kernel.org 2.6.11.7
[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 <asm/param.h>
26 #include <asm/uaccess.h>
27 #include <asm/unistd.h>
28 #include <asm/siginfo.h>
29
30 extern void k_getrusage(struct task_struct *, int, struct rusage *);
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, int flags)
264 {
265         struct sigqueue *q = NULL;
266
267         if (atomic_read(&t->user->sigpending) <
268                         t->signal->rlim[RLIMIT_SIGPENDING].rlim_cur)
269                 q = kmem_cache_alloc(sigqueue_cachep, flags);
270         if (q) {
271                 INIT_LIST_HEAD(&q->list);
272                 q->flags = 0;
273                 q->lock = NULL;
274                 q->user = get_uid(t->user);
275                 atomic_inc(&q->user->sigpending);
276         }
277         return(q);
278 }
279
280 static inline void __sigqueue_free(struct sigqueue *q)
281 {
282         if (q->flags & SIGQUEUE_PREALLOC)
283                 return;
284         atomic_dec(&q->user->sigpending);
285         free_uid(q->user);
286         kmem_cache_free(sigqueue_cachep, q);
287 }
288
289 static void flush_sigqueue(struct sigpending *queue)
290 {
291         struct sigqueue *q;
292
293         sigemptyset(&queue->signal);
294         while (!list_empty(&queue->list)) {
295                 q = list_entry(queue->list.next, struct sigqueue , list);
296                 list_del_init(&q->list);
297                 __sigqueue_free(q);
298         }
299 }
300
301 /*
302  * Flush all pending signals for a task.
303  */
304
305 void
306 flush_signals(struct task_struct *t)
307 {
308         unsigned long flags;
309
310         spin_lock_irqsave(&t->sighand->siglock, flags);
311         clear_tsk_thread_flag(t,TIF_SIGPENDING);
312         flush_sigqueue(&t->pending);
313         flush_sigqueue(&t->signal->shared_pending);
314         spin_unlock_irqrestore(&t->sighand->siglock, flags);
315 }
316
317 /*
318  * This function expects the tasklist_lock write-locked.
319  */
320 void __exit_sighand(struct task_struct *tsk)
321 {
322         struct sighand_struct * sighand = tsk->sighand;
323
324         /* Ok, we're done with the signal handlers */
325         tsk->sighand = NULL;
326         if (atomic_dec_and_test(&sighand->count))
327                 kmem_cache_free(sighand_cachep, sighand);
328 }
329
330 void exit_sighand(struct task_struct *tsk)
331 {
332         write_lock_irq(&tasklist_lock);
333         __exit_sighand(tsk);
334         write_unlock_irq(&tasklist_lock);
335 }
336
337 /*
338  * This function expects the tasklist_lock write-locked.
339  */
340 void __exit_signal(struct task_struct *tsk)
341 {
342         struct signal_struct * sig = tsk->signal;
343         struct sighand_struct * sighand = tsk->sighand;
344
345         if (!sig)
346                 BUG();
347         if (!atomic_read(&sig->count))
348                 BUG();
349         spin_lock(&sighand->siglock);
350         if (atomic_dec_and_test(&sig->count)) {
351                 if (tsk == sig->curr_target)
352                         sig->curr_target = next_thread(tsk);
353                 tsk->signal = NULL;
354                 spin_unlock(&sighand->siglock);
355                 flush_sigqueue(&sig->shared_pending);
356         } else {
357                 /*
358                  * If there is any task waiting for the group exit
359                  * then notify it:
360                  */
361                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
362                         wake_up_process(sig->group_exit_task);
363                         sig->group_exit_task = NULL;
364                 }
365                 if (tsk == sig->curr_target)
366                         sig->curr_target = next_thread(tsk);
367                 tsk->signal = NULL;
368                 /*
369                  * Accumulate here the counters for all threads but the
370                  * group leader as they die, so they can be added into
371                  * the process-wide totals when those are taken.
372                  * The group leader stays around as a zombie as long
373                  * as there are other threads.  When it gets reaped,
374                  * the exit.c code will add its counts into these totals.
375                  * We won't ever get here for the group leader, since it
376                  * will have been the last reference on the signal_struct.
377                  */
378                 sig->utime = cputime_add(sig->utime, tsk->utime);
379                 sig->stime = cputime_add(sig->stime, tsk->stime);
380                 sig->min_flt += tsk->min_flt;
381                 sig->maj_flt += tsk->maj_flt;
382                 sig->nvcsw += tsk->nvcsw;
383                 sig->nivcsw += tsk->nivcsw;
384                 spin_unlock(&sighand->siglock);
385                 sig = NULL;     /* Marker for below.  */
386         }
387         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
388         flush_sigqueue(&tsk->pending);
389         if (sig) {
390                 /*
391                  * We are cleaning up the signal_struct here.  We delayed
392                  * calling exit_itimers until after flush_sigqueue, just in
393                  * case our thread-local pending queue contained a queued
394                  * timer signal that would have been cleared in
395                  * exit_itimers.  When that called sigqueue_free, it would
396                  * attempt to re-take the tasklist_lock and deadlock.  This
397                  * can never happen if we ensure that all queues the
398                  * timer's signal might be queued on have been flushed
399                  * first.  The shared_pending queue, and our own pending
400                  * queue are the only queues the timer could be on, since
401                  * there are no other threads left in the group and timer
402                  * signals are constrained to threads inside the group.
403                  */
404                 exit_itimers(sig);
405                 kmem_cache_free(signal_cachep, sig);
406         }
407 }
408
409 void exit_signal(struct task_struct *tsk)
410 {
411         write_lock_irq(&tasklist_lock);
412         __exit_signal(tsk);
413         write_unlock_irq(&tasklist_lock);
414 }
415
416 /*
417  * Flush all handlers for a task.
418  */
419
420 void
421 flush_signal_handlers(struct task_struct *t, int force_default)
422 {
423         int i;
424         struct k_sigaction *ka = &t->sighand->action[0];
425         for (i = _NSIG ; i != 0 ; i--) {
426                 if (force_default || ka->sa.sa_handler != SIG_IGN)
427                         ka->sa.sa_handler = SIG_DFL;
428                 ka->sa.sa_flags = 0;
429                 sigemptyset(&ka->sa.sa_mask);
430                 ka++;
431         }
432 }
433
434
435 /* Notify the system that a driver wants to block all signals for this
436  * process, and wants to be notified if any signals at all were to be
437  * sent/acted upon.  If the notifier routine returns non-zero, then the
438  * signal will be acted upon after all.  If the notifier routine returns 0,
439  * then then signal will be blocked.  Only one block per process is
440  * allowed.  priv is a pointer to private data that the notifier routine
441  * can use to determine if the signal should be blocked or not.  */
442
443 void
444 block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
445 {
446         unsigned long flags;
447
448         spin_lock_irqsave(&current->sighand->siglock, flags);
449         current->notifier_mask = mask;
450         current->notifier_data = priv;
451         current->notifier = notifier;
452         spin_unlock_irqrestore(&current->sighand->siglock, flags);
453 }
454
455 /* Notify the system that blocking has ended. */
456
457 void
458 unblock_all_signals(void)
459 {
460         unsigned long flags;
461
462         spin_lock_irqsave(&current->sighand->siglock, flags);
463         current->notifier = NULL;
464         current->notifier_data = NULL;
465         recalc_sigpending();
466         spin_unlock_irqrestore(&current->sighand->siglock, flags);
467 }
468
469 static inline int collect_signal(int sig, struct sigpending *list, siginfo_t *info)
470 {
471         struct sigqueue *q, *first = NULL;
472         int still_pending = 0;
473
474         if (unlikely(!sigismember(&list->signal, sig)))
475                 return 0;
476
477         /*
478          * Collect the siginfo appropriate to this signal.  Check if
479          * there is another siginfo for the same signal.
480         */
481         list_for_each_entry(q, &list->list, list) {
482                 if (q->info.si_signo == sig) {
483                         if (first) {
484                                 still_pending = 1;
485                                 break;
486                         }
487                         first = q;
488                 }
489         }
490         if (first) {
491                 list_del_init(&first->list);
492                 copy_siginfo(info, &first->info);
493                 __sigqueue_free(first);
494                 if (!still_pending)
495                         sigdelset(&list->signal, sig);
496         } else {
497
498                 /* Ok, it wasn't in the queue.  This must be
499                    a fast-pathed signal or we must have been
500                    out of queue space.  So zero out the info.
501                  */
502                 sigdelset(&list->signal, sig);
503                 info->si_signo = sig;
504                 info->si_errno = 0;
505                 info->si_code = 0;
506                 info->si_pid = 0;
507                 info->si_uid = 0;
508         }
509         return 1;
510 }
511
512 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
513                         siginfo_t *info)
514 {
515         int sig = 0;
516
517         sig = next_signal(pending, mask);
518         if (sig) {
519                 if (current->notifier) {
520                         if (sigismember(current->notifier_mask, sig)) {
521                                 if (!(current->notifier)(current->notifier_data)) {
522                                         clear_thread_flag(TIF_SIGPENDING);
523                                         return 0;
524                                 }
525                         }
526                 }
527
528                 if (!collect_signal(sig, pending, info))
529                         sig = 0;
530                                 
531         }
532         recalc_sigpending();
533
534         return sig;
535 }
536
537 /*
538  * Dequeue a signal and return the element to the caller, which is 
539  * expected to free it.
540  *
541  * All callers have to hold the siglock.
542  */
543 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
544 {
545         int signr = __dequeue_signal(&tsk->pending, mask, info);
546         if (!signr)
547                 signr = __dequeue_signal(&tsk->signal->shared_pending,
548                                          mask, info);
549         if (signr && unlikely(sig_kernel_stop(signr))) {
550                 /*
551                  * Set a marker that we have dequeued a stop signal.  Our
552                  * caller might release the siglock and then the pending
553                  * stop signal it is about to process is no longer in the
554                  * pending bitmasks, but must still be cleared by a SIGCONT
555                  * (and overruled by a SIGKILL).  So those cases clear this
556                  * shared flag after we've set it.  Note that this flag may
557                  * remain set after the signal we return is ignored or
558                  * handled.  That doesn't matter because its only purpose
559                  * is to alert stop-signal processing code when another
560                  * processor has come along and cleared the flag.
561                  */
562                 tsk->signal->flags |= SIGNAL_STOP_DEQUEUED;
563         }
564         if ( signr &&
565              ((info->si_code & __SI_MASK) == __SI_TIMER) &&
566              info->si_sys_private){
567                 do_schedule_next_timer(info);
568         }
569         return signr;
570 }
571
572 /*
573  * Tell a process that it has a new active signal..
574  *
575  * NOTE! we rely on the previous spin_lock to
576  * lock interrupts for us! We can only be called with
577  * "siglock" held, and the local interrupt must
578  * have been disabled when that got acquired!
579  *
580  * No need to set need_resched since signal event passing
581  * goes through ->blocked
582  */
583 void signal_wake_up(struct task_struct *t, int resume)
584 {
585         unsigned int mask;
586
587         set_tsk_thread_flag(t, TIF_SIGPENDING);
588
589         /*
590          * For SIGKILL, we want to wake it up in the stopped/traced case.
591          * We don't check t->state here because there is a race with it
592          * executing another processor and just now entering stopped state.
593          * By using wake_up_state, we ensure the process will wake up and
594          * handle its death signal.
595          */
596         mask = TASK_INTERRUPTIBLE;
597         if (resume)
598                 mask |= TASK_STOPPED | TASK_TRACED;
599         if (!wake_up_state(t, mask))
600                 kick_process(t);
601 }
602
603 /*
604  * Remove signals in mask from the pending set and queue.
605  * Returns 1 if any signals were found.
606  *
607  * All callers must be holding the siglock.
608  */
609 static int rm_from_queue(unsigned long mask, struct sigpending *s)
610 {
611         struct sigqueue *q, *n;
612
613         if (!sigtestsetmask(&s->signal, mask))
614                 return 0;
615
616         sigdelsetmask(&s->signal, mask);
617         list_for_each_entry_safe(q, n, &s->list, list) {
618                 if (q->info.si_signo < SIGRTMIN &&
619                     (mask & sigmask(q->info.si_signo))) {
620                         list_del_init(&q->list);
621                         __sigqueue_free(q);
622                 }
623         }
624         return 1;
625 }
626
627 /*
628  * Bad permissions for sending the signal
629  */
630 static int check_kill_permission(int sig, struct siginfo *info,
631                                  struct task_struct *t)
632 {
633         int error = -EINVAL;
634         if (sig < 0 || sig > _NSIG)
635                 return error;
636         error = -EPERM;
637         if ((!info || ((unsigned long)info != 1 &&
638                         (unsigned long)info != 2 && SI_FROMUSER(info)))
639             && ((sig != SIGCONT) ||
640                 (current->signal->session != t->signal->session))
641             && (current->euid ^ t->suid) && (current->euid ^ t->uid)
642             && (current->uid ^ t->suid) && (current->uid ^ t->uid)
643             && !capable(CAP_KILL))
644                 return error;
645         return security_task_kill(t, info, sig);
646 }
647
648 /* forward decl */
649 static void do_notify_parent_cldstop(struct task_struct *tsk,
650                                      struct task_struct *parent,
651                                      int why);
652
653 /*
654  * Handle magic process-wide effects of stop/continue signals.
655  * Unlike the signal actions, these happen immediately at signal-generation
656  * time regardless of blocking, ignoring, or handling.  This does the
657  * actual continuing for SIGCONT, but not the actual stopping for stop
658  * signals.  The process stop is done as a signal action for SIG_DFL.
659  */
660 static void handle_stop_signal(int sig, struct task_struct *p)
661 {
662         struct task_struct *t;
663
664         if (p->flags & SIGNAL_GROUP_EXIT)
665                 /*
666                  * The process is in the middle of dying already.
667                  */
668                 return;
669
670         if (sig_kernel_stop(sig)) {
671                 /*
672                  * This is a stop signal.  Remove SIGCONT from all queues.
673                  */
674                 rm_from_queue(sigmask(SIGCONT), &p->signal->shared_pending);
675                 t = p;
676                 do {
677                         rm_from_queue(sigmask(SIGCONT), &t->pending);
678                         t = next_thread(t);
679                 } while (t != p);
680         } else if (sig == SIGCONT) {
681                 /*
682                  * Remove all stop signals from all queues,
683                  * and wake all threads.
684                  */
685                 if (unlikely(p->signal->group_stop_count > 0)) {
686                         /*
687                          * There was a group stop in progress.  We'll
688                          * pretend it finished before we got here.  We are
689                          * obliged to report it to the parent: if the
690                          * SIGSTOP happened "after" this SIGCONT, then it
691                          * would have cleared this pending SIGCONT.  If it
692                          * happened "before" this SIGCONT, then the parent
693                          * got the SIGCHLD about the stop finishing before
694                          * the continue happened.  We do the notification
695                          * now, and it's as if the stop had finished and
696                          * the SIGCHLD was pending on entry to this kill.
697                          */
698                         p->signal->group_stop_count = 0;
699                         p->signal->flags = SIGNAL_STOP_CONTINUED;
700                         spin_unlock(&p->sighand->siglock);
701                         if (p->ptrace & PT_PTRACED)
702                                 do_notify_parent_cldstop(p, p->parent,
703                                                          CLD_STOPPED);
704                         else
705                                 do_notify_parent_cldstop(
706                                         p->group_leader,
707                                         p->group_leader->real_parent,
708                                                          CLD_STOPPED);
709                         spin_lock(&p->sighand->siglock);
710                 }
711                 rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
712                 t = p;
713                 do {
714                         unsigned int state;
715                         rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
716                         
717                         /*
718                          * If there is a handler for SIGCONT, we must make
719                          * sure that no thread returns to user mode before
720                          * we post the signal, in case it was the only
721                          * thread eligible to run the signal handler--then
722                          * it must not do anything between resuming and
723                          * running the handler.  With the TIF_SIGPENDING
724                          * flag set, the thread will pause and acquire the
725                          * siglock that we hold now and until we've queued
726                          * the pending signal. 
727                          *
728                          * Wake up the stopped thread _after_ setting
729                          * TIF_SIGPENDING
730                          */
731                         state = TASK_STOPPED;
732                         if (sig_user_defined(t, SIGCONT) && !sigismember(&t->blocked, SIGCONT)) {
733                                 set_tsk_thread_flag(t, TIF_SIGPENDING);
734                                 state |= TASK_INTERRUPTIBLE;
735                         }
736                         wake_up_state(t, state);
737
738                         t = next_thread(t);
739                 } while (t != p);
740
741                 if (p->signal->flags & SIGNAL_STOP_STOPPED) {
742                         /*
743                          * We were in fact stopped, and are now continued.
744                          * Notify the parent with CLD_CONTINUED.
745                          */
746                         p->signal->flags = SIGNAL_STOP_CONTINUED;
747                         p->signal->group_exit_code = 0;
748                         spin_unlock(&p->sighand->siglock);
749                         if (p->ptrace & PT_PTRACED)
750                                 do_notify_parent_cldstop(p, p->parent,
751                                                          CLD_CONTINUED);
752                         else
753                                 do_notify_parent_cldstop(
754                                         p->group_leader,
755                                         p->group_leader->real_parent,
756                                                          CLD_CONTINUED);
757                         spin_lock(&p->sighand->siglock);
758                 } else {
759                         /*
760                          * We are not stopped, but there could be a stop
761                          * signal in the middle of being processed after
762                          * being removed from the queue.  Clear that too.
763                          */
764                         p->signal->flags = 0;
765                 }
766         } else if (sig == SIGKILL) {
767                 /*
768                  * Make sure that any pending stop signal already dequeued
769                  * is undone by the wakeup for SIGKILL.
770                  */
771                 p->signal->flags = 0;
772         }
773 }
774
775 static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
776                         struct sigpending *signals)
777 {
778         struct sigqueue * q = NULL;
779         int ret = 0;
780
781         /*
782          * fast-pathed signals for kernel-internal things like SIGSTOP
783          * or SIGKILL.
784          */
785         if ((unsigned long)info == 2)
786                 goto out_set;
787
788         /* Real-time signals must be queued if sent by sigqueue, or
789            some other real-time mechanism.  It is implementation
790            defined whether kill() does so.  We attempt to do so, on
791            the principle of least surprise, but since kill is not
792            allowed to fail with EAGAIN when low on memory we just
793            make sure at least one signal gets delivered and don't
794            pass on the info struct.  */
795
796         q = __sigqueue_alloc(t, GFP_ATOMIC);
797         if (q) {
798                 list_add_tail(&q->list, &signals->list);
799                 switch ((unsigned long) info) {
800                 case 0:
801                         q->info.si_signo = sig;
802                         q->info.si_errno = 0;
803                         q->info.si_code = SI_USER;
804                         q->info.si_pid = current->pid;
805                         q->info.si_uid = current->uid;
806                         break;
807                 case 1:
808                         q->info.si_signo = sig;
809                         q->info.si_errno = 0;
810                         q->info.si_code = SI_KERNEL;
811                         q->info.si_pid = 0;
812                         q->info.si_uid = 0;
813                         break;
814                 default:
815                         copy_siginfo(&q->info, info);
816                         break;
817                 }
818         } else {
819                 if (sig >= SIGRTMIN && info && (unsigned long)info != 1
820                    && info->si_code != SI_USER)
821                 /*
822                  * Queue overflow, abort.  We may abort if the signal was rt
823                  * and sent by user using something other than kill().
824                  */
825                         return -EAGAIN;
826                 if (((unsigned long)info > 1) && (info->si_code == SI_TIMER))
827                         /*
828                          * Set up a return to indicate that we dropped 
829                          * the signal.
830                          */
831                         ret = info->si_sys_private;
832         }
833
834 out_set:
835         sigaddset(&signals->signal, sig);
836         return ret;
837 }
838
839 #define LEGACY_QUEUE(sigptr, sig) \
840         (((sig) < SIGRTMIN) && sigismember(&(sigptr)->signal, (sig)))
841
842
843 static int
844 specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
845 {
846         int ret = 0;
847
848         if (!irqs_disabled())
849                 BUG();
850         assert_spin_locked(&t->sighand->siglock);
851
852         if (((unsigned long)info > 2) && (info->si_code == SI_TIMER))
853                 /*
854                  * Set up a return to indicate that we dropped the signal.
855                  */
856                 ret = info->si_sys_private;
857
858         /* Short-circuit ignored signals.  */
859         if (sig_ignored(t, sig))
860                 goto out;
861
862         /* Support queueing exactly one non-rt signal, so that we
863            can get more detailed information about the cause of
864            the signal. */
865         if (LEGACY_QUEUE(&t->pending, sig))
866                 goto out;
867
868         ret = send_signal(sig, info, t, &t->pending);
869         if (!ret && !sigismember(&t->blocked, sig))
870                 signal_wake_up(t, sig == SIGKILL);
871 out:
872         return ret;
873 }
874
875 /*
876  * Force a signal that the process can't ignore: if necessary
877  * we unblock the signal and change any SIG_IGN to SIG_DFL.
878  */
879
880 int
881 force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
882 {
883         unsigned long int flags;
884         int ret;
885
886         spin_lock_irqsave(&t->sighand->siglock, flags);
887         if (sigismember(&t->blocked, sig) || t->sighand->action[sig-1].sa.sa_handler == SIG_IGN) {
888                 t->sighand->action[sig-1].sa.sa_handler = SIG_DFL;
889                 sigdelset(&t->blocked, sig);
890                 recalc_sigpending_tsk(t);
891         }
892         ret = specific_send_sig_info(sig, info, t);
893         spin_unlock_irqrestore(&t->sighand->siglock, flags);
894
895         return ret;
896 }
897
898 void
899 force_sig_specific(int sig, struct task_struct *t)
900 {
901         unsigned long int flags;
902
903         spin_lock_irqsave(&t->sighand->siglock, flags);
904         if (t->sighand->action[sig-1].sa.sa_handler == SIG_IGN)
905                 t->sighand->action[sig-1].sa.sa_handler = SIG_DFL;
906         sigdelset(&t->blocked, sig);
907         recalc_sigpending_tsk(t);
908         specific_send_sig_info(sig, (void *)2, t);
909         spin_unlock_irqrestore(&t->sighand->siglock, flags);
910 }
911
912 /*
913  * Test if P wants to take SIG.  After we've checked all threads with this,
914  * it's equivalent to finding no threads not blocking SIG.  Any threads not
915  * blocking SIG were ruled out because they are not running and already
916  * have pending signals.  Such threads will dequeue from the shared queue
917  * as soon as they're available, so putting the signal on the shared queue
918  * will be equivalent to sending it to one such thread.
919  */
920 #define wants_signal(sig, p, mask)                      \
921         (!sigismember(&(p)->blocked, sig)               \
922          && !((p)->state & mask)                        \
923          && !((p)->flags & PF_EXITING)                  \
924          && (task_curr(p) || !signal_pending(p)))
925
926
927 static void
928 __group_complete_signal(int sig, struct task_struct *p)
929 {
930         unsigned int mask;
931         struct task_struct *t;
932
933         /*
934          * Don't bother traced and stopped tasks (but
935          * SIGKILL will punch through that).
936          */
937         mask = TASK_STOPPED | TASK_TRACED;
938         if (sig == SIGKILL)
939                 mask = 0;
940
941         /*
942          * Now find a thread we can wake up to take the signal off the queue.
943          *
944          * If the main thread wants the signal, it gets first crack.
945          * Probably the least surprising to the average bear.
946          */
947         if (wants_signal(sig, p, mask))
948                 t = p;
949         else if (thread_group_empty(p))
950                 /*
951                  * There is just one thread and it does not need to be woken.
952                  * It will dequeue unblocked signals before it runs again.
953                  */
954                 return;
955         else {
956                 /*
957                  * Otherwise try to find a suitable thread.
958                  */
959                 t = p->signal->curr_target;
960                 if (t == NULL)
961                         /* restart balancing at this thread */
962                         t = p->signal->curr_target = p;
963                 BUG_ON(t->tgid != p->tgid);
964
965                 while (!wants_signal(sig, t, mask)) {
966                         t = next_thread(t);
967                         if (t == p->signal->curr_target)
968                                 /*
969                                  * No thread needs to be woken.
970                                  * Any eligible threads will see
971                                  * the signal in the queue soon.
972                                  */
973                                 return;
974                 }
975                 p->signal->curr_target = t;
976         }
977
978         /*
979          * Found a killable thread.  If the signal will be fatal,
980          * then start taking the whole group down immediately.
981          */
982         if (sig_fatal(p, sig) && !(p->signal->flags & SIGNAL_GROUP_EXIT) &&
983             !sigismember(&t->real_blocked, sig) &&
984             (sig == SIGKILL || !(t->ptrace & PT_PTRACED))) {
985                 /*
986                  * This signal will be fatal to the whole group.
987                  */
988                 if (!sig_kernel_coredump(sig)) {
989                         /*
990                          * Start a group exit and wake everybody up.
991                          * This way we don't have other threads
992                          * running and doing things after a slower
993                          * thread has the fatal signal pending.
994                          */
995                         p->signal->flags = SIGNAL_GROUP_EXIT;
996                         p->signal->group_exit_code = sig;
997                         p->signal->group_stop_count = 0;
998                         t = p;
999                         do {
1000                                 sigaddset(&t->pending.signal, SIGKILL);
1001                                 signal_wake_up(t, 1);
1002                                 t = next_thread(t);
1003                         } while (t != p);
1004                         return;
1005                 }
1006
1007                 /*
1008                  * There will be a core dump.  We make all threads other
1009                  * than the chosen one go into a group stop so that nothing
1010                  * happens until it gets scheduled, takes the signal off
1011                  * the shared queue, and does the core dump.  This is a
1012                  * little more complicated than strictly necessary, but it
1013                  * keeps the signal state that winds up in the core dump
1014                  * unchanged from the death state, e.g. which thread had
1015                  * the core-dump signal unblocked.
1016                  */
1017                 rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
1018                 rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
1019                 p->signal->group_stop_count = 0;
1020                 p->signal->group_exit_task = t;
1021                 t = p;
1022                 do {
1023                         p->signal->group_stop_count++;
1024                         signal_wake_up(t, 0);
1025                         t = next_thread(t);
1026                 } while (t != p);
1027                 wake_up_process(p->signal->group_exit_task);
1028                 return;
1029         }
1030
1031         /*
1032          * The signal is already in the shared-pending queue.
1033          * Tell the chosen thread to wake up and dequeue it.
1034          */
1035         signal_wake_up(t, sig == SIGKILL);
1036         return;
1037 }
1038
1039 static int
1040 __group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1041 {
1042         int ret = 0;
1043
1044         assert_spin_locked(&p->sighand->siglock);
1045         handle_stop_signal(sig, p);
1046
1047         if (((unsigned long)info > 2) && (info->si_code == SI_TIMER))
1048                 /*
1049                  * Set up a return to indicate that we dropped the signal.
1050                  */
1051                 ret = info->si_sys_private;
1052
1053         /* Short-circuit ignored signals.  */
1054         if (sig_ignored(p, sig))
1055                 return ret;
1056
1057         if (LEGACY_QUEUE(&p->signal->shared_pending, sig))
1058                 /* This is a non-RT signal and we already have one queued.  */
1059                 return ret;
1060
1061         /*
1062          * Put this signal on the shared-pending queue, or fail with EAGAIN.
1063          * We always use the shared queue for process-wide signals,
1064          * to avoid several races.
1065          */
1066         ret = send_signal(sig, info, p, &p->signal->shared_pending);
1067         if (unlikely(ret))
1068                 return ret;
1069
1070         __group_complete_signal(sig, p);
1071         return 0;
1072 }
1073
1074 /*
1075  * Nuke all other threads in the group.
1076  */
1077 void zap_other_threads(struct task_struct *p)
1078 {
1079         struct task_struct *t;
1080
1081         p->signal->flags = SIGNAL_GROUP_EXIT;
1082         p->signal->group_stop_count = 0;
1083
1084         if (thread_group_empty(p))
1085                 return;
1086
1087         for (t = next_thread(p); t != p; t = next_thread(t)) {
1088                 /*
1089                  * Don't bother with already dead threads
1090                  */
1091                 if (t->exit_state)
1092                         continue;
1093
1094                 /*
1095                  * We don't want to notify the parent, since we are
1096                  * killed as part of a thread group due to another
1097                  * thread doing an execve() or similar. So set the
1098                  * exit signal to -1 to allow immediate reaping of
1099                  * the process.  But don't detach the thread group
1100                  * leader.
1101                  */
1102                 if (t != p->group_leader)
1103                         t->exit_signal = -1;
1104
1105                 sigaddset(&t->pending.signal, SIGKILL);
1106                 rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
1107                 signal_wake_up(t, 1);
1108         }
1109 }
1110
1111 /*
1112  * Must be called with the tasklist_lock held for reading!
1113  */
1114 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1115 {
1116         unsigned long flags;
1117         int ret;
1118
1119         ret = check_kill_permission(sig, info, p);
1120         if (!ret && sig && p->sighand) {
1121                 spin_lock_irqsave(&p->sighand->siglock, flags);
1122                 ret = __group_send_sig_info(sig, info, p);
1123                 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1124         }
1125
1126         return ret;
1127 }
1128
1129 /*
1130  * kill_pg_info() sends a signal to a process group: this is what the tty
1131  * control characters do (^C, ^Z etc)
1132  */
1133
1134 int __kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)
1135 {
1136         struct task_struct *p = NULL;
1137         int retval, success;
1138
1139         if (pgrp <= 0)
1140                 return -EINVAL;
1141
1142         success = 0;
1143         retval = -ESRCH;
1144         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
1145                 int err = group_send_sig_info(sig, info, p);
1146                 success |= !err;
1147                 retval = err;
1148         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
1149         return success ? 0 : retval;
1150 }
1151
1152 int
1153 kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)
1154 {
1155         int retval;
1156
1157         read_lock(&tasklist_lock);
1158         retval = __kill_pg_info(sig, info, pgrp);
1159         read_unlock(&tasklist_lock);
1160
1161         return retval;
1162 }
1163
1164 int
1165 kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1166 {
1167         int error;
1168         struct task_struct *p;
1169
1170         read_lock(&tasklist_lock);
1171         p = find_task_by_pid(pid);
1172         error = -ESRCH;
1173         if (p)
1174                 error = group_send_sig_info(sig, info, p);
1175         read_unlock(&tasklist_lock);
1176         return error;
1177 }
1178
1179
1180 /*
1181  * kill_something_info() interprets pid in interesting ways just like kill(2).
1182  *
1183  * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1184  * is probably wrong.  Should make it like BSD or SYSV.
1185  */
1186
1187 static int kill_something_info(int sig, struct siginfo *info, int pid)
1188 {
1189         if (!pid) {
1190                 return kill_pg_info(sig, info, process_group(current));
1191         } else if (pid == -1) {
1192                 int retval = 0, count = 0;
1193                 struct task_struct * p;
1194
1195                 read_lock(&tasklist_lock);
1196                 for_each_process(p) {
1197                         if (p->pid > 1 && p->tgid != current->tgid) {
1198                                 int err = group_send_sig_info(sig, info, p);
1199                                 ++count;
1200                                 if (err != -EPERM)
1201                                         retval = err;
1202                         }
1203                 }
1204                 read_unlock(&tasklist_lock);
1205                 return count ? retval : -ESRCH;
1206         } else if (pid < 0) {
1207                 return kill_pg_info(sig, info, -pid);
1208         } else {
1209                 return kill_proc_info(sig, info, pid);
1210         }
1211 }
1212
1213 /*
1214  * These are for backward compatibility with the rest of the kernel source.
1215  */
1216
1217 /*
1218  * These two are the most common entry points.  They send a signal
1219  * just to the specific thread.
1220  */
1221 int
1222 send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1223 {
1224         int ret;
1225         unsigned long flags;
1226
1227         /*
1228          * Make sure legacy kernel users don't send in bad values
1229          * (normal paths check this in check_kill_permission).
1230          */
1231         if (sig < 0 || sig > _NSIG)
1232                 return -EINVAL;
1233
1234         /*
1235          * We need the tasklist lock even for the specific
1236          * thread case (when we don't need to follow the group
1237          * lists) in order to avoid races with "p->sighand"
1238          * going away or changing from under us.
1239          */
1240         read_lock(&tasklist_lock);  
1241         spin_lock_irqsave(&p->sighand->siglock, flags);
1242         ret = specific_send_sig_info(sig, info, p);
1243         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1244         read_unlock(&tasklist_lock);
1245         return ret;
1246 }
1247
1248 int
1249 send_sig(int sig, struct task_struct *p, int priv)
1250 {
1251         return send_sig_info(sig, (void*)(long)(priv != 0), p);
1252 }
1253
1254 /*
1255  * This is the entry point for "process-wide" signals.
1256  * They will go to an appropriate thread in the thread group.
1257  */
1258 int
1259 send_group_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1260 {
1261         int ret;
1262         read_lock(&tasklist_lock);
1263         ret = group_send_sig_info(sig, info, p);
1264         read_unlock(&tasklist_lock);
1265         return ret;
1266 }
1267
1268 void
1269 force_sig(int sig, struct task_struct *p)
1270 {
1271         force_sig_info(sig, (void*)1L, p);
1272 }
1273
1274 /*
1275  * When things go south during signal handling, we
1276  * will force a SIGSEGV. And if the signal that caused
1277  * the problem was already a SIGSEGV, we'll want to
1278  * make sure we don't even try to deliver the signal..
1279  */
1280 int
1281 force_sigsegv(int sig, struct task_struct *p)
1282 {
1283         if (sig == SIGSEGV) {
1284                 unsigned long flags;
1285                 spin_lock_irqsave(&p->sighand->siglock, flags);
1286                 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1287                 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1288         }
1289         force_sig(SIGSEGV, p);
1290         return 0;
1291 }
1292
1293 int
1294 kill_pg(pid_t pgrp, int sig, int priv)
1295 {
1296         return kill_pg_info(sig, (void *)(long)(priv != 0), pgrp);
1297 }
1298
1299 int
1300 kill_proc(pid_t pid, int sig, int priv)
1301 {
1302         return kill_proc_info(sig, (void *)(long)(priv != 0), pid);
1303 }
1304
1305 /*
1306  * These functions support sending signals using preallocated sigqueue
1307  * structures.  This is needed "because realtime applications cannot
1308  * afford to lose notifications of asynchronous events, like timer
1309  * expirations or I/O completions".  In the case of Posix Timers 
1310  * we allocate the sigqueue structure from the timer_create.  If this
1311  * allocation fails we are able to report the failure to the application
1312  * with an EAGAIN error.
1313  */
1314  
1315 struct sigqueue *sigqueue_alloc(void)
1316 {
1317         struct sigqueue *q;
1318
1319         if ((q = __sigqueue_alloc(current, GFP_KERNEL)))
1320                 q->flags |= SIGQUEUE_PREALLOC;
1321         return(q);
1322 }
1323
1324 void sigqueue_free(struct sigqueue *q)
1325 {
1326         unsigned long flags;
1327         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1328         /*
1329          * If the signal is still pending remove it from the
1330          * pending queue.
1331          */
1332         if (unlikely(!list_empty(&q->list))) {
1333                 read_lock(&tasklist_lock);  
1334                 spin_lock_irqsave(q->lock, flags);
1335                 if (!list_empty(&q->list))
1336                         list_del_init(&q->list);
1337                 spin_unlock_irqrestore(q->lock, flags);
1338                 read_unlock(&tasklist_lock);
1339         }
1340         q->flags &= ~SIGQUEUE_PREALLOC;
1341         __sigqueue_free(q);
1342 }
1343
1344 int
1345 send_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
1346 {
1347         unsigned long flags;
1348         int ret = 0;
1349
1350         /*
1351          * We need the tasklist lock even for the specific
1352          * thread case (when we don't need to follow the group
1353          * lists) in order to avoid races with "p->sighand"
1354          * going away or changing from under us.
1355          */
1356         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1357         read_lock(&tasklist_lock);  
1358         spin_lock_irqsave(&p->sighand->siglock, flags);
1359         
1360         if (unlikely(!list_empty(&q->list))) {
1361                 /*
1362                  * If an SI_TIMER entry is already queue just increment
1363                  * the overrun count.
1364                  */
1365                 if (q->info.si_code != SI_TIMER)
1366                         BUG();
1367                 q->info.si_overrun++;
1368                 goto out;
1369         } 
1370         /* Short-circuit ignored signals.  */
1371         if (sig_ignored(p, sig)) {
1372                 ret = 1;
1373                 goto out;
1374         }
1375
1376         q->lock = &p->sighand->siglock;
1377         list_add_tail(&q->list, &p->pending.list);
1378         sigaddset(&p->pending.signal, sig);
1379         if (!sigismember(&p->blocked, sig))
1380                 signal_wake_up(p, sig == SIGKILL);
1381
1382 out:
1383         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1384         read_unlock(&tasklist_lock);
1385         return(ret);
1386 }
1387
1388 int
1389 send_group_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
1390 {
1391         unsigned long flags;
1392         int ret = 0;
1393
1394         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1395         read_lock(&tasklist_lock);
1396         spin_lock_irqsave(&p->sighand->siglock, flags);
1397         handle_stop_signal(sig, p);
1398
1399         /* Short-circuit ignored signals.  */
1400         if (sig_ignored(p, sig)) {
1401                 ret = 1;
1402                 goto out;
1403         }
1404
1405         if (unlikely(!list_empty(&q->list))) {
1406                 /*
1407                  * If an SI_TIMER entry is already queue just increment
1408                  * the overrun count.  Other uses should not try to
1409                  * send the signal multiple times.
1410                  */
1411                 if (q->info.si_code != SI_TIMER)
1412                         BUG();
1413                 q->info.si_overrun++;
1414                 goto out;
1415         } 
1416
1417         /*
1418          * Put this signal on the shared-pending queue.
1419          * We always use the shared queue for process-wide signals,
1420          * to avoid several races.
1421          */
1422         q->lock = &p->sighand->siglock;
1423         list_add_tail(&q->list, &p->signal->shared_pending.list);
1424         sigaddset(&p->signal->shared_pending.signal, sig);
1425
1426         __group_complete_signal(sig, p);
1427 out:
1428         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1429         read_unlock(&tasklist_lock);
1430         return(ret);
1431 }
1432
1433 /*
1434  * Wake up any threads in the parent blocked in wait* syscalls.
1435  */
1436 static inline void __wake_up_parent(struct task_struct *p,
1437                                     struct task_struct *parent)
1438 {
1439         wake_up_interruptible_sync(&parent->signal->wait_chldexit);
1440 }
1441
1442 /*
1443  * Let a parent know about the death of a child.
1444  * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1445  */
1446
1447 void do_notify_parent(struct task_struct *tsk, int sig)
1448 {
1449         struct siginfo info;
1450         unsigned long flags;
1451         struct sighand_struct *psig;
1452
1453         BUG_ON(sig == -1);
1454
1455         /* do_notify_parent_cldstop should have been called instead.  */
1456         BUG_ON(tsk->state & (TASK_STOPPED|TASK_TRACED));
1457
1458         BUG_ON(!tsk->ptrace &&
1459                (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1460
1461         info.si_signo = sig;
1462         info.si_errno = 0;
1463         info.si_pid = tsk->pid;
1464         info.si_uid = tsk->uid;
1465
1466         /* FIXME: find out whether or not this is supposed to be c*time. */
1467         info.si_utime = cputime_to_jiffies(cputime_add(tsk->utime,
1468                                                        tsk->signal->utime));
1469         info.si_stime = cputime_to_jiffies(cputime_add(tsk->stime,
1470                                                        tsk->signal->stime));
1471
1472         info.si_status = tsk->exit_code & 0x7f;
1473         if (tsk->exit_code & 0x80)
1474                 info.si_code = CLD_DUMPED;
1475         else if (tsk->exit_code & 0x7f)
1476                 info.si_code = CLD_KILLED;
1477         else {
1478                 info.si_code = CLD_EXITED;
1479                 info.si_status = tsk->exit_code >> 8;
1480         }
1481
1482         psig = tsk->parent->sighand;
1483         spin_lock_irqsave(&psig->siglock, flags);
1484         if (sig == SIGCHLD &&
1485             (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1486              (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1487                 /*
1488                  * We are exiting and our parent doesn't care.  POSIX.1
1489                  * defines special semantics for setting SIGCHLD to SIG_IGN
1490                  * or setting the SA_NOCLDWAIT flag: we should be reaped
1491                  * automatically and not left for our parent's wait4 call.
1492                  * Rather than having the parent do it as a magic kind of
1493                  * signal handler, we just set this to tell do_exit that we
1494                  * can be cleaned up without becoming a zombie.  Note that
1495                  * we still call __wake_up_parent in this case, because a
1496                  * blocked sys_wait4 might now return -ECHILD.
1497                  *
1498                  * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1499                  * is implementation-defined: we do (if you don't want
1500                  * it, just use SIG_IGN instead).
1501                  */
1502                 tsk->exit_signal = -1;
1503                 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1504                         sig = 0;
1505         }
1506         if (sig > 0 && sig <= _NSIG)
1507                 __group_send_sig_info(sig, &info, tsk->parent);
1508         __wake_up_parent(tsk, tsk->parent);
1509         spin_unlock_irqrestore(&psig->siglock, flags);
1510 }
1511
1512 static void
1513 do_notify_parent_cldstop(struct task_struct *tsk, struct task_struct *parent,
1514                          int why)
1515 {
1516         struct siginfo info;
1517         unsigned long flags;
1518         struct sighand_struct *sighand;
1519
1520         info.si_signo = SIGCHLD;
1521         info.si_errno = 0;
1522         info.si_pid = tsk->pid;
1523         info.si_uid = tsk->uid;
1524
1525         /* FIXME: find out whether or not this is supposed to be c*time. */
1526         info.si_utime = cputime_to_jiffies(tsk->utime);
1527         info.si_stime = cputime_to_jiffies(tsk->stime);
1528
1529         info.si_code = why;
1530         switch (why) {
1531         case CLD_CONTINUED:
1532                 info.si_status = SIGCONT;
1533                 break;
1534         case CLD_STOPPED:
1535                 info.si_status = tsk->signal->group_exit_code & 0x7f;
1536                 break;
1537         case CLD_TRAPPED:
1538                 info.si_status = tsk->exit_code & 0x7f;
1539                 break;
1540         default:
1541                 BUG();
1542         }
1543
1544         sighand = parent->sighand;
1545         spin_lock_irqsave(&sighand->siglock, flags);
1546         if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1547             !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1548                 __group_send_sig_info(SIGCHLD, &info, parent);
1549         /*
1550          * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1551          */
1552         __wake_up_parent(tsk, parent);
1553         spin_unlock_irqrestore(&sighand->siglock, flags);
1554 }
1555
1556 /*
1557  * This must be called with current->sighand->siglock held.
1558  *
1559  * This should be the path for all ptrace stops.
1560  * We always set current->last_siginfo while stopped here.
1561  * That makes it a way to test a stopped process for
1562  * being ptrace-stopped vs being job-control-stopped.
1563  *
1564  * If we actually decide not to stop at all because the tracer is gone,
1565  * we leave nostop_code in current->exit_code.
1566  */
1567 static void ptrace_stop(int exit_code, int nostop_code, siginfo_t *info)
1568 {
1569         /*
1570          * If there is a group stop in progress,
1571          * we must participate in the bookkeeping.
1572          */
1573         if (current->signal->group_stop_count > 0)
1574                 --current->signal->group_stop_count;
1575
1576         current->last_siginfo = info;
1577         current->exit_code = exit_code;
1578
1579         /* Let the debugger run.  */
1580         set_current_state(TASK_TRACED);
1581         spin_unlock_irq(&current->sighand->siglock);
1582         read_lock(&tasklist_lock);
1583         if (likely(current->ptrace & PT_PTRACED) &&
1584             likely(current->parent != current->real_parent ||
1585                    !(current->ptrace & PT_ATTACHED)) &&
1586             (likely(current->parent->signal != current->signal) ||
1587              !unlikely(current->signal->flags & SIGNAL_GROUP_EXIT))) {
1588                 do_notify_parent_cldstop(current, current->parent,
1589                                          CLD_TRAPPED);
1590                 read_unlock(&tasklist_lock);
1591                 schedule();
1592         } else {
1593                 /*
1594                  * By the time we got the lock, our tracer went away.
1595                  * Don't stop here.
1596                  */
1597                 read_unlock(&tasklist_lock);
1598                 set_current_state(TASK_RUNNING);
1599                 current->exit_code = nostop_code;
1600         }
1601
1602         /*
1603          * We are back.  Now reacquire the siglock before touching
1604          * last_siginfo, so that we are sure to have synchronized with
1605          * any signal-sending on another CPU that wants to examine it.
1606          */
1607         spin_lock_irq(&current->sighand->siglock);
1608         current->last_siginfo = NULL;
1609
1610         /*
1611          * Queued signals ignored us while we were stopped for tracing.
1612          * So check for any that we should take before resuming user mode.
1613          */
1614         recalc_sigpending();
1615 }
1616
1617 void ptrace_notify(int exit_code)
1618 {
1619         siginfo_t info;
1620
1621         BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1622
1623         memset(&info, 0, sizeof info);
1624         info.si_signo = SIGTRAP;
1625         info.si_code = exit_code;
1626         info.si_pid = current->pid;
1627         info.si_uid = current->uid;
1628
1629         /* Let the debugger run.  */
1630         spin_lock_irq(&current->sighand->siglock);
1631         ptrace_stop(exit_code, 0, &info);
1632         spin_unlock_irq(&current->sighand->siglock);
1633 }
1634
1635 #ifndef HAVE_ARCH_GET_SIGNAL_TO_DELIVER
1636
1637 static void
1638 finish_stop(int stop_count)
1639 {
1640         /*
1641          * If there are no other threads in the group, or if there is
1642          * a group stop in progress and we are the last to stop,
1643          * report to the parent.  When ptraced, every thread reports itself.
1644          */
1645         if (stop_count < 0 || (current->ptrace & PT_PTRACED)) {
1646                 read_lock(&tasklist_lock);
1647                 do_notify_parent_cldstop(current, current->parent,
1648                                          CLD_STOPPED);
1649                 read_unlock(&tasklist_lock);
1650         }
1651         else if (stop_count == 0) {
1652                 read_lock(&tasklist_lock);
1653                 do_notify_parent_cldstop(current->group_leader,
1654                                          current->group_leader->real_parent,
1655                                          CLD_STOPPED);
1656                 read_unlock(&tasklist_lock);
1657         }
1658
1659         schedule();
1660         /*
1661          * Now we don't run again until continued.
1662          */
1663         current->exit_code = 0;
1664 }
1665
1666 /*
1667  * This performs the stopping for SIGSTOP and other stop signals.
1668  * We have to stop all threads in the thread group.
1669  * Returns nonzero if we've actually stopped and released the siglock.
1670  * Returns zero if we didn't stop and still hold the siglock.
1671  */
1672 static int
1673 do_signal_stop(int signr)
1674 {
1675         struct signal_struct *sig = current->signal;
1676         struct sighand_struct *sighand = current->sighand;
1677         int stop_count = -1;
1678
1679         if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED))
1680                 return 0;
1681
1682         if (sig->group_stop_count > 0) {
1683                 /*
1684                  * There is a group stop in progress.  We don't need to
1685                  * start another one.
1686                  */
1687                 signr = sig->group_exit_code;
1688                 stop_count = --sig->group_stop_count;
1689                 current->exit_code = signr;
1690                 set_current_state(TASK_STOPPED);
1691                 if (stop_count == 0)
1692                         sig->flags = SIGNAL_STOP_STOPPED;
1693                 spin_unlock_irq(&sighand->siglock);
1694         }
1695         else if (thread_group_empty(current)) {
1696                 /*
1697                  * Lock must be held through transition to stopped state.
1698                  */
1699                 current->exit_code = current->signal->group_exit_code = signr;
1700                 set_current_state(TASK_STOPPED);
1701                 sig->flags = SIGNAL_STOP_STOPPED;
1702                 spin_unlock_irq(&sighand->siglock);
1703         }
1704         else {
1705                 /*
1706                  * There is no group stop already in progress.
1707                  * We must initiate one now, but that requires
1708                  * dropping siglock to get both the tasklist lock
1709                  * and siglock again in the proper order.  Note that
1710                  * this allows an intervening SIGCONT to be posted.
1711                  * We need to check for that and bail out if necessary.
1712                  */
1713                 struct task_struct *t;
1714
1715                 spin_unlock_irq(&sighand->siglock);
1716
1717                 /* signals can be posted during this window */
1718
1719                 read_lock(&tasklist_lock);
1720                 spin_lock_irq(&sighand->siglock);
1721
1722                 if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED)) {
1723                         /*
1724                          * Another stop or continue happened while we
1725                          * didn't have the lock.  We can just swallow this
1726                          * signal now.  If we raced with a SIGCONT, that
1727                          * should have just cleared it now.  If we raced
1728                          * with another processor delivering a stop signal,
1729                          * then the SIGCONT that wakes us up should clear it.
1730                          */
1731                         read_unlock(&tasklist_lock);
1732                         return 0;
1733                 }
1734
1735                 if (sig->group_stop_count == 0) {
1736                         sig->group_exit_code = signr;
1737                         stop_count = 0;
1738                         for (t = next_thread(current); t != current;
1739                              t = next_thread(t))
1740                                 /*
1741                                  * Setting state to TASK_STOPPED for a group
1742                                  * stop is always done with the siglock held,
1743                                  * so this check has no races.
1744                                  */
1745                                 if (t->state < TASK_STOPPED) {
1746                                         stop_count++;
1747                                         signal_wake_up(t, 0);
1748                                 }
1749                         sig->group_stop_count = stop_count;
1750                 }
1751                 else {
1752                         /* A race with another thread while unlocked.  */
1753                         signr = sig->group_exit_code;
1754                         stop_count = --sig->group_stop_count;
1755                 }
1756
1757                 current->exit_code = signr;
1758                 set_current_state(TASK_STOPPED);
1759                 if (stop_count == 0)
1760                         sig->flags = SIGNAL_STOP_STOPPED;
1761
1762                 spin_unlock_irq(&sighand->siglock);
1763                 read_unlock(&tasklist_lock);
1764         }
1765
1766         finish_stop(stop_count);
1767         return 1;
1768 }
1769
1770 /*
1771  * Do appropriate magic when group_stop_count > 0.
1772  * We return nonzero if we stopped, after releasing the siglock.
1773  * We return zero if we still hold the siglock and should look
1774  * for another signal without checking group_stop_count again.
1775  */
1776 static inline int handle_group_stop(void)
1777 {
1778         int stop_count;
1779
1780         if (current->signal->group_exit_task == current) {
1781                 /*
1782                  * Group stop is so we can do a core dump,
1783                  * We are the initiating thread, so get on with it.
1784                  */
1785                 current->signal->group_exit_task = NULL;
1786                 return 0;
1787         }
1788
1789         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1790                 /*
1791                  * Group stop is so another thread can do a core dump,
1792                  * or else we are racing against a death signal.
1793                  * Just punt the stop so we can get the next signal.
1794                  */
1795                 return 0;
1796
1797         /*
1798          * There is a group stop in progress.  We stop
1799          * without any associated signal being in our queue.
1800          */
1801         stop_count = --current->signal->group_stop_count;
1802         if (stop_count == 0)
1803                 current->signal->flags = SIGNAL_STOP_STOPPED;
1804         current->exit_code = current->signal->group_exit_code;
1805         set_current_state(TASK_STOPPED);
1806         spin_unlock_irq(&current->sighand->siglock);
1807         finish_stop(stop_count);
1808         return 1;
1809 }
1810
1811 int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
1812                           struct pt_regs *regs, void *cookie)
1813 {
1814         sigset_t *mask = &current->blocked;
1815         int signr = 0;
1816
1817 relock:
1818         spin_lock_irq(&current->sighand->siglock);
1819         for (;;) {
1820                 struct k_sigaction *ka;
1821
1822                 if (unlikely(current->signal->group_stop_count > 0) &&
1823                     handle_group_stop())
1824                         goto relock;
1825
1826                 signr = dequeue_signal(current, mask, info);
1827
1828                 if (!signr)
1829                         break; /* will return 0 */
1830
1831                 if ((current->ptrace & PT_PTRACED) && signr != SIGKILL) {
1832                         ptrace_signal_deliver(regs, cookie);
1833
1834                         /* Let the debugger run.  */
1835                         ptrace_stop(signr, signr, info);
1836
1837                         /* We're back.  Did the debugger cancel the sig?  */
1838                         signr = current->exit_code;
1839                         if (signr == 0)
1840                                 continue;
1841
1842                         current->exit_code = 0;
1843
1844                         /* Update the siginfo structure if the signal has
1845                            changed.  If the debugger wanted something
1846                            specific in the siginfo structure then it should
1847                            have updated *info via PTRACE_SETSIGINFO.  */
1848                         if (signr != info->si_signo) {
1849                                 info->si_signo = signr;
1850                                 info->si_errno = 0;
1851                                 info->si_code = SI_USER;
1852                                 info->si_pid = current->parent->pid;
1853                                 info->si_uid = current->parent->uid;
1854                         }
1855
1856                         /* If the (new) signal is now blocked, requeue it.  */
1857                         if (sigismember(&current->blocked, signr)) {
1858                                 specific_send_sig_info(signr, info, current);
1859                                 continue;
1860                         }
1861                 }
1862
1863                 ka = &current->sighand->action[signr-1];
1864                 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
1865                         continue;
1866                 if (ka->sa.sa_handler != SIG_DFL) {
1867                         /* Run the handler.  */
1868                         *return_ka = *ka;
1869
1870                         if (ka->sa.sa_flags & SA_ONESHOT)
1871                                 ka->sa.sa_handler = SIG_DFL;
1872
1873                         break; /* will return non-zero "signr" value */
1874                 }
1875
1876                 /*
1877                  * Now we are doing the default action for this signal.
1878                  */
1879                 if (sig_kernel_ignore(signr)) /* Default is nothing. */
1880                         continue;
1881
1882                 /* Init gets no signals it doesn't want.  */
1883                 if (current->pid == 1)
1884                         continue;
1885
1886                 if (sig_kernel_stop(signr)) {
1887                         /*
1888                          * The default action is to stop all threads in
1889                          * the thread group.  The job control signals
1890                          * do nothing in an orphaned pgrp, but SIGSTOP
1891                          * always works.  Note that siglock needs to be
1892                          * dropped during the call to is_orphaned_pgrp()
1893                          * because of lock ordering with tasklist_lock.
1894                          * This allows an intervening SIGCONT to be posted.
1895                          * We need to check for that and bail out if necessary.
1896                          */
1897                         if (signr != SIGSTOP) {
1898                                 spin_unlock_irq(&current->sighand->siglock);
1899
1900                                 /* signals can be posted during this window */
1901
1902                                 if (is_orphaned_pgrp(process_group(current)))
1903                                         goto relock;
1904
1905                                 spin_lock_irq(&current->sighand->siglock);
1906                         }
1907
1908                         if (likely(do_signal_stop(signr))) {
1909                                 /* It released the siglock.  */
1910                                 goto relock;
1911                         }
1912
1913                         /*
1914                          * We didn't actually stop, due to a race
1915                          * with SIGCONT or something like that.
1916                          */
1917                         continue;
1918                 }
1919
1920                 spin_unlock_irq(&current->sighand->siglock);
1921
1922                 /*
1923                  * Anything else is fatal, maybe with a core dump.
1924                  */
1925                 current->flags |= PF_SIGNALED;
1926                 if (sig_kernel_coredump(signr)) {
1927                         /*
1928                          * If it was able to dump core, this kills all
1929                          * other threads in the group and synchronizes with
1930                          * their demise.  If we lost the race with another
1931                          * thread getting here, it set group_exit_code
1932                          * first and our do_group_exit call below will use
1933                          * that value and ignore the one we pass it.
1934                          */
1935                         do_coredump((long)signr, signr, regs);
1936                 }
1937
1938                 /*
1939                  * Death signals, no core dump.
1940                  */
1941                 do_group_exit(signr);
1942                 /* NOTREACHED */
1943         }
1944         spin_unlock_irq(&current->sighand->siglock);
1945         return signr;
1946 }
1947
1948 #endif
1949
1950 EXPORT_SYMBOL(recalc_sigpending);
1951 EXPORT_SYMBOL_GPL(dequeue_signal);
1952 EXPORT_SYMBOL(flush_signals);
1953 EXPORT_SYMBOL(force_sig);
1954 EXPORT_SYMBOL(kill_pg);
1955 EXPORT_SYMBOL(kill_proc);
1956 EXPORT_SYMBOL(ptrace_notify);
1957 EXPORT_SYMBOL(send_sig);
1958 EXPORT_SYMBOL(send_sig_info);
1959 EXPORT_SYMBOL(sigprocmask);
1960 EXPORT_SYMBOL(block_all_signals);
1961 EXPORT_SYMBOL(unblock_all_signals);
1962
1963
1964 /*
1965  * System call entry points.
1966  */
1967
1968 asmlinkage long sys_restart_syscall(void)
1969 {
1970         struct restart_block *restart = &current_thread_info()->restart_block;
1971         return restart->fn(restart);
1972 }
1973
1974 long do_no_restart_syscall(struct restart_block *param)
1975 {
1976         return -EINTR;
1977 }
1978
1979 /*
1980  * We don't need to get the kernel lock - this is all local to this
1981  * particular thread.. (and that's good, because this is _heavily_
1982  * used by various programs)
1983  */
1984
1985 /*
1986  * This is also useful for kernel threads that want to temporarily
1987  * (or permanently) block certain signals.
1988  *
1989  * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
1990  * interface happily blocks "unblockable" signals like SIGKILL
1991  * and friends.
1992  */
1993 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
1994 {
1995         int error;
1996         sigset_t old_block;
1997
1998         spin_lock_irq(&current->sighand->siglock);
1999         old_block = current->blocked;
2000         error = 0;
2001         switch (how) {
2002         case SIG_BLOCK:
2003                 sigorsets(&current->blocked, &current->blocked, set);
2004                 break;
2005         case SIG_UNBLOCK:
2006                 signandsets(&current->blocked, &current->blocked, set);
2007                 break;
2008         case SIG_SETMASK:
2009                 current->blocked = *set;
2010                 break;
2011         default:
2012                 error = -EINVAL;
2013         }
2014         recalc_sigpending();
2015         spin_unlock_irq(&current->sighand->siglock);
2016         if (oldset)
2017                 *oldset = old_block;
2018         return error;
2019 }
2020
2021 asmlinkage long
2022 sys_rt_sigprocmask(int how, sigset_t __user *set, sigset_t __user *oset, size_t sigsetsize)
2023 {
2024         int error = -EINVAL;
2025         sigset_t old_set, new_set;
2026
2027         /* XXX: Don't preclude handling different sized sigset_t's.  */
2028         if (sigsetsize != sizeof(sigset_t))
2029                 goto out;
2030
2031         if (set) {
2032                 error = -EFAULT;
2033                 if (copy_from_user(&new_set, set, sizeof(*set)))
2034                         goto out;
2035                 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2036
2037                 error = sigprocmask(how, &new_set, &old_set);
2038                 if (error)
2039                         goto out;
2040                 if (oset)
2041                         goto set_old;
2042         } else if (oset) {
2043                 spin_lock_irq(&current->sighand->siglock);
2044                 old_set = current->blocked;
2045                 spin_unlock_irq(&current->sighand->siglock);
2046
2047         set_old:
2048                 error = -EFAULT;
2049                 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2050                         goto out;
2051         }
2052         error = 0;
2053 out:
2054         return error;
2055 }
2056
2057 long do_sigpending(void __user *set, unsigned long sigsetsize)
2058 {
2059         long error = -EINVAL;
2060         sigset_t pending;
2061
2062         if (sigsetsize > sizeof(sigset_t))
2063                 goto out;
2064
2065         spin_lock_irq(&current->sighand->siglock);
2066         sigorsets(&pending, &current->pending.signal,
2067                   &current->signal->shared_pending.signal);
2068         spin_unlock_irq(&current->sighand->siglock);
2069
2070         /* Outside the lock because only this thread touches it.  */
2071         sigandsets(&pending, &current->blocked, &pending);
2072
2073         error = -EFAULT;
2074         if (!copy_to_user(set, &pending, sigsetsize))
2075                 error = 0;
2076
2077 out:
2078         return error;
2079 }       
2080
2081 asmlinkage long
2082 sys_rt_sigpending(sigset_t __user *set, size_t sigsetsize)
2083 {
2084         return do_sigpending(set, sigsetsize);
2085 }
2086
2087 #ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2088
2089 int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2090 {
2091         int err;
2092
2093         if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2094                 return -EFAULT;
2095         if (from->si_code < 0)
2096                 return __copy_to_user(to, from, sizeof(siginfo_t))
2097                         ? -EFAULT : 0;
2098         /*
2099          * If you change siginfo_t structure, please be sure
2100          * this code is fixed accordingly.
2101          * It should never copy any pad contained in the structure
2102          * to avoid security leaks, but must copy the generic
2103          * 3 ints plus the relevant union member.
2104          */
2105         err = __put_user(from->si_signo, &to->si_signo);
2106         err |= __put_user(from->si_errno, &to->si_errno);
2107         err |= __put_user((short)from->si_code, &to->si_code);
2108         switch (from->si_code & __SI_MASK) {
2109         case __SI_KILL:
2110                 err |= __put_user(from->si_pid, &to->si_pid);
2111                 err |= __put_user(from->si_uid, &to->si_uid);
2112                 break;
2113         case __SI_TIMER:
2114                  err |= __put_user(from->si_tid, &to->si_tid);
2115                  err |= __put_user(from->si_overrun, &to->si_overrun);
2116                  err |= __put_user(from->si_ptr, &to->si_ptr);
2117                 break;
2118         case __SI_POLL:
2119                 err |= __put_user(from->si_band, &to->si_band);
2120                 err |= __put_user(from->si_fd, &to->si_fd);
2121                 break;
2122         case __SI_FAULT:
2123                 err |= __put_user(from->si_addr, &to->si_addr);
2124 #ifdef __ARCH_SI_TRAPNO
2125                 err |= __put_user(from->si_trapno, &to->si_trapno);
2126 #endif
2127                 break;
2128         case __SI_CHLD:
2129                 err |= __put_user(from->si_pid, &to->si_pid);
2130                 err |= __put_user(from->si_uid, &to->si_uid);
2131                 err |= __put_user(from->si_status, &to->si_status);
2132                 err |= __put_user(from->si_utime, &to->si_utime);
2133                 err |= __put_user(from->si_stime, &to->si_stime);
2134                 break;
2135         case __SI_RT: /* This is not generated by the kernel as of now. */
2136         case __SI_MESGQ: /* But this is */
2137                 err |= __put_user(from->si_pid, &to->si_pid);
2138                 err |= __put_user(from->si_uid, &to->si_uid);
2139                 err |= __put_user(from->si_ptr, &to->si_ptr);
2140                 break;
2141         default: /* this is just in case for now ... */
2142                 err |= __put_user(from->si_pid, &to->si_pid);
2143                 err |= __put_user(from->si_uid, &to->si_uid);
2144                 break;
2145         }
2146         return err;
2147 }
2148
2149 #endif
2150
2151 asmlinkage long
2152 sys_rt_sigtimedwait(const sigset_t __user *uthese,
2153                     siginfo_t __user *uinfo,
2154                     const struct timespec __user *uts,
2155                     size_t sigsetsize)
2156 {
2157         int ret, sig;
2158         sigset_t these;
2159         struct timespec ts;
2160         siginfo_t info;
2161         long timeout = 0;
2162
2163         /* XXX: Don't preclude handling different sized sigset_t's.  */
2164         if (sigsetsize != sizeof(sigset_t))
2165                 return -EINVAL;
2166
2167         if (copy_from_user(&these, uthese, sizeof(these)))
2168                 return -EFAULT;
2169                 
2170         /*
2171          * Invert the set of allowed signals to get those we
2172          * want to block.
2173          */
2174         sigdelsetmask(&these, sigmask(SIGKILL)|sigmask(SIGSTOP));
2175         signotset(&these);
2176
2177         if (uts) {
2178                 if (copy_from_user(&ts, uts, sizeof(ts)))
2179                         return -EFAULT;
2180                 if (ts.tv_nsec >= 1000000000L || ts.tv_nsec < 0
2181                     || ts.tv_sec < 0)
2182                         return -EINVAL;
2183         }
2184
2185         spin_lock_irq(&current->sighand->siglock);
2186         sig = dequeue_signal(current, &these, &info);
2187         if (!sig) {
2188                 timeout = MAX_SCHEDULE_TIMEOUT;
2189                 if (uts)
2190                         timeout = (timespec_to_jiffies(&ts)
2191                                    + (ts.tv_sec || ts.tv_nsec));
2192
2193                 if (timeout) {
2194                         /* None ready -- temporarily unblock those we're
2195                          * interested while we are sleeping in so that we'll
2196                          * be awakened when they arrive.  */
2197                         current->real_blocked = current->blocked;
2198                         sigandsets(&current->blocked, &current->blocked, &these);
2199                         recalc_sigpending();
2200                         spin_unlock_irq(&current->sighand->siglock);
2201
2202                         current->state = TASK_INTERRUPTIBLE;
2203                         timeout = schedule_timeout(timeout);
2204
2205                         spin_lock_irq(&current->sighand->siglock);
2206                         sig = dequeue_signal(current, &these, &info);
2207                         current->blocked = current->real_blocked;
2208                         siginitset(&current->real_blocked, 0);
2209                         recalc_sigpending();
2210                 }
2211         }
2212         spin_unlock_irq(&current->sighand->siglock);
2213
2214         if (sig) {
2215                 ret = sig;
2216                 if (uinfo) {
2217                         if (copy_siginfo_to_user(uinfo, &info))
2218                                 ret = -EFAULT;
2219                 }
2220         } else {
2221                 ret = -EAGAIN;
2222                 if (timeout)
2223                         ret = -EINTR;
2224         }
2225
2226         return ret;
2227 }
2228
2229 asmlinkage long
2230 sys_kill(int pid, int sig)
2231 {
2232         struct siginfo info;
2233
2234         info.si_signo = sig;
2235         info.si_errno = 0;
2236         info.si_code = SI_USER;
2237         info.si_pid = current->tgid;
2238         info.si_uid = current->uid;
2239
2240         return kill_something_info(sig, &info, pid);
2241 }
2242
2243 /**
2244  *  sys_tgkill - send signal to one specific thread
2245  *  @tgid: the thread group ID of the thread
2246  *  @pid: the PID of the thread
2247  *  @sig: signal to be sent
2248  *
2249  *  This syscall also checks the tgid and returns -ESRCH even if the PID
2250  *  exists but it's not belonging to the target process anymore. This
2251  *  method solves the problem of threads exiting and PIDs getting reused.
2252  */
2253 asmlinkage long sys_tgkill(int tgid, int pid, int sig)
2254 {
2255         struct siginfo info;
2256         int error;
2257         struct task_struct *p;
2258
2259         /* This is only valid for single tasks */
2260         if (pid <= 0 || tgid <= 0)
2261                 return -EINVAL;
2262
2263         info.si_signo = sig;
2264         info.si_errno = 0;
2265         info.si_code = SI_TKILL;
2266         info.si_pid = current->tgid;
2267         info.si_uid = current->uid;
2268
2269         read_lock(&tasklist_lock);
2270         p = find_task_by_pid(pid);
2271         error = -ESRCH;
2272         if (p && (p->tgid == tgid)) {
2273                 error = check_kill_permission(sig, &info, p);
2274                 /*
2275                  * The null signal is a permissions and process existence
2276                  * probe.  No signal is actually delivered.
2277                  */
2278                 if (!error && sig && p->sighand) {
2279                         spin_lock_irq(&p->sighand->siglock);
2280                         handle_stop_signal(sig, p);
2281                         error = specific_send_sig_info(sig, &info, p);
2282                         spin_unlock_irq(&p->sighand->siglock);
2283                 }
2284         }
2285         read_unlock(&tasklist_lock);
2286         return error;
2287 }
2288
2289 /*
2290  *  Send a signal to only one task, even if it's a CLONE_THREAD task.
2291  */
2292 asmlinkage long
2293 sys_tkill(int pid, int sig)
2294 {
2295         struct siginfo info;
2296         int error;
2297         struct task_struct *p;
2298
2299         /* This is only valid for single tasks */
2300         if (pid <= 0)
2301                 return -EINVAL;
2302
2303         info.si_signo = sig;
2304         info.si_errno = 0;
2305         info.si_code = SI_TKILL;
2306         info.si_pid = current->tgid;
2307         info.si_uid = current->uid;
2308
2309         read_lock(&tasklist_lock);
2310         p = find_task_by_pid(pid);
2311         error = -ESRCH;
2312         if (p) {
2313                 error = check_kill_permission(sig, &info, p);
2314                 /*
2315                  * The null signal is a permissions and process existence
2316                  * probe.  No signal is actually delivered.
2317                  */
2318                 if (!error && sig && p->sighand) {
2319                         spin_lock_irq(&p->sighand->siglock);
2320                         handle_stop_signal(sig, p);
2321                         error = specific_send_sig_info(sig, &info, p);
2322                         spin_unlock_irq(&p->sighand->siglock);
2323                 }
2324         }
2325         read_unlock(&tasklist_lock);
2326         return error;
2327 }
2328
2329 asmlinkage long
2330 sys_rt_sigqueueinfo(int pid, int sig, siginfo_t __user *uinfo)
2331 {
2332         siginfo_t info;
2333
2334         if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2335                 return -EFAULT;
2336
2337         /* Not even root can pretend to send signals from the kernel.
2338            Nor can they impersonate a kill(), which adds source info.  */
2339         if (info.si_code >= 0)
2340                 return -EPERM;
2341         info.si_signo = sig;
2342
2343         /* POSIX.1b doesn't mention process groups.  */
2344         return kill_proc_info(sig, &info, pid);
2345 }
2346
2347 int
2348 do_sigaction(int sig, const struct k_sigaction *act, struct k_sigaction *oact)
2349 {
2350         struct k_sigaction *k;
2351
2352         if (sig < 1 || sig > _NSIG || (act && sig_kernel_only(sig)))
2353                 return -EINVAL;
2354
2355         k = &current->sighand->action[sig-1];
2356
2357         spin_lock_irq(&current->sighand->siglock);
2358         if (signal_pending(current)) {
2359                 /*
2360                  * If there might be a fatal signal pending on multiple
2361                  * threads, make sure we take it before changing the action.
2362                  */
2363                 spin_unlock_irq(&current->sighand->siglock);
2364                 return -ERESTARTNOINTR;
2365         }
2366
2367         if (oact)
2368                 *oact = *k;
2369
2370         if (act) {
2371                 /*
2372                  * POSIX 3.3.1.3:
2373                  *  "Setting a signal action to SIG_IGN for a signal that is
2374                  *   pending shall cause the pending signal to be discarded,
2375                  *   whether or not it is blocked."
2376                  *
2377                  *  "Setting a signal action to SIG_DFL for a signal that is
2378                  *   pending and whose default action is to ignore the signal
2379                  *   (for example, SIGCHLD), shall cause the pending signal to
2380                  *   be discarded, whether or not it is blocked"
2381                  */
2382                 if (act->sa.sa_handler == SIG_IGN ||
2383                     (act->sa.sa_handler == SIG_DFL &&
2384                      sig_kernel_ignore(sig))) {
2385                         /*
2386                          * This is a fairly rare case, so we only take the
2387                          * tasklist_lock once we're sure we'll need it.
2388                          * Now we must do this little unlock and relock
2389                          * dance to maintain the lock hierarchy.
2390                          */
2391                         struct task_struct *t = current;
2392                         spin_unlock_irq(&t->sighand->siglock);
2393                         read_lock(&tasklist_lock);
2394                         spin_lock_irq(&t->sighand->siglock);
2395                         *k = *act;
2396                         sigdelsetmask(&k->sa.sa_mask,
2397                                       sigmask(SIGKILL) | sigmask(SIGSTOP));
2398                         rm_from_queue(sigmask(sig), &t->signal->shared_pending);
2399                         do {
2400                                 rm_from_queue(sigmask(sig), &t->pending);
2401                                 recalc_sigpending_tsk(t);
2402                                 t = next_thread(t);
2403                         } while (t != current);
2404                         spin_unlock_irq(&current->sighand->siglock);
2405                         read_unlock(&tasklist_lock);
2406                         return 0;
2407                 }
2408
2409                 *k = *act;
2410                 sigdelsetmask(&k->sa.sa_mask,
2411                               sigmask(SIGKILL) | sigmask(SIGSTOP));
2412         }
2413
2414         spin_unlock_irq(&current->sighand->siglock);
2415         return 0;
2416 }
2417
2418 int 
2419 do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
2420 {
2421         stack_t oss;
2422         int error;
2423
2424         if (uoss) {
2425                 oss.ss_sp = (void __user *) current->sas_ss_sp;
2426                 oss.ss_size = current->sas_ss_size;
2427                 oss.ss_flags = sas_ss_flags(sp);
2428         }
2429
2430         if (uss) {
2431                 void __user *ss_sp;
2432                 size_t ss_size;
2433                 int ss_flags;
2434
2435                 error = -EFAULT;
2436                 if (verify_area(VERIFY_READ, uss, sizeof(*uss))
2437                     || __get_user(ss_sp, &uss->ss_sp)
2438                     || __get_user(ss_flags, &uss->ss_flags)
2439                     || __get_user(ss_size, &uss->ss_size))
2440                         goto out;
2441
2442                 error = -EPERM;
2443                 if (on_sig_stack(sp))
2444                         goto out;
2445
2446                 error = -EINVAL;
2447                 /*
2448                  *
2449                  * Note - this code used to test ss_flags incorrectly
2450                  *        old code may have been written using ss_flags==0
2451                  *        to mean ss_flags==SS_ONSTACK (as this was the only
2452                  *        way that worked) - this fix preserves that older
2453                  *        mechanism
2454                  */
2455                 if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
2456                         goto out;
2457
2458                 if (ss_flags == SS_DISABLE) {
2459                         ss_size = 0;
2460                         ss_sp = NULL;
2461                 } else {
2462                         error = -ENOMEM;
2463                         if (ss_size < MINSIGSTKSZ)
2464                                 goto out;
2465                 }
2466
2467                 current->sas_ss_sp = (unsigned long) ss_sp;
2468                 current->sas_ss_size = ss_size;
2469         }
2470
2471         if (uoss) {
2472                 error = -EFAULT;
2473                 if (copy_to_user(uoss, &oss, sizeof(oss)))
2474                         goto out;
2475         }
2476
2477         error = 0;
2478 out:
2479         return error;
2480 }
2481
2482 #ifdef __ARCH_WANT_SYS_SIGPENDING
2483
2484 asmlinkage long
2485 sys_sigpending(old_sigset_t __user *set)
2486 {
2487         return do_sigpending(set, sizeof(*set));
2488 }
2489
2490 #endif
2491
2492 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
2493 /* Some platforms have their own version with special arguments others
2494    support only sys_rt_sigprocmask.  */
2495
2496 asmlinkage long
2497 sys_sigprocmask(int how, old_sigset_t __user *set, old_sigset_t __user *oset)
2498 {
2499         int error;
2500         old_sigset_t old_set, new_set;
2501
2502         if (set) {
2503                 error = -EFAULT;
2504                 if (copy_from_user(&new_set, set, sizeof(*set)))
2505                         goto out;
2506                 new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
2507
2508                 spin_lock_irq(&current->sighand->siglock);
2509                 old_set = current->blocked.sig[0];
2510
2511                 error = 0;
2512                 switch (how) {
2513                 default:
2514                         error = -EINVAL;
2515                         break;
2516                 case SIG_BLOCK:
2517                         sigaddsetmask(&current->blocked, new_set);
2518                         break;
2519                 case SIG_UNBLOCK:
2520                         sigdelsetmask(&current->blocked, new_set);
2521                         break;
2522                 case SIG_SETMASK:
2523                         current->blocked.sig[0] = new_set;
2524                         break;
2525                 }
2526
2527                 recalc_sigpending();
2528                 spin_unlock_irq(&current->sighand->siglock);
2529                 if (error)
2530                         goto out;
2531                 if (oset)
2532                         goto set_old;
2533         } else if (oset) {
2534                 old_set = current->blocked.sig[0];
2535         set_old:
2536                 error = -EFAULT;
2537                 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2538                         goto out;
2539         }
2540         error = 0;
2541 out:
2542         return error;
2543 }
2544 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
2545
2546 #ifdef __ARCH_WANT_SYS_RT_SIGACTION
2547 asmlinkage long
2548 sys_rt_sigaction(int sig,
2549                  const struct sigaction __user *act,
2550                  struct sigaction __user *oact,
2551                  size_t sigsetsize)
2552 {
2553         struct k_sigaction new_sa, old_sa;
2554         int ret = -EINVAL;
2555
2556         /* XXX: Don't preclude handling different sized sigset_t's.  */
2557         if (sigsetsize != sizeof(sigset_t))
2558                 goto out;
2559
2560         if (act) {
2561                 if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
2562                         return -EFAULT;
2563         }
2564
2565         ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
2566
2567         if (!ret && oact) {
2568                 if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
2569                         return -EFAULT;
2570         }
2571 out:
2572         return ret;
2573 }
2574 #endif /* __ARCH_WANT_SYS_RT_SIGACTION */
2575
2576 #ifdef __ARCH_WANT_SYS_SGETMASK
2577
2578 /*
2579  * For backwards compatibility.  Functionality superseded by sigprocmask.
2580  */
2581 asmlinkage long
2582 sys_sgetmask(void)
2583 {
2584         /* SMP safe */
2585         return current->blocked.sig[0];
2586 }
2587
2588 asmlinkage long
2589 sys_ssetmask(int newmask)
2590 {
2591         int old;
2592
2593         spin_lock_irq(&current->sighand->siglock);
2594         old = current->blocked.sig[0];
2595
2596         siginitset(&current->blocked, newmask & ~(sigmask(SIGKILL)|
2597                                                   sigmask(SIGSTOP)));
2598         recalc_sigpending();
2599         spin_unlock_irq(&current->sighand->siglock);
2600
2601         return old;
2602 }
2603 #endif /* __ARCH_WANT_SGETMASK */
2604
2605 #ifdef __ARCH_WANT_SYS_SIGNAL
2606 /*
2607  * For backwards compatibility.  Functionality superseded by sigaction.
2608  */
2609 asmlinkage unsigned long
2610 sys_signal(int sig, __sighandler_t handler)
2611 {
2612         struct k_sigaction new_sa, old_sa;
2613         int ret;
2614
2615         new_sa.sa.sa_handler = handler;
2616         new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
2617
2618         ret = do_sigaction(sig, &new_sa, &old_sa);
2619
2620         return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
2621 }
2622 #endif /* __ARCH_WANT_SYS_SIGNAL */
2623
2624 #ifdef __ARCH_WANT_SYS_PAUSE
2625
2626 asmlinkage long
2627 sys_pause(void)
2628 {
2629         current->state = TASK_INTERRUPTIBLE;
2630         schedule();
2631         return -ERESTARTNOHAND;
2632 }
2633
2634 #endif
2635
2636 void __init signals_init(void)
2637 {
2638         sigqueue_cachep =
2639                 kmem_cache_create("sigqueue",
2640                                   sizeof(struct sigqueue),
2641                                   __alignof__(struct sigqueue),
2642                                   SLAB_PANIC, NULL, NULL);
2643 }