Merge branch 'master'
[linux-2.6] / kernel / exit.c
1 /*
2  *  linux/kernel/exit.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 #include <linux/config.h>
8 #include <linux/mm.h>
9 #include <linux/slab.h>
10 #include <linux/interrupt.h>
11 #include <linux/smp_lock.h>
12 #include <linux/module.h>
13 #include <linux/completion.h>
14 #include <linux/personality.h>
15 #include <linux/tty.h>
16 #include <linux/namespace.h>
17 #include <linux/key.h>
18 #include <linux/security.h>
19 #include <linux/cpu.h>
20 #include <linux/acct.h>
21 #include <linux/file.h>
22 #include <linux/binfmts.h>
23 #include <linux/ptrace.h>
24 #include <linux/profile.h>
25 #include <linux/mount.h>
26 #include <linux/proc_fs.h>
27 #include <linux/mempolicy.h>
28 #include <linux/cpuset.h>
29 #include <linux/syscalls.h>
30 #include <linux/signal.h>
31
32 #include <asm/uaccess.h>
33 #include <asm/unistd.h>
34 #include <asm/pgtable.h>
35 #include <asm/mmu_context.h>
36
37 extern void sem_exit (void);
38 extern struct task_struct *child_reaper;
39
40 int getrusage(struct task_struct *, int, struct rusage __user *);
41
42 static void exit_mm(struct task_struct * tsk);
43
44 static void __unhash_process(struct task_struct *p)
45 {
46         nr_threads--;
47         detach_pid(p, PIDTYPE_PID);
48         detach_pid(p, PIDTYPE_TGID);
49         if (thread_group_leader(p)) {
50                 detach_pid(p, PIDTYPE_PGID);
51                 detach_pid(p, PIDTYPE_SID);
52                 if (p->pid)
53                         __get_cpu_var(process_counts)--;
54         }
55
56         REMOVE_LINKS(p);
57 }
58
59 void release_task(struct task_struct * p)
60 {
61         int zap_leader;
62         task_t *leader;
63         struct dentry *proc_dentry;
64
65 repeat: 
66         atomic_dec(&p->user->processes);
67         spin_lock(&p->proc_lock);
68         proc_dentry = proc_pid_unhash(p);
69         write_lock_irq(&tasklist_lock);
70         if (unlikely(p->ptrace))
71                 __ptrace_unlink(p);
72         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
73         __exit_signal(p);
74         __exit_sighand(p);
75         /*
76          * Note that the fastpath in sys_times depends on __exit_signal having
77          * updated the counters before a task is removed from the tasklist of
78          * the process by __unhash_process.
79          */
80         __unhash_process(p);
81
82         /*
83          * If we are the last non-leader member of the thread
84          * group, and the leader is zombie, then notify the
85          * group leader's parent process. (if it wants notification.)
86          */
87         zap_leader = 0;
88         leader = p->group_leader;
89         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
90                 BUG_ON(leader->exit_signal == -1);
91                 do_notify_parent(leader, leader->exit_signal);
92                 /*
93                  * If we were the last child thread and the leader has
94                  * exited already, and the leader's parent ignores SIGCHLD,
95                  * then we are the one who should release the leader.
96                  *
97                  * do_notify_parent() will have marked it self-reaping in
98                  * that case.
99                  */
100                 zap_leader = (leader->exit_signal == -1);
101         }
102
103         sched_exit(p);
104         write_unlock_irq(&tasklist_lock);
105         spin_unlock(&p->proc_lock);
106         proc_pid_flush(proc_dentry);
107         release_thread(p);
108         put_task_struct(p);
109
110         p = leader;
111         if (unlikely(zap_leader))
112                 goto repeat;
113 }
114
115 /* we are using it only for SMP init */
116
117 void unhash_process(struct task_struct *p)
118 {
119         struct dentry *proc_dentry;
120
121         spin_lock(&p->proc_lock);
122         proc_dentry = proc_pid_unhash(p);
123         write_lock_irq(&tasklist_lock);
124         __unhash_process(p);
125         write_unlock_irq(&tasklist_lock);
126         spin_unlock(&p->proc_lock);
127         proc_pid_flush(proc_dentry);
128 }
129
130 /*
131  * This checks not only the pgrp, but falls back on the pid if no
132  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
133  * without this...
134  */
135 int session_of_pgrp(int pgrp)
136 {
137         struct task_struct *p;
138         int sid = -1;
139
140         read_lock(&tasklist_lock);
141         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
142                 if (p->signal->session > 0) {
143                         sid = p->signal->session;
144                         goto out;
145                 }
146         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
147         p = find_task_by_pid(pgrp);
148         if (p)
149                 sid = p->signal->session;
150 out:
151         read_unlock(&tasklist_lock);
152         
153         return sid;
154 }
155
156 /*
157  * Determine if a process group is "orphaned", according to the POSIX
158  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
159  * by terminal-generated stop signals.  Newly orphaned process groups are
160  * to receive a SIGHUP and a SIGCONT.
161  *
162  * "I ask you, have you ever known what it is to be an orphan?"
163  */
164 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
165 {
166         struct task_struct *p;
167         int ret = 1;
168
169         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
170                 if (p == ignored_task
171                                 || p->exit_state
172                                 || p->real_parent->pid == 1)
173                         continue;
174                 if (process_group(p->real_parent) != pgrp
175                             && p->real_parent->signal->session == p->signal->session) {
176                         ret = 0;
177                         break;
178                 }
179         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
180         return ret;     /* (sighing) "Often!" */
181 }
182
183 int is_orphaned_pgrp(int pgrp)
184 {
185         int retval;
186
187         read_lock(&tasklist_lock);
188         retval = will_become_orphaned_pgrp(pgrp, NULL);
189         read_unlock(&tasklist_lock);
190
191         return retval;
192 }
193
194 static inline int has_stopped_jobs(int pgrp)
195 {
196         int retval = 0;
197         struct task_struct *p;
198
199         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
200                 if (p->state != TASK_STOPPED)
201                         continue;
202
203                 /* If p is stopped by a debugger on a signal that won't
204                    stop it, then don't count p as stopped.  This isn't
205                    perfect but it's a good approximation.  */
206                 if (unlikely (p->ptrace)
207                     && p->exit_code != SIGSTOP
208                     && p->exit_code != SIGTSTP
209                     && p->exit_code != SIGTTOU
210                     && p->exit_code != SIGTTIN)
211                         continue;
212
213                 retval = 1;
214                 break;
215         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
216         return retval;
217 }
218
219 /**
220  * reparent_to_init - Reparent the calling kernel thread to the init task.
221  *
222  * If a kernel thread is launched as a result of a system call, or if
223  * it ever exits, it should generally reparent itself to init so that
224  * it is correctly cleaned up on exit.
225  *
226  * The various task state such as scheduling policy and priority may have
227  * been inherited from a user process, so we reset them to sane values here.
228  *
229  * NOTE that reparent_to_init() gives the caller full capabilities.
230  */
231 static inline void reparent_to_init(void)
232 {
233         write_lock_irq(&tasklist_lock);
234
235         ptrace_unlink(current);
236         /* Reparent to init */
237         REMOVE_LINKS(current);
238         current->parent = child_reaper;
239         current->real_parent = child_reaper;
240         SET_LINKS(current);
241
242         /* Set the exit signal to SIGCHLD so we signal init on exit */
243         current->exit_signal = SIGCHLD;
244
245         if ((current->policy == SCHED_NORMAL) && (task_nice(current) < 0))
246                 set_user_nice(current, 0);
247         /* cpus_allowed? */
248         /* rt_priority? */
249         /* signals? */
250         security_task_reparent_to_init(current);
251         memcpy(current->signal->rlim, init_task.signal->rlim,
252                sizeof(current->signal->rlim));
253         atomic_inc(&(INIT_USER->__count));
254         write_unlock_irq(&tasklist_lock);
255         switch_uid(INIT_USER);
256 }
257
258 void __set_special_pids(pid_t session, pid_t pgrp)
259 {
260         struct task_struct *curr = current;
261
262         if (curr->signal->session != session) {
263                 detach_pid(curr, PIDTYPE_SID);
264                 curr->signal->session = session;
265                 attach_pid(curr, PIDTYPE_SID, session);
266         }
267         if (process_group(curr) != pgrp) {
268                 detach_pid(curr, PIDTYPE_PGID);
269                 curr->signal->pgrp = pgrp;
270                 attach_pid(curr, PIDTYPE_PGID, pgrp);
271         }
272 }
273
274 void set_special_pids(pid_t session, pid_t pgrp)
275 {
276         write_lock_irq(&tasklist_lock);
277         __set_special_pids(session, pgrp);
278         write_unlock_irq(&tasklist_lock);
279 }
280
281 /*
282  * Let kernel threads use this to say that they
283  * allow a certain signal (since daemonize() will
284  * have disabled all of them by default).
285  */
286 int allow_signal(int sig)
287 {
288         if (!valid_signal(sig) || sig < 1)
289                 return -EINVAL;
290
291         spin_lock_irq(&current->sighand->siglock);
292         sigdelset(&current->blocked, sig);
293         if (!current->mm) {
294                 /* Kernel threads handle their own signals.
295                    Let the signal code know it'll be handled, so
296                    that they don't get converted to SIGKILL or
297                    just silently dropped */
298                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
299         }
300         recalc_sigpending();
301         spin_unlock_irq(&current->sighand->siglock);
302         return 0;
303 }
304
305 EXPORT_SYMBOL(allow_signal);
306
307 int disallow_signal(int sig)
308 {
309         if (!valid_signal(sig) || sig < 1)
310                 return -EINVAL;
311
312         spin_lock_irq(&current->sighand->siglock);
313         sigaddset(&current->blocked, sig);
314         recalc_sigpending();
315         spin_unlock_irq(&current->sighand->siglock);
316         return 0;
317 }
318
319 EXPORT_SYMBOL(disallow_signal);
320
321 /*
322  *      Put all the gunge required to become a kernel thread without
323  *      attached user resources in one place where it belongs.
324  */
325
326 void daemonize(const char *name, ...)
327 {
328         va_list args;
329         struct fs_struct *fs;
330         sigset_t blocked;
331
332         va_start(args, name);
333         vsnprintf(current->comm, sizeof(current->comm), name, args);
334         va_end(args);
335
336         /*
337          * If we were started as result of loading a module, close all of the
338          * user space pages.  We don't need them, and if we didn't close them
339          * they would be locked into memory.
340          */
341         exit_mm(current);
342
343         set_special_pids(1, 1);
344         down(&tty_sem);
345         current->signal->tty = NULL;
346         up(&tty_sem);
347
348         /* Block and flush all signals */
349         sigfillset(&blocked);
350         sigprocmask(SIG_BLOCK, &blocked, NULL);
351         flush_signals(current);
352
353         /* Become as one with the init task */
354
355         exit_fs(current);       /* current->fs->count--; */
356         fs = init_task.fs;
357         current->fs = fs;
358         atomic_inc(&fs->count);
359         exit_files(current);
360         current->files = init_task.files;
361         atomic_inc(&current->files->count);
362
363         reparent_to_init();
364 }
365
366 EXPORT_SYMBOL(daemonize);
367
368 static inline void close_files(struct files_struct * files)
369 {
370         int i, j;
371         struct fdtable *fdt;
372
373         j = 0;
374
375         /*
376          * It is safe to dereference the fd table without RCU or
377          * ->file_lock because this is the last reference to the
378          * files structure.
379          */
380         fdt = files_fdtable(files);
381         for (;;) {
382                 unsigned long set;
383                 i = j * __NFDBITS;
384                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
385                         break;
386                 set = fdt->open_fds->fds_bits[j++];
387                 while (set) {
388                         if (set & 1) {
389                                 struct file * file = xchg(&fdt->fd[i], NULL);
390                                 if (file)
391                                         filp_close(file, files);
392                         }
393                         i++;
394                         set >>= 1;
395                 }
396         }
397 }
398
399 struct files_struct *get_files_struct(struct task_struct *task)
400 {
401         struct files_struct *files;
402
403         task_lock(task);
404         files = task->files;
405         if (files)
406                 atomic_inc(&files->count);
407         task_unlock(task);
408
409         return files;
410 }
411
412 void fastcall put_files_struct(struct files_struct *files)
413 {
414         struct fdtable *fdt;
415
416         if (atomic_dec_and_test(&files->count)) {
417                 close_files(files);
418                 /*
419                  * Free the fd and fdset arrays if we expanded them.
420                  * If the fdtable was embedded, pass files for freeing
421                  * at the end of the RCU grace period. Otherwise,
422                  * you can free files immediately.
423                  */
424                 fdt = files_fdtable(files);
425                 if (fdt == &files->fdtab)
426                         fdt->free_files = files;
427                 else
428                         kmem_cache_free(files_cachep, files);
429                 free_fdtable(fdt);
430         }
431 }
432
433 EXPORT_SYMBOL(put_files_struct);
434
435 static inline void __exit_files(struct task_struct *tsk)
436 {
437         struct files_struct * files = tsk->files;
438
439         if (files) {
440                 task_lock(tsk);
441                 tsk->files = NULL;
442                 task_unlock(tsk);
443                 put_files_struct(files);
444         }
445 }
446
447 void exit_files(struct task_struct *tsk)
448 {
449         __exit_files(tsk);
450 }
451
452 static inline void __put_fs_struct(struct fs_struct *fs)
453 {
454         /* No need to hold fs->lock if we are killing it */
455         if (atomic_dec_and_test(&fs->count)) {
456                 dput(fs->root);
457                 mntput(fs->rootmnt);
458                 dput(fs->pwd);
459                 mntput(fs->pwdmnt);
460                 if (fs->altroot) {
461                         dput(fs->altroot);
462                         mntput(fs->altrootmnt);
463                 }
464                 kmem_cache_free(fs_cachep, fs);
465         }
466 }
467
468 void put_fs_struct(struct fs_struct *fs)
469 {
470         __put_fs_struct(fs);
471 }
472
473 static inline void __exit_fs(struct task_struct *tsk)
474 {
475         struct fs_struct * fs = tsk->fs;
476
477         if (fs) {
478                 task_lock(tsk);
479                 tsk->fs = NULL;
480                 task_unlock(tsk);
481                 __put_fs_struct(fs);
482         }
483 }
484
485 void exit_fs(struct task_struct *tsk)
486 {
487         __exit_fs(tsk);
488 }
489
490 EXPORT_SYMBOL_GPL(exit_fs);
491
492 /*
493  * Turn us into a lazy TLB process if we
494  * aren't already..
495  */
496 static void exit_mm(struct task_struct * tsk)
497 {
498         struct mm_struct *mm = tsk->mm;
499
500         mm_release(tsk, mm);
501         if (!mm)
502                 return;
503         /*
504          * Serialize with any possible pending coredump.
505          * We must hold mmap_sem around checking core_waiters
506          * and clearing tsk->mm.  The core-inducing thread
507          * will increment core_waiters for each thread in the
508          * group with ->mm != NULL.
509          */
510         down_read(&mm->mmap_sem);
511         if (mm->core_waiters) {
512                 up_read(&mm->mmap_sem);
513                 down_write(&mm->mmap_sem);
514                 if (!--mm->core_waiters)
515                         complete(mm->core_startup_done);
516                 up_write(&mm->mmap_sem);
517
518                 wait_for_completion(&mm->core_done);
519                 down_read(&mm->mmap_sem);
520         }
521         atomic_inc(&mm->mm_count);
522         if (mm != tsk->active_mm) BUG();
523         /* more a memory barrier than a real lock */
524         task_lock(tsk);
525         tsk->mm = NULL;
526         up_read(&mm->mmap_sem);
527         enter_lazy_tlb(mm, current);
528         task_unlock(tsk);
529         mmput(mm);
530 }
531
532 static inline void choose_new_parent(task_t *p, task_t *reaper, task_t *child_reaper)
533 {
534         /*
535          * Make sure we're not reparenting to ourselves and that
536          * the parent is not a zombie.
537          */
538         BUG_ON(p == reaper || reaper->exit_state >= EXIT_ZOMBIE);
539         p->real_parent = reaper;
540 }
541
542 static inline void reparent_thread(task_t *p, task_t *father, int traced)
543 {
544         /* We don't want people slaying init.  */
545         if (p->exit_signal != -1)
546                 p->exit_signal = SIGCHLD;
547
548         if (p->pdeath_signal)
549                 /* We already hold the tasklist_lock here.  */
550                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
551
552         /* Move the child from its dying parent to the new one.  */
553         if (unlikely(traced)) {
554                 /* Preserve ptrace links if someone else is tracing this child.  */
555                 list_del_init(&p->ptrace_list);
556                 if (p->parent != p->real_parent)
557                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
558         } else {
559                 /* If this child is being traced, then we're the one tracing it
560                  * anyway, so let go of it.
561                  */
562                 p->ptrace = 0;
563                 list_del_init(&p->sibling);
564                 p->parent = p->real_parent;
565                 list_add_tail(&p->sibling, &p->parent->children);
566
567                 /* If we'd notified the old parent about this child's death,
568                  * also notify the new parent.
569                  */
570                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
571                     thread_group_empty(p))
572                         do_notify_parent(p, p->exit_signal);
573                 else if (p->state == TASK_TRACED) {
574                         /*
575                          * If it was at a trace stop, turn it into
576                          * a normal stop since it's no longer being
577                          * traced.
578                          */
579                         ptrace_untrace(p);
580                 }
581         }
582
583         /*
584          * process group orphan check
585          * Case ii: Our child is in a different pgrp
586          * than we are, and it was the only connection
587          * outside, so the child pgrp is now orphaned.
588          */
589         if ((process_group(p) != process_group(father)) &&
590             (p->signal->session == father->signal->session)) {
591                 int pgrp = process_group(p);
592
593                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
594                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
595                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
596                 }
597         }
598 }
599
600 /*
601  * When we die, we re-parent all our children.
602  * Try to give them to another thread in our thread
603  * group, and if no such member exists, give it to
604  * the global child reaper process (ie "init")
605  */
606 static inline void forget_original_parent(struct task_struct * father,
607                                           struct list_head *to_release)
608 {
609         struct task_struct *p, *reaper = father;
610         struct list_head *_p, *_n;
611
612         do {
613                 reaper = next_thread(reaper);
614                 if (reaper == father) {
615                         reaper = child_reaper;
616                         break;
617                 }
618         } while (reaper->exit_state);
619
620         /*
621          * There are only two places where our children can be:
622          *
623          * - in our child list
624          * - in our ptraced child list
625          *
626          * Search them and reparent children.
627          */
628         list_for_each_safe(_p, _n, &father->children) {
629                 int ptrace;
630                 p = list_entry(_p,struct task_struct,sibling);
631
632                 ptrace = p->ptrace;
633
634                 /* if father isn't the real parent, then ptrace must be enabled */
635                 BUG_ON(father != p->real_parent && !ptrace);
636
637                 if (father == p->real_parent) {
638                         /* reparent with a reaper, real father it's us */
639                         choose_new_parent(p, reaper, child_reaper);
640                         reparent_thread(p, father, 0);
641                 } else {
642                         /* reparent ptraced task to its real parent */
643                         __ptrace_unlink (p);
644                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
645                             thread_group_empty(p))
646                                 do_notify_parent(p, p->exit_signal);
647                 }
648
649                 /*
650                  * if the ptraced child is a zombie with exit_signal == -1
651                  * we must collect it before we exit, or it will remain
652                  * zombie forever since we prevented it from self-reap itself
653                  * while it was being traced by us, to be able to see it in wait4.
654                  */
655                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
656                         list_add(&p->ptrace_list, to_release);
657         }
658         list_for_each_safe(_p, _n, &father->ptrace_children) {
659                 p = list_entry(_p,struct task_struct,ptrace_list);
660                 choose_new_parent(p, reaper, child_reaper);
661                 reparent_thread(p, father, 1);
662         }
663 }
664
665 /*
666  * Send signals to all our closest relatives so that they know
667  * to properly mourn us..
668  */
669 static void exit_notify(struct task_struct *tsk)
670 {
671         int state;
672         struct task_struct *t;
673         struct list_head ptrace_dead, *_p, *_n;
674
675         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
676             && !thread_group_empty(tsk)) {
677                 /*
678                  * This occurs when there was a race between our exit
679                  * syscall and a group signal choosing us as the one to
680                  * wake up.  It could be that we are the only thread
681                  * alerted to check for pending signals, but another thread
682                  * should be woken now to take the signal since we will not.
683                  * Now we'll wake all the threads in the group just to make
684                  * sure someone gets all the pending signals.
685                  */
686                 read_lock(&tasklist_lock);
687                 spin_lock_irq(&tsk->sighand->siglock);
688                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
689                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
690                                 recalc_sigpending_tsk(t);
691                                 if (signal_pending(t))
692                                         signal_wake_up(t, 0);
693                         }
694                 spin_unlock_irq(&tsk->sighand->siglock);
695                 read_unlock(&tasklist_lock);
696         }
697
698         write_lock_irq(&tasklist_lock);
699
700         /*
701          * This does two things:
702          *
703          * A.  Make init inherit all the child processes
704          * B.  Check to see if any process groups have become orphaned
705          *      as a result of our exiting, and if they have any stopped
706          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
707          */
708
709         INIT_LIST_HEAD(&ptrace_dead);
710         forget_original_parent(tsk, &ptrace_dead);
711         BUG_ON(!list_empty(&tsk->children));
712         BUG_ON(!list_empty(&tsk->ptrace_children));
713
714         /*
715          * Check to see if any process groups have become orphaned
716          * as a result of our exiting, and if they have any stopped
717          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
718          *
719          * Case i: Our father is in a different pgrp than we are
720          * and we were the only connection outside, so our pgrp
721          * is about to become orphaned.
722          */
723          
724         t = tsk->real_parent;
725         
726         if ((process_group(t) != process_group(tsk)) &&
727             (t->signal->session == tsk->signal->session) &&
728             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
729             has_stopped_jobs(process_group(tsk))) {
730                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
731                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
732         }
733
734         /* Let father know we died 
735          *
736          * Thread signals are configurable, but you aren't going to use
737          * that to send signals to arbitary processes. 
738          * That stops right now.
739          *
740          * If the parent exec id doesn't match the exec id we saved
741          * when we started then we know the parent has changed security
742          * domain.
743          *
744          * If our self_exec id doesn't match our parent_exec_id then
745          * we have changed execution domain as these two values started
746          * the same after a fork.
747          *      
748          */
749         
750         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
751             ( tsk->parent_exec_id != t->self_exec_id  ||
752               tsk->self_exec_id != tsk->parent_exec_id)
753             && !capable(CAP_KILL))
754                 tsk->exit_signal = SIGCHLD;
755
756
757         /* If something other than our normal parent is ptracing us, then
758          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
759          * only has special meaning to our real parent.
760          */
761         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
762                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
763                 do_notify_parent(tsk, signal);
764         } else if (tsk->ptrace) {
765                 do_notify_parent(tsk, SIGCHLD);
766         }
767
768         state = EXIT_ZOMBIE;
769         if (tsk->exit_signal == -1 &&
770             (likely(tsk->ptrace == 0) ||
771              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
772                 state = EXIT_DEAD;
773         tsk->exit_state = state;
774
775         write_unlock_irq(&tasklist_lock);
776
777         list_for_each_safe(_p, _n, &ptrace_dead) {
778                 list_del_init(_p);
779                 t = list_entry(_p,struct task_struct,ptrace_list);
780                 release_task(t);
781         }
782
783         /* If the process is dead, release it - nobody will wait for it */
784         if (state == EXIT_DEAD)
785                 release_task(tsk);
786 }
787
788 fastcall NORET_TYPE void do_exit(long code)
789 {
790         struct task_struct *tsk = current;
791         int group_dead;
792
793         profile_task_exit(tsk);
794
795         WARN_ON(atomic_read(&tsk->fs_excl));
796
797         if (unlikely(in_interrupt()))
798                 panic("Aiee, killing interrupt handler!");
799         if (unlikely(!tsk->pid))
800                 panic("Attempted to kill the idle task!");
801         if (unlikely(tsk->pid == 1))
802                 panic("Attempted to kill init!");
803         if (tsk->io_context)
804                 exit_io_context();
805
806         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
807                 current->ptrace_message = code;
808                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
809         }
810
811         /*
812          * We're taking recursive faults here in do_exit. Safest is to just
813          * leave this task alone and wait for reboot.
814          */
815         if (unlikely(tsk->flags & PF_EXITING)) {
816                 printk(KERN_ALERT
817                         "Fixing recursive fault but reboot is needed!\n");
818                 set_current_state(TASK_UNINTERRUPTIBLE);
819                 schedule();
820         }
821
822         tsk->flags |= PF_EXITING;
823
824         /*
825          * Make sure we don't try to process any timer firings
826          * while we are already exiting.
827          */
828         tsk->it_virt_expires = cputime_zero;
829         tsk->it_prof_expires = cputime_zero;
830         tsk->it_sched_expires = 0;
831
832         if (unlikely(in_atomic()))
833                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
834                                 current->comm, current->pid,
835                                 preempt_count());
836
837         acct_update_integrals(tsk);
838         if (tsk->mm) {
839                 update_hiwater_rss(tsk->mm);
840                 update_hiwater_vm(tsk->mm);
841         }
842         group_dead = atomic_dec_and_test(&tsk->signal->live);
843         if (group_dead) {
844                 del_timer_sync(&tsk->signal->real_timer);
845                 exit_itimers(tsk->signal);
846                 acct_process(code);
847         }
848         exit_mm(tsk);
849
850         exit_sem(tsk);
851         __exit_files(tsk);
852         __exit_fs(tsk);
853         exit_namespace(tsk);
854         exit_thread();
855         cpuset_exit(tsk);
856         exit_keys(tsk);
857
858         if (group_dead && tsk->signal->leader)
859                 disassociate_ctty(1);
860
861         module_put(tsk->thread_info->exec_domain->module);
862         if (tsk->binfmt)
863                 module_put(tsk->binfmt->module);
864
865         tsk->exit_code = code;
866         exit_notify(tsk);
867 #ifdef CONFIG_NUMA
868         mpol_free(tsk->mempolicy);
869         tsk->mempolicy = NULL;
870 #endif
871
872         /* PF_DEAD causes final put_task_struct after we schedule. */
873         preempt_disable();
874         BUG_ON(tsk->flags & PF_DEAD);
875         tsk->flags |= PF_DEAD;
876
877         schedule();
878         BUG();
879         /* Avoid "noreturn function does return".  */
880         for (;;) ;
881 }
882
883 EXPORT_SYMBOL_GPL(do_exit);
884
885 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
886 {
887         if (comp)
888                 complete(comp);
889         
890         do_exit(code);
891 }
892
893 EXPORT_SYMBOL(complete_and_exit);
894
895 asmlinkage long sys_exit(int error_code)
896 {
897         do_exit((error_code&0xff)<<8);
898 }
899
900 task_t fastcall *next_thread(const task_t *p)
901 {
902         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
903 }
904
905 EXPORT_SYMBOL(next_thread);
906
907 /*
908  * Take down every thread in the group.  This is called by fatal signals
909  * as well as by sys_exit_group (below).
910  */
911 NORET_TYPE void
912 do_group_exit(int exit_code)
913 {
914         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
915
916         if (current->signal->flags & SIGNAL_GROUP_EXIT)
917                 exit_code = current->signal->group_exit_code;
918         else if (!thread_group_empty(current)) {
919                 struct signal_struct *const sig = current->signal;
920                 struct sighand_struct *const sighand = current->sighand;
921                 read_lock(&tasklist_lock);
922                 spin_lock_irq(&sighand->siglock);
923                 if (sig->flags & SIGNAL_GROUP_EXIT)
924                         /* Another thread got here before we took the lock.  */
925                         exit_code = sig->group_exit_code;
926                 else {
927                         sig->flags = SIGNAL_GROUP_EXIT;
928                         sig->group_exit_code = exit_code;
929                         zap_other_threads(current);
930                 }
931                 spin_unlock_irq(&sighand->siglock);
932                 read_unlock(&tasklist_lock);
933         }
934
935         do_exit(exit_code);
936         /* NOTREACHED */
937 }
938
939 /*
940  * this kills every thread in the thread group. Note that any externally
941  * wait4()-ing process will get the correct exit code - even if this
942  * thread is not the thread group leader.
943  */
944 asmlinkage void sys_exit_group(int error_code)
945 {
946         do_group_exit((error_code & 0xff) << 8);
947 }
948
949 static int eligible_child(pid_t pid, int options, task_t *p)
950 {
951         if (pid > 0) {
952                 if (p->pid != pid)
953                         return 0;
954         } else if (!pid) {
955                 if (process_group(p) != process_group(current))
956                         return 0;
957         } else if (pid != -1) {
958                 if (process_group(p) != -pid)
959                         return 0;
960         }
961
962         /*
963          * Do not consider detached threads that are
964          * not ptraced:
965          */
966         if (p->exit_signal == -1 && !p->ptrace)
967                 return 0;
968
969         /* Wait for all children (clone and not) if __WALL is set;
970          * otherwise, wait for clone children *only* if __WCLONE is
971          * set; otherwise, wait for non-clone children *only*.  (Note:
972          * A "clone" child here is one that reports to its parent
973          * using a signal other than SIGCHLD.) */
974         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
975             && !(options & __WALL))
976                 return 0;
977         /*
978          * Do not consider thread group leaders that are
979          * in a non-empty thread group:
980          */
981         if (current->tgid != p->tgid && delay_group_leader(p))
982                 return 2;
983
984         if (security_task_wait(p))
985                 return 0;
986
987         return 1;
988 }
989
990 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
991                                int why, int status,
992                                struct siginfo __user *infop,
993                                struct rusage __user *rusagep)
994 {
995         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
996         put_task_struct(p);
997         if (!retval)
998                 retval = put_user(SIGCHLD, &infop->si_signo);
999         if (!retval)
1000                 retval = put_user(0, &infop->si_errno);
1001         if (!retval)
1002                 retval = put_user((short)why, &infop->si_code);
1003         if (!retval)
1004                 retval = put_user(pid, &infop->si_pid);
1005         if (!retval)
1006                 retval = put_user(uid, &infop->si_uid);
1007         if (!retval)
1008                 retval = put_user(status, &infop->si_status);
1009         if (!retval)
1010                 retval = pid;
1011         return retval;
1012 }
1013
1014 /*
1015  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1016  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1017  * the lock and this task is uninteresting.  If we return nonzero, we have
1018  * released the lock and the system call should return.
1019  */
1020 static int wait_task_zombie(task_t *p, int noreap,
1021                             struct siginfo __user *infop,
1022                             int __user *stat_addr, struct rusage __user *ru)
1023 {
1024         unsigned long state;
1025         int retval;
1026         int status;
1027
1028         if (unlikely(noreap)) {
1029                 pid_t pid = p->pid;
1030                 uid_t uid = p->uid;
1031                 int exit_code = p->exit_code;
1032                 int why, status;
1033
1034                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1035                         return 0;
1036                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1037                         return 0;
1038                 get_task_struct(p);
1039                 read_unlock(&tasklist_lock);
1040                 if ((exit_code & 0x7f) == 0) {
1041                         why = CLD_EXITED;
1042                         status = exit_code >> 8;
1043                 } else {
1044                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1045                         status = exit_code & 0x7f;
1046                 }
1047                 return wait_noreap_copyout(p, pid, uid, why,
1048                                            status, infop, ru);
1049         }
1050
1051         /*
1052          * Try to move the task's state to DEAD
1053          * only one thread is allowed to do this:
1054          */
1055         state = xchg(&p->exit_state, EXIT_DEAD);
1056         if (state != EXIT_ZOMBIE) {
1057                 BUG_ON(state != EXIT_DEAD);
1058                 return 0;
1059         }
1060         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1061                 /*
1062                  * This can only happen in a race with a ptraced thread
1063                  * dying on another processor.
1064                  */
1065                 return 0;
1066         }
1067
1068         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1069                 /*
1070                  * The resource counters for the group leader are in its
1071                  * own task_struct.  Those for dead threads in the group
1072                  * are in its signal_struct, as are those for the child
1073                  * processes it has previously reaped.  All these
1074                  * accumulate in the parent's signal_struct c* fields.
1075                  *
1076                  * We don't bother to take a lock here to protect these
1077                  * p->signal fields, because they are only touched by
1078                  * __exit_signal, which runs with tasklist_lock
1079                  * write-locked anyway, and so is excluded here.  We do
1080                  * need to protect the access to p->parent->signal fields,
1081                  * as other threads in the parent group can be right
1082                  * here reaping other children at the same time.
1083                  */
1084                 spin_lock_irq(&p->parent->sighand->siglock);
1085                 p->parent->signal->cutime =
1086                         cputime_add(p->parent->signal->cutime,
1087                         cputime_add(p->utime,
1088                         cputime_add(p->signal->utime,
1089                                     p->signal->cutime)));
1090                 p->parent->signal->cstime =
1091                         cputime_add(p->parent->signal->cstime,
1092                         cputime_add(p->stime,
1093                         cputime_add(p->signal->stime,
1094                                     p->signal->cstime)));
1095                 p->parent->signal->cmin_flt +=
1096                         p->min_flt + p->signal->min_flt + p->signal->cmin_flt;
1097                 p->parent->signal->cmaj_flt +=
1098                         p->maj_flt + p->signal->maj_flt + p->signal->cmaj_flt;
1099                 p->parent->signal->cnvcsw +=
1100                         p->nvcsw + p->signal->nvcsw + p->signal->cnvcsw;
1101                 p->parent->signal->cnivcsw +=
1102                         p->nivcsw + p->signal->nivcsw + p->signal->cnivcsw;
1103                 spin_unlock_irq(&p->parent->sighand->siglock);
1104         }
1105
1106         /*
1107          * Now we are sure this task is interesting, and no other
1108          * thread can reap it because we set its state to EXIT_DEAD.
1109          */
1110         read_unlock(&tasklist_lock);
1111
1112         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1113         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1114                 ? p->signal->group_exit_code : p->exit_code;
1115         if (!retval && stat_addr)
1116                 retval = put_user(status, stat_addr);
1117         if (!retval && infop)
1118                 retval = put_user(SIGCHLD, &infop->si_signo);
1119         if (!retval && infop)
1120                 retval = put_user(0, &infop->si_errno);
1121         if (!retval && infop) {
1122                 int why;
1123
1124                 if ((status & 0x7f) == 0) {
1125                         why = CLD_EXITED;
1126                         status >>= 8;
1127                 } else {
1128                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1129                         status &= 0x7f;
1130                 }
1131                 retval = put_user((short)why, &infop->si_code);
1132                 if (!retval)
1133                         retval = put_user(status, &infop->si_status);
1134         }
1135         if (!retval && infop)
1136                 retval = put_user(p->pid, &infop->si_pid);
1137         if (!retval && infop)
1138                 retval = put_user(p->uid, &infop->si_uid);
1139         if (retval) {
1140                 // TODO: is this safe?
1141                 p->exit_state = EXIT_ZOMBIE;
1142                 return retval;
1143         }
1144         retval = p->pid;
1145         if (p->real_parent != p->parent) {
1146                 write_lock_irq(&tasklist_lock);
1147                 /* Double-check with lock held.  */
1148                 if (p->real_parent != p->parent) {
1149                         __ptrace_unlink(p);
1150                         // TODO: is this safe?
1151                         p->exit_state = EXIT_ZOMBIE;
1152                         /*
1153                          * If this is not a detached task, notify the parent.
1154                          * If it's still not detached after that, don't release
1155                          * it now.
1156                          */
1157                         if (p->exit_signal != -1) {
1158                                 do_notify_parent(p, p->exit_signal);
1159                                 if (p->exit_signal != -1)
1160                                         p = NULL;
1161                         }
1162                 }
1163                 write_unlock_irq(&tasklist_lock);
1164         }
1165         if (p != NULL)
1166                 release_task(p);
1167         BUG_ON(!retval);
1168         return retval;
1169 }
1170
1171 /*
1172  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1173  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1174  * the lock and this task is uninteresting.  If we return nonzero, we have
1175  * released the lock and the system call should return.
1176  */
1177 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1178                              struct siginfo __user *infop,
1179                              int __user *stat_addr, struct rusage __user *ru)
1180 {
1181         int retval, exit_code;
1182
1183         if (!p->exit_code)
1184                 return 0;
1185         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1186             p->signal && p->signal->group_stop_count > 0)
1187                 /*
1188                  * A group stop is in progress and this is the group leader.
1189                  * We won't report until all threads have stopped.
1190                  */
1191                 return 0;
1192
1193         /*
1194          * Now we are pretty sure this task is interesting.
1195          * Make sure it doesn't get reaped out from under us while we
1196          * give up the lock and then examine it below.  We don't want to
1197          * keep holding onto the tasklist_lock while we call getrusage and
1198          * possibly take page faults for user memory.
1199          */
1200         get_task_struct(p);
1201         read_unlock(&tasklist_lock);
1202
1203         if (unlikely(noreap)) {
1204                 pid_t pid = p->pid;
1205                 uid_t uid = p->uid;
1206                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1207
1208                 exit_code = p->exit_code;
1209                 if (unlikely(!exit_code) ||
1210                     unlikely(p->state & TASK_TRACED))
1211                         goto bail_ref;
1212                 return wait_noreap_copyout(p, pid, uid,
1213                                            why, (exit_code << 8) | 0x7f,
1214                                            infop, ru);
1215         }
1216
1217         write_lock_irq(&tasklist_lock);
1218
1219         /*
1220          * This uses xchg to be atomic with the thread resuming and setting
1221          * it.  It must also be done with the write lock held to prevent a
1222          * race with the EXIT_ZOMBIE case.
1223          */
1224         exit_code = xchg(&p->exit_code, 0);
1225         if (unlikely(p->exit_state)) {
1226                 /*
1227                  * The task resumed and then died.  Let the next iteration
1228                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1229                  * already be zero here if it resumed and did _exit(0).
1230                  * The task itself is dead and won't touch exit_code again;
1231                  * other processors in this function are locked out.
1232                  */
1233                 p->exit_code = exit_code;
1234                 exit_code = 0;
1235         }
1236         if (unlikely(exit_code == 0)) {
1237                 /*
1238                  * Another thread in this function got to it first, or it
1239                  * resumed, or it resumed and then died.
1240                  */
1241                 write_unlock_irq(&tasklist_lock);
1242 bail_ref:
1243                 put_task_struct(p);
1244                 /*
1245                  * We are returning to the wait loop without having successfully
1246                  * removed the process and having released the lock. We cannot
1247                  * continue, since the "p" task pointer is potentially stale.
1248                  *
1249                  * Return -EAGAIN, and do_wait() will restart the loop from the
1250                  * beginning. Do _not_ re-acquire the lock.
1251                  */
1252                 return -EAGAIN;
1253         }
1254
1255         /* move to end of parent's list to avoid starvation */
1256         remove_parent(p);
1257         add_parent(p, p->parent);
1258
1259         write_unlock_irq(&tasklist_lock);
1260
1261         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1262         if (!retval && stat_addr)
1263                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1264         if (!retval && infop)
1265                 retval = put_user(SIGCHLD, &infop->si_signo);
1266         if (!retval && infop)
1267                 retval = put_user(0, &infop->si_errno);
1268         if (!retval && infop)
1269                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1270                                           ? CLD_TRAPPED : CLD_STOPPED),
1271                                   &infop->si_code);
1272         if (!retval && infop)
1273                 retval = put_user(exit_code, &infop->si_status);
1274         if (!retval && infop)
1275                 retval = put_user(p->pid, &infop->si_pid);
1276         if (!retval && infop)
1277                 retval = put_user(p->uid, &infop->si_uid);
1278         if (!retval)
1279                 retval = p->pid;
1280         put_task_struct(p);
1281
1282         BUG_ON(!retval);
1283         return retval;
1284 }
1285
1286 /*
1287  * Handle do_wait work for one task in a live, non-stopped state.
1288  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1289  * the lock and this task is uninteresting.  If we return nonzero, we have
1290  * released the lock and the system call should return.
1291  */
1292 static int wait_task_continued(task_t *p, int noreap,
1293                                struct siginfo __user *infop,
1294                                int __user *stat_addr, struct rusage __user *ru)
1295 {
1296         int retval;
1297         pid_t pid;
1298         uid_t uid;
1299
1300         if (unlikely(!p->signal))
1301                 return 0;
1302
1303         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1304                 return 0;
1305
1306         spin_lock_irq(&p->sighand->siglock);
1307         /* Re-check with the lock held.  */
1308         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1309                 spin_unlock_irq(&p->sighand->siglock);
1310                 return 0;
1311         }
1312         if (!noreap)
1313                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1314         spin_unlock_irq(&p->sighand->siglock);
1315
1316         pid = p->pid;
1317         uid = p->uid;
1318         get_task_struct(p);
1319         read_unlock(&tasklist_lock);
1320
1321         if (!infop) {
1322                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1323                 put_task_struct(p);
1324                 if (!retval && stat_addr)
1325                         retval = put_user(0xffff, stat_addr);
1326                 if (!retval)
1327                         retval = p->pid;
1328         } else {
1329                 retval = wait_noreap_copyout(p, pid, uid,
1330                                              CLD_CONTINUED, SIGCONT,
1331                                              infop, ru);
1332                 BUG_ON(retval == 0);
1333         }
1334
1335         return retval;
1336 }
1337
1338
1339 static inline int my_ptrace_child(struct task_struct *p)
1340 {
1341         if (!(p->ptrace & PT_PTRACED))
1342                 return 0;
1343         if (!(p->ptrace & PT_ATTACHED))
1344                 return 1;
1345         /*
1346          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1347          * we are the attacher.  If we are the real parent, this is a race
1348          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1349          * which we have to switch the parent links, but has already set
1350          * the flags in p->ptrace.
1351          */
1352         return (p->parent != p->real_parent);
1353 }
1354
1355 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1356                     int __user *stat_addr, struct rusage __user *ru)
1357 {
1358         DECLARE_WAITQUEUE(wait, current);
1359         struct task_struct *tsk;
1360         int flag, retval;
1361
1362         add_wait_queue(&current->signal->wait_chldexit,&wait);
1363 repeat:
1364         /*
1365          * We will set this flag if we see any child that might later
1366          * match our criteria, even if we are not able to reap it yet.
1367          */
1368         flag = 0;
1369         current->state = TASK_INTERRUPTIBLE;
1370         read_lock(&tasklist_lock);
1371         tsk = current;
1372         do {
1373                 struct task_struct *p;
1374                 struct list_head *_p;
1375                 int ret;
1376
1377                 list_for_each(_p,&tsk->children) {
1378                         p = list_entry(_p,struct task_struct,sibling);
1379
1380                         ret = eligible_child(pid, options, p);
1381                         if (!ret)
1382                                 continue;
1383
1384                         switch (p->state) {
1385                         case TASK_TRACED:
1386                                 /*
1387                                  * When we hit the race with PTRACE_ATTACH,
1388                                  * we will not report this child.  But the
1389                                  * race means it has not yet been moved to
1390                                  * our ptrace_children list, so we need to
1391                                  * set the flag here to avoid a spurious ECHILD
1392                                  * when the race happens with the only child.
1393                                  */
1394                                 flag = 1;
1395                                 if (!my_ptrace_child(p))
1396                                         continue;
1397                                 /*FALLTHROUGH*/
1398                         case TASK_STOPPED:
1399                                 /*
1400                                  * It's stopped now, so it might later
1401                                  * continue, exit, or stop again.
1402                                  */
1403                                 flag = 1;
1404                                 if (!(options & WUNTRACED) &&
1405                                     !my_ptrace_child(p))
1406                                         continue;
1407                                 retval = wait_task_stopped(p, ret == 2,
1408                                                            (options & WNOWAIT),
1409                                                            infop,
1410                                                            stat_addr, ru);
1411                                 if (retval == -EAGAIN)
1412                                         goto repeat;
1413                                 if (retval != 0) /* He released the lock.  */
1414                                         goto end;
1415                                 break;
1416                         default:
1417                         // case EXIT_DEAD:
1418                                 if (p->exit_state == EXIT_DEAD)
1419                                         continue;
1420                         // case EXIT_ZOMBIE:
1421                                 if (p->exit_state == EXIT_ZOMBIE) {
1422                                         /*
1423                                          * Eligible but we cannot release
1424                                          * it yet:
1425                                          */
1426                                         if (ret == 2)
1427                                                 goto check_continued;
1428                                         if (!likely(options & WEXITED))
1429                                                 continue;
1430                                         retval = wait_task_zombie(
1431                                                 p, (options & WNOWAIT),
1432                                                 infop, stat_addr, ru);
1433                                         /* He released the lock.  */
1434                                         if (retval != 0)
1435                                                 goto end;
1436                                         break;
1437                                 }
1438 check_continued:
1439                                 /*
1440                                  * It's running now, so it might later
1441                                  * exit, stop, or stop and then continue.
1442                                  */
1443                                 flag = 1;
1444                                 if (!unlikely(options & WCONTINUED))
1445                                         continue;
1446                                 retval = wait_task_continued(
1447                                         p, (options & WNOWAIT),
1448                                         infop, stat_addr, ru);
1449                                 if (retval != 0) /* He released the lock.  */
1450                                         goto end;
1451                                 break;
1452                         }
1453                 }
1454                 if (!flag) {
1455                         list_for_each(_p, &tsk->ptrace_children) {
1456                                 p = list_entry(_p, struct task_struct,
1457                                                 ptrace_list);
1458                                 if (!eligible_child(pid, options, p))
1459                                         continue;
1460                                 flag = 1;
1461                                 break;
1462                         }
1463                 }
1464                 if (options & __WNOTHREAD)
1465                         break;
1466                 tsk = next_thread(tsk);
1467                 if (tsk->signal != current->signal)
1468                         BUG();
1469         } while (tsk != current);
1470
1471         read_unlock(&tasklist_lock);
1472         if (flag) {
1473                 retval = 0;
1474                 if (options & WNOHANG)
1475                         goto end;
1476                 retval = -ERESTARTSYS;
1477                 if (signal_pending(current))
1478                         goto end;
1479                 schedule();
1480                 goto repeat;
1481         }
1482         retval = -ECHILD;
1483 end:
1484         current->state = TASK_RUNNING;
1485         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1486         if (infop) {
1487                 if (retval > 0)
1488                 retval = 0;
1489                 else {
1490                         /*
1491                          * For a WNOHANG return, clear out all the fields
1492                          * we would set so the user can easily tell the
1493                          * difference.
1494                          */
1495                         if (!retval)
1496                                 retval = put_user(0, &infop->si_signo);
1497                         if (!retval)
1498                                 retval = put_user(0, &infop->si_errno);
1499                         if (!retval)
1500                                 retval = put_user(0, &infop->si_code);
1501                         if (!retval)
1502                                 retval = put_user(0, &infop->si_pid);
1503                         if (!retval)
1504                                 retval = put_user(0, &infop->si_uid);
1505                         if (!retval)
1506                                 retval = put_user(0, &infop->si_status);
1507                 }
1508         }
1509         return retval;
1510 }
1511
1512 asmlinkage long sys_waitid(int which, pid_t pid,
1513                            struct siginfo __user *infop, int options,
1514                            struct rusage __user *ru)
1515 {
1516         long ret;
1517
1518         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1519                 return -EINVAL;
1520         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1521                 return -EINVAL;
1522
1523         switch (which) {
1524         case P_ALL:
1525                 pid = -1;
1526                 break;
1527         case P_PID:
1528                 if (pid <= 0)
1529                         return -EINVAL;
1530                 break;
1531         case P_PGID:
1532                 if (pid <= 0)
1533                         return -EINVAL;
1534                 pid = -pid;
1535                 break;
1536         default:
1537                 return -EINVAL;
1538         }
1539
1540         ret = do_wait(pid, options, infop, NULL, ru);
1541
1542         /* avoid REGPARM breakage on x86: */
1543         prevent_tail_call(ret);
1544         return ret;
1545 }
1546
1547 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1548                           int options, struct rusage __user *ru)
1549 {
1550         long ret;
1551
1552         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1553                         __WNOTHREAD|__WCLONE|__WALL))
1554                 return -EINVAL;
1555         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1556
1557         /* avoid REGPARM breakage on x86: */
1558         prevent_tail_call(ret);
1559         return ret;
1560 }
1561
1562 #ifdef __ARCH_WANT_SYS_WAITPID
1563
1564 /*
1565  * sys_waitpid() remains for compatibility. waitpid() should be
1566  * implemented by calling sys_wait4() from libc.a.
1567  */
1568 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1569 {
1570         return sys_wait4(pid, stat_addr, options, NULL);
1571 }
1572
1573 #endif