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