asmlinkage_protect replaces prevent_tail_call
[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/mm.h>
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/mnt_namespace.h>
16 #include <linux/key.h>
17 #include <linux/security.h>
18 #include <linux/cpu.h>
19 #include <linux/acct.h>
20 #include <linux/tsacct_kern.h>
21 #include <linux/file.h>
22 #include <linux/binfmts.h>
23 #include <linux/nsproxy.h>
24 #include <linux/pid_namespace.h>
25 #include <linux/ptrace.h>
26 #include <linux/profile.h>
27 #include <linux/mount.h>
28 #include <linux/proc_fs.h>
29 #include <linux/kthread.h>
30 #include <linux/mempolicy.h>
31 #include <linux/taskstats_kern.h>
32 #include <linux/delayacct.h>
33 #include <linux/freezer.h>
34 #include <linux/cgroup.h>
35 #include <linux/syscalls.h>
36 #include <linux/signal.h>
37 #include <linux/posix-timers.h>
38 #include <linux/cn_proc.h>
39 #include <linux/mutex.h>
40 #include <linux/futex.h>
41 #include <linux/compat.h>
42 #include <linux/pipe_fs_i.h>
43 #include <linux/audit.h> /* for audit_free() */
44 #include <linux/resource.h>
45 #include <linux/blkdev.h>
46 #include <linux/task_io_accounting_ops.h>
47
48 #include <asm/uaccess.h>
49 #include <asm/unistd.h>
50 #include <asm/pgtable.h>
51 #include <asm/mmu_context.h>
52
53 static void exit_mm(struct task_struct * tsk);
54
55 static void __unhash_process(struct task_struct *p)
56 {
57         nr_threads--;
58         detach_pid(p, PIDTYPE_PID);
59         if (thread_group_leader(p)) {
60                 detach_pid(p, PIDTYPE_PGID);
61                 detach_pid(p, PIDTYPE_SID);
62
63                 list_del_rcu(&p->tasks);
64                 __get_cpu_var(process_counts)--;
65         }
66         list_del_rcu(&p->thread_group);
67         remove_parent(p);
68 }
69
70 /*
71  * This function expects the tasklist_lock write-locked.
72  */
73 static void __exit_signal(struct task_struct *tsk)
74 {
75         struct signal_struct *sig = tsk->signal;
76         struct sighand_struct *sighand;
77
78         BUG_ON(!sig);
79         BUG_ON(!atomic_read(&sig->count));
80
81         rcu_read_lock();
82         sighand = rcu_dereference(tsk->sighand);
83         spin_lock(&sighand->siglock);
84
85         posix_cpu_timers_exit(tsk);
86         if (atomic_dec_and_test(&sig->count))
87                 posix_cpu_timers_exit_group(tsk);
88         else {
89                 /*
90                  * If there is any task waiting for the group exit
91                  * then notify it:
92                  */
93                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
94                         wake_up_process(sig->group_exit_task);
95
96                 if (tsk == sig->curr_target)
97                         sig->curr_target = next_thread(tsk);
98                 /*
99                  * Accumulate here the counters for all threads but the
100                  * group leader as they die, so they can be added into
101                  * the process-wide totals when those are taken.
102                  * The group leader stays around as a zombie as long
103                  * as there are other threads.  When it gets reaped,
104                  * the exit.c code will add its counts into these totals.
105                  * We won't ever get here for the group leader, since it
106                  * will have been the last reference on the signal_struct.
107                  */
108                 sig->utime = cputime_add(sig->utime, tsk->utime);
109                 sig->stime = cputime_add(sig->stime, tsk->stime);
110                 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
111                 sig->min_flt += tsk->min_flt;
112                 sig->maj_flt += tsk->maj_flt;
113                 sig->nvcsw += tsk->nvcsw;
114                 sig->nivcsw += tsk->nivcsw;
115                 sig->inblock += task_io_get_inblock(tsk);
116                 sig->oublock += task_io_get_oublock(tsk);
117                 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
118                 sig = NULL; /* Marker for below. */
119         }
120
121         __unhash_process(tsk);
122
123         tsk->signal = NULL;
124         tsk->sighand = NULL;
125         spin_unlock(&sighand->siglock);
126         rcu_read_unlock();
127
128         __cleanup_sighand(sighand);
129         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
130         flush_sigqueue(&tsk->pending);
131         if (sig) {
132                 flush_sigqueue(&sig->shared_pending);
133                 taskstats_tgid_free(sig);
134                 __cleanup_signal(sig);
135         }
136 }
137
138 static void delayed_put_task_struct(struct rcu_head *rhp)
139 {
140         put_task_struct(container_of(rhp, struct task_struct, rcu));
141 }
142
143 void release_task(struct task_struct * p)
144 {
145         struct task_struct *leader;
146         int zap_leader;
147 repeat:
148         atomic_dec(&p->user->processes);
149         proc_flush_task(p);
150         write_lock_irq(&tasklist_lock);
151         ptrace_unlink(p);
152         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
153         __exit_signal(p);
154
155         /*
156          * If we are the last non-leader member of the thread
157          * group, and the leader is zombie, then notify the
158          * group leader's parent process. (if it wants notification.)
159          */
160         zap_leader = 0;
161         leader = p->group_leader;
162         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
163                 BUG_ON(leader->exit_signal == -1);
164                 do_notify_parent(leader, leader->exit_signal);
165                 /*
166                  * If we were the last child thread and the leader has
167                  * exited already, and the leader's parent ignores SIGCHLD,
168                  * then we are the one who should release the leader.
169                  *
170                  * do_notify_parent() will have marked it self-reaping in
171                  * that case.
172                  */
173                 zap_leader = (leader->exit_signal == -1);
174         }
175
176         write_unlock_irq(&tasklist_lock);
177         release_thread(p);
178         call_rcu(&p->rcu, delayed_put_task_struct);
179
180         p = leader;
181         if (unlikely(zap_leader))
182                 goto repeat;
183 }
184
185 /*
186  * This checks not only the pgrp, but falls back on the pid if no
187  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
188  * without this...
189  *
190  * The caller must hold rcu lock or the tasklist lock.
191  */
192 struct pid *session_of_pgrp(struct pid *pgrp)
193 {
194         struct task_struct *p;
195         struct pid *sid = NULL;
196
197         p = pid_task(pgrp, PIDTYPE_PGID);
198         if (p == NULL)
199                 p = pid_task(pgrp, PIDTYPE_PID);
200         if (p != NULL)
201                 sid = task_session(p);
202
203         return sid;
204 }
205
206 /*
207  * Determine if a process group is "orphaned", according to the POSIX
208  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
209  * by terminal-generated stop signals.  Newly orphaned process groups are
210  * to receive a SIGHUP and a SIGCONT.
211  *
212  * "I ask you, have you ever known what it is to be an orphan?"
213  */
214 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
215 {
216         struct task_struct *p;
217
218         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
219                 if ((p == ignored_task) ||
220                     (p->exit_state && thread_group_empty(p)) ||
221                     is_global_init(p->real_parent))
222                         continue;
223
224                 if (task_pgrp(p->real_parent) != pgrp &&
225                     task_session(p->real_parent) == task_session(p))
226                         return 0;
227         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
228
229         return 1;
230 }
231
232 int is_current_pgrp_orphaned(void)
233 {
234         int retval;
235
236         read_lock(&tasklist_lock);
237         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
238         read_unlock(&tasklist_lock);
239
240         return retval;
241 }
242
243 static int has_stopped_jobs(struct pid *pgrp)
244 {
245         int retval = 0;
246         struct task_struct *p;
247
248         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
249                 if (!task_is_stopped(p))
250                         continue;
251                 retval = 1;
252                 break;
253         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
254         return retval;
255 }
256
257 /*
258  * Check to see if any process groups have become orphaned as
259  * a result of our exiting, and if they have any stopped jobs,
260  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
261  */
262 static void
263 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
264 {
265         struct pid *pgrp = task_pgrp(tsk);
266         struct task_struct *ignored_task = tsk;
267
268         if (!parent)
269                  /* exit: our father is in a different pgrp than
270                   * we are and we were the only connection outside.
271                   */
272                 parent = tsk->real_parent;
273         else
274                 /* reparent: our child is in a different pgrp than
275                  * we are, and it was the only connection outside.
276                  */
277                 ignored_task = NULL;
278
279         if (task_pgrp(parent) != pgrp &&
280             task_session(parent) == task_session(tsk) &&
281             will_become_orphaned_pgrp(pgrp, ignored_task) &&
282             has_stopped_jobs(pgrp)) {
283                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
284                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
285         }
286 }
287
288 /**
289  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
290  *
291  * If a kernel thread is launched as a result of a system call, or if
292  * it ever exits, it should generally reparent itself to kthreadd so it
293  * isn't in the way of other processes and is correctly cleaned up on exit.
294  *
295  * The various task state such as scheduling policy and priority may have
296  * been inherited from a user process, so we reset them to sane values here.
297  *
298  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
299  */
300 static void reparent_to_kthreadd(void)
301 {
302         write_lock_irq(&tasklist_lock);
303
304         ptrace_unlink(current);
305         /* Reparent to init */
306         remove_parent(current);
307         current->real_parent = current->parent = kthreadd_task;
308         add_parent(current);
309
310         /* Set the exit signal to SIGCHLD so we signal init on exit */
311         current->exit_signal = SIGCHLD;
312
313         if (task_nice(current) < 0)
314                 set_user_nice(current, 0);
315         /* cpus_allowed? */
316         /* rt_priority? */
317         /* signals? */
318         security_task_reparent_to_init(current);
319         memcpy(current->signal->rlim, init_task.signal->rlim,
320                sizeof(current->signal->rlim));
321         atomic_inc(&(INIT_USER->__count));
322         write_unlock_irq(&tasklist_lock);
323         switch_uid(INIT_USER);
324 }
325
326 void __set_special_pids(struct pid *pid)
327 {
328         struct task_struct *curr = current->group_leader;
329         pid_t nr = pid_nr(pid);
330
331         if (task_session(curr) != pid) {
332                 detach_pid(curr, PIDTYPE_SID);
333                 attach_pid(curr, PIDTYPE_SID, pid);
334                 set_task_session(curr, nr);
335         }
336         if (task_pgrp(curr) != pid) {
337                 detach_pid(curr, PIDTYPE_PGID);
338                 attach_pid(curr, PIDTYPE_PGID, pid);
339                 set_task_pgrp(curr, nr);
340         }
341 }
342
343 static void set_special_pids(struct pid *pid)
344 {
345         write_lock_irq(&tasklist_lock);
346         __set_special_pids(pid);
347         write_unlock_irq(&tasklist_lock);
348 }
349
350 /*
351  * Let kernel threads use this to say that they
352  * allow a certain signal (since daemonize() will
353  * have disabled all of them by default).
354  */
355 int allow_signal(int sig)
356 {
357         if (!valid_signal(sig) || sig < 1)
358                 return -EINVAL;
359
360         spin_lock_irq(&current->sighand->siglock);
361         sigdelset(&current->blocked, sig);
362         if (!current->mm) {
363                 /* Kernel threads handle their own signals.
364                    Let the signal code know it'll be handled, so
365                    that they don't get converted to SIGKILL or
366                    just silently dropped */
367                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
368         }
369         recalc_sigpending();
370         spin_unlock_irq(&current->sighand->siglock);
371         return 0;
372 }
373
374 EXPORT_SYMBOL(allow_signal);
375
376 int disallow_signal(int sig)
377 {
378         if (!valid_signal(sig) || sig < 1)
379                 return -EINVAL;
380
381         spin_lock_irq(&current->sighand->siglock);
382         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
383         recalc_sigpending();
384         spin_unlock_irq(&current->sighand->siglock);
385         return 0;
386 }
387
388 EXPORT_SYMBOL(disallow_signal);
389
390 /*
391  *      Put all the gunge required to become a kernel thread without
392  *      attached user resources in one place where it belongs.
393  */
394
395 void daemonize(const char *name, ...)
396 {
397         va_list args;
398         struct fs_struct *fs;
399         sigset_t blocked;
400
401         va_start(args, name);
402         vsnprintf(current->comm, sizeof(current->comm), name, args);
403         va_end(args);
404
405         /*
406          * If we were started as result of loading a module, close all of the
407          * user space pages.  We don't need them, and if we didn't close them
408          * they would be locked into memory.
409          */
410         exit_mm(current);
411         /*
412          * We don't want to have TIF_FREEZE set if the system-wide hibernation
413          * or suspend transition begins right now.
414          */
415         current->flags |= PF_NOFREEZE;
416
417         if (current->nsproxy != &init_nsproxy) {
418                 get_nsproxy(&init_nsproxy);
419                 switch_task_namespaces(current, &init_nsproxy);
420         }
421         set_special_pids(&init_struct_pid);
422         proc_clear_tty(current);
423
424         /* Block and flush all signals */
425         sigfillset(&blocked);
426         sigprocmask(SIG_BLOCK, &blocked, NULL);
427         flush_signals(current);
428
429         /* Become as one with the init task */
430
431         exit_fs(current);       /* current->fs->count--; */
432         fs = init_task.fs;
433         current->fs = fs;
434         atomic_inc(&fs->count);
435
436         exit_files(current);
437         current->files = init_task.files;
438         atomic_inc(&current->files->count);
439
440         reparent_to_kthreadd();
441 }
442
443 EXPORT_SYMBOL(daemonize);
444
445 static void close_files(struct files_struct * files)
446 {
447         int i, j;
448         struct fdtable *fdt;
449
450         j = 0;
451
452         /*
453          * It is safe to dereference the fd table without RCU or
454          * ->file_lock because this is the last reference to the
455          * files structure.
456          */
457         fdt = files_fdtable(files);
458         for (;;) {
459                 unsigned long set;
460                 i = j * __NFDBITS;
461                 if (i >= fdt->max_fds)
462                         break;
463                 set = fdt->open_fds->fds_bits[j++];
464                 while (set) {
465                         if (set & 1) {
466                                 struct file * file = xchg(&fdt->fd[i], NULL);
467                                 if (file) {
468                                         filp_close(file, files);
469                                         cond_resched();
470                                 }
471                         }
472                         i++;
473                         set >>= 1;
474                 }
475         }
476 }
477
478 struct files_struct *get_files_struct(struct task_struct *task)
479 {
480         struct files_struct *files;
481
482         task_lock(task);
483         files = task->files;
484         if (files)
485                 atomic_inc(&files->count);
486         task_unlock(task);
487
488         return files;
489 }
490
491 void put_files_struct(struct files_struct *files)
492 {
493         struct fdtable *fdt;
494
495         if (atomic_dec_and_test(&files->count)) {
496                 close_files(files);
497                 /*
498                  * Free the fd and fdset arrays if we expanded them.
499                  * If the fdtable was embedded, pass files for freeing
500                  * at the end of the RCU grace period. Otherwise,
501                  * you can free files immediately.
502                  */
503                 fdt = files_fdtable(files);
504                 if (fdt != &files->fdtab)
505                         kmem_cache_free(files_cachep, files);
506                 free_fdtable(fdt);
507         }
508 }
509
510 EXPORT_SYMBOL(put_files_struct);
511
512 void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
513 {
514         struct files_struct *old;
515
516         old = tsk->files;
517         task_lock(tsk);
518         tsk->files = files;
519         task_unlock(tsk);
520         put_files_struct(old);
521 }
522 EXPORT_SYMBOL(reset_files_struct);
523
524 static void __exit_files(struct task_struct *tsk)
525 {
526         struct files_struct * files = tsk->files;
527
528         if (files) {
529                 task_lock(tsk);
530                 tsk->files = NULL;
531                 task_unlock(tsk);
532                 put_files_struct(files);
533         }
534 }
535
536 void exit_files(struct task_struct *tsk)
537 {
538         __exit_files(tsk);
539 }
540
541 static void __put_fs_struct(struct fs_struct *fs)
542 {
543         /* No need to hold fs->lock if we are killing it */
544         if (atomic_dec_and_test(&fs->count)) {
545                 path_put(&fs->root);
546                 path_put(&fs->pwd);
547                 if (fs->altroot.dentry)
548                         path_put(&fs->altroot);
549                 kmem_cache_free(fs_cachep, fs);
550         }
551 }
552
553 void put_fs_struct(struct fs_struct *fs)
554 {
555         __put_fs_struct(fs);
556 }
557
558 static void __exit_fs(struct task_struct *tsk)
559 {
560         struct fs_struct * fs = tsk->fs;
561
562         if (fs) {
563                 task_lock(tsk);
564                 tsk->fs = NULL;
565                 task_unlock(tsk);
566                 __put_fs_struct(fs);
567         }
568 }
569
570 void exit_fs(struct task_struct *tsk)
571 {
572         __exit_fs(tsk);
573 }
574
575 EXPORT_SYMBOL_GPL(exit_fs);
576
577 /*
578  * Turn us into a lazy TLB process if we
579  * aren't already..
580  */
581 static void exit_mm(struct task_struct * tsk)
582 {
583         struct mm_struct *mm = tsk->mm;
584
585         mm_release(tsk, mm);
586         if (!mm)
587                 return;
588         /*
589          * Serialize with any possible pending coredump.
590          * We must hold mmap_sem around checking core_waiters
591          * and clearing tsk->mm.  The core-inducing thread
592          * will increment core_waiters for each thread in the
593          * group with ->mm != NULL.
594          */
595         down_read(&mm->mmap_sem);
596         if (mm->core_waiters) {
597                 up_read(&mm->mmap_sem);
598                 down_write(&mm->mmap_sem);
599                 if (!--mm->core_waiters)
600                         complete(mm->core_startup_done);
601                 up_write(&mm->mmap_sem);
602
603                 wait_for_completion(&mm->core_done);
604                 down_read(&mm->mmap_sem);
605         }
606         atomic_inc(&mm->mm_count);
607         BUG_ON(mm != tsk->active_mm);
608         /* more a memory barrier than a real lock */
609         task_lock(tsk);
610         tsk->mm = NULL;
611         up_read(&mm->mmap_sem);
612         enter_lazy_tlb(mm, current);
613         /* We don't want this task to be frozen prematurely */
614         clear_freeze_flag(tsk);
615         task_unlock(tsk);
616         mmput(mm);
617 }
618
619 static void
620 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
621 {
622         if (p->pdeath_signal)
623                 /* We already hold the tasklist_lock here.  */
624                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
625
626         /* Move the child from its dying parent to the new one.  */
627         if (unlikely(traced)) {
628                 /* Preserve ptrace links if someone else is tracing this child.  */
629                 list_del_init(&p->ptrace_list);
630                 if (p->parent != p->real_parent)
631                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
632         } else {
633                 /* If this child is being traced, then we're the one tracing it
634                  * anyway, so let go of it.
635                  */
636                 p->ptrace = 0;
637                 remove_parent(p);
638                 p->parent = p->real_parent;
639                 add_parent(p);
640
641                 if (task_is_traced(p)) {
642                         /*
643                          * If it was at a trace stop, turn it into
644                          * a normal stop since it's no longer being
645                          * traced.
646                          */
647                         ptrace_untrace(p);
648                 }
649         }
650
651         /* If this is a threaded reparent there is no need to
652          * notify anyone anything has happened.
653          */
654         if (p->real_parent->group_leader == father->group_leader)
655                 return;
656
657         /* We don't want people slaying init.  */
658         if (p->exit_signal != -1)
659                 p->exit_signal = SIGCHLD;
660
661         /* If we'd notified the old parent about this child's death,
662          * also notify the new parent.
663          */
664         if (!traced && p->exit_state == EXIT_ZOMBIE &&
665             p->exit_signal != -1 && thread_group_empty(p))
666                 do_notify_parent(p, p->exit_signal);
667
668         kill_orphaned_pgrp(p, father);
669 }
670
671 /*
672  * When we die, we re-parent all our children.
673  * Try to give them to another thread in our thread
674  * group, and if no such member exists, give it to
675  * the child reaper process (ie "init") in our pid
676  * space.
677  */
678 static void forget_original_parent(struct task_struct *father)
679 {
680         struct task_struct *p, *n, *reaper = father;
681         struct list_head ptrace_dead;
682
683         INIT_LIST_HEAD(&ptrace_dead);
684
685         write_lock_irq(&tasklist_lock);
686
687         do {
688                 reaper = next_thread(reaper);
689                 if (reaper == father) {
690                         reaper = task_child_reaper(father);
691                         break;
692                 }
693         } while (reaper->flags & PF_EXITING);
694
695         /*
696          * There are only two places where our children can be:
697          *
698          * - in our child list
699          * - in our ptraced child list
700          *
701          * Search them and reparent children.
702          */
703         list_for_each_entry_safe(p, n, &father->children, sibling) {
704                 int ptrace;
705
706                 ptrace = p->ptrace;
707
708                 /* if father isn't the real parent, then ptrace must be enabled */
709                 BUG_ON(father != p->real_parent && !ptrace);
710
711                 if (father == p->real_parent) {
712                         /* reparent with a reaper, real father it's us */
713                         p->real_parent = reaper;
714                         reparent_thread(p, father, 0);
715                 } else {
716                         /* reparent ptraced task to its real parent */
717                         __ptrace_unlink (p);
718                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
719                             thread_group_empty(p))
720                                 do_notify_parent(p, p->exit_signal);
721                 }
722
723                 /*
724                  * if the ptraced child is a zombie with exit_signal == -1
725                  * we must collect it before we exit, or it will remain
726                  * zombie forever since we prevented it from self-reap itself
727                  * while it was being traced by us, to be able to see it in wait4.
728                  */
729                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
730                         list_add(&p->ptrace_list, &ptrace_dead);
731         }
732
733         list_for_each_entry_safe(p, n, &father->ptrace_children, ptrace_list) {
734                 p->real_parent = reaper;
735                 reparent_thread(p, father, 1);
736         }
737
738         write_unlock_irq(&tasklist_lock);
739         BUG_ON(!list_empty(&father->children));
740         BUG_ON(!list_empty(&father->ptrace_children));
741
742         list_for_each_entry_safe(p, n, &ptrace_dead, ptrace_list) {
743                 list_del_init(&p->ptrace_list);
744                 release_task(p);
745         }
746
747 }
748
749 /*
750  * Send signals to all our closest relatives so that they know
751  * to properly mourn us..
752  */
753 static void exit_notify(struct task_struct *tsk, int group_dead)
754 {
755         int state;
756
757         /*
758          * This does two things:
759          *
760          * A.  Make init inherit all the child processes
761          * B.  Check to see if any process groups have become orphaned
762          *      as a result of our exiting, and if they have any stopped
763          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
764          */
765         forget_original_parent(tsk);
766         exit_task_namespaces(tsk);
767
768         write_lock_irq(&tasklist_lock);
769         if (group_dead)
770                 kill_orphaned_pgrp(tsk->group_leader, NULL);
771
772         /* Let father know we died
773          *
774          * Thread signals are configurable, but you aren't going to use
775          * that to send signals to arbitary processes.
776          * That stops right now.
777          *
778          * If the parent exec id doesn't match the exec id we saved
779          * when we started then we know the parent has changed security
780          * domain.
781          *
782          * If our self_exec id doesn't match our parent_exec_id then
783          * we have changed execution domain as these two values started
784          * the same after a fork.
785          */
786         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
787             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
788              tsk->self_exec_id != tsk->parent_exec_id)
789             && !capable(CAP_KILL))
790                 tsk->exit_signal = SIGCHLD;
791
792
793         /* If something other than our normal parent is ptracing us, then
794          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
795          * only has special meaning to our real parent.
796          */
797         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
798                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
799                 do_notify_parent(tsk, signal);
800         } else if (tsk->ptrace) {
801                 do_notify_parent(tsk, SIGCHLD);
802         }
803
804         state = EXIT_ZOMBIE;
805         if (tsk->exit_signal == -1 && likely(!tsk->ptrace))
806                 state = EXIT_DEAD;
807         tsk->exit_state = state;
808
809         if (thread_group_leader(tsk) &&
810             tsk->signal->notify_count < 0 &&
811             tsk->signal->group_exit_task)
812                 wake_up_process(tsk->signal->group_exit_task);
813
814         write_unlock_irq(&tasklist_lock);
815
816         /* If the process is dead, release it - nobody will wait for it */
817         if (state == EXIT_DEAD)
818                 release_task(tsk);
819 }
820
821 #ifdef CONFIG_DEBUG_STACK_USAGE
822 static void check_stack_usage(void)
823 {
824         static DEFINE_SPINLOCK(low_water_lock);
825         static int lowest_to_date = THREAD_SIZE;
826         unsigned long *n = end_of_stack(current);
827         unsigned long free;
828
829         while (*n == 0)
830                 n++;
831         free = (unsigned long)n - (unsigned long)end_of_stack(current);
832
833         if (free >= lowest_to_date)
834                 return;
835
836         spin_lock(&low_water_lock);
837         if (free < lowest_to_date) {
838                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
839                                 "left\n",
840                                 current->comm, free);
841                 lowest_to_date = free;
842         }
843         spin_unlock(&low_water_lock);
844 }
845 #else
846 static inline void check_stack_usage(void) {}
847 #endif
848
849 static inline void exit_child_reaper(struct task_struct *tsk)
850 {
851         if (likely(tsk->group_leader != task_child_reaper(tsk)))
852                 return;
853
854         if (tsk->nsproxy->pid_ns == &init_pid_ns)
855                 panic("Attempted to kill init!");
856
857         /*
858          * @tsk is the last thread in the 'cgroup-init' and is exiting.
859          * Terminate all remaining processes in the namespace and reap them
860          * before exiting @tsk.
861          *
862          * Note that @tsk (last thread of cgroup-init) may not necessarily
863          * be the child-reaper (i.e main thread of cgroup-init) of the
864          * namespace i.e the child_reaper may have already exited.
865          *
866          * Even after a child_reaper exits, we let it inherit orphaned children,
867          * because, pid_ns->child_reaper remains valid as long as there is
868          * at least one living sub-thread in the cgroup init.
869
870          * This living sub-thread of the cgroup-init will be notified when
871          * a child inherited by the 'child-reaper' exits (do_notify_parent()
872          * uses __group_send_sig_info()). Further, when reaping child processes,
873          * do_wait() iterates over children of all living sub threads.
874
875          * i.e even though 'child_reaper' thread is listed as the parent of the
876          * orphaned children, any living sub-thread in the cgroup-init can
877          * perform the role of the child_reaper.
878          */
879         zap_pid_ns_processes(tsk->nsproxy->pid_ns);
880 }
881
882 NORET_TYPE void do_exit(long code)
883 {
884         struct task_struct *tsk = current;
885         int group_dead;
886
887         profile_task_exit(tsk);
888
889         WARN_ON(atomic_read(&tsk->fs_excl));
890
891         if (unlikely(in_interrupt()))
892                 panic("Aiee, killing interrupt handler!");
893         if (unlikely(!tsk->pid))
894                 panic("Attempted to kill the idle task!");
895
896         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
897                 current->ptrace_message = code;
898                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
899         }
900
901         /*
902          * We're taking recursive faults here in do_exit. Safest is to just
903          * leave this task alone and wait for reboot.
904          */
905         if (unlikely(tsk->flags & PF_EXITING)) {
906                 printk(KERN_ALERT
907                         "Fixing recursive fault but reboot is needed!\n");
908                 /*
909                  * We can do this unlocked here. The futex code uses
910                  * this flag just to verify whether the pi state
911                  * cleanup has been done or not. In the worst case it
912                  * loops once more. We pretend that the cleanup was
913                  * done as there is no way to return. Either the
914                  * OWNER_DIED bit is set by now or we push the blocked
915                  * task into the wait for ever nirwana as well.
916                  */
917                 tsk->flags |= PF_EXITPIDONE;
918                 if (tsk->io_context)
919                         exit_io_context();
920                 set_current_state(TASK_UNINTERRUPTIBLE);
921                 schedule();
922         }
923
924         exit_signals(tsk);  /* sets PF_EXITING */
925         /*
926          * tsk->flags are checked in the futex code to protect against
927          * an exiting task cleaning up the robust pi futexes.
928          */
929         smp_mb();
930         spin_unlock_wait(&tsk->pi_lock);
931
932         if (unlikely(in_atomic()))
933                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
934                                 current->comm, task_pid_nr(current),
935                                 preempt_count());
936
937         acct_update_integrals(tsk);
938         if (tsk->mm) {
939                 update_hiwater_rss(tsk->mm);
940                 update_hiwater_vm(tsk->mm);
941         }
942         group_dead = atomic_dec_and_test(&tsk->signal->live);
943         if (group_dead) {
944                 exit_child_reaper(tsk);
945                 hrtimer_cancel(&tsk->signal->real_timer);
946                 exit_itimers(tsk->signal);
947         }
948         acct_collect(code, group_dead);
949 #ifdef CONFIG_FUTEX
950         if (unlikely(tsk->robust_list))
951                 exit_robust_list(tsk);
952 #ifdef CONFIG_COMPAT
953         if (unlikely(tsk->compat_robust_list))
954                 compat_exit_robust_list(tsk);
955 #endif
956 #endif
957         if (group_dead)
958                 tty_audit_exit();
959         if (unlikely(tsk->audit_context))
960                 audit_free(tsk);
961
962         tsk->exit_code = code;
963         taskstats_exit(tsk, group_dead);
964
965         exit_mm(tsk);
966
967         if (group_dead)
968                 acct_process();
969         exit_sem(tsk);
970         __exit_files(tsk);
971         __exit_fs(tsk);
972         check_stack_usage();
973         exit_thread();
974         cgroup_exit(tsk, 1);
975         exit_keys(tsk);
976
977         if (group_dead && tsk->signal->leader)
978                 disassociate_ctty(1);
979
980         module_put(task_thread_info(tsk)->exec_domain->module);
981         if (tsk->binfmt)
982                 module_put(tsk->binfmt->module);
983
984         proc_exit_connector(tsk);
985         exit_notify(tsk, group_dead);
986 #ifdef CONFIG_NUMA
987         mpol_free(tsk->mempolicy);
988         tsk->mempolicy = NULL;
989 #endif
990 #ifdef CONFIG_FUTEX
991         /*
992          * This must happen late, after the PID is not
993          * hashed anymore:
994          */
995         if (unlikely(!list_empty(&tsk->pi_state_list)))
996                 exit_pi_state_list(tsk);
997         if (unlikely(current->pi_state_cache))
998                 kfree(current->pi_state_cache);
999 #endif
1000         /*
1001          * Make sure we are holding no locks:
1002          */
1003         debug_check_no_locks_held(tsk);
1004         /*
1005          * We can do this unlocked here. The futex code uses this flag
1006          * just to verify whether the pi state cleanup has been done
1007          * or not. In the worst case it loops once more.
1008          */
1009         tsk->flags |= PF_EXITPIDONE;
1010
1011         if (tsk->io_context)
1012                 exit_io_context();
1013
1014         if (tsk->splice_pipe)
1015                 __free_pipe_info(tsk->splice_pipe);
1016
1017         preempt_disable();
1018         /* causes final put_task_struct in finish_task_switch(). */
1019         tsk->state = TASK_DEAD;
1020
1021         schedule();
1022         BUG();
1023         /* Avoid "noreturn function does return".  */
1024         for (;;)
1025                 cpu_relax();    /* For when BUG is null */
1026 }
1027
1028 EXPORT_SYMBOL_GPL(do_exit);
1029
1030 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1031 {
1032         if (comp)
1033                 complete(comp);
1034
1035         do_exit(code);
1036 }
1037
1038 EXPORT_SYMBOL(complete_and_exit);
1039
1040 asmlinkage long sys_exit(int error_code)
1041 {
1042         do_exit((error_code&0xff)<<8);
1043 }
1044
1045 /*
1046  * Take down every thread in the group.  This is called by fatal signals
1047  * as well as by sys_exit_group (below).
1048  */
1049 NORET_TYPE void
1050 do_group_exit(int exit_code)
1051 {
1052         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1053
1054         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1055                 exit_code = current->signal->group_exit_code;
1056         else if (!thread_group_empty(current)) {
1057                 struct signal_struct *const sig = current->signal;
1058                 struct sighand_struct *const sighand = current->sighand;
1059                 spin_lock_irq(&sighand->siglock);
1060                 if (signal_group_exit(sig))
1061                         /* Another thread got here before we took the lock.  */
1062                         exit_code = sig->group_exit_code;
1063                 else {
1064                         sig->group_exit_code = exit_code;
1065                         sig->flags = SIGNAL_GROUP_EXIT;
1066                         zap_other_threads(current);
1067                 }
1068                 spin_unlock_irq(&sighand->siglock);
1069         }
1070
1071         do_exit(exit_code);
1072         /* NOTREACHED */
1073 }
1074
1075 /*
1076  * this kills every thread in the thread group. Note that any externally
1077  * wait4()-ing process will get the correct exit code - even if this
1078  * thread is not the thread group leader.
1079  */
1080 asmlinkage void sys_exit_group(int error_code)
1081 {
1082         do_group_exit((error_code & 0xff) << 8);
1083 }
1084
1085 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1086 {
1087         struct pid *pid = NULL;
1088         if (type == PIDTYPE_PID)
1089                 pid = task->pids[type].pid;
1090         else if (type < PIDTYPE_MAX)
1091                 pid = task->group_leader->pids[type].pid;
1092         return pid;
1093 }
1094
1095 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1096                           struct task_struct *p)
1097 {
1098         int err;
1099
1100         if (type < PIDTYPE_MAX) {
1101                 if (task_pid_type(p, type) != pid)
1102                         return 0;
1103         }
1104
1105         /*
1106          * Do not consider detached threads that are
1107          * not ptraced:
1108          */
1109         if (p->exit_signal == -1 && !p->ptrace)
1110                 return 0;
1111
1112         /* Wait for all children (clone and not) if __WALL is set;
1113          * otherwise, wait for clone children *only* if __WCLONE is
1114          * set; otherwise, wait for non-clone children *only*.  (Note:
1115          * A "clone" child here is one that reports to its parent
1116          * using a signal other than SIGCHLD.) */
1117         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1118             && !(options & __WALL))
1119                 return 0;
1120
1121         err = security_task_wait(p);
1122         if (likely(!err))
1123                 return 1;
1124
1125         if (type != PIDTYPE_PID)
1126                 return 0;
1127         /* This child was explicitly requested, abort */
1128         read_unlock(&tasklist_lock);
1129         return err;
1130 }
1131
1132 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1133                                int why, int status,
1134                                struct siginfo __user *infop,
1135                                struct rusage __user *rusagep)
1136 {
1137         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1138
1139         put_task_struct(p);
1140         if (!retval)
1141                 retval = put_user(SIGCHLD, &infop->si_signo);
1142         if (!retval)
1143                 retval = put_user(0, &infop->si_errno);
1144         if (!retval)
1145                 retval = put_user((short)why, &infop->si_code);
1146         if (!retval)
1147                 retval = put_user(pid, &infop->si_pid);
1148         if (!retval)
1149                 retval = put_user(uid, &infop->si_uid);
1150         if (!retval)
1151                 retval = put_user(status, &infop->si_status);
1152         if (!retval)
1153                 retval = pid;
1154         return retval;
1155 }
1156
1157 /*
1158  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1159  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1160  * the lock and this task is uninteresting.  If we return nonzero, we have
1161  * released the lock and the system call should return.
1162  */
1163 static int wait_task_zombie(struct task_struct *p, int noreap,
1164                             struct siginfo __user *infop,
1165                             int __user *stat_addr, struct rusage __user *ru)
1166 {
1167         unsigned long state;
1168         int retval, status, traced;
1169         pid_t pid = task_pid_vnr(p);
1170
1171         if (unlikely(noreap)) {
1172                 uid_t uid = p->uid;
1173                 int exit_code = p->exit_code;
1174                 int why, status;
1175
1176                 get_task_struct(p);
1177                 read_unlock(&tasklist_lock);
1178                 if ((exit_code & 0x7f) == 0) {
1179                         why = CLD_EXITED;
1180                         status = exit_code >> 8;
1181                 } else {
1182                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1183                         status = exit_code & 0x7f;
1184                 }
1185                 return wait_noreap_copyout(p, pid, uid, why,
1186                                            status, infop, ru);
1187         }
1188
1189         /*
1190          * Try to move the task's state to DEAD
1191          * only one thread is allowed to do this:
1192          */
1193         state = xchg(&p->exit_state, EXIT_DEAD);
1194         if (state != EXIT_ZOMBIE) {
1195                 BUG_ON(state != EXIT_DEAD);
1196                 return 0;
1197         }
1198
1199         /* traced means p->ptrace, but not vice versa */
1200         traced = (p->real_parent != p->parent);
1201
1202         if (likely(!traced)) {
1203                 struct signal_struct *psig;
1204                 struct signal_struct *sig;
1205
1206                 /*
1207                  * The resource counters for the group leader are in its
1208                  * own task_struct.  Those for dead threads in the group
1209                  * are in its signal_struct, as are those for the child
1210                  * processes it has previously reaped.  All these
1211                  * accumulate in the parent's signal_struct c* fields.
1212                  *
1213                  * We don't bother to take a lock here to protect these
1214                  * p->signal fields, because they are only touched by
1215                  * __exit_signal, which runs with tasklist_lock
1216                  * write-locked anyway, and so is excluded here.  We do
1217                  * need to protect the access to p->parent->signal fields,
1218                  * as other threads in the parent group can be right
1219                  * here reaping other children at the same time.
1220                  */
1221                 spin_lock_irq(&p->parent->sighand->siglock);
1222                 psig = p->parent->signal;
1223                 sig = p->signal;
1224                 psig->cutime =
1225                         cputime_add(psig->cutime,
1226                         cputime_add(p->utime,
1227                         cputime_add(sig->utime,
1228                                     sig->cutime)));
1229                 psig->cstime =
1230                         cputime_add(psig->cstime,
1231                         cputime_add(p->stime,
1232                         cputime_add(sig->stime,
1233                                     sig->cstime)));
1234                 psig->cgtime =
1235                         cputime_add(psig->cgtime,
1236                         cputime_add(p->gtime,
1237                         cputime_add(sig->gtime,
1238                                     sig->cgtime)));
1239                 psig->cmin_flt +=
1240                         p->min_flt + sig->min_flt + sig->cmin_flt;
1241                 psig->cmaj_flt +=
1242                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1243                 psig->cnvcsw +=
1244                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1245                 psig->cnivcsw +=
1246                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1247                 psig->cinblock +=
1248                         task_io_get_inblock(p) +
1249                         sig->inblock + sig->cinblock;
1250                 psig->coublock +=
1251                         task_io_get_oublock(p) +
1252                         sig->oublock + sig->coublock;
1253                 spin_unlock_irq(&p->parent->sighand->siglock);
1254         }
1255
1256         /*
1257          * Now we are sure this task is interesting, and no other
1258          * thread can reap it because we set its state to EXIT_DEAD.
1259          */
1260         read_unlock(&tasklist_lock);
1261
1262         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1263         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1264                 ? p->signal->group_exit_code : p->exit_code;
1265         if (!retval && stat_addr)
1266                 retval = put_user(status, stat_addr);
1267         if (!retval && infop)
1268                 retval = put_user(SIGCHLD, &infop->si_signo);
1269         if (!retval && infop)
1270                 retval = put_user(0, &infop->si_errno);
1271         if (!retval && infop) {
1272                 int why;
1273
1274                 if ((status & 0x7f) == 0) {
1275                         why = CLD_EXITED;
1276                         status >>= 8;
1277                 } else {
1278                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1279                         status &= 0x7f;
1280                 }
1281                 retval = put_user((short)why, &infop->si_code);
1282                 if (!retval)
1283                         retval = put_user(status, &infop->si_status);
1284         }
1285         if (!retval && infop)
1286                 retval = put_user(pid, &infop->si_pid);
1287         if (!retval && infop)
1288                 retval = put_user(p->uid, &infop->si_uid);
1289         if (!retval)
1290                 retval = pid;
1291
1292         if (traced) {
1293                 write_lock_irq(&tasklist_lock);
1294                 /* We dropped tasklist, ptracer could die and untrace */
1295                 ptrace_unlink(p);
1296                 /*
1297                  * If this is not a detached task, notify the parent.
1298                  * If it's still not detached after that, don't release
1299                  * it now.
1300                  */
1301                 if (p->exit_signal != -1) {
1302                         do_notify_parent(p, p->exit_signal);
1303                         if (p->exit_signal != -1) {
1304                                 p->exit_state = EXIT_ZOMBIE;
1305                                 p = NULL;
1306                         }
1307                 }
1308                 write_unlock_irq(&tasklist_lock);
1309         }
1310         if (p != NULL)
1311                 release_task(p);
1312
1313         return retval;
1314 }
1315
1316 /*
1317  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1318  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1319  * the lock and this task is uninteresting.  If we return nonzero, we have
1320  * released the lock and the system call should return.
1321  */
1322 static int wait_task_stopped(struct task_struct *p,
1323                              int noreap, struct siginfo __user *infop,
1324                              int __user *stat_addr, struct rusage __user *ru)
1325 {
1326         int retval, exit_code, why;
1327         uid_t uid = 0; /* unneeded, required by compiler */
1328         pid_t pid;
1329
1330         exit_code = 0;
1331         spin_lock_irq(&p->sighand->siglock);
1332
1333         if (unlikely(!task_is_stopped_or_traced(p)))
1334                 goto unlock_sig;
1335
1336         if (!(p->ptrace & PT_PTRACED) && p->signal->group_stop_count > 0)
1337                 /*
1338                  * A group stop is in progress and this is the group leader.
1339                  * We won't report until all threads have stopped.
1340                  */
1341                 goto unlock_sig;
1342
1343         exit_code = p->exit_code;
1344         if (!exit_code)
1345                 goto unlock_sig;
1346
1347         if (!noreap)
1348                 p->exit_code = 0;
1349
1350         uid = p->uid;
1351 unlock_sig:
1352         spin_unlock_irq(&p->sighand->siglock);
1353         if (!exit_code)
1354                 return 0;
1355
1356         /*
1357          * Now we are pretty sure this task is interesting.
1358          * Make sure it doesn't get reaped out from under us while we
1359          * give up the lock and then examine it below.  We don't want to
1360          * keep holding onto the tasklist_lock while we call getrusage and
1361          * possibly take page faults for user memory.
1362          */
1363         get_task_struct(p);
1364         pid = task_pid_vnr(p);
1365         why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1366         read_unlock(&tasklist_lock);
1367
1368         if (unlikely(noreap))
1369                 return wait_noreap_copyout(p, pid, uid,
1370                                            why, exit_code,
1371                                            infop, ru);
1372
1373         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1374         if (!retval && stat_addr)
1375                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1376         if (!retval && infop)
1377                 retval = put_user(SIGCHLD, &infop->si_signo);
1378         if (!retval && infop)
1379                 retval = put_user(0, &infop->si_errno);
1380         if (!retval && infop)
1381                 retval = put_user((short)why, &infop->si_code);
1382         if (!retval && infop)
1383                 retval = put_user(exit_code, &infop->si_status);
1384         if (!retval && infop)
1385                 retval = put_user(pid, &infop->si_pid);
1386         if (!retval && infop)
1387                 retval = put_user(uid, &infop->si_uid);
1388         if (!retval)
1389                 retval = pid;
1390         put_task_struct(p);
1391
1392         BUG_ON(!retval);
1393         return retval;
1394 }
1395
1396 /*
1397  * Handle do_wait work for one task in a live, non-stopped state.
1398  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1399  * the lock and this task is uninteresting.  If we return nonzero, we have
1400  * released the lock and the system call should return.
1401  */
1402 static int wait_task_continued(struct task_struct *p, int noreap,
1403                                struct siginfo __user *infop,
1404                                int __user *stat_addr, struct rusage __user *ru)
1405 {
1406         int retval;
1407         pid_t pid;
1408         uid_t uid;
1409
1410         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1411                 return 0;
1412
1413         spin_lock_irq(&p->sighand->siglock);
1414         /* Re-check with the lock held.  */
1415         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1416                 spin_unlock_irq(&p->sighand->siglock);
1417                 return 0;
1418         }
1419         if (!noreap)
1420                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1421         spin_unlock_irq(&p->sighand->siglock);
1422
1423         pid = task_pid_vnr(p);
1424         uid = p->uid;
1425         get_task_struct(p);
1426         read_unlock(&tasklist_lock);
1427
1428         if (!infop) {
1429                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1430                 put_task_struct(p);
1431                 if (!retval && stat_addr)
1432                         retval = put_user(0xffff, stat_addr);
1433                 if (!retval)
1434                         retval = pid;
1435         } else {
1436                 retval = wait_noreap_copyout(p, pid, uid,
1437                                              CLD_CONTINUED, SIGCONT,
1438                                              infop, ru);
1439                 BUG_ON(retval == 0);
1440         }
1441
1442         return retval;
1443 }
1444
1445 static long do_wait(enum pid_type type, struct pid *pid, int options,
1446                     struct siginfo __user *infop, int __user *stat_addr,
1447                     struct rusage __user *ru)
1448 {
1449         DECLARE_WAITQUEUE(wait, current);
1450         struct task_struct *tsk;
1451         int flag, retval;
1452
1453         add_wait_queue(&current->signal->wait_chldexit,&wait);
1454 repeat:
1455         /* If there is nothing that can match our critier just get out */
1456         retval = -ECHILD;
1457         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1458                 goto end;
1459
1460         /*
1461          * We will set this flag if we see any child that might later
1462          * match our criteria, even if we are not able to reap it yet.
1463          */
1464         flag = retval = 0;
1465         current->state = TASK_INTERRUPTIBLE;
1466         read_lock(&tasklist_lock);
1467         tsk = current;
1468         do {
1469                 struct task_struct *p;
1470
1471                 list_for_each_entry(p, &tsk->children, sibling) {
1472                         int ret = eligible_child(type, pid, options, p);
1473                         if (!ret)
1474                                 continue;
1475
1476                         if (unlikely(ret < 0)) {
1477                                 retval = ret;
1478                         } else if (task_is_stopped_or_traced(p)) {
1479                                 /*
1480                                  * It's stopped now, so it might later
1481                                  * continue, exit, or stop again.
1482                                  */
1483                                 flag = 1;
1484                                 if (!(p->ptrace & PT_PTRACED) &&
1485                                     !(options & WUNTRACED))
1486                                         continue;
1487
1488                                 retval = wait_task_stopped(p,
1489                                                 (options & WNOWAIT), infop,
1490                                                 stat_addr, ru);
1491                         } else if (p->exit_state == EXIT_ZOMBIE &&
1492                                         !delay_group_leader(p)) {
1493                                 /*
1494                                  * We don't reap group leaders with subthreads.
1495                                  */
1496                                 if (!likely(options & WEXITED))
1497                                         continue;
1498                                 retval = wait_task_zombie(p,
1499                                                 (options & WNOWAIT), infop,
1500                                                 stat_addr, ru);
1501                         } else if (p->exit_state != EXIT_DEAD) {
1502                                 /*
1503                                  * It's running now, so it might later
1504                                  * exit, stop, or stop and then continue.
1505                                  */
1506                                 flag = 1;
1507                                 if (!unlikely(options & WCONTINUED))
1508                                         continue;
1509                                 retval = wait_task_continued(p,
1510                                                 (options & WNOWAIT), infop,
1511                                                 stat_addr, ru);
1512                         }
1513                         if (retval != 0) /* tasklist_lock released */
1514                                 goto end;
1515                 }
1516                 if (!flag) {
1517                         list_for_each_entry(p, &tsk->ptrace_children,
1518                                                                 ptrace_list) {
1519                                 flag = eligible_child(type, pid, options, p);
1520                                 if (!flag)
1521                                         continue;
1522                                 if (likely(flag > 0))
1523                                         break;
1524                                 retval = flag;
1525                                 goto end;
1526                         }
1527                 }
1528                 if (options & __WNOTHREAD)
1529                         break;
1530                 tsk = next_thread(tsk);
1531                 BUG_ON(tsk->signal != current->signal);
1532         } while (tsk != current);
1533         read_unlock(&tasklist_lock);
1534
1535         if (flag) {
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 upid,
1575                            struct siginfo __user *infop, int options,
1576                            struct rusage __user *ru)
1577 {
1578         struct pid *pid = NULL;
1579         enum pid_type type;
1580         long ret;
1581
1582         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1583                 return -EINVAL;
1584         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1585                 return -EINVAL;
1586
1587         switch (which) {
1588         case P_ALL:
1589                 type = PIDTYPE_MAX;
1590                 break;
1591         case P_PID:
1592                 type = PIDTYPE_PID;
1593                 if (upid <= 0)
1594                         return -EINVAL;
1595                 break;
1596         case P_PGID:
1597                 type = PIDTYPE_PGID;
1598                 if (upid <= 0)
1599                         return -EINVAL;
1600                 break;
1601         default:
1602                 return -EINVAL;
1603         }
1604
1605         if (type < PIDTYPE_MAX)
1606                 pid = find_get_pid(upid);
1607         ret = do_wait(type, pid, options, infop, NULL, ru);
1608         put_pid(pid);
1609
1610         /* avoid REGPARM breakage on x86: */
1611         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1612         return ret;
1613 }
1614
1615 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1616                           int options, struct rusage __user *ru)
1617 {
1618         struct pid *pid = NULL;
1619         enum pid_type type;
1620         long ret;
1621
1622         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1623                         __WNOTHREAD|__WCLONE|__WALL))
1624                 return -EINVAL;
1625
1626         if (upid == -1)
1627                 type = PIDTYPE_MAX;
1628         else if (upid < 0) {
1629                 type = PIDTYPE_PGID;
1630                 pid = find_get_pid(-upid);
1631         } else if (upid == 0) {
1632                 type = PIDTYPE_PGID;
1633                 pid = get_pid(task_pgrp(current));
1634         } else /* upid > 0 */ {
1635                 type = PIDTYPE_PID;
1636                 pid = find_get_pid(upid);
1637         }
1638
1639         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1640         put_pid(pid);
1641
1642         /* avoid REGPARM breakage on x86: */
1643         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1644         return ret;
1645 }
1646
1647 #ifdef __ARCH_WANT_SYS_WAITPID
1648
1649 /*
1650  * sys_waitpid() remains for compatibility. waitpid() should be
1651  * implemented by calling sys_wait4() from libc.a.
1652  */
1653 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1654 {
1655         return sys_wait4(pid, stat_addr, options, NULL);
1656 }
1657
1658 #endif