This commit was manufactured by cvs2svn to create tag
[linux-2.6.git] / kernel / signal.c
1 /*
2  *  linux/kernel/signal.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  *
6  *  1997-11-02  Modified for POSIX.1b signals by Richard Henderson
7  *
8  *  2003-06-02  Jim Houston - Concurrent Computer Corp.
9  *              Changes to use preallocated sigqueue structures
10  *              to allow signals to be sent reliably.
11  */
12
13 #include <linux/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/ptrace.h>
24 #include <asm/param.h>
25 #include <asm/uaccess.h>
26 #include <asm/unistd.h>
27 #include <asm/siginfo.h>
28
29 /*
30  * SLAB caches for signal bits.
31  */
32
33 static kmem_cache_t *sigqueue_cachep;
34
35 /*
36  * In POSIX a signal is sent either to a specific thread (Linux task)
37  * or to the process as a whole (Linux thread group).  How the signal
38  * is sent determines whether it's to one thread or the whole group,
39  * which determines which signal mask(s) are involved in blocking it
40  * from being delivered until later.  When the signal is delivered,
41  * either it's caught or ignored by a user handler or it has a default
42  * effect that applies to the whole thread group (POSIX process).
43  *
44  * The possible effects an unblocked signal set to SIG_DFL can have are:
45  *   ignore     - Nothing Happens
46  *   terminate  - kill the process, i.e. all threads in the group,
47  *                similar to exit_group.  The group leader (only) reports
48  *                WIFSIGNALED status to its parent.
49  *   coredump   - write a core dump file describing all threads using
50  *                the same mm and then kill all those threads
51  *   stop       - stop all the threads in the group, i.e. TASK_STOPPED state
52  *
53  * SIGKILL and SIGSTOP cannot be caught, blocked, or ignored.
54  * Other signals when not blocked and set to SIG_DFL behaves as follows.
55  * The job control signals also have other special effects.
56  *
57  *      +--------------------+------------------+
58  *      |  POSIX signal      |  default action  |
59  *      +--------------------+------------------+
60  *      |  SIGHUP            |  terminate       |
61  *      |  SIGINT            |  terminate       |
62  *      |  SIGQUIT           |  coredump        |
63  *      |  SIGILL            |  coredump        |
64  *      |  SIGTRAP           |  coredump        |
65  *      |  SIGABRT/SIGIOT    |  coredump        |
66  *      |  SIGBUS            |  coredump        |
67  *      |  SIGFPE            |  coredump        |
68  *      |  SIGKILL           |  terminate(+)    |
69  *      |  SIGUSR1           |  terminate       |
70  *      |  SIGSEGV           |  coredump        |
71  *      |  SIGUSR2           |  terminate       |
72  *      |  SIGPIPE           |  terminate       |
73  *      |  SIGALRM           |  terminate       |
74  *      |  SIGTERM           |  terminate       |
75  *      |  SIGCHLD           |  ignore          |
76  *      |  SIGCONT           |  ignore(*)       |
77  *      |  SIGSTOP           |  stop(*)(+)      |
78  *      |  SIGTSTP           |  stop(*)         |
79  *      |  SIGTTIN           |  stop(*)         |
80  *      |  SIGTTOU           |  stop(*)         |
81  *      |  SIGURG            |  ignore          |
82  *      |  SIGXCPU           |  coredump        |
83  *      |  SIGXFSZ           |  coredump        |
84  *      |  SIGVTALRM         |  terminate       |
85  *      |  SIGPROF           |  terminate       |
86  *      |  SIGPOLL/SIGIO     |  terminate       |
87  *      |  SIGSYS/SIGUNUSED  |  coredump        |
88  *      |  SIGSTKFLT         |  terminate       |
89  *      |  SIGWINCH          |  ignore          |
90  *      |  SIGPWR            |  terminate       |
91  *      |  SIGRTMIN-SIGRTMAX |  terminate       |
92  *      +--------------------+------------------+
93  *      |  non-POSIX signal  |  default action  |
94  *      +--------------------+------------------+
95  *      |  SIGEMT            |  coredump        |
96  *      +--------------------+------------------+
97  *
98  * (+) For SIGKILL and SIGSTOP the action is "always", not just "default".
99  * (*) Special job control effects:
100  * When SIGCONT is sent, it resumes the process (all threads in the group)
101  * from TASK_STOPPED state and also clears any pending/queued stop signals
102  * (any of those marked with "stop(*)").  This happens regardless of blocking,
103  * catching, or ignoring SIGCONT.  When any stop signal is sent, it clears
104  * any pending/queued SIGCONT signals; this happens regardless of blocking,
105  * catching, or ignored the stop signal, though (except for SIGSTOP) the
106  * default action of stopping the process may happen later or never.
107  */
108
109 #ifdef SIGEMT
110 #define M_SIGEMT        M(SIGEMT)
111 #else
112 #define M_SIGEMT        0
113 #endif
114
115 #if SIGRTMIN > BITS_PER_LONG
116 #define M(sig) (1ULL << ((sig)-1))
117 #else
118 #define M(sig) (1UL << ((sig)-1))
119 #endif
120 #define T(sig, mask) (M(sig) & (mask))
121
122 #define SIG_KERNEL_ONLY_MASK (\
123         M(SIGKILL)   |  M(SIGSTOP)                                   )
124
125 #define SIG_KERNEL_STOP_MASK (\
126         M(SIGSTOP)   |  M(SIGTSTP)   |  M(SIGTTIN)   |  M(SIGTTOU)   )
127
128 #define SIG_KERNEL_COREDUMP_MASK (\
129         M(SIGQUIT)   |  M(SIGILL)    |  M(SIGTRAP)   |  M(SIGABRT)   | \
130         M(SIGFPE)    |  M(SIGSEGV)   |  M(SIGBUS)    |  M(SIGSYS)    | \
131         M(SIGXCPU)   |  M(SIGXFSZ)   |  M_SIGEMT                     )
132
133 #define SIG_KERNEL_IGNORE_MASK (\
134         M(SIGCONT)   |  M(SIGCHLD)   |  M(SIGWINCH)  |  M(SIGURG)    )
135
136 #define sig_kernel_only(sig) \
137                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_ONLY_MASK))
138 #define sig_kernel_coredump(sig) \
139                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_COREDUMP_MASK))
140 #define sig_kernel_ignore(sig) \
141                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_IGNORE_MASK))
142 #define sig_kernel_stop(sig) \
143                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_STOP_MASK))
144
145 #define sig_user_defined(t, signr) \
146         (((t)->sighand->action[(signr)-1].sa.sa_handler != SIG_DFL) &&  \
147          ((t)->sighand->action[(signr)-1].sa.sa_handler != SIG_IGN))
148
149 #define sig_fatal(t, signr) \
150         (!T(signr, SIG_KERNEL_IGNORE_MASK|SIG_KERNEL_STOP_MASK) && \
151          (t)->sighand->action[(signr)-1].sa.sa_handler == SIG_DFL)
152
153 #define sig_avoid_stop_race() \
154         (sigtestsetmask(&current->pending.signal, M(SIGCONT) | M(SIGKILL)) || \
155          sigtestsetmask(&current->signal->shared_pending.signal, \
156                                                   M(SIGCONT) | M(SIGKILL)))
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             PENDING(&t->pending, &t->blocked) ||
218             PENDING(&t->signal->shared_pending, &t->blocked))
219                 set_tsk_thread_flag(t, TIF_SIGPENDING);
220         else
221                 clear_tsk_thread_flag(t, TIF_SIGPENDING);
222 }
223
224 void recalc_sigpending(void)
225 {
226         recalc_sigpending_tsk(current);
227 }
228
229 /* Given the mask, find the first available signal that should be serviced. */
230
231 static int
232 next_signal(struct sigpending *pending, sigset_t *mask)
233 {
234         unsigned long i, *s, *m, x;
235         int sig = 0;
236         
237         s = pending->signal.sig;
238         m = mask->sig;
239         switch (_NSIG_WORDS) {
240         default:
241                 for (i = 0; i < _NSIG_WORDS; ++i, ++s, ++m)
242                         if ((x = *s &~ *m) != 0) {
243                                 sig = ffz(~x) + i*_NSIG_BPW + 1;
244                                 break;
245                         }
246                 break;
247
248         case 2: if ((x = s[0] &~ m[0]) != 0)
249                         sig = 1;
250                 else if ((x = s[1] &~ m[1]) != 0)
251                         sig = _NSIG_BPW + 1;
252                 else
253                         break;
254                 sig += ffz(~x);
255                 break;
256
257         case 1: if ((x = *s &~ *m) != 0)
258                         sig = ffz(~x) + 1;
259                 break;
260         }
261         
262         return sig;
263 }
264
265 static struct sigqueue *__sigqueue_alloc(void)
266 {
267         struct sigqueue *q = NULL;
268
269         if (atomic_read(&current->user->sigpending) <
270                         current->rlim[RLIMIT_SIGPENDING].rlim_cur)
271                 q = kmem_cache_alloc(sigqueue_cachep, GFP_ATOMIC);
272         if (q) {
273                 INIT_LIST_HEAD(&q->list);
274                 q->flags = 0;
275                 q->lock = NULL;
276                 q->user = get_uid(current->user);
277                 atomic_inc(&q->user->sigpending);
278         }
279         return(q);
280 }
281
282 static inline void __sigqueue_free(struct sigqueue *q)
283 {
284         if (q->flags & SIGQUEUE_PREALLOC)
285                 return;
286         atomic_dec(&q->user->sigpending);
287         free_uid(q->user);
288         kmem_cache_free(sigqueue_cachep, q);
289 }
290
291 static void flush_sigqueue(struct sigpending *queue)
292 {
293         struct sigqueue *q;
294
295         sigemptyset(&queue->signal);
296         while (!list_empty(&queue->list)) {
297                 q = list_entry(queue->list.next, struct sigqueue , list);
298                 list_del_init(&q->list);
299                 __sigqueue_free(q);
300         }
301 }
302
303 /*
304  * Flush all pending signals for a task.
305  */
306
307 void
308 flush_signals(struct task_struct *t)
309 {
310         unsigned long flags;
311
312         spin_lock_irqsave(&t->sighand->siglock, flags);
313         clear_tsk_thread_flag(t,TIF_SIGPENDING);
314         flush_sigqueue(&t->pending);
315         flush_sigqueue(&t->signal->shared_pending);
316         spin_unlock_irqrestore(&t->sighand->siglock, flags);
317 }
318
319 /*
320  * This function expects the tasklist_lock write-locked.
321  */
322 void __exit_sighand(struct task_struct *tsk)
323 {
324         struct sighand_struct * sighand = tsk->sighand;
325
326         /* Ok, we're done with the signal handlers */
327         tsk->sighand = NULL;
328         if (atomic_dec_and_test(&sighand->count))
329                 kmem_cache_free(sighand_cachep, sighand);
330 }
331
332 void exit_sighand(struct task_struct *tsk)
333 {
334         write_lock_irq(&tasklist_lock);
335         __exit_sighand(tsk);
336         write_unlock_irq(&tasklist_lock);
337 }
338
339 /*
340  * This function expects the tasklist_lock write-locked.
341  */
342 void __exit_signal(struct task_struct *tsk)
343 {
344         struct signal_struct * sig = tsk->signal;
345         struct sighand_struct * sighand = tsk->sighand;
346
347         if (!sig)
348                 BUG();
349         if (!atomic_read(&sig->count))
350                 BUG();
351         spin_lock(&sighand->siglock);
352         if (atomic_dec_and_test(&sig->count)) {
353                 if (tsk == sig->curr_target)
354                         sig->curr_target = next_thread(tsk);
355                 tsk->signal = NULL;
356                 spin_unlock(&sighand->siglock);
357                 flush_sigqueue(&sig->shared_pending);
358         } else {
359                 /*
360                  * If there is any task waiting for the group exit
361                  * then notify it:
362                  */
363                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
364                         wake_up_process(sig->group_exit_task);
365                         sig->group_exit_task = NULL;
366                 }
367                 if (tsk == sig->curr_target)
368                         sig->curr_target = next_thread(tsk);
369                 tsk->signal = NULL;
370                 spin_unlock(&sighand->siglock);
371                 sig = NULL;     /* Marker for below.  */
372         }
373         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
374         flush_sigqueue(&tsk->pending);
375         if (sig) {
376                 /*
377                  * We are cleaning up the signal_struct here.  We delayed
378                  * calling exit_itimers until after flush_sigqueue, just in
379                  * case our thread-local pending queue contained a queued
380                  * timer signal that would have been cleared in
381                  * exit_itimers.  When that called sigqueue_free, it would
382                  * attempt to re-take the tasklist_lock and deadlock.  This
383                  * can never happen if we ensure that all queues the
384                  * timer's signal might be queued on have been flushed
385                  * first.  The shared_pending queue, and our own pending
386                  * queue are the only queues the timer could be on, since
387                  * there are no other threads left in the group and timer
388                  * signals are constrained to threads inside the group.
389                  */
390                 exit_itimers(sig);
391                 kmem_cache_free(signal_cachep, sig);
392         }
393 }
394
395 void exit_signal(struct task_struct *tsk)
396 {
397         write_lock_irq(&tasklist_lock);
398         __exit_signal(tsk);
399         write_unlock_irq(&tasklist_lock);
400 }
401
402 /*
403  * Flush all handlers for a task.
404  */
405
406 void
407 flush_signal_handlers(struct task_struct *t, int force_default)
408 {
409         int i;
410         struct k_sigaction *ka = &t->sighand->action[0];
411         for (i = _NSIG ; i != 0 ; i--) {
412                 if (force_default || ka->sa.sa_handler != SIG_IGN)
413                         ka->sa.sa_handler = SIG_DFL;
414                 ka->sa.sa_flags = 0;
415                 sigemptyset(&ka->sa.sa_mask);
416                 ka++;
417         }
418 }
419
420 EXPORT_SYMBOL_GPL(flush_signal_handlers);
421
422 /* Notify the system that a driver wants to block all signals for this
423  * process, and wants to be notified if any signals at all were to be
424  * sent/acted upon.  If the notifier routine returns non-zero, then the
425  * signal will be acted upon after all.  If the notifier routine returns 0,
426  * then then signal will be blocked.  Only one block per process is
427  * allowed.  priv is a pointer to private data that the notifier routine
428  * can use to determine if the signal should be blocked or not.  */
429
430 void
431 block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
432 {
433         unsigned long flags;
434
435         spin_lock_irqsave(&current->sighand->siglock, flags);
436         current->notifier_mask = mask;
437         current->notifier_data = priv;
438         current->notifier = notifier;
439         spin_unlock_irqrestore(&current->sighand->siglock, flags);
440 }
441
442 /* Notify the system that blocking has ended. */
443
444 void
445 unblock_all_signals(void)
446 {
447         unsigned long flags;
448
449         spin_lock_irqsave(&current->sighand->siglock, flags);
450         current->notifier = NULL;
451         current->notifier_data = NULL;
452         recalc_sigpending();
453         spin_unlock_irqrestore(&current->sighand->siglock, flags);
454 }
455
456 static inline int collect_signal(int sig, struct sigpending *list, siginfo_t *info)
457 {
458         struct sigqueue *q, *first = NULL;
459         int still_pending = 0;
460
461         if (unlikely(!sigismember(&list->signal, sig)))
462                 return 0;
463
464         /*
465          * Collect the siginfo appropriate to this signal.  Check if
466          * there is another siginfo for the same signal.
467         */
468         list_for_each_entry(q, &list->list, list) {
469                 if (q->info.si_signo == sig) {
470                         if (first) {
471                                 still_pending = 1;
472                                 break;
473                         }
474                         first = q;
475                 }
476         }
477         if (first) {
478                 list_del_init(&first->list);
479                 copy_siginfo(info, &first->info);
480                 __sigqueue_free(first);
481                 if (!still_pending)
482                         sigdelset(&list->signal, sig);
483         } else {
484
485                 /* Ok, it wasn't in the queue.  This must be
486                    a fast-pathed signal or we must have been
487                    out of queue space.  So zero out the info.
488                  */
489                 sigdelset(&list->signal, sig);
490                 info->si_signo = sig;
491                 info->si_errno = 0;
492                 info->si_code = 0;
493                 info->si_pid = 0;
494                 info->si_uid = 0;
495         }
496         return 1;
497 }
498
499 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
500                         siginfo_t *info)
501 {
502         int sig = 0;
503
504         sig = next_signal(pending, mask);
505         if (sig) {
506                 if (current->notifier) {
507                         if (sigismember(current->notifier_mask, sig)) {
508                                 if (!(current->notifier)(current->notifier_data)) {
509                                         clear_thread_flag(TIF_SIGPENDING);
510                                         return 0;
511                                 }
512                         }
513                 }
514
515                 if (!collect_signal(sig, pending, info))
516                         sig = 0;
517                                 
518         }
519         recalc_sigpending();
520
521         return sig;
522 }
523
524 /*
525  * Dequeue a signal and return the element to the caller, which is 
526  * expected to free it.
527  *
528  * All callers have to hold the siglock.
529  */
530 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
531 {
532         int signr = __dequeue_signal(&tsk->pending, mask, info);
533         if (!signr)
534                 signr = __dequeue_signal(&tsk->signal->shared_pending,
535                                          mask, info);
536         if ( signr &&
537              ((info->si_code & __SI_MASK) == __SI_TIMER) &&
538              info->si_sys_private){
539                 do_schedule_next_timer(info);
540         }
541         return signr;
542 }
543
544 /*
545  * Tell a process that it has a new active signal..
546  *
547  * NOTE! we rely on the previous spin_lock to
548  * lock interrupts for us! We can only be called with
549  * "siglock" held, and the local interrupt must
550  * have been disabled when that got acquired!
551  *
552  * No need to set need_resched since signal event passing
553  * goes through ->blocked
554  */
555 void signal_wake_up(struct task_struct *t, int resume)
556 {
557         unsigned int mask;
558
559         set_tsk_thread_flag(t, TIF_SIGPENDING);
560
561         /*
562          * If resume is set, we want to wake it up in the TASK_STOPPED case.
563          * We don't check for TASK_STOPPED because there is a race with it
564          * executing another processor and just now entering stopped state.
565          * By calling wake_up_process any time resume is set, we ensure
566          * the process will wake up and handle its stop or death signal.
567          */
568         mask = TASK_INTERRUPTIBLE;
569         if (resume)
570                 mask |= TASK_STOPPED;
571         if (!wake_up_state(t, mask))
572                 kick_process(t);
573 }
574
575 /*
576  * Remove signals in mask from the pending set and queue.
577  * Returns 1 if any signals were found.
578  *
579  * All callers must be holding the siglock.
580  */
581 static int rm_from_queue(unsigned long mask, struct sigpending *s)
582 {
583         struct sigqueue *q, *n;
584
585         if (!sigtestsetmask(&s->signal, mask))
586                 return 0;
587
588         sigdelsetmask(&s->signal, mask);
589         list_for_each_entry_safe(q, n, &s->list, list) {
590                 if (q->info.si_signo < SIGRTMIN &&
591                     (mask & sigmask(q->info.si_signo))) {
592                         list_del_init(&q->list);
593                         __sigqueue_free(q);
594                 }
595         }
596         return 1;
597 }
598
599 /*
600  * Bad permissions for sending the signal
601  */
602 static int check_kill_permission(int sig, struct siginfo *info,
603                                  struct task_struct *t)
604 {
605         int error = -EINVAL;
606         if (sig < 0 || sig > _NSIG)
607                 return error;
608         error = -EPERM;
609         if ((!info || ((unsigned long)info != 1 &&
610                         (unsigned long)info != 2 && SI_FROMUSER(info)))
611             && ((sig != SIGCONT) ||
612                 (current->signal->session != t->signal->session))
613             && (current->euid ^ t->suid) && (current->euid ^ t->uid)
614             && (current->uid ^ t->suid) && (current->uid ^ t->uid)
615             && !capable(CAP_KILL))
616                 return error;
617         return security_task_kill(t, info, sig);
618 }
619
620 /* forward decl */
621 static void do_notify_parent_cldstop(struct task_struct *tsk,
622                                      struct task_struct *parent);
623
624 /*
625  * Handle magic process-wide effects of stop/continue signals.
626  * Unlike the signal actions, these happen immediately at signal-generation
627  * time regardless of blocking, ignoring, or handling.  This does the
628  * actual continuing for SIGCONT, but not the actual stopping for stop
629  * signals.  The process stop is done as a signal action for SIG_DFL.
630  */
631 static void handle_stop_signal(int sig, struct task_struct *p)
632 {
633         struct task_struct *t;
634
635         if (sig_kernel_stop(sig)) {
636                 /*
637                  * This is a stop signal.  Remove SIGCONT from all queues.
638                  */
639                 rm_from_queue(sigmask(SIGCONT), &p->signal->shared_pending);
640                 t = p;
641                 do {
642                         rm_from_queue(sigmask(SIGCONT), &t->pending);
643                         t = next_thread(t);
644                 } while (t != p);
645         } else if (sig == SIGCONT) {
646                 /*
647                  * Remove all stop signals from all queues,
648                  * and wake all threads.
649                  */
650                 if (unlikely(p->signal->group_stop_count > 0)) {
651                         /*
652                          * There was a group stop in progress.  We'll
653                          * pretend it finished before we got here.  We are
654                          * obliged to report it to the parent: if the
655                          * SIGSTOP happened "after" this SIGCONT, then it
656                          * would have cleared this pending SIGCONT.  If it
657                          * happened "before" this SIGCONT, then the parent
658                          * got the SIGCHLD about the stop finishing before
659                          * the continue happened.  We do the notification
660                          * now, and it's as if the stop had finished and
661                          * the SIGCHLD was pending on entry to this kill.
662                          */
663                         p->signal->group_stop_count = 0;
664                         if (p->ptrace & PT_PTRACED)
665                                 do_notify_parent_cldstop(p, p->parent);
666                         else
667                                 do_notify_parent_cldstop(
668                                         p->group_leader,
669                                         p->group_leader->real_parent);
670                 }
671                 rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
672                 t = p;
673                 do {
674                         unsigned int state;
675                         rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
676                         
677                         /*
678                          * If there is a handler for SIGCONT, we must make
679                          * sure that no thread returns to user mode before
680                          * we post the signal, in case it was the only
681                          * thread eligible to run the signal handler--then
682                          * it must not do anything between resuming and
683                          * running the handler.  With the TIF_SIGPENDING
684                          * flag set, the thread will pause and acquire the
685                          * siglock that we hold now and until we've queued
686                          * the pending signal. 
687                          *
688                          * Wake up the stopped thread _after_ setting
689                          * TIF_SIGPENDING
690                          */
691                         state = TASK_STOPPED;
692                         if (sig_user_defined(t, SIGCONT) && !sigismember(&t->blocked, SIGCONT)) {
693                                 set_tsk_thread_flag(t, TIF_SIGPENDING);
694                                 state |= TASK_INTERRUPTIBLE;
695                         }
696                         wake_up_state(t, state);
697
698                         t = next_thread(t);
699                 } while (t != p);
700         }
701 }
702
703 static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
704                         struct sigpending *signals)
705 {
706         struct sigqueue * q = NULL;
707         int ret = 0;
708
709         /*
710          * fast-pathed signals for kernel-internal things like SIGSTOP
711          * or SIGKILL.
712          */
713         if ((unsigned long)info == 2)
714                 goto out_set;
715
716         /* Real-time signals must be queued if sent by sigqueue, or
717            some other real-time mechanism.  It is implementation
718            defined whether kill() does so.  We attempt to do so, on
719            the principle of least surprise, but since kill is not
720            allowed to fail with EAGAIN when low on memory we just
721            make sure at least one signal gets delivered and don't
722            pass on the info struct.  */
723
724         if (atomic_read(&t->user->sigpending) <
725                         t->rlim[RLIMIT_SIGPENDING].rlim_cur)
726                 q = kmem_cache_alloc(sigqueue_cachep, GFP_ATOMIC);
727
728         if (q) {
729                 q->flags = 0;
730                 q->user = get_uid(t->user);
731                 atomic_inc(&q->user->sigpending);
732                 list_add_tail(&q->list, &signals->list);
733                 switch ((unsigned long) info) {
734                 case 0:
735                         q->info.si_signo = sig;
736                         q->info.si_errno = 0;
737                         q->info.si_code = SI_USER;
738                         q->info.si_pid = current->pid;
739                         q->info.si_uid = current->uid;
740                         break;
741                 case 1:
742                         q->info.si_signo = sig;
743                         q->info.si_errno = 0;
744                         q->info.si_code = SI_KERNEL;
745                         q->info.si_pid = 0;
746                         q->info.si_uid = 0;
747                         break;
748                 default:
749                         copy_siginfo(&q->info, info);
750                         break;
751                 }
752         } else {
753                 if (sig >= SIGRTMIN && info && (unsigned long)info != 1
754                    && info->si_code != SI_USER)
755                 /*
756                  * Queue overflow, abort.  We may abort if the signal was rt
757                  * and sent by user using something other than kill().
758                  */
759                         return -EAGAIN;
760                 if (((unsigned long)info > 1) && (info->si_code == SI_TIMER))
761                         /*
762                          * Set up a return to indicate that we dropped 
763                          * the signal.
764                          */
765                         ret = info->si_sys_private;
766         }
767
768 out_set:
769         sigaddset(&signals->signal, sig);
770         return ret;
771 }
772
773 #define LEGACY_QUEUE(sigptr, sig) \
774         (((sig) < SIGRTMIN) && sigismember(&(sigptr)->signal, (sig)))
775
776
777 static int
778 specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
779 {
780         int ret = 0;
781
782         if (!irqs_disabled())
783                 BUG();
784 #ifdef CONFIG_SMP
785         if (!spin_is_locked(&t->sighand->siglock))
786                 BUG();
787 #endif
788
789         if (((unsigned long)info > 2) && (info->si_code == SI_TIMER))
790                 /*
791                  * Set up a return to indicate that we dropped the signal.
792                  */
793                 ret = info->si_sys_private;
794
795         /* Short-circuit ignored signals.  */
796         if (sig_ignored(t, sig))
797                 goto out;
798
799         /* Support queueing exactly one non-rt signal, so that we
800            can get more detailed information about the cause of
801            the signal. */
802         if (LEGACY_QUEUE(&t->pending, sig))
803                 goto out;
804
805         ret = send_signal(sig, info, t, &t->pending);
806         if (!ret && !sigismember(&t->blocked, sig))
807                 signal_wake_up(t, sig == SIGKILL);
808 out:
809         return ret;
810 }
811
812 /*
813  * Force a signal that the process can't ignore: if necessary
814  * we unblock the signal and change any SIG_IGN to SIG_DFL.
815  */
816
817 int
818 force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
819 {
820         unsigned long int flags;
821         int ret;
822
823         spin_lock_irqsave(&t->sighand->siglock, flags);
824         if (sigismember(&t->blocked, sig) || t->sighand->action[sig-1].sa.sa_handler == SIG_IGN) {
825                 t->sighand->action[sig-1].sa.sa_handler = SIG_DFL;
826                 sigdelset(&t->blocked, sig);
827                 recalc_sigpending_tsk(t);
828         }
829         ret = specific_send_sig_info(sig, info, t);
830         spin_unlock_irqrestore(&t->sighand->siglock, flags);
831
832         return ret;
833 }
834
835 void
836 force_sig_specific(int sig, struct task_struct *t)
837 {
838         unsigned long int flags;
839
840         spin_lock_irqsave(&t->sighand->siglock, flags);
841         if (t->sighand->action[sig-1].sa.sa_handler == SIG_IGN)
842                 t->sighand->action[sig-1].sa.sa_handler = SIG_DFL;
843         sigdelset(&t->blocked, sig);
844         recalc_sigpending_tsk(t);
845         specific_send_sig_info(sig, (void *)2, t);
846         spin_unlock_irqrestore(&t->sighand->siglock, flags);
847 }
848
849 /*
850  * Test if P wants to take SIG.  After we've checked all threads with this,
851  * it's equivalent to finding no threads not blocking SIG.  Any threads not
852  * blocking SIG were ruled out because they are not running and already
853  * have pending signals.  Such threads will dequeue from the shared queue
854  * as soon as they're available, so putting the signal on the shared queue
855  * will be equivalent to sending it to one such thread.
856  */
857 #define wants_signal(sig, p, mask)                      \
858         (!sigismember(&(p)->blocked, sig)               \
859          && !((p)->state & mask)                        \
860          && !((p)->flags & PF_EXITING)                  \
861          && (task_curr(p) || !signal_pending(p)))
862
863
864 static void
865 __group_complete_signal(int sig, struct task_struct *p, unsigned int mask)
866 {
867         struct task_struct *t;
868
869         /*
870          * Now find a thread we can wake up to take the signal off the queue.
871          *
872          * If the main thread wants the signal, it gets first crack.
873          * Probably the least surprising to the average bear.
874          */
875         if (wants_signal(sig, p, mask))
876                 t = p;
877         else if (thread_group_empty(p))
878                 /*
879                  * There is just one thread and it does not need to be woken.
880                  * It will dequeue unblocked signals before it runs again.
881                  */
882                 return;
883         else {
884                 /*
885                  * Otherwise try to find a suitable thread.
886                  */
887                 t = p->signal->curr_target;
888                 if (t == NULL)
889                         /* restart balancing at this thread */
890                         t = p->signal->curr_target = p;
891                 BUG_ON(t->tgid != p->tgid);
892
893                 while (!wants_signal(sig, t, mask)) {
894                         t = next_thread(t);
895                         if (t == p->signal->curr_target)
896                                 /*
897                                  * No thread needs to be woken.
898                                  * Any eligible threads will see
899                                  * the signal in the queue soon.
900                                  */
901                                 return;
902                 }
903                 p->signal->curr_target = t;
904         }
905
906         /*
907          * Found a killable thread.  If the signal will be fatal,
908          * then start taking the whole group down immediately.
909          */
910         if (sig_fatal(p, sig) && !p->signal->group_exit &&
911             !sigismember(&t->real_blocked, sig) &&
912             (sig == SIGKILL || !(t->ptrace & PT_PTRACED))) {
913                 /*
914                  * This signal will be fatal to the whole group.
915                  */
916                 if (!sig_kernel_coredump(sig)) {
917                         /*
918                          * Start a group exit and wake everybody up.
919                          * This way we don't have other threads
920                          * running and doing things after a slower
921                          * thread has the fatal signal pending.
922                          */
923                         p->signal->group_exit = 1;
924                         p->signal->group_exit_code = sig;
925                         p->signal->group_stop_count = 0;
926                         t = p;
927                         do {
928                                 sigaddset(&t->pending.signal, SIGKILL);
929                                 signal_wake_up(t, 1);
930                                 t = next_thread(t);
931                         } while (t != p);
932                         return;
933                 }
934
935                 /*
936                  * There will be a core dump.  We make all threads other
937                  * than the chosen one go into a group stop so that nothing
938                  * happens until it gets scheduled, takes the signal off
939                  * the shared queue, and does the core dump.  This is a
940                  * little more complicated than strictly necessary, but it
941                  * keeps the signal state that winds up in the core dump
942                  * unchanged from the death state, e.g. which thread had
943                  * the core-dump signal unblocked.
944                  */
945                 rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
946                 rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
947                 p->signal->group_stop_count = 0;
948                 p->signal->group_exit_task = t;
949                 t = p;
950                 do {
951                         p->signal->group_stop_count++;
952                         signal_wake_up(t, 0);
953                         t = next_thread(t);
954                 } while (t != p);
955                 wake_up_process(p->signal->group_exit_task);
956                 return;
957         }
958
959         /*
960          * The signal is already in the shared-pending queue.
961          * Tell the chosen thread to wake up and dequeue it.
962          */
963         signal_wake_up(t, sig == SIGKILL);
964         return;
965 }
966
967 static int
968 __group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
969 {
970         unsigned int mask;
971         int ret = 0;
972
973 #ifdef CONFIG_SMP
974         if (!spin_is_locked(&p->sighand->siglock))
975                 BUG();
976 #endif
977         handle_stop_signal(sig, p);
978
979         if (((unsigned long)info > 2) && (info->si_code == SI_TIMER))
980                 /*
981                  * Set up a return to indicate that we dropped the signal.
982                  */
983                 ret = info->si_sys_private;
984
985         /* Short-circuit ignored signals.  */
986         if (sig_ignored(p, sig))
987                 return ret;
988
989         if (LEGACY_QUEUE(&p->signal->shared_pending, sig))
990                 /* This is a non-RT signal and we already have one queued.  */
991                 return ret;
992
993         /*
994          * Don't bother zombies and stopped tasks (but
995          * SIGKILL will punch through stopped state)
996          */
997         mask = TASK_DEAD | TASK_ZOMBIE;
998         if (sig != SIGKILL)
999                 mask |= TASK_STOPPED;
1000
1001         /*
1002          * Put this signal on the shared-pending queue, or fail with EAGAIN.
1003          * We always use the shared queue for process-wide signals,
1004          * to avoid several races.
1005          */
1006         ret = send_signal(sig, info, p, &p->signal->shared_pending);
1007         if (unlikely(ret))
1008                 return ret;
1009
1010         __group_complete_signal(sig, p, mask);
1011         return 0;
1012 }
1013
1014 /*
1015  * Nuke all other threads in the group.
1016  */
1017 void zap_other_threads(struct task_struct *p)
1018 {
1019         struct task_struct *t;
1020
1021         p->signal->group_stop_count = 0;
1022
1023         if (thread_group_empty(p))
1024                 return;
1025
1026         for (t = next_thread(p); t != p; t = next_thread(t)) {
1027                 /*
1028                  * Don't bother with already dead threads
1029                  */
1030                 if (t->state & (TASK_ZOMBIE|TASK_DEAD))
1031                         continue;
1032
1033                 /*
1034                  * We don't want to notify the parent, since we are
1035                  * killed as part of a thread group due to another
1036                  * thread doing an execve() or similar. So set the
1037                  * exit signal to -1 to allow immediate reaping of
1038                  * the process.  But don't detach the thread group
1039                  * leader.
1040                  */
1041                 if (t != p->group_leader)
1042                         t->exit_signal = -1;
1043
1044                 sigaddset(&t->pending.signal, SIGKILL);
1045                 rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
1046                 signal_wake_up(t, 1);
1047         }
1048 }
1049
1050 /*
1051  * Must be called with the tasklist_lock held for reading!
1052  */
1053 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1054 {
1055         unsigned long flags;
1056         int ret;
1057
1058         if (!vx_check(vx_task_xid(p), VX_ADMIN|VX_WATCH|VX_IDENT))
1059                 return -ESRCH;
1060
1061         ret = check_kill_permission(sig, info, p);
1062         if (!ret && sig && p->sighand) {
1063                 spin_lock_irqsave(&p->sighand->siglock, flags);
1064                 ret = __group_send_sig_info(sig, info, p);
1065                 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1066         }
1067
1068         return ret;
1069 }
1070
1071 /*
1072  * kill_pg_info() sends a signal to a process group: this is what the tty
1073  * control characters do (^C, ^Z etc)
1074  */
1075
1076 int __kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)
1077 {
1078         struct task_struct *p;
1079         struct list_head *l;
1080         struct pid *pid;
1081         int retval, success;
1082
1083         if (pgrp <= 0)
1084                 return -EINVAL;
1085
1086         success = 0;
1087         retval = -ESRCH;
1088         for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
1089                 int err = group_send_sig_info(sig, info, p);
1090                 success |= !err;
1091                 retval = err;
1092         }
1093         return success ? 0 : retval;
1094 }
1095
1096 int
1097 kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)
1098 {
1099         int retval;
1100
1101         read_lock(&tasklist_lock);
1102         retval = __kill_pg_info(sig, info, pgrp);
1103         read_unlock(&tasklist_lock);
1104
1105         return retval;
1106 }
1107
1108 /*
1109  * kill_sl_info() sends a signal to the session leader: this is used
1110  * to send SIGHUP to the controlling process of a terminal when
1111  * the connection is lost.
1112  */
1113
1114
1115 int
1116 kill_sl_info(int sig, struct siginfo *info, pid_t sid)
1117 {
1118         int err, retval = -EINVAL;
1119         struct pid *pid;
1120         struct list_head *l;
1121         struct task_struct *p;
1122
1123         if (sid <= 0)
1124                 goto out;
1125
1126         retval = -ESRCH;
1127         read_lock(&tasklist_lock);
1128         for_each_task_pid(sid, PIDTYPE_SID, p, l, pid) {
1129                 if (!p->signal->leader)
1130                         continue;
1131                 err = group_send_sig_info(sig, info, p);
1132                 if (retval)
1133                         retval = err;
1134         }
1135         read_unlock(&tasklist_lock);
1136 out:
1137         return retval;
1138 }
1139
1140 int
1141 kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1142 {
1143         int error;
1144         struct task_struct *p;
1145
1146         read_lock(&tasklist_lock);
1147         p = find_task_by_pid(pid);
1148         error = -ESRCH;
1149         if (p)
1150                 error = group_send_sig_info(sig, info, p);
1151         read_unlock(&tasklist_lock);
1152         return error;
1153 }
1154
1155
1156 /*
1157  * kill_something_info() interprets pid in interesting ways just like kill(2).
1158  *
1159  * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1160  * is probably wrong.  Should make it like BSD or SYSV.
1161  */
1162
1163 static int kill_something_info(int sig, struct siginfo *info, int pid)
1164 {
1165         if (!pid) {
1166                 return kill_pg_info(sig, info, process_group(current));
1167         } else if (pid == -1) {
1168                 int retval = 0, count = 0;
1169                 struct task_struct * p;
1170
1171                 read_lock(&tasklist_lock);
1172                 for_each_process(p) {
1173                         if (p->pid > 1 && p->tgid != current->tgid) {
1174                                 int err = group_send_sig_info(sig, info, p);
1175                                 ++count;
1176                                 if (err != -EPERM)
1177                                         retval = err;
1178                         }
1179                 }
1180                 read_unlock(&tasklist_lock);
1181                 return count ? retval : -ESRCH;
1182         } else if (pid < 0) {
1183                 return kill_pg_info(sig, info, -pid);
1184         } else {
1185                 return kill_proc_info(sig, info, pid);
1186         }
1187 }
1188
1189 /*
1190  * These are for backward compatibility with the rest of the kernel source.
1191  */
1192
1193 /*
1194  * These two are the most common entry points.  They send a signal
1195  * just to the specific thread.
1196  */
1197 int
1198 send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1199 {
1200         int ret;
1201         unsigned long flags;
1202
1203         /*
1204          * Make sure legacy kernel users don't send in bad values
1205          * (normal paths check this in check_kill_permission).
1206          */
1207         if (sig < 0 || sig > _NSIG)
1208                 return -EINVAL;
1209
1210         /*
1211          * We need the tasklist lock even for the specific
1212          * thread case (when we don't need to follow the group
1213          * lists) in order to avoid races with "p->sighand"
1214          * going away or changing from under us.
1215          */
1216         read_lock(&tasklist_lock);  
1217         spin_lock_irqsave(&p->sighand->siglock, flags);
1218         ret = specific_send_sig_info(sig, info, p);
1219         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1220         read_unlock(&tasklist_lock);
1221         return ret;
1222 }
1223
1224 int
1225 send_sig(int sig, struct task_struct *p, int priv)
1226 {
1227         return send_sig_info(sig, (void*)(long)(priv != 0), p);
1228 }
1229
1230 /*
1231  * This is the entry point for "process-wide" signals.
1232  * They will go to an appropriate thread in the thread group.
1233  */
1234 int
1235 send_group_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1236 {
1237         int ret;
1238         read_lock(&tasklist_lock);
1239         ret = group_send_sig_info(sig, info, p);
1240         read_unlock(&tasklist_lock);
1241         return ret;
1242 }
1243
1244 void
1245 force_sig(int sig, struct task_struct *p)
1246 {
1247         force_sig_info(sig, (void*)1L, p);
1248 }
1249
1250 int
1251 kill_pg(pid_t pgrp, int sig, int priv)
1252 {
1253         return kill_pg_info(sig, (void *)(long)(priv != 0), pgrp);
1254 }
1255
1256 int
1257 kill_sl(pid_t sess, int sig, int priv)
1258 {
1259         return kill_sl_info(sig, (void *)(long)(priv != 0), sess);
1260 }
1261
1262 int
1263 kill_proc(pid_t pid, int sig, int priv)
1264 {
1265         return kill_proc_info(sig, (void *)(long)(priv != 0), pid);
1266 }
1267
1268 /*
1269  * These functions support sending signals using preallocated sigqueue
1270  * structures.  This is needed "because realtime applications cannot
1271  * afford to lose notifications of asynchronous events, like timer
1272  * expirations or I/O completions".  In the case of Posix Timers 
1273  * we allocate the sigqueue structure from the timer_create.  If this
1274  * allocation fails we are able to report the failure to the application
1275  * with an EAGAIN error.
1276  */
1277  
1278 struct sigqueue *sigqueue_alloc(void)
1279 {
1280         struct sigqueue *q;
1281
1282         if ((q = __sigqueue_alloc()))
1283                 q->flags |= SIGQUEUE_PREALLOC;
1284         return(q);
1285 }
1286
1287 void sigqueue_free(struct sigqueue *q)
1288 {
1289         unsigned long flags;
1290         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1291         /*
1292          * If the signal is still pending remove it from the
1293          * pending queue.
1294          */
1295         if (unlikely(!list_empty(&q->list))) {
1296                 read_lock(&tasklist_lock);  
1297                 spin_lock_irqsave(q->lock, flags);
1298                 if (!list_empty(&q->list))
1299                         list_del_init(&q->list);
1300                 spin_unlock_irqrestore(q->lock, flags);
1301                 read_unlock(&tasklist_lock);
1302         }
1303         q->flags &= ~SIGQUEUE_PREALLOC;
1304         __sigqueue_free(q);
1305 }
1306
1307 int
1308 send_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
1309 {
1310         unsigned long flags;
1311         int ret = 0;
1312
1313         /*
1314          * We need the tasklist lock even for the specific
1315          * thread case (when we don't need to follow the group
1316          * lists) in order to avoid races with "p->sighand"
1317          * going away or changing from under us.
1318          */
1319         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1320         read_lock(&tasklist_lock);  
1321         spin_lock_irqsave(&p->sighand->siglock, flags);
1322         
1323         if (unlikely(!list_empty(&q->list))) {
1324                 /*
1325                  * If an SI_TIMER entry is already queue just increment
1326                  * the overrun count.
1327                  */
1328                 if (q->info.si_code != SI_TIMER)
1329                         BUG();
1330                 q->info.si_overrun++;
1331                 goto out;
1332         } 
1333         /* Short-circuit ignored signals.  */
1334         if (sig_ignored(p, sig)) {
1335                 ret = 1;
1336                 goto out;
1337         }
1338
1339         q->lock = &p->sighand->siglock;
1340         list_add_tail(&q->list, &p->pending.list);
1341         sigaddset(&p->pending.signal, sig);
1342         if (!sigismember(&p->blocked, sig))
1343                 signal_wake_up(p, sig == SIGKILL);
1344
1345 out:
1346         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1347         read_unlock(&tasklist_lock);
1348         return(ret);
1349 }
1350
1351 int
1352 send_group_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
1353 {
1354         unsigned long flags;
1355         unsigned int mask;
1356         int ret = 0;
1357
1358         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1359         read_lock(&tasklist_lock);
1360         spin_lock_irqsave(&p->sighand->siglock, flags);
1361         handle_stop_signal(sig, p);
1362
1363         /* Short-circuit ignored signals.  */
1364         if (sig_ignored(p, sig)) {
1365                 ret = 1;
1366                 goto out;
1367         }
1368
1369         if (unlikely(!list_empty(&q->list))) {
1370                 /*
1371                  * If an SI_TIMER entry is already queue just increment
1372                  * the overrun count.  Other uses should not try to
1373                  * send the signal multiple times.
1374                  */
1375                 if (q->info.si_code != SI_TIMER)
1376                         BUG();
1377                 q->info.si_overrun++;
1378                 goto out;
1379         } 
1380         /*
1381          * Don't bother zombies and stopped tasks (but
1382          * SIGKILL will punch through stopped state)
1383          */
1384         mask = TASK_DEAD | TASK_ZOMBIE;
1385         if (sig != SIGKILL)
1386                 mask |= TASK_STOPPED;
1387
1388         /*
1389          * Put this signal on the shared-pending queue.
1390          * We always use the shared queue for process-wide signals,
1391          * to avoid several races.
1392          */
1393         q->lock = &p->sighand->siglock;
1394         list_add_tail(&q->list, &p->signal->shared_pending.list);
1395         sigaddset(&p->signal->shared_pending.signal, sig);
1396
1397         __group_complete_signal(sig, p, mask);
1398 out:
1399         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1400         read_unlock(&tasklist_lock);
1401         return(ret);
1402 }
1403
1404 /*
1405  * Joy. Or not. Pthread wants us to wake up every thread
1406  * in our parent group.
1407  */
1408 static void __wake_up_parent(struct task_struct *p,
1409                                     struct task_struct *parent)
1410 {
1411         struct task_struct *tsk = parent;
1412
1413         /*
1414          * Fortunately this is not necessary for thread groups:
1415          */
1416         if (p->tgid == tsk->tgid) {
1417                 wake_up_interruptible_sync(&tsk->wait_chldexit);
1418                 return;
1419         }
1420
1421         do {
1422                 wake_up_interruptible_sync(&tsk->wait_chldexit);
1423                 tsk = next_thread(tsk);
1424                 if (tsk->signal != parent->signal)
1425                         BUG();
1426         } while (tsk != parent);
1427 }
1428
1429 /*
1430  * Let a parent know about a status change of a child.
1431  */
1432
1433 void do_notify_parent(struct task_struct *tsk, int sig)
1434 {
1435         struct siginfo info;
1436         unsigned long flags;
1437         int why, status;
1438         struct sighand_struct *psig;
1439
1440         if (sig == -1)
1441                 BUG();
1442
1443         BUG_ON(tsk->group_leader != tsk && tsk->group_leader->state != TASK_ZOMBIE && !tsk->ptrace);
1444         BUG_ON(tsk->group_leader == tsk && !thread_group_empty(tsk) && !tsk->ptrace);
1445
1446         info.si_signo = sig;
1447         info.si_errno = 0;
1448         info.si_pid = tsk->pid;
1449         info.si_uid = tsk->uid;
1450
1451         /* FIXME: find out whether or not this is supposed to be c*time. */
1452         info.si_utime = tsk->utime;
1453         info.si_stime = tsk->stime;
1454
1455         status = tsk->exit_code & 0x7f;
1456         why = SI_KERNEL;        /* shouldn't happen */
1457         switch (tsk->state) {
1458         case TASK_STOPPED:
1459                 /* FIXME -- can we deduce CLD_TRAPPED or CLD_CONTINUED? */
1460                 if (tsk->ptrace & PT_PTRACED)
1461                         why = CLD_TRAPPED;
1462                 else
1463                         why = CLD_STOPPED;
1464                 break;
1465
1466         default:
1467                 if (tsk->exit_code & 0x80)
1468                         why = CLD_DUMPED;
1469                 else if (tsk->exit_code & 0x7f)
1470                         why = CLD_KILLED;
1471                 else {
1472                         why = CLD_EXITED;
1473                         status = tsk->exit_code >> 8;
1474                 }
1475                 break;
1476         }
1477         info.si_code = why;
1478         info.si_status = status;
1479
1480         psig = tsk->parent->sighand;
1481         spin_lock_irqsave(&psig->siglock, flags);
1482         if (sig == SIGCHLD && tsk->state != TASK_STOPPED &&
1483             (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1484              (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1485                 /*
1486                  * We are exiting and our parent doesn't care.  POSIX.1
1487                  * defines special semantics for setting SIGCHLD to SIG_IGN
1488                  * or setting the SA_NOCLDWAIT flag: we should be reaped
1489                  * automatically and not left for our parent's wait4 call.
1490                  * Rather than having the parent do it as a magic kind of
1491                  * signal handler, we just set this to tell do_exit that we
1492                  * can be cleaned up without becoming a zombie.  Note that
1493                  * we still call __wake_up_parent in this case, because a
1494                  * blocked sys_wait4 might now return -ECHILD.
1495                  *
1496                  * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1497                  * is implementation-defined: we do (if you don't want
1498                  * it, just use SIG_IGN instead).
1499                  */
1500                 tsk->exit_signal = -1;
1501                 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1502                         sig = 0;
1503         }
1504         if (sig > 0 && sig <= _NSIG)
1505                 __group_send_sig_info(sig, &info, tsk->parent);
1506         __wake_up_parent(tsk, tsk->parent);
1507         spin_unlock_irqrestore(&psig->siglock, flags);
1508 }
1509
1510
1511 /*
1512  * We need the tasklist lock because it's the only
1513  * thing that protects out "parent" pointer.
1514  *
1515  * exit.c calls "do_notify_parent()" directly, because
1516  * it already has the tasklist lock.
1517  */
1518 void
1519 notify_parent(struct task_struct *tsk, int sig)
1520 {
1521         if (sig != -1) {
1522                 read_lock(&tasklist_lock);
1523                 do_notify_parent(tsk, sig);
1524                 read_unlock(&tasklist_lock);
1525         }
1526 }
1527
1528 static void
1529 do_notify_parent_cldstop(struct task_struct *tsk, struct task_struct *parent)
1530 {
1531         struct siginfo info;
1532         unsigned long flags;
1533         struct sighand_struct *sighand;
1534
1535         info.si_signo = SIGCHLD;
1536         info.si_errno = 0;
1537         info.si_pid = tsk->pid;
1538         info.si_uid = tsk->uid;
1539
1540         /* FIXME: find out whether or not this is supposed to be c*time. */
1541         info.si_utime = tsk->utime;
1542         info.si_stime = tsk->stime;
1543
1544         info.si_status = tsk->exit_code & 0x7f;
1545         info.si_code = CLD_STOPPED;
1546
1547         sighand = parent->sighand;
1548         spin_lock_irqsave(&sighand->siglock, flags);
1549         if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1550             !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1551                 __group_send_sig_info(SIGCHLD, &info, parent);
1552         /*
1553          * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1554          */
1555         __wake_up_parent(tsk, parent);
1556         spin_unlock_irqrestore(&sighand->siglock, flags);
1557 }
1558
1559 int print_fatal_signals = 0;
1560
1561 static void print_fatal_signal(struct pt_regs *regs, int signr)
1562 {
1563         int i;
1564         unsigned char insn;
1565         printk("%s/%d: potentially unexpected fatal signal %d.\n",
1566                 current->comm, current->pid, signr);
1567                 
1568 #ifdef __i386__
1569         printk("code at %08lx: ", regs->eip);
1570         for (i = 0; i < 16; i++) {
1571                 __get_user(insn, (unsigned char *)(regs->eip + i));
1572                 printk("%02x ", insn);
1573         }
1574 #endif  
1575         printk("\n");
1576         show_regs(regs);
1577 }
1578
1579 static int __init setup_print_fatal_signals(char *str)
1580 {
1581         get_option (&str, &print_fatal_signals);
1582
1583         return 1;
1584 }
1585
1586 __setup("print-fatal-signals=", setup_print_fatal_signals);
1587
1588 #ifndef HAVE_ARCH_GET_SIGNAL_TO_DELIVER
1589
1590 static void
1591 finish_stop(int stop_count)
1592 {
1593         /*
1594          * If there are no other threads in the group, or if there is
1595          * a group stop in progress and we are the last to stop,
1596          * report to the parent.  When ptraced, every thread reports itself.
1597          */
1598         if (stop_count < 0 || (current->ptrace & PT_PTRACED)) {
1599                 read_lock(&tasklist_lock);
1600                 do_notify_parent_cldstop(current, current->parent);
1601                 read_unlock(&tasklist_lock);
1602         }
1603         else if (stop_count == 0) {
1604                 read_lock(&tasklist_lock);
1605                 do_notify_parent_cldstop(current->group_leader,
1606                                          current->group_leader->real_parent);
1607                 read_unlock(&tasklist_lock);
1608         }
1609
1610         schedule();
1611         /*
1612          * Now we don't run again until continued.
1613          */
1614         current->exit_code = 0;
1615 }
1616
1617 /*
1618  * This performs the stopping for SIGSTOP and other stop signals.
1619  * We have to stop all threads in the thread group.
1620  */
1621 static void
1622 do_signal_stop(int signr)
1623 {
1624         struct signal_struct *sig = current->signal;
1625         struct sighand_struct *sighand = current->sighand;
1626         int stop_count = -1;
1627
1628         /* spin_lock_irq(&sighand->siglock) is now done in caller */
1629
1630         if (sig->group_stop_count > 0) {
1631                 /*
1632                  * There is a group stop in progress.  We don't need to
1633                  * start another one.
1634                  */
1635                 signr = sig->group_exit_code;
1636                 stop_count = --sig->group_stop_count;
1637                 current->exit_code = signr;
1638                 set_current_state(TASK_STOPPED);
1639                 spin_unlock_irq(&sighand->siglock);
1640         }
1641         else if (thread_group_empty(current)) {
1642                 /*
1643                  * Lock must be held through transition to stopped state.
1644                  */
1645                 current->exit_code = signr;
1646                 set_current_state(TASK_STOPPED);
1647                 spin_unlock_irq(&sighand->siglock);
1648         }
1649         else {
1650                 /*
1651                  * There is no group stop already in progress.
1652                  * We must initiate one now, but that requires
1653                  * dropping siglock to get both the tasklist lock
1654                  * and siglock again in the proper order.  Note that
1655                  * this allows an intervening SIGCONT to be posted.
1656                  * We need to check for that and bail out if necessary.
1657                  */
1658                 struct task_struct *t;
1659
1660                 spin_unlock_irq(&sighand->siglock);
1661
1662                 /* signals can be posted during this window */
1663
1664                 read_lock(&tasklist_lock);
1665                 spin_lock_irq(&sighand->siglock);
1666
1667                 if (unlikely(sig->group_exit)) {
1668                         /*
1669                          * There is a group exit in progress now.
1670                          * We'll just ignore the stop and process the
1671                          * associated fatal signal.
1672                          */
1673                         spin_unlock_irq(&sighand->siglock);
1674                         read_unlock(&tasklist_lock);
1675                         return;
1676                 }
1677
1678                 if (unlikely(sig_avoid_stop_race())) {
1679                         /*
1680                          * Either a SIGCONT or a SIGKILL signal was
1681                          * posted in the siglock-not-held window.
1682                          */
1683                         spin_unlock_irq(&sighand->siglock);
1684                         read_unlock(&tasklist_lock);
1685                         return;
1686                 }
1687
1688                 if (sig->group_stop_count == 0) {
1689                         sig->group_exit_code = signr;
1690                         stop_count = 0;
1691                         for (t = next_thread(current); t != current;
1692                              t = next_thread(t))
1693                                 /*
1694                                  * Setting state to TASK_STOPPED for a group
1695                                  * stop is always done with the siglock held,
1696                                  * so this check has no races.
1697                                  */
1698                                 if (t->state < TASK_STOPPED) {
1699                                         stop_count++;
1700                                         signal_wake_up(t, 0);
1701                                 }
1702                         sig->group_stop_count = stop_count;
1703                 }
1704                 else {
1705                         /* A race with another thread while unlocked.  */
1706                         signr = sig->group_exit_code;
1707                         stop_count = --sig->group_stop_count;
1708                 }
1709
1710                 current->exit_code = signr;
1711                 set_current_state(TASK_STOPPED);
1712
1713                 spin_unlock_irq(&sighand->siglock);
1714                 read_unlock(&tasklist_lock);
1715         }
1716
1717         finish_stop(stop_count);
1718 }
1719
1720 /*
1721  * Do appropriate magic when group_stop_count > 0.
1722  * We return nonzero if we stopped, after releasing the siglock.
1723  * We return zero if we still hold the siglock and should look
1724  * for another signal without checking group_stop_count again.
1725  */
1726 static inline int handle_group_stop(void)
1727 {
1728         int stop_count;
1729
1730         if (current->signal->group_exit_task == current) {
1731                 /*
1732                  * Group stop is so we can do a core dump,
1733                  * We are the initiating thread, so get on with it.
1734                  */
1735                 current->signal->group_exit_task = NULL;
1736                 return 0;
1737         }
1738
1739         if (current->signal->group_exit)
1740                 /*
1741                  * Group stop is so another thread can do a core dump,
1742                  * or else we are racing against a death signal.
1743                  * Just punt the stop so we can get the next signal.
1744                  */
1745                 return 0;
1746
1747         /*
1748          * There is a group stop in progress.  We stop
1749          * without any associated signal being in our queue.
1750          */
1751         stop_count = --current->signal->group_stop_count;
1752         current->exit_code = current->signal->group_exit_code;
1753         set_current_state(TASK_STOPPED);
1754         spin_unlock_irq(&current->sighand->siglock);
1755         finish_stop(stop_count);
1756         return 1;
1757 }
1758
1759 int get_signal_to_deliver(siginfo_t *info, struct pt_regs *regs, void *cookie)
1760 {
1761         sigset_t *mask = &current->blocked;
1762         int signr = 0;
1763
1764 relock:
1765         spin_lock_irq(&current->sighand->siglock);
1766         for (;;) {
1767                 struct k_sigaction *ka;
1768
1769                 if (unlikely(current->signal->group_stop_count > 0) &&
1770                     handle_group_stop())
1771                         goto relock;
1772
1773                 signr = dequeue_signal(current, mask, info);
1774
1775                 if (!signr)
1776                         break; /* will return 0 */
1777
1778                 if ((signr == SIGSEGV) && print_fatal_signals) {
1779                         spin_unlock_irq(&current->sighand->siglock);
1780                         print_fatal_signal(regs, signr);
1781                         spin_lock_irq(&current->sighand->siglock);
1782                 }
1783                 if ((current->ptrace & PT_PTRACED) && signr != SIGKILL) {
1784                         ptrace_signal_deliver(regs, cookie);
1785
1786                         /*
1787                          * If there is a group stop in progress,
1788                          * we must participate in the bookkeeping.
1789                          */
1790                         if (current->signal->group_stop_count > 0)
1791                                 --current->signal->group_stop_count;
1792
1793                         /* Let the debugger run.  */
1794                         current->exit_code = signr;
1795                         current->last_siginfo = info;
1796                         set_current_state(TASK_STOPPED);
1797                         spin_unlock_irq(&current->sighand->siglock);
1798                         notify_parent(current, SIGCHLD);
1799                         schedule();
1800
1801                         current->last_siginfo = NULL;
1802
1803                         /* We're back.  Did the debugger cancel the sig?  */
1804                         spin_lock_irq(&current->sighand->siglock);
1805                         signr = current->exit_code;
1806                         if (signr == 0)
1807                                 continue;
1808
1809                         current->exit_code = 0;
1810
1811                         /* Update the siginfo structure if the signal has
1812                            changed.  If the debugger wanted something
1813                            specific in the siginfo structure then it should
1814                            have updated *info via PTRACE_SETSIGINFO.  */
1815                         if (signr != info->si_signo) {
1816                                 info->si_signo = signr;
1817                                 info->si_errno = 0;
1818                                 info->si_code = SI_USER;
1819                                 info->si_pid = current->parent->pid;
1820                                 info->si_uid = current->parent->uid;
1821                         }
1822
1823                         /* If the (new) signal is now blocked, requeue it.  */
1824                         if (sigismember(&current->blocked, signr)) {
1825                                 specific_send_sig_info(signr, info, current);
1826                                 continue;
1827                         }
1828                 }
1829
1830                 ka = &current->sighand->action[signr-1];
1831                 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
1832                         continue;
1833                 if (ka->sa.sa_handler != SIG_DFL) /* Run the handler.  */
1834                         break; /* will return non-zero "signr" value */
1835
1836                 /*
1837                  * Now we are doing the default action for this signal.
1838                  */
1839                 if (sig_kernel_ignore(signr)) /* Default is nothing. */
1840                         continue;
1841
1842                 /* Init gets no signals it doesn't want.  */
1843                 if (current->pid == 1)
1844                         continue;
1845
1846                 if (sig_kernel_stop(signr)) {
1847                         /*
1848                          * The default action is to stop all threads in
1849                          * the thread group.  The job control signals
1850                          * do nothing in an orphaned pgrp, but SIGSTOP
1851                          * always works.  Note that siglock needs to be
1852                          * dropped during the call to is_orphaned_pgrp()
1853                          * because of lock ordering with tasklist_lock.
1854                          * This allows an intervening SIGCONT to be posted.
1855                          * We need to check for that and bail out if necessary.
1856                          */
1857                         if (signr == SIGSTOP) {
1858                                 do_signal_stop(signr); /* releases siglock */
1859                                 goto relock;
1860                         }
1861                         spin_unlock_irq(&current->sighand->siglock);
1862
1863                         /* signals can be posted during this window */
1864
1865                         if (is_orphaned_pgrp(process_group(current)))
1866                                 goto relock;
1867
1868                         spin_lock_irq(&current->sighand->siglock);
1869                         if (unlikely(sig_avoid_stop_race())) {
1870                                 /*
1871                                  * Either a SIGCONT or a SIGKILL signal was
1872                                  * posted in the siglock-not-held window.
1873                                  */
1874                                 continue;
1875                         }
1876
1877                         do_signal_stop(signr); /* releases siglock */
1878                         goto relock;
1879                 }
1880
1881                 spin_unlock_irq(&current->sighand->siglock);
1882
1883                 /*
1884                  * Anything else is fatal, maybe with a core dump.
1885                  */
1886                 current->flags |= PF_SIGNALED;
1887                 if (print_fatal_signals)
1888                         print_fatal_signal(regs, signr);
1889                 if (sig_kernel_coredump(signr) &&
1890                     do_coredump((long)signr, signr, regs)) {
1891                         /*
1892                          * That killed all other threads in the group and
1893                          * synchronized with their demise, so there can't
1894                          * be any more left to kill now.  The group_exit
1895                          * flags are set by do_coredump.  Note that
1896                          * thread_group_empty won't always be true yet,
1897                          * because those threads were blocked in __exit_mm
1898                          * and we just let them go to finish dying.
1899                          */
1900                         const int code = signr | 0x80;
1901                         BUG_ON(!current->signal->group_exit);
1902                         BUG_ON(current->signal->group_exit_code != code);
1903                         do_exit(code);
1904                         /* NOTREACHED */
1905                 }
1906
1907                 /*
1908                  * Death signals, no core dump.
1909                  */
1910                 do_group_exit(signr);
1911                 /* NOTREACHED */
1912         }
1913         spin_unlock_irq(&current->sighand->siglock);
1914         return signr;
1915 }
1916
1917 #endif
1918
1919 EXPORT_SYMBOL(recalc_sigpending);
1920 EXPORT_SYMBOL_GPL(dequeue_signal);
1921 EXPORT_SYMBOL(flush_signals);
1922 EXPORT_SYMBOL(force_sig);
1923 EXPORT_SYMBOL(force_sig_info);
1924 EXPORT_SYMBOL(kill_pg);
1925 EXPORT_SYMBOL(kill_pg_info);
1926 EXPORT_SYMBOL(kill_proc);
1927 EXPORT_SYMBOL(kill_proc_info);
1928 EXPORT_SYMBOL(kill_sl);
1929 EXPORT_SYMBOL(kill_sl_info);
1930 EXPORT_SYMBOL(notify_parent);
1931 EXPORT_SYMBOL(send_sig);
1932 EXPORT_SYMBOL(send_sig_info);
1933 EXPORT_SYMBOL(send_group_sig_info);
1934 EXPORT_SYMBOL(sigqueue_alloc);
1935 EXPORT_SYMBOL(sigqueue_free);
1936 EXPORT_SYMBOL(send_sigqueue);
1937 EXPORT_SYMBOL(send_group_sigqueue);
1938 EXPORT_SYMBOL(sigprocmask);
1939 EXPORT_SYMBOL(block_all_signals);
1940 EXPORT_SYMBOL(unblock_all_signals);
1941
1942
1943 /*
1944  * System call entry points.
1945  */
1946
1947 asmlinkage long sys_restart_syscall(void)
1948 {
1949         struct restart_block *restart = &current_thread_info()->restart_block;
1950         return restart->fn(restart);
1951 }
1952
1953 long do_no_restart_syscall(struct restart_block *param)
1954 {
1955         return -EINTR;
1956 }
1957
1958 /*
1959  * We don't need to get the kernel lock - this is all local to this
1960  * particular thread.. (and that's good, because this is _heavily_
1961  * used by various programs)
1962  */
1963
1964 /*
1965  * This is also useful for kernel threads that want to temporarily
1966  * (or permanently) block certain signals.
1967  *
1968  * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
1969  * interface happily blocks "unblockable" signals like SIGKILL
1970  * and friends.
1971  */
1972 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
1973 {
1974         int error;
1975         sigset_t old_block;
1976
1977         spin_lock_irq(&current->sighand->siglock);
1978         old_block = current->blocked;
1979         error = 0;
1980         switch (how) {
1981         case SIG_BLOCK:
1982                 sigorsets(&current->blocked, &current->blocked, set);
1983                 break;
1984         case SIG_UNBLOCK:
1985                 signandsets(&current->blocked, &current->blocked, set);
1986                 break;
1987         case SIG_SETMASK:
1988                 current->blocked = *set;
1989                 break;
1990         default:
1991                 error = -EINVAL;
1992         }
1993         recalc_sigpending();
1994         spin_unlock_irq(&current->sighand->siglock);
1995         if (oldset)
1996                 *oldset = old_block;
1997         return error;
1998 }
1999
2000 asmlinkage long
2001 sys_rt_sigprocmask(int how, sigset_t __user *set, sigset_t __user *oset, size_t sigsetsize)
2002 {
2003         int error = -EINVAL;
2004         sigset_t old_set, new_set;
2005
2006         /* XXX: Don't preclude handling different sized sigset_t's.  */
2007         if (sigsetsize != sizeof(sigset_t))
2008                 goto out;
2009
2010         if (set) {
2011                 error = -EFAULT;
2012                 if (copy_from_user(&new_set, set, sizeof(*set)))
2013                         goto out;
2014                 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2015
2016                 error = sigprocmask(how, &new_set, &old_set);
2017                 if (error)
2018                         goto out;
2019                 if (oset)
2020                         goto set_old;
2021         } else if (oset) {
2022                 spin_lock_irq(&current->sighand->siglock);
2023                 old_set = current->blocked;
2024                 spin_unlock_irq(&current->sighand->siglock);
2025
2026         set_old:
2027                 error = -EFAULT;
2028                 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2029                         goto out;
2030         }
2031         error = 0;
2032 out:
2033         return error;
2034 }
2035
2036 long do_sigpending(void __user *set, unsigned long sigsetsize)
2037 {
2038         long error = -EINVAL;
2039         sigset_t pending;
2040
2041         if (sigsetsize > sizeof(sigset_t))
2042                 goto out;
2043
2044         spin_lock_irq(&current->sighand->siglock);
2045         sigorsets(&pending, &current->pending.signal,
2046                   &current->signal->shared_pending.signal);
2047         spin_unlock_irq(&current->sighand->siglock);
2048
2049         /* Outside the lock because only this thread touches it.  */
2050         sigandsets(&pending, &current->blocked, &pending);
2051
2052         error = -EFAULT;
2053         if (!copy_to_user(set, &pending, sigsetsize))
2054                 error = 0;
2055
2056 out:
2057         return error;
2058 }       
2059
2060 asmlinkage long
2061 sys_rt_sigpending(sigset_t __user *set, size_t sigsetsize)
2062 {
2063         return do_sigpending(set, sigsetsize);
2064 }
2065
2066 #ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2067
2068 int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2069 {
2070         int err;
2071
2072         if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2073                 return -EFAULT;
2074         if (from->si_code < 0)
2075                 return __copy_to_user(to, from, sizeof(siginfo_t))
2076                         ? -EFAULT : 0;
2077         /*
2078          * If you change siginfo_t structure, please be sure
2079          * this code is fixed accordingly.
2080          * It should never copy any pad contained in the structure
2081          * to avoid security leaks, but must copy the generic
2082          * 3 ints plus the relevant union member.
2083          */
2084         err = __put_user(from->si_signo, &to->si_signo);
2085         err |= __put_user(from->si_errno, &to->si_errno);
2086         err |= __put_user((short)from->si_code, &to->si_code);
2087         switch (from->si_code & __SI_MASK) {
2088         case __SI_KILL:
2089                 err |= __put_user(from->si_pid, &to->si_pid);
2090                 err |= __put_user(from->si_uid, &to->si_uid);
2091                 break;
2092         case __SI_TIMER:
2093                  err |= __put_user(from->si_tid, &to->si_tid);
2094                  err |= __put_user(from->si_overrun, &to->si_overrun);
2095                  err |= __put_user(from->si_ptr, &to->si_ptr);
2096                 break;
2097         case __SI_POLL:
2098                 err |= __put_user(from->si_band, &to->si_band);
2099                 err |= __put_user(from->si_fd, &to->si_fd);
2100                 break;
2101         case __SI_FAULT:
2102                 err |= __put_user(from->si_addr, &to->si_addr);
2103 #ifdef __ARCH_SI_TRAPNO
2104                 err |= __put_user(from->si_trapno, &to->si_trapno);
2105 #endif
2106                 break;
2107         case __SI_CHLD:
2108                 err |= __put_user(from->si_pid, &to->si_pid);
2109                 err |= __put_user(from->si_uid, &to->si_uid);
2110                 err |= __put_user(from->si_status, &to->si_status);
2111                 err |= __put_user(from->si_utime, &to->si_utime);
2112                 err |= __put_user(from->si_stime, &to->si_stime);
2113                 break;
2114         case __SI_RT: /* This is not generated by the kernel as of now. */
2115         case __SI_MESGQ: /* But this is */
2116                 err |= __put_user(from->si_pid, &to->si_pid);
2117                 err |= __put_user(from->si_uid, &to->si_uid);
2118                 err |= __put_user(from->si_ptr, &to->si_ptr);
2119                 break;
2120         default: /* this is just in case for now ... */
2121                 err |= __put_user(from->si_pid, &to->si_pid);
2122                 err |= __put_user(from->si_uid, &to->si_uid);
2123                 break;
2124         }
2125         return err;
2126 }
2127
2128 #endif
2129
2130 asmlinkage long
2131 sys_rt_sigtimedwait(const sigset_t __user *uthese,
2132                     siginfo_t __user *uinfo,
2133                     const struct timespec __user *uts,
2134                     size_t sigsetsize)
2135 {
2136         int ret, sig;
2137         sigset_t these;
2138         struct timespec ts;
2139         siginfo_t info;
2140         long timeout = 0;
2141
2142         /* XXX: Don't preclude handling different sized sigset_t's.  */
2143         if (sigsetsize != sizeof(sigset_t))
2144                 return -EINVAL;
2145
2146         if (copy_from_user(&these, uthese, sizeof(these)))
2147                 return -EFAULT;
2148                 
2149         /*
2150          * Invert the set of allowed signals to get those we
2151          * want to block.
2152          */
2153         sigdelsetmask(&these, sigmask(SIGKILL)|sigmask(SIGSTOP));
2154         signotset(&these);
2155
2156         if (uts) {
2157                 if (copy_from_user(&ts, uts, sizeof(ts)))
2158                         return -EFAULT;
2159                 if (ts.tv_nsec >= 1000000000L || ts.tv_nsec < 0
2160                     || ts.tv_sec < 0)
2161                         return -EINVAL;
2162         }
2163
2164         spin_lock_irq(&current->sighand->siglock);
2165         sig = dequeue_signal(current, &these, &info);
2166         if (!sig) {
2167                 timeout = MAX_SCHEDULE_TIMEOUT;
2168                 if (uts)
2169                         timeout = (timespec_to_jiffies(&ts)
2170                                    + (ts.tv_sec || ts.tv_nsec));
2171
2172                 if (timeout) {
2173                         /* None ready -- temporarily unblock those we're
2174                          * interested while we are sleeping in so that we'll
2175                          * be awakened when they arrive.  */
2176                         current->real_blocked = current->blocked;
2177                         sigandsets(&current->blocked, &current->blocked, &these);
2178                         recalc_sigpending();
2179                         spin_unlock_irq(&current->sighand->siglock);
2180
2181                         current->state = TASK_INTERRUPTIBLE;
2182                         timeout = schedule_timeout(timeout);
2183
2184                         spin_lock_irq(&current->sighand->siglock);
2185                         sig = dequeue_signal(current, &these, &info);
2186                         current->blocked = current->real_blocked;
2187                         siginitset(&current->real_blocked, 0);
2188                         recalc_sigpending();
2189                 }
2190         }
2191         spin_unlock_irq(&current->sighand->siglock);
2192
2193         if (sig) {
2194                 ret = sig;
2195                 if (uinfo) {
2196                         if (copy_siginfo_to_user(uinfo, &info))
2197                                 ret = -EFAULT;
2198                 }
2199         } else {
2200                 ret = -EAGAIN;
2201                 if (timeout)
2202                         ret = -EINTR;
2203         }
2204
2205         return ret;
2206 }
2207
2208 asmlinkage long
2209 sys_kill(int pid, int sig)
2210 {
2211         struct siginfo info;
2212
2213         info.si_signo = sig;
2214         info.si_errno = 0;
2215         info.si_code = SI_USER;
2216         info.si_pid = current->tgid;
2217         info.si_uid = current->uid;
2218
2219         return kill_something_info(sig, &info, pid);
2220 }
2221
2222 /**
2223  *  sys_tgkill - send signal to one specific thread
2224  *  @tgid: the thread group ID of the thread
2225  *  @pid: the PID of the thread
2226  *  @sig: signal to be sent
2227  *
2228  *  This syscall also checks the tgid and returns -ESRCH even if the PID
2229  *  exists but it's not belonging to the target process anymore. This
2230  *  method solves the problem of threads exiting and PIDs getting reused.
2231  */
2232 asmlinkage long sys_tgkill(int tgid, int pid, int sig)
2233 {
2234         struct siginfo info;
2235         int error;
2236         struct task_struct *p;
2237
2238         /* This is only valid for single tasks */
2239         if (pid <= 0 || tgid <= 0)
2240                 return -EINVAL;
2241
2242         info.si_signo = sig;
2243         info.si_errno = 0;
2244         info.si_code = SI_TKILL;
2245         info.si_pid = current->tgid;
2246         info.si_uid = current->uid;
2247
2248         read_lock(&tasklist_lock);
2249         p = find_task_by_pid(pid);
2250         error = -ESRCH;
2251         if (p && (p->tgid == tgid)) {
2252                 error = check_kill_permission(sig, &info, p);
2253                 /*
2254                  * The null signal is a permissions and process existence
2255                  * probe.  No signal is actually delivered.
2256                  */
2257                 if (!error && sig && p->sighand) {
2258                         spin_lock_irq(&p->sighand->siglock);
2259                         handle_stop_signal(sig, p);
2260                         error = specific_send_sig_info(sig, &info, p);
2261                         spin_unlock_irq(&p->sighand->siglock);
2262                 }
2263         }
2264         read_unlock(&tasklist_lock);
2265         return error;
2266 }
2267
2268 /*
2269  *  Send a signal to only one task, even if it's a CLONE_THREAD task.
2270  */
2271 asmlinkage long
2272 sys_tkill(int pid, int sig)
2273 {
2274         struct siginfo info;
2275         int error;
2276         struct task_struct *p;
2277
2278         /* This is only valid for single tasks */
2279         if (pid <= 0)
2280                 return -EINVAL;
2281
2282         info.si_signo = sig;
2283         info.si_errno = 0;
2284         info.si_code = SI_TKILL;
2285         info.si_pid = current->tgid;
2286         info.si_uid = current->uid;
2287
2288         read_lock(&tasklist_lock);
2289         p = find_task_by_pid(pid);
2290         error = -ESRCH;
2291         if (p) {
2292                 error = check_kill_permission(sig, &info, p);
2293                 /*
2294                  * The null signal is a permissions and process existence
2295                  * probe.  No signal is actually delivered.
2296                  */
2297                 if (!error && sig && p->sighand) {
2298                         spin_lock_irq(&p->sighand->siglock);
2299                         handle_stop_signal(sig, p);
2300                         error = specific_send_sig_info(sig, &info, p);
2301                         spin_unlock_irq(&p->sighand->siglock);
2302                 }
2303         }
2304         read_unlock(&tasklist_lock);
2305         return error;
2306 }
2307
2308 asmlinkage long
2309 sys_rt_sigqueueinfo(int pid, int sig, siginfo_t __user *uinfo)
2310 {
2311         siginfo_t info;
2312
2313         if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2314                 return -EFAULT;
2315
2316         /* Not even root can pretend to send signals from the kernel.
2317            Nor can they impersonate a kill(), which adds source info.  */
2318         if (info.si_code >= 0)
2319                 return -EPERM;
2320         info.si_signo = sig;
2321
2322         /* POSIX.1b doesn't mention process groups.  */
2323         return kill_proc_info(sig, &info, pid);
2324 }
2325
2326 int
2327 do_sigaction(int sig, const struct k_sigaction *act, struct k_sigaction *oact)
2328 {
2329         struct k_sigaction *k;
2330
2331         if (sig < 1 || sig > _NSIG || (act && sig_kernel_only(sig)))
2332                 return -EINVAL;
2333
2334         k = &current->sighand->action[sig-1];
2335
2336         spin_lock_irq(&current->sighand->siglock);
2337         if (signal_pending(current)) {
2338                 /*
2339                  * If there might be a fatal signal pending on multiple
2340                  * threads, make sure we take it before changing the action.
2341                  */
2342                 spin_unlock_irq(&current->sighand->siglock);
2343                 return -ERESTARTNOINTR;
2344         }
2345
2346         if (oact)
2347                 *oact = *k;
2348
2349         if (act) {
2350                 /*
2351                  * POSIX 3.3.1.3:
2352                  *  "Setting a signal action to SIG_IGN for a signal that is
2353                  *   pending shall cause the pending signal to be discarded,
2354                  *   whether or not it is blocked."
2355                  *
2356                  *  "Setting a signal action to SIG_DFL for a signal that is
2357                  *   pending and whose default action is to ignore the signal
2358                  *   (for example, SIGCHLD), shall cause the pending signal to
2359                  *   be discarded, whether or not it is blocked"
2360                  */
2361                 if (act->sa.sa_handler == SIG_IGN ||
2362                     (act->sa.sa_handler == SIG_DFL &&
2363                      sig_kernel_ignore(sig))) {
2364                         /*
2365                          * This is a fairly rare case, so we only take the
2366                          * tasklist_lock once we're sure we'll need it.
2367                          * Now we must do this little unlock and relock
2368                          * dance to maintain the lock hierarchy.
2369                          */
2370                         struct task_struct *t = current;
2371                         spin_unlock_irq(&t->sighand->siglock);
2372                         read_lock(&tasklist_lock);
2373                         spin_lock_irq(&t->sighand->siglock);
2374                         *k = *act;
2375                         sigdelsetmask(&k->sa.sa_mask,
2376                                       sigmask(SIGKILL) | sigmask(SIGSTOP));
2377                         rm_from_queue(sigmask(sig), &t->signal->shared_pending);
2378                         do {
2379                                 rm_from_queue(sigmask(sig), &t->pending);
2380                                 recalc_sigpending_tsk(t);
2381                                 t = next_thread(t);
2382                         } while (t != current);
2383                         spin_unlock_irq(&current->sighand->siglock);
2384                         read_unlock(&tasklist_lock);
2385                         return 0;
2386                 }
2387
2388                 *k = *act;
2389                 sigdelsetmask(&k->sa.sa_mask,
2390                               sigmask(SIGKILL) | sigmask(SIGSTOP));
2391         }
2392
2393         spin_unlock_irq(&current->sighand->siglock);
2394         return 0;
2395 }
2396
2397 int 
2398 do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
2399 {
2400         stack_t oss;
2401         int error;
2402
2403         if (uoss) {
2404                 oss.ss_sp = (void __user *) current->sas_ss_sp;
2405                 oss.ss_size = current->sas_ss_size;
2406                 oss.ss_flags = sas_ss_flags(sp);
2407         }
2408
2409         if (uss) {
2410                 void __user *ss_sp;
2411                 size_t ss_size;
2412                 int ss_flags;
2413
2414                 error = -EFAULT;
2415                 if (verify_area(VERIFY_READ, uss, sizeof(*uss))
2416                     || __get_user(ss_sp, &uss->ss_sp)
2417                     || __get_user(ss_flags, &uss->ss_flags)
2418                     || __get_user(ss_size, &uss->ss_size))
2419                         goto out;
2420
2421                 error = -EPERM;
2422                 if (on_sig_stack(sp))
2423                         goto out;
2424
2425                 error = -EINVAL;
2426                 /*
2427                  *
2428                  * Note - this code used to test ss_flags incorrectly
2429                  *        old code may have been written using ss_flags==0
2430                  *        to mean ss_flags==SS_ONSTACK (as this was the only
2431                  *        way that worked) - this fix preserves that older
2432                  *        mechanism
2433                  */
2434                 if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
2435                         goto out;
2436
2437                 if (ss_flags == SS_DISABLE) {
2438                         ss_size = 0;
2439                         ss_sp = NULL;
2440                 } else {
2441                         error = -ENOMEM;
2442                         if (ss_size < MINSIGSTKSZ)
2443                                 goto out;
2444                 }
2445
2446                 current->sas_ss_sp = (unsigned long) ss_sp;
2447                 current->sas_ss_size = ss_size;
2448         }
2449
2450         if (uoss) {
2451                 error = -EFAULT;
2452                 if (copy_to_user(uoss, &oss, sizeof(oss)))
2453                         goto out;
2454         }
2455
2456         error = 0;
2457 out:
2458         return error;
2459 }
2460
2461 #ifdef __ARCH_WANT_SYS_SIGPENDING
2462
2463 asmlinkage long
2464 sys_sigpending(old_sigset_t __user *set)
2465 {
2466         return do_sigpending(set, sizeof(*set));
2467 }
2468
2469 #endif
2470
2471 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
2472 /* Some platforms have their own version with special arguments others
2473    support only sys_rt_sigprocmask.  */
2474
2475 asmlinkage long
2476 sys_sigprocmask(int how, old_sigset_t __user *set, old_sigset_t __user *oset)
2477 {
2478         int error;
2479         old_sigset_t old_set, new_set;
2480
2481         if (set) {
2482                 error = -EFAULT;
2483                 if (copy_from_user(&new_set, set, sizeof(*set)))
2484                         goto out;
2485                 new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
2486
2487                 spin_lock_irq(&current->sighand->siglock);
2488                 old_set = current->blocked.sig[0];
2489
2490                 error = 0;
2491                 switch (how) {
2492                 default:
2493                         error = -EINVAL;
2494                         break;
2495                 case SIG_BLOCK:
2496                         sigaddsetmask(&current->blocked, new_set);
2497                         break;
2498                 case SIG_UNBLOCK:
2499                         sigdelsetmask(&current->blocked, new_set);
2500                         break;
2501                 case SIG_SETMASK:
2502                         current->blocked.sig[0] = new_set;
2503                         break;
2504                 }
2505
2506                 recalc_sigpending();
2507                 spin_unlock_irq(&current->sighand->siglock);
2508                 if (error)
2509                         goto out;
2510                 if (oset)
2511                         goto set_old;
2512         } else if (oset) {
2513                 old_set = current->blocked.sig[0];
2514         set_old:
2515                 error = -EFAULT;
2516                 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2517                         goto out;
2518         }
2519         error = 0;
2520 out:
2521         return error;
2522 }
2523 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
2524
2525 #ifdef __ARCH_WANT_SYS_RT_SIGACTION
2526 asmlinkage long
2527 sys_rt_sigaction(int sig,
2528                  const struct sigaction __user *act,
2529                  struct sigaction __user *oact,
2530                  size_t sigsetsize)
2531 {
2532         struct k_sigaction new_sa, old_sa;
2533         int ret = -EINVAL;
2534
2535         /* XXX: Don't preclude handling different sized sigset_t's.  */
2536         if (sigsetsize != sizeof(sigset_t))
2537                 goto out;
2538
2539         if (act) {
2540                 if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
2541                         return -EFAULT;
2542         }
2543
2544         ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
2545
2546         if (!ret && oact) {
2547                 if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
2548                         return -EFAULT;
2549         }
2550 out:
2551         return ret;
2552 }
2553 #endif /* __ARCH_WANT_SYS_RT_SIGACTION */
2554
2555 #ifdef __ARCH_WANT_SYS_SGETMASK
2556
2557 /*
2558  * For backwards compatibility.  Functionality superseded by sigprocmask.
2559  */
2560 asmlinkage long
2561 sys_sgetmask(void)
2562 {
2563         /* SMP safe */
2564         return current->blocked.sig[0];
2565 }
2566
2567 asmlinkage long
2568 sys_ssetmask(int newmask)
2569 {
2570         int old;
2571
2572         spin_lock_irq(&current->sighand->siglock);
2573         old = current->blocked.sig[0];
2574
2575         siginitset(&current->blocked, newmask & ~(sigmask(SIGKILL)|
2576                                                   sigmask(SIGSTOP)));
2577         recalc_sigpending();
2578         spin_unlock_irq(&current->sighand->siglock);
2579
2580         return old;
2581 }
2582 #endif /* __ARCH_WANT_SGETMASK */
2583
2584 #ifdef __ARCH_WANT_SYS_SIGNAL
2585 /*
2586  * For backwards compatibility.  Functionality superseded by sigaction.
2587  */
2588 asmlinkage unsigned long
2589 sys_signal(int sig, __sighandler_t handler)
2590 {
2591         struct k_sigaction new_sa, old_sa;
2592         int ret;
2593
2594         new_sa.sa.sa_handler = handler;
2595         new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
2596
2597         ret = do_sigaction(sig, &new_sa, &old_sa);
2598
2599         return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
2600 }
2601 #endif /* __ARCH_WANT_SYS_SIGNAL */
2602
2603 #ifdef __ARCH_WANT_SYS_PAUSE
2604
2605 asmlinkage long
2606 sys_pause(void)
2607 {
2608         current->state = TASK_INTERRUPTIBLE;
2609         schedule();
2610         return -ERESTARTNOHAND;
2611 }
2612
2613 #endif
2614
2615 void __init signals_init(void)
2616 {
2617         sigqueue_cachep =
2618                 kmem_cache_create("sigqueue",
2619                                   sizeof(struct sigqueue),
2620                                   __alignof__(struct sigqueue),
2621                                   SLAB_PANIC, NULL, NULL);
2622 }