[PATCH] Refactor sys_reboot into reusable parts
[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
372         j = 0;
373         for (;;) {
374                 unsigned long set;
375                 i = j * __NFDBITS;
376                 if (i >= files->max_fdset || i >= files->max_fds)
377                         break;
378                 set = files->open_fds->fds_bits[j++];
379                 while (set) {
380                         if (set & 1) {
381                                 struct file * file = xchg(&files->fd[i], NULL);
382                                 if (file)
383                                         filp_close(file, files);
384                         }
385                         i++;
386                         set >>= 1;
387                 }
388         }
389 }
390
391 struct files_struct *get_files_struct(struct task_struct *task)
392 {
393         struct files_struct *files;
394
395         task_lock(task);
396         files = task->files;
397         if (files)
398                 atomic_inc(&files->count);
399         task_unlock(task);
400
401         return files;
402 }
403
404 void fastcall put_files_struct(struct files_struct *files)
405 {
406         if (atomic_dec_and_test(&files->count)) {
407                 close_files(files);
408                 /*
409                  * Free the fd and fdset arrays if we expanded them.
410                  */
411                 if (files->fd != &files->fd_array[0])
412                         free_fd_array(files->fd, files->max_fds);
413                 if (files->max_fdset > __FD_SETSIZE) {
414                         free_fdset(files->open_fds, files->max_fdset);
415                         free_fdset(files->close_on_exec, files->max_fdset);
416                 }
417                 kmem_cache_free(files_cachep, files);
418         }
419 }
420
421 EXPORT_SYMBOL(put_files_struct);
422
423 static inline void __exit_files(struct task_struct *tsk)
424 {
425         struct files_struct * files = tsk->files;
426
427         if (files) {
428                 task_lock(tsk);
429                 tsk->files = NULL;
430                 task_unlock(tsk);
431                 put_files_struct(files);
432         }
433 }
434
435 void exit_files(struct task_struct *tsk)
436 {
437         __exit_files(tsk);
438 }
439
440 static inline void __put_fs_struct(struct fs_struct *fs)
441 {
442         /* No need to hold fs->lock if we are killing it */
443         if (atomic_dec_and_test(&fs->count)) {
444                 dput(fs->root);
445                 mntput(fs->rootmnt);
446                 dput(fs->pwd);
447                 mntput(fs->pwdmnt);
448                 if (fs->altroot) {
449                         dput(fs->altroot);
450                         mntput(fs->altrootmnt);
451                 }
452                 kmem_cache_free(fs_cachep, fs);
453         }
454 }
455
456 void put_fs_struct(struct fs_struct *fs)
457 {
458         __put_fs_struct(fs);
459 }
460
461 static inline void __exit_fs(struct task_struct *tsk)
462 {
463         struct fs_struct * fs = tsk->fs;
464
465         if (fs) {
466                 task_lock(tsk);
467                 tsk->fs = NULL;
468                 task_unlock(tsk);
469                 __put_fs_struct(fs);
470         }
471 }
472
473 void exit_fs(struct task_struct *tsk)
474 {
475         __exit_fs(tsk);
476 }
477
478 EXPORT_SYMBOL_GPL(exit_fs);
479
480 /*
481  * Turn us into a lazy TLB process if we
482  * aren't already..
483  */
484 static void exit_mm(struct task_struct * tsk)
485 {
486         struct mm_struct *mm = tsk->mm;
487
488         mm_release(tsk, mm);
489         if (!mm)
490                 return;
491         /*
492          * Serialize with any possible pending coredump.
493          * We must hold mmap_sem around checking core_waiters
494          * and clearing tsk->mm.  The core-inducing thread
495          * will increment core_waiters for each thread in the
496          * group with ->mm != NULL.
497          */
498         down_read(&mm->mmap_sem);
499         if (mm->core_waiters) {
500                 up_read(&mm->mmap_sem);
501                 down_write(&mm->mmap_sem);
502                 if (!--mm->core_waiters)
503                         complete(mm->core_startup_done);
504                 up_write(&mm->mmap_sem);
505
506                 wait_for_completion(&mm->core_done);
507                 down_read(&mm->mmap_sem);
508         }
509         atomic_inc(&mm->mm_count);
510         if (mm != tsk->active_mm) BUG();
511         /* more a memory barrier than a real lock */
512         task_lock(tsk);
513         tsk->mm = NULL;
514         up_read(&mm->mmap_sem);
515         enter_lazy_tlb(mm, current);
516         task_unlock(tsk);
517         mmput(mm);
518 }
519
520 static inline void choose_new_parent(task_t *p, task_t *reaper, task_t *child_reaper)
521 {
522         /*
523          * Make sure we're not reparenting to ourselves and that
524          * the parent is not a zombie.
525          */
526         BUG_ON(p == reaper || reaper->exit_state >= EXIT_ZOMBIE);
527         p->real_parent = reaper;
528 }
529
530 static inline void reparent_thread(task_t *p, task_t *father, int traced)
531 {
532         /* We don't want people slaying init.  */
533         if (p->exit_signal != -1)
534                 p->exit_signal = SIGCHLD;
535
536         if (p->pdeath_signal)
537                 /* We already hold the tasklist_lock here.  */
538                 group_send_sig_info(p->pdeath_signal, (void *) 0, p);
539
540         /* Move the child from its dying parent to the new one.  */
541         if (unlikely(traced)) {
542                 /* Preserve ptrace links if someone else is tracing this child.  */
543                 list_del_init(&p->ptrace_list);
544                 if (p->parent != p->real_parent)
545                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
546         } else {
547                 /* If this child is being traced, then we're the one tracing it
548                  * anyway, so let go of it.
549                  */
550                 p->ptrace = 0;
551                 list_del_init(&p->sibling);
552                 p->parent = p->real_parent;
553                 list_add_tail(&p->sibling, &p->parent->children);
554
555                 /* If we'd notified the old parent about this child's death,
556                  * also notify the new parent.
557                  */
558                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
559                     thread_group_empty(p))
560                         do_notify_parent(p, p->exit_signal);
561                 else if (p->state == TASK_TRACED) {
562                         /*
563                          * If it was at a trace stop, turn it into
564                          * a normal stop since it's no longer being
565                          * traced.
566                          */
567                         ptrace_untrace(p);
568                 }
569         }
570
571         /*
572          * process group orphan check
573          * Case ii: Our child is in a different pgrp
574          * than we are, and it was the only connection
575          * outside, so the child pgrp is now orphaned.
576          */
577         if ((process_group(p) != process_group(father)) &&
578             (p->signal->session == father->signal->session)) {
579                 int pgrp = process_group(p);
580
581                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
582                         __kill_pg_info(SIGHUP, (void *)1, pgrp);
583                         __kill_pg_info(SIGCONT, (void *)1, pgrp);
584                 }
585         }
586 }
587
588 /*
589  * When we die, we re-parent all our children.
590  * Try to give them to another thread in our thread
591  * group, and if no such member exists, give it to
592  * the global child reaper process (ie "init")
593  */
594 static inline void forget_original_parent(struct task_struct * father,
595                                           struct list_head *to_release)
596 {
597         struct task_struct *p, *reaper = father;
598         struct list_head *_p, *_n;
599
600         do {
601                 reaper = next_thread(reaper);
602                 if (reaper == father) {
603                         reaper = child_reaper;
604                         break;
605                 }
606         } while (reaper->exit_state);
607
608         /*
609          * There are only two places where our children can be:
610          *
611          * - in our child list
612          * - in our ptraced child list
613          *
614          * Search them and reparent children.
615          */
616         list_for_each_safe(_p, _n, &father->children) {
617                 int ptrace;
618                 p = list_entry(_p,struct task_struct,sibling);
619
620                 ptrace = p->ptrace;
621
622                 /* if father isn't the real parent, then ptrace must be enabled */
623                 BUG_ON(father != p->real_parent && !ptrace);
624
625                 if (father == p->real_parent) {
626                         /* reparent with a reaper, real father it's us */
627                         choose_new_parent(p, reaper, child_reaper);
628                         reparent_thread(p, father, 0);
629                 } else {
630                         /* reparent ptraced task to its real parent */
631                         __ptrace_unlink (p);
632                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
633                             thread_group_empty(p))
634                                 do_notify_parent(p, p->exit_signal);
635                 }
636
637                 /*
638                  * if the ptraced child is a zombie with exit_signal == -1
639                  * we must collect it before we exit, or it will remain
640                  * zombie forever since we prevented it from self-reap itself
641                  * while it was being traced by us, to be able to see it in wait4.
642                  */
643                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
644                         list_add(&p->ptrace_list, to_release);
645         }
646         list_for_each_safe(_p, _n, &father->ptrace_children) {
647                 p = list_entry(_p,struct task_struct,ptrace_list);
648                 choose_new_parent(p, reaper, child_reaper);
649                 reparent_thread(p, father, 1);
650         }
651 }
652
653 /*
654  * Send signals to all our closest relatives so that they know
655  * to properly mourn us..
656  */
657 static void exit_notify(struct task_struct *tsk)
658 {
659         int state;
660         struct task_struct *t;
661         struct list_head ptrace_dead, *_p, *_n;
662
663         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
664             && !thread_group_empty(tsk)) {
665                 /*
666                  * This occurs when there was a race between our exit
667                  * syscall and a group signal choosing us as the one to
668                  * wake up.  It could be that we are the only thread
669                  * alerted to check for pending signals, but another thread
670                  * should be woken now to take the signal since we will not.
671                  * Now we'll wake all the threads in the group just to make
672                  * sure someone gets all the pending signals.
673                  */
674                 read_lock(&tasklist_lock);
675                 spin_lock_irq(&tsk->sighand->siglock);
676                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
677                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
678                                 recalc_sigpending_tsk(t);
679                                 if (signal_pending(t))
680                                         signal_wake_up(t, 0);
681                         }
682                 spin_unlock_irq(&tsk->sighand->siglock);
683                 read_unlock(&tasklist_lock);
684         }
685
686         write_lock_irq(&tasklist_lock);
687
688         /*
689          * This does two things:
690          *
691          * A.  Make init inherit all the child processes
692          * B.  Check to see if any process groups have become orphaned
693          *      as a result of our exiting, and if they have any stopped
694          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
695          */
696
697         INIT_LIST_HEAD(&ptrace_dead);
698         forget_original_parent(tsk, &ptrace_dead);
699         BUG_ON(!list_empty(&tsk->children));
700         BUG_ON(!list_empty(&tsk->ptrace_children));
701
702         /*
703          * Check to see if any process groups have become orphaned
704          * as a result of our exiting, and if they have any stopped
705          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
706          *
707          * Case i: Our father is in a different pgrp than we are
708          * and we were the only connection outside, so our pgrp
709          * is about to become orphaned.
710          */
711          
712         t = tsk->real_parent;
713         
714         if ((process_group(t) != process_group(tsk)) &&
715             (t->signal->session == tsk->signal->session) &&
716             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
717             has_stopped_jobs(process_group(tsk))) {
718                 __kill_pg_info(SIGHUP, (void *)1, process_group(tsk));
719                 __kill_pg_info(SIGCONT, (void *)1, process_group(tsk));
720         }
721
722         /* Let father know we died 
723          *
724          * Thread signals are configurable, but you aren't going to use
725          * that to send signals to arbitary processes. 
726          * That stops right now.
727          *
728          * If the parent exec id doesn't match the exec id we saved
729          * when we started then we know the parent has changed security
730          * domain.
731          *
732          * If our self_exec id doesn't match our parent_exec_id then
733          * we have changed execution domain as these two values started
734          * the same after a fork.
735          *      
736          */
737         
738         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
739             ( tsk->parent_exec_id != t->self_exec_id  ||
740               tsk->self_exec_id != tsk->parent_exec_id)
741             && !capable(CAP_KILL))
742                 tsk->exit_signal = SIGCHLD;
743
744
745         /* If something other than our normal parent is ptracing us, then
746          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
747          * only has special meaning to our real parent.
748          */
749         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
750                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
751                 do_notify_parent(tsk, signal);
752         } else if (tsk->ptrace) {
753                 do_notify_parent(tsk, SIGCHLD);
754         }
755
756         state = EXIT_ZOMBIE;
757         if (tsk->exit_signal == -1 &&
758             (likely(tsk->ptrace == 0) ||
759              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
760                 state = EXIT_DEAD;
761         tsk->exit_state = state;
762
763         write_unlock_irq(&tasklist_lock);
764
765         list_for_each_safe(_p, _n, &ptrace_dead) {
766                 list_del_init(_p);
767                 t = list_entry(_p,struct task_struct,ptrace_list);
768                 release_task(t);
769         }
770
771         /* If the process is dead, release it - nobody will wait for it */
772         if (state == EXIT_DEAD)
773                 release_task(tsk);
774
775         /* PF_DEAD causes final put_task_struct after we schedule. */
776         preempt_disable();
777         tsk->flags |= PF_DEAD;
778 }
779
780 fastcall NORET_TYPE void do_exit(long code)
781 {
782         struct task_struct *tsk = current;
783         int group_dead;
784
785         profile_task_exit(tsk);
786
787         WARN_ON(atomic_read(&tsk->fs_excl));
788
789         if (unlikely(in_interrupt()))
790                 panic("Aiee, killing interrupt handler!");
791         if (unlikely(!tsk->pid))
792                 panic("Attempted to kill the idle task!");
793         if (unlikely(tsk->pid == 1))
794                 panic("Attempted to kill init!");
795         if (tsk->io_context)
796                 exit_io_context();
797
798         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
799                 current->ptrace_message = code;
800                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
801         }
802
803         /*
804          * We're taking recursive faults here in do_exit. Safest is to just
805          * leave this task alone and wait for reboot.
806          */
807         if (unlikely(tsk->flags & PF_EXITING)) {
808                 printk(KERN_ALERT
809                         "Fixing recursive fault but reboot is needed!\n");
810                 set_current_state(TASK_UNINTERRUPTIBLE);
811                 schedule();
812         }
813
814         tsk->flags |= PF_EXITING;
815
816         /*
817          * Make sure we don't try to process any timer firings
818          * while we are already exiting.
819          */
820         tsk->it_virt_expires = cputime_zero;
821         tsk->it_prof_expires = cputime_zero;
822         tsk->it_sched_expires = 0;
823
824         if (unlikely(in_atomic()))
825                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
826                                 current->comm, current->pid,
827                                 preempt_count());
828
829         acct_update_integrals(tsk);
830         update_mem_hiwater(tsk);
831         group_dead = atomic_dec_and_test(&tsk->signal->live);
832         if (group_dead)
833                 acct_process(code);
834         exit_mm(tsk);
835
836         exit_sem(tsk);
837         __exit_files(tsk);
838         __exit_fs(tsk);
839         exit_namespace(tsk);
840         exit_thread();
841         cpuset_exit(tsk);
842         exit_keys(tsk);
843
844         if (group_dead && tsk->signal->leader)
845                 disassociate_ctty(1);
846
847         module_put(tsk->thread_info->exec_domain->module);
848         if (tsk->binfmt)
849                 module_put(tsk->binfmt->module);
850
851         tsk->exit_code = code;
852         exit_notify(tsk);
853 #ifdef CONFIG_NUMA
854         mpol_free(tsk->mempolicy);
855         tsk->mempolicy = NULL;
856 #endif
857
858         BUG_ON(!(current->flags & PF_DEAD));
859         schedule();
860         BUG();
861         /* Avoid "noreturn function does return".  */
862         for (;;) ;
863 }
864
865 EXPORT_SYMBOL_GPL(do_exit);
866
867 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
868 {
869         if (comp)
870                 complete(comp);
871         
872         do_exit(code);
873 }
874
875 EXPORT_SYMBOL(complete_and_exit);
876
877 asmlinkage long sys_exit(int error_code)
878 {
879         do_exit((error_code&0xff)<<8);
880 }
881
882 task_t fastcall *next_thread(const task_t *p)
883 {
884         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
885 }
886
887 EXPORT_SYMBOL(next_thread);
888
889 /*
890  * Take down every thread in the group.  This is called by fatal signals
891  * as well as by sys_exit_group (below).
892  */
893 NORET_TYPE void
894 do_group_exit(int exit_code)
895 {
896         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
897
898         if (current->signal->flags & SIGNAL_GROUP_EXIT)
899                 exit_code = current->signal->group_exit_code;
900         else if (!thread_group_empty(current)) {
901                 struct signal_struct *const sig = current->signal;
902                 struct sighand_struct *const sighand = current->sighand;
903                 read_lock(&tasklist_lock);
904                 spin_lock_irq(&sighand->siglock);
905                 if (sig->flags & SIGNAL_GROUP_EXIT)
906                         /* Another thread got here before we took the lock.  */
907                         exit_code = sig->group_exit_code;
908                 else {
909                         sig->flags = SIGNAL_GROUP_EXIT;
910                         sig->group_exit_code = exit_code;
911                         zap_other_threads(current);
912                 }
913                 spin_unlock_irq(&sighand->siglock);
914                 read_unlock(&tasklist_lock);
915         }
916
917         do_exit(exit_code);
918         /* NOTREACHED */
919 }
920
921 /*
922  * this kills every thread in the thread group. Note that any externally
923  * wait4()-ing process will get the correct exit code - even if this
924  * thread is not the thread group leader.
925  */
926 asmlinkage void sys_exit_group(int error_code)
927 {
928         do_group_exit((error_code & 0xff) << 8);
929 }
930
931 static int eligible_child(pid_t pid, int options, task_t *p)
932 {
933         if (pid > 0) {
934                 if (p->pid != pid)
935                         return 0;
936         } else if (!pid) {
937                 if (process_group(p) != process_group(current))
938                         return 0;
939         } else if (pid != -1) {
940                 if (process_group(p) != -pid)
941                         return 0;
942         }
943
944         /*
945          * Do not consider detached threads that are
946          * not ptraced:
947          */
948         if (p->exit_signal == -1 && !p->ptrace)
949                 return 0;
950
951         /* Wait for all children (clone and not) if __WALL is set;
952          * otherwise, wait for clone children *only* if __WCLONE is
953          * set; otherwise, wait for non-clone children *only*.  (Note:
954          * A "clone" child here is one that reports to its parent
955          * using a signal other than SIGCHLD.) */
956         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
957             && !(options & __WALL))
958                 return 0;
959         /*
960          * Do not consider thread group leaders that are
961          * in a non-empty thread group:
962          */
963         if (current->tgid != p->tgid && delay_group_leader(p))
964                 return 2;
965
966         if (security_task_wait(p))
967                 return 0;
968
969         return 1;
970 }
971
972 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
973                                int why, int status,
974                                struct siginfo __user *infop,
975                                struct rusage __user *rusagep)
976 {
977         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
978         put_task_struct(p);
979         if (!retval)
980                 retval = put_user(SIGCHLD, &infop->si_signo);
981         if (!retval)
982                 retval = put_user(0, &infop->si_errno);
983         if (!retval)
984                 retval = put_user((short)why, &infop->si_code);
985         if (!retval)
986                 retval = put_user(pid, &infop->si_pid);
987         if (!retval)
988                 retval = put_user(uid, &infop->si_uid);
989         if (!retval)
990                 retval = put_user(status, &infop->si_status);
991         if (!retval)
992                 retval = pid;
993         return retval;
994 }
995
996 /*
997  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
998  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
999  * the lock and this task is uninteresting.  If we return nonzero, we have
1000  * released the lock and the system call should return.
1001  */
1002 static int wait_task_zombie(task_t *p, int noreap,
1003                             struct siginfo __user *infop,
1004                             int __user *stat_addr, struct rusage __user *ru)
1005 {
1006         unsigned long state;
1007         int retval;
1008         int status;
1009
1010         if (unlikely(noreap)) {
1011                 pid_t pid = p->pid;
1012                 uid_t uid = p->uid;
1013                 int exit_code = p->exit_code;
1014                 int why, status;
1015
1016                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1017                         return 0;
1018                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1019                         return 0;
1020                 get_task_struct(p);
1021                 read_unlock(&tasklist_lock);
1022                 if ((exit_code & 0x7f) == 0) {
1023                         why = CLD_EXITED;
1024                         status = exit_code >> 8;
1025                 } else {
1026                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1027                         status = exit_code & 0x7f;
1028                 }
1029                 return wait_noreap_copyout(p, pid, uid, why,
1030                                            status, infop, ru);
1031         }
1032
1033         /*
1034          * Try to move the task's state to DEAD
1035          * only one thread is allowed to do this:
1036          */
1037         state = xchg(&p->exit_state, EXIT_DEAD);
1038         if (state != EXIT_ZOMBIE) {
1039                 BUG_ON(state != EXIT_DEAD);
1040                 return 0;
1041         }
1042         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1043                 /*
1044                  * This can only happen in a race with a ptraced thread
1045                  * dying on another processor.
1046                  */
1047                 return 0;
1048         }
1049
1050         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1051                 /*
1052                  * The resource counters for the group leader are in its
1053                  * own task_struct.  Those for dead threads in the group
1054                  * are in its signal_struct, as are those for the child
1055                  * processes it has previously reaped.  All these
1056                  * accumulate in the parent's signal_struct c* fields.
1057                  *
1058                  * We don't bother to take a lock here to protect these
1059                  * p->signal fields, because they are only touched by
1060                  * __exit_signal, which runs with tasklist_lock
1061                  * write-locked anyway, and so is excluded here.  We do
1062                  * need to protect the access to p->parent->signal fields,
1063                  * as other threads in the parent group can be right
1064                  * here reaping other children at the same time.
1065                  */
1066                 spin_lock_irq(&p->parent->sighand->siglock);
1067                 p->parent->signal->cutime =
1068                         cputime_add(p->parent->signal->cutime,
1069                         cputime_add(p->utime,
1070                         cputime_add(p->signal->utime,
1071                                     p->signal->cutime)));
1072                 p->parent->signal->cstime =
1073                         cputime_add(p->parent->signal->cstime,
1074                         cputime_add(p->stime,
1075                         cputime_add(p->signal->stime,
1076                                     p->signal->cstime)));
1077                 p->parent->signal->cmin_flt +=
1078                         p->min_flt + p->signal->min_flt + p->signal->cmin_flt;
1079                 p->parent->signal->cmaj_flt +=
1080                         p->maj_flt + p->signal->maj_flt + p->signal->cmaj_flt;
1081                 p->parent->signal->cnvcsw +=
1082                         p->nvcsw + p->signal->nvcsw + p->signal->cnvcsw;
1083                 p->parent->signal->cnivcsw +=
1084                         p->nivcsw + p->signal->nivcsw + p->signal->cnivcsw;
1085                 spin_unlock_irq(&p->parent->sighand->siglock);
1086         }
1087
1088         /*
1089          * Now we are sure this task is interesting, and no other
1090          * thread can reap it because we set its state to EXIT_DEAD.
1091          */
1092         read_unlock(&tasklist_lock);
1093
1094         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1095         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1096                 ? p->signal->group_exit_code : p->exit_code;
1097         if (!retval && stat_addr)
1098                 retval = put_user(status, stat_addr);
1099         if (!retval && infop)
1100                 retval = put_user(SIGCHLD, &infop->si_signo);
1101         if (!retval && infop)
1102                 retval = put_user(0, &infop->si_errno);
1103         if (!retval && infop) {
1104                 int why;
1105
1106                 if ((status & 0x7f) == 0) {
1107                         why = CLD_EXITED;
1108                         status >>= 8;
1109                 } else {
1110                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1111                         status &= 0x7f;
1112                 }
1113                 retval = put_user((short)why, &infop->si_code);
1114                 if (!retval)
1115                         retval = put_user(status, &infop->si_status);
1116         }
1117         if (!retval && infop)
1118                 retval = put_user(p->pid, &infop->si_pid);
1119         if (!retval && infop)
1120                 retval = put_user(p->uid, &infop->si_uid);
1121         if (retval) {
1122                 // TODO: is this safe?
1123                 p->exit_state = EXIT_ZOMBIE;
1124                 return retval;
1125         }
1126         retval = p->pid;
1127         if (p->real_parent != p->parent) {
1128                 write_lock_irq(&tasklist_lock);
1129                 /* Double-check with lock held.  */
1130                 if (p->real_parent != p->parent) {
1131                         __ptrace_unlink(p);
1132                         // TODO: is this safe?
1133                         p->exit_state = EXIT_ZOMBIE;
1134                         /*
1135                          * If this is not a detached task, notify the parent.
1136                          * If it's still not detached after that, don't release
1137                          * it now.
1138                          */
1139                         if (p->exit_signal != -1) {
1140                                 do_notify_parent(p, p->exit_signal);
1141                                 if (p->exit_signal != -1)
1142                                         p = NULL;
1143                         }
1144                 }
1145                 write_unlock_irq(&tasklist_lock);
1146         }
1147         if (p != NULL)
1148                 release_task(p);
1149         BUG_ON(!retval);
1150         return retval;
1151 }
1152
1153 /*
1154  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1155  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1156  * the lock and this task is uninteresting.  If we return nonzero, we have
1157  * released the lock and the system call should return.
1158  */
1159 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1160                              struct siginfo __user *infop,
1161                              int __user *stat_addr, struct rusage __user *ru)
1162 {
1163         int retval, exit_code;
1164
1165         if (!p->exit_code)
1166                 return 0;
1167         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1168             p->signal && p->signal->group_stop_count > 0)
1169                 /*
1170                  * A group stop is in progress and this is the group leader.
1171                  * We won't report until all threads have stopped.
1172                  */
1173                 return 0;
1174
1175         /*
1176          * Now we are pretty sure this task is interesting.
1177          * Make sure it doesn't get reaped out from under us while we
1178          * give up the lock and then examine it below.  We don't want to
1179          * keep holding onto the tasklist_lock while we call getrusage and
1180          * possibly take page faults for user memory.
1181          */
1182         get_task_struct(p);
1183         read_unlock(&tasklist_lock);
1184
1185         if (unlikely(noreap)) {
1186                 pid_t pid = p->pid;
1187                 uid_t uid = p->uid;
1188                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1189
1190                 exit_code = p->exit_code;
1191                 if (unlikely(!exit_code) ||
1192                     unlikely(p->state > TASK_STOPPED))
1193                         goto bail_ref;
1194                 return wait_noreap_copyout(p, pid, uid,
1195                                            why, (exit_code << 8) | 0x7f,
1196                                            infop, ru);
1197         }
1198
1199         write_lock_irq(&tasklist_lock);
1200
1201         /*
1202          * This uses xchg to be atomic with the thread resuming and setting
1203          * it.  It must also be done with the write lock held to prevent a
1204          * race with the EXIT_ZOMBIE case.
1205          */
1206         exit_code = xchg(&p->exit_code, 0);
1207         if (unlikely(p->exit_state)) {
1208                 /*
1209                  * The task resumed and then died.  Let the next iteration
1210                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1211                  * already be zero here if it resumed and did _exit(0).
1212                  * The task itself is dead and won't touch exit_code again;
1213                  * other processors in this function are locked out.
1214                  */
1215                 p->exit_code = exit_code;
1216                 exit_code = 0;
1217         }
1218         if (unlikely(exit_code == 0)) {
1219                 /*
1220                  * Another thread in this function got to it first, or it
1221                  * resumed, or it resumed and then died.
1222                  */
1223                 write_unlock_irq(&tasklist_lock);
1224 bail_ref:
1225                 put_task_struct(p);
1226                 /*
1227                  * We are returning to the wait loop without having successfully
1228                  * removed the process and having released the lock. We cannot
1229                  * continue, since the "p" task pointer is potentially stale.
1230                  *
1231                  * Return -EAGAIN, and do_wait() will restart the loop from the
1232                  * beginning. Do _not_ re-acquire the lock.
1233                  */
1234                 return -EAGAIN;
1235         }
1236
1237         /* move to end of parent's list to avoid starvation */
1238         remove_parent(p);
1239         add_parent(p, p->parent);
1240
1241         write_unlock_irq(&tasklist_lock);
1242
1243         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1244         if (!retval && stat_addr)
1245                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1246         if (!retval && infop)
1247                 retval = put_user(SIGCHLD, &infop->si_signo);
1248         if (!retval && infop)
1249                 retval = put_user(0, &infop->si_errno);
1250         if (!retval && infop)
1251                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1252                                           ? CLD_TRAPPED : CLD_STOPPED),
1253                                   &infop->si_code);
1254         if (!retval && infop)
1255                 retval = put_user(exit_code, &infop->si_status);
1256         if (!retval && infop)
1257                 retval = put_user(p->pid, &infop->si_pid);
1258         if (!retval && infop)
1259                 retval = put_user(p->uid, &infop->si_uid);
1260         if (!retval)
1261                 retval = p->pid;
1262         put_task_struct(p);
1263
1264         BUG_ON(!retval);
1265         return retval;
1266 }
1267
1268 /*
1269  * Handle do_wait work for one task in a live, non-stopped state.
1270  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1271  * the lock and this task is uninteresting.  If we return nonzero, we have
1272  * released the lock and the system call should return.
1273  */
1274 static int wait_task_continued(task_t *p, int noreap,
1275                                struct siginfo __user *infop,
1276                                int __user *stat_addr, struct rusage __user *ru)
1277 {
1278         int retval;
1279         pid_t pid;
1280         uid_t uid;
1281
1282         if (unlikely(!p->signal))
1283                 return 0;
1284
1285         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1286                 return 0;
1287
1288         spin_lock_irq(&p->sighand->siglock);
1289         /* Re-check with the lock held.  */
1290         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1291                 spin_unlock_irq(&p->sighand->siglock);
1292                 return 0;
1293         }
1294         if (!noreap)
1295                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1296         spin_unlock_irq(&p->sighand->siglock);
1297
1298         pid = p->pid;
1299         uid = p->uid;
1300         get_task_struct(p);
1301         read_unlock(&tasklist_lock);
1302
1303         if (!infop) {
1304                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1305                 put_task_struct(p);
1306                 if (!retval && stat_addr)
1307                         retval = put_user(0xffff, stat_addr);
1308                 if (!retval)
1309                         retval = p->pid;
1310         } else {
1311                 retval = wait_noreap_copyout(p, pid, uid,
1312                                              CLD_CONTINUED, SIGCONT,
1313                                              infop, ru);
1314                 BUG_ON(retval == 0);
1315         }
1316
1317         return retval;
1318 }
1319
1320
1321 static inline int my_ptrace_child(struct task_struct *p)
1322 {
1323         if (!(p->ptrace & PT_PTRACED))
1324                 return 0;
1325         if (!(p->ptrace & PT_ATTACHED))
1326                 return 1;
1327         /*
1328          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1329          * we are the attacher.  If we are the real parent, this is a race
1330          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1331          * which we have to switch the parent links, but has already set
1332          * the flags in p->ptrace.
1333          */
1334         return (p->parent != p->real_parent);
1335 }
1336
1337 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1338                     int __user *stat_addr, struct rusage __user *ru)
1339 {
1340         DECLARE_WAITQUEUE(wait, current);
1341         struct task_struct *tsk;
1342         int flag, retval;
1343
1344         add_wait_queue(&current->signal->wait_chldexit,&wait);
1345 repeat:
1346         /*
1347          * We will set this flag if we see any child that might later
1348          * match our criteria, even if we are not able to reap it yet.
1349          */
1350         flag = 0;
1351         current->state = TASK_INTERRUPTIBLE;
1352         read_lock(&tasklist_lock);
1353         tsk = current;
1354         do {
1355                 struct task_struct *p;
1356                 struct list_head *_p;
1357                 int ret;
1358
1359                 list_for_each(_p,&tsk->children) {
1360                         p = list_entry(_p,struct task_struct,sibling);
1361
1362                         ret = eligible_child(pid, options, p);
1363                         if (!ret)
1364                                 continue;
1365
1366                         switch (p->state) {
1367                         case TASK_TRACED:
1368                                 if (!my_ptrace_child(p))
1369                                         continue;
1370                                 /*FALLTHROUGH*/
1371                         case TASK_STOPPED:
1372                                 /*
1373                                  * It's stopped now, so it might later
1374                                  * continue, exit, or stop again.
1375                                  */
1376                                 flag = 1;
1377                                 if (!(options & WUNTRACED) &&
1378                                     !my_ptrace_child(p))
1379                                         continue;
1380                                 retval = wait_task_stopped(p, ret == 2,
1381                                                            (options & WNOWAIT),
1382                                                            infop,
1383                                                            stat_addr, ru);
1384                                 if (retval == -EAGAIN)
1385                                         goto repeat;
1386                                 if (retval != 0) /* He released the lock.  */
1387                                         goto end;
1388                                 break;
1389                         default:
1390                         // case EXIT_DEAD:
1391                                 if (p->exit_state == EXIT_DEAD)
1392                                         continue;
1393                         // case EXIT_ZOMBIE:
1394                                 if (p->exit_state == EXIT_ZOMBIE) {
1395                                         /*
1396                                          * Eligible but we cannot release
1397                                          * it yet:
1398                                          */
1399                                         if (ret == 2)
1400                                                 goto check_continued;
1401                                         if (!likely(options & WEXITED))
1402                                                 continue;
1403                                         retval = wait_task_zombie(
1404                                                 p, (options & WNOWAIT),
1405                                                 infop, stat_addr, ru);
1406                                         /* He released the lock.  */
1407                                         if (retval != 0)
1408                                                 goto end;
1409                                         break;
1410                                 }
1411 check_continued:
1412                                 /*
1413                                  * It's running now, so it might later
1414                                  * exit, stop, or stop and then continue.
1415                                  */
1416                                 flag = 1;
1417                                 if (!unlikely(options & WCONTINUED))
1418                                         continue;
1419                                 retval = wait_task_continued(
1420                                         p, (options & WNOWAIT),
1421                                         infop, stat_addr, ru);
1422                                 if (retval != 0) /* He released the lock.  */
1423                                         goto end;
1424                                 break;
1425                         }
1426                 }
1427                 if (!flag) {
1428                         list_for_each(_p, &tsk->ptrace_children) {
1429                                 p = list_entry(_p, struct task_struct,
1430                                                 ptrace_list);
1431                                 if (!eligible_child(pid, options, p))
1432                                         continue;
1433                                 flag = 1;
1434                                 break;
1435                         }
1436                 }
1437                 if (options & __WNOTHREAD)
1438                         break;
1439                 tsk = next_thread(tsk);
1440                 if (tsk->signal != current->signal)
1441                         BUG();
1442         } while (tsk != current);
1443
1444         read_unlock(&tasklist_lock);
1445         if (flag) {
1446                 retval = 0;
1447                 if (options & WNOHANG)
1448                         goto end;
1449                 retval = -ERESTARTSYS;
1450                 if (signal_pending(current))
1451                         goto end;
1452                 schedule();
1453                 goto repeat;
1454         }
1455         retval = -ECHILD;
1456 end:
1457         current->state = TASK_RUNNING;
1458         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1459         if (infop) {
1460                 if (retval > 0)
1461                 retval = 0;
1462                 else {
1463                         /*
1464                          * For a WNOHANG return, clear out all the fields
1465                          * we would set so the user can easily tell the
1466                          * difference.
1467                          */
1468                         if (!retval)
1469                                 retval = put_user(0, &infop->si_signo);
1470                         if (!retval)
1471                                 retval = put_user(0, &infop->si_errno);
1472                         if (!retval)
1473                                 retval = put_user(0, &infop->si_code);
1474                         if (!retval)
1475                                 retval = put_user(0, &infop->si_pid);
1476                         if (!retval)
1477                                 retval = put_user(0, &infop->si_uid);
1478                         if (!retval)
1479                                 retval = put_user(0, &infop->si_status);
1480                 }
1481         }
1482         return retval;
1483 }
1484
1485 asmlinkage long sys_waitid(int which, pid_t pid,
1486                            struct siginfo __user *infop, int options,
1487                            struct rusage __user *ru)
1488 {
1489         long ret;
1490
1491         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1492                 return -EINVAL;
1493         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1494                 return -EINVAL;
1495
1496         switch (which) {
1497         case P_ALL:
1498                 pid = -1;
1499                 break;
1500         case P_PID:
1501                 if (pid <= 0)
1502                         return -EINVAL;
1503                 break;
1504         case P_PGID:
1505                 if (pid <= 0)
1506                         return -EINVAL;
1507                 pid = -pid;
1508                 break;
1509         default:
1510                 return -EINVAL;
1511         }
1512
1513         ret = do_wait(pid, options, infop, NULL, ru);
1514
1515         /* avoid REGPARM breakage on x86: */
1516         prevent_tail_call(ret);
1517         return ret;
1518 }
1519
1520 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1521                           int options, struct rusage __user *ru)
1522 {
1523         long ret;
1524
1525         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1526                         __WNOTHREAD|__WCLONE|__WALL))
1527                 return -EINVAL;
1528         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1529
1530         /* avoid REGPARM breakage on x86: */
1531         prevent_tail_call(ret);
1532         return ret;
1533 }
1534
1535 #ifdef __ARCH_WANT_SYS_WAITPID
1536
1537 /*
1538  * sys_waitpid() remains for compatibility. waitpid() should be
1539  * implemented by calling sys_wait4() from libc.a.
1540  */
1541 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1542 {
1543         return sys_wait4(pid, stat_addr, options, NULL);
1544 }
1545
1546 #endif