps_siginfo, sig);
842 		} else {
843 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
844 			SIGDELSET(ps->ps_siginfo, sig);
845 		}
846 		if (!sigact_flag_test(act, SA_RESTART))
847 			SIGADDSET(ps->ps_sigintr, sig);
848 		else
849 			SIGDELSET(ps->ps_sigintr, sig);
850 		if (sigact_flag_test(act, SA_ONSTACK))
851 			SIGADDSET(ps->ps_sigonstack, sig);
852 		else
853 			SIGDELSET(ps->ps_sigonstack, sig);
854 		if (sigact_flag_test(act, SA_RESETHAND))
855 			SIGADDSET(ps->ps_sigreset, sig);
856 		else
857 			SIGDELSET(ps->ps_sigreset, sig);
858 		if (sigact_flag_test(act, SA_NODEFER))
859 			SIGADDSET(ps->ps_signodefer, sig);
860 		else
861 			SIGDELSET(ps->ps_signodefer, sig);
862 		if (sig == SIGCHLD) {
863 			if (act->sa_flags & SA_NOCLDSTOP)
864 				ps->ps_flag |= PS_NOCLDSTOP;
865 			else
866 				ps->ps_flag &= ~PS_NOCLDSTOP;
867 			if (act->sa_flags & SA_NOCLDWAIT) {
868 				/*
869 				 * Paranoia: since SA_NOCLDWAIT is implemented
870 				 * by reparenting the dying child to PID 1 (and
871 				 * trust it to reap the zombie), PID 1 itself
872 				 * is forbidden to set SA_NOCLDWAIT.
873 				 */
874 				if (p->p_pid == 1)
875 					ps->ps_flag &= ~PS_NOCLDWAIT;
876 				else
877 					ps->ps_flag |= PS_NOCLDWAIT;
878 			} else
879 				ps->ps_flag &= ~PS_NOCLDWAIT;
880 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
881 				ps->ps_flag |= PS_CLDSIGIGN;
882 			else
883 				ps->ps_flag &= ~PS_CLDSIGIGN;
884 		}
885 		/*
886 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
887 		 * and for signals set to SIG_DFL where the default is to
888 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
889 		 * have to restart the process.
890 		 */
891 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
892 		    (sigprop(sig) & SIGPROP_IGNORE &&
893 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
894 			/* never to be seen again */
895 			sigqueue_delete_proc(p, sig);
896 			if (sig != SIGCONT)
897 				/* easier in psignal */
898 				SIGADDSET(ps->ps_sigignore, sig);
899 			SIGDELSET(ps->ps_sigcatch, sig);
900 		} else {
901 			SIGDELSET(ps->ps_sigignore, sig);
902 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
903 				SIGDELSET(ps->ps_sigcatch, sig);
904 			else
905 				SIGADDSET(ps->ps_sigcatch, sig);
906 		}
907 #ifdef COMPAT_FREEBSD4
908 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
909 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
910 		    (flags & KSA_FREEBSD4) == 0)
911 			SIGDELSET(ps->ps_freebsd4, sig);
912 		else
913 			SIGADDSET(ps->ps_freebsd4, sig);
914 #endif
915 #ifdef COMPAT_43
916 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
917 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
918 		    (flags & KSA_OSIGSET) == 0)
919 			SIGDELSET(ps->ps_osigset, sig);
920 		else
921 			SIGADDSET(ps->ps_osigset, sig);
922 #endif
923 	}
924 	mtx_unlock(&ps->ps_mtx);
925 	PROC_UNLOCK(p);
926 	return (0);
927 }
928 
929 #ifndef _SYS_SYSPROTO_H_
930 struct sigaction_args {
931 	int	sig;
932 	struct	sigaction *act;
933 	struct	sigaction *oact;
934 };
935 #endif
936 int
sys_sigaction(struct thread * td,struct sigaction_args * uap)937 sys_sigaction(struct thread *td, struct sigaction_args *uap)
938 {
939 	struct sigaction act, oact;
940 	struct sigaction *actp, *oactp;
941 	int error;
942 
943 	actp = (uap->act != NULL) ? &act : NULL;
944 	oactp = (uap->oact != NULL) ? &oact : NULL;
945 	if (actp) {
946 		error = copyin(uap->act, actp, sizeof(act));
947 		if (error)
948 			return (error);
949 	}
950 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
951 	if (oactp && !error)
952 		error = copyout(oactp, uap->oact, sizeof(oact));
953 	return (error);
954 }
955 
956 #ifdef COMPAT_FREEBSD4
957 #ifndef _SYS_SYSPROTO_H_
958 struct freebsd4_sigaction_args {
959 	int	sig;
960 	struct	sigaction *act;
961 	struct	sigaction *oact;
962 };
963 #endif
964 int
freebsd4_sigaction(struct thread * td,struct freebsd4_sigaction_args * uap)965 freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
966 {
967 	struct sigaction act, oact;
968 	struct sigaction *actp, *oactp;
969 	int error;
970 
971 	actp = (uap->act != NULL) ? &act : NULL;
972 	oactp = (uap->oact != NULL) ? &oact : NULL;
973 	if (actp) {
974 		error = copyin(uap->act, actp, sizeof(act));
975 		if (error)
976 			return (error);
977 	}
978 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
979 	if (oactp && !error)
980 		error = copyout(oactp, uap->oact, sizeof(oact));
981 	return (error);
982 }
983 #endif	/* COMAPT_FREEBSD4 */
984 
985 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
986 #ifndef _SYS_SYSPROTO_H_
987 struct osigaction_args {
988 	int	signum;
989 	struct	osigaction *nsa;
990 	struct	osigaction *osa;
991 };
992 #endif
993 int
osigaction(struct thread * td,struct osigaction_args * uap)994 osigaction(struct thread *td, struct osigaction_args *uap)
995 {
996 	struct osigaction sa;
997 	struct sigaction nsa, osa;
998 	struct sigaction *nsap, *osap;
999 	int error;
1000 
1001 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1002 		return (EINVAL);
1003 
1004 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
1005 	osap = (uap->osa != NULL) ? &osa : NULL;
1006 
1007 	if (nsap) {
1008 		error = copyin(uap->nsa, &sa, sizeof(sa));
1009 		if (error)
1010 			return (error);
1011 		nsap->sa_handler = sa.sa_handler;
1012 		nsap->sa_flags = sa.sa_flags;
1013 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
1014 	}
1015 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1016 	if (osap && !error) {
1017 		sa.sa_handler = osap->sa_handler;
1018 		sa.sa_flags = osap->sa_flags;
1019 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
1020 		error = copyout(&sa, uap->osa, sizeof(sa));
1021 	}
1022 	return (error);
1023 }
1024 
1025 #if !defined(__i386__)
1026 /* Avoid replicating the same stub everywhere */
1027 int
osigreturn(struct thread * td,struct osigreturn_args * uap)1028 osigreturn(struct thread *td, struct osigreturn_args *uap)
1029 {
1030 
1031 	return (nosys(td, (struct nosys_args *)uap));
1032 }
1033 #endif
1034 #endif /* COMPAT_43 */
1035 
1036 /*
1037  * Initialize signal state for process 0;
1038  * set to ignore signals that are ignored by default.
1039  */
1040 void
siginit(struct proc * p)1041 siginit(struct proc *p)
1042 {
1043 	int i;
1044 	struct sigacts *ps;
1045 
1046 	PROC_LOCK(p);
1047 	ps = p->p_sigacts;
1048 	mtx_lock(&ps->ps_mtx);
1049 	for (i = 1; i <= NSIG; i++) {
1050 		if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) {
1051 			SIGADDSET(ps->ps_sigignore, i);
1052 		}
1053 	}
1054 	mtx_unlock(&ps->ps_mtx);
1055 	PROC_UNLOCK(p);
1056 }
1057 
1058 /*
1059  * Reset specified signal to the default disposition.
1060  */
1061 static void
sigdflt(struct sigacts * ps,int sig)1062 sigdflt(struct sigacts *ps, int sig)
1063 {
1064 
1065 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1066 	SIGDELSET(ps->ps_sigcatch, sig);
1067 	if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT)
1068 		SIGADDSET(ps->ps_sigignore, sig);
1069 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1070 	SIGDELSET(ps->ps_siginfo, sig);
1071 }
1072 
1073 /*
1074  * Reset signals for an exec of the specified process.
1075  */
1076 void
execsigs(struct proc * p)1077 execsigs(struct proc *p)
1078 {
1079 	struct sigacts *ps;
1080 	struct thread *td;
1081 
1082 	/*
1083 	 * Reset caught signals.  Held signals remain held
1084 	 * through td_sigmask (unless they were caught,
1085 	 * and are now ignored by default).
1086 	 */
1087 	PROC_LOCK_ASSERT(p, MA_OWNED);
1088 	ps = p->p_sigacts;
1089 	mtx_lock(&ps->ps_mtx);
1090 	sig_drop_caught(p);
1091 
1092 	/*
1093 	 * Reset stack state to the user stack.
1094 	 * Clear set of signals caught on the signal stack.
1095 	 */
1096 	td = curthread;
1097 	MPASS(td->td_proc == p);
1098 	td->td_sigstk.ss_flags = SS_DISABLE;
1099 	td->td_sigstk.ss_size = 0;
1100 	td->td_sigstk.ss_sp = 0;
1101 	td->td_pflags &= ~TDP_ALTSTACK;
1102 	/*
1103 	 * Reset no zombies if child dies flag as Solaris does.
1104 	 */
1105 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1106 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1107 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1108 	mtx_unlock(&ps->ps_mtx);
1109 }
1110 
1111 /*
1112  * kern_sigprocmask()
1113  *
1114  *	Manipulate signal mask.
1115  */
1116 int
kern_sigprocmask(struct thread * td,int how,sigset_t * set,sigset_t * oset,int flags)1117 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1118     int flags)
1119 {
1120 	sigset_t new_block, oset1;
1121 	struct proc *p;
1122 	int error;
1123 
1124 	p = td->td_proc;
1125 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1126 		PROC_LOCK_ASSERT(p, MA_OWNED);
1127 	else
1128 		PROC_LOCK(p);
1129 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1130 	    ? MA_OWNED : MA_NOTOWNED);
1131 	if (oset != NULL)
1132 		*oset = td->td_sigmask;
1133 
1134 	error = 0;
1135 	if (set != NULL) {
1136 		switch (how) {
1137 		case SIG_BLOCK:
1138 			SIG_CANTMASK(*set);
1139 			oset1 = td->td_sigmask;
1140 			SIGSETOR(td->td_sigmask, *set);
1141 			new_block = td->td_sigmask;
1142 			SIGSETNAND(new_block, oset1);
1143 			break;
1144 		case SIG_UNBLOCK:
1145 			SIGSETNAND(td->td_sigmask, *set);
1146 			signotify(td);
1147 			goto out;
1148 		case SIG_SETMASK:
1149 			SIG_CANTMASK(*set);
1150 			oset1 = td->td_sigmask;
1151 			if (flags & SIGPROCMASK_OLD)
1152 				SIGSETLO(td->td_sigmask, *set);
1153 			else
1154 				td->td_sigmask = *set;
1155 			new_block = td->td_sigmask;
1156 			SIGSETNAND(new_block, oset1);
1157 			signotify(td);
1158 			break;
1159 		default:
1160 			error = EINVAL;
1161 			goto out;
1162 		}
1163 
1164 		/*
1165 		 * The new_block set contains signals that were not previously
1166 		 * blocked, but are blocked now.
1167 		 *
1168 		 * In case we block any signal that was not previously blocked
1169 		 * for td, and process has the signal pending, try to schedule
1170 		 * signal delivery to some thread that does not block the
1171 		 * signal, possibly waking it up.
1172 		 */
1173 		if (p->p_numthreads != 1)
1174 			reschedule_signals(p, new_block, flags);
1175 	}
1176 
1177 out:
1178 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1179 		PROC_UNLOCK(p);
1180 	return (error);
1181 }
1182 
1183 #ifndef _SYS_SYSPROTO_H_
1184 struct sigprocmask_args {
1185 	int	how;
1186 	const sigset_t *set;
1187 	sigset_t *oset;
1188 };
1189 #endif
1190 int
sys_sigprocmask(struct thread * td,struct sigprocmask_args * uap)1191 sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1192 {
1193 	sigset_t set, oset;
1194 	sigset_t *setp, *osetp;
1195 	int error;
1196 
1197 	setp = (uap->set != NULL) ? &set : NULL;
1198 	osetp = (uap->oset != NULL) ? &oset : NULL;
1199 	if (setp) {
1200 		error = copyin(uap->set, setp, sizeof(set));
1201 		if (error)
1202 			return (error);
1203 	}
1204 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1205 	if (osetp && !error) {
1206 		error = copyout(osetp, uap->oset, sizeof(oset));
1207 	}
1208 	return (error);
1209 }
1210 
1211 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1212 #ifndef _SYS_SYSPROTO_H_
1213 struct osigprocmask_args {
1214 	int	how;
1215 	osigset_t mask;
1216 };
1217 #endif
1218 int
osigprocmask(struct thread * td,struct osigprocmask_args * uap)1219 osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1220 {
1221 	sigset_t set, oset;
1222 	int error;
1223 
1224 	OSIG2SIG(uap->mask, set);
1225 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1226 	SIG2OSIG(oset, td->td_retval[0]);
1227 	return (error);
1228 }
1229 #endif /* COMPAT_43 */
1230 
1231 int
sys_sigwait(struct thread * td,struct sigwait_args * uap)1232 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1233 {
1234 	ksiginfo_t ksi;
1235 	sigset_t set;
1236 	int error;
1237 
1238 	error = copyin(uap->set, &set, sizeof(set));
1239 	if (error) {
1240 		td->td_retval[0] = error;
1241 		return (0);
1242 	}
1243 
1244 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1245 	if (error) {
1246 		/*
1247 		 * sigwait() function shall not return EINTR, but
1248 		 * the syscall does.  Non-ancient libc provides the
1249 		 * wrapper which hides EINTR.  Otherwise, EINTR return
1250 		 * is used by libthr to handle required cancellation
1251 		 * point in the sigwait().
1252 		 */
1253 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1254 			return (ERESTART);
1255 		td->td_retval[0] = error;
1256 		return (0);
1257 	}
1258 
1259 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1260 	td->td_retval[0] = error;
1261 	return (0);
1262 }
1263 
1264 int
sys_sigtimedwait(struct thread * td,struct sigtimedwait_args * uap)1265 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1266 {
1267 	struct timespec ts;
1268 	struct timespec *timeout;
1269 	sigset_t set;
1270 	ksiginfo_t ksi;
1271 	int error;
1272 
1273 	if (uap->timeout) {
1274 		error = copyin(uap->timeout, &ts, sizeof(ts));
1275 		if (error)
1276 			return (error);
1277 
1278 		timeout = &ts;
1279 	} else
1280 		timeout = NULL;
1281 
1282 	error = copyin(uap->set, &set, sizeof(set));
1283 	if (error)
1284 		return (error);
1285 
1286 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1287 	if (error)
1288 		return (error);
1289 
1290 	if (uap->info)
1291 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1292 
1293 	if (error == 0)
1294 		td->td_retval[0] = ksi.ksi_signo;
1295 	return (error);
1296 }
1297 
1298 int
sys_sigwaitinfo(struct thread * td,struct sigwaitinfo_args * uap)1299 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1300 {
1301 	ksiginfo_t ksi;
1302 	sigset_t set;
1303 	int error;
1304 
1305 	error = copyin(uap->set, &set, sizeof(set));
1306 	if (error)
1307 		return (error);
1308 
1309 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1310 	if (error)
1311 		return (error);
1312 
1313 	if (uap->info)
1314 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1315 
1316 	if (error == 0)
1317 		td->td_retval[0] = ksi.ksi_signo;
1318 	return (error);
1319 }
1320 
1321 static void
proc_td_siginfo_capture(struct thread * td,siginfo_t * si)1322 proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1323 {
1324 	struct thread *thr;
1325 
1326 	FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1327 		if (thr == td)
1328 			thr->td_si = *si;
1329 		else
1330 			thr->td_si.si_signo = 0;
1331 	}
1332 }
1333 
1334 int
kern_sigtimedwait(struct thread * td,sigset_t waitset,ksiginfo_t * ksi,struct timespec * timeout)1335 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1336 	struct timespec *timeout)
1337 {
1338 	struct sigacts *ps;
1339 	sigset_t saved_mask, new_block;
1340 	struct proc *p;
1341 	int error, sig, timevalid = 0;
1342 	sbintime_t sbt, precision, tsbt;
1343 	struct timespec ts;
1344 	bool traced;
1345 
1346 	p = td->td_proc;
1347 	error = 0;
1348 	traced = false;
1349 
1350 	/* Ensure the sigfastblock value is up to date. */
1351 	sigfastblock_fetch(td);
1352 
1353 	if (timeout != NULL) {
1354 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1355 			timevalid = 1;
1356 			ts = *timeout;
1357 			if (ts.tv_sec < INT32_MAX / 2) {
1358 				tsbt = tstosbt(ts);
1359 				precision = tsbt;
1360 				precision >>= tc_precexp;
1361 				if (TIMESEL(&sbt, tsbt))
1362 					sbt += tc_tick_sbt;
1363 				sbt += tsbt;
1364 			} else
1365 				precision = sbt = 0;
1366 		}
1367 	} else
1368 		precision = sbt = 0;
1369 	ksiginfo_init(ksi);
1370 	/* Some signals can not be waited for. */
1371 	SIG_CANTMASK(waitset);
1372 	ps = p->p_sigacts;
1373 	PROC_LOCK(p);
1374 	saved_mask = td->td_sigmask;
1375 	SIGSETNAND(td->td_sigmask, waitset);
1376 	if ((p->p_sysent->sv_flags & SV_SIG_DISCIGN) != 0 ||
1377 	    !kern_sig_discard_ign) {
1378 		thread_lock(td);
1379 		td->td_flags |= TDF_SIGWAIT;
1380 		thread_unlock(td);
1381 	}
1382 	for (;;) {
1383 		mtx_lock(&ps->ps_mtx);
1384 		sig = cursig(td);
1385 		mtx_unlock(&ps->ps_mtx);
1386 		KASSERT(sig >= 0, ("sig %d", sig));
1387 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1388 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1389 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1390 				error = 0;
1391 				break;
1392 			}
1393 		}
1394 
1395 		if (error != 0)
1396 			break;
1397 
1398 		/*
1399 		 * POSIX says this must be checked after looking for pending
1400 		 * signals.
1401 		 */
1402 		if (timeout != NULL && !timevalid) {
1403 			error = EINVAL;
1404 			break;
1405 		}
1406 
1407 		if (traced) {
1408 			error = EINTR;
1409 			break;
1410 		}
1411 
1412 		error = msleep_sbt(&p->p_sigacts, &p->p_mtx, PPAUSE | PCATCH,
1413 		    "sigwait", sbt, precision, C_ABSOLUTE);
1414 
1415 		/* The syscalls can not be restarted. */
1416 		if (error == ERESTART)
1417 			error = EINTR;
1418 
1419 		/*
1420 		 * If PTRACE_SCE or PTRACE_SCX were set after
1421 		 * userspace entered the syscall, return spurious
1422 		 * EINTR after wait was done.  Only do this as last
1423 		 * resort after rechecking for possible queued signals
1424 		 * and expired timeouts.
1425 		 */
1426 		if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1427 			traced = true;
1428 	}
1429 	thread_lock(td);
1430 	td->td_flags &= ~TDF_SIGWAIT;
1431 	thread_unlock(td);
1432 
1433 	new_block = saved_mask;
1434 	SIGSETNAND(new_block, td->td_sigmask);
1435 	td->td_sigmask = saved_mask;
1436 	/*
1437 	 * Fewer signals can be delivered to us, reschedule signal
1438 	 * notification.
1439 	 */
1440 	if (p->p_numthreads != 1)
1441 		reschedule_signals(p, new_block, 0);
1442 
1443 	if (error == 0) {
1444 		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1445 
1446 		if (ksi->ksi_code == SI_TIMER)
1447 			itimer_accept(p, ksi->ksi_timerid, ksi);
1448 
1449 #ifdef KTRACE
1450 		if (KTRPOINT(td, KTR_PSIG)) {
1451 			sig_t action;
1452 
1453 			mtx_lock(&ps->ps_mtx);
1454 			action = ps->ps_sigact[_SIG_IDX(sig)];
1455 			mtx_unlock(&ps->ps_mtx);
1456 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1457 		}
1458 #endif
1459 		if (sig == SIGKILL) {
1460 			proc_td_siginfo_capture(td, &ksi->ksi_info);
1461 			sigexit(td, sig);
1462 		}
1463 	}
1464 	PROC_UNLOCK(p);
1465 	return (error);
1466 }
1467 
1468 #ifndef _SYS_SYSPROTO_H_
1469 struct sigpending_args {
1470 	sigset_t	*set;
1471 };
1472 #endif
1473 int
sys_sigpending(struct thread * td,struct sigpending_args * uap)1474 sys_sigpending(struct thread *td, struct sigpending_args *uap)
1475 {
1476 	struct proc *p = td->td_proc;
1477 	sigset_t pending;
1478 
1479 	PROC_LOCK(p);
1480 	pending = p->p_sigqueue.sq_signals;
1481 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1482 	PROC_UNLOCK(p);
1483 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1484 }
1485 
1486 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1487 #ifndef _SYS_SYSPROTO_H_
1488 struct osigpending_args {
1489 	int	dummy;
1490 };
1491 #endif
1492 int
osigpending(struct thread * td,struct osigpending_args * uap)1493 osigpending(struct thread *td, struct osigpending_args *uap)
1494 {
1495 	struct proc *p = td->td_proc;
1496 	sigset_t pending;
1497 
1498 	PROC_LOCK(p);
1499 	pending = p->p_sigqueue.sq_signals;
1500 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1501 	PROC_UNLOCK(p);
1502 	SIG2OSIG(pending, td->td_retval[0]);
1503 	return (0);
1504 }
1505 #endif /* COMPAT_43 */
1506 
1507 #if defined(COMPAT_43)
1508 /*
1509  * Generalized interface signal handler, 4.3-compatible.
1510  */
1511 #ifndef _SYS_SYSPROTO_H_
1512 struct osigvec_args {
1513 	int	signum;
1514 	struct	sigvec *nsv;
1515 	struct	sigvec *osv;
1516 };
1517 #endif
1518 /* ARGSUSED */
1519 int
osigvec(struct thread * td,struct osigvec_args * uap)1520 osigvec(struct thread *td, struct osigvec_args *uap)
1521 {
1522 	struct sigvec vec;
1523 	struct sigaction nsa, osa;
1524 	struct sigaction *nsap, *osap;
1525 	int error;
1526 
1527 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1528 		return (EINVAL);
1529 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1530 	osap = (uap->osv != NULL) ? &osa : NULL;
1531 	if (nsap) {
1532 		error = copyin(uap->nsv, &vec, sizeof(vec));
1533 		if (error)
1534 			return (error);
1535 		nsap->sa_handler = vec.sv_handler;
1536 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1537 		nsap->sa_flags = vec.sv_flags;
1538 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1539 	}
1540 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1541 	if (osap && !error) {
1542 		vec.sv_handler = osap->sa_handler;
1543 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1544 		vec.sv_flags = osap->sa_flags;
1545 		vec.sv_flags &= ~SA_NOCLDWAIT;
1546 		vec.sv_flags ^= SA_RESTART;
1547 		error = copyout(&vec, uap->osv, sizeof(vec));
1548 	}
1549 	return (error);
1550 }
1551 
1552 #ifndef _SYS_SYSPROTO_H_
1553 struct osigblock_args {
1554 	int	mask;
1555 };
1556 #endif
1557 int
osigblock(struct thread * td,struct osigblock_args * uap)1558 osigblock(struct thread *td, struct osigblock_args *uap)
1559 {
1560 	sigset_t set, oset;
1561 
1562 	OSIG2SIG(uap->mask, set);
1563 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1564 	SIG2OSIG(oset, td->td_retval[0]);
1565 	return (0);
1566 }
1567 
1568 #ifndef _SYS_SYSPROTO_H_
1569 struct osigsetmask_args {
1570 	int	mask;
1571 };
1572 #endif
1573 int
osigsetmask(struct thread * td,struct osigsetmask_args * uap)1574 osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1575 {
1576 	sigset_t set, oset;
1577 
1578 	OSIG2SIG(uap->mask, set);
1579 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1580 	SIG2OSIG(oset, td->td_retval[0]);
1581 	return (0);
1582 }
1583 #endif /* COMPAT_43 */
1584 
1585 /*
1586  * Suspend calling thread until signal, providing mask to be set in the
1587  * meantime.
1588  */
1589 #ifndef _SYS_SYSPROTO_H_
1590 struct sigsuspend_args {
1591 	const sigset_t *sigmask;
1592 };
1593 #endif
1594 /* ARGSUSED */
1595 int
sys_sigsuspend(struct thread * td,struct sigsuspend_args * uap)1596 sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1597 {
1598 	sigset_t mask;
1599 	int error;
1600 
1601 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1602 	if (error)
1603 		return (error);
1604 	return (kern_sigsuspend(td, mask));
1605 }
1606 
1607 int
kern_sigsuspend(struct thread * td,sigset_t mask)1608 kern_sigsuspend(struct thread *td, sigset_t mask)
1609 {
1610 	struct proc *p = td->td_proc;
1611 	int has_sig, sig;
1612 
1613 	/* Ensure the sigfastblock value is up to date. */
1614 	sigfastblock_fetch(td);
1615 
1616 	/*
1617 	 * When returning from sigsuspend, we want
1618 	 * the old mask to be restored after the
1619 	 * signal handler has finished.  Thus, we
1620 	 * save it here and mark the sigacts structure
1621 	 * to indicate this.
1622 	 */
1623 	PROC_LOCK(p);
1624 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1625 	    SIGPROCMASK_PROC_LOCKED);
1626 	td->td_pflags |= TDP_OLDMASK;
1627 	ast_sched(td, TDA_SIGSUSPEND);
1628 
1629 	/*
1630 	 * Process signals now. Otherwise, we can get spurious wakeup
1631 	 * due to signal entered process queue, but delivered to other
1632 	 * thread. But sigsuspend should return only on signal
1633 	 * delivery.
1634 	 */
1635 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1636 	for (has_sig = 0; !has_sig;) {
1637 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1638 			0) == 0)
1639 			/* void */;
1640 		thread_suspend_check(0);
1641 		mtx_lock(&p->p_sigacts->ps_mtx);
1642 		while ((sig = cursig(td)) != 0) {
1643 			KASSERT(sig >= 0, ("sig %d", sig));
1644 			has_sig += postsig(sig);
1645 		}
1646 		mtx_unlock(&p->p_sigacts->ps_mtx);
1647 
1648 		/*
1649 		 * If PTRACE_SCE or PTRACE_SCX were set after
1650 		 * userspace entered the syscall, return spurious
1651 		 * EINTR.
1652 		 */
1653 		if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1654 			has_sig += 1;
1655 	}
1656 	PROC_UNLOCK(p);
1657 	td->td_errno = EINTR;
1658 	td->td_pflags |= TDP_NERRNO;
1659 	return (EJUSTRETURN);
1660 }
1661 
1662 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1663 /*
1664  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1665  * convention: libc stub passes mask, not pointer, to save a copyin.
1666  */
1667 #ifndef _SYS_SYSPROTO_H_
1668 struct osigsuspend_args {
1669 	osigset_t mask;
1670 };
1671 #endif
1672 /* ARGSUSED */
1673 int
osigsuspend(struct thread * td,struct osigsuspend_args * uap)1674 osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1675 {
1676 	sigset_t mask;
1677 
1678 	OSIG2SIG(uap->mask, mask);
1679 	return (kern_sigsuspend(td, mask));
1680 }
1681 #endif /* COMPAT_43 */
1682 
1683 #if defined(COMPAT_43)
1684 #ifndef _SYS_SYSPROTO_H_
1685 struct osigstack_args {
1686 	struct	sigstack *nss;
1687 	struct	sigstack *oss;
1688 };
1689 #endif
1690 /* ARGSUSED */
1691 int
osigstack(struct thread * td,struct osigstack_args * uap)1692 osigstack(struct thread *td, struct osigstack_args *uap)
1693 {
1694 	struct sigstack nss, oss;
1695 	int error = 0;
1696 
1697 	if (uap->nss != NULL) {
1698 		error = copyin(uap->nss, &nss, sizeof(nss));
1699 		if (error)
1700 			return (error);
1701 	}
1702 	oss.ss_sp = td->td_sigstk.ss_sp;
1703 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1704 	if (uap->nss != NULL) {
1705 		td->td_sigstk.ss_sp = nss.ss_sp;
1706 		td->td_sigstk.ss_size = 0;
1707 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1708 		td->td_pflags |= TDP_ALTSTACK;
1709 	}
1710 	if (uap->oss != NULL)
1711 		error = copyout(&oss, uap->oss, sizeof(oss));
1712 
1713 	return (error);
1714 }
1715 #endif /* COMPAT_43 */
1716 
1717 #ifndef _SYS_SYSPROTO_H_
1718 struct sigaltstack_args {
1719 	stack_t	*ss;
1720 	stack_t	*oss;
1721 };
1722 #endif
1723 /* ARGSUSED */
1724 int
sys_sigaltstack(struct thread * td,struct sigaltstack_args * uap)1725 sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1726 {
1727 	stack_t ss, oss;
1728 	int error;
1729 
1730 	if (uap->ss != NULL) {
1731 		error = copyin(uap->ss, &ss, sizeof(ss));
1732 		if (error)
1733 			return (error);
1734 	}
1735 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1736 	    (uap->oss != NULL) ? &oss : NULL);
1737 	if (error)
1738 		return (error);
1739 	if (uap->oss != NULL)
1740 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1741 	return (error);
1742 }
1743 
1744 int
kern_sigaltstack(struct thread * td,stack_t * ss,stack_t * oss)1745 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1746 {
1747 	struct proc *p = td->td_proc;
1748 	int oonstack;
1749 
1750 	oonstack = sigonstack(cpu_getstack(td));
1751 
1752 	if (oss != NULL) {
1753 		*oss = td->td_sigstk;
1754 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1755 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1756 	}
1757 
1758 	if (ss != NULL) {
1759 		if (oonstack)
1760 			return (EPERM);
1761 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1762 			return (EINVAL);
1763 		if (!(ss->ss_flags & SS_DISABLE)) {
1764 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1765 				return (ENOMEM);
1766 
1767 			td->td_sigstk = *ss;
1768 			td->td_pflags |= TDP_ALTSTACK;
1769 		} else {
1770 			td->td_pflags &= ~TDP_ALTSTACK;
1771 		}
1772 	}
1773 	return (0);
1774 }
1775 
1776 struct killpg1_ctx {
1777 	struct thread *td;
1778 	ksiginfo_t *ksi;
1779 	int sig;
1780 	bool sent;
1781 	bool found;
1782 	int ret;
1783 };
1784 
1785 static void
killpg1_sendsig_locked(struct proc * p,struct killpg1_ctx * arg)1786 killpg1_sendsig_locked(struct proc *p, struct killpg1_ctx *arg)
1787 {
1788 	int err;
1789 
1790 	err = p_cansignal(arg->td, p, arg->sig);
1791 	if (err == 0 && arg->sig != 0)
1792 		pksignal(p, arg->sig, arg->ksi);
1793 	if (err != ESRCH)
1794 		arg->found = true;
1795 	if (err == 0)
1796 		arg->sent = true;
1797 	else if (arg->ret == 0 && err != ESRCH && err != EPERM)
1798 		arg->ret = err;
1799 }
1800 
1801 static void
killpg1_sendsig(struct proc * p,bool notself,struct killpg1_ctx * arg)1802 killpg1_sendsig(struct proc *p, bool notself, struct killpg1_ctx *arg)
1803 {
1804 
1805 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1806 	    (notself && p == arg->td->td_proc) || p->p_state == PRS_NEW)
1807 		return;
1808 
1809 	PROC_LOCK(p);
1810 	killpg1_sendsig_locked(p, arg);
1811 	PROC_UNLOCK(p);
1812 }
1813 
1814 static void
kill_processes_prison_cb(struct proc * p,void * arg)1815 kill_processes_prison_cb(struct proc *p, void *arg)
1816 {
1817 	struct killpg1_ctx *ctx = arg;
1818 
1819 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1820 	    (p == ctx->td->td_proc) || p->p_state == PRS_NEW)
1821 		return;
1822 
1823 	killpg1_sendsig_locked(p, ctx);
1824 }
1825 
1826 /*
1827  * Common code for kill process group/broadcast kill.
1828  * td is the calling thread, as usual.
1829  */
1830 static int
killpg1(struct thread * td,int sig,int pgid,int all,ksiginfo_t * ksi)1831 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1832 {
1833 	struct proc *p;
1834 	struct pgrp *pgrp;
1835 	struct killpg1_ctx arg;
1836 
1837 	arg.td = td;
1838 	arg.ksi = ksi;
1839 	arg.sig = sig;
1840 	arg.sent = false;
1841 	arg.found = false;
1842 	arg.ret = 0;
1843 	if (all) {
1844 		/*
1845 		 * broadcast
1846 		 */
1847 		prison_proc_iterate(td->td_ucred->cr_prison,
1848 		    kill_processes_prison_cb, &arg);
1849 	} else {
1850 again:
1851 		sx_slock(&proctree_lock);
1852 		if (pgid == 0) {
1853 			/*
1854 			 * zero pgid means send to my process group.
1855 			 */
1856 			pgrp = td->td_proc->p_pgrp;
1857 			PGRP_LOCK(pgrp);
1858 		} else {
1859 			pgrp = pgfind(pgid);
1860 			if (pgrp == NULL) {
1861 				sx_sunlock(&proctree_lock);
1862 				return (ESRCH);
1863 			}
1864 		}
1865 		sx_sunlock(&proctree_lock);
1866 		if (!sx_try_xlock(&pgrp->pg_killsx)) {
1867 			PGRP_UNLOCK(pgrp);
1868 			sx_xlock(&pgrp->pg_killsx);
1869 			sx_xunlock(&pgrp->pg_killsx);
1870 			goto again;
1871 		}
1872 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1873 			killpg1_sendsig(p, false, &arg);
1874 		}
1875 		PGRP_UNLOCK(pgrp);
1876 		sx_xunlock(&pgrp->pg_killsx);
1877 	}
1878 	MPASS(arg.ret != 0 || arg.found || !arg.sent);
1879 	if (arg.ret == 0 && !arg.sent)
1880 		arg.ret = arg.found ? EPERM : ESRCH;
1881 	return (arg.ret);
1882 }
1883 
1884 #ifndef _SYS_SYSPROTO_H_
1885 struct kill_args {
1886 	int	pid;
1887 	int	signum;
1888 };
1889 #endif
1890 /* ARGSUSED */
1891 int
sys_kill(struct thread * td,struct kill_args * uap)1892 sys_kill(struct thread *td, struct kill_args *uap)
1893 {
1894 
1895 	return (kern_kill(td, uap->pid, uap->signum));
1896 }
1897 
1898 int
kern_kill(struct thread * td,pid_t pid,int signum)1899 kern_kill(struct thread *td, pid_t pid, int signum)
1900 {
1901 	ksiginfo_t ksi;
1902 	struct proc *p;
1903 	int error;
1904 
1905 	/*
1906 	 * A process in capability mode can send signals only to himself.
1907 	 * The main rationale behind this is that abort(3) is implemented as
1908 	 * kill(getpid(), SIGABRT).
1909 	 */
1910 	if (pid != td->td_proc->p_pid) {
1911 		if (CAP_TRACING(td))
1912 			ktrcapfail(CAPFAIL_SIGNAL, &signum);
1913 		if (IN_CAPABILITY_MODE(td))
1914 			return (ECAPMODE);
1915 	}
1916 
1917 	AUDIT_ARG_SIGNUM(signum);
1918 	AUDIT_ARG_PID(pid);
1919 	if ((u_int)signum > _SIG_MAXSIG)
1920 		return (EINVAL);
1921 
1922 	ksiginfo_init(&ksi);
1923 	ksi.ksi_signo = signum;
1924 	ksi.ksi_code = SI_USER;
1925 	ksi.ksi_pid = td->td_proc->p_pid;
1926 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1927 
1928 	if (pid > 0) {
1929 		/* kill single process */
1930 		if ((p = pfind_any(pid)) == NULL)
1931 			return (ESRCH);
1932 		AUDIT_ARG_PROCESS(p);
1933 		error = p_cansignal(td, p, signum);
1934 		if (error == 0 && signum)
1935 			pksignal(p, signum, &ksi);
1936 		PROC_UNLOCK(p);
1937 		return (error);
1938 	}
1939 	switch (pid) {
1940 	case -1:		/* broadcast signal */
1941 		return (killpg1(td, signum, 0, 1, &ksi));
1942 	case 0:			/* signal own process group */
1943 		return (killpg1(td, signum, 0, 0, &ksi));
1944 	default:		/* negative explicit process group */
1945 		return (killpg1(td, signum, -pid, 0, &ksi));
1946 	}
1947 	/* NOTREACHED */
1948 }
1949 
1950 int
sys_pdkill(struct thread * td,struct pdkill_args * uap)1951 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1952 {
1953 	struct proc *p;
1954 	int error;
1955 
1956 	AUDIT_ARG_SIGNUM(uap->signum);
1957 	AUDIT_ARG_FD(uap->fd);
1958 	if ((u_int)uap->signum > _SIG_MAXSIG)
1959 		return (EINVAL);
1960 
1961 	error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1962 	if (error)
1963 		return (error);
1964 	AUDIT_ARG_PROCESS(p);
1965 	error = p_cansignal(td, p, uap->signum);
1966 	if (error == 0 && uap->signum)
1967 		kern_psignal(p, uap->signum);
1968 	PROC_UNLOCK(p);
1969 	return (error);
1970 }
1971 
1972 #if defined(COMPAT_43)
1973 #ifndef _SYS_SYSPROTO_H_
1974 struct okillpg_args {
1975 	int	pgid;
1976 	int	signum;
1977 };
1978 #endif
1979 /* ARGSUSED */
1980 int
okillpg(struct thread * td,struct okillpg_args * uap)1981 okillpg(struct thread *td, struct okillpg_args *uap)
1982 {
1983 	ksiginfo_t ksi;
1984 
1985 	AUDIT_ARG_SIGNUM(uap->signum);
1986 	AUDIT_ARG_PID(uap->pgid);
1987 	if ((u_int)uap->signum > _SIG_MAXSIG)
1988 		return (EINVAL);
1989 
1990 	ksiginfo_init(&ksi);
1991 	ksi.ksi_signo = uap->signum;
1992 	ksi.ksi_code = SI_USER;
1993 	ksi.ksi_pid = td->td_proc->p_pid;
1994 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1995 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1996 }
1997 #endif /* COMPAT_43 */
1998 
1999 #ifndef _SYS_SYSPROTO_H_
2000 struct sigqueue_args {
2001 	pid_t pid;
2002 	int signum;
2003 	/* union sigval */ void *value;
2004 };
2005 #endif
2006 int
sys_sigqueue(struct thread * td,struct sigqueue_args * uap)2007 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
2008 {
2009 	union sigval sv;
2010 
2011 	sv.sival_ptr = uap->value;
2012 
2013 	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
2014 }
2015 
2016 int
kern_sigqueue(struct thread * td,pid_t pid,int signumf,union sigval * value)2017 kern_sigqueue(struct thread *td, pid_t pid, int signumf, union sigval *value)
2018 {
2019 	ksiginfo_t ksi;
2020 	struct proc *p;
2021 	struct thread *td2;
2022 	u_int signum;
2023 	int error;
2024 
2025 	signum = signumf & ~__SIGQUEUE_TID;
2026 	if (signum > _SIG_MAXSIG)
2027 		return (EINVAL);
2028 
2029 	/*
2030 	 * Specification says sigqueue can only send signal to
2031 	 * single process.
2032 	 */
2033 	if (pid <= 0)
2034 		return (EINVAL);
2035 
2036 	if ((signumf & __SIGQUEUE_TID) == 0) {
2037 		if ((p = pfind_any(pid)) == NULL)
2038 			return (ESRCH);
2039 		td2 = NULL;
2040 	} else {
2041 		p = td->td_proc;
2042 		td2 = tdfind((lwpid_t)pid, p->p_pid);
2043 		if (td2 == NULL)
2044 			return (ESRCH);
2045 	}
2046 
2047 	error = p_cansignal(td, p, signum);
2048 	if (error == 0 && signum != 0) {
2049 		ksiginfo_init(&ksi);
2050 		ksi.ksi_flags = KSI_SIGQ;
2051 		ksi.ksi_signo = signum;
2052 		ksi.ksi_code = SI_QUEUE;
2053 		ksi.ksi_pid = td->td_proc->p_pid;
2054 		ksi.ksi_uid = td->td_ucred->cr_ruid;
2055 		ksi.ksi_value = *value;
2056 		error = tdsendsignal(p, td2, ksi.ksi_signo, &ksi);
2057 	}
2058 	PROC_UNLOCK(p);
2059 	return (error);
2060 }
2061 
2062 /*
2063  * Send a signal to a process group.  If checktty is 1,
2064  * limit to members which have a controlling terminal.
2065  */
2066 void
pgsignal(struct pgrp * pgrp,int sig,int checkctty,ksiginfo_t * ksi)2067 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
2068 {
2069 	struct proc *p;
2070 
2071 	if (pgrp) {
2072 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
2073 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
2074 			PROC_LOCK(p);
2075 			if (p->p_state == PRS_NORMAL &&
2076 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
2077 				pksignal(p, sig, ksi);
2078 			PROC_UNLOCK(p);
2079 		}
2080 	}
2081 }
2082 
2083 /*
2084  * Recalculate the signal mask and reset the signal disposition after
2085  * usermode frame for delivery is formed.  Should be called after
2086  * mach-specific routine, because sysent->sv_sendsig() needs correct
2087  * ps_siginfo and signal mask.
2088  */
2089 static void
postsig_done(int sig,struct thread * td,struct sigacts * ps)2090 postsig_done(int sig, struct thread *td, struct sigacts *ps)
2091 {
2092 	sigset_t mask;
2093 
2094 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2095 	td->td_ru.ru_nsignals++;
2096 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
2097 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
2098 		SIGADDSET(mask, sig);
2099 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
2100 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
2101 	if (SIGISMEMBER(ps->ps_sigreset, sig))
2102 		sigdflt(ps, sig);
2103 }
2104 
2105 /*
2106  * Send a signal caused by a trap to the current thread.  If it will be
2107  * caught immediately, deliver it with correct code.  Otherwise, post it
2108  * normally.
2109  */
2110 void
trapsignal(struct thread * td,ksiginfo_t * ksi)2111 trapsignal(struct thread *td, ksiginfo_t *ksi)
2112 {
2113 	struct sigacts *ps;
2114 	struct proc *p;
2115 	sigset_t sigmask;
2116 	int sig;
2117 
2118 	p = td->td_proc;
2119 	sig = ksi->ksi_signo;
2120 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
2121 
2122 	sigfastblock_fetch(td);
2123 	PROC_LOCK(p);
2124 	ps = p->p_sigacts;
2125 	mtx_lock(&ps->ps_mtx);
2126 	sigmask = td->td_sigmask;
2127 	if (td->td_sigblock_val != 0)
2128 		SIGSETOR(sigmask, fastblock_mask);
2129 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2130 	    !SIGISMEMBER(sigmask, sig)) {
2131 #ifdef KTRACE
2132 		if (KTRPOINT(curthread, KTR_PSIG))
2133 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2134 			    &td->td_sigmask, ksi->ksi_code);
2135 #endif
2136 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2137 		    ksi, &td->td_sigmask);
2138 		postsig_done(sig, td, ps);
2139 		mtx_unlock(&ps->ps_mtx);
2140 	} else {
2141 		/*
2142 		 * Avoid a possible infinite loop if the thread
2143 		 * masking the signal or process is ignoring the
2144 		 * signal.
2145 		 */
2146 		if (kern_forcesigexit && (SIGISMEMBER(sigmask, sig) ||
2147 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2148 			SIGDELSET(td->td_sigmask, sig);
2149 			SIGDELSET(ps->ps_sigcatch, sig);
2150 			SIGDELSET(ps->ps_sigignore, sig);
2151 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2152 			td->td_pflags &= ~TDP_SIGFASTBLOCK;
2153 			td->td_sigblock_val = 0;
2154 		}
2155 		mtx_unlock(&ps->ps_mtx);
2156 		p->p_sig = sig;		/* XXX to verify code */
2157 		tdsendsignal(p, td, sig, ksi);
2158 	}
2159 	PROC_UNLOCK(p);
2160 }
2161 
2162 static struct thread *
sigtd(struct proc * p,int sig,bool fast_sigblock)2163 sigtd(struct proc *p, int sig, bool fast_sigblock)
2164 {
2165 	struct thread *td, *signal_td;
2166 
2167 	PROC_LOCK_ASSERT(p, MA_OWNED);
2168 	MPASS(!fast_sigblock || p == curproc);
2169 
2170 	/*
2171 	 * Check if current thread can handle the signal without
2172 	 * switching context to another thread.
2173 	 */
2174 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig) &&
2175 	    (!fast_sigblock || curthread->td_sigblock_val == 0))
2176 		return (curthread);
2177 
2178 	/* Find a non-stopped thread that does not mask the signal. */
2179 	signal_td = NULL;
2180 	FOREACH_THREAD_IN_PROC(p, td) {
2181 		if (!SIGISMEMBER(td->td_sigmask, sig) && (!fast_sigblock ||
2182 		    td != curthread || td->td_sigblock_val == 0) &&
2183 		    (td->td_flags & TDF_BOUNDARY) == 0) {
2184 			signal_td = td;
2185 			break;
2186 		}
2187 	}
2188 	/* Select random (first) thread if no better match was found. */
2189 	if (signal_td == NULL)
2190 		signal_td = FIRST_THREAD_IN_PROC(p);
2191 	return (signal_td);
2192 }
2193 
2194 /*
2195  * Send the signal to the process.  If the signal has an action, the action
2196  * is usually performed by the target process rather than the caller; we add
2197  * the signal to the set of pending signals for the process.
2198  *
2199  * Exceptions:
2200  *   o When a stop signal is sent to a sleeping process that takes the
2201  *     default action, the process is stopped without awakening it.
2202  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2203  *     regardless of the signal action (eg, blocked or ignored).
2204  *
2205  * Other ignored signals are discarded immediately.
2206  *
2207  * NB: This function may be entered from the debugger via the "kill" DDB
2208  * command.  There is little that can be done to mitigate the possibly messy
2209  * side effects of this unwise possibility.
2210  */
2211 void
kern_psignal(struct proc * p,int sig)2212 kern_psignal(struct proc *p, int sig)
2213 {
2214 	ksiginfo_t ksi;
2215 
2216 	ksiginfo_init(&ksi);
2217 	ksi.ksi_signo = sig;
2218 	ksi.ksi_code = SI_KERNEL;
2219 	(void) tdsendsignal(p, NULL, sig, &ksi);
2220 }
2221 
2222 int
pksignal(struct proc * p,int sig,ksiginfo_t * ksi)2223 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2224 {
2225 
2226 	return (tdsendsignal(p, NULL, sig, ksi));
2227 }
2228 
2229 /* Utility function for finding a thread to send signal event to. */
2230 int
sigev_findtd(struct proc * p,struct sigevent * sigev,struct thread ** ttd)2231 sigev_findtd(struct proc *p, struct sigevent *sigev, struct thread **ttd)
2232 {
2233 	struct thread *td;
2234 
2235 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2236 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2237 		if (td == NULL)
2238 			return (ESRCH);
2239 		*ttd = td;
2240 	} else {
2241 		*ttd = NULL;
2242 		PROC_LOCK(p);
2243 	}
2244 	return (0);
2245 }
2246 
2247 void
tdsignal(struct thread * td,int sig)2248 tdsignal(struct thread *td, int sig)
2249 {
2250 	ksiginfo_t ksi;
2251 
2252 	ksiginfo_init(&ksi);
2253 	ksi.ksi_signo = sig;
2254 	ksi.ksi_code = SI_KERNEL;
2255 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2256 }
2257 
2258 void
tdksignal(struct thread * td,int sig,ksiginfo_t * ksi)2259 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2260 {
2261 
2262 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2263 }
2264 
2265 static int
sig_sleepq_abort(struct thread * td,int intrval)2266 sig_sleepq_abort(struct thread *td, int intrval)
2267 {
2268 	THREAD_LOCK_ASSERT(td, MA_OWNED);
2269 
2270 	if (intrval == 0 && (td->td_flags & TDF_SIGWAIT) == 0) {
2271 		thread_unlock(td);
2272 		return (0);
2273 	}
2274 	return (sleepq_abort(td, intrval));
2275 }
2276 
2277 int
tdsendsignal(struct proc * p,struct thread * td,int sig,ksiginfo_t * ksi)2278 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2279 {
2280 	sig_t action;
2281 	sigqueue_t *sigqueue;
2282 	int prop;
2283 	struct sigacts *ps;
2284 	int intrval;
2285 	int ret = 0;
2286 	int wakeup_swapper;
2287 
2288 	MPASS(td == NULL || p == td->td_proc);
2289 	PROC_LOCK_ASSERT(p, MA_OWNED);
2290 
2291 	if (!_SIG_VALID(sig))
2292 		panic("%s(): invalid signal %d", __func__, sig);
2293 
2294 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2295 
2296 	/*
2297 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2298 	 */
2299 	if (p->p_state == PRS_ZOMBIE) {
2300 		if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2301 			ksiginfo_tryfree(ksi);
2302 		return (ret);
2303 	}
2304 
2305 	ps = p->p_sigacts;
2306 	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2307 	prop = sigprop(sig);
2308 
2309 	if (td == NULL) {
2310 		td = sigtd(p, sig, false);
2311 		sigqueue = &p->p_sigqueue;
2312 	} else
2313 		sigqueue = &td->td_sigqueue;
2314 
2315 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2316 
2317 	/*
2318 	 * If the signal is being ignored, then we forget about it
2319 	 * immediately, except when the target process executes
2320 	 * sigwait().  (Note: we don't set SIGCONT in ps_sigignore,
2321 	 * and if it is set to SIG_IGN, action will be SIG_DFL here.)
2322 	 */
2323 	mtx_lock(&ps->ps_mtx);
2324 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2325 		if (kern_sig_discard_ign &&
2326 		    (p->p_sysent->sv_flags & SV_SIG_DISCIGN) == 0) {
2327 			SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2328 
2329 			mtx_unlock(&ps->ps_mtx);
2330 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2331 				ksiginfo_tryfree(ksi);
2332 			return (ret);
2333 		} else {
2334 			action = SIG_CATCH;
2335 			intrval = 0;
2336 		}
2337 	} else {
2338 		if (SIGISMEMBER(td->td_sigmask, sig))
2339 			action = SIG_HOLD;
2340 		else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2341 			action = SIG_CATCH;
2342 		else
2343 			action = SIG_DFL;
2344 		if (SIGISMEMBER(ps->ps_sigintr, sig))
2345 			intrval = EINTR;
2346 		else
2347 			intrval = ERESTART;
2348 	}
2349 	mtx_unlock(&ps->ps_mtx);
2350 
2351 	if (prop & SIGPROP_CONT)
2352 		sigqueue_delete_stopmask_proc(p);
2353 	else if (prop & SIGPROP_STOP) {
2354 		/*
2355 		 * If sending a tty stop signal to a member of an orphaned
2356 		 * process group, discard the signal here if the action
2357 		 * is default; don't stop the process below if sleeping,
2358 		 * and don't clear any pending SIGCONT.
2359 		 */
2360 		if ((prop & SIGPROP_TTYSTOP) != 0 &&
2361 		    (p->p_pgrp->pg_flags & PGRP_ORPHANED) != 0 &&
2362 		    action == SIG_DFL) {
2363 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2364 				ksiginfo_tryfree(ksi);
2365 			return (ret);
2366 		}
2367 		sigqueue_delete_proc(p, SIGCONT);
2368 		if (p->p_flag & P_CONTINUED) {
2369 			p->p_flag &= ~P_CONTINUED;
2370 			PROC_LOCK(p->p_pptr);
2371 			sigqueue_take(p->p_ksi);
2372 			PROC_UNLOCK(p->p_pptr);
2373 		}
2374 	}
2375 
2376 	ret = sigqueue_add(sigqueue, sig, ksi);
2377 	if (ret != 0)
2378 		return (ret);
2379 	signotify(td);
2380 	/*
2381 	 * Defer further processing for signals which are held,
2382 	 * except that stopped processes must be continued by SIGCONT.
2383 	 */
2384 	if (action == SIG_HOLD &&
2385 	    !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2386 		return (ret);
2387 
2388 	wakeup_swapper = 0;
2389 
2390 	/*
2391 	 * Some signals have a process-wide effect and a per-thread
2392 	 * component.  Most processing occurs when the process next
2393 	 * tries to cross the user boundary, however there are some
2394 	 * times when processing needs to be done immediately, such as
2395 	 * waking up threads so that they can cross the user boundary.
2396 	 * We try to do the per-process part here.
2397 	 */
2398 	if (P_SHOULDSTOP(p)) {
2399 		KASSERT(!(p->p_flag & P_WEXIT),
2400 		    ("signal to stopped but exiting process"));
2401 		if (sig == SIGKILL) {
2402 			/*
2403 			 * If traced process is already stopped,
2404 			 * then no further action is necessary.
2405 			 */
2406 			if (p->p_flag & P_TRACED)
2407 				goto out;
2408 			/*
2409 			 * SIGKILL sets process running.
2410 			 * It will die elsewhere.
2411 			 * All threads must be restarted.
2412 			 */
2413 			p->p_flag &= ~P_STOPPED_SIG;
2414 			goto runfast;
2415 		}
2416 
2417 		if (prop & SIGPROP_CONT) {
2418 			/*
2419 			 * If traced process is already stopped,
2420 			 * then no further action is necessary.
2421 			 */
2422 			if (p->p_flag & P_TRACED)
2423 				goto out;
2424 			/*
2425 			 * If SIGCONT is default (or ignored), we continue the
2426 			 * process but don't leave the signal in sigqueue as
2427 			 * it has no further action.  If SIGCONT is held, we
2428 			 * continue the process and leave the signal in
2429 			 * sigqueue.  If the process catches SIGCONT, let it
2430 			 * handle the signal itself.  If it isn't waiting on
2431 			 * an event, it goes back to run state.
2432 			 * Otherwise, process goes back to sleep state.
2433 			 */
2434 			p->p_flag &= ~P_STOPPED_SIG;
2435 			PROC_SLOCK(p);
2436 			if (p->p_numthreads == p->p_suspcount) {
2437 				PROC_SUNLOCK(p);
2438 				p->p_flag |= P_CONTINUED;
2439 				p->p_xsig = SIGCONT;
2440 				PROC_LOCK(p->p_pptr);
2441 				childproc_continued(p);
2442 				PROC_UNLOCK(p->p_pptr);
2443 				PROC_SLOCK(p);
2444 			}
2445 			if (action == SIG_DFL) {
2446 				thread_unsuspend(p);
2447 				PROC_SUNLOCK(p);
2448 				sigqueue_delete(sigqueue, sig);
2449 				goto out_cont;
2450 			}
2451 			if (action == SIG_CATCH) {
2452 				/*
2453 				 * The process wants to catch it so it needs
2454 				 * to run at least one thread, but which one?
2455 				 */
2456 				PROC_SUNLOCK(p);
2457 				goto runfast;
2458 			}
2459 			/*
2460 			 * The signal is not ignored or caught.
2461 			 */
2462 			thread_unsuspend(p);
2463 			PROC_SUNLOCK(p);
2464 			goto out_cont;
2465 		}
2466 
2467 		if (prop & SIGPROP_STOP) {
2468 			/*
2469 			 * If traced process is already stopped,
2470 			 * then no further action is necessary.
2471 			 */
2472 			if (p->p_flag & P_TRACED)
2473 				goto out;
2474 			/*
2475 			 * Already stopped, don't need to stop again
2476 			 * (If we did the shell could get confused).
2477 			 * Just make sure the signal STOP bit set.
2478 			 */
2479 			p->p_flag |= P_STOPPED_SIG;
2480 			sigqueue_delete(sigqueue, sig);
2481 			goto out;
2482 		}
2483 
2484 		/*
2485 		 * All other kinds of signals:
2486 		 * If a thread is sleeping interruptibly, simulate a
2487 		 * wakeup so that when it is continued it will be made
2488 		 * runnable and can look at the signal.  However, don't make
2489 		 * the PROCESS runnable, leave it stopped.
2490 		 * It may run a bit until it hits a thread_suspend_check().
2491 		 */
2492 		PROC_SLOCK(p);
2493 		thread_lock(td);
2494 		if (TD_CAN_ABORT(td))
2495 			wakeup_swapper = sig_sleepq_abort(td, intrval);
2496 		else
2497 			thread_unlock(td);
2498 		PROC_SUNLOCK(p);
2499 		goto out;
2500 		/*
2501 		 * Mutexes are short lived. Threads waiting on them will
2502 		 * hit thread_suspend_check() soon.
2503 		 */
2504 	} else if (p->p_state == PRS_NORMAL) {
2505 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2506 			tdsigwakeup(td, sig, action, intrval);
2507 			goto out;
2508 		}
2509 
2510 		MPASS(action == SIG_DFL);
2511 
2512 		if (prop & SIGPROP_STOP) {
2513 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2514 				goto out;
2515 			p->p_flag |= P_STOPPED_SIG;
2516 			p->p_xsig = sig;
2517 			PROC_SLOCK(p);
2518 			wakeup_swapper = sig_suspend_threads(td, p);
2519 			if (p->p_numthreads == p->p_suspcount) {
2520 				/*
2521 				 * only thread sending signal to another
2522 				 * process can reach here, if thread is sending
2523 				 * signal to its process, because thread does
2524 				 * not suspend itself here, p_numthreads
2525 				 * should never be equal to p_suspcount.
2526 				 */
2527 				thread_stopped(p);
2528 				PROC_SUNLOCK(p);
2529 				sigqueue_delete_proc(p, p->p_xsig);
2530 			} else
2531 				PROC_SUNLOCK(p);
2532 			goto out;
2533 		}
2534 	} else {
2535 		/* Not in "NORMAL" state. discard the signal. */
2536 		sigqueue_delete(sigqueue, sig);
2537 		goto out;
2538 	}
2539 
2540 	/*
2541 	 * The process is not stopped so we need to apply the signal to all the
2542 	 * running threads.
2543 	 */
2544 runfast:
2545 	tdsigwakeup(td, sig, action, intrval);
2546 	PROC_SLOCK(p);
2547 	thread_unsuspend(p);
2548 	PROC_SUNLOCK(p);
2549 out_cont:
2550 	itimer_proc_continue(p);
2551 	kqtimer_proc_continue(p);
2552 out:
2553 	/* If we jump here, proc slock should not be owned. */
2554 	PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2555 	if (wakeup_swapper)
2556 		kick_proc0();
2557 
2558 	return (ret);
2559 }
2560 
2561 /*
2562  * The force of a signal has been directed against a single
2563  * thread.  We need to see what we can do about knocking it
2564  * out of any sleep it may be in etc.
2565  */
2566 static void
tdsigwakeup(struct thread * td,int sig,sig_t action,int intrval)2567 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2568 {
2569 	struct proc *p = td->td_proc;
2570 	int prop, wakeup_swapper;
2571 
2572 	PROC_LOCK_ASSERT(p, MA_OWNED);
2573 	prop = sigprop(sig);
2574 
2575 	PROC_SLOCK(p);
2576 	thread_lock(td);
2577 	/*
2578 	 * Bring the priority of a thread up if we want it to get
2579 	 * killed in this lifetime.  Be careful to avoid bumping the
2580 	 * priority of the idle thread, since we still allow to signal
2581 	 * kernel processes.
2582 	 */
2583 	if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2584 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2585 		sched_prio(td, PUSER);
2586 	if (TD_ON_SLEEPQ(td)) {
2587 		/*
2588 		 * If thread is sleeping uninterruptibly
2589 		 * we can't interrupt the sleep... the signal will
2590 		 * be noticed when the process returns through
2591 		 * trap() or syscall().
2592 		 */
2593 		if ((td->td_flags & TDF_SINTR) == 0)
2594 			goto out;
2595 		/*
2596 		 * If SIGCONT is default (or ignored) and process is
2597 		 * asleep, we are finished; the process should not
2598 		 * be awakened.
2599 		 */
2600 		if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2601 			thread_unlock(td);
2602 			PROC_SUNLOCK(p);
2603 			sigqueue_delete(&p->p_sigqueue, sig);
2604 			/*
2605 			 * It may be on either list in this state.
2606 			 * Remove from both for now.
2607 			 */
2608 			sigqueue_delete(&td->td_sigqueue, sig);
2609 			return;
2610 		}
2611 
2612 		/*
2613 		 * Don't awaken a sleeping thread for SIGSTOP if the
2614 		 * STOP signal is deferred.
2615 		 */
2616 		if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2617 		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2618 			goto out;
2619 
2620 		/*
2621 		 * Give low priority threads a better chance to run.
2622 		 */
2623 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2624 			sched_prio(td, PUSER);
2625 
2626 		wakeup_swapper = sig_sleepq_abort(td, intrval);
2627 		PROC_SUNLOCK(p);
2628 		if (wakeup_swapper)
2629 			kick_proc0();
2630 		return;
2631 	}
2632 
2633 	/*
2634 	 * Other states do nothing with the signal immediately,
2635 	 * other than kicking ourselves if we are running.
2636 	 * It will either never be noticed, or noticed very soon.
2637 	 */
2638 #ifdef SMP
2639 	if (TD_IS_RUNNING(td) && td != curthread)
2640 		forward_signal(td);
2641 #endif
2642 
2643 out:
2644 	PROC_SUNLOCK(p);
2645 	thread_unlock(td);
2646 }
2647 
2648 static void
ptrace_coredumpreq(struct thread * td,struct proc * p,struct thr_coredump_req * tcq)2649 ptrace_coredumpreq(struct thread *td, struct proc *p,
2650     struct thr_coredump_req *tcq)
2651 {
2652 	void *rl_cookie;
2653 
2654 	if (p->p_sysent->sv_coredump == NULL) {
2655 		tcq->tc_error = ENOSYS;
2656 		return;
2657 	}
2658 
2659 	rl_cookie = vn_rangelock_wlock(tcq->tc_vp, 0, OFF_MAX);
2660 	tcq->tc_error = p->p_sysent->sv_coredump(td, tcq->tc_vp,
2661 	    tcq->tc_limit, tcq->tc_flags);
2662 	vn_rangelock_unlock(tcq->tc_vp, rl_cookie);
2663 }
2664 
2665 static void
ptrace_syscallreq(struct thread * td,struct proc * p,struct thr_syscall_req * tsr)2666 ptrace_syscallreq(struct thread *td, struct proc *p,
2667     struct thr_syscall_req *tsr)
2668 {
2669 	struct sysentvec *sv;
2670 	struct sysent *se;
2671 	register_t rv_saved[2];
2672 	int error, nerror;
2673 	int sc;
2674 	bool audited, sy_thr_static;
2675 
2676 	sv = p->p_sysent;
2677 	if (sv->sv_table == NULL || sv->sv_size < tsr->ts_sa.code) {
2678 		tsr->ts_ret.sr_error = ENOSYS;
2679 		return;
2680 	}
2681 
2682 	sc = tsr->ts_sa.code;
2683 	if (sc == SYS_syscall || sc == SYS___syscall) {
2684 		sc = tsr->ts_sa.args[0];
2685 		memmove(&tsr->ts_sa.args[0], &tsr->ts_sa.args[1],
2686 		    sizeof(register_t) * (tsr->ts_nargs - 1));
2687 	}
2688 
2689 	tsr->ts_sa.callp = se = &sv->sv_table[sc];
2690 
2691 	VM_CNT_INC(v_syscall);
2692 	td->td_pticks = 0;
2693 	if (__predict_false(td->td_cowgen != atomic_load_int(
2694 	    &td->td_proc->p_cowgen)))
2695 		thread_cow_update(td);
2696 
2697 	td->td_sa = tsr->ts_sa;
2698 
2699 #ifdef CAPABILITY_MODE
2700 	if ((se->sy_flags & SYF_CAPENABLED) == 0) {
2701 		if (CAP_TRACING(td))
2702 			ktrcapfail(CAPFAIL_SYSCALL, NULL);
2703 		if (IN_CAPABILITY_MODE(td)) {
2704 			tsr->ts_ret.sr_error = ECAPMODE;
2705 			return;
2706 		}
2707 	}
2708 #endif
2709 
2710 	sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2711 	audited = AUDIT_SYSCALL_ENTER(sc, td) != 0;
2712 
2713 	if (!sy_thr_static) {
2714 		error = syscall_thread_enter(td, &se);
2715 		sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2716 		if (error != 0) {
2717 			tsr->ts_ret.sr_error = error;
2718 			return;
2719 		}
2720 	}
2721 
2722 	rv_saved[0] = td->td_retval[0];
2723 	rv_saved[1] = td->td_retval[1];
2724 	nerror = td->td_errno;
2725 	td->td_retval[0] = 0;
2726 	td->td_retval[1] = 0;
2727 
2728 #ifdef KDTRACE_HOOKS
2729 	if (se->sy_entry != 0)
2730 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_ENTRY, 0);
2731 #endif
2732 	tsr->ts_ret.sr_error = se->sy_call(td, tsr->ts_sa.args);
2733 #ifdef KDTRACE_HOOKS
2734 	if (se->sy_return != 0)
2735 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_RETURN,
2736 		    tsr->ts_ret.sr_error != 0 ? -1 : td->td_retval[0]);
2737 #endif
2738 
2739 	tsr->ts_ret.sr_retval[0] = td->td_retval[0];
2740 	tsr->ts_ret.sr_retval[1] = td->td_retval[1];
2741 	td->td_retval[0] = rv_saved[0];
2742 	td->td_retval[1] = rv_saved[1];
2743 	td->td_errno = nerror;
2744 
2745 	if (audited)
2746 		AUDIT_SYSCALL_EXIT(error, td);
2747 	if (!sy_thr_static)
2748 		syscall_thread_exit(td, se);
2749 }
2750 
2751 static void
ptrace_remotereq(struct thread * td,int flag)2752 ptrace_remotereq(struct thread *td, int flag)
2753 {
2754 	struct proc *p;
2755 
2756 	MPASS(td == curthread);
2757 	p = td->td_proc;
2758 	PROC_LOCK_ASSERT(p, MA_OWNED);
2759 	if ((td->td_dbgflags & flag) == 0)
2760 		return;
2761 	KASSERT((p->p_flag & P_STOPPED_TRACE) != 0, ("not stopped"));
2762 	KASSERT(td->td_remotereq != NULL, ("td_remotereq is NULL"));
2763 
2764 	PROC_UNLOCK(p);
2765 	switch (flag) {
2766 	case TDB_COREDUMPREQ:
2767 		ptrace_coredumpreq(td, p, td->td_remotereq);
2768 		break;
2769 	case TDB_SCREMOTEREQ:
2770 		ptrace_syscallreq(td, p, td->td_remotereq);
2771 		break;
2772 	default:
2773 		__unreachable();
2774 	}
2775 	PROC_LOCK(p);
2776 
2777 	MPASS((td->td_dbgflags & flag) != 0);
2778 	td->td_dbgflags &= ~flag;
2779 	td->td_remotereq = NULL;
2780 	wakeup(p);
2781 }
2782 
2783 static int
sig_suspend_threads(struct thread * td,struct proc * p)2784 sig_suspend_threads(struct thread *td, struct proc *p)
2785 {
2786 	struct thread *td2;
2787 	int wakeup_swapper;
2788 
2789 	PROC_LOCK_ASSERT(p, MA_OWNED);
2790 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2791 
2792 	wakeup_swapper = 0;
2793 	FOREACH_THREAD_IN_PROC(p, td2) {
2794 		thread_lock(td2);
2795 		ast_sched_locked(td2, TDA_SUSPEND);
2796 		if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2797 		    (td2->td_flags & TDF_SINTR)) {
2798 			if (td2->td_flags & TDF_SBDRY) {
2799 				/*
2800 				 * Once a thread is asleep with
2801 				 * TDF_SBDRY and without TDF_SERESTART
2802 				 * or TDF_SEINTR set, it should never
2803 				 * become suspended due to this check.
2804 				 */
2805 				KASSERT(!TD_IS_SUSPENDED(td2),
2806 				    ("thread with deferred stops suspended"));
2807 				if (TD_SBDRY_INTR(td2)) {
2808 					wakeup_swapper |= sleepq_abort(td2,
2809 					    TD_SBDRY_ERRNO(td2));
2810 					continue;
2811 				}
2812 			} else if (!TD_IS_SUSPENDED(td2))
2813 				thread_suspend_one(td2);
2814 		} else if (!TD_IS_SUSPENDED(td2)) {
2815 #ifdef SMP
2816 			if (TD_IS_RUNNING(td2) && td2 != td)
2817 				forward_signal(td2);
2818 #endif
2819 		}
2820 		thread_unlock(td2);
2821 	}
2822 	return (wakeup_swapper);
2823 }
2824 
2825 /*
2826  * Stop the process for an event deemed interesting to the debugger. If si is
2827  * non-NULL, this is a signal exchange; the new signal requested by the
2828  * debugger will be returned for handling. If si is NULL, this is some other
2829  * type of interesting event. The debugger may request a signal be delivered in
2830  * that case as well, however it will be deferred until it can be handled.
2831  */
2832 int
ptracestop(struct thread * td,int sig,ksiginfo_t * si)2833 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2834 {
2835 	struct proc *p = td->td_proc;
2836 	struct thread *td2;
2837 	ksiginfo_t ksi;
2838 
2839 	PROC_LOCK_ASSERT(p, MA_OWNED);
2840 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2841 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2842 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2843 
2844 	td->td_xsig = sig;
2845 
2846 	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2847 		td->td_dbgflags |= TDB_XSIG;
2848 		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2849 		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2850 		PROC_SLOCK(p);
2851 		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2852 			if (P_KILLED(p)) {
2853 				/*
2854 				 * Ensure that, if we've been PT_KILLed, the
2855 				 * exit status reflects that. Another thread
2856 				 * may also be in ptracestop(), having just
2857 				 * received the SIGKILL, but this thread was
2858 				 * unsuspended first.
2859 				 */
2860 				td->td_dbgflags &= ~TDB_XSIG;
2861 				td->td_xsig = SIGKILL;
2862 				p->p_ptevents = 0;
2863 				break;
2864 			}
2865 			if (p->p_flag & P_SINGLE_EXIT &&
2866 			    !(td->td_dbgflags & TDB_EXIT)) {
2867 				/*
2868 				 * Ignore ptrace stops except for thread exit
2869 				 * events when the process exits.
2870 				 */
2871 				td->td_dbgflags &= ~TDB_XSIG;
2872 				PROC_SUNLOCK(p);
2873 				return (0);
2874 			}
2875 
2876 			/*
2877 			 * Make wait(2) work.  Ensure that right after the
2878 			 * attach, the thread which was decided to become the
2879 			 * leader of attach gets reported to the waiter.
2880 			 * Otherwise, just avoid overwriting another thread's
2881 			 * assignment to p_xthread.  If another thread has
2882 			 * already set p_xthread, the current thread will get
2883 			 * a chance to report itself upon the next iteration.
2884 			 */
2885 			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2886 			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2887 			    p->p_xthread == NULL)) {
2888 				p->p_xsig = sig;
2889 				p->p_xthread = td;
2890 
2891 				/*
2892 				 * If we are on sleepqueue already,
2893 				 * let sleepqueue code decide if it
2894 				 * needs to go sleep after attach.
2895 				 */
2896 				if (td->td_wchan == NULL)
2897 					td->td_dbgflags &= ~TDB_FSTP;
2898 
2899 				p->p_flag2 &= ~P2_PTRACE_FSTP;
2900 				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2901 				sig_suspend_threads(td, p);
2902 			}
2903 			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2904 				td->td_dbgflags &= ~TDB_STOPATFORK;
2905 			}
2906 stopme:
2907 			td->td_dbgflags |= TDB_SSWITCH;
2908 			thread_suspend_switch(td, p);
2909 			td->td_dbgflags &= ~TDB_SSWITCH;
2910 			if ((td->td_dbgflags & (TDB_COREDUMPREQ |
2911 			    TDB_SCREMOTEREQ)) != 0) {
2912 				MPASS((td->td_dbgflags & (TDB_COREDUMPREQ |
2913 				    TDB_SCREMOTEREQ)) !=
2914 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2915 				PROC_SUNLOCK(p);
2916 				ptrace_remotereq(td, td->td_dbgflags &
2917 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2918 				PROC_SLOCK(p);
2919 				goto stopme;
2920 			}
2921 			if (p->p_xthread == td)
2922 				p->p_xthread = NULL;
2923 			if (!(p->p_flag & P_TRACED))
2924 				break;
2925 			if (td->td_dbgflags & TDB_SUSPEND) {
2926 				if (p->p_flag & P_SINGLE_EXIT)
2927 					break;
2928 				goto stopme;
2929 			}
2930 		}
2931 		PROC_SUNLOCK(p);
2932 	}
2933 
2934 	if (si != NULL && sig == td->td_xsig) {
2935 		/* Parent wants us to take the original signal unchanged. */
2936 		si->ksi_flags |= KSI_HEAD;
2937 		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2938 			si->ksi_signo = 0;
2939 	} else if (td->td_xsig != 0) {
2940 		/*
2941 		 * If parent wants us to take a new signal, then it will leave
2942 		 * it in td->td_xsig; otherwise we just look for signals again.
2943 		 */
2944 		ksiginfo_init(&ksi);
2945 		ksi.ksi_signo = td->td_xsig;
2946 		ksi.ksi_flags |= KSI_PTRACE;
2947 		td2 = sigtd(p, td->td_xsig, false);
2948 		tdsendsignal(p, td2, td->td_xsig, &ksi);
2949 		if (td != td2)
2950 			return (0);
2951 	}
2952 
2953 	return (td->td_xsig);
2954 }
2955 
2956 static void
reschedule_signals(struct proc * p,sigset_t block,int flags)2957 reschedule_signals(struct proc *p, sigset_t block, int flags)
2958 {
2959 	struct sigacts *ps;
2960 	struct thread *td;
2961 	int sig;
2962 	bool fastblk, pslocked;
2963 
2964 	PROC_LOCK_ASSERT(p, MA_OWNED);
2965 	ps = p->p_sigacts;
2966 	pslocked = (flags & SIGPROCMASK_PS_LOCKED) != 0;
2967 	mtx_assert(&ps->ps_mtx, pslocked ? MA_OWNED : MA_NOTOWNED);
2968 	if (SIGISEMPTY(p->p_siglist))
2969 		return;
2970 	SIGSETAND(block, p->p_siglist);
2971 	fastblk = (flags & SIGPROCMASK_FASTBLK) != 0;
2972 	SIG_FOREACH(sig, &block) {
2973 		td = sigtd(p, sig, fastblk);
2974 
2975 		/*
2976 		 * If sigtd() selected us despite sigfastblock is
2977 		 * blocking, do not activate AST or wake us, to avoid
2978 		 * loop in AST handler.
2979 		 */
2980 		if (fastblk && td == curthread)
2981 			continue;
2982 
2983 		signotify(td);
2984 		if (!pslocked)
2985 			mtx_lock(&ps->ps_mtx);
2986 		if (p->p_flag & P_TRACED ||
2987 		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2988 		    !SIGISMEMBER(td->td_sigmask, sig))) {
2989 			tdsigwakeup(td, sig, SIG_CATCH,
2990 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2991 			    ERESTART));
2992 		}
2993 		if (!pslocked)
2994 			mtx_unlock(&ps->ps_mtx);
2995 	}
2996 }
2997 
2998 void
tdsigcleanup(struct thread * td)2999 tdsigcleanup(struct thread *td)
3000 {
3001 	struct proc *p;
3002 	sigset_t unblocked;
3003 
3004 	p = td->td_proc;
3005 	PROC_LOCK_ASSERT(p, MA_OWNED);
3006 
3007 	sigqueue_flush(&td->td_sigqueue);
3008 	if (p->p_numthreads == 1)
3009 		return;
3010 
3011 	/*
3012 	 * Since we cannot handle signals, notify signal post code
3013 	 * about this by filling the sigmask.
3014 	 *
3015 	 * Also, if needed, wake up thread(s) that do not block the
3016 	 * same signals as the exiting thread, since the thread might
3017 	 * have been selected for delivery and woken up.
3018 	 */
3019 	SIGFILLSET(unblocked);
3020 	SIGSETNAND(unblocked, td->td_sigmask);
3021 	SIGFILLSET(td->td_sigmask);
3022 	reschedule_signals(p, unblocked, 0);
3023 
3024 }
3025 
3026 static int
sigdeferstop_curr_flags(int cflags)3027 sigdeferstop_curr_flags(int cflags)
3028 {
3029 
3030 	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
3031 	    (cflags & TDF_SBDRY) != 0);
3032 	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
3033 }
3034 
3035 /*
3036  * Defer the delivery of SIGSTOP for the current thread, according to
3037  * the requested mode.  Returns previous flags, which must be restored
3038  * by sigallowstop().
3039  *
3040  * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
3041  * cleared by the current thread, which allow the lock-less read-only
3042  * accesses below.
3043  */
3044 int
sigdeferstop_impl(int mode)3045 sigdeferstop_impl(int mode)
3046 {
3047 	struct thread *td;
3048 	int cflags, nflags;
3049 
3050 	td = curthread;
3051 	cflags = sigdeferstop_curr_flags(td->td_flags);
3052 	switch (mode) {
3053 	case SIGDEFERSTOP_NOP:
3054 		nflags = cflags;
3055 		break;
3056 	case SIGDEFERSTOP_OFF:
3057 		nflags = 0;
3058 		break;
3059 	case SIGDEFERSTOP_SILENT:
3060 		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
3061 		break;
3062 	case SIGDEFERSTOP_EINTR:
3063 		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
3064 		break;
3065 	case SIGDEFERSTOP_ERESTART:
3066 		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
3067 		break;
3068 	default:
3069 		panic("sigdeferstop: invalid mode %x", mode);
3070 		break;
3071 	}
3072 	if (cflags == nflags)
3073 		return (SIGDEFERSTOP_VAL_NCHG);
3074 	thread_lock(td);
3075 	td->td_flags = (td->td_flags & ~cflags) | nflags;
3076 	thread_unlock(td);
3077 	return (cflags);
3078 }
3079 
3080 /*
3081  * Restores the STOP handling mode, typically permitting the delivery
3082  * of SIGSTOP for the current thread.  This does not immediately
3083  * suspend if a stop was posted.  Instead, the thread will suspend
3084  * either via ast() or a subsequent interruptible sleep.
3085  */
3086 void
sigallowstop_impl(int prev)3087 sigallowstop_impl(int prev)
3088 {
3089 	struct thread *td;
3090 	int cflags;
3091 
3092 	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
3093 	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
3094 	    ("sigallowstop: incorrect previous mode %x", prev));
3095 	td = curthread;
3096 	cflags = sigdeferstop_curr_flags(td->td_flags);
3097 	if (cflags != prev) {
3098 		thread_lock(td);
3099 		td->td_flags = (td->td_flags & ~cflags) | prev;
3100 		thread_unlock(td);
3101 	}
3102 }
3103 
3104 enum sigstatus {
3105 	SIGSTATUS_HANDLE,
3106 	SIGSTATUS_HANDLED,
3107 	SIGSTATUS_IGNORE,
3108 	SIGSTATUS_SBDRY_STOP,
3109 };
3110 
3111 /*
3112  * The thread has signal "sig" pending.  Figure out what to do with it:
3113  *
3114  * _HANDLE     -> the caller should handle the signal
3115  * _HANDLED    -> handled internally, reload pending signal set
3116  * _IGNORE     -> ignored, remove from the set of pending signals and try the
3117  *                next pending signal
3118  * _SBDRY_STOP -> the signal should stop the thread but this is not
3119  *                permitted in the current context
3120  */
3121 static enum sigstatus
sigprocess(struct thread * td,int sig)3122 sigprocess(struct thread *td, int sig)
3123 {
3124 	struct proc *p;
3125 	struct sigacts *ps;
3126 	struct sigqueue *queue;
3127 	ksiginfo_t ksi;
3128 	int prop;
3129 
3130 	KASSERT(_SIG_VALID(sig), ("%s: invalid signal %d", __func__, sig));
3131 
3132 	p = td->td_proc;
3133 	ps = p->p_sigacts;
3134 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3135 	PROC_LOCK_ASSERT(p, MA_OWNED);
3136 
3137 	/*
3138 	 * We should allow pending but ignored signals below
3139 	 * if there is sigwait() active, or P_TRACED was
3140 	 * on when they were posted.
3141 	 */
3142 	if (SIGISMEMBER(ps->ps_sigignore, sig) &&
3143 	    (p->p_flag & P_TRACED) == 0 &&
3144 	    (td->td_flags & TDF_SIGWAIT) == 0) {
3145 		return (SIGSTATUS_IGNORE);
3146 	}
3147 
3148 	/*
3149 	 * If the process is going to single-thread mode to prepare
3150 	 * for exit, there is no sense in delivering any signal
3151 	 * to usermode.  Another important consequence is that
3152 	 * msleep(..., PCATCH, ...) now is only interruptible by a
3153 	 * suspend request.
3154 	 */
3155 	if ((p->p_flag2 & P2_WEXIT) != 0)
3156 		return (SIGSTATUS_IGNORE);
3157 
3158 	if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
3159 		/*
3160 		 * If traced, always stop.
3161 		 * Remove old signal from queue before the stop.
3162 		 * XXX shrug off debugger, it causes siginfo to
3163 		 * be thrown away.
3164 		 */
3165 		queue = &td->td_sigqueue;
3166 		ksiginfo_init(&ksi);
3167 		if (sigqueue_get(queue, sig, &ksi) == 0) {
3168 			queue = &p->p_sigqueue;
3169 			sigqueue_get(queue, sig, &ksi);
3170 		}
3171 		td->td_si = ksi.ksi_info;
3172 
3173 		mtx_unlock(&ps->ps_mtx);
3174 		sig = ptracestop(td, sig, &ksi);
3175 		mtx_lock(&ps->ps_mtx);
3176 
3177 		td->td_si.si_signo = 0;
3178 
3179 		/*
3180 		 * Keep looking if the debugger discarded or
3181 		 * replaced the signal.
3182 		 */
3183 		if (sig == 0)
3184 			return (SIGSTATUS_HANDLED);
3185 
3186 		/*
3187 		 * If the signal became masked, re-queue it.
3188 		 */
3189 		if (SIGISMEMBER(td->td_sigmask, sig)) {
3190 			ksi.ksi_flags |= KSI_HEAD;
3191 			sigqueue_add(&p->p_sigqueue, sig, &ksi);
3192 			return (SIGSTATUS_HANDLED);
3193 		}
3194 
3195 		/*
3196 		 * If the traced bit got turned off, requeue the signal and
3197 		 * reload the set of pending signals.  This ensures that p_sig*
3198 		 * and p_sigact are consistent.
3199 		 */
3200 		if ((p->p_flag & P_TRACED) == 0) {
3201 			if ((ksi.ksi_flags & KSI_PTRACE) == 0) {
3202 				ksi.ksi_flags |= KSI_HEAD;
3203 				sigqueue_add(queue, sig, &ksi);
3204 			}
3205 			return (SIGSTATUS_HANDLED);
3206 		}
3207 	}
3208 
3209 	/*
3210 	 * Decide whether the signal should be returned.
3211 	 * Return the signal's number, or fall through
3212 	 * to clear it from the pending mask.
3213 	 */
3214 	switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
3215 	case (intptr_t)SIG_DFL:
3216 		/*
3217 		 * Don't take default actions on system processes.
3218 		 */
3219 		if (p->p_pid <= 1) {
3220 #ifdef DIAGNOSTIC
3221 			/*
3222 			 * Are you sure you want to ignore SIGSEGV
3223 			 * in init? XXX
3224 			 */
3225 			printf("Process (pid %lu) got signal %d\n",
3226 				(u_long)p->p_pid, sig);
3227 #endif
3228 			return (SIGSTATUS_IGNORE);
3229 		}
3230 
3231 		/*
3232 		 * If there is a pending stop signal to process with
3233 		 * default action, stop here, then clear the signal.
3234 		 * Traced or exiting processes should ignore stops.
3235 		 * Additionally, a member of an orphaned process group
3236 		 * should ignore tty stops.
3237 		 */
3238 		prop = sigprop(sig);
3239 		if (prop & SIGPROP_STOP) {
3240 			mtx_unlock(&ps->ps_mtx);
3241 			if ((p->p_flag & (P_TRACED | P_WEXIT |
3242 			    P_SINGLE_EXIT)) != 0 || ((p->p_pgrp->
3243 			    pg_flags & PGRP_ORPHANED) != 0 &&
3244 			    (prop & SIGPROP_TTYSTOP) != 0)) {
3245 				mtx_lock(&ps->ps_mtx);
3246 				return (SIGSTATUS_IGNORE);
3247 			}
3248 			if (TD_SBDRY_INTR(td)) {
3249 				KASSERT((td->td_flags & TDF_SBDRY) != 0,
3250 				    ("lost TDF_SBDRY"));
3251 				mtx_lock(&ps->ps_mtx);
3252 				return (SIGSTATUS_SBDRY_STOP);
3253 			}
3254 			WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
3255 			    &p->p_mtx.lock_object, "Catching SIGSTOP");
3256 			sigqueue_delete(&td->td_sigqueue, sig);
3257 			sigqueue_delete(&p->p_sigqueue, sig);
3258 			p->p_flag |= P_STOPPED_SIG;
3259 			p->p_xsig = sig;
3260 			PROC_SLOCK(p);
3261 			sig_suspend_threads(td, p);
3262 			thread_suspend_switch(td, p);
3263 			PROC_SUNLOCK(p);
3264 			mtx_lock(&ps->ps_mtx);
3265 			return (SIGSTATUS_HANDLED);
3266 		} else if ((prop & SIGPROP_IGNORE) != 0 &&
3267 		    (td->td_flags & TDF_SIGWAIT) == 0) {
3268 			/*
3269 			 * Default action is to ignore; drop it if
3270 			 * not in kern_sigtimedwait().
3271 			 */
3272 			return (SIGSTATUS_IGNORE);
3273 		} else {
3274 			return (SIGSTATUS_HANDLE);
3275 		}
3276 
3277 	case (intptr_t)SIG_IGN:
3278 		if ((td->td_flags & TDF_SIGWAIT) == 0)
3279 			return (SIGSTATUS_IGNORE);
3280 		else
3281 			return (SIGSTATUS_HANDLE);
3282 
3283 	default:
3284 		/*
3285 		 * This signal has an action, let postsig() process it.
3286 		 */
3287 		return (SIGSTATUS_HANDLE);
3288 	}
3289 }
3290 
3291 /*
3292  * If the current process has received a signal (should be caught or cause
3293  * termination, should interrupt current syscall), return the signal number.
3294  * Stop signals with default action are processed immediately, then cleared;
3295  * they aren't returned.  This is checked after each entry to the system for
3296  * a syscall or trap (though this can usually be done without calling
3297  * issignal by checking the pending signal masks in cursig.) The normal call
3298  * sequence is
3299  *
3300  *	while (sig = cursig(curthread))
3301  *		postsig(sig);
3302  */
3303 static int
issignal(struct thread * td)3304 issignal(struct thread *td)
3305 {
3306 	struct proc *p;
3307 	sigset_t sigpending;
3308 	int sig;
3309 
3310 	p = td->td_proc;
3311 	PROC_LOCK_ASSERT(p, MA_OWNED);
3312 
3313 	for (;;) {
3314 		sigpending = td->td_sigqueue.sq_signals;
3315 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
3316 		SIGSETNAND(sigpending, td->td_sigmask);
3317 
3318 		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
3319 		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
3320 			SIG_STOPSIGMASK(sigpending);
3321 		if (SIGISEMPTY(sigpending))	/* no signal to send */
3322 			return (0);
3323 
3324 		/*
3325 		 * Do fast sigblock if requested by usermode.  Since
3326 		 * we do know that there was a signal pending at this
3327 		 * point, set the FAST_SIGBLOCK_PEND as indicator for
3328 		 * usermode to perform a dummy call to
3329 		 * FAST_SIGBLOCK_UNBLOCK, which causes immediate
3330 		 * delivery of postponed pending signal.
3331 		 */
3332 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
3333 			if (td->td_sigblock_val != 0)
3334 				SIGSETNAND(sigpending, fastblock_mask);
3335 			if (SIGISEMPTY(sigpending)) {
3336 				td->td_pflags |= TDP_SIGFASTPENDING;
3337 				return (0);
3338 			}
3339 		}
3340 
3341 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
3342 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
3343 		    SIGISMEMBER(sigpending, SIGSTOP)) {
3344 			/*
3345 			 * If debugger just attached, always consume
3346 			 * SIGSTOP from ptrace(PT_ATTACH) first, to
3347 			 * execute the debugger attach ritual in
3348 			 * order.
3349 			 */
3350 			td->td_dbgflags |= TDB_FSTP;
3351 			SIGEMPTYSET(sigpending);
3352 			SIGADDSET(sigpending, SIGSTOP);
3353 		}
3354 
3355 		SIG_FOREACH(sig, &sigpending) {
3356 			switch (sigprocess(td, sig)) {
3357 			case SIGSTATUS_HANDLE:
3358 				return (sig);
3359 			case SIGSTATUS_HANDLED:
3360 				goto next;
3361 			case SIGSTATUS_IGNORE:
3362 				sigqueue_delete(&td->td_sigqueue, sig);
3363 				sigqueue_delete(&p->p_sigqueue, sig);
3364 				break;
3365 			case SIGSTATUS_SBDRY_STOP:
3366 				return (-1);
3367 			}
3368 		}
3369 next:;
3370 	}
3371 }
3372 
3373 void
thread_stopped(struct proc * p)3374 thread_stopped(struct proc *p)
3375 {
3376 	int n;
3377 
3378 	PROC_LOCK_ASSERT(p, MA_OWNED);
3379 	PROC_SLOCK_ASSERT(p, MA_OWNED);
3380 	n = p->p_suspcount;
3381 	if (p == curproc)
3382 		n++;
3383 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3384 		PROC_SUNLOCK(p);
3385 		p->p_flag &= ~P_WAITED;
3386 		PROC_LOCK(p->p_pptr);
3387 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
3388 			CLD_TRAPPED : CLD_STOPPED);
3389 		PROC_UNLOCK(p->p_pptr);
3390 		PROC_SLOCK(p);
3391 	}
3392 }
3393 
3394 /*
3395  * Take the action for the specified signal
3396  * from the current set of pending signals.
3397  */
3398 int
postsig(int sig)3399 postsig(int sig)
3400 {
3401 	struct thread *td;
3402 	struct proc *p;
3403 	struct sigacts *ps;
3404 	sig_t action;
3405 	ksiginfo_t ksi;
3406 	sigset_t returnmask;
3407 
3408 	KASSERT(sig != 0, ("postsig"));
3409 
3410 	td = curthread;
3411 	p = td->td_proc;
3412 	PROC_LOCK_ASSERT(p, MA_OWNED);
3413 	ps = p->p_sigacts;
3414 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3415 	ksiginfo_init(&ksi);
3416 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3417 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3418 		return (0);
3419 	ksi.ksi_signo = sig;
3420 	if (ksi.ksi_code == SI_TIMER)
3421 		itimer_accept(p, ksi.ksi_timerid, &ksi);
3422 	action = ps->ps_sigact[_SIG_IDX(sig)];
3423 #ifdef KTRACE
3424 	if (KTRPOINT(td, KTR_PSIG))
3425 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3426 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3427 #endif
3428 
3429 	if (action == SIG_DFL) {
3430 		/*
3431 		 * Default action, where the default is to kill
3432 		 * the process.  (Other cases were ignored above.)
3433 		 */
3434 		mtx_unlock(&ps->ps_mtx);
3435 		proc_td_siginfo_capture(td, &ksi.ksi_info);
3436 		sigexit(td, sig);
3437 		/* NOTREACHED */
3438 	} else {
3439 		/*
3440 		 * If we get here, the signal must be caught.
3441 		 */
3442 		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3443 		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3444 		    ("postsig action: blocked sig %d", sig));
3445 
3446 		/*
3447 		 * Set the new mask value and also defer further
3448 		 * occurrences of this signal.
3449 		 *
3450 		 * Special case: user has done a sigsuspend.  Here the
3451 		 * current mask is not of interest, but rather the
3452 		 * mask from before the sigsuspend is what we want
3453 		 * restored after the signal processing is completed.
3454 		 */
3455 		if (td->td_pflags & TDP_OLDMASK) {
3456 			returnmask = td->td_oldsigmask;
3457 			td->td_pflags &= ~TDP_OLDMASK;
3458 		} else
3459 			returnmask = td->td_sigmask;
3460 
3461 		if (p->p_sig == sig) {
3462 			p->p_sig = 0;
3463 		}
3464 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3465 		postsig_done(sig, td, ps);
3466 	}
3467 	return (1);
3468 }
3469 
3470 int
sig_ast_checksusp(struct thread * td)3471 sig_ast_checksusp(struct thread *td)
3472 {
3473 	struct proc *p __diagused;
3474 	int ret;
3475 
3476 	p = td->td_proc;
3477 	PROC_LOCK_ASSERT(p, MA_OWNED);
3478 
3479 	if (!td_ast_pending(td, TDA_SUSPEND))
3480 		return (0);
3481 
3482 	ret = thread_suspend_check(1);
3483 	MPASS(ret == 0 || ret == EINTR || ret == ERESTART);
3484 	return (ret);
3485 }
3486 
3487 int
sig_ast_needsigchk(struct thread * td)3488 sig_ast_needsigchk(struct thread *td)
3489 {
3490 	struct proc *p;
3491 	struct sigacts *ps;
3492 	int ret, sig;
3493 
3494 	p = td->td_proc;
3495 	PROC_LOCK_ASSERT(p, MA_OWNED);
3496 
3497 	if (!td_ast_pending(td, TDA_SIG))
3498 		return (0);
3499 
3500 	ps = p->p_sigacts;
3501 	mtx_lock(&ps->ps_mtx);
3502 	sig = cursig(td);
3503 	if (sig == -1) {
3504 		mtx_unlock(&ps->ps_mtx);
3505 		KASSERT((td->td_flags & TDF_SBDRY) != 0, ("lost TDF_SBDRY"));
3506 		KASSERT(TD_SBDRY_INTR(td),
3507 		    ("lost TDF_SERESTART of TDF_SEINTR"));
3508 		KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
3509 		    (TDF_SEINTR | TDF_SERESTART),
3510 		    ("both TDF_SEINTR and TDF_SERESTART"));
3511 		ret = TD_SBDRY_ERRNO(td);
3512 	} else if (sig != 0) {
3513 		ret = SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR : ERESTART;
3514 		mtx_unlock(&ps->ps_mtx);
3515 	} else {
3516 		mtx_unlock(&ps->ps_mtx);
3517 		ret = 0;
3518 	}
3519 
3520 	/*
3521 	 * Do not go into sleep if this thread was the ptrace(2)
3522 	 * attach leader.  cursig() consumed SIGSTOP from PT_ATTACH,
3523 	 * but we usually act on the signal by interrupting sleep, and
3524 	 * should do that here as well.
3525 	 */
3526 	if ((td->td_dbgflags & TDB_FSTP) != 0) {
3527 		if (ret == 0)
3528 			ret = EINTR;
3529 		td->td_dbgflags &= ~TDB_FSTP;
3530 	}
3531 
3532 	return (ret);
3533 }
3534 
3535 int
sig_intr(void)3536 sig_intr(void)
3537 {
3538 	struct thread *td;
3539 	struct proc *p;
3540 	int ret;
3541 
3542 	td = curthread;
3543 	if (!td_ast_pending(td, TDA_SIG) && !td_ast_pending(td, TDA_SUSPEND))
3544 		return (0);
3545 
3546 	p = td->td_proc;
3547 
3548 	PROC_LOCK(p);
3549 	ret = sig_ast_checksusp(td);
3550 	if (ret == 0)
3551 		ret = sig_ast_needsigchk(td);
3552 	PROC_UNLOCK(p);
3553 	return (ret);
3554 }
3555 
3556 bool
curproc_sigkilled(void)3557 curproc_sigkilled(void)
3558 {
3559 	struct thread *td;
3560 	struct proc *p;
3561 	struct sigacts *ps;
3562 	bool res;
3563 
3564 	td = curthread;
3565 	if (!td_ast_pending(td, TDA_SIG))
3566 		return (false);
3567 
3568 	p = td->td_proc;
3569 	PROC_LOCK(p);
3570 	ps = p->p_sigacts;
3571 	mtx_lock(&ps->ps_mtx);
3572 	res = SIGISMEMBER(td->td_sigqueue.sq_signals, SIGKILL) ||
3573 	    SIGISMEMBER(p->p_sigqueue.sq_signals, SIGKILL);
3574 	mtx_unlock(&ps->ps_mtx);
3575 	PROC_UNLOCK(p);
3576 	return (res);
3577 }
3578 
3579 void
proc_wkilled(struct proc * p)3580 proc_wkilled(struct proc *p)
3581 {
3582 
3583 	PROC_LOCK_ASSERT(p, MA_OWNED);
3584 	if ((p->p_flag & P_WKILLED) == 0) {
3585 		p->p_flag |= P_WKILLED;
3586 		/*
3587 		 * Notify swapper that there is a process to swap in.
3588 		 * The notification is racy, at worst it would take 10
3589 		 * seconds for the swapper process to notice.
3590 		 */
3591 		if ((p->p_flag & (P_INMEM | P_SWAPPINGIN)) == 0)
3592 			wakeup(&proc0);
3593 	}
3594 }
3595 
3596 /*
3597  * Kill the current process for stated reason.
3598  */
3599 void
killproc(struct proc * p,const char * why)3600 killproc(struct proc *p, const char *why)
3601 {
3602 
3603 	PROC_LOCK_ASSERT(p, MA_OWNED);
3604 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3605 	    p->p_comm);
3606 	log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3607 	    p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3608 	    p->p_ucred->cr_uid, why);
3609 	proc_wkilled(p);
3610 	kern_psignal(p, SIGKILL);
3611 }
3612 
3613 /*
3614  * Force the current process to exit with the specified signal, dumping core
3615  * if appropriate.  We bypass the normal tests for masked and caught signals,
3616  * allowing unrecoverable failures to terminate the process without changing
3617  * signal state.  Mark the accounting record with the signal termination.
3618  * If dumping core, save the signal number for the debugger.  Calls exit and
3619  * does not return.
3620  */
3621 void
sigexit(struct thread * td,int sig)3622 sigexit(struct thread *td, int sig)
3623 {
3624 	struct proc *p = td->td_proc;
3625 	const char *coreinfo;
3626 	int rv;
3627 
3628 	PROC_LOCK_ASSERT(p, MA_OWNED);
3629 	proc_set_p2_wexit(p);
3630 
3631 	p->p_acflag |= AXSIG;
3632 	/*
3633 	 * We must be single-threading to generate a core dump.  This
3634 	 * ensures that the registers in the core file are up-to-date.
3635 	 * Also, the ELF dump handler assumes that the thread list doesn't
3636 	 * change out from under it.
3637 	 *
3638 	 * XXX If another thread attempts to single-thread before us
3639 	 *     (e.g. via fork()), we won't get a dump at all.
3640 	 */
3641 	if ((sigprop(sig) & SIGPROP_CORE) &&
3642 	    thread_single(p, SINGLE_NO_EXIT) == 0) {
3643 		p->p_sig = sig;
3644 		/*
3645 		 * Log signals which would cause core dumps
3646 		 * (Log as LOG_INFO to appease those who don't want
3647 		 * these messages.)
3648 		 * XXX : Todo, as well as euid, write out ruid too
3649 		 * Note that coredump() drops proc lock.
3650 		 */
3651 		rv = coredump(td);
3652 		switch (rv) {
3653 		case 0:
3654 			sig |= WCOREFLAG;
3655 			coreinfo = " (core dumped)";
3656 			break;
3657 		case EFAULT:
3658 			coreinfo = " (no core dump - bad address)";
3659 			break;
3660 		case EINVAL:
3661 			coreinfo = " (no core dump - invalid argument)";
3662 			break;
3663 		case EFBIG:
3664 			coreinfo = " (no core dump - too large)";
3665 			break;
3666 		default:
3667 			coreinfo = " (no core dump - other error)";
3668 			break;
3669 		}
3670 		if (kern_logsigexit)
3671 			log(LOG_INFO,
3672 			    "pid %d (%s), jid %d, uid %d: exited on "
3673 			    "signal %d%s\n", p->p_pid, p->p_comm,
3674 			    p->p_ucred->cr_prison->pr_id,
3675 			    td->td_ucred->cr_uid,
3676 			    sig &~ WCOREFLAG, coreinfo);
3677 	} else
3678 		PROC_UNLOCK(p);
3679 	exit1(td, 0, sig);
3680 	/* NOTREACHED */
3681 }
3682 
3683 /*
3684  * Send queued SIGCHLD to parent when child process's state
3685  * is changed.
3686  */
3687 static void
sigparent(struct proc * p,int reason,int status)3688 sigparent(struct proc *p, int reason, int status)
3689 {
3690 	PROC_LOCK_ASSERT(p, MA_OWNED);
3691 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3692 
3693 	if (p->p_ksi != NULL) {
3694 		p->p_ksi->ksi_signo  = SIGCHLD;
3695 		p->p_ksi->ksi_code   = reason;
3696 		p->p_ksi->ksi_status = status;
3697 		p->p_ksi->ksi_pid    = p->p_pid;
3698 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3699 		if (KSI_ONQ(p->p_ksi))
3700 			return;
3701 	}
3702 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3703 }
3704 
3705 static void
childproc_jobstate(struct proc * p,int reason,int sig)3706 childproc_jobstate(struct proc *p, int reason, int sig)
3707 {
3708 	struct sigacts *ps;
3709 
3710 	PROC_LOCK_ASSERT(p, MA_OWNED);
3711 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3712 
3713 	/*
3714 	 * Wake up parent sleeping in kern_wait(), also send
3715 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3716 	 * that parent will awake, because parent may masked
3717 	 * the signal.
3718 	 */
3719 	p->p_pptr->p_flag |= P_STATCHILD;
3720 	wakeup(p->p_pptr);
3721 
3722 	ps = p->p_pptr->p_sigacts;
3723 	mtx_lock(&ps->ps_mtx);
3724 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3725 		mtx_unlock(&ps->ps_mtx);
3726 		sigparent(p, reason, sig);
3727 	} else
3728 		mtx_unlock(&ps->ps_mtx);
3729 }
3730 
3731 void
childproc_stopped(struct proc * p,int reason)3732 childproc_stopped(struct proc *p, int reason)
3733 {
3734 
3735 	childproc_jobstate(p, reason, p->p_xsig);
3736 }
3737 
3738 void
childproc_continued(struct proc * p)3739 childproc_continued(struct proc *p)
3740 {
3741 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3742 }
3743 
3744 void
childproc_exited(struct proc * p)3745 childproc_exited(struct proc *p)
3746 {
3747 	int reason, status;
3748 
3749 	if (WCOREDUMP(p->p_xsig)) {
3750 		reason = CLD_DUMPED;
3751 		status = WTERMSIG(p->p_xsig);
3752 	} else if (WIFSIGNALED(p->p_xsig)) {
3753 		reason = CLD_KILLED;
3754 		status = WTERMSIG(p->p_xsig);
3755 	} else {
3756 		reason = CLD_EXITED;
3757 		status = p->p_xexit;
3758 	}
3759 	/*
3760 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3761 	 * done in exit1().
3762 	 */
3763 	sigparent(p, reason, status);
3764 }
3765 
3766 #define	MAX_NUM_CORE_FILES 100000
3767 #ifndef NUM_CORE_FILES
3768 #define	NUM_CORE_FILES 5
3769 #endif
3770 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3771 static int num_cores = NUM_CORE_FILES;
3772 
3773 static int
sysctl_debug_num_cores_check(SYSCTL_HANDLER_ARGS)3774 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3775 {
3776 	int error;
3777 	int new_val;
3778 
3779 	new_val = num_cores;
3780 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3781 	if (error != 0 || req->newptr == NULL)
3782 		return (error);
3783 	if (new_val > MAX_NUM_CORE_FILES)
3784 		new_val = MAX_NUM_CORE_FILES;
3785 	if (new_val < 0)
3786 		new_val = 0;
3787 	num_cores = new_val;
3788 	return (0);
3789 }
3790 SYSCTL_PROC(_debug, OID_AUTO, ncores,
3791     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
3792     sysctl_debug_num_cores_check, "I",
3793     "Maximum number of generated process corefiles while using index format");
3794 
3795 #define	GZIP_SUFFIX	".gz"
3796 #define	ZSTD_SUFFIX	".zst"
3797 
3798 int compress_user_cores = 0;
3799 
3800 static int
sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)3801 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3802 {
3803 	int error, val;
3804 
3805 	val = compress_user_cores;
3806 	error = sysctl_handle_int(oidp, &val, 0, req);
3807 	if (error != 0 || req->newptr == NULL)
3808 		return (error);
3809 	if (val != 0 && !compressor_avail(val))
3810 		return (EINVAL);
3811 	compress_user_cores = val;
3812 	return (error);
3813 }
3814 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores,
3815     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, 0, sizeof(int),
3816     sysctl_compress_user_cores, "I",
3817     "Enable compression of user corefiles ("
3818     __XSTRING(COMPRESS_GZIP) " = gzip, "
3819     __XSTRING(COMPRESS_ZSTD) " = zstd)");
3820 
3821 int compress_user_cores_level = 6;
3822 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3823     &compress_user_cores_level, 0,
3824     "Corefile compression level");
3825 
3826 /*
3827  * Protect the access to corefilename[] by allproc_lock.
3828  */
3829 #define	corefilename_lock	allproc_lock
3830 
3831 static char corefilename[MAXPATHLEN] = {"%N.core"};
3832 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3833 
3834 static int
sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)3835 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3836 {
3837 	int error;
3838 
3839 	sx_xlock(&corefilename_lock);
3840 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3841 	    req);
3842 	sx_xunlock(&corefilename_lock);
3843 
3844 	return (error);
3845 }
3846 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3847     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3848     "Process corefile name format string");
3849 
3850 static void
vnode_close_locked(struct thread * td,struct vnode * vp)3851 vnode_close_locked(struct thread *td, struct vnode *vp)
3852 {
3853 
3854 	VOP_UNLOCK(vp);
3855 	vn_close(vp, FWRITE, td->td_ucred, td);
3856 }
3857 
3858 /*
3859  * If the core format has a %I in it, then we need to check
3860  * for existing corefiles before defining a name.
3861  * To do this we iterate over 0..ncores to find a
3862  * non-existing core file name to use. If all core files are
3863  * already used we choose the oldest one.
3864  */
3865 static int
corefile_open_last(struct thread * td,char * name,int indexpos,int indexlen,int ncores,struct vnode ** vpp)3866 corefile_open_last(struct thread *td, char *name, int indexpos,
3867     int indexlen, int ncores, struct vnode **vpp)
3868 {
3869 	struct vnode *oldvp, *nextvp, *vp;
3870 	struct vattr vattr;
3871 	struct nameidata nd;
3872 	int error, i, flags, oflags, cmode;
3873 	char ch;
3874 	struct timespec lasttime;
3875 
3876 	nextvp = oldvp = NULL;
3877 	cmode = S_IRUSR | S_IWUSR;
3878 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3879 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3880 
3881 	for (i = 0; i < ncores; i++) {
3882 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3883 
3884 		ch = name[indexpos + indexlen];
3885 		(void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3886 		    i);
3887 		name[indexpos + indexlen] = ch;
3888 
3889 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
3890 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3891 		    NULL);
3892 		if (error != 0)
3893 			break;
3894 
3895 		vp = nd.ni_vp;
3896 		NDFREE_PNBUF(&nd);
3897 		if ((flags & O_CREAT) == O_CREAT) {
3898 			nextvp = vp;
3899 			break;
3900 		}
3901 
3902 		error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3903 		if (error != 0) {
3904 			vnode_close_locked(td, vp);
3905 			break;
3906 		}
3907 
3908 		if (oldvp == NULL ||
3909 		    lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3910 		    (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3911 		    lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3912 			if (oldvp != NULL)
3913 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3914 			oldvp = vp;
3915 			VOP_UNLOCK(oldvp);
3916 			lasttime = vattr.va_mtime;
3917 		} else {
3918 			vnode_close_locked(td, vp);
3919 		}
3920 	}
3921 
3922 	if (oldvp != NULL) {
3923 		if (nextvp == NULL) {
3924 			if ((td->td_proc->p_flag & P_SUGID) != 0) {
3925 				error = EFAULT;
3926 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3927 			} else {
3928 				nextvp = oldvp;
3929 				error = vn_lock(nextvp, LK_EXCLUSIVE);
3930 				if (error != 0) {
3931 					vn_close(nextvp, FWRITE, td->td_ucred,
3932 					    td);
3933 					nextvp = NULL;
3934 				}
3935 			}
3936 		} else {
3937 			vn_close(oldvp, FWRITE, td->td_ucred, td);
3938 		}
3939 	}
3940 	if (error != 0) {
3941 		if (nextvp != NULL)
3942 			vnode_close_locked(td, oldvp);
3943 	} else {
3944 		*vpp = nextvp;
3945 	}
3946 
3947 	return (error);
3948 }
3949 
3950 /*
3951  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3952  * Expand the name described in corefilename, using name, uid, and pid
3953  * and open/create core file.
3954  * corefilename is a printf-like string, with three format specifiers:
3955  *	%N	name of process ("name")
3956  *	%P	process id (pid)
3957  *	%U	user id (uid)
3958  * For example, "%N.core" is the default; they can be disabled completely
3959  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3960  * This is controlled by the sysctl variable kern.corefile (see above).
3961  */
3962 static int
corefile_open(const char * comm,uid_t uid,pid_t pid,struct thread * td,int compress,int signum,struct vnode ** vpp,char ** namep)3963 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3964     int compress, int signum, struct vnode **vpp, char **namep)
3965 {
3966 	struct sbuf sb;
3967 	struct nameidata nd;
3968 	const char *format;
3969 	char *hostname, *name;
3970 	int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
3971 
3972 	hostname = NULL;
3973 	format = corefilename;
3974 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3975 	indexlen = 0;
3976 	indexpos = -1;
3977 	ncores = num_cores;
3978 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3979 	sx_slock(&corefilename_lock);
3980 	for (i = 0; format[i] != '\0'; i++) {
3981 		switch (format[i]) {
3982 		case '%':	/* Format character */
3983 			i++;
3984 			switch (format[i]) {
3985 			case '%':
3986 				sbuf_putc(&sb, '%');
3987 				break;
3988 			case 'H':	/* hostname */
3989 				if (hostname == NULL) {
3990 					hostname = malloc(MAXHOSTNAMELEN,
3991 					    M_TEMP, M_WAITOK);
3992 				}
3993 				getcredhostname(td->td_ucred, hostname,
3994 				    MAXHOSTNAMELEN);
3995 				sbuf_printf(&sb, "%s", hostname);
3996 				break;
3997 			case 'I':	/* autoincrementing index */
3998 				if (indexpos != -1) {
3999 					sbuf_printf(&sb, "%%I");
4000 					break;
4001 				}
4002 
4003 				indexpos = sbuf_len(&sb);
4004 				sbuf_printf(&sb, "%u", ncores - 1);
4005 				indexlen = sbuf_len(&sb) - indexpos;
4006 				break;
4007 			case 'N':	/* process name */
4008 				sbuf_printf(&sb, "%s", comm);
4009 				break;
4010 			case 'P':	/* process id */
4011 				sbuf_printf(&sb, "%u", pid);
4012 				break;
4013 			case 'S':	/* signal number */
4014 				sbuf_printf(&sb, "%i", signum);
4015 				break;
4016 			case 'U':	/* user id */
4017 				sbuf_printf(&sb, "%u", uid);
4018 				break;
4019 			default:
4020 				log(LOG_ERR,
4021 				    "Unknown format character %c in "
4022 				    "corename `%s'\n", format[i], format);
4023 				break;
4024 			}
4025 			break;
4026 		default:
4027 			sbuf_putc(&sb, format[i]);
4028 			break;
4029 		}
4030 	}
4031 	sx_sunlock(&corefilename_lock);
4032 	free(hostname, M_TEMP);
4033 	if (compress == COMPRESS_GZIP)
4034 		sbuf_printf(&sb, GZIP_SUFFIX);
4035 	else if (compress == COMPRESS_ZSTD)
4036 		sbuf_printf(&sb, ZSTD_SUFFIX);
4037 	if (sbuf_error(&sb) != 0) {
4038 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
4039 		    "long\n", (long)pid, comm, (u_long)uid);
4040 		sbuf_delete(&sb);
4041 		free(name, M_TEMP);
4042 		return (ENOMEM);
4043 	}
4044 	sbuf_finish(&sb);
4045 	sbuf_delete(&sb);
4046 
4047 	if (indexpos != -1) {
4048 		error = corefile_open_last(td, name, indexpos, indexlen, ncores,
4049 		    vpp);
4050 		if (error != 0) {
4051 			log(LOG_ERR,
4052 			    "pid %d (%s), uid (%u):  Path `%s' failed "
4053 			    "on initial open test, error = %d\n",
4054 			    pid, comm, uid, name, error);
4055 		}
4056 	} else {
4057 		cmode = S_IRUSR | S_IWUSR;
4058 		oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
4059 		    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
4060 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
4061 		if ((td->td_proc->p_flag & P_SUGID) != 0)
4062 			flags |= O_EXCL;
4063 
4064 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
4065 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
4066 		    NULL);
4067 		if (error == 0) {
4068 			*vpp = nd.ni_vp;
4069 			NDFREE_PNBUF(&nd);
4070 		}
4071 	}
4072 
4073 	if (error != 0) {
4074 #ifdef AUDIT
4075 		audit_proc_coredump(td, name, error);
4076 #endif
4077 		free(name, M_TEMP);
4078 		return (error);
4079 	}
4080 	*namep = name;
4081 	return (0);
4082 }
4083 
4084 /*
4085  * Dump a process' core.  The main routine does some
4086  * policy checking, and creates the name of the coredump;
4087  * then it passes on a vnode and a size limit to the process-specific
4088  * coredump routine if there is one; if there _is not_ one, it returns
4089  * ENOSYS; otherwise it returns the error from the process-specific routine.
4090  */
4091 
4092 static int
coredump(struct thread * td)4093 coredump(struct thread *td)
4094 {
4095 	struct proc *p = td->td_proc;
4096 	struct ucred *cred = td->td_ucred;
4097 	struct vnode *vp;
4098 	struct flock lf;
4099 	struct vattr vattr;
4100 	size_t fullpathsize;
4101 	int error, error1, locked;
4102 	char *name;			/* name of corefile */
4103 	void *rl_cookie;
4104 	off_t limit;
4105 	char *fullpath, *freepath = NULL;
4106 	struct sbuf *sb;
4107 
4108 	PROC_LOCK_ASSERT(p, MA_OWNED);
4109 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
4110 
4111 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
4112 	    (p->p_flag2 & P2_NOTRACE) != 0) {
4113 		PROC_UNLOCK(p);
4114 		return (EFAULT);
4115 	}
4116 
4117 	/*
4118 	 * Note that the bulk of limit checking is done after
4119 	 * the corefile is created.  The exception is if the limit
4120 	 * for corefiles is 0, in which case we don't bother
4121 	 * creating the corefile at all.  This layout means that
4122 	 * a corefile is truncated instead of not being created,
4123 	 * if it is larger than the limit.
4124 	 */
4125 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
4126 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
4127 		PROC_UNLOCK(p);
4128 		return (EFBIG);
4129 	}
4130 	PROC_UNLOCK(p);
4131 
4132 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
4133 	    compress_user_cores, p->p_sig, &vp, &name);
4134 	if (error != 0)
4135 		return (error);
4136 
4137 	/*
4138 	 * Don't dump to non-regular files or files with links.
4139 	 * Do not dump into system files. Effective user must own the corefile.
4140 	 */
4141 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
4142 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
4143 	    vattr.va_uid != cred->cr_uid) {
4144 		VOP_UNLOCK(vp);
4145 		error = EFAULT;
4146 		goto out;
4147 	}
4148 
4149 	VOP_UNLOCK(vp);
4150 
4151 	/* Postpone other writers, including core dumps of other processes. */
4152 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
4153 
4154 	lf.l_whence = SEEK_SET;
4155 	lf.l_start = 0;
4156 	lf.l_len = 0;
4157 	lf.l_type = F_WRLCK;
4158 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
4159 
4160 	VATTR_NULL(&vattr);
4161 	vattr.va_size = 0;
4162 	if (set_core_nodump_flag)
4163 		vattr.va_flags = UF_NODUMP;
4164 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4165 	VOP_SETATTR(vp, &vattr, cred);
4166 	VOP_UNLOCK(vp);
4167 	PROC_LOCK(p);
4168 	p->p_acflag |= ACORE;
4169 	PROC_UNLOCK(p);
4170 
4171 	if (p->p_sysent->sv_coredump != NULL) {
4172 		error = p->p_sysent->sv_coredump(td, vp, limit, 0);
4173 	} else {
4174 		error = ENOSYS;
4175 	}
4176 
4177 	if (locked) {
4178 		lf.l_type = F_UNLCK;
4179 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
4180 	}
4181 	vn_rangelock_unlock(vp, rl_cookie);
4182 
4183 	/*
4184 	 * Notify the userland helper that a process triggered a core dump.
4185 	 * This allows the helper to run an automated debugging session.
4186 	 */
4187 	if (error != 0 || coredump_devctl == 0)
4188 		goto out;
4189 	sb = sbuf_new_auto();
4190 	if (vn_fullpath_global(p->p_textvp, &fullpath, &freepath) != 0)
4191 		goto out2;
4192 	sbuf_printf(sb, "comm=\"");
4193 	devctl_safe_quote_sb(sb, fullpath);
4194 	free(freepath, M_TEMP);
4195 	sbuf_printf(sb, "\" core=\"");
4196 
4197 	/*
4198 	 * We can't lookup core file vp directly. When we're replacing a core, and
4199 	 * other random times, we flush the name cache, so it will fail. Instead,
4200 	 * if the path of the core is relative, add the current dir in front if it.
4201 	 */
4202 	if (name[0] != '/') {
4203 		fullpathsize = MAXPATHLEN;
4204 		freepath = malloc(fullpathsize, M_TEMP, M_WAITOK);
4205 		if (vn_getcwd(freepath, &fullpath, &fullpathsize) != 0) {
4206 			free(freepath, M_TEMP);
4207 			goto out2;
4208 		}
4209 		devctl_safe_quote_sb(sb, fullpath);
4210 		free(freepath, M_TEMP);
4211 		sbuf_putc(sb, '/');
4212 	}
4213 	devctl_safe_quote_sb(sb, name);
4214 	sbuf_printf(sb, "\"");
4215 	if (sbuf_finish(sb) == 0)
4216 		devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
4217 out2:
4218 	sbuf_delete(sb);
4219 out:
4220 	error1 = vn_close(vp, FWRITE, cred, td);
4221 	if (error == 0)
4222 		error = error1;
4223 #ifdef AUDIT
4224 	audit_proc_coredump(td, name, error);
4225 #endif
4226 	free(name, M_TEMP);
4227 	return (error);
4228 }
4229 
4230 /*
4231  * Nonexistent system call-- signal process (may want to handle it).  Flag
4232  * error in case process won't see signal immediately (blocked or ignored).
4233  */
4234 #ifndef _SYS_SYSPROTO_H_
4235 struct nosys_args {
4236 	int	dummy;
4237 };
4238 #endif
4239 /* ARGSUSED */
4240 int
nosys(struct thread * td,struct nosys_args * args)4241 nosys(struct thread *td, struct nosys_args *args)
4242 {
4243 	struct proc *p;
4244 
4245 	p = td->td_proc;
4246 
4247 	if (SV_PROC_FLAG(p, SV_SIGSYS) != 0 && kern_signosys) {
4248 		PROC_LOCK(p);
4249 		tdsignal(td, SIGSYS);
4250 		PROC_UNLOCK(p);
4251 	}
4252 	if (kern_lognosys == 1 || kern_lognosys == 3) {
4253 		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4254 		    td->td_sa.code);
4255 	}
4256 	if (kern_lognosys == 2 || kern_lognosys == 3 ||
4257 	    (p->p_pid == 1 && (kern_lognosys & 3) == 0)) {
4258 		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4259 		    td->td_sa.code);
4260 	}
4261 	return (ENOSYS);
4262 }
4263 
4264 /*
4265  * Send a SIGIO or SIGURG signal to a process or process group using stored
4266  * credentials rather than those of the current process.
4267  */
4268 void
pgsigio(struct sigio ** sigiop,int sig,int checkctty)4269 pgsigio(struct sigio **sigiop, int sig, int checkctty)
4270 {
4271 	ksiginfo_t ksi;
4272 	struct sigio *sigio;
4273 
4274 	ksiginfo_init(&ksi);
4275 	ksi.ksi_signo = sig;
4276 	ksi.ksi_code = SI_KERNEL;
4277 
4278 	SIGIO_LOCK();
4279 	sigio = *sigiop;
4280 	if (sigio == NULL) {
4281 		SIGIO_UNLOCK();
4282 		return;
4283 	}
4284 	if (sigio->sio_pgid > 0) {
4285 		PROC_LOCK(sigio->sio_proc);
4286 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
4287 			kern_psignal(sigio->sio_proc, sig);
4288 		PROC_UNLOCK(sigio->sio_proc);
4289 	} else if (sigio->sio_pgid < 0) {
4290 		struct proc *p;
4291 
4292 		PGRP_LOCK(sigio->sio_pgrp);
4293 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
4294 			PROC_LOCK(p);
4295 			if (p->p_state == PRS_NORMAL &&
4296 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
4297 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
4298 				kern_psignal(p, sig);
4299 			PROC_UNLOCK(p);
4300 		}
4301 		PGRP_UNLOCK(sigio->sio_pgrp);
4302 	}
4303 	SIGIO_UNLOCK();
4304 }
4305 
4306 static int
filt_sigattach(struct knote * kn)4307 filt_sigattach(struct knote *kn)
4308 {
4309 	struct proc *p = curproc;
4310 
4311 	kn->kn_ptr.p_proc = p;
4312 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
4313 
4314 	knlist_add(p->p_klist, kn, 0);
4315 
4316 	return (0);
4317 }
4318 
4319 static void
filt_sigdetach(struct knote * kn)4320 filt_sigdetach(struct knote *kn)
4321 {
4322 	knlist_remove(kn->kn_knlist, kn, 0);
4323 }
4324 
4325 /*
4326  * signal knotes are shared with proc knotes, so we apply a mask to
4327  * the hint in order to differentiate them from process hints.  This
4328  * could be avoided by using a signal-specific knote list, but probably
4329  * isn't worth the trouble.
4330  */
4331 static int
filt_signal(struct knote * kn,long hint)4332 filt_signal(struct knote *kn, long hint)
4333 {
4334 
4335 	if (hint & NOTE_SIGNAL) {
4336 		hint &= ~NOTE_SIGNAL;
4337 
4338 		if (kn->kn_id == hint)
4339 			kn->kn_data++;
4340 	}
4341 	return (kn->kn_data != 0);
4342 }
4343 
4344 struct sigacts *
sigacts_alloc(void)4345 sigacts_alloc(void)
4346 {
4347 	struct sigacts *ps;
4348 
4349 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
4350 	refcount_init(&ps->ps_refcnt, 1);
4351 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
4352 	return (ps);
4353 }
4354 
4355 void
sigacts_free(struct sigacts * ps)4356 sigacts_free(struct sigacts *ps)
4357 {
4358 
4359 	if (refcount_release(&ps->ps_refcnt) == 0)
4360 		return;
4361 	mtx_destroy(&ps->ps_mtx);
4362 	free(ps, M_SUBPROC);
4363 }
4364 
4365 struct sigacts *
sigacts_hold(struct sigacts * ps)4366 sigacts_hold(struct sigacts *ps)
4367 {
4368 
4369 	refcount_acquire(&ps->ps_refcnt);
4370 	return (ps);
4371 }
4372 
4373 void
sigacts_copy(struct sigacts * dest,struct sigacts * src)4374 sigacts_copy(struct sigacts *dest, struct sigacts *src)
4375 {
4376 
4377 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
4378 	mtx_lock(&src->ps_mtx);
4379 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
4380 	mtx_unlock(&src->ps_mtx);
4381 }
4382 
4383 int
sigacts_shared(struct sigacts * ps)4384 sigacts_shared(struct sigacts *ps)
4385 {
4386 
4387 	return (ps->ps_refcnt > 1);
4388 }
4389 
4390 void
sig_drop_caught(struct proc * p)4391 sig_drop_caught(struct proc *p)
4392 {
4393 	int sig;
4394 	struct sigacts *ps;
4395 
4396 	ps = p->p_sigacts;
4397 	PROC_LOCK_ASSERT(p, MA_OWNED);
4398 	mtx_assert(&ps->ps_mtx, MA_OWNED);
4399 	SIG_FOREACH(sig, &ps->ps_sigcatch) {
4400 		sigdflt(ps, sig);
4401 		if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
4402 			sigqueue_delete_proc(p, sig);
4403 	}
4404 }
4405 
4406 static void
sigfastblock_failed(struct thread * td,bool sendsig,bool write)4407 sigfastblock_failed(struct thread *td, bool sendsig, bool write)
4408 {
4409 	ksiginfo_t ksi;
4410 
4411 	/*
4412 	 * Prevent further fetches and SIGSEGVs, allowing thread to
4413 	 * issue syscalls despite corruption.
4414 	 */
4415 	sigfastblock_clear(td);
4416 
4417 	if (!sendsig)
4418 		return;
4419 	ksiginfo_init_trap(&ksi);
4420 	ksi.ksi_signo = SIGSEGV;
4421 	ksi.ksi_code = write ? SEGV_ACCERR : SEGV_MAPERR;
4422 	ksi.ksi_addr = td->td_sigblock_ptr;
4423 	trapsignal(td, &ksi);
4424 }
4425 
4426 static bool
sigfastblock_fetch_sig(struct thread * td,bool sendsig,uint32_t * valp)4427 sigfastblock_fetch_sig(struct thread *td, bool sendsig, uint32_t *valp)
4428 {
4429 	uint32_t res;
4430 
4431 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4432 		return (true);
4433 	if (fueword32((void *)td->td_sigblock_ptr, &res) == -1) {
4434 		sigfastblock_failed(td, sendsig, false);
4435 		return (false);
4436 	}
4437 	*valp = res;
4438 	td->td_sigblock_val = res & ~SIGFASTBLOCK_FLAGS;
4439 	return (true);
4440 }
4441 
4442 static void
sigfastblock_resched(struct thread * td,bool resched)4443 sigfastblock_resched(struct thread *td, bool resched)
4444 {
4445 	struct proc *p;
4446 
4447 	if (resched) {
4448 		p = td->td_proc;
4449 		PROC_LOCK(p);
4450 		reschedule_signals(p, td->td_sigmask, 0);
4451 		PROC_UNLOCK(p);
4452 	}
4453 	ast_sched(td, TDA_SIG);
4454 }
4455 
4456 int
sys_sigfastblock(struct thread * td,struct sigfastblock_args * uap)4457 sys_sigfastblock(struct thread *td, struct sigfastblock_args *uap)
4458 {
4459 	struct proc *p;
4460 	int error, res;
4461 	uint32_t oldval;
4462 
4463 	error = 0;
4464 	p = td->td_proc;
4465 	switch (uap->cmd) {
4466 	case SIGFASTBLOCK_SETPTR:
4467 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
4468 			error = EBUSY;
4469 			break;
4470 		}
4471 		if (((uintptr_t)(uap->ptr) & (sizeof(uint32_t) - 1)) != 0) {
4472 			error = EINVAL;
4473 			break;
4474 		}
4475 		td->td_pflags |= TDP_SIGFASTBLOCK;
4476 		td->td_sigblock_ptr = uap->ptr;
4477 		break;
4478 
4479 	case SIGFASTBLOCK_UNBLOCK:
4480 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4481 			error = EINVAL;
4482 			break;
4483 		}
4484 
4485 		for (;;) {
4486 			res = casueword32(td->td_sigblock_ptr,
4487 			    SIGFASTBLOCK_PEND, &oldval, 0);
4488 			if (res == -1) {
4489 				error = EFAULT;
4490 				sigfastblock_failed(td, false, true);
4491 				break;
4492 			}
4493 			if (res == 0)
4494 				break;
4495 			MPASS(res == 1);
4496 			if (oldval != SIGFASTBLOCK_PEND) {
4497 				error = EBUSY;
4498 				break;
4499 			}
4500 			error = thread_check_susp(td, false);
4501 			if (error != 0)
4502 				break;
4503 		}
4504 		if (error != 0)
4505 			break;
4506 
4507 		/*
4508 		 * td_sigblock_val is cleared there, but not on a
4509 		 * syscall exit.  The end effect is that a single
4510 		 * interruptible sleep, while user sigblock word is
4511 		 * set, might return EINTR or ERESTART to usermode
4512 		 * without delivering signal.  All further sleeps,
4513 		 * until userspace clears the word and does
4514 		 * sigfastblock(UNBLOCK), observe current word and no
4515 		 * longer get interrupted.  It is slight
4516 		 * non-conformance, with alternative to have read the
4517 		 * sigblock word on each syscall entry.
4518 		 */
4519 		td->td_sigblock_val = 0;
4520 
4521 		/*
4522 		 * Rely on normal ast mechanism to deliver pending
4523 		 * signals to current thread.  But notify others about
4524 		 * fake unblock.
4525 		 */
4526 		sigfastblock_resched(td, error == 0 && p->p_numthreads != 1);
4527 
4528 		break;
4529 
4530 	case SIGFASTBLOCK_UNSETPTR:
4531 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4532 			error = EINVAL;
4533 			break;
4534 		}
4535 		if (!sigfastblock_fetch_sig(td, false, &oldval)) {
4536 			error = EFAULT;
4537 			break;
4538 		}
4539 		if (oldval != 0 && oldval != SIGFASTBLOCK_PEND) {
4540 			error = EBUSY;
4541 			break;
4542 		}
4543 		sigfastblock_clear(td);
4544 		break;
4545 
4546 	default:
4547 		error = EINVAL;
4548 		break;
4549 	}
4550 	return (error);
4551 }
4552 
4553 void
sigfastblock_clear(struct thread * td)4554 sigfastblock_clear(struct thread *td)
4555 {
4556 	bool resched;
4557 
4558 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4559 		return;
4560 	td->td_sigblock_val = 0;
4561 	resched = (td->td_pflags & TDP_SIGFASTPENDING) != 0 ||
4562 	    SIGPENDING(td);
4563 	td->td_pflags &= ~(TDP_SIGFASTBLOCK | TDP_SIGFASTPENDING);
4564 	sigfastblock_resched(td, resched);
4565 }
4566 
4567 void
sigfastblock_fetch(struct thread * td)4568 sigfastblock_fetch(struct thread *td)
4569 {
4570 	uint32_t val;
4571 
4572 	(void)sigfastblock_fetch_sig(td, true, &val);
4573 }
4574 
4575 static void
sigfastblock_setpend1(struct thread * td)4576 sigfastblock_setpend1(struct thread *td)
4577 {
4578 	int res;
4579 	uint32_t oldval;
4580 
4581 	if ((td->td_pflags & TDP_SIGFASTPENDING) == 0)
4582 		return;
4583 	res = fueword32((void *)td->td_sigblock_ptr, &oldval);
4584 	if (res == -1) {
4585 		sigfastblock_failed(td, true, false);
4586 		return;
4587 	}
4588 	for (;;) {
4589 		res = casueword32(td->td_sigblock_ptr, oldval, &oldval,
4590 		    oldval | SIGFASTBLOCK_PEND);
4591 		if (res == -1) {
4592 			sigfastblock_failed(td, true, true);
4593 			return;
4594 		}
4595 		if (res == 0) {
4596 			td->td_sigblock_val = oldval & ~SIGFASTBLOCK_FLAGS;
4597 			td->td_pflags &= ~TDP_SIGFASTPENDING;
4598 			break;
4599 		}
4600 		MPASS(res == 1);
4601 		if (thread_check_susp(td, false) != 0)
4602 			break;
4603 	}
4604 }
4605 
4606 static void
sigfastblock_setpend(struct thread * td,bool resched)4607 sigfastblock_setpend(struct thread *td, bool resched)
4608 {
4609 	struct proc *p;
4610 
4611 	sigfastblock_setpend1(td);
4612 	if (resched) {
4613 		p = td->td_proc;
4614 		PROC_LOCK(p);
4615 		reschedule_signals(p, fastblock_mask, SIGPROCMASK_FASTBLK);
4616 		PROC_UNLOCK(p);
4617 	}
4618 }
4619