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/sysmsg.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 send the signal sig to process q? Only processes within
98 * the current reaper or children of the current reaper can be signaled.
99 * Normally the reaper itself cannot be signalled, unless initok is set.
100 */
101 #define CANSIGNAL(q, sig, initok) \
102 ((!p_trespass(curproc->p_ucred, (q)->p_ucred) && \
103 reaper_sigtest(curproc, p, initok)) || \
104 ((sig) == SIGCONT && (q)->p_session == curproc->p_session))
105
106 /*
107 * Policy -- Can real uid ruid with ucred uc send a signal to process q?
108 */
109 #define CANSIGIO(ruid, uc, q) \
110 ((uc)->cr_uid == 0 || \
111 (ruid) == (q)->p_ucred->cr_ruid || \
112 (uc)->cr_uid == (q)->p_ucred->cr_ruid || \
113 (ruid) == (q)->p_ucred->cr_uid || \
114 (uc)->cr_uid == (q)->p_ucred->cr_uid)
115
116 int sugid_coredump;
117 SYSCTL_INT(_kern, OID_AUTO, sugid_coredump, CTLFLAG_RW,
118 &sugid_coredump, 0, "Enable coredumping set user/group ID processes");
119
120 static int do_coredump = 1;
121 SYSCTL_INT(_kern, OID_AUTO, coredump, CTLFLAG_RW,
122 &do_coredump, 0, "Enable/Disable coredumps");
123
124 /*
125 * Signal properties and actions.
126 * The array below categorizes the signals and their default actions
127 * according to the following properties:
128 */
129 #define SA_KILL 0x01 /* terminates process by default */
130 #define SA_CORE 0x02 /* ditto and coredumps */
131 #define SA_STOP 0x04 /* suspend process */
132 #define SA_TTYSTOP 0x08 /* ditto, from tty */
133 #define SA_IGNORE 0x10 /* ignore by default */
134 #define SA_CONT 0x20 /* continue if suspended */
135 #define SA_CANTMASK 0x40 /* non-maskable, catchable */
136 #define SA_CKPT 0x80 /* checkpoint process */
137
138
139 static int sigproptbl[NSIG] = {
140 SA_KILL, /* SIGHUP */
141 SA_KILL, /* SIGINT */
142 SA_KILL|SA_CORE, /* SIGQUIT */
143 SA_KILL|SA_CORE, /* SIGILL */
144 SA_KILL|SA_CORE, /* SIGTRAP */
145 SA_KILL|SA_CORE, /* SIGABRT */
146 SA_KILL|SA_CORE, /* SIGEMT */
147 SA_KILL|SA_CORE, /* SIGFPE */
148 SA_KILL, /* SIGKILL */
149 SA_KILL|SA_CORE, /* SIGBUS */
150 SA_KILL|SA_CORE, /* SIGSEGV */
151 SA_KILL|SA_CORE, /* SIGSYS */
152 SA_KILL, /* SIGPIPE */
153 SA_KILL, /* SIGALRM */
154 SA_KILL, /* SIGTERM */
155 SA_IGNORE, /* SIGURG */
156 SA_STOP, /* SIGSTOP */
157 SA_STOP|SA_TTYSTOP, /* SIGTSTP */
158 SA_IGNORE|SA_CONT, /* SIGCONT */
159 SA_IGNORE, /* SIGCHLD */
160 SA_STOP|SA_TTYSTOP, /* SIGTTIN */
161 SA_STOP|SA_TTYSTOP, /* SIGTTOU */
162 SA_IGNORE, /* SIGIO */
163 SA_KILL, /* SIGXCPU */
164 SA_KILL, /* SIGXFSZ */
165 SA_KILL, /* SIGVTALRM */
166 SA_KILL, /* SIGPROF */
167 SA_IGNORE, /* SIGWINCH */
168 SA_IGNORE, /* SIGINFO */
169 SA_KILL, /* SIGUSR1 */
170 SA_KILL, /* SIGUSR2 */
171 SA_IGNORE, /* SIGTHR */
172 SA_CKPT, /* SIGCKPT */
173 SA_KILL|SA_CKPT, /* SIGCKPTEXIT */
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 SA_IGNORE,
202 SA_IGNORE,
203 SA_IGNORE,
204 };
205
206 __read_mostly sigset_t sigcantmask_mask;
207
208 static __inline int
sigprop(int sig)209 sigprop(int sig)
210 {
211
212 if (sig > 0 && sig < NSIG)
213 return (sigproptbl[_SIG_IDX(sig)]);
214
215 return (0);
216 }
217
218 static __inline int
sig_ffs(sigset_t * set)219 sig_ffs(sigset_t *set)
220 {
221 int i;
222
223 for (i = 0; i < _SIG_WORDS; i++)
224 if (set->__bits[i])
225 return (ffs(set->__bits[i]) + (i * 32));
226 return (0);
227 }
228
229 /*
230 * Allows us to populate siginfo->si_pid and si_uid in the target process
231 * (p) from the originating thread (td). This function must work properly
232 * even if a kernel thread is sending the signal.
233 *
234 * NOTE: Signals are not queued, so if multiple signals are received the
235 * signal handler will only see the most recent pid and uid for any
236 * given signal number.
237 */
238 static __inline void
sigsetfrompid(thread_t td,struct proc * p,int sig)239 sigsetfrompid(thread_t td, struct proc *p, int sig)
240 {
241 struct sigacts *sap;
242
243 if ((sap = p->p_sigacts) == NULL)
244 return;
245 if (td->td_proc) {
246 sap->ps_frominfo[sig].pid = td->td_proc->p_pid;
247 sap->ps_frominfo[sig].uid = td->td_ucred->cr_uid;
248 } else {
249 sap->ps_frominfo[sig].pid = 0;
250 sap->ps_frominfo[sig].uid = 0;
251 }
252 }
253
254 /*
255 * No requirements.
256 */
257 int
kern_sigaction(int sig,struct sigaction * act,struct sigaction * oact)258 kern_sigaction(int sig, struct sigaction *act, struct sigaction *oact)
259 {
260 struct thread *td = curthread;
261 struct proc *p = td->td_proc;
262 struct lwp *lp;
263 struct sigacts *ps = p->p_sigacts;
264
265 if (sig <= 0 || sig >= _SIG_MAXSIG)
266 return (EINVAL);
267
268 lwkt_gettoken(&p->p_token);
269
270 if (oact) {
271 oact->sa_handler = ps->ps_sigact[_SIG_IDX(sig)];
272 oact->sa_mask = ps->ps_catchmask[_SIG_IDX(sig)];
273 oact->sa_flags = 0;
274 if (SIGISMEMBER(ps->ps_sigonstack, sig))
275 oact->sa_flags |= SA_ONSTACK;
276 if (!SIGISMEMBER(ps->ps_sigintr, sig))
277 oact->sa_flags |= SA_RESTART;
278 if (SIGISMEMBER(ps->ps_sigreset, sig))
279 oact->sa_flags |= SA_RESETHAND;
280 if (SIGISMEMBER(ps->ps_signodefer, sig))
281 oact->sa_flags |= SA_NODEFER;
282 if (SIGISMEMBER(ps->ps_siginfo, sig))
283 oact->sa_flags |= SA_SIGINFO;
284 if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDSTOP)
285 oact->sa_flags |= SA_NOCLDSTOP;
286 if (sig == SIGCHLD && p->p_sigacts->ps_flag & PS_NOCLDWAIT)
287 oact->sa_flags |= SA_NOCLDWAIT;
288 }
289 if (act) {
290 /*
291 * Check for invalid requests. KILL and STOP cannot be
292 * caught.
293 */
294 if (sig == SIGKILL || sig == SIGSTOP) {
295 if (act->sa_handler != SIG_DFL) {
296 lwkt_reltoken(&p->p_token);
297 return (EINVAL);
298 }
299 }
300
301 /*
302 * Change setting atomically.
303 */
304 ps->ps_catchmask[_SIG_IDX(sig)] = act->sa_mask;
305 SIG_CANTMASK(ps->ps_catchmask[_SIG_IDX(sig)]);
306 if (act->sa_flags & SA_SIGINFO) {
307 ps->ps_sigact[_SIG_IDX(sig)] =
308 (__sighandler_t *)act->sa_sigaction;
309 SIGADDSET(ps->ps_siginfo, sig);
310 } else {
311 ps->ps_sigact[_SIG_IDX(sig)] = act->sa_handler;
312 SIGDELSET(ps->ps_siginfo, sig);
313 }
314 if (!(act->sa_flags & SA_RESTART))
315 SIGADDSET(ps->ps_sigintr, sig);
316 else
317 SIGDELSET(ps->ps_sigintr, sig);
318 if (act->sa_flags & SA_ONSTACK)
319 SIGADDSET(ps->ps_sigonstack, sig);
320 else
321 SIGDELSET(ps->ps_sigonstack, sig);
322 if (act->sa_flags & SA_RESETHAND)
323 SIGADDSET(ps->ps_sigreset, sig);
324 else
325 SIGDELSET(ps->ps_sigreset, sig);
326 if (act->sa_flags & SA_NODEFER)
327 SIGADDSET(ps->ps_signodefer, sig);
328 else
329 SIGDELSET(ps->ps_signodefer, sig);
330 if (sig == SIGCHLD) {
331 if (act->sa_flags & SA_NOCLDSTOP)
332 p->p_sigacts->ps_flag |= PS_NOCLDSTOP;
333 else
334 p->p_sigacts->ps_flag &= ~PS_NOCLDSTOP;
335 if (act->sa_flags & SA_NOCLDWAIT) {
336 /*
337 * Paranoia: since SA_NOCLDWAIT is implemented
338 * by reparenting the dying child to PID 1 (and
339 * trust it to reap the zombie), PID 1 itself
340 * is forbidden to set SA_NOCLDWAIT.
341 */
342 if (p->p_pid == 1)
343 p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
344 else
345 p->p_sigacts->ps_flag |= PS_NOCLDWAIT;
346 } else {
347 p->p_sigacts->ps_flag &= ~PS_NOCLDWAIT;
348 }
349 if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
350 ps->ps_flag |= PS_CLDSIGIGN;
351 else
352 ps->ps_flag &= ~PS_CLDSIGIGN;
353 }
354 /*
355 * Set bit in p_sigignore for signals that are set to SIG_IGN,
356 * and for signals set to SIG_DFL where the default is to
357 * ignore. However, don't put SIGCONT in p_sigignore, as we
358 * have to restart the process.
359 *
360 * Also remove the signal from the process and lwp signal
361 * list.
362 */
363 if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
364 (sigprop(sig) & SA_IGNORE &&
365 ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
366 SIGDELSET_ATOMIC(p->p_siglist, sig);
367 FOREACH_LWP_IN_PROC(lp, p) {
368 spin_lock(&lp->lwp_spin);
369 SIGDELSET(lp->lwp_siglist, sig);
370 spin_unlock(&lp->lwp_spin);
371 }
372 if (sig != SIGCONT) {
373 /* easier in ksignal */
374 SIGADDSET(p->p_sigignore, sig);
375 }
376 SIGDELSET(p->p_sigcatch, sig);
377 } else {
378 SIGDELSET(p->p_sigignore, sig);
379 if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)
380 SIGDELSET(p->p_sigcatch, sig);
381 else
382 SIGADDSET(p->p_sigcatch, sig);
383 }
384 }
385 lwkt_reltoken(&p->p_token);
386 return (0);
387 }
388
389 int
sys_sigaction(struct sysmsg * sysmsg,const struct sigaction_args * uap)390 sys_sigaction(struct sysmsg *sysmsg, const struct sigaction_args *uap)
391 {
392 struct sigaction act, oact;
393 struct sigaction *actp, *oactp;
394 int error;
395
396 actp = (uap->act != NULL) ? &act : NULL;
397 oactp = (uap->oact != NULL) ? &oact : NULL;
398 if (actp) {
399 error = copyin(uap->act, actp, sizeof(act));
400 if (error)
401 return (error);
402 }
403 error = kern_sigaction(uap->sig, actp, oactp);
404 if (oactp && !error) {
405 error = copyout(oactp, uap->oact, sizeof(oact));
406 }
407 return (error);
408 }
409
410 /*
411 * Initialize signal state for process 0;
412 * set to ignore signals that are ignored by default.
413 */
414 void
siginit(struct proc * p)415 siginit(struct proc *p)
416 {
417 int i;
418
419 for (i = 1; i <= NSIG; i++) {
420 if (sigprop(i) & SA_IGNORE && i != SIGCONT)
421 SIGADDSET(p->p_sigignore, i);
422 }
423
424 /*
425 * Also initialize signal-related global state.
426 */
427 SIGSETOR_CANTMASK(sigcantmask_mask);
428 }
429
430 /*
431 * Reset signals for an exec of the specified process.
432 */
433 void
execsigs(struct proc * p)434 execsigs(struct proc *p)
435 {
436 struct sigacts *ps = p->p_sigacts;
437 struct lwp *lp;
438 int sig;
439
440 lp = ONLY_LWP_IN_PROC(p);
441
442 /*
443 * Reset caught signals. Held signals remain held
444 * through p_sigmask (unless they were caught,
445 * and are now ignored by default).
446 */
447 while (SIGNOTEMPTY(p->p_sigcatch)) {
448 sig = sig_ffs(&p->p_sigcatch);
449 SIGDELSET(p->p_sigcatch, sig);
450 if (sigprop(sig) & SA_IGNORE) {
451 if (sig != SIGCONT)
452 SIGADDSET(p->p_sigignore, sig);
453 SIGDELSET_ATOMIC(p->p_siglist, sig);
454 /* don't need spinlock */
455 SIGDELSET(lp->lwp_siglist, sig);
456 }
457 ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
458 }
459
460 /*
461 * Reset stack state to the user stack.
462 * Clear set of signals caught on the signal stack.
463 */
464 lp->lwp_sigstk.ss_flags = SS_DISABLE;
465 lp->lwp_sigstk.ss_size = 0;
466 lp->lwp_sigstk.ss_sp = NULL;
467 lp->lwp_flags &= ~LWP_ALTSTACK;
468 /*
469 * Reset no zombies if child dies flag as Solaris does.
470 */
471 p->p_sigacts->ps_flag &= ~(PS_NOCLDWAIT | PS_CLDSIGIGN);
472 if (ps->ps_sigact[_SIG_IDX(SIGCHLD)] == SIG_IGN)
473 ps->ps_sigact[_SIG_IDX(SIGCHLD)] = SIG_DFL;
474 }
475
476 /*
477 * kern_sigprocmask() - MP SAFE ONLY IF p == curproc
478 *
479 * Manipulate signal mask. This routine is MP SAFE *ONLY* if
480 * p == curproc.
481 */
482 int
kern_sigprocmask(int how,sigset_t * set,sigset_t * oset)483 kern_sigprocmask(int how, sigset_t *set, sigset_t *oset)
484 {
485 struct thread *td = curthread;
486 struct lwp *lp = td->td_lwp;
487 struct proc *p = td->td_proc;
488 int error;
489
490 lwkt_gettoken(&p->p_token);
491
492 if (oset != NULL)
493 *oset = lp->lwp_sigmask;
494
495 error = 0;
496 if (set != NULL) {
497 switch (how) {
498 case SIG_BLOCK:
499 SIG_CANTMASK(*set);
500 SIGSETOR(lp->lwp_sigmask, *set);
501 break;
502 case SIG_UNBLOCK:
503 SIGSETNAND(lp->lwp_sigmask, *set);
504 break;
505 case SIG_SETMASK:
506 SIG_CANTMASK(*set);
507 lp->lwp_sigmask = *set;
508 sigirefs_wait(p);
509 break;
510 default:
511 error = EINVAL;
512 break;
513 }
514 }
515
516 lwkt_reltoken(&p->p_token);
517
518 return (error);
519 }
520
521 /*
522 * sigprocmask()
523 *
524 * MPSAFE
525 */
526 int
sys_sigprocmask(struct sysmsg * sysmsg,const struct sigprocmask_args * uap)527 sys_sigprocmask(struct sysmsg *sysmsg, const struct sigprocmask_args *uap)
528 {
529 sigset_t set, oset;
530 sigset_t *setp, *osetp;
531 int error;
532
533 setp = (uap->set != NULL) ? &set : NULL;
534 osetp = (uap->oset != NULL) ? &oset : NULL;
535 if (setp) {
536 error = copyin(uap->set, setp, sizeof(set));
537 if (error)
538 return (error);
539 }
540 error = kern_sigprocmask(uap->how, setp, osetp);
541 if (osetp && !error) {
542 error = copyout(osetp, uap->oset, sizeof(oset));
543 }
544 return (error);
545 }
546
547 /*
548 * MPSAFE
549 */
550 int
kern_sigpending(sigset_t * set)551 kern_sigpending(sigset_t *set)
552 {
553 struct lwp *lp = curthread->td_lwp;
554
555 *set = lwp_sigpend(lp);
556
557 return (0);
558 }
559
560 /*
561 * MPSAFE
562 */
563 int
sys_sigpending(struct sysmsg * sysmsg,const struct sigpending_args * uap)564 sys_sigpending(struct sysmsg *sysmsg, const struct sigpending_args *uap)
565 {
566 sigset_t set;
567 int error;
568
569 error = kern_sigpending(&set);
570
571 if (error == 0)
572 error = copyout(&set, uap->set, sizeof(set));
573 return (error);
574 }
575
576 /*
577 * Suspend process until signal, providing mask to be set
578 * in the meantime.
579 *
580 * MPSAFE
581 */
582 int
kern_sigsuspend(sigset_t * set)583 kern_sigsuspend(sigset_t *set)
584 {
585 struct thread *td = curthread;
586 struct lwp *lp = td->td_lwp;
587 struct proc *p = td->td_proc;
588 struct sigacts *ps = p->p_sigacts;
589
590 /*
591 * When returning from sigsuspend, we want the old mask to be
592 * restored after the signal handler has finished. Thus, we
593 * save it here and mark the sigacts structure to indicate this.
594 *
595 * To interlock signal deliveries which may race this function, we
596 * must hold the LWP token, otherwise the signal may be made pending
597 * to the process rather than the lwp during execution of the tsleep()
598 * (which does not hold the process token to interlock that) and be
599 * missed by the tsleep().
600 */
601 lwkt_gettoken(&lp->lwp_token);
602 lp->lwp_oldsigmask = lp->lwp_sigmask;
603 lp->lwp_flags |= LWP_OLDMASK;
604 SIG_CANTMASK(*set);
605 lp->lwp_sigmask = *set;
606 lwkt_reltoken(&lp->lwp_token);
607 sigirefs_wait(p);
608
609 while (tsleep(ps, PCATCH, "pause", 0) == 0)
610 /* void */;
611 /* always return EINTR rather than ERESTART... */
612 return (EINTR);
613 }
614
615 /*
616 * Note nonstandard calling convention: libc stub passes mask, not
617 * pointer, to save a copyin.
618 *
619 * MPSAFE
620 */
621 int
sys_sigsuspend(struct sysmsg * sysmsg,const struct sigsuspend_args * uap)622 sys_sigsuspend(struct sysmsg *sysmsg, const struct sigsuspend_args *uap)
623 {
624 sigset_t mask;
625 int error;
626
627 error = copyin(uap->sigmask, &mask, sizeof(mask));
628 if (error)
629 return (error);
630
631 error = kern_sigsuspend(&mask);
632
633 return (error);
634 }
635
636 /*
637 * MPSAFE
638 */
639 int
kern_sigaltstack(stack_t * ss,stack_t * oss)640 kern_sigaltstack(stack_t *ss, stack_t *oss)
641 {
642 struct thread *td = curthread;
643 struct lwp *lp = td->td_lwp;
644 struct proc *p = td->td_proc;
645
646 if ((lp->lwp_flags & LWP_ALTSTACK) == 0)
647 lp->lwp_sigstk.ss_flags |= SS_DISABLE;
648
649 if (oss)
650 *oss = lp->lwp_sigstk;
651
652 if (ss) {
653 if (ss->ss_flags & ~SS_DISABLE)
654 return (EINVAL);
655 if (ss->ss_flags & SS_DISABLE) {
656 if (lp->lwp_sigstk.ss_flags & SS_ONSTACK)
657 return (EPERM);
658 lp->lwp_flags &= ~LWP_ALTSTACK;
659 lp->lwp_sigstk.ss_flags = ss->ss_flags;
660 } else {
661 if (ss->ss_size < p->p_sysent->sv_minsigstksz)
662 return (ENOMEM);
663 lp->lwp_flags |= LWP_ALTSTACK;
664 lp->lwp_sigstk = *ss;
665 }
666 }
667
668 return (0);
669 }
670
671 /*
672 * MPSAFE
673 */
674 int
sys_sigaltstack(struct sysmsg * sysmsg,const struct sigaltstack_args * uap)675 sys_sigaltstack(struct sysmsg *sysmsg, const struct sigaltstack_args *uap)
676 {
677 stack_t ss, oss;
678 int error;
679
680 if (uap->ss) {
681 error = copyin(uap->ss, &ss, sizeof(ss));
682 if (error)
683 return (error);
684 }
685
686 error = kern_sigaltstack(uap->ss ? &ss : NULL, uap->oss ? &oss : NULL);
687
688 if (error == 0 && uap->oss)
689 error = copyout(&oss, uap->oss, sizeof(*uap->oss));
690 return (error);
691 }
692
693 /*
694 * Common code for kill process group/broadcast kill.
695 * cp is calling process.
696 */
697 struct killpg_info {
698 int nfound;
699 int sig;
700 };
701
702 static int killpg_all_callback(struct proc *p, void *data);
703
704 static int
dokillpg(int sig,int pgid,int all)705 dokillpg(int sig, int pgid, int all)
706 {
707 struct killpg_info info;
708 struct proc *cp = curproc;
709 struct proc *p;
710 struct pgrp *pgrp;
711
712 info.nfound = 0;
713 info.sig = sig;
714
715 if (all) {
716 /*
717 * broadcast
718 */
719 allproc_scan(killpg_all_callback, &info, 0);
720 } else {
721 if (pgid == 0) {
722 /*
723 * zero pgid means send to my process group.
724 */
725 pgrp = cp->p_pgrp;
726 pgref(pgrp);
727 } else {
728 pgrp = pgfind(pgid);
729 if (pgrp == NULL)
730 return (ESRCH);
731 }
732
733 /*
734 * Must interlock all signals against fork
735 */
736 lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
737 LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
738 if (p->p_pid <= 1 ||
739 p->p_stat == SZOMB ||
740 (p->p_flags & P_SYSTEM) ||
741 !CANSIGNAL(p, sig, 0)) {
742 continue;
743 }
744 ++info.nfound;
745 if (sig)
746 ksignal(p, sig);
747 }
748 lockmgr(&pgrp->pg_lock, LK_RELEASE);
749 pgrel(pgrp);
750 }
751 return (info.nfound ? 0 : ESRCH);
752 }
753
754 static int
killpg_all_callback(struct proc * p,void * data)755 killpg_all_callback(struct proc *p, void *data)
756 {
757 struct killpg_info *info = data;
758
759 if (p->p_pid <= 1 || (p->p_flags & P_SYSTEM) ||
760 p == curproc || !CANSIGNAL(p, info->sig, 0)) {
761 return (0);
762 }
763 ++info->nfound;
764 if (info->sig)
765 ksignal(p, info->sig);
766 return(0);
767 }
768
769 /*
770 * Send a general signal to a process or LWPs within that process.
771 *
772 * Note that new signals cannot be sent if a process is exiting or already
773 * a zombie, but we return success anyway as userland is likely to not handle
774 * the race properly.
775 *
776 * No requirements.
777 */
778 int
kern_kill(int sig,pid_t pid,lwpid_t tid)779 kern_kill(int sig, pid_t pid, lwpid_t tid)
780 {
781 int t;
782
783 if ((u_int)sig >= _SIG_MAXSIG)
784 return (EINVAL);
785
786 if (pid > 0) {
787 struct proc *p;
788 struct lwp *lp = NULL;
789
790 /*
791 * Sending a signal to pid 1 as root requires that we
792 * are not reboot-restricted.
793 */
794 if (pid == 1 && caps_priv_check_self(SYSCAP_NOREBOOT))
795 return EPERM;
796
797 /*
798 * Send a signal to a single process. If the kill() is
799 * racing an exiting process which has not yet been reaped
800 * act as though the signal was delivered successfully but
801 * don't actually try to deliver the signal.
802 */
803 if ((p = pfind(pid)) == NULL) {
804 if ((p = zpfind(pid)) == NULL)
805 return (ESRCH);
806 PRELE(p);
807 return (0);
808 }
809 if (p != curproc) {
810 lwkt_gettoken_shared(&p->p_token);
811 if (!CANSIGNAL(p, sig, 1)) {
812 lwkt_reltoken(&p->p_token);
813 PRELE(p);
814 return (EPERM);
815 }
816 lwkt_reltoken(&p->p_token);
817 }
818
819 /*
820 * NOP if the process is exiting. Note that lwpsignal() is
821 * called directly with P_WEXIT set to kill individual LWPs
822 * during exit, which is allowed.
823 */
824 if (p->p_flags & P_WEXIT) {
825 PRELE(p);
826 return (0);
827 }
828 if (tid != -1) {
829 lwkt_gettoken_shared(&p->p_token);
830 lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
831 if (lp == NULL) {
832 lwkt_reltoken(&p->p_token);
833 PRELE(p);
834 return (ESRCH);
835 }
836 LWPHOLD(lp);
837 lwkt_reltoken(&p->p_token);
838 }
839 if (sig)
840 lwpsignal(p, lp, sig);
841 if (lp)
842 LWPRELE(lp);
843 PRELE(p);
844
845 return (0);
846 }
847
848 /*
849 * If we come here, pid is a special broadcast pid.
850 * This doesn't mix with a tid.
851 */
852 if (tid != -1)
853 return (EINVAL);
854
855 switch (pid) {
856 case -1: /* broadcast signal */
857 t = (dokillpg(sig, 0, 1));
858 break;
859 case 0: /* signal own process group */
860 t = (dokillpg(sig, 0, 0));
861 break;
862 default: /* negative explicit process group */
863 t = (dokillpg(sig, -pid, 0));
864 break;
865 }
866 return t;
867 }
868
869 int
sys_kill(struct sysmsg * sysmsg,const struct kill_args * uap)870 sys_kill(struct sysmsg *sysmsg, const struct kill_args *uap)
871 {
872 int error;
873
874 error = kern_kill(uap->signum, uap->pid, -1);
875 return (error);
876 }
877
878 int
sys_lwp_kill(struct sysmsg * sysmsg,const struct lwp_kill_args * uap)879 sys_lwp_kill(struct sysmsg *sysmsg, const struct lwp_kill_args *uap)
880 {
881 int error;
882 pid_t pid = uap->pid;
883
884 /*
885 * A tid is mandatory for lwp_kill(), otherwise
886 * you could simply use kill().
887 */
888 if (uap->tid == -1)
889 return (EINVAL);
890
891 /*
892 * To save on a getpid() function call for intra-process
893 * signals, pid == -1 means current process.
894 */
895 if (pid == -1)
896 pid = curproc->p_pid;
897
898 error = kern_kill(uap->signum, pid, uap->tid);
899 return (error);
900 }
901
902 /*
903 * Send a signal to a process group.
904 */
905 void
gsignal(int pgid,int sig)906 gsignal(int pgid, int sig)
907 {
908 struct pgrp *pgrp;
909
910 if (pgid && (pgrp = pgfind(pgid)))
911 pgsignal(pgrp, sig, 0);
912 }
913
914 /*
915 * Send a signal to a process group. If checktty is 1,
916 * limit to members which have a controlling terminal.
917 *
918 * pg_lock interlocks against a fork that might be in progress, to
919 * ensure that the new child process picks up the signal.
920 */
921 void
pgsignal(struct pgrp * pgrp,int sig,int checkctty)922 pgsignal(struct pgrp *pgrp, int sig, int checkctty)
923 {
924 struct proc *p;
925
926 /*
927 * Must interlock all signals against fork
928 */
929 if (pgrp) {
930 pgref(pgrp);
931 lockmgr(&pgrp->pg_lock, LK_EXCLUSIVE);
932 LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
933 if (checkctty == 0 || p->p_flags & P_CONTROLT)
934 ksignal(p, sig);
935 }
936 lockmgr(&pgrp->pg_lock, LK_RELEASE);
937 pgrel(pgrp);
938 }
939 }
940
941 /*
942 * Send a signal caused by a trap to the current lwp. If it will be caught
943 * immediately, deliver it with correct code. Otherwise, post it normally.
944 *
945 * These signals may ONLY be delivered to the specified lwp and may never
946 * be delivered to the process generically.
947 *
948 * lpmap->blockallsigs is ignored.
949 */
950 void
trapsignal(struct lwp * lp,int sig,u_long code)951 trapsignal(struct lwp *lp, int sig, u_long code)
952 {
953 struct proc *p = lp->lwp_proc;
954 struct sigacts *ps = p->p_sigacts;
955
956 /*
957 * If we are a virtual kernel running an emulated user process
958 * context, switch back to the virtual kernel context before
959 * trying to post the signal.
960 */
961 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
962 struct trapframe *tf = lp->lwp_md.md_regs;
963 tf->tf_trapno = 0;
964 vkernel_trap(lp, tf);
965 }
966
967 if ((p->p_flags & P_TRACED) == 0 && SIGISMEMBER(p->p_sigcatch, sig) &&
968 !SIGISMEMBER(lp->lwp_sigmask, sig)) {
969 lp->lwp_ru.ru_nsignals++;
970 #ifdef KTRACE
971 if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
972 ktrpsig(lp, sig, ps->ps_sigact[_SIG_IDX(sig)],
973 &lp->lwp_sigmask, code);
974 #endif
975 (*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)], sig,
976 &lp->lwp_sigmask, code);
977 SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
978 if (!SIGISMEMBER(ps->ps_signodefer, sig))
979 SIGADDSET(lp->lwp_sigmask, sig);
980 if (SIGISMEMBER(ps->ps_sigreset, sig)) {
981 /*
982 * See kern_sigaction() for origin of this code.
983 */
984 SIGDELSET(p->p_sigcatch, sig);
985 if (sig != SIGCONT &&
986 sigprop(sig) & SA_IGNORE)
987 SIGADDSET(p->p_sigignore, sig);
988 ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
989 }
990 } else {
991 lp->lwp_code = code; /* XXX for core dump/debugger */
992 lp->lwp_sig = sig; /* XXX to verify code */
993 lwpsignal(p, lp, sig);
994 }
995 }
996
997 /*
998 * Find a suitable lwp to deliver the signal to. Returns NULL if all
999 * lwps hold the signal blocked.
1000 *
1001 * Caller must hold p->p_token.
1002 *
1003 * Returns a lp or NULL. If non-NULL the lp is held and its token is
1004 * acquired.
1005 */
1006 static struct lwp *
find_lwp_for_signal(struct proc * p,int sig)1007 find_lwp_for_signal(struct proc *p, int sig)
1008 {
1009 struct lwp *lp;
1010 struct lwp *run, *sleep, *stop;
1011
1012 /*
1013 * If the running/preempted thread belongs to the proc to which
1014 * the signal is being delivered and this thread does not block
1015 * the signal, then we can avoid a context switch by delivering
1016 * the signal to this thread, because it will return to userland
1017 * soon anyways.
1018 */
1019 lp = lwkt_preempted_proc();
1020 if (lp != NULL && lp->lwp_proc == p) {
1021 LWPHOLD(lp);
1022 lwkt_gettoken(&lp->lwp_token);
1023 if (!SIGISMEMBER(lp->lwp_sigmask, sig)) {
1024 /* return w/ token held */
1025 return (lp);
1026 }
1027 lwkt_reltoken(&lp->lwp_token);
1028 LWPRELE(lp);
1029 }
1030
1031 run = sleep = stop = NULL;
1032 FOREACH_LWP_IN_PROC(lp, p) {
1033 /*
1034 * If the signal is being blocked by the lwp, then this
1035 * lwp is not eligible for receiving the signal.
1036 */
1037 LWPHOLD(lp);
1038 lwkt_gettoken(&lp->lwp_token);
1039
1040 if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
1041 lwkt_reltoken(&lp->lwp_token);
1042 LWPRELE(lp);
1043 continue;
1044 }
1045
1046 switch (lp->lwp_stat) {
1047 case LSRUN:
1048 if (sleep) {
1049 lwkt_token_swap();
1050 lwkt_reltoken(&sleep->lwp_token);
1051 LWPRELE(sleep);
1052 sleep = NULL;
1053 run = lp;
1054 } else if (stop) {
1055 lwkt_token_swap();
1056 lwkt_reltoken(&stop->lwp_token);
1057 LWPRELE(stop);
1058 stop = NULL;
1059 run = lp;
1060 } else {
1061 run = lp;
1062 }
1063 break;
1064 case LSSLEEP:
1065 if (lp->lwp_flags & LWP_SINTR) {
1066 if (sleep) {
1067 lwkt_reltoken(&lp->lwp_token);
1068 LWPRELE(lp);
1069 } else if (stop) {
1070 lwkt_token_swap();
1071 lwkt_reltoken(&stop->lwp_token);
1072 LWPRELE(stop);
1073 stop = NULL;
1074 sleep = lp;
1075 } else {
1076 sleep = lp;
1077 }
1078 } else {
1079 lwkt_reltoken(&lp->lwp_token);
1080 LWPRELE(lp);
1081 }
1082 break;
1083 case LSSTOP:
1084 if (sleep) {
1085 lwkt_reltoken(&lp->lwp_token);
1086 LWPRELE(lp);
1087 } else if (stop) {
1088 lwkt_reltoken(&lp->lwp_token);
1089 LWPRELE(lp);
1090 } else {
1091 stop = lp;
1092 }
1093 break;
1094 }
1095 if (run)
1096 break;
1097 }
1098
1099 if (run != NULL)
1100 return (run);
1101 else if (sleep != NULL)
1102 return (sleep);
1103 else
1104 return (stop);
1105 }
1106
1107 /*
1108 * Send the signal to the process. If the signal has an action, the action
1109 * is usually performed by the target process rather than the caller; we add
1110 * the signal to the set of pending signals for the process.
1111 *
1112 * Exceptions:
1113 * o When a stop signal is sent to a sleeping process that takes the
1114 * default action, the process is stopped without awakening it.
1115 * o SIGCONT restarts stopped processes (or puts them back to sleep)
1116 * regardless of the signal action (eg, blocked or ignored).
1117 *
1118 * Other ignored signals are discarded immediately.
1119 *
1120 * If the caller wishes to call this function from a hard code section the
1121 * caller must already hold p->p_token (see kern_clock.c).
1122 *
1123 * No requirements.
1124 */
1125 void
ksignal(struct proc * p,int sig)1126 ksignal(struct proc *p, int sig)
1127 {
1128 lwpsignal(p, NULL, sig);
1129 }
1130
1131 /*
1132 * The core for ksignal. lp may be NULL, then a suitable thread
1133 * will be chosen. If not, lp MUST be a member of p.
1134 *
1135 * If the caller wishes to call this function from a hard code section the
1136 * caller must already hold p->p_token.
1137 *
1138 * No requirements.
1139 */
1140 void
lwpsignal(struct proc * p,struct lwp * lp,int sig)1141 lwpsignal(struct proc *p, struct lwp *lp, int sig)
1142 {
1143 struct proc *q;
1144 sig_t action;
1145 int prop;
1146
1147 if (sig >= _SIG_MAXSIG || sig <= 0) {
1148 kprintf("lwpsignal: signal %d\n", sig);
1149 panic("lwpsignal signal number");
1150 }
1151
1152 KKASSERT(lp == NULL || lp->lwp_proc == p);
1153
1154 /*
1155 * We don't want to race... well, all sorts of things. Get appropriate
1156 * tokens.
1157 *
1158 * Don't try to deliver a generic signal to an exiting process,
1159 * the signal structures could be in flux. We check the LWP later
1160 * on.
1161 */
1162 PHOLD(p);
1163 if (lp) {
1164 LWPHOLD(lp);
1165 lwkt_gettoken(&lp->lwp_token);
1166 } else {
1167 lwkt_gettoken(&p->p_token);
1168 if (p->p_flags & P_WEXIT)
1169 goto out;
1170 }
1171
1172 prop = sigprop(sig);
1173
1174 /*
1175 * If proc is traced, always give parent a chance;
1176 * if signal event is tracked by procfs, give *that*
1177 * a chance, as well.
1178 */
1179 if ((p->p_flags & P_TRACED) || (p->p_stops & S_SIG)) {
1180 action = SIG_DFL;
1181 } else {
1182 /*
1183 * Do not try to deliver signals to an exiting lwp other
1184 * than SIGKILL. Note that we must still deliver the signal
1185 * if P_WEXIT is set in the process flags.
1186 */
1187 if (lp && (lp->lwp_mpflags & LWP_MP_WEXIT) && sig != SIGKILL) {
1188 lwkt_reltoken(&lp->lwp_token);
1189 LWPRELE(lp);
1190 PRELE(p);
1191 return;
1192 }
1193
1194 /*
1195 * If the signal is being ignored, then we forget about
1196 * it immediately. NOTE: We don't set SIGCONT in p_sigignore,
1197 * and if it is set to SIG_IGN, action will be SIG_DFL here.
1198 */
1199 if (SIGISMEMBER(p->p_sigignore, sig)) {
1200 /*
1201 * Even if a signal is set SIG_IGN, it may still be
1202 * lurking in a kqueue.
1203 */
1204 KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1205 if (lp) {
1206 lwkt_reltoken(&lp->lwp_token);
1207 LWPRELE(lp);
1208 } else {
1209 lwkt_reltoken(&p->p_token);
1210 }
1211 PRELE(p);
1212 return;
1213 }
1214 if (SIGISMEMBER(p->p_sigcatch, sig))
1215 action = SIG_CATCH;
1216 else
1217 action = SIG_DFL;
1218 }
1219
1220 /*
1221 * If continuing, clear any pending STOP signals for the whole
1222 * process.
1223 */
1224 if (prop & SA_CONT) {
1225 lwkt_gettoken(&p->p_token);
1226 SIG_STOPSIGMASK_ATOMIC(p->p_siglist);
1227 lwkt_reltoken(&p->p_token);
1228 }
1229
1230 if (prop & SA_STOP) {
1231 /*
1232 * If sending a tty stop signal to a member of an orphaned
1233 * process group, discard the signal here if the action
1234 * is default; don't stop the process below if sleeping,
1235 * and don't clear any pending SIGCONT.
1236 */
1237 if ((prop & SA_TTYSTOP) && p->p_pgrp->pg_jobc == 0 &&
1238 action == SIG_DFL) {
1239 if (lp) {
1240 lwkt_reltoken(&lp->lwp_token);
1241 LWPRELE(lp);
1242 } else {
1243 lwkt_reltoken(&p->p_token);
1244 }
1245 PRELE(p);
1246 return;
1247 }
1248 lwkt_gettoken(&p->p_token);
1249 SIG_CONTSIGMASK_ATOMIC(p->p_siglist);
1250 p->p_flags &= ~P_CONTINUED;
1251 lwkt_reltoken(&p->p_token);
1252 }
1253
1254 if (p->p_stat == SSTOP) {
1255 /*
1256 * Nobody can handle this signal, add it to the lwp or
1257 * process pending list
1258 */
1259 lwkt_gettoken(&p->p_token);
1260 if (p->p_stat != SSTOP) {
1261 lwkt_reltoken(&p->p_token);
1262 goto not_stopped;
1263 }
1264 sigsetfrompid(curthread, p, sig);
1265 if (lp) {
1266 spin_lock(&lp->lwp_spin);
1267 SIGADDSET(lp->lwp_siglist, sig);
1268 spin_unlock(&lp->lwp_spin);
1269 } else {
1270 SIGADDSET_ATOMIC(p->p_siglist, sig);
1271 }
1272
1273 /*
1274 * If the process is stopped and is being traced, then no
1275 * further action is necessary.
1276 */
1277 if (p->p_flags & P_TRACED) {
1278 lwkt_reltoken(&p->p_token);
1279 goto out;
1280 }
1281
1282 /*
1283 * If the process is stopped and receives a KILL signal,
1284 * make the process runnable.
1285 */
1286 if (sig == SIGKILL) {
1287 proc_unstop(p, SSTOP);
1288 lwkt_reltoken(&p->p_token);
1289 goto active_process;
1290 }
1291
1292 /*
1293 * If the process is stopped and receives a CONT signal,
1294 * then try to make the process runnable again.
1295 */
1296 if (prop & SA_CONT) {
1297 /*
1298 * If SIGCONT is default (or ignored), we continue the
1299 * process but don't leave the signal in p_siglist, as
1300 * it has no further action. If SIGCONT is held, we
1301 * continue the process and leave the signal in
1302 * p_siglist. If the process catches SIGCONT, let it
1303 * handle the signal itself.
1304 *
1305 * XXX what if the signal is being held blocked?
1306 *
1307 * Token required to interlock kern_wait().
1308 * Reparenting can also cause a race so we have to
1309 * hold (q).
1310 */
1311 q = p->p_pptr;
1312 PHOLD(q);
1313 lwkt_gettoken(&q->p_token);
1314 p->p_flags |= P_CONTINUED;
1315 wakeup(q);
1316 if (action == SIG_DFL)
1317 SIGDELSET_ATOMIC(p->p_siglist, sig);
1318 proc_unstop(p, SSTOP);
1319 lwkt_reltoken(&q->p_token);
1320 PRELE(q);
1321 lwkt_reltoken(&p->p_token);
1322 if (action == SIG_CATCH)
1323 goto active_process;
1324 goto out;
1325 }
1326
1327 /*
1328 * If the process is stopped and receives another STOP
1329 * signal, we do not need to stop it again. If we did
1330 * the shell could get confused.
1331 *
1332 * However, if the current/preempted lwp is part of the
1333 * process receiving the signal, we need to keep it,
1334 * so that this lwp can stop in issignal() later, as
1335 * we don't want to wait until it reaches userret!
1336 */
1337 if (prop & SA_STOP) {
1338 if (lwkt_preempted_proc() == NULL ||
1339 lwkt_preempted_proc()->lwp_proc != p) {
1340 SIGDELSET_ATOMIC(p->p_siglist, sig);
1341 }
1342 }
1343
1344 /*
1345 * Otherwise the process is stopped and it received some
1346 * signal, which does not change its stopped state. When
1347 * the process is continued a wakeup(p) will be issued which
1348 * will wakeup any threads sleeping in tstop().
1349 */
1350 lwkt_reltoken(&p->p_token);
1351 goto out;
1352 /* NOTREACHED */
1353 }
1354 not_stopped:
1355 ;
1356 /* else not stopped */
1357 active_process:
1358
1359 /*
1360 * Never deliver a lwp-specific signal to a random lwp.
1361 *
1362 * When delivering an untargetted signal, use p_sigirefs to
1363 * inform lwps of potential collisions.
1364 */
1365 if (lp == NULL) {
1366 /* NOTE: returns lp w/ token held */
1367 sigirefs_hold(p);
1368 lp = find_lwp_for_signal(p, sig);
1369 if (lp) {
1370 if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
1371 lwkt_reltoken(&lp->lwp_token);
1372 LWPRELE(lp);
1373 lp = NULL;
1374 /* maintain proc token */
1375 /* maintain sigirefs */
1376 } else {
1377 lwkt_token_swap();
1378 lwkt_reltoken(&p->p_token);
1379 /* maintain lp token */
1380 sigirefs_drop(p);
1381 }
1382 }
1383 }
1384
1385 /*
1386 * Deliver to the process generically if (1) the signal is being
1387 * sent to any thread or (2) we could not find a thread to deliver
1388 * it to.
1389 *
1390 * Drop p_sigirefs after the signal has been resolved to interlock
1391 * against sigsuspend/ppoll/pselect.
1392 */
1393 if (lp == NULL) {
1394 sigsetfrompid(curthread, p, sig);
1395 KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
1396 SIGADDSET_ATOMIC(p->p_siglist, sig);
1397 sigirefs_drop(p);
1398 goto out;
1399 }
1400
1401 /*
1402 * Deliver to a specific LWP whether it masks it or not. It will
1403 * not be dispatched if masked but we must still deliver it.
1404 */
1405 if (p->p_nice > NZERO && action == SIG_DFL && (prop & SA_KILL) &&
1406 (p->p_flags & P_TRACED) == 0) {
1407 lwkt_gettoken(&p->p_token);
1408 p->p_nice = NZERO;
1409 lwkt_reltoken(&p->p_token);
1410 }
1411
1412 /*
1413 * If the process receives a STOP signal which indeed needs to
1414 * stop the process, do so. If the process chose to catch the
1415 * signal, it will be treated like any other signal.
1416 */
1417 if ((prop & SA_STOP) && action == SIG_DFL) {
1418 /*
1419 * If a child holding parent blocked, stopping
1420 * could cause deadlock. Take no action at this
1421 * time.
1422 */
1423 lwkt_gettoken(&p->p_token);
1424 if (p->p_flags & P_PPWAIT) {
1425 sigsetfrompid(curthread, p, sig);
1426 SIGADDSET_ATOMIC(p->p_siglist, sig);
1427 lwkt_reltoken(&p->p_token);
1428 goto out;
1429 }
1430
1431 /*
1432 * Do not actually try to manipulate the process, but simply
1433 * stop it. Lwps will stop as soon as they safely can.
1434 *
1435 * Ignore stop if the process is exiting.
1436 */
1437 if ((p->p_flags & P_WEXIT) == 0) {
1438 p->p_xstat = sig;
1439 proc_stop(p, SSTOP);
1440 }
1441 lwkt_reltoken(&p->p_token);
1442 goto out;
1443 }
1444
1445 /*
1446 * If it is a CONT signal with default action, just ignore it.
1447 */
1448 if ((prop & SA_CONT) && action == SIG_DFL)
1449 goto out;
1450
1451 /*
1452 * Mark signal pending at this specific thread.
1453 */
1454 sigsetfrompid(curthread, p, sig);
1455 spin_lock(&lp->lwp_spin);
1456 SIGADDSET(lp->lwp_siglist, sig);
1457 spin_unlock(&lp->lwp_spin);
1458
1459 lwp_signotify(lp);
1460
1461 out:
1462 if (lp) {
1463 lwkt_reltoken(&lp->lwp_token);
1464 LWPRELE(lp);
1465 } else {
1466 lwkt_reltoken(&p->p_token);
1467 }
1468 PRELE(p);
1469 }
1470
1471 /*
1472 * Notify the LWP that a signal has arrived. The LWP does not have to be
1473 * sleeping on the current cpu.
1474 *
1475 * p->p_token and lp->lwp_token must be held on call.
1476 *
1477 * We can only safely schedule the thread on its current cpu and only if
1478 * one of the SINTR flags is set. If an SINTR flag is set AND we are on
1479 * the correct cpu we are properly interlocked, otherwise we could be
1480 * racing other thread transition states (or the lwp is on the user scheduler
1481 * runq but not scheduled) and must not do anything.
1482 *
1483 * Since we hold the lwp token we know the lwp cannot be ripped out from
1484 * under us so we can safely hold it to prevent it from being ripped out
1485 * from under us if we are forced to IPI another cpu to make the local
1486 * checks there.
1487 *
1488 * Adjustment of lp->lwp_stat can only occur when we hold the lwp_token,
1489 * which we won't in an IPI so any fixups have to be done here, effectively
1490 * replicating part of what setrunnable() does.
1491 */
1492 static void
lwp_signotify(struct lwp * lp)1493 lwp_signotify(struct lwp *lp)
1494 {
1495 thread_t dtd;
1496
1497 ASSERT_LWKT_TOKEN_HELD(&lp->lwp_token);
1498 dtd = lp->lwp_thread;
1499
1500 crit_enter();
1501 if (lp == lwkt_preempted_proc()) {
1502 /*
1503 * lwp is on the current cpu AND it is currently running
1504 * (we preempted it).
1505 */
1506 signotify();
1507 } else if (lp->lwp_flags & LWP_SINTR) {
1508 /*
1509 * lwp is sitting in tsleep() with PCATCH set
1510 */
1511 if (dtd->td_gd == mycpu) {
1512 setrunnable(lp);
1513 } else {
1514 /*
1515 * We can only adjust lwp_stat while we hold the
1516 * lwp_token, and we won't in the IPI function.
1517 */
1518 LWPHOLD(lp);
1519 if (lp->lwp_stat == LSSTOP)
1520 lp->lwp_stat = LSSLEEP;
1521 lwkt_send_ipiq(dtd->td_gd, lwp_signotify_remote, lp);
1522 }
1523 } else if (dtd->td_flags & TDF_SINTR) {
1524 /*
1525 * lwp is sitting in lwkt_sleep() with PCATCH set.
1526 */
1527 if (dtd->td_gd == mycpu) {
1528 setrunnable(lp);
1529 } else {
1530 /*
1531 * We can only adjust lwp_stat while we hold the
1532 * lwp_token, and we won't in the IPI function.
1533 */
1534 LWPHOLD(lp);
1535 if (lp->lwp_stat == LSSTOP)
1536 lp->lwp_stat = LSSLEEP;
1537 lwkt_send_ipiq(dtd->td_gd, lwp_signotify_remote, lp);
1538 }
1539 } else {
1540 /*
1541 * Otherwise the lwp is either in some uninterruptible state
1542 * or it is on the userland scheduler's runqueue waiting to
1543 * be scheduled to a cpu, or it is running in userland. We
1544 * generally want to send an IPI so a running target gets the
1545 * signal ASAP, otherwise a scheduler-tick worth of latency
1546 * will occur.
1547 *
1548 * Issue an IPI to the remote cpu to knock it into the kernel,
1549 * remote cpu will issue the cpu-local signotify() if the IPI
1550 * preempts the desired thread.
1551 */
1552 if (dtd->td_gd != mycpu) {
1553 LWPHOLD(lp);
1554 lwkt_send_ipiq(dtd->td_gd, lwp_signotify_remote, lp);
1555 }
1556 }
1557 crit_exit();
1558 }
1559
1560 /*
1561 * This function is called via an IPI so we cannot call setrunnable() here
1562 * (because while we hold the lp we don't own its token, and can't get it
1563 * from an IPI).
1564 *
1565 * We are interlocked by virtue of being on the same cpu as the target. If
1566 * we still are and LWP_SINTR or TDF_SINTR is set we can safely schedule
1567 * the target thread.
1568 */
1569 static void
lwp_signotify_remote(void * arg)1570 lwp_signotify_remote(void *arg)
1571 {
1572 struct lwp *lp = arg;
1573 thread_t td = lp->lwp_thread;
1574
1575 if (lp == lwkt_preempted_proc()) {
1576 signotify();
1577 LWPRELE(lp);
1578 } else if (td->td_gd == mycpu) {
1579 if ((lp->lwp_flags & LWP_SINTR) ||
1580 (td->td_flags & TDF_SINTR)) {
1581 lwkt_schedule(td);
1582 }
1583 LWPRELE(lp);
1584 } else {
1585 lwkt_send_ipiq(td->td_gd, lwp_signotify_remote, lp);
1586 /* LWPHOLD() is forwarded to the target cpu */
1587 }
1588 }
1589
1590 /*
1591 * Caller must hold p->p_token
1592 */
1593 void
proc_stop(struct proc * p,int stat)1594 proc_stop(struct proc *p, int stat)
1595 {
1596 struct proc *q;
1597 struct lwp *lp;
1598
1599 ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1600
1601 /*
1602 * If somebody raced us, be happy with it. SCORE overrides SSTOP.
1603 */
1604 if (stat == SCORE) {
1605 if (p->p_stat == SCORE || p->p_stat == SZOMB)
1606 return;
1607 } else {
1608 if (p->p_stat == SSTOP || p->p_stat == SCORE ||
1609 p->p_stat == SZOMB) {
1610 return;
1611 }
1612 }
1613 p->p_stat = stat;
1614
1615 FOREACH_LWP_IN_PROC(lp, p) {
1616 LWPHOLD(lp);
1617 lwkt_gettoken(&lp->lwp_token);
1618
1619 switch (lp->lwp_stat) {
1620 case LSSTOP:
1621 /*
1622 * Do nothing, we are already counted in
1623 * p_nstopped.
1624 */
1625 break;
1626
1627 case LSSLEEP:
1628 /*
1629 * We're sleeping, but we will stop before
1630 * returning to userspace, so count us
1631 * as stopped as well. We set LWP_MP_WSTOP
1632 * to signal the lwp that it should not
1633 * increase p_nstopped when reaching tstop().
1634 *
1635 * LWP_MP_WSTOP is protected by lp->lwp_token.
1636 */
1637 if ((lp->lwp_mpflags & LWP_MP_WSTOP) == 0) {
1638 atomic_set_int(&lp->lwp_mpflags, LWP_MP_WSTOP);
1639 ++p->p_nstopped;
1640 }
1641 break;
1642
1643 case LSRUN:
1644 /*
1645 * We might notify ourself, but that's not
1646 * a problem.
1647 */
1648 lwp_signotify(lp);
1649 break;
1650 }
1651 lwkt_reltoken(&lp->lwp_token);
1652 LWPRELE(lp);
1653 }
1654
1655 if (p->p_nstopped == p->p_nthreads) {
1656 /*
1657 * Token required to interlock kern_wait(). Reparenting can
1658 * also cause a race so we have to hold (q).
1659 */
1660 q = p->p_pptr;
1661 PHOLD(q);
1662 lwkt_gettoken(&q->p_token);
1663 p->p_flags &= ~P_WAITED;
1664 wakeup(q);
1665 if ((q->p_sigacts->ps_flag & PS_NOCLDSTOP) == 0)
1666 ksignal(p->p_pptr, SIGCHLD);
1667 lwkt_reltoken(&q->p_token);
1668 PRELE(q);
1669 }
1670 }
1671
1672 /*
1673 * Caller must hold p_token
1674 */
1675 void
proc_unstop(struct proc * p,int stat)1676 proc_unstop(struct proc *p, int stat)
1677 {
1678 struct lwp *lp;
1679
1680 ASSERT_LWKT_TOKEN_HELD(&p->p_token);
1681
1682 if (p->p_stat != stat)
1683 return;
1684
1685 p->p_stat = SACTIVE;
1686
1687 FOREACH_LWP_IN_PROC(lp, p) {
1688 LWPHOLD(lp);
1689 lwkt_gettoken(&lp->lwp_token);
1690
1691 switch (lp->lwp_stat) {
1692 case LSRUN:
1693 /*
1694 * Uh? Not stopped? Well, I guess that's okay.
1695 */
1696 if (bootverbose)
1697 kprintf("proc_unstop: lwp %d/%d not sleeping\n",
1698 p->p_pid, lp->lwp_tid);
1699 break;
1700
1701 case LSSLEEP:
1702 /*
1703 * Still sleeping. Don't bother waking it up.
1704 * However, if this thread was counted as
1705 * stopped, undo this.
1706 *
1707 * Nevertheless we call setrunnable() so that it
1708 * will wake up in case a signal or timeout arrived
1709 * in the meantime.
1710 *
1711 * LWP_MP_WSTOP is protected by lp->lwp_token.
1712 */
1713 if (lp->lwp_mpflags & LWP_MP_WSTOP) {
1714 atomic_clear_int(&lp->lwp_mpflags,
1715 LWP_MP_WSTOP);
1716 --p->p_nstopped;
1717 } else {
1718 if (bootverbose)
1719 kprintf("proc_unstop: lwp %d/%d sleeping, not stopped\n",
1720 p->p_pid, lp->lwp_tid);
1721 }
1722 /* FALLTHROUGH */
1723
1724 case LSSTOP:
1725 /*
1726 * This handles any lwp's waiting in a tsleep with
1727 * SIGCATCH.
1728 */
1729 lwp_signotify(lp);
1730 break;
1731
1732 }
1733 lwkt_reltoken(&lp->lwp_token);
1734 LWPRELE(lp);
1735 }
1736
1737 /*
1738 * This handles any lwp's waiting in tstop(). We have interlocked
1739 * the setting of p_stat by acquiring and releasing each lpw's
1740 * token.
1741 */
1742 wakeup(p);
1743 }
1744
1745 /*
1746 * Wait for all threads except the current thread to stop.
1747 */
1748 static void
proc_stopwait(struct proc * p)1749 proc_stopwait(struct proc *p)
1750 {
1751 while ((p->p_stat == SSTOP || p->p_stat == SCORE) &&
1752 p->p_nstopped < p->p_nthreads - 1) {
1753 tsleep_interlock(&p->p_nstopped, 0);
1754 if (p->p_nstopped < p->p_nthreads - 1) {
1755 tsleep(&p->p_nstopped, PINTERLOCKED, "stopwt", hz);
1756 }
1757 }
1758 }
1759
1760 /*
1761 * No requirements.
1762 */
1763 static int
kern_sigtimedwait(sigset_t waitset,siginfo_t * info,struct timespec * timeout)1764 kern_sigtimedwait(sigset_t waitset, siginfo_t *info, struct timespec *timeout)
1765 {
1766 sigset_t savedmask, set;
1767 struct proc *p = curproc;
1768 struct lwp *lp = curthread->td_lwp;
1769 int error, sig, hz, timevalid = 0;
1770 struct timespec rts, ets, ts;
1771 struct timeval tv;
1772
1773 error = 0;
1774 sig = 0;
1775 ets.tv_sec = 0; /* silence compiler warning */
1776 ets.tv_nsec = 0; /* silence compiler warning */
1777 SIG_CANTMASK(waitset);
1778 savedmask = lp->lwp_sigmask;
1779
1780 if (timeout) {
1781 if (timeout->tv_sec >= 0 && timeout->tv_nsec >= 0 &&
1782 timeout->tv_nsec < 1000000000) {
1783 timevalid = 1;
1784 getnanouptime(&rts);
1785 timespecadd(&rts, timeout, &ets);
1786 }
1787 }
1788
1789 for (;;) {
1790 set = lwp_sigpend(lp);
1791 SIGSETAND(set, waitset);
1792 if ((sig = sig_ffs(&set)) != 0) {
1793 SIGFILLSET(lp->lwp_sigmask);
1794 SIGDELSET(lp->lwp_sigmask, sig);
1795 SIG_CANTMASK(lp->lwp_sigmask);
1796 sig = issignal(lp, 1, 0);
1797 /*
1798 * It may be a STOP signal, in the case, issignal
1799 * returns 0, because we may stop there, and new
1800 * signal can come in, we should restart if we got
1801 * nothing.
1802 */
1803 if (sig == 0)
1804 continue;
1805 else
1806 break;
1807 }
1808
1809 /*
1810 * Previous checking got nothing, and we retried but still
1811 * got nothing, we should return the error status.
1812 */
1813 if (error)
1814 break;
1815
1816 /*
1817 * POSIX says this must be checked after looking for pending
1818 * signals.
1819 */
1820 if (timeout) {
1821 if (timevalid == 0) {
1822 error = EINVAL;
1823 break;
1824 }
1825 getnanouptime(&rts);
1826 if (timespeccmp(&rts, &ets, >=)) {
1827 error = EAGAIN;
1828 break;
1829 }
1830 timespecsub(&ets, &rts, &ts);
1831 TIMESPEC_TO_TIMEVAL(&tv, &ts);
1832 hz = tvtohz_high(&tv);
1833 } else {
1834 hz = 0;
1835 }
1836
1837 lp->lwp_sigmask = savedmask;
1838 SIGSETNAND(lp->lwp_sigmask, waitset);
1839 sigirefs_wait(p);
1840
1841 /*
1842 * We won't ever be woken up. Instead, our sleep will
1843 * be broken in lwpsignal().
1844 */
1845 error = tsleep(&p->p_sigacts, PCATCH, "sigwt", hz);
1846 if (timeout) {
1847 if (error == ERESTART) {
1848 /* can not restart a timeout wait. */
1849 error = EINTR;
1850 } else if (error == EAGAIN) {
1851 /* will calculate timeout by ourself. */
1852 error = 0;
1853 }
1854 }
1855 /* Retry ... */
1856 }
1857
1858 lp->lwp_sigmask = savedmask;
1859 sigirefs_wait(p);
1860
1861 if (sig) {
1862 error = 0;
1863 bzero(info, sizeof(*info));
1864 info->si_signo = sig;
1865 spin_lock(&lp->lwp_spin);
1866 lwp_delsig(lp, sig, 1); /* take the signal! */
1867 spin_unlock(&lp->lwp_spin);
1868
1869 if (sig == SIGKILL) {
1870 sigexit(lp, sig);
1871 /* NOT REACHED */
1872 }
1873 }
1874
1875 return (error);
1876 }
1877
1878 /*
1879 * MPALMOSTSAFE
1880 */
1881 int
sys_sigtimedwait(struct sysmsg * sysmsg,const struct sigtimedwait_args * uap)1882 sys_sigtimedwait(struct sysmsg *sysmsg, const struct sigtimedwait_args *uap)
1883 {
1884 struct timespec ts;
1885 struct timespec *timeout;
1886 sigset_t set;
1887 siginfo_t info;
1888 int error;
1889
1890 if (uap->timeout) {
1891 error = copyin(uap->timeout, &ts, sizeof(ts));
1892 if (error)
1893 return (error);
1894 timeout = &ts;
1895 } else {
1896 timeout = NULL;
1897 }
1898 error = copyin(uap->set, &set, sizeof(set));
1899 if (error)
1900 return (error);
1901 error = kern_sigtimedwait(set, &info, timeout);
1902 if (error)
1903 return (error);
1904 if (uap->info)
1905 error = copyout(&info, uap->info, sizeof(info));
1906 /* Repost if we got an error. */
1907 /*
1908 * XXX lwp
1909 *
1910 * This could transform a thread-specific signal to another
1911 * thread / process pending signal.
1912 */
1913 if (error) {
1914 ksignal(curproc, info.si_signo);
1915 } else {
1916 sysmsg->sysmsg_result = info.si_signo;
1917 }
1918 return (error);
1919 }
1920
1921 /*
1922 * MPALMOSTSAFE
1923 */
1924 int
sys_sigwaitinfo(struct sysmsg * sysmsg,const struct sigwaitinfo_args * uap)1925 sys_sigwaitinfo(struct sysmsg *sysmsg, const struct sigwaitinfo_args *uap)
1926 {
1927 siginfo_t info;
1928 sigset_t set;
1929 int error;
1930
1931 error = copyin(uap->set, &set, sizeof(set));
1932 if (error)
1933 return (error);
1934 error = kern_sigtimedwait(set, &info, NULL);
1935 if (error)
1936 return (error);
1937 if (uap->info)
1938 error = copyout(&info, uap->info, sizeof(info));
1939 /* Repost if we got an error. */
1940 /*
1941 * XXX lwp
1942 *
1943 * This could transform a thread-specific signal to another
1944 * thread / process pending signal.
1945 */
1946 if (error) {
1947 ksignal(curproc, info.si_signo);
1948 } else {
1949 sysmsg->sysmsg_result = info.si_signo;
1950 }
1951 return (error);
1952 }
1953
1954 /*
1955 * If the current process has received a signal that would interrupt a
1956 * system call, return EINTR or ERESTART as appropriate.
1957 */
1958 int
iscaught(struct lwp * lp)1959 iscaught(struct lwp *lp)
1960 {
1961 struct proc *p = lp->lwp_proc;
1962 int sig;
1963
1964 if (p) {
1965 if ((sig = CURSIG(lp)) != 0) {
1966 if (SIGISMEMBER(p->p_sigacts->ps_sigintr, sig))
1967 return (EINTR);
1968 return (ERESTART);
1969 }
1970 }
1971 return(EWOULDBLOCK);
1972 }
1973
1974 /*
1975 * If the current lwp/proc has received a signal (should be caught or cause
1976 * termination, should interrupt current syscall), return the signal number.
1977 * Stop signals with default action are processed immediately, then cleared;
1978 * they aren't returned. This is checked after each entry to the system for
1979 * a syscall or trap (though this can usually be done without calling issignal
1980 * by checking the pending signal masks in the CURSIG macro).
1981 *
1982 * This routine is called via CURSIG/__cursig. We will acquire and release
1983 * p->p_token but if the caller needs to interlock the test the caller must
1984 * also hold p->p_token.
1985 *
1986 * while (sig = CURSIG(curproc))
1987 * postsig(sig);
1988 */
1989 int
issignal(struct lwp * lp,int maytrace,int * ptokp)1990 issignal(struct lwp *lp, int maytrace, int *ptokp)
1991 {
1992 struct proc *p = lp->lwp_proc;
1993 sigset_t mask;
1994 int sig, prop;
1995 int haveptok;
1996
1997 for (;;) {
1998 int traced = (p->p_flags & P_TRACED) || (p->p_stops & S_SIG);
1999
2000 haveptok = 0;
2001
2002 /*
2003 * NOTE: Do not tstop here. Issue the proc_stop()
2004 * so other parties see that we know we need
2005 * to stop, but don't block here. Locks might
2006 * be held.
2007 *
2008 * XXX If this process is supposed to stop, stop this thread.
2009 * removed.
2010 */
2011 #if 0
2012 if (STOPLWP(p, lp)) {
2013 lwkt_gettoken(&p->p_token);
2014 tstop();
2015 lwkt_reltoken(&p->p_token);
2016 }
2017 #endif
2018
2019 /*
2020 * Quick check without token
2021 */
2022 mask = lwp_sigpend(lp);
2023 SIGSETNAND(mask, lp->lwp_sigmask);
2024 if (p->p_flags & P_PPWAIT)
2025 SIG_STOPSIGMASK(mask);
2026 SIG_CONDBLOCKALLSIGS(mask, lp);
2027
2028 if (SIGISEMPTY(mask)) /* no signal to send */
2029 return (0);
2030
2031 /*
2032 * If the signal is a member of the process signal set
2033 * we need p_token (even if it is also a member of the
2034 * lwp signal set).
2035 */
2036 sig = sig_ffs(&mask);
2037 if (SIGISMEMBER(p->p_siglist, sig)) {
2038 /*
2039 * Recheck with token
2040 */
2041 haveptok = 1;
2042 lwkt_gettoken(&p->p_token);
2043
2044 mask = lwp_sigpend(lp);
2045 SIGSETNAND(mask, lp->lwp_sigmask);
2046 if (p->p_flags & P_PPWAIT)
2047 SIG_STOPSIGMASK(mask);
2048 if (SIGISEMPTY(mask)) { /* no signal to send */
2049 /* haveptok is TRUE */
2050 lwkt_reltoken(&p->p_token);
2051 return (0);
2052 }
2053 sig = sig_ffs(&mask);
2054 }
2055
2056 STOPEVENT(p, S_SIG, sig);
2057
2058 /*
2059 * We should see pending but ignored signals
2060 * only if P_TRACED was on when they were posted.
2061 */
2062 if (SIGISMEMBER(p->p_sigignore, sig) && (traced == 0)) {
2063 spin_lock(&lp->lwp_spin);
2064 lwp_delsig(lp, sig, haveptok);
2065 spin_unlock(&lp->lwp_spin);
2066 if (haveptok)
2067 lwkt_reltoken(&p->p_token);
2068 continue;
2069 }
2070 if (maytrace &&
2071 (p->p_flags & P_TRACED) &&
2072 (p->p_flags & P_PPWAIT) == 0) {
2073 /*
2074 * If traced, always stop, and stay stopped until
2075 * released by the parent.
2076 *
2077 * NOTE: SSTOP may get cleared during the loop, but
2078 * we do not re-notify the parent if we have
2079 * to loop several times waiting for the parent
2080 * to let us continue. XXX not sure if this is
2081 * still true
2082 *
2083 * NOTE: Do not tstop here. Issue the proc_stop()
2084 * so other parties see that we know we need
2085 * to stop, but don't block here. Locks might
2086 * be held.
2087 */
2088 if (haveptok == 0) {
2089 lwkt_gettoken(&p->p_token);
2090 haveptok = 1;
2091 }
2092 p->p_xstat = sig;
2093 proc_stop(p, SSTOP);
2094
2095 /*
2096 * Normally we don't stop until we return to userland, but
2097 * make an exception when tracing and 'maytrace' is asserted.
2098 */
2099 if (p->p_flags & P_TRACED)
2100 tstop();
2101
2102 /*
2103 * If parent wants us to take the signal,
2104 * then it will leave it in p->p_xstat;
2105 * otherwise we just look for signals again.
2106 */
2107 spin_lock(&lp->lwp_spin);
2108 lwp_delsig(lp, sig, 1); /* clear old signal */
2109 spin_unlock(&lp->lwp_spin);
2110 sig = p->p_xstat;
2111 if (sig == 0) {
2112 /* haveptok is TRUE */
2113 lwkt_reltoken(&p->p_token);
2114 continue;
2115 }
2116
2117 /*
2118 * Put the new signal into p_siglist. If the
2119 * signal is being masked, look for other signals.
2120 *
2121 * XXX lwp might need a call to ksignal()
2122 */
2123 SIGADDSET_ATOMIC(p->p_siglist, sig);
2124 if (SIGISMEMBER(lp->lwp_sigmask, sig)) {
2125 /* haveptok is TRUE */
2126 lwkt_reltoken(&p->p_token);
2127 continue;
2128 }
2129
2130 /*
2131 * If the traced bit got turned off, go back up
2132 * to the top to rescan signals. This ensures
2133 * that p_sig* and ps_sigact are consistent.
2134 */
2135 if ((p->p_flags & P_TRACED) == 0) {
2136 /* haveptok is TRUE */
2137 lwkt_reltoken(&p->p_token);
2138 continue;
2139 }
2140 }
2141
2142 /*
2143 * p_token may be held here
2144 */
2145 prop = sigprop(sig);
2146
2147 /*
2148 * Decide whether the signal should be returned.
2149 * Return the signal's number, or fall through
2150 * to clear it from the pending mask.
2151 */
2152 switch ((intptr_t)p->p_sigacts->ps_sigact[_SIG_IDX(sig)]) {
2153 case (intptr_t)SIG_DFL:
2154 /*
2155 * Don't take default actions on system processes.
2156 */
2157 if (p->p_pid <= 1) {
2158 #ifdef DIAGNOSTIC
2159 /*
2160 * Are you sure you want to ignore SIGSEGV
2161 * in init? XXX
2162 */
2163 kprintf("Process (pid %lu) got signal %d\n",
2164 (u_long)p->p_pid, sig);
2165 #endif
2166 break; /* == ignore */
2167 }
2168
2169 /*
2170 * Handle the in-kernel checkpoint action
2171 */
2172 if (prop & SA_CKPT) {
2173 if (haveptok == 0) {
2174 lwkt_gettoken(&p->p_token);
2175 haveptok = 1;
2176 }
2177 checkpoint_signal_handler(lp);
2178 break;
2179 }
2180
2181 /*
2182 * If there is a pending stop signal to process
2183 * with default action, stop here,
2184 * then clear the signal. However,
2185 * if process is member of an orphaned
2186 * process group, ignore tty stop signals.
2187 */
2188 if (prop & SA_STOP) {
2189 if (haveptok == 0) {
2190 lwkt_gettoken(&p->p_token);
2191 haveptok = 1;
2192 }
2193 if (p->p_flags & P_TRACED ||
2194 (p->p_pgrp->pg_jobc == 0 &&
2195 prop & SA_TTYSTOP))
2196 break; /* == ignore */
2197 if ((p->p_flags & P_WEXIT) == 0) {
2198 /*
2199 * NOTE: We do not block here. Issue
2200 * stopthe stop so other parties
2201 * see that we know we need to
2202 * stop. Locks might be held.
2203 */
2204 p->p_xstat = sig;
2205 proc_stop(p, SSTOP);
2206
2207 #if 0
2208 tstop();
2209 #endif
2210 }
2211 break;
2212 } else if (prop & SA_IGNORE) {
2213 /*
2214 * Except for SIGCONT, shouldn't get here.
2215 * Default action is to ignore; drop it.
2216 */
2217 break; /* == ignore */
2218 } else {
2219 if (ptokp)
2220 *ptokp = haveptok;
2221 else if (haveptok)
2222 lwkt_reltoken(&p->p_token);
2223 return (sig);
2224 }
2225
2226 /*NOTREACHED*/
2227
2228 case (intptr_t)SIG_IGN:
2229 /*
2230 * Masking above should prevent us ever trying
2231 * to take action on an ignored signal other
2232 * than SIGCONT, unless process is traced.
2233 */
2234 if ((prop & SA_CONT) == 0 &&
2235 (p->p_flags & P_TRACED) == 0)
2236 kprintf("issignal\n");
2237 break; /* == ignore */
2238
2239 default:
2240 /*
2241 * This signal has an action, let
2242 * postsig() process it.
2243 */
2244 if (ptokp)
2245 *ptokp = haveptok;
2246 else if (haveptok)
2247 lwkt_reltoken(&p->p_token);
2248 return (sig);
2249 }
2250 spin_lock(&lp->lwp_spin);
2251 lwp_delsig(lp, sig, haveptok); /* take the signal! */
2252 spin_unlock(&lp->lwp_spin);
2253
2254 if (haveptok)
2255 lwkt_reltoken(&p->p_token);
2256 }
2257 /* NOTREACHED */
2258 }
2259
2260 /*
2261 * Take the action for the specified signal from the current set of
2262 * pending signals.
2263 *
2264 * haveptok indicates whether the caller is holding p->p_token. If the
2265 * caller is, we are responsible for releasing it.
2266 *
2267 * This routine can only be called from the top-level trap from usermode.
2268 * It is expecting to be able to modify the top-level stack frame.
2269 */
2270 void
postsig(int sig,int haveptok)2271 postsig(int sig, int haveptok)
2272 {
2273 struct lwp *lp = curthread->td_lwp;
2274 struct proc *p = lp->lwp_proc;
2275 struct sigacts *ps = p->p_sigacts;
2276 sig_t action;
2277 sigset_t returnmask;
2278 int code;
2279
2280 KASSERT(sig != 0, ("postsig"));
2281
2282 /*
2283 * If we are a virtual kernel running an emulated user process
2284 * context, switch back to the virtual kernel context before
2285 * trying to post the signal.
2286 */
2287 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
2288 struct trapframe *tf = lp->lwp_md.md_regs;
2289 tf->tf_trapno = 0;
2290 vkernel_trap(lp, tf);
2291 }
2292
2293 KNOTE(&p->p_klist, NOTE_SIGNAL | sig);
2294
2295 spin_lock(&lp->lwp_spin);
2296 lwp_delsig(lp, sig, haveptok);
2297 spin_unlock(&lp->lwp_spin);
2298 action = ps->ps_sigact[_SIG_IDX(sig)];
2299 #ifdef KTRACE
2300 if (KTRPOINT(lp->lwp_thread, KTR_PSIG))
2301 ktrpsig(lp, sig, action, lp->lwp_flags & LWP_OLDMASK ?
2302 &lp->lwp_oldsigmask : &lp->lwp_sigmask, 0);
2303 #endif
2304 /*
2305 * We don't need p_token after this point.
2306 */
2307 if (haveptok)
2308 lwkt_reltoken(&p->p_token);
2309
2310 STOPEVENT(p, S_SIG, sig);
2311
2312 if (action == SIG_DFL) {
2313 /*
2314 * Default action, where the default is to kill
2315 * the process. (Other cases were ignored above.)
2316 */
2317 sigexit(lp, sig);
2318 /* NOTREACHED */
2319 } else {
2320 /*
2321 * If we get here, the signal must be caught.
2322 */
2323 KASSERT(action != SIG_IGN && !SIGISMEMBER(lp->lwp_sigmask, sig),
2324 ("postsig action"));
2325
2326 /*
2327 * Reset the signal handler if asked to
2328 */
2329 if (SIGISMEMBER(ps->ps_sigreset, sig)) {
2330 /*
2331 * See kern_sigaction() for origin of this code.
2332 */
2333 SIGDELSET(p->p_sigcatch, sig);
2334 if (sig != SIGCONT &&
2335 sigprop(sig) & SA_IGNORE)
2336 SIGADDSET(p->p_sigignore, sig);
2337 ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL;
2338 }
2339
2340 /*
2341 * Set the signal mask and calculate the mask to restore
2342 * when the signal function returns.
2343 *
2344 * Special case: user has done a sigsuspend. Here the
2345 * current mask is not of interest, but rather the
2346 * mask from before the sigsuspend is what we want
2347 * restored after the signal processing is completed.
2348 */
2349 if (lp->lwp_flags & LWP_OLDMASK) {
2350 returnmask = lp->lwp_oldsigmask;
2351 lp->lwp_flags &= ~LWP_OLDMASK;
2352 } else {
2353 returnmask = lp->lwp_sigmask;
2354 }
2355
2356 SIGSETOR(lp->lwp_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
2357 if (!SIGISMEMBER(ps->ps_signodefer, sig))
2358 SIGADDSET(lp->lwp_sigmask, sig);
2359
2360 lp->lwp_ru.ru_nsignals++;
2361 if (lp->lwp_sig != sig) {
2362 code = 0;
2363 } else {
2364 code = lp->lwp_code;
2365 lp->lwp_code = 0;
2366 lp->lwp_sig = 0;
2367 }
2368 (*p->p_sysent->sv_sendsig)(action, sig, &returnmask, code);
2369 }
2370 }
2371
2372 /*
2373 * Kill the current process for stated reason.
2374 */
2375 void
killproc(struct proc * p,char * why)2376 killproc(struct proc *p, char *why)
2377 {
2378 log(LOG_ERR, "pid %d (%s), uid %d, was killed: %s\n",
2379 p->p_pid, p->p_comm,
2380 p->p_ucred ? p->p_ucred->cr_uid : -1, why);
2381 ksignal(p, SIGKILL);
2382 }
2383
2384 /*
2385 * Force the current process to exit with the specified signal, dumping core
2386 * if appropriate. We bypass the normal tests for masked and caught signals,
2387 * allowing unrecoverable failures to terminate the process without changing
2388 * signal state. Mark the accounting record with the signal termination.
2389 * If dumping core, save the signal number for the debugger. Calls exit and
2390 * does not return.
2391 *
2392 * This routine does not return.
2393 */
2394 void
sigexit(struct lwp * lp,int sig)2395 sigexit(struct lwp *lp, int sig)
2396 {
2397 struct proc *p = lp->lwp_proc;
2398
2399 lwkt_gettoken(&p->p_token);
2400 p->p_acflag |= AXSIG;
2401 if (sigprop(sig) & SA_CORE) {
2402 lp->lwp_sig = sig;
2403
2404 /*
2405 * All threads must be stopped before we can safely coredump.
2406 * Stop threads using SCORE, which cannot be overridden.
2407 */
2408 if (p->p_stat != SCORE) {
2409 proc_stop(p, SCORE);
2410 proc_stopwait(p);
2411
2412 if (coredump(lp, sig) == 0)
2413 sig |= WCOREFLAG;
2414 p->p_stat = SSTOP;
2415 }
2416
2417 /*
2418 * Log signals which would cause core dumps
2419 * (Log as LOG_INFO to appease those who don't want
2420 * these messages.)
2421 * XXX : Todo, as well as euid, write out ruid too
2422 */
2423 if (kern_logsigexit) {
2424 log(LOG_INFO,
2425 "pid %d (%s), uid %d: exited on signal %d%s\n",
2426 p->p_pid, p->p_comm,
2427 p->p_ucred ? p->p_ucred->cr_uid : -1,
2428 sig &~ WCOREFLAG,
2429 sig & WCOREFLAG ? " (core dumped)" : "");
2430 if (kern_logsigexit > 1)
2431 kprintf("DEBUG - waiting on kern.logsigexit\n");
2432 while (kern_logsigexit > 1) {
2433 tsleep(&kern_logsigexit, 0, "DEBUG", hz);
2434 }
2435 }
2436 }
2437 lwkt_reltoken(&p->p_token);
2438 exit1(W_EXITCODE(0, sig));
2439 /* NOTREACHED */
2440 }
2441
2442 static char corefilename[MAXPATHLEN+1] = {"%N.core"};
2443 SYSCTL_STRING(_kern, OID_AUTO, corefile, CTLFLAG_RW, corefilename,
2444 sizeof(corefilename), "process corefile name format string");
2445
2446 /*
2447 * expand_name(name, uid, pid)
2448 * Expand the name described in corefilename, using name, uid, and pid.
2449 * corefilename is a kprintf-like string, with three format specifiers:
2450 * %N name of process ("name")
2451 * %P process id (pid)
2452 * %U user id (uid)
2453 * For example, "%N.core" is the default; they can be disabled completely
2454 * by using "/dev/null", or all core files can be stored in "/cores/%U/%N-%P".
2455 * This is controlled by the sysctl variable kern.corefile (see above).
2456 */
2457
2458 static char *
expand_name(const char * name,uid_t uid,pid_t pid)2459 expand_name(const char *name, uid_t uid, pid_t pid)
2460 {
2461 char *temp;
2462 char buf[11]; /* Buffer for pid/uid -- max 4B */
2463 int i, n;
2464 char *format = corefilename;
2465 size_t namelen;
2466
2467 temp = kmalloc(MAXPATHLEN + 1, M_TEMP, M_NOWAIT);
2468 if (temp == NULL)
2469 return NULL;
2470 namelen = strlen(name);
2471 for (i = 0, n = 0; n < MAXPATHLEN && format[i]; i++) {
2472 int l;
2473 switch (format[i]) {
2474 case '%': /* Format character */
2475 i++;
2476 switch (format[i]) {
2477 case '%':
2478 temp[n++] = '%';
2479 break;
2480 case 'N': /* process name */
2481 if ((n + namelen) > MAXPATHLEN) {
2482 log(LOG_ERR, "pid %d (%s), uid (%u): Path `%s%s' is too long\n",
2483 pid, name, uid, temp, name);
2484 kfree(temp, M_TEMP);
2485 return NULL;
2486 }
2487 memcpy(temp+n, name, namelen);
2488 n += namelen;
2489 break;
2490 case 'P': /* process id */
2491 l = ksprintf(buf, "%u", pid);
2492 if ((n + l) > MAXPATHLEN) {
2493 log(LOG_ERR, "pid %d (%s), uid (%u): Path `%s%s' is too long\n",
2494 pid, name, uid, temp, name);
2495 kfree(temp, M_TEMP);
2496 return NULL;
2497 }
2498 memcpy(temp+n, buf, l);
2499 n += l;
2500 break;
2501 case 'U': /* user id */
2502 l = ksprintf(buf, "%u", uid);
2503 if ((n + l) > MAXPATHLEN) {
2504 log(LOG_ERR, "pid %d (%s), uid (%u): Path `%s%s' is too long\n",
2505 pid, name, uid, temp, name);
2506 kfree(temp, M_TEMP);
2507 return NULL;
2508 }
2509 memcpy(temp+n, buf, l);
2510 n += l;
2511 break;
2512 default:
2513 log(LOG_ERR, "Unknown format character %c in `%s'\n", format[i], format);
2514 }
2515 break;
2516 default:
2517 temp[n++] = format[i];
2518 }
2519 }
2520 temp[n] = '\0';
2521 return temp;
2522 }
2523
2524 /*
2525 * Dump a process' core. The main routine does some
2526 * policy checking, and creates the name of the coredump;
2527 * then it passes on a vnode and a size limit to the process-specific
2528 * coredump routine if there is one; if there _is not_ one, it returns
2529 * ENOSYS; otherwise it returns the error from the process-specific routine.
2530 *
2531 * The parameter `lp' is the lwp which triggered the coredump.
2532 */
2533
2534 static int
coredump(struct lwp * lp,int sig)2535 coredump(struct lwp *lp, int sig)
2536 {
2537 struct proc *p = lp->lwp_proc;
2538 struct vnode *vp;
2539 struct ucred *cred = p->p_ucred;
2540 struct flock lf;
2541 struct nlookupdata nd;
2542 struct vattr vattr;
2543 int error, error1;
2544 char *name; /* name of corefile */
2545 off_t limit;
2546
2547 STOPEVENT(p, S_CORE, 0);
2548
2549 if (((sugid_coredump == 0) && (p->p_flags & P_SUGID)) ||
2550 do_coredump == 0)
2551 {
2552 return (EFAULT);
2553 }
2554
2555 /*
2556 * Note that the bulk of limit checking is done after
2557 * the corefile is created. The exception is if the limit
2558 * for corefiles is 0, in which case we don't bother
2559 * creating the corefile at all. This layout means that
2560 * a corefile is truncated instead of not being created,
2561 * if it is larger than the limit.
2562 */
2563 limit = p->p_rlimit[RLIMIT_CORE].rlim_cur;
2564 if (limit == 0)
2565 return EFBIG;
2566
2567 name = expand_name(p->p_comm, p->p_ucred->cr_uid, p->p_pid);
2568 if (name == NULL)
2569 return (EINVAL);
2570 error = nlookup_init(&nd, name, UIO_SYSSPACE, NLC_LOCKVP);
2571 if (error == 0)
2572 error = vn_open(&nd, NULL,
2573 O_CREAT | FWRITE | O_NOFOLLOW,
2574 S_IRUSR | S_IWUSR);
2575 kfree(name, M_TEMP);
2576 if (error) {
2577 nlookup_done(&nd);
2578 return (error);
2579 }
2580 vp = nd.nl_open_vp;
2581 nd.nl_open_vp = NULL;
2582 nlookup_done(&nd);
2583
2584 vn_unlock(vp);
2585 lf.l_whence = SEEK_SET;
2586 lf.l_start = 0;
2587 lf.l_len = 0;
2588 lf.l_type = F_WRLCK;
2589 error = VOP_ADVLOCK(vp, (caddr_t)p, F_SETLK, &lf, 0);
2590 if (error)
2591 goto out2;
2592
2593 /* Don't dump to non-regular files or files with links. */
2594 if (vp->v_type != VREG ||
2595 VOP_GETATTR(vp, &vattr) || vattr.va_nlink != 1) {
2596 error = EFAULT;
2597 goto out1;
2598 }
2599
2600 /* Don't dump to files current user does not own */
2601 if (vattr.va_uid != p->p_ucred->cr_uid) {
2602 error = EFAULT;
2603 goto out1;
2604 }
2605
2606 VATTR_NULL(&vattr);
2607 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2608 vattr.va_size = 0;
2609 VOP_SETATTR(vp, &vattr, cred);
2610 p->p_acflag |= ACORE;
2611 vn_unlock(vp);
2612
2613 error = p->p_sysent->sv_coredump ?
2614 p->p_sysent->sv_coredump(lp, sig, vp, limit) : ENOSYS;
2615
2616 out1:
2617 lf.l_type = F_UNLCK;
2618 VOP_ADVLOCK(vp, (caddr_t)p, F_UNLCK, &lf, 0);
2619 out2:
2620 error1 = vn_close(vp, FWRITE, NULL);
2621 if (error == 0)
2622 error = error1;
2623 return (error);
2624 }
2625
2626 /*
2627 * Nonexistent system call-- signal process (may want to handle it).
2628 * Flag error in case process won't see signal immediately (blocked or ignored).
2629 *
2630 * MPALMOSTSAFE
2631 */
2632 /* ARGSUSED */
2633 int
sys_nosys(struct sysmsg * sysmsg,const struct nosys_args * args)2634 sys_nosys(struct sysmsg *sysmsg, const struct nosys_args *args)
2635 {
2636 lwpsignal(curproc, curthread->td_lwp, SIGSYS);
2637 return (EINVAL);
2638 }
2639
2640 /*
2641 * Send a SIGIO or SIGURG signal to a process or process group using
2642 * stored credentials rather than those of the current process.
2643 */
2644 void
pgsigio(struct sigio * sigio,int sig,int checkctty)2645 pgsigio(struct sigio *sigio, int sig, int checkctty)
2646 {
2647 if (sigio == NULL)
2648 return;
2649
2650 if (sigio->sio_pgid > 0) {
2651 if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred,
2652 sigio->sio_proc))
2653 ksignal(sigio->sio_proc, sig);
2654 } else if (sigio->sio_pgid < 0) {
2655 struct proc *p;
2656 struct pgrp *pg = sigio->sio_pgrp;
2657
2658 /*
2659 * Must interlock all signals against fork
2660 */
2661 pgref(pg);
2662 lockmgr(&pg->pg_lock, LK_EXCLUSIVE);
2663 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
2664 if (CANSIGIO(sigio->sio_ruid, sigio->sio_ucred, p) &&
2665 (checkctty == 0 || (p->p_flags & P_CONTROLT)))
2666 ksignal(p, sig);
2667 }
2668 lockmgr(&pg->pg_lock, LK_RELEASE);
2669 pgrel(pg);
2670 }
2671 }
2672
2673 static int
filt_sigattach(struct knote * kn)2674 filt_sigattach(struct knote *kn)
2675 {
2676 struct proc *p = curproc;
2677
2678 kn->kn_ptr.p_proc = p;
2679 kn->kn_flags |= EV_CLEAR; /* automatically set */
2680
2681 /* XXX lock the proc here while adding to the list? */
2682 knote_insert(&p->p_klist, kn);
2683
2684 return (0);
2685 }
2686
2687 static void
filt_sigdetach(struct knote * kn)2688 filt_sigdetach(struct knote *kn)
2689 {
2690 struct proc *p = kn->kn_ptr.p_proc;
2691
2692 knote_remove(&p->p_klist, kn);
2693 }
2694
2695 /*
2696 * signal knotes are shared with proc knotes, so we apply a mask to
2697 * the hint in order to differentiate them from process hints. This
2698 * could be avoided by using a signal-specific knote list, but probably
2699 * isn't worth the trouble.
2700 */
2701 static int
filt_signal(struct knote * kn,long hint)2702 filt_signal(struct knote *kn, long hint)
2703 {
2704 if (hint & NOTE_SIGNAL) {
2705 hint &= ~NOTE_SIGNAL;
2706
2707 if (kn->kn_id == hint)
2708 kn->kn_data++;
2709 }
2710 return (kn->kn_data != 0);
2711 }
2712