patch-2_6_7-vs1_9_1_12
[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 atomic_t nr_queued_signals;
36 int max_queued_signals = 1024;
37
38 /*
39  * In POSIX a signal is sent either to a specific thread (Linux task)
40  * or to the process as a whole (Linux thread group).  How the signal
41  * is sent determines whether it's to one thread or the whole group,
42  * which determines which signal mask(s) are involved in blocking it
43  * from being delivered until later.  When the signal is delivered,
44  * either it's caught or ignored by a user handler or it has a default
45  * effect that applies to the whole thread group (POSIX process).
46  *
47  * The possible effects an unblocked signal set to SIG_DFL can have are:
48  *   ignore     - Nothing Happens
49  *   terminate  - kill the process, i.e. all threads in the group,
50  *                similar to exit_group.  The group leader (only) reports
51  *                WIFSIGNALED status to its parent.
52  *   coredump   - write a core dump file describing all threads using
53  *                the same mm and then kill all those threads
54  *   stop       - stop all the threads in the group, i.e. TASK_STOPPED state
55  *
56  * SIGKILL and SIGSTOP cannot be caught, blocked, or ignored.
57  * Other signals when not blocked and set to SIG_DFL behaves as follows.
58  * The job control signals also have other special effects.
59  *
60  *      +--------------------+------------------+
61  *      |  POSIX signal      |  default action  |
62  *      +--------------------+------------------+
63  *      |  SIGHUP            |  terminate       |
64  *      |  SIGINT            |  terminate       |
65  *      |  SIGQUIT           |  coredump        |
66  *      |  SIGILL            |  coredump        |
67  *      |  SIGTRAP           |  coredump        |
68  *      |  SIGABRT/SIGIOT    |  coredump        |
69  *      |  SIGBUS            |  coredump        |
70  *      |  SIGFPE            |  coredump        |
71  *      |  SIGKILL           |  terminate(+)    |
72  *      |  SIGUSR1           |  terminate       |
73  *      |  SIGSEGV           |  coredump        |
74  *      |  SIGUSR2           |  terminate       |
75  *      |  SIGPIPE           |  terminate       |
76  *      |  SIGALRM           |  terminate       |
77  *      |  SIGTERM           |  terminate       |
78  *      |  SIGCHLD           |  ignore          |
79  *      |  SIGCONT           |  ignore(*)       |
80  *      |  SIGSTOP           |  stop(*)(+)      |
81  *      |  SIGTSTP           |  stop(*)         |
82  *      |  SIGTTIN           |  stop(*)         |
83  *      |  SIGTTOU           |  stop(*)         |
84  *      |  SIGURG            |  ignore          |
85  *      |  SIGXCPU           |  coredump        |
86  *      |  SIGXFSZ           |  coredump        |
87  *      |  SIGVTALRM         |  terminate       |
88  *      |  SIGPROF           |  terminate       |
89  *      |  SIGPOLL/SIGIO     |  terminate       |
90  *      |  SIGSYS/SIGUNUSED  |  coredump        |
91  *      |  SIGSTKFLT         |  terminate       |
92  *      |  SIGWINCH          |  ignore          |
93  *      |  SIGPWR            |  terminate       |
94  *      |  SIGRTMIN-SIGRTMAX |  terminate       |
95  *      +--------------------+------------------+
96  *      |  non-POSIX signal  |  default action  |
97  *      +--------------------+------------------+
98  *      |  SIGEMT            |  coredump        |
99  *      +--------------------+------------------+
100  *
101  * (+) For SIGKILL and SIGSTOP the action is "always", not just "default".
102  * (*) Special job control effects:
103  * When SIGCONT is sent, it resumes the process (all threads in the group)
104  * from TASK_STOPPED state and also clears any pending/queued stop signals
105  * (any of those marked with "stop(*)").  This happens regardless of blocking,
106  * catching, or ignoring SIGCONT.  When any stop signal is sent, it clears
107  * any pending/queued SIGCONT signals; this happens regardless of blocking,
108  * catching, or ignored the stop signal, though (except for SIGSTOP) the
109  * default action of stopping the process may happen later or never.
110  */
111
112 #ifdef SIGEMT
113 #define M_SIGEMT        M(SIGEMT)
114 #else
115 #define M_SIGEMT        0
116 #endif
117
118 #if SIGRTMIN > BITS_PER_LONG
119 #define M(sig) (1ULL << ((sig)-1))
120 #else
121 #define M(sig) (1UL << ((sig)-1))
122 #endif
123 #define T(sig, mask) (M(sig) & (mask))
124
125 #define SIG_KERNEL_ONLY_MASK (\
126         M(SIGKILL)   |  M(SIGSTOP)                                   )
127
128 #define SIG_KERNEL_STOP_MASK (\
129         M(SIGSTOP)   |  M(SIGTSTP)   |  M(SIGTTIN)   |  M(SIGTTOU)   )
130
131 #define SIG_KERNEL_COREDUMP_MASK (\
132         M(SIGQUIT)   |  M(SIGILL)    |  M(SIGTRAP)   |  M(SIGABRT)   | \
133         M(SIGFPE)    |  M(SIGSEGV)   |  M(SIGBUS)    |  M(SIGSYS)    | \
134         M(SIGXCPU)   |  M(SIGXFSZ)   |  M_SIGEMT                     )
135
136 #define SIG_KERNEL_IGNORE_MASK (\
137         M(SIGCONT)   |  M(SIGCHLD)   |  M(SIGWINCH)  |  M(SIGURG)    )
138
139 #define sig_kernel_only(sig) \
140                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_ONLY_MASK))
141 #define sig_kernel_coredump(sig) \
142                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_COREDUMP_MASK))
143 #define sig_kernel_ignore(sig) \
144                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_IGNORE_MASK))
145 #define sig_kernel_stop(sig) \
146                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_STOP_MASK))
147
148 #define sig_user_defined(t, signr) \
149         (((t)->sighand->action[(signr)-1].sa.sa_handler != SIG_DFL) &&  \
150          ((t)->sighand->action[(signr)-1].sa.sa_handler != SIG_IGN))
151
152 #define sig_fatal(t, signr) \
153         (!T(signr, SIG_KERNEL_IGNORE_MASK|SIG_KERNEL_STOP_MASK) && \
154          (t)->sighand->action[(signr)-1].sa.sa_handler == SIG_DFL)
155
156 #define sig_avoid_stop_race() \
157         (sigtestsetmask(&current->pending.signal, M(SIGCONT) | M(SIGKILL)) || \
158          sigtestsetmask(&current->signal->shared_pending.signal, \
159                                                   M(SIGCONT) | M(SIGKILL)))
160
161 static int sig_ignored(struct task_struct *t, int sig)
162 {
163         void * handler;
164
165         /*
166          * Tracers always want to know about signals..
167          */
168         if (t->ptrace & PT_PTRACED)
169                 return 0;
170
171         /*
172          * Blocked signals are never ignored, since the
173          * signal handler may change by the time it is
174          * unblocked.
175          */
176         if (sigismember(&t->blocked, sig))
177                 return 0;
178
179         /* Is it explicitly or implicitly ignored? */
180         handler = t->sighand->action[sig-1].sa.sa_handler;
181         return   handler == SIG_IGN ||
182                 (handler == SIG_DFL && sig_kernel_ignore(sig));
183 }
184
185 /*
186  * Re-calculate pending state from the set of locally pending
187  * signals, globally pending signals, and blocked signals.
188  */
189 static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
190 {
191         unsigned long ready;
192         long i;
193
194         switch (_NSIG_WORDS) {
195         default:
196                 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
197                         ready |= signal->sig[i] &~ blocked->sig[i];
198                 break;
199
200         case 4: ready  = signal->sig[3] &~ blocked->sig[3];
201                 ready |= signal->sig[2] &~ blocked->sig[2];
202                 ready |= signal->sig[1] &~ blocked->sig[1];
203                 ready |= signal->sig[0] &~ blocked->sig[0];
204                 break;
205
206         case 2: ready  = signal->sig[1] &~ blocked->sig[1];
207                 ready |= signal->sig[0] &~ blocked->sig[0];
208                 break;
209
210         case 1: ready  = signal->sig[0] &~ blocked->sig[0];
211         }
212         return ready != 0;
213 }
214
215 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
216
217 fastcall void recalc_sigpending_tsk(struct task_struct *t)
218 {
219         if (t->signal->group_stop_count > 0 ||
220             PENDING(&t->pending, &t->blocked) ||
221             PENDING(&t->signal->shared_pending, &t->blocked))
222                 set_tsk_thread_flag(t, TIF_SIGPENDING);
223         else
224                 clear_tsk_thread_flag(t, TIF_SIGPENDING);
225 }
226
227 void recalc_sigpending(void)
228 {
229         recalc_sigpending_tsk(current);
230 }
231
232 /* Given the mask, find the first available signal that should be serviced. */
233
234 static int
235 next_signal(struct sigpending *pending, sigset_t *mask)
236 {
237         unsigned long i, *s, *m, x;
238         int sig = 0;
239         
240         s = pending->signal.sig;
241         m = mask->sig;
242         switch (_NSIG_WORDS) {
243         default:
244                 for (i = 0; i < _NSIG_WORDS; ++i, ++s, ++m)
245                         if ((x = *s &~ *m) != 0) {
246                                 sig = ffz(~x) + i*_NSIG_BPW + 1;
247                                 break;
248                         }
249                 break;
250
251         case 2: if ((x = s[0] &~ m[0]) != 0)
252                         sig = 1;
253                 else if ((x = s[1] &~ m[1]) != 0)
254                         sig = _NSIG_BPW + 1;
255                 else
256                         break;
257                 sig += ffz(~x);
258                 break;
259
260         case 1: if ((x = *s &~ *m) != 0)
261                         sig = ffz(~x) + 1;
262                 break;
263         }
264         
265         return sig;
266 }
267
268 struct sigqueue *__sigqueue_alloc(void)
269 {
270         struct sigqueue *q = 0;
271
272         if (atomic_read(&nr_queued_signals) < max_queued_signals)
273                 q = kmem_cache_alloc(sigqueue_cachep, GFP_ATOMIC);
274         if (q) {
275                 atomic_inc(&nr_queued_signals);
276                 INIT_LIST_HEAD(&q->list);
277                 q->flags = 0;
278                 q->lock = 0;
279         }
280         return(q);
281 }
282
283 static inline void __sigqueue_free(struct sigqueue *q)
284 {
285         if (q->flags & SIGQUEUE_PREALLOC)
286                 return;
287         kmem_cache_free(sigqueue_cachep, q);
288         atomic_dec(&nr_queued_signals);
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
421 /* Notify the system that a driver wants to block all signals for this
422  * process, and wants to be notified if any signals at all were to be
423  * sent/acted upon.  If the notifier routine returns non-zero, then the
424  * signal will be acted upon after all.  If the notifier routine returns 0,
425  * then then signal will be blocked.  Only one block per process is
426  * allowed.  priv is a pointer to private data that the notifier routine
427  * can use to determine if the signal should be blocked or not.  */
428
429 void
430 block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
431 {
432         unsigned long flags;
433
434         spin_lock_irqsave(&current->sighand->siglock, flags);
435         current->notifier_mask = mask;
436         current->notifier_data = priv;
437         current->notifier = notifier;
438         spin_unlock_irqrestore(&current->sighand->siglock, flags);
439 }
440
441 /* Notify the system that blocking has ended. */
442
443 void
444 unblock_all_signals(void)
445 {
446         unsigned long flags;
447
448         spin_lock_irqsave(&current->sighand->siglock, flags);
449         current->notifier = NULL;
450         current->notifier_data = NULL;
451         recalc_sigpending();
452         spin_unlock_irqrestore(&current->sighand->siglock, flags);
453 }
454
455 static inline int collect_signal(int sig, struct sigpending *list, siginfo_t *info)
456 {
457         struct sigqueue *q, *first = 0;
458         int still_pending = 0;
459
460         if (unlikely(!sigismember(&list->signal, sig)))
461                 return 0;
462
463         /*
464          * Collect the siginfo appropriate to this signal.  Check if
465          * there is another siginfo for the same signal.
466         */
467         list_for_each_entry(q, &list->list, list) {
468                 if (q->info.si_signo == sig) {
469                         if (first) {
470                                 still_pending = 1;
471                                 break;
472                         }
473                         first = q;
474                 }
475         }
476         if (first) {
477                 list_del_init(&first->list);
478                 copy_siginfo(info, &first->info);
479                 __sigqueue_free(first);
480                 if (!still_pending)
481                         sigdelset(&list->signal, sig);
482         } else {
483
484                 /* Ok, it wasn't in the queue.  This must be
485                    a fast-pathed signal or we must have been
486                    out of queue space.  So zero out the info.
487                  */
488                 sigdelset(&list->signal, sig);
489                 info->si_signo = sig;
490                 info->si_errno = 0;
491                 info->si_code = 0;
492                 info->si_pid = 0;
493                 info->si_uid = 0;
494         }
495         return 1;
496 }
497
498 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
499                         siginfo_t *info)
500 {
501         int sig = 0;
502
503         sig = next_signal(pending, mask);
504         if (sig) {
505                 if (current->notifier) {
506                         if (sigismember(current->notifier_mask, sig)) {
507                                 if (!(current->notifier)(current->notifier_data)) {
508                                         clear_thread_flag(TIF_SIGPENDING);
509                                         return 0;
510                                 }
511                         }
512                 }
513
514                 if (!collect_signal(sig, pending, info))
515                         sig = 0;
516                                 
517         }
518         recalc_sigpending();
519
520         return sig;
521 }
522
523 /*
524  * Dequeue a signal and return the element to the caller, which is 
525  * expected to free it.
526  *
527  * All callers have to hold the siglock.
528  */
529 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
530 {
531         int signr = __dequeue_signal(&tsk->pending, mask, info);
532         if (!signr)
533                 signr = __dequeue_signal(&tsk->signal->shared_pending,
534                                          mask, info);
535         if ( signr &&
536              ((info->si_code & __SI_MASK) == __SI_TIMER) &&
537              info->si_sys_private){
538                 do_schedule_next_timer(info);
539         }
540         return signr;
541 }
542
543 /*
544  * Tell a process that it has a new active signal..
545  *
546  * NOTE! we rely on the previous spin_lock to
547  * lock interrupts for us! We can only be called with
548  * "siglock" held, and the local interrupt must
549  * have been disabled when that got acquired!
550  *
551  * No need to set need_resched since signal event passing
552  * goes through ->blocked
553  */
554 void signal_wake_up(struct task_struct *t, int resume)
555 {
556         unsigned int mask;
557
558         set_tsk_thread_flag(t, TIF_SIGPENDING);
559
560         /*
561          * If resume is set, we want to wake it up in the TASK_STOPPED case.
562          * We don't check for TASK_STOPPED because there is a race with it
563          * executing another processor and just now entering stopped state.
564          * By calling wake_up_process any time resume is set, we ensure
565          * the process will wake up and handle its stop or death signal.
566          */
567         mask = TASK_INTERRUPTIBLE;
568         if (resume)
569                 mask |= TASK_STOPPED;
570         if (!wake_up_state(t, mask))
571                 kick_process(t);
572 }
573
574 /*
575  * Remove signals in mask from the pending set and queue.
576  * Returns 1 if any signals were found.
577  *
578  * All callers must be holding the siglock.
579  */
580 static int rm_from_queue(unsigned long mask, struct sigpending *s)
581 {
582         struct sigqueue *q, *n;
583
584         if (!sigtestsetmask(&s->signal, mask))
585                 return 0;
586
587         sigdelsetmask(&s->signal, mask);
588         list_for_each_entry_safe(q, n, &s->list, list) {
589                 if (q->info.si_signo < SIGRTMIN &&
590                     (mask & sigmask(q->info.si_signo))) {
591                         list_del_init(&q->list);
592                         __sigqueue_free(q);
593                 }
594         }
595         return 1;
596 }
597
598 /*
599  * Bad permissions for sending the signal
600  */
601 static int check_kill_permission(int sig, struct siginfo *info,
602                                  struct task_struct *t)
603 {
604         int error = -EINVAL;
605         if (sig < 0 || sig > _NSIG)
606                 return error;
607         error = -EPERM;
608         if ((!info || ((unsigned long)info != 1 &&
609                         (unsigned long)info != 2 && SI_FROMUSER(info)))
610             && ((sig != SIGCONT) ||
611                 (current->signal->session != t->signal->session))
612             && (current->euid ^ t->suid) && (current->euid ^ t->uid)
613             && (current->uid ^ t->suid) && (current->uid ^ t->uid)
614             && !capable(CAP_KILL))
615                 return error;
616         return security_task_kill(t, info, sig);
617 }
618
619 /* forward decl */
620 static void do_notify_parent_cldstop(struct task_struct *tsk,
621                                      struct task_struct *parent);
622
623 /*
624  * Handle magic process-wide effects of stop/continue signals.
625  * Unlike the signal actions, these happen immediately at signal-generation
626  * time regardless of blocking, ignoring, or handling.  This does the
627  * actual continuing for SIGCONT, but not the actual stopping for stop
628  * signals.  The process stop is done as a signal action for SIG_DFL.
629  */
630 static void handle_stop_signal(int sig, struct task_struct *p)
631 {
632         struct task_struct *t;
633
634         if (sig_kernel_stop(sig)) {
635                 /*
636                  * This is a stop signal.  Remove SIGCONT from all queues.
637                  */
638                 rm_from_queue(sigmask(SIGCONT), &p->signal->shared_pending);
639                 t = p;
640                 do {
641                         rm_from_queue(sigmask(SIGCONT), &t->pending);
642                         t = next_thread(t);
643                 } while (t != p);
644         } else if (sig == SIGCONT) {
645                 /*
646                  * Remove all stop signals from all queues,
647                  * and wake all threads.
648                  */
649                 if (unlikely(p->signal->group_stop_count > 0)) {
650                         /*
651                          * There was a group stop in progress.  We'll
652                          * pretend it finished before we got here.  We are
653                          * obliged to report it to the parent: if the
654                          * SIGSTOP happened "after" this SIGCONT, then it
655                          * would have cleared this pending SIGCONT.  If it
656                          * happened "before" this SIGCONT, then the parent
657                          * got the SIGCHLD about the stop finishing before
658                          * the continue happened.  We do the notification
659                          * now, and it's as if the stop had finished and
660                          * the SIGCHLD was pending on entry to this kill.
661                          */
662                         p->signal->group_stop_count = 0;
663                         if (p->ptrace & PT_PTRACED)
664                                 do_notify_parent_cldstop(p, p->parent);
665                         else
666                                 do_notify_parent_cldstop(
667                                         p->group_leader,
668                                         p->group_leader->real_parent);
669                 }
670                 rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
671                 t = p;
672                 do {
673                         unsigned int state;
674                         rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
675                         
676                         /*
677                          * If there is a handler for SIGCONT, we must make
678                          * sure that no thread returns to user mode before
679                          * we post the signal, in case it was the only
680                          * thread eligible to run the signal handler--then
681                          * it must not do anything between resuming and
682                          * running the handler.  With the TIF_SIGPENDING
683                          * flag set, the thread will pause and acquire the
684                          * siglock that we hold now and until we've queued
685                          * the pending signal. 
686                          *
687                          * Wake up the stopped thread _after_ setting
688                          * TIF_SIGPENDING
689                          */
690                         state = TASK_STOPPED;
691                         if (sig_user_defined(t, SIGCONT) && !sigismember(&t->blocked, SIGCONT)) {
692                                 set_tsk_thread_flag(t, TIF_SIGPENDING);
693                                 state |= TASK_INTERRUPTIBLE;
694                         }
695                         wake_up_state(t, state);
696
697                         t = next_thread(t);
698                 } while (t != p);
699         }
700 }
701
702 static int send_signal(int sig, struct siginfo *info, struct sigpending *signals)
703 {
704         struct sigqueue * q = NULL;
705         int ret = 0;
706
707         /*
708          * fast-pathed signals for kernel-internal things like SIGSTOP
709          * or SIGKILL.
710          */
711         if ((unsigned long)info == 2)
712                 goto out_set;
713
714         /* Real-time signals must be queued if sent by sigqueue, or
715            some other real-time mechanism.  It is implementation
716            defined whether kill() does so.  We attempt to do so, on
717            the principle of least surprise, but since kill is not
718            allowed to fail with EAGAIN when low on memory we just
719            make sure at least one signal gets delivered and don't
720            pass on the info struct.  */
721
722         if (atomic_read(&nr_queued_signals) < max_queued_signals)
723                 q = kmem_cache_alloc(sigqueue_cachep, GFP_ATOMIC);
724
725         if (q) {
726                 atomic_inc(&nr_queued_signals);
727                 q->flags = 0;
728                 list_add_tail(&q->list, &signals->list);
729                 switch ((unsigned long) info) {
730                 case 0:
731                         q->info.si_signo = sig;
732                         q->info.si_errno = 0;
733                         q->info.si_code = SI_USER;
734                         q->info.si_pid = current->pid;
735                         q->info.si_uid = current->uid;
736                         break;
737                 case 1:
738                         q->info.si_signo = sig;
739                         q->info.si_errno = 0;
740                         q->info.si_code = SI_KERNEL;
741                         q->info.si_pid = 0;
742                         q->info.si_uid = 0;
743                         break;
744                 default:
745                         copy_siginfo(&q->info, info);
746                         break;
747                 }
748         } else {
749                 if (sig >= SIGRTMIN && info && (unsigned long)info != 1
750                    && info->si_code != SI_USER)
751                 /*
752                  * Queue overflow, abort.  We may abort if the signal was rt
753                  * and sent by user using something other than kill().
754                  */
755                         return -EAGAIN;
756                 if (((unsigned long)info > 1) && (info->si_code == SI_TIMER))
757                         /*
758                          * Set up a return to indicate that we dropped 
759                          * the signal.
760                          */
761                         ret = info->si_sys_private;
762         }
763
764 out_set:
765         sigaddset(&signals->signal, sig);
766         return ret;
767 }
768
769 #define LEGACY_QUEUE(sigptr, sig) \
770         (((sig) < SIGRTMIN) && sigismember(&(sigptr)->signal, (sig)))
771
772
773 static int
774 specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
775 {
776         int ret = 0;
777
778         if (!irqs_disabled())
779                 BUG();
780 #ifdef CONFIG_SMP
781         if (!spin_is_locked(&t->sighand->siglock))
782                 BUG();
783 #endif
784
785         if (((unsigned long)info > 2) && (info->si_code == SI_TIMER))
786                 /*
787                  * Set up a return to indicate that we dropped the signal.
788                  */
789                 ret = info->si_sys_private;
790
791         /* Short-circuit ignored signals.  */
792         if (sig_ignored(t, sig))
793                 goto out;
794
795         /* Support queueing exactly one non-rt signal, so that we
796            can get more detailed information about the cause of
797            the signal. */
798         if (LEGACY_QUEUE(&t->pending, sig))
799                 goto out;
800
801         ret = send_signal(sig, info, &t->pending);
802         if (!ret && !sigismember(&t->blocked, sig))
803                 signal_wake_up(t, sig == SIGKILL);
804 out:
805         return ret;
806 }
807
808 /*
809  * Force a signal that the process can't ignore: if necessary
810  * we unblock the signal and change any SIG_IGN to SIG_DFL.
811  */
812
813 int
814 force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
815 {
816         unsigned long int flags;
817         int ret;
818
819         spin_lock_irqsave(&t->sighand->siglock, flags);
820         if (sigismember(&t->blocked, sig) || t->sighand->action[sig-1].sa.sa_handler == SIG_IGN) {
821                 t->sighand->action[sig-1].sa.sa_handler = SIG_DFL;
822                 sigdelset(&t->blocked, sig);
823                 recalc_sigpending_tsk(t);
824         }
825         ret = specific_send_sig_info(sig, info, t);
826         spin_unlock_irqrestore(&t->sighand->siglock, flags);
827
828         return ret;
829 }
830
831 void
832 force_sig_specific(int sig, struct task_struct *t)
833 {
834         unsigned long int flags;
835
836         spin_lock_irqsave(&t->sighand->siglock, flags);
837         if (t->sighand->action[sig-1].sa.sa_handler == SIG_IGN)
838                 t->sighand->action[sig-1].sa.sa_handler = SIG_DFL;
839         sigdelset(&t->blocked, sig);
840         recalc_sigpending_tsk(t);
841         specific_send_sig_info(sig, (void *)2, t);
842         spin_unlock_irqrestore(&t->sighand->siglock, flags);
843 }
844
845 /*
846  * Test if P wants to take SIG.  After we've checked all threads with this,
847  * it's equivalent to finding no threads not blocking SIG.  Any threads not
848  * blocking SIG were ruled out because they are not running and already
849  * have pending signals.  Such threads will dequeue from the shared queue
850  * as soon as they're available, so putting the signal on the shared queue
851  * will be equivalent to sending it to one such thread.
852  */
853 #define wants_signal(sig, p, mask)                      \
854         (!sigismember(&(p)->blocked, sig)               \
855          && !((p)->state & mask)                        \
856          && !((p)->flags & PF_EXITING)                  \
857          && (task_curr(p) || !signal_pending(p)))
858
859
860 static void
861 __group_complete_signal(int sig, struct task_struct *p, unsigned int mask)
862 {
863         struct task_struct *t;
864
865         /*
866          * Now find a thread we can wake up to take the signal off the queue.
867          *
868          * If the main thread wants the signal, it gets first crack.
869          * Probably the least surprising to the average bear.
870          */
871         if (wants_signal(sig, p, mask))
872                 t = p;
873         else if (thread_group_empty(p))
874                 /*
875                  * There is just one thread and it does not need to be woken.
876                  * It will dequeue unblocked signals before it runs again.
877                  */
878                 return;
879         else {
880                 /*
881                  * Otherwise try to find a suitable thread.
882                  */
883                 t = p->signal->curr_target;
884                 if (t == NULL)
885                         /* restart balancing at this thread */
886                         t = p->signal->curr_target = p;
887                 BUG_ON(t->tgid != p->tgid);
888
889                 while (!wants_signal(sig, t, mask)) {
890                         t = next_thread(t);
891                         if (t == p->signal->curr_target)
892                                 /*
893                                  * No thread needs to be woken.
894                                  * Any eligible threads will see
895                                  * the signal in the queue soon.
896                                  */
897                                 return;
898                 }
899                 p->signal->curr_target = t;
900         }
901
902         /*
903          * Found a killable thread.  If the signal will be fatal,
904          * then start taking the whole group down immediately.
905          */
906         if (sig_fatal(p, sig) && !p->signal->group_exit &&
907             !sigismember(&t->real_blocked, sig) &&
908             (sig == SIGKILL || !(t->ptrace & PT_PTRACED))) {
909                 /*
910                  * This signal will be fatal to the whole group.
911                  */
912                 if (!sig_kernel_coredump(sig)) {
913                         /*
914                          * Start a group exit and wake everybody up.
915                          * This way we don't have other threads
916                          * running and doing things after a slower
917                          * thread has the fatal signal pending.
918                          */
919                         p->signal->group_exit = 1;
920                         p->signal->group_exit_code = sig;
921                         p->signal->group_stop_count = 0;
922                         t = p;
923                         do {
924                                 sigaddset(&t->pending.signal, SIGKILL);
925                                 signal_wake_up(t, 1);
926                                 t = next_thread(t);
927                         } while (t != p);
928                         return;
929                 }
930
931                 /*
932                  * There will be a core dump.  We make all threads other
933                  * than the chosen one go into a group stop so that nothing
934                  * happens until it gets scheduled, takes the signal off
935                  * the shared queue, and does the core dump.  This is a
936                  * little more complicated than strictly necessary, but it
937                  * keeps the signal state that winds up in the core dump
938                  * unchanged from the death state, e.g. which thread had
939                  * the core-dump signal unblocked.
940                  */
941                 rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
942                 rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
943                 p->signal->group_stop_count = 0;
944                 p->signal->group_exit_task = t;
945                 t = p;
946                 do {
947                         p->signal->group_stop_count++;
948                         signal_wake_up(t, 0);
949                         t = next_thread(t);
950                 } while (t != p);
951                 wake_up_process(p->signal->group_exit_task);
952                 return;
953         }
954
955         /*
956          * The signal is already in the shared-pending queue.
957          * Tell the chosen thread to wake up and dequeue it.
958          */
959         signal_wake_up(t, sig == SIGKILL);
960         return;
961 }
962
963 static int
964 __group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
965 {
966         unsigned int mask;
967         int ret = 0;
968
969 #ifdef CONFIG_SMP
970         if (!spin_is_locked(&p->sighand->siglock))
971                 BUG();
972 #endif
973         handle_stop_signal(sig, p);
974
975         if (((unsigned long)info > 2) && (info->si_code == SI_TIMER))
976                 /*
977                  * Set up a return to indicate that we dropped the signal.
978                  */
979                 ret = info->si_sys_private;
980
981         /* Short-circuit ignored signals.  */
982         if (sig_ignored(p, sig))
983                 return ret;
984
985         if (LEGACY_QUEUE(&p->signal->shared_pending, sig))
986                 /* This is a non-RT signal and we already have one queued.  */
987                 return ret;
988
989         /*
990          * Don't bother zombies and stopped tasks (but
991          * SIGKILL will punch through stopped state)
992          */
993         mask = TASK_DEAD | TASK_ZOMBIE;
994         if (sig != SIGKILL)
995                 mask |= TASK_STOPPED;
996
997         /*
998          * Put this signal on the shared-pending queue, or fail with EAGAIN.
999          * We always use the shared queue for process-wide signals,
1000          * to avoid several races.
1001          */
1002         ret = send_signal(sig, info, &p->signal->shared_pending);
1003         if (unlikely(ret))
1004                 return ret;
1005
1006         __group_complete_signal(sig, p, mask);
1007         return 0;
1008 }
1009
1010 /*
1011  * Nuke all other threads in the group.
1012  */
1013 void zap_other_threads(struct task_struct *p)
1014 {
1015         struct task_struct *t;
1016
1017         p->signal->group_stop_count = 0;
1018
1019         if (thread_group_empty(p))
1020                 return;
1021
1022         for (t = next_thread(p); t != p; t = next_thread(t)) {
1023                 /*
1024                  * Don't bother with already dead threads
1025                  */
1026                 if (t->state & (TASK_ZOMBIE|TASK_DEAD))
1027                         continue;
1028
1029                 /*
1030                  * We don't want to notify the parent, since we are
1031                  * killed as part of a thread group due to another
1032                  * thread doing an execve() or similar. So set the
1033                  * exit signal to -1 to allow immediate reaping of
1034                  * the process.  But don't detach the thread group
1035                  * leader.
1036                  */
1037                 if (t != p->group_leader)
1038                         t->exit_signal = -1;
1039
1040                 sigaddset(&t->pending.signal, SIGKILL);
1041                 rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
1042                 signal_wake_up(t, 1);
1043         }
1044 }
1045
1046 /*
1047  * Must be called with the tasklist_lock held for reading!
1048  */
1049 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1050 {
1051         unsigned long flags;
1052         int ret;
1053
1054         if (!vx_check(vx_task_xid(p), VX_ADMIN|VX_WATCH|VX_IDENT))
1055                 return -ESRCH;
1056
1057         ret = check_kill_permission(sig, info, p);
1058         if (!ret && sig && p->sighand) {
1059                 spin_lock_irqsave(&p->sighand->siglock, flags);
1060                 ret = __group_send_sig_info(sig, info, p);
1061                 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1062         }
1063
1064         return ret;
1065 }
1066
1067 /*
1068  * kill_pg_info() sends a signal to a process group: this is what the tty
1069  * control characters do (^C, ^Z etc)
1070  */
1071
1072 int __kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)
1073 {
1074         struct task_struct *p;
1075         struct list_head *l;
1076         struct pid *pid;
1077         int retval;
1078         int found;
1079
1080         if (pgrp <= 0)
1081                 return -EINVAL;
1082
1083         found = 0;
1084         retval = 0;
1085         for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
1086                 int err;
1087
1088                 found = 1;
1089                 err = group_send_sig_info(sig, info, p);
1090                 if (!retval)
1091                         retval = err;
1092         }
1093         return found ? retval : -ESRCH;
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          * We need the tasklist lock even for the specific
1205          * thread case (when we don't need to follow the group
1206          * lists) in order to avoid races with "p->sighand"
1207          * going away or changing from under us.
1208          */
1209         read_lock(&tasklist_lock);  
1210         spin_lock_irqsave(&p->sighand->siglock, flags);
1211         ret = specific_send_sig_info(sig, info, p);
1212         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1213         read_unlock(&tasklist_lock);
1214         return ret;
1215 }
1216
1217 int
1218 send_sig(int sig, struct task_struct *p, int priv)
1219 {
1220         return send_sig_info(sig, (void*)(long)(priv != 0), p);
1221 }
1222
1223 /*
1224  * This is the entry point for "process-wide" signals.
1225  * They will go to an appropriate thread in the thread group.
1226  */
1227 int
1228 send_group_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1229 {
1230         int ret;
1231         read_lock(&tasklist_lock);
1232         ret = group_send_sig_info(sig, info, p);
1233         read_unlock(&tasklist_lock);
1234         return ret;
1235 }
1236
1237 void
1238 force_sig(int sig, struct task_struct *p)
1239 {
1240         force_sig_info(sig, (void*)1L, p);
1241 }
1242
1243 int
1244 kill_pg(pid_t pgrp, int sig, int priv)
1245 {
1246         return kill_pg_info(sig, (void *)(long)(priv != 0), pgrp);
1247 }
1248
1249 int
1250 kill_sl(pid_t sess, int sig, int priv)
1251 {
1252         return kill_sl_info(sig, (void *)(long)(priv != 0), sess);
1253 }
1254
1255 int
1256 kill_proc(pid_t pid, int sig, int priv)
1257 {
1258         return kill_proc_info(sig, (void *)(long)(priv != 0), pid);
1259 }
1260
1261 /*
1262  * These functions support sending signals using preallocated sigqueue
1263  * structures.  This is needed "because realtime applications cannot
1264  * afford to lose notifications of asynchronous events, like timer
1265  * expirations or I/O completions".  In the case of Posix Timers 
1266  * we allocate the sigqueue structure from the timer_create.  If this
1267  * allocation fails we are able to report the failure to the application
1268  * with an EAGAIN error.
1269  */
1270  
1271 struct sigqueue *sigqueue_alloc(void)
1272 {
1273         struct sigqueue *q;
1274
1275         if ((q = __sigqueue_alloc()))
1276                 q->flags |= SIGQUEUE_PREALLOC;
1277         return(q);
1278 }
1279
1280 void sigqueue_free(struct sigqueue *q)
1281 {
1282         unsigned long flags;
1283         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1284         /*
1285          * If the signal is still pending remove it from the
1286          * pending queue.
1287          */
1288         if (unlikely(!list_empty(&q->list))) {
1289                 read_lock(&tasklist_lock);  
1290                 spin_lock_irqsave(q->lock, flags);
1291                 if (!list_empty(&q->list))
1292                         list_del_init(&q->list);
1293                 spin_unlock_irqrestore(q->lock, flags);
1294                 read_unlock(&tasklist_lock);
1295         }
1296         q->flags &= ~SIGQUEUE_PREALLOC;
1297         __sigqueue_free(q);
1298 }
1299
1300 int
1301 send_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
1302 {
1303         unsigned long flags;
1304         int ret = 0;
1305
1306         /*
1307          * We need the tasklist lock even for the specific
1308          * thread case (when we don't need to follow the group
1309          * lists) in order to avoid races with "p->sighand"
1310          * going away or changing from under us.
1311          */
1312         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1313         read_lock(&tasklist_lock);  
1314         spin_lock_irqsave(&p->sighand->siglock, flags);
1315         
1316         if (unlikely(!list_empty(&q->list))) {
1317                 /*
1318                  * If an SI_TIMER entry is already queue just increment
1319                  * the overrun count.
1320                  */
1321                 if (q->info.si_code != SI_TIMER)
1322                         BUG();
1323                 q->info.si_overrun++;
1324                 goto out;
1325         } 
1326         /* Short-circuit ignored signals.  */
1327         if (sig_ignored(p, sig)) {
1328                 ret = 1;
1329                 goto out;
1330         }
1331
1332         q->lock = &p->sighand->siglock;
1333         list_add_tail(&q->list, &p->pending.list);
1334         sigaddset(&p->pending.signal, sig);
1335         if (!sigismember(&p->blocked, sig))
1336                 signal_wake_up(p, sig == SIGKILL);
1337
1338 out:
1339         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1340         read_unlock(&tasklist_lock);
1341         return(ret);
1342 }
1343
1344 int
1345 send_group_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
1346 {
1347         unsigned long flags;
1348         unsigned int mask;
1349         int ret = 0;
1350
1351         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1352         read_lock(&tasklist_lock);
1353         spin_lock_irqsave(&p->sighand->siglock, flags);
1354         handle_stop_signal(sig, p);
1355
1356         /* Short-circuit ignored signals.  */
1357         if (sig_ignored(p, sig)) {
1358                 ret = 1;
1359                 goto out;
1360         }
1361
1362         if (unlikely(!list_empty(&q->list))) {
1363                 /*
1364                  * If an SI_TIMER entry is already queue just increment
1365                  * the overrun count.  Other uses should not try to
1366                  * send the signal multiple times.
1367                  */
1368                 if (q->info.si_code != SI_TIMER)
1369                         BUG();
1370                 q->info.si_overrun++;
1371                 goto out;
1372         } 
1373         /*
1374          * Don't bother zombies and stopped tasks (but
1375          * SIGKILL will punch through stopped state)
1376          */
1377         mask = TASK_DEAD | TASK_ZOMBIE;
1378         if (sig != SIGKILL)
1379                 mask |= TASK_STOPPED;
1380
1381         /*
1382          * Put this signal on the shared-pending queue.
1383          * We always use the shared queue for process-wide signals,
1384          * to avoid several races.
1385          */
1386         q->lock = &p->sighand->siglock;
1387         list_add_tail(&q->list, &p->signal->shared_pending.list);
1388         sigaddset(&p->signal->shared_pending.signal, sig);
1389
1390         __group_complete_signal(sig, p, mask);
1391 out:
1392         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1393         read_unlock(&tasklist_lock);
1394         return(ret);
1395 }
1396
1397 /*
1398  * Joy. Or not. Pthread wants us to wake up every thread
1399  * in our parent group.
1400  */
1401 static void __wake_up_parent(struct task_struct *p,
1402                                     struct task_struct *parent)
1403 {
1404         struct task_struct *tsk = parent;
1405
1406         /*
1407          * Fortunately this is not necessary for thread groups:
1408          */
1409         if (p->tgid == tsk->tgid) {
1410                 wake_up_interruptible_sync(&tsk->wait_chldexit);
1411                 return;
1412         }
1413
1414         do {
1415                 wake_up_interruptible_sync(&tsk->wait_chldexit);
1416                 tsk = next_thread(tsk);
1417                 if (tsk->signal != parent->signal)
1418                         BUG();
1419         } while (tsk != parent);
1420 }
1421
1422 /*
1423  * Let a parent know about a status change of a child.
1424  */
1425
1426 void do_notify_parent(struct task_struct *tsk, int sig)
1427 {
1428         struct siginfo info;
1429         unsigned long flags;
1430         int why, status;
1431         struct sighand_struct *psig;
1432
1433         if (sig == -1)
1434                 BUG();
1435
1436         BUG_ON(tsk->group_leader != tsk && tsk->group_leader->state != TASK_ZOMBIE && !tsk->ptrace);
1437         BUG_ON(tsk->group_leader == tsk && !thread_group_empty(tsk) && !tsk->ptrace);
1438
1439         info.si_signo = sig;
1440         info.si_errno = 0;
1441         info.si_pid = tsk->pid;
1442         info.si_uid = tsk->uid;
1443
1444         /* FIXME: find out whether or not this is supposed to be c*time. */
1445         info.si_utime = tsk->utime;
1446         info.si_stime = tsk->stime;
1447
1448         status = tsk->exit_code & 0x7f;
1449         why = SI_KERNEL;        /* shouldn't happen */
1450         switch (tsk->state) {
1451         case TASK_STOPPED:
1452                 /* FIXME -- can we deduce CLD_TRAPPED or CLD_CONTINUED? */
1453                 if (tsk->ptrace & PT_PTRACED)
1454                         why = CLD_TRAPPED;
1455                 else
1456                         why = CLD_STOPPED;
1457                 break;
1458
1459         default:
1460                 if (tsk->exit_code & 0x80)
1461                         why = CLD_DUMPED;
1462                 else if (tsk->exit_code & 0x7f)
1463                         why = CLD_KILLED;
1464                 else {
1465                         why = CLD_EXITED;
1466                         status = tsk->exit_code >> 8;
1467                 }
1468                 break;
1469         }
1470         info.si_code = why;
1471         info.si_status = status;
1472
1473         psig = tsk->parent->sighand;
1474         spin_lock_irqsave(&psig->siglock, flags);
1475         if (sig == SIGCHLD && tsk->state != TASK_STOPPED &&
1476             (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1477              (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1478                 /*
1479                  * We are exiting and our parent doesn't care.  POSIX.1
1480                  * defines special semantics for setting SIGCHLD to SIG_IGN
1481                  * or setting the SA_NOCLDWAIT flag: we should be reaped
1482                  * automatically and not left for our parent's wait4 call.
1483                  * Rather than having the parent do it as a magic kind of
1484                  * signal handler, we just set this to tell do_exit that we
1485                  * can be cleaned up without becoming a zombie.  Note that
1486                  * we still call __wake_up_parent in this case, because a
1487                  * blocked sys_wait4 might now return -ECHILD.
1488                  *
1489                  * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1490                  * is implementation-defined: we do (if you don't want
1491                  * it, just use SIG_IGN instead).
1492                  */
1493                 tsk->exit_signal = -1;
1494                 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1495                         sig = 0;
1496         }
1497         if (sig > 0 && sig <= _NSIG)
1498                 __group_send_sig_info(sig, &info, tsk->parent);
1499         __wake_up_parent(tsk, tsk->parent);
1500         spin_unlock_irqrestore(&psig->siglock, flags);
1501 }
1502
1503
1504 /*
1505  * We need the tasklist lock because it's the only
1506  * thing that protects out "parent" pointer.
1507  *
1508  * exit.c calls "do_notify_parent()" directly, because
1509  * it already has the tasklist lock.
1510  */
1511 void
1512 notify_parent(struct task_struct *tsk, int sig)
1513 {
1514         if (sig != -1) {
1515                 read_lock(&tasklist_lock);
1516                 do_notify_parent(tsk, sig);
1517                 read_unlock(&tasklist_lock);
1518         }
1519 }
1520
1521 static void
1522 do_notify_parent_cldstop(struct task_struct *tsk, struct task_struct *parent)
1523 {
1524         struct siginfo info;
1525         unsigned long flags;
1526         struct sighand_struct *sighand;
1527
1528         info.si_signo = SIGCHLD;
1529         info.si_errno = 0;
1530         info.si_pid = tsk->pid;
1531         info.si_uid = tsk->uid;
1532
1533         /* FIXME: find out whether or not this is supposed to be c*time. */
1534         info.si_utime = tsk->utime;
1535         info.si_stime = tsk->stime;
1536
1537         info.si_status = tsk->exit_code & 0x7f;
1538         info.si_code = CLD_STOPPED;
1539
1540         sighand = parent->sighand;
1541         spin_lock_irqsave(&sighand->siglock, flags);
1542         if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1543             !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1544                 __group_send_sig_info(SIGCHLD, &info, parent);
1545         /*
1546          * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1547          */
1548         __wake_up_parent(tsk, parent);
1549         spin_unlock_irqrestore(&sighand->siglock, flags);
1550 }
1551
1552
1553 #ifndef HAVE_ARCH_GET_SIGNAL_TO_DELIVER
1554
1555 static void
1556 finish_stop(int stop_count)
1557 {
1558         /*
1559          * If there are no other threads in the group, or if there is
1560          * a group stop in progress and we are the last to stop,
1561          * report to the parent.  When ptraced, every thread reports itself.
1562          */
1563         if (stop_count < 0 || (current->ptrace & PT_PTRACED)) {
1564                 read_lock(&tasklist_lock);
1565                 do_notify_parent_cldstop(current, current->parent);
1566                 read_unlock(&tasklist_lock);
1567         }
1568         else if (stop_count == 0) {
1569                 read_lock(&tasklist_lock);
1570                 do_notify_parent_cldstop(current->group_leader,
1571                                          current->group_leader->real_parent);
1572                 read_unlock(&tasklist_lock);
1573         }
1574
1575         schedule();
1576         /*
1577          * Now we don't run again until continued.
1578          */
1579         current->exit_code = 0;
1580 }
1581
1582 /*
1583  * This performs the stopping for SIGSTOP and other stop signals.
1584  * We have to stop all threads in the thread group.
1585  */
1586 static void
1587 do_signal_stop(int signr)
1588 {
1589         struct signal_struct *sig = current->signal;
1590         struct sighand_struct *sighand = current->sighand;
1591         int stop_count = -1;
1592
1593         /* spin_lock_irq(&sighand->siglock) is now done in caller */
1594
1595         if (sig->group_stop_count > 0) {
1596                 /*
1597                  * There is a group stop in progress.  We don't need to
1598                  * start another one.
1599                  */
1600                 signr = sig->group_exit_code;
1601                 stop_count = --sig->group_stop_count;
1602                 current->exit_code = signr;
1603                 set_current_state(TASK_STOPPED);
1604                 spin_unlock_irq(&sighand->siglock);
1605         }
1606         else if (thread_group_empty(current)) {
1607                 /*
1608                  * Lock must be held through transition to stopped state.
1609                  */
1610                 current->exit_code = signr;
1611                 set_current_state(TASK_STOPPED);
1612                 spin_unlock_irq(&sighand->siglock);
1613         }
1614         else {
1615                 /*
1616                  * There is no group stop already in progress.
1617                  * We must initiate one now, but that requires
1618                  * dropping siglock to get both the tasklist lock
1619                  * and siglock again in the proper order.  Note that
1620                  * this allows an intervening SIGCONT to be posted.
1621                  * We need to check for that and bail out if necessary.
1622                  */
1623                 struct task_struct *t;
1624
1625                 spin_unlock_irq(&sighand->siglock);
1626
1627                 /* signals can be posted during this window */
1628
1629                 read_lock(&tasklist_lock);
1630                 spin_lock_irq(&sighand->siglock);
1631
1632                 if (unlikely(sig->group_exit)) {
1633                         /*
1634                          * There is a group exit in progress now.
1635                          * We'll just ignore the stop and process the
1636                          * associated fatal signal.
1637                          */
1638                         spin_unlock_irq(&sighand->siglock);
1639                         read_unlock(&tasklist_lock);
1640                         return;
1641                 }
1642
1643                 if (unlikely(sig_avoid_stop_race())) {
1644                         /*
1645                          * Either a SIGCONT or a SIGKILL signal was
1646                          * posted in the siglock-not-held window.
1647                          */
1648                         spin_unlock_irq(&sighand->siglock);
1649                         read_unlock(&tasklist_lock);
1650                         return;
1651                 }
1652
1653                 if (sig->group_stop_count == 0) {
1654                         sig->group_exit_code = signr;
1655                         stop_count = 0;
1656                         for (t = next_thread(current); t != current;
1657                              t = next_thread(t))
1658                                 /*
1659                                  * Setting state to TASK_STOPPED for a group
1660                                  * stop is always done with the siglock held,
1661                                  * so this check has no races.
1662                                  */
1663                                 if (t->state < TASK_STOPPED) {
1664                                         stop_count++;
1665                                         signal_wake_up(t, 0);
1666                                 }
1667                         sig->group_stop_count = stop_count;
1668                 }
1669                 else {
1670                         /* A race with another thread while unlocked.  */
1671                         signr = sig->group_exit_code;
1672                         stop_count = --sig->group_stop_count;
1673                 }
1674
1675                 current->exit_code = signr;
1676                 set_current_state(TASK_STOPPED);
1677
1678                 spin_unlock_irq(&sighand->siglock);
1679                 read_unlock(&tasklist_lock);
1680         }
1681
1682         finish_stop(stop_count);
1683 }
1684
1685 /*
1686  * Do appropriate magic when group_stop_count > 0.
1687  * We return nonzero if we stopped, after releasing the siglock.
1688  * We return zero if we still hold the siglock and should look
1689  * for another signal without checking group_stop_count again.
1690  */
1691 static inline int handle_group_stop(void)
1692 {
1693         int stop_count;
1694
1695         if (current->signal->group_exit_task == current) {
1696                 /*
1697                  * Group stop is so we can do a core dump,
1698                  * We are the initiating thread, so get on with it.
1699                  */
1700                 current->signal->group_exit_task = NULL;
1701                 return 0;
1702         }
1703
1704         if (current->signal->group_exit)
1705                 /*
1706                  * Group stop is so another thread can do a core dump,
1707                  * or else we are racing against a death signal.
1708                  * Just punt the stop so we can get the next signal.
1709                  */
1710                 return 0;
1711
1712         /*
1713          * There is a group stop in progress.  We stop
1714          * without any associated signal being in our queue.
1715          */
1716         stop_count = --current->signal->group_stop_count;
1717         current->exit_code = current->signal->group_exit_code;
1718         set_current_state(TASK_STOPPED);
1719         spin_unlock_irq(&current->sighand->siglock);
1720         finish_stop(stop_count);
1721         return 1;
1722 }
1723
1724 int get_signal_to_deliver(siginfo_t *info, struct pt_regs *regs, void *cookie)
1725 {
1726         sigset_t *mask = &current->blocked;
1727         int signr = 0;
1728
1729 relock:
1730         spin_lock_irq(&current->sighand->siglock);
1731         for (;;) {
1732                 struct k_sigaction *ka;
1733
1734                 if (unlikely(current->signal->group_stop_count > 0) &&
1735                     handle_group_stop())
1736                         goto relock;
1737
1738                 signr = dequeue_signal(current, mask, info);
1739
1740                 if (!signr)
1741                         break; /* will return 0 */
1742
1743                 if ((current->ptrace & PT_PTRACED) && signr != SIGKILL) {
1744                         ptrace_signal_deliver(regs, cookie);
1745
1746                         /*
1747                          * If there is a group stop in progress,
1748                          * we must participate in the bookkeeping.
1749                          */
1750                         if (current->signal->group_stop_count > 0)
1751                                 --current->signal->group_stop_count;
1752
1753                         /* Let the debugger run.  */
1754                         current->exit_code = signr;
1755                         current->last_siginfo = info;
1756                         set_current_state(TASK_STOPPED);
1757                         spin_unlock_irq(&current->sighand->siglock);
1758                         notify_parent(current, SIGCHLD);
1759                         schedule();
1760
1761                         current->last_siginfo = NULL;
1762
1763                         /* We're back.  Did the debugger cancel the sig?  */
1764                         spin_lock_irq(&current->sighand->siglock);
1765                         signr = current->exit_code;
1766                         if (signr == 0)
1767                                 continue;
1768
1769                         current->exit_code = 0;
1770
1771                         /* Update the siginfo structure if the signal has
1772                            changed.  If the debugger wanted something
1773                            specific in the siginfo structure then it should
1774                            have updated *info via PTRACE_SETSIGINFO.  */
1775                         if (signr != info->si_signo) {
1776                                 info->si_signo = signr;
1777                                 info->si_errno = 0;
1778                                 info->si_code = SI_USER;
1779                                 info->si_pid = current->parent->pid;
1780                                 info->si_uid = current->parent->uid;
1781                         }
1782
1783                         /* If the (new) signal is now blocked, requeue it.  */
1784                         if (sigismember(&current->blocked, signr)) {
1785                                 specific_send_sig_info(signr, info, current);
1786                                 continue;
1787                         }
1788                 }
1789
1790                 ka = &current->sighand->action[signr-1];
1791                 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
1792                         continue;
1793                 if (ka->sa.sa_handler != SIG_DFL) /* Run the handler.  */
1794                         break; /* will return non-zero "signr" value */
1795
1796                 /*
1797                  * Now we are doing the default action for this signal.
1798                  */
1799                 if (sig_kernel_ignore(signr)) /* Default is nothing. */
1800                         continue;
1801
1802                 /* Init gets no signals it doesn't want.  */
1803                 if (current->pid == 1)
1804                         continue;
1805
1806                 if (sig_kernel_stop(signr)) {
1807                         /*
1808                          * The default action is to stop all threads in
1809                          * the thread group.  The job control signals
1810                          * do nothing in an orphaned pgrp, but SIGSTOP
1811                          * always works.  Note that siglock needs to be
1812                          * dropped during the call to is_orphaned_pgrp()
1813                          * because of lock ordering with tasklist_lock.
1814                          * This allows an intervening SIGCONT to be posted.
1815                          * We need to check for that and bail out if necessary.
1816                          */
1817                         if (signr == SIGSTOP) {
1818                                 do_signal_stop(signr); /* releases siglock */
1819                                 goto relock;
1820                         }
1821                         spin_unlock_irq(&current->sighand->siglock);
1822
1823                         /* signals can be posted during this window */
1824
1825                         if (is_orphaned_pgrp(process_group(current)))
1826                                 goto relock;
1827
1828                         spin_lock_irq(&current->sighand->siglock);
1829                         if (unlikely(sig_avoid_stop_race())) {
1830                                 /*
1831                                  * Either a SIGCONT or a SIGKILL signal was
1832                                  * posted in the siglock-not-held window.
1833                                  */
1834                                 continue;
1835                         }
1836
1837                         do_signal_stop(signr); /* releases siglock */
1838                         goto relock;
1839                 }
1840
1841                 spin_unlock_irq(&current->sighand->siglock);
1842
1843                 /*
1844                  * Anything else is fatal, maybe with a core dump.
1845                  */
1846                 current->flags |= PF_SIGNALED;
1847                 if (sig_kernel_coredump(signr) &&
1848                     do_coredump((long)signr, signr, regs)) {
1849                         /*
1850                          * That killed all other threads in the group and
1851                          * synchronized with their demise, so there can't
1852                          * be any more left to kill now.  The group_exit
1853                          * flags are set by do_coredump.  Note that
1854                          * thread_group_empty won't always be true yet,
1855                          * because those threads were blocked in __exit_mm
1856                          * and we just let them go to finish dying.
1857                          */
1858                         const int code = signr | 0x80;
1859                         BUG_ON(!current->signal->group_exit);
1860                         BUG_ON(current->signal->group_exit_code != code);
1861                         do_exit(code);
1862                         /* NOTREACHED */
1863                 }
1864
1865                 /*
1866                  * Death signals, no core dump.
1867                  */
1868                 do_group_exit(signr);
1869                 /* NOTREACHED */
1870         }
1871         spin_unlock_irq(&current->sighand->siglock);
1872         return signr;
1873 }
1874
1875 #endif
1876
1877 EXPORT_SYMBOL(recalc_sigpending);
1878 EXPORT_SYMBOL_GPL(dequeue_signal);
1879 EXPORT_SYMBOL(flush_signals);
1880 EXPORT_SYMBOL(force_sig);
1881 EXPORT_SYMBOL(force_sig_info);
1882 EXPORT_SYMBOL(kill_pg);
1883 EXPORT_SYMBOL(kill_pg_info);
1884 EXPORT_SYMBOL(kill_proc);
1885 EXPORT_SYMBOL(kill_proc_info);
1886 EXPORT_SYMBOL(kill_sl);
1887 EXPORT_SYMBOL(kill_sl_info);
1888 EXPORT_SYMBOL(notify_parent);
1889 EXPORT_SYMBOL(send_sig);
1890 EXPORT_SYMBOL(send_sig_info);
1891 EXPORT_SYMBOL(send_group_sig_info);
1892 EXPORT_SYMBOL(sigqueue_alloc);
1893 EXPORT_SYMBOL(sigqueue_free);
1894 EXPORT_SYMBOL(send_sigqueue);
1895 EXPORT_SYMBOL(send_group_sigqueue);
1896 EXPORT_SYMBOL(sigprocmask);
1897 EXPORT_SYMBOL(block_all_signals);
1898 EXPORT_SYMBOL(unblock_all_signals);
1899
1900
1901 /*
1902  * System call entry points.
1903  */
1904
1905 asmlinkage long sys_restart_syscall(void)
1906 {
1907         struct restart_block *restart = &current_thread_info()->restart_block;
1908         return restart->fn(restart);
1909 }
1910
1911 long do_no_restart_syscall(struct restart_block *param)
1912 {
1913         return -EINTR;
1914 }
1915
1916 /*
1917  * We don't need to get the kernel lock - this is all local to this
1918  * particular thread.. (and that's good, because this is _heavily_
1919  * used by various programs)
1920  */
1921
1922 /*
1923  * This is also useful for kernel threads that want to temporarily
1924  * (or permanently) block certain signals.
1925  *
1926  * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
1927  * interface happily blocks "unblockable" signals like SIGKILL
1928  * and friends.
1929  */
1930 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
1931 {
1932         int error;
1933         sigset_t old_block;
1934
1935         spin_lock_irq(&current->sighand->siglock);
1936         old_block = current->blocked;
1937         error = 0;
1938         switch (how) {
1939         case SIG_BLOCK:
1940                 sigorsets(&current->blocked, &current->blocked, set);
1941                 break;
1942         case SIG_UNBLOCK:
1943                 signandsets(&current->blocked, &current->blocked, set);
1944                 break;
1945         case SIG_SETMASK:
1946                 current->blocked = *set;
1947                 break;
1948         default:
1949                 error = -EINVAL;
1950         }
1951         recalc_sigpending();
1952         spin_unlock_irq(&current->sighand->siglock);
1953         if (oldset)
1954                 *oldset = old_block;
1955         return error;
1956 }
1957
1958 asmlinkage long
1959 sys_rt_sigprocmask(int how, sigset_t __user *set, sigset_t __user *oset, size_t sigsetsize)
1960 {
1961         int error = -EINVAL;
1962         sigset_t old_set, new_set;
1963
1964         /* XXX: Don't preclude handling different sized sigset_t's.  */
1965         if (sigsetsize != sizeof(sigset_t))
1966                 goto out;
1967
1968         if (set) {
1969                 error = -EFAULT;
1970                 if (copy_from_user(&new_set, set, sizeof(*set)))
1971                         goto out;
1972                 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
1973
1974                 error = sigprocmask(how, &new_set, &old_set);
1975                 if (error)
1976                         goto out;
1977                 if (oset)
1978                         goto set_old;
1979         } else if (oset) {
1980                 spin_lock_irq(&current->sighand->siglock);
1981                 old_set = current->blocked;
1982                 spin_unlock_irq(&current->sighand->siglock);
1983
1984         set_old:
1985                 error = -EFAULT;
1986                 if (copy_to_user(oset, &old_set, sizeof(*oset)))
1987                         goto out;
1988         }
1989         error = 0;
1990 out:
1991         return error;
1992 }
1993
1994 long do_sigpending(void __user *set, unsigned long sigsetsize)
1995 {
1996         long error = -EINVAL;
1997         sigset_t pending;
1998
1999         if (sigsetsize > sizeof(sigset_t))
2000                 goto out;
2001
2002         spin_lock_irq(&current->sighand->siglock);
2003         sigorsets(&pending, &current->pending.signal,
2004                   &current->signal->shared_pending.signal);
2005         spin_unlock_irq(&current->sighand->siglock);
2006
2007         /* Outside the lock because only this thread touches it.  */
2008         sigandsets(&pending, &current->blocked, &pending);
2009
2010         error = -EFAULT;
2011         if (!copy_to_user(set, &pending, sigsetsize))
2012                 error = 0;
2013
2014 out:
2015         return error;
2016 }       
2017
2018 asmlinkage long
2019 sys_rt_sigpending(sigset_t __user *set, size_t sigsetsize)
2020 {
2021         return do_sigpending(set, sigsetsize);
2022 }
2023
2024 #ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2025
2026 int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2027 {
2028         int err;
2029
2030         if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2031                 return -EFAULT;
2032         if (from->si_code < 0)
2033                 return __copy_to_user(to, from, sizeof(siginfo_t))
2034                         ? -EFAULT : 0;
2035         /*
2036          * If you change siginfo_t structure, please be sure
2037          * this code is fixed accordingly.
2038          * It should never copy any pad contained in the structure
2039          * to avoid security leaks, but must copy the generic
2040          * 3 ints plus the relevant union member.
2041          */
2042         err = __put_user(from->si_signo, &to->si_signo);
2043         err |= __put_user(from->si_errno, &to->si_errno);
2044         err |= __put_user((short)from->si_code, &to->si_code);
2045         switch (from->si_code & __SI_MASK) {
2046         case __SI_KILL:
2047                 err |= __put_user(from->si_pid, &to->si_pid);
2048                 err |= __put_user(from->si_uid, &to->si_uid);
2049                 break;
2050         case __SI_TIMER:
2051                  err |= __put_user(from->si_tid, &to->si_tid);
2052                  err |= __put_user(from->si_overrun, &to->si_overrun);
2053                  err |= __put_user(from->si_ptr, &to->si_ptr);
2054                 break;
2055         case __SI_POLL:
2056                 err |= __put_user(from->si_band, &to->si_band);
2057                 err |= __put_user(from->si_fd, &to->si_fd);
2058                 break;
2059         case __SI_FAULT:
2060                 err |= __put_user(from->si_addr, &to->si_addr);
2061 #ifdef __ARCH_SI_TRAPNO
2062                 err |= __put_user(from->si_trapno, &to->si_trapno);
2063 #endif
2064                 break;
2065         case __SI_CHLD:
2066                 err |= __put_user(from->si_pid, &to->si_pid);
2067                 err |= __put_user(from->si_uid, &to->si_uid);
2068                 err |= __put_user(from->si_status, &to->si_status);
2069                 err |= __put_user(from->si_utime, &to->si_utime);
2070                 err |= __put_user(from->si_stime, &to->si_stime);
2071                 break;
2072         case __SI_RT: /* This is not generated by the kernel as of now. */
2073         case __SI_MESGQ: /* But this is */
2074                 err |= __put_user(from->si_pid, &to->si_pid);
2075                 err |= __put_user(from->si_uid, &to->si_uid);
2076                 err |= __put_user(from->si_ptr, &to->si_ptr);
2077                 break;
2078         default: /* this is just in case for now ... */
2079                 err |= __put_user(from->si_pid, &to->si_pid);
2080                 err |= __put_user(from->si_uid, &to->si_uid);
2081                 break;
2082         }
2083         return err;
2084 }
2085
2086 #endif
2087
2088 asmlinkage long
2089 sys_rt_sigtimedwait(const sigset_t __user *uthese,
2090                     siginfo_t __user *uinfo,
2091                     const struct timespec __user *uts,
2092                     size_t sigsetsize)
2093 {
2094         int ret, sig;
2095         sigset_t these;
2096         struct timespec ts;
2097         siginfo_t info;
2098         long timeout = 0;
2099
2100         /* XXX: Don't preclude handling different sized sigset_t's.  */
2101         if (sigsetsize != sizeof(sigset_t))
2102                 return -EINVAL;
2103
2104         if (copy_from_user(&these, uthese, sizeof(these)))
2105                 return -EFAULT;
2106                 
2107         /*
2108          * Invert the set of allowed signals to get those we
2109          * want to block.
2110          */
2111         sigdelsetmask(&these, sigmask(SIGKILL)|sigmask(SIGSTOP));
2112         signotset(&these);
2113
2114         if (uts) {
2115                 if (copy_from_user(&ts, uts, sizeof(ts)))
2116                         return -EFAULT;
2117                 if (ts.tv_nsec >= 1000000000L || ts.tv_nsec < 0
2118                     || ts.tv_sec < 0)
2119                         return -EINVAL;
2120         }
2121
2122         spin_lock_irq(&current->sighand->siglock);
2123         sig = dequeue_signal(current, &these, &info);
2124         if (!sig) {
2125                 timeout = MAX_SCHEDULE_TIMEOUT;
2126                 if (uts)
2127                         timeout = (timespec_to_jiffies(&ts)
2128                                    + (ts.tv_sec || ts.tv_nsec));
2129
2130                 if (timeout) {
2131                         /* None ready -- temporarily unblock those we're
2132                          * interested while we are sleeping in so that we'll
2133                          * be awakened when they arrive.  */
2134                         current->real_blocked = current->blocked;
2135                         sigandsets(&current->blocked, &current->blocked, &these);
2136                         recalc_sigpending();
2137                         spin_unlock_irq(&current->sighand->siglock);
2138
2139                         current->state = TASK_INTERRUPTIBLE;
2140                         timeout = schedule_timeout(timeout);
2141
2142                         spin_lock_irq(&current->sighand->siglock);
2143                         sig = dequeue_signal(current, &these, &info);
2144                         current->blocked = current->real_blocked;
2145                         siginitset(&current->real_blocked, 0);
2146                         recalc_sigpending();
2147                 }
2148         }
2149         spin_unlock_irq(&current->sighand->siglock);
2150
2151         if (sig) {
2152                 ret = sig;
2153                 if (uinfo) {
2154                         if (copy_siginfo_to_user(uinfo, &info))
2155                                 ret = -EFAULT;
2156                 }
2157         } else {
2158                 ret = -EAGAIN;
2159                 if (timeout)
2160                         ret = -EINTR;
2161         }
2162
2163         return ret;
2164 }
2165
2166 asmlinkage long
2167 sys_kill(int pid, int sig)
2168 {
2169         struct siginfo info;
2170
2171         info.si_signo = sig;
2172         info.si_errno = 0;
2173         info.si_code = SI_USER;
2174         info.si_pid = current->tgid;
2175         info.si_uid = current->uid;
2176
2177         return kill_something_info(sig, &info, pid);
2178 }
2179
2180 /**
2181  *  sys_tgkill - send signal to one specific thread
2182  *  @tgid: the thread group ID of the thread
2183  *  @pid: the PID of the thread
2184  *  @sig: signal to be sent
2185  *
2186  *  This syscall also checks the tgid and returns -ESRCH even if the PID
2187  *  exists but it's not belonging to the target process anymore. This
2188  *  method solves the problem of threads exiting and PIDs getting reused.
2189  */
2190 asmlinkage long sys_tgkill(int tgid, int pid, int sig)
2191 {
2192         struct siginfo info;
2193         int error;
2194         struct task_struct *p;
2195
2196         /* This is only valid for single tasks */
2197         if (pid <= 0 || tgid <= 0)
2198                 return -EINVAL;
2199
2200         info.si_signo = sig;
2201         info.si_errno = 0;
2202         info.si_code = SI_TKILL;
2203         info.si_pid = current->tgid;
2204         info.si_uid = current->uid;
2205
2206         read_lock(&tasklist_lock);
2207         p = find_task_by_pid(pid);
2208         error = -ESRCH;
2209         if (p && (p->tgid == tgid)) {
2210                 error = check_kill_permission(sig, &info, p);
2211                 /*
2212                  * The null signal is a permissions and process existence
2213                  * probe.  No signal is actually delivered.
2214                  */
2215                 if (!error && sig && p->sighand) {
2216                         spin_lock_irq(&p->sighand->siglock);
2217                         handle_stop_signal(sig, p);
2218                         error = specific_send_sig_info(sig, &info, p);
2219                         spin_unlock_irq(&p->sighand->siglock);
2220                 }
2221         }
2222         read_unlock(&tasklist_lock);
2223         return error;
2224 }
2225
2226 /*
2227  *  Send a signal to only one task, even if it's a CLONE_THREAD task.
2228  */
2229 asmlinkage long
2230 sys_tkill(int pid, int sig)
2231 {
2232         struct siginfo info;
2233         int error;
2234         struct task_struct *p;
2235
2236         /* This is only valid for single tasks */
2237         if (pid <= 0)
2238                 return -EINVAL;
2239
2240         info.si_signo = sig;
2241         info.si_errno = 0;
2242         info.si_code = SI_TKILL;
2243         info.si_pid = current->tgid;
2244         info.si_uid = current->uid;
2245
2246         read_lock(&tasklist_lock);
2247         p = find_task_by_pid(pid);
2248         error = -ESRCH;
2249         if (p) {
2250                 error = check_kill_permission(sig, &info, p);
2251                 /*
2252                  * The null signal is a permissions and process existence
2253                  * probe.  No signal is actually delivered.
2254                  */
2255                 if (!error && sig && p->sighand) {
2256                         spin_lock_irq(&p->sighand->siglock);
2257                         handle_stop_signal(sig, p);
2258                         error = specific_send_sig_info(sig, &info, p);
2259                         spin_unlock_irq(&p->sighand->siglock);
2260                 }
2261         }
2262         read_unlock(&tasklist_lock);
2263         return error;
2264 }
2265
2266 asmlinkage long
2267 sys_rt_sigqueueinfo(int pid, int sig, siginfo_t __user *uinfo)
2268 {
2269         siginfo_t info;
2270
2271         if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2272                 return -EFAULT;
2273
2274         /* Not even root can pretend to send signals from the kernel.
2275            Nor can they impersonate a kill(), which adds source info.  */
2276         if (info.si_code >= 0)
2277                 return -EPERM;
2278         info.si_signo = sig;
2279
2280         /* POSIX.1b doesn't mention process groups.  */
2281         return kill_proc_info(sig, &info, pid);
2282 }
2283
2284 int
2285 do_sigaction(int sig, const struct k_sigaction *act, struct k_sigaction *oact)
2286 {
2287         struct k_sigaction *k;
2288
2289         if (sig < 1 || sig > _NSIG || (act && sig_kernel_only(sig)))
2290                 return -EINVAL;
2291
2292         k = &current->sighand->action[sig-1];
2293
2294         spin_lock_irq(&current->sighand->siglock);
2295         if (signal_pending(current)) {
2296                 /*
2297                  * If there might be a fatal signal pending on multiple
2298                  * threads, make sure we take it before changing the action.
2299                  */
2300                 spin_unlock_irq(&current->sighand->siglock);
2301                 return -ERESTARTNOINTR;
2302         }
2303
2304         if (oact)
2305                 *oact = *k;
2306
2307         if (act) {
2308                 /*
2309                  * POSIX 3.3.1.3:
2310                  *  "Setting a signal action to SIG_IGN for a signal that is
2311                  *   pending shall cause the pending signal to be discarded,
2312                  *   whether or not it is blocked."
2313                  *
2314                  *  "Setting a signal action to SIG_DFL for a signal that is
2315                  *   pending and whose default action is to ignore the signal
2316                  *   (for example, SIGCHLD), shall cause the pending signal to
2317                  *   be discarded, whether or not it is blocked"
2318                  */
2319                 if (act->sa.sa_handler == SIG_IGN ||
2320                     (act->sa.sa_handler == SIG_DFL &&
2321                      sig_kernel_ignore(sig))) {
2322                         /*
2323                          * This is a fairly rare case, so we only take the
2324                          * tasklist_lock once we're sure we'll need it.
2325                          * Now we must do this little unlock and relock
2326                          * dance to maintain the lock hierarchy.
2327                          */
2328                         struct task_struct *t = current;
2329                         spin_unlock_irq(&t->sighand->siglock);
2330                         read_lock(&tasklist_lock);
2331                         spin_lock_irq(&t->sighand->siglock);
2332                         *k = *act;
2333                         sigdelsetmask(&k->sa.sa_mask,
2334                                       sigmask(SIGKILL) | sigmask(SIGSTOP));
2335                         rm_from_queue(sigmask(sig), &t->signal->shared_pending);
2336                         do {
2337                                 rm_from_queue(sigmask(sig), &t->pending);
2338                                 recalc_sigpending_tsk(t);
2339                                 t = next_thread(t);
2340                         } while (t != current);
2341                         spin_unlock_irq(&current->sighand->siglock);
2342                         read_unlock(&tasklist_lock);
2343                         return 0;
2344                 }
2345
2346                 *k = *act;
2347                 sigdelsetmask(&k->sa.sa_mask,
2348                               sigmask(SIGKILL) | sigmask(SIGSTOP));
2349         }
2350
2351         spin_unlock_irq(&current->sighand->siglock);
2352         return 0;
2353 }
2354
2355 int 
2356 do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
2357 {
2358         stack_t oss;
2359         int error;
2360
2361         if (uoss) {
2362                 oss.ss_sp = (void *) current->sas_ss_sp;
2363                 oss.ss_size = current->sas_ss_size;
2364                 oss.ss_flags = sas_ss_flags(sp);
2365         }
2366
2367         if (uss) {
2368                 void *ss_sp;
2369                 size_t ss_size;
2370                 int ss_flags;
2371
2372                 error = -EFAULT;
2373                 if (verify_area(VERIFY_READ, uss, sizeof(*uss))
2374                     || __get_user(ss_sp, &uss->ss_sp)
2375                     || __get_user(ss_flags, &uss->ss_flags)
2376                     || __get_user(ss_size, &uss->ss_size))
2377                         goto out;
2378
2379                 error = -EPERM;
2380                 if (on_sig_stack(sp))
2381                         goto out;
2382
2383                 error = -EINVAL;
2384                 /*
2385                  *
2386                  * Note - this code used to test ss_flags incorrectly
2387                  *        old code may have been written using ss_flags==0
2388                  *        to mean ss_flags==SS_ONSTACK (as this was the only
2389                  *        way that worked) - this fix preserves that older
2390                  *        mechanism
2391                  */
2392                 if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
2393                         goto out;
2394
2395                 if (ss_flags == SS_DISABLE) {
2396                         ss_size = 0;
2397                         ss_sp = NULL;
2398                 } else {
2399                         error = -ENOMEM;
2400                         if (ss_size < MINSIGSTKSZ)
2401                                 goto out;
2402                 }
2403
2404                 current->sas_ss_sp = (unsigned long) ss_sp;
2405                 current->sas_ss_size = ss_size;
2406         }
2407
2408         if (uoss) {
2409                 error = -EFAULT;
2410                 if (copy_to_user(uoss, &oss, sizeof(oss)))
2411                         goto out;
2412         }
2413
2414         error = 0;
2415 out:
2416         return error;
2417 }
2418
2419 #ifdef __ARCH_WANT_SYS_SIGPENDING
2420
2421 asmlinkage long
2422 sys_sigpending(old_sigset_t __user *set)
2423 {
2424         return do_sigpending(set, sizeof(*set));
2425 }
2426
2427 #endif
2428
2429 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
2430 /* Some platforms have their own version with special arguments others
2431    support only sys_rt_sigprocmask.  */
2432
2433 asmlinkage long
2434 sys_sigprocmask(int how, old_sigset_t __user *set, old_sigset_t __user *oset)
2435 {
2436         int error;
2437         old_sigset_t old_set, new_set;
2438
2439         if (set) {
2440                 error = -EFAULT;
2441                 if (copy_from_user(&new_set, set, sizeof(*set)))
2442                         goto out;
2443                 new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
2444
2445                 spin_lock_irq(&current->sighand->siglock);
2446                 old_set = current->blocked.sig[0];
2447
2448                 error = 0;
2449                 switch (how) {
2450                 default:
2451                         error = -EINVAL;
2452                         break;
2453                 case SIG_BLOCK:
2454                         sigaddsetmask(&current->blocked, new_set);
2455                         break;
2456                 case SIG_UNBLOCK:
2457                         sigdelsetmask(&current->blocked, new_set);
2458                         break;
2459                 case SIG_SETMASK:
2460                         current->blocked.sig[0] = new_set;
2461                         break;
2462                 }
2463
2464                 recalc_sigpending();
2465                 spin_unlock_irq(&current->sighand->siglock);
2466                 if (error)
2467                         goto out;
2468                 if (oset)
2469                         goto set_old;
2470         } else if (oset) {
2471                 old_set = current->blocked.sig[0];
2472         set_old:
2473                 error = -EFAULT;
2474                 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2475                         goto out;
2476         }
2477         error = 0;
2478 out:
2479         return error;
2480 }
2481 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
2482
2483 #ifdef __ARCH_WANT_SYS_RT_SIGACTION
2484 asmlinkage long
2485 sys_rt_sigaction(int sig,
2486                  const struct sigaction __user *act,
2487                  struct sigaction __user *oact,
2488                  size_t sigsetsize)
2489 {
2490         struct k_sigaction new_sa, old_sa;
2491         int ret = -EINVAL;
2492
2493         /* XXX: Don't preclude handling different sized sigset_t's.  */
2494         if (sigsetsize != sizeof(sigset_t))
2495                 goto out;
2496
2497         if (act) {
2498                 if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
2499                         return -EFAULT;
2500         }
2501
2502         ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
2503
2504         if (!ret && oact) {
2505                 if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
2506                         return -EFAULT;
2507         }
2508 out:
2509         return ret;
2510 }
2511 #endif /* __ARCH_WANT_SYS_RT_SIGACTION */
2512
2513 #ifdef __ARCH_WANT_SYS_SGETMASK
2514
2515 /*
2516  * For backwards compatibility.  Functionality superseded by sigprocmask.
2517  */
2518 asmlinkage long
2519 sys_sgetmask(void)
2520 {
2521         /* SMP safe */
2522         return current->blocked.sig[0];
2523 }
2524
2525 asmlinkage long
2526 sys_ssetmask(int newmask)
2527 {
2528         int old;
2529
2530         spin_lock_irq(&current->sighand->siglock);
2531         old = current->blocked.sig[0];
2532
2533         siginitset(&current->blocked, newmask & ~(sigmask(SIGKILL)|
2534                                                   sigmask(SIGSTOP)));
2535         recalc_sigpending();
2536         spin_unlock_irq(&current->sighand->siglock);
2537
2538         return old;
2539 }
2540 #endif /* __ARCH_WANT_SGETMASK */
2541
2542 #ifdef __ARCH_WANT_SYS_SIGNAL
2543 /*
2544  * For backwards compatibility.  Functionality superseded by sigaction.
2545  */
2546 asmlinkage unsigned long
2547 sys_signal(int sig, __sighandler_t handler)
2548 {
2549         struct k_sigaction new_sa, old_sa;
2550         int ret;
2551
2552         new_sa.sa.sa_handler = handler;
2553         new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
2554
2555         ret = do_sigaction(sig, &new_sa, &old_sa);
2556
2557         return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
2558 }
2559 #endif /* __ARCH_WANT_SYS_SIGNAL */
2560
2561 #ifdef __ARCH_WANT_SYS_PAUSE
2562
2563 asmlinkage long
2564 sys_pause(void)
2565 {
2566         current->state = TASK_INTERRUPTIBLE;
2567         schedule();
2568         return -ERESTARTNOHAND;
2569 }
2570
2571 #endif
2572
2573 void __init signals_init(void)
2574 {
2575         sigqueue_cachep =
2576                 kmem_cache_create("sigqueue",
2577                                   sizeof(struct sigqueue),
2578                                   __alignof__(struct sigqueue),
2579                                   SLAB_PANIC, NULL, NULL);
2580 }