]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - kernel/signal.c
Merge branch 'master' of /usr/src/ntfs-2.6/
[karo-tx-linux.git] / kernel / signal.c
1 /*
2  *  linux/kernel/signal.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  *
6  *  1997-11-02  Modified for POSIX.1b signals by Richard Henderson
7  *
8  *  2003-06-02  Jim Houston - Concurrent Computer Corp.
9  *              Changes to use preallocated sigqueue structures
10  *              to allow signals to be sent reliably.
11  */
12
13 #include <linux/config.h>
14 #include <linux/slab.h>
15 #include <linux/module.h>
16 #include <linux/smp_lock.h>
17 #include <linux/init.h>
18 #include <linux/sched.h>
19 #include <linux/fs.h>
20 #include <linux/tty.h>
21 #include <linux/binfmts.h>
22 #include <linux/security.h>
23 #include <linux/syscalls.h>
24 #include <linux/ptrace.h>
25 #include <linux/posix-timers.h>
26 #include <linux/signal.h>
27 #include <linux/audit.h>
28 #include <asm/param.h>
29 #include <asm/uaccess.h>
30 #include <asm/unistd.h>
31 #include <asm/siginfo.h>
32
33 /*
34  * SLAB caches for signal bits.
35  */
36
37 static kmem_cache_t *sigqueue_cachep;
38
39 /*
40  * In POSIX a signal is sent either to a specific thread (Linux task)
41  * or to the process as a whole (Linux thread group).  How the signal
42  * is sent determines whether it's to one thread or the whole group,
43  * which determines which signal mask(s) are involved in blocking it
44  * from being delivered until later.  When the signal is delivered,
45  * either it's caught or ignored by a user handler or it has a default
46  * effect that applies to the whole thread group (POSIX process).
47  *
48  * The possible effects an unblocked signal set to SIG_DFL can have are:
49  *   ignore     - Nothing Happens
50  *   terminate  - kill the process, i.e. all threads in the group,
51  *                similar to exit_group.  The group leader (only) reports
52  *                WIFSIGNALED status to its parent.
53  *   coredump   - write a core dump file describing all threads using
54  *                the same mm and then kill all those threads
55  *   stop       - stop all the threads in the group, i.e. TASK_STOPPED state
56  *
57  * SIGKILL and SIGSTOP cannot be caught, blocked, or ignored.
58  * Other signals when not blocked and set to SIG_DFL behaves as follows.
59  * The job control signals also have other special effects.
60  *
61  *      +--------------------+------------------+
62  *      |  POSIX signal      |  default action  |
63  *      +--------------------+------------------+
64  *      |  SIGHUP            |  terminate       |
65  *      |  SIGINT            |  terminate       |
66  *      |  SIGQUIT           |  coredump        |
67  *      |  SIGILL            |  coredump        |
68  *      |  SIGTRAP           |  coredump        |
69  *      |  SIGABRT/SIGIOT    |  coredump        |
70  *      |  SIGBUS            |  coredump        |
71  *      |  SIGFPE            |  coredump        |
72  *      |  SIGKILL           |  terminate(+)    |
73  *      |  SIGUSR1           |  terminate       |
74  *      |  SIGSEGV           |  coredump        |
75  *      |  SIGUSR2           |  terminate       |
76  *      |  SIGPIPE           |  terminate       |
77  *      |  SIGALRM           |  terminate       |
78  *      |  SIGTERM           |  terminate       |
79  *      |  SIGCHLD           |  ignore          |
80  *      |  SIGCONT           |  ignore(*)       |
81  *      |  SIGSTOP           |  stop(*)(+)      |
82  *      |  SIGTSTP           |  stop(*)         |
83  *      |  SIGTTIN           |  stop(*)         |
84  *      |  SIGTTOU           |  stop(*)         |
85  *      |  SIGURG            |  ignore          |
86  *      |  SIGXCPU           |  coredump        |
87  *      |  SIGXFSZ           |  coredump        |
88  *      |  SIGVTALRM         |  terminate       |
89  *      |  SIGPROF           |  terminate       |
90  *      |  SIGPOLL/SIGIO     |  terminate       |
91  *      |  SIGSYS/SIGUNUSED  |  coredump        |
92  *      |  SIGSTKFLT         |  terminate       |
93  *      |  SIGWINCH          |  ignore          |
94  *      |  SIGPWR            |  terminate       |
95  *      |  SIGRTMIN-SIGRTMAX |  terminate       |
96  *      +--------------------+------------------+
97  *      |  non-POSIX signal  |  default action  |
98  *      +--------------------+------------------+
99  *      |  SIGEMT            |  coredump        |
100  *      +--------------------+------------------+
101  *
102  * (+) For SIGKILL and SIGSTOP the action is "always", not just "default".
103  * (*) Special job control effects:
104  * When SIGCONT is sent, it resumes the process (all threads in the group)
105  * from TASK_STOPPED state and also clears any pending/queued stop signals
106  * (any of those marked with "stop(*)").  This happens regardless of blocking,
107  * catching, or ignoring SIGCONT.  When any stop signal is sent, it clears
108  * any pending/queued SIGCONT signals; this happens regardless of blocking,
109  * catching, or ignored the stop signal, though (except for SIGSTOP) the
110  * default action of stopping the process may happen later or never.
111  */
112
113 #ifdef SIGEMT
114 #define M_SIGEMT        M(SIGEMT)
115 #else
116 #define M_SIGEMT        0
117 #endif
118
119 #if SIGRTMIN > BITS_PER_LONG
120 #define M(sig) (1ULL << ((sig)-1))
121 #else
122 #define M(sig) (1UL << ((sig)-1))
123 #endif
124 #define T(sig, mask) (M(sig) & (mask))
125
126 #define SIG_KERNEL_ONLY_MASK (\
127         M(SIGKILL)   |  M(SIGSTOP)                                   )
128
129 #define SIG_KERNEL_STOP_MASK (\
130         M(SIGSTOP)   |  M(SIGTSTP)   |  M(SIGTTIN)   |  M(SIGTTOU)   )
131
132 #define SIG_KERNEL_COREDUMP_MASK (\
133         M(SIGQUIT)   |  M(SIGILL)    |  M(SIGTRAP)   |  M(SIGABRT)   | \
134         M(SIGFPE)    |  M(SIGSEGV)   |  M(SIGBUS)    |  M(SIGSYS)    | \
135         M(SIGXCPU)   |  M(SIGXFSZ)   |  M_SIGEMT                     )
136
137 #define SIG_KERNEL_IGNORE_MASK (\
138         M(SIGCONT)   |  M(SIGCHLD)   |  M(SIGWINCH)  |  M(SIGURG)    )
139
140 #define sig_kernel_only(sig) \
141                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_ONLY_MASK))
142 #define sig_kernel_coredump(sig) \
143                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_COREDUMP_MASK))
144 #define sig_kernel_ignore(sig) \
145                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_IGNORE_MASK))
146 #define sig_kernel_stop(sig) \
147                 (((sig) < SIGRTMIN)  && T(sig, SIG_KERNEL_STOP_MASK))
148
149 #define sig_user_defined(t, signr) \
150         (((t)->sighand->action[(signr)-1].sa.sa_handler != SIG_DFL) &&  \
151          ((t)->sighand->action[(signr)-1].sa.sa_handler != SIG_IGN))
152
153 #define sig_fatal(t, signr) \
154         (!T(signr, SIG_KERNEL_IGNORE_MASK|SIG_KERNEL_STOP_MASK) && \
155          (t)->sighand->action[(signr)-1].sa.sa_handler == SIG_DFL)
156
157 static int sig_ignored(struct task_struct *t, int sig)
158 {
159         void __user * handler;
160
161         /*
162          * Tracers always want to know about signals..
163          */
164         if (t->ptrace & PT_PTRACED)
165                 return 0;
166
167         /*
168          * Blocked signals are never ignored, since the
169          * signal handler may change by the time it is
170          * unblocked.
171          */
172         if (sigismember(&t->blocked, sig))
173                 return 0;
174
175         /* Is it explicitly or implicitly ignored? */
176         handler = t->sighand->action[sig-1].sa.sa_handler;
177         return   handler == SIG_IGN ||
178                 (handler == SIG_DFL && sig_kernel_ignore(sig));
179 }
180
181 /*
182  * Re-calculate pending state from the set of locally pending
183  * signals, globally pending signals, and blocked signals.
184  */
185 static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
186 {
187         unsigned long ready;
188         long i;
189
190         switch (_NSIG_WORDS) {
191         default:
192                 for (i = _NSIG_WORDS, ready = 0; --i >= 0 ;)
193                         ready |= signal->sig[i] &~ blocked->sig[i];
194                 break;
195
196         case 4: ready  = signal->sig[3] &~ blocked->sig[3];
197                 ready |= signal->sig[2] &~ blocked->sig[2];
198                 ready |= signal->sig[1] &~ blocked->sig[1];
199                 ready |= signal->sig[0] &~ blocked->sig[0];
200                 break;
201
202         case 2: ready  = signal->sig[1] &~ blocked->sig[1];
203                 ready |= signal->sig[0] &~ blocked->sig[0];
204                 break;
205
206         case 1: ready  = signal->sig[0] &~ blocked->sig[0];
207         }
208         return ready != 0;
209 }
210
211 #define PENDING(p,b) has_pending_signals(&(p)->signal, (b))
212
213 fastcall void recalc_sigpending_tsk(struct task_struct *t)
214 {
215         if (t->signal->group_stop_count > 0 ||
216             (freezing(t)) ||
217             PENDING(&t->pending, &t->blocked) ||
218             PENDING(&t->signal->shared_pending, &t->blocked))
219                 set_tsk_thread_flag(t, TIF_SIGPENDING);
220         else
221                 clear_tsk_thread_flag(t, TIF_SIGPENDING);
222 }
223
224 void recalc_sigpending(void)
225 {
226         recalc_sigpending_tsk(current);
227 }
228
229 /* Given the mask, find the first available signal that should be serviced. */
230
231 static int
232 next_signal(struct sigpending *pending, sigset_t *mask)
233 {
234         unsigned long i, *s, *m, x;
235         int sig = 0;
236         
237         s = pending->signal.sig;
238         m = mask->sig;
239         switch (_NSIG_WORDS) {
240         default:
241                 for (i = 0; i < _NSIG_WORDS; ++i, ++s, ++m)
242                         if ((x = *s &~ *m) != 0) {
243                                 sig = ffz(~x) + i*_NSIG_BPW + 1;
244                                 break;
245                         }
246                 break;
247
248         case 2: if ((x = s[0] &~ m[0]) != 0)
249                         sig = 1;
250                 else if ((x = s[1] &~ m[1]) != 0)
251                         sig = _NSIG_BPW + 1;
252                 else
253                         break;
254                 sig += ffz(~x);
255                 break;
256
257         case 1: if ((x = *s &~ *m) != 0)
258                         sig = ffz(~x) + 1;
259                 break;
260         }
261         
262         return sig;
263 }
264
265 static struct sigqueue *__sigqueue_alloc(struct task_struct *t, gfp_t flags,
266                                          int override_rlimit)
267 {
268         struct sigqueue *q = NULL;
269
270         atomic_inc(&t->user->sigpending);
271         if (override_rlimit ||
272             atomic_read(&t->user->sigpending) <=
273                         t->signal->rlim[RLIMIT_SIGPENDING].rlim_cur)
274                 q = kmem_cache_alloc(sigqueue_cachep, flags);
275         if (unlikely(q == NULL)) {
276                 atomic_dec(&t->user->sigpending);
277         } else {
278                 INIT_LIST_HEAD(&q->list);
279                 q->flags = 0;
280                 q->lock = NULL;
281                 q->user = get_uid(t->user);
282         }
283         return(q);
284 }
285
286 static inline void __sigqueue_free(struct sigqueue *q)
287 {
288         if (q->flags & SIGQUEUE_PREALLOC)
289                 return;
290         atomic_dec(&q->user->sigpending);
291         free_uid(q->user);
292         kmem_cache_free(sigqueue_cachep, q);
293 }
294
295 static void flush_sigqueue(struct sigpending *queue)
296 {
297         struct sigqueue *q;
298
299         sigemptyset(&queue->signal);
300         while (!list_empty(&queue->list)) {
301                 q = list_entry(queue->list.next, struct sigqueue , list);
302                 list_del_init(&q->list);
303                 __sigqueue_free(q);
304         }
305 }
306
307 /*
308  * Flush all pending signals for a task.
309  */
310
311 void
312 flush_signals(struct task_struct *t)
313 {
314         unsigned long flags;
315
316         spin_lock_irqsave(&t->sighand->siglock, flags);
317         clear_tsk_thread_flag(t,TIF_SIGPENDING);
318         flush_sigqueue(&t->pending);
319         flush_sigqueue(&t->signal->shared_pending);
320         spin_unlock_irqrestore(&t->sighand->siglock, flags);
321 }
322
323 /*
324  * This function expects the tasklist_lock write-locked.
325  */
326 void __exit_sighand(struct task_struct *tsk)
327 {
328         struct sighand_struct * sighand = tsk->sighand;
329
330         /* Ok, we're done with the signal handlers */
331         tsk->sighand = NULL;
332         if (atomic_dec_and_test(&sighand->count))
333                 kmem_cache_free(sighand_cachep, sighand);
334 }
335
336 void exit_sighand(struct task_struct *tsk)
337 {
338         write_lock_irq(&tasklist_lock);
339         __exit_sighand(tsk);
340         write_unlock_irq(&tasklist_lock);
341 }
342
343 /*
344  * This function expects the tasklist_lock write-locked.
345  */
346 void __exit_signal(struct task_struct *tsk)
347 {
348         struct signal_struct * sig = tsk->signal;
349         struct sighand_struct * sighand = tsk->sighand;
350
351         if (!sig)
352                 BUG();
353         if (!atomic_read(&sig->count))
354                 BUG();
355         spin_lock(&sighand->siglock);
356         posix_cpu_timers_exit(tsk);
357         if (atomic_dec_and_test(&sig->count)) {
358                 posix_cpu_timers_exit_group(tsk);
359                 if (tsk == sig->curr_target)
360                         sig->curr_target = next_thread(tsk);
361                 tsk->signal = NULL;
362                 spin_unlock(&sighand->siglock);
363                 flush_sigqueue(&sig->shared_pending);
364         } else {
365                 /*
366                  * If there is any task waiting for the group exit
367                  * then notify it:
368                  */
369                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
370                         wake_up_process(sig->group_exit_task);
371                         sig->group_exit_task = NULL;
372                 }
373                 if (tsk == sig->curr_target)
374                         sig->curr_target = next_thread(tsk);
375                 tsk->signal = NULL;
376                 /*
377                  * Accumulate here the counters for all threads but the
378                  * group leader as they die, so they can be added into
379                  * the process-wide totals when those are taken.
380                  * The group leader stays around as a zombie as long
381                  * as there are other threads.  When it gets reaped,
382                  * the exit.c code will add its counts into these totals.
383                  * We won't ever get here for the group leader, since it
384                  * will have been the last reference on the signal_struct.
385                  */
386                 sig->utime = cputime_add(sig->utime, tsk->utime);
387                 sig->stime = cputime_add(sig->stime, tsk->stime);
388                 sig->min_flt += tsk->min_flt;
389                 sig->maj_flt += tsk->maj_flt;
390                 sig->nvcsw += tsk->nvcsw;
391                 sig->nivcsw += tsk->nivcsw;
392                 sig->sched_time += tsk->sched_time;
393                 spin_unlock(&sighand->siglock);
394                 sig = NULL;     /* Marker for below.  */
395         }
396         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
397         flush_sigqueue(&tsk->pending);
398         if (sig) {
399                 /*
400                  * We are cleaning up the signal_struct here.  We delayed
401                  * calling exit_itimers until after flush_sigqueue, just in
402                  * case our thread-local pending queue contained a queued
403                  * timer signal that would have been cleared in
404                  * exit_itimers.  When that called sigqueue_free, it would
405                  * attempt to re-take the tasklist_lock and deadlock.  This
406                  * can never happen if we ensure that all queues the
407                  * timer's signal might be queued on have been flushed
408                  * first.  The shared_pending queue, and our own pending
409                  * queue are the only queues the timer could be on, since
410                  * there are no other threads left in the group and timer
411                  * signals are constrained to threads inside the group.
412                  */
413                 exit_itimers(sig);
414                 exit_thread_group_keys(sig);
415                 kmem_cache_free(signal_cachep, sig);
416         }
417 }
418
419 void exit_signal(struct task_struct *tsk)
420 {
421         write_lock_irq(&tasklist_lock);
422         __exit_signal(tsk);
423         write_unlock_irq(&tasklist_lock);
424 }
425
426 /*
427  * Flush all handlers for a task.
428  */
429
430 void
431 flush_signal_handlers(struct task_struct *t, int force_default)
432 {
433         int i;
434         struct k_sigaction *ka = &t->sighand->action[0];
435         for (i = _NSIG ; i != 0 ; i--) {
436                 if (force_default || ka->sa.sa_handler != SIG_IGN)
437                         ka->sa.sa_handler = SIG_DFL;
438                 ka->sa.sa_flags = 0;
439                 sigemptyset(&ka->sa.sa_mask);
440                 ka++;
441         }
442 }
443
444
445 /* Notify the system that a driver wants to block all signals for this
446  * process, and wants to be notified if any signals at all were to be
447  * sent/acted upon.  If the notifier routine returns non-zero, then the
448  * signal will be acted upon after all.  If the notifier routine returns 0,
449  * then then signal will be blocked.  Only one block per process is
450  * allowed.  priv is a pointer to private data that the notifier routine
451  * can use to determine if the signal should be blocked or not.  */
452
453 void
454 block_all_signals(int (*notifier)(void *priv), void *priv, sigset_t *mask)
455 {
456         unsigned long flags;
457
458         spin_lock_irqsave(&current->sighand->siglock, flags);
459         current->notifier_mask = mask;
460         current->notifier_data = priv;
461         current->notifier = notifier;
462         spin_unlock_irqrestore(&current->sighand->siglock, flags);
463 }
464
465 /* Notify the system that blocking has ended. */
466
467 void
468 unblock_all_signals(void)
469 {
470         unsigned long flags;
471
472         spin_lock_irqsave(&current->sighand->siglock, flags);
473         current->notifier = NULL;
474         current->notifier_data = NULL;
475         recalc_sigpending();
476         spin_unlock_irqrestore(&current->sighand->siglock, flags);
477 }
478
479 static inline int collect_signal(int sig, struct sigpending *list, siginfo_t *info)
480 {
481         struct sigqueue *q, *first = NULL;
482         int still_pending = 0;
483
484         if (unlikely(!sigismember(&list->signal, sig)))
485                 return 0;
486
487         /*
488          * Collect the siginfo appropriate to this signal.  Check if
489          * there is another siginfo for the same signal.
490         */
491         list_for_each_entry(q, &list->list, list) {
492                 if (q->info.si_signo == sig) {
493                         if (first) {
494                                 still_pending = 1;
495                                 break;
496                         }
497                         first = q;
498                 }
499         }
500         if (first) {
501                 list_del_init(&first->list);
502                 copy_siginfo(info, &first->info);
503                 __sigqueue_free(first);
504                 if (!still_pending)
505                         sigdelset(&list->signal, sig);
506         } else {
507
508                 /* Ok, it wasn't in the queue.  This must be
509                    a fast-pathed signal or we must have been
510                    out of queue space.  So zero out the info.
511                  */
512                 sigdelset(&list->signal, sig);
513                 info->si_signo = sig;
514                 info->si_errno = 0;
515                 info->si_code = 0;
516                 info->si_pid = 0;
517                 info->si_uid = 0;
518         }
519         return 1;
520 }
521
522 static int __dequeue_signal(struct sigpending *pending, sigset_t *mask,
523                         siginfo_t *info)
524 {
525         int sig = 0;
526
527         /* SIGKILL must have priority, otherwise it is quite easy
528          * to create an unkillable process, sending sig < SIGKILL
529          * to self */
530         if (unlikely(sigismember(&pending->signal, SIGKILL))) {
531                 if (!sigismember(mask, SIGKILL))
532                         sig = SIGKILL;
533         }
534
535         if (likely(!sig))
536                 sig = next_signal(pending, mask);
537         if (sig) {
538                 if (current->notifier) {
539                         if (sigismember(current->notifier_mask, sig)) {
540                                 if (!(current->notifier)(current->notifier_data)) {
541                                         clear_thread_flag(TIF_SIGPENDING);
542                                         return 0;
543                                 }
544                         }
545                 }
546
547                 if (!collect_signal(sig, pending, info))
548                         sig = 0;
549                                 
550         }
551         recalc_sigpending();
552
553         return sig;
554 }
555
556 /*
557  * Dequeue a signal and return the element to the caller, which is 
558  * expected to free it.
559  *
560  * All callers have to hold the siglock.
561  */
562 int dequeue_signal(struct task_struct *tsk, sigset_t *mask, siginfo_t *info)
563 {
564         int signr = __dequeue_signal(&tsk->pending, mask, info);
565         if (!signr)
566                 signr = __dequeue_signal(&tsk->signal->shared_pending,
567                                          mask, info);
568         if (signr && unlikely(sig_kernel_stop(signr))) {
569                 /*
570                  * Set a marker that we have dequeued a stop signal.  Our
571                  * caller might release the siglock and then the pending
572                  * stop signal it is about to process is no longer in the
573                  * pending bitmasks, but must still be cleared by a SIGCONT
574                  * (and overruled by a SIGKILL).  So those cases clear this
575                  * shared flag after we've set it.  Note that this flag may
576                  * remain set after the signal we return is ignored or
577                  * handled.  That doesn't matter because its only purpose
578                  * is to alert stop-signal processing code when another
579                  * processor has come along and cleared the flag.
580                  */
581                 if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT))
582                         tsk->signal->flags |= SIGNAL_STOP_DEQUEUED;
583         }
584         if ( signr &&
585              ((info->si_code & __SI_MASK) == __SI_TIMER) &&
586              info->si_sys_private){
587                 /*
588                  * Release the siglock to ensure proper locking order
589                  * of timer locks outside of siglocks.  Note, we leave
590                  * irqs disabled here, since the posix-timers code is
591                  * about to disable them again anyway.
592                  */
593                 spin_unlock(&tsk->sighand->siglock);
594                 do_schedule_next_timer(info);
595                 spin_lock(&tsk->sighand->siglock);
596         }
597         return signr;
598 }
599
600 /*
601  * Tell a process that it has a new active signal..
602  *
603  * NOTE! we rely on the previous spin_lock to
604  * lock interrupts for us! We can only be called with
605  * "siglock" held, and the local interrupt must
606  * have been disabled when that got acquired!
607  *
608  * No need to set need_resched since signal event passing
609  * goes through ->blocked
610  */
611 void signal_wake_up(struct task_struct *t, int resume)
612 {
613         unsigned int mask;
614
615         set_tsk_thread_flag(t, TIF_SIGPENDING);
616
617         /*
618          * For SIGKILL, we want to wake it up in the stopped/traced case.
619          * We don't check t->state here because there is a race with it
620          * executing another processor and just now entering stopped state.
621          * By using wake_up_state, we ensure the process will wake up and
622          * handle its death signal.
623          */
624         mask = TASK_INTERRUPTIBLE;
625         if (resume)
626                 mask |= TASK_STOPPED | TASK_TRACED;
627         if (!wake_up_state(t, mask))
628                 kick_process(t);
629 }
630
631 /*
632  * Remove signals in mask from the pending set and queue.
633  * Returns 1 if any signals were found.
634  *
635  * All callers must be holding the siglock.
636  */
637 static int rm_from_queue(unsigned long mask, struct sigpending *s)
638 {
639         struct sigqueue *q, *n;
640
641         if (!sigtestsetmask(&s->signal, mask))
642                 return 0;
643
644         sigdelsetmask(&s->signal, mask);
645         list_for_each_entry_safe(q, n, &s->list, list) {
646                 if (q->info.si_signo < SIGRTMIN &&
647                     (mask & sigmask(q->info.si_signo))) {
648                         list_del_init(&q->list);
649                         __sigqueue_free(q);
650                 }
651         }
652         return 1;
653 }
654
655 /*
656  * Bad permissions for sending the signal
657  */
658 static int check_kill_permission(int sig, struct siginfo *info,
659                                  struct task_struct *t)
660 {
661         int error = -EINVAL;
662         if (!valid_signal(sig))
663                 return error;
664         error = -EPERM;
665         if ((!info || ((unsigned long)info != 1 &&
666                         (unsigned long)info != 2 && SI_FROMUSER(info)))
667             && ((sig != SIGCONT) ||
668                 (current->signal->session != t->signal->session))
669             && (current->euid ^ t->suid) && (current->euid ^ t->uid)
670             && (current->uid ^ t->suid) && (current->uid ^ t->uid)
671             && !capable(CAP_KILL))
672                 return error;
673
674         error = security_task_kill(t, info, sig);
675         if (!error)
676                 audit_signal_info(sig, t); /* Let audit system see the signal */
677         return error;
678 }
679
680 /* forward decl */
681 static void do_notify_parent_cldstop(struct task_struct *tsk,
682                                      int to_self,
683                                      int why);
684
685 /*
686  * Handle magic process-wide effects of stop/continue signals.
687  * Unlike the signal actions, these happen immediately at signal-generation
688  * time regardless of blocking, ignoring, or handling.  This does the
689  * actual continuing for SIGCONT, but not the actual stopping for stop
690  * signals.  The process stop is done as a signal action for SIG_DFL.
691  */
692 static void handle_stop_signal(int sig, struct task_struct *p)
693 {
694         struct task_struct *t;
695
696         if (p->signal->flags & SIGNAL_GROUP_EXIT)
697                 /*
698                  * The process is in the middle of dying already.
699                  */
700                 return;
701
702         if (sig_kernel_stop(sig)) {
703                 /*
704                  * This is a stop signal.  Remove SIGCONT from all queues.
705                  */
706                 rm_from_queue(sigmask(SIGCONT), &p->signal->shared_pending);
707                 t = p;
708                 do {
709                         rm_from_queue(sigmask(SIGCONT), &t->pending);
710                         t = next_thread(t);
711                 } while (t != p);
712         } else if (sig == SIGCONT) {
713                 /*
714                  * Remove all stop signals from all queues,
715                  * and wake all threads.
716                  */
717                 if (unlikely(p->signal->group_stop_count > 0)) {
718                         /*
719                          * There was a group stop in progress.  We'll
720                          * pretend it finished before we got here.  We are
721                          * obliged to report it to the parent: if the
722                          * SIGSTOP happened "after" this SIGCONT, then it
723                          * would have cleared this pending SIGCONT.  If it
724                          * happened "before" this SIGCONT, then the parent
725                          * got the SIGCHLD about the stop finishing before
726                          * the continue happened.  We do the notification
727                          * now, and it's as if the stop had finished and
728                          * the SIGCHLD was pending on entry to this kill.
729                          */
730                         p->signal->group_stop_count = 0;
731                         p->signal->flags = SIGNAL_STOP_CONTINUED;
732                         spin_unlock(&p->sighand->siglock);
733                         do_notify_parent_cldstop(p, (p->ptrace & PT_PTRACED), CLD_STOPPED);
734                         spin_lock(&p->sighand->siglock);
735                 }
736                 rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
737                 t = p;
738                 do {
739                         unsigned int state;
740                         rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
741                         
742                         /*
743                          * If there is a handler for SIGCONT, we must make
744                          * sure that no thread returns to user mode before
745                          * we post the signal, in case it was the only
746                          * thread eligible to run the signal handler--then
747                          * it must not do anything between resuming and
748                          * running the handler.  With the TIF_SIGPENDING
749                          * flag set, the thread will pause and acquire the
750                          * siglock that we hold now and until we've queued
751                          * the pending signal. 
752                          *
753                          * Wake up the stopped thread _after_ setting
754                          * TIF_SIGPENDING
755                          */
756                         state = TASK_STOPPED;
757                         if (sig_user_defined(t, SIGCONT) && !sigismember(&t->blocked, SIGCONT)) {
758                                 set_tsk_thread_flag(t, TIF_SIGPENDING);
759                                 state |= TASK_INTERRUPTIBLE;
760                         }
761                         wake_up_state(t, state);
762
763                         t = next_thread(t);
764                 } while (t != p);
765
766                 if (p->signal->flags & SIGNAL_STOP_STOPPED) {
767                         /*
768                          * We were in fact stopped, and are now continued.
769                          * Notify the parent with CLD_CONTINUED.
770                          */
771                         p->signal->flags = SIGNAL_STOP_CONTINUED;
772                         p->signal->group_exit_code = 0;
773                         spin_unlock(&p->sighand->siglock);
774                         do_notify_parent_cldstop(p, (p->ptrace & PT_PTRACED), CLD_CONTINUED);
775                         spin_lock(&p->sighand->siglock);
776                 } else {
777                         /*
778                          * We are not stopped, but there could be a stop
779                          * signal in the middle of being processed after
780                          * being removed from the queue.  Clear that too.
781                          */
782                         p->signal->flags = 0;
783                 }
784         } else if (sig == SIGKILL) {
785                 /*
786                  * Make sure that any pending stop signal already dequeued
787                  * is undone by the wakeup for SIGKILL.
788                  */
789                 p->signal->flags = 0;
790         }
791 }
792
793 static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
794                         struct sigpending *signals)
795 {
796         struct sigqueue * q = NULL;
797         int ret = 0;
798
799         /*
800          * fast-pathed signals for kernel-internal things like SIGSTOP
801          * or SIGKILL.
802          */
803         if ((unsigned long)info == 2)
804                 goto out_set;
805
806         /* Real-time signals must be queued if sent by sigqueue, or
807            some other real-time mechanism.  It is implementation
808            defined whether kill() does so.  We attempt to do so, on
809            the principle of least surprise, but since kill is not
810            allowed to fail with EAGAIN when low on memory we just
811            make sure at least one signal gets delivered and don't
812            pass on the info struct.  */
813
814         q = __sigqueue_alloc(t, GFP_ATOMIC, (sig < SIGRTMIN &&
815                                              ((unsigned long) info < 2 ||
816                                               info->si_code >= 0)));
817         if (q) {
818                 list_add_tail(&q->list, &signals->list);
819                 switch ((unsigned long) info) {
820                 case 0:
821                         q->info.si_signo = sig;
822                         q->info.si_errno = 0;
823                         q->info.si_code = SI_USER;
824                         q->info.si_pid = current->pid;
825                         q->info.si_uid = current->uid;
826                         break;
827                 case 1:
828                         q->info.si_signo = sig;
829                         q->info.si_errno = 0;
830                         q->info.si_code = SI_KERNEL;
831                         q->info.si_pid = 0;
832                         q->info.si_uid = 0;
833                         break;
834                 default:
835                         copy_siginfo(&q->info, info);
836                         break;
837                 }
838         } else {
839                 if (sig >= SIGRTMIN && info && (unsigned long)info != 1
840                    && info->si_code != SI_USER)
841                 /*
842                  * Queue overflow, abort.  We may abort if the signal was rt
843                  * and sent by user using something other than kill().
844                  */
845                         return -EAGAIN;
846                 if (((unsigned long)info > 1) && (info->si_code == SI_TIMER))
847                         /*
848                          * Set up a return to indicate that we dropped 
849                          * the signal.
850                          */
851                         ret = info->si_sys_private;
852         }
853
854 out_set:
855         sigaddset(&signals->signal, sig);
856         return ret;
857 }
858
859 #define LEGACY_QUEUE(sigptr, sig) \
860         (((sig) < SIGRTMIN) && sigismember(&(sigptr)->signal, (sig)))
861
862
863 static int
864 specific_send_sig_info(int sig, struct siginfo *info, struct task_struct *t)
865 {
866         int ret = 0;
867
868         if (!irqs_disabled())
869                 BUG();
870         assert_spin_locked(&t->sighand->siglock);
871
872         if (((unsigned long)info > 2) && (info->si_code == SI_TIMER))
873                 /*
874                  * Set up a return to indicate that we dropped the signal.
875                  */
876                 ret = info->si_sys_private;
877
878         /* Short-circuit ignored signals.  */
879         if (sig_ignored(t, sig))
880                 goto out;
881
882         /* Support queueing exactly one non-rt signal, so that we
883            can get more detailed information about the cause of
884            the signal. */
885         if (LEGACY_QUEUE(&t->pending, sig))
886                 goto out;
887
888         ret = send_signal(sig, info, t, &t->pending);
889         if (!ret && !sigismember(&t->blocked, sig))
890                 signal_wake_up(t, sig == SIGKILL);
891 out:
892         return ret;
893 }
894
895 /*
896  * Force a signal that the process can't ignore: if necessary
897  * we unblock the signal and change any SIG_IGN to SIG_DFL.
898  */
899
900 int
901 force_sig_info(int sig, struct siginfo *info, struct task_struct *t)
902 {
903         unsigned long int flags;
904         int ret;
905
906         spin_lock_irqsave(&t->sighand->siglock, flags);
907         if (sigismember(&t->blocked, sig) || t->sighand->action[sig-1].sa.sa_handler == SIG_IGN) {
908                 t->sighand->action[sig-1].sa.sa_handler = SIG_DFL;
909                 sigdelset(&t->blocked, sig);
910                 recalc_sigpending_tsk(t);
911         }
912         ret = specific_send_sig_info(sig, info, t);
913         spin_unlock_irqrestore(&t->sighand->siglock, flags);
914
915         return ret;
916 }
917
918 void
919 force_sig_specific(int sig, struct task_struct *t)
920 {
921         unsigned long int flags;
922
923         spin_lock_irqsave(&t->sighand->siglock, flags);
924         if (t->sighand->action[sig-1].sa.sa_handler == SIG_IGN)
925                 t->sighand->action[sig-1].sa.sa_handler = SIG_DFL;
926         sigdelset(&t->blocked, sig);
927         recalc_sigpending_tsk(t);
928         specific_send_sig_info(sig, (void *)2, t);
929         spin_unlock_irqrestore(&t->sighand->siglock, flags);
930 }
931
932 /*
933  * Test if P wants to take SIG.  After we've checked all threads with this,
934  * it's equivalent to finding no threads not blocking SIG.  Any threads not
935  * blocking SIG were ruled out because they are not running and already
936  * have pending signals.  Such threads will dequeue from the shared queue
937  * as soon as they're available, so putting the signal on the shared queue
938  * will be equivalent to sending it to one such thread.
939  */
940 static inline int wants_signal(int sig, struct task_struct *p)
941 {
942         if (sigismember(&p->blocked, sig))
943                 return 0;
944         if (p->flags & PF_EXITING)
945                 return 0;
946         if (sig == SIGKILL)
947                 return 1;
948         if (p->state & (TASK_STOPPED | TASK_TRACED))
949                 return 0;
950         return task_curr(p) || !signal_pending(p);
951 }
952
953 static void
954 __group_complete_signal(int sig, struct task_struct *p)
955 {
956         struct task_struct *t;
957
958         /*
959          * Now find a thread we can wake up to take the signal off the queue.
960          *
961          * If the main thread wants the signal, it gets first crack.
962          * Probably the least surprising to the average bear.
963          */
964         if (wants_signal(sig, p))
965                 t = p;
966         else if (thread_group_empty(p))
967                 /*
968                  * There is just one thread and it does not need to be woken.
969                  * It will dequeue unblocked signals before it runs again.
970                  */
971                 return;
972         else {
973                 /*
974                  * Otherwise try to find a suitable thread.
975                  */
976                 t = p->signal->curr_target;
977                 if (t == NULL)
978                         /* restart balancing at this thread */
979                         t = p->signal->curr_target = p;
980                 BUG_ON(t->tgid != p->tgid);
981
982                 while (!wants_signal(sig, t)) {
983                         t = next_thread(t);
984                         if (t == p->signal->curr_target)
985                                 /*
986                                  * No thread needs to be woken.
987                                  * Any eligible threads will see
988                                  * the signal in the queue soon.
989                                  */
990                                 return;
991                 }
992                 p->signal->curr_target = t;
993         }
994
995         /*
996          * Found a killable thread.  If the signal will be fatal,
997          * then start taking the whole group down immediately.
998          */
999         if (sig_fatal(p, sig) && !(p->signal->flags & SIGNAL_GROUP_EXIT) &&
1000             !sigismember(&t->real_blocked, sig) &&
1001             (sig == SIGKILL || !(t->ptrace & PT_PTRACED))) {
1002                 /*
1003                  * This signal will be fatal to the whole group.
1004                  */
1005                 if (!sig_kernel_coredump(sig)) {
1006                         /*
1007                          * Start a group exit and wake everybody up.
1008                          * This way we don't have other threads
1009                          * running and doing things after a slower
1010                          * thread has the fatal signal pending.
1011                          */
1012                         p->signal->flags = SIGNAL_GROUP_EXIT;
1013                         p->signal->group_exit_code = sig;
1014                         p->signal->group_stop_count = 0;
1015                         t = p;
1016                         do {
1017                                 sigaddset(&t->pending.signal, SIGKILL);
1018                                 signal_wake_up(t, 1);
1019                                 t = next_thread(t);
1020                         } while (t != p);
1021                         return;
1022                 }
1023
1024                 /*
1025                  * There will be a core dump.  We make all threads other
1026                  * than the chosen one go into a group stop so that nothing
1027                  * happens until it gets scheduled, takes the signal off
1028                  * the shared queue, and does the core dump.  This is a
1029                  * little more complicated than strictly necessary, but it
1030                  * keeps the signal state that winds up in the core dump
1031                  * unchanged from the death state, e.g. which thread had
1032                  * the core-dump signal unblocked.
1033                  */
1034                 rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
1035                 rm_from_queue(SIG_KERNEL_STOP_MASK, &p->signal->shared_pending);
1036                 p->signal->group_stop_count = 0;
1037                 p->signal->group_exit_task = t;
1038                 t = p;
1039                 do {
1040                         p->signal->group_stop_count++;
1041                         signal_wake_up(t, 0);
1042                         t = next_thread(t);
1043                 } while (t != p);
1044                 wake_up_process(p->signal->group_exit_task);
1045                 return;
1046         }
1047
1048         /*
1049          * The signal is already in the shared-pending queue.
1050          * Tell the chosen thread to wake up and dequeue it.
1051          */
1052         signal_wake_up(t, sig == SIGKILL);
1053         return;
1054 }
1055
1056 int
1057 __group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1058 {
1059         int ret = 0;
1060
1061         assert_spin_locked(&p->sighand->siglock);
1062         handle_stop_signal(sig, p);
1063
1064         if (((unsigned long)info > 2) && (info->si_code == SI_TIMER))
1065                 /*
1066                  * Set up a return to indicate that we dropped the signal.
1067                  */
1068                 ret = info->si_sys_private;
1069
1070         /* Short-circuit ignored signals.  */
1071         if (sig_ignored(p, sig))
1072                 return ret;
1073
1074         if (LEGACY_QUEUE(&p->signal->shared_pending, sig))
1075                 /* This is a non-RT signal and we already have one queued.  */
1076                 return ret;
1077
1078         /*
1079          * Put this signal on the shared-pending queue, or fail with EAGAIN.
1080          * We always use the shared queue for process-wide signals,
1081          * to avoid several races.
1082          */
1083         ret = send_signal(sig, info, p, &p->signal->shared_pending);
1084         if (unlikely(ret))
1085                 return ret;
1086
1087         __group_complete_signal(sig, p);
1088         return 0;
1089 }
1090
1091 /*
1092  * Nuke all other threads in the group.
1093  */
1094 void zap_other_threads(struct task_struct *p)
1095 {
1096         struct task_struct *t;
1097
1098         p->signal->flags = SIGNAL_GROUP_EXIT;
1099         p->signal->group_stop_count = 0;
1100
1101         if (thread_group_empty(p))
1102                 return;
1103
1104         for (t = next_thread(p); t != p; t = next_thread(t)) {
1105                 /*
1106                  * Don't bother with already dead threads
1107                  */
1108                 if (t->exit_state)
1109                         continue;
1110
1111                 /*
1112                  * We don't want to notify the parent, since we are
1113                  * killed as part of a thread group due to another
1114                  * thread doing an execve() or similar. So set the
1115                  * exit signal to -1 to allow immediate reaping of
1116                  * the process.  But don't detach the thread group
1117                  * leader.
1118                  */
1119                 if (t != p->group_leader)
1120                         t->exit_signal = -1;
1121
1122                 sigaddset(&t->pending.signal, SIGKILL);
1123                 rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
1124                 signal_wake_up(t, 1);
1125         }
1126 }
1127
1128 /*
1129  * Must be called with the tasklist_lock held for reading!
1130  */
1131 int group_send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1132 {
1133         unsigned long flags;
1134         int ret;
1135
1136         ret = check_kill_permission(sig, info, p);
1137         if (!ret && sig && p->sighand) {
1138                 spin_lock_irqsave(&p->sighand->siglock, flags);
1139                 ret = __group_send_sig_info(sig, info, p);
1140                 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1141         }
1142
1143         return ret;
1144 }
1145
1146 /*
1147  * kill_pg_info() sends a signal to a process group: this is what the tty
1148  * control characters do (^C, ^Z etc)
1149  */
1150
1151 int __kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)
1152 {
1153         struct task_struct *p = NULL;
1154         int retval, success;
1155
1156         if (pgrp <= 0)
1157                 return -EINVAL;
1158
1159         success = 0;
1160         retval = -ESRCH;
1161         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
1162                 int err = group_send_sig_info(sig, info, p);
1163                 success |= !err;
1164                 retval = err;
1165         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
1166         return success ? 0 : retval;
1167 }
1168
1169 int
1170 kill_pg_info(int sig, struct siginfo *info, pid_t pgrp)
1171 {
1172         int retval;
1173
1174         read_lock(&tasklist_lock);
1175         retval = __kill_pg_info(sig, info, pgrp);
1176         read_unlock(&tasklist_lock);
1177
1178         return retval;
1179 }
1180
1181 int
1182 kill_proc_info(int sig, struct siginfo *info, pid_t pid)
1183 {
1184         int error;
1185         struct task_struct *p;
1186
1187         read_lock(&tasklist_lock);
1188         p = find_task_by_pid(pid);
1189         error = -ESRCH;
1190         if (p)
1191                 error = group_send_sig_info(sig, info, p);
1192         read_unlock(&tasklist_lock);
1193         return error;
1194 }
1195
1196 /* like kill_proc_info(), but doesn't use uid/euid of "current" */
1197 int kill_proc_info_as_uid(int sig, struct siginfo *info, pid_t pid,
1198                       uid_t uid, uid_t euid)
1199 {
1200         int ret = -EINVAL;
1201         struct task_struct *p;
1202
1203         if (!valid_signal(sig))
1204                 return ret;
1205
1206         read_lock(&tasklist_lock);
1207         p = find_task_by_pid(pid);
1208         if (!p) {
1209                 ret = -ESRCH;
1210                 goto out_unlock;
1211         }
1212         if ((!info || ((unsigned long)info != 1 &&
1213                         (unsigned long)info != 2 && SI_FROMUSER(info)))
1214             && (euid != p->suid) && (euid != p->uid)
1215             && (uid != p->suid) && (uid != p->uid)) {
1216                 ret = -EPERM;
1217                 goto out_unlock;
1218         }
1219         if (sig && p->sighand) {
1220                 unsigned long flags;
1221                 spin_lock_irqsave(&p->sighand->siglock, flags);
1222                 ret = __group_send_sig_info(sig, info, p);
1223                 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1224         }
1225 out_unlock:
1226         read_unlock(&tasklist_lock);
1227         return ret;
1228 }
1229 EXPORT_SYMBOL_GPL(kill_proc_info_as_uid);
1230
1231 /*
1232  * kill_something_info() interprets pid in interesting ways just like kill(2).
1233  *
1234  * POSIX specifies that kill(-1,sig) is unspecified, but what we have
1235  * is probably wrong.  Should make it like BSD or SYSV.
1236  */
1237
1238 static int kill_something_info(int sig, struct siginfo *info, int pid)
1239 {
1240         if (!pid) {
1241                 return kill_pg_info(sig, info, process_group(current));
1242         } else if (pid == -1) {
1243                 int retval = 0, count = 0;
1244                 struct task_struct * p;
1245
1246                 read_lock(&tasklist_lock);
1247                 for_each_process(p) {
1248                         if (p->pid > 1 && p->tgid != current->tgid) {
1249                                 int err = group_send_sig_info(sig, info, p);
1250                                 ++count;
1251                                 if (err != -EPERM)
1252                                         retval = err;
1253                         }
1254                 }
1255                 read_unlock(&tasklist_lock);
1256                 return count ? retval : -ESRCH;
1257         } else if (pid < 0) {
1258                 return kill_pg_info(sig, info, -pid);
1259         } else {
1260                 return kill_proc_info(sig, info, pid);
1261         }
1262 }
1263
1264 /*
1265  * These are for backward compatibility with the rest of the kernel source.
1266  */
1267
1268 /*
1269  * These two are the most common entry points.  They send a signal
1270  * just to the specific thread.
1271  */
1272 int
1273 send_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1274 {
1275         int ret;
1276         unsigned long flags;
1277
1278         /*
1279          * Make sure legacy kernel users don't send in bad values
1280          * (normal paths check this in check_kill_permission).
1281          */
1282         if (!valid_signal(sig))
1283                 return -EINVAL;
1284
1285         /*
1286          * We need the tasklist lock even for the specific
1287          * thread case (when we don't need to follow the group
1288          * lists) in order to avoid races with "p->sighand"
1289          * going away or changing from under us.
1290          */
1291         read_lock(&tasklist_lock);  
1292         spin_lock_irqsave(&p->sighand->siglock, flags);
1293         ret = specific_send_sig_info(sig, info, p);
1294         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1295         read_unlock(&tasklist_lock);
1296         return ret;
1297 }
1298
1299 int
1300 send_sig(int sig, struct task_struct *p, int priv)
1301 {
1302         return send_sig_info(sig, (void*)(long)(priv != 0), p);
1303 }
1304
1305 /*
1306  * This is the entry point for "process-wide" signals.
1307  * They will go to an appropriate thread in the thread group.
1308  */
1309 int
1310 send_group_sig_info(int sig, struct siginfo *info, struct task_struct *p)
1311 {
1312         int ret;
1313         read_lock(&tasklist_lock);
1314         ret = group_send_sig_info(sig, info, p);
1315         read_unlock(&tasklist_lock);
1316         return ret;
1317 }
1318
1319 void
1320 force_sig(int sig, struct task_struct *p)
1321 {
1322         force_sig_info(sig, (void*)1L, p);
1323 }
1324
1325 /*
1326  * When things go south during signal handling, we
1327  * will force a SIGSEGV. And if the signal that caused
1328  * the problem was already a SIGSEGV, we'll want to
1329  * make sure we don't even try to deliver the signal..
1330  */
1331 int
1332 force_sigsegv(int sig, struct task_struct *p)
1333 {
1334         if (sig == SIGSEGV) {
1335                 unsigned long flags;
1336                 spin_lock_irqsave(&p->sighand->siglock, flags);
1337                 p->sighand->action[sig - 1].sa.sa_handler = SIG_DFL;
1338                 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1339         }
1340         force_sig(SIGSEGV, p);
1341         return 0;
1342 }
1343
1344 int
1345 kill_pg(pid_t pgrp, int sig, int priv)
1346 {
1347         return kill_pg_info(sig, (void *)(long)(priv != 0), pgrp);
1348 }
1349
1350 int
1351 kill_proc(pid_t pid, int sig, int priv)
1352 {
1353         return kill_proc_info(sig, (void *)(long)(priv != 0), pid);
1354 }
1355
1356 /*
1357  * These functions support sending signals using preallocated sigqueue
1358  * structures.  This is needed "because realtime applications cannot
1359  * afford to lose notifications of asynchronous events, like timer
1360  * expirations or I/O completions".  In the case of Posix Timers 
1361  * we allocate the sigqueue structure from the timer_create.  If this
1362  * allocation fails we are able to report the failure to the application
1363  * with an EAGAIN error.
1364  */
1365  
1366 struct sigqueue *sigqueue_alloc(void)
1367 {
1368         struct sigqueue *q;
1369
1370         if ((q = __sigqueue_alloc(current, GFP_KERNEL, 0)))
1371                 q->flags |= SIGQUEUE_PREALLOC;
1372         return(q);
1373 }
1374
1375 void sigqueue_free(struct sigqueue *q)
1376 {
1377         unsigned long flags;
1378         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1379         /*
1380          * If the signal is still pending remove it from the
1381          * pending queue.
1382          */
1383         if (unlikely(!list_empty(&q->list))) {
1384                 read_lock(&tasklist_lock);  
1385                 spin_lock_irqsave(q->lock, flags);
1386                 if (!list_empty(&q->list))
1387                         list_del_init(&q->list);
1388                 spin_unlock_irqrestore(q->lock, flags);
1389                 read_unlock(&tasklist_lock);
1390         }
1391         q->flags &= ~SIGQUEUE_PREALLOC;
1392         __sigqueue_free(q);
1393 }
1394
1395 int
1396 send_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
1397 {
1398         unsigned long flags;
1399         int ret = 0;
1400
1401         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1402         read_lock(&tasklist_lock);
1403
1404         if (unlikely(p->flags & PF_EXITING)) {
1405                 ret = -1;
1406                 goto out_err;
1407         }
1408
1409         spin_lock_irqsave(&p->sighand->siglock, flags);
1410
1411         if (unlikely(!list_empty(&q->list))) {
1412                 /*
1413                  * If an SI_TIMER entry is already queue just increment
1414                  * the overrun count.
1415                  */
1416                 if (q->info.si_code != SI_TIMER)
1417                         BUG();
1418                 q->info.si_overrun++;
1419                 goto out;
1420         }
1421         /* Short-circuit ignored signals.  */
1422         if (sig_ignored(p, sig)) {
1423                 ret = 1;
1424                 goto out;
1425         }
1426
1427         q->lock = &p->sighand->siglock;
1428         list_add_tail(&q->list, &p->pending.list);
1429         sigaddset(&p->pending.signal, sig);
1430         if (!sigismember(&p->blocked, sig))
1431                 signal_wake_up(p, sig == SIGKILL);
1432
1433 out:
1434         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1435 out_err:
1436         read_unlock(&tasklist_lock);
1437
1438         return ret;
1439 }
1440
1441 int
1442 send_group_sigqueue(int sig, struct sigqueue *q, struct task_struct *p)
1443 {
1444         unsigned long flags;
1445         int ret = 0;
1446
1447         BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1448         read_lock(&tasklist_lock);
1449         spin_lock_irqsave(&p->sighand->siglock, flags);
1450         handle_stop_signal(sig, p);
1451
1452         /* Short-circuit ignored signals.  */
1453         if (sig_ignored(p, sig)) {
1454                 ret = 1;
1455                 goto out;
1456         }
1457
1458         if (unlikely(!list_empty(&q->list))) {
1459                 /*
1460                  * If an SI_TIMER entry is already queue just increment
1461                  * the overrun count.  Other uses should not try to
1462                  * send the signal multiple times.
1463                  */
1464                 if (q->info.si_code != SI_TIMER)
1465                         BUG();
1466                 q->info.si_overrun++;
1467                 goto out;
1468         } 
1469
1470         /*
1471          * Put this signal on the shared-pending queue.
1472          * We always use the shared queue for process-wide signals,
1473          * to avoid several races.
1474          */
1475         q->lock = &p->sighand->siglock;
1476         list_add_tail(&q->list, &p->signal->shared_pending.list);
1477         sigaddset(&p->signal->shared_pending.signal, sig);
1478
1479         __group_complete_signal(sig, p);
1480 out:
1481         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1482         read_unlock(&tasklist_lock);
1483         return(ret);
1484 }
1485
1486 /*
1487  * Wake up any threads in the parent blocked in wait* syscalls.
1488  */
1489 static inline void __wake_up_parent(struct task_struct *p,
1490                                     struct task_struct *parent)
1491 {
1492         wake_up_interruptible_sync(&parent->signal->wait_chldexit);
1493 }
1494
1495 /*
1496  * Let a parent know about the death of a child.
1497  * For a stopped/continued status change, use do_notify_parent_cldstop instead.
1498  */
1499
1500 void do_notify_parent(struct task_struct *tsk, int sig)
1501 {
1502         struct siginfo info;
1503         unsigned long flags;
1504         struct sighand_struct *psig;
1505
1506         BUG_ON(sig == -1);
1507
1508         /* do_notify_parent_cldstop should have been called instead.  */
1509         BUG_ON(tsk->state & (TASK_STOPPED|TASK_TRACED));
1510
1511         BUG_ON(!tsk->ptrace &&
1512                (tsk->group_leader != tsk || !thread_group_empty(tsk)));
1513
1514         info.si_signo = sig;
1515         info.si_errno = 0;
1516         info.si_pid = tsk->pid;
1517         info.si_uid = tsk->uid;
1518
1519         /* FIXME: find out whether or not this is supposed to be c*time. */
1520         info.si_utime = cputime_to_jiffies(cputime_add(tsk->utime,
1521                                                        tsk->signal->utime));
1522         info.si_stime = cputime_to_jiffies(cputime_add(tsk->stime,
1523                                                        tsk->signal->stime));
1524
1525         info.si_status = tsk->exit_code & 0x7f;
1526         if (tsk->exit_code & 0x80)
1527                 info.si_code = CLD_DUMPED;
1528         else if (tsk->exit_code & 0x7f)
1529                 info.si_code = CLD_KILLED;
1530         else {
1531                 info.si_code = CLD_EXITED;
1532                 info.si_status = tsk->exit_code >> 8;
1533         }
1534
1535         psig = tsk->parent->sighand;
1536         spin_lock_irqsave(&psig->siglock, flags);
1537         if (sig == SIGCHLD &&
1538             (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
1539              (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT))) {
1540                 /*
1541                  * We are exiting and our parent doesn't care.  POSIX.1
1542                  * defines special semantics for setting SIGCHLD to SIG_IGN
1543                  * or setting the SA_NOCLDWAIT flag: we should be reaped
1544                  * automatically and not left for our parent's wait4 call.
1545                  * Rather than having the parent do it as a magic kind of
1546                  * signal handler, we just set this to tell do_exit that we
1547                  * can be cleaned up without becoming a zombie.  Note that
1548                  * we still call __wake_up_parent in this case, because a
1549                  * blocked sys_wait4 might now return -ECHILD.
1550                  *
1551                  * Whether we send SIGCHLD or not for SA_NOCLDWAIT
1552                  * is implementation-defined: we do (if you don't want
1553                  * it, just use SIG_IGN instead).
1554                  */
1555                 tsk->exit_signal = -1;
1556                 if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
1557                         sig = 0;
1558         }
1559         if (valid_signal(sig) && sig > 0)
1560                 __group_send_sig_info(sig, &info, tsk->parent);
1561         __wake_up_parent(tsk, tsk->parent);
1562         spin_unlock_irqrestore(&psig->siglock, flags);
1563 }
1564
1565 static void do_notify_parent_cldstop(struct task_struct *tsk, int to_self, int why)
1566 {
1567         struct siginfo info;
1568         unsigned long flags;
1569         struct task_struct *parent;
1570         struct sighand_struct *sighand;
1571
1572         if (to_self)
1573                 parent = tsk->parent;
1574         else {
1575                 tsk = tsk->group_leader;
1576                 parent = tsk->real_parent;
1577         }
1578
1579         info.si_signo = SIGCHLD;
1580         info.si_errno = 0;
1581         info.si_pid = tsk->pid;
1582         info.si_uid = tsk->uid;
1583
1584         /* FIXME: find out whether or not this is supposed to be c*time. */
1585         info.si_utime = cputime_to_jiffies(tsk->utime);
1586         info.si_stime = cputime_to_jiffies(tsk->stime);
1587
1588         info.si_code = why;
1589         switch (why) {
1590         case CLD_CONTINUED:
1591                 info.si_status = SIGCONT;
1592                 break;
1593         case CLD_STOPPED:
1594                 info.si_status = tsk->signal->group_exit_code & 0x7f;
1595                 break;
1596         case CLD_TRAPPED:
1597                 info.si_status = tsk->exit_code & 0x7f;
1598                 break;
1599         default:
1600                 BUG();
1601         }
1602
1603         sighand = parent->sighand;
1604         spin_lock_irqsave(&sighand->siglock, flags);
1605         if (sighand->action[SIGCHLD-1].sa.sa_handler != SIG_IGN &&
1606             !(sighand->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDSTOP))
1607                 __group_send_sig_info(SIGCHLD, &info, parent);
1608         /*
1609          * Even if SIGCHLD is not generated, we must wake up wait4 calls.
1610          */
1611         __wake_up_parent(tsk, parent);
1612         spin_unlock_irqrestore(&sighand->siglock, flags);
1613 }
1614
1615 /*
1616  * This must be called with current->sighand->siglock held.
1617  *
1618  * This should be the path for all ptrace stops.
1619  * We always set current->last_siginfo while stopped here.
1620  * That makes it a way to test a stopped process for
1621  * being ptrace-stopped vs being job-control-stopped.
1622  *
1623  * If we actually decide not to stop at all because the tracer is gone,
1624  * we leave nostop_code in current->exit_code.
1625  */
1626 static void ptrace_stop(int exit_code, int nostop_code, siginfo_t *info)
1627 {
1628         /*
1629          * If there is a group stop in progress,
1630          * we must participate in the bookkeeping.
1631          */
1632         if (current->signal->group_stop_count > 0)
1633                 --current->signal->group_stop_count;
1634
1635         current->last_siginfo = info;
1636         current->exit_code = exit_code;
1637
1638         /* Let the debugger run.  */
1639         set_current_state(TASK_TRACED);
1640         spin_unlock_irq(&current->sighand->siglock);
1641         read_lock(&tasklist_lock);
1642         if (likely(current->ptrace & PT_PTRACED) &&
1643             likely(current->parent != current->real_parent ||
1644                    !(current->ptrace & PT_ATTACHED)) &&
1645             (likely(current->parent->signal != current->signal) ||
1646              !unlikely(current->signal->flags & SIGNAL_GROUP_EXIT))) {
1647                 do_notify_parent_cldstop(current, 1, CLD_TRAPPED);
1648                 read_unlock(&tasklist_lock);
1649                 schedule();
1650         } else {
1651                 /*
1652                  * By the time we got the lock, our tracer went away.
1653                  * Don't stop here.
1654                  */
1655                 read_unlock(&tasklist_lock);
1656                 set_current_state(TASK_RUNNING);
1657                 current->exit_code = nostop_code;
1658         }
1659
1660         /*
1661          * We are back.  Now reacquire the siglock before touching
1662          * last_siginfo, so that we are sure to have synchronized with
1663          * any signal-sending on another CPU that wants to examine it.
1664          */
1665         spin_lock_irq(&current->sighand->siglock);
1666         current->last_siginfo = NULL;
1667
1668         /*
1669          * Queued signals ignored us while we were stopped for tracing.
1670          * So check for any that we should take before resuming user mode.
1671          */
1672         recalc_sigpending();
1673 }
1674
1675 void ptrace_notify(int exit_code)
1676 {
1677         siginfo_t info;
1678
1679         BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1680
1681         memset(&info, 0, sizeof info);
1682         info.si_signo = SIGTRAP;
1683         info.si_code = exit_code;
1684         info.si_pid = current->pid;
1685         info.si_uid = current->uid;
1686
1687         /* Let the debugger run.  */
1688         spin_lock_irq(&current->sighand->siglock);
1689         ptrace_stop(exit_code, 0, &info);
1690         spin_unlock_irq(&current->sighand->siglock);
1691 }
1692
1693 static void
1694 finish_stop(int stop_count)
1695 {
1696         int to_self;
1697
1698         /*
1699          * If there are no other threads in the group, or if there is
1700          * a group stop in progress and we are the last to stop,
1701          * report to the parent.  When ptraced, every thread reports itself.
1702          */
1703         if (stop_count < 0 || (current->ptrace & PT_PTRACED))
1704                 to_self = 1;
1705         else if (stop_count == 0)
1706                 to_self = 0;
1707         else
1708                 goto out;
1709
1710         read_lock(&tasklist_lock);
1711         do_notify_parent_cldstop(current, to_self, CLD_STOPPED);
1712         read_unlock(&tasklist_lock);
1713
1714 out:
1715         schedule();
1716         /*
1717          * Now we don't run again until continued.
1718          */
1719         current->exit_code = 0;
1720 }
1721
1722 /*
1723  * This performs the stopping for SIGSTOP and other stop signals.
1724  * We have to stop all threads in the thread group.
1725  * Returns nonzero if we've actually stopped and released the siglock.
1726  * Returns zero if we didn't stop and still hold the siglock.
1727  */
1728 static int
1729 do_signal_stop(int signr)
1730 {
1731         struct signal_struct *sig = current->signal;
1732         struct sighand_struct *sighand = current->sighand;
1733         int stop_count = -1;
1734
1735         if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED))
1736                 return 0;
1737
1738         if (sig->group_stop_count > 0) {
1739                 /*
1740                  * There is a group stop in progress.  We don't need to
1741                  * start another one.
1742                  */
1743                 signr = sig->group_exit_code;
1744                 stop_count = --sig->group_stop_count;
1745                 current->exit_code = signr;
1746                 set_current_state(TASK_STOPPED);
1747                 if (stop_count == 0)
1748                         sig->flags = SIGNAL_STOP_STOPPED;
1749                 spin_unlock_irq(&sighand->siglock);
1750         }
1751         else if (thread_group_empty(current)) {
1752                 /*
1753                  * Lock must be held through transition to stopped state.
1754                  */
1755                 current->exit_code = current->signal->group_exit_code = signr;
1756                 set_current_state(TASK_STOPPED);
1757                 sig->flags = SIGNAL_STOP_STOPPED;
1758                 spin_unlock_irq(&sighand->siglock);
1759         }
1760         else {
1761                 /*
1762                  * There is no group stop already in progress.
1763                  * We must initiate one now, but that requires
1764                  * dropping siglock to get both the tasklist lock
1765                  * and siglock again in the proper order.  Note that
1766                  * this allows an intervening SIGCONT to be posted.
1767                  * We need to check for that and bail out if necessary.
1768                  */
1769                 struct task_struct *t;
1770
1771                 spin_unlock_irq(&sighand->siglock);
1772
1773                 /* signals can be posted during this window */
1774
1775                 read_lock(&tasklist_lock);
1776                 spin_lock_irq(&sighand->siglock);
1777
1778                 if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED)) {
1779                         /*
1780                          * Another stop or continue happened while we
1781                          * didn't have the lock.  We can just swallow this
1782                          * signal now.  If we raced with a SIGCONT, that
1783                          * should have just cleared it now.  If we raced
1784                          * with another processor delivering a stop signal,
1785                          * then the SIGCONT that wakes us up should clear it.
1786                          */
1787                         read_unlock(&tasklist_lock);
1788                         return 0;
1789                 }
1790
1791                 if (sig->group_stop_count == 0) {
1792                         sig->group_exit_code = signr;
1793                         stop_count = 0;
1794                         for (t = next_thread(current); t != current;
1795                              t = next_thread(t))
1796                                 /*
1797                                  * Setting state to TASK_STOPPED for a group
1798                                  * stop is always done with the siglock held,
1799                                  * so this check has no races.
1800                                  */
1801                                 if (!t->exit_state &&
1802                                     !(t->state & (TASK_STOPPED|TASK_TRACED))) {
1803                                         stop_count++;
1804                                         signal_wake_up(t, 0);
1805                                 }
1806                         sig->group_stop_count = stop_count;
1807                 }
1808                 else {
1809                         /* A race with another thread while unlocked.  */
1810                         signr = sig->group_exit_code;
1811                         stop_count = --sig->group_stop_count;
1812                 }
1813
1814                 current->exit_code = signr;
1815                 set_current_state(TASK_STOPPED);
1816                 if (stop_count == 0)
1817                         sig->flags = SIGNAL_STOP_STOPPED;
1818
1819                 spin_unlock_irq(&sighand->siglock);
1820                 read_unlock(&tasklist_lock);
1821         }
1822
1823         finish_stop(stop_count);
1824         return 1;
1825 }
1826
1827 /*
1828  * Do appropriate magic when group_stop_count > 0.
1829  * We return nonzero if we stopped, after releasing the siglock.
1830  * We return zero if we still hold the siglock and should look
1831  * for another signal without checking group_stop_count again.
1832  */
1833 static inline int handle_group_stop(void)
1834 {
1835         int stop_count;
1836
1837         if (current->signal->group_exit_task == current) {
1838                 /*
1839                  * Group stop is so we can do a core dump,
1840                  * We are the initiating thread, so get on with it.
1841                  */
1842                 current->signal->group_exit_task = NULL;
1843                 return 0;
1844         }
1845
1846         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1847                 /*
1848                  * Group stop is so another thread can do a core dump,
1849                  * or else we are racing against a death signal.
1850                  * Just punt the stop so we can get the next signal.
1851                  */
1852                 return 0;
1853
1854         /*
1855          * There is a group stop in progress.  We stop
1856          * without any associated signal being in our queue.
1857          */
1858         stop_count = --current->signal->group_stop_count;
1859         if (stop_count == 0)
1860                 current->signal->flags = SIGNAL_STOP_STOPPED;
1861         current->exit_code = current->signal->group_exit_code;
1862         set_current_state(TASK_STOPPED);
1863         spin_unlock_irq(&current->sighand->siglock);
1864         finish_stop(stop_count);
1865         return 1;
1866 }
1867
1868 int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
1869                           struct pt_regs *regs, void *cookie)
1870 {
1871         sigset_t *mask = &current->blocked;
1872         int signr = 0;
1873
1874 relock:
1875         spin_lock_irq(&current->sighand->siglock);
1876         for (;;) {
1877                 struct k_sigaction *ka;
1878
1879                 if (unlikely(current->signal->group_stop_count > 0) &&
1880                     handle_group_stop())
1881                         goto relock;
1882
1883                 signr = dequeue_signal(current, mask, info);
1884
1885                 if (!signr)
1886                         break; /* will return 0 */
1887
1888                 if ((current->ptrace & PT_PTRACED) && signr != SIGKILL) {
1889                         ptrace_signal_deliver(regs, cookie);
1890
1891                         /* Let the debugger run.  */
1892                         ptrace_stop(signr, signr, info);
1893
1894                         /* We're back.  Did the debugger cancel the sig?  */
1895                         signr = current->exit_code;
1896                         if (signr == 0)
1897                                 continue;
1898
1899                         current->exit_code = 0;
1900
1901                         /* Update the siginfo structure if the signal has
1902                            changed.  If the debugger wanted something
1903                            specific in the siginfo structure then it should
1904                            have updated *info via PTRACE_SETSIGINFO.  */
1905                         if (signr != info->si_signo) {
1906                                 info->si_signo = signr;
1907                                 info->si_errno = 0;
1908                                 info->si_code = SI_USER;
1909                                 info->si_pid = current->parent->pid;
1910                                 info->si_uid = current->parent->uid;
1911                         }
1912
1913                         /* If the (new) signal is now blocked, requeue it.  */
1914                         if (sigismember(&current->blocked, signr)) {
1915                                 specific_send_sig_info(signr, info, current);
1916                                 continue;
1917                         }
1918                 }
1919
1920                 ka = &current->sighand->action[signr-1];
1921                 if (ka->sa.sa_handler == SIG_IGN) /* Do nothing.  */
1922                         continue;
1923                 if (ka->sa.sa_handler != SIG_DFL) {
1924                         /* Run the handler.  */
1925                         *return_ka = *ka;
1926
1927                         if (ka->sa.sa_flags & SA_ONESHOT)
1928                                 ka->sa.sa_handler = SIG_DFL;
1929
1930                         break; /* will return non-zero "signr" value */
1931                 }
1932
1933                 /*
1934                  * Now we are doing the default action for this signal.
1935                  */
1936                 if (sig_kernel_ignore(signr)) /* Default is nothing. */
1937                         continue;
1938
1939                 /* Init gets no signals it doesn't want.  */
1940                 if (current->pid == 1)
1941                         continue;
1942
1943                 if (sig_kernel_stop(signr)) {
1944                         /*
1945                          * The default action is to stop all threads in
1946                          * the thread group.  The job control signals
1947                          * do nothing in an orphaned pgrp, but SIGSTOP
1948                          * always works.  Note that siglock needs to be
1949                          * dropped during the call to is_orphaned_pgrp()
1950                          * because of lock ordering with tasklist_lock.
1951                          * This allows an intervening SIGCONT to be posted.
1952                          * We need to check for that and bail out if necessary.
1953                          */
1954                         if (signr != SIGSTOP) {
1955                                 spin_unlock_irq(&current->sighand->siglock);
1956
1957                                 /* signals can be posted during this window */
1958
1959                                 if (is_orphaned_pgrp(process_group(current)))
1960                                         goto relock;
1961
1962                                 spin_lock_irq(&current->sighand->siglock);
1963                         }
1964
1965                         if (likely(do_signal_stop(signr))) {
1966                                 /* It released the siglock.  */
1967                                 goto relock;
1968                         }
1969
1970                         /*
1971                          * We didn't actually stop, due to a race
1972                          * with SIGCONT or something like that.
1973                          */
1974                         continue;
1975                 }
1976
1977                 spin_unlock_irq(&current->sighand->siglock);
1978
1979                 /*
1980                  * Anything else is fatal, maybe with a core dump.
1981                  */
1982                 current->flags |= PF_SIGNALED;
1983                 if (sig_kernel_coredump(signr)) {
1984                         /*
1985                          * If it was able to dump core, this kills all
1986                          * other threads in the group and synchronizes with
1987                          * their demise.  If we lost the race with another
1988                          * thread getting here, it set group_exit_code
1989                          * first and our do_group_exit call below will use
1990                          * that value and ignore the one we pass it.
1991                          */
1992                         do_coredump((long)signr, signr, regs);
1993                 }
1994
1995                 /*
1996                  * Death signals, no core dump.
1997                  */
1998                 do_group_exit(signr);
1999                 /* NOTREACHED */
2000         }
2001         spin_unlock_irq(&current->sighand->siglock);
2002         return signr;
2003 }
2004
2005 EXPORT_SYMBOL(recalc_sigpending);
2006 EXPORT_SYMBOL_GPL(dequeue_signal);
2007 EXPORT_SYMBOL(flush_signals);
2008 EXPORT_SYMBOL(force_sig);
2009 EXPORT_SYMBOL(kill_pg);
2010 EXPORT_SYMBOL(kill_proc);
2011 EXPORT_SYMBOL(ptrace_notify);
2012 EXPORT_SYMBOL(send_sig);
2013 EXPORT_SYMBOL(send_sig_info);
2014 EXPORT_SYMBOL(sigprocmask);
2015 EXPORT_SYMBOL(block_all_signals);
2016 EXPORT_SYMBOL(unblock_all_signals);
2017
2018
2019 /*
2020  * System call entry points.
2021  */
2022
2023 asmlinkage long sys_restart_syscall(void)
2024 {
2025         struct restart_block *restart = &current_thread_info()->restart_block;
2026         return restart->fn(restart);
2027 }
2028
2029 long do_no_restart_syscall(struct restart_block *param)
2030 {
2031         return -EINTR;
2032 }
2033
2034 /*
2035  * We don't need to get the kernel lock - this is all local to this
2036  * particular thread.. (and that's good, because this is _heavily_
2037  * used by various programs)
2038  */
2039
2040 /*
2041  * This is also useful for kernel threads that want to temporarily
2042  * (or permanently) block certain signals.
2043  *
2044  * NOTE! Unlike the user-mode sys_sigprocmask(), the kernel
2045  * interface happily blocks "unblockable" signals like SIGKILL
2046  * and friends.
2047  */
2048 int sigprocmask(int how, sigset_t *set, sigset_t *oldset)
2049 {
2050         int error;
2051         sigset_t old_block;
2052
2053         spin_lock_irq(&current->sighand->siglock);
2054         old_block = current->blocked;
2055         error = 0;
2056         switch (how) {
2057         case SIG_BLOCK:
2058                 sigorsets(&current->blocked, &current->blocked, set);
2059                 break;
2060         case SIG_UNBLOCK:
2061                 signandsets(&current->blocked, &current->blocked, set);
2062                 break;
2063         case SIG_SETMASK:
2064                 current->blocked = *set;
2065                 break;
2066         default:
2067                 error = -EINVAL;
2068         }
2069         recalc_sigpending();
2070         spin_unlock_irq(&current->sighand->siglock);
2071         if (oldset)
2072                 *oldset = old_block;
2073         return error;
2074 }
2075
2076 asmlinkage long
2077 sys_rt_sigprocmask(int how, sigset_t __user *set, sigset_t __user *oset, size_t sigsetsize)
2078 {
2079         int error = -EINVAL;
2080         sigset_t old_set, new_set;
2081
2082         /* XXX: Don't preclude handling different sized sigset_t's.  */
2083         if (sigsetsize != sizeof(sigset_t))
2084                 goto out;
2085
2086         if (set) {
2087                 error = -EFAULT;
2088                 if (copy_from_user(&new_set, set, sizeof(*set)))
2089                         goto out;
2090                 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
2091
2092                 error = sigprocmask(how, &new_set, &old_set);
2093                 if (error)
2094                         goto out;
2095                 if (oset)
2096                         goto set_old;
2097         } else if (oset) {
2098                 spin_lock_irq(&current->sighand->siglock);
2099                 old_set = current->blocked;
2100                 spin_unlock_irq(&current->sighand->siglock);
2101
2102         set_old:
2103                 error = -EFAULT;
2104                 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2105                         goto out;
2106         }
2107         error = 0;
2108 out:
2109         return error;
2110 }
2111
2112 long do_sigpending(void __user *set, unsigned long sigsetsize)
2113 {
2114         long error = -EINVAL;
2115         sigset_t pending;
2116
2117         if (sigsetsize > sizeof(sigset_t))
2118                 goto out;
2119
2120         spin_lock_irq(&current->sighand->siglock);
2121         sigorsets(&pending, &current->pending.signal,
2122                   &current->signal->shared_pending.signal);
2123         spin_unlock_irq(&current->sighand->siglock);
2124
2125         /* Outside the lock because only this thread touches it.  */
2126         sigandsets(&pending, &current->blocked, &pending);
2127
2128         error = -EFAULT;
2129         if (!copy_to_user(set, &pending, sigsetsize))
2130                 error = 0;
2131
2132 out:
2133         return error;
2134 }       
2135
2136 asmlinkage long
2137 sys_rt_sigpending(sigset_t __user *set, size_t sigsetsize)
2138 {
2139         return do_sigpending(set, sigsetsize);
2140 }
2141
2142 #ifndef HAVE_ARCH_COPY_SIGINFO_TO_USER
2143
2144 int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2145 {
2146         int err;
2147
2148         if (!access_ok (VERIFY_WRITE, to, sizeof(siginfo_t)))
2149                 return -EFAULT;
2150         if (from->si_code < 0)
2151                 return __copy_to_user(to, from, sizeof(siginfo_t))
2152                         ? -EFAULT : 0;
2153         /*
2154          * If you change siginfo_t structure, please be sure
2155          * this code is fixed accordingly.
2156          * It should never copy any pad contained in the structure
2157          * to avoid security leaks, but must copy the generic
2158          * 3 ints plus the relevant union member.
2159          */
2160         err = __put_user(from->si_signo, &to->si_signo);
2161         err |= __put_user(from->si_errno, &to->si_errno);
2162         err |= __put_user((short)from->si_code, &to->si_code);
2163         switch (from->si_code & __SI_MASK) {
2164         case __SI_KILL:
2165                 err |= __put_user(from->si_pid, &to->si_pid);
2166                 err |= __put_user(from->si_uid, &to->si_uid);
2167                 break;
2168         case __SI_TIMER:
2169                  err |= __put_user(from->si_tid, &to->si_tid);
2170                  err |= __put_user(from->si_overrun, &to->si_overrun);
2171                  err |= __put_user(from->si_ptr, &to->si_ptr);
2172                 break;
2173         case __SI_POLL:
2174                 err |= __put_user(from->si_band, &to->si_band);
2175                 err |= __put_user(from->si_fd, &to->si_fd);
2176                 break;
2177         case __SI_FAULT:
2178                 err |= __put_user(from->si_addr, &to->si_addr);
2179 #ifdef __ARCH_SI_TRAPNO
2180                 err |= __put_user(from->si_trapno, &to->si_trapno);
2181 #endif
2182                 break;
2183         case __SI_CHLD:
2184                 err |= __put_user(from->si_pid, &to->si_pid);
2185                 err |= __put_user(from->si_uid, &to->si_uid);
2186                 err |= __put_user(from->si_status, &to->si_status);
2187                 err |= __put_user(from->si_utime, &to->si_utime);
2188                 err |= __put_user(from->si_stime, &to->si_stime);
2189                 break;
2190         case __SI_RT: /* This is not generated by the kernel as of now. */
2191         case __SI_MESGQ: /* But this is */
2192                 err |= __put_user(from->si_pid, &to->si_pid);
2193                 err |= __put_user(from->si_uid, &to->si_uid);
2194                 err |= __put_user(from->si_ptr, &to->si_ptr);
2195                 break;
2196         default: /* this is just in case for now ... */
2197                 err |= __put_user(from->si_pid, &to->si_pid);
2198                 err |= __put_user(from->si_uid, &to->si_uid);
2199                 break;
2200         }
2201         return err;
2202 }
2203
2204 #endif
2205
2206 asmlinkage long
2207 sys_rt_sigtimedwait(const sigset_t __user *uthese,
2208                     siginfo_t __user *uinfo,
2209                     const struct timespec __user *uts,
2210                     size_t sigsetsize)
2211 {
2212         int ret, sig;
2213         sigset_t these;
2214         struct timespec ts;
2215         siginfo_t info;
2216         long timeout = 0;
2217
2218         /* XXX: Don't preclude handling different sized sigset_t's.  */
2219         if (sigsetsize != sizeof(sigset_t))
2220                 return -EINVAL;
2221
2222         if (copy_from_user(&these, uthese, sizeof(these)))
2223                 return -EFAULT;
2224                 
2225         /*
2226          * Invert the set of allowed signals to get those we
2227          * want to block.
2228          */
2229         sigdelsetmask(&these, sigmask(SIGKILL)|sigmask(SIGSTOP));
2230         signotset(&these);
2231
2232         if (uts) {
2233                 if (copy_from_user(&ts, uts, sizeof(ts)))
2234                         return -EFAULT;
2235                 if (ts.tv_nsec >= 1000000000L || ts.tv_nsec < 0
2236                     || ts.tv_sec < 0)
2237                         return -EINVAL;
2238         }
2239
2240         spin_lock_irq(&current->sighand->siglock);
2241         sig = dequeue_signal(current, &these, &info);
2242         if (!sig) {
2243                 timeout = MAX_SCHEDULE_TIMEOUT;
2244                 if (uts)
2245                         timeout = (timespec_to_jiffies(&ts)
2246                                    + (ts.tv_sec || ts.tv_nsec));
2247
2248                 if (timeout) {
2249                         /* None ready -- temporarily unblock those we're
2250                          * interested while we are sleeping in so that we'll
2251                          * be awakened when they arrive.  */
2252                         current->real_blocked = current->blocked;
2253                         sigandsets(&current->blocked, &current->blocked, &these);
2254                         recalc_sigpending();
2255                         spin_unlock_irq(&current->sighand->siglock);
2256
2257                         timeout = schedule_timeout_interruptible(timeout);
2258
2259                         try_to_freeze();
2260                         spin_lock_irq(&current->sighand->siglock);
2261                         sig = dequeue_signal(current, &these, &info);
2262                         current->blocked = current->real_blocked;
2263                         siginitset(&current->real_blocked, 0);
2264                         recalc_sigpending();
2265                 }
2266         }
2267         spin_unlock_irq(&current->sighand->siglock);
2268
2269         if (sig) {
2270                 ret = sig;
2271                 if (uinfo) {
2272                         if (copy_siginfo_to_user(uinfo, &info))
2273                                 ret = -EFAULT;
2274                 }
2275         } else {
2276                 ret = -EAGAIN;
2277                 if (timeout)
2278                         ret = -EINTR;
2279         }
2280
2281         return ret;
2282 }
2283
2284 asmlinkage long
2285 sys_kill(int pid, int sig)
2286 {
2287         struct siginfo info;
2288
2289         info.si_signo = sig;
2290         info.si_errno = 0;
2291         info.si_code = SI_USER;
2292         info.si_pid = current->tgid;
2293         info.si_uid = current->uid;
2294
2295         return kill_something_info(sig, &info, pid);
2296 }
2297
2298 /**
2299  *  sys_tgkill - send signal to one specific thread
2300  *  @tgid: the thread group ID of the thread
2301  *  @pid: the PID of the thread
2302  *  @sig: signal to be sent
2303  *
2304  *  This syscall also checks the tgid and returns -ESRCH even if the PID
2305  *  exists but it's not belonging to the target process anymore. This
2306  *  method solves the problem of threads exiting and PIDs getting reused.
2307  */
2308 asmlinkage long sys_tgkill(int tgid, int pid, int sig)
2309 {
2310         struct siginfo info;
2311         int error;
2312         struct task_struct *p;
2313
2314         /* This is only valid for single tasks */
2315         if (pid <= 0 || tgid <= 0)
2316                 return -EINVAL;
2317
2318         info.si_signo = sig;
2319         info.si_errno = 0;
2320         info.si_code = SI_TKILL;
2321         info.si_pid = current->tgid;
2322         info.si_uid = current->uid;
2323
2324         read_lock(&tasklist_lock);
2325         p = find_task_by_pid(pid);
2326         error = -ESRCH;
2327         if (p && (p->tgid == tgid)) {
2328                 error = check_kill_permission(sig, &info, p);
2329                 /*
2330                  * The null signal is a permissions and process existence
2331                  * probe.  No signal is actually delivered.
2332                  */
2333                 if (!error && sig && p->sighand) {
2334                         spin_lock_irq(&p->sighand->siglock);
2335                         handle_stop_signal(sig, p);
2336                         error = specific_send_sig_info(sig, &info, p);
2337                         spin_unlock_irq(&p->sighand->siglock);
2338                 }
2339         }
2340         read_unlock(&tasklist_lock);
2341         return error;
2342 }
2343
2344 /*
2345  *  Send a signal to only one task, even if it's a CLONE_THREAD task.
2346  */
2347 asmlinkage long
2348 sys_tkill(int pid, int sig)
2349 {
2350         struct siginfo info;
2351         int error;
2352         struct task_struct *p;
2353
2354         /* This is only valid for single tasks */
2355         if (pid <= 0)
2356                 return -EINVAL;
2357
2358         info.si_signo = sig;
2359         info.si_errno = 0;
2360         info.si_code = SI_TKILL;
2361         info.si_pid = current->tgid;
2362         info.si_uid = current->uid;
2363
2364         read_lock(&tasklist_lock);
2365         p = find_task_by_pid(pid);
2366         error = -ESRCH;
2367         if (p) {
2368                 error = check_kill_permission(sig, &info, p);
2369                 /*
2370                  * The null signal is a permissions and process existence
2371                  * probe.  No signal is actually delivered.
2372                  */
2373                 if (!error && sig && p->sighand) {
2374                         spin_lock_irq(&p->sighand->siglock);
2375                         handle_stop_signal(sig, p);
2376                         error = specific_send_sig_info(sig, &info, p);
2377                         spin_unlock_irq(&p->sighand->siglock);
2378                 }
2379         }
2380         read_unlock(&tasklist_lock);
2381         return error;
2382 }
2383
2384 asmlinkage long
2385 sys_rt_sigqueueinfo(int pid, int sig, siginfo_t __user *uinfo)
2386 {
2387         siginfo_t info;
2388
2389         if (copy_from_user(&info, uinfo, sizeof(siginfo_t)))
2390                 return -EFAULT;
2391
2392         /* Not even root can pretend to send signals from the kernel.
2393            Nor can they impersonate a kill(), which adds source info.  */
2394         if (info.si_code >= 0)
2395                 return -EPERM;
2396         info.si_signo = sig;
2397
2398         /* POSIX.1b doesn't mention process groups.  */
2399         return kill_proc_info(sig, &info, pid);
2400 }
2401
2402 int
2403 do_sigaction(int sig, const struct k_sigaction *act, struct k_sigaction *oact)
2404 {
2405         struct k_sigaction *k;
2406
2407         if (!valid_signal(sig) || sig < 1 || (act && sig_kernel_only(sig)))
2408                 return -EINVAL;
2409
2410         k = &current->sighand->action[sig-1];
2411
2412         spin_lock_irq(&current->sighand->siglock);
2413         if (signal_pending(current)) {
2414                 /*
2415                  * If there might be a fatal signal pending on multiple
2416                  * threads, make sure we take it before changing the action.
2417                  */
2418                 spin_unlock_irq(&current->sighand->siglock);
2419                 return -ERESTARTNOINTR;
2420         }
2421
2422         if (oact)
2423                 *oact = *k;
2424
2425         if (act) {
2426                 /*
2427                  * POSIX 3.3.1.3:
2428                  *  "Setting a signal action to SIG_IGN for a signal that is
2429                  *   pending shall cause the pending signal to be discarded,
2430                  *   whether or not it is blocked."
2431                  *
2432                  *  "Setting a signal action to SIG_DFL for a signal that is
2433                  *   pending and whose default action is to ignore the signal
2434                  *   (for example, SIGCHLD), shall cause the pending signal to
2435                  *   be discarded, whether or not it is blocked"
2436                  */
2437                 if (act->sa.sa_handler == SIG_IGN ||
2438                     (act->sa.sa_handler == SIG_DFL &&
2439                      sig_kernel_ignore(sig))) {
2440                         /*
2441                          * This is a fairly rare case, so we only take the
2442                          * tasklist_lock once we're sure we'll need it.
2443                          * Now we must do this little unlock and relock
2444                          * dance to maintain the lock hierarchy.
2445                          */
2446                         struct task_struct *t = current;
2447                         spin_unlock_irq(&t->sighand->siglock);
2448                         read_lock(&tasklist_lock);
2449                         spin_lock_irq(&t->sighand->siglock);
2450                         *k = *act;
2451                         sigdelsetmask(&k->sa.sa_mask,
2452                                       sigmask(SIGKILL) | sigmask(SIGSTOP));
2453                         rm_from_queue(sigmask(sig), &t->signal->shared_pending);
2454                         do {
2455                                 rm_from_queue(sigmask(sig), &t->pending);
2456                                 recalc_sigpending_tsk(t);
2457                                 t = next_thread(t);
2458                         } while (t != current);
2459                         spin_unlock_irq(&current->sighand->siglock);
2460                         read_unlock(&tasklist_lock);
2461                         return 0;
2462                 }
2463
2464                 *k = *act;
2465                 sigdelsetmask(&k->sa.sa_mask,
2466                               sigmask(SIGKILL) | sigmask(SIGSTOP));
2467         }
2468
2469         spin_unlock_irq(&current->sighand->siglock);
2470         return 0;
2471 }
2472
2473 int 
2474 do_sigaltstack (const stack_t __user *uss, stack_t __user *uoss, unsigned long sp)
2475 {
2476         stack_t oss;
2477         int error;
2478
2479         if (uoss) {
2480                 oss.ss_sp = (void __user *) current->sas_ss_sp;
2481                 oss.ss_size = current->sas_ss_size;
2482                 oss.ss_flags = sas_ss_flags(sp);
2483         }
2484
2485         if (uss) {
2486                 void __user *ss_sp;
2487                 size_t ss_size;
2488                 int ss_flags;
2489
2490                 error = -EFAULT;
2491                 if (!access_ok(VERIFY_READ, uss, sizeof(*uss))
2492                     || __get_user(ss_sp, &uss->ss_sp)
2493                     || __get_user(ss_flags, &uss->ss_flags)
2494                     || __get_user(ss_size, &uss->ss_size))
2495                         goto out;
2496
2497                 error = -EPERM;
2498                 if (on_sig_stack(sp))
2499                         goto out;
2500
2501                 error = -EINVAL;
2502                 /*
2503                  *
2504                  * Note - this code used to test ss_flags incorrectly
2505                  *        old code may have been written using ss_flags==0
2506                  *        to mean ss_flags==SS_ONSTACK (as this was the only
2507                  *        way that worked) - this fix preserves that older
2508                  *        mechanism
2509                  */
2510                 if (ss_flags != SS_DISABLE && ss_flags != SS_ONSTACK && ss_flags != 0)
2511                         goto out;
2512
2513                 if (ss_flags == SS_DISABLE) {
2514                         ss_size = 0;
2515                         ss_sp = NULL;
2516                 } else {
2517                         error = -ENOMEM;
2518                         if (ss_size < MINSIGSTKSZ)
2519                                 goto out;
2520                 }
2521
2522                 current->sas_ss_sp = (unsigned long) ss_sp;
2523                 current->sas_ss_size = ss_size;
2524         }
2525
2526         if (uoss) {
2527                 error = -EFAULT;
2528                 if (copy_to_user(uoss, &oss, sizeof(oss)))
2529                         goto out;
2530         }
2531
2532         error = 0;
2533 out:
2534         return error;
2535 }
2536
2537 #ifdef __ARCH_WANT_SYS_SIGPENDING
2538
2539 asmlinkage long
2540 sys_sigpending(old_sigset_t __user *set)
2541 {
2542         return do_sigpending(set, sizeof(*set));
2543 }
2544
2545 #endif
2546
2547 #ifdef __ARCH_WANT_SYS_SIGPROCMASK
2548 /* Some platforms have their own version with special arguments others
2549    support only sys_rt_sigprocmask.  */
2550
2551 asmlinkage long
2552 sys_sigprocmask(int how, old_sigset_t __user *set, old_sigset_t __user *oset)
2553 {
2554         int error;
2555         old_sigset_t old_set, new_set;
2556
2557         if (set) {
2558                 error = -EFAULT;
2559                 if (copy_from_user(&new_set, set, sizeof(*set)))
2560                         goto out;
2561                 new_set &= ~(sigmask(SIGKILL) | sigmask(SIGSTOP));
2562
2563                 spin_lock_irq(&current->sighand->siglock);
2564                 old_set = current->blocked.sig[0];
2565
2566                 error = 0;
2567                 switch (how) {
2568                 default:
2569                         error = -EINVAL;
2570                         break;
2571                 case SIG_BLOCK:
2572                         sigaddsetmask(&current->blocked, new_set);
2573                         break;
2574                 case SIG_UNBLOCK:
2575                         sigdelsetmask(&current->blocked, new_set);
2576                         break;
2577                 case SIG_SETMASK:
2578                         current->blocked.sig[0] = new_set;
2579                         break;
2580                 }
2581
2582                 recalc_sigpending();
2583                 spin_unlock_irq(&current->sighand->siglock);
2584                 if (error)
2585                         goto out;
2586                 if (oset)
2587                         goto set_old;
2588         } else if (oset) {
2589                 old_set = current->blocked.sig[0];
2590         set_old:
2591                 error = -EFAULT;
2592                 if (copy_to_user(oset, &old_set, sizeof(*oset)))
2593                         goto out;
2594         }
2595         error = 0;
2596 out:
2597         return error;
2598 }
2599 #endif /* __ARCH_WANT_SYS_SIGPROCMASK */
2600
2601 #ifdef __ARCH_WANT_SYS_RT_SIGACTION
2602 asmlinkage long
2603 sys_rt_sigaction(int sig,
2604                  const struct sigaction __user *act,
2605                  struct sigaction __user *oact,
2606                  size_t sigsetsize)
2607 {
2608         struct k_sigaction new_sa, old_sa;
2609         int ret = -EINVAL;
2610
2611         /* XXX: Don't preclude handling different sized sigset_t's.  */
2612         if (sigsetsize != sizeof(sigset_t))
2613                 goto out;
2614
2615         if (act) {
2616                 if (copy_from_user(&new_sa.sa, act, sizeof(new_sa.sa)))
2617                         return -EFAULT;
2618         }
2619
2620         ret = do_sigaction(sig, act ? &new_sa : NULL, oact ? &old_sa : NULL);
2621
2622         if (!ret && oact) {
2623                 if (copy_to_user(oact, &old_sa.sa, sizeof(old_sa.sa)))
2624                         return -EFAULT;
2625         }
2626 out:
2627         return ret;
2628 }
2629 #endif /* __ARCH_WANT_SYS_RT_SIGACTION */
2630
2631 #ifdef __ARCH_WANT_SYS_SGETMASK
2632
2633 /*
2634  * For backwards compatibility.  Functionality superseded by sigprocmask.
2635  */
2636 asmlinkage long
2637 sys_sgetmask(void)
2638 {
2639         /* SMP safe */
2640         return current->blocked.sig[0];
2641 }
2642
2643 asmlinkage long
2644 sys_ssetmask(int newmask)
2645 {
2646         int old;
2647
2648         spin_lock_irq(&current->sighand->siglock);
2649         old = current->blocked.sig[0];
2650
2651         siginitset(&current->blocked, newmask & ~(sigmask(SIGKILL)|
2652                                                   sigmask(SIGSTOP)));
2653         recalc_sigpending();
2654         spin_unlock_irq(&current->sighand->siglock);
2655
2656         return old;
2657 }
2658 #endif /* __ARCH_WANT_SGETMASK */
2659
2660 #ifdef __ARCH_WANT_SYS_SIGNAL
2661 /*
2662  * For backwards compatibility.  Functionality superseded by sigaction.
2663  */
2664 asmlinkage unsigned long
2665 sys_signal(int sig, __sighandler_t handler)
2666 {
2667         struct k_sigaction new_sa, old_sa;
2668         int ret;
2669
2670         new_sa.sa.sa_handler = handler;
2671         new_sa.sa.sa_flags = SA_ONESHOT | SA_NOMASK;
2672
2673         ret = do_sigaction(sig, &new_sa, &old_sa);
2674
2675         return ret ? ret : (unsigned long)old_sa.sa.sa_handler;
2676 }
2677 #endif /* __ARCH_WANT_SYS_SIGNAL */
2678
2679 #ifdef __ARCH_WANT_SYS_PAUSE
2680
2681 asmlinkage long
2682 sys_pause(void)
2683 {
2684         current->state = TASK_INTERRUPTIBLE;
2685         schedule();
2686         return -ERESTARTNOHAND;
2687 }
2688
2689 #endif
2690
2691 void __init signals_init(void)
2692 {
2693         sigqueue_cachep =
2694                 kmem_cache_create("sigqueue",
2695                                   sizeof(struct sigqueue),
2696                                   __alignof__(struct sigqueue),
2697                                   SLAB_PANIC, NULL, NULL);
2698 }