xref: /freebsd/sys/kern/kern_sig.c (revision 26061e4e542d220c577fb3437a9a9f108dc27698)
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 bool	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 static void	sig_handle_first_stop(struct thread *td, struct proc *p,
119     int sig);
120 
121 static uma_zone_t	ksiginfo_zone = NULL;
122 const struct filterops sig_filtops = {
123 	.f_isfd = 0,
124 	.f_attach = filt_sigattach,
125 	.f_detach = filt_sigdetach,
126 	.f_event = filt_signal,
127 };
128 
129 static int	kern_logsigexit = 1;
130 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
131     &kern_logsigexit, 0,
132     "Log processes quitting on abnormal signals to syslog(3)");
133 
134 static int	kern_forcesigexit = 1;
135 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
136     &kern_forcesigexit, 0, "Force trap signal to be handled");
137 
138 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
139     "POSIX real time signal");
140 
141 static int	max_pending_per_proc = 128;
142 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
143     &max_pending_per_proc, 0, "Max pending signals per proc");
144 
145 static int	preallocate_siginfo = 1024;
146 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
147     &preallocate_siginfo, 0, "Preallocated signal memory size");
148 
149 static int	signal_overflow = 0;
150 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
151     &signal_overflow, 0, "Number of signals overflew");
152 
153 static int	signal_alloc_fail = 0;
154 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
155     &signal_alloc_fail, 0, "signals failed to be allocated");
156 
157 static int	kern_lognosys = 0;
158 SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
159     "Log invalid syscalls");
160 
161 static int	kern_signosys = 1;
162 SYSCTL_INT(_kern, OID_AUTO, signosys, CTLFLAG_RWTUN, &kern_signosys, 0,
163     "Send SIGSYS on return from invalid syscall");
164 
165 __read_frequently bool sigfastblock_fetch_always = false;
166 SYSCTL_BOOL(_kern, OID_AUTO, sigfastblock_fetch_always, CTLFLAG_RWTUN,
167     &sigfastblock_fetch_always, 0,
168     "Fetch sigfastblock word on each syscall entry for proper "
169     "blocking semantic");
170 
171 static bool	kern_sig_discard_ign = true;
172 SYSCTL_BOOL(_kern, OID_AUTO, sig_discard_ign, CTLFLAG_RWTUN,
173     &kern_sig_discard_ign, 0,
174     "Discard ignored signals on delivery, otherwise queue them to "
175     "the target queue");
176 
177 bool pt_attach_transparent = true;
178 SYSCTL_BOOL(_debug, OID_AUTO, ptrace_attach_transparent, CTLFLAG_RWTUN,
179     &pt_attach_transparent, 0,
180     "Hide wakes from PT_ATTACH on interruptible sleeps");
181 
182 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
183 
184 /*
185  * Policy -- Can ucred cr1 send SIGIO to process cr2?
186  * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
187  * in the right situations.
188  */
189 #define CANSIGIO(cr1, cr2) \
190 	((cr1)->cr_uid == 0 || \
191 	    (cr1)->cr_ruid == (cr2)->cr_ruid || \
192 	    (cr1)->cr_uid == (cr2)->cr_ruid || \
193 	    (cr1)->cr_ruid == (cr2)->cr_uid || \
194 	    (cr1)->cr_uid == (cr2)->cr_uid)
195 
196 static int	sugid_coredump;
197 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
198     &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
199 
200 static int	capmode_coredump;
201 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
202     &capmode_coredump, 0, "Allow processes in capability mode to dump core");
203 
204 static int	do_coredump = 1;
205 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
206 	&do_coredump, 0, "Enable/Disable coredumps");
207 
208 static int	set_core_nodump_flag = 0;
209 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
210 	0, "Enable setting the NODUMP flag on coredump files");
211 
212 static int	coredump_devctl = 0;
213 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
214 	0, "Generate a devctl notification when processes coredump");
215 
216 /*
217  * Signal properties and actions.
218  * The array below categorizes the signals and their default actions
219  * according to the following properties:
220  */
221 #define	SIGPROP_KILL		0x01	/* terminates process by default */
222 #define	SIGPROP_CORE		0x02	/* ditto and coredumps */
223 #define	SIGPROP_STOP		0x04	/* suspend process */
224 #define	SIGPROP_TTYSTOP		0x08	/* ditto, from tty */
225 #define	SIGPROP_IGNORE		0x10	/* ignore by default */
226 #define	SIGPROP_CONT		0x20	/* continue if suspended */
227 
228 static const int sigproptbl[NSIG] = {
229 	[SIGHUP] =	SIGPROP_KILL,
230 	[SIGINT] =	SIGPROP_KILL,
231 	[SIGQUIT] =	SIGPROP_KILL | SIGPROP_CORE,
232 	[SIGILL] =	SIGPROP_KILL | SIGPROP_CORE,
233 	[SIGTRAP] =	SIGPROP_KILL | SIGPROP_CORE,
234 	[SIGABRT] =	SIGPROP_KILL | SIGPROP_CORE,
235 	[SIGEMT] =	SIGPROP_KILL | SIGPROP_CORE,
236 	[SIGFPE] =	SIGPROP_KILL | SIGPROP_CORE,
237 	[SIGKILL] =	SIGPROP_KILL,
238 	[SIGBUS] =	SIGPROP_KILL | SIGPROP_CORE,
239 	[SIGSEGV] =	SIGPROP_KILL | SIGPROP_CORE,
240 	[SIGSYS] =	SIGPROP_KILL | SIGPROP_CORE,
241 	[SIGPIPE] =	SIGPROP_KILL,
242 	[SIGALRM] =	SIGPROP_KILL,
243 	[SIGTERM] =	SIGPROP_KILL,
244 	[SIGURG] =	SIGPROP_IGNORE,
245 	[SIGSTOP] =	SIGPROP_STOP,
246 	[SIGTSTP] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
247 	[SIGCONT] =	SIGPROP_IGNORE | SIGPROP_CONT,
248 	[SIGCHLD] =	SIGPROP_IGNORE,
249 	[SIGTTIN] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
250 	[SIGTTOU] =	SIGPROP_STOP | SIGPROP_TTYSTOP,
251 	[SIGIO] =	SIGPROP_IGNORE,
252 	[SIGXCPU] =	SIGPROP_KILL,
253 	[SIGXFSZ] =	SIGPROP_KILL,
254 	[SIGVTALRM] =	SIGPROP_KILL,
255 	[SIGPROF] =	SIGPROP_KILL,
256 	[SIGWINCH] =	SIGPROP_IGNORE,
257 	[SIGINFO] =	SIGPROP_IGNORE,
258 	[SIGUSR1] =	SIGPROP_KILL,
259 	[SIGUSR2] =	SIGPROP_KILL,
260 };
261 
262 #define	_SIG_FOREACH_ADVANCE(i, set) ({					\
263 	int __found;							\
264 	for (;;) {							\
265 		if (__bits != 0) {					\
266 			int __sig = ffs(__bits);			\
267 			__bits &= ~(1u << (__sig - 1));			\
268 			sig = __i * sizeof((set)->__bits[0]) * NBBY + __sig; \
269 			__found = 1;					\
270 			break;						\
271 		}							\
272 		if (++__i == _SIG_WORDS) {				\
273 			__found = 0;					\
274 			break;						\
275 		}							\
276 		__bits = (set)->__bits[__i];				\
277 	}								\
278 	__found != 0;							\
279 })
280 
281 #define	SIG_FOREACH(i, set)						\
282 	for (int32_t __i = -1, __bits = 0;				\
283 	    _SIG_FOREACH_ADVANCE(i, set); )				\
284 
285 static sigset_t fastblock_mask;
286 
287 static void
ast_sig(struct thread * td,int tda)288 ast_sig(struct thread *td, int tda)
289 {
290 	struct proc *p;
291 	int old_boundary, sig;
292 	bool resched_sigs;
293 
294 	p = td->td_proc;
295 
296 #ifdef DIAGNOSTIC
297 	if (p->p_numthreads == 1 && (tda & (TDAI(TDA_SIG) |
298 	    TDAI(TDA_AST))) == 0) {
299 		PROC_LOCK(p);
300 		thread_lock(td);
301 		/*
302 		 * Note that TDA_SIG should be re-read from
303 		 * td_ast, since signal might have been delivered
304 		 * after we cleared td_flags above.  This is one of
305 		 * the reason for looping check for AST condition.
306 		 * See comment in userret() about P_PPWAIT.
307 		 */
308 		if ((p->p_flag & P_PPWAIT) == 0 &&
309 		    (td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
310 			if (SIGPENDING(td) && ((tda | td->td_ast) &
311 			    (TDAI(TDA_SIG) | TDAI(TDA_AST))) == 0) {
312 				thread_unlock(td); /* fix dumps */
313 				panic(
314 				    "failed2 to set signal flags for ast p %p "
315 				    "td %p tda %#x td_ast %#x fl %#x",
316 				    p, td, tda, td->td_ast, td->td_flags);
317 			}
318 		}
319 		thread_unlock(td);
320 		PROC_UNLOCK(p);
321 	}
322 #endif
323 
324 	/*
325 	 * Check for signals. Unlocked reads of p_pendingcnt or
326 	 * p_siglist might cause process-directed signal to be handled
327 	 * later.
328 	 */
329 	if ((tda & TDAI(TDA_SIG)) != 0 || p->p_pendingcnt > 0 ||
330 	    !SIGISEMPTY(p->p_siglist)) {
331 		sigfastblock_fetch(td);
332 		PROC_LOCK(p);
333 		old_boundary = ~TDB_BOUNDARY | (td->td_dbgflags & TDB_BOUNDARY);
334 		td->td_dbgflags |= TDB_BOUNDARY;
335 		mtx_lock(&p->p_sigacts->ps_mtx);
336 		while ((sig = cursig(td)) != 0) {
337 			KASSERT(sig >= 0, ("sig %d", sig));
338 			postsig(sig);
339 		}
340 		mtx_unlock(&p->p_sigacts->ps_mtx);
341 		td->td_dbgflags &= old_boundary;
342 		PROC_UNLOCK(p);
343 		resched_sigs = true;
344 	} else {
345 		resched_sigs = false;
346 	}
347 
348 	/*
349 	 * Handle deferred update of the fast sigblock value, after
350 	 * the postsig() loop was performed.
351 	 */
352 	sigfastblock_setpend(td, resched_sigs);
353 
354 	/*
355 	 * Clear td_sa.code: signal to ptrace that syscall arguments
356 	 * are unavailable after this point. This AST handler is the
357 	 * last chance for ptracestop() to signal the tracer before
358 	 * the tracee returns to userspace.
359 	 */
360 	td->td_sa.code = 0;
361 }
362 
363 static void
ast_sigsuspend(struct thread * td,int tda __unused)364 ast_sigsuspend(struct thread *td, int tda __unused)
365 {
366 	MPASS((td->td_pflags & TDP_OLDMASK) != 0);
367 	td->td_pflags &= ~TDP_OLDMASK;
368 	kern_sigprocmask(td, SIG_SETMASK, &td->td_oldsigmask, NULL, 0);
369 }
370 
371 static void
sigqueue_start(void)372 sigqueue_start(void)
373 {
374 	ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
375 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
376 	uma_prealloc(ksiginfo_zone, preallocate_siginfo);
377 	p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
378 	p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
379 	p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
380 	SIGFILLSET(fastblock_mask);
381 	SIG_CANTMASK(fastblock_mask);
382 	ast_register(TDA_SIG, ASTR_UNCOND, 0, ast_sig);
383 
384 	/*
385 	 * TDA_PSELECT is for the case where the signal mask should be restored
386 	 * before delivering any signals so that we do not deliver any that are
387 	 * blocked by the normal thread mask.  It is mutually exclusive with
388 	 * TDA_SIGSUSPEND, which should be used if we *do* want to deliver
389 	 * signals that are normally blocked, e.g., if it interrupted our sleep.
390 	 */
391 	ast_register(TDA_PSELECT, ASTR_ASTF_REQUIRED | ASTR_TDP,
392 	    TDP_OLDMASK, ast_sigsuspend);
393 	ast_register(TDA_SIGSUSPEND, ASTR_ASTF_REQUIRED | ASTR_TDP,
394 	    TDP_OLDMASK, ast_sigsuspend);
395 }
396 
397 ksiginfo_t *
ksiginfo_alloc(int mwait)398 ksiginfo_alloc(int mwait)
399 {
400 	MPASS(mwait == M_WAITOK || mwait == M_NOWAIT);
401 
402 	if (ksiginfo_zone == NULL)
403 		return (NULL);
404 	return (uma_zalloc(ksiginfo_zone, mwait | M_ZERO));
405 }
406 
407 void
ksiginfo_free(ksiginfo_t * ksi)408 ksiginfo_free(ksiginfo_t *ksi)
409 {
410 	uma_zfree(ksiginfo_zone, ksi);
411 }
412 
413 static __inline bool
ksiginfo_tryfree(ksiginfo_t * ksi)414 ksiginfo_tryfree(ksiginfo_t *ksi)
415 {
416 	if ((ksi->ksi_flags & KSI_EXT) == 0) {
417 		uma_zfree(ksiginfo_zone, ksi);
418 		return (true);
419 	}
420 	return (false);
421 }
422 
423 void
sigqueue_init(sigqueue_t * list,struct proc * p)424 sigqueue_init(sigqueue_t *list, struct proc *p)
425 {
426 	SIGEMPTYSET(list->sq_signals);
427 	SIGEMPTYSET(list->sq_kill);
428 	SIGEMPTYSET(list->sq_ptrace);
429 	TAILQ_INIT(&list->sq_list);
430 	list->sq_proc = p;
431 	list->sq_flags = SQ_INIT;
432 }
433 
434 /*
435  * Get a signal's ksiginfo.
436  * Return:
437  *	0	-	signal not found
438  *	others	-	signal number
439  */
440 static int
sigqueue_get(sigqueue_t * sq,int signo,ksiginfo_t * si)441 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
442 {
443 	struct proc *p = sq->sq_proc;
444 	struct ksiginfo *ksi, *next;
445 	int count = 0;
446 
447 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
448 
449 	if (!SIGISMEMBER(sq->sq_signals, signo))
450 		return (0);
451 
452 	if (SIGISMEMBER(sq->sq_ptrace, signo)) {
453 		count++;
454 		SIGDELSET(sq->sq_ptrace, signo);
455 		si->ksi_flags |= KSI_PTRACE;
456 	}
457 	if (SIGISMEMBER(sq->sq_kill, signo)) {
458 		count++;
459 		if (count == 1)
460 			SIGDELSET(sq->sq_kill, signo);
461 	}
462 
463 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
464 		if (ksi->ksi_signo == signo) {
465 			if (count == 0) {
466 				TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
467 				ksi->ksi_sigq = NULL;
468 				ksiginfo_copy(ksi, si);
469 				if (ksiginfo_tryfree(ksi) && p != NULL)
470 					p->p_pendingcnt--;
471 			}
472 			if (++count > 1)
473 				break;
474 		}
475 	}
476 
477 	if (count <= 1)
478 		SIGDELSET(sq->sq_signals, signo);
479 	si->ksi_signo = signo;
480 	return (signo);
481 }
482 
483 void
sigqueue_take(ksiginfo_t * ksi)484 sigqueue_take(ksiginfo_t *ksi)
485 {
486 	struct ksiginfo *kp;
487 	struct proc	*p;
488 	sigqueue_t	*sq;
489 
490 	if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
491 		return;
492 
493 	p = sq->sq_proc;
494 	TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
495 	ksi->ksi_sigq = NULL;
496 	if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
497 		p->p_pendingcnt--;
498 
499 	for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
500 	     kp = TAILQ_NEXT(kp, ksi_link)) {
501 		if (kp->ksi_signo == ksi->ksi_signo)
502 			break;
503 	}
504 	if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
505 	    !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
506 		SIGDELSET(sq->sq_signals, ksi->ksi_signo);
507 }
508 
509 static int
sigqueue_add(sigqueue_t * sq,int signo,ksiginfo_t * si)510 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
511 {
512 	struct proc *p = sq->sq_proc;
513 	struct ksiginfo *ksi;
514 	int ret = 0;
515 
516 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
517 
518 	/*
519 	 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
520 	 * for these signals.
521 	 */
522 	if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
523 		SIGADDSET(sq->sq_kill, signo);
524 		goto out_set_bit;
525 	}
526 
527 	/* directly insert the ksi, don't copy it */
528 	if (si->ksi_flags & KSI_INS) {
529 		if (si->ksi_flags & KSI_HEAD)
530 			TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
531 		else
532 			TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
533 		si->ksi_sigq = sq;
534 		goto out_set_bit;
535 	}
536 
537 	if (__predict_false(ksiginfo_zone == NULL)) {
538 		SIGADDSET(sq->sq_kill, signo);
539 		goto out_set_bit;
540 	}
541 
542 	if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
543 		signal_overflow++;
544 		ret = EAGAIN;
545 	} else if ((ksi = ksiginfo_alloc(M_NOWAIT)) == NULL) {
546 		signal_alloc_fail++;
547 		ret = EAGAIN;
548 	} else {
549 		if (p != NULL)
550 			p->p_pendingcnt++;
551 		ksiginfo_copy(si, ksi);
552 		ksi->ksi_signo = signo;
553 		if (si->ksi_flags & KSI_HEAD)
554 			TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
555 		else
556 			TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
557 		ksi->ksi_sigq = sq;
558 	}
559 
560 	if (ret != 0) {
561 		if ((si->ksi_flags & KSI_PTRACE) != 0) {
562 			SIGADDSET(sq->sq_ptrace, signo);
563 			ret = 0;
564 			goto out_set_bit;
565 		} else if ((si->ksi_flags & KSI_TRAP) != 0 ||
566 		    (si->ksi_flags & KSI_SIGQ) == 0) {
567 			SIGADDSET(sq->sq_kill, signo);
568 			ret = 0;
569 			goto out_set_bit;
570 		}
571 		return (ret);
572 	}
573 
574 out_set_bit:
575 	SIGADDSET(sq->sq_signals, signo);
576 	return (ret);
577 }
578 
579 void
sigqueue_flush(sigqueue_t * sq)580 sigqueue_flush(sigqueue_t *sq)
581 {
582 	struct proc *p = sq->sq_proc;
583 	ksiginfo_t *ksi;
584 
585 	KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
586 
587 	if (p != NULL)
588 		PROC_LOCK_ASSERT(p, MA_OWNED);
589 
590 	while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
591 		TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
592 		ksi->ksi_sigq = NULL;
593 		if (ksiginfo_tryfree(ksi) && p != NULL)
594 			p->p_pendingcnt--;
595 	}
596 
597 	SIGEMPTYSET(sq->sq_signals);
598 	SIGEMPTYSET(sq->sq_kill);
599 	SIGEMPTYSET(sq->sq_ptrace);
600 }
601 
602 static void
sigqueue_move_set(sigqueue_t * src,sigqueue_t * dst,const sigset_t * set)603 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
604 {
605 	sigset_t tmp;
606 	struct proc *p1, *p2;
607 	ksiginfo_t *ksi, *next;
608 
609 	KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
610 	KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
611 	p1 = src->sq_proc;
612 	p2 = dst->sq_proc;
613 	/* Move siginfo to target list */
614 	TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
615 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
616 			TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
617 			if (p1 != NULL)
618 				p1->p_pendingcnt--;
619 			TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
620 			ksi->ksi_sigq = dst;
621 			if (p2 != NULL)
622 				p2->p_pendingcnt++;
623 		}
624 	}
625 
626 	/* Move pending bits to target list */
627 	tmp = src->sq_kill;
628 	SIGSETAND(tmp, *set);
629 	SIGSETOR(dst->sq_kill, tmp);
630 	SIGSETNAND(src->sq_kill, tmp);
631 
632 	tmp = src->sq_ptrace;
633 	SIGSETAND(tmp, *set);
634 	SIGSETOR(dst->sq_ptrace, tmp);
635 	SIGSETNAND(src->sq_ptrace, tmp);
636 
637 	tmp = src->sq_signals;
638 	SIGSETAND(tmp, *set);
639 	SIGSETOR(dst->sq_signals, tmp);
640 	SIGSETNAND(src->sq_signals, tmp);
641 }
642 
643 #if 0
644 static void
645 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
646 {
647 	sigset_t set;
648 
649 	SIGEMPTYSET(set);
650 	SIGADDSET(set, signo);
651 	sigqueue_move_set(src, dst, &set);
652 }
653 #endif
654 
655 static void
sigqueue_delete_set(sigqueue_t * sq,const sigset_t * set)656 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
657 {
658 	struct proc *p = sq->sq_proc;
659 	ksiginfo_t *ksi, *next;
660 
661 	KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
662 
663 	/* Remove siginfo queue */
664 	TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
665 		if (SIGISMEMBER(*set, ksi->ksi_signo)) {
666 			TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
667 			ksi->ksi_sigq = NULL;
668 			if (ksiginfo_tryfree(ksi) && p != NULL)
669 				p->p_pendingcnt--;
670 		}
671 	}
672 	SIGSETNAND(sq->sq_kill, *set);
673 	SIGSETNAND(sq->sq_ptrace, *set);
674 	SIGSETNAND(sq->sq_signals, *set);
675 }
676 
677 void
sigqueue_delete(sigqueue_t * sq,int signo)678 sigqueue_delete(sigqueue_t *sq, int signo)
679 {
680 	sigset_t set;
681 
682 	SIGEMPTYSET(set);
683 	SIGADDSET(set, signo);
684 	sigqueue_delete_set(sq, &set);
685 }
686 
687 /* Remove a set of signals for a process */
688 static void
sigqueue_delete_set_proc(struct proc * p,const sigset_t * set)689 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
690 {
691 	sigqueue_t worklist;
692 	struct thread *td0;
693 
694 	PROC_LOCK_ASSERT(p, MA_OWNED);
695 
696 	sigqueue_init(&worklist, NULL);
697 	sigqueue_move_set(&p->p_sigqueue, &worklist, set);
698 
699 	FOREACH_THREAD_IN_PROC(p, td0)
700 		sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
701 
702 	sigqueue_flush(&worklist);
703 }
704 
705 void
sigqueue_delete_proc(struct proc * p,int signo)706 sigqueue_delete_proc(struct proc *p, int signo)
707 {
708 	sigset_t set;
709 
710 	SIGEMPTYSET(set);
711 	SIGADDSET(set, signo);
712 	sigqueue_delete_set_proc(p, &set);
713 }
714 
715 static void
sigqueue_delete_stopmask_proc(struct proc * p)716 sigqueue_delete_stopmask_proc(struct proc *p)
717 {
718 	sigset_t set;
719 
720 	SIGEMPTYSET(set);
721 	SIGADDSET(set, SIGSTOP);
722 	SIGADDSET(set, SIGTSTP);
723 	SIGADDSET(set, SIGTTIN);
724 	SIGADDSET(set, SIGTTOU);
725 	sigqueue_delete_set_proc(p, &set);
726 }
727 
728 /*
729  * Determine signal that should be delivered to thread td, the current
730  * thread, 0 if none.  If there is a pending stop signal with default
731  * action, the process stops in issignal().
732  */
733 int
cursig(struct thread * td)734 cursig(struct thread *td)
735 {
736 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
737 	mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
738 	THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
739 	return (SIGPENDING(td) ? issignal(td) : 0);
740 }
741 
742 /*
743  * Arrange for ast() to handle unmasked pending signals on return to user
744  * mode.  This must be called whenever a signal is added to td_sigqueue or
745  * unmasked in td_sigmask.
746  */
747 void
signotify(struct thread * td)748 signotify(struct thread *td)
749 {
750 
751 	PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
752 
753 	if (SIGPENDING(td))
754 		ast_sched(td, TDA_SIG);
755 }
756 
757 /*
758  * Returns 1 (true) if altstack is configured for the thread, and the
759  * passed stack bottom address falls into the altstack range.  Handles
760  * the 43 compat special case where the alt stack size is zero.
761  */
762 int
sigonstack(size_t sp)763 sigonstack(size_t sp)
764 {
765 	struct thread *td;
766 
767 	td = curthread;
768 	if ((td->td_pflags & TDP_ALTSTACK) == 0)
769 		return (0);
770 #if defined(COMPAT_43)
771 	if (SV_PROC_FLAG(td->td_proc, SV_AOUT) && td->td_sigstk.ss_size == 0)
772 		return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
773 #endif
774 	return (sp >= (size_t)td->td_sigstk.ss_sp &&
775 	    sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
776 }
777 
778 static __inline int
sigprop(int sig)779 sigprop(int sig)
780 {
781 
782 	if (sig > 0 && sig < nitems(sigproptbl))
783 		return (sigproptbl[sig]);
784 	return (0);
785 }
786 
787 static bool
sigact_flag_test(const struct sigaction * act,int flag)788 sigact_flag_test(const struct sigaction *act, int flag)
789 {
790 
791 	/*
792 	 * SA_SIGINFO is reset when signal disposition is set to
793 	 * ignore or default.  Other flags are kept according to user
794 	 * settings.
795 	 */
796 	return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
797 	    ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
798 	    (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
799 }
800 
801 /*
802  * kern_sigaction
803  * sigaction
804  * freebsd4_sigaction
805  * osigaction
806  */
807 int
kern_sigaction(struct thread * td,int sig,const struct sigaction * act,struct sigaction * oact,int flags)808 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
809     struct sigaction *oact, int flags)
810 {
811 	struct sigacts *ps;
812 	struct proc *p = td->td_proc;
813 
814 	if (!_SIG_VALID(sig))
815 		return (EINVAL);
816 	if (act != NULL && act->sa_handler != SIG_DFL &&
817 	    act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
818 	    SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
819 	    SA_NOCLDWAIT | SA_SIGINFO)) != 0)
820 		return (EINVAL);
821 
822 	PROC_LOCK(p);
823 	ps = p->p_sigacts;
824 	mtx_lock(&ps->ps_mtx);
825 	if (oact) {
826 		memset(oact, 0, sizeof(*oact));
827 		oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
828 		if (SIGISMEMBER(ps->ps_sigonstack, sig))
829 			oact->sa_flags |= SA_ONSTACK;
830 		if (!SIGISMEMBER(ps->ps_sigintr, sig))
831 			oact->sa_flags |= SA_RESTART;
832 		if (SIGISMEMBER(ps->ps_sigreset, sig))
833 			oact->sa_flags |= SA_RESETHAND;
834 		if (SIGISMEMBER(ps->ps_signodefer, sig))
835 			oact->sa_flags |= SA_NODEFER;
836 		if (SIGISMEMBER(ps->ps_siginfo, sig)) {
837 			oact->sa_flags |= SA_SIGINFO;
838 			oact->sa_sigaction =
839 			    (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
840 		} else
841 			oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
842 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
843 			oact->sa_flags |= SA_NOCLDSTOP;
844 		if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
845 			oact->sa_flags |= SA_NOCLDWAIT;
846 	}
847 	if (act) {
848 		if ((sig == SIGKILL || sig == SIGSTOP) &&
849 		    act->sa_handler != SIG_DFL) {
850 			mtx_unlock(&ps->ps_mtx);
851 			PROC_UNLOCK(p);
852 			return (EINVAL);
853 		}
854 
855 		/*
856 		 * Change setting atomically.
857 		 */
858 
859 		ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
860 		SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
861 		if (sigact_flag_test(act, SA_SIGINFO)) {
862 			ps->ps_sigact[_SIG_IDX(sig)] =
863 			    (__sighandler_t *)act->sa_sigaction;
864 			SIGADDSET(ps->ps_siginfo, sig);
865 		} else {
866 			ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
867 			SIGDELSET(ps->ps_siginfo, sig);
868 		}
869 		if (!sigact_flag_test(act, SA_RESTART))
870 			SIGADDSET(ps->ps_sigintr, sig);
871 		else
872 			SIGDELSET(ps->ps_sigintr, sig);
873 		if (sigact_flag_test(act, SA_ONSTACK))
874 			SIGADDSET(ps->ps_sigonstack, sig);
875 		else
876 			SIGDELSET(ps->ps_sigonstack, sig);
877 		if (sigact_flag_test(act, SA_RESETHAND))
878 			SIGADDSET(ps->ps_sigreset, sig);
879 		else
880 			SIGDELSET(ps->ps_sigreset, sig);
881 		if (sigact_flag_test(act, SA_NODEFER))
882 			SIGADDSET(ps->ps_signodefer, sig);
883 		else
884 			SIGDELSET(ps->ps_signodefer, sig);
885 		if (sig == SIGCHLD) {
886 			if (act->sa_flags & SA_NOCLDSTOP)
887 				ps->ps_flag |= PS_NOCLDSTOP;
888 			else
889 				ps->ps_flag &= ~PS_NOCLDSTOP;
890 			if (act->sa_flags & SA_NOCLDWAIT) {
891 				/*
892 				 * Paranoia: since SA_NOCLDWAIT is implemented
893 				 * by reparenting the dying child to PID 1 (and
894 				 * trust it to reap the zombie), PID 1 itself
895 				 * is forbidden to set SA_NOCLDWAIT.
896 				 */
897 				if (p->p_pid == 1)
898 					ps->ps_flag &= ~PS_NOCLDWAIT;
899 				else
900 					ps->ps_flag |= PS_NOCLDWAIT;
901 			} else
902 				ps->ps_flag &= ~PS_NOCLDWAIT;
903 			if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
904 				ps->ps_flag |= PS_CLDSIGIGN;
905 			else
906 				ps->ps_flag &= ~PS_CLDSIGIGN;
907 		}
908 		/*
909 		 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
910 		 * and for signals set to SIG_DFL where the default is to
911 		 * ignore. However, don't put SIGCONT in ps_sigignore, as we
912 		 * have to restart the process.
913 		 */
914 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
915 		    (sigprop(sig) & SIGPROP_IGNORE &&
916 		     ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
917 			/* never to be seen again */
918 			sigqueue_delete_proc(p, sig);
919 			if (sig != SIGCONT)
920 				/* easier in psignal */
921 				SIGADDSET(ps->ps_sigignore, sig);
922 			SIGDELSET(ps->ps_sigcatch, sig);
923 		} else {
924 			SIGDELSET(ps->ps_sigignore, sig);
925 			if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
926 				SIGDELSET(ps->ps_sigcatch, sig);
927 			else
928 				SIGADDSET(ps->ps_sigcatch, sig);
929 		}
930 #ifdef COMPAT_FREEBSD4
931 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
932 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
933 		    (flags & KSA_FREEBSD4) == 0)
934 			SIGDELSET(ps->ps_freebsd4, sig);
935 		else
936 			SIGADDSET(ps->ps_freebsd4, sig);
937 #endif
938 #ifdef COMPAT_43
939 		if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
940 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
941 		    (flags & KSA_OSIGSET) == 0)
942 			SIGDELSET(ps->ps_osigset, sig);
943 		else
944 			SIGADDSET(ps->ps_osigset, sig);
945 #endif
946 	}
947 	mtx_unlock(&ps->ps_mtx);
948 	PROC_UNLOCK(p);
949 	return (0);
950 }
951 
952 #ifndef _SYS_SYSPROTO_H_
953 struct sigaction_args {
954 	int	sig;
955 	struct	sigaction *act;
956 	struct	sigaction *oact;
957 };
958 #endif
959 int
sys_sigaction(struct thread * td,struct sigaction_args * uap)960 sys_sigaction(struct thread *td, struct sigaction_args *uap)
961 {
962 	struct sigaction act, oact;
963 	struct sigaction *actp, *oactp;
964 	int error;
965 
966 	actp = (uap->act != NULL) ? &act : NULL;
967 	oactp = (uap->oact != NULL) ? &oact : NULL;
968 	if (actp) {
969 		error = copyin(uap->act, actp, sizeof(act));
970 		if (error)
971 			return (error);
972 	}
973 	error = kern_sigaction(td, uap->sig, actp, oactp, 0);
974 	if (oactp && !error)
975 		error = copyout(oactp, uap->oact, sizeof(oact));
976 	return (error);
977 }
978 
979 #ifdef COMPAT_FREEBSD4
980 #ifndef _SYS_SYSPROTO_H_
981 struct freebsd4_sigaction_args {
982 	int	sig;
983 	struct	sigaction *act;
984 	struct	sigaction *oact;
985 };
986 #endif
987 int
freebsd4_sigaction(struct thread * td,struct freebsd4_sigaction_args * uap)988 freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
989 {
990 	struct sigaction act, oact;
991 	struct sigaction *actp, *oactp;
992 	int error;
993 
994 	actp = (uap->act != NULL) ? &act : NULL;
995 	oactp = (uap->oact != NULL) ? &oact : NULL;
996 	if (actp) {
997 		error = copyin(uap->act, actp, sizeof(act));
998 		if (error)
999 			return (error);
1000 	}
1001 	error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
1002 	if (oactp && !error)
1003 		error = copyout(oactp, uap->oact, sizeof(oact));
1004 	return (error);
1005 }
1006 #endif	/* COMAPT_FREEBSD4 */
1007 
1008 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1009 #ifndef _SYS_SYSPROTO_H_
1010 struct osigaction_args {
1011 	int	signum;
1012 	struct	osigaction *nsa;
1013 	struct	osigaction *osa;
1014 };
1015 #endif
1016 int
osigaction(struct thread * td,struct osigaction_args * uap)1017 osigaction(struct thread *td, struct osigaction_args *uap)
1018 {
1019 	struct osigaction sa;
1020 	struct sigaction nsa, osa;
1021 	struct sigaction *nsap, *osap;
1022 	int error;
1023 
1024 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1025 		return (EINVAL);
1026 
1027 	nsap = (uap->nsa != NULL) ? &nsa : NULL;
1028 	osap = (uap->osa != NULL) ? &osa : NULL;
1029 
1030 	if (nsap) {
1031 		error = copyin(uap->nsa, &sa, sizeof(sa));
1032 		if (error)
1033 			return (error);
1034 		nsap->sa_handler = sa.sa_handler;
1035 		nsap->sa_flags = sa.sa_flags;
1036 		OSIG2SIG(sa.sa_mask, nsap->sa_mask);
1037 	}
1038 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1039 	if (osap && !error) {
1040 		sa.sa_handler = osap->sa_handler;
1041 		sa.sa_flags = osap->sa_flags;
1042 		SIG2OSIG(osap->sa_mask, sa.sa_mask);
1043 		error = copyout(&sa, uap->osa, sizeof(sa));
1044 	}
1045 	return (error);
1046 }
1047 
1048 #if !defined(__i386__)
1049 /* Avoid replicating the same stub everywhere */
1050 int
osigreturn(struct thread * td,struct osigreturn_args * uap)1051 osigreturn(struct thread *td, struct osigreturn_args *uap)
1052 {
1053 	return (kern_nosys(td, 0));
1054 }
1055 #endif
1056 #endif /* COMPAT_43 */
1057 
1058 /*
1059  * Initialize signal state for process 0;
1060  * set to ignore signals that are ignored by default.
1061  */
1062 void
siginit(struct proc * p)1063 siginit(struct proc *p)
1064 {
1065 	int i;
1066 	struct sigacts *ps;
1067 
1068 	PROC_LOCK(p);
1069 	ps = p->p_sigacts;
1070 	mtx_lock(&ps->ps_mtx);
1071 	for (i = 1; i <= NSIG; i++) {
1072 		if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) {
1073 			SIGADDSET(ps->ps_sigignore, i);
1074 		}
1075 	}
1076 	mtx_unlock(&ps->ps_mtx);
1077 	PROC_UNLOCK(p);
1078 }
1079 
1080 /*
1081  * Reset specified signal to the default disposition.
1082  */
1083 static void
sigdflt(struct sigacts * ps,int sig)1084 sigdflt(struct sigacts *ps, int sig)
1085 {
1086 
1087 	mtx_assert(&ps->ps_mtx, MA_OWNED);
1088 	SIGDELSET(ps->ps_sigcatch, sig);
1089 	if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT)
1090 		SIGADDSET(ps->ps_sigignore, sig);
1091 	ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
1092 	SIGDELSET(ps->ps_siginfo, sig);
1093 }
1094 
1095 /*
1096  * Reset signals for an exec of the specified process.
1097  */
1098 void
execsigs(struct proc * p)1099 execsigs(struct proc *p)
1100 {
1101 	struct sigacts *ps;
1102 	struct thread *td;
1103 
1104 	/*
1105 	 * Reset caught signals.  Held signals remain held
1106 	 * through td_sigmask (unless they were caught,
1107 	 * and are now ignored by default).
1108 	 */
1109 	PROC_LOCK_ASSERT(p, MA_OWNED);
1110 	ps = p->p_sigacts;
1111 	mtx_lock(&ps->ps_mtx);
1112 	sig_drop_caught(p);
1113 
1114 	/*
1115 	 * Reset stack state to the user stack.
1116 	 * Clear set of signals caught on the signal stack.
1117 	 */
1118 	td = curthread;
1119 	MPASS(td->td_proc == p);
1120 	td->td_sigstk.ss_flags = SS_DISABLE;
1121 	td->td_sigstk.ss_size = 0;
1122 	td->td_sigstk.ss_sp = 0;
1123 	td->td_pflags &= ~TDP_ALTSTACK;
1124 	/*
1125 	 * Reset no zombies if child dies flag as Solaris does.
1126 	 */
1127 	ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1128 	if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1129 		ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1130 	mtx_unlock(&ps->ps_mtx);
1131 }
1132 
1133 /*
1134  * kern_sigprocmask()
1135  *
1136  *	Manipulate signal mask.
1137  */
1138 int
kern_sigprocmask(struct thread * td,int how,sigset_t * set,sigset_t * oset,int flags)1139 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1140     int flags)
1141 {
1142 	sigset_t new_block, oset1;
1143 	struct proc *p;
1144 	int error;
1145 
1146 	p = td->td_proc;
1147 	if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1148 		PROC_LOCK_ASSERT(p, MA_OWNED);
1149 	else
1150 		PROC_LOCK(p);
1151 	mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1152 	    ? MA_OWNED : MA_NOTOWNED);
1153 	if (oset != NULL)
1154 		*oset = td->td_sigmask;
1155 
1156 	error = 0;
1157 	if (set != NULL) {
1158 		switch (how) {
1159 		case SIG_BLOCK:
1160 			SIG_CANTMASK(*set);
1161 			oset1 = td->td_sigmask;
1162 			SIGSETOR(td->td_sigmask, *set);
1163 			new_block = td->td_sigmask;
1164 			SIGSETNAND(new_block, oset1);
1165 			break;
1166 		case SIG_UNBLOCK:
1167 			SIGSETNAND(td->td_sigmask, *set);
1168 			signotify(td);
1169 			goto out;
1170 		case SIG_SETMASK:
1171 			SIG_CANTMASK(*set);
1172 			oset1 = td->td_sigmask;
1173 			if (flags & SIGPROCMASK_OLD)
1174 				SIGSETLO(td->td_sigmask, *set);
1175 			else
1176 				td->td_sigmask = *set;
1177 			new_block = td->td_sigmask;
1178 			SIGSETNAND(new_block, oset1);
1179 			signotify(td);
1180 			break;
1181 		default:
1182 			error = EINVAL;
1183 			goto out;
1184 		}
1185 
1186 		/*
1187 		 * The new_block set contains signals that were not previously
1188 		 * blocked, but are blocked now.
1189 		 *
1190 		 * In case we block any signal that was not previously blocked
1191 		 * for td, and process has the signal pending, try to schedule
1192 		 * signal delivery to some thread that does not block the
1193 		 * signal, possibly waking it up.
1194 		 */
1195 		if (p->p_numthreads != 1)
1196 			reschedule_signals(p, new_block, flags);
1197 	}
1198 
1199 out:
1200 	if (!(flags & SIGPROCMASK_PROC_LOCKED))
1201 		PROC_UNLOCK(p);
1202 	return (error);
1203 }
1204 
1205 #ifndef _SYS_SYSPROTO_H_
1206 struct sigprocmask_args {
1207 	int	how;
1208 	const sigset_t *set;
1209 	sigset_t *oset;
1210 };
1211 #endif
1212 int
sys_sigprocmask(struct thread * td,struct sigprocmask_args * uap)1213 sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1214 {
1215 	sigset_t set, oset;
1216 	sigset_t *setp, *osetp;
1217 	int error;
1218 
1219 	setp = (uap->set != NULL) ? &set : NULL;
1220 	osetp = (uap->oset != NULL) ? &oset : NULL;
1221 	if (setp) {
1222 		error = copyin(uap->set, setp, sizeof(set));
1223 		if (error)
1224 			return (error);
1225 	}
1226 	error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1227 	if (osetp && !error) {
1228 		error = copyout(osetp, uap->oset, sizeof(oset));
1229 	}
1230 	return (error);
1231 }
1232 
1233 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1234 #ifndef _SYS_SYSPROTO_H_
1235 struct osigprocmask_args {
1236 	int	how;
1237 	osigset_t mask;
1238 };
1239 #endif
1240 int
osigprocmask(struct thread * td,struct osigprocmask_args * uap)1241 osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1242 {
1243 	sigset_t set, oset;
1244 	int error;
1245 
1246 	OSIG2SIG(uap->mask, set);
1247 	error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1248 	SIG2OSIG(oset, td->td_retval[0]);
1249 	return (error);
1250 }
1251 #endif /* COMPAT_43 */
1252 
1253 int
sys_sigwait(struct thread * td,struct sigwait_args * uap)1254 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1255 {
1256 	ksiginfo_t ksi;
1257 	sigset_t set;
1258 	int error;
1259 
1260 	error = copyin(uap->set, &set, sizeof(set));
1261 	if (error) {
1262 		td->td_retval[0] = error;
1263 		return (0);
1264 	}
1265 
1266 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1267 	if (error) {
1268 		/*
1269 		 * sigwait() function shall not return EINTR, but
1270 		 * the syscall does.  Non-ancient libc provides the
1271 		 * wrapper which hides EINTR.  Otherwise, EINTR return
1272 		 * is used by libthr to handle required cancellation
1273 		 * point in the sigwait().
1274 		 */
1275 		if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1276 			return (ERESTART);
1277 		td->td_retval[0] = error;
1278 		return (0);
1279 	}
1280 
1281 	error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1282 	td->td_retval[0] = error;
1283 	return (0);
1284 }
1285 
1286 int
sys_sigtimedwait(struct thread * td,struct sigtimedwait_args * uap)1287 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1288 {
1289 	struct timespec ts;
1290 	struct timespec *timeout;
1291 	sigset_t set;
1292 	ksiginfo_t ksi;
1293 	int error;
1294 
1295 	if (uap->timeout) {
1296 		error = copyin(uap->timeout, &ts, sizeof(ts));
1297 		if (error)
1298 			return (error);
1299 
1300 		timeout = &ts;
1301 	} else
1302 		timeout = NULL;
1303 
1304 	error = copyin(uap->set, &set, sizeof(set));
1305 	if (error)
1306 		return (error);
1307 
1308 	error = kern_sigtimedwait(td, set, &ksi, timeout);
1309 	if (error)
1310 		return (error);
1311 
1312 	if (uap->info)
1313 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1314 
1315 	if (error == 0)
1316 		td->td_retval[0] = ksi.ksi_signo;
1317 	return (error);
1318 }
1319 
1320 int
sys_sigwaitinfo(struct thread * td,struct sigwaitinfo_args * uap)1321 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1322 {
1323 	ksiginfo_t ksi;
1324 	sigset_t set;
1325 	int error;
1326 
1327 	error = copyin(uap->set, &set, sizeof(set));
1328 	if (error)
1329 		return (error);
1330 
1331 	error = kern_sigtimedwait(td, set, &ksi, NULL);
1332 	if (error)
1333 		return (error);
1334 
1335 	if (uap->info)
1336 		error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1337 
1338 	if (error == 0)
1339 		td->td_retval[0] = ksi.ksi_signo;
1340 	return (error);
1341 }
1342 
1343 static void
proc_td_siginfo_capture(struct thread * td,siginfo_t * si)1344 proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1345 {
1346 	struct thread *thr;
1347 
1348 	FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1349 		if (thr == td)
1350 			thr->td_si = *si;
1351 		else
1352 			thr->td_si.si_signo = 0;
1353 	}
1354 }
1355 
1356 int
kern_sigtimedwait(struct thread * td,sigset_t waitset,ksiginfo_t * ksi,struct timespec * timeout)1357 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1358 	struct timespec *timeout)
1359 {
1360 	struct sigacts *ps;
1361 	sigset_t saved_mask, new_block;
1362 	struct proc *p;
1363 	int error, sig, timevalid = 0;
1364 	sbintime_t sbt, precision, tsbt;
1365 	struct timespec ts;
1366 	bool traced;
1367 
1368 	p = td->td_proc;
1369 	error = 0;
1370 	traced = false;
1371 
1372 	/* Ensure the sigfastblock value is up to date. */
1373 	sigfastblock_fetch(td);
1374 
1375 	if (timeout != NULL) {
1376 		if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1377 			timevalid = 1;
1378 			ts = *timeout;
1379 			if (ts.tv_sec < INT32_MAX / 2) {
1380 				tsbt = tstosbt(ts);
1381 				precision = tsbt;
1382 				precision >>= tc_precexp;
1383 				if (TIMESEL(&sbt, tsbt))
1384 					sbt += tc_tick_sbt;
1385 				sbt += tsbt;
1386 			} else
1387 				precision = sbt = 0;
1388 		}
1389 	} else
1390 		precision = sbt = 0;
1391 	ksiginfo_init(ksi);
1392 	/* Some signals can not be waited for. */
1393 	SIG_CANTMASK(waitset);
1394 	ps = p->p_sigacts;
1395 	PROC_LOCK(p);
1396 	saved_mask = td->td_sigmask;
1397 	SIGSETNAND(td->td_sigmask, waitset);
1398 	if ((p->p_sysent->sv_flags & SV_SIG_DISCIGN) != 0 ||
1399 	    !kern_sig_discard_ign) {
1400 		thread_lock(td);
1401 		td->td_flags |= TDF_SIGWAIT;
1402 		thread_unlock(td);
1403 	}
1404 	for (;;) {
1405 		mtx_lock(&ps->ps_mtx);
1406 		sig = cursig(td);
1407 		mtx_unlock(&ps->ps_mtx);
1408 		KASSERT(sig >= 0, ("sig %d", sig));
1409 		if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1410 			if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1411 			    sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1412 				error = 0;
1413 				break;
1414 			}
1415 		}
1416 
1417 		if (error != 0)
1418 			break;
1419 
1420 		/*
1421 		 * POSIX says this must be checked after looking for pending
1422 		 * signals.
1423 		 */
1424 		if (timeout != NULL && !timevalid) {
1425 			error = EINVAL;
1426 			break;
1427 		}
1428 
1429 		if (traced) {
1430 			error = EINTR;
1431 			break;
1432 		}
1433 
1434 		error = msleep_sbt(&p->p_sigacts, &p->p_mtx, PPAUSE | PCATCH,
1435 		    "sigwait", sbt, precision, C_ABSOLUTE);
1436 
1437 		/* The syscalls can not be restarted. */
1438 		if (error == ERESTART)
1439 			error = EINTR;
1440 
1441 		/*
1442 		 * If PTRACE_SCE or PTRACE_SCX were set after
1443 		 * userspace entered the syscall, return spurious
1444 		 * EINTR after wait was done.  Only do this as last
1445 		 * resort after rechecking for possible queued signals
1446 		 * and expired timeouts.
1447 		 */
1448 		if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1449 			traced = true;
1450 	}
1451 	thread_lock(td);
1452 	td->td_flags &= ~TDF_SIGWAIT;
1453 	thread_unlock(td);
1454 
1455 	new_block = saved_mask;
1456 	SIGSETNAND(new_block, td->td_sigmask);
1457 	td->td_sigmask = saved_mask;
1458 	/*
1459 	 * Fewer signals can be delivered to us, reschedule signal
1460 	 * notification.
1461 	 */
1462 	if (p->p_numthreads != 1)
1463 		reschedule_signals(p, new_block, 0);
1464 
1465 	if (error == 0) {
1466 		SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1467 
1468 		if (ksi->ksi_code == SI_TIMER)
1469 			itimer_accept(p, ksi->ksi_timerid, ksi);
1470 
1471 #ifdef KTRACE
1472 		if (KTRPOINT(td, KTR_PSIG)) {
1473 			sig_t action;
1474 
1475 			mtx_lock(&ps->ps_mtx);
1476 			action = ps->ps_sigact[_SIG_IDX(sig)];
1477 			mtx_unlock(&ps->ps_mtx);
1478 			ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1479 		}
1480 #endif
1481 		if (sig == SIGKILL) {
1482 			proc_td_siginfo_capture(td, &ksi->ksi_info);
1483 			sigexit(td, sig);
1484 		}
1485 	}
1486 	PROC_UNLOCK(p);
1487 	return (error);
1488 }
1489 
1490 #ifndef _SYS_SYSPROTO_H_
1491 struct sigpending_args {
1492 	sigset_t	*set;
1493 };
1494 #endif
1495 int
sys_sigpending(struct thread * td,struct sigpending_args * uap)1496 sys_sigpending(struct thread *td, struct sigpending_args *uap)
1497 {
1498 	struct proc *p = td->td_proc;
1499 	sigset_t pending;
1500 
1501 	PROC_LOCK(p);
1502 	pending = p->p_sigqueue.sq_signals;
1503 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1504 	PROC_UNLOCK(p);
1505 	return (copyout(&pending, uap->set, sizeof(sigset_t)));
1506 }
1507 
1508 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1509 #ifndef _SYS_SYSPROTO_H_
1510 struct osigpending_args {
1511 	int	dummy;
1512 };
1513 #endif
1514 int
osigpending(struct thread * td,struct osigpending_args * uap)1515 osigpending(struct thread *td, struct osigpending_args *uap)
1516 {
1517 	struct proc *p = td->td_proc;
1518 	sigset_t pending;
1519 
1520 	PROC_LOCK(p);
1521 	pending = p->p_sigqueue.sq_signals;
1522 	SIGSETOR(pending, td->td_sigqueue.sq_signals);
1523 	PROC_UNLOCK(p);
1524 	SIG2OSIG(pending, td->td_retval[0]);
1525 	return (0);
1526 }
1527 #endif /* COMPAT_43 */
1528 
1529 #if defined(COMPAT_43)
1530 /*
1531  * Generalized interface signal handler, 4.3-compatible.
1532  */
1533 #ifndef _SYS_SYSPROTO_H_
1534 struct osigvec_args {
1535 	int	signum;
1536 	struct	sigvec *nsv;
1537 	struct	sigvec *osv;
1538 };
1539 #endif
1540 /* ARGSUSED */
1541 int
osigvec(struct thread * td,struct osigvec_args * uap)1542 osigvec(struct thread *td, struct osigvec_args *uap)
1543 {
1544 	struct sigvec vec;
1545 	struct sigaction nsa, osa;
1546 	struct sigaction *nsap, *osap;
1547 	int error;
1548 
1549 	if (uap->signum <= 0 || uap->signum >= ONSIG)
1550 		return (EINVAL);
1551 	nsap = (uap->nsv != NULL) ? &nsa : NULL;
1552 	osap = (uap->osv != NULL) ? &osa : NULL;
1553 	if (nsap) {
1554 		error = copyin(uap->nsv, &vec, sizeof(vec));
1555 		if (error)
1556 			return (error);
1557 		nsap->sa_handler = vec.sv_handler;
1558 		OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1559 		nsap->sa_flags = vec.sv_flags;
1560 		nsap->sa_flags ^= SA_RESTART;	/* opposite of SV_INTERRUPT */
1561 	}
1562 	error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1563 	if (osap && !error) {
1564 		vec.sv_handler = osap->sa_handler;
1565 		SIG2OSIG(osap->sa_mask, vec.sv_mask);
1566 		vec.sv_flags = osap->sa_flags;
1567 		vec.sv_flags &= ~SA_NOCLDWAIT;
1568 		vec.sv_flags ^= SA_RESTART;
1569 		error = copyout(&vec, uap->osv, sizeof(vec));
1570 	}
1571 	return (error);
1572 }
1573 
1574 #ifndef _SYS_SYSPROTO_H_
1575 struct osigblock_args {
1576 	int	mask;
1577 };
1578 #endif
1579 int
osigblock(struct thread * td,struct osigblock_args * uap)1580 osigblock(struct thread *td, struct osigblock_args *uap)
1581 {
1582 	sigset_t set, oset;
1583 
1584 	OSIG2SIG(uap->mask, set);
1585 	kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1586 	SIG2OSIG(oset, td->td_retval[0]);
1587 	return (0);
1588 }
1589 
1590 #ifndef _SYS_SYSPROTO_H_
1591 struct osigsetmask_args {
1592 	int	mask;
1593 };
1594 #endif
1595 int
osigsetmask(struct thread * td,struct osigsetmask_args * uap)1596 osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1597 {
1598 	sigset_t set, oset;
1599 
1600 	OSIG2SIG(uap->mask, set);
1601 	kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1602 	SIG2OSIG(oset, td->td_retval[0]);
1603 	return (0);
1604 }
1605 #endif /* COMPAT_43 */
1606 
1607 /*
1608  * Suspend calling thread until signal, providing mask to be set in the
1609  * meantime.
1610  */
1611 #ifndef _SYS_SYSPROTO_H_
1612 struct sigsuspend_args {
1613 	const sigset_t *sigmask;
1614 };
1615 #endif
1616 /* ARGSUSED */
1617 int
sys_sigsuspend(struct thread * td,struct sigsuspend_args * uap)1618 sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1619 {
1620 	sigset_t mask;
1621 	int error;
1622 
1623 	error = copyin(uap->sigmask, &mask, sizeof(mask));
1624 	if (error)
1625 		return (error);
1626 	return (kern_sigsuspend(td, mask));
1627 }
1628 
1629 int
kern_sigsuspend(struct thread * td,sigset_t mask)1630 kern_sigsuspend(struct thread *td, sigset_t mask)
1631 {
1632 	struct proc *p = td->td_proc;
1633 	int has_sig, sig;
1634 
1635 	/* Ensure the sigfastblock value is up to date. */
1636 	sigfastblock_fetch(td);
1637 
1638 	/*
1639 	 * When returning from sigsuspend, we want
1640 	 * the old mask to be restored after the
1641 	 * signal handler has finished.  Thus, we
1642 	 * save it here and mark the sigacts structure
1643 	 * to indicate this.
1644 	 */
1645 	PROC_LOCK(p);
1646 	kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1647 	    SIGPROCMASK_PROC_LOCKED);
1648 	td->td_pflags |= TDP_OLDMASK;
1649 	ast_sched(td, TDA_SIGSUSPEND);
1650 
1651 	/*
1652 	 * Process signals now. Otherwise, we can get spurious wakeup
1653 	 * due to signal entered process queue, but delivered to other
1654 	 * thread. But sigsuspend should return only on signal
1655 	 * delivery.
1656 	 */
1657 	(p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1658 	for (has_sig = 0; !has_sig;) {
1659 		while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE | PCATCH,
1660 		    "sigsusp", 0) == 0)
1661 			/* void */;
1662 		thread_suspend_check(0);
1663 		mtx_lock(&p->p_sigacts->ps_mtx);
1664 		while ((sig = cursig(td)) != 0) {
1665 			KASSERT(sig >= 0, ("sig %d", sig));
1666 			has_sig += postsig(sig);
1667 		}
1668 		mtx_unlock(&p->p_sigacts->ps_mtx);
1669 
1670 		/*
1671 		 * If PTRACE_SCE or PTRACE_SCX were set after
1672 		 * userspace entered the syscall, return spurious
1673 		 * EINTR.
1674 		 */
1675 		if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1676 			has_sig += 1;
1677 	}
1678 	PROC_UNLOCK(p);
1679 	td->td_errno = EINTR;
1680 	td->td_pflags |= TDP_NERRNO;
1681 	return (EJUSTRETURN);
1682 }
1683 
1684 #ifdef COMPAT_43	/* XXX - COMPAT_FBSD3 */
1685 /*
1686  * Compatibility sigsuspend call for old binaries.  Note nonstandard calling
1687  * convention: libc stub passes mask, not pointer, to save a copyin.
1688  */
1689 #ifndef _SYS_SYSPROTO_H_
1690 struct osigsuspend_args {
1691 	osigset_t mask;
1692 };
1693 #endif
1694 /* ARGSUSED */
1695 int
osigsuspend(struct thread * td,struct osigsuspend_args * uap)1696 osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1697 {
1698 	sigset_t mask;
1699 
1700 	OSIG2SIG(uap->mask, mask);
1701 	return (kern_sigsuspend(td, mask));
1702 }
1703 #endif /* COMPAT_43 */
1704 
1705 #if defined(COMPAT_43)
1706 #ifndef _SYS_SYSPROTO_H_
1707 struct osigstack_args {
1708 	struct	sigstack *nss;
1709 	struct	sigstack *oss;
1710 };
1711 #endif
1712 /* ARGSUSED */
1713 int
osigstack(struct thread * td,struct osigstack_args * uap)1714 osigstack(struct thread *td, struct osigstack_args *uap)
1715 {
1716 	struct sigstack nss, oss;
1717 	int error = 0;
1718 
1719 	if (uap->nss != NULL) {
1720 		error = copyin(uap->nss, &nss, sizeof(nss));
1721 		if (error)
1722 			return (error);
1723 	}
1724 	oss.ss_sp = td->td_sigstk.ss_sp;
1725 	oss.ss_onstack = sigonstack(cpu_getstack(td));
1726 	if (uap->nss != NULL) {
1727 		td->td_sigstk.ss_sp = nss.ss_sp;
1728 		td->td_sigstk.ss_size = 0;
1729 		td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1730 		td->td_pflags |= TDP_ALTSTACK;
1731 	}
1732 	if (uap->oss != NULL)
1733 		error = copyout(&oss, uap->oss, sizeof(oss));
1734 
1735 	return (error);
1736 }
1737 #endif /* COMPAT_43 */
1738 
1739 #ifndef _SYS_SYSPROTO_H_
1740 struct sigaltstack_args {
1741 	stack_t	*ss;
1742 	stack_t	*oss;
1743 };
1744 #endif
1745 /* ARGSUSED */
1746 int
sys_sigaltstack(struct thread * td,struct sigaltstack_args * uap)1747 sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1748 {
1749 	stack_t ss, oss;
1750 	int error;
1751 
1752 	if (uap->ss != NULL) {
1753 		error = copyin(uap->ss, &ss, sizeof(ss));
1754 		if (error)
1755 			return (error);
1756 	}
1757 	error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1758 	    (uap->oss != NULL) ? &oss : NULL);
1759 	if (error)
1760 		return (error);
1761 	if (uap->oss != NULL)
1762 		error = copyout(&oss, uap->oss, sizeof(stack_t));
1763 	return (error);
1764 }
1765 
1766 int
kern_sigaltstack(struct thread * td,stack_t * ss,stack_t * oss)1767 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1768 {
1769 	struct proc *p = td->td_proc;
1770 	int oonstack;
1771 
1772 	oonstack = sigonstack(cpu_getstack(td));
1773 
1774 	if (oss != NULL) {
1775 		*oss = td->td_sigstk;
1776 		oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1777 		    ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1778 	}
1779 
1780 	if (ss != NULL) {
1781 		if (oonstack)
1782 			return (EPERM);
1783 		if ((ss->ss_flags & ~SS_DISABLE) != 0)
1784 			return (EINVAL);
1785 		if (!(ss->ss_flags & SS_DISABLE)) {
1786 			if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1787 				return (ENOMEM);
1788 
1789 			td->td_sigstk = *ss;
1790 			td->td_pflags |= TDP_ALTSTACK;
1791 		} else {
1792 			td->td_pflags &= ~TDP_ALTSTACK;
1793 		}
1794 	}
1795 	return (0);
1796 }
1797 
1798 struct killpg1_ctx {
1799 	struct thread *td;
1800 	ksiginfo_t *ksi;
1801 	int sig;
1802 	bool sent;
1803 	bool found;
1804 	int ret;
1805 };
1806 
1807 static void
killpg1_sendsig_locked(struct proc * p,struct killpg1_ctx * arg)1808 killpg1_sendsig_locked(struct proc *p, struct killpg1_ctx *arg)
1809 {
1810 	int err;
1811 
1812 	err = p_cansignal(arg->td, p, arg->sig);
1813 	if (err == 0 && arg->sig != 0)
1814 		pksignal(p, arg->sig, arg->ksi);
1815 	if (err != ESRCH)
1816 		arg->found = true;
1817 	if (err == 0)
1818 		arg->sent = true;
1819 	else if (arg->ret == 0 && err != ESRCH && err != EPERM)
1820 		arg->ret = err;
1821 }
1822 
1823 static void
killpg1_sendsig(struct proc * p,bool notself,struct killpg1_ctx * arg)1824 killpg1_sendsig(struct proc *p, bool notself, struct killpg1_ctx *arg)
1825 {
1826 
1827 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1828 	    (notself && p == arg->td->td_proc) || p->p_state == PRS_NEW)
1829 		return;
1830 
1831 	PROC_LOCK(p);
1832 	killpg1_sendsig_locked(p, arg);
1833 	PROC_UNLOCK(p);
1834 }
1835 
1836 static void
kill_processes_prison_cb(struct proc * p,void * arg)1837 kill_processes_prison_cb(struct proc *p, void *arg)
1838 {
1839 	struct killpg1_ctx *ctx = arg;
1840 
1841 	if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1842 	    (p == ctx->td->td_proc) || p->p_state == PRS_NEW)
1843 		return;
1844 
1845 	killpg1_sendsig_locked(p, ctx);
1846 }
1847 
1848 /*
1849  * Common code for kill process group/broadcast kill.
1850  * td is the calling thread, as usual.
1851  */
1852 static int
killpg1(struct thread * td,int sig,int pgid,int all,ksiginfo_t * ksi)1853 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1854 {
1855 	struct proc *p;
1856 	struct pgrp *pgrp;
1857 	struct killpg1_ctx arg;
1858 
1859 	arg.td = td;
1860 	arg.ksi = ksi;
1861 	arg.sig = sig;
1862 	arg.sent = false;
1863 	arg.found = false;
1864 	arg.ret = 0;
1865 	if (all) {
1866 		/*
1867 		 * broadcast
1868 		 */
1869 		prison_proc_iterate(td->td_ucred->cr_prison,
1870 		    kill_processes_prison_cb, &arg);
1871 	} else {
1872 again:
1873 		sx_slock(&proctree_lock);
1874 		if (pgid == 0) {
1875 			/*
1876 			 * zero pgid means send to my process group.
1877 			 */
1878 			pgrp = td->td_proc->p_pgrp;
1879 			PGRP_LOCK(pgrp);
1880 		} else {
1881 			pgrp = pgfind(pgid);
1882 			if (pgrp == NULL) {
1883 				sx_sunlock(&proctree_lock);
1884 				return (ESRCH);
1885 			}
1886 		}
1887 		sx_sunlock(&proctree_lock);
1888 		if (!sx_try_xlock(&pgrp->pg_killsx)) {
1889 			PGRP_UNLOCK(pgrp);
1890 			sx_xlock(&pgrp->pg_killsx);
1891 			sx_xunlock(&pgrp->pg_killsx);
1892 			goto again;
1893 		}
1894 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1895 			killpg1_sendsig(p, false, &arg);
1896 		}
1897 		PGRP_UNLOCK(pgrp);
1898 		sx_xunlock(&pgrp->pg_killsx);
1899 	}
1900 	MPASS(arg.ret != 0 || arg.found || !arg.sent);
1901 	if (arg.ret == 0 && !arg.sent)
1902 		arg.ret = arg.found ? EPERM : ESRCH;
1903 	return (arg.ret);
1904 }
1905 
1906 #ifndef _SYS_SYSPROTO_H_
1907 struct kill_args {
1908 	int	pid;
1909 	int	signum;
1910 };
1911 #endif
1912 /* ARGSUSED */
1913 int
sys_kill(struct thread * td,struct kill_args * uap)1914 sys_kill(struct thread *td, struct kill_args *uap)
1915 {
1916 
1917 	return (kern_kill(td, uap->pid, uap->signum));
1918 }
1919 
1920 int
kern_kill(struct thread * td,pid_t pid,int signum)1921 kern_kill(struct thread *td, pid_t pid, int signum)
1922 {
1923 	ksiginfo_t ksi;
1924 	struct proc *p;
1925 	int error;
1926 
1927 	/*
1928 	 * A process in capability mode can send signals only to himself.
1929 	 * The main rationale behind this is that abort(3) is implemented as
1930 	 * kill(getpid(), SIGABRT).
1931 	 */
1932 	if (pid != td->td_proc->p_pid) {
1933 		if (CAP_TRACING(td))
1934 			ktrcapfail(CAPFAIL_SIGNAL, &signum);
1935 		if (IN_CAPABILITY_MODE(td))
1936 			return (ECAPMODE);
1937 	}
1938 
1939 	AUDIT_ARG_SIGNUM(signum);
1940 	AUDIT_ARG_PID(pid);
1941 	if ((u_int)signum > _SIG_MAXSIG)
1942 		return (EINVAL);
1943 
1944 	ksiginfo_init(&ksi);
1945 	ksi.ksi_signo = signum;
1946 	ksi.ksi_code = SI_USER;
1947 	ksi.ksi_pid = td->td_proc->p_pid;
1948 	ksi.ksi_uid = td->td_ucred->cr_ruid;
1949 
1950 	if (pid > 0) {
1951 		/* kill single process */
1952 		if ((p = pfind_any(pid)) == NULL)
1953 			return (ESRCH);
1954 		AUDIT_ARG_PROCESS(p);
1955 		error = p_cansignal(td, p, signum);
1956 		if (error == 0 && signum)
1957 			pksignal(p, signum, &ksi);
1958 		PROC_UNLOCK(p);
1959 		return (error);
1960 	}
1961 	switch (pid) {
1962 	case -1:		/* broadcast signal */
1963 		return (killpg1(td, signum, 0, 1, &ksi));
1964 	case 0:			/* signal own process group */
1965 		return (killpg1(td, signum, 0, 0, &ksi));
1966 	default:		/* negative explicit process group */
1967 		return (killpg1(td, signum, -pid, 0, &ksi));
1968 	}
1969 	/* NOTREACHED */
1970 }
1971 
1972 int
sys_pdkill(struct thread * td,struct pdkill_args * uap)1973 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1974 {
1975 	struct proc *p;
1976 	int error;
1977 
1978 	AUDIT_ARG_SIGNUM(uap->signum);
1979 	AUDIT_ARG_FD(uap->fd);
1980 	if ((u_int)uap->signum > _SIG_MAXSIG)
1981 		return (EINVAL);
1982 
1983 	error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1984 	if (error)
1985 		return (error);
1986 	AUDIT_ARG_PROCESS(p);
1987 	error = p_cansignal(td, p, uap->signum);
1988 	if (error == 0 && uap->signum)
1989 		kern_psignal(p, uap->signum);
1990 	PROC_UNLOCK(p);
1991 	return (error);
1992 }
1993 
1994 #if defined(COMPAT_43)
1995 #ifndef _SYS_SYSPROTO_H_
1996 struct okillpg_args {
1997 	int	pgid;
1998 	int	signum;
1999 };
2000 #endif
2001 /* ARGSUSED */
2002 int
okillpg(struct thread * td,struct okillpg_args * uap)2003 okillpg(struct thread *td, struct okillpg_args *uap)
2004 {
2005 	ksiginfo_t ksi;
2006 
2007 	AUDIT_ARG_SIGNUM(uap->signum);
2008 	AUDIT_ARG_PID(uap->pgid);
2009 	if ((u_int)uap->signum > _SIG_MAXSIG)
2010 		return (EINVAL);
2011 
2012 	ksiginfo_init(&ksi);
2013 	ksi.ksi_signo = uap->signum;
2014 	ksi.ksi_code = SI_USER;
2015 	ksi.ksi_pid = td->td_proc->p_pid;
2016 	ksi.ksi_uid = td->td_ucred->cr_ruid;
2017 	return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
2018 }
2019 #endif /* COMPAT_43 */
2020 
2021 #ifndef _SYS_SYSPROTO_H_
2022 struct sigqueue_args {
2023 	pid_t pid;
2024 	int signum;
2025 	/* union sigval */ void *value;
2026 };
2027 #endif
2028 int
sys_sigqueue(struct thread * td,struct sigqueue_args * uap)2029 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
2030 {
2031 	union sigval sv;
2032 
2033 	sv.sival_ptr = uap->value;
2034 
2035 	return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
2036 }
2037 
2038 int
kern_sigqueue(struct thread * td,pid_t pid,int signumf,union sigval * value)2039 kern_sigqueue(struct thread *td, pid_t pid, int signumf, union sigval *value)
2040 {
2041 	ksiginfo_t ksi;
2042 	struct proc *p;
2043 	struct thread *td2;
2044 	u_int signum;
2045 	int error;
2046 
2047 	signum = signumf & ~__SIGQUEUE_TID;
2048 	if (signum > _SIG_MAXSIG)
2049 		return (EINVAL);
2050 
2051 	/*
2052 	 * Specification says sigqueue can only send signal to
2053 	 * single process.
2054 	 */
2055 	if (pid <= 0)
2056 		return (EINVAL);
2057 
2058 	if ((signumf & __SIGQUEUE_TID) == 0) {
2059 		if ((p = pfind_any(pid)) == NULL)
2060 			return (ESRCH);
2061 		td2 = NULL;
2062 	} else {
2063 		p = td->td_proc;
2064 		td2 = tdfind((lwpid_t)pid, p->p_pid);
2065 		if (td2 == NULL)
2066 			return (ESRCH);
2067 	}
2068 
2069 	error = p_cansignal(td, p, signum);
2070 	if (error == 0 && signum != 0) {
2071 		ksiginfo_init(&ksi);
2072 		ksi.ksi_flags = KSI_SIGQ;
2073 		ksi.ksi_signo = signum;
2074 		ksi.ksi_code = SI_QUEUE;
2075 		ksi.ksi_pid = td->td_proc->p_pid;
2076 		ksi.ksi_uid = td->td_ucred->cr_ruid;
2077 		ksi.ksi_value = *value;
2078 		error = tdsendsignal(p, td2, ksi.ksi_signo, &ksi);
2079 	}
2080 	PROC_UNLOCK(p);
2081 	return (error);
2082 }
2083 
2084 /*
2085  * Send a signal to a process group.  If checktty is 1,
2086  * limit to members which have a controlling terminal.
2087  */
2088 void
pgsignal(struct pgrp * pgrp,int sig,int checkctty,ksiginfo_t * ksi)2089 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
2090 {
2091 	struct proc *p;
2092 
2093 	if (pgrp) {
2094 		PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
2095 		LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
2096 			PROC_LOCK(p);
2097 			if (p->p_state == PRS_NORMAL &&
2098 			    (checkctty == 0 || p->p_flag & P_CONTROLT))
2099 				pksignal(p, sig, ksi);
2100 			PROC_UNLOCK(p);
2101 		}
2102 	}
2103 }
2104 
2105 /*
2106  * Recalculate the signal mask and reset the signal disposition after
2107  * usermode frame for delivery is formed.  Should be called after
2108  * mach-specific routine, because sysent->sv_sendsig() needs correct
2109  * ps_siginfo and signal mask.
2110  */
2111 static void
postsig_done(int sig,struct thread * td,struct sigacts * ps)2112 postsig_done(int sig, struct thread *td, struct sigacts *ps)
2113 {
2114 	sigset_t mask;
2115 
2116 	mtx_assert(&ps->ps_mtx, MA_OWNED);
2117 	td->td_ru.ru_nsignals++;
2118 	mask = ps->ps_catchmask[_SIG_IDX(sig)];
2119 	if (!SIGISMEMBER(ps->ps_signodefer, sig))
2120 		SIGADDSET(mask, sig);
2121 	kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
2122 	    SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
2123 	if (SIGISMEMBER(ps->ps_sigreset, sig))
2124 		sigdflt(ps, sig);
2125 }
2126 
2127 /*
2128  * Send a signal caused by a trap to the current thread.  If it will be
2129  * caught immediately, deliver it with correct code.  Otherwise, post it
2130  * normally.
2131  */
2132 void
trapsignal(struct thread * td,ksiginfo_t * ksi)2133 trapsignal(struct thread *td, ksiginfo_t *ksi)
2134 {
2135 	struct sigacts *ps;
2136 	struct proc *p;
2137 	sigset_t sigmask;
2138 	int sig;
2139 
2140 	p = td->td_proc;
2141 	sig = ksi->ksi_signo;
2142 	KASSERT(_SIG_VALID(sig), ("invalid signal"));
2143 
2144 	sigfastblock_fetch(td);
2145 	PROC_LOCK(p);
2146 	ps = p->p_sigacts;
2147 	mtx_lock(&ps->ps_mtx);
2148 	sigmask = td->td_sigmask;
2149 	if (td->td_sigblock_val != 0)
2150 		SIGSETOR(sigmask, fastblock_mask);
2151 	if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2152 	    !SIGISMEMBER(sigmask, sig)) {
2153 #ifdef KTRACE
2154 		if (KTRPOINT(curthread, KTR_PSIG))
2155 			ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2156 			    &td->td_sigmask, ksi->ksi_code);
2157 #endif
2158 		(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2159 		    ksi, &td->td_sigmask);
2160 		postsig_done(sig, td, ps);
2161 		mtx_unlock(&ps->ps_mtx);
2162 	} else {
2163 		/*
2164 		 * Avoid a possible infinite loop if the thread
2165 		 * masking the signal or process is ignoring the
2166 		 * signal.
2167 		 */
2168 		if (kern_forcesigexit && (SIGISMEMBER(sigmask, sig) ||
2169 		    ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2170 			SIGDELSET(td->td_sigmask, sig);
2171 			SIGDELSET(ps->ps_sigcatch, sig);
2172 			SIGDELSET(ps->ps_sigignore, sig);
2173 			ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2174 			td->td_pflags &= ~TDP_SIGFASTBLOCK;
2175 			td->td_sigblock_val = 0;
2176 		}
2177 		mtx_unlock(&ps->ps_mtx);
2178 		p->p_sig = sig;		/* XXX to verify code */
2179 		tdsendsignal(p, td, sig, ksi);
2180 	}
2181 	PROC_UNLOCK(p);
2182 }
2183 
2184 static struct thread *
sigtd(struct proc * p,int sig,bool fast_sigblock)2185 sigtd(struct proc *p, int sig, bool fast_sigblock)
2186 {
2187 	struct thread *td, *signal_td;
2188 
2189 	PROC_LOCK_ASSERT(p, MA_OWNED);
2190 	MPASS(!fast_sigblock || p == curproc);
2191 
2192 	/*
2193 	 * Check if current thread can handle the signal without
2194 	 * switching context to another thread.
2195 	 */
2196 	if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig) &&
2197 	    (!fast_sigblock || curthread->td_sigblock_val == 0))
2198 		return (curthread);
2199 
2200 	/* Find a non-stopped thread that does not mask the signal. */
2201 	signal_td = NULL;
2202 	FOREACH_THREAD_IN_PROC(p, td) {
2203 		if (!SIGISMEMBER(td->td_sigmask, sig) && (!fast_sigblock ||
2204 		    td != curthread || td->td_sigblock_val == 0) &&
2205 		    (td->td_flags & TDF_BOUNDARY) == 0) {
2206 			signal_td = td;
2207 			break;
2208 		}
2209 	}
2210 	/* Select random (first) thread if no better match was found. */
2211 	if (signal_td == NULL)
2212 		signal_td = FIRST_THREAD_IN_PROC(p);
2213 	return (signal_td);
2214 }
2215 
2216 /*
2217  * Send the signal to the process.  If the signal has an action, the action
2218  * is usually performed by the target process rather than the caller; we add
2219  * the signal to the set of pending signals for the process.
2220  *
2221  * Exceptions:
2222  *   o When a stop signal is sent to a sleeping process that takes the
2223  *     default action, the process is stopped without awakening it.
2224  *   o SIGCONT restarts stopped processes (or puts them back to sleep)
2225  *     regardless of the signal action (eg, blocked or ignored).
2226  *
2227  * Other ignored signals are discarded immediately.
2228  *
2229  * NB: This function may be entered from the debugger via the "kill" DDB
2230  * command.  There is little that can be done to mitigate the possibly messy
2231  * side effects of this unwise possibility.
2232  */
2233 void
kern_psignal(struct proc * p,int sig)2234 kern_psignal(struct proc *p, int sig)
2235 {
2236 	ksiginfo_t ksi;
2237 
2238 	ksiginfo_init(&ksi);
2239 	ksi.ksi_signo = sig;
2240 	ksi.ksi_code = SI_KERNEL;
2241 	(void) tdsendsignal(p, NULL, sig, &ksi);
2242 }
2243 
2244 int
pksignal(struct proc * p,int sig,ksiginfo_t * ksi)2245 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2246 {
2247 
2248 	return (tdsendsignal(p, NULL, sig, ksi));
2249 }
2250 
2251 /* Utility function for finding a thread to send signal event to. */
2252 int
sigev_findtd(struct proc * p,struct sigevent * sigev,struct thread ** ttd)2253 sigev_findtd(struct proc *p, struct sigevent *sigev, struct thread **ttd)
2254 {
2255 	struct thread *td;
2256 
2257 	if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2258 		td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2259 		if (td == NULL)
2260 			return (ESRCH);
2261 		*ttd = td;
2262 	} else {
2263 		*ttd = NULL;
2264 		PROC_LOCK(p);
2265 	}
2266 	return (0);
2267 }
2268 
2269 void
tdsignal(struct thread * td,int sig)2270 tdsignal(struct thread *td, int sig)
2271 {
2272 	ksiginfo_t ksi;
2273 
2274 	ksiginfo_init(&ksi);
2275 	ksi.ksi_signo = sig;
2276 	ksi.ksi_code = SI_KERNEL;
2277 	(void) tdsendsignal(td->td_proc, td, sig, &ksi);
2278 }
2279 
2280 void
tdksignal(struct thread * td,int sig,ksiginfo_t * ksi)2281 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2282 {
2283 
2284 	(void) tdsendsignal(td->td_proc, td, sig, ksi);
2285 }
2286 
2287 static void
sig_sleepq_abort(struct thread * td,int intrval)2288 sig_sleepq_abort(struct thread *td, int intrval)
2289 {
2290 	THREAD_LOCK_ASSERT(td, MA_OWNED);
2291 
2292 	if (intrval == 0 && (td->td_flags & TDF_SIGWAIT) == 0)
2293 		thread_unlock(td);
2294 	else
2295 		sleepq_abort(td, intrval);
2296 }
2297 
2298 int
tdsendsignal(struct proc * p,struct thread * td,int sig,ksiginfo_t * ksi)2299 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2300 {
2301 	sig_t action;
2302 	sigqueue_t *sigqueue;
2303 	struct sigacts *ps;
2304 	int intrval, prop, ret;
2305 
2306 	MPASS(td == NULL || p == td->td_proc);
2307 	PROC_LOCK_ASSERT(p, MA_OWNED);
2308 
2309 	if (!_SIG_VALID(sig))
2310 		panic("%s(): invalid signal %d", __func__, sig);
2311 
2312 	KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2313 
2314 	/*
2315 	 * IEEE Std 1003.1-2001: return success when killing a zombie.
2316 	 */
2317 	if (p->p_state == PRS_ZOMBIE) {
2318 		if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2319 			ksiginfo_tryfree(ksi);
2320 		return (0);
2321 	}
2322 
2323 	ps = p->p_sigacts;
2324 	KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2325 	prop = sigprop(sig);
2326 
2327 	if (td == NULL) {
2328 		td = sigtd(p, sig, false);
2329 		sigqueue = &p->p_sigqueue;
2330 	} else
2331 		sigqueue = &td->td_sigqueue;
2332 
2333 	SDT_PROBE3(proc, , , signal__send, td, p, sig);
2334 
2335 	/*
2336 	 * If the signal is being ignored, then we forget about it
2337 	 * immediately, except when the target process executes
2338 	 * sigwait().  (Note: we don't set SIGCONT in ps_sigignore,
2339 	 * and if it is set to SIG_IGN, action will be SIG_DFL here.)
2340 	 */
2341 	mtx_lock(&ps->ps_mtx);
2342 	if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2343 		if (kern_sig_discard_ign &&
2344 		    (p->p_sysent->sv_flags & SV_SIG_DISCIGN) == 0) {
2345 			SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2346 
2347 			mtx_unlock(&ps->ps_mtx);
2348 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2349 				ksiginfo_tryfree(ksi);
2350 			return (0);
2351 		} else {
2352 			action = SIG_CATCH;
2353 			intrval = 0;
2354 		}
2355 	} else {
2356 		if (SIGISMEMBER(td->td_sigmask, sig))
2357 			action = SIG_HOLD;
2358 		else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2359 			action = SIG_CATCH;
2360 		else
2361 			action = SIG_DFL;
2362 		if (SIGISMEMBER(ps->ps_sigintr, sig))
2363 			intrval = EINTR;
2364 		else
2365 			intrval = ERESTART;
2366 	}
2367 	mtx_unlock(&ps->ps_mtx);
2368 
2369 	if (prop & SIGPROP_CONT)
2370 		sigqueue_delete_stopmask_proc(p);
2371 	else if (prop & SIGPROP_STOP) {
2372 		if (pt_attach_transparent &&
2373 		    (p->p_flag & P_TRACED) != 0 &&
2374 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0) {
2375 			PROC_SLOCK(p);
2376 			sig_handle_first_stop(NULL, p, sig);
2377 			PROC_SUNLOCK(p);
2378 			return (0);
2379 		}
2380 
2381 		/*
2382 		 * If sending a tty stop signal to a member of an orphaned
2383 		 * process group, discard the signal here if the action
2384 		 * is default; don't stop the process below if sleeping,
2385 		 * and don't clear any pending SIGCONT.
2386 		 */
2387 		if ((prop & SIGPROP_TTYSTOP) != 0 &&
2388 		    (p->p_pgrp->pg_flags & PGRP_ORPHANED) != 0 &&
2389 		    action == SIG_DFL) {
2390 			if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2391 				ksiginfo_tryfree(ksi);
2392 			return (0);
2393 		}
2394 		sigqueue_delete_proc(p, SIGCONT);
2395 		if (p->p_flag & P_CONTINUED) {
2396 			p->p_flag &= ~P_CONTINUED;
2397 			PROC_LOCK(p->p_pptr);
2398 			sigqueue_take(p->p_ksi);
2399 			PROC_UNLOCK(p->p_pptr);
2400 		}
2401 	}
2402 
2403 	ret = sigqueue_add(sigqueue, sig, ksi);
2404 	if (ret != 0)
2405 		return (ret);
2406 	signotify(td);
2407 	/*
2408 	 * Defer further processing for signals which are held,
2409 	 * except that stopped processes must be continued by SIGCONT.
2410 	 */
2411 	if (action == SIG_HOLD &&
2412 	    !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2413 		return (0);
2414 
2415 	/*
2416 	 * Some signals have a process-wide effect and a per-thread
2417 	 * component.  Most processing occurs when the process next
2418 	 * tries to cross the user boundary, however there are some
2419 	 * times when processing needs to be done immediately, such as
2420 	 * waking up threads so that they can cross the user boundary.
2421 	 * We try to do the per-process part here.
2422 	 */
2423 	if (P_SHOULDSTOP(p)) {
2424 		KASSERT(!(p->p_flag & P_WEXIT),
2425 		    ("signal to stopped but exiting process"));
2426 		if (sig == SIGKILL) {
2427 			/*
2428 			 * If traced process is already stopped,
2429 			 * then no further action is necessary.
2430 			 */
2431 			if (p->p_flag & P_TRACED)
2432 				return (0);
2433 			/*
2434 			 * SIGKILL sets process running.
2435 			 * It will die elsewhere.
2436 			 * All threads must be restarted.
2437 			 */
2438 			p->p_flag &= ~P_STOPPED_SIG;
2439 			goto runfast;
2440 		}
2441 
2442 		if (prop & SIGPROP_CONT) {
2443 			/*
2444 			 * If traced process is already stopped,
2445 			 * then no further action is necessary.
2446 			 */
2447 			if (p->p_flag & P_TRACED)
2448 				return (0);
2449 			/*
2450 			 * If SIGCONT is default (or ignored), we continue the
2451 			 * process but don't leave the signal in sigqueue as
2452 			 * it has no further action.  If SIGCONT is held, we
2453 			 * continue the process and leave the signal in
2454 			 * sigqueue.  If the process catches SIGCONT, let it
2455 			 * handle the signal itself.  If it isn't waiting on
2456 			 * an event, it goes back to run state.
2457 			 * Otherwise, process goes back to sleep state.
2458 			 */
2459 			p->p_flag &= ~P_STOPPED_SIG;
2460 			PROC_SLOCK(p);
2461 			if (p->p_numthreads == p->p_suspcount) {
2462 				PROC_SUNLOCK(p);
2463 				PROC_LOCK(p->p_pptr);
2464 				childproc_continued(p);
2465 				PROC_UNLOCK(p->p_pptr);
2466 				PROC_SLOCK(p);
2467 			}
2468 			if (action == SIG_DFL) {
2469 				thread_unsuspend(p);
2470 				PROC_SUNLOCK(p);
2471 				sigqueue_delete(sigqueue, sig);
2472 				goto out_cont;
2473 			}
2474 			if (action == SIG_CATCH) {
2475 				/*
2476 				 * The process wants to catch it so it needs
2477 				 * to run at least one thread, but which one?
2478 				 */
2479 				PROC_SUNLOCK(p);
2480 				goto runfast;
2481 			}
2482 			/*
2483 			 * The signal is not ignored or caught.
2484 			 */
2485 			thread_unsuspend(p);
2486 			PROC_SUNLOCK(p);
2487 			goto out_cont;
2488 		}
2489 
2490 		if (prop & SIGPROP_STOP) {
2491 			/*
2492 			 * If traced process is already stopped,
2493 			 * then no further action is necessary.
2494 			 */
2495 			if (p->p_flag & P_TRACED)
2496 				return (0);
2497 			/*
2498 			 * Already stopped, don't need to stop again
2499 			 * (If we did the shell could get confused).
2500 			 * Just make sure the signal STOP bit set.
2501 			 */
2502 			p->p_flag |= P_STOPPED_SIG;
2503 			sigqueue_delete(sigqueue, sig);
2504 			return (0);
2505 		}
2506 
2507 		/*
2508 		 * All other kinds of signals:
2509 		 * If a thread is sleeping interruptibly, simulate a
2510 		 * wakeup so that when it is continued it will be made
2511 		 * runnable and can look at the signal.  However, don't make
2512 		 * the PROCESS runnable, leave it stopped.
2513 		 * It may run a bit until it hits a thread_suspend_check().
2514 		 */
2515 		PROC_SLOCK(p);
2516 		thread_lock(td);
2517 		if (TD_CAN_ABORT(td))
2518 			sig_sleepq_abort(td, intrval);
2519 		else
2520 			thread_unlock(td);
2521 		PROC_SUNLOCK(p);
2522 		return (0);
2523 		/*
2524 		 * Mutexes are short lived. Threads waiting on them will
2525 		 * hit thread_suspend_check() soon.
2526 		 */
2527 	} else if (p->p_state == PRS_NORMAL) {
2528 		if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2529 			tdsigwakeup(td, sig, action, intrval);
2530 			return (0);
2531 		}
2532 
2533 		MPASS(action == SIG_DFL);
2534 
2535 		if (prop & SIGPROP_STOP) {
2536 			if (p->p_flag & (P_PPWAIT|P_WEXIT))
2537 				return (0);
2538 			p->p_flag |= P_STOPPED_SIG;
2539 			p->p_xsig = sig;
2540 			PROC_SLOCK(p);
2541 			sig_suspend_threads(td, p);
2542 			if (p->p_numthreads == p->p_suspcount) {
2543 				/*
2544 				 * only thread sending signal to another
2545 				 * process can reach here, if thread is sending
2546 				 * signal to its process, because thread does
2547 				 * not suspend itself here, p_numthreads
2548 				 * should never be equal to p_suspcount.
2549 				 */
2550 				thread_stopped(p);
2551 				PROC_SUNLOCK(p);
2552 				sigqueue_delete_proc(p, p->p_xsig);
2553 			} else
2554 				PROC_SUNLOCK(p);
2555 			return (0);
2556 		}
2557 	} else {
2558 		/* Not in "NORMAL" state. discard the signal. */
2559 		sigqueue_delete(sigqueue, sig);
2560 		return (0);
2561 	}
2562 
2563 	/*
2564 	 * The process is not stopped so we need to apply the signal to all the
2565 	 * running threads.
2566 	 */
2567 runfast:
2568 	tdsigwakeup(td, sig, action, intrval);
2569 	PROC_SLOCK(p);
2570 	thread_unsuspend(p);
2571 	PROC_SUNLOCK(p);
2572 out_cont:
2573 	itimer_proc_continue(p);
2574 	kqtimer_proc_continue(p);
2575 
2576 	return (0);
2577 }
2578 
2579 /*
2580  * The force of a signal has been directed against a single
2581  * thread.  We need to see what we can do about knocking it
2582  * out of any sleep it may be in etc.
2583  */
2584 static void
tdsigwakeup(struct thread * td,int sig,sig_t action,int intrval)2585 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2586 {
2587 	struct proc *p = td->td_proc;
2588 	int prop;
2589 
2590 	PROC_LOCK_ASSERT(p, MA_OWNED);
2591 	prop = sigprop(sig);
2592 
2593 	PROC_SLOCK(p);
2594 	thread_lock(td);
2595 	/*
2596 	 * Bring the priority of a thread up if we want it to get
2597 	 * killed in this lifetime.  Be careful to avoid bumping the
2598 	 * priority of the idle thread, since we still allow to signal
2599 	 * kernel processes.
2600 	 */
2601 	if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2602 	    td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2603 		sched_prio(td, PUSER);
2604 	if (TD_ON_SLEEPQ(td)) {
2605 		/*
2606 		 * If thread is sleeping uninterruptibly
2607 		 * we can't interrupt the sleep... the signal will
2608 		 * be noticed when the process returns through
2609 		 * trap() or syscall().
2610 		 */
2611 		if ((td->td_flags & TDF_SINTR) == 0)
2612 			goto out;
2613 		/*
2614 		 * If SIGCONT is default (or ignored) and process is
2615 		 * asleep, we are finished; the process should not
2616 		 * be awakened.
2617 		 */
2618 		if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2619 			thread_unlock(td);
2620 			PROC_SUNLOCK(p);
2621 			sigqueue_delete(&p->p_sigqueue, sig);
2622 			/*
2623 			 * It may be on either list in this state.
2624 			 * Remove from both for now.
2625 			 */
2626 			sigqueue_delete(&td->td_sigqueue, sig);
2627 			return;
2628 		}
2629 
2630 		/*
2631 		 * Don't awaken a sleeping thread for SIGSTOP if the
2632 		 * STOP signal is deferred.
2633 		 */
2634 		if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2635 		    TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2636 			goto out;
2637 
2638 		/*
2639 		 * Give low priority threads a better chance to run.
2640 		 */
2641 		if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2642 			sched_prio(td, PUSER);
2643 
2644 		sig_sleepq_abort(td, intrval);
2645 		PROC_SUNLOCK(p);
2646 		return;
2647 	}
2648 
2649 	/*
2650 	 * Other states do nothing with the signal immediately,
2651 	 * other than kicking ourselves if we are running.
2652 	 * It will either never be noticed, or noticed very soon.
2653 	 */
2654 #ifdef SMP
2655 	if (TD_IS_RUNNING(td) && td != curthread)
2656 		forward_signal(td);
2657 #endif
2658 
2659 out:
2660 	PROC_SUNLOCK(p);
2661 	thread_unlock(td);
2662 }
2663 
2664 static void
ptrace_coredumpreq(struct thread * td,struct proc * p,struct thr_coredump_req * tcq)2665 ptrace_coredumpreq(struct thread *td, struct proc *p,
2666     struct thr_coredump_req *tcq)
2667 {
2668 	void *rl_cookie;
2669 
2670 	if (p->p_sysent->sv_coredump == NULL) {
2671 		tcq->tc_error = ENOSYS;
2672 		return;
2673 	}
2674 
2675 	rl_cookie = vn_rangelock_wlock(tcq->tc_vp, 0, OFF_MAX);
2676 	tcq->tc_error = p->p_sysent->sv_coredump(td, tcq->tc_vp,
2677 	    tcq->tc_limit, tcq->tc_flags);
2678 	vn_rangelock_unlock(tcq->tc_vp, rl_cookie);
2679 }
2680 
2681 static void
ptrace_syscallreq(struct thread * td,struct proc * p,struct thr_syscall_req * tsr)2682 ptrace_syscallreq(struct thread *td, struct proc *p,
2683     struct thr_syscall_req *tsr)
2684 {
2685 	struct sysentvec *sv;
2686 	struct sysent *se;
2687 	register_t rv_saved[2];
2688 	int error, nerror;
2689 	int sc;
2690 	bool audited, sy_thr_static;
2691 
2692 	sv = p->p_sysent;
2693 	if (sv->sv_table == NULL || sv->sv_size < tsr->ts_sa.code) {
2694 		tsr->ts_ret.sr_error = ENOSYS;
2695 		return;
2696 	}
2697 
2698 	sc = tsr->ts_sa.code;
2699 	if (sc == SYS_syscall || sc == SYS___syscall) {
2700 		sc = tsr->ts_sa.args[0];
2701 		memmove(&tsr->ts_sa.args[0], &tsr->ts_sa.args[1],
2702 		    sizeof(register_t) * (tsr->ts_nargs - 1));
2703 	}
2704 
2705 	tsr->ts_sa.callp = se = &sv->sv_table[sc];
2706 
2707 	VM_CNT_INC(v_syscall);
2708 	td->td_pticks = 0;
2709 	if (__predict_false(td->td_cowgen != atomic_load_int(
2710 	    &td->td_proc->p_cowgen)))
2711 		thread_cow_update(td);
2712 
2713 	td->td_sa = tsr->ts_sa;
2714 
2715 #ifdef CAPABILITY_MODE
2716 	if ((se->sy_flags & SYF_CAPENABLED) == 0) {
2717 		if (CAP_TRACING(td))
2718 			ktrcapfail(CAPFAIL_SYSCALL, NULL);
2719 		if (IN_CAPABILITY_MODE(td)) {
2720 			tsr->ts_ret.sr_error = ECAPMODE;
2721 			return;
2722 		}
2723 	}
2724 #endif
2725 
2726 	sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2727 	audited = AUDIT_SYSCALL_ENTER(sc, td) != 0;
2728 
2729 	if (!sy_thr_static) {
2730 		error = syscall_thread_enter(td, &se);
2731 		sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2732 		if (error != 0) {
2733 			tsr->ts_ret.sr_error = error;
2734 			return;
2735 		}
2736 	}
2737 
2738 	rv_saved[0] = td->td_retval[0];
2739 	rv_saved[1] = td->td_retval[1];
2740 	nerror = td->td_errno;
2741 	td->td_retval[0] = 0;
2742 	td->td_retval[1] = 0;
2743 
2744 #ifdef KDTRACE_HOOKS
2745 	if (se->sy_entry != 0)
2746 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_ENTRY, 0);
2747 #endif
2748 	tsr->ts_ret.sr_error = se->sy_call(td, tsr->ts_sa.args);
2749 #ifdef KDTRACE_HOOKS
2750 	if (se->sy_return != 0)
2751 		(*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_RETURN,
2752 		    tsr->ts_ret.sr_error != 0 ? -1 : td->td_retval[0]);
2753 #endif
2754 
2755 	tsr->ts_ret.sr_retval[0] = td->td_retval[0];
2756 	tsr->ts_ret.sr_retval[1] = td->td_retval[1];
2757 	td->td_retval[0] = rv_saved[0];
2758 	td->td_retval[1] = rv_saved[1];
2759 	td->td_errno = nerror;
2760 
2761 	if (audited)
2762 		AUDIT_SYSCALL_EXIT(error, td);
2763 	if (!sy_thr_static)
2764 		syscall_thread_exit(td, se);
2765 }
2766 
2767 static void
ptrace_remotereq(struct thread * td,int flag)2768 ptrace_remotereq(struct thread *td, int flag)
2769 {
2770 	struct proc *p;
2771 
2772 	MPASS(td == curthread);
2773 	p = td->td_proc;
2774 	PROC_LOCK_ASSERT(p, MA_OWNED);
2775 	if ((td->td_dbgflags & flag) == 0)
2776 		return;
2777 	KASSERT((p->p_flag & P_STOPPED_TRACE) != 0, ("not stopped"));
2778 	KASSERT(td->td_remotereq != NULL, ("td_remotereq is NULL"));
2779 
2780 	PROC_UNLOCK(p);
2781 	switch (flag) {
2782 	case TDB_COREDUMPREQ:
2783 		ptrace_coredumpreq(td, p, td->td_remotereq);
2784 		break;
2785 	case TDB_SCREMOTEREQ:
2786 		ptrace_syscallreq(td, p, td->td_remotereq);
2787 		break;
2788 	default:
2789 		__unreachable();
2790 	}
2791 	PROC_LOCK(p);
2792 
2793 	MPASS((td->td_dbgflags & flag) != 0);
2794 	td->td_dbgflags &= ~flag;
2795 	td->td_remotereq = NULL;
2796 	wakeup(p);
2797 }
2798 
2799 /*
2800  * Suspend threads of the process p, either by directly setting the
2801  * inhibitor for the thread sleeping interruptibly, or by making the
2802  * thread suspend at the userspace boundary by scheduling a suspend AST.
2803  *
2804  * Returns true if some threads were suspended directly from the
2805  * sleeping state, and false if all threads are forced to process AST.
2806  */
2807 static bool
sig_suspend_threads(struct thread * td,struct proc * p)2808 sig_suspend_threads(struct thread *td, struct proc *p)
2809 {
2810 	struct thread *td2;
2811 	bool res;
2812 
2813 	PROC_LOCK_ASSERT(p, MA_OWNED);
2814 	PROC_SLOCK_ASSERT(p, MA_OWNED);
2815 
2816 	res = false;
2817 	FOREACH_THREAD_IN_PROC(p, td2) {
2818 		thread_lock(td2);
2819 		ast_sched_locked(td2, TDA_SUSPEND);
2820 		if (TD_IS_SLEEPING(td2) && (td2->td_flags & TDF_SINTR) != 0) {
2821 			if (td2->td_flags & TDF_SBDRY) {
2822 				/*
2823 				 * Once a thread is asleep with
2824 				 * TDF_SBDRY and without TDF_SERESTART
2825 				 * or TDF_SEINTR set, it should never
2826 				 * become suspended due to this check.
2827 				 */
2828 				KASSERT(!TD_IS_SUSPENDED(td2),
2829 				    ("thread with deferred stops suspended"));
2830 				if (TD_SBDRY_INTR(td2)) {
2831 					sleepq_abort(td2, TD_SBDRY_ERRNO(td2));
2832 					continue;
2833 				}
2834 			} else if (!TD_IS_SUSPENDED(td2)) {
2835 				thread_suspend_one(td2);
2836 				res = true;
2837 			}
2838 		} else if (!TD_IS_SUSPENDED(td2)) {
2839 #ifdef SMP
2840 			if (TD_IS_RUNNING(td2) && td2 != td)
2841 				forward_signal(td2);
2842 #endif
2843 		}
2844 		thread_unlock(td2);
2845 	}
2846 	return (res);
2847 }
2848 
2849 static void
sig_handle_first_stop(struct thread * td,struct proc * p,int sig)2850 sig_handle_first_stop(struct thread *td, struct proc *p, int sig)
2851 {
2852 	if (td != NULL && (td->td_dbgflags & TDB_FSTP) == 0 &&
2853 	    ((p->p_flag2 & P2_PTRACE_FSTP) != 0 || p->p_xthread != NULL))
2854 		return;
2855 
2856 	p->p_xsig = sig;
2857 	p->p_xthread = td;
2858 
2859 	/*
2860 	 * If we are on sleepqueue already, let sleepqueue
2861 	 * code decide if it needs to go sleep after attach.
2862 	 */
2863 	if (td != NULL && td->td_wchan == NULL)
2864 		td->td_dbgflags &= ~TDB_FSTP;
2865 
2866 	p->p_flag2 &= ~P2_PTRACE_FSTP;
2867 	p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2868 	if (sig_suspend_threads(td, p) && td == NULL)
2869 		thread_stopped(p);
2870 }
2871 
2872 /*
2873  * Stop the process for an event deemed interesting to the debugger. If si is
2874  * non-NULL, this is a signal exchange; the new signal requested by the
2875  * debugger will be returned for handling. If si is NULL, this is some other
2876  * type of interesting event. The debugger may request a signal be delivered in
2877  * that case as well, however it will be deferred until it can be handled.
2878  */
2879 int
ptracestop(struct thread * td,int sig,ksiginfo_t * si)2880 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2881 {
2882 	struct proc *p = td->td_proc;
2883 	struct thread *td2;
2884 	ksiginfo_t ksi;
2885 
2886 	PROC_LOCK_ASSERT(p, MA_OWNED);
2887 	KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2888 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2889 	    &p->p_mtx.lock_object, "Stopping for traced signal");
2890 
2891 	td->td_xsig = sig;
2892 
2893 	if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2894 		td->td_dbgflags |= TDB_XSIG;
2895 		CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2896 		    td->td_tid, p->p_pid, td->td_dbgflags, sig);
2897 		PROC_SLOCK(p);
2898 		while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2899 			if (P_KILLED(p)) {
2900 				/*
2901 				 * Ensure that, if we've been PT_KILLed, the
2902 				 * exit status reflects that. Another thread
2903 				 * may also be in ptracestop(), having just
2904 				 * received the SIGKILL, but this thread was
2905 				 * unsuspended first.
2906 				 */
2907 				td->td_dbgflags &= ~TDB_XSIG;
2908 				td->td_xsig = SIGKILL;
2909 				p->p_ptevents = 0;
2910 				break;
2911 			}
2912 			if (p->p_flag & P_SINGLE_EXIT &&
2913 			    !(td->td_dbgflags & TDB_EXIT)) {
2914 				/*
2915 				 * Ignore ptrace stops except for thread exit
2916 				 * events when the process exits.
2917 				 */
2918 				td->td_dbgflags &= ~TDB_XSIG;
2919 				PROC_SUNLOCK(p);
2920 				return (0);
2921 			}
2922 
2923 			/*
2924 			 * Make wait(2) work.  Ensure that right after the
2925 			 * attach, the thread which was decided to become the
2926 			 * leader of attach gets reported to the waiter.
2927 			 * Otherwise, just avoid overwriting another thread's
2928 			 * assignment to p_xthread.  If another thread has
2929 			 * already set p_xthread, the current thread will get
2930 			 * a chance to report itself upon the next iteration.
2931 			 */
2932 			sig_handle_first_stop(td, p, sig);
2933 
2934 			if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2935 				td->td_dbgflags &= ~TDB_STOPATFORK;
2936 			}
2937 stopme:
2938 			td->td_dbgflags |= TDB_SSWITCH;
2939 			thread_suspend_switch(td, p);
2940 			td->td_dbgflags &= ~TDB_SSWITCH;
2941 			if ((td->td_dbgflags & (TDB_COREDUMPREQ |
2942 			    TDB_SCREMOTEREQ)) != 0) {
2943 				MPASS((td->td_dbgflags & (TDB_COREDUMPREQ |
2944 				    TDB_SCREMOTEREQ)) !=
2945 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2946 				PROC_SUNLOCK(p);
2947 				ptrace_remotereq(td, td->td_dbgflags &
2948 				    (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2949 				PROC_SLOCK(p);
2950 				goto stopme;
2951 			}
2952 			if (p->p_xthread == td)
2953 				p->p_xthread = NULL;
2954 			if (!(p->p_flag & P_TRACED))
2955 				break;
2956 			if (td->td_dbgflags & TDB_SUSPEND) {
2957 				if (p->p_flag & P_SINGLE_EXIT)
2958 					break;
2959 				goto stopme;
2960 			}
2961 		}
2962 		PROC_SUNLOCK(p);
2963 	}
2964 
2965 	if (si != NULL && sig == td->td_xsig) {
2966 		/* Parent wants us to take the original signal unchanged. */
2967 		si->ksi_flags |= KSI_HEAD;
2968 		if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2969 			si->ksi_signo = 0;
2970 	} else if (td->td_xsig != 0) {
2971 		/*
2972 		 * If parent wants us to take a new signal, then it will leave
2973 		 * it in td->td_xsig; otherwise we just look for signals again.
2974 		 */
2975 		ksiginfo_init(&ksi);
2976 		ksi.ksi_signo = td->td_xsig;
2977 		ksi.ksi_flags |= KSI_PTRACE;
2978 		td2 = sigtd(p, td->td_xsig, false);
2979 		tdsendsignal(p, td2, td->td_xsig, &ksi);
2980 		if (td != td2)
2981 			return (0);
2982 	}
2983 
2984 	return (td->td_xsig);
2985 }
2986 
2987 static void
reschedule_signals(struct proc * p,sigset_t block,int flags)2988 reschedule_signals(struct proc *p, sigset_t block, int flags)
2989 {
2990 	struct sigacts *ps;
2991 	struct thread *td;
2992 	int sig;
2993 	bool fastblk, pslocked;
2994 
2995 	PROC_LOCK_ASSERT(p, MA_OWNED);
2996 	ps = p->p_sigacts;
2997 	pslocked = (flags & SIGPROCMASK_PS_LOCKED) != 0;
2998 	mtx_assert(&ps->ps_mtx, pslocked ? MA_OWNED : MA_NOTOWNED);
2999 	if (SIGISEMPTY(p->p_siglist))
3000 		return;
3001 	SIGSETAND(block, p->p_siglist);
3002 	fastblk = (flags & SIGPROCMASK_FASTBLK) != 0;
3003 	SIG_FOREACH(sig, &block) {
3004 		td = sigtd(p, sig, fastblk);
3005 
3006 		/*
3007 		 * If sigtd() selected us despite sigfastblock is
3008 		 * blocking, do not activate AST or wake us, to avoid
3009 		 * loop in AST handler.
3010 		 */
3011 		if (fastblk && td == curthread)
3012 			continue;
3013 
3014 		signotify(td);
3015 		if (!pslocked)
3016 			mtx_lock(&ps->ps_mtx);
3017 		if (p->p_flag & P_TRACED ||
3018 		    (SIGISMEMBER(ps->ps_sigcatch, sig) &&
3019 		    !SIGISMEMBER(td->td_sigmask, sig))) {
3020 			tdsigwakeup(td, sig, SIG_CATCH,
3021 			    (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
3022 			    ERESTART));
3023 		}
3024 		if (!pslocked)
3025 			mtx_unlock(&ps->ps_mtx);
3026 	}
3027 }
3028 
3029 void
tdsigcleanup(struct thread * td)3030 tdsigcleanup(struct thread *td)
3031 {
3032 	struct proc *p;
3033 	sigset_t unblocked;
3034 
3035 	p = td->td_proc;
3036 	PROC_LOCK_ASSERT(p, MA_OWNED);
3037 
3038 	sigqueue_flush(&td->td_sigqueue);
3039 	if (p->p_numthreads == 1)
3040 		return;
3041 
3042 	/*
3043 	 * Since we cannot handle signals, notify signal post code
3044 	 * about this by filling the sigmask.
3045 	 *
3046 	 * Also, if needed, wake up thread(s) that do not block the
3047 	 * same signals as the exiting thread, since the thread might
3048 	 * have been selected for delivery and woken up.
3049 	 */
3050 	SIGFILLSET(unblocked);
3051 	SIGSETNAND(unblocked, td->td_sigmask);
3052 	SIGFILLSET(td->td_sigmask);
3053 	reschedule_signals(p, unblocked, 0);
3054 
3055 }
3056 
3057 static int
sigdeferstop_curr_flags(int cflags)3058 sigdeferstop_curr_flags(int cflags)
3059 {
3060 
3061 	MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
3062 	    (cflags & TDF_SBDRY) != 0);
3063 	return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
3064 }
3065 
3066 /*
3067  * Defer the delivery of SIGSTOP for the current thread, according to
3068  * the requested mode.  Returns previous flags, which must be restored
3069  * by sigallowstop().
3070  *
3071  * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
3072  * cleared by the current thread, which allow the lock-less read-only
3073  * accesses below.
3074  */
3075 int
sigdeferstop_impl(int mode)3076 sigdeferstop_impl(int mode)
3077 {
3078 	struct thread *td;
3079 	int cflags, nflags;
3080 
3081 	td = curthread;
3082 	cflags = sigdeferstop_curr_flags(td->td_flags);
3083 	switch (mode) {
3084 	case SIGDEFERSTOP_NOP:
3085 		nflags = cflags;
3086 		break;
3087 	case SIGDEFERSTOP_OFF:
3088 		nflags = 0;
3089 		break;
3090 	case SIGDEFERSTOP_SILENT:
3091 		nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
3092 		break;
3093 	case SIGDEFERSTOP_EINTR:
3094 		nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
3095 		break;
3096 	case SIGDEFERSTOP_ERESTART:
3097 		nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
3098 		break;
3099 	default:
3100 		panic("sigdeferstop: invalid mode %x", mode);
3101 		break;
3102 	}
3103 	if (cflags == nflags)
3104 		return (SIGDEFERSTOP_VAL_NCHG);
3105 	thread_lock(td);
3106 	td->td_flags = (td->td_flags & ~cflags) | nflags;
3107 	thread_unlock(td);
3108 	return (cflags);
3109 }
3110 
3111 /*
3112  * Restores the STOP handling mode, typically permitting the delivery
3113  * of SIGSTOP for the current thread.  This does not immediately
3114  * suspend if a stop was posted.  Instead, the thread will suspend
3115  * either via ast() or a subsequent interruptible sleep.
3116  */
3117 void
sigallowstop_impl(int prev)3118 sigallowstop_impl(int prev)
3119 {
3120 	struct thread *td;
3121 	int cflags;
3122 
3123 	KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
3124 	KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
3125 	    ("sigallowstop: incorrect previous mode %x", prev));
3126 	td = curthread;
3127 	cflags = sigdeferstop_curr_flags(td->td_flags);
3128 	if (cflags != prev) {
3129 		thread_lock(td);
3130 		td->td_flags = (td->td_flags & ~cflags) | prev;
3131 		thread_unlock(td);
3132 	}
3133 }
3134 
3135 enum sigstatus {
3136 	SIGSTATUS_HANDLE,
3137 	SIGSTATUS_HANDLED,
3138 	SIGSTATUS_IGNORE,
3139 	SIGSTATUS_SBDRY_STOP,
3140 };
3141 
3142 /*
3143  * The thread has signal "sig" pending.  Figure out what to do with it:
3144  *
3145  * _HANDLE     -> the caller should handle the signal
3146  * _HANDLED    -> handled internally, reload pending signal set
3147  * _IGNORE     -> ignored, remove from the set of pending signals and try the
3148  *                next pending signal
3149  * _SBDRY_STOP -> the signal should stop the thread but this is not
3150  *                permitted in the current context
3151  */
3152 static enum sigstatus
sigprocess(struct thread * td,int sig)3153 sigprocess(struct thread *td, int sig)
3154 {
3155 	struct proc *p;
3156 	struct sigacts *ps;
3157 	struct sigqueue *queue;
3158 	ksiginfo_t ksi;
3159 	int prop;
3160 
3161 	KASSERT(_SIG_VALID(sig), ("%s: invalid signal %d", __func__, sig));
3162 
3163 	p = td->td_proc;
3164 	ps = p->p_sigacts;
3165 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3166 	PROC_LOCK_ASSERT(p, MA_OWNED);
3167 
3168 	/*
3169 	 * We should allow pending but ignored signals below
3170 	 * if there is sigwait() active, or P_TRACED was
3171 	 * on when they were posted.
3172 	 */
3173 	if (SIGISMEMBER(ps->ps_sigignore, sig) &&
3174 	    (p->p_flag & P_TRACED) == 0 &&
3175 	    (td->td_flags & TDF_SIGWAIT) == 0) {
3176 		return (SIGSTATUS_IGNORE);
3177 	}
3178 
3179 	/*
3180 	 * If the process is going to single-thread mode to prepare
3181 	 * for exit, there is no sense in delivering any signal
3182 	 * to usermode.  Another important consequence is that
3183 	 * msleep(..., PCATCH, ...) now is only interruptible by a
3184 	 * suspend request.
3185 	 */
3186 	if ((p->p_flag2 & P2_WEXIT) != 0)
3187 		return (SIGSTATUS_IGNORE);
3188 
3189 	if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
3190 		/*
3191 		 * If traced, always stop.
3192 		 * Remove old signal from queue before the stop.
3193 		 * XXX shrug off debugger, it causes siginfo to
3194 		 * be thrown away.
3195 		 */
3196 		queue = &td->td_sigqueue;
3197 		ksiginfo_init(&ksi);
3198 		if (sigqueue_get(queue, sig, &ksi) == 0) {
3199 			queue = &p->p_sigqueue;
3200 			sigqueue_get(queue, sig, &ksi);
3201 		}
3202 		td->td_si = ksi.ksi_info;
3203 
3204 		mtx_unlock(&ps->ps_mtx);
3205 		sig = ptracestop(td, sig, &ksi);
3206 		mtx_lock(&ps->ps_mtx);
3207 
3208 		td->td_si.si_signo = 0;
3209 
3210 		/*
3211 		 * Keep looking if the debugger discarded or
3212 		 * replaced the signal.
3213 		 */
3214 		if (sig == 0)
3215 			return (SIGSTATUS_HANDLED);
3216 
3217 		/*
3218 		 * If the signal became masked, re-queue it.
3219 		 */
3220 		if (SIGISMEMBER(td->td_sigmask, sig)) {
3221 			ksi.ksi_flags |= KSI_HEAD;
3222 			sigqueue_add(&p->p_sigqueue, sig, &ksi);
3223 			return (SIGSTATUS_HANDLED);
3224 		}
3225 
3226 		/*
3227 		 * If the traced bit got turned off, requeue the signal and
3228 		 * reload the set of pending signals.  This ensures that p_sig*
3229 		 * and p_sigact are consistent.
3230 		 */
3231 		if ((p->p_flag & P_TRACED) == 0) {
3232 			if ((ksi.ksi_flags & KSI_PTRACE) == 0) {
3233 				ksi.ksi_flags |= KSI_HEAD;
3234 				sigqueue_add(queue, sig, &ksi);
3235 			}
3236 			return (SIGSTATUS_HANDLED);
3237 		}
3238 	}
3239 
3240 	/*
3241 	 * Decide whether the signal should be returned.
3242 	 * Return the signal's number, or fall through
3243 	 * to clear it from the pending mask.
3244 	 */
3245 	switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
3246 	case (intptr_t)SIG_DFL:
3247 		/*
3248 		 * Don't take default actions on system processes.
3249 		 */
3250 		if (p->p_pid <= 1) {
3251 #ifdef DIAGNOSTIC
3252 			/*
3253 			 * Are you sure you want to ignore SIGSEGV
3254 			 * in init? XXX
3255 			 */
3256 			printf("Process (pid %lu) got signal %d\n",
3257 				(u_long)p->p_pid, sig);
3258 #endif
3259 			return (SIGSTATUS_IGNORE);
3260 		}
3261 
3262 		/*
3263 		 * If there is a pending stop signal to process with
3264 		 * default action, stop here, then clear the signal.
3265 		 * Traced or exiting processes should ignore stops.
3266 		 * Additionally, a member of an orphaned process group
3267 		 * should ignore tty stops.
3268 		 */
3269 		prop = sigprop(sig);
3270 		if (prop & SIGPROP_STOP) {
3271 			mtx_unlock(&ps->ps_mtx);
3272 			if ((p->p_flag & (P_TRACED | P_WEXIT |
3273 			    P_SINGLE_EXIT)) != 0 || ((p->p_pgrp->
3274 			    pg_flags & PGRP_ORPHANED) != 0 &&
3275 			    (prop & SIGPROP_TTYSTOP) != 0)) {
3276 				mtx_lock(&ps->ps_mtx);
3277 				return (SIGSTATUS_IGNORE);
3278 			}
3279 			if (TD_SBDRY_INTR(td)) {
3280 				KASSERT((td->td_flags & TDF_SBDRY) != 0,
3281 				    ("lost TDF_SBDRY"));
3282 				mtx_lock(&ps->ps_mtx);
3283 				return (SIGSTATUS_SBDRY_STOP);
3284 			}
3285 			WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
3286 			    &p->p_mtx.lock_object, "Catching SIGSTOP");
3287 			sigqueue_delete(&td->td_sigqueue, sig);
3288 			sigqueue_delete(&p->p_sigqueue, sig);
3289 			p->p_flag |= P_STOPPED_SIG;
3290 			p->p_xsig = sig;
3291 			PROC_SLOCK(p);
3292 			sig_suspend_threads(td, p);
3293 			thread_suspend_switch(td, p);
3294 			PROC_SUNLOCK(p);
3295 			mtx_lock(&ps->ps_mtx);
3296 			return (SIGSTATUS_HANDLED);
3297 		} else if ((prop & SIGPROP_IGNORE) != 0 &&
3298 		    (td->td_flags & TDF_SIGWAIT) == 0) {
3299 			/*
3300 			 * Default action is to ignore; drop it if
3301 			 * not in kern_sigtimedwait().
3302 			 */
3303 			return (SIGSTATUS_IGNORE);
3304 		} else {
3305 			return (SIGSTATUS_HANDLE);
3306 		}
3307 
3308 	case (intptr_t)SIG_IGN:
3309 		if ((td->td_flags & TDF_SIGWAIT) == 0)
3310 			return (SIGSTATUS_IGNORE);
3311 		else
3312 			return (SIGSTATUS_HANDLE);
3313 
3314 	default:
3315 		/*
3316 		 * This signal has an action, let postsig() process it.
3317 		 */
3318 		return (SIGSTATUS_HANDLE);
3319 	}
3320 }
3321 
3322 /*
3323  * If the current process has received a signal (should be caught or cause
3324  * termination, should interrupt current syscall), return the signal number.
3325  * Stop signals with default action are processed immediately, then cleared;
3326  * they aren't returned.  This is checked after each entry to the system for
3327  * a syscall or trap (though this can usually be done without calling
3328  * issignal by checking the pending signal masks in cursig.) The normal call
3329  * sequence is
3330  *
3331  *	while (sig = cursig(curthread))
3332  *		postsig(sig);
3333  */
3334 static int
issignal(struct thread * td)3335 issignal(struct thread *td)
3336 {
3337 	struct proc *p;
3338 	sigset_t sigpending;
3339 	int sig;
3340 
3341 	p = td->td_proc;
3342 	PROC_LOCK_ASSERT(p, MA_OWNED);
3343 
3344 	for (;;) {
3345 		sigpending = td->td_sigqueue.sq_signals;
3346 		SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
3347 		SIGSETNAND(sigpending, td->td_sigmask);
3348 
3349 		if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
3350 		    (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
3351 			SIG_STOPSIGMASK(sigpending);
3352 		if (SIGISEMPTY(sigpending))	/* no signal to send */
3353 			return (0);
3354 
3355 		/*
3356 		 * Do fast sigblock if requested by usermode.  Since
3357 		 * we do know that there was a signal pending at this
3358 		 * point, set the FAST_SIGBLOCK_PEND as indicator for
3359 		 * usermode to perform a dummy call to
3360 		 * FAST_SIGBLOCK_UNBLOCK, which causes immediate
3361 		 * delivery of postponed pending signal.
3362 		 */
3363 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
3364 			if (td->td_sigblock_val != 0)
3365 				SIGSETNAND(sigpending, fastblock_mask);
3366 			if (SIGISEMPTY(sigpending)) {
3367 				td->td_pflags |= TDP_SIGFASTPENDING;
3368 				return (0);
3369 			}
3370 		}
3371 
3372 		if (!pt_attach_transparent &&
3373 		    (p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
3374 		    (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
3375 		    SIGISMEMBER(sigpending, SIGSTOP)) {
3376 			/*
3377 			 * If debugger just attached, always consume
3378 			 * SIGSTOP from ptrace(PT_ATTACH) first, to
3379 			 * execute the debugger attach ritual in
3380 			 * order.
3381 			 */
3382 			td->td_dbgflags |= TDB_FSTP;
3383 			SIGEMPTYSET(sigpending);
3384 			SIGADDSET(sigpending, SIGSTOP);
3385 		}
3386 
3387 		SIG_FOREACH(sig, &sigpending) {
3388 			switch (sigprocess(td, sig)) {
3389 			case SIGSTATUS_HANDLE:
3390 				return (sig);
3391 			case SIGSTATUS_HANDLED:
3392 				goto next;
3393 			case SIGSTATUS_IGNORE:
3394 				sigqueue_delete(&td->td_sigqueue, sig);
3395 				sigqueue_delete(&p->p_sigqueue, sig);
3396 				break;
3397 			case SIGSTATUS_SBDRY_STOP:
3398 				return (-1);
3399 			}
3400 		}
3401 next:;
3402 	}
3403 }
3404 
3405 void
thread_stopped(struct proc * p)3406 thread_stopped(struct proc *p)
3407 {
3408 	int n;
3409 
3410 	PROC_LOCK_ASSERT(p, MA_OWNED);
3411 	PROC_SLOCK_ASSERT(p, MA_OWNED);
3412 	n = p->p_suspcount;
3413 	if (p == curproc)
3414 		n++;
3415 	if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3416 		PROC_SUNLOCK(p);
3417 		p->p_flag &= ~P_WAITED;
3418 		PROC_LOCK(p->p_pptr);
3419 		childproc_stopped(p, (p->p_flag & P_TRACED) ?
3420 			CLD_TRAPPED : CLD_STOPPED);
3421 		PROC_UNLOCK(p->p_pptr);
3422 		PROC_SLOCK(p);
3423 	}
3424 }
3425 
3426 /*
3427  * Take the action for the specified signal
3428  * from the current set of pending signals.
3429  */
3430 int
postsig(int sig)3431 postsig(int sig)
3432 {
3433 	struct thread *td;
3434 	struct proc *p;
3435 	struct sigacts *ps;
3436 	sig_t action;
3437 	ksiginfo_t ksi;
3438 	sigset_t returnmask;
3439 
3440 	KASSERT(sig != 0, ("postsig"));
3441 
3442 	td = curthread;
3443 	p = td->td_proc;
3444 	PROC_LOCK_ASSERT(p, MA_OWNED);
3445 	ps = p->p_sigacts;
3446 	mtx_assert(&ps->ps_mtx, MA_OWNED);
3447 	ksiginfo_init(&ksi);
3448 	if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3449 	    sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3450 		return (0);
3451 	ksi.ksi_signo = sig;
3452 	if (ksi.ksi_code == SI_TIMER)
3453 		itimer_accept(p, ksi.ksi_timerid, &ksi);
3454 	action = ps->ps_sigact[_SIG_IDX(sig)];
3455 #ifdef KTRACE
3456 	if (KTRPOINT(td, KTR_PSIG))
3457 		ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3458 		    &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3459 #endif
3460 
3461 	if (action == SIG_DFL) {
3462 		/*
3463 		 * Default action, where the default is to kill
3464 		 * the process.  (Other cases were ignored above.)
3465 		 */
3466 		mtx_unlock(&ps->ps_mtx);
3467 		proc_td_siginfo_capture(td, &ksi.ksi_info);
3468 		sigexit(td, sig);
3469 		/* NOTREACHED */
3470 	} else {
3471 		/*
3472 		 * If we get here, the signal must be caught.
3473 		 */
3474 		KASSERT(action != SIG_IGN, ("postsig action %p", action));
3475 		KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3476 		    ("postsig action: blocked sig %d", sig));
3477 
3478 		/*
3479 		 * Set the new mask value and also defer further
3480 		 * occurrences of this signal.
3481 		 *
3482 		 * Special case: user has done a sigsuspend.  Here the
3483 		 * current mask is not of interest, but rather the
3484 		 * mask from before the sigsuspend is what we want
3485 		 * restored after the signal processing is completed.
3486 		 */
3487 		if (td->td_pflags & TDP_OLDMASK) {
3488 			returnmask = td->td_oldsigmask;
3489 			td->td_pflags &= ~TDP_OLDMASK;
3490 		} else
3491 			returnmask = td->td_sigmask;
3492 
3493 		if (p->p_sig == sig) {
3494 			p->p_sig = 0;
3495 		}
3496 		(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3497 		postsig_done(sig, td, ps);
3498 	}
3499 	return (1);
3500 }
3501 
3502 int
sig_ast_checksusp(struct thread * td)3503 sig_ast_checksusp(struct thread *td)
3504 {
3505 	struct proc *p __diagused;
3506 	int ret;
3507 
3508 	p = td->td_proc;
3509 	PROC_LOCK_ASSERT(p, MA_OWNED);
3510 
3511 	if (!td_ast_pending(td, TDA_SUSPEND))
3512 		return (0);
3513 
3514 	ret = thread_suspend_check(1);
3515 	MPASS(ret == 0 || ret == EINTR || ret == ERESTART);
3516 	return (ret);
3517 }
3518 
3519 int
sig_ast_needsigchk(struct thread * td)3520 sig_ast_needsigchk(struct thread *td)
3521 {
3522 	struct proc *p;
3523 	struct sigacts *ps;
3524 	int ret, sig;
3525 
3526 	p = td->td_proc;
3527 	PROC_LOCK_ASSERT(p, MA_OWNED);
3528 
3529 	if (!td_ast_pending(td, TDA_SIG))
3530 		return (0);
3531 
3532 	ps = p->p_sigacts;
3533 	mtx_lock(&ps->ps_mtx);
3534 	sig = cursig(td);
3535 	if (sig == -1) {
3536 		mtx_unlock(&ps->ps_mtx);
3537 		KASSERT((td->td_flags & TDF_SBDRY) != 0, ("lost TDF_SBDRY"));
3538 		KASSERT(TD_SBDRY_INTR(td),
3539 		    ("lost TDF_SERESTART of TDF_SEINTR"));
3540 		KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
3541 		    (TDF_SEINTR | TDF_SERESTART),
3542 		    ("both TDF_SEINTR and TDF_SERESTART"));
3543 		ret = TD_SBDRY_ERRNO(td);
3544 	} else if (sig != 0) {
3545 		ret = SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR : ERESTART;
3546 		mtx_unlock(&ps->ps_mtx);
3547 	} else {
3548 		mtx_unlock(&ps->ps_mtx);
3549 		ret = 0;
3550 	}
3551 
3552 	/*
3553 	 * Do not go into sleep if this thread was the ptrace(2)
3554 	 * attach leader.  cursig() consumed SIGSTOP from PT_ATTACH,
3555 	 * but we usually act on the signal by interrupting sleep, and
3556 	 * should do that here as well.
3557 	 */
3558 	if ((td->td_dbgflags & TDB_FSTP) != 0) {
3559 		if (ret == 0)
3560 			ret = EINTR;
3561 		td->td_dbgflags &= ~TDB_FSTP;
3562 	}
3563 
3564 	return (ret);
3565 }
3566 
3567 int
sig_intr(void)3568 sig_intr(void)
3569 {
3570 	struct thread *td;
3571 	struct proc *p;
3572 	int ret;
3573 
3574 	td = curthread;
3575 	if (!td_ast_pending(td, TDA_SIG) && !td_ast_pending(td, TDA_SUSPEND))
3576 		return (0);
3577 
3578 	p = td->td_proc;
3579 
3580 	PROC_LOCK(p);
3581 	ret = sig_ast_checksusp(td);
3582 	if (ret == 0)
3583 		ret = sig_ast_needsigchk(td);
3584 	PROC_UNLOCK(p);
3585 	return (ret);
3586 }
3587 
3588 bool
curproc_sigkilled(void)3589 curproc_sigkilled(void)
3590 {
3591 	struct thread *td;
3592 	struct proc *p;
3593 	struct sigacts *ps;
3594 	bool res;
3595 
3596 	td = curthread;
3597 	if (!td_ast_pending(td, TDA_SIG))
3598 		return (false);
3599 
3600 	p = td->td_proc;
3601 	PROC_LOCK(p);
3602 	ps = p->p_sigacts;
3603 	mtx_lock(&ps->ps_mtx);
3604 	res = SIGISMEMBER(td->td_sigqueue.sq_signals, SIGKILL) ||
3605 	    SIGISMEMBER(p->p_sigqueue.sq_signals, SIGKILL);
3606 	mtx_unlock(&ps->ps_mtx);
3607 	PROC_UNLOCK(p);
3608 	return (res);
3609 }
3610 
3611 void
proc_wkilled(struct proc * p)3612 proc_wkilled(struct proc *p)
3613 {
3614 
3615 	PROC_LOCK_ASSERT(p, MA_OWNED);
3616 	if ((p->p_flag & P_WKILLED) == 0)
3617 		p->p_flag |= P_WKILLED;
3618 }
3619 
3620 /*
3621  * Kill the current process for stated reason.
3622  */
3623 void
killproc(struct proc * p,const char * why)3624 killproc(struct proc *p, const char *why)
3625 {
3626 
3627 	PROC_LOCK_ASSERT(p, MA_OWNED);
3628 	CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3629 	    p->p_comm);
3630 	log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3631 	    p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3632 	    p->p_ucred->cr_uid, why);
3633 	proc_wkilled(p);
3634 	kern_psignal(p, SIGKILL);
3635 }
3636 
3637 /*
3638  * Force the current process to exit with the specified signal, dumping core
3639  * if appropriate.  We bypass the normal tests for masked and caught signals,
3640  * allowing unrecoverable failures to terminate the process without changing
3641  * signal state.  Mark the accounting record with the signal termination.
3642  * If dumping core, save the signal number for the debugger.  Calls exit and
3643  * does not return.
3644  */
3645 void
sigexit(struct thread * td,int sig)3646 sigexit(struct thread *td, int sig)
3647 {
3648 	struct proc *p = td->td_proc;
3649 	const char *coreinfo;
3650 	int rv;
3651 	bool logexit;
3652 
3653 	PROC_LOCK_ASSERT(p, MA_OWNED);
3654 	proc_set_p2_wexit(p);
3655 
3656 	p->p_acflag |= AXSIG;
3657 	if ((p->p_flag2 & P2_LOGSIGEXIT_CTL) == 0)
3658 		logexit = kern_logsigexit != 0;
3659 	else
3660 		logexit = (p->p_flag2 & P2_LOGSIGEXIT_ENABLE) != 0;
3661 
3662 	/*
3663 	 * We must be single-threading to generate a core dump.  This
3664 	 * ensures that the registers in the core file are up-to-date.
3665 	 * Also, the ELF dump handler assumes that the thread list doesn't
3666 	 * change out from under it.
3667 	 *
3668 	 * XXX If another thread attempts to single-thread before us
3669 	 *     (e.g. via fork()), we won't get a dump at all.
3670 	 */
3671 	if ((sigprop(sig) & SIGPROP_CORE) &&
3672 	    thread_single(p, SINGLE_NO_EXIT) == 0) {
3673 		p->p_sig = sig;
3674 		/*
3675 		 * Log signals which would cause core dumps
3676 		 * (Log as LOG_INFO to appease those who don't want
3677 		 * these messages.)
3678 		 * XXX : Todo, as well as euid, write out ruid too
3679 		 * Note that coredump() drops proc lock.
3680 		 */
3681 		rv = coredump(td);
3682 		switch (rv) {
3683 		case 0:
3684 			sig |= WCOREFLAG;
3685 			coreinfo = " (core dumped)";
3686 			break;
3687 		case EFAULT:
3688 			coreinfo = " (no core dump - bad address)";
3689 			break;
3690 		case EINVAL:
3691 			coreinfo = " (no core dump - invalid argument)";
3692 			break;
3693 		case EFBIG:
3694 			coreinfo = " (no core dump - too large)";
3695 			break;
3696 		default:
3697 			coreinfo = " (no core dump - other error)";
3698 			break;
3699 		}
3700 		if (logexit)
3701 			log(LOG_INFO,
3702 			    "pid %d (%s), jid %d, uid %d: exited on "
3703 			    "signal %d%s\n", p->p_pid, p->p_comm,
3704 			    p->p_ucred->cr_prison->pr_id,
3705 			    td->td_ucred->cr_uid,
3706 			    sig &~ WCOREFLAG, coreinfo);
3707 	} else
3708 		PROC_UNLOCK(p);
3709 	exit1(td, 0, sig);
3710 	/* NOTREACHED */
3711 }
3712 
3713 /*
3714  * Send queued SIGCHLD to parent when child process's state
3715  * is changed.
3716  */
3717 static void
sigparent(struct proc * p,int reason,int status)3718 sigparent(struct proc *p, int reason, int status)
3719 {
3720 	PROC_LOCK_ASSERT(p, MA_OWNED);
3721 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3722 
3723 	if (p->p_ksi != NULL) {
3724 		p->p_ksi->ksi_signo  = SIGCHLD;
3725 		p->p_ksi->ksi_code   = reason;
3726 		p->p_ksi->ksi_status = status;
3727 		p->p_ksi->ksi_pid    = p->p_pid;
3728 		p->p_ksi->ksi_uid    = p->p_ucred->cr_ruid;
3729 		if (KSI_ONQ(p->p_ksi))
3730 			return;
3731 	}
3732 
3733 	/*
3734 	 * Do not consume p_ksi if parent is zombie, since signal is
3735 	 * dropped immediately.  Instead, keep it since it might be
3736 	 * useful for reaper.
3737 	 */
3738 	if (p->p_pptr->p_state != PRS_ZOMBIE)
3739 		pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3740 }
3741 
3742 static void
childproc_jobstate(struct proc * p,int reason,int sig)3743 childproc_jobstate(struct proc *p, int reason, int sig)
3744 {
3745 	struct sigacts *ps;
3746 
3747 	PROC_LOCK_ASSERT(p, MA_OWNED);
3748 	PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3749 
3750 	/*
3751 	 * Wake up parent sleeping in kern_wait(), also send
3752 	 * SIGCHLD to parent, but SIGCHLD does not guarantee
3753 	 * that parent will awake, because parent may masked
3754 	 * the signal.
3755 	 */
3756 	p->p_pptr->p_flag |= P_STATCHILD;
3757 	wakeup(p->p_pptr);
3758 
3759 	ps = p->p_pptr->p_sigacts;
3760 	mtx_lock(&ps->ps_mtx);
3761 	if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3762 		mtx_unlock(&ps->ps_mtx);
3763 		sigparent(p, reason, sig);
3764 	} else
3765 		mtx_unlock(&ps->ps_mtx);
3766 }
3767 
3768 void
childproc_stopped(struct proc * p,int reason)3769 childproc_stopped(struct proc *p, int reason)
3770 {
3771 
3772 	childproc_jobstate(p, reason, p->p_xsig);
3773 }
3774 
3775 void
childproc_continued(struct proc * p)3776 childproc_continued(struct proc *p)
3777 {
3778 	PROC_LOCK_ASSERT(p, MA_OWNED);
3779 	p->p_flag |= P_CONTINUED;
3780 	p->p_xsig = SIGCONT;
3781 	childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3782 }
3783 
3784 void
childproc_exited(struct proc * p)3785 childproc_exited(struct proc *p)
3786 {
3787 	int reason, status;
3788 
3789 	if (WCOREDUMP(p->p_xsig)) {
3790 		reason = CLD_DUMPED;
3791 		status = WTERMSIG(p->p_xsig);
3792 	} else if (WIFSIGNALED(p->p_xsig)) {
3793 		reason = CLD_KILLED;
3794 		status = WTERMSIG(p->p_xsig);
3795 	} else {
3796 		reason = CLD_EXITED;
3797 		status = p->p_xexit;
3798 	}
3799 	/*
3800 	 * XXX avoid calling wakeup(p->p_pptr), the work is
3801 	 * done in exit1().
3802 	 */
3803 	sigparent(p, reason, status);
3804 }
3805 
3806 #define	MAX_NUM_CORE_FILES 100000
3807 #ifndef NUM_CORE_FILES
3808 #define	NUM_CORE_FILES 5
3809 #endif
3810 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3811 static int num_cores = NUM_CORE_FILES;
3812 
3813 static int
sysctl_debug_num_cores_check(SYSCTL_HANDLER_ARGS)3814 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3815 {
3816 	int error;
3817 	int new_val;
3818 
3819 	new_val = num_cores;
3820 	error = sysctl_handle_int(oidp, &new_val, 0, req);
3821 	if (error != 0 || req->newptr == NULL)
3822 		return (error);
3823 	if (new_val > MAX_NUM_CORE_FILES)
3824 		new_val = MAX_NUM_CORE_FILES;
3825 	if (new_val < 0)
3826 		new_val = 0;
3827 	num_cores = new_val;
3828 	return (0);
3829 }
3830 SYSCTL_PROC(_debug, OID_AUTO, ncores,
3831     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
3832     sysctl_debug_num_cores_check, "I",
3833     "Maximum number of generated process corefiles while using index format");
3834 
3835 #define	GZIP_SUFFIX	".gz"
3836 #define	ZSTD_SUFFIX	".zst"
3837 
3838 int compress_user_cores = 0;
3839 
3840 static int
sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)3841 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3842 {
3843 	int error, val;
3844 
3845 	val = compress_user_cores;
3846 	error = sysctl_handle_int(oidp, &val, 0, req);
3847 	if (error != 0 || req->newptr == NULL)
3848 		return (error);
3849 	if (val != 0 && !compressor_avail(val))
3850 		return (EINVAL);
3851 	compress_user_cores = val;
3852 	return (error);
3853 }
3854 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores,
3855     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, 0, sizeof(int),
3856     sysctl_compress_user_cores, "I",
3857     "Enable compression of user corefiles ("
3858     __XSTRING(COMPRESS_GZIP) " = gzip, "
3859     __XSTRING(COMPRESS_ZSTD) " = zstd)");
3860 
3861 int compress_user_cores_level = 6;
3862 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3863     &compress_user_cores_level, 0,
3864     "Corefile compression level");
3865 
3866 /*
3867  * Protect the access to corefilename[] by allproc_lock.
3868  */
3869 #define	corefilename_lock	allproc_lock
3870 
3871 static char corefilename[MAXPATHLEN] = {"%N.core"};
3872 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3873 
3874 static int
sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)3875 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3876 {
3877 	int error;
3878 
3879 	sx_xlock(&corefilename_lock);
3880 	error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3881 	    req);
3882 	sx_xunlock(&corefilename_lock);
3883 
3884 	return (error);
3885 }
3886 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3887     CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3888     "Process corefile name format string");
3889 
3890 static void
vnode_close_locked(struct thread * td,struct vnode * vp)3891 vnode_close_locked(struct thread *td, struct vnode *vp)
3892 {
3893 
3894 	VOP_UNLOCK(vp);
3895 	vn_close(vp, FWRITE, td->td_ucred, td);
3896 }
3897 
3898 /*
3899  * If the core format has a %I in it, then we need to check
3900  * for existing corefiles before defining a name.
3901  * To do this we iterate over 0..ncores to find a
3902  * non-existing core file name to use. If all core files are
3903  * already used we choose the oldest one.
3904  */
3905 static int
corefile_open_last(struct thread * td,char * name,int indexpos,int indexlen,int ncores,struct vnode ** vpp)3906 corefile_open_last(struct thread *td, char *name, int indexpos,
3907     int indexlen, int ncores, struct vnode **vpp)
3908 {
3909 	struct vnode *oldvp, *nextvp, *vp;
3910 	struct vattr vattr;
3911 	struct nameidata nd;
3912 	int error, i, flags, oflags, cmode;
3913 	char ch;
3914 	struct timespec lasttime;
3915 
3916 	nextvp = oldvp = NULL;
3917 	cmode = S_IRUSR | S_IWUSR;
3918 	oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3919 	    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3920 
3921 	for (i = 0; i < ncores; i++) {
3922 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
3923 
3924 		ch = name[indexpos + indexlen];
3925 		(void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3926 		    i);
3927 		name[indexpos + indexlen] = ch;
3928 
3929 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
3930 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3931 		    NULL);
3932 		if (error != 0)
3933 			break;
3934 
3935 		vp = nd.ni_vp;
3936 		NDFREE_PNBUF(&nd);
3937 		if ((flags & O_CREAT) == O_CREAT) {
3938 			nextvp = vp;
3939 			break;
3940 		}
3941 
3942 		error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3943 		if (error != 0) {
3944 			vnode_close_locked(td, vp);
3945 			break;
3946 		}
3947 
3948 		if (oldvp == NULL ||
3949 		    lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3950 		    (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3951 		    lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3952 			if (oldvp != NULL)
3953 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3954 			oldvp = vp;
3955 			VOP_UNLOCK(oldvp);
3956 			lasttime = vattr.va_mtime;
3957 		} else {
3958 			vnode_close_locked(td, vp);
3959 		}
3960 	}
3961 
3962 	if (oldvp != NULL) {
3963 		if (nextvp == NULL) {
3964 			if ((td->td_proc->p_flag & P_SUGID) != 0) {
3965 				error = EFAULT;
3966 				vn_close(oldvp, FWRITE, td->td_ucred, td);
3967 			} else {
3968 				nextvp = oldvp;
3969 				error = vn_lock(nextvp, LK_EXCLUSIVE);
3970 				if (error != 0) {
3971 					vn_close(nextvp, FWRITE, td->td_ucred,
3972 					    td);
3973 					nextvp = NULL;
3974 				}
3975 			}
3976 		} else {
3977 			vn_close(oldvp, FWRITE, td->td_ucred, td);
3978 		}
3979 	}
3980 	if (error != 0) {
3981 		if (nextvp != NULL)
3982 			vnode_close_locked(td, oldvp);
3983 	} else {
3984 		*vpp = nextvp;
3985 	}
3986 
3987 	return (error);
3988 }
3989 
3990 /*
3991  * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3992  * Expand the name described in corefilename, using name, uid, and pid
3993  * and open/create core file.
3994  * corefilename is a printf-like string, with three format specifiers:
3995  *	%N	name of process ("name")
3996  *	%P	process id (pid)
3997  *	%U	user id (uid)
3998  * For example, "%N.core" is the default; they can be disabled completely
3999  * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
4000  * This is controlled by the sysctl variable kern.corefile (see above).
4001  */
4002 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)4003 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
4004     int compress, int signum, struct vnode **vpp, char **namep)
4005 {
4006 	struct sbuf sb;
4007 	struct nameidata nd;
4008 	const char *format;
4009 	char *hostname, *name;
4010 	int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
4011 
4012 	hostname = NULL;
4013 	format = corefilename;
4014 	name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
4015 	indexlen = 0;
4016 	indexpos = -1;
4017 	ncores = num_cores;
4018 	(void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
4019 	sx_slock(&corefilename_lock);
4020 	for (i = 0; format[i] != '\0'; i++) {
4021 		switch (format[i]) {
4022 		case '%':	/* Format character */
4023 			i++;
4024 			switch (format[i]) {
4025 			case '%':
4026 				sbuf_putc(&sb, '%');
4027 				break;
4028 			case 'H':	/* hostname */
4029 				if (hostname == NULL) {
4030 					hostname = malloc(MAXHOSTNAMELEN,
4031 					    M_TEMP, M_WAITOK);
4032 				}
4033 				getcredhostname(td->td_ucred, hostname,
4034 				    MAXHOSTNAMELEN);
4035 				sbuf_cat(&sb, hostname);
4036 				break;
4037 			case 'I':	/* autoincrementing index */
4038 				if (indexpos != -1) {
4039 					sbuf_printf(&sb, "%%I");
4040 					break;
4041 				}
4042 
4043 				indexpos = sbuf_len(&sb);
4044 				sbuf_printf(&sb, "%u", ncores - 1);
4045 				indexlen = sbuf_len(&sb) - indexpos;
4046 				break;
4047 			case 'N':	/* process name */
4048 				sbuf_printf(&sb, "%s", comm);
4049 				break;
4050 			case 'P':	/* process id */
4051 				sbuf_printf(&sb, "%u", pid);
4052 				break;
4053 			case 'S':	/* signal number */
4054 				sbuf_printf(&sb, "%i", signum);
4055 				break;
4056 			case 'U':	/* user id */
4057 				sbuf_printf(&sb, "%u", uid);
4058 				break;
4059 			default:
4060 				log(LOG_ERR,
4061 				    "Unknown format character %c in "
4062 				    "corename `%s'\n", format[i], format);
4063 				break;
4064 			}
4065 			break;
4066 		default:
4067 			sbuf_putc(&sb, format[i]);
4068 			break;
4069 		}
4070 	}
4071 	sx_sunlock(&corefilename_lock);
4072 	free(hostname, M_TEMP);
4073 	if (compress == COMPRESS_GZIP)
4074 		sbuf_cat(&sb, GZIP_SUFFIX);
4075 	else if (compress == COMPRESS_ZSTD)
4076 		sbuf_cat(&sb, ZSTD_SUFFIX);
4077 	if (sbuf_error(&sb) != 0) {
4078 		log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
4079 		    "long\n", (long)pid, comm, (u_long)uid);
4080 		sbuf_delete(&sb);
4081 		free(name, M_TEMP);
4082 		return (ENOMEM);
4083 	}
4084 	sbuf_finish(&sb);
4085 	sbuf_delete(&sb);
4086 
4087 	if (indexpos != -1) {
4088 		error = corefile_open_last(td, name, indexpos, indexlen, ncores,
4089 		    vpp);
4090 		if (error != 0) {
4091 			log(LOG_ERR,
4092 			    "pid %d (%s), uid (%u):  Path `%s' failed "
4093 			    "on initial open test, error = %d\n",
4094 			    pid, comm, uid, name, error);
4095 		}
4096 	} else {
4097 		cmode = S_IRUSR | S_IWUSR;
4098 		oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
4099 		    (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
4100 		flags = O_CREAT | FWRITE | O_NOFOLLOW;
4101 		if ((td->td_proc->p_flag & P_SUGID) != 0)
4102 			flags |= O_EXCL;
4103 
4104 		NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name);
4105 		error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
4106 		    NULL);
4107 		if (error == 0) {
4108 			*vpp = nd.ni_vp;
4109 			NDFREE_PNBUF(&nd);
4110 		}
4111 	}
4112 
4113 	if (error != 0) {
4114 #ifdef AUDIT
4115 		audit_proc_coredump(td, name, error);
4116 #endif
4117 		free(name, M_TEMP);
4118 		return (error);
4119 	}
4120 	*namep = name;
4121 	return (0);
4122 }
4123 
4124 /*
4125  * Dump a process' core.  The main routine does some
4126  * policy checking, and creates the name of the coredump;
4127  * then it passes on a vnode and a size limit to the process-specific
4128  * coredump routine if there is one; if there _is not_ one, it returns
4129  * ENOSYS; otherwise it returns the error from the process-specific routine.
4130  */
4131 
4132 static int
coredump(struct thread * td)4133 coredump(struct thread *td)
4134 {
4135 	struct proc *p = td->td_proc;
4136 	struct ucred *cred = td->td_ucred;
4137 	struct vnode *vp;
4138 	struct flock lf;
4139 	struct vattr vattr;
4140 	size_t fullpathsize;
4141 	int error, error1, jid, locked, ppid, sig;
4142 	char *name;			/* name of corefile */
4143 	void *rl_cookie;
4144 	off_t limit;
4145 	char *fullpath, *freepath = NULL;
4146 	struct sbuf *sb;
4147 
4148 	PROC_LOCK_ASSERT(p, MA_OWNED);
4149 	MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
4150 
4151 	if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
4152 	    (p->p_flag2 & P2_NOTRACE) != 0) {
4153 		PROC_UNLOCK(p);
4154 		return (EFAULT);
4155 	}
4156 
4157 	/*
4158 	 * Note that the bulk of limit checking is done after
4159 	 * the corefile is created.  The exception is if the limit
4160 	 * for corefiles is 0, in which case we don't bother
4161 	 * creating the corefile at all.  This layout means that
4162 	 * a corefile is truncated instead of not being created,
4163 	 * if it is larger than the limit.
4164 	 */
4165 	limit = (off_t)lim_cur(td, RLIMIT_CORE);
4166 	if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
4167 		PROC_UNLOCK(p);
4168 		return (EFBIG);
4169 	}
4170 
4171 	ppid = p->p_oppid;
4172 	sig = p->p_sig;
4173 	jid = p->p_ucred->cr_prison->pr_id;
4174 	PROC_UNLOCK(p);
4175 
4176 	error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
4177 	    compress_user_cores, p->p_sig, &vp, &name);
4178 	if (error != 0)
4179 		return (error);
4180 
4181 	/*
4182 	 * Don't dump to non-regular files or files with links.
4183 	 * Do not dump into system files. Effective user must own the corefile.
4184 	 */
4185 	if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
4186 	    vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
4187 	    vattr.va_uid != cred->cr_uid) {
4188 		VOP_UNLOCK(vp);
4189 		error = EFAULT;
4190 		goto out;
4191 	}
4192 
4193 	VOP_UNLOCK(vp);
4194 
4195 	/* Postpone other writers, including core dumps of other processes. */
4196 	rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
4197 
4198 	lf.l_whence = SEEK_SET;
4199 	lf.l_start = 0;
4200 	lf.l_len = 0;
4201 	lf.l_type = F_WRLCK;
4202 	locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
4203 
4204 	VATTR_NULL(&vattr);
4205 	vattr.va_size = 0;
4206 	if (set_core_nodump_flag)
4207 		vattr.va_flags = UF_NODUMP;
4208 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4209 	VOP_SETATTR(vp, &vattr, cred);
4210 	VOP_UNLOCK(vp);
4211 	PROC_LOCK(p);
4212 	p->p_acflag |= ACORE;
4213 	PROC_UNLOCK(p);
4214 
4215 	if (p->p_sysent->sv_coredump != NULL) {
4216 		error = p->p_sysent->sv_coredump(td, vp, limit, 0);
4217 	} else {
4218 		error = ENOSYS;
4219 	}
4220 
4221 	if (locked) {
4222 		lf.l_type = F_UNLCK;
4223 		VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
4224 	}
4225 	vn_rangelock_unlock(vp, rl_cookie);
4226 
4227 	/*
4228 	 * Notify the userland helper that a process triggered a core dump.
4229 	 * This allows the helper to run an automated debugging session.
4230 	 */
4231 	if (error != 0 || coredump_devctl == 0)
4232 		goto out;
4233 	sb = sbuf_new_auto();
4234 	if (vn_fullpath_global(p->p_textvp, &fullpath, &freepath) != 0)
4235 		goto out2;
4236 	sbuf_cat(sb, "comm=\"");
4237 	devctl_safe_quote_sb(sb, fullpath);
4238 	free(freepath, M_TEMP);
4239 	sbuf_cat(sb, "\" core=\"");
4240 
4241 	/*
4242 	 * We can't lookup core file vp directly. When we're replacing a core, and
4243 	 * other random times, we flush the name cache, so it will fail. Instead,
4244 	 * if the path of the core is relative, add the current dir in front if it.
4245 	 */
4246 	if (name[0] != '/') {
4247 		fullpathsize = MAXPATHLEN;
4248 		freepath = malloc(fullpathsize, M_TEMP, M_WAITOK);
4249 		if (vn_getcwd(freepath, &fullpath, &fullpathsize) != 0) {
4250 			free(freepath, M_TEMP);
4251 			goto out2;
4252 		}
4253 		devctl_safe_quote_sb(sb, fullpath);
4254 		free(freepath, M_TEMP);
4255 		sbuf_putc(sb, '/');
4256 	}
4257 	devctl_safe_quote_sb(sb, name);
4258 	sbuf_putc(sb, '"');
4259 
4260 	sbuf_printf(sb, " jid=%d pid=%d ppid=%d signo=%d",
4261 	    jid, p->p_pid, ppid, sig);
4262 	if (sbuf_finish(sb) == 0)
4263 		devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
4264 out2:
4265 	sbuf_delete(sb);
4266 out:
4267 	error1 = vn_close(vp, FWRITE, cred, td);
4268 	if (error == 0)
4269 		error = error1;
4270 #ifdef AUDIT
4271 	audit_proc_coredump(td, name, error);
4272 #endif
4273 	free(name, M_TEMP);
4274 	return (error);
4275 }
4276 
4277 /*
4278  * Nonexistent system call-- signal process (may want to handle it).  Flag
4279  * error in case process won't see signal immediately (blocked or ignored).
4280  */
4281 #ifndef _SYS_SYSPROTO_H_
4282 struct nosys_args {
4283 	int	dummy;
4284 };
4285 #endif
4286 /* ARGSUSED */
4287 int
nosys(struct thread * td,struct nosys_args * args)4288 nosys(struct thread *td, struct nosys_args *args)
4289 {
4290 	return (kern_nosys(td, args->dummy));
4291 }
4292 
4293 int
kern_nosys(struct thread * td,int dummy)4294 kern_nosys(struct thread *td, int dummy)
4295 {
4296 	struct proc *p;
4297 
4298 	p = td->td_proc;
4299 
4300 	if (SV_PROC_FLAG(p, SV_SIGSYS) != 0 && kern_signosys) {
4301 		PROC_LOCK(p);
4302 		tdsignal(td, SIGSYS);
4303 		PROC_UNLOCK(p);
4304 	}
4305 	if (kern_lognosys == 1 || kern_lognosys == 3) {
4306 		uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4307 		    td->td_sa.code);
4308 	}
4309 	if (kern_lognosys == 2 || kern_lognosys == 3 ||
4310 	    (p->p_pid == 1 && (kern_lognosys & 3) == 0)) {
4311 		printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4312 		    td->td_sa.code);
4313 	}
4314 	return (ENOSYS);
4315 }
4316 
4317 /*
4318  * Send a SIGIO or SIGURG signal to a process or process group using stored
4319  * credentials rather than those of the current process.
4320  */
4321 void
pgsigio(struct sigio ** sigiop,int sig,int checkctty)4322 pgsigio(struct sigio **sigiop, int sig, int checkctty)
4323 {
4324 	ksiginfo_t ksi;
4325 	struct sigio *sigio;
4326 
4327 	ksiginfo_init(&ksi);
4328 	ksi.ksi_signo = sig;
4329 	ksi.ksi_code = SI_KERNEL;
4330 
4331 	SIGIO_LOCK();
4332 	sigio = *sigiop;
4333 	if (sigio == NULL) {
4334 		SIGIO_UNLOCK();
4335 		return;
4336 	}
4337 	if (sigio->sio_pgid > 0) {
4338 		PROC_LOCK(sigio->sio_proc);
4339 		if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
4340 			kern_psignal(sigio->sio_proc, sig);
4341 		PROC_UNLOCK(sigio->sio_proc);
4342 	} else if (sigio->sio_pgid < 0) {
4343 		struct proc *p;
4344 
4345 		PGRP_LOCK(sigio->sio_pgrp);
4346 		LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
4347 			PROC_LOCK(p);
4348 			if (p->p_state == PRS_NORMAL &&
4349 			    CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
4350 			    (checkctty == 0 || (p->p_flag & P_CONTROLT)))
4351 				kern_psignal(p, sig);
4352 			PROC_UNLOCK(p);
4353 		}
4354 		PGRP_UNLOCK(sigio->sio_pgrp);
4355 	}
4356 	SIGIO_UNLOCK();
4357 }
4358 
4359 static int
filt_sigattach(struct knote * kn)4360 filt_sigattach(struct knote *kn)
4361 {
4362 	struct proc *p = curproc;
4363 
4364 	kn->kn_ptr.p_proc = p;
4365 	kn->kn_flags |= EV_CLEAR;		/* automatically set */
4366 
4367 	knlist_add(p->p_klist, kn, 0);
4368 
4369 	return (0);
4370 }
4371 
4372 static void
filt_sigdetach(struct knote * kn)4373 filt_sigdetach(struct knote *kn)
4374 {
4375 	knlist_remove(kn->kn_knlist, kn, 0);
4376 }
4377 
4378 /*
4379  * signal knotes are shared with proc knotes, so we apply a mask to
4380  * the hint in order to differentiate them from process hints.  This
4381  * could be avoided by using a signal-specific knote list, but probably
4382  * isn't worth the trouble.
4383  */
4384 static int
filt_signal(struct knote * kn,long hint)4385 filt_signal(struct knote *kn, long hint)
4386 {
4387 
4388 	if (hint & NOTE_SIGNAL) {
4389 		hint &= ~NOTE_SIGNAL;
4390 
4391 		if (kn->kn_id == hint)
4392 			kn->kn_data++;
4393 	}
4394 	return (kn->kn_data != 0);
4395 }
4396 
4397 struct sigacts *
sigacts_alloc(void)4398 sigacts_alloc(void)
4399 {
4400 	struct sigacts *ps;
4401 
4402 	ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
4403 	refcount_init(&ps->ps_refcnt, 1);
4404 	mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
4405 	return (ps);
4406 }
4407 
4408 void
sigacts_free(struct sigacts * ps)4409 sigacts_free(struct sigacts *ps)
4410 {
4411 
4412 	if (refcount_release(&ps->ps_refcnt) == 0)
4413 		return;
4414 	mtx_destroy(&ps->ps_mtx);
4415 	free(ps, M_SUBPROC);
4416 }
4417 
4418 struct sigacts *
sigacts_hold(struct sigacts * ps)4419 sigacts_hold(struct sigacts *ps)
4420 {
4421 
4422 	refcount_acquire(&ps->ps_refcnt);
4423 	return (ps);
4424 }
4425 
4426 void
sigacts_copy(struct sigacts * dest,struct sigacts * src)4427 sigacts_copy(struct sigacts *dest, struct sigacts *src)
4428 {
4429 
4430 	KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
4431 	mtx_lock(&src->ps_mtx);
4432 	bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
4433 	mtx_unlock(&src->ps_mtx);
4434 }
4435 
4436 int
sigacts_shared(struct sigacts * ps)4437 sigacts_shared(struct sigacts *ps)
4438 {
4439 
4440 	return (ps->ps_refcnt > 1);
4441 }
4442 
4443 void
sig_drop_caught(struct proc * p)4444 sig_drop_caught(struct proc *p)
4445 {
4446 	int sig;
4447 	struct sigacts *ps;
4448 
4449 	ps = p->p_sigacts;
4450 	PROC_LOCK_ASSERT(p, MA_OWNED);
4451 	mtx_assert(&ps->ps_mtx, MA_OWNED);
4452 	SIG_FOREACH(sig, &ps->ps_sigcatch) {
4453 		sigdflt(ps, sig);
4454 		if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
4455 			sigqueue_delete_proc(p, sig);
4456 	}
4457 }
4458 
4459 static void
sigfastblock_failed(struct thread * td,bool sendsig,bool write)4460 sigfastblock_failed(struct thread *td, bool sendsig, bool write)
4461 {
4462 	ksiginfo_t ksi;
4463 
4464 	/*
4465 	 * Prevent further fetches and SIGSEGVs, allowing thread to
4466 	 * issue syscalls despite corruption.
4467 	 */
4468 	sigfastblock_clear(td);
4469 
4470 	if (!sendsig)
4471 		return;
4472 	ksiginfo_init_trap(&ksi);
4473 	ksi.ksi_signo = SIGSEGV;
4474 	ksi.ksi_code = write ? SEGV_ACCERR : SEGV_MAPERR;
4475 	ksi.ksi_addr = td->td_sigblock_ptr;
4476 	trapsignal(td, &ksi);
4477 }
4478 
4479 static bool
sigfastblock_fetch_sig(struct thread * td,bool sendsig,uint32_t * valp)4480 sigfastblock_fetch_sig(struct thread *td, bool sendsig, uint32_t *valp)
4481 {
4482 	uint32_t res;
4483 
4484 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4485 		return (true);
4486 	if (fueword32((void *)td->td_sigblock_ptr, &res) == -1) {
4487 		sigfastblock_failed(td, sendsig, false);
4488 		return (false);
4489 	}
4490 	*valp = res;
4491 	td->td_sigblock_val = res & ~SIGFASTBLOCK_FLAGS;
4492 	return (true);
4493 }
4494 
4495 static void
sigfastblock_resched(struct thread * td,bool resched)4496 sigfastblock_resched(struct thread *td, bool resched)
4497 {
4498 	struct proc *p;
4499 
4500 	if (resched) {
4501 		p = td->td_proc;
4502 		PROC_LOCK(p);
4503 		reschedule_signals(p, td->td_sigmask, 0);
4504 		PROC_UNLOCK(p);
4505 	}
4506 	ast_sched(td, TDA_SIG);
4507 }
4508 
4509 int
sys_sigfastblock(struct thread * td,struct sigfastblock_args * uap)4510 sys_sigfastblock(struct thread *td, struct sigfastblock_args *uap)
4511 {
4512 	struct proc *p;
4513 	int error, res;
4514 	uint32_t oldval;
4515 
4516 	error = 0;
4517 	p = td->td_proc;
4518 	switch (uap->cmd) {
4519 	case SIGFASTBLOCK_SETPTR:
4520 		if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
4521 			error = EBUSY;
4522 			break;
4523 		}
4524 		if (((uintptr_t)(uap->ptr) & (sizeof(uint32_t) - 1)) != 0) {
4525 			error = EINVAL;
4526 			break;
4527 		}
4528 		td->td_pflags |= TDP_SIGFASTBLOCK;
4529 		td->td_sigblock_ptr = uap->ptr;
4530 		break;
4531 
4532 	case SIGFASTBLOCK_UNBLOCK:
4533 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4534 			error = EINVAL;
4535 			break;
4536 		}
4537 
4538 		for (;;) {
4539 			res = casueword32(td->td_sigblock_ptr,
4540 			    SIGFASTBLOCK_PEND, &oldval, 0);
4541 			if (res == -1) {
4542 				error = EFAULT;
4543 				sigfastblock_failed(td, false, true);
4544 				break;
4545 			}
4546 			if (res == 0)
4547 				break;
4548 			MPASS(res == 1);
4549 			if (oldval != SIGFASTBLOCK_PEND) {
4550 				error = EBUSY;
4551 				break;
4552 			}
4553 			error = thread_check_susp(td, false);
4554 			if (error != 0)
4555 				break;
4556 		}
4557 		if (error != 0)
4558 			break;
4559 
4560 		/*
4561 		 * td_sigblock_val is cleared there, but not on a
4562 		 * syscall exit.  The end effect is that a single
4563 		 * interruptible sleep, while user sigblock word is
4564 		 * set, might return EINTR or ERESTART to usermode
4565 		 * without delivering signal.  All further sleeps,
4566 		 * until userspace clears the word and does
4567 		 * sigfastblock(UNBLOCK), observe current word and no
4568 		 * longer get interrupted.  It is slight
4569 		 * non-conformance, with alternative to have read the
4570 		 * sigblock word on each syscall entry.
4571 		 */
4572 		td->td_sigblock_val = 0;
4573 
4574 		/*
4575 		 * Rely on normal ast mechanism to deliver pending
4576 		 * signals to current thread.  But notify others about
4577 		 * fake unblock.
4578 		 */
4579 		sigfastblock_resched(td, error == 0 && p->p_numthreads != 1);
4580 
4581 		break;
4582 
4583 	case SIGFASTBLOCK_UNSETPTR:
4584 		if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4585 			error = EINVAL;
4586 			break;
4587 		}
4588 		if (!sigfastblock_fetch_sig(td, false, &oldval)) {
4589 			error = EFAULT;
4590 			break;
4591 		}
4592 		if (oldval != 0 && oldval != SIGFASTBLOCK_PEND) {
4593 			error = EBUSY;
4594 			break;
4595 		}
4596 		sigfastblock_clear(td);
4597 		break;
4598 
4599 	default:
4600 		error = EINVAL;
4601 		break;
4602 	}
4603 	return (error);
4604 }
4605 
4606 void
sigfastblock_clear(struct thread * td)4607 sigfastblock_clear(struct thread *td)
4608 {
4609 	bool resched;
4610 
4611 	if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4612 		return;
4613 	td->td_sigblock_val = 0;
4614 	resched = (td->td_pflags & TDP_SIGFASTPENDING) != 0 ||
4615 	    SIGPENDING(td);
4616 	td->td_pflags &= ~(TDP_SIGFASTBLOCK | TDP_SIGFASTPENDING);
4617 	sigfastblock_resched(td, resched);
4618 }
4619 
4620 void
sigfastblock_fetch(struct thread * td)4621 sigfastblock_fetch(struct thread *td)
4622 {
4623 	uint32_t val;
4624 
4625 	(void)sigfastblock_fetch_sig(td, true, &val);
4626 }
4627 
4628 static void
sigfastblock_setpend1(struct thread * td)4629 sigfastblock_setpend1(struct thread *td)
4630 {
4631 	int res;
4632 	uint32_t oldval;
4633 
4634 	if ((td->td_pflags & TDP_SIGFASTPENDING) == 0)
4635 		return;
4636 	res = fueword32((void *)td->td_sigblock_ptr, &oldval);
4637 	if (res == -1) {
4638 		sigfastblock_failed(td, true, false);
4639 		return;
4640 	}
4641 	for (;;) {
4642 		res = casueword32(td->td_sigblock_ptr, oldval, &oldval,
4643 		    oldval | SIGFASTBLOCK_PEND);
4644 		if (res == -1) {
4645 			sigfastblock_failed(td, true, true);
4646 			return;
4647 		}
4648 		if (res == 0) {
4649 			td->td_sigblock_val = oldval & ~SIGFASTBLOCK_FLAGS;
4650 			td->td_pflags &= ~TDP_SIGFASTPENDING;
4651 			break;
4652 		}
4653 		MPASS(res == 1);
4654 		if (thread_check_susp(td, false) != 0)
4655 			break;
4656 	}
4657 }
4658 
4659 static void
sigfastblock_setpend(struct thread * td,bool resched)4660 sigfastblock_setpend(struct thread *td, bool resched)
4661 {
4662 	struct proc *p;
4663 
4664 	sigfastblock_setpend1(td);
4665 	if (resched) {
4666 		p = td->td_proc;
4667 		PROC_LOCK(p);
4668 		reschedule_signals(p, fastblock_mask, SIGPROCMASK_FASTBLK);
4669 		PROC_UNLOCK(p);
4670 	}
4671 }
4672