]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - kernel/signal.c
Linux 2.6.27.39
[karo-tx-linux.git] / 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
31 #include <asm/param.h>
32 #include <asm/uaccess.h>
33 #include <asm/unistd.h>
34 #include <asm/siginfo.h>
35 #include "audit.h"      /* audit_signal_info() */
36
37 /*
38  * SLAB caches for signal bits.
39  */
40
41 static struct kmem_cache *sigqueue_cachep;
42
43 static void __user *sig_handler(struct task_struct *t, int sig)
44 {
45         return t->sighand->action[sig - 1].sa.sa_handler;
46 }
47
48 static int sig_handler_ignored(void __user *handler, int sig)
49 {
50         /* Is it explicitly or implicitly ignored? */
51         return handler == SIG_IGN ||
52                 (handler == SIG_DFL && sig_kernel_ignore(sig));
53 }
54
55 static int sig_ignored(struct task_struct *t, int sig)
56 {
57         void __user *handler;
58
59         /*
60          * Blocked signals are never ignored, since the
61          * signal handler may change by the time it is
62          * unblocked.
63          */
64         if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
65                 return 0;
66
67         handler = sig_handler(t, sig);
68         if (!sig_handler_ignored(handler, sig))
69                 return 0;
70
71         /*
72          * Tracers may want to know about even ignored signals.
73          */
74         return !tracehook_consider_ignored_signal(t, sig, handler);
75 }
76
77 /*
78  * Re-calculate pending state from the set of locally pending
79  * signals, globally pending signals, and blocked signals.
80  */
81 static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
82 {
83         unsigned long ready;
84         long i;
85
86         switch (_NSIG_WORDS) {
87         default:
88                 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
89                         ready |= signal->sig[i] &~ blocked->sig[i];
90                 break;
91
92         case 4: ready  = signal->sig[3] &~ blocked->sig[3];
93                 ready |= signal->sig[2] &~ blocked->sig[2];
94                 ready |= signal->sig[1] &~ blocked->sig[1];
95                 ready |= signal->sig[0] &~ blocked->sig[0];
96                 break;
97
98         case 2: ready  = signal->sig[1] &~ blocked->sig[1];
99                 ready |= signal->sig[0] &~ blocked->sig[0];
100                 break;
101
102         case 1: ready  = signal->sig[0] &~ blocked->sig[0];
103         }
104         return ready != 0;
105 }
106
107 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
108
109 static int recalc_sigpending_tsk(struct task_struct *t)
110 {
111         if (t->signal->group_stop_count > 0 ||
112             PENDING(&t->pending, &t->blocked) ||
113             PENDING(&t->signal->shared_pending, &t->blocked)) {
114                 set_tsk_thread_flag(t, TIF_SIGPENDING);
115                 return 1;
116         }
117         /*
118          * We must never clear the flag in another thread, or in current
119          * when it's possible the current syscall is returning -ERESTART*.
120          * So we don't clear it here, and only callers who know they should do.
121          */
122         return 0;
123 }
124
125 /*
126  * After recalculating TIF_SIGPENDING, we need to make sure the task wakes up.
127  * This is superfluous when called on current, the wakeup is a harmless no-op.
128  */
129 void recalc_sigpending_and_wake(struct task_struct *t)
130 {
131         if (recalc_sigpending_tsk(t))
132                 signal_wake_up(t, 0);
133 }
134
135 void recalc_sigpending(void)
136 {
137         if (unlikely(tracehook_force_sigpending()))
138                 set_thread_flag(TIF_SIGPENDING);
139         else if (!recalc_sigpending_tsk(current) && !freezing(current))
140                 clear_thread_flag(TIF_SIGPENDING);
141
142 }
143
144 /* Given the mask, find the first available signal that should be serviced. */
145
146 int next_signal(struct sigpending *pending, sigset_t *mask)
147 {
148         unsigned long i, *s, *m, x;
149         int sig = 0;
150         
151         s = pending->signal.sig;
152         m = mask->sig;
153         switch (_NSIG_WORDS) {
154         default:
155                 for (i = 0; i < _NSIG_WORDS; ++i, ++s, ++m)
156                         if ((x = *s &~ *m) != 0) {
157                                 sig = ffz(~x) + i*_NSIG_BPW + 1;
158                                 break;
159                         }
160                 break;
161
162         case 2: if ((x = s[0] &~ m[0]) != 0)
163                         sig = 1;
164                 else if ((x = s[1] &~ m[1]) != 0)
165                         sig = _NSIG_BPW + 1;
166                 else
167                         break;
168                 sig += ffz(~x);
169                 break;
170
171         case 1: if ((x = *s &~ *m) != 0)
172                         sig = ffz(~x) + 1;
173                 break;
174         }
175         
176         return sig;
177 }
178
179 static struct sigqueue *__sigqueue_alloc(struct task_struct *t, gfp_t flags,
180                                          int override_rlimit)
181 {
182         struct sigqueue *q = NULL;
183         struct user_struct *user;
184
185         /*
186          * In order to avoid problems with "switch_user()", we want to make
187          * sure that the compiler doesn't re-load "t->user"
188          */
189         user = t->user;
190         barrier();
191         atomic_inc(&user->sigpending);
192         if (override_rlimit ||
193             atomic_read(&user->sigpending) <=
194                         t->signal->rlim[RLIMIT_SIGPENDING].rlim_cur)
195                 q = kmem_cache_alloc(sigqueue_cachep, flags);
196         if (unlikely(q == NULL)) {
197                 atomic_dec(&user->sigpending);
198         } else {
199                 INIT_LIST_HEAD(&q->list);
200                 q->flags = 0;
201                 q->user = get_uid(user);
202         }
203         return(q);
204 }
205
206 static void __sigqueue_free(struct sigqueue *q)
207 {
208         if (q->flags & SIGQUEUE_PREALLOC)
209                 return;
210         atomic_dec(&q->user->sigpending);
211         free_uid(q->user);
212         kmem_cache_free(sigqueue_cachep, q);
213 }
214
215 void flush_sigqueue(struct sigpending *queue)
216 {
217         struct sigqueue *q;
218
219         sigemptyset(&queue->signal);
220         while (!list_empty(&queue->list)) {
221                 q = list_entry(queue->list.next, struct sigqueue , list);
222                 list_del_init(&q->list);
223                 __sigqueue_free(q);
224         }
225 }
226
227 /*
228  * Flush all pending signals for a task.
229  */
230 void flush_signals(struct task_struct *t)
231 {
232         unsigned long flags;
233
234         spin_lock_irqsave(&t->sighand->siglock, flags);
235         clear_tsk_thread_flag(t, TIF_SIGPENDING);
236         flush_sigqueue(&t->pending);
237         flush_sigqueue(&t->signal->shared_pending);
238         spin_unlock_irqrestore(&t->sighand->siglock, flags);
239 }
240
241 static void __flush_itimer_signals(struct sigpending *pending)
242 {
243         sigset_t signal, retain;
244         struct sigqueue *q, *n;
245
246         signal = pending->signal;
247         sigemptyset(&retain);
248
249         list_for_each_entry_safe(q, n, &pending->list, list) {
250                 int sig = q->info.si_signo;
251
252                 if (likely(q->info.si_code != SI_TIMER)) {
253                         sigaddset(&retain, sig);
254                 } else {
255                         sigdelset(&signal, sig);
256                         list_del_init(&q->list);
257                         __sigqueue_free(q);
258                 }
259         }
260
261         sigorsets(&pending->signal, &signal, &retain);
262 }
263
264 void flush_itimer_signals(void)
265 {
266         struct task_struct *tsk = current;
267         unsigned long flags;
268
269         spin_lock_irqsave(&tsk->sighand->siglock, flags);
270         __flush_itimer_signals(&tsk->pending);
271         __flush_itimer_signals(&tsk->signal->shared_pending);
272         spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
273 }
274
275 void ignore_signals(struct task_struct *t)
276 {
277         int i;
278
279         for (i = 0; i < _NSIG; ++i)
280                 t->sighand->action[i].sa.sa_handler = SIG_IGN;
281
282         flush_signals(t);
283 }
284
285 /*
286  * Flush all handlers for a task.
287  */
288
289 void
290 flush_signal_handlers(struct task_struct *t, int force_default)
291 {
292         int i;
293         struct k_sigaction *ka = &t->sighand->action[0];
294         for (i = _NSIG ; i != 0 ; i--) {
295                 if (force_default || ka->sa.sa_handler != SIG_IGN)
296                         ka->sa.sa_handler = SIG_DFL;
297                 ka->sa.sa_flags = 0;
298                 sigemptyset(&ka->sa.sa_mask);
299                 ka++;
300         }
301 }
302
303 int unhandled_signal(struct task_struct *tsk, int sig)
304 {
305         void __user *handler = tsk->sighand->action[sig-1].sa.sa_handler;
306         if (is_global_init(tsk))
307                 return 1;
308         if (handler != SIG_IGN && handler != SIG_DFL)
309                 return 0;
310         return !tracehook_consider_fatal_signal(tsk, sig, handler);
311 }
312
313
314 /* Notify the system that a driver wants to block all signals for this
315  * process, and wants to be notified if any signals at all were to be
316  * sent/acted upon.  If the notifier routine returns non-zero, then the
317  * signal will be acted upon after all.  If the notifier routine returns 0,
318  * then then signal will be blocked.  Only one block per process is
319  * allowed.  priv is a pointer to private data that the notifier routine
320  * can use to determine if the signal should be blocked or not.  */
321
322 void
323 block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
324 {
325         unsigned long flags;
326
327         spin_lock_irqsave(&current->sighand->siglock, flags);
328         current->notifier_mask = mask;
329         current->notifier_data = priv;
330         current->notifier = notifier;
331         spin_unlock_irqrestore(&current->sighand->siglock, flags);
332 }
333
334 /* Notify the system that blocking has ended. */
335
336 void
337 unblock_all_signals(void)
338 {
339         unsigned long flags;
340
341         spin_lock_irqsave(&current->sighand->siglock, flags);
342         current->notifier = NULL;
343         current->notifier_data = NULL;
344         recalc_sigpending();
345         spin_unlock_irqrestore(&current->sighand->siglock, flags);
346 }
347
348 static void collect_signal(int sig, struct sigpending *list, siginfo_t *info)
349 {
350         struct sigqueue *q, *first = NULL;
351
352         /*
353          * Collect the siginfo appropriate to this signal.  Check if
354          * there is another siginfo for the same signal.
355         */
356         list_for_each_entry(q, &list->list, list) {
357                 if (q->info.si_signo == sig) {
358                         if (first)
359                                 goto still_pending;
360                         first = q;
361                 }
362         }
363
364         sigdelset(&list->signal, sig);
365
366         if (first) {
367 still_pending:
368                 list_del_init(&first->list);
369                 copy_siginfo(info, &first->info);
370                 __sigqueue_free(first);
371         } else {
372                 /* Ok, it wasn't in the queue.  This must be
373                    a fast-pathed signal or we must have been
374                    out of queue space.  So zero out the info.
375                  */
376                 info->si_signo = sig;
377                 info->si_errno = 0;
378                 info->si_code = 0;
379                 info->si_pid = 0;
380                 info->si_uid = 0;
381         }
382 }
383
384 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
385                         siginfo_t *info)
386 {
387         int sig = next_signal(pending, mask);
388
389         if (sig) {
390                 if (current->notifier) {
391                         if (sigismember(current->notifier_mask, sig)) {
392                                 if (!(current->notifier)(current->notifier_data)) {
393                                         clear_thread_flag(TIF_SIGPENDING);
394                                         return 0;
395                                 }
396                         }
397                 }
398
399                 collect_signal(sig, pending, info);
400         }
401
402         return sig;
403 }
404
405 /*
406  * Dequeue a signal and return the element to the caller, which is 
407  * expected to free it.
408  *
409  * All callers have to hold the siglock.
410  */
411 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
412 {
413         int signr;
414
415         /* We only dequeue private signals from ourselves, we don't let
416          * signalfd steal them
417          */
418         signr = __dequeue_signal(&tsk->pending, mask, info);
419         if (!signr) {
420                 signr = __dequeue_signal(&tsk->signal->shared_pending,
421                                          mask, info);
422                 /*
423                  * itimer signal ?
424                  *
425                  * itimers are process shared and we restart periodic
426                  * itimers in the signal delivery path to prevent DoS
427                  * attacks in the high resolution timer case. This is
428                  * compliant with the old way of self restarting
429                  * itimers, as the SIGALRM is a legacy signal and only
430                  * queued once. Changing the restart behaviour to
431                  * restart the timer in the signal dequeue path is
432                  * reducing the timer noise on heavy loaded !highres
433                  * systems too.
434                  */
435                 if (unlikely(signr == SIGALRM)) {
436                         struct hrtimer *tmr = &tsk->signal->real_timer;
437
438                         if (!hrtimer_is_queued(tmr) &&
439                             tsk->signal->it_real_incr.tv64 != 0) {
440                                 hrtimer_forward(tmr, tmr->base->get_time(),
441                                                 tsk->signal->it_real_incr);
442                                 hrtimer_restart(tmr);
443                         }
444                 }
445         }
446
447         recalc_sigpending();
448         if (!signr)
449                 return 0;
450
451         if (unlikely(sig_kernel_stop(signr))) {
452                 /*
453                  * Set a marker that we have dequeued a stop signal.  Our
454                  * caller might release the siglock and then the pending
455                  * stop signal it is about to process is no longer in the
456                  * pending bitmasks, but must still be cleared by a SIGCONT
457                  * (and overruled by a SIGKILL).  So those cases clear this
458                  * shared flag after we've set it.  Note that this flag may
459                  * remain set after the signal we return is ignored or
460                  * handled.  That doesn't matter because its only purpose
461                  * is to alert stop-signal processing code when another
462                  * processor has come along and cleared the flag.
463                  */
464                 tsk->signal->flags |= SIGNAL_STOP_DEQUEUED;
465         }
466         if ((info->si_code & __SI_MASK) == __SI_TIMER && info->si_sys_private) {
467                 /*
468                  * Release the siglock to ensure proper locking order
469                  * of timer locks outside of siglocks.  Note, we leave
470                  * irqs disabled here, since the posix-timers code is
471                  * about to disable them again anyway.
472                  */
473                 spin_unlock(&tsk->sighand->siglock);
474                 do_schedule_next_timer(info);
475                 spin_lock(&tsk->sighand->siglock);
476         }
477         return signr;
478 }
479
480 /*
481  * Tell a process that it has a new active signal..
482  *
483  * NOTE! we rely on the previous spin_lock to
484  * lock interrupts for us! We can only be called with
485  * "siglock" held, and the local interrupt must
486  * have been disabled when that got acquired!
487  *
488  * No need to set need_resched since signal event passing
489  * goes through ->blocked
490  */
491 void signal_wake_up(struct task_struct *t, int resume)
492 {
493         unsigned int mask;
494
495         set_tsk_thread_flag(t, TIF_SIGPENDING);
496
497         /*
498          * For SIGKILL, we want to wake it up in the stopped/traced/killable
499          * case. We don't check t->state here because there is a race with it
500          * executing another processor and just now entering stopped state.
501          * By using wake_up_state, we ensure the process will wake up and
502          * handle its death signal.
503          */
504         mask = TASK_INTERRUPTIBLE;
505         if (resume)
506                 mask |= TASK_WAKEKILL;
507         if (!wake_up_state(t, mask))
508                 kick_process(t);
509 }
510
511 /*
512  * Remove signals in mask from the pending set and queue.
513  * Returns 1 if any signals were found.
514  *
515  * All callers must be holding the siglock.
516  *
517  * This version takes a sigset mask and looks at all signals,
518  * not just those in the first mask word.
519  */
520 static int rm_from_queue_full(sigset_t *mask, struct sigpending *s)
521 {
522         struct sigqueue *q, *n;
523         sigset_t m;
524
525         sigandsets(&m, mask, &s->signal);
526         if (sigisemptyset(&m))
527                 return 0;
528
529         signandsets(&s->signal, &s->signal, mask);
530         list_for_each_entry_safe(q, n, &s->list, list) {
531                 if (sigismember(mask, q->info.si_signo)) {
532                         list_del_init(&q->list);
533                         __sigqueue_free(q);
534                 }
535         }
536         return 1;
537 }
538 /*
539  * Remove signals in mask from the pending set and queue.
540  * Returns 1 if any signals were found.
541  *
542  * All callers must be holding the siglock.
543  */
544 static int rm_from_queue(unsigned long mask, struct sigpending *s)
545 {
546         struct sigqueue *q, *n;
547
548         if (!sigtestsetmask(&s->signal, mask))
549                 return 0;
550
551         sigdelsetmask(&s->signal, mask);
552         list_for_each_entry_safe(q, n, &s->list, list) {
553                 if (q->info.si_signo < SIGRTMIN &&
554                     (mask & sigmask(q->info.si_signo))) {
555                         list_del_init(&q->list);
556                         __sigqueue_free(q);
557                 }
558         }
559         return 1;
560 }
561
562 /*
563  * Bad permissions for sending the signal
564  */
565 static int check_kill_permission(int sig, struct siginfo *info,
566                                  struct task_struct *t)
567 {
568         struct pid *sid;
569         int error;
570
571         if (!valid_signal(sig))
572                 return -EINVAL;
573
574         if (info != SEND_SIG_NOINFO && (is_si_special(info) || SI_FROMKERNEL(info)))
575                 return 0;
576
577         error = audit_signal_info(sig, t); /* Let audit system see the signal */
578         if (error)
579                 return error;
580
581         if ((current->euid ^ t->suid) && (current->euid ^ t->uid) &&
582             (current->uid  ^ t->suid) && (current->uid  ^ t->uid) &&
583             !capable(CAP_KILL)) {
584                 switch (sig) {
585                 case SIGCONT:
586                         sid = task_session(t);
587                         /*
588                          * We don't return the error if sid == NULL. The
589                          * task was unhashed, the caller must notice this.
590                          */
591                         if (!sid || sid == task_session(current))
592                                 break;
593                 default:
594                         return -EPERM;
595                 }
596         }
597
598         return security_task_kill(t, info, sig, 0);
599 }
600
601 /*
602  * Handle magic process-wide effects of stop/continue signals. Unlike
603  * the signal actions, these happen immediately at signal-generation
604  * time regardless of blocking, ignoring, or handling.  This does the
605  * actual continuing for SIGCONT, but not the actual stopping for stop
606  * signals. The process stop is done as a signal action for SIG_DFL.
607  *
608  * Returns true if the signal should be actually delivered, otherwise
609  * it should be dropped.
610  */
611 static int prepare_signal(int sig, struct task_struct *p)
612 {
613         struct signal_struct *signal = p->signal;
614         struct task_struct *t;
615
616         if (unlikely(signal->flags & SIGNAL_GROUP_EXIT)) {
617                 /*
618                  * The process is in the middle of dying, nothing to do.
619                  */
620         } else if (sig_kernel_stop(sig)) {
621                 /*
622                  * This is a stop signal.  Remove SIGCONT from all queues.
623                  */
624                 rm_from_queue(sigmask(SIGCONT), &signal->shared_pending);
625                 t = p;
626                 do {
627                         rm_from_queue(sigmask(SIGCONT), &t->pending);
628                 } while_each_thread(p, t);
629         } else if (sig == SIGCONT) {
630                 unsigned int why;
631                 /*
632                  * Remove all stop signals from all queues,
633                  * and wake all threads.
634                  */
635                 rm_from_queue(SIG_KERNEL_STOP_MASK, &signal->shared_pending);
636                 t = p;
637                 do {
638                         unsigned int state;
639                         rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
640                         /*
641                          * If there is a handler for SIGCONT, we must make
642                          * sure that no thread returns to user mode before
643                          * we post the signal, in case it was the only
644                          * thread eligible to run the signal handler--then
645                          * it must not do anything between resuming and
646                          * running the handler.  With the TIF_SIGPENDING
647                          * flag set, the thread will pause and acquire the
648                          * siglock that we hold now and until we've queued
649                          * the pending signal.
650                          *
651                          * Wake up the stopped thread _after_ setting
652                          * TIF_SIGPENDING
653                          */
654                         state = __TASK_STOPPED;
655                         if (sig_user_defined(t, SIGCONT) && !sigismember(&t->blocked, SIGCONT)) {
656                                 set_tsk_thread_flag(t, TIF_SIGPENDING);
657                                 state |= TASK_INTERRUPTIBLE;
658                         }
659                         wake_up_state(t, state);
660                 } while_each_thread(p, t);
661
662                 /*
663                  * Notify the parent with CLD_CONTINUED if we were stopped.
664                  *
665                  * If we were in the middle of a group stop, we pretend it
666                  * was already finished, and then continued. Since SIGCHLD
667                  * doesn't queue we report only CLD_STOPPED, as if the next
668                  * CLD_CONTINUED was dropped.
669                  */
670                 why = 0;
671                 if (signal->flags & SIGNAL_STOP_STOPPED)
672                         why |= SIGNAL_CLD_CONTINUED;
673                 else if (signal->group_stop_count)
674                         why |= SIGNAL_CLD_STOPPED;
675
676                 if (why) {
677                         /*
678                          * The first thread which returns from finish_stop()
679                          * will take ->siglock, notice SIGNAL_CLD_MASK, and
680                          * notify its parent. See get_signal_to_deliver().
681                          */
682                         signal->flags = why | SIGNAL_STOP_CONTINUED;
683                         signal->group_stop_count = 0;
684                         signal->group_exit_code = 0;
685                 } else {
686                         /*
687                          * We are not stopped, but there could be a stop
688                          * signal in the middle of being processed after
689                          * being removed from the queue.  Clear that too.
690                          */
691                         signal->flags &= ~SIGNAL_STOP_DEQUEUED;
692                 }
693         }
694
695         return !sig_ignored(p, sig);
696 }
697
698 /*
699  * Test if P wants to take SIG.  After we've checked all threads with this,
700  * it's equivalent to finding no threads not blocking SIG.  Any threads not
701  * blocking SIG were ruled out because they are not running and already
702  * have pending signals.  Such threads will dequeue from the shared queue
703  * as soon as they're available, so putting the signal on the shared queue
704  * will be equivalent to sending it to one such thread.
705  */
706 static inline int wants_signal(int sig, struct task_struct *p)
707 {
708         if (sigismember(&p->blocked, sig))
709                 return 0;
710         if (p->flags & PF_EXITING)
711                 return 0;
712         if (sig == SIGKILL)
713                 return 1;
714         if (task_is_stopped_or_traced(p))
715                 return 0;
716         return task_curr(p) || !signal_pending(p);
717 }
718
719 static void complete_signal(int sig, struct task_struct *p, int group)
720 {
721         struct signal_struct *signal = p->signal;
722         struct task_struct *t;
723
724         /*
725          * Now find a thread we can wake up to take the signal off the queue.
726          *
727          * If the main thread wants the signal, it gets first crack.
728          * Probably the least surprising to the average bear.
729          */
730         if (wants_signal(sig, p))
731                 t = p;
732         else if (!group || thread_group_empty(p))
733                 /*
734                  * There is just one thread and it does not need to be woken.
735                  * It will dequeue unblocked signals before it runs again.
736                  */
737                 return;
738         else {
739                 /*
740                  * Otherwise try to find a suitable thread.
741                  */
742                 t = signal->curr_target;
743                 while (!wants_signal(sig, t)) {
744                         t = next_thread(t);
745                         if (t == signal->curr_target)
746                                 /*
747                                  * No thread needs to be woken.
748                                  * Any eligible threads will see
749                                  * the signal in the queue soon.
750                                  */
751                                 return;
752                 }
753                 signal->curr_target = t;
754         }
755
756         /*
757          * Found a killable thread.  If the signal will be fatal,
758          * then start taking the whole group down immediately.
759          */
760         if (sig_fatal(p, sig) &&
761             !(signal->flags & (SIGNAL_UNKILLABLE | SIGNAL_GROUP_EXIT)) &&
762             !sigismember(&t->real_blocked, sig) &&
763             (sig == SIGKILL ||
764              !tracehook_consider_fatal_signal(t, sig, SIG_DFL))) {
765                 /*
766                  * This signal will be fatal to the whole group.
767                  */
768                 if (!sig_kernel_coredump(sig)) {
769                         /*
770                          * Start a group exit and wake everybody up.
771                          * This way we don't have other threads
772                          * running and doing things after a slower
773                          * thread has the fatal signal pending.
774                          */
775                         signal->flags = SIGNAL_GROUP_EXIT;
776                         signal->group_exit_code = sig;
777                         signal->group_stop_count = 0;
778                         t = p;
779                         do {
780                                 sigaddset(&t->pending.signal, SIGKILL);
781                                 signal_wake_up(t, 1);
782                         } while_each_thread(p, t);
783                         return;
784                 }
785         }
786
787         /*
788          * The signal is already in the shared-pending queue.
789          * Tell the chosen thread to wake up and dequeue it.
790          */
791         signal_wake_up(t, sig == SIGKILL);
792         return;
793 }
794
795 static inline int legacy_queue(struct sigpending *signals, int sig)
796 {
797         return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
798 }
799
800 static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
801                         int group)
802 {
803         struct sigpending *pending;
804         struct sigqueue *q;
805
806         assert_spin_locked(&t->sighand->siglock);
807         if (!prepare_signal(sig, t))
808                 return 0;
809
810         pending = group ? &t->signal->shared_pending : &t->pending;
811         /*
812          * Short-circuit ignored signals and support queuing
813          * exactly one non-rt signal, so that we can get more
814          * detailed information about the cause of the signal.
815          */
816         if (legacy_queue(pending, sig))
817                 return 0;
818         /*
819          * fast-pathed signals for kernel-internal things like SIGSTOP
820          * or SIGKILL.
821          */
822         if (info == SEND_SIG_FORCED)
823                 goto out_set;
824
825         /* Real-time signals must be queued if sent by sigqueue, or
826            some other real-time mechanism.  It is implementation
827            defined whether kill() does so.  We attempt to do so, on
828            the principle of least surprise, but since kill is not
829            allowed to fail with EAGAIN when low on memory we just
830            make sure at least one signal gets delivered and don't
831            pass on the info struct.  */
832
833         q = __sigqueue_alloc(t, GFP_ATOMIC, (sig < SIGRTMIN &&
834                                              (is_si_special(info) ||
835                                               info->si_code >= 0)));
836         if (q) {
837                 list_add_tail(&q->list, &pending->list);
838                 switch ((unsigned long) info) {
839                 case (unsigned long) SEND_SIG_NOINFO:
840                         q->info.si_signo = sig;
841                         q->info.si_errno = 0;
842                         q->info.si_code = SI_USER;
843                         q->info.si_pid = task_pid_vnr(current);
844                         q->info.si_uid = current->uid;
845                         break;
846                 case (unsigned long) SEND_SIG_PRIV:
847                         q->info.si_signo = sig;
848                         q->info.si_errno = 0;
849                         q->info.si_code = SI_KERNEL;
850                         q->info.si_pid = 0;
851                         q->info.si_uid = 0;
852                         break;
853                 default:
854                         copy_siginfo(&q->info, info);
855                         break;
856                 }
857         } else if (!is_si_special(info)) {
858                 if (sig >= SIGRTMIN && info->si_code != SI_USER)
859                 /*
860                  * Queue overflow, abort.  We may abort if the signal was rt
861                  * and sent by user using something other than kill().
862                  */
863                         return -EAGAIN;
864         }
865
866 out_set:
867         signalfd_notify(t, sig);
868         sigaddset(&pending->signal, sig);
869         complete_signal(sig, t, group);
870         return 0;
871 }
872
873 int print_fatal_signals;
874
875 static void print_fatal_signal(struct pt_regs *regs, int signr)
876 {
877         printk("%s/%d: potentially unexpected fatal signal %d.\n",
878                 current->comm, task_pid_nr(current), signr);
879
880 #if defined(__i386__) && !defined(__arch_um__)
881         printk("code at %08lx: ", regs->ip);
882         {
883                 int i;
884                 for (i = 0; i < 16; i++) {
885                         unsigned char insn;
886
887                         __get_user(insn, (unsigned char *)(regs->ip + i));
888                         printk("%02x ", insn);
889                 }
890         }
891 #endif
892         printk("\n");
893         show_regs(regs);
894 }
895
896 static int __init setup_print_fatal_signals(char *str)
897 {
898         get_option (&str, &print_fatal_signals);
899
900         return 1;
901 }
902
903 __setup("print-fatal-signals=", setup_print_fatal_signals);
904
905 int
906 __group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
907 {
908         return send_signal(sig, info, p, 1);
909 }
910
911 static int
912 specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
913 {
914         return send_signal(sig, info, t, 0);
915 }
916
917 /*
918  * Force a signal that the process can't ignore: if necessary
919  * we unblock the signal and change any SIG_IGN to SIG_DFL.
920  *
921  * Note: If we unblock the signal, we always reset it to SIG_DFL,
922  * since we do not want to have a signal handler that was blocked
923  * be invoked when user space had explicitly blocked it.
924  *
925  * We don't want to have recursive SIGSEGV's etc, for example,
926  * that is why we also clear SIGNAL_UNKILLABLE.
927  */
928 int
929 force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
930 {
931         unsigned long int flags;
932         int ret, blocked, ignored;
933         struct k_sigaction *action;
934
935         spin_lock_irqsave(&t->sighand->siglock, flags);
936         action = &t->sighand->action[sig-1];
937         ignored = action->sa.sa_handler == SIG_IGN;
938         blocked = sigismember(&t->blocked, sig);
939         if (blocked || ignored) {
940                 action->sa.sa_handler = SIG_DFL;
941                 if (blocked) {
942                         sigdelset(&t->blocked, sig);
943                         recalc_sigpending_and_wake(t);
944                 }
945         }
946         if (action->sa.sa_handler == SIG_DFL)
947                 t->signal->flags &= ~SIGNAL_UNKILLABLE;
948         ret = specific_send_sig_info(sig, info, t);
949         spin_unlock_irqrestore(&t->sighand->siglock, flags);
950
951         return ret;
952 }
953
954 void
955 force_sig_specific(int sig, struct task_struct *t)
956 {
957         force_sig_info(sig, SEND_SIG_FORCED, t);
958 }
959
960 /*
961  * Nuke all other threads in the group.
962  */
963 void zap_other_threads(struct task_struct *p)
964 {
965         struct task_struct *t;
966
967         p->signal->group_stop_count = 0;
968
969         for (t = next_thread(p); t != p; t = next_thread(t)) {
970                 /*
971                  * Don't bother with already dead threads
972                  */
973                 if (t->exit_state)
974                         continue;
975
976                 /* SIGKILL will be handled before any pending SIGSTOP */
977                 sigaddset(&t->pending.signal, SIGKILL);
978                 signal_wake_up(t, 1);
979         }
980 }
981
982 int __fatal_signal_pending(struct task_struct *tsk)
983 {
984         return sigismember(&tsk->pending.signal, SIGKILL);
985 }
986 EXPORT_SYMBOL(__fatal_signal_pending);
987
988 struct sighand_struct *lock_task_sighand(struct task_struct *tsk, unsigned long *flags)
989 {
990         struct sighand_struct *sighand;
991
992         rcu_read_lock();
993         for (;;) {
994                 sighand = rcu_dereference(tsk->sighand);
995                 if (unlikely(sighand == NULL))
996                         break;
997
998                 spin_lock_irqsave(&sighand->siglock, *flags);
999                 if (likely(sighand == tsk->sighand))
1000                         break;
1001                 spin_unlock_irqrestore(&sighand->siglock, *flags);
1002         }
1003         rcu_read_unlock();
1004
1005         return sighand;
1006 }
1007
1008 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1009 {
1010         unsigned long flags;
1011         int ret;
1012
1013         ret = check_kill_permission(sig, info, p);
1014
1015         if (!ret && sig) {
1016                 ret = -ESRCH;
1017                 if (lock_task_sighand(p, &flags)) {
1018                         ret = __group_send_sig_info(sig, info, p);
1019                         unlock_task_sighand(p, &flags);
1020                 }
1021         }
1022
1023         return ret;
1024 }
1025
1026 /*
1027  * __kill_pgrp_info() sends a signal to a process group: this is what the tty
1028  * control characters do (^C, ^Z etc)
1029  */
1030
1031 int __kill_pgrp_info(int sig, struct siginfo *info, struct pid *pgrp)
1032 {
1033         struct task_struct *p = NULL;
1034         int retval, success;
1035
1036         success = 0;
1037         retval = -ESRCH;
1038         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
1039                 int err = group_send_sig_info(sig, info, p);
1040                 success |= !err;
1041                 retval = err;
1042         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
1043         return success ? 0 : retval;
1044 }
1045
1046 int kill_pid_info(int sig, struct siginfo *info, struct pid *pid)
1047 {
1048         int error = -ESRCH;
1049         struct task_struct *p;
1050
1051         rcu_read_lock();
1052 retry:
1053         p = pid_task(pid, PIDTYPE_PID);
1054         if (p) {
1055                 error = group_send_sig_info(sig, info, p);
1056                 if (unlikely(error == -ESRCH))
1057                         /*
1058                          * The task was unhashed in between, try again.
1059                          * If it is dead, pid_task() will return NULL,
1060                          * if we race with de_thread() it will find the
1061                          * new leader.
1062                          */
1063                         goto retry;
1064         }
1065         rcu_read_unlock();
1066
1067         return error;
1068 }
1069
1070 int
1071 kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1072 {
1073         int error;
1074         rcu_read_lock();
1075         error = kill_pid_info(sig, info, find_vpid(pid));
1076         rcu_read_unlock();
1077         return error;
1078 }
1079
1080 /* like kill_pid_info(), but doesn't use uid/euid of "current" */
1081 int kill_pid_info_as_uid(int sig, struct siginfo *info, struct pid *pid,
1082                       uid_t uid, uid_t euid, u32 secid)
1083 {
1084         int ret = -EINVAL;
1085         struct task_struct *p;
1086
1087         if (!valid_signal(sig))
1088                 return ret;
1089
1090         read_lock(&tasklist_lock);
1091         p = pid_task(pid, PIDTYPE_PID);
1092         if (!p) {
1093                 ret = -ESRCH;
1094                 goto out_unlock;
1095         }
1096         if ((info == SEND_SIG_NOINFO || (!is_si_special(info) && SI_FROMUSER(info)))
1097             && (euid != p->suid) && (euid != p->uid)
1098             && (uid != p->suid) && (uid != p->uid)) {
1099                 ret = -EPERM;
1100                 goto out_unlock;
1101         }
1102         ret = security_task_kill(p, info, sig, secid);
1103         if (ret)
1104                 goto out_unlock;
1105         if (sig && p->sighand) {
1106                 unsigned long flags;
1107                 spin_lock_irqsave(&p->sighand->siglock, flags);
1108                 ret = __group_send_sig_info(sig, info, p);
1109                 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1110         }
1111 out_unlock:
1112         read_unlock(&tasklist_lock);
1113         return ret;
1114 }
1115 EXPORT_SYMBOL_GPL(kill_pid_info_as_uid);
1116
1117 /*
1118  * kill_something_info() interprets pid in interesting ways just like kill(2).
1119  *
1120  * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1121  * is probably wrong.  Should make it like BSD or SYSV.
1122  */
1123
1124 static int kill_something_info(int sig, struct siginfo *info, pid_t pid)
1125 {
1126         int ret;
1127
1128         if (pid > 0) {
1129                 rcu_read_lock();
1130                 ret = kill_pid_info(sig, info, find_vpid(pid));
1131                 rcu_read_unlock();
1132                 return ret;
1133         }
1134
1135         read_lock(&tasklist_lock);
1136         if (pid != -1) {
1137                 ret = __kill_pgrp_info(sig, info,
1138                                 pid ? find_vpid(-pid) : task_pgrp(current));
1139         } else {
1140                 int retval = 0, count = 0;
1141                 struct task_struct * p;
1142
1143                 for_each_process(p) {
1144                         if (task_pid_vnr(p) > 1 &&
1145                                         !same_thread_group(p, current)) {
1146                                 int err = group_send_sig_info(sig, info, p);
1147                                 ++count;
1148                                 if (err != -EPERM)
1149                                         retval = err;
1150                         }
1151                 }
1152                 ret = count ? retval : -ESRCH;
1153         }
1154         read_unlock(&tasklist_lock);
1155
1156         return ret;
1157 }
1158
1159 /*
1160  * These are for backward compatibility with the rest of the kernel source.
1161  */
1162
1163 /*
1164  * The caller must ensure the task can't exit.
1165  */
1166 int
1167 send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1168 {
1169         int ret;
1170         unsigned long flags;
1171
1172         /*
1173          * Make sure legacy kernel users don't send in bad values
1174          * (normal paths check this in check_kill_permission).
1175          */
1176         if (!valid_signal(sig))
1177                 return -EINVAL;
1178
1179         spin_lock_irqsave(&p->sighand->siglock, flags);
1180         ret = specific_send_sig_info(sig, info, p);
1181         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1182         return ret;
1183 }
1184
1185 #define __si_special(priv) \
1186         ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1187
1188 int
1189 send_sig(int sig, struct task_struct *p, int priv)
1190 {
1191         return send_sig_info(sig, __si_special(priv), p);
1192 }
1193
1194 void
1195 force_sig(int sig, struct task_struct *p)
1196 {
1197         force_sig_info(sig, SEND_SIG_PRIV, p);
1198 }
1199
1200 /*
1201  * When things go south during signal handling, we
1202  * will force a SIGSEGV. And if the signal that caused
1203  * the problem was already a SIGSEGV, we'll want to
1204  * make sure we don't even try to deliver the signal..
1205  */
1206 int
1207 force_sigsegv(int sig, struct task_struct *p)
1208 {
1209         if (sig == SIGSEGV) {
1210                 unsigned long flags;
1211                 spin_lock_irqsave(&p->sighand->siglock, flags);
1212                 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1213                 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1214         }
1215         force_sig(SIGSEGV, p);
1216         return 0;
1217 }
1218
1219 int kill_pgrp(struct pid *pid, int sig, int priv)
1220 {
1221         int ret;
1222
1223         read_lock(&tasklist_lock);
1224         ret = __kill_pgrp_info(sig, __si_special(priv), pid);
1225         read_unlock(&tasklist_lock);
1226
1227         return ret;
1228 }
1229 EXPORT_SYMBOL(kill_pgrp);
1230
1231 int kill_pid(struct pid *pid, int sig, int priv)
1232 {
1233         return kill_pid_info(sig, __si_special(priv), pid);
1234 }
1235 EXPORT_SYMBOL(kill_pid);
1236
1237 /*
1238  * These functions support sending signals using preallocated sigqueue
1239  * structures.  This is needed "because realtime applications cannot
1240  * afford to lose notifications of asynchronous events, like timer
1241  * expirations or I/O completions".  In the case of Posix Timers 
1242  * we allocate the sigqueue structure from the timer_create.  If this
1243  * allocation fails we are able to report the failure to the application
1244  * with an EAGAIN error.
1245  */
1246  
1247 struct sigqueue *sigqueue_alloc(void)
1248 {
1249         struct sigqueue *q;
1250
1251         if ((q = __sigqueue_alloc(current, GFP_KERNEL, 0)))
1252                 q->flags |= SIGQUEUE_PREALLOC;
1253         return(q);
1254 }
1255
1256 void sigqueue_free(struct sigqueue *q)
1257 {
1258         unsigned long flags;
1259         spinlock_t *lock = &current->sighand->siglock;
1260
1261         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1262         /*
1263          * We must hold ->siglock while testing q->list
1264          * to serialize with collect_signal() or with
1265          * __exit_signal()->flush_sigqueue().
1266          */
1267         spin_lock_irqsave(lock, flags);
1268         q->flags &= ~SIGQUEUE_PREALLOC;
1269         /*
1270          * If it is queued it will be freed when dequeued,
1271          * like the "regular" sigqueue.
1272          */
1273         if (!list_empty(&q->list))
1274                 q = NULL;
1275         spin_unlock_irqrestore(lock, flags);
1276
1277         if (q)
1278                 __sigqueue_free(q);
1279 }
1280
1281 int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group)
1282 {
1283         int sig = q->info.si_signo;
1284         struct sigpending *pending;
1285         unsigned long flags;
1286         int ret;
1287
1288         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1289
1290         ret = -1;
1291         if (!likely(lock_task_sighand(t, &flags)))
1292                 goto ret;
1293
1294         ret = 1; /* the signal is ignored */
1295         if (!prepare_signal(sig, t))
1296                 goto out;
1297
1298         ret = 0;
1299         if (unlikely(!list_empty(&q->list))) {
1300                 /*
1301                  * If an SI_TIMER entry is already queue just increment
1302                  * the overrun count.
1303                  */
1304                 BUG_ON(q->info.si_code != SI_TIMER);
1305                 q->info.si_overrun++;
1306                 goto out;
1307         }
1308         q->info.si_overrun = 0;
1309
1310         signalfd_notify(t, sig);
1311         pending = group ? &t->signal->shared_pending : &t->pending;
1312         list_add_tail(&q->list, &pending->list);
1313         sigaddset(&pending->signal, sig);
1314         complete_signal(sig, t, group);
1315 out:
1316         unlock_task_sighand(t, &flags);
1317 ret:
1318         return ret;
1319 }
1320
1321 /*
1322  * Wake up any threads in the parent blocked in wait* syscalls.
1323  */
1324 static inline void __wake_up_parent(struct task_struct *p,
1325                                     struct task_struct *parent)
1326 {
1327         wake_up_interruptible_sync(&parent->signal->wait_chldexit);
1328 }
1329
1330 /*
1331  * Let a parent know about the death of a child.
1332  * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1333  *
1334  * Returns -1 if our parent ignored us and so we've switched to
1335  * self-reaping, or else @sig.
1336  */
1337 int do_notify_parent(struct task_struct *tsk, int sig)
1338 {
1339         struct siginfo info;
1340         unsigned long flags;
1341         struct sighand_struct *psig;
1342         int ret = sig;
1343
1344         BUG_ON(sig == -1);
1345
1346         /* do_notify_parent_cldstop should have been called instead.  */
1347         BUG_ON(task_is_stopped_or_traced(tsk));
1348
1349         BUG_ON(!tsk->ptrace &&
1350                (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1351
1352         info.si_signo = sig;
1353         info.si_errno = 0;
1354         /*
1355          * we are under tasklist_lock here so our parent is tied to
1356          * us and cannot exit and release its namespace.
1357          *
1358          * the only it can is to switch its nsproxy with sys_unshare,
1359          * bu uncharing pid namespaces is not allowed, so we'll always
1360          * see relevant namespace
1361          *
1362          * write_lock() currently calls preempt_disable() which is the
1363          * same as rcu_read_lock(), but according to Oleg, this is not
1364          * correct to rely on this
1365          */
1366         rcu_read_lock();
1367         info.si_pid = task_pid_nr_ns(tsk, tsk->parent->nsproxy->pid_ns);
1368         rcu_read_unlock();
1369
1370         info.si_uid = tsk->uid;
1371
1372         info.si_utime = cputime_to_clock_t(cputime_add(tsk->utime,
1373                                                        tsk->signal->utime));
1374         info.si_stime = cputime_to_clock_t(cputime_add(tsk->stime,
1375                                                        tsk->signal->stime));
1376
1377         info.si_status = tsk->exit_code & 0x7f;
1378         if (tsk->exit_code & 0x80)
1379                 info.si_code = CLD_DUMPED;
1380         else if (tsk->exit_code & 0x7f)
1381                 info.si_code = CLD_KILLED;
1382         else {
1383                 info.si_code = CLD_EXITED;
1384                 info.si_status = tsk->exit_code >> 8;
1385         }
1386
1387         psig = tsk->parent->sighand;
1388         spin_lock_irqsave(&psig->siglock, flags);
1389         if (!tsk->ptrace && sig == SIGCHLD &&
1390             (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1391              (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1392                 /*
1393                  * We are exiting and our parent doesn't care.  POSIX.1
1394                  * defines special semantics for setting SIGCHLD to SIG_IGN
1395                  * or setting the SA_NOCLDWAIT flag: we should be reaped
1396                  * automatically and not left for our parent's wait4 call.
1397                  * Rather than having the parent do it as a magic kind of
1398                  * signal handler, we just set this to tell do_exit that we
1399                  * can be cleaned up without becoming a zombie.  Note that
1400                  * we still call __wake_up_parent in this case, because a
1401                  * blocked sys_wait4 might now return -ECHILD.
1402                  *
1403                  * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1404                  * is implementation-defined: we do (if you don't want
1405                  * it, just use SIG_IGN instead).
1406                  */
1407                 ret = tsk->exit_signal = -1;
1408                 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1409                         sig = -1;
1410         }
1411         if (valid_signal(sig) && sig > 0)
1412                 __group_send_sig_info(sig, &info, tsk->parent);
1413         __wake_up_parent(tsk, tsk->parent);
1414         spin_unlock_irqrestore(&psig->siglock, flags);
1415
1416         return ret;
1417 }
1418
1419 static void do_notify_parent_cldstop(struct task_struct *tsk, int why)
1420 {
1421         struct siginfo info;
1422         unsigned long flags;
1423         struct task_struct *parent;
1424         struct sighand_struct *sighand;
1425
1426         if (tsk->ptrace & PT_PTRACED)
1427                 parent = tsk->parent;
1428         else {
1429                 tsk = tsk->group_leader;
1430                 parent = tsk->real_parent;
1431         }
1432
1433         info.si_signo = SIGCHLD;
1434         info.si_errno = 0;
1435         /*
1436          * see comment in do_notify_parent() abot the following 3 lines
1437          */
1438         rcu_read_lock();
1439         info.si_pid = task_pid_nr_ns(tsk, tsk->parent->nsproxy->pid_ns);
1440         rcu_read_unlock();
1441
1442         info.si_uid = tsk->uid;
1443
1444         info.si_utime = cputime_to_clock_t(tsk->utime);
1445         info.si_stime = cputime_to_clock_t(tsk->stime);
1446
1447         info.si_code = why;
1448         switch (why) {
1449         case CLD_CONTINUED:
1450                 info.si_status = SIGCONT;
1451                 break;
1452         case CLD_STOPPED:
1453                 info.si_status = tsk->signal->group_exit_code & 0x7f;
1454                 break;
1455         case CLD_TRAPPED:
1456                 info.si_status = tsk->exit_code & 0x7f;
1457                 break;
1458         default:
1459                 BUG();
1460         }
1461
1462         sighand = parent->sighand;
1463         spin_lock_irqsave(&sighand->siglock, flags);
1464         if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1465             !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1466                 __group_send_sig_info(SIGCHLD, &info, parent);
1467         /*
1468          * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1469          */
1470         __wake_up_parent(tsk, parent);
1471         spin_unlock_irqrestore(&sighand->siglock, flags);
1472 }
1473
1474 static inline int may_ptrace_stop(void)
1475 {
1476         if (!likely(current->ptrace & PT_PTRACED))
1477                 return 0;
1478         /*
1479          * Are we in the middle of do_coredump?
1480          * If so and our tracer is also part of the coredump stopping
1481          * is a deadlock situation, and pointless because our tracer
1482          * is dead so don't allow us to stop.
1483          * If SIGKILL was already sent before the caller unlocked
1484          * ->siglock we must see ->core_state != NULL. Otherwise it
1485          * is safe to enter schedule().
1486          */
1487         if (unlikely(current->mm->core_state) &&
1488             unlikely(current->mm == current->parent->mm))
1489                 return 0;
1490
1491         return 1;
1492 }
1493
1494 /*
1495  * Return nonzero if there is a SIGKILL that should be waking us up.
1496  * Called with the siglock held.
1497  */
1498 static int sigkill_pending(struct task_struct *tsk)
1499 {
1500         return  sigismember(&tsk->pending.signal, SIGKILL) ||
1501                 sigismember(&tsk->signal->shared_pending.signal, SIGKILL);
1502 }
1503
1504 /*
1505  * This must be called with current->sighand->siglock held.
1506  *
1507  * This should be the path for all ptrace stops.
1508  * We always set current->last_siginfo while stopped here.
1509  * That makes it a way to test a stopped process for
1510  * being ptrace-stopped vs being job-control-stopped.
1511  *
1512  * If we actually decide not to stop at all because the tracer
1513  * is gone, we keep current->exit_code unless clear_code.
1514  */
1515 static void ptrace_stop(int exit_code, int clear_code, siginfo_t *info)
1516 {
1517         if (arch_ptrace_stop_needed(exit_code, info)) {
1518                 /*
1519                  * The arch code has something special to do before a
1520                  * ptrace stop.  This is allowed to block, e.g. for faults
1521                  * on user stack pages.  We can't keep the siglock while
1522                  * calling arch_ptrace_stop, so we must release it now.
1523                  * To preserve proper semantics, we must do this before
1524                  * any signal bookkeeping like checking group_stop_count.
1525                  * Meanwhile, a SIGKILL could come in before we retake the
1526                  * siglock.  That must prevent us from sleeping in TASK_TRACED.
1527                  * So after regaining the lock, we must check for SIGKILL.
1528                  */
1529                 spin_unlock_irq(&current->sighand->siglock);
1530                 arch_ptrace_stop(exit_code, info);
1531                 spin_lock_irq(&current->sighand->siglock);
1532                 if (sigkill_pending(current))
1533                         return;
1534         }
1535
1536         /*
1537          * If there is a group stop in progress,
1538          * we must participate in the bookkeeping.
1539          */
1540         if (current->signal->group_stop_count > 0)
1541                 --current->signal->group_stop_count;
1542
1543         current->last_siginfo = info;
1544         current->exit_code = exit_code;
1545
1546         /* Let the debugger run.  */
1547         __set_current_state(TASK_TRACED);
1548         spin_unlock_irq(&current->sighand->siglock);
1549         read_lock(&tasklist_lock);
1550         if (may_ptrace_stop()) {
1551                 do_notify_parent_cldstop(current, CLD_TRAPPED);
1552                 /*
1553                  * Don't want to allow preemption here, because
1554                  * sys_ptrace() needs this task to be inactive.
1555                  *
1556                  * XXX: implement read_unlock_no_resched().
1557                  */
1558                 preempt_disable();
1559                 read_unlock(&tasklist_lock);
1560                 preempt_enable_no_resched();
1561                 schedule();
1562         } else {
1563                 /*
1564                  * By the time we got the lock, our tracer went away.
1565                  * Don't drop the lock yet, another tracer may come.
1566                  */
1567                 __set_current_state(TASK_RUNNING);
1568                 if (clear_code)
1569                         current->exit_code = 0;
1570                 read_unlock(&tasklist_lock);
1571         }
1572
1573         /*
1574          * While in TASK_TRACED, we were considered "frozen enough".
1575          * Now that we woke up, it's crucial if we're supposed to be
1576          * frozen that we freeze now before running anything substantial.
1577          */
1578         try_to_freeze();
1579
1580         /*
1581          * We are back.  Now reacquire the siglock before touching
1582          * last_siginfo, so that we are sure to have synchronized with
1583          * any signal-sending on another CPU that wants to examine it.
1584          */
1585         spin_lock_irq(&current->sighand->siglock);
1586         current->last_siginfo = NULL;
1587
1588         /*
1589          * Queued signals ignored us while we were stopped for tracing.
1590          * So check for any that we should take before resuming user mode.
1591          * This sets TIF_SIGPENDING, but never clears it.
1592          */
1593         recalc_sigpending_tsk(current);
1594 }
1595
1596 void ptrace_notify(int exit_code)
1597 {
1598         siginfo_t info;
1599
1600         BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1601
1602         memset(&info, 0, sizeof info);
1603         info.si_signo = SIGTRAP;
1604         info.si_code = exit_code;
1605         info.si_pid = task_pid_vnr(current);
1606         info.si_uid = current->uid;
1607
1608         /* Let the debugger run.  */
1609         spin_lock_irq(&current->sighand->siglock);
1610         ptrace_stop(exit_code, 1, &info);
1611         spin_unlock_irq(&current->sighand->siglock);
1612 }
1613
1614 static void
1615 finish_stop(int stop_count)
1616 {
1617         /*
1618          * If there are no other threads in the group, or if there is
1619          * a group stop in progress and we are the last to stop,
1620          * report to the parent.  When ptraced, every thread reports itself.
1621          */
1622         if (tracehook_notify_jctl(stop_count == 0, CLD_STOPPED)) {
1623                 read_lock(&tasklist_lock);
1624                 do_notify_parent_cldstop(current, CLD_STOPPED);
1625                 read_unlock(&tasklist_lock);
1626         }
1627
1628         do {
1629                 schedule();
1630         } while (try_to_freeze());
1631         /*
1632          * Now we don't run again until continued.
1633          */
1634         current->exit_code = 0;
1635 }
1636
1637 /*
1638  * This performs the stopping for SIGSTOP and other stop signals.
1639  * We have to stop all threads in the thread group.
1640  * Returns nonzero if we've actually stopped and released the siglock.
1641  * Returns zero if we didn't stop and still hold the siglock.
1642  */
1643 static int do_signal_stop(int signr)
1644 {
1645         struct signal_struct *sig = current->signal;
1646         int stop_count;
1647
1648         if (sig->group_stop_count > 0) {
1649                 /*
1650                  * There is a group stop in progress.  We don't need to
1651                  * start another one.
1652                  */
1653                 stop_count = --sig->group_stop_count;
1654         } else {
1655                 struct task_struct *t;
1656
1657                 if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED) ||
1658                     unlikely(signal_group_exit(sig)))
1659                         return 0;
1660                 /*
1661                  * There is no group stop already in progress.
1662                  * We must initiate one now.
1663                  */
1664                 sig->group_exit_code = signr;
1665
1666                 stop_count = 0;
1667                 for (t = next_thread(current); t != current; t = next_thread(t))
1668                         /*
1669                          * Setting state to TASK_STOPPED for a group
1670                          * stop is always done with the siglock held,
1671                          * so this check has no races.
1672                          */
1673                         if (!(t->flags & PF_EXITING) &&
1674                             !task_is_stopped_or_traced(t)) {
1675                                 stop_count++;
1676                                 signal_wake_up(t, 0);
1677                         }
1678                 sig->group_stop_count = stop_count;
1679         }
1680
1681         if (stop_count == 0)
1682                 sig->flags = SIGNAL_STOP_STOPPED;
1683         current->exit_code = sig->group_exit_code;
1684         __set_current_state(TASK_STOPPED);
1685
1686         spin_unlock_irq(&current->sighand->siglock);
1687         finish_stop(stop_count);
1688         return 1;
1689 }
1690
1691 static int ptrace_signal(int signr, siginfo_t *info,
1692                          struct pt_regs *regs, void *cookie)
1693 {
1694         if (!(current->ptrace & PT_PTRACED))
1695                 return signr;
1696
1697         ptrace_signal_deliver(regs, cookie);
1698
1699         /* Let the debugger run.  */
1700         ptrace_stop(signr, 0, info);
1701
1702         /* We're back.  Did the debugger cancel the sig?  */
1703         signr = current->exit_code;
1704         if (signr == 0)
1705                 return signr;
1706
1707         current->exit_code = 0;
1708
1709         /* Update the siginfo structure if the signal has
1710            changed.  If the debugger wanted something
1711            specific in the siginfo structure then it should
1712            have updated *info via PTRACE_SETSIGINFO.  */
1713         if (signr != info->si_signo) {
1714                 info->si_signo = signr;
1715                 info->si_errno = 0;
1716                 info->si_code = SI_USER;
1717                 info->si_pid = task_pid_vnr(current->parent);
1718                 info->si_uid = current->parent->uid;
1719         }
1720
1721         /* If the (new) signal is now blocked, requeue it.  */
1722         if (sigismember(&current->blocked, signr)) {
1723                 specific_send_sig_info(signr, info, current);
1724                 signr = 0;
1725         }
1726
1727         return signr;
1728 }
1729
1730 int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
1731                           struct pt_regs *regs, void *cookie)
1732 {
1733         struct sighand_struct *sighand = current->sighand;
1734         struct signal_struct *signal = current->signal;
1735         int signr;
1736
1737 relock:
1738         /*
1739          * We'll jump back here after any time we were stopped in TASK_STOPPED.
1740          * While in TASK_STOPPED, we were considered "frozen enough".
1741          * Now that we woke up, it's crucial if we're supposed to be
1742          * frozen that we freeze now before running anything substantial.
1743          */
1744         try_to_freeze();
1745
1746         spin_lock_irq(&sighand->siglock);
1747         /*
1748          * Every stopped thread goes here after wakeup. Check to see if
1749          * we should notify the parent, prepare_signal(SIGCONT) encodes
1750          * the CLD_ si_code into SIGNAL_CLD_MASK bits.
1751          */
1752         if (unlikely(signal->flags & SIGNAL_CLD_MASK)) {
1753                 int why = (signal->flags & SIGNAL_STOP_CONTINUED)
1754                                 ? CLD_CONTINUED : CLD_STOPPED;
1755                 signal->flags &= ~SIGNAL_CLD_MASK;
1756                 spin_unlock_irq(&sighand->siglock);
1757
1758                 if (unlikely(!tracehook_notify_jctl(1, why)))
1759                         goto relock;
1760
1761                 read_lock(&tasklist_lock);
1762                 do_notify_parent_cldstop(current->group_leader, why);
1763                 read_unlock(&tasklist_lock);
1764                 goto relock;
1765         }
1766
1767         for (;;) {
1768                 struct k_sigaction *ka;
1769
1770                 if (unlikely(signal->group_stop_count > 0) &&
1771                     do_signal_stop(0))
1772                         goto relock;
1773
1774                 /*
1775                  * Tracing can induce an artifical signal and choose sigaction.
1776                  * The return value in @signr determines the default action,
1777                  * but @info->si_signo is the signal number we will report.
1778                  */
1779                 signr = tracehook_get_signal(current, regs, info, return_ka);
1780                 if (unlikely(signr < 0))
1781                         goto relock;
1782                 if (unlikely(signr != 0))
1783                         ka = return_ka;
1784                 else {
1785                         signr = dequeue_signal(current, &current->blocked,
1786                                                info);
1787
1788                         if (!signr)
1789                                 break; /* will return 0 */
1790
1791                         if (signr != SIGKILL) {
1792                                 signr = ptrace_signal(signr, info,
1793                                                       regs, cookie);
1794                                 if (!signr)
1795                                         continue;
1796                         }
1797
1798                         ka = &sighand->action[signr-1];
1799                 }
1800
1801                 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
1802                         continue;
1803                 if (ka->sa.sa_handler != SIG_DFL) {
1804                         /* Run the handler.  */
1805                         *return_ka = *ka;
1806
1807                         if (ka->sa.sa_flags & SA_ONESHOT)
1808                                 ka->sa.sa_handler = SIG_DFL;
1809
1810                         break; /* will return non-zero "signr" value */
1811                 }
1812
1813                 /*
1814                  * Now we are doing the default action for this signal.
1815                  */
1816                 if (sig_kernel_ignore(signr)) /* Default is nothing. */
1817                         continue;
1818
1819                 /*
1820                  * Global init gets no signals it doesn't want.
1821                  */
1822                 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
1823                     !signal_group_exit(signal))
1824                         continue;
1825
1826                 if (sig_kernel_stop(signr)) {
1827                         /*
1828                          * The default action is to stop all threads in
1829                          * the thread group.  The job control signals
1830                          * do nothing in an orphaned pgrp, but SIGSTOP
1831                          * always works.  Note that siglock needs to be
1832                          * dropped during the call to is_orphaned_pgrp()
1833                          * because of lock ordering with tasklist_lock.
1834                          * This allows an intervening SIGCONT to be posted.
1835                          * We need to check for that and bail out if necessary.
1836                          */
1837                         if (signr != SIGSTOP) {
1838                                 spin_unlock_irq(&sighand->siglock);
1839
1840                                 /* signals can be posted during this window */
1841
1842                                 if (is_current_pgrp_orphaned())
1843                                         goto relock;
1844
1845                                 spin_lock_irq(&sighand->siglock);
1846                         }
1847
1848                         if (likely(do_signal_stop(info->si_signo))) {
1849                                 /* It released the siglock.  */
1850                                 goto relock;
1851                         }
1852
1853                         /*
1854                          * We didn't actually stop, due to a race
1855                          * with SIGCONT or something like that.
1856                          */
1857                         continue;
1858                 }
1859
1860                 spin_unlock_irq(&sighand->siglock);
1861
1862                 /*
1863                  * Anything else is fatal, maybe with a core dump.
1864                  */
1865                 current->flags |= PF_SIGNALED;
1866
1867                 if (sig_kernel_coredump(signr)) {
1868                         if (print_fatal_signals)
1869                                 print_fatal_signal(regs, info->si_signo);
1870                         /*
1871                          * If it was able to dump core, this kills all
1872                          * other threads in the group and synchronizes with
1873                          * their demise.  If we lost the race with another
1874                          * thread getting here, it set group_exit_code
1875                          * first and our do_group_exit call below will use
1876                          * that value and ignore the one we pass it.
1877                          */
1878                         do_coredump(info->si_signo, info->si_signo, regs);
1879                 }
1880
1881                 /*
1882                  * Death signals, no core dump.
1883                  */
1884                 do_group_exit(info->si_signo);
1885                 /* NOTREACHED */
1886         }
1887         spin_unlock_irq(&sighand->siglock);
1888         return signr;
1889 }
1890
1891 void exit_signals(struct task_struct *tsk)
1892 {
1893         int group_stop = 0;
1894         struct task_struct *t;
1895
1896         if (thread_group_empty(tsk) || signal_group_exit(tsk->signal)) {
1897                 tsk->flags |= PF_EXITING;
1898                 return;
1899         }
1900
1901         spin_lock_irq(&tsk->sighand->siglock);
1902         /*
1903          * From now this task is not visible for group-wide signals,
1904          * see wants_signal(), do_signal_stop().
1905          */
1906         tsk->flags |= PF_EXITING;
1907         if (!signal_pending(tsk))
1908                 goto out;
1909
1910         /* It could be that __group_complete_signal() choose us to
1911          * notify about group-wide signal. Another thread should be
1912          * woken now to take the signal since we will not.
1913          */
1914         for (t = tsk; (t = next_thread(t)) != tsk; )
1915                 if (!signal_pending(t) && !(t->flags & PF_EXITING))
1916                         recalc_sigpending_and_wake(t);
1917
1918         if (unlikely(tsk->signal->group_stop_count) &&
1919                         !--tsk->signal->group_stop_count) {
1920                 tsk->signal->flags = SIGNAL_STOP_STOPPED;
1921                 group_stop = 1;
1922         }
1923 out:
1924         spin_unlock_irq(&tsk->sighand->siglock);
1925
1926         if (unlikely(group_stop) && tracehook_notify_jctl(1, CLD_STOPPED)) {
1927                 read_lock(&tasklist_lock);
1928                 do_notify_parent_cldstop(tsk, CLD_STOPPED);
1929                 read_unlock(&tasklist_lock);
1930         }
1931 }
1932
1933 EXPORT_SYMBOL(recalc_sigpending);
1934 EXPORT_SYMBOL_GPL(dequeue_signal);
1935 EXPORT_SYMBOL(flush_signals);
1936 EXPORT_SYMBOL(force_sig);
1937 EXPORT_SYMBOL(send_sig);
1938 EXPORT_SYMBOL(send_sig_info);
1939 EXPORT_SYMBOL(sigprocmask);
1940 EXPORT_SYMBOL(block_all_signals);
1941 EXPORT_SYMBOL(unblock_all_signals);
1942
1943
1944 /*
1945  * System call entry points.
1946  */
1947
1948 SYSCALL_DEFINE0(restart_syscall)
1949 {
1950         struct restart_block *restart = &current_thread_info()->restart_block;
1951         return restart->fn(restart);
1952 }
1953
1954 long do_no_restart_syscall(struct restart_block *param)
1955 {
1956         return -EINTR;
1957 }
1958
1959 /*
1960  * We don't need to get the kernel lock - this is all local to this
1961  * particular thread.. (and that's good, because this is _heavily_
1962  * used by various programs)
1963  */
1964
1965 /*
1966  * This is also useful for kernel threads that want to temporarily
1967  * (or permanently) block certain signals.
1968  *
1969  * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
1970  * interface happily blocks "unblockable" signals like SIGKILL
1971  * and friends.
1972  */
1973 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
1974 {
1975         int error;
1976
1977         spin_lock_irq(&current->sighand->siglock);
1978         if (oldset)
1979                 *oldset = current->blocked;
1980
1981         error = 0;
1982         switch (how) {
1983         case SIG_BLOCK:
1984                 sigorsets(&current->blocked, &current->blocked, set);
1985                 break;
1986         case SIG_UNBLOCK:
1987                 signandsets(&current->blocked, &current->blocked, set);
1988                 break;
1989         case SIG_SETMASK:
1990                 current->blocked = *set;
1991                 break;
1992         default:
1993                 error = -EINVAL;
1994         }
1995         recalc_sigpending();
1996         spin_unlock_irq(&current->sighand->siglock);
1997
1998         return error;
1999 }
2000
2001 SYSCALL_DEFINE4(rt_sigprocmask, int, how, sigset_t __user *, set,
2002                 sigset_t __user *, oset, size_t, sigsetsize)
2003 {
2004         int error = -EINVAL;
2005         sigset_t old_set, new_set;
2006
2007         /* XXX: Don't preclude handling different sized sigset_t's.  */
2008         if (sigsetsize != sizeof(sigset_t))
2009                 goto out;
2010
2011         if (set) {
2012                 error = -EFAULT;
2013                 if (copy_from_user(&new_set, set, sizeof(*set)))
2014                         goto out;
2015                 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2016
2017                 error = sigprocmask(how, &new_set, &old_set);
2018                 if (error)
2019                         goto out;
2020                 if (oset)
2021                         goto set_old;
2022         } else if (oset) {
2023                 spin_lock_irq(&current->sighand->siglock);
2024                 old_set = current->blocked;
2025                 spin_unlock_irq(&current->sighand->siglock);
2026
2027         set_old:
2028                 error = -EFAULT;
2029                 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2030                         goto out;
2031         }
2032         error = 0;
2033 out:
2034         return error;
2035 }
2036
2037 long do_sigpending(void __user *set, unsigned long sigsetsize)
2038 {
2039         long error = -EINVAL;
2040         sigset_t pending;
2041
2042         if (sigsetsize > sizeof(sigset_t))
2043                 goto out;
2044
2045         spin_lock_irq(&current->sighand->siglock);
2046         sigorsets(&pending, &current->pending.signal,
2047                   &current->signal->shared_pending.signal);
2048         spin_unlock_irq(&current->sighand->siglock);
2049
2050         /* Outside the lock because only this thread touches it.  */
2051         sigandsets(&pending, &current->blocked, &pending);
2052
2053         error = -EFAULT;
2054         if (!copy_to_user(set, &pending, sigsetsize))
2055                 error = 0;
2056
2057 out:
2058         return error;
2059 }       
2060
2061 SYSCALL_DEFINE2(rt_sigpending, sigset_t __user *, set, size_t, sigsetsize)
2062 {
2063         return do_sigpending(set, sigsetsize);
2064 }
2065
2066 #ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2067
2068 int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2069 {
2070         int err;
2071
2072         if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2073                 return -EFAULT;
2074         if (from->si_code < 0)
2075                 return __copy_to_user(to, from, sizeof(siginfo_t))
2076                         ? -EFAULT : 0;
2077         /*
2078          * If you change siginfo_t structure, please be sure
2079          * this code is fixed accordingly.
2080          * Please remember to update the signalfd_copyinfo() function
2081          * inside fs/signalfd.c too, in case siginfo_t changes.
2082          * It should never copy any pad contained in the structure
2083          * to avoid security leaks, but must copy the generic
2084          * 3 ints plus the relevant union member.
2085          */
2086         err = __put_user(from->si_signo, &to->si_signo);
2087         err |= __put_user(from->si_errno, &to->si_errno);
2088         err |= __put_user((short)from->si_code, &to->si_code);
2089         switch (from->si_code & __SI_MASK) {
2090         case __SI_KILL:
2091                 err |= __put_user(from->si_pid, &to->si_pid);
2092                 err |= __put_user(from->si_uid, &to->si_uid);
2093                 break;
2094         case __SI_TIMER:
2095                  err |= __put_user(from->si_tid, &to->si_tid);
2096                  err |= __put_user(from->si_overrun, &to->si_overrun);
2097                  err |= __put_user(from->si_ptr, &to->si_ptr);
2098                 break;
2099         case __SI_POLL:
2100                 err |= __put_user(from->si_band, &to->si_band);
2101                 err |= __put_user(from->si_fd, &to->si_fd);
2102                 break;
2103         case __SI_FAULT:
2104                 err |= __put_user(from->si_addr, &to->si_addr);
2105 #ifdef __ARCH_SI_TRAPNO
2106                 err |= __put_user(from->si_trapno, &to->si_trapno);
2107 #endif
2108                 break;
2109         case __SI_CHLD:
2110                 err |= __put_user(from->si_pid, &to->si_pid);
2111                 err |= __put_user(from->si_uid, &to->si_uid);
2112                 err |= __put_user(from->si_status, &to->si_status);
2113                 err |= __put_user(from->si_utime, &to->si_utime);
2114                 err |= __put_user(from->si_stime, &to->si_stime);
2115                 break;
2116         case __SI_RT: /* This is not generated by the kernel as of now. */
2117         case __SI_MESGQ: /* But this is */
2118                 err |= __put_user(from->si_pid, &to->si_pid);
2119                 err |= __put_user(from->si_uid, &to->si_uid);
2120                 err |= __put_user(from->si_ptr, &to->si_ptr);
2121                 break;
2122         default: /* this is just in case for now ... */
2123                 err |= __put_user(from->si_pid, &to->si_pid);
2124                 err |= __put_user(from->si_uid, &to->si_uid);
2125                 break;
2126         }
2127         return err;
2128 }
2129
2130 #endif
2131
2132 SYSCALL_DEFINE4(rt_sigtimedwait, const sigset_t __user *, uthese,
2133                 siginfo_t __user *, uinfo, const struct timespec __user *, uts,
2134                 size_t, sigsetsize)
2135 {
2136         int ret, sig;
2137         sigset_t these;
2138         struct timespec ts;
2139         siginfo_t info;
2140         long timeout = 0;
2141
2142         /* XXX: Don't preclude handling different sized sigset_t's.  */
2143         if (sigsetsize != sizeof(sigset_t))
2144                 return -EINVAL;
2145
2146         if (copy_from_user(&these, uthese, sizeof(these)))
2147                 return -EFAULT;
2148                 
2149         /*
2150          * Invert the set of allowed signals to get those we
2151          * want to block.
2152          */
2153         sigdelsetmask(&these, sigmask(SIGKILL)|sigmask(SIGSTOP));
2154         signotset(&these);
2155
2156         if (uts) {
2157                 if (copy_from_user(&ts, uts, sizeof(ts)))
2158                         return -EFAULT;
2159                 if (ts.tv_nsec >= 1000000000L || ts.tv_nsec < 0
2160                     || ts.tv_sec < 0)
2161                         return -EINVAL;
2162         }
2163
2164         spin_lock_irq(&current->sighand->siglock);
2165         sig = dequeue_signal(current, &these, &info);
2166         if (!sig) {
2167                 timeout = MAX_SCHEDULE_TIMEOUT;
2168                 if (uts)
2169                         timeout = (timespec_to_jiffies(&ts)
2170                                    + (ts.tv_sec || ts.tv_nsec));
2171
2172                 if (timeout) {
2173                         /* None ready -- temporarily unblock those we're
2174                          * interested while we are sleeping in so that we'll
2175                          * be awakened when they arrive.  */
2176                         current->real_blocked = current->blocked;
2177                         sigandsets(&current->blocked, &current->blocked, &these);
2178                         recalc_sigpending();
2179                         spin_unlock_irq(&current->sighand->siglock);
2180
2181                         timeout = schedule_timeout_interruptible(timeout);
2182
2183                         spin_lock_irq(&current->sighand->siglock);
2184                         sig = dequeue_signal(current, &these, &info);
2185                         current->blocked = current->real_blocked;
2186                         siginitset(&current->real_blocked, 0);
2187                         recalc_sigpending();
2188                 }
2189         }
2190         spin_unlock_irq(&current->sighand->siglock);
2191
2192         if (sig) {
2193                 ret = sig;
2194                 if (uinfo) {
2195                         if (copy_siginfo_to_user(uinfo, &info))
2196                                 ret = -EFAULT;
2197                 }
2198         } else {
2199                 ret = -EAGAIN;
2200                 if (timeout)
2201                         ret = -EINTR;
2202         }
2203
2204         return ret;
2205 }
2206
2207 SYSCALL_DEFINE2(kill, pid_t, pid, int, sig)
2208 {
2209         struct siginfo info;
2210
2211         info.si_signo = sig;
2212         info.si_errno = 0;
2213         info.si_code = SI_USER;
2214         info.si_pid = task_tgid_vnr(current);
2215         info.si_uid = current->uid;
2216
2217         return kill_something_info(sig, &info, pid);
2218 }
2219
2220 static int do_tkill(pid_t tgid, pid_t pid, int sig)
2221 {
2222         int error;
2223         struct siginfo info;
2224         struct task_struct *p;
2225         unsigned long flags;
2226
2227         error = -ESRCH;
2228         info.si_signo = sig;
2229         info.si_errno = 0;
2230         info.si_code = SI_TKILL;
2231         info.si_pid = task_tgid_vnr(current);
2232         info.si_uid = current->uid;
2233
2234         rcu_read_lock();
2235         p = find_task_by_vpid(pid);
2236         if (p && (tgid <= 0 || task_tgid_vnr(p) == tgid)) {
2237                 error = check_kill_permission(sig, &info, p);
2238                 /*
2239                  * The null signal is a permissions and process existence
2240                  * probe.  No signal is actually delivered.
2241                  *
2242                  * If lock_task_sighand() fails we pretend the task dies
2243                  * after receiving the signal. The window is tiny, and the
2244                  * signal is private anyway.
2245                  */
2246                 if (!error && sig && lock_task_sighand(p, &flags)) {
2247                         error = specific_send_sig_info(sig, &info, p);
2248                         unlock_task_sighand(p, &flags);
2249                 }
2250         }
2251         rcu_read_unlock();
2252
2253         return error;
2254 }
2255
2256 /**
2257  *  sys_tgkill - send signal to one specific thread
2258  *  @tgid: the thread group ID of the thread
2259  *  @pid: the PID of the thread
2260  *  @sig: signal to be sent
2261  *
2262  *  This syscall also checks the @tgid and returns -ESRCH even if the PID
2263  *  exists but it's not belonging to the target process anymore. This
2264  *  method solves the problem of threads exiting and PIDs getting reused.
2265  */
2266 SYSCALL_DEFINE3(tgkill, pid_t, tgid, pid_t, pid, int, sig)
2267 {
2268         /* This is only valid for single tasks */
2269         if (pid <= 0 || tgid <= 0)
2270                 return -EINVAL;
2271
2272         return do_tkill(tgid, pid, sig);
2273 }
2274
2275 /*
2276  *  Send a signal to only one task, even if it's a CLONE_THREAD task.
2277  */
2278 SYSCALL_DEFINE2(tkill, pid_t, pid, int, sig)
2279 {
2280         /* This is only valid for single tasks */
2281         if (pid <= 0)
2282                 return -EINVAL;
2283
2284         return do_tkill(0, pid, sig);
2285 }
2286
2287 SYSCALL_DEFINE3(rt_sigqueueinfo, pid_t, pid, int, sig,
2288                 siginfo_t __user *, uinfo)
2289 {
2290         siginfo_t info;
2291
2292         if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2293                 return -EFAULT;
2294
2295         /* Not even root can pretend to send signals from the kernel.
2296            Nor can they impersonate a kill(), which adds source info.  */
2297         if (info.si_code >= 0)
2298                 return -EPERM;
2299         info.si_signo = sig;
2300
2301         /* POSIX.1b doesn't mention process groups.  */
2302         return kill_proc_info(sig, &info, pid);
2303 }
2304
2305 int do_sigaction(int sig, struct k_sigaction *act, struct k_sigaction *oact)
2306 {
2307         struct task_struct *t = current;
2308         struct k_sigaction *k;
2309         sigset_t mask;
2310
2311         if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
2312                 return -EINVAL;
2313
2314         k = &t->sighand->action[sig-1];
2315
2316         spin_lock_irq(&current->sighand->siglock);
2317         if (oact)
2318                 *oact = *k;
2319
2320         if (act) {
2321                 sigdelsetmask(&act->sa.sa_mask,
2322                               sigmask(SIGKILL) | sigmask(SIGSTOP));
2323                 *k = *act;
2324                 /*
2325                  * POSIX 3.3.1.3:
2326                  *  "Setting a signal action to SIG_IGN for a signal that is
2327                  *   pending shall cause the pending signal to be discarded,
2328                  *   whether or not it is blocked."
2329                  *
2330                  *  "Setting a signal action to SIG_DFL for a signal that is
2331                  *   pending and whose default action is to ignore the signal
2332                  *   (for example, SIGCHLD), shall cause the pending signal to
2333                  *   be discarded, whether or not it is blocked"
2334                  */
2335                 if (sig_handler_ignored(sig_handler(t, sig), sig)) {
2336                         sigemptyset(&mask);
2337                         sigaddset(&mask, sig);
2338                         rm_from_queue_full(&mask, &t->signal->shared_pending);
2339                         do {
2340                                 rm_from_queue_full(&mask, &t->pending);
2341                                 t = next_thread(t);
2342                         } while (t != current);
2343                 }
2344         }
2345
2346         spin_unlock_irq(&current->sighand->siglock);
2347         return 0;
2348 }
2349
2350 int 
2351 do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
2352 {
2353         stack_t oss;
2354         int error;
2355
2356         oss.ss_sp = (void __user *) current->sas_ss_sp;
2357         oss.ss_size = current->sas_ss_size;
2358         oss.ss_flags = sas_ss_flags(sp);
2359
2360         if (uss) {
2361                 void __user *ss_sp;
2362                 size_t ss_size;
2363                 int ss_flags;
2364
2365                 error = -EFAULT;
2366                 if (!access_ok(VERIFY_READ, uss, sizeof(*uss))
2367                     || __get_user(ss_sp, &uss->ss_sp)
2368                     || __get_user(ss_flags, &uss->ss_flags)
2369                     || __get_user(ss_size, &uss->ss_size))
2370                         goto out;
2371
2372                 error = -EPERM;
2373                 if (on_sig_stack(sp))
2374                         goto out;
2375
2376                 error = -EINVAL;
2377                 /*
2378                  *
2379                  * Note - this code used to test ss_flags incorrectly
2380                  *        old code may have been written using ss_flags==0
2381                  *        to mean ss_flags==SS_ONSTACK (as this was the only
2382                  *        way that worked) - this fix preserves that older
2383                  *        mechanism
2384                  */
2385                 if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
2386                         goto out;
2387
2388                 if (ss_flags == SS_DISABLE) {
2389                         ss_size = 0;
2390                         ss_sp = NULL;
2391                 } else {
2392                         error = -ENOMEM;
2393                         if (ss_size < MINSIGSTKSZ)
2394                                 goto out;
2395                 }
2396
2397                 current->sas_ss_sp = (unsigned long) ss_sp;
2398                 current->sas_ss_size = ss_size;
2399         }
2400
2401         error = 0;
2402         if (uoss) {
2403                 error = -EFAULT;
2404                 if (!access_ok(VERIFY_WRITE, uoss, sizeof(*uoss)))
2405                         goto out;
2406                 error = __put_user(oss.ss_sp, &uoss->ss_sp) |
2407                         __put_user(oss.ss_size, &uoss->ss_size) |
2408                         __put_user(oss.ss_flags, &uoss->ss_flags);
2409         }
2410
2411 out:
2412         return error;
2413 }
2414
2415 #ifdef __ARCH_WANT_SYS_SIGPENDING
2416
2417 SYSCALL_DEFINE1(sigpending, old_sigset_t __user *, set)
2418 {
2419         return do_sigpending(set, sizeof(*set));
2420 }
2421
2422 #endif
2423
2424 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
2425 /* Some platforms have their own version with special arguments others
2426    support only sys_rt_sigprocmask.  */
2427
2428 SYSCALL_DEFINE3(sigprocmask, int, how, old_sigset_t __user *, set,
2429                 old_sigset_t __user *, oset)
2430 {
2431         int error;
2432         old_sigset_t old_set, new_set;
2433
2434         if (set) {
2435                 error = -EFAULT;
2436                 if (copy_from_user(&new_set, set, sizeof(*set)))
2437                         goto out;
2438                 new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
2439
2440                 spin_lock_irq(&current->sighand->siglock);
2441                 old_set = current->blocked.sig[0];
2442
2443                 error = 0;
2444                 switch (how) {
2445                 default:
2446                         error = -EINVAL;
2447                         break;
2448                 case SIG_BLOCK:
2449                         sigaddsetmask(&current->blocked, new_set);
2450                         break;
2451                 case SIG_UNBLOCK:
2452                         sigdelsetmask(&current->blocked, new_set);
2453                         break;
2454                 case SIG_SETMASK:
2455                         current->blocked.sig[0] = new_set;
2456                         break;
2457                 }
2458
2459                 recalc_sigpending();
2460                 spin_unlock_irq(&current->sighand->siglock);
2461                 if (error)
2462                         goto out;
2463                 if (oset)
2464                         goto set_old;
2465         } else if (oset) {
2466                 old_set = current->blocked.sig[0];
2467         set_old:
2468                 error = -EFAULT;
2469                 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2470                         goto out;
2471         }
2472         error = 0;
2473 out:
2474         return error;
2475 }
2476 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
2477
2478 #ifdef __ARCH_WANT_SYS_RT_SIGACTION
2479 SYSCALL_DEFINE4(rt_sigaction, int, sig,
2480                 const struct sigaction __user *, act,
2481                 struct sigaction __user *, oact,
2482                 size_t, sigsetsize)
2483 {
2484         struct k_sigaction new_sa, old_sa;
2485         int ret = -EINVAL;
2486
2487         /* XXX: Don't preclude handling different sized sigset_t's.  */
2488         if (sigsetsize != sizeof(sigset_t))
2489                 goto out;
2490
2491         if (act) {
2492                 if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
2493                         return -EFAULT;
2494         }
2495
2496         ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
2497
2498         if (!ret && oact) {
2499                 if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
2500                         return -EFAULT;
2501         }
2502 out:
2503         return ret;
2504 }
2505 #endif /* __ARCH_WANT_SYS_RT_SIGACTION */
2506
2507 #ifdef __ARCH_WANT_SYS_SGETMASK
2508
2509 /*
2510  * For backwards compatibility.  Functionality superseded by sigprocmask.
2511  */
2512 SYSCALL_DEFINE0(sgetmask)
2513 {
2514         /* SMP safe */
2515         return current->blocked.sig[0];
2516 }
2517
2518 SYSCALL_DEFINE1(ssetmask, int, newmask)
2519 {
2520         int old;
2521
2522         spin_lock_irq(&current->sighand->siglock);
2523         old = current->blocked.sig[0];
2524
2525         siginitset(&current->blocked, newmask & ~(sigmask(SIGKILL)|
2526                                                   sigmask(SIGSTOP)));
2527         recalc_sigpending();
2528         spin_unlock_irq(&current->sighand->siglock);
2529
2530         return old;
2531 }
2532 #endif /* __ARCH_WANT_SGETMASK */
2533
2534 #ifdef __ARCH_WANT_SYS_SIGNAL
2535 /*
2536  * For backwards compatibility.  Functionality superseded by sigaction.
2537  */
2538 SYSCALL_DEFINE2(signal, int, sig, __sighandler_t, handler)
2539 {
2540         struct k_sigaction new_sa, old_sa;
2541         int ret;
2542
2543         new_sa.sa.sa_handler = handler;
2544         new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
2545         sigemptyset(&new_sa.sa.sa_mask);
2546
2547         ret = do_sigaction(sig, &new_sa, &old_sa);
2548
2549         return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
2550 }
2551 #endif /* __ARCH_WANT_SYS_SIGNAL */
2552
2553 #ifdef __ARCH_WANT_SYS_PAUSE
2554
2555 SYSCALL_DEFINE0(pause)
2556 {
2557         current->state = TASK_INTERRUPTIBLE;
2558         schedule();
2559         return -ERESTARTNOHAND;
2560 }
2561
2562 #endif
2563
2564 #ifdef __ARCH_WANT_SYS_RT_SIGSUSPEND
2565 SYSCALL_DEFINE2(rt_sigsuspend, sigset_t __user *, unewset, size_t, sigsetsize)
2566 {
2567         sigset_t newset;
2568
2569         /* XXX: Don't preclude handling different sized sigset_t's.  */
2570         if (sigsetsize != sizeof(sigset_t))
2571                 return -EINVAL;
2572
2573         if (copy_from_user(&newset, unewset, sizeof(newset)))
2574                 return -EFAULT;
2575         sigdelsetmask(&newset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2576
2577         spin_lock_irq(&current->sighand->siglock);
2578         current->saved_sigmask = current->blocked;
2579         current->blocked = newset;
2580         recalc_sigpending();
2581         spin_unlock_irq(&current->sighand->siglock);
2582
2583         current->state = TASK_INTERRUPTIBLE;
2584         schedule();
2585         set_restore_sigmask();
2586         return -ERESTARTNOHAND;
2587 }
2588 #endif /* __ARCH_WANT_SYS_RT_SIGSUSPEND */
2589
2590 __attribute__((weak)) const char *arch_vma_name(struct vm_area_struct *vma)
2591 {
2592         return NULL;
2593 }
2594
2595 void __init signals_init(void)
2596 {
2597         sigqueue_cachep = KMEM_CACHE(sigqueue, SLAB_PANIC);
2598 }