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