Merge branches 'core/debug', 'core/futexes', 'core/locking', 'core/rcu', 'core/signal...
[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/iocontext.h>
17 #include <linux/key.h>
18 #include <linux/security.h>
19 #include <linux/cpu.h>
20 #include <linux/acct.h>
21 #include <linux/tsacct_kern.h>
22 #include <linux/file.h>
23 #include <linux/fdtable.h>
24 #include <linux/binfmts.h>
25 #include <linux/nsproxy.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/ptrace.h>
28 #include <linux/profile.h>
29 #include <linux/mount.h>
30 #include <linux/proc_fs.h>
31 #include <linux/kthread.h>
32 #include <linux/mempolicy.h>
33 #include <linux/taskstats_kern.h>
34 #include <linux/delayacct.h>
35 #include <linux/freezer.h>
36 #include <linux/cgroup.h>
37 #include <linux/syscalls.h>
38 #include <linux/signal.h>
39 #include <linux/posix-timers.h>
40 #include <linux/cn_proc.h>
41 #include <linux/mutex.h>
42 #include <linux/futex.h>
43 #include <linux/pipe_fs_i.h>
44 #include <linux/audit.h> /* for audit_free() */
45 #include <linux/resource.h>
46 #include <linux/blkdev.h>
47 #include <linux/task_io_accounting_ops.h>
48 #include <linux/tracehook.h>
49 #include <trace/sched.h>
50
51 #include <asm/uaccess.h>
52 #include <asm/unistd.h>
53 #include <asm/pgtable.h>
54 #include <asm/mmu_context.h>
55
56 static void exit_mm(struct task_struct * tsk);
57
58 static inline int task_detached(struct task_struct *p)
59 {
60         return p->exit_signal == -1;
61 }
62
63 static void __unhash_process(struct task_struct *p)
64 {
65         nr_threads--;
66         detach_pid(p, PIDTYPE_PID);
67         if (thread_group_leader(p)) {
68                 detach_pid(p, PIDTYPE_PGID);
69                 detach_pid(p, PIDTYPE_SID);
70
71                 list_del_rcu(&p->tasks);
72                 __get_cpu_var(process_counts)--;
73         }
74         list_del_rcu(&p->thread_group);
75         list_del_init(&p->sibling);
76 }
77
78 /*
79  * This function expects the tasklist_lock write-locked.
80  */
81 static void __exit_signal(struct task_struct *tsk)
82 {
83         struct signal_struct *sig = tsk->signal;
84         struct sighand_struct *sighand;
85
86         BUG_ON(!sig);
87         BUG_ON(!atomic_read(&sig->count));
88
89         sighand = rcu_dereference(tsk->sighand);
90         spin_lock(&sighand->siglock);
91
92         posix_cpu_timers_exit(tsk);
93         if (atomic_dec_and_test(&sig->count))
94                 posix_cpu_timers_exit_group(tsk);
95         else {
96                 /*
97                  * If there is any task waiting for the group exit
98                  * then notify it:
99                  */
100                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
101                         wake_up_process(sig->group_exit_task);
102
103                 if (tsk == sig->curr_target)
104                         sig->curr_target = next_thread(tsk);
105                 /*
106                  * Accumulate here the counters for all threads but the
107                  * group leader as they die, so they can be added into
108                  * the process-wide totals when those are taken.
109                  * The group leader stays around as a zombie as long
110                  * as there are other threads.  When it gets reaped,
111                  * the exit.c code will add its counts into these totals.
112                  * We won't ever get here for the group leader, since it
113                  * will have been the last reference on the signal_struct.
114                  */
115                 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
116                 sig->min_flt += tsk->min_flt;
117                 sig->maj_flt += tsk->maj_flt;
118                 sig->nvcsw += tsk->nvcsw;
119                 sig->nivcsw += tsk->nivcsw;
120                 sig->inblock += task_io_get_inblock(tsk);
121                 sig->oublock += task_io_get_oublock(tsk);
122                 task_io_accounting_add(&sig->ioac, &tsk->ioac);
123                 sig = NULL; /* Marker for below. */
124         }
125
126         __unhash_process(tsk);
127
128         /*
129          * Do this under ->siglock, we can race with another thread
130          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
131          */
132         flush_sigqueue(&tsk->pending);
133
134         tsk->signal = NULL;
135         tsk->sighand = NULL;
136         spin_unlock(&sighand->siglock);
137
138         __cleanup_sighand(sighand);
139         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
140         if (sig) {
141                 flush_sigqueue(&sig->shared_pending);
142                 taskstats_tgid_free(sig);
143                 /*
144                  * Make sure ->signal can't go away under rq->lock,
145                  * see account_group_exec_runtime().
146                  */
147                 task_rq_unlock_wait(tsk);
148                 __cleanup_signal(sig);
149         }
150 }
151
152 static void delayed_put_task_struct(struct rcu_head *rhp)
153 {
154         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
155
156         trace_sched_process_free(tsk);
157         put_task_struct(tsk);
158 }
159
160
161 void release_task(struct task_struct * p)
162 {
163         struct task_struct *leader;
164         int zap_leader;
165 repeat:
166         tracehook_prepare_release_task(p);
167         atomic_dec(&p->user->processes);
168         proc_flush_task(p);
169         write_lock_irq(&tasklist_lock);
170         tracehook_finish_release_task(p);
171         __exit_signal(p);
172
173         /*
174          * If we are the last non-leader member of the thread
175          * group, and the leader is zombie, then notify the
176          * group leader's parent process. (if it wants notification.)
177          */
178         zap_leader = 0;
179         leader = p->group_leader;
180         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
181                 BUG_ON(task_detached(leader));
182                 do_notify_parent(leader, leader->exit_signal);
183                 /*
184                  * If we were the last child thread and the leader has
185                  * exited already, and the leader's parent ignores SIGCHLD,
186                  * then we are the one who should release the leader.
187                  *
188                  * do_notify_parent() will have marked it self-reaping in
189                  * that case.
190                  */
191                 zap_leader = task_detached(leader);
192
193                 /*
194                  * This maintains the invariant that release_task()
195                  * only runs on a task in EXIT_DEAD, just for sanity.
196                  */
197                 if (zap_leader)
198                         leader->exit_state = EXIT_DEAD;
199         }
200
201         write_unlock_irq(&tasklist_lock);
202         release_thread(p);
203         call_rcu(&p->rcu, delayed_put_task_struct);
204
205         p = leader;
206         if (unlikely(zap_leader))
207                 goto repeat;
208 }
209
210 /*
211  * This checks not only the pgrp, but falls back on the pid if no
212  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
213  * without this...
214  *
215  * The caller must hold rcu lock or the tasklist lock.
216  */
217 struct pid *session_of_pgrp(struct pid *pgrp)
218 {
219         struct task_struct *p;
220         struct pid *sid = NULL;
221
222         p = pid_task(pgrp, PIDTYPE_PGID);
223         if (p == NULL)
224                 p = pid_task(pgrp, PIDTYPE_PID);
225         if (p != NULL)
226                 sid = task_session(p);
227
228         return sid;
229 }
230
231 /*
232  * Determine if a process group is "orphaned", according to the POSIX
233  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
234  * by terminal-generated stop signals.  Newly orphaned process groups are
235  * to receive a SIGHUP and a SIGCONT.
236  *
237  * "I ask you, have you ever known what it is to be an orphan?"
238  */
239 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
240 {
241         struct task_struct *p;
242
243         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
244                 if ((p == ignored_task) ||
245                     (p->exit_state && thread_group_empty(p)) ||
246                     is_global_init(p->real_parent))
247                         continue;
248
249                 if (task_pgrp(p->real_parent) != pgrp &&
250                     task_session(p->real_parent) == task_session(p))
251                         return 0;
252         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
253
254         return 1;
255 }
256
257 int is_current_pgrp_orphaned(void)
258 {
259         int retval;
260
261         read_lock(&tasklist_lock);
262         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
263         read_unlock(&tasklist_lock);
264
265         return retval;
266 }
267
268 static int has_stopped_jobs(struct pid *pgrp)
269 {
270         int retval = 0;
271         struct task_struct *p;
272
273         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
274                 if (!task_is_stopped(p))
275                         continue;
276                 retval = 1;
277                 break;
278         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
279         return retval;
280 }
281
282 /*
283  * Check to see if any process groups have become orphaned as
284  * a result of our exiting, and if they have any stopped jobs,
285  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
286  */
287 static void
288 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
289 {
290         struct pid *pgrp = task_pgrp(tsk);
291         struct task_struct *ignored_task = tsk;
292
293         if (!parent)
294                  /* exit: our father is in a different pgrp than
295                   * we are and we were the only connection outside.
296                   */
297                 parent = tsk->real_parent;
298         else
299                 /* reparent: our child is in a different pgrp than
300                  * we are, and it was the only connection outside.
301                  */
302                 ignored_task = NULL;
303
304         if (task_pgrp(parent) != pgrp &&
305             task_session(parent) == task_session(tsk) &&
306             will_become_orphaned_pgrp(pgrp, ignored_task) &&
307             has_stopped_jobs(pgrp)) {
308                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
309                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
310         }
311 }
312
313 /**
314  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
315  *
316  * If a kernel thread is launched as a result of a system call, or if
317  * it ever exits, it should generally reparent itself to kthreadd so it
318  * isn't in the way of other processes and is correctly cleaned up on exit.
319  *
320  * The various task state such as scheduling policy and priority may have
321  * been inherited from a user process, so we reset them to sane values here.
322  *
323  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
324  */
325 static void reparent_to_kthreadd(void)
326 {
327         write_lock_irq(&tasklist_lock);
328
329         ptrace_unlink(current);
330         /* Reparent to init */
331         current->real_parent = current->parent = kthreadd_task;
332         list_move_tail(&current->sibling, &current->real_parent->children);
333
334         /* Set the exit signal to SIGCHLD so we signal init on exit */
335         current->exit_signal = SIGCHLD;
336
337         if (task_nice(current) < 0)
338                 set_user_nice(current, 0);
339         /* cpus_allowed? */
340         /* rt_priority? */
341         /* signals? */
342         security_task_reparent_to_init(current);
343         memcpy(current->signal->rlim, init_task.signal->rlim,
344                sizeof(current->signal->rlim));
345         atomic_inc(&(INIT_USER->__count));
346         write_unlock_irq(&tasklist_lock);
347         switch_uid(INIT_USER);
348 }
349
350 void __set_special_pids(struct pid *pid)
351 {
352         struct task_struct *curr = current->group_leader;
353         pid_t nr = pid_nr(pid);
354
355         if (task_session(curr) != pid) {
356                 change_pid(curr, PIDTYPE_SID, pid);
357                 set_task_session(curr, nr);
358         }
359         if (task_pgrp(curr) != pid) {
360                 change_pid(curr, PIDTYPE_PGID, pid);
361                 set_task_pgrp(curr, nr);
362         }
363 }
364
365 static void set_special_pids(struct pid *pid)
366 {
367         write_lock_irq(&tasklist_lock);
368         __set_special_pids(pid);
369         write_unlock_irq(&tasklist_lock);
370 }
371
372 /*
373  * Let kernel threads use this to say that they
374  * allow a certain signal (since daemonize() will
375  * have disabled all of them by default).
376  */
377 int allow_signal(int sig)
378 {
379         if (!valid_signal(sig) || sig < 1)
380                 return -EINVAL;
381
382         spin_lock_irq(&current->sighand->siglock);
383         sigdelset(&current->blocked, sig);
384         if (!current->mm) {
385                 /* Kernel threads handle their own signals.
386                    Let the signal code know it'll be handled, so
387                    that they don't get converted to SIGKILL or
388                    just silently dropped */
389                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
390         }
391         recalc_sigpending();
392         spin_unlock_irq(&current->sighand->siglock);
393         return 0;
394 }
395
396 EXPORT_SYMBOL(allow_signal);
397
398 int disallow_signal(int sig)
399 {
400         if (!valid_signal(sig) || sig < 1)
401                 return -EINVAL;
402
403         spin_lock_irq(&current->sighand->siglock);
404         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
405         recalc_sigpending();
406         spin_unlock_irq(&current->sighand->siglock);
407         return 0;
408 }
409
410 EXPORT_SYMBOL(disallow_signal);
411
412 /*
413  *      Put all the gunge required to become a kernel thread without
414  *      attached user resources in one place where it belongs.
415  */
416
417 void daemonize(const char *name, ...)
418 {
419         va_list args;
420         struct fs_struct *fs;
421         sigset_t blocked;
422
423         va_start(args, name);
424         vsnprintf(current->comm, sizeof(current->comm), name, args);
425         va_end(args);
426
427         /*
428          * If we were started as result of loading a module, close all of the
429          * user space pages.  We don't need them, and if we didn't close them
430          * they would be locked into memory.
431          */
432         exit_mm(current);
433         /*
434          * We don't want to have TIF_FREEZE set if the system-wide hibernation
435          * or suspend transition begins right now.
436          */
437         current->flags |= (PF_NOFREEZE | PF_KTHREAD);
438
439         if (current->nsproxy != &init_nsproxy) {
440                 get_nsproxy(&init_nsproxy);
441                 switch_task_namespaces(current, &init_nsproxy);
442         }
443         set_special_pids(&init_struct_pid);
444         proc_clear_tty(current);
445
446         /* Block and flush all signals */
447         sigfillset(&blocked);
448         sigprocmask(SIG_BLOCK, &blocked, NULL);
449         flush_signals(current);
450
451         /* Become as one with the init task */
452
453         exit_fs(current);       /* current->fs->count--; */
454         fs = init_task.fs;
455         current->fs = fs;
456         atomic_inc(&fs->count);
457
458         exit_files(current);
459         current->files = init_task.files;
460         atomic_inc(&current->files->count);
461
462         reparent_to_kthreadd();
463 }
464
465 EXPORT_SYMBOL(daemonize);
466
467 static void close_files(struct files_struct * files)
468 {
469         int i, j;
470         struct fdtable *fdt;
471
472         j = 0;
473
474         /*
475          * It is safe to dereference the fd table without RCU or
476          * ->file_lock because this is the last reference to the
477          * files structure.
478          */
479         fdt = files_fdtable(files);
480         for (;;) {
481                 unsigned long set;
482                 i = j * __NFDBITS;
483                 if (i >= fdt->max_fds)
484                         break;
485                 set = fdt->open_fds->fds_bits[j++];
486                 while (set) {
487                         if (set & 1) {
488                                 struct file * file = xchg(&fdt->fd[i], NULL);
489                                 if (file) {
490                                         filp_close(file, files);
491                                         cond_resched();
492                                 }
493                         }
494                         i++;
495                         set >>= 1;
496                 }
497         }
498 }
499
500 struct files_struct *get_files_struct(struct task_struct *task)
501 {
502         struct files_struct *files;
503
504         task_lock(task);
505         files = task->files;
506         if (files)
507                 atomic_inc(&files->count);
508         task_unlock(task);
509
510         return files;
511 }
512
513 void put_files_struct(struct files_struct *files)
514 {
515         struct fdtable *fdt;
516
517         if (atomic_dec_and_test(&files->count)) {
518                 close_files(files);
519                 /*
520                  * Free the fd and fdset arrays if we expanded them.
521                  * If the fdtable was embedded, pass files for freeing
522                  * at the end of the RCU grace period. Otherwise,
523                  * you can free files immediately.
524                  */
525                 fdt = files_fdtable(files);
526                 if (fdt != &files->fdtab)
527                         kmem_cache_free(files_cachep, files);
528                 free_fdtable(fdt);
529         }
530 }
531
532 void reset_files_struct(struct files_struct *files)
533 {
534         struct task_struct *tsk = current;
535         struct files_struct *old;
536
537         old = tsk->files;
538         task_lock(tsk);
539         tsk->files = files;
540         task_unlock(tsk);
541         put_files_struct(old);
542 }
543
544 void exit_files(struct task_struct *tsk)
545 {
546         struct files_struct * files = tsk->files;
547
548         if (files) {
549                 task_lock(tsk);
550                 tsk->files = NULL;
551                 task_unlock(tsk);
552                 put_files_struct(files);
553         }
554 }
555
556 void put_fs_struct(struct fs_struct *fs)
557 {
558         /* No need to hold fs->lock if we are killing it */
559         if (atomic_dec_and_test(&fs->count)) {
560                 path_put(&fs->root);
561                 path_put(&fs->pwd);
562                 kmem_cache_free(fs_cachep, fs);
563         }
564 }
565
566 void exit_fs(struct task_struct *tsk)
567 {
568         struct fs_struct * fs = tsk->fs;
569
570         if (fs) {
571                 task_lock(tsk);
572                 tsk->fs = NULL;
573                 task_unlock(tsk);
574                 put_fs_struct(fs);
575         }
576 }
577
578 EXPORT_SYMBOL_GPL(exit_fs);
579
580 #ifdef CONFIG_MM_OWNER
581 /*
582  * Task p is exiting and it owned mm, lets find a new owner for it
583  */
584 static inline int
585 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
586 {
587         /*
588          * If there are other users of the mm and the owner (us) is exiting
589          * we need to find a new owner to take on the responsibility.
590          */
591         if (atomic_read(&mm->mm_users) <= 1)
592                 return 0;
593         if (mm->owner != p)
594                 return 0;
595         return 1;
596 }
597
598 void mm_update_next_owner(struct mm_struct *mm)
599 {
600         struct task_struct *c, *g, *p = current;
601
602 retry:
603         if (!mm_need_new_owner(mm, p))
604                 return;
605
606         read_lock(&tasklist_lock);
607         /*
608          * Search in the children
609          */
610         list_for_each_entry(c, &p->children, sibling) {
611                 if (c->mm == mm)
612                         goto assign_new_owner;
613         }
614
615         /*
616          * Search in the siblings
617          */
618         list_for_each_entry(c, &p->parent->children, sibling) {
619                 if (c->mm == mm)
620                         goto assign_new_owner;
621         }
622
623         /*
624          * Search through everything else. We should not get
625          * here often
626          */
627         do_each_thread(g, c) {
628                 if (c->mm == mm)
629                         goto assign_new_owner;
630         } while_each_thread(g, c);
631
632         read_unlock(&tasklist_lock);
633         /*
634          * We found no owner yet mm_users > 1: this implies that we are
635          * most likely racing with swapoff (try_to_unuse()) or /proc or
636          * ptrace or page migration (get_task_mm()).  Mark owner as NULL,
637          * so that subsystems can understand the callback and take action.
638          */
639         down_write(&mm->mmap_sem);
640         cgroup_mm_owner_callbacks(mm->owner, NULL);
641         mm->owner = NULL;
642         up_write(&mm->mmap_sem);
643         return;
644
645 assign_new_owner:
646         BUG_ON(c == p);
647         get_task_struct(c);
648         read_unlock(&tasklist_lock);
649         down_write(&mm->mmap_sem);
650         /*
651          * The task_lock protects c->mm from changing.
652          * We always want mm->owner->mm == mm
653          */
654         task_lock(c);
655         if (c->mm != mm) {
656                 task_unlock(c);
657                 up_write(&mm->mmap_sem);
658                 put_task_struct(c);
659                 goto retry;
660         }
661         cgroup_mm_owner_callbacks(mm->owner, c);
662         mm->owner = c;
663         task_unlock(c);
664         up_write(&mm->mmap_sem);
665         put_task_struct(c);
666 }
667 #endif /* CONFIG_MM_OWNER */
668
669 /*
670  * Turn us into a lazy TLB process if we
671  * aren't already..
672  */
673 static void exit_mm(struct task_struct * tsk)
674 {
675         struct mm_struct *mm = tsk->mm;
676         struct core_state *core_state;
677
678         mm_release(tsk, mm);
679         if (!mm)
680                 return;
681         /*
682          * Serialize with any possible pending coredump.
683          * We must hold mmap_sem around checking core_state
684          * and clearing tsk->mm.  The core-inducing thread
685          * will increment ->nr_threads for each thread in the
686          * group with ->mm != NULL.
687          */
688         down_read(&mm->mmap_sem);
689         core_state = mm->core_state;
690         if (core_state) {
691                 struct core_thread self;
692                 up_read(&mm->mmap_sem);
693
694                 self.task = tsk;
695                 self.next = xchg(&core_state->dumper.next, &self);
696                 /*
697                  * Implies mb(), the result of xchg() must be visible
698                  * to core_state->dumper.
699                  */
700                 if (atomic_dec_and_test(&core_state->nr_threads))
701                         complete(&core_state->startup);
702
703                 for (;;) {
704                         set_task_state(tsk, TASK_UNINTERRUPTIBLE);
705                         if (!self.task) /* see coredump_finish() */
706                                 break;
707                         schedule();
708                 }
709                 __set_task_state(tsk, TASK_RUNNING);
710                 down_read(&mm->mmap_sem);
711         }
712         atomic_inc(&mm->mm_count);
713         BUG_ON(mm != tsk->active_mm);
714         /* more a memory barrier than a real lock */
715         task_lock(tsk);
716         tsk->mm = NULL;
717         up_read(&mm->mmap_sem);
718         enter_lazy_tlb(mm, current);
719         /* We don't want this task to be frozen prematurely */
720         clear_freeze_flag(tsk);
721         task_unlock(tsk);
722         mm_update_next_owner(mm);
723         mmput(mm);
724 }
725
726 /*
727  * Return nonzero if @parent's children should reap themselves.
728  *
729  * Called with write_lock_irq(&tasklist_lock) held.
730  */
731 static int ignoring_children(struct task_struct *parent)
732 {
733         int ret;
734         struct sighand_struct *psig = parent->sighand;
735         unsigned long flags;
736         spin_lock_irqsave(&psig->siglock, flags);
737         ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
738                (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
739         spin_unlock_irqrestore(&psig->siglock, flags);
740         return ret;
741 }
742
743 /*
744  * Detach all tasks we were using ptrace on.
745  * Any that need to be release_task'd are put on the @dead list.
746  *
747  * Called with write_lock(&tasklist_lock) held.
748  */
749 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
750 {
751         struct task_struct *p, *n;
752         int ign = -1;
753
754         list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
755                 __ptrace_unlink(p);
756
757                 if (p->exit_state != EXIT_ZOMBIE)
758                         continue;
759
760                 /*
761                  * If it's a zombie, our attachedness prevented normal
762                  * parent notification or self-reaping.  Do notification
763                  * now if it would have happened earlier.  If it should
764                  * reap itself, add it to the @dead list.  We can't call
765                  * release_task() here because we already hold tasklist_lock.
766                  *
767                  * If it's our own child, there is no notification to do.
768                  * But if our normal children self-reap, then this child
769                  * was prevented by ptrace and we must reap it now.
770                  */
771                 if (!task_detached(p) && thread_group_empty(p)) {
772                         if (!same_thread_group(p->real_parent, parent))
773                                 do_notify_parent(p, p->exit_signal);
774                         else {
775                                 if (ign < 0)
776                                         ign = ignoring_children(parent);
777                                 if (ign)
778                                         p->exit_signal = -1;
779                         }
780                 }
781
782                 if (task_detached(p)) {
783                         /*
784                          * Mark it as in the process of being reaped.
785                          */
786                         p->exit_state = EXIT_DEAD;
787                         list_add(&p->ptrace_entry, dead);
788                 }
789         }
790 }
791
792 /*
793  * Finish up exit-time ptrace cleanup.
794  *
795  * Called without locks.
796  */
797 static void ptrace_exit_finish(struct task_struct *parent,
798                                struct list_head *dead)
799 {
800         struct task_struct *p, *n;
801
802         BUG_ON(!list_empty(&parent->ptraced));
803
804         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
805                 list_del_init(&p->ptrace_entry);
806                 release_task(p);
807         }
808 }
809
810 static void reparent_thread(struct task_struct *p, struct task_struct *father)
811 {
812         if (p->pdeath_signal)
813                 /* We already hold the tasklist_lock here.  */
814                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
815
816         list_move_tail(&p->sibling, &p->real_parent->children);
817
818         /* If this is a threaded reparent there is no need to
819          * notify anyone anything has happened.
820          */
821         if (same_thread_group(p->real_parent, father))
822                 return;
823
824         /* We don't want people slaying init.  */
825         if (!task_detached(p))
826                 p->exit_signal = SIGCHLD;
827
828         /* If we'd notified the old parent about this child's death,
829          * also notify the new parent.
830          */
831         if (!ptrace_reparented(p) &&
832             p->exit_state == EXIT_ZOMBIE &&
833             !task_detached(p) && thread_group_empty(p))
834                 do_notify_parent(p, p->exit_signal);
835
836         kill_orphaned_pgrp(p, father);
837 }
838
839 /*
840  * When we die, we re-parent all our children.
841  * Try to give them to another thread in our thread
842  * group, and if no such member exists, give it to
843  * the child reaper process (ie "init") in our pid
844  * space.
845  */
846 static struct task_struct *find_new_reaper(struct task_struct *father)
847 {
848         struct pid_namespace *pid_ns = task_active_pid_ns(father);
849         struct task_struct *thread;
850
851         thread = father;
852         while_each_thread(father, thread) {
853                 if (thread->flags & PF_EXITING)
854                         continue;
855                 if (unlikely(pid_ns->child_reaper == father))
856                         pid_ns->child_reaper = thread;
857                 return thread;
858         }
859
860         if (unlikely(pid_ns->child_reaper == father)) {
861                 write_unlock_irq(&tasklist_lock);
862                 if (unlikely(pid_ns == &init_pid_ns))
863                         panic("Attempted to kill init!");
864
865                 zap_pid_ns_processes(pid_ns);
866                 write_lock_irq(&tasklist_lock);
867                 /*
868                  * We can not clear ->child_reaper or leave it alone.
869                  * There may by stealth EXIT_DEAD tasks on ->children,
870                  * forget_original_parent() must move them somewhere.
871                  */
872                 pid_ns->child_reaper = init_pid_ns.child_reaper;
873         }
874
875         return pid_ns->child_reaper;
876 }
877
878 static void forget_original_parent(struct task_struct *father)
879 {
880         struct task_struct *p, *n, *reaper;
881         LIST_HEAD(ptrace_dead);
882
883         write_lock_irq(&tasklist_lock);
884         reaper = find_new_reaper(father);
885         /*
886          * First clean up ptrace if we were using it.
887          */
888         ptrace_exit(father, &ptrace_dead);
889
890         list_for_each_entry_safe(p, n, &father->children, sibling) {
891                 p->real_parent = reaper;
892                 if (p->parent == father) {
893                         BUG_ON(p->ptrace);
894                         p->parent = p->real_parent;
895                 }
896                 reparent_thread(p, father);
897         }
898
899         write_unlock_irq(&tasklist_lock);
900         BUG_ON(!list_empty(&father->children));
901
902         ptrace_exit_finish(father, &ptrace_dead);
903 }
904
905 /*
906  * Send signals to all our closest relatives so that they know
907  * to properly mourn us..
908  */
909 static void exit_notify(struct task_struct *tsk, int group_dead)
910 {
911         int signal;
912         void *cookie;
913
914         /*
915          * This does two things:
916          *
917          * A.  Make init inherit all the child processes
918          * B.  Check to see if any process groups have become orphaned
919          *      as a result of our exiting, and if they have any stopped
920          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
921          */
922         forget_original_parent(tsk);
923         exit_task_namespaces(tsk);
924
925         write_lock_irq(&tasklist_lock);
926         if (group_dead)
927                 kill_orphaned_pgrp(tsk->group_leader, NULL);
928
929         /* Let father know we died
930          *
931          * Thread signals are configurable, but you aren't going to use
932          * that to send signals to arbitary processes.
933          * That stops right now.
934          *
935          * If the parent exec id doesn't match the exec id we saved
936          * when we started then we know the parent has changed security
937          * domain.
938          *
939          * If our self_exec id doesn't match our parent_exec_id then
940          * we have changed execution domain as these two values started
941          * the same after a fork.
942          */
943         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
944             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
945              tsk->self_exec_id != tsk->parent_exec_id) &&
946             !capable(CAP_KILL))
947                 tsk->exit_signal = SIGCHLD;
948
949         signal = tracehook_notify_death(tsk, &cookie, group_dead);
950         if (signal >= 0)
951                 signal = do_notify_parent(tsk, signal);
952
953         tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
954
955         /* mt-exec, de_thread() is waiting for us */
956         if (thread_group_leader(tsk) &&
957             tsk->signal->group_exit_task &&
958             tsk->signal->notify_count < 0)
959                 wake_up_process(tsk->signal->group_exit_task);
960
961         write_unlock_irq(&tasklist_lock);
962
963         tracehook_report_death(tsk, signal, cookie, group_dead);
964
965         /* If the process is dead, release it - nobody will wait for it */
966         if (signal == DEATH_REAP)
967                 release_task(tsk);
968 }
969
970 #ifdef CONFIG_DEBUG_STACK_USAGE
971 static void check_stack_usage(void)
972 {
973         static DEFINE_SPINLOCK(low_water_lock);
974         static int lowest_to_date = THREAD_SIZE;
975         unsigned long *n = end_of_stack(current);
976         unsigned long free;
977
978         while (*n == 0)
979                 n++;
980         free = (unsigned long)n - (unsigned long)end_of_stack(current);
981
982         if (free >= lowest_to_date)
983                 return;
984
985         spin_lock(&low_water_lock);
986         if (free < lowest_to_date) {
987                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
988                                 "left\n",
989                                 current->comm, free);
990                 lowest_to_date = free;
991         }
992         spin_unlock(&low_water_lock);
993 }
994 #else
995 static inline void check_stack_usage(void) {}
996 #endif
997
998 NORET_TYPE void do_exit(long code)
999 {
1000         struct task_struct *tsk = current;
1001         int group_dead;
1002
1003         profile_task_exit(tsk);
1004
1005         WARN_ON(atomic_read(&tsk->fs_excl));
1006
1007         if (unlikely(in_interrupt()))
1008                 panic("Aiee, killing interrupt handler!");
1009         if (unlikely(!tsk->pid))
1010                 panic("Attempted to kill the idle task!");
1011
1012         tracehook_report_exit(&code);
1013
1014         /*
1015          * We're taking recursive faults here in do_exit. Safest is to just
1016          * leave this task alone and wait for reboot.
1017          */
1018         if (unlikely(tsk->flags & PF_EXITING)) {
1019                 printk(KERN_ALERT
1020                         "Fixing recursive fault but reboot is needed!\n");
1021                 /*
1022                  * We can do this unlocked here. The futex code uses
1023                  * this flag just to verify whether the pi state
1024                  * cleanup has been done or not. In the worst case it
1025                  * loops once more. We pretend that the cleanup was
1026                  * done as there is no way to return. Either the
1027                  * OWNER_DIED bit is set by now or we push the blocked
1028                  * task into the wait for ever nirwana as well.
1029                  */
1030                 tsk->flags |= PF_EXITPIDONE;
1031                 if (tsk->io_context)
1032                         exit_io_context();
1033                 set_current_state(TASK_UNINTERRUPTIBLE);
1034                 schedule();
1035         }
1036
1037         exit_signals(tsk);  /* sets PF_EXITING */
1038         /*
1039          * tsk->flags are checked in the futex code to protect against
1040          * an exiting task cleaning up the robust pi futexes.
1041          */
1042         smp_mb();
1043         spin_unlock_wait(&tsk->pi_lock);
1044
1045         if (unlikely(in_atomic()))
1046                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1047                                 current->comm, task_pid_nr(current),
1048                                 preempt_count());
1049
1050         acct_update_integrals(tsk);
1051         if (tsk->mm) {
1052                 update_hiwater_rss(tsk->mm);
1053                 update_hiwater_vm(tsk->mm);
1054         }
1055         group_dead = atomic_dec_and_test(&tsk->signal->live);
1056         if (group_dead) {
1057                 hrtimer_cancel(&tsk->signal->real_timer);
1058                 exit_itimers(tsk->signal);
1059         }
1060         acct_collect(code, group_dead);
1061         if (group_dead)
1062                 tty_audit_exit();
1063         if (unlikely(tsk->audit_context))
1064                 audit_free(tsk);
1065
1066         tsk->exit_code = code;
1067         taskstats_exit(tsk, group_dead);
1068
1069         exit_mm(tsk);
1070
1071         if (group_dead)
1072                 acct_process();
1073         trace_sched_process_exit(tsk);
1074
1075         exit_sem(tsk);
1076         exit_files(tsk);
1077         exit_fs(tsk);
1078         check_stack_usage();
1079         exit_thread();
1080         cgroup_exit(tsk, 1);
1081         exit_keys(tsk);
1082
1083         if (group_dead && tsk->signal->leader)
1084                 disassociate_ctty(1);
1085
1086         module_put(task_thread_info(tsk)->exec_domain->module);
1087         if (tsk->binfmt)
1088                 module_put(tsk->binfmt->module);
1089
1090         proc_exit_connector(tsk);
1091         exit_notify(tsk, group_dead);
1092 #ifdef CONFIG_NUMA
1093         mpol_put(tsk->mempolicy);
1094         tsk->mempolicy = NULL;
1095 #endif
1096 #ifdef CONFIG_FUTEX
1097         /*
1098          * This must happen late, after the PID is not
1099          * hashed anymore:
1100          */
1101         if (unlikely(!list_empty(&tsk->pi_state_list)))
1102                 exit_pi_state_list(tsk);
1103         if (unlikely(current->pi_state_cache))
1104                 kfree(current->pi_state_cache);
1105 #endif
1106         /*
1107          * Make sure we are holding no locks:
1108          */
1109         debug_check_no_locks_held(tsk);
1110         /*
1111          * We can do this unlocked here. The futex code uses this flag
1112          * just to verify whether the pi state cleanup has been done
1113          * or not. In the worst case it loops once more.
1114          */
1115         tsk->flags |= PF_EXITPIDONE;
1116
1117         if (tsk->io_context)
1118                 exit_io_context();
1119
1120         if (tsk->splice_pipe)
1121                 __free_pipe_info(tsk->splice_pipe);
1122
1123         preempt_disable();
1124         /* causes final put_task_struct in finish_task_switch(). */
1125         tsk->state = TASK_DEAD;
1126
1127         schedule();
1128         BUG();
1129         /* Avoid "noreturn function does return".  */
1130         for (;;)
1131                 cpu_relax();    /* For when BUG is null */
1132 }
1133
1134 EXPORT_SYMBOL_GPL(do_exit);
1135
1136 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1137 {
1138         if (comp)
1139                 complete(comp);
1140
1141         do_exit(code);
1142 }
1143
1144 EXPORT_SYMBOL(complete_and_exit);
1145
1146 asmlinkage long sys_exit(int error_code)
1147 {
1148         do_exit((error_code&0xff)<<8);
1149 }
1150
1151 /*
1152  * Take down every thread in the group.  This is called by fatal signals
1153  * as well as by sys_exit_group (below).
1154  */
1155 NORET_TYPE void
1156 do_group_exit(int exit_code)
1157 {
1158         struct signal_struct *sig = current->signal;
1159
1160         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1161
1162         if (signal_group_exit(sig))
1163                 exit_code = sig->group_exit_code;
1164         else if (!thread_group_empty(current)) {
1165                 struct sighand_struct *const sighand = current->sighand;
1166                 spin_lock_irq(&sighand->siglock);
1167                 if (signal_group_exit(sig))
1168                         /* Another thread got here before we took the lock.  */
1169                         exit_code = sig->group_exit_code;
1170                 else {
1171                         sig->group_exit_code = exit_code;
1172                         sig->flags = SIGNAL_GROUP_EXIT;
1173                         zap_other_threads(current);
1174                 }
1175                 spin_unlock_irq(&sighand->siglock);
1176         }
1177
1178         do_exit(exit_code);
1179         /* NOTREACHED */
1180 }
1181
1182 /*
1183  * this kills every thread in the thread group. Note that any externally
1184  * wait4()-ing process will get the correct exit code - even if this
1185  * thread is not the thread group leader.
1186  */
1187 asmlinkage void sys_exit_group(int error_code)
1188 {
1189         do_group_exit((error_code & 0xff) << 8);
1190 }
1191
1192 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1193 {
1194         struct pid *pid = NULL;
1195         if (type == PIDTYPE_PID)
1196                 pid = task->pids[type].pid;
1197         else if (type < PIDTYPE_MAX)
1198                 pid = task->group_leader->pids[type].pid;
1199         return pid;
1200 }
1201
1202 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1203                           struct task_struct *p)
1204 {
1205         int err;
1206
1207         if (type < PIDTYPE_MAX) {
1208                 if (task_pid_type(p, type) != pid)
1209                         return 0;
1210         }
1211
1212         /* Wait for all children (clone and not) if __WALL is set;
1213          * otherwise, wait for clone children *only* if __WCLONE is
1214          * set; otherwise, wait for non-clone children *only*.  (Note:
1215          * A "clone" child here is one that reports to its parent
1216          * using a signal other than SIGCHLD.) */
1217         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1218             && !(options & __WALL))
1219                 return 0;
1220
1221         err = security_task_wait(p);
1222         if (err)
1223                 return err;
1224
1225         return 1;
1226 }
1227
1228 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1229                                int why, int status,
1230                                struct siginfo __user *infop,
1231                                struct rusage __user *rusagep)
1232 {
1233         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1234
1235         put_task_struct(p);
1236         if (!retval)
1237                 retval = put_user(SIGCHLD, &infop->si_signo);
1238         if (!retval)
1239                 retval = put_user(0, &infop->si_errno);
1240         if (!retval)
1241                 retval = put_user((short)why, &infop->si_code);
1242         if (!retval)
1243                 retval = put_user(pid, &infop->si_pid);
1244         if (!retval)
1245                 retval = put_user(uid, &infop->si_uid);
1246         if (!retval)
1247                 retval = put_user(status, &infop->si_status);
1248         if (!retval)
1249                 retval = pid;
1250         return retval;
1251 }
1252
1253 /*
1254  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1255  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1256  * the lock and this task is uninteresting.  If we return nonzero, we have
1257  * released the lock and the system call should return.
1258  */
1259 static int wait_task_zombie(struct task_struct *p, int options,
1260                             struct siginfo __user *infop,
1261                             int __user *stat_addr, struct rusage __user *ru)
1262 {
1263         unsigned long state;
1264         int retval, status, traced;
1265         pid_t pid = task_pid_vnr(p);
1266
1267         if (!likely(options & WEXITED))
1268                 return 0;
1269
1270         if (unlikely(options & WNOWAIT)) {
1271                 uid_t uid = p->uid;
1272                 int exit_code = p->exit_code;
1273                 int why, status;
1274
1275                 get_task_struct(p);
1276                 read_unlock(&tasklist_lock);
1277                 if ((exit_code & 0x7f) == 0) {
1278                         why = CLD_EXITED;
1279                         status = exit_code >> 8;
1280                 } else {
1281                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1282                         status = exit_code & 0x7f;
1283                 }
1284                 return wait_noreap_copyout(p, pid, uid, why,
1285                                            status, infop, ru);
1286         }
1287
1288         /*
1289          * Try to move the task's state to DEAD
1290          * only one thread is allowed to do this:
1291          */
1292         state = xchg(&p->exit_state, EXIT_DEAD);
1293         if (state != EXIT_ZOMBIE) {
1294                 BUG_ON(state != EXIT_DEAD);
1295                 return 0;
1296         }
1297
1298         traced = ptrace_reparented(p);
1299
1300         if (likely(!traced)) {
1301                 struct signal_struct *psig;
1302                 struct signal_struct *sig;
1303                 struct task_cputime cputime;
1304
1305                 /*
1306                  * The resource counters for the group leader are in its
1307                  * own task_struct.  Those for dead threads in the group
1308                  * are in its signal_struct, as are those for the child
1309                  * processes it has previously reaped.  All these
1310                  * accumulate in the parent's signal_struct c* fields.
1311                  *
1312                  * We don't bother to take a lock here to protect these
1313                  * p->signal fields, because they are only touched by
1314                  * __exit_signal, which runs with tasklist_lock
1315                  * write-locked anyway, and so is excluded here.  We do
1316                  * need to protect the access to p->parent->signal fields,
1317                  * as other threads in the parent group can be right
1318                  * here reaping other children at the same time.
1319                  *
1320                  * We use thread_group_cputime() to get times for the thread
1321                  * group, which consolidates times for all threads in the
1322                  * group including the group leader.
1323                  */
1324                 thread_group_cputime(p, &cputime);
1325                 spin_lock_irq(&p->parent->sighand->siglock);
1326                 psig = p->parent->signal;
1327                 sig = p->signal;
1328                 psig->cutime =
1329                         cputime_add(psig->cutime,
1330                         cputime_add(cputime.utime,
1331                                     sig->cutime));
1332                 psig->cstime =
1333                         cputime_add(psig->cstime,
1334                         cputime_add(cputime.stime,
1335                                     sig->cstime));
1336                 psig->cgtime =
1337                         cputime_add(psig->cgtime,
1338                         cputime_add(p->gtime,
1339                         cputime_add(sig->gtime,
1340                                     sig->cgtime)));
1341                 psig->cmin_flt +=
1342                         p->min_flt + sig->min_flt + sig->cmin_flt;
1343                 psig->cmaj_flt +=
1344                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1345                 psig->cnvcsw +=
1346                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1347                 psig->cnivcsw +=
1348                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1349                 psig->cinblock +=
1350                         task_io_get_inblock(p) +
1351                         sig->inblock + sig->cinblock;
1352                 psig->coublock +=
1353                         task_io_get_oublock(p) +
1354                         sig->oublock + sig->coublock;
1355                 task_io_accounting_add(&psig->ioac, &p->ioac);
1356                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1357                 spin_unlock_irq(&p->parent->sighand->siglock);
1358         }
1359
1360         /*
1361          * Now we are sure this task is interesting, and no other
1362          * thread can reap it because we set its state to EXIT_DEAD.
1363          */
1364         read_unlock(&tasklist_lock);
1365
1366         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1367         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1368                 ? p->signal->group_exit_code : p->exit_code;
1369         if (!retval && stat_addr)
1370                 retval = put_user(status, stat_addr);
1371         if (!retval && infop)
1372                 retval = put_user(SIGCHLD, &infop->si_signo);
1373         if (!retval && infop)
1374                 retval = put_user(0, &infop->si_errno);
1375         if (!retval && infop) {
1376                 int why;
1377
1378                 if ((status & 0x7f) == 0) {
1379                         why = CLD_EXITED;
1380                         status >>= 8;
1381                 } else {
1382                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1383                         status &= 0x7f;
1384                 }
1385                 retval = put_user((short)why, &infop->si_code);
1386                 if (!retval)
1387                         retval = put_user(status, &infop->si_status);
1388         }
1389         if (!retval && infop)
1390                 retval = put_user(pid, &infop->si_pid);
1391         if (!retval && infop)
1392                 retval = put_user(p->uid, &infop->si_uid);
1393         if (!retval)
1394                 retval = pid;
1395
1396         if (traced) {
1397                 write_lock_irq(&tasklist_lock);
1398                 /* We dropped tasklist, ptracer could die and untrace */
1399                 ptrace_unlink(p);
1400                 /*
1401                  * If this is not a detached task, notify the parent.
1402                  * If it's still not detached after that, don't release
1403                  * it now.
1404                  */
1405                 if (!task_detached(p)) {
1406                         do_notify_parent(p, p->exit_signal);
1407                         if (!task_detached(p)) {
1408                                 p->exit_state = EXIT_ZOMBIE;
1409                                 p = NULL;
1410                         }
1411                 }
1412                 write_unlock_irq(&tasklist_lock);
1413         }
1414         if (p != NULL)
1415                 release_task(p);
1416
1417         return retval;
1418 }
1419
1420 /*
1421  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1422  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1423  * the lock and this task is uninteresting.  If we return nonzero, we have
1424  * released the lock and the system call should return.
1425  */
1426 static int wait_task_stopped(int ptrace, struct task_struct *p,
1427                              int options, struct siginfo __user *infop,
1428                              int __user *stat_addr, struct rusage __user *ru)
1429 {
1430         int retval, exit_code, why;
1431         uid_t uid = 0; /* unneeded, required by compiler */
1432         pid_t pid;
1433
1434         if (!(options & WUNTRACED))
1435                 return 0;
1436
1437         exit_code = 0;
1438         spin_lock_irq(&p->sighand->siglock);
1439
1440         if (unlikely(!task_is_stopped_or_traced(p)))
1441                 goto unlock_sig;
1442
1443         if (!ptrace && p->signal->group_stop_count > 0)
1444                 /*
1445                  * A group stop is in progress and this is the group leader.
1446                  * We won't report until all threads have stopped.
1447                  */
1448                 goto unlock_sig;
1449
1450         exit_code = p->exit_code;
1451         if (!exit_code)
1452                 goto unlock_sig;
1453
1454         if (!unlikely(options & WNOWAIT))
1455                 p->exit_code = 0;
1456
1457         uid = p->uid;
1458 unlock_sig:
1459         spin_unlock_irq(&p->sighand->siglock);
1460         if (!exit_code)
1461                 return 0;
1462
1463         /*
1464          * Now we are pretty sure this task is interesting.
1465          * Make sure it doesn't get reaped out from under us while we
1466          * give up the lock and then examine it below.  We don't want to
1467          * keep holding onto the tasklist_lock while we call getrusage and
1468          * possibly take page faults for user memory.
1469          */
1470         get_task_struct(p);
1471         pid = task_pid_vnr(p);
1472         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1473         read_unlock(&tasklist_lock);
1474
1475         if (unlikely(options & WNOWAIT))
1476                 return wait_noreap_copyout(p, pid, uid,
1477                                            why, exit_code,
1478                                            infop, ru);
1479
1480         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1481         if (!retval && stat_addr)
1482                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1483         if (!retval && infop)
1484                 retval = put_user(SIGCHLD, &infop->si_signo);
1485         if (!retval && infop)
1486                 retval = put_user(0, &infop->si_errno);
1487         if (!retval && infop)
1488                 retval = put_user((short)why, &infop->si_code);
1489         if (!retval && infop)
1490                 retval = put_user(exit_code, &infop->si_status);
1491         if (!retval && infop)
1492                 retval = put_user(pid, &infop->si_pid);
1493         if (!retval && infop)
1494                 retval = put_user(uid, &infop->si_uid);
1495         if (!retval)
1496                 retval = pid;
1497         put_task_struct(p);
1498
1499         BUG_ON(!retval);
1500         return retval;
1501 }
1502
1503 /*
1504  * Handle do_wait work for one task in a live, non-stopped state.
1505  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1506  * the lock and this task is uninteresting.  If we return nonzero, we have
1507  * released the lock and the system call should return.
1508  */
1509 static int wait_task_continued(struct task_struct *p, int options,
1510                                struct siginfo __user *infop,
1511                                int __user *stat_addr, struct rusage __user *ru)
1512 {
1513         int retval;
1514         pid_t pid;
1515         uid_t uid;
1516
1517         if (!unlikely(options & WCONTINUED))
1518                 return 0;
1519
1520         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1521                 return 0;
1522
1523         spin_lock_irq(&p->sighand->siglock);
1524         /* Re-check with the lock held.  */
1525         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1526                 spin_unlock_irq(&p->sighand->siglock);
1527                 return 0;
1528         }
1529         if (!unlikely(options & WNOWAIT))
1530                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1531         spin_unlock_irq(&p->sighand->siglock);
1532
1533         pid = task_pid_vnr(p);
1534         uid = p->uid;
1535         get_task_struct(p);
1536         read_unlock(&tasklist_lock);
1537
1538         if (!infop) {
1539                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1540                 put_task_struct(p);
1541                 if (!retval && stat_addr)
1542                         retval = put_user(0xffff, stat_addr);
1543                 if (!retval)
1544                         retval = pid;
1545         } else {
1546                 retval = wait_noreap_copyout(p, pid, uid,
1547                                              CLD_CONTINUED, SIGCONT,
1548                                              infop, ru);
1549                 BUG_ON(retval == 0);
1550         }
1551
1552         return retval;
1553 }
1554
1555 /*
1556  * Consider @p for a wait by @parent.
1557  *
1558  * -ECHILD should be in *@notask_error before the first call.
1559  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1560  * Returns zero if the search for a child should continue;
1561  * then *@notask_error is 0 if @p is an eligible child,
1562  * or another error from security_task_wait(), or still -ECHILD.
1563  */
1564 static int wait_consider_task(struct task_struct *parent, int ptrace,
1565                               struct task_struct *p, int *notask_error,
1566                               enum pid_type type, struct pid *pid, int options,
1567                               struct siginfo __user *infop,
1568                               int __user *stat_addr, struct rusage __user *ru)
1569 {
1570         int ret = eligible_child(type, pid, options, p);
1571         if (!ret)
1572                 return ret;
1573
1574         if (unlikely(ret < 0)) {
1575                 /*
1576                  * If we have not yet seen any eligible child,
1577                  * then let this error code replace -ECHILD.
1578                  * A permission error will give the user a clue
1579                  * to look for security policy problems, rather
1580                  * than for mysterious wait bugs.
1581                  */
1582                 if (*notask_error)
1583                         *notask_error = ret;
1584         }
1585
1586         if (likely(!ptrace) && unlikely(p->ptrace)) {
1587                 /*
1588                  * This child is hidden by ptrace.
1589                  * We aren't allowed to see it now, but eventually we will.
1590                  */
1591                 *notask_error = 0;
1592                 return 0;
1593         }
1594
1595         if (p->exit_state == EXIT_DEAD)
1596                 return 0;
1597
1598         /*
1599          * We don't reap group leaders with subthreads.
1600          */
1601         if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1602                 return wait_task_zombie(p, options, infop, stat_addr, ru);
1603
1604         /*
1605          * It's stopped or running now, so it might
1606          * later continue, exit, or stop again.
1607          */
1608         *notask_error = 0;
1609
1610         if (task_is_stopped_or_traced(p))
1611                 return wait_task_stopped(ptrace, p, options,
1612                                          infop, stat_addr, ru);
1613
1614         return wait_task_continued(p, options, infop, stat_addr, ru);
1615 }
1616
1617 /*
1618  * Do the work of do_wait() for one thread in the group, @tsk.
1619  *
1620  * -ECHILD should be in *@notask_error before the first call.
1621  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1622  * Returns zero if the search for a child should continue; then
1623  * *@notask_error is 0 if there were any eligible children,
1624  * or another error from security_task_wait(), or still -ECHILD.
1625  */
1626 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1627                           enum pid_type type, struct pid *pid, int options,
1628                           struct siginfo __user *infop, int __user *stat_addr,
1629                           struct rusage __user *ru)
1630 {
1631         struct task_struct *p;
1632
1633         list_for_each_entry(p, &tsk->children, sibling) {
1634                 /*
1635                  * Do not consider detached threads.
1636                  */
1637                 if (!task_detached(p)) {
1638                         int ret = wait_consider_task(tsk, 0, p, notask_error,
1639                                                      type, pid, options,
1640                                                      infop, stat_addr, ru);
1641                         if (ret)
1642                                 return ret;
1643                 }
1644         }
1645
1646         return 0;
1647 }
1648
1649 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1650                           enum pid_type type, struct pid *pid, int options,
1651                           struct siginfo __user *infop, int __user *stat_addr,
1652                           struct rusage __user *ru)
1653 {
1654         struct task_struct *p;
1655
1656         /*
1657          * Traditionally we see ptrace'd stopped tasks regardless of options.
1658          */
1659         options |= WUNTRACED;
1660
1661         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1662                 int ret = wait_consider_task(tsk, 1, p, notask_error,
1663                                              type, pid, options,
1664                                              infop, stat_addr, ru);
1665                 if (ret)
1666                         return ret;
1667         }
1668
1669         return 0;
1670 }
1671
1672 static long do_wait(enum pid_type type, struct pid *pid, int options,
1673                     struct siginfo __user *infop, int __user *stat_addr,
1674                     struct rusage __user *ru)
1675 {
1676         DECLARE_WAITQUEUE(wait, current);
1677         struct task_struct *tsk;
1678         int retval;
1679
1680         trace_sched_process_wait(pid);
1681
1682         add_wait_queue(&current->signal->wait_chldexit,&wait);
1683 repeat:
1684         /*
1685          * If there is nothing that can match our critiera just get out.
1686          * We will clear @retval to zero if we see any child that might later
1687          * match our criteria, even if we are not able to reap it yet.
1688          */
1689         retval = -ECHILD;
1690         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1691                 goto end;
1692
1693         current->state = TASK_INTERRUPTIBLE;
1694         read_lock(&tasklist_lock);
1695         tsk = current;
1696         do {
1697                 int tsk_result = do_wait_thread(tsk, &retval,
1698                                                 type, pid, options,
1699                                                 infop, stat_addr, ru);
1700                 if (!tsk_result)
1701                         tsk_result = ptrace_do_wait(tsk, &retval,
1702                                                     type, pid, options,
1703                                                     infop, stat_addr, ru);
1704                 if (tsk_result) {
1705                         /*
1706                          * tasklist_lock is unlocked and we have a final result.
1707                          */
1708                         retval = tsk_result;
1709                         goto end;
1710                 }
1711
1712                 if (options & __WNOTHREAD)
1713                         break;
1714                 tsk = next_thread(tsk);
1715                 BUG_ON(tsk->signal != current->signal);
1716         } while (tsk != current);
1717         read_unlock(&tasklist_lock);
1718
1719         if (!retval && !(options & WNOHANG)) {
1720                 retval = -ERESTARTSYS;
1721                 if (!signal_pending(current)) {
1722                         schedule();
1723                         goto repeat;
1724                 }
1725         }
1726
1727 end:
1728         current->state = TASK_RUNNING;
1729         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1730         if (infop) {
1731                 if (retval > 0)
1732                         retval = 0;
1733                 else {
1734                         /*
1735                          * For a WNOHANG return, clear out all the fields
1736                          * we would set so the user can easily tell the
1737                          * difference.
1738                          */
1739                         if (!retval)
1740                                 retval = put_user(0, &infop->si_signo);
1741                         if (!retval)
1742                                 retval = put_user(0, &infop->si_errno);
1743                         if (!retval)
1744                                 retval = put_user(0, &infop->si_code);
1745                         if (!retval)
1746                                 retval = put_user(0, &infop->si_pid);
1747                         if (!retval)
1748                                 retval = put_user(0, &infop->si_uid);
1749                         if (!retval)
1750                                 retval = put_user(0, &infop->si_status);
1751                 }
1752         }
1753         return retval;
1754 }
1755
1756 asmlinkage long sys_waitid(int which, pid_t upid,
1757                            struct siginfo __user *infop, int options,
1758                            struct rusage __user *ru)
1759 {
1760         struct pid *pid = NULL;
1761         enum pid_type type;
1762         long ret;
1763
1764         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1765                 return -EINVAL;
1766         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1767                 return -EINVAL;
1768
1769         switch (which) {
1770         case P_ALL:
1771                 type = PIDTYPE_MAX;
1772                 break;
1773         case P_PID:
1774                 type = PIDTYPE_PID;
1775                 if (upid <= 0)
1776                         return -EINVAL;
1777                 break;
1778         case P_PGID:
1779                 type = PIDTYPE_PGID;
1780                 if (upid <= 0)
1781                         return -EINVAL;
1782                 break;
1783         default:
1784                 return -EINVAL;
1785         }
1786
1787         if (type < PIDTYPE_MAX)
1788                 pid = find_get_pid(upid);
1789         ret = do_wait(type, pid, options, infop, NULL, ru);
1790         put_pid(pid);
1791
1792         /* avoid REGPARM breakage on x86: */
1793         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1794         return ret;
1795 }
1796
1797 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1798                           int options, struct rusage __user *ru)
1799 {
1800         struct pid *pid = NULL;
1801         enum pid_type type;
1802         long ret;
1803
1804         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1805                         __WNOTHREAD|__WCLONE|__WALL))
1806                 return -EINVAL;
1807
1808         if (upid == -1)
1809                 type = PIDTYPE_MAX;
1810         else if (upid < 0) {
1811                 type = PIDTYPE_PGID;
1812                 pid = find_get_pid(-upid);
1813         } else if (upid == 0) {
1814                 type = PIDTYPE_PGID;
1815                 pid = get_pid(task_pgrp(current));
1816         } else /* upid > 0 */ {
1817                 type = PIDTYPE_PID;
1818                 pid = find_get_pid(upid);
1819         }
1820
1821         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1822         put_pid(pid);
1823
1824         /* avoid REGPARM breakage on x86: */
1825         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1826         return ret;
1827 }
1828
1829 #ifdef __ARCH_WANT_SYS_WAITPID
1830
1831 /*
1832  * sys_waitpid() remains for compatibility. waitpid() should be
1833  * implemented by calling sys_wait4() from libc.a.
1834  */
1835 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1836 {
1837         return sys_wait4(pid, stat_addr, options, NULL);
1838 }
1839
1840 #endif