1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * @(#)kern_sig.c 8.7 (Berkeley) 4/18/94
37 */
38
39 #include <sys/cdefs.h>
40 #include "opt_capsicum.h"
41 #include "opt_ktrace.h"
42
43 #include <sys/param.h>
44 #include <sys/capsicum.h>
45 #include <sys/ctype.h>
46 #include <sys/systm.h>
47 #include <sys/signalvar.h>
48 #include <sys/vnode.h>
49 #include <sys/acct.h>
50 #include <sys/capsicum.h>
51 #include <sys/compressor.h>
52 #include <sys/condvar.h>
53 #include <sys/devctl.h>
54 #include <sys/event.h>
55 #include <sys/fcntl.h>
56 #include <sys/imgact.h>
57 #include <sys/kernel.h>
58 #include <sys/ktr.h>
59 #include <sys/ktrace.h>
60 #include <sys/limits.h>
61 #include <sys/lock.h>
62 #include <sys/malloc.h>
63 #include <sys/mutex.h>
64 #include <sys/refcount.h>
65 #include <sys/namei.h>
66 #include <sys/proc.h>
67 #include <sys/procdesc.h>
68 #include <sys/ptrace.h>
69 #include <sys/posix4.h>
70 #include <sys/racct.h>
71 #include <sys/resourcevar.h>
72 #include <sys/sdt.h>
73 #include <sys/sbuf.h>
74 #include <sys/sleepqueue.h>
75 #include <sys/smp.h>
76 #include <sys/stat.h>
77 #include <sys/sx.h>
78 #include <sys/syscall.h>
79 #include <sys/syscallsubr.h>
80 #include <sys/sysctl.h>
81 #include <sys/sysent.h>
82 #include <sys/syslog.h>
83 #include <sys/sysproto.h>
84 #include <sys/timers.h>
85 #include <sys/unistd.h>
86 #include <sys/vmmeter.h>
87 #include <sys/wait.h>
88 #include <vm/vm.h>
89 #include <vm/vm_extern.h>
90 #include <vm/uma.h>
91
92 #include <sys/jail.h>
93
94 #include <machine/cpu.h>
95
96 #include <security/audit/audit.h>
97
98 #define ONSIG 32 /* NSIG for osig* syscalls. XXX. */
99
100 SDT_PROVIDER_DECLARE(proc);
101 SDT_PROBE_DEFINE3(proc, , , signal__send,
102 "struct thread *", "struct proc *", "int");
103 SDT_PROBE_DEFINE2(proc, , , signal__clear,
104 "int", "ksiginfo_t *");
105 SDT_PROBE_DEFINE3(proc, , , signal__discard,
106 "struct thread *", "struct proc *", "int");
107
108 static int coredump(struct thread *);
109 static int killpg1(struct thread *td, int sig, int pgid, int all,
110 ksiginfo_t *ksi);
111 static int issignal(struct thread *td);
112 static void reschedule_signals(struct proc *p, sigset_t block, int flags);
113 static int sigprop(int sig);
114 static void tdsigwakeup(struct thread *, int, sig_t, int);
115 static int sig_suspend_threads(struct thread *, struct proc *, int);
116 static int filt_sigattach(struct knote *kn);
117 static void filt_sigdetach(struct knote *kn);
118 static int filt_signal(struct knote *kn, long hint);
119 static struct thread *sigtd(struct proc *p, int sig, bool fast_sigblock);
120 static void sigqueue_start(void);
121
122 static uma_zone_t ksiginfo_zone = NULL;
123 struct filterops sig_filtops = {
124 .f_isfd = 0,
125 .f_attach = filt_sigattach,
126 .f_detach = filt_sigdetach,
127 .f_event = filt_signal,
128 };
129
130 static int kern_logsigexit = 1;
131 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
132 &kern_logsigexit, 0,
133 "Log processes quitting on abnormal signals to syslog(3)");
134
135 static int kern_forcesigexit = 1;
136 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
137 &kern_forcesigexit, 0, "Force trap signal to be handled");
138
139 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
140 "POSIX real time signal");
141
142 static int max_pending_per_proc = 128;
143 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
144 &max_pending_per_proc, 0, "Max pending signals per proc");
145
146 static int preallocate_siginfo = 1024;
147 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
148 &preallocate_siginfo, 0, "Preallocated signal memory size");
149
150 static int signal_overflow = 0;
151 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
152 &signal_overflow, 0, "Number of signals overflew");
153
154 static int signal_alloc_fail = 0;
155 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
156 &signal_alloc_fail, 0, "signals failed to be allocated");
157
158 static int kern_lognosys = 0;
159 SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
160 "Log invalid syscalls");
161
162 static int kern_signosys = 1;
163 SYSCTL_INT(_kern, OID_AUTO, signosys, CTLFLAG_RWTUN, &kern_signosys, 0,
164 "Send SIGSYS on return from invalid syscall");
165
166 __read_frequently bool sigfastblock_fetch_always = false;
167 SYSCTL_BOOL(_kern, OID_AUTO, sigfastblock_fetch_always, CTLFLAG_RWTUN,
168 &sigfastblock_fetch_always, 0,
169 "Fetch sigfastblock word on each syscall entry for proper "
170 "blocking semantic");
171
172 static bool kern_sig_discard_ign = true;
173 SYSCTL_BOOL(_kern, OID_AUTO, sig_discard_ign, CTLFLAG_RWTUN,
174 &kern_sig_discard_ign, 0,
175 "Discard ignored signals on delivery, otherwise queue them to "
176 "the target queue");
177
178 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
179
180 /*
181 * Policy -- Can ucred cr1 send SIGIO to process cr2?
182 * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
183 * in the right situations.
184 */
185 #define CANSIGIO(cr1, cr2) \
186 ((cr1)->cr_uid == 0 || \
187 (cr1)->cr_ruid == (cr2)->cr_ruid || \
188 (cr1)->cr_uid == (cr2)->cr_ruid || \
189 (cr1)->cr_ruid == (cr2)->cr_uid || \
190 (cr1)->cr_uid == (cr2)->cr_uid)
191
192 static int sugid_coredump;
193 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
194 &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
195
196 static int capmode_coredump;
197 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
198 &capmode_coredump, 0, "Allow processes in capability mode to dump core");
199
200 static int do_coredump = 1;
201 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
202 &do_coredump, 0, "Enable/Disable coredumps");
203
204 static int set_core_nodump_flag = 0;
205 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
206 0, "Enable setting the NODUMP flag on coredump files");
207
208 static int coredump_devctl = 0;
209 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
210 0, "Generate a devctl notification when processes coredump");
211
212 /*
213 * Signal properties and actions.
214 * The array below categorizes the signals and their default actions
215 * according to the following properties:
216 */
217 #define SIGPROP_KILL 0x01 /* terminates process by default */
218 #define SIGPROP_CORE 0x02 /* ditto and coredumps */
219 #define SIGPROP_STOP 0x04 /* suspend process */
220 #define SIGPROP_TTYSTOP 0x08 /* ditto, from tty */
221 #define SIGPROP_IGNORE 0x10 /* ignore by default */
222 #define SIGPROP_CONT 0x20 /* continue if suspended */
223
224 static const int sigproptbl[NSIG] = {
225 [SIGHUP] = SIGPROP_KILL,
226 [SIGINT] = SIGPROP_KILL,
227 [SIGQUIT] = SIGPROP_KILL | SIGPROP_CORE,
228 [SIGILL] = SIGPROP_KILL | SIGPROP_CORE,
229 [SIGTRAP] = SIGPROP_KILL | SIGPROP_CORE,
230 [SIGABRT] = SIGPROP_KILL | SIGPROP_CORE,
231 [SIGEMT] = SIGPROP_KILL | SIGPROP_CORE,
232 [SIGFPE] = SIGPROP_KILL | SIGPROP_CORE,
233 [SIGKILL] = SIGPROP_KILL,
234 [SIGBUS] = SIGPROP_KILL | SIGPROP_CORE,
235 [SIGSEGV] = SIGPROP_KILL | SIGPROP_CORE,
236 [SIGSYS] = SIGPROP_KILL | SIGPROP_CORE,
237 [SIGPIPE] = SIGPROP_KILL,
238 [SIGALRM] = SIGPROP_KILL,
239 [SIGTERM] = SIGPROP_KILL,
240 [SIGURG] = SIGPROP_IGNORE,
241 [SIGSTOP] = SIGPROP_STOP,
242 [SIGTSTP] = SIGPROP_STOP | SIGPROP_TTYSTOP,
243 [SIGCONT] = SIGPROP_IGNORE | SIGPROP_CONT,
244 [SIGCHLD] = SIGPROP_IGNORE,
245 [SIGTTIN] = SIGPROP_STOP | SIGPROP_TTYSTOP,
246 [SIGTTOU] = SIGPROP_STOP | SIGPROP_TTYSTOP,
247 [SIGIO] = SIGPROP_IGNORE,
248 [SIGXCPU] = SIGPROP_KILL,
249 [SIGXFSZ] = SIGPROP_KILL,
250 [SIGVTALRM] = SIGPROP_KILL,
251 [SIGPROF] = SIGPROP_KILL,
252 [SIGWINCH] = SIGPROP_IGNORE,
253 [SIGINFO] = SIGPROP_IGNORE,
254 [SIGUSR1] = SIGPROP_KILL,
255 [SIGUSR2] = SIGPROP_KILL,
256 };
257
258 #define _SIG_FOREACH_ADVANCE(i, set) ({ \
259 int __found; \
260 for (;;) { \
261 if (__bits != 0) { \
262 int __sig = ffs(__bits); \
263 __bits &= ~(1u << (__sig - 1)); \
264 sig = __i * sizeof((set)->__bits[0]) * NBBY + __sig; \
265 __found = 1; \
266 break; \
267 } \
268 if (++__i == _SIG_WORDS) { \
269 __found = 0; \
270 break; \
271 } \
272 __bits = (set)->__bits[__i]; \
273 } \
274 __found != 0; \
275 })
276
277 #define SIG_FOREACH(i, set) \
278 for (int32_t __i = -1, __bits = 0; \
279 _SIG_FOREACH_ADVANCE(i, set); ) \
280
281 sigset_t fastblock_mask;
282
283 static void
sigqueue_start(void)284 sigqueue_start(void)
285 {
286 ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
287 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
288 uma_prealloc(ksiginfo_zone, preallocate_siginfo);
289 p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
290 p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
291 p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
292 SIGFILLSET(fastblock_mask);
293 SIG_CANTMASK(fastblock_mask);
294 }
295
296 ksiginfo_t *
ksiginfo_alloc(int mwait)297 ksiginfo_alloc(int mwait)
298 {
299 MPASS(mwait == M_WAITOK || mwait == M_NOWAIT);
300
301 if (ksiginfo_zone == NULL)
302 return (NULL);
303 return (uma_zalloc(ksiginfo_zone, mwait | M_ZERO));
304 }
305
306 void
ksiginfo_free(ksiginfo_t * ksi)307 ksiginfo_free(ksiginfo_t *ksi)
308 {
309 uma_zfree(ksiginfo_zone, ksi);
310 }
311
312 static __inline bool
ksiginfo_tryfree(ksiginfo_t * ksi)313 ksiginfo_tryfree(ksiginfo_t *ksi)
314 {
315 if ((ksi->ksi_flags & KSI_EXT) == 0) {
316 uma_zfree(ksiginfo_zone, ksi);
317 return (true);
318 }
319 return (false);
320 }
321
322 void
sigqueue_init(sigqueue_t * list,struct proc * p)323 sigqueue_init(sigqueue_t *list, struct proc *p)
324 {
325 SIGEMPTYSET(list->sq_signals);
326 SIGEMPTYSET(list->sq_kill);
327 SIGEMPTYSET(list->sq_ptrace);
328 TAILQ_INIT(&list->sq_list);
329 list->sq_proc = p;
330 list->sq_flags = SQ_INIT;
331 }
332
333 /*
334 * Get a signal's ksiginfo.
335 * Return:
336 * 0 - signal not found
337 * others - signal number
338 */
339 static int
sigqueue_get(sigqueue_t * sq,int signo,ksiginfo_t * si)340 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
341 {
342 struct proc *p = sq->sq_proc;
343 struct ksiginfo *ksi, *next;
344 int count = 0;
345
346 KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
347
348 if (!SIGISMEMBER(sq->sq_signals, signo))
349 return (0);
350
351 if (SIGISMEMBER(sq->sq_ptrace, signo)) {
352 count++;
353 SIGDELSET(sq->sq_ptrace, signo);
354 si->ksi_flags |= KSI_PTRACE;
355 }
356 if (SIGISMEMBER(sq->sq_kill, signo)) {
357 count++;
358 if (count == 1)
359 SIGDELSET(sq->sq_kill, signo);
360 }
361
362 TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
363 if (ksi->ksi_signo == signo) {
364 if (count == 0) {
365 TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
366 ksi->ksi_sigq = NULL;
367 ksiginfo_copy(ksi, si);
368 if (ksiginfo_tryfree(ksi) && p != NULL)
369 p->p_pendingcnt--;
370 }
371 if (++count > 1)
372 break;
373 }
374 }
375
376 if (count <= 1)
377 SIGDELSET(sq->sq_signals, signo);
378 si->ksi_signo = signo;
379 return (signo);
380 }
381
382 void
sigqueue_take(ksiginfo_t * ksi)383 sigqueue_take(ksiginfo_t *ksi)
384 {
385 struct ksiginfo *kp;
386 struct proc *p;
387 sigqueue_t *sq;
388
389 if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
390 return;
391
392 p = sq->sq_proc;
393 TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
394 ksi->ksi_sigq = NULL;
395 if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
396 p->p_pendingcnt--;
397
398 for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
399 kp = TAILQ_NEXT(kp, ksi_link)) {
400 if (kp->ksi_signo == ksi->ksi_signo)
401 break;
402 }
403 if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
404 !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
405 SIGDELSET(sq->sq_signals, ksi->ksi_signo);
406 }
407
408 static int
sigqueue_add(sigqueue_t * sq,int signo,ksiginfo_t * si)409 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
410 {
411 struct proc *p = sq->sq_proc;
412 struct ksiginfo *ksi;
413 int ret = 0;
414
415 KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
416
417 /*
418 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
419 * for these signals.
420 */
421 if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
422 SIGADDSET(sq->sq_kill, signo);
423 goto out_set_bit;
424 }
425
426 /* directly insert the ksi, don't copy it */
427 if (si->ksi_flags & KSI_INS) {
428 if (si->ksi_flags & KSI_HEAD)
429 TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
430 else
431 TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
432 si->ksi_sigq = sq;
433 goto out_set_bit;
434 }
435
436 if (__predict_false(ksiginfo_zone == NULL)) {
437 SIGADDSET(sq->sq_kill, signo);
438 goto out_set_bit;
439 }
440
441 if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
442 signal_overflow++;
443 ret = EAGAIN;
444 } else if ((ksi = ksiginfo_alloc(M_NOWAIT)) == NULL) {
445 signal_alloc_fail++;
446 ret = EAGAIN;
447 } else {
448 if (p != NULL)
449 p->p_pendingcnt++;
450 ksiginfo_copy(si, ksi);
451 ksi->ksi_signo = signo;
452 if (si->ksi_flags & KSI_HEAD)
453 TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
454 else
455 TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
456 ksi->ksi_sigq = sq;
457 }
458
459 if (ret != 0) {
460 if ((si->ksi_flags & KSI_PTRACE) != 0) {
461 SIGADDSET(sq->sq_ptrace, signo);
462 ret = 0;
463 goto out_set_bit;
464 } else if ((si->ksi_flags & KSI_TRAP) != 0 ||
465 (si->ksi_flags & KSI_SIGQ) == 0) {
466 SIGADDSET(sq->sq_kill, signo);
467 ret = 0;
468 goto out_set_bit;
469 }
470 return (ret);
471 }
472
473 out_set_bit:
474 SIGADDSET(sq->sq_signals, signo);
475 return (ret);
476 }
477
478 void
sigqueue_flush(sigqueue_t * sq)479 sigqueue_flush(sigqueue_t *sq)
480 {
481 struct proc *p = sq->sq_proc;
482 ksiginfo_t *ksi;
483
484 KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
485
486 if (p != NULL)
487 PROC_LOCK_ASSERT(p, MA_OWNED);
488
489 while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
490 TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
491 ksi->ksi_sigq = NULL;
492 if (ksiginfo_tryfree(ksi) && p != NULL)
493 p->p_pendingcnt--;
494 }
495
496 SIGEMPTYSET(sq->sq_signals);
497 SIGEMPTYSET(sq->sq_kill);
498 SIGEMPTYSET(sq->sq_ptrace);
499 }
500
501 static void
sigqueue_move_set(sigqueue_t * src,sigqueue_t * dst,const sigset_t * set)502 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
503 {
504 sigset_t tmp;
505 struct proc *p1, *p2;
506 ksiginfo_t *ksi, *next;
507
508 KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
509 KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
510 p1 = src->sq_proc;
511 p2 = dst->sq_proc;
512 /* Move siginfo to target list */
513 TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
514 if (SIGISMEMBER(*set, ksi->ksi_signo)) {
515 TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
516 if (p1 != NULL)
517 p1->p_pendingcnt--;
518 TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
519 ksi->ksi_sigq = dst;
520 if (p2 != NULL)
521 p2->p_pendingcnt++;
522 }
523 }
524
525 /* Move pending bits to target list */
526 tmp = src->sq_kill;
527 SIGSETAND(tmp, *set);
528 SIGSETOR(dst->sq_kill, tmp);
529 SIGSETNAND(src->sq_kill, tmp);
530
531 tmp = src->sq_ptrace;
532 SIGSETAND(tmp, *set);
533 SIGSETOR(dst->sq_ptrace, tmp);
534 SIGSETNAND(src->sq_ptrace, tmp);
535
536 tmp = src->sq_signals;
537 SIGSETAND(tmp, *set);
538 SIGSETOR(dst->sq_signals, tmp);
539 SIGSETNAND(src->sq_signals, tmp);
540 }
541
542 #if 0
543 static void
544 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
545 {
546 sigset_t set;
547
548 SIGEMPTYSET(set);
549 SIGADDSET(set, signo);
550 sigqueue_move_set(src, dst, &set);
551 }
552 #endif
553
554 static void
sigqueue_delete_set(sigqueue_t * sq,const sigset_t * set)555 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
556 {
557 struct proc *p = sq->sq_proc;
558 ksiginfo_t *ksi, *next;
559
560 KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
561
562 /* Remove siginfo queue */
563 TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
564 if (SIGISMEMBER(*set, ksi->ksi_signo)) {
565 TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
566 ksi->ksi_sigq = NULL;
567 if (ksiginfo_tryfree(ksi) && p != NULL)
568 p->p_pendingcnt--;
569 }
570 }
571 SIGSETNAND(sq->sq_kill, *set);
572 SIGSETNAND(sq->sq_ptrace, *set);
573 SIGSETNAND(sq->sq_signals, *set);
574 }
575
576 void
sigqueue_delete(sigqueue_t * sq,int signo)577 sigqueue_delete(sigqueue_t *sq, int signo)
578 {
579 sigset_t set;
580
581 SIGEMPTYSET(set);
582 SIGADDSET(set, signo);
583 sigqueue_delete_set(sq, &set);
584 }
585
586 /* Remove a set of signals for a process */
587 static void
sigqueue_delete_set_proc(struct proc * p,const sigset_t * set)588 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
589 {
590 sigqueue_t worklist;
591 struct thread *td0;
592
593 PROC_LOCK_ASSERT(p, MA_OWNED);
594
595 sigqueue_init(&worklist, NULL);
596 sigqueue_move_set(&p->p_sigqueue, &worklist, set);
597
598 FOREACH_THREAD_IN_PROC(p, td0)
599 sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
600
601 sigqueue_flush(&worklist);
602 }
603
604 void
sigqueue_delete_proc(struct proc * p,int signo)605 sigqueue_delete_proc(struct proc *p, int signo)
606 {
607 sigset_t set;
608
609 SIGEMPTYSET(set);
610 SIGADDSET(set, signo);
611 sigqueue_delete_set_proc(p, &set);
612 }
613
614 static void
sigqueue_delete_stopmask_proc(struct proc * p)615 sigqueue_delete_stopmask_proc(struct proc *p)
616 {
617 sigset_t set;
618
619 SIGEMPTYSET(set);
620 SIGADDSET(set, SIGSTOP);
621 SIGADDSET(set, SIGTSTP);
622 SIGADDSET(set, SIGTTIN);
623 SIGADDSET(set, SIGTTOU);
624 sigqueue_delete_set_proc(p, &set);
625 }
626
627 /*
628 * Determine signal that should be delivered to thread td, the current
629 * thread, 0 if none. If there is a pending stop signal with default
630 * action, the process stops in issignal().
631 */
632 int
cursig(struct thread * td)633 cursig(struct thread *td)
634 {
635 PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
636 mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
637 THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
638 return (SIGPENDING(td) ? issignal(td) : 0);
639 }
640
641 /*
642 * Arrange for ast() to handle unmasked pending signals on return to user
643 * mode. This must be called whenever a signal is added to td_sigqueue or
644 * unmasked in td_sigmask.
645 */
646 void
signotify(struct thread * td)647 signotify(struct thread *td)
648 {
649
650 PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
651
652 if (SIGPENDING(td)) {
653 thread_lock(td);
654 td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING;
655 thread_unlock(td);
656 }
657 }
658
659 /*
660 * Returns 1 (true) if altstack is configured for the thread, and the
661 * passed stack bottom address falls into the altstack range. Handles
662 * the 43 compat special case where the alt stack size is zero.
663 */
664 int
sigonstack(size_t sp)665 sigonstack(size_t sp)
666 {
667 struct thread *td;
668
669 td = curthread;
670 if ((td->td_pflags & TDP_ALTSTACK) == 0)
671 return (0);
672 #if defined(COMPAT_43)
673 if (SV_PROC_FLAG(td->td_proc, SV_AOUT) && td->td_sigstk.ss_size == 0)
674 return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
675 #endif
676 return (sp >= (size_t)td->td_sigstk.ss_sp &&
677 sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
678 }
679
680 static __inline int
sigprop(int sig)681 sigprop(int sig)
682 {
683
684 if (sig > 0 && sig < nitems(sigproptbl))
685 return (sigproptbl[sig]);
686 return (0);
687 }
688
689 static bool
sigact_flag_test(const struct sigaction * act,int flag)690 sigact_flag_test(const struct sigaction *act, int flag)
691 {
692
693 /*
694 * SA_SIGINFO is reset when signal disposition is set to
695 * ignore or default. Other flags are kept according to user
696 * settings.
697 */
698 return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
699 ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
700 (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
701 }
702
703 /*
704 * kern_sigaction
705 * sigaction
706 * freebsd4_sigaction
707 * osigaction
708 */
709 int
kern_sigaction(struct thread * td,int sig,const struct sigaction * act,struct sigaction * oact,int flags)710 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
711 struct sigaction *oact, int flags)
712 {
713 struct sigacts *ps;
714 struct proc *p = td->td_proc;
715
716 if (!_SIG_VALID(sig))
717 return (EINVAL);
718 if (act != NULL && act->sa_handler != SIG_DFL &&
719 act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
720 SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
721 SA_NOCLDWAIT | SA_SIGINFO)) != 0)
722 return (EINVAL);
723
724 PROC_LOCK(p);
725 ps = p->p_sigacts;
726 mtx_lock(&ps->ps_mtx);
727 if (oact) {
728 memset(oact, 0, sizeof(*oact));
729 oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
730 if (SIGISMEMBER(ps->ps_sigonstack, sig))
731 oact->sa_flags |= SA_ONSTACK;
732 if (!SIGISMEMBER(ps->ps_sigintr, sig))
733 oact->sa_flags |= SA_RESTART;
734 if (SIGISMEMBER(ps->ps_sigreset, sig))
735 oact->sa_flags |= SA_RESETHAND;
736 if (SIGISMEMBER(ps->ps_signodefer, sig))
737 oact->sa_flags |= SA_NODEFER;
738 if (SIGISMEMBER(ps->ps_siginfo, sig)) {
739 oact->sa_flags |= SA_SIGINFO;
740 oact->sa_sigaction =
741 (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
742 } else
743 oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
744 if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
745 oact->sa_flags |= SA_NOCLDSTOP;
746 if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
747 oact->sa_flags |= SA_NOCLDWAIT;
748 }
749 if (act) {
750 if ((sig == SIGKILL || sig == SIGSTOP) &&
751 act->sa_handler != SIG_DFL) {
752 mtx_unlock(&ps->ps_mtx);
753 PROC_UNLOCK(p);
754 return (EINVAL);
755 }
756
757 /*
758 * Change setting atomically.
759 */
760
761 ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
762 SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
763 if (sigact_flag_test(act, SA_SIGINFO)) {
764 ps->ps_sigact[_SIG_IDX(sig)] =
765 (__sighandler_t *)act->sa_sigaction;
766 SIGADDSET(ps->ps_siginfo, sig);
767 } else {
768 ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
769 SIGDELSET(ps->ps_siginfo, sig);
770 }
771 if (!sigact_flag_test(act, SA_RESTART))
772 SIGADDSET(ps->ps_sigintr, sig);
773 else
774 SIGDELSET(ps->ps_sigintr, sig);
775 if (sigact_flag_test(act, SA_ONSTACK))
776 SIGADDSET(ps->ps_sigonstack, sig);
777 else
778 SIGDELSET(ps->ps_sigonstack, sig);
779 if (sigact_flag_test(act, SA_RESETHAND))
780 SIGADDSET(ps->ps_sigreset, sig);
781 else
782 SIGDELSET(ps->ps_sigreset, sig);
783 if (sigact_flag_test(act, SA_NODEFER))
784 SIGADDSET(ps->ps_signodefer, sig);
785 else
786 SIGDELSET(ps->ps_signodefer, sig);
787 if (sig == SIGCHLD) {
788 if (act->sa_flags & SA_NOCLDSTOP)
789 ps->ps_flag |= PS_NOCLDSTOP;
790 else
791 ps->ps_flag &= ~PS_NOCLDSTOP;
792 if (act->sa_flags & SA_NOCLDWAIT) {
793 /*
794 * Paranoia: since SA_NOCLDWAIT is implemented
795 * by reparenting the dying child to PID 1 (and
796 * trust it to reap the zombie), PID 1 itself
797 * is forbidden to set SA_NOCLDWAIT.
798 */
799 if (p->p_pid == 1)
800 ps->ps_flag &= ~PS_NOCLDWAIT;
801 else
802 ps->ps_flag |= PS_NOCLDWAIT;
803 } else
804 ps->ps_flag &= ~PS_NOCLDWAIT;
805 if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
806 ps->ps_flag |= PS_CLDSIGIGN;
807 else
808 ps->ps_flag &= ~PS_CLDSIGIGN;
809 }
810 /*
811 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
812 * and for signals set to SIG_DFL where the default is to
813 * ignore. However, don't put SIGCONT in ps_sigignore, as we
814 * have to restart the process.
815 */
816 if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
817 (sigprop(sig) & SIGPROP_IGNORE &&
818 ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
819 /* never to be seen again */
820 sigqueue_delete_proc(p, sig);
821 if (sig != SIGCONT)
822 /* easier in psignal */
823 SIGADDSET(ps->ps_sigignore, sig);
824 SIGDELSET(ps->ps_sigcatch, sig);
825 } else {
826 SIGDELSET(ps->ps_sigignore, sig);
827 if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
828 SIGDELSET(ps->ps_sigcatch, sig);
829 else
830 SIGADDSET(ps->ps_sigcatch, sig);
831 }
832 #ifdef COMPAT_FREEBSD4
833 if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
834 ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
835 (flags & KSA_FREEBSD4) == 0)
836 SIGDELSET(ps->ps_freebsd4, sig);
837 else
838 SIGADDSET(ps->ps_freebsd4, sig);
839 #endif
840 #ifdef COMPAT_43
841 if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
842 ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
843 (flags & KSA_OSIGSET) == 0)
844 SIGDELSET(ps->ps_osigset, sig);
845 else
846 SIGADDSET(ps->ps_osigset, sig);
847 #endif
848 }
849 mtx_unlock(&ps->ps_mtx);
850 PROC_UNLOCK(p);
851 return (0);
852 }
853
854 #ifndef _SYS_SYSPROTO_H_
855 struct sigaction_args {
856 int sig;
857 struct sigaction *act;
858 struct sigaction *oact;
859 };
860 #endif
861 int
sys_sigaction(struct thread * td,struct sigaction_args * uap)862 sys_sigaction(struct thread *td, struct sigaction_args *uap)
863 {
864 struct sigaction act, oact;
865 struct sigaction *actp, *oactp;
866 int error;
867
868 actp = (uap->act != NULL) ? &act : NULL;
869 oactp = (uap->oact != NULL) ? &oact : NULL;
870 if (actp) {
871 error = copyin(uap->act, actp, sizeof(act));
872 if (error)
873 return (error);
874 }
875 error = kern_sigaction(td, uap->sig, actp, oactp, 0);
876 if (oactp && !error)
877 error = copyout(oactp, uap->oact, sizeof(oact));
878 return (error);
879 }
880
881 #ifdef COMPAT_FREEBSD4
882 #ifndef _SYS_SYSPROTO_H_
883 struct freebsd4_sigaction_args {
884 int sig;
885 struct sigaction *act;
886 struct sigaction *oact;
887 };
888 #endif
889 int
freebsd4_sigaction(struct thread * td,struct freebsd4_sigaction_args * uap)890 freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
891 {
892 struct sigaction act, oact;
893 struct sigaction *actp, *oactp;
894 int error;
895
896 actp = (uap->act != NULL) ? &act : NULL;
897 oactp = (uap->oact != NULL) ? &oact : NULL;
898 if (actp) {
899 error = copyin(uap->act, actp, sizeof(act));
900 if (error)
901 return (error);
902 }
903 error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
904 if (oactp && !error)
905 error = copyout(oactp, uap->oact, sizeof(oact));
906 return (error);
907 }
908 #endif /* COMAPT_FREEBSD4 */
909
910 #ifdef COMPAT_43 /* XXX - COMPAT_FBSD3 */
911 #ifndef _SYS_SYSPROTO_H_
912 struct osigaction_args {
913 int signum;
914 struct osigaction *nsa;
915 struct osigaction *osa;
916 };
917 #endif
918 int
osigaction(struct thread * td,struct osigaction_args * uap)919 osigaction(struct thread *td, struct osigaction_args *uap)
920 {
921 struct osigaction sa;
922 struct sigaction nsa, osa;
923 struct sigaction *nsap, *osap;
924 int error;
925
926 if (uap->signum <= 0 || uap->signum >= ONSIG)
927 return (EINVAL);
928
929 nsap = (uap->nsa != NULL) ? &nsa : NULL;
930 osap = (uap->osa != NULL) ? &osa : NULL;
931
932 if (nsap) {
933 error = copyin(uap->nsa, &sa, sizeof(sa));
934 if (error)
935 return (error);
936 nsap->sa_handler = sa.sa_handler;
937 nsap->sa_flags = sa.sa_flags;
938 OSIG2SIG(sa.sa_mask, nsap->sa_mask);
939 }
940 error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
941 if (osap && !error) {
942 sa.sa_handler = osap->sa_handler;
943 sa.sa_flags = osap->sa_flags;
944 SIG2OSIG(osap->sa_mask, sa.sa_mask);
945 error = copyout(&sa, uap->osa, sizeof(sa));
946 }
947 return (error);
948 }
949
950 #if !defined(__i386__)
951 /* Avoid replicating the same stub everywhere */
952 int
osigreturn(struct thread * td,struct osigreturn_args * uap)953 osigreturn(struct thread *td, struct osigreturn_args *uap)
954 {
955
956 return (nosys(td, (struct nosys_args *)uap));
957 }
958 #endif
959 #endif /* COMPAT_43 */
960
961 /*
962 * Initialize signal state for process 0;
963 * set to ignore signals that are ignored by default.
964 */
965 void
siginit(struct proc * p)966 siginit(struct proc *p)
967 {
968 int i;
969 struct sigacts *ps;
970
971 PROC_LOCK(p);
972 ps = p->p_sigacts;
973 mtx_lock(&ps->ps_mtx);
974 for (i = 1; i <= NSIG; i++) {
975 if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) {
976 SIGADDSET(ps->ps_sigignore, i);
977 }
978 }
979 mtx_unlock(&ps->ps_mtx);
980 PROC_UNLOCK(p);
981 }
982
983 /*
984 * Reset specified signal to the default disposition.
985 */
986 static void
sigdflt(struct sigacts * ps,int sig)987 sigdflt(struct sigacts *ps, int sig)
988 {
989
990 mtx_assert(&ps->ps_mtx, MA_OWNED);
991 SIGDELSET(ps->ps_sigcatch, sig);
992 if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT)
993 SIGADDSET(ps->ps_sigignore, sig);
994 ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
995 SIGDELSET(ps->ps_siginfo, sig);
996 }
997
998 /*
999 * Reset signals for an exec of the specified process.
1000 */
1001 void
execsigs(struct proc * p)1002 execsigs(struct proc *p)
1003 {
1004 sigset_t osigignore;
1005 struct sigacts *ps;
1006 int sig;
1007 struct thread *td;
1008
1009 /*
1010 * Reset caught signals. Held signals remain held
1011 * through td_sigmask (unless they were caught,
1012 * and are now ignored by default).
1013 */
1014 PROC_LOCK_ASSERT(p, MA_OWNED);
1015 ps = p->p_sigacts;
1016 mtx_lock(&ps->ps_mtx);
1017 sig_drop_caught(p);
1018
1019 /*
1020 * As CloudABI processes cannot modify signal handlers, fully
1021 * reset all signals to their default behavior. Do ignore
1022 * SIGPIPE, as it would otherwise be impossible to recover from
1023 * writes to broken pipes and sockets.
1024 */
1025 if (SV_PROC_ABI(p) == SV_ABI_CLOUDABI) {
1026 osigignore = ps->ps_sigignore;
1027 SIG_FOREACH(sig, &osigignore) {
1028 if (sig != SIGPIPE)
1029 sigdflt(ps, sig);
1030 }
1031 SIGADDSET(ps->ps_sigignore, SIGPIPE);
1032 }
1033
1034 /*
1035 * Reset stack state to the user stack.
1036 * Clear set of signals caught on the signal stack.
1037 */
1038 td = curthread;
1039 MPASS(td->td_proc == p);
1040 td->td_sigstk.ss_flags = SS_DISABLE;
1041 td->td_sigstk.ss_size = 0;
1042 td->td_sigstk.ss_sp = 0;
1043 td->td_pflags &= ~TDP_ALTSTACK;
1044 /*
1045 * Reset no zombies if child dies flag as Solaris does.
1046 */
1047 ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1048 if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1049 ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1050 mtx_unlock(&ps->ps_mtx);
1051 }
1052
1053 /*
1054 * kern_sigprocmask()
1055 *
1056 * Manipulate signal mask.
1057 */
1058 int
kern_sigprocmask(struct thread * td,int how,sigset_t * set,sigset_t * oset,int flags)1059 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1060 int flags)
1061 {
1062 sigset_t new_block, oset1;
1063 struct proc *p;
1064 int error;
1065
1066 p = td->td_proc;
1067 if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1068 PROC_LOCK_ASSERT(p, MA_OWNED);
1069 else
1070 PROC_LOCK(p);
1071 mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1072 ? MA_OWNED : MA_NOTOWNED);
1073 if (oset != NULL)
1074 *oset = td->td_sigmask;
1075
1076 error = 0;
1077 if (set != NULL) {
1078 switch (how) {
1079 case SIG_BLOCK:
1080 SIG_CANTMASK(*set);
1081 oset1 = td->td_sigmask;
1082 SIGSETOR(td->td_sigmask, *set);
1083 new_block = td->td_sigmask;
1084 SIGSETNAND(new_block, oset1);
1085 break;
1086 case SIG_UNBLOCK:
1087 SIGSETNAND(td->td_sigmask, *set);
1088 signotify(td);
1089 goto out;
1090 case SIG_SETMASK:
1091 SIG_CANTMASK(*set);
1092 oset1 = td->td_sigmask;
1093 if (flags & SIGPROCMASK_OLD)
1094 SIGSETLO(td->td_sigmask, *set);
1095 else
1096 td->td_sigmask = *set;
1097 new_block = td->td_sigmask;
1098 SIGSETNAND(new_block, oset1);
1099 signotify(td);
1100 break;
1101 default:
1102 error = EINVAL;
1103 goto out;
1104 }
1105
1106 /*
1107 * The new_block set contains signals that were not previously
1108 * blocked, but are blocked now.
1109 *
1110 * In case we block any signal that was not previously blocked
1111 * for td, and process has the signal pending, try to schedule
1112 * signal delivery to some thread that does not block the
1113 * signal, possibly waking it up.
1114 */
1115 if (p->p_numthreads != 1)
1116 reschedule_signals(p, new_block, flags);
1117 }
1118
1119 out:
1120 if (!(flags & SIGPROCMASK_PROC_LOCKED))
1121 PROC_UNLOCK(p);
1122 return (error);
1123 }
1124
1125 #ifndef _SYS_SYSPROTO_H_
1126 struct sigprocmask_args {
1127 int how;
1128 const sigset_t *set;
1129 sigset_t *oset;
1130 };
1131 #endif
1132 int
sys_sigprocmask(struct thread * td,struct sigprocmask_args * uap)1133 sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1134 {
1135 sigset_t set, oset;
1136 sigset_t *setp, *osetp;
1137 int error;
1138
1139 setp = (uap->set != NULL) ? &set : NULL;
1140 osetp = (uap->oset != NULL) ? &oset : NULL;
1141 if (setp) {
1142 error = copyin(uap->set, setp, sizeof(set));
1143 if (error)
1144 return (error);
1145 }
1146 error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1147 if (osetp && !error) {
1148 error = copyout(osetp, uap->oset, sizeof(oset));
1149 }
1150 return (error);
1151 }
1152
1153 #ifdef COMPAT_43 /* XXX - COMPAT_FBSD3 */
1154 #ifndef _SYS_SYSPROTO_H_
1155 struct osigprocmask_args {
1156 int how;
1157 osigset_t mask;
1158 };
1159 #endif
1160 int
osigprocmask(struct thread * td,struct osigprocmask_args * uap)1161 osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1162 {
1163 sigset_t set, oset;
1164 int error;
1165
1166 OSIG2SIG(uap->mask, set);
1167 error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1168 SIG2OSIG(oset, td->td_retval[0]);
1169 return (error);
1170 }
1171 #endif /* COMPAT_43 */
1172
1173 int
sys_sigwait(struct thread * td,struct sigwait_args * uap)1174 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1175 {
1176 ksiginfo_t ksi;
1177 sigset_t set;
1178 int error;
1179
1180 error = copyin(uap->set, &set, sizeof(set));
1181 if (error) {
1182 td->td_retval[0] = error;
1183 return (0);
1184 }
1185
1186 error = kern_sigtimedwait(td, set, &ksi, NULL);
1187 if (error) {
1188 /*
1189 * sigwait() function shall not return EINTR, but
1190 * the syscall does. Non-ancient libc provides the
1191 * wrapper which hides EINTR. Otherwise, EINTR return
1192 * is used by libthr to handle required cancellation
1193 * point in the sigwait().
1194 */
1195 if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1196 return (ERESTART);
1197 td->td_retval[0] = error;
1198 return (0);
1199 }
1200
1201 error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1202 td->td_retval[0] = error;
1203 return (0);
1204 }
1205
1206 int
sys_sigtimedwait(struct thread * td,struct sigtimedwait_args * uap)1207 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1208 {
1209 struct timespec ts;
1210 struct timespec *timeout;
1211 sigset_t set;
1212 ksiginfo_t ksi;
1213 int error;
1214
1215 if (uap->timeout) {
1216 error = copyin(uap->timeout, &ts, sizeof(ts));
1217 if (error)
1218 return (error);
1219
1220 timeout = &ts;
1221 } else
1222 timeout = NULL;
1223
1224 error = copyin(uap->set, &set, sizeof(set));
1225 if (error)
1226 return (error);
1227
1228 error = kern_sigtimedwait(td, set, &ksi, timeout);
1229 if (error)
1230 return (error);
1231
1232 if (uap->info)
1233 error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1234
1235 if (error == 0)
1236 td->td_retval[0] = ksi.ksi_signo;
1237 return (error);
1238 }
1239
1240 int
sys_sigwaitinfo(struct thread * td,struct sigwaitinfo_args * uap)1241 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1242 {
1243 ksiginfo_t ksi;
1244 sigset_t set;
1245 int error;
1246
1247 error = copyin(uap->set, &set, sizeof(set));
1248 if (error)
1249 return (error);
1250
1251 error = kern_sigtimedwait(td, set, &ksi, NULL);
1252 if (error)
1253 return (error);
1254
1255 if (uap->info)
1256 error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1257
1258 if (error == 0)
1259 td->td_retval[0] = ksi.ksi_signo;
1260 return (error);
1261 }
1262
1263 static void
proc_td_siginfo_capture(struct thread * td,siginfo_t * si)1264 proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1265 {
1266 struct thread *thr;
1267
1268 FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1269 if (thr == td)
1270 thr->td_si = *si;
1271 else
1272 thr->td_si.si_signo = 0;
1273 }
1274 }
1275
1276 int
kern_sigtimedwait(struct thread * td,sigset_t waitset,ksiginfo_t * ksi,struct timespec * timeout)1277 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1278 struct timespec *timeout)
1279 {
1280 struct sigacts *ps;
1281 sigset_t saved_mask, new_block;
1282 struct proc *p;
1283 int error, sig, timevalid = 0;
1284 sbintime_t sbt, precision, tsbt;
1285 struct timespec ts;
1286 bool traced;
1287
1288 p = td->td_proc;
1289 error = 0;
1290 traced = false;
1291
1292 /* Ensure the sigfastblock value is up to date. */
1293 sigfastblock_fetch(td);
1294
1295 if (timeout != NULL) {
1296 if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1297 timevalid = 1;
1298 ts = *timeout;
1299 if (ts.tv_sec < INT32_MAX / 2) {
1300 tsbt = tstosbt(ts);
1301 precision = tsbt;
1302 precision >>= tc_precexp;
1303 if (TIMESEL(&sbt, tsbt))
1304 sbt += tc_tick_sbt;
1305 sbt += tsbt;
1306 } else
1307 precision = sbt = 0;
1308 }
1309 } else
1310 precision = sbt = 0;
1311 ksiginfo_init(ksi);
1312 /* Some signals can not be waited for. */
1313 SIG_CANTMASK(waitset);
1314 ps = p->p_sigacts;
1315 PROC_LOCK(p);
1316 saved_mask = td->td_sigmask;
1317 SIGSETNAND(td->td_sigmask, waitset);
1318 if ((p->p_sysent->sv_flags & SV_SIG_DISCIGN) != 0 ||
1319 !kern_sig_discard_ign) {
1320 thread_lock(td);
1321 td->td_flags |= TDF_SIGWAIT;
1322 thread_unlock(td);
1323 }
1324 for (;;) {
1325 mtx_lock(&ps->ps_mtx);
1326 sig = cursig(td);
1327 mtx_unlock(&ps->ps_mtx);
1328 KASSERT(sig >= 0, ("sig %d", sig));
1329 if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1330 if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1331 sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1332 error = 0;
1333 break;
1334 }
1335 }
1336
1337 if (error != 0)
1338 break;
1339
1340 /*
1341 * POSIX says this must be checked after looking for pending
1342 * signals.
1343 */
1344 if (timeout != NULL && !timevalid) {
1345 error = EINVAL;
1346 break;
1347 }
1348
1349 if (traced) {
1350 error = EINTR;
1351 break;
1352 }
1353
1354 error = msleep_sbt(&p->p_sigacts, &p->p_mtx, PPAUSE | PCATCH,
1355 "sigwait", sbt, precision, C_ABSOLUTE);
1356
1357 /* The syscalls can not be restarted. */
1358 if (error == ERESTART)
1359 error = EINTR;
1360
1361 /*
1362 * If PTRACE_SCE or PTRACE_SCX were set after
1363 * userspace entered the syscall, return spurious
1364 * EINTR after wait was done. Only do this as last
1365 * resort after rechecking for possible queued signals
1366 * and expired timeouts.
1367 */
1368 if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1369 traced = true;
1370 }
1371 thread_lock(td);
1372 td->td_flags &= ~TDF_SIGWAIT;
1373 thread_unlock(td);
1374
1375 new_block = saved_mask;
1376 SIGSETNAND(new_block, td->td_sigmask);
1377 td->td_sigmask = saved_mask;
1378 /*
1379 * Fewer signals can be delivered to us, reschedule signal
1380 * notification.
1381 */
1382 if (p->p_numthreads != 1)
1383 reschedule_signals(p, new_block, 0);
1384
1385 if (error == 0) {
1386 SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1387
1388 if (ksi->ksi_code == SI_TIMER)
1389 itimer_accept(p, ksi->ksi_timerid, ksi);
1390
1391 #ifdef KTRACE
1392 if (KTRPOINT(td, KTR_PSIG)) {
1393 sig_t action;
1394
1395 mtx_lock(&ps->ps_mtx);
1396 action = ps->ps_sigact[_SIG_IDX(sig)];
1397 mtx_unlock(&ps->ps_mtx);
1398 ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1399 }
1400 #endif
1401 if (sig == SIGKILL) {
1402 proc_td_siginfo_capture(td, &ksi->ksi_info);
1403 sigexit(td, sig);
1404 }
1405 }
1406 PROC_UNLOCK(p);
1407 return (error);
1408 }
1409
1410 #ifndef _SYS_SYSPROTO_H_
1411 struct sigpending_args {
1412 sigset_t *set;
1413 };
1414 #endif
1415 int
sys_sigpending(struct thread * td,struct sigpending_args * uap)1416 sys_sigpending(struct thread *td, struct sigpending_args *uap)
1417 {
1418 struct proc *p = td->td_proc;
1419 sigset_t pending;
1420
1421 PROC_LOCK(p);
1422 pending = p->p_sigqueue.sq_signals;
1423 SIGSETOR(pending, td->td_sigqueue.sq_signals);
1424 PROC_UNLOCK(p);
1425 return (copyout(&pending, uap->set, sizeof(sigset_t)));
1426 }
1427
1428 #ifdef COMPAT_43 /* XXX - COMPAT_FBSD3 */
1429 #ifndef _SYS_SYSPROTO_H_
1430 struct osigpending_args {
1431 int dummy;
1432 };
1433 #endif
1434 int
osigpending(struct thread * td,struct osigpending_args * uap)1435 osigpending(struct thread *td, struct osigpending_args *uap)
1436 {
1437 struct proc *p = td->td_proc;
1438 sigset_t pending;
1439
1440 PROC_LOCK(p);
1441 pending = p->p_sigqueue.sq_signals;
1442 SIGSETOR(pending, td->td_sigqueue.sq_signals);
1443 PROC_UNLOCK(p);
1444 SIG2OSIG(pending, td->td_retval[0]);
1445 return (0);
1446 }
1447 #endif /* COMPAT_43 */
1448
1449 #if defined(COMPAT_43)
1450 /*
1451 * Generalized interface signal handler, 4.3-compatible.
1452 */
1453 #ifndef _SYS_SYSPROTO_H_
1454 struct osigvec_args {
1455 int signum;
1456 struct sigvec *nsv;
1457 struct sigvec *osv;
1458 };
1459 #endif
1460 /* ARGSUSED */
1461 int
osigvec(struct thread * td,struct osigvec_args * uap)1462 osigvec(struct thread *td, struct osigvec_args *uap)
1463 {
1464 struct sigvec vec;
1465 struct sigaction nsa, osa;
1466 struct sigaction *nsap, *osap;
1467 int error;
1468
1469 if (uap->signum <= 0 || uap->signum >= ONSIG)
1470 return (EINVAL);
1471 nsap = (uap->nsv != NULL) ? &nsa : NULL;
1472 osap = (uap->osv != NULL) ? &osa : NULL;
1473 if (nsap) {
1474 error = copyin(uap->nsv, &vec, sizeof(vec));
1475 if (error)
1476 return (error);
1477 nsap->sa_handler = vec.sv_handler;
1478 OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1479 nsap->sa_flags = vec.sv_flags;
1480 nsap->sa_flags ^= SA_RESTART; /* opposite of SV_INTERRUPT */
1481 }
1482 error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1483 if (osap && !error) {
1484 vec.sv_handler = osap->sa_handler;
1485 SIG2OSIG(osap->sa_mask, vec.sv_mask);
1486 vec.sv_flags = osap->sa_flags;
1487 vec.sv_flags &= ~SA_NOCLDWAIT;
1488 vec.sv_flags ^= SA_RESTART;
1489 error = copyout(&vec, uap->osv, sizeof(vec));
1490 }
1491 return (error);
1492 }
1493
1494 #ifndef _SYS_SYSPROTO_H_
1495 struct osigblock_args {
1496 int mask;
1497 };
1498 #endif
1499 int
osigblock(struct thread * td,struct osigblock_args * uap)1500 osigblock(struct thread *td, struct osigblock_args *uap)
1501 {
1502 sigset_t set, oset;
1503
1504 OSIG2SIG(uap->mask, set);
1505 kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1506 SIG2OSIG(oset, td->td_retval[0]);
1507 return (0);
1508 }
1509
1510 #ifndef _SYS_SYSPROTO_H_
1511 struct osigsetmask_args {
1512 int mask;
1513 };
1514 #endif
1515 int
osigsetmask(struct thread * td,struct osigsetmask_args * uap)1516 osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1517 {
1518 sigset_t set, oset;
1519
1520 OSIG2SIG(uap->mask, set);
1521 kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1522 SIG2OSIG(oset, td->td_retval[0]);
1523 return (0);
1524 }
1525 #endif /* COMPAT_43 */
1526
1527 /*
1528 * Suspend calling thread until signal, providing mask to be set in the
1529 * meantime.
1530 */
1531 #ifndef _SYS_SYSPROTO_H_
1532 struct sigsuspend_args {
1533 const sigset_t *sigmask;
1534 };
1535 #endif
1536 /* ARGSUSED */
1537 int
sys_sigsuspend(struct thread * td,struct sigsuspend_args * uap)1538 sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1539 {
1540 sigset_t mask;
1541 int error;
1542
1543 error = copyin(uap->sigmask, &mask, sizeof(mask));
1544 if (error)
1545 return (error);
1546 return (kern_sigsuspend(td, mask));
1547 }
1548
1549 int
kern_sigsuspend(struct thread * td,sigset_t mask)1550 kern_sigsuspend(struct thread *td, sigset_t mask)
1551 {
1552 struct proc *p = td->td_proc;
1553 int has_sig, sig;
1554
1555 /* Ensure the sigfastblock value is up to date. */
1556 sigfastblock_fetch(td);
1557
1558 /*
1559 * When returning from sigsuspend, we want
1560 * the old mask to be restored after the
1561 * signal handler has finished. Thus, we
1562 * save it here and mark the sigacts structure
1563 * to indicate this.
1564 */
1565 PROC_LOCK(p);
1566 kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1567 SIGPROCMASK_PROC_LOCKED);
1568 td->td_pflags |= TDP_OLDMASK;
1569
1570 /*
1571 * Process signals now. Otherwise, we can get spurious wakeup
1572 * due to signal entered process queue, but delivered to other
1573 * thread. But sigsuspend should return only on signal
1574 * delivery.
1575 */
1576 (p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1577 for (has_sig = 0; !has_sig;) {
1578 while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1579 0) == 0)
1580 /* void */;
1581 thread_suspend_check(0);
1582 mtx_lock(&p->p_sigacts->ps_mtx);
1583 while ((sig = cursig(td)) != 0) {
1584 KASSERT(sig >= 0, ("sig %d", sig));
1585 has_sig += postsig(sig);
1586 }
1587 mtx_unlock(&p->p_sigacts->ps_mtx);
1588
1589 /*
1590 * If PTRACE_SCE or PTRACE_SCX were set after
1591 * userspace entered the syscall, return spurious
1592 * EINTR.
1593 */
1594 if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1595 has_sig += 1;
1596 }
1597 PROC_UNLOCK(p);
1598 td->td_errno = EINTR;
1599 td->td_pflags |= TDP_NERRNO;
1600 return (EJUSTRETURN);
1601 }
1602
1603 #ifdef COMPAT_43 /* XXX - COMPAT_FBSD3 */
1604 /*
1605 * Compatibility sigsuspend call for old binaries. Note nonstandard calling
1606 * convention: libc stub passes mask, not pointer, to save a copyin.
1607 */
1608 #ifndef _SYS_SYSPROTO_H_
1609 struct osigsuspend_args {
1610 osigset_t mask;
1611 };
1612 #endif
1613 /* ARGSUSED */
1614 int
osigsuspend(struct thread * td,struct osigsuspend_args * uap)1615 osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1616 {
1617 sigset_t mask;
1618
1619 OSIG2SIG(uap->mask, mask);
1620 return (kern_sigsuspend(td, mask));
1621 }
1622 #endif /* COMPAT_43 */
1623
1624 #if defined(COMPAT_43)
1625 #ifndef _SYS_SYSPROTO_H_
1626 struct osigstack_args {
1627 struct sigstack *nss;
1628 struct sigstack *oss;
1629 };
1630 #endif
1631 /* ARGSUSED */
1632 int
osigstack(struct thread * td,struct osigstack_args * uap)1633 osigstack(struct thread *td, struct osigstack_args *uap)
1634 {
1635 struct sigstack nss, oss;
1636 int error = 0;
1637
1638 if (uap->nss != NULL) {
1639 error = copyin(uap->nss, &nss, sizeof(nss));
1640 if (error)
1641 return (error);
1642 }
1643 oss.ss_sp = td->td_sigstk.ss_sp;
1644 oss.ss_onstack = sigonstack(cpu_getstack(td));
1645 if (uap->nss != NULL) {
1646 td->td_sigstk.ss_sp = nss.ss_sp;
1647 td->td_sigstk.ss_size = 0;
1648 td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1649 td->td_pflags |= TDP_ALTSTACK;
1650 }
1651 if (uap->oss != NULL)
1652 error = copyout(&oss, uap->oss, sizeof(oss));
1653
1654 return (error);
1655 }
1656 #endif /* COMPAT_43 */
1657
1658 #ifndef _SYS_SYSPROTO_H_
1659 struct sigaltstack_args {
1660 stack_t *ss;
1661 stack_t *oss;
1662 };
1663 #endif
1664 /* ARGSUSED */
1665 int
sys_sigaltstack(struct thread * td,struct sigaltstack_args * uap)1666 sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1667 {
1668 stack_t ss, oss;
1669 int error;
1670
1671 if (uap->ss != NULL) {
1672 error = copyin(uap->ss, &ss, sizeof(ss));
1673 if (error)
1674 return (error);
1675 }
1676 error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1677 (uap->oss != NULL) ? &oss : NULL);
1678 if (error)
1679 return (error);
1680 if (uap->oss != NULL)
1681 error = copyout(&oss, uap->oss, sizeof(stack_t));
1682 return (error);
1683 }
1684
1685 int
kern_sigaltstack(struct thread * td,stack_t * ss,stack_t * oss)1686 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1687 {
1688 struct proc *p = td->td_proc;
1689 int oonstack;
1690
1691 oonstack = sigonstack(cpu_getstack(td));
1692
1693 if (oss != NULL) {
1694 *oss = td->td_sigstk;
1695 oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1696 ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1697 }
1698
1699 if (ss != NULL) {
1700 if (oonstack)
1701 return (EPERM);
1702 if ((ss->ss_flags & ~SS_DISABLE) != 0)
1703 return (EINVAL);
1704 if (!(ss->ss_flags & SS_DISABLE)) {
1705 if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1706 return (ENOMEM);
1707
1708 td->td_sigstk = *ss;
1709 td->td_pflags |= TDP_ALTSTACK;
1710 } else {
1711 td->td_pflags &= ~TDP_ALTSTACK;
1712 }
1713 }
1714 return (0);
1715 }
1716
1717 struct killpg1_ctx {
1718 struct thread *td;
1719 ksiginfo_t *ksi;
1720 int sig;
1721 bool sent;
1722 bool found;
1723 int ret;
1724 };
1725
1726 static void
killpg1_sendsig(struct proc * p,bool notself,struct killpg1_ctx * arg)1727 killpg1_sendsig(struct proc *p, bool notself, struct killpg1_ctx *arg)
1728 {
1729 int err;
1730
1731 if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1732 (notself && p == arg->td->td_proc) || p->p_state == PRS_NEW)
1733 return;
1734 PROC_LOCK(p);
1735 err = p_cansignal(arg->td, p, arg->sig);
1736 if (err == 0 && arg->sig != 0)
1737 pksignal(p, arg->sig, arg->ksi);
1738 PROC_UNLOCK(p);
1739 if (err != ESRCH)
1740 arg->found = true;
1741 if (err == 0)
1742 arg->sent = true;
1743 else if (arg->ret == 0 && err != ESRCH && err != EPERM)
1744 arg->ret = err;
1745 }
1746
1747 /*
1748 * Common code for kill process group/broadcast kill.
1749 * td is the calling thread, as usual.
1750 */
1751 static int
killpg1(struct thread * td,int sig,int pgid,int all,ksiginfo_t * ksi)1752 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1753 {
1754 struct proc *p;
1755 struct pgrp *pgrp;
1756 struct killpg1_ctx arg;
1757
1758 arg.td = td;
1759 arg.ksi = ksi;
1760 arg.sig = sig;
1761 arg.sent = false;
1762 arg.found = false;
1763 arg.ret = 0;
1764 if (all) {
1765 /*
1766 * broadcast
1767 */
1768 sx_slock(&allproc_lock);
1769 FOREACH_PROC_IN_SYSTEM(p) {
1770 killpg1_sendsig(p, true, &arg);
1771 }
1772 sx_sunlock(&allproc_lock);
1773 } else {
1774 again:
1775 sx_slock(&proctree_lock);
1776 if (pgid == 0) {
1777 /*
1778 * zero pgid means send to my process group.
1779 */
1780 pgrp = td->td_proc->p_pgrp;
1781 PGRP_LOCK(pgrp);
1782 } else {
1783 pgrp = pgfind(pgid);
1784 if (pgrp == NULL) {
1785 sx_sunlock(&proctree_lock);
1786 return (ESRCH);
1787 }
1788 }
1789 sx_sunlock(&proctree_lock);
1790 if (!sx_try_xlock(&pgrp->pg_killsx)) {
1791 PGRP_UNLOCK(pgrp);
1792 sx_xlock(&pgrp->pg_killsx);
1793 sx_xunlock(&pgrp->pg_killsx);
1794 goto again;
1795 }
1796 LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1797 killpg1_sendsig(p, false, &arg);
1798 }
1799 PGRP_UNLOCK(pgrp);
1800 sx_xunlock(&pgrp->pg_killsx);
1801 }
1802 MPASS(arg.ret != 0 || arg.found || !arg.sent);
1803 if (arg.ret == 0 && !arg.sent)
1804 arg.ret = arg.found ? EPERM : ESRCH;
1805 return (arg.ret);
1806 }
1807
1808 #ifndef _SYS_SYSPROTO_H_
1809 struct kill_args {
1810 int pid;
1811 int signum;
1812 };
1813 #endif
1814 /* ARGSUSED */
1815 int
sys_kill(struct thread * td,struct kill_args * uap)1816 sys_kill(struct thread *td, struct kill_args *uap)
1817 {
1818
1819 return (kern_kill(td, uap->pid, uap->signum));
1820 }
1821
1822 int
kern_kill(struct thread * td,pid_t pid,int signum)1823 kern_kill(struct thread *td, pid_t pid, int signum)
1824 {
1825 ksiginfo_t ksi;
1826 struct proc *p;
1827 int error;
1828
1829 /*
1830 * A process in capability mode can send signals only to himself.
1831 * The main rationale behind this is that abort(3) is implemented as
1832 * kill(getpid(), SIGABRT).
1833 */
1834 if (IN_CAPABILITY_MODE(td) && pid != td->td_proc->p_pid)
1835 return (ECAPMODE);
1836
1837 AUDIT_ARG_SIGNUM(signum);
1838 AUDIT_ARG_PID(pid);
1839 if ((u_int)signum > _SIG_MAXSIG)
1840 return (EINVAL);
1841
1842 ksiginfo_init(&ksi);
1843 ksi.ksi_signo = signum;
1844 ksi.ksi_code = SI_USER;
1845 ksi.ksi_pid = td->td_proc->p_pid;
1846 ksi.ksi_uid = td->td_ucred->cr_ruid;
1847
1848 if (pid > 0) {
1849 /* kill single process */
1850 if ((p = pfind_any(pid)) == NULL)
1851 return (ESRCH);
1852 AUDIT_ARG_PROCESS(p);
1853 error = p_cansignal(td, p, signum);
1854 if (error == 0 && signum)
1855 pksignal(p, signum, &ksi);
1856 PROC_UNLOCK(p);
1857 return (error);
1858 }
1859 switch (pid) {
1860 case -1: /* broadcast signal */
1861 return (killpg1(td, signum, 0, 1, &ksi));
1862 case 0: /* signal own process group */
1863 return (killpg1(td, signum, 0, 0, &ksi));
1864 default: /* negative explicit process group */
1865 return (killpg1(td, signum, -pid, 0, &ksi));
1866 }
1867 /* NOTREACHED */
1868 }
1869
1870 int
sys_pdkill(struct thread * td,struct pdkill_args * uap)1871 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1872 {
1873 struct proc *p;
1874 int error;
1875
1876 AUDIT_ARG_SIGNUM(uap->signum);
1877 AUDIT_ARG_FD(uap->fd);
1878 if ((u_int)uap->signum > _SIG_MAXSIG)
1879 return (EINVAL);
1880
1881 error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1882 if (error)
1883 return (error);
1884 AUDIT_ARG_PROCESS(p);
1885 error = p_cansignal(td, p, uap->signum);
1886 if (error == 0 && uap->signum)
1887 kern_psignal(p, uap->signum);
1888 PROC_UNLOCK(p);
1889 return (error);
1890 }
1891
1892 #if defined(COMPAT_43)
1893 #ifndef _SYS_SYSPROTO_H_
1894 struct okillpg_args {
1895 int pgid;
1896 int signum;
1897 };
1898 #endif
1899 /* ARGSUSED */
1900 int
okillpg(struct thread * td,struct okillpg_args * uap)1901 okillpg(struct thread *td, struct okillpg_args *uap)
1902 {
1903 ksiginfo_t ksi;
1904
1905 AUDIT_ARG_SIGNUM(uap->signum);
1906 AUDIT_ARG_PID(uap->pgid);
1907 if ((u_int)uap->signum > _SIG_MAXSIG)
1908 return (EINVAL);
1909
1910 ksiginfo_init(&ksi);
1911 ksi.ksi_signo = uap->signum;
1912 ksi.ksi_code = SI_USER;
1913 ksi.ksi_pid = td->td_proc->p_pid;
1914 ksi.ksi_uid = td->td_ucred->cr_ruid;
1915 return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1916 }
1917 #endif /* COMPAT_43 */
1918
1919 #ifndef _SYS_SYSPROTO_H_
1920 struct sigqueue_args {
1921 pid_t pid;
1922 int signum;
1923 /* union sigval */ void *value;
1924 };
1925 #endif
1926 int
sys_sigqueue(struct thread * td,struct sigqueue_args * uap)1927 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1928 {
1929 union sigval sv;
1930
1931 sv.sival_ptr = uap->value;
1932
1933 return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
1934 }
1935
1936 int
kern_sigqueue(struct thread * td,pid_t pid,int signum,union sigval * value)1937 kern_sigqueue(struct thread *td, pid_t pid, int signum, union sigval *value)
1938 {
1939 ksiginfo_t ksi;
1940 struct proc *p;
1941 int error;
1942
1943 if ((u_int)signum > _SIG_MAXSIG)
1944 return (EINVAL);
1945
1946 /*
1947 * Specification says sigqueue can only send signal to
1948 * single process.
1949 */
1950 if (pid <= 0)
1951 return (EINVAL);
1952
1953 if ((p = pfind_any(pid)) == NULL)
1954 return (ESRCH);
1955 error = p_cansignal(td, p, signum);
1956 if (error == 0 && signum != 0) {
1957 ksiginfo_init(&ksi);
1958 ksi.ksi_flags = KSI_SIGQ;
1959 ksi.ksi_signo = signum;
1960 ksi.ksi_code = SI_QUEUE;
1961 ksi.ksi_pid = td->td_proc->p_pid;
1962 ksi.ksi_uid = td->td_ucred->cr_ruid;
1963 ksi.ksi_value = *value;
1964 error = pksignal(p, ksi.ksi_signo, &ksi);
1965 }
1966 PROC_UNLOCK(p);
1967 return (error);
1968 }
1969
1970 /*
1971 * Send a signal to a process group. If checktty is 1,
1972 * limit to members which have a controlling terminal.
1973 */
1974 void
pgsignal(struct pgrp * pgrp,int sig,int checkctty,ksiginfo_t * ksi)1975 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
1976 {
1977 struct proc *p;
1978
1979 if (pgrp) {
1980 PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
1981 LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1982 PROC_LOCK(p);
1983 if (p->p_state == PRS_NORMAL &&
1984 (checkctty == 0 || p->p_flag & P_CONTROLT))
1985 pksignal(p, sig, ksi);
1986 PROC_UNLOCK(p);
1987 }
1988 }
1989 }
1990
1991 /*
1992 * Recalculate the signal mask and reset the signal disposition after
1993 * usermode frame for delivery is formed. Should be called after
1994 * mach-specific routine, because sysent->sv_sendsig() needs correct
1995 * ps_siginfo and signal mask.
1996 */
1997 static void
postsig_done(int sig,struct thread * td,struct sigacts * ps)1998 postsig_done(int sig, struct thread *td, struct sigacts *ps)
1999 {
2000 sigset_t mask;
2001
2002 mtx_assert(&ps->ps_mtx, MA_OWNED);
2003 td->td_ru.ru_nsignals++;
2004 mask = ps->ps_catchmask[_SIG_IDX(sig)];
2005 if (!SIGISMEMBER(ps->ps_signodefer, sig))
2006 SIGADDSET(mask, sig);
2007 kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
2008 SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
2009 if (SIGISMEMBER(ps->ps_sigreset, sig))
2010 sigdflt(ps, sig);
2011 }
2012
2013 /*
2014 * Send a signal caused by a trap to the current thread. If it will be
2015 * caught immediately, deliver it with correct code. Otherwise, post it
2016 * normally.
2017 */
2018 void
trapsignal(struct thread * td,ksiginfo_t * ksi)2019 trapsignal(struct thread *td, ksiginfo_t *ksi)
2020 {
2021 struct sigacts *ps;
2022 struct proc *p;
2023 sigset_t sigmask;
2024 int code, sig;
2025
2026 p = td->td_proc;
2027 sig = ksi->ksi_signo;
2028 code = ksi->ksi_code;
2029 KASSERT(_SIG_VALID(sig), ("invalid signal"));
2030
2031 sigfastblock_fetch(td);
2032 PROC_LOCK(p);
2033 ps = p->p_sigacts;
2034 mtx_lock(&ps->ps_mtx);
2035 sigmask = td->td_sigmask;
2036 if (td->td_sigblock_val != 0)
2037 SIGSETOR(sigmask, fastblock_mask);
2038 if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2039 !SIGISMEMBER(sigmask, sig)) {
2040 #ifdef KTRACE
2041 if (KTRPOINT(curthread, KTR_PSIG))
2042 ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2043 &td->td_sigmask, code);
2044 #endif
2045 (*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2046 ksi, &td->td_sigmask);
2047 postsig_done(sig, td, ps);
2048 mtx_unlock(&ps->ps_mtx);
2049 } else {
2050 /*
2051 * Avoid a possible infinite loop if the thread
2052 * masking the signal or process is ignoring the
2053 * signal.
2054 */
2055 if (kern_forcesigexit && (SIGISMEMBER(sigmask, sig) ||
2056 ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2057 SIGDELSET(td->td_sigmask, sig);
2058 SIGDELSET(ps->ps_sigcatch, sig);
2059 SIGDELSET(ps->ps_sigignore, sig);
2060 ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2061 td->td_pflags &= ~TDP_SIGFASTBLOCK;
2062 td->td_sigblock_val = 0;
2063 }
2064 mtx_unlock(&ps->ps_mtx);
2065 p->p_sig = sig; /* XXX to verify code */
2066 tdsendsignal(p, td, sig, ksi);
2067 }
2068 PROC_UNLOCK(p);
2069 }
2070
2071 static struct thread *
sigtd(struct proc * p,int sig,bool fast_sigblock)2072 sigtd(struct proc *p, int sig, bool fast_sigblock)
2073 {
2074 struct thread *td, *signal_td;
2075
2076 PROC_LOCK_ASSERT(p, MA_OWNED);
2077 MPASS(!fast_sigblock || p == curproc);
2078
2079 /*
2080 * Check if current thread can handle the signal without
2081 * switching context to another thread.
2082 */
2083 if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig) &&
2084 (!fast_sigblock || curthread->td_sigblock_val == 0))
2085 return (curthread);
2086
2087 /* Find a non-stopped thread that does not mask the signal. */
2088 signal_td = NULL;
2089 FOREACH_THREAD_IN_PROC(p, td) {
2090 if (!SIGISMEMBER(td->td_sigmask, sig) && (!fast_sigblock ||
2091 td != curthread || td->td_sigblock_val == 0) &&
2092 (td->td_flags & TDF_BOUNDARY) == 0) {
2093 signal_td = td;
2094 break;
2095 }
2096 }
2097 /* Select random (first) thread if no better match was found. */
2098 if (signal_td == NULL)
2099 signal_td = FIRST_THREAD_IN_PROC(p);
2100 return (signal_td);
2101 }
2102
2103 /*
2104 * Send the signal to the process. If the signal has an action, the action
2105 * is usually performed by the target process rather than the caller; we add
2106 * the signal to the set of pending signals for the process.
2107 *
2108 * Exceptions:
2109 * o When a stop signal is sent to a sleeping process that takes the
2110 * default action, the process is stopped without awakening it.
2111 * o SIGCONT restarts stopped processes (or puts them back to sleep)
2112 * regardless of the signal action (eg, blocked or ignored).
2113 *
2114 * Other ignored signals are discarded immediately.
2115 *
2116 * NB: This function may be entered from the debugger via the "kill" DDB
2117 * command. There is little that can be done to mitigate the possibly messy
2118 * side effects of this unwise possibility.
2119 */
2120 void
kern_psignal(struct proc * p,int sig)2121 kern_psignal(struct proc *p, int sig)
2122 {
2123 ksiginfo_t ksi;
2124
2125 ksiginfo_init(&ksi);
2126 ksi.ksi_signo = sig;
2127 ksi.ksi_code = SI_KERNEL;
2128 (void) tdsendsignal(p, NULL, sig, &ksi);
2129 }
2130
2131 int
pksignal(struct proc * p,int sig,ksiginfo_t * ksi)2132 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2133 {
2134
2135 return (tdsendsignal(p, NULL, sig, ksi));
2136 }
2137
2138 /* Utility function for finding a thread to send signal event to. */
2139 int
sigev_findtd(struct proc * p,struct sigevent * sigev,struct thread ** ttd)2140 sigev_findtd(struct proc *p, struct sigevent *sigev, struct thread **ttd)
2141 {
2142 struct thread *td;
2143
2144 if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2145 td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2146 if (td == NULL)
2147 return (ESRCH);
2148 *ttd = td;
2149 } else {
2150 *ttd = NULL;
2151 PROC_LOCK(p);
2152 }
2153 return (0);
2154 }
2155
2156 void
tdsignal(struct thread * td,int sig)2157 tdsignal(struct thread *td, int sig)
2158 {
2159 ksiginfo_t ksi;
2160
2161 ksiginfo_init(&ksi);
2162 ksi.ksi_signo = sig;
2163 ksi.ksi_code = SI_KERNEL;
2164 (void) tdsendsignal(td->td_proc, td, sig, &ksi);
2165 }
2166
2167 void
tdksignal(struct thread * td,int sig,ksiginfo_t * ksi)2168 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2169 {
2170
2171 (void) tdsendsignal(td->td_proc, td, sig, ksi);
2172 }
2173
2174 static int
sig_sleepq_abort(struct thread * td,int intrval)2175 sig_sleepq_abort(struct thread *td, int intrval)
2176 {
2177 THREAD_LOCK_ASSERT(td, MA_OWNED);
2178
2179 if (intrval == 0 && (td->td_flags & TDF_SIGWAIT) == 0) {
2180 thread_unlock(td);
2181 return (0);
2182 }
2183 return (sleepq_abort(td, intrval));
2184 }
2185
2186 int
tdsendsignal(struct proc * p,struct thread * td,int sig,ksiginfo_t * ksi)2187 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2188 {
2189 sig_t action;
2190 sigqueue_t *sigqueue;
2191 int prop;
2192 struct sigacts *ps;
2193 int intrval;
2194 int ret = 0;
2195 int wakeup_swapper;
2196
2197 MPASS(td == NULL || p == td->td_proc);
2198 PROC_LOCK_ASSERT(p, MA_OWNED);
2199
2200 if (!_SIG_VALID(sig))
2201 panic("%s(): invalid signal %d", __func__, sig);
2202
2203 KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2204
2205 /*
2206 * IEEE Std 1003.1-2001: return success when killing a zombie.
2207 */
2208 if (p->p_state == PRS_ZOMBIE) {
2209 if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2210 ksiginfo_tryfree(ksi);
2211 return (ret);
2212 }
2213
2214 ps = p->p_sigacts;
2215 KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2216 prop = sigprop(sig);
2217
2218 if (td == NULL) {
2219 td = sigtd(p, sig, false);
2220 sigqueue = &p->p_sigqueue;
2221 } else
2222 sigqueue = &td->td_sigqueue;
2223
2224 SDT_PROBE3(proc, , , signal__send, td, p, sig);
2225
2226 /*
2227 * If the signal is being ignored, then we forget about it
2228 * immediately, except when the target process executes
2229 * sigwait(). (Note: we don't set SIGCONT in ps_sigignore,
2230 * and if it is set to SIG_IGN, action will be SIG_DFL here.)
2231 */
2232 mtx_lock(&ps->ps_mtx);
2233 if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2234 if (kern_sig_discard_ign &&
2235 (p->p_sysent->sv_flags & SV_SIG_DISCIGN) == 0) {
2236 SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2237
2238 mtx_unlock(&ps->ps_mtx);
2239 if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2240 ksiginfo_tryfree(ksi);
2241 return (ret);
2242 } else {
2243 action = SIG_CATCH;
2244 intrval = 0;
2245 }
2246 } else {
2247 if (SIGISMEMBER(td->td_sigmask, sig))
2248 action = SIG_HOLD;
2249 else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2250 action = SIG_CATCH;
2251 else
2252 action = SIG_DFL;
2253 if (SIGISMEMBER(ps->ps_sigintr, sig))
2254 intrval = EINTR;
2255 else
2256 intrval = ERESTART;
2257 }
2258 mtx_unlock(&ps->ps_mtx);
2259
2260 if (prop & SIGPROP_CONT)
2261 sigqueue_delete_stopmask_proc(p);
2262 else if (prop & SIGPROP_STOP) {
2263 /*
2264 * If sending a tty stop signal to a member of an orphaned
2265 * process group, discard the signal here if the action
2266 * is default; don't stop the process below if sleeping,
2267 * and don't clear any pending SIGCONT.
2268 */
2269 if ((prop & SIGPROP_TTYSTOP) != 0 &&
2270 (p->p_pgrp->pg_flags & PGRP_ORPHANED) != 0 &&
2271 action == SIG_DFL) {
2272 if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2273 ksiginfo_tryfree(ksi);
2274 return (ret);
2275 }
2276 sigqueue_delete_proc(p, SIGCONT);
2277 if (p->p_flag & P_CONTINUED) {
2278 p->p_flag &= ~P_CONTINUED;
2279 PROC_LOCK(p->p_pptr);
2280 sigqueue_take(p->p_ksi);
2281 PROC_UNLOCK(p->p_pptr);
2282 }
2283 }
2284
2285 ret = sigqueue_add(sigqueue, sig, ksi);
2286 if (ret != 0)
2287 return (ret);
2288 signotify(td);
2289 /*
2290 * Defer further processing for signals which are held,
2291 * except that stopped processes must be continued by SIGCONT.
2292 */
2293 if (action == SIG_HOLD &&
2294 !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2295 return (ret);
2296
2297 wakeup_swapper = 0;
2298
2299 /*
2300 * Some signals have a process-wide effect and a per-thread
2301 * component. Most processing occurs when the process next
2302 * tries to cross the user boundary, however there are some
2303 * times when processing needs to be done immediately, such as
2304 * waking up threads so that they can cross the user boundary.
2305 * We try to do the per-process part here.
2306 */
2307 if (P_SHOULDSTOP(p)) {
2308 KASSERT(!(p->p_flag & P_WEXIT),
2309 ("signal to stopped but exiting process"));
2310 if (sig == SIGKILL) {
2311 /*
2312 * If traced process is already stopped,
2313 * then no further action is necessary.
2314 */
2315 if (p->p_flag & P_TRACED)
2316 goto out;
2317 /*
2318 * SIGKILL sets process running.
2319 * It will die elsewhere.
2320 * All threads must be restarted.
2321 */
2322 p->p_flag &= ~P_STOPPED_SIG;
2323 goto runfast;
2324 }
2325
2326 if (prop & SIGPROP_CONT) {
2327 /*
2328 * If traced process is already stopped,
2329 * then no further action is necessary.
2330 */
2331 if (p->p_flag & P_TRACED)
2332 goto out;
2333 /*
2334 * If SIGCONT is default (or ignored), we continue the
2335 * process but don't leave the signal in sigqueue as
2336 * it has no further action. If SIGCONT is held, we
2337 * continue the process and leave the signal in
2338 * sigqueue. If the process catches SIGCONT, let it
2339 * handle the signal itself. If it isn't waiting on
2340 * an event, it goes back to run state.
2341 * Otherwise, process goes back to sleep state.
2342 */
2343 p->p_flag &= ~P_STOPPED_SIG;
2344 PROC_SLOCK(p);
2345 if (p->p_numthreads == p->p_suspcount) {
2346 PROC_SUNLOCK(p);
2347 p->p_flag |= P_CONTINUED;
2348 p->p_xsig = SIGCONT;
2349 PROC_LOCK(p->p_pptr);
2350 childproc_continued(p);
2351 PROC_UNLOCK(p->p_pptr);
2352 PROC_SLOCK(p);
2353 }
2354 if (action == SIG_DFL) {
2355 thread_unsuspend(p);
2356 PROC_SUNLOCK(p);
2357 sigqueue_delete(sigqueue, sig);
2358 goto out_cont;
2359 }
2360 if (action == SIG_CATCH) {
2361 /*
2362 * The process wants to catch it so it needs
2363 * to run at least one thread, but which one?
2364 */
2365 PROC_SUNLOCK(p);
2366 goto runfast;
2367 }
2368 /*
2369 * The signal is not ignored or caught.
2370 */
2371 thread_unsuspend(p);
2372 PROC_SUNLOCK(p);
2373 goto out_cont;
2374 }
2375
2376 if (prop & SIGPROP_STOP) {
2377 /*
2378 * If traced process is already stopped,
2379 * then no further action is necessary.
2380 */
2381 if (p->p_flag & P_TRACED)
2382 goto out;
2383 /*
2384 * Already stopped, don't need to stop again
2385 * (If we did the shell could get confused).
2386 * Just make sure the signal STOP bit set.
2387 */
2388 p->p_flag |= P_STOPPED_SIG;
2389 sigqueue_delete(sigqueue, sig);
2390 goto out;
2391 }
2392
2393 /*
2394 * All other kinds of signals:
2395 * If a thread is sleeping interruptibly, simulate a
2396 * wakeup so that when it is continued it will be made
2397 * runnable and can look at the signal. However, don't make
2398 * the PROCESS runnable, leave it stopped.
2399 * It may run a bit until it hits a thread_suspend_check().
2400 */
2401 PROC_SLOCK(p);
2402 thread_lock(td);
2403 if (TD_CAN_ABORT(td))
2404 wakeup_swapper = sig_sleepq_abort(td, intrval);
2405 else
2406 thread_unlock(td);
2407 PROC_SUNLOCK(p);
2408 goto out;
2409 /*
2410 * Mutexes are short lived. Threads waiting on them will
2411 * hit thread_suspend_check() soon.
2412 */
2413 } else if (p->p_state == PRS_NORMAL) {
2414 if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2415 tdsigwakeup(td, sig, action, intrval);
2416 goto out;
2417 }
2418
2419 MPASS(action == SIG_DFL);
2420
2421 if (prop & SIGPROP_STOP) {
2422 if (p->p_flag & (P_PPWAIT|P_WEXIT))
2423 goto out;
2424 p->p_flag |= P_STOPPED_SIG;
2425 p->p_xsig = sig;
2426 PROC_SLOCK(p);
2427 wakeup_swapper = sig_suspend_threads(td, p, 1);
2428 if (p->p_numthreads == p->p_suspcount) {
2429 /*
2430 * only thread sending signal to another
2431 * process can reach here, if thread is sending
2432 * signal to its process, because thread does
2433 * not suspend itself here, p_numthreads
2434 * should never be equal to p_suspcount.
2435 */
2436 thread_stopped(p);
2437 PROC_SUNLOCK(p);
2438 sigqueue_delete_proc(p, p->p_xsig);
2439 } else
2440 PROC_SUNLOCK(p);
2441 goto out;
2442 }
2443 } else {
2444 /* Not in "NORMAL" state. discard the signal. */
2445 sigqueue_delete(sigqueue, sig);
2446 goto out;
2447 }
2448
2449 /*
2450 * The process is not stopped so we need to apply the signal to all the
2451 * running threads.
2452 */
2453 runfast:
2454 tdsigwakeup(td, sig, action, intrval);
2455 PROC_SLOCK(p);
2456 thread_unsuspend(p);
2457 PROC_SUNLOCK(p);
2458 out_cont:
2459 itimer_proc_continue(p);
2460 kqtimer_proc_continue(p);
2461 out:
2462 /* If we jump here, proc slock should not be owned. */
2463 PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2464 if (wakeup_swapper)
2465 kick_proc0();
2466
2467 return (ret);
2468 }
2469
2470 /*
2471 * The force of a signal has been directed against a single
2472 * thread. We need to see what we can do about knocking it
2473 * out of any sleep it may be in etc.
2474 */
2475 static void
tdsigwakeup(struct thread * td,int sig,sig_t action,int intrval)2476 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2477 {
2478 struct proc *p = td->td_proc;
2479 int prop, wakeup_swapper;
2480
2481 PROC_LOCK_ASSERT(p, MA_OWNED);
2482 prop = sigprop(sig);
2483
2484 PROC_SLOCK(p);
2485 thread_lock(td);
2486 /*
2487 * Bring the priority of a thread up if we want it to get
2488 * killed in this lifetime. Be careful to avoid bumping the
2489 * priority of the idle thread, since we still allow to signal
2490 * kernel processes.
2491 */
2492 if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2493 td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2494 sched_prio(td, PUSER);
2495 if (TD_ON_SLEEPQ(td)) {
2496 /*
2497 * If thread is sleeping uninterruptibly
2498 * we can't interrupt the sleep... the signal will
2499 * be noticed when the process returns through
2500 * trap() or syscall().
2501 */
2502 if ((td->td_flags & TDF_SINTR) == 0)
2503 goto out;
2504 /*
2505 * If SIGCONT is default (or ignored) and process is
2506 * asleep, we are finished; the process should not
2507 * be awakened.
2508 */
2509 if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2510 thread_unlock(td);
2511 PROC_SUNLOCK(p);
2512 sigqueue_delete(&p->p_sigqueue, sig);
2513 /*
2514 * It may be on either list in this state.
2515 * Remove from both for now.
2516 */
2517 sigqueue_delete(&td->td_sigqueue, sig);
2518 return;
2519 }
2520
2521 /*
2522 * Don't awaken a sleeping thread for SIGSTOP if the
2523 * STOP signal is deferred.
2524 */
2525 if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2526 TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2527 goto out;
2528
2529 /*
2530 * Give low priority threads a better chance to run.
2531 */
2532 if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2533 sched_prio(td, PUSER);
2534
2535 wakeup_swapper = sig_sleepq_abort(td, intrval);
2536 PROC_SUNLOCK(p);
2537 if (wakeup_swapper)
2538 kick_proc0();
2539 return;
2540 }
2541
2542 /*
2543 * Other states do nothing with the signal immediately,
2544 * other than kicking ourselves if we are running.
2545 * It will either never be noticed, or noticed very soon.
2546 */
2547 #ifdef SMP
2548 if (TD_IS_RUNNING(td) && td != curthread)
2549 forward_signal(td);
2550 #endif
2551
2552 out:
2553 PROC_SUNLOCK(p);
2554 thread_unlock(td);
2555 }
2556
2557 static void
ptrace_coredumpreq(struct thread * td,struct proc * p,struct thr_coredump_req * tcq)2558 ptrace_coredumpreq(struct thread *td, struct proc *p,
2559 struct thr_coredump_req *tcq)
2560 {
2561 void *rl_cookie;
2562
2563 if (p->p_sysent->sv_coredump == NULL) {
2564 tcq->tc_error = ENOSYS;
2565 return;
2566 }
2567
2568 rl_cookie = vn_rangelock_wlock(tcq->tc_vp, 0, OFF_MAX);
2569 tcq->tc_error = p->p_sysent->sv_coredump(td, tcq->tc_vp,
2570 tcq->tc_limit, tcq->tc_flags);
2571 vn_rangelock_unlock(tcq->tc_vp, rl_cookie);
2572 }
2573
2574 static void
ptrace_syscallreq(struct thread * td,struct proc * p,struct thr_syscall_req * tsr)2575 ptrace_syscallreq(struct thread *td, struct proc *p,
2576 struct thr_syscall_req *tsr)
2577 {
2578 struct sysentvec *sv;
2579 struct sysent *se;
2580 register_t rv_saved[2];
2581 int error, nerror;
2582 int sc;
2583 bool audited, sy_thr_static;
2584
2585 sv = p->p_sysent;
2586 if (sv->sv_table == NULL || sv->sv_size < tsr->ts_sa.code) {
2587 tsr->ts_ret.sr_error = ENOSYS;
2588 return;
2589 }
2590
2591 sc = tsr->ts_sa.code;
2592 if (sc == SYS_syscall || sc == SYS___syscall) {
2593 sc = tsr->ts_sa.args[0];
2594 memmove(&tsr->ts_sa.args[0], &tsr->ts_sa.args[1],
2595 sizeof(register_t) * (tsr->ts_nargs - 1));
2596 }
2597
2598 tsr->ts_sa.callp = se = &sv->sv_table[sc];
2599
2600 VM_CNT_INC(v_syscall);
2601 td->td_pticks = 0;
2602 if (__predict_false(td->td_cowgen != atomic_load_int(
2603 &td->td_proc->p_cowgen)))
2604 thread_cow_update(td);
2605
2606 #ifdef CAPABILITY_MODE
2607 if (IN_CAPABILITY_MODE(td) && (se->sy_flags & SYF_CAPENABLED) == 0) {
2608 tsr->ts_ret.sr_error = ECAPMODE;
2609 return;
2610 }
2611 #endif
2612
2613 sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2614 audited = AUDIT_SYSCALL_ENTER(sc, td) != 0;
2615
2616 if (!sy_thr_static) {
2617 error = syscall_thread_enter(td, &se);
2618 sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2619 if (error != 0) {
2620 tsr->ts_ret.sr_error = error;
2621 return;
2622 }
2623 }
2624
2625 rv_saved[0] = td->td_retval[0];
2626 rv_saved[1] = td->td_retval[1];
2627 nerror = td->td_errno;
2628 td->td_retval[0] = 0;
2629 td->td_retval[1] = 0;
2630
2631 #ifdef KDTRACE_HOOKS
2632 if (se->sy_entry != 0)
2633 (*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_ENTRY, 0);
2634 #endif
2635 tsr->ts_ret.sr_error = se->sy_call(td, tsr->ts_sa.args);
2636 #ifdef KDTRACE_HOOKS
2637 if (se->sy_return != 0)
2638 (*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_RETURN,
2639 tsr->ts_ret.sr_error != 0 ? -1 : td->td_retval[0]);
2640 #endif
2641
2642 tsr->ts_ret.sr_retval[0] = td->td_retval[0];
2643 tsr->ts_ret.sr_retval[1] = td->td_retval[1];
2644 td->td_retval[0] = rv_saved[0];
2645 td->td_retval[1] = rv_saved[1];
2646 td->td_errno = nerror;
2647
2648 if (audited)
2649 AUDIT_SYSCALL_EXIT(error, td);
2650 if (!sy_thr_static)
2651 syscall_thread_exit(td, se);
2652 }
2653
2654 static void
ptrace_remotereq(struct thread * td,int flag)2655 ptrace_remotereq(struct thread *td, int flag)
2656 {
2657 struct proc *p;
2658
2659 MPASS(td == curthread);
2660 p = td->td_proc;
2661 PROC_LOCK_ASSERT(p, MA_OWNED);
2662 if ((td->td_dbgflags & flag) == 0)
2663 return;
2664 KASSERT((p->p_flag & P_STOPPED_TRACE) != 0, ("not stopped"));
2665 KASSERT(td->td_remotereq != NULL, ("td_remotereq is NULL"));
2666
2667 PROC_UNLOCK(p);
2668 switch (flag) {
2669 case TDB_COREDUMPREQ:
2670 ptrace_coredumpreq(td, p, td->td_remotereq);
2671 break;
2672 case TDB_SCREMOTEREQ:
2673 ptrace_syscallreq(td, p, td->td_remotereq);
2674 break;
2675 default:
2676 __unreachable();
2677 }
2678 PROC_LOCK(p);
2679
2680 MPASS((td->td_dbgflags & flag) != 0);
2681 td->td_dbgflags &= ~flag;
2682 td->td_remotereq = NULL;
2683 wakeup(p);
2684 }
2685
2686 static int
sig_suspend_threads(struct thread * td,struct proc * p,int sending)2687 sig_suspend_threads(struct thread *td, struct proc *p, int sending)
2688 {
2689 struct thread *td2;
2690 int wakeup_swapper;
2691
2692 PROC_LOCK_ASSERT(p, MA_OWNED);
2693 PROC_SLOCK_ASSERT(p, MA_OWNED);
2694 MPASS(sending || td == curthread);
2695
2696 wakeup_swapper = 0;
2697 FOREACH_THREAD_IN_PROC(p, td2) {
2698 thread_lock(td2);
2699 td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
2700 if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2701 (td2->td_flags & TDF_SINTR)) {
2702 if (td2->td_flags & TDF_SBDRY) {
2703 /*
2704 * Once a thread is asleep with
2705 * TDF_SBDRY and without TDF_SERESTART
2706 * or TDF_SEINTR set, it should never
2707 * become suspended due to this check.
2708 */
2709 KASSERT(!TD_IS_SUSPENDED(td2),
2710 ("thread with deferred stops suspended"));
2711 if (TD_SBDRY_INTR(td2)) {
2712 wakeup_swapper |= sleepq_abort(td2,
2713 TD_SBDRY_ERRNO(td2));
2714 continue;
2715 }
2716 } else if (!TD_IS_SUSPENDED(td2))
2717 thread_suspend_one(td2);
2718 } else if (!TD_IS_SUSPENDED(td2)) {
2719 if (sending || td != td2)
2720 td2->td_flags |= TDF_ASTPENDING;
2721 #ifdef SMP
2722 if (TD_IS_RUNNING(td2) && td2 != td)
2723 forward_signal(td2);
2724 #endif
2725 }
2726 thread_unlock(td2);
2727 }
2728 return (wakeup_swapper);
2729 }
2730
2731 /*
2732 * Stop the process for an event deemed interesting to the debugger. If si is
2733 * non-NULL, this is a signal exchange; the new signal requested by the
2734 * debugger will be returned for handling. If si is NULL, this is some other
2735 * type of interesting event. The debugger may request a signal be delivered in
2736 * that case as well, however it will be deferred until it can be handled.
2737 */
2738 int
ptracestop(struct thread * td,int sig,ksiginfo_t * si)2739 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2740 {
2741 struct proc *p = td->td_proc;
2742 struct thread *td2;
2743 ksiginfo_t ksi;
2744
2745 PROC_LOCK_ASSERT(p, MA_OWNED);
2746 KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2747 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2748 &p->p_mtx.lock_object, "Stopping for traced signal");
2749
2750 td->td_xsig = sig;
2751
2752 if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2753 td->td_dbgflags |= TDB_XSIG;
2754 CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2755 td->td_tid, p->p_pid, td->td_dbgflags, sig);
2756 PROC_SLOCK(p);
2757 while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2758 if (P_KILLED(p)) {
2759 /*
2760 * Ensure that, if we've been PT_KILLed, the
2761 * exit status reflects that. Another thread
2762 * may also be in ptracestop(), having just
2763 * received the SIGKILL, but this thread was
2764 * unsuspended first.
2765 */
2766 td->td_dbgflags &= ~TDB_XSIG;
2767 td->td_xsig = SIGKILL;
2768 p->p_ptevents = 0;
2769 break;
2770 }
2771 if (p->p_flag & P_SINGLE_EXIT &&
2772 !(td->td_dbgflags & TDB_EXIT)) {
2773 /*
2774 * Ignore ptrace stops except for thread exit
2775 * events when the process exits.
2776 */
2777 td->td_dbgflags &= ~TDB_XSIG;
2778 PROC_SUNLOCK(p);
2779 return (0);
2780 }
2781
2782 /*
2783 * Make wait(2) work. Ensure that right after the
2784 * attach, the thread which was decided to become the
2785 * leader of attach gets reported to the waiter.
2786 * Otherwise, just avoid overwriting another thread's
2787 * assignment to p_xthread. If another thread has
2788 * already set p_xthread, the current thread will get
2789 * a chance to report itself upon the next iteration.
2790 */
2791 if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2792 ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2793 p->p_xthread == NULL)) {
2794 p->p_xsig = sig;
2795 p->p_xthread = td;
2796
2797 /*
2798 * If we are on sleepqueue already,
2799 * let sleepqueue code decide if it
2800 * needs to go sleep after attach.
2801 */
2802 if (td->td_wchan == NULL)
2803 td->td_dbgflags &= ~TDB_FSTP;
2804
2805 p->p_flag2 &= ~P2_PTRACE_FSTP;
2806 p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2807 sig_suspend_threads(td, p, 0);
2808 }
2809 if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2810 td->td_dbgflags &= ~TDB_STOPATFORK;
2811 }
2812 stopme:
2813 td->td_dbgflags |= TDB_SSWITCH;
2814 thread_suspend_switch(td, p);
2815 td->td_dbgflags &= ~TDB_SSWITCH;
2816 if ((td->td_dbgflags & (TDB_COREDUMPREQ |
2817 TDB_SCREMOTEREQ)) != 0) {
2818 MPASS((td->td_dbgflags & (TDB_COREDUMPREQ |
2819 TDB_SCREMOTEREQ)) !=
2820 (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2821 PROC_SUNLOCK(p);
2822 ptrace_remotereq(td, td->td_dbgflags &
2823 (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2824 PROC_SLOCK(p);
2825 goto stopme;
2826 }
2827 if (p->p_xthread == td)
2828 p->p_xthread = NULL;
2829 if (!(p->p_flag & P_TRACED))
2830 break;
2831 if (td->td_dbgflags & TDB_SUSPEND) {
2832 if (p->p_flag & P_SINGLE_EXIT)
2833 break;
2834 goto stopme;
2835 }
2836 }
2837 PROC_SUNLOCK(p);
2838 }
2839
2840 if (si != NULL && sig == td->td_xsig) {
2841 /* Parent wants us to take the original signal unchanged. */
2842 si->ksi_flags |= KSI_HEAD;
2843 if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2844 si->ksi_signo = 0;
2845 } else if (td->td_xsig != 0) {
2846 /*
2847 * If parent wants us to take a new signal, then it will leave
2848 * it in td->td_xsig; otherwise we just look for signals again.
2849 */
2850 ksiginfo_init(&ksi);
2851 ksi.ksi_signo = td->td_xsig;
2852 ksi.ksi_flags |= KSI_PTRACE;
2853 td2 = sigtd(p, td->td_xsig, false);
2854 tdsendsignal(p, td2, td->td_xsig, &ksi);
2855 if (td != td2)
2856 return (0);
2857 }
2858
2859 return (td->td_xsig);
2860 }
2861
2862 static void
reschedule_signals(struct proc * p,sigset_t block,int flags)2863 reschedule_signals(struct proc *p, sigset_t block, int flags)
2864 {
2865 struct sigacts *ps;
2866 struct thread *td;
2867 int sig;
2868 bool fastblk, pslocked;
2869
2870 PROC_LOCK_ASSERT(p, MA_OWNED);
2871 ps = p->p_sigacts;
2872 pslocked = (flags & SIGPROCMASK_PS_LOCKED) != 0;
2873 mtx_assert(&ps->ps_mtx, pslocked ? MA_OWNED : MA_NOTOWNED);
2874 if (SIGISEMPTY(p->p_siglist))
2875 return;
2876 SIGSETAND(block, p->p_siglist);
2877 fastblk = (flags & SIGPROCMASK_FASTBLK) != 0;
2878 SIG_FOREACH(sig, &block) {
2879 td = sigtd(p, sig, fastblk);
2880
2881 /*
2882 * If sigtd() selected us despite sigfastblock is
2883 * blocking, do not activate AST or wake us, to avoid
2884 * loop in AST handler.
2885 */
2886 if (fastblk && td == curthread)
2887 continue;
2888
2889 signotify(td);
2890 if (!pslocked)
2891 mtx_lock(&ps->ps_mtx);
2892 if (p->p_flag & P_TRACED ||
2893 (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2894 !SIGISMEMBER(td->td_sigmask, sig))) {
2895 tdsigwakeup(td, sig, SIG_CATCH,
2896 (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2897 ERESTART));
2898 }
2899 if (!pslocked)
2900 mtx_unlock(&ps->ps_mtx);
2901 }
2902 }
2903
2904 void
tdsigcleanup(struct thread * td)2905 tdsigcleanup(struct thread *td)
2906 {
2907 struct proc *p;
2908 sigset_t unblocked;
2909
2910 p = td->td_proc;
2911 PROC_LOCK_ASSERT(p, MA_OWNED);
2912
2913 sigqueue_flush(&td->td_sigqueue);
2914 if (p->p_numthreads == 1)
2915 return;
2916
2917 /*
2918 * Since we cannot handle signals, notify signal post code
2919 * about this by filling the sigmask.
2920 *
2921 * Also, if needed, wake up thread(s) that do not block the
2922 * same signals as the exiting thread, since the thread might
2923 * have been selected for delivery and woken up.
2924 */
2925 SIGFILLSET(unblocked);
2926 SIGSETNAND(unblocked, td->td_sigmask);
2927 SIGFILLSET(td->td_sigmask);
2928 reschedule_signals(p, unblocked, 0);
2929
2930 }
2931
2932 static int
sigdeferstop_curr_flags(int cflags)2933 sigdeferstop_curr_flags(int cflags)
2934 {
2935
2936 MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
2937 (cflags & TDF_SBDRY) != 0);
2938 return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
2939 }
2940
2941 /*
2942 * Defer the delivery of SIGSTOP for the current thread, according to
2943 * the requested mode. Returns previous flags, which must be restored
2944 * by sigallowstop().
2945 *
2946 * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
2947 * cleared by the current thread, which allow the lock-less read-only
2948 * accesses below.
2949 */
2950 int
sigdeferstop_impl(int mode)2951 sigdeferstop_impl(int mode)
2952 {
2953 struct thread *td;
2954 int cflags, nflags;
2955
2956 td = curthread;
2957 cflags = sigdeferstop_curr_flags(td->td_flags);
2958 switch (mode) {
2959 case SIGDEFERSTOP_NOP:
2960 nflags = cflags;
2961 break;
2962 case SIGDEFERSTOP_OFF:
2963 nflags = 0;
2964 break;
2965 case SIGDEFERSTOP_SILENT:
2966 nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
2967 break;
2968 case SIGDEFERSTOP_EINTR:
2969 nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
2970 break;
2971 case SIGDEFERSTOP_ERESTART:
2972 nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
2973 break;
2974 default:
2975 panic("sigdeferstop: invalid mode %x", mode);
2976 break;
2977 }
2978 if (cflags == nflags)
2979 return (SIGDEFERSTOP_VAL_NCHG);
2980 thread_lock(td);
2981 td->td_flags = (td->td_flags & ~cflags) | nflags;
2982 thread_unlock(td);
2983 return (cflags);
2984 }
2985
2986 /*
2987 * Restores the STOP handling mode, typically permitting the delivery
2988 * of SIGSTOP for the current thread. This does not immediately
2989 * suspend if a stop was posted. Instead, the thread will suspend
2990 * either via ast() or a subsequent interruptible sleep.
2991 */
2992 void
sigallowstop_impl(int prev)2993 sigallowstop_impl(int prev)
2994 {
2995 struct thread *td;
2996 int cflags;
2997
2998 KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
2999 KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
3000 ("sigallowstop: incorrect previous mode %x", prev));
3001 td = curthread;
3002 cflags = sigdeferstop_curr_flags(td->td_flags);
3003 if (cflags != prev) {
3004 thread_lock(td);
3005 td->td_flags = (td->td_flags & ~cflags) | prev;
3006 thread_unlock(td);
3007 }
3008 }
3009
3010 enum sigstatus {
3011 SIGSTATUS_HANDLE,
3012 SIGSTATUS_HANDLED,
3013 SIGSTATUS_IGNORE,
3014 SIGSTATUS_SBDRY_STOP,
3015 };
3016
3017 /*
3018 * The thread has signal "sig" pending. Figure out what to do with it:
3019 *
3020 * _HANDLE -> the caller should handle the signal
3021 * _HANDLED -> handled internally, reload pending signal set
3022 * _IGNORE -> ignored, remove from the set of pending signals and try the
3023 * next pending signal
3024 * _SBDRY_STOP -> the signal should stop the thread but this is not
3025 * permitted in the current context
3026 */
3027 static enum sigstatus
sigprocess(struct thread * td,int sig)3028 sigprocess(struct thread *td, int sig)
3029 {
3030 struct proc *p;
3031 struct sigacts *ps;
3032 struct sigqueue *queue;
3033 ksiginfo_t ksi;
3034 int prop;
3035
3036 KASSERT(_SIG_VALID(sig), ("%s: invalid signal %d", __func__, sig));
3037
3038 p = td->td_proc;
3039 ps = p->p_sigacts;
3040 mtx_assert(&ps->ps_mtx, MA_OWNED);
3041 PROC_LOCK_ASSERT(p, MA_OWNED);
3042
3043 /*
3044 * We should allow pending but ignored signals below
3045 * if there is sigwait() active, or P_TRACED was
3046 * on when they were posted.
3047 */
3048 if (SIGISMEMBER(ps->ps_sigignore, sig) &&
3049 (p->p_flag & P_TRACED) == 0 &&
3050 (td->td_flags & TDF_SIGWAIT) == 0) {
3051 return (SIGSTATUS_IGNORE);
3052 }
3053
3054 /*
3055 * If the process is going to single-thread mode to prepare
3056 * for exit, there is no sense in delivering any signal
3057 * to usermode. Another important consequence is that
3058 * msleep(..., PCATCH, ...) now is only interruptible by a
3059 * suspend request.
3060 */
3061 if ((p->p_flag2 & P2_WEXIT) != 0)
3062 return (SIGSTATUS_IGNORE);
3063
3064 if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
3065 /*
3066 * If traced, always stop.
3067 * Remove old signal from queue before the stop.
3068 * XXX shrug off debugger, it causes siginfo to
3069 * be thrown away.
3070 */
3071 queue = &td->td_sigqueue;
3072 ksiginfo_init(&ksi);
3073 if (sigqueue_get(queue, sig, &ksi) == 0) {
3074 queue = &p->p_sigqueue;
3075 sigqueue_get(queue, sig, &ksi);
3076 }
3077 td->td_si = ksi.ksi_info;
3078
3079 mtx_unlock(&ps->ps_mtx);
3080 sig = ptracestop(td, sig, &ksi);
3081 mtx_lock(&ps->ps_mtx);
3082
3083 td->td_si.si_signo = 0;
3084
3085 /*
3086 * Keep looking if the debugger discarded or
3087 * replaced the signal.
3088 */
3089 if (sig == 0)
3090 return (SIGSTATUS_HANDLED);
3091
3092 /*
3093 * If the signal became masked, re-queue it.
3094 */
3095 if (SIGISMEMBER(td->td_sigmask, sig)) {
3096 ksi.ksi_flags |= KSI_HEAD;
3097 sigqueue_add(&p->p_sigqueue, sig, &ksi);
3098 return (SIGSTATUS_HANDLED);
3099 }
3100
3101 /*
3102 * If the traced bit got turned off, requeue the signal and
3103 * reload the set of pending signals. This ensures that p_sig*
3104 * and p_sigact are consistent.
3105 */
3106 if ((p->p_flag & P_TRACED) == 0) {
3107 if ((ksi.ksi_flags & KSI_PTRACE) == 0) {
3108 ksi.ksi_flags |= KSI_HEAD;
3109 sigqueue_add(queue, sig, &ksi);
3110 }
3111 return (SIGSTATUS_HANDLED);
3112 }
3113 }
3114
3115 /*
3116 * Decide whether the signal should be returned.
3117 * Return the signal's number, or fall through
3118 * to clear it from the pending mask.
3119 */
3120 switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
3121 case (intptr_t)SIG_DFL:
3122 /*
3123 * Don't take default actions on system processes.
3124 */
3125 if (p->p_pid <= 1) {
3126 #ifdef DIAGNOSTIC
3127 /*
3128 * Are you sure you want to ignore SIGSEGV
3129 * in init? XXX
3130 */
3131 printf("Process (pid %lu) got signal %d\n",
3132 (u_long)p->p_pid, sig);
3133 #endif
3134 return (SIGSTATUS_IGNORE);
3135 }
3136
3137 /*
3138 * If there is a pending stop signal to process with
3139 * default action, stop here, then clear the signal.
3140 * Traced or exiting processes should ignore stops.
3141 * Additionally, a member of an orphaned process group
3142 * should ignore tty stops.
3143 */
3144 prop = sigprop(sig);
3145 if (prop & SIGPROP_STOP) {
3146 mtx_unlock(&ps->ps_mtx);
3147 if ((p->p_flag & (P_TRACED | P_WEXIT |
3148 P_SINGLE_EXIT)) != 0 || ((p->p_pgrp->
3149 pg_flags & PGRP_ORPHANED) != 0 &&
3150 (prop & SIGPROP_TTYSTOP) != 0)) {
3151 mtx_lock(&ps->ps_mtx);
3152 return (SIGSTATUS_IGNORE);
3153 }
3154 if (TD_SBDRY_INTR(td)) {
3155 KASSERT((td->td_flags & TDF_SBDRY) != 0,
3156 ("lost TDF_SBDRY"));
3157 mtx_lock(&ps->ps_mtx);
3158 return (SIGSTATUS_SBDRY_STOP);
3159 }
3160 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
3161 &p->p_mtx.lock_object, "Catching SIGSTOP");
3162 sigqueue_delete(&td->td_sigqueue, sig);
3163 sigqueue_delete(&p->p_sigqueue, sig);
3164 p->p_flag |= P_STOPPED_SIG;
3165 p->p_xsig = sig;
3166 PROC_SLOCK(p);
3167 sig_suspend_threads(td, p, 0);
3168 thread_suspend_switch(td, p);
3169 PROC_SUNLOCK(p);
3170 mtx_lock(&ps->ps_mtx);
3171 return (SIGSTATUS_HANDLED);
3172 } else if ((prop & SIGPROP_IGNORE) != 0 &&
3173 (td->td_flags & TDF_SIGWAIT) == 0) {
3174 /*
3175 * Default action is to ignore; drop it if
3176 * not in kern_sigtimedwait().
3177 */
3178 return (SIGSTATUS_IGNORE);
3179 } else {
3180 return (SIGSTATUS_HANDLE);
3181 }
3182
3183 case (intptr_t)SIG_IGN:
3184 if ((td->td_flags & TDF_SIGWAIT) == 0)
3185 return (SIGSTATUS_IGNORE);
3186 else
3187 return (SIGSTATUS_HANDLE);
3188
3189 default:
3190 /*
3191 * This signal has an action, let postsig() process it.
3192 */
3193 return (SIGSTATUS_HANDLE);
3194 }
3195 }
3196
3197 /*
3198 * If the current process has received a signal (should be caught or cause
3199 * termination, should interrupt current syscall), return the signal number.
3200 * Stop signals with default action are processed immediately, then cleared;
3201 * they aren't returned. This is checked after each entry to the system for
3202 * a syscall or trap (though this can usually be done without calling
3203 * issignal by checking the pending signal masks in cursig.) The normal call
3204 * sequence is
3205 *
3206 * while (sig = cursig(curthread))
3207 * postsig(sig);
3208 */
3209 static int
issignal(struct thread * td)3210 issignal(struct thread *td)
3211 {
3212 struct proc *p;
3213 sigset_t sigpending;
3214 int sig;
3215
3216 p = td->td_proc;
3217 PROC_LOCK_ASSERT(p, MA_OWNED);
3218
3219 for (;;) {
3220 sigpending = td->td_sigqueue.sq_signals;
3221 SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
3222 SIGSETNAND(sigpending, td->td_sigmask);
3223
3224 if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
3225 (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
3226 SIG_STOPSIGMASK(sigpending);
3227 if (SIGISEMPTY(sigpending)) /* no signal to send */
3228 return (0);
3229
3230 /*
3231 * Do fast sigblock if requested by usermode. Since
3232 * we do know that there was a signal pending at this
3233 * point, set the FAST_SIGBLOCK_PEND as indicator for
3234 * usermode to perform a dummy call to
3235 * FAST_SIGBLOCK_UNBLOCK, which causes immediate
3236 * delivery of postponed pending signal.
3237 */
3238 if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
3239 if (td->td_sigblock_val != 0)
3240 SIGSETNAND(sigpending, fastblock_mask);
3241 if (SIGISEMPTY(sigpending)) {
3242 td->td_pflags |= TDP_SIGFASTPENDING;
3243 return (0);
3244 }
3245 }
3246
3247 if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
3248 (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
3249 SIGISMEMBER(sigpending, SIGSTOP)) {
3250 /*
3251 * If debugger just attached, always consume
3252 * SIGSTOP from ptrace(PT_ATTACH) first, to
3253 * execute the debugger attach ritual in
3254 * order.
3255 */
3256 td->td_dbgflags |= TDB_FSTP;
3257 SIGEMPTYSET(sigpending);
3258 SIGADDSET(sigpending, SIGSTOP);
3259 }
3260
3261 SIG_FOREACH(sig, &sigpending) {
3262 switch (sigprocess(td, sig)) {
3263 case SIGSTATUS_HANDLE:
3264 return (sig);
3265 case SIGSTATUS_HANDLED:
3266 goto next;
3267 case SIGSTATUS_IGNORE:
3268 sigqueue_delete(&td->td_sigqueue, sig);
3269 sigqueue_delete(&p->p_sigqueue, sig);
3270 break;
3271 case SIGSTATUS_SBDRY_STOP:
3272 return (-1);
3273 }
3274 }
3275 next:;
3276 }
3277 }
3278
3279 void
thread_stopped(struct proc * p)3280 thread_stopped(struct proc *p)
3281 {
3282 int n;
3283
3284 PROC_LOCK_ASSERT(p, MA_OWNED);
3285 PROC_SLOCK_ASSERT(p, MA_OWNED);
3286 n = p->p_suspcount;
3287 if (p == curproc)
3288 n++;
3289 if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3290 PROC_SUNLOCK(p);
3291 p->p_flag &= ~P_WAITED;
3292 PROC_LOCK(p->p_pptr);
3293 childproc_stopped(p, (p->p_flag & P_TRACED) ?
3294 CLD_TRAPPED : CLD_STOPPED);
3295 PROC_UNLOCK(p->p_pptr);
3296 PROC_SLOCK(p);
3297 }
3298 }
3299
3300 /*
3301 * Take the action for the specified signal
3302 * from the current set of pending signals.
3303 */
3304 int
postsig(int sig)3305 postsig(int sig)
3306 {
3307 struct thread *td;
3308 struct proc *p;
3309 struct sigacts *ps;
3310 sig_t action;
3311 ksiginfo_t ksi;
3312 sigset_t returnmask;
3313
3314 KASSERT(sig != 0, ("postsig"));
3315
3316 td = curthread;
3317 p = td->td_proc;
3318 PROC_LOCK_ASSERT(p, MA_OWNED);
3319 ps = p->p_sigacts;
3320 mtx_assert(&ps->ps_mtx, MA_OWNED);
3321 ksiginfo_init(&ksi);
3322 if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3323 sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3324 return (0);
3325 ksi.ksi_signo = sig;
3326 if (ksi.ksi_code == SI_TIMER)
3327 itimer_accept(p, ksi.ksi_timerid, &ksi);
3328 action = ps->ps_sigact[_SIG_IDX(sig)];
3329 #ifdef KTRACE
3330 if (KTRPOINT(td, KTR_PSIG))
3331 ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3332 &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3333 #endif
3334
3335 if (action == SIG_DFL) {
3336 /*
3337 * Default action, where the default is to kill
3338 * the process. (Other cases were ignored above.)
3339 */
3340 mtx_unlock(&ps->ps_mtx);
3341 proc_td_siginfo_capture(td, &ksi.ksi_info);
3342 sigexit(td, sig);
3343 /* NOTREACHED */
3344 } else {
3345 /*
3346 * If we get here, the signal must be caught.
3347 */
3348 KASSERT(action != SIG_IGN, ("postsig action %p", action));
3349 KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3350 ("postsig action: blocked sig %d", sig));
3351
3352 /*
3353 * Set the new mask value and also defer further
3354 * occurrences of this signal.
3355 *
3356 * Special case: user has done a sigsuspend. Here the
3357 * current mask is not of interest, but rather the
3358 * mask from before the sigsuspend is what we want
3359 * restored after the signal processing is completed.
3360 */
3361 if (td->td_pflags & TDP_OLDMASK) {
3362 returnmask = td->td_oldsigmask;
3363 td->td_pflags &= ~TDP_OLDMASK;
3364 } else
3365 returnmask = td->td_sigmask;
3366
3367 if (p->p_sig == sig) {
3368 p->p_sig = 0;
3369 }
3370 (*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3371 postsig_done(sig, td, ps);
3372 }
3373 return (1);
3374 }
3375
3376 int
sig_ast_checksusp(struct thread * td)3377 sig_ast_checksusp(struct thread *td)
3378 {
3379 struct proc *p __diagused;
3380 int ret;
3381
3382 p = td->td_proc;
3383 PROC_LOCK_ASSERT(p, MA_OWNED);
3384
3385 if ((td->td_flags & TDF_NEEDSUSPCHK) == 0)
3386 return (0);
3387
3388 ret = thread_suspend_check(1);
3389 MPASS(ret == 0 || ret == EINTR || ret == ERESTART);
3390 return (ret);
3391 }
3392
3393 int
sig_ast_needsigchk(struct thread * td)3394 sig_ast_needsigchk(struct thread *td)
3395 {
3396 struct proc *p;
3397 struct sigacts *ps;
3398 int ret, sig;
3399
3400 p = td->td_proc;
3401 PROC_LOCK_ASSERT(p, MA_OWNED);
3402
3403 if ((td->td_flags & TDF_NEEDSIGCHK) == 0)
3404 return (0);
3405
3406 ps = p->p_sigacts;
3407 mtx_lock(&ps->ps_mtx);
3408 sig = cursig(td);
3409 if (sig == -1) {
3410 mtx_unlock(&ps->ps_mtx);
3411 KASSERT((td->td_flags & TDF_SBDRY) != 0, ("lost TDF_SBDRY"));
3412 KASSERT(TD_SBDRY_INTR(td),
3413 ("lost TDF_SERESTART of TDF_SEINTR"));
3414 KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
3415 (TDF_SEINTR | TDF_SERESTART),
3416 ("both TDF_SEINTR and TDF_SERESTART"));
3417 ret = TD_SBDRY_ERRNO(td);
3418 } else if (sig != 0) {
3419 ret = SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR : ERESTART;
3420 mtx_unlock(&ps->ps_mtx);
3421 } else {
3422 mtx_unlock(&ps->ps_mtx);
3423 ret = 0;
3424 }
3425
3426 /*
3427 * Do not go into sleep if this thread was the ptrace(2)
3428 * attach leader. cursig() consumed SIGSTOP from PT_ATTACH,
3429 * but we usually act on the signal by interrupting sleep, and
3430 * should do that here as well.
3431 */
3432 if ((td->td_dbgflags & TDB_FSTP) != 0) {
3433 if (ret == 0)
3434 ret = EINTR;
3435 td->td_dbgflags &= ~TDB_FSTP;
3436 }
3437
3438 return (ret);
3439 }
3440
3441 int
sig_intr(void)3442 sig_intr(void)
3443 {
3444 struct thread *td;
3445 struct proc *p;
3446 int ret;
3447
3448 td = curthread;
3449 if ((td->td_flags & (TDF_NEEDSIGCHK | TDF_NEEDSUSPCHK)) == 0)
3450 return (0);
3451
3452 p = td->td_proc;
3453
3454 PROC_LOCK(p);
3455 ret = sig_ast_checksusp(td);
3456 if (ret == 0)
3457 ret = sig_ast_needsigchk(td);
3458 PROC_UNLOCK(p);
3459 return (ret);
3460 }
3461
3462 bool
curproc_sigkilled(void)3463 curproc_sigkilled(void)
3464 {
3465 struct thread *td;
3466 struct proc *p;
3467 struct sigacts *ps;
3468 bool res;
3469
3470 td = curthread;
3471 if ((td->td_flags & TDF_NEEDSIGCHK) == 0)
3472 return (false);
3473
3474 p = td->td_proc;
3475 PROC_LOCK(p);
3476 ps = p->p_sigacts;
3477 mtx_lock(&ps->ps_mtx);
3478 res = SIGISMEMBER(td->td_sigqueue.sq_signals, SIGKILL) ||
3479 SIGISMEMBER(p->p_sigqueue.sq_signals, SIGKILL);
3480 mtx_unlock(&ps->ps_mtx);
3481 PROC_UNLOCK(p);
3482 return (res);
3483 }
3484
3485 void
proc_wkilled(struct proc * p)3486 proc_wkilled(struct proc *p)
3487 {
3488
3489 PROC_LOCK_ASSERT(p, MA_OWNED);
3490 if ((p->p_flag & P_WKILLED) == 0) {
3491 p->p_flag |= P_WKILLED;
3492 /*
3493 * Notify swapper that there is a process to swap in.
3494 * The notification is racy, at worst it would take 10
3495 * seconds for the swapper process to notice.
3496 */
3497 if ((p->p_flag & (P_INMEM | P_SWAPPINGIN)) == 0)
3498 wakeup(&proc0);
3499 }
3500 }
3501
3502 /*
3503 * Kill the current process for stated reason.
3504 */
3505 void
killproc(struct proc * p,const char * why)3506 killproc(struct proc *p, const char *why)
3507 {
3508
3509 PROC_LOCK_ASSERT(p, MA_OWNED);
3510 CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3511 p->p_comm);
3512 log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3513 p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3514 p->p_ucred->cr_uid, why);
3515 proc_wkilled(p);
3516 kern_psignal(p, SIGKILL);
3517 }
3518
3519 /*
3520 * Force the current process to exit with the specified signal, dumping core
3521 * if appropriate. We bypass the normal tests for masked and caught signals,
3522 * allowing unrecoverable failures to terminate the process without changing
3523 * signal state. Mark the accounting record with the signal termination.
3524 * If dumping core, save the signal number for the debugger. Calls exit and
3525 * does not return.
3526 */
3527 void
sigexit(struct thread * td,int sig)3528 sigexit(struct thread *td, int sig)
3529 {
3530 struct proc *p = td->td_proc;
3531
3532 PROC_LOCK_ASSERT(p, MA_OWNED);
3533 proc_set_p2_wexit(p);
3534
3535 p->p_acflag |= AXSIG;
3536 /*
3537 * We must be single-threading to generate a core dump. This
3538 * ensures that the registers in the core file are up-to-date.
3539 * Also, the ELF dump handler assumes that the thread list doesn't
3540 * change out from under it.
3541 *
3542 * XXX If another thread attempts to single-thread before us
3543 * (e.g. via fork()), we won't get a dump at all.
3544 */
3545 if ((sigprop(sig) & SIGPROP_CORE) &&
3546 thread_single(p, SINGLE_NO_EXIT) == 0) {
3547 p->p_sig = sig;
3548 /*
3549 * Log signals which would cause core dumps
3550 * (Log as LOG_INFO to appease those who don't want
3551 * these messages.)
3552 * XXX : Todo, as well as euid, write out ruid too
3553 * Note that coredump() drops proc lock.
3554 */
3555 if (coredump(td) == 0)
3556 sig |= WCOREFLAG;
3557 if (kern_logsigexit)
3558 log(LOG_INFO,
3559 "pid %d (%s), jid %d, uid %d: exited on "
3560 "signal %d%s\n", p->p_pid, p->p_comm,
3561 p->p_ucred->cr_prison->pr_id,
3562 td->td_ucred->cr_uid,
3563 sig &~ WCOREFLAG,
3564 sig & WCOREFLAG ? " (core dumped)" : "");
3565 } else
3566 PROC_UNLOCK(p);
3567 exit1(td, 0, sig);
3568 /* NOTREACHED */
3569 }
3570
3571 /*
3572 * Send queued SIGCHLD to parent when child process's state
3573 * is changed.
3574 */
3575 static void
sigparent(struct proc * p,int reason,int status)3576 sigparent(struct proc *p, int reason, int status)
3577 {
3578 PROC_LOCK_ASSERT(p, MA_OWNED);
3579 PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3580
3581 if (p->p_ksi != NULL) {
3582 p->p_ksi->ksi_signo = SIGCHLD;
3583 p->p_ksi->ksi_code = reason;
3584 p->p_ksi->ksi_status = status;
3585 p->p_ksi->ksi_pid = p->p_pid;
3586 p->p_ksi->ksi_uid = p->p_ucred->cr_ruid;
3587 if (KSI_ONQ(p->p_ksi))
3588 return;
3589 }
3590 pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3591 }
3592
3593 static void
childproc_jobstate(struct proc * p,int reason,int sig)3594 childproc_jobstate(struct proc *p, int reason, int sig)
3595 {
3596 struct sigacts *ps;
3597
3598 PROC_LOCK_ASSERT(p, MA_OWNED);
3599 PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3600
3601 /*
3602 * Wake up parent sleeping in kern_wait(), also send
3603 * SIGCHLD to parent, but SIGCHLD does not guarantee
3604 * that parent will awake, because parent may masked
3605 * the signal.
3606 */
3607 p->p_pptr->p_flag |= P_STATCHILD;
3608 wakeup(p->p_pptr);
3609
3610 ps = p->p_pptr->p_sigacts;
3611 mtx_lock(&ps->ps_mtx);
3612 if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3613 mtx_unlock(&ps->ps_mtx);
3614 sigparent(p, reason, sig);
3615 } else
3616 mtx_unlock(&ps->ps_mtx);
3617 }
3618
3619 void
childproc_stopped(struct proc * p,int reason)3620 childproc_stopped(struct proc *p, int reason)
3621 {
3622
3623 childproc_jobstate(p, reason, p->p_xsig);
3624 }
3625
3626 void
childproc_continued(struct proc * p)3627 childproc_continued(struct proc *p)
3628 {
3629 childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3630 }
3631
3632 void
childproc_exited(struct proc * p)3633 childproc_exited(struct proc *p)
3634 {
3635 int reason, status;
3636
3637 if (WCOREDUMP(p->p_xsig)) {
3638 reason = CLD_DUMPED;
3639 status = WTERMSIG(p->p_xsig);
3640 } else if (WIFSIGNALED(p->p_xsig)) {
3641 reason = CLD_KILLED;
3642 status = WTERMSIG(p->p_xsig);
3643 } else {
3644 reason = CLD_EXITED;
3645 status = p->p_xexit;
3646 }
3647 /*
3648 * XXX avoid calling wakeup(p->p_pptr), the work is
3649 * done in exit1().
3650 */
3651 sigparent(p, reason, status);
3652 }
3653
3654 #define MAX_NUM_CORE_FILES 100000
3655 #ifndef NUM_CORE_FILES
3656 #define NUM_CORE_FILES 5
3657 #endif
3658 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3659 static int num_cores = NUM_CORE_FILES;
3660
3661 static int
sysctl_debug_num_cores_check(SYSCTL_HANDLER_ARGS)3662 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3663 {
3664 int error;
3665 int new_val;
3666
3667 new_val = num_cores;
3668 error = sysctl_handle_int(oidp, &new_val, 0, req);
3669 if (error != 0 || req->newptr == NULL)
3670 return (error);
3671 if (new_val > MAX_NUM_CORE_FILES)
3672 new_val = MAX_NUM_CORE_FILES;
3673 if (new_val < 0)
3674 new_val = 0;
3675 num_cores = new_val;
3676 return (0);
3677 }
3678 SYSCTL_PROC(_debug, OID_AUTO, ncores,
3679 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
3680 sysctl_debug_num_cores_check, "I",
3681 "Maximum number of generated process corefiles while using index format");
3682
3683 #define GZIP_SUFFIX ".gz"
3684 #define ZSTD_SUFFIX ".zst"
3685
3686 int compress_user_cores = 0;
3687
3688 static int
sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)3689 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3690 {
3691 int error, val;
3692
3693 val = compress_user_cores;
3694 error = sysctl_handle_int(oidp, &val, 0, req);
3695 if (error != 0 || req->newptr == NULL)
3696 return (error);
3697 if (val != 0 && !compressor_avail(val))
3698 return (EINVAL);
3699 compress_user_cores = val;
3700 return (error);
3701 }
3702 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores,
3703 CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, 0, sizeof(int),
3704 sysctl_compress_user_cores, "I",
3705 "Enable compression of user corefiles ("
3706 __XSTRING(COMPRESS_GZIP) " = gzip, "
3707 __XSTRING(COMPRESS_ZSTD) " = zstd)");
3708
3709 int compress_user_cores_level = 6;
3710 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3711 &compress_user_cores_level, 0,
3712 "Corefile compression level");
3713
3714 /*
3715 * Protect the access to corefilename[] by allproc_lock.
3716 */
3717 #define corefilename_lock allproc_lock
3718
3719 static char corefilename[MAXPATHLEN] = {"%N.core"};
3720 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3721
3722 static int
sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)3723 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3724 {
3725 int error;
3726
3727 sx_xlock(&corefilename_lock);
3728 error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3729 req);
3730 sx_xunlock(&corefilename_lock);
3731
3732 return (error);
3733 }
3734 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3735 CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3736 "Process corefile name format string");
3737
3738 static void
vnode_close_locked(struct thread * td,struct vnode * vp)3739 vnode_close_locked(struct thread *td, struct vnode *vp)
3740 {
3741
3742 VOP_UNLOCK(vp);
3743 vn_close(vp, FWRITE, td->td_ucred, td);
3744 }
3745
3746 /*
3747 * If the core format has a %I in it, then we need to check
3748 * for existing corefiles before defining a name.
3749 * To do this we iterate over 0..ncores to find a
3750 * non-existing core file name to use. If all core files are
3751 * already used we choose the oldest one.
3752 */
3753 static int
corefile_open_last(struct thread * td,char * name,int indexpos,int indexlen,int ncores,struct vnode ** vpp)3754 corefile_open_last(struct thread *td, char *name, int indexpos,
3755 int indexlen, int ncores, struct vnode **vpp)
3756 {
3757 struct vnode *oldvp, *nextvp, *vp;
3758 struct vattr vattr;
3759 struct nameidata nd;
3760 int error, i, flags, oflags, cmode;
3761 char ch;
3762 struct timespec lasttime;
3763
3764 nextvp = oldvp = NULL;
3765 cmode = S_IRUSR | S_IWUSR;
3766 oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3767 (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3768
3769 for (i = 0; i < ncores; i++) {
3770 flags = O_CREAT | FWRITE | O_NOFOLLOW;
3771
3772 ch = name[indexpos + indexlen];
3773 (void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3774 i);
3775 name[indexpos + indexlen] = ch;
3776
3777 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3778 error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3779 NULL);
3780 if (error != 0)
3781 break;
3782
3783 vp = nd.ni_vp;
3784 NDFREE(&nd, NDF_ONLY_PNBUF);
3785 if ((flags & O_CREAT) == O_CREAT) {
3786 nextvp = vp;
3787 break;
3788 }
3789
3790 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3791 if (error != 0) {
3792 vnode_close_locked(td, vp);
3793 break;
3794 }
3795
3796 if (oldvp == NULL ||
3797 lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3798 (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3799 lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3800 if (oldvp != NULL)
3801 vn_close(oldvp, FWRITE, td->td_ucred, td);
3802 oldvp = vp;
3803 VOP_UNLOCK(oldvp);
3804 lasttime = vattr.va_mtime;
3805 } else {
3806 vnode_close_locked(td, vp);
3807 }
3808 }
3809
3810 if (oldvp != NULL) {
3811 if (nextvp == NULL) {
3812 if ((td->td_proc->p_flag & P_SUGID) != 0) {
3813 error = EFAULT;
3814 vn_close(oldvp, FWRITE, td->td_ucred, td);
3815 } else {
3816 nextvp = oldvp;
3817 error = vn_lock(nextvp, LK_EXCLUSIVE);
3818 if (error != 0) {
3819 vn_close(nextvp, FWRITE, td->td_ucred,
3820 td);
3821 nextvp = NULL;
3822 }
3823 }
3824 } else {
3825 vn_close(oldvp, FWRITE, td->td_ucred, td);
3826 }
3827 }
3828 if (error != 0) {
3829 if (nextvp != NULL)
3830 vnode_close_locked(td, oldvp);
3831 } else {
3832 *vpp = nextvp;
3833 }
3834
3835 return (error);
3836 }
3837
3838 /*
3839 * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3840 * Expand the name described in corefilename, using name, uid, and pid
3841 * and open/create core file.
3842 * corefilename is a printf-like string, with three format specifiers:
3843 * %N name of process ("name")
3844 * %P process id (pid)
3845 * %U user id (uid)
3846 * For example, "%N.core" is the default; they can be disabled completely
3847 * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3848 * This is controlled by the sysctl variable kern.corefile (see above).
3849 */
3850 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)3851 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3852 int compress, int signum, struct vnode **vpp, char **namep)
3853 {
3854 struct sbuf sb;
3855 struct nameidata nd;
3856 const char *format;
3857 char *hostname, *name;
3858 int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
3859
3860 hostname = NULL;
3861 format = corefilename;
3862 name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3863 indexlen = 0;
3864 indexpos = -1;
3865 ncores = num_cores;
3866 (void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3867 sx_slock(&corefilename_lock);
3868 for (i = 0; format[i] != '\0'; i++) {
3869 switch (format[i]) {
3870 case '%': /* Format character */
3871 i++;
3872 switch (format[i]) {
3873 case '%':
3874 sbuf_putc(&sb, '%');
3875 break;
3876 case 'H': /* hostname */
3877 if (hostname == NULL) {
3878 hostname = malloc(MAXHOSTNAMELEN,
3879 M_TEMP, M_WAITOK);
3880 }
3881 getcredhostname(td->td_ucred, hostname,
3882 MAXHOSTNAMELEN);
3883 sbuf_printf(&sb, "%s", hostname);
3884 break;
3885 case 'I': /* autoincrementing index */
3886 if (indexpos != -1) {
3887 sbuf_printf(&sb, "%%I");
3888 break;
3889 }
3890
3891 indexpos = sbuf_len(&sb);
3892 sbuf_printf(&sb, "%u", ncores - 1);
3893 indexlen = sbuf_len(&sb) - indexpos;
3894 break;
3895 case 'N': /* process name */
3896 sbuf_printf(&sb, "%s", comm);
3897 break;
3898 case 'P': /* process id */
3899 sbuf_printf(&sb, "%u", pid);
3900 break;
3901 case 'S': /* signal number */
3902 sbuf_printf(&sb, "%i", signum);
3903 break;
3904 case 'U': /* user id */
3905 sbuf_printf(&sb, "%u", uid);
3906 break;
3907 default:
3908 log(LOG_ERR,
3909 "Unknown format character %c in "
3910 "corename `%s'\n", format[i], format);
3911 break;
3912 }
3913 break;
3914 default:
3915 sbuf_putc(&sb, format[i]);
3916 break;
3917 }
3918 }
3919 sx_sunlock(&corefilename_lock);
3920 free(hostname, M_TEMP);
3921 if (compress == COMPRESS_GZIP)
3922 sbuf_printf(&sb, GZIP_SUFFIX);
3923 else if (compress == COMPRESS_ZSTD)
3924 sbuf_printf(&sb, ZSTD_SUFFIX);
3925 if (sbuf_error(&sb) != 0) {
3926 log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3927 "long\n", (long)pid, comm, (u_long)uid);
3928 sbuf_delete(&sb);
3929 free(name, M_TEMP);
3930 return (ENOMEM);
3931 }
3932 sbuf_finish(&sb);
3933 sbuf_delete(&sb);
3934
3935 if (indexpos != -1) {
3936 error = corefile_open_last(td, name, indexpos, indexlen, ncores,
3937 vpp);
3938 if (error != 0) {
3939 log(LOG_ERR,
3940 "pid %d (%s), uid (%u): Path `%s' failed "
3941 "on initial open test, error = %d\n",
3942 pid, comm, uid, name, error);
3943 }
3944 } else {
3945 cmode = S_IRUSR | S_IWUSR;
3946 oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3947 (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3948 flags = O_CREAT | FWRITE | O_NOFOLLOW;
3949 if ((td->td_proc->p_flag & P_SUGID) != 0)
3950 flags |= O_EXCL;
3951
3952 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3953 error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3954 NULL);
3955 if (error == 0) {
3956 *vpp = nd.ni_vp;
3957 NDFREE(&nd, NDF_ONLY_PNBUF);
3958 }
3959 }
3960
3961 if (error != 0) {
3962 #ifdef AUDIT
3963 audit_proc_coredump(td, name, error);
3964 #endif
3965 free(name, M_TEMP);
3966 return (error);
3967 }
3968 *namep = name;
3969 return (0);
3970 }
3971
3972 /*
3973 * Dump a process' core. The main routine does some
3974 * policy checking, and creates the name of the coredump;
3975 * then it passes on a vnode and a size limit to the process-specific
3976 * coredump routine if there is one; if there _is not_ one, it returns
3977 * ENOSYS; otherwise it returns the error from the process-specific routine.
3978 */
3979
3980 static int
coredump(struct thread * td)3981 coredump(struct thread *td)
3982 {
3983 struct proc *p = td->td_proc;
3984 struct ucred *cred = td->td_ucred;
3985 struct vnode *vp;
3986 struct flock lf;
3987 struct vattr vattr;
3988 size_t fullpathsize;
3989 int error, error1, locked;
3990 char *name; /* name of corefile */
3991 void *rl_cookie;
3992 off_t limit;
3993 char *fullpath, *freepath = NULL;
3994 struct sbuf *sb;
3995
3996 PROC_LOCK_ASSERT(p, MA_OWNED);
3997 MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
3998
3999 if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
4000 (p->p_flag2 & P2_NOTRACE) != 0) {
4001 PROC_UNLOCK(p);
4002 return (EFAULT);
4003 }
4004
4005 /*
4006 * Note that the bulk of limit checking is done after
4007 * the corefile is created. The exception is if the limit
4008 * for corefiles is 0, in which case we don't bother
4009 * creating the corefile at all. This layout means that
4010 * a corefile is truncated instead of not being created,
4011 * if it is larger than the limit.
4012 */
4013 limit = (off_t)lim_cur(td, RLIMIT_CORE);
4014 if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
4015 PROC_UNLOCK(p);
4016 return (EFBIG);
4017 }
4018 PROC_UNLOCK(p);
4019
4020 error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
4021 compress_user_cores, p->p_sig, &vp, &name);
4022 if (error != 0)
4023 return (error);
4024
4025 /*
4026 * Don't dump to non-regular files or files with links.
4027 * Do not dump into system files. Effective user must own the corefile.
4028 */
4029 if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
4030 vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
4031 vattr.va_uid != cred->cr_uid) {
4032 VOP_UNLOCK(vp);
4033 error = EFAULT;
4034 goto out;
4035 }
4036
4037 VOP_UNLOCK(vp);
4038
4039 /* Postpone other writers, including core dumps of other processes. */
4040 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
4041
4042 lf.l_whence = SEEK_SET;
4043 lf.l_start = 0;
4044 lf.l_len = 0;
4045 lf.l_type = F_WRLCK;
4046 locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
4047
4048 VATTR_NULL(&vattr);
4049 vattr.va_size = 0;
4050 if (set_core_nodump_flag)
4051 vattr.va_flags = UF_NODUMP;
4052 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4053 VOP_SETATTR(vp, &vattr, cred);
4054 VOP_UNLOCK(vp);
4055 PROC_LOCK(p);
4056 p->p_acflag |= ACORE;
4057 PROC_UNLOCK(p);
4058
4059 if (p->p_sysent->sv_coredump != NULL) {
4060 error = p->p_sysent->sv_coredump(td, vp, limit, 0);
4061 } else {
4062 error = ENOSYS;
4063 }
4064
4065 if (locked) {
4066 lf.l_type = F_UNLCK;
4067 VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
4068 }
4069 vn_rangelock_unlock(vp, rl_cookie);
4070
4071 /*
4072 * Notify the userland helper that a process triggered a core dump.
4073 * This allows the helper to run an automated debugging session.
4074 */
4075 if (error != 0 || coredump_devctl == 0)
4076 goto out;
4077 sb = sbuf_new_auto();
4078 if (vn_fullpath_global(p->p_textvp, &fullpath, &freepath) != 0)
4079 goto out2;
4080 sbuf_printf(sb, "comm=\"");
4081 devctl_safe_quote_sb(sb, fullpath);
4082 free(freepath, M_TEMP);
4083 sbuf_printf(sb, "\" core=\"");
4084
4085 /*
4086 * We can't lookup core file vp directly. When we're replacing a core, and
4087 * other random times, we flush the name cache, so it will fail. Instead,
4088 * if the path of the core is relative, add the current dir in front if it.
4089 */
4090 if (name[0] != '/') {
4091 fullpathsize = MAXPATHLEN;
4092 freepath = malloc(fullpathsize, M_TEMP, M_WAITOK);
4093 if (vn_getcwd(freepath, &fullpath, &fullpathsize) != 0) {
4094 free(freepath, M_TEMP);
4095 goto out2;
4096 }
4097 devctl_safe_quote_sb(sb, fullpath);
4098 free(freepath, M_TEMP);
4099 sbuf_putc(sb, '/');
4100 }
4101 devctl_safe_quote_sb(sb, name);
4102 sbuf_printf(sb, "\"");
4103 if (sbuf_finish(sb) == 0)
4104 devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
4105 out2:
4106 sbuf_delete(sb);
4107 out:
4108 error1 = vn_close(vp, FWRITE, cred, td);
4109 if (error == 0)
4110 error = error1;
4111 #ifdef AUDIT
4112 audit_proc_coredump(td, name, error);
4113 #endif
4114 free(name, M_TEMP);
4115 return (error);
4116 }
4117
4118 /*
4119 * Nonexistent system call-- signal process (may want to handle it). Flag
4120 * error in case process won't see signal immediately (blocked or ignored).
4121 */
4122 #ifndef _SYS_SYSPROTO_H_
4123 struct nosys_args {
4124 int dummy;
4125 };
4126 #endif
4127 /* ARGSUSED */
4128 int
nosys(struct thread * td,struct nosys_args * args)4129 nosys(struct thread *td, struct nosys_args *args)
4130 {
4131 struct proc *p;
4132
4133 p = td->td_proc;
4134
4135 if (SV_PROC_FLAG(p, SV_SIGSYS) != 0 && kern_signosys) {
4136 PROC_LOCK(p);
4137 tdsignal(td, SIGSYS);
4138 PROC_UNLOCK(p);
4139 }
4140 if (kern_lognosys == 1 || kern_lognosys == 3) {
4141 uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4142 td->td_sa.code);
4143 }
4144 if (kern_lognosys == 2 || kern_lognosys == 3 ||
4145 (p->p_pid == 1 && (kern_lognosys & 3) == 0)) {
4146 printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4147 td->td_sa.code);
4148 }
4149 return (ENOSYS);
4150 }
4151
4152 /*
4153 * Send a SIGIO or SIGURG signal to a process or process group using stored
4154 * credentials rather than those of the current process.
4155 */
4156 void
pgsigio(struct sigio ** sigiop,int sig,int checkctty)4157 pgsigio(struct sigio **sigiop, int sig, int checkctty)
4158 {
4159 ksiginfo_t ksi;
4160 struct sigio *sigio;
4161
4162 ksiginfo_init(&ksi);
4163 ksi.ksi_signo = sig;
4164 ksi.ksi_code = SI_KERNEL;
4165
4166 SIGIO_LOCK();
4167 sigio = *sigiop;
4168 if (sigio == NULL) {
4169 SIGIO_UNLOCK();
4170 return;
4171 }
4172 if (sigio->sio_pgid > 0) {
4173 PROC_LOCK(sigio->sio_proc);
4174 if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
4175 kern_psignal(sigio->sio_proc, sig);
4176 PROC_UNLOCK(sigio->sio_proc);
4177 } else if (sigio->sio_pgid < 0) {
4178 struct proc *p;
4179
4180 PGRP_LOCK(sigio->sio_pgrp);
4181 LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
4182 PROC_LOCK(p);
4183 if (p->p_state == PRS_NORMAL &&
4184 CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
4185 (checkctty == 0 || (p->p_flag & P_CONTROLT)))
4186 kern_psignal(p, sig);
4187 PROC_UNLOCK(p);
4188 }
4189 PGRP_UNLOCK(sigio->sio_pgrp);
4190 }
4191 SIGIO_UNLOCK();
4192 }
4193
4194 static int
filt_sigattach(struct knote * kn)4195 filt_sigattach(struct knote *kn)
4196 {
4197 struct proc *p = curproc;
4198
4199 kn->kn_ptr.p_proc = p;
4200 kn->kn_flags |= EV_CLEAR; /* automatically set */
4201
4202 knlist_add(p->p_klist, kn, 0);
4203
4204 return (0);
4205 }
4206
4207 static void
filt_sigdetach(struct knote * kn)4208 filt_sigdetach(struct knote *kn)
4209 {
4210 knlist_remove(kn->kn_knlist, kn, 0);
4211 }
4212
4213 /*
4214 * signal knotes are shared with proc knotes, so we apply a mask to
4215 * the hint in order to differentiate them from process hints. This
4216 * could be avoided by using a signal-specific knote list, but probably
4217 * isn't worth the trouble.
4218 */
4219 static int
filt_signal(struct knote * kn,long hint)4220 filt_signal(struct knote *kn, long hint)
4221 {
4222
4223 if (hint & NOTE_SIGNAL) {
4224 hint &= ~NOTE_SIGNAL;
4225
4226 if (kn->kn_id == hint)
4227 kn->kn_data++;
4228 }
4229 return (kn->kn_data != 0);
4230 }
4231
4232 struct sigacts *
sigacts_alloc(void)4233 sigacts_alloc(void)
4234 {
4235 struct sigacts *ps;
4236
4237 ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
4238 refcount_init(&ps->ps_refcnt, 1);
4239 mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
4240 return (ps);
4241 }
4242
4243 void
sigacts_free(struct sigacts * ps)4244 sigacts_free(struct sigacts *ps)
4245 {
4246
4247 if (refcount_release(&ps->ps_refcnt) == 0)
4248 return;
4249 mtx_destroy(&ps->ps_mtx);
4250 free(ps, M_SUBPROC);
4251 }
4252
4253 struct sigacts *
sigacts_hold(struct sigacts * ps)4254 sigacts_hold(struct sigacts *ps)
4255 {
4256
4257 refcount_acquire(&ps->ps_refcnt);
4258 return (ps);
4259 }
4260
4261 void
sigacts_copy(struct sigacts * dest,struct sigacts * src)4262 sigacts_copy(struct sigacts *dest, struct sigacts *src)
4263 {
4264
4265 KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
4266 mtx_lock(&src->ps_mtx);
4267 bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
4268 mtx_unlock(&src->ps_mtx);
4269 }
4270
4271 int
sigacts_shared(struct sigacts * ps)4272 sigacts_shared(struct sigacts *ps)
4273 {
4274
4275 return (ps->ps_refcnt > 1);
4276 }
4277
4278 void
sig_drop_caught(struct proc * p)4279 sig_drop_caught(struct proc *p)
4280 {
4281 int sig;
4282 struct sigacts *ps;
4283
4284 ps = p->p_sigacts;
4285 PROC_LOCK_ASSERT(p, MA_OWNED);
4286 mtx_assert(&ps->ps_mtx, MA_OWNED);
4287 SIG_FOREACH(sig, &ps->ps_sigcatch) {
4288 sigdflt(ps, sig);
4289 if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
4290 sigqueue_delete_proc(p, sig);
4291 }
4292 }
4293
4294 static void
sigfastblock_failed(struct thread * td,bool sendsig,bool write)4295 sigfastblock_failed(struct thread *td, bool sendsig, bool write)
4296 {
4297 ksiginfo_t ksi;
4298
4299 /*
4300 * Prevent further fetches and SIGSEGVs, allowing thread to
4301 * issue syscalls despite corruption.
4302 */
4303 sigfastblock_clear(td);
4304
4305 if (!sendsig)
4306 return;
4307 ksiginfo_init_trap(&ksi);
4308 ksi.ksi_signo = SIGSEGV;
4309 ksi.ksi_code = write ? SEGV_ACCERR : SEGV_MAPERR;
4310 ksi.ksi_addr = td->td_sigblock_ptr;
4311 trapsignal(td, &ksi);
4312 }
4313
4314 static bool
sigfastblock_fetch_sig(struct thread * td,bool sendsig,uint32_t * valp)4315 sigfastblock_fetch_sig(struct thread *td, bool sendsig, uint32_t *valp)
4316 {
4317 uint32_t res;
4318
4319 if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4320 return (true);
4321 if (fueword32((void *)td->td_sigblock_ptr, &res) == -1) {
4322 sigfastblock_failed(td, sendsig, false);
4323 return (false);
4324 }
4325 *valp = res;
4326 td->td_sigblock_val = res & ~SIGFASTBLOCK_FLAGS;
4327 return (true);
4328 }
4329
4330 static void
sigfastblock_resched(struct thread * td,bool resched)4331 sigfastblock_resched(struct thread *td, bool resched)
4332 {
4333 struct proc *p;
4334
4335 if (resched) {
4336 p = td->td_proc;
4337 PROC_LOCK(p);
4338 reschedule_signals(p, td->td_sigmask, 0);
4339 PROC_UNLOCK(p);
4340 }
4341 thread_lock(td);
4342 td->td_flags |= TDF_ASTPENDING | TDF_NEEDSIGCHK;
4343 thread_unlock(td);
4344 }
4345
4346 int
sys_sigfastblock(struct thread * td,struct sigfastblock_args * uap)4347 sys_sigfastblock(struct thread *td, struct sigfastblock_args *uap)
4348 {
4349 struct proc *p;
4350 int error, res;
4351 uint32_t oldval;
4352
4353 error = 0;
4354 p = td->td_proc;
4355 switch (uap->cmd) {
4356 case SIGFASTBLOCK_SETPTR:
4357 if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
4358 error = EBUSY;
4359 break;
4360 }
4361 if (((uintptr_t)(uap->ptr) & (sizeof(uint32_t) - 1)) != 0) {
4362 error = EINVAL;
4363 break;
4364 }
4365 td->td_pflags |= TDP_SIGFASTBLOCK;
4366 td->td_sigblock_ptr = uap->ptr;
4367 break;
4368
4369 case SIGFASTBLOCK_UNBLOCK:
4370 if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4371 error = EINVAL;
4372 break;
4373 }
4374
4375 for (;;) {
4376 res = casueword32(td->td_sigblock_ptr,
4377 SIGFASTBLOCK_PEND, &oldval, 0);
4378 if (res == -1) {
4379 error = EFAULT;
4380 sigfastblock_failed(td, false, true);
4381 break;
4382 }
4383 if (res == 0)
4384 break;
4385 MPASS(res == 1);
4386 if (oldval != SIGFASTBLOCK_PEND) {
4387 error = EBUSY;
4388 break;
4389 }
4390 error = thread_check_susp(td, false);
4391 if (error != 0)
4392 break;
4393 }
4394 if (error != 0)
4395 break;
4396
4397 /*
4398 * td_sigblock_val is cleared there, but not on a
4399 * syscall exit. The end effect is that a single
4400 * interruptible sleep, while user sigblock word is
4401 * set, might return EINTR or ERESTART to usermode
4402 * without delivering signal. All further sleeps,
4403 * until userspace clears the word and does
4404 * sigfastblock(UNBLOCK), observe current word and no
4405 * longer get interrupted. It is slight
4406 * non-conformance, with alternative to have read the
4407 * sigblock word on each syscall entry.
4408 */
4409 td->td_sigblock_val = 0;
4410
4411 /*
4412 * Rely on normal ast mechanism to deliver pending
4413 * signals to current thread. But notify others about
4414 * fake unblock.
4415 */
4416 sigfastblock_resched(td, error == 0 && p->p_numthreads != 1);
4417
4418 break;
4419
4420 case SIGFASTBLOCK_UNSETPTR:
4421 if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4422 error = EINVAL;
4423 break;
4424 }
4425 if (!sigfastblock_fetch_sig(td, false, &oldval)) {
4426 error = EFAULT;
4427 break;
4428 }
4429 if (oldval != 0 && oldval != SIGFASTBLOCK_PEND) {
4430 error = EBUSY;
4431 break;
4432 }
4433 sigfastblock_clear(td);
4434 break;
4435
4436 default:
4437 error = EINVAL;
4438 break;
4439 }
4440 return (error);
4441 }
4442
4443 void
sigfastblock_clear(struct thread * td)4444 sigfastblock_clear(struct thread *td)
4445 {
4446 bool resched;
4447
4448 if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4449 return;
4450 td->td_sigblock_val = 0;
4451 resched = (td->td_pflags & TDP_SIGFASTPENDING) != 0 ||
4452 SIGPENDING(td);
4453 td->td_pflags &= ~(TDP_SIGFASTBLOCK | TDP_SIGFASTPENDING);
4454 sigfastblock_resched(td, resched);
4455 }
4456
4457 void
sigfastblock_fetch(struct thread * td)4458 sigfastblock_fetch(struct thread *td)
4459 {
4460 uint32_t val;
4461
4462 (void)sigfastblock_fetch_sig(td, true, &val);
4463 }
4464
4465 static void
sigfastblock_setpend1(struct thread * td)4466 sigfastblock_setpend1(struct thread *td)
4467 {
4468 int res;
4469 uint32_t oldval;
4470
4471 if ((td->td_pflags & TDP_SIGFASTPENDING) == 0)
4472 return;
4473 res = fueword32((void *)td->td_sigblock_ptr, &oldval);
4474 if (res == -1) {
4475 sigfastblock_failed(td, true, false);
4476 return;
4477 }
4478 for (;;) {
4479 res = casueword32(td->td_sigblock_ptr, oldval, &oldval,
4480 oldval | SIGFASTBLOCK_PEND);
4481 if (res == -1) {
4482 sigfastblock_failed(td, true, true);
4483 return;
4484 }
4485 if (res == 0) {
4486 td->td_sigblock_val = oldval & ~SIGFASTBLOCK_FLAGS;
4487 td->td_pflags &= ~TDP_SIGFASTPENDING;
4488 break;
4489 }
4490 MPASS(res == 1);
4491 if (thread_check_susp(td, false) != 0)
4492 break;
4493 }
4494 }
4495
4496 void
sigfastblock_setpend(struct thread * td,bool resched)4497 sigfastblock_setpend(struct thread *td, bool resched)
4498 {
4499 struct proc *p;
4500
4501 sigfastblock_setpend1(td);
4502 if (resched) {
4503 p = td->td_proc;
4504 PROC_LOCK(p);
4505 reschedule_signals(p, fastblock_mask, SIGPROCMASK_FASTBLK);
4506 PROC_UNLOCK(p);
4507 }
4508 }
4509