ACPICA: Preserve all PM control reserved and ignored bits
[linux-2.6] / kernel / signal.c
1 /*
2  *  linux/kernel/signal.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  *
6  *  1997-11-02  Modified for POSIX.1b signals by Richard Henderson
7  *
8  *  2003-06-02  Jim Houston - Concurrent Computer Corp.
9  *              Changes to use preallocated sigqueue structures
10  *              to allow signals to be sent reliably.
11  */
12
13 #include <linux/slab.h>
14 #include <linux/module.h>
15 #include <linux/init.h>
16 #include <linux/sched.h>
17 #include <linux/fs.h>
18 #include <linux/tty.h>
19 #include <linux/binfmts.h>
20 #include <linux/security.h>
21 #include <linux/syscalls.h>
22 #include <linux/ptrace.h>
23 #include <linux/signal.h>
24 #include <linux/signalfd.h>
25 #include <linux/tracehook.h>
26 #include <linux/capability.h>
27 #include <linux/freezer.h>
28 #include <linux/pid_namespace.h>
29 #include <linux/nsproxy.h>
30 #include <trace/sched.h>
31
32 #include <asm/param.h>
33 #include <asm/uaccess.h>
34 #include <asm/unistd.h>
35 #include <asm/siginfo.h>
36 #include "audit.h"      /* audit_signal_info() */
37
38 /*
39  * SLAB caches for signal bits.
40  */
41
42 static struct kmem_cache *sigqueue_cachep;
43
44 DEFINE_TRACE(sched_signal_send);
45
46 static void __user *sig_handler(struct task_struct *t, int sig)
47 {
48         return t->sighand->action[sig - 1].sa.sa_handler;
49 }
50
51 static int sig_handler_ignored(void __user *handler, int sig)
52 {
53         /* Is it explicitly or implicitly ignored? */
54         return handler == SIG_IGN ||
55                 (handler == SIG_DFL && sig_kernel_ignore(sig));
56 }
57
58 static int sig_ignored(struct task_struct *t, int sig)
59 {
60         void __user *handler;
61
62         /*
63          * Blocked signals are never ignored, since the
64          * signal handler may change by the time it is
65          * unblocked.
66          */
67         if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
68                 return 0;
69
70         handler = sig_handler(t, sig);
71         if (!sig_handler_ignored(handler, sig))
72                 return 0;
73
74         /*
75          * Tracers may want to know about even ignored signals.
76          */
77         return !tracehook_consider_ignored_signal(t, sig, handler);
78 }
79
80 /*
81  * Re-calculate pending state from the set of locally pending
82  * signals, globally pending signals, and blocked signals.
83  */
84 static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
85 {
86         unsigned long ready;
87         long i;
88
89         switch (_NSIG_WORDS) {
90         default:
91                 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
92                         ready |= signal->sig[i] &~ blocked->sig[i];
93                 break;
94
95         case 4: ready  = signal->sig[3] &~ blocked->sig[3];
96                 ready |= signal->sig[2] &~ blocked->sig[2];
97                 ready |= signal->sig[1] &~ blocked->sig[1];
98                 ready |= signal->sig[0] &~ blocked->sig[0];
99                 break;
100
101         case 2: ready  = signal->sig[1] &~ blocked->sig[1];
102                 ready |= signal->sig[0] &~ blocked->sig[0];
103                 break;
104
105         case 1: ready  = signal->sig[0] &~ blocked->sig[0];
106         }
107         return ready != 0;
108 }
109
110 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
111
112 static int recalc_sigpending_tsk(struct task_struct *t)
113 {
114         if (t->signal->group_stop_count > 0 ||
115             PENDING(&t->pending, &t->blocked) ||
116             PENDING(&t->signal->shared_pending, &t->blocked)) {
117                 set_tsk_thread_flag(t, TIF_SIGPENDING);
118                 return 1;
119         }
120         /*
121          * We must never clear the flag in another thread, or in current
122          * when it's possible the current syscall is returning -ERESTART*.
123          * So we don't clear it here, and only callers who know they should do.
124          */
125         return 0;
126 }
127
128 /*
129  * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
130  * This is superfluous when called on current, the wakeup is a harmless no-op.
131  */
132 void recalc_sigpending_and_wake(struct task_struct *t)
133 {
134         if (recalc_sigpending_tsk(t))
135                 signal_wake_up(t, 0);
136 }
137
138 void recalc_sigpending(void)
139 {
140         if (unlikely(tracehook_force_sigpending()))
141                 set_thread_flag(TIF_SIGPENDING);
142         else if (!recalc_sigpending_tsk(current) && !freezing(current))
143                 clear_thread_flag(TIF_SIGPENDING);
144
145 }
146
147 /* Given the mask, find the first available signal that should be serviced. */
148
149 int next_signal(struct sigpending *pending, sigset_t *mask)
150 {
151         unsigned long i, *s, *m, x;
152         int sig = 0;
153         
154         s = pending->signal.sig;
155         m = mask->sig;
156         switch (_NSIG_WORDS) {
157         default:
158                 for (i = 0; i < _NSIG_WORDS; ++i, ++s, ++m)
159                         if ((x = *s &~ *m) != 0) {
160                                 sig = ffz(~x) + i*_NSIG_BPW + 1;
161                                 break;
162                         }
163                 break;
164
165         case 2: if ((x = s[0] &~ m[0]) != 0)
166                         sig = 1;
167                 else if ((x = s[1] &~ m[1]) != 0)
168                         sig = _NSIG_BPW + 1;
169                 else
170                         break;
171                 sig += ffz(~x);
172                 break;
173
174         case 1: if ((x = *s &~ *m) != 0)
175                         sig = ffz(~x) + 1;
176                 break;
177         }
178         
179         return sig;
180 }
181
182 /*
183  * allocate a new signal queue record
184  * - this may be called without locks if and only if t == current, otherwise an
185  *   appopriate lock must be held to stop the target task from exiting
186  */
187 static struct sigqueue *__sigqueue_alloc(struct task_struct *t, gfp_t flags,
188                                          int override_rlimit)
189 {
190         struct sigqueue *q = NULL;
191         struct user_struct *user;
192
193         /*
194          * We won't get problems with the target's UID changing under us
195          * because changing it requires RCU be used, and if t != current, the
196          * caller must be holding the RCU readlock (by way of a spinlock) and
197          * we use RCU protection here
198          */
199         user = get_uid(__task_cred(t)->user);
200         atomic_inc(&user->sigpending);
201         if (override_rlimit ||
202             atomic_read(&user->sigpending) <=
203                         t->signal->rlim[RLIMIT_SIGPENDING].rlim_cur)
204                 q = kmem_cache_alloc(sigqueue_cachep, flags);
205         if (unlikely(q == NULL)) {
206                 atomic_dec(&user->sigpending);
207                 free_uid(user);
208         } else {
209                 INIT_LIST_HEAD(&q->list);
210                 q->flags = 0;
211                 q->user = user;
212         }
213
214         return q;
215 }
216
217 static void __sigqueue_free(struct sigqueue *q)
218 {
219         if (q->flags & SIGQUEUE_PREALLOC)
220                 return;
221         atomic_dec(&q->user->sigpending);
222         free_uid(q->user);
223         kmem_cache_free(sigqueue_cachep, q);
224 }
225
226 void flush_sigqueue(struct sigpending *queue)
227 {
228         struct sigqueue *q;
229
230         sigemptyset(&queue->signal);
231         while (!list_empty(&queue->list)) {
232                 q = list_entry(queue->list.next, struct sigqueue , list);
233                 list_del_init(&q->list);
234                 __sigqueue_free(q);
235         }
236 }
237
238 /*
239  * Flush all pending signals for a task.
240  */
241 void flush_signals(struct task_struct *t)
242 {
243         unsigned long flags;
244
245         spin_lock_irqsave(&t->sighand->siglock, flags);
246         clear_tsk_thread_flag(t, TIF_SIGPENDING);
247         flush_sigqueue(&t->pending);
248         flush_sigqueue(&t->signal->shared_pending);
249         spin_unlock_irqrestore(&t->sighand->siglock, flags);
250 }
251
252 static void __flush_itimer_signals(struct sigpending *pending)
253 {
254         sigset_t signal, retain;
255         struct sigqueue *q, *n;
256
257         signal = pending->signal;
258         sigemptyset(&retain);
259
260         list_for_each_entry_safe(q, n, &pending->list, list) {
261                 int sig = q->info.si_signo;
262
263                 if (likely(q->info.si_code != SI_TIMER)) {
264                         sigaddset(&retain, sig);
265                 } else {
266                         sigdelset(&signal, sig);
267                         list_del_init(&q->list);
268                         __sigqueue_free(q);
269                 }
270         }
271
272         sigorsets(&pending->signal, &signal, &retain);
273 }
274
275 void flush_itimer_signals(void)
276 {
277         struct task_struct *tsk = current;
278         unsigned long flags;
279
280         spin_lock_irqsave(&tsk->sighand->siglock, flags);
281         __flush_itimer_signals(&tsk->pending);
282         __flush_itimer_signals(&tsk->signal->shared_pending);
283         spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
284 }
285
286 void ignore_signals(struct task_struct *t)
287 {
288         int i;
289
290         for (i = 0; i < _NSIG; ++i)
291                 t->sighand->action[i].sa.sa_handler = SIG_IGN;
292
293         flush_signals(t);
294 }
295
296 /*
297  * Flush all handlers for a task.
298  */
299
300 void
301 flush_signal_handlers(struct task_struct *t, int force_default)
302 {
303         int i;
304         struct k_sigaction *ka = &t->sighand->action[0];
305         for (i = _NSIG ; i != 0 ; i--) {
306                 if (force_default || ka->sa.sa_handler != SIG_IGN)
307                         ka->sa.sa_handler = SIG_DFL;
308                 ka->sa.sa_flags = 0;
309                 sigemptyset(&ka->sa.sa_mask);
310                 ka++;
311         }
312 }
313
314 int unhandled_signal(struct task_struct *tsk, int sig)
315 {
316         void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
317         if (is_global_init(tsk))
318                 return 1;
319         if (handler != SIG_IGN && handler != SIG_DFL)
320                 return 0;
321         return !tracehook_consider_fatal_signal(tsk, sig, handler);
322 }
323
324
325 /* Notify the system that a driver wants to block all signals for this
326  * process, and wants to be notified if any signals at all were to be
327  * sent/acted upon.  If the notifier routine returns non-zero, then the
328  * signal will be acted upon after all.  If the notifier routine returns 0,
329  * then then signal will be blocked.  Only one block per process is
330  * allowed.  priv is a pointer to private data that the notifier routine
331  * can use to determine if the signal should be blocked or not.  */
332
333 void
334 block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
335 {
336         unsigned long flags;
337
338         spin_lock_irqsave(&current->sighand->siglock, flags);
339         current->notifier_mask = mask;
340         current->notifier_data = priv;
341         current->notifier = notifier;
342         spin_unlock_irqrestore(&current->sighand->siglock, flags);
343 }
344
345 /* Notify the system that blocking has ended. */
346
347 void
348 unblock_all_signals(void)
349 {
350         unsigned long flags;
351
352         spin_lock_irqsave(&current->sighand->siglock, flags);
353         current->notifier = NULL;
354         current->notifier_data = NULL;
355         recalc_sigpending();
356         spin_unlock_irqrestore(&current->sighand->siglock, flags);
357 }
358
359 static void collect_signal(int sig, struct sigpending *list, siginfo_t *info)
360 {
361         struct sigqueue *q, *first = NULL;
362
363         /*
364          * Collect the siginfo appropriate to this signal.  Check if
365          * there is another siginfo for the same signal.
366         */
367         list_for_each_entry(q, &list->list, list) {
368                 if (q->info.si_signo == sig) {
369                         if (first)
370                                 goto still_pending;
371                         first = q;
372                 }
373         }
374
375         sigdelset(&list->signal, sig);
376
377         if (first) {
378 still_pending:
379                 list_del_init(&first->list);
380                 copy_siginfo(info, &first->info);
381                 __sigqueue_free(first);
382         } else {
383                 /* Ok, it wasn't in the queue.  This must be
384                    a fast-pathed signal or we must have been
385                    out of queue space.  So zero out the info.
386                  */
387                 info->si_signo = sig;
388                 info->si_errno = 0;
389                 info->si_code = 0;
390                 info->si_pid = 0;
391                 info->si_uid = 0;
392         }
393 }
394
395 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
396                         siginfo_t *info)
397 {
398         int sig = next_signal(pending, mask);
399
400         if (sig) {
401                 if (current->notifier) {
402                         if (sigismember(current->notifier_mask, sig)) {
403                                 if (!(current->notifier)(current->notifier_data)) {
404                                         clear_thread_flag(TIF_SIGPENDING);
405                                         return 0;
406                                 }
407                         }
408                 }
409
410                 collect_signal(sig, pending, info);
411         }
412
413         return sig;
414 }
415
416 /*
417  * Dequeue a signal and return the element to the caller, which is 
418  * expected to free it.
419  *
420  * All callers have to hold the siglock.
421  */
422 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
423 {
424         int signr;
425
426         /* We only dequeue private signals from ourselves, we don't let
427          * signalfd steal them
428          */
429         signr = __dequeue_signal(&tsk->pending, mask, info);
430         if (!signr) {
431                 signr = __dequeue_signal(&tsk->signal->shared_pending,
432                                          mask, info);
433                 /*
434                  * itimer signal ?
435                  *
436                  * itimers are process shared and we restart periodic
437                  * itimers in the signal delivery path to prevent DoS
438                  * attacks in the high resolution timer case. This is
439                  * compliant with the old way of self restarting
440                  * itimers, as the SIGALRM is a legacy signal and only
441                  * queued once. Changing the restart behaviour to
442                  * restart the timer in the signal dequeue path is
443                  * reducing the timer noise on heavy loaded !highres
444                  * systems too.
445                  */
446                 if (unlikely(signr == SIGALRM)) {
447                         struct hrtimer *tmr = &tsk->signal->real_timer;
448
449                         if (!hrtimer_is_queued(tmr) &&
450                             tsk->signal->it_real_incr.tv64 != 0) {
451                                 hrtimer_forward(tmr, tmr->base->get_time(),
452                                                 tsk->signal->it_real_incr);
453                                 hrtimer_restart(tmr);
454                         }
455                 }
456         }
457
458         recalc_sigpending();
459         if (!signr)
460                 return 0;
461
462         if (unlikely(sig_kernel_stop(signr))) {
463                 /*
464                  * Set a marker that we have dequeued a stop signal.  Our
465                  * caller might release the siglock and then the pending
466                  * stop signal it is about to process is no longer in the
467                  * pending bitmasks, but must still be cleared by a SIGCONT
468                  * (and overruled by a SIGKILL).  So those cases clear this
469                  * shared flag after we've set it.  Note that this flag may
470                  * remain set after the signal we return is ignored or
471                  * handled.  That doesn't matter because its only purpose
472                  * is to alert stop-signal processing code when another
473                  * processor has come along and cleared the flag.
474                  */
475                 tsk->signal->flags |= SIGNAL_STOP_DEQUEUED;
476         }
477         if ((info->si_code & __SI_MASK) == __SI_TIMER && info->si_sys_private) {
478                 /*
479                  * Release the siglock to ensure proper locking order
480                  * of timer locks outside of siglocks.  Note, we leave
481                  * irqs disabled here, since the posix-timers code is
482                  * about to disable them again anyway.
483                  */
484                 spin_unlock(&tsk->sighand->siglock);
485                 do_schedule_next_timer(info);
486                 spin_lock(&tsk->sighand->siglock);
487         }
488         return signr;
489 }
490
491 /*
492  * Tell a process that it has a new active signal..
493  *
494  * NOTE! we rely on the previous spin_lock to
495  * lock interrupts for us! We can only be called with
496  * "siglock" held, and the local interrupt must
497  * have been disabled when that got acquired!
498  *
499  * No need to set need_resched since signal event passing
500  * goes through ->blocked
501  */
502 void signal_wake_up(struct task_struct *t, int resume)
503 {
504         unsigned int mask;
505
506         set_tsk_thread_flag(t, TIF_SIGPENDING);
507
508         /*
509          * For SIGKILL, we want to wake it up in the stopped/traced/killable
510          * case. We don't check t->state here because there is a race with it
511          * executing another processor and just now entering stopped state.
512          * By using wake_up_state, we ensure the process will wake up and
513          * handle its death signal.
514          */
515         mask = TASK_INTERRUPTIBLE;
516         if (resume)
517                 mask |= TASK_WAKEKILL;
518         if (!wake_up_state(t, mask))
519                 kick_process(t);
520 }
521
522 /*
523  * Remove signals in mask from the pending set and queue.
524  * Returns 1 if any signals were found.
525  *
526  * All callers must be holding the siglock.
527  *
528  * This version takes a sigset mask and looks at all signals,
529  * not just those in the first mask word.
530  */
531 static int rm_from_queue_full(sigset_t *mask, struct sigpending *s)
532 {
533         struct sigqueue *q, *n;
534         sigset_t m;
535
536         sigandsets(&m, mask, &s->signal);
537         if (sigisemptyset(&m))
538                 return 0;
539
540         signandsets(&s->signal, &s->signal, mask);
541         list_for_each_entry_safe(q, n, &s->list, list) {
542                 if (sigismember(mask, q->info.si_signo)) {
543                         list_del_init(&q->list);
544                         __sigqueue_free(q);
545                 }
546         }
547         return 1;
548 }
549 /*
550  * Remove signals in mask from the pending set and queue.
551  * Returns 1 if any signals were found.
552  *
553  * All callers must be holding the siglock.
554  */
555 static int rm_from_queue(unsigned long mask, struct sigpending *s)
556 {
557         struct sigqueue *q, *n;
558
559         if (!sigtestsetmask(&s->signal, mask))
560                 return 0;
561
562         sigdelsetmask(&s->signal, mask);
563         list_for_each_entry_safe(q, n, &s->list, list) {
564                 if (q->info.si_signo < SIGRTMIN &&
565                     (mask & sigmask(q->info.si_signo))) {
566                         list_del_init(&q->list);
567                         __sigqueue_free(q);
568                 }
569         }
570         return 1;
571 }
572
573 /*
574  * Bad permissions for sending the signal
575  * - the caller must hold at least the RCU read lock
576  */
577 static int check_kill_permission(int sig, struct siginfo *info,
578                                  struct task_struct *t)
579 {
580         const struct cred *cred = current_cred(), *tcred;
581         struct pid *sid;
582         int error;
583
584         if (!valid_signal(sig))
585                 return -EINVAL;
586
587         if (info != SEND_SIG_NOINFO && (is_si_special(info) || SI_FROMKERNEL(info)))
588                 return 0;
589
590         error = audit_signal_info(sig, t); /* Let audit system see the signal */
591         if (error)
592                 return error;
593
594         tcred = __task_cred(t);
595         if ((cred->euid ^ tcred->suid) &&
596             (cred->euid ^ tcred->uid) &&
597             (cred->uid  ^ tcred->suid) &&
598             (cred->uid  ^ tcred->uid) &&
599             !capable(CAP_KILL)) {
600                 switch (sig) {
601                 case SIGCONT:
602                         sid = task_session(t);
603                         /*
604                          * We don't return the error if sid == NULL. The
605                          * task was unhashed, the caller must notice this.
606                          */
607                         if (!sid || sid == task_session(current))
608                                 break;
609                 default:
610                         return -EPERM;
611                 }
612         }
613
614         return security_task_kill(t, info, sig, 0);
615 }
616
617 /*
618  * Handle magic process-wide effects of stop/continue signals. Unlike
619  * the signal actions, these happen immediately at signal-generation
620  * time regardless of blocking, ignoring, or handling.  This does the
621  * actual continuing for SIGCONT, but not the actual stopping for stop
622  * signals. The process stop is done as a signal action for SIG_DFL.
623  *
624  * Returns true if the signal should be actually delivered, otherwise
625  * it should be dropped.
626  */
627 static int prepare_signal(int sig, struct task_struct *p)
628 {
629         struct signal_struct *signal = p->signal;
630         struct task_struct *t;
631
632         if (unlikely(signal->flags & SIGNAL_GROUP_EXIT)) {
633                 /*
634                  * The process is in the middle of dying, nothing to do.
635                  */
636         } else if (sig_kernel_stop(sig)) {
637                 /*
638                  * This is a stop signal.  Remove SIGCONT from all queues.
639                  */
640                 rm_from_queue(sigmask(SIGCONT), &signal->shared_pending);
641                 t = p;
642                 do {
643                         rm_from_queue(sigmask(SIGCONT), &t->pending);
644                 } while_each_thread(p, t);
645         } else if (sig == SIGCONT) {
646                 unsigned int why;
647                 /*
648                  * Remove all stop signals from all queues,
649                  * and wake all threads.
650                  */
651                 rm_from_queue(SIG_KERNEL_STOP_MASK, &signal->shared_pending);
652                 t = p;
653                 do {
654                         unsigned int state;
655                         rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
656                         /*
657                          * If there is a handler for SIGCONT, we must make
658                          * sure that no thread returns to user mode before
659                          * we post the signal, in case it was the only
660                          * thread eligible to run the signal handler--then
661                          * it must not do anything between resuming and
662                          * running the handler.  With the TIF_SIGPENDING
663                          * flag set, the thread will pause and acquire the
664                          * siglock that we hold now and until we've queued
665                          * the pending signal.
666                          *
667                          * Wake up the stopped thread _after_ setting
668                          * TIF_SIGPENDING
669                          */
670                         state = __TASK_STOPPED;
671                         if (sig_user_defined(t, SIGCONT) && !sigismember(&t->blocked, SIGCONT)) {
672                                 set_tsk_thread_flag(t, TIF_SIGPENDING);
673                                 state |= TASK_INTERRUPTIBLE;
674                         }
675                         wake_up_state(t, state);
676                 } while_each_thread(p, t);
677
678                 /*
679                  * Notify the parent with CLD_CONTINUED if we were stopped.
680                  *
681                  * If we were in the middle of a group stop, we pretend it
682                  * was already finished, and then continued. Since SIGCHLD
683                  * doesn't queue we report only CLD_STOPPED, as if the next
684                  * CLD_CONTINUED was dropped.
685                  */
686                 why = 0;
687                 if (signal->flags & SIGNAL_STOP_STOPPED)
688                         why |= SIGNAL_CLD_CONTINUED;
689                 else if (signal->group_stop_count)
690                         why |= SIGNAL_CLD_STOPPED;
691
692                 if (why) {
693                         /*
694                          * The first thread which returns from finish_stop()
695                          * will take ->siglock, notice SIGNAL_CLD_MASK, and
696                          * notify its parent. See get_signal_to_deliver().
697                          */
698                         signal->flags = why | SIGNAL_STOP_CONTINUED;
699                         signal->group_stop_count = 0;
700                         signal->group_exit_code = 0;
701                 } else {
702                         /*
703                          * We are not stopped, but there could be a stop
704                          * signal in the middle of being processed after
705                          * being removed from the queue.  Clear that too.
706                          */
707                         signal->flags &= ~SIGNAL_STOP_DEQUEUED;
708                 }
709         }
710
711         return !sig_ignored(p, sig);
712 }
713
714 /*
715  * Test if P wants to take SIG.  After we've checked all threads with this,
716  * it's equivalent to finding no threads not blocking SIG.  Any threads not
717  * blocking SIG were ruled out because they are not running and already
718  * have pending signals.  Such threads will dequeue from the shared queue
719  * as soon as they're available, so putting the signal on the shared queue
720  * will be equivalent to sending it to one such thread.
721  */
722 static inline int wants_signal(int sig, struct task_struct *p)
723 {
724         if (sigismember(&p->blocked, sig))
725                 return 0;
726         if (p->flags & PF_EXITING)
727                 return 0;
728         if (sig == SIGKILL)
729                 return 1;
730         if (task_is_stopped_or_traced(p))
731                 return 0;
732         return task_curr(p) || !signal_pending(p);
733 }
734
735 static void complete_signal(int sig, struct task_struct *p, int group)
736 {
737         struct signal_struct *signal = p->signal;
738         struct task_struct *t;
739
740         /*
741          * Now find a thread we can wake up to take the signal off the queue.
742          *
743          * If the main thread wants the signal, it gets first crack.
744          * Probably the least surprising to the average bear.
745          */
746         if (wants_signal(sig, p))
747                 t = p;
748         else if (!group || thread_group_empty(p))
749                 /*
750                  * There is just one thread and it does not need to be woken.
751                  * It will dequeue unblocked signals before it runs again.
752                  */
753                 return;
754         else {
755                 /*
756                  * Otherwise try to find a suitable thread.
757                  */
758                 t = signal->curr_target;
759                 while (!wants_signal(sig, t)) {
760                         t = next_thread(t);
761                         if (t == signal->curr_target)
762                                 /*
763                                  * No thread needs to be woken.
764                                  * Any eligible threads will see
765                                  * the signal in the queue soon.
766                                  */
767                                 return;
768                 }
769                 signal->curr_target = t;
770         }
771
772         /*
773          * Found a killable thread.  If the signal will be fatal,
774          * then start taking the whole group down immediately.
775          */
776         if (sig_fatal(p, sig) &&
777             !(signal->flags & (SIGNAL_UNKILLABLE | SIGNAL_GROUP_EXIT)) &&
778             !sigismember(&t->real_blocked, sig) &&
779             (sig == SIGKILL ||
780              !tracehook_consider_fatal_signal(t, sig, SIG_DFL))) {
781                 /*
782                  * This signal will be fatal to the whole group.
783                  */
784                 if (!sig_kernel_coredump(sig)) {
785                         /*
786                          * Start a group exit and wake everybody up.
787                          * This way we don't have other threads
788                          * running and doing things after a slower
789                          * thread has the fatal signal pending.
790                          */
791                         signal->flags = SIGNAL_GROUP_EXIT;
792                         signal->group_exit_code = sig;
793                         signal->group_stop_count = 0;
794                         t = p;
795                         do {
796                                 sigaddset(&t->pending.signal, SIGKILL);
797                                 signal_wake_up(t, 1);
798                         } while_each_thread(p, t);
799                         return;
800                 }
801         }
802
803         /*
804          * The signal is already in the shared-pending queue.
805          * Tell the chosen thread to wake up and dequeue it.
806          */
807         signal_wake_up(t, sig == SIGKILL);
808         return;
809 }
810
811 static inline int legacy_queue(struct sigpending *signals, int sig)
812 {
813         return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
814 }
815
816 static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
817                         int group)
818 {
819         struct sigpending *pending;
820         struct sigqueue *q;
821
822         trace_sched_signal_send(sig, t);
823
824         assert_spin_locked(&t->sighand->siglock);
825         if (!prepare_signal(sig, t))
826                 return 0;
827
828         pending = group ? &t->signal->shared_pending : &t->pending;
829         /*
830          * Short-circuit ignored signals and support queuing
831          * exactly one non-rt signal, so that we can get more
832          * detailed information about the cause of the signal.
833          */
834         if (legacy_queue(pending, sig))
835                 return 0;
836         /*
837          * fast-pathed signals for kernel-internal things like SIGSTOP
838          * or SIGKILL.
839          */
840         if (info == SEND_SIG_FORCED)
841                 goto out_set;
842
843         /* Real-time signals must be queued if sent by sigqueue, or
844            some other real-time mechanism.  It is implementation
845            defined whether kill() does so.  We attempt to do so, on
846            the principle of least surprise, but since kill is not
847            allowed to fail with EAGAIN when low on memory we just
848            make sure at least one signal gets delivered and don't
849            pass on the info struct.  */
850
851         q = __sigqueue_alloc(t, GFP_ATOMIC, (sig < SIGRTMIN &&
852                                              (is_si_special(info) ||
853                                               info->si_code >= 0)));
854         if (q) {
855                 list_add_tail(&q->list, &pending->list);
856                 switch ((unsigned long) info) {
857                 case (unsigned long) SEND_SIG_NOINFO:
858                         q->info.si_signo = sig;
859                         q->info.si_errno = 0;
860                         q->info.si_code = SI_USER;
861                         q->info.si_pid = task_tgid_nr_ns(current,
862                                                         task_active_pid_ns(t));
863                         q->info.si_uid = current_uid();
864                         break;
865                 case (unsigned long) SEND_SIG_PRIV:
866                         q->info.si_signo = sig;
867                         q->info.si_errno = 0;
868                         q->info.si_code = SI_KERNEL;
869                         q->info.si_pid = 0;
870                         q->info.si_uid = 0;
871                         break;
872                 default:
873                         copy_siginfo(&q->info, info);
874                         break;
875                 }
876         } else if (!is_si_special(info)) {
877                 if (sig >= SIGRTMIN && info->si_code != SI_USER)
878                 /*
879                  * Queue overflow, abort.  We may abort if the signal was rt
880                  * and sent by user using something other than kill().
881                  */
882                         return -EAGAIN;
883         }
884
885 out_set:
886         signalfd_notify(t, sig);
887         sigaddset(&pending->signal, sig);
888         complete_signal(sig, t, group);
889         return 0;
890 }
891
892 int print_fatal_signals;
893
894 static void print_fatal_signal(struct pt_regs *regs, int signr)
895 {
896         printk("%s/%d: potentially unexpected fatal signal %d.\n",
897                 current->comm, task_pid_nr(current), signr);
898
899 #if defined(__i386__) && !defined(__arch_um__)
900         printk("code at %08lx: ", regs->ip);
901         {
902                 int i;
903                 for (i = 0; i < 16; i++) {
904                         unsigned char insn;
905
906                         __get_user(insn, (unsigned char *)(regs->ip + i));
907                         printk("%02x ", insn);
908                 }
909         }
910 #endif
911         printk("\n");
912         preempt_disable();
913         show_regs(regs);
914         preempt_enable();
915 }
916
917 static int __init setup_print_fatal_signals(char *str)
918 {
919         get_option (&str, &print_fatal_signals);
920
921         return 1;
922 }
923
924 __setup("print-fatal-signals=", setup_print_fatal_signals);
925
926 int
927 __group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
928 {
929         return send_signal(sig, info, p, 1);
930 }
931
932 static int
933 specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
934 {
935         return send_signal(sig, info, t, 0);
936 }
937
938 /*
939  * Force a signal that the process can't ignore: if necessary
940  * we unblock the signal and change any SIG_IGN to SIG_DFL.
941  *
942  * Note: If we unblock the signal, we always reset it to SIG_DFL,
943  * since we do not want to have a signal handler that was blocked
944  * be invoked when user space had explicitly blocked it.
945  *
946  * We don't want to have recursive SIGSEGV's etc, for example,
947  * that is why we also clear SIGNAL_UNKILLABLE.
948  */
949 int
950 force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
951 {
952         unsigned long int flags;
953         int ret, blocked, ignored;
954         struct k_sigaction *action;
955
956         spin_lock_irqsave(&t->sighand->siglock, flags);
957         action = &t->sighand->action[sig-1];
958         ignored = action->sa.sa_handler == SIG_IGN;
959         blocked = sigismember(&t->blocked, sig);
960         if (blocked || ignored) {
961                 action->sa.sa_handler = SIG_DFL;
962                 if (blocked) {
963                         sigdelset(&t->blocked, sig);
964                         recalc_sigpending_and_wake(t);
965                 }
966         }
967         if (action->sa.sa_handler == SIG_DFL)
968                 t->signal->flags &= ~SIGNAL_UNKILLABLE;
969         ret = specific_send_sig_info(sig, info, t);
970         spin_unlock_irqrestore(&t->sighand->siglock, flags);
971
972         return ret;
973 }
974
975 void
976 force_sig_specific(int sig, struct task_struct *t)
977 {
978         force_sig_info(sig, SEND_SIG_FORCED, t);
979 }
980
981 /*
982  * Nuke all other threads in the group.
983  */
984 void zap_other_threads(struct task_struct *p)
985 {
986         struct task_struct *t;
987
988         p->signal->group_stop_count = 0;
989
990         for (t = next_thread(p); t != p; t = next_thread(t)) {
991                 /*
992                  * Don't bother with already dead threads
993                  */
994                 if (t->exit_state)
995                         continue;
996
997                 /* SIGKILL will be handled before any pending SIGSTOP */
998                 sigaddset(&t->pending.signal, SIGKILL);
999                 signal_wake_up(t, 1);
1000         }
1001 }
1002
1003 int __fatal_signal_pending(struct task_struct *tsk)
1004 {
1005         return sigismember(&tsk->pending.signal, SIGKILL);
1006 }
1007 EXPORT_SYMBOL(__fatal_signal_pending);
1008
1009 struct sighand_struct *lock_task_sighand(struct task_struct *tsk, unsigned long *flags)
1010 {
1011         struct sighand_struct *sighand;
1012
1013         rcu_read_lock();
1014         for (;;) {
1015                 sighand = rcu_dereference(tsk->sighand);
1016                 if (unlikely(sighand == NULL))
1017                         break;
1018
1019                 spin_lock_irqsave(&sighand->siglock, *flags);
1020                 if (likely(sighand == tsk->sighand))
1021                         break;
1022                 spin_unlock_irqrestore(&sighand->siglock, *flags);
1023         }
1024         rcu_read_unlock();
1025
1026         return sighand;
1027 }
1028
1029 /*
1030  * send signal info to all the members of a group
1031  * - the caller must hold the RCU read lock at least
1032  */
1033 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1034 {
1035         unsigned long flags;
1036         int ret;
1037
1038         ret = check_kill_permission(sig, info, p);
1039
1040         if (!ret && sig) {
1041                 ret = -ESRCH;
1042                 if (lock_task_sighand(p, &flags)) {
1043                         ret = __group_send_sig_info(sig, info, p);
1044                         unlock_task_sighand(p, &flags);
1045                 }
1046         }
1047
1048         return ret;
1049 }
1050
1051 /*
1052  * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1053  * control characters do (^C, ^Z etc)
1054  * - the caller must hold at least a readlock on tasklist_lock
1055  */
1056 int __kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
1057 {
1058         struct task_struct *p = NULL;
1059         int retval, success;
1060
1061         success = 0;
1062         retval = -ESRCH;
1063         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1064                 int err = group_send_sig_info(sig, info, p);
1065                 success |= !err;
1066                 retval = err;
1067         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1068         return success ? 0 : retval;
1069 }
1070
1071 int kill_pid_info(int sig, struct siginfo *info, struct pid *pid)
1072 {
1073         int error = -ESRCH;
1074         struct task_struct *p;
1075
1076         rcu_read_lock();
1077 retry:
1078         p = pid_task(pid, PIDTYPE_PID);
1079         if (p) {
1080                 error = group_send_sig_info(sig, info, p);
1081                 if (unlikely(error == -ESRCH))
1082                         /*
1083                          * The task was unhashed in between, try again.
1084                          * If it is dead, pid_task() will return NULL,
1085                          * if we race with de_thread() it will find the
1086                          * new leader.
1087                          */
1088                         goto retry;
1089         }
1090         rcu_read_unlock();
1091
1092         return error;
1093 }
1094
1095 int
1096 kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1097 {
1098         int error;
1099         rcu_read_lock();
1100         error = kill_pid_info(sig, info, find_vpid(pid));
1101         rcu_read_unlock();
1102         return error;
1103 }
1104
1105 /* like kill_pid_info(), but doesn't use uid/euid of "current" */
1106 int kill_pid_info_as_uid(int sig, struct siginfo *info, struct pid *pid,
1107                       uid_t uid, uid_t euid, u32 secid)
1108 {
1109         int ret = -EINVAL;
1110         struct task_struct *p;
1111         const struct cred *pcred;
1112
1113         if (!valid_signal(sig))
1114                 return ret;
1115
1116         read_lock(&tasklist_lock);
1117         p = pid_task(pid, PIDTYPE_PID);
1118         if (!p) {
1119                 ret = -ESRCH;
1120                 goto out_unlock;
1121         }
1122         pcred = __task_cred(p);
1123         if ((info == SEND_SIG_NOINFO ||
1124              (!is_si_special(info) && SI_FROMUSER(info))) &&
1125             euid != pcred->suid && euid != pcred->uid &&
1126             uid  != pcred->suid && uid  != pcred->uid) {
1127                 ret = -EPERM;
1128                 goto out_unlock;
1129         }
1130         ret = security_task_kill(p, info, sig, secid);
1131         if (ret)
1132                 goto out_unlock;
1133         if (sig && p->sighand) {
1134                 unsigned long flags;
1135                 spin_lock_irqsave(&p->sighand->siglock, flags);
1136                 ret = __group_send_sig_info(sig, info, p);
1137                 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1138         }
1139 out_unlock:
1140         read_unlock(&tasklist_lock);
1141         return ret;
1142 }
1143 EXPORT_SYMBOL_GPL(kill_pid_info_as_uid);
1144
1145 /*
1146  * kill_something_info() interprets pid in interesting ways just like kill(2).
1147  *
1148  * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1149  * is probably wrong.  Should make it like BSD or SYSV.
1150  */
1151
1152 static int kill_something_info(int sig, struct siginfo *info, pid_t pid)
1153 {
1154         int ret;
1155
1156         if (pid > 0) {
1157                 rcu_read_lock();
1158                 ret = kill_pid_info(sig, info, find_vpid(pid));
1159                 rcu_read_unlock();
1160                 return ret;
1161         }
1162
1163         read_lock(&tasklist_lock);
1164         if (pid != -1) {
1165                 ret = __kill_pgrp_info(sig, info,
1166                                 pid ? find_vpid(-pid) : task_pgrp(current));
1167         } else {
1168                 int retval = 0, count = 0;
1169                 struct task_struct * p;
1170
1171                 for_each_process(p) {
1172                         if (task_pid_vnr(p) > 1 &&
1173                                         !same_thread_group(p, current)) {
1174                                 int err = group_send_sig_info(sig, info, p);
1175                                 ++count;
1176                                 if (err != -EPERM)
1177                                         retval = err;
1178                         }
1179                 }
1180                 ret = count ? retval : -ESRCH;
1181         }
1182         read_unlock(&tasklist_lock);
1183
1184         return ret;
1185 }
1186
1187 /*
1188  * These are for backward compatibility with the rest of the kernel source.
1189  */
1190
1191 /*
1192  * The caller must ensure the task can't exit.
1193  */
1194 int
1195 send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1196 {
1197         int ret;
1198         unsigned long flags;
1199
1200         /*
1201          * Make sure legacy kernel users don't send in bad values
1202          * (normal paths check this in check_kill_permission).
1203          */
1204         if (!valid_signal(sig))
1205                 return -EINVAL;
1206
1207         spin_lock_irqsave(&p->sighand->siglock, flags);
1208         ret = specific_send_sig_info(sig, info, p);
1209         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1210         return ret;
1211 }
1212
1213 #define __si_special(priv) \
1214         ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1215
1216 int
1217 send_sig(int sig, struct task_struct *p, int priv)
1218 {
1219         return send_sig_info(sig, __si_special(priv), p);
1220 }
1221
1222 void
1223 force_sig(int sig, struct task_struct *p)
1224 {
1225         force_sig_info(sig, SEND_SIG_PRIV, p);
1226 }
1227
1228 /*
1229  * When things go south during signal handling, we
1230  * will force a SIGSEGV. And if the signal that caused
1231  * the problem was already a SIGSEGV, we'll want to
1232  * make sure we don't even try to deliver the signal..
1233  */
1234 int
1235 force_sigsegv(int sig, struct task_struct *p)
1236 {
1237         if (sig == SIGSEGV) {
1238                 unsigned long flags;
1239                 spin_lock_irqsave(&p->sighand->siglock, flags);
1240                 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1241                 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1242         }
1243         force_sig(SIGSEGV, p);
1244         return 0;
1245 }
1246
1247 int kill_pgrp(struct pid *pid, int sig, int priv)
1248 {
1249         int ret;
1250
1251         read_lock(&tasklist_lock);
1252         ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1253         read_unlock(&tasklist_lock);
1254
1255         return ret;
1256 }
1257 EXPORT_SYMBOL(kill_pgrp);
1258
1259 int kill_pid(struct pid *pid, int sig, int priv)
1260 {
1261         return kill_pid_info(sig, __si_special(priv), pid);
1262 }
1263 EXPORT_SYMBOL(kill_pid);
1264
1265 /*
1266  * These functions support sending signals using preallocated sigqueue
1267  * structures.  This is needed "because realtime applications cannot
1268  * afford to lose notifications of asynchronous events, like timer
1269  * expirations or I/O completions".  In the case of Posix Timers 
1270  * we allocate the sigqueue structure from the timer_create.  If this
1271  * allocation fails we are able to report the failure to the application
1272  * with an EAGAIN error.
1273  */
1274  
1275 struct sigqueue *sigqueue_alloc(void)
1276 {
1277         struct sigqueue *q;
1278
1279         if ((q = __sigqueue_alloc(current, GFP_KERNEL, 0)))
1280                 q->flags |= SIGQUEUE_PREALLOC;
1281         return(q);
1282 }
1283
1284 void sigqueue_free(struct sigqueue *q)
1285 {
1286         unsigned long flags;
1287         spinlock_t *lock = &current->sighand->siglock;
1288
1289         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1290         /*
1291          * We must hold ->siglock while testing q->list
1292          * to serialize with collect_signal() or with
1293          * __exit_signal()->flush_sigqueue().
1294          */
1295         spin_lock_irqsave(lock, flags);
1296         q->flags &= ~SIGQUEUE_PREALLOC;
1297         /*
1298          * If it is queued it will be freed when dequeued,
1299          * like the "regular" sigqueue.
1300          */
1301         if (!list_empty(&q->list))
1302                 q = NULL;
1303         spin_unlock_irqrestore(lock, flags);
1304
1305         if (q)
1306                 __sigqueue_free(q);
1307 }
1308
1309 int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group)
1310 {
1311         int sig = q->info.si_signo;
1312         struct sigpending *pending;
1313         unsigned long flags;
1314         int ret;
1315
1316         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1317
1318         ret = -1;
1319         if (!likely(lock_task_sighand(t, &flags)))
1320                 goto ret;
1321
1322         ret = 1; /* the signal is ignored */
1323         if (!prepare_signal(sig, t))
1324                 goto out;
1325
1326         ret = 0;
1327         if (unlikely(!list_empty(&q->list))) {
1328                 /*
1329                  * If an SI_TIMER entry is already queue just increment
1330                  * the overrun count.
1331                  */
1332                 BUG_ON(q->info.si_code != SI_TIMER);
1333                 q->info.si_overrun++;
1334                 goto out;
1335         }
1336         q->info.si_overrun = 0;
1337
1338         signalfd_notify(t, sig);
1339         pending = group ? &t->signal->shared_pending : &t->pending;
1340         list_add_tail(&q->list, &pending->list);
1341         sigaddset(&pending->signal, sig);
1342         complete_signal(sig, t, group);
1343 out:
1344         unlock_task_sighand(t, &flags);
1345 ret:
1346         return ret;
1347 }
1348
1349 /*
1350  * Wake up any threads in the parent blocked in wait* syscalls.
1351  */
1352 static inline void __wake_up_parent(struct task_struct *p,
1353                                     struct task_struct *parent)
1354 {
1355         wake_up_interruptible_sync(&parent->signal->wait_chldexit);
1356 }
1357
1358 /*
1359  * Let a parent know about the death of a child.
1360  * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1361  *
1362  * Returns -1 if our parent ignored us and so we've switched to
1363  * self-reaping, or else @sig.
1364  */
1365 int do_notify_parent(struct task_struct *tsk, int sig)
1366 {
1367         struct siginfo info;
1368         unsigned long flags;
1369         struct sighand_struct *psig;
1370         int ret = sig;
1371
1372         BUG_ON(sig == -1);
1373
1374         /* do_notify_parent_cldstop should have been called instead.  */
1375         BUG_ON(task_is_stopped_or_traced(tsk));
1376
1377         BUG_ON(!tsk->ptrace &&
1378                (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1379
1380         info.si_signo = sig;
1381         info.si_errno = 0;
1382         /*
1383          * we are under tasklist_lock here so our parent is tied to
1384          * us and cannot exit and release its namespace.
1385          *
1386          * the only it can is to switch its nsproxy with sys_unshare,
1387          * bu uncharing pid namespaces is not allowed, so we'll always
1388          * see relevant namespace
1389          *
1390          * write_lock() currently calls preempt_disable() which is the
1391          * same as rcu_read_lock(), but according to Oleg, this is not
1392          * correct to rely on this
1393          */
1394         rcu_read_lock();
1395         info.si_pid = task_pid_nr_ns(tsk, tsk->parent->nsproxy->pid_ns);
1396         info.si_uid = __task_cred(tsk)->uid;
1397         rcu_read_unlock();
1398
1399         info.si_utime = cputime_to_clock_t(cputime_add(tsk->utime,
1400                                 tsk->signal->utime));
1401         info.si_stime = cputime_to_clock_t(cputime_add(tsk->stime,
1402                                 tsk->signal->stime));
1403
1404         info.si_status = tsk->exit_code & 0x7f;
1405         if (tsk->exit_code & 0x80)
1406                 info.si_code = CLD_DUMPED;
1407         else if (tsk->exit_code & 0x7f)
1408                 info.si_code = CLD_KILLED;
1409         else {
1410                 info.si_code = CLD_EXITED;
1411                 info.si_status = tsk->exit_code >> 8;
1412         }
1413
1414         psig = tsk->parent->sighand;
1415         spin_lock_irqsave(&psig->siglock, flags);
1416         if (!tsk->ptrace && sig == SIGCHLD &&
1417             (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1418              (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1419                 /*
1420                  * We are exiting and our parent doesn't care.  POSIX.1
1421                  * defines special semantics for setting SIGCHLD to SIG_IGN
1422                  * or setting the SA_NOCLDWAIT flag: we should be reaped
1423                  * automatically and not left for our parent's wait4 call.
1424                  * Rather than having the parent do it as a magic kind of
1425                  * signal handler, we just set this to tell do_exit that we
1426                  * can be cleaned up without becoming a zombie.  Note that
1427                  * we still call __wake_up_parent in this case, because a
1428                  * blocked sys_wait4 might now return -ECHILD.
1429                  *
1430                  * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1431                  * is implementation-defined: we do (if you don't want
1432                  * it, just use SIG_IGN instead).
1433                  */
1434                 ret = tsk->exit_signal = -1;
1435                 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1436                         sig = -1;
1437         }
1438         if (valid_signal(sig) && sig > 0)
1439                 __group_send_sig_info(sig, &info, tsk->parent);
1440         __wake_up_parent(tsk, tsk->parent);
1441         spin_unlock_irqrestore(&psig->siglock, flags);
1442
1443         return ret;
1444 }
1445
1446 static void do_notify_parent_cldstop(struct task_struct *tsk, int why)
1447 {
1448         struct siginfo info;
1449         unsigned long flags;
1450         struct task_struct *parent;
1451         struct sighand_struct *sighand;
1452
1453         if (tsk->ptrace & PT_PTRACED)
1454                 parent = tsk->parent;
1455         else {
1456                 tsk = tsk->group_leader;
1457                 parent = tsk->real_parent;
1458         }
1459
1460         info.si_signo = SIGCHLD;
1461         info.si_errno = 0;
1462         /*
1463          * see comment in do_notify_parent() abot the following 3 lines
1464          */
1465         rcu_read_lock();
1466         info.si_pid = task_pid_nr_ns(tsk, tsk->parent->nsproxy->pid_ns);
1467         info.si_uid = __task_cred(tsk)->uid;
1468         rcu_read_unlock();
1469
1470         info.si_utime = cputime_to_clock_t(tsk->utime);
1471         info.si_stime = cputime_to_clock_t(tsk->stime);
1472
1473         info.si_code = why;
1474         switch (why) {
1475         case CLD_CONTINUED:
1476                 info.si_status = SIGCONT;
1477                 break;
1478         case CLD_STOPPED:
1479                 info.si_status = tsk->signal->group_exit_code & 0x7f;
1480                 break;
1481         case CLD_TRAPPED:
1482                 info.si_status = tsk->exit_code & 0x7f;
1483                 break;
1484         default:
1485                 BUG();
1486         }
1487
1488         sighand = parent->sighand;
1489         spin_lock_irqsave(&sighand->siglock, flags);
1490         if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1491             !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1492                 __group_send_sig_info(SIGCHLD, &info, parent);
1493         /*
1494          * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1495          */
1496         __wake_up_parent(tsk, parent);
1497         spin_unlock_irqrestore(&sighand->siglock, flags);
1498 }
1499
1500 static inline int may_ptrace_stop(void)
1501 {
1502         if (!likely(current->ptrace & PT_PTRACED))
1503                 return 0;
1504         /*
1505          * Are we in the middle of do_coredump?
1506          * If so and our tracer is also part of the coredump stopping
1507          * is a deadlock situation, and pointless because our tracer
1508          * is dead so don't allow us to stop.
1509          * If SIGKILL was already sent before the caller unlocked
1510          * ->siglock we must see ->core_state != NULL. Otherwise it
1511          * is safe to enter schedule().
1512          */
1513         if (unlikely(current->mm->core_state) &&
1514             unlikely(current->mm == current->parent->mm))
1515                 return 0;
1516
1517         return 1;
1518 }
1519
1520 /*
1521  * Return nonzero if there is a SIGKILL that should be waking us up.
1522  * Called with the siglock held.
1523  */
1524 static int sigkill_pending(struct task_struct *tsk)
1525 {
1526         return  sigismember(&tsk->pending.signal, SIGKILL) ||
1527                 sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
1528 }
1529
1530 /*
1531  * This must be called with current->sighand->siglock held.
1532  *
1533  * This should be the path for all ptrace stops.
1534  * We always set current->last_siginfo while stopped here.
1535  * That makes it a way to test a stopped process for
1536  * being ptrace-stopped vs being job-control-stopped.
1537  *
1538  * If we actually decide not to stop at all because the tracer
1539  * is gone, we keep current->exit_code unless clear_code.
1540  */
1541 static void ptrace_stop(int exit_code, int clear_code, siginfo_t *info)
1542 {
1543         if (arch_ptrace_stop_needed(exit_code, info)) {
1544                 /*
1545                  * The arch code has something special to do before a
1546                  * ptrace stop.  This is allowed to block, e.g. for faults
1547                  * on user stack pages.  We can't keep the siglock while
1548                  * calling arch_ptrace_stop, so we must release it now.
1549                  * To preserve proper semantics, we must do this before
1550                  * any signal bookkeeping like checking group_stop_count.
1551                  * Meanwhile, a SIGKILL could come in before we retake the
1552                  * siglock.  That must prevent us from sleeping in TASK_TRACED.
1553                  * So after regaining the lock, we must check for SIGKILL.
1554                  */
1555                 spin_unlock_irq(&current->sighand->siglock);
1556                 arch_ptrace_stop(exit_code, info);
1557                 spin_lock_irq(&current->sighand->siglock);
1558                 if (sigkill_pending(current))
1559                         return;
1560         }
1561
1562         /*
1563          * If there is a group stop in progress,
1564          * we must participate in the bookkeeping.
1565          */
1566         if (current->signal->group_stop_count > 0)
1567                 --current->signal->group_stop_count;
1568
1569         current->last_siginfo = info;
1570         current->exit_code = exit_code;
1571
1572         /* Let the debugger run.  */
1573         __set_current_state(TASK_TRACED);
1574         spin_unlock_irq(&current->sighand->siglock);
1575         read_lock(&tasklist_lock);
1576         if (may_ptrace_stop()) {
1577                 do_notify_parent_cldstop(current, CLD_TRAPPED);
1578                 /*
1579                  * Don't want to allow preemption here, because
1580                  * sys_ptrace() needs this task to be inactive.
1581                  *
1582                  * XXX: implement read_unlock_no_resched().
1583                  */
1584                 preempt_disable();
1585                 read_unlock(&tasklist_lock);
1586                 preempt_enable_no_resched();
1587                 schedule();
1588         } else {
1589                 /*
1590                  * By the time we got the lock, our tracer went away.
1591                  * Don't drop the lock yet, another tracer may come.
1592                  */
1593                 __set_current_state(TASK_RUNNING);
1594                 if (clear_code)
1595                         current->exit_code = 0;
1596                 read_unlock(&tasklist_lock);
1597         }
1598
1599         /*
1600          * While in TASK_TRACED, we were considered "frozen enough".
1601          * Now that we woke up, it's crucial if we're supposed to be
1602          * frozen that we freeze now before running anything substantial.
1603          */
1604         try_to_freeze();
1605
1606         /*
1607          * We are back.  Now reacquire the siglock before touching
1608          * last_siginfo, so that we are sure to have synchronized with
1609          * any signal-sending on another CPU that wants to examine it.
1610          */
1611         spin_lock_irq(&current->sighand->siglock);
1612         current->last_siginfo = NULL;
1613
1614         /*
1615          * Queued signals ignored us while we were stopped for tracing.
1616          * So check for any that we should take before resuming user mode.
1617          * This sets TIF_SIGPENDING, but never clears it.
1618          */
1619         recalc_sigpending_tsk(current);
1620 }
1621
1622 void ptrace_notify(int exit_code)
1623 {
1624         siginfo_t info;
1625
1626         BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1627
1628         memset(&info, 0, sizeof info);
1629         info.si_signo = SIGTRAP;
1630         info.si_code = exit_code;
1631         info.si_pid = task_pid_vnr(current);
1632         info.si_uid = current_uid();
1633
1634         /* Let the debugger run.  */
1635         spin_lock_irq(&current->sighand->siglock);
1636         ptrace_stop(exit_code, 1, &info);
1637         spin_unlock_irq(&current->sighand->siglock);
1638 }
1639
1640 static void
1641 finish_stop(int stop_count)
1642 {
1643         /*
1644          * If there are no other threads in the group, or if there is
1645          * a group stop in progress and we are the last to stop,
1646          * report to the parent.  When ptraced, every thread reports itself.
1647          */
1648         if (tracehook_notify_jctl(stop_count == 0, CLD_STOPPED)) {
1649                 read_lock(&tasklist_lock);
1650                 do_notify_parent_cldstop(current, CLD_STOPPED);
1651                 read_unlock(&tasklist_lock);
1652         }
1653
1654         do {
1655                 schedule();
1656         } while (try_to_freeze());
1657         /*
1658          * Now we don't run again until continued.
1659          */
1660         current->exit_code = 0;
1661 }
1662
1663 /*
1664  * This performs the stopping for SIGSTOP and other stop signals.
1665  * We have to stop all threads in the thread group.
1666  * Returns nonzero if we've actually stopped and released the siglock.
1667  * Returns zero if we didn't stop and still hold the siglock.
1668  */
1669 static int do_signal_stop(int signr)
1670 {
1671         struct signal_struct *sig = current->signal;
1672         int stop_count;
1673
1674         if (sig->group_stop_count > 0) {
1675                 /*
1676                  * There is a group stop in progress.  We don't need to
1677                  * start another one.
1678                  */
1679                 stop_count = --sig->group_stop_count;
1680         } else {
1681                 struct task_struct *t;
1682
1683                 if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED) ||
1684                     unlikely(signal_group_exit(sig)))
1685                         return 0;
1686                 /*
1687                  * There is no group stop already in progress.
1688                  * We must initiate one now.
1689                  */
1690                 sig->group_exit_code = signr;
1691
1692                 stop_count = 0;
1693                 for (t = next_thread(current); t != current; t = next_thread(t))
1694                         /*
1695                          * Setting state to TASK_STOPPED for a group
1696                          * stop is always done with the siglock held,
1697                          * so this check has no races.
1698                          */
1699                         if (!(t->flags & PF_EXITING) &&
1700                             !task_is_stopped_or_traced(t)) {
1701                                 stop_count++;
1702                                 signal_wake_up(t, 0);
1703                         }
1704                 sig->group_stop_count = stop_count;
1705         }
1706
1707         if (stop_count == 0)
1708                 sig->flags = SIGNAL_STOP_STOPPED;
1709         current->exit_code = sig->group_exit_code;
1710         __set_current_state(TASK_STOPPED);
1711
1712         spin_unlock_irq(&current->sighand->siglock);
1713         finish_stop(stop_count);
1714         return 1;
1715 }
1716
1717 static int ptrace_signal(int signr, siginfo_t *info,
1718                          struct pt_regs *regs, void *cookie)
1719 {
1720         if (!(current->ptrace & PT_PTRACED))
1721                 return signr;
1722
1723         ptrace_signal_deliver(regs, cookie);
1724
1725         /* Let the debugger run.  */
1726         ptrace_stop(signr, 0, info);
1727
1728         /* We're back.  Did the debugger cancel the sig?  */
1729         signr = current->exit_code;
1730         if (signr == 0)
1731                 return signr;
1732
1733         current->exit_code = 0;
1734
1735         /* Update the siginfo structure if the signal has
1736            changed.  If the debugger wanted something
1737            specific in the siginfo structure then it should
1738            have updated *info via PTRACE_SETSIGINFO.  */
1739         if (signr != info->si_signo) {
1740                 info->si_signo = signr;
1741                 info->si_errno = 0;
1742                 info->si_code = SI_USER;
1743                 info->si_pid = task_pid_vnr(current->parent);
1744                 info->si_uid = task_uid(current->parent);
1745         }
1746
1747         /* If the (new) signal is now blocked, requeue it.  */
1748         if (sigismember(&current->blocked, signr)) {
1749                 specific_send_sig_info(signr, info, current);
1750                 signr = 0;
1751         }
1752
1753         return signr;
1754 }
1755
1756 int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
1757                           struct pt_regs *regs, void *cookie)
1758 {
1759         struct sighand_struct *sighand = current->sighand;
1760         struct signal_struct *signal = current->signal;
1761         int signr;
1762
1763 relock:
1764         /*
1765          * We'll jump back here after any time we were stopped in TASK_STOPPED.
1766          * While in TASK_STOPPED, we were considered "frozen enough".
1767          * Now that we woke up, it's crucial if we're supposed to be
1768          * frozen that we freeze now before running anything substantial.
1769          */
1770         try_to_freeze();
1771
1772         spin_lock_irq(&sighand->siglock);
1773         /*
1774          * Every stopped thread goes here after wakeup. Check to see if
1775          * we should notify the parent, prepare_signal(SIGCONT) encodes
1776          * the CLD_ si_code into SIGNAL_CLD_MASK bits.
1777          */
1778         if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
1779                 int why = (signal->flags & SIGNAL_STOP_CONTINUED)
1780                                 ? CLD_CONTINUED : CLD_STOPPED;
1781                 signal->flags &= ~SIGNAL_CLD_MASK;
1782                 spin_unlock_irq(&sighand->siglock);
1783
1784                 if (unlikely(!tracehook_notify_jctl(1, why)))
1785                         goto relock;
1786
1787                 read_lock(&tasklist_lock);
1788                 do_notify_parent_cldstop(current->group_leader, why);
1789                 read_unlock(&tasklist_lock);
1790                 goto relock;
1791         }
1792
1793         for (;;) {
1794                 struct k_sigaction *ka;
1795
1796                 if (unlikely(signal->group_stop_count > 0) &&
1797                     do_signal_stop(0))
1798                         goto relock;
1799
1800                 /*
1801                  * Tracing can induce an artifical signal and choose sigaction.
1802                  * The return value in @signr determines the default action,
1803                  * but @info->si_signo is the signal number we will report.
1804                  */
1805                 signr = tracehook_get_signal(current, regs, info, return_ka);
1806                 if (unlikely(signr < 0))
1807                         goto relock;
1808                 if (unlikely(signr != 0))
1809                         ka = return_ka;
1810                 else {
1811                         signr = dequeue_signal(current, &current->blocked,
1812                                                info);
1813
1814                         if (!signr)
1815                                 break; /* will return 0 */
1816
1817                         if (signr != SIGKILL) {
1818                                 signr = ptrace_signal(signr, info,
1819                                                       regs, cookie);
1820                                 if (!signr)
1821                                         continue;
1822                         }
1823
1824                         ka = &sighand->action[signr-1];
1825                 }
1826
1827                 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
1828                         continue;
1829                 if (ka->sa.sa_handler != SIG_DFL) {
1830                         /* Run the handler.  */
1831                         *return_ka = *ka;
1832
1833                         if (ka->sa.sa_flags & SA_ONESHOT)
1834                                 ka->sa.sa_handler = SIG_DFL;
1835
1836                         break; /* will return non-zero "signr" value */
1837                 }
1838
1839                 /*
1840                  * Now we are doing the default action for this signal.
1841                  */
1842                 if (sig_kernel_ignore(signr)) /* Default is nothing. */
1843                         continue;
1844
1845                 /*
1846                  * Global init gets no signals it doesn't want.
1847                  */
1848                 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
1849                     !signal_group_exit(signal))
1850                         continue;
1851
1852                 if (sig_kernel_stop(signr)) {
1853                         /*
1854                          * The default action is to stop all threads in
1855                          * the thread group.  The job control signals
1856                          * do nothing in an orphaned pgrp, but SIGSTOP
1857                          * always works.  Note that siglock needs to be
1858                          * dropped during the call to is_orphaned_pgrp()
1859                          * because of lock ordering with tasklist_lock.
1860                          * This allows an intervening SIGCONT to be posted.
1861                          * We need to check for that and bail out if necessary.
1862                          */
1863                         if (signr != SIGSTOP) {
1864                                 spin_unlock_irq(&sighand->siglock);
1865
1866                                 /* signals can be posted during this window */
1867
1868                                 if (is_current_pgrp_orphaned())
1869                                         goto relock;
1870
1871                                 spin_lock_irq(&sighand->siglock);
1872                         }
1873
1874                         if (likely(do_signal_stop(info->si_signo))) {
1875                                 /* It released the siglock.  */
1876                                 goto relock;
1877                         }
1878
1879                         /*
1880                          * We didn't actually stop, due to a race
1881                          * with SIGCONT or something like that.
1882                          */
1883                         continue;
1884                 }
1885
1886                 spin_unlock_irq(&sighand->siglock);
1887
1888                 /*
1889                  * Anything else is fatal, maybe with a core dump.
1890                  */
1891                 current->flags |= PF_SIGNALED;
1892
1893                 if (sig_kernel_coredump(signr)) {
1894                         if (print_fatal_signals)
1895                                 print_fatal_signal(regs, info->si_signo);
1896                         /*
1897                          * If it was able to dump core, this kills all
1898                          * other threads in the group and synchronizes with
1899                          * their demise.  If we lost the race with another
1900                          * thread getting here, it set group_exit_code
1901                          * first and our do_group_exit call below will use
1902                          * that value and ignore the one we pass it.
1903                          */
1904                         do_coredump(info->si_signo, info->si_signo, regs);
1905                 }
1906
1907                 /*
1908                  * Death signals, no core dump.
1909                  */
1910                 do_group_exit(info->si_signo);
1911                 /* NOTREACHED */
1912         }
1913         spin_unlock_irq(&sighand->siglock);
1914         return signr;
1915 }
1916
1917 void exit_signals(struct task_struct *tsk)
1918 {
1919         int group_stop = 0;
1920         struct task_struct *t;
1921
1922         if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
1923                 tsk->flags |= PF_EXITING;
1924                 return;
1925         }
1926
1927         spin_lock_irq(&tsk->sighand->siglock);
1928         /*
1929          * From now this task is not visible for group-wide signals,
1930          * see wants_signal(), do_signal_stop().
1931          */
1932         tsk->flags |= PF_EXITING;
1933         if (!signal_pending(tsk))
1934                 goto out;
1935
1936         /* It could be that __group_complete_signal() choose us to
1937          * notify about group-wide signal. Another thread should be
1938          * woken now to take the signal since we will not.
1939          */
1940         for (t = tsk; (t = next_thread(t)) != tsk; )
1941                 if (!signal_pending(t) && !(t->flags & PF_EXITING))
1942                         recalc_sigpending_and_wake(t);
1943
1944         if (unlikely(tsk->signal->group_stop_count) &&
1945                         !--tsk->signal->group_stop_count) {
1946                 tsk->signal->flags = SIGNAL_STOP_STOPPED;
1947                 group_stop = 1;
1948         }
1949 out:
1950         spin_unlock_irq(&tsk->sighand->siglock);
1951
1952         if (unlikely(group_stop) && tracehook_notify_jctl(1, CLD_STOPPED)) {
1953                 read_lock(&tasklist_lock);
1954                 do_notify_parent_cldstop(tsk, CLD_STOPPED);
1955                 read_unlock(&tasklist_lock);
1956         }
1957 }
1958
1959 EXPORT_SYMBOL(recalc_sigpending);
1960 EXPORT_SYMBOL_GPL(dequeue_signal);
1961 EXPORT_SYMBOL(flush_signals);
1962 EXPORT_SYMBOL(force_sig);
1963 EXPORT_SYMBOL(send_sig);
1964 EXPORT_SYMBOL(send_sig_info);
1965 EXPORT_SYMBOL(sigprocmask);
1966 EXPORT_SYMBOL(block_all_signals);
1967 EXPORT_SYMBOL(unblock_all_signals);
1968
1969
1970 /*
1971  * System call entry points.
1972  */
1973
1974 SYSCALL_DEFINE0(restart_syscall)
1975 {
1976         struct restart_block *restart = &current_thread_info()->restart_block;
1977         return restart->fn(restart);
1978 }
1979
1980 long do_no_restart_syscall(struct restart_block *param)
1981 {
1982         return -EINTR;
1983 }
1984
1985 /*
1986  * We don't need to get the kernel lock - this is all local to this
1987  * particular thread.. (and that's good, because this is _heavily_
1988  * used by various programs)
1989  */
1990
1991 /*
1992  * This is also useful for kernel threads that want to temporarily
1993  * (or permanently) block certain signals.
1994  *
1995  * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
1996  * interface happily blocks "unblockable" signals like SIGKILL
1997  * and friends.
1998  */
1999 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2000 {
2001         int error;
2002
2003         spin_lock_irq(&current->sighand->siglock);
2004         if (oldset)
2005                 *oldset = current->blocked;
2006
2007         error = 0;
2008         switch (how) {
2009         case SIG_BLOCK:
2010                 sigorsets(&current->blocked, &current->blocked, set);
2011                 break;
2012         case SIG_UNBLOCK:
2013                 signandsets(&current->blocked, &current->blocked, set);
2014                 break;
2015         case SIG_SETMASK:
2016                 current->blocked = *set;
2017                 break;
2018         default:
2019                 error = -EINVAL;
2020         }
2021         recalc_sigpending();
2022         spin_unlock_irq(&current->sighand->siglock);
2023
2024         return error;
2025 }
2026
2027 SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, set,
2028                 sigset_t __user *, oset, size_t, sigsetsize)
2029 {
2030         int error = -EINVAL;
2031         sigset_t old_set, new_set;
2032
2033         /* XXX: Don't preclude handling different sized sigset_t's.  */
2034         if (sigsetsize != sizeof(sigset_t))
2035                 goto out;
2036
2037         if (set) {
2038                 error = -EFAULT;
2039                 if (copy_from_user(&new_set, set, sizeof(*set)))
2040                         goto out;
2041                 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2042
2043                 error = sigprocmask(how, &new_set, &old_set);
2044                 if (error)
2045                         goto out;
2046                 if (oset)
2047                         goto set_old;
2048         } else if (oset) {
2049                 spin_lock_irq(&current->sighand->siglock);
2050                 old_set = current->blocked;
2051                 spin_unlock_irq(&current->sighand->siglock);
2052
2053         set_old:
2054                 error = -EFAULT;
2055                 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2056                         goto out;
2057         }
2058         error = 0;
2059 out:
2060         return error;
2061 }
2062
2063 long do_sigpending(void __user *set, unsigned long sigsetsize)
2064 {
2065         long error = -EINVAL;
2066         sigset_t pending;
2067
2068         if (sigsetsize > sizeof(sigset_t))
2069                 goto out;
2070
2071         spin_lock_irq(&current->sighand->siglock);
2072         sigorsets(&pending, &current->pending.signal,
2073                   &current->signal->shared_pending.signal);
2074         spin_unlock_irq(&current->sighand->siglock);
2075
2076         /* Outside the lock because only this thread touches it.  */
2077         sigandsets(&pending, &current->blocked, &pending);
2078
2079         error = -EFAULT;
2080         if (!copy_to_user(set, &pending, sigsetsize))
2081                 error = 0;
2082
2083 out:
2084         return error;
2085 }       
2086
2087 SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, set, size_t, sigsetsize)
2088 {
2089         return do_sigpending(set, sigsetsize);
2090 }
2091
2092 #ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2093
2094 int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2095 {
2096         int err;
2097
2098         if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2099                 return -EFAULT;
2100         if (from->si_code < 0)
2101                 return __copy_to_user(to, from, sizeof(siginfo_t))
2102                         ? -EFAULT : 0;
2103         /*
2104          * If you change siginfo_t structure, please be sure
2105          * this code is fixed accordingly.
2106          * Please remember to update the signalfd_copyinfo() function
2107          * inside fs/signalfd.c too, in case siginfo_t changes.
2108          * It should never copy any pad contained in the structure
2109          * to avoid security leaks, but must copy the generic
2110          * 3 ints plus the relevant union member.
2111          */
2112         err = __put_user(from->si_signo, &to->si_signo);
2113         err |= __put_user(from->si_errno, &to->si_errno);
2114         err |= __put_user((short)from->si_code, &to->si_code);
2115         switch (from->si_code & __SI_MASK) {
2116         case __SI_KILL:
2117                 err |= __put_user(from->si_pid, &to->si_pid);
2118                 err |= __put_user(from->si_uid, &to->si_uid);
2119                 break;
2120         case __SI_TIMER:
2121                  err |= __put_user(from->si_tid, &to->si_tid);
2122                  err |= __put_user(from->si_overrun, &to->si_overrun);
2123                  err |= __put_user(from->si_ptr, &to->si_ptr);
2124                 break;
2125         case __SI_POLL:
2126                 err |= __put_user(from->si_band, &to->si_band);
2127                 err |= __put_user(from->si_fd, &to->si_fd);
2128                 break;
2129         case __SI_FAULT:
2130                 err |= __put_user(from->si_addr, &to->si_addr);
2131 #ifdef __ARCH_SI_TRAPNO
2132                 err |= __put_user(from->si_trapno, &to->si_trapno);
2133 #endif
2134                 break;
2135         case __SI_CHLD:
2136                 err |= __put_user(from->si_pid, &to->si_pid);
2137                 err |= __put_user(from->si_uid, &to->si_uid);
2138                 err |= __put_user(from->si_status, &to->si_status);
2139                 err |= __put_user(from->si_utime, &to->si_utime);
2140                 err |= __put_user(from->si_stime, &to->si_stime);
2141                 break;
2142         case __SI_RT: /* This is not generated by the kernel as of now. */
2143         case __SI_MESGQ: /* But this is */
2144                 err |= __put_user(from->si_pid, &to->si_pid);
2145                 err |= __put_user(from->si_uid, &to->si_uid);
2146                 err |= __put_user(from->si_ptr, &to->si_ptr);
2147                 break;
2148         default: /* this is just in case for now ... */
2149                 err |= __put_user(from->si_pid, &to->si_pid);
2150                 err |= __put_user(from->si_uid, &to->si_uid);
2151                 break;
2152         }
2153         return err;
2154 }
2155
2156 #endif
2157
2158 SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
2159                 siginfo_t __user *, uinfo, const struct timespec __user *, uts,
2160                 size_t, sigsetsize)
2161 {
2162         int ret, sig;
2163         sigset_t these;
2164         struct timespec ts;
2165         siginfo_t info;
2166         long timeout = 0;
2167
2168         /* XXX: Don't preclude handling different sized sigset_t's.  */
2169         if (sigsetsize != sizeof(sigset_t))
2170                 return -EINVAL;
2171
2172         if (copy_from_user(&these, uthese, sizeof(these)))
2173                 return -EFAULT;
2174                 
2175         /*
2176          * Invert the set of allowed signals to get those we
2177          * want to block.
2178          */
2179         sigdelsetmask(&these, sigmask(SIGKILL)|sigmask(SIGSTOP));
2180         signotset(&these);
2181
2182         if (uts) {
2183                 if (copy_from_user(&ts, uts, sizeof(ts)))
2184                         return -EFAULT;
2185                 if (ts.tv_nsec >= 1000000000L || ts.tv_nsec < 0
2186                     || ts.tv_sec < 0)
2187                         return -EINVAL;
2188         }
2189
2190         spin_lock_irq(&current->sighand->siglock);
2191         sig = dequeue_signal(current, &these, &info);
2192         if (!sig) {
2193                 timeout = MAX_SCHEDULE_TIMEOUT;
2194                 if (uts)
2195                         timeout = (timespec_to_jiffies(&ts)
2196                                    + (ts.tv_sec || ts.tv_nsec));
2197
2198                 if (timeout) {
2199                         /* None ready -- temporarily unblock those we're
2200                          * interested while we are sleeping in so that we'll
2201                          * be awakened when they arrive.  */
2202                         current->real_blocked = current->blocked;
2203                         sigandsets(&current->blocked, &current->blocked, &these);
2204                         recalc_sigpending();
2205                         spin_unlock_irq(&current->sighand->siglock);
2206
2207                         timeout = schedule_timeout_interruptible(timeout);
2208
2209                         spin_lock_irq(&current->sighand->siglock);
2210                         sig = dequeue_signal(current, &these, &info);
2211                         current->blocked = current->real_blocked;
2212                         siginitset(&current->real_blocked, 0);
2213                         recalc_sigpending();
2214                 }
2215         }
2216         spin_unlock_irq(&current->sighand->siglock);
2217
2218         if (sig) {
2219                 ret = sig;
2220                 if (uinfo) {
2221                         if (copy_siginfo_to_user(uinfo, &info))
2222                                 ret = -EFAULT;
2223                 }
2224         } else {
2225                 ret = -EAGAIN;
2226                 if (timeout)
2227                         ret = -EINTR;
2228         }
2229
2230         return ret;
2231 }
2232
2233 SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
2234 {
2235         struct siginfo info;
2236
2237         info.si_signo = sig;
2238         info.si_errno = 0;
2239         info.si_code = SI_USER;
2240         info.si_pid = task_tgid_vnr(current);
2241         info.si_uid = current_uid();
2242
2243         return kill_something_info(sig, &info, pid);
2244 }
2245
2246 static int do_tkill(pid_t tgid, pid_t pid, int sig)
2247 {
2248         int error;
2249         struct siginfo info;
2250         struct task_struct *p;
2251         unsigned long flags;
2252
2253         error = -ESRCH;
2254         info.si_signo = sig;
2255         info.si_errno = 0;
2256         info.si_code = SI_TKILL;
2257         info.si_pid = task_tgid_vnr(current);
2258         info.si_uid = current_uid();
2259
2260         rcu_read_lock();
2261         p = find_task_by_vpid(pid);
2262         if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
2263                 error = check_kill_permission(sig, &info, p);
2264                 /*
2265                  * The null signal is a permissions and process existence
2266                  * probe.  No signal is actually delivered.
2267                  *
2268                  * If lock_task_sighand() fails we pretend the task dies
2269                  * after receiving the signal. The window is tiny, and the
2270                  * signal is private anyway.
2271                  */
2272                 if (!error && sig && lock_task_sighand(p, &flags)) {
2273                         error = specific_send_sig_info(sig, &info, p);
2274                         unlock_task_sighand(p, &flags);
2275                 }
2276         }
2277         rcu_read_unlock();
2278
2279         return error;
2280 }
2281
2282 /**
2283  *  sys_tgkill - send signal to one specific thread
2284  *  @tgid: the thread group ID of the thread
2285  *  @pid: the PID of the thread
2286  *  @sig: signal to be sent
2287  *
2288  *  This syscall also checks the @tgid and returns -ESRCH even if the PID
2289  *  exists but it's not belonging to the target process anymore. This
2290  *  method solves the problem of threads exiting and PIDs getting reused.
2291  */
2292 SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
2293 {
2294         /* This is only valid for single tasks */
2295         if (pid <= 0 || tgid <= 0)
2296                 return -EINVAL;
2297
2298         return do_tkill(tgid, pid, sig);
2299 }
2300
2301 /*
2302  *  Send a signal to only one task, even if it's a CLONE_THREAD task.
2303  */
2304 SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
2305 {
2306         /* This is only valid for single tasks */
2307         if (pid <= 0)
2308                 return -EINVAL;
2309
2310         return do_tkill(0, pid, sig);
2311 }
2312
2313 SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
2314                 siginfo_t __user *, uinfo)
2315 {
2316         siginfo_t info;
2317
2318         if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2319                 return -EFAULT;
2320
2321         /* Not even root can pretend to send signals from the kernel.
2322            Nor can they impersonate a kill(), which adds source info.  */
2323         if (info.si_code >= 0)
2324                 return -EPERM;
2325         info.si_signo = sig;
2326
2327         /* POSIX.1b doesn't mention process groups.  */
2328         return kill_proc_info(sig, &info, pid);
2329 }
2330
2331 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
2332 {
2333         struct task_struct *t = current;
2334         struct k_sigaction *k;
2335         sigset_t mask;
2336
2337         if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
2338                 return -EINVAL;
2339
2340         k = &t->sighand->action[sig-1];
2341
2342         spin_lock_irq(&current->sighand->siglock);
2343         if (oact)
2344                 *oact = *k;
2345
2346         if (act) {
2347                 sigdelsetmask(&act->sa.sa_mask,
2348                               sigmask(SIGKILL) | sigmask(SIGSTOP));
2349                 *k = *act;
2350                 /*
2351                  * POSIX 3.3.1.3:
2352                  *  "Setting a signal action to SIG_IGN for a signal that is
2353                  *   pending shall cause the pending signal to be discarded,
2354                  *   whether or not it is blocked."
2355                  *
2356                  *  "Setting a signal action to SIG_DFL for a signal that is
2357                  *   pending and whose default action is to ignore the signal
2358                  *   (for example, SIGCHLD), shall cause the pending signal to
2359                  *   be discarded, whether or not it is blocked"
2360                  */
2361                 if (sig_handler_ignored(sig_handler(t, sig), sig)) {
2362                         sigemptyset(&mask);
2363                         sigaddset(&mask, sig);
2364                         rm_from_queue_full(&mask, &t->signal->shared_pending);
2365                         do {
2366                                 rm_from_queue_full(&mask, &t->pending);
2367                                 t = next_thread(t);
2368                         } while (t != current);
2369                 }
2370         }
2371
2372         spin_unlock_irq(&current->sighand->siglock);
2373         return 0;
2374 }
2375
2376 int 
2377 do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
2378 {
2379         stack_t oss;
2380         int error;
2381
2382         if (uoss) {
2383                 oss.ss_sp = (void __user *) current->sas_ss_sp;
2384                 oss.ss_size = current->sas_ss_size;
2385                 oss.ss_flags = sas_ss_flags(sp);
2386         }
2387
2388         if (uss) {
2389                 void __user *ss_sp;
2390                 size_t ss_size;
2391                 int ss_flags;
2392
2393                 error = -EFAULT;
2394                 if (!access_ok(VERIFY_READ, uss, sizeof(*uss))
2395                     || __get_user(ss_sp, &uss->ss_sp)
2396                     || __get_user(ss_flags, &uss->ss_flags)
2397                     || __get_user(ss_size, &uss->ss_size))
2398                         goto out;
2399
2400                 error = -EPERM;
2401                 if (on_sig_stack(sp))
2402                         goto out;
2403
2404                 error = -EINVAL;
2405                 /*
2406                  *
2407                  * Note - this code used to test ss_flags incorrectly
2408                  *        old code may have been written using ss_flags==0
2409                  *        to mean ss_flags==SS_ONSTACK (as this was the only
2410                  *        way that worked) - this fix preserves that older
2411                  *        mechanism
2412                  */
2413                 if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
2414                         goto out;
2415
2416                 if (ss_flags == SS_DISABLE) {
2417                         ss_size = 0;
2418                         ss_sp = NULL;
2419                 } else {
2420                         error = -ENOMEM;
2421                         if (ss_size < MINSIGSTKSZ)
2422                                 goto out;
2423                 }
2424
2425                 current->sas_ss_sp = (unsigned long) ss_sp;
2426                 current->sas_ss_size = ss_size;
2427         }
2428
2429         if (uoss) {
2430                 error = -EFAULT;
2431                 if (copy_to_user(uoss, &oss, sizeof(oss)))
2432                         goto out;
2433         }
2434
2435         error = 0;
2436 out:
2437         return error;
2438 }
2439
2440 #ifdef __ARCH_WANT_SYS_SIGPENDING
2441
2442 SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, set)
2443 {
2444         return do_sigpending(set, sizeof(*set));
2445 }
2446
2447 #endif
2448
2449 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
2450 /* Some platforms have their own version with special arguments others
2451    support only sys_rt_sigprocmask.  */
2452
2453 SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, set,
2454                 old_sigset_t __user *, oset)
2455 {
2456         int error;
2457         old_sigset_t old_set, new_set;
2458
2459         if (set) {
2460                 error = -EFAULT;
2461                 if (copy_from_user(&new_set, set, sizeof(*set)))
2462                         goto out;
2463                 new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
2464
2465                 spin_lock_irq(&current->sighand->siglock);
2466                 old_set = current->blocked.sig[0];
2467
2468                 error = 0;
2469                 switch (how) {
2470                 default:
2471                         error = -EINVAL;
2472                         break;
2473                 case SIG_BLOCK:
2474                         sigaddsetmask(&current->blocked, new_set);
2475                         break;
2476                 case SIG_UNBLOCK:
2477                         sigdelsetmask(&current->blocked, new_set);
2478                         break;
2479                 case SIG_SETMASK:
2480                         current->blocked.sig[0] = new_set;
2481                         break;
2482                 }
2483
2484                 recalc_sigpending();
2485                 spin_unlock_irq(&current->sighand->siglock);
2486                 if (error)
2487                         goto out;
2488                 if (oset)
2489                         goto set_old;
2490         } else if (oset) {
2491                 old_set = current->blocked.sig[0];
2492         set_old:
2493                 error = -EFAULT;
2494                 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2495                         goto out;
2496         }
2497         error = 0;
2498 out:
2499         return error;
2500 }
2501 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
2502
2503 #ifdef __ARCH_WANT_SYS_RT_SIGACTION
2504 SYSCALL_DEFINE4(rt_sigaction, int, sig,
2505                 const struct sigaction __user *, act,
2506                 struct sigaction __user *, oact,
2507                 size_t, sigsetsize)
2508 {
2509         struct k_sigaction new_sa, old_sa;
2510         int ret = -EINVAL;
2511
2512         /* XXX: Don't preclude handling different sized sigset_t's.  */
2513         if (sigsetsize != sizeof(sigset_t))
2514                 goto out;
2515
2516         if (act) {
2517                 if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
2518                         return -EFAULT;
2519         }
2520
2521         ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
2522
2523         if (!ret && oact) {
2524                 if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
2525                         return -EFAULT;
2526         }
2527 out:
2528         return ret;
2529 }
2530 #endif /* __ARCH_WANT_SYS_RT_SIGACTION */
2531
2532 #ifdef __ARCH_WANT_SYS_SGETMASK
2533
2534 /*
2535  * For backwards compatibility.  Functionality superseded by sigprocmask.
2536  */
2537 SYSCALL_DEFINE0(sgetmask)
2538 {
2539         /* SMP safe */
2540         return current->blocked.sig[0];
2541 }
2542
2543 SYSCALL_DEFINE1(ssetmask, int, newmask)
2544 {
2545         int old;
2546
2547         spin_lock_irq(&current->sighand->siglock);
2548         old = current->blocked.sig[0];
2549
2550         siginitset(&current->blocked, newmask & ~(sigmask(SIGKILL)|
2551                                                   sigmask(SIGSTOP)));
2552         recalc_sigpending();
2553         spin_unlock_irq(&current->sighand->siglock);
2554
2555         return old;
2556 }
2557 #endif /* __ARCH_WANT_SGETMASK */
2558
2559 #ifdef __ARCH_WANT_SYS_SIGNAL
2560 /*
2561  * For backwards compatibility.  Functionality superseded by sigaction.
2562  */
2563 SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
2564 {
2565         struct k_sigaction new_sa, old_sa;
2566         int ret;
2567
2568         new_sa.sa.sa_handler = handler;
2569         new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
2570         sigemptyset(&new_sa.sa.sa_mask);
2571
2572         ret = do_sigaction(sig, &new_sa, &old_sa);
2573
2574         return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
2575 }
2576 #endif /* __ARCH_WANT_SYS_SIGNAL */
2577
2578 #ifdef __ARCH_WANT_SYS_PAUSE
2579
2580 SYSCALL_DEFINE0(pause)
2581 {
2582         current->state = TASK_INTERRUPTIBLE;
2583         schedule();
2584         return -ERESTARTNOHAND;
2585 }
2586
2587 #endif
2588
2589 #ifdef __ARCH_WANT_SYS_RT_SIGSUSPEND
2590 SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
2591 {
2592         sigset_t newset;
2593
2594         /* XXX: Don't preclude handling different sized sigset_t's.  */
2595         if (sigsetsize != sizeof(sigset_t))
2596                 return -EINVAL;
2597
2598         if (copy_from_user(&newset, unewset, sizeof(newset)))
2599                 return -EFAULT;
2600         sigdelsetmask(&newset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2601
2602         spin_lock_irq(&current->sighand->siglock);
2603         current->saved_sigmask = current->blocked;
2604         current->blocked = newset;
2605         recalc_sigpending();
2606         spin_unlock_irq(&current->sighand->siglock);
2607
2608         current->state = TASK_INTERRUPTIBLE;
2609         schedule();
2610         set_restore_sigmask();
2611         return -ERESTARTNOHAND;
2612 }
2613 #endif /* __ARCH_WANT_SYS_RT_SIGSUSPEND */
2614
2615 __attribute__((weak)) const char *arch_vma_name(struct vm_area_struct *vma)
2616 {
2617         return NULL;
2618 }
2619
2620 void __init signals_init(void)
2621 {
2622         sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
2623 }