41c1edace97a34fc44d5725ed4544ad07bf33dd5
[sfrench/cifs-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)
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         kill_orphaned_pgrp(tsk, NULL);
770
771         /* Let father know we died
772          *
773          * Thread signals are configurable, but you aren't going to use
774          * that to send signals to arbitary processes.
775          * That stops right now.
776          *
777          * If the parent exec id doesn't match the exec id we saved
778          * when we started then we know the parent has changed security
779          * domain.
780          *
781          * If our self_exec id doesn't match our parent_exec_id then
782          * we have changed execution domain as these two values started
783          * the same after a fork.
784          */
785         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
786             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
787              tsk->self_exec_id != tsk->parent_exec_id)
788             && !capable(CAP_KILL))
789                 tsk->exit_signal = SIGCHLD;
790
791
792         /* If something other than our normal parent is ptracing us, then
793          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
794          * only has special meaning to our real parent.
795          */
796         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
797                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
798                 do_notify_parent(tsk, signal);
799         } else if (tsk->ptrace) {
800                 do_notify_parent(tsk, SIGCHLD);
801         }
802
803         state = EXIT_ZOMBIE;
804         if (tsk->exit_signal == -1 && likely(!tsk->ptrace))
805                 state = EXIT_DEAD;
806         tsk->exit_state = state;
807
808         if (thread_group_leader(tsk) &&
809             tsk->signal->notify_count < 0 &&
810             tsk->signal->group_exit_task)
811                 wake_up_process(tsk->signal->group_exit_task);
812
813         write_unlock_irq(&tasklist_lock);
814
815         /* If the process is dead, release it - nobody will wait for it */
816         if (state == EXIT_DEAD)
817                 release_task(tsk);
818 }
819
820 #ifdef CONFIG_DEBUG_STACK_USAGE
821 static void check_stack_usage(void)
822 {
823         static DEFINE_SPINLOCK(low_water_lock);
824         static int lowest_to_date = THREAD_SIZE;
825         unsigned long *n = end_of_stack(current);
826         unsigned long free;
827
828         while (*n == 0)
829                 n++;
830         free = (unsigned long)n - (unsigned long)end_of_stack(current);
831
832         if (free >= lowest_to_date)
833                 return;
834
835         spin_lock(&low_water_lock);
836         if (free < lowest_to_date) {
837                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
838                                 "left\n",
839                                 current->comm, free);
840                 lowest_to_date = free;
841         }
842         spin_unlock(&low_water_lock);
843 }
844 #else
845 static inline void check_stack_usage(void) {}
846 #endif
847
848 static inline void exit_child_reaper(struct task_struct *tsk)
849 {
850         if (likely(tsk->group_leader != task_child_reaper(tsk)))
851                 return;
852
853         if (tsk->nsproxy->pid_ns == &init_pid_ns)
854                 panic("Attempted to kill init!");
855
856         /*
857          * @tsk is the last thread in the 'cgroup-init' and is exiting.
858          * Terminate all remaining processes in the namespace and reap them
859          * before exiting @tsk.
860          *
861          * Note that @tsk (last thread of cgroup-init) may not necessarily
862          * be the child-reaper (i.e main thread of cgroup-init) of the
863          * namespace i.e the child_reaper may have already exited.
864          *
865          * Even after a child_reaper exits, we let it inherit orphaned children,
866          * because, pid_ns->child_reaper remains valid as long as there is
867          * at least one living sub-thread in the cgroup init.
868
869          * This living sub-thread of the cgroup-init will be notified when
870          * a child inherited by the 'child-reaper' exits (do_notify_parent()
871          * uses __group_send_sig_info()). Further, when reaping child processes,
872          * do_wait() iterates over children of all living sub threads.
873
874          * i.e even though 'child_reaper' thread is listed as the parent of the
875          * orphaned children, any living sub-thread in the cgroup-init can
876          * perform the role of the child_reaper.
877          */
878         zap_pid_ns_processes(tsk->nsproxy->pid_ns);
879 }
880
881 NORET_TYPE void do_exit(long code)
882 {
883         struct task_struct *tsk = current;
884         int group_dead;
885
886         profile_task_exit(tsk);
887
888         WARN_ON(atomic_read(&tsk->fs_excl));
889
890         if (unlikely(in_interrupt()))
891                 panic("Aiee, killing interrupt handler!");
892         if (unlikely(!tsk->pid))
893                 panic("Attempted to kill the idle task!");
894
895         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
896                 current->ptrace_message = code;
897                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
898         }
899
900         /*
901          * We're taking recursive faults here in do_exit. Safest is to just
902          * leave this task alone and wait for reboot.
903          */
904         if (unlikely(tsk->flags & PF_EXITING)) {
905                 printk(KERN_ALERT
906                         "Fixing recursive fault but reboot is needed!\n");
907                 /*
908                  * We can do this unlocked here. The futex code uses
909                  * this flag just to verify whether the pi state
910                  * cleanup has been done or not. In the worst case it
911                  * loops once more. We pretend that the cleanup was
912                  * done as there is no way to return. Either the
913                  * OWNER_DIED bit is set by now or we push the blocked
914                  * task into the wait for ever nirwana as well.
915                  */
916                 tsk->flags |= PF_EXITPIDONE;
917                 if (tsk->io_context)
918                         exit_io_context();
919                 set_current_state(TASK_UNINTERRUPTIBLE);
920                 schedule();
921         }
922
923         exit_signals(tsk);  /* sets PF_EXITING */
924         /*
925          * tsk->flags are checked in the futex code to protect against
926          * an exiting task cleaning up the robust pi futexes.
927          */
928         smp_mb();
929         spin_unlock_wait(&tsk->pi_lock);
930
931         if (unlikely(in_atomic()))
932                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
933                                 current->comm, task_pid_nr(current),
934                                 preempt_count());
935
936         acct_update_integrals(tsk);
937         if (tsk->mm) {
938                 update_hiwater_rss(tsk->mm);
939                 update_hiwater_vm(tsk->mm);
940         }
941         group_dead = atomic_dec_and_test(&tsk->signal->live);
942         if (group_dead) {
943                 exit_child_reaper(tsk);
944                 hrtimer_cancel(&tsk->signal->real_timer);
945                 exit_itimers(tsk->signal);
946         }
947         acct_collect(code, group_dead);
948 #ifdef CONFIG_FUTEX
949         if (unlikely(tsk->robust_list))
950                 exit_robust_list(tsk);
951 #ifdef CONFIG_COMPAT
952         if (unlikely(tsk->compat_robust_list))
953                 compat_exit_robust_list(tsk);
954 #endif
955 #endif
956         if (group_dead)
957                 tty_audit_exit();
958         if (unlikely(tsk->audit_context))
959                 audit_free(tsk);
960
961         tsk->exit_code = code;
962         taskstats_exit(tsk, group_dead);
963
964         exit_mm(tsk);
965
966         if (group_dead)
967                 acct_process();
968         exit_sem(tsk);
969         __exit_files(tsk);
970         __exit_fs(tsk);
971         check_stack_usage();
972         exit_thread();
973         cgroup_exit(tsk, 1);
974         exit_keys(tsk);
975
976         if (group_dead && tsk->signal->leader)
977                 disassociate_ctty(1);
978
979         module_put(task_thread_info(tsk)->exec_domain->module);
980         if (tsk->binfmt)
981                 module_put(tsk->binfmt->module);
982
983         proc_exit_connector(tsk);
984         exit_notify(tsk);
985 #ifdef CONFIG_NUMA
986         mpol_free(tsk->mempolicy);
987         tsk->mempolicy = NULL;
988 #endif
989 #ifdef CONFIG_FUTEX
990         /*
991          * This must happen late, after the PID is not
992          * hashed anymore:
993          */
994         if (unlikely(!list_empty(&tsk->pi_state_list)))
995                 exit_pi_state_list(tsk);
996         if (unlikely(current->pi_state_cache))
997                 kfree(current->pi_state_cache);
998 #endif
999         /*
1000          * Make sure we are holding no locks:
1001          */
1002         debug_check_no_locks_held(tsk);
1003         /*
1004          * We can do this unlocked here. The futex code uses this flag
1005          * just to verify whether the pi state cleanup has been done
1006          * or not. In the worst case it loops once more.
1007          */
1008         tsk->flags |= PF_EXITPIDONE;
1009
1010         if (tsk->io_context)
1011                 exit_io_context();
1012
1013         if (tsk->splice_pipe)
1014                 __free_pipe_info(tsk->splice_pipe);
1015
1016         preempt_disable();
1017         /* causes final put_task_struct in finish_task_switch(). */
1018         tsk->state = TASK_DEAD;
1019
1020         schedule();
1021         BUG();
1022         /* Avoid "noreturn function does return".  */
1023         for (;;)
1024                 cpu_relax();    /* For when BUG is null */
1025 }
1026
1027 EXPORT_SYMBOL_GPL(do_exit);
1028
1029 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1030 {
1031         if (comp)
1032                 complete(comp);
1033
1034         do_exit(code);
1035 }
1036
1037 EXPORT_SYMBOL(complete_and_exit);
1038
1039 asmlinkage long sys_exit(int error_code)
1040 {
1041         do_exit((error_code&0xff)<<8);
1042 }
1043
1044 /*
1045  * Take down every thread in the group.  This is called by fatal signals
1046  * as well as by sys_exit_group (below).
1047  */
1048 NORET_TYPE void
1049 do_group_exit(int exit_code)
1050 {
1051         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1052
1053         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1054                 exit_code = current->signal->group_exit_code;
1055         else if (!thread_group_empty(current)) {
1056                 struct signal_struct *const sig = current->signal;
1057                 struct sighand_struct *const sighand = current->sighand;
1058                 spin_lock_irq(&sighand->siglock);
1059                 if (signal_group_exit(sig))
1060                         /* Another thread got here before we took the lock.  */
1061                         exit_code = sig->group_exit_code;
1062                 else {
1063                         sig->group_exit_code = exit_code;
1064                         sig->flags = SIGNAL_GROUP_EXIT;
1065                         zap_other_threads(current);
1066                 }
1067                 spin_unlock_irq(&sighand->siglock);
1068         }
1069
1070         do_exit(exit_code);
1071         /* NOTREACHED */
1072 }
1073
1074 /*
1075  * this kills every thread in the thread group. Note that any externally
1076  * wait4()-ing process will get the correct exit code - even if this
1077  * thread is not the thread group leader.
1078  */
1079 asmlinkage void sys_exit_group(int error_code)
1080 {
1081         do_group_exit((error_code & 0xff) << 8);
1082 }
1083
1084 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1085 {
1086         struct pid *pid = NULL;
1087         if (type == PIDTYPE_PID)
1088                 pid = task->pids[type].pid;
1089         else if (type < PIDTYPE_MAX)
1090                 pid = task->group_leader->pids[type].pid;
1091         return pid;
1092 }
1093
1094 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1095                           struct task_struct *p)
1096 {
1097         int err;
1098
1099         if (type < PIDTYPE_MAX) {
1100                 if (task_pid_type(p, type) != pid)
1101                         return 0;
1102         }
1103
1104         /*
1105          * Do not consider detached threads that are
1106          * not ptraced:
1107          */
1108         if (p->exit_signal == -1 && !p->ptrace)
1109                 return 0;
1110
1111         /* Wait for all children (clone and not) if __WALL is set;
1112          * otherwise, wait for clone children *only* if __WCLONE is
1113          * set; otherwise, wait for non-clone children *only*.  (Note:
1114          * A "clone" child here is one that reports to its parent
1115          * using a signal other than SIGCHLD.) */
1116         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1117             && !(options & __WALL))
1118                 return 0;
1119
1120         err = security_task_wait(p);
1121         if (likely(!err))
1122                 return 1;
1123
1124         if (type != PIDTYPE_PID)
1125                 return 0;
1126         /* This child was explicitly requested, abort */
1127         read_unlock(&tasklist_lock);
1128         return err;
1129 }
1130
1131 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1132                                int why, int status,
1133                                struct siginfo __user *infop,
1134                                struct rusage __user *rusagep)
1135 {
1136         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1137
1138         put_task_struct(p);
1139         if (!retval)
1140                 retval = put_user(SIGCHLD, &infop->si_signo);
1141         if (!retval)
1142                 retval = put_user(0, &infop->si_errno);
1143         if (!retval)
1144                 retval = put_user((short)why, &infop->si_code);
1145         if (!retval)
1146                 retval = put_user(pid, &infop->si_pid);
1147         if (!retval)
1148                 retval = put_user(uid, &infop->si_uid);
1149         if (!retval)
1150                 retval = put_user(status, &infop->si_status);
1151         if (!retval)
1152                 retval = pid;
1153         return retval;
1154 }
1155
1156 /*
1157  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1158  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1159  * the lock and this task is uninteresting.  If we return nonzero, we have
1160  * released the lock and the system call should return.
1161  */
1162 static int wait_task_zombie(struct task_struct *p, int noreap,
1163                             struct siginfo __user *infop,
1164                             int __user *stat_addr, struct rusage __user *ru)
1165 {
1166         unsigned long state;
1167         int retval, status, traced;
1168         pid_t pid = task_pid_vnr(p);
1169
1170         if (unlikely(noreap)) {
1171                 uid_t uid = p->uid;
1172                 int exit_code = p->exit_code;
1173                 int why, status;
1174
1175                 get_task_struct(p);
1176                 read_unlock(&tasklist_lock);
1177                 if ((exit_code & 0x7f) == 0) {
1178                         why = CLD_EXITED;
1179                         status = exit_code >> 8;
1180                 } else {
1181                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1182                         status = exit_code & 0x7f;
1183                 }
1184                 return wait_noreap_copyout(p, pid, uid, why,
1185                                            status, infop, ru);
1186         }
1187
1188         /*
1189          * Try to move the task's state to DEAD
1190          * only one thread is allowed to do this:
1191          */
1192         state = xchg(&p->exit_state, EXIT_DEAD);
1193         if (state != EXIT_ZOMBIE) {
1194                 BUG_ON(state != EXIT_DEAD);
1195                 return 0;
1196         }
1197
1198         /* traced means p->ptrace, but not vice versa */
1199         traced = (p->real_parent != p->parent);
1200
1201         if (likely(!traced)) {
1202                 struct signal_struct *psig;
1203                 struct signal_struct *sig;
1204
1205                 /*
1206                  * The resource counters for the group leader are in its
1207                  * own task_struct.  Those for dead threads in the group
1208                  * are in its signal_struct, as are those for the child
1209                  * processes it has previously reaped.  All these
1210                  * accumulate in the parent's signal_struct c* fields.
1211                  *
1212                  * We don't bother to take a lock here to protect these
1213                  * p->signal fields, because they are only touched by
1214                  * __exit_signal, which runs with tasklist_lock
1215                  * write-locked anyway, and so is excluded here.  We do
1216                  * need to protect the access to p->parent->signal fields,
1217                  * as other threads in the parent group can be right
1218                  * here reaping other children at the same time.
1219                  */
1220                 spin_lock_irq(&p->parent->sighand->siglock);
1221                 psig = p->parent->signal;
1222                 sig = p->signal;
1223                 psig->cutime =
1224                         cputime_add(psig->cutime,
1225                         cputime_add(p->utime,
1226                         cputime_add(sig->utime,
1227                                     sig->cutime)));
1228                 psig->cstime =
1229                         cputime_add(psig->cstime,
1230                         cputime_add(p->stime,
1231                         cputime_add(sig->stime,
1232                                     sig->cstime)));
1233                 psig->cgtime =
1234                         cputime_add(psig->cgtime,
1235                         cputime_add(p->gtime,
1236                         cputime_add(sig->gtime,
1237                                     sig->cgtime)));
1238                 psig->cmin_flt +=
1239                         p->min_flt + sig->min_flt + sig->cmin_flt;
1240                 psig->cmaj_flt +=
1241                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1242                 psig->cnvcsw +=
1243                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1244                 psig->cnivcsw +=
1245                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1246                 psig->cinblock +=
1247                         task_io_get_inblock(p) +
1248                         sig->inblock + sig->cinblock;
1249                 psig->coublock +=
1250                         task_io_get_oublock(p) +
1251                         sig->oublock + sig->coublock;
1252                 spin_unlock_irq(&p->parent->sighand->siglock);
1253         }
1254
1255         /*
1256          * Now we are sure this task is interesting, and no other
1257          * thread can reap it because we set its state to EXIT_DEAD.
1258          */
1259         read_unlock(&tasklist_lock);
1260
1261         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1262         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1263                 ? p->signal->group_exit_code : p->exit_code;
1264         if (!retval && stat_addr)
1265                 retval = put_user(status, stat_addr);
1266         if (!retval && infop)
1267                 retval = put_user(SIGCHLD, &infop->si_signo);
1268         if (!retval && infop)
1269                 retval = put_user(0, &infop->si_errno);
1270         if (!retval && infop) {
1271                 int why;
1272
1273                 if ((status & 0x7f) == 0) {
1274                         why = CLD_EXITED;
1275                         status >>= 8;
1276                 } else {
1277                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1278                         status &= 0x7f;
1279                 }
1280                 retval = put_user((short)why, &infop->si_code);
1281                 if (!retval)
1282                         retval = put_user(status, &infop->si_status);
1283         }
1284         if (!retval && infop)
1285                 retval = put_user(pid, &infop->si_pid);
1286         if (!retval && infop)
1287                 retval = put_user(p->uid, &infop->si_uid);
1288         if (!retval)
1289                 retval = pid;
1290
1291         if (traced) {
1292                 write_lock_irq(&tasklist_lock);
1293                 /* We dropped tasklist, ptracer could die and untrace */
1294                 ptrace_unlink(p);
1295                 /*
1296                  * If this is not a detached task, notify the parent.
1297                  * If it's still not detached after that, don't release
1298                  * it now.
1299                  */
1300                 if (p->exit_signal != -1) {
1301                         do_notify_parent(p, p->exit_signal);
1302                         if (p->exit_signal != -1) {
1303                                 p->exit_state = EXIT_ZOMBIE;
1304                                 p = NULL;
1305                         }
1306                 }
1307                 write_unlock_irq(&tasklist_lock);
1308         }
1309         if (p != NULL)
1310                 release_task(p);
1311
1312         return retval;
1313 }
1314
1315 /*
1316  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1317  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1318  * the lock and this task is uninteresting.  If we return nonzero, we have
1319  * released the lock and the system call should return.
1320  */
1321 static int wait_task_stopped(struct task_struct *p,
1322                              int noreap, struct siginfo __user *infop,
1323                              int __user *stat_addr, struct rusage __user *ru)
1324 {
1325         int retval, exit_code, why;
1326         uid_t uid = 0; /* unneeded, required by compiler */
1327         pid_t pid;
1328
1329         exit_code = 0;
1330         spin_lock_irq(&p->sighand->siglock);
1331
1332         if (unlikely(!task_is_stopped_or_traced(p)))
1333                 goto unlock_sig;
1334
1335         if (!(p->ptrace & PT_PTRACED) && p->signal->group_stop_count > 0)
1336                 /*
1337                  * A group stop is in progress and this is the group leader.
1338                  * We won't report until all threads have stopped.
1339                  */
1340                 goto unlock_sig;
1341
1342         exit_code = p->exit_code;
1343         if (!exit_code)
1344                 goto unlock_sig;
1345
1346         if (!noreap)
1347                 p->exit_code = 0;
1348
1349         uid = p->uid;
1350 unlock_sig:
1351         spin_unlock_irq(&p->sighand->siglock);
1352         if (!exit_code)
1353                 return 0;
1354
1355         /*
1356          * Now we are pretty sure this task is interesting.
1357          * Make sure it doesn't get reaped out from under us while we
1358          * give up the lock and then examine it below.  We don't want to
1359          * keep holding onto the tasklist_lock while we call getrusage and
1360          * possibly take page faults for user memory.
1361          */
1362         get_task_struct(p);
1363         pid = task_pid_vnr(p);
1364         why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1365         read_unlock(&tasklist_lock);
1366
1367         if (unlikely(noreap))
1368                 return wait_noreap_copyout(p, pid, uid,
1369                                            why, exit_code,
1370                                            infop, ru);
1371
1372         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1373         if (!retval && stat_addr)
1374                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1375         if (!retval && infop)
1376                 retval = put_user(SIGCHLD, &infop->si_signo);
1377         if (!retval && infop)
1378                 retval = put_user(0, &infop->si_errno);
1379         if (!retval && infop)
1380                 retval = put_user(why, &infop->si_code);
1381         if (!retval && infop)
1382                 retval = put_user(exit_code, &infop->si_status);
1383         if (!retval && infop)
1384                 retval = put_user(pid, &infop->si_pid);
1385         if (!retval && infop)
1386                 retval = put_user(uid, &infop->si_uid);
1387         if (!retval)
1388                 retval = pid;
1389         put_task_struct(p);
1390
1391         BUG_ON(!retval);
1392         return retval;
1393 }
1394
1395 /*
1396  * Handle do_wait work for one task in a live, non-stopped state.
1397  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1398  * the lock and this task is uninteresting.  If we return nonzero, we have
1399  * released the lock and the system call should return.
1400  */
1401 static int wait_task_continued(struct task_struct *p, int noreap,
1402                                struct siginfo __user *infop,
1403                                int __user *stat_addr, struct rusage __user *ru)
1404 {
1405         int retval;
1406         pid_t pid;
1407         uid_t uid;
1408
1409         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1410                 return 0;
1411
1412         spin_lock_irq(&p->sighand->siglock);
1413         /* Re-check with the lock held.  */
1414         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1415                 spin_unlock_irq(&p->sighand->siglock);
1416                 return 0;
1417         }
1418         if (!noreap)
1419                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1420         spin_unlock_irq(&p->sighand->siglock);
1421
1422         pid = task_pid_vnr(p);
1423         uid = p->uid;
1424         get_task_struct(p);
1425         read_unlock(&tasklist_lock);
1426
1427         if (!infop) {
1428                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1429                 put_task_struct(p);
1430                 if (!retval && stat_addr)
1431                         retval = put_user(0xffff, stat_addr);
1432                 if (!retval)
1433                         retval = pid;
1434         } else {
1435                 retval = wait_noreap_copyout(p, pid, uid,
1436                                              CLD_CONTINUED, SIGCONT,
1437                                              infop, ru);
1438                 BUG_ON(retval == 0);
1439         }
1440
1441         return retval;
1442 }
1443
1444 static long do_wait(enum pid_type type, struct pid *pid, int options,
1445                     struct siginfo __user *infop, int __user *stat_addr,
1446                     struct rusage __user *ru)
1447 {
1448         DECLARE_WAITQUEUE(wait, current);
1449         struct task_struct *tsk;
1450         int flag, retval;
1451
1452         add_wait_queue(&current->signal->wait_chldexit,&wait);
1453 repeat:
1454         /* If there is nothing that can match our critier just get out */
1455         retval = -ECHILD;
1456         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1457                 goto end;
1458
1459         /*
1460          * We will set this flag if we see any child that might later
1461          * match our criteria, even if we are not able to reap it yet.
1462          */
1463         flag = retval = 0;
1464         current->state = TASK_INTERRUPTIBLE;
1465         read_lock(&tasklist_lock);
1466         tsk = current;
1467         do {
1468                 struct task_struct *p;
1469
1470                 list_for_each_entry(p, &tsk->children, sibling) {
1471                         int ret = eligible_child(type, pid, options, p);
1472                         if (!ret)
1473                                 continue;
1474
1475                         if (unlikely(ret < 0)) {
1476                                 retval = ret;
1477                         } else if (task_is_stopped_or_traced(p)) {
1478                                 /*
1479                                  * It's stopped now, so it might later
1480                                  * continue, exit, or stop again.
1481                                  */
1482                                 flag = 1;
1483                                 if (!(p->ptrace & PT_PTRACED) &&
1484                                     !(options & WUNTRACED))
1485                                         continue;
1486
1487                                 retval = wait_task_stopped(p,
1488                                                 (options & WNOWAIT), infop,
1489                                                 stat_addr, ru);
1490                         } else if (p->exit_state == EXIT_ZOMBIE &&
1491                                         !delay_group_leader(p)) {
1492                                 /*
1493                                  * We don't reap group leaders with subthreads.
1494                                  */
1495                                 if (!likely(options & WEXITED))
1496                                         continue;
1497                                 retval = wait_task_zombie(p,
1498                                                 (options & WNOWAIT), infop,
1499                                                 stat_addr, ru);
1500                         } else if (p->exit_state != EXIT_DEAD) {
1501                                 /*
1502                                  * It's running now, so it might later
1503                                  * exit, stop, or stop and then continue.
1504                                  */
1505                                 flag = 1;
1506                                 if (!unlikely(options & WCONTINUED))
1507                                         continue;
1508                                 retval = wait_task_continued(p,
1509                                                 (options & WNOWAIT), infop,
1510                                                 stat_addr, ru);
1511                         }
1512                         if (retval != 0) /* tasklist_lock released */
1513                                 goto end;
1514                 }
1515                 if (!flag) {
1516                         list_for_each_entry(p, &tsk->ptrace_children,
1517                                                                 ptrace_list) {
1518                                 flag = eligible_child(type, pid, options, p);
1519                                 if (!flag)
1520                                         continue;
1521                                 if (likely(flag > 0))
1522                                         break;
1523                                 retval = flag;
1524                                 goto end;
1525                         }
1526                 }
1527                 if (options & __WNOTHREAD)
1528                         break;
1529                 tsk = next_thread(tsk);
1530                 BUG_ON(tsk->signal != current->signal);
1531         } while (tsk != current);
1532         read_unlock(&tasklist_lock);
1533
1534         if (flag) {
1535                 if (options & WNOHANG)
1536                         goto end;
1537                 retval = -ERESTARTSYS;
1538                 if (signal_pending(current))
1539                         goto end;
1540                 schedule();
1541                 goto repeat;
1542         }
1543         retval = -ECHILD;
1544 end:
1545         current->state = TASK_RUNNING;
1546         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1547         if (infop) {
1548                 if (retval > 0)
1549                         retval = 0;
1550                 else {
1551                         /*
1552                          * For a WNOHANG return, clear out all the fields
1553                          * we would set so the user can easily tell the
1554                          * difference.
1555                          */
1556                         if (!retval)
1557                                 retval = put_user(0, &infop->si_signo);
1558                         if (!retval)
1559                                 retval = put_user(0, &infop->si_errno);
1560                         if (!retval)
1561                                 retval = put_user(0, &infop->si_code);
1562                         if (!retval)
1563                                 retval = put_user(0, &infop->si_pid);
1564                         if (!retval)
1565                                 retval = put_user(0, &infop->si_uid);
1566                         if (!retval)
1567                                 retval = put_user(0, &infop->si_status);
1568                 }
1569         }
1570         return retval;
1571 }
1572
1573 asmlinkage long sys_waitid(int which, pid_t upid,
1574                            struct siginfo __user *infop, int options,
1575                            struct rusage __user *ru)
1576 {
1577         struct pid *pid = NULL;
1578         enum pid_type type;
1579         long ret;
1580
1581         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1582                 return -EINVAL;
1583         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1584                 return -EINVAL;
1585
1586         switch (which) {
1587         case P_ALL:
1588                 type = PIDTYPE_MAX;
1589                 break;
1590         case P_PID:
1591                 type = PIDTYPE_PID;
1592                 if (upid <= 0)
1593                         return -EINVAL;
1594                 break;
1595         case P_PGID:
1596                 type = PIDTYPE_PGID;
1597                 if (upid <= 0)
1598                         return -EINVAL;
1599                 break;
1600         default:
1601                 return -EINVAL;
1602         }
1603
1604         if (type < PIDTYPE_MAX)
1605                 pid = find_get_pid(upid);
1606         ret = do_wait(type, pid, options, infop, NULL, ru);
1607         put_pid(pid);
1608
1609         /* avoid REGPARM breakage on x86: */
1610         prevent_tail_call(ret);
1611         return ret;
1612 }
1613
1614 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1615                           int options, struct rusage __user *ru)
1616 {
1617         struct pid *pid = NULL;
1618         enum pid_type type;
1619         long ret;
1620
1621         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1622                         __WNOTHREAD|__WCLONE|__WALL))
1623                 return -EINVAL;
1624
1625         if (upid == -1)
1626                 type = PIDTYPE_MAX;
1627         else if (upid < 0) {
1628                 type = PIDTYPE_PGID;
1629                 pid = find_get_pid(-upid);
1630         } else if (upid == 0) {
1631                 type = PIDTYPE_PGID;
1632                 pid = get_pid(task_pgrp(current));
1633         } else /* upid > 0 */ {
1634                 type = PIDTYPE_PID;
1635                 pid = find_get_pid(upid);
1636         }
1637
1638         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1639         put_pid(pid);
1640
1641         /* avoid REGPARM breakage on x86: */
1642         prevent_tail_call(ret);
1643         return ret;
1644 }
1645
1646 #ifdef __ARCH_WANT_SYS_WAITPID
1647
1648 /*
1649  * sys_waitpid() remains for compatibility. waitpid() should be
1650  * implemented by calling sys_wait4() from libc.a.
1651  */
1652 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1653 {
1654         return sys_wait4(pid, stat_addr, options, NULL);
1655 }
1656
1657 #endif