[PATCH] kernel/hrtimer.c sparse warning fix
[linux-2.6] / kernel / signal.c
1 /*
2  *  linux/kernel/signal.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  *
6  *  1997-11-02  Modified for POSIX.1b signals by Richard Henderson
7  *
8  *  2003-06-02  Jim Houston - Concurrent Computer Corp.
9  *              Changes to use preallocated sigqueue structures
10  *              to allow signals to be sent reliably.
11  */
12
13 #include <linux/config.h>
14 #include <linux/slab.h>
15 #include <linux/module.h>
16 #include <linux/smp_lock.h>
17 #include <linux/init.h>
18 #include <linux/sched.h>
19 #include <linux/fs.h>
20 #include <linux/tty.h>
21 #include <linux/binfmts.h>
22 #include <linux/security.h>
23 #include <linux/syscalls.h>
24 #include <linux/ptrace.h>
25 #include <linux/posix-timers.h>
26 #include <linux/signal.h>
27 #include <linux/audit.h>
28 #include <linux/capability.h>
29 #include <asm/param.h>
30 #include <asm/uaccess.h>
31 #include <asm/unistd.h>
32 #include <asm/siginfo.h>
33
34 /*
35  * SLAB caches for signal bits.
36  */
37
38 static kmem_cache_t *sigqueue_cachep;
39
40 /*
41  * In POSIX a signal is sent either to a specific thread (Linux task)
42  * or to the process as a whole (Linux thread group).  How the signal
43  * is sent determines whether it's to one thread or the whole group,
44  * which determines which signal mask(s) are involved in blocking it
45  * from being delivered until later.  When the signal is delivered,
46  * either it's caught or ignored by a user handler or it has a default
47  * effect that applies to the whole thread group (POSIX process).
48  *
49  * The possible effects an unblocked signal set to SIG_DFL can have are:
50  *   ignore     - Nothing Happens
51  *   terminate  - kill the process, i.e. all threads in the group,
52  *                similar to exit_group.  The group leader (only) reports
53  *                WIFSIGNALED status to its parent.
54  *   coredump   - write a core dump file describing all threads using
55  *                the same mm and then kill all those threads
56  *   stop       - stop all the threads in the group, i.e. TASK_STOPPED state
57  *
58  * SIGKILL and SIGSTOP cannot be caught, blocked, or ignored.
59  * Other signals when not blocked and set to SIG_DFL behaves as follows.
60  * The job control signals also have other special effects.
61  *
62  *      +--------------------+------------------+
63  *      |  POSIX signal      |  default action  |
64  *      +--------------------+------------------+
65  *      |  SIGHUP            |  terminate       |
66  *      |  SIGINT            |  terminate       |
67  *      |  SIGQUIT           |  coredump        |
68  *      |  SIGILL            |  coredump        |
69  *      |  SIGTRAP           |  coredump        |
70  *      |  SIGABRT/SIGIOT    |  coredump        |
71  *      |  SIGBUS            |  coredump        |
72  *      |  SIGFPE            |  coredump        |
73  *      |  SIGKILL           |  terminate(+)    |
74  *      |  SIGUSR1           |  terminate       |
75  *      |  SIGSEGV           |  coredump        |
76  *      |  SIGUSR2           |  terminate       |
77  *      |  SIGPIPE           |  terminate       |
78  *      |  SIGALRM           |  terminate       |
79  *      |  SIGTERM           |  terminate       |
80  *      |  SIGCHLD           |  ignore          |
81  *      |  SIGCONT           |  ignore(*)       |
82  *      |  SIGSTOP           |  stop(*)(+)      |
83  *      |  SIGTSTP           |  stop(*)         |
84  *      |  SIGTTIN           |  stop(*)         |
85  *      |  SIGTTOU           |  stop(*)         |
86  *      |  SIGURG            |  ignore          |
87  *      |  SIGXCPU           |  coredump        |
88  *      |  SIGXFSZ           |  coredump        |
89  *      |  SIGVTALRM         |  terminate       |
90  *      |  SIGPROF           |  terminate       |
91  *      |  SIGPOLL/SIGIO     |  terminate       |
92  *      |  SIGSYS/SIGUNUSED  |  coredump        |
93  *      |  SIGSTKFLT         |  terminate       |
94  *      |  SIGWINCH          |  ignore          |
95  *      |  SIGPWR            |  terminate       |
96  *      |  SIGRTMIN-SIGRTMAX |  terminate       |
97  *      +--------------------+------------------+
98  *      |  non-POSIX signal  |  default action  |
99  *      +--------------------+------------------+
100  *      |  SIGEMT            |  coredump        |
101  *      +--------------------+------------------+
102  *
103  * (+) For SIGKILL and SIGSTOP the action is "always", not just "default".
104  * (*) Special job control effects:
105  * When SIGCONT is sent, it resumes the process (all threads in the group)
106  * from TASK_STOPPED state and also clears any pending/queued stop signals
107  * (any of those marked with "stop(*)").  This happens regardless of blocking,
108  * catching, or ignoring SIGCONT.  When any stop signal is sent, it clears
109  * any pending/queued SIGCONT signals; this happens regardless of blocking,
110  * catching, or ignored the stop signal, though (except for SIGSTOP) the
111  * default action of stopping the process may happen later or never.
112  */
113
114 #ifdef SIGEMT
115 #define M_SIGEMT        M(SIGEMT)
116 #else
117 #define M_SIGEMT        0
118 #endif
119
120 #if SIGRTMIN > BITS_PER_LONG
121 #define M(sig) (1ULL << ((sig)-1))
122 #else
123 #define M(sig) (1UL << ((sig)-1))
124 #endif
125 #define T(sig, mask) (M(sig) & (mask))
126
127 #define SIG_KERNEL_ONLY_MASK (\
128         M(SIGKILL)   |  M(SIGSTOP)                                   )
129
130 #define SIG_KERNEL_STOP_MASK (\
131         M(SIGSTOP)   |  M(SIGTSTP)   |  M(SIGTTIN)   |  M(SIGTTOU)   )
132
133 #define SIG_KERNEL_COREDUMP_MASK (\
134         M(SIGQUIT)   |  M(SIGILL)    |  M(SIGTRAP)   |  M(SIGABRT)   | \
135         M(SIGFPE)    |  M(SIGSEGV)   |  M(SIGBUS)    |  M(SIGSYS)    | \
136         M(SIGXCPU)   |  M(SIGXFSZ)   |  M_SIGEMT                     )
137
138 #define SIG_KERNEL_IGNORE_MASK (\
139         M(SIGCONT)   |  M(SIGCHLD)   |  M(SIGWINCH)  |  M(SIGURG)    )
140
141 #define sig_kernel_only(sig) \
142                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_ONLY_MASK))
143 #define sig_kernel_coredump(sig) \
144                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_COREDUMP_MASK))
145 #define sig_kernel_ignore(sig) \
146                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_IGNORE_MASK))
147 #define sig_kernel_stop(sig) \
148                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_STOP_MASK))
149
150 #define sig_user_defined(t, signr) \
151         (((t)->sighand->action[(signr)-1].sa.sa_handler != SIG_DFL) &&  \
152          ((t)->sighand->action[(signr)-1].sa.sa_handler != SIG_IGN))
153
154 #define sig_fatal(t, signr) \
155         (!T(signr, SIG_KERNEL_IGNORE_MASK|SIG_KERNEL_STOP_MASK) && \
156          (t)->sighand->action[(signr)-1].sa.sa_handler == SIG_DFL)
157
158 static int sig_ignored(struct task_struct *t, int sig)
159 {
160         void __user * handler;
161
162         /*
163          * Tracers always want to know about signals..
164          */
165         if (t->ptrace & PT_PTRACED)
166                 return 0;
167
168         /*
169          * Blocked signals are never ignored, since the
170          * signal handler may change by the time it is
171          * unblocked.
172          */
173         if (sigismember(&t->blocked, sig))
174                 return 0;
175
176         /* Is it explicitly or implicitly ignored? */
177         handler = t->sighand->action[sig-1].sa.sa_handler;
178         return   handler == SIG_IGN ||
179                 (handler == SIG_DFL && sig_kernel_ignore(sig));
180 }
181
182 /*
183  * Re-calculate pending state from the set of locally pending
184  * signals, globally pending signals, and blocked signals.
185  */
186 static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
187 {
188         unsigned long ready;
189         long i;
190
191         switch (_NSIG_WORDS) {
192         default:
193                 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
194                         ready |= signal->sig[i] &~ blocked->sig[i];
195                 break;
196
197         case 4: ready  = signal->sig[3] &~ blocked->sig[3];
198                 ready |= signal->sig[2] &~ blocked->sig[2];
199                 ready |= signal->sig[1] &~ blocked->sig[1];
200                 ready |= signal->sig[0] &~ blocked->sig[0];
201                 break;
202
203         case 2: ready  = signal->sig[1] &~ blocked->sig[1];
204                 ready |= signal->sig[0] &~ blocked->sig[0];
205                 break;
206
207         case 1: ready  = signal->sig[0] &~ blocked->sig[0];
208         }
209         return ready != 0;
210 }
211
212 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
213
214 fastcall void recalc_sigpending_tsk(struct task_struct *t)
215 {
216         if (t->signal->group_stop_count > 0 ||
217             (freezing(t)) ||
218             PENDING(&t->pending, &t->blocked) ||
219             PENDING(&t->signal->shared_pending, &t->blocked))
220                 set_tsk_thread_flag(t, TIF_SIGPENDING);
221         else
222                 clear_tsk_thread_flag(t, TIF_SIGPENDING);
223 }
224
225 void recalc_sigpending(void)
226 {
227         recalc_sigpending_tsk(current);
228 }
229
230 /* Given the mask, find the first available signal that should be serviced. */
231
232 static int
233 next_signal(struct sigpending *pending, sigset_t *mask)
234 {
235         unsigned long i, *s, *m, x;
236         int sig = 0;
237         
238         s = pending->signal.sig;
239         m = mask->sig;
240         switch (_NSIG_WORDS) {
241         default:
242                 for (i = 0; i < _NSIG_WORDS; ++i, ++s, ++m)
243                         if ((x = *s &~ *m) != 0) {
244                                 sig = ffz(~x) + i*_NSIG_BPW + 1;
245                                 break;
246                         }
247                 break;
248
249         case 2: if ((x = s[0] &~ m[0]) != 0)
250                         sig = 1;
251                 else if ((x = s[1] &~ m[1]) != 0)
252                         sig = _NSIG_BPW + 1;
253                 else
254                         break;
255                 sig += ffz(~x);
256                 break;
257
258         case 1: if ((x = *s &~ *m) != 0)
259                         sig = ffz(~x) + 1;
260                 break;
261         }
262         
263         return sig;
264 }
265
266 static struct sigqueue *__sigqueue_alloc(struct task_struct *t, gfp_t flags,
267                                          int override_rlimit)
268 {
269         struct sigqueue *q = NULL;
270
271         atomic_inc(&t->user->sigpending);
272         if (override_rlimit ||
273             atomic_read(&t->user->sigpending) <=
274                         t->signal->rlim[RLIMIT_SIGPENDING].rlim_cur)
275                 q = kmem_cache_alloc(sigqueue_cachep, flags);
276         if (unlikely(q == NULL)) {
277                 atomic_dec(&t->user->sigpending);
278         } else {
279                 INIT_LIST_HEAD(&q->list);
280                 q->flags = 0;
281                 q->user = get_uid(t->user);
282         }
283         return(q);
284 }
285
286 static inline void __sigqueue_free(struct sigqueue *q)
287 {
288         if (q->flags & SIGQUEUE_PREALLOC)
289                 return;
290         atomic_dec(&q->user->sigpending);
291         free_uid(q->user);
292         kmem_cache_free(sigqueue_cachep, q);
293 }
294
295 static void flush_sigqueue(struct sigpending *queue)
296 {
297         struct sigqueue *q;
298
299         sigemptyset(&queue->signal);
300         while (!list_empty(&queue->list)) {
301                 q = list_entry(queue->list.next, struct sigqueue , list);
302                 list_del_init(&q->list);
303                 __sigqueue_free(q);
304         }
305 }
306
307 /*
308  * Flush all pending signals for a task.
309  */
310
311 void
312 flush_signals(struct task_struct *t)
313 {
314         unsigned long flags;
315
316         spin_lock_irqsave(&t->sighand->siglock, flags);
317         clear_tsk_thread_flag(t,TIF_SIGPENDING);
318         flush_sigqueue(&t->pending);
319         flush_sigqueue(&t->signal->shared_pending);
320         spin_unlock_irqrestore(&t->sighand->siglock, flags);
321 }
322
323 /*
324  * This function expects the tasklist_lock write-locked.
325  */
326 void __exit_sighand(struct task_struct *tsk)
327 {
328         struct sighand_struct * sighand = tsk->sighand;
329
330         /* Ok, we're done with the signal handlers */
331         tsk->sighand = NULL;
332         if (atomic_dec_and_test(&sighand->count))
333                 sighand_free(sighand);
334 }
335
336 void exit_sighand(struct task_struct *tsk)
337 {
338         write_lock_irq(&tasklist_lock);
339         rcu_read_lock();
340         if (tsk->sighand != NULL) {
341                 struct sighand_struct *sighand = rcu_dereference(tsk->sighand);
342                 spin_lock(&sighand->siglock);
343                 __exit_sighand(tsk);
344                 spin_unlock(&sighand->siglock);
345         }
346         rcu_read_unlock();
347         write_unlock_irq(&tasklist_lock);
348 }
349
350 /*
351  * This function expects the tasklist_lock write-locked.
352  */
353 void __exit_signal(struct task_struct *tsk)
354 {
355         struct signal_struct * sig = tsk->signal;
356         struct sighand_struct * sighand;
357
358         if (!sig)
359                 BUG();
360         if (!atomic_read(&sig->count))
361                 BUG();
362         rcu_read_lock();
363         sighand = rcu_dereference(tsk->sighand);
364         spin_lock(&sighand->siglock);
365         posix_cpu_timers_exit(tsk);
366         if (atomic_dec_and_test(&sig->count)) {
367                 posix_cpu_timers_exit_group(tsk);
368                 tsk->signal = NULL;
369                 __exit_sighand(tsk);
370                 spin_unlock(&sighand->siglock);
371                 flush_sigqueue(&sig->shared_pending);
372         } else {
373                 /*
374                  * If there is any task waiting for the group exit
375                  * then notify it:
376                  */
377                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
378                         wake_up_process(sig->group_exit_task);
379                         sig->group_exit_task = NULL;
380                 }
381                 if (tsk == sig->curr_target)
382                         sig->curr_target = next_thread(tsk);
383                 tsk->signal = NULL;
384                 /*
385                  * Accumulate here the counters for all threads but the
386                  * group leader as they die, so they can be added into
387                  * the process-wide totals when those are taken.
388                  * The group leader stays around as a zombie as long
389                  * as there are other threads.  When it gets reaped,
390                  * the exit.c code will add its counts into these totals.
391                  * We won't ever get here for the group leader, since it
392                  * will have been the last reference on the signal_struct.
393                  */
394                 sig->utime = cputime_add(sig->utime, tsk->utime);
395                 sig->stime = cputime_add(sig->stime, tsk->stime);
396                 sig->min_flt += tsk->min_flt;
397                 sig->maj_flt += tsk->maj_flt;
398                 sig->nvcsw += tsk->nvcsw;
399                 sig->nivcsw += tsk->nivcsw;
400                 sig->sched_time += tsk->sched_time;
401                 __exit_sighand(tsk);
402                 spin_unlock(&sighand->siglock);
403                 sig = NULL;     /* Marker for below.  */
404         }
405         rcu_read_unlock();
406         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
407         flush_sigqueue(&tsk->pending);
408         if (sig) {
409                 /*
410                  * We are cleaning up the signal_struct here.
411                  */
412                 exit_thread_group_keys(sig);
413                 kmem_cache_free(signal_cachep, sig);
414         }
415 }
416
417 void exit_signal(struct task_struct *tsk)
418 {
419         atomic_dec(&tsk->signal->live);
420
421         write_lock_irq(&tasklist_lock);
422         __exit_signal(tsk);
423         write_unlock_irq(&tasklist_lock);
424 }
425
426 /*
427  * Flush all handlers for a task.
428  */
429
430 void
431 flush_signal_handlers(struct task_struct *t, int force_default)
432 {
433         int i;
434         struct k_sigaction *ka = &t->sighand->action[0];
435         for (i = _NSIG ; i != 0 ; i--) {
436                 if (force_default || ka->sa.sa_handler != SIG_IGN)
437                         ka->sa.sa_handler = SIG_DFL;
438                 ka->sa.sa_flags = 0;
439                 sigemptyset(&ka->sa.sa_mask);
440                 ka++;
441         }
442 }
443
444
445 /* Notify the system that a driver wants to block all signals for this
446  * process, and wants to be notified if any signals at all were to be
447  * sent/acted upon.  If the notifier routine returns non-zero, then the
448  * signal will be acted upon after all.  If the notifier routine returns 0,
449  * then then signal will be blocked.  Only one block per process is
450  * allowed.  priv is a pointer to private data that the notifier routine
451  * can use to determine if the signal should be blocked or not.  */
452
453 void
454 block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
455 {
456         unsigned long flags;
457
458         spin_lock_irqsave(&current->sighand->siglock, flags);
459         current->notifier_mask = mask;
460         current->notifier_data = priv;
461         current->notifier = notifier;
462         spin_unlock_irqrestore(&current->sighand->siglock, flags);
463 }
464
465 /* Notify the system that blocking has ended. */
466
467 void
468 unblock_all_signals(void)
469 {
470         unsigned long flags;
471
472         spin_lock_irqsave(&current->sighand->siglock, flags);
473         current->notifier = NULL;
474         current->notifier_data = NULL;
475         recalc_sigpending();
476         spin_unlock_irqrestore(&current->sighand->siglock, flags);
477 }
478
479 static int collect_signal(int sig, struct sigpending *list, siginfo_t *info)
480 {
481         struct sigqueue *q, *first = NULL;
482         int still_pending = 0;
483
484         if (unlikely(!sigismember(&list->signal, sig)))
485                 return 0;
486
487         /*
488          * Collect the siginfo appropriate to this signal.  Check if
489          * there is another siginfo for the same signal.
490         */
491         list_for_each_entry(q, &list->list, list) {
492                 if (q->info.si_signo == sig) {
493                         if (first) {
494                                 still_pending = 1;
495                                 break;
496                         }
497                         first = q;
498                 }
499         }
500         if (first) {
501                 list_del_init(&first->list);
502                 copy_siginfo(info, &first->info);
503                 __sigqueue_free(first);
504                 if (!still_pending)
505                         sigdelset(&list->signal, sig);
506         } else {
507
508                 /* Ok, it wasn't in the queue.  This must be
509                    a fast-pathed signal or we must have been
510                    out of queue space.  So zero out the info.
511                  */
512                 sigdelset(&list->signal, sig);
513                 info->si_signo = sig;
514                 info->si_errno = 0;
515                 info->si_code = 0;
516                 info->si_pid = 0;
517                 info->si_uid = 0;
518         }
519         return 1;
520 }
521
522 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
523                         siginfo_t *info)
524 {
525         int sig = 0;
526
527         sig = next_signal(pending, mask);
528         if (sig) {
529                 if (current->notifier) {
530                         if (sigismember(current->notifier_mask, sig)) {
531                                 if (!(current->notifier)(current->notifier_data)) {
532                                         clear_thread_flag(TIF_SIGPENDING);
533                                         return 0;
534                                 }
535                         }
536                 }
537
538                 if (!collect_signal(sig, pending, info))
539                         sig = 0;
540                                 
541         }
542         recalc_sigpending();
543
544         return sig;
545 }
546
547 /*
548  * Dequeue a signal and return the element to the caller, which is 
549  * expected to free it.
550  *
551  * All callers have to hold the siglock.
552  */
553 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
554 {
555         int signr = __dequeue_signal(&tsk->pending, mask, info);
556         if (!signr)
557                 signr = __dequeue_signal(&tsk->signal->shared_pending,
558                                          mask, info);
559         if (signr && unlikely(sig_kernel_stop(signr))) {
560                 /*
561                  * Set a marker that we have dequeued a stop signal.  Our
562                  * caller might release the siglock and then the pending
563                  * stop signal it is about to process is no longer in the
564                  * pending bitmasks, but must still be cleared by a SIGCONT
565                  * (and overruled by a SIGKILL).  So those cases clear this
566                  * shared flag after we've set it.  Note that this flag may
567                  * remain set after the signal we return is ignored or
568                  * handled.  That doesn't matter because its only purpose
569                  * is to alert stop-signal processing code when another
570                  * processor has come along and cleared the flag.
571                  */
572                 if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT))
573                         tsk->signal->flags |= SIGNAL_STOP_DEQUEUED;
574         }
575         if ( signr &&
576              ((info->si_code & __SI_MASK) == __SI_TIMER) &&
577              info->si_sys_private){
578                 /*
579                  * Release the siglock to ensure proper locking order
580                  * of timer locks outside of siglocks.  Note, we leave
581                  * irqs disabled here, since the posix-timers code is
582                  * about to disable them again anyway.
583                  */
584                 spin_unlock(&tsk->sighand->siglock);
585                 do_schedule_next_timer(info);
586                 spin_lock(&tsk->sighand->siglock);
587         }
588         return signr;
589 }
590
591 /*
592  * Tell a process that it has a new active signal..
593  *
594  * NOTE! we rely on the previous spin_lock to
595  * lock interrupts for us! We can only be called with
596  * "siglock" held, and the local interrupt must
597  * have been disabled when that got acquired!
598  *
599  * No need to set need_resched since signal event passing
600  * goes through ->blocked
601  */
602 void signal_wake_up(struct task_struct *t, int resume)
603 {
604         unsigned int mask;
605
606         set_tsk_thread_flag(t, TIF_SIGPENDING);
607
608         /*
609          * For SIGKILL, we want to wake it up in the stopped/traced case.
610          * We don't check t->state here because there is a race with it
611          * executing another processor and just now entering stopped state.
612          * By using wake_up_state, we ensure the process will wake up and
613          * handle its death signal.
614          */
615         mask = TASK_INTERRUPTIBLE;
616         if (resume)
617                 mask |= TASK_STOPPED | TASK_TRACED;
618         if (!wake_up_state(t, mask))
619                 kick_process(t);
620 }
621
622 /*
623  * Remove signals in mask from the pending set and queue.
624  * Returns 1 if any signals were found.
625  *
626  * All callers must be holding the siglock.
627  *
628  * This version takes a sigset mask and looks at all signals,
629  * not just those in the first mask word.
630  */
631 static int rm_from_queue_full(sigset_t *mask, struct sigpending *s)
632 {
633         struct sigqueue *q, *n;
634         sigset_t m;
635
636         sigandsets(&m, mask, &s->signal);
637         if (sigisemptyset(&m))
638                 return 0;
639
640         signandsets(&s->signal, &s->signal, mask);
641         list_for_each_entry_safe(q, n, &s->list, list) {
642                 if (sigismember(mask, q->info.si_signo)) {
643                         list_del_init(&q->list);
644                         __sigqueue_free(q);
645                 }
646         }
647         return 1;
648 }
649 /*
650  * Remove signals in mask from the pending set and queue.
651  * Returns 1 if any signals were found.
652  *
653  * All callers must be holding the siglock.
654  */
655 static int rm_from_queue(unsigned long mask, struct sigpending *s)
656 {
657         struct sigqueue *q, *n;
658
659         if (!sigtestsetmask(&s->signal, mask))
660                 return 0;
661
662         sigdelsetmask(&s->signal, mask);
663         list_for_each_entry_safe(q, n, &s->list, list) {
664                 if (q->info.si_signo < SIGRTMIN &&
665                     (mask & sigmask(q->info.si_signo))) {
666                         list_del_init(&q->list);
667                         __sigqueue_free(q);
668                 }
669         }
670         return 1;
671 }
672
673 /*
674  * Bad permissions for sending the signal
675  */
676 static int check_kill_permission(int sig, struct siginfo *info,
677                                  struct task_struct *t)
678 {
679         int error = -EINVAL;
680         if (!valid_signal(sig))
681                 return error;
682         error = -EPERM;
683         if ((info == SEND_SIG_NOINFO || (!is_si_special(info) && SI_FROMUSER(info)))
684             && ((sig != SIGCONT) ||
685                 (current->signal->session != t->signal->session))
686             && (current->euid ^ t->suid) && (current->euid ^ t->uid)
687             && (current->uid ^ t->suid) && (current->uid ^ t->uid)
688             && !capable(CAP_KILL))
689                 return error;
690
691         error = security_task_kill(t, info, sig);
692         if (!error)
693                 audit_signal_info(sig, t); /* Let audit system see the signal */
694         return error;
695 }
696
697 /* forward decl */
698 static void do_notify_parent_cldstop(struct task_struct *tsk,
699                                      int to_self,
700                                      int why);
701
702 /*
703  * Handle magic process-wide effects of stop/continue signals.
704  * Unlike the signal actions, these happen immediately at signal-generation
705  * time regardless of blocking, ignoring, or handling.  This does the
706  * actual continuing for SIGCONT, but not the actual stopping for stop
707  * signals.  The process stop is done as a signal action for SIG_DFL.
708  */
709 static void handle_stop_signal(int sig, struct task_struct *p)
710 {
711         struct task_struct *t;
712
713         if (p->signal->flags & SIGNAL_GROUP_EXIT)
714                 /*
715                  * The process is in the middle of dying already.
716                  */
717                 return;
718
719         if (sig_kernel_stop(sig)) {
720                 /*
721                  * This is a stop signal.  Remove SIGCONT from all queues.
722                  */
723                 rm_from_queue(sigmask(SIGCONT), &p->signal->shared_pending);
724                 t = p;
725                 do {
726                         rm_from_queue(sigmask(SIGCONT), &t->pending);
727                         t = next_thread(t);
728                 } while (t != p);
729         } else if (sig == SIGCONT) {
730                 /*
731                  * Remove all stop signals from all queues,
732                  * and wake all threads.
733                  */
734                 if (unlikely(p->signal->group_stop_count > 0)) {
735                         /*
736                          * There was a group stop in progress.  We'll
737                          * pretend it finished before we got here.  We are
738                          * obliged to report it to the parent: if the
739                          * SIGSTOP happened "after" this SIGCONT, then it
740                          * would have cleared this pending SIGCONT.  If it
741                          * happened "before" this SIGCONT, then the parent
742                          * got the SIGCHLD about the stop finishing before
743                          * the continue happened.  We do the notification
744                          * now, and it's as if the stop had finished and
745                          * the SIGCHLD was pending on entry to this kill.
746                          */
747                         p->signal->group_stop_count = 0;
748                         p->signal->flags = SIGNAL_STOP_CONTINUED;
749                         spin_unlock(&p->sighand->siglock);
750                         do_notify_parent_cldstop(p, (p->ptrace & PT_PTRACED), CLD_STOPPED);
751                         spin_lock(&p->sighand->siglock);
752                 }
753                 rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
754                 t = p;
755                 do {
756                         unsigned int state;
757                         rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
758                         
759                         /*
760                          * If there is a handler for SIGCONT, we must make
761                          * sure that no thread returns to user mode before
762                          * we post the signal, in case it was the only
763                          * thread eligible to run the signal handler--then
764                          * it must not do anything between resuming and
765                          * running the handler.  With the TIF_SIGPENDING
766                          * flag set, the thread will pause and acquire the
767                          * siglock that we hold now and until we've queued
768                          * the pending signal. 
769                          *
770                          * Wake up the stopped thread _after_ setting
771                          * TIF_SIGPENDING
772                          */
773                         state = TASK_STOPPED;
774                         if (sig_user_defined(t, SIGCONT) && !sigismember(&t->blocked, SIGCONT)) {
775                                 set_tsk_thread_flag(t, TIF_SIGPENDING);
776                                 state |= TASK_INTERRUPTIBLE;
777                         }
778                         wake_up_state(t, state);
779
780                         t = next_thread(t);
781                 } while (t != p);
782
783                 if (p->signal->flags & SIGNAL_STOP_STOPPED) {
784                         /*
785                          * We were in fact stopped, and are now continued.
786                          * Notify the parent with CLD_CONTINUED.
787                          */
788                         p->signal->flags = SIGNAL_STOP_CONTINUED;
789                         p->signal->group_exit_code = 0;
790                         spin_unlock(&p->sighand->siglock);
791                         do_notify_parent_cldstop(p, (p->ptrace & PT_PTRACED), CLD_CONTINUED);
792                         spin_lock(&p->sighand->siglock);
793                 } else {
794                         /*
795                          * We are not stopped, but there could be a stop
796                          * signal in the middle of being processed after
797                          * being removed from the queue.  Clear that too.
798                          */
799                         p->signal->flags = 0;
800                 }
801         } else if (sig == SIGKILL) {
802                 /*
803                  * Make sure that any pending stop signal already dequeued
804                  * is undone by the wakeup for SIGKILL.
805                  */
806                 p->signal->flags = 0;
807         }
808 }
809
810 static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
811                         struct sigpending *signals)
812 {
813         struct sigqueue * q = NULL;
814         int ret = 0;
815
816         /*
817          * fast-pathed signals for kernel-internal things like SIGSTOP
818          * or SIGKILL.
819          */
820         if (info == SEND_SIG_FORCED)
821                 goto out_set;
822
823         /* Real-time signals must be queued if sent by sigqueue, or
824            some other real-time mechanism.  It is implementation
825            defined whether kill() does so.  We attempt to do so, on
826            the principle of least surprise, but since kill is not
827            allowed to fail with EAGAIN when low on memory we just
828            make sure at least one signal gets delivered and don't
829            pass on the info struct.  */
830
831         q = __sigqueue_alloc(t, GFP_ATOMIC, (sig < SIGRTMIN &&
832                                              (is_si_special(info) ||
833                                               info->si_code >= 0)));
834         if (q) {
835                 list_add_tail(&q->list, &signals->list);
836                 switch ((unsigned long) info) {
837                 case (unsigned long) SEND_SIG_NOINFO:
838                         q->info.si_signo = sig;
839                         q->info.si_errno = 0;
840                         q->info.si_code = SI_USER;
841                         q->info.si_pid = current->pid;
842                         q->info.si_uid = current->uid;
843                         break;
844                 case (unsigned long) SEND_SIG_PRIV:
845                         q->info.si_signo = sig;
846                         q->info.si_errno = 0;
847                         q->info.si_code = SI_KERNEL;
848                         q->info.si_pid = 0;
849                         q->info.si_uid = 0;
850                         break;
851                 default:
852                         copy_siginfo(&q->info, info);
853                         break;
854                 }
855         } else if (!is_si_special(info)) {
856                 if (sig >= SIGRTMIN && info->si_code != SI_USER)
857                 /*
858                  * Queue overflow, abort.  We may abort if the signal was rt
859                  * and sent by user using something other than kill().
860                  */
861                         return -EAGAIN;
862         }
863
864 out_set:
865         sigaddset(&signals->signal, sig);
866         return ret;
867 }
868
869 #define LEGACY_QUEUE(sigptr, sig) \
870         (((sig) < SIGRTMIN) && sigismember(&(sigptr)->signal, (sig)))
871
872
873 static int
874 specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
875 {
876         int ret = 0;
877
878         if (!irqs_disabled())
879                 BUG();
880         assert_spin_locked(&t->sighand->siglock);
881
882         /* Short-circuit ignored signals.  */
883         if (sig_ignored(t, sig))
884                 goto out;
885
886         /* Support queueing exactly one non-rt signal, so that we
887            can get more detailed information about the cause of
888            the signal. */
889         if (LEGACY_QUEUE(&t->pending, sig))
890                 goto out;
891
892         ret = send_signal(sig, info, t, &t->pending);
893         if (!ret && !sigismember(&t->blocked, sig))
894                 signal_wake_up(t, sig == SIGKILL);
895 out:
896         return ret;
897 }
898
899 /*
900  * Force a signal that the process can't ignore: if necessary
901  * we unblock the signal and change any SIG_IGN to SIG_DFL.
902  */
903
904 int
905 force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
906 {
907         unsigned long int flags;
908         int ret;
909
910         spin_lock_irqsave(&t->sighand->siglock, flags);
911         if (t->sighand->action[sig-1].sa.sa_handler == SIG_IGN) {
912                 t->sighand->action[sig-1].sa.sa_handler = SIG_DFL;
913         }
914         if (sigismember(&t->blocked, sig)) {
915                 sigdelset(&t->blocked, sig);
916         }
917         recalc_sigpending_tsk(t);
918         ret = specific_send_sig_info(sig, info, t);
919         spin_unlock_irqrestore(&t->sighand->siglock, flags);
920
921         return ret;
922 }
923
924 void
925 force_sig_specific(int sig, struct task_struct *t)
926 {
927         force_sig_info(sig, SEND_SIG_FORCED, t);
928 }
929
930 /*
931  * Test if P wants to take SIG.  After we've checked all threads with this,
932  * it's equivalent to finding no threads not blocking SIG.  Any threads not
933  * blocking SIG were ruled out because they are not running and already
934  * have pending signals.  Such threads will dequeue from the shared queue
935  * as soon as they're available, so putting the signal on the shared queue
936  * will be equivalent to sending it to one such thread.
937  */
938 static inline int wants_signal(int sig, struct task_struct *p)
939 {
940         if (sigismember(&p->blocked, sig))
941                 return 0;
942         if (p->flags & PF_EXITING)
943                 return 0;
944         if (sig == SIGKILL)
945                 return 1;
946         if (p->state & (TASK_STOPPED | TASK_TRACED))
947                 return 0;
948         return task_curr(p) || !signal_pending(p);
949 }
950
951 static void
952 __group_complete_signal(int sig, struct task_struct *p)
953 {
954         struct task_struct *t;
955
956         /*
957          * Now find a thread we can wake up to take the signal off the queue.
958          *
959          * If the main thread wants the signal, it gets first crack.
960          * Probably the least surprising to the average bear.
961          */
962         if (wants_signal(sig, p))
963                 t = p;
964         else if (thread_group_empty(p))
965                 /*
966                  * There is just one thread and it does not need to be woken.
967                  * It will dequeue unblocked signals before it runs again.
968                  */
969                 return;
970         else {
971                 /*
972                  * Otherwise try to find a suitable thread.
973                  */
974                 t = p->signal->curr_target;
975                 if (t == NULL)
976                         /* restart balancing at this thread */
977                         t = p->signal->curr_target = p;
978                 BUG_ON(t->tgid != p->tgid);
979
980                 while (!wants_signal(sig, t)) {
981                         t = next_thread(t);
982                         if (t == p->signal->curr_target)
983                                 /*
984                                  * No thread needs to be woken.
985                                  * Any eligible threads will see
986                                  * the signal in the queue soon.
987                                  */
988                                 return;
989                 }
990                 p->signal->curr_target = t;
991         }
992
993         /*
994          * Found a killable thread.  If the signal will be fatal,
995          * then start taking the whole group down immediately.
996          */
997         if (sig_fatal(p, sig) && !(p->signal->flags & SIGNAL_GROUP_EXIT) &&
998             !sigismember(&t->real_blocked, sig) &&
999             (sig == SIGKILL || !(t->ptrace & PT_PTRACED))) {
1000                 /*
1001                  * This signal will be fatal to the whole group.
1002                  */
1003                 if (!sig_kernel_coredump(sig)) {
1004                         /*
1005                          * Start a group exit and wake everybody up.
1006                          * This way we don't have other threads
1007                          * running and doing things after a slower
1008                          * thread has the fatal signal pending.
1009                          */
1010                         p->signal->flags = SIGNAL_GROUP_EXIT;
1011                         p->signal->group_exit_code = sig;
1012                         p->signal->group_stop_count = 0;
1013                         t = p;
1014                         do {
1015                                 sigaddset(&t->pending.signal, SIGKILL);
1016                                 signal_wake_up(t, 1);
1017                                 t = next_thread(t);
1018                         } while (t != p);
1019                         return;
1020                 }
1021
1022                 /*
1023                  * There will be a core dump.  We make all threads other
1024                  * than the chosen one go into a group stop so that nothing
1025                  * happens until it gets scheduled, takes the signal off
1026                  * the shared queue, and does the core dump.  This is a
1027                  * little more complicated than strictly necessary, but it
1028                  * keeps the signal state that winds up in the core dump
1029                  * unchanged from the death state, e.g. which thread had
1030                  * the core-dump signal unblocked.
1031                  */
1032                 rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
1033                 rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
1034                 p->signal->group_stop_count = 0;
1035                 p->signal->group_exit_task = t;
1036                 t = p;
1037                 do {
1038                         p->signal->group_stop_count++;
1039                         signal_wake_up(t, 0);
1040                         t = next_thread(t);
1041                 } while (t != p);
1042                 wake_up_process(p->signal->group_exit_task);
1043                 return;
1044         }
1045
1046         /*
1047          * The signal is already in the shared-pending queue.
1048          * Tell the chosen thread to wake up and dequeue it.
1049          */
1050         signal_wake_up(t, sig == SIGKILL);
1051         return;
1052 }
1053
1054 int
1055 __group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1056 {
1057         int ret = 0;
1058
1059         assert_spin_locked(&p->sighand->siglock);
1060         handle_stop_signal(sig, p);
1061
1062         /* Short-circuit ignored signals.  */
1063         if (sig_ignored(p, sig))
1064                 return ret;
1065
1066         if (LEGACY_QUEUE(&p->signal->shared_pending, sig))
1067                 /* This is a non-RT signal and we already have one queued.  */
1068                 return ret;
1069
1070         /*
1071          * Put this signal on the shared-pending queue, or fail with EAGAIN.
1072          * We always use the shared queue for process-wide signals,
1073          * to avoid several races.
1074          */
1075         ret = send_signal(sig, info, p, &p->signal->shared_pending);
1076         if (unlikely(ret))
1077                 return ret;
1078
1079         __group_complete_signal(sig, p);
1080         return 0;
1081 }
1082
1083 /*
1084  * Nuke all other threads in the group.
1085  */
1086 void zap_other_threads(struct task_struct *p)
1087 {
1088         struct task_struct *t;
1089
1090         p->signal->flags = SIGNAL_GROUP_EXIT;
1091         p->signal->group_stop_count = 0;
1092
1093         if (thread_group_empty(p))
1094                 return;
1095
1096         for (t = next_thread(p); t != p; t = next_thread(t)) {
1097                 /*
1098                  * Don't bother with already dead threads
1099                  */
1100                 if (t->exit_state)
1101                         continue;
1102
1103                 /*
1104                  * We don't want to notify the parent, since we are
1105                  * killed as part of a thread group due to another
1106                  * thread doing an execve() or similar. So set the
1107                  * exit signal to -1 to allow immediate reaping of
1108                  * the process.  But don't detach the thread group
1109                  * leader.
1110                  */
1111                 if (t != p->group_leader)
1112                         t->exit_signal = -1;
1113
1114                 /* SIGKILL will be handled before any pending SIGSTOP */
1115                 sigaddset(&t->pending.signal, SIGKILL);
1116                 signal_wake_up(t, 1);
1117         }
1118 }
1119
1120 /*
1121  * Must be called under rcu_read_lock() or with tasklist_lock read-held.
1122  */
1123 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1124 {
1125         unsigned long flags;
1126         struct sighand_struct *sp;
1127         int ret;
1128
1129 retry:
1130         ret = check_kill_permission(sig, info, p);
1131         if (!ret && sig && (sp = rcu_dereference(p->sighand))) {
1132                 spin_lock_irqsave(&sp->siglock, flags);
1133                 if (p->sighand != sp) {
1134                         spin_unlock_irqrestore(&sp->siglock, flags);
1135                         goto retry;
1136                 }
1137                 if ((atomic_read(&sp->count) == 0) ||
1138                                 (atomic_read(&p->usage) == 0)) {
1139                         spin_unlock_irqrestore(&sp->siglock, flags);
1140                         return -ESRCH;
1141                 }
1142                 ret = __group_send_sig_info(sig, info, p);
1143                 spin_unlock_irqrestore(&sp->siglock, flags);
1144         }
1145
1146         return ret;
1147 }
1148
1149 /*
1150  * kill_pg_info() sends a signal to a process group: this is what the tty
1151  * control characters do (^C, ^Z etc)
1152  */
1153
1154 int __kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)
1155 {
1156         struct task_struct *p = NULL;
1157         int retval, success;
1158
1159         if (pgrp <= 0)
1160                 return -EINVAL;
1161
1162         success = 0;
1163         retval = -ESRCH;
1164         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
1165                 int err = group_send_sig_info(sig, info, p);
1166                 success |= !err;
1167                 retval = err;
1168         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
1169         return success ? 0 : retval;
1170 }
1171
1172 int
1173 kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)
1174 {
1175         int retval;
1176
1177         read_lock(&tasklist_lock);
1178         retval = __kill_pg_info(sig, info, pgrp);
1179         read_unlock(&tasklist_lock);
1180
1181         return retval;
1182 }
1183
1184 int
1185 kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1186 {
1187         int error;
1188         int acquired_tasklist_lock = 0;
1189         struct task_struct *p;
1190
1191         rcu_read_lock();
1192         if (unlikely(sig_kernel_stop(sig) || sig == SIGCONT)) {
1193                 read_lock(&tasklist_lock);
1194                 acquired_tasklist_lock = 1;
1195         }
1196         p = find_task_by_pid(pid);
1197         error = -ESRCH;
1198         if (p)
1199                 error = group_send_sig_info(sig, info, p);
1200         if (unlikely(acquired_tasklist_lock))
1201                 read_unlock(&tasklist_lock);
1202         rcu_read_unlock();
1203         return error;
1204 }
1205
1206 /* like kill_proc_info(), but doesn't use uid/euid of "current" */
1207 int kill_proc_info_as_uid(int sig, struct siginfo *info, pid_t pid,
1208                       uid_t uid, uid_t euid)
1209 {
1210         int ret = -EINVAL;
1211         struct task_struct *p;
1212
1213         if (!valid_signal(sig))
1214                 return ret;
1215
1216         read_lock(&tasklist_lock);
1217         p = find_task_by_pid(pid);
1218         if (!p) {
1219                 ret = -ESRCH;
1220                 goto out_unlock;
1221         }
1222         if ((info == SEND_SIG_NOINFO || (!is_si_special(info) && SI_FROMUSER(info)))
1223             && (euid != p->suid) && (euid != p->uid)
1224             && (uid != p->suid) && (uid != p->uid)) {
1225                 ret = -EPERM;
1226                 goto out_unlock;
1227         }
1228         if (sig && p->sighand) {
1229                 unsigned long flags;
1230                 spin_lock_irqsave(&p->sighand->siglock, flags);
1231                 ret = __group_send_sig_info(sig, info, p);
1232                 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1233         }
1234 out_unlock:
1235         read_unlock(&tasklist_lock);
1236         return ret;
1237 }
1238 EXPORT_SYMBOL_GPL(kill_proc_info_as_uid);
1239
1240 /*
1241  * kill_something_info() interprets pid in interesting ways just like kill(2).
1242  *
1243  * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1244  * is probably wrong.  Should make it like BSD or SYSV.
1245  */
1246
1247 static int kill_something_info(int sig, struct siginfo *info, int pid)
1248 {
1249         if (!pid) {
1250                 return kill_pg_info(sig, info, process_group(current));
1251         } else if (pid == -1) {
1252                 int retval = 0, count = 0;
1253                 struct task_struct * p;
1254
1255                 read_lock(&tasklist_lock);
1256                 for_each_process(p) {
1257                         if (p->pid > 1 && p->tgid != current->tgid) {
1258                                 int err = group_send_sig_info(sig, info, p);
1259                                 ++count;
1260                                 if (err != -EPERM)
1261                                         retval = err;
1262                         }
1263                 }
1264                 read_unlock(&tasklist_lock);
1265                 return count ? retval : -ESRCH;
1266         } else if (pid < 0) {
1267                 return kill_pg_info(sig, info, -pid);
1268         } else {
1269                 return kill_proc_info(sig, info, pid);
1270         }
1271 }
1272
1273 /*
1274  * These are for backward compatibility with the rest of the kernel source.
1275  */
1276
1277 /*
1278  * These two are the most common entry points.  They send a signal
1279  * just to the specific thread.
1280  */
1281 int
1282 send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1283 {
1284         int ret;
1285         unsigned long flags;
1286
1287         /*
1288          * Make sure legacy kernel users don't send in bad values
1289          * (normal paths check this in check_kill_permission).
1290          */
1291         if (!valid_signal(sig))
1292                 return -EINVAL;
1293
1294         /*
1295          * We need the tasklist lock even for the specific
1296          * thread case (when we don't need to follow the group
1297          * lists) in order to avoid races with "p->sighand"
1298          * going away or changing from under us.
1299          */
1300         read_lock(&tasklist_lock);  
1301         spin_lock_irqsave(&p->sighand->siglock, flags);
1302         ret = specific_send_sig_info(sig, info, p);
1303         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1304         read_unlock(&tasklist_lock);
1305         return ret;
1306 }
1307
1308 #define __si_special(priv) \
1309         ((priv) ? SEND_SIG_PRIV : SEND_SIG_NOINFO)
1310
1311 int
1312 send_sig(int sig, struct task_struct *p, int priv)
1313 {
1314         return send_sig_info(sig, __si_special(priv), p);
1315 }
1316
1317 /*
1318  * This is the entry point for "process-wide" signals.
1319  * They will go to an appropriate thread in the thread group.
1320  */
1321 int
1322 send_group_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1323 {
1324         int ret;
1325         read_lock(&tasklist_lock);
1326         ret = group_send_sig_info(sig, info, p);
1327         read_unlock(&tasklist_lock);
1328         return ret;
1329 }
1330
1331 void
1332 force_sig(int sig, struct task_struct *p)
1333 {
1334         force_sig_info(sig, SEND_SIG_PRIV, p);
1335 }
1336
1337 /*
1338  * When things go south during signal handling, we
1339  * will force a SIGSEGV. And if the signal that caused
1340  * the problem was already a SIGSEGV, we'll want to
1341  * make sure we don't even try to deliver the signal..
1342  */
1343 int
1344 force_sigsegv(int sig, struct task_struct *p)
1345 {
1346         if (sig == SIGSEGV) {
1347                 unsigned long flags;
1348                 spin_lock_irqsave(&p->sighand->siglock, flags);
1349                 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1350                 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1351         }
1352         force_sig(SIGSEGV, p);
1353         return 0;
1354 }
1355
1356 int
1357 kill_pg(pid_t pgrp, int sig, int priv)
1358 {
1359         return kill_pg_info(sig, __si_special(priv), pgrp);
1360 }
1361
1362 int
1363 kill_proc(pid_t pid, int sig, int priv)
1364 {
1365         return kill_proc_info(sig, __si_special(priv), pid);
1366 }
1367
1368 /*
1369  * These functions support sending signals using preallocated sigqueue
1370  * structures.  This is needed "because realtime applications cannot
1371  * afford to lose notifications of asynchronous events, like timer
1372  * expirations or I/O completions".  In the case of Posix Timers 
1373  * we allocate the sigqueue structure from the timer_create.  If this
1374  * allocation fails we are able to report the failure to the application
1375  * with an EAGAIN error.
1376  */
1377  
1378 struct sigqueue *sigqueue_alloc(void)
1379 {
1380         struct sigqueue *q;
1381
1382         if ((q = __sigqueue_alloc(current, GFP_KERNEL, 0)))
1383                 q->flags |= SIGQUEUE_PREALLOC;
1384         return(q);
1385 }
1386
1387 void sigqueue_free(struct sigqueue *q)
1388 {
1389         unsigned long flags;
1390         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1391         /*
1392          * If the signal is still pending remove it from the
1393          * pending queue.
1394          */
1395         if (unlikely(!list_empty(&q->list))) {
1396                 spinlock_t *lock = &current->sighand->siglock;
1397                 read_lock(&tasklist_lock);
1398                 spin_lock_irqsave(lock, flags);
1399                 if (!list_empty(&q->list))
1400                         list_del_init(&q->list);
1401                 spin_unlock_irqrestore(lock, flags);
1402                 read_unlock(&tasklist_lock);
1403         }
1404         q->flags &= ~SIGQUEUE_PREALLOC;
1405         __sigqueue_free(q);
1406 }
1407
1408 int
1409 send_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
1410 {
1411         unsigned long flags;
1412         int ret = 0;
1413         struct sighand_struct *sh;
1414
1415         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1416
1417         /*
1418          * The rcu based delayed sighand destroy makes it possible to
1419          * run this without tasklist lock held. The task struct itself
1420          * cannot go away as create_timer did get_task_struct().
1421          *
1422          * We return -1, when the task is marked exiting, so
1423          * posix_timer_event can redirect it to the group leader
1424          */
1425         rcu_read_lock();
1426
1427         if (unlikely(p->flags & PF_EXITING)) {
1428                 ret = -1;
1429                 goto out_err;
1430         }
1431
1432 retry:
1433         sh = rcu_dereference(p->sighand);
1434
1435         spin_lock_irqsave(&sh->siglock, flags);
1436         if (p->sighand != sh) {
1437                 /* We raced with exec() in a multithreaded process... */
1438                 spin_unlock_irqrestore(&sh->siglock, flags);
1439                 goto retry;
1440         }
1441
1442         /*
1443          * We do the check here again to handle the following scenario:
1444          *
1445          * CPU 0                CPU 1
1446          * send_sigqueue
1447          * check PF_EXITING
1448          * interrupt            exit code running
1449          *                      __exit_signal
1450          *                      lock sighand->siglock
1451          *                      unlock sighand->siglock
1452          * lock sh->siglock
1453          * add(tsk->pending)    flush_sigqueue(tsk->pending)
1454          *
1455          */
1456
1457         if (unlikely(p->flags & PF_EXITING)) {
1458                 ret = -1;
1459                 goto out;
1460         }
1461
1462         if (unlikely(!list_empty(&q->list))) {
1463                 /*
1464                  * If an SI_TIMER entry is already queue just increment
1465                  * the overrun count.
1466                  */
1467                 if (q->info.si_code != SI_TIMER)
1468                         BUG();
1469                 q->info.si_overrun++;
1470                 goto out;
1471         }
1472         /* Short-circuit ignored signals.  */
1473         if (sig_ignored(p, sig)) {
1474                 ret = 1;
1475                 goto out;
1476         }
1477
1478         list_add_tail(&q->list, &p->pending.list);
1479         sigaddset(&p->pending.signal, sig);
1480         if (!sigismember(&p->blocked, sig))
1481                 signal_wake_up(p, sig == SIGKILL);
1482
1483 out:
1484         spin_unlock_irqrestore(&sh->siglock, flags);
1485 out_err:
1486         rcu_read_unlock();
1487
1488         return ret;
1489 }
1490
1491 int
1492 send_group_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
1493 {
1494         unsigned long flags;
1495         int ret = 0;
1496
1497         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1498
1499         read_lock(&tasklist_lock);
1500         /* Since it_lock is held, p->sighand cannot be NULL. */
1501         spin_lock_irqsave(&p->sighand->siglock, flags);
1502         handle_stop_signal(sig, p);
1503
1504         /* Short-circuit ignored signals.  */
1505         if (sig_ignored(p, sig)) {
1506                 ret = 1;
1507                 goto out;
1508         }
1509
1510         if (unlikely(!list_empty(&q->list))) {
1511                 /*
1512                  * If an SI_TIMER entry is already queue just increment
1513                  * the overrun count.  Other uses should not try to
1514                  * send the signal multiple times.
1515                  */
1516                 if (q->info.si_code != SI_TIMER)
1517                         BUG();
1518                 q->info.si_overrun++;
1519                 goto out;
1520         } 
1521
1522         /*
1523          * Put this signal on the shared-pending queue.
1524          * We always use the shared queue for process-wide signals,
1525          * to avoid several races.
1526          */
1527         list_add_tail(&q->list, &p->signal->shared_pending.list);
1528         sigaddset(&p->signal->shared_pending.signal, sig);
1529
1530         __group_complete_signal(sig, p);
1531 out:
1532         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1533         read_unlock(&tasklist_lock);
1534         return ret;
1535 }
1536
1537 /*
1538  * Wake up any threads in the parent blocked in wait* syscalls.
1539  */
1540 static inline void __wake_up_parent(struct task_struct *p,
1541                                     struct task_struct *parent)
1542 {
1543         wake_up_interruptible_sync(&parent->signal->wait_chldexit);
1544 }
1545
1546 /*
1547  * Let a parent know about the death of a child.
1548  * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1549  */
1550
1551 void do_notify_parent(struct task_struct *tsk, int sig)
1552 {
1553         struct siginfo info;
1554         unsigned long flags;
1555         struct sighand_struct *psig;
1556
1557         BUG_ON(sig == -1);
1558
1559         /* do_notify_parent_cldstop should have been called instead.  */
1560         BUG_ON(tsk->state & (TASK_STOPPED|TASK_TRACED));
1561
1562         BUG_ON(!tsk->ptrace &&
1563                (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1564
1565         info.si_signo = sig;
1566         info.si_errno = 0;
1567         info.si_pid = tsk->pid;
1568         info.si_uid = tsk->uid;
1569
1570         /* FIXME: find out whether or not this is supposed to be c*time. */
1571         info.si_utime = cputime_to_jiffies(cputime_add(tsk->utime,
1572                                                        tsk->signal->utime));
1573         info.si_stime = cputime_to_jiffies(cputime_add(tsk->stime,
1574                                                        tsk->signal->stime));
1575
1576         info.si_status = tsk->exit_code & 0x7f;
1577         if (tsk->exit_code & 0x80)
1578                 info.si_code = CLD_DUMPED;
1579         else if (tsk->exit_code & 0x7f)
1580                 info.si_code = CLD_KILLED;
1581         else {
1582                 info.si_code = CLD_EXITED;
1583                 info.si_status = tsk->exit_code >> 8;
1584         }
1585
1586         psig = tsk->parent->sighand;
1587         spin_lock_irqsave(&psig->siglock, flags);
1588         if (!tsk->ptrace && sig == SIGCHLD &&
1589             (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1590              (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1591                 /*
1592                  * We are exiting and our parent doesn't care.  POSIX.1
1593                  * defines special semantics for setting SIGCHLD to SIG_IGN
1594                  * or setting the SA_NOCLDWAIT flag: we should be reaped
1595                  * automatically and not left for our parent's wait4 call.
1596                  * Rather than having the parent do it as a magic kind of
1597                  * signal handler, we just set this to tell do_exit that we
1598                  * can be cleaned up without becoming a zombie.  Note that
1599                  * we still call __wake_up_parent in this case, because a
1600                  * blocked sys_wait4 might now return -ECHILD.
1601                  *
1602                  * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1603                  * is implementation-defined: we do (if you don't want
1604                  * it, just use SIG_IGN instead).
1605                  */
1606                 tsk->exit_signal = -1;
1607                 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1608                         sig = 0;
1609         }
1610         if (valid_signal(sig) && sig > 0)
1611                 __group_send_sig_info(sig, &info, tsk->parent);
1612         __wake_up_parent(tsk, tsk->parent);
1613         spin_unlock_irqrestore(&psig->siglock, flags);
1614 }
1615
1616 static void do_notify_parent_cldstop(struct task_struct *tsk, int to_self, int why)
1617 {
1618         struct siginfo info;
1619         unsigned long flags;
1620         struct task_struct *parent;
1621         struct sighand_struct *sighand;
1622
1623         if (to_self)
1624                 parent = tsk->parent;
1625         else {
1626                 tsk = tsk->group_leader;
1627                 parent = tsk->real_parent;
1628         }
1629
1630         info.si_signo = SIGCHLD;
1631         info.si_errno = 0;
1632         info.si_pid = tsk->pid;
1633         info.si_uid = tsk->uid;
1634
1635         /* FIXME: find out whether or not this is supposed to be c*time. */
1636         info.si_utime = cputime_to_jiffies(tsk->utime);
1637         info.si_stime = cputime_to_jiffies(tsk->stime);
1638
1639         info.si_code = why;
1640         switch (why) {
1641         case CLD_CONTINUED:
1642                 info.si_status = SIGCONT;
1643                 break;
1644         case CLD_STOPPED:
1645                 info.si_status = tsk->signal->group_exit_code & 0x7f;
1646                 break;
1647         case CLD_TRAPPED:
1648                 info.si_status = tsk->exit_code & 0x7f;
1649                 break;
1650         default:
1651                 BUG();
1652         }
1653
1654         sighand = parent->sighand;
1655         spin_lock_irqsave(&sighand->siglock, flags);
1656         if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1657             !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1658                 __group_send_sig_info(SIGCHLD, &info, parent);
1659         /*
1660          * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1661          */
1662         __wake_up_parent(tsk, parent);
1663         spin_unlock_irqrestore(&sighand->siglock, flags);
1664 }
1665
1666 /*
1667  * This must be called with current->sighand->siglock held.
1668  *
1669  * This should be the path for all ptrace stops.
1670  * We always set current->last_siginfo while stopped here.
1671  * That makes it a way to test a stopped process for
1672  * being ptrace-stopped vs being job-control-stopped.
1673  *
1674  * If we actually decide not to stop at all because the tracer is gone,
1675  * we leave nostop_code in current->exit_code.
1676  */
1677 static void ptrace_stop(int exit_code, int nostop_code, siginfo_t *info)
1678 {
1679         /*
1680          * If there is a group stop in progress,
1681          * we must participate in the bookkeeping.
1682          */
1683         if (current->signal->group_stop_count > 0)
1684                 --current->signal->group_stop_count;
1685
1686         current->last_siginfo = info;
1687         current->exit_code = exit_code;
1688
1689         /* Let the debugger run.  */
1690         set_current_state(TASK_TRACED);
1691         spin_unlock_irq(&current->sighand->siglock);
1692         read_lock(&tasklist_lock);
1693         if (likely(current->ptrace & PT_PTRACED) &&
1694             likely(current->parent != current->real_parent ||
1695                    !(current->ptrace & PT_ATTACHED)) &&
1696             (likely(current->parent->signal != current->signal) ||
1697              !unlikely(current->signal->flags & SIGNAL_GROUP_EXIT))) {
1698                 do_notify_parent_cldstop(current, 1, CLD_TRAPPED);
1699                 read_unlock(&tasklist_lock);
1700                 schedule();
1701         } else {
1702                 /*
1703                  * By the time we got the lock, our tracer went away.
1704                  * Don't stop here.
1705                  */
1706                 read_unlock(&tasklist_lock);
1707                 set_current_state(TASK_RUNNING);
1708                 current->exit_code = nostop_code;
1709         }
1710
1711         /*
1712          * We are back.  Now reacquire the siglock before touching
1713          * last_siginfo, so that we are sure to have synchronized with
1714          * any signal-sending on another CPU that wants to examine it.
1715          */
1716         spin_lock_irq(&current->sighand->siglock);
1717         current->last_siginfo = NULL;
1718
1719         /*
1720          * Queued signals ignored us while we were stopped for tracing.
1721          * So check for any that we should take before resuming user mode.
1722          */
1723         recalc_sigpending();
1724 }
1725
1726 void ptrace_notify(int exit_code)
1727 {
1728         siginfo_t info;
1729
1730         BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1731
1732         memset(&info, 0, sizeof info);
1733         info.si_signo = SIGTRAP;
1734         info.si_code = exit_code;
1735         info.si_pid = current->pid;
1736         info.si_uid = current->uid;
1737
1738         /* Let the debugger run.  */
1739         spin_lock_irq(&current->sighand->siglock);
1740         ptrace_stop(exit_code, 0, &info);
1741         spin_unlock_irq(&current->sighand->siglock);
1742 }
1743
1744 static void
1745 finish_stop(int stop_count)
1746 {
1747         int to_self;
1748
1749         /*
1750          * If there are no other threads in the group, or if there is
1751          * a group stop in progress and we are the last to stop,
1752          * report to the parent.  When ptraced, every thread reports itself.
1753          */
1754         if (stop_count < 0 || (current->ptrace & PT_PTRACED))
1755                 to_self = 1;
1756         else if (stop_count == 0)
1757                 to_self = 0;
1758         else
1759                 goto out;
1760
1761         read_lock(&tasklist_lock);
1762         do_notify_parent_cldstop(current, to_self, CLD_STOPPED);
1763         read_unlock(&tasklist_lock);
1764
1765 out:
1766         schedule();
1767         /*
1768          * Now we don't run again until continued.
1769          */
1770         current->exit_code = 0;
1771 }
1772
1773 /*
1774  * This performs the stopping for SIGSTOP and other stop signals.
1775  * We have to stop all threads in the thread group.
1776  * Returns nonzero if we've actually stopped and released the siglock.
1777  * Returns zero if we didn't stop and still hold the siglock.
1778  */
1779 static int
1780 do_signal_stop(int signr)
1781 {
1782         struct signal_struct *sig = current->signal;
1783         struct sighand_struct *sighand = current->sighand;
1784         int stop_count = -1;
1785
1786         if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED))
1787                 return 0;
1788
1789         if (sig->group_stop_count > 0) {
1790                 /*
1791                  * There is a group stop in progress.  We don't need to
1792                  * start another one.
1793                  */
1794                 signr = sig->group_exit_code;
1795                 stop_count = --sig->group_stop_count;
1796                 current->exit_code = signr;
1797                 set_current_state(TASK_STOPPED);
1798                 if (stop_count == 0)
1799                         sig->flags = SIGNAL_STOP_STOPPED;
1800                 spin_unlock_irq(&sighand->siglock);
1801         }
1802         else if (thread_group_empty(current)) {
1803                 /*
1804                  * Lock must be held through transition to stopped state.
1805                  */
1806                 current->exit_code = current->signal->group_exit_code = signr;
1807                 set_current_state(TASK_STOPPED);
1808                 sig->flags = SIGNAL_STOP_STOPPED;
1809                 spin_unlock_irq(&sighand->siglock);
1810         }
1811         else {
1812                 /*
1813                  * There is no group stop already in progress.
1814                  * We must initiate one now, but that requires
1815                  * dropping siglock to get both the tasklist lock
1816                  * and siglock again in the proper order.  Note that
1817                  * this allows an intervening SIGCONT to be posted.
1818                  * We need to check for that and bail out if necessary.
1819                  */
1820                 struct task_struct *t;
1821
1822                 spin_unlock_irq(&sighand->siglock);
1823
1824                 /* signals can be posted during this window */
1825
1826                 read_lock(&tasklist_lock);
1827                 spin_lock_irq(&sighand->siglock);
1828
1829                 if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED)) {
1830                         /*
1831                          * Another stop or continue happened while we
1832                          * didn't have the lock.  We can just swallow this
1833                          * signal now.  If we raced with a SIGCONT, that
1834                          * should have just cleared it now.  If we raced
1835                          * with another processor delivering a stop signal,
1836                          * then the SIGCONT that wakes us up should clear it.
1837                          */
1838                         read_unlock(&tasklist_lock);
1839                         return 0;
1840                 }
1841
1842                 if (sig->group_stop_count == 0) {
1843                         sig->group_exit_code = signr;
1844                         stop_count = 0;
1845                         for (t = next_thread(current); t != current;
1846                              t = next_thread(t))
1847                                 /*
1848                                  * Setting state to TASK_STOPPED for a group
1849                                  * stop is always done with the siglock held,
1850                                  * so this check has no races.
1851                                  */
1852                                 if (!t->exit_state &&
1853                                     !(t->state & (TASK_STOPPED|TASK_TRACED))) {
1854                                         stop_count++;
1855                                         signal_wake_up(t, 0);
1856                                 }
1857                         sig->group_stop_count = stop_count;
1858                 }
1859                 else {
1860                         /* A race with another thread while unlocked.  */
1861                         signr = sig->group_exit_code;
1862                         stop_count = --sig->group_stop_count;
1863                 }
1864
1865                 current->exit_code = signr;
1866                 set_current_state(TASK_STOPPED);
1867                 if (stop_count == 0)
1868                         sig->flags = SIGNAL_STOP_STOPPED;
1869
1870                 spin_unlock_irq(&sighand->siglock);
1871                 read_unlock(&tasklist_lock);
1872         }
1873
1874         finish_stop(stop_count);
1875         return 1;
1876 }
1877
1878 /*
1879  * Do appropriate magic when group_stop_count > 0.
1880  * We return nonzero if we stopped, after releasing the siglock.
1881  * We return zero if we still hold the siglock and should look
1882  * for another signal without checking group_stop_count again.
1883  */
1884 static int handle_group_stop(void)
1885 {
1886         int stop_count;
1887
1888         if (current->signal->group_exit_task == current) {
1889                 /*
1890                  * Group stop is so we can do a core dump,
1891                  * We are the initiating thread, so get on with it.
1892                  */
1893                 current->signal->group_exit_task = NULL;
1894                 return 0;
1895         }
1896
1897         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1898                 /*
1899                  * Group stop is so another thread can do a core dump,
1900                  * or else we are racing against a death signal.
1901                  * Just punt the stop so we can get the next signal.
1902                  */
1903                 return 0;
1904
1905         /*
1906          * There is a group stop in progress.  We stop
1907          * without any associated signal being in our queue.
1908          */
1909         stop_count = --current->signal->group_stop_count;
1910         if (stop_count == 0)
1911                 current->signal->flags = SIGNAL_STOP_STOPPED;
1912         current->exit_code = current->signal->group_exit_code;
1913         set_current_state(TASK_STOPPED);
1914         spin_unlock_irq(&current->sighand->siglock);
1915         finish_stop(stop_count);
1916         return 1;
1917 }
1918
1919 int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
1920                           struct pt_regs *regs, void *cookie)
1921 {
1922         sigset_t *mask = &current->blocked;
1923         int signr = 0;
1924
1925 relock:
1926         spin_lock_irq(&current->sighand->siglock);
1927         for (;;) {
1928                 struct k_sigaction *ka;
1929
1930                 if (unlikely(current->signal->group_stop_count > 0) &&
1931                     handle_group_stop())
1932                         goto relock;
1933
1934                 signr = dequeue_signal(current, mask, info);
1935
1936                 if (!signr)
1937                         break; /* will return 0 */
1938
1939                 if ((current->ptrace & PT_PTRACED) && signr != SIGKILL) {
1940                         ptrace_signal_deliver(regs, cookie);
1941
1942                         /* Let the debugger run.  */
1943                         ptrace_stop(signr, signr, info);
1944
1945                         /* We're back.  Did the debugger cancel the sig or group_exit? */
1946                         signr = current->exit_code;
1947                         if (signr == 0 || current->signal->flags & SIGNAL_GROUP_EXIT)
1948                                 continue;
1949
1950                         current->exit_code = 0;
1951
1952                         /* Update the siginfo structure if the signal has
1953                            changed.  If the debugger wanted something
1954                            specific in the siginfo structure then it should
1955                            have updated *info via PTRACE_SETSIGINFO.  */
1956                         if (signr != info->si_signo) {
1957                                 info->si_signo = signr;
1958                                 info->si_errno = 0;
1959                                 info->si_code = SI_USER;
1960                                 info->si_pid = current->parent->pid;
1961                                 info->si_uid = current->parent->uid;
1962                         }
1963
1964                         /* If the (new) signal is now blocked, requeue it.  */
1965                         if (sigismember(&current->blocked, signr)) {
1966                                 specific_send_sig_info(signr, info, current);
1967                                 continue;
1968                         }
1969                 }
1970
1971                 ka = &current->sighand->action[signr-1];
1972                 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
1973                         continue;
1974                 if (ka->sa.sa_handler != SIG_DFL) {
1975                         /* Run the handler.  */
1976                         *return_ka = *ka;
1977
1978                         if (ka->sa.sa_flags & SA_ONESHOT)
1979                                 ka->sa.sa_handler = SIG_DFL;
1980
1981                         break; /* will return non-zero "signr" value */
1982                 }
1983
1984                 /*
1985                  * Now we are doing the default action for this signal.
1986                  */
1987                 if (sig_kernel_ignore(signr)) /* Default is nothing. */
1988                         continue;
1989
1990                 /* Init gets no signals it doesn't want.  */
1991                 if (current->pid == 1)
1992                         continue;
1993
1994                 if (sig_kernel_stop(signr)) {
1995                         /*
1996                          * The default action is to stop all threads in
1997                          * the thread group.  The job control signals
1998                          * do nothing in an orphaned pgrp, but SIGSTOP
1999                          * always works.  Note that siglock needs to be
2000                          * dropped during the call to is_orphaned_pgrp()
2001                          * because of lock ordering with tasklist_lock.
2002                          * This allows an intervening SIGCONT to be posted.
2003                          * We need to check for that and bail out if necessary.
2004                          */
2005                         if (signr != SIGSTOP) {
2006                                 spin_unlock_irq(&current->sighand->siglock);
2007
2008                                 /* signals can be posted during this window */
2009
2010                                 if (is_orphaned_pgrp(process_group(current)))
2011                                         goto relock;
2012
2013                                 spin_lock_irq(&current->sighand->siglock);
2014                         }
2015
2016                         if (likely(do_signal_stop(signr))) {
2017                                 /* It released the siglock.  */
2018                                 goto relock;
2019                         }
2020
2021                         /*
2022                          * We didn't actually stop, due to a race
2023                          * with SIGCONT or something like that.
2024                          */
2025                         continue;
2026                 }
2027
2028                 spin_unlock_irq(&current->sighand->siglock);
2029
2030                 /*
2031                  * Anything else is fatal, maybe with a core dump.
2032                  */
2033                 current->flags |= PF_SIGNALED;
2034                 if (sig_kernel_coredump(signr)) {
2035                         /*
2036                          * If it was able to dump core, this kills all
2037                          * other threads in the group and synchronizes with
2038                          * their demise.  If we lost the race with another
2039                          * thread getting here, it set group_exit_code
2040                          * first and our do_group_exit call below will use
2041                          * that value and ignore the one we pass it.
2042                          */
2043                         do_coredump((long)signr, signr, regs);
2044                 }
2045
2046                 /*
2047                  * Death signals, no core dump.
2048                  */
2049                 do_group_exit(signr);
2050                 /* NOTREACHED */
2051         }
2052         spin_unlock_irq(&current->sighand->siglock);
2053         return signr;
2054 }
2055
2056 EXPORT_SYMBOL(recalc_sigpending);
2057 EXPORT_SYMBOL_GPL(dequeue_signal);
2058 EXPORT_SYMBOL(flush_signals);
2059 EXPORT_SYMBOL(force_sig);
2060 EXPORT_SYMBOL(kill_pg);
2061 EXPORT_SYMBOL(kill_proc);
2062 EXPORT_SYMBOL(ptrace_notify);
2063 EXPORT_SYMBOL(send_sig);
2064 EXPORT_SYMBOL(send_sig_info);
2065 EXPORT_SYMBOL(sigprocmask);
2066 EXPORT_SYMBOL(block_all_signals);
2067 EXPORT_SYMBOL(unblock_all_signals);
2068
2069
2070 /*
2071  * System call entry points.
2072  */
2073
2074 asmlinkage long sys_restart_syscall(void)
2075 {
2076         struct restart_block *restart = &current_thread_info()->restart_block;
2077         return restart->fn(restart);
2078 }
2079
2080 long do_no_restart_syscall(struct restart_block *param)
2081 {
2082         return -EINTR;
2083 }
2084
2085 /*
2086  * We don't need to get the kernel lock - this is all local to this
2087  * particular thread.. (and that's good, because this is _heavily_
2088  * used by various programs)
2089  */
2090
2091 /*
2092  * This is also useful for kernel threads that want to temporarily
2093  * (or permanently) block certain signals.
2094  *
2095  * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
2096  * interface happily blocks "unblockable" signals like SIGKILL
2097  * and friends.
2098  */
2099 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2100 {
2101         int error;
2102         sigset_t old_block;
2103
2104         spin_lock_irq(&current->sighand->siglock);
2105         old_block = current->blocked;
2106         error = 0;
2107         switch (how) {
2108         case SIG_BLOCK:
2109                 sigorsets(&current->blocked, &current->blocked, set);
2110                 break;
2111         case SIG_UNBLOCK:
2112                 signandsets(&current->blocked, &current->blocked, set);
2113                 break;
2114         case SIG_SETMASK:
2115                 current->blocked = *set;
2116                 break;
2117         default:
2118                 error = -EINVAL;
2119         }
2120         recalc_sigpending();
2121         spin_unlock_irq(&current->sighand->siglock);
2122         if (oldset)
2123                 *oldset = old_block;
2124         return error;
2125 }
2126
2127 asmlinkage long
2128 sys_rt_sigprocmask(int how, sigset_t __user *set, sigset_t __user *oset, size_t sigsetsize)
2129 {
2130         int error = -EINVAL;
2131         sigset_t old_set, new_set;
2132
2133         /* XXX: Don't preclude handling different sized sigset_t's.  */
2134         if (sigsetsize != sizeof(sigset_t))
2135                 goto out;
2136
2137         if (set) {
2138                 error = -EFAULT;
2139                 if (copy_from_user(&new_set, set, sizeof(*set)))
2140                         goto out;
2141                 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2142
2143                 error = sigprocmask(how, &new_set, &old_set);
2144                 if (error)
2145                         goto out;
2146                 if (oset)
2147                         goto set_old;
2148         } else if (oset) {
2149                 spin_lock_irq(&current->sighand->siglock);
2150                 old_set = current->blocked;
2151                 spin_unlock_irq(&current->sighand->siglock);
2152
2153         set_old:
2154                 error = -EFAULT;
2155                 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2156                         goto out;
2157         }
2158         error = 0;
2159 out:
2160         return error;
2161 }
2162
2163 long do_sigpending(void __user *set, unsigned long sigsetsize)
2164 {
2165         long error = -EINVAL;
2166         sigset_t pending;
2167
2168         if (sigsetsize > sizeof(sigset_t))
2169                 goto out;
2170
2171         spin_lock_irq(&current->sighand->siglock);
2172         sigorsets(&pending, &current->pending.signal,
2173                   &current->signal->shared_pending.signal);
2174         spin_unlock_irq(&current->sighand->siglock);
2175
2176         /* Outside the lock because only this thread touches it.  */
2177         sigandsets(&pending, &current->blocked, &pending);
2178
2179         error = -EFAULT;
2180         if (!copy_to_user(set, &pending, sigsetsize))
2181                 error = 0;
2182
2183 out:
2184         return error;
2185 }       
2186
2187 asmlinkage long
2188 sys_rt_sigpending(sigset_t __user *set, size_t sigsetsize)
2189 {
2190         return do_sigpending(set, sigsetsize);
2191 }
2192
2193 #ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2194
2195 int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2196 {
2197         int err;
2198
2199         if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2200                 return -EFAULT;
2201         if (from->si_code < 0)
2202                 return __copy_to_user(to, from, sizeof(siginfo_t))
2203                         ? -EFAULT : 0;
2204         /*
2205          * If you change siginfo_t structure, please be sure
2206          * this code is fixed accordingly.
2207          * It should never copy any pad contained in the structure
2208          * to avoid security leaks, but must copy the generic
2209          * 3 ints plus the relevant union member.
2210          */
2211         err = __put_user(from->si_signo, &to->si_signo);
2212         err |= __put_user(from->si_errno, &to->si_errno);
2213         err |= __put_user((short)from->si_code, &to->si_code);
2214         switch (from->si_code & __SI_MASK) {
2215         case __SI_KILL:
2216                 err |= __put_user(from->si_pid, &to->si_pid);
2217                 err |= __put_user(from->si_uid, &to->si_uid);
2218                 break;
2219         case __SI_TIMER:
2220                  err |= __put_user(from->si_tid, &to->si_tid);
2221                  err |= __put_user(from->si_overrun, &to->si_overrun);
2222                  err |= __put_user(from->si_ptr, &to->si_ptr);
2223                 break;
2224         case __SI_POLL:
2225                 err |= __put_user(from->si_band, &to->si_band);
2226                 err |= __put_user(from->si_fd, &to->si_fd);
2227                 break;
2228         case __SI_FAULT:
2229                 err |= __put_user(from->si_addr, &to->si_addr);
2230 #ifdef __ARCH_SI_TRAPNO
2231                 err |= __put_user(from->si_trapno, &to->si_trapno);
2232 #endif
2233                 break;
2234         case __SI_CHLD:
2235                 err |= __put_user(from->si_pid, &to->si_pid);
2236                 err |= __put_user(from->si_uid, &to->si_uid);
2237                 err |= __put_user(from->si_status, &to->si_status);
2238                 err |= __put_user(from->si_utime, &to->si_utime);
2239                 err |= __put_user(from->si_stime, &to->si_stime);
2240                 break;
2241         case __SI_RT: /* This is not generated by the kernel as of now. */
2242         case __SI_MESGQ: /* But this is */
2243                 err |= __put_user(from->si_pid, &to->si_pid);
2244                 err |= __put_user(from->si_uid, &to->si_uid);
2245                 err |= __put_user(from->si_ptr, &to->si_ptr);
2246                 break;
2247         default: /* this is just in case for now ... */
2248                 err |= __put_user(from->si_pid, &to->si_pid);
2249                 err |= __put_user(from->si_uid, &to->si_uid);
2250                 break;
2251         }
2252         return err;
2253 }
2254
2255 #endif
2256
2257 asmlinkage long
2258 sys_rt_sigtimedwait(const sigset_t __user *uthese,
2259                     siginfo_t __user *uinfo,
2260                     const struct timespec __user *uts,
2261                     size_t sigsetsize)
2262 {
2263         int ret, sig;
2264         sigset_t these;
2265         struct timespec ts;
2266         siginfo_t info;
2267         long timeout = 0;
2268
2269         /* XXX: Don't preclude handling different sized sigset_t's.  */
2270         if (sigsetsize != sizeof(sigset_t))
2271                 return -EINVAL;
2272
2273         if (copy_from_user(&these, uthese, sizeof(these)))
2274                 return -EFAULT;
2275                 
2276         /*
2277          * Invert the set of allowed signals to get those we
2278          * want to block.
2279          */
2280         sigdelsetmask(&these, sigmask(SIGKILL)|sigmask(SIGSTOP));
2281         signotset(&these);
2282
2283         if (uts) {
2284                 if (copy_from_user(&ts, uts, sizeof(ts)))
2285                         return -EFAULT;
2286                 if (ts.tv_nsec >= 1000000000L || ts.tv_nsec < 0
2287                     || ts.tv_sec < 0)
2288                         return -EINVAL;
2289         }
2290
2291         spin_lock_irq(&current->sighand->siglock);
2292         sig = dequeue_signal(current, &these, &info);
2293         if (!sig) {
2294                 timeout = MAX_SCHEDULE_TIMEOUT;
2295                 if (uts)
2296                         timeout = (timespec_to_jiffies(&ts)
2297                                    + (ts.tv_sec || ts.tv_nsec));
2298
2299                 if (timeout) {
2300                         /* None ready -- temporarily unblock those we're
2301                          * interested while we are sleeping in so that we'll
2302                          * be awakened when they arrive.  */
2303                         current->real_blocked = current->blocked;
2304                         sigandsets(&current->blocked, &current->blocked, &these);
2305                         recalc_sigpending();
2306                         spin_unlock_irq(&current->sighand->siglock);
2307
2308                         timeout = schedule_timeout_interruptible(timeout);
2309
2310                         try_to_freeze();
2311                         spin_lock_irq(&current->sighand->siglock);
2312                         sig = dequeue_signal(current, &these, &info);
2313                         current->blocked = current->real_blocked;
2314                         siginitset(&current->real_blocked, 0);
2315                         recalc_sigpending();
2316                 }
2317         }
2318         spin_unlock_irq(&current->sighand->siglock);
2319
2320         if (sig) {
2321                 ret = sig;
2322                 if (uinfo) {
2323                         if (copy_siginfo_to_user(uinfo, &info))
2324                                 ret = -EFAULT;
2325                 }
2326         } else {
2327                 ret = -EAGAIN;
2328                 if (timeout)
2329                         ret = -EINTR;
2330         }
2331
2332         return ret;
2333 }
2334
2335 asmlinkage long
2336 sys_kill(int pid, int sig)
2337 {
2338         struct siginfo info;
2339
2340         info.si_signo = sig;
2341         info.si_errno = 0;
2342         info.si_code = SI_USER;
2343         info.si_pid = current->tgid;
2344         info.si_uid = current->uid;
2345
2346         return kill_something_info(sig, &info, pid);
2347 }
2348
2349 static int do_tkill(int tgid, int pid, int sig)
2350 {
2351         int error;
2352         struct siginfo info;
2353         struct task_struct *p;
2354
2355         error = -ESRCH;
2356         info.si_signo = sig;
2357         info.si_errno = 0;
2358         info.si_code = SI_TKILL;
2359         info.si_pid = current->tgid;
2360         info.si_uid = current->uid;
2361
2362         read_lock(&tasklist_lock);
2363         p = find_task_by_pid(pid);
2364         if (p && (tgid <= 0 || p->tgid == tgid)) {
2365                 error = check_kill_permission(sig, &info, p);
2366                 /*
2367                  * The null signal is a permissions and process existence
2368                  * probe.  No signal is actually delivered.
2369                  */
2370                 if (!error && sig && p->sighand) {
2371                         spin_lock_irq(&p->sighand->siglock);
2372                         handle_stop_signal(sig, p);
2373                         error = specific_send_sig_info(sig, &info, p);
2374                         spin_unlock_irq(&p->sighand->siglock);
2375                 }
2376         }
2377         read_unlock(&tasklist_lock);
2378
2379         return error;
2380 }
2381
2382 /**
2383  *  sys_tgkill - send signal to one specific thread
2384  *  @tgid: the thread group ID of the thread
2385  *  @pid: the PID of the thread
2386  *  @sig: signal to be sent
2387  *
2388  *  This syscall also checks the tgid and returns -ESRCH even if the PID
2389  *  exists but it's not belonging to the target process anymore. This
2390  *  method solves the problem of threads exiting and PIDs getting reused.
2391  */
2392 asmlinkage long sys_tgkill(int tgid, int pid, int sig)
2393 {
2394         /* This is only valid for single tasks */
2395         if (pid <= 0 || tgid <= 0)
2396                 return -EINVAL;
2397
2398         return do_tkill(tgid, pid, sig);
2399 }
2400
2401 /*
2402  *  Send a signal to only one task, even if it's a CLONE_THREAD task.
2403  */
2404 asmlinkage long
2405 sys_tkill(int pid, int sig)
2406 {
2407         /* This is only valid for single tasks */
2408         if (pid <= 0)
2409                 return -EINVAL;
2410
2411         return do_tkill(0, pid, sig);
2412 }
2413
2414 asmlinkage long
2415 sys_rt_sigqueueinfo(int pid, int sig, siginfo_t __user *uinfo)
2416 {
2417         siginfo_t info;
2418
2419         if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2420                 return -EFAULT;
2421
2422         /* Not even root can pretend to send signals from the kernel.
2423            Nor can they impersonate a kill(), which adds source info.  */
2424         if (info.si_code >= 0)
2425                 return -EPERM;
2426         info.si_signo = sig;
2427
2428         /* POSIX.1b doesn't mention process groups.  */
2429         return kill_proc_info(sig, &info, pid);
2430 }
2431
2432 int
2433 do_sigaction(int sig, const struct k_sigaction *act, struct k_sigaction *oact)
2434 {
2435         struct k_sigaction *k;
2436         sigset_t mask;
2437
2438         if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
2439                 return -EINVAL;
2440
2441         k = &current->sighand->action[sig-1];
2442
2443         spin_lock_irq(&current->sighand->siglock);
2444         if (signal_pending(current)) {
2445                 /*
2446                  * If there might be a fatal signal pending on multiple
2447                  * threads, make sure we take it before changing the action.
2448                  */
2449                 spin_unlock_irq(&current->sighand->siglock);
2450                 return -ERESTARTNOINTR;
2451         }
2452
2453         if (oact)
2454                 *oact = *k;
2455
2456         if (act) {
2457                 /*
2458                  * POSIX 3.3.1.3:
2459                  *  "Setting a signal action to SIG_IGN for a signal that is
2460                  *   pending shall cause the pending signal to be discarded,
2461                  *   whether or not it is blocked."
2462                  *
2463                  *  "Setting a signal action to SIG_DFL for a signal that is
2464                  *   pending and whose default action is to ignore the signal
2465                  *   (for example, SIGCHLD), shall cause the pending signal to
2466                  *   be discarded, whether or not it is blocked"
2467                  */
2468                 if (act->sa.sa_handler == SIG_IGN ||
2469                     (act->sa.sa_handler == SIG_DFL &&
2470                      sig_kernel_ignore(sig))) {
2471                         /*
2472                          * This is a fairly rare case, so we only take the
2473                          * tasklist_lock once we're sure we'll need it.
2474                          * Now we must do this little unlock and relock
2475                          * dance to maintain the lock hierarchy.
2476                          */
2477                         struct task_struct *t = current;
2478                         spin_unlock_irq(&t->sighand->siglock);
2479                         read_lock(&tasklist_lock);
2480                         spin_lock_irq(&t->sighand->siglock);
2481                         *k = *act;
2482                         sigdelsetmask(&k->sa.sa_mask,
2483                                       sigmask(SIGKILL) | sigmask(SIGSTOP));
2484                         sigemptyset(&mask);
2485                         sigaddset(&mask, sig);
2486                         rm_from_queue_full(&mask, &t->signal->shared_pending);
2487                         do {
2488                                 rm_from_queue_full(&mask, &t->pending);
2489                                 recalc_sigpending_tsk(t);
2490                                 t = next_thread(t);
2491                         } while (t != current);
2492                         spin_unlock_irq(&current->sighand->siglock);
2493                         read_unlock(&tasklist_lock);
2494                         return 0;
2495                 }
2496
2497                 *k = *act;
2498                 sigdelsetmask(&k->sa.sa_mask,
2499                               sigmask(SIGKILL) | sigmask(SIGSTOP));
2500         }
2501
2502         spin_unlock_irq(&current->sighand->siglock);
2503         return 0;
2504 }
2505
2506 int 
2507 do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
2508 {
2509         stack_t oss;
2510         int error;
2511
2512         if (uoss) {
2513                 oss.ss_sp = (void __user *) current->sas_ss_sp;
2514                 oss.ss_size = current->sas_ss_size;
2515                 oss.ss_flags = sas_ss_flags(sp);
2516         }
2517
2518         if (uss) {
2519                 void __user *ss_sp;
2520                 size_t ss_size;
2521                 int ss_flags;
2522
2523                 error = -EFAULT;
2524                 if (!access_ok(VERIFY_READ, uss, sizeof(*uss))
2525                     || __get_user(ss_sp, &uss->ss_sp)
2526                     || __get_user(ss_flags, &uss->ss_flags)
2527                     || __get_user(ss_size, &uss->ss_size))
2528                         goto out;
2529
2530                 error = -EPERM;
2531                 if (on_sig_stack(sp))
2532                         goto out;
2533
2534                 error = -EINVAL;
2535                 /*
2536                  *
2537                  * Note - this code used to test ss_flags incorrectly
2538                  *        old code may have been written using ss_flags==0
2539                  *        to mean ss_flags==SS_ONSTACK (as this was the only
2540                  *        way that worked) - this fix preserves that older
2541                  *        mechanism
2542                  */
2543                 if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
2544                         goto out;
2545
2546                 if (ss_flags == SS_DISABLE) {
2547                         ss_size = 0;
2548                         ss_sp = NULL;
2549                 } else {
2550                         error = -ENOMEM;
2551                         if (ss_size < MINSIGSTKSZ)
2552                                 goto out;
2553                 }
2554
2555                 current->sas_ss_sp = (unsigned long) ss_sp;
2556                 current->sas_ss_size = ss_size;
2557         }
2558
2559         if (uoss) {
2560                 error = -EFAULT;
2561                 if (copy_to_user(uoss, &oss, sizeof(oss)))
2562                         goto out;
2563         }
2564
2565         error = 0;
2566 out:
2567         return error;
2568 }
2569
2570 #ifdef __ARCH_WANT_SYS_SIGPENDING
2571
2572 asmlinkage long
2573 sys_sigpending(old_sigset_t __user *set)
2574 {
2575         return do_sigpending(set, sizeof(*set));
2576 }
2577
2578 #endif
2579
2580 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
2581 /* Some platforms have their own version with special arguments others
2582    support only sys_rt_sigprocmask.  */
2583
2584 asmlinkage long
2585 sys_sigprocmask(int how, old_sigset_t __user *set, old_sigset_t __user *oset)
2586 {
2587         int error;
2588         old_sigset_t old_set, new_set;
2589
2590         if (set) {
2591                 error = -EFAULT;
2592                 if (copy_from_user(&new_set, set, sizeof(*set)))
2593                         goto out;
2594                 new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
2595
2596                 spin_lock_irq(&current->sighand->siglock);
2597                 old_set = current->blocked.sig[0];
2598
2599                 error = 0;
2600                 switch (how) {
2601                 default:
2602                         error = -EINVAL;
2603                         break;
2604                 case SIG_BLOCK:
2605                         sigaddsetmask(&current->blocked, new_set);
2606                         break;
2607                 case SIG_UNBLOCK:
2608                         sigdelsetmask(&current->blocked, new_set);
2609                         break;
2610                 case SIG_SETMASK:
2611                         current->blocked.sig[0] = new_set;
2612                         break;
2613                 }
2614
2615                 recalc_sigpending();
2616                 spin_unlock_irq(&current->sighand->siglock);
2617                 if (error)
2618                         goto out;
2619                 if (oset)
2620                         goto set_old;
2621         } else if (oset) {
2622                 old_set = current->blocked.sig[0];
2623         set_old:
2624                 error = -EFAULT;
2625                 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2626                         goto out;
2627         }
2628         error = 0;
2629 out:
2630         return error;
2631 }
2632 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
2633
2634 #ifdef __ARCH_WANT_SYS_RT_SIGACTION
2635 asmlinkage long
2636 sys_rt_sigaction(int sig,
2637                  const struct sigaction __user *act,
2638                  struct sigaction __user *oact,
2639                  size_t sigsetsize)
2640 {
2641         struct k_sigaction new_sa, old_sa;
2642         int ret = -EINVAL;
2643
2644         /* XXX: Don't preclude handling different sized sigset_t's.  */
2645         if (sigsetsize != sizeof(sigset_t))
2646                 goto out;
2647
2648         if (act) {
2649                 if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
2650                         return -EFAULT;
2651         }
2652
2653         ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
2654
2655         if (!ret && oact) {
2656                 if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
2657                         return -EFAULT;
2658         }
2659 out:
2660         return ret;
2661 }
2662 #endif /* __ARCH_WANT_SYS_RT_SIGACTION */
2663
2664 #ifdef __ARCH_WANT_SYS_SGETMASK
2665
2666 /*
2667  * For backwards compatibility.  Functionality superseded by sigprocmask.
2668  */
2669 asmlinkage long
2670 sys_sgetmask(void)
2671 {
2672         /* SMP safe */
2673         return current->blocked.sig[0];
2674 }
2675
2676 asmlinkage long
2677 sys_ssetmask(int newmask)
2678 {
2679         int old;
2680
2681         spin_lock_irq(&current->sighand->siglock);
2682         old = current->blocked.sig[0];
2683
2684         siginitset(&current->blocked, newmask & ~(sigmask(SIGKILL)|
2685                                                   sigmask(SIGSTOP)));
2686         recalc_sigpending();
2687         spin_unlock_irq(&current->sighand->siglock);
2688
2689         return old;
2690 }
2691 #endif /* __ARCH_WANT_SGETMASK */
2692
2693 #ifdef __ARCH_WANT_SYS_SIGNAL
2694 /*
2695  * For backwards compatibility.  Functionality superseded by sigaction.
2696  */
2697 asmlinkage unsigned long
2698 sys_signal(int sig, __sighandler_t handler)
2699 {
2700         struct k_sigaction new_sa, old_sa;
2701         int ret;
2702
2703         new_sa.sa.sa_handler = handler;
2704         new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
2705
2706         ret = do_sigaction(sig, &new_sa, &old_sa);
2707
2708         return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
2709 }
2710 #endif /* __ARCH_WANT_SYS_SIGNAL */
2711
2712 #ifdef __ARCH_WANT_SYS_PAUSE
2713
2714 asmlinkage long
2715 sys_pause(void)
2716 {
2717         current->state = TASK_INTERRUPTIBLE;
2718         schedule();
2719         return -ERESTARTNOHAND;
2720 }
2721
2722 #endif
2723
2724 void __init signals_init(void)
2725 {
2726         sigqueue_cachep =
2727                 kmem_cache_create("sigqueue",
2728                                   sizeof(struct sigqueue),
2729                                   __alignof__(struct sigqueue),
2730                                   SLAB_PANIC, NULL, NULL);
2731 }