xref: /linux/kernel/signal.c (revision 843aef4930b9953c9ca624a990b201440304b56f)
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/tracehook.h>
26 #include <linux/capability.h>
27 #include <linux/freezer.h>
28 #include <linux/pid_namespace.h>
29 #include <linux/nsproxy.h>
30 #include <trace/sched.h>
31 
32 #include <asm/param.h>
33 #include <asm/uaccess.h>
34 #include <asm/unistd.h>
35 #include <asm/siginfo.h>
36 #include "audit.h"	/* audit_signal_info() */
37 
38 /*
39  * SLAB caches for signal bits.
40  */
41 
42 static struct kmem_cache *sigqueue_cachep;
43 
44 DEFINE_TRACE(sched_signal_send);
45 
46 static void __user *sig_handler(struct task_struct *t, int sig)
47 {
48 	return t->sighand->action[sig - 1].sa.sa_handler;
49 }
50 
51 static int sig_handler_ignored(void __user *handler, int sig)
52 {
53 	/* Is it explicitly or implicitly ignored? */
54 	return handler == SIG_IGN ||
55 		(handler == SIG_DFL && sig_kernel_ignore(sig));
56 }
57 
58 static int sig_ignored(struct task_struct *t, int sig)
59 {
60 	void __user *handler;
61 
62 	/*
63 	 * Blocked signals are never ignored, since the
64 	 * signal handler may change by the time it is
65 	 * unblocked.
66 	 */
67 	if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
68 		return 0;
69 
70 	handler = sig_handler(t, sig);
71 	if (!sig_handler_ignored(handler, sig))
72 		return 0;
73 
74 	/*
75 	 * Tracers may want to know about even ignored signals.
76 	 */
77 	return !tracehook_consider_ignored_signal(t, sig, handler);
78 }
79 
80 /*
81  * Re-calculate pending state from the set of locally pending
82  * signals, globally pending signals, and blocked signals.
83  */
84 static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
85 {
86 	unsigned long ready;
87 	long i;
88 
89 	switch (_NSIG_WORDS) {
90 	default:
91 		for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
92 			ready |= signal->sig[i] &~ blocked->sig[i];
93 		break;
94 
95 	case 4: ready  = signal->sig[3] &~ blocked->sig[3];
96 		ready |= signal->sig[2] &~ blocked->sig[2];
97 		ready |= signal->sig[1] &~ blocked->sig[1];
98 		ready |= signal->sig[0] &~ blocked->sig[0];
99 		break;
100 
101 	case 2: ready  = signal->sig[1] &~ blocked->sig[1];
102 		ready |= signal->sig[0] &~ blocked->sig[0];
103 		break;
104 
105 	case 1: ready  = signal->sig[0] &~ blocked->sig[0];
106 	}
107 	return ready !=	0;
108 }
109 
110 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
111 
112 static int recalc_sigpending_tsk(struct task_struct *t)
113 {
114 	if (t->signal->group_stop_count > 0 ||
115 	    PENDING(&t->pending, &t->blocked) ||
116 	    PENDING(&t->signal->shared_pending, &t->blocked)) {
117 		set_tsk_thread_flag(t, TIF_SIGPENDING);
118 		return 1;
119 	}
120 	/*
121 	 * We must never clear the flag in another thread, or in current
122 	 * when it's possible the current syscall is returning -ERESTART*.
123 	 * So we don't clear it here, and only callers who know they should do.
124 	 */
125 	return 0;
126 }
127 
128 /*
129  * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
130  * This is superfluous when called on current, the wakeup is a harmless no-op.
131  */
132 void recalc_sigpending_and_wake(struct task_struct *t)
133 {
134 	if (recalc_sigpending_tsk(t))
135 		signal_wake_up(t, 0);
136 }
137 
138 void recalc_sigpending(void)
139 {
140 	if (unlikely(tracehook_force_sigpending()))
141 		set_thread_flag(TIF_SIGPENDING);
142 	else if (!recalc_sigpending_tsk(current) && !freezing(current))
143 		clear_thread_flag(TIF_SIGPENDING);
144 
145 }
146 
147 /* Given the mask, find the first available signal that should be serviced. */
148 
149 int next_signal(struct sigpending *pending, sigset_t *mask)
150 {
151 	unsigned long i, *s, *m, x;
152 	int sig = 0;
153 
154 	s = pending->signal.sig;
155 	m = mask->sig;
156 	switch (_NSIG_WORDS) {
157 	default:
158 		for (i = 0; i < _NSIG_WORDS; ++i, ++s, ++m)
159 			if ((x = *s &~ *m) != 0) {
160 				sig = ffz(~x) + i*_NSIG_BPW + 1;
161 				break;
162 			}
163 		break;
164 
165 	case 2: if ((x = s[0] &~ m[0]) != 0)
166 			sig = 1;
167 		else if ((x = s[1] &~ m[1]) != 0)
168 			sig = _NSIG_BPW + 1;
169 		else
170 			break;
171 		sig += ffz(~x);
172 		break;
173 
174 	case 1: if ((x = *s &~ *m) != 0)
175 			sig = ffz(~x) + 1;
176 		break;
177 	}
178 
179 	return sig;
180 }
181 
182 /*
183  * allocate a new signal queue record
184  * - this may be called without locks if and only if t == current, otherwise an
185  *   appopriate lock must be held to stop the target task from exiting
186  */
187 static struct sigqueue *__sigqueue_alloc(struct task_struct *t, gfp_t flags,
188 					 int override_rlimit)
189 {
190 	struct sigqueue *q = NULL;
191 	struct user_struct *user;
192 
193 	/*
194 	 * We won't get problems with the target's UID changing under us
195 	 * because changing it requires RCU be used, and if t != current, the
196 	 * caller must be holding the RCU readlock (by way of a spinlock) and
197 	 * we use RCU protection here
198 	 */
199 	user = get_uid(__task_cred(t)->user);
200 	atomic_inc(&user->sigpending);
201 	if (override_rlimit ||
202 	    atomic_read(&user->sigpending) <=
203 			t->signal->rlim[RLIMIT_SIGPENDING].rlim_cur)
204 		q = kmem_cache_alloc(sigqueue_cachep, flags);
205 	if (unlikely(q == NULL)) {
206 		atomic_dec(&user->sigpending);
207 		free_uid(user);
208 	} else {
209 		INIT_LIST_HEAD(&q->list);
210 		q->flags = 0;
211 		q->user = user;
212 	}
213 
214 	return q;
215 }
216 
217 static void __sigqueue_free(struct sigqueue *q)
218 {
219 	if (q->flags & SIGQUEUE_PREALLOC)
220 		return;
221 	atomic_dec(&q->user->sigpending);
222 	free_uid(q->user);
223 	kmem_cache_free(sigqueue_cachep, q);
224 }
225 
226 void flush_sigqueue(struct sigpending *queue)
227 {
228 	struct sigqueue *q;
229 
230 	sigemptyset(&queue->signal);
231 	while (!list_empty(&queue->list)) {
232 		q = list_entry(queue->list.next, struct sigqueue , list);
233 		list_del_init(&q->list);
234 		__sigqueue_free(q);
235 	}
236 }
237 
238 /*
239  * Flush all pending signals for a task.
240  */
241 void flush_signals(struct task_struct *t)
242 {
243 	unsigned long flags;
244 
245 	spin_lock_irqsave(&t->sighand->siglock, flags);
246 	clear_tsk_thread_flag(t, TIF_SIGPENDING);
247 	flush_sigqueue(&t->pending);
248 	flush_sigqueue(&t->signal->shared_pending);
249 	spin_unlock_irqrestore(&t->sighand->siglock, flags);
250 }
251 
252 static void __flush_itimer_signals(struct sigpending *pending)
253 {
254 	sigset_t signal, retain;
255 	struct sigqueue *q, *n;
256 
257 	signal = pending->signal;
258 	sigemptyset(&retain);
259 
260 	list_for_each_entry_safe(q, n, &pending->list, list) {
261 		int sig = q->info.si_signo;
262 
263 		if (likely(q->info.si_code != SI_TIMER)) {
264 			sigaddset(&retain, sig);
265 		} else {
266 			sigdelset(&signal, sig);
267 			list_del_init(&q->list);
268 			__sigqueue_free(q);
269 		}
270 	}
271 
272 	sigorsets(&pending->signal, &signal, &retain);
273 }
274 
275 void flush_itimer_signals(void)
276 {
277 	struct task_struct *tsk = current;
278 	unsigned long flags;
279 
280 	spin_lock_irqsave(&tsk->sighand->siglock, flags);
281 	__flush_itimer_signals(&tsk->pending);
282 	__flush_itimer_signals(&tsk->signal->shared_pending);
283 	spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
284 }
285 
286 void ignore_signals(struct task_struct *t)
287 {
288 	int i;
289 
290 	for (i = 0; i < _NSIG; ++i)
291 		t->sighand->action[i].sa.sa_handler = SIG_IGN;
292 
293 	flush_signals(t);
294 }
295 
296 /*
297  * Flush all handlers for a task.
298  */
299 
300 void
301 flush_signal_handlers(struct task_struct *t, int force_default)
302 {
303 	int i;
304 	struct k_sigaction *ka = &t->sighand->action[0];
305 	for (i = _NSIG ; i != 0 ; i--) {
306 		if (force_default || ka->sa.sa_handler != SIG_IGN)
307 			ka->sa.sa_handler = SIG_DFL;
308 		ka->sa.sa_flags = 0;
309 		sigemptyset(&ka->sa.sa_mask);
310 		ka++;
311 	}
312 }
313 
314 int unhandled_signal(struct task_struct *tsk, int sig)
315 {
316 	void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
317 	if (is_global_init(tsk))
318 		return 1;
319 	if (handler != SIG_IGN && handler != SIG_DFL)
320 		return 0;
321 	return !tracehook_consider_fatal_signal(tsk, sig, handler);
322 }
323 
324 
325 /* Notify the system that a driver wants to block all signals for this
326  * process, and wants to be notified if any signals at all were to be
327  * sent/acted upon.  If the notifier routine returns non-zero, then the
328  * signal will be acted upon after all.  If the notifier routine returns 0,
329  * then then signal will be blocked.  Only one block per process is
330  * allowed.  priv is a pointer to private data that the notifier routine
331  * can use to determine if the signal should be blocked or not.  */
332 
333 void
334 block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
335 {
336 	unsigned long flags;
337 
338 	spin_lock_irqsave(&current->sighand->siglock, flags);
339 	current->notifier_mask = mask;
340 	current->notifier_data = priv;
341 	current->notifier = notifier;
342 	spin_unlock_irqrestore(&current->sighand->siglock, flags);
343 }
344 
345 /* Notify the system that blocking has ended. */
346 
347 void
348 unblock_all_signals(void)
349 {
350 	unsigned long flags;
351 
352 	spin_lock_irqsave(&current->sighand->siglock, flags);
353 	current->notifier = NULL;
354 	current->notifier_data = NULL;
355 	recalc_sigpending();
356 	spin_unlock_irqrestore(&current->sighand->siglock, flags);
357 }
358 
359 static void collect_signal(int sig, struct sigpending *list, siginfo_t *info)
360 {
361 	struct sigqueue *q, *first = NULL;
362 
363 	/*
364 	 * Collect the siginfo appropriate to this signal.  Check if
365 	 * there is another siginfo for the same signal.
366 	*/
367 	list_for_each_entry(q, &list->list, list) {
368 		if (q->info.si_signo == sig) {
369 			if (first)
370 				goto still_pending;
371 			first = q;
372 		}
373 	}
374 
375 	sigdelset(&list->signal, sig);
376 
377 	if (first) {
378 still_pending:
379 		list_del_init(&first->list);
380 		copy_siginfo(info, &first->info);
381 		__sigqueue_free(first);
382 	} else {
383 		/* Ok, it wasn't in the queue.  This must be
384 		   a fast-pathed signal or we must have been
385 		   out of queue space.  So zero out the info.
386 		 */
387 		info->si_signo = sig;
388 		info->si_errno = 0;
389 		info->si_code = 0;
390 		info->si_pid = 0;
391 		info->si_uid = 0;
392 	}
393 }
394 
395 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
396 			siginfo_t *info)
397 {
398 	int sig = next_signal(pending, mask);
399 
400 	if (sig) {
401 		if (current->notifier) {
402 			if (sigismember(current->notifier_mask, sig)) {
403 				if (!(current->notifier)(current->notifier_data)) {
404 					clear_thread_flag(TIF_SIGPENDING);
405 					return 0;
406 				}
407 			}
408 		}
409 
410 		collect_signal(sig, pending, info);
411 	}
412 
413 	return sig;
414 }
415 
416 /*
417  * Dequeue a signal and return the element to the caller, which is
418  * expected to free it.
419  *
420  * All callers have to hold the siglock.
421  */
422 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
423 {
424 	int signr;
425 
426 	/* We only dequeue private signals from ourselves, we don't let
427 	 * signalfd steal them
428 	 */
429 	signr = __dequeue_signal(&tsk->pending, mask, info);
430 	if (!signr) {
431 		signr = __dequeue_signal(&tsk->signal->shared_pending,
432 					 mask, info);
433 		/*
434 		 * itimer signal ?
435 		 *
436 		 * itimers are process shared and we restart periodic
437 		 * itimers in the signal delivery path to prevent DoS
438 		 * attacks in the high resolution timer case. This is
439 		 * compliant with the old way of self restarting
440 		 * itimers, as the SIGALRM is a legacy signal and only
441 		 * queued once. Changing the restart behaviour to
442 		 * restart the timer in the signal dequeue path is
443 		 * reducing the timer noise on heavy loaded !highres
444 		 * systems too.
445 		 */
446 		if (unlikely(signr == SIGALRM)) {
447 			struct hrtimer *tmr = &tsk->signal->real_timer;
448 
449 			if (!hrtimer_is_queued(tmr) &&
450 			    tsk->signal->it_real_incr.tv64 != 0) {
451 				hrtimer_forward(tmr, tmr->base->get_time(),
452 						tsk->signal->it_real_incr);
453 				hrtimer_restart(tmr);
454 			}
455 		}
456 	}
457 
458 	recalc_sigpending();
459 	if (!signr)
460 		return 0;
461 
462 	if (unlikely(sig_kernel_stop(signr))) {
463 		/*
464 		 * Set a marker that we have dequeued a stop signal.  Our
465 		 * caller might release the siglock and then the pending
466 		 * stop signal it is about to process is no longer in the
467 		 * pending bitmasks, but must still be cleared by a SIGCONT
468 		 * (and overruled by a SIGKILL).  So those cases clear this
469 		 * shared flag after we've set it.  Note that this flag may
470 		 * remain set after the signal we return is ignored or
471 		 * handled.  That doesn't matter because its only purpose
472 		 * is to alert stop-signal processing code when another
473 		 * processor has come along and cleared the flag.
474 		 */
475 		tsk->signal->flags |= SIGNAL_STOP_DEQUEUED;
476 	}
477 	if ((info->si_code & __SI_MASK) == __SI_TIMER && info->si_sys_private) {
478 		/*
479 		 * Release the siglock to ensure proper locking order
480 		 * of timer locks outside of siglocks.  Note, we leave
481 		 * irqs disabled here, since the posix-timers code is
482 		 * about to disable them again anyway.
483 		 */
484 		spin_unlock(&tsk->sighand->siglock);
485 		do_schedule_next_timer(info);
486 		spin_lock(&tsk->sighand->siglock);
487 	}
488 	return signr;
489 }
490 
491 /*
492  * Tell a process that it has a new active signal..
493  *
494  * NOTE! we rely on the previous spin_lock to
495  * lock interrupts for us! We can only be called with
496  * "siglock" held, and the local interrupt must
497  * have been disabled when that got acquired!
498  *
499  * No need to set need_resched since signal event passing
500  * goes through ->blocked
501  */
502 void signal_wake_up(struct task_struct *t, int resume)
503 {
504 	unsigned int mask;
505 
506 	set_tsk_thread_flag(t, TIF_SIGPENDING);
507 
508 	/*
509 	 * For SIGKILL, we want to wake it up in the stopped/traced/killable
510 	 * case. We don't check t->state here because there is a race with it
511 	 * executing another processor and just now entering stopped state.
512 	 * By using wake_up_state, we ensure the process will wake up and
513 	 * handle its death signal.
514 	 */
515 	mask = TASK_INTERRUPTIBLE;
516 	if (resume)
517 		mask |= TASK_WAKEKILL;
518 	if (!wake_up_state(t, mask))
519 		kick_process(t);
520 }
521 
522 /*
523  * Remove signals in mask from the pending set and queue.
524  * Returns 1 if any signals were found.
525  *
526  * All callers must be holding the siglock.
527  *
528  * This version takes a sigset mask and looks at all signals,
529  * not just those in the first mask word.
530  */
531 static int rm_from_queue_full(sigset_t *mask, struct sigpending *s)
532 {
533 	struct sigqueue *q, *n;
534 	sigset_t m;
535 
536 	sigandsets(&m, mask, &s->signal);
537 	if (sigisemptyset(&m))
538 		return 0;
539 
540 	signandsets(&s->signal, &s->signal, mask);
541 	list_for_each_entry_safe(q, n, &s->list, list) {
542 		if (sigismember(mask, q->info.si_signo)) {
543 			list_del_init(&q->list);
544 			__sigqueue_free(q);
545 		}
546 	}
547 	return 1;
548 }
549 /*
550  * Remove signals in mask from the pending set and queue.
551  * Returns 1 if any signals were found.
552  *
553  * All callers must be holding the siglock.
554  */
555 static int rm_from_queue(unsigned long mask, struct sigpending *s)
556 {
557 	struct sigqueue *q, *n;
558 
559 	if (!sigtestsetmask(&s->signal, mask))
560 		return 0;
561 
562 	sigdelsetmask(&s->signal, mask);
563 	list_for_each_entry_safe(q, n, &s->list, list) {
564 		if (q->info.si_signo < SIGRTMIN &&
565 		    (mask & sigmask(q->info.si_signo))) {
566 			list_del_init(&q->list);
567 			__sigqueue_free(q);
568 		}
569 	}
570 	return 1;
571 }
572 
573 /*
574  * Bad permissions for sending the signal
575  * - the caller must hold at least the RCU read lock
576  */
577 static int check_kill_permission(int sig, struct siginfo *info,
578 				 struct task_struct *t)
579 {
580 	const struct cred *cred = current_cred(), *tcred;
581 	struct pid *sid;
582 	int error;
583 
584 	if (!valid_signal(sig))
585 		return -EINVAL;
586 
587 	if (info != SEND_SIG_NOINFO && (is_si_special(info) || SI_FROMKERNEL(info)))
588 		return 0;
589 
590 	error = audit_signal_info(sig, t); /* Let audit system see the signal */
591 	if (error)
592 		return error;
593 
594 	tcred = __task_cred(t);
595 	if ((cred->euid ^ tcred->suid) &&
596 	    (cred->euid ^ tcred->uid) &&
597 	    (cred->uid  ^ tcred->suid) &&
598 	    (cred->uid  ^ tcred->uid) &&
599 	    !capable(CAP_KILL)) {
600 		switch (sig) {
601 		case SIGCONT:
602 			sid = task_session(t);
603 			/*
604 			 * We don't return the error if sid == NULL. The
605 			 * task was unhashed, the caller must notice this.
606 			 */
607 			if (!sid || sid == task_session(current))
608 				break;
609 		default:
610 			return -EPERM;
611 		}
612 	}
613 
614 	return security_task_kill(t, info, sig, 0);
615 }
616 
617 /*
618  * Handle magic process-wide effects of stop/continue signals. Unlike
619  * the signal actions, these happen immediately at signal-generation
620  * time regardless of blocking, ignoring, or handling.  This does the
621  * actual continuing for SIGCONT, but not the actual stopping for stop
622  * signals. The process stop is done as a signal action for SIG_DFL.
623  *
624  * Returns true if the signal should be actually delivered, otherwise
625  * it should be dropped.
626  */
627 static int prepare_signal(int sig, struct task_struct *p)
628 {
629 	struct signal_struct *signal = p->signal;
630 	struct task_struct *t;
631 
632 	if (unlikely(signal->flags & SIGNAL_GROUP_EXIT)) {
633 		/*
634 		 * The process is in the middle of dying, nothing to do.
635 		 */
636 	} else if (sig_kernel_stop(sig)) {
637 		/*
638 		 * This is a stop signal.  Remove SIGCONT from all queues.
639 		 */
640 		rm_from_queue(sigmask(SIGCONT), &signal->shared_pending);
641 		t = p;
642 		do {
643 			rm_from_queue(sigmask(SIGCONT), &t->pending);
644 		} while_each_thread(p, t);
645 	} else if (sig == SIGCONT) {
646 		unsigned int why;
647 		/*
648 		 * Remove all stop signals from all queues,
649 		 * and wake all threads.
650 		 */
651 		rm_from_queue(SIG_KERNEL_STOP_MASK, &signal->shared_pending);
652 		t = p;
653 		do {
654 			unsigned int state;
655 			rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
656 			/*
657 			 * If there is a handler for SIGCONT, we must make
658 			 * sure that no thread returns to user mode before
659 			 * we post the signal, in case it was the only
660 			 * thread eligible to run the signal handler--then
661 			 * it must not do anything between resuming and
662 			 * running the handler.  With the TIF_SIGPENDING
663 			 * flag set, the thread will pause and acquire the
664 			 * siglock that we hold now and until we've queued
665 			 * the pending signal.
666 			 *
667 			 * Wake up the stopped thread _after_ setting
668 			 * TIF_SIGPENDING
669 			 */
670 			state = __TASK_STOPPED;
671 			if (sig_user_defined(t, SIGCONT) && !sigismember(&t->blocked, SIGCONT)) {
672 				set_tsk_thread_flag(t, TIF_SIGPENDING);
673 				state |= TASK_INTERRUPTIBLE;
674 			}
675 			wake_up_state(t, state);
676 		} while_each_thread(p, t);
677 
678 		/*
679 		 * Notify the parent with CLD_CONTINUED if we were stopped.
680 		 *
681 		 * If we were in the middle of a group stop, we pretend it
682 		 * was already finished, and then continued. Since SIGCHLD
683 		 * doesn't queue we report only CLD_STOPPED, as if the next
684 		 * CLD_CONTINUED was dropped.
685 		 */
686 		why = 0;
687 		if (signal->flags & SIGNAL_STOP_STOPPED)
688 			why |= SIGNAL_CLD_CONTINUED;
689 		else if (signal->group_stop_count)
690 			why |= SIGNAL_CLD_STOPPED;
691 
692 		if (why) {
693 			/*
694 			 * The first thread which returns from finish_stop()
695 			 * will take ->siglock, notice SIGNAL_CLD_MASK, and
696 			 * notify its parent. See get_signal_to_deliver().
697 			 */
698 			signal->flags = why | SIGNAL_STOP_CONTINUED;
699 			signal->group_stop_count = 0;
700 			signal->group_exit_code = 0;
701 		} else {
702 			/*
703 			 * We are not stopped, but there could be a stop
704 			 * signal in the middle of being processed after
705 			 * being removed from the queue.  Clear that too.
706 			 */
707 			signal->flags &= ~SIGNAL_STOP_DEQUEUED;
708 		}
709 	}
710 
711 	return !sig_ignored(p, sig);
712 }
713 
714 /*
715  * Test if P wants to take SIG.  After we've checked all threads with this,
716  * it's equivalent to finding no threads not blocking SIG.  Any threads not
717  * blocking SIG were ruled out because they are not running and already
718  * have pending signals.  Such threads will dequeue from the shared queue
719  * as soon as they're available, so putting the signal on the shared queue
720  * will be equivalent to sending it to one such thread.
721  */
722 static inline int wants_signal(int sig, struct task_struct *p)
723 {
724 	if (sigismember(&p->blocked, sig))
725 		return 0;
726 	if (p->flags & PF_EXITING)
727 		return 0;
728 	if (sig == SIGKILL)
729 		return 1;
730 	if (task_is_stopped_or_traced(p))
731 		return 0;
732 	return task_curr(p) || !signal_pending(p);
733 }
734 
735 static void complete_signal(int sig, struct task_struct *p, int group)
736 {
737 	struct signal_struct *signal = p->signal;
738 	struct task_struct *t;
739 
740 	/*
741 	 * Now find a thread we can wake up to take the signal off the queue.
742 	 *
743 	 * If the main thread wants the signal, it gets first crack.
744 	 * Probably the least surprising to the average bear.
745 	 */
746 	if (wants_signal(sig, p))
747 		t = p;
748 	else if (!group || thread_group_empty(p))
749 		/*
750 		 * There is just one thread and it does not need to be woken.
751 		 * It will dequeue unblocked signals before it runs again.
752 		 */
753 		return;
754 	else {
755 		/*
756 		 * Otherwise try to find a suitable thread.
757 		 */
758 		t = signal->curr_target;
759 		while (!wants_signal(sig, t)) {
760 			t = next_thread(t);
761 			if (t == signal->curr_target)
762 				/*
763 				 * No thread needs to be woken.
764 				 * Any eligible threads will see
765 				 * the signal in the queue soon.
766 				 */
767 				return;
768 		}
769 		signal->curr_target = t;
770 	}
771 
772 	/*
773 	 * Found a killable thread.  If the signal will be fatal,
774 	 * then start taking the whole group down immediately.
775 	 */
776 	if (sig_fatal(p, sig) &&
777 	    !(signal->flags & (SIGNAL_UNKILLABLE | SIGNAL_GROUP_EXIT)) &&
778 	    !sigismember(&t->real_blocked, sig) &&
779 	    (sig == SIGKILL ||
780 	     !tracehook_consider_fatal_signal(t, sig, SIG_DFL))) {
781 		/*
782 		 * This signal will be fatal to the whole group.
783 		 */
784 		if (!sig_kernel_coredump(sig)) {
785 			/*
786 			 * Start a group exit and wake everybody up.
787 			 * This way we don't have other threads
788 			 * running and doing things after a slower
789 			 * thread has the fatal signal pending.
790 			 */
791 			signal->flags = SIGNAL_GROUP_EXIT;
792 			signal->group_exit_code = sig;
793 			signal->group_stop_count = 0;
794 			t = p;
795 			do {
796 				sigaddset(&t->pending.signal, SIGKILL);
797 				signal_wake_up(t, 1);
798 			} while_each_thread(p, t);
799 			return;
800 		}
801 	}
802 
803 	/*
804 	 * The signal is already in the shared-pending queue.
805 	 * Tell the chosen thread to wake up and dequeue it.
806 	 */
807 	signal_wake_up(t, sig == SIGKILL);
808 	return;
809 }
810 
811 static inline int legacy_queue(struct sigpending *signals, int sig)
812 {
813 	return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
814 }
815 
816 static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
817 			int group)
818 {
819 	struct sigpending *pending;
820 	struct sigqueue *q;
821 
822 	trace_sched_signal_send(sig, t);
823 
824 	assert_spin_locked(&t->sighand->siglock);
825 	if (!prepare_signal(sig, t))
826 		return 0;
827 
828 	pending = group ? &t->signal->shared_pending : &t->pending;
829 	/*
830 	 * Short-circuit ignored signals and support queuing
831 	 * exactly one non-rt signal, so that we can get more
832 	 * detailed information about the cause of the signal.
833 	 */
834 	if (legacy_queue(pending, sig))
835 		return 0;
836 	/*
837 	 * fast-pathed signals for kernel-internal things like SIGSTOP
838 	 * or SIGKILL.
839 	 */
840 	if (info == SEND_SIG_FORCED)
841 		goto out_set;
842 
843 	/* Real-time signals must be queued if sent by sigqueue, or
844 	   some other real-time mechanism.  It is implementation
845 	   defined whether kill() does so.  We attempt to do so, on
846 	   the principle of least surprise, but since kill is not
847 	   allowed to fail with EAGAIN when low on memory we just
848 	   make sure at least one signal gets delivered and don't
849 	   pass on the info struct.  */
850 
851 	q = __sigqueue_alloc(t, GFP_ATOMIC, (sig < SIGRTMIN &&
852 					     (is_si_special(info) ||
853 					      info->si_code >= 0)));
854 	if (q) {
855 		list_add_tail(&q->list, &pending->list);
856 		switch ((unsigned long) info) {
857 		case (unsigned long) SEND_SIG_NOINFO:
858 			q->info.si_signo = sig;
859 			q->info.si_errno = 0;
860 			q->info.si_code = SI_USER;
861 			q->info.si_pid = task_tgid_nr_ns(current,
862 							task_active_pid_ns(t));
863 			q->info.si_uid = current_uid();
864 			break;
865 		case (unsigned long) SEND_SIG_PRIV:
866 			q->info.si_signo = sig;
867 			q->info.si_errno = 0;
868 			q->info.si_code = SI_KERNEL;
869 			q->info.si_pid = 0;
870 			q->info.si_uid = 0;
871 			break;
872 		default:
873 			copy_siginfo(&q->info, info);
874 			break;
875 		}
876 	} else if (!is_si_special(info)) {
877 		if (sig >= SIGRTMIN && info->si_code != SI_USER)
878 		/*
879 		 * Queue overflow, abort.  We may abort if the signal was rt
880 		 * and sent by user using something other than kill().
881 		 */
882 			return -EAGAIN;
883 	}
884 
885 out_set:
886 	signalfd_notify(t, sig);
887 	sigaddset(&pending->signal, sig);
888 	complete_signal(sig, t, group);
889 	return 0;
890 }
891 
892 int print_fatal_signals;
893 
894 static void print_fatal_signal(struct pt_regs *regs, int signr)
895 {
896 	printk("%s/%d: potentially unexpected fatal signal %d.\n",
897 		current->comm, task_pid_nr(current), signr);
898 
899 #if defined(__i386__) && !defined(__arch_um__)
900 	printk("code at %08lx: ", regs->ip);
901 	{
902 		int i;
903 		for (i = 0; i < 16; i++) {
904 			unsigned char insn;
905 
906 			__get_user(insn, (unsigned char *)(regs->ip + i));
907 			printk("%02x ", insn);
908 		}
909 	}
910 #endif
911 	printk("\n");
912 	show_regs(regs);
913 }
914 
915 static int __init setup_print_fatal_signals(char *str)
916 {
917 	get_option (&str, &print_fatal_signals);
918 
919 	return 1;
920 }
921 
922 __setup("print-fatal-signals=", setup_print_fatal_signals);
923 
924 int
925 __group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
926 {
927 	return send_signal(sig, info, p, 1);
928 }
929 
930 static int
931 specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
932 {
933 	return send_signal(sig, info, t, 0);
934 }
935 
936 /*
937  * Force a signal that the process can't ignore: if necessary
938  * we unblock the signal and change any SIG_IGN to SIG_DFL.
939  *
940  * Note: If we unblock the signal, we always reset it to SIG_DFL,
941  * since we do not want to have a signal handler that was blocked
942  * be invoked when user space had explicitly blocked it.
943  *
944  * We don't want to have recursive SIGSEGV's etc, for example,
945  * that is why we also clear SIGNAL_UNKILLABLE.
946  */
947 int
948 force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
949 {
950 	unsigned long int flags;
951 	int ret, blocked, ignored;
952 	struct k_sigaction *action;
953 
954 	spin_lock_irqsave(&t->sighand->siglock, flags);
955 	action = &t->sighand->action[sig-1];
956 	ignored = action->sa.sa_handler == SIG_IGN;
957 	blocked = sigismember(&t->blocked, sig);
958 	if (blocked || ignored) {
959 		action->sa.sa_handler = SIG_DFL;
960 		if (blocked) {
961 			sigdelset(&t->blocked, sig);
962 			recalc_sigpending_and_wake(t);
963 		}
964 	}
965 	if (action->sa.sa_handler == SIG_DFL)
966 		t->signal->flags &= ~SIGNAL_UNKILLABLE;
967 	ret = specific_send_sig_info(sig, info, t);
968 	spin_unlock_irqrestore(&t->sighand->siglock, flags);
969 
970 	return ret;
971 }
972 
973 void
974 force_sig_specific(int sig, struct task_struct *t)
975 {
976 	force_sig_info(sig, SEND_SIG_FORCED, t);
977 }
978 
979 /*
980  * Nuke all other threads in the group.
981  */
982 void zap_other_threads(struct task_struct *p)
983 {
984 	struct task_struct *t;
985 
986 	p->signal->group_stop_count = 0;
987 
988 	for (t = next_thread(p); t != p; t = next_thread(t)) {
989 		/*
990 		 * Don't bother with already dead threads
991 		 */
992 		if (t->exit_state)
993 			continue;
994 
995 		/* SIGKILL will be handled before any pending SIGSTOP */
996 		sigaddset(&t->pending.signal, SIGKILL);
997 		signal_wake_up(t, 1);
998 	}
999 }
1000 
1001 int __fatal_signal_pending(struct task_struct *tsk)
1002 {
1003 	return sigismember(&tsk->pending.signal, SIGKILL);
1004 }
1005 EXPORT_SYMBOL(__fatal_signal_pending);
1006 
1007 struct sighand_struct *lock_task_sighand(struct task_struct *tsk, unsigned long *flags)
1008 {
1009 	struct sighand_struct *sighand;
1010 
1011 	rcu_read_lock();
1012 	for (;;) {
1013 		sighand = rcu_dereference(tsk->sighand);
1014 		if (unlikely(sighand == NULL))
1015 			break;
1016 
1017 		spin_lock_irqsave(&sighand->siglock, *flags);
1018 		if (likely(sighand == tsk->sighand))
1019 			break;
1020 		spin_unlock_irqrestore(&sighand->siglock, *flags);
1021 	}
1022 	rcu_read_unlock();
1023 
1024 	return sighand;
1025 }
1026 
1027 /*
1028  * send signal info to all the members of a group
1029  * - the caller must hold the RCU read lock at least
1030  */
1031 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1032 {
1033 	unsigned long flags;
1034 	int ret;
1035 
1036 	ret = check_kill_permission(sig, info, p);
1037 
1038 	if (!ret && sig) {
1039 		ret = -ESRCH;
1040 		if (lock_task_sighand(p, &flags)) {
1041 			ret = __group_send_sig_info(sig, info, p);
1042 			unlock_task_sighand(p, &flags);
1043 		}
1044 	}
1045 
1046 	return ret;
1047 }
1048 
1049 /*
1050  * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1051  * control characters do (^C, ^Z etc)
1052  * - the caller must hold at least a readlock on tasklist_lock
1053  */
1054 int __kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
1055 {
1056 	struct task_struct *p = NULL;
1057 	int retval, success;
1058 
1059 	success = 0;
1060 	retval = -ESRCH;
1061 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1062 		int err = group_send_sig_info(sig, info, p);
1063 		success |= !err;
1064 		retval = err;
1065 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1066 	return success ? 0 : retval;
1067 }
1068 
1069 int kill_pid_info(int sig, struct siginfo *info, struct pid *pid)
1070 {
1071 	int error = -ESRCH;
1072 	struct task_struct *p;
1073 
1074 	rcu_read_lock();
1075 retry:
1076 	p = pid_task(pid, PIDTYPE_PID);
1077 	if (p) {
1078 		error = group_send_sig_info(sig, info, p);
1079 		if (unlikely(error == -ESRCH))
1080 			/*
1081 			 * The task was unhashed in between, try again.
1082 			 * If it is dead, pid_task() will return NULL,
1083 			 * if we race with de_thread() it will find the
1084 			 * new leader.
1085 			 */
1086 			goto retry;
1087 	}
1088 	rcu_read_unlock();
1089 
1090 	return error;
1091 }
1092 
1093 int
1094 kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1095 {
1096 	int error;
1097 	rcu_read_lock();
1098 	error = kill_pid_info(sig, info, find_vpid(pid));
1099 	rcu_read_unlock();
1100 	return error;
1101 }
1102 
1103 /* like kill_pid_info(), but doesn't use uid/euid of "current" */
1104 int kill_pid_info_as_uid(int sig, struct siginfo *info, struct pid *pid,
1105 		      uid_t uid, uid_t euid, u32 secid)
1106 {
1107 	int ret = -EINVAL;
1108 	struct task_struct *p;
1109 	const struct cred *pcred;
1110 
1111 	if (!valid_signal(sig))
1112 		return ret;
1113 
1114 	read_lock(&tasklist_lock);
1115 	p = pid_task(pid, PIDTYPE_PID);
1116 	if (!p) {
1117 		ret = -ESRCH;
1118 		goto out_unlock;
1119 	}
1120 	pcred = __task_cred(p);
1121 	if ((info == SEND_SIG_NOINFO ||
1122 	     (!is_si_special(info) && SI_FROMUSER(info))) &&
1123 	    euid != pcred->suid && euid != pcred->uid &&
1124 	    uid  != pcred->suid && uid  != pcred->uid) {
1125 		ret = -EPERM;
1126 		goto out_unlock;
1127 	}
1128 	ret = security_task_kill(p, info, sig, secid);
1129 	if (ret)
1130 		goto out_unlock;
1131 	if (sig && p->sighand) {
1132 		unsigned long flags;
1133 		spin_lock_irqsave(&p->sighand->siglock, flags);
1134 		ret = __group_send_sig_info(sig, info, p);
1135 		spin_unlock_irqrestore(&p->sighand->siglock, flags);
1136 	}
1137 out_unlock:
1138 	read_unlock(&tasklist_lock);
1139 	return ret;
1140 }
1141 EXPORT_SYMBOL_GPL(kill_pid_info_as_uid);
1142 
1143 /*
1144  * kill_something_info() interprets pid in interesting ways just like kill(2).
1145  *
1146  * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1147  * is probably wrong.  Should make it like BSD or SYSV.
1148  */
1149 
1150 static int kill_something_info(int sig, struct siginfo *info, pid_t pid)
1151 {
1152 	int ret;
1153 
1154 	if (pid > 0) {
1155 		rcu_read_lock();
1156 		ret = kill_pid_info(sig, info, find_vpid(pid));
1157 		rcu_read_unlock();
1158 		return ret;
1159 	}
1160 
1161 	read_lock(&tasklist_lock);
1162 	if (pid != -1) {
1163 		ret = __kill_pgrp_info(sig, info,
1164 				pid ? find_vpid(-pid) : task_pgrp(current));
1165 	} else {
1166 		int retval = 0, count = 0;
1167 		struct task_struct * p;
1168 
1169 		for_each_process(p) {
1170 			if (task_pid_vnr(p) > 1 &&
1171 					!same_thread_group(p, current)) {
1172 				int err = group_send_sig_info(sig, info, p);
1173 				++count;
1174 				if (err != -EPERM)
1175 					retval = err;
1176 			}
1177 		}
1178 		ret = count ? retval : -ESRCH;
1179 	}
1180 	read_unlock(&tasklist_lock);
1181 
1182 	return ret;
1183 }
1184 
1185 /*
1186  * These are for backward compatibility with the rest of the kernel source.
1187  */
1188 
1189 /*
1190  * The caller must ensure the task can't exit.
1191  */
1192 int
1193 send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1194 {
1195 	int ret;
1196 	unsigned long flags;
1197 
1198 	/*
1199 	 * Make sure legacy kernel users don't send in bad values
1200 	 * (normal paths check this in check_kill_permission).
1201 	 */
1202 	if (!valid_signal(sig))
1203 		return -EINVAL;
1204 
1205 	spin_lock_irqsave(&p->sighand->siglock, flags);
1206 	ret = specific_send_sig_info(sig, info, p);
1207 	spin_unlock_irqrestore(&p->sighand->siglock, flags);
1208 	return ret;
1209 }
1210 
1211 #define __si_special(priv) \
1212 	((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1213 
1214 int
1215 send_sig(int sig, struct task_struct *p, int priv)
1216 {
1217 	return send_sig_info(sig, __si_special(priv), p);
1218 }
1219 
1220 void
1221 force_sig(int sig, struct task_struct *p)
1222 {
1223 	force_sig_info(sig, SEND_SIG_PRIV, p);
1224 }
1225 
1226 /*
1227  * When things go south during signal handling, we
1228  * will force a SIGSEGV. And if the signal that caused
1229  * the problem was already a SIGSEGV, we'll want to
1230  * make sure we don't even try to deliver the signal..
1231  */
1232 int
1233 force_sigsegv(int sig, struct task_struct *p)
1234 {
1235 	if (sig == SIGSEGV) {
1236 		unsigned long flags;
1237 		spin_lock_irqsave(&p->sighand->siglock, flags);
1238 		p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1239 		spin_unlock_irqrestore(&p->sighand->siglock, flags);
1240 	}
1241 	force_sig(SIGSEGV, p);
1242 	return 0;
1243 }
1244 
1245 int kill_pgrp(struct pid *pid, int sig, int priv)
1246 {
1247 	int ret;
1248 
1249 	read_lock(&tasklist_lock);
1250 	ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1251 	read_unlock(&tasklist_lock);
1252 
1253 	return ret;
1254 }
1255 EXPORT_SYMBOL(kill_pgrp);
1256 
1257 int kill_pid(struct pid *pid, int sig, int priv)
1258 {
1259 	return kill_pid_info(sig, __si_special(priv), pid);
1260 }
1261 EXPORT_SYMBOL(kill_pid);
1262 
1263 /*
1264  * These functions support sending signals using preallocated sigqueue
1265  * structures.  This is needed "because realtime applications cannot
1266  * afford to lose notifications of asynchronous events, like timer
1267  * expirations or I/O completions".  In the case of Posix Timers
1268  * we allocate the sigqueue structure from the timer_create.  If this
1269  * allocation fails we are able to report the failure to the application
1270  * with an EAGAIN error.
1271  */
1272 
1273 struct sigqueue *sigqueue_alloc(void)
1274 {
1275 	struct sigqueue *q;
1276 
1277 	if ((q = __sigqueue_alloc(current, GFP_KERNEL, 0)))
1278 		q->flags |= SIGQUEUE_PREALLOC;
1279 	return(q);
1280 }
1281 
1282 void sigqueue_free(struct sigqueue *q)
1283 {
1284 	unsigned long flags;
1285 	spinlock_t *lock = &current->sighand->siglock;
1286 
1287 	BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1288 	/*
1289 	 * We must hold ->siglock while testing q->list
1290 	 * to serialize with collect_signal() or with
1291 	 * __exit_signal()->flush_sigqueue().
1292 	 */
1293 	spin_lock_irqsave(lock, flags);
1294 	q->flags &= ~SIGQUEUE_PREALLOC;
1295 	/*
1296 	 * If it is queued it will be freed when dequeued,
1297 	 * like the "regular" sigqueue.
1298 	 */
1299 	if (!list_empty(&q->list))
1300 		q = NULL;
1301 	spin_unlock_irqrestore(lock, flags);
1302 
1303 	if (q)
1304 		__sigqueue_free(q);
1305 }
1306 
1307 int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group)
1308 {
1309 	int sig = q->info.si_signo;
1310 	struct sigpending *pending;
1311 	unsigned long flags;
1312 	int ret;
1313 
1314 	BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1315 
1316 	ret = -1;
1317 	if (!likely(lock_task_sighand(t, &flags)))
1318 		goto ret;
1319 
1320 	ret = 1; /* the signal is ignored */
1321 	if (!prepare_signal(sig, t))
1322 		goto out;
1323 
1324 	ret = 0;
1325 	if (unlikely(!list_empty(&q->list))) {
1326 		/*
1327 		 * If an SI_TIMER entry is already queue just increment
1328 		 * the overrun count.
1329 		 */
1330 		BUG_ON(q->info.si_code != SI_TIMER);
1331 		q->info.si_overrun++;
1332 		goto out;
1333 	}
1334 	q->info.si_overrun = 0;
1335 
1336 	signalfd_notify(t, sig);
1337 	pending = group ? &t->signal->shared_pending : &t->pending;
1338 	list_add_tail(&q->list, &pending->list);
1339 	sigaddset(&pending->signal, sig);
1340 	complete_signal(sig, t, group);
1341 out:
1342 	unlock_task_sighand(t, &flags);
1343 ret:
1344 	return ret;
1345 }
1346 
1347 /*
1348  * Wake up any threads in the parent blocked in wait* syscalls.
1349  */
1350 static inline void __wake_up_parent(struct task_struct *p,
1351 				    struct task_struct *parent)
1352 {
1353 	wake_up_interruptible_sync(&parent->signal->wait_chldexit);
1354 }
1355 
1356 /*
1357  * Let a parent know about the death of a child.
1358  * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1359  *
1360  * Returns -1 if our parent ignored us and so we've switched to
1361  * self-reaping, or else @sig.
1362  */
1363 int do_notify_parent(struct task_struct *tsk, int sig)
1364 {
1365 	struct siginfo info;
1366 	unsigned long flags;
1367 	struct sighand_struct *psig;
1368 	struct task_cputime cputime;
1369 	int ret = sig;
1370 
1371 	BUG_ON(sig == -1);
1372 
1373  	/* do_notify_parent_cldstop should have been called instead.  */
1374  	BUG_ON(task_is_stopped_or_traced(tsk));
1375 
1376 	BUG_ON(!tsk->ptrace &&
1377 	       (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1378 
1379 	info.si_signo = sig;
1380 	info.si_errno = 0;
1381 	/*
1382 	 * we are under tasklist_lock here so our parent is tied to
1383 	 * us and cannot exit and release its namespace.
1384 	 *
1385 	 * the only it can is to switch its nsproxy with sys_unshare,
1386 	 * bu uncharing pid namespaces is not allowed, so we'll always
1387 	 * see relevant namespace
1388 	 *
1389 	 * write_lock() currently calls preempt_disable() which is the
1390 	 * same as rcu_read_lock(), but according to Oleg, this is not
1391 	 * correct to rely on this
1392 	 */
1393 	rcu_read_lock();
1394 	info.si_pid = task_pid_nr_ns(tsk, tsk->parent->nsproxy->pid_ns);
1395 	info.si_uid = __task_cred(tsk)->uid;
1396 	rcu_read_unlock();
1397 
1398 	thread_group_cputime(tsk, &cputime);
1399 	info.si_utime = cputime_to_jiffies(cputime.utime);
1400 	info.si_stime = cputime_to_jiffies(cputime.stime);
1401 
1402 	info.si_status = tsk->exit_code & 0x7f;
1403 	if (tsk->exit_code & 0x80)
1404 		info.si_code = CLD_DUMPED;
1405 	else if (tsk->exit_code & 0x7f)
1406 		info.si_code = CLD_KILLED;
1407 	else {
1408 		info.si_code = CLD_EXITED;
1409 		info.si_status = tsk->exit_code >> 8;
1410 	}
1411 
1412 	psig = tsk->parent->sighand;
1413 	spin_lock_irqsave(&psig->siglock, flags);
1414 	if (!tsk->ptrace && sig == SIGCHLD &&
1415 	    (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1416 	     (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1417 		/*
1418 		 * We are exiting and our parent doesn't care.  POSIX.1
1419 		 * defines special semantics for setting SIGCHLD to SIG_IGN
1420 		 * or setting the SA_NOCLDWAIT flag: we should be reaped
1421 		 * automatically and not left for our parent's wait4 call.
1422 		 * Rather than having the parent do it as a magic kind of
1423 		 * signal handler, we just set this to tell do_exit that we
1424 		 * can be cleaned up without becoming a zombie.  Note that
1425 		 * we still call __wake_up_parent in this case, because a
1426 		 * blocked sys_wait4 might now return -ECHILD.
1427 		 *
1428 		 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1429 		 * is implementation-defined: we do (if you don't want
1430 		 * it, just use SIG_IGN instead).
1431 		 */
1432 		ret = tsk->exit_signal = -1;
1433 		if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1434 			sig = -1;
1435 	}
1436 	if (valid_signal(sig) && sig > 0)
1437 		__group_send_sig_info(sig, &info, tsk->parent);
1438 	__wake_up_parent(tsk, tsk->parent);
1439 	spin_unlock_irqrestore(&psig->siglock, flags);
1440 
1441 	return ret;
1442 }
1443 
1444 static void do_notify_parent_cldstop(struct task_struct *tsk, int why)
1445 {
1446 	struct siginfo info;
1447 	unsigned long flags;
1448 	struct task_struct *parent;
1449 	struct sighand_struct *sighand;
1450 
1451 	if (tsk->ptrace & PT_PTRACED)
1452 		parent = tsk->parent;
1453 	else {
1454 		tsk = tsk->group_leader;
1455 		parent = tsk->real_parent;
1456 	}
1457 
1458 	info.si_signo = SIGCHLD;
1459 	info.si_errno = 0;
1460 	/*
1461 	 * see comment in do_notify_parent() abot the following 3 lines
1462 	 */
1463 	rcu_read_lock();
1464 	info.si_pid = task_pid_nr_ns(tsk, tsk->parent->nsproxy->pid_ns);
1465 	info.si_uid = __task_cred(tsk)->uid;
1466 	rcu_read_unlock();
1467 
1468 	info.si_utime = cputime_to_clock_t(tsk->utime);
1469 	info.si_stime = cputime_to_clock_t(tsk->stime);
1470 
1471  	info.si_code = why;
1472  	switch (why) {
1473  	case CLD_CONTINUED:
1474  		info.si_status = SIGCONT;
1475  		break;
1476  	case CLD_STOPPED:
1477  		info.si_status = tsk->signal->group_exit_code & 0x7f;
1478  		break;
1479  	case CLD_TRAPPED:
1480  		info.si_status = tsk->exit_code & 0x7f;
1481  		break;
1482  	default:
1483  		BUG();
1484  	}
1485 
1486 	sighand = parent->sighand;
1487 	spin_lock_irqsave(&sighand->siglock, flags);
1488 	if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1489 	    !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1490 		__group_send_sig_info(SIGCHLD, &info, parent);
1491 	/*
1492 	 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1493 	 */
1494 	__wake_up_parent(tsk, parent);
1495 	spin_unlock_irqrestore(&sighand->siglock, flags);
1496 }
1497 
1498 static inline int may_ptrace_stop(void)
1499 {
1500 	if (!likely(current->ptrace & PT_PTRACED))
1501 		return 0;
1502 	/*
1503 	 * Are we in the middle of do_coredump?
1504 	 * If so and our tracer is also part of the coredump stopping
1505 	 * is a deadlock situation, and pointless because our tracer
1506 	 * is dead so don't allow us to stop.
1507 	 * If SIGKILL was already sent before the caller unlocked
1508 	 * ->siglock we must see ->core_state != NULL. Otherwise it
1509 	 * is safe to enter schedule().
1510 	 */
1511 	if (unlikely(current->mm->core_state) &&
1512 	    unlikely(current->mm == current->parent->mm))
1513 		return 0;
1514 
1515 	return 1;
1516 }
1517 
1518 /*
1519  * Return nonzero if there is a SIGKILL that should be waking us up.
1520  * Called with the siglock held.
1521  */
1522 static int sigkill_pending(struct task_struct *tsk)
1523 {
1524 	return	sigismember(&tsk->pending.signal, SIGKILL) ||
1525 		sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
1526 }
1527 
1528 /*
1529  * This must be called with current->sighand->siglock held.
1530  *
1531  * This should be the path for all ptrace stops.
1532  * We always set current->last_siginfo while stopped here.
1533  * That makes it a way to test a stopped process for
1534  * being ptrace-stopped vs being job-control-stopped.
1535  *
1536  * If we actually decide not to stop at all because the tracer
1537  * is gone, we keep current->exit_code unless clear_code.
1538  */
1539 static void ptrace_stop(int exit_code, int clear_code, siginfo_t *info)
1540 {
1541 	if (arch_ptrace_stop_needed(exit_code, info)) {
1542 		/*
1543 		 * The arch code has something special to do before a
1544 		 * ptrace stop.  This is allowed to block, e.g. for faults
1545 		 * on user stack pages.  We can't keep the siglock while
1546 		 * calling arch_ptrace_stop, so we must release it now.
1547 		 * To preserve proper semantics, we must do this before
1548 		 * any signal bookkeeping like checking group_stop_count.
1549 		 * Meanwhile, a SIGKILL could come in before we retake the
1550 		 * siglock.  That must prevent us from sleeping in TASK_TRACED.
1551 		 * So after regaining the lock, we must check for SIGKILL.
1552 		 */
1553 		spin_unlock_irq(&current->sighand->siglock);
1554 		arch_ptrace_stop(exit_code, info);
1555 		spin_lock_irq(&current->sighand->siglock);
1556 		if (sigkill_pending(current))
1557 			return;
1558 	}
1559 
1560 	/*
1561 	 * If there is a group stop in progress,
1562 	 * we must participate in the bookkeeping.
1563 	 */
1564 	if (current->signal->group_stop_count > 0)
1565 		--current->signal->group_stop_count;
1566 
1567 	current->last_siginfo = info;
1568 	current->exit_code = exit_code;
1569 
1570 	/* Let the debugger run.  */
1571 	__set_current_state(TASK_TRACED);
1572 	spin_unlock_irq(&current->sighand->siglock);
1573 	read_lock(&tasklist_lock);
1574 	if (may_ptrace_stop()) {
1575 		do_notify_parent_cldstop(current, CLD_TRAPPED);
1576 		read_unlock(&tasklist_lock);
1577 		schedule();
1578 	} else {
1579 		/*
1580 		 * By the time we got the lock, our tracer went away.
1581 		 * Don't drop the lock yet, another tracer may come.
1582 		 */
1583 		__set_current_state(TASK_RUNNING);
1584 		if (clear_code)
1585 			current->exit_code = 0;
1586 		read_unlock(&tasklist_lock);
1587 	}
1588 
1589 	/*
1590 	 * While in TASK_TRACED, we were considered "frozen enough".
1591 	 * Now that we woke up, it's crucial if we're supposed to be
1592 	 * frozen that we freeze now before running anything substantial.
1593 	 */
1594 	try_to_freeze();
1595 
1596 	/*
1597 	 * We are back.  Now reacquire the siglock before touching
1598 	 * last_siginfo, so that we are sure to have synchronized with
1599 	 * any signal-sending on another CPU that wants to examine it.
1600 	 */
1601 	spin_lock_irq(&current->sighand->siglock);
1602 	current->last_siginfo = NULL;
1603 
1604 	/*
1605 	 * Queued signals ignored us while we were stopped for tracing.
1606 	 * So check for any that we should take before resuming user mode.
1607 	 * This sets TIF_SIGPENDING, but never clears it.
1608 	 */
1609 	recalc_sigpending_tsk(current);
1610 }
1611 
1612 void ptrace_notify(int exit_code)
1613 {
1614 	siginfo_t info;
1615 
1616 	BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1617 
1618 	memset(&info, 0, sizeof info);
1619 	info.si_signo = SIGTRAP;
1620 	info.si_code = exit_code;
1621 	info.si_pid = task_pid_vnr(current);
1622 	info.si_uid = current_uid();
1623 
1624 	/* Let the debugger run.  */
1625 	spin_lock_irq(&current->sighand->siglock);
1626 	ptrace_stop(exit_code, 1, &info);
1627 	spin_unlock_irq(&current->sighand->siglock);
1628 }
1629 
1630 static void
1631 finish_stop(int stop_count)
1632 {
1633 	/*
1634 	 * If there are no other threads in the group, or if there is
1635 	 * a group stop in progress and we are the last to stop,
1636 	 * report to the parent.  When ptraced, every thread reports itself.
1637 	 */
1638 	if (tracehook_notify_jctl(stop_count == 0, CLD_STOPPED)) {
1639 		read_lock(&tasklist_lock);
1640 		do_notify_parent_cldstop(current, CLD_STOPPED);
1641 		read_unlock(&tasklist_lock);
1642 	}
1643 
1644 	do {
1645 		schedule();
1646 	} while (try_to_freeze());
1647 	/*
1648 	 * Now we don't run again until continued.
1649 	 */
1650 	current->exit_code = 0;
1651 }
1652 
1653 /*
1654  * This performs the stopping for SIGSTOP and other stop signals.
1655  * We have to stop all threads in the thread group.
1656  * Returns nonzero if we've actually stopped and released the siglock.
1657  * Returns zero if we didn't stop and still hold the siglock.
1658  */
1659 static int do_signal_stop(int signr)
1660 {
1661 	struct signal_struct *sig = current->signal;
1662 	int stop_count;
1663 
1664 	if (sig->group_stop_count > 0) {
1665 		/*
1666 		 * There is a group stop in progress.  We don't need to
1667 		 * start another one.
1668 		 */
1669 		stop_count = --sig->group_stop_count;
1670 	} else {
1671 		struct task_struct *t;
1672 
1673 		if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED) ||
1674 		    unlikely(signal_group_exit(sig)))
1675 			return 0;
1676 		/*
1677 		 * There is no group stop already in progress.
1678 		 * We must initiate one now.
1679 		 */
1680 		sig->group_exit_code = signr;
1681 
1682 		stop_count = 0;
1683 		for (t = next_thread(current); t != current; t = next_thread(t))
1684 			/*
1685 			 * Setting state to TASK_STOPPED for a group
1686 			 * stop is always done with the siglock held,
1687 			 * so this check has no races.
1688 			 */
1689 			if (!(t->flags & PF_EXITING) &&
1690 			    !task_is_stopped_or_traced(t)) {
1691 				stop_count++;
1692 				signal_wake_up(t, 0);
1693 			}
1694 		sig->group_stop_count = stop_count;
1695 	}
1696 
1697 	if (stop_count == 0)
1698 		sig->flags = SIGNAL_STOP_STOPPED;
1699 	current->exit_code = sig->group_exit_code;
1700 	__set_current_state(TASK_STOPPED);
1701 
1702 	spin_unlock_irq(&current->sighand->siglock);
1703 	finish_stop(stop_count);
1704 	return 1;
1705 }
1706 
1707 static int ptrace_signal(int signr, siginfo_t *info,
1708 			 struct pt_regs *regs, void *cookie)
1709 {
1710 	if (!(current->ptrace & PT_PTRACED))
1711 		return signr;
1712 
1713 	ptrace_signal_deliver(regs, cookie);
1714 
1715 	/* Let the debugger run.  */
1716 	ptrace_stop(signr, 0, info);
1717 
1718 	/* We're back.  Did the debugger cancel the sig?  */
1719 	signr = current->exit_code;
1720 	if (signr == 0)
1721 		return signr;
1722 
1723 	current->exit_code = 0;
1724 
1725 	/* Update the siginfo structure if the signal has
1726 	   changed.  If the debugger wanted something
1727 	   specific in the siginfo structure then it should
1728 	   have updated *info via PTRACE_SETSIGINFO.  */
1729 	if (signr != info->si_signo) {
1730 		info->si_signo = signr;
1731 		info->si_errno = 0;
1732 		info->si_code = SI_USER;
1733 		info->si_pid = task_pid_vnr(current->parent);
1734 		info->si_uid = task_uid(current->parent);
1735 	}
1736 
1737 	/* If the (new) signal is now blocked, requeue it.  */
1738 	if (sigismember(&current->blocked, signr)) {
1739 		specific_send_sig_info(signr, info, current);
1740 		signr = 0;
1741 	}
1742 
1743 	return signr;
1744 }
1745 
1746 int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
1747 			  struct pt_regs *regs, void *cookie)
1748 {
1749 	struct sighand_struct *sighand = current->sighand;
1750 	struct signal_struct *signal = current->signal;
1751 	int signr;
1752 
1753 relock:
1754 	/*
1755 	 * We'll jump back here after any time we were stopped in TASK_STOPPED.
1756 	 * While in TASK_STOPPED, we were considered "frozen enough".
1757 	 * Now that we woke up, it's crucial if we're supposed to be
1758 	 * frozen that we freeze now before running anything substantial.
1759 	 */
1760 	try_to_freeze();
1761 
1762 	spin_lock_irq(&sighand->siglock);
1763 	/*
1764 	 * Every stopped thread goes here after wakeup. Check to see if
1765 	 * we should notify the parent, prepare_signal(SIGCONT) encodes
1766 	 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
1767 	 */
1768 	if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
1769 		int why = (signal->flags & SIGNAL_STOP_CONTINUED)
1770 				? CLD_CONTINUED : CLD_STOPPED;
1771 		signal->flags &= ~SIGNAL_CLD_MASK;
1772 		spin_unlock_irq(&sighand->siglock);
1773 
1774 		if (unlikely(!tracehook_notify_jctl(1, why)))
1775 			goto relock;
1776 
1777 		read_lock(&tasklist_lock);
1778 		do_notify_parent_cldstop(current->group_leader, why);
1779 		read_unlock(&tasklist_lock);
1780 		goto relock;
1781 	}
1782 
1783 	for (;;) {
1784 		struct k_sigaction *ka;
1785 
1786 		if (unlikely(signal->group_stop_count > 0) &&
1787 		    do_signal_stop(0))
1788 			goto relock;
1789 
1790 		/*
1791 		 * Tracing can induce an artifical signal and choose sigaction.
1792 		 * The return value in @signr determines the default action,
1793 		 * but @info->si_signo is the signal number we will report.
1794 		 */
1795 		signr = tracehook_get_signal(current, regs, info, return_ka);
1796 		if (unlikely(signr < 0))
1797 			goto relock;
1798 		if (unlikely(signr != 0))
1799 			ka = return_ka;
1800 		else {
1801 			signr = dequeue_signal(current, &current->blocked,
1802 					       info);
1803 
1804 			if (!signr)
1805 				break; /* will return 0 */
1806 
1807 			if (signr != SIGKILL) {
1808 				signr = ptrace_signal(signr, info,
1809 						      regs, cookie);
1810 				if (!signr)
1811 					continue;
1812 			}
1813 
1814 			ka = &sighand->action[signr-1];
1815 		}
1816 
1817 		if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
1818 			continue;
1819 		if (ka->sa.sa_handler != SIG_DFL) {
1820 			/* Run the handler.  */
1821 			*return_ka = *ka;
1822 
1823 			if (ka->sa.sa_flags & SA_ONESHOT)
1824 				ka->sa.sa_handler = SIG_DFL;
1825 
1826 			break; /* will return non-zero "signr" value */
1827 		}
1828 
1829 		/*
1830 		 * Now we are doing the default action for this signal.
1831 		 */
1832 		if (sig_kernel_ignore(signr)) /* Default is nothing. */
1833 			continue;
1834 
1835 		/*
1836 		 * Global init gets no signals it doesn't want.
1837 		 */
1838 		if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
1839 		    !signal_group_exit(signal))
1840 			continue;
1841 
1842 		if (sig_kernel_stop(signr)) {
1843 			/*
1844 			 * The default action is to stop all threads in
1845 			 * the thread group.  The job control signals
1846 			 * do nothing in an orphaned pgrp, but SIGSTOP
1847 			 * always works.  Note that siglock needs to be
1848 			 * dropped during the call to is_orphaned_pgrp()
1849 			 * because of lock ordering with tasklist_lock.
1850 			 * This allows an intervening SIGCONT to be posted.
1851 			 * We need to check for that and bail out if necessary.
1852 			 */
1853 			if (signr != SIGSTOP) {
1854 				spin_unlock_irq(&sighand->siglock);
1855 
1856 				/* signals can be posted during this window */
1857 
1858 				if (is_current_pgrp_orphaned())
1859 					goto relock;
1860 
1861 				spin_lock_irq(&sighand->siglock);
1862 			}
1863 
1864 			if (likely(do_signal_stop(info->si_signo))) {
1865 				/* It released the siglock.  */
1866 				goto relock;
1867 			}
1868 
1869 			/*
1870 			 * We didn't actually stop, due to a race
1871 			 * with SIGCONT or something like that.
1872 			 */
1873 			continue;
1874 		}
1875 
1876 		spin_unlock_irq(&sighand->siglock);
1877 
1878 		/*
1879 		 * Anything else is fatal, maybe with a core dump.
1880 		 */
1881 		current->flags |= PF_SIGNALED;
1882 
1883 		if (sig_kernel_coredump(signr)) {
1884 			if (print_fatal_signals)
1885 				print_fatal_signal(regs, info->si_signo);
1886 			/*
1887 			 * If it was able to dump core, this kills all
1888 			 * other threads in the group and synchronizes with
1889 			 * their demise.  If we lost the race with another
1890 			 * thread getting here, it set group_exit_code
1891 			 * first and our do_group_exit call below will use
1892 			 * that value and ignore the one we pass it.
1893 			 */
1894 			do_coredump(info->si_signo, info->si_signo, regs);
1895 		}
1896 
1897 		/*
1898 		 * Death signals, no core dump.
1899 		 */
1900 		do_group_exit(info->si_signo);
1901 		/* NOTREACHED */
1902 	}
1903 	spin_unlock_irq(&sighand->siglock);
1904 	return signr;
1905 }
1906 
1907 void exit_signals(struct task_struct *tsk)
1908 {
1909 	int group_stop = 0;
1910 	struct task_struct *t;
1911 
1912 	if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
1913 		tsk->flags |= PF_EXITING;
1914 		return;
1915 	}
1916 
1917 	spin_lock_irq(&tsk->sighand->siglock);
1918 	/*
1919 	 * From now this task is not visible for group-wide signals,
1920 	 * see wants_signal(), do_signal_stop().
1921 	 */
1922 	tsk->flags |= PF_EXITING;
1923 	if (!signal_pending(tsk))
1924 		goto out;
1925 
1926 	/* It could be that __group_complete_signal() choose us to
1927 	 * notify about group-wide signal. Another thread should be
1928 	 * woken now to take the signal since we will not.
1929 	 */
1930 	for (t = tsk; (t = next_thread(t)) != tsk; )
1931 		if (!signal_pending(t) && !(t->flags & PF_EXITING))
1932 			recalc_sigpending_and_wake(t);
1933 
1934 	if (unlikely(tsk->signal->group_stop_count) &&
1935 			!--tsk->signal->group_stop_count) {
1936 		tsk->signal->flags = SIGNAL_STOP_STOPPED;
1937 		group_stop = 1;
1938 	}
1939 out:
1940 	spin_unlock_irq(&tsk->sighand->siglock);
1941 
1942 	if (unlikely(group_stop) && tracehook_notify_jctl(1, CLD_STOPPED)) {
1943 		read_lock(&tasklist_lock);
1944 		do_notify_parent_cldstop(tsk, CLD_STOPPED);
1945 		read_unlock(&tasklist_lock);
1946 	}
1947 }
1948 
1949 EXPORT_SYMBOL(recalc_sigpending);
1950 EXPORT_SYMBOL_GPL(dequeue_signal);
1951 EXPORT_SYMBOL(flush_signals);
1952 EXPORT_SYMBOL(force_sig);
1953 EXPORT_SYMBOL(send_sig);
1954 EXPORT_SYMBOL(send_sig_info);
1955 EXPORT_SYMBOL(sigprocmask);
1956 EXPORT_SYMBOL(block_all_signals);
1957 EXPORT_SYMBOL(unblock_all_signals);
1958 
1959 
1960 /*
1961  * System call entry points.
1962  */
1963 
1964 asmlinkage long sys_restart_syscall(void)
1965 {
1966 	struct restart_block *restart = &current_thread_info()->restart_block;
1967 	return restart->fn(restart);
1968 }
1969 
1970 long do_no_restart_syscall(struct restart_block *param)
1971 {
1972 	return -EINTR;
1973 }
1974 
1975 /*
1976  * We don't need to get the kernel lock - this is all local to this
1977  * particular thread.. (and that's good, because this is _heavily_
1978  * used by various programs)
1979  */
1980 
1981 /*
1982  * This is also useful for kernel threads that want to temporarily
1983  * (or permanently) block certain signals.
1984  *
1985  * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
1986  * interface happily blocks "unblockable" signals like SIGKILL
1987  * and friends.
1988  */
1989 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
1990 {
1991 	int error;
1992 
1993 	spin_lock_irq(&current->sighand->siglock);
1994 	if (oldset)
1995 		*oldset = current->blocked;
1996 
1997 	error = 0;
1998 	switch (how) {
1999 	case SIG_BLOCK:
2000 		sigorsets(&current->blocked, &current->blocked, set);
2001 		break;
2002 	case SIG_UNBLOCK:
2003 		signandsets(&current->blocked, &current->blocked, set);
2004 		break;
2005 	case SIG_SETMASK:
2006 		current->blocked = *set;
2007 		break;
2008 	default:
2009 		error = -EINVAL;
2010 	}
2011 	recalc_sigpending();
2012 	spin_unlock_irq(&current->sighand->siglock);
2013 
2014 	return error;
2015 }
2016 
2017 asmlinkage long
2018 sys_rt_sigprocmask(int how, sigset_t __user *set, sigset_t __user *oset, size_t sigsetsize)
2019 {
2020 	int error = -EINVAL;
2021 	sigset_t old_set, new_set;
2022 
2023 	/* XXX: Don't preclude handling different sized sigset_t's.  */
2024 	if (sigsetsize != sizeof(sigset_t))
2025 		goto out;
2026 
2027 	if (set) {
2028 		error = -EFAULT;
2029 		if (copy_from_user(&new_set, set, sizeof(*set)))
2030 			goto out;
2031 		sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2032 
2033 		error = sigprocmask(how, &new_set, &old_set);
2034 		if (error)
2035 			goto out;
2036 		if (oset)
2037 			goto set_old;
2038 	} else if (oset) {
2039 		spin_lock_irq(&current->sighand->siglock);
2040 		old_set = current->blocked;
2041 		spin_unlock_irq(&current->sighand->siglock);
2042 
2043 	set_old:
2044 		error = -EFAULT;
2045 		if (copy_to_user(oset, &old_set, sizeof(*oset)))
2046 			goto out;
2047 	}
2048 	error = 0;
2049 out:
2050 	return error;
2051 }
2052 
2053 long do_sigpending(void __user *set, unsigned long sigsetsize)
2054 {
2055 	long error = -EINVAL;
2056 	sigset_t pending;
2057 
2058 	if (sigsetsize > sizeof(sigset_t))
2059 		goto out;
2060 
2061 	spin_lock_irq(&current->sighand->siglock);
2062 	sigorsets(&pending, &current->pending.signal,
2063 		  &current->signal->shared_pending.signal);
2064 	spin_unlock_irq(&current->sighand->siglock);
2065 
2066 	/* Outside the lock because only this thread touches it.  */
2067 	sigandsets(&pending, &current->blocked, &pending);
2068 
2069 	error = -EFAULT;
2070 	if (!copy_to_user(set, &pending, sigsetsize))
2071 		error = 0;
2072 
2073 out:
2074 	return error;
2075 }
2076 
2077 asmlinkage long
2078 sys_rt_sigpending(sigset_t __user *set, size_t sigsetsize)
2079 {
2080 	return do_sigpending(set, sigsetsize);
2081 }
2082 
2083 #ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2084 
2085 int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2086 {
2087 	int err;
2088 
2089 	if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2090 		return -EFAULT;
2091 	if (from->si_code < 0)
2092 		return __copy_to_user(to, from, sizeof(siginfo_t))
2093 			? -EFAULT : 0;
2094 	/*
2095 	 * If you change siginfo_t structure, please be sure
2096 	 * this code is fixed accordingly.
2097 	 * Please remember to update the signalfd_copyinfo() function
2098 	 * inside fs/signalfd.c too, in case siginfo_t changes.
2099 	 * It should never copy any pad contained in the structure
2100 	 * to avoid security leaks, but must copy the generic
2101 	 * 3 ints plus the relevant union member.
2102 	 */
2103 	err = __put_user(from->si_signo, &to->si_signo);
2104 	err |= __put_user(from->si_errno, &to->si_errno);
2105 	err |= __put_user((short)from->si_code, &to->si_code);
2106 	switch (from->si_code & __SI_MASK) {
2107 	case __SI_KILL:
2108 		err |= __put_user(from->si_pid, &to->si_pid);
2109 		err |= __put_user(from->si_uid, &to->si_uid);
2110 		break;
2111 	case __SI_TIMER:
2112 		 err |= __put_user(from->si_tid, &to->si_tid);
2113 		 err |= __put_user(from->si_overrun, &to->si_overrun);
2114 		 err |= __put_user(from->si_ptr, &to->si_ptr);
2115 		break;
2116 	case __SI_POLL:
2117 		err |= __put_user(from->si_band, &to->si_band);
2118 		err |= __put_user(from->si_fd, &to->si_fd);
2119 		break;
2120 	case __SI_FAULT:
2121 		err |= __put_user(from->si_addr, &to->si_addr);
2122 #ifdef __ARCH_SI_TRAPNO
2123 		err |= __put_user(from->si_trapno, &to->si_trapno);
2124 #endif
2125 		break;
2126 	case __SI_CHLD:
2127 		err |= __put_user(from->si_pid, &to->si_pid);
2128 		err |= __put_user(from->si_uid, &to->si_uid);
2129 		err |= __put_user(from->si_status, &to->si_status);
2130 		err |= __put_user(from->si_utime, &to->si_utime);
2131 		err |= __put_user(from->si_stime, &to->si_stime);
2132 		break;
2133 	case __SI_RT: /* This is not generated by the kernel as of now. */
2134 	case __SI_MESGQ: /* But this is */
2135 		err |= __put_user(from->si_pid, &to->si_pid);
2136 		err |= __put_user(from->si_uid, &to->si_uid);
2137 		err |= __put_user(from->si_ptr, &to->si_ptr);
2138 		break;
2139 	default: /* this is just in case for now ... */
2140 		err |= __put_user(from->si_pid, &to->si_pid);
2141 		err |= __put_user(from->si_uid, &to->si_uid);
2142 		break;
2143 	}
2144 	return err;
2145 }
2146 
2147 #endif
2148 
2149 asmlinkage long
2150 sys_rt_sigtimedwait(const sigset_t __user *uthese,
2151 		    siginfo_t __user *uinfo,
2152 		    const struct timespec __user *uts,
2153 		    size_t sigsetsize)
2154 {
2155 	int ret, sig;
2156 	sigset_t these;
2157 	struct timespec ts;
2158 	siginfo_t info;
2159 	long timeout = 0;
2160 
2161 	/* XXX: Don't preclude handling different sized sigset_t's.  */
2162 	if (sigsetsize != sizeof(sigset_t))
2163 		return -EINVAL;
2164 
2165 	if (copy_from_user(&these, uthese, sizeof(these)))
2166 		return -EFAULT;
2167 
2168 	/*
2169 	 * Invert the set of allowed signals to get those we
2170 	 * want to block.
2171 	 */
2172 	sigdelsetmask(&these, sigmask(SIGKILL)|sigmask(SIGSTOP));
2173 	signotset(&these);
2174 
2175 	if (uts) {
2176 		if (copy_from_user(&ts, uts, sizeof(ts)))
2177 			return -EFAULT;
2178 		if (ts.tv_nsec >= 1000000000L || ts.tv_nsec < 0
2179 		    || ts.tv_sec < 0)
2180 			return -EINVAL;
2181 	}
2182 
2183 	spin_lock_irq(&current->sighand->siglock);
2184 	sig = dequeue_signal(current, &these, &info);
2185 	if (!sig) {
2186 		timeout = MAX_SCHEDULE_TIMEOUT;
2187 		if (uts)
2188 			timeout = (timespec_to_jiffies(&ts)
2189 				   + (ts.tv_sec || ts.tv_nsec));
2190 
2191 		if (timeout) {
2192 			/* None ready -- temporarily unblock those we're
2193 			 * interested while we are sleeping in so that we'll
2194 			 * be awakened when they arrive.  */
2195 			current->real_blocked = current->blocked;
2196 			sigandsets(&current->blocked, &current->blocked, &these);
2197 			recalc_sigpending();
2198 			spin_unlock_irq(&current->sighand->siglock);
2199 
2200 			timeout = schedule_timeout_interruptible(timeout);
2201 
2202 			spin_lock_irq(&current->sighand->siglock);
2203 			sig = dequeue_signal(current, &these, &info);
2204 			current->blocked = current->real_blocked;
2205 			siginitset(&current->real_blocked, 0);
2206 			recalc_sigpending();
2207 		}
2208 	}
2209 	spin_unlock_irq(&current->sighand->siglock);
2210 
2211 	if (sig) {
2212 		ret = sig;
2213 		if (uinfo) {
2214 			if (copy_siginfo_to_user(uinfo, &info))
2215 				ret = -EFAULT;
2216 		}
2217 	} else {
2218 		ret = -EAGAIN;
2219 		if (timeout)
2220 			ret = -EINTR;
2221 	}
2222 
2223 	return ret;
2224 }
2225 
2226 asmlinkage long
2227 sys_kill(pid_t pid, int sig)
2228 {
2229 	struct siginfo info;
2230 
2231 	info.si_signo = sig;
2232 	info.si_errno = 0;
2233 	info.si_code = SI_USER;
2234 	info.si_pid = task_tgid_vnr(current);
2235 	info.si_uid = current_uid();
2236 
2237 	return kill_something_info(sig, &info, pid);
2238 }
2239 
2240 static int do_tkill(pid_t tgid, pid_t pid, int sig)
2241 {
2242 	int error;
2243 	struct siginfo info;
2244 	struct task_struct *p;
2245 	unsigned long flags;
2246 
2247 	error = -ESRCH;
2248 	info.si_signo = sig;
2249 	info.si_errno = 0;
2250 	info.si_code = SI_TKILL;
2251 	info.si_pid = task_tgid_vnr(current);
2252 	info.si_uid = current_uid();
2253 
2254 	rcu_read_lock();
2255 	p = find_task_by_vpid(pid);
2256 	if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
2257 		error = check_kill_permission(sig, &info, p);
2258 		/*
2259 		 * The null signal is a permissions and process existence
2260 		 * probe.  No signal is actually delivered.
2261 		 *
2262 		 * If lock_task_sighand() fails we pretend the task dies
2263 		 * after receiving the signal. The window is tiny, and the
2264 		 * signal is private anyway.
2265 		 */
2266 		if (!error && sig && lock_task_sighand(p, &flags)) {
2267 			error = specific_send_sig_info(sig, &info, p);
2268 			unlock_task_sighand(p, &flags);
2269 		}
2270 	}
2271 	rcu_read_unlock();
2272 
2273 	return error;
2274 }
2275 
2276 /**
2277  *  sys_tgkill - send signal to one specific thread
2278  *  @tgid: the thread group ID of the thread
2279  *  @pid: the PID of the thread
2280  *  @sig: signal to be sent
2281  *
2282  *  This syscall also checks the @tgid and returns -ESRCH even if the PID
2283  *  exists but it's not belonging to the target process anymore. This
2284  *  method solves the problem of threads exiting and PIDs getting reused.
2285  */
2286 asmlinkage long sys_tgkill(pid_t tgid, pid_t pid, int sig)
2287 {
2288 	/* This is only valid for single tasks */
2289 	if (pid <= 0 || tgid <= 0)
2290 		return -EINVAL;
2291 
2292 	return do_tkill(tgid, pid, sig);
2293 }
2294 
2295 /*
2296  *  Send a signal to only one task, even if it's a CLONE_THREAD task.
2297  */
2298 asmlinkage long
2299 sys_tkill(pid_t pid, int sig)
2300 {
2301 	/* This is only valid for single tasks */
2302 	if (pid <= 0)
2303 		return -EINVAL;
2304 
2305 	return do_tkill(0, pid, sig);
2306 }
2307 
2308 asmlinkage long
2309 sys_rt_sigqueueinfo(pid_t pid, int sig, siginfo_t __user *uinfo)
2310 {
2311 	siginfo_t info;
2312 
2313 	if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2314 		return -EFAULT;
2315 
2316 	/* Not even root can pretend to send signals from the kernel.
2317 	   Nor can they impersonate a kill(), which adds source info.  */
2318 	if (info.si_code >= 0)
2319 		return -EPERM;
2320 	info.si_signo = sig;
2321 
2322 	/* POSIX.1b doesn't mention process groups.  */
2323 	return kill_proc_info(sig, &info, pid);
2324 }
2325 
2326 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
2327 {
2328 	struct task_struct *t = current;
2329 	struct k_sigaction *k;
2330 	sigset_t mask;
2331 
2332 	if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
2333 		return -EINVAL;
2334 
2335 	k = &t->sighand->action[sig-1];
2336 
2337 	spin_lock_irq(&current->sighand->siglock);
2338 	if (oact)
2339 		*oact = *k;
2340 
2341 	if (act) {
2342 		sigdelsetmask(&act->sa.sa_mask,
2343 			      sigmask(SIGKILL) | sigmask(SIGSTOP));
2344 		*k = *act;
2345 		/*
2346 		 * POSIX 3.3.1.3:
2347 		 *  "Setting a signal action to SIG_IGN for a signal that is
2348 		 *   pending shall cause the pending signal to be discarded,
2349 		 *   whether or not it is blocked."
2350 		 *
2351 		 *  "Setting a signal action to SIG_DFL for a signal that is
2352 		 *   pending and whose default action is to ignore the signal
2353 		 *   (for example, SIGCHLD), shall cause the pending signal to
2354 		 *   be discarded, whether or not it is blocked"
2355 		 */
2356 		if (sig_handler_ignored(sig_handler(t, sig), sig)) {
2357 			sigemptyset(&mask);
2358 			sigaddset(&mask, sig);
2359 			rm_from_queue_full(&mask, &t->signal->shared_pending);
2360 			do {
2361 				rm_from_queue_full(&mask, &t->pending);
2362 				t = next_thread(t);
2363 			} while (t != current);
2364 		}
2365 	}
2366 
2367 	spin_unlock_irq(&current->sighand->siglock);
2368 	return 0;
2369 }
2370 
2371 int
2372 do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
2373 {
2374 	stack_t oss;
2375 	int error;
2376 
2377 	if (uoss) {
2378 		oss.ss_sp = (void __user *) current->sas_ss_sp;
2379 		oss.ss_size = current->sas_ss_size;
2380 		oss.ss_flags = sas_ss_flags(sp);
2381 	}
2382 
2383 	if (uss) {
2384 		void __user *ss_sp;
2385 		size_t ss_size;
2386 		int ss_flags;
2387 
2388 		error = -EFAULT;
2389 		if (!access_ok(VERIFY_READ, uss, sizeof(*uss))
2390 		    || __get_user(ss_sp, &uss->ss_sp)
2391 		    || __get_user(ss_flags, &uss->ss_flags)
2392 		    || __get_user(ss_size, &uss->ss_size))
2393 			goto out;
2394 
2395 		error = -EPERM;
2396 		if (on_sig_stack(sp))
2397 			goto out;
2398 
2399 		error = -EINVAL;
2400 		/*
2401 		 *
2402 		 * Note - this code used to test ss_flags incorrectly
2403 		 *  	  old code may have been written using ss_flags==0
2404 		 *	  to mean ss_flags==SS_ONSTACK (as this was the only
2405 		 *	  way that worked) - this fix preserves that older
2406 		 *	  mechanism
2407 		 */
2408 		if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
2409 			goto out;
2410 
2411 		if (ss_flags == SS_DISABLE) {
2412 			ss_size = 0;
2413 			ss_sp = NULL;
2414 		} else {
2415 			error = -ENOMEM;
2416 			if (ss_size < MINSIGSTKSZ)
2417 				goto out;
2418 		}
2419 
2420 		current->sas_ss_sp = (unsigned long) ss_sp;
2421 		current->sas_ss_size = ss_size;
2422 	}
2423 
2424 	if (uoss) {
2425 		error = -EFAULT;
2426 		if (copy_to_user(uoss, &oss, sizeof(oss)))
2427 			goto out;
2428 	}
2429 
2430 	error = 0;
2431 out:
2432 	return error;
2433 }
2434 
2435 #ifdef __ARCH_WANT_SYS_SIGPENDING
2436 
2437 asmlinkage long
2438 sys_sigpending(old_sigset_t __user *set)
2439 {
2440 	return do_sigpending(set, sizeof(*set));
2441 }
2442 
2443 #endif
2444 
2445 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
2446 /* Some platforms have their own version with special arguments others
2447    support only sys_rt_sigprocmask.  */
2448 
2449 asmlinkage long
2450 sys_sigprocmask(int how, old_sigset_t __user *set, old_sigset_t __user *oset)
2451 {
2452 	int error;
2453 	old_sigset_t old_set, new_set;
2454 
2455 	if (set) {
2456 		error = -EFAULT;
2457 		if (copy_from_user(&new_set, set, sizeof(*set)))
2458 			goto out;
2459 		new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
2460 
2461 		spin_lock_irq(&current->sighand->siglock);
2462 		old_set = current->blocked.sig[0];
2463 
2464 		error = 0;
2465 		switch (how) {
2466 		default:
2467 			error = -EINVAL;
2468 			break;
2469 		case SIG_BLOCK:
2470 			sigaddsetmask(&current->blocked, new_set);
2471 			break;
2472 		case SIG_UNBLOCK:
2473 			sigdelsetmask(&current->blocked, new_set);
2474 			break;
2475 		case SIG_SETMASK:
2476 			current->blocked.sig[0] = new_set;
2477 			break;
2478 		}
2479 
2480 		recalc_sigpending();
2481 		spin_unlock_irq(&current->sighand->siglock);
2482 		if (error)
2483 			goto out;
2484 		if (oset)
2485 			goto set_old;
2486 	} else if (oset) {
2487 		old_set = current->blocked.sig[0];
2488 	set_old:
2489 		error = -EFAULT;
2490 		if (copy_to_user(oset, &old_set, sizeof(*oset)))
2491 			goto out;
2492 	}
2493 	error = 0;
2494 out:
2495 	return error;
2496 }
2497 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
2498 
2499 #ifdef __ARCH_WANT_SYS_RT_SIGACTION
2500 asmlinkage long
2501 sys_rt_sigaction(int sig,
2502 		 const struct sigaction __user *act,
2503 		 struct sigaction __user *oact,
2504 		 size_t sigsetsize)
2505 {
2506 	struct k_sigaction new_sa, old_sa;
2507 	int ret = -EINVAL;
2508 
2509 	/* XXX: Don't preclude handling different sized sigset_t's.  */
2510 	if (sigsetsize != sizeof(sigset_t))
2511 		goto out;
2512 
2513 	if (act) {
2514 		if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
2515 			return -EFAULT;
2516 	}
2517 
2518 	ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
2519 
2520 	if (!ret && oact) {
2521 		if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
2522 			return -EFAULT;
2523 	}
2524 out:
2525 	return ret;
2526 }
2527 #endif /* __ARCH_WANT_SYS_RT_SIGACTION */
2528 
2529 #ifdef __ARCH_WANT_SYS_SGETMASK
2530 
2531 /*
2532  * For backwards compatibility.  Functionality superseded by sigprocmask.
2533  */
2534 asmlinkage long
2535 sys_sgetmask(void)
2536 {
2537 	/* SMP safe */
2538 	return current->blocked.sig[0];
2539 }
2540 
2541 asmlinkage long
2542 sys_ssetmask(int newmask)
2543 {
2544 	int old;
2545 
2546 	spin_lock_irq(&current->sighand->siglock);
2547 	old = current->blocked.sig[0];
2548 
2549 	siginitset(&current->blocked, newmask & ~(sigmask(SIGKILL)|
2550 						  sigmask(SIGSTOP)));
2551 	recalc_sigpending();
2552 	spin_unlock_irq(&current->sighand->siglock);
2553 
2554 	return old;
2555 }
2556 #endif /* __ARCH_WANT_SGETMASK */
2557 
2558 #ifdef __ARCH_WANT_SYS_SIGNAL
2559 /*
2560  * For backwards compatibility.  Functionality superseded by sigaction.
2561  */
2562 asmlinkage unsigned long
2563 sys_signal(int sig, __sighandler_t handler)
2564 {
2565 	struct k_sigaction new_sa, old_sa;
2566 	int ret;
2567 
2568 	new_sa.sa.sa_handler = handler;
2569 	new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
2570 	sigemptyset(&new_sa.sa.sa_mask);
2571 
2572 	ret = do_sigaction(sig, &new_sa, &old_sa);
2573 
2574 	return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
2575 }
2576 #endif /* __ARCH_WANT_SYS_SIGNAL */
2577 
2578 #ifdef __ARCH_WANT_SYS_PAUSE
2579 
2580 asmlinkage long
2581 sys_pause(void)
2582 {
2583 	current->state = TASK_INTERRUPTIBLE;
2584 	schedule();
2585 	return -ERESTARTNOHAND;
2586 }
2587 
2588 #endif
2589 
2590 #ifdef __ARCH_WANT_SYS_RT_SIGSUSPEND
2591 asmlinkage long sys_rt_sigsuspend(sigset_t __user *unewset, size_t sigsetsize)
2592 {
2593 	sigset_t newset;
2594 
2595 	/* XXX: Don't preclude handling different sized sigset_t's.  */
2596 	if (sigsetsize != sizeof(sigset_t))
2597 		return -EINVAL;
2598 
2599 	if (copy_from_user(&newset, unewset, sizeof(newset)))
2600 		return -EFAULT;
2601 	sigdelsetmask(&newset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2602 
2603 	spin_lock_irq(&current->sighand->siglock);
2604 	current->saved_sigmask = current->blocked;
2605 	current->blocked = newset;
2606 	recalc_sigpending();
2607 	spin_unlock_irq(&current->sighand->siglock);
2608 
2609 	current->state = TASK_INTERRUPTIBLE;
2610 	schedule();
2611 	set_restore_sigmask();
2612 	return -ERESTARTNOHAND;
2613 }
2614 #endif /* __ARCH_WANT_SYS_RT_SIGSUSPEND */
2615 
2616 __attribute__((weak)) const char *arch_vma_name(struct vm_area_struct *vma)
2617 {
2618 	return NULL;
2619 }
2620 
2621 void __init signals_init(void)
2622 {
2623 	sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
2624 }
2625