FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_sig.c
1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * @(#)kern_sig.c 8.7 (Berkeley) 4/18/94
37 */
38
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41
42 #include "opt_capsicum.h"
43 #include "opt_ktrace.h"
44
45 #include <sys/param.h>
46 #include <sys/capsicum.h>
47 #include <sys/ctype.h>
48 #include <sys/systm.h>
49 #include <sys/signalvar.h>
50 #include <sys/vnode.h>
51 #include <sys/acct.h>
52 #include <sys/capsicum.h>
53 #include <sys/compressor.h>
54 #include <sys/condvar.h>
55 #include <sys/devctl.h>
56 #include <sys/event.h>
57 #include <sys/fcntl.h>
58 #include <sys/imgact.h>
59 #include <sys/kernel.h>
60 #include <sys/ktr.h>
61 #include <sys/ktrace.h>
62 #include <sys/limits.h>
63 #include <sys/lock.h>
64 #include <sys/malloc.h>
65 #include <sys/mutex.h>
66 #include <sys/refcount.h>
67 #include <sys/namei.h>
68 #include <sys/proc.h>
69 #include <sys/procdesc.h>
70 #include <sys/ptrace.h>
71 #include <sys/posix4.h>
72 #include <sys/racct.h>
73 #include <sys/resourcevar.h>
74 #include <sys/sdt.h>
75 #include <sys/sbuf.h>
76 #include <sys/sleepqueue.h>
77 #include <sys/smp.h>
78 #include <sys/stat.h>
79 #include <sys/sx.h>
80 #include <sys/syscall.h>
81 #include <sys/syscallsubr.h>
82 #include <sys/sysctl.h>
83 #include <sys/sysent.h>
84 #include <sys/syslog.h>
85 #include <sys/sysproto.h>
86 #include <sys/timers.h>
87 #include <sys/unistd.h>
88 #include <sys/vmmeter.h>
89 #include <sys/wait.h>
90 #include <vm/vm.h>
91 #include <vm/vm_extern.h>
92 #include <vm/uma.h>
93
94 #include <sys/jail.h>
95
96 #include <machine/cpu.h>
97
98 #include <security/audit/audit.h>
99
100 #define ONSIG 32 /* NSIG for osig* syscalls. XXX. */
101
102 SDT_PROVIDER_DECLARE(proc);
103 SDT_PROBE_DEFINE3(proc, , , signal__send,
104 "struct thread *", "struct proc *", "int");
105 SDT_PROBE_DEFINE2(proc, , , signal__clear,
106 "int", "ksiginfo_t *");
107 SDT_PROBE_DEFINE3(proc, , , signal__discard,
108 "struct thread *", "struct proc *", "int");
109
110 static int coredump(struct thread *);
111 static int killpg1(struct thread *td, int sig, int pgid, int all,
112 ksiginfo_t *ksi);
113 static int issignal(struct thread *td);
114 static void reschedule_signals(struct proc *p, sigset_t block, int flags);
115 static int sigprop(int sig);
116 static void tdsigwakeup(struct thread *, int, sig_t, int);
117 static int sig_suspend_threads(struct thread *, struct proc *, int);
118 static int filt_sigattach(struct knote *kn);
119 static void filt_sigdetach(struct knote *kn);
120 static int filt_signal(struct knote *kn, long hint);
121 static struct thread *sigtd(struct proc *p, int sig, bool fast_sigblock);
122 static void sigqueue_start(void);
123
124 static uma_zone_t ksiginfo_zone = NULL;
125 struct filterops sig_filtops = {
126 .f_isfd = 0,
127 .f_attach = filt_sigattach,
128 .f_detach = filt_sigdetach,
129 .f_event = filt_signal,
130 };
131
132 static int kern_logsigexit = 1;
133 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
134 &kern_logsigexit, 0,
135 "Log processes quitting on abnormal signals to syslog(3)");
136
137 static int kern_forcesigexit = 1;
138 SYSCTL_INT(_kern, OID_AUTO, forcesigexit, CTLFLAG_RW,
139 &kern_forcesigexit, 0, "Force trap signal to be handled");
140
141 static SYSCTL_NODE(_kern, OID_AUTO, sigqueue, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
142 "POSIX real time signal");
143
144 static int max_pending_per_proc = 128;
145 SYSCTL_INT(_kern_sigqueue, OID_AUTO, max_pending_per_proc, CTLFLAG_RW,
146 &max_pending_per_proc, 0, "Max pending signals per proc");
147
148 static int preallocate_siginfo = 1024;
149 SYSCTL_INT(_kern_sigqueue, OID_AUTO, preallocate, CTLFLAG_RDTUN,
150 &preallocate_siginfo, 0, "Preallocated signal memory size");
151
152 static int signal_overflow = 0;
153 SYSCTL_INT(_kern_sigqueue, OID_AUTO, overflow, CTLFLAG_RD,
154 &signal_overflow, 0, "Number of signals overflew");
155
156 static int signal_alloc_fail = 0;
157 SYSCTL_INT(_kern_sigqueue, OID_AUTO, alloc_fail, CTLFLAG_RD,
158 &signal_alloc_fail, 0, "signals failed to be allocated");
159
160 static int kern_lognosys = 0;
161 SYSCTL_INT(_kern, OID_AUTO, lognosys, CTLFLAG_RWTUN, &kern_lognosys, 0,
162 "Log invalid syscalls");
163
164 __read_frequently bool sigfastblock_fetch_always = false;
165 SYSCTL_BOOL(_kern, OID_AUTO, sigfastblock_fetch_always, CTLFLAG_RWTUN,
166 &sigfastblock_fetch_always, 0,
167 "Fetch sigfastblock word on each syscall entry for proper "
168 "blocking semantic");
169
170 static bool kern_sig_discard_ign = true;
171 SYSCTL_BOOL(_kern, OID_AUTO, sig_discard_ign, CTLFLAG_RWTUN,
172 &kern_sig_discard_ign, 0,
173 "Discard ignored signals on delivery, otherwise queue them to "
174 "the target queue");
175
176 SYSINIT(signal, SI_SUB_P1003_1B, SI_ORDER_FIRST+3, sigqueue_start, NULL);
177
178 /*
179 * Policy -- Can ucred cr1 send SIGIO to process cr2?
180 * Should use cr_cansignal() once cr_cansignal() allows SIGIO and SIGURG
181 * in the right situations.
182 */
183 #define CANSIGIO(cr1, cr2) \
184 ((cr1)->cr_uid == 0 || \
185 (cr1)->cr_ruid == (cr2)->cr_ruid || \
186 (cr1)->cr_uid == (cr2)->cr_ruid || \
187 (cr1)->cr_ruid == (cr2)->cr_uid || \
188 (cr1)->cr_uid == (cr2)->cr_uid)
189
190 static int sugid_coredump;
191 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RWTUN,
192 &sugid_coredump, 0, "Allow setuid and setgid processes to dump core");
193
194 static int capmode_coredump;
195 SYSCTL_INT(_kern, OID_AUTO, capmode_coredump, CTLFLAG_RWTUN,
196 &capmode_coredump, 0, "Allow processes in capability mode to dump core");
197
198 static int do_coredump = 1;
199 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
200 &do_coredump, 0, "Enable/Disable coredumps");
201
202 static int set_core_nodump_flag = 0;
203 SYSCTL_INT(_kern, OID_AUTO, nodump_coredump, CTLFLAG_RW, &set_core_nodump_flag,
204 0, "Enable setting the NODUMP flag on coredump files");
205
206 static int coredump_devctl = 0;
207 SYSCTL_INT(_kern, OID_AUTO, coredump_devctl, CTLFLAG_RW, &coredump_devctl,
208 0, "Generate a devctl notification when processes coredump");
209
210 /*
211 * Signal properties and actions.
212 * The array below categorizes the signals and their default actions
213 * according to the following properties:
214 */
215 #define SIGPROP_KILL 0x01 /* terminates process by default */
216 #define SIGPROP_CORE 0x02 /* ditto and coredumps */
217 #define SIGPROP_STOP 0x04 /* suspend process */
218 #define SIGPROP_TTYSTOP 0x08 /* ditto, from tty */
219 #define SIGPROP_IGNORE 0x10 /* ignore by default */
220 #define SIGPROP_CONT 0x20 /* continue if suspended */
221
222 static int sigproptbl[NSIG] = {
223 [SIGHUP] = SIGPROP_KILL,
224 [SIGINT] = SIGPROP_KILL,
225 [SIGQUIT] = SIGPROP_KILL | SIGPROP_CORE,
226 [SIGILL] = SIGPROP_KILL | SIGPROP_CORE,
227 [SIGTRAP] = SIGPROP_KILL | SIGPROP_CORE,
228 [SIGABRT] = SIGPROP_KILL | SIGPROP_CORE,
229 [SIGEMT] = SIGPROP_KILL | SIGPROP_CORE,
230 [SIGFPE] = SIGPROP_KILL | SIGPROP_CORE,
231 [SIGKILL] = SIGPROP_KILL,
232 [SIGBUS] = SIGPROP_KILL | SIGPROP_CORE,
233 [SIGSEGV] = SIGPROP_KILL | SIGPROP_CORE,
234 [SIGSYS] = SIGPROP_KILL | SIGPROP_CORE,
235 [SIGPIPE] = SIGPROP_KILL,
236 [SIGALRM] = SIGPROP_KILL,
237 [SIGTERM] = SIGPROP_KILL,
238 [SIGURG] = SIGPROP_IGNORE,
239 [SIGSTOP] = SIGPROP_STOP,
240 [SIGTSTP] = SIGPROP_STOP | SIGPROP_TTYSTOP,
241 [SIGCONT] = SIGPROP_IGNORE | SIGPROP_CONT,
242 [SIGCHLD] = SIGPROP_IGNORE,
243 [SIGTTIN] = SIGPROP_STOP | SIGPROP_TTYSTOP,
244 [SIGTTOU] = SIGPROP_STOP | SIGPROP_TTYSTOP,
245 [SIGIO] = SIGPROP_IGNORE,
246 [SIGXCPU] = SIGPROP_KILL,
247 [SIGXFSZ] = SIGPROP_KILL,
248 [SIGVTALRM] = SIGPROP_KILL,
249 [SIGPROF] = SIGPROP_KILL,
250 [SIGWINCH] = SIGPROP_IGNORE,
251 [SIGINFO] = SIGPROP_IGNORE,
252 [SIGUSR1] = SIGPROP_KILL,
253 [SIGUSR2] = SIGPROP_KILL,
254 };
255
256 #define _SIG_FOREACH_ADVANCE(i, set) ({ \
257 int __found; \
258 for (;;) { \
259 if (__bits != 0) { \
260 int __sig = ffs(__bits); \
261 __bits &= ~(1u << (__sig - 1)); \
262 sig = __i * sizeof((set)->__bits[0]) * NBBY + __sig; \
263 __found = 1; \
264 break; \
265 } \
266 if (++__i == _SIG_WORDS) { \
267 __found = 0; \
268 break; \
269 } \
270 __bits = (set)->__bits[__i]; \
271 } \
272 __found != 0; \
273 })
274
275 #define SIG_FOREACH(i, set) \
276 for (int32_t __i = -1, __bits = 0; \
277 _SIG_FOREACH_ADVANCE(i, set); ) \
278
279 sigset_t fastblock_mask;
280
281 static void
282 sigqueue_start(void)
283 {
284 ksiginfo_zone = uma_zcreate("ksiginfo", sizeof(ksiginfo_t),
285 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
286 uma_prealloc(ksiginfo_zone, preallocate_siginfo);
287 p31b_setcfg(CTL_P1003_1B_REALTIME_SIGNALS, _POSIX_REALTIME_SIGNALS);
288 p31b_setcfg(CTL_P1003_1B_RTSIG_MAX, SIGRTMAX - SIGRTMIN + 1);
289 p31b_setcfg(CTL_P1003_1B_SIGQUEUE_MAX, max_pending_per_proc);
290 SIGFILLSET(fastblock_mask);
291 SIG_CANTMASK(fastblock_mask);
292 }
293
294 ksiginfo_t *
295 ksiginfo_alloc(int mwait)
296 {
297 MPASS(mwait == M_WAITOK || mwait == M_NOWAIT);
298
299 if (ksiginfo_zone == NULL)
300 return (NULL);
301 return (uma_zalloc(ksiginfo_zone, mwait | M_ZERO));
302 }
303
304 void
305 ksiginfo_free(ksiginfo_t *ksi)
306 {
307 uma_zfree(ksiginfo_zone, ksi);
308 }
309
310 static __inline bool
311 ksiginfo_tryfree(ksiginfo_t *ksi)
312 {
313 if ((ksi->ksi_flags & KSI_EXT) == 0) {
314 uma_zfree(ksiginfo_zone, ksi);
315 return (true);
316 }
317 return (false);
318 }
319
320 void
321 sigqueue_init(sigqueue_t *list, struct proc *p)
322 {
323 SIGEMPTYSET(list->sq_signals);
324 SIGEMPTYSET(list->sq_kill);
325 SIGEMPTYSET(list->sq_ptrace);
326 TAILQ_INIT(&list->sq_list);
327 list->sq_proc = p;
328 list->sq_flags = SQ_INIT;
329 }
330
331 /*
332 * Get a signal's ksiginfo.
333 * Return:
334 * 0 - signal not found
335 * others - signal number
336 */
337 static int
338 sigqueue_get(sigqueue_t *sq, int signo, ksiginfo_t *si)
339 {
340 struct proc *p = sq->sq_proc;
341 struct ksiginfo *ksi, *next;
342 int count = 0;
343
344 KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
345
346 if (!SIGISMEMBER(sq->sq_signals, signo))
347 return (0);
348
349 if (SIGISMEMBER(sq->sq_ptrace, signo)) {
350 count++;
351 SIGDELSET(sq->sq_ptrace, signo);
352 si->ksi_flags |= KSI_PTRACE;
353 }
354 if (SIGISMEMBER(sq->sq_kill, signo)) {
355 count++;
356 if (count == 1)
357 SIGDELSET(sq->sq_kill, signo);
358 }
359
360 TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
361 if (ksi->ksi_signo == signo) {
362 if (count == 0) {
363 TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
364 ksi->ksi_sigq = NULL;
365 ksiginfo_copy(ksi, si);
366 if (ksiginfo_tryfree(ksi) && p != NULL)
367 p->p_pendingcnt--;
368 }
369 if (++count > 1)
370 break;
371 }
372 }
373
374 if (count <= 1)
375 SIGDELSET(sq->sq_signals, signo);
376 si->ksi_signo = signo;
377 return (signo);
378 }
379
380 void
381 sigqueue_take(ksiginfo_t *ksi)
382 {
383 struct ksiginfo *kp;
384 struct proc *p;
385 sigqueue_t *sq;
386
387 if (ksi == NULL || (sq = ksi->ksi_sigq) == NULL)
388 return;
389
390 p = sq->sq_proc;
391 TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
392 ksi->ksi_sigq = NULL;
393 if (!(ksi->ksi_flags & KSI_EXT) && p != NULL)
394 p->p_pendingcnt--;
395
396 for (kp = TAILQ_FIRST(&sq->sq_list); kp != NULL;
397 kp = TAILQ_NEXT(kp, ksi_link)) {
398 if (kp->ksi_signo == ksi->ksi_signo)
399 break;
400 }
401 if (kp == NULL && !SIGISMEMBER(sq->sq_kill, ksi->ksi_signo) &&
402 !SIGISMEMBER(sq->sq_ptrace, ksi->ksi_signo))
403 SIGDELSET(sq->sq_signals, ksi->ksi_signo);
404 }
405
406 static int
407 sigqueue_add(sigqueue_t *sq, int signo, ksiginfo_t *si)
408 {
409 struct proc *p = sq->sq_proc;
410 struct ksiginfo *ksi;
411 int ret = 0;
412
413 KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
414
415 /*
416 * SIGKILL/SIGSTOP cannot be caught or masked, so take the fast path
417 * for these signals.
418 */
419 if (signo == SIGKILL || signo == SIGSTOP || si == NULL) {
420 SIGADDSET(sq->sq_kill, signo);
421 goto out_set_bit;
422 }
423
424 /* directly insert the ksi, don't copy it */
425 if (si->ksi_flags & KSI_INS) {
426 if (si->ksi_flags & KSI_HEAD)
427 TAILQ_INSERT_HEAD(&sq->sq_list, si, ksi_link);
428 else
429 TAILQ_INSERT_TAIL(&sq->sq_list, si, ksi_link);
430 si->ksi_sigq = sq;
431 goto out_set_bit;
432 }
433
434 if (__predict_false(ksiginfo_zone == NULL)) {
435 SIGADDSET(sq->sq_kill, signo);
436 goto out_set_bit;
437 }
438
439 if (p != NULL && p->p_pendingcnt >= max_pending_per_proc) {
440 signal_overflow++;
441 ret = EAGAIN;
442 } else if ((ksi = ksiginfo_alloc(M_NOWAIT)) == NULL) {
443 signal_alloc_fail++;
444 ret = EAGAIN;
445 } else {
446 if (p != NULL)
447 p->p_pendingcnt++;
448 ksiginfo_copy(si, ksi);
449 ksi->ksi_signo = signo;
450 if (si->ksi_flags & KSI_HEAD)
451 TAILQ_INSERT_HEAD(&sq->sq_list, ksi, ksi_link);
452 else
453 TAILQ_INSERT_TAIL(&sq->sq_list, ksi, ksi_link);
454 ksi->ksi_sigq = sq;
455 }
456
457 if (ret != 0) {
458 if ((si->ksi_flags & KSI_PTRACE) != 0) {
459 SIGADDSET(sq->sq_ptrace, signo);
460 ret = 0;
461 goto out_set_bit;
462 } else if ((si->ksi_flags & KSI_TRAP) != 0 ||
463 (si->ksi_flags & KSI_SIGQ) == 0) {
464 SIGADDSET(sq->sq_kill, signo);
465 ret = 0;
466 goto out_set_bit;
467 }
468 return (ret);
469 }
470
471 out_set_bit:
472 SIGADDSET(sq->sq_signals, signo);
473 return (ret);
474 }
475
476 void
477 sigqueue_flush(sigqueue_t *sq)
478 {
479 struct proc *p = sq->sq_proc;
480 ksiginfo_t *ksi;
481
482 KASSERT(sq->sq_flags & SQ_INIT, ("sigqueue not inited"));
483
484 if (p != NULL)
485 PROC_LOCK_ASSERT(p, MA_OWNED);
486
487 while ((ksi = TAILQ_FIRST(&sq->sq_list)) != NULL) {
488 TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
489 ksi->ksi_sigq = NULL;
490 if (ksiginfo_tryfree(ksi) && p != NULL)
491 p->p_pendingcnt--;
492 }
493
494 SIGEMPTYSET(sq->sq_signals);
495 SIGEMPTYSET(sq->sq_kill);
496 SIGEMPTYSET(sq->sq_ptrace);
497 }
498
499 static void
500 sigqueue_move_set(sigqueue_t *src, sigqueue_t *dst, const sigset_t *set)
501 {
502 sigset_t tmp;
503 struct proc *p1, *p2;
504 ksiginfo_t *ksi, *next;
505
506 KASSERT(src->sq_flags & SQ_INIT, ("src sigqueue not inited"));
507 KASSERT(dst->sq_flags & SQ_INIT, ("dst sigqueue not inited"));
508 p1 = src->sq_proc;
509 p2 = dst->sq_proc;
510 /* Move siginfo to target list */
511 TAILQ_FOREACH_SAFE(ksi, &src->sq_list, ksi_link, next) {
512 if (SIGISMEMBER(*set, ksi->ksi_signo)) {
513 TAILQ_REMOVE(&src->sq_list, ksi, ksi_link);
514 if (p1 != NULL)
515 p1->p_pendingcnt--;
516 TAILQ_INSERT_TAIL(&dst->sq_list, ksi, ksi_link);
517 ksi->ksi_sigq = dst;
518 if (p2 != NULL)
519 p2->p_pendingcnt++;
520 }
521 }
522
523 /* Move pending bits to target list */
524 tmp = src->sq_kill;
525 SIGSETAND(tmp, *set);
526 SIGSETOR(dst->sq_kill, tmp);
527 SIGSETNAND(src->sq_kill, tmp);
528
529 tmp = src->sq_ptrace;
530 SIGSETAND(tmp, *set);
531 SIGSETOR(dst->sq_ptrace, tmp);
532 SIGSETNAND(src->sq_ptrace, tmp);
533
534 tmp = src->sq_signals;
535 SIGSETAND(tmp, *set);
536 SIGSETOR(dst->sq_signals, tmp);
537 SIGSETNAND(src->sq_signals, tmp);
538 }
539
540 #if 0
541 static void
542 sigqueue_move(sigqueue_t *src, sigqueue_t *dst, int signo)
543 {
544 sigset_t set;
545
546 SIGEMPTYSET(set);
547 SIGADDSET(set, signo);
548 sigqueue_move_set(src, dst, &set);
549 }
550 #endif
551
552 static void
553 sigqueue_delete_set(sigqueue_t *sq, const sigset_t *set)
554 {
555 struct proc *p = sq->sq_proc;
556 ksiginfo_t *ksi, *next;
557
558 KASSERT(sq->sq_flags & SQ_INIT, ("src sigqueue not inited"));
559
560 /* Remove siginfo queue */
561 TAILQ_FOREACH_SAFE(ksi, &sq->sq_list, ksi_link, next) {
562 if (SIGISMEMBER(*set, ksi->ksi_signo)) {
563 TAILQ_REMOVE(&sq->sq_list, ksi, ksi_link);
564 ksi->ksi_sigq = NULL;
565 if (ksiginfo_tryfree(ksi) && p != NULL)
566 p->p_pendingcnt--;
567 }
568 }
569 SIGSETNAND(sq->sq_kill, *set);
570 SIGSETNAND(sq->sq_ptrace, *set);
571 SIGSETNAND(sq->sq_signals, *set);
572 }
573
574 void
575 sigqueue_delete(sigqueue_t *sq, int signo)
576 {
577 sigset_t set;
578
579 SIGEMPTYSET(set);
580 SIGADDSET(set, signo);
581 sigqueue_delete_set(sq, &set);
582 }
583
584 /* Remove a set of signals for a process */
585 static void
586 sigqueue_delete_set_proc(struct proc *p, const sigset_t *set)
587 {
588 sigqueue_t worklist;
589 struct thread *td0;
590
591 PROC_LOCK_ASSERT(p, MA_OWNED);
592
593 sigqueue_init(&worklist, NULL);
594 sigqueue_move_set(&p->p_sigqueue, &worklist, set);
595
596 FOREACH_THREAD_IN_PROC(p, td0)
597 sigqueue_move_set(&td0->td_sigqueue, &worklist, set);
598
599 sigqueue_flush(&worklist);
600 }
601
602 void
603 sigqueue_delete_proc(struct proc *p, int signo)
604 {
605 sigset_t set;
606
607 SIGEMPTYSET(set);
608 SIGADDSET(set, signo);
609 sigqueue_delete_set_proc(p, &set);
610 }
611
612 static void
613 sigqueue_delete_stopmask_proc(struct proc *p)
614 {
615 sigset_t set;
616
617 SIGEMPTYSET(set);
618 SIGADDSET(set, SIGSTOP);
619 SIGADDSET(set, SIGTSTP);
620 SIGADDSET(set, SIGTTIN);
621 SIGADDSET(set, SIGTTOU);
622 sigqueue_delete_set_proc(p, &set);
623 }
624
625 /*
626 * Determine signal that should be delivered to thread td, the current
627 * thread, 0 if none. If there is a pending stop signal with default
628 * action, the process stops in issignal().
629 */
630 int
631 cursig(struct thread *td)
632 {
633 PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
634 mtx_assert(&td->td_proc->p_sigacts->ps_mtx, MA_OWNED);
635 THREAD_LOCK_ASSERT(td, MA_NOTOWNED);
636 return (SIGPENDING(td) ? issignal(td) : 0);
637 }
638
639 /*
640 * Arrange for ast() to handle unmasked pending signals on return to user
641 * mode. This must be called whenever a signal is added to td_sigqueue or
642 * unmasked in td_sigmask.
643 */
644 void
645 signotify(struct thread *td)
646 {
647
648 PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);
649
650 if (SIGPENDING(td)) {
651 thread_lock(td);
652 td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING;
653 thread_unlock(td);
654 }
655 }
656
657 /*
658 * Returns 1 (true) if altstack is configured for the thread, and the
659 * passed stack bottom address falls into the altstack range. Handles
660 * the 43 compat special case where the alt stack size is zero.
661 */
662 int
663 sigonstack(size_t sp)
664 {
665 struct thread *td;
666
667 td = curthread;
668 if ((td->td_pflags & TDP_ALTSTACK) == 0)
669 return (0);
670 #if defined(COMPAT_43)
671 if (SV_PROC_FLAG(td->td_proc, SV_AOUT) && td->td_sigstk.ss_size == 0)
672 return ((td->td_sigstk.ss_flags & SS_ONSTACK) != 0);
673 #endif
674 return (sp >= (size_t)td->td_sigstk.ss_sp &&
675 sp < td->td_sigstk.ss_size + (size_t)td->td_sigstk.ss_sp);
676 }
677
678 static __inline int
679 sigprop(int sig)
680 {
681
682 if (sig > 0 && sig < nitems(sigproptbl))
683 return (sigproptbl[sig]);
684 return (0);
685 }
686
687 static bool
688 sigact_flag_test(const struct sigaction *act, int flag)
689 {
690
691 /*
692 * SA_SIGINFO is reset when signal disposition is set to
693 * ignore or default. Other flags are kept according to user
694 * settings.
695 */
696 return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO ||
697 ((__sighandler_t *)act->sa_sigaction != SIG_IGN &&
698 (__sighandler_t *)act->sa_sigaction != SIG_DFL)));
699 }
700
701 /*
702 * kern_sigaction
703 * sigaction
704 * freebsd4_sigaction
705 * osigaction
706 */
707 int
708 kern_sigaction(struct thread *td, int sig, const struct sigaction *act,
709 struct sigaction *oact, int flags)
710 {
711 struct sigacts *ps;
712 struct proc *p = td->td_proc;
713
714 if (!_SIG_VALID(sig))
715 return (EINVAL);
716 if (act != NULL && act->sa_handler != SIG_DFL &&
717 act->sa_handler != SIG_IGN && (act->sa_flags & ~(SA_ONSTACK |
718 SA_RESTART | SA_RESETHAND | SA_NOCLDSTOP | SA_NODEFER |
719 SA_NOCLDWAIT | SA_SIGINFO)) != 0)
720 return (EINVAL);
721
722 PROC_LOCK(p);
723 ps = p->p_sigacts;
724 mtx_lock(&ps->ps_mtx);
725 if (oact) {
726 memset(oact, 0, sizeof(*oact));
727 oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
728 if (SIGISMEMBER(ps->ps_sigonstack, sig))
729 oact->sa_flags |= SA_ONSTACK;
730 if (!SIGISMEMBER(ps->ps_sigintr, sig))
731 oact->sa_flags |= SA_RESTART;
732 if (SIGISMEMBER(ps->ps_sigreset, sig))
733 oact->sa_flags |= SA_RESETHAND;
734 if (SIGISMEMBER(ps->ps_signodefer, sig))
735 oact->sa_flags |= SA_NODEFER;
736 if (SIGISMEMBER(ps->ps_siginfo, sig)) {
737 oact->sa_flags |= SA_SIGINFO;
738 oact->sa_sigaction =
739 (__siginfohandler_t *)ps->ps_sigact[_SIG_IDX(sig)];
740 } else
741 oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
742 if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDSTOP)
743 oact->sa_flags |= SA_NOCLDSTOP;
744 if (sig == SIGCHLD && ps->ps_flag & PS_NOCLDWAIT)
745 oact->sa_flags |= SA_NOCLDWAIT;
746 }
747 if (act) {
748 if ((sig == SIGKILL || sig == SIGSTOP) &&
749 act->sa_handler != SIG_DFL) {
750 mtx_unlock(&ps->ps_mtx);
751 PROC_UNLOCK(p);
752 return (EINVAL);
753 }
754
755 /*
756 * Change setting atomically.
757 */
758
759 ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
760 SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
761 if (sigact_flag_test(act, SA_SIGINFO)) {
762 ps->ps_sigact[_SIG_IDX(sig)] =
763 (__sighandler_t *)act->sa_sigaction;
764 SIGADDSET(ps->ps_siginfo, sig);
765 } else {
766 ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
767 SIGDELSET(ps->ps_siginfo, sig);
768 }
769 if (!sigact_flag_test(act, SA_RESTART))
770 SIGADDSET(ps->ps_sigintr, sig);
771 else
772 SIGDELSET(ps->ps_sigintr, sig);
773 if (sigact_flag_test(act, SA_ONSTACK))
774 SIGADDSET(ps->ps_sigonstack, sig);
775 else
776 SIGDELSET(ps->ps_sigonstack, sig);
777 if (sigact_flag_test(act, SA_RESETHAND))
778 SIGADDSET(ps->ps_sigreset, sig);
779 else
780 SIGDELSET(ps->ps_sigreset, sig);
781 if (sigact_flag_test(act, SA_NODEFER))
782 SIGADDSET(ps->ps_signodefer, sig);
783 else
784 SIGDELSET(ps->ps_signodefer, sig);
785 if (sig == SIGCHLD) {
786 if (act->sa_flags & SA_NOCLDSTOP)
787 ps->ps_flag |= PS_NOCLDSTOP;
788 else
789 ps->ps_flag &= ~PS_NOCLDSTOP;
790 if (act->sa_flags & SA_NOCLDWAIT) {
791 /*
792 * Paranoia: since SA_NOCLDWAIT is implemented
793 * by reparenting the dying child to PID 1 (and
794 * trust it to reap the zombie), PID 1 itself
795 * is forbidden to set SA_NOCLDWAIT.
796 */
797 if (p->p_pid == 1)
798 ps->ps_flag &= ~PS_NOCLDWAIT;
799 else
800 ps->ps_flag |= PS_NOCLDWAIT;
801 } else
802 ps->ps_flag &= ~PS_NOCLDWAIT;
803 if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
804 ps->ps_flag |= PS_CLDSIGIGN;
805 else
806 ps->ps_flag &= ~PS_CLDSIGIGN;
807 }
808 /*
809 * Set bit in ps_sigignore for signals that are set to SIG_IGN,
810 * and for signals set to SIG_DFL where the default is to
811 * ignore. However, don't put SIGCONT in ps_sigignore, as we
812 * have to restart the process.
813 */
814 if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
815 (sigprop(sig) & SIGPROP_IGNORE &&
816 ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
817 /* never to be seen again */
818 sigqueue_delete_proc(p, sig);
819 if (sig != SIGCONT)
820 /* easier in psignal */
821 SIGADDSET(ps->ps_sigignore, sig);
822 SIGDELSET(ps->ps_sigcatch, sig);
823 } else {
824 SIGDELSET(ps->ps_sigignore, sig);
825 if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
826 SIGDELSET(ps->ps_sigcatch, sig);
827 else
828 SIGADDSET(ps->ps_sigcatch, sig);
829 }
830 #ifdef COMPAT_FREEBSD4
831 if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
832 ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
833 (flags & KSA_FREEBSD4) == 0)
834 SIGDELSET(ps->ps_freebsd4, sig);
835 else
836 SIGADDSET(ps->ps_freebsd4, sig);
837 #endif
838 #ifdef COMPAT_43
839 if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
840 ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL ||
841 (flags & KSA_OSIGSET) == 0)
842 SIGDELSET(ps->ps_osigset, sig);
843 else
844 SIGADDSET(ps->ps_osigset, sig);
845 #endif
846 }
847 mtx_unlock(&ps->ps_mtx);
848 PROC_UNLOCK(p);
849 return (0);
850 }
851
852 #ifndef _SYS_SYSPROTO_H_
853 struct sigaction_args {
854 int sig;
855 struct sigaction *act;
856 struct sigaction *oact;
857 };
858 #endif
859 int
860 sys_sigaction(struct thread *td, struct sigaction_args *uap)
861 {
862 struct sigaction act, oact;
863 struct sigaction *actp, *oactp;
864 int error;
865
866 actp = (uap->act != NULL) ? &act : NULL;
867 oactp = (uap->oact != NULL) ? &oact : NULL;
868 if (actp) {
869 error = copyin(uap->act, actp, sizeof(act));
870 if (error)
871 return (error);
872 }
873 error = kern_sigaction(td, uap->sig, actp, oactp, 0);
874 if (oactp && !error)
875 error = copyout(oactp, uap->oact, sizeof(oact));
876 return (error);
877 }
878
879 #ifdef COMPAT_FREEBSD4
880 #ifndef _SYS_SYSPROTO_H_
881 struct freebsd4_sigaction_args {
882 int sig;
883 struct sigaction *act;
884 struct sigaction *oact;
885 };
886 #endif
887 int
888 freebsd4_sigaction(struct thread *td, struct freebsd4_sigaction_args *uap)
889 {
890 struct sigaction act, oact;
891 struct sigaction *actp, *oactp;
892 int error;
893
894 actp = (uap->act != NULL) ? &act : NULL;
895 oactp = (uap->oact != NULL) ? &oact : NULL;
896 if (actp) {
897 error = copyin(uap->act, actp, sizeof(act));
898 if (error)
899 return (error);
900 }
901 error = kern_sigaction(td, uap->sig, actp, oactp, KSA_FREEBSD4);
902 if (oactp && !error)
903 error = copyout(oactp, uap->oact, sizeof(oact));
904 return (error);
905 }
906 #endif /* COMAPT_FREEBSD4 */
907
908 #ifdef COMPAT_43 /* XXX - COMPAT_FBSD3 */
909 #ifndef _SYS_SYSPROTO_H_
910 struct osigaction_args {
911 int signum;
912 struct osigaction *nsa;
913 struct osigaction *osa;
914 };
915 #endif
916 int
917 osigaction(struct thread *td, struct osigaction_args *uap)
918 {
919 struct osigaction sa;
920 struct sigaction nsa, osa;
921 struct sigaction *nsap, *osap;
922 int error;
923
924 if (uap->signum <= 0 || uap->signum >= ONSIG)
925 return (EINVAL);
926
927 nsap = (uap->nsa != NULL) ? &nsa : NULL;
928 osap = (uap->osa != NULL) ? &osa : NULL;
929
930 if (nsap) {
931 error = copyin(uap->nsa, &sa, sizeof(sa));
932 if (error)
933 return (error);
934 nsap->sa_handler = sa.sa_handler;
935 nsap->sa_flags = sa.sa_flags;
936 OSIG2SIG(sa.sa_mask, nsap->sa_mask);
937 }
938 error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
939 if (osap && !error) {
940 sa.sa_handler = osap->sa_handler;
941 sa.sa_flags = osap->sa_flags;
942 SIG2OSIG(osap->sa_mask, sa.sa_mask);
943 error = copyout(&sa, uap->osa, sizeof(sa));
944 }
945 return (error);
946 }
947
948 #if !defined(__i386__)
949 /* Avoid replicating the same stub everywhere */
950 int
951 osigreturn(struct thread *td, struct osigreturn_args *uap)
952 {
953
954 return (nosys(td, (struct nosys_args *)uap));
955 }
956 #endif
957 #endif /* COMPAT_43 */
958
959 /*
960 * Initialize signal state for process 0;
961 * set to ignore signals that are ignored by default.
962 */
963 void
964 siginit(struct proc *p)
965 {
966 int i;
967 struct sigacts *ps;
968
969 PROC_LOCK(p);
970 ps = p->p_sigacts;
971 mtx_lock(&ps->ps_mtx);
972 for (i = 1; i <= NSIG; i++) {
973 if (sigprop(i) & SIGPROP_IGNORE && i != SIGCONT) {
974 SIGADDSET(ps->ps_sigignore, i);
975 }
976 }
977 mtx_unlock(&ps->ps_mtx);
978 PROC_UNLOCK(p);
979 }
980
981 /*
982 * Reset specified signal to the default disposition.
983 */
984 static void
985 sigdflt(struct sigacts *ps, int sig)
986 {
987
988 mtx_assert(&ps->ps_mtx, MA_OWNED);
989 SIGDELSET(ps->ps_sigcatch, sig);
990 if ((sigprop(sig) & SIGPROP_IGNORE) != 0 && sig != SIGCONT)
991 SIGADDSET(ps->ps_sigignore, sig);
992 ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
993 SIGDELSET(ps->ps_siginfo, sig);
994 }
995
996 /*
997 * Reset signals for an exec of the specified process.
998 */
999 void
1000 execsigs(struct proc *p)
1001 {
1002 sigset_t osigignore;
1003 struct sigacts *ps;
1004 int sig;
1005 struct thread *td;
1006
1007 /*
1008 * Reset caught signals. Held signals remain held
1009 * through td_sigmask (unless they were caught,
1010 * and are now ignored by default).
1011 */
1012 PROC_LOCK_ASSERT(p, MA_OWNED);
1013 ps = p->p_sigacts;
1014 mtx_lock(&ps->ps_mtx);
1015 sig_drop_caught(p);
1016
1017 /*
1018 * As CloudABI processes cannot modify signal handlers, fully
1019 * reset all signals to their default behavior. Do ignore
1020 * SIGPIPE, as it would otherwise be impossible to recover from
1021 * writes to broken pipes and sockets.
1022 */
1023 if (SV_PROC_ABI(p) == SV_ABI_CLOUDABI) {
1024 osigignore = ps->ps_sigignore;
1025 SIG_FOREACH(sig, &osigignore) {
1026 if (sig != SIGPIPE)
1027 sigdflt(ps, sig);
1028 }
1029 SIGADDSET(ps->ps_sigignore, SIGPIPE);
1030 }
1031
1032 /*
1033 * Reset stack state to the user stack.
1034 * Clear set of signals caught on the signal stack.
1035 */
1036 td = curthread;
1037 MPASS(td->td_proc == p);
1038 td->td_sigstk.ss_flags = SS_DISABLE;
1039 td->td_sigstk.ss_size = 0;
1040 td->td_sigstk.ss_sp = 0;
1041 td->td_pflags &= ~TDP_ALTSTACK;
1042 /*
1043 * Reset no zombies if child dies flag as Solaris does.
1044 */
1045 ps->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
1046 if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
1047 ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
1048 mtx_unlock(&ps->ps_mtx);
1049 }
1050
1051 /*
1052 * kern_sigprocmask()
1053 *
1054 * Manipulate signal mask.
1055 */
1056 int
1057 kern_sigprocmask(struct thread *td, int how, sigset_t *set, sigset_t *oset,
1058 int flags)
1059 {
1060 sigset_t new_block, oset1;
1061 struct proc *p;
1062 int error;
1063
1064 p = td->td_proc;
1065 if ((flags & SIGPROCMASK_PROC_LOCKED) != 0)
1066 PROC_LOCK_ASSERT(p, MA_OWNED);
1067 else
1068 PROC_LOCK(p);
1069 mtx_assert(&p->p_sigacts->ps_mtx, (flags & SIGPROCMASK_PS_LOCKED) != 0
1070 ? MA_OWNED : MA_NOTOWNED);
1071 if (oset != NULL)
1072 *oset = td->td_sigmask;
1073
1074 error = 0;
1075 if (set != NULL) {
1076 switch (how) {
1077 case SIG_BLOCK:
1078 SIG_CANTMASK(*set);
1079 oset1 = td->td_sigmask;
1080 SIGSETOR(td->td_sigmask, *set);
1081 new_block = td->td_sigmask;
1082 SIGSETNAND(new_block, oset1);
1083 break;
1084 case SIG_UNBLOCK:
1085 SIGSETNAND(td->td_sigmask, *set);
1086 signotify(td);
1087 goto out;
1088 case SIG_SETMASK:
1089 SIG_CANTMASK(*set);
1090 oset1 = td->td_sigmask;
1091 if (flags & SIGPROCMASK_OLD)
1092 SIGSETLO(td->td_sigmask, *set);
1093 else
1094 td->td_sigmask = *set;
1095 new_block = td->td_sigmask;
1096 SIGSETNAND(new_block, oset1);
1097 signotify(td);
1098 break;
1099 default:
1100 error = EINVAL;
1101 goto out;
1102 }
1103
1104 /*
1105 * The new_block set contains signals that were not previously
1106 * blocked, but are blocked now.
1107 *
1108 * In case we block any signal that was not previously blocked
1109 * for td, and process has the signal pending, try to schedule
1110 * signal delivery to some thread that does not block the
1111 * signal, possibly waking it up.
1112 */
1113 if (p->p_numthreads != 1)
1114 reschedule_signals(p, new_block, flags);
1115 }
1116
1117 out:
1118 if (!(flags & SIGPROCMASK_PROC_LOCKED))
1119 PROC_UNLOCK(p);
1120 return (error);
1121 }
1122
1123 #ifndef _SYS_SYSPROTO_H_
1124 struct sigprocmask_args {
1125 int how;
1126 const sigset_t *set;
1127 sigset_t *oset;
1128 };
1129 #endif
1130 int
1131 sys_sigprocmask(struct thread *td, struct sigprocmask_args *uap)
1132 {
1133 sigset_t set, oset;
1134 sigset_t *setp, *osetp;
1135 int error;
1136
1137 setp = (uap->set != NULL) ? &set : NULL;
1138 osetp = (uap->oset != NULL) ? &oset : NULL;
1139 if (setp) {
1140 error = copyin(uap->set, setp, sizeof(set));
1141 if (error)
1142 return (error);
1143 }
1144 error = kern_sigprocmask(td, uap->how, setp, osetp, 0);
1145 if (osetp && !error) {
1146 error = copyout(osetp, uap->oset, sizeof(oset));
1147 }
1148 return (error);
1149 }
1150
1151 #ifdef COMPAT_43 /* XXX - COMPAT_FBSD3 */
1152 #ifndef _SYS_SYSPROTO_H_
1153 struct osigprocmask_args {
1154 int how;
1155 osigset_t mask;
1156 };
1157 #endif
1158 int
1159 osigprocmask(struct thread *td, struct osigprocmask_args *uap)
1160 {
1161 sigset_t set, oset;
1162 int error;
1163
1164 OSIG2SIG(uap->mask, set);
1165 error = kern_sigprocmask(td, uap->how, &set, &oset, 1);
1166 SIG2OSIG(oset, td->td_retval[0]);
1167 return (error);
1168 }
1169 #endif /* COMPAT_43 */
1170
1171 int
1172 sys_sigwait(struct thread *td, struct sigwait_args *uap)
1173 {
1174 ksiginfo_t ksi;
1175 sigset_t set;
1176 int error;
1177
1178 error = copyin(uap->set, &set, sizeof(set));
1179 if (error) {
1180 td->td_retval[0] = error;
1181 return (0);
1182 }
1183
1184 error = kern_sigtimedwait(td, set, &ksi, NULL);
1185 if (error) {
1186 /*
1187 * sigwait() function shall not return EINTR, but
1188 * the syscall does. Non-ancient libc provides the
1189 * wrapper which hides EINTR. Otherwise, EINTR return
1190 * is used by libthr to handle required cancellation
1191 * point in the sigwait().
1192 */
1193 if (error == EINTR && td->td_proc->p_osrel < P_OSREL_SIGWAIT)
1194 return (ERESTART);
1195 td->td_retval[0] = error;
1196 return (0);
1197 }
1198
1199 error = copyout(&ksi.ksi_signo, uap->sig, sizeof(ksi.ksi_signo));
1200 td->td_retval[0] = error;
1201 return (0);
1202 }
1203
1204 int
1205 sys_sigtimedwait(struct thread *td, struct sigtimedwait_args *uap)
1206 {
1207 struct timespec ts;
1208 struct timespec *timeout;
1209 sigset_t set;
1210 ksiginfo_t ksi;
1211 int error;
1212
1213 if (uap->timeout) {
1214 error = copyin(uap->timeout, &ts, sizeof(ts));
1215 if (error)
1216 return (error);
1217
1218 timeout = &ts;
1219 } else
1220 timeout = NULL;
1221
1222 error = copyin(uap->set, &set, sizeof(set));
1223 if (error)
1224 return (error);
1225
1226 error = kern_sigtimedwait(td, set, &ksi, timeout);
1227 if (error)
1228 return (error);
1229
1230 if (uap->info)
1231 error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1232
1233 if (error == 0)
1234 td->td_retval[0] = ksi.ksi_signo;
1235 return (error);
1236 }
1237
1238 int
1239 sys_sigwaitinfo(struct thread *td, struct sigwaitinfo_args *uap)
1240 {
1241 ksiginfo_t ksi;
1242 sigset_t set;
1243 int error;
1244
1245 error = copyin(uap->set, &set, sizeof(set));
1246 if (error)
1247 return (error);
1248
1249 error = kern_sigtimedwait(td, set, &ksi, NULL);
1250 if (error)
1251 return (error);
1252
1253 if (uap->info)
1254 error = copyout(&ksi.ksi_info, uap->info, sizeof(siginfo_t));
1255
1256 if (error == 0)
1257 td->td_retval[0] = ksi.ksi_signo;
1258 return (error);
1259 }
1260
1261 static void
1262 proc_td_siginfo_capture(struct thread *td, siginfo_t *si)
1263 {
1264 struct thread *thr;
1265
1266 FOREACH_THREAD_IN_PROC(td->td_proc, thr) {
1267 if (thr == td)
1268 thr->td_si = *si;
1269 else
1270 thr->td_si.si_signo = 0;
1271 }
1272 }
1273
1274 int
1275 kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
1276 struct timespec *timeout)
1277 {
1278 struct sigacts *ps;
1279 sigset_t saved_mask, new_block;
1280 struct proc *p;
1281 int error, sig, timevalid = 0;
1282 sbintime_t sbt, precision, tsbt;
1283 struct timespec ts;
1284 bool traced;
1285
1286 p = td->td_proc;
1287 error = 0;
1288 traced = false;
1289
1290 /* Ensure the sigfastblock value is up to date. */
1291 sigfastblock_fetch(td);
1292
1293 if (timeout != NULL) {
1294 if (timeout->tv_nsec >= 0 && timeout->tv_nsec < 1000000000) {
1295 timevalid = 1;
1296 ts = *timeout;
1297 if (ts.tv_sec < INT32_MAX / 2) {
1298 tsbt = tstosbt(ts);
1299 precision = tsbt;
1300 precision >>= tc_precexp;
1301 if (TIMESEL(&sbt, tsbt))
1302 sbt += tc_tick_sbt;
1303 sbt += tsbt;
1304 } else
1305 precision = sbt = 0;
1306 }
1307 } else
1308 precision = sbt = 0;
1309 ksiginfo_init(ksi);
1310 /* Some signals can not be waited for. */
1311 SIG_CANTMASK(waitset);
1312 ps = p->p_sigacts;
1313 PROC_LOCK(p);
1314 saved_mask = td->td_sigmask;
1315 SIGSETNAND(td->td_sigmask, waitset);
1316 if ((p->p_sysent->sv_flags & SV_SIG_DISCIGN) != 0 ||
1317 !kern_sig_discard_ign) {
1318 thread_lock(td);
1319 td->td_flags |= TDF_SIGWAIT;
1320 thread_unlock(td);
1321 }
1322 for (;;) {
1323 mtx_lock(&ps->ps_mtx);
1324 sig = cursig(td);
1325 mtx_unlock(&ps->ps_mtx);
1326 KASSERT(sig >= 0, ("sig %d", sig));
1327 if (sig != 0 && SIGISMEMBER(waitset, sig)) {
1328 if (sigqueue_get(&td->td_sigqueue, sig, ksi) != 0 ||
1329 sigqueue_get(&p->p_sigqueue, sig, ksi) != 0) {
1330 error = 0;
1331 break;
1332 }
1333 }
1334
1335 if (error != 0)
1336 break;
1337
1338 /*
1339 * POSIX says this must be checked after looking for pending
1340 * signals.
1341 */
1342 if (timeout != NULL && !timevalid) {
1343 error = EINVAL;
1344 break;
1345 }
1346
1347 if (traced) {
1348 error = EINTR;
1349 break;
1350 }
1351
1352 error = msleep_sbt(&p->p_sigacts, &p->p_mtx, PPAUSE | PCATCH,
1353 "sigwait", sbt, precision, C_ABSOLUTE);
1354
1355 /* The syscalls can not be restarted. */
1356 if (error == ERESTART)
1357 error = EINTR;
1358
1359 /*
1360 * If PTRACE_SCE or PTRACE_SCX were set after
1361 * userspace entered the syscall, return spurious
1362 * EINTR after wait was done. Only do this as last
1363 * resort after rechecking for possible queued signals
1364 * and expired timeouts.
1365 */
1366 if (error == 0 && (p->p_ptevents & PTRACE_SYSCALL) != 0)
1367 traced = true;
1368 }
1369 thread_lock(td);
1370 td->td_flags &= ~TDF_SIGWAIT;
1371 thread_unlock(td);
1372
1373 new_block = saved_mask;
1374 SIGSETNAND(new_block, td->td_sigmask);
1375 td->td_sigmask = saved_mask;
1376 /*
1377 * Fewer signals can be delivered to us, reschedule signal
1378 * notification.
1379 */
1380 if (p->p_numthreads != 1)
1381 reschedule_signals(p, new_block, 0);
1382
1383 if (error == 0) {
1384 SDT_PROBE2(proc, , , signal__clear, sig, ksi);
1385
1386 if (ksi->ksi_code == SI_TIMER)
1387 itimer_accept(p, ksi->ksi_timerid, ksi);
1388
1389 #ifdef KTRACE
1390 if (KTRPOINT(td, KTR_PSIG)) {
1391 sig_t action;
1392
1393 mtx_lock(&ps->ps_mtx);
1394 action = ps->ps_sigact[_SIG_IDX(sig)];
1395 mtx_unlock(&ps->ps_mtx);
1396 ktrpsig(sig, action, &td->td_sigmask, ksi->ksi_code);
1397 }
1398 #endif
1399 if (sig == SIGKILL) {
1400 proc_td_siginfo_capture(td, &ksi->ksi_info);
1401 sigexit(td, sig);
1402 }
1403 }
1404 PROC_UNLOCK(p);
1405 return (error);
1406 }
1407
1408 #ifndef _SYS_SYSPROTO_H_
1409 struct sigpending_args {
1410 sigset_t *set;
1411 };
1412 #endif
1413 int
1414 sys_sigpending(struct thread *td, struct sigpending_args *uap)
1415 {
1416 struct proc *p = td->td_proc;
1417 sigset_t pending;
1418
1419 PROC_LOCK(p);
1420 pending = p->p_sigqueue.sq_signals;
1421 SIGSETOR(pending, td->td_sigqueue.sq_signals);
1422 PROC_UNLOCK(p);
1423 return (copyout(&pending, uap->set, sizeof(sigset_t)));
1424 }
1425
1426 #ifdef COMPAT_43 /* XXX - COMPAT_FBSD3 */
1427 #ifndef _SYS_SYSPROTO_H_
1428 struct osigpending_args {
1429 int dummy;
1430 };
1431 #endif
1432 int
1433 osigpending(struct thread *td, struct osigpending_args *uap)
1434 {
1435 struct proc *p = td->td_proc;
1436 sigset_t pending;
1437
1438 PROC_LOCK(p);
1439 pending = p->p_sigqueue.sq_signals;
1440 SIGSETOR(pending, td->td_sigqueue.sq_signals);
1441 PROC_UNLOCK(p);
1442 SIG2OSIG(pending, td->td_retval[0]);
1443 return (0);
1444 }
1445 #endif /* COMPAT_43 */
1446
1447 #if defined(COMPAT_43)
1448 /*
1449 * Generalized interface signal handler, 4.3-compatible.
1450 */
1451 #ifndef _SYS_SYSPROTO_H_
1452 struct osigvec_args {
1453 int signum;
1454 struct sigvec *nsv;
1455 struct sigvec *osv;
1456 };
1457 #endif
1458 /* ARGSUSED */
1459 int
1460 osigvec(struct thread *td, struct osigvec_args *uap)
1461 {
1462 struct sigvec vec;
1463 struct sigaction nsa, osa;
1464 struct sigaction *nsap, *osap;
1465 int error;
1466
1467 if (uap->signum <= 0 || uap->signum >= ONSIG)
1468 return (EINVAL);
1469 nsap = (uap->nsv != NULL) ? &nsa : NULL;
1470 osap = (uap->osv != NULL) ? &osa : NULL;
1471 if (nsap) {
1472 error = copyin(uap->nsv, &vec, sizeof(vec));
1473 if (error)
1474 return (error);
1475 nsap->sa_handler = vec.sv_handler;
1476 OSIG2SIG(vec.sv_mask, nsap->sa_mask);
1477 nsap->sa_flags = vec.sv_flags;
1478 nsap->sa_flags ^= SA_RESTART; /* opposite of SV_INTERRUPT */
1479 }
1480 error = kern_sigaction(td, uap->signum, nsap, osap, KSA_OSIGSET);
1481 if (osap && !error) {
1482 vec.sv_handler = osap->sa_handler;
1483 SIG2OSIG(osap->sa_mask, vec.sv_mask);
1484 vec.sv_flags = osap->sa_flags;
1485 vec.sv_flags &= ~SA_NOCLDWAIT;
1486 vec.sv_flags ^= SA_RESTART;
1487 error = copyout(&vec, uap->osv, sizeof(vec));
1488 }
1489 return (error);
1490 }
1491
1492 #ifndef _SYS_SYSPROTO_H_
1493 struct osigblock_args {
1494 int mask;
1495 };
1496 #endif
1497 int
1498 osigblock(struct thread *td, struct osigblock_args *uap)
1499 {
1500 sigset_t set, oset;
1501
1502 OSIG2SIG(uap->mask, set);
1503 kern_sigprocmask(td, SIG_BLOCK, &set, &oset, 0);
1504 SIG2OSIG(oset, td->td_retval[0]);
1505 return (0);
1506 }
1507
1508 #ifndef _SYS_SYSPROTO_H_
1509 struct osigsetmask_args {
1510 int mask;
1511 };
1512 #endif
1513 int
1514 osigsetmask(struct thread *td, struct osigsetmask_args *uap)
1515 {
1516 sigset_t set, oset;
1517
1518 OSIG2SIG(uap->mask, set);
1519 kern_sigprocmask(td, SIG_SETMASK, &set, &oset, 0);
1520 SIG2OSIG(oset, td->td_retval[0]);
1521 return (0);
1522 }
1523 #endif /* COMPAT_43 */
1524
1525 /*
1526 * Suspend calling thread until signal, providing mask to be set in the
1527 * meantime.
1528 */
1529 #ifndef _SYS_SYSPROTO_H_
1530 struct sigsuspend_args {
1531 const sigset_t *sigmask;
1532 };
1533 #endif
1534 /* ARGSUSED */
1535 int
1536 sys_sigsuspend(struct thread *td, struct sigsuspend_args *uap)
1537 {
1538 sigset_t mask;
1539 int error;
1540
1541 error = copyin(uap->sigmask, &mask, sizeof(mask));
1542 if (error)
1543 return (error);
1544 return (kern_sigsuspend(td, mask));
1545 }
1546
1547 int
1548 kern_sigsuspend(struct thread *td, sigset_t mask)
1549 {
1550 struct proc *p = td->td_proc;
1551 int has_sig, sig;
1552
1553 /* Ensure the sigfastblock value is up to date. */
1554 sigfastblock_fetch(td);
1555
1556 /*
1557 * When returning from sigsuspend, we want
1558 * the old mask to be restored after the
1559 * signal handler has finished. Thus, we
1560 * save it here and mark the sigacts structure
1561 * to indicate this.
1562 */
1563 PROC_LOCK(p);
1564 kern_sigprocmask(td, SIG_SETMASK, &mask, &td->td_oldsigmask,
1565 SIGPROCMASK_PROC_LOCKED);
1566 td->td_pflags |= TDP_OLDMASK;
1567
1568 /*
1569 * Process signals now. Otherwise, we can get spurious wakeup
1570 * due to signal entered process queue, but delivered to other
1571 * thread. But sigsuspend should return only on signal
1572 * delivery.
1573 */
1574 (p->p_sysent->sv_set_syscall_retval)(td, EINTR);
1575 for (has_sig = 0; !has_sig;) {
1576 while (msleep(&p->p_sigacts, &p->p_mtx, PPAUSE|PCATCH, "pause",
1577 0) == 0)
1578 /* void */;
1579 thread_suspend_check(0);
1580 mtx_lock(&p->p_sigacts->ps_mtx);
1581 while ((sig = cursig(td)) != 0) {
1582 KASSERT(sig >= 0, ("sig %d", sig));
1583 has_sig += postsig(sig);
1584 }
1585 mtx_unlock(&p->p_sigacts->ps_mtx);
1586
1587 /*
1588 * If PTRACE_SCE or PTRACE_SCX were set after
1589 * userspace entered the syscall, return spurious
1590 * EINTR.
1591 */
1592 if ((p->p_ptevents & PTRACE_SYSCALL) != 0)
1593 has_sig += 1;
1594 }
1595 PROC_UNLOCK(p);
1596 td->td_errno = EINTR;
1597 td->td_pflags |= TDP_NERRNO;
1598 return (EJUSTRETURN);
1599 }
1600
1601 #ifdef COMPAT_43 /* XXX - COMPAT_FBSD3 */
1602 /*
1603 * Compatibility sigsuspend call for old binaries. Note nonstandard calling
1604 * convention: libc stub passes mask, not pointer, to save a copyin.
1605 */
1606 #ifndef _SYS_SYSPROTO_H_
1607 struct osigsuspend_args {
1608 osigset_t mask;
1609 };
1610 #endif
1611 /* ARGSUSED */
1612 int
1613 osigsuspend(struct thread *td, struct osigsuspend_args *uap)
1614 {
1615 sigset_t mask;
1616
1617 OSIG2SIG(uap->mask, mask);
1618 return (kern_sigsuspend(td, mask));
1619 }
1620 #endif /* COMPAT_43 */
1621
1622 #if defined(COMPAT_43)
1623 #ifndef _SYS_SYSPROTO_H_
1624 struct osigstack_args {
1625 struct sigstack *nss;
1626 struct sigstack *oss;
1627 };
1628 #endif
1629 /* ARGSUSED */
1630 int
1631 osigstack(struct thread *td, struct osigstack_args *uap)
1632 {
1633 struct sigstack nss, oss;
1634 int error = 0;
1635
1636 if (uap->nss != NULL) {
1637 error = copyin(uap->nss, &nss, sizeof(nss));
1638 if (error)
1639 return (error);
1640 }
1641 oss.ss_sp = td->td_sigstk.ss_sp;
1642 oss.ss_onstack = sigonstack(cpu_getstack(td));
1643 if (uap->nss != NULL) {
1644 td->td_sigstk.ss_sp = nss.ss_sp;
1645 td->td_sigstk.ss_size = 0;
1646 td->td_sigstk.ss_flags |= nss.ss_onstack & SS_ONSTACK;
1647 td->td_pflags |= TDP_ALTSTACK;
1648 }
1649 if (uap->oss != NULL)
1650 error = copyout(&oss, uap->oss, sizeof(oss));
1651
1652 return (error);
1653 }
1654 #endif /* COMPAT_43 */
1655
1656 #ifndef _SYS_SYSPROTO_H_
1657 struct sigaltstack_args {
1658 stack_t *ss;
1659 stack_t *oss;
1660 };
1661 #endif
1662 /* ARGSUSED */
1663 int
1664 sys_sigaltstack(struct thread *td, struct sigaltstack_args *uap)
1665 {
1666 stack_t ss, oss;
1667 int error;
1668
1669 if (uap->ss != NULL) {
1670 error = copyin(uap->ss, &ss, sizeof(ss));
1671 if (error)
1672 return (error);
1673 }
1674 error = kern_sigaltstack(td, (uap->ss != NULL) ? &ss : NULL,
1675 (uap->oss != NULL) ? &oss : NULL);
1676 if (error)
1677 return (error);
1678 if (uap->oss != NULL)
1679 error = copyout(&oss, uap->oss, sizeof(stack_t));
1680 return (error);
1681 }
1682
1683 int
1684 kern_sigaltstack(struct thread *td, stack_t *ss, stack_t *oss)
1685 {
1686 struct proc *p = td->td_proc;
1687 int oonstack;
1688
1689 oonstack = sigonstack(cpu_getstack(td));
1690
1691 if (oss != NULL) {
1692 *oss = td->td_sigstk;
1693 oss->ss_flags = (td->td_pflags & TDP_ALTSTACK)
1694 ? ((oonstack) ? SS_ONSTACK : 0) : SS_DISABLE;
1695 }
1696
1697 if (ss != NULL) {
1698 if (oonstack)
1699 return (EPERM);
1700 if ((ss->ss_flags & ~SS_DISABLE) != 0)
1701 return (EINVAL);
1702 if (!(ss->ss_flags & SS_DISABLE)) {
1703 if (ss->ss_size < p->p_sysent->sv_minsigstksz)
1704 return (ENOMEM);
1705
1706 td->td_sigstk = *ss;
1707 td->td_pflags |= TDP_ALTSTACK;
1708 } else {
1709 td->td_pflags &= ~TDP_ALTSTACK;
1710 }
1711 }
1712 return (0);
1713 }
1714
1715 struct killpg1_ctx {
1716 struct thread *td;
1717 ksiginfo_t *ksi;
1718 int sig;
1719 bool sent;
1720 bool found;
1721 int ret;
1722 };
1723
1724 static void
1725 killpg1_sendsig(struct proc *p, bool notself, struct killpg1_ctx *arg)
1726 {
1727 int err;
1728
1729 if (p->p_pid <= 1 || (p->p_flag & P_SYSTEM) != 0 ||
1730 (notself && p == arg->td->td_proc) || p->p_state == PRS_NEW)
1731 return;
1732 PROC_LOCK(p);
1733 err = p_cansignal(arg->td, p, arg->sig);
1734 if (err == 0 && arg->sig != 0)
1735 pksignal(p, arg->sig, arg->ksi);
1736 PROC_UNLOCK(p);
1737 if (err != ESRCH)
1738 arg->found = true;
1739 if (err == 0)
1740 arg->sent = true;
1741 else if (arg->ret == 0 && err != ESRCH && err != EPERM)
1742 arg->ret = err;
1743 }
1744
1745 /*
1746 * Common code for kill process group/broadcast kill.
1747 * cp is calling process.
1748 */
1749 static int
1750 killpg1(struct thread *td, int sig, int pgid, int all, ksiginfo_t *ksi)
1751 {
1752 struct proc *p;
1753 struct pgrp *pgrp;
1754 struct killpg1_ctx arg;
1755
1756 arg.td = td;
1757 arg.ksi = ksi;
1758 arg.sig = sig;
1759 arg.sent = false;
1760 arg.found = false;
1761 arg.ret = 0;
1762 if (all) {
1763 /*
1764 * broadcast
1765 */
1766 sx_slock(&allproc_lock);
1767 FOREACH_PROC_IN_SYSTEM(p) {
1768 killpg1_sendsig(p, true, &arg);
1769 }
1770 sx_sunlock(&allproc_lock);
1771 } else {
1772 sx_slock(&proctree_lock);
1773 if (pgid == 0) {
1774 /*
1775 * zero pgid means send to my process group.
1776 */
1777 pgrp = td->td_proc->p_pgrp;
1778 PGRP_LOCK(pgrp);
1779 } else {
1780 pgrp = pgfind(pgid);
1781 if (pgrp == NULL) {
1782 sx_sunlock(&proctree_lock);
1783 return (ESRCH);
1784 }
1785 }
1786 sx_sunlock(&proctree_lock);
1787 LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1788 killpg1_sendsig(p, false, &arg);
1789 }
1790 PGRP_UNLOCK(pgrp);
1791 }
1792 MPASS(arg.ret != 0 || arg.found || !arg.sent);
1793 if (arg.ret == 0 && !arg.sent)
1794 arg.ret = arg.found ? EPERM : ESRCH;
1795 return (arg.ret);
1796 }
1797
1798 #ifndef _SYS_SYSPROTO_H_
1799 struct kill_args {
1800 int pid;
1801 int signum;
1802 };
1803 #endif
1804 /* ARGSUSED */
1805 int
1806 sys_kill(struct thread *td, struct kill_args *uap)
1807 {
1808
1809 return (kern_kill(td, uap->pid, uap->signum));
1810 }
1811
1812 int
1813 kern_kill(struct thread *td, pid_t pid, int signum)
1814 {
1815 ksiginfo_t ksi;
1816 struct proc *p;
1817 int error;
1818
1819 /*
1820 * A process in capability mode can send signals only to himself.
1821 * The main rationale behind this is that abort(3) is implemented as
1822 * kill(getpid(), SIGABRT).
1823 */
1824 if (IN_CAPABILITY_MODE(td) && pid != td->td_proc->p_pid)
1825 return (ECAPMODE);
1826
1827 AUDIT_ARG_SIGNUM(signum);
1828 AUDIT_ARG_PID(pid);
1829 if ((u_int)signum > _SIG_MAXSIG)
1830 return (EINVAL);
1831
1832 ksiginfo_init(&ksi);
1833 ksi.ksi_signo = signum;
1834 ksi.ksi_code = SI_USER;
1835 ksi.ksi_pid = td->td_proc->p_pid;
1836 ksi.ksi_uid = td->td_ucred->cr_ruid;
1837
1838 if (pid > 0) {
1839 /* kill single process */
1840 if ((p = pfind_any(pid)) == NULL)
1841 return (ESRCH);
1842 AUDIT_ARG_PROCESS(p);
1843 error = p_cansignal(td, p, signum);
1844 if (error == 0 && signum)
1845 pksignal(p, signum, &ksi);
1846 PROC_UNLOCK(p);
1847 return (error);
1848 }
1849 switch (pid) {
1850 case -1: /* broadcast signal */
1851 return (killpg1(td, signum, 0, 1, &ksi));
1852 case 0: /* signal own process group */
1853 return (killpg1(td, signum, 0, 0, &ksi));
1854 default: /* negative explicit process group */
1855 return (killpg1(td, signum, -pid, 0, &ksi));
1856 }
1857 /* NOTREACHED */
1858 }
1859
1860 int
1861 sys_pdkill(struct thread *td, struct pdkill_args *uap)
1862 {
1863 struct proc *p;
1864 int error;
1865
1866 AUDIT_ARG_SIGNUM(uap->signum);
1867 AUDIT_ARG_FD(uap->fd);
1868 if ((u_int)uap->signum > _SIG_MAXSIG)
1869 return (EINVAL);
1870
1871 error = procdesc_find(td, uap->fd, &cap_pdkill_rights, &p);
1872 if (error)
1873 return (error);
1874 AUDIT_ARG_PROCESS(p);
1875 error = p_cansignal(td, p, uap->signum);
1876 if (error == 0 && uap->signum)
1877 kern_psignal(p, uap->signum);
1878 PROC_UNLOCK(p);
1879 return (error);
1880 }
1881
1882 #if defined(COMPAT_43)
1883 #ifndef _SYS_SYSPROTO_H_
1884 struct okillpg_args {
1885 int pgid;
1886 int signum;
1887 };
1888 #endif
1889 /* ARGSUSED */
1890 int
1891 okillpg(struct thread *td, struct okillpg_args *uap)
1892 {
1893 ksiginfo_t ksi;
1894
1895 AUDIT_ARG_SIGNUM(uap->signum);
1896 AUDIT_ARG_PID(uap->pgid);
1897 if ((u_int)uap->signum > _SIG_MAXSIG)
1898 return (EINVAL);
1899
1900 ksiginfo_init(&ksi);
1901 ksi.ksi_signo = uap->signum;
1902 ksi.ksi_code = SI_USER;
1903 ksi.ksi_pid = td->td_proc->p_pid;
1904 ksi.ksi_uid = td->td_ucred->cr_ruid;
1905 return (killpg1(td, uap->signum, uap->pgid, 0, &ksi));
1906 }
1907 #endif /* COMPAT_43 */
1908
1909 #ifndef _SYS_SYSPROTO_H_
1910 struct sigqueue_args {
1911 pid_t pid;
1912 int signum;
1913 /* union sigval */ void *value;
1914 };
1915 #endif
1916 int
1917 sys_sigqueue(struct thread *td, struct sigqueue_args *uap)
1918 {
1919 union sigval sv;
1920
1921 sv.sival_ptr = uap->value;
1922
1923 return (kern_sigqueue(td, uap->pid, uap->signum, &sv));
1924 }
1925
1926 int
1927 kern_sigqueue(struct thread *td, pid_t pid, int signum, union sigval *value)
1928 {
1929 ksiginfo_t ksi;
1930 struct proc *p;
1931 int error;
1932
1933 if ((u_int)signum > _SIG_MAXSIG)
1934 return (EINVAL);
1935
1936 /*
1937 * Specification says sigqueue can only send signal to
1938 * single process.
1939 */
1940 if (pid <= 0)
1941 return (EINVAL);
1942
1943 if ((p = pfind_any(pid)) == NULL)
1944 return (ESRCH);
1945 error = p_cansignal(td, p, signum);
1946 if (error == 0 && signum != 0) {
1947 ksiginfo_init(&ksi);
1948 ksi.ksi_flags = KSI_SIGQ;
1949 ksi.ksi_signo = signum;
1950 ksi.ksi_code = SI_QUEUE;
1951 ksi.ksi_pid = td->td_proc->p_pid;
1952 ksi.ksi_uid = td->td_ucred->cr_ruid;
1953 ksi.ksi_value = *value;
1954 error = pksignal(p, ksi.ksi_signo, &ksi);
1955 }
1956 PROC_UNLOCK(p);
1957 return (error);
1958 }
1959
1960 /*
1961 * Send a signal to a process group.
1962 */
1963 void
1964 gsignal(int pgid, int sig, ksiginfo_t *ksi)
1965 {
1966 struct pgrp *pgrp;
1967
1968 if (pgid != 0) {
1969 sx_slock(&proctree_lock);
1970 pgrp = pgfind(pgid);
1971 sx_sunlock(&proctree_lock);
1972 if (pgrp != NULL) {
1973 pgsignal(pgrp, sig, 0, ksi);
1974 PGRP_UNLOCK(pgrp);
1975 }
1976 }
1977 }
1978
1979 /*
1980 * Send a signal to a process group. If checktty is 1,
1981 * limit to members which have a controlling terminal.
1982 */
1983 void
1984 pgsignal(struct pgrp *pgrp, int sig, int checkctty, ksiginfo_t *ksi)
1985 {
1986 struct proc *p;
1987
1988 if (pgrp) {
1989 PGRP_LOCK_ASSERT(pgrp, MA_OWNED);
1990 LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1991 PROC_LOCK(p);
1992 if (p->p_state == PRS_NORMAL &&
1993 (checkctty == 0 || p->p_flag & P_CONTROLT))
1994 pksignal(p, sig, ksi);
1995 PROC_UNLOCK(p);
1996 }
1997 }
1998 }
1999
2000 /*
2001 * Recalculate the signal mask and reset the signal disposition after
2002 * usermode frame for delivery is formed. Should be called after
2003 * mach-specific routine, because sysent->sv_sendsig() needs correct
2004 * ps_siginfo and signal mask.
2005 */
2006 static void
2007 postsig_done(int sig, struct thread *td, struct sigacts *ps)
2008 {
2009 sigset_t mask;
2010
2011 mtx_assert(&ps->ps_mtx, MA_OWNED);
2012 td->td_ru.ru_nsignals++;
2013 mask = ps->ps_catchmask[_SIG_IDX(sig)];
2014 if (!SIGISMEMBER(ps->ps_signodefer, sig))
2015 SIGADDSET(mask, sig);
2016 kern_sigprocmask(td, SIG_BLOCK, &mask, NULL,
2017 SIGPROCMASK_PROC_LOCKED | SIGPROCMASK_PS_LOCKED);
2018 if (SIGISMEMBER(ps->ps_sigreset, sig))
2019 sigdflt(ps, sig);
2020 }
2021
2022 /*
2023 * Send a signal caused by a trap to the current thread. If it will be
2024 * caught immediately, deliver it with correct code. Otherwise, post it
2025 * normally.
2026 */
2027 void
2028 trapsignal(struct thread *td, ksiginfo_t *ksi)
2029 {
2030 struct sigacts *ps;
2031 struct proc *p;
2032 sigset_t sigmask;
2033 int code, sig;
2034
2035 p = td->td_proc;
2036 sig = ksi->ksi_signo;
2037 code = ksi->ksi_code;
2038 KASSERT(_SIG_VALID(sig), ("invalid signal"));
2039
2040 sigfastblock_fetch(td);
2041 PROC_LOCK(p);
2042 ps = p->p_sigacts;
2043 mtx_lock(&ps->ps_mtx);
2044 sigmask = td->td_sigmask;
2045 if (td->td_sigblock_val != 0)
2046 SIGSETOR(sigmask, fastblock_mask);
2047 if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
2048 !SIGISMEMBER(sigmask, sig)) {
2049 #ifdef KTRACE
2050 if (KTRPOINT(curthread, KTR_PSIG))
2051 ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
2052 &td->td_sigmask, code);
2053 #endif
2054 (*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
2055 ksi, &td->td_sigmask);
2056 postsig_done(sig, td, ps);
2057 mtx_unlock(&ps->ps_mtx);
2058 } else {
2059 /*
2060 * Avoid a possible infinite loop if the thread
2061 * masking the signal or process is ignoring the
2062 * signal.
2063 */
2064 if (kern_forcesigexit && (SIGISMEMBER(sigmask, sig) ||
2065 ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN)) {
2066 SIGDELSET(td->td_sigmask, sig);
2067 SIGDELSET(ps->ps_sigcatch, sig);
2068 SIGDELSET(ps->ps_sigignore, sig);
2069 ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2070 td->td_pflags &= ~TDP_SIGFASTBLOCK;
2071 td->td_sigblock_val = 0;
2072 }
2073 mtx_unlock(&ps->ps_mtx);
2074 p->p_sig = sig; /* XXX to verify code */
2075 tdsendsignal(p, td, sig, ksi);
2076 }
2077 PROC_UNLOCK(p);
2078 }
2079
2080 static struct thread *
2081 sigtd(struct proc *p, int sig, bool fast_sigblock)
2082 {
2083 struct thread *td, *signal_td;
2084
2085 PROC_LOCK_ASSERT(p, MA_OWNED);
2086 MPASS(!fast_sigblock || p == curproc);
2087
2088 /*
2089 * Check if current thread can handle the signal without
2090 * switching context to another thread.
2091 */
2092 if (curproc == p && !SIGISMEMBER(curthread->td_sigmask, sig) &&
2093 (!fast_sigblock || curthread->td_sigblock_val == 0))
2094 return (curthread);
2095 signal_td = NULL;
2096 FOREACH_THREAD_IN_PROC(p, td) {
2097 if (!SIGISMEMBER(td->td_sigmask, sig) && (!fast_sigblock ||
2098 td != curthread || td->td_sigblock_val == 0)) {
2099 signal_td = td;
2100 break;
2101 }
2102 }
2103 if (signal_td == NULL)
2104 signal_td = FIRST_THREAD_IN_PROC(p);
2105 return (signal_td);
2106 }
2107
2108 /*
2109 * Send the signal to the process. If the signal has an action, the action
2110 * is usually performed by the target process rather than the caller; we add
2111 * the signal to the set of pending signals for the process.
2112 *
2113 * Exceptions:
2114 * o When a stop signal is sent to a sleeping process that takes the
2115 * default action, the process is stopped without awakening it.
2116 * o SIGCONT restarts stopped processes (or puts them back to sleep)
2117 * regardless of the signal action (eg, blocked or ignored).
2118 *
2119 * Other ignored signals are discarded immediately.
2120 *
2121 * NB: This function may be entered from the debugger via the "kill" DDB
2122 * command. There is little that can be done to mitigate the possibly messy
2123 * side effects of this unwise possibility.
2124 */
2125 void
2126 kern_psignal(struct proc *p, int sig)
2127 {
2128 ksiginfo_t ksi;
2129
2130 ksiginfo_init(&ksi);
2131 ksi.ksi_signo = sig;
2132 ksi.ksi_code = SI_KERNEL;
2133 (void) tdsendsignal(p, NULL, sig, &ksi);
2134 }
2135
2136 int
2137 pksignal(struct proc *p, int sig, ksiginfo_t *ksi)
2138 {
2139
2140 return (tdsendsignal(p, NULL, sig, ksi));
2141 }
2142
2143 /* Utility function for finding a thread to send signal event to. */
2144 int
2145 sigev_findtd(struct proc *p, struct sigevent *sigev, struct thread **ttd)
2146 {
2147 struct thread *td;
2148
2149 if (sigev->sigev_notify == SIGEV_THREAD_ID) {
2150 td = tdfind(sigev->sigev_notify_thread_id, p->p_pid);
2151 if (td == NULL)
2152 return (ESRCH);
2153 *ttd = td;
2154 } else {
2155 *ttd = NULL;
2156 PROC_LOCK(p);
2157 }
2158 return (0);
2159 }
2160
2161 void
2162 tdsignal(struct thread *td, int sig)
2163 {
2164 ksiginfo_t ksi;
2165
2166 ksiginfo_init(&ksi);
2167 ksi.ksi_signo = sig;
2168 ksi.ksi_code = SI_KERNEL;
2169 (void) tdsendsignal(td->td_proc, td, sig, &ksi);
2170 }
2171
2172 void
2173 tdksignal(struct thread *td, int sig, ksiginfo_t *ksi)
2174 {
2175
2176 (void) tdsendsignal(td->td_proc, td, sig, ksi);
2177 }
2178
2179 static int
2180 sig_sleepq_abort(struct thread *td, int intrval)
2181 {
2182 THREAD_LOCK_ASSERT(td, MA_OWNED);
2183
2184 if (intrval == 0 && (td->td_flags & TDF_SIGWAIT) == 0) {
2185 thread_unlock(td);
2186 return (0);
2187 }
2188 return (sleepq_abort(td, intrval));
2189 }
2190
2191 int
2192 tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
2193 {
2194 sig_t action;
2195 sigqueue_t *sigqueue;
2196 int prop;
2197 struct sigacts *ps;
2198 int intrval;
2199 int ret = 0;
2200 int wakeup_swapper;
2201
2202 MPASS(td == NULL || p == td->td_proc);
2203 PROC_LOCK_ASSERT(p, MA_OWNED);
2204
2205 if (!_SIG_VALID(sig))
2206 panic("%s(): invalid signal %d", __func__, sig);
2207
2208 KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("%s: ksi on queue", __func__));
2209
2210 /*
2211 * IEEE Std 1003.1-2001: return success when killing a zombie.
2212 */
2213 if (p->p_state == PRS_ZOMBIE) {
2214 if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2215 ksiginfo_tryfree(ksi);
2216 return (ret);
2217 }
2218
2219 ps = p->p_sigacts;
2220 KNOTE_LOCKED(p->p_klist, NOTE_SIGNAL | sig);
2221 prop = sigprop(sig);
2222
2223 if (td == NULL) {
2224 td = sigtd(p, sig, false);
2225 sigqueue = &p->p_sigqueue;
2226 } else
2227 sigqueue = &td->td_sigqueue;
2228
2229 SDT_PROBE3(proc, , , signal__send, td, p, sig);
2230
2231 /*
2232 * If the signal is being ignored, then we forget about it
2233 * immediately, except when the target process executes
2234 * sigwait(). (Note: we don't set SIGCONT in ps_sigignore,
2235 * and if it is set to SIG_IGN, action will be SIG_DFL here.)
2236 */
2237 mtx_lock(&ps->ps_mtx);
2238 if (SIGISMEMBER(ps->ps_sigignore, sig)) {
2239 if (kern_sig_discard_ign &&
2240 (p->p_sysent->sv_flags & SV_SIG_DISCIGN) == 0) {
2241 SDT_PROBE3(proc, , , signal__discard, td, p, sig);
2242
2243 mtx_unlock(&ps->ps_mtx);
2244 if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2245 ksiginfo_tryfree(ksi);
2246 return (ret);
2247 } else {
2248 action = SIG_CATCH;
2249 intrval = 0;
2250 }
2251 } else {
2252 if (SIGISMEMBER(td->td_sigmask, sig))
2253 action = SIG_HOLD;
2254 else if (SIGISMEMBER(ps->ps_sigcatch, sig))
2255 action = SIG_CATCH;
2256 else
2257 action = SIG_DFL;
2258 if (SIGISMEMBER(ps->ps_sigintr, sig))
2259 intrval = EINTR;
2260 else
2261 intrval = ERESTART;
2262 }
2263 mtx_unlock(&ps->ps_mtx);
2264
2265 if (prop & SIGPROP_CONT)
2266 sigqueue_delete_stopmask_proc(p);
2267 else if (prop & SIGPROP_STOP) {
2268 /*
2269 * If sending a tty stop signal to a member of an orphaned
2270 * process group, discard the signal here if the action
2271 * is default; don't stop the process below if sleeping,
2272 * and don't clear any pending SIGCONT.
2273 */
2274 if ((prop & SIGPROP_TTYSTOP) != 0 &&
2275 (p->p_pgrp->pg_flags & PGRP_ORPHANED) != 0 &&
2276 action == SIG_DFL) {
2277 if (ksi != NULL && (ksi->ksi_flags & KSI_INS) != 0)
2278 ksiginfo_tryfree(ksi);
2279 return (ret);
2280 }
2281 sigqueue_delete_proc(p, SIGCONT);
2282 if (p->p_flag & P_CONTINUED) {
2283 p->p_flag &= ~P_CONTINUED;
2284 PROC_LOCK(p->p_pptr);
2285 sigqueue_take(p->p_ksi);
2286 PROC_UNLOCK(p->p_pptr);
2287 }
2288 }
2289
2290 ret = sigqueue_add(sigqueue, sig, ksi);
2291 if (ret != 0)
2292 return (ret);
2293 signotify(td);
2294 /*
2295 * Defer further processing for signals which are held,
2296 * except that stopped processes must be continued by SIGCONT.
2297 */
2298 if (action == SIG_HOLD &&
2299 !((prop & SIGPROP_CONT) && (p->p_flag & P_STOPPED_SIG)))
2300 return (ret);
2301
2302 wakeup_swapper = 0;
2303
2304 /*
2305 * Some signals have a process-wide effect and a per-thread
2306 * component. Most processing occurs when the process next
2307 * tries to cross the user boundary, however there are some
2308 * times when processing needs to be done immediately, such as
2309 * waking up threads so that they can cross the user boundary.
2310 * We try to do the per-process part here.
2311 */
2312 if (P_SHOULDSTOP(p)) {
2313 KASSERT(!(p->p_flag & P_WEXIT),
2314 ("signal to stopped but exiting process"));
2315 if (sig == SIGKILL) {
2316 /*
2317 * If traced process is already stopped,
2318 * then no further action is necessary.
2319 */
2320 if (p->p_flag & P_TRACED)
2321 goto out;
2322 /*
2323 * SIGKILL sets process running.
2324 * It will die elsewhere.
2325 * All threads must be restarted.
2326 */
2327 p->p_flag &= ~P_STOPPED_SIG;
2328 goto runfast;
2329 }
2330
2331 if (prop & SIGPROP_CONT) {
2332 /*
2333 * If traced process is already stopped,
2334 * then no further action is necessary.
2335 */
2336 if (p->p_flag & P_TRACED)
2337 goto out;
2338 /*
2339 * If SIGCONT is default (or ignored), we continue the
2340 * process but don't leave the signal in sigqueue as
2341 * it has no further action. If SIGCONT is held, we
2342 * continue the process and leave the signal in
2343 * sigqueue. If the process catches SIGCONT, let it
2344 * handle the signal itself. If it isn't waiting on
2345 * an event, it goes back to run state.
2346 * Otherwise, process goes back to sleep state.
2347 */
2348 p->p_flag &= ~P_STOPPED_SIG;
2349 PROC_SLOCK(p);
2350 if (p->p_numthreads == p->p_suspcount) {
2351 PROC_SUNLOCK(p);
2352 p->p_flag |= P_CONTINUED;
2353 p->p_xsig = SIGCONT;
2354 PROC_LOCK(p->p_pptr);
2355 childproc_continued(p);
2356 PROC_UNLOCK(p->p_pptr);
2357 PROC_SLOCK(p);
2358 }
2359 if (action == SIG_DFL) {
2360 thread_unsuspend(p);
2361 PROC_SUNLOCK(p);
2362 sigqueue_delete(sigqueue, sig);
2363 goto out_cont;
2364 }
2365 if (action == SIG_CATCH) {
2366 /*
2367 * The process wants to catch it so it needs
2368 * to run at least one thread, but which one?
2369 */
2370 PROC_SUNLOCK(p);
2371 goto runfast;
2372 }
2373 /*
2374 * The signal is not ignored or caught.
2375 */
2376 thread_unsuspend(p);
2377 PROC_SUNLOCK(p);
2378 goto out_cont;
2379 }
2380
2381 if (prop & SIGPROP_STOP) {
2382 /*
2383 * If traced process is already stopped,
2384 * then no further action is necessary.
2385 */
2386 if (p->p_flag & P_TRACED)
2387 goto out;
2388 /*
2389 * Already stopped, don't need to stop again
2390 * (If we did the shell could get confused).
2391 * Just make sure the signal STOP bit set.
2392 */
2393 p->p_flag |= P_STOPPED_SIG;
2394 sigqueue_delete(sigqueue, sig);
2395 goto out;
2396 }
2397
2398 /*
2399 * All other kinds of signals:
2400 * If a thread is sleeping interruptibly, simulate a
2401 * wakeup so that when it is continued it will be made
2402 * runnable and can look at the signal. However, don't make
2403 * the PROCESS runnable, leave it stopped.
2404 * It may run a bit until it hits a thread_suspend_check().
2405 */
2406 PROC_SLOCK(p);
2407 thread_lock(td);
2408 if (TD_CAN_ABORT(td))
2409 wakeup_swapper = sig_sleepq_abort(td, intrval);
2410 else
2411 thread_unlock(td);
2412 PROC_SUNLOCK(p);
2413 goto out;
2414 /*
2415 * Mutexes are short lived. Threads waiting on them will
2416 * hit thread_suspend_check() soon.
2417 */
2418 } else if (p->p_state == PRS_NORMAL) {
2419 if (p->p_flag & P_TRACED || action == SIG_CATCH) {
2420 tdsigwakeup(td, sig, action, intrval);
2421 goto out;
2422 }
2423
2424 MPASS(action == SIG_DFL);
2425
2426 if (prop & SIGPROP_STOP) {
2427 if (p->p_flag & (P_PPWAIT|P_WEXIT))
2428 goto out;
2429 p->p_flag |= P_STOPPED_SIG;
2430 p->p_xsig = sig;
2431 PROC_SLOCK(p);
2432 wakeup_swapper = sig_suspend_threads(td, p, 1);
2433 if (p->p_numthreads == p->p_suspcount) {
2434 /*
2435 * only thread sending signal to another
2436 * process can reach here, if thread is sending
2437 * signal to its process, because thread does
2438 * not suspend itself here, p_numthreads
2439 * should never be equal to p_suspcount.
2440 */
2441 thread_stopped(p);
2442 PROC_SUNLOCK(p);
2443 sigqueue_delete_proc(p, p->p_xsig);
2444 } else
2445 PROC_SUNLOCK(p);
2446 goto out;
2447 }
2448 } else {
2449 /* Not in "NORMAL" state. discard the signal. */
2450 sigqueue_delete(sigqueue, sig);
2451 goto out;
2452 }
2453
2454 /*
2455 * The process is not stopped so we need to apply the signal to all the
2456 * running threads.
2457 */
2458 runfast:
2459 tdsigwakeup(td, sig, action, intrval);
2460 PROC_SLOCK(p);
2461 thread_unsuspend(p);
2462 PROC_SUNLOCK(p);
2463 out_cont:
2464 itimer_proc_continue(p);
2465 kqtimer_proc_continue(p);
2466 out:
2467 /* If we jump here, proc slock should not be owned. */
2468 PROC_SLOCK_ASSERT(p, MA_NOTOWNED);
2469 if (wakeup_swapper)
2470 kick_proc0();
2471
2472 return (ret);
2473 }
2474
2475 /*
2476 * The force of a signal has been directed against a single
2477 * thread. We need to see what we can do about knocking it
2478 * out of any sleep it may be in etc.
2479 */
2480 static void
2481 tdsigwakeup(struct thread *td, int sig, sig_t action, int intrval)
2482 {
2483 struct proc *p = td->td_proc;
2484 int prop, wakeup_swapper;
2485
2486 PROC_LOCK_ASSERT(p, MA_OWNED);
2487 prop = sigprop(sig);
2488
2489 PROC_SLOCK(p);
2490 thread_lock(td);
2491 /*
2492 * Bring the priority of a thread up if we want it to get
2493 * killed in this lifetime. Be careful to avoid bumping the
2494 * priority of the idle thread, since we still allow to signal
2495 * kernel processes.
2496 */
2497 if (action == SIG_DFL && (prop & SIGPROP_KILL) != 0 &&
2498 td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2499 sched_prio(td, PUSER);
2500 if (TD_ON_SLEEPQ(td)) {
2501 /*
2502 * If thread is sleeping uninterruptibly
2503 * we can't interrupt the sleep... the signal will
2504 * be noticed when the process returns through
2505 * trap() or syscall().
2506 */
2507 if ((td->td_flags & TDF_SINTR) == 0)
2508 goto out;
2509 /*
2510 * If SIGCONT is default (or ignored) and process is
2511 * asleep, we are finished; the process should not
2512 * be awakened.
2513 */
2514 if ((prop & SIGPROP_CONT) && action == SIG_DFL) {
2515 thread_unlock(td);
2516 PROC_SUNLOCK(p);
2517 sigqueue_delete(&p->p_sigqueue, sig);
2518 /*
2519 * It may be on either list in this state.
2520 * Remove from both for now.
2521 */
2522 sigqueue_delete(&td->td_sigqueue, sig);
2523 return;
2524 }
2525
2526 /*
2527 * Don't awaken a sleeping thread for SIGSTOP if the
2528 * STOP signal is deferred.
2529 */
2530 if ((prop & SIGPROP_STOP) != 0 && (td->td_flags & (TDF_SBDRY |
2531 TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
2532 goto out;
2533
2534 /*
2535 * Give low priority threads a better chance to run.
2536 */
2537 if (td->td_priority > PUSER && !TD_IS_IDLETHREAD(td))
2538 sched_prio(td, PUSER);
2539
2540 wakeup_swapper = sig_sleepq_abort(td, intrval);
2541 PROC_SUNLOCK(p);
2542 if (wakeup_swapper)
2543 kick_proc0();
2544 return;
2545 }
2546
2547 /*
2548 * Other states do nothing with the signal immediately,
2549 * other than kicking ourselves if we are running.
2550 * It will either never be noticed, or noticed very soon.
2551 */
2552 #ifdef SMP
2553 if (TD_IS_RUNNING(td) && td != curthread)
2554 forward_signal(td);
2555 #endif
2556
2557 out:
2558 PROC_SUNLOCK(p);
2559 thread_unlock(td);
2560 }
2561
2562 static void
2563 ptrace_coredumpreq(struct thread *td, struct proc *p,
2564 struct thr_coredump_req *tcq)
2565 {
2566 void *rl_cookie;
2567
2568 if (p->p_sysent->sv_coredump == NULL) {
2569 tcq->tc_error = ENOSYS;
2570 return;
2571 }
2572
2573 rl_cookie = vn_rangelock_wlock(tcq->tc_vp, 0, OFF_MAX);
2574 tcq->tc_error = p->p_sysent->sv_coredump(td, tcq->tc_vp,
2575 tcq->tc_limit, tcq->tc_flags);
2576 vn_rangelock_unlock(tcq->tc_vp, rl_cookie);
2577 }
2578
2579 static void
2580 ptrace_syscallreq(struct thread *td, struct proc *p,
2581 struct thr_syscall_req *tsr)
2582 {
2583 struct sysentvec *sv;
2584 struct sysent *se;
2585 register_t rv_saved[2];
2586 int error, nerror;
2587 int sc;
2588 bool audited, sy_thr_static;
2589
2590 sv = p->p_sysent;
2591 if (sv->sv_table == NULL || sv->sv_size < tsr->ts_sa.code) {
2592 tsr->ts_ret.sr_error = ENOSYS;
2593 return;
2594 }
2595
2596 sc = tsr->ts_sa.code;
2597 if (sc == SYS_syscall || sc == SYS___syscall) {
2598 sc = tsr->ts_sa.args[0];
2599 memmove(&tsr->ts_sa.args[0], &tsr->ts_sa.args[1],
2600 sizeof(register_t) * (tsr->ts_nargs - 1));
2601 }
2602
2603 tsr->ts_sa.callp = se = &sv->sv_table[sc];
2604
2605 VM_CNT_INC(v_syscall);
2606 td->td_pticks = 0;
2607 if (__predict_false(td->td_cowgen != atomic_load_int(
2608 &td->td_proc->p_cowgen)))
2609 thread_cow_update(td);
2610
2611 #ifdef CAPABILITY_MODE
2612 if (IN_CAPABILITY_MODE(td) && (se->sy_flags & SYF_CAPENABLED) == 0) {
2613 tsr->ts_ret.sr_error = ECAPMODE;
2614 return;
2615 }
2616 #endif
2617
2618 sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0;
2619 audited = AUDIT_SYSCALL_ENTER(sc, td) != 0;
2620
2621 if (!sy_thr_static) {
2622 error = syscall_thread_enter(td, se);
2623 if (error != 0) {
2624 tsr->ts_ret.sr_error = error;
2625 return;
2626 }
2627 }
2628
2629 rv_saved[0] = td->td_retval[0];
2630 rv_saved[1] = td->td_retval[1];
2631 nerror = td->td_errno;
2632 td->td_retval[0] = 0;
2633 td->td_retval[1] = 0;
2634
2635 #ifdef KDTRACE_HOOKS
2636 if (se->sy_entry != 0)
2637 (*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_ENTRY, 0);
2638 #endif
2639 tsr->ts_ret.sr_error = se->sy_call(td, tsr->ts_sa.args);
2640 #ifdef KDTRACE_HOOKS
2641 if (se->sy_return != 0)
2642 (*systrace_probe_func)(&tsr->ts_sa, SYSTRACE_RETURN,
2643 tsr->ts_ret.sr_error != 0 ? -1 : td->td_retval[0]);
2644 #endif
2645
2646 tsr->ts_ret.sr_retval[0] = td->td_retval[0];
2647 tsr->ts_ret.sr_retval[1] = td->td_retval[1];
2648 td->td_retval[0] = rv_saved[0];
2649 td->td_retval[1] = rv_saved[1];
2650 td->td_errno = nerror;
2651
2652 if (audited)
2653 AUDIT_SYSCALL_EXIT(error, td);
2654 if (!sy_thr_static)
2655 syscall_thread_exit(td, se);
2656 }
2657
2658 static void
2659 ptrace_remotereq(struct thread *td, int flag)
2660 {
2661 struct proc *p;
2662
2663 MPASS(td == curthread);
2664 p = td->td_proc;
2665 PROC_LOCK_ASSERT(p, MA_OWNED);
2666 if ((td->td_dbgflags & flag) == 0)
2667 return;
2668 KASSERT((p->p_flag & P_STOPPED_TRACE) != 0, ("not stopped"));
2669 KASSERT(td->td_remotereq != NULL, ("td_remotereq is NULL"));
2670
2671 PROC_UNLOCK(p);
2672 switch (flag) {
2673 case TDB_COREDUMPREQ:
2674 ptrace_coredumpreq(td, p, td->td_remotereq);
2675 break;
2676 case TDB_SCREMOTEREQ:
2677 ptrace_syscallreq(td, p, td->td_remotereq);
2678 break;
2679 default:
2680 __unreachable();
2681 }
2682 PROC_LOCK(p);
2683
2684 MPASS((td->td_dbgflags & flag) != 0);
2685 td->td_dbgflags &= ~flag;
2686 td->td_remotereq = NULL;
2687 wakeup(p);
2688 }
2689
2690 static int
2691 sig_suspend_threads(struct thread *td, struct proc *p, int sending)
2692 {
2693 struct thread *td2;
2694 int wakeup_swapper;
2695
2696 PROC_LOCK_ASSERT(p, MA_OWNED);
2697 PROC_SLOCK_ASSERT(p, MA_OWNED);
2698 MPASS(sending || td == curthread);
2699
2700 wakeup_swapper = 0;
2701 FOREACH_THREAD_IN_PROC(p, td2) {
2702 thread_lock(td2);
2703 td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
2704 if ((TD_IS_SLEEPING(td2) || TD_IS_SWAPPED(td2)) &&
2705 (td2->td_flags & TDF_SINTR)) {
2706 if (td2->td_flags & TDF_SBDRY) {
2707 /*
2708 * Once a thread is asleep with
2709 * TDF_SBDRY and without TDF_SERESTART
2710 * or TDF_SEINTR set, it should never
2711 * become suspended due to this check.
2712 */
2713 KASSERT(!TD_IS_SUSPENDED(td2),
2714 ("thread with deferred stops suspended"));
2715 if (TD_SBDRY_INTR(td2)) {
2716 wakeup_swapper |= sleepq_abort(td2,
2717 TD_SBDRY_ERRNO(td2));
2718 continue;
2719 }
2720 } else if (!TD_IS_SUSPENDED(td2))
2721 thread_suspend_one(td2);
2722 } else if (!TD_IS_SUSPENDED(td2)) {
2723 if (sending || td != td2)
2724 td2->td_flags |= TDF_ASTPENDING;
2725 #ifdef SMP
2726 if (TD_IS_RUNNING(td2) && td2 != td)
2727 forward_signal(td2);
2728 #endif
2729 }
2730 thread_unlock(td2);
2731 }
2732 return (wakeup_swapper);
2733 }
2734
2735 /*
2736 * Stop the process for an event deemed interesting to the debugger. If si is
2737 * non-NULL, this is a signal exchange; the new signal requested by the
2738 * debugger will be returned for handling. If si is NULL, this is some other
2739 * type of interesting event. The debugger may request a signal be delivered in
2740 * that case as well, however it will be deferred until it can be handled.
2741 */
2742 int
2743 ptracestop(struct thread *td, int sig, ksiginfo_t *si)
2744 {
2745 struct proc *p = td->td_proc;
2746 struct thread *td2;
2747 ksiginfo_t ksi;
2748
2749 PROC_LOCK_ASSERT(p, MA_OWNED);
2750 KASSERT(!(p->p_flag & P_WEXIT), ("Stopping exiting process"));
2751 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
2752 &p->p_mtx.lock_object, "Stopping for traced signal");
2753
2754 td->td_xsig = sig;
2755
2756 if (si == NULL || (si->ksi_flags & KSI_PTRACE) == 0) {
2757 td->td_dbgflags |= TDB_XSIG;
2758 CTR4(KTR_PTRACE, "ptracestop: tid %d (pid %d) flags %#x sig %d",
2759 td->td_tid, p->p_pid, td->td_dbgflags, sig);
2760 PROC_SLOCK(p);
2761 while ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_XSIG)) {
2762 if (P_KILLED(p)) {
2763 /*
2764 * Ensure that, if we've been PT_KILLed, the
2765 * exit status reflects that. Another thread
2766 * may also be in ptracestop(), having just
2767 * received the SIGKILL, but this thread was
2768 * unsuspended first.
2769 */
2770 td->td_dbgflags &= ~TDB_XSIG;
2771 td->td_xsig = SIGKILL;
2772 p->p_ptevents = 0;
2773 break;
2774 }
2775 if (p->p_flag & P_SINGLE_EXIT &&
2776 !(td->td_dbgflags & TDB_EXIT)) {
2777 /*
2778 * Ignore ptrace stops except for thread exit
2779 * events when the process exits.
2780 */
2781 td->td_dbgflags &= ~TDB_XSIG;
2782 PROC_SUNLOCK(p);
2783 return (0);
2784 }
2785
2786 /*
2787 * Make wait(2) work. Ensure that right after the
2788 * attach, the thread which was decided to become the
2789 * leader of attach gets reported to the waiter.
2790 * Otherwise, just avoid overwriting another thread's
2791 * assignment to p_xthread. If another thread has
2792 * already set p_xthread, the current thread will get
2793 * a chance to report itself upon the next iteration.
2794 */
2795 if ((td->td_dbgflags & TDB_FSTP) != 0 ||
2796 ((p->p_flag2 & P2_PTRACE_FSTP) == 0 &&
2797 p->p_xthread == NULL)) {
2798 p->p_xsig = sig;
2799 p->p_xthread = td;
2800
2801 /*
2802 * If we are on sleepqueue already,
2803 * let sleepqueue code decide if it
2804 * needs to go sleep after attach.
2805 */
2806 if (td->td_wchan == NULL)
2807 td->td_dbgflags &= ~TDB_FSTP;
2808
2809 p->p_flag2 &= ~P2_PTRACE_FSTP;
2810 p->p_flag |= P_STOPPED_SIG | P_STOPPED_TRACE;
2811 sig_suspend_threads(td, p, 0);
2812 }
2813 if ((td->td_dbgflags & TDB_STOPATFORK) != 0) {
2814 td->td_dbgflags &= ~TDB_STOPATFORK;
2815 }
2816 stopme:
2817 td->td_dbgflags |= TDB_SSWITCH;
2818 thread_suspend_switch(td, p);
2819 td->td_dbgflags &= ~TDB_SSWITCH;
2820 if ((td->td_dbgflags & (TDB_COREDUMPREQ |
2821 TDB_SCREMOTEREQ)) != 0) {
2822 MPASS((td->td_dbgflags & (TDB_COREDUMPREQ |
2823 TDB_SCREMOTEREQ)) !=
2824 (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2825 PROC_SUNLOCK(p);
2826 ptrace_remotereq(td, td->td_dbgflags &
2827 (TDB_COREDUMPREQ | TDB_SCREMOTEREQ));
2828 PROC_SLOCK(p);
2829 goto stopme;
2830 }
2831 if (p->p_xthread == td)
2832 p->p_xthread = NULL;
2833 if (!(p->p_flag & P_TRACED))
2834 break;
2835 if (td->td_dbgflags & TDB_SUSPEND) {
2836 if (p->p_flag & P_SINGLE_EXIT)
2837 break;
2838 goto stopme;
2839 }
2840 }
2841 PROC_SUNLOCK(p);
2842 }
2843
2844 if (si != NULL && sig == td->td_xsig) {
2845 /* Parent wants us to take the original signal unchanged. */
2846 si->ksi_flags |= KSI_HEAD;
2847 if (sigqueue_add(&td->td_sigqueue, sig, si) != 0)
2848 si->ksi_signo = 0;
2849 } else if (td->td_xsig != 0) {
2850 /*
2851 * If parent wants us to take a new signal, then it will leave
2852 * it in td->td_xsig; otherwise we just look for signals again.
2853 */
2854 ksiginfo_init(&ksi);
2855 ksi.ksi_signo = td->td_xsig;
2856 ksi.ksi_flags |= KSI_PTRACE;
2857 td2 = sigtd(p, td->td_xsig, false);
2858 tdsendsignal(p, td2, td->td_xsig, &ksi);
2859 if (td != td2)
2860 return (0);
2861 }
2862
2863 return (td->td_xsig);
2864 }
2865
2866 static void
2867 reschedule_signals(struct proc *p, sigset_t block, int flags)
2868 {
2869 struct sigacts *ps;
2870 struct thread *td;
2871 int sig;
2872 bool fastblk, pslocked;
2873
2874 PROC_LOCK_ASSERT(p, MA_OWNED);
2875 ps = p->p_sigacts;
2876 pslocked = (flags & SIGPROCMASK_PS_LOCKED) != 0;
2877 mtx_assert(&ps->ps_mtx, pslocked ? MA_OWNED : MA_NOTOWNED);
2878 if (SIGISEMPTY(p->p_siglist))
2879 return;
2880 SIGSETAND(block, p->p_siglist);
2881 fastblk = (flags & SIGPROCMASK_FASTBLK) != 0;
2882 SIG_FOREACH(sig, &block) {
2883 td = sigtd(p, sig, fastblk);
2884
2885 /*
2886 * If sigtd() selected us despite sigfastblock is
2887 * blocking, do not activate AST or wake us, to avoid
2888 * loop in AST handler.
2889 */
2890 if (fastblk && td == curthread)
2891 continue;
2892
2893 signotify(td);
2894 if (!pslocked)
2895 mtx_lock(&ps->ps_mtx);
2896 if (p->p_flag & P_TRACED ||
2897 (SIGISMEMBER(ps->ps_sigcatch, sig) &&
2898 !SIGISMEMBER(td->td_sigmask, sig))) {
2899 tdsigwakeup(td, sig, SIG_CATCH,
2900 (SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR :
2901 ERESTART));
2902 }
2903 if (!pslocked)
2904 mtx_unlock(&ps->ps_mtx);
2905 }
2906 }
2907
2908 void
2909 tdsigcleanup(struct thread *td)
2910 {
2911 struct proc *p;
2912 sigset_t unblocked;
2913
2914 p = td->td_proc;
2915 PROC_LOCK_ASSERT(p, MA_OWNED);
2916
2917 sigqueue_flush(&td->td_sigqueue);
2918 if (p->p_numthreads == 1)
2919 return;
2920
2921 /*
2922 * Since we cannot handle signals, notify signal post code
2923 * about this by filling the sigmask.
2924 *
2925 * Also, if needed, wake up thread(s) that do not block the
2926 * same signals as the exiting thread, since the thread might
2927 * have been selected for delivery and woken up.
2928 */
2929 SIGFILLSET(unblocked);
2930 SIGSETNAND(unblocked, td->td_sigmask);
2931 SIGFILLSET(td->td_sigmask);
2932 reschedule_signals(p, unblocked, 0);
2933
2934 }
2935
2936 static int
2937 sigdeferstop_curr_flags(int cflags)
2938 {
2939
2940 MPASS((cflags & (TDF_SEINTR | TDF_SERESTART)) == 0 ||
2941 (cflags & TDF_SBDRY) != 0);
2942 return (cflags & (TDF_SBDRY | TDF_SEINTR | TDF_SERESTART));
2943 }
2944
2945 /*
2946 * Defer the delivery of SIGSTOP for the current thread, according to
2947 * the requested mode. Returns previous flags, which must be restored
2948 * by sigallowstop().
2949 *
2950 * TDF_SBDRY, TDF_SEINTR, and TDF_SERESTART flags are only set and
2951 * cleared by the current thread, which allow the lock-less read-only
2952 * accesses below.
2953 */
2954 int
2955 sigdeferstop_impl(int mode)
2956 {
2957 struct thread *td;
2958 int cflags, nflags;
2959
2960 td = curthread;
2961 cflags = sigdeferstop_curr_flags(td->td_flags);
2962 switch (mode) {
2963 case SIGDEFERSTOP_NOP:
2964 nflags = cflags;
2965 break;
2966 case SIGDEFERSTOP_OFF:
2967 nflags = 0;
2968 break;
2969 case SIGDEFERSTOP_SILENT:
2970 nflags = (cflags | TDF_SBDRY) & ~(TDF_SEINTR | TDF_SERESTART);
2971 break;
2972 case SIGDEFERSTOP_EINTR:
2973 nflags = (cflags | TDF_SBDRY | TDF_SEINTR) & ~TDF_SERESTART;
2974 break;
2975 case SIGDEFERSTOP_ERESTART:
2976 nflags = (cflags | TDF_SBDRY | TDF_SERESTART) & ~TDF_SEINTR;
2977 break;
2978 default:
2979 panic("sigdeferstop: invalid mode %x", mode);
2980 break;
2981 }
2982 if (cflags == nflags)
2983 return (SIGDEFERSTOP_VAL_NCHG);
2984 thread_lock(td);
2985 td->td_flags = (td->td_flags & ~cflags) | nflags;
2986 thread_unlock(td);
2987 return (cflags);
2988 }
2989
2990 /*
2991 * Restores the STOP handling mode, typically permitting the delivery
2992 * of SIGSTOP for the current thread. This does not immediately
2993 * suspend if a stop was posted. Instead, the thread will suspend
2994 * either via ast() or a subsequent interruptible sleep.
2995 */
2996 void
2997 sigallowstop_impl(int prev)
2998 {
2999 struct thread *td;
3000 int cflags;
3001
3002 KASSERT(prev != SIGDEFERSTOP_VAL_NCHG, ("failed sigallowstop"));
3003 KASSERT((prev & ~(TDF_SBDRY | TDF_SEINTR | TDF_SERESTART)) == 0,
3004 ("sigallowstop: incorrect previous mode %x", prev));
3005 td = curthread;
3006 cflags = sigdeferstop_curr_flags(td->td_flags);
3007 if (cflags != prev) {
3008 thread_lock(td);
3009 td->td_flags = (td->td_flags & ~cflags) | prev;
3010 thread_unlock(td);
3011 }
3012 }
3013
3014 enum sigstatus {
3015 SIGSTATUS_HANDLE,
3016 SIGSTATUS_HANDLED,
3017 SIGSTATUS_IGNORE,
3018 SIGSTATUS_SBDRY_STOP,
3019 };
3020
3021 /*
3022 * The thread has signal "sig" pending. Figure out what to do with it:
3023 *
3024 * _HANDLE -> the caller should handle the signal
3025 * _HANDLED -> handled internally, reload pending signal set
3026 * _IGNORE -> ignored, remove from the set of pending signals and try the
3027 * next pending signal
3028 * _SBDRY_STOP -> the signal should stop the thread but this is not
3029 * permitted in the current context
3030 */
3031 static enum sigstatus
3032 sigprocess(struct thread *td, int sig)
3033 {
3034 struct proc *p;
3035 struct sigacts *ps;
3036 struct sigqueue *queue;
3037 ksiginfo_t ksi;
3038 int prop;
3039
3040 KASSERT(_SIG_VALID(sig), ("%s: invalid signal %d", __func__, sig));
3041
3042 p = td->td_proc;
3043 ps = p->p_sigacts;
3044 mtx_assert(&ps->ps_mtx, MA_OWNED);
3045 PROC_LOCK_ASSERT(p, MA_OWNED);
3046
3047 /*
3048 * We should allow pending but ignored signals below
3049 * if there is sigwait() active, or P_TRACED was
3050 * on when they were posted.
3051 */
3052 if (SIGISMEMBER(ps->ps_sigignore, sig) &&
3053 (p->p_flag & P_TRACED) == 0 &&
3054 (td->td_flags & TDF_SIGWAIT) == 0) {
3055 return (SIGSTATUS_IGNORE);
3056 }
3057
3058 /*
3059 * If the process is going to single-thread mode to prepare
3060 * for exit, there is no sense in delivering any signal
3061 * to usermode. Another important consequence is that
3062 * msleep(..., PCATCH, ...) now is only interruptible by a
3063 * suspend request.
3064 */
3065 if ((p->p_flag2 & P2_WEXIT) != 0)
3066 return (SIGSTATUS_IGNORE);
3067
3068 if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED) {
3069 /*
3070 * If traced, always stop.
3071 * Remove old signal from queue before the stop.
3072 * XXX shrug off debugger, it causes siginfo to
3073 * be thrown away.
3074 */
3075 queue = &td->td_sigqueue;
3076 ksiginfo_init(&ksi);
3077 if (sigqueue_get(queue, sig, &ksi) == 0) {
3078 queue = &p->p_sigqueue;
3079 sigqueue_get(queue, sig, &ksi);
3080 }
3081 td->td_si = ksi.ksi_info;
3082
3083 mtx_unlock(&ps->ps_mtx);
3084 sig = ptracestop(td, sig, &ksi);
3085 mtx_lock(&ps->ps_mtx);
3086
3087 td->td_si.si_signo = 0;
3088
3089 /*
3090 * Keep looking if the debugger discarded or
3091 * replaced the signal.
3092 */
3093 if (sig == 0)
3094 return (SIGSTATUS_HANDLED);
3095
3096 /*
3097 * If the signal became masked, re-queue it.
3098 */
3099 if (SIGISMEMBER(td->td_sigmask, sig)) {
3100 ksi.ksi_flags |= KSI_HEAD;
3101 sigqueue_add(&p->p_sigqueue, sig, &ksi);
3102 return (SIGSTATUS_HANDLED);
3103 }
3104
3105 /*
3106 * If the traced bit got turned off, requeue the signal and
3107 * reload the set of pending signals. This ensures that p_sig*
3108 * and p_sigact are consistent.
3109 */
3110 if ((p->p_flag & P_TRACED) == 0) {
3111 if ((ksi.ksi_flags & KSI_PTRACE) == 0) {
3112 ksi.ksi_flags |= KSI_HEAD;
3113 sigqueue_add(queue, sig, &ksi);
3114 }
3115 return (SIGSTATUS_HANDLED);
3116 }
3117 }
3118
3119 /*
3120 * Decide whether the signal should be returned.
3121 * Return the signal's number, or fall through
3122 * to clear it from the pending mask.
3123 */
3124 switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
3125 case (intptr_t)SIG_DFL:
3126 /*
3127 * Don't take default actions on system processes.
3128 */
3129 if (p->p_pid <= 1) {
3130 #ifdef DIAGNOSTIC
3131 /*
3132 * Are you sure you want to ignore SIGSEGV
3133 * in init? XXX
3134 */
3135 printf("Process (pid %lu) got signal %d\n",
3136 (u_long)p->p_pid, sig);
3137 #endif
3138 return (SIGSTATUS_IGNORE);
3139 }
3140
3141 /*
3142 * If there is a pending stop signal to process with
3143 * default action, stop here, then clear the signal.
3144 * Traced or exiting processes should ignore stops.
3145 * Additionally, a member of an orphaned process group
3146 * should ignore tty stops.
3147 */
3148 prop = sigprop(sig);
3149 if (prop & SIGPROP_STOP) {
3150 mtx_unlock(&ps->ps_mtx);
3151 if ((p->p_flag & (P_TRACED | P_WEXIT |
3152 P_SINGLE_EXIT)) != 0 || ((p->p_pgrp->
3153 pg_flags & PGRP_ORPHANED) != 0 &&
3154 (prop & SIGPROP_TTYSTOP) != 0)) {
3155 mtx_lock(&ps->ps_mtx);
3156 return (SIGSTATUS_IGNORE);
3157 }
3158 if (TD_SBDRY_INTR(td)) {
3159 KASSERT((td->td_flags & TDF_SBDRY) != 0,
3160 ("lost TDF_SBDRY"));
3161 mtx_lock(&ps->ps_mtx);
3162 return (SIGSTATUS_SBDRY_STOP);
3163 }
3164 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK,
3165 &p->p_mtx.lock_object, "Catching SIGSTOP");
3166 sigqueue_delete(&td->td_sigqueue, sig);
3167 sigqueue_delete(&p->p_sigqueue, sig);
3168 p->p_flag |= P_STOPPED_SIG;
3169 p->p_xsig = sig;
3170 PROC_SLOCK(p);
3171 sig_suspend_threads(td, p, 0);
3172 thread_suspend_switch(td, p);
3173 PROC_SUNLOCK(p);
3174 mtx_lock(&ps->ps_mtx);
3175 return (SIGSTATUS_HANDLED);
3176 } else if ((prop & SIGPROP_IGNORE) != 0 &&
3177 (td->td_flags & TDF_SIGWAIT) == 0) {
3178 /*
3179 * Default action is to ignore; drop it if
3180 * not in kern_sigtimedwait().
3181 */
3182 return (SIGSTATUS_IGNORE);
3183 } else {
3184 return (SIGSTATUS_HANDLE);
3185 }
3186
3187 case (intptr_t)SIG_IGN:
3188 if ((td->td_flags & TDF_SIGWAIT) == 0)
3189 return (SIGSTATUS_IGNORE);
3190 else
3191 return (SIGSTATUS_HANDLE);
3192
3193 default:
3194 /*
3195 * This signal has an action, let postsig() process it.
3196 */
3197 return (SIGSTATUS_HANDLE);
3198 }
3199 }
3200
3201 /*
3202 * If the current process has received a signal (should be caught or cause
3203 * termination, should interrupt current syscall), return the signal number.
3204 * Stop signals with default action are processed immediately, then cleared;
3205 * they aren't returned. This is checked after each entry to the system for
3206 * a syscall or trap (though this can usually be done without calling
3207 * issignal by checking the pending signal masks in cursig.) The normal call
3208 * sequence is
3209 *
3210 * while (sig = cursig(curthread))
3211 * postsig(sig);
3212 */
3213 static int
3214 issignal(struct thread *td)
3215 {
3216 struct proc *p;
3217 sigset_t sigpending;
3218 int sig;
3219
3220 p = td->td_proc;
3221 PROC_LOCK_ASSERT(p, MA_OWNED);
3222
3223 for (;;) {
3224 sigpending = td->td_sigqueue.sq_signals;
3225 SIGSETOR(sigpending, p->p_sigqueue.sq_signals);
3226 SIGSETNAND(sigpending, td->td_sigmask);
3227
3228 if ((p->p_flag & P_PPWAIT) != 0 || (td->td_flags &
3229 (TDF_SBDRY | TDF_SERESTART | TDF_SEINTR)) == TDF_SBDRY)
3230 SIG_STOPSIGMASK(sigpending);
3231 if (SIGISEMPTY(sigpending)) /* no signal to send */
3232 return (0);
3233
3234 /*
3235 * Do fast sigblock if requested by usermode. Since
3236 * we do know that there was a signal pending at this
3237 * point, set the FAST_SIGBLOCK_PEND as indicator for
3238 * usermode to perform a dummy call to
3239 * FAST_SIGBLOCK_UNBLOCK, which causes immediate
3240 * delivery of postponed pending signal.
3241 */
3242 if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
3243 if (td->td_sigblock_val != 0)
3244 SIGSETNAND(sigpending, fastblock_mask);
3245 if (SIGISEMPTY(sigpending)) {
3246 td->td_pflags |= TDP_SIGFASTPENDING;
3247 return (0);
3248 }
3249 }
3250
3251 if ((p->p_flag & (P_TRACED | P_PPTRACE)) == P_TRACED &&
3252 (p->p_flag2 & P2_PTRACE_FSTP) != 0 &&
3253 SIGISMEMBER(sigpending, SIGSTOP)) {
3254 /*
3255 * If debugger just attached, always consume
3256 * SIGSTOP from ptrace(PT_ATTACH) first, to
3257 * execute the debugger attach ritual in
3258 * order.
3259 */
3260 td->td_dbgflags |= TDB_FSTP;
3261 SIGEMPTYSET(sigpending);
3262 SIGADDSET(sigpending, SIGSTOP);
3263 }
3264
3265 SIG_FOREACH(sig, &sigpending) {
3266 switch (sigprocess(td, sig)) {
3267 case SIGSTATUS_HANDLE:
3268 return (sig);
3269 case SIGSTATUS_HANDLED:
3270 goto next;
3271 case SIGSTATUS_IGNORE:
3272 sigqueue_delete(&td->td_sigqueue, sig);
3273 sigqueue_delete(&p->p_sigqueue, sig);
3274 break;
3275 case SIGSTATUS_SBDRY_STOP:
3276 return (-1);
3277 }
3278 }
3279 next:;
3280 }
3281 }
3282
3283 void
3284 thread_stopped(struct proc *p)
3285 {
3286 int n;
3287
3288 PROC_LOCK_ASSERT(p, MA_OWNED);
3289 PROC_SLOCK_ASSERT(p, MA_OWNED);
3290 n = p->p_suspcount;
3291 if (p == curproc)
3292 n++;
3293 if ((p->p_flag & P_STOPPED_SIG) && (n == p->p_numthreads)) {
3294 PROC_SUNLOCK(p);
3295 p->p_flag &= ~P_WAITED;
3296 PROC_LOCK(p->p_pptr);
3297 childproc_stopped(p, (p->p_flag & P_TRACED) ?
3298 CLD_TRAPPED : CLD_STOPPED);
3299 PROC_UNLOCK(p->p_pptr);
3300 PROC_SLOCK(p);
3301 }
3302 }
3303
3304 /*
3305 * Take the action for the specified signal
3306 * from the current set of pending signals.
3307 */
3308 int
3309 postsig(int sig)
3310 {
3311 struct thread *td;
3312 struct proc *p;
3313 struct sigacts *ps;
3314 sig_t action;
3315 ksiginfo_t ksi;
3316 sigset_t returnmask;
3317
3318 KASSERT(sig != 0, ("postsig"));
3319
3320 td = curthread;
3321 p = td->td_proc;
3322 PROC_LOCK_ASSERT(p, MA_OWNED);
3323 ps = p->p_sigacts;
3324 mtx_assert(&ps->ps_mtx, MA_OWNED);
3325 ksiginfo_init(&ksi);
3326 if (sigqueue_get(&td->td_sigqueue, sig, &ksi) == 0 &&
3327 sigqueue_get(&p->p_sigqueue, sig, &ksi) == 0)
3328 return (0);
3329 ksi.ksi_signo = sig;
3330 if (ksi.ksi_code == SI_TIMER)
3331 itimer_accept(p, ksi.ksi_timerid, &ksi);
3332 action = ps->ps_sigact[_SIG_IDX(sig)];
3333 #ifdef KTRACE
3334 if (KTRPOINT(td, KTR_PSIG))
3335 ktrpsig(sig, action, td->td_pflags & TDP_OLDMASK ?
3336 &td->td_oldsigmask : &td->td_sigmask, ksi.ksi_code);
3337 #endif
3338
3339 if (action == SIG_DFL) {
3340 /*
3341 * Default action, where the default is to kill
3342 * the process. (Other cases were ignored above.)
3343 */
3344 mtx_unlock(&ps->ps_mtx);
3345 proc_td_siginfo_capture(td, &ksi.ksi_info);
3346 sigexit(td, sig);
3347 /* NOTREACHED */
3348 } else {
3349 /*
3350 * If we get here, the signal must be caught.
3351 */
3352 KASSERT(action != SIG_IGN, ("postsig action %p", action));
3353 KASSERT(!SIGISMEMBER(td->td_sigmask, sig),
3354 ("postsig action: blocked sig %d", sig));
3355
3356 /*
3357 * Set the new mask value and also defer further
3358 * occurrences of this signal.
3359 *
3360 * Special case: user has done a sigsuspend. Here the
3361 * current mask is not of interest, but rather the
3362 * mask from before the sigsuspend is what we want
3363 * restored after the signal processing is completed.
3364 */
3365 if (td->td_pflags & TDP_OLDMASK) {
3366 returnmask = td->td_oldsigmask;
3367 td->td_pflags &= ~TDP_OLDMASK;
3368 } else
3369 returnmask = td->td_sigmask;
3370
3371 if (p->p_sig == sig) {
3372 p->p_sig = 0;
3373 }
3374 (*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
3375 postsig_done(sig, td, ps);
3376 }
3377 return (1);
3378 }
3379
3380 int
3381 sig_ast_checksusp(struct thread *td)
3382 {
3383 struct proc *p __diagused;
3384 int ret;
3385
3386 p = td->td_proc;
3387 PROC_LOCK_ASSERT(p, MA_OWNED);
3388
3389 if ((td->td_flags & TDF_NEEDSUSPCHK) == 0)
3390 return (0);
3391
3392 ret = thread_suspend_check(1);
3393 MPASS(ret == 0 || ret == EINTR || ret == ERESTART);
3394 return (ret);
3395 }
3396
3397 int
3398 sig_ast_needsigchk(struct thread *td)
3399 {
3400 struct proc *p;
3401 struct sigacts *ps;
3402 int ret, sig;
3403
3404 p = td->td_proc;
3405 PROC_LOCK_ASSERT(p, MA_OWNED);
3406
3407 if ((td->td_flags & TDF_NEEDSIGCHK) == 0)
3408 return (0);
3409
3410 ps = p->p_sigacts;
3411 mtx_lock(&ps->ps_mtx);
3412 sig = cursig(td);
3413 if (sig == -1) {
3414 mtx_unlock(&ps->ps_mtx);
3415 KASSERT((td->td_flags & TDF_SBDRY) != 0, ("lost TDF_SBDRY"));
3416 KASSERT(TD_SBDRY_INTR(td),
3417 ("lost TDF_SERESTART of TDF_SEINTR"));
3418 KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
3419 (TDF_SEINTR | TDF_SERESTART),
3420 ("both TDF_SEINTR and TDF_SERESTART"));
3421 ret = TD_SBDRY_ERRNO(td);
3422 } else if (sig != 0) {
3423 ret = SIGISMEMBER(ps->ps_sigintr, sig) ? EINTR : ERESTART;
3424 mtx_unlock(&ps->ps_mtx);
3425 } else {
3426 mtx_unlock(&ps->ps_mtx);
3427 ret = 0;
3428 }
3429
3430 /*
3431 * Do not go into sleep if this thread was the ptrace(2)
3432 * attach leader. cursig() consumed SIGSTOP from PT_ATTACH,
3433 * but we usually act on the signal by interrupting sleep, and
3434 * should do that here as well.
3435 */
3436 if ((td->td_dbgflags & TDB_FSTP) != 0) {
3437 if (ret == 0)
3438 ret = EINTR;
3439 td->td_dbgflags &= ~TDB_FSTP;
3440 }
3441
3442 return (ret);
3443 }
3444
3445 int
3446 sig_intr(void)
3447 {
3448 struct thread *td;
3449 struct proc *p;
3450 int ret;
3451
3452 td = curthread;
3453 if ((td->td_flags & (TDF_NEEDSIGCHK | TDF_NEEDSUSPCHK)) == 0)
3454 return (0);
3455
3456 p = td->td_proc;
3457
3458 PROC_LOCK(p);
3459 ret = sig_ast_checksusp(td);
3460 if (ret == 0)
3461 ret = sig_ast_needsigchk(td);
3462 PROC_UNLOCK(p);
3463 return (ret);
3464 }
3465
3466 bool
3467 curproc_sigkilled(void)
3468 {
3469 struct thread *td;
3470 struct proc *p;
3471 struct sigacts *ps;
3472 bool res;
3473
3474 td = curthread;
3475 if ((td->td_flags & TDF_NEEDSIGCHK) == 0)
3476 return (false);
3477
3478 p = td->td_proc;
3479 PROC_LOCK(p);
3480 ps = p->p_sigacts;
3481 mtx_lock(&ps->ps_mtx);
3482 res = SIGISMEMBER(td->td_sigqueue.sq_signals, SIGKILL) ||
3483 SIGISMEMBER(p->p_sigqueue.sq_signals, SIGKILL);
3484 mtx_unlock(&ps->ps_mtx);
3485 PROC_UNLOCK(p);
3486 return (res);
3487 }
3488
3489 void
3490 proc_wkilled(struct proc *p)
3491 {
3492
3493 PROC_LOCK_ASSERT(p, MA_OWNED);
3494 if ((p->p_flag & P_WKILLED) == 0) {
3495 p->p_flag |= P_WKILLED;
3496 /*
3497 * Notify swapper that there is a process to swap in.
3498 * The notification is racy, at worst it would take 10
3499 * seconds for the swapper process to notice.
3500 */
3501 if ((p->p_flag & (P_INMEM | P_SWAPPINGIN)) == 0)
3502 wakeup(&proc0);
3503 }
3504 }
3505
3506 /*
3507 * Kill the current process for stated reason.
3508 */
3509 void
3510 killproc(struct proc *p, const char *why)
3511 {
3512
3513 PROC_LOCK_ASSERT(p, MA_OWNED);
3514 CTR3(KTR_PROC, "killproc: proc %p (pid %d, %s)", p, p->p_pid,
3515 p->p_comm);
3516 log(LOG_ERR, "pid %d (%s), jid %d, uid %d, was killed: %s\n",
3517 p->p_pid, p->p_comm, p->p_ucred->cr_prison->pr_id,
3518 p->p_ucred->cr_uid, why);
3519 proc_wkilled(p);
3520 kern_psignal(p, SIGKILL);
3521 }
3522
3523 /*
3524 * Force the current process to exit with the specified signal, dumping core
3525 * if appropriate. We bypass the normal tests for masked and caught signals,
3526 * allowing unrecoverable failures to terminate the process without changing
3527 * signal state. Mark the accounting record with the signal termination.
3528 * If dumping core, save the signal number for the debugger. Calls exit and
3529 * does not return.
3530 */
3531 void
3532 sigexit(struct thread *td, int sig)
3533 {
3534 struct proc *p = td->td_proc;
3535
3536 PROC_LOCK_ASSERT(p, MA_OWNED);
3537 proc_set_p2_wexit(p);
3538
3539 p->p_acflag |= AXSIG;
3540 /*
3541 * We must be single-threading to generate a core dump. This
3542 * ensures that the registers in the core file are up-to-date.
3543 * Also, the ELF dump handler assumes that the thread list doesn't
3544 * change out from under it.
3545 *
3546 * XXX If another thread attempts to single-thread before us
3547 * (e.g. via fork()), we won't get a dump at all.
3548 */
3549 if ((sigprop(sig) & SIGPROP_CORE) &&
3550 thread_single(p, SINGLE_NO_EXIT) == 0) {
3551 p->p_sig = sig;
3552 /*
3553 * Log signals which would cause core dumps
3554 * (Log as LOG_INFO to appease those who don't want
3555 * these messages.)
3556 * XXX : Todo, as well as euid, write out ruid too
3557 * Note that coredump() drops proc lock.
3558 */
3559 if (coredump(td) == 0)
3560 sig |= WCOREFLAG;
3561 if (kern_logsigexit)
3562 log(LOG_INFO,
3563 "pid %d (%s), jid %d, uid %d: exited on "
3564 "signal %d%s\n", p->p_pid, p->p_comm,
3565 p->p_ucred->cr_prison->pr_id,
3566 td->td_ucred->cr_uid,
3567 sig &~ WCOREFLAG,
3568 sig & WCOREFLAG ? " (core dumped)" : "");
3569 } else
3570 PROC_UNLOCK(p);
3571 exit1(td, 0, sig);
3572 /* NOTREACHED */
3573 }
3574
3575 /*
3576 * Send queued SIGCHLD to parent when child process's state
3577 * is changed.
3578 */
3579 static void
3580 sigparent(struct proc *p, int reason, int status)
3581 {
3582 PROC_LOCK_ASSERT(p, MA_OWNED);
3583 PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3584
3585 if (p->p_ksi != NULL) {
3586 p->p_ksi->ksi_signo = SIGCHLD;
3587 p->p_ksi->ksi_code = reason;
3588 p->p_ksi->ksi_status = status;
3589 p->p_ksi->ksi_pid = p->p_pid;
3590 p->p_ksi->ksi_uid = p->p_ucred->cr_ruid;
3591 if (KSI_ONQ(p->p_ksi))
3592 return;
3593 }
3594 pksignal(p->p_pptr, SIGCHLD, p->p_ksi);
3595 }
3596
3597 static void
3598 childproc_jobstate(struct proc *p, int reason, int sig)
3599 {
3600 struct sigacts *ps;
3601
3602 PROC_LOCK_ASSERT(p, MA_OWNED);
3603 PROC_LOCK_ASSERT(p->p_pptr, MA_OWNED);
3604
3605 /*
3606 * Wake up parent sleeping in kern_wait(), also send
3607 * SIGCHLD to parent, but SIGCHLD does not guarantee
3608 * that parent will awake, because parent may masked
3609 * the signal.
3610 */
3611 p->p_pptr->p_flag |= P_STATCHILD;
3612 wakeup(p->p_pptr);
3613
3614 ps = p->p_pptr->p_sigacts;
3615 mtx_lock(&ps->ps_mtx);
3616 if ((ps->ps_flag & PS_NOCLDSTOP) == 0) {
3617 mtx_unlock(&ps->ps_mtx);
3618 sigparent(p, reason, sig);
3619 } else
3620 mtx_unlock(&ps->ps_mtx);
3621 }
3622
3623 void
3624 childproc_stopped(struct proc *p, int reason)
3625 {
3626
3627 childproc_jobstate(p, reason, p->p_xsig);
3628 }
3629
3630 void
3631 childproc_continued(struct proc *p)
3632 {
3633 childproc_jobstate(p, CLD_CONTINUED, SIGCONT);
3634 }
3635
3636 void
3637 childproc_exited(struct proc *p)
3638 {
3639 int reason, status;
3640
3641 if (WCOREDUMP(p->p_xsig)) {
3642 reason = CLD_DUMPED;
3643 status = WTERMSIG(p->p_xsig);
3644 } else if (WIFSIGNALED(p->p_xsig)) {
3645 reason = CLD_KILLED;
3646 status = WTERMSIG(p->p_xsig);
3647 } else {
3648 reason = CLD_EXITED;
3649 status = p->p_xexit;
3650 }
3651 /*
3652 * XXX avoid calling wakeup(p->p_pptr), the work is
3653 * done in exit1().
3654 */
3655 sigparent(p, reason, status);
3656 }
3657
3658 #define MAX_NUM_CORE_FILES 100000
3659 #ifndef NUM_CORE_FILES
3660 #define NUM_CORE_FILES 5
3661 #endif
3662 CTASSERT(NUM_CORE_FILES >= 0 && NUM_CORE_FILES <= MAX_NUM_CORE_FILES);
3663 static int num_cores = NUM_CORE_FILES;
3664
3665 static int
3666 sysctl_debug_num_cores_check (SYSCTL_HANDLER_ARGS)
3667 {
3668 int error;
3669 int new_val;
3670
3671 new_val = num_cores;
3672 error = sysctl_handle_int(oidp, &new_val, 0, req);
3673 if (error != 0 || req->newptr == NULL)
3674 return (error);
3675 if (new_val > MAX_NUM_CORE_FILES)
3676 new_val = MAX_NUM_CORE_FILES;
3677 if (new_val < 0)
3678 new_val = 0;
3679 num_cores = new_val;
3680 return (0);
3681 }
3682 SYSCTL_PROC(_debug, OID_AUTO, ncores,
3683 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
3684 sysctl_debug_num_cores_check, "I",
3685 "Maximum number of generated process corefiles while using index format");
3686
3687 #define GZIP_SUFFIX ".gz"
3688 #define ZSTD_SUFFIX ".zst"
3689
3690 int compress_user_cores = 0;
3691
3692 static int
3693 sysctl_compress_user_cores(SYSCTL_HANDLER_ARGS)
3694 {
3695 int error, val;
3696
3697 val = compress_user_cores;
3698 error = sysctl_handle_int(oidp, &val, 0, req);
3699 if (error != 0 || req->newptr == NULL)
3700 return (error);
3701 if (val != 0 && !compressor_avail(val))
3702 return (EINVAL);
3703 compress_user_cores = val;
3704 return (error);
3705 }
3706 SYSCTL_PROC(_kern, OID_AUTO, compress_user_cores,
3707 CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, 0, sizeof(int),
3708 sysctl_compress_user_cores, "I",
3709 "Enable compression of user corefiles ("
3710 __XSTRING(COMPRESS_GZIP) " = gzip, "
3711 __XSTRING(COMPRESS_ZSTD) " = zstd)");
3712
3713 int compress_user_cores_level = 6;
3714 SYSCTL_INT(_kern, OID_AUTO, compress_user_cores_level, CTLFLAG_RWTUN,
3715 &compress_user_cores_level, 0,
3716 "Corefile compression level");
3717
3718 /*
3719 * Protect the access to corefilename[] by allproc_lock.
3720 */
3721 #define corefilename_lock allproc_lock
3722
3723 static char corefilename[MAXPATHLEN] = {"%N.core"};
3724 TUNABLE_STR("kern.corefile", corefilename, sizeof(corefilename));
3725
3726 static int
3727 sysctl_kern_corefile(SYSCTL_HANDLER_ARGS)
3728 {
3729 int error;
3730
3731 sx_xlock(&corefilename_lock);
3732 error = sysctl_handle_string(oidp, corefilename, sizeof(corefilename),
3733 req);
3734 sx_xunlock(&corefilename_lock);
3735
3736 return (error);
3737 }
3738 SYSCTL_PROC(_kern, OID_AUTO, corefile, CTLTYPE_STRING | CTLFLAG_RW |
3739 CTLFLAG_MPSAFE, 0, 0, sysctl_kern_corefile, "A",
3740 "Process corefile name format string");
3741
3742 static void
3743 vnode_close_locked(struct thread *td, struct vnode *vp)
3744 {
3745
3746 VOP_UNLOCK(vp);
3747 vn_close(vp, FWRITE, td->td_ucred, td);
3748 }
3749
3750 /*
3751 * If the core format has a %I in it, then we need to check
3752 * for existing corefiles before defining a name.
3753 * To do this we iterate over 0..ncores to find a
3754 * non-existing core file name to use. If all core files are
3755 * already used we choose the oldest one.
3756 */
3757 static int
3758 corefile_open_last(struct thread *td, char *name, int indexpos,
3759 int indexlen, int ncores, struct vnode **vpp)
3760 {
3761 struct vnode *oldvp, *nextvp, *vp;
3762 struct vattr vattr;
3763 struct nameidata nd;
3764 int error, i, flags, oflags, cmode;
3765 char ch;
3766 struct timespec lasttime;
3767
3768 nextvp = oldvp = NULL;
3769 cmode = S_IRUSR | S_IWUSR;
3770 oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3771 (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3772
3773 for (i = 0; i < ncores; i++) {
3774 flags = O_CREAT | FWRITE | O_NOFOLLOW;
3775
3776 ch = name[indexpos + indexlen];
3777 (void)snprintf(name + indexpos, indexlen + 1, "%.*u", indexlen,
3778 i);
3779 name[indexpos + indexlen] = ch;
3780
3781 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3782 error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3783 NULL);
3784 if (error != 0)
3785 break;
3786
3787 vp = nd.ni_vp;
3788 NDFREE(&nd, NDF_ONLY_PNBUF);
3789 if ((flags & O_CREAT) == O_CREAT) {
3790 nextvp = vp;
3791 break;
3792 }
3793
3794 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
3795 if (error != 0) {
3796 vnode_close_locked(td, vp);
3797 break;
3798 }
3799
3800 if (oldvp == NULL ||
3801 lasttime.tv_sec > vattr.va_mtime.tv_sec ||
3802 (lasttime.tv_sec == vattr.va_mtime.tv_sec &&
3803 lasttime.tv_nsec >= vattr.va_mtime.tv_nsec)) {
3804 if (oldvp != NULL)
3805 vn_close(oldvp, FWRITE, td->td_ucred, td);
3806 oldvp = vp;
3807 VOP_UNLOCK(oldvp);
3808 lasttime = vattr.va_mtime;
3809 } else {
3810 vnode_close_locked(td, vp);
3811 }
3812 }
3813
3814 if (oldvp != NULL) {
3815 if (nextvp == NULL) {
3816 if ((td->td_proc->p_flag & P_SUGID) != 0) {
3817 error = EFAULT;
3818 vn_close(oldvp, FWRITE, td->td_ucred, td);
3819 } else {
3820 nextvp = oldvp;
3821 error = vn_lock(nextvp, LK_EXCLUSIVE);
3822 if (error != 0) {
3823 vn_close(nextvp, FWRITE, td->td_ucred,
3824 td);
3825 nextvp = NULL;
3826 }
3827 }
3828 } else {
3829 vn_close(oldvp, FWRITE, td->td_ucred, td);
3830 }
3831 }
3832 if (error != 0) {
3833 if (nextvp != NULL)
3834 vnode_close_locked(td, oldvp);
3835 } else {
3836 *vpp = nextvp;
3837 }
3838
3839 return (error);
3840 }
3841
3842 /*
3843 * corefile_open(comm, uid, pid, td, compress, vpp, namep)
3844 * Expand the name described in corefilename, using name, uid, and pid
3845 * and open/create core file.
3846 * corefilename is a printf-like string, with three format specifiers:
3847 * %N name of process ("name")
3848 * %P process id (pid)
3849 * %U user id (uid)
3850 * For example, "%N.core" is the default; they can be disabled completely
3851 * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
3852 * This is controlled by the sysctl variable kern.corefile (see above).
3853 */
3854 static int
3855 corefile_open(const char *comm, uid_t uid, pid_t pid, struct thread *td,
3856 int compress, int signum, struct vnode **vpp, char **namep)
3857 {
3858 struct sbuf sb;
3859 struct nameidata nd;
3860 const char *format;
3861 char *hostname, *name;
3862 int cmode, error, flags, i, indexpos, indexlen, oflags, ncores;
3863
3864 hostname = NULL;
3865 format = corefilename;
3866 name = malloc(MAXPATHLEN, M_TEMP, M_WAITOK | M_ZERO);
3867 indexlen = 0;
3868 indexpos = -1;
3869 ncores = num_cores;
3870 (void)sbuf_new(&sb, name, MAXPATHLEN, SBUF_FIXEDLEN);
3871 sx_slock(&corefilename_lock);
3872 for (i = 0; format[i] != '\0'; i++) {
3873 switch (format[i]) {
3874 case '%': /* Format character */
3875 i++;
3876 switch (format[i]) {
3877 case '%':
3878 sbuf_putc(&sb, '%');
3879 break;
3880 case 'H': /* hostname */
3881 if (hostname == NULL) {
3882 hostname = malloc(MAXHOSTNAMELEN,
3883 M_TEMP, M_WAITOK);
3884 }
3885 getcredhostname(td->td_ucred, hostname,
3886 MAXHOSTNAMELEN);
3887 sbuf_printf(&sb, "%s", hostname);
3888 break;
3889 case 'I': /* autoincrementing index */
3890 if (indexpos != -1) {
3891 sbuf_printf(&sb, "%%I");
3892 break;
3893 }
3894
3895 indexpos = sbuf_len(&sb);
3896 sbuf_printf(&sb, "%u", ncores - 1);
3897 indexlen = sbuf_len(&sb) - indexpos;
3898 break;
3899 case 'N': /* process name */
3900 sbuf_printf(&sb, "%s", comm);
3901 break;
3902 case 'P': /* process id */
3903 sbuf_printf(&sb, "%u", pid);
3904 break;
3905 case 'S': /* signal number */
3906 sbuf_printf(&sb, "%i", signum);
3907 break;
3908 case 'U': /* user id */
3909 sbuf_printf(&sb, "%u", uid);
3910 break;
3911 default:
3912 log(LOG_ERR,
3913 "Unknown format character %c in "
3914 "corename `%s'\n", format[i], format);
3915 break;
3916 }
3917 break;
3918 default:
3919 sbuf_putc(&sb, format[i]);
3920 break;
3921 }
3922 }
3923 sx_sunlock(&corefilename_lock);
3924 free(hostname, M_TEMP);
3925 if (compress == COMPRESS_GZIP)
3926 sbuf_printf(&sb, GZIP_SUFFIX);
3927 else if (compress == COMPRESS_ZSTD)
3928 sbuf_printf(&sb, ZSTD_SUFFIX);
3929 if (sbuf_error(&sb) != 0) {
3930 log(LOG_ERR, "pid %ld (%s), uid (%lu): corename is too "
3931 "long\n", (long)pid, comm, (u_long)uid);
3932 sbuf_delete(&sb);
3933 free(name, M_TEMP);
3934 return (ENOMEM);
3935 }
3936 sbuf_finish(&sb);
3937 sbuf_delete(&sb);
3938
3939 if (indexpos != -1) {
3940 error = corefile_open_last(td, name, indexpos, indexlen, ncores,
3941 vpp);
3942 if (error != 0) {
3943 log(LOG_ERR,
3944 "pid %d (%s), uid (%u): Path `%s' failed "
3945 "on initial open test, error = %d\n",
3946 pid, comm, uid, name, error);
3947 }
3948 } else {
3949 cmode = S_IRUSR | S_IWUSR;
3950 oflags = VN_OPEN_NOAUDIT | VN_OPEN_NAMECACHE |
3951 (capmode_coredump ? VN_OPEN_NOCAPCHECK : 0);
3952 flags = O_CREAT | FWRITE | O_NOFOLLOW;
3953 if ((td->td_proc->p_flag & P_SUGID) != 0)
3954 flags |= O_EXCL;
3955
3956 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, name, td);
3957 error = vn_open_cred(&nd, &flags, cmode, oflags, td->td_ucred,
3958 NULL);
3959 if (error == 0) {
3960 *vpp = nd.ni_vp;
3961 NDFREE(&nd, NDF_ONLY_PNBUF);
3962 }
3963 }
3964
3965 if (error != 0) {
3966 #ifdef AUDIT
3967 audit_proc_coredump(td, name, error);
3968 #endif
3969 free(name, M_TEMP);
3970 return (error);
3971 }
3972 *namep = name;
3973 return (0);
3974 }
3975
3976 /*
3977 * Dump a process' core. The main routine does some
3978 * policy checking, and creates the name of the coredump;
3979 * then it passes on a vnode and a size limit to the process-specific
3980 * coredump routine if there is one; if there _is not_ one, it returns
3981 * ENOSYS; otherwise it returns the error from the process-specific routine.
3982 */
3983
3984 static int
3985 coredump(struct thread *td)
3986 {
3987 struct proc *p = td->td_proc;
3988 struct ucred *cred = td->td_ucred;
3989 struct vnode *vp;
3990 struct flock lf;
3991 struct vattr vattr;
3992 size_t fullpathsize;
3993 int error, error1, locked;
3994 char *name; /* name of corefile */
3995 void *rl_cookie;
3996 off_t limit;
3997 char *fullpath, *freepath = NULL;
3998 struct sbuf *sb;
3999
4000 PROC_LOCK_ASSERT(p, MA_OWNED);
4001 MPASS((p->p_flag & P_HADTHREADS) == 0 || p->p_singlethread == td);
4002
4003 if (!do_coredump || (!sugid_coredump && (p->p_flag & P_SUGID) != 0) ||
4004 (p->p_flag2 & P2_NOTRACE) != 0) {
4005 PROC_UNLOCK(p);
4006 return (EFAULT);
4007 }
4008
4009 /*
4010 * Note that the bulk of limit checking is done after
4011 * the corefile is created. The exception is if the limit
4012 * for corefiles is 0, in which case we don't bother
4013 * creating the corefile at all. This layout means that
4014 * a corefile is truncated instead of not being created,
4015 * if it is larger than the limit.
4016 */
4017 limit = (off_t)lim_cur(td, RLIMIT_CORE);
4018 if (limit == 0 || racct_get_available(p, RACCT_CORE) == 0) {
4019 PROC_UNLOCK(p);
4020 return (EFBIG);
4021 }
4022 PROC_UNLOCK(p);
4023
4024 error = corefile_open(p->p_comm, cred->cr_uid, p->p_pid, td,
4025 compress_user_cores, p->p_sig, &vp, &name);
4026 if (error != 0)
4027 return (error);
4028
4029 /*
4030 * Don't dump to non-regular files or files with links.
4031 * Do not dump into system files. Effective user must own the corefile.
4032 */
4033 if (vp->v_type != VREG || VOP_GETATTR(vp, &vattr, cred) != 0 ||
4034 vattr.va_nlink != 1 || (vp->v_vflag & VV_SYSTEM) != 0 ||
4035 vattr.va_uid != cred->cr_uid) {
4036 VOP_UNLOCK(vp);
4037 error = EFAULT;
4038 goto out;
4039 }
4040
4041 VOP_UNLOCK(vp);
4042
4043 /* Postpone other writers, including core dumps of other processes. */
4044 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
4045
4046 lf.l_whence = SEEK_SET;
4047 lf.l_start = 0;
4048 lf.l_len = 0;
4049 lf.l_type = F_WRLCK;
4050 locked = (VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, F_FLOCK) == 0);
4051
4052 VATTR_NULL(&vattr);
4053 vattr.va_size = 0;
4054 if (set_core_nodump_flag)
4055 vattr.va_flags = UF_NODUMP;
4056 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4057 VOP_SETATTR(vp, &vattr, cred);
4058 VOP_UNLOCK(vp);
4059 PROC_LOCK(p);
4060 p->p_acflag |= ACORE;
4061 PROC_UNLOCK(p);
4062
4063 if (p->p_sysent->sv_coredump != NULL) {
4064 error = p->p_sysent->sv_coredump(td, vp, limit, 0);
4065 } else {
4066 error = ENOSYS;
4067 }
4068
4069 if (locked) {
4070 lf.l_type = F_UNLCK;
4071 VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, F_FLOCK);
4072 }
4073 vn_rangelock_unlock(vp, rl_cookie);
4074
4075 /*
4076 * Notify the userland helper that a process triggered a core dump.
4077 * This allows the helper to run an automated debugging session.
4078 */
4079 if (error != 0 || coredump_devctl == 0)
4080 goto out;
4081 sb = sbuf_new_auto();
4082 if (vn_fullpath_global(p->p_textvp, &fullpath, &freepath) != 0)
4083 goto out2;
4084 sbuf_printf(sb, "comm=\"");
4085 devctl_safe_quote_sb(sb, fullpath);
4086 free(freepath, M_TEMP);
4087 sbuf_printf(sb, "\" core=\"");
4088
4089 /*
4090 * We can't lookup core file vp directly. When we're replacing a core, and
4091 * other random times, we flush the name cache, so it will fail. Instead,
4092 * if the path of the core is relative, add the current dir in front if it.
4093 */
4094 if (name[0] != '/') {
4095 fullpathsize = MAXPATHLEN;
4096 freepath = malloc(fullpathsize, M_TEMP, M_WAITOK);
4097 if (vn_getcwd(freepath, &fullpath, &fullpathsize) != 0) {
4098 free(freepath, M_TEMP);
4099 goto out2;
4100 }
4101 devctl_safe_quote_sb(sb, fullpath);
4102 free(freepath, M_TEMP);
4103 sbuf_putc(sb, '/');
4104 }
4105 devctl_safe_quote_sb(sb, name);
4106 sbuf_printf(sb, "\"");
4107 if (sbuf_finish(sb) == 0)
4108 devctl_notify("kernel", "signal", "coredump", sbuf_data(sb));
4109 out2:
4110 sbuf_delete(sb);
4111 out:
4112 error1 = vn_close(vp, FWRITE, cred, td);
4113 if (error == 0)
4114 error = error1;
4115 #ifdef AUDIT
4116 audit_proc_coredump(td, name, error);
4117 #endif
4118 free(name, M_TEMP);
4119 return (error);
4120 }
4121
4122 /*
4123 * Nonexistent system call-- signal process (may want to handle it). Flag
4124 * error in case process won't see signal immediately (blocked or ignored).
4125 */
4126 #ifndef _SYS_SYSPROTO_H_
4127 struct nosys_args {
4128 int dummy;
4129 };
4130 #endif
4131 /* ARGSUSED */
4132 int
4133 nosys(struct thread *td, struct nosys_args *args)
4134 {
4135 struct proc *p;
4136
4137 p = td->td_proc;
4138
4139 PROC_LOCK(p);
4140 tdsignal(td, SIGSYS);
4141 PROC_UNLOCK(p);
4142 if (kern_lognosys == 1 || kern_lognosys == 3) {
4143 uprintf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4144 td->td_sa.code);
4145 }
4146 if (kern_lognosys == 2 || kern_lognosys == 3 ||
4147 (p->p_pid == 1 && (kern_lognosys & 3) == 0)) {
4148 printf("pid %d comm %s: nosys %d\n", p->p_pid, p->p_comm,
4149 td->td_sa.code);
4150 }
4151 return (ENOSYS);
4152 }
4153
4154 /*
4155 * Send a SIGIO or SIGURG signal to a process or process group using stored
4156 * credentials rather than those of the current process.
4157 */
4158 void
4159 pgsigio(struct sigio **sigiop, int sig, int checkctty)
4160 {
4161 ksiginfo_t ksi;
4162 struct sigio *sigio;
4163
4164 ksiginfo_init(&ksi);
4165 ksi.ksi_signo = sig;
4166 ksi.ksi_code = SI_KERNEL;
4167
4168 SIGIO_LOCK();
4169 sigio = *sigiop;
4170 if (sigio == NULL) {
4171 SIGIO_UNLOCK();
4172 return;
4173 }
4174 if (sigio->sio_pgid > 0) {
4175 PROC_LOCK(sigio->sio_proc);
4176 if (CANSIGIO(sigio->sio_ucred, sigio->sio_proc->p_ucred))
4177 kern_psignal(sigio->sio_proc, sig);
4178 PROC_UNLOCK(sigio->sio_proc);
4179 } else if (sigio->sio_pgid < 0) {
4180 struct proc *p;
4181
4182 PGRP_LOCK(sigio->sio_pgrp);
4183 LIST_FOREACH(p, &sigio->sio_pgrp->pg_members, p_pglist) {
4184 PROC_LOCK(p);
4185 if (p->p_state == PRS_NORMAL &&
4186 CANSIGIO(sigio->sio_ucred, p->p_ucred) &&
4187 (checkctty == 0 || (p->p_flag & P_CONTROLT)))
4188 kern_psignal(p, sig);
4189 PROC_UNLOCK(p);
4190 }
4191 PGRP_UNLOCK(sigio->sio_pgrp);
4192 }
4193 SIGIO_UNLOCK();
4194 }
4195
4196 static int
4197 filt_sigattach(struct knote *kn)
4198 {
4199 struct proc *p = curproc;
4200
4201 kn->kn_ptr.p_proc = p;
4202 kn->kn_flags |= EV_CLEAR; /* automatically set */
4203
4204 knlist_add(p->p_klist, kn, 0);
4205
4206 return (0);
4207 }
4208
4209 static void
4210 filt_sigdetach(struct knote *kn)
4211 {
4212 struct proc *p = kn->kn_ptr.p_proc;
4213
4214 knlist_remove(p->p_klist, kn, 0);
4215 }
4216
4217 /*
4218 * signal knotes are shared with proc knotes, so we apply a mask to
4219 * the hint in order to differentiate them from process hints. This
4220 * could be avoided by using a signal-specific knote list, but probably
4221 * isn't worth the trouble.
4222 */
4223 static int
4224 filt_signal(struct knote *kn, long hint)
4225 {
4226
4227 if (hint & NOTE_SIGNAL) {
4228 hint &= ~NOTE_SIGNAL;
4229
4230 if (kn->kn_id == hint)
4231 kn->kn_data++;
4232 }
4233 return (kn->kn_data != 0);
4234 }
4235
4236 struct sigacts *
4237 sigacts_alloc(void)
4238 {
4239 struct sigacts *ps;
4240
4241 ps = malloc(sizeof(struct sigacts), M_SUBPROC, M_WAITOK | M_ZERO);
4242 refcount_init(&ps->ps_refcnt, 1);
4243 mtx_init(&ps->ps_mtx, "sigacts", NULL, MTX_DEF);
4244 return (ps);
4245 }
4246
4247 void
4248 sigacts_free(struct sigacts *ps)
4249 {
4250
4251 if (refcount_release(&ps->ps_refcnt) == 0)
4252 return;
4253 mtx_destroy(&ps->ps_mtx);
4254 free(ps, M_SUBPROC);
4255 }
4256
4257 struct sigacts *
4258 sigacts_hold(struct sigacts *ps)
4259 {
4260
4261 refcount_acquire(&ps->ps_refcnt);
4262 return (ps);
4263 }
4264
4265 void
4266 sigacts_copy(struct sigacts *dest, struct sigacts *src)
4267 {
4268
4269 KASSERT(dest->ps_refcnt == 1, ("sigacts_copy to shared dest"));
4270 mtx_lock(&src->ps_mtx);
4271 bcopy(src, dest, offsetof(struct sigacts, ps_refcnt));
4272 mtx_unlock(&src->ps_mtx);
4273 }
4274
4275 int
4276 sigacts_shared(struct sigacts *ps)
4277 {
4278
4279 return (ps->ps_refcnt > 1);
4280 }
4281
4282 void
4283 sig_drop_caught(struct proc *p)
4284 {
4285 int sig;
4286 struct sigacts *ps;
4287
4288 ps = p->p_sigacts;
4289 PROC_LOCK_ASSERT(p, MA_OWNED);
4290 mtx_assert(&ps->ps_mtx, MA_OWNED);
4291 SIG_FOREACH(sig, &ps->ps_sigcatch) {
4292 sigdflt(ps, sig);
4293 if ((sigprop(sig) & SIGPROP_IGNORE) != 0)
4294 sigqueue_delete_proc(p, sig);
4295 }
4296 }
4297
4298 static void
4299 sigfastblock_failed(struct thread *td, bool sendsig, bool write)
4300 {
4301 ksiginfo_t ksi;
4302
4303 /*
4304 * Prevent further fetches and SIGSEGVs, allowing thread to
4305 * issue syscalls despite corruption.
4306 */
4307 sigfastblock_clear(td);
4308
4309 if (!sendsig)
4310 return;
4311 ksiginfo_init_trap(&ksi);
4312 ksi.ksi_signo = SIGSEGV;
4313 ksi.ksi_code = write ? SEGV_ACCERR : SEGV_MAPERR;
4314 ksi.ksi_addr = td->td_sigblock_ptr;
4315 trapsignal(td, &ksi);
4316 }
4317
4318 static bool
4319 sigfastblock_fetch_sig(struct thread *td, bool sendsig, uint32_t *valp)
4320 {
4321 uint32_t res;
4322
4323 if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4324 return (true);
4325 if (fueword32((void *)td->td_sigblock_ptr, &res) == -1) {
4326 sigfastblock_failed(td, sendsig, false);
4327 return (false);
4328 }
4329 *valp = res;
4330 td->td_sigblock_val = res & ~SIGFASTBLOCK_FLAGS;
4331 return (true);
4332 }
4333
4334 static void
4335 sigfastblock_resched(struct thread *td, bool resched)
4336 {
4337 struct proc *p;
4338
4339 if (resched) {
4340 p = td->td_proc;
4341 PROC_LOCK(p);
4342 reschedule_signals(p, td->td_sigmask, 0);
4343 PROC_UNLOCK(p);
4344 }
4345 thread_lock(td);
4346 td->td_flags |= TDF_ASTPENDING | TDF_NEEDSIGCHK;
4347 thread_unlock(td);
4348 }
4349
4350 int
4351 sys_sigfastblock(struct thread *td, struct sigfastblock_args *uap)
4352 {
4353 struct proc *p;
4354 int error, res;
4355 uint32_t oldval;
4356
4357 error = 0;
4358 p = td->td_proc;
4359 switch (uap->cmd) {
4360 case SIGFASTBLOCK_SETPTR:
4361 if ((td->td_pflags & TDP_SIGFASTBLOCK) != 0) {
4362 error = EBUSY;
4363 break;
4364 }
4365 if (((uintptr_t)(uap->ptr) & (sizeof(uint32_t) - 1)) != 0) {
4366 error = EINVAL;
4367 break;
4368 }
4369 td->td_pflags |= TDP_SIGFASTBLOCK;
4370 td->td_sigblock_ptr = uap->ptr;
4371 break;
4372
4373 case SIGFASTBLOCK_UNBLOCK:
4374 if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4375 error = EINVAL;
4376 break;
4377 }
4378
4379 for (;;) {
4380 res = casueword32(td->td_sigblock_ptr,
4381 SIGFASTBLOCK_PEND, &oldval, 0);
4382 if (res == -1) {
4383 error = EFAULT;
4384 sigfastblock_failed(td, false, true);
4385 break;
4386 }
4387 if (res == 0)
4388 break;
4389 MPASS(res == 1);
4390 if (oldval != SIGFASTBLOCK_PEND) {
4391 error = EBUSY;
4392 break;
4393 }
4394 error = thread_check_susp(td, false);
4395 if (error != 0)
4396 break;
4397 }
4398 if (error != 0)
4399 break;
4400
4401 /*
4402 * td_sigblock_val is cleared there, but not on a
4403 * syscall exit. The end effect is that a single
4404 * interruptible sleep, while user sigblock word is
4405 * set, might return EINTR or ERESTART to usermode
4406 * without delivering signal. All further sleeps,
4407 * until userspace clears the word and does
4408 * sigfastblock(UNBLOCK), observe current word and no
4409 * longer get interrupted. It is slight
4410 * non-conformance, with alternative to have read the
4411 * sigblock word on each syscall entry.
4412 */
4413 td->td_sigblock_val = 0;
4414
4415 /*
4416 * Rely on normal ast mechanism to deliver pending
4417 * signals to current thread. But notify others about
4418 * fake unblock.
4419 */
4420 sigfastblock_resched(td, error == 0 && p->p_numthreads != 1);
4421
4422 break;
4423
4424 case SIGFASTBLOCK_UNSETPTR:
4425 if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0) {
4426 error = EINVAL;
4427 break;
4428 }
4429 if (!sigfastblock_fetch_sig(td, false, &oldval)) {
4430 error = EFAULT;
4431 break;
4432 }
4433 if (oldval != 0 && oldval != SIGFASTBLOCK_PEND) {
4434 error = EBUSY;
4435 break;
4436 }
4437 sigfastblock_clear(td);
4438 break;
4439
4440 default:
4441 error = EINVAL;
4442 break;
4443 }
4444 return (error);
4445 }
4446
4447 void
4448 sigfastblock_clear(struct thread *td)
4449 {
4450 bool resched;
4451
4452 if ((td->td_pflags & TDP_SIGFASTBLOCK) == 0)
4453 return;
4454 td->td_sigblock_val = 0;
4455 resched = (td->td_pflags & TDP_SIGFASTPENDING) != 0 ||
4456 SIGPENDING(td);
4457 td->td_pflags &= ~(TDP_SIGFASTBLOCK | TDP_SIGFASTPENDING);
4458 sigfastblock_resched(td, resched);
4459 }
4460
4461 void
4462 sigfastblock_fetch(struct thread *td)
4463 {
4464 uint32_t val;
4465
4466 (void)sigfastblock_fetch_sig(td, true, &val);
4467 }
4468
4469 static void
4470 sigfastblock_setpend1(struct thread *td)
4471 {
4472 int res;
4473 uint32_t oldval;
4474
4475 if ((td->td_pflags & TDP_SIGFASTPENDING) == 0)
4476 return;
4477 res = fueword32((void *)td->td_sigblock_ptr, &oldval);
4478 if (res == -1) {
4479 sigfastblock_failed(td, true, false);
4480 return;
4481 }
4482 for (;;) {
4483 res = casueword32(td->td_sigblock_ptr, oldval, &oldval,
4484 oldval | SIGFASTBLOCK_PEND);
4485 if (res == -1) {
4486 sigfastblock_failed(td, true, true);
4487 return;
4488 }
4489 if (res == 0) {
4490 td->td_sigblock_val = oldval & ~SIGFASTBLOCK_FLAGS;
4491 td->td_pflags &= ~TDP_SIGFASTPENDING;
4492 break;
4493 }
4494 MPASS(res == 1);
4495 if (thread_check_susp(td, false) != 0)
4496 break;
4497 }
4498 }
4499
4500 void
4501 sigfastblock_setpend(struct thread *td, bool resched)
4502 {
4503 struct proc *p;
4504
4505 sigfastblock_setpend1(td);
4506 if (resched) {
4507 p = td->td_proc;
4508 PROC_LOCK(p);
4509 reschedule_signals(p, fastblock_mask, SIGPROCMASK_FASTBLK);
4510 PROC_UNLOCK(p);
4511 }
4512 }
Cache object: 3352f2259d6a6eb0f2d14cf04c650a1b
|