pid_ns: (BUG 11391) change ->child_reaper when init->group_leader exits
[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/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 #include <linux/tracehook.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->utime = cputime_add(sig->utime, tsk->utime);
116                 sig->stime = cputime_add(sig->stime, tsk->stime);
117                 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
118                 sig->min_flt += tsk->min_flt;
119                 sig->maj_flt += tsk->maj_flt;
120                 sig->nvcsw += tsk->nvcsw;
121                 sig->nivcsw += tsk->nivcsw;
122                 sig->inblock += task_io_get_inblock(tsk);
123                 sig->oublock += task_io_get_oublock(tsk);
124                 task_io_accounting_add(&sig->ioac, &tsk->ioac);
125                 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
126                 sig = NULL; /* Marker for below. */
127         }
128
129         __unhash_process(tsk);
130
131         /*
132          * Do this under ->siglock, we can race with another thread
133          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
134          */
135         flush_sigqueue(&tsk->pending);
136
137         tsk->signal = NULL;
138         tsk->sighand = NULL;
139         spin_unlock(&sighand->siglock);
140
141         __cleanup_sighand(sighand);
142         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
143         if (sig) {
144                 flush_sigqueue(&sig->shared_pending);
145                 taskstats_tgid_free(sig);
146                 __cleanup_signal(sig);
147         }
148 }
149
150 static void delayed_put_task_struct(struct rcu_head *rhp)
151 {
152         put_task_struct(container_of(rhp, struct task_struct, rcu));
153 }
154
155
156 void release_task(struct task_struct * p)
157 {
158         struct task_struct *leader;
159         int zap_leader;
160 repeat:
161         tracehook_prepare_release_task(p);
162         atomic_dec(&p->user->processes);
163         proc_flush_task(p);
164         write_lock_irq(&tasklist_lock);
165         tracehook_finish_release_task(p);
166         __exit_signal(p);
167
168         /*
169          * If we are the last non-leader member of the thread
170          * group, and the leader is zombie, then notify the
171          * group leader's parent process. (if it wants notification.)
172          */
173         zap_leader = 0;
174         leader = p->group_leader;
175         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
176                 BUG_ON(task_detached(leader));
177                 do_notify_parent(leader, leader->exit_signal);
178                 /*
179                  * If we were the last child thread and the leader has
180                  * exited already, and the leader's parent ignores SIGCHLD,
181                  * then we are the one who should release the leader.
182                  *
183                  * do_notify_parent() will have marked it self-reaping in
184                  * that case.
185                  */
186                 zap_leader = task_detached(leader);
187
188                 /*
189                  * This maintains the invariant that release_task()
190                  * only runs on a task in EXIT_DEAD, just for sanity.
191                  */
192                 if (zap_leader)
193                         leader->exit_state = EXIT_DEAD;
194         }
195
196         write_unlock_irq(&tasklist_lock);
197         release_thread(p);
198         call_rcu(&p->rcu, delayed_put_task_struct);
199
200         p = leader;
201         if (unlikely(zap_leader))
202                 goto repeat;
203 }
204
205 /*
206  * This checks not only the pgrp, but falls back on the pid if no
207  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
208  * without this...
209  *
210  * The caller must hold rcu lock or the tasklist lock.
211  */
212 struct pid *session_of_pgrp(struct pid *pgrp)
213 {
214         struct task_struct *p;
215         struct pid *sid = NULL;
216
217         p = pid_task(pgrp, PIDTYPE_PGID);
218         if (p == NULL)
219                 p = pid_task(pgrp, PIDTYPE_PID);
220         if (p != NULL)
221                 sid = task_session(p);
222
223         return sid;
224 }
225
226 /*
227  * Determine if a process group is "orphaned", according to the POSIX
228  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
229  * by terminal-generated stop signals.  Newly orphaned process groups are
230  * to receive a SIGHUP and a SIGCONT.
231  *
232  * "I ask you, have you ever known what it is to be an orphan?"
233  */
234 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
235 {
236         struct task_struct *p;
237
238         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
239                 if ((p == ignored_task) ||
240                     (p->exit_state && thread_group_empty(p)) ||
241                     is_global_init(p->real_parent))
242                         continue;
243
244                 if (task_pgrp(p->real_parent) != pgrp &&
245                     task_session(p->real_parent) == task_session(p))
246                         return 0;
247         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
248
249         return 1;
250 }
251
252 int is_current_pgrp_orphaned(void)
253 {
254         int retval;
255
256         read_lock(&tasklist_lock);
257         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
258         read_unlock(&tasklist_lock);
259
260         return retval;
261 }
262
263 static int has_stopped_jobs(struct pid *pgrp)
264 {
265         int retval = 0;
266         struct task_struct *p;
267
268         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
269                 if (!task_is_stopped(p))
270                         continue;
271                 retval = 1;
272                 break;
273         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
274         return retval;
275 }
276
277 /*
278  * Check to see if any process groups have become orphaned as
279  * a result of our exiting, and if they have any stopped jobs,
280  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
281  */
282 static void
283 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
284 {
285         struct pid *pgrp = task_pgrp(tsk);
286         struct task_struct *ignored_task = tsk;
287
288         if (!parent)
289                  /* exit: our father is in a different pgrp than
290                   * we are and we were the only connection outside.
291                   */
292                 parent = tsk->real_parent;
293         else
294                 /* reparent: our child is in a different pgrp than
295                  * we are, and it was the only connection outside.
296                  */
297                 ignored_task = NULL;
298
299         if (task_pgrp(parent) != pgrp &&
300             task_session(parent) == task_session(tsk) &&
301             will_become_orphaned_pgrp(pgrp, ignored_task) &&
302             has_stopped_jobs(pgrp)) {
303                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
304                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
305         }
306 }
307
308 /**
309  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
310  *
311  * If a kernel thread is launched as a result of a system call, or if
312  * it ever exits, it should generally reparent itself to kthreadd so it
313  * isn't in the way of other processes and is correctly cleaned up on exit.
314  *
315  * The various task state such as scheduling policy and priority may have
316  * been inherited from a user process, so we reset them to sane values here.
317  *
318  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
319  */
320 static void reparent_to_kthreadd(void)
321 {
322         write_lock_irq(&tasklist_lock);
323
324         ptrace_unlink(current);
325         /* Reparent to init */
326         current->real_parent = current->parent = kthreadd_task;
327         list_move_tail(&current->sibling, &current->real_parent->children);
328
329         /* Set the exit signal to SIGCHLD so we signal init on exit */
330         current->exit_signal = SIGCHLD;
331
332         if (task_nice(current) < 0)
333                 set_user_nice(current, 0);
334         /* cpus_allowed? */
335         /* rt_priority? */
336         /* signals? */
337         security_task_reparent_to_init(current);
338         memcpy(current->signal->rlim, init_task.signal->rlim,
339                sizeof(current->signal->rlim));
340         atomic_inc(&(INIT_USER->__count));
341         write_unlock_irq(&tasklist_lock);
342         switch_uid(INIT_USER);
343 }
344
345 void __set_special_pids(struct pid *pid)
346 {
347         struct task_struct *curr = current->group_leader;
348         pid_t nr = pid_nr(pid);
349
350         if (task_session(curr) != pid) {
351                 change_pid(curr, PIDTYPE_SID, pid);
352                 set_task_session(curr, nr);
353         }
354         if (task_pgrp(curr) != pid) {
355                 change_pid(curr, PIDTYPE_PGID, pid);
356                 set_task_pgrp(curr, nr);
357         }
358 }
359
360 static void set_special_pids(struct pid *pid)
361 {
362         write_lock_irq(&tasklist_lock);
363         __set_special_pids(pid);
364         write_unlock_irq(&tasklist_lock);
365 }
366
367 /*
368  * Let kernel threads use this to say that they
369  * allow a certain signal (since daemonize() will
370  * have disabled all of them by default).
371  */
372 int allow_signal(int sig)
373 {
374         if (!valid_signal(sig) || sig < 1)
375                 return -EINVAL;
376
377         spin_lock_irq(&current->sighand->siglock);
378         sigdelset(&current->blocked, sig);
379         if (!current->mm) {
380                 /* Kernel threads handle their own signals.
381                    Let the signal code know it'll be handled, so
382                    that they don't get converted to SIGKILL or
383                    just silently dropped */
384                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
385         }
386         recalc_sigpending();
387         spin_unlock_irq(&current->sighand->siglock);
388         return 0;
389 }
390
391 EXPORT_SYMBOL(allow_signal);
392
393 int disallow_signal(int sig)
394 {
395         if (!valid_signal(sig) || sig < 1)
396                 return -EINVAL;
397
398         spin_lock_irq(&current->sighand->siglock);
399         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
400         recalc_sigpending();
401         spin_unlock_irq(&current->sighand->siglock);
402         return 0;
403 }
404
405 EXPORT_SYMBOL(disallow_signal);
406
407 /*
408  *      Put all the gunge required to become a kernel thread without
409  *      attached user resources in one place where it belongs.
410  */
411
412 void daemonize(const char *name, ...)
413 {
414         va_list args;
415         struct fs_struct *fs;
416         sigset_t blocked;
417
418         va_start(args, name);
419         vsnprintf(current->comm, sizeof(current->comm), name, args);
420         va_end(args);
421
422         /*
423          * If we were started as result of loading a module, close all of the
424          * user space pages.  We don't need them, and if we didn't close them
425          * they would be locked into memory.
426          */
427         exit_mm(current);
428         /*
429          * We don't want to have TIF_FREEZE set if the system-wide hibernation
430          * or suspend transition begins right now.
431          */
432         current->flags |= (PF_NOFREEZE | PF_KTHREAD);
433
434         if (current->nsproxy != &init_nsproxy) {
435                 get_nsproxy(&init_nsproxy);
436                 switch_task_namespaces(current, &init_nsproxy);
437         }
438         set_special_pids(&init_struct_pid);
439         proc_clear_tty(current);
440
441         /* Block and flush all signals */
442         sigfillset(&blocked);
443         sigprocmask(SIG_BLOCK, &blocked, NULL);
444         flush_signals(current);
445
446         /* Become as one with the init task */
447
448         exit_fs(current);       /* current->fs->count--; */
449         fs = init_task.fs;
450         current->fs = fs;
451         atomic_inc(&fs->count);
452
453         exit_files(current);
454         current->files = init_task.files;
455         atomic_inc(&current->files->count);
456
457         reparent_to_kthreadd();
458 }
459
460 EXPORT_SYMBOL(daemonize);
461
462 static void close_files(struct files_struct * files)
463 {
464         int i, j;
465         struct fdtable *fdt;
466
467         j = 0;
468
469         /*
470          * It is safe to dereference the fd table without RCU or
471          * ->file_lock because this is the last reference to the
472          * files structure.
473          */
474         fdt = files_fdtable(files);
475         for (;;) {
476                 unsigned long set;
477                 i = j * __NFDBITS;
478                 if (i >= fdt->max_fds)
479                         break;
480                 set = fdt->open_fds->fds_bits[j++];
481                 while (set) {
482                         if (set & 1) {
483                                 struct file * file = xchg(&fdt->fd[i], NULL);
484                                 if (file) {
485                                         filp_close(file, files);
486                                         cond_resched();
487                                 }
488                         }
489                         i++;
490                         set >>= 1;
491                 }
492         }
493 }
494
495 struct files_struct *get_files_struct(struct task_struct *task)
496 {
497         struct files_struct *files;
498
499         task_lock(task);
500         files = task->files;
501         if (files)
502                 atomic_inc(&files->count);
503         task_unlock(task);
504
505         return files;
506 }
507
508 void put_files_struct(struct files_struct *files)
509 {
510         struct fdtable *fdt;
511
512         if (atomic_dec_and_test(&files->count)) {
513                 close_files(files);
514                 /*
515                  * Free the fd and fdset arrays if we expanded them.
516                  * If the fdtable was embedded, pass files for freeing
517                  * at the end of the RCU grace period. Otherwise,
518                  * you can free files immediately.
519                  */
520                 fdt = files_fdtable(files);
521                 if (fdt != &files->fdtab)
522                         kmem_cache_free(files_cachep, files);
523                 free_fdtable(fdt);
524         }
525 }
526
527 void reset_files_struct(struct files_struct *files)
528 {
529         struct task_struct *tsk = current;
530         struct files_struct *old;
531
532         old = tsk->files;
533         task_lock(tsk);
534         tsk->files = files;
535         task_unlock(tsk);
536         put_files_struct(old);
537 }
538
539 void exit_files(struct task_struct *tsk)
540 {
541         struct files_struct * files = tsk->files;
542
543         if (files) {
544                 task_lock(tsk);
545                 tsk->files = NULL;
546                 task_unlock(tsk);
547                 put_files_struct(files);
548         }
549 }
550
551 void put_fs_struct(struct fs_struct *fs)
552 {
553         /* No need to hold fs->lock if we are killing it */
554         if (atomic_dec_and_test(&fs->count)) {
555                 path_put(&fs->root);
556                 path_put(&fs->pwd);
557                 kmem_cache_free(fs_cachep, fs);
558         }
559 }
560
561 void exit_fs(struct task_struct *tsk)
562 {
563         struct fs_struct * fs = tsk->fs;
564
565         if (fs) {
566                 task_lock(tsk);
567                 tsk->fs = NULL;
568                 task_unlock(tsk);
569                 put_fs_struct(fs);
570         }
571 }
572
573 EXPORT_SYMBOL_GPL(exit_fs);
574
575 #ifdef CONFIG_MM_OWNER
576 /*
577  * Task p is exiting and it owned mm, lets find a new owner for it
578  */
579 static inline int
580 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
581 {
582         /*
583          * If there are other users of the mm and the owner (us) is exiting
584          * we need to find a new owner to take on the responsibility.
585          */
586         if (!mm)
587                 return 0;
588         if (atomic_read(&mm->mm_users) <= 1)
589                 return 0;
590         if (mm->owner != p)
591                 return 0;
592         return 1;
593 }
594
595 void mm_update_next_owner(struct mm_struct *mm)
596 {
597         struct task_struct *c, *g, *p = current;
598
599 retry:
600         if (!mm_need_new_owner(mm, p))
601                 return;
602
603         read_lock(&tasklist_lock);
604         /*
605          * Search in the children
606          */
607         list_for_each_entry(c, &p->children, sibling) {
608                 if (c->mm == mm)
609                         goto assign_new_owner;
610         }
611
612         /*
613          * Search in the siblings
614          */
615         list_for_each_entry(c, &p->parent->children, sibling) {
616                 if (c->mm == mm)
617                         goto assign_new_owner;
618         }
619
620         /*
621          * Search through everything else. We should not get
622          * here often
623          */
624         do_each_thread(g, c) {
625                 if (c->mm == mm)
626                         goto assign_new_owner;
627         } while_each_thread(g, c);
628
629         read_unlock(&tasklist_lock);
630         return;
631
632 assign_new_owner:
633         BUG_ON(c == p);
634         get_task_struct(c);
635         /*
636          * The task_lock protects c->mm from changing.
637          * We always want mm->owner->mm == mm
638          */
639         task_lock(c);
640         /*
641          * Delay read_unlock() till we have the task_lock()
642          * to ensure that c does not slip away underneath us
643          */
644         read_unlock(&tasklist_lock);
645         if (c->mm != mm) {
646                 task_unlock(c);
647                 put_task_struct(c);
648                 goto retry;
649         }
650         cgroup_mm_owner_callbacks(mm->owner, c);
651         mm->owner = c;
652         task_unlock(c);
653         put_task_struct(c);
654 }
655 #endif /* CONFIG_MM_OWNER */
656
657 /*
658  * Turn us into a lazy TLB process if we
659  * aren't already..
660  */
661 static void exit_mm(struct task_struct * tsk)
662 {
663         struct mm_struct *mm = tsk->mm;
664         struct core_state *core_state;
665
666         mm_release(tsk, mm);
667         if (!mm)
668                 return;
669         /*
670          * Serialize with any possible pending coredump.
671          * We must hold mmap_sem around checking core_state
672          * and clearing tsk->mm.  The core-inducing thread
673          * will increment ->nr_threads for each thread in the
674          * group with ->mm != NULL.
675          */
676         down_read(&mm->mmap_sem);
677         core_state = mm->core_state;
678         if (core_state) {
679                 struct core_thread self;
680                 up_read(&mm->mmap_sem);
681
682                 self.task = tsk;
683                 self.next = xchg(&core_state->dumper.next, &self);
684                 /*
685                  * Implies mb(), the result of xchg() must be visible
686                  * to core_state->dumper.
687                  */
688                 if (atomic_dec_and_test(&core_state->nr_threads))
689                         complete(&core_state->startup);
690
691                 for (;;) {
692                         set_task_state(tsk, TASK_UNINTERRUPTIBLE);
693                         if (!self.task) /* see coredump_finish() */
694                                 break;
695                         schedule();
696                 }
697                 __set_task_state(tsk, TASK_RUNNING);
698                 down_read(&mm->mmap_sem);
699         }
700         atomic_inc(&mm->mm_count);
701         BUG_ON(mm != tsk->active_mm);
702         /* more a memory barrier than a real lock */
703         task_lock(tsk);
704         tsk->mm = NULL;
705         up_read(&mm->mmap_sem);
706         enter_lazy_tlb(mm, current);
707         /* We don't want this task to be frozen prematurely */
708         clear_freeze_flag(tsk);
709         task_unlock(tsk);
710         mm_update_next_owner(mm);
711         mmput(mm);
712 }
713
714 /*
715  * Return nonzero if @parent's children should reap themselves.
716  *
717  * Called with write_lock_irq(&tasklist_lock) held.
718  */
719 static int ignoring_children(struct task_struct *parent)
720 {
721         int ret;
722         struct sighand_struct *psig = parent->sighand;
723         unsigned long flags;
724         spin_lock_irqsave(&psig->siglock, flags);
725         ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
726                (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
727         spin_unlock_irqrestore(&psig->siglock, flags);
728         return ret;
729 }
730
731 /*
732  * Detach all tasks we were using ptrace on.
733  * Any that need to be release_task'd are put on the @dead list.
734  *
735  * Called with write_lock(&tasklist_lock) held.
736  */
737 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
738 {
739         struct task_struct *p, *n;
740         int ign = -1;
741
742         list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
743                 __ptrace_unlink(p);
744
745                 if (p->exit_state != EXIT_ZOMBIE)
746                         continue;
747
748                 /*
749                  * If it's a zombie, our attachedness prevented normal
750                  * parent notification or self-reaping.  Do notification
751                  * now if it would have happened earlier.  If it should
752                  * reap itself, add it to the @dead list.  We can't call
753                  * release_task() here because we already hold tasklist_lock.
754                  *
755                  * If it's our own child, there is no notification to do.
756                  * But if our normal children self-reap, then this child
757                  * was prevented by ptrace and we must reap it now.
758                  */
759                 if (!task_detached(p) && thread_group_empty(p)) {
760                         if (!same_thread_group(p->real_parent, parent))
761                                 do_notify_parent(p, p->exit_signal);
762                         else {
763                                 if (ign < 0)
764                                         ign = ignoring_children(parent);
765                                 if (ign)
766                                         p->exit_signal = -1;
767                         }
768                 }
769
770                 if (task_detached(p)) {
771                         /*
772                          * Mark it as in the process of being reaped.
773                          */
774                         p->exit_state = EXIT_DEAD;
775                         list_add(&p->ptrace_entry, dead);
776                 }
777         }
778 }
779
780 /*
781  * Finish up exit-time ptrace cleanup.
782  *
783  * Called without locks.
784  */
785 static void ptrace_exit_finish(struct task_struct *parent,
786                                struct list_head *dead)
787 {
788         struct task_struct *p, *n;
789
790         BUG_ON(!list_empty(&parent->ptraced));
791
792         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
793                 list_del_init(&p->ptrace_entry);
794                 release_task(p);
795         }
796 }
797
798 static void reparent_thread(struct task_struct *p, struct task_struct *father)
799 {
800         if (p->pdeath_signal)
801                 /* We already hold the tasklist_lock here.  */
802                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
803
804         list_move_tail(&p->sibling, &p->real_parent->children);
805
806         /* If this is a threaded reparent there is no need to
807          * notify anyone anything has happened.
808          */
809         if (same_thread_group(p->real_parent, father))
810                 return;
811
812         /* We don't want people slaying init.  */
813         if (!task_detached(p))
814                 p->exit_signal = SIGCHLD;
815
816         /* If we'd notified the old parent about this child's death,
817          * also notify the new parent.
818          */
819         if (!ptrace_reparented(p) &&
820             p->exit_state == EXIT_ZOMBIE &&
821             !task_detached(p) && thread_group_empty(p))
822                 do_notify_parent(p, p->exit_signal);
823
824         kill_orphaned_pgrp(p, father);
825 }
826
827 /*
828  * When we die, we re-parent all our children.
829  * Try to give them to another thread in our thread
830  * group, and if no such member exists, give it to
831  * the child reaper process (ie "init") in our pid
832  * space.
833  */
834 static struct task_struct *find_new_reaper(struct task_struct *father)
835 {
836         struct pid_namespace *pid_ns = task_active_pid_ns(father);
837         struct task_struct *thread;
838
839         thread = father;
840         while_each_thread(father, thread) {
841                 if (thread->flags & PF_EXITING)
842                         continue;
843                 if (unlikely(pid_ns->child_reaper == father))
844                         pid_ns->child_reaper = thread;
845                 return thread;
846         }
847
848         if (unlikely(pid_ns->child_reaper == father)) {
849                 write_unlock_irq(&tasklist_lock);
850                 if (unlikely(pid_ns == &init_pid_ns))
851                         panic("Attempted to kill init!");
852
853                 zap_pid_ns_processes(pid_ns);
854                 write_lock_irq(&tasklist_lock);
855                 /*
856                  * We can not clear ->child_reaper or leave it alone.
857                  * There may by stealth EXIT_DEAD tasks on ->children,
858                  * forget_original_parent() must move them somewhere.
859                  */
860                 pid_ns->child_reaper = init_pid_ns.child_reaper;
861         }
862
863         return pid_ns->child_reaper;
864 }
865
866 static void forget_original_parent(struct task_struct *father)
867 {
868         struct task_struct *p, *n, *reaper;
869         LIST_HEAD(ptrace_dead);
870
871         write_lock_irq(&tasklist_lock);
872         reaper = find_new_reaper(father);
873         /*
874          * First clean up ptrace if we were using it.
875          */
876         ptrace_exit(father, &ptrace_dead);
877
878         list_for_each_entry_safe(p, n, &father->children, sibling) {
879                 p->real_parent = reaper;
880                 if (p->parent == father) {
881                         BUG_ON(p->ptrace);
882                         p->parent = p->real_parent;
883                 }
884                 reparent_thread(p, father);
885         }
886
887         write_unlock_irq(&tasklist_lock);
888         BUG_ON(!list_empty(&father->children));
889
890         ptrace_exit_finish(father, &ptrace_dead);
891 }
892
893 /*
894  * Send signals to all our closest relatives so that they know
895  * to properly mourn us..
896  */
897 static void exit_notify(struct task_struct *tsk, int group_dead)
898 {
899         int signal;
900         void *cookie;
901
902         /*
903          * This does two things:
904          *
905          * A.  Make init inherit all the child processes
906          * B.  Check to see if any process groups have become orphaned
907          *      as a result of our exiting, and if they have any stopped
908          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
909          */
910         forget_original_parent(tsk);
911         exit_task_namespaces(tsk);
912
913         write_lock_irq(&tasklist_lock);
914         if (group_dead)
915                 kill_orphaned_pgrp(tsk->group_leader, NULL);
916
917         /* Let father know we died
918          *
919          * Thread signals are configurable, but you aren't going to use
920          * that to send signals to arbitary processes.
921          * That stops right now.
922          *
923          * If the parent exec id doesn't match the exec id we saved
924          * when we started then we know the parent has changed security
925          * domain.
926          *
927          * If our self_exec id doesn't match our parent_exec_id then
928          * we have changed execution domain as these two values started
929          * the same after a fork.
930          */
931         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
932             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
933              tsk->self_exec_id != tsk->parent_exec_id) &&
934             !capable(CAP_KILL))
935                 tsk->exit_signal = SIGCHLD;
936
937         signal = tracehook_notify_death(tsk, &cookie, group_dead);
938         if (signal >= 0)
939                 signal = do_notify_parent(tsk, signal);
940
941         tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
942
943         /* mt-exec, de_thread() is waiting for us */
944         if (thread_group_leader(tsk) &&
945             tsk->signal->group_exit_task &&
946             tsk->signal->notify_count < 0)
947                 wake_up_process(tsk->signal->group_exit_task);
948
949         write_unlock_irq(&tasklist_lock);
950
951         tracehook_report_death(tsk, signal, cookie, group_dead);
952
953         /* If the process is dead, release it - nobody will wait for it */
954         if (signal == DEATH_REAP)
955                 release_task(tsk);
956 }
957
958 #ifdef CONFIG_DEBUG_STACK_USAGE
959 static void check_stack_usage(void)
960 {
961         static DEFINE_SPINLOCK(low_water_lock);
962         static int lowest_to_date = THREAD_SIZE;
963         unsigned long *n = end_of_stack(current);
964         unsigned long free;
965
966         while (*n == 0)
967                 n++;
968         free = (unsigned long)n - (unsigned long)end_of_stack(current);
969
970         if (free >= lowest_to_date)
971                 return;
972
973         spin_lock(&low_water_lock);
974         if (free < lowest_to_date) {
975                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
976                                 "left\n",
977                                 current->comm, free);
978                 lowest_to_date = free;
979         }
980         spin_unlock(&low_water_lock);
981 }
982 #else
983 static inline void check_stack_usage(void) {}
984 #endif
985
986 NORET_TYPE void do_exit(long code)
987 {
988         struct task_struct *tsk = current;
989         int group_dead;
990
991         profile_task_exit(tsk);
992
993         WARN_ON(atomic_read(&tsk->fs_excl));
994
995         if (unlikely(in_interrupt()))
996                 panic("Aiee, killing interrupt handler!");
997         if (unlikely(!tsk->pid))
998                 panic("Attempted to kill the idle task!");
999
1000         tracehook_report_exit(&code);
1001
1002         /*
1003          * We're taking recursive faults here in do_exit. Safest is to just
1004          * leave this task alone and wait for reboot.
1005          */
1006         if (unlikely(tsk->flags & PF_EXITING)) {
1007                 printk(KERN_ALERT
1008                         "Fixing recursive fault but reboot is needed!\n");
1009                 /*
1010                  * We can do this unlocked here. The futex code uses
1011                  * this flag just to verify whether the pi state
1012                  * cleanup has been done or not. In the worst case it
1013                  * loops once more. We pretend that the cleanup was
1014                  * done as there is no way to return. Either the
1015                  * OWNER_DIED bit is set by now or we push the blocked
1016                  * task into the wait for ever nirwana as well.
1017                  */
1018                 tsk->flags |= PF_EXITPIDONE;
1019                 if (tsk->io_context)
1020                         exit_io_context();
1021                 set_current_state(TASK_UNINTERRUPTIBLE);
1022                 schedule();
1023         }
1024
1025         exit_signals(tsk);  /* sets PF_EXITING */
1026         /*
1027          * tsk->flags are checked in the futex code to protect against
1028          * an exiting task cleaning up the robust pi futexes.
1029          */
1030         smp_mb();
1031         spin_unlock_wait(&tsk->pi_lock);
1032
1033         if (unlikely(in_atomic()))
1034                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1035                                 current->comm, task_pid_nr(current),
1036                                 preempt_count());
1037
1038         acct_update_integrals(tsk);
1039         if (tsk->mm) {
1040                 update_hiwater_rss(tsk->mm);
1041                 update_hiwater_vm(tsk->mm);
1042         }
1043         group_dead = atomic_dec_and_test(&tsk->signal->live);
1044         if (group_dead) {
1045                 hrtimer_cancel(&tsk->signal->real_timer);
1046                 exit_itimers(tsk->signal);
1047         }
1048         acct_collect(code, group_dead);
1049 #ifdef CONFIG_FUTEX
1050         if (unlikely(tsk->robust_list))
1051                 exit_robust_list(tsk);
1052 #ifdef CONFIG_COMPAT
1053         if (unlikely(tsk->compat_robust_list))
1054                 compat_exit_robust_list(tsk);
1055 #endif
1056 #endif
1057         if (group_dead)
1058                 tty_audit_exit();
1059         if (unlikely(tsk->audit_context))
1060                 audit_free(tsk);
1061
1062         tsk->exit_code = code;
1063         taskstats_exit(tsk, group_dead);
1064
1065         exit_mm(tsk);
1066
1067         if (group_dead)
1068                 acct_process();
1069         exit_sem(tsk);
1070         exit_files(tsk);
1071         exit_fs(tsk);
1072         check_stack_usage();
1073         exit_thread();
1074         cgroup_exit(tsk, 1);
1075         exit_keys(tsk);
1076
1077         if (group_dead && tsk->signal->leader)
1078                 disassociate_ctty(1);
1079
1080         module_put(task_thread_info(tsk)->exec_domain->module);
1081         if (tsk->binfmt)
1082                 module_put(tsk->binfmt->module);
1083
1084         proc_exit_connector(tsk);
1085         exit_notify(tsk, group_dead);
1086 #ifdef CONFIG_NUMA
1087         mpol_put(tsk->mempolicy);
1088         tsk->mempolicy = NULL;
1089 #endif
1090 #ifdef CONFIG_FUTEX
1091         /*
1092          * This must happen late, after the PID is not
1093          * hashed anymore:
1094          */
1095         if (unlikely(!list_empty(&tsk->pi_state_list)))
1096                 exit_pi_state_list(tsk);
1097         if (unlikely(current->pi_state_cache))
1098                 kfree(current->pi_state_cache);
1099 #endif
1100         /*
1101          * Make sure we are holding no locks:
1102          */
1103         debug_check_no_locks_held(tsk);
1104         /*
1105          * We can do this unlocked here. The futex code uses this flag
1106          * just to verify whether the pi state cleanup has been done
1107          * or not. In the worst case it loops once more.
1108          */
1109         tsk->flags |= PF_EXITPIDONE;
1110
1111         if (tsk->io_context)
1112                 exit_io_context();
1113
1114         if (tsk->splice_pipe)
1115                 __free_pipe_info(tsk->splice_pipe);
1116
1117         preempt_disable();
1118         /* causes final put_task_struct in finish_task_switch(). */
1119         tsk->state = TASK_DEAD;
1120
1121         schedule();
1122         BUG();
1123         /* Avoid "noreturn function does return".  */
1124         for (;;)
1125                 cpu_relax();    /* For when BUG is null */
1126 }
1127
1128 EXPORT_SYMBOL_GPL(do_exit);
1129
1130 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1131 {
1132         if (comp)
1133                 complete(comp);
1134
1135         do_exit(code);
1136 }
1137
1138 EXPORT_SYMBOL(complete_and_exit);
1139
1140 asmlinkage long sys_exit(int error_code)
1141 {
1142         do_exit((error_code&0xff)<<8);
1143 }
1144
1145 /*
1146  * Take down every thread in the group.  This is called by fatal signals
1147  * as well as by sys_exit_group (below).
1148  */
1149 NORET_TYPE void
1150 do_group_exit(int exit_code)
1151 {
1152         struct signal_struct *sig = current->signal;
1153
1154         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1155
1156         if (signal_group_exit(sig))
1157                 exit_code = sig->group_exit_code;
1158         else if (!thread_group_empty(current)) {
1159                 struct sighand_struct *const sighand = current->sighand;
1160                 spin_lock_irq(&sighand->siglock);
1161                 if (signal_group_exit(sig))
1162                         /* Another thread got here before we took the lock.  */
1163                         exit_code = sig->group_exit_code;
1164                 else {
1165                         sig->group_exit_code = exit_code;
1166                         sig->flags = SIGNAL_GROUP_EXIT;
1167                         zap_other_threads(current);
1168                 }
1169                 spin_unlock_irq(&sighand->siglock);
1170         }
1171
1172         do_exit(exit_code);
1173         /* NOTREACHED */
1174 }
1175
1176 /*
1177  * this kills every thread in the thread group. Note that any externally
1178  * wait4()-ing process will get the correct exit code - even if this
1179  * thread is not the thread group leader.
1180  */
1181 asmlinkage void sys_exit_group(int error_code)
1182 {
1183         do_group_exit((error_code & 0xff) << 8);
1184 }
1185
1186 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1187 {
1188         struct pid *pid = NULL;
1189         if (type == PIDTYPE_PID)
1190                 pid = task->pids[type].pid;
1191         else if (type < PIDTYPE_MAX)
1192                 pid = task->group_leader->pids[type].pid;
1193         return pid;
1194 }
1195
1196 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1197                           struct task_struct *p)
1198 {
1199         int err;
1200
1201         if (type < PIDTYPE_MAX) {
1202                 if (task_pid_type(p, type) != pid)
1203                         return 0;
1204         }
1205
1206         /* Wait for all children (clone and not) if __WALL is set;
1207          * otherwise, wait for clone children *only* if __WCLONE is
1208          * set; otherwise, wait for non-clone children *only*.  (Note:
1209          * A "clone" child here is one that reports to its parent
1210          * using a signal other than SIGCHLD.) */
1211         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1212             && !(options & __WALL))
1213                 return 0;
1214
1215         err = security_task_wait(p);
1216         if (err)
1217                 return err;
1218
1219         return 1;
1220 }
1221
1222 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1223                                int why, int status,
1224                                struct siginfo __user *infop,
1225                                struct rusage __user *rusagep)
1226 {
1227         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1228
1229         put_task_struct(p);
1230         if (!retval)
1231                 retval = put_user(SIGCHLD, &infop->si_signo);
1232         if (!retval)
1233                 retval = put_user(0, &infop->si_errno);
1234         if (!retval)
1235                 retval = put_user((short)why, &infop->si_code);
1236         if (!retval)
1237                 retval = put_user(pid, &infop->si_pid);
1238         if (!retval)
1239                 retval = put_user(uid, &infop->si_uid);
1240         if (!retval)
1241                 retval = put_user(status, &infop->si_status);
1242         if (!retval)
1243                 retval = pid;
1244         return retval;
1245 }
1246
1247 /*
1248  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1249  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1250  * the lock and this task is uninteresting.  If we return nonzero, we have
1251  * released the lock and the system call should return.
1252  */
1253 static int wait_task_zombie(struct task_struct *p, int options,
1254                             struct siginfo __user *infop,
1255                             int __user *stat_addr, struct rusage __user *ru)
1256 {
1257         unsigned long state;
1258         int retval, status, traced;
1259         pid_t pid = task_pid_vnr(p);
1260
1261         if (!likely(options & WEXITED))
1262                 return 0;
1263
1264         if (unlikely(options & WNOWAIT)) {
1265                 uid_t uid = p->uid;
1266                 int exit_code = p->exit_code;
1267                 int why, status;
1268
1269                 get_task_struct(p);
1270                 read_unlock(&tasklist_lock);
1271                 if ((exit_code & 0x7f) == 0) {
1272                         why = CLD_EXITED;
1273                         status = exit_code >> 8;
1274                 } else {
1275                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1276                         status = exit_code & 0x7f;
1277                 }
1278                 return wait_noreap_copyout(p, pid, uid, why,
1279                                            status, infop, ru);
1280         }
1281
1282         /*
1283          * Try to move the task's state to DEAD
1284          * only one thread is allowed to do this:
1285          */
1286         state = xchg(&p->exit_state, EXIT_DEAD);
1287         if (state != EXIT_ZOMBIE) {
1288                 BUG_ON(state != EXIT_DEAD);
1289                 return 0;
1290         }
1291
1292         traced = ptrace_reparented(p);
1293
1294         if (likely(!traced)) {
1295                 struct signal_struct *psig;
1296                 struct signal_struct *sig;
1297
1298                 /*
1299                  * The resource counters for the group leader are in its
1300                  * own task_struct.  Those for dead threads in the group
1301                  * are in its signal_struct, as are those for the child
1302                  * processes it has previously reaped.  All these
1303                  * accumulate in the parent's signal_struct c* fields.
1304                  *
1305                  * We don't bother to take a lock here to protect these
1306                  * p->signal fields, because they are only touched by
1307                  * __exit_signal, which runs with tasklist_lock
1308                  * write-locked anyway, and so is excluded here.  We do
1309                  * need to protect the access to p->parent->signal fields,
1310                  * as other threads in the parent group can be right
1311                  * here reaping other children at the same time.
1312                  */
1313                 spin_lock_irq(&p->parent->sighand->siglock);
1314                 psig = p->parent->signal;
1315                 sig = p->signal;
1316                 psig->cutime =
1317                         cputime_add(psig->cutime,
1318                         cputime_add(p->utime,
1319                         cputime_add(sig->utime,
1320                                     sig->cutime)));
1321                 psig->cstime =
1322                         cputime_add(psig->cstime,
1323                         cputime_add(p->stime,
1324                         cputime_add(sig->stime,
1325                                     sig->cstime)));
1326                 psig->cgtime =
1327                         cputime_add(psig->cgtime,
1328                         cputime_add(p->gtime,
1329                         cputime_add(sig->gtime,
1330                                     sig->cgtime)));
1331                 psig->cmin_flt +=
1332                         p->min_flt + sig->min_flt + sig->cmin_flt;
1333                 psig->cmaj_flt +=
1334                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1335                 psig->cnvcsw +=
1336                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1337                 psig->cnivcsw +=
1338                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1339                 psig->cinblock +=
1340                         task_io_get_inblock(p) +
1341                         sig->inblock + sig->cinblock;
1342                 psig->coublock +=
1343                         task_io_get_oublock(p) +
1344                         sig->oublock + sig->coublock;
1345                 task_io_accounting_add(&psig->ioac, &p->ioac);
1346                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1347                 spin_unlock_irq(&p->parent->sighand->siglock);
1348         }
1349
1350         /*
1351          * Now we are sure this task is interesting, and no other
1352          * thread can reap it because we set its state to EXIT_DEAD.
1353          */
1354         read_unlock(&tasklist_lock);
1355
1356         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1357         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1358                 ? p->signal->group_exit_code : p->exit_code;
1359         if (!retval && stat_addr)
1360                 retval = put_user(status, stat_addr);
1361         if (!retval && infop)
1362                 retval = put_user(SIGCHLD, &infop->si_signo);
1363         if (!retval && infop)
1364                 retval = put_user(0, &infop->si_errno);
1365         if (!retval && infop) {
1366                 int why;
1367
1368                 if ((status & 0x7f) == 0) {
1369                         why = CLD_EXITED;
1370                         status >>= 8;
1371                 } else {
1372                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1373                         status &= 0x7f;
1374                 }
1375                 retval = put_user((short)why, &infop->si_code);
1376                 if (!retval)
1377                         retval = put_user(status, &infop->si_status);
1378         }
1379         if (!retval && infop)
1380                 retval = put_user(pid, &infop->si_pid);
1381         if (!retval && infop)
1382                 retval = put_user(p->uid, &infop->si_uid);
1383         if (!retval)
1384                 retval = pid;
1385
1386         if (traced) {
1387                 write_lock_irq(&tasklist_lock);
1388                 /* We dropped tasklist, ptracer could die and untrace */
1389                 ptrace_unlink(p);
1390                 /*
1391                  * If this is not a detached task, notify the parent.
1392                  * If it's still not detached after that, don't release
1393                  * it now.
1394                  */
1395                 if (!task_detached(p)) {
1396                         do_notify_parent(p, p->exit_signal);
1397                         if (!task_detached(p)) {
1398                                 p->exit_state = EXIT_ZOMBIE;
1399                                 p = NULL;
1400                         }
1401                 }
1402                 write_unlock_irq(&tasklist_lock);
1403         }
1404         if (p != NULL)
1405                 release_task(p);
1406
1407         return retval;
1408 }
1409
1410 /*
1411  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1412  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1413  * the lock and this task is uninteresting.  If we return nonzero, we have
1414  * released the lock and the system call should return.
1415  */
1416 static int wait_task_stopped(int ptrace, struct task_struct *p,
1417                              int options, struct siginfo __user *infop,
1418                              int __user *stat_addr, struct rusage __user *ru)
1419 {
1420         int retval, exit_code, why;
1421         uid_t uid = 0; /* unneeded, required by compiler */
1422         pid_t pid;
1423
1424         if (!(options & WUNTRACED))
1425                 return 0;
1426
1427         exit_code = 0;
1428         spin_lock_irq(&p->sighand->siglock);
1429
1430         if (unlikely(!task_is_stopped_or_traced(p)))
1431                 goto unlock_sig;
1432
1433         if (!ptrace && p->signal->group_stop_count > 0)
1434                 /*
1435                  * A group stop is in progress and this is the group leader.
1436                  * We won't report until all threads have stopped.
1437                  */
1438                 goto unlock_sig;
1439
1440         exit_code = p->exit_code;
1441         if (!exit_code)
1442                 goto unlock_sig;
1443
1444         if (!unlikely(options & WNOWAIT))
1445                 p->exit_code = 0;
1446
1447         uid = p->uid;
1448 unlock_sig:
1449         spin_unlock_irq(&p->sighand->siglock);
1450         if (!exit_code)
1451                 return 0;
1452
1453         /*
1454          * Now we are pretty sure this task is interesting.
1455          * Make sure it doesn't get reaped out from under us while we
1456          * give up the lock and then examine it below.  We don't want to
1457          * keep holding onto the tasklist_lock while we call getrusage and
1458          * possibly take page faults for user memory.
1459          */
1460         get_task_struct(p);
1461         pid = task_pid_vnr(p);
1462         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1463         read_unlock(&tasklist_lock);
1464
1465         if (unlikely(options & WNOWAIT))
1466                 return wait_noreap_copyout(p, pid, uid,
1467                                            why, exit_code,
1468                                            infop, ru);
1469
1470         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1471         if (!retval && stat_addr)
1472                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1473         if (!retval && infop)
1474                 retval = put_user(SIGCHLD, &infop->si_signo);
1475         if (!retval && infop)
1476                 retval = put_user(0, &infop->si_errno);
1477         if (!retval && infop)
1478                 retval = put_user((short)why, &infop->si_code);
1479         if (!retval && infop)
1480                 retval = put_user(exit_code, &infop->si_status);
1481         if (!retval && infop)
1482                 retval = put_user(pid, &infop->si_pid);
1483         if (!retval && infop)
1484                 retval = put_user(uid, &infop->si_uid);
1485         if (!retval)
1486                 retval = pid;
1487         put_task_struct(p);
1488
1489         BUG_ON(!retval);
1490         return retval;
1491 }
1492
1493 /*
1494  * Handle do_wait work for one task in a live, non-stopped state.
1495  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1496  * the lock and this task is uninteresting.  If we return nonzero, we have
1497  * released the lock and the system call should return.
1498  */
1499 static int wait_task_continued(struct task_struct *p, int options,
1500                                struct siginfo __user *infop,
1501                                int __user *stat_addr, struct rusage __user *ru)
1502 {
1503         int retval;
1504         pid_t pid;
1505         uid_t uid;
1506
1507         if (!unlikely(options & WCONTINUED))
1508                 return 0;
1509
1510         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1511                 return 0;
1512
1513         spin_lock_irq(&p->sighand->siglock);
1514         /* Re-check with the lock held.  */
1515         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1516                 spin_unlock_irq(&p->sighand->siglock);
1517                 return 0;
1518         }
1519         if (!unlikely(options & WNOWAIT))
1520                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1521         spin_unlock_irq(&p->sighand->siglock);
1522
1523         pid = task_pid_vnr(p);
1524         uid = p->uid;
1525         get_task_struct(p);
1526         read_unlock(&tasklist_lock);
1527
1528         if (!infop) {
1529                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1530                 put_task_struct(p);
1531                 if (!retval && stat_addr)
1532                         retval = put_user(0xffff, stat_addr);
1533                 if (!retval)
1534                         retval = pid;
1535         } else {
1536                 retval = wait_noreap_copyout(p, pid, uid,
1537                                              CLD_CONTINUED, SIGCONT,
1538                                              infop, ru);
1539                 BUG_ON(retval == 0);
1540         }
1541
1542         return retval;
1543 }
1544
1545 /*
1546  * Consider @p for a wait by @parent.
1547  *
1548  * -ECHILD should be in *@notask_error before the first call.
1549  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1550  * Returns zero if the search for a child should continue;
1551  * then *@notask_error is 0 if @p is an eligible child,
1552  * or another error from security_task_wait(), or still -ECHILD.
1553  */
1554 static int wait_consider_task(struct task_struct *parent, int ptrace,
1555                               struct task_struct *p, int *notask_error,
1556                               enum pid_type type, struct pid *pid, int options,
1557                               struct siginfo __user *infop,
1558                               int __user *stat_addr, struct rusage __user *ru)
1559 {
1560         int ret = eligible_child(type, pid, options, p);
1561         if (!ret)
1562                 return ret;
1563
1564         if (unlikely(ret < 0)) {
1565                 /*
1566                  * If we have not yet seen any eligible child,
1567                  * then let this error code replace -ECHILD.
1568                  * A permission error will give the user a clue
1569                  * to look for security policy problems, rather
1570                  * than for mysterious wait bugs.
1571                  */
1572                 if (*notask_error)
1573                         *notask_error = ret;
1574         }
1575
1576         if (likely(!ptrace) && unlikely(p->ptrace)) {
1577                 /*
1578                  * This child is hidden by ptrace.
1579                  * We aren't allowed to see it now, but eventually we will.
1580                  */
1581                 *notask_error = 0;
1582                 return 0;
1583         }
1584
1585         if (p->exit_state == EXIT_DEAD)
1586                 return 0;
1587
1588         /*
1589          * We don't reap group leaders with subthreads.
1590          */
1591         if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1592                 return wait_task_zombie(p, options, infop, stat_addr, ru);
1593
1594         /*
1595          * It's stopped or running now, so it might
1596          * later continue, exit, or stop again.
1597          */
1598         *notask_error = 0;
1599
1600         if (task_is_stopped_or_traced(p))
1601                 return wait_task_stopped(ptrace, p, options,
1602                                          infop, stat_addr, ru);
1603
1604         return wait_task_continued(p, options, infop, stat_addr, ru);
1605 }
1606
1607 /*
1608  * Do the work of do_wait() for one thread in the group, @tsk.
1609  *
1610  * -ECHILD should be in *@notask_error before the first call.
1611  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1612  * Returns zero if the search for a child should continue; then
1613  * *@notask_error is 0 if there were any eligible children,
1614  * or another error from security_task_wait(), or still -ECHILD.
1615  */
1616 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1617                           enum pid_type type, struct pid *pid, int options,
1618                           struct siginfo __user *infop, int __user *stat_addr,
1619                           struct rusage __user *ru)
1620 {
1621         struct task_struct *p;
1622
1623         list_for_each_entry(p, &tsk->children, sibling) {
1624                 /*
1625                  * Do not consider detached threads.
1626                  */
1627                 if (!task_detached(p)) {
1628                         int ret = wait_consider_task(tsk, 0, p, notask_error,
1629                                                      type, pid, options,
1630                                                      infop, stat_addr, ru);
1631                         if (ret)
1632                                 return ret;
1633                 }
1634         }
1635
1636         return 0;
1637 }
1638
1639 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1640                           enum pid_type type, struct pid *pid, int options,
1641                           struct siginfo __user *infop, int __user *stat_addr,
1642                           struct rusage __user *ru)
1643 {
1644         struct task_struct *p;
1645
1646         /*
1647          * Traditionally we see ptrace'd stopped tasks regardless of options.
1648          */
1649         options |= WUNTRACED;
1650
1651         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1652                 int ret = wait_consider_task(tsk, 1, p, notask_error,
1653                                              type, pid, options,
1654                                              infop, stat_addr, ru);
1655                 if (ret)
1656                         return ret;
1657         }
1658
1659         return 0;
1660 }
1661
1662 static long do_wait(enum pid_type type, struct pid *pid, int options,
1663                     struct siginfo __user *infop, int __user *stat_addr,
1664                     struct rusage __user *ru)
1665 {
1666         DECLARE_WAITQUEUE(wait, current);
1667         struct task_struct *tsk;
1668         int retval;
1669
1670         add_wait_queue(&current->signal->wait_chldexit,&wait);
1671 repeat:
1672         /*
1673          * If there is nothing that can match our critiera just get out.
1674          * We will clear @retval to zero if we see any child that might later
1675          * match our criteria, even if we are not able to reap it yet.
1676          */
1677         retval = -ECHILD;
1678         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1679                 goto end;
1680
1681         current->state = TASK_INTERRUPTIBLE;
1682         read_lock(&tasklist_lock);
1683         tsk = current;
1684         do {
1685                 int tsk_result = do_wait_thread(tsk, &retval,
1686                                                 type, pid, options,
1687                                                 infop, stat_addr, ru);
1688                 if (!tsk_result)
1689                         tsk_result = ptrace_do_wait(tsk, &retval,
1690                                                     type, pid, options,
1691                                                     infop, stat_addr, ru);
1692                 if (tsk_result) {
1693                         /*
1694                          * tasklist_lock is unlocked and we have a final result.
1695                          */
1696                         retval = tsk_result;
1697                         goto end;
1698                 }
1699
1700                 if (options & __WNOTHREAD)
1701                         break;
1702                 tsk = next_thread(tsk);
1703                 BUG_ON(tsk->signal != current->signal);
1704         } while (tsk != current);
1705         read_unlock(&tasklist_lock);
1706
1707         if (!retval && !(options & WNOHANG)) {
1708                 retval = -ERESTARTSYS;
1709                 if (!signal_pending(current)) {
1710                         schedule();
1711                         goto repeat;
1712                 }
1713         }
1714
1715 end:
1716         current->state = TASK_RUNNING;
1717         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1718         if (infop) {
1719                 if (retval > 0)
1720                         retval = 0;
1721                 else {
1722                         /*
1723                          * For a WNOHANG return, clear out all the fields
1724                          * we would set so the user can easily tell the
1725                          * difference.
1726                          */
1727                         if (!retval)
1728                                 retval = put_user(0, &infop->si_signo);
1729                         if (!retval)
1730                                 retval = put_user(0, &infop->si_errno);
1731                         if (!retval)
1732                                 retval = put_user(0, &infop->si_code);
1733                         if (!retval)
1734                                 retval = put_user(0, &infop->si_pid);
1735                         if (!retval)
1736                                 retval = put_user(0, &infop->si_uid);
1737                         if (!retval)
1738                                 retval = put_user(0, &infop->si_status);
1739                 }
1740         }
1741         return retval;
1742 }
1743
1744 asmlinkage long sys_waitid(int which, pid_t upid,
1745                            struct siginfo __user *infop, int options,
1746                            struct rusage __user *ru)
1747 {
1748         struct pid *pid = NULL;
1749         enum pid_type type;
1750         long ret;
1751
1752         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1753                 return -EINVAL;
1754         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1755                 return -EINVAL;
1756
1757         switch (which) {
1758         case P_ALL:
1759                 type = PIDTYPE_MAX;
1760                 break;
1761         case P_PID:
1762                 type = PIDTYPE_PID;
1763                 if (upid <= 0)
1764                         return -EINVAL;
1765                 break;
1766         case P_PGID:
1767                 type = PIDTYPE_PGID;
1768                 if (upid <= 0)
1769                         return -EINVAL;
1770                 break;
1771         default:
1772                 return -EINVAL;
1773         }
1774
1775         if (type < PIDTYPE_MAX)
1776                 pid = find_get_pid(upid);
1777         ret = do_wait(type, pid, options, infop, NULL, ru);
1778         put_pid(pid);
1779
1780         /* avoid REGPARM breakage on x86: */
1781         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1782         return ret;
1783 }
1784
1785 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1786                           int options, struct rusage __user *ru)
1787 {
1788         struct pid *pid = NULL;
1789         enum pid_type type;
1790         long ret;
1791
1792         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1793                         __WNOTHREAD|__WCLONE|__WALL))
1794                 return -EINVAL;
1795
1796         if (upid == -1)
1797                 type = PIDTYPE_MAX;
1798         else if (upid < 0) {
1799                 type = PIDTYPE_PGID;
1800                 pid = find_get_pid(-upid);
1801         } else if (upid == 0) {
1802                 type = PIDTYPE_PGID;
1803                 pid = get_pid(task_pgrp(current));
1804         } else /* upid > 0 */ {
1805                 type = PIDTYPE_PID;
1806                 pid = find_get_pid(upid);
1807         }
1808
1809         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1810         put_pid(pid);
1811
1812         /* avoid REGPARM breakage on x86: */
1813         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1814         return ret;
1815 }
1816
1817 #ifdef __ARCH_WANT_SYS_WAITPID
1818
1819 /*
1820  * sys_waitpid() remains for compatibility. waitpid() should be
1821  * implemented by calling sys_wait4() from libc.a.
1822  */
1823 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1824 {
1825         return sys_wait4(pid, stat_addr, options, NULL);
1826 }
1827
1828 #endif