This commit was manufactured by cvs2svn to create tag
[linux-2.6.git] / 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/security.h>
18 #include <linux/acct.h>
19 #include <linux/file.h>
20 #include <linux/binfmts.h>
21 #include <linux/ptrace.h>
22 #include <linux/profile.h>
23 #include <linux/mount.h>
24 #include <linux/proc_fs.h>
25 #include <linux/mempolicy.h>
26 #include <linux/vs_limit.h>
27
28 #include <linux/ckrm.h>
29 #include <linux/ckrm_tsk.h>
30
31 #include <asm/uaccess.h>
32 #include <asm/unistd.h>
33 #include <asm/pgtable.h>
34 #include <asm/mmu_context.h>
35
36 extern void sem_exit (void);
37 extern struct task_struct *child_reaper;
38
39 int getrusage(struct task_struct *, int, struct rusage __user *);
40
41 static void __unhash_process(struct task_struct *p)
42 {
43         nr_threads--;
44         /* tasklist_lock is held, is this sufficient? */
45         if (p->vx_info) {
46                 atomic_dec(&p->vx_info->cacct.nr_threads);
47                 vx_nproc_dec(p->vx_info);
48                 // atomic_dec(&p->vx_info->limit.res[RLIMIT_NPROC]);
49         }
50         detach_pid(p, PIDTYPE_PID);
51         detach_pid(p, PIDTYPE_TGID);
52         if (thread_group_leader(p)) {
53                 detach_pid(p, PIDTYPE_PGID);
54                 detach_pid(p, PIDTYPE_SID);
55                 if (p->pid)
56                         __get_cpu_var(process_counts)--;
57         }
58
59         REMOVE_LINKS(p);
60 }
61
62 void release_task(struct task_struct * p)
63 {
64         int zap_leader;
65         task_t *leader;
66         struct dentry *proc_dentry;
67
68 repeat: 
69         BUG_ON(p->state < TASK_ZOMBIE);
70  
71         atomic_dec(&p->user->processes);
72         spin_lock(&p->proc_lock);
73         proc_dentry = proc_pid_unhash(p);
74         write_lock_irq(&tasklist_lock);
75         if (unlikely(p->ptrace))
76                 __ptrace_unlink(p);
77         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
78         __exit_signal(p);
79         __exit_sighand(p);
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->state == TASK_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         p->parent->cutime += p->utime + p->cutime;
104         p->parent->cstime += p->stime + p->cstime;
105         p->parent->cmin_flt += p->min_flt + p->cmin_flt;
106         p->parent->cmaj_flt += p->maj_flt + p->cmaj_flt;
107         p->parent->cnvcsw += p->nvcsw + p->cnvcsw;
108         p->parent->cnivcsw += p->nivcsw + p->cnivcsw;
109         sched_exit(p);
110         write_unlock_irq(&tasklist_lock);
111         spin_unlock(&p->proc_lock);
112         proc_pid_flush(proc_dentry);
113         release_thread(p);
114         put_task_struct(p);
115
116         p = leader;
117         if (unlikely(zap_leader))
118                 goto repeat;
119 }
120
121 /* we are using it only for SMP init */
122
123 void unhash_process(struct task_struct *p)
124 {
125         struct dentry *proc_dentry;
126
127         spin_lock(&p->proc_lock);
128         proc_dentry = proc_pid_unhash(p);
129         write_lock_irq(&tasklist_lock);
130         __unhash_process(p);
131         write_unlock_irq(&tasklist_lock);
132         spin_unlock(&p->proc_lock);
133         proc_pid_flush(proc_dentry);
134 }
135
136 /*
137  * This checks not only the pgrp, but falls back on the pid if no
138  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
139  * without this...
140  */
141 int session_of_pgrp(int pgrp)
142 {
143         struct task_struct *p;
144         struct list_head *l;
145         struct pid *pid;
146         int sid = -1;
147
148         read_lock(&tasklist_lock);
149         for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid)
150                 if (p->signal->session > 0) {
151                         sid = p->signal->session;
152                         goto out;
153                 }
154         p = find_task_by_pid(pgrp);
155         if (p)
156                 sid = p->signal->session;
157 out:
158         read_unlock(&tasklist_lock);
159         
160         return sid;
161 }
162
163 /*
164  * Determine if a process group is "orphaned", according to the POSIX
165  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
166  * by terminal-generated stop signals.  Newly orphaned process groups are
167  * to receive a SIGHUP and a SIGCONT.
168  *
169  * "I ask you, have you ever known what it is to be an orphan?"
170  */
171 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
172 {
173         struct task_struct *p;
174         struct list_head *l;
175         struct pid *pid;
176         int ret = 1;
177
178         for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
179                 if (p == ignored_task
180                                 || p->state >= TASK_ZOMBIE 
181                                 || p->real_parent->pid == 1)
182                         continue;
183                 if (process_group(p->real_parent) != pgrp
184                             && p->real_parent->signal->session == p->signal->session) {
185                         ret = 0;
186                         break;
187                 }
188         }
189         return ret;     /* (sighing) "Often!" */
190 }
191
192 int is_orphaned_pgrp(int pgrp)
193 {
194         int retval;
195
196         read_lock(&tasklist_lock);
197         retval = will_become_orphaned_pgrp(pgrp, NULL);
198         read_unlock(&tasklist_lock);
199
200         return retval;
201 }
202
203 static inline int has_stopped_jobs(int pgrp)
204 {
205         int retval = 0;
206         struct task_struct *p;
207         struct list_head *l;
208         struct pid *pid;
209
210         for_each_task_pid(pgrp, PIDTYPE_PGID, p, l, pid) {
211                 if (p->state != TASK_STOPPED)
212                         continue;
213
214                 /* If p is stopped by a debugger on a signal that won't
215                    stop it, then don't count p as stopped.  This isn't
216                    perfect but it's a good approximation.  */
217                 if (unlikely (p->ptrace)
218                     && p->exit_code != SIGSTOP
219                     && p->exit_code != SIGTSTP
220                     && p->exit_code != SIGTTOU
221                     && p->exit_code != SIGTTIN)
222                         continue;
223
224                 retval = 1;
225                 break;
226         }
227         return retval;
228 }
229
230 /**
231  * reparent_to_init() - Reparent the calling kernel thread to the init task.
232  *
233  * If a kernel thread is launched as a result of a system call, or if
234  * it ever exits, it should generally reparent itself to init so that
235  * it is correctly cleaned up on exit.
236  *
237  * The various task state such as scheduling policy and priority may have
238  * been inherited from a user process, so we reset them to sane values here.
239  *
240  * NOTE that reparent_to_init() gives the caller full capabilities.
241  */
242 void reparent_to_init(void)
243 {
244         write_lock_irq(&tasklist_lock);
245
246         ptrace_unlink(current);
247         /* Reparent to init */
248         REMOVE_LINKS(current);
249         /* FIXME handle vchild_reaper/initpid */
250         current->parent = child_reaper;
251         current->real_parent = child_reaper;
252         SET_LINKS(current);
253
254         /* Set the exit signal to SIGCHLD so we signal init on exit */
255         current->exit_signal = SIGCHLD;
256
257         if ((current->policy == SCHED_NORMAL) && (task_nice(current) < 0))
258                 set_user_nice(current, 0);
259         /* cpus_allowed? */
260         /* rt_priority? */
261         /* signals? */
262         security_task_reparent_to_init(current);
263         memcpy(current->rlim, init_task.rlim, sizeof(*(current->rlim)));
264         atomic_inc(&(INIT_USER->__count));
265         switch_uid(INIT_USER);
266
267         write_unlock_irq(&tasklist_lock);
268 }
269
270 void __set_special_pids(pid_t session, pid_t pgrp)
271 {
272         struct task_struct *curr = current;
273
274         if (curr->signal->session != session) {
275                 detach_pid(curr, PIDTYPE_SID);
276                 curr->signal->session = session;
277                 attach_pid(curr, PIDTYPE_SID, session);
278         }
279         if (process_group(curr) != pgrp) {
280                 detach_pid(curr, PIDTYPE_PGID);
281                 curr->signal->pgrp = pgrp;
282                 attach_pid(curr, PIDTYPE_PGID, pgrp);
283         }
284 }
285
286 void set_special_pids(pid_t session, pid_t pgrp)
287 {
288         write_lock_irq(&tasklist_lock);
289         __set_special_pids(session, pgrp);
290         write_unlock_irq(&tasklist_lock);
291 }
292
293 /*
294  * Let kernel threads use this to say that they
295  * allow a certain signal (since daemonize() will
296  * have disabled all of them by default).
297  */
298 int allow_signal(int sig)
299 {
300         if (sig < 1 || sig > _NSIG)
301                 return -EINVAL;
302
303         spin_lock_irq(&current->sighand->siglock);
304         sigdelset(&current->blocked, sig);
305         if (!current->mm) {
306                 /* Kernel threads handle their own signals.
307                    Let the signal code know it'll be handled, so
308                    that they don't get converted to SIGKILL or
309                    just silently dropped */
310                 current->sighand->action[(sig)-1].sa.sa_handler = (void *)2;
311         }
312         recalc_sigpending();
313         spin_unlock_irq(&current->sighand->siglock);
314         return 0;
315 }
316
317 EXPORT_SYMBOL(allow_signal);
318
319 int disallow_signal(int sig)
320 {
321         if (sig < 1 || sig > _NSIG)
322                 return -EINVAL;
323
324         spin_lock_irq(&current->sighand->siglock);
325         sigaddset(&current->blocked, sig);
326         recalc_sigpending();
327         spin_unlock_irq(&current->sighand->siglock);
328         return 0;
329 }
330
331 EXPORT_SYMBOL(disallow_signal);
332
333 /*
334  *      Put all the gunge required to become a kernel thread without
335  *      attached user resources in one place where it belongs.
336  */
337
338 void daemonize(const char *name, ...)
339 {
340         va_list args;
341         struct fs_struct *fs;
342         sigset_t blocked;
343
344         va_start(args, name);
345         vsnprintf(current->comm, sizeof(current->comm), name, args);
346         va_end(args);
347
348         /*
349          * If we were started as result of loading a module, close all of the
350          * user space pages.  We don't need them, and if we didn't close them
351          * they would be locked into memory.
352          */
353         exit_mm(current);
354
355         set_special_pids(1, 1);
356         current->signal->tty = NULL;
357
358         /* Block and flush all signals */
359         sigfillset(&blocked);
360         sigprocmask(SIG_BLOCK, &blocked, NULL);
361         flush_signals(current);
362
363         /* Become as one with the init task */
364
365         exit_fs(current);       /* current->fs->count--; */
366         fs = init_task.fs;
367         current->fs = fs;
368         atomic_inc(&fs->count);
369         exit_files(current);
370         current->files = init_task.files;
371         atomic_inc(&current->files->count);
372
373         reparent_to_init();
374 }
375
376 EXPORT_SYMBOL(daemonize);
377
378 static inline void close_files(struct files_struct * files)
379 {
380         int i, j;
381
382         j = 0;
383         for (;;) {
384                 unsigned long set;
385                 i = j * __NFDBITS;
386                 if (i >= files->max_fdset || i >= files->max_fds)
387                         break;
388                 set = files->open_fds->fds_bits[j++];
389                 while (set) {
390                         if (set & 1) {
391                                 struct file * file = xchg(&files->fd[i], NULL);
392                                 if (file)
393                                         filp_close(file, files);
394                                 vx_openfd_dec(fd);
395                         }
396                         i++;
397                         set >>= 1;
398                 }
399         }
400 }
401
402 struct files_struct *get_files_struct(struct task_struct *task)
403 {
404         struct files_struct *files;
405
406         task_lock(task);
407         files = task->files;
408         if (files)
409                 atomic_inc(&files->count);
410         task_unlock(task);
411
412         return files;
413 }
414
415 void fastcall put_files_struct(struct files_struct *files)
416 {
417         if (atomic_dec_and_test(&files->count)) {
418                 close_files(files);
419                 /*
420                  * Free the fd and fdset arrays if we expanded them.
421                  */
422                 if (files->fd != &files->fd_array[0])
423                         free_fd_array(files->fd, files->max_fds);
424                 if (files->max_fdset > __FD_SETSIZE) {
425                         free_fdset(files->open_fds, files->max_fdset);
426                         free_fdset(files->close_on_exec, files->max_fdset);
427                 }
428                 kmem_cache_free(files_cachep, files);
429         }
430 }
431
432 EXPORT_SYMBOL(put_files_struct);
433
434 static inline void __exit_files(struct task_struct *tsk)
435 {
436         struct files_struct * files = tsk->files;
437
438         if (files) {
439                 task_lock(tsk);
440                 tsk->files = NULL;
441                 task_unlock(tsk);
442                 put_files_struct(files);
443         }
444 }
445
446 void exit_files(struct task_struct *tsk)
447 {
448         __exit_files(tsk);
449 }
450
451 static inline void __put_fs_struct(struct fs_struct *fs)
452 {
453         /* No need to hold fs->lock if we are killing it */
454         if (atomic_dec_and_test(&fs->count)) {
455                 dput(fs->root);
456                 mntput(fs->rootmnt);
457                 dput(fs->pwd);
458                 mntput(fs->pwdmnt);
459                 if (fs->altroot) {
460                         dput(fs->altroot);
461                         mntput(fs->altrootmnt);
462                 }
463                 kmem_cache_free(fs_cachep, fs);
464         }
465 }
466
467 void put_fs_struct(struct fs_struct *fs)
468 {
469         __put_fs_struct(fs);
470 }
471
472 static inline void __exit_fs(struct task_struct *tsk)
473 {
474         struct fs_struct * fs = tsk->fs;
475
476         if (fs) {
477                 task_lock(tsk);
478                 tsk->fs = NULL;
479                 task_unlock(tsk);
480                 __put_fs_struct(fs);
481         }
482 }
483
484 void exit_fs(struct task_struct *tsk)
485 {
486         __exit_fs(tsk);
487 }
488
489 EXPORT_SYMBOL_GPL(exit_fs);
490
491 /*
492  * Turn us into a lazy TLB process if we
493  * aren't already..
494  */
495 static inline void __exit_mm(struct task_struct * tsk)
496 {
497         struct mm_struct *mm = tsk->mm;
498
499         mm_release(tsk, mm);
500         if (!mm)
501                 return;
502         /*
503          * Serialize with any possible pending coredump.
504          * We must hold mmap_sem around checking core_waiters
505          * and clearing tsk->mm.  The core-inducing thread
506          * will increment core_waiters for each thread in the
507          * group with ->mm != NULL.
508          */
509         down_read(&mm->mmap_sem);
510         if (mm->core_waiters) {
511                 up_read(&mm->mmap_sem);
512                 down_write(&mm->mmap_sem);
513                 if (!--mm->core_waiters)
514                         complete(mm->core_startup_done);
515                 up_write(&mm->mmap_sem);
516
517                 wait_for_completion(&mm->core_done);
518                 down_read(&mm->mmap_sem);
519         }
520         atomic_inc(&mm->mm_count);
521         if (mm != tsk->active_mm) BUG();
522         /* more a memory barrier than a real lock */
523         task_lock(tsk);
524         tsk->mm = NULL;
525         up_read(&mm->mmap_sem);
526         enter_lazy_tlb(mm, current);
527         task_unlock(tsk);
528         mmput(mm);
529 }
530
531 void exit_mm(struct task_struct *tsk)
532 {
533         __exit_mm(tsk);
534 }
535
536 EXPORT_SYMBOL(exit_mm);
537
538 static inline void choose_new_parent(task_t *p, task_t *reaper, task_t *child_reaper)
539 {
540         /*
541          * Make sure we're not reparenting to ourselves and that
542          * the parent is not a zombie.
543          */
544         if (p == reaper || reaper->state >= TASK_ZOMBIE)
545                 p->real_parent = child_reaper;
546         else
547                 p->real_parent = reaper;
548         if (p->parent == p->real_parent)
549                 BUG();
550 }
551
552 static inline void reparent_thread(task_t *p, task_t *father, int traced)
553 {
554         /* We don't want people slaying init.  */
555         if (p->exit_signal != -1)
556                 p->exit_signal = SIGCHLD;
557         p->self_exec_id++;
558
559         if (p->pdeath_signal)
560                 /* We already hold the tasklist_lock here.  */
561                 group_send_sig_info(p->pdeath_signal, (void *) 0, p);
562
563         /* Move the child from its dying parent to the new one.  */
564         if (unlikely(traced)) {
565                 /* Preserve ptrace links if someone else is tracing this child.  */
566                 list_del_init(&p->ptrace_list);
567                 if (p->parent != p->real_parent)
568                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
569         } else {
570                 /* If this child is being traced, then we're the one tracing it
571                  * anyway, so let go of it.
572                  */
573                 p->ptrace = 0;
574                 list_del_init(&p->sibling);
575                 p->parent = p->real_parent;
576                 list_add_tail(&p->sibling, &p->parent->children);
577
578                 /* If we'd notified the old parent about this child's death,
579                  * also notify the new parent.
580                  */
581                 if (p->state == TASK_ZOMBIE && p->exit_signal != -1 &&
582                     thread_group_empty(p))
583                         do_notify_parent(p, p->exit_signal);
584         }
585
586         /*
587          * process group orphan check
588          * Case ii: Our child is in a different pgrp
589          * than we are, and it was the only connection
590          * outside, so the child pgrp is now orphaned.
591          */
592         if ((process_group(p) != process_group(father)) &&
593             (p->signal->session == father->signal->session)) {
594                 int pgrp = process_group(p);
595
596                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
597                         __kill_pg_info(SIGHUP, (void *)1, pgrp);
598                         __kill_pg_info(SIGCONT, (void *)1, pgrp);
599                 }
600         }
601 }
602
603 /*
604  * When we die, we re-parent all our children.
605  * Try to give them to another thread in our thread
606  * group, and if no such member exists, give it to
607  * the global child reaper process (ie "init")
608  */
609 static inline void forget_original_parent(struct task_struct * father)
610 {
611         struct task_struct *p, *reaper = father;
612         struct list_head *_p, *_n;
613
614         /* FIXME handle vchild_reaper/initpid */
615         reaper = father->group_leader;
616         if (reaper == father)
617                 reaper = child_reaper;
618
619         /*
620          * There are only two places where our children can be:
621          *
622          * - in our child list
623          * - in our ptraced child list
624          *
625          * Search them and reparent children.
626          */
627         list_for_each_safe(_p, _n, &father->children) {
628                 p = list_entry(_p,struct task_struct,sibling);
629                 if (father == p->real_parent) {
630                         choose_new_parent(p, reaper, child_reaper);
631                         reparent_thread(p, father, 0);
632                 } else {
633                         ptrace_unlink (p);
634                         if (p->state == TASK_ZOMBIE && p->exit_signal != -1 &&
635                             thread_group_empty(p))
636                                 do_notify_parent(p, p->exit_signal);
637                 }
638         }
639         list_for_each_safe(_p, _n, &father->ptrace_children) {
640                 p = list_entry(_p,struct task_struct,ptrace_list);
641                 choose_new_parent(p, reaper, child_reaper);
642                 reparent_thread(p, father, 1);
643         }
644 }
645
646 /*
647  * Send signals to all our closest relatives so that they know
648  * to properly mourn us..
649  */
650 static void exit_notify(struct task_struct *tsk)
651 {
652         int state;
653         struct task_struct *t;
654
655         ckrm_cb_exit(tsk);
656
657         if (signal_pending(tsk) && !tsk->signal->group_exit
658             && !thread_group_empty(tsk)) {
659                 /*
660                  * This occurs when there was a race between our exit
661                  * syscall and a group signal choosing us as the one to
662                  * wake up.  It could be that we are the only thread
663                  * alerted to check for pending signals, but another thread
664                  * should be woken now to take the signal since we will not.
665                  * Now we'll wake all the threads in the group just to make
666                  * sure someone gets all the pending signals.
667                  */
668                 read_lock(&tasklist_lock);
669                 spin_lock_irq(&tsk->sighand->siglock);
670                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
671                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
672                                 recalc_sigpending_tsk(t);
673                                 if (signal_pending(t))
674                                         signal_wake_up(t, 0);
675                         }
676                 spin_unlock_irq(&tsk->sighand->siglock);
677                 read_unlock(&tasklist_lock);
678         }
679
680         write_lock_irq(&tasklist_lock);
681
682         /*
683          * This does two things:
684          *
685          * A.  Make init inherit all the child processes
686          * B.  Check to see if any process groups have become orphaned
687          *      as a result of our exiting, and if they have any stopped
688          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
689          */
690
691         forget_original_parent(tsk);
692         BUG_ON(!list_empty(&tsk->children));
693
694         /*
695          * Check to see if any process groups have become orphaned
696          * as a result of our exiting, and if they have any stopped
697          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
698          *
699          * Case i: Our father is in a different pgrp than we are
700          * and we were the only connection outside, so our pgrp
701          * is about to become orphaned.
702          */
703          
704         t = tsk->real_parent;
705         
706         if ((process_group(t) != process_group(tsk)) &&
707             (t->signal->session == tsk->signal->session) &&
708             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
709             has_stopped_jobs(process_group(tsk))) {
710                 __kill_pg_info(SIGHUP, (void *)1, process_group(tsk));
711                 __kill_pg_info(SIGCONT, (void *)1, process_group(tsk));
712         }
713
714         /* Let father know we died 
715          *
716          * Thread signals are configurable, but you aren't going to use
717          * that to send signals to arbitary processes. 
718          * That stops right now.
719          *
720          * If the parent exec id doesn't match the exec id we saved
721          * when we started then we know the parent has changed security
722          * domain.
723          *
724          * If our self_exec id doesn't match our parent_exec_id then
725          * we have changed execution domain as these two values started
726          * the same after a fork.
727          *      
728          */
729         
730         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
731             ( tsk->parent_exec_id != t->self_exec_id  ||
732               tsk->self_exec_id != tsk->parent_exec_id)
733             && !capable(CAP_KILL))
734                 tsk->exit_signal = SIGCHLD;
735
736
737         /* If something other than our normal parent is ptracing us, then
738          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
739          * only has special meaning to our real parent.
740          */
741         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
742                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
743                 do_notify_parent(tsk, signal);
744         } else if (tsk->ptrace) {
745                 do_notify_parent(tsk, SIGCHLD);
746         }
747
748         state = TASK_ZOMBIE;
749         if (tsk->exit_signal == -1 && tsk->ptrace == 0)
750                 state = TASK_DEAD;
751         tsk->state = state;
752         tsk->flags |= PF_DEAD;
753
754         /*
755          * Clear these here so that update_process_times() won't try to deliver
756          * itimer, profile or rlimit signals to this task while it is in late exit.
757          */
758         tsk->it_virt_value = 0;
759         tsk->it_prof_value = 0;
760         tsk->rlim[RLIMIT_CPU].rlim_cur = RLIM_INFINITY;
761
762         /*
763          * In the preemption case it must be impossible for the task
764          * to get runnable again, so use "_raw_" unlock to keep
765          * preempt_count elevated until we schedule().
766          *
767          * To avoid deadlock on SMP, interrupts must be unmasked.  If we
768          * don't, subsequently called functions (e.g, wait_task_inactive()
769          * via release_task()) will spin, with interrupt flags
770          * unwittingly blocked, until the other task sleeps.  That task
771          * may itself be waiting for smp_call_function() to answer and
772          * complete, and with interrupts blocked that will never happen.
773          */
774         _raw_write_unlock(&tasklist_lock);
775         local_irq_enable();
776
777         /* If the process is dead, release it - nobody will wait for it */
778         if (state == TASK_DEAD)
779                 release_task(tsk);
780
781 }
782
783 asmlinkage NORET_TYPE void do_exit(long code)
784 {
785         struct task_struct *tsk = current;
786
787         if (unlikely(in_interrupt()))
788                 panic("Aiee, killing interrupt handler!");
789         if (unlikely(!tsk->pid))
790                 panic("Attempted to kill the idle task!");
791         if (unlikely(tsk->pid == 1))
792                 panic("Attempted to kill init!");
793         if (tsk->io_context)
794                 exit_io_context();
795         tsk->flags |= PF_EXITING;
796         del_timer_sync(&tsk->real_timer);
797
798         if (unlikely(in_atomic()))
799                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
800                                 current->comm, current->pid,
801                                 preempt_count());
802
803         profile_exit_task(tsk);
804  
805         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
806                 current->ptrace_message = code;
807                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
808         }
809
810         acct_process(code);
811         if (current->tux_info) {
812 #ifdef CONFIG_TUX_DEBUG
813                 printk("Possibly unexpected TUX-thread exit(%ld) at %p?\n",
814                         code, __builtin_return_address(0));
815 #endif
816                 current->tux_exit();
817         }
818         __exit_mm(tsk);
819
820         exit_sem(tsk);
821         __exit_files(tsk);
822         __exit_fs(tsk);
823         exit_namespace(tsk);
824         exit_thread();
825 #ifdef CONFIG_NUMA
826         mpol_free(tsk->mempolicy);
827 #endif
828
829         if (tsk->signal->leader)
830                 disassociate_ctty(1);
831
832         module_put(tsk->thread_info->exec_domain->module);
833         if (tsk->binfmt)
834                 module_put(tsk->binfmt->module);
835
836         tsk->exit_code = code;
837 #ifdef CONFIG_CKRM_TYPE_TASKCLASS
838         numtasks_put_ref(tsk->taskclass);
839 #endif
840         exit_notify(tsk);
841         schedule();
842         BUG();
843         /* Avoid "noreturn function does return".  */
844         for (;;) ;
845 }
846
847 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
848 {
849         if (comp)
850                 complete(comp);
851         
852         do_exit(code);
853 }
854
855 EXPORT_SYMBOL(complete_and_exit);
856
857 asmlinkage long sys_exit(int error_code)
858 {
859         do_exit((error_code&0xff)<<8);
860 }
861
862 task_t fastcall *next_thread(const task_t *p)
863 {
864         const struct pid_link *link = p->pids + PIDTYPE_TGID;
865         const struct list_head *tmp, *head = &link->pidptr->task_list;
866
867 #ifdef CONFIG_SMP
868         if (!p->sighand)
869                 BUG();
870         if (!spin_is_locked(&p->sighand->siglock) &&
871                                 !rwlock_is_locked(&tasklist_lock))
872                 BUG();
873 #endif
874         tmp = link->pid_chain.next;
875         if (tmp == head)
876                 tmp = head->next;
877
878         return pid_task(tmp, PIDTYPE_TGID);
879 }
880
881 EXPORT_SYMBOL(next_thread);
882
883 /*
884  * Take down every thread in the group.  This is called by fatal signals
885  * as well as by sys_exit_group (below).
886  */
887 NORET_TYPE void
888 do_group_exit(int exit_code)
889 {
890         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
891
892         if (current->signal->group_exit)
893                 exit_code = current->signal->group_exit_code;
894         else if (!thread_group_empty(current)) {
895                 struct signal_struct *const sig = current->signal;
896                 struct sighand_struct *const sighand = current->sighand;
897                 read_lock(&tasklist_lock);
898                 spin_lock_irq(&sighand->siglock);
899                 if (sig->group_exit)
900                         /* Another thread got here before we took the lock.  */
901                         exit_code = sig->group_exit_code;
902                 else {
903                         sig->group_exit = 1;
904                         sig->group_exit_code = exit_code;
905                         zap_other_threads(current);
906                 }
907                 spin_unlock_irq(&sighand->siglock);
908                 read_unlock(&tasklist_lock);
909         }
910
911         do_exit(exit_code);
912         /* NOTREACHED */
913 }
914
915 /*
916  * this kills every thread in the thread group. Note that any externally
917  * wait4()-ing process will get the correct exit code - even if this
918  * thread is not the thread group leader.
919  */
920 asmlinkage void sys_exit_group(int error_code)
921 {
922         do_group_exit((error_code & 0xff) << 8);
923 }
924
925 static int eligible_child(pid_t pid, int options, task_t *p)
926 {
927         if (pid > 0) {
928                 if (p->pid != pid)
929                         return 0;
930         } else if (!pid) {
931                 if (process_group(p) != process_group(current))
932                         return 0;
933         } else if (pid != -1) {
934                 if (process_group(p) != -pid)
935                         return 0;
936         }
937
938         /*
939          * Do not consider detached threads that are
940          * not ptraced:
941          */
942         if (p->exit_signal == -1 && !p->ptrace)
943                 return 0;
944
945         /* Wait for all children (clone and not) if __WALL is set;
946          * otherwise, wait for clone children *only* if __WCLONE is
947          * set; otherwise, wait for non-clone children *only*.  (Note:
948          * A "clone" child here is one that reports to its parent
949          * using a signal other than SIGCHLD.) */
950         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
951             && !(options & __WALL))
952                 return 0;
953         /*
954          * Do not consider thread group leaders that are
955          * in a non-empty thread group:
956          */
957         if (current->tgid != p->tgid && delay_group_leader(p))
958                 return 2;
959
960         if (security_task_wait(p))
961                 return 0;
962
963         return 1;
964 }
965
966 /*
967  * Handle sys_wait4 work for one task in state TASK_ZOMBIE.  We hold
968  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
969  * the lock and this task is uninteresting.  If we return nonzero, we have
970  * released the lock and the system call should return.
971  */
972 static int wait_task_zombie(task_t *p, unsigned int __user *stat_addr, struct rusage __user *ru)
973 {
974         unsigned long state;
975         int retval;
976
977         /*
978          * Try to move the task's state to DEAD
979          * only one thread is allowed to do this:
980          */
981         state = xchg(&p->state, TASK_DEAD);
982         if (state != TASK_ZOMBIE) {
983                 BUG_ON(state != TASK_DEAD);
984                 return 0;
985         }
986         if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
987                 /*
988                  * This can only happen in a race with a ptraced thread
989                  * dying on another processor.
990                  */
991                 return 0;
992
993         /*
994          * Now we are sure this task is interesting, and no other
995          * thread can reap it because we set its state to TASK_DEAD.
996          */
997         read_unlock(&tasklist_lock);
998
999         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1000         if (!retval && stat_addr) {
1001                 if (p->signal->group_exit)
1002                         retval = put_user(p->signal->group_exit_code, stat_addr);
1003                 else
1004                         retval = put_user(p->exit_code, stat_addr);
1005         }
1006         if (retval) {
1007                 p->state = TASK_ZOMBIE;
1008                 return retval;
1009         }
1010         retval = p->pid;
1011         if (p->real_parent != p->parent) {
1012                 write_lock_irq(&tasklist_lock);
1013                 /* Double-check with lock held.  */
1014                 if (p->real_parent != p->parent) {
1015                         __ptrace_unlink(p);
1016                         p->state = TASK_ZOMBIE;
1017                         /* If this is a detached thread, this is where it goes away.  */
1018                         if (p->exit_signal == -1) {
1019                                 /* release_task takes the lock itself.  */
1020                                 write_unlock_irq(&tasklist_lock);
1021                                 release_task (p);
1022                         }
1023                         else {
1024                                 do_notify_parent(p, p->exit_signal);
1025                                 write_unlock_irq(&tasklist_lock);
1026                         }
1027                         p = NULL;
1028                 }
1029                 else
1030                         write_unlock_irq(&tasklist_lock);
1031         }
1032         if (p != NULL)
1033                 release_task(p);
1034         BUG_ON(!retval);
1035         return retval;
1036 }
1037
1038 /*
1039  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1040  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1041  * the lock and this task is uninteresting.  If we return nonzero, we have
1042  * released the lock and the system call should return.
1043  */
1044 static int wait_task_stopped(task_t *p, int delayed_group_leader,
1045                              unsigned int __user *stat_addr,
1046                              struct rusage __user *ru)
1047 {
1048         int retval, exit_code;
1049
1050         if (!p->exit_code)
1051                 return 0;
1052         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1053             p->signal && p->signal->group_stop_count > 0)
1054                 /*
1055                  * A group stop is in progress and this is the group leader.
1056                  * We won't report until all threads have stopped.
1057                  */
1058                 return 0;
1059
1060         /*
1061          * Now we are pretty sure this task is interesting.
1062          * Make sure it doesn't get reaped out from under us while we
1063          * give up the lock and then examine it below.  We don't want to
1064          * keep holding onto the tasklist_lock while we call getrusage and
1065          * possibly take page faults for user memory.
1066          */
1067         get_task_struct(p);
1068         read_unlock(&tasklist_lock);
1069         write_lock_irq(&tasklist_lock);
1070
1071         /*
1072          * This uses xchg to be atomic with the thread resuming and setting
1073          * it.  It must also be done with the write lock held to prevent a
1074          * race with the TASK_ZOMBIE case.
1075          */
1076         exit_code = xchg(&p->exit_code, 0);
1077         if (unlikely(p->state > TASK_STOPPED)) {
1078                 /*
1079                  * The task resumed and then died.  Let the next iteration
1080                  * catch it in TASK_ZOMBIE.  Note that exit_code might
1081                  * already be zero here if it resumed and did _exit(0).
1082                  * The task itself is dead and won't touch exit_code again;
1083                  * other processors in this function are locked out.
1084                  */
1085                 p->exit_code = exit_code;
1086                 exit_code = 0;
1087         }
1088         if (unlikely(exit_code == 0)) {
1089                 /*
1090                  * Another thread in this function got to it first, or it
1091                  * resumed, or it resumed and then died.
1092                  */
1093                 write_unlock_irq(&tasklist_lock);
1094                 put_task_struct(p);
1095                 read_lock(&tasklist_lock);
1096                 return 0;
1097         }
1098
1099         /* move to end of parent's list to avoid starvation */
1100         remove_parent(p);
1101         add_parent(p, p->parent);
1102
1103         write_unlock_irq(&tasklist_lock);
1104
1105         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1106         if (!retval && stat_addr)
1107                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1108         if (!retval)
1109                 retval = p->pid;
1110         put_task_struct(p);
1111
1112         BUG_ON(!retval);
1113         return retval;
1114 }
1115
1116 asmlinkage long sys_wait4(pid_t pid,unsigned int __user *stat_addr, int options, struct rusage __user *ru)
1117 {
1118         DECLARE_WAITQUEUE(wait, current);
1119         struct task_struct *tsk;
1120         int flag, retval;
1121
1122         if (options & ~(WNOHANG|WUNTRACED|__WNOTHREAD|__WCLONE|__WALL))
1123                 return -EINVAL;
1124
1125         add_wait_queue(&current->wait_chldexit,&wait);
1126 repeat:
1127         flag = 0;
1128         current->state = TASK_INTERRUPTIBLE;
1129         read_lock(&tasklist_lock);
1130         tsk = current;
1131         do {
1132                 struct task_struct *p;
1133                 struct list_head *_p;
1134                 int ret;
1135
1136                 list_for_each(_p,&tsk->children) {
1137                         p = list_entry(_p,struct task_struct,sibling);
1138
1139                         ret = eligible_child(pid, options, p);
1140                         if (!ret)
1141                                 continue;
1142                         flag = 1;
1143
1144                         switch (p->state) {
1145                         case TASK_STOPPED:
1146                                 if (!(options & WUNTRACED) &&
1147                                     !(p->ptrace & PT_PTRACED))
1148                                         continue;
1149                                 retval = wait_task_stopped(p, ret == 2,
1150                                                            stat_addr, ru);
1151                                 if (retval != 0) /* He released the lock.  */
1152                                         goto end_wait4;
1153                                 break;
1154                         case TASK_ZOMBIE:
1155                                 /*
1156                                  * Eligible but we cannot release it yet:
1157                                  */
1158                                 if (ret == 2)
1159                                         continue;
1160                                 retval = wait_task_zombie(p, stat_addr, ru);
1161                                 if (retval != 0) /* He released the lock.  */
1162                                         goto end_wait4;
1163                                 break;
1164                         }
1165                 }
1166                 if (!flag) {
1167                         list_for_each (_p,&tsk->ptrace_children) {
1168                                 p = list_entry(_p,struct task_struct,ptrace_list);
1169                                 if (!eligible_child(pid, options, p))
1170                                         continue;
1171                                 flag = 1;
1172                                 break;
1173                         }
1174                 }
1175                 if (options & __WNOTHREAD)
1176                         break;
1177                 tsk = next_thread(tsk);
1178                 if (tsk->signal != current->signal)
1179                         BUG();
1180         } while (tsk != current);
1181         read_unlock(&tasklist_lock);
1182         if (flag) {
1183                 retval = 0;
1184                 if (options & WNOHANG)
1185                         goto end_wait4;
1186                 retval = -ERESTARTSYS;
1187                 if (signal_pending(current))
1188                         goto end_wait4;
1189                 schedule();
1190                 goto repeat;
1191         }
1192         retval = -ECHILD;
1193 end_wait4:
1194         current->state = TASK_RUNNING;
1195         remove_wait_queue(&current->wait_chldexit,&wait);
1196         return retval;
1197 }
1198
1199 #ifdef __ARCH_WANT_SYS_WAITPID
1200
1201 /*
1202  * sys_waitpid() remains for compatibility. waitpid() should be
1203  * implemented by calling sys_wait4() from libc.a.
1204  */
1205 asmlinkage long sys_waitpid(pid_t pid, unsigned __user *stat_addr, int options)
1206 {
1207         return sys_wait4(pid, stat_addr, options, NULL);
1208 }
1209
1210 #endif