Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / kernel / signal.c
1 /*
2 * linux/kernel/signal.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * 1997-11-02 Modified for POSIX.1b signals by Richard Henderson
7 *
8 * 2003-06-02 Jim Houston - Concurrent Computer Corp.
9 * Changes to use preallocated sigqueue structures
10 * to allow signals to be sent reliably.
11 */
12
13 #include <linux/slab.h>
14 #include <linux/module.h>
15 #include <linux/init.h>
16 #include <linux/sched.h>
17 #include <linux/fs.h>
18 #include <linux/tty.h>
19 #include <linux/binfmts.h>
20 #include <linux/security.h>
21 #include <linux/syscalls.h>
22 #include <linux/ptrace.h>
23 #include <linux/signal.h>
24 #include <linux/signalfd.h>
25 #include <linux/ratelimit.h>
26 #include <linux/tracehook.h>
27 #include <linux/capability.h>
28 #include <linux/freezer.h>
29 #include <linux/pid_namespace.h>
30 #include <linux/nsproxy.h>
31 #define CREATE_TRACE_POINTS
32 #include <trace/events/signal.h>
33
34 #include <asm/param.h>
35 #include <asm/uaccess.h>
36 #include <asm/unistd.h>
37 #include <asm/siginfo.h>
38 #include "audit.h" /* audit_signal_info() */
39
40 /*
41 * SLAB caches for signal bits.
42 */
43
44 static struct kmem_cache *sigqueue_cachep;
45
46 int print_fatal_signals __read_mostly;
47
48 static void __user *sig_handler(struct task_struct *t, int sig)
49 {
50 return t->sighand->action[sig - 1].sa.sa_handler;
51 }
52
53 static int sig_handler_ignored(void __user *handler, int sig)
54 {
55 /* Is it explicitly or implicitly ignored? */
56 return handler == SIG_IGN ||
57 (handler == SIG_DFL && sig_kernel_ignore(sig));
58 }
59
60 static int sig_task_ignored(struct task_struct *t, int sig,
61 int from_ancestor_ns)
62 {
63 void __user *handler;
64
65 handler = sig_handler(t, sig);
66
67 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
68 handler == SIG_DFL && !from_ancestor_ns)
69 return 1;
70
71 return sig_handler_ignored(handler, sig);
72 }
73
74 static int sig_ignored(struct task_struct *t, int sig, int from_ancestor_ns)
75 {
76 /*
77 * Blocked signals are never ignored, since the
78 * signal handler may change by the time it is
79 * unblocked.
80 */
81 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
82 return 0;
83
84 if (!sig_task_ignored(t, sig, from_ancestor_ns))
85 return 0;
86
87 /*
88 * Tracers may want to know about even ignored signals.
89 */
90 return !tracehook_consider_ignored_signal(t, sig);
91 }
92
93 /*
94 * Re-calculate pending state from the set of locally pending
95 * signals, globally pending signals, and blocked signals.
96 */
97 static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
98 {
99 unsigned long ready;
100 long i;
101
102 switch (_NSIG_WORDS) {
103 default:
104 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
105 ready |= signal->sig[i] &~ blocked->sig[i];
106 break;
107
108 case 4: ready = signal->sig[3] &~ blocked->sig[3];
109 ready |= signal->sig[2] &~ blocked->sig[2];
110 ready |= signal->sig[1] &~ blocked->sig[1];
111 ready |= signal->sig[0] &~ blocked->sig[0];
112 break;
113
114 case 2: ready = signal->sig[1] &~ blocked->sig[1];
115 ready |= signal->sig[0] &~ blocked->sig[0];
116 break;
117
118 case 1: ready = signal->sig[0] &~ blocked->sig[0];
119 }
120 return ready != 0;
121 }
122
123 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
124
125 static int recalc_sigpending_tsk(struct task_struct *t)
126 {
127 if (t->signal->group_stop_count > 0 ||
128 PENDING(&t->pending, &t->blocked) ||
129 PENDING(&t->signal->shared_pending, &t->blocked)) {
130 set_tsk_thread_flag(t, TIF_SIGPENDING);
131 return 1;
132 }
133 /*
134 * We must never clear the flag in another thread, or in current
135 * when it's possible the current syscall is returning -ERESTART*.
136 * So we don't clear it here, and only callers who know they should do.
137 */
138 return 0;
139 }
140
141 /*
142 * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
143 * This is superfluous when called on current, the wakeup is a harmless no-op.
144 */
145 void recalc_sigpending_and_wake(struct task_struct *t)
146 {
147 if (recalc_sigpending_tsk(t))
148 signal_wake_up(t, 0);
149 }
150
151 void recalc_sigpending(void)
152 {
153 if (unlikely(tracehook_force_sigpending()))
154 set_thread_flag(TIF_SIGPENDING);
155 else if (!recalc_sigpending_tsk(current) && !freezing(current))
156 clear_thread_flag(TIF_SIGPENDING);
157
158 }
159
160 /* Given the mask, find the first available signal that should be serviced. */
161
162 #define SYNCHRONOUS_MASK \
163 (sigmask(SIGSEGV) | sigmask(SIGBUS) | sigmask(SIGILL) | \
164 sigmask(SIGTRAP) | sigmask(SIGFPE))
165
166 int next_signal(struct sigpending *pending, sigset_t *mask)
167 {
168 unsigned long i, *s, *m, x;
169 int sig = 0;
170
171 s = pending->signal.sig;
172 m = mask->sig;
173
174 /*
175 * Handle the first word specially: it contains the
176 * synchronous signals that need to be dequeued first.
177 */
178 x = *s &~ *m;
179 if (x) {
180 if (x & SYNCHRONOUS_MASK)
181 x &= SYNCHRONOUS_MASK;
182 sig = ffz(~x) + 1;
183 return sig;
184 }
185
186 switch (_NSIG_WORDS) {
187 default:
188 for (i = 1; i < _NSIG_WORDS; ++i) {
189 x = *++s &~ *++m;
190 if (!x)
191 continue;
192 sig = ffz(~x) + i*_NSIG_BPW + 1;
193 break;
194 }
195 break;
196
197 case 2:
198 x = s[1] &~ m[1];
199 if (!x)
200 break;
201 sig = ffz(~x) + _NSIG_BPW + 1;
202 break;
203
204 case 1:
205 /* Nothing to do */
206 break;
207 }
208
209 return sig;
210 }
211
212 static inline void print_dropped_signal(int sig)
213 {
214 static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
215
216 if (!print_fatal_signals)
217 return;
218
219 if (!__ratelimit(&ratelimit_state))
220 return;
221
222 printk(KERN_INFO "%s/%d: reached RLIMIT_SIGPENDING, dropped signal %d\n",
223 current->comm, current->pid, sig);
224 }
225
226 /*
227 * allocate a new signal queue record
228 * - this may be called without locks if and only if t == current, otherwise an
229 * appopriate lock must be held to stop the target task from exiting
230 */
231 static struct sigqueue *
232 __sigqueue_alloc(int sig, struct task_struct *t, gfp_t flags, int override_rlimit)
233 {
234 struct sigqueue *q = NULL;
235 struct user_struct *user;
236
237 /*
238 * Protect access to @t credentials. This can go away when all
239 * callers hold rcu read lock.
240 */
241 rcu_read_lock();
242 user = get_uid(__task_cred(t)->user);
243 atomic_inc(&user->sigpending);
244 rcu_read_unlock();
245
246 if (override_rlimit ||
247 atomic_read(&user->sigpending) <=
248 task_rlimit(t, RLIMIT_SIGPENDING)) {
249 q = kmem_cache_alloc(sigqueue_cachep, flags);
250 } else {
251 print_dropped_signal(sig);
252 }
253
254 if (unlikely(q == NULL)) {
255 atomic_dec(&user->sigpending);
256 free_uid(user);
257 } else {
258 INIT_LIST_HEAD(&q->list);
259 q->flags = 0;
260 q->user = user;
261 }
262
263 return q;
264 }
265
266 static void __sigqueue_free(struct sigqueue *q)
267 {
268 if (q->flags & SIGQUEUE_PREALLOC)
269 return;
270 atomic_dec(&q->user->sigpending);
271 free_uid(q->user);
272 kmem_cache_free(sigqueue_cachep, q);
273 }
274
275 void flush_sigqueue(struct sigpending *queue)
276 {
277 struct sigqueue *q;
278
279 sigemptyset(&queue->signal);
280 while (!list_empty(&queue->list)) {
281 q = list_entry(queue->list.next, struct sigqueue , list);
282 list_del_init(&q->list);
283 __sigqueue_free(q);
284 }
285 }
286
287 /*
288 * Flush all pending signals for a task.
289 */
290 void __flush_signals(struct task_struct *t)
291 {
292 clear_tsk_thread_flag(t, TIF_SIGPENDING);
293 flush_sigqueue(&t->pending);
294 flush_sigqueue(&t->signal->shared_pending);
295 }
296
297 void flush_signals(struct task_struct *t)
298 {
299 unsigned long flags;
300
301 spin_lock_irqsave(&t->sighand->siglock, flags);
302 __flush_signals(t);
303 spin_unlock_irqrestore(&t->sighand->siglock, flags);
304 }
305
306 static void __flush_itimer_signals(struct sigpending *pending)
307 {
308 sigset_t signal, retain;
309 struct sigqueue *q, *n;
310
311 signal = pending->signal;
312 sigemptyset(&retain);
313
314 list_for_each_entry_safe(q, n, &pending->list, list) {
315 int sig = q->info.si_signo;
316
317 if (likely(q->info.si_code != SI_TIMER)) {
318 sigaddset(&retain, sig);
319 } else {
320 sigdelset(&signal, sig);
321 list_del_init(&q->list);
322 __sigqueue_free(q);
323 }
324 }
325
326 sigorsets(&pending->signal, &signal, &retain);
327 }
328
329 void flush_itimer_signals(void)
330 {
331 struct task_struct *tsk = current;
332 unsigned long flags;
333
334 spin_lock_irqsave(&tsk->sighand->siglock, flags);
335 __flush_itimer_signals(&tsk->pending);
336 __flush_itimer_signals(&tsk->signal->shared_pending);
337 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
338 }
339
340 void ignore_signals(struct task_struct *t)
341 {
342 int i;
343
344 for (i = 0; i < _NSIG; ++i)
345 t->sighand->action[i].sa.sa_handler = SIG_IGN;
346
347 flush_signals(t);
348 }
349
350 /*
351 * Flush all handlers for a task.
352 */
353
354 void
355 flush_signal_handlers(struct task_struct *t, int force_default)
356 {
357 int i;
358 struct k_sigaction *ka = &t->sighand->action[0];
359 for (i = _NSIG ; i != 0 ; i--) {
360 if (force_default || ka->sa.sa_handler != SIG_IGN)
361 ka->sa.sa_handler = SIG_DFL;
362 ka->sa.sa_flags = 0;
363 sigemptyset(&ka->sa.sa_mask);
364 ka++;
365 }
366 }
367
368 int unhandled_signal(struct task_struct *tsk, int sig)
369 {
370 void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
371 if (is_global_init(tsk))
372 return 1;
373 if (handler != SIG_IGN && handler != SIG_DFL)
374 return 0;
375 return !tracehook_consider_fatal_signal(tsk, sig);
376 }
377
378
379 /* Notify the system that a driver wants to block all signals for this
380 * process, and wants to be notified if any signals at all were to be
381 * sent/acted upon. If the notifier routine returns non-zero, then the
382 * signal will be acted upon after all. If the notifier routine returns 0,
383 * then then signal will be blocked. Only one block per process is
384 * allowed. priv is a pointer to private data that the notifier routine
385 * can use to determine if the signal should be blocked or not. */
386
387 void
388 block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
389 {
390 unsigned long flags;
391
392 spin_lock_irqsave(&current->sighand->siglock, flags);
393 current->notifier_mask = mask;
394 current->notifier_data = priv;
395 current->notifier = notifier;
396 spin_unlock_irqrestore(&current->sighand->siglock, flags);
397 }
398
399 /* Notify the system that blocking has ended. */
400
401 void
402 unblock_all_signals(void)
403 {
404 unsigned long flags;
405
406 spin_lock_irqsave(&current->sighand->siglock, flags);
407 current->notifier = NULL;
408 current->notifier_data = NULL;
409 recalc_sigpending();
410 spin_unlock_irqrestore(&current->sighand->siglock, flags);
411 }
412
413 static void collect_signal(int sig, struct sigpending *list, siginfo_t *info)
414 {
415 struct sigqueue *q, *first = NULL;
416
417 /*
418 * Collect the siginfo appropriate to this signal. Check if
419 * there is another siginfo for the same signal.
420 */
421 list_for_each_entry(q, &list->list, list) {
422 if (q->info.si_signo == sig) {
423 if (first)
424 goto still_pending;
425 first = q;
426 }
427 }
428
429 sigdelset(&list->signal, sig);
430
431 if (first) {
432 still_pending:
433 list_del_init(&first->list);
434 copy_siginfo(info, &first->info);
435 __sigqueue_free(first);
436 } else {
437 /* Ok, it wasn't in the queue. This must be
438 a fast-pathed signal or we must have been
439 out of queue space. So zero out the info.
440 */
441 info->si_signo = sig;
442 info->si_errno = 0;
443 info->si_code = SI_USER;
444 info->si_pid = 0;
445 info->si_uid = 0;
446 }
447 }
448
449 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
450 siginfo_t *info)
451 {
452 int sig = next_signal(pending, mask);
453
454 if (sig) {
455 if (current->notifier) {
456 if (sigismember(current->notifier_mask, sig)) {
457 if (!(current->notifier)(current->notifier_data)) {
458 clear_thread_flag(TIF_SIGPENDING);
459 return 0;
460 }
461 }
462 }
463
464 collect_signal(sig, pending, info);
465 }
466
467 return sig;
468 }
469
470 /*
471 * Dequeue a signal and return the element to the caller, which is
472 * expected to free it.
473 *
474 * All callers have to hold the siglock.
475 */
476 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
477 {
478 int signr;
479
480 /* We only dequeue private signals from ourselves, we don't let
481 * signalfd steal them
482 */
483 signr = __dequeue_signal(&tsk->pending, mask, info);
484 if (!signr) {
485 signr = __dequeue_signal(&tsk->signal->shared_pending,
486 mask, info);
487 /*
488 * itimer signal ?
489 *
490 * itimers are process shared and we restart periodic
491 * itimers in the signal delivery path to prevent DoS
492 * attacks in the high resolution timer case. This is
493 * compliant with the old way of self restarting
494 * itimers, as the SIGALRM is a legacy signal and only
495 * queued once. Changing the restart behaviour to
496 * restart the timer in the signal dequeue path is
497 * reducing the timer noise on heavy loaded !highres
498 * systems too.
499 */
500 if (unlikely(signr == SIGALRM)) {
501 struct hrtimer *tmr = &tsk->signal->real_timer;
502
503 if (!hrtimer_is_queued(tmr) &&
504 tsk->signal->it_real_incr.tv64 != 0) {
505 hrtimer_forward(tmr, tmr->base->get_time(),
506 tsk->signal->it_real_incr);
507 hrtimer_restart(tmr);
508 }
509 }
510 }
511
512 recalc_sigpending();
513 if (!signr)
514 return 0;
515
516 if (unlikely(sig_kernel_stop(signr))) {
517 /*
518 * Set a marker that we have dequeued a stop signal. Our
519 * caller might release the siglock and then the pending
520 * stop signal it is about to process is no longer in the
521 * pending bitmasks, but must still be cleared by a SIGCONT
522 * (and overruled by a SIGKILL). So those cases clear this
523 * shared flag after we've set it. Note that this flag may
524 * remain set after the signal we return is ignored or
525 * handled. That doesn't matter because its only purpose
526 * is to alert stop-signal processing code when another
527 * processor has come along and cleared the flag.
528 */
529 tsk->signal->flags |= SIGNAL_STOP_DEQUEUED;
530 }
531 if ((info->si_code & __SI_MASK) == __SI_TIMER && info->si_sys_private) {
532 /*
533 * Release the siglock to ensure proper locking order
534 * of timer locks outside of siglocks. Note, we leave
535 * irqs disabled here, since the posix-timers code is
536 * about to disable them again anyway.
537 */
538 spin_unlock(&tsk->sighand->siglock);
539 do_schedule_next_timer(info);
540 spin_lock(&tsk->sighand->siglock);
541 }
542 return signr;
543 }
544
545 /*
546 * Tell a process that it has a new active signal..
547 *
548 * NOTE! we rely on the previous spin_lock to
549 * lock interrupts for us! We can only be called with
550 * "siglock" held, and the local interrupt must
551 * have been disabled when that got acquired!
552 *
553 * No need to set need_resched since signal event passing
554 * goes through ->blocked
555 */
556 void signal_wake_up(struct task_struct *t, int resume)
557 {
558 unsigned int mask;
559
560 set_tsk_thread_flag(t, TIF_SIGPENDING);
561
562 /*
563 * For SIGKILL, we want to wake it up in the stopped/traced/killable
564 * case. We don't check t->state here because there is a race with it
565 * executing another processor and just now entering stopped state.
566 * By using wake_up_state, we ensure the process will wake up and
567 * handle its death signal.
568 */
569 mask = TASK_INTERRUPTIBLE;
570 if (resume)
571 mask |= TASK_WAKEKILL;
572 if (!wake_up_state(t, mask))
573 kick_process(t);
574 }
575
576 /*
577 * Remove signals in mask from the pending set and queue.
578 * Returns 1 if any signals were found.
579 *
580 * All callers must be holding the siglock.
581 *
582 * This version takes a sigset mask and looks at all signals,
583 * not just those in the first mask word.
584 */
585 static int rm_from_queue_full(sigset_t *mask, struct sigpending *s)
586 {
587 struct sigqueue *q, *n;
588 sigset_t m;
589
590 sigandsets(&m, mask, &s->signal);
591 if (sigisemptyset(&m))
592 return 0;
593
594 signandsets(&s->signal, &s->signal, mask);
595 list_for_each_entry_safe(q, n, &s->list, list) {
596 if (sigismember(mask, q->info.si_signo)) {
597 list_del_init(&q->list);
598 __sigqueue_free(q);
599 }
600 }
601 return 1;
602 }
603 /*
604 * Remove signals in mask from the pending set and queue.
605 * Returns 1 if any signals were found.
606 *
607 * All callers must be holding the siglock.
608 */
609 static int rm_from_queue(unsigned long mask, struct sigpending *s)
610 {
611 struct sigqueue *q, *n;
612
613 if (!sigtestsetmask(&s->signal, mask))
614 return 0;
615
616 sigdelsetmask(&s->signal, mask);
617 list_for_each_entry_safe(q, n, &s->list, list) {
618 if (q->info.si_signo < SIGRTMIN &&
619 (mask & sigmask(q->info.si_signo))) {
620 list_del_init(&q->list);
621 __sigqueue_free(q);
622 }
623 }
624 return 1;
625 }
626
627 static inline int is_si_special(const struct siginfo *info)
628 {
629 return info <= SEND_SIG_FORCED;
630 }
631
632 static inline bool si_fromuser(const struct siginfo *info)
633 {
634 return info == SEND_SIG_NOINFO ||
635 (!is_si_special(info) && SI_FROMUSER(info));
636 }
637
638 /*
639 * called with RCU read lock from check_kill_permission()
640 */
641 static int kill_ok_by_cred(struct task_struct *t)
642 {
643 const struct cred *cred = current_cred();
644 const struct cred *tcred = __task_cred(t);
645
646 if (cred->user->user_ns == tcred->user->user_ns &&
647 (cred->euid == tcred->suid ||
648 cred->euid == tcred->uid ||
649 cred->uid == tcred->suid ||
650 cred->uid == tcred->uid))
651 return 1;
652
653 if (ns_capable(tcred->user->user_ns, CAP_KILL))
654 return 1;
655
656 return 0;
657 }
658
659 /*
660 * Bad permissions for sending the signal
661 * - the caller must hold the RCU read lock
662 */
663 static int check_kill_permission(int sig, struct siginfo *info,
664 struct task_struct *t)
665 {
666 struct pid *sid;
667 int error;
668
669 if (!valid_signal(sig))
670 return -EINVAL;
671
672 if (!si_fromuser(info))
673 return 0;
674
675 error = audit_signal_info(sig, t); /* Let audit system see the signal */
676 if (error)
677 return error;
678
679 if (!same_thread_group(current, t) &&
680 !kill_ok_by_cred(t)) {
681 switch (sig) {
682 case SIGCONT:
683 sid = task_session(t);
684 /*
685 * We don't return the error if sid == NULL. The
686 * task was unhashed, the caller must notice this.
687 */
688 if (!sid || sid == task_session(current))
689 break;
690 default:
691 return -EPERM;
692 }
693 }
694
695 return security_task_kill(t, info, sig, 0);
696 }
697
698 /*
699 * Handle magic process-wide effects of stop/continue signals. Unlike
700 * the signal actions, these happen immediately at signal-generation
701 * time regardless of blocking, ignoring, or handling. This does the
702 * actual continuing for SIGCONT, but not the actual stopping for stop
703 * signals. The process stop is done as a signal action for SIG_DFL.
704 *
705 * Returns true if the signal should be actually delivered, otherwise
706 * it should be dropped.
707 */
708 static int prepare_signal(int sig, struct task_struct *p, int from_ancestor_ns)
709 {
710 struct signal_struct *signal = p->signal;
711 struct task_struct *t;
712
713 if (unlikely(signal->flags & SIGNAL_GROUP_EXIT)) {
714 /*
715 * The process is in the middle of dying, nothing to do.
716 */
717 } else if (sig_kernel_stop(sig)) {
718 /*
719 * This is a stop signal. Remove SIGCONT from all queues.
720 */
721 rm_from_queue(sigmask(SIGCONT), &signal->shared_pending);
722 t = p;
723 do {
724 rm_from_queue(sigmask(SIGCONT), &t->pending);
725 } while_each_thread(p, t);
726 } else if (sig == SIGCONT) {
727 unsigned int why;
728 /*
729 * Remove all stop signals from all queues,
730 * and wake all threads.
731 */
732 rm_from_queue(SIG_KERNEL_STOP_MASK, &signal->shared_pending);
733 t = p;
734 do {
735 unsigned int state;
736 rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
737 /*
738 * If there is a handler for SIGCONT, we must make
739 * sure that no thread returns to user mode before
740 * we post the signal, in case it was the only
741 * thread eligible to run the signal handler--then
742 * it must not do anything between resuming and
743 * running the handler. With the TIF_SIGPENDING
744 * flag set, the thread will pause and acquire the
745 * siglock that we hold now and until we've queued
746 * the pending signal.
747 *
748 * Wake up the stopped thread _after_ setting
749 * TIF_SIGPENDING
750 */
751 state = __TASK_STOPPED;
752 if (sig_user_defined(t, SIGCONT) && !sigismember(&t->blocked, SIGCONT)) {
753 set_tsk_thread_flag(t, TIF_SIGPENDING);
754 state |= TASK_INTERRUPTIBLE;
755 }
756 wake_up_state(t, state);
757 } while_each_thread(p, t);
758
759 /*
760 * Notify the parent with CLD_CONTINUED if we were stopped.
761 *
762 * If we were in the middle of a group stop, we pretend it
763 * was already finished, and then continued. Since SIGCHLD
764 * doesn't queue we report only CLD_STOPPED, as if the next
765 * CLD_CONTINUED was dropped.
766 */
767 why = 0;
768 if (signal->flags & SIGNAL_STOP_STOPPED)
769 why |= SIGNAL_CLD_CONTINUED;
770 else if (signal->group_stop_count)
771 why |= SIGNAL_CLD_STOPPED;
772
773 if (why) {
774 /*
775 * The first thread which returns from do_signal_stop()
776 * will take ->siglock, notice SIGNAL_CLD_MASK, and
777 * notify its parent. See get_signal_to_deliver().
778 */
779 signal->flags = why | SIGNAL_STOP_CONTINUED;
780 signal->group_stop_count = 0;
781 signal->group_exit_code = 0;
782 } else {
783 /*
784 * We are not stopped, but there could be a stop
785 * signal in the middle of being processed after
786 * being removed from the queue. Clear that too.
787 */
788 signal->flags &= ~SIGNAL_STOP_DEQUEUED;
789 }
790 }
791
792 return !sig_ignored(p, sig, from_ancestor_ns);
793 }
794
795 /*
796 * Test if P wants to take SIG. After we've checked all threads with this,
797 * it's equivalent to finding no threads not blocking SIG. Any threads not
798 * blocking SIG were ruled out because they are not running and already
799 * have pending signals. Such threads will dequeue from the shared queue
800 * as soon as they're available, so putting the signal on the shared queue
801 * will be equivalent to sending it to one such thread.
802 */
803 static inline int wants_signal(int sig, struct task_struct *p)
804 {
805 if (sigismember(&p->blocked, sig))
806 return 0;
807 if (p->flags & PF_EXITING)
808 return 0;
809 if (sig == SIGKILL)
810 return 1;
811 if (task_is_stopped_or_traced(p))
812 return 0;
813 return task_curr(p) || !signal_pending(p);
814 }
815
816 static void complete_signal(int sig, struct task_struct *p, int group)
817 {
818 struct signal_struct *signal = p->signal;
819 struct task_struct *t;
820
821 /*
822 * Now find a thread we can wake up to take the signal off the queue.
823 *
824 * If the main thread wants the signal, it gets first crack.
825 * Probably the least surprising to the average bear.
826 */
827 if (wants_signal(sig, p))
828 t = p;
829 else if (!group || thread_group_empty(p))
830 /*
831 * There is just one thread and it does not need to be woken.
832 * It will dequeue unblocked signals before it runs again.
833 */
834 return;
835 else {
836 /*
837 * Otherwise try to find a suitable thread.
838 */
839 t = signal->curr_target;
840 while (!wants_signal(sig, t)) {
841 t = next_thread(t);
842 if (t == signal->curr_target)
843 /*
844 * No thread needs to be woken.
845 * Any eligible threads will see
846 * the signal in the queue soon.
847 */
848 return;
849 }
850 signal->curr_target = t;
851 }
852
853 /*
854 * Found a killable thread. If the signal will be fatal,
855 * then start taking the whole group down immediately.
856 */
857 if (sig_fatal(p, sig) &&
858 !(signal->flags & (SIGNAL_UNKILLABLE | SIGNAL_GROUP_EXIT)) &&
859 !sigismember(&t->real_blocked, sig) &&
860 (sig == SIGKILL ||
861 !tracehook_consider_fatal_signal(t, sig))) {
862 /*
863 * This signal will be fatal to the whole group.
864 */
865 if (!sig_kernel_coredump(sig)) {
866 /*
867 * Start a group exit and wake everybody up.
868 * This way we don't have other threads
869 * running and doing things after a slower
870 * thread has the fatal signal pending.
871 */
872 signal->flags = SIGNAL_GROUP_EXIT;
873 signal->group_exit_code = sig;
874 signal->group_stop_count = 0;
875 t = p;
876 do {
877 sigaddset(&t->pending.signal, SIGKILL);
878 signal_wake_up(t, 1);
879 } while_each_thread(p, t);
880 return;
881 }
882 }
883
884 /*
885 * The signal is already in the shared-pending queue.
886 * Tell the chosen thread to wake up and dequeue it.
887 */
888 signal_wake_up(t, sig == SIGKILL);
889 return;
890 }
891
892 static inline int legacy_queue(struct sigpending *signals, int sig)
893 {
894 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
895 }
896
897 static int __send_signal(int sig, struct siginfo *info, struct task_struct *t,
898 int group, int from_ancestor_ns)
899 {
900 struct sigpending *pending;
901 struct sigqueue *q;
902 int override_rlimit;
903
904 trace_signal_generate(sig, info, t);
905
906 assert_spin_locked(&t->sighand->siglock);
907
908 if (!prepare_signal(sig, t, from_ancestor_ns))
909 return 0;
910
911 pending = group ? &t->signal->shared_pending : &t->pending;
912 /*
913 * Short-circuit ignored signals and support queuing
914 * exactly one non-rt signal, so that we can get more
915 * detailed information about the cause of the signal.
916 */
917 if (legacy_queue(pending, sig))
918 return 0;
919 /*
920 * fast-pathed signals for kernel-internal things like SIGSTOP
921 * or SIGKILL.
922 */
923 if (info == SEND_SIG_FORCED)
924 goto out_set;
925
926 /* Real-time signals must be queued if sent by sigqueue, or
927 some other real-time mechanism. It is implementation
928 defined whether kill() does so. We attempt to do so, on
929 the principle of least surprise, but since kill is not
930 allowed to fail with EAGAIN when low on memory we just
931 make sure at least one signal gets delivered and don't
932 pass on the info struct. */
933
934 if (sig < SIGRTMIN)
935 override_rlimit = (is_si_special(info) || info->si_code >= 0);
936 else
937 override_rlimit = 0;
938
939 q = __sigqueue_alloc(sig, t, GFP_ATOMIC | __GFP_NOTRACK_FALSE_POSITIVE,
940 override_rlimit);
941 if (q) {
942 list_add_tail(&q->list, &pending->list);
943 switch ((unsigned long) info) {
944 case (unsigned long) SEND_SIG_NOINFO:
945 q->info.si_signo = sig;
946 q->info.si_errno = 0;
947 q->info.si_code = SI_USER;
948 q->info.si_pid = task_tgid_nr_ns(current,
949 task_active_pid_ns(t));
950 q->info.si_uid = current_uid();
951 break;
952 case (unsigned long) SEND_SIG_PRIV:
953 q->info.si_signo = sig;
954 q->info.si_errno = 0;
955 q->info.si_code = SI_KERNEL;
956 q->info.si_pid = 0;
957 q->info.si_uid = 0;
958 break;
959 default:
960 copy_siginfo(&q->info, info);
961 if (from_ancestor_ns)
962 q->info.si_pid = 0;
963 break;
964 }
965 } else if (!is_si_special(info)) {
966 if (sig >= SIGRTMIN && info->si_code != SI_USER) {
967 /*
968 * Queue overflow, abort. We may abort if the
969 * signal was rt and sent by user using something
970 * other than kill().
971 */
972 trace_signal_overflow_fail(sig, group, info);
973 return -EAGAIN;
974 } else {
975 /*
976 * This is a silent loss of information. We still
977 * send the signal, but the *info bits are lost.
978 */
979 trace_signal_lose_info(sig, group, info);
980 }
981 }
982
983 out_set:
984 signalfd_notify(t, sig);
985 sigaddset(&pending->signal, sig);
986 complete_signal(sig, t, group);
987 return 0;
988 }
989
990 static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
991 int group)
992 {
993 int from_ancestor_ns = 0;
994
995 #ifdef CONFIG_PID_NS
996 from_ancestor_ns = si_fromuser(info) &&
997 !task_pid_nr_ns(current, task_active_pid_ns(t));
998 #endif
999
1000 return __send_signal(sig, info, t, group, from_ancestor_ns);
1001 }
1002
1003 static void print_fatal_signal(struct pt_regs *regs, int signr)
1004 {
1005 printk("%s/%d: potentially unexpected fatal signal %d.\n",
1006 current->comm, task_pid_nr(current), signr);
1007
1008 #if defined(__i386__) && !defined(__arch_um__)
1009 printk("code at %08lx: ", regs->ip);
1010 {
1011 int i;
1012 for (i = 0; i < 16; i++) {
1013 unsigned char insn;
1014
1015 if (get_user(insn, (unsigned char *)(regs->ip + i)))
1016 break;
1017 printk("%02x ", insn);
1018 }
1019 }
1020 #endif
1021 printk("\n");
1022 preempt_disable();
1023 show_regs(regs);
1024 preempt_enable();
1025 }
1026
1027 static int __init setup_print_fatal_signals(char *str)
1028 {
1029 get_option (&str, &print_fatal_signals);
1030
1031 return 1;
1032 }
1033
1034 __setup("print-fatal-signals=", setup_print_fatal_signals);
1035
1036 int
1037 __group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1038 {
1039 return send_signal(sig, info, p, 1);
1040 }
1041
1042 static int
1043 specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
1044 {
1045 return send_signal(sig, info, t, 0);
1046 }
1047
1048 int do_send_sig_info(int sig, struct siginfo *info, struct task_struct *p,
1049 bool group)
1050 {
1051 unsigned long flags;
1052 int ret = -ESRCH;
1053
1054 if (lock_task_sighand(p, &flags)) {
1055 ret = send_signal(sig, info, p, group);
1056 unlock_task_sighand(p, &flags);
1057 }
1058
1059 return ret;
1060 }
1061
1062 /*
1063 * Force a signal that the process can't ignore: if necessary
1064 * we unblock the signal and change any SIG_IGN to SIG_DFL.
1065 *
1066 * Note: If we unblock the signal, we always reset it to SIG_DFL,
1067 * since we do not want to have a signal handler that was blocked
1068 * be invoked when user space had explicitly blocked it.
1069 *
1070 * We don't want to have recursive SIGSEGV's etc, for example,
1071 * that is why we also clear SIGNAL_UNKILLABLE.
1072 */
1073 int
1074 force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
1075 {
1076 unsigned long int flags;
1077 int ret, blocked, ignored;
1078 struct k_sigaction *action;
1079
1080 spin_lock_irqsave(&t->sighand->siglock, flags);
1081 action = &t->sighand->action[sig-1];
1082 ignored = action->sa.sa_handler == SIG_IGN;
1083 blocked = sigismember(&t->blocked, sig);
1084 if (blocked || ignored) {
1085 action->sa.sa_handler = SIG_DFL;
1086 if (blocked) {
1087 sigdelset(&t->blocked, sig);
1088 recalc_sigpending_and_wake(t);
1089 }
1090 }
1091 if (action->sa.sa_handler == SIG_DFL)
1092 t->signal->flags &= ~SIGNAL_UNKILLABLE;
1093 ret = specific_send_sig_info(sig, info, t);
1094 spin_unlock_irqrestore(&t->sighand->siglock, flags);
1095
1096 return ret;
1097 }
1098
1099 /*
1100 * Nuke all other threads in the group.
1101 */
1102 int zap_other_threads(struct task_struct *p)
1103 {
1104 struct task_struct *t = p;
1105 int count = 0;
1106
1107 p->signal->group_stop_count = 0;
1108
1109 while_each_thread(p, t) {
1110 count++;
1111
1112 /* Don't bother with already dead threads */
1113 if (t->exit_state)
1114 continue;
1115 sigaddset(&t->pending.signal, SIGKILL);
1116 signal_wake_up(t, 1);
1117 }
1118
1119 return count;
1120 }
1121
1122 struct sighand_struct *__lock_task_sighand(struct task_struct *tsk,
1123 unsigned long *flags)
1124 {
1125 struct sighand_struct *sighand;
1126
1127 rcu_read_lock();
1128 for (;;) {
1129 sighand = rcu_dereference(tsk->sighand);
1130 if (unlikely(sighand == NULL))
1131 break;
1132
1133 spin_lock_irqsave(&sighand->siglock, *flags);
1134 if (likely(sighand == tsk->sighand))
1135 break;
1136 spin_unlock_irqrestore(&sighand->siglock, *flags);
1137 }
1138 rcu_read_unlock();
1139
1140 return sighand;
1141 }
1142
1143 /*
1144 * send signal info to all the members of a group
1145 */
1146 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1147 {
1148 int ret;
1149
1150 rcu_read_lock();
1151 ret = check_kill_permission(sig, info, p);
1152 rcu_read_unlock();
1153
1154 if (!ret && sig)
1155 ret = do_send_sig_info(sig, info, p, true);
1156
1157 return ret;
1158 }
1159
1160 /*
1161 * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1162 * control characters do (^C, ^Z etc)
1163 * - the caller must hold at least a readlock on tasklist_lock
1164 */
1165 int __kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
1166 {
1167 struct task_struct *p = NULL;
1168 int retval, success;
1169
1170 success = 0;
1171 retval = -ESRCH;
1172 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1173 int err = group_send_sig_info(sig, info, p);
1174 success |= !err;
1175 retval = err;
1176 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1177 return success ? 0 : retval;
1178 }
1179
1180 int kill_pid_info(int sig, struct siginfo *info, struct pid *pid)
1181 {
1182 int error = -ESRCH;
1183 struct task_struct *p;
1184
1185 rcu_read_lock();
1186 retry:
1187 p = pid_task(pid, PIDTYPE_PID);
1188 if (p) {
1189 error = group_send_sig_info(sig, info, p);
1190 if (unlikely(error == -ESRCH))
1191 /*
1192 * The task was unhashed in between, try again.
1193 * If it is dead, pid_task() will return NULL,
1194 * if we race with de_thread() it will find the
1195 * new leader.
1196 */
1197 goto retry;
1198 }
1199 rcu_read_unlock();
1200
1201 return error;
1202 }
1203
1204 int
1205 kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1206 {
1207 int error;
1208 rcu_read_lock();
1209 error = kill_pid_info(sig, info, find_vpid(pid));
1210 rcu_read_unlock();
1211 return error;
1212 }
1213
1214 /* like kill_pid_info(), but doesn't use uid/euid of "current" */
1215 int kill_pid_info_as_uid(int sig, struct siginfo *info, struct pid *pid,
1216 uid_t uid, uid_t euid, u32 secid)
1217 {
1218 int ret = -EINVAL;
1219 struct task_struct *p;
1220 const struct cred *pcred;
1221 unsigned long flags;
1222
1223 if (!valid_signal(sig))
1224 return ret;
1225
1226 rcu_read_lock();
1227 p = pid_task(pid, PIDTYPE_PID);
1228 if (!p) {
1229 ret = -ESRCH;
1230 goto out_unlock;
1231 }
1232 pcred = __task_cred(p);
1233 if (si_fromuser(info) &&
1234 euid != pcred->suid && euid != pcred->uid &&
1235 uid != pcred->suid && uid != pcred->uid) {
1236 ret = -EPERM;
1237 goto out_unlock;
1238 }
1239 ret = security_task_kill(p, info, sig, secid);
1240 if (ret)
1241 goto out_unlock;
1242
1243 if (sig) {
1244 if (lock_task_sighand(p, &flags)) {
1245 ret = __send_signal(sig, info, p, 1, 0);
1246 unlock_task_sighand(p, &flags);
1247 } else
1248 ret = -ESRCH;
1249 }
1250 out_unlock:
1251 rcu_read_unlock();
1252 return ret;
1253 }
1254 EXPORT_SYMBOL_GPL(kill_pid_info_as_uid);
1255
1256 /*
1257 * kill_something_info() interprets pid in interesting ways just like kill(2).
1258 *
1259 * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1260 * is probably wrong. Should make it like BSD or SYSV.
1261 */
1262
1263 static int kill_something_info(int sig, struct siginfo *info, pid_t pid)
1264 {
1265 int ret;
1266
1267 if (pid > 0) {
1268 rcu_read_lock();
1269 ret = kill_pid_info(sig, info, find_vpid(pid));
1270 rcu_read_unlock();
1271 return ret;
1272 }
1273
1274 read_lock(&tasklist_lock);
1275 if (pid != -1) {
1276 ret = __kill_pgrp_info(sig, info,
1277 pid ? find_vpid(-pid) : task_pgrp(current));
1278 } else {
1279 int retval = 0, count = 0;
1280 struct task_struct * p;
1281
1282 for_each_process(p) {
1283 if (task_pid_vnr(p) > 1 &&
1284 !same_thread_group(p, current)) {
1285 int err = group_send_sig_info(sig, info, p);
1286 ++count;
1287 if (err != -EPERM)
1288 retval = err;
1289 }
1290 }
1291 ret = count ? retval : -ESRCH;
1292 }
1293 read_unlock(&tasklist_lock);
1294
1295 return ret;
1296 }
1297
1298 /*
1299 * These are for backward compatibility with the rest of the kernel source.
1300 */
1301
1302 int
1303 send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1304 {
1305 /*
1306 * Make sure legacy kernel users don't send in bad values
1307 * (normal paths check this in check_kill_permission).
1308 */
1309 if (!valid_signal(sig))
1310 return -EINVAL;
1311
1312 return do_send_sig_info(sig, info, p, false);
1313 }
1314
1315 #define __si_special(priv) \
1316 ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1317
1318 int
1319 send_sig(int sig, struct task_struct *p, int priv)
1320 {
1321 return send_sig_info(sig, __si_special(priv), p);
1322 }
1323
1324 void
1325 force_sig(int sig, struct task_struct *p)
1326 {
1327 force_sig_info(sig, SEND_SIG_PRIV, p);
1328 }
1329
1330 /*
1331 * When things go south during signal handling, we
1332 * will force a SIGSEGV. And if the signal that caused
1333 * the problem was already a SIGSEGV, we'll want to
1334 * make sure we don't even try to deliver the signal..
1335 */
1336 int
1337 force_sigsegv(int sig, struct task_struct *p)
1338 {
1339 if (sig == SIGSEGV) {
1340 unsigned long flags;
1341 spin_lock_irqsave(&p->sighand->siglock, flags);
1342 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1343 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1344 }
1345 force_sig(SIGSEGV, p);
1346 return 0;
1347 }
1348
1349 int kill_pgrp(struct pid *pid, int sig, int priv)
1350 {
1351 int ret;
1352
1353 read_lock(&tasklist_lock);
1354 ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1355 read_unlock(&tasklist_lock);
1356
1357 return ret;
1358 }
1359 EXPORT_SYMBOL(kill_pgrp);
1360
1361 int kill_pid(struct pid *pid, int sig, int priv)
1362 {
1363 return kill_pid_info(sig, __si_special(priv), pid);
1364 }
1365 EXPORT_SYMBOL(kill_pid);
1366
1367 /*
1368 * These functions support sending signals using preallocated sigqueue
1369 * structures. This is needed "because realtime applications cannot
1370 * afford to lose notifications of asynchronous events, like timer
1371 * expirations or I/O completions". In the case of Posix Timers
1372 * we allocate the sigqueue structure from the timer_create. If this
1373 * allocation fails we are able to report the failure to the application
1374 * with an EAGAIN error.
1375 */
1376 struct sigqueue *sigqueue_alloc(void)
1377 {
1378 struct sigqueue *q = __sigqueue_alloc(-1, current, GFP_KERNEL, 0);
1379
1380 if (q)
1381 q->flags |= SIGQUEUE_PREALLOC;
1382
1383 return q;
1384 }
1385
1386 void sigqueue_free(struct sigqueue *q)
1387 {
1388 unsigned long flags;
1389 spinlock_t *lock = &current->sighand->siglock;
1390
1391 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1392 /*
1393 * We must hold ->siglock while testing q->list
1394 * to serialize with collect_signal() or with
1395 * __exit_signal()->flush_sigqueue().
1396 */
1397 spin_lock_irqsave(lock, flags);
1398 q->flags &= ~SIGQUEUE_PREALLOC;
1399 /*
1400 * If it is queued it will be freed when dequeued,
1401 * like the "regular" sigqueue.
1402 */
1403 if (!list_empty(&q->list))
1404 q = NULL;
1405 spin_unlock_irqrestore(lock, flags);
1406
1407 if (q)
1408 __sigqueue_free(q);
1409 }
1410
1411 int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group)
1412 {
1413 int sig = q->info.si_signo;
1414 struct sigpending *pending;
1415 unsigned long flags;
1416 int ret;
1417
1418 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1419
1420 ret = -1;
1421 if (!likely(lock_task_sighand(t, &flags)))
1422 goto ret;
1423
1424 ret = 1; /* the signal is ignored */
1425 if (!prepare_signal(sig, t, 0))
1426 goto out;
1427
1428 ret = 0;
1429 if (unlikely(!list_empty(&q->list))) {
1430 /*
1431 * If an SI_TIMER entry is already queue just increment
1432 * the overrun count.
1433 */
1434 BUG_ON(q->info.si_code != SI_TIMER);
1435 q->info.si_overrun++;
1436 goto out;
1437 }
1438 q->info.si_overrun = 0;
1439
1440 signalfd_notify(t, sig);
1441 pending = group ? &t->signal->shared_pending : &t->pending;
1442 list_add_tail(&q->list, &pending->list);
1443 sigaddset(&pending->signal, sig);
1444 complete_signal(sig, t, group);
1445 out:
1446 unlock_task_sighand(t, &flags);
1447 ret:
1448 return ret;
1449 }
1450
1451 /*
1452 * Let a parent know about the death of a child.
1453 * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1454 *
1455 * Returns -1 if our parent ignored us and so we've switched to
1456 * self-reaping, or else @sig.
1457 */
1458 int do_notify_parent(struct task_struct *tsk, int sig)
1459 {
1460 struct siginfo info;
1461 unsigned long flags;
1462 struct sighand_struct *psig;
1463 int ret = sig;
1464
1465 BUG_ON(sig == -1);
1466
1467 /* do_notify_parent_cldstop should have been called instead. */
1468 BUG_ON(task_is_stopped_or_traced(tsk));
1469
1470 BUG_ON(!task_ptrace(tsk) &&
1471 (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1472
1473 info.si_signo = sig;
1474 info.si_errno = 0;
1475 /*
1476 * we are under tasklist_lock here so our parent is tied to
1477 * us and cannot exit and release its namespace.
1478 *
1479 * the only it can is to switch its nsproxy with sys_unshare,
1480 * bu uncharing pid namespaces is not allowed, so we'll always
1481 * see relevant namespace
1482 *
1483 * write_lock() currently calls preempt_disable() which is the
1484 * same as rcu_read_lock(), but according to Oleg, this is not
1485 * correct to rely on this
1486 */
1487 rcu_read_lock();
1488 info.si_pid = task_pid_nr_ns(tsk, tsk->parent->nsproxy->pid_ns);
1489 info.si_uid = __task_cred(tsk)->uid;
1490 rcu_read_unlock();
1491
1492 info.si_utime = cputime_to_clock_t(cputime_add(tsk->utime,
1493 tsk->signal->utime));
1494 info.si_stime = cputime_to_clock_t(cputime_add(tsk->stime,
1495 tsk->signal->stime));
1496
1497 info.si_status = tsk->exit_code & 0x7f;
1498 if (tsk->exit_code & 0x80)
1499 info.si_code = CLD_DUMPED;
1500 else if (tsk->exit_code & 0x7f)
1501 info.si_code = CLD_KILLED;
1502 else {
1503 info.si_code = CLD_EXITED;
1504 info.si_status = tsk->exit_code >> 8;
1505 }
1506
1507 psig = tsk->parent->sighand;
1508 spin_lock_irqsave(&psig->siglock, flags);
1509 if (!task_ptrace(tsk) && sig == SIGCHLD &&
1510 (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1511 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1512 /*
1513 * We are exiting and our parent doesn't care. POSIX.1
1514 * defines special semantics for setting SIGCHLD to SIG_IGN
1515 * or setting the SA_NOCLDWAIT flag: we should be reaped
1516 * automatically and not left for our parent's wait4 call.
1517 * Rather than having the parent do it as a magic kind of
1518 * signal handler, we just set this to tell do_exit that we
1519 * can be cleaned up without becoming a zombie. Note that
1520 * we still call __wake_up_parent in this case, because a
1521 * blocked sys_wait4 might now return -ECHILD.
1522 *
1523 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1524 * is implementation-defined: we do (if you don't want
1525 * it, just use SIG_IGN instead).
1526 */
1527 ret = tsk->exit_signal = -1;
1528 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1529 sig = -1;
1530 }
1531 if (valid_signal(sig) && sig > 0)
1532 __group_send_sig_info(sig, &info, tsk->parent);
1533 __wake_up_parent(tsk, tsk->parent);
1534 spin_unlock_irqrestore(&psig->siglock, flags);
1535
1536 return ret;
1537 }
1538
1539 static void do_notify_parent_cldstop(struct task_struct *tsk, int why)
1540 {
1541 struct siginfo info;
1542 unsigned long flags;
1543 struct task_struct *parent;
1544 struct sighand_struct *sighand;
1545
1546 if (task_ptrace(tsk))
1547 parent = tsk->parent;
1548 else {
1549 tsk = tsk->group_leader;
1550 parent = tsk->real_parent;
1551 }
1552
1553 info.si_signo = SIGCHLD;
1554 info.si_errno = 0;
1555 /*
1556 * see comment in do_notify_parent() abot the following 3 lines
1557 */
1558 rcu_read_lock();
1559 info.si_pid = task_pid_nr_ns(tsk, parent->nsproxy->pid_ns);
1560 info.si_uid = __task_cred(tsk)->uid;
1561 rcu_read_unlock();
1562
1563 info.si_utime = cputime_to_clock_t(tsk->utime);
1564 info.si_stime = cputime_to_clock_t(tsk->stime);
1565
1566 info.si_code = why;
1567 switch (why) {
1568 case CLD_CONTINUED:
1569 info.si_status = SIGCONT;
1570 break;
1571 case CLD_STOPPED:
1572 info.si_status = tsk->signal->group_exit_code & 0x7f;
1573 break;
1574 case CLD_TRAPPED:
1575 info.si_status = tsk->exit_code & 0x7f;
1576 break;
1577 default:
1578 BUG();
1579 }
1580
1581 sighand = parent->sighand;
1582 spin_lock_irqsave(&sighand->siglock, flags);
1583 if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1584 !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1585 __group_send_sig_info(SIGCHLD, &info, parent);
1586 /*
1587 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1588 */
1589 __wake_up_parent(tsk, parent);
1590 spin_unlock_irqrestore(&sighand->siglock, flags);
1591 }
1592
1593 static inline int may_ptrace_stop(void)
1594 {
1595 if (!likely(task_ptrace(current)))
1596 return 0;
1597 /*
1598 * Are we in the middle of do_coredump?
1599 * If so and our tracer is also part of the coredump stopping
1600 * is a deadlock situation, and pointless because our tracer
1601 * is dead so don't allow us to stop.
1602 * If SIGKILL was already sent before the caller unlocked
1603 * ->siglock we must see ->core_state != NULL. Otherwise it
1604 * is safe to enter schedule().
1605 */
1606 if (unlikely(current->mm->core_state) &&
1607 unlikely(current->mm == current->parent->mm))
1608 return 0;
1609
1610 return 1;
1611 }
1612
1613 /*
1614 * Return nonzero if there is a SIGKILL that should be waking us up.
1615 * Called with the siglock held.
1616 */
1617 static int sigkill_pending(struct task_struct *tsk)
1618 {
1619 return sigismember(&tsk->pending.signal, SIGKILL) ||
1620 sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
1621 }
1622
1623 /*
1624 * This must be called with current->sighand->siglock held.
1625 *
1626 * This should be the path for all ptrace stops.
1627 * We always set current->last_siginfo while stopped here.
1628 * That makes it a way to test a stopped process for
1629 * being ptrace-stopped vs being job-control-stopped.
1630 *
1631 * If we actually decide not to stop at all because the tracer
1632 * is gone, we keep current->exit_code unless clear_code.
1633 */
1634 static void ptrace_stop(int exit_code, int clear_code, siginfo_t *info)
1635 __releases(&current->sighand->siglock)
1636 __acquires(&current->sighand->siglock)
1637 {
1638 if (arch_ptrace_stop_needed(exit_code, info)) {
1639 /*
1640 * The arch code has something special to do before a
1641 * ptrace stop. This is allowed to block, e.g. for faults
1642 * on user stack pages. We can't keep the siglock while
1643 * calling arch_ptrace_stop, so we must release it now.
1644 * To preserve proper semantics, we must do this before
1645 * any signal bookkeeping like checking group_stop_count.
1646 * Meanwhile, a SIGKILL could come in before we retake the
1647 * siglock. That must prevent us from sleeping in TASK_TRACED.
1648 * So after regaining the lock, we must check for SIGKILL.
1649 */
1650 spin_unlock_irq(&current->sighand->siglock);
1651 arch_ptrace_stop(exit_code, info);
1652 spin_lock_irq(&current->sighand->siglock);
1653 if (sigkill_pending(current))
1654 return;
1655 }
1656
1657 /*
1658 * If there is a group stop in progress,
1659 * we must participate in the bookkeeping.
1660 */
1661 if (current->signal->group_stop_count > 0)
1662 --current->signal->group_stop_count;
1663
1664 current->last_siginfo = info;
1665 current->exit_code = exit_code;
1666
1667 /* Let the debugger run. */
1668 __set_current_state(TASK_TRACED);
1669 spin_unlock_irq(&current->sighand->siglock);
1670 read_lock(&tasklist_lock);
1671 if (may_ptrace_stop()) {
1672 do_notify_parent_cldstop(current, CLD_TRAPPED);
1673 /*
1674 * Don't want to allow preemption here, because
1675 * sys_ptrace() needs this task to be inactive.
1676 *
1677 * XXX: implement read_unlock_no_resched().
1678 */
1679 preempt_disable();
1680 read_unlock(&tasklist_lock);
1681 preempt_enable_no_resched();
1682 schedule();
1683 } else {
1684 /*
1685 * By the time we got the lock, our tracer went away.
1686 * Don't drop the lock yet, another tracer may come.
1687 */
1688 __set_current_state(TASK_RUNNING);
1689 if (clear_code)
1690 current->exit_code = 0;
1691 read_unlock(&tasklist_lock);
1692 }
1693
1694 /*
1695 * While in TASK_TRACED, we were considered "frozen enough".
1696 * Now that we woke up, it's crucial if we're supposed to be
1697 * frozen that we freeze now before running anything substantial.
1698 */
1699 try_to_freeze();
1700
1701 /*
1702 * We are back. Now reacquire the siglock before touching
1703 * last_siginfo, so that we are sure to have synchronized with
1704 * any signal-sending on another CPU that wants to examine it.
1705 */
1706 spin_lock_irq(&current->sighand->siglock);
1707 current->last_siginfo = NULL;
1708
1709 /*
1710 * Queued signals ignored us while we were stopped for tracing.
1711 * So check for any that we should take before resuming user mode.
1712 * This sets TIF_SIGPENDING, but never clears it.
1713 */
1714 recalc_sigpending_tsk(current);
1715 }
1716
1717 void ptrace_notify(int exit_code)
1718 {
1719 siginfo_t info;
1720
1721 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1722
1723 memset(&info, 0, sizeof info);
1724 info.si_signo = SIGTRAP;
1725 info.si_code = exit_code;
1726 info.si_pid = task_pid_vnr(current);
1727 info.si_uid = current_uid();
1728
1729 /* Let the debugger run. */
1730 spin_lock_irq(&current->sighand->siglock);
1731 ptrace_stop(exit_code, 1, &info);
1732 spin_unlock_irq(&current->sighand->siglock);
1733 }
1734
1735 /*
1736 * This performs the stopping for SIGSTOP and other stop signals.
1737 * We have to stop all threads in the thread group.
1738 * Returns nonzero if we've actually stopped and released the siglock.
1739 * Returns zero if we didn't stop and still hold the siglock.
1740 */
1741 static int do_signal_stop(int signr)
1742 {
1743 struct signal_struct *sig = current->signal;
1744 int notify;
1745
1746 if (!sig->group_stop_count) {
1747 struct task_struct *t;
1748
1749 if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED) ||
1750 unlikely(signal_group_exit(sig)))
1751 return 0;
1752 /*
1753 * There is no group stop already in progress.
1754 * We must initiate one now.
1755 */
1756 sig->group_exit_code = signr;
1757
1758 sig->group_stop_count = 1;
1759 for (t = next_thread(current); t != current; t = next_thread(t))
1760 /*
1761 * Setting state to TASK_STOPPED for a group
1762 * stop is always done with the siglock held,
1763 * so this check has no races.
1764 */
1765 if (!(t->flags & PF_EXITING) &&
1766 !task_is_stopped_or_traced(t)) {
1767 sig->group_stop_count++;
1768 signal_wake_up(t, 0);
1769 }
1770 }
1771 /*
1772 * If there are no other threads in the group, or if there is
1773 * a group stop in progress and we are the last to stop, report
1774 * to the parent. When ptraced, every thread reports itself.
1775 */
1776 notify = sig->group_stop_count == 1 ? CLD_STOPPED : 0;
1777 notify = tracehook_notify_jctl(notify, CLD_STOPPED);
1778 /*
1779 * tracehook_notify_jctl() can drop and reacquire siglock, so
1780 * we keep ->group_stop_count != 0 before the call. If SIGCONT
1781 * or SIGKILL comes in between ->group_stop_count == 0.
1782 */
1783 if (sig->group_stop_count) {
1784 if (!--sig->group_stop_count)
1785 sig->flags = SIGNAL_STOP_STOPPED;
1786 current->exit_code = sig->group_exit_code;
1787 __set_current_state(TASK_STOPPED);
1788 }
1789 spin_unlock_irq(&current->sighand->siglock);
1790
1791 if (notify) {
1792 read_lock(&tasklist_lock);
1793 do_notify_parent_cldstop(current, notify);
1794 read_unlock(&tasklist_lock);
1795 }
1796
1797 /* Now we don't run again until woken by SIGCONT or SIGKILL */
1798 do {
1799 schedule();
1800 } while (try_to_freeze());
1801
1802 tracehook_finish_jctl();
1803 current->exit_code = 0;
1804
1805 return 1;
1806 }
1807
1808 static int ptrace_signal(int signr, siginfo_t *info,
1809 struct pt_regs *regs, void *cookie)
1810 {
1811 if (!task_ptrace(current))
1812 return signr;
1813
1814 ptrace_signal_deliver(regs, cookie);
1815
1816 /* Let the debugger run. */
1817 ptrace_stop(signr, 0, info);
1818
1819 /* We're back. Did the debugger cancel the sig? */
1820 signr = current->exit_code;
1821 if (signr == 0)
1822 return signr;
1823
1824 current->exit_code = 0;
1825
1826 /* Update the siginfo structure if the signal has
1827 changed. If the debugger wanted something
1828 specific in the siginfo structure then it should
1829 have updated *info via PTRACE_SETSIGINFO. */
1830 if (signr != info->si_signo) {
1831 info->si_signo = signr;
1832 info->si_errno = 0;
1833 info->si_code = SI_USER;
1834 info->si_pid = task_pid_vnr(current->parent);
1835 info->si_uid = task_uid(current->parent);
1836 }
1837
1838 /* If the (new) signal is now blocked, requeue it. */
1839 if (sigismember(&current->blocked, signr)) {
1840 specific_send_sig_info(signr, info, current);
1841 signr = 0;
1842 }
1843
1844 return signr;
1845 }
1846
1847 int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
1848 struct pt_regs *regs, void *cookie)
1849 {
1850 struct sighand_struct *sighand = current->sighand;
1851 struct signal_struct *signal = current->signal;
1852 int signr;
1853
1854 relock:
1855 /*
1856 * We'll jump back here after any time we were stopped in TASK_STOPPED.
1857 * While in TASK_STOPPED, we were considered "frozen enough".
1858 * Now that we woke up, it's crucial if we're supposed to be
1859 * frozen that we freeze now before running anything substantial.
1860 */
1861 try_to_freeze();
1862
1863 spin_lock_irq(&sighand->siglock);
1864 /*
1865 * Every stopped thread goes here after wakeup. Check to see if
1866 * we should notify the parent, prepare_signal(SIGCONT) encodes
1867 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
1868 */
1869 if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
1870 int why = (signal->flags & SIGNAL_STOP_CONTINUED)
1871 ? CLD_CONTINUED : CLD_STOPPED;
1872 signal->flags &= ~SIGNAL_CLD_MASK;
1873
1874 why = tracehook_notify_jctl(why, CLD_CONTINUED);
1875 spin_unlock_irq(&sighand->siglock);
1876
1877 if (why) {
1878 read_lock(&tasklist_lock);
1879 do_notify_parent_cldstop(current->group_leader, why);
1880 read_unlock(&tasklist_lock);
1881 }
1882 goto relock;
1883 }
1884
1885 for (;;) {
1886 struct k_sigaction *ka;
1887 /*
1888 * Tracing can induce an artifical signal and choose sigaction.
1889 * The return value in @signr determines the default action,
1890 * but @info->si_signo is the signal number we will report.
1891 */
1892 signr = tracehook_get_signal(current, regs, info, return_ka);
1893 if (unlikely(signr < 0))
1894 goto relock;
1895 if (unlikely(signr != 0))
1896 ka = return_ka;
1897 else {
1898 if (unlikely(signal->group_stop_count > 0) &&
1899 do_signal_stop(0))
1900 goto relock;
1901
1902 signr = dequeue_signal(current, &current->blocked,
1903 info);
1904
1905 if (!signr)
1906 break; /* will return 0 */
1907
1908 if (signr != SIGKILL) {
1909 signr = ptrace_signal(signr, info,
1910 regs, cookie);
1911 if (!signr)
1912 continue;
1913 }
1914
1915 ka = &sighand->action[signr-1];
1916 }
1917
1918 /* Trace actually delivered signals. */
1919 trace_signal_deliver(signr, info, ka);
1920
1921 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing. */
1922 continue;
1923 if (ka->sa.sa_handler != SIG_DFL) {
1924 /* Run the handler. */
1925 *return_ka = *ka;
1926
1927 if (ka->sa.sa_flags & SA_ONESHOT)
1928 ka->sa.sa_handler = SIG_DFL;
1929
1930 break; /* will return non-zero "signr" value */
1931 }
1932
1933 /*
1934 * Now we are doing the default action for this signal.
1935 */
1936 if (sig_kernel_ignore(signr)) /* Default is nothing. */
1937 continue;
1938
1939 /*
1940 * Global init gets no signals it doesn't want.
1941 * Container-init gets no signals it doesn't want from same
1942 * container.
1943 *
1944 * Note that if global/container-init sees a sig_kernel_only()
1945 * signal here, the signal must have been generated internally
1946 * or must have come from an ancestor namespace. In either
1947 * case, the signal cannot be dropped.
1948 */
1949 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
1950 !sig_kernel_only(signr))
1951 continue;
1952
1953 if (sig_kernel_stop(signr)) {
1954 /*
1955 * The default action is to stop all threads in
1956 * the thread group. The job control signals
1957 * do nothing in an orphaned pgrp, but SIGSTOP
1958 * always works. Note that siglock needs to be
1959 * dropped during the call to is_orphaned_pgrp()
1960 * because of lock ordering with tasklist_lock.
1961 * This allows an intervening SIGCONT to be posted.
1962 * We need to check for that and bail out if necessary.
1963 */
1964 if (signr != SIGSTOP) {
1965 spin_unlock_irq(&sighand->siglock);
1966
1967 /* signals can be posted during this window */
1968
1969 if (is_current_pgrp_orphaned())
1970 goto relock;
1971
1972 spin_lock_irq(&sighand->siglock);
1973 }
1974
1975 if (likely(do_signal_stop(info->si_signo))) {
1976 /* It released the siglock. */
1977 goto relock;
1978 }
1979
1980 /*
1981 * We didn't actually stop, due to a race
1982 * with SIGCONT or something like that.
1983 */
1984 continue;
1985 }
1986
1987 spin_unlock_irq(&sighand->siglock);
1988
1989 /*
1990 * Anything else is fatal, maybe with a core dump.
1991 */
1992 current->flags |= PF_SIGNALED;
1993
1994 if (sig_kernel_coredump(signr)) {
1995 if (print_fatal_signals)
1996 print_fatal_signal(regs, info->si_signo);
1997 /*
1998 * If it was able to dump core, this kills all
1999 * other threads in the group and synchronizes with
2000 * their demise. If we lost the race with another
2001 * thread getting here, it set group_exit_code
2002 * first and our do_group_exit call below will use
2003 * that value and ignore the one we pass it.
2004 */
2005 do_coredump(info->si_signo, info->si_signo, regs);
2006 }
2007
2008 /*
2009 * Death signals, no core dump.
2010 */
2011 do_group_exit(info->si_signo);
2012 /* NOTREACHED */
2013 }
2014 spin_unlock_irq(&sighand->siglock);
2015 return signr;
2016 }
2017
2018 void exit_signals(struct task_struct *tsk)
2019 {
2020 int group_stop = 0;
2021 struct task_struct *t;
2022
2023 if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
2024 tsk->flags |= PF_EXITING;
2025 return;
2026 }
2027
2028 spin_lock_irq(&tsk->sighand->siglock);
2029 /*
2030 * From now this task is not visible for group-wide signals,
2031 * see wants_signal(), do_signal_stop().
2032 */
2033 tsk->flags |= PF_EXITING;
2034 if (!signal_pending(tsk))
2035 goto out;
2036
2037 /* It could be that __group_complete_signal() choose us to
2038 * notify about group-wide signal. Another thread should be
2039 * woken now to take the signal since we will not.
2040 */
2041 for (t = tsk; (t = next_thread(t)) != tsk; )
2042 if (!signal_pending(t) && !(t->flags & PF_EXITING))
2043 recalc_sigpending_and_wake(t);
2044
2045 if (unlikely(tsk->signal->group_stop_count) &&
2046 !--tsk->signal->group_stop_count) {
2047 tsk->signal->flags = SIGNAL_STOP_STOPPED;
2048 group_stop = tracehook_notify_jctl(CLD_STOPPED, CLD_STOPPED);
2049 }
2050 out:
2051 spin_unlock_irq(&tsk->sighand->siglock);
2052
2053 if (unlikely(group_stop)) {
2054 read_lock(&tasklist_lock);
2055 do_notify_parent_cldstop(tsk, group_stop);
2056 read_unlock(&tasklist_lock);
2057 }
2058 }
2059
2060 EXPORT_SYMBOL(recalc_sigpending);
2061 EXPORT_SYMBOL_GPL(dequeue_signal);
2062 EXPORT_SYMBOL(flush_signals);
2063 EXPORT_SYMBOL(force_sig);
2064 EXPORT_SYMBOL(send_sig);
2065 EXPORT_SYMBOL(send_sig_info);
2066 EXPORT_SYMBOL(sigprocmask);
2067 EXPORT_SYMBOL(block_all_signals);
2068 EXPORT_SYMBOL(unblock_all_signals);
2069
2070
2071 /*
2072 * System call entry points.
2073 */
2074
2075 SYSCALL_DEFINE0(restart_syscall)
2076 {
2077 struct restart_block *restart = &current_thread_info()->restart_block;
2078 return restart->fn(restart);
2079 }
2080
2081 long do_no_restart_syscall(struct restart_block *param)
2082 {
2083 return -EINTR;
2084 }
2085
2086 /*
2087 * We don't need to get the kernel lock - this is all local to this
2088 * particular thread.. (and that's good, because this is _heavily_
2089 * used by various programs)
2090 */
2091
2092 /*
2093 * This is also useful for kernel threads that want to temporarily
2094 * (or permanently) block certain signals.
2095 *
2096 * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
2097 * interface happily blocks "unblockable" signals like SIGKILL
2098 * and friends.
2099 */
2100 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2101 {
2102 int error;
2103
2104 spin_lock_irq(&current->sighand->siglock);
2105 if (oldset)
2106 *oldset = current->blocked;
2107
2108 error = 0;
2109 switch (how) {
2110 case SIG_BLOCK:
2111 sigorsets(&current->blocked, &current->blocked, set);
2112 break;
2113 case SIG_UNBLOCK:
2114 signandsets(&current->blocked, &current->blocked, set);
2115 break;
2116 case SIG_SETMASK:
2117 current->blocked = *set;
2118 break;
2119 default:
2120 error = -EINVAL;
2121 }
2122 recalc_sigpending();
2123 spin_unlock_irq(&current->sighand->siglock);
2124
2125 return error;
2126 }
2127
2128 SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, set,
2129 sigset_t __user *, oset, size_t, sigsetsize)
2130 {
2131 int error = -EINVAL;
2132 sigset_t old_set, new_set;
2133
2134 /* XXX: Don't preclude handling different sized sigset_t's. */
2135 if (sigsetsize != sizeof(sigset_t))
2136 goto out;
2137
2138 if (set) {
2139 error = -EFAULT;
2140 if (copy_from_user(&new_set, set, sizeof(*set)))
2141 goto out;
2142 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2143
2144 error = sigprocmask(how, &new_set, &old_set);
2145 if (error)
2146 goto out;
2147 if (oset)
2148 goto set_old;
2149 } else if (oset) {
2150 spin_lock_irq(&current->sighand->siglock);
2151 old_set = current->blocked;
2152 spin_unlock_irq(&current->sighand->siglock);
2153
2154 set_old:
2155 error = -EFAULT;
2156 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2157 goto out;
2158 }
2159 error = 0;
2160 out:
2161 return error;
2162 }
2163
2164 long do_sigpending(void __user *set, unsigned long sigsetsize)
2165 {
2166 long error = -EINVAL;
2167 sigset_t pending;
2168
2169 if (sigsetsize > sizeof(sigset_t))
2170 goto out;
2171
2172 spin_lock_irq(&current->sighand->siglock);
2173 sigorsets(&pending, &current->pending.signal,
2174 &current->signal->shared_pending.signal);
2175 spin_unlock_irq(&current->sighand->siglock);
2176
2177 /* Outside the lock because only this thread touches it. */
2178 sigandsets(&pending, &current->blocked, &pending);
2179
2180 error = -EFAULT;
2181 if (!copy_to_user(set, &pending, sigsetsize))
2182 error = 0;
2183
2184 out:
2185 return error;
2186 }
2187
2188 SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, set, size_t, sigsetsize)
2189 {
2190 return do_sigpending(set, sigsetsize);
2191 }
2192
2193 #ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2194
2195 int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2196 {
2197 int err;
2198
2199 if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2200 return -EFAULT;
2201 if (from->si_code < 0)
2202 return __copy_to_user(to, from, sizeof(siginfo_t))
2203 ? -EFAULT : 0;
2204 /*
2205 * If you change siginfo_t structure, please be sure
2206 * this code is fixed accordingly.
2207 * Please remember to update the signalfd_copyinfo() function
2208 * inside fs/signalfd.c too, in case siginfo_t changes.
2209 * It should never copy any pad contained in the structure
2210 * to avoid security leaks, but must copy the generic
2211 * 3 ints plus the relevant union member.
2212 */
2213 err = __put_user(from->si_signo, &to->si_signo);
2214 err |= __put_user(from->si_errno, &to->si_errno);
2215 err |= __put_user((short)from->si_code, &to->si_code);
2216 switch (from->si_code & __SI_MASK) {
2217 case __SI_KILL:
2218 err |= __put_user(from->si_pid, &to->si_pid);
2219 err |= __put_user(from->si_uid, &to->si_uid);
2220 break;
2221 case __SI_TIMER:
2222 err |= __put_user(from->si_tid, &to->si_tid);
2223 err |= __put_user(from->si_overrun, &to->si_overrun);
2224 err |= __put_user(from->si_ptr, &to->si_ptr);
2225 break;
2226 case __SI_POLL:
2227 err |= __put_user(from->si_band, &to->si_band);
2228 err |= __put_user(from->si_fd, &to->si_fd);
2229 break;
2230 case __SI_FAULT:
2231 err |= __put_user(from->si_addr, &to->si_addr);
2232 #ifdef __ARCH_SI_TRAPNO
2233 err |= __put_user(from->si_trapno, &to->si_trapno);
2234 #endif
2235 #ifdef BUS_MCEERR_AO
2236 /*
2237 * Other callers might not initialize the si_lsb field,
2238 * so check explicitely for the right codes here.
2239 */
2240 if (from->si_code == BUS_MCEERR_AR || from->si_code == BUS_MCEERR_AO)
2241 err |= __put_user(from->si_addr_lsb, &to->si_addr_lsb);
2242 #endif
2243 break;
2244 case __SI_CHLD:
2245 err |= __put_user(from->si_pid, &to->si_pid);
2246 err |= __put_user(from->si_uid, &to->si_uid);
2247 err |= __put_user(from->si_status, &to->si_status);
2248 err |= __put_user(from->si_utime, &to->si_utime);
2249 err |= __put_user(from->si_stime, &to->si_stime);
2250 break;
2251 case __SI_RT: /* This is not generated by the kernel as of now. */
2252 case __SI_MESGQ: /* But this is */
2253 err |= __put_user(from->si_pid, &to->si_pid);
2254 err |= __put_user(from->si_uid, &to->si_uid);
2255 err |= __put_user(from->si_ptr, &to->si_ptr);
2256 break;
2257 default: /* this is just in case for now ... */
2258 err |= __put_user(from->si_pid, &to->si_pid);
2259 err |= __put_user(from->si_uid, &to->si_uid);
2260 break;
2261 }
2262 return err;
2263 }
2264
2265 #endif
2266
2267 SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
2268 siginfo_t __user *, uinfo, const struct timespec __user *, uts,
2269 size_t, sigsetsize)
2270 {
2271 int ret, sig;
2272 sigset_t these;
2273 struct timespec ts;
2274 siginfo_t info;
2275 long timeout = 0;
2276
2277 /* XXX: Don't preclude handling different sized sigset_t's. */
2278 if (sigsetsize != sizeof(sigset_t))
2279 return -EINVAL;
2280
2281 if (copy_from_user(&these, uthese, sizeof(these)))
2282 return -EFAULT;
2283
2284 /*
2285 * Invert the set of allowed signals to get those we
2286 * want to block.
2287 */
2288 sigdelsetmask(&these, sigmask(SIGKILL)|sigmask(SIGSTOP));
2289 signotset(&these);
2290
2291 if (uts) {
2292 if (copy_from_user(&ts, uts, sizeof(ts)))
2293 return -EFAULT;
2294 if (ts.tv_nsec >= 1000000000L || ts.tv_nsec < 0
2295 || ts.tv_sec < 0)
2296 return -EINVAL;
2297 }
2298
2299 spin_lock_irq(&current->sighand->siglock);
2300 sig = dequeue_signal(current, &these, &info);
2301 if (!sig) {
2302 timeout = MAX_SCHEDULE_TIMEOUT;
2303 if (uts)
2304 timeout = (timespec_to_jiffies(&ts)
2305 + (ts.tv_sec || ts.tv_nsec));
2306
2307 if (timeout) {
2308 /* None ready -- temporarily unblock those we're
2309 * interested while we are sleeping in so that we'll
2310 * be awakened when they arrive. */
2311 current->real_blocked = current->blocked;
2312 sigandsets(&current->blocked, &current->blocked, &these);
2313 recalc_sigpending();
2314 spin_unlock_irq(&current->sighand->siglock);
2315
2316 timeout = schedule_timeout_interruptible(timeout);
2317
2318 spin_lock_irq(&current->sighand->siglock);
2319 sig = dequeue_signal(current, &these, &info);
2320 current->blocked = current->real_blocked;
2321 siginitset(&current->real_blocked, 0);
2322 recalc_sigpending();
2323 }
2324 }
2325 spin_unlock_irq(&current->sighand->siglock);
2326
2327 if (sig) {
2328 ret = sig;
2329 if (uinfo) {
2330 if (copy_siginfo_to_user(uinfo, &info))
2331 ret = -EFAULT;
2332 }
2333 } else {
2334 ret = -EAGAIN;
2335 if (timeout)
2336 ret = -EINTR;
2337 }
2338
2339 return ret;
2340 }
2341
2342 SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
2343 {
2344 struct siginfo info;
2345
2346 info.si_signo = sig;
2347 info.si_errno = 0;
2348 info.si_code = SI_USER;
2349 info.si_pid = task_tgid_vnr(current);
2350 info.si_uid = current_uid();
2351
2352 return kill_something_info(sig, &info, pid);
2353 }
2354
2355 static int
2356 do_send_specific(pid_t tgid, pid_t pid, int sig, struct siginfo *info)
2357 {
2358 struct task_struct *p;
2359 int error = -ESRCH;
2360
2361 rcu_read_lock();
2362 p = find_task_by_vpid(pid);
2363 if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
2364 error = check_kill_permission(sig, info, p);
2365 /*
2366 * The null signal is a permissions and process existence
2367 * probe. No signal is actually delivered.
2368 */
2369 if (!error && sig) {
2370 error = do_send_sig_info(sig, info, p, false);
2371 /*
2372 * If lock_task_sighand() failed we pretend the task
2373 * dies after receiving the signal. The window is tiny,
2374 * and the signal is private anyway.
2375 */
2376 if (unlikely(error == -ESRCH))
2377 error = 0;
2378 }
2379 }
2380 rcu_read_unlock();
2381
2382 return error;
2383 }
2384
2385 static int do_tkill(pid_t tgid, pid_t pid, int sig)
2386 {
2387 struct siginfo info;
2388
2389 info.si_signo = sig;
2390 info.si_errno = 0;
2391 info.si_code = SI_TKILL;
2392 info.si_pid = task_tgid_vnr(current);
2393 info.si_uid = current_uid();
2394
2395 return do_send_specific(tgid, pid, sig, &info);
2396 }
2397
2398 /**
2399 * sys_tgkill - send signal to one specific thread
2400 * @tgid: the thread group ID of the thread
2401 * @pid: the PID of the thread
2402 * @sig: signal to be sent
2403 *
2404 * This syscall also checks the @tgid and returns -ESRCH even if the PID
2405 * exists but it's not belonging to the target process anymore. This
2406 * method solves the problem of threads exiting and PIDs getting reused.
2407 */
2408 SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
2409 {
2410 /* This is only valid for single tasks */
2411 if (pid <= 0 || tgid <= 0)
2412 return -EINVAL;
2413
2414 return do_tkill(tgid, pid, sig);
2415 }
2416
2417 /*
2418 * Send a signal to only one task, even if it's a CLONE_THREAD task.
2419 */
2420 SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
2421 {
2422 /* This is only valid for single tasks */
2423 if (pid <= 0)
2424 return -EINVAL;
2425
2426 return do_tkill(0, pid, sig);
2427 }
2428
2429 SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
2430 siginfo_t __user *, uinfo)
2431 {
2432 siginfo_t info;
2433
2434 if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2435 return -EFAULT;
2436
2437 /* Not even root can pretend to send signals from the kernel.
2438 * Nor can they impersonate a kill()/tgkill(), which adds source info.
2439 */
2440 if (info.si_code != SI_QUEUE) {
2441 /* We used to allow any < 0 si_code */
2442 WARN_ON_ONCE(info.si_code < 0);
2443 return -EPERM;
2444 }
2445 info.si_signo = sig;
2446
2447 /* POSIX.1b doesn't mention process groups. */
2448 return kill_proc_info(sig, &info, pid);
2449 }
2450
2451 long do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, siginfo_t *info)
2452 {
2453 /* This is only valid for single tasks */
2454 if (pid <= 0 || tgid <= 0)
2455 return -EINVAL;
2456
2457 /* Not even root can pretend to send signals from the kernel.
2458 * Nor can they impersonate a kill()/tgkill(), which adds source info.
2459 */
2460 if (info->si_code != SI_QUEUE) {
2461 /* We used to allow any < 0 si_code */
2462 WARN_ON_ONCE(info->si_code < 0);
2463 return -EPERM;
2464 }
2465 info->si_signo = sig;
2466
2467 return do_send_specific(tgid, pid, sig, info);
2468 }
2469
2470 SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
2471 siginfo_t __user *, uinfo)
2472 {
2473 siginfo_t info;
2474
2475 if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2476 return -EFAULT;
2477
2478 return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
2479 }
2480
2481 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
2482 {
2483 struct task_struct *t = current;
2484 struct k_sigaction *k;
2485 sigset_t mask;
2486
2487 if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
2488 return -EINVAL;
2489
2490 k = &t->sighand->action[sig-1];
2491
2492 spin_lock_irq(&current->sighand->siglock);
2493 if (oact)
2494 *oact = *k;
2495
2496 if (act) {
2497 sigdelsetmask(&act->sa.sa_mask,
2498 sigmask(SIGKILL) | sigmask(SIGSTOP));
2499 *k = *act;
2500 /*
2501 * POSIX 3.3.1.3:
2502 * "Setting a signal action to SIG_IGN for a signal that is
2503 * pending shall cause the pending signal to be discarded,
2504 * whether or not it is blocked."
2505 *
2506 * "Setting a signal action to SIG_DFL for a signal that is
2507 * pending and whose default action is to ignore the signal
2508 * (for example, SIGCHLD), shall cause the pending signal to
2509 * be discarded, whether or not it is blocked"
2510 */
2511 if (sig_handler_ignored(sig_handler(t, sig), sig)) {
2512 sigemptyset(&mask);
2513 sigaddset(&mask, sig);
2514 rm_from_queue_full(&mask, &t->signal->shared_pending);
2515 do {
2516 rm_from_queue_full(&mask, &t->pending);
2517 t = next_thread(t);
2518 } while (t != current);
2519 }
2520 }
2521
2522 spin_unlock_irq(&current->sighand->siglock);
2523 return 0;
2524 }
2525
2526 int
2527 do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
2528 {
2529 stack_t oss;
2530 int error;
2531
2532 oss.ss_sp = (void __user *) current->sas_ss_sp;
2533 oss.ss_size = current->sas_ss_size;
2534 oss.ss_flags = sas_ss_flags(sp);
2535
2536 if (uss) {
2537 void __user *ss_sp;
2538 size_t ss_size;
2539 int ss_flags;
2540
2541 error = -EFAULT;
2542 if (!access_ok(VERIFY_READ, uss, sizeof(*uss)))
2543 goto out;
2544 error = __get_user(ss_sp, &uss->ss_sp) |
2545 __get_user(ss_flags, &uss->ss_flags) |
2546 __get_user(ss_size, &uss->ss_size);
2547 if (error)
2548 goto out;
2549
2550 error = -EPERM;
2551 if (on_sig_stack(sp))
2552 goto out;
2553
2554 error = -EINVAL;
2555 /*
2556 *
2557 * Note - this code used to test ss_flags incorrectly
2558 * old code may have been written using ss_flags==0
2559 * to mean ss_flags==SS_ONSTACK (as this was the only
2560 * way that worked) - this fix preserves that older
2561 * mechanism
2562 */
2563 if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
2564 goto out;
2565
2566 if (ss_flags == SS_DISABLE) {
2567 ss_size = 0;
2568 ss_sp = NULL;
2569 } else {
2570 error = -ENOMEM;
2571 if (ss_size < MINSIGSTKSZ)
2572 goto out;
2573 }
2574
2575 current->sas_ss_sp = (unsigned long) ss_sp;
2576 current->sas_ss_size = ss_size;
2577 }
2578
2579 error = 0;
2580 if (uoss) {
2581 error = -EFAULT;
2582 if (!access_ok(VERIFY_WRITE, uoss, sizeof(*uoss)))
2583 goto out;
2584 error = __put_user(oss.ss_sp, &uoss->ss_sp) |
2585 __put_user(oss.ss_size, &uoss->ss_size) |
2586 __put_user(oss.ss_flags, &uoss->ss_flags);
2587 }
2588
2589 out:
2590 return error;
2591 }
2592
2593 #ifdef __ARCH_WANT_SYS_SIGPENDING
2594
2595 SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, set)
2596 {
2597 return do_sigpending(set, sizeof(*set));
2598 }
2599
2600 #endif
2601
2602 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
2603 /* Some platforms have their own version with special arguments others
2604 support only sys_rt_sigprocmask. */
2605
2606 SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, set,
2607 old_sigset_t __user *, oset)
2608 {
2609 int error;
2610 old_sigset_t old_set, new_set;
2611
2612 if (set) {
2613 error = -EFAULT;
2614 if (copy_from_user(&new_set, set, sizeof(*set)))
2615 goto out;
2616 new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
2617
2618 spin_lock_irq(&current->sighand->siglock);
2619 old_set = current->blocked.sig[0];
2620
2621 error = 0;
2622 switch (how) {
2623 default:
2624 error = -EINVAL;
2625 break;
2626 case SIG_BLOCK:
2627 sigaddsetmask(&current->blocked, new_set);
2628 break;
2629 case SIG_UNBLOCK:
2630 sigdelsetmask(&current->blocked, new_set);
2631 break;
2632 case SIG_SETMASK:
2633 current->blocked.sig[0] = new_set;
2634 break;
2635 }
2636
2637 recalc_sigpending();
2638 spin_unlock_irq(&current->sighand->siglock);
2639 if (error)
2640 goto out;
2641 if (oset)
2642 goto set_old;
2643 } else if (oset) {
2644 old_set = current->blocked.sig[0];
2645 set_old:
2646 error = -EFAULT;
2647 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2648 goto out;
2649 }
2650 error = 0;
2651 out:
2652 return error;
2653 }
2654 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
2655
2656 #ifdef __ARCH_WANT_SYS_RT_SIGACTION
2657 SYSCALL_DEFINE4(rt_sigaction, int, sig,
2658 const struct sigaction __user *, act,
2659 struct sigaction __user *, oact,
2660 size_t, sigsetsize)
2661 {
2662 struct k_sigaction new_sa, old_sa;
2663 int ret = -EINVAL;
2664
2665 /* XXX: Don't preclude handling different sized sigset_t's. */
2666 if (sigsetsize != sizeof(sigset_t))
2667 goto out;
2668
2669 if (act) {
2670 if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
2671 return -EFAULT;
2672 }
2673
2674 ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
2675
2676 if (!ret && oact) {
2677 if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
2678 return -EFAULT;
2679 }
2680 out:
2681 return ret;
2682 }
2683 #endif /* __ARCH_WANT_SYS_RT_SIGACTION */
2684
2685 #ifdef __ARCH_WANT_SYS_SGETMASK
2686
2687 /*
2688 * For backwards compatibility. Functionality superseded by sigprocmask.
2689 */
2690 SYSCALL_DEFINE0(sgetmask)
2691 {
2692 /* SMP safe */
2693 return current->blocked.sig[0];
2694 }
2695
2696 SYSCALL_DEFINE1(ssetmask, int, newmask)
2697 {
2698 int old;
2699
2700 spin_lock_irq(&current->sighand->siglock);
2701 old = current->blocked.sig[0];
2702
2703 siginitset(&current->blocked, newmask & ~(sigmask(SIGKILL)|
2704 sigmask(SIGSTOP)));
2705 recalc_sigpending();
2706 spin_unlock_irq(&current->sighand->siglock);
2707
2708 return old;
2709 }
2710 #endif /* __ARCH_WANT_SGETMASK */
2711
2712 #ifdef __ARCH_WANT_SYS_SIGNAL
2713 /*
2714 * For backwards compatibility. Functionality superseded by sigaction.
2715 */
2716 SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
2717 {
2718 struct k_sigaction new_sa, old_sa;
2719 int ret;
2720
2721 new_sa.sa.sa_handler = handler;
2722 new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
2723 sigemptyset(&new_sa.sa.sa_mask);
2724
2725 ret = do_sigaction(sig, &new_sa, &old_sa);
2726
2727 return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
2728 }
2729 #endif /* __ARCH_WANT_SYS_SIGNAL */
2730
2731 #ifdef __ARCH_WANT_SYS_PAUSE
2732
2733 SYSCALL_DEFINE0(pause)
2734 {
2735 current->state = TASK_INTERRUPTIBLE;
2736 schedule();
2737 return -ERESTARTNOHAND;
2738 }
2739
2740 #endif
2741
2742 #ifdef __ARCH_WANT_SYS_RT_SIGSUSPEND
2743 SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
2744 {
2745 sigset_t newset;
2746
2747 /* XXX: Don't preclude handling different sized sigset_t's. */
2748 if (sigsetsize != sizeof(sigset_t))
2749 return -EINVAL;
2750
2751 if (copy_from_user(&newset, unewset, sizeof(newset)))
2752 return -EFAULT;
2753 sigdelsetmask(&newset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2754
2755 spin_lock_irq(&current->sighand->siglock);
2756 current->saved_sigmask = current->blocked;
2757 current->blocked = newset;
2758 recalc_sigpending();
2759 spin_unlock_irq(&current->sighand->siglock);
2760
2761 current->state = TASK_INTERRUPTIBLE;
2762 schedule();
2763 set_restore_sigmask();
2764 return -ERESTARTNOHAND;
2765 }
2766 #endif /* __ARCH_WANT_SYS_RT_SIGSUSPEND */
2767
2768 __attribute__((weak)) const char *arch_vma_name(struct vm_area_struct *vma)
2769 {
2770 return NULL;
2771 }
2772
2773 void __init signals_init(void)
2774 {
2775 sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
2776 }
2777
2778 #ifdef CONFIG_KGDB_KDB
2779 #include <linux/kdb.h>
2780 /*
2781 * kdb_send_sig_info - Allows kdb to send signals without exposing
2782 * signal internals. This function checks if the required locks are
2783 * available before calling the main signal code, to avoid kdb
2784 * deadlocks.
2785 */
2786 void
2787 kdb_send_sig_info(struct task_struct *t, struct siginfo *info)
2788 {
2789 static struct task_struct *kdb_prev_t;
2790 int sig, new_t;
2791 if (!spin_trylock(&t->sighand->siglock)) {
2792 kdb_printf("Can't do kill command now.\n"
2793 "The sigmask lock is held somewhere else in "
2794 "kernel, try again later\n");
2795 return;
2796 }
2797 spin_unlock(&t->sighand->siglock);
2798 new_t = kdb_prev_t != t;
2799 kdb_prev_t = t;
2800 if (t->state != TASK_RUNNING && new_t) {
2801 kdb_printf("Process is not RUNNING, sending a signal from "
2802 "kdb risks deadlock\n"
2803 "on the run queue locks. "
2804 "The signal has _not_ been sent.\n"
2805 "Reissue the kill command if you want to risk "
2806 "the deadlock.\n");
2807 return;
2808 }
2809 sig = info->si_signo;
2810 if (send_sig_info(sig, info, t))
2811 kdb_printf("Fail to deliver Signal %d to process %d.\n",
2812 sig, t->pid);
2813 else
2814 kdb_printf("Signal %d is sent to process %d.\n", sig, t->pid);
2815 }
2816 #endif /* CONFIG_KGDB_KDB */