xref: /openbmc/linux/kernel/signal.c (revision ecba1060)
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/events/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 static void __user *sig_handler(struct task_struct *t, int sig)
45 {
46 	return t->sighand->action[sig - 1].sa.sa_handler;
47 }
48 
49 static int sig_handler_ignored(void __user *handler, int sig)
50 {
51 	/* Is it explicitly or implicitly ignored? */
52 	return handler == SIG_IGN ||
53 		(handler == SIG_DFL && sig_kernel_ignore(sig));
54 }
55 
56 static int sig_task_ignored(struct task_struct *t, int sig,
57 		int from_ancestor_ns)
58 {
59 	void __user *handler;
60 
61 	handler = sig_handler(t, sig);
62 
63 	if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
64 			handler == SIG_DFL && !from_ancestor_ns)
65 		return 1;
66 
67 	return sig_handler_ignored(handler, sig);
68 }
69 
70 static int sig_ignored(struct task_struct *t, int sig, int from_ancestor_ns)
71 {
72 	/*
73 	 * Blocked signals are never ignored, since the
74 	 * signal handler may change by the time it is
75 	 * unblocked.
76 	 */
77 	if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
78 		return 0;
79 
80 	if (!sig_task_ignored(t, sig, from_ancestor_ns))
81 		return 0;
82 
83 	/*
84 	 * Tracers may want to know about even ignored signals.
85 	 */
86 	return !tracehook_consider_ignored_signal(t, sig);
87 }
88 
89 /*
90  * Re-calculate pending state from the set of locally pending
91  * signals, globally pending signals, and blocked signals.
92  */
93 static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
94 {
95 	unsigned long ready;
96 	long i;
97 
98 	switch (_NSIG_WORDS) {
99 	default:
100 		for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
101 			ready |= signal->sig[i] &~ blocked->sig[i];
102 		break;
103 
104 	case 4: ready  = signal->sig[3] &~ blocked->sig[3];
105 		ready |= signal->sig[2] &~ blocked->sig[2];
106 		ready |= signal->sig[1] &~ blocked->sig[1];
107 		ready |= signal->sig[0] &~ blocked->sig[0];
108 		break;
109 
110 	case 2: ready  = signal->sig[1] &~ blocked->sig[1];
111 		ready |= signal->sig[0] &~ blocked->sig[0];
112 		break;
113 
114 	case 1: ready  = signal->sig[0] &~ blocked->sig[0];
115 	}
116 	return ready !=	0;
117 }
118 
119 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
120 
121 static int recalc_sigpending_tsk(struct task_struct *t)
122 {
123 	if (t->signal->group_stop_count > 0 ||
124 	    PENDING(&t->pending, &t->blocked) ||
125 	    PENDING(&t->signal->shared_pending, &t->blocked)) {
126 		set_tsk_thread_flag(t, TIF_SIGPENDING);
127 		return 1;
128 	}
129 	/*
130 	 * We must never clear the flag in another thread, or in current
131 	 * when it's possible the current syscall is returning -ERESTART*.
132 	 * So we don't clear it here, and only callers who know they should do.
133 	 */
134 	return 0;
135 }
136 
137 /*
138  * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
139  * This is superfluous when called on current, the wakeup is a harmless no-op.
140  */
141 void recalc_sigpending_and_wake(struct task_struct *t)
142 {
143 	if (recalc_sigpending_tsk(t))
144 		signal_wake_up(t, 0);
145 }
146 
147 void recalc_sigpending(void)
148 {
149 	if (unlikely(tracehook_force_sigpending()))
150 		set_thread_flag(TIF_SIGPENDING);
151 	else if (!recalc_sigpending_tsk(current) && !freezing(current))
152 		clear_thread_flag(TIF_SIGPENDING);
153 
154 }
155 
156 /* Given the mask, find the first available signal that should be serviced. */
157 
158 int next_signal(struct sigpending *pending, sigset_t *mask)
159 {
160 	unsigned long i, *s, *m, x;
161 	int sig = 0;
162 
163 	s = pending->signal.sig;
164 	m = mask->sig;
165 	switch (_NSIG_WORDS) {
166 	default:
167 		for (i = 0; i < _NSIG_WORDS; ++i, ++s, ++m)
168 			if ((x = *s &~ *m) != 0) {
169 				sig = ffz(~x) + i*_NSIG_BPW + 1;
170 				break;
171 			}
172 		break;
173 
174 	case 2: if ((x = s[0] &~ m[0]) != 0)
175 			sig = 1;
176 		else if ((x = s[1] &~ m[1]) != 0)
177 			sig = _NSIG_BPW + 1;
178 		else
179 			break;
180 		sig += ffz(~x);
181 		break;
182 
183 	case 1: if ((x = *s &~ *m) != 0)
184 			sig = ffz(~x) + 1;
185 		break;
186 	}
187 
188 	return sig;
189 }
190 
191 /*
192  * allocate a new signal queue record
193  * - this may be called without locks if and only if t == current, otherwise an
194  *   appopriate lock must be held to stop the target task from exiting
195  */
196 static struct sigqueue *__sigqueue_alloc(struct task_struct *t, gfp_t flags,
197 					 int override_rlimit)
198 {
199 	struct sigqueue *q = NULL;
200 	struct user_struct *user;
201 
202 	/*
203 	 * We won't get problems with the target's UID changing under us
204 	 * because changing it requires RCU be used, and if t != current, the
205 	 * caller must be holding the RCU readlock (by way of a spinlock) and
206 	 * we use RCU protection here
207 	 */
208 	user = get_uid(__task_cred(t)->user);
209 	atomic_inc(&user->sigpending);
210 	if (override_rlimit ||
211 	    atomic_read(&user->sigpending) <=
212 			t->signal->rlim[RLIMIT_SIGPENDING].rlim_cur)
213 		q = kmem_cache_alloc(sigqueue_cachep, flags);
214 	if (unlikely(q == NULL)) {
215 		atomic_dec(&user->sigpending);
216 		free_uid(user);
217 	} else {
218 		INIT_LIST_HEAD(&q->list);
219 		q->flags = 0;
220 		q->user = user;
221 	}
222 
223 	return q;
224 }
225 
226 static void __sigqueue_free(struct sigqueue *q)
227 {
228 	if (q->flags & SIGQUEUE_PREALLOC)
229 		return;
230 	atomic_dec(&q->user->sigpending);
231 	free_uid(q->user);
232 	kmem_cache_free(sigqueue_cachep, q);
233 }
234 
235 void flush_sigqueue(struct sigpending *queue)
236 {
237 	struct sigqueue *q;
238 
239 	sigemptyset(&queue->signal);
240 	while (!list_empty(&queue->list)) {
241 		q = list_entry(queue->list.next, struct sigqueue , list);
242 		list_del_init(&q->list);
243 		__sigqueue_free(q);
244 	}
245 }
246 
247 /*
248  * Flush all pending signals for a task.
249  */
250 void __flush_signals(struct task_struct *t)
251 {
252 	clear_tsk_thread_flag(t, TIF_SIGPENDING);
253 	flush_sigqueue(&t->pending);
254 	flush_sigqueue(&t->signal->shared_pending);
255 }
256 
257 void flush_signals(struct task_struct *t)
258 {
259 	unsigned long flags;
260 
261 	spin_lock_irqsave(&t->sighand->siglock, flags);
262 	__flush_signals(t);
263 	spin_unlock_irqrestore(&t->sighand->siglock, flags);
264 }
265 
266 static void __flush_itimer_signals(struct sigpending *pending)
267 {
268 	sigset_t signal, retain;
269 	struct sigqueue *q, *n;
270 
271 	signal = pending->signal;
272 	sigemptyset(&retain);
273 
274 	list_for_each_entry_safe(q, n, &pending->list, list) {
275 		int sig = q->info.si_signo;
276 
277 		if (likely(q->info.si_code != SI_TIMER)) {
278 			sigaddset(&retain, sig);
279 		} else {
280 			sigdelset(&signal, sig);
281 			list_del_init(&q->list);
282 			__sigqueue_free(q);
283 		}
284 	}
285 
286 	sigorsets(&pending->signal, &signal, &retain);
287 }
288 
289 void flush_itimer_signals(void)
290 {
291 	struct task_struct *tsk = current;
292 	unsigned long flags;
293 
294 	spin_lock_irqsave(&tsk->sighand->siglock, flags);
295 	__flush_itimer_signals(&tsk->pending);
296 	__flush_itimer_signals(&tsk->signal->shared_pending);
297 	spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
298 }
299 
300 void ignore_signals(struct task_struct *t)
301 {
302 	int i;
303 
304 	for (i = 0; i < _NSIG; ++i)
305 		t->sighand->action[i].sa.sa_handler = SIG_IGN;
306 
307 	flush_signals(t);
308 }
309 
310 /*
311  * Flush all handlers for a task.
312  */
313 
314 void
315 flush_signal_handlers(struct task_struct *t, int force_default)
316 {
317 	int i;
318 	struct k_sigaction *ka = &t->sighand->action[0];
319 	for (i = _NSIG ; i != 0 ; i--) {
320 		if (force_default || ka->sa.sa_handler != SIG_IGN)
321 			ka->sa.sa_handler = SIG_DFL;
322 		ka->sa.sa_flags = 0;
323 		sigemptyset(&ka->sa.sa_mask);
324 		ka++;
325 	}
326 }
327 
328 int unhandled_signal(struct task_struct *tsk, int sig)
329 {
330 	void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
331 	if (is_global_init(tsk))
332 		return 1;
333 	if (handler != SIG_IGN && handler != SIG_DFL)
334 		return 0;
335 	return !tracehook_consider_fatal_signal(tsk, sig);
336 }
337 
338 
339 /* Notify the system that a driver wants to block all signals for this
340  * process, and wants to be notified if any signals at all were to be
341  * sent/acted upon.  If the notifier routine returns non-zero, then the
342  * signal will be acted upon after all.  If the notifier routine returns 0,
343  * then then signal will be blocked.  Only one block per process is
344  * allowed.  priv is a pointer to private data that the notifier routine
345  * can use to determine if the signal should be blocked or not.  */
346 
347 void
348 block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
349 {
350 	unsigned long flags;
351 
352 	spin_lock_irqsave(&current->sighand->siglock, flags);
353 	current->notifier_mask = mask;
354 	current->notifier_data = priv;
355 	current->notifier = notifier;
356 	spin_unlock_irqrestore(&current->sighand->siglock, flags);
357 }
358 
359 /* Notify the system that blocking has ended. */
360 
361 void
362 unblock_all_signals(void)
363 {
364 	unsigned long flags;
365 
366 	spin_lock_irqsave(&current->sighand->siglock, flags);
367 	current->notifier = NULL;
368 	current->notifier_data = NULL;
369 	recalc_sigpending();
370 	spin_unlock_irqrestore(&current->sighand->siglock, flags);
371 }
372 
373 static void collect_signal(int sig, struct sigpending *list, siginfo_t *info)
374 {
375 	struct sigqueue *q, *first = NULL;
376 
377 	/*
378 	 * Collect the siginfo appropriate to this signal.  Check if
379 	 * there is another siginfo for the same signal.
380 	*/
381 	list_for_each_entry(q, &list->list, list) {
382 		if (q->info.si_signo == sig) {
383 			if (first)
384 				goto still_pending;
385 			first = q;
386 		}
387 	}
388 
389 	sigdelset(&list->signal, sig);
390 
391 	if (first) {
392 still_pending:
393 		list_del_init(&first->list);
394 		copy_siginfo(info, &first->info);
395 		__sigqueue_free(first);
396 	} else {
397 		/* Ok, it wasn't in the queue.  This must be
398 		   a fast-pathed signal or we must have been
399 		   out of queue space.  So zero out the info.
400 		 */
401 		info->si_signo = sig;
402 		info->si_errno = 0;
403 		info->si_code = 0;
404 		info->si_pid = 0;
405 		info->si_uid = 0;
406 	}
407 }
408 
409 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
410 			siginfo_t *info)
411 {
412 	int sig = next_signal(pending, mask);
413 
414 	if (sig) {
415 		if (current->notifier) {
416 			if (sigismember(current->notifier_mask, sig)) {
417 				if (!(current->notifier)(current->notifier_data)) {
418 					clear_thread_flag(TIF_SIGPENDING);
419 					return 0;
420 				}
421 			}
422 		}
423 
424 		collect_signal(sig, pending, info);
425 	}
426 
427 	return sig;
428 }
429 
430 /*
431  * Dequeue a signal and return the element to the caller, which is
432  * expected to free it.
433  *
434  * All callers have to hold the siglock.
435  */
436 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
437 {
438 	int signr;
439 
440 	/* We only dequeue private signals from ourselves, we don't let
441 	 * signalfd steal them
442 	 */
443 	signr = __dequeue_signal(&tsk->pending, mask, info);
444 	if (!signr) {
445 		signr = __dequeue_signal(&tsk->signal->shared_pending,
446 					 mask, info);
447 		/*
448 		 * itimer signal ?
449 		 *
450 		 * itimers are process shared and we restart periodic
451 		 * itimers in the signal delivery path to prevent DoS
452 		 * attacks in the high resolution timer case. This is
453 		 * compliant with the old way of self restarting
454 		 * itimers, as the SIGALRM is a legacy signal and only
455 		 * queued once. Changing the restart behaviour to
456 		 * restart the timer in the signal dequeue path is
457 		 * reducing the timer noise on heavy loaded !highres
458 		 * systems too.
459 		 */
460 		if (unlikely(signr == SIGALRM)) {
461 			struct hrtimer *tmr = &tsk->signal->real_timer;
462 
463 			if (!hrtimer_is_queued(tmr) &&
464 			    tsk->signal->it_real_incr.tv64 != 0) {
465 				hrtimer_forward(tmr, tmr->base->get_time(),
466 						tsk->signal->it_real_incr);
467 				hrtimer_restart(tmr);
468 			}
469 		}
470 	}
471 
472 	recalc_sigpending();
473 	if (!signr)
474 		return 0;
475 
476 	if (unlikely(sig_kernel_stop(signr))) {
477 		/*
478 		 * Set a marker that we have dequeued a stop signal.  Our
479 		 * caller might release the siglock and then the pending
480 		 * stop signal it is about to process is no longer in the
481 		 * pending bitmasks, but must still be cleared by a SIGCONT
482 		 * (and overruled by a SIGKILL).  So those cases clear this
483 		 * shared flag after we've set it.  Note that this flag may
484 		 * remain set after the signal we return is ignored or
485 		 * handled.  That doesn't matter because its only purpose
486 		 * is to alert stop-signal processing code when another
487 		 * processor has come along and cleared the flag.
488 		 */
489 		tsk->signal->flags |= SIGNAL_STOP_DEQUEUED;
490 	}
491 	if ((info->si_code & __SI_MASK) == __SI_TIMER && info->si_sys_private) {
492 		/*
493 		 * Release the siglock to ensure proper locking order
494 		 * of timer locks outside of siglocks.  Note, we leave
495 		 * irqs disabled here, since the posix-timers code is
496 		 * about to disable them again anyway.
497 		 */
498 		spin_unlock(&tsk->sighand->siglock);
499 		do_schedule_next_timer(info);
500 		spin_lock(&tsk->sighand->siglock);
501 	}
502 	return signr;
503 }
504 
505 /*
506  * Tell a process that it has a new active signal..
507  *
508  * NOTE! we rely on the previous spin_lock to
509  * lock interrupts for us! We can only be called with
510  * "siglock" held, and the local interrupt must
511  * have been disabled when that got acquired!
512  *
513  * No need to set need_resched since signal event passing
514  * goes through ->blocked
515  */
516 void signal_wake_up(struct task_struct *t, int resume)
517 {
518 	unsigned int mask;
519 
520 	set_tsk_thread_flag(t, TIF_SIGPENDING);
521 
522 	/*
523 	 * For SIGKILL, we want to wake it up in the stopped/traced/killable
524 	 * case. We don't check t->state here because there is a race with it
525 	 * executing another processor and just now entering stopped state.
526 	 * By using wake_up_state, we ensure the process will wake up and
527 	 * handle its death signal.
528 	 */
529 	mask = TASK_INTERRUPTIBLE;
530 	if (resume)
531 		mask |= TASK_WAKEKILL;
532 	if (!wake_up_state(t, mask))
533 		kick_process(t);
534 }
535 
536 /*
537  * Remove signals in mask from the pending set and queue.
538  * Returns 1 if any signals were found.
539  *
540  * All callers must be holding the siglock.
541  *
542  * This version takes a sigset mask and looks at all signals,
543  * not just those in the first mask word.
544  */
545 static int rm_from_queue_full(sigset_t *mask, struct sigpending *s)
546 {
547 	struct sigqueue *q, *n;
548 	sigset_t m;
549 
550 	sigandsets(&m, mask, &s->signal);
551 	if (sigisemptyset(&m))
552 		return 0;
553 
554 	signandsets(&s->signal, &s->signal, mask);
555 	list_for_each_entry_safe(q, n, &s->list, list) {
556 		if (sigismember(mask, q->info.si_signo)) {
557 			list_del_init(&q->list);
558 			__sigqueue_free(q);
559 		}
560 	}
561 	return 1;
562 }
563 /*
564  * Remove signals in mask from the pending set and queue.
565  * Returns 1 if any signals were found.
566  *
567  * All callers must be holding the siglock.
568  */
569 static int rm_from_queue(unsigned long mask, struct sigpending *s)
570 {
571 	struct sigqueue *q, *n;
572 
573 	if (!sigtestsetmask(&s->signal, mask))
574 		return 0;
575 
576 	sigdelsetmask(&s->signal, mask);
577 	list_for_each_entry_safe(q, n, &s->list, list) {
578 		if (q->info.si_signo < SIGRTMIN &&
579 		    (mask & sigmask(q->info.si_signo))) {
580 			list_del_init(&q->list);
581 			__sigqueue_free(q);
582 		}
583 	}
584 	return 1;
585 }
586 
587 /*
588  * Bad permissions for sending the signal
589  * - the caller must hold at least the RCU read lock
590  */
591 static int check_kill_permission(int sig, struct siginfo *info,
592 				 struct task_struct *t)
593 {
594 	const struct cred *cred = current_cred(), *tcred;
595 	struct pid *sid;
596 	int error;
597 
598 	if (!valid_signal(sig))
599 		return -EINVAL;
600 
601 	if (info != SEND_SIG_NOINFO && (is_si_special(info) || SI_FROMKERNEL(info)))
602 		return 0;
603 
604 	error = audit_signal_info(sig, t); /* Let audit system see the signal */
605 	if (error)
606 		return error;
607 
608 	tcred = __task_cred(t);
609 	if ((cred->euid ^ tcred->suid) &&
610 	    (cred->euid ^ tcred->uid) &&
611 	    (cred->uid  ^ tcred->suid) &&
612 	    (cred->uid  ^ tcred->uid) &&
613 	    !capable(CAP_KILL)) {
614 		switch (sig) {
615 		case SIGCONT:
616 			sid = task_session(t);
617 			/*
618 			 * We don't return the error if sid == NULL. The
619 			 * task was unhashed, the caller must notice this.
620 			 */
621 			if (!sid || sid == task_session(current))
622 				break;
623 		default:
624 			return -EPERM;
625 		}
626 	}
627 
628 	return security_task_kill(t, info, sig, 0);
629 }
630 
631 /*
632  * Handle magic process-wide effects of stop/continue signals. Unlike
633  * the signal actions, these happen immediately at signal-generation
634  * time regardless of blocking, ignoring, or handling.  This does the
635  * actual continuing for SIGCONT, but not the actual stopping for stop
636  * signals. The process stop is done as a signal action for SIG_DFL.
637  *
638  * Returns true if the signal should be actually delivered, otherwise
639  * it should be dropped.
640  */
641 static int prepare_signal(int sig, struct task_struct *p, int from_ancestor_ns)
642 {
643 	struct signal_struct *signal = p->signal;
644 	struct task_struct *t;
645 
646 	if (unlikely(signal->flags & SIGNAL_GROUP_EXIT)) {
647 		/*
648 		 * The process is in the middle of dying, nothing to do.
649 		 */
650 	} else if (sig_kernel_stop(sig)) {
651 		/*
652 		 * This is a stop signal.  Remove SIGCONT from all queues.
653 		 */
654 		rm_from_queue(sigmask(SIGCONT), &signal->shared_pending);
655 		t = p;
656 		do {
657 			rm_from_queue(sigmask(SIGCONT), &t->pending);
658 		} while_each_thread(p, t);
659 	} else if (sig == SIGCONT) {
660 		unsigned int why;
661 		/*
662 		 * Remove all stop signals from all queues,
663 		 * and wake all threads.
664 		 */
665 		rm_from_queue(SIG_KERNEL_STOP_MASK, &signal->shared_pending);
666 		t = p;
667 		do {
668 			unsigned int state;
669 			rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
670 			/*
671 			 * If there is a handler for SIGCONT, we must make
672 			 * sure that no thread returns to user mode before
673 			 * we post the signal, in case it was the only
674 			 * thread eligible to run the signal handler--then
675 			 * it must not do anything between resuming and
676 			 * running the handler.  With the TIF_SIGPENDING
677 			 * flag set, the thread will pause and acquire the
678 			 * siglock that we hold now and until we've queued
679 			 * the pending signal.
680 			 *
681 			 * Wake up the stopped thread _after_ setting
682 			 * TIF_SIGPENDING
683 			 */
684 			state = __TASK_STOPPED;
685 			if (sig_user_defined(t, SIGCONT) && !sigismember(&t->blocked, SIGCONT)) {
686 				set_tsk_thread_flag(t, TIF_SIGPENDING);
687 				state |= TASK_INTERRUPTIBLE;
688 			}
689 			wake_up_state(t, state);
690 		} while_each_thread(p, t);
691 
692 		/*
693 		 * Notify the parent with CLD_CONTINUED if we were stopped.
694 		 *
695 		 * If we were in the middle of a group stop, we pretend it
696 		 * was already finished, and then continued. Since SIGCHLD
697 		 * doesn't queue we report only CLD_STOPPED, as if the next
698 		 * CLD_CONTINUED was dropped.
699 		 */
700 		why = 0;
701 		if (signal->flags & SIGNAL_STOP_STOPPED)
702 			why |= SIGNAL_CLD_CONTINUED;
703 		else if (signal->group_stop_count)
704 			why |= SIGNAL_CLD_STOPPED;
705 
706 		if (why) {
707 			/*
708 			 * The first thread which returns from finish_stop()
709 			 * will take ->siglock, notice SIGNAL_CLD_MASK, and
710 			 * notify its parent. See get_signal_to_deliver().
711 			 */
712 			signal->flags = why | SIGNAL_STOP_CONTINUED;
713 			signal->group_stop_count = 0;
714 			signal->group_exit_code = 0;
715 		} else {
716 			/*
717 			 * We are not stopped, but there could be a stop
718 			 * signal in the middle of being processed after
719 			 * being removed from the queue.  Clear that too.
720 			 */
721 			signal->flags &= ~SIGNAL_STOP_DEQUEUED;
722 		}
723 	}
724 
725 	return !sig_ignored(p, sig, from_ancestor_ns);
726 }
727 
728 /*
729  * Test if P wants to take SIG.  After we've checked all threads with this,
730  * it's equivalent to finding no threads not blocking SIG.  Any threads not
731  * blocking SIG were ruled out because they are not running and already
732  * have pending signals.  Such threads will dequeue from the shared queue
733  * as soon as they're available, so putting the signal on the shared queue
734  * will be equivalent to sending it to one such thread.
735  */
736 static inline int wants_signal(int sig, struct task_struct *p)
737 {
738 	if (sigismember(&p->blocked, sig))
739 		return 0;
740 	if (p->flags & PF_EXITING)
741 		return 0;
742 	if (sig == SIGKILL)
743 		return 1;
744 	if (task_is_stopped_or_traced(p))
745 		return 0;
746 	return task_curr(p) || !signal_pending(p);
747 }
748 
749 static void complete_signal(int sig, struct task_struct *p, int group)
750 {
751 	struct signal_struct *signal = p->signal;
752 	struct task_struct *t;
753 
754 	/*
755 	 * Now find a thread we can wake up to take the signal off the queue.
756 	 *
757 	 * If the main thread wants the signal, it gets first crack.
758 	 * Probably the least surprising to the average bear.
759 	 */
760 	if (wants_signal(sig, p))
761 		t = p;
762 	else if (!group || thread_group_empty(p))
763 		/*
764 		 * There is just one thread and it does not need to be woken.
765 		 * It will dequeue unblocked signals before it runs again.
766 		 */
767 		return;
768 	else {
769 		/*
770 		 * Otherwise try to find a suitable thread.
771 		 */
772 		t = signal->curr_target;
773 		while (!wants_signal(sig, t)) {
774 			t = next_thread(t);
775 			if (t == signal->curr_target)
776 				/*
777 				 * No thread needs to be woken.
778 				 * Any eligible threads will see
779 				 * the signal in the queue soon.
780 				 */
781 				return;
782 		}
783 		signal->curr_target = t;
784 	}
785 
786 	/*
787 	 * Found a killable thread.  If the signal will be fatal,
788 	 * then start taking the whole group down immediately.
789 	 */
790 	if (sig_fatal(p, sig) &&
791 	    !(signal->flags & (SIGNAL_UNKILLABLE | SIGNAL_GROUP_EXIT)) &&
792 	    !sigismember(&t->real_blocked, sig) &&
793 	    (sig == SIGKILL ||
794 	     !tracehook_consider_fatal_signal(t, sig))) {
795 		/*
796 		 * This signal will be fatal to the whole group.
797 		 */
798 		if (!sig_kernel_coredump(sig)) {
799 			/*
800 			 * Start a group exit and wake everybody up.
801 			 * This way we don't have other threads
802 			 * running and doing things after a slower
803 			 * thread has the fatal signal pending.
804 			 */
805 			signal->flags = SIGNAL_GROUP_EXIT;
806 			signal->group_exit_code = sig;
807 			signal->group_stop_count = 0;
808 			t = p;
809 			do {
810 				sigaddset(&t->pending.signal, SIGKILL);
811 				signal_wake_up(t, 1);
812 			} while_each_thread(p, t);
813 			return;
814 		}
815 	}
816 
817 	/*
818 	 * The signal is already in the shared-pending queue.
819 	 * Tell the chosen thread to wake up and dequeue it.
820 	 */
821 	signal_wake_up(t, sig == SIGKILL);
822 	return;
823 }
824 
825 static inline int legacy_queue(struct sigpending *signals, int sig)
826 {
827 	return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
828 }
829 
830 static int __send_signal(int sig, struct siginfo *info, struct task_struct *t,
831 			int group, int from_ancestor_ns)
832 {
833 	struct sigpending *pending;
834 	struct sigqueue *q;
835 	int override_rlimit;
836 
837 	trace_sched_signal_send(sig, t);
838 
839 	assert_spin_locked(&t->sighand->siglock);
840 
841 	if (!prepare_signal(sig, t, from_ancestor_ns))
842 		return 0;
843 
844 	pending = group ? &t->signal->shared_pending : &t->pending;
845 	/*
846 	 * Short-circuit ignored signals and support queuing
847 	 * exactly one non-rt signal, so that we can get more
848 	 * detailed information about the cause of the signal.
849 	 */
850 	if (legacy_queue(pending, sig))
851 		return 0;
852 	/*
853 	 * fast-pathed signals for kernel-internal things like SIGSTOP
854 	 * or SIGKILL.
855 	 */
856 	if (info == SEND_SIG_FORCED)
857 		goto out_set;
858 
859 	/* Real-time signals must be queued if sent by sigqueue, or
860 	   some other real-time mechanism.  It is implementation
861 	   defined whether kill() does so.  We attempt to do so, on
862 	   the principle of least surprise, but since kill is not
863 	   allowed to fail with EAGAIN when low on memory we just
864 	   make sure at least one signal gets delivered and don't
865 	   pass on the info struct.  */
866 
867 	if (sig < SIGRTMIN)
868 		override_rlimit = (is_si_special(info) || info->si_code >= 0);
869 	else
870 		override_rlimit = 0;
871 
872 	q = __sigqueue_alloc(t, GFP_ATOMIC | __GFP_NOTRACK_FALSE_POSITIVE,
873 		override_rlimit);
874 	if (q) {
875 		list_add_tail(&q->list, &pending->list);
876 		switch ((unsigned long) info) {
877 		case (unsigned long) SEND_SIG_NOINFO:
878 			q->info.si_signo = sig;
879 			q->info.si_errno = 0;
880 			q->info.si_code = SI_USER;
881 			q->info.si_pid = task_tgid_nr_ns(current,
882 							task_active_pid_ns(t));
883 			q->info.si_uid = current_uid();
884 			break;
885 		case (unsigned long) SEND_SIG_PRIV:
886 			q->info.si_signo = sig;
887 			q->info.si_errno = 0;
888 			q->info.si_code = SI_KERNEL;
889 			q->info.si_pid = 0;
890 			q->info.si_uid = 0;
891 			break;
892 		default:
893 			copy_siginfo(&q->info, info);
894 			if (from_ancestor_ns)
895 				q->info.si_pid = 0;
896 			break;
897 		}
898 	} else if (!is_si_special(info)) {
899 		if (sig >= SIGRTMIN && info->si_code != SI_USER)
900 		/*
901 		 * Queue overflow, abort.  We may abort if the signal was rt
902 		 * and sent by user using something other than kill().
903 		 */
904 			return -EAGAIN;
905 	}
906 
907 out_set:
908 	signalfd_notify(t, sig);
909 	sigaddset(&pending->signal, sig);
910 	complete_signal(sig, t, group);
911 	return 0;
912 }
913 
914 static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
915 			int group)
916 {
917 	int from_ancestor_ns = 0;
918 
919 #ifdef CONFIG_PID_NS
920 	if (!is_si_special(info) && SI_FROMUSER(info) &&
921 			task_pid_nr_ns(current, task_active_pid_ns(t)) <= 0)
922 		from_ancestor_ns = 1;
923 #endif
924 
925 	return __send_signal(sig, info, t, group, from_ancestor_ns);
926 }
927 
928 int print_fatal_signals;
929 
930 static void print_fatal_signal(struct pt_regs *regs, int signr)
931 {
932 	printk("%s/%d: potentially unexpected fatal signal %d.\n",
933 		current->comm, task_pid_nr(current), signr);
934 
935 #if defined(__i386__) && !defined(__arch_um__)
936 	printk("code at %08lx: ", regs->ip);
937 	{
938 		int i;
939 		for (i = 0; i < 16; i++) {
940 			unsigned char insn;
941 
942 			__get_user(insn, (unsigned char *)(regs->ip + i));
943 			printk("%02x ", insn);
944 		}
945 	}
946 #endif
947 	printk("\n");
948 	preempt_disable();
949 	show_regs(regs);
950 	preempt_enable();
951 }
952 
953 static int __init setup_print_fatal_signals(char *str)
954 {
955 	get_option (&str, &print_fatal_signals);
956 
957 	return 1;
958 }
959 
960 __setup("print-fatal-signals=", setup_print_fatal_signals);
961 
962 int
963 __group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
964 {
965 	return send_signal(sig, info, p, 1);
966 }
967 
968 static int
969 specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
970 {
971 	return send_signal(sig, info, t, 0);
972 }
973 
974 /*
975  * Force a signal that the process can't ignore: if necessary
976  * we unblock the signal and change any SIG_IGN to SIG_DFL.
977  *
978  * Note: If we unblock the signal, we always reset it to SIG_DFL,
979  * since we do not want to have a signal handler that was blocked
980  * be invoked when user space had explicitly blocked it.
981  *
982  * We don't want to have recursive SIGSEGV's etc, for example,
983  * that is why we also clear SIGNAL_UNKILLABLE.
984  */
985 int
986 force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
987 {
988 	unsigned long int flags;
989 	int ret, blocked, ignored;
990 	struct k_sigaction *action;
991 
992 	spin_lock_irqsave(&t->sighand->siglock, flags);
993 	action = &t->sighand->action[sig-1];
994 	ignored = action->sa.sa_handler == SIG_IGN;
995 	blocked = sigismember(&t->blocked, sig);
996 	if (blocked || ignored) {
997 		action->sa.sa_handler = SIG_DFL;
998 		if (blocked) {
999 			sigdelset(&t->blocked, sig);
1000 			recalc_sigpending_and_wake(t);
1001 		}
1002 	}
1003 	if (action->sa.sa_handler == SIG_DFL)
1004 		t->signal->flags &= ~SIGNAL_UNKILLABLE;
1005 	ret = specific_send_sig_info(sig, info, t);
1006 	spin_unlock_irqrestore(&t->sighand->siglock, flags);
1007 
1008 	return ret;
1009 }
1010 
1011 void
1012 force_sig_specific(int sig, struct task_struct *t)
1013 {
1014 	force_sig_info(sig, SEND_SIG_FORCED, t);
1015 }
1016 
1017 /*
1018  * Nuke all other threads in the group.
1019  */
1020 void zap_other_threads(struct task_struct *p)
1021 {
1022 	struct task_struct *t;
1023 
1024 	p->signal->group_stop_count = 0;
1025 
1026 	for (t = next_thread(p); t != p; t = next_thread(t)) {
1027 		/*
1028 		 * Don't bother with already dead threads
1029 		 */
1030 		if (t->exit_state)
1031 			continue;
1032 
1033 		/* SIGKILL will be handled before any pending SIGSTOP */
1034 		sigaddset(&t->pending.signal, SIGKILL);
1035 		signal_wake_up(t, 1);
1036 	}
1037 }
1038 
1039 int __fatal_signal_pending(struct task_struct *tsk)
1040 {
1041 	return sigismember(&tsk->pending.signal, SIGKILL);
1042 }
1043 EXPORT_SYMBOL(__fatal_signal_pending);
1044 
1045 struct sighand_struct *lock_task_sighand(struct task_struct *tsk, unsigned long *flags)
1046 {
1047 	struct sighand_struct *sighand;
1048 
1049 	rcu_read_lock();
1050 	for (;;) {
1051 		sighand = rcu_dereference(tsk->sighand);
1052 		if (unlikely(sighand == NULL))
1053 			break;
1054 
1055 		spin_lock_irqsave(&sighand->siglock, *flags);
1056 		if (likely(sighand == tsk->sighand))
1057 			break;
1058 		spin_unlock_irqrestore(&sighand->siglock, *flags);
1059 	}
1060 	rcu_read_unlock();
1061 
1062 	return sighand;
1063 }
1064 
1065 /*
1066  * send signal info to all the members of a group
1067  * - the caller must hold the RCU read lock at least
1068  */
1069 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1070 {
1071 	unsigned long flags;
1072 	int ret;
1073 
1074 	ret = check_kill_permission(sig, info, p);
1075 
1076 	if (!ret && sig) {
1077 		ret = -ESRCH;
1078 		if (lock_task_sighand(p, &flags)) {
1079 			ret = __group_send_sig_info(sig, info, p);
1080 			unlock_task_sighand(p, &flags);
1081 		}
1082 	}
1083 
1084 	return ret;
1085 }
1086 
1087 /*
1088  * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1089  * control characters do (^C, ^Z etc)
1090  * - the caller must hold at least a readlock on tasklist_lock
1091  */
1092 int __kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
1093 {
1094 	struct task_struct *p = NULL;
1095 	int retval, success;
1096 
1097 	success = 0;
1098 	retval = -ESRCH;
1099 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1100 		int err = group_send_sig_info(sig, info, p);
1101 		success |= !err;
1102 		retval = err;
1103 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1104 	return success ? 0 : retval;
1105 }
1106 
1107 int kill_pid_info(int sig, struct siginfo *info, struct pid *pid)
1108 {
1109 	int error = -ESRCH;
1110 	struct task_struct *p;
1111 
1112 	rcu_read_lock();
1113 retry:
1114 	p = pid_task(pid, PIDTYPE_PID);
1115 	if (p) {
1116 		error = group_send_sig_info(sig, info, p);
1117 		if (unlikely(error == -ESRCH))
1118 			/*
1119 			 * The task was unhashed in between, try again.
1120 			 * If it is dead, pid_task() will return NULL,
1121 			 * if we race with de_thread() it will find the
1122 			 * new leader.
1123 			 */
1124 			goto retry;
1125 	}
1126 	rcu_read_unlock();
1127 
1128 	return error;
1129 }
1130 
1131 int
1132 kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1133 {
1134 	int error;
1135 	rcu_read_lock();
1136 	error = kill_pid_info(sig, info, find_vpid(pid));
1137 	rcu_read_unlock();
1138 	return error;
1139 }
1140 
1141 /* like kill_pid_info(), but doesn't use uid/euid of "current" */
1142 int kill_pid_info_as_uid(int sig, struct siginfo *info, struct pid *pid,
1143 		      uid_t uid, uid_t euid, u32 secid)
1144 {
1145 	int ret = -EINVAL;
1146 	struct task_struct *p;
1147 	const struct cred *pcred;
1148 
1149 	if (!valid_signal(sig))
1150 		return ret;
1151 
1152 	read_lock(&tasklist_lock);
1153 	p = pid_task(pid, PIDTYPE_PID);
1154 	if (!p) {
1155 		ret = -ESRCH;
1156 		goto out_unlock;
1157 	}
1158 	pcred = __task_cred(p);
1159 	if ((info == SEND_SIG_NOINFO ||
1160 	     (!is_si_special(info) && SI_FROMUSER(info))) &&
1161 	    euid != pcred->suid && euid != pcred->uid &&
1162 	    uid  != pcred->suid && uid  != pcred->uid) {
1163 		ret = -EPERM;
1164 		goto out_unlock;
1165 	}
1166 	ret = security_task_kill(p, info, sig, secid);
1167 	if (ret)
1168 		goto out_unlock;
1169 	if (sig && p->sighand) {
1170 		unsigned long flags;
1171 		spin_lock_irqsave(&p->sighand->siglock, flags);
1172 		ret = __send_signal(sig, info, p, 1, 0);
1173 		spin_unlock_irqrestore(&p->sighand->siglock, flags);
1174 	}
1175 out_unlock:
1176 	read_unlock(&tasklist_lock);
1177 	return ret;
1178 }
1179 EXPORT_SYMBOL_GPL(kill_pid_info_as_uid);
1180 
1181 /*
1182  * kill_something_info() interprets pid in interesting ways just like kill(2).
1183  *
1184  * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1185  * is probably wrong.  Should make it like BSD or SYSV.
1186  */
1187 
1188 static int kill_something_info(int sig, struct siginfo *info, pid_t pid)
1189 {
1190 	int ret;
1191 
1192 	if (pid > 0) {
1193 		rcu_read_lock();
1194 		ret = kill_pid_info(sig, info, find_vpid(pid));
1195 		rcu_read_unlock();
1196 		return ret;
1197 	}
1198 
1199 	read_lock(&tasklist_lock);
1200 	if (pid != -1) {
1201 		ret = __kill_pgrp_info(sig, info,
1202 				pid ? find_vpid(-pid) : task_pgrp(current));
1203 	} else {
1204 		int retval = 0, count = 0;
1205 		struct task_struct * p;
1206 
1207 		for_each_process(p) {
1208 			if (task_pid_vnr(p) > 1 &&
1209 					!same_thread_group(p, current)) {
1210 				int err = group_send_sig_info(sig, info, p);
1211 				++count;
1212 				if (err != -EPERM)
1213 					retval = err;
1214 			}
1215 		}
1216 		ret = count ? retval : -ESRCH;
1217 	}
1218 	read_unlock(&tasklist_lock);
1219 
1220 	return ret;
1221 }
1222 
1223 /*
1224  * These are for backward compatibility with the rest of the kernel source.
1225  */
1226 
1227 /*
1228  * The caller must ensure the task can't exit.
1229  */
1230 int
1231 send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1232 {
1233 	int ret;
1234 	unsigned long flags;
1235 
1236 	/*
1237 	 * Make sure legacy kernel users don't send in bad values
1238 	 * (normal paths check this in check_kill_permission).
1239 	 */
1240 	if (!valid_signal(sig))
1241 		return -EINVAL;
1242 
1243 	spin_lock_irqsave(&p->sighand->siglock, flags);
1244 	ret = specific_send_sig_info(sig, info, p);
1245 	spin_unlock_irqrestore(&p->sighand->siglock, flags);
1246 	return ret;
1247 }
1248 
1249 #define __si_special(priv) \
1250 	((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1251 
1252 int
1253 send_sig(int sig, struct task_struct *p, int priv)
1254 {
1255 	return send_sig_info(sig, __si_special(priv), p);
1256 }
1257 
1258 void
1259 force_sig(int sig, struct task_struct *p)
1260 {
1261 	force_sig_info(sig, SEND_SIG_PRIV, p);
1262 }
1263 
1264 /*
1265  * When things go south during signal handling, we
1266  * will force a SIGSEGV. And if the signal that caused
1267  * the problem was already a SIGSEGV, we'll want to
1268  * make sure we don't even try to deliver the signal..
1269  */
1270 int
1271 force_sigsegv(int sig, struct task_struct *p)
1272 {
1273 	if (sig == SIGSEGV) {
1274 		unsigned long flags;
1275 		spin_lock_irqsave(&p->sighand->siglock, flags);
1276 		p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1277 		spin_unlock_irqrestore(&p->sighand->siglock, flags);
1278 	}
1279 	force_sig(SIGSEGV, p);
1280 	return 0;
1281 }
1282 
1283 int kill_pgrp(struct pid *pid, int sig, int priv)
1284 {
1285 	int ret;
1286 
1287 	read_lock(&tasklist_lock);
1288 	ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1289 	read_unlock(&tasklist_lock);
1290 
1291 	return ret;
1292 }
1293 EXPORT_SYMBOL(kill_pgrp);
1294 
1295 int kill_pid(struct pid *pid, int sig, int priv)
1296 {
1297 	return kill_pid_info(sig, __si_special(priv), pid);
1298 }
1299 EXPORT_SYMBOL(kill_pid);
1300 
1301 /*
1302  * These functions support sending signals using preallocated sigqueue
1303  * structures.  This is needed "because realtime applications cannot
1304  * afford to lose notifications of asynchronous events, like timer
1305  * expirations or I/O completions".  In the case of Posix Timers
1306  * we allocate the sigqueue structure from the timer_create.  If this
1307  * allocation fails we are able to report the failure to the application
1308  * with an EAGAIN error.
1309  */
1310 
1311 struct sigqueue *sigqueue_alloc(void)
1312 {
1313 	struct sigqueue *q;
1314 
1315 	if ((q = __sigqueue_alloc(current, GFP_KERNEL, 0)))
1316 		q->flags |= SIGQUEUE_PREALLOC;
1317 	return(q);
1318 }
1319 
1320 void sigqueue_free(struct sigqueue *q)
1321 {
1322 	unsigned long flags;
1323 	spinlock_t *lock = &current->sighand->siglock;
1324 
1325 	BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1326 	/*
1327 	 * We must hold ->siglock while testing q->list
1328 	 * to serialize with collect_signal() or with
1329 	 * __exit_signal()->flush_sigqueue().
1330 	 */
1331 	spin_lock_irqsave(lock, flags);
1332 	q->flags &= ~SIGQUEUE_PREALLOC;
1333 	/*
1334 	 * If it is queued it will be freed when dequeued,
1335 	 * like the "regular" sigqueue.
1336 	 */
1337 	if (!list_empty(&q->list))
1338 		q = NULL;
1339 	spin_unlock_irqrestore(lock, flags);
1340 
1341 	if (q)
1342 		__sigqueue_free(q);
1343 }
1344 
1345 int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group)
1346 {
1347 	int sig = q->info.si_signo;
1348 	struct sigpending *pending;
1349 	unsigned long flags;
1350 	int ret;
1351 
1352 	BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1353 
1354 	ret = -1;
1355 	if (!likely(lock_task_sighand(t, &flags)))
1356 		goto ret;
1357 
1358 	ret = 1; /* the signal is ignored */
1359 	if (!prepare_signal(sig, t, 0))
1360 		goto out;
1361 
1362 	ret = 0;
1363 	if (unlikely(!list_empty(&q->list))) {
1364 		/*
1365 		 * If an SI_TIMER entry is already queue just increment
1366 		 * the overrun count.
1367 		 */
1368 		BUG_ON(q->info.si_code != SI_TIMER);
1369 		q->info.si_overrun++;
1370 		goto out;
1371 	}
1372 	q->info.si_overrun = 0;
1373 
1374 	signalfd_notify(t, sig);
1375 	pending = group ? &t->signal->shared_pending : &t->pending;
1376 	list_add_tail(&q->list, &pending->list);
1377 	sigaddset(&pending->signal, sig);
1378 	complete_signal(sig, t, group);
1379 out:
1380 	unlock_task_sighand(t, &flags);
1381 ret:
1382 	return ret;
1383 }
1384 
1385 /*
1386  * Wake up any threads in the parent blocked in wait* syscalls.
1387  */
1388 static inline void __wake_up_parent(struct task_struct *p,
1389 				    struct task_struct *parent)
1390 {
1391 	wake_up_interruptible_sync(&parent->signal->wait_chldexit);
1392 }
1393 
1394 /*
1395  * Let a parent know about the death of a child.
1396  * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1397  *
1398  * Returns -1 if our parent ignored us and so we've switched to
1399  * self-reaping, or else @sig.
1400  */
1401 int do_notify_parent(struct task_struct *tsk, int sig)
1402 {
1403 	struct siginfo info;
1404 	unsigned long flags;
1405 	struct sighand_struct *psig;
1406 	int ret = sig;
1407 
1408 	BUG_ON(sig == -1);
1409 
1410  	/* do_notify_parent_cldstop should have been called instead.  */
1411  	BUG_ON(task_is_stopped_or_traced(tsk));
1412 
1413 	BUG_ON(!task_ptrace(tsk) &&
1414 	       (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1415 
1416 	info.si_signo = sig;
1417 	info.si_errno = 0;
1418 	/*
1419 	 * we are under tasklist_lock here so our parent is tied to
1420 	 * us and cannot exit and release its namespace.
1421 	 *
1422 	 * the only it can is to switch its nsproxy with sys_unshare,
1423 	 * bu uncharing pid namespaces is not allowed, so we'll always
1424 	 * see relevant namespace
1425 	 *
1426 	 * write_lock() currently calls preempt_disable() which is the
1427 	 * same as rcu_read_lock(), but according to Oleg, this is not
1428 	 * correct to rely on this
1429 	 */
1430 	rcu_read_lock();
1431 	info.si_pid = task_pid_nr_ns(tsk, tsk->parent->nsproxy->pid_ns);
1432 	info.si_uid = __task_cred(tsk)->uid;
1433 	rcu_read_unlock();
1434 
1435 	info.si_utime = cputime_to_clock_t(cputime_add(tsk->utime,
1436 				tsk->signal->utime));
1437 	info.si_stime = cputime_to_clock_t(cputime_add(tsk->stime,
1438 				tsk->signal->stime));
1439 
1440 	info.si_status = tsk->exit_code & 0x7f;
1441 	if (tsk->exit_code & 0x80)
1442 		info.si_code = CLD_DUMPED;
1443 	else if (tsk->exit_code & 0x7f)
1444 		info.si_code = CLD_KILLED;
1445 	else {
1446 		info.si_code = CLD_EXITED;
1447 		info.si_status = tsk->exit_code >> 8;
1448 	}
1449 
1450 	psig = tsk->parent->sighand;
1451 	spin_lock_irqsave(&psig->siglock, flags);
1452 	if (!task_ptrace(tsk) && sig == SIGCHLD &&
1453 	    (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1454 	     (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1455 		/*
1456 		 * We are exiting and our parent doesn't care.  POSIX.1
1457 		 * defines special semantics for setting SIGCHLD to SIG_IGN
1458 		 * or setting the SA_NOCLDWAIT flag: we should be reaped
1459 		 * automatically and not left for our parent's wait4 call.
1460 		 * Rather than having the parent do it as a magic kind of
1461 		 * signal handler, we just set this to tell do_exit that we
1462 		 * can be cleaned up without becoming a zombie.  Note that
1463 		 * we still call __wake_up_parent in this case, because a
1464 		 * blocked sys_wait4 might now return -ECHILD.
1465 		 *
1466 		 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1467 		 * is implementation-defined: we do (if you don't want
1468 		 * it, just use SIG_IGN instead).
1469 		 */
1470 		ret = tsk->exit_signal = -1;
1471 		if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1472 			sig = -1;
1473 	}
1474 	if (valid_signal(sig) && sig > 0)
1475 		__group_send_sig_info(sig, &info, tsk->parent);
1476 	__wake_up_parent(tsk, tsk->parent);
1477 	spin_unlock_irqrestore(&psig->siglock, flags);
1478 
1479 	return ret;
1480 }
1481 
1482 static void do_notify_parent_cldstop(struct task_struct *tsk, int why)
1483 {
1484 	struct siginfo info;
1485 	unsigned long flags;
1486 	struct task_struct *parent;
1487 	struct sighand_struct *sighand;
1488 
1489 	if (task_ptrace(tsk))
1490 		parent = tsk->parent;
1491 	else {
1492 		tsk = tsk->group_leader;
1493 		parent = tsk->real_parent;
1494 	}
1495 
1496 	info.si_signo = SIGCHLD;
1497 	info.si_errno = 0;
1498 	/*
1499 	 * see comment in do_notify_parent() abot the following 3 lines
1500 	 */
1501 	rcu_read_lock();
1502 	info.si_pid = task_pid_nr_ns(tsk, parent->nsproxy->pid_ns);
1503 	info.si_uid = __task_cred(tsk)->uid;
1504 	rcu_read_unlock();
1505 
1506 	info.si_utime = cputime_to_clock_t(tsk->utime);
1507 	info.si_stime = cputime_to_clock_t(tsk->stime);
1508 
1509  	info.si_code = why;
1510  	switch (why) {
1511  	case CLD_CONTINUED:
1512  		info.si_status = SIGCONT;
1513  		break;
1514  	case CLD_STOPPED:
1515  		info.si_status = tsk->signal->group_exit_code & 0x7f;
1516  		break;
1517  	case CLD_TRAPPED:
1518  		info.si_status = tsk->exit_code & 0x7f;
1519  		break;
1520  	default:
1521  		BUG();
1522  	}
1523 
1524 	sighand = parent->sighand;
1525 	spin_lock_irqsave(&sighand->siglock, flags);
1526 	if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1527 	    !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1528 		__group_send_sig_info(SIGCHLD, &info, parent);
1529 	/*
1530 	 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1531 	 */
1532 	__wake_up_parent(tsk, parent);
1533 	spin_unlock_irqrestore(&sighand->siglock, flags);
1534 }
1535 
1536 static inline int may_ptrace_stop(void)
1537 {
1538 	if (!likely(task_ptrace(current)))
1539 		return 0;
1540 	/*
1541 	 * Are we in the middle of do_coredump?
1542 	 * If so and our tracer is also part of the coredump stopping
1543 	 * is a deadlock situation, and pointless because our tracer
1544 	 * is dead so don't allow us to stop.
1545 	 * If SIGKILL was already sent before the caller unlocked
1546 	 * ->siglock we must see ->core_state != NULL. Otherwise it
1547 	 * is safe to enter schedule().
1548 	 */
1549 	if (unlikely(current->mm->core_state) &&
1550 	    unlikely(current->mm == current->parent->mm))
1551 		return 0;
1552 
1553 	return 1;
1554 }
1555 
1556 /*
1557  * Return nonzero if there is a SIGKILL that should be waking us up.
1558  * Called with the siglock held.
1559  */
1560 static int sigkill_pending(struct task_struct *tsk)
1561 {
1562 	return	sigismember(&tsk->pending.signal, SIGKILL) ||
1563 		sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
1564 }
1565 
1566 /*
1567  * This must be called with current->sighand->siglock held.
1568  *
1569  * This should be the path for all ptrace stops.
1570  * We always set current->last_siginfo while stopped here.
1571  * That makes it a way to test a stopped process for
1572  * being ptrace-stopped vs being job-control-stopped.
1573  *
1574  * If we actually decide not to stop at all because the tracer
1575  * is gone, we keep current->exit_code unless clear_code.
1576  */
1577 static void ptrace_stop(int exit_code, int clear_code, siginfo_t *info)
1578 {
1579 	if (arch_ptrace_stop_needed(exit_code, info)) {
1580 		/*
1581 		 * The arch code has something special to do before a
1582 		 * ptrace stop.  This is allowed to block, e.g. for faults
1583 		 * on user stack pages.  We can't keep the siglock while
1584 		 * calling arch_ptrace_stop, so we must release it now.
1585 		 * To preserve proper semantics, we must do this before
1586 		 * any signal bookkeeping like checking group_stop_count.
1587 		 * Meanwhile, a SIGKILL could come in before we retake the
1588 		 * siglock.  That must prevent us from sleeping in TASK_TRACED.
1589 		 * So after regaining the lock, we must check for SIGKILL.
1590 		 */
1591 		spin_unlock_irq(&current->sighand->siglock);
1592 		arch_ptrace_stop(exit_code, info);
1593 		spin_lock_irq(&current->sighand->siglock);
1594 		if (sigkill_pending(current))
1595 			return;
1596 	}
1597 
1598 	/*
1599 	 * If there is a group stop in progress,
1600 	 * we must participate in the bookkeeping.
1601 	 */
1602 	if (current->signal->group_stop_count > 0)
1603 		--current->signal->group_stop_count;
1604 
1605 	current->last_siginfo = info;
1606 	current->exit_code = exit_code;
1607 
1608 	/* Let the debugger run.  */
1609 	__set_current_state(TASK_TRACED);
1610 	spin_unlock_irq(&current->sighand->siglock);
1611 	read_lock(&tasklist_lock);
1612 	if (may_ptrace_stop()) {
1613 		do_notify_parent_cldstop(current, CLD_TRAPPED);
1614 		/*
1615 		 * Don't want to allow preemption here, because
1616 		 * sys_ptrace() needs this task to be inactive.
1617 		 *
1618 		 * XXX: implement read_unlock_no_resched().
1619 		 */
1620 		preempt_disable();
1621 		read_unlock(&tasklist_lock);
1622 		preempt_enable_no_resched();
1623 		schedule();
1624 	} else {
1625 		/*
1626 		 * By the time we got the lock, our tracer went away.
1627 		 * Don't drop the lock yet, another tracer may come.
1628 		 */
1629 		__set_current_state(TASK_RUNNING);
1630 		if (clear_code)
1631 			current->exit_code = 0;
1632 		read_unlock(&tasklist_lock);
1633 	}
1634 
1635 	/*
1636 	 * While in TASK_TRACED, we were considered "frozen enough".
1637 	 * Now that we woke up, it's crucial if we're supposed to be
1638 	 * frozen that we freeze now before running anything substantial.
1639 	 */
1640 	try_to_freeze();
1641 
1642 	/*
1643 	 * We are back.  Now reacquire the siglock before touching
1644 	 * last_siginfo, so that we are sure to have synchronized with
1645 	 * any signal-sending on another CPU that wants to examine it.
1646 	 */
1647 	spin_lock_irq(&current->sighand->siglock);
1648 	current->last_siginfo = NULL;
1649 
1650 	/*
1651 	 * Queued signals ignored us while we were stopped for tracing.
1652 	 * So check for any that we should take before resuming user mode.
1653 	 * This sets TIF_SIGPENDING, but never clears it.
1654 	 */
1655 	recalc_sigpending_tsk(current);
1656 }
1657 
1658 void ptrace_notify(int exit_code)
1659 {
1660 	siginfo_t info;
1661 
1662 	BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1663 
1664 	memset(&info, 0, sizeof info);
1665 	info.si_signo = SIGTRAP;
1666 	info.si_code = exit_code;
1667 	info.si_pid = task_pid_vnr(current);
1668 	info.si_uid = current_uid();
1669 
1670 	/* Let the debugger run.  */
1671 	spin_lock_irq(&current->sighand->siglock);
1672 	ptrace_stop(exit_code, 1, &info);
1673 	spin_unlock_irq(&current->sighand->siglock);
1674 }
1675 
1676 static void
1677 finish_stop(int stop_count)
1678 {
1679 	/*
1680 	 * If there are no other threads in the group, or if there is
1681 	 * a group stop in progress and we are the last to stop,
1682 	 * report to the parent.  When ptraced, every thread reports itself.
1683 	 */
1684 	if (tracehook_notify_jctl(stop_count == 0, CLD_STOPPED)) {
1685 		read_lock(&tasklist_lock);
1686 		do_notify_parent_cldstop(current, CLD_STOPPED);
1687 		read_unlock(&tasklist_lock);
1688 	}
1689 
1690 	do {
1691 		schedule();
1692 	} while (try_to_freeze());
1693 	/*
1694 	 * Now we don't run again until continued.
1695 	 */
1696 	current->exit_code = 0;
1697 }
1698 
1699 /*
1700  * This performs the stopping for SIGSTOP and other stop signals.
1701  * We have to stop all threads in the thread group.
1702  * Returns nonzero if we've actually stopped and released the siglock.
1703  * Returns zero if we didn't stop and still hold the siglock.
1704  */
1705 static int do_signal_stop(int signr)
1706 {
1707 	struct signal_struct *sig = current->signal;
1708 	int stop_count;
1709 
1710 	if (sig->group_stop_count > 0) {
1711 		/*
1712 		 * There is a group stop in progress.  We don't need to
1713 		 * start another one.
1714 		 */
1715 		stop_count = --sig->group_stop_count;
1716 	} else {
1717 		struct task_struct *t;
1718 
1719 		if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED) ||
1720 		    unlikely(signal_group_exit(sig)))
1721 			return 0;
1722 		/*
1723 		 * There is no group stop already in progress.
1724 		 * We must initiate one now.
1725 		 */
1726 		sig->group_exit_code = signr;
1727 
1728 		stop_count = 0;
1729 		for (t = next_thread(current); t != current; t = next_thread(t))
1730 			/*
1731 			 * Setting state to TASK_STOPPED for a group
1732 			 * stop is always done with the siglock held,
1733 			 * so this check has no races.
1734 			 */
1735 			if (!(t->flags & PF_EXITING) &&
1736 			    !task_is_stopped_or_traced(t)) {
1737 				stop_count++;
1738 				signal_wake_up(t, 0);
1739 			}
1740 		sig->group_stop_count = stop_count;
1741 	}
1742 
1743 	if (stop_count == 0)
1744 		sig->flags = SIGNAL_STOP_STOPPED;
1745 	current->exit_code = sig->group_exit_code;
1746 	__set_current_state(TASK_STOPPED);
1747 
1748 	spin_unlock_irq(&current->sighand->siglock);
1749 	finish_stop(stop_count);
1750 	return 1;
1751 }
1752 
1753 static int ptrace_signal(int signr, siginfo_t *info,
1754 			 struct pt_regs *regs, void *cookie)
1755 {
1756 	if (!task_ptrace(current))
1757 		return signr;
1758 
1759 	ptrace_signal_deliver(regs, cookie);
1760 
1761 	/* Let the debugger run.  */
1762 	ptrace_stop(signr, 0, info);
1763 
1764 	/* We're back.  Did the debugger cancel the sig?  */
1765 	signr = current->exit_code;
1766 	if (signr == 0)
1767 		return signr;
1768 
1769 	current->exit_code = 0;
1770 
1771 	/* Update the siginfo structure if the signal has
1772 	   changed.  If the debugger wanted something
1773 	   specific in the siginfo structure then it should
1774 	   have updated *info via PTRACE_SETSIGINFO.  */
1775 	if (signr != info->si_signo) {
1776 		info->si_signo = signr;
1777 		info->si_errno = 0;
1778 		info->si_code = SI_USER;
1779 		info->si_pid = task_pid_vnr(current->parent);
1780 		info->si_uid = task_uid(current->parent);
1781 	}
1782 
1783 	/* If the (new) signal is now blocked, requeue it.  */
1784 	if (sigismember(&current->blocked, signr)) {
1785 		specific_send_sig_info(signr, info, current);
1786 		signr = 0;
1787 	}
1788 
1789 	return signr;
1790 }
1791 
1792 int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
1793 			  struct pt_regs *regs, void *cookie)
1794 {
1795 	struct sighand_struct *sighand = current->sighand;
1796 	struct signal_struct *signal = current->signal;
1797 	int signr;
1798 
1799 relock:
1800 	/*
1801 	 * We'll jump back here after any time we were stopped in TASK_STOPPED.
1802 	 * While in TASK_STOPPED, we were considered "frozen enough".
1803 	 * Now that we woke up, it's crucial if we're supposed to be
1804 	 * frozen that we freeze now before running anything substantial.
1805 	 */
1806 	try_to_freeze();
1807 
1808 	spin_lock_irq(&sighand->siglock);
1809 	/*
1810 	 * Every stopped thread goes here after wakeup. Check to see if
1811 	 * we should notify the parent, prepare_signal(SIGCONT) encodes
1812 	 * the CLD_ si_code into SIGNAL_CLD_MASK bits.
1813 	 */
1814 	if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
1815 		int why = (signal->flags & SIGNAL_STOP_CONTINUED)
1816 				? CLD_CONTINUED : CLD_STOPPED;
1817 		signal->flags &= ~SIGNAL_CLD_MASK;
1818 		spin_unlock_irq(&sighand->siglock);
1819 
1820 		if (unlikely(!tracehook_notify_jctl(1, why)))
1821 			goto relock;
1822 
1823 		read_lock(&tasklist_lock);
1824 		do_notify_parent_cldstop(current->group_leader, why);
1825 		read_unlock(&tasklist_lock);
1826 		goto relock;
1827 	}
1828 
1829 	for (;;) {
1830 		struct k_sigaction *ka;
1831 
1832 		if (unlikely(signal->group_stop_count > 0) &&
1833 		    do_signal_stop(0))
1834 			goto relock;
1835 
1836 		/*
1837 		 * Tracing can induce an artifical signal and choose sigaction.
1838 		 * The return value in @signr determines the default action,
1839 		 * but @info->si_signo is the signal number we will report.
1840 		 */
1841 		signr = tracehook_get_signal(current, regs, info, return_ka);
1842 		if (unlikely(signr < 0))
1843 			goto relock;
1844 		if (unlikely(signr != 0))
1845 			ka = return_ka;
1846 		else {
1847 			signr = dequeue_signal(current, &current->blocked,
1848 					       info);
1849 
1850 			if (!signr)
1851 				break; /* will return 0 */
1852 
1853 			if (signr != SIGKILL) {
1854 				signr = ptrace_signal(signr, info,
1855 						      regs, cookie);
1856 				if (!signr)
1857 					continue;
1858 			}
1859 
1860 			ka = &sighand->action[signr-1];
1861 		}
1862 
1863 		if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
1864 			continue;
1865 		if (ka->sa.sa_handler != SIG_DFL) {
1866 			/* Run the handler.  */
1867 			*return_ka = *ka;
1868 
1869 			if (ka->sa.sa_flags & SA_ONESHOT)
1870 				ka->sa.sa_handler = SIG_DFL;
1871 
1872 			break; /* will return non-zero "signr" value */
1873 		}
1874 
1875 		/*
1876 		 * Now we are doing the default action for this signal.
1877 		 */
1878 		if (sig_kernel_ignore(signr)) /* Default is nothing. */
1879 			continue;
1880 
1881 		/*
1882 		 * Global init gets no signals it doesn't want.
1883 		 * Container-init gets no signals it doesn't want from same
1884 		 * container.
1885 		 *
1886 		 * Note that if global/container-init sees a sig_kernel_only()
1887 		 * signal here, the signal must have been generated internally
1888 		 * or must have come from an ancestor namespace. In either
1889 		 * case, the signal cannot be dropped.
1890 		 */
1891 		if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
1892 				!sig_kernel_only(signr))
1893 			continue;
1894 
1895 		if (sig_kernel_stop(signr)) {
1896 			/*
1897 			 * The default action is to stop all threads in
1898 			 * the thread group.  The job control signals
1899 			 * do nothing in an orphaned pgrp, but SIGSTOP
1900 			 * always works.  Note that siglock needs to be
1901 			 * dropped during the call to is_orphaned_pgrp()
1902 			 * because of lock ordering with tasklist_lock.
1903 			 * This allows an intervening SIGCONT to be posted.
1904 			 * We need to check for that and bail out if necessary.
1905 			 */
1906 			if (signr != SIGSTOP) {
1907 				spin_unlock_irq(&sighand->siglock);
1908 
1909 				/* signals can be posted during this window */
1910 
1911 				if (is_current_pgrp_orphaned())
1912 					goto relock;
1913 
1914 				spin_lock_irq(&sighand->siglock);
1915 			}
1916 
1917 			if (likely(do_signal_stop(info->si_signo))) {
1918 				/* It released the siglock.  */
1919 				goto relock;
1920 			}
1921 
1922 			/*
1923 			 * We didn't actually stop, due to a race
1924 			 * with SIGCONT or something like that.
1925 			 */
1926 			continue;
1927 		}
1928 
1929 		spin_unlock_irq(&sighand->siglock);
1930 
1931 		/*
1932 		 * Anything else is fatal, maybe with a core dump.
1933 		 */
1934 		current->flags |= PF_SIGNALED;
1935 
1936 		if (sig_kernel_coredump(signr)) {
1937 			if (print_fatal_signals)
1938 				print_fatal_signal(regs, info->si_signo);
1939 			/*
1940 			 * If it was able to dump core, this kills all
1941 			 * other threads in the group and synchronizes with
1942 			 * their demise.  If we lost the race with another
1943 			 * thread getting here, it set group_exit_code
1944 			 * first and our do_group_exit call below will use
1945 			 * that value and ignore the one we pass it.
1946 			 */
1947 			do_coredump(info->si_signo, info->si_signo, regs);
1948 		}
1949 
1950 		/*
1951 		 * Death signals, no core dump.
1952 		 */
1953 		do_group_exit(info->si_signo);
1954 		/* NOTREACHED */
1955 	}
1956 	spin_unlock_irq(&sighand->siglock);
1957 	return signr;
1958 }
1959 
1960 void exit_signals(struct task_struct *tsk)
1961 {
1962 	int group_stop = 0;
1963 	struct task_struct *t;
1964 
1965 	if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
1966 		tsk->flags |= PF_EXITING;
1967 		return;
1968 	}
1969 
1970 	spin_lock_irq(&tsk->sighand->siglock);
1971 	/*
1972 	 * From now this task is not visible for group-wide signals,
1973 	 * see wants_signal(), do_signal_stop().
1974 	 */
1975 	tsk->flags |= PF_EXITING;
1976 	if (!signal_pending(tsk))
1977 		goto out;
1978 
1979 	/* It could be that __group_complete_signal() choose us to
1980 	 * notify about group-wide signal. Another thread should be
1981 	 * woken now to take the signal since we will not.
1982 	 */
1983 	for (t = tsk; (t = next_thread(t)) != tsk; )
1984 		if (!signal_pending(t) && !(t->flags & PF_EXITING))
1985 			recalc_sigpending_and_wake(t);
1986 
1987 	if (unlikely(tsk->signal->group_stop_count) &&
1988 			!--tsk->signal->group_stop_count) {
1989 		tsk->signal->flags = SIGNAL_STOP_STOPPED;
1990 		group_stop = 1;
1991 	}
1992 out:
1993 	spin_unlock_irq(&tsk->sighand->siglock);
1994 
1995 	if (unlikely(group_stop) && tracehook_notify_jctl(1, CLD_STOPPED)) {
1996 		read_lock(&tasklist_lock);
1997 		do_notify_parent_cldstop(tsk, CLD_STOPPED);
1998 		read_unlock(&tasklist_lock);
1999 	}
2000 }
2001 
2002 EXPORT_SYMBOL(recalc_sigpending);
2003 EXPORT_SYMBOL_GPL(dequeue_signal);
2004 EXPORT_SYMBOL(flush_signals);
2005 EXPORT_SYMBOL(force_sig);
2006 EXPORT_SYMBOL(send_sig);
2007 EXPORT_SYMBOL(send_sig_info);
2008 EXPORT_SYMBOL(sigprocmask);
2009 EXPORT_SYMBOL(block_all_signals);
2010 EXPORT_SYMBOL(unblock_all_signals);
2011 
2012 
2013 /*
2014  * System call entry points.
2015  */
2016 
2017 SYSCALL_DEFINE0(restart_syscall)
2018 {
2019 	struct restart_block *restart = &current_thread_info()->restart_block;
2020 	return restart->fn(restart);
2021 }
2022 
2023 long do_no_restart_syscall(struct restart_block *param)
2024 {
2025 	return -EINTR;
2026 }
2027 
2028 /*
2029  * We don't need to get the kernel lock - this is all local to this
2030  * particular thread.. (and that's good, because this is _heavily_
2031  * used by various programs)
2032  */
2033 
2034 /*
2035  * This is also useful for kernel threads that want to temporarily
2036  * (or permanently) block certain signals.
2037  *
2038  * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
2039  * interface happily blocks "unblockable" signals like SIGKILL
2040  * and friends.
2041  */
2042 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2043 {
2044 	int error;
2045 
2046 	spin_lock_irq(&current->sighand->siglock);
2047 	if (oldset)
2048 		*oldset = current->blocked;
2049 
2050 	error = 0;
2051 	switch (how) {
2052 	case SIG_BLOCK:
2053 		sigorsets(&current->blocked, &current->blocked, set);
2054 		break;
2055 	case SIG_UNBLOCK:
2056 		signandsets(&current->blocked, &current->blocked, set);
2057 		break;
2058 	case SIG_SETMASK:
2059 		current->blocked = *set;
2060 		break;
2061 	default:
2062 		error = -EINVAL;
2063 	}
2064 	recalc_sigpending();
2065 	spin_unlock_irq(&current->sighand->siglock);
2066 
2067 	return error;
2068 }
2069 
2070 SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, set,
2071 		sigset_t __user *, oset, size_t, sigsetsize)
2072 {
2073 	int error = -EINVAL;
2074 	sigset_t old_set, new_set;
2075 
2076 	/* XXX: Don't preclude handling different sized sigset_t's.  */
2077 	if (sigsetsize != sizeof(sigset_t))
2078 		goto out;
2079 
2080 	if (set) {
2081 		error = -EFAULT;
2082 		if (copy_from_user(&new_set, set, sizeof(*set)))
2083 			goto out;
2084 		sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2085 
2086 		error = sigprocmask(how, &new_set, &old_set);
2087 		if (error)
2088 			goto out;
2089 		if (oset)
2090 			goto set_old;
2091 	} else if (oset) {
2092 		spin_lock_irq(&current->sighand->siglock);
2093 		old_set = current->blocked;
2094 		spin_unlock_irq(&current->sighand->siglock);
2095 
2096 	set_old:
2097 		error = -EFAULT;
2098 		if (copy_to_user(oset, &old_set, sizeof(*oset)))
2099 			goto out;
2100 	}
2101 	error = 0;
2102 out:
2103 	return error;
2104 }
2105 
2106 long do_sigpending(void __user *set, unsigned long sigsetsize)
2107 {
2108 	long error = -EINVAL;
2109 	sigset_t pending;
2110 
2111 	if (sigsetsize > sizeof(sigset_t))
2112 		goto out;
2113 
2114 	spin_lock_irq(&current->sighand->siglock);
2115 	sigorsets(&pending, &current->pending.signal,
2116 		  &current->signal->shared_pending.signal);
2117 	spin_unlock_irq(&current->sighand->siglock);
2118 
2119 	/* Outside the lock because only this thread touches it.  */
2120 	sigandsets(&pending, &current->blocked, &pending);
2121 
2122 	error = -EFAULT;
2123 	if (!copy_to_user(set, &pending, sigsetsize))
2124 		error = 0;
2125 
2126 out:
2127 	return error;
2128 }
2129 
2130 SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, set, size_t, sigsetsize)
2131 {
2132 	return do_sigpending(set, sigsetsize);
2133 }
2134 
2135 #ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2136 
2137 int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2138 {
2139 	int err;
2140 
2141 	if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2142 		return -EFAULT;
2143 	if (from->si_code < 0)
2144 		return __copy_to_user(to, from, sizeof(siginfo_t))
2145 			? -EFAULT : 0;
2146 	/*
2147 	 * If you change siginfo_t structure, please be sure
2148 	 * this code is fixed accordingly.
2149 	 * Please remember to update the signalfd_copyinfo() function
2150 	 * inside fs/signalfd.c too, in case siginfo_t changes.
2151 	 * It should never copy any pad contained in the structure
2152 	 * to avoid security leaks, but must copy the generic
2153 	 * 3 ints plus the relevant union member.
2154 	 */
2155 	err = __put_user(from->si_signo, &to->si_signo);
2156 	err |= __put_user(from->si_errno, &to->si_errno);
2157 	err |= __put_user((short)from->si_code, &to->si_code);
2158 	switch (from->si_code & __SI_MASK) {
2159 	case __SI_KILL:
2160 		err |= __put_user(from->si_pid, &to->si_pid);
2161 		err |= __put_user(from->si_uid, &to->si_uid);
2162 		break;
2163 	case __SI_TIMER:
2164 		 err |= __put_user(from->si_tid, &to->si_tid);
2165 		 err |= __put_user(from->si_overrun, &to->si_overrun);
2166 		 err |= __put_user(from->si_ptr, &to->si_ptr);
2167 		break;
2168 	case __SI_POLL:
2169 		err |= __put_user(from->si_band, &to->si_band);
2170 		err |= __put_user(from->si_fd, &to->si_fd);
2171 		break;
2172 	case __SI_FAULT:
2173 		err |= __put_user(from->si_addr, &to->si_addr);
2174 #ifdef __ARCH_SI_TRAPNO
2175 		err |= __put_user(from->si_trapno, &to->si_trapno);
2176 #endif
2177 		break;
2178 	case __SI_CHLD:
2179 		err |= __put_user(from->si_pid, &to->si_pid);
2180 		err |= __put_user(from->si_uid, &to->si_uid);
2181 		err |= __put_user(from->si_status, &to->si_status);
2182 		err |= __put_user(from->si_utime, &to->si_utime);
2183 		err |= __put_user(from->si_stime, &to->si_stime);
2184 		break;
2185 	case __SI_RT: /* This is not generated by the kernel as of now. */
2186 	case __SI_MESGQ: /* But this is */
2187 		err |= __put_user(from->si_pid, &to->si_pid);
2188 		err |= __put_user(from->si_uid, &to->si_uid);
2189 		err |= __put_user(from->si_ptr, &to->si_ptr);
2190 		break;
2191 	default: /* this is just in case for now ... */
2192 		err |= __put_user(from->si_pid, &to->si_pid);
2193 		err |= __put_user(from->si_uid, &to->si_uid);
2194 		break;
2195 	}
2196 	return err;
2197 }
2198 
2199 #endif
2200 
2201 SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
2202 		siginfo_t __user *, uinfo, const struct timespec __user *, uts,
2203 		size_t, sigsetsize)
2204 {
2205 	int ret, sig;
2206 	sigset_t these;
2207 	struct timespec ts;
2208 	siginfo_t info;
2209 	long timeout = 0;
2210 
2211 	/* XXX: Don't preclude handling different sized sigset_t's.  */
2212 	if (sigsetsize != sizeof(sigset_t))
2213 		return -EINVAL;
2214 
2215 	if (copy_from_user(&these, uthese, sizeof(these)))
2216 		return -EFAULT;
2217 
2218 	/*
2219 	 * Invert the set of allowed signals to get those we
2220 	 * want to block.
2221 	 */
2222 	sigdelsetmask(&these, sigmask(SIGKILL)|sigmask(SIGSTOP));
2223 	signotset(&these);
2224 
2225 	if (uts) {
2226 		if (copy_from_user(&ts, uts, sizeof(ts)))
2227 			return -EFAULT;
2228 		if (ts.tv_nsec >= 1000000000L || ts.tv_nsec < 0
2229 		    || ts.tv_sec < 0)
2230 			return -EINVAL;
2231 	}
2232 
2233 	spin_lock_irq(&current->sighand->siglock);
2234 	sig = dequeue_signal(current, &these, &info);
2235 	if (!sig) {
2236 		timeout = MAX_SCHEDULE_TIMEOUT;
2237 		if (uts)
2238 			timeout = (timespec_to_jiffies(&ts)
2239 				   + (ts.tv_sec || ts.tv_nsec));
2240 
2241 		if (timeout) {
2242 			/* None ready -- temporarily unblock those we're
2243 			 * interested while we are sleeping in so that we'll
2244 			 * be awakened when they arrive.  */
2245 			current->real_blocked = current->blocked;
2246 			sigandsets(&current->blocked, &current->blocked, &these);
2247 			recalc_sigpending();
2248 			spin_unlock_irq(&current->sighand->siglock);
2249 
2250 			timeout = schedule_timeout_interruptible(timeout);
2251 
2252 			spin_lock_irq(&current->sighand->siglock);
2253 			sig = dequeue_signal(current, &these, &info);
2254 			current->blocked = current->real_blocked;
2255 			siginitset(&current->real_blocked, 0);
2256 			recalc_sigpending();
2257 		}
2258 	}
2259 	spin_unlock_irq(&current->sighand->siglock);
2260 
2261 	if (sig) {
2262 		ret = sig;
2263 		if (uinfo) {
2264 			if (copy_siginfo_to_user(uinfo, &info))
2265 				ret = -EFAULT;
2266 		}
2267 	} else {
2268 		ret = -EAGAIN;
2269 		if (timeout)
2270 			ret = -EINTR;
2271 	}
2272 
2273 	return ret;
2274 }
2275 
2276 SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
2277 {
2278 	struct siginfo info;
2279 
2280 	info.si_signo = sig;
2281 	info.si_errno = 0;
2282 	info.si_code = SI_USER;
2283 	info.si_pid = task_tgid_vnr(current);
2284 	info.si_uid = current_uid();
2285 
2286 	return kill_something_info(sig, &info, pid);
2287 }
2288 
2289 static int
2290 do_send_specific(pid_t tgid, pid_t pid, int sig, struct siginfo *info)
2291 {
2292 	struct task_struct *p;
2293 	unsigned long flags;
2294 	int error = -ESRCH;
2295 
2296 	rcu_read_lock();
2297 	p = find_task_by_vpid(pid);
2298 	if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
2299 		error = check_kill_permission(sig, info, p);
2300 		/*
2301 		 * The null signal is a permissions and process existence
2302 		 * probe.  No signal is actually delivered.
2303 		 *
2304 		 * If lock_task_sighand() fails we pretend the task dies
2305 		 * after receiving the signal. The window is tiny, and the
2306 		 * signal is private anyway.
2307 		 */
2308 		if (!error && sig && lock_task_sighand(p, &flags)) {
2309 			error = specific_send_sig_info(sig, info, p);
2310 			unlock_task_sighand(p, &flags);
2311 		}
2312 	}
2313 	rcu_read_unlock();
2314 
2315 	return error;
2316 }
2317 
2318 static int do_tkill(pid_t tgid, pid_t pid, int sig)
2319 {
2320 	struct siginfo info;
2321 
2322 	info.si_signo = sig;
2323 	info.si_errno = 0;
2324 	info.si_code = SI_TKILL;
2325 	info.si_pid = task_tgid_vnr(current);
2326 	info.si_uid = current_uid();
2327 
2328 	return do_send_specific(tgid, pid, sig, &info);
2329 }
2330 
2331 /**
2332  *  sys_tgkill - send signal to one specific thread
2333  *  @tgid: the thread group ID of the thread
2334  *  @pid: the PID of the thread
2335  *  @sig: signal to be sent
2336  *
2337  *  This syscall also checks the @tgid and returns -ESRCH even if the PID
2338  *  exists but it's not belonging to the target process anymore. This
2339  *  method solves the problem of threads exiting and PIDs getting reused.
2340  */
2341 SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
2342 {
2343 	/* This is only valid for single tasks */
2344 	if (pid <= 0 || tgid <= 0)
2345 		return -EINVAL;
2346 
2347 	return do_tkill(tgid, pid, sig);
2348 }
2349 
2350 /*
2351  *  Send a signal to only one task, even if it's a CLONE_THREAD task.
2352  */
2353 SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
2354 {
2355 	/* This is only valid for single tasks */
2356 	if (pid <= 0)
2357 		return -EINVAL;
2358 
2359 	return do_tkill(0, pid, sig);
2360 }
2361 
2362 SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
2363 		siginfo_t __user *, uinfo)
2364 {
2365 	siginfo_t info;
2366 
2367 	if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2368 		return -EFAULT;
2369 
2370 	/* Not even root can pretend to send signals from the kernel.
2371 	   Nor can they impersonate a kill(), which adds source info.  */
2372 	if (info.si_code >= 0)
2373 		return -EPERM;
2374 	info.si_signo = sig;
2375 
2376 	/* POSIX.1b doesn't mention process groups.  */
2377 	return kill_proc_info(sig, &info, pid);
2378 }
2379 
2380 long do_rt_tgsigqueueinfo(pid_t tgid, pid_t pid, int sig, siginfo_t *info)
2381 {
2382 	/* This is only valid for single tasks */
2383 	if (pid <= 0 || tgid <= 0)
2384 		return -EINVAL;
2385 
2386 	/* Not even root can pretend to send signals from the kernel.
2387 	   Nor can they impersonate a kill(), which adds source info.  */
2388 	if (info->si_code >= 0)
2389 		return -EPERM;
2390 	info->si_signo = sig;
2391 
2392 	return do_send_specific(tgid, pid, sig, info);
2393 }
2394 
2395 SYSCALL_DEFINE4(rt_tgsigqueueinfo, pid_t, tgid, pid_t, pid, int, sig,
2396 		siginfo_t __user *, uinfo)
2397 {
2398 	siginfo_t info;
2399 
2400 	if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2401 		return -EFAULT;
2402 
2403 	return do_rt_tgsigqueueinfo(tgid, pid, sig, &info);
2404 }
2405 
2406 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
2407 {
2408 	struct task_struct *t = current;
2409 	struct k_sigaction *k;
2410 	sigset_t mask;
2411 
2412 	if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
2413 		return -EINVAL;
2414 
2415 	k = &t->sighand->action[sig-1];
2416 
2417 	spin_lock_irq(&current->sighand->siglock);
2418 	if (oact)
2419 		*oact = *k;
2420 
2421 	if (act) {
2422 		sigdelsetmask(&act->sa.sa_mask,
2423 			      sigmask(SIGKILL) | sigmask(SIGSTOP));
2424 		*k = *act;
2425 		/*
2426 		 * POSIX 3.3.1.3:
2427 		 *  "Setting a signal action to SIG_IGN for a signal that is
2428 		 *   pending shall cause the pending signal to be discarded,
2429 		 *   whether or not it is blocked."
2430 		 *
2431 		 *  "Setting a signal action to SIG_DFL for a signal that is
2432 		 *   pending and whose default action is to ignore the signal
2433 		 *   (for example, SIGCHLD), shall cause the pending signal to
2434 		 *   be discarded, whether or not it is blocked"
2435 		 */
2436 		if (sig_handler_ignored(sig_handler(t, sig), sig)) {
2437 			sigemptyset(&mask);
2438 			sigaddset(&mask, sig);
2439 			rm_from_queue_full(&mask, &t->signal->shared_pending);
2440 			do {
2441 				rm_from_queue_full(&mask, &t->pending);
2442 				t = next_thread(t);
2443 			} while (t != current);
2444 		}
2445 	}
2446 
2447 	spin_unlock_irq(&current->sighand->siglock);
2448 	return 0;
2449 }
2450 
2451 int
2452 do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
2453 {
2454 	stack_t oss;
2455 	int error;
2456 
2457 	oss.ss_sp = (void __user *) current->sas_ss_sp;
2458 	oss.ss_size = current->sas_ss_size;
2459 	oss.ss_flags = sas_ss_flags(sp);
2460 
2461 	if (uss) {
2462 		void __user *ss_sp;
2463 		size_t ss_size;
2464 		int ss_flags;
2465 
2466 		error = -EFAULT;
2467 		if (!access_ok(VERIFY_READ, uss, sizeof(*uss)))
2468 			goto out;
2469 		error = __get_user(ss_sp, &uss->ss_sp) |
2470 			__get_user(ss_flags, &uss->ss_flags) |
2471 			__get_user(ss_size, &uss->ss_size);
2472 		if (error)
2473 			goto out;
2474 
2475 		error = -EPERM;
2476 		if (on_sig_stack(sp))
2477 			goto out;
2478 
2479 		error = -EINVAL;
2480 		/*
2481 		 *
2482 		 * Note - this code used to test ss_flags incorrectly
2483 		 *  	  old code may have been written using ss_flags==0
2484 		 *	  to mean ss_flags==SS_ONSTACK (as this was the only
2485 		 *	  way that worked) - this fix preserves that older
2486 		 *	  mechanism
2487 		 */
2488 		if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
2489 			goto out;
2490 
2491 		if (ss_flags == SS_DISABLE) {
2492 			ss_size = 0;
2493 			ss_sp = NULL;
2494 		} else {
2495 			error = -ENOMEM;
2496 			if (ss_size < MINSIGSTKSZ)
2497 				goto out;
2498 		}
2499 
2500 		current->sas_ss_sp = (unsigned long) ss_sp;
2501 		current->sas_ss_size = ss_size;
2502 	}
2503 
2504 	error = 0;
2505 	if (uoss) {
2506 		error = -EFAULT;
2507 		if (!access_ok(VERIFY_WRITE, uoss, sizeof(*uoss)))
2508 			goto out;
2509 		error = __put_user(oss.ss_sp, &uoss->ss_sp) |
2510 			__put_user(oss.ss_size, &uoss->ss_size) |
2511 			__put_user(oss.ss_flags, &uoss->ss_flags);
2512 	}
2513 
2514 out:
2515 	return error;
2516 }
2517 
2518 #ifdef __ARCH_WANT_SYS_SIGPENDING
2519 
2520 SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, set)
2521 {
2522 	return do_sigpending(set, sizeof(*set));
2523 }
2524 
2525 #endif
2526 
2527 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
2528 /* Some platforms have their own version with special arguments others
2529    support only sys_rt_sigprocmask.  */
2530 
2531 SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, set,
2532 		old_sigset_t __user *, oset)
2533 {
2534 	int error;
2535 	old_sigset_t old_set, new_set;
2536 
2537 	if (set) {
2538 		error = -EFAULT;
2539 		if (copy_from_user(&new_set, set, sizeof(*set)))
2540 			goto out;
2541 		new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
2542 
2543 		spin_lock_irq(&current->sighand->siglock);
2544 		old_set = current->blocked.sig[0];
2545 
2546 		error = 0;
2547 		switch (how) {
2548 		default:
2549 			error = -EINVAL;
2550 			break;
2551 		case SIG_BLOCK:
2552 			sigaddsetmask(&current->blocked, new_set);
2553 			break;
2554 		case SIG_UNBLOCK:
2555 			sigdelsetmask(&current->blocked, new_set);
2556 			break;
2557 		case SIG_SETMASK:
2558 			current->blocked.sig[0] = new_set;
2559 			break;
2560 		}
2561 
2562 		recalc_sigpending();
2563 		spin_unlock_irq(&current->sighand->siglock);
2564 		if (error)
2565 			goto out;
2566 		if (oset)
2567 			goto set_old;
2568 	} else if (oset) {
2569 		old_set = current->blocked.sig[0];
2570 	set_old:
2571 		error = -EFAULT;
2572 		if (copy_to_user(oset, &old_set, sizeof(*oset)))
2573 			goto out;
2574 	}
2575 	error = 0;
2576 out:
2577 	return error;
2578 }
2579 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
2580 
2581 #ifdef __ARCH_WANT_SYS_RT_SIGACTION
2582 SYSCALL_DEFINE4(rt_sigaction, int, sig,
2583 		const struct sigaction __user *, act,
2584 		struct sigaction __user *, oact,
2585 		size_t, sigsetsize)
2586 {
2587 	struct k_sigaction new_sa, old_sa;
2588 	int ret = -EINVAL;
2589 
2590 	/* XXX: Don't preclude handling different sized sigset_t's.  */
2591 	if (sigsetsize != sizeof(sigset_t))
2592 		goto out;
2593 
2594 	if (act) {
2595 		if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
2596 			return -EFAULT;
2597 	}
2598 
2599 	ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
2600 
2601 	if (!ret && oact) {
2602 		if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
2603 			return -EFAULT;
2604 	}
2605 out:
2606 	return ret;
2607 }
2608 #endif /* __ARCH_WANT_SYS_RT_SIGACTION */
2609 
2610 #ifdef __ARCH_WANT_SYS_SGETMASK
2611 
2612 /*
2613  * For backwards compatibility.  Functionality superseded by sigprocmask.
2614  */
2615 SYSCALL_DEFINE0(sgetmask)
2616 {
2617 	/* SMP safe */
2618 	return current->blocked.sig[0];
2619 }
2620 
2621 SYSCALL_DEFINE1(ssetmask, int, newmask)
2622 {
2623 	int old;
2624 
2625 	spin_lock_irq(&current->sighand->siglock);
2626 	old = current->blocked.sig[0];
2627 
2628 	siginitset(&current->blocked, newmask & ~(sigmask(SIGKILL)|
2629 						  sigmask(SIGSTOP)));
2630 	recalc_sigpending();
2631 	spin_unlock_irq(&current->sighand->siglock);
2632 
2633 	return old;
2634 }
2635 #endif /* __ARCH_WANT_SGETMASK */
2636 
2637 #ifdef __ARCH_WANT_SYS_SIGNAL
2638 /*
2639  * For backwards compatibility.  Functionality superseded by sigaction.
2640  */
2641 SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
2642 {
2643 	struct k_sigaction new_sa, old_sa;
2644 	int ret;
2645 
2646 	new_sa.sa.sa_handler = handler;
2647 	new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
2648 	sigemptyset(&new_sa.sa.sa_mask);
2649 
2650 	ret = do_sigaction(sig, &new_sa, &old_sa);
2651 
2652 	return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
2653 }
2654 #endif /* __ARCH_WANT_SYS_SIGNAL */
2655 
2656 #ifdef __ARCH_WANT_SYS_PAUSE
2657 
2658 SYSCALL_DEFINE0(pause)
2659 {
2660 	current->state = TASK_INTERRUPTIBLE;
2661 	schedule();
2662 	return -ERESTARTNOHAND;
2663 }
2664 
2665 #endif
2666 
2667 #ifdef __ARCH_WANT_SYS_RT_SIGSUSPEND
2668 SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
2669 {
2670 	sigset_t newset;
2671 
2672 	/* XXX: Don't preclude handling different sized sigset_t's.  */
2673 	if (sigsetsize != sizeof(sigset_t))
2674 		return -EINVAL;
2675 
2676 	if (copy_from_user(&newset, unewset, sizeof(newset)))
2677 		return -EFAULT;
2678 	sigdelsetmask(&newset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2679 
2680 	spin_lock_irq(&current->sighand->siglock);
2681 	current->saved_sigmask = current->blocked;
2682 	current->blocked = newset;
2683 	recalc_sigpending();
2684 	spin_unlock_irq(&current->sighand->siglock);
2685 
2686 	current->state = TASK_INTERRUPTIBLE;
2687 	schedule();
2688 	set_restore_sigmask();
2689 	return -ERESTARTNOHAND;
2690 }
2691 #endif /* __ARCH_WANT_SYS_RT_SIGSUSPEND */
2692 
2693 __attribute__((weak)) const char *arch_vma_name(struct vm_area_struct *vma)
2694 {
2695 	return NULL;
2696 }
2697 
2698 void __init signals_init(void)
2699 {
2700 	sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
2701 }
2702