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