vserver 2.0 rc7
[linux-2.6.git] / kernel / signal.c
1 /*
2  *  linux/kernel/signal.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  *
6  *  1997-11-02  Modified for POSIX.1b signals by Richard Henderson
7  *
8  *  2003-06-02  Jim Houston - Concurrent Computer Corp.
9  *              Changes to use preallocated sigqueue structures
10  *              to allow signals to be sent reliably.
11  */
12
13 #include <linux/config.h>
14 #include <linux/slab.h>
15 #include <linux/module.h>
16 #include <linux/smp_lock.h>
17 #include <linux/init.h>
18 #include <linux/sched.h>
19 #include <linux/fs.h>
20 #include <linux/tty.h>
21 #include <linux/binfmts.h>
22 #include <linux/security.h>
23 #include <linux/syscalls.h>
24 #include <linux/ptrace.h>
25 #include <linux/posix-timers.h>
26 #include <linux/signal.h>
27 #include <asm/param.h>
28 #include <asm/uaccess.h>
29 #include <asm/unistd.h>
30 #include <asm/siginfo.h>
31
32 /*
33  * SLAB caches for signal bits.
34  */
35
36 static kmem_cache_t *sigqueue_cachep;
37
38 /*
39  * In POSIX a signal is sent either to a specific thread (Linux task)
40  * or to the process as a whole (Linux thread group).  How the signal
41  * is sent determines whether it's to one thread or the whole group,
42  * which determines which signal mask(s) are involved in blocking it
43  * from being delivered until later.  When the signal is delivered,
44  * either it's caught or ignored by a user handler or it has a default
45  * effect that applies to the whole thread group (POSIX process).
46  *
47  * The possible effects an unblocked signal set to SIG_DFL can have are:
48  *   ignore     - Nothing Happens
49  *   terminate  - kill the process, i.e. all threads in the group,
50  *                similar to exit_group.  The group leader (only) reports
51  *                WIFSIGNALED status to its parent.
52  *   coredump   - write a core dump file describing all threads using
53  *                the same mm and then kill all those threads
54  *   stop       - stop all the threads in the group, i.e. TASK_STOPPED state
55  *
56  * SIGKILL and SIGSTOP cannot be caught, blocked, or ignored.
57  * Other signals when not blocked and set to SIG_DFL behaves as follows.
58  * The job control signals also have other special effects.
59  *
60  *      +--------------------+------------------+
61  *      |  POSIX signal      |  default action  |
62  *      +--------------------+------------------+
63  *      |  SIGHUP            |  terminate       |
64  *      |  SIGINT            |  terminate       |
65  *      |  SIGQUIT           |  coredump        |
66  *      |  SIGILL            |  coredump        |
67  *      |  SIGTRAP           |  coredump        |
68  *      |  SIGABRT/SIGIOT    |  coredump        |
69  *      |  SIGBUS            |  coredump        |
70  *      |  SIGFPE            |  coredump        |
71  *      |  SIGKILL           |  terminate(+)    |
72  *      |  SIGUSR1           |  terminate       |
73  *      |  SIGSEGV           |  coredump        |
74  *      |  SIGUSR2           |  terminate       |
75  *      |  SIGPIPE           |  terminate       |
76  *      |  SIGALRM           |  terminate       |
77  *      |  SIGTERM           |  terminate       |
78  *      |  SIGCHLD           |  ignore          |
79  *      |  SIGCONT           |  ignore(*)       |
80  *      |  SIGSTOP           |  stop(*)(+)      |
81  *      |  SIGTSTP           |  stop(*)         |
82  *      |  SIGTTIN           |  stop(*)         |
83  *      |  SIGTTOU           |  stop(*)         |
84  *      |  SIGURG            |  ignore          |
85  *      |  SIGXCPU           |  coredump        |
86  *      |  SIGXFSZ           |  coredump        |
87  *      |  SIGVTALRM         |  terminate       |
88  *      |  SIGPROF           |  terminate       |
89  *      |  SIGPOLL/SIGIO     |  terminate       |
90  *      |  SIGSYS/SIGUNUSED  |  coredump        |
91  *      |  SIGSTKFLT         |  terminate       |
92  *      |  SIGWINCH          |  ignore          |
93  *      |  SIGPWR            |  terminate       |
94  *      |  SIGRTMIN-SIGRTMAX |  terminate       |
95  *      +--------------------+------------------+
96  *      |  non-POSIX signal  |  default action  |
97  *      +--------------------+------------------+
98  *      |  SIGEMT            |  coredump        |
99  *      +--------------------+------------------+
100  *
101  * (+) For SIGKILL and SIGSTOP the action is "always", not just "default".
102  * (*) Special job control effects:
103  * When SIGCONT is sent, it resumes the process (all threads in the group)
104  * from TASK_STOPPED state and also clears any pending/queued stop signals
105  * (any of those marked with "stop(*)").  This happens regardless of blocking,
106  * catching, or ignoring SIGCONT.  When any stop signal is sent, it clears
107  * any pending/queued SIGCONT signals; this happens regardless of blocking,
108  * catching, or ignored the stop signal, though (except for SIGSTOP) the
109  * default action of stopping the process may happen later or never.
110  */
111
112 #ifdef SIGEMT
113 #define M_SIGEMT        M(SIGEMT)
114 #else
115 #define M_SIGEMT        0
116 #endif
117
118 #if SIGRTMIN > BITS_PER_LONG
119 #define M(sig) (1ULL << ((sig)-1))
120 #else
121 #define M(sig) (1UL << ((sig)-1))
122 #endif
123 #define T(sig, mask) (M(sig) & (mask))
124
125 #define SIG_KERNEL_ONLY_MASK (\
126         M(SIGKILL)   |  M(SIGSTOP)                                   )
127
128 #define SIG_KERNEL_STOP_MASK (\
129         M(SIGSTOP)   |  M(SIGTSTP)   |  M(SIGTTIN)   |  M(SIGTTOU)   )
130
131 #define SIG_KERNEL_COREDUMP_MASK (\
132         M(SIGQUIT)   |  M(SIGILL)    |  M(SIGTRAP)   |  M(SIGABRT)   | \
133         M(SIGFPE)    |  M(SIGSEGV)   |  M(SIGBUS)    |  M(SIGSYS)    | \
134         M(SIGXCPU)   |  M(SIGXFSZ)   |  M_SIGEMT                     )
135
136 #define SIG_KERNEL_IGNORE_MASK (\
137         M(SIGCONT)   |  M(SIGCHLD)   |  M(SIGWINCH)  |  M(SIGURG)    )
138
139 #define sig_kernel_only(sig) \
140                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_ONLY_MASK))
141 #define sig_kernel_coredump(sig) \
142                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_COREDUMP_MASK))
143 #define sig_kernel_ignore(sig) \
144                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_IGNORE_MASK))
145 #define sig_kernel_stop(sig) \
146                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_STOP_MASK))
147
148 #define sig_user_defined(t, signr) \
149         (((t)->sighand->action[(signr)-1].sa.sa_handler != SIG_DFL) &&  \
150          ((t)->sighand->action[(signr)-1].sa.sa_handler != SIG_IGN))
151
152 #define sig_fatal(t, signr) \
153         (!T(signr, SIG_KERNEL_IGNORE_MASK|SIG_KERNEL_STOP_MASK) && \
154          (t)->sighand->action[(signr)-1].sa.sa_handler == SIG_DFL)
155
156 static int sig_ignored(struct task_struct *t, int sig)
157 {
158         void __user * handler;
159
160         /*
161          * Tracers always want to know about signals..
162          */
163         if (t->ptrace & PT_PTRACED)
164                 return 0;
165
166         /*
167          * Blocked signals are never ignored, since the
168          * signal handler may change by the time it is
169          * unblocked.
170          */
171         if (sigismember(&t->blocked, sig))
172                 return 0;
173
174         /* Is it explicitly or implicitly ignored? */
175         handler = t->sighand->action[sig-1].sa.sa_handler;
176         return   handler == SIG_IGN ||
177                 (handler == SIG_DFL && sig_kernel_ignore(sig));
178 }
179
180 /*
181  * Re-calculate pending state from the set of locally pending
182  * signals, globally pending signals, and blocked signals.
183  */
184 static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
185 {
186         unsigned long ready;
187         long i;
188
189         switch (_NSIG_WORDS) {
190         default:
191                 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
192                         ready |= signal->sig[i] &~ blocked->sig[i];
193                 break;
194
195         case 4: ready  = signal->sig[3] &~ blocked->sig[3];
196                 ready |= signal->sig[2] &~ blocked->sig[2];
197                 ready |= signal->sig[1] &~ blocked->sig[1];
198                 ready |= signal->sig[0] &~ blocked->sig[0];
199                 break;
200
201         case 2: ready  = signal->sig[1] &~ blocked->sig[1];
202                 ready |= signal->sig[0] &~ blocked->sig[0];
203                 break;
204
205         case 1: ready  = signal->sig[0] &~ blocked->sig[0];
206         }
207         return ready != 0;
208 }
209
210 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
211
212 fastcall void recalc_sigpending_tsk(struct task_struct *t)
213 {
214         if (t->signal->group_stop_count > 0 ||
215             PENDING(&t->pending, &t->blocked) ||
216             PENDING(&t->signal->shared_pending, &t->blocked))
217                 set_tsk_thread_flag(t, TIF_SIGPENDING);
218         else
219                 clear_tsk_thread_flag(t, TIF_SIGPENDING);
220 }
221
222 void recalc_sigpending(void)
223 {
224         recalc_sigpending_tsk(current);
225 }
226
227 /* Given the mask, find the first available signal that should be serviced. */
228
229 static int
230 next_signal(struct sigpending *pending, sigset_t *mask)
231 {
232         unsigned long i, *s, *m, x;
233         int sig = 0;
234         
235         s = pending->signal.sig;
236         m = mask->sig;
237         switch (_NSIG_WORDS) {
238         default:
239                 for (i = 0; i < _NSIG_WORDS; ++i, ++s, ++m)
240                         if ((x = *s &~ *m) != 0) {
241                                 sig = ffz(~x) + i*_NSIG_BPW + 1;
242                                 break;
243                         }
244                 break;
245
246         case 2: if ((x = s[0] &~ m[0]) != 0)
247                         sig = 1;
248                 else if ((x = s[1] &~ m[1]) != 0)
249                         sig = _NSIG_BPW + 1;
250                 else
251                         break;
252                 sig += ffz(~x);
253                 break;
254
255         case 1: if ((x = *s &~ *m) != 0)
256                         sig = ffz(~x) + 1;
257                 break;
258         }
259         
260         return sig;
261 }
262
263 static struct sigqueue *__sigqueue_alloc(struct task_struct *t, unsigned int __nocast flags,
264                                          int override_rlimit)
265 {
266         struct sigqueue *q = NULL;
267
268         atomic_inc(&t->user->sigpending);
269         if (override_rlimit ||
270             atomic_read(&t->user->sigpending) <=
271                         t->signal->rlim[RLIMIT_SIGPENDING].rlim_cur)
272                 q = kmem_cache_alloc(sigqueue_cachep, flags);
273         if (unlikely(q == NULL)) {
274                 atomic_dec(&t->user->sigpending);
275         } else {
276                 INIT_LIST_HEAD(&q->list);
277                 q->flags = 0;
278                 q->lock = NULL;
279                 q->user = get_uid(t->user);
280         }
281         return(q);
282 }
283
284 static inline void __sigqueue_free(struct sigqueue *q)
285 {
286         if (q->flags & SIGQUEUE_PREALLOC)
287                 return;
288         atomic_dec(&q->user->sigpending);
289         free_uid(q->user);
290         kmem_cache_free(sigqueue_cachep, q);
291 }
292
293 static void flush_sigqueue(struct sigpending *queue)
294 {
295         struct sigqueue *q;
296
297         sigemptyset(&queue->signal);
298         while (!list_empty(&queue->list)) {
299                 q = list_entry(queue->list.next, struct sigqueue , list);
300                 list_del_init(&q->list);
301                 __sigqueue_free(q);
302         }
303 }
304
305 /*
306  * Flush all pending signals for a task.
307  */
308
309 void
310 flush_signals(struct task_struct *t)
311 {
312         unsigned long flags;
313
314         spin_lock_irqsave(&t->sighand->siglock, flags);
315         clear_tsk_thread_flag(t,TIF_SIGPENDING);
316         flush_sigqueue(&t->pending);
317         flush_sigqueue(&t->signal->shared_pending);
318         spin_unlock_irqrestore(&t->sighand->siglock, flags);
319 }
320
321 /*
322  * This function expects the tasklist_lock write-locked.
323  */
324 void __exit_sighand(struct task_struct *tsk)
325 {
326         struct sighand_struct * sighand = tsk->sighand;
327
328         /* Ok, we're done with the signal handlers */
329         tsk->sighand = NULL;
330         if (atomic_dec_and_test(&sighand->count))
331                 kmem_cache_free(sighand_cachep, sighand);
332 }
333
334 void exit_sighand(struct task_struct *tsk)
335 {
336         write_lock_irq(&tasklist_lock);
337         __exit_sighand(tsk);
338         write_unlock_irq(&tasklist_lock);
339 }
340
341 /*
342  * This function expects the tasklist_lock write-locked.
343  */
344 void __exit_signal(struct task_struct *tsk)
345 {
346         struct signal_struct * sig = tsk->signal;
347         struct sighand_struct * sighand = tsk->sighand;
348
349         if (!sig)
350                 BUG();
351         if (!atomic_read(&sig->count))
352                 BUG();
353         spin_lock(&sighand->siglock);
354         posix_cpu_timers_exit(tsk);
355         if (atomic_dec_and_test(&sig->count)) {
356                 posix_cpu_timers_exit_group(tsk);
357                 if (tsk == sig->curr_target)
358                         sig->curr_target = next_thread(tsk);
359                 tsk->signal = NULL;
360                 spin_unlock(&sighand->siglock);
361                 flush_sigqueue(&sig->shared_pending);
362         } else {
363                 /*
364                  * If there is any task waiting for the group exit
365                  * then notify it:
366                  */
367                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
368                         wake_up_process(sig->group_exit_task);
369                         sig->group_exit_task = NULL;
370                 }
371                 if (tsk == sig->curr_target)
372                         sig->curr_target = next_thread(tsk);
373                 tsk->signal = NULL;
374                 /*
375                  * Accumulate here the counters for all threads but the
376                  * group leader as they die, so they can be added into
377                  * the process-wide totals when those are taken.
378                  * The group leader stays around as a zombie as long
379                  * as there are other threads.  When it gets reaped,
380                  * the exit.c code will add its counts into these totals.
381                  * We won't ever get here for the group leader, since it
382                  * will have been the last reference on the signal_struct.
383                  */
384                 sig->utime = cputime_add(sig->utime, tsk->utime);
385                 sig->stime = cputime_add(sig->stime, tsk->stime);
386                 sig->min_flt += tsk->min_flt;
387                 sig->maj_flt += tsk->maj_flt;
388                 sig->nvcsw += tsk->nvcsw;
389                 sig->nivcsw += tsk->nivcsw;
390                 sig->sched_time += tsk->sched_time;
391                 spin_unlock(&sighand->siglock);
392                 sig = NULL;     /* Marker for below.  */
393         }
394         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
395         flush_sigqueue(&tsk->pending);
396         if (sig) {
397                 /*
398                  * We are cleaning up the signal_struct here.  We delayed
399                  * calling exit_itimers until after flush_sigqueue, just in
400                  * case our thread-local pending queue contained a queued
401                  * timer signal that would have been cleared in
402                  * exit_itimers.  When that called sigqueue_free, it would
403                  * attempt to re-take the tasklist_lock and deadlock.  This
404                  * can never happen if we ensure that all queues the
405                  * timer's signal might be queued on have been flushed
406                  * first.  The shared_pending queue, and our own pending
407                  * queue are the only queues the timer could be on, since
408                  * there are no other threads left in the group and timer
409                  * signals are constrained to threads inside the group.
410                  */
411                 exit_itimers(sig);
412                 exit_thread_group_keys(sig);
413                 kmem_cache_free(signal_cachep, sig);
414         }
415 }
416
417 void exit_signal(struct task_struct *tsk)
418 {
419         write_lock_irq(&tasklist_lock);
420         __exit_signal(tsk);
421         write_unlock_irq(&tasklist_lock);
422 }
423
424 /*
425  * Flush all handlers for a task.
426  */
427
428 void
429 flush_signal_handlers(struct task_struct *t, int force_default)
430 {
431         int i;
432         struct k_sigaction *ka = &t->sighand->action[0];
433         for (i = _NSIG ; i != 0 ; i--) {
434                 if (force_default || ka->sa.sa_handler != SIG_IGN)
435                         ka->sa.sa_handler = SIG_DFL;
436                 ka->sa.sa_flags = 0;
437                 sigemptyset(&ka->sa.sa_mask);
438                 ka++;
439         }
440 }
441
442
443 /* Notify the system that a driver wants to block all signals for this
444  * process, and wants to be notified if any signals at all were to be
445  * sent/acted upon.  If the notifier routine returns non-zero, then the
446  * signal will be acted upon after all.  If the notifier routine returns 0,
447  * then then signal will be blocked.  Only one block per process is
448  * allowed.  priv is a pointer to private data that the notifier routine
449  * can use to determine if the signal should be blocked or not.  */
450
451 void
452 block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
453 {
454         unsigned long flags;
455
456         spin_lock_irqsave(&current->sighand->siglock, flags);
457         current->notifier_mask = mask;
458         current->notifier_data = priv;
459         current->notifier = notifier;
460         spin_unlock_irqrestore(&current->sighand->siglock, flags);
461 }
462
463 /* Notify the system that blocking has ended. */
464
465 void
466 unblock_all_signals(void)
467 {
468         unsigned long flags;
469
470         spin_lock_irqsave(&current->sighand->siglock, flags);
471         current->notifier = NULL;
472         current->notifier_data = NULL;
473         recalc_sigpending();
474         spin_unlock_irqrestore(&current->sighand->siglock, flags);
475 }
476
477 static inline int collect_signal(int sig, struct sigpending *list, siginfo_t *info)
478 {
479         struct sigqueue *q, *first = NULL;
480         int still_pending = 0;
481
482         if (unlikely(!sigismember(&list->signal, sig)))
483                 return 0;
484
485         /*
486          * Collect the siginfo appropriate to this signal.  Check if
487          * there is another siginfo for the same signal.
488         */
489         list_for_each_entry(q, &list->list, list) {
490                 if (q->info.si_signo == sig) {
491                         if (first) {
492                                 still_pending = 1;
493                                 break;
494                         }
495                         first = q;
496                 }
497         }
498         if (first) {
499                 list_del_init(&first->list);
500                 copy_siginfo(info, &first->info);
501                 __sigqueue_free(first);
502                 if (!still_pending)
503                         sigdelset(&list->signal, sig);
504         } else {
505
506                 /* Ok, it wasn't in the queue.  This must be
507                    a fast-pathed signal or we must have been
508                    out of queue space.  So zero out the info.
509                  */
510                 sigdelset(&list->signal, sig);
511                 info->si_signo = sig;
512                 info->si_errno = 0;
513                 info->si_code = 0;
514                 info->si_pid = 0;
515                 info->si_uid = 0;
516         }
517         return 1;
518 }
519
520 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
521                         siginfo_t *info)
522 {
523         int sig = 0;
524
525         /* SIGKILL must have priority, otherwise it is quite easy
526          * to create an unkillable process, sending sig < SIGKILL
527          * to self */
528         if (unlikely(sigismember(&pending->signal, SIGKILL))) {
529                 if (!sigismember(mask, SIGKILL))
530                         sig = SIGKILL;
531         }
532
533         if (likely(!sig))
534                 sig = next_signal(pending, mask);
535         if (sig) {
536                 if (current->notifier) {
537                         if (sigismember(current->notifier_mask, sig)) {
538                                 if (!(current->notifier)(current->notifier_data)) {
539                                         clear_thread_flag(TIF_SIGPENDING);
540                                         return 0;
541                                 }
542                         }
543                 }
544
545                 if (!collect_signal(sig, pending, info))
546                         sig = 0;
547                                 
548         }
549         recalc_sigpending();
550
551         return sig;
552 }
553
554 /*
555  * Dequeue a signal and return the element to the caller, which is 
556  * expected to free it.
557  *
558  * All callers have to hold the siglock.
559  */
560 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
561 {
562         int signr = __dequeue_signal(&tsk->pending, mask, info);
563         if (!signr)
564                 signr = __dequeue_signal(&tsk->signal->shared_pending,
565                                          mask, info);
566         if (signr && unlikely(sig_kernel_stop(signr))) {
567                 /*
568                  * Set a marker that we have dequeued a stop signal.  Our
569                  * caller might release the siglock and then the pending
570                  * stop signal it is about to process is no longer in the
571                  * pending bitmasks, but must still be cleared by a SIGCONT
572                  * (and overruled by a SIGKILL).  So those cases clear this
573                  * shared flag after we've set it.  Note that this flag may
574                  * remain set after the signal we return is ignored or
575                  * handled.  That doesn't matter because its only purpose
576                  * is to alert stop-signal processing code when another
577                  * processor has come along and cleared the flag.
578                  */
579                 tsk->signal->flags |= SIGNAL_STOP_DEQUEUED;
580         }
581         if ( signr &&
582              ((info->si_code & __SI_MASK) == __SI_TIMER) &&
583              info->si_sys_private){
584                 /*
585                  * Release the siglock to ensure proper locking order
586                  * of timer locks outside of siglocks.  Note, we leave
587                  * irqs disabled here, since the posix-timers code is
588                  * about to disable them again anyway.
589                  */
590                 spin_unlock(&tsk->sighand->siglock);
591                 do_schedule_next_timer(info);
592                 spin_lock(&tsk->sighand->siglock);
593         }
594         return signr;
595 }
596
597 /*
598  * Tell a process that it has a new active signal..
599  *
600  * NOTE! we rely on the previous spin_lock to
601  * lock interrupts for us! We can only be called with
602  * "siglock" held, and the local interrupt must
603  * have been disabled when that got acquired!
604  *
605  * No need to set need_resched since signal event passing
606  * goes through ->blocked
607  */
608 void signal_wake_up(struct task_struct *t, int resume)
609 {
610         unsigned int mask;
611
612         set_tsk_thread_flag(t, TIF_SIGPENDING);
613
614         /*
615          * For SIGKILL, we want to wake it up in the stopped/traced case.
616          * We don't check t->state here because there is a race with it
617          * executing another processor and just now entering stopped state.
618          * By using wake_up_state, we ensure the process will wake up and
619          * handle its death signal.
620          */
621         mask = TASK_INTERRUPTIBLE;
622         if (resume)
623                 mask |= TASK_STOPPED | TASK_TRACED;
624         if (!wake_up_state(t, mask))
625                 kick_process(t);
626 }
627
628 /*
629  * Remove signals in mask from the pending set and queue.
630  * Returns 1 if any signals were found.
631  *
632  * All callers must be holding the siglock.
633  */
634 static int rm_from_queue(unsigned long mask, struct sigpending *s)
635 {
636         struct sigqueue *q, *n;
637
638         if (!sigtestsetmask(&s->signal, mask))
639                 return 0;
640
641         sigdelsetmask(&s->signal, mask);
642         list_for_each_entry_safe(q, n, &s->list, list) {
643                 if (q->info.si_signo < SIGRTMIN &&
644                     (mask & sigmask(q->info.si_signo))) {
645                         list_del_init(&q->list);
646                         __sigqueue_free(q);
647                 }
648         }
649         return 1;
650 }
651
652 /*
653  * Bad permissions for sending the signal
654  */
655 static int check_kill_permission(int sig, struct siginfo *info,
656                                  struct task_struct *t)
657 {
658         int error = -EINVAL;
659         int user;
660
661         if (!valid_signal(sig))
662                 return error;
663
664         user = (!info || ((unsigned long)info != 1 &&
665                 (unsigned long)info != 2 && SI_FROMUSER(info)));
666
667         error = -EPERM;
668         if (user && ((sig != SIGCONT) ||
669                 (current->signal->session != t->signal->session))
670             && (current->euid ^ t->suid) && (current->euid ^ t->uid)
671             && (current->uid ^ t->suid) && (current->uid ^ t->uid)
672             && !capable(CAP_KILL))
673                 return error;
674
675         error = -ESRCH;
676         if (user && !vx_check(vx_task_xid(t), VX_ADMIN|VX_IDENT))
677                 return error;
678
679         return security_task_kill(t, info, sig);
680 }
681
682 /* forward decl */
683 static void do_notify_parent_cldstop(struct task_struct *tsk,
684                                      struct task_struct *parent,
685                                      int why);
686
687 /*
688  * Handle magic process-wide effects of stop/continue signals.
689  * Unlike the signal actions, these happen immediately at signal-generation
690  * time regardless of blocking, ignoring, or handling.  This does the
691  * actual continuing for SIGCONT, but not the actual stopping for stop
692  * signals.  The process stop is done as a signal action for SIG_DFL.
693  */
694 static void handle_stop_signal(int sig, struct task_struct *p)
695 {
696         struct task_struct *t;
697
698         if (p->flags & SIGNAL_GROUP_EXIT)
699                 /*
700                  * The process is in the middle of dying already.
701                  */
702                 return;
703
704         if (sig_kernel_stop(sig)) {
705                 /*
706                  * This is a stop signal.  Remove SIGCONT from all queues.
707                  */
708                 rm_from_queue(sigmask(SIGCONT), &p->signal->shared_pending);
709                 t = p;
710                 do {
711                         rm_from_queue(sigmask(SIGCONT), &t->pending);
712                         t = next_thread(t);
713                 } while (t != p);
714         } else if (sig == SIGCONT) {
715                 /*
716                  * Remove all stop signals from all queues,
717                  * and wake all threads.
718                  */
719                 if (unlikely(p->signal->group_stop_count > 0)) {
720                         /*
721                          * There was a group stop in progress.  We'll
722                          * pretend it finished before we got here.  We are
723                          * obliged to report it to the parent: if the
724                          * SIGSTOP happened "after" this SIGCONT, then it
725                          * would have cleared this pending SIGCONT.  If it
726                          * happened "before" this SIGCONT, then the parent
727                          * got the SIGCHLD about the stop finishing before
728                          * the continue happened.  We do the notification
729                          * now, and it's as if the stop had finished and
730                          * the SIGCHLD was pending on entry to this kill.
731                          */
732                         p->signal->group_stop_count = 0;
733                         p->signal->flags = SIGNAL_STOP_CONTINUED;
734                         spin_unlock(&p->sighand->siglock);
735                         if (p->ptrace & PT_PTRACED)
736                                 do_notify_parent_cldstop(p, p->parent,
737                                                          CLD_STOPPED);
738                         else
739                                 do_notify_parent_cldstop(
740                                         p->group_leader,
741                                         p->group_leader->real_parent,
742                                                          CLD_STOPPED);
743                         spin_lock(&p->sighand->siglock);
744                 }
745                 rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
746                 t = p;
747                 do {
748                         unsigned int state;
749                         rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
750                         
751                         /*
752                          * If there is a handler for SIGCONT, we must make
753                          * sure that no thread returns to user mode before
754                          * we post the signal, in case it was the only
755                          * thread eligible to run the signal handler--then
756                          * it must not do anything between resuming and
757                          * running the handler.  With the TIF_SIGPENDING
758                          * flag set, the thread will pause and acquire the
759                          * siglock that we hold now and until we've queued
760                          * the pending signal. 
761                          *
762                          * Wake up the stopped thread _after_ setting
763                          * TIF_SIGPENDING
764                          */
765                         state = TASK_STOPPED;
766                         if (sig_user_defined(t, SIGCONT) && !sigismember(&t->blocked, SIGCONT)) {
767                                 set_tsk_thread_flag(t, TIF_SIGPENDING);
768                                 state |= TASK_INTERRUPTIBLE;
769                         }
770                         wake_up_state(t, state);
771
772                         t = next_thread(t);
773                 } while (t != p);
774
775                 if (p->signal->flags & SIGNAL_STOP_STOPPED) {
776                         /*
777                          * We were in fact stopped, and are now continued.
778                          * Notify the parent with CLD_CONTINUED.
779                          */
780                         p->signal->flags = SIGNAL_STOP_CONTINUED;
781                         p->signal->group_exit_code = 0;
782                         spin_unlock(&p->sighand->siglock);
783                         if (p->ptrace & PT_PTRACED)
784                                 do_notify_parent_cldstop(p, p->parent,
785                                                          CLD_CONTINUED);
786                         else
787                                 do_notify_parent_cldstop(
788                                         p->group_leader,
789                                         p->group_leader->real_parent,
790                                                          CLD_CONTINUED);
791                         spin_lock(&p->sighand->siglock);
792                 } else {
793                         /*
794                          * We are not stopped, but there could be a stop
795                          * signal in the middle of being processed after
796                          * being removed from the queue.  Clear that too.
797                          */
798                         p->signal->flags = 0;
799                 }
800         } else if (sig == SIGKILL) {
801                 /*
802                  * Make sure that any pending stop signal already dequeued
803                  * is undone by the wakeup for SIGKILL.
804                  */
805                 p->signal->flags = 0;
806         }
807 }
808
809 static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
810                         struct sigpending *signals)
811 {
812         struct sigqueue * q = NULL;
813         int ret = 0;
814
815         /*
816          * fast-pathed signals for kernel-internal things like SIGSTOP
817          * or SIGKILL.
818          */
819         if ((unsigned long)info == 2)
820                 goto out_set;
821
822         /* Real-time signals must be queued if sent by sigqueue, or
823            some other real-time mechanism.  It is implementation
824            defined whether kill() does so.  We attempt to do so, on
825            the principle of least surprise, but since kill is not
826            allowed to fail with EAGAIN when low on memory we just
827            make sure at least one signal gets delivered and don't
828            pass on the info struct.  */
829
830         q = __sigqueue_alloc(t, GFP_ATOMIC, (sig < SIGRTMIN &&
831                                              ((unsigned long) info < 2 ||
832                                               info->si_code >= 0)));
833         if (q) {
834                 list_add_tail(&q->list, &signals->list);
835                 switch ((unsigned long) info) {
836                 case 0:
837                         q->info.si_signo = sig;
838                         q->info.si_errno = 0;
839                         q->info.si_code = SI_USER;
840                         q->info.si_pid = current->pid;
841                         q->info.si_uid = current->uid;
842                         break;
843                 case 1:
844                         q->info.si_signo = sig;
845                         q->info.si_errno = 0;
846                         q->info.si_code = SI_KERNEL;
847                         q->info.si_pid = 0;
848                         q->info.si_uid = 0;
849                         break;
850                 default:
851                         copy_siginfo(&q->info, info);
852                         break;
853                 }
854         } else {
855                 if (sig >= SIGRTMIN && info && (unsigned long)info != 1
856                    && info->si_code != SI_USER)
857                 /*
858                  * Queue overflow, abort.  We may abort if the signal was rt
859                  * and sent by user using something other than kill().
860                  */
861                         return -EAGAIN;
862                 if (((unsigned long)info > 1) && (info->si_code == SI_TIMER))
863                         /*
864                          * Set up a return to indicate that we dropped 
865                          * the signal.
866                          */
867                         ret = info->si_sys_private;
868         }
869
870 out_set:
871         sigaddset(&signals->signal, sig);
872         return ret;
873 }
874
875 #define LEGACY_QUEUE(sigptr, sig) \
876         (((sig) < SIGRTMIN) && sigismember(&(sigptr)->signal, (sig)))
877
878
879 static int
880 specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
881 {
882         int ret = 0;
883
884         if (!irqs_disabled())
885                 BUG();
886         assert_spin_locked(&t->sighand->siglock);
887
888         if (((unsigned long)info > 2) && (info->si_code == SI_TIMER))
889                 /*
890                  * Set up a return to indicate that we dropped the signal.
891                  */
892                 ret = info->si_sys_private;
893
894         /* Short-circuit ignored signals.  */
895         if (sig_ignored(t, sig))
896                 goto out;
897
898         /* Support queueing exactly one non-rt signal, so that we
899            can get more detailed information about the cause of
900            the signal. */
901         if (LEGACY_QUEUE(&t->pending, sig))
902                 goto out;
903
904         ret = send_signal(sig, info, t, &t->pending);
905         if (!ret && !sigismember(&t->blocked, sig))
906                 signal_wake_up(t, sig == SIGKILL);
907 out:
908         return ret;
909 }
910
911 /*
912  * Force a signal that the process can't ignore: if necessary
913  * we unblock the signal and change any SIG_IGN to SIG_DFL.
914  */
915
916 int
917 force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
918 {
919         unsigned long int flags;
920         int ret;
921
922         spin_lock_irqsave(&t->sighand->siglock, flags);
923         if (sigismember(&t->blocked, sig) || 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         }
928         ret = specific_send_sig_info(sig, info, t);
929         spin_unlock_irqrestore(&t->sighand->siglock, flags);
930
931         return ret;
932 }
933
934 void
935 force_sig_specific(int sig, struct task_struct *t)
936 {
937         unsigned long int flags;
938
939         spin_lock_irqsave(&t->sighand->siglock, flags);
940         if (t->sighand->action[sig-1].sa.sa_handler == SIG_IGN)
941                 t->sighand->action[sig-1].sa.sa_handler = SIG_DFL;
942         sigdelset(&t->blocked, sig);
943         recalc_sigpending_tsk(t);
944         specific_send_sig_info(sig, (void *)2, t);
945         spin_unlock_irqrestore(&t->sighand->siglock, flags);
946 }
947
948 /*
949  * Test if P wants to take SIG.  After we've checked all threads with this,
950  * it's equivalent to finding no threads not blocking SIG.  Any threads not
951  * blocking SIG were ruled out because they are not running and already
952  * have pending signals.  Such threads will dequeue from the shared queue
953  * as soon as they're available, so putting the signal on the shared queue
954  * will be equivalent to sending it to one such thread.
955  */
956 #define wants_signal(sig, p, mask)                      \
957         (!sigismember(&(p)->blocked, sig)               \
958          && !((p)->state & mask)                        \
959          && !((p)->flags & PF_EXITING)                  \
960          && (task_curr(p) || !signal_pending(p)))
961
962
963 static void
964 __group_complete_signal(int sig, struct task_struct *p)
965 {
966         unsigned int mask;
967         struct task_struct *t;
968
969         /*
970          * Don't bother traced and stopped tasks (but
971          * SIGKILL will punch through that).
972          */
973         mask = TASK_STOPPED | TASK_TRACED;
974         if (sig == SIGKILL)
975                 mask = 0;
976
977         /*
978          * Now find a thread we can wake up to take the signal off the queue.
979          *
980          * If the main thread wants the signal, it gets first crack.
981          * Probably the least surprising to the average bear.
982          */
983         if (wants_signal(sig, p, mask))
984                 t = p;
985         else if (thread_group_empty(p))
986                 /*
987                  * There is just one thread and it does not need to be woken.
988                  * It will dequeue unblocked signals before it runs again.
989                  */
990                 return;
991         else {
992                 /*
993                  * Otherwise try to find a suitable thread.
994                  */
995                 t = p->signal->curr_target;
996                 if (t == NULL)
997                         /* restart balancing at this thread */
998                         t = p->signal->curr_target = p;
999                 BUG_ON(t->tgid != p->tgid);
1000
1001                 while (!wants_signal(sig, t, mask)) {
1002                         t = next_thread(t);
1003                         if (t == p->signal->curr_target)
1004                                 /*
1005                                  * No thread needs to be woken.
1006                                  * Any eligible threads will see
1007                                  * the signal in the queue soon.
1008                                  */
1009                                 return;
1010                 }
1011                 p->signal->curr_target = t;
1012         }
1013
1014         /*
1015          * Found a killable thread.  If the signal will be fatal,
1016          * then start taking the whole group down immediately.
1017          */
1018         if (sig_fatal(p, sig) && !(p->signal->flags & SIGNAL_GROUP_EXIT) &&
1019             !sigismember(&t->real_blocked, sig) &&
1020             (sig == SIGKILL || !(t->ptrace & PT_PTRACED))) {
1021                 /*
1022                  * This signal will be fatal to the whole group.
1023                  */
1024                 if (!sig_kernel_coredump(sig)) {
1025                         /*
1026                          * Start a group exit and wake everybody up.
1027                          * This way we don't have other threads
1028                          * running and doing things after a slower
1029                          * thread has the fatal signal pending.
1030                          */
1031                         p->signal->flags = SIGNAL_GROUP_EXIT;
1032                         p->signal->group_exit_code = sig;
1033                         p->signal->group_stop_count = 0;
1034                         t = p;
1035                         do {
1036                                 sigaddset(&t->pending.signal, SIGKILL);
1037                                 signal_wake_up(t, 1);
1038                                 t = next_thread(t);
1039                         } while (t != p);
1040                         return;
1041                 }
1042
1043                 /*
1044                  * There will be a core dump.  We make all threads other
1045                  * than the chosen one go into a group stop so that nothing
1046                  * happens until it gets scheduled, takes the signal off
1047                  * the shared queue, and does the core dump.  This is a
1048                  * little more complicated than strictly necessary, but it
1049                  * keeps the signal state that winds up in the core dump
1050                  * unchanged from the death state, e.g. which thread had
1051                  * the core-dump signal unblocked.
1052                  */
1053                 rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
1054                 rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
1055                 p->signal->group_stop_count = 0;
1056                 p->signal->group_exit_task = t;
1057                 t = p;
1058                 do {
1059                         p->signal->group_stop_count++;
1060                         signal_wake_up(t, 0);
1061                         t = next_thread(t);
1062                 } while (t != p);
1063                 wake_up_process(p->signal->group_exit_task);
1064                 return;
1065         }
1066
1067         /*
1068          * The signal is already in the shared-pending queue.
1069          * Tell the chosen thread to wake up and dequeue it.
1070          */
1071         signal_wake_up(t, sig == SIGKILL);
1072         return;
1073 }
1074
1075 int
1076 __group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1077 {
1078         int ret = 0;
1079
1080         assert_spin_locked(&p->sighand->siglock);
1081         handle_stop_signal(sig, p);
1082
1083         if (((unsigned long)info > 2) && (info->si_code == SI_TIMER))
1084                 /*
1085                  * Set up a return to indicate that we dropped the signal.
1086                  */
1087                 ret = info->si_sys_private;
1088
1089         /* Short-circuit ignored signals.  */
1090         if (sig_ignored(p, sig))
1091                 return ret;
1092
1093         if (LEGACY_QUEUE(&p->signal->shared_pending, sig))
1094                 /* This is a non-RT signal and we already have one queued.  */
1095                 return ret;
1096
1097         /*
1098          * Put this signal on the shared-pending queue, or fail with EAGAIN.
1099          * We always use the shared queue for process-wide signals,
1100          * to avoid several races.
1101          */
1102         ret = send_signal(sig, info, p, &p->signal->shared_pending);
1103         if (unlikely(ret))
1104                 return ret;
1105
1106         __group_complete_signal(sig, p);
1107         return 0;
1108 }
1109
1110 /*
1111  * Nuke all other threads in the group.
1112  */
1113 void zap_other_threads(struct task_struct *p)
1114 {
1115         struct task_struct *t;
1116
1117         p->signal->flags = SIGNAL_GROUP_EXIT;
1118         p->signal->group_stop_count = 0;
1119
1120         if (thread_group_empty(p))
1121                 return;
1122
1123         for (t = next_thread(p); t != p; t = next_thread(t)) {
1124                 /*
1125                  * Don't bother with already dead threads
1126                  */
1127                 if (t->exit_state)
1128                         continue;
1129
1130                 /*
1131                  * We don't want to notify the parent, since we are
1132                  * killed as part of a thread group due to another
1133                  * thread doing an execve() or similar. So set the
1134                  * exit signal to -1 to allow immediate reaping of
1135                  * the process.  But don't detach the thread group
1136                  * leader.
1137                  */
1138                 if (t != p->group_leader)
1139                         t->exit_signal = -1;
1140
1141                 sigaddset(&t->pending.signal, SIGKILL);
1142                 rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
1143                 signal_wake_up(t, 1);
1144         }
1145 }
1146
1147 /*
1148  * Must be called with the tasklist_lock held for reading!
1149  */
1150 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1151 {
1152         unsigned long flags;
1153         int ret;
1154
1155         ret = check_kill_permission(sig, info, p);
1156         if (!ret && sig && p->sighand) {
1157                 spin_lock_irqsave(&p->sighand->siglock, flags);
1158                 ret = __group_send_sig_info(sig, info, p);
1159                 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1160         }
1161
1162         return ret;
1163 }
1164
1165 /*
1166  * kill_pg_info() sends a signal to a process group: this is what the tty
1167  * control characters do (^C, ^Z etc)
1168  */
1169
1170 int __kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)
1171 {
1172         struct task_struct *p = NULL;
1173         int retval, success;
1174
1175         if (pgrp <= 0)
1176                 return -EINVAL;
1177
1178         success = 0;
1179         retval = -ESRCH;
1180         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
1181                 int err = group_send_sig_info(sig, info, p);
1182                 success |= !err;
1183                 retval = err;
1184         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
1185         return success ? 0 : retval;
1186 }
1187
1188 int
1189 kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)
1190 {
1191         int retval;
1192
1193         read_lock(&tasklist_lock);
1194         retval = __kill_pg_info(sig, info, pgrp);
1195         read_unlock(&tasklist_lock);
1196
1197         return retval;
1198 }
1199
1200 int
1201 kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1202 {
1203         int error;
1204         struct task_struct *p;
1205
1206         read_lock(&tasklist_lock);
1207         p = find_task_by_pid(pid);
1208         error = -ESRCH;
1209         if (p)
1210                 error = group_send_sig_info(sig, info, p);
1211         read_unlock(&tasklist_lock);
1212         return error;
1213 }
1214
1215
1216 /*
1217  * kill_something_info() interprets pid in interesting ways just like kill(2).
1218  *
1219  * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1220  * is probably wrong.  Should make it like BSD or SYSV.
1221  */
1222
1223 static int kill_something_info(int sig, struct siginfo *info, int pid)
1224 {
1225         if (!pid) {
1226                 return kill_pg_info(sig, info, process_group(current));
1227         } else if (pid == -1) {
1228                 int retval = 0, count = 0;
1229                 struct task_struct * p;
1230
1231                 read_lock(&tasklist_lock);
1232                 for_each_process(p) {
1233                         if (p->pid > 1 && p->tgid != current->tgid) {
1234                                 int err = group_send_sig_info(sig, info, p);
1235                                 ++count;
1236                                 if (err != -EPERM)
1237                                         retval = err;
1238                         }
1239                 }
1240                 read_unlock(&tasklist_lock);
1241                 return count ? retval : -ESRCH;
1242         } else if (pid < 0) {
1243                 return kill_pg_info(sig, info, -pid);
1244         } else {
1245                 return kill_proc_info(sig, info, pid);
1246         }
1247 }
1248
1249 /*
1250  * These are for backward compatibility with the rest of the kernel source.
1251  */
1252
1253 /*
1254  * These two are the most common entry points.  They send a signal
1255  * just to the specific thread.
1256  */
1257 int
1258 send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1259 {
1260         int ret;
1261         unsigned long flags;
1262
1263         /*
1264          * Make sure legacy kernel users don't send in bad values
1265          * (normal paths check this in check_kill_permission).
1266          */
1267         if (!valid_signal(sig))
1268                 return -EINVAL;
1269
1270         /*
1271          * We need the tasklist lock even for the specific
1272          * thread case (when we don't need to follow the group
1273          * lists) in order to avoid races with "p->sighand"
1274          * going away or changing from under us.
1275          */
1276         read_lock(&tasklist_lock);  
1277         spin_lock_irqsave(&p->sighand->siglock, flags);
1278         ret = specific_send_sig_info(sig, info, p);
1279         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1280         read_unlock(&tasklist_lock);
1281         return ret;
1282 }
1283
1284 int
1285 send_sig(int sig, struct task_struct *p, int priv)
1286 {
1287         return send_sig_info(sig, (void*)(long)(priv != 0), p);
1288 }
1289
1290 /*
1291  * This is the entry point for "process-wide" signals.
1292  * They will go to an appropriate thread in the thread group.
1293  */
1294 int
1295 send_group_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1296 {
1297         int ret;
1298         read_lock(&tasklist_lock);
1299         ret = group_send_sig_info(sig, info, p);
1300         read_unlock(&tasklist_lock);
1301         return ret;
1302 }
1303
1304 void
1305 force_sig(int sig, struct task_struct *p)
1306 {
1307         force_sig_info(sig, (void*)1L, p);
1308 }
1309
1310 /*
1311  * When things go south during signal handling, we
1312  * will force a SIGSEGV. And if the signal that caused
1313  * the problem was already a SIGSEGV, we'll want to
1314  * make sure we don't even try to deliver the signal..
1315  */
1316 int
1317 force_sigsegv(int sig, struct task_struct *p)
1318 {
1319         if (sig == SIGSEGV) {
1320                 unsigned long flags;
1321                 spin_lock_irqsave(&p->sighand->siglock, flags);
1322                 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1323                 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1324         }
1325         force_sig(SIGSEGV, p);
1326         return 0;
1327 }
1328
1329 int
1330 kill_pg(pid_t pgrp, int sig, int priv)
1331 {
1332         return kill_pg_info(sig, (void *)(long)(priv != 0), pgrp);
1333 }
1334
1335 int
1336 kill_proc(pid_t pid, int sig, int priv)
1337 {
1338         return kill_proc_info(sig, (void *)(long)(priv != 0), pid);
1339 }
1340
1341 /*
1342  * These functions support sending signals using preallocated sigqueue
1343  * structures.  This is needed "because realtime applications cannot
1344  * afford to lose notifications of asynchronous events, like timer
1345  * expirations or I/O completions".  In the case of Posix Timers 
1346  * we allocate the sigqueue structure from the timer_create.  If this
1347  * allocation fails we are able to report the failure to the application
1348  * with an EAGAIN error.
1349  */
1350  
1351 struct sigqueue *sigqueue_alloc(void)
1352 {
1353         struct sigqueue *q;
1354
1355         if ((q = __sigqueue_alloc(current, GFP_KERNEL, 0)))
1356                 q->flags |= SIGQUEUE_PREALLOC;
1357         return(q);
1358 }
1359
1360 void sigqueue_free(struct sigqueue *q)
1361 {
1362         unsigned long flags;
1363         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1364         /*
1365          * If the signal is still pending remove it from the
1366          * pending queue.
1367          */
1368         if (unlikely(!list_empty(&q->list))) {
1369                 read_lock(&tasklist_lock);  
1370                 spin_lock_irqsave(q->lock, flags);
1371                 if (!list_empty(&q->list))
1372                         list_del_init(&q->list);
1373                 spin_unlock_irqrestore(q->lock, flags);
1374                 read_unlock(&tasklist_lock);
1375         }
1376         q->flags &= ~SIGQUEUE_PREALLOC;
1377         __sigqueue_free(q);
1378 }
1379
1380 int
1381 send_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
1382 {
1383         unsigned long flags;
1384         int ret = 0;
1385
1386         /*
1387          * We need the tasklist lock even for the specific
1388          * thread case (when we don't need to follow the group
1389          * lists) in order to avoid races with "p->sighand"
1390          * going away or changing from under us.
1391          */
1392         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1393         read_lock(&tasklist_lock);  
1394         spin_lock_irqsave(&p->sighand->siglock, flags);
1395         
1396         if (unlikely(!list_empty(&q->list))) {
1397                 /*
1398                  * If an SI_TIMER entry is already queue just increment
1399                  * the overrun count.
1400                  */
1401                 if (q->info.si_code != SI_TIMER)
1402                         BUG();
1403                 q->info.si_overrun++;
1404                 goto out;
1405         } 
1406         /* Short-circuit ignored signals.  */
1407         if (sig_ignored(p, sig)) {
1408                 ret = 1;
1409                 goto out;
1410         }
1411
1412         q->lock = &p->sighand->siglock;
1413         list_add_tail(&q->list, &p->pending.list);
1414         sigaddset(&p->pending.signal, sig);
1415         if (!sigismember(&p->blocked, sig))
1416                 signal_wake_up(p, sig == SIGKILL);
1417
1418 out:
1419         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1420         read_unlock(&tasklist_lock);
1421         return(ret);
1422 }
1423
1424 int
1425 send_group_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
1426 {
1427         unsigned long flags;
1428         int ret = 0;
1429
1430         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1431         read_lock(&tasklist_lock);
1432         spin_lock_irqsave(&p->sighand->siglock, flags);
1433         handle_stop_signal(sig, p);
1434
1435         /* Short-circuit ignored signals.  */
1436         if (sig_ignored(p, sig)) {
1437                 ret = 1;
1438                 goto out;
1439         }
1440
1441         if (unlikely(!list_empty(&q->list))) {
1442                 /*
1443                  * If an SI_TIMER entry is already queue just increment
1444                  * the overrun count.  Other uses should not try to
1445                  * send the signal multiple times.
1446                  */
1447                 if (q->info.si_code != SI_TIMER)
1448                         BUG();
1449                 q->info.si_overrun++;
1450                 goto out;
1451         } 
1452
1453         /*
1454          * Put this signal on the shared-pending queue.
1455          * We always use the shared queue for process-wide signals,
1456          * to avoid several races.
1457          */
1458         q->lock = &p->sighand->siglock;
1459         list_add_tail(&q->list, &p->signal->shared_pending.list);
1460         sigaddset(&p->signal->shared_pending.signal, sig);
1461
1462         __group_complete_signal(sig, p);
1463 out:
1464         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1465         read_unlock(&tasklist_lock);
1466         return(ret);
1467 }
1468
1469 /*
1470  * Wake up any threads in the parent blocked in wait* syscalls.
1471  */
1472 static inline void __wake_up_parent(struct task_struct *p,
1473                                     struct task_struct *parent)
1474 {
1475         wake_up_interruptible_sync(&parent->signal->wait_chldexit);
1476 }
1477
1478 /*
1479  * Let a parent know about the death of a child.
1480  * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1481  */
1482
1483 void do_notify_parent(struct task_struct *tsk, int sig)
1484 {
1485         struct siginfo info;
1486         unsigned long flags;
1487         struct sighand_struct *psig;
1488
1489         BUG_ON(sig == -1);
1490
1491         /* do_notify_parent_cldstop should have been called instead.  */
1492         BUG_ON(tsk->state & (TASK_STOPPED|TASK_TRACED));
1493
1494         BUG_ON(!tsk->ptrace &&
1495                (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1496
1497         info.si_signo = sig;
1498         info.si_errno = 0;
1499         info.si_pid = tsk->pid;
1500         info.si_uid = tsk->uid;
1501
1502         /* FIXME: find out whether or not this is supposed to be c*time. */
1503         info.si_utime = cputime_to_jiffies(cputime_add(tsk->utime,
1504                                                        tsk->signal->utime));
1505         info.si_stime = cputime_to_jiffies(cputime_add(tsk->stime,
1506                                                        tsk->signal->stime));
1507
1508         info.si_status = tsk->exit_code & 0x7f;
1509         if (tsk->exit_code & 0x80)
1510                 info.si_code = CLD_DUMPED;
1511         else if (tsk->exit_code & 0x7f)
1512                 info.si_code = CLD_KILLED;
1513         else {
1514                 info.si_code = CLD_EXITED;
1515                 info.si_status = tsk->exit_code >> 8;
1516         }
1517
1518         psig = tsk->parent->sighand;
1519         spin_lock_irqsave(&psig->siglock, flags);
1520         if (sig == SIGCHLD &&
1521             (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1522              (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1523                 /*
1524                  * We are exiting and our parent doesn't care.  POSIX.1
1525                  * defines special semantics for setting SIGCHLD to SIG_IGN
1526                  * or setting the SA_NOCLDWAIT flag: we should be reaped
1527                  * automatically and not left for our parent's wait4 call.
1528                  * Rather than having the parent do it as a magic kind of
1529                  * signal handler, we just set this to tell do_exit that we
1530                  * can be cleaned up without becoming a zombie.  Note that
1531                  * we still call __wake_up_parent in this case, because a
1532                  * blocked sys_wait4 might now return -ECHILD.
1533                  *
1534                  * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1535                  * is implementation-defined: we do (if you don't want
1536                  * it, just use SIG_IGN instead).
1537                  */
1538                 tsk->exit_signal = -1;
1539                 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1540                         sig = 0;
1541         }
1542         if (valid_signal(sig) && sig > 0)
1543                 __group_send_sig_info(sig, &info, tsk->parent);
1544         __wake_up_parent(tsk, tsk->parent);
1545         spin_unlock_irqrestore(&psig->siglock, flags);
1546 }
1547
1548 static void
1549 do_notify_parent_cldstop(struct task_struct *tsk, struct task_struct *parent,
1550                          int why)
1551 {
1552         struct siginfo info;
1553         unsigned long flags;
1554         struct sighand_struct *sighand;
1555
1556         info.si_signo = SIGCHLD;
1557         info.si_errno = 0;
1558         info.si_pid = tsk->pid;
1559         info.si_uid = tsk->uid;
1560
1561         /* FIXME: find out whether or not this is supposed to be c*time. */
1562         info.si_utime = cputime_to_jiffies(tsk->utime);
1563         info.si_stime = cputime_to_jiffies(tsk->stime);
1564
1565         info.si_code = why;
1566         switch (why) {
1567         case CLD_CONTINUED:
1568                 info.si_status = SIGCONT;
1569                 break;
1570         case CLD_STOPPED:
1571                 info.si_status = tsk->signal->group_exit_code & 0x7f;
1572                 break;
1573         case CLD_TRAPPED:
1574                 info.si_status = tsk->exit_code & 0x7f;
1575                 break;
1576         default:
1577                 BUG();
1578         }
1579
1580         sighand = parent->sighand;
1581         spin_lock_irqsave(&sighand->siglock, flags);
1582         if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1583             !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1584                 __group_send_sig_info(SIGCHLD, &info, parent);
1585         /*
1586          * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1587          */
1588         __wake_up_parent(tsk, parent);
1589         spin_unlock_irqrestore(&sighand->siglock, flags);
1590 }
1591
1592 /*
1593  * This must be called with current->sighand->siglock held.
1594  *
1595  * This should be the path for all ptrace stops.
1596  * We always set current->last_siginfo while stopped here.
1597  * That makes it a way to test a stopped process for
1598  * being ptrace-stopped vs being job-control-stopped.
1599  *
1600  * If we actually decide not to stop at all because the tracer is gone,
1601  * we leave nostop_code in current->exit_code.
1602  */
1603 static void ptrace_stop(int exit_code, int nostop_code, siginfo_t *info)
1604 {
1605         /*
1606          * If there is a group stop in progress,
1607          * we must participate in the bookkeeping.
1608          */
1609         if (current->signal->group_stop_count > 0)
1610                 --current->signal->group_stop_count;
1611
1612         current->last_siginfo = info;
1613         current->exit_code = exit_code;
1614
1615         /* Let the debugger run.  */
1616         set_current_state(TASK_TRACED);
1617         spin_unlock_irq(&current->sighand->siglock);
1618         read_lock(&tasklist_lock);
1619         if (likely(current->ptrace & PT_PTRACED) &&
1620             likely(current->parent != current->real_parent ||
1621                    !(current->ptrace & PT_ATTACHED)) &&
1622             (likely(current->parent->signal != current->signal) ||
1623              !unlikely(current->signal->flags & SIGNAL_GROUP_EXIT))) {
1624                 do_notify_parent_cldstop(current, current->parent,
1625                                          CLD_TRAPPED);
1626                 read_unlock(&tasklist_lock);
1627                 schedule();
1628         } else {
1629                 /*
1630                  * By the time we got the lock, our tracer went away.
1631                  * Don't stop here.
1632                  */
1633                 read_unlock(&tasklist_lock);
1634                 set_current_state(TASK_RUNNING);
1635                 current->exit_code = nostop_code;
1636         }
1637
1638         /*
1639          * We are back.  Now reacquire the siglock before touching
1640          * last_siginfo, so that we are sure to have synchronized with
1641          * any signal-sending on another CPU that wants to examine it.
1642          */
1643         spin_lock_irq(&current->sighand->siglock);
1644         current->last_siginfo = NULL;
1645
1646         /*
1647          * Queued signals ignored us while we were stopped for tracing.
1648          * So check for any that we should take before resuming user mode.
1649          */
1650         recalc_sigpending();
1651 }
1652
1653 void ptrace_notify(int exit_code)
1654 {
1655         siginfo_t info;
1656
1657         BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1658
1659         memset(&info, 0, sizeof info);
1660         info.si_signo = SIGTRAP;
1661         info.si_code = exit_code;
1662         info.si_pid = current->pid;
1663         info.si_uid = current->uid;
1664
1665         /* Let the debugger run.  */
1666         spin_lock_irq(&current->sighand->siglock);
1667         ptrace_stop(exit_code, 0, &info);
1668         spin_unlock_irq(&current->sighand->siglock);
1669 }
1670
1671 static void
1672 finish_stop(int stop_count)
1673 {
1674         /*
1675          * If there are no other threads in the group, or if there is
1676          * a group stop in progress and we are the last to stop,
1677          * report to the parent.  When ptraced, every thread reports itself.
1678          */
1679         if (stop_count < 0 || (current->ptrace & PT_PTRACED)) {
1680                 read_lock(&tasklist_lock);
1681                 do_notify_parent_cldstop(current, current->parent,
1682                                          CLD_STOPPED);
1683                 read_unlock(&tasklist_lock);
1684         }
1685         else if (stop_count == 0) {
1686                 read_lock(&tasklist_lock);
1687                 do_notify_parent_cldstop(current->group_leader,
1688                                          current->group_leader->real_parent,
1689                                          CLD_STOPPED);
1690                 read_unlock(&tasklist_lock);
1691         }
1692
1693         schedule();
1694         /*
1695          * Now we don't run again until continued.
1696          */
1697         current->exit_code = 0;
1698 }
1699
1700 /*
1701  * This performs the stopping for SIGSTOP and other stop signals.
1702  * We have to stop all threads in the thread group.
1703  * Returns nonzero if we've actually stopped and released the siglock.
1704  * Returns zero if we didn't stop and still hold the siglock.
1705  */
1706 static int
1707 do_signal_stop(int signr)
1708 {
1709         struct signal_struct *sig = current->signal;
1710         struct sighand_struct *sighand = current->sighand;
1711         int stop_count = -1;
1712
1713         if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED))
1714                 return 0;
1715
1716         if (sig->group_stop_count > 0) {
1717                 /*
1718                  * There is a group stop in progress.  We don't need to
1719                  * start another one.
1720                  */
1721                 signr = sig->group_exit_code;
1722                 stop_count = --sig->group_stop_count;
1723                 current->exit_code = signr;
1724                 set_current_state(TASK_STOPPED);
1725                 if (stop_count == 0)
1726                         sig->flags = SIGNAL_STOP_STOPPED;
1727                 spin_unlock_irq(&sighand->siglock);
1728         }
1729         else if (thread_group_empty(current)) {
1730                 /*
1731                  * Lock must be held through transition to stopped state.
1732                  */
1733                 current->exit_code = current->signal->group_exit_code = signr;
1734                 set_current_state(TASK_STOPPED);
1735                 sig->flags = SIGNAL_STOP_STOPPED;
1736                 spin_unlock_irq(&sighand->siglock);
1737         }
1738         else {
1739                 /*
1740                  * There is no group stop already in progress.
1741                  * We must initiate one now, but that requires
1742                  * dropping siglock to get both the tasklist lock
1743                  * and siglock again in the proper order.  Note that
1744                  * this allows an intervening SIGCONT to be posted.
1745                  * We need to check for that and bail out if necessary.
1746                  */
1747                 struct task_struct *t;
1748
1749                 spin_unlock_irq(&sighand->siglock);
1750
1751                 /* signals can be posted during this window */
1752
1753                 read_lock(&tasklist_lock);
1754                 spin_lock_irq(&sighand->siglock);
1755
1756                 if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED)) {
1757                         /*
1758                          * Another stop or continue happened while we
1759                          * didn't have the lock.  We can just swallow this
1760                          * signal now.  If we raced with a SIGCONT, that
1761                          * should have just cleared it now.  If we raced
1762                          * with another processor delivering a stop signal,
1763                          * then the SIGCONT that wakes us up should clear it.
1764                          */
1765                         read_unlock(&tasklist_lock);
1766                         return 0;
1767                 }
1768
1769                 if (sig->group_stop_count == 0) {
1770                         sig->group_exit_code = signr;
1771                         stop_count = 0;
1772                         for (t = next_thread(current); t != current;
1773                              t = next_thread(t))
1774                                 /*
1775                                  * Setting state to TASK_STOPPED for a group
1776                                  * stop is always done with the siglock held,
1777                                  * so this check has no races.
1778                                  */
1779                                 if (t->state < TASK_STOPPED) {
1780                                         stop_count++;
1781                                         signal_wake_up(t, 0);
1782                                 }
1783                         sig->group_stop_count = stop_count;
1784                 }
1785                 else {
1786                         /* A race with another thread while unlocked.  */
1787                         signr = sig->group_exit_code;
1788                         stop_count = --sig->group_stop_count;
1789                 }
1790
1791                 current->exit_code = signr;
1792                 set_current_state(TASK_STOPPED);
1793                 if (stop_count == 0)
1794                         sig->flags = SIGNAL_STOP_STOPPED;
1795
1796                 spin_unlock_irq(&sighand->siglock);
1797                 read_unlock(&tasklist_lock);
1798         }
1799
1800         finish_stop(stop_count);
1801         return 1;
1802 }
1803
1804 /*
1805  * Do appropriate magic when group_stop_count > 0.
1806  * We return nonzero if we stopped, after releasing the siglock.
1807  * We return zero if we still hold the siglock and should look
1808  * for another signal without checking group_stop_count again.
1809  */
1810 static inline int handle_group_stop(void)
1811 {
1812         int stop_count;
1813
1814         if (current->signal->group_exit_task == current) {
1815                 /*
1816                  * Group stop is so we can do a core dump,
1817                  * We are the initiating thread, so get on with it.
1818                  */
1819                 current->signal->group_exit_task = NULL;
1820                 return 0;
1821         }
1822
1823         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1824                 /*
1825                  * Group stop is so another thread can do a core dump,
1826                  * or else we are racing against a death signal.
1827                  * Just punt the stop so we can get the next signal.
1828                  */
1829                 return 0;
1830
1831         /*
1832          * There is a group stop in progress.  We stop
1833          * without any associated signal being in our queue.
1834          */
1835         stop_count = --current->signal->group_stop_count;
1836         if (stop_count == 0)
1837                 current->signal->flags = SIGNAL_STOP_STOPPED;
1838         current->exit_code = current->signal->group_exit_code;
1839         set_current_state(TASK_STOPPED);
1840         spin_unlock_irq(&current->sighand->siglock);
1841         finish_stop(stop_count);
1842         return 1;
1843 }
1844
1845 int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
1846                           struct pt_regs *regs, void *cookie)
1847 {
1848         sigset_t *mask = &current->blocked;
1849         int signr = 0;
1850
1851 relock:
1852         spin_lock_irq(&current->sighand->siglock);
1853         for (;;) {
1854                 struct k_sigaction *ka;
1855
1856                 if (unlikely(current->signal->group_stop_count > 0) &&
1857                     handle_group_stop())
1858                         goto relock;
1859
1860                 signr = dequeue_signal(current, mask, info);
1861
1862                 if (!signr)
1863                         break; /* will return 0 */
1864
1865                 if ((current->ptrace & PT_PTRACED) && signr != SIGKILL) {
1866                         ptrace_signal_deliver(regs, cookie);
1867
1868                         /* Let the debugger run.  */
1869                         ptrace_stop(signr, signr, info);
1870
1871                         /* We're back.  Did the debugger cancel the sig?  */
1872                         signr = current->exit_code;
1873                         if (signr == 0)
1874                                 continue;
1875
1876                         current->exit_code = 0;
1877
1878                         /* Update the siginfo structure if the signal has
1879                            changed.  If the debugger wanted something
1880                            specific in the siginfo structure then it should
1881                            have updated *info via PTRACE_SETSIGINFO.  */
1882                         if (signr != info->si_signo) {
1883                                 info->si_signo = signr;
1884                                 info->si_errno = 0;
1885                                 info->si_code = SI_USER;
1886                                 info->si_pid = current->parent->pid;
1887                                 info->si_uid = current->parent->uid;
1888                         }
1889
1890                         /* If the (new) signal is now blocked, requeue it.  */
1891                         if (sigismember(&current->blocked, signr)) {
1892                                 specific_send_sig_info(signr, info, current);
1893                                 continue;
1894                         }
1895                 }
1896
1897                 ka = &current->sighand->action[signr-1];
1898                 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
1899                         continue;
1900                 if (ka->sa.sa_handler != SIG_DFL) {
1901                         /* Run the handler.  */
1902                         *return_ka = *ka;
1903
1904                         if (ka->sa.sa_flags & SA_ONESHOT)
1905                                 ka->sa.sa_handler = SIG_DFL;
1906
1907                         break; /* will return non-zero "signr" value */
1908                 }
1909
1910                 /*
1911                  * Now we are doing the default action for this signal.
1912                  */
1913                 if (sig_kernel_ignore(signr)) /* Default is nothing. */
1914                         continue;
1915
1916                 /* Init gets no signals it doesn't want.  */
1917                 if (current->pid == 1)
1918                         continue;
1919
1920                 /* virtual init is protected against user signals */
1921                 if ((info->si_code == SI_USER) &&
1922                         vx_current_initpid(current->pid))
1923                         continue;
1924
1925                 if (sig_kernel_stop(signr)) {
1926                         /*
1927                          * The default action is to stop all threads in
1928                          * the thread group.  The job control signals
1929                          * do nothing in an orphaned pgrp, but SIGSTOP
1930                          * always works.  Note that siglock needs to be
1931                          * dropped during the call to is_orphaned_pgrp()
1932                          * because of lock ordering with tasklist_lock.
1933                          * This allows an intervening SIGCONT to be posted.
1934                          * We need to check for that and bail out if necessary.
1935                          */
1936                         if (signr != SIGSTOP) {
1937                                 spin_unlock_irq(&current->sighand->siglock);
1938
1939                                 /* signals can be posted during this window */
1940
1941                                 if (is_orphaned_pgrp(process_group(current)))
1942                                         goto relock;
1943
1944                                 spin_lock_irq(&current->sighand->siglock);
1945                         }
1946
1947                         if (likely(do_signal_stop(signr))) {
1948                                 /* It released the siglock.  */
1949                                 goto relock;
1950                         }
1951
1952                         /*
1953                          * We didn't actually stop, due to a race
1954                          * with SIGCONT or something like that.
1955                          */
1956                         continue;
1957                 }
1958
1959                 spin_unlock_irq(&current->sighand->siglock);
1960
1961                 /*
1962                  * Anything else is fatal, maybe with a core dump.
1963                  */
1964                 current->flags |= PF_SIGNALED;
1965                 if (sig_kernel_coredump(signr)) {
1966                         /*
1967                          * If it was able to dump core, this kills all
1968                          * other threads in the group and synchronizes with
1969                          * their demise.  If we lost the race with another
1970                          * thread getting here, it set group_exit_code
1971                          * first and our do_group_exit call below will use
1972                          * that value and ignore the one we pass it.
1973                          */
1974                         do_coredump((long)signr, signr, regs);
1975                 }
1976
1977                 /*
1978                  * Death signals, no core dump.
1979                  */
1980                 do_group_exit(signr);
1981                 /* NOTREACHED */
1982         }
1983         spin_unlock_irq(&current->sighand->siglock);
1984         return signr;
1985 }
1986
1987 EXPORT_SYMBOL(recalc_sigpending);
1988 EXPORT_SYMBOL_GPL(dequeue_signal);
1989 EXPORT_SYMBOL(flush_signals);
1990 EXPORT_SYMBOL(force_sig);
1991 EXPORT_SYMBOL(kill_pg);
1992 EXPORT_SYMBOL(kill_proc);
1993 EXPORT_SYMBOL(ptrace_notify);
1994 EXPORT_SYMBOL(send_sig);
1995 EXPORT_SYMBOL(send_sig_info);
1996 EXPORT_SYMBOL(sigprocmask);
1997 EXPORT_SYMBOL(block_all_signals);
1998 EXPORT_SYMBOL(unblock_all_signals);
1999
2000
2001 /*
2002  * System call entry points.
2003  */
2004
2005 asmlinkage long sys_restart_syscall(void)
2006 {
2007         struct restart_block *restart = &current_thread_info()->restart_block;
2008         return restart->fn(restart);
2009 }
2010
2011 long do_no_restart_syscall(struct restart_block *param)
2012 {
2013         return -EINTR;
2014 }
2015
2016 /*
2017  * We don't need to get the kernel lock - this is all local to this
2018  * particular thread.. (and that's good, because this is _heavily_
2019  * used by various programs)
2020  */
2021
2022 /*
2023  * This is also useful for kernel threads that want to temporarily
2024  * (or permanently) block certain signals.
2025  *
2026  * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
2027  * interface happily blocks "unblockable" signals like SIGKILL
2028  * and friends.
2029  */
2030 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2031 {
2032         int error;
2033         sigset_t old_block;
2034
2035         spin_lock_irq(&current->sighand->siglock);
2036         old_block = current->blocked;
2037         error = 0;
2038         switch (how) {
2039         case SIG_BLOCK:
2040                 sigorsets(&current->blocked, &current->blocked, set);
2041                 break;
2042         case SIG_UNBLOCK:
2043                 signandsets(&current->blocked, &current->blocked, set);
2044                 break;
2045         case SIG_SETMASK:
2046                 current->blocked = *set;
2047                 break;
2048         default:
2049                 error = -EINVAL;
2050         }
2051         recalc_sigpending();
2052         spin_unlock_irq(&current->sighand->siglock);
2053         if (oldset)
2054                 *oldset = old_block;
2055         return error;
2056 }
2057
2058 asmlinkage long
2059 sys_rt_sigprocmask(int how, sigset_t __user *set, sigset_t __user *oset, size_t sigsetsize)
2060 {
2061         int error = -EINVAL;
2062         sigset_t old_set, new_set;
2063
2064         /* XXX: Don't preclude handling different sized sigset_t's.  */
2065         if (sigsetsize != sizeof(sigset_t))
2066                 goto out;
2067
2068         if (set) {
2069                 error = -EFAULT;
2070                 if (copy_from_user(&new_set, set, sizeof(*set)))
2071                         goto out;
2072                 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2073
2074                 error = sigprocmask(how, &new_set, &old_set);
2075                 if (error)
2076                         goto out;
2077                 if (oset)
2078                         goto set_old;
2079         } else if (oset) {
2080                 spin_lock_irq(&current->sighand->siglock);
2081                 old_set = current->blocked;
2082                 spin_unlock_irq(&current->sighand->siglock);
2083
2084         set_old:
2085                 error = -EFAULT;
2086                 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2087                         goto out;
2088         }
2089         error = 0;
2090 out:
2091         return error;
2092 }
2093
2094 long do_sigpending(void __user *set, unsigned long sigsetsize)
2095 {
2096         long error = -EINVAL;
2097         sigset_t pending;
2098
2099         if (sigsetsize > sizeof(sigset_t))
2100                 goto out;
2101
2102         spin_lock_irq(&current->sighand->siglock);
2103         sigorsets(&pending, &current->pending.signal,
2104                   &current->signal->shared_pending.signal);
2105         spin_unlock_irq(&current->sighand->siglock);
2106
2107         /* Outside the lock because only this thread touches it.  */
2108         sigandsets(&pending, &current->blocked, &pending);
2109
2110         error = -EFAULT;
2111         if (!copy_to_user(set, &pending, sigsetsize))
2112                 error = 0;
2113
2114 out:
2115         return error;
2116 }       
2117
2118 asmlinkage long
2119 sys_rt_sigpending(sigset_t __user *set, size_t sigsetsize)
2120 {
2121         return do_sigpending(set, sigsetsize);
2122 }
2123
2124 #ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2125
2126 int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2127 {
2128         int err;
2129
2130         if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2131                 return -EFAULT;
2132         if (from->si_code < 0)
2133                 return __copy_to_user(to, from, sizeof(siginfo_t))
2134                         ? -EFAULT : 0;
2135         /*
2136          * If you change siginfo_t structure, please be sure
2137          * this code is fixed accordingly.
2138          * It should never copy any pad contained in the structure
2139          * to avoid security leaks, but must copy the generic
2140          * 3 ints plus the relevant union member.
2141          */
2142         err = __put_user(from->si_signo, &to->si_signo);
2143         err |= __put_user(from->si_errno, &to->si_errno);
2144         err |= __put_user((short)from->si_code, &to->si_code);
2145         switch (from->si_code & __SI_MASK) {
2146         case __SI_KILL:
2147                 err |= __put_user(from->si_pid, &to->si_pid);
2148                 err |= __put_user(from->si_uid, &to->si_uid);
2149                 break;
2150         case __SI_TIMER:
2151                  err |= __put_user(from->si_tid, &to->si_tid);
2152                  err |= __put_user(from->si_overrun, &to->si_overrun);
2153                  err |= __put_user(from->si_ptr, &to->si_ptr);
2154                 break;
2155         case __SI_POLL:
2156                 err |= __put_user(from->si_band, &to->si_band);
2157                 err |= __put_user(from->si_fd, &to->si_fd);
2158                 break;
2159         case __SI_FAULT:
2160                 err |= __put_user(from->si_addr, &to->si_addr);
2161 #ifdef __ARCH_SI_TRAPNO
2162                 err |= __put_user(from->si_trapno, &to->si_trapno);
2163 #endif
2164                 break;
2165         case __SI_CHLD:
2166                 err |= __put_user(from->si_pid, &to->si_pid);
2167                 err |= __put_user(from->si_uid, &to->si_uid);
2168                 err |= __put_user(from->si_status, &to->si_status);
2169                 err |= __put_user(from->si_utime, &to->si_utime);
2170                 err |= __put_user(from->si_stime, &to->si_stime);
2171                 break;
2172         case __SI_RT: /* This is not generated by the kernel as of now. */
2173         case __SI_MESGQ: /* But this is */
2174                 err |= __put_user(from->si_pid, &to->si_pid);
2175                 err |= __put_user(from->si_uid, &to->si_uid);
2176                 err |= __put_user(from->si_ptr, &to->si_ptr);
2177                 break;
2178         default: /* this is just in case for now ... */
2179                 err |= __put_user(from->si_pid, &to->si_pid);
2180                 err |= __put_user(from->si_uid, &to->si_uid);
2181                 break;
2182         }
2183         return err;
2184 }
2185
2186 #endif
2187
2188 asmlinkage long
2189 sys_rt_sigtimedwait(const sigset_t __user *uthese,
2190                     siginfo_t __user *uinfo,
2191                     const struct timespec __user *uts,
2192                     size_t sigsetsize)
2193 {
2194         int ret, sig;
2195         sigset_t these;
2196         struct timespec ts;
2197         siginfo_t info;
2198         long timeout = 0;
2199
2200         /* XXX: Don't preclude handling different sized sigset_t's.  */
2201         if (sigsetsize != sizeof(sigset_t))
2202                 return -EINVAL;
2203
2204         if (copy_from_user(&these, uthese, sizeof(these)))
2205                 return -EFAULT;
2206                 
2207         /*
2208          * Invert the set of allowed signals to get those we
2209          * want to block.
2210          */
2211         sigdelsetmask(&these, sigmask(SIGKILL)|sigmask(SIGSTOP));
2212         signotset(&these);
2213
2214         if (uts) {
2215                 if (copy_from_user(&ts, uts, sizeof(ts)))
2216                         return -EFAULT;
2217                 if (ts.tv_nsec >= 1000000000L || ts.tv_nsec < 0
2218                     || ts.tv_sec < 0)
2219                         return -EINVAL;
2220         }
2221
2222         spin_lock_irq(&current->sighand->siglock);
2223         sig = dequeue_signal(current, &these, &info);
2224         if (!sig) {
2225                 timeout = MAX_SCHEDULE_TIMEOUT;
2226                 if (uts)
2227                         timeout = (timespec_to_jiffies(&ts)
2228                                    + (ts.tv_sec || ts.tv_nsec));
2229
2230                 if (timeout) {
2231                         /* None ready -- temporarily unblock those we're
2232                          * interested while we are sleeping in so that we'll
2233                          * be awakened when they arrive.  */
2234                         current->real_blocked = current->blocked;
2235                         sigandsets(&current->blocked, &current->blocked, &these);
2236                         recalc_sigpending();
2237                         spin_unlock_irq(&current->sighand->siglock);
2238
2239                         current->state = TASK_INTERRUPTIBLE;
2240                         timeout = schedule_timeout(timeout);
2241
2242                         if (current->flags & PF_FREEZE)
2243                                 refrigerator(PF_FREEZE);
2244                         spin_lock_irq(&current->sighand->siglock);
2245                         sig = dequeue_signal(current, &these, &info);
2246                         current->blocked = current->real_blocked;
2247                         siginitset(&current->real_blocked, 0);
2248                         recalc_sigpending();
2249                 }
2250         }
2251         spin_unlock_irq(&current->sighand->siglock);
2252
2253         if (sig) {
2254                 ret = sig;
2255                 if (uinfo) {
2256                         if (copy_siginfo_to_user(uinfo, &info))
2257                                 ret = -EFAULT;
2258                 }
2259         } else {
2260                 ret = -EAGAIN;
2261                 if (timeout)
2262                         ret = -EINTR;
2263         }
2264
2265         return ret;
2266 }
2267
2268 asmlinkage long
2269 sys_kill(int pid, int sig)
2270 {
2271         struct siginfo info;
2272
2273         info.si_signo = sig;
2274         info.si_errno = 0;
2275         info.si_code = SI_USER;
2276         info.si_pid = current->tgid;
2277         info.si_uid = current->uid;
2278
2279         return kill_something_info(sig, &info, pid);
2280 }
2281
2282 /**
2283  *  sys_tgkill - send signal to one specific thread
2284  *  @tgid: the thread group ID of the thread
2285  *  @pid: the PID of the thread
2286  *  @sig: signal to be sent
2287  *
2288  *  This syscall also checks the tgid and returns -ESRCH even if the PID
2289  *  exists but it's not belonging to the target process anymore. This
2290  *  method solves the problem of threads exiting and PIDs getting reused.
2291  */
2292 asmlinkage long sys_tgkill(int tgid, int pid, int sig)
2293 {
2294         struct siginfo info;
2295         int error;
2296         struct task_struct *p;
2297
2298         /* This is only valid for single tasks */
2299         if (pid <= 0 || tgid <= 0)
2300                 return -EINVAL;
2301
2302         info.si_signo = sig;
2303         info.si_errno = 0;
2304         info.si_code = SI_TKILL;
2305         info.si_pid = current->tgid;
2306         info.si_uid = current->uid;
2307
2308         read_lock(&tasklist_lock);
2309         p = find_task_by_pid(pid);
2310         error = -ESRCH;
2311         if (p && (p->tgid == tgid)) {
2312                 error = check_kill_permission(sig, &info, p);
2313                 /*
2314                  * The null signal is a permissions and process existence
2315                  * probe.  No signal is actually delivered.
2316                  */
2317                 if (!error && sig && p->sighand) {
2318                         spin_lock_irq(&p->sighand->siglock);
2319                         handle_stop_signal(sig, p);
2320                         error = specific_send_sig_info(sig, &info, p);
2321                         spin_unlock_irq(&p->sighand->siglock);
2322                 }
2323         }
2324         read_unlock(&tasklist_lock);
2325         return error;
2326 }
2327
2328 /*
2329  *  Send a signal to only one task, even if it's a CLONE_THREAD task.
2330  */
2331 asmlinkage long
2332 sys_tkill(int pid, int sig)
2333 {
2334         struct siginfo info;
2335         int error;
2336         struct task_struct *p;
2337
2338         /* This is only valid for single tasks */
2339         if (pid <= 0)
2340                 return -EINVAL;
2341
2342         info.si_signo = sig;
2343         info.si_errno = 0;
2344         info.si_code = SI_TKILL;
2345         info.si_pid = current->tgid;
2346         info.si_uid = current->uid;
2347
2348         read_lock(&tasklist_lock);
2349         p = find_task_by_pid(pid);
2350         error = -ESRCH;
2351         if (p) {
2352                 error = check_kill_permission(sig, &info, p);
2353                 /*
2354                  * The null signal is a permissions and process existence
2355                  * probe.  No signal is actually delivered.
2356                  */
2357                 if (!error && sig && p->sighand) {
2358                         spin_lock_irq(&p->sighand->siglock);
2359                         handle_stop_signal(sig, p);
2360                         error = specific_send_sig_info(sig, &info, p);
2361                         spin_unlock_irq(&p->sighand->siglock);
2362                 }
2363         }
2364         read_unlock(&tasklist_lock);
2365         return error;
2366 }
2367
2368 asmlinkage long
2369 sys_rt_sigqueueinfo(int pid, int sig, siginfo_t __user *uinfo)
2370 {
2371         siginfo_t info;
2372
2373         if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2374                 return -EFAULT;
2375
2376         /* Not even root can pretend to send signals from the kernel.
2377            Nor can they impersonate a kill(), which adds source info.  */
2378         if (info.si_code >= 0)
2379                 return -EPERM;
2380         info.si_signo = sig;
2381
2382         /* POSIX.1b doesn't mention process groups.  */
2383         return kill_proc_info(sig, &info, pid);
2384 }
2385
2386 int
2387 do_sigaction(int sig, const struct k_sigaction *act, struct k_sigaction *oact)
2388 {
2389         struct k_sigaction *k;
2390
2391         if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
2392                 return -EINVAL;
2393
2394         k = &current->sighand->action[sig-1];
2395
2396         spin_lock_irq(&current->sighand->siglock);
2397         if (signal_pending(current)) {
2398                 /*
2399                  * If there might be a fatal signal pending on multiple
2400                  * threads, make sure we take it before changing the action.
2401                  */
2402                 spin_unlock_irq(&current->sighand->siglock);
2403                 return -ERESTARTNOINTR;
2404         }
2405
2406         if (oact)
2407                 *oact = *k;
2408
2409         if (act) {
2410                 /*
2411                  * POSIX 3.3.1.3:
2412                  *  "Setting a signal action to SIG_IGN for a signal that is
2413                  *   pending shall cause the pending signal to be discarded,
2414                  *   whether or not it is blocked."
2415                  *
2416                  *  "Setting a signal action to SIG_DFL for a signal that is
2417                  *   pending and whose default action is to ignore the signal
2418                  *   (for example, SIGCHLD), shall cause the pending signal to
2419                  *   be discarded, whether or not it is blocked"
2420                  */
2421                 if (act->sa.sa_handler == SIG_IGN ||
2422                     (act->sa.sa_handler == SIG_DFL &&
2423                      sig_kernel_ignore(sig))) {
2424                         /*
2425                          * This is a fairly rare case, so we only take the
2426                          * tasklist_lock once we're sure we'll need it.
2427                          * Now we must do this little unlock and relock
2428                          * dance to maintain the lock hierarchy.
2429                          */
2430                         struct task_struct *t = current;
2431                         spin_unlock_irq(&t->sighand->siglock);
2432                         read_lock(&tasklist_lock);
2433                         spin_lock_irq(&t->sighand->siglock);
2434                         *k = *act;
2435                         sigdelsetmask(&k->sa.sa_mask,
2436                                       sigmask(SIGKILL) | sigmask(SIGSTOP));
2437                         rm_from_queue(sigmask(sig), &t->signal->shared_pending);
2438                         do {
2439                                 rm_from_queue(sigmask(sig), &t->pending);
2440                                 recalc_sigpending_tsk(t);
2441                                 t = next_thread(t);
2442                         } while (t != current);
2443                         spin_unlock_irq(&current->sighand->siglock);
2444                         read_unlock(&tasklist_lock);
2445                         return 0;
2446                 }
2447
2448                 *k = *act;
2449                 sigdelsetmask(&k->sa.sa_mask,
2450                               sigmask(SIGKILL) | sigmask(SIGSTOP));
2451         }
2452
2453         spin_unlock_irq(&current->sighand->siglock);
2454         return 0;
2455 }
2456
2457 int 
2458 do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
2459 {
2460         stack_t oss;
2461         int error;
2462
2463         if (uoss) {
2464                 oss.ss_sp = (void __user *) current->sas_ss_sp;
2465                 oss.ss_size = current->sas_ss_size;
2466                 oss.ss_flags = sas_ss_flags(sp);
2467         }
2468
2469         if (uss) {
2470                 void __user *ss_sp;
2471                 size_t ss_size;
2472                 int ss_flags;
2473
2474                 error = -EFAULT;
2475                 if (!access_ok(VERIFY_READ, uss, sizeof(*uss))
2476                     || __get_user(ss_sp, &uss->ss_sp)
2477                     || __get_user(ss_flags, &uss->ss_flags)
2478                     || __get_user(ss_size, &uss->ss_size))
2479                         goto out;
2480
2481                 error = -EPERM;
2482                 if (on_sig_stack(sp))
2483                         goto out;
2484
2485                 error = -EINVAL;
2486                 /*
2487                  *
2488                  * Note - this code used to test ss_flags incorrectly
2489                  *        old code may have been written using ss_flags==0
2490                  *        to mean ss_flags==SS_ONSTACK (as this was the only
2491                  *        way that worked) - this fix preserves that older
2492                  *        mechanism
2493                  */
2494                 if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
2495                         goto out;
2496
2497                 if (ss_flags == SS_DISABLE) {
2498                         ss_size = 0;
2499                         ss_sp = NULL;
2500                 } else {
2501                         error = -ENOMEM;
2502                         if (ss_size < MINSIGSTKSZ)
2503                                 goto out;
2504                 }
2505
2506                 current->sas_ss_sp = (unsigned long) ss_sp;
2507                 current->sas_ss_size = ss_size;
2508         }
2509
2510         if (uoss) {
2511                 error = -EFAULT;
2512                 if (copy_to_user(uoss, &oss, sizeof(oss)))
2513                         goto out;
2514         }
2515
2516         error = 0;
2517 out:
2518         return error;
2519 }
2520
2521 #ifdef __ARCH_WANT_SYS_SIGPENDING
2522
2523 asmlinkage long
2524 sys_sigpending(old_sigset_t __user *set)
2525 {
2526         return do_sigpending(set, sizeof(*set));
2527 }
2528
2529 #endif
2530
2531 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
2532 /* Some platforms have their own version with special arguments others
2533    support only sys_rt_sigprocmask.  */
2534
2535 asmlinkage long
2536 sys_sigprocmask(int how, old_sigset_t __user *set, old_sigset_t __user *oset)
2537 {
2538         int error;
2539         old_sigset_t old_set, new_set;
2540
2541         if (set) {
2542                 error = -EFAULT;
2543                 if (copy_from_user(&new_set, set, sizeof(*set)))
2544                         goto out;
2545                 new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
2546
2547                 spin_lock_irq(&current->sighand->siglock);
2548                 old_set = current->blocked.sig[0];
2549
2550                 error = 0;
2551                 switch (how) {
2552                 default:
2553                         error = -EINVAL;
2554                         break;
2555                 case SIG_BLOCK:
2556                         sigaddsetmask(&current->blocked, new_set);
2557                         break;
2558                 case SIG_UNBLOCK:
2559                         sigdelsetmask(&current->blocked, new_set);
2560                         break;
2561                 case SIG_SETMASK:
2562                         current->blocked.sig[0] = new_set;
2563                         break;
2564                 }
2565
2566                 recalc_sigpending();
2567                 spin_unlock_irq(&current->sighand->siglock);
2568                 if (error)
2569                         goto out;
2570                 if (oset)
2571                         goto set_old;
2572         } else if (oset) {
2573                 old_set = current->blocked.sig[0];
2574         set_old:
2575                 error = -EFAULT;
2576                 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2577                         goto out;
2578         }
2579         error = 0;
2580 out:
2581         return error;
2582 }
2583 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
2584
2585 #ifdef __ARCH_WANT_SYS_RT_SIGACTION
2586 asmlinkage long
2587 sys_rt_sigaction(int sig,
2588                  const struct sigaction __user *act,
2589                  struct sigaction __user *oact,
2590                  size_t sigsetsize)
2591 {
2592         struct k_sigaction new_sa, old_sa;
2593         int ret = -EINVAL;
2594
2595         /* XXX: Don't preclude handling different sized sigset_t's.  */
2596         if (sigsetsize != sizeof(sigset_t))
2597                 goto out;
2598
2599         if (act) {
2600                 if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
2601                         return -EFAULT;
2602         }
2603
2604         ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
2605
2606         if (!ret && oact) {
2607                 if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
2608                         return -EFAULT;
2609         }
2610 out:
2611         return ret;
2612 }
2613 #endif /* __ARCH_WANT_SYS_RT_SIGACTION */
2614
2615 #ifdef __ARCH_WANT_SYS_SGETMASK
2616
2617 /*
2618  * For backwards compatibility.  Functionality superseded by sigprocmask.
2619  */
2620 asmlinkage long
2621 sys_sgetmask(void)
2622 {
2623         /* SMP safe */
2624         return current->blocked.sig[0];
2625 }
2626
2627 asmlinkage long
2628 sys_ssetmask(int newmask)
2629 {
2630         int old;
2631
2632         spin_lock_irq(&current->sighand->siglock);
2633         old = current->blocked.sig[0];
2634
2635         siginitset(&current->blocked, newmask & ~(sigmask(SIGKILL)|
2636                                                   sigmask(SIGSTOP)));
2637         recalc_sigpending();
2638         spin_unlock_irq(&current->sighand->siglock);
2639
2640         return old;
2641 }
2642 #endif /* __ARCH_WANT_SGETMASK */
2643
2644 #ifdef __ARCH_WANT_SYS_SIGNAL
2645 /*
2646  * For backwards compatibility.  Functionality superseded by sigaction.
2647  */
2648 asmlinkage unsigned long
2649 sys_signal(int sig, __sighandler_t handler)
2650 {
2651         struct k_sigaction new_sa, old_sa;
2652         int ret;
2653
2654         new_sa.sa.sa_handler = handler;
2655         new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
2656
2657         ret = do_sigaction(sig, &new_sa, &old_sa);
2658
2659         return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
2660 }
2661 #endif /* __ARCH_WANT_SYS_SIGNAL */
2662
2663 #ifdef __ARCH_WANT_SYS_PAUSE
2664
2665 asmlinkage long
2666 sys_pause(void)
2667 {
2668         current->state = TASK_INTERRUPTIBLE;
2669         schedule();
2670         return -ERESTARTNOHAND;
2671 }
2672
2673 #endif
2674
2675 void __init signals_init(void)
2676 {
2677         sigqueue_cachep =
2678                 kmem_cache_create("sigqueue",
2679                                   sizeof(struct sigqueue),
2680                                   __alignof__(struct sigqueue),
2681                                   SLAB_PANIC, NULL, NULL);
2682 }