Merge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6
[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/capability.h>
14 #include <linux/completion.h>
15 #include <linux/personality.h>
16 #include <linux/tty.h>
17 #include <linux/namespace.h>
18 #include <linux/key.h>
19 #include <linux/security.h>
20 #include <linux/cpu.h>
21 #include <linux/acct.h>
22 #include <linux/file.h>
23 #include <linux/binfmts.h>
24 #include <linux/ptrace.h>
25 #include <linux/profile.h>
26 #include <linux/mount.h>
27 #include <linux/proc_fs.h>
28 #include <linux/mempolicy.h>
29 #include <linux/cpuset.h>
30 #include <linux/syscalls.h>
31 #include <linux/signal.h>
32 #include <linux/posix-timers.h>
33 #include <linux/cn_proc.h>
34 #include <linux/mutex.h>
35 #include <linux/futex.h>
36 #include <linux/compat.h>
37
38 #include <asm/uaccess.h>
39 #include <asm/unistd.h>
40 #include <asm/pgtable.h>
41 #include <asm/mmu_context.h>
42
43 extern void sem_exit (void);
44 extern struct task_struct *child_reaper;
45
46 int getrusage(struct task_struct *, int, struct rusage __user *);
47
48 static void exit_mm(struct task_struct * tsk);
49
50 static void __unhash_process(struct task_struct *p)
51 {
52         nr_threads--;
53         detach_pid(p, PIDTYPE_PID);
54         if (thread_group_leader(p)) {
55                 detach_pid(p, PIDTYPE_PGID);
56                 detach_pid(p, PIDTYPE_SID);
57
58                 list_del_init(&p->tasks);
59                 __get_cpu_var(process_counts)--;
60         }
61         list_del_rcu(&p->thread_group);
62         remove_parent(p);
63 }
64
65 /*
66  * This function expects the tasklist_lock write-locked.
67  */
68 static void __exit_signal(struct task_struct *tsk)
69 {
70         struct signal_struct *sig = tsk->signal;
71         struct sighand_struct *sighand;
72
73         BUG_ON(!sig);
74         BUG_ON(!atomic_read(&sig->count));
75
76         rcu_read_lock();
77         sighand = rcu_dereference(tsk->sighand);
78         spin_lock(&sighand->siglock);
79
80         posix_cpu_timers_exit(tsk);
81         if (atomic_dec_and_test(&sig->count))
82                 posix_cpu_timers_exit_group(tsk);
83         else {
84                 /*
85                  * If there is any task waiting for the group exit
86                  * then notify it:
87                  */
88                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
89                         wake_up_process(sig->group_exit_task);
90                         sig->group_exit_task = NULL;
91                 }
92                 if (tsk == sig->curr_target)
93                         sig->curr_target = next_thread(tsk);
94                 /*
95                  * Accumulate here the counters for all threads but the
96                  * group leader as they die, so they can be added into
97                  * the process-wide totals when those are taken.
98                  * The group leader stays around as a zombie as long
99                  * as there are other threads.  When it gets reaped,
100                  * the exit.c code will add its counts into these totals.
101                  * We won't ever get here for the group leader, since it
102                  * will have been the last reference on the signal_struct.
103                  */
104                 sig->utime = cputime_add(sig->utime, tsk->utime);
105                 sig->stime = cputime_add(sig->stime, tsk->stime);
106                 sig->min_flt += tsk->min_flt;
107                 sig->maj_flt += tsk->maj_flt;
108                 sig->nvcsw += tsk->nvcsw;
109                 sig->nivcsw += tsk->nivcsw;
110                 sig->sched_time += tsk->sched_time;
111                 sig = NULL; /* Marker for below. */
112         }
113
114         __unhash_process(tsk);
115
116         tsk->signal = NULL;
117         tsk->sighand = NULL;
118         spin_unlock(&sighand->siglock);
119         rcu_read_unlock();
120
121         __cleanup_sighand(sighand);
122         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
123         flush_sigqueue(&tsk->pending);
124         if (sig) {
125                 flush_sigqueue(&sig->shared_pending);
126                 __cleanup_signal(sig);
127         }
128 }
129
130 void release_task(struct task_struct * p)
131 {
132         int zap_leader;
133         task_t *leader;
134         struct dentry *proc_dentry;
135
136 repeat:
137         atomic_dec(&p->user->processes);
138         spin_lock(&p->proc_lock);
139         proc_dentry = proc_pid_unhash(p);
140         write_lock_irq(&tasklist_lock);
141         ptrace_unlink(p);
142         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
143         __exit_signal(p);
144
145         /*
146          * If we are the last non-leader member of the thread
147          * group, and the leader is zombie, then notify the
148          * group leader's parent process. (if it wants notification.)
149          */
150         zap_leader = 0;
151         leader = p->group_leader;
152         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
153                 BUG_ON(leader->exit_signal == -1);
154                 do_notify_parent(leader, leader->exit_signal);
155                 /*
156                  * If we were the last child thread and the leader has
157                  * exited already, and the leader's parent ignores SIGCHLD,
158                  * then we are the one who should release the leader.
159                  *
160                  * do_notify_parent() will have marked it self-reaping in
161                  * that case.
162                  */
163                 zap_leader = (leader->exit_signal == -1);
164         }
165
166         sched_exit(p);
167         write_unlock_irq(&tasklist_lock);
168         spin_unlock(&p->proc_lock);
169         proc_pid_flush(proc_dentry);
170         release_thread(p);
171         put_task_struct(p);
172
173         p = leader;
174         if (unlikely(zap_leader))
175                 goto repeat;
176 }
177
178 /*
179  * This checks not only the pgrp, but falls back on the pid if no
180  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
181  * without this...
182  */
183 int session_of_pgrp(int pgrp)
184 {
185         struct task_struct *p;
186         int sid = -1;
187
188         read_lock(&tasklist_lock);
189         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
190                 if (p->signal->session > 0) {
191                         sid = p->signal->session;
192                         goto out;
193                 }
194         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
195         p = find_task_by_pid(pgrp);
196         if (p)
197                 sid = p->signal->session;
198 out:
199         read_unlock(&tasklist_lock);
200         
201         return sid;
202 }
203
204 /*
205  * Determine if a process group is "orphaned", according to the POSIX
206  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
207  * by terminal-generated stop signals.  Newly orphaned process groups are
208  * to receive a SIGHUP and a SIGCONT.
209  *
210  * "I ask you, have you ever known what it is to be an orphan?"
211  */
212 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
213 {
214         struct task_struct *p;
215         int ret = 1;
216
217         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
218                 if (p == ignored_task
219                                 || p->exit_state
220                                 || p->real_parent->pid == 1)
221                         continue;
222                 if (process_group(p->real_parent) != pgrp
223                             && p->real_parent->signal->session == p->signal->session) {
224                         ret = 0;
225                         break;
226                 }
227         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
228         return ret;     /* (sighing) "Often!" */
229 }
230
231 int is_orphaned_pgrp(int pgrp)
232 {
233         int retval;
234
235         read_lock(&tasklist_lock);
236         retval = will_become_orphaned_pgrp(pgrp, NULL);
237         read_unlock(&tasklist_lock);
238
239         return retval;
240 }
241
242 static int has_stopped_jobs(int pgrp)
243 {
244         int retval = 0;
245         struct task_struct *p;
246
247         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
248                 if (p->state != TASK_STOPPED)
249                         continue;
250
251                 /* If p is stopped by a debugger on a signal that won't
252                    stop it, then don't count p as stopped.  This isn't
253                    perfect but it's a good approximation.  */
254                 if (unlikely (p->ptrace)
255                     && p->exit_code != SIGSTOP
256                     && p->exit_code != SIGTSTP
257                     && p->exit_code != SIGTTOU
258                     && p->exit_code != SIGTTIN)
259                         continue;
260
261                 retval = 1;
262                 break;
263         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
264         return retval;
265 }
266
267 /**
268  * reparent_to_init - Reparent the calling kernel thread to the init task.
269  *
270  * If a kernel thread is launched as a result of a system call, or if
271  * it ever exits, it should generally reparent itself to init so that
272  * it is correctly cleaned up on exit.
273  *
274  * The various task state such as scheduling policy and priority may have
275  * been inherited from a user process, so we reset them to sane values here.
276  *
277  * NOTE that reparent_to_init() gives the caller full capabilities.
278  */
279 static void reparent_to_init(void)
280 {
281         write_lock_irq(&tasklist_lock);
282
283         ptrace_unlink(current);
284         /* Reparent to init */
285         remove_parent(current);
286         current->parent = child_reaper;
287         current->real_parent = child_reaper;
288         add_parent(current);
289
290         /* Set the exit signal to SIGCHLD so we signal init on exit */
291         current->exit_signal = SIGCHLD;
292
293         if ((current->policy == SCHED_NORMAL ||
294                         current->policy == SCHED_BATCH)
295                                 && (task_nice(current) < 0))
296                 set_user_nice(current, 0);
297         /* cpus_allowed? */
298         /* rt_priority? */
299         /* signals? */
300         security_task_reparent_to_init(current);
301         memcpy(current->signal->rlim, init_task.signal->rlim,
302                sizeof(current->signal->rlim));
303         atomic_inc(&(INIT_USER->__count));
304         write_unlock_irq(&tasklist_lock);
305         switch_uid(INIT_USER);
306 }
307
308 void __set_special_pids(pid_t session, pid_t pgrp)
309 {
310         struct task_struct *curr = current->group_leader;
311
312         if (curr->signal->session != session) {
313                 detach_pid(curr, PIDTYPE_SID);
314                 curr->signal->session = session;
315                 attach_pid(curr, PIDTYPE_SID, session);
316         }
317         if (process_group(curr) != pgrp) {
318                 detach_pid(curr, PIDTYPE_PGID);
319                 curr->signal->pgrp = pgrp;
320                 attach_pid(curr, PIDTYPE_PGID, pgrp);
321         }
322 }
323
324 void set_special_pids(pid_t session, pid_t pgrp)
325 {
326         write_lock_irq(&tasklist_lock);
327         __set_special_pids(session, pgrp);
328         write_unlock_irq(&tasklist_lock);
329 }
330
331 /*
332  * Let kernel threads use this to say that they
333  * allow a certain signal (since daemonize() will
334  * have disabled all of them by default).
335  */
336 int allow_signal(int sig)
337 {
338         if (!valid_signal(sig) || sig < 1)
339                 return -EINVAL;
340
341         spin_lock_irq(&current->sighand->siglock);
342         sigdelset(&current->blocked, sig);
343         if (!current->mm) {
344                 /* Kernel threads handle their own signals.
345                    Let the signal code know it'll be handled, so
346                    that they don't get converted to SIGKILL or
347                    just silently dropped */
348                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
349         }
350         recalc_sigpending();
351         spin_unlock_irq(&current->sighand->siglock);
352         return 0;
353 }
354
355 EXPORT_SYMBOL(allow_signal);
356
357 int disallow_signal(int sig)
358 {
359         if (!valid_signal(sig) || sig < 1)
360                 return -EINVAL;
361
362         spin_lock_irq(&current->sighand->siglock);
363         sigaddset(&current->blocked, sig);
364         recalc_sigpending();
365         spin_unlock_irq(&current->sighand->siglock);
366         return 0;
367 }
368
369 EXPORT_SYMBOL(disallow_signal);
370
371 /*
372  *      Put all the gunge required to become a kernel thread without
373  *      attached user resources in one place where it belongs.
374  */
375
376 void daemonize(const char *name, ...)
377 {
378         va_list args;
379         struct fs_struct *fs;
380         sigset_t blocked;
381
382         va_start(args, name);
383         vsnprintf(current->comm, sizeof(current->comm), name, args);
384         va_end(args);
385
386         /*
387          * If we were started as result of loading a module, close all of the
388          * user space pages.  We don't need them, and if we didn't close them
389          * they would be locked into memory.
390          */
391         exit_mm(current);
392
393         set_special_pids(1, 1);
394         mutex_lock(&tty_mutex);
395         current->signal->tty = NULL;
396         mutex_unlock(&tty_mutex);
397
398         /* Block and flush all signals */
399         sigfillset(&blocked);
400         sigprocmask(SIG_BLOCK, &blocked, NULL);
401         flush_signals(current);
402
403         /* Become as one with the init task */
404
405         exit_fs(current);       /* current->fs->count--; */
406         fs = init_task.fs;
407         current->fs = fs;
408         atomic_inc(&fs->count);
409         exit_namespace(current);
410         current->namespace = init_task.namespace;
411         get_namespace(current->namespace);
412         exit_files(current);
413         current->files = init_task.files;
414         atomic_inc(&current->files->count);
415
416         reparent_to_init();
417 }
418
419 EXPORT_SYMBOL(daemonize);
420
421 static void close_files(struct files_struct * files)
422 {
423         int i, j;
424         struct fdtable *fdt;
425
426         j = 0;
427
428         /*
429          * It is safe to dereference the fd table without RCU or
430          * ->file_lock because this is the last reference to the
431          * files structure.
432          */
433         fdt = files_fdtable(files);
434         for (;;) {
435                 unsigned long set;
436                 i = j * __NFDBITS;
437                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
438                         break;
439                 set = fdt->open_fds->fds_bits[j++];
440                 while (set) {
441                         if (set & 1) {
442                                 struct file * file = xchg(&fdt->fd[i], NULL);
443                                 if (file)
444                                         filp_close(file, files);
445                         }
446                         i++;
447                         set >>= 1;
448                 }
449         }
450 }
451
452 struct files_struct *get_files_struct(struct task_struct *task)
453 {
454         struct files_struct *files;
455
456         task_lock(task);
457         files = task->files;
458         if (files)
459                 atomic_inc(&files->count);
460         task_unlock(task);
461
462         return files;
463 }
464
465 void fastcall put_files_struct(struct files_struct *files)
466 {
467         struct fdtable *fdt;
468
469         if (atomic_dec_and_test(&files->count)) {
470                 close_files(files);
471                 /*
472                  * Free the fd and fdset arrays if we expanded them.
473                  * If the fdtable was embedded, pass files for freeing
474                  * at the end of the RCU grace period. Otherwise,
475                  * you can free files immediately.
476                  */
477                 fdt = files_fdtable(files);
478                 if (fdt == &files->fdtab)
479                         fdt->free_files = files;
480                 else
481                         kmem_cache_free(files_cachep, files);
482                 free_fdtable(fdt);
483         }
484 }
485
486 EXPORT_SYMBOL(put_files_struct);
487
488 static inline void __exit_files(struct task_struct *tsk)
489 {
490         struct files_struct * files = tsk->files;
491
492         if (files) {
493                 task_lock(tsk);
494                 tsk->files = NULL;
495                 task_unlock(tsk);
496                 put_files_struct(files);
497         }
498 }
499
500 void exit_files(struct task_struct *tsk)
501 {
502         __exit_files(tsk);
503 }
504
505 static inline void __put_fs_struct(struct fs_struct *fs)
506 {
507         /* No need to hold fs->lock if we are killing it */
508         if (atomic_dec_and_test(&fs->count)) {
509                 dput(fs->root);
510                 mntput(fs->rootmnt);
511                 dput(fs->pwd);
512                 mntput(fs->pwdmnt);
513                 if (fs->altroot) {
514                         dput(fs->altroot);
515                         mntput(fs->altrootmnt);
516                 }
517                 kmem_cache_free(fs_cachep, fs);
518         }
519 }
520
521 void put_fs_struct(struct fs_struct *fs)
522 {
523         __put_fs_struct(fs);
524 }
525
526 static inline void __exit_fs(struct task_struct *tsk)
527 {
528         struct fs_struct * fs = tsk->fs;
529
530         if (fs) {
531                 task_lock(tsk);
532                 tsk->fs = NULL;
533                 task_unlock(tsk);
534                 __put_fs_struct(fs);
535         }
536 }
537
538 void exit_fs(struct task_struct *tsk)
539 {
540         __exit_fs(tsk);
541 }
542
543 EXPORT_SYMBOL_GPL(exit_fs);
544
545 /*
546  * Turn us into a lazy TLB process if we
547  * aren't already..
548  */
549 static void exit_mm(struct task_struct * tsk)
550 {
551         struct mm_struct *mm = tsk->mm;
552
553         mm_release(tsk, mm);
554         if (!mm)
555                 return;
556         /*
557          * Serialize with any possible pending coredump.
558          * We must hold mmap_sem around checking core_waiters
559          * and clearing tsk->mm.  The core-inducing thread
560          * will increment core_waiters for each thread in the
561          * group with ->mm != NULL.
562          */
563         down_read(&mm->mmap_sem);
564         if (mm->core_waiters) {
565                 up_read(&mm->mmap_sem);
566                 down_write(&mm->mmap_sem);
567                 if (!--mm->core_waiters)
568                         complete(mm->core_startup_done);
569                 up_write(&mm->mmap_sem);
570
571                 wait_for_completion(&mm->core_done);
572                 down_read(&mm->mmap_sem);
573         }
574         atomic_inc(&mm->mm_count);
575         if (mm != tsk->active_mm) BUG();
576         /* more a memory barrier than a real lock */
577         task_lock(tsk);
578         tsk->mm = NULL;
579         up_read(&mm->mmap_sem);
580         enter_lazy_tlb(mm, current);
581         task_unlock(tsk);
582         mmput(mm);
583 }
584
585 static inline void choose_new_parent(task_t *p, task_t *reaper)
586 {
587         /*
588          * Make sure we're not reparenting to ourselves and that
589          * the parent is not a zombie.
590          */
591         BUG_ON(p == reaper || reaper->exit_state);
592         p->real_parent = reaper;
593 }
594
595 static void reparent_thread(task_t *p, task_t *father, int traced)
596 {
597         /* We don't want people slaying init.  */
598         if (p->exit_signal != -1)
599                 p->exit_signal = SIGCHLD;
600
601         if (p->pdeath_signal)
602                 /* We already hold the tasklist_lock here.  */
603                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
604
605         /* Move the child from its dying parent to the new one.  */
606         if (unlikely(traced)) {
607                 /* Preserve ptrace links if someone else is tracing this child.  */
608                 list_del_init(&p->ptrace_list);
609                 if (p->parent != p->real_parent)
610                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
611         } else {
612                 /* If this child is being traced, then we're the one tracing it
613                  * anyway, so let go of it.
614                  */
615                 p->ptrace = 0;
616                 remove_parent(p);
617                 p->parent = p->real_parent;
618                 add_parent(p);
619
620                 /* If we'd notified the old parent about this child's death,
621                  * also notify the new parent.
622                  */
623                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
624                     thread_group_empty(p))
625                         do_notify_parent(p, p->exit_signal);
626                 else if (p->state == TASK_TRACED) {
627                         /*
628                          * If it was at a trace stop, turn it into
629                          * a normal stop since it's no longer being
630                          * traced.
631                          */
632                         ptrace_untrace(p);
633                 }
634         }
635
636         /*
637          * process group orphan check
638          * Case ii: Our child is in a different pgrp
639          * than we are, and it was the only connection
640          * outside, so the child pgrp is now orphaned.
641          */
642         if ((process_group(p) != process_group(father)) &&
643             (p->signal->session == father->signal->session)) {
644                 int pgrp = process_group(p);
645
646                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
647                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
648                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
649                 }
650         }
651 }
652
653 /*
654  * When we die, we re-parent all our children.
655  * Try to give them to another thread in our thread
656  * group, and if no such member exists, give it to
657  * the global child reaper process (ie "init")
658  */
659 static void forget_original_parent(struct task_struct * father,
660                                           struct list_head *to_release)
661 {
662         struct task_struct *p, *reaper = father;
663         struct list_head *_p, *_n;
664
665         do {
666                 reaper = next_thread(reaper);
667                 if (reaper == father) {
668                         reaper = child_reaper;
669                         break;
670                 }
671         } while (reaper->exit_state);
672
673         /*
674          * There are only two places where our children can be:
675          *
676          * - in our child list
677          * - in our ptraced child list
678          *
679          * Search them and reparent children.
680          */
681         list_for_each_safe(_p, _n, &father->children) {
682                 int ptrace;
683                 p = list_entry(_p,struct task_struct,sibling);
684
685                 ptrace = p->ptrace;
686
687                 /* if father isn't the real parent, then ptrace must be enabled */
688                 BUG_ON(father != p->real_parent && !ptrace);
689
690                 if (father == p->real_parent) {
691                         /* reparent with a reaper, real father it's us */
692                         choose_new_parent(p, reaper);
693                         reparent_thread(p, father, 0);
694                 } else {
695                         /* reparent ptraced task to its real parent */
696                         __ptrace_unlink (p);
697                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
698                             thread_group_empty(p))
699                                 do_notify_parent(p, p->exit_signal);
700                 }
701
702                 /*
703                  * if the ptraced child is a zombie with exit_signal == -1
704                  * we must collect it before we exit, or it will remain
705                  * zombie forever since we prevented it from self-reap itself
706                  * while it was being traced by us, to be able to see it in wait4.
707                  */
708                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
709                         list_add(&p->ptrace_list, to_release);
710         }
711         list_for_each_safe(_p, _n, &father->ptrace_children) {
712                 p = list_entry(_p,struct task_struct,ptrace_list);
713                 choose_new_parent(p, reaper);
714                 reparent_thread(p, father, 1);
715         }
716 }
717
718 /*
719  * Send signals to all our closest relatives so that they know
720  * to properly mourn us..
721  */
722 static void exit_notify(struct task_struct *tsk)
723 {
724         int state;
725         struct task_struct *t;
726         struct list_head ptrace_dead, *_p, *_n;
727
728         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
729             && !thread_group_empty(tsk)) {
730                 /*
731                  * This occurs when there was a race between our exit
732                  * syscall and a group signal choosing us as the one to
733                  * wake up.  It could be that we are the only thread
734                  * alerted to check for pending signals, but another thread
735                  * should be woken now to take the signal since we will not.
736                  * Now we'll wake all the threads in the group just to make
737                  * sure someone gets all the pending signals.
738                  */
739                 read_lock(&tasklist_lock);
740                 spin_lock_irq(&tsk->sighand->siglock);
741                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
742                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
743                                 recalc_sigpending_tsk(t);
744                                 if (signal_pending(t))
745                                         signal_wake_up(t, 0);
746                         }
747                 spin_unlock_irq(&tsk->sighand->siglock);
748                 read_unlock(&tasklist_lock);
749         }
750
751         write_lock_irq(&tasklist_lock);
752
753         /*
754          * This does two things:
755          *
756          * A.  Make init inherit all the child processes
757          * B.  Check to see if any process groups have become orphaned
758          *      as a result of our exiting, and if they have any stopped
759          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
760          */
761
762         INIT_LIST_HEAD(&ptrace_dead);
763         forget_original_parent(tsk, &ptrace_dead);
764         BUG_ON(!list_empty(&tsk->children));
765         BUG_ON(!list_empty(&tsk->ptrace_children));
766
767         /*
768          * Check to see if any process groups have become orphaned
769          * as a result of our exiting, and if they have any stopped
770          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
771          *
772          * Case i: Our father is in a different pgrp than we are
773          * and we were the only connection outside, so our pgrp
774          * is about to become orphaned.
775          */
776          
777         t = tsk->real_parent;
778         
779         if ((process_group(t) != process_group(tsk)) &&
780             (t->signal->session == tsk->signal->session) &&
781             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
782             has_stopped_jobs(process_group(tsk))) {
783                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
784                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
785         }
786
787         /* Let father know we died 
788          *
789          * Thread signals are configurable, but you aren't going to use
790          * that to send signals to arbitary processes. 
791          * That stops right now.
792          *
793          * If the parent exec id doesn't match the exec id we saved
794          * when we started then we know the parent has changed security
795          * domain.
796          *
797          * If our self_exec id doesn't match our parent_exec_id then
798          * we have changed execution domain as these two values started
799          * the same after a fork.
800          *      
801          */
802         
803         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
804             ( tsk->parent_exec_id != t->self_exec_id  ||
805               tsk->self_exec_id != tsk->parent_exec_id)
806             && !capable(CAP_KILL))
807                 tsk->exit_signal = SIGCHLD;
808
809
810         /* If something other than our normal parent is ptracing us, then
811          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
812          * only has special meaning to our real parent.
813          */
814         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
815                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
816                 do_notify_parent(tsk, signal);
817         } else if (tsk->ptrace) {
818                 do_notify_parent(tsk, SIGCHLD);
819         }
820
821         state = EXIT_ZOMBIE;
822         if (tsk->exit_signal == -1 &&
823             (likely(tsk->ptrace == 0) ||
824              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
825                 state = EXIT_DEAD;
826         tsk->exit_state = state;
827
828         write_unlock_irq(&tasklist_lock);
829
830         list_for_each_safe(_p, _n, &ptrace_dead) {
831                 list_del_init(_p);
832                 t = list_entry(_p,struct task_struct,ptrace_list);
833                 release_task(t);
834         }
835
836         /* If the process is dead, release it - nobody will wait for it */
837         if (state == EXIT_DEAD)
838                 release_task(tsk);
839 }
840
841 fastcall NORET_TYPE void do_exit(long code)
842 {
843         struct task_struct *tsk = current;
844         int group_dead;
845
846         profile_task_exit(tsk);
847
848         WARN_ON(atomic_read(&tsk->fs_excl));
849
850         if (unlikely(in_interrupt()))
851                 panic("Aiee, killing interrupt handler!");
852         if (unlikely(!tsk->pid))
853                 panic("Attempted to kill the idle task!");
854         if (unlikely(tsk == child_reaper))
855                 panic("Attempted to kill init!");
856
857         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
858                 current->ptrace_message = code;
859                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
860         }
861
862         /*
863          * We're taking recursive faults here in do_exit. Safest is to just
864          * leave this task alone and wait for reboot.
865          */
866         if (unlikely(tsk->flags & PF_EXITING)) {
867                 printk(KERN_ALERT
868                         "Fixing recursive fault but reboot is needed!\n");
869                 if (tsk->io_context)
870                         exit_io_context();
871                 set_current_state(TASK_UNINTERRUPTIBLE);
872                 schedule();
873         }
874
875         tsk->flags |= PF_EXITING;
876
877         /*
878          * Make sure we don't try to process any timer firings
879          * while we are already exiting.
880          */
881         tsk->it_virt_expires = cputime_zero;
882         tsk->it_prof_expires = cputime_zero;
883         tsk->it_sched_expires = 0;
884
885         if (unlikely(in_atomic()))
886                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
887                                 current->comm, current->pid,
888                                 preempt_count());
889
890         acct_update_integrals(tsk);
891         if (tsk->mm) {
892                 update_hiwater_rss(tsk->mm);
893                 update_hiwater_vm(tsk->mm);
894         }
895         group_dead = atomic_dec_and_test(&tsk->signal->live);
896         if (group_dead) {
897                 hrtimer_cancel(&tsk->signal->real_timer);
898                 exit_itimers(tsk->signal);
899                 acct_process(code);
900         }
901         if (unlikely(tsk->robust_list))
902                 exit_robust_list(tsk);
903 #ifdef CONFIG_COMPAT
904         if (unlikely(tsk->compat_robust_list))
905                 compat_exit_robust_list(tsk);
906 #endif
907         exit_mm(tsk);
908
909         exit_sem(tsk);
910         __exit_files(tsk);
911         __exit_fs(tsk);
912         exit_namespace(tsk);
913         exit_thread();
914         cpuset_exit(tsk);
915         exit_keys(tsk);
916
917         if (group_dead && tsk->signal->leader)
918                 disassociate_ctty(1);
919
920         module_put(task_thread_info(tsk)->exec_domain->module);
921         if (tsk->binfmt)
922                 module_put(tsk->binfmt->module);
923
924         tsk->exit_code = code;
925         proc_exit_connector(tsk);
926         exit_notify(tsk);
927 #ifdef CONFIG_NUMA
928         mpol_free(tsk->mempolicy);
929         tsk->mempolicy = NULL;
930 #endif
931         /*
932          * If DEBUG_MUTEXES is on, make sure we are holding no locks:
933          */
934         mutex_debug_check_no_locks_held(tsk);
935
936         if (tsk->io_context)
937                 exit_io_context();
938
939         /* PF_DEAD causes final put_task_struct after we schedule. */
940         preempt_disable();
941         BUG_ON(tsk->flags & PF_DEAD);
942         tsk->flags |= PF_DEAD;
943
944         schedule();
945         BUG();
946         /* Avoid "noreturn function does return".  */
947         for (;;) ;
948 }
949
950 EXPORT_SYMBOL_GPL(do_exit);
951
952 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
953 {
954         if (comp)
955                 complete(comp);
956         
957         do_exit(code);
958 }
959
960 EXPORT_SYMBOL(complete_and_exit);
961
962 asmlinkage long sys_exit(int error_code)
963 {
964         do_exit((error_code&0xff)<<8);
965 }
966
967 /*
968  * Take down every thread in the group.  This is called by fatal signals
969  * as well as by sys_exit_group (below).
970  */
971 NORET_TYPE void
972 do_group_exit(int exit_code)
973 {
974         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
975
976         if (current->signal->flags & SIGNAL_GROUP_EXIT)
977                 exit_code = current->signal->group_exit_code;
978         else if (!thread_group_empty(current)) {
979                 struct signal_struct *const sig = current->signal;
980                 struct sighand_struct *const sighand = current->sighand;
981                 spin_lock_irq(&sighand->siglock);
982                 if (sig->flags & SIGNAL_GROUP_EXIT)
983                         /* Another thread got here before we took the lock.  */
984                         exit_code = sig->group_exit_code;
985                 else {
986                         sig->group_exit_code = exit_code;
987                         zap_other_threads(current);
988                 }
989                 spin_unlock_irq(&sighand->siglock);
990         }
991
992         do_exit(exit_code);
993         /* NOTREACHED */
994 }
995
996 /*
997  * this kills every thread in the thread group. Note that any externally
998  * wait4()-ing process will get the correct exit code - even if this
999  * thread is not the thread group leader.
1000  */
1001 asmlinkage void sys_exit_group(int error_code)
1002 {
1003         do_group_exit((error_code & 0xff) << 8);
1004 }
1005
1006 static int eligible_child(pid_t pid, int options, task_t *p)
1007 {
1008         if (pid > 0) {
1009                 if (p->pid != pid)
1010                         return 0;
1011         } else if (!pid) {
1012                 if (process_group(p) != process_group(current))
1013                         return 0;
1014         } else if (pid != -1) {
1015                 if (process_group(p) != -pid)
1016                         return 0;
1017         }
1018
1019         /*
1020          * Do not consider detached threads that are
1021          * not ptraced:
1022          */
1023         if (p->exit_signal == -1 && !p->ptrace)
1024                 return 0;
1025
1026         /* Wait for all children (clone and not) if __WALL is set;
1027          * otherwise, wait for clone children *only* if __WCLONE is
1028          * set; otherwise, wait for non-clone children *only*.  (Note:
1029          * A "clone" child here is one that reports to its parent
1030          * using a signal other than SIGCHLD.) */
1031         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1032             && !(options & __WALL))
1033                 return 0;
1034         /*
1035          * Do not consider thread group leaders that are
1036          * in a non-empty thread group:
1037          */
1038         if (current->tgid != p->tgid && delay_group_leader(p))
1039                 return 2;
1040
1041         if (security_task_wait(p))
1042                 return 0;
1043
1044         return 1;
1045 }
1046
1047 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
1048                                int why, int status,
1049                                struct siginfo __user *infop,
1050                                struct rusage __user *rusagep)
1051 {
1052         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1053         put_task_struct(p);
1054         if (!retval)
1055                 retval = put_user(SIGCHLD, &infop->si_signo);
1056         if (!retval)
1057                 retval = put_user(0, &infop->si_errno);
1058         if (!retval)
1059                 retval = put_user((short)why, &infop->si_code);
1060         if (!retval)
1061                 retval = put_user(pid, &infop->si_pid);
1062         if (!retval)
1063                 retval = put_user(uid, &infop->si_uid);
1064         if (!retval)
1065                 retval = put_user(status, &infop->si_status);
1066         if (!retval)
1067                 retval = pid;
1068         return retval;
1069 }
1070
1071 /*
1072  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1073  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1074  * the lock and this task is uninteresting.  If we return nonzero, we have
1075  * released the lock and the system call should return.
1076  */
1077 static int wait_task_zombie(task_t *p, int noreap,
1078                             struct siginfo __user *infop,
1079                             int __user *stat_addr, struct rusage __user *ru)
1080 {
1081         unsigned long state;
1082         int retval;
1083         int status;
1084
1085         if (unlikely(noreap)) {
1086                 pid_t pid = p->pid;
1087                 uid_t uid = p->uid;
1088                 int exit_code = p->exit_code;
1089                 int why, status;
1090
1091                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1092                         return 0;
1093                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1094                         return 0;
1095                 get_task_struct(p);
1096                 read_unlock(&tasklist_lock);
1097                 if ((exit_code & 0x7f) == 0) {
1098                         why = CLD_EXITED;
1099                         status = exit_code >> 8;
1100                 } else {
1101                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1102                         status = exit_code & 0x7f;
1103                 }
1104                 return wait_noreap_copyout(p, pid, uid, why,
1105                                            status, infop, ru);
1106         }
1107
1108         /*
1109          * Try to move the task's state to DEAD
1110          * only one thread is allowed to do this:
1111          */
1112         state = xchg(&p->exit_state, EXIT_DEAD);
1113         if (state != EXIT_ZOMBIE) {
1114                 BUG_ON(state != EXIT_DEAD);
1115                 return 0;
1116         }
1117         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1118                 /*
1119                  * This can only happen in a race with a ptraced thread
1120                  * dying on another processor.
1121                  */
1122                 return 0;
1123         }
1124
1125         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1126                 struct signal_struct *psig;
1127                 struct signal_struct *sig;
1128
1129                 /*
1130                  * The resource counters for the group leader are in its
1131                  * own task_struct.  Those for dead threads in the group
1132                  * are in its signal_struct, as are those for the child
1133                  * processes it has previously reaped.  All these
1134                  * accumulate in the parent's signal_struct c* fields.
1135                  *
1136                  * We don't bother to take a lock here to protect these
1137                  * p->signal fields, because they are only touched by
1138                  * __exit_signal, which runs with tasklist_lock
1139                  * write-locked anyway, and so is excluded here.  We do
1140                  * need to protect the access to p->parent->signal fields,
1141                  * as other threads in the parent group can be right
1142                  * here reaping other children at the same time.
1143                  */
1144                 spin_lock_irq(&p->parent->sighand->siglock);
1145                 psig = p->parent->signal;
1146                 sig = p->signal;
1147                 psig->cutime =
1148                         cputime_add(psig->cutime,
1149                         cputime_add(p->utime,
1150                         cputime_add(sig->utime,
1151                                     sig->cutime)));
1152                 psig->cstime =
1153                         cputime_add(psig->cstime,
1154                         cputime_add(p->stime,
1155                         cputime_add(sig->stime,
1156                                     sig->cstime)));
1157                 psig->cmin_flt +=
1158                         p->min_flt + sig->min_flt + sig->cmin_flt;
1159                 psig->cmaj_flt +=
1160                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1161                 psig->cnvcsw +=
1162                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1163                 psig->cnivcsw +=
1164                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1165                 spin_unlock_irq(&p->parent->sighand->siglock);
1166         }
1167
1168         /*
1169          * Now we are sure this task is interesting, and no other
1170          * thread can reap it because we set its state to EXIT_DEAD.
1171          */
1172         read_unlock(&tasklist_lock);
1173
1174         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1175         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1176                 ? p->signal->group_exit_code : p->exit_code;
1177         if (!retval && stat_addr)
1178                 retval = put_user(status, stat_addr);
1179         if (!retval && infop)
1180                 retval = put_user(SIGCHLD, &infop->si_signo);
1181         if (!retval && infop)
1182                 retval = put_user(0, &infop->si_errno);
1183         if (!retval && infop) {
1184                 int why;
1185
1186                 if ((status & 0x7f) == 0) {
1187                         why = CLD_EXITED;
1188                         status >>= 8;
1189                 } else {
1190                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1191                         status &= 0x7f;
1192                 }
1193                 retval = put_user((short)why, &infop->si_code);
1194                 if (!retval)
1195                         retval = put_user(status, &infop->si_status);
1196         }
1197         if (!retval && infop)
1198                 retval = put_user(p->pid, &infop->si_pid);
1199         if (!retval && infop)
1200                 retval = put_user(p->uid, &infop->si_uid);
1201         if (retval) {
1202                 // TODO: is this safe?
1203                 p->exit_state = EXIT_ZOMBIE;
1204                 return retval;
1205         }
1206         retval = p->pid;
1207         if (p->real_parent != p->parent) {
1208                 write_lock_irq(&tasklist_lock);
1209                 /* Double-check with lock held.  */
1210                 if (p->real_parent != p->parent) {
1211                         __ptrace_unlink(p);
1212                         // TODO: is this safe?
1213                         p->exit_state = EXIT_ZOMBIE;
1214                         /*
1215                          * If this is not a detached task, notify the parent.
1216                          * If it's still not detached after that, don't release
1217                          * it now.
1218                          */
1219                         if (p->exit_signal != -1) {
1220                                 do_notify_parent(p, p->exit_signal);
1221                                 if (p->exit_signal != -1)
1222                                         p = NULL;
1223                         }
1224                 }
1225                 write_unlock_irq(&tasklist_lock);
1226         }
1227         if (p != NULL)
1228                 release_task(p);
1229         BUG_ON(!retval);
1230         return retval;
1231 }
1232
1233 /*
1234  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1235  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1236  * the lock and this task is uninteresting.  If we return nonzero, we have
1237  * released the lock and the system call should return.
1238  */
1239 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1240                              struct siginfo __user *infop,
1241                              int __user *stat_addr, struct rusage __user *ru)
1242 {
1243         int retval, exit_code;
1244
1245         if (!p->exit_code)
1246                 return 0;
1247         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1248             p->signal && p->signal->group_stop_count > 0)
1249                 /*
1250                  * A group stop is in progress and this is the group leader.
1251                  * We won't report until all threads have stopped.
1252                  */
1253                 return 0;
1254
1255         /*
1256          * Now we are pretty sure this task is interesting.
1257          * Make sure it doesn't get reaped out from under us while we
1258          * give up the lock and then examine it below.  We don't want to
1259          * keep holding onto the tasklist_lock while we call getrusage and
1260          * possibly take page faults for user memory.
1261          */
1262         get_task_struct(p);
1263         read_unlock(&tasklist_lock);
1264
1265         if (unlikely(noreap)) {
1266                 pid_t pid = p->pid;
1267                 uid_t uid = p->uid;
1268                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1269
1270                 exit_code = p->exit_code;
1271                 if (unlikely(!exit_code) ||
1272                     unlikely(p->state & TASK_TRACED))
1273                         goto bail_ref;
1274                 return wait_noreap_copyout(p, pid, uid,
1275                                            why, (exit_code << 8) | 0x7f,
1276                                            infop, ru);
1277         }
1278
1279         write_lock_irq(&tasklist_lock);
1280
1281         /*
1282          * This uses xchg to be atomic with the thread resuming and setting
1283          * it.  It must also be done with the write lock held to prevent a
1284          * race with the EXIT_ZOMBIE case.
1285          */
1286         exit_code = xchg(&p->exit_code, 0);
1287         if (unlikely(p->exit_state)) {
1288                 /*
1289                  * The task resumed and then died.  Let the next iteration
1290                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1291                  * already be zero here if it resumed and did _exit(0).
1292                  * The task itself is dead and won't touch exit_code again;
1293                  * other processors in this function are locked out.
1294                  */
1295                 p->exit_code = exit_code;
1296                 exit_code = 0;
1297         }
1298         if (unlikely(exit_code == 0)) {
1299                 /*
1300                  * Another thread in this function got to it first, or it
1301                  * resumed, or it resumed and then died.
1302                  */
1303                 write_unlock_irq(&tasklist_lock);
1304 bail_ref:
1305                 put_task_struct(p);
1306                 /*
1307                  * We are returning to the wait loop without having successfully
1308                  * removed the process and having released the lock. We cannot
1309                  * continue, since the "p" task pointer is potentially stale.
1310                  *
1311                  * Return -EAGAIN, and do_wait() will restart the loop from the
1312                  * beginning. Do _not_ re-acquire the lock.
1313                  */
1314                 return -EAGAIN;
1315         }
1316
1317         /* move to end of parent's list to avoid starvation */
1318         remove_parent(p);
1319         add_parent(p);
1320
1321         write_unlock_irq(&tasklist_lock);
1322
1323         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1324         if (!retval && stat_addr)
1325                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1326         if (!retval && infop)
1327                 retval = put_user(SIGCHLD, &infop->si_signo);
1328         if (!retval && infop)
1329                 retval = put_user(0, &infop->si_errno);
1330         if (!retval && infop)
1331                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1332                                           ? CLD_TRAPPED : CLD_STOPPED),
1333                                   &infop->si_code);
1334         if (!retval && infop)
1335                 retval = put_user(exit_code, &infop->si_status);
1336         if (!retval && infop)
1337                 retval = put_user(p->pid, &infop->si_pid);
1338         if (!retval && infop)
1339                 retval = put_user(p->uid, &infop->si_uid);
1340         if (!retval)
1341                 retval = p->pid;
1342         put_task_struct(p);
1343
1344         BUG_ON(!retval);
1345         return retval;
1346 }
1347
1348 /*
1349  * Handle do_wait work for one task in a live, non-stopped state.
1350  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1351  * the lock and this task is uninteresting.  If we return nonzero, we have
1352  * released the lock and the system call should return.
1353  */
1354 static int wait_task_continued(task_t *p, int noreap,
1355                                struct siginfo __user *infop,
1356                                int __user *stat_addr, struct rusage __user *ru)
1357 {
1358         int retval;
1359         pid_t pid;
1360         uid_t uid;
1361
1362         if (unlikely(!p->signal))
1363                 return 0;
1364
1365         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1366                 return 0;
1367
1368         spin_lock_irq(&p->sighand->siglock);
1369         /* Re-check with the lock held.  */
1370         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1371                 spin_unlock_irq(&p->sighand->siglock);
1372                 return 0;
1373         }
1374         if (!noreap)
1375                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1376         spin_unlock_irq(&p->sighand->siglock);
1377
1378         pid = p->pid;
1379         uid = p->uid;
1380         get_task_struct(p);
1381         read_unlock(&tasklist_lock);
1382
1383         if (!infop) {
1384                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1385                 put_task_struct(p);
1386                 if (!retval && stat_addr)
1387                         retval = put_user(0xffff, stat_addr);
1388                 if (!retval)
1389                         retval = p->pid;
1390         } else {
1391                 retval = wait_noreap_copyout(p, pid, uid,
1392                                              CLD_CONTINUED, SIGCONT,
1393                                              infop, ru);
1394                 BUG_ON(retval == 0);
1395         }
1396
1397         return retval;
1398 }
1399
1400
1401 static inline int my_ptrace_child(struct task_struct *p)
1402 {
1403         if (!(p->ptrace & PT_PTRACED))
1404                 return 0;
1405         if (!(p->ptrace & PT_ATTACHED))
1406                 return 1;
1407         /*
1408          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1409          * we are the attacher.  If we are the real parent, this is a race
1410          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1411          * which we have to switch the parent links, but has already set
1412          * the flags in p->ptrace.
1413          */
1414         return (p->parent != p->real_parent);
1415 }
1416
1417 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1418                     int __user *stat_addr, struct rusage __user *ru)
1419 {
1420         DECLARE_WAITQUEUE(wait, current);
1421         struct task_struct *tsk;
1422         int flag, retval;
1423
1424         add_wait_queue(&current->signal->wait_chldexit,&wait);
1425 repeat:
1426         /*
1427          * We will set this flag if we see any child that might later
1428          * match our criteria, even if we are not able to reap it yet.
1429          */
1430         flag = 0;
1431         current->state = TASK_INTERRUPTIBLE;
1432         read_lock(&tasklist_lock);
1433         tsk = current;
1434         do {
1435                 struct task_struct *p;
1436                 struct list_head *_p;
1437                 int ret;
1438
1439                 list_for_each(_p,&tsk->children) {
1440                         p = list_entry(_p,struct task_struct,sibling);
1441
1442                         ret = eligible_child(pid, options, p);
1443                         if (!ret)
1444                                 continue;
1445
1446                         switch (p->state) {
1447                         case TASK_TRACED:
1448                                 /*
1449                                  * When we hit the race with PTRACE_ATTACH,
1450                                  * we will not report this child.  But the
1451                                  * race means it has not yet been moved to
1452                                  * our ptrace_children list, so we need to
1453                                  * set the flag here to avoid a spurious ECHILD
1454                                  * when the race happens with the only child.
1455                                  */
1456                                 flag = 1;
1457                                 if (!my_ptrace_child(p))
1458                                         continue;
1459                                 /*FALLTHROUGH*/
1460                         case TASK_STOPPED:
1461                                 /*
1462                                  * It's stopped now, so it might later
1463                                  * continue, exit, or stop again.
1464                                  */
1465                                 flag = 1;
1466                                 if (!(options & WUNTRACED) &&
1467                                     !my_ptrace_child(p))
1468                                         continue;
1469                                 retval = wait_task_stopped(p, ret == 2,
1470                                                            (options & WNOWAIT),
1471                                                            infop,
1472                                                            stat_addr, ru);
1473                                 if (retval == -EAGAIN)
1474                                         goto repeat;
1475                                 if (retval != 0) /* He released the lock.  */
1476                                         goto end;
1477                                 break;
1478                         default:
1479                         // case EXIT_DEAD:
1480                                 if (p->exit_state == EXIT_DEAD)
1481                                         continue;
1482                         // case EXIT_ZOMBIE:
1483                                 if (p->exit_state == EXIT_ZOMBIE) {
1484                                         /*
1485                                          * Eligible but we cannot release
1486                                          * it yet:
1487                                          */
1488                                         if (ret == 2)
1489                                                 goto check_continued;
1490                                         if (!likely(options & WEXITED))
1491                                                 continue;
1492                                         retval = wait_task_zombie(
1493                                                 p, (options & WNOWAIT),
1494                                                 infop, stat_addr, ru);
1495                                         /* He released the lock.  */
1496                                         if (retval != 0)
1497                                                 goto end;
1498                                         break;
1499                                 }
1500 check_continued:
1501                                 /*
1502                                  * It's running now, so it might later
1503                                  * exit, stop, or stop and then continue.
1504                                  */
1505                                 flag = 1;
1506                                 if (!unlikely(options & WCONTINUED))
1507                                         continue;
1508                                 retval = wait_task_continued(
1509                                         p, (options & WNOWAIT),
1510                                         infop, stat_addr, ru);
1511                                 if (retval != 0) /* He released the lock.  */
1512                                         goto end;
1513                                 break;
1514                         }
1515                 }
1516                 if (!flag) {
1517                         list_for_each(_p, &tsk->ptrace_children) {
1518                                 p = list_entry(_p, struct task_struct,
1519                                                 ptrace_list);
1520                                 if (!eligible_child(pid, options, p))
1521                                         continue;
1522                                 flag = 1;
1523                                 break;
1524                         }
1525                 }
1526                 if (options & __WNOTHREAD)
1527                         break;
1528                 tsk = next_thread(tsk);
1529                 if (tsk->signal != current->signal)
1530                         BUG();
1531         } while (tsk != current);
1532
1533         read_unlock(&tasklist_lock);
1534         if (flag) {
1535                 retval = 0;
1536                 if (options & WNOHANG)
1537                         goto end;
1538                 retval = -ERESTARTSYS;
1539                 if (signal_pending(current))
1540                         goto end;
1541                 schedule();
1542                 goto repeat;
1543         }
1544         retval = -ECHILD;
1545 end:
1546         current->state = TASK_RUNNING;
1547         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1548         if (infop) {
1549                 if (retval > 0)
1550                 retval = 0;
1551                 else {
1552                         /*
1553                          * For a WNOHANG return, clear out all the fields
1554                          * we would set so the user can easily tell the
1555                          * difference.
1556                          */
1557                         if (!retval)
1558                                 retval = put_user(0, &infop->si_signo);
1559                         if (!retval)
1560                                 retval = put_user(0, &infop->si_errno);
1561                         if (!retval)
1562                                 retval = put_user(0, &infop->si_code);
1563                         if (!retval)
1564                                 retval = put_user(0, &infop->si_pid);
1565                         if (!retval)
1566                                 retval = put_user(0, &infop->si_uid);
1567                         if (!retval)
1568                                 retval = put_user(0, &infop->si_status);
1569                 }
1570         }
1571         return retval;
1572 }
1573
1574 asmlinkage long sys_waitid(int which, pid_t pid,
1575                            struct siginfo __user *infop, int options,
1576                            struct rusage __user *ru)
1577 {
1578         long ret;
1579
1580         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1581                 return -EINVAL;
1582         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1583                 return -EINVAL;
1584
1585         switch (which) {
1586         case P_ALL:
1587                 pid = -1;
1588                 break;
1589         case P_PID:
1590                 if (pid <= 0)
1591                         return -EINVAL;
1592                 break;
1593         case P_PGID:
1594                 if (pid <= 0)
1595                         return -EINVAL;
1596                 pid = -pid;
1597                 break;
1598         default:
1599                 return -EINVAL;
1600         }
1601
1602         ret = do_wait(pid, options, infop, NULL, ru);
1603
1604         /* avoid REGPARM breakage on x86: */
1605         prevent_tail_call(ret);
1606         return ret;
1607 }
1608
1609 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1610                           int options, struct rusage __user *ru)
1611 {
1612         long ret;
1613
1614         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1615                         __WNOTHREAD|__WCLONE|__WALL))
1616                 return -EINVAL;
1617         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1618
1619         /* avoid REGPARM breakage on x86: */
1620         prevent_tail_call(ret);
1621         return ret;
1622 }
1623
1624 #ifdef __ARCH_WANT_SYS_WAITPID
1625
1626 /*
1627  * sys_waitpid() remains for compatibility. waitpid() should be
1628  * implemented by calling sys_wait4() from libc.a.
1629  */
1630 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1631 {
1632         return sys_wait4(pid, stat_addr, options, NULL);
1633 }
1634
1635 #endif