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