Merge git://git.infradead.org/~dwmw2/random-2.6
[pandora-kernel.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/compat.h>
44 #include <linux/pipe_fs_i.h>
45 #include <linux/audit.h> /* for audit_free() */
46 #include <linux/resource.h>
47 #include <linux/blkdev.h>
48 #include <linux/task_io_accounting_ops.h>
49
50 #include <asm/uaccess.h>
51 #include <asm/unistd.h>
52 #include <asm/pgtable.h>
53 #include <asm/mmu_context.h>
54
55 static void exit_mm(struct task_struct * tsk);
56
57 static inline int task_detached(struct task_struct *p)
58 {
59         return p->exit_signal == -1;
60 }
61
62 static void __unhash_process(struct task_struct *p)
63 {
64         nr_threads--;
65         detach_pid(p, PIDTYPE_PID);
66         if (thread_group_leader(p)) {
67                 detach_pid(p, PIDTYPE_PGID);
68                 detach_pid(p, PIDTYPE_SID);
69
70                 list_del_rcu(&p->tasks);
71                 __get_cpu_var(process_counts)--;
72         }
73         list_del_rcu(&p->thread_group);
74         list_del_init(&p->sibling);
75 }
76
77 /*
78  * This function expects the tasklist_lock write-locked.
79  */
80 static void __exit_signal(struct task_struct *tsk)
81 {
82         struct signal_struct *sig = tsk->signal;
83         struct sighand_struct *sighand;
84
85         BUG_ON(!sig);
86         BUG_ON(!atomic_read(&sig->count));
87
88         sighand = rcu_dereference(tsk->sighand);
89         spin_lock(&sighand->siglock);
90
91         posix_cpu_timers_exit(tsk);
92         if (atomic_dec_and_test(&sig->count))
93                 posix_cpu_timers_exit_group(tsk);
94         else {
95                 /*
96                  * If there is any task waiting for the group exit
97                  * then notify it:
98                  */
99                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
100                         wake_up_process(sig->group_exit_task);
101
102                 if (tsk == sig->curr_target)
103                         sig->curr_target = next_thread(tsk);
104                 /*
105                  * Accumulate here the counters for all threads but the
106                  * group leader as they die, so they can be added into
107                  * the process-wide totals when those are taken.
108                  * The group leader stays around as a zombie as long
109                  * as there are other threads.  When it gets reaped,
110                  * the exit.c code will add its counts into these totals.
111                  * We won't ever get here for the group leader, since it
112                  * will have been the last reference on the signal_struct.
113                  */
114                 sig->utime = cputime_add(sig->utime, tsk->utime);
115                 sig->stime = cputime_add(sig->stime, tsk->stime);
116                 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
117                 sig->min_flt += tsk->min_flt;
118                 sig->maj_flt += tsk->maj_flt;
119                 sig->nvcsw += tsk->nvcsw;
120                 sig->nivcsw += tsk->nivcsw;
121                 sig->inblock += task_io_get_inblock(tsk);
122                 sig->oublock += task_io_get_oublock(tsk);
123 #ifdef CONFIG_TASK_XACCT
124                 sig->rchar += tsk->rchar;
125                 sig->wchar += tsk->wchar;
126                 sig->syscr += tsk->syscr;
127                 sig->syscw += tsk->syscw;
128 #endif /* CONFIG_TASK_XACCT */
129 #ifdef CONFIG_TASK_IO_ACCOUNTING
130                 sig->ioac.read_bytes += tsk->ioac.read_bytes;
131                 sig->ioac.write_bytes += tsk->ioac.write_bytes;
132                 sig->ioac.cancelled_write_bytes +=
133                                         tsk->ioac.cancelled_write_bytes;
134 #endif /* CONFIG_TASK_IO_ACCOUNTING */
135                 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
136                 sig = NULL; /* Marker for below. */
137         }
138
139         __unhash_process(tsk);
140
141         /*
142          * Do this under ->siglock, we can race with another thread
143          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
144          */
145         flush_sigqueue(&tsk->pending);
146
147         tsk->signal = NULL;
148         tsk->sighand = NULL;
149         spin_unlock(&sighand->siglock);
150
151         __cleanup_sighand(sighand);
152         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
153         if (sig) {
154                 flush_sigqueue(&sig->shared_pending);
155                 taskstats_tgid_free(sig);
156                 __cleanup_signal(sig);
157         }
158 }
159
160 static void delayed_put_task_struct(struct rcu_head *rhp)
161 {
162         put_task_struct(container_of(rhp, struct task_struct, rcu));
163 }
164
165 /*
166  * Do final ptrace-related cleanup of a zombie being reaped.
167  *
168  * Called with write_lock(&tasklist_lock) held.
169  */
170 static void ptrace_release_task(struct task_struct *p)
171 {
172         BUG_ON(!list_empty(&p->ptraced));
173         ptrace_unlink(p);
174         BUG_ON(!list_empty(&p->ptrace_entry));
175 }
176
177 void release_task(struct task_struct * p)
178 {
179         struct task_struct *leader;
180         int zap_leader;
181 repeat:
182         atomic_dec(&p->user->processes);
183         proc_flush_task(p);
184         write_lock_irq(&tasklist_lock);
185         ptrace_release_task(p);
186         __exit_signal(p);
187
188         /*
189          * If we are the last non-leader member of the thread
190          * group, and the leader is zombie, then notify the
191          * group leader's parent process. (if it wants notification.)
192          */
193         zap_leader = 0;
194         leader = p->group_leader;
195         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
196                 BUG_ON(task_detached(leader));
197                 do_notify_parent(leader, leader->exit_signal);
198                 /*
199                  * If we were the last child thread and the leader has
200                  * exited already, and the leader's parent ignores SIGCHLD,
201                  * then we are the one who should release the leader.
202                  *
203                  * do_notify_parent() will have marked it self-reaping in
204                  * that case.
205                  */
206                 zap_leader = task_detached(leader);
207         }
208
209         write_unlock_irq(&tasklist_lock);
210         release_thread(p);
211         call_rcu(&p->rcu, delayed_put_task_struct);
212
213         p = leader;
214         if (unlikely(zap_leader))
215                 goto repeat;
216 }
217
218 /*
219  * This checks not only the pgrp, but falls back on the pid if no
220  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
221  * without this...
222  *
223  * The caller must hold rcu lock or the tasklist lock.
224  */
225 struct pid *session_of_pgrp(struct pid *pgrp)
226 {
227         struct task_struct *p;
228         struct pid *sid = NULL;
229
230         p = pid_task(pgrp, PIDTYPE_PGID);
231         if (p == NULL)
232                 p = pid_task(pgrp, PIDTYPE_PID);
233         if (p != NULL)
234                 sid = task_session(p);
235
236         return sid;
237 }
238
239 /*
240  * Determine if a process group is "orphaned", according to the POSIX
241  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
242  * by terminal-generated stop signals.  Newly orphaned process groups are
243  * to receive a SIGHUP and a SIGCONT.
244  *
245  * "I ask you, have you ever known what it is to be an orphan?"
246  */
247 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
248 {
249         struct task_struct *p;
250
251         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
252                 if ((p == ignored_task) ||
253                     (p->exit_state && thread_group_empty(p)) ||
254                     is_global_init(p->real_parent))
255                         continue;
256
257                 if (task_pgrp(p->real_parent) != pgrp &&
258                     task_session(p->real_parent) == task_session(p))
259                         return 0;
260         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
261
262         return 1;
263 }
264
265 int is_current_pgrp_orphaned(void)
266 {
267         int retval;
268
269         read_lock(&tasklist_lock);
270         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
271         read_unlock(&tasklist_lock);
272
273         return retval;
274 }
275
276 static int has_stopped_jobs(struct pid *pgrp)
277 {
278         int retval = 0;
279         struct task_struct *p;
280
281         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
282                 if (!task_is_stopped(p))
283                         continue;
284                 retval = 1;
285                 break;
286         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
287         return retval;
288 }
289
290 /*
291  * Check to see if any process groups have become orphaned as
292  * a result of our exiting, and if they have any stopped jobs,
293  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
294  */
295 static void
296 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
297 {
298         struct pid *pgrp = task_pgrp(tsk);
299         struct task_struct *ignored_task = tsk;
300
301         if (!parent)
302                  /* exit: our father is in a different pgrp than
303                   * we are and we were the only connection outside.
304                   */
305                 parent = tsk->real_parent;
306         else
307                 /* reparent: our child is in a different pgrp than
308                  * we are, and it was the only connection outside.
309                  */
310                 ignored_task = NULL;
311
312         if (task_pgrp(parent) != pgrp &&
313             task_session(parent) == task_session(tsk) &&
314             will_become_orphaned_pgrp(pgrp, ignored_task) &&
315             has_stopped_jobs(pgrp)) {
316                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
317                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
318         }
319 }
320
321 /**
322  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
323  *
324  * If a kernel thread is launched as a result of a system call, or if
325  * it ever exits, it should generally reparent itself to kthreadd so it
326  * isn't in the way of other processes and is correctly cleaned up on exit.
327  *
328  * The various task state such as scheduling policy and priority may have
329  * been inherited from a user process, so we reset them to sane values here.
330  *
331  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
332  */
333 static void reparent_to_kthreadd(void)
334 {
335         write_lock_irq(&tasklist_lock);
336
337         ptrace_unlink(current);
338         /* Reparent to init */
339         current->real_parent = current->parent = kthreadd_task;
340         list_move_tail(&current->sibling, &current->real_parent->children);
341
342         /* Set the exit signal to SIGCHLD so we signal init on exit */
343         current->exit_signal = SIGCHLD;
344
345         if (task_nice(current) < 0)
346                 set_user_nice(current, 0);
347         /* cpus_allowed? */
348         /* rt_priority? */
349         /* signals? */
350         security_task_reparent_to_init(current);
351         memcpy(current->signal->rlim, init_task.signal->rlim,
352                sizeof(current->signal->rlim));
353         atomic_inc(&(INIT_USER->__count));
354         write_unlock_irq(&tasklist_lock);
355         switch_uid(INIT_USER);
356 }
357
358 void __set_special_pids(struct pid *pid)
359 {
360         struct task_struct *curr = current->group_leader;
361         pid_t nr = pid_nr(pid);
362
363         if (task_session(curr) != pid) {
364                 change_pid(curr, PIDTYPE_SID, pid);
365                 set_task_session(curr, nr);
366         }
367         if (task_pgrp(curr) != pid) {
368                 change_pid(curr, PIDTYPE_PGID, pid);
369                 set_task_pgrp(curr, nr);
370         }
371 }
372
373 static void set_special_pids(struct pid *pid)
374 {
375         write_lock_irq(&tasklist_lock);
376         __set_special_pids(pid);
377         write_unlock_irq(&tasklist_lock);
378 }
379
380 /*
381  * Let kernel threads use this to say that they
382  * allow a certain signal (since daemonize() will
383  * have disabled all of them by default).
384  */
385 int allow_signal(int sig)
386 {
387         if (!valid_signal(sig) || sig < 1)
388                 return -EINVAL;
389
390         spin_lock_irq(&current->sighand->siglock);
391         sigdelset(&current->blocked, sig);
392         if (!current->mm) {
393                 /* Kernel threads handle their own signals.
394                    Let the signal code know it'll be handled, so
395                    that they don't get converted to SIGKILL or
396                    just silently dropped */
397                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
398         }
399         recalc_sigpending();
400         spin_unlock_irq(&current->sighand->siglock);
401         return 0;
402 }
403
404 EXPORT_SYMBOL(allow_signal);
405
406 int disallow_signal(int sig)
407 {
408         if (!valid_signal(sig) || sig < 1)
409                 return -EINVAL;
410
411         spin_lock_irq(&current->sighand->siglock);
412         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
413         recalc_sigpending();
414         spin_unlock_irq(&current->sighand->siglock);
415         return 0;
416 }
417
418 EXPORT_SYMBOL(disallow_signal);
419
420 /*
421  *      Put all the gunge required to become a kernel thread without
422  *      attached user resources in one place where it belongs.
423  */
424
425 void daemonize(const char *name, ...)
426 {
427         va_list args;
428         struct fs_struct *fs;
429         sigset_t blocked;
430
431         va_start(args, name);
432         vsnprintf(current->comm, sizeof(current->comm), name, args);
433         va_end(args);
434
435         /*
436          * If we were started as result of loading a module, close all of the
437          * user space pages.  We don't need them, and if we didn't close them
438          * they would be locked into memory.
439          */
440         exit_mm(current);
441         /*
442          * We don't want to have TIF_FREEZE set if the system-wide hibernation
443          * or suspend transition begins right now.
444          */
445         current->flags |= (PF_NOFREEZE | PF_KTHREAD);
446
447         if (current->nsproxy != &init_nsproxy) {
448                 get_nsproxy(&init_nsproxy);
449                 switch_task_namespaces(current, &init_nsproxy);
450         }
451         set_special_pids(&init_struct_pid);
452         proc_clear_tty(current);
453
454         /* Block and flush all signals */
455         sigfillset(&blocked);
456         sigprocmask(SIG_BLOCK, &blocked, NULL);
457         flush_signals(current);
458
459         /* Become as one with the init task */
460
461         exit_fs(current);       /* current->fs->count--; */
462         fs = init_task.fs;
463         current->fs = fs;
464         atomic_inc(&fs->count);
465
466         exit_files(current);
467         current->files = init_task.files;
468         atomic_inc(&current->files->count);
469
470         reparent_to_kthreadd();
471 }
472
473 EXPORT_SYMBOL(daemonize);
474
475 static void close_files(struct files_struct * files)
476 {
477         int i, j;
478         struct fdtable *fdt;
479
480         j = 0;
481
482         /*
483          * It is safe to dereference the fd table without RCU or
484          * ->file_lock because this is the last reference to the
485          * files structure.
486          */
487         fdt = files_fdtable(files);
488         for (;;) {
489                 unsigned long set;
490                 i = j * __NFDBITS;
491                 if (i >= fdt->max_fds)
492                         break;
493                 set = fdt->open_fds->fds_bits[j++];
494                 while (set) {
495                         if (set & 1) {
496                                 struct file * file = xchg(&fdt->fd[i], NULL);
497                                 if (file) {
498                                         filp_close(file, files);
499                                         cond_resched();
500                                 }
501                         }
502                         i++;
503                         set >>= 1;
504                 }
505         }
506 }
507
508 struct files_struct *get_files_struct(struct task_struct *task)
509 {
510         struct files_struct *files;
511
512         task_lock(task);
513         files = task->files;
514         if (files)
515                 atomic_inc(&files->count);
516         task_unlock(task);
517
518         return files;
519 }
520
521 void put_files_struct(struct files_struct *files)
522 {
523         struct fdtable *fdt;
524
525         if (atomic_dec_and_test(&files->count)) {
526                 close_files(files);
527                 /*
528                  * Free the fd and fdset arrays if we expanded them.
529                  * If the fdtable was embedded, pass files for freeing
530                  * at the end of the RCU grace period. Otherwise,
531                  * you can free files immediately.
532                  */
533                 fdt = files_fdtable(files);
534                 if (fdt != &files->fdtab)
535                         kmem_cache_free(files_cachep, files);
536                 free_fdtable(fdt);
537         }
538 }
539
540 void reset_files_struct(struct files_struct *files)
541 {
542         struct task_struct *tsk = current;
543         struct files_struct *old;
544
545         old = tsk->files;
546         task_lock(tsk);
547         tsk->files = files;
548         task_unlock(tsk);
549         put_files_struct(old);
550 }
551
552 void exit_files(struct task_struct *tsk)
553 {
554         struct files_struct * files = tsk->files;
555
556         if (files) {
557                 task_lock(tsk);
558                 tsk->files = NULL;
559                 task_unlock(tsk);
560                 put_files_struct(files);
561         }
562 }
563
564 void put_fs_struct(struct fs_struct *fs)
565 {
566         /* No need to hold fs->lock if we are killing it */
567         if (atomic_dec_and_test(&fs->count)) {
568                 path_put(&fs->root);
569                 path_put(&fs->pwd);
570                 if (fs->altroot.dentry)
571                         path_put(&fs->altroot);
572                 kmem_cache_free(fs_cachep, fs);
573         }
574 }
575
576 void exit_fs(struct task_struct *tsk)
577 {
578         struct fs_struct * fs = tsk->fs;
579
580         if (fs) {
581                 task_lock(tsk);
582                 tsk->fs = NULL;
583                 task_unlock(tsk);
584                 put_fs_struct(fs);
585         }
586 }
587
588 EXPORT_SYMBOL_GPL(exit_fs);
589
590 #ifdef CONFIG_MM_OWNER
591 /*
592  * Task p is exiting and it owned mm, lets find a new owner for it
593  */
594 static inline int
595 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
596 {
597         /*
598          * If there are other users of the mm and the owner (us) is exiting
599          * we need to find a new owner to take on the responsibility.
600          */
601         if (!mm)
602                 return 0;
603         if (atomic_read(&mm->mm_users) <= 1)
604                 return 0;
605         if (mm->owner != p)
606                 return 0;
607         return 1;
608 }
609
610 void mm_update_next_owner(struct mm_struct *mm)
611 {
612         struct task_struct *c, *g, *p = current;
613
614 retry:
615         if (!mm_need_new_owner(mm, p))
616                 return;
617
618         read_lock(&tasklist_lock);
619         /*
620          * Search in the children
621          */
622         list_for_each_entry(c, &p->children, sibling) {
623                 if (c->mm == mm)
624                         goto assign_new_owner;
625         }
626
627         /*
628          * Search in the siblings
629          */
630         list_for_each_entry(c, &p->parent->children, sibling) {
631                 if (c->mm == mm)
632                         goto assign_new_owner;
633         }
634
635         /*
636          * Search through everything else. We should not get
637          * here often
638          */
639         do_each_thread(g, c) {
640                 if (c->mm == mm)
641                         goto assign_new_owner;
642         } while_each_thread(g, c);
643
644         read_unlock(&tasklist_lock);
645         return;
646
647 assign_new_owner:
648         BUG_ON(c == p);
649         get_task_struct(c);
650         /*
651          * The task_lock protects c->mm from changing.
652          * We always want mm->owner->mm == mm
653          */
654         task_lock(c);
655         /*
656          * Delay read_unlock() till we have the task_lock()
657          * to ensure that c does not slip away underneath us
658          */
659         read_unlock(&tasklist_lock);
660         if (c->mm != mm) {
661                 task_unlock(c);
662                 put_task_struct(c);
663                 goto retry;
664         }
665         cgroup_mm_owner_callbacks(mm->owner, c);
666         mm->owner = c;
667         task_unlock(c);
668         put_task_struct(c);
669 }
670 #endif /* CONFIG_MM_OWNER */
671
672 /*
673  * Turn us into a lazy TLB process if we
674  * aren't already..
675  */
676 static void exit_mm(struct task_struct * tsk)
677 {
678         struct mm_struct *mm = tsk->mm;
679         struct core_state *core_state;
680
681         mm_release(tsk, mm);
682         if (!mm)
683                 return;
684         /*
685          * Serialize with any possible pending coredump.
686          * We must hold mmap_sem around checking core_state
687          * and clearing tsk->mm.  The core-inducing thread
688          * will increment ->nr_threads for each thread in the
689          * group with ->mm != NULL.
690          */
691         down_read(&mm->mmap_sem);
692         core_state = mm->core_state;
693         if (core_state) {
694                 struct core_thread self;
695                 up_read(&mm->mmap_sem);
696
697                 self.task = tsk;
698                 self.next = xchg(&core_state->dumper.next, &self);
699                 /*
700                  * Implies mb(), the result of xchg() must be visible
701                  * to core_state->dumper.
702                  */
703                 if (atomic_dec_and_test(&core_state->nr_threads))
704                         complete(&core_state->startup);
705
706                 for (;;) {
707                         set_task_state(tsk, TASK_UNINTERRUPTIBLE);
708                         if (!self.task) /* see coredump_finish() */
709                                 break;
710                         schedule();
711                 }
712                 __set_task_state(tsk, TASK_RUNNING);
713                 down_read(&mm->mmap_sem);
714         }
715         atomic_inc(&mm->mm_count);
716         BUG_ON(mm != tsk->active_mm);
717         /* more a memory barrier than a real lock */
718         task_lock(tsk);
719         tsk->mm = NULL;
720         up_read(&mm->mmap_sem);
721         enter_lazy_tlb(mm, current);
722         /* We don't want this task to be frozen prematurely */
723         clear_freeze_flag(tsk);
724         task_unlock(tsk);
725         mm_update_next_owner(mm);
726         mmput(mm);
727 }
728
729 /*
730  * Return nonzero if @parent's children should reap themselves.
731  *
732  * Called with write_lock_irq(&tasklist_lock) held.
733  */
734 static int ignoring_children(struct task_struct *parent)
735 {
736         int ret;
737         struct sighand_struct *psig = parent->sighand;
738         unsigned long flags;
739         spin_lock_irqsave(&psig->siglock, flags);
740         ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
741                (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
742         spin_unlock_irqrestore(&psig->siglock, flags);
743         return ret;
744 }
745
746 /*
747  * Detach all tasks we were using ptrace on.
748  * Any that need to be release_task'd are put on the @dead list.
749  *
750  * Called with write_lock(&tasklist_lock) held.
751  */
752 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
753 {
754         struct task_struct *p, *n;
755         int ign = -1;
756
757         list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
758                 __ptrace_unlink(p);
759
760                 if (p->exit_state != EXIT_ZOMBIE)
761                         continue;
762
763                 /*
764                  * If it's a zombie, our attachedness prevented normal
765                  * parent notification or self-reaping.  Do notification
766                  * now if it would have happened earlier.  If it should
767                  * reap itself, add it to the @dead list.  We can't call
768                  * release_task() here because we already hold tasklist_lock.
769                  *
770                  * If it's our own child, there is no notification to do.
771                  * But if our normal children self-reap, then this child
772                  * was prevented by ptrace and we must reap it now.
773                  */
774                 if (!task_detached(p) && thread_group_empty(p)) {
775                         if (!same_thread_group(p->real_parent, parent))
776                                 do_notify_parent(p, p->exit_signal);
777                         else {
778                                 if (ign < 0)
779                                         ign = ignoring_children(parent);
780                                 if (ign)
781                                         p->exit_signal = -1;
782                         }
783                 }
784
785                 if (task_detached(p)) {
786                         /*
787                          * Mark it as in the process of being reaped.
788                          */
789                         p->exit_state = EXIT_DEAD;
790                         list_add(&p->ptrace_entry, dead);
791                 }
792         }
793 }
794
795 /*
796  * Finish up exit-time ptrace cleanup.
797  *
798  * Called without locks.
799  */
800 static void ptrace_exit_finish(struct task_struct *parent,
801                                struct list_head *dead)
802 {
803         struct task_struct *p, *n;
804
805         BUG_ON(!list_empty(&parent->ptraced));
806
807         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
808                 list_del_init(&p->ptrace_entry);
809                 release_task(p);
810         }
811 }
812
813 static void reparent_thread(struct task_struct *p, struct task_struct *father)
814 {
815         if (p->pdeath_signal)
816                 /* We already hold the tasklist_lock here.  */
817                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
818
819         list_move_tail(&p->sibling, &p->real_parent->children);
820
821         /* If this is a threaded reparent there is no need to
822          * notify anyone anything has happened.
823          */
824         if (same_thread_group(p->real_parent, father))
825                 return;
826
827         /* We don't want people slaying init.  */
828         if (!task_detached(p))
829                 p->exit_signal = SIGCHLD;
830
831         /* If we'd notified the old parent about this child's death,
832          * also notify the new parent.
833          */
834         if (!ptrace_reparented(p) &&
835             p->exit_state == EXIT_ZOMBIE &&
836             !task_detached(p) && thread_group_empty(p))
837                 do_notify_parent(p, p->exit_signal);
838
839         kill_orphaned_pgrp(p, father);
840 }
841
842 /*
843  * When we die, we re-parent all our children.
844  * Try to give them to another thread in our thread
845  * group, and if no such member exists, give it to
846  * the child reaper process (ie "init") in our pid
847  * space.
848  */
849 static void forget_original_parent(struct task_struct *father)
850 {
851         struct task_struct *p, *n, *reaper = father;
852         LIST_HEAD(ptrace_dead);
853
854         write_lock_irq(&tasklist_lock);
855
856         /*
857          * First clean up ptrace if we were using it.
858          */
859         ptrace_exit(father, &ptrace_dead);
860
861         do {
862                 reaper = next_thread(reaper);
863                 if (reaper == father) {
864                         reaper = task_child_reaper(father);
865                         break;
866                 }
867         } while (reaper->flags & PF_EXITING);
868
869         list_for_each_entry_safe(p, n, &father->children, sibling) {
870                 p->real_parent = reaper;
871                 if (p->parent == father) {
872                         BUG_ON(p->ptrace);
873                         p->parent = p->real_parent;
874                 }
875                 reparent_thread(p, father);
876         }
877
878         write_unlock_irq(&tasklist_lock);
879         BUG_ON(!list_empty(&father->children));
880
881         ptrace_exit_finish(father, &ptrace_dead);
882 }
883
884 /*
885  * Send signals to all our closest relatives so that they know
886  * to properly mourn us..
887  */
888 static void exit_notify(struct task_struct *tsk, int group_dead)
889 {
890         int state;
891
892         /*
893          * This does two things:
894          *
895          * A.  Make init inherit all the child processes
896          * B.  Check to see if any process groups have become orphaned
897          *      as a result of our exiting, and if they have any stopped
898          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
899          */
900         forget_original_parent(tsk);
901         exit_task_namespaces(tsk);
902
903         write_lock_irq(&tasklist_lock);
904         if (group_dead)
905                 kill_orphaned_pgrp(tsk->group_leader, NULL);
906
907         /* Let father know we died
908          *
909          * Thread signals are configurable, but you aren't going to use
910          * that to send signals to arbitary processes.
911          * That stops right now.
912          *
913          * If the parent exec id doesn't match the exec id we saved
914          * when we started then we know the parent has changed security
915          * domain.
916          *
917          * If our self_exec id doesn't match our parent_exec_id then
918          * we have changed execution domain as these two values started
919          * the same after a fork.
920          */
921         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
922             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
923              tsk->self_exec_id != tsk->parent_exec_id) &&
924             !capable(CAP_KILL))
925                 tsk->exit_signal = SIGCHLD;
926
927         /* If something other than our normal parent is ptracing us, then
928          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
929          * only has special meaning to our real parent.
930          */
931         if (!task_detached(tsk) && thread_group_empty(tsk)) {
932                 int signal = ptrace_reparented(tsk) ?
933                                 SIGCHLD : tsk->exit_signal;
934                 do_notify_parent(tsk, signal);
935         } else if (tsk->ptrace) {
936                 do_notify_parent(tsk, SIGCHLD);
937         }
938
939         state = EXIT_ZOMBIE;
940         if (task_detached(tsk) && likely(!tsk->ptrace))
941                 state = EXIT_DEAD;
942         tsk->exit_state = state;
943
944         /* mt-exec, de_thread() is waiting for us */
945         if (thread_group_leader(tsk) &&
946             tsk->signal->notify_count < 0 &&
947             tsk->signal->group_exit_task)
948                 wake_up_process(tsk->signal->group_exit_task);
949
950         write_unlock_irq(&tasklist_lock);
951
952         /* If the process is dead, release it - nobody will wait for it */
953         if (state == EXIT_DEAD)
954                 release_task(tsk);
955 }
956
957 #ifdef CONFIG_DEBUG_STACK_USAGE
958 static void check_stack_usage(void)
959 {
960         static DEFINE_SPINLOCK(low_water_lock);
961         static int lowest_to_date = THREAD_SIZE;
962         unsigned long *n = end_of_stack(current);
963         unsigned long free;
964
965         while (*n == 0)
966                 n++;
967         free = (unsigned long)n - (unsigned long)end_of_stack(current);
968
969         if (free >= lowest_to_date)
970                 return;
971
972         spin_lock(&low_water_lock);
973         if (free < lowest_to_date) {
974                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
975                                 "left\n",
976                                 current->comm, free);
977                 lowest_to_date = free;
978         }
979         spin_unlock(&low_water_lock);
980 }
981 #else
982 static inline void check_stack_usage(void) {}
983 #endif
984
985 static inline void exit_child_reaper(struct task_struct *tsk)
986 {
987         if (likely(tsk->group_leader != task_child_reaper(tsk)))
988                 return;
989
990         if (tsk->nsproxy->pid_ns == &init_pid_ns)
991                 panic("Attempted to kill init!");
992
993         /*
994          * @tsk is the last thread in the 'cgroup-init' and is exiting.
995          * Terminate all remaining processes in the namespace and reap them
996          * before exiting @tsk.
997          *
998          * Note that @tsk (last thread of cgroup-init) may not necessarily
999          * be the child-reaper (i.e main thread of cgroup-init) of the
1000          * namespace i.e the child_reaper may have already exited.
1001          *
1002          * Even after a child_reaper exits, we let it inherit orphaned children,
1003          * because, pid_ns->child_reaper remains valid as long as there is
1004          * at least one living sub-thread in the cgroup init.
1005
1006          * This living sub-thread of the cgroup-init will be notified when
1007          * a child inherited by the 'child-reaper' exits (do_notify_parent()
1008          * uses __group_send_sig_info()). Further, when reaping child processes,
1009          * do_wait() iterates over children of all living sub threads.
1010
1011          * i.e even though 'child_reaper' thread is listed as the parent of the
1012          * orphaned children, any living sub-thread in the cgroup-init can
1013          * perform the role of the child_reaper.
1014          */
1015         zap_pid_ns_processes(tsk->nsproxy->pid_ns);
1016 }
1017
1018 NORET_TYPE void do_exit(long code)
1019 {
1020         struct task_struct *tsk = current;
1021         int group_dead;
1022
1023         profile_task_exit(tsk);
1024
1025         WARN_ON(atomic_read(&tsk->fs_excl));
1026
1027         if (unlikely(in_interrupt()))
1028                 panic("Aiee, killing interrupt handler!");
1029         if (unlikely(!tsk->pid))
1030                 panic("Attempted to kill the idle task!");
1031
1032         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
1033                 current->ptrace_message = code;
1034                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
1035         }
1036
1037         /*
1038          * We're taking recursive faults here in do_exit. Safest is to just
1039          * leave this task alone and wait for reboot.
1040          */
1041         if (unlikely(tsk->flags & PF_EXITING)) {
1042                 printk(KERN_ALERT
1043                         "Fixing recursive fault but reboot is needed!\n");
1044                 /*
1045                  * We can do this unlocked here. The futex code uses
1046                  * this flag just to verify whether the pi state
1047                  * cleanup has been done or not. In the worst case it
1048                  * loops once more. We pretend that the cleanup was
1049                  * done as there is no way to return. Either the
1050                  * OWNER_DIED bit is set by now or we push the blocked
1051                  * task into the wait for ever nirwana as well.
1052                  */
1053                 tsk->flags |= PF_EXITPIDONE;
1054                 if (tsk->io_context)
1055                         exit_io_context();
1056                 set_current_state(TASK_UNINTERRUPTIBLE);
1057                 schedule();
1058         }
1059
1060         exit_signals(tsk);  /* sets PF_EXITING */
1061         /*
1062          * tsk->flags are checked in the futex code to protect against
1063          * an exiting task cleaning up the robust pi futexes.
1064          */
1065         smp_mb();
1066         spin_unlock_wait(&tsk->pi_lock);
1067
1068         if (unlikely(in_atomic()))
1069                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1070                                 current->comm, task_pid_nr(current),
1071                                 preempt_count());
1072
1073         acct_update_integrals(tsk);
1074         if (tsk->mm) {
1075                 update_hiwater_rss(tsk->mm);
1076                 update_hiwater_vm(tsk->mm);
1077         }
1078         group_dead = atomic_dec_and_test(&tsk->signal->live);
1079         if (group_dead) {
1080                 exit_child_reaper(tsk);
1081                 hrtimer_cancel(&tsk->signal->real_timer);
1082                 exit_itimers(tsk->signal);
1083         }
1084         acct_collect(code, group_dead);
1085 #ifdef CONFIG_FUTEX
1086         if (unlikely(tsk->robust_list))
1087                 exit_robust_list(tsk);
1088 #ifdef CONFIG_COMPAT
1089         if (unlikely(tsk->compat_robust_list))
1090                 compat_exit_robust_list(tsk);
1091 #endif
1092 #endif
1093         if (group_dead)
1094                 tty_audit_exit();
1095         if (unlikely(tsk->audit_context))
1096                 audit_free(tsk);
1097
1098         tsk->exit_code = code;
1099         taskstats_exit(tsk, group_dead);
1100
1101         exit_mm(tsk);
1102
1103         if (group_dead)
1104                 acct_process();
1105         exit_sem(tsk);
1106         exit_files(tsk);
1107         exit_fs(tsk);
1108         check_stack_usage();
1109         exit_thread();
1110         cgroup_exit(tsk, 1);
1111         exit_keys(tsk);
1112
1113         if (group_dead && tsk->signal->leader)
1114                 disassociate_ctty(1);
1115
1116         module_put(task_thread_info(tsk)->exec_domain->module);
1117         if (tsk->binfmt)
1118                 module_put(tsk->binfmt->module);
1119
1120         proc_exit_connector(tsk);
1121         exit_notify(tsk, group_dead);
1122 #ifdef CONFIG_NUMA
1123         mpol_put(tsk->mempolicy);
1124         tsk->mempolicy = NULL;
1125 #endif
1126 #ifdef CONFIG_FUTEX
1127         /*
1128          * This must happen late, after the PID is not
1129          * hashed anymore:
1130          */
1131         if (unlikely(!list_empty(&tsk->pi_state_list)))
1132                 exit_pi_state_list(tsk);
1133         if (unlikely(current->pi_state_cache))
1134                 kfree(current->pi_state_cache);
1135 #endif
1136         /*
1137          * Make sure we are holding no locks:
1138          */
1139         debug_check_no_locks_held(tsk);
1140         /*
1141          * We can do this unlocked here. The futex code uses this flag
1142          * just to verify whether the pi state cleanup has been done
1143          * or not. In the worst case it loops once more.
1144          */
1145         tsk->flags |= PF_EXITPIDONE;
1146
1147         if (tsk->io_context)
1148                 exit_io_context();
1149
1150         if (tsk->splice_pipe)
1151                 __free_pipe_info(tsk->splice_pipe);
1152
1153         preempt_disable();
1154         /* causes final put_task_struct in finish_task_switch(). */
1155         tsk->state = TASK_DEAD;
1156
1157         schedule();
1158         BUG();
1159         /* Avoid "noreturn function does return".  */
1160         for (;;)
1161                 cpu_relax();    /* For when BUG is null */
1162 }
1163
1164 EXPORT_SYMBOL_GPL(do_exit);
1165
1166 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1167 {
1168         if (comp)
1169                 complete(comp);
1170
1171         do_exit(code);
1172 }
1173
1174 EXPORT_SYMBOL(complete_and_exit);
1175
1176 asmlinkage long sys_exit(int error_code)
1177 {
1178         do_exit((error_code&0xff)<<8);
1179 }
1180
1181 /*
1182  * Take down every thread in the group.  This is called by fatal signals
1183  * as well as by sys_exit_group (below).
1184  */
1185 NORET_TYPE void
1186 do_group_exit(int exit_code)
1187 {
1188         struct signal_struct *sig = current->signal;
1189
1190         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1191
1192         if (signal_group_exit(sig))
1193                 exit_code = sig->group_exit_code;
1194         else if (!thread_group_empty(current)) {
1195                 struct sighand_struct *const sighand = current->sighand;
1196                 spin_lock_irq(&sighand->siglock);
1197                 if (signal_group_exit(sig))
1198                         /* Another thread got here before we took the lock.  */
1199                         exit_code = sig->group_exit_code;
1200                 else {
1201                         sig->group_exit_code = exit_code;
1202                         sig->flags = SIGNAL_GROUP_EXIT;
1203                         zap_other_threads(current);
1204                 }
1205                 spin_unlock_irq(&sighand->siglock);
1206         }
1207
1208         do_exit(exit_code);
1209         /* NOTREACHED */
1210 }
1211
1212 /*
1213  * this kills every thread in the thread group. Note that any externally
1214  * wait4()-ing process will get the correct exit code - even if this
1215  * thread is not the thread group leader.
1216  */
1217 asmlinkage void sys_exit_group(int error_code)
1218 {
1219         do_group_exit((error_code & 0xff) << 8);
1220 }
1221
1222 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1223 {
1224         struct pid *pid = NULL;
1225         if (type == PIDTYPE_PID)
1226                 pid = task->pids[type].pid;
1227         else if (type < PIDTYPE_MAX)
1228                 pid = task->group_leader->pids[type].pid;
1229         return pid;
1230 }
1231
1232 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1233                           struct task_struct *p)
1234 {
1235         int err;
1236
1237         if (type < PIDTYPE_MAX) {
1238                 if (task_pid_type(p, type) != pid)
1239                         return 0;
1240         }
1241
1242         /* Wait for all children (clone and not) if __WALL is set;
1243          * otherwise, wait for clone children *only* if __WCLONE is
1244          * set; otherwise, wait for non-clone children *only*.  (Note:
1245          * A "clone" child here is one that reports to its parent
1246          * using a signal other than SIGCHLD.) */
1247         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1248             && !(options & __WALL))
1249                 return 0;
1250
1251         err = security_task_wait(p);
1252         if (err)
1253                 return err;
1254
1255         return 1;
1256 }
1257
1258 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1259                                int why, int status,
1260                                struct siginfo __user *infop,
1261                                struct rusage __user *rusagep)
1262 {
1263         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1264
1265         put_task_struct(p);
1266         if (!retval)
1267                 retval = put_user(SIGCHLD, &infop->si_signo);
1268         if (!retval)
1269                 retval = put_user(0, &infop->si_errno);
1270         if (!retval)
1271                 retval = put_user((short)why, &infop->si_code);
1272         if (!retval)
1273                 retval = put_user(pid, &infop->si_pid);
1274         if (!retval)
1275                 retval = put_user(uid, &infop->si_uid);
1276         if (!retval)
1277                 retval = put_user(status, &infop->si_status);
1278         if (!retval)
1279                 retval = pid;
1280         return retval;
1281 }
1282
1283 /*
1284  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1285  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1286  * the lock and this task is uninteresting.  If we return nonzero, we have
1287  * released the lock and the system call should return.
1288  */
1289 static int wait_task_zombie(struct task_struct *p, int options,
1290                             struct siginfo __user *infop,
1291                             int __user *stat_addr, struct rusage __user *ru)
1292 {
1293         unsigned long state;
1294         int retval, status, traced;
1295         pid_t pid = task_pid_vnr(p);
1296
1297         if (!likely(options & WEXITED))
1298                 return 0;
1299
1300         if (unlikely(options & WNOWAIT)) {
1301                 uid_t uid = p->uid;
1302                 int exit_code = p->exit_code;
1303                 int why, status;
1304
1305                 get_task_struct(p);
1306                 read_unlock(&tasklist_lock);
1307                 if ((exit_code & 0x7f) == 0) {
1308                         why = CLD_EXITED;
1309                         status = exit_code >> 8;
1310                 } else {
1311                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1312                         status = exit_code & 0x7f;
1313                 }
1314                 return wait_noreap_copyout(p, pid, uid, why,
1315                                            status, infop, ru);
1316         }
1317
1318         /*
1319          * Try to move the task's state to DEAD
1320          * only one thread is allowed to do this:
1321          */
1322         state = xchg(&p->exit_state, EXIT_DEAD);
1323         if (state != EXIT_ZOMBIE) {
1324                 BUG_ON(state != EXIT_DEAD);
1325                 return 0;
1326         }
1327
1328         traced = ptrace_reparented(p);
1329
1330         if (likely(!traced)) {
1331                 struct signal_struct *psig;
1332                 struct signal_struct *sig;
1333
1334                 /*
1335                  * The resource counters for the group leader are in its
1336                  * own task_struct.  Those for dead threads in the group
1337                  * are in its signal_struct, as are those for the child
1338                  * processes it has previously reaped.  All these
1339                  * accumulate in the parent's signal_struct c* fields.
1340                  *
1341                  * We don't bother to take a lock here to protect these
1342                  * p->signal fields, because they are only touched by
1343                  * __exit_signal, which runs with tasklist_lock
1344                  * write-locked anyway, and so is excluded here.  We do
1345                  * need to protect the access to p->parent->signal fields,
1346                  * as other threads in the parent group can be right
1347                  * here reaping other children at the same time.
1348                  */
1349                 spin_lock_irq(&p->parent->sighand->siglock);
1350                 psig = p->parent->signal;
1351                 sig = p->signal;
1352                 psig->cutime =
1353                         cputime_add(psig->cutime,
1354                         cputime_add(p->utime,
1355                         cputime_add(sig->utime,
1356                                     sig->cutime)));
1357                 psig->cstime =
1358                         cputime_add(psig->cstime,
1359                         cputime_add(p->stime,
1360                         cputime_add(sig->stime,
1361                                     sig->cstime)));
1362                 psig->cgtime =
1363                         cputime_add(psig->cgtime,
1364                         cputime_add(p->gtime,
1365                         cputime_add(sig->gtime,
1366                                     sig->cgtime)));
1367                 psig->cmin_flt +=
1368                         p->min_flt + sig->min_flt + sig->cmin_flt;
1369                 psig->cmaj_flt +=
1370                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1371                 psig->cnvcsw +=
1372                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1373                 psig->cnivcsw +=
1374                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1375                 psig->cinblock +=
1376                         task_io_get_inblock(p) +
1377                         sig->inblock + sig->cinblock;
1378                 psig->coublock +=
1379                         task_io_get_oublock(p) +
1380                         sig->oublock + sig->coublock;
1381 #ifdef CONFIG_TASK_XACCT
1382                 psig->rchar += p->rchar + sig->rchar;
1383                 psig->wchar += p->wchar + sig->wchar;
1384                 psig->syscr += p->syscr + sig->syscr;
1385                 psig->syscw += p->syscw + sig->syscw;
1386 #endif /* CONFIG_TASK_XACCT */
1387 #ifdef CONFIG_TASK_IO_ACCOUNTING
1388                 psig->ioac.read_bytes +=
1389                         p->ioac.read_bytes + sig->ioac.read_bytes;
1390                 psig->ioac.write_bytes +=
1391                         p->ioac.write_bytes + sig->ioac.write_bytes;
1392                 psig->ioac.cancelled_write_bytes +=
1393                                 p->ioac.cancelled_write_bytes +
1394                                 sig->ioac.cancelled_write_bytes;
1395 #endif /* CONFIG_TASK_IO_ACCOUNTING */
1396                 spin_unlock_irq(&p->parent->sighand->siglock);
1397         }
1398
1399         /*
1400          * Now we are sure this task is interesting, and no other
1401          * thread can reap it because we set its state to EXIT_DEAD.
1402          */
1403         read_unlock(&tasklist_lock);
1404
1405         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1406         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1407                 ? p->signal->group_exit_code : p->exit_code;
1408         if (!retval && stat_addr)
1409                 retval = put_user(status, stat_addr);
1410         if (!retval && infop)
1411                 retval = put_user(SIGCHLD, &infop->si_signo);
1412         if (!retval && infop)
1413                 retval = put_user(0, &infop->si_errno);
1414         if (!retval && infop) {
1415                 int why;
1416
1417                 if ((status & 0x7f) == 0) {
1418                         why = CLD_EXITED;
1419                         status >>= 8;
1420                 } else {
1421                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1422                         status &= 0x7f;
1423                 }
1424                 retval = put_user((short)why, &infop->si_code);
1425                 if (!retval)
1426                         retval = put_user(status, &infop->si_status);
1427         }
1428         if (!retval && infop)
1429                 retval = put_user(pid, &infop->si_pid);
1430         if (!retval && infop)
1431                 retval = put_user(p->uid, &infop->si_uid);
1432         if (!retval)
1433                 retval = pid;
1434
1435         if (traced) {
1436                 write_lock_irq(&tasklist_lock);
1437                 /* We dropped tasklist, ptracer could die and untrace */
1438                 ptrace_unlink(p);
1439                 /*
1440                  * If this is not a detached task, notify the parent.
1441                  * If it's still not detached after that, don't release
1442                  * it now.
1443                  */
1444                 if (!task_detached(p)) {
1445                         do_notify_parent(p, p->exit_signal);
1446                         if (!task_detached(p)) {
1447                                 p->exit_state = EXIT_ZOMBIE;
1448                                 p = NULL;
1449                         }
1450                 }
1451                 write_unlock_irq(&tasklist_lock);
1452         }
1453         if (p != NULL)
1454                 release_task(p);
1455
1456         return retval;
1457 }
1458
1459 /*
1460  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1461  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1462  * the lock and this task is uninteresting.  If we return nonzero, we have
1463  * released the lock and the system call should return.
1464  */
1465 static int wait_task_stopped(int ptrace, struct task_struct *p,
1466                              int options, struct siginfo __user *infop,
1467                              int __user *stat_addr, struct rusage __user *ru)
1468 {
1469         int retval, exit_code, why;
1470         uid_t uid = 0; /* unneeded, required by compiler */
1471         pid_t pid;
1472
1473         if (!(options & WUNTRACED))
1474                 return 0;
1475
1476         exit_code = 0;
1477         spin_lock_irq(&p->sighand->siglock);
1478
1479         if (unlikely(!task_is_stopped_or_traced(p)))
1480                 goto unlock_sig;
1481
1482         if (!ptrace && p->signal->group_stop_count > 0)
1483                 /*
1484                  * A group stop is in progress and this is the group leader.
1485                  * We won't report until all threads have stopped.
1486                  */
1487                 goto unlock_sig;
1488
1489         exit_code = p->exit_code;
1490         if (!exit_code)
1491                 goto unlock_sig;
1492
1493         if (!unlikely(options & WNOWAIT))
1494                 p->exit_code = 0;
1495
1496         uid = p->uid;
1497 unlock_sig:
1498         spin_unlock_irq(&p->sighand->siglock);
1499         if (!exit_code)
1500                 return 0;
1501
1502         /*
1503          * Now we are pretty sure this task is interesting.
1504          * Make sure it doesn't get reaped out from under us while we
1505          * give up the lock and then examine it below.  We don't want to
1506          * keep holding onto the tasklist_lock while we call getrusage and
1507          * possibly take page faults for user memory.
1508          */
1509         get_task_struct(p);
1510         pid = task_pid_vnr(p);
1511         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1512         read_unlock(&tasklist_lock);
1513
1514         if (unlikely(options & WNOWAIT))
1515                 return wait_noreap_copyout(p, pid, uid,
1516                                            why, exit_code,
1517                                            infop, ru);
1518
1519         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1520         if (!retval && stat_addr)
1521                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1522         if (!retval && infop)
1523                 retval = put_user(SIGCHLD, &infop->si_signo);
1524         if (!retval && infop)
1525                 retval = put_user(0, &infop->si_errno);
1526         if (!retval && infop)
1527                 retval = put_user((short)why, &infop->si_code);
1528         if (!retval && infop)
1529                 retval = put_user(exit_code, &infop->si_status);
1530         if (!retval && infop)
1531                 retval = put_user(pid, &infop->si_pid);
1532         if (!retval && infop)
1533                 retval = put_user(uid, &infop->si_uid);
1534         if (!retval)
1535                 retval = pid;
1536         put_task_struct(p);
1537
1538         BUG_ON(!retval);
1539         return retval;
1540 }
1541
1542 /*
1543  * Handle do_wait work for one task in a live, non-stopped state.
1544  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1545  * the lock and this task is uninteresting.  If we return nonzero, we have
1546  * released the lock and the system call should return.
1547  */
1548 static int wait_task_continued(struct task_struct *p, int options,
1549                                struct siginfo __user *infop,
1550                                int __user *stat_addr, struct rusage __user *ru)
1551 {
1552         int retval;
1553         pid_t pid;
1554         uid_t uid;
1555
1556         if (!unlikely(options & WCONTINUED))
1557                 return 0;
1558
1559         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1560                 return 0;
1561
1562         spin_lock_irq(&p->sighand->siglock);
1563         /* Re-check with the lock held.  */
1564         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1565                 spin_unlock_irq(&p->sighand->siglock);
1566                 return 0;
1567         }
1568         if (!unlikely(options & WNOWAIT))
1569                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1570         spin_unlock_irq(&p->sighand->siglock);
1571
1572         pid = task_pid_vnr(p);
1573         uid = p->uid;
1574         get_task_struct(p);
1575         read_unlock(&tasklist_lock);
1576
1577         if (!infop) {
1578                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1579                 put_task_struct(p);
1580                 if (!retval && stat_addr)
1581                         retval = put_user(0xffff, stat_addr);
1582                 if (!retval)
1583                         retval = pid;
1584         } else {
1585                 retval = wait_noreap_copyout(p, pid, uid,
1586                                              CLD_CONTINUED, SIGCONT,
1587                                              infop, ru);
1588                 BUG_ON(retval == 0);
1589         }
1590
1591         return retval;
1592 }
1593
1594 /*
1595  * Consider @p for a wait by @parent.
1596  *
1597  * -ECHILD should be in *@notask_error before the first call.
1598  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1599  * Returns zero if the search for a child should continue;
1600  * then *@notask_error is 0 if @p is an eligible child,
1601  * or another error from security_task_wait(), or still -ECHILD.
1602  */
1603 static int wait_consider_task(struct task_struct *parent, int ptrace,
1604                               struct task_struct *p, int *notask_error,
1605                               enum pid_type type, struct pid *pid, int options,
1606                               struct siginfo __user *infop,
1607                               int __user *stat_addr, struct rusage __user *ru)
1608 {
1609         int ret = eligible_child(type, pid, options, p);
1610         if (!ret)
1611                 return ret;
1612
1613         if (unlikely(ret < 0)) {
1614                 /*
1615                  * If we have not yet seen any eligible child,
1616                  * then let this error code replace -ECHILD.
1617                  * A permission error will give the user a clue
1618                  * to look for security policy problems, rather
1619                  * than for mysterious wait bugs.
1620                  */
1621                 if (*notask_error)
1622                         *notask_error = ret;
1623         }
1624
1625         if (likely(!ptrace) && unlikely(p->ptrace)) {
1626                 /*
1627                  * This child is hidden by ptrace.
1628                  * We aren't allowed to see it now, but eventually we will.
1629                  */
1630                 *notask_error = 0;
1631                 return 0;
1632         }
1633
1634         if (p->exit_state == EXIT_DEAD)
1635                 return 0;
1636
1637         /*
1638          * We don't reap group leaders with subthreads.
1639          */
1640         if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1641                 return wait_task_zombie(p, options, infop, stat_addr, ru);
1642
1643         /*
1644          * It's stopped or running now, so it might
1645          * later continue, exit, or stop again.
1646          */
1647         *notask_error = 0;
1648
1649         if (task_is_stopped_or_traced(p))
1650                 return wait_task_stopped(ptrace, p, options,
1651                                          infop, stat_addr, ru);
1652
1653         return wait_task_continued(p, options, infop, stat_addr, ru);
1654 }
1655
1656 /*
1657  * Do the work of do_wait() for one thread in the group, @tsk.
1658  *
1659  * -ECHILD should be in *@notask_error before the first call.
1660  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1661  * Returns zero if the search for a child should continue; then
1662  * *@notask_error is 0 if there were any eligible children,
1663  * or another error from security_task_wait(), or still -ECHILD.
1664  */
1665 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1666                           enum pid_type type, struct pid *pid, int options,
1667                           struct siginfo __user *infop, int __user *stat_addr,
1668                           struct rusage __user *ru)
1669 {
1670         struct task_struct *p;
1671
1672         list_for_each_entry(p, &tsk->children, sibling) {
1673                 /*
1674                  * Do not consider detached threads.
1675                  */
1676                 if (!task_detached(p)) {
1677                         int ret = wait_consider_task(tsk, 0, p, notask_error,
1678                                                      type, pid, options,
1679                                                      infop, stat_addr, ru);
1680                         if (ret)
1681                                 return ret;
1682                 }
1683         }
1684
1685         return 0;
1686 }
1687
1688 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1689                           enum pid_type type, struct pid *pid, int options,
1690                           struct siginfo __user *infop, int __user *stat_addr,
1691                           struct rusage __user *ru)
1692 {
1693         struct task_struct *p;
1694
1695         /*
1696          * Traditionally we see ptrace'd stopped tasks regardless of options.
1697          */
1698         options |= WUNTRACED;
1699
1700         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1701                 int ret = wait_consider_task(tsk, 1, p, notask_error,
1702                                              type, pid, options,
1703                                              infop, stat_addr, ru);
1704                 if (ret)
1705                         return ret;
1706         }
1707
1708         return 0;
1709 }
1710
1711 static long do_wait(enum pid_type type, struct pid *pid, int options,
1712                     struct siginfo __user *infop, int __user *stat_addr,
1713                     struct rusage __user *ru)
1714 {
1715         DECLARE_WAITQUEUE(wait, current);
1716         struct task_struct *tsk;
1717         int retval;
1718
1719         add_wait_queue(&current->signal->wait_chldexit,&wait);
1720 repeat:
1721         /*
1722          * If there is nothing that can match our critiera just get out.
1723          * We will clear @retval to zero if we see any child that might later
1724          * match our criteria, even if we are not able to reap it yet.
1725          */
1726         retval = -ECHILD;
1727         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1728                 goto end;
1729
1730         current->state = TASK_INTERRUPTIBLE;
1731         read_lock(&tasklist_lock);
1732         tsk = current;
1733         do {
1734                 int tsk_result = do_wait_thread(tsk, &retval,
1735                                                 type, pid, options,
1736                                                 infop, stat_addr, ru);
1737                 if (!tsk_result)
1738                         tsk_result = ptrace_do_wait(tsk, &retval,
1739                                                     type, pid, options,
1740                                                     infop, stat_addr, ru);
1741                 if (tsk_result) {
1742                         /*
1743                          * tasklist_lock is unlocked and we have a final result.
1744                          */
1745                         retval = tsk_result;
1746                         goto end;
1747                 }
1748
1749                 if (options & __WNOTHREAD)
1750                         break;
1751                 tsk = next_thread(tsk);
1752                 BUG_ON(tsk->signal != current->signal);
1753         } while (tsk != current);
1754         read_unlock(&tasklist_lock);
1755
1756         if (!retval && !(options & WNOHANG)) {
1757                 retval = -ERESTARTSYS;
1758                 if (!signal_pending(current)) {
1759                         schedule();
1760                         goto repeat;
1761                 }
1762         }
1763
1764 end:
1765         current->state = TASK_RUNNING;
1766         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1767         if (infop) {
1768                 if (retval > 0)
1769                         retval = 0;
1770                 else {
1771                         /*
1772                          * For a WNOHANG return, clear out all the fields
1773                          * we would set so the user can easily tell the
1774                          * difference.
1775                          */
1776                         if (!retval)
1777                                 retval = put_user(0, &infop->si_signo);
1778                         if (!retval)
1779                                 retval = put_user(0, &infop->si_errno);
1780                         if (!retval)
1781                                 retval = put_user(0, &infop->si_code);
1782                         if (!retval)
1783                                 retval = put_user(0, &infop->si_pid);
1784                         if (!retval)
1785                                 retval = put_user(0, &infop->si_uid);
1786                         if (!retval)
1787                                 retval = put_user(0, &infop->si_status);
1788                 }
1789         }
1790         return retval;
1791 }
1792
1793 asmlinkage long sys_waitid(int which, pid_t upid,
1794                            struct siginfo __user *infop, int options,
1795                            struct rusage __user *ru)
1796 {
1797         struct pid *pid = NULL;
1798         enum pid_type type;
1799         long ret;
1800
1801         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1802                 return -EINVAL;
1803         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1804                 return -EINVAL;
1805
1806         switch (which) {
1807         case P_ALL:
1808                 type = PIDTYPE_MAX;
1809                 break;
1810         case P_PID:
1811                 type = PIDTYPE_PID;
1812                 if (upid <= 0)
1813                         return -EINVAL;
1814                 break;
1815         case P_PGID:
1816                 type = PIDTYPE_PGID;
1817                 if (upid <= 0)
1818                         return -EINVAL;
1819                 break;
1820         default:
1821                 return -EINVAL;
1822         }
1823
1824         if (type < PIDTYPE_MAX)
1825                 pid = find_get_pid(upid);
1826         ret = do_wait(type, pid, options, infop, NULL, ru);
1827         put_pid(pid);
1828
1829         /* avoid REGPARM breakage on x86: */
1830         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1831         return ret;
1832 }
1833
1834 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1835                           int options, struct rusage __user *ru)
1836 {
1837         struct pid *pid = NULL;
1838         enum pid_type type;
1839         long ret;
1840
1841         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1842                         __WNOTHREAD|__WCLONE|__WALL))
1843                 return -EINVAL;
1844
1845         if (upid == -1)
1846                 type = PIDTYPE_MAX;
1847         else if (upid < 0) {
1848                 type = PIDTYPE_PGID;
1849                 pid = find_get_pid(-upid);
1850         } else if (upid == 0) {
1851                 type = PIDTYPE_PGID;
1852                 pid = get_pid(task_pgrp(current));
1853         } else /* upid > 0 */ {
1854                 type = PIDTYPE_PID;
1855                 pid = find_get_pid(upid);
1856         }
1857
1858         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1859         put_pid(pid);
1860
1861         /* avoid REGPARM breakage on x86: */
1862         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1863         return ret;
1864 }
1865
1866 #ifdef __ARCH_WANT_SYS_WAITPID
1867
1868 /*
1869  * sys_waitpid() remains for compatibility. waitpid() should be
1870  * implemented by calling sys_wait4() from libc.a.
1871  */
1872 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1873 {
1874         return sys_wait4(pid, stat_addr, options, NULL);
1875 }
1876
1877 #endif