xref: /linux/kernel/signal.c (revision cd354f1ae75e6466a7e31b727faede57a1f89ca5)
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/smp_lock.h>
16 #include <linux/init.h>
17 #include <linux/sched.h>
18 #include <linux/fs.h>
19 #include <linux/tty.h>
20 #include <linux/binfmts.h>
21 #include <linux/security.h>
22 #include <linux/syscalls.h>
23 #include <linux/ptrace.h>
24 #include <linux/signal.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  * In POSIX a signal is sent either to a specific thread (Linux task)
44  * or to the process as a whole (Linux thread group).  How the signal
45  * is sent determines whether it's to one thread or the whole group,
46  * which determines which signal mask(s) are involved in blocking it
47  * from being delivered until later.  When the signal is delivered,
48  * either it's caught or ignored by a user handler or it has a default
49  * effect that applies to the whole thread group (POSIX process).
50  *
51  * The possible effects an unblocked signal set to SIG_DFL can have are:
52  *   ignore	- Nothing Happens
53  *   terminate	- kill the process, i.e. all threads in the group,
54  * 		  similar to exit_group.  The group leader (only) reports
55  *		  WIFSIGNALED status to its parent.
56  *   coredump	- write a core dump file describing all threads using
57  *		  the same mm and then kill all those threads
58  *   stop 	- stop all the threads in the group, i.e. TASK_STOPPED state
59  *
60  * SIGKILL and SIGSTOP cannot be caught, blocked, or ignored.
61  * Other signals when not blocked and set to SIG_DFL behaves as follows.
62  * The job control signals also have other special effects.
63  *
64  *	+--------------------+------------------+
65  *	|  POSIX signal      |  default action  |
66  *	+--------------------+------------------+
67  *	|  SIGHUP            |  terminate	|
68  *	|  SIGINT            |	terminate	|
69  *	|  SIGQUIT           |	coredump 	|
70  *	|  SIGILL            |	coredump 	|
71  *	|  SIGTRAP           |	coredump 	|
72  *	|  SIGABRT/SIGIOT    |	coredump 	|
73  *	|  SIGBUS            |	coredump 	|
74  *	|  SIGFPE            |	coredump 	|
75  *	|  SIGKILL           |	terminate(+)	|
76  *	|  SIGUSR1           |	terminate	|
77  *	|  SIGSEGV           |	coredump 	|
78  *	|  SIGUSR2           |	terminate	|
79  *	|  SIGPIPE           |	terminate	|
80  *	|  SIGALRM           |	terminate	|
81  *	|  SIGTERM           |	terminate	|
82  *	|  SIGCHLD           |	ignore   	|
83  *	|  SIGCONT           |	ignore(*)	|
84  *	|  SIGSTOP           |	stop(*)(+)  	|
85  *	|  SIGTSTP           |	stop(*)  	|
86  *	|  SIGTTIN           |	stop(*)  	|
87  *	|  SIGTTOU           |	stop(*)  	|
88  *	|  SIGURG            |	ignore   	|
89  *	|  SIGXCPU           |	coredump 	|
90  *	|  SIGXFSZ           |	coredump 	|
91  *	|  SIGVTALRM         |	terminate	|
92  *	|  SIGPROF           |	terminate	|
93  *	|  SIGPOLL/SIGIO     |	terminate	|
94  *	|  SIGSYS/SIGUNUSED  |	coredump 	|
95  *	|  SIGSTKFLT         |	terminate	|
96  *	|  SIGWINCH          |	ignore   	|
97  *	|  SIGPWR            |	terminate	|
98  *	|  SIGRTMIN-SIGRTMAX |	terminate       |
99  *	+--------------------+------------------+
100  *	|  non-POSIX signal  |  default action  |
101  *	+--------------------+------------------+
102  *	|  SIGEMT            |  coredump	|
103  *	+--------------------+------------------+
104  *
105  * (+) For SIGKILL and SIGSTOP the action is "always", not just "default".
106  * (*) Special job control effects:
107  * When SIGCONT is sent, it resumes the process (all threads in the group)
108  * from TASK_STOPPED state and also clears any pending/queued stop signals
109  * (any of those marked with "stop(*)").  This happens regardless of blocking,
110  * catching, or ignoring SIGCONT.  When any stop signal is sent, it clears
111  * any pending/queued SIGCONT signals; this happens regardless of blocking,
112  * catching, or ignored the stop signal, though (except for SIGSTOP) the
113  * default action of stopping the process may happen later or never.
114  */
115 
116 #ifdef SIGEMT
117 #define M_SIGEMT	M(SIGEMT)
118 #else
119 #define M_SIGEMT	0
120 #endif
121 
122 #if SIGRTMIN > BITS_PER_LONG
123 #define M(sig) (1ULL << ((sig)-1))
124 #else
125 #define M(sig) (1UL << ((sig)-1))
126 #endif
127 #define T(sig, mask) (M(sig) & (mask))
128 
129 #define SIG_KERNEL_ONLY_MASK (\
130 	M(SIGKILL)   |  M(SIGSTOP)                                   )
131 
132 #define SIG_KERNEL_STOP_MASK (\
133 	M(SIGSTOP)   |  M(SIGTSTP)   |  M(SIGTTIN)   |  M(SIGTTOU)   )
134 
135 #define SIG_KERNEL_COREDUMP_MASK (\
136         M(SIGQUIT)   |  M(SIGILL)    |  M(SIGTRAP)   |  M(SIGABRT)   | \
137         M(SIGFPE)    |  M(SIGSEGV)   |  M(SIGBUS)    |  M(SIGSYS)    | \
138         M(SIGXCPU)   |  M(SIGXFSZ)   |  M_SIGEMT                     )
139 
140 #define SIG_KERNEL_IGNORE_MASK (\
141         M(SIGCONT)   |  M(SIGCHLD)   |  M(SIGWINCH)  |  M(SIGURG)    )
142 
143 #define sig_kernel_only(sig) \
144 		(((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_ONLY_MASK))
145 #define sig_kernel_coredump(sig) \
146 		(((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_COREDUMP_MASK))
147 #define sig_kernel_ignore(sig) \
148 		(((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_IGNORE_MASK))
149 #define sig_kernel_stop(sig) \
150 		(((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_STOP_MASK))
151 
152 #define sig_needs_tasklist(sig)	((sig) == SIGCONT)
153 
154 #define sig_user_defined(t, signr) \
155 	(((t)->sighand->action[(signr)-1].sa.sa_handler != SIG_DFL) &&	\
156 	 ((t)->sighand->action[(signr)-1].sa.sa_handler != SIG_IGN))
157 
158 #define sig_fatal(t, signr) \
159 	(!T(signr, SIG_KERNEL_IGNORE_MASK|SIG_KERNEL_STOP_MASK) && \
160 	 (t)->sighand->action[(signr)-1].sa.sa_handler == SIG_DFL)
161 
162 static int sig_ignored(struct task_struct *t, int sig)
163 {
164 	void __user * handler;
165 
166 	/*
167 	 * Tracers always want to know about signals..
168 	 */
169 	if (t->ptrace & PT_PTRACED)
170 		return 0;
171 
172 	/*
173 	 * Blocked signals are never ignored, since the
174 	 * signal handler may change by the time it is
175 	 * unblocked.
176 	 */
177 	if (sigismember(&t->blocked, sig))
178 		return 0;
179 
180 	/* Is it explicitly or implicitly ignored? */
181 	handler = t->sighand->action[sig-1].sa.sa_handler;
182 	return   handler == SIG_IGN ||
183 		(handler == SIG_DFL && sig_kernel_ignore(sig));
184 }
185 
186 /*
187  * Re-calculate pending state from the set of locally pending
188  * signals, globally pending signals, and blocked signals.
189  */
190 static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
191 {
192 	unsigned long ready;
193 	long i;
194 
195 	switch (_NSIG_WORDS) {
196 	default:
197 		for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
198 			ready |= signal->sig[i] &~ blocked->sig[i];
199 		break;
200 
201 	case 4: ready  = signal->sig[3] &~ blocked->sig[3];
202 		ready |= signal->sig[2] &~ blocked->sig[2];
203 		ready |= signal->sig[1] &~ blocked->sig[1];
204 		ready |= signal->sig[0] &~ blocked->sig[0];
205 		break;
206 
207 	case 2: ready  = signal->sig[1] &~ blocked->sig[1];
208 		ready |= signal->sig[0] &~ blocked->sig[0];
209 		break;
210 
211 	case 1: ready  = signal->sig[0] &~ blocked->sig[0];
212 	}
213 	return ready !=	0;
214 }
215 
216 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
217 
218 fastcall void recalc_sigpending_tsk(struct task_struct *t)
219 {
220 	if (t->signal->group_stop_count > 0 ||
221 	    (freezing(t)) ||
222 	    PENDING(&t->pending, &t->blocked) ||
223 	    PENDING(&t->signal->shared_pending, &t->blocked))
224 		set_tsk_thread_flag(t, TIF_SIGPENDING);
225 	else
226 		clear_tsk_thread_flag(t, TIF_SIGPENDING);
227 }
228 
229 void recalc_sigpending(void)
230 {
231 	recalc_sigpending_tsk(current);
232 }
233 
234 /* Given the mask, find the first available signal that should be serviced. */
235 
236 static int
237 next_signal(struct sigpending *pending, sigset_t *mask)
238 {
239 	unsigned long i, *s, *m, x;
240 	int sig = 0;
241 
242 	s = pending->signal.sig;
243 	m = mask->sig;
244 	switch (_NSIG_WORDS) {
245 	default:
246 		for (i = 0; i < _NSIG_WORDS; ++i, ++s, ++m)
247 			if ((x = *s &~ *m) != 0) {
248 				sig = ffz(~x) + i*_NSIG_BPW + 1;
249 				break;
250 			}
251 		break;
252 
253 	case 2: if ((x = s[0] &~ m[0]) != 0)
254 			sig = 1;
255 		else if ((x = s[1] &~ m[1]) != 0)
256 			sig = _NSIG_BPW + 1;
257 		else
258 			break;
259 		sig += ffz(~x);
260 		break;
261 
262 	case 1: if ((x = *s &~ *m) != 0)
263 			sig = ffz(~x) + 1;
264 		break;
265 	}
266 
267 	return sig;
268 }
269 
270 static struct sigqueue *__sigqueue_alloc(struct task_struct *t, gfp_t flags,
271 					 int override_rlimit)
272 {
273 	struct sigqueue *q = NULL;
274 	struct user_struct *user;
275 
276 	/*
277 	 * In order to avoid problems with "switch_user()", we want to make
278 	 * sure that the compiler doesn't re-load "t->user"
279 	 */
280 	user = t->user;
281 	barrier();
282 	atomic_inc(&user->sigpending);
283 	if (override_rlimit ||
284 	    atomic_read(&user->sigpending) <=
285 			t->signal->rlim[RLIMIT_SIGPENDING].rlim_cur)
286 		q = kmem_cache_alloc(sigqueue_cachep, flags);
287 	if (unlikely(q == NULL)) {
288 		atomic_dec(&user->sigpending);
289 	} else {
290 		INIT_LIST_HEAD(&q->list);
291 		q->flags = 0;
292 		q->user = get_uid(user);
293 	}
294 	return(q);
295 }
296 
297 static void __sigqueue_free(struct sigqueue *q)
298 {
299 	if (q->flags & SIGQUEUE_PREALLOC)
300 		return;
301 	atomic_dec(&q->user->sigpending);
302 	free_uid(q->user);
303 	kmem_cache_free(sigqueue_cachep, q);
304 }
305 
306 void flush_sigqueue(struct sigpending *queue)
307 {
308 	struct sigqueue *q;
309 
310 	sigemptyset(&queue->signal);
311 	while (!list_empty(&queue->list)) {
312 		q = list_entry(queue->list.next, struct sigqueue , list);
313 		list_del_init(&q->list);
314 		__sigqueue_free(q);
315 	}
316 }
317 
318 /*
319  * Flush all pending signals for a task.
320  */
321 void flush_signals(struct task_struct *t)
322 {
323 	unsigned long flags;
324 
325 	spin_lock_irqsave(&t->sighand->siglock, flags);
326 	clear_tsk_thread_flag(t,TIF_SIGPENDING);
327 	flush_sigqueue(&t->pending);
328 	flush_sigqueue(&t->signal->shared_pending);
329 	spin_unlock_irqrestore(&t->sighand->siglock, flags);
330 }
331 
332 /*
333  * Flush all handlers for a task.
334  */
335 
336 void
337 flush_signal_handlers(struct task_struct *t, int force_default)
338 {
339 	int i;
340 	struct k_sigaction *ka = &t->sighand->action[0];
341 	for (i = _NSIG ; i != 0 ; i--) {
342 		if (force_default || ka->sa.sa_handler != SIG_IGN)
343 			ka->sa.sa_handler = SIG_DFL;
344 		ka->sa.sa_flags = 0;
345 		sigemptyset(&ka->sa.sa_mask);
346 		ka++;
347 	}
348 }
349 
350 
351 /* Notify the system that a driver wants to block all signals for this
352  * process, and wants to be notified if any signals at all were to be
353  * sent/acted upon.  If the notifier routine returns non-zero, then the
354  * signal will be acted upon after all.  If the notifier routine returns 0,
355  * then then signal will be blocked.  Only one block per process is
356  * allowed.  priv is a pointer to private data that the notifier routine
357  * can use to determine if the signal should be blocked or not.  */
358 
359 void
360 block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
361 {
362 	unsigned long flags;
363 
364 	spin_lock_irqsave(&current->sighand->siglock, flags);
365 	current->notifier_mask = mask;
366 	current->notifier_data = priv;
367 	current->notifier = notifier;
368 	spin_unlock_irqrestore(&current->sighand->siglock, flags);
369 }
370 
371 /* Notify the system that blocking has ended. */
372 
373 void
374 unblock_all_signals(void)
375 {
376 	unsigned long flags;
377 
378 	spin_lock_irqsave(&current->sighand->siglock, flags);
379 	current->notifier = NULL;
380 	current->notifier_data = NULL;
381 	recalc_sigpending();
382 	spin_unlock_irqrestore(&current->sighand->siglock, flags);
383 }
384 
385 static int collect_signal(int sig, struct sigpending *list, siginfo_t *info)
386 {
387 	struct sigqueue *q, *first = NULL;
388 	int still_pending = 0;
389 
390 	if (unlikely(!sigismember(&list->signal, sig)))
391 		return 0;
392 
393 	/*
394 	 * Collect the siginfo appropriate to this signal.  Check if
395 	 * there is another siginfo for the same signal.
396 	*/
397 	list_for_each_entry(q, &list->list, list) {
398 		if (q->info.si_signo == sig) {
399 			if (first) {
400 				still_pending = 1;
401 				break;
402 			}
403 			first = q;
404 		}
405 	}
406 	if (first) {
407 		list_del_init(&first->list);
408 		copy_siginfo(info, &first->info);
409 		__sigqueue_free(first);
410 		if (!still_pending)
411 			sigdelset(&list->signal, sig);
412 	} else {
413 
414 		/* Ok, it wasn't in the queue.  This must be
415 		   a fast-pathed signal or we must have been
416 		   out of queue space.  So zero out the info.
417 		 */
418 		sigdelset(&list->signal, sig);
419 		info->si_signo = sig;
420 		info->si_errno = 0;
421 		info->si_code = 0;
422 		info->si_pid = 0;
423 		info->si_uid = 0;
424 	}
425 	return 1;
426 }
427 
428 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
429 			siginfo_t *info)
430 {
431 	int sig = next_signal(pending, mask);
432 
433 	if (sig) {
434 		if (current->notifier) {
435 			if (sigismember(current->notifier_mask, sig)) {
436 				if (!(current->notifier)(current->notifier_data)) {
437 					clear_thread_flag(TIF_SIGPENDING);
438 					return 0;
439 				}
440 			}
441 		}
442 
443 		if (!collect_signal(sig, pending, info))
444 			sig = 0;
445 	}
446 
447 	return sig;
448 }
449 
450 /*
451  * Dequeue a signal and return the element to the caller, which is
452  * expected to free it.
453  *
454  * All callers have to hold the siglock.
455  */
456 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
457 {
458 	int signr = __dequeue_signal(&tsk->pending, mask, info);
459 	if (!signr)
460 		signr = __dequeue_signal(&tsk->signal->shared_pending,
461 					 mask, info);
462 	recalc_sigpending_tsk(tsk);
463  	if (signr && unlikely(sig_kernel_stop(signr))) {
464  		/*
465  		 * Set a marker that we have dequeued a stop signal.  Our
466  		 * caller might release the siglock and then the pending
467  		 * stop signal it is about to process is no longer in the
468  		 * pending bitmasks, but must still be cleared by a SIGCONT
469  		 * (and overruled by a SIGKILL).  So those cases clear this
470  		 * shared flag after we've set it.  Note that this flag may
471  		 * remain set after the signal we return is ignored or
472  		 * handled.  That doesn't matter because its only purpose
473  		 * is to alert stop-signal processing code when another
474  		 * processor has come along and cleared the flag.
475  		 */
476  		if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT))
477  			tsk->signal->flags |= SIGNAL_STOP_DEQUEUED;
478  	}
479 	if ( signr &&
480 	     ((info->si_code & __SI_MASK) == __SI_TIMER) &&
481 	     info->si_sys_private){
482 		/*
483 		 * Release the siglock to ensure proper locking order
484 		 * of timer locks outside of siglocks.  Note, we leave
485 		 * irqs disabled here, since the posix-timers code is
486 		 * about to disable them again anyway.
487 		 */
488 		spin_unlock(&tsk->sighand->siglock);
489 		do_schedule_next_timer(info);
490 		spin_lock(&tsk->sighand->siglock);
491 	}
492 	return signr;
493 }
494 
495 /*
496  * Tell a process that it has a new active signal..
497  *
498  * NOTE! we rely on the previous spin_lock to
499  * lock interrupts for us! We can only be called with
500  * "siglock" held, and the local interrupt must
501  * have been disabled when that got acquired!
502  *
503  * No need to set need_resched since signal event passing
504  * goes through ->blocked
505  */
506 void signal_wake_up(struct task_struct *t, int resume)
507 {
508 	unsigned int mask;
509 
510 	set_tsk_thread_flag(t, TIF_SIGPENDING);
511 
512 	/*
513 	 * For SIGKILL, we want to wake it up in the stopped/traced case.
514 	 * We don't check t->state here because there is a race with it
515 	 * executing another processor and just now entering stopped state.
516 	 * By using wake_up_state, we ensure the process will wake up and
517 	 * handle its death signal.
518 	 */
519 	mask = TASK_INTERRUPTIBLE;
520 	if (resume)
521 		mask |= TASK_STOPPED | TASK_TRACED;
522 	if (!wake_up_state(t, mask))
523 		kick_process(t);
524 }
525 
526 /*
527  * Remove signals in mask from the pending set and queue.
528  * Returns 1 if any signals were found.
529  *
530  * All callers must be holding the siglock.
531  *
532  * This version takes a sigset mask and looks at all signals,
533  * not just those in the first mask word.
534  */
535 static int rm_from_queue_full(sigset_t *mask, struct sigpending *s)
536 {
537 	struct sigqueue *q, *n;
538 	sigset_t m;
539 
540 	sigandsets(&m, mask, &s->signal);
541 	if (sigisemptyset(&m))
542 		return 0;
543 
544 	signandsets(&s->signal, &s->signal, mask);
545 	list_for_each_entry_safe(q, n, &s->list, list) {
546 		if (sigismember(mask, q->info.si_signo)) {
547 			list_del_init(&q->list);
548 			__sigqueue_free(q);
549 		}
550 	}
551 	return 1;
552 }
553 /*
554  * Remove signals in mask from the pending set and queue.
555  * Returns 1 if any signals were found.
556  *
557  * All callers must be holding the siglock.
558  */
559 static int rm_from_queue(unsigned long mask, struct sigpending *s)
560 {
561 	struct sigqueue *q, *n;
562 
563 	if (!sigtestsetmask(&s->signal, mask))
564 		return 0;
565 
566 	sigdelsetmask(&s->signal, mask);
567 	list_for_each_entry_safe(q, n, &s->list, list) {
568 		if (q->info.si_signo < SIGRTMIN &&
569 		    (mask & sigmask(q->info.si_signo))) {
570 			list_del_init(&q->list);
571 			__sigqueue_free(q);
572 		}
573 	}
574 	return 1;
575 }
576 
577 /*
578  * Bad permissions for sending the signal
579  */
580 static int check_kill_permission(int sig, struct siginfo *info,
581 				 struct task_struct *t)
582 {
583 	int error = -EINVAL;
584 	if (!valid_signal(sig))
585 		return error;
586 	error = -EPERM;
587 	if ((info == SEND_SIG_NOINFO || (!is_si_special(info) && SI_FROMUSER(info)))
588 	    && ((sig != SIGCONT) ||
589 		(process_session(current) != process_session(t)))
590 	    && (current->euid ^ t->suid) && (current->euid ^ t->uid)
591 	    && (current->uid ^ t->suid) && (current->uid ^ t->uid)
592 	    && !capable(CAP_KILL))
593 		return error;
594 
595 	error = security_task_kill(t, info, sig, 0);
596 	if (!error)
597 		audit_signal_info(sig, t); /* Let audit system see the signal */
598 	return error;
599 }
600 
601 /* forward decl */
602 static void do_notify_parent_cldstop(struct task_struct *tsk, int why);
603 
604 /*
605  * Handle magic process-wide effects of stop/continue signals.
606  * Unlike the signal actions, these happen immediately at signal-generation
607  * time regardless of blocking, ignoring, or handling.  This does the
608  * actual continuing for SIGCONT, but not the actual stopping for stop
609  * signals.  The process stop is done as a signal action for SIG_DFL.
610  */
611 static void handle_stop_signal(int sig, struct task_struct *p)
612 {
613 	struct task_struct *t;
614 
615 	if (p->signal->flags & SIGNAL_GROUP_EXIT)
616 		/*
617 		 * The process is in the middle of dying already.
618 		 */
619 		return;
620 
621 	if (sig_kernel_stop(sig)) {
622 		/*
623 		 * This is a stop signal.  Remove SIGCONT from all queues.
624 		 */
625 		rm_from_queue(sigmask(SIGCONT), &p->signal->shared_pending);
626 		t = p;
627 		do {
628 			rm_from_queue(sigmask(SIGCONT), &t->pending);
629 			t = next_thread(t);
630 		} while (t != p);
631 	} else if (sig == SIGCONT) {
632 		/*
633 		 * Remove all stop signals from all queues,
634 		 * and wake all threads.
635 		 */
636 		if (unlikely(p->signal->group_stop_count > 0)) {
637 			/*
638 			 * There was a group stop in progress.  We'll
639 			 * pretend it finished before we got here.  We are
640 			 * obliged to report it to the parent: if the
641 			 * SIGSTOP happened "after" this SIGCONT, then it
642 			 * would have cleared this pending SIGCONT.  If it
643 			 * happened "before" this SIGCONT, then the parent
644 			 * got the SIGCHLD about the stop finishing before
645 			 * the continue happened.  We do the notification
646 			 * now, and it's as if the stop had finished and
647 			 * the SIGCHLD was pending on entry to this kill.
648 			 */
649 			p->signal->group_stop_count = 0;
650 			p->signal->flags = SIGNAL_STOP_CONTINUED;
651 			spin_unlock(&p->sighand->siglock);
652 			do_notify_parent_cldstop(p, CLD_STOPPED);
653 			spin_lock(&p->sighand->siglock);
654 		}
655 		rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
656 		t = p;
657 		do {
658 			unsigned int state;
659 			rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
660 
661 			/*
662 			 * If there is a handler for SIGCONT, we must make
663 			 * sure that no thread returns to user mode before
664 			 * we post the signal, in case it was the only
665 			 * thread eligible to run the signal handler--then
666 			 * it must not do anything between resuming and
667 			 * running the handler.  With the TIF_SIGPENDING
668 			 * flag set, the thread will pause and acquire the
669 			 * siglock that we hold now and until we've queued
670 			 * the pending signal.
671 			 *
672 			 * Wake up the stopped thread _after_ setting
673 			 * TIF_SIGPENDING
674 			 */
675 			state = TASK_STOPPED;
676 			if (sig_user_defined(t, SIGCONT) && !sigismember(&t->blocked, SIGCONT)) {
677 				set_tsk_thread_flag(t, TIF_SIGPENDING);
678 				state |= TASK_INTERRUPTIBLE;
679 			}
680 			wake_up_state(t, state);
681 
682 			t = next_thread(t);
683 		} while (t != p);
684 
685 		if (p->signal->flags & SIGNAL_STOP_STOPPED) {
686 			/*
687 			 * We were in fact stopped, and are now continued.
688 			 * Notify the parent with CLD_CONTINUED.
689 			 */
690 			p->signal->flags = SIGNAL_STOP_CONTINUED;
691 			p->signal->group_exit_code = 0;
692 			spin_unlock(&p->sighand->siglock);
693 			do_notify_parent_cldstop(p, CLD_CONTINUED);
694 			spin_lock(&p->sighand->siglock);
695 		} else {
696 			/*
697 			 * We are not stopped, but there could be a stop
698 			 * signal in the middle of being processed after
699 			 * being removed from the queue.  Clear that too.
700 			 */
701 			p->signal->flags = 0;
702 		}
703 	} else if (sig == SIGKILL) {
704 		/*
705 		 * Make sure that any pending stop signal already dequeued
706 		 * is undone by the wakeup for SIGKILL.
707 		 */
708 		p->signal->flags = 0;
709 	}
710 }
711 
712 static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
713 			struct sigpending *signals)
714 {
715 	struct sigqueue * q = NULL;
716 	int ret = 0;
717 
718 	/*
719 	 * fast-pathed signals for kernel-internal things like SIGSTOP
720 	 * or SIGKILL.
721 	 */
722 	if (info == SEND_SIG_FORCED)
723 		goto out_set;
724 
725 	/* Real-time signals must be queued if sent by sigqueue, or
726 	   some other real-time mechanism.  It is implementation
727 	   defined whether kill() does so.  We attempt to do so, on
728 	   the principle of least surprise, but since kill is not
729 	   allowed to fail with EAGAIN when low on memory we just
730 	   make sure at least one signal gets delivered and don't
731 	   pass on the info struct.  */
732 
733 	q = __sigqueue_alloc(t, GFP_ATOMIC, (sig < SIGRTMIN &&
734 					     (is_si_special(info) ||
735 					      info->si_code >= 0)));
736 	if (q) {
737 		list_add_tail(&q->list, &signals->list);
738 		switch ((unsigned long) info) {
739 		case (unsigned long) SEND_SIG_NOINFO:
740 			q->info.si_signo = sig;
741 			q->info.si_errno = 0;
742 			q->info.si_code = SI_USER;
743 			q->info.si_pid = current->pid;
744 			q->info.si_uid = current->uid;
745 			break;
746 		case (unsigned long) SEND_SIG_PRIV:
747 			q->info.si_signo = sig;
748 			q->info.si_errno = 0;
749 			q->info.si_code = SI_KERNEL;
750 			q->info.si_pid = 0;
751 			q->info.si_uid = 0;
752 			break;
753 		default:
754 			copy_siginfo(&q->info, info);
755 			break;
756 		}
757 	} else if (!is_si_special(info)) {
758 		if (sig >= SIGRTMIN && info->si_code != SI_USER)
759 		/*
760 		 * Queue overflow, abort.  We may abort if the signal was rt
761 		 * and sent by user using something other than kill().
762 		 */
763 			return -EAGAIN;
764 	}
765 
766 out_set:
767 	sigaddset(&signals->signal, sig);
768 	return ret;
769 }
770 
771 #define LEGACY_QUEUE(sigptr, sig) \
772 	(((sig) < SIGRTMIN) && sigismember(&(sigptr)->signal, (sig)))
773 
774 
775 static int
776 specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
777 {
778 	int ret = 0;
779 
780 	BUG_ON(!irqs_disabled());
781 	assert_spin_locked(&t->sighand->siglock);
782 
783 	/* Short-circuit ignored signals.  */
784 	if (sig_ignored(t, sig))
785 		goto out;
786 
787 	/* Support queueing exactly one non-rt signal, so that we
788 	   can get more detailed information about the cause of
789 	   the signal. */
790 	if (LEGACY_QUEUE(&t->pending, sig))
791 		goto out;
792 
793 	ret = send_signal(sig, info, t, &t->pending);
794 	if (!ret && !sigismember(&t->blocked, sig))
795 		signal_wake_up(t, sig == SIGKILL);
796 out:
797 	return ret;
798 }
799 
800 /*
801  * Force a signal that the process can't ignore: if necessary
802  * we unblock the signal and change any SIG_IGN to SIG_DFL.
803  *
804  * Note: If we unblock the signal, we always reset it to SIG_DFL,
805  * since we do not want to have a signal handler that was blocked
806  * be invoked when user space had explicitly blocked it.
807  *
808  * We don't want to have recursive SIGSEGV's etc, for example.
809  */
810 int
811 force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
812 {
813 	unsigned long int flags;
814 	int ret, blocked, ignored;
815 	struct k_sigaction *action;
816 
817 	spin_lock_irqsave(&t->sighand->siglock, flags);
818 	action = &t->sighand->action[sig-1];
819 	ignored = action->sa.sa_handler == SIG_IGN;
820 	blocked = sigismember(&t->blocked, sig);
821 	if (blocked || ignored) {
822 		action->sa.sa_handler = SIG_DFL;
823 		if (blocked) {
824 			sigdelset(&t->blocked, sig);
825 			recalc_sigpending_tsk(t);
826 		}
827 	}
828 	ret = specific_send_sig_info(sig, info, t);
829 	spin_unlock_irqrestore(&t->sighand->siglock, flags);
830 
831 	return ret;
832 }
833 
834 void
835 force_sig_specific(int sig, struct task_struct *t)
836 {
837 	force_sig_info(sig, SEND_SIG_FORCED, t);
838 }
839 
840 /*
841  * Test if P wants to take SIG.  After we've checked all threads with this,
842  * it's equivalent to finding no threads not blocking SIG.  Any threads not
843  * blocking SIG were ruled out because they are not running and already
844  * have pending signals.  Such threads will dequeue from the shared queue
845  * as soon as they're available, so putting the signal on the shared queue
846  * will be equivalent to sending it to one such thread.
847  */
848 static inline int wants_signal(int sig, struct task_struct *p)
849 {
850 	if (sigismember(&p->blocked, sig))
851 		return 0;
852 	if (p->flags & PF_EXITING)
853 		return 0;
854 	if (sig == SIGKILL)
855 		return 1;
856 	if (p->state & (TASK_STOPPED | TASK_TRACED))
857 		return 0;
858 	return task_curr(p) || !signal_pending(p);
859 }
860 
861 static void
862 __group_complete_signal(int sig, struct task_struct *p)
863 {
864 	struct task_struct *t;
865 
866 	/*
867 	 * Now find a thread we can wake up to take the signal off the queue.
868 	 *
869 	 * If the main thread wants the signal, it gets first crack.
870 	 * Probably the least surprising to the average bear.
871 	 */
872 	if (wants_signal(sig, p))
873 		t = p;
874 	else if (thread_group_empty(p))
875 		/*
876 		 * There is just one thread and it does not need to be woken.
877 		 * It will dequeue unblocked signals before it runs again.
878 		 */
879 		return;
880 	else {
881 		/*
882 		 * Otherwise try to find a suitable thread.
883 		 */
884 		t = p->signal->curr_target;
885 		if (t == NULL)
886 			/* restart balancing at this thread */
887 			t = p->signal->curr_target = p;
888 
889 		while (!wants_signal(sig, t)) {
890 			t = next_thread(t);
891 			if (t == p->signal->curr_target)
892 				/*
893 				 * No thread needs to be woken.
894 				 * Any eligible threads will see
895 				 * the signal in the queue soon.
896 				 */
897 				return;
898 		}
899 		p->signal->curr_target = t;
900 	}
901 
902 	/*
903 	 * Found a killable thread.  If the signal will be fatal,
904 	 * then start taking the whole group down immediately.
905 	 */
906 	if (sig_fatal(p, sig) && !(p->signal->flags & SIGNAL_GROUP_EXIT) &&
907 	    !sigismember(&t->real_blocked, sig) &&
908 	    (sig == SIGKILL || !(t->ptrace & PT_PTRACED))) {
909 		/*
910 		 * This signal will be fatal to the whole group.
911 		 */
912 		if (!sig_kernel_coredump(sig)) {
913 			/*
914 			 * Start a group exit and wake everybody up.
915 			 * This way we don't have other threads
916 			 * running and doing things after a slower
917 			 * thread has the fatal signal pending.
918 			 */
919 			p->signal->flags = SIGNAL_GROUP_EXIT;
920 			p->signal->group_exit_code = sig;
921 			p->signal->group_stop_count = 0;
922 			t = p;
923 			do {
924 				sigaddset(&t->pending.signal, SIGKILL);
925 				signal_wake_up(t, 1);
926 				t = next_thread(t);
927 			} while (t != p);
928 			return;
929 		}
930 
931 		/*
932 		 * There will be a core dump.  We make all threads other
933 		 * than the chosen one go into a group stop so that nothing
934 		 * happens until it gets scheduled, takes the signal off
935 		 * the shared queue, and does the core dump.  This is a
936 		 * little more complicated than strictly necessary, but it
937 		 * keeps the signal state that winds up in the core dump
938 		 * unchanged from the death state, e.g. which thread had
939 		 * the core-dump signal unblocked.
940 		 */
941 		rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
942 		rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
943 		p->signal->group_stop_count = 0;
944 		p->signal->group_exit_task = t;
945 		t = p;
946 		do {
947 			p->signal->group_stop_count++;
948 			signal_wake_up(t, 0);
949 			t = next_thread(t);
950 		} while (t != p);
951 		wake_up_process(p->signal->group_exit_task);
952 		return;
953 	}
954 
955 	/*
956 	 * The signal is already in the shared-pending queue.
957 	 * Tell the chosen thread to wake up and dequeue it.
958 	 */
959 	signal_wake_up(t, sig == SIGKILL);
960 	return;
961 }
962 
963 int
964 __group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
965 {
966 	int ret = 0;
967 
968 	assert_spin_locked(&p->sighand->siglock);
969 	handle_stop_signal(sig, p);
970 
971 	/* Short-circuit ignored signals.  */
972 	if (sig_ignored(p, sig))
973 		return ret;
974 
975 	if (LEGACY_QUEUE(&p->signal->shared_pending, sig))
976 		/* This is a non-RT signal and we already have one queued.  */
977 		return ret;
978 
979 	/*
980 	 * Put this signal on the shared-pending queue, or fail with EAGAIN.
981 	 * We always use the shared queue for process-wide signals,
982 	 * to avoid several races.
983 	 */
984 	ret = send_signal(sig, info, p, &p->signal->shared_pending);
985 	if (unlikely(ret))
986 		return ret;
987 
988 	__group_complete_signal(sig, p);
989 	return 0;
990 }
991 
992 /*
993  * Nuke all other threads in the group.
994  */
995 void zap_other_threads(struct task_struct *p)
996 {
997 	struct task_struct *t;
998 
999 	p->signal->flags = SIGNAL_GROUP_EXIT;
1000 	p->signal->group_stop_count = 0;
1001 
1002 	if (thread_group_empty(p))
1003 		return;
1004 
1005 	for (t = next_thread(p); t != p; t = next_thread(t)) {
1006 		/*
1007 		 * Don't bother with already dead threads
1008 		 */
1009 		if (t->exit_state)
1010 			continue;
1011 
1012 		/*
1013 		 * We don't want to notify the parent, since we are
1014 		 * killed as part of a thread group due to another
1015 		 * thread doing an execve() or similar. So set the
1016 		 * exit signal to -1 to allow immediate reaping of
1017 		 * the process.  But don't detach the thread group
1018 		 * leader.
1019 		 */
1020 		if (t != p->group_leader)
1021 			t->exit_signal = -1;
1022 
1023 		/* SIGKILL will be handled before any pending SIGSTOP */
1024 		sigaddset(&t->pending.signal, SIGKILL);
1025 		signal_wake_up(t, 1);
1026 	}
1027 }
1028 
1029 /*
1030  * Must be called under rcu_read_lock() or with tasklist_lock read-held.
1031  */
1032 struct sighand_struct *lock_task_sighand(struct task_struct *tsk, unsigned long *flags)
1033 {
1034 	struct sighand_struct *sighand;
1035 
1036 	for (;;) {
1037 		sighand = rcu_dereference(tsk->sighand);
1038 		if (unlikely(sighand == NULL))
1039 			break;
1040 
1041 		spin_lock_irqsave(&sighand->siglock, *flags);
1042 		if (likely(sighand == tsk->sighand))
1043 			break;
1044 		spin_unlock_irqrestore(&sighand->siglock, *flags);
1045 	}
1046 
1047 	return sighand;
1048 }
1049 
1050 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1051 {
1052 	unsigned long flags;
1053 	int ret;
1054 
1055 	ret = check_kill_permission(sig, info, p);
1056 
1057 	if (!ret && sig) {
1058 		ret = -ESRCH;
1059 		if (lock_task_sighand(p, &flags)) {
1060 			ret = __group_send_sig_info(sig, info, p);
1061 			unlock_task_sighand(p, &flags);
1062 		}
1063 	}
1064 
1065 	return ret;
1066 }
1067 
1068 /*
1069  * kill_pgrp_info() sends a signal to a process group: this is what the tty
1070  * control characters do (^C, ^Z etc)
1071  */
1072 
1073 int __kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
1074 {
1075 	struct task_struct *p = NULL;
1076 	int retval, success;
1077 
1078 	success = 0;
1079 	retval = -ESRCH;
1080 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1081 		int err = group_send_sig_info(sig, info, p);
1082 		success |= !err;
1083 		retval = err;
1084 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1085 	return success ? 0 : retval;
1086 }
1087 
1088 int kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
1089 {
1090 	int retval;
1091 
1092 	read_lock(&tasklist_lock);
1093 	retval = __kill_pgrp_info(sig, info, pgrp);
1094 	read_unlock(&tasklist_lock);
1095 
1096 	return retval;
1097 }
1098 
1099 int kill_pid_info(int sig, struct siginfo *info, struct pid *pid)
1100 {
1101 	int error;
1102 	struct task_struct *p;
1103 
1104 	rcu_read_lock();
1105 	if (unlikely(sig_needs_tasklist(sig)))
1106 		read_lock(&tasklist_lock);
1107 
1108 	p = pid_task(pid, PIDTYPE_PID);
1109 	error = -ESRCH;
1110 	if (p)
1111 		error = group_send_sig_info(sig, info, p);
1112 
1113 	if (unlikely(sig_needs_tasklist(sig)))
1114 		read_unlock(&tasklist_lock);
1115 	rcu_read_unlock();
1116 	return error;
1117 }
1118 
1119 static int kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1120 {
1121 	int error;
1122 	rcu_read_lock();
1123 	error = kill_pid_info(sig, info, find_pid(pid));
1124 	rcu_read_unlock();
1125 	return error;
1126 }
1127 
1128 /* like kill_pid_info(), but doesn't use uid/euid of "current" */
1129 int kill_pid_info_as_uid(int sig, struct siginfo *info, struct pid *pid,
1130 		      uid_t uid, uid_t euid, u32 secid)
1131 {
1132 	int ret = -EINVAL;
1133 	struct task_struct *p;
1134 
1135 	if (!valid_signal(sig))
1136 		return ret;
1137 
1138 	read_lock(&tasklist_lock);
1139 	p = pid_task(pid, PIDTYPE_PID);
1140 	if (!p) {
1141 		ret = -ESRCH;
1142 		goto out_unlock;
1143 	}
1144 	if ((info == SEND_SIG_NOINFO || (!is_si_special(info) && SI_FROMUSER(info)))
1145 	    && (euid != p->suid) && (euid != p->uid)
1146 	    && (uid != p->suid) && (uid != p->uid)) {
1147 		ret = -EPERM;
1148 		goto out_unlock;
1149 	}
1150 	ret = security_task_kill(p, info, sig, secid);
1151 	if (ret)
1152 		goto out_unlock;
1153 	if (sig && p->sighand) {
1154 		unsigned long flags;
1155 		spin_lock_irqsave(&p->sighand->siglock, flags);
1156 		ret = __group_send_sig_info(sig, info, p);
1157 		spin_unlock_irqrestore(&p->sighand->siglock, flags);
1158 	}
1159 out_unlock:
1160 	read_unlock(&tasklist_lock);
1161 	return ret;
1162 }
1163 EXPORT_SYMBOL_GPL(kill_pid_info_as_uid);
1164 
1165 /*
1166  * kill_something_info() interprets pid in interesting ways just like kill(2).
1167  *
1168  * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1169  * is probably wrong.  Should make it like BSD or SYSV.
1170  */
1171 
1172 static int kill_something_info(int sig, struct siginfo *info, int pid)
1173 {
1174 	int ret;
1175 	rcu_read_lock();
1176 	if (!pid) {
1177 		ret = kill_pgrp_info(sig, info, task_pgrp(current));
1178 	} else if (pid == -1) {
1179 		int retval = 0, count = 0;
1180 		struct task_struct * p;
1181 
1182 		read_lock(&tasklist_lock);
1183 		for_each_process(p) {
1184 			if (p->pid > 1 && p->tgid != current->tgid) {
1185 				int err = group_send_sig_info(sig, info, p);
1186 				++count;
1187 				if (err != -EPERM)
1188 					retval = err;
1189 			}
1190 		}
1191 		read_unlock(&tasklist_lock);
1192 		ret = count ? retval : -ESRCH;
1193 	} else if (pid < 0) {
1194 		ret = kill_pgrp_info(sig, info, find_pid(-pid));
1195 	} else {
1196 		ret = kill_pid_info(sig, info, find_pid(pid));
1197 	}
1198 	rcu_read_unlock();
1199 	return ret;
1200 }
1201 
1202 /*
1203  * These are for backward compatibility with the rest of the kernel source.
1204  */
1205 
1206 /*
1207  * These two are the most common entry points.  They send a signal
1208  * just to the specific thread.
1209  */
1210 int
1211 send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1212 {
1213 	int ret;
1214 	unsigned long flags;
1215 
1216 	/*
1217 	 * Make sure legacy kernel users don't send in bad values
1218 	 * (normal paths check this in check_kill_permission).
1219 	 */
1220 	if (!valid_signal(sig))
1221 		return -EINVAL;
1222 
1223 	/*
1224 	 * We need the tasklist lock even for the specific
1225 	 * thread case (when we don't need to follow the group
1226 	 * lists) in order to avoid races with "p->sighand"
1227 	 * going away or changing from under us.
1228 	 */
1229 	read_lock(&tasklist_lock);
1230 	spin_lock_irqsave(&p->sighand->siglock, flags);
1231 	ret = specific_send_sig_info(sig, info, p);
1232 	spin_unlock_irqrestore(&p->sighand->siglock, flags);
1233 	read_unlock(&tasklist_lock);
1234 	return ret;
1235 }
1236 
1237 #define __si_special(priv) \
1238 	((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1239 
1240 int
1241 send_sig(int sig, struct task_struct *p, int priv)
1242 {
1243 	return send_sig_info(sig, __si_special(priv), p);
1244 }
1245 
1246 /*
1247  * This is the entry point for "process-wide" signals.
1248  * They will go to an appropriate thread in the thread group.
1249  */
1250 int
1251 send_group_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1252 {
1253 	int ret;
1254 	read_lock(&tasklist_lock);
1255 	ret = group_send_sig_info(sig, info, p);
1256 	read_unlock(&tasklist_lock);
1257 	return ret;
1258 }
1259 
1260 void
1261 force_sig(int sig, struct task_struct *p)
1262 {
1263 	force_sig_info(sig, SEND_SIG_PRIV, p);
1264 }
1265 
1266 /*
1267  * When things go south during signal handling, we
1268  * will force a SIGSEGV. And if the signal that caused
1269  * the problem was already a SIGSEGV, we'll want to
1270  * make sure we don't even try to deliver the signal..
1271  */
1272 int
1273 force_sigsegv(int sig, struct task_struct *p)
1274 {
1275 	if (sig == SIGSEGV) {
1276 		unsigned long flags;
1277 		spin_lock_irqsave(&p->sighand->siglock, flags);
1278 		p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1279 		spin_unlock_irqrestore(&p->sighand->siglock, flags);
1280 	}
1281 	force_sig(SIGSEGV, p);
1282 	return 0;
1283 }
1284 
1285 int kill_pgrp(struct pid *pid, int sig, int priv)
1286 {
1287 	return kill_pgrp_info(sig, __si_special(priv), pid);
1288 }
1289 EXPORT_SYMBOL(kill_pgrp);
1290 
1291 int kill_pid(struct pid *pid, int sig, int priv)
1292 {
1293 	return kill_pid_info(sig, __si_special(priv), pid);
1294 }
1295 EXPORT_SYMBOL(kill_pid);
1296 
1297 int
1298 kill_proc(pid_t pid, int sig, int priv)
1299 {
1300 	return kill_proc_info(sig, __si_special(priv), pid);
1301 }
1302 
1303 /*
1304  * These functions support sending signals using preallocated sigqueue
1305  * structures.  This is needed "because realtime applications cannot
1306  * afford to lose notifications of asynchronous events, like timer
1307  * expirations or I/O completions".  In the case of Posix Timers
1308  * we allocate the sigqueue structure from the timer_create.  If this
1309  * allocation fails we are able to report the failure to the application
1310  * with an EAGAIN error.
1311  */
1312 
1313 struct sigqueue *sigqueue_alloc(void)
1314 {
1315 	struct sigqueue *q;
1316 
1317 	if ((q = __sigqueue_alloc(current, GFP_KERNEL, 0)))
1318 		q->flags |= SIGQUEUE_PREALLOC;
1319 	return(q);
1320 }
1321 
1322 void sigqueue_free(struct sigqueue *q)
1323 {
1324 	unsigned long flags;
1325 	BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1326 	/*
1327 	 * If the signal is still pending remove it from the
1328 	 * pending queue.
1329 	 */
1330 	if (unlikely(!list_empty(&q->list))) {
1331 		spinlock_t *lock = &current->sighand->siglock;
1332 		read_lock(&tasklist_lock);
1333 		spin_lock_irqsave(lock, flags);
1334 		if (!list_empty(&q->list))
1335 			list_del_init(&q->list);
1336 		spin_unlock_irqrestore(lock, flags);
1337 		read_unlock(&tasklist_lock);
1338 	}
1339 	q->flags &= ~SIGQUEUE_PREALLOC;
1340 	__sigqueue_free(q);
1341 }
1342 
1343 int send_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
1344 {
1345 	unsigned long flags;
1346 	int ret = 0;
1347 
1348 	BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1349 
1350 	/*
1351 	 * The rcu based delayed sighand destroy makes it possible to
1352 	 * run this without tasklist lock held. The task struct itself
1353 	 * cannot go away as create_timer did get_task_struct().
1354 	 *
1355 	 * We return -1, when the task is marked exiting, so
1356 	 * posix_timer_event can redirect it to the group leader
1357 	 */
1358 	rcu_read_lock();
1359 
1360 	if (!likely(lock_task_sighand(p, &flags))) {
1361 		ret = -1;
1362 		goto out_err;
1363 	}
1364 
1365 	if (unlikely(!list_empty(&q->list))) {
1366 		/*
1367 		 * If an SI_TIMER entry is already queue just increment
1368 		 * the overrun count.
1369 		 */
1370 		BUG_ON(q->info.si_code != SI_TIMER);
1371 		q->info.si_overrun++;
1372 		goto out;
1373 	}
1374 	/* Short-circuit ignored signals.  */
1375 	if (sig_ignored(p, sig)) {
1376 		ret = 1;
1377 		goto out;
1378 	}
1379 
1380 	list_add_tail(&q->list, &p->pending.list);
1381 	sigaddset(&p->pending.signal, sig);
1382 	if (!sigismember(&p->blocked, sig))
1383 		signal_wake_up(p, sig == SIGKILL);
1384 
1385 out:
1386 	unlock_task_sighand(p, &flags);
1387 out_err:
1388 	rcu_read_unlock();
1389 
1390 	return ret;
1391 }
1392 
1393 int
1394 send_group_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
1395 {
1396 	unsigned long flags;
1397 	int ret = 0;
1398 
1399 	BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1400 
1401 	read_lock(&tasklist_lock);
1402 	/* Since it_lock is held, p->sighand cannot be NULL. */
1403 	spin_lock_irqsave(&p->sighand->siglock, flags);
1404 	handle_stop_signal(sig, p);
1405 
1406 	/* Short-circuit ignored signals.  */
1407 	if (sig_ignored(p, sig)) {
1408 		ret = 1;
1409 		goto out;
1410 	}
1411 
1412 	if (unlikely(!list_empty(&q->list))) {
1413 		/*
1414 		 * If an SI_TIMER entry is already queue just increment
1415 		 * the overrun count.  Other uses should not try to
1416 		 * send the signal multiple times.
1417 		 */
1418 		BUG_ON(q->info.si_code != SI_TIMER);
1419 		q->info.si_overrun++;
1420 		goto out;
1421 	}
1422 
1423 	/*
1424 	 * Put this signal on the shared-pending queue.
1425 	 * We always use the shared queue for process-wide signals,
1426 	 * to avoid several races.
1427 	 */
1428 	list_add_tail(&q->list, &p->signal->shared_pending.list);
1429 	sigaddset(&p->signal->shared_pending.signal, sig);
1430 
1431 	__group_complete_signal(sig, p);
1432 out:
1433 	spin_unlock_irqrestore(&p->sighand->siglock, flags);
1434 	read_unlock(&tasklist_lock);
1435 	return ret;
1436 }
1437 
1438 /*
1439  * Wake up any threads in the parent blocked in wait* syscalls.
1440  */
1441 static inline void __wake_up_parent(struct task_struct *p,
1442 				    struct task_struct *parent)
1443 {
1444 	wake_up_interruptible_sync(&parent->signal->wait_chldexit);
1445 }
1446 
1447 /*
1448  * Let a parent know about the death of a child.
1449  * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1450  */
1451 
1452 void do_notify_parent(struct task_struct *tsk, int sig)
1453 {
1454 	struct siginfo info;
1455 	unsigned long flags;
1456 	struct sighand_struct *psig;
1457 
1458 	BUG_ON(sig == -1);
1459 
1460  	/* do_notify_parent_cldstop should have been called instead.  */
1461  	BUG_ON(tsk->state & (TASK_STOPPED|TASK_TRACED));
1462 
1463 	BUG_ON(!tsk->ptrace &&
1464 	       (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1465 
1466 	info.si_signo = sig;
1467 	info.si_errno = 0;
1468 	info.si_pid = tsk->pid;
1469 	info.si_uid = tsk->uid;
1470 
1471 	/* FIXME: find out whether or not this is supposed to be c*time. */
1472 	info.si_utime = cputime_to_jiffies(cputime_add(tsk->utime,
1473 						       tsk->signal->utime));
1474 	info.si_stime = cputime_to_jiffies(cputime_add(tsk->stime,
1475 						       tsk->signal->stime));
1476 
1477 	info.si_status = tsk->exit_code & 0x7f;
1478 	if (tsk->exit_code & 0x80)
1479 		info.si_code = CLD_DUMPED;
1480 	else if (tsk->exit_code & 0x7f)
1481 		info.si_code = CLD_KILLED;
1482 	else {
1483 		info.si_code = CLD_EXITED;
1484 		info.si_status = tsk->exit_code >> 8;
1485 	}
1486 
1487 	psig = tsk->parent->sighand;
1488 	spin_lock_irqsave(&psig->siglock, flags);
1489 	if (!tsk->ptrace && sig == SIGCHLD &&
1490 	    (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1491 	     (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1492 		/*
1493 		 * We are exiting and our parent doesn't care.  POSIX.1
1494 		 * defines special semantics for setting SIGCHLD to SIG_IGN
1495 		 * or setting the SA_NOCLDWAIT flag: we should be reaped
1496 		 * automatically and not left for our parent's wait4 call.
1497 		 * Rather than having the parent do it as a magic kind of
1498 		 * signal handler, we just set this to tell do_exit that we
1499 		 * can be cleaned up without becoming a zombie.  Note that
1500 		 * we still call __wake_up_parent in this case, because a
1501 		 * blocked sys_wait4 might now return -ECHILD.
1502 		 *
1503 		 * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1504 		 * is implementation-defined: we do (if you don't want
1505 		 * it, just use SIG_IGN instead).
1506 		 */
1507 		tsk->exit_signal = -1;
1508 		if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1509 			sig = 0;
1510 	}
1511 	if (valid_signal(sig) && sig > 0)
1512 		__group_send_sig_info(sig, &info, tsk->parent);
1513 	__wake_up_parent(tsk, tsk->parent);
1514 	spin_unlock_irqrestore(&psig->siglock, flags);
1515 }
1516 
1517 static void do_notify_parent_cldstop(struct task_struct *tsk, int why)
1518 {
1519 	struct siginfo info;
1520 	unsigned long flags;
1521 	struct task_struct *parent;
1522 	struct sighand_struct *sighand;
1523 
1524 	if (tsk->ptrace & PT_PTRACED)
1525 		parent = tsk->parent;
1526 	else {
1527 		tsk = tsk->group_leader;
1528 		parent = tsk->real_parent;
1529 	}
1530 
1531 	info.si_signo = SIGCHLD;
1532 	info.si_errno = 0;
1533 	info.si_pid = tsk->pid;
1534 	info.si_uid = tsk->uid;
1535 
1536 	/* FIXME: find out whether or not this is supposed to be c*time. */
1537 	info.si_utime = cputime_to_jiffies(tsk->utime);
1538 	info.si_stime = cputime_to_jiffies(tsk->stime);
1539 
1540  	info.si_code = why;
1541  	switch (why) {
1542  	case CLD_CONTINUED:
1543  		info.si_status = SIGCONT;
1544  		break;
1545  	case CLD_STOPPED:
1546  		info.si_status = tsk->signal->group_exit_code & 0x7f;
1547  		break;
1548  	case CLD_TRAPPED:
1549  		info.si_status = tsk->exit_code & 0x7f;
1550  		break;
1551  	default:
1552  		BUG();
1553  	}
1554 
1555 	sighand = parent->sighand;
1556 	spin_lock_irqsave(&sighand->siglock, flags);
1557 	if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1558 	    !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1559 		__group_send_sig_info(SIGCHLD, &info, parent);
1560 	/*
1561 	 * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1562 	 */
1563 	__wake_up_parent(tsk, parent);
1564 	spin_unlock_irqrestore(&sighand->siglock, flags);
1565 }
1566 
1567 static inline int may_ptrace_stop(void)
1568 {
1569 	if (!likely(current->ptrace & PT_PTRACED))
1570 		return 0;
1571 
1572 	if (unlikely(current->parent == current->real_parent &&
1573 		    (current->ptrace & PT_ATTACHED)))
1574 		return 0;
1575 
1576 	if (unlikely(current->signal == current->parent->signal) &&
1577 	    unlikely(current->signal->flags & SIGNAL_GROUP_EXIT))
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 	 */
1650 	recalc_sigpending();
1651 }
1652 
1653 void ptrace_notify(int exit_code)
1654 {
1655 	siginfo_t info;
1656 
1657 	BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1658 
1659 	memset(&info, 0, sizeof info);
1660 	info.si_signo = SIGTRAP;
1661 	info.si_code = exit_code;
1662 	info.si_pid = current->pid;
1663 	info.si_uid = current->uid;
1664 
1665 	/* Let the debugger run.  */
1666 	spin_lock_irq(&current->sighand->siglock);
1667 	ptrace_stop(exit_code, 0, &info);
1668 	spin_unlock_irq(&current->sighand->siglock);
1669 }
1670 
1671 static void
1672 finish_stop(int stop_count)
1673 {
1674 	/*
1675 	 * If there are no other threads in the group, or if there is
1676 	 * a group stop in progress and we are the last to stop,
1677 	 * report to the parent.  When ptraced, every thread reports itself.
1678 	 */
1679 	if (stop_count == 0 || (current->ptrace & PT_PTRACED)) {
1680 		read_lock(&tasklist_lock);
1681 		do_notify_parent_cldstop(current, CLD_STOPPED);
1682 		read_unlock(&tasklist_lock);
1683 	}
1684 
1685 	do {
1686 		schedule();
1687 	} while (try_to_freeze());
1688 	/*
1689 	 * Now we don't run again until continued.
1690 	 */
1691 	current->exit_code = 0;
1692 }
1693 
1694 /*
1695  * This performs the stopping for SIGSTOP and other stop signals.
1696  * We have to stop all threads in the thread group.
1697  * Returns nonzero if we've actually stopped and released the siglock.
1698  * Returns zero if we didn't stop and still hold the siglock.
1699  */
1700 static int do_signal_stop(int signr)
1701 {
1702 	struct signal_struct *sig = current->signal;
1703 	int stop_count;
1704 
1705 	if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED))
1706 		return 0;
1707 
1708 	if (sig->group_stop_count > 0) {
1709 		/*
1710 		 * There is a group stop in progress.  We don't need to
1711 		 * start another one.
1712 		 */
1713 		stop_count = --sig->group_stop_count;
1714 	} else {
1715 		/*
1716 		 * There is no group stop already in progress.
1717 		 * We must initiate one now.
1718 		 */
1719 		struct task_struct *t;
1720 
1721 		sig->group_exit_code = signr;
1722 
1723 		stop_count = 0;
1724 		for (t = next_thread(current); t != current; t = next_thread(t))
1725 			/*
1726 			 * Setting state to TASK_STOPPED for a group
1727 			 * stop is always done with the siglock held,
1728 			 * so this check has no races.
1729 			 */
1730 			if (!t->exit_state &&
1731 			    !(t->state & (TASK_STOPPED|TASK_TRACED))) {
1732 				stop_count++;
1733 				signal_wake_up(t, 0);
1734 			}
1735 		sig->group_stop_count = stop_count;
1736 	}
1737 
1738 	if (stop_count == 0)
1739 		sig->flags = SIGNAL_STOP_STOPPED;
1740 	current->exit_code = sig->group_exit_code;
1741 	__set_current_state(TASK_STOPPED);
1742 
1743 	spin_unlock_irq(&current->sighand->siglock);
1744 	finish_stop(stop_count);
1745 	return 1;
1746 }
1747 
1748 /*
1749  * Do appropriate magic when group_stop_count > 0.
1750  * We return nonzero if we stopped, after releasing the siglock.
1751  * We return zero if we still hold the siglock and should look
1752  * for another signal without checking group_stop_count again.
1753  */
1754 static int handle_group_stop(void)
1755 {
1756 	int stop_count;
1757 
1758 	if (current->signal->group_exit_task == current) {
1759 		/*
1760 		 * Group stop is so we can do a core dump,
1761 		 * We are the initiating thread, so get on with it.
1762 		 */
1763 		current->signal->group_exit_task = NULL;
1764 		return 0;
1765 	}
1766 
1767 	if (current->signal->flags & SIGNAL_GROUP_EXIT)
1768 		/*
1769 		 * Group stop is so another thread can do a core dump,
1770 		 * or else we are racing against a death signal.
1771 		 * Just punt the stop so we can get the next signal.
1772 		 */
1773 		return 0;
1774 
1775 	/*
1776 	 * There is a group stop in progress.  We stop
1777 	 * without any associated signal being in our queue.
1778 	 */
1779 	stop_count = --current->signal->group_stop_count;
1780 	if (stop_count == 0)
1781 		current->signal->flags = SIGNAL_STOP_STOPPED;
1782 	current->exit_code = current->signal->group_exit_code;
1783 	set_current_state(TASK_STOPPED);
1784 	spin_unlock_irq(&current->sighand->siglock);
1785 	finish_stop(stop_count);
1786 	return 1;
1787 }
1788 
1789 int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
1790 			  struct pt_regs *regs, void *cookie)
1791 {
1792 	sigset_t *mask = &current->blocked;
1793 	int signr = 0;
1794 
1795 	try_to_freeze();
1796 
1797 relock:
1798 	spin_lock_irq(&current->sighand->siglock);
1799 	for (;;) {
1800 		struct k_sigaction *ka;
1801 
1802 		if (unlikely(current->signal->group_stop_count > 0) &&
1803 		    handle_group_stop())
1804 			goto relock;
1805 
1806 		signr = dequeue_signal(current, mask, info);
1807 
1808 		if (!signr)
1809 			break; /* will return 0 */
1810 
1811 		if ((current->ptrace & PT_PTRACED) && signr != SIGKILL) {
1812 			ptrace_signal_deliver(regs, cookie);
1813 
1814 			/* Let the debugger run.  */
1815 			ptrace_stop(signr, signr, info);
1816 
1817 			/* We're back.  Did the debugger cancel the sig?  */
1818 			signr = current->exit_code;
1819 			if (signr == 0)
1820 				continue;
1821 
1822 			current->exit_code = 0;
1823 
1824 			/* Update the siginfo structure if the signal has
1825 			   changed.  If the debugger wanted something
1826 			   specific in the siginfo structure then it should
1827 			   have updated *info via PTRACE_SETSIGINFO.  */
1828 			if (signr != info->si_signo) {
1829 				info->si_signo = signr;
1830 				info->si_errno = 0;
1831 				info->si_code = SI_USER;
1832 				info->si_pid = current->parent->pid;
1833 				info->si_uid = current->parent->uid;
1834 			}
1835 
1836 			/* If the (new) signal is now blocked, requeue it.  */
1837 			if (sigismember(&current->blocked, signr)) {
1838 				specific_send_sig_info(signr, info, current);
1839 				continue;
1840 			}
1841 		}
1842 
1843 		ka = &current->sighand->action[signr-1];
1844 		if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
1845 			continue;
1846 		if (ka->sa.sa_handler != SIG_DFL) {
1847 			/* Run the handler.  */
1848 			*return_ka = *ka;
1849 
1850 			if (ka->sa.sa_flags & SA_ONESHOT)
1851 				ka->sa.sa_handler = SIG_DFL;
1852 
1853 			break; /* will return non-zero "signr" value */
1854 		}
1855 
1856 		/*
1857 		 * Now we are doing the default action for this signal.
1858 		 */
1859 		if (sig_kernel_ignore(signr)) /* Default is nothing. */
1860 			continue;
1861 
1862 		/*
1863 		 * Init of a pid space gets no signals it doesn't want from
1864 		 * within that pid space. It can of course get signals from
1865 		 * its parent pid space.
1866 		 */
1867 		if (current == child_reaper(current))
1868 			continue;
1869 
1870 		if (sig_kernel_stop(signr)) {
1871 			/*
1872 			 * The default action is to stop all threads in
1873 			 * the thread group.  The job control signals
1874 			 * do nothing in an orphaned pgrp, but SIGSTOP
1875 			 * always works.  Note that siglock needs to be
1876 			 * dropped during the call to is_orphaned_pgrp()
1877 			 * because of lock ordering with tasklist_lock.
1878 			 * This allows an intervening SIGCONT to be posted.
1879 			 * We need to check for that and bail out if necessary.
1880 			 */
1881 			if (signr != SIGSTOP) {
1882 				spin_unlock_irq(&current->sighand->siglock);
1883 
1884 				/* signals can be posted during this window */
1885 
1886 				if (is_current_pgrp_orphaned())
1887 					goto relock;
1888 
1889 				spin_lock_irq(&current->sighand->siglock);
1890 			}
1891 
1892 			if (likely(do_signal_stop(signr))) {
1893 				/* It released the siglock.  */
1894 				goto relock;
1895 			}
1896 
1897 			/*
1898 			 * We didn't actually stop, due to a race
1899 			 * with SIGCONT or something like that.
1900 			 */
1901 			continue;
1902 		}
1903 
1904 		spin_unlock_irq(&current->sighand->siglock);
1905 
1906 		/*
1907 		 * Anything else is fatal, maybe with a core dump.
1908 		 */
1909 		current->flags |= PF_SIGNALED;
1910 		if (sig_kernel_coredump(signr)) {
1911 			/*
1912 			 * If it was able to dump core, this kills all
1913 			 * other threads in the group and synchronizes with
1914 			 * their demise.  If we lost the race with another
1915 			 * thread getting here, it set group_exit_code
1916 			 * first and our do_group_exit call below will use
1917 			 * that value and ignore the one we pass it.
1918 			 */
1919 			do_coredump((long)signr, signr, regs);
1920 		}
1921 
1922 		/*
1923 		 * Death signals, no core dump.
1924 		 */
1925 		do_group_exit(signr);
1926 		/* NOTREACHED */
1927 	}
1928 	spin_unlock_irq(&current->sighand->siglock);
1929 	return signr;
1930 }
1931 
1932 EXPORT_SYMBOL(recalc_sigpending);
1933 EXPORT_SYMBOL_GPL(dequeue_signal);
1934 EXPORT_SYMBOL(flush_signals);
1935 EXPORT_SYMBOL(force_sig);
1936 EXPORT_SYMBOL(kill_proc);
1937 EXPORT_SYMBOL(ptrace_notify);
1938 EXPORT_SYMBOL(send_sig);
1939 EXPORT_SYMBOL(send_sig_info);
1940 EXPORT_SYMBOL(sigprocmask);
1941 EXPORT_SYMBOL(block_all_signals);
1942 EXPORT_SYMBOL(unblock_all_signals);
1943 
1944 
1945 /*
1946  * System call entry points.
1947  */
1948 
1949 asmlinkage long sys_restart_syscall(void)
1950 {
1951 	struct restart_block *restart = &current_thread_info()->restart_block;
1952 	return restart->fn(restart);
1953 }
1954 
1955 long do_no_restart_syscall(struct restart_block *param)
1956 {
1957 	return -EINTR;
1958 }
1959 
1960 /*
1961  * We don't need to get the kernel lock - this is all local to this
1962  * particular thread.. (and that's good, because this is _heavily_
1963  * used by various programs)
1964  */
1965 
1966 /*
1967  * This is also useful for kernel threads that want to temporarily
1968  * (or permanently) block certain signals.
1969  *
1970  * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
1971  * interface happily blocks "unblockable" signals like SIGKILL
1972  * and friends.
1973  */
1974 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
1975 {
1976 	int error;
1977 
1978 	spin_lock_irq(&current->sighand->siglock);
1979 	if (oldset)
1980 		*oldset = current->blocked;
1981 
1982 	error = 0;
1983 	switch (how) {
1984 	case SIG_BLOCK:
1985 		sigorsets(&current->blocked, &current->blocked, set);
1986 		break;
1987 	case SIG_UNBLOCK:
1988 		signandsets(&current->blocked, &current->blocked, set);
1989 		break;
1990 	case SIG_SETMASK:
1991 		current->blocked = *set;
1992 		break;
1993 	default:
1994 		error = -EINVAL;
1995 	}
1996 	recalc_sigpending();
1997 	spin_unlock_irq(&current->sighand->siglock);
1998 
1999 	return error;
2000 }
2001 
2002 asmlinkage long
2003 sys_rt_sigprocmask(int how, sigset_t __user *set, sigset_t __user *oset, size_t sigsetsize)
2004 {
2005 	int error = -EINVAL;
2006 	sigset_t old_set, new_set;
2007 
2008 	/* XXX: Don't preclude handling different sized sigset_t's.  */
2009 	if (sigsetsize != sizeof(sigset_t))
2010 		goto out;
2011 
2012 	if (set) {
2013 		error = -EFAULT;
2014 		if (copy_from_user(&new_set, set, sizeof(*set)))
2015 			goto out;
2016 		sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2017 
2018 		error = sigprocmask(how, &new_set, &old_set);
2019 		if (error)
2020 			goto out;
2021 		if (oset)
2022 			goto set_old;
2023 	} else if (oset) {
2024 		spin_lock_irq(&current->sighand->siglock);
2025 		old_set = current->blocked;
2026 		spin_unlock_irq(&current->sighand->siglock);
2027 
2028 	set_old:
2029 		error = -EFAULT;
2030 		if (copy_to_user(oset, &old_set, sizeof(*oset)))
2031 			goto out;
2032 	}
2033 	error = 0;
2034 out:
2035 	return error;
2036 }
2037 
2038 long do_sigpending(void __user *set, unsigned long sigsetsize)
2039 {
2040 	long error = -EINVAL;
2041 	sigset_t pending;
2042 
2043 	if (sigsetsize > sizeof(sigset_t))
2044 		goto out;
2045 
2046 	spin_lock_irq(&current->sighand->siglock);
2047 	sigorsets(&pending, &current->pending.signal,
2048 		  &current->signal->shared_pending.signal);
2049 	spin_unlock_irq(&current->sighand->siglock);
2050 
2051 	/* Outside the lock because only this thread touches it.  */
2052 	sigandsets(&pending, &current->blocked, &pending);
2053 
2054 	error = -EFAULT;
2055 	if (!copy_to_user(set, &pending, sigsetsize))
2056 		error = 0;
2057 
2058 out:
2059 	return error;
2060 }
2061 
2062 asmlinkage long
2063 sys_rt_sigpending(sigset_t __user *set, size_t sigsetsize)
2064 {
2065 	return do_sigpending(set, sigsetsize);
2066 }
2067 
2068 #ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2069 
2070 int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2071 {
2072 	int err;
2073 
2074 	if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2075 		return -EFAULT;
2076 	if (from->si_code < 0)
2077 		return __copy_to_user(to, from, sizeof(siginfo_t))
2078 			? -EFAULT : 0;
2079 	/*
2080 	 * If you change siginfo_t structure, please be sure
2081 	 * this code is fixed accordingly.
2082 	 * It should never copy any pad contained in the structure
2083 	 * to avoid security leaks, but must copy the generic
2084 	 * 3 ints plus the relevant union member.
2085 	 */
2086 	err = __put_user(from->si_signo, &to->si_signo);
2087 	err |= __put_user(from->si_errno, &to->si_errno);
2088 	err |= __put_user((short)from->si_code, &to->si_code);
2089 	switch (from->si_code & __SI_MASK) {
2090 	case __SI_KILL:
2091 		err |= __put_user(from->si_pid, &to->si_pid);
2092 		err |= __put_user(from->si_uid, &to->si_uid);
2093 		break;
2094 	case __SI_TIMER:
2095 		 err |= __put_user(from->si_tid, &to->si_tid);
2096 		 err |= __put_user(from->si_overrun, &to->si_overrun);
2097 		 err |= __put_user(from->si_ptr, &to->si_ptr);
2098 		break;
2099 	case __SI_POLL:
2100 		err |= __put_user(from->si_band, &to->si_band);
2101 		err |= __put_user(from->si_fd, &to->si_fd);
2102 		break;
2103 	case __SI_FAULT:
2104 		err |= __put_user(from->si_addr, &to->si_addr);
2105 #ifdef __ARCH_SI_TRAPNO
2106 		err |= __put_user(from->si_trapno, &to->si_trapno);
2107 #endif
2108 		break;
2109 	case __SI_CHLD:
2110 		err |= __put_user(from->si_pid, &to->si_pid);
2111 		err |= __put_user(from->si_uid, &to->si_uid);
2112 		err |= __put_user(from->si_status, &to->si_status);
2113 		err |= __put_user(from->si_utime, &to->si_utime);
2114 		err |= __put_user(from->si_stime, &to->si_stime);
2115 		break;
2116 	case __SI_RT: /* This is not generated by the kernel as of now. */
2117 	case __SI_MESGQ: /* But this is */
2118 		err |= __put_user(from->si_pid, &to->si_pid);
2119 		err |= __put_user(from->si_uid, &to->si_uid);
2120 		err |= __put_user(from->si_ptr, &to->si_ptr);
2121 		break;
2122 	default: /* this is just in case for now ... */
2123 		err |= __put_user(from->si_pid, &to->si_pid);
2124 		err |= __put_user(from->si_uid, &to->si_uid);
2125 		break;
2126 	}
2127 	return err;
2128 }
2129 
2130 #endif
2131 
2132 asmlinkage long
2133 sys_rt_sigtimedwait(const sigset_t __user *uthese,
2134 		    siginfo_t __user *uinfo,
2135 		    const struct timespec __user *uts,
2136 		    size_t sigsetsize)
2137 {
2138 	int ret, sig;
2139 	sigset_t these;
2140 	struct timespec ts;
2141 	siginfo_t info;
2142 	long timeout = 0;
2143 
2144 	/* XXX: Don't preclude handling different sized sigset_t's.  */
2145 	if (sigsetsize != sizeof(sigset_t))
2146 		return -EINVAL;
2147 
2148 	if (copy_from_user(&these, uthese, sizeof(these)))
2149 		return -EFAULT;
2150 
2151 	/*
2152 	 * Invert the set of allowed signals to get those we
2153 	 * want to block.
2154 	 */
2155 	sigdelsetmask(&these, sigmask(SIGKILL)|sigmask(SIGSTOP));
2156 	signotset(&these);
2157 
2158 	if (uts) {
2159 		if (copy_from_user(&ts, uts, sizeof(ts)))
2160 			return -EFAULT;
2161 		if (ts.tv_nsec >= 1000000000L || ts.tv_nsec < 0
2162 		    || ts.tv_sec < 0)
2163 			return -EINVAL;
2164 	}
2165 
2166 	spin_lock_irq(&current->sighand->siglock);
2167 	sig = dequeue_signal(current, &these, &info);
2168 	if (!sig) {
2169 		timeout = MAX_SCHEDULE_TIMEOUT;
2170 		if (uts)
2171 			timeout = (timespec_to_jiffies(&ts)
2172 				   + (ts.tv_sec || ts.tv_nsec));
2173 
2174 		if (timeout) {
2175 			/* None ready -- temporarily unblock those we're
2176 			 * interested while we are sleeping in so that we'll
2177 			 * be awakened when they arrive.  */
2178 			current->real_blocked = current->blocked;
2179 			sigandsets(&current->blocked, &current->blocked, &these);
2180 			recalc_sigpending();
2181 			spin_unlock_irq(&current->sighand->siglock);
2182 
2183 			timeout = schedule_timeout_interruptible(timeout);
2184 
2185 			spin_lock_irq(&current->sighand->siglock);
2186 			sig = dequeue_signal(current, &these, &info);
2187 			current->blocked = current->real_blocked;
2188 			siginitset(&current->real_blocked, 0);
2189 			recalc_sigpending();
2190 		}
2191 	}
2192 	spin_unlock_irq(&current->sighand->siglock);
2193 
2194 	if (sig) {
2195 		ret = sig;
2196 		if (uinfo) {
2197 			if (copy_siginfo_to_user(uinfo, &info))
2198 				ret = -EFAULT;
2199 		}
2200 	} else {
2201 		ret = -EAGAIN;
2202 		if (timeout)
2203 			ret = -EINTR;
2204 	}
2205 
2206 	return ret;
2207 }
2208 
2209 asmlinkage long
2210 sys_kill(int pid, int sig)
2211 {
2212 	struct siginfo info;
2213 
2214 	info.si_signo = sig;
2215 	info.si_errno = 0;
2216 	info.si_code = SI_USER;
2217 	info.si_pid = current->tgid;
2218 	info.si_uid = current->uid;
2219 
2220 	return kill_something_info(sig, &info, pid);
2221 }
2222 
2223 static int do_tkill(int tgid, int pid, int sig)
2224 {
2225 	int error;
2226 	struct siginfo info;
2227 	struct task_struct *p;
2228 
2229 	error = -ESRCH;
2230 	info.si_signo = sig;
2231 	info.si_errno = 0;
2232 	info.si_code = SI_TKILL;
2233 	info.si_pid = current->tgid;
2234 	info.si_uid = current->uid;
2235 
2236 	read_lock(&tasklist_lock);
2237 	p = find_task_by_pid(pid);
2238 	if (p && (tgid <= 0 || p->tgid == tgid)) {
2239 		error = check_kill_permission(sig, &info, p);
2240 		/*
2241 		 * The null signal is a permissions and process existence
2242 		 * probe.  No signal is actually delivered.
2243 		 */
2244 		if (!error && sig && p->sighand) {
2245 			spin_lock_irq(&p->sighand->siglock);
2246 			handle_stop_signal(sig, p);
2247 			error = specific_send_sig_info(sig, &info, p);
2248 			spin_unlock_irq(&p->sighand->siglock);
2249 		}
2250 	}
2251 	read_unlock(&tasklist_lock);
2252 
2253 	return error;
2254 }
2255 
2256 /**
2257  *  sys_tgkill - send signal to one specific thread
2258  *  @tgid: the thread group ID of the thread
2259  *  @pid: the PID of the thread
2260  *  @sig: signal to be sent
2261  *
2262  *  This syscall also checks the @tgid and returns -ESRCH even if the PID
2263  *  exists but it's not belonging to the target process anymore. This
2264  *  method solves the problem of threads exiting and PIDs getting reused.
2265  */
2266 asmlinkage long sys_tgkill(int tgid, int pid, int sig)
2267 {
2268 	/* This is only valid for single tasks */
2269 	if (pid <= 0 || tgid <= 0)
2270 		return -EINVAL;
2271 
2272 	return do_tkill(tgid, pid, sig);
2273 }
2274 
2275 /*
2276  *  Send a signal to only one task, even if it's a CLONE_THREAD task.
2277  */
2278 asmlinkage long
2279 sys_tkill(int pid, int sig)
2280 {
2281 	/* This is only valid for single tasks */
2282 	if (pid <= 0)
2283 		return -EINVAL;
2284 
2285 	return do_tkill(0, pid, sig);
2286 }
2287 
2288 asmlinkage long
2289 sys_rt_sigqueueinfo(int pid, int sig, siginfo_t __user *uinfo)
2290 {
2291 	siginfo_t info;
2292 
2293 	if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2294 		return -EFAULT;
2295 
2296 	/* Not even root can pretend to send signals from the kernel.
2297 	   Nor can they impersonate a kill(), which adds source info.  */
2298 	if (info.si_code >= 0)
2299 		return -EPERM;
2300 	info.si_signo = sig;
2301 
2302 	/* POSIX.1b doesn't mention process groups.  */
2303 	return kill_proc_info(sig, &info, pid);
2304 }
2305 
2306 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
2307 {
2308 	struct k_sigaction *k;
2309 	sigset_t mask;
2310 
2311 	if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
2312 		return -EINVAL;
2313 
2314 	k = &current->sighand->action[sig-1];
2315 
2316 	spin_lock_irq(&current->sighand->siglock);
2317 	if (signal_pending(current)) {
2318 		/*
2319 		 * If there might be a fatal signal pending on multiple
2320 		 * threads, make sure we take it before changing the action.
2321 		 */
2322 		spin_unlock_irq(&current->sighand->siglock);
2323 		return -ERESTARTNOINTR;
2324 	}
2325 
2326 	if (oact)
2327 		*oact = *k;
2328 
2329 	if (act) {
2330 		sigdelsetmask(&act->sa.sa_mask,
2331 			      sigmask(SIGKILL) | sigmask(SIGSTOP));
2332 		*k = *act;
2333 		/*
2334 		 * POSIX 3.3.1.3:
2335 		 *  "Setting a signal action to SIG_IGN for a signal that is
2336 		 *   pending shall cause the pending signal to be discarded,
2337 		 *   whether or not it is blocked."
2338 		 *
2339 		 *  "Setting a signal action to SIG_DFL for a signal that is
2340 		 *   pending and whose default action is to ignore the signal
2341 		 *   (for example, SIGCHLD), shall cause the pending signal to
2342 		 *   be discarded, whether or not it is blocked"
2343 		 */
2344 		if (act->sa.sa_handler == SIG_IGN ||
2345 		   (act->sa.sa_handler == SIG_DFL && sig_kernel_ignore(sig))) {
2346 			struct task_struct *t = current;
2347 			sigemptyset(&mask);
2348 			sigaddset(&mask, sig);
2349 			rm_from_queue_full(&mask, &t->signal->shared_pending);
2350 			do {
2351 				rm_from_queue_full(&mask, &t->pending);
2352 				recalc_sigpending_tsk(t);
2353 				t = next_thread(t);
2354 			} while (t != current);
2355 		}
2356 	}
2357 
2358 	spin_unlock_irq(&current->sighand->siglock);
2359 	return 0;
2360 }
2361 
2362 int
2363 do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
2364 {
2365 	stack_t oss;
2366 	int error;
2367 
2368 	if (uoss) {
2369 		oss.ss_sp = (void __user *) current->sas_ss_sp;
2370 		oss.ss_size = current->sas_ss_size;
2371 		oss.ss_flags = sas_ss_flags(sp);
2372 	}
2373 
2374 	if (uss) {
2375 		void __user *ss_sp;
2376 		size_t ss_size;
2377 		int ss_flags;
2378 
2379 		error = -EFAULT;
2380 		if (!access_ok(VERIFY_READ, uss, sizeof(*uss))
2381 		    || __get_user(ss_sp, &uss->ss_sp)
2382 		    || __get_user(ss_flags, &uss->ss_flags)
2383 		    || __get_user(ss_size, &uss->ss_size))
2384 			goto out;
2385 
2386 		error = -EPERM;
2387 		if (on_sig_stack(sp))
2388 			goto out;
2389 
2390 		error = -EINVAL;
2391 		/*
2392 		 *
2393 		 * Note - this code used to test ss_flags incorrectly
2394 		 *  	  old code may have been written using ss_flags==0
2395 		 *	  to mean ss_flags==SS_ONSTACK (as this was the only
2396 		 *	  way that worked) - this fix preserves that older
2397 		 *	  mechanism
2398 		 */
2399 		if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
2400 			goto out;
2401 
2402 		if (ss_flags == SS_DISABLE) {
2403 			ss_size = 0;
2404 			ss_sp = NULL;
2405 		} else {
2406 			error = -ENOMEM;
2407 			if (ss_size < MINSIGSTKSZ)
2408 				goto out;
2409 		}
2410 
2411 		current->sas_ss_sp = (unsigned long) ss_sp;
2412 		current->sas_ss_size = ss_size;
2413 	}
2414 
2415 	if (uoss) {
2416 		error = -EFAULT;
2417 		if (copy_to_user(uoss, &oss, sizeof(oss)))
2418 			goto out;
2419 	}
2420 
2421 	error = 0;
2422 out:
2423 	return error;
2424 }
2425 
2426 #ifdef __ARCH_WANT_SYS_SIGPENDING
2427 
2428 asmlinkage long
2429 sys_sigpending(old_sigset_t __user *set)
2430 {
2431 	return do_sigpending(set, sizeof(*set));
2432 }
2433 
2434 #endif
2435 
2436 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
2437 /* Some platforms have their own version with special arguments others
2438    support only sys_rt_sigprocmask.  */
2439 
2440 asmlinkage long
2441 sys_sigprocmask(int how, old_sigset_t __user *set, old_sigset_t __user *oset)
2442 {
2443 	int error;
2444 	old_sigset_t old_set, new_set;
2445 
2446 	if (set) {
2447 		error = -EFAULT;
2448 		if (copy_from_user(&new_set, set, sizeof(*set)))
2449 			goto out;
2450 		new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
2451 
2452 		spin_lock_irq(&current->sighand->siglock);
2453 		old_set = current->blocked.sig[0];
2454 
2455 		error = 0;
2456 		switch (how) {
2457 		default:
2458 			error = -EINVAL;
2459 			break;
2460 		case SIG_BLOCK:
2461 			sigaddsetmask(&current->blocked, new_set);
2462 			break;
2463 		case SIG_UNBLOCK:
2464 			sigdelsetmask(&current->blocked, new_set);
2465 			break;
2466 		case SIG_SETMASK:
2467 			current->blocked.sig[0] = new_set;
2468 			break;
2469 		}
2470 
2471 		recalc_sigpending();
2472 		spin_unlock_irq(&current->sighand->siglock);
2473 		if (error)
2474 			goto out;
2475 		if (oset)
2476 			goto set_old;
2477 	} else if (oset) {
2478 		old_set = current->blocked.sig[0];
2479 	set_old:
2480 		error = -EFAULT;
2481 		if (copy_to_user(oset, &old_set, sizeof(*oset)))
2482 			goto out;
2483 	}
2484 	error = 0;
2485 out:
2486 	return error;
2487 }
2488 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
2489 
2490 #ifdef __ARCH_WANT_SYS_RT_SIGACTION
2491 asmlinkage long
2492 sys_rt_sigaction(int sig,
2493 		 const struct sigaction __user *act,
2494 		 struct sigaction __user *oact,
2495 		 size_t sigsetsize)
2496 {
2497 	struct k_sigaction new_sa, old_sa;
2498 	int ret = -EINVAL;
2499 
2500 	/* XXX: Don't preclude handling different sized sigset_t's.  */
2501 	if (sigsetsize != sizeof(sigset_t))
2502 		goto out;
2503 
2504 	if (act) {
2505 		if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
2506 			return -EFAULT;
2507 	}
2508 
2509 	ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
2510 
2511 	if (!ret && oact) {
2512 		if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
2513 			return -EFAULT;
2514 	}
2515 out:
2516 	return ret;
2517 }
2518 #endif /* __ARCH_WANT_SYS_RT_SIGACTION */
2519 
2520 #ifdef __ARCH_WANT_SYS_SGETMASK
2521 
2522 /*
2523  * For backwards compatibility.  Functionality superseded by sigprocmask.
2524  */
2525 asmlinkage long
2526 sys_sgetmask(void)
2527 {
2528 	/* SMP safe */
2529 	return current->blocked.sig[0];
2530 }
2531 
2532 asmlinkage long
2533 sys_ssetmask(int newmask)
2534 {
2535 	int old;
2536 
2537 	spin_lock_irq(&current->sighand->siglock);
2538 	old = current->blocked.sig[0];
2539 
2540 	siginitset(&current->blocked, newmask & ~(sigmask(SIGKILL)|
2541 						  sigmask(SIGSTOP)));
2542 	recalc_sigpending();
2543 	spin_unlock_irq(&current->sighand->siglock);
2544 
2545 	return old;
2546 }
2547 #endif /* __ARCH_WANT_SGETMASK */
2548 
2549 #ifdef __ARCH_WANT_SYS_SIGNAL
2550 /*
2551  * For backwards compatibility.  Functionality superseded by sigaction.
2552  */
2553 asmlinkage unsigned long
2554 sys_signal(int sig, __sighandler_t handler)
2555 {
2556 	struct k_sigaction new_sa, old_sa;
2557 	int ret;
2558 
2559 	new_sa.sa.sa_handler = handler;
2560 	new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
2561 	sigemptyset(&new_sa.sa.sa_mask);
2562 
2563 	ret = do_sigaction(sig, &new_sa, &old_sa);
2564 
2565 	return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
2566 }
2567 #endif /* __ARCH_WANT_SYS_SIGNAL */
2568 
2569 #ifdef __ARCH_WANT_SYS_PAUSE
2570 
2571 asmlinkage long
2572 sys_pause(void)
2573 {
2574 	current->state = TASK_INTERRUPTIBLE;
2575 	schedule();
2576 	return -ERESTARTNOHAND;
2577 }
2578 
2579 #endif
2580 
2581 #ifdef __ARCH_WANT_SYS_RT_SIGSUSPEND
2582 asmlinkage long sys_rt_sigsuspend(sigset_t __user *unewset, size_t sigsetsize)
2583 {
2584 	sigset_t newset;
2585 
2586 	/* XXX: Don't preclude handling different sized sigset_t's.  */
2587 	if (sigsetsize != sizeof(sigset_t))
2588 		return -EINVAL;
2589 
2590 	if (copy_from_user(&newset, unewset, sizeof(newset)))
2591 		return -EFAULT;
2592 	sigdelsetmask(&newset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2593 
2594 	spin_lock_irq(&current->sighand->siglock);
2595 	current->saved_sigmask = current->blocked;
2596 	current->blocked = newset;
2597 	recalc_sigpending();
2598 	spin_unlock_irq(&current->sighand->siglock);
2599 
2600 	current->state = TASK_INTERRUPTIBLE;
2601 	schedule();
2602 	set_thread_flag(TIF_RESTORE_SIGMASK);
2603 	return -ERESTARTNOHAND;
2604 }
2605 #endif /* __ARCH_WANT_SYS_RT_SIGSUSPEND */
2606 
2607 __attribute__((weak)) const char *arch_vma_name(struct vm_area_struct *vma)
2608 {
2609 	return NULL;
2610 }
2611 
2612 void __init signals_init(void)
2613 {
2614 	sigqueue_cachep =
2615 		kmem_cache_create("sigqueue",
2616 				  sizeof(struct sigqueue),
2617 				  __alignof__(struct sigqueue),
2618 				  SLAB_PANIC, NULL, NULL);
2619 }
2620