xref: /openbmc/linux/kernel/signal.c (revision 96de0e252cedffad61b3cb5e05662c591898e69a)
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/capability.h>
26 #include <linux/freezer.h>
27 #include <linux/pid_namespace.h>
28 #include <linux/nsproxy.h>
29 
30 #include <asm/param.h>
31 #include <asm/uaccess.h>
32 #include <asm/unistd.h>
33 #include <asm/siginfo.h>
34 #include "audit.h"	/* audit_signal_info() */
35 
36 /*
37  * SLAB caches for signal bits.
38  */
39 
40 static struct kmem_cache *sigqueue_cachep;
41 
42 
43 static int sig_ignored(struct task_struct *t, int sig)
44 {
45 	void __user * handler;
46 
47 	/*
48 	 * Tracers always want to know about signals..
49 	 */
50 	if (t->ptrace & PT_PTRACED)
51 		return 0;
52 
53 	/*
54 	 * Blocked signals are never ignored, since the
55 	 * signal handler may change by the time it is
56 	 * unblocked.
57 	 */
58 	if (sigismember(&t->blocked, sig))
59 		return 0;
60 
61 	/* Is it explicitly or implicitly ignored? */
62 	handler = t->sighand->action[sig-1].sa.sa_handler;
63 	return   handler == SIG_IGN ||
64 		(handler == SIG_DFL && sig_kernel_ignore(sig));
65 }
66 
67 /*
68  * Re-calculate pending state from the set of locally pending
69  * signals, globally pending signals, and blocked signals.
70  */
71 static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
72 {
73 	unsigned long ready;
74 	long i;
75 
76 	switch (_NSIG_WORDS) {
77 	default:
78 		for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
79 			ready |= signal->sig[i] &~ blocked->sig[i];
80 		break;
81 
82 	case 4: ready  = signal->sig[3] &~ blocked->sig[3];
83 		ready |= signal->sig[2] &~ blocked->sig[2];
84 		ready |= signal->sig[1] &~ blocked->sig[1];
85 		ready |= signal->sig[0] &~ blocked->sig[0];
86 		break;
87 
88 	case 2: ready  = signal->sig[1] &~ blocked->sig[1];
89 		ready |= signal->sig[0] &~ blocked->sig[0];
90 		break;
91 
92 	case 1: ready  = signal->sig[0] &~ blocked->sig[0];
93 	}
94 	return ready !=	0;
95 }
96 
97 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
98 
99 static int recalc_sigpending_tsk(struct task_struct *t)
100 {
101 	if (t->signal->group_stop_count > 0 ||
102 	    PENDING(&t->pending, &t->blocked) ||
103 	    PENDING(&t->signal->shared_pending, &t->blocked)) {
104 		set_tsk_thread_flag(t, TIF_SIGPENDING);
105 		return 1;
106 	}
107 	/*
108 	 * We must never clear the flag in another thread, or in current
109 	 * when it's possible the current syscall is returning -ERESTART*.
110 	 * So we don't clear it here, and only callers who know they should do.
111 	 */
112 	return 0;
113 }
114 
115 /*
116  * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
117  * This is superfluous when called on current, the wakeup is a harmless no-op.
118  */
119 void recalc_sigpending_and_wake(struct task_struct *t)
120 {
121 	if (recalc_sigpending_tsk(t))
122 		signal_wake_up(t, 0);
123 }
124 
125 void recalc_sigpending(void)
126 {
127 	if (!recalc_sigpending_tsk(current))
128 		clear_thread_flag(TIF_SIGPENDING);
129 
130 }
131 
132 /* Given the mask, find the first available signal that should be serviced. */
133 
134 int next_signal(struct sigpending *pending, sigset_t *mask)
135 {
136 	unsigned long i, *s, *m, x;
137 	int sig = 0;
138 
139 	s = pending->signal.sig;
140 	m = mask->sig;
141 	switch (_NSIG_WORDS) {
142 	default:
143 		for (i = 0; i < _NSIG_WORDS; ++i, ++s, ++m)
144 			if ((x = *s &~ *m) != 0) {
145 				sig = ffz(~x) + i*_NSIG_BPW + 1;
146 				break;
147 			}
148 		break;
149 
150 	case 2: if ((x = s[0] &~ m[0]) != 0)
151 			sig = 1;
152 		else if ((x = s[1] &~ m[1]) != 0)
153 			sig = _NSIG_BPW + 1;
154 		else
155 			break;
156 		sig += ffz(~x);
157 		break;
158 
159 	case 1: if ((x = *s &~ *m) != 0)
160 			sig = ffz(~x) + 1;
161 		break;
162 	}
163 
164 	return sig;
165 }
166 
167 static struct sigqueue *__sigqueue_alloc(struct task_struct *t, gfp_t flags,
168 					 int override_rlimit)
169 {
170 	struct sigqueue *q = NULL;
171 	struct user_struct *user;
172 
173 	/*
174 	 * In order to avoid problems with "switch_user()", we want to make
175 	 * sure that the compiler doesn't re-load "t->user"
176 	 */
177 	user = t->user;
178 	barrier();
179 	atomic_inc(&user->sigpending);
180 	if (override_rlimit ||
181 	    atomic_read(&user->sigpending) <=
182 			t->signal->rlim[RLIMIT_SIGPENDING].rlim_cur)
183 		q = kmem_cache_alloc(sigqueue_cachep, flags);
184 	if (unlikely(q == NULL)) {
185 		atomic_dec(&user->sigpending);
186 	} else {
187 		INIT_LIST_HEAD(&q->list);
188 		q->flags = 0;
189 		q->user = get_uid(user);
190 	}
191 	return(q);
192 }
193 
194 static void __sigqueue_free(struct sigqueue *q)
195 {
196 	if (q->flags & SIGQUEUE_PREALLOC)
197 		return;
198 	atomic_dec(&q->user->sigpending);
199 	free_uid(q->user);
200 	kmem_cache_free(sigqueue_cachep, q);
201 }
202 
203 void flush_sigqueue(struct sigpending *queue)
204 {
205 	struct sigqueue *q;
206 
207 	sigemptyset(&queue->signal);
208 	while (!list_empty(&queue->list)) {
209 		q = list_entry(queue->list.next, struct sigqueue , list);
210 		list_del_init(&q->list);
211 		__sigqueue_free(q);
212 	}
213 }
214 
215 /*
216  * Flush all pending signals for a task.
217  */
218 void flush_signals(struct task_struct *t)
219 {
220 	unsigned long flags;
221 
222 	spin_lock_irqsave(&t->sighand->siglock, flags);
223 	clear_tsk_thread_flag(t,TIF_SIGPENDING);
224 	flush_sigqueue(&t->pending);
225 	flush_sigqueue(&t->signal->shared_pending);
226 	spin_unlock_irqrestore(&t->sighand->siglock, flags);
227 }
228 
229 void ignore_signals(struct task_struct *t)
230 {
231 	int i;
232 
233 	for (i = 0; i < _NSIG; ++i)
234 		t->sighand->action[i].sa.sa_handler = SIG_IGN;
235 
236 	flush_signals(t);
237 }
238 
239 /*
240  * Flush all handlers for a task.
241  */
242 
243 void
244 flush_signal_handlers(struct task_struct *t, int force_default)
245 {
246 	int i;
247 	struct k_sigaction *ka = &t->sighand->action[0];
248 	for (i = _NSIG ; i != 0 ; i--) {
249 		if (force_default || ka->sa.sa_handler != SIG_IGN)
250 			ka->sa.sa_handler = SIG_DFL;
251 		ka->sa.sa_flags = 0;
252 		sigemptyset(&ka->sa.sa_mask);
253 		ka++;
254 	}
255 }
256 
257 int unhandled_signal(struct task_struct *tsk, int sig)
258 {
259 	if (is_global_init(tsk))
260 		return 1;
261 	if (tsk->ptrace & PT_PTRACED)
262 		return 0;
263 	return (tsk->sighand->action[sig-1].sa.sa_handler == SIG_IGN) ||
264 		(tsk->sighand->action[sig-1].sa.sa_handler == SIG_DFL);
265 }
266 
267 
268 /* Notify the system that a driver wants to block all signals for this
269  * process, and wants to be notified if any signals at all were to be
270  * sent/acted upon.  If the notifier routine returns non-zero, then the
271  * signal will be acted upon after all.  If the notifier routine returns 0,
272  * then then signal will be blocked.  Only one block per process is
273  * allowed.  priv is a pointer to private data that the notifier routine
274  * can use to determine if the signal should be blocked or not.  */
275 
276 void
277 block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
278 {
279 	unsigned long flags;
280 
281 	spin_lock_irqsave(&current->sighand->siglock, flags);
282 	current->notifier_mask = mask;
283 	current->notifier_data = priv;
284 	current->notifier = notifier;
285 	spin_unlock_irqrestore(&current->sighand->siglock, flags);
286 }
287 
288 /* Notify the system that blocking has ended. */
289 
290 void
291 unblock_all_signals(void)
292 {
293 	unsigned long flags;
294 
295 	spin_lock_irqsave(&current->sighand->siglock, flags);
296 	current->notifier = NULL;
297 	current->notifier_data = NULL;
298 	recalc_sigpending();
299 	spin_unlock_irqrestore(&current->sighand->siglock, flags);
300 }
301 
302 static int collect_signal(int sig, struct sigpending *list, siginfo_t *info)
303 {
304 	struct sigqueue *q, *first = NULL;
305 	int still_pending = 0;
306 
307 	if (unlikely(!sigismember(&list->signal, sig)))
308 		return 0;
309 
310 	/*
311 	 * Collect the siginfo appropriate to this signal.  Check if
312 	 * there is another siginfo for the same signal.
313 	*/
314 	list_for_each_entry(q, &list->list, list) {
315 		if (q->info.si_signo == sig) {
316 			if (first) {
317 				still_pending = 1;
318 				break;
319 			}
320 			first = q;
321 		}
322 	}
323 	if (first) {
324 		list_del_init(&first->list);
325 		copy_siginfo(info, &first->info);
326 		__sigqueue_free(first);
327 		if (!still_pending)
328 			sigdelset(&list->signal, sig);
329 	} else {
330 
331 		/* Ok, it wasn't in the queue.  This must be
332 		   a fast-pathed signal or we must have been
333 		   out of queue space.  So zero out the info.
334 		 */
335 		sigdelset(&list->signal, sig);
336 		info->si_signo = sig;
337 		info->si_errno = 0;
338 		info->si_code = 0;
339 		info->si_pid = 0;
340 		info->si_uid = 0;
341 	}
342 	return 1;
343 }
344 
345 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
346 			siginfo_t *info)
347 {
348 	int sig = next_signal(pending, mask);
349 
350 	if (sig) {
351 		if (current->notifier) {
352 			if (sigismember(current->notifier_mask, sig)) {
353 				if (!(current->notifier)(current->notifier_data)) {
354 					clear_thread_flag(TIF_SIGPENDING);
355 					return 0;
356 				}
357 			}
358 		}
359 
360 		if (!collect_signal(sig, pending, info))
361 			sig = 0;
362 	}
363 
364 	return sig;
365 }
366 
367 /*
368  * Dequeue a signal and return the element to the caller, which is
369  * expected to free it.
370  *
371  * All callers have to hold the siglock.
372  */
373 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
374 {
375 	int signr = 0;
376 
377 	/* We only dequeue private signals from ourselves, we don't let
378 	 * signalfd steal them
379 	 */
380 	signr = __dequeue_signal(&tsk->pending, mask, info);
381 	if (!signr) {
382 		signr = __dequeue_signal(&tsk->signal->shared_pending,
383 					 mask, info);
384 		/*
385 		 * itimer signal ?
386 		 *
387 		 * itimers are process shared and we restart periodic
388 		 * itimers in the signal delivery path to prevent DoS
389 		 * attacks in the high resolution timer case. This is
390 		 * compliant with the old way of self restarting
391 		 * itimers, as the SIGALRM is a legacy signal and only
392 		 * queued once. Changing the restart behaviour to
393 		 * restart the timer in the signal dequeue path is
394 		 * reducing the timer noise on heavy loaded !highres
395 		 * systems too.
396 		 */
397 		if (unlikely(signr == SIGALRM)) {
398 			struct hrtimer *tmr = &tsk->signal->real_timer;
399 
400 			if (!hrtimer_is_queued(tmr) &&
401 			    tsk->signal->it_real_incr.tv64 != 0) {
402 				hrtimer_forward(tmr, tmr->base->get_time(),
403 						tsk->signal->it_real_incr);
404 				hrtimer_restart(tmr);
405 			}
406 		}
407 	}
408 	recalc_sigpending();
409 	if (signr && unlikely(sig_kernel_stop(signr))) {
410 		/*
411 		 * Set a marker that we have dequeued a stop signal.  Our
412 		 * caller might release the siglock and then the pending
413 		 * stop signal it is about to process is no longer in the
414 		 * pending bitmasks, but must still be cleared by a SIGCONT
415 		 * (and overruled by a SIGKILL).  So those cases clear this
416 		 * shared flag after we've set it.  Note that this flag may
417 		 * remain set after the signal we return is ignored or
418 		 * handled.  That doesn't matter because its only purpose
419 		 * is to alert stop-signal processing code when another
420 		 * processor has come along and cleared the flag.
421 		 */
422 		if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT))
423 			tsk->signal->flags |= SIGNAL_STOP_DEQUEUED;
424 	}
425 	if (signr &&
426 	     ((info->si_code & __SI_MASK) == __SI_TIMER) &&
427 	     info->si_sys_private){
428 		/*
429 		 * Release the siglock to ensure proper locking order
430 		 * of timer locks outside of siglocks.  Note, we leave
431 		 * irqs disabled here, since the posix-timers code is
432 		 * about to disable them again anyway.
433 		 */
434 		spin_unlock(&tsk->sighand->siglock);
435 		do_schedule_next_timer(info);
436 		spin_lock(&tsk->sighand->siglock);
437 	}
438 	return signr;
439 }
440 
441 /*
442  * Tell a process that it has a new active signal..
443  *
444  * NOTE! we rely on the previous spin_lock to
445  * lock interrupts for us! We can only be called with
446  * "siglock" held, and the local interrupt must
447  * have been disabled when that got acquired!
448  *
449  * No need to set need_resched since signal event passing
450  * goes through ->blocked
451  */
452 void signal_wake_up(struct task_struct *t, int resume)
453 {
454 	unsigned int mask;
455 
456 	set_tsk_thread_flag(t, TIF_SIGPENDING);
457 
458 	/*
459 	 * For SIGKILL, we want to wake it up in the stopped/traced case.
460 	 * We don't check t->state here because there is a race with it
461 	 * executing another processor and just now entering stopped state.
462 	 * By using wake_up_state, we ensure the process will wake up and
463 	 * handle its death signal.
464 	 */
465 	mask = TASK_INTERRUPTIBLE;
466 	if (resume)
467 		mask |= TASK_STOPPED | TASK_TRACED;
468 	if (!wake_up_state(t, mask))
469 		kick_process(t);
470 }
471 
472 /*
473  * Remove signals in mask from the pending set and queue.
474  * Returns 1 if any signals were found.
475  *
476  * All callers must be holding the siglock.
477  *
478  * This version takes a sigset mask and looks at all signals,
479  * not just those in the first mask word.
480  */
481 static int rm_from_queue_full(sigset_t *mask, struct sigpending *s)
482 {
483 	struct sigqueue *q, *n;
484 	sigset_t m;
485 
486 	sigandsets(&m, mask, &s->signal);
487 	if (sigisemptyset(&m))
488 		return 0;
489 
490 	signandsets(&s->signal, &s->signal, mask);
491 	list_for_each_entry_safe(q, n, &s->list, list) {
492 		if (sigismember(mask, q->info.si_signo)) {
493 			list_del_init(&q->list);
494 			__sigqueue_free(q);
495 		}
496 	}
497 	return 1;
498 }
499 /*
500  * Remove signals in mask from the pending set and queue.
501  * Returns 1 if any signals were found.
502  *
503  * All callers must be holding the siglock.
504  */
505 static int rm_from_queue(unsigned long mask, struct sigpending *s)
506 {
507 	struct sigqueue *q, *n;
508 
509 	if (!sigtestsetmask(&s->signal, mask))
510 		return 0;
511 
512 	sigdelsetmask(&s->signal, mask);
513 	list_for_each_entry_safe(q, n, &s->list, list) {
514 		if (q->info.si_signo < SIGRTMIN &&
515 		    (mask & sigmask(q->info.si_signo))) {
516 			list_del_init(&q->list);
517 			__sigqueue_free(q);
518 		}
519 	}
520 	return 1;
521 }
522 
523 /*
524  * Bad permissions for sending the signal
525  */
526 static int check_kill_permission(int sig, struct siginfo *info,
527 				 struct task_struct *t)
528 {
529 	int error = -EINVAL;
530 	if (!valid_signal(sig))
531 		return error;
532 
533 	if (info == SEND_SIG_NOINFO || (!is_si_special(info) && SI_FROMUSER(info))) {
534 		error = audit_signal_info(sig, t); /* Let audit system see the signal */
535 		if (error)
536 			return error;
537 		error = -EPERM;
538 		if (((sig != SIGCONT) ||
539 			(task_session_nr(current) != task_session_nr(t)))
540 		    && (current->euid ^ t->suid) && (current->euid ^ t->uid)
541 		    && (current->uid ^ t->suid) && (current->uid ^ t->uid)
542 		    && !capable(CAP_KILL))
543 		return error;
544 	}
545 
546 	return security_task_kill(t, info, sig, 0);
547 }
548 
549 /* forward decl */
550 static void do_notify_parent_cldstop(struct task_struct *tsk, int why);
551 
552 /*
553  * Handle magic process-wide effects of stop/continue signals.
554  * Unlike the signal actions, these happen immediately at signal-generation
555  * time regardless of blocking, ignoring, or handling.  This does the
556  * actual continuing for SIGCONT, but not the actual stopping for stop
557  * signals.  The process stop is done as a signal action for SIG_DFL.
558  */
559 static void handle_stop_signal(int sig, struct task_struct *p)
560 {
561 	struct task_struct *t;
562 
563 	if (p->signal->flags & SIGNAL_GROUP_EXIT)
564 		/*
565 		 * The process is in the middle of dying already.
566 		 */
567 		return;
568 
569 	if (sig_kernel_stop(sig)) {
570 		/*
571 		 * This is a stop signal.  Remove SIGCONT from all queues.
572 		 */
573 		rm_from_queue(sigmask(SIGCONT), &p->signal->shared_pending);
574 		t = p;
575 		do {
576 			rm_from_queue(sigmask(SIGCONT), &t->pending);
577 			t = next_thread(t);
578 		} while (t != p);
579 	} else if (sig == SIGCONT) {
580 		/*
581 		 * Remove all stop signals from all queues,
582 		 * and wake all threads.
583 		 */
584 		if (unlikely(p->signal->group_stop_count > 0)) {
585 			/*
586 			 * There was a group stop in progress.  We'll
587 			 * pretend it finished before we got here.  We are
588 			 * obliged to report it to the parent: if the
589 			 * SIGSTOP happened "after" this SIGCONT, then it
590 			 * would have cleared this pending SIGCONT.  If it
591 			 * happened "before" this SIGCONT, then the parent
592 			 * got the SIGCHLD about the stop finishing before
593 			 * the continue happened.  We do the notification
594 			 * now, and it's as if the stop had finished and
595 			 * the SIGCHLD was pending on entry to this kill.
596 			 */
597 			p->signal->group_stop_count = 0;
598 			p->signal->flags = SIGNAL_STOP_CONTINUED;
599 			spin_unlock(&p->sighand->siglock);
600 			do_notify_parent_cldstop(p, CLD_STOPPED);
601 			spin_lock(&p->sighand->siglock);
602 		}
603 		rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
604 		t = p;
605 		do {
606 			unsigned int state;
607 			rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
608 
609 			/*
610 			 * If there is a handler for SIGCONT, we must make
611 			 * sure that no thread returns to user mode before
612 			 * we post the signal, in case it was the only
613 			 * thread eligible to run the signal handler--then
614 			 * it must not do anything between resuming and
615 			 * running the handler.  With the TIF_SIGPENDING
616 			 * flag set, the thread will pause and acquire the
617 			 * siglock that we hold now and until we've queued
618 			 * the pending signal.
619 			 *
620 			 * Wake up the stopped thread _after_ setting
621 			 * TIF_SIGPENDING
622 			 */
623 			state = TASK_STOPPED;
624 			if (sig_user_defined(t, SIGCONT) && !sigismember(&t->blocked, SIGCONT)) {
625 				set_tsk_thread_flag(t, TIF_SIGPENDING);
626 				state |= TASK_INTERRUPTIBLE;
627 			}
628 			wake_up_state(t, state);
629 
630 			t = next_thread(t);
631 		} while (t != p);
632 
633 		if (p->signal->flags & SIGNAL_STOP_STOPPED) {
634 			/*
635 			 * We were in fact stopped, and are now continued.
636 			 * Notify the parent with CLD_CONTINUED.
637 			 */
638 			p->signal->flags = SIGNAL_STOP_CONTINUED;
639 			p->signal->group_exit_code = 0;
640 			spin_unlock(&p->sighand->siglock);
641 			do_notify_parent_cldstop(p, CLD_CONTINUED);
642 			spin_lock(&p->sighand->siglock);
643 		} else {
644 			/*
645 			 * We are not stopped, but there could be a stop
646 			 * signal in the middle of being processed after
647 			 * being removed from the queue.  Clear that too.
648 			 */
649 			p->signal->flags = 0;
650 		}
651 	} else if (sig == SIGKILL) {
652 		/*
653 		 * Make sure that any pending stop signal already dequeued
654 		 * is undone by the wakeup for SIGKILL.
655 		 */
656 		p->signal->flags = 0;
657 	}
658 }
659 
660 static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
661 			struct sigpending *signals)
662 {
663 	struct sigqueue * q = NULL;
664 	int ret = 0;
665 
666 	/*
667 	 * Deliver the signal to listening signalfds. This must be called
668 	 * with the sighand lock held.
669 	 */
670 	signalfd_notify(t, sig);
671 
672 	/*
673 	 * fast-pathed signals for kernel-internal things like SIGSTOP
674 	 * or SIGKILL.
675 	 */
676 	if (info == SEND_SIG_FORCED)
677 		goto out_set;
678 
679 	/* Real-time signals must be queued if sent by sigqueue, or
680 	   some other real-time mechanism.  It is implementation
681 	   defined whether kill() does so.  We attempt to do so, on
682 	   the principle of least surprise, but since kill is not
683 	   allowed to fail with EAGAIN when low on memory we just
684 	   make sure at least one signal gets delivered and don't
685 	   pass on the info struct.  */
686 
687 	q = __sigqueue_alloc(t, GFP_ATOMIC, (sig < SIGRTMIN &&
688 					     (is_si_special(info) ||
689 					      info->si_code >= 0)));
690 	if (q) {
691 		list_add_tail(&q->list, &signals->list);
692 		switch ((unsigned long) info) {
693 		case (unsigned long) SEND_SIG_NOINFO:
694 			q->info.si_signo = sig;
695 			q->info.si_errno = 0;
696 			q->info.si_code = SI_USER;
697 			q->info.si_pid = task_pid_vnr(current);
698 			q->info.si_uid = current->uid;
699 			break;
700 		case (unsigned long) SEND_SIG_PRIV:
701 			q->info.si_signo = sig;
702 			q->info.si_errno = 0;
703 			q->info.si_code = SI_KERNEL;
704 			q->info.si_pid = 0;
705 			q->info.si_uid = 0;
706 			break;
707 		default:
708 			copy_siginfo(&q->info, info);
709 			break;
710 		}
711 	} else if (!is_si_special(info)) {
712 		if (sig >= SIGRTMIN && info->si_code != SI_USER)
713 		/*
714 		 * Queue overflow, abort.  We may abort if the signal was rt
715 		 * and sent by user using something other than kill().
716 		 */
717 			return -EAGAIN;
718 	}
719 
720 out_set:
721 	sigaddset(&signals->signal, sig);
722 	return ret;
723 }
724 
725 #define LEGACY_QUEUE(sigptr, sig) \
726 	(((sig) < SIGRTMIN) && sigismember(&(sigptr)->signal, (sig)))
727 
728 int print_fatal_signals;
729 
730 static void print_fatal_signal(struct pt_regs *regs, int signr)
731 {
732 	printk("%s/%d: potentially unexpected fatal signal %d.\n",
733 		current->comm, task_pid_nr(current), signr);
734 
735 #ifdef __i386__
736 	printk("code at %08lx: ", regs->eip);
737 	{
738 		int i;
739 		for (i = 0; i < 16; i++) {
740 			unsigned char insn;
741 
742 			__get_user(insn, (unsigned char *)(regs->eip + i));
743 			printk("%02x ", insn);
744 		}
745 	}
746 #endif
747 	printk("\n");
748 	show_regs(regs);
749 }
750 
751 static int __init setup_print_fatal_signals(char *str)
752 {
753 	get_option (&str, &print_fatal_signals);
754 
755 	return 1;
756 }
757 
758 __setup("print-fatal-signals=", setup_print_fatal_signals);
759 
760 static int
761 specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
762 {
763 	int ret = 0;
764 
765 	BUG_ON(!irqs_disabled());
766 	assert_spin_locked(&t->sighand->siglock);
767 
768 	/* Short-circuit ignored signals.  */
769 	if (sig_ignored(t, sig))
770 		goto out;
771 
772 	/* Support queueing exactly one non-rt signal, so that we
773 	   can get more detailed information about the cause of
774 	   the signal. */
775 	if (LEGACY_QUEUE(&t->pending, sig))
776 		goto out;
777 
778 	ret = send_signal(sig, info, t, &t->pending);
779 	if (!ret && !sigismember(&t->blocked, sig))
780 		signal_wake_up(t, sig == SIGKILL);
781 out:
782 	return ret;
783 }
784 
785 /*
786  * Force a signal that the process can't ignore: if necessary
787  * we unblock the signal and change any SIG_IGN to SIG_DFL.
788  *
789  * Note: If we unblock the signal, we always reset it to SIG_DFL,
790  * since we do not want to have a signal handler that was blocked
791  * be invoked when user space had explicitly blocked it.
792  *
793  * We don't want to have recursive SIGSEGV's etc, for example.
794  */
795 int
796 force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
797 {
798 	unsigned long int flags;
799 	int ret, blocked, ignored;
800 	struct k_sigaction *action;
801 
802 	spin_lock_irqsave(&t->sighand->siglock, flags);
803 	action = &t->sighand->action[sig-1];
804 	ignored = action->sa.sa_handler == SIG_IGN;
805 	blocked = sigismember(&t->blocked, sig);
806 	if (blocked || ignored) {
807 		action->sa.sa_handler = SIG_DFL;
808 		if (blocked) {
809 			sigdelset(&t->blocked, sig);
810 			recalc_sigpending_and_wake(t);
811 		}
812 	}
813 	ret = specific_send_sig_info(sig, info, t);
814 	spin_unlock_irqrestore(&t->sighand->siglock, flags);
815 
816 	return ret;
817 }
818 
819 void
820 force_sig_specific(int sig, struct task_struct *t)
821 {
822 	force_sig_info(sig, SEND_SIG_FORCED, t);
823 }
824 
825 /*
826  * Test if P wants to take SIG.  After we've checked all threads with this,
827  * it's equivalent to finding no threads not blocking SIG.  Any threads not
828  * blocking SIG were ruled out because they are not running and already
829  * have pending signals.  Such threads will dequeue from the shared queue
830  * as soon as they're available, so putting the signal on the shared queue
831  * will be equivalent to sending it to one such thread.
832  */
833 static inline int wants_signal(int sig, struct task_struct *p)
834 {
835 	if (sigismember(&p->blocked, sig))
836 		return 0;
837 	if (p->flags & PF_EXITING)
838 		return 0;
839 	if (sig == SIGKILL)
840 		return 1;
841 	if (p->state & (TASK_STOPPED | TASK_TRACED))
842 		return 0;
843 	return task_curr(p) || !signal_pending(p);
844 }
845 
846 static void
847 __group_complete_signal(int sig, struct task_struct *p)
848 {
849 	struct task_struct *t;
850 
851 	/*
852 	 * Now find a thread we can wake up to take the signal off the queue.
853 	 *
854 	 * If the main thread wants the signal, it gets first crack.
855 	 * Probably the least surprising to the average bear.
856 	 */
857 	if (wants_signal(sig, p))
858 		t = p;
859 	else if (thread_group_empty(p))
860 		/*
861 		 * There is just one thread and it does not need to be woken.
862 		 * It will dequeue unblocked signals before it runs again.
863 		 */
864 		return;
865 	else {
866 		/*
867 		 * Otherwise try to find a suitable thread.
868 		 */
869 		t = p->signal->curr_target;
870 		if (t == NULL)
871 			/* restart balancing at this thread */
872 			t = p->signal->curr_target = p;
873 
874 		while (!wants_signal(sig, t)) {
875 			t = next_thread(t);
876 			if (t == p->signal->curr_target)
877 				/*
878 				 * No thread needs to be woken.
879 				 * Any eligible threads will see
880 				 * the signal in the queue soon.
881 				 */
882 				return;
883 		}
884 		p->signal->curr_target = t;
885 	}
886 
887 	/*
888 	 * Found a killable thread.  If the signal will be fatal,
889 	 * then start taking the whole group down immediately.
890 	 */
891 	if (sig_fatal(p, sig) && !(p->signal->flags & SIGNAL_GROUP_EXIT) &&
892 	    !sigismember(&t->real_blocked, sig) &&
893 	    (sig == SIGKILL || !(t->ptrace & PT_PTRACED))) {
894 		/*
895 		 * This signal will be fatal to the whole group.
896 		 */
897 		if (!sig_kernel_coredump(sig)) {
898 			/*
899 			 * Start a group exit and wake everybody up.
900 			 * This way we don't have other threads
901 			 * running and doing things after a slower
902 			 * thread has the fatal signal pending.
903 			 */
904 			p->signal->flags = SIGNAL_GROUP_EXIT;
905 			p->signal->group_exit_code = sig;
906 			p->signal->group_stop_count = 0;
907 			t = p;
908 			do {
909 				sigaddset(&t->pending.signal, SIGKILL);
910 				signal_wake_up(t, 1);
911 			} while_each_thread(p, t);
912 			return;
913 		}
914 
915 		/*
916 		 * There will be a core dump.  We make all threads other
917 		 * than the chosen one go into a group stop so that nothing
918 		 * happens until it gets scheduled, takes the signal off
919 		 * the shared queue, and does the core dump.  This is a
920 		 * little more complicated than strictly necessary, but it
921 		 * keeps the signal state that winds up in the core dump
922 		 * unchanged from the death state, e.g. which thread had
923 		 * the core-dump signal unblocked.
924 		 */
925 		rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
926 		rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
927 		p->signal->group_stop_count = 0;
928 		p->signal->group_exit_task = t;
929 		p = t;
930 		do {
931 			p->signal->group_stop_count++;
932 			signal_wake_up(t, t == p);
933 		} while_each_thread(p, t);
934 		return;
935 	}
936 
937 	/*
938 	 * The signal is already in the shared-pending queue.
939 	 * Tell the chosen thread to wake up and dequeue it.
940 	 */
941 	signal_wake_up(t, sig == SIGKILL);
942 	return;
943 }
944 
945 int
946 __group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
947 {
948 	int ret = 0;
949 
950 	assert_spin_locked(&p->sighand->siglock);
951 	handle_stop_signal(sig, p);
952 
953 	/* Short-circuit ignored signals.  */
954 	if (sig_ignored(p, sig))
955 		return ret;
956 
957 	if (LEGACY_QUEUE(&p->signal->shared_pending, sig))
958 		/* This is a non-RT signal and we already have one queued.  */
959 		return ret;
960 
961 	/*
962 	 * Put this signal on the shared-pending queue, or fail with EAGAIN.
963 	 * We always use the shared queue for process-wide signals,
964 	 * to avoid several races.
965 	 */
966 	ret = send_signal(sig, info, p, &p->signal->shared_pending);
967 	if (unlikely(ret))
968 		return ret;
969 
970 	__group_complete_signal(sig, p);
971 	return 0;
972 }
973 
974 /*
975  * Nuke all other threads in the group.
976  */
977 void zap_other_threads(struct task_struct *p)
978 {
979 	struct task_struct *t;
980 
981 	p->signal->flags = SIGNAL_GROUP_EXIT;
982 	p->signal->group_stop_count = 0;
983 
984 	for (t = next_thread(p); t != p; t = next_thread(t)) {
985 		/*
986 		 * Don't bother with already dead threads
987 		 */
988 		if (t->exit_state)
989 			continue;
990 
991 		/* SIGKILL will be handled before any pending SIGSTOP */
992 		sigaddset(&t->pending.signal, SIGKILL);
993 		signal_wake_up(t, 1);
994 	}
995 }
996 
997 /*
998  * Must be called under rcu_read_lock() or with tasklist_lock read-held.
999  */
1000 struct sighand_struct *lock_task_sighand(struct task_struct *tsk, unsigned long *flags)
1001 {
1002 	struct sighand_struct *sighand;
1003 
1004 	for (;;) {
1005 		sighand = rcu_dereference(tsk->sighand);
1006 		if (unlikely(sighand == NULL))
1007 			break;
1008 
1009 		spin_lock_irqsave(&sighand->siglock, *flags);
1010 		if (likely(sighand == tsk->sighand))
1011 			break;
1012 		spin_unlock_irqrestore(&sighand->siglock, *flags);
1013 	}
1014 
1015 	return sighand;
1016 }
1017 
1018 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1019 {
1020 	unsigned long flags;
1021 	int ret;
1022 
1023 	ret = check_kill_permission(sig, info, p);
1024 
1025 	if (!ret && sig) {
1026 		ret = -ESRCH;
1027 		if (lock_task_sighand(p, &flags)) {
1028 			ret = __group_send_sig_info(sig, info, p);
1029 			unlock_task_sighand(p, &flags);
1030 		}
1031 	}
1032 
1033 	return ret;
1034 }
1035 
1036 /*
1037  * kill_pgrp_info() sends a signal to a process group: this is what the tty
1038  * control characters do (^C, ^Z etc)
1039  */
1040 
1041 int __kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
1042 {
1043 	struct task_struct *p = NULL;
1044 	int retval, success;
1045 
1046 	success = 0;
1047 	retval = -ESRCH;
1048 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1049 		int err = group_send_sig_info(sig, info, p);
1050 		success |= !err;
1051 		retval = err;
1052 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1053 	return success ? 0 : retval;
1054 }
1055 
1056 int kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
1057 {
1058 	int retval;
1059 
1060 	read_lock(&tasklist_lock);
1061 	retval = __kill_pgrp_info(sig, info, pgrp);
1062 	read_unlock(&tasklist_lock);
1063 
1064 	return retval;
1065 }
1066 
1067 int kill_pid_info(int sig, struct siginfo *info, struct pid *pid)
1068 {
1069 	int error;
1070 	struct task_struct *p;
1071 
1072 	rcu_read_lock();
1073 	if (unlikely(sig_needs_tasklist(sig)))
1074 		read_lock(&tasklist_lock);
1075 
1076 	p = pid_task(pid, PIDTYPE_PID);
1077 	error = -ESRCH;
1078 	if (p)
1079 		error = group_send_sig_info(sig, info, p);
1080 
1081 	if (unlikely(sig_needs_tasklist(sig)))
1082 		read_unlock(&tasklist_lock);
1083 	rcu_read_unlock();
1084 	return error;
1085 }
1086 
1087 int
1088 kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1089 {
1090 	int error;
1091 	rcu_read_lock();
1092 	error = kill_pid_info(sig, info, find_vpid(pid));
1093 	rcu_read_unlock();
1094 	return error;
1095 }
1096 
1097 /* like kill_pid_info(), but doesn't use uid/euid of "current" */
1098 int kill_pid_info_as_uid(int sig, struct siginfo *info, struct pid *pid,
1099 		      uid_t uid, uid_t euid, u32 secid)
1100 {
1101 	int ret = -EINVAL;
1102 	struct task_struct *p;
1103 
1104 	if (!valid_signal(sig))
1105 		return ret;
1106 
1107 	read_lock(&tasklist_lock);
1108 	p = pid_task(pid, PIDTYPE_PID);
1109 	if (!p) {
1110 		ret = -ESRCH;
1111 		goto out_unlock;
1112 	}
1113 	if ((info == SEND_SIG_NOINFO || (!is_si_special(info) && SI_FROMUSER(info)))
1114 	    && (euid != p->suid) && (euid != p->uid)
1115 	    && (uid != p->suid) && (uid != p->uid)) {
1116 		ret = -EPERM;
1117 		goto out_unlock;
1118 	}
1119 	ret = security_task_kill(p, info, sig, secid);
1120 	if (ret)
1121 		goto out_unlock;
1122 	if (sig && p->sighand) {
1123 		unsigned long flags;
1124 		spin_lock_irqsave(&p->sighand->siglock, flags);
1125 		ret = __group_send_sig_info(sig, info, p);
1126 		spin_unlock_irqrestore(&p->sighand->siglock, flags);
1127 	}
1128 out_unlock:
1129 	read_unlock(&tasklist_lock);
1130 	return ret;
1131 }
1132 EXPORT_SYMBOL_GPL(kill_pid_info_as_uid);
1133 
1134 /*
1135  * kill_something_info() interprets pid in interesting ways just like kill(2).
1136  *
1137  * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1138  * is probably wrong.  Should make it like BSD or SYSV.
1139  */
1140 
1141 static int kill_something_info(int sig, struct siginfo *info, int pid)
1142 {
1143 	int ret;
1144 	rcu_read_lock();
1145 	if (!pid) {
1146 		ret = kill_pgrp_info(sig, info, task_pgrp(current));
1147 	} else if (pid == -1) {
1148 		int retval = 0, count = 0;
1149 		struct task_struct * p;
1150 
1151 		read_lock(&tasklist_lock);
1152 		for_each_process(p) {
1153 			if (p->pid > 1 && !same_thread_group(p, current)) {
1154 				int err = group_send_sig_info(sig, info, p);
1155 				++count;
1156 				if (err != -EPERM)
1157 					retval = err;
1158 			}
1159 		}
1160 		read_unlock(&tasklist_lock);
1161 		ret = count ? retval : -ESRCH;
1162 	} else if (pid < 0) {
1163 		ret = kill_pgrp_info(sig, info, find_vpid(-pid));
1164 	} else {
1165 		ret = kill_pid_info(sig, info, find_vpid(pid));
1166 	}
1167 	rcu_read_unlock();
1168 	return ret;
1169 }
1170 
1171 /*
1172  * These are for backward compatibility with the rest of the kernel source.
1173  */
1174 
1175 /*
1176  * These two are the most common entry points.  They send a signal
1177  * just to the specific thread.
1178  */
1179 int
1180 send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1181 {
1182 	int ret;
1183 	unsigned long flags;
1184 
1185 	/*
1186 	 * Make sure legacy kernel users don't send in bad values
1187 	 * (normal paths check this in check_kill_permission).
1188 	 */
1189 	if (!valid_signal(sig))
1190 		return -EINVAL;
1191 
1192 	/*
1193 	 * We need the tasklist lock even for the specific
1194 	 * thread case (when we don't need to follow the group
1195 	 * lists) in order to avoid races with "p->sighand"
1196 	 * going away or changing from under us.
1197 	 */
1198 	read_lock(&tasklist_lock);
1199 	spin_lock_irqsave(&p->sighand->siglock, flags);
1200 	ret = specific_send_sig_info(sig, info, p);
1201 	spin_unlock_irqrestore(&p->sighand->siglock, flags);
1202 	read_unlock(&tasklist_lock);
1203 	return ret;
1204 }
1205 
1206 #define __si_special(priv) \
1207 	((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1208 
1209 int
1210 send_sig(int sig, struct task_struct *p, int priv)
1211 {
1212 	return send_sig_info(sig, __si_special(priv), p);
1213 }
1214 
1215 /*
1216  * This is the entry point for "process-wide" signals.
1217  * They will go to an appropriate thread in the thread group.
1218  */
1219 int
1220 send_group_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1221 {
1222 	int ret;
1223 	read_lock(&tasklist_lock);
1224 	ret = group_send_sig_info(sig, info, p);
1225 	read_unlock(&tasklist_lock);
1226 	return ret;
1227 }
1228 
1229 void
1230 force_sig(int sig, struct task_struct *p)
1231 {
1232 	force_sig_info(sig, SEND_SIG_PRIV, p);
1233 }
1234 
1235 /*
1236  * When things go south during signal handling, we
1237  * will force a SIGSEGV. And if the signal that caused
1238  * the problem was already a SIGSEGV, we'll want to
1239  * make sure we don't even try to deliver the signal..
1240  */
1241 int
1242 force_sigsegv(int sig, struct task_struct *p)
1243 {
1244 	if (sig == SIGSEGV) {
1245 		unsigned long flags;
1246 		spin_lock_irqsave(&p->sighand->siglock, flags);
1247 		p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1248 		spin_unlock_irqrestore(&p->sighand->siglock, flags);
1249 	}
1250 	force_sig(SIGSEGV, p);
1251 	return 0;
1252 }
1253 
1254 int kill_pgrp(struct pid *pid, int sig, int priv)
1255 {
1256 	return kill_pgrp_info(sig, __si_special(priv), pid);
1257 }
1258 EXPORT_SYMBOL(kill_pgrp);
1259 
1260 int kill_pid(struct pid *pid, int sig, int priv)
1261 {
1262 	return kill_pid_info(sig, __si_special(priv), pid);
1263 }
1264 EXPORT_SYMBOL(kill_pid);
1265 
1266 int
1267 kill_proc(pid_t pid, int sig, int priv)
1268 {
1269 	int ret;
1270 
1271 	rcu_read_lock();
1272 	ret = kill_pid_info(sig, __si_special(priv), find_pid(pid));
1273 	rcu_read_unlock();
1274 	return ret;
1275 }
1276 
1277 /*
1278  * These functions support sending signals using preallocated sigqueue
1279  * structures.  This is needed "because realtime applications cannot
1280  * afford to lose notifications of asynchronous events, like timer
1281  * expirations or I/O completions".  In the case of Posix Timers
1282  * we allocate the sigqueue structure from the timer_create.  If this
1283  * allocation fails we are able to report the failure to the application
1284  * with an EAGAIN error.
1285  */
1286 
1287 struct sigqueue *sigqueue_alloc(void)
1288 {
1289 	struct sigqueue *q;
1290 
1291 	if ((q = __sigqueue_alloc(current, GFP_KERNEL, 0)))
1292 		q->flags |= SIGQUEUE_PREALLOC;
1293 	return(q);
1294 }
1295 
1296 void sigqueue_free(struct sigqueue *q)
1297 {
1298 	unsigned long flags;
1299 	spinlock_t *lock = &current->sighand->siglock;
1300 
1301 	BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1302 	/*
1303 	 * If the signal is still pending remove it from the
1304 	 * pending queue. We must hold ->siglock while testing
1305 	 * q->list to serialize with collect_signal().
1306 	 */
1307 	spin_lock_irqsave(lock, flags);
1308 	if (!list_empty(&q->list))
1309 		list_del_init(&q->list);
1310 	spin_unlock_irqrestore(lock, flags);
1311 
1312 	q->flags &= ~SIGQUEUE_PREALLOC;
1313 	__sigqueue_free(q);
1314 }
1315 
1316 int send_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
1317 {
1318 	unsigned long flags;
1319 	int ret = 0;
1320 
1321 	BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1322 
1323 	/*
1324 	 * The rcu based delayed sighand destroy makes it possible to
1325 	 * run this without tasklist lock held. The task struct itself
1326 	 * cannot go away as create_timer did get_task_struct().
1327 	 *
1328 	 * We return -1, when the task is marked exiting, so
1329 	 * posix_timer_event can redirect it to the group leader
1330 	 */
1331 	rcu_read_lock();
1332 
1333 	if (!likely(lock_task_sighand(p, &flags))) {
1334 		ret = -1;
1335 		goto out_err;
1336 	}
1337 
1338 	if (unlikely(!list_empty(&q->list))) {
1339 		/*
1340 		 * If an SI_TIMER entry is already queue just increment
1341 		 * the overrun count.
1342 		 */
1343 		BUG_ON(q->info.si_code != SI_TIMER);
1344 		q->info.si_overrun++;
1345 		goto out;
1346 	}
1347 	/* Short-circuit ignored signals.  */
1348 	if (sig_ignored(p, sig)) {
1349 		ret = 1;
1350 		goto out;
1351 	}
1352 	/*
1353 	 * Deliver the signal to listening signalfds. This must be called
1354 	 * with the sighand lock held.
1355 	 */
1356 	signalfd_notify(p, sig);
1357 
1358 	list_add_tail(&q->list, &p->pending.list);
1359 	sigaddset(&p->pending.signal, sig);
1360 	if (!sigismember(&p->blocked, sig))
1361 		signal_wake_up(p, sig == SIGKILL);
1362 
1363 out:
1364 	unlock_task_sighand(p, &flags);
1365 out_err:
1366 	rcu_read_unlock();
1367 
1368 	return ret;
1369 }
1370 
1371 int
1372 send_group_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
1373 {
1374 	unsigned long flags;
1375 	int ret = 0;
1376 
1377 	BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1378 
1379 	read_lock(&tasklist_lock);
1380 	/* Since it_lock is held, p->sighand cannot be NULL. */
1381 	spin_lock_irqsave(&p->sighand->siglock, flags);
1382 	handle_stop_signal(sig, p);
1383 
1384 	/* Short-circuit ignored signals.  */
1385 	if (sig_ignored(p, sig)) {
1386 		ret = 1;
1387 		goto out;
1388 	}
1389 
1390 	if (unlikely(!list_empty(&q->list))) {
1391 		/*
1392 		 * If an SI_TIMER entry is already queue just increment
1393 		 * the overrun count.  Other uses should not try to
1394 		 * send the signal multiple times.
1395 		 */
1396 		BUG_ON(q->info.si_code != SI_TIMER);
1397 		q->info.si_overrun++;
1398 		goto out;
1399 	}
1400 	/*
1401 	 * Deliver the signal to listening signalfds. This must be called
1402 	 * with the sighand lock held.
1403 	 */
1404 	signalfd_notify(p, sig);
1405 
1406 	/*
1407 	 * Put this signal on the shared-pending queue.
1408 	 * We always use the shared queue for process-wide signals,
1409 	 * to avoid several races.
1410 	 */
1411 	list_add_tail(&q->list, &p->signal->shared_pending.list);
1412 	sigaddset(&p->signal->shared_pending.signal, sig);
1413 
1414 	__group_complete_signal(sig, p);
1415 out:
1416 	spin_unlock_irqrestore(&p->sighand->siglock, flags);
1417 	read_unlock(&tasklist_lock);
1418 	return ret;
1419 }
1420 
1421 /*
1422  * Wake up any threads in the parent blocked in wait* syscalls.
1423  */
1424 static inline void __wake_up_parent(struct task_struct *p,
1425 				    struct task_struct *parent)
1426 {
1427 	wake_up_interruptible_sync(&parent->signal->wait_chldexit);
1428 }
1429 
1430 /*
1431  * Let a parent know about the death of a child.
1432  * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1433  */
1434 
1435 void do_notify_parent(struct task_struct *tsk, int sig)
1436 {
1437 	struct siginfo info;
1438 	unsigned long flags;
1439 	struct sighand_struct *psig;
1440 
1441 	BUG_ON(sig == -1);
1442 
1443  	/* do_notify_parent_cldstop should have been called instead.  */
1444  	BUG_ON(tsk->state & (TASK_STOPPED|TASK_TRACED));
1445 
1446 	BUG_ON(!tsk->ptrace &&
1447 	       (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1448 
1449 	info.si_signo = sig;
1450 	info.si_errno = 0;
1451 	/*
1452 	 * we are under tasklist_lock here so our parent is tied to
1453 	 * us and cannot exit and release its namespace.
1454 	 *
1455 	 * the only it can is to switch its nsproxy with sys_unshare,
1456 	 * bu uncharing pid namespaces is not allowed, so we'll always
1457 	 * see relevant namespace
1458 	 *
1459 	 * write_lock() currently calls preempt_disable() which is the
1460 	 * same as rcu_read_lock(), but according to Oleg, this is not
1461 	 * correct to rely on this
1462 	 */
1463 	rcu_read_lock();
1464 	info.si_pid = task_pid_nr_ns(tsk, tsk->parent->nsproxy->pid_ns);
1465 	rcu_read_unlock();
1466 
1467 	info.si_uid = tsk->uid;
1468 
1469 	/* FIXME: find out whether or not this is supposed to be c*time. */
1470 	info.si_utime = cputime_to_jiffies(cputime_add(tsk->utime,
1471 						       tsk->signal->utime));
1472 	info.si_stime = cputime_to_jiffies(cputime_add(tsk->stime,
1473 						       tsk->signal->stime));
1474 
1475 	info.si_status = tsk->exit_code & 0x7f;
1476 	if (tsk->exit_code & 0x80)
1477 		info.si_code = CLD_DUMPED;
1478 	else if (tsk->exit_code & 0x7f)
1479 		info.si_code = CLD_KILLED;
1480 	else {
1481 		info.si_code = CLD_EXITED;
1482 		info.si_status = tsk->exit_code >> 8;
1483 	}
1484 
1485 	psig = tsk->parent->sighand;
1486 	spin_lock_irqsave(&psig->siglock, flags);
1487 	if (!tsk->ptrace && sig == SIGCHLD &&
1488 	    (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1489 	     (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1490 		/*
1491 		 * We are exiting and our parent doesn't care.  POSIX.1
1492 		 * defines special semantics for setting SIGCHLD to SIG_IGN
1493 		 * or setting the SA_NOCLDWAIT flag: we should be reaped
1494 		 * automatically and not left for our parent's wait4 call.
1495 		 * Rather than having the parent do it as a magic kind of
1496 		 * signal handler, we just set this to tell do_exit that we
1497 		 * can be cleaned up without becoming a zombie.  Note that
1498 		 * we still call __wake_up_parent in this case, because a
1499 		 * blocked sys_wait4 might now return -ECHILD.
1500 		 *
1501 		 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1502 		 * is implementation-defined: we do (if you don't want
1503 		 * it, just use SIG_IGN instead).
1504 		 */
1505 		tsk->exit_signal = -1;
1506 		if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1507 			sig = 0;
1508 	}
1509 	if (valid_signal(sig) && sig > 0)
1510 		__group_send_sig_info(sig, &info, tsk->parent);
1511 	__wake_up_parent(tsk, tsk->parent);
1512 	spin_unlock_irqrestore(&psig->siglock, flags);
1513 }
1514 
1515 static void do_notify_parent_cldstop(struct task_struct *tsk, int why)
1516 {
1517 	struct siginfo info;
1518 	unsigned long flags;
1519 	struct task_struct *parent;
1520 	struct sighand_struct *sighand;
1521 
1522 	if (tsk->ptrace & PT_PTRACED)
1523 		parent = tsk->parent;
1524 	else {
1525 		tsk = tsk->group_leader;
1526 		parent = tsk->real_parent;
1527 	}
1528 
1529 	info.si_signo = SIGCHLD;
1530 	info.si_errno = 0;
1531 	/*
1532 	 * see comment in do_notify_parent() abot the following 3 lines
1533 	 */
1534 	rcu_read_lock();
1535 	info.si_pid = task_pid_nr_ns(tsk, tsk->parent->nsproxy->pid_ns);
1536 	rcu_read_unlock();
1537 
1538 	info.si_uid = tsk->uid;
1539 
1540 	/* FIXME: find out whether or not this is supposed to be c*time. */
1541 	info.si_utime = cputime_to_jiffies(tsk->utime);
1542 	info.si_stime = cputime_to_jiffies(tsk->stime);
1543 
1544  	info.si_code = why;
1545  	switch (why) {
1546  	case CLD_CONTINUED:
1547  		info.si_status = SIGCONT;
1548  		break;
1549  	case CLD_STOPPED:
1550  		info.si_status = tsk->signal->group_exit_code & 0x7f;
1551  		break;
1552  	case CLD_TRAPPED:
1553  		info.si_status = tsk->exit_code & 0x7f;
1554  		break;
1555  	default:
1556  		BUG();
1557  	}
1558 
1559 	sighand = parent->sighand;
1560 	spin_lock_irqsave(&sighand->siglock, flags);
1561 	if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1562 	    !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1563 		__group_send_sig_info(SIGCHLD, &info, parent);
1564 	/*
1565 	 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1566 	 */
1567 	__wake_up_parent(tsk, parent);
1568 	spin_unlock_irqrestore(&sighand->siglock, flags);
1569 }
1570 
1571 static inline int may_ptrace_stop(void)
1572 {
1573 	if (!likely(current->ptrace & PT_PTRACED))
1574 		return 0;
1575 
1576 	if (unlikely(current->parent == current->real_parent &&
1577 		    (current->ptrace & PT_ATTACHED)))
1578 		return 0;
1579 
1580 	/*
1581 	 * Are we in the middle of do_coredump?
1582 	 * If so and our tracer is also part of the coredump stopping
1583 	 * is a deadlock situation, and pointless because our tracer
1584 	 * is dead so don't allow us to stop.
1585 	 * If SIGKILL was already sent before the caller unlocked
1586 	 * ->siglock we must see ->core_waiters != 0. Otherwise it
1587 	 * is safe to enter schedule().
1588 	 */
1589 	if (unlikely(current->mm->core_waiters) &&
1590 	    unlikely(current->mm == current->parent->mm))
1591 		return 0;
1592 
1593 	return 1;
1594 }
1595 
1596 /*
1597  * This must be called with current->sighand->siglock held.
1598  *
1599  * This should be the path for all ptrace stops.
1600  * We always set current->last_siginfo while stopped here.
1601  * That makes it a way to test a stopped process for
1602  * being ptrace-stopped vs being job-control-stopped.
1603  *
1604  * If we actually decide not to stop at all because the tracer is gone,
1605  * we leave nostop_code in current->exit_code.
1606  */
1607 static void ptrace_stop(int exit_code, int nostop_code, siginfo_t *info)
1608 {
1609 	/*
1610 	 * If there is a group stop in progress,
1611 	 * we must participate in the bookkeeping.
1612 	 */
1613 	if (current->signal->group_stop_count > 0)
1614 		--current->signal->group_stop_count;
1615 
1616 	current->last_siginfo = info;
1617 	current->exit_code = exit_code;
1618 
1619 	/* Let the debugger run.  */
1620 	set_current_state(TASK_TRACED);
1621 	spin_unlock_irq(&current->sighand->siglock);
1622 	try_to_freeze();
1623 	read_lock(&tasklist_lock);
1624 	if (may_ptrace_stop()) {
1625 		do_notify_parent_cldstop(current, CLD_TRAPPED);
1626 		read_unlock(&tasklist_lock);
1627 		schedule();
1628 	} else {
1629 		/*
1630 		 * By the time we got the lock, our tracer went away.
1631 		 * Don't stop here.
1632 		 */
1633 		read_unlock(&tasklist_lock);
1634 		set_current_state(TASK_RUNNING);
1635 		current->exit_code = nostop_code;
1636 	}
1637 
1638 	/*
1639 	 * We are back.  Now reacquire the siglock before touching
1640 	 * last_siginfo, so that we are sure to have synchronized with
1641 	 * any signal-sending on another CPU that wants to examine it.
1642 	 */
1643 	spin_lock_irq(&current->sighand->siglock);
1644 	current->last_siginfo = NULL;
1645 
1646 	/*
1647 	 * Queued signals ignored us while we were stopped for tracing.
1648 	 * So check for any that we should take before resuming user mode.
1649 	 * This sets TIF_SIGPENDING, but never clears it.
1650 	 */
1651 	recalc_sigpending_tsk(current);
1652 }
1653 
1654 void ptrace_notify(int exit_code)
1655 {
1656 	siginfo_t info;
1657 
1658 	BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1659 
1660 	memset(&info, 0, sizeof info);
1661 	info.si_signo = SIGTRAP;
1662 	info.si_code = exit_code;
1663 	info.si_pid = task_pid_vnr(current);
1664 	info.si_uid = current->uid;
1665 
1666 	/* Let the debugger run.  */
1667 	spin_lock_irq(&current->sighand->siglock);
1668 	ptrace_stop(exit_code, 0, &info);
1669 	spin_unlock_irq(&current->sighand->siglock);
1670 }
1671 
1672 static void
1673 finish_stop(int stop_count)
1674 {
1675 	/*
1676 	 * If there are no other threads in the group, or if there is
1677 	 * a group stop in progress and we are the last to stop,
1678 	 * report to the parent.  When ptraced, every thread reports itself.
1679 	 */
1680 	if (stop_count == 0 || (current->ptrace & PT_PTRACED)) {
1681 		read_lock(&tasklist_lock);
1682 		do_notify_parent_cldstop(current, CLD_STOPPED);
1683 		read_unlock(&tasklist_lock);
1684 	}
1685 
1686 	do {
1687 		schedule();
1688 	} while (try_to_freeze());
1689 	/*
1690 	 * Now we don't run again until continued.
1691 	 */
1692 	current->exit_code = 0;
1693 }
1694 
1695 /*
1696  * This performs the stopping for SIGSTOP and other stop signals.
1697  * We have to stop all threads in the thread group.
1698  * Returns nonzero if we've actually stopped and released the siglock.
1699  * Returns zero if we didn't stop and still hold the siglock.
1700  */
1701 static int do_signal_stop(int signr)
1702 {
1703 	struct signal_struct *sig = current->signal;
1704 	int stop_count;
1705 
1706 	if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED))
1707 		return 0;
1708 
1709 	if (sig->group_stop_count > 0) {
1710 		/*
1711 		 * There is a group stop in progress.  We don't need to
1712 		 * start another one.
1713 		 */
1714 		stop_count = --sig->group_stop_count;
1715 	} else {
1716 		/*
1717 		 * There is no group stop already in progress.
1718 		 * We must initiate one now.
1719 		 */
1720 		struct task_struct *t;
1721 
1722 		sig->group_exit_code = signr;
1723 
1724 		stop_count = 0;
1725 		for (t = next_thread(current); t != current; t = next_thread(t))
1726 			/*
1727 			 * Setting state to TASK_STOPPED for a group
1728 			 * stop is always done with the siglock held,
1729 			 * so this check has no races.
1730 			 */
1731 			if (!t->exit_state &&
1732 			    !(t->state & (TASK_STOPPED|TASK_TRACED))) {
1733 				stop_count++;
1734 				signal_wake_up(t, 0);
1735 			}
1736 		sig->group_stop_count = stop_count;
1737 	}
1738 
1739 	if (stop_count == 0)
1740 		sig->flags = SIGNAL_STOP_STOPPED;
1741 	current->exit_code = sig->group_exit_code;
1742 	__set_current_state(TASK_STOPPED);
1743 
1744 	spin_unlock_irq(&current->sighand->siglock);
1745 	finish_stop(stop_count);
1746 	return 1;
1747 }
1748 
1749 /*
1750  * Do appropriate magic when group_stop_count > 0.
1751  * We return nonzero if we stopped, after releasing the siglock.
1752  * We return zero if we still hold the siglock and should look
1753  * for another signal without checking group_stop_count again.
1754  */
1755 static int handle_group_stop(void)
1756 {
1757 	int stop_count;
1758 
1759 	if (current->signal->group_exit_task == current) {
1760 		/*
1761 		 * Group stop is so we can do a core dump,
1762 		 * We are the initiating thread, so get on with it.
1763 		 */
1764 		current->signal->group_exit_task = NULL;
1765 		return 0;
1766 	}
1767 
1768 	if (current->signal->flags & SIGNAL_GROUP_EXIT)
1769 		/*
1770 		 * Group stop is so another thread can do a core dump,
1771 		 * or else we are racing against a death signal.
1772 		 * Just punt the stop so we can get the next signal.
1773 		 */
1774 		return 0;
1775 
1776 	/*
1777 	 * There is a group stop in progress.  We stop
1778 	 * without any associated signal being in our queue.
1779 	 */
1780 	stop_count = --current->signal->group_stop_count;
1781 	if (stop_count == 0)
1782 		current->signal->flags = SIGNAL_STOP_STOPPED;
1783 	current->exit_code = current->signal->group_exit_code;
1784 	set_current_state(TASK_STOPPED);
1785 	spin_unlock_irq(&current->sighand->siglock);
1786 	finish_stop(stop_count);
1787 	return 1;
1788 }
1789 
1790 int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
1791 			  struct pt_regs *regs, void *cookie)
1792 {
1793 	sigset_t *mask = &current->blocked;
1794 	int signr = 0;
1795 
1796 	try_to_freeze();
1797 
1798 relock:
1799 	spin_lock_irq(&current->sighand->siglock);
1800 	for (;;) {
1801 		struct k_sigaction *ka;
1802 
1803 		if (unlikely(current->signal->group_stop_count > 0) &&
1804 		    handle_group_stop())
1805 			goto relock;
1806 
1807 		signr = dequeue_signal(current, mask, info);
1808 
1809 		if (!signr)
1810 			break; /* will return 0 */
1811 
1812 		if ((current->ptrace & PT_PTRACED) && signr != SIGKILL) {
1813 			ptrace_signal_deliver(regs, cookie);
1814 
1815 			/* Let the debugger run.  */
1816 			ptrace_stop(signr, signr, info);
1817 
1818 			/* We're back.  Did the debugger cancel the sig?  */
1819 			signr = current->exit_code;
1820 			if (signr == 0)
1821 				continue;
1822 
1823 			current->exit_code = 0;
1824 
1825 			/* Update the siginfo structure if the signal has
1826 			   changed.  If the debugger wanted something
1827 			   specific in the siginfo structure then it should
1828 			   have updated *info via PTRACE_SETSIGINFO.  */
1829 			if (signr != info->si_signo) {
1830 				info->si_signo = signr;
1831 				info->si_errno = 0;
1832 				info->si_code = SI_USER;
1833 				info->si_pid = task_pid_vnr(current->parent);
1834 				info->si_uid = current->parent->uid;
1835 			}
1836 
1837 			/* If the (new) signal is now blocked, requeue it.  */
1838 			if (sigismember(&current->blocked, signr)) {
1839 				specific_send_sig_info(signr, info, current);
1840 				continue;
1841 			}
1842 		}
1843 
1844 		ka = &current->sighand->action[signr-1];
1845 		if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
1846 			continue;
1847 		if (ka->sa.sa_handler != SIG_DFL) {
1848 			/* Run the handler.  */
1849 			*return_ka = *ka;
1850 
1851 			if (ka->sa.sa_flags & SA_ONESHOT)
1852 				ka->sa.sa_handler = SIG_DFL;
1853 
1854 			break; /* will return non-zero "signr" value */
1855 		}
1856 
1857 		/*
1858 		 * Now we are doing the default action for this signal.
1859 		 */
1860 		if (sig_kernel_ignore(signr)) /* Default is nothing. */
1861 			continue;
1862 
1863 		/*
1864 		 * Global init gets no signals it doesn't want.
1865 		 */
1866 		if (is_global_init(current))
1867 			continue;
1868 
1869 		if (sig_kernel_stop(signr)) {
1870 			/*
1871 			 * The default action is to stop all threads in
1872 			 * the thread group.  The job control signals
1873 			 * do nothing in an orphaned pgrp, but SIGSTOP
1874 			 * always works.  Note that siglock needs to be
1875 			 * dropped during the call to is_orphaned_pgrp()
1876 			 * because of lock ordering with tasklist_lock.
1877 			 * This allows an intervening SIGCONT to be posted.
1878 			 * We need to check for that and bail out if necessary.
1879 			 */
1880 			if (signr != SIGSTOP) {
1881 				spin_unlock_irq(&current->sighand->siglock);
1882 
1883 				/* signals can be posted during this window */
1884 
1885 				if (is_current_pgrp_orphaned())
1886 					goto relock;
1887 
1888 				spin_lock_irq(&current->sighand->siglock);
1889 			}
1890 
1891 			if (likely(do_signal_stop(signr))) {
1892 				/* It released the siglock.  */
1893 				goto relock;
1894 			}
1895 
1896 			/*
1897 			 * We didn't actually stop, due to a race
1898 			 * with SIGCONT or something like that.
1899 			 */
1900 			continue;
1901 		}
1902 
1903 		spin_unlock_irq(&current->sighand->siglock);
1904 
1905 		/*
1906 		 * Anything else is fatal, maybe with a core dump.
1907 		 */
1908 		current->flags |= PF_SIGNALED;
1909 		if ((signr != SIGKILL) && print_fatal_signals)
1910 			print_fatal_signal(regs, signr);
1911 		if (sig_kernel_coredump(signr)) {
1912 			/*
1913 			 * If it was able to dump core, this kills all
1914 			 * other threads in the group and synchronizes with
1915 			 * their demise.  If we lost the race with another
1916 			 * thread getting here, it set group_exit_code
1917 			 * first and our do_group_exit call below will use
1918 			 * that value and ignore the one we pass it.
1919 			 */
1920 			do_coredump((long)signr, signr, regs);
1921 		}
1922 
1923 		/*
1924 		 * Death signals, no core dump.
1925 		 */
1926 		do_group_exit(signr);
1927 		/* NOTREACHED */
1928 	}
1929 	spin_unlock_irq(&current->sighand->siglock);
1930 	return signr;
1931 }
1932 
1933 EXPORT_SYMBOL(recalc_sigpending);
1934 EXPORT_SYMBOL_GPL(dequeue_signal);
1935 EXPORT_SYMBOL(flush_signals);
1936 EXPORT_SYMBOL(force_sig);
1937 EXPORT_SYMBOL(kill_proc);
1938 EXPORT_SYMBOL(ptrace_notify);
1939 EXPORT_SYMBOL(send_sig);
1940 EXPORT_SYMBOL(send_sig_info);
1941 EXPORT_SYMBOL(sigprocmask);
1942 EXPORT_SYMBOL(block_all_signals);
1943 EXPORT_SYMBOL(unblock_all_signals);
1944 
1945 
1946 /*
1947  * System call entry points.
1948  */
1949 
1950 asmlinkage long sys_restart_syscall(void)
1951 {
1952 	struct restart_block *restart = &current_thread_info()->restart_block;
1953 	return restart->fn(restart);
1954 }
1955 
1956 long do_no_restart_syscall(struct restart_block *param)
1957 {
1958 	return -EINTR;
1959 }
1960 
1961 /*
1962  * We don't need to get the kernel lock - this is all local to this
1963  * particular thread.. (and that's good, because this is _heavily_
1964  * used by various programs)
1965  */
1966 
1967 /*
1968  * This is also useful for kernel threads that want to temporarily
1969  * (or permanently) block certain signals.
1970  *
1971  * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
1972  * interface happily blocks "unblockable" signals like SIGKILL
1973  * and friends.
1974  */
1975 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
1976 {
1977 	int error;
1978 
1979 	spin_lock_irq(&current->sighand->siglock);
1980 	if (oldset)
1981 		*oldset = current->blocked;
1982 
1983 	error = 0;
1984 	switch (how) {
1985 	case SIG_BLOCK:
1986 		sigorsets(&current->blocked, &current->blocked, set);
1987 		break;
1988 	case SIG_UNBLOCK:
1989 		signandsets(&current->blocked, &current->blocked, set);
1990 		break;
1991 	case SIG_SETMASK:
1992 		current->blocked = *set;
1993 		break;
1994 	default:
1995 		error = -EINVAL;
1996 	}
1997 	recalc_sigpending();
1998 	spin_unlock_irq(&current->sighand->siglock);
1999 
2000 	return error;
2001 }
2002 
2003 asmlinkage long
2004 sys_rt_sigprocmask(int how, sigset_t __user *set, sigset_t __user *oset, size_t sigsetsize)
2005 {
2006 	int error = -EINVAL;
2007 	sigset_t old_set, new_set;
2008 
2009 	/* XXX: Don't preclude handling different sized sigset_t's.  */
2010 	if (sigsetsize != sizeof(sigset_t))
2011 		goto out;
2012 
2013 	if (set) {
2014 		error = -EFAULT;
2015 		if (copy_from_user(&new_set, set, sizeof(*set)))
2016 			goto out;
2017 		sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2018 
2019 		error = sigprocmask(how, &new_set, &old_set);
2020 		if (error)
2021 			goto out;
2022 		if (oset)
2023 			goto set_old;
2024 	} else if (oset) {
2025 		spin_lock_irq(&current->sighand->siglock);
2026 		old_set = current->blocked;
2027 		spin_unlock_irq(&current->sighand->siglock);
2028 
2029 	set_old:
2030 		error = -EFAULT;
2031 		if (copy_to_user(oset, &old_set, sizeof(*oset)))
2032 			goto out;
2033 	}
2034 	error = 0;
2035 out:
2036 	return error;
2037 }
2038 
2039 long do_sigpending(void __user *set, unsigned long sigsetsize)
2040 {
2041 	long error = -EINVAL;
2042 	sigset_t pending;
2043 
2044 	if (sigsetsize > sizeof(sigset_t))
2045 		goto out;
2046 
2047 	spin_lock_irq(&current->sighand->siglock);
2048 	sigorsets(&pending, &current->pending.signal,
2049 		  &current->signal->shared_pending.signal);
2050 	spin_unlock_irq(&current->sighand->siglock);
2051 
2052 	/* Outside the lock because only this thread touches it.  */
2053 	sigandsets(&pending, &current->blocked, &pending);
2054 
2055 	error = -EFAULT;
2056 	if (!copy_to_user(set, &pending, sigsetsize))
2057 		error = 0;
2058 
2059 out:
2060 	return error;
2061 }
2062 
2063 asmlinkage long
2064 sys_rt_sigpending(sigset_t __user *set, size_t sigsetsize)
2065 {
2066 	return do_sigpending(set, sigsetsize);
2067 }
2068 
2069 #ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2070 
2071 int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2072 {
2073 	int err;
2074 
2075 	if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2076 		return -EFAULT;
2077 	if (from->si_code < 0)
2078 		return __copy_to_user(to, from, sizeof(siginfo_t))
2079 			? -EFAULT : 0;
2080 	/*
2081 	 * If you change siginfo_t structure, please be sure
2082 	 * this code is fixed accordingly.
2083 	 * Please remember to update the signalfd_copyinfo() function
2084 	 * inside fs/signalfd.c too, in case siginfo_t changes.
2085 	 * It should never copy any pad contained in the structure
2086 	 * to avoid security leaks, but must copy the generic
2087 	 * 3 ints plus the relevant union member.
2088 	 */
2089 	err = __put_user(from->si_signo, &to->si_signo);
2090 	err |= __put_user(from->si_errno, &to->si_errno);
2091 	err |= __put_user((short)from->si_code, &to->si_code);
2092 	switch (from->si_code & __SI_MASK) {
2093 	case __SI_KILL:
2094 		err |= __put_user(from->si_pid, &to->si_pid);
2095 		err |= __put_user(from->si_uid, &to->si_uid);
2096 		break;
2097 	case __SI_TIMER:
2098 		 err |= __put_user(from->si_tid, &to->si_tid);
2099 		 err |= __put_user(from->si_overrun, &to->si_overrun);
2100 		 err |= __put_user(from->si_ptr, &to->si_ptr);
2101 		break;
2102 	case __SI_POLL:
2103 		err |= __put_user(from->si_band, &to->si_band);
2104 		err |= __put_user(from->si_fd, &to->si_fd);
2105 		break;
2106 	case __SI_FAULT:
2107 		err |= __put_user(from->si_addr, &to->si_addr);
2108 #ifdef __ARCH_SI_TRAPNO
2109 		err |= __put_user(from->si_trapno, &to->si_trapno);
2110 #endif
2111 		break;
2112 	case __SI_CHLD:
2113 		err |= __put_user(from->si_pid, &to->si_pid);
2114 		err |= __put_user(from->si_uid, &to->si_uid);
2115 		err |= __put_user(from->si_status, &to->si_status);
2116 		err |= __put_user(from->si_utime, &to->si_utime);
2117 		err |= __put_user(from->si_stime, &to->si_stime);
2118 		break;
2119 	case __SI_RT: /* This is not generated by the kernel as of now. */
2120 	case __SI_MESGQ: /* But this is */
2121 		err |= __put_user(from->si_pid, &to->si_pid);
2122 		err |= __put_user(from->si_uid, &to->si_uid);
2123 		err |= __put_user(from->si_ptr, &to->si_ptr);
2124 		break;
2125 	default: /* this is just in case for now ... */
2126 		err |= __put_user(from->si_pid, &to->si_pid);
2127 		err |= __put_user(from->si_uid, &to->si_uid);
2128 		break;
2129 	}
2130 	return err;
2131 }
2132 
2133 #endif
2134 
2135 asmlinkage long
2136 sys_rt_sigtimedwait(const sigset_t __user *uthese,
2137 		    siginfo_t __user *uinfo,
2138 		    const struct timespec __user *uts,
2139 		    size_t sigsetsize)
2140 {
2141 	int ret, sig;
2142 	sigset_t these;
2143 	struct timespec ts;
2144 	siginfo_t info;
2145 	long timeout = 0;
2146 
2147 	/* XXX: Don't preclude handling different sized sigset_t's.  */
2148 	if (sigsetsize != sizeof(sigset_t))
2149 		return -EINVAL;
2150 
2151 	if (copy_from_user(&these, uthese, sizeof(these)))
2152 		return -EFAULT;
2153 
2154 	/*
2155 	 * Invert the set of allowed signals to get those we
2156 	 * want to block.
2157 	 */
2158 	sigdelsetmask(&these, sigmask(SIGKILL)|sigmask(SIGSTOP));
2159 	signotset(&these);
2160 
2161 	if (uts) {
2162 		if (copy_from_user(&ts, uts, sizeof(ts)))
2163 			return -EFAULT;
2164 		if (ts.tv_nsec >= 1000000000L || ts.tv_nsec < 0
2165 		    || ts.tv_sec < 0)
2166 			return -EINVAL;
2167 	}
2168 
2169 	spin_lock_irq(&current->sighand->siglock);
2170 	sig = dequeue_signal(current, &these, &info);
2171 	if (!sig) {
2172 		timeout = MAX_SCHEDULE_TIMEOUT;
2173 		if (uts)
2174 			timeout = (timespec_to_jiffies(&ts)
2175 				   + (ts.tv_sec || ts.tv_nsec));
2176 
2177 		if (timeout) {
2178 			/* None ready -- temporarily unblock those we're
2179 			 * interested while we are sleeping in so that we'll
2180 			 * be awakened when they arrive.  */
2181 			current->real_blocked = current->blocked;
2182 			sigandsets(&current->blocked, &current->blocked, &these);
2183 			recalc_sigpending();
2184 			spin_unlock_irq(&current->sighand->siglock);
2185 
2186 			timeout = schedule_timeout_interruptible(timeout);
2187 
2188 			spin_lock_irq(&current->sighand->siglock);
2189 			sig = dequeue_signal(current, &these, &info);
2190 			current->blocked = current->real_blocked;
2191 			siginitset(&current->real_blocked, 0);
2192 			recalc_sigpending();
2193 		}
2194 	}
2195 	spin_unlock_irq(&current->sighand->siglock);
2196 
2197 	if (sig) {
2198 		ret = sig;
2199 		if (uinfo) {
2200 			if (copy_siginfo_to_user(uinfo, &info))
2201 				ret = -EFAULT;
2202 		}
2203 	} else {
2204 		ret = -EAGAIN;
2205 		if (timeout)
2206 			ret = -EINTR;
2207 	}
2208 
2209 	return ret;
2210 }
2211 
2212 asmlinkage long
2213 sys_kill(int pid, int sig)
2214 {
2215 	struct siginfo info;
2216 
2217 	info.si_signo = sig;
2218 	info.si_errno = 0;
2219 	info.si_code = SI_USER;
2220 	info.si_pid = task_tgid_vnr(current);
2221 	info.si_uid = current->uid;
2222 
2223 	return kill_something_info(sig, &info, pid);
2224 }
2225 
2226 static int do_tkill(int tgid, int pid, int sig)
2227 {
2228 	int error;
2229 	struct siginfo info;
2230 	struct task_struct *p;
2231 
2232 	error = -ESRCH;
2233 	info.si_signo = sig;
2234 	info.si_errno = 0;
2235 	info.si_code = SI_TKILL;
2236 	info.si_pid = task_tgid_vnr(current);
2237 	info.si_uid = current->uid;
2238 
2239 	read_lock(&tasklist_lock);
2240 	p = find_task_by_vpid(pid);
2241 	if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
2242 		error = check_kill_permission(sig, &info, p);
2243 		/*
2244 		 * The null signal is a permissions and process existence
2245 		 * probe.  No signal is actually delivered.
2246 		 */
2247 		if (!error && sig && p->sighand) {
2248 			spin_lock_irq(&p->sighand->siglock);
2249 			handle_stop_signal(sig, p);
2250 			error = specific_send_sig_info(sig, &info, p);
2251 			spin_unlock_irq(&p->sighand->siglock);
2252 		}
2253 	}
2254 	read_unlock(&tasklist_lock);
2255 
2256 	return error;
2257 }
2258 
2259 /**
2260  *  sys_tgkill - send signal to one specific thread
2261  *  @tgid: the thread group ID of the thread
2262  *  @pid: the PID of the thread
2263  *  @sig: signal to be sent
2264  *
2265  *  This syscall also checks the @tgid and returns -ESRCH even if the PID
2266  *  exists but it's not belonging to the target process anymore. This
2267  *  method solves the problem of threads exiting and PIDs getting reused.
2268  */
2269 asmlinkage long sys_tgkill(int tgid, int pid, int sig)
2270 {
2271 	/* This is only valid for single tasks */
2272 	if (pid <= 0 || tgid <= 0)
2273 		return -EINVAL;
2274 
2275 	return do_tkill(tgid, pid, sig);
2276 }
2277 
2278 /*
2279  *  Send a signal to only one task, even if it's a CLONE_THREAD task.
2280  */
2281 asmlinkage long
2282 sys_tkill(int pid, int sig)
2283 {
2284 	/* This is only valid for single tasks */
2285 	if (pid <= 0)
2286 		return -EINVAL;
2287 
2288 	return do_tkill(0, pid, sig);
2289 }
2290 
2291 asmlinkage long
2292 sys_rt_sigqueueinfo(int pid, int sig, siginfo_t __user *uinfo)
2293 {
2294 	siginfo_t info;
2295 
2296 	if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2297 		return -EFAULT;
2298 
2299 	/* Not even root can pretend to send signals from the kernel.
2300 	   Nor can they impersonate a kill(), which adds source info.  */
2301 	if (info.si_code >= 0)
2302 		return -EPERM;
2303 	info.si_signo = sig;
2304 
2305 	/* POSIX.1b doesn't mention process groups.  */
2306 	return kill_proc_info(sig, &info, pid);
2307 }
2308 
2309 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
2310 {
2311 	struct k_sigaction *k;
2312 	sigset_t mask;
2313 
2314 	if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
2315 		return -EINVAL;
2316 
2317 	k = &current->sighand->action[sig-1];
2318 
2319 	spin_lock_irq(&current->sighand->siglock);
2320 	if (oact)
2321 		*oact = *k;
2322 
2323 	if (act) {
2324 		sigdelsetmask(&act->sa.sa_mask,
2325 			      sigmask(SIGKILL) | sigmask(SIGSTOP));
2326 		*k = *act;
2327 		/*
2328 		 * POSIX 3.3.1.3:
2329 		 *  "Setting a signal action to SIG_IGN for a signal that is
2330 		 *   pending shall cause the pending signal to be discarded,
2331 		 *   whether or not it is blocked."
2332 		 *
2333 		 *  "Setting a signal action to SIG_DFL for a signal that is
2334 		 *   pending and whose default action is to ignore the signal
2335 		 *   (for example, SIGCHLD), shall cause the pending signal to
2336 		 *   be discarded, whether or not it is blocked"
2337 		 */
2338 		if (act->sa.sa_handler == SIG_IGN ||
2339 		   (act->sa.sa_handler == SIG_DFL && sig_kernel_ignore(sig))) {
2340 			struct task_struct *t = current;
2341 			sigemptyset(&mask);
2342 			sigaddset(&mask, sig);
2343 			rm_from_queue_full(&mask, &t->signal->shared_pending);
2344 			do {
2345 				rm_from_queue_full(&mask, &t->pending);
2346 				t = next_thread(t);
2347 			} while (t != current);
2348 		}
2349 	}
2350 
2351 	spin_unlock_irq(&current->sighand->siglock);
2352 	return 0;
2353 }
2354 
2355 int
2356 do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
2357 {
2358 	stack_t oss;
2359 	int error;
2360 
2361 	if (uoss) {
2362 		oss.ss_sp = (void __user *) current->sas_ss_sp;
2363 		oss.ss_size = current->sas_ss_size;
2364 		oss.ss_flags = sas_ss_flags(sp);
2365 	}
2366 
2367 	if (uss) {
2368 		void __user *ss_sp;
2369 		size_t ss_size;
2370 		int ss_flags;
2371 
2372 		error = -EFAULT;
2373 		if (!access_ok(VERIFY_READ, uss, sizeof(*uss))
2374 		    || __get_user(ss_sp, &uss->ss_sp)
2375 		    || __get_user(ss_flags, &uss->ss_flags)
2376 		    || __get_user(ss_size, &uss->ss_size))
2377 			goto out;
2378 
2379 		error = -EPERM;
2380 		if (on_sig_stack(sp))
2381 			goto out;
2382 
2383 		error = -EINVAL;
2384 		/*
2385 		 *
2386 		 * Note - this code used to test ss_flags incorrectly
2387 		 *  	  old code may have been written using ss_flags==0
2388 		 *	  to mean ss_flags==SS_ONSTACK (as this was the only
2389 		 *	  way that worked) - this fix preserves that older
2390 		 *	  mechanism
2391 		 */
2392 		if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
2393 			goto out;
2394 
2395 		if (ss_flags == SS_DISABLE) {
2396 			ss_size = 0;
2397 			ss_sp = NULL;
2398 		} else {
2399 			error = -ENOMEM;
2400 			if (ss_size < MINSIGSTKSZ)
2401 				goto out;
2402 		}
2403 
2404 		current->sas_ss_sp = (unsigned long) ss_sp;
2405 		current->sas_ss_size = ss_size;
2406 	}
2407 
2408 	if (uoss) {
2409 		error = -EFAULT;
2410 		if (copy_to_user(uoss, &oss, sizeof(oss)))
2411 			goto out;
2412 	}
2413 
2414 	error = 0;
2415 out:
2416 	return error;
2417 }
2418 
2419 #ifdef __ARCH_WANT_SYS_SIGPENDING
2420 
2421 asmlinkage long
2422 sys_sigpending(old_sigset_t __user *set)
2423 {
2424 	return do_sigpending(set, sizeof(*set));
2425 }
2426 
2427 #endif
2428 
2429 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
2430 /* Some platforms have their own version with special arguments others
2431    support only sys_rt_sigprocmask.  */
2432 
2433 asmlinkage long
2434 sys_sigprocmask(int how, old_sigset_t __user *set, old_sigset_t __user *oset)
2435 {
2436 	int error;
2437 	old_sigset_t old_set, new_set;
2438 
2439 	if (set) {
2440 		error = -EFAULT;
2441 		if (copy_from_user(&new_set, set, sizeof(*set)))
2442 			goto out;
2443 		new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
2444 
2445 		spin_lock_irq(&current->sighand->siglock);
2446 		old_set = current->blocked.sig[0];
2447 
2448 		error = 0;
2449 		switch (how) {
2450 		default:
2451 			error = -EINVAL;
2452 			break;
2453 		case SIG_BLOCK:
2454 			sigaddsetmask(&current->blocked, new_set);
2455 			break;
2456 		case SIG_UNBLOCK:
2457 			sigdelsetmask(&current->blocked, new_set);
2458 			break;
2459 		case SIG_SETMASK:
2460 			current->blocked.sig[0] = new_set;
2461 			break;
2462 		}
2463 
2464 		recalc_sigpending();
2465 		spin_unlock_irq(&current->sighand->siglock);
2466 		if (error)
2467 			goto out;
2468 		if (oset)
2469 			goto set_old;
2470 	} else if (oset) {
2471 		old_set = current->blocked.sig[0];
2472 	set_old:
2473 		error = -EFAULT;
2474 		if (copy_to_user(oset, &old_set, sizeof(*oset)))
2475 			goto out;
2476 	}
2477 	error = 0;
2478 out:
2479 	return error;
2480 }
2481 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
2482 
2483 #ifdef __ARCH_WANT_SYS_RT_SIGACTION
2484 asmlinkage long
2485 sys_rt_sigaction(int sig,
2486 		 const struct sigaction __user *act,
2487 		 struct sigaction __user *oact,
2488 		 size_t sigsetsize)
2489 {
2490 	struct k_sigaction new_sa, old_sa;
2491 	int ret = -EINVAL;
2492 
2493 	/* XXX: Don't preclude handling different sized sigset_t's.  */
2494 	if (sigsetsize != sizeof(sigset_t))
2495 		goto out;
2496 
2497 	if (act) {
2498 		if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
2499 			return -EFAULT;
2500 	}
2501 
2502 	ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
2503 
2504 	if (!ret && oact) {
2505 		if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
2506 			return -EFAULT;
2507 	}
2508 out:
2509 	return ret;
2510 }
2511 #endif /* __ARCH_WANT_SYS_RT_SIGACTION */
2512 
2513 #ifdef __ARCH_WANT_SYS_SGETMASK
2514 
2515 /*
2516  * For backwards compatibility.  Functionality superseded by sigprocmask.
2517  */
2518 asmlinkage long
2519 sys_sgetmask(void)
2520 {
2521 	/* SMP safe */
2522 	return current->blocked.sig[0];
2523 }
2524 
2525 asmlinkage long
2526 sys_ssetmask(int newmask)
2527 {
2528 	int old;
2529 
2530 	spin_lock_irq(&current->sighand->siglock);
2531 	old = current->blocked.sig[0];
2532 
2533 	siginitset(&current->blocked, newmask & ~(sigmask(SIGKILL)|
2534 						  sigmask(SIGSTOP)));
2535 	recalc_sigpending();
2536 	spin_unlock_irq(&current->sighand->siglock);
2537 
2538 	return old;
2539 }
2540 #endif /* __ARCH_WANT_SGETMASK */
2541 
2542 #ifdef __ARCH_WANT_SYS_SIGNAL
2543 /*
2544  * For backwards compatibility.  Functionality superseded by sigaction.
2545  */
2546 asmlinkage unsigned long
2547 sys_signal(int sig, __sighandler_t handler)
2548 {
2549 	struct k_sigaction new_sa, old_sa;
2550 	int ret;
2551 
2552 	new_sa.sa.sa_handler = handler;
2553 	new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
2554 	sigemptyset(&new_sa.sa.sa_mask);
2555 
2556 	ret = do_sigaction(sig, &new_sa, &old_sa);
2557 
2558 	return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
2559 }
2560 #endif /* __ARCH_WANT_SYS_SIGNAL */
2561 
2562 #ifdef __ARCH_WANT_SYS_PAUSE
2563 
2564 asmlinkage long
2565 sys_pause(void)
2566 {
2567 	current->state = TASK_INTERRUPTIBLE;
2568 	schedule();
2569 	return -ERESTARTNOHAND;
2570 }
2571 
2572 #endif
2573 
2574 #ifdef __ARCH_WANT_SYS_RT_SIGSUSPEND
2575 asmlinkage long sys_rt_sigsuspend(sigset_t __user *unewset, size_t sigsetsize)
2576 {
2577 	sigset_t newset;
2578 
2579 	/* XXX: Don't preclude handling different sized sigset_t's.  */
2580 	if (sigsetsize != sizeof(sigset_t))
2581 		return -EINVAL;
2582 
2583 	if (copy_from_user(&newset, unewset, sizeof(newset)))
2584 		return -EFAULT;
2585 	sigdelsetmask(&newset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2586 
2587 	spin_lock_irq(&current->sighand->siglock);
2588 	current->saved_sigmask = current->blocked;
2589 	current->blocked = newset;
2590 	recalc_sigpending();
2591 	spin_unlock_irq(&current->sighand->siglock);
2592 
2593 	current->state = TASK_INTERRUPTIBLE;
2594 	schedule();
2595 	set_thread_flag(TIF_RESTORE_SIGMASK);
2596 	return -ERESTARTNOHAND;
2597 }
2598 #endif /* __ARCH_WANT_SYS_RT_SIGSUSPEND */
2599 
2600 __attribute__((weak)) const char *arch_vma_name(struct vm_area_struct *vma)
2601 {
2602 	return NULL;
2603 }
2604 
2605 void __init signals_init(void)
2606 {
2607 	sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
2608 }
2609