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