?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
kernel - Optimize lwp-specific signaling.
[dragonfly.git] / sys / kern / kern_sig.c
blob06f14e303393342d24c998ad5b7dbb850a7942b0
1 /*
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)kern_sig.c 8.7 (Berkeley) 4/18/94
35 * $FreeBSD: src/sys/kern/kern_sig.c,v 1.72.2.17 2003/05/16 16:34:34 obrien Exp $
36 */
37
38 #include "opt_ktrace.h"
39
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/sysproto.h>
44 #include <sys/signalvar.h>
45 #include <sys/resourcevar.h>
46 #include <sys/vnode.h>
47 #include <sys/event.h>
48 #include <sys/proc.h>
49 #include <sys/nlookup.h>
50 #include <sys/pioctl.h>
51 #include <sys/acct.h>
52 #include <sys/fcntl.h>
53 #include <sys/lock.h>
54 #include <sys/wait.h>
55 #include <sys/ktrace.h>
56 #include <sys/syslog.h>
57 #include <sys/stat.h>
58 #include <sys/sysent.h>
59 #include <sys/sysctl.h>
60 #include <sys/malloc.h>
61 #include <sys/interrupt.h>
62 #include <sys/unistd.h>
63 #include <sys/kern_syscall.h>
64 #include <sys/vkernel.h>
65
66 #include <sys/signal2.h>
67 #include <sys/thread2.h>
68 #include <sys/spinlock2.h>
69
70 #include <machine/cpu.h>
71 #include <machine/smp.h>
72
73 static int coredump(struct lwp *, int);
74 static char *expand_name(const char *, uid_t, pid_t);
75 static int dokillpg(int sig, int pgid, int all);
76 static int sig_ffs(sigset_t *set);
77 static int sigprop(int sig);
78 static void lwp_signotify(struct lwp *lp);
79 static void lwp_signotify_remote(void *arg);
80 static int kern_sigtimedwait(sigset_t set, siginfo_t *info,
81 struct timespec *timeout);
82 static void proc_stopwait(struct proc *p);
83
84 static int filt_sigattach(struct knote *kn);
85 static void filt_sigdetach(struct knote *kn);
86 static int filt_signal(struct knote *kn, long hint);
87
88 struct filterops sig_filtops =
89 { FILTEROP_MPSAFE, filt_sigattach, filt_sigdetach, filt_signal };
90
91 static int kern_logsigexit = 1;
92 SYSCTL_INT(_kern, KERN_LOGSIGEXIT, logsigexit, CTLFLAG_RW,
93 &kern_logsigexit, 0,
94 "Log processes quitting on abnormal signals to syslog(3)");
95
96 /*
97 * Can process p, with pcred pc, send the signal sig to process q?
98 */
99 #define CANSIGNAL(q, sig) \
100 (!p_trespass(curproc->p_ucred, (q)->p_ucred) || \
101 ((sig) == SIGCONT && (q)->p_session == curproc->p_session))
102
103 /*
104 * Policy -- Can real uid ruid with ucred uc send a signal to process q?
105 */
106 #define CANSIGIO(ruid, uc, q) \
107 ((uc)->cr_uid == 0 || \
108 (ruid) == (q)->p_ucred->cr_ruid || \
109 (uc)->cr_uid == (q)->p_ucred->cr_ruid || \
110 (ruid) == (q)->p_ucred->cr_uid || \
111 (uc)->cr_uid == (q)->p_ucred->cr_uid)
112
113 int sugid_coredump;
114 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
115 &sugid_coredump, 0, "Enable coredumping set user/group ID processes");
116
117 static int do_coredump = 1;
118 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
119 &do_coredump, 0, "Enable/Disable coredumps");
120
121 /*
122 * Signal properties and actions.
123 * The array below categorizes the signals and their default actions
124 * according to the following properties:
125 */
126 #define SA_KILL 0x01 /* terminates process by default */
127 #define SA_CORE 0x02 /* ditto and coredumps */
128 #define SA_STOP 0x04 /* suspend process */
129 #define SA_TTYSTOP 0x08 /* ditto, from tty */
130 #define SA_IGNORE 0x10 /* ignore by default */
131 #define SA_CONT 0x20 /* continue if suspended */
132 #define SA_CANTMASK 0x40 /* non-maskable, catchable */
133 #define SA_CKPT 0x80 /* checkpoint process */
134
135
136 static int sigproptbl[NSIG] = {
137 SA_KILL, /* SIGHUP */
138 SA_KILL, /* SIGINT */
139 SA_KILL|SA_CORE, /* SIGQUIT */
140 SA_KILL|SA_CORE, /* SIGILL */
141 SA_KILL|SA_CORE, /* SIGTRAP */
142 SA_KILL|SA_CORE, /* SIGABRT */
143 SA_KILL|SA_CORE, /* SIGEMT */
144 SA_KILL|SA_CORE, /* SIGFPE */
145 SA_KILL, /* SIGKILL */
146 SA_KILL|SA_CORE, /* SIGBUS */
147 SA_KILL|SA_CORE, /* SIGSEGV */
148 SA_KILL|SA_CORE, /* SIGSYS */
149 SA_KILL, /* SIGPIPE */
150 SA_KILL, /* SIGALRM */
151 SA_KILL, /* SIGTERM */
152 SA_IGNORE, /* SIGURG */
153 SA_STOP, /* SIGSTOP */
154 SA_STOP|SA_TTYSTOP, /* SIGTSTP */
155 SA_IGNORE|SA_CONT, /* SIGCONT */
156 SA_IGNORE, /* SIGCHLD */
157 SA_STOP|SA_TTYSTOP, /* SIGTTIN */
158 SA_STOP|SA_TTYSTOP, /* SIGTTOU */
159 SA_IGNORE, /* SIGIO */
160 SA_KILL, /* SIGXCPU */
161 SA_KILL, /* SIGXFSZ */
162 SA_KILL, /* SIGVTALRM */
163 SA_KILL, /* SIGPROF */
164 SA_IGNORE, /* SIGWINCH */
165 SA_IGNORE, /* SIGINFO */
166 SA_KILL, /* SIGUSR1 */
167 SA_KILL, /* SIGUSR2 */
168 SA_IGNORE, /* SIGTHR */
169 SA_CKPT, /* SIGCKPT */
170 SA_KILL|SA_CKPT, /* SIGCKPTEXIT */
171 SA_IGNORE,
172 SA_IGNORE,
173 SA_IGNORE,
174 SA_IGNORE,
175 SA_IGNORE,
176 SA_IGNORE,
177 SA_IGNORE,
178 SA_IGNORE,
179 SA_IGNORE,
180 SA_IGNORE,
181 SA_IGNORE,
182 SA_IGNORE,
183 SA_IGNORE,
184 SA_IGNORE,
185 SA_IGNORE,
186 SA_IGNORE,
187 SA_IGNORE,
188 SA_IGNORE,
189 SA_IGNORE,
190 SA_IGNORE,
191 SA_IGNORE,
192 SA_IGNORE,
193 SA_IGNORE,
194 SA_IGNORE,
195 SA_IGNORE,
196 SA_IGNORE,
197 SA_IGNORE,
198 SA_IGNORE,
199 SA_IGNORE,
200 SA_IGNORE,
201
202 };
203
204 static __inline int
205 sigprop(int sig)
206 {
207
208 if (sig > 0 && sig < NSIG)
209 return (sigproptbl[_SIG_IDX(sig)]);
210 return (0);
211 }
212
213 static __inline int
214 sig_ffs(sigset_t *set)
215 {
216 int i;
217
218 for (i = 0; i < _SIG_WORDS; i++)
219 if (set->__bits[i])
220 return (ffs(set->__bits[i]) + (i * 32));
221 return (0);
222 }
223
224 /*
225 * No requirements.
226 */
227 int
228 kern_sigaction(int sig, struct sigaction *act, struct sigaction *oact)
229 {
230 struct thread *td = curthread;
231 struct proc *p = td->td_proc;
232 struct lwp *lp;
233 struct sigacts *ps = p->p_sigacts;
234
235 if (sig <= 0 || sig > _SIG_MAXSIG)
236 return (EINVAL);
237
238 lwkt_gettoken(&p->p_token);
239
240 if (oact) {
241 oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
242 oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
243 oact->sa_flags = 0;
244 if (SIGISMEMBER(ps->ps_sigonstack, sig))
245 oact->sa_flags |= SA_ONSTACK;
246 if (!SIGISMEMBER(ps->ps_sigintr, sig))
247 oact->sa_flags |= SA_RESTART;
248 if (SIGISMEMBER(ps->ps_sigreset, sig))
249 oact->sa_flags |= SA_RESETHAND;
250 if (SIGISMEMBER(ps->ps_signodefer, sig))
251 oact->sa_flags |= SA_NODEFER;
252 if (SIGISMEMBER(ps->ps_siginfo, sig))
253 oact->sa_flags |= SA_SIGINFO;
254 if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDSTOP)
255 oact->sa_flags |= SA_NOCLDSTOP;
256 if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDWAIT)
257 oact->sa_flags |= SA_NOCLDWAIT;
258 }
259 if (act) {
260 /*
261 * Check for invalid requests. KILL and STOP cannot be
262 * caught.
263 */
264 if (sig == SIGKILL || sig == SIGSTOP) {
265 if (act->sa_handler != SIG_DFL) {
266 lwkt_reltoken(&p->p_token);
267 return (EINVAL);
268 }
269 }
270
271 /*
272 * Change setting atomically.
273 */
274 ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
275 SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
276 if (act->sa_flags & SA_SIGINFO) {
277 ps->ps_sigact[_SIG_IDX(sig)] =
278 (__sighandler_t *)act->sa_sigaction;
279 SIGADDSET(ps->ps_siginfo, sig);
280 } else {
281 ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
282 SIGDELSET(ps->ps_siginfo, sig);
283 }
284 if (!(act->sa_flags & SA_RESTART))
285 SIGADDSET(ps->ps_sigintr, sig);
286 else
287 SIGDELSET(ps->ps_sigintr, sig);
288 if (act->sa_flags & SA_ONSTACK)
289 SIGADDSET(ps->ps_sigonstack, sig);
290 else
291 SIGDELSET(ps->ps_sigonstack, sig);
292 if (act->sa_flags & SA_RESETHAND)
293 SIGADDSET(ps->ps_sigreset, sig);
294 else
295 SIGDELSET(ps->ps_sigreset, sig);
296 if (act->sa_flags & SA_NODEFER)
297 SIGADDSET(ps->ps_signodefer, sig);
298 else
299 SIGDELSET(ps->ps_signodefer, sig);
300 if (sig == SIGCHLD) {
301 if (act->sa_flags & SA_NOCLDSTOP)
302 p->p_sigacts->ps_flag |= PS_NOCLDSTOP;
303 else
304 p->p_sigacts->ps_flag &= ~PS_NOCLDSTOP;
305 if (act->sa_flags & SA_NOCLDWAIT) {
306 /*
307 * Paranoia: since SA_NOCLDWAIT is implemented
308 * by reparenting the dying child to PID 1 (and
309 * trust it to reap the zombie), PID 1 itself
310 * is forbidden to set SA_NOCLDWAIT.
311 */
312 if (p->p_pid == 1)
313 p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
314 else
315 p->p_sigacts->ps_flag |= PS_NOCLDWAIT;
316 } else {
317 p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
318 }
319 if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
320 ps->ps_flag |= PS_CLDSIGIGN;
321 else
322 ps->ps_flag &= ~PS_CLDSIGIGN;
323 }
324 /*
325 * Set bit in p_sigignore for signals that are set to SIG_IGN,
326 * and for signals set to SIG_DFL where the default is to
327 * ignore. However, don't put SIGCONT in p_sigignore, as we
328 * have to restart the process.
329 *
330 * Also remove the signal from the process and lwp signal
331 * list.
332 */
333 if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
334 (sigprop(sig) & SA_IGNORE &&
335 ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
336 SIGDELSET_ATOMIC(p->p_siglist, sig);
337 FOREACH_LWP_IN_PROC(lp, p) {
338 spin_lock(&lp->lwp_spin);
339 SIGDELSET(lp->lwp_siglist, sig);
340 spin_unlock(&lp->lwp_spin);
341 }
342 if (sig != SIGCONT) {
343 /* easier in ksignal */
344 SIGADDSET(p->p_sigignore, sig);
345 }
346 SIGDELSET(p->p_sigcatch, sig);
347 } else {
348 SIGDELSET(p->p_sigignore, sig);
349 if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
350 SIGDELSET(p->p_sigcatch, sig);
351 else
352 SIGADDSET(p->p_sigcatch, sig);
353 }
354 }
355 lwkt_reltoken(&p->p_token);
356 return (0);
357 }
358
359 int
360 sys_sigaction(struct sigaction_args *uap)
361 {
362 struct sigaction act, oact;
363 struct sigaction *actp, *oactp;
364 int error;
365
366 actp = (uap->act != NULL) ? &act : NULL;
367 oactp = (uap->oact != NULL) ? &oact : NULL;
368 if (actp) {
369 error = copyin(uap->act, actp, sizeof(act));
370 if (error)
371 return (error);
372 }
373 error = kern_sigaction(uap->sig, actp, oactp);
374 if (oactp && !error) {
375 error = copyout(oactp, uap->oact, sizeof(oact));
376 }
377 return (error);
378 }
379
380 /*
381 * Initialize signal state for process 0;
382 * set to ignore signals that are ignored by default.
383 */
384 void
385 siginit(struct proc *p)
386 {
387 int i;
388
389 for (i = 1; i <= NSIG; i++)
390 if (sigprop(i) & SA_IGNORE && i != SIGCONT)
391 SIGADDSET(p->p_sigignore, i);
392 }
393
394 /*
395 * Reset signals for an exec of the specified process.
396 */
397 void
398 execsigs(struct proc *p)
399 {
400 struct sigacts *ps = p->p_sigacts;
401 struct lwp *lp;
402 int sig;
403
404 lp = ONLY_LWP_IN_PROC(p);
405
406 /*
407 * Reset caught signals. Held signals remain held
408 * through p_sigmask (unless they were caught,
409 * and are now ignored by default).
410 */
411 while (SIGNOTEMPTY(p->p_sigcatch)) {
412 sig = sig_ffs(&p->p_sigcatch);
413 SIGDELSET(p->p_sigcatch, sig);
414 if (sigprop(sig) & SA_IGNORE) {
415 if (sig != SIGCONT)
416 SIGADDSET(p->p_sigignore, sig);
417 SIGDELSET_ATOMIC(p->p_siglist, sig);
418 /* don't need spinlock */
419 SIGDELSET(lp->lwp_siglist, sig);
420 }
421 ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
422 }
423
424 /*
425 * Reset stack state to the user stack.
426 * Clear set of signals caught on the signal stack.
427 */
428 lp->lwp_sigstk.ss_flags = SS_DISABLE;
429 lp->lwp_sigstk.ss_size = 0;
430 lp->lwp_sigstk.ss_sp = NULL;
431 lp->lwp_flags &= ~LWP_ALTSTACK;
432 /*
433 * Reset no zombies if child dies flag as Solaris does.
434 */
435 p->p_sigacts->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
436 if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
437 ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
438 }
439
440 /*
441 * kern_sigprocmask() - MP SAFE ONLY IF p == curproc
442 *
443 * Manipulate signal mask. This routine is MP SAFE *ONLY* if
444 * p == curproc.
445 */
446 int
447 kern_sigprocmask(int how, sigset_t *set, sigset_t *oset)
448 {
449 struct thread *td = curthread;
450 struct lwp *lp = td->td_lwp;
451 struct proc *p = td->td_proc;
452 int error;
453
454 lwkt_gettoken(&p->p_token);
455
456 if (oset != NULL)
457 *oset = lp->lwp_sigmask;
458
459 error = 0;
460 if (set != NULL) {
461 switch (how) {
462 case SIG_BLOCK:
463 SIG_CANTMASK(*set);
464 SIGSETOR(lp->lwp_sigmask, *set);
465 break;
466 case SIG_UNBLOCK:
467 SIGSETNAND(lp->lwp_sigmask, *set);
468 break;
469 case SIG_SETMASK:
470 SIG_CANTMASK(*set);
471 lp->lwp_sigmask = *set;
472 break;
473 default:
474 error = EINVAL;
475 break;
476 }
477 }
478
479 lwkt_reltoken(&p->p_token);
480
481 return (error);
482 }
483
484 /*
485 * sigprocmask()
486 *
487 * MPSAFE
488 */
489 int
490 sys_sigprocmask(struct sigprocmask_args *uap)
491 {
492 sigset_t set, oset;
493 sigset_t *setp, *osetp;
494 int error;
495
496 setp = (uap->set != NULL) ? &set : NULL;
497 osetp = (uap->oset != NULL) ? &oset : NULL;
498 if (setp) {
499 error = copyin(uap->set, setp, sizeof(set));
500 if (error)
501 return (error);
502 }
503 error = kern_sigprocmask(uap->how, setp, osetp);
504 if (osetp && !error) {
505 error = copyout(osetp, uap->oset, sizeof(oset));
506 }
507 return (error);
508 }
509
510 /*
511 * MPSAFE
512 */
513 int
514 kern_sigpending(struct __sigset *set)
515 {
516 struct lwp *lp = curthread->td_lwp;
517
518 *set = lwp_sigpend(lp);
519
520 return (0);
521 }
522
523 /*
524 * MPSAFE
525 */
526 int
527 sys_sigpending(struct sigpending_args *uap)
528 {
529 sigset_t set;
530 int error;
531
532 error = kern_sigpending(&set);
533
534 if (error == 0)
535 error = copyout(&set, uap->set, sizeof(set));
536 return (error);
537 }
538
539 /*
540 * Suspend process until signal, providing mask to be set
541 * in the meantime.
542 *
543 * MPSAFE
544 */
545 int
546 kern_sigsuspend(struct __sigset *set)
547 {
548 struct thread *td = curthread;
549 struct lwp *lp = td->td_lwp;
550 struct proc *p = td->td_proc;
551 struct sigacts *ps = p->p_sigacts;
552
553 /*
554 * When returning from sigsuspend, we want
555 * the old mask to be restored after the
556 * signal handler has finished. Thus, we
557 * save it here and mark the sigacts structure
558 * to indicate this.
559 */
560 lp->lwp_oldsigmask = lp->lwp_sigmask;
561 lp->lwp_flags |= LWP_OLDMASK;
562
563 SIG_CANTMASK(*set);
564 lp->lwp_sigmask = *set;
565 while (tsleep(ps, PCATCH, "pause", 0) == 0)
566 /* void */;
567 /* always return EINTR rather than ERESTART... */
568 return (EINTR);
569 }
570
571 /*
572 * Note nonstandard calling convention: libc stub passes mask, not
573 * pointer, to save a copyin.
574 *
575 * MPSAFE
576 */
577 int
578 sys_sigsuspend(struct sigsuspend_args *uap)
579 {
580 sigset_t mask;
581 int error;
582
583 error = copyin(uap->sigmask, &mask, sizeof(mask));
584 if (error)
585 return (error);
586
587 error = kern_sigsuspend(&mask);
588
589 return (error);
590 }
591
592 /*
593 * MPSAFE
594 */
595 int
596 kern_sigaltstack(struct sigaltstack *ss, struct sigaltstack *oss)
597 {
598 struct thread *td = curthread;
599 struct lwp *lp = td->td_lwp;
600 struct proc *p = td->td_proc;
601
602 if ((lp->lwp_flags & LWP_ALTSTACK) == 0)
603 lp->lwp_sigstk.ss_flags |= SS_DISABLE;
604
605 if (oss)
606 *oss = lp->lwp_sigstk;
607
608 if (ss) {
609 if (ss->ss_flags & ~SS_DISABLE)
610 return (EINVAL);
611 if (ss->ss_flags & SS_DISABLE) {
612 if (lp->lwp_sigstk.ss_flags & SS_ONSTACK)
613 return (EPERM);
614 lp->lwp_flags &= ~LWP_ALTSTACK;
615 lp->lwp_sigstk.ss_flags = ss->ss_flags;
616 } else {
617 if (ss->ss_size < p->p_sysent->sv_minsigstksz)
618 return (ENOMEM);
619 lp->lwp_flags |= LWP_ALTSTACK;
620 lp->lwp_sigstk = *ss;
621 }
622 }
623
624 return (0);
625 }
626
627 /*
628 * MPSAFE
629 */
630 int
631 sys_sigaltstack(struct sigaltstack_args *uap)
632 {
633 stack_t ss, oss;
634 int error;
635
636 if (uap->ss) {
637 error = copyin(uap->ss, &ss, sizeof(ss));
638 if (error)
639 return (error);
640 }
641
642 error = kern_sigaltstack(uap->ss ? &ss : NULL, uap->oss ? &oss : NULL);
643
644 if (error == 0 && uap->oss)
645 error = copyout(&oss, uap->oss, sizeof(*uap->oss));
646 return (error);
647 }
648
649 /*
650 * Common code for kill process group/broadcast kill.
651 * cp is calling process.
652 */
653 struct killpg_info {
654 int nfound;
655 int sig;
656 };
657
658 static int killpg_all_callback(struct proc *p, void *data);
659
660 static int
661 dokillpg(int sig, int pgid, int all)
662 {
663 struct killpg_info info;
664 struct proc *cp = curproc;
665 struct proc *p;
666 struct pgrp *pgrp;
667
668 info.nfound = 0;
669 info.sig = sig;
670
671 if (all) {
672 /*
673 * broadcast
674 */
675 allproc_scan(killpg_all_callback, &info);
676 } else {
677 if (pgid == 0) {
678 /*
679 * zero pgid means send to my process group.
680 */
681 pgrp = cp->p_pgrp;
682 pgref(pgrp);
683 } else {
684 pgrp = pgfind(pgid);
685 if (pgrp == NULL)
686 return (ESRCH);
687 }
688
689 /*
690 * Must interlock all signals against fork
691 */
692 lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
693 LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
694 if (p->p_pid <= 1 ||
695 p->p_stat == SZOMB ||
696 (p->p_flags & P_SYSTEM) ||
697 !CANSIGNAL(p, sig)) {
698 continue;
699 }
700 ++info.nfound;
701 if (sig)
702 ksignal(p, sig);
703 }
704 lockmgr(&pgrp->pg_lock, LK_RELEASE);
705 pgrel(pgrp);
706 }
707 return (info.nfound ? 0 : ESRCH);
708 }
709
710 static int
711 killpg_all_callback(struct proc *p, void *data)
712 {
713 struct killpg_info *info = data;
714
715 if (p->p_pid <= 1 || (p->p_flags & P_SYSTEM) ||
716 p == curproc || !CANSIGNAL(p, info->sig)) {
717 return (0);
718 }
719 ++info->nfound;
720 if (info->sig)
721 ksignal(p, info->sig);
722 return(0);
723 }
724
725 /*
726 * Send a general signal to a process or LWPs within that process.
727 *
728 * Note that new signals cannot be sent if a process is exiting or already
729 * a zombie, but we return success anyway as userland is likely to not handle
730 * the race properly.
731 *
732 * No requirements.
733 */
734 int
735 kern_kill(int sig, pid_t pid, lwpid_t tid)
736 {
737 int t;
738
739 if ((u_int)sig > _SIG_MAXSIG)
740 return (EINVAL);
741
742 if (pid > 0) {
743 struct proc *p;
744 struct lwp *lp = NULL;
745
746 /*
747 * Send a signal to a single process. If the kill() is
748 * racing an exiting process which has not yet been reaped
749 * act as though the signal was delivered successfully but
750 * don't actually try to deliver the signal.
751 */
752 if ((p = pfind(pid)) == NULL) {
753 if ((p = zpfind(pid)) == NULL)
754 return (ESRCH);
755 PRELE(p);
756 return (0);
757 }
758 lwkt_gettoken(&p->p_token);
759 if (!CANSIGNAL(p, sig)) {
760 lwkt_reltoken(&p->p_token);
761 PRELE(p);
762 return (EPERM);
763 }
764
765 /*
766 * NOP if the process is exiting. Note that lwpsignal() is
767 * called directly with P_WEXIT set to kill individual LWPs
768 * during exit, which is allowed.
769 */
770 if (p->p_flags & P_WEXIT) {
771 lwkt_reltoken(&p->p_token);
772 PRELE(p);
773 return (0);
774 }
775 if (tid != -1) {
776 lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
777 if (lp == NULL) {
778 lwkt_reltoken(&p->p_token);
779 PRELE(p);
780 return (ESRCH);
781 }
782 }
783 if (sig)
784 lwpsignal(p, lp, sig);
785 lwkt_reltoken(&p->p_token);
786 PRELE(p);
787
788 return (0);
789 }
790
791 /*
792 * If we come here, pid is a special broadcast pid.
793 * This doesn't mix with a tid.
794 */
795 if (tid != -1)
796 return (EINVAL);
797
798 switch (pid) {
799 case -1: /* broadcast signal */
800 t = (dokillpg(sig, 0, 1));
801 break;
802 case 0: /* signal own process group */
803 t = (dokillpg(sig, 0, 0));
804 break;
805 default: /* negative explicit process group */
806 t = (dokillpg(sig, -pid, 0));
807 break;
808 }
809 return t;
810 }
811
812 int
813 sys_kill(struct kill_args *uap)
814 {
815 int error;
816
817 error = kern_kill(uap->signum, uap->pid, -1);
818 return (error);
819 }
820
821 int
822 sys_lwp_kill(struct lwp_kill_args *uap)
823 {
824 int error;
825 pid_t pid = uap->pid;
826
827 /*
828 * A tid is mandatory for lwp_kill(), otherwise
829 * you could simply use kill().
830 */
831 if (uap->tid == -1)
832 return (EINVAL);
833
834 /*
835 * To save on a getpid() function call for intra-process
836 * signals, pid == -1 means current process.
837 */
838 if (pid == -1)
839 pid = curproc->p_pid;
840
841 error = kern_kill(uap->signum, pid, uap->tid);
842 return (error);
843 }
844
845 /*
846 * Send a signal to a process group.
847 */
848 void
849 gsignal(int pgid, int sig)
850 {
851 struct pgrp *pgrp;
852
853 if (pgid && (pgrp = pgfind(pgid)))
854 pgsignal(pgrp, sig, 0);
855 }
856
857 /*
858 * Send a signal to a process group. If checktty is 1,
859 * limit to members which have a controlling terminal.
860 *
861 * pg_lock interlocks against a fork that might be in progress, to
862 * ensure that the new child process picks up the signal.
863 */
864 void
865 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
866 {
867 struct proc *p;
868
869 /*
870 * Must interlock all signals against fork
871 */
872 if (pgrp) {
873 pgref(pgrp);
874 lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
875 LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
876 if (checkctty == 0 || p->p_flags & P_CONTROLT)
877 ksignal(p, sig);
878 }
879 lockmgr(&pgrp->pg_lock, LK_RELEASE);
880 pgrel(pgrp);
881 }
882 }
883
884 /*
885 * Send a signal caused by a trap to the current lwp. If it will be caught
886 * immediately, deliver it with correct code. Otherwise, post it normally.
887 *
888 * These signals may ONLY be delivered to the specified lwp and may never
889 * be delivered to the process generically.
890 */
891 void
892 trapsignal(struct lwp *lp, int sig, u_long code)
893 {
894 struct proc *p = lp->lwp_proc;
895 struct sigacts *ps = p->p_sigacts;
896
897 /*
898 * If we are a virtual kernel running an emulated user process
899 * context, switch back to the virtual kernel context before
900 * trying to post the signal.
901 */
902 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
903 struct trapframe *tf = lp->lwp_md.md_regs;
904 tf->tf_trapno = 0;
905 vkernel_trap(lp, tf);
906 }
907
908
909 if ((p->p_flags & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
910 !SIGISMEMBER(lp->lwp_sigmask, sig)) {
911 lp->lwp_ru.ru_nsignals++;
912 #ifdef KTRACE
913 if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
914 ktrpsig(lp, sig, ps->ps_sigact[_SIG_IDX(sig)],
915 &lp->lwp_sigmask, code);
916 #endif
917 (*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
918 &lp->lwp_sigmask, code);
919 SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
920 if (!SIGISMEMBER(ps->ps_signodefer, sig))
921 SIGADDSET(lp->lwp_sigmask, sig);
922 if (SIGISMEMBER(ps->ps_sigreset, sig)) {
923 /*
924 * See kern_sigaction() for origin of this code.
925 */
926 SIGDELSET(p->p_sigcatch, sig);
927 if (sig != SIGCONT &&
928 sigprop(sig) & SA_IGNORE)
929 SIGADDSET(p->p_sigignore, sig);
930 ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
931 }
932 } else {
933 lp->lwp_code = code; /* XXX for core dump/debugger */
934 lp->lwp_sig = sig; /* XXX to verify code */
935 lwpsignal(p, lp, sig);
936 }
937 }
938
939 /*
940 * Find a suitable lwp to deliver the signal to. Returns NULL if all
941 * lwps hold the signal blocked.
942 *
943 * Caller must hold p->p_token.
944 *
945 * Returns a lp or NULL. If non-NULL the lp is held and its token is
946 * acquired.
947 */
948 static struct lwp *
949 find_lwp_for_signal(struct proc *p, int sig)
950 {
951 struct lwp *lp;
952 struct lwp *run, *sleep, *stop;
953
954 /*
955 * If the running/preempted thread belongs to the proc to which
956 * the signal is being delivered and this thread does not block
957 * the signal, then we can avoid a context switch by delivering
958 * the signal to this thread, because it will return to userland
959 * soon anyways.
960 */
961 lp = lwkt_preempted_proc();
962 if (lp != NULL && lp->lwp_proc == p) {
963 LWPHOLD(lp);
964 lwkt_gettoken(&lp->lwp_token);
965 if (!SIGISMEMBER(lp->lwp_sigmask, sig)) {
966 /* return w/ token held */
967 return (lp);
968 }
969 lwkt_reltoken(&lp->lwp_token);
970 LWPRELE(lp);
971 }
972
973 run = sleep = stop = NULL;
974 FOREACH_LWP_IN_PROC(lp, p) {
975 /*
976 * If the signal is being blocked by the lwp, then this
977 * lwp is not eligible for receiving the signal.
978 */
979 LWPHOLD(lp);
980 lwkt_gettoken(&lp->lwp_token);
981
982 if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
983 lwkt_reltoken(&lp->lwp_token);
984 LWPRELE(lp);
985 continue;
986 }
987
988 switch (lp->lwp_stat) {
989 case LSRUN:
990 if (sleep) {
991 lwkt_token_swap();
992 lwkt_reltoken(&sleep->lwp_token);
993 LWPRELE(sleep);
994 sleep = NULL;
995 run = lp;
996 } else if (stop) {
997 lwkt_token_swap();
998 lwkt_reltoken(&stop->lwp_token);
999 LWPRELE(stop);
1000 stop = NULL;
1001 run = lp;
1002 } else {
1003 run = lp;
1004 }
1005 break;
1006 case LSSLEEP:
1007 if (lp->lwp_flags & LWP_SINTR) {
1008 if (sleep) {
1009 lwkt_reltoken(&lp->lwp_token);
1010 LWPRELE(lp);
1011 } else if (stop) {
1012 lwkt_token_swap();
1013 lwkt_reltoken(&stop->lwp_token);
1014 LWPRELE(stop);
1015 stop = NULL;
1016 sleep = lp;
1017 } else {
1018 sleep = lp;
1019 }
1020 } else {
1021 lwkt_reltoken(&lp->lwp_token);
1022 LWPRELE(lp);
1023 }
1024 break;
1025 case LSSTOP:
1026 if (sleep) {
1027 lwkt_reltoken(&lp->lwp_token);
1028 LWPRELE(lp);
1029 } else if (stop) {
1030 lwkt_reltoken(&lp->lwp_token);
1031 LWPRELE(lp);
1032 } else {
1033 stop = lp;
1034 }
1035 break;
1036 }
1037 if (run)
1038 break;
1039 }
1040
1041 if (run != NULL)
1042 return (run);
1043 else if (sleep != NULL)
1044 return (sleep);
1045 else
1046 return (stop);
1047 }
1048
1049 /*
1050 * Send the signal to the process. If the signal has an action, the action
1051 * is usually performed by the target process rather than the caller; we add
1052 * the signal to the set of pending signals for the process.
1053 *
1054 * Exceptions:
1055 * o When a stop signal is sent to a sleeping process that takes the
1056 * default action, the process is stopped without awakening it.
1057 * o SIGCONT restarts stopped processes (or puts them back to sleep)
1058 * regardless of the signal action (eg, blocked or ignored).
1059 *
1060 * Other ignored signals are discarded immediately.
1061 *
1062 * If the caller wishes to call this function from a hard code section the
1063 * caller must already hold p->p_token (see kern_clock.c).
1064 *
1065 * No requirements.
1066 */
1067 void
1068 ksignal(struct proc *p, int sig)
1069 {
1070 lwpsignal(p, NULL, sig);
1071 }
1072
1073 /*
1074 * The core for ksignal. lp may be NULL, then a suitable thread
1075 * will be chosen. If not, lp MUST be a member of p.
1076 *
1077 * If the caller wishes to call this function from a hard code section the
1078 * caller must already hold p->p_token.
1079 *
1080 * No requirements.
1081 */
1082 void
1083 lwpsignal(struct proc *p, struct lwp *lp, int sig)
1084 {
1085 struct proc *q;
1086 sig_t action;
1087 int prop;
1088
1089 if (sig > _SIG_MAXSIG || sig <= 0) {
1090 kprintf("lwpsignal: signal %d\n", sig);
1091 panic("lwpsignal signal number");
1092 }
1093
1094 KKASSERT(lp == NULL || lp->lwp_proc == p);
1095
1096 /*
1097 * We don't want to race... well, all sorts of things. Get appropriate
1098 * tokens.
1099 *
1100 * Don't try to deliver a generic signal to an exiting process,
1101 * the signal structures could be in flux. We check the LWP later
1102 * on.
1103 */
1104 PHOLD(p);
1105 if (lp) {
1106 LWPHOLD(lp);
1107 lwkt_gettoken(&lp->lwp_token);
1108 } else {
1109 lwkt_gettoken(&p->p_token);
1110 if (p->p_flags & P_WEXIT)
1111 goto out;
1112 }
1113
1114 prop = sigprop(sig);
1115
1116 /*
1117 * If proc is traced, always give parent a chance;
1118 * if signal event is tracked by procfs, give *that*
1119 * a chance, as well.
1120 */
1121 if ((p->p_flags & P_TRACED) || (p->p_stops & S_SIG)) {
1122 action = SIG_DFL;
1123 } else {
1124 /*
1125 * Do not try to deliver signals to an exiting lwp other
1126 * than SIGKILL. Note that we must still deliver the signal
1127 * if P_WEXIT is set in the process flags.
1128 */
1129 if (lp && (lp->lwp_mpflags & LWP_MP_WEXIT) && sig != SIGKILL) {
1130 lwkt_reltoken(&lp->lwp_token);
1131 LWPRELE(lp);
1132 PRELE(p);
1133 return;
1134 }
1135
1136 /*
1137 * If the signal is being ignored, then we forget about
1138 * it immediately. NOTE: We don't set SIGCONT in p_sigignore,
1139 * and if it is set to SIG_IGN, action will be SIG_DFL here.
1140 */
1141 if (SIGISMEMBER(p->p_sigignore, sig)) {
1142 /*
1143 * Even if a signal is set SIG_IGN, it may still be
1144 * lurking in a kqueue.
1145 */
1146 KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1147 if (lp) {
1148 lwkt_reltoken(&lp->lwp_token);
1149 LWPRELE(lp);
1150 } else {
1151 lwkt_reltoken(&p->p_token);
1152 }
1153 PRELE(p);
1154 return;
1155 }
1156 if (SIGISMEMBER(p->p_sigcatch, sig))
1157 action = SIG_CATCH;
1158 else
1159 action = SIG_DFL;
1160 }
1161
1162 /*
1163 * If continuing, clear any pending STOP signals for the whole
1164 * process.
1165 */
1166 if (prop & SA_CONT) {
1167 lwkt_gettoken(&p->p_token);
1168 SIG_STOPSIGMASK_ATOMIC(p->p_siglist);
1169 lwkt_reltoken(&p->p_token);
1170 }
1171
1172 if (prop & SA_STOP) {
1173 /*
1174 * If sending a tty stop signal to a member of an orphaned
1175 * process group, discard the signal here if the action
1176 * is default; don't stop the process below if sleeping,
1177 * and don't clear any pending SIGCONT.
1178 */
1179 if ((prop & SA_TTYSTOP) && p->p_pgrp->pg_jobc == 0 &&
1180 action == SIG_DFL) {
1181 if (lp) {
1182 lwkt_reltoken(&lp->lwp_token);
1183 LWPRELE(lp);
1184 } else {
1185 lwkt_reltoken(&p->p_token);
1186 }
1187 PRELE(p);
1188 return;
1189 }
1190 lwkt_gettoken(&p->p_token);
1191 SIG_CONTSIGMASK_ATOMIC(p->p_siglist);
1192 p->p_flags &= ~P_CONTINUED;
1193 lwkt_reltoken(&p->p_token);
1194 }
1195
1196 if (p->p_stat == SSTOP) {
1197 /*
1198 * Nobody can handle this signal, add it to the lwp or
1199 * process pending list
1200 */
1201 lwkt_gettoken(&p->p_token);
1202 if (p->p_stat != SSTOP) {
1203 lwkt_reltoken(&p->p_token);
1204 goto not_stopped;
1205 }
1206 if (lp) {
1207 spin_lock(&lp->lwp_spin);
1208 SIGADDSET(lp->lwp_siglist, sig);
1209 spin_unlock(&lp->lwp_spin);
1210 } else {
1211 SIGADDSET_ATOMIC(p->p_siglist, sig);
1212 }
1213
1214 /*
1215 * If the process is stopped and is being traced, then no
1216 * further action is necessary.
1217 */
1218 if (p->p_flags & P_TRACED) {
1219 lwkt_reltoken(&p->p_token);
1220 goto out;
1221 }
1222
1223 /*
1224 * If the process is stopped and receives a KILL signal,
1225 * make the process runnable.
1226 */
1227 if (sig == SIGKILL) {
1228 proc_unstop(p, SSTOP);
1229 lwkt_reltoken(&p->p_token);
1230 goto active_process;
1231 }
1232
1233 /*
1234 * If the process is stopped and receives a CONT signal,
1235 * then try to make the process runnable again.
1236 */
1237 if (prop & SA_CONT) {
1238 /*
1239 * If SIGCONT is default (or ignored), we continue the
1240 * process but don't leave the signal in p_siglist, as
1241 * it has no further action. If SIGCONT is held, we
1242 * continue the process and leave the signal in
1243 * p_siglist. If the process catches SIGCONT, let it
1244 * handle the signal itself.
1245 *
1246 * XXX what if the signal is being held blocked?
1247 *
1248 * Token required to interlock kern_wait().
1249 * Reparenting can also cause a race so we have to
1250 * hold (q).
1251 */
1252 q = p->p_pptr;
1253 PHOLD(q);
1254 lwkt_gettoken(&q->p_token);
1255 p->p_flags |= P_CONTINUED;
1256 wakeup(q);
1257 if (action == SIG_DFL)
1258 SIGDELSET_ATOMIC(p->p_siglist, sig);
1259 proc_unstop(p, SSTOP);
1260 lwkt_reltoken(&q->p_token);
1261 PRELE(q);
1262 lwkt_reltoken(&p->p_token);
1263 if (action == SIG_CATCH)
1264 goto active_process;
1265 goto out;
1266 }
1267
1268 /*
1269 * If the process is stopped and receives another STOP
1270 * signal, we do not need to stop it again. If we did
1271 * the shell could get confused.
1272 *
1273 * However, if the current/preempted lwp is part of the
1274 * process receiving the signal, we need to keep it,
1275 * so that this lwp can stop in issignal() later, as
1276 * we don't want to wait until it reaches userret!
1277 */
1278 if (prop & SA_STOP) {
1279 lwkt_gettoken(&p->p_token);
1280 if (lwkt_preempted_proc() == NULL ||
1281 lwkt_preempted_proc()->lwp_proc != p) {
1282 SIGDELSET_ATOMIC(p->p_siglist, sig);
1283 }
1284 lwkt_reltoken(&p->p_token);
1285 }
1286
1287 /*
1288 * Otherwise the process is stopped and it received some
1289 * signal, which does not change its stopped state. When
1290 * the process is continued a wakeup(p) will be issued which
1291 * will wakeup any threads sleeping in tstop().
1292 */
1293 goto out;
1294 /* NOTREACHED */
1295 }
1296 not_stopped:
1297 ;
1298 /* else not stopped */
1299 active_process:
1300
1301 /*
1302 * Never deliver a lwp-specific signal to a random lwp.
1303 */
1304 if (lp == NULL) {
1305 /* NOTE: returns lp w/ token held */
1306 lp = find_lwp_for_signal(p, sig);
1307 if (lp) {
1308 if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
1309 lwkt_reltoken(&lp->lwp_token);
1310 LWPRELE(lp);
1311 lp = NULL;
1312 } else {
1313 lwkt_token_swap();
1314 lwkt_reltoken(&p->p_token);
1315 }
1316 }
1317 }
1318
1319 /*
1320 * Deliver to the process generically if (1) the signal is being
1321 * sent to any thread or (2) we could not find a thread to deliver
1322 * it to.
1323 */
1324 if (lp == NULL) {
1325 KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1326 SIGADDSET_ATOMIC(p->p_siglist, sig);
1327 goto out;
1328 }
1329
1330 /*
1331 * Deliver to a specific LWP whether it masks it or not. It will
1332 * not be dispatched if masked but we must still deliver it.
1333 */
1334 if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1335 (p->p_flags & P_TRACED) == 0) {
1336 lwkt_gettoken(&p->p_token);
1337 p->p_nice = NZERO;
1338 lwkt_reltoken(&p->p_token);
1339 }
1340
1341 /*
1342 * If the process receives a STOP signal which indeed needs to
1343 * stop the process, do so. If the process chose to catch the
1344 * signal, it will be treated like any other signal.
1345 */
1346 if ((prop & SA_STOP) && action == SIG_DFL) {
1347 /*
1348 * If a child holding parent blocked, stopping
1349 * could cause deadlock. Take no action at this
1350 * time.
1351 */
1352 lwkt_gettoken(&p->p_token);
1353 if (p->p_flags & P_PPWAIT) {
1354 SIGADDSET_ATOMIC(p->p_siglist, sig);
1355 lwkt_reltoken(&p->p_token);
1356 goto out;
1357 }
1358
1359 /*
1360 * Do not actually try to manipulate the process, but simply
1361 * stop it. Lwps will stop as soon as they safely can.
1362 *
1363 * Ignore stop if the process is exiting.
1364 */
1365 if ((p->p_flags & P_WEXIT) == 0) {
1366 p->p_xstat = sig;
1367 proc_stop(p, SSTOP);
1368 }
1369 lwkt_reltoken(&p->p_token);
1370 goto out;
1371 }
1372
1373 /*
1374 * If it is a CONT signal with default action, just ignore it.
1375 */
1376 if ((prop & SA_CONT) && action == SIG_DFL)
1377 goto out;
1378
1379 /*
1380 * Mark signal pending at this specific thread.
1381 */
1382 spin_lock(&lp->lwp_spin);
1383 SIGADDSET(lp->lwp_siglist, sig);
1384 spin_unlock(&lp->lwp_spin);
1385
1386 lwp_signotify(lp);
1387
1388 out:
1389 if (lp) {
1390 lwkt_reltoken(&lp->lwp_token);
1391 LWPRELE(lp);
1392 } else {
1393 lwkt_reltoken(&p->p_token);
1394 }
1395 PRELE(p);
1396 }
1397
1398 /*
1399 * Notify the LWP that a signal has arrived. The LWP does not have to be
1400 * sleeping on the current cpu.
1401 *
1402 * p->p_token and lp->lwp_token must be held on call.
1403 *
1404 * We can only safely schedule the thread on its current cpu and only if
1405 * one of the SINTR flags is set. If an SINTR flag is set AND we are on
1406 * the correct cpu we are properly interlocked, otherwise we could be
1407 * racing other thread transition states (or the lwp is on the user scheduler
1408 * runq but not scheduled) and must not do anything.
1409 *
1410 * Since we hold the lwp token we know the lwp cannot be ripped out from
1411 * under us so we can safely hold it to prevent it from being ripped out
1412 * from under us if we are forced to IPI another cpu to make the local
1413 * checks there.
1414 *
1415 * Adjustment of lp->lwp_stat can only occur when we hold the lwp_token,
1416 * which we won't in an IPI so any fixups have to be done here, effectively
1417 * replicating part of what setrunnable() does.
1418 */
1419 static void
1420 lwp_signotify(struct lwp *lp)
1421 {
1422 thread_t dtd;
1423
1424 ASSERT_LWKT_TOKEN_HELD(&lp->lwp_token);
1425 dtd = lp->lwp_thread;
1426
1427 crit_enter();
1428 if (lp == lwkt_preempted_proc()) {
1429 /*
1430 * lwp is on the current cpu AND it is currently running
1431 * (we preempted it).
1432 */
1433 signotify();
1434 } else if (lp->lwp_flags & LWP_SINTR) {
1435 /*
1436 * lwp is sitting in tsleep() with PCATCH set
1437 */
1438 if (dtd->td_gd == mycpu) {
1439 setrunnable(lp);
1440 } else {
1441 /*
1442 * We can only adjust lwp_stat while we hold the
1443 * lwp_token, and we won't in the IPI function.
1444 */
1445 LWPHOLD(lp);
1446 if (lp->lwp_stat == LSSTOP)
1447 lp->lwp_stat = LSSLEEP;
1448 lwkt_send_ipiq(dtd->td_gd, lwp_signotify_remote, lp);
1449 }
1450 } else if (dtd->td_flags & TDF_SINTR) {
1451 /*
1452 * lwp is sitting in lwkt_sleep() with PCATCH set.
1453 */
1454 if (dtd->td_gd == mycpu) {
1455 setrunnable(lp);
1456 } else {
1457 /*
1458 * We can only adjust lwp_stat while we hold the
1459 * lwp_token, and we won't in the IPI function.
1460 */
1461 LWPHOLD(lp);
1462 if (lp->lwp_stat == LSSTOP)
1463 lp->lwp_stat = LSSLEEP;
1464 lwkt_send_ipiq(dtd->td_gd, lwp_signotify_remote, lp);
1465 }
1466 } else {
1467 /*
1468 * Otherwise the lwp is either in some uninterruptible state
1469 * or it is on the userland scheduler's runqueue waiting to
1470 * be scheduled to a cpu, or it is running in userland. We
1471 * generally want to send an IPI so a running target gets the
1472 * signal ASAP, otherwise a scheduler-tick worth of latency
1473 * will occur.
1474 *
1475 * Issue an IPI to the remote cpu to knock it into the kernel,
1476 * remote cpu will issue the cpu-local signotify() if the IPI
1477 * preempts the desired thread.
1478 */
1479 if (dtd->td_gd != mycpu) {
1480 LWPHOLD(lp);
1481 lwkt_send_ipiq(dtd->td_gd, lwp_signotify_remote, lp);
1482 }
1483 }
1484 crit_exit();
1485 }
1486
1487 /*
1488 * This function is called via an IPI so we cannot call setrunnable() here
1489 * (because while we hold the lp we don't own its token, and can't get it
1490 * from an IPI).
1491 *
1492 * We are interlocked by virtue of being on the same cpu as the target. If
1493 * we still are and LWP_SINTR or TDF_SINTR is set we can safely schedule
1494 * the target thread.
1495 */
1496 static void
1497 lwp_signotify_remote(void *arg)
1498 {
1499 struct lwp *lp = arg;
1500 thread_t td = lp->lwp_thread;
1501
1502 if (lp == lwkt_preempted_proc()) {
1503 signotify();
1504 LWPRELE(lp);
1505 } else if (td->td_gd == mycpu) {
1506 if ((lp->lwp_flags & LWP_SINTR) ||
1507 (td->td_flags & TDF_SINTR)) {
1508 lwkt_schedule(td);
1509 }
1510 LWPRELE(lp);
1511 } else {
1512 lwkt_send_ipiq(td->td_gd, lwp_signotify_remote, lp);
1513 /* LWPHOLD() is forwarded to the target cpu */
1514 }
1515 }
1516
1517 /*
1518 * Caller must hold p->p_token
1519 */
1520 void
1521 proc_stop(struct proc *p, int sig)
1522 {
1523 struct proc *q;
1524 struct lwp *lp;
1525
1526 ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1527
1528 /*
1529 * If somebody raced us, be happy with it. SCORE overrides SSTOP.
1530 */
1531 if (sig == SCORE) {
1532 if (p->p_stat == SCORE || p->p_stat == SZOMB)
1533 return;
1534 } else {
1535 if (p->p_stat == SSTOP || p->p_stat == SCORE ||
1536 p->p_stat == SZOMB) {
1537 return;
1538 }
1539 }
1540 p->p_stat = sig;
1541
1542 FOREACH_LWP_IN_PROC(lp, p) {
1543 LWPHOLD(lp);
1544 lwkt_gettoken(&lp->lwp_token);
1545
1546 switch (lp->lwp_stat) {
1547 case LSSTOP:
1548 /*
1549 * Do nothing, we are already counted in
1550 * p_nstopped.
1551 */
1552 break;
1553
1554 case LSSLEEP:
1555 /*
1556 * We're sleeping, but we will stop before
1557 * returning to userspace, so count us
1558 * as stopped as well. We set LWP_MP_WSTOP
1559 * to signal the lwp that it should not
1560 * increase p_nstopped when reaching tstop().
1561 *
1562 * LWP_MP_WSTOP is protected by lp->lwp_token.
1563 */
1564 if ((lp->lwp_mpflags & LWP_MP_WSTOP) == 0) {
1565 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
1566 ++p->p_nstopped;
1567 }
1568 break;
1569
1570 case LSRUN:
1571 /*
1572 * We might notify ourself, but that's not
1573 * a problem.
1574 */
1575 lwp_signotify(lp);
1576 break;
1577 }
1578 lwkt_reltoken(&lp->lwp_token);
1579 LWPRELE(lp);
1580 }
1581
1582 if (p->p_nstopped == p->p_nthreads) {
1583 /*
1584 * Token required to interlock kern_wait(). Reparenting can
1585 * also cause a race so we have to hold (q).
1586 */
1587 q = p->p_pptr;
1588 PHOLD(q);
1589 lwkt_gettoken(&q->p_token);
1590 p->p_flags &= ~P_WAITED;
1591 wakeup(q);
1592 if ((q->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1593 ksignal(p->p_pptr, SIGCHLD);
1594 lwkt_reltoken(&q->p_token);
1595 PRELE(q);
1596 }
1597 }
1598
1599 /*
1600 * Caller must hold p_token
1601 */
1602 void
1603 proc_unstop(struct proc *p, int sig)
1604 {
1605 struct lwp *lp;
1606
1607 ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1608
1609 if (p->p_stat != sig)
1610 return;
1611
1612 p->p_stat = SACTIVE;
1613
1614 FOREACH_LWP_IN_PROC(lp, p) {
1615 LWPHOLD(lp);
1616 lwkt_gettoken(&lp->lwp_token);
1617
1618 switch (lp->lwp_stat) {
1619 case LSRUN:
1620 /*
1621 * Uh? Not stopped? Well, I guess that's okay.
1622 */
1623 if (bootverbose)
1624 kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1625 p->p_pid, lp->lwp_tid);
1626 break;
1627
1628 case LSSLEEP:
1629 /*
1630 * Still sleeping. Don't bother waking it up.
1631 * However, if this thread was counted as
1632 * stopped, undo this.
1633 *
1634 * Nevertheless we call setrunnable() so that it
1635 * will wake up in case a signal or timeout arrived
1636 * in the meantime.
1637 *
1638 * LWP_MP_WSTOP is protected by lp->lwp_token.
1639 */
1640 if (lp->lwp_mpflags & LWP_MP_WSTOP) {
1641 atomic_clear_int(&lp->lwp_mpflags,
1642 LWP_MP_WSTOP);
1643 --p->p_nstopped;
1644 } else {
1645 if (bootverbose)
1646 kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1647 p->p_pid, lp->lwp_tid);
1648 }
1649 /* FALLTHROUGH */
1650
1651 case LSSTOP:
1652 /*
1653 * This handles any lwp's waiting in a tsleep with
1654 * SIGCATCH.
1655 */
1656 lwp_signotify(lp);
1657 break;
1658
1659 }
1660 lwkt_reltoken(&lp->lwp_token);
1661 LWPRELE(lp);
1662 }
1663
1664 /*
1665 * This handles any lwp's waiting in tstop(). We have interlocked
1666 * the setting of p_stat by acquiring and releasing each lpw's
1667 * token.
1668 */
1669 wakeup(p);
1670 }
1671
1672 /*
1673 * Wait for all threads except the current thread to stop.
1674 */
1675 static void
1676 proc_stopwait(struct proc *p)
1677 {
1678 while ((p->p_stat == SSTOP || p->p_stat == SCORE) &&
1679 p->p_nstopped < p->p_nthreads - 1) {
1680 tsleep_interlock(&p->p_nstopped, 0);
1681 if (p->p_nstopped < p->p_nthreads - 1) {
1682 tsleep(&p->p_nstopped, PINTERLOCKED, "stopwt", hz);
1683 }
1684 }
1685 }
1686
1687 /*
1688 * No requirements.
1689 */
1690 static int
1691 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1692 {
1693 sigset_t savedmask, set;
1694 struct proc *p = curproc;
1695 struct lwp *lp = curthread->td_lwp;
1696 int error, sig, hz, timevalid = 0;
1697 struct timespec rts, ets, ts;
1698 struct timeval tv;
1699
1700 error = 0;
1701 sig = 0;
1702 ets.tv_sec = 0; /* silence compiler warning */
1703 ets.tv_nsec = 0; /* silence compiler warning */
1704 SIG_CANTMASK(waitset);
1705 savedmask = lp->lwp_sigmask;
1706
1707 if (timeout) {
1708 if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1709 timeout->tv_nsec < 1000000000) {
1710 timevalid = 1;
1711 getnanouptime(&rts);
1712 ets = rts;
1713 timespecadd(&ets, timeout);
1714 }
1715 }
1716
1717 for (;;) {
1718 set = lwp_sigpend(lp);
1719 SIGSETAND(set, waitset);
1720 if ((sig = sig_ffs(&set)) != 0) {
1721 SIGFILLSET(lp->lwp_sigmask);
1722 SIGDELSET(lp->lwp_sigmask, sig);
1723 SIG_CANTMASK(lp->lwp_sigmask);
1724 sig = issignal(lp, 1, 0);
1725 /*
1726 * It may be a STOP signal, in the case, issignal
1727 * returns 0, because we may stop there, and new
1728 * signal can come in, we should restart if we got
1729 * nothing.
1730 */
1731 if (sig == 0)
1732 continue;
1733 else
1734 break;
1735 }
1736
1737 /*
1738 * Previous checking got nothing, and we retried but still
1739 * got nothing, we should return the error status.
1740 */
1741 if (error)
1742 break;
1743
1744 /*
1745 * POSIX says this must be checked after looking for pending
1746 * signals.
1747 */
1748 if (timeout) {
1749 if (timevalid == 0) {
1750 error = EINVAL;
1751 break;
1752 }
1753 getnanouptime(&rts);
1754 if (timespeccmp(&rts, &ets, >=)) {
1755 error = EAGAIN;
1756 break;
1757 }
1758 ts = ets;
1759 timespecsub(&ts, &rts);
1760 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1761 hz = tvtohz_high(&tv);
1762 } else {
1763 hz = 0;
1764 }
1765
1766 lp->lwp_sigmask = savedmask;
1767 SIGSETNAND(lp->lwp_sigmask, waitset);
1768 /*
1769 * We won't ever be woken up. Instead, our sleep will
1770 * be broken in lwpsignal().
1771 */
1772 error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1773 if (timeout) {
1774 if (error == ERESTART) {
1775 /* can not restart a timeout wait. */
1776 error = EINTR;
1777 } else if (error == EAGAIN) {
1778 /* will calculate timeout by ourself. */
1779 error = 0;
1780 }
1781 }
1782 /* Retry ... */
1783 }
1784
1785 lp->lwp_sigmask = savedmask;
1786 if (sig) {
1787 error = 0;
1788 bzero(info, sizeof(*info));
1789 info->si_signo = sig;
1790 spin_lock(&lp->lwp_spin);
1791 lwp_delsig(lp, sig, 1); /* take the signal! */
1792 spin_unlock(&lp->lwp_spin);
1793
1794 if (sig == SIGKILL) {
1795 sigexit(lp, sig);
1796 /* NOT REACHED */
1797 }
1798 }
1799
1800 return (error);
1801 }
1802
1803 /*
1804 * MPALMOSTSAFE
1805 */
1806 int
1807 sys_sigtimedwait(struct sigtimedwait_args *uap)
1808 {
1809 struct timespec ts;
1810 struct timespec *timeout;
1811 sigset_t set;
1812 siginfo_t info;
1813 int error;
1814
1815 if (uap->timeout) {
1816 error = copyin(uap->timeout, &ts, sizeof(ts));
1817 if (error)
1818 return (error);
1819 timeout = &ts;
1820 } else {
1821 timeout = NULL;
1822 }
1823 error = copyin(uap->set, &set, sizeof(set));
1824 if (error)
1825 return (error);
1826 error = kern_sigtimedwait(set, &info, timeout);
1827 if (error)
1828 return (error);
1829 if (uap->info)
1830 error = copyout(&info, uap->info, sizeof(info));
1831 /* Repost if we got an error. */
1832 /*
1833 * XXX lwp
1834 *
1835 * This could transform a thread-specific signal to another
1836 * thread / process pending signal.
1837 */
1838 if (error) {
1839 ksignal(curproc, info.si_signo);
1840 } else {
1841 uap->sysmsg_result = info.si_signo;
1842 }
1843 return (error);
1844 }
1845
1846 /*
1847 * MPALMOSTSAFE
1848 */
1849 int
1850 sys_sigwaitinfo(struct sigwaitinfo_args *uap)
1851 {
1852 siginfo_t info;
1853 sigset_t set;
1854 int error;
1855
1856 error = copyin(uap->set, &set, sizeof(set));
1857 if (error)
1858 return (error);
1859 error = kern_sigtimedwait(set, &info, NULL);
1860 if (error)
1861 return (error);
1862 if (uap->info)
1863 error = copyout(&info, uap->info, sizeof(info));
1864 /* Repost if we got an error. */
1865 /*
1866 * XXX lwp
1867 *
1868 * This could transform a thread-specific signal to another
1869 * thread / process pending signal.
1870 */
1871 if (error) {
1872 ksignal(curproc, info.si_signo);
1873 } else {
1874 uap->sysmsg_result = info.si_signo;
1875 }
1876 return (error);
1877 }
1878
1879 /*
1880 * If the current process has received a signal that would interrupt a
1881 * system call, return EINTR or ERESTART as appropriate.
1882 */
1883 int
1884 iscaught(struct lwp *lp)
1885 {
1886 struct proc *p = lp->lwp_proc;
1887 int sig;
1888
1889 if (p) {
1890 if ((sig = CURSIG(lp)) != 0) {
1891 if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1892 return (EINTR);
1893 return (ERESTART);
1894 }
1895 }
1896 return(EWOULDBLOCK);
1897 }
1898
1899 /*
1900 * If the current lwp/proc has received a signal (should be caught or cause
1901 * termination, should interrupt current syscall), return the signal number.
1902 * Stop signals with default action are processed immediately, then cleared;
1903 * they aren't returned. This is checked after each entry to the system for
1904 * a syscall or trap (though this can usually be done without calling issignal
1905 * by checking the pending signal masks in the CURSIG macro).
1906 *
1907 * This routine is called via CURSIG/__cursig. We will acquire and release
1908 * p->p_token but if the caller needs to interlock the test the caller must
1909 * also hold p->p_token.
1910 *
1911 * while (sig = CURSIG(curproc))
1912 * postsig(sig);
1913 */
1914 int
1915 issignal(struct lwp *lp, int maytrace, int *ptokp)
1916 {
1917 struct proc *p = lp->lwp_proc;
1918 sigset_t mask;
1919 int sig, prop;
1920 int haveptok;
1921
1922 for (;;) {
1923 int traced = (p->p_flags & P_TRACED) || (p->p_stops & S_SIG);
1924
1925 haveptok = 0;
1926
1927 /*
1928 * If this process is supposed to stop, stop this thread.
1929 */
1930 if (STOPLWP(p, lp)) {
1931 lwkt_gettoken(&p->p_token);
1932 tstop();
1933 lwkt_reltoken(&p->p_token);
1934 }
1935
1936 /*
1937 * Quick check without token
1938 */
1939 mask = lwp_sigpend(lp);
1940 SIGSETNAND(mask, lp->lwp_sigmask);
1941 if (p->p_flags & P_PPWAIT)
1942 SIG_STOPSIGMASK(mask);
1943 if (SIGISEMPTY(mask)) /* no signal to send */
1944 return (0);
1945
1946 /*
1947 * If the signal is a member of the process signal set
1948 * we need p_token (even if it is also a member of the
1949 * lwp signal set).
1950 */
1951 sig = sig_ffs(&mask);
1952 if (SIGISMEMBER(p->p_siglist, sig)) {
1953 /*
1954 * Recheck with token
1955 */
1956 haveptok = 1;
1957 lwkt_gettoken(&p->p_token);
1958
1959 mask = lwp_sigpend(lp);
1960 SIGSETNAND(mask, lp->lwp_sigmask);
1961 if (p->p_flags & P_PPWAIT)
1962 SIG_STOPSIGMASK(mask);
1963 if (SIGISEMPTY(mask)) { /* no signal to send */
1964 /* haveptok is TRUE */
1965 lwkt_reltoken(&p->p_token);
1966 return (0);
1967 }
1968 sig = sig_ffs(&mask);
1969 }
1970
1971 STOPEVENT(p, S_SIG, sig);
1972
1973 /*
1974 * We should see pending but ignored signals
1975 * only if P_TRACED was on when they were posted.
1976 */
1977 if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
1978 spin_lock(&lp->lwp_spin);
1979 lwp_delsig(lp, sig, haveptok);
1980 spin_unlock(&lp->lwp_spin);
1981 if (haveptok)
1982 lwkt_reltoken(&p->p_token);
1983 continue;
1984 }
1985 if (maytrace &&
1986 (p->p_flags & P_TRACED) &&
1987 (p->p_flags & P_PPWAIT) == 0) {
1988 /*
1989 * If traced, always stop, and stay stopped until
1990 * released by the parent.
1991 *
1992 * NOTE: SSTOP may get cleared during the loop,
1993 * but we do not re-notify the parent if we have
1994 * to loop several times waiting for the parent
1995 * to let us continue.
1996 *
1997 * XXX not sure if this is still true
1998 */
1999 if (haveptok == 0) {
2000 lwkt_gettoken(&p->p_token);
2001 haveptok = 1;
2002 }
2003 p->p_xstat = sig;
2004 proc_stop(p, SSTOP);
2005 do {
2006 tstop();
2007 } while (!trace_req(p) && (p->p_flags & P_TRACED));
2008
2009 /*
2010 * If parent wants us to take the signal,
2011 * then it will leave it in p->p_xstat;
2012 * otherwise we just look for signals again.
2013 */
2014 spin_lock(&lp->lwp_spin);
2015 lwp_delsig(lp, sig, 1); /* clear old signal */
2016 spin_unlock(&lp->lwp_spin);
2017 sig = p->p_xstat;
2018 if (sig == 0) {
2019 /* haveptok is TRUE */
2020 lwkt_reltoken(&p->p_token);
2021 continue;
2022 }
2023
2024 /*
2025 * Put the new signal into p_siglist. If the
2026 * signal is being masked, look for other signals.
2027 *
2028 * XXX lwp might need a call to ksignal()
2029 */
2030 SIGADDSET_ATOMIC(p->p_siglist, sig);
2031 if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
2032 /* haveptok is TRUE */
2033 lwkt_reltoken(&p->p_token);
2034 continue;
2035 }
2036
2037 /*
2038 * If the traced bit got turned off, go back up
2039 * to the top to rescan signals. This ensures
2040 * that p_sig* and ps_sigact are consistent.
2041 */
2042 if ((p->p_flags & P_TRACED) == 0) {
2043 /* haveptok is TRUE */
2044 lwkt_reltoken(&p->p_token);
2045 continue;
2046 }
2047 }
2048
2049 /*
2050 * p_token may be held here
2051 */
2052 prop = sigprop(sig);
2053
2054 /*
2055 * Decide whether the signal should be returned.
2056 * Return the signal's number, or fall through
2057 * to clear it from the pending mask.
2058 */
2059 switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2060 case (intptr_t)SIG_DFL:
2061 /*
2062 * Don't take default actions on system processes.
2063 */
2064 if (p->p_pid <= 1) {
2065 #ifdef DIAGNOSTIC
2066 /*
2067 * Are you sure you want to ignore SIGSEGV
2068 * in init? XXX
2069 */
2070 kprintf("Process (pid %lu) got signal %d\n",
2071 (u_long)p->p_pid, sig);
2072 #endif
2073 break; /* == ignore */
2074 }
2075
2076 /*
2077 * Handle the in-kernel checkpoint action
2078 */
2079 if (prop & SA_CKPT) {
2080 if (haveptok == 0) {
2081 lwkt_gettoken(&p->p_token);
2082 haveptok = 1;
2083 }
2084 checkpoint_signal_handler(lp);
2085 break;
2086 }
2087
2088 /*
2089 * If there is a pending stop signal to process
2090 * with default action, stop here,
2091 * then clear the signal. However,
2092 * if process is member of an orphaned
2093 * process group, ignore tty stop signals.
2094 */
2095 if (prop & SA_STOP) {
2096 if (haveptok == 0) {
2097 lwkt_gettoken(&p->p_token);
2098 haveptok = 1;
2099 }
2100 if (p->p_flags & P_TRACED ||
2101 (p->p_pgrp->pg_jobc == 0 &&
2102 prop & SA_TTYSTOP))
2103 break; /* == ignore */
2104 if ((p->p_flags & P_WEXIT) == 0) {
2105 p->p_xstat = sig;
2106 proc_stop(p, SSTOP);
2107 tstop();
2108 }
2109 break;
2110 } else if (prop & SA_IGNORE) {
2111 /*
2112 * Except for SIGCONT, shouldn't get here.
2113 * Default action is to ignore; drop it.
2114 */
2115 break; /* == ignore */
2116 } else {
2117 if (ptokp)
2118 *ptokp = haveptok;
2119 else if (haveptok)
2120 lwkt_reltoken(&p->p_token);
2121 return (sig);
2122 }
2123
2124 /*NOTREACHED*/
2125
2126 case (intptr_t)SIG_IGN:
2127 /*
2128 * Masking above should prevent us ever trying
2129 * to take action on an ignored signal other
2130 * than SIGCONT, unless process is traced.
2131 */
2132 if ((prop & SA_CONT) == 0 &&
2133 (p->p_flags & P_TRACED) == 0)
2134 kprintf("issignal\n");
2135 break; /* == ignore */
2136
2137 default:
2138 /*
2139 * This signal has an action, let
2140 * postsig() process it.
2141 */
2142 if (ptokp)
2143 *ptokp = haveptok;
2144 else if (haveptok)
2145 lwkt_reltoken(&p->p_token);
2146 return (sig);
2147 }
2148 spin_lock(&lp->lwp_spin);
2149 lwp_delsig(lp, sig, haveptok); /* take the signal! */
2150 spin_unlock(&lp->lwp_spin);
2151
2152 if (haveptok)
2153 lwkt_reltoken(&p->p_token);
2154 }
2155 /* NOTREACHED */
2156 }
2157
2158 /*
2159 * Take the action for the specified signal
2160 * from the current set of pending signals.
2161 *
2162 * haveptok indicates whether the caller is holding
2163 * p->p_token. If the caller is, we are responsible
2164 * for releasing it.
2165 */
2166 void
2167 postsig(int sig, int haveptok)
2168 {
2169 struct lwp *lp = curthread->td_lwp;
2170 struct proc *p = lp->lwp_proc;
2171 struct sigacts *ps = p->p_sigacts;
2172 sig_t action;
2173 sigset_t returnmask;
2174 int code;
2175
2176 KASSERT(sig != 0, ("postsig"));
2177
2178 KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
2179
2180 /*
2181 * If we are a virtual kernel running an emulated user process
2182 * context, switch back to the virtual kernel context before
2183 * trying to post the signal.
2184 */
2185 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
2186 struct trapframe *tf = lp->lwp_md.md_regs;
2187 tf->tf_trapno = 0;
2188 vkernel_trap(lp, tf);
2189 }
2190
2191 spin_lock(&lp->lwp_spin);
2192 lwp_delsig(lp, sig, haveptok);
2193 spin_unlock(&lp->lwp_spin);
2194 action = ps->ps_sigact[_SIG_IDX(sig)];
2195 #ifdef KTRACE
2196 if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
2197 ktrpsig(lp, sig, action, lp->lwp_flags & LWP_OLDMASK ?
2198 &lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
2199 #endif
2200 /*
2201 * We don't need p_token after this point.
2202 */
2203 if (haveptok)
2204 lwkt_reltoken(&p->p_token);
2205
2206 STOPEVENT(p, S_SIG, sig);
2207
2208 if (action == SIG_DFL) {
2209 /*
2210 * Default action, where the default is to kill
2211 * the process. (Other cases were ignored above.)
2212 */
2213 sigexit(lp, sig);
2214 /* NOTREACHED */
2215 } else {
2216 /*
2217 * If we get here, the signal must be caught.
2218 */
2219 KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
2220 ("postsig action"));
2221
2222 /*
2223 * Reset the signal handler if asked to
2224 */
2225 if (SIGISMEMBER(ps->ps_sigreset, sig)) {
2226 /*
2227 * See kern_sigaction() for origin of this code.
2228 */
2229 SIGDELSET(p->p_sigcatch, sig);
2230 if (sig != SIGCONT &&
2231 sigprop(sig) & SA_IGNORE)
2232 SIGADDSET(p->p_sigignore, sig);
2233 ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2234 }
2235
2236 /*
2237 * Set the signal mask and calculate the mask to restore
2238 * when the signal function returns.
2239 *
2240 * Special case: user has done a sigsuspend. Here the
2241 * current mask is not of interest, but rather the
2242 * mask from before the sigsuspend is what we want
2243 * restored after the signal processing is completed.
2244 */
2245 if (lp->lwp_flags & LWP_OLDMASK) {
2246 returnmask = lp->lwp_oldsigmask;
2247 lp->lwp_flags &= ~LWP_OLDMASK;
2248 } else {
2249 returnmask = lp->lwp_sigmask;
2250 }
2251
2252 SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
2253 if (!SIGISMEMBER(ps->ps_signodefer, sig))
2254 SIGADDSET(lp->lwp_sigmask, sig);
2255
2256 lp->lwp_ru.ru_nsignals++;
2257 if (lp->lwp_sig != sig) {
2258 code = 0;
2259 } else {
2260 code = lp->lwp_code;
2261 lp->lwp_code = 0;
2262 lp->lwp_sig = 0;
2263 }
2264 (*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
2265 }
2266 }
2267
2268 /*
2269 * Kill the current process for stated reason.
2270 */
2271 void
2272 killproc(struct proc *p, char *why)
2273 {
2274 log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
2275 p->p_pid, p->p_comm,
2276 p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2277 ksignal(p, SIGKILL);
2278 }
2279
2280 /*
2281 * Force the current process to exit with the specified signal, dumping core
2282 * if appropriate. We bypass the normal tests for masked and caught signals,
2283 * allowing unrecoverable failures to terminate the process without changing
2284 * signal state. Mark the accounting record with the signal termination.
2285 * If dumping core, save the signal number for the debugger. Calls exit and
2286 * does not return.
2287 *
2288 * This routine does not return.
2289 */
2290 void
2291 sigexit(struct lwp *lp, int sig)
2292 {
2293 struct proc *p = lp->lwp_proc;
2294
2295 lwkt_gettoken(&p->p_token);
2296 p->p_acflag |= AXSIG;
2297 if (sigprop(sig) & SA_CORE) {
2298 lp->lwp_sig = sig;
2299
2300 /*
2301 * All threads must be stopped before we can safely coredump.
2302 * Stop threads using SCORE, which cannot be overridden.
2303 */
2304 if (p->p_stat != SCORE) {
2305 proc_stop(p, SCORE);
2306 proc_stopwait(p);
2307
2308 if (coredump(lp, sig) == 0)
2309 sig |= WCOREFLAG;
2310 p->p_stat = SSTOP;
2311 }
2312
2313 /*
2314 * Log signals which would cause core dumps
2315 * (Log as LOG_INFO to appease those who don't want
2316 * these messages.)
2317 * XXX : Todo, as well as euid, write out ruid too
2318 */
2319 if (kern_logsigexit)
2320 log(LOG_INFO,
2321 "pid %d (%s), uid %d: exited on signal %d%s\n",
2322 p->p_pid, p->p_comm,
2323 p->p_ucred ? p->p_ucred->cr_uid : -1,
2324 sig &~ WCOREFLAG,
2325 sig & WCOREFLAG ? " (core dumped)" : "");
2326 }
2327 lwkt_reltoken(&p->p_token);
2328 exit1(W_EXITCODE(0, sig));
2329 /* NOTREACHED */
2330 }
2331
2332 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
2333 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2334 sizeof(corefilename), "process corefile name format string");
2335
2336 /*
2337 * expand_name(name, uid, pid)
2338 * Expand the name described in corefilename, using name, uid, and pid.
2339 * corefilename is a kprintf-like string, with three format specifiers:
2340 * %N name of process ("name")
2341 * %P process id (pid)
2342 * %U user id (uid)
2343 * For example, "%N.core" is the default; they can be disabled completely
2344 * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2345 * This is controlled by the sysctl variable kern.corefile (see above).
2346 */
2347
2348 static char *
2349 expand_name(const char *name, uid_t uid, pid_t pid)
2350 {
2351 char *temp;
2352 char buf[11]; /* Buffer for pid/uid -- max 4B */
2353 int i, n;
2354 char *format = corefilename;
2355 size_t namelen;
2356
2357 temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
2358 if (temp == NULL)
2359 return NULL;
2360 namelen = strlen(name);
2361 for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2362 int l;
2363 switch (format[i]) {
2364 case '%': /* Format character */
2365 i++;
2366 switch (format[i]) {
2367 case '%':
2368 temp[n++] = '%';
2369 break;
2370 case 'N': /* process name */
2371 if ((n + namelen) > MAXPATHLEN) {
2372 log(LOG_ERR, "pid %d (%s), uid (%u): Path `%s%s' is too long\n",
2373 pid, name, uid, temp, name);
2374 kfree(temp, M_TEMP);
2375 return NULL;
2376 }
2377 memcpy(temp+n, name, namelen);
2378 n += namelen;
2379 break;
2380 case 'P': /* process id */
2381 l = ksprintf(buf, "%u", pid);
2382 if ((n + l) > MAXPATHLEN) {
2383 log(LOG_ERR, "pid %d (%s), uid (%u): Path `%s%s' is too long\n",
2384 pid, name, uid, temp, name);
2385 kfree(temp, M_TEMP);
2386 return NULL;
2387 }
2388 memcpy(temp+n, buf, l);
2389 n += l;
2390 break;
2391 case 'U': /* user id */
2392 l = ksprintf(buf, "%u", uid);
2393 if ((n + l) > MAXPATHLEN) {
2394 log(LOG_ERR, "pid %d (%s), uid (%u): Path `%s%s' is too long\n",
2395 pid, name, uid, temp, name);
2396 kfree(temp, M_TEMP);
2397 return NULL;
2398 }
2399 memcpy(temp+n, buf, l);
2400 n += l;
2401 break;
2402 default:
2403 log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2404 }
2405 break;
2406 default:
2407 temp[n++] = format[i];
2408 }
2409 }
2410 temp[n] = '\0';
2411 return temp;
2412 }
2413
2414 /*
2415 * Dump a process' core. The main routine does some
2416 * policy checking, and creates the name of the coredump;
2417 * then it passes on a vnode and a size limit to the process-specific
2418 * coredump routine if there is one; if there _is not_ one, it returns
2419 * ENOSYS; otherwise it returns the error from the process-specific routine.
2420 *
2421 * The parameter `lp' is the lwp which triggered the coredump.
2422 */
2423
2424 static int
2425 coredump(struct lwp *lp, int sig)
2426 {
2427 struct proc *p = lp->lwp_proc;
2428 struct vnode *vp;
2429 struct ucred *cred = p->p_ucred;
2430 struct flock lf;
2431 struct nlookupdata nd;
2432 struct vattr vattr;
2433 int error, error1;
2434 char *name; /* name of corefile */
2435 off_t limit;
2436
2437 STOPEVENT(p, S_CORE, 0);
2438
2439 if (((sugid_coredump == 0) && p->p_flags & P_SUGID) || do_coredump == 0)
2440 return (EFAULT);
2441
2442 /*
2443 * Note that the bulk of limit checking is done after
2444 * the corefile is created. The exception is if the limit
2445 * for corefiles is 0, in which case we don't bother
2446 * creating the corefile at all. This layout means that
2447 * a corefile is truncated instead of not being created,
2448 * if it is larger than the limit.
2449 */
2450 limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2451 if (limit == 0)
2452 return EFBIG;
2453
2454 name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2455 if (name == NULL)
2456 return (EINVAL);
2457 error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2458 if (error == 0)
2459 error = vn_open(&nd, NULL,
2460 O_CREAT | FWRITE | O_NOFOLLOW,
2461 S_IRUSR | S_IWUSR);
2462 kfree(name, M_TEMP);
2463 if (error) {
2464 nlookup_done(&nd);
2465 return (error);
2466 }
2467 vp = nd.nl_open_vp;
2468 nd.nl_open_vp = NULL;
2469 nlookup_done(&nd);
2470
2471 vn_unlock(vp);
2472 lf.l_whence = SEEK_SET;
2473 lf.l_start = 0;
2474 lf.l_len = 0;
2475 lf.l_type = F_WRLCK;
2476 error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2477 if (error)
2478 goto out2;
2479
2480 /* Don't dump to non-regular files or files with links. */
2481 if (vp->v_type != VREG ||
2482 VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2483 error = EFAULT;
2484 goto out1;
2485 }
2486
2487 /* Don't dump to files current user does not own */
2488 if (vattr.va_uid != p->p_ucred->cr_uid) {
2489 error = EFAULT;
2490 goto out1;
2491 }
2492
2493 VATTR_NULL(&vattr);
2494 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2495 vattr.va_size = 0;
2496 VOP_SETATTR(vp, &vattr, cred);
2497 p->p_acflag |= ACORE;
2498 vn_unlock(vp);
2499
2500 error = p->p_sysent->sv_coredump ?
2501 p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2502
2503 out1:
2504 lf.l_type = F_UNLCK;
2505 VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2506 out2:
2507 error1 = vn_close(vp, FWRITE, NULL);
2508 if (error == 0)
2509 error = error1;
2510 return (error);
2511 }
2512
2513 /*
2514 * Nonexistent system call-- signal process (may want to handle it).
2515 * Flag error in case process won't see signal immediately (blocked or ignored).
2516 *
2517 * MPALMOSTSAFE
2518 */
2519 /* ARGSUSED */
2520 int
2521 sys_nosys(struct nosys_args *args)
2522 {
2523 lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2524 return (EINVAL);
2525 }
2526
2527 /*
2528 * Send a SIGIO or SIGURG signal to a process or process group using
2529 * stored credentials rather than those of the current process.
2530 */
2531 void
2532 pgsigio(struct sigio *sigio, int sig, int checkctty)
2533 {
2534 if (sigio == NULL)
2535 return;
2536
2537 if (sigio->sio_pgid > 0) {
2538 if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2539 sigio->sio_proc))
2540 ksignal(sigio->sio_proc, sig);
2541 } else if (sigio->sio_pgid < 0) {
2542 struct proc *p;
2543 struct pgrp *pg = sigio->sio_pgrp;
2544
2545 /*
2546 * Must interlock all signals against fork
2547 */
2548 pgref(pg);
2549 lockmgr(&pg->pg_lock, LK_EXCLUSIVE);
2550 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
2551 if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2552 (checkctty == 0 || (p->p_flags & P_CONTROLT)))
2553 ksignal(p, sig);
2554 }
2555 lockmgr(&pg->pg_lock, LK_RELEASE);
2556 pgrel(pg);
2557 }
2558 }
2559
2560 static int
2561 filt_sigattach(struct knote *kn)
2562 {
2563 struct proc *p = curproc;
2564
2565 kn->kn_ptr.p_proc = p;
2566 kn->kn_flags |= EV_CLEAR; /* automatically set */
2567
2568 /* XXX lock the proc here while adding to the list? */
2569 knote_insert(&p->p_klist, kn);
2570
2571 return (0);
2572 }
2573
2574 static void
2575 filt_sigdetach(struct knote *kn)
2576 {
2577 struct proc *p = kn->kn_ptr.p_proc;
2578
2579 knote_remove(&p->p_klist, kn);
2580 }
2581
2582 /*
2583 * signal knotes are shared with proc knotes, so we apply a mask to
2584 * the hint in order to differentiate them from process hints. This
2585 * could be avoided by using a signal-specific knote list, but probably
2586 * isn't worth the trouble.
2587 */
2588 static int
2589 filt_signal(struct knote *kn, long hint)
2590 {
2591 if (hint & NOTE_SIGNAL) {
2592 hint &= ~NOTE_SIGNAL;
2593
2594 if (kn->kn_id == hint)
2595 kn->kn_data++;
2596 }
2597 return (kn->kn_data != 0);
2598 }
DragonFly BSD