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