xref: /freebsd/sys/kern/kern_sig.c (revision dabf006a638fdc44cdcf69731de8ac83959db731)
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 
37 #include "opt_capsicum.h"
38 #include "opt_ktrace.h"
39 
40 #include <sys/param.h>
41 #include <sys/capsicum.h>
42 #include <sys/ctype.h>
43 #include <sys/systm.h>
44 #include <sys/signalvar.h>
45 #include <sys/vnode.h>
46 #include <sys/acct.h>
47 #include <sys/capsicum.h>
48 #include <sys/compressor.h>
49 #include <sys/condvar.h>
50 #include <sys/devctl.h>
51 #include <sys/event.h>
52 #include <sys/fcntl.h>
53 #include <sys/imgact.h>
54 #include <sys/jail.h>
55 #include <sys/kernel.h>
56 #include <sys/ktr.h>
57 #include <sys/ktrace.h>
58 #include <sys/limits.h>
59 #include <sys/lock.h>
60 #include <sys/malloc.h>
61 #include <sys/mutex.h>
62 #include <sys/refcount.h>
63 #include <sys/namei.h>
64 #include <sys/proc.h>
65 #include <sys/procdesc.h>
66 #include <sys/ptrace.h>
67 #include <sys/posix4.h>
68 #include <sys/racct.h>
69 #include <sys/resourcevar.h>
70 #include <sys/sdt.h>
71 #include <sys/sbuf.h>
72 #include <sys/sleepqueue.h>
73 #include <sys/smp.h>
74 #include <sys/stat.h>
75 #include <sys/sx.h>
76 #include <sys/syscall.h>
77 #include <sys/syscallsubr.h>
78 #include <sys/sysctl.h>
79 #include <sys/sysent.h>
80 #include <sys/syslog.h>
81 #include <sys/sysproto.h>
82 #include <sys/timers.h>
83 #include <sys/unistd.h>
84 #include <sys/vmmeter.h>
85 #include <sys/wait.h>
86 #include <vm/vm.h>
87 #include <vm/vm_extern.h>
88 #include <vm/uma.h>
89 
90 #include <machine/cpu.h>
91 
92 #include <security/audit/audit.h>
93 
94 #define	ONSIG	32		/* NSIG for osig* syscalls.  XXX. */
95 
96 SDT_PROVIDER_DECLARE(proc);
97 SDT_PROBE_DEFINE3(proc, , , signal__send,
98     "struct thread *", "struct proc *", "int");
99 SDT_PROBE_DEFINE2(proc, , , signal__clear,
100     "int", "ksiginfo_t *");
101 SDT_PROBE_DEFINE3(proc, , , signal__discard,
102     "struct thread *", "struct proc *", "int");
103 
104 static int	coredump(struct thread *);
105 static int	killpg1(struct thread *td, int sig, int pgid, int all,
106 		    ksiginfo_t *ksi);
107 static int	issignal(struct thread *td);
108 static void	reschedule_signals(struct proc *p, sigset_t block, int flags);
109 static int	sigprop(int sig);
110 static void	tdsigwakeup(struct thread *, int, sig_t, int);
111 static void	sig_suspend_threads(struct thread *, struct proc *);
112 static int	filt_sigattach(struct knote *kn);
113 static void	filt_sigdetach(struct knote *kn);
114 static int	filt_signal(struct knote *kn, long hint);
115 static struct thread *sigtd(struct proc *p, int sig, bool fast_sigblock);
116 static void	sigqueue_start(void);
117 static void	sigfastblock_setpend(struct thread *td, bool resched);
118 
119 static uma_zone_t	ksiginfo_zone = NULL;
120 const struct filterops sig_filtops = {
121 	.f_isfd = 0,
122 	.f_attach = filt_sigattach,
123 	.f_detach = filt_sigdetach,
124 	.f_event = filt_signal,
125 };
126 
127 static int	kern_logsigexit = 1;
128 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
129     &kern_logsigexit, 0,
130     "Log processes quitting on abnormal signals to syslog(3)");
131 
132 static int	kern_forcesigexit = 1;
133 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
134     &kern_forcesigexit, 0, "Force trap signal to be handled");
135 
136 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
137     "POSIX real time signal");
138 
139 static int	max_pending_per_proc = 128;
140 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
141     &max_pending_per_proc, 0, "Max pending signals per proc");
142 
143 static int	preallocate_siginfo = 1024;
144 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
145     &preallocate_siginfo, 0, "Preallocated signal memory size");
146 
147 static int	signal_overflow = 0;
148 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
149     &signal_overflow, 0, "Number of signals overflew");
150 
151 static int	signal_alloc_fail = 0;
152 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
153     &signal_alloc_fail, 0, "signals failed to be allocated");
154 
155 static int	kern_lognosys = 0;
156 SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
157     "Log invalid syscalls");
158 
159 static int	kern_signosys = 1;
160 SYSCTL_INT(_kern, OID_AUTO, signosys, CTLFLAG_RWTUN, &kern_signosys, 0,
161     "Send SIGSYS on return from invalid syscall");
162 
163 __read_frequently bool sigfastblock_fetch_always = false;
164 SYSCTL_BOOL(_kern, OID_AUTO, sigfastblock_fetch_always, CTLFLAG_RWTUN,
165     &sigfastblock_fetch_always, 0,
166     "Fetch sigfastblock word on each syscall entry for proper "
167     "blocking semantic");
168 
169 static bool	kern_sig_discard_ign = true;
170 SYSCTL_BOOL(_kern, OID_AUTO, sig_discard_ign, CTLFLAG_RWTUN,
171     &kern_sig_discard_ign, 0,
172     "Discard ignored signals on delivery, otherwise queue them to "
173     "the target queue");
174 
175 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
176 
177 /*
178  * Policy -- Can ucred cr1 send SIGIO to process cr2?
179  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
180  * in the right situations.
181  */
182 #define CANSIGIO(cr1, cr2) \
183 	((cr1)->cr_uid == 0 || \
184 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
185 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
186 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
187 	    (cr1)->cr_uid == (cr2)->cr_uid)
188 
189 static int	sugid_coredump;
190 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
191     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
192 
193 static int	capmode_coredump;
194 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
195     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
196 
197 static int	do_coredump = 1;
198 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
199 	&do_coredump, 0, "Enable/Disable coredumps");
200 
201 static int	set_core_nodump_flag = 0;
202 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
203 	0, "Enable setting the NODUMP flag on coredump files");
204 
205 static int	coredump_devctl = 0;
206 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
207 	0, "Generate a devctl notification when processes coredump");
208 
209 /*
210  * Signal properties and actions.
211  * The array below categorizes the signals and their default actions
212  * according to the following properties:
213  */
214 #define	SIGPROP_KILL		0x01	/* terminates process by default */
215 #define	SIGPROP_CORE		0x02	/* ditto and coredumps */
216 #define	SIGPROP_STOP		0x04	/* suspend process */
217 #define	SIGPROP_TTYSTOP		0x08	/* ditto, from tty */
218 #define	SIGPROP_IGNORE		0x10	/* ignore by default */
219 #define	SIGPROP_CONT		0x20	/* continue if suspended */
220 
221 static const int sigproptbl[NSIG] = {
222 	[SIGHUP] =	SIGPROP_KILL,
223 	[SIGINT] =	SIGPROP_KILL,
224 	[SIGQUIT] =	SIGPROP_KILL | SIGPROP_CORE,
225 	[SIGILL] =	SIGPROP_KILL | SIGPROP_CORE,
226 	[SIGTRAP] =	SIGPROP_KILL | SIGPROP_CORE,
227 	[SIGABRT] =	SIGPROP_KILL | SIGPROP_CORE,
228 	[SIGEMT] =	SIGPROP_KILL | SIGPROP_CORE,
229 	[SIGFPE] =	SIGPROP_KILL | SIGPROP_CORE,
230 	[SIGKILL] =	SIGPROP_KILL,
231 	[SIGBUS] =	SIGPROP_KILL | SIGPROP_CORE,
232 	[SIGSEGV] =	SIGPROP_KILL | SIGPROP_CORE,
233 	[SIGSYS] =	SIGPROP_KILL | SIGPROP_CORE,
234 	[SIGPIPE] =	SIGPROP_KILL,
235 	[SIGALRM] =	SIGPROP_KILL,
236 	[SIGTERM] =	SIGPROP_KILL,
237 	[SIGURG] =	SIGPROP_IGNORE,
238 	[SIGSTOP] =	SIGPROP_STOP,
239 	[SIGTSTP] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
240 	[SIGCONT] =	SIGPROP_IGNORE | SIGPROP_CONT,
241 	[SIGCHLD] =	SIGPROP_IGNORE,
242 	[SIGTTIN] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
243 	[SIGTTOU] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
244 	[SIGIO] =	SIGPROP_IGNORE,
245 	[SIGXCPU] =	SIGPROP_KILL,
246 	[SIGXFSZ] =	SIGPROP_KILL,
247 	[SIGVTALRM] =	SIGPROP_KILL,
248 	[SIGPROF] =	SIGPROP_KILL,
249 	[SIGWINCH] =	SIGPROP_IGNORE,
250 	[SIGINFO] =	SIGPROP_IGNORE,
251 	[SIGUSR1] =	SIGPROP_KILL,
252 	[SIGUSR2] =	SIGPROP_KILL,
253 };
254 
255 #define	_SIG_FOREACH_ADVANCE(i, set) ({					\
256 	int __found;							\
257 	for (;;) {							\
258 		if (__bits != 0) {					\
259 			int __sig = ffs(__bits);			\
260 			__bits &= ~(1u << (__sig - 1));			\
261 			sig = __i * sizeof((set)->__bits[0]) * NBBY + __sig; \
262 			__found = 1;					\
263 			break;						\
264 		}							\
265 		if (++__i == _SIG_WORDS) {				\
266 			__found = 0;					\
267 			break;						\
268 		}							\
269 		__bits = (set)->__bits[__i];				\
270 	}								\
271 	__found != 0;							\
272 })
273 
274 #define	SIG_FOREACH(i, set)						\
275 	for (int32_t __i = -1, __bits = 0;				\
276 	    _SIG_FOREACH_ADVANCE(i, set); )				\
277 
278 static sigset_t fastblock_mask;
279 
280 static void
ast_sig(struct thread * td,int tda)281 ast_sig(struct thread *td, int tda)
282 {
283 	struct proc *p;
284 	int old_boundary, sig;
285 	bool resched_sigs;
286 
287 	p = td->td_proc;
288 
289 #ifdef DIAGNOSTIC
290 	if (p->p_numthreads == 1 && (tda & (TDAI(TDA_SIG) |
291 	    TDAI(TDA_AST))) == 0) {
292 		PROC_LOCK(p);
293 		thread_lock(td);
294 		/*
295 		 * Note that TDA_SIG should be re-read from
296 		 * td_ast, since signal might have been delivered
297 		 * after we cleared td_flags above.  This is one of
298 		 * the reason for looping check for AST condition.
299 		 * See comment in userret() about P_PPWAIT.
300 		 */
301 		if ((p->p_flag & P_PPWAIT) == 0 &&
302 		    (td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
303 			if (SIGPENDING(td) && ((tda | td->td_ast) &
304 			    (TDAI(TDA_SIG) | TDAI(TDA_AST))) == 0) {
305 				thread_unlock(td); /* fix dumps */
306 				panic(
307 				    "failed2 to set signal flags for ast p %p "
308 				    "td %p tda %#x td_ast %#x fl %#x",
309 				    p, td, tda, td->td_ast, td->td_flags);
310 			}
311 		}
312 		thread_unlock(td);
313 		PROC_UNLOCK(p);
314 	}
315 #endif
316 
317 	/*
318 	 * Check for signals. Unlocked reads of p_pendingcnt or
319 	 * p_siglist might cause process-directed signal to be handled
320 	 * later.
321 	 */
322 	if ((tda & TDAI(TDA_SIG)) != 0 || p->p_pendingcnt > 0 ||
323 	    !SIGISEMPTY(p->p_siglist)) {
324 		sigfastblock_fetch(td);
325 		PROC_LOCK(p);
326 		old_boundary = ~TDB_BOUNDARY | (td->td_dbgflags & TDB_BOUNDARY);
327 		td->td_dbgflags |= TDB_BOUNDARY;
328 		mtx_lock(&p->p_sigacts->ps_mtx);
329 		while ((sig = cursig(td)) != 0) {
330 			KASSERT(sig >= 0, ("sig %d", sig));
331 			postsig(sig);
332 		}
333 		mtx_unlock(&p->p_sigacts->ps_mtx);
334 		td->td_dbgflags &= old_boundary;
335 		PROC_UNLOCK(p);
336 		resched_sigs = true;
337 	} else {
338 		resched_sigs = false;
339 	}
340 
341 	/*
342 	 * Handle deferred update of the fast sigblock value, after
343 	 * the postsig() loop was performed.
344 	 */
345 	sigfastblock_setpend(td, resched_sigs);
346 }
347 
348 static void
ast_sigsuspend(struct thread * td,int tda __unused)349 ast_sigsuspend(struct thread *td, int tda __unused)
350 {
351 	MPASS((td->td_pflags & TDP_OLDMASK) != 0);
352 	td->td_pflags &= ~TDP_OLDMASK;
353 	kern_sigprocmask(td, SIG_SETMASK, &td->td_oldsigmask, NULL, 0);
354 }
355 
356 static void
sigqueue_start(void)357 sigqueue_start(void)
358 {
359 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
360 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
361 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
362 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
363 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
364 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
365 	SIGFILLSET(fastblock_mask);
366 	SIG_CANTMASK(fastblock_mask);
367 	ast_register(TDA_SIG, ASTR_UNCOND, 0, ast_sig);
368 
369 	/*
370 	 * TDA_PSELECT is for the case where the signal mask should be restored
371 	 * before delivering any signals so that we do not deliver any that are
372 	 * blocked by the normal thread mask.  It is mutually exclusive with
373 	 * TDA_SIGSUSPEND, which should be used if we *do* want to deliver
374 	 * signals that are normally blocked, e.g., if it interrupted our sleep.
375 	 */
376 	ast_register(TDA_PSELECT, ASTR_ASTF_REQUIRED | ASTR_TDP,
377 	    TDP_OLDMASK, ast_sigsuspend);
378 	ast_register(TDA_SIGSUSPEND, ASTR_ASTF_REQUIRED | ASTR_TDP,
379 	    TDP_OLDMASK, ast_sigsuspend);
380 }
381 
382 ksiginfo_t *
ksiginfo_alloc(int mwait)383 ksiginfo_alloc(int mwait)
384 {
385 	MPASS(mwait == M_WAITOK || mwait == M_NOWAIT);
386 
387 	if (ksiginfo_zone == NULL)
388 		return (NULL);
389 	return (uma_zalloc(ksiginfo_zone, mwait | M_ZERO));
390 }
391 
392 void
ksiginfo_free(ksiginfo_t * ksi)393 ksiginfo_free(ksiginfo_t *ksi)
394 {
395 	uma_zfree(ksiginfo_zone, ksi);
396 }
397 
398 static __inline bool
ksiginfo_tryfree(ksiginfo_t * ksi)399 ksiginfo_tryfree(ksiginfo_t *ksi)
400 {
401 	if ((ksi->ksi_flags & KSI_EXT) == 0) {
402 		uma_zfree(ksiginfo_zone, ksi);
403 		return (true);
404 	}
405 	return (false);
406 }
407 
408 void
sigqueue_init(sigqueue_t * list,struct proc * p)409 sigqueue_init(sigqueue_t *list, struct proc *p)
410 {
411 	SIGEMPTYSET(list->sq_signals);
412 	SIGEMPTYSET(list->sq_kill);
413 	SIGEMPTYSET(list->sq_ptrace);
414 	TAILQ_INIT(&list->sq_list);
415 	list->sq_proc = p;
416 	list->sq_flags = SQ_INIT;
417 }
418 
419 /*
420  * Get a signal's ksiginfo.
421  * Return:
422  *	0	-	signal not found
423  *	others	-	signal number
424  */
425 static int
sigqueue_get(sigqueue_t * sq,int signo,ksiginfo_t * si)426 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
427 {
428 	struct proc *p = sq->sq_proc;
429 	struct ksiginfo *ksi, *next;
430 	int count = 0;
431 
432 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
433 
434 	if (!SIGISMEMBER(sq->sq_signals, signo))
435 		return (0);
436 
437 	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
438 		count++;
439 		SIGDELSET(sq->sq_ptrace, signo);
440 		si->ksi_flags |= KSI_PTRACE;
441 	}
442 	if (SIGISMEMBER(sq->sq_kill, signo)) {
443 		count++;
444 		if (count == 1)
445 			SIGDELSET(sq->sq_kill, signo);
446 	}
447 
448 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
449 		if (ksi->ksi_signo == signo) {
450 			if (count == 0) {
451 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
452 				ksi->ksi_sigq = NULL;
453 				ksiginfo_copy(ksi, si);
454 				if (ksiginfo_tryfree(ksi) && p != NULL)
455 					p->p_pendingcnt--;
456 			}
457 			if (++count > 1)
458 				break;
459 		}
460 	}
461 
462 	if (count <= 1)
463 		SIGDELSET(sq->sq_signals, signo);
464 	si->ksi_signo = signo;
465 	return (signo);
466 }
467 
468 void
sigqueue_take(ksiginfo_t * ksi)469 sigqueue_take(ksiginfo_t *ksi)
470 {
471 	struct ksiginfo *kp;
472 	struct proc	*p;
473 	sigqueue_t	*sq;
474 
475 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
476 		return;
477 
478 	p = sq->sq_proc;
479 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
480 	ksi->ksi_sigq = NULL;
481 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
482 		p->p_pendingcnt--;
483 
484 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
485 	     kp = TAILQ_NEXT(kp, ksi_link)) {
486 		if (kp->ksi_signo == ksi->ksi_signo)
487 			break;
488 	}
489 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
490 	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
491 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
492 }
493 
494 static int
sigqueue_add(sigqueue_t * sq,int signo,ksiginfo_t * si)495 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
496 {
497 	struct proc *p = sq->sq_proc;
498 	struct ksiginfo *ksi;
499 	int ret = 0;
500 
501 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
502 
503 	/*
504 	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
505 	 * for these signals.
506 	 */
507 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
508 		SIGADDSET(sq->sq_kill, signo);
509 		goto out_set_bit;
510 	}
511 
512 	/* directly insert the ksi, don't copy it */
513 	if (si->ksi_flags & KSI_INS) {
514 		if (si->ksi_flags & KSI_HEAD)
515 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
516 		else
517 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
518 		si->ksi_sigq = sq;
519 		goto out_set_bit;
520 	}
521 
522 	if (__predict_false(ksiginfo_zone == NULL)) {
523 		SIGADDSET(sq->sq_kill, signo);
524 		goto out_set_bit;
525 	}
526 
527 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
528 		signal_overflow++;
529 		ret = EAGAIN;
530 	} else if ((ksi = ksiginfo_alloc(M_NOWAIT)) == NULL) {
531 		signal_alloc_fail++;
532 		ret = EAGAIN;
533 	} else {
534 		if (p != NULL)
535 			p->p_pendingcnt++;
536 		ksiginfo_copy(si, ksi);
537 		ksi->ksi_signo = signo;
538 		if (si->ksi_flags & KSI_HEAD)
539 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
540 		else
541 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
542 		ksi->ksi_sigq = sq;
543 	}
544 
545 	if (ret != 0) {
546 		if ((si->ksi_flags & KSI_PTRACE) != 0) {
547 			SIGADDSET(sq->sq_ptrace, signo);
548 			ret = 0;
549 			goto out_set_bit;
550 		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
551 		    (si->ksi_flags & KSI_SIGQ) == 0) {
552 			SIGADDSET(sq->sq_kill, signo);
553 			ret = 0;
554 			goto out_set_bit;
555 		}
556 		return (ret);
557 	}
558 
559 out_set_bit:
560 	SIGADDSET(sq->sq_signals, signo);
561 	return (ret);
562 }
563 
564 void
sigqueue_flush(sigqueue_t * sq)565 sigqueue_flush(sigqueue_t *sq)
566 {
567 	struct proc *p = sq->sq_proc;
568 	ksiginfo_t *ksi;
569 
570 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
571 
572 	if (p != NULL)
573 		PROC_LOCK_ASSERT(p, MA_OWNED);
574 
575 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
576 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
577 		ksi->ksi_sigq = NULL;
578 		if (ksiginfo_tryfree(ksi) && p != NULL)
579 			p->p_pendingcnt--;
580 	}
581 
582 	SIGEMPTYSET(sq->sq_signals);
583 	SIGEMPTYSET(sq->sq_kill);
584 	SIGEMPTYSET(sq->sq_ptrace);
585 }
586 
587 static void
sigqueue_move_set(sigqueue_t * src,sigqueue_t * dst,const sigset_t * set)588 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
589 {
590 	sigset_t tmp;
591 	struct proc *p1, *p2;
592 	ksiginfo_t *ksi, *next;
593 
594 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
595 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
596 	p1 = src->sq_proc;
597 	p2 = dst->sq_proc;
598 	/* Move siginfo to target list */
599 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
600 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
601 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
602 			if (p1 != NULL)
603 				p1->p_pendingcnt--;
604 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
605 			ksi->ksi_sigq = dst;
606 			if (p2 != NULL)
607 				p2->p_pendingcnt++;
608 		}
609 	}
610 
611 	/* Move pending bits to target list */
612 	tmp = src->sq_kill;
613 	SIGSETAND(tmp, *set);
614 	SIGSETOR(dst->sq_kill, tmp);
615 	SIGSETNAND(src->sq_kill, tmp);
616 
617 	tmp = src->sq_ptrace;
618 	SIGSETAND(tmp, *set);
619 	SIGSETOR(dst->sq_ptrace, tmp);
620 	SIGSETNAND(src->sq_ptrace, tmp);
621 
622 	tmp = src->sq_signals;
623 	SIGSETAND(tmp, *set);
624 	SIGSETOR(dst->sq_signals, tmp);
625 	SIGSETNAND(src->sq_signals, tmp);
626 }
627 
628 #if 0
629 static void
630 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
631 {
632 	sigset_t set;
633 
634 	SIGEMPTYSET(set);
635 	SIGADDSET(set, signo);
636 	sigqueue_move_set(src, dst, &set);
637 }
638 #endif
639 
640 static void
sigqueue_delete_set(sigqueue_t * sq,const sigset_t * set)641 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
642 {
643 	struct proc *p = sq->sq_proc;
644 	ksiginfo_t *ksi, *next;
645 
646 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
647 
648 	/* Remove siginfo queue */
649 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
650 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
651 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
652 			ksi->ksi_sigq = NULL;
653 			if (ksiginfo_tryfree(ksi) && p != NULL)
654 				p->p_pendingcnt--;
655 		}
656 	}
657 	SIGSETNAND(sq->sq_kill, *set);
658 	SIGSETNAND(sq->sq_ptrace, *set);
659 	SIGSETNAND(sq->sq_signals, *set);
660 }
661 
662 void
sigqueue_delete(sigqueue_t * sq,int signo)663 sigqueue_delete(sigqueue_t *sq, int signo)
664 {
665 	sigset_t set;
666 
667 	SIGEMPTYSET(set);
668 	SIGADDSET(set, signo);
669 	sigqueue_delete_set(sq, &set);
670 }
671 
672 /* Remove a set of signals for a process */
673 static void
sigqueue_delete_set_proc(struct proc * p,const sigset_t * set)674 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
675 {
676 	sigqueue_t worklist;
677 	struct thread *td0;
678 
679 	PROC_LOCK_ASSERT(p, MA_OWNED);
680 
681 	sigqueue_init(&worklist, NULL);
682 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
683 
684 	FOREACH_THREAD_IN_PROC(p, td0)
685 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
686 
687 	sigqueue_flush(&worklist);
688 }
689 
690 void
sigqueue_delete_proc(struct proc * p,int signo)691 sigqueue_delete_proc(struct proc *p, int signo)
692 {
693 	sigset_t set;
694 
695 	SIGEMPTYSET(set);
696 	SIGADDSET(set, signo);
697 	sigqueue_delete_set_proc(p, &set);
698 }
699 
700 static void
sigqueue_delete_stopmask_proc(struct proc * p)701 sigqueue_delete_stopmask_proc(struct proc *p)
702 {
703 	sigset_t set;
704 
705 	SIGEMPTYSET(set);
706 	SIGADDSET(set, SIGSTOP);
707 	SIGADDSET(set, SIGTSTP);
708 	SIGADDSET(set, SIGTTIN);
709 	SIGADDSET(set, SIGTTOU);
710 	sigqueue_delete_set_proc(p, &set);
711 }
712 
713 /*
714  * Determine signal that should be delivered to thread td, the current
715  * thread, 0 if none.  If there is a pending stop signal with default
716  * action, the process stops in issignal().
717  */
718 int
cursig(struct thread * td)719 cursig(struct thread *td)
720 {
721 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
722 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
723 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
724 	return (SIGPENDING(td) ? issignal(td) : 0);
725 }
726 
727 /*
728  * Arrange for ast() to handle unmasked pending signals on return to user
729  * mode.  This must be called whenever a signal is added to td_sigqueue or
730  * unmasked in td_sigmask.
731  */
732 void
signotify(struct thread * td)733 signotify(struct thread *td)
734 {
735 
736 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
737 
738 	if (SIGPENDING(td))
739 		ast_sched(td, TDA_SIG);
740 }
741 
742 /*
743  * Returns 1 (true) if altstack is configured for the thread, and the
744  * passed stack bottom address falls into the altstack range.  Handles
745  * the 43 compat special case where the alt stack size is zero.
746  */
747 int
sigonstack(size_t sp)748 sigonstack(size_t sp)
749 {
750 	struct thread *td;
751 
752 	td = curthread;
753 	if ((td->td_pflags & TDP_ALTSTACK) == 0)
754 		return (0);
755 #if defined(COMPAT_43)
756 	if (SV_PROC_FLAG(td->td_proc, SV_AOUT) && td->td_sigstk.ss_size == 0)
757 		return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
758 #endif
759 	return (sp >= (size_t)td->td_sigstk.ss_sp &&
760 	    sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
761 }
762 
763 static __inline int
sigprop(int sig)764 sigprop(int sig)
765 {
766 
767 	if (sig > 0 && sig < nitems(sigproptbl))
768 		return (sigproptbl[sig]);
769 	return (0);
770 }
771 
772 static bool
sigact_flag_test(const struct sigaction * act,int flag)773 sigact_flag_test(const struct sigaction *act, int flag)
774 {
775 
776 	/*
777 	 * SA_SIGINFO is reset when signal disposition is set to
778 	 * ignore or default.  Other flags are kept according to user
779 	 * settings.
780 	 */
781 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
782 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
783 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
784 }
785 
786 /*
787  * kern_sigaction
788  * sigaction
789  * freebsd4_sigaction
790  * osigaction
791  */
792 int
kern_sigaction(struct thread * td,int sig,const struct sigaction * act,struct sigaction * oact,int flags)793 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
794     struct sigaction *oact, int flags)
795 {
796 	struct sigacts *ps;
797 	struct proc *p = td->td_proc;
798 
799 	if (!_SIG_VALID(sig))
800 		return (EINVAL);
801 	if (act != NULL && act->sa_handler != SIG_DFL &&
802 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
803 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
804 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
805 		return (EINVAL);
806 
807 	PROC_LOCK(p);
808 	ps = p->p_sigacts;
809 	mtx_lock(&ps->ps_mtx);
810 	if (oact) {
811 		memset(oact, 0, sizeof(*oact));
812 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
813 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
814 			oact->sa_flags |= SA_ONSTACK;
815 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
816 			oact->sa_flags |= SA_RESTART;
817 		if (SIGISMEMBER(ps->ps_sigreset, sig))
818 			oact->sa_flags |= SA_RESETHAND;
819 		if (SIGISMEMBER(ps->ps_signodefer, sig))
820 			oact->sa_flags |= SA_NODEFER;
821 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
822 			oact->sa_flags |= SA_SIGINFO;
823 			oact->sa_sigaction =
824 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
825 		} else
826 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
827 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
828 			oact->sa_flags |= SA_NOCLDSTOP;
829 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
830 			oact->sa_flags |= SA_NOCLDWAIT;
831 	}
832 	if (act) {
833 		if ((sig == SIGKILL || sig == SIGSTOP) &&
834 		    act->sa_handler != SIG_DFL) {
835 			mtx_unlock(&ps->ps_mtx);
836 			PROC_UNLOCK(p);
837 			return (EINVAL);
838 		}
839 
840 		/*
841 		 * Change setting atomically.
842 		 */
843 
844 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
845 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
846 		if (sigact_flag_test(act, SA_SIGINFO)) {
847 			ps->ps_sigact[_SIG_IDX(sig)] =
848 			    (__sighandler_t *)act->sa_sigaction;
849 			SIGADDSET(ps->ps_siginfo, sig);
850 		} else {
851 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
852 			SIGDELSET(ps->ps_siginfo, sig);
853 		}
854 		if (!sigact_flag_test(act, SA_RESTART))
855 			SIGADDSET(ps->ps_sigintr, sig);
856 		else
857 			SIGDELSET(ps->ps_sigintr, sig);
858 		if (sigact_flag_test(act, SA_ONSTACK))
859 			SIGADDSET(ps->ps_sigonstack, sig);
860 		else
861 			SIGDELSET(ps->ps_sigonstack, sig);
862 		if (sigact_flag_test(act, SA_RESETHAND))
863 			SIGADDSET(ps->ps_sigreset, sig);
864 		else
865 			SIGDELSET(ps->ps_sigreset, sig);
866 		if (sigact_flag_test(act, SA_NODEFER))
867 			SIGADDSET(ps->ps_signodefer, sig);
868 		else
869 			SIGDELSET(ps->ps_signodefer, sig);
870 		if (sig == SIGCHLD) {
871 			if (act->sa_flags & SA_NOCLDSTOP)
872 				ps->ps_flag |= PS_NOCLDSTOP;
873 			else
874 				ps->ps_flag &= ~PS_NOCLDSTOP;
875 			if (act->sa_flags & SA_NOCLDWAIT) {
876 				/*
877 				 * Paranoia: since SA_NOCLDWAIT is implemented
878 				 * by reparenting the dying child to PID 1 (and
879 				 * trust it to reap the zombie), PID 1 itself
880 				 * is forbidden to set SA_NOCLDWAIT.
881 				 */
882 				if (p->p_pid == 1)
883 					ps->ps_flag &= ~PS_NOCLDWAIT;
884 				else
885 					ps->ps_flag |= PS_NOCLDWAIT;
886 			} else
887 				ps->ps_flag &= ~PS_NOCLDWAIT;
888 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
889 				ps->ps_flag |= PS_CLDSIGIGN;
890 			else
891 				ps->ps_flag &= ~PS_CLDSIGIGN;
892 		}
893 		/*
894 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
895 		 * and for signals set to SIG_DFL where the default is to
896 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
897 		 * have to restart the process.
898 		 */
899 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
900 		    (sigprop(sig) & SIGPROP_IGNORE &&
901 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
902 			/* never to be seen again */
903 			sigqueue_delete_proc(p, sig);
904 			if (sig != SIGCONT)
905 				/* easier in psignal */
906 				SIGADDSET(ps->ps_sigignore, sig);
907 			SIGDELSET(ps->ps_sigcatch, sig);
908 		} else {
909 			SIGDELSET(ps->ps_sigignore, sig);
910 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
911 				SIGDELSET(ps->ps_sigcatch, sig);
912 			else
913 				SIGADDSET(ps->ps_sigcatch, sig);
914 		}
915 #ifdef COMPAT_FREEBSD4
916 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
917 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
918 		    (flags & KSA_FREEBSD4) == 0)
919 			SIGDELSET(ps->ps_freebsd4, sig);
920 		else
921 			SIGADDSET(ps->ps_freebsd4, sig);
922 #endif
923 #ifdef COMPAT_43
924 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
925 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
926 		    (flags & KSA_OSIGSET) == 0)
927 			SIGDELSET(ps->ps_osigset, sig);
928 		else
929 			SIGADDSET(ps->ps_osigset, sig);
930 #endif
931 	}
932 	mtx_unlock(&ps->ps_mtx);
933 	PROC_UNLOCK(p);
934 	return (0);
935 }
936 
937 #ifndef _SYS_SYSPROTO_H_
938 struct sigaction_args {
939 	int	sig;
940 	struct	sigaction *act;
941 	struct	sigaction *oact;
942 };
943 #endif
944 int
sys_sigaction(struct thread * td,struct sigaction_args * uap)945 sys_sigaction(struct thread *td, struct sigaction_args *uap)
946 {
947 	struct sigaction act, oact;
948 	struct sigaction *actp, *oactp;
949 	int error;
950 
951 	actp = (uap->act != NULL) ? &act : NULL;
952 	oactp = (uap->oact != NULL) ? &oact : NULL;
953 	if (actp) {
954 		error = copyin(uap->act, actp, sizeof(act));
955 		if (error)
956 			return (error);
957 	}
958 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
959 	if (oactp && !error)
960 		error = copyout(oactp, uap->oact, sizeof(oact));
961 	return (error);
962 }
963 
964 #ifdef COMPAT_FREEBSD4
965 #ifndef _SYS_SYSPROTO_H_
966 struct freebsd4_sigaction_args {
967 	int	sig;
968 	struct	sigaction *act;
969 	struct	sigaction *oact;
970 };
971 #endif
972 int
freebsd4_sigaction(struct thread * td,struct freebsd4_sigaction_args * uap)973 freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
974 {
975 	struct sigaction act, oact;
976 	struct sigaction *actp, *oactp;
977 	int error;
978 
979 	actp = (uap->act != NULL) ? &act : NULL;
980 	oactp = (uap->oact != NULL) ? &oact : NULL;
981 	if (actp) {
982 		error = copyin(uap->act, actp, sizeof(act));
983 		if (error)
984 			return (error);
985 	}
986 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
987 	if (oactp && !error)
988 		error = copyout(oactp, uap->oact, sizeof(oact));
989 	return (error);
990 }
991 #endif	/* COMAPT_FREEBSD4 */
992 
993 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
994 #ifndef _SYS_SYSPROTO_H_
995 struct osigaction_args {
996 	int	signum;
997 	struct	osigaction *nsa;
998 	struct	osigaction *osa;
999 };
1000 #endif
1001 int
osigaction(struct thread * td,struct osigaction_args * uap)1002 osigaction(struct thread *td, struct osigaction_args *uap)
1003 {
1004 	struct osigaction sa;
1005 	struct sigaction nsa, osa;
1006 	struct sigaction *nsap, *osap;
1007 	int error;
1008 
1009 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1010 		return (EINVAL);
1011 
1012 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
1013 	osap = (uap->osa != NULL) ? &osa : NULL;
1014 
1015 	if (nsap) {
1016 		error = copyin(uap->nsa, &sa, sizeof(sa));
1017 		if (error)
1018 			return (error);
1019 		nsap->sa_handler = sa.sa_handler;
1020 		nsap->sa_flags = sa.sa_flags;
1021 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
1022 	}
1023 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1024 	if (osap && !error) {
1025 		sa.sa_handler = osap->sa_handler;
1026 		sa.sa_flags = osap->sa_flags;
1027 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
1028 		error = copyout(&sa, uap->osa, sizeof(sa));
1029 	}
1030 	return (error);
1031 }
1032 
1033 #if !defined(__i386__)
1034 /* Avoid replicating the same stub everywhere */
1035 int
osigreturn(struct thread * td,struct osigreturn_args * uap)1036 osigreturn(struct thread *td, struct osigreturn_args *uap)
1037 {
1038 
1039 	return (nosys(td, (struct nosys_args *)uap));
1040 }
1041 #endif
1042 #endif /* COMPAT_43 */
1043 
1044 /*
1045  * Initialize signal state for process 0;
1046  * set to ignore signals that are ignored by default.
1047  */
1048 void
siginit(struct proc * p)1049 siginit(struct proc *p)
1050 {
1051 	int i;
1052 	struct sigacts *ps;
1053 
1054 	PROC_LOCK(p);
1055 	ps = p->p_sigacts;
1056 	mtx_lock(&ps->ps_mtx);
1057 	for (i = 1; i <= NSIG; i++) {
1058 		if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) {
1059 			SIGADDSET(ps->ps_sigignore, i);
1060 		}
1061 	}
1062 	mtx_unlock(&ps->ps_mtx);
1063 	PROC_UNLOCK(p);
1064 }
1065 
1066 /*
1067  * Reset specified signal to the default disposition.
1068  */
1069 static void
sigdflt(struct sigacts * ps,int sig)1070 sigdflt(struct sigacts *ps, int sig)
1071 {
1072 
1073 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1074 	SIGDELSET(ps->ps_sigcatch, sig);
1075 	if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT)
1076 		SIGADDSET(ps->ps_sigignore, sig);
1077 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1078 	SIGDELSET(ps->ps_siginfo, sig);
1079 }
1080 
1081 /*
1082  * Reset signals for an exec of the specified process.
1083  */
1084 void
execsigs(struct proc * p)1085 execsigs(struct proc *p)
1086 {
1087 	struct sigacts *ps;
1088 	struct thread *td;
1089 
1090 	/*
1091 	 * Reset caught signals.  Held signals remain held
1092 	 * through td_sigmask (unless they were caught,
1093 	 * and are now ignored by default).
1094 	 */
1095 	PROC_LOCK_ASSERT(p, MA_OWNED);
1096 	ps = p->p_sigacts;
1097 	mtx_lock(&ps->ps_mtx);
1098 	sig_drop_caught(p);
1099 
1100 	/*
1101 	 * Reset stack state to the user stack.
1102 	 * Clear set of signals caught on the signal stack.
1103 	 */
1104 	td = curthread;
1105 	MPASS(td->td_proc == p);
1106 	td->td_sigstk.ss_flags = SS_DISABLE;
1107 	td->td_sigstk.ss_size = 0;
1108 	td->td_sigstk.ss_sp = 0;
1109 	td->td_pflags &= ~TDP_ALTSTACK;
1110 	/*
1111 	 * Reset no zombies if child dies flag as Solaris does.
1112 	 */
1113 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1114 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1115 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1116 	mtx_unlock(&ps->ps_mtx);
1117 }
1118 
1119 /*
1120  * kern_sigprocmask()
1121  *
1122  *	Manipulate signal mask.
1123  */
1124 int
kern_sigprocmask(struct thread * td,int how,sigset_t * set,sigset_t * oset,int flags)1125 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1126     int flags)
1127 {
1128 	sigset_t new_block, oset1;
1129 	struct proc *p;
1130 	int error;
1131 
1132 	p = td->td_proc;
1133 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1134 		PROC_LOCK_ASSERT(p, MA_OWNED);
1135 	else
1136 		PROC_LOCK(p);
1137 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1138 	    ? MA_OWNED : MA_NOTOWNED);
1139 	if (oset != NULL)
1140 		*oset = td->td_sigmask;
1141 
1142 	error = 0;
1143 	if (set != NULL) {
1144 		switch (how) {
1145 		case SIG_BLOCK:
1146 			SIG_CANTMASK(*set);
1147 			oset1 = td->td_sigmask;
1148 			SIGSETOR(td->td_sigmask, *set);
1149 			new_block = td->td_sigmask;
1150 			SIGSETNAND(new_block, oset1);
1151 			break;
1152 		case SIG_UNBLOCK:
1153 			SIGSETNAND(td->td_sigmask, *set);
1154 			signotify(td);
1155 			goto out;
1156 		case SIG_SETMASK:
1157 			SIG_CANTMASK(*set);
1158 			oset1 = td->td_sigmask;
1159 			if (flags & SIGPROCMASK_OLD)
1160 				SIGSETLO(td->td_sigmask, *set);
1161 			else
1162 				td->td_sigmask = *set;
1163 			new_block = td->td_sigmask;
1164 			SIGSETNAND(new_block, oset1);
1165 			signotify(td);
1166 			break;
1167 		default:
1168 			error = EINVAL;
1169 			goto out;
1170 		}
1171 
1172 		/*
1173 		 * The new_block set contains signals that were not previously
1174 		 * blocked, but are blocked now.
1175 		 *
1176 		 * In case we block any signal that was not previously blocked
1177 		 * for td, and process has the signal pending, try to schedule
1178 		 * signal delivery to some thread that does not block the
1179 		 * signal, possibly waking it up.
1180 		 */
1181 		if (p->p_numthreads != 1)
1182 			reschedule_signals(p, new_block, flags);
1183 	}
1184 
1185 out:
1186 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1187 		PROC_UNLOCK(p);
1188 	return (error);
1189 }
1190 
1191 #ifndef _SYS_SYSPROTO_H_
1192 struct sigprocmask_args {
1193 	int	how;
1194 	const sigset_t *set;
1195 	sigset_t *oset;
1196 };
1197 #endif
1198 int
sys_sigprocmask(struct thread * td,struct sigprocmask_args * uap)1199 sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1200 {
1201 	sigset_t set, oset;
1202 	sigset_t *setp, *osetp;
1203 	int error;
1204 
1205 	setp = (uap->set != NULL) ? &set : NULL;
1206 	osetp = (uap->oset != NULL) ? &oset : NULL;
1207 	if (setp) {
1208 		error = copyin(uap->set, setp, sizeof(set));
1209 		if (error)
1210 			return (error);
1211 	}
1212 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1213 	if (osetp && !error) {
1214 		error = copyout(osetp, uap->oset, sizeof(oset));
1215 	}
1216 	return (error);
1217 }
1218 
1219 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1220 #ifndef _SYS_SYSPROTO_H_
1221 struct osigprocmask_args {
1222 	int	how;
1223 	osigset_t mask;
1224 };
1225 #endif
1226 int
osigprocmask(struct thread * td,struct osigprocmask_args * uap)1227 osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1228 {
1229 	sigset_t set, oset;
1230 	int error;
1231 
1232 	OSIG2SIG(uap->mask, set);
1233 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1234 	SIG2OSIG(oset, td->td_retval[0]);
1235 	return (error);
1236 }
1237 #endif /* COMPAT_43 */
1238 
1239 int
sys_sigwait(struct thread * td,struct sigwait_args * uap)1240 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1241 {
1242 	ksiginfo_t ksi;
1243 	sigset_t set;
1244 	int error;
1245 
1246 	error = copyin(uap->set, &set, sizeof(set));
1247 	if (error) {
1248 		td->td_retval[0] = error;
1249 		return (0);
1250 	}
1251 
1252 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1253 	if (error) {
1254 		/*
1255 		 * sigwait() function shall not return EINTR, but
1256 		 * the syscall does.  Non-ancient libc provides the
1257 		 * wrapper which hides EINTR.  Otherwise, EINTR return
1258 		 * is used by libthr to handle required cancellation
1259 		 * point in the sigwait().
1260 		 */
1261 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1262 			return (ERESTART);
1263 		td->td_retval[0] = error;
1264 		return (0);
1265 	}
1266 
1267 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1268 	td->td_retval[0] = error;
1269 	return (0);
1270 }
1271 
1272 int
sys_sigtimedwait(struct thread * td,struct sigtimedwait_args * uap)1273 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1274 {
1275 	struct timespec ts;
1276 	struct timespec *timeout;
1277 	sigset_t set;
1278 	ksiginfo_t ksi;
1279 	int error;
1280 
1281 	if (uap->timeout) {
1282 		error = copyin(uap->timeout, &ts, sizeof(ts));
1283 		if (error)
1284 			return (error);
1285 
1286 		timeout = &ts;
1287 	} else
1288 		timeout = NULL;
1289 
1290 	error = copyin(uap->set, &set, sizeof(set));
1291 	if (error)
1292 		return (error);
1293 
1294 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1295 	if (error)
1296 		return (error);
1297 
1298 	if (uap->info)
1299 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1300 
1301 	if (error == 0)
1302 		td->td_retval[0] = ksi.ksi_signo;
1303 	return (error);
1304 }
1305 
1306 int
sys_sigwaitinfo(struct thread * td,struct sigwaitinfo_args * uap)1307 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1308 {
1309 	ksiginfo_t ksi;
1310 	sigset_t set;
1311 	int error;
1312 
1313 	error = copyin(uap->set, &set, sizeof(set));
1314 	if (error)
1315 		return (error);
1316 
1317 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1318 	if (error)
1319 		return (error);
1320 
1321 	if (uap->info)
1322 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1323 
1324 	if (error == 0)
1325 		td->td_retval[0] = ksi.ksi_signo;
1326 	return (error);
1327 }
1328 
1329 static void
proc_td_siginfo_capture(struct thread * td,siginfo_t * si)1330 proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1331 {
1332 	struct thread *thr;
1333 
1334 	FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1335 		if (thr == td)
1336 			thr->td_si = *si;
1337 		else
1338 			thr->td_si.si_signo = 0;
1339 	}
1340 }
1341 
1342 int
kern_sigtimedwait(struct thread * td,sigset_t waitset,ksiginfo_t * ksi,struct timespec * timeout)1343 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1344 	struct timespec *timeout)
1345 {
1346 	struct sigacts *ps;
1347 	sigset_t saved_mask, new_block;
1348 	struct proc *p;
1349 	int error, sig, timevalid = 0;
1350 	sbintime_t sbt, precision, tsbt;
1351 	struct timespec ts;
1352 	bool traced;
1353 
1354 	p = td->td_proc;
1355 	error = 0;
1356 	traced = false;
1357 
1358 	/* Ensure the sigfastblock value is up to date. */
1359 	sigfastblock_fetch(td);
1360 
1361 	if (timeout != NULL) {
1362 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1363 			timevalid = 1;
1364 			ts = *timeout;
1365 			if (ts.tv_sec < INT32_MAX / 2) {
1366 				tsbt = tstosbt(ts);
1367 				precision = tsbt;
1368 				precision >>= tc_precexp;
1369 				if (TIMESEL(&sbt, tsbt))
1370 					sbt += tc_tick_sbt;
1371 				sbt += tsbt;
1372 			} else
1373 				precision = sbt = 0;
1374 		}
1375 	} else
1376 		precision = sbt = 0;
1377 	ksiginfo_init(ksi);
1378 	/* Some signals can not be waited for. */
1379 	SIG_CANTMASK(waitset);
1380 	ps = p->p_sigacts;
1381 	PROC_LOCK(p);
1382 	saved_mask = td->td_sigmask;
1383 	SIGSETNAND(td->td_sigmask, waitset);
1384 	if ((p->p_sysent->sv_flags & SV_SIG_DISCIGN) != 0 ||
1385 	    !kern_sig_discard_ign) {
1386 		thread_lock(td);
1387 		td->td_flags |= TDF_SIGWAIT;
1388 		thread_unlock(td);
1389 	}
1390 	for (;;) {
1391 		mtx_lock(&ps->ps_mtx);
1392 		sig = cursig(td);
1393 		mtx_unlock(&ps->ps_mtx);
1394 		KASSERT(sig >= 0, ("sig %d", sig));
1395 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1396 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1397 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1398 				error = 0;
1399 				break;
1400 			}
1401 		}
1402 
1403 		if (error != 0)
1404 			break;
1405 
1406 		/*
1407 		 * POSIX says this must be checked after looking for pending
1408 		 * signals.
1409 		 */
1410 		if (timeout != NULL && !timevalid) {
1411 			error = EINVAL;
1412 			break;
1413 		}
1414 
1415 		if (traced) {
1416 			error = EINTR;
1417 			break;
1418 		}
1419 
1420 		error = msleep_sbt(&p->p_sigacts, &p->p_mtx, PPAUSE | PCATCH,
1421 		    "sigwait", sbt, precision, C_ABSOLUTE);
1422 
1423 		/* The syscalls can not be restarted. */
1424 		if (error == ERESTART)
1425 			error = EINTR;
1426 
1427 		/*
1428 		 * If PTRACE_SCE or PTRACE_SCX were set after
1429 		 * userspace entered the syscall, return spurious
1430 		 * EINTR after wait was done.  Only do this as last
1431 		 * resort after rechecking for possible queued signals
1432 		 * and expired timeouts.
1433 		 */
1434 		if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1435 			traced = true;
1436 	}
1437 	thread_lock(td);
1438 	td->td_flags &= ~TDF_SIGWAIT;
1439 	thread_unlock(td);
1440 
1441 	new_block = saved_mask;
1442 	SIGSETNAND(new_block, td->td_sigmask);
1443 	td->td_sigmask = saved_mask;
1444 	/*
1445 	 * Fewer signals can be delivered to us, reschedule signal
1446 	 * notification.
1447 	 */
1448 	if (p->p_numthreads != 1)
1449 		reschedule_signals(p, new_block, 0);
1450 
1451 	if (error == 0) {
1452 		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1453 
1454 		if (ksi->ksi_code == SI_TIMER)
1455 			itimer_accept(p, ksi->ksi_timerid, ksi);
1456 
1457 #ifdef KTRACE
1458 		if (KTRPOINT(td, KTR_PSIG)) {
1459 			sig_t action;
1460 
1461 			mtx_lock(&ps->ps_mtx);
1462 			action = ps->ps_sigact[_SIG_IDX(sig)];
1463 			mtx_unlock(&ps->ps_mtx);
1464 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1465 		}
1466 #endif
1467 		if (sig == SIGKILL) {
1468 			proc_td_siginfo_capture(td, &ksi->ksi_info);
1469 			sigexit(td, sig);
1470 		}
1471 	}
1472 	PROC_UNLOCK(p);
1473 	return (error);
1474 }
1475 
1476 #ifndef _SYS_SYSPROTO_H_
1477 struct sigpending_args {
1478 	sigset_t	*set;
1479 };
1480 #endif
1481 int
sys_sigpending(struct thread * td,struct sigpending_args * uap)1482 sys_sigpending(struct thread *td, struct sigpending_args *uap)
1483 {
1484 	struct proc *p = td->td_proc;
1485 	sigset_t pending;
1486 
1487 	PROC_LOCK(p);
1488 	pending = p->p_sigqueue.sq_signals;
1489 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1490 	PROC_UNLOCK(p);
1491 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1492 }
1493 
1494 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1495 #ifndef _SYS_SYSPROTO_H_
1496 struct osigpending_args {
1497 	int	dummy;
1498 };
1499 #endif
1500 int
osigpending(struct thread * td,struct osigpending_args * uap)1501 osigpending(struct thread *td, struct osigpending_args *uap)
1502 {
1503 	struct proc *p = td->td_proc;
1504 	sigset_t pending;
1505 
1506 	PROC_LOCK(p);
1507 	pending = p->p_sigqueue.sq_signals;
1508 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1509 	PROC_UNLOCK(p);
1510 	SIG2OSIG(pending, td->td_retval[0]);
1511 	return (0);
1512 }
1513 #endif /* COMPAT_43 */
1514 
1515 #if defined(COMPAT_43)
1516 /*
1517  * Generalized interface signal handler, 4.3-compatible.
1518  */
1519 #ifndef _SYS_SYSPROTO_H_
1520 struct osigvec_args {
1521 	int	signum;
1522 	struct	sigvec *nsv;
1523 	struct	sigvec *osv;
1524 };
1525 #endif
1526 /* ARGSUSED */
1527 int
osigvec(struct thread * td,struct osigvec_args * uap)1528 osigvec(struct thread *td, struct osigvec_args *uap)
1529 {
1530 	struct sigvec vec;
1531 	struct sigaction nsa, osa;
1532 	struct sigaction *nsap, *osap;
1533 	int error;
1534 
1535 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1536 		return (EINVAL);
1537 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1538 	osap = (uap->osv != NULL) ? &osa : NULL;
1539 	if (nsap) {
1540 		error = copyin(uap->nsv, &vec, sizeof(vec));
1541 		if (error)
1542 			return (error);
1543 		nsap->sa_handler = vec.sv_handler;
1544 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1545 		nsap->sa_flags = vec.sv_flags;
1546 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1547 	}
1548 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1549 	if (osap && !error) {
1550 		vec.sv_handler = osap->sa_handler;
1551 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1552 		vec.sv_flags = osap->sa_flags;
1553 		vec.sv_flags &= ~SA_NOCLDWAIT;
1554 		vec.sv_flags ^= SA_RESTART;
1555 		error = copyout(&vec, uap->osv, sizeof(vec));
1556 	}
1557 	return (error);
1558 }
1559 
1560 #ifndef _SYS_SYSPROTO_H_
1561 struct osigblock_args {
1562 	int	mask;
1563 };
1564 #endif
1565 int
osigblock(struct thread * td,struct osigblock_args * uap)1566 osigblock(struct thread *td, struct osigblock_args *uap)
1567 {
1568 	sigset_t set, oset;
1569 
1570 	OSIG2SIG(uap->mask, set);
1571 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1572 	SIG2OSIG(oset, td->td_retval[0]);
1573 	return (0);
1574 }
1575 
1576 #ifndef _SYS_SYSPROTO_H_
1577 struct osigsetmask_args {
1578 	int	mask;
1579 };
1580 #endif
1581 int
osigsetmask(struct thread * td,struct osigsetmask_args * uap)1582 osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1583 {
1584 	sigset_t set, oset;
1585 
1586 	OSIG2SIG(uap->mask, set);
1587 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1588 	SIG2OSIG(oset, td->td_retval[0]);
1589 	return (0);
1590 }
1591 #endif /* COMPAT_43 */
1592 
1593 /*
1594  * Suspend calling thread until signal, providing mask to be set in the
1595  * meantime.
1596  */
1597 #ifndef _SYS_SYSPROTO_H_
1598 struct sigsuspend_args {
1599 	const sigset_t *sigmask;
1600 };
1601 #endif
1602 /* ARGSUSED */
1603 int
sys_sigsuspend(struct thread * td,struct sigsuspend_args * uap)1604 sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1605 {
1606 	sigset_t mask;
1607 	int error;
1608 
1609 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1610 	if (error)
1611 		return (error);
1612 	return (kern_sigsuspend(td, mask));
1613 }
1614 
1615 int
kern_sigsuspend(struct thread * td,sigset_t mask)1616 kern_sigsuspend(struct thread *td, sigset_t mask)
1617 {
1618 	struct proc *p = td->td_proc;
1619 	int has_sig, sig;
1620 
1621 	/* Ensure the sigfastblock value is up to date. */
1622 	sigfastblock_fetch(td);
1623 
1624 	/*
1625 	 * When returning from sigsuspend, we want
1626 	 * the old mask to be restored after the
1627 	 * signal handler has finished.  Thus, we
1628 	 * save it here and mark the sigacts structure
1629 	 * to indicate this.
1630 	 */
1631 	PROC_LOCK(p);
1632 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1633 	    SIGPROCMASK_PROC_LOCKED);
1634 	td->td_pflags |= TDP_OLDMASK;
1635 	ast_sched(td, TDA_SIGSUSPEND);
1636 
1637 	/*
1638 	 * Process signals now. Otherwise, we can get spurious wakeup
1639 	 * due to signal entered process queue, but delivered to other
1640 	 * thread. But sigsuspend should return only on signal
1641 	 * delivery.
1642 	 */
1643 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1644 	for (has_sig = 0; !has_sig;) {
1645 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1646 			0) == 0)
1647 			/* void */;
1648 		thread_suspend_check(0);
1649 		mtx_lock(&p->p_sigacts->ps_mtx);
1650 		while ((sig = cursig(td)) != 0) {
1651 			KASSERT(sig >= 0, ("sig %d", sig));
1652 			has_sig += postsig(sig);
1653 		}
1654 		mtx_unlock(&p->p_sigacts->ps_mtx);
1655 
1656 		/*
1657 		 * If PTRACE_SCE or PTRACE_SCX were set after
1658 		 * userspace entered the syscall, return spurious
1659 		 * EINTR.
1660 		 */
1661 		if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1662 			has_sig += 1;
1663 	}
1664 	PROC_UNLOCK(p);
1665 	td->td_errno = EINTR;
1666 	td->td_pflags |= TDP_NERRNO;
1667 	return (EJUSTRETURN);
1668 }
1669 
1670 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1671 /*
1672  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1673  * convention: libc stub passes mask, not pointer, to save a copyin.
1674  */
1675 #ifndef _SYS_SYSPROTO_H_
1676 struct osigsuspend_args {
1677 	osigset_t mask;
1678 };
1679 #endif
1680 /* ARGSUSED */
1681 int
osigsuspend(struct thread * td,struct osigsuspend_args * uap)1682 osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1683 {
1684 	sigset_t mask;
1685 
1686 	OSIG2SIG(uap->mask, mask);
1687 	return (kern_sigsuspend(td, mask));
1688 }
1689 #endif /* COMPAT_43 */
1690 
1691 #if defined(COMPAT_43)
1692 #ifndef _SYS_SYSPROTO_H_
1693 struct osigstack_args {
1694 	struct	sigstack *nss;
1695 	struct	sigstack *oss;
1696 };
1697 #endif
1698 /* ARGSUSED */
1699 int
osigstack(struct thread * td,struct osigstack_args * uap)1700 osigstack(struct thread *td, struct osigstack_args *uap)
1701 {
1702 	struct sigstack nss, oss;
1703 	int error = 0;
1704 
1705 	if (uap->nss != NULL) {
1706 		error = copyin(uap->nss, &nss, sizeof(nss));
1707 		if (error)
1708 			return (error);
1709 	}
1710 	oss.ss_sp = td->td_sigstk.ss_sp;
1711 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1712 	if (uap->nss != NULL) {
1713 		td->td_sigstk.ss_sp = nss.ss_sp;
1714 		td->td_sigstk.ss_size = 0;
1715 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1716 		td->td_pflags |= TDP_ALTSTACK;
1717 	}
1718 	if (uap->oss != NULL)
1719 		error = copyout(&oss, uap->oss, sizeof(oss));
1720 
1721 	return (error);
1722 }
1723 #endif /* COMPAT_43 */
1724 
1725 #ifndef _SYS_SYSPROTO_H_
1726 struct sigaltstack_args {
1727 	stack_t	*ss;
1728 	stack_t	*oss;
1729 };
1730 #endif
1731 /* ARGSUSED */
1732 int
sys_sigaltstack(struct thread * td,struct sigaltstack_args * uap)1733 sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1734 {
1735 	stack_t ss, oss;
1736 	int error;
1737 
1738 	if (uap->ss != NULL) {
1739 		error = copyin(uap->ss, &ss, sizeof(ss));
1740 		if (error)
1741 			return (error);
1742 	}
1743 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1744 	    (uap->oss != NULL) ? &oss : NULL);
1745 	if (error)
1746 		return (error);
1747 	if (uap->oss != NULL)
1748 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1749 	return (error);
1750 }
1751 
1752 int
kern_sigaltstack(struct thread * td,stack_t * ss,stack_t * oss)1753 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1754 {
1755 	struct proc *p = td->td_proc;
1756 	int oonstack;
1757 
1758 	oonstack = sigonstack(cpu_getstack(td));
1759 
1760 	if (oss != NULL) {
1761 		*oss = td->td_sigstk;
1762 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1763 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1764 	}
1765 
1766 	if (ss != NULL) {
1767 		if (oonstack)
1768 			return (EPERM);
1769 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1770 			return (EINVAL);
1771 		if (!(ss->ss_flags & SS_DISABLE)) {
1772 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1773 				return (ENOMEM);
1774 
1775 			td->td_sigstk = *ss;
1776 			td->td_pflags |= TDP_ALTSTACK;
1777 		} else {
1778 			td->td_pflags &= ~TDP_ALTSTACK;
1779 		}
1780 	}
1781 	return (0);
1782 }
1783 
1784 struct killpg1_ctx {
1785 	struct thread *td;
1786 	ksiginfo_t *ksi;
1787 	int sig;
1788 	bool sent;
1789 	bool found;
1790 	int ret;
1791 };
1792 
1793 static void
killpg1_sendsig_locked(struct proc * p,struct killpg1_ctx * arg)1794 killpg1_sendsig_locked(struct proc *p, struct killpg1_ctx *arg)
1795 {
1796 	int err;
1797 
1798 	err = p_cansignal(arg->td, p, arg->sig);
1799 	if (err == 0 && arg->sig != 0)
1800 		pksignal(p, arg->sig, arg->ksi);
1801 	if (err != ESRCH)
1802 		arg->found = true;
1803 	if (err == 0)
1804 		arg->sent = true;
1805 	else if (arg->ret == 0 && err != ESRCH && err != EPERM)
1806 		arg->ret = err;
1807 }
1808 
1809 static void
killpg1_sendsig(struct proc * p,bool notself,struct killpg1_ctx * arg)1810 killpg1_sendsig(struct proc *p, bool notself, struct killpg1_ctx *arg)
1811 {
1812 
1813 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1814 	    (notself && p == arg->td->td_proc) || p->p_state == PRS_NEW)
1815 		return;
1816 
1817 	PROC_LOCK(p);
1818 	killpg1_sendsig_locked(p, arg);
1819 	PROC_UNLOCK(p);
1820 }
1821 
1822 static void
kill_processes_prison_cb(struct proc * p,void * arg)1823 kill_processes_prison_cb(struct proc *p, void *arg)
1824 {
1825 	struct killpg1_ctx *ctx = arg;
1826 
1827 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1828 	    (p == ctx->td->td_proc) || p->p_state == PRS_NEW)
1829 		return;
1830 
1831 	killpg1_sendsig_locked(p, ctx);
1832 }
1833 
1834 /*
1835  * Common code for kill process group/broadcast kill.
1836  * td is the calling thread, as usual.
1837  */
1838 static int
killpg1(struct thread * td,int sig,int pgid,int all,ksiginfo_t * ksi)1839 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1840 {
1841 	struct proc *p;
1842 	struct pgrp *pgrp;
1843 	struct killpg1_ctx arg;
1844 
1845 	arg.td = td;
1846 	arg.ksi = ksi;
1847 	arg.sig = sig;
1848 	arg.sent = false;
1849 	arg.found = false;
1850 	arg.ret = 0;
1851 	if (all) {
1852 		/*
1853 		 * broadcast
1854 		 */
1855 		prison_proc_iterate(td->td_ucred->cr_prison,
1856 		    kill_processes_prison_cb, &arg);
1857 	} else {
1858 again:
1859 		sx_slock(&proctree_lock);
1860 		if (pgid == 0) {
1861 			/*
1862 			 * zero pgid means send to my process group.
1863 			 */
1864 			pgrp = td->td_proc->p_pgrp;
1865 			PGRP_LOCK(pgrp);
1866 		} else {
1867 			pgrp = pgfind(pgid);
1868 			if (pgrp == NULL) {
1869 				sx_sunlock(&proctree_lock);
1870 				return (ESRCH);
1871 			}
1872 		}
1873 		sx_sunlock(&proctree_lock);
1874 		if (!sx_try_xlock(&pgrp->pg_killsx)) {
1875 			PGRP_UNLOCK(pgrp);
1876 			sx_xlock(&pgrp->pg_killsx);
1877 			sx_xunlock(&pgrp->pg_killsx);
1878 			goto again;
1879 		}
1880 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1881 			killpg1_sendsig(p, false, &arg);
1882 		}
1883 		PGRP_UNLOCK(pgrp);
1884 		sx_xunlock(&pgrp->pg_killsx);
1885 	}
1886 	MPASS(arg.ret != 0 || arg.found || !arg.sent);
1887 	if (arg.ret == 0 && !arg.sent)
1888 		arg.ret = arg.found ? EPERM : ESRCH;
1889 	return (arg.ret);
1890 }
1891 
1892 #ifndef _SYS_SYSPROTO_H_
1893 struct kill_args {
1894 	int	pid;
1895 	int	signum;
1896 };
1897 #endif
1898 /* ARGSUSED */
1899 int
sys_kill(struct thread * td,struct kill_args * uap)1900 sys_kill(struct thread *td, struct kill_args *uap)
1901 {
1902 
1903 	return (kern_kill(td, uap->pid, uap->signum));
1904 }
1905 
1906 int
kern_kill(struct thread * td,pid_t pid,int signum)1907 kern_kill(struct thread *td, pid_t pid, int signum)
1908 {
1909 	ksiginfo_t ksi;
1910 	struct proc *p;
1911 	int error;
1912 
1913 	/*
1914 	 * A process in capability mode can send signals only to himself.
1915 	 * The main rationale behind this is that abort(3) is implemented as
1916 	 * kill(getpid(), SIGABRT).
1917 	 */
1918 	if (pid != td->td_proc->p_pid) {
1919 		if (CAP_TRACING(td))
1920 			ktrcapfail(CAPFAIL_SIGNAL, &signum);
1921 		if (IN_CAPABILITY_MODE(td))
1922 			return (ECAPMODE);
1923 	}
1924 
1925 	AUDIT_ARG_SIGNUM(signum);
1926 	AUDIT_ARG_PID(pid);
1927 	if ((u_int)signum > _SIG_MAXSIG)
1928 		return (EINVAL);
1929 
1930 	ksiginfo_init(&ksi);
1931 	ksi.ksi_signo = signum;
1932 	ksi.ksi_code = SI_USER;
1933 	ksi.ksi_pid = td->td_proc->p_pid;
1934 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1935 
1936 	if (pid > 0) {
1937 		/* kill single process */
1938 		if ((p = pfind_any(pid)) == NULL)
1939 			return (ESRCH);
1940 		AUDIT_ARG_PROCESS(p);
1941 		error = p_cansignal(td, p, signum);
1942 		if (error == 0 && signum)
1943 			pksignal(p, signum, &ksi);
1944 		PROC_UNLOCK(p);
1945 		return (error);
1946 	}
1947 	switch (pid) {
1948 	case -1:		/* broadcast signal */
1949 		return (killpg1(td, signum, 0, 1, &ksi));
1950 	case 0:			/* signal own process group */
1951 		return (killpg1(td, signum, 0, 0, &ksi));
1952 	default:		/* negative explicit process group */
1953 		return (killpg1(td, signum, -pid, 0, &ksi));
1954 	}
1955 	/* NOTREACHED */
1956 }
1957 
1958 int
sys_pdkill(struct thread * td,struct pdkill_args * uap)1959 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1960 {
1961 	struct proc *p;
1962 	int error;
1963 
1964 	AUDIT_ARG_SIGNUM(uap->signum);
1965 	AUDIT_ARG_FD(uap->fd);
1966 	if ((u_int)uap->signum > _SIG_MAXSIG)
1967 		return (EINVAL);
1968 
1969 	error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1970 	if (error)
1971 		return (error);
1972 	AUDIT_ARG_PROCESS(p);
1973 	error = p_cansignal(td, p, uap->signum);
1974 	if (error == 0 && uap->signum)
1975 		kern_psignal(p, uap->signum);
1976 	PROC_UNLOCK(p);
1977 	return (error);
1978 }
1979 
1980 #if defined(COMPAT_43)
1981 #ifndef _SYS_SYSPROTO_H_
1982 struct okillpg_args {
1983 	int	pgid;
1984 	int	signum;
1985 };
1986 #endif
1987 /* ARGSUSED */
1988 int
okillpg(struct thread * td,struct okillpg_args * uap)1989 okillpg(struct thread *td, struct okillpg_args *uap)
1990 {
1991 	ksiginfo_t ksi;
1992 
1993 	AUDIT_ARG_SIGNUM(uap->signum);
1994 	AUDIT_ARG_PID(uap->pgid);
1995 	if ((u_int)uap->signum > _SIG_MAXSIG)
1996 		return (EINVAL);
1997 
1998 	ksiginfo_init(&ksi);
1999 	ksi.ksi_signo = uap->signum;
2000 	ksi.ksi_code = SI_USER;
2001 	ksi.ksi_pid = td->td_proc->p_pid;
2002 	ksi.ksi_uid = td->td_ucred->cr_ruid;
2003 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
2004 }
2005 #endif /* COMPAT_43 */
2006 
2007 #ifndef _SYS_SYSPROTO_H_
2008 struct sigqueue_args {
2009 	pid_t pid;
2010 	int signum;
2011 	/* union sigval */ void *value;
2012 };
2013 #endif
2014 int
sys_sigqueue(struct thread * td,struct sigqueue_args * uap)2015 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
2016 {
2017 	union sigval sv;
2018 
2019 	sv.sival_ptr = uap->value;
2020 
2021 	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
2022 }
2023 
2024 int
kern_sigqueue(struct thread * td,pid_t pid,int signumf,union sigval * value)2025 kern_sigqueue(struct thread *td, pid_t pid, int signumf, union sigval *value)
2026 {
2027 	ksiginfo_t ksi;
2028 	struct proc *p;
2029 	struct thread *td2;
2030 	u_int signum;
2031 	int error;
2032 
2033 	signum = signumf & ~__SIGQUEUE_TID;
2034 	if (signum > _SIG_MAXSIG)
2035 		return (EINVAL);
2036 
2037 	/*
2038 	 * Specification says sigqueue can only send signal to
2039 	 * single process.
2040 	 */
2041 	if (pid <= 0)
2042 		return (EINVAL);
2043 
2044 	if ((signumf & __SIGQUEUE_TID) == 0) {
2045 		if ((p = pfind_any(pid)) == NULL)
2046 			return (ESRCH);
2047 		td2 = NULL;
2048 	} else {
2049 		p = td->td_proc;
2050 		td2 = tdfind((lwpid_t)pid, p->p_pid);
2051 		if (td2 == NULL)
2052 			return (ESRCH);
2053 	}
2054 
2055 	error = p_cansignal(td, p, signum);
2056 	if (error == 0 && signum != 0) {
2057 		ksiginfo_init(&ksi);
2058 		ksi.ksi_flags = KSI_SIGQ;
2059 		ksi.ksi_signo = signum;
2060 		ksi.ksi_code = SI_QUEUE;
2061 		ksi.ksi_pid = td->td_proc->p_pid;
2062 		ksi.ksi_uid = td->td_ucred->cr_ruid;
2063 		ksi.ksi_value = *value;
2064 		error = tdsendsignal(p, td2, ksi.ksi_signo, &ksi);
2065 	}
2066 	PROC_UNLOCK(p);
2067 	return (error);
2068 }
2069 
2070 /*
2071  * Send a signal to a process group.  If checktty is 1,
2072  * limit to members which have a controlling terminal.
2073  */
2074 void
pgsignal(struct pgrp * pgrp,int sig,int checkctty,ksiginfo_t * ksi)2075 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
2076 {
2077 	struct proc *p;
2078 
2079 	if (pgrp) {
2080 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
2081 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
2082 			PROC_LOCK(p);
2083 			if (p->p_state == PRS_NORMAL &&
2084 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
2085 				pksignal(p, sig, ksi);
2086 			PROC_UNLOCK(p);
2087 		}
2088 	}
2089 }
2090 
2091 /*
2092  * Recalculate the signal mask and reset the signal disposition after
2093  * usermode frame for delivery is formed.  Should be called after
2094  * mach-specific routine, because sysent->sv_sendsig() needs correct
2095  * ps_siginfo and signal mask.
2096  */
2097 static void
postsig_done(int sig,struct thread * td,struct sigacts * ps)2098 postsig_done(int sig, struct thread *td, struct sigacts *ps)
2099 {
2100 	sigset_t mask;
2101 
2102 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2103 	td->td_ru.ru_nsignals++;
2104 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
2105 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
2106 		SIGADDSET(mask, sig);
2107 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
2108 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
2109 	if (SIGISMEMBER(ps->ps_sigreset, sig))
2110 		sigdflt(ps, sig);
2111 }
2112 
2113 /*
2114  * Send a signal caused by a trap to the current thread.  If it will be
2115  * caught immediately, deliver it with correct code.  Otherwise, post it
2116  * normally.
2117  */
2118 void
trapsignal(struct thread * td,ksiginfo_t * ksi)2119 trapsignal(struct thread *td, ksiginfo_t *ksi)
2120 {
2121 	struct sigacts *ps;
2122 	struct proc *p;
2123 	sigset_t sigmask;
2124 	int sig;
2125 
2126 	p = td->td_proc;
2127 	sig = ksi->ksi_signo;
2128 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
2129 
2130 	sigfastblock_fetch(td);
2131 	PROC_LOCK(p);
2132 	ps = p->p_sigacts;
2133 	mtx_lock(&ps->ps_mtx);
2134 	sigmask = td->td_sigmask;
2135 	if (td->td_sigblock_val != 0)
2136 		SIGSETOR(sigmask, fastblock_mask);
2137 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2138 	    !SIGISMEMBER(sigmask, sig)) {
2139 #ifdef KTRACE
2140 		if (KTRPOINT(curthread, KTR_PSIG))
2141 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2142 			    &td->td_sigmask, ksi->ksi_code);
2143 #endif
2144 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2145 		    ksi, &td->td_sigmask);
2146 		postsig_done(sig, td, ps);
2147 		mtx_unlock(&ps->ps_mtx);
2148 	} else {
2149 		/*
2150 		 * Avoid a possible infinite loop if the thread
2151 		 * masking the signal or process is ignoring the
2152 		 * signal.
2153 		 */
2154 		if (kern_forcesigexit && (SIGISMEMBER(sigmask, sig) ||
2155 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2156 			SIGDELSET(td->td_sigmask, sig);
2157 			SIGDELSET(ps->ps_sigcatch, sig);
2158 			SIGDELSET(ps->ps_sigignore, sig);
2159 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2160 			td->td_pflags &= ~TDP_SIGFASTBLOCK;
2161 			td->td_sigblock_val = 0;
2162 		}
2163 		mtx_unlock(&ps->ps_mtx);
2164 		p->p_sig = sig;		/* XXX to verify code */
2165 		tdsendsignal(p, td, sig, ksi);
2166 	}
2167 	PROC_UNLOCK(p);
2168 }
2169 
2170 static struct thread *
sigtd(struct proc * p,int sig,bool fast_sigblock)2171 sigtd(struct proc *p, int sig, bool fast_sigblock)
2172 {
2173 	struct thread *td, *signal_td;
2174 
2175 	PROC_LOCK_ASSERT(p, MA_OWNED);
2176 	MPASS(!fast_sigblock || p == curproc);
2177 
2178 	/*
2179 	 * Check if current thread can handle the signal without
2180 	 * switching context to another thread.
2181 	 */
2182 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig) &&
2183 	    (!fast_sigblock || curthread->td_sigblock_val == 0))
2184 		return (curthread);
2185 
2186 	/* Find a non-stopped thread that does not mask the signal. */
2187 	signal_td = NULL;
2188 	FOREACH_THREAD_IN_PROC(p, td) {
2189 		if (!SIGISMEMBER(td->td_sigmask, sig) && (!fast_sigblock ||
2190 		    td != curthread || td->td_sigblock_val == 0) &&
2191 		    (td->td_flags & TDF_BOUNDARY) == 0) {
2192 			signal_td = td;
2193 			break;
2194 		}
2195 	}
2196 	/* Select random (first) thread if no better match was found. */
2197 	if (signal_td == NULL)
2198 		signal_td = FIRST_THREAD_IN_PROC(p);
2199 	return (signal_td);
2200 }
2201 
2202 /*
2203  * Send the signal to the process.  If the signal has an action, the action
2204  * is usually performed by the target process rather than the caller; we add
2205  * the signal to the set of pending signals for the process.
2206  *
2207  * Exceptions:
2208  *   o When a stop signal is sent to a sleeping process that takes the
2209  *     default action, the process is stopped without awakening it.
2210  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2211  *     regardless of the signal action (eg, blocked or ignored).
2212  *
2213  * Other ignored signals are discarded immediately.
2214  *
2215  * NB: This function may be entered from the debugger via the "kill" DDB
2216  * command.  There is little that can be done to mitigate the possibly messy
2217  * side effects of this unwise possibility.
2218  */
2219 void
kern_psignal(struct proc * p,int sig)2220 kern_psignal(struct proc *p, int sig)
2221 {
2222 	ksiginfo_t ksi;
2223 
2224 	ksiginfo_init(&ksi);
2225 	ksi.ksi_signo = sig;
2226 	ksi.ksi_code = SI_KERNEL;
2227 	(void) tdsendsignal(p, NULL, sig, &ksi);
2228 }
2229 
2230 int
pksignal(struct proc * p,int sig,ksiginfo_t * ksi)2231 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2232 {
2233 
2234 	return (tdsendsignal(p, NULL, sig, ksi));
2235 }
2236 
2237 /* Utility function for finding a thread to send signal event to. */
2238 int
sigev_findtd(struct proc * p,struct sigevent * sigev,struct thread ** ttd)2239 sigev_findtd(struct proc *p, struct sigevent *sigev, struct thread **ttd)
2240 {
2241 	struct thread *td;
2242 
2243 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2244 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2245 		if (td == NULL)
2246 			return (ESRCH);
2247 		*ttd = td;
2248 	} else {
2249 		*ttd = NULL;
2250 		PROC_LOCK(p);
2251 	}
2252 	return (0);
2253 }
2254 
2255 void
tdsignal(struct thread * td,int sig)2256 tdsignal(struct thread *td, int sig)
2257 {
2258 	ksiginfo_t ksi;
2259 
2260 	ksiginfo_init(&ksi);
2261 	ksi.ksi_signo = sig;
2262 	ksi.ksi_code = SI_KERNEL;
2263 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2264 }
2265 
2266 void
tdksignal(struct thread * td,int sig,ksiginfo_t * ksi)2267 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2268 {
2269 
2270 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2271 }
2272 
2273 static void
sig_sleepq_abort(struct thread * td,int intrval)2274 sig_sleepq_abort(struct thread *td, int intrval)
2275 {
2276 	THREAD_LOCK_ASSERT(td, MA_OWNED);
2277 
2278 	if (intrval == 0 && (td->td_flags & TDF_SIGWAIT) == 0)
2279 		thread_unlock(td);
2280 	else
2281 		sleepq_abort(td, intrval);
2282 }
2283 
2284 int
tdsendsignal(struct proc * p,struct thread * td,int sig,ksiginfo_t * ksi)2285 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2286 {
2287 	sig_t action;
2288 	sigqueue_t *sigqueue;
2289 	struct sigacts *ps;
2290 	int intrval, prop, ret;
2291 
2292 	MPASS(td == NULL || p == td->td_proc);
2293 	PROC_LOCK_ASSERT(p, MA_OWNED);
2294 
2295 	if (!_SIG_VALID(sig))
2296 		panic("%s(): invalid signal %d", __func__, sig);
2297 
2298 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2299 
2300 	/*
2301 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2302 	 */
2303 	if (p->p_state == PRS_ZOMBIE) {
2304 		if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2305 			ksiginfo_tryfree(ksi);
2306 		return (0);
2307 	}
2308 
2309 	ps = p->p_sigacts;
2310 	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2311 	prop = sigprop(sig);
2312 
2313 	if (td == NULL) {
2314 		td = sigtd(p, sig, false);
2315 		sigqueue = &p->p_sigqueue;
2316 	} else
2317 		sigqueue = &td->td_sigqueue;
2318 
2319 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2320 
2321 	/*
2322 	 * If the signal is being ignored, then we forget about it
2323 	 * immediately, except when the target process executes
2324 	 * sigwait().  (Note: we don't set SIGCONT in ps_sigignore,
2325 	 * and if it is set to SIG_IGN, action will be SIG_DFL here.)
2326 	 */
2327 	mtx_lock(&ps->ps_mtx);
2328 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2329 		if (kern_sig_discard_ign &&
2330 		    (p->p_sysent->sv_flags & SV_SIG_DISCIGN) == 0) {
2331 			SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2332 
2333 			mtx_unlock(&ps->ps_mtx);
2334 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2335 				ksiginfo_tryfree(ksi);
2336 			return (0);
2337 		} else {
2338 			action = SIG_CATCH;
2339 			intrval = 0;
2340 		}
2341 	} else {
2342 		if (SIGISMEMBER(td->td_sigmask, sig))
2343 			action = SIG_HOLD;
2344 		else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2345 			action = SIG_CATCH;
2346 		else
2347 			action = SIG_DFL;
2348 		if (SIGISMEMBER(ps->ps_sigintr, sig))
2349 			intrval = EINTR;
2350 		else
2351 			intrval = ERESTART;
2352 	}
2353 	mtx_unlock(&ps->ps_mtx);
2354 
2355 	if (prop & SIGPROP_CONT)
2356 		sigqueue_delete_stopmask_proc(p);
2357 	else if (prop & SIGPROP_STOP) {
2358 		/*
2359 		 * If sending a tty stop signal to a member of an orphaned
2360 		 * process group, discard the signal here if the action
2361 		 * is default; don't stop the process below if sleeping,
2362 		 * and don't clear any pending SIGCONT.
2363 		 */
2364 		if ((prop & SIGPROP_TTYSTOP) != 0 &&
2365 		    (p->p_pgrp->pg_flags & PGRP_ORPHANED) != 0 &&
2366 		    action == SIG_DFL) {
2367 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2368 				ksiginfo_tryfree(ksi);
2369 			return (0);
2370 		}
2371 		sigqueue_delete_proc(p, SIGCONT);
2372 		if (p->p_flag & P_CONTINUED) {
2373 			p->p_flag &= ~P_CONTINUED;
2374 			PROC_LOCK(p->p_pptr);
2375 			sigqueue_take(p->p_ksi);
2376 			PROC_UNLOCK(p->p_pptr);
2377 		}
2378 	}
2379 
2380 	ret = sigqueue_add(sigqueue, sig, ksi);
2381 	if (ret != 0)
2382 		return (ret);
2383 	signotify(td);
2384 	/*
2385 	 * Defer further processing for signals which are held,
2386 	 * except that stopped processes must be continued by SIGCONT.
2387 	 */
2388 	if (action == SIG_HOLD &&
2389 	    !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2390 		return (0);
2391 
2392 	/*
2393 	 * Some signals have a process-wide effect and a per-thread
2394 	 * component.  Most processing occurs when the process next
2395 	 * tries to cross the user boundary, however there are some
2396 	 * times when processing needs to be done immediately, such as
2397 	 * waking up threads so that they can cross the user boundary.
2398 	 * We try to do the per-process part here.
2399 	 */
2400 	if (P_SHOULDSTOP(p)) {
2401 		KASSERT(!(p->p_flag & P_WEXIT),
2402 		    ("signal to stopped but exiting process"));
2403 		if (sig == SIGKILL) {
2404 			/*
2405 			 * If traced process is already stopped,
2406 			 * then no further action is necessary.
2407 			 */
2408 			if (p->p_flag & P_TRACED)
2409 				return (0);
2410 			/*
2411 			 * SIGKILL sets process running.
2412 			 * It will die elsewhere.
2413 			 * All threads must be restarted.
2414 			 */
2415 			p->p_flag &= ~P_STOPPED_SIG;
2416 			goto runfast;
2417 		}
2418 
2419 		if (prop & SIGPROP_CONT) {
2420 			/*
2421 			 * If traced process is already stopped,
2422 			 * then no further action is necessary.
2423 			 */
2424 			if (p->p_flag & P_TRACED)
2425 				return (0);
2426 			/*
2427 			 * If SIGCONT is default (or ignored), we continue the
2428 			 * process but don't leave the signal in sigqueue as
2429 			 * it has no further action.  If SIGCONT is held, we
2430 			 * continue the process and leave the signal in
2431 			 * sigqueue.  If the process catches SIGCONT, let it
2432 			 * handle the signal itself.  If it isn't waiting on
2433 			 * an event, it goes back to run state.
2434 			 * Otherwise, process goes back to sleep state.
2435 			 */
2436 			p->p_flag &= ~P_STOPPED_SIG;
2437 			PROC_SLOCK(p);
2438 			if (p->p_numthreads == p->p_suspcount) {
2439 				PROC_SUNLOCK(p);
2440 				p->p_flag |= P_CONTINUED;
2441 				p->p_xsig = SIGCONT;
2442 				PROC_LOCK(p->p_pptr);
2443 				childproc_continued(p);
2444 				PROC_UNLOCK(p->p_pptr);
2445 				PROC_SLOCK(p);
2446 			}
2447 			if (action == SIG_DFL) {
2448 				thread_unsuspend(p);
2449 				PROC_SUNLOCK(p);
2450 				sigqueue_delete(sigqueue, sig);
2451 				goto out_cont;
2452 			}
2453 			if (action == SIG_CATCH) {
2454 				/*
2455 				 * The process wants to catch it so it needs
2456 				 * to run at least one thread, but which one?
2457 				 */
2458 				PROC_SUNLOCK(p);
2459 				goto runfast;
2460 			}
2461 			/*
2462 			 * The signal is not ignored or caught.
2463 			 */
2464 			thread_unsuspend(p);
2465 			PROC_SUNLOCK(p);
2466 			goto out_cont;
2467 		}
2468 
2469 		if (prop & SIGPROP_STOP) {
2470 			/*
2471 			 * If traced process is already stopped,
2472 			 * then no further action is necessary.
2473 			 */
2474 			if (p->p_flag & P_TRACED)
2475 				return (0);
2476 			/*
2477 			 * Already stopped, don't need to stop again
2478 			 * (If we did the shell could get confused).
2479 			 * Just make sure the signal STOP bit set.
2480 			 */
2481 			p->p_flag |= P_STOPPED_SIG;
2482 			sigqueue_delete(sigqueue, sig);
2483 			return (0);
2484 		}
2485 
2486 		/*
2487 		 * All other kinds of signals:
2488 		 * If a thread is sleeping interruptibly, simulate a
2489 		 * wakeup so that when it is continued it will be made
2490 		 * runnable and can look at the signal.  However, don't make
2491 		 * the PROCESS runnable, leave it stopped.
2492 		 * It may run a bit until it hits a thread_suspend_check().
2493 		 */
2494 		PROC_SLOCK(p);
2495 		thread_lock(td);
2496 		if (TD_CAN_ABORT(td))
2497 			sig_sleepq_abort(td, intrval);
2498 		else
2499 			thread_unlock(td);
2500 		PROC_SUNLOCK(p);
2501 		return (0);
2502 		/*
2503 		 * Mutexes are short lived. Threads waiting on them will
2504 		 * hit thread_suspend_check() soon.
2505 		 */
2506 	} else if (p->p_state == PRS_NORMAL) {
2507 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2508 			tdsigwakeup(td, sig, action, intrval);
2509 			return (0);
2510 		}
2511 
2512 		MPASS(action == SIG_DFL);
2513 
2514 		if (prop & SIGPROP_STOP) {
2515 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2516 				return (0);
2517 			p->p_flag |= P_STOPPED_SIG;
2518 			p->p_xsig = sig;
2519 			PROC_SLOCK(p);
2520 			sig_suspend_threads(td, p);
2521 			if (p->p_numthreads == p->p_suspcount) {
2522 				/*
2523 				 * only thread sending signal to another
2524 				 * process can reach here, if thread is sending
2525 				 * signal to its process, because thread does
2526 				 * not suspend itself here, p_numthreads
2527 				 * should never be equal to p_suspcount.
2528 				 */
2529 				thread_stopped(p);
2530 				PROC_SUNLOCK(p);
2531 				sigqueue_delete_proc(p, p->p_xsig);
2532 			} else
2533 				PROC_SUNLOCK(p);
2534 			return (0);
2535 		}
2536 	} else {
2537 		/* Not in "NORMAL" state. discard the signal. */
2538 		sigqueue_delete(sigqueue, sig);
2539 		return (0);
2540 	}
2541 
2542 	/*
2543 	 * The process is not stopped so we need to apply the signal to all the
2544 	 * running threads.
2545 	 */
2546 runfast:
2547 	tdsigwakeup(td, sig, action, intrval);
2548 	PROC_SLOCK(p);
2549 	thread_unsuspend(p);
2550 	PROC_SUNLOCK(p);
2551 out_cont:
2552 	itimer_proc_continue(p);
2553 	kqtimer_proc_continue(p);
2554 
2555 	return (0);
2556 }
2557 
2558 /*
2559  * The force of a signal has been directed against a single
2560  * thread.  We need to see what we can do about knocking it
2561  * out of any sleep it may be in etc.
2562  */
2563 static void
tdsigwakeup(struct thread * td,int sig,sig_t action,int intrval)2564 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2565 {
2566 	struct proc *p = td->td_proc;
2567 	int prop;
2568 
2569 	PROC_LOCK_ASSERT(p, MA_OWNED);
2570 	prop = sigprop(sig);
2571 
2572 	PROC_SLOCK(p);
2573 	thread_lock(td);
2574 	/*
2575 	 * Bring the priority of a thread up if we want it to get
2576 	 * killed in this lifetime.  Be careful to avoid bumping the
2577 	 * priority of the idle thread, since we still allow to signal
2578 	 * kernel processes.
2579 	 */
2580 	if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2581 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2582 		sched_prio(td, PUSER);
2583 	if (TD_ON_SLEEPQ(td)) {
2584 		/*
2585 		 * If thread is sleeping uninterruptibly
2586 		 * we can't interrupt the sleep... the signal will
2587 		 * be noticed when the process returns through
2588 		 * trap() or syscall().
2589 		 */
2590 		if ((td->td_flags & TDF_SINTR) == 0)
2591 			goto out;
2592 		/*
2593 		 * If SIGCONT is default (or ignored) and process is
2594 		 * asleep, we are finished; the process should not
2595 		 * be awakened.
2596 		 */
2597 		if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2598 			thread_unlock(td);
2599 			PROC_SUNLOCK(p);
2600 			sigqueue_delete(&p->p_sigqueue, sig);
2601 			/*
2602 			 * It may be on either list in this state.
2603 			 * Remove from both for now.
2604 			 */
2605 			sigqueue_delete(&td->td_sigqueue, sig);
2606 			return;
2607 		}
2608 
2609 		/*
2610 		 * Don't awaken a sleeping thread for SIGSTOP if the
2611 		 * STOP signal is deferred.
2612 		 */
2613 		if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2614 		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2615 			goto out;
2616 
2617 		/*
2618 		 * Give low priority threads a better chance to run.
2619 		 */
2620 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2621 			sched_prio(td, PUSER);
2622 
2623 		sig_sleepq_abort(td, intrval);
2624 		PROC_SUNLOCK(p);
2625 		return;
2626 	}
2627 
2628 	/*
2629 	 * Other states do nothing with the signal immediately,
2630 	 * other than kicking ourselves if we are running.
2631 	 * It will either never be noticed, or noticed very soon.
2632 	 */
2633 #ifdef SMP
2634 	if (TD_IS_RUNNING(td) && td != curthread)
2635 		forward_signal(td);
2636 #endif
2637 
2638 out:
2639 	PROC_SUNLOCK(p);
2640 	thread_unlock(td);
2641 }
2642 
2643 static void
ptrace_coredumpreq(struct thread * td,struct proc * p,struct thr_coredump_req * tcq)2644 ptrace_coredumpreq(struct thread *td, struct proc *p,
2645     struct thr_coredump_req *tcq)
2646 {
2647 	void *rl_cookie;
2648 
2649 	if (p->p_sysent->sv_coredump == NULL) {
2650 		tcq->tc_error = ENOSYS;
2651 		return;
2652 	}
2653 
2654 	rl_cookie = vn_rangelock_wlock(tcq->tc_vp, 0, OFF_MAX);
2655 	tcq->tc_error = p->p_sysent->sv_coredump(td, tcq->tc_vp,
2656 	    tcq->tc_limit, tcq->tc_flags);
2657 	vn_rangelock_unlock(tcq->tc_vp, rl_cookie);
2658 }
2659 
2660 static void
ptrace_syscallreq(struct thread * td,struct proc * p,struct thr_syscall_req * tsr)2661 ptrace_syscallreq(struct thread *td, struct proc *p,
2662     struct thr_syscall_req *tsr)
2663 {
2664 	struct sysentvec *sv;
2665 	struct sysent *se;
2666 	register_t rv_saved[2];
2667 	int error, nerror;
2668 	int sc;
2669 	bool audited, sy_thr_static;
2670 
2671 	sv = p->p_sysent;
2672 	if (sv->sv_table == NULL || sv->sv_size < tsr->ts_sa.code) {
2673 		tsr->ts_ret.sr_error = ENOSYS;
2674 		return;
2675 	}
2676 
2677 	sc = tsr->ts_sa.code;
2678 	if (sc == SYS_syscall || sc == SYS___syscall) {
2679 		sc = tsr->ts_sa.args[0];
2680 		memmove(&tsr->ts_sa.args[0], &tsr->ts_sa.args[1],
2681 		    sizeof(register_t) * (tsr->ts_nargs - 1));
2682 	}
2683 
2684 	tsr->ts_sa.callp = se = &sv->sv_table[sc];
2685 
2686 	VM_CNT_INC(v_syscall);
2687 	td->td_pticks = 0;
2688 	if (__predict_false(td->td_cowgen != atomic_load_int(
2689 	    &td->td_proc->p_cowgen)))
2690 		thread_cow_update(td);
2691 
2692 	td->td_sa = tsr->ts_sa;
2693 
2694 #ifdef CAPABILITY_MODE
2695 	if ((se->sy_flags & SYF_CAPENABLED) == 0) {
2696 		if (CAP_TRACING(td))
2697 			ktrcapfail(CAPFAIL_SYSCALL, NULL);
2698 		if (IN_CAPABILITY_MODE(td)) {
2699 			tsr->ts_ret.sr_error = ECAPMODE;
2700 			return;
2701 		}
2702 	}
2703 #endif
2704 
2705 	sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2706 	audited = AUDIT_SYSCALL_ENTER(sc, td) != 0;
2707 
2708 	if (!sy_thr_static) {
2709 		error = syscall_thread_enter(td, &se);
2710 		sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2711 		if (error != 0) {
2712 			tsr->ts_ret.sr_error = error;
2713 			return;
2714 		}
2715 	}
2716 
2717 	rv_saved[0] = td->td_retval[0];
2718 	rv_saved[1] = td->td_retval[1];
2719 	nerror = td->td_errno;
2720 	td->td_retval[0] = 0;
2721 	td->td_retval[1] = 0;
2722 
2723 #ifdef KDTRACE_HOOKS
2724 	if (se->sy_entry != 0)
2725 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_ENTRY, 0);
2726 #endif
2727 	tsr->ts_ret.sr_error = se->sy_call(td, tsr->ts_sa.args);
2728 #ifdef KDTRACE_HOOKS
2729 	if (se->sy_return != 0)
2730 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_RETURN,
2731 		    tsr->ts_ret.sr_error != 0 ? -1 : td->td_retval[0]);
2732 #endif
2733 
2734 	tsr->ts_ret.sr_retval[0] = td->td_retval[0];
2735 	tsr->ts_ret.sr_retval[1] = td->td_retval[1];
2736 	td->td_retval[0] = rv_saved[0];
2737 	td->td_retval[1] = rv_saved[1];
2738 	td->td_errno = nerror;
2739 
2740 	if (audited)
2741 		AUDIT_SYSCALL_EXIT(error, td);
2742 	if (!sy_thr_static)
2743 		syscall_thread_exit(td, se);
2744 }
2745 
2746 static void
ptrace_remotereq(struct thread * td,int flag)2747 ptrace_remotereq(struct thread *td, int flag)
2748 {
2749 	struct proc *p;
2750 
2751 	MPASS(td == curthread);
2752 	p = td->td_proc;
2753 	PROC_LOCK_ASSERT(p, MA_OWNED);
2754 	if ((td->td_dbgflags & flag) == 0)
2755 		return;
2756 	KASSERT((p->p_flag & P_STOPPED_TRACE) != 0, ("not stopped"));
2757 	KASSERT(td->td_remotereq != NULL, ("td_remotereq is NULL"));
2758 
2759 	PROC_UNLOCK(p);
2760 	switch (flag) {
2761 	case TDB_COREDUMPREQ:
2762 		ptrace_coredumpreq(td, p, td->td_remotereq);
2763 		break;
2764 	case TDB_SCREMOTEREQ:
2765 		ptrace_syscallreq(td, p, td->td_remotereq);
2766 		break;
2767 	default:
2768 		__unreachable();
2769 	}
2770 	PROC_LOCK(p);
2771 
2772 	MPASS((td->td_dbgflags & flag) != 0);
2773 	td->td_dbgflags &= ~flag;
2774 	td->td_remotereq = NULL;
2775 	wakeup(p);
2776 }
2777 
2778 static void
sig_suspend_threads(struct thread * td,struct proc * p)2779 sig_suspend_threads(struct thread *td, struct proc *p)
2780 {
2781 	struct thread *td2;
2782 
2783 	PROC_LOCK_ASSERT(p, MA_OWNED);
2784 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2785 
2786 	FOREACH_THREAD_IN_PROC(p, td2) {
2787 		thread_lock(td2);
2788 		ast_sched_locked(td2, TDA_SUSPEND);
2789 		if (TD_IS_SLEEPING(td2) && (td2->td_flags & TDF_SINTR) != 0) {
2790 			if (td2->td_flags & TDF_SBDRY) {
2791 				/*
2792 				 * Once a thread is asleep with
2793 				 * TDF_SBDRY and without TDF_SERESTART
2794 				 * or TDF_SEINTR set, it should never
2795 				 * become suspended due to this check.
2796 				 */
2797 				KASSERT(!TD_IS_SUSPENDED(td2),
2798 				    ("thread with deferred stops suspended"));
2799 				if (TD_SBDRY_INTR(td2)) {
2800 					sleepq_abort(td2, TD_SBDRY_ERRNO(td2));
2801 					continue;
2802 				}
2803 			} else if (!TD_IS_SUSPENDED(td2))
2804 				thread_suspend_one(td2);
2805 		} else if (!TD_IS_SUSPENDED(td2)) {
2806 #ifdef SMP
2807 			if (TD_IS_RUNNING(td2) && td2 != td)
2808 				forward_signal(td2);
2809 #endif
2810 		}
2811 		thread_unlock(td2);
2812 	}
2813 }
2814 
2815 /*
2816  * Stop the process for an event deemed interesting to the debugger. If si is
2817  * non-NULL, this is a signal exchange; the new signal requested by the
2818  * debugger will be returned for handling. If si is NULL, this is some other
2819  * type of interesting event. The debugger may request a signal be delivered in
2820  * that case as well, however it will be deferred until it can be handled.
2821  */
2822 int
ptracestop(struct thread * td,int sig,ksiginfo_t * si)2823 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2824 {
2825 	struct proc *p = td->td_proc;
2826 	struct thread *td2;
2827 	ksiginfo_t ksi;
2828 
2829 	PROC_LOCK_ASSERT(p, MA_OWNED);
2830 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2831 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2832 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2833 
2834 	td->td_xsig = sig;
2835 
2836 	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2837 		td->td_dbgflags |= TDB_XSIG;
2838 		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2839 		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2840 		PROC_SLOCK(p);
2841 		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2842 			if (P_KILLED(p)) {
2843 				/*
2844 				 * Ensure that, if we've been PT_KILLed, the
2845 				 * exit status reflects that. Another thread
2846 				 * may also be in ptracestop(), having just
2847 				 * received the SIGKILL, but this thread was
2848 				 * unsuspended first.
2849 				 */
2850 				td->td_dbgflags &= ~TDB_XSIG;
2851 				td->td_xsig = SIGKILL;
2852 				p->p_ptevents = 0;
2853 				break;
2854 			}
2855 			if (p->p_flag & P_SINGLE_EXIT &&
2856 			    !(td->td_dbgflags & TDB_EXIT)) {
2857 				/*
2858 				 * Ignore ptrace stops except for thread exit
2859 				 * events when the process exits.
2860 				 */
2861 				td->td_dbgflags &= ~TDB_XSIG;
2862 				PROC_SUNLOCK(p);
2863 				return (0);
2864 			}
2865 
2866 			/*
2867 			 * Make wait(2) work.  Ensure that right after the
2868 			 * attach, the thread which was decided to become the
2869 			 * leader of attach gets reported to the waiter.
2870 			 * Otherwise, just avoid overwriting another thread's
2871 			 * assignment to p_xthread.  If another thread has
2872 			 * already set p_xthread, the current thread will get
2873 			 * a chance to report itself upon the next iteration.
2874 			 */
2875 			if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2876 			    ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2877 			    p->p_xthread == NULL)) {
2878 				p->p_xsig = sig;
2879 				p->p_xthread = td;
2880 
2881 				/*
2882 				 * If we are on sleepqueue already,
2883 				 * let sleepqueue code decide if it
2884 				 * needs to go sleep after attach.
2885 				 */
2886 				if (td->td_wchan == NULL)
2887 					td->td_dbgflags &= ~TDB_FSTP;
2888 
2889 				p->p_flag2 &= ~P2_PTRACE_FSTP;
2890 				p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2891 				sig_suspend_threads(td, p);
2892 			}
2893 			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2894 				td->td_dbgflags &= ~TDB_STOPATFORK;
2895 			}
2896 stopme:
2897 			td->td_dbgflags |= TDB_SSWITCH;
2898 			thread_suspend_switch(td, p);
2899 			td->td_dbgflags &= ~TDB_SSWITCH;
2900 			if ((td->td_dbgflags & (TDB_COREDUMPREQ |
2901 			    TDB_SCREMOTEREQ)) != 0) {
2902 				MPASS((td->td_dbgflags & (TDB_COREDUMPREQ |
2903 				    TDB_SCREMOTEREQ)) !=
2904 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2905 				PROC_SUNLOCK(p);
2906 				ptrace_remotereq(td, td->td_dbgflags &
2907 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2908 				PROC_SLOCK(p);
2909 				goto stopme;
2910 			}
2911 			if (p->p_xthread == td)
2912 				p->p_xthread = NULL;
2913 			if (!(p->p_flag & P_TRACED))
2914 				break;
2915 			if (td->td_dbgflags & TDB_SUSPEND) {
2916 				if (p->p_flag & P_SINGLE_EXIT)
2917 					break;
2918 				goto stopme;
2919 			}
2920 		}
2921 		PROC_SUNLOCK(p);
2922 	}
2923 
2924 	if (si != NULL && sig == td->td_xsig) {
2925 		/* Parent wants us to take the original signal unchanged. */
2926 		si->ksi_flags |= KSI_HEAD;
2927 		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2928 			si->ksi_signo = 0;
2929 	} else if (td->td_xsig != 0) {
2930 		/*
2931 		 * If parent wants us to take a new signal, then it will leave
2932 		 * it in td->td_xsig; otherwise we just look for signals again.
2933 		 */
2934 		ksiginfo_init(&ksi);
2935 		ksi.ksi_signo = td->td_xsig;
2936 		ksi.ksi_flags |= KSI_PTRACE;
2937 		td2 = sigtd(p, td->td_xsig, false);
2938 		tdsendsignal(p, td2, td->td_xsig, &ksi);
2939 		if (td != td2)
2940 			return (0);
2941 	}
2942 
2943 	return (td->td_xsig);
2944 }
2945 
2946 static void
reschedule_signals(struct proc * p,sigset_t block,int flags)2947 reschedule_signals(struct proc *p, sigset_t block, int flags)
2948 {
2949 	struct sigacts *ps;
2950 	struct thread *td;
2951 	int sig;
2952 	bool fastblk, pslocked;
2953 
2954 	PROC_LOCK_ASSERT(p, MA_OWNED);
2955 	ps = p->p_sigacts;
2956 	pslocked = (flags & SIGPROCMASK_PS_LOCKED) != 0;
2957 	mtx_assert(&ps->ps_mtx, pslocked ? MA_OWNED : MA_NOTOWNED);
2958 	if (SIGISEMPTY(p->p_siglist))
2959 		return;
2960 	SIGSETAND(block, p->p_siglist);
2961 	fastblk = (flags & SIGPROCMASK_FASTBLK) != 0;
2962 	SIG_FOREACH(sig, &block) {
2963 		td = sigtd(p, sig, fastblk);
2964 
2965 		/*
2966 		 * If sigtd() selected us despite sigfastblock is
2967 		 * blocking, do not activate AST or wake us, to avoid
2968 		 * loop in AST handler.
2969 		 */
2970 		if (fastblk && td == curthread)
2971 			continue;
2972 
2973 		signotify(td);
2974 		if (!pslocked)
2975 			mtx_lock(&ps->ps_mtx);
2976 		if (p->p_flag & P_TRACED ||
2977 		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2978 		    !SIGISMEMBER(td->td_sigmask, sig))) {
2979 			tdsigwakeup(td, sig, SIG_CATCH,
2980 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2981 			    ERESTART));
2982 		}
2983 		if (!pslocked)
2984 			mtx_unlock(&ps->ps_mtx);
2985 	}
2986 }
2987 
2988 void
tdsigcleanup(struct thread * td)2989 tdsigcleanup(struct thread *td)
2990 {
2991 	struct proc *p;
2992 	sigset_t unblocked;
2993 
2994 	p = td->td_proc;
2995 	PROC_LOCK_ASSERT(p, MA_OWNED);
2996 
2997 	sigqueue_flush(&td->td_sigqueue);
2998 	if (p->p_numthreads == 1)
2999 		return;
3000 
3001 	/*
3002 	 * Since we cannot handle signals, notify signal post code
3003 	 * about this by filling the sigmask.
3004 	 *
3005 	 * Also, if needed, wake up thread(s) that do not block the
3006 	 * same signals as the exiting thread, since the thread might
3007 	 * have been selected for delivery and woken up.
3008 	 */
3009 	SIGFILLSET(unblocked);
3010 	SIGSETNAND(unblocked, td->td_sigmask);
3011 	SIGFILLSET(td->td_sigmask);
3012 	reschedule_signals(p, unblocked, 0);
3013 
3014 }
3015 
3016 static int
sigdeferstop_curr_flags(int cflags)3017 sigdeferstop_curr_flags(int cflags)
3018 {
3019 
3020 	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
3021 	    (cflags & TDF_SBDRY) != 0);
3022 	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
3023 }
3024 
3025 /*
3026  * Defer the delivery of SIGSTOP for the current thread, according to
3027  * the requested mode.  Returns previous flags, which must be restored
3028  * by sigallowstop().
3029  *
3030  * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
3031  * cleared by the current thread, which allow the lock-less read-only
3032  * accesses below.
3033  */
3034 int
sigdeferstop_impl(int mode)3035 sigdeferstop_impl(int mode)
3036 {
3037 	struct thread *td;
3038 	int cflags, nflags;
3039 
3040 	td = curthread;
3041 	cflags = sigdeferstop_curr_flags(td->td_flags);
3042 	switch (mode) {
3043 	case SIGDEFERSTOP_NOP:
3044 		nflags = cflags;
3045 		break;
3046 	case SIGDEFERSTOP_OFF:
3047 		nflags = 0;
3048 		break;
3049 	case SIGDEFERSTOP_SILENT:
3050 		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
3051 		break;
3052 	case SIGDEFERSTOP_EINTR:
3053 		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
3054 		break;
3055 	case SIGDEFERSTOP_ERESTART:
3056 		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
3057 		break;
3058 	default:
3059 		panic("sigdeferstop: invalid mode %x", mode);
3060 		break;
3061 	}
3062 	if (cflags == nflags)
3063 		return (SIGDEFERSTOP_VAL_NCHG);
3064 	thread_lock(td);
3065 	td->td_flags = (td->td_flags & ~cflags) | nflags;
3066 	thread_unlock(td);
3067 	return (cflags);
3068 }
3069 
3070 /*
3071  * Restores the STOP handling mode, typically permitting the delivery
3072  * of SIGSTOP for the current thread.  This does not immediately
3073  * suspend if a stop was posted.  Instead, the thread will suspend
3074  * either via ast() or a subsequent interruptible sleep.
3075  */
3076 void
sigallowstop_impl(int prev)3077 sigallowstop_impl(int prev)
3078 {
3079 	struct thread *td;
3080 	int cflags;
3081 
3082 	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
3083 	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
3084 	    ("sigallowstop: incorrect previous mode %x", prev));
3085 	td = curthread;
3086 	cflags = sigdeferstop_curr_flags(td->td_flags);
3087 	if (cflags != prev) {
3088 		thread_lock(td);
3089 		td->td_flags = (td->td_flags & ~cflags) | prev;
3090 		thread_unlock(td);
3091 	}
3092 }
3093 
3094 enum sigstatus {
3095 	SIGSTATUS_HANDLE,
3096 	SIGSTATUS_HANDLED,
3097 	SIGSTATUS_IGNORE,
3098 	SIGSTATUS_SBDRY_STOP,
3099 };
3100 
3101 /*
3102  * The thread has signal "sig" pending.  Figure out what to do with it:
3103  *
3104  * _HANDLE     -> the caller should handle the signal
3105  * _HANDLED    -> handled internally, reload pending signal set
3106  * _IGNORE     -> ignored, remove from the set of pending signals and try the
3107  *                next pending signal
3108  * _SBDRY_STOP -> the signal should stop the thread but this is not
3109  *                permitted in the current context
3110  */
3111 static enum sigstatus
sigprocess(struct thread * td,int sig)3112 sigprocess(struct thread *td, int sig)
3113 {
3114 	struct proc *p;
3115 	struct sigacts *ps;
3116 	struct sigqueue *queue;
3117 	ksiginfo_t ksi;
3118 	int prop;
3119 
3120 	KASSERT(_SIG_VALID(sig), ("%s: invalid signal %d", __func__, sig));
3121 
3122 	p = td->td_proc;
3123 	ps = p->p_sigacts;
3124 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3125 	PROC_LOCK_ASSERT(p, MA_OWNED);
3126 
3127 	/*
3128 	 * We should allow pending but ignored signals below
3129 	 * if there is sigwait() active, or P_TRACED was
3130 	 * on when they were posted.
3131 	 */
3132 	if (SIGISMEMBER(ps->ps_sigignore, sig) &&
3133 	    (p->p_flag & P_TRACED) == 0 &&
3134 	    (td->td_flags & TDF_SIGWAIT) == 0) {
3135 		return (SIGSTATUS_IGNORE);
3136 	}
3137 
3138 	/*
3139 	 * If the process is going to single-thread mode to prepare
3140 	 * for exit, there is no sense in delivering any signal
3141 	 * to usermode.  Another important consequence is that
3142 	 * msleep(..., PCATCH, ...) now is only interruptible by a
3143 	 * suspend request.
3144 	 */
3145 	if ((p->p_flag2 & P2_WEXIT) != 0)
3146 		return (SIGSTATUS_IGNORE);
3147 
3148 	if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
3149 		/*
3150 		 * If traced, always stop.
3151 		 * Remove old signal from queue before the stop.
3152 		 * XXX shrug off debugger, it causes siginfo to
3153 		 * be thrown away.
3154 		 */
3155 		queue = &td->td_sigqueue;
3156 		ksiginfo_init(&ksi);
3157 		if (sigqueue_get(queue, sig, &ksi) == 0) {
3158 			queue = &p->p_sigqueue;
3159 			sigqueue_get(queue, sig, &ksi);
3160 		}
3161 		td->td_si = ksi.ksi_info;
3162 
3163 		mtx_unlock(&ps->ps_mtx);
3164 		sig = ptracestop(td, sig, &ksi);
3165 		mtx_lock(&ps->ps_mtx);
3166 
3167 		td->td_si.si_signo = 0;
3168 
3169 		/*
3170 		 * Keep looking if the debugger discarded or
3171 		 * replaced the signal.
3172 		 */
3173 		if (sig == 0)
3174 			return (SIGSTATUS_HANDLED);
3175 
3176 		/*
3177 		 * If the signal became masked, re-queue it.
3178 		 */
3179 		if (SIGISMEMBER(td->td_sigmask, sig)) {
3180 			ksi.ksi_flags |= KSI_HEAD;
3181 			sigqueue_add(&p->p_sigqueue, sig, &ksi);
3182 			return (SIGSTATUS_HANDLED);
3183 		}
3184 
3185 		/*
3186 		 * If the traced bit got turned off, requeue the signal and
3187 		 * reload the set of pending signals.  This ensures that p_sig*
3188 		 * and p_sigact are consistent.
3189 		 */
3190 		if ((p->p_flag & P_TRACED) == 0) {
3191 			if ((ksi.ksi_flags & KSI_PTRACE) == 0) {
3192 				ksi.ksi_flags |= KSI_HEAD;
3193 				sigqueue_add(queue, sig, &ksi);
3194 			}
3195 			return (SIGSTATUS_HANDLED);
3196 		}
3197 	}
3198 
3199 	/*
3200 	 * Decide whether the signal should be returned.
3201 	 * Return the signal's number, or fall through
3202 	 * to clear it from the pending mask.
3203 	 */
3204 	switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
3205 	case (intptr_t)SIG_DFL:
3206 		/*
3207 		 * Don't take default actions on system processes.
3208 		 */
3209 		if (p->p_pid <= 1) {
3210 #ifdef DIAGNOSTIC
3211 			/*
3212 			 * Are you sure you want to ignore SIGSEGV
3213 			 * in init? XXX
3214 			 */
3215 			printf("Process (pid %lu) got signal %d\n",
3216 				(u_long)p->p_pid, sig);
3217 #endif
3218 			return (SIGSTATUS_IGNORE);
3219 		}
3220 
3221 		/*
3222 		 * If there is a pending stop signal to process with
3223 		 * default action, stop here, then clear the signal.
3224 		 * Traced or exiting processes should ignore stops.
3225 		 * Additionally, a member of an orphaned process group
3226 		 * should ignore tty stops.
3227 		 */
3228 		prop = sigprop(sig);
3229 		if (prop & SIGPROP_STOP) {
3230 			mtx_unlock(&ps->ps_mtx);
3231 			if ((p->p_flag & (P_TRACED | P_WEXIT |
3232 			    P_SINGLE_EXIT)) != 0 || ((p->p_pgrp->
3233 			    pg_flags & PGRP_ORPHANED) != 0 &&
3234 			    (prop & SIGPROP_TTYSTOP) != 0)) {
3235 				mtx_lock(&ps->ps_mtx);
3236 				return (SIGSTATUS_IGNORE);
3237 			}
3238 			if (TD_SBDRY_INTR(td)) {
3239 				KASSERT((td->td_flags & TDF_SBDRY) != 0,
3240 				    ("lost TDF_SBDRY"));
3241 				mtx_lock(&ps->ps_mtx);
3242 				return (SIGSTATUS_SBDRY_STOP);
3243 			}
3244 			WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
3245 			    &p->p_mtx.lock_object, "Catching SIGSTOP");
3246 			sigqueue_delete(&td->td_sigqueue, sig);
3247 			sigqueue_delete(&p->p_sigqueue, sig);
3248 			p->p_flag |= P_STOPPED_SIG;
3249 			p->p_xsig = sig;
3250 			PROC_SLOCK(p);
3251 			sig_suspend_threads(td, p);
3252 			thread_suspend_switch(td, p);
3253 			PROC_SUNLOCK(p);
3254 			mtx_lock(&ps->ps_mtx);
3255 			return (SIGSTATUS_HANDLED);
3256 		} else if ((prop & SIGPROP_IGNORE) != 0 &&
3257 		    (td->td_flags & TDF_SIGWAIT) == 0) {
3258 			/*
3259 			 * Default action is to ignore; drop it if
3260 			 * not in kern_sigtimedwait().
3261 			 */
3262 			return (SIGSTATUS_IGNORE);
3263 		} else {
3264 			return (SIGSTATUS_HANDLE);
3265 		}
3266 
3267 	case (intptr_t)SIG_IGN:
3268 		if ((td->td_flags & TDF_SIGWAIT) == 0)
3269 			return (SIGSTATUS_IGNORE);
3270 		else
3271 			return (SIGSTATUS_HANDLE);
3272 
3273 	default:
3274 		/*
3275 		 * This signal has an action, let postsig() process it.
3276 		 */
3277 		return (SIGSTATUS_HANDLE);
3278 	}
3279 }
3280 
3281 /*
3282  * If the current process has received a signal (should be caught or cause
3283  * termination, should interrupt current syscall), return the signal number.
3284  * Stop signals with default action are processed immediately, then cleared;
3285  * they aren't returned.  This is checked after each entry to the system for
3286  * a syscall or trap (though this can usually be done without calling
3287  * issignal by checking the pending signal masks in cursig.) The normal call
3288  * sequence is
3289  *
3290  *	while (sig = cursig(curthread))
3291  *		postsig(sig);
3292  */
3293 static int
issignal(struct thread * td)3294 issignal(struct thread *td)
3295 {
3296 	struct proc *p;
3297 	sigset_t sigpending;
3298 	int sig;
3299 
3300 	p = td->td_proc;
3301 	PROC_LOCK_ASSERT(p, MA_OWNED);
3302 
3303 	for (;;) {
3304 		sigpending = td->td_sigqueue.sq_signals;
3305 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
3306 		SIGSETNAND(sigpending, td->td_sigmask);
3307 
3308 		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
3309 		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
3310 			SIG_STOPSIGMASK(sigpending);
3311 		if (SIGISEMPTY(sigpending))	/* no signal to send */
3312 			return (0);
3313 
3314 		/*
3315 		 * Do fast sigblock if requested by usermode.  Since
3316 		 * we do know that there was a signal pending at this
3317 		 * point, set the FAST_SIGBLOCK_PEND as indicator for
3318 		 * usermode to perform a dummy call to
3319 		 * FAST_SIGBLOCK_UNBLOCK, which causes immediate
3320 		 * delivery of postponed pending signal.
3321 		 */
3322 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
3323 			if (td->td_sigblock_val != 0)
3324 				SIGSETNAND(sigpending, fastblock_mask);
3325 			if (SIGISEMPTY(sigpending)) {
3326 				td->td_pflags |= TDP_SIGFASTPENDING;
3327 				return (0);
3328 			}
3329 		}
3330 
3331 		if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
3332 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
3333 		    SIGISMEMBER(sigpending, SIGSTOP)) {
3334 			/*
3335 			 * If debugger just attached, always consume
3336 			 * SIGSTOP from ptrace(PT_ATTACH) first, to
3337 			 * execute the debugger attach ritual in
3338 			 * order.
3339 			 */
3340 			td->td_dbgflags |= TDB_FSTP;
3341 			SIGEMPTYSET(sigpending);
3342 			SIGADDSET(sigpending, SIGSTOP);
3343 		}
3344 
3345 		SIG_FOREACH(sig, &sigpending) {
3346 			switch (sigprocess(td, sig)) {
3347 			case SIGSTATUS_HANDLE:
3348 				return (sig);
3349 			case SIGSTATUS_HANDLED:
3350 				goto next;
3351 			case SIGSTATUS_IGNORE:
3352 				sigqueue_delete(&td->td_sigqueue, sig);
3353 				sigqueue_delete(&p->p_sigqueue, sig);
3354 				break;
3355 			case SIGSTATUS_SBDRY_STOP:
3356 				return (-1);
3357 			}
3358 		}
3359 next:;
3360 	}
3361 }
3362 
3363 void
thread_stopped(struct proc * p)3364 thread_stopped(struct proc *p)
3365 {
3366 	int n;
3367 
3368 	PROC_LOCK_ASSERT(p, MA_OWNED);
3369 	PROC_SLOCK_ASSERT(p, MA_OWNED);
3370 	n = p->p_suspcount;
3371 	if (p == curproc)
3372 		n++;
3373 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3374 		PROC_SUNLOCK(p);
3375 		p->p_flag &= ~P_WAITED;
3376 		PROC_LOCK(p->p_pptr);
3377 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
3378 			CLD_TRAPPED : CLD_STOPPED);
3379 		PROC_UNLOCK(p->p_pptr);
3380 		PROC_SLOCK(p);
3381 	}
3382 }
3383 
3384 /*
3385  * Take the action for the specified signal
3386  * from the current set of pending signals.
3387  */
3388 int
postsig(int sig)3389 postsig(int sig)
3390 {
3391 	struct thread *td;
3392 	struct proc *p;
3393 	struct sigacts *ps;
3394 	sig_t action;
3395 	ksiginfo_t ksi;
3396 	sigset_t returnmask;
3397 
3398 	KASSERT(sig != 0, ("postsig"));
3399 
3400 	td = curthread;
3401 	p = td->td_proc;
3402 	PROC_LOCK_ASSERT(p, MA_OWNED);
3403 	ps = p->p_sigacts;
3404 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3405 	ksiginfo_init(&ksi);
3406 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3407 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3408 		return (0);
3409 	ksi.ksi_signo = sig;
3410 	if (ksi.ksi_code == SI_TIMER)
3411 		itimer_accept(p, ksi.ksi_timerid, &ksi);
3412 	action = ps->ps_sigact[_SIG_IDX(sig)];
3413 #ifdef KTRACE
3414 	if (KTRPOINT(td, KTR_PSIG))
3415 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3416 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3417 #endif
3418 
3419 	if (action == SIG_DFL) {
3420 		/*
3421 		 * Default action, where the default is to kill
3422 		 * the process.  (Other cases were ignored above.)
3423 		 */
3424 		mtx_unlock(&ps->ps_mtx);
3425 		proc_td_siginfo_capture(td, &ksi.ksi_info);
3426 		sigexit(td, sig);
3427 		/* NOTREACHED */
3428 	} else {
3429 		/*
3430 		 * If we get here, the signal must be caught.
3431 		 */
3432 		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3433 		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3434 		    ("postsig action: blocked sig %d", sig));
3435 
3436 		/*
3437 		 * Set the new mask value and also defer further
3438 		 * occurrences of this signal.
3439 		 *
3440 		 * Special case: user has done a sigsuspend.  Here the
3441 		 * current mask is not of interest, but rather the
3442 		 * mask from before the sigsuspend is what we want
3443 		 * restored after the signal processing is completed.
3444 		 */
3445 		if (td->td_pflags & TDP_OLDMASK) {
3446 			returnmask = td->td_oldsigmask;
3447 			td->td_pflags &= ~TDP_OLDMASK;
3448 		} else
3449 			returnmask = td->td_sigmask;
3450 
3451 		if (p->p_sig == sig) {
3452 			p->p_sig = 0;
3453 		}
3454 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3455 		postsig_done(sig, td, ps);
3456 	}
3457 	return (1);
3458 }
3459 
3460 int
sig_ast_checksusp(struct thread * td)3461 sig_ast_checksusp(struct thread *td)
3462 {
3463 	struct proc *p __diagused;
3464 	int ret;
3465 
3466 	p = td->td_proc;
3467 	PROC_LOCK_ASSERT(p, MA_OWNED);
3468 
3469 	if (!td_ast_pending(td, TDA_SUSPEND))
3470 		return (0);
3471 
3472 	ret = thread_suspend_check(1);
3473 	MPASS(ret == 0 || ret == EINTR || ret == ERESTART);
3474 	return (ret);
3475 }
3476 
3477 int
sig_ast_needsigchk(struct thread * td)3478 sig_ast_needsigchk(struct thread *td)
3479 {
3480 	struct proc *p;
3481 	struct sigacts *ps;
3482 	int ret, sig;
3483 
3484 	p = td->td_proc;
3485 	PROC_LOCK_ASSERT(p, MA_OWNED);
3486 
3487 	if (!td_ast_pending(td, TDA_SIG))
3488 		return (0);
3489 
3490 	ps = p->p_sigacts;
3491 	mtx_lock(&ps->ps_mtx);
3492 	sig = cursig(td);
3493 	if (sig == -1) {
3494 		mtx_unlock(&ps->ps_mtx);
3495 		KASSERT((td->td_flags & TDF_SBDRY) != 0, ("lost TDF_SBDRY"));
3496 		KASSERT(TD_SBDRY_INTR(td),
3497 		    ("lost TDF_SERESTART of TDF_SEINTR"));
3498 		KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
3499 		    (TDF_SEINTR | TDF_SERESTART),
3500 		    ("both TDF_SEINTR and TDF_SERESTART"));
3501 		ret = TD_SBDRY_ERRNO(td);
3502 	} else if (sig != 0) {
3503 		ret = SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR : ERESTART;
3504 		mtx_unlock(&ps->ps_mtx);
3505 	} else {
3506 		mtx_unlock(&ps->ps_mtx);
3507 		ret = 0;
3508 	}
3509 
3510 	/*
3511 	 * Do not go into sleep if this thread was the ptrace(2)
3512 	 * attach leader.  cursig() consumed SIGSTOP from PT_ATTACH,
3513 	 * but we usually act on the signal by interrupting sleep, and
3514 	 * should do that here as well.
3515 	 */
3516 	if ((td->td_dbgflags & TDB_FSTP) != 0) {
3517 		if (ret == 0)
3518 			ret = EINTR;
3519 		td->td_dbgflags &= ~TDB_FSTP;
3520 	}
3521 
3522 	return (ret);
3523 }
3524 
3525 int
sig_intr(void)3526 sig_intr(void)
3527 {
3528 	struct thread *td;
3529 	struct proc *p;
3530 	int ret;
3531 
3532 	td = curthread;
3533 	if (!td_ast_pending(td, TDA_SIG) && !td_ast_pending(td, TDA_SUSPEND))
3534 		return (0);
3535 
3536 	p = td->td_proc;
3537 
3538 	PROC_LOCK(p);
3539 	ret = sig_ast_checksusp(td);
3540 	if (ret == 0)
3541 		ret = sig_ast_needsigchk(td);
3542 	PROC_UNLOCK(p);
3543 	return (ret);
3544 }
3545 
3546 bool
curproc_sigkilled(void)3547 curproc_sigkilled(void)
3548 {
3549 	struct thread *td;
3550 	struct proc *p;
3551 	struct sigacts *ps;
3552 	bool res;
3553 
3554 	td = curthread;
3555 	if (!td_ast_pending(td, TDA_SIG))
3556 		return (false);
3557 
3558 	p = td->td_proc;
3559 	PROC_LOCK(p);
3560 	ps = p->p_sigacts;
3561 	mtx_lock(&ps->ps_mtx);
3562 	res = SIGISMEMBER(td->td_sigqueue.sq_signals, SIGKILL) ||
3563 	    SIGISMEMBER(p->p_sigqueue.sq_signals, SIGKILL);
3564 	mtx_unlock(&ps->ps_mtx);
3565 	PROC_UNLOCK(p);
3566 	return (res);
3567 }
3568 
3569 void
proc_wkilled(struct proc * p)3570 proc_wkilled(struct proc *p)
3571 {
3572 
3573 	PROC_LOCK_ASSERT(p, MA_OWNED);
3574 	if ((p->p_flag & P_WKILLED) == 0)
3575 		p->p_flag |= P_WKILLED;
3576 }
3577 
3578 /*
3579  * Kill the current process for stated reason.
3580  */
3581 void
killproc(struct proc * p,const char * why)3582 killproc(struct proc *p, const char *why)
3583 {
3584 
3585 	PROC_LOCK_ASSERT(p, MA_OWNED);
3586 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3587 	    p->p_comm);
3588 	log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3589 	    p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3590 	    p->p_ucred->cr_uid, why);
3591 	proc_wkilled(p);
3592 	kern_psignal(p, SIGKILL);
3593 }
3594 
3595 /*
3596  * Force the current process to exit with the specified signal, dumping core
3597  * if appropriate.  We bypass the normal tests for masked and caught signals,
3598  * allowing unrecoverable failures to terminate the process without changing
3599  * signal state.  Mark the accounting record with the signal termination.
3600  * If dumping core, save the signal number for the debugger.  Calls exit and
3601  * does not return.
3602  */
3603 void
sigexit(struct thread * td,int sig)3604 sigexit(struct thread *td, int sig)
3605 {
3606 	struct proc *p = td->td_proc;
3607 	const char *coreinfo;
3608 	int rv;
3609 	bool logexit;
3610 
3611 	PROC_LOCK_ASSERT(p, MA_OWNED);
3612 	proc_set_p2_wexit(p);
3613 
3614 	p->p_acflag |= AXSIG;
3615 	if ((p->p_flag2 & P2_LOGSIGEXIT_CTL) == 0)
3616 		logexit = kern_logsigexit != 0;
3617 	else
3618 		logexit = (p->p_flag2 & P2_LOGSIGEXIT_ENABLE) != 0;
3619 
3620 	/*
3621 	 * We must be single-threading to generate a core dump.  This
3622 	 * ensures that the registers in the core file are up-to-date.
3623 	 * Also, the ELF dump handler assumes that the thread list doesn't
3624 	 * change out from under it.
3625 	 *
3626 	 * XXX If another thread attempts to single-thread before us
3627 	 *     (e.g. via fork()), we won't get a dump at all.
3628 	 */
3629 	if ((sigprop(sig) & SIGPROP_CORE) &&
3630 	    thread_single(p, SINGLE_NO_EXIT) == 0) {
3631 		p->p_sig = sig;
3632 		/*
3633 		 * Log signals which would cause core dumps
3634 		 * (Log as LOG_INFO to appease those who don't want
3635 		 * these messages.)
3636 		 * XXX : Todo, as well as euid, write out ruid too
3637 		 * Note that coredump() drops proc lock.
3638 		 */
3639 		rv = coredump(td);
3640 		switch (rv) {
3641 		case 0:
3642 			sig |= WCOREFLAG;
3643 			coreinfo = " (core dumped)";
3644 			break;
3645 		case EFAULT:
3646 			coreinfo = " (no core dump - bad address)";
3647 			break;
3648 		case EINVAL:
3649 			coreinfo = " (no core dump - invalid argument)";
3650 			break;
3651 		case EFBIG:
3652 			coreinfo = " (no core dump - too large)";
3653 			break;
3654 		default:
3655 			coreinfo = " (no core dump - other error)";
3656 			break;
3657 		}
3658 		if (logexit)
3659 			log(LOG_INFO,
3660 			    "pid %d (%s), jid %d, uid %d: exited on "
3661 			    "signal %d%s\n", p->p_pid, p->p_comm,
3662 			    p->p_ucred->cr_prison->pr_id,
3663 			    td->td_ucred->cr_uid,
3664 			    sig &~ WCOREFLAG, coreinfo);
3665 	} else
3666 		PROC_UNLOCK(p);
3667 	exit1(td, 0, sig);
3668 	/* NOTREACHED */
3669 }
3670 
3671 /*
3672  * Send queued SIGCHLD to parent when child process's state
3673  * is changed.
3674  */
3675 static void
sigparent(struct proc * p,int reason,int status)3676 sigparent(struct proc *p, int reason, int status)
3677 {
3678 	PROC_LOCK_ASSERT(p, MA_OWNED);
3679 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3680 
3681 	if (p->p_ksi != NULL) {
3682 		p->p_ksi->ksi_signo  = SIGCHLD;
3683 		p->p_ksi->ksi_code   = reason;
3684 		p->p_ksi->ksi_status = status;
3685 		p->p_ksi->ksi_pid    = p->p_pid;
3686 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3687 		if (KSI_ONQ(p->p_ksi))
3688 			return;
3689 	}
3690 	pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3691 }
3692 
3693 static void
childproc_jobstate(struct proc * p,int reason,int sig)3694 childproc_jobstate(struct proc *p, int reason, int sig)
3695 {
3696 	struct sigacts *ps;
3697 
3698 	PROC_LOCK_ASSERT(p, MA_OWNED);
3699 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3700 
3701 	/*
3702 	 * Wake up parent sleeping in kern_wait(), also send
3703 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3704 	 * that parent will awake, because parent may masked
3705 	 * the signal.
3706 	 */
3707 	p->p_pptr->p_flag |= P_STATCHILD;
3708 	wakeup(p->p_pptr);
3709 
3710 	ps = p->p_pptr->p_sigacts;
3711 	mtx_lock(&ps->ps_mtx);
3712 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3713 		mtx_unlock(&ps->ps_mtx);
3714 		sigparent(p, reason, sig);
3715 	} else
3716 		mtx_unlock(&ps->ps_mtx);
3717 }
3718 
3719 void
childproc_stopped(struct proc * p,int reason)3720 childproc_stopped(struct proc *p, int reason)
3721 {
3722 
3723 	childproc_jobstate(p, reason, p->p_xsig);
3724 }
3725 
3726 void
childproc_continued(struct proc * p)3727 childproc_continued(struct proc *p)
3728 {
3729 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3730 }
3731 
3732 void
childproc_exited(struct proc * p)3733 childproc_exited(struct proc *p)
3734 {
3735 	int reason, status;
3736 
3737 	if (WCOREDUMP(p->p_xsig)) {
3738 		reason = CLD_DUMPED;
3739 		status = WTERMSIG(p->p_xsig);
3740 	} else if (WIFSIGNALED(p->p_xsig)) {
3741 		reason = CLD_KILLED;
3742 		status = WTERMSIG(p->p_xsig);
3743 	} else {
3744 		reason = CLD_EXITED;
3745 		status = p->p_xexit;
3746 	}
3747 	/*
3748 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3749 	 * done in exit1().
3750 	 */
3751 	sigparent(p, reason, status);
3752 }
3753 
3754 #define	MAX_NUM_CORE_FILES 100000
3755 #ifndef NUM_CORE_FILES
3756 #define	NUM_CORE_FILES 5
3757 #endif
3758 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3759 static int num_cores = NUM_CORE_FILES;
3760 
3761 static int
sysctl_debug_num_cores_check(SYSCTL_HANDLER_ARGS)3762 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3763 {
3764 	int error;
3765 	int new_val;
3766 
3767 	new_val = num_cores;
3768 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3769 	if (error != 0 || req->newptr == NULL)
3770 		return (error);
3771 	if (new_val > MAX_NUM_CORE_FILES)
3772 		new_val = MAX_NUM_CORE_FILES;
3773 	if (new_val < 0)
3774 		new_val = 0;
3775 	num_cores = new_val;
3776 	return (0);
3777 }
3778 SYSCTL_PROC(_debug, OID_AUTO, ncores,
3779     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
3780     sysctl_debug_num_cores_check, "I",
3781     "Maximum number of generated process corefiles while using index format");
3782 
3783 #define	GZIP_SUFFIX	".gz"
3784 #define	ZSTD_SUFFIX	".zst"
3785 
3786 int compress_user_cores = 0;
3787 
3788 static int
sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)3789 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3790 {
3791 	int error, val;
3792 
3793 	val = compress_user_cores;
3794 	error = sysctl_handle_int(oidp, &val, 0, req);
3795 	if (error != 0 || req->newptr == NULL)
3796 		return (error);
3797 	if (val != 0 && !compressor_avail(val))
3798 		return (EINVAL);
3799 	compress_user_cores = val;
3800 	return (error);
3801 }
3802 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores,
3803     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, 0, sizeof(int),
3804     sysctl_compress_user_cores, "I",
3805     "Enable compression of user corefiles ("
3806     __XSTRING(COMPRESS_GZIP) " = gzip, "
3807     __XSTRING(COMPRESS_ZSTD) " = zstd)");
3808 
3809 int compress_user_cores_level = 6;
3810 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3811     &compress_user_cores_level, 0,
3812     "Corefile compression level");
3813 
3814 /*
3815  * Protect the access to corefilename[] by allproc_lock.
3816  */
3817 #define	corefilename_lock	allproc_lock
3818 
3819 static char corefilename[MAXPATHLEN] = {"%N.core"};
3820 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3821 
3822 static int
sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)3823 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3824 {
3825 	int error;
3826 
3827 	sx_xlock(&corefilename_lock);
3828 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3829 	    req);
3830 	sx_xunlock(&corefilename_lock);
3831 
3832 	return (error);
3833 }
3834 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3835     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3836     "Process corefile name format string");
3837 
3838 static void
vnode_close_locked(struct thread * td,struct vnode * vp)3839 vnode_close_locked(struct thread *td, struct vnode *vp)
3840 {
3841 
3842 	VOP_UNLOCK(vp);
3843 	vn_close(vp, FWRITE, td->td_ucred, td);
3844 }
3845 
3846 /*
3847  * If the core format has a %I in it, then we need to check
3848  * for existing corefiles before defining a name.
3849  * To do this we iterate over 0..ncores to find a
3850  * non-existing core file name to use. If all core files are
3851  * already used we choose the oldest one.
3852  */
3853 static int
corefile_open_last(struct thread * td,char * name,int indexpos,int indexlen,int ncores,struct vnode ** vpp)3854 corefile_open_last(struct thread *td, char *name, int indexpos,
3855     int indexlen, int ncores, struct vnode **vpp)
3856 {
3857 	struct vnode *oldvp, *nextvp, *vp;
3858 	struct vattr vattr;
3859 	struct nameidata nd;
3860 	int error, i, flags, oflags, cmode;
3861 	char ch;
3862 	struct timespec lasttime;
3863 
3864 	nextvp = oldvp = NULL;
3865 	cmode = S_IRUSR | S_IWUSR;
3866 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3867 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3868 
3869 	for (i = 0; i < ncores; i++) {
3870 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3871 
3872 		ch = name[indexpos + indexlen];
3873 		(void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3874 		    i);
3875 		name[indexpos + indexlen] = ch;
3876 
3877 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
3878 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3879 		    NULL);
3880 		if (error != 0)
3881 			break;
3882 
3883 		vp = nd.ni_vp;
3884 		NDFREE_PNBUF(&nd);
3885 		if ((flags & O_CREAT) == O_CREAT) {
3886 			nextvp = vp;
3887 			break;
3888 		}
3889 
3890 		error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3891 		if (error != 0) {
3892 			vnode_close_locked(td, vp);
3893 			break;
3894 		}
3895 
3896 		if (oldvp == NULL ||
3897 		    lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3898 		    (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3899 		    lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3900 			if (oldvp != NULL)
3901 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3902 			oldvp = vp;
3903 			VOP_UNLOCK(oldvp);
3904 			lasttime = vattr.va_mtime;
3905 		} else {
3906 			vnode_close_locked(td, vp);
3907 		}
3908 	}
3909 
3910 	if (oldvp != NULL) {
3911 		if (nextvp == NULL) {
3912 			if ((td->td_proc->p_flag & P_SUGID) != 0) {
3913 				error = EFAULT;
3914 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3915 			} else {
3916 				nextvp = oldvp;
3917 				error = vn_lock(nextvp, LK_EXCLUSIVE);
3918 				if (error != 0) {
3919 					vn_close(nextvp, FWRITE, td->td_ucred,
3920 					    td);
3921 					nextvp = NULL;
3922 				}
3923 			}
3924 		} else {
3925 			vn_close(oldvp, FWRITE, td->td_ucred, td);
3926 		}
3927 	}
3928 	if (error != 0) {
3929 		if (nextvp != NULL)
3930 			vnode_close_locked(td, oldvp);
3931 	} else {
3932 		*vpp = nextvp;
3933 	}
3934 
3935 	return (error);
3936 }
3937 
3938 /*
3939  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3940  * Expand the name described in corefilename, using name, uid, and pid
3941  * and open/create core file.
3942  * corefilename is a printf-like string, with three format specifiers:
3943  *	%N	name of process ("name")
3944  *	%P	process id (pid)
3945  *	%U	user id (uid)
3946  * For example, "%N.core" is the default; they can be disabled completely
3947  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3948  * This is controlled by the sysctl variable kern.corefile (see above).
3949  */
3950 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)3951 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3952     int compress, int signum, struct vnode **vpp, char **namep)
3953 {
3954 	struct sbuf sb;
3955 	struct nameidata nd;
3956 	const char *format;
3957 	char *hostname, *name;
3958 	int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
3959 
3960 	hostname = NULL;
3961 	format = corefilename;
3962 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3963 	indexlen = 0;
3964 	indexpos = -1;
3965 	ncores = num_cores;
3966 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3967 	sx_slock(&corefilename_lock);
3968 	for (i = 0; format[i] != '\0'; i++) {
3969 		switch (format[i]) {
3970 		case '%':	/* Format character */
3971 			i++;
3972 			switch (format[i]) {
3973 			case '%':
3974 				sbuf_putc(&sb, '%');
3975 				break;
3976 			case 'H':	/* hostname */
3977 				if (hostname == NULL) {
3978 					hostname = malloc(MAXHOSTNAMELEN,
3979 					    M_TEMP, M_WAITOK);
3980 				}
3981 				getcredhostname(td->td_ucred, hostname,
3982 				    MAXHOSTNAMELEN);
3983 				sbuf_cat(&sb, hostname);
3984 				break;
3985 			case 'I':	/* autoincrementing index */
3986 				if (indexpos != -1) {
3987 					sbuf_printf(&sb, "%%I");
3988 					break;
3989 				}
3990 
3991 				indexpos = sbuf_len(&sb);
3992 				sbuf_printf(&sb, "%u", ncores - 1);
3993 				indexlen = sbuf_len(&sb) - indexpos;
3994 				break;
3995 			case 'N':	/* process name */
3996 				sbuf_printf(&sb, "%s", comm);
3997 				break;
3998 			case 'P':	/* process id */
3999 				sbuf_printf(&sb, "%u", pid);
4000 				break;
4001 			case 'S':	/* signal number */
4002 				sbuf_printf(&sb, "%i", signum);
4003 				break;
4004 			case 'U':	/* user id */
4005 				sbuf_printf(&sb, "%u", uid);
4006 				break;
4007 			default:
4008 				log(LOG_ERR,
4009 				    "Unknown format character %c in "
4010 				    "corename `%s'\n", format[i], format);
4011 				break;
4012 			}
4013 			break;
4014 		default:
4015 			sbuf_putc(&sb, format[i]);
4016 			break;
4017 		}
4018 	}
4019 	sx_sunlock(&corefilename_lock);
4020 	free(hostname, M_TEMP);
4021 	if (compress == COMPRESS_GZIP)
4022 		sbuf_cat(&sb, GZIP_SUFFIX);
4023 	else if (compress == COMPRESS_ZSTD)
4024 		sbuf_cat(&sb, ZSTD_SUFFIX);
4025 	if (sbuf_error(&sb) != 0) {
4026 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
4027 		    "long\n", (long)pid, comm, (u_long)uid);
4028 		sbuf_delete(&sb);
4029 		free(name, M_TEMP);
4030 		return (ENOMEM);
4031 	}
4032 	sbuf_finish(&sb);
4033 	sbuf_delete(&sb);
4034 
4035 	if (indexpos != -1) {
4036 		error = corefile_open_last(td, name, indexpos, indexlen, ncores,
4037 		    vpp);
4038 		if (error != 0) {
4039 			log(LOG_ERR,
4040 			    "pid %d (%s), uid (%u):  Path `%s' failed "
4041 			    "on initial open test, error = %d\n",
4042 			    pid, comm, uid, name, error);
4043 		}
4044 	} else {
4045 		cmode = S_IRUSR | S_IWUSR;
4046 		oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
4047 		    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
4048 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
4049 		if ((td->td_proc->p_flag & P_SUGID) != 0)
4050 			flags |= O_EXCL;
4051 
4052 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
4053 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
4054 		    NULL);
4055 		if (error == 0) {
4056 			*vpp = nd.ni_vp;
4057 			NDFREE_PNBUF(&nd);
4058 		}
4059 	}
4060 
4061 	if (error != 0) {
4062 #ifdef AUDIT
4063 		audit_proc_coredump(td, name, error);
4064 #endif
4065 		free(name, M_TEMP);
4066 		return (error);
4067 	}
4068 	*namep = name;
4069 	return (0);
4070 }
4071 
4072 /*
4073  * Dump a process' core.  The main routine does some
4074  * policy checking, and creates the name of the coredump;
4075  * then it passes on a vnode and a size limit to the process-specific
4076  * coredump routine if there is one; if there _is not_ one, it returns
4077  * ENOSYS; otherwise it returns the error from the process-specific routine.
4078  */
4079 
4080 static int
coredump(struct thread * td)4081 coredump(struct thread *td)
4082 {
4083 	struct proc *p = td->td_proc;
4084 	struct ucred *cred = td->td_ucred;
4085 	struct vnode *vp;
4086 	struct flock lf;
4087 	struct vattr vattr;
4088 	size_t fullpathsize;
4089 	int error, error1, locked;
4090 	char *name;			/* name of corefile */
4091 	void *rl_cookie;
4092 	off_t limit;
4093 	char *fullpath, *freepath = NULL;
4094 	struct sbuf *sb;
4095 
4096 	PROC_LOCK_ASSERT(p, MA_OWNED);
4097 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
4098 
4099 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
4100 	    (p->p_flag2 & P2_NOTRACE) != 0) {
4101 		PROC_UNLOCK(p);
4102 		return (EFAULT);
4103 	}
4104 
4105 	/*
4106 	 * Note that the bulk of limit checking is done after
4107 	 * the corefile is created.  The exception is if the limit
4108 	 * for corefiles is 0, in which case we don't bother
4109 	 * creating the corefile at all.  This layout means that
4110 	 * a corefile is truncated instead of not being created,
4111 	 * if it is larger than the limit.
4112 	 */
4113 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
4114 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
4115 		PROC_UNLOCK(p);
4116 		return (EFBIG);
4117 	}
4118 	PROC_UNLOCK(p);
4119 
4120 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
4121 	    compress_user_cores, p->p_sig, &vp, &name);
4122 	if (error != 0)
4123 		return (error);
4124 
4125 	/*
4126 	 * Don't dump to non-regular files or files with links.
4127 	 * Do not dump into system files. Effective user must own the corefile.
4128 	 */
4129 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
4130 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
4131 	    vattr.va_uid != cred->cr_uid) {
4132 		VOP_UNLOCK(vp);
4133 		error = EFAULT;
4134 		goto out;
4135 	}
4136 
4137 	VOP_UNLOCK(vp);
4138 
4139 	/* Postpone other writers, including core dumps of other processes. */
4140 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
4141 
4142 	lf.l_whence = SEEK_SET;
4143 	lf.l_start = 0;
4144 	lf.l_len = 0;
4145 	lf.l_type = F_WRLCK;
4146 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
4147 
4148 	VATTR_NULL(&vattr);
4149 	vattr.va_size = 0;
4150 	if (set_core_nodump_flag)
4151 		vattr.va_flags = UF_NODUMP;
4152 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4153 	VOP_SETATTR(vp, &vattr, cred);
4154 	VOP_UNLOCK(vp);
4155 	PROC_LOCK(p);
4156 	p->p_acflag |= ACORE;
4157 	PROC_UNLOCK(p);
4158 
4159 	if (p->p_sysent->sv_coredump != NULL) {
4160 		error = p->p_sysent->sv_coredump(td, vp, limit, 0);
4161 	} else {
4162 		error = ENOSYS;
4163 	}
4164 
4165 	if (locked) {
4166 		lf.l_type = F_UNLCK;
4167 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
4168 	}
4169 	vn_rangelock_unlock(vp, rl_cookie);
4170 
4171 	/*
4172 	 * Notify the userland helper that a process triggered a core dump.
4173 	 * This allows the helper to run an automated debugging session.
4174 	 */
4175 	if (error != 0 || coredump_devctl == 0)
4176 		goto out;
4177 	sb = sbuf_new_auto();
4178 	if (vn_fullpath_global(p->p_textvp, &fullpath, &freepath) != 0)
4179 		goto out2;
4180 	sbuf_cat(sb, "comm=\"");
4181 	devctl_safe_quote_sb(sb, fullpath);
4182 	free(freepath, M_TEMP);
4183 	sbuf_cat(sb, "\" core=\"");
4184 
4185 	/*
4186 	 * We can't lookup core file vp directly. When we're replacing a core, and
4187 	 * other random times, we flush the name cache, so it will fail. Instead,
4188 	 * if the path of the core is relative, add the current dir in front if it.
4189 	 */
4190 	if (name[0] != '/') {
4191 		fullpathsize = MAXPATHLEN;
4192 		freepath = malloc(fullpathsize, M_TEMP, M_WAITOK);
4193 		if (vn_getcwd(freepath, &fullpath, &fullpathsize) != 0) {
4194 			free(freepath, M_TEMP);
4195 			goto out2;
4196 		}
4197 		devctl_safe_quote_sb(sb, fullpath);
4198 		free(freepath, M_TEMP);
4199 		sbuf_putc(sb, '/');
4200 	}
4201 	devctl_safe_quote_sb(sb, name);
4202 	sbuf_putc(sb, '"');
4203 	if (sbuf_finish(sb) == 0)
4204 		devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
4205 out2:
4206 	sbuf_delete(sb);
4207 out:
4208 	error1 = vn_close(vp, FWRITE, cred, td);
4209 	if (error == 0)
4210 		error = error1;
4211 #ifdef AUDIT
4212 	audit_proc_coredump(td, name, error);
4213 #endif
4214 	free(name, M_TEMP);
4215 	return (error);
4216 }
4217 
4218 /*
4219  * Nonexistent system call-- signal process (may want to handle it).  Flag
4220  * error in case process won't see signal immediately (blocked or ignored).
4221  */
4222 #ifndef _SYS_SYSPROTO_H_
4223 struct nosys_args {
4224 	int	dummy;
4225 };
4226 #endif
4227 /* ARGSUSED */
4228 int
nosys(struct thread * td,struct nosys_args * args)4229 nosys(struct thread *td, struct nosys_args *args)
4230 {
4231 	struct proc *p;
4232 
4233 	p = td->td_proc;
4234 
4235 	if (SV_PROC_FLAG(p, SV_SIGSYS) != 0 && kern_signosys) {
4236 		PROC_LOCK(p);
4237 		tdsignal(td, SIGSYS);
4238 		PROC_UNLOCK(p);
4239 	}
4240 	if (kern_lognosys == 1 || kern_lognosys == 3) {
4241 		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4242 		    td->td_sa.code);
4243 	}
4244 	if (kern_lognosys == 2 || kern_lognosys == 3 ||
4245 	    (p->p_pid == 1 && (kern_lognosys & 3) == 0)) {
4246 		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4247 		    td->td_sa.code);
4248 	}
4249 	return (ENOSYS);
4250 }
4251 
4252 /*
4253  * Send a SIGIO or SIGURG signal to a process or process group using stored
4254  * credentials rather than those of the current process.
4255  */
4256 void
pgsigio(struct sigio ** sigiop,int sig,int checkctty)4257 pgsigio(struct sigio **sigiop, int sig, int checkctty)
4258 {
4259 	ksiginfo_t ksi;
4260 	struct sigio *sigio;
4261 
4262 	ksiginfo_init(&ksi);
4263 	ksi.ksi_signo = sig;
4264 	ksi.ksi_code = SI_KERNEL;
4265 
4266 	SIGIO_LOCK();
4267 	sigio = *sigiop;
4268 	if (sigio == NULL) {
4269 		SIGIO_UNLOCK();
4270 		return;
4271 	}
4272 	if (sigio->sio_pgid > 0) {
4273 		PROC_LOCK(sigio->sio_proc);
4274 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
4275 			kern_psignal(sigio->sio_proc, sig);
4276 		PROC_UNLOCK(sigio->sio_proc);
4277 	} else if (sigio->sio_pgid < 0) {
4278 		struct proc *p;
4279 
4280 		PGRP_LOCK(sigio->sio_pgrp);
4281 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
4282 			PROC_LOCK(p);
4283 			if (p->p_state == PRS_NORMAL &&
4284 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
4285 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
4286 				kern_psignal(p, sig);
4287 			PROC_UNLOCK(p);
4288 		}
4289 		PGRP_UNLOCK(sigio->sio_pgrp);
4290 	}
4291 	SIGIO_UNLOCK();
4292 }
4293 
4294 static int
filt_sigattach(struct knote * kn)4295 filt_sigattach(struct knote *kn)
4296 {
4297 	struct proc *p = curproc;
4298 
4299 	kn->kn_ptr.p_proc = p;
4300 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
4301 
4302 	knlist_add(p->p_klist, kn, 0);
4303 
4304 	return (0);
4305 }
4306 
4307 static void
filt_sigdetach(struct knote * kn)4308 filt_sigdetach(struct knote *kn)
4309 {
4310 	knlist_remove(kn->kn_knlist, kn, 0);
4311 }
4312 
4313 /*
4314  * signal knotes are shared with proc knotes, so we apply a mask to
4315  * the hint in order to differentiate them from process hints.  This
4316  * could be avoided by using a signal-specific knote list, but probably
4317  * isn't worth the trouble.
4318  */
4319 static int
filt_signal(struct knote * kn,long hint)4320 filt_signal(struct knote *kn, long hint)
4321 {
4322 
4323 	if (hint & NOTE_SIGNAL) {
4324 		hint &= ~NOTE_SIGNAL;
4325 
4326 		if (kn->kn_id == hint)
4327 			kn->kn_data++;
4328 	}
4329 	return (kn->kn_data != 0);
4330 }
4331 
4332 struct sigacts *
sigacts_alloc(void)4333 sigacts_alloc(void)
4334 {
4335 	struct sigacts *ps;
4336 
4337 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
4338 	refcount_init(&ps->ps_refcnt, 1);
4339 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
4340 	return (ps);
4341 }
4342 
4343 void
sigacts_free(struct sigacts * ps)4344 sigacts_free(struct sigacts *ps)
4345 {
4346 
4347 	if (refcount_release(&ps->ps_refcnt) == 0)
4348 		return;
4349 	mtx_destroy(&ps->ps_mtx);
4350 	free(ps, M_SUBPROC);
4351 }
4352 
4353 struct sigacts *
sigacts_hold(struct sigacts * ps)4354 sigacts_hold(struct sigacts *ps)
4355 {
4356 
4357 	refcount_acquire(&ps->ps_refcnt);
4358 	return (ps);
4359 }
4360 
4361 void
sigacts_copy(struct sigacts * dest,struct sigacts * src)4362 sigacts_copy(struct sigacts *dest, struct sigacts *src)
4363 {
4364 
4365 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
4366 	mtx_lock(&src->ps_mtx);
4367 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
4368 	mtx_unlock(&src->ps_mtx);
4369 }
4370 
4371 int
sigacts_shared(struct sigacts * ps)4372 sigacts_shared(struct sigacts *ps)
4373 {
4374 
4375 	return (ps->ps_refcnt > 1);
4376 }
4377 
4378 void
sig_drop_caught(struct proc * p)4379 sig_drop_caught(struct proc *p)
4380 {
4381 	int sig;
4382 	struct sigacts *ps;
4383 
4384 	ps = p->p_sigacts;
4385 	PROC_LOCK_ASSERT(p, MA_OWNED);
4386 	mtx_assert(&ps->ps_mtx, MA_OWNED);
4387 	SIG_FOREACH(sig, &ps->ps_sigcatch) {
4388 		sigdflt(ps, sig);
4389 		if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
4390 			sigqueue_delete_proc(p, sig);
4391 	}
4392 }
4393 
4394 static void
sigfastblock_failed(struct thread * td,bool sendsig,bool write)4395 sigfastblock_failed(struct thread *td, bool sendsig, bool write)
4396 {
4397 	ksiginfo_t ksi;
4398 
4399 	/*
4400 	 * Prevent further fetches and SIGSEGVs, allowing thread to
4401 	 * issue syscalls despite corruption.
4402 	 */
4403 	sigfastblock_clear(td);
4404 
4405 	if (!sendsig)
4406 		return;
4407 	ksiginfo_init_trap(&ksi);
4408 	ksi.ksi_signo = SIGSEGV;
4409 	ksi.ksi_code = write ? SEGV_ACCERR : SEGV_MAPERR;
4410 	ksi.ksi_addr = td->td_sigblock_ptr;
4411 	trapsignal(td, &ksi);
4412 }
4413 
4414 static bool
sigfastblock_fetch_sig(struct thread * td,bool sendsig,uint32_t * valp)4415 sigfastblock_fetch_sig(struct thread *td, bool sendsig, uint32_t *valp)
4416 {
4417 	uint32_t res;
4418 
4419 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4420 		return (true);
4421 	if (fueword32((void *)td->td_sigblock_ptr, &res) == -1) {
4422 		sigfastblock_failed(td, sendsig, false);
4423 		return (false);
4424 	}
4425 	*valp = res;
4426 	td->td_sigblock_val = res & ~SIGFASTBLOCK_FLAGS;
4427 	return (true);
4428 }
4429 
4430 static void
sigfastblock_resched(struct thread * td,bool resched)4431 sigfastblock_resched(struct thread *td, bool resched)
4432 {
4433 	struct proc *p;
4434 
4435 	if (resched) {
4436 		p = td->td_proc;
4437 		PROC_LOCK(p);
4438 		reschedule_signals(p, td->td_sigmask, 0);
4439 		PROC_UNLOCK(p);
4440 	}
4441 	ast_sched(td, TDA_SIG);
4442 }
4443 
4444 int
sys_sigfastblock(struct thread * td,struct sigfastblock_args * uap)4445 sys_sigfastblock(struct thread *td, struct sigfastblock_args *uap)
4446 {
4447 	struct proc *p;
4448 	int error, res;
4449 	uint32_t oldval;
4450 
4451 	error = 0;
4452 	p = td->td_proc;
4453 	switch (uap->cmd) {
4454 	case SIGFASTBLOCK_SETPTR:
4455 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
4456 			error = EBUSY;
4457 			break;
4458 		}
4459 		if (((uintptr_t)(uap->ptr) & (sizeof(uint32_t) - 1)) != 0) {
4460 			error = EINVAL;
4461 			break;
4462 		}
4463 		td->td_pflags |= TDP_SIGFASTBLOCK;
4464 		td->td_sigblock_ptr = uap->ptr;
4465 		break;
4466 
4467 	case SIGFASTBLOCK_UNBLOCK:
4468 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4469 			error = EINVAL;
4470 			break;
4471 		}
4472 
4473 		for (;;) {
4474 			res = casueword32(td->td_sigblock_ptr,
4475 			    SIGFASTBLOCK_PEND, &oldval, 0);
4476 			if (res == -1) {
4477 				error = EFAULT;
4478 				sigfastblock_failed(td, false, true);
4479 				break;
4480 			}
4481 			if (res == 0)
4482 				break;
4483 			MPASS(res == 1);
4484 			if (oldval != SIGFASTBLOCK_PEND) {
4485 				error = EBUSY;
4486 				break;
4487 			}
4488 			error = thread_check_susp(td, false);
4489 			if (error != 0)
4490 				break;
4491 		}
4492 		if (error != 0)
4493 			break;
4494 
4495 		/*
4496 		 * td_sigblock_val is cleared there, but not on a
4497 		 * syscall exit.  The end effect is that a single
4498 		 * interruptible sleep, while user sigblock word is
4499 		 * set, might return EINTR or ERESTART to usermode
4500 		 * without delivering signal.  All further sleeps,
4501 		 * until userspace clears the word and does
4502 		 * sigfastblock(UNBLOCK), observe current word and no
4503 		 * longer get interrupted.  It is slight
4504 		 * non-conformance, with alternative to have read the
4505 		 * sigblock word on each syscall entry.
4506 		 */
4507 		td->td_sigblock_val = 0;
4508 
4509 		/*
4510 		 * Rely on normal ast mechanism to deliver pending
4511 		 * signals to current thread.  But notify others about
4512 		 * fake unblock.
4513 		 */
4514 		sigfastblock_resched(td, error == 0 && p->p_numthreads != 1);
4515 
4516 		break;
4517 
4518 	case SIGFASTBLOCK_UNSETPTR:
4519 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4520 			error = EINVAL;
4521 			break;
4522 		}
4523 		if (!sigfastblock_fetch_sig(td, false, &oldval)) {
4524 			error = EFAULT;
4525 			break;
4526 		}
4527 		if (oldval != 0 && oldval != SIGFASTBLOCK_PEND) {
4528 			error = EBUSY;
4529 			break;
4530 		}
4531 		sigfastblock_clear(td);
4532 		break;
4533 
4534 	default:
4535 		error = EINVAL;
4536 		break;
4537 	}
4538 	return (error);
4539 }
4540 
4541 void
sigfastblock_clear(struct thread * td)4542 sigfastblock_clear(struct thread *td)
4543 {
4544 	bool resched;
4545 
4546 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4547 		return;
4548 	td->td_sigblock_val = 0;
4549 	resched = (td->td_pflags & TDP_SIGFASTPENDING) != 0 ||
4550 	    SIGPENDING(td);
4551 	td->td_pflags &= ~(TDP_SIGFASTBLOCK | TDP_SIGFASTPENDING);
4552 	sigfastblock_resched(td, resched);
4553 }
4554 
4555 void
sigfastblock_fetch(struct thread * td)4556 sigfastblock_fetch(struct thread *td)
4557 {
4558 	uint32_t val;
4559 
4560 	(void)sigfastblock_fetch_sig(td, true, &val);
4561 }
4562 
4563 static void
sigfastblock_setpend1(struct thread * td)4564 sigfastblock_setpend1(struct thread *td)
4565 {
4566 	int res;
4567 	uint32_t oldval;
4568 
4569 	if ((td->td_pflags & TDP_SIGFASTPENDING) == 0)
4570 		return;
4571 	res = fueword32((void *)td->td_sigblock_ptr, &oldval);
4572 	if (res == -1) {
4573 		sigfastblock_failed(td, true, false);
4574 		return;
4575 	}
4576 	for (;;) {
4577 		res = casueword32(td->td_sigblock_ptr, oldval, &oldval,
4578 		    oldval | SIGFASTBLOCK_PEND);
4579 		if (res == -1) {
4580 			sigfastblock_failed(td, true, true);
4581 			return;
4582 		}
4583 		if (res == 0) {
4584 			td->td_sigblock_val = oldval & ~SIGFASTBLOCK_FLAGS;
4585 			td->td_pflags &= ~TDP_SIGFASTPENDING;
4586 			break;
4587 		}
4588 		MPASS(res == 1);
4589 		if (thread_check_susp(td, false) != 0)
4590 			break;
4591 	}
4592 }
4593 
4594 static void
sigfastblock_setpend(struct thread * td,bool resched)4595 sigfastblock_setpend(struct thread *td, bool resched)
4596 {
4597 	struct proc *p;
4598 
4599 	sigfastblock_setpend1(td);
4600 	if (resched) {
4601 		p = td->td_proc;
4602 		PROC_LOCK(p);
4603 		reschedule_signals(p, fastblock_mask, SIGPROCMASK_FASTBLK);
4604 		PROC_UNLOCK(p);
4605 	}
4606 }
4607