pids: improve get_task_pid() to fix the unsafe sys_wait4()->task_pgrp()
[linux-2.6.git] / kernel / exit.c
1 /*
2  *  linux/kernel/exit.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 #include <linux/mm.h>
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/mnt_namespace.h>
16 #include <linux/iocontext.h>
17 #include <linux/key.h>
18 #include <linux/security.h>
19 #include <linux/cpu.h>
20 #include <linux/acct.h>
21 #include <linux/tsacct_kern.h>
22 #include <linux/file.h>
23 #include <linux/fdtable.h>
24 #include <linux/binfmts.h>
25 #include <linux/nsproxy.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/ptrace.h>
28 #include <linux/profile.h>
29 #include <linux/mount.h>
30 #include <linux/proc_fs.h>
31 #include <linux/kthread.h>
32 #include <linux/mempolicy.h>
33 #include <linux/taskstats_kern.h>
34 #include <linux/delayacct.h>
35 #include <linux/freezer.h>
36 #include <linux/cgroup.h>
37 #include <linux/syscalls.h>
38 #include <linux/signal.h>
39 #include <linux/posix-timers.h>
40 #include <linux/cn_proc.h>
41 #include <linux/mutex.h>
42 #include <linux/futex.h>
43 #include <linux/pipe_fs_i.h>
44 #include <linux/audit.h> /* for audit_free() */
45 #include <linux/resource.h>
46 #include <linux/blkdev.h>
47 #include <linux/task_io_accounting_ops.h>
48 #include <linux/tracehook.h>
49 #include <linux/init_task.h>
50 #include <trace/sched.h>
51
52 #include <asm/uaccess.h>
53 #include <asm/unistd.h>
54 #include <asm/pgtable.h>
55 #include <asm/mmu_context.h>
56 #include "cred-internals.h"
57
58 DEFINE_TRACE(sched_process_free);
59 DEFINE_TRACE(sched_process_exit);
60 DEFINE_TRACE(sched_process_wait);
61
62 static void exit_mm(struct task_struct * tsk);
63
64 static void __unhash_process(struct task_struct *p)
65 {
66         nr_threads--;
67         detach_pid(p, PIDTYPE_PID);
68         if (thread_group_leader(p)) {
69                 detach_pid(p, PIDTYPE_PGID);
70                 detach_pid(p, PIDTYPE_SID);
71
72                 list_del_rcu(&p->tasks);
73                 __get_cpu_var(process_counts)--;
74         }
75         list_del_rcu(&p->thread_group);
76         list_del_init(&p->sibling);
77 }
78
79 /*
80  * This function expects the tasklist_lock write-locked.
81  */
82 static void __exit_signal(struct task_struct *tsk)
83 {
84         struct signal_struct *sig = tsk->signal;
85         struct sighand_struct *sighand;
86
87         BUG_ON(!sig);
88         BUG_ON(!atomic_read(&sig->count));
89
90         sighand = rcu_dereference(tsk->sighand);
91         spin_lock(&sighand->siglock);
92
93         posix_cpu_timers_exit(tsk);
94         if (atomic_dec_and_test(&sig->count))
95                 posix_cpu_timers_exit_group(tsk);
96         else {
97                 /*
98                  * If there is any task waiting for the group exit
99                  * then notify it:
100                  */
101                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
102                         wake_up_process(sig->group_exit_task);
103
104                 if (tsk == sig->curr_target)
105                         sig->curr_target = next_thread(tsk);
106                 /*
107                  * Accumulate here the counters for all threads but the
108                  * group leader as they die, so they can be added into
109                  * the process-wide totals when those are taken.
110                  * The group leader stays around as a zombie as long
111                  * as there are other threads.  When it gets reaped,
112                  * the exit.c code will add its counts into these totals.
113                  * We won't ever get here for the group leader, since it
114                  * will have been the last reference on the signal_struct.
115                  */
116                 sig->utime = cputime_add(sig->utime, task_utime(tsk));
117                 sig->stime = cputime_add(sig->stime, task_stime(tsk));
118                 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
119                 sig->min_flt += tsk->min_flt;
120                 sig->maj_flt += tsk->maj_flt;
121                 sig->nvcsw += tsk->nvcsw;
122                 sig->nivcsw += tsk->nivcsw;
123                 sig->inblock += task_io_get_inblock(tsk);
124                 sig->oublock += task_io_get_oublock(tsk);
125                 task_io_accounting_add(&sig->ioac, &tsk->ioac);
126                 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
127                 sig = NULL; /* Marker for below. */
128         }
129
130         __unhash_process(tsk);
131
132         /*
133          * Do this under ->siglock, we can race with another thread
134          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
135          */
136         flush_sigqueue(&tsk->pending);
137
138         tsk->signal = NULL;
139         tsk->sighand = NULL;
140         spin_unlock(&sighand->siglock);
141
142         __cleanup_sighand(sighand);
143         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
144         if (sig) {
145                 flush_sigqueue(&sig->shared_pending);
146                 taskstats_tgid_free(sig);
147                 /*
148                  * Make sure ->signal can't go away under rq->lock,
149                  * see account_group_exec_runtime().
150                  */
151                 task_rq_unlock_wait(tsk);
152                 __cleanup_signal(sig);
153         }
154 }
155
156 static void delayed_put_task_struct(struct rcu_head *rhp)
157 {
158         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
159
160         trace_sched_process_free(tsk);
161         put_task_struct(tsk);
162 }
163
164
165 void release_task(struct task_struct * p)
166 {
167         struct task_struct *leader;
168         int zap_leader;
169 repeat:
170         tracehook_prepare_release_task(p);
171         /* don't need to get the RCU readlock here - the process is dead and
172          * can't be modifying its own credentials */
173         atomic_dec(&__task_cred(p)->user->processes);
174
175         proc_flush_task(p);
176         write_lock_irq(&tasklist_lock);
177         tracehook_finish_release_task(p);
178         __exit_signal(p);
179
180         /*
181          * If we are the last non-leader member of the thread
182          * group, and the leader is zombie, then notify the
183          * group leader's parent process. (if it wants notification.)
184          */
185         zap_leader = 0;
186         leader = p->group_leader;
187         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
188                 BUG_ON(task_detached(leader));
189                 do_notify_parent(leader, leader->exit_signal);
190                 /*
191                  * If we were the last child thread and the leader has
192                  * exited already, and the leader's parent ignores SIGCHLD,
193                  * then we are the one who should release the leader.
194                  *
195                  * do_notify_parent() will have marked it self-reaping in
196                  * that case.
197                  */
198                 zap_leader = task_detached(leader);
199
200                 /*
201                  * This maintains the invariant that release_task()
202                  * only runs on a task in EXIT_DEAD, just for sanity.
203                  */
204                 if (zap_leader)
205                         leader->exit_state = EXIT_DEAD;
206         }
207
208         write_unlock_irq(&tasklist_lock);
209         release_thread(p);
210         call_rcu(&p->rcu, delayed_put_task_struct);
211
212         p = leader;
213         if (unlikely(zap_leader))
214                 goto repeat;
215 }
216
217 /*
218  * This checks not only the pgrp, but falls back on the pid if no
219  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
220  * without this...
221  *
222  * The caller must hold rcu lock or the tasklist lock.
223  */
224 struct pid *session_of_pgrp(struct pid *pgrp)
225 {
226         struct task_struct *p;
227         struct pid *sid = NULL;
228
229         p = pid_task(pgrp, PIDTYPE_PGID);
230         if (p == NULL)
231                 p = pid_task(pgrp, PIDTYPE_PID);
232         if (p != NULL)
233                 sid = task_session(p);
234
235         return sid;
236 }
237
238 /*
239  * Determine if a process group is "orphaned", according to the POSIX
240  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
241  * by terminal-generated stop signals.  Newly orphaned process groups are
242  * to receive a SIGHUP and a SIGCONT.
243  *
244  * "I ask you, have you ever known what it is to be an orphan?"
245  */
246 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
247 {
248         struct task_struct *p;
249
250         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
251                 if ((p == ignored_task) ||
252                     (p->exit_state && thread_group_empty(p)) ||
253                     is_global_init(p->real_parent))
254                         continue;
255
256                 if (task_pgrp(p->real_parent) != pgrp &&
257                     task_session(p->real_parent) == task_session(p))
258                         return 0;
259         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
260
261         return 1;
262 }
263
264 int is_current_pgrp_orphaned(void)
265 {
266         int retval;
267
268         read_lock(&tasklist_lock);
269         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
270         read_unlock(&tasklist_lock);
271
272         return retval;
273 }
274
275 static int has_stopped_jobs(struct pid *pgrp)
276 {
277         int retval = 0;
278         struct task_struct *p;
279
280         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
281                 if (!task_is_stopped(p))
282                         continue;
283                 retval = 1;
284                 break;
285         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
286         return retval;
287 }
288
289 /*
290  * Check to see if any process groups have become orphaned as
291  * a result of our exiting, and if they have any stopped jobs,
292  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
293  */
294 static void
295 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
296 {
297         struct pid *pgrp = task_pgrp(tsk);
298         struct task_struct *ignored_task = tsk;
299
300         if (!parent)
301                  /* exit: our father is in a different pgrp than
302                   * we are and we were the only connection outside.
303                   */
304                 parent = tsk->real_parent;
305         else
306                 /* reparent: our child is in a different pgrp than
307                  * we are, and it was the only connection outside.
308                  */
309                 ignored_task = NULL;
310
311         if (task_pgrp(parent) != pgrp &&
312             task_session(parent) == task_session(tsk) &&
313             will_become_orphaned_pgrp(pgrp, ignored_task) &&
314             has_stopped_jobs(pgrp)) {
315                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
316                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
317         }
318 }
319
320 /**
321  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
322  *
323  * If a kernel thread is launched as a result of a system call, or if
324  * it ever exits, it should generally reparent itself to kthreadd so it
325  * isn't in the way of other processes and is correctly cleaned up on exit.
326  *
327  * The various task state such as scheduling policy and priority may have
328  * been inherited from a user process, so we reset them to sane values here.
329  *
330  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
331  */
332 static void reparent_to_kthreadd(void)
333 {
334         write_lock_irq(&tasklist_lock);
335
336         ptrace_unlink(current);
337         /* Reparent to init */
338         current->real_parent = current->parent = kthreadd_task;
339         list_move_tail(&current->sibling, &current->real_parent->children);
340
341         /* Set the exit signal to SIGCHLD so we signal init on exit */
342         current->exit_signal = SIGCHLD;
343
344         if (task_nice(current) < 0)
345                 set_user_nice(current, 0);
346         /* cpus_allowed? */
347         /* rt_priority? */
348         /* signals? */
349         memcpy(current->signal->rlim, init_task.signal->rlim,
350                sizeof(current->signal->rlim));
351
352         atomic_inc(&init_cred.usage);
353         commit_creds(&init_cred);
354         write_unlock_irq(&tasklist_lock);
355 }
356
357 void __set_special_pids(struct pid *pid)
358 {
359         struct task_struct *curr = current->group_leader;
360         pid_t nr = pid_nr(pid);
361
362         if (task_session(curr) != pid) {
363                 change_pid(curr, PIDTYPE_SID, pid);
364                 set_task_session(curr, nr);
365         }
366         if (task_pgrp(curr) != pid) {
367                 change_pid(curr, PIDTYPE_PGID, pid);
368                 set_task_pgrp(curr, nr);
369         }
370 }
371
372 static void set_special_pids(struct pid *pid)
373 {
374         write_lock_irq(&tasklist_lock);
375         __set_special_pids(pid);
376         write_unlock_irq(&tasklist_lock);
377 }
378
379 /*
380  * Let kernel threads use this to say that they
381  * allow a certain signal (since daemonize() will
382  * have disabled all of them by default).
383  */
384 int allow_signal(int sig)
385 {
386         if (!valid_signal(sig) || sig < 1)
387                 return -EINVAL;
388
389         spin_lock_irq(&current->sighand->siglock);
390         sigdelset(&current->blocked, sig);
391         if (!current->mm) {
392                 /* Kernel threads handle their own signals.
393                    Let the signal code know it'll be handled, so
394                    that they don't get converted to SIGKILL or
395                    just silently dropped */
396                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
397         }
398         recalc_sigpending();
399         spin_unlock_irq(&current->sighand->siglock);
400         return 0;
401 }
402
403 EXPORT_SYMBOL(allow_signal);
404
405 int disallow_signal(int sig)
406 {
407         if (!valid_signal(sig) || sig < 1)
408                 return -EINVAL;
409
410         spin_lock_irq(&current->sighand->siglock);
411         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
412         recalc_sigpending();
413         spin_unlock_irq(&current->sighand->siglock);
414         return 0;
415 }
416
417 EXPORT_SYMBOL(disallow_signal);
418
419 /*
420  *      Put all the gunge required to become a kernel thread without
421  *      attached user resources in one place where it belongs.
422  */
423
424 void daemonize(const char *name, ...)
425 {
426         va_list args;
427         struct fs_struct *fs;
428         sigset_t blocked;
429
430         va_start(args, name);
431         vsnprintf(current->comm, sizeof(current->comm), name, args);
432         va_end(args);
433
434         /*
435          * If we were started as result of loading a module, close all of the
436          * user space pages.  We don't need them, and if we didn't close them
437          * they would be locked into memory.
438          */
439         exit_mm(current);
440         /*
441          * We don't want to have TIF_FREEZE set if the system-wide hibernation
442          * or suspend transition begins right now.
443          */
444         current->flags |= (PF_NOFREEZE | PF_KTHREAD);
445
446         if (current->nsproxy != &init_nsproxy) {
447                 get_nsproxy(&init_nsproxy);
448                 switch_task_namespaces(current, &init_nsproxy);
449         }
450         set_special_pids(&init_struct_pid);
451         proc_clear_tty(current);
452
453         /* Block and flush all signals */
454         sigfillset(&blocked);
455         sigprocmask(SIG_BLOCK, &blocked, NULL);
456         flush_signals(current);
457
458         /* Become as one with the init task */
459
460         exit_fs(current);       /* current->fs->count--; */
461         fs = init_task.fs;
462         current->fs = fs;
463         atomic_inc(&fs->count);
464
465         exit_files(current);
466         current->files = init_task.files;
467         atomic_inc(&current->files->count);
468
469         reparent_to_kthreadd();
470 }
471
472 EXPORT_SYMBOL(daemonize);
473
474 static void close_files(struct files_struct * files)
475 {
476         int i, j;
477         struct fdtable *fdt;
478
479         j = 0;
480
481         /*
482          * It is safe to dereference the fd table without RCU or
483          * ->file_lock because this is the last reference to the
484          * files structure.
485          */
486         fdt = files_fdtable(files);
487         for (;;) {
488                 unsigned long set;
489                 i = j * __NFDBITS;
490                 if (i >= fdt->max_fds)
491                         break;
492                 set = fdt->open_fds->fds_bits[j++];
493                 while (set) {
494                         if (set & 1) {
495                                 struct file * file = xchg(&fdt->fd[i], NULL);
496                                 if (file) {
497                                         filp_close(file, files);
498                                         cond_resched();
499                                 }
500                         }
501                         i++;
502                         set >>= 1;
503                 }
504         }
505 }
506
507 struct files_struct *get_files_struct(struct task_struct *task)
508 {
509         struct files_struct *files;
510
511         task_lock(task);
512         files = task->files;
513         if (files)
514                 atomic_inc(&files->count);
515         task_unlock(task);
516
517         return files;
518 }
519
520 void put_files_struct(struct files_struct *files)
521 {
522         struct fdtable *fdt;
523
524         if (atomic_dec_and_test(&files->count)) {
525                 close_files(files);
526                 /*
527                  * Free the fd and fdset arrays if we expanded them.
528                  * If the fdtable was embedded, pass files for freeing
529                  * at the end of the RCU grace period. Otherwise,
530                  * you can free files immediately.
531                  */
532                 fdt = files_fdtable(files);
533                 if (fdt != &files->fdtab)
534                         kmem_cache_free(files_cachep, files);
535                 free_fdtable(fdt);
536         }
537 }
538
539 void reset_files_struct(struct files_struct *files)
540 {
541         struct task_struct *tsk = current;
542         struct files_struct *old;
543
544         old = tsk->files;
545         task_lock(tsk);
546         tsk->files = files;
547         task_unlock(tsk);
548         put_files_struct(old);
549 }
550
551 void exit_files(struct task_struct *tsk)
552 {
553         struct files_struct * files = tsk->files;
554
555         if (files) {
556                 task_lock(tsk);
557                 tsk->files = NULL;
558                 task_unlock(tsk);
559                 put_files_struct(files);
560         }
561 }
562
563 void put_fs_struct(struct fs_struct *fs)
564 {
565         /* No need to hold fs->lock if we are killing it */
566         if (atomic_dec_and_test(&fs->count)) {
567                 path_put(&fs->root);
568                 path_put(&fs->pwd);
569                 kmem_cache_free(fs_cachep, fs);
570         }
571 }
572
573 void exit_fs(struct task_struct *tsk)
574 {
575         struct fs_struct * fs = tsk->fs;
576
577         if (fs) {
578                 task_lock(tsk);
579                 tsk->fs = NULL;
580                 task_unlock(tsk);
581                 put_fs_struct(fs);
582         }
583 }
584
585 EXPORT_SYMBOL_GPL(exit_fs);
586
587 #ifdef CONFIG_MM_OWNER
588 /*
589  * Task p is exiting and it owned mm, lets find a new owner for it
590  */
591 static inline int
592 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
593 {
594         /*
595          * If there are other users of the mm and the owner (us) is exiting
596          * we need to find a new owner to take on the responsibility.
597          */
598         if (atomic_read(&mm->mm_users) <= 1)
599                 return 0;
600         if (mm->owner != p)
601                 return 0;
602         return 1;
603 }
604
605 void mm_update_next_owner(struct mm_struct *mm)
606 {
607         struct task_struct *c, *g, *p = current;
608
609 retry:
610         if (!mm_need_new_owner(mm, p))
611                 return;
612
613         read_lock(&tasklist_lock);
614         /*
615          * Search in the children
616          */
617         list_for_each_entry(c, &p->children, sibling) {
618                 if (c->mm == mm)
619                         goto assign_new_owner;
620         }
621
622         /*
623          * Search in the siblings
624          */
625         list_for_each_entry(c, &p->parent->children, sibling) {
626                 if (c->mm == mm)
627                         goto assign_new_owner;
628         }
629
630         /*
631          * Search through everything else. We should not get
632          * here often
633          */
634         do_each_thread(g, c) {
635                 if (c->mm == mm)
636                         goto assign_new_owner;
637         } while_each_thread(g, c);
638
639         read_unlock(&tasklist_lock);
640         /*
641          * We found no owner yet mm_users > 1: this implies that we are
642          * most likely racing with swapoff (try_to_unuse()) or /proc or
643          * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
644          */
645         mm->owner = NULL;
646         return;
647
648 assign_new_owner:
649         BUG_ON(c == p);
650         get_task_struct(c);
651         /*
652          * The task_lock protects c->mm from changing.
653          * We always want mm->owner->mm == mm
654          */
655         task_lock(c);
656         /*
657          * Delay read_unlock() till we have the task_lock()
658          * to ensure that c does not slip away underneath us
659          */
660         read_unlock(&tasklist_lock);
661         if (c->mm != mm) {
662                 task_unlock(c);
663                 put_task_struct(c);
664                 goto retry;
665         }
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  * When we die, we re-parent all our children.
731  * Try to give them to another thread in our thread
732  * group, and if no such member exists, give it to
733  * the child reaper process (ie "init") in our pid
734  * space.
735  */
736 static struct task_struct *find_new_reaper(struct task_struct *father)
737 {
738         struct pid_namespace *pid_ns = task_active_pid_ns(father);
739         struct task_struct *thread;
740
741         thread = father;
742         while_each_thread(father, thread) {
743                 if (thread->flags & PF_EXITING)
744                         continue;
745                 if (unlikely(pid_ns->child_reaper == father))
746                         pid_ns->child_reaper = thread;
747                 return thread;
748         }
749
750         if (unlikely(pid_ns->child_reaper == father)) {
751                 write_unlock_irq(&tasklist_lock);
752                 if (unlikely(pid_ns == &init_pid_ns))
753                         panic("Attempted to kill init!");
754
755                 zap_pid_ns_processes(pid_ns);
756                 write_lock_irq(&tasklist_lock);
757                 /*
758                  * We can not clear ->child_reaper or leave it alone.
759                  * There may by stealth EXIT_DEAD tasks on ->children,
760                  * forget_original_parent() must move them somewhere.
761                  */
762                 pid_ns->child_reaper = init_pid_ns.child_reaper;
763         }
764
765         return pid_ns->child_reaper;
766 }
767
768 /*
769 * Any that need to be release_task'd are put on the @dead list.
770  */
771 static void reparent_thread(struct task_struct *father, struct task_struct *p,
772                                 struct list_head *dead)
773 {
774         if (p->pdeath_signal)
775                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
776
777         list_move_tail(&p->sibling, &p->real_parent->children);
778
779         if (task_detached(p))
780                 return;
781         /*
782          * If this is a threaded reparent there is no need to
783          * notify anyone anything has happened.
784          */
785         if (same_thread_group(p->real_parent, father))
786                 return;
787
788         /* We don't want people slaying init.  */
789         p->exit_signal = SIGCHLD;
790
791         /* If it has exited notify the new parent about this child's death. */
792         if (!p->ptrace &&
793             p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
794                 do_notify_parent(p, p->exit_signal);
795                 if (task_detached(p)) {
796                         p->exit_state = EXIT_DEAD;
797                         list_move_tail(&p->sibling, dead);
798                 }
799         }
800
801         kill_orphaned_pgrp(p, father);
802 }
803
804 static void forget_original_parent(struct task_struct *father)
805 {
806         struct task_struct *p, *n, *reaper;
807         LIST_HEAD(dead_children);
808
809         exit_ptrace(father);
810
811         write_lock_irq(&tasklist_lock);
812         reaper = find_new_reaper(father);
813
814         list_for_each_entry_safe(p, n, &father->children, sibling) {
815                 p->real_parent = reaper;
816                 if (p->parent == father) {
817                         BUG_ON(p->ptrace);
818                         p->parent = p->real_parent;
819                 }
820                 reparent_thread(father, p, &dead_children);
821         }
822         write_unlock_irq(&tasklist_lock);
823
824         BUG_ON(!list_empty(&father->children));
825
826         list_for_each_entry_safe(p, n, &dead_children, sibling) {
827                 list_del_init(&p->sibling);
828                 release_task(p);
829         }
830 }
831
832 /*
833  * Send signals to all our closest relatives so that they know
834  * to properly mourn us..
835  */
836 static void exit_notify(struct task_struct *tsk, int group_dead)
837 {
838         int signal;
839         void *cookie;
840
841         /*
842          * This does two things:
843          *
844          * A.  Make init inherit all the child processes
845          * B.  Check to see if any process groups have become orphaned
846          *      as a result of our exiting, and if they have any stopped
847          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
848          */
849         forget_original_parent(tsk);
850         exit_task_namespaces(tsk);
851
852         write_lock_irq(&tasklist_lock);
853         if (group_dead)
854                 kill_orphaned_pgrp(tsk->group_leader, NULL);
855
856         /* Let father know we died
857          *
858          * Thread signals are configurable, but you aren't going to use
859          * that to send signals to arbitary processes.
860          * That stops right now.
861          *
862          * If the parent exec id doesn't match the exec id we saved
863          * when we started then we know the parent has changed security
864          * domain.
865          *
866          * If our self_exec id doesn't match our parent_exec_id then
867          * we have changed execution domain as these two values started
868          * the same after a fork.
869          */
870         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
871             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
872              tsk->self_exec_id != tsk->parent_exec_id) &&
873             !capable(CAP_KILL))
874                 tsk->exit_signal = SIGCHLD;
875
876         signal = tracehook_notify_death(tsk, &cookie, group_dead);
877         if (signal >= 0)
878                 signal = do_notify_parent(tsk, signal);
879
880         tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
881
882         /* mt-exec, de_thread() is waiting for us */
883         if (thread_group_leader(tsk) &&
884             tsk->signal->group_exit_task &&
885             tsk->signal->notify_count < 0)
886                 wake_up_process(tsk->signal->group_exit_task);
887
888         write_unlock_irq(&tasklist_lock);
889
890         tracehook_report_death(tsk, signal, cookie, group_dead);
891
892         /* If the process is dead, release it - nobody will wait for it */
893         if (signal == DEATH_REAP)
894                 release_task(tsk);
895 }
896
897 #ifdef CONFIG_DEBUG_STACK_USAGE
898 static void check_stack_usage(void)
899 {
900         static DEFINE_SPINLOCK(low_water_lock);
901         static int lowest_to_date = THREAD_SIZE;
902         unsigned long free;
903
904         free = stack_not_used(current);
905
906         if (free >= lowest_to_date)
907                 return;
908
909         spin_lock(&low_water_lock);
910         if (free < lowest_to_date) {
911                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
912                                 "left\n",
913                                 current->comm, free);
914                 lowest_to_date = free;
915         }
916         spin_unlock(&low_water_lock);
917 }
918 #else
919 static inline void check_stack_usage(void) {}
920 #endif
921
922 NORET_TYPE void do_exit(long code)
923 {
924         struct task_struct *tsk = current;
925         int group_dead;
926
927         profile_task_exit(tsk);
928
929         WARN_ON(atomic_read(&tsk->fs_excl));
930
931         if (unlikely(in_interrupt()))
932                 panic("Aiee, killing interrupt handler!");
933         if (unlikely(!tsk->pid))
934                 panic("Attempted to kill the idle task!");
935
936         tracehook_report_exit(&code);
937
938         /*
939          * We're taking recursive faults here in do_exit. Safest is to just
940          * leave this task alone and wait for reboot.
941          */
942         if (unlikely(tsk->flags & PF_EXITING)) {
943                 printk(KERN_ALERT
944                         "Fixing recursive fault but reboot is needed!\n");
945                 /*
946                  * We can do this unlocked here. The futex code uses
947                  * this flag just to verify whether the pi state
948                  * cleanup has been done or not. In the worst case it
949                  * loops once more. We pretend that the cleanup was
950                  * done as there is no way to return. Either the
951                  * OWNER_DIED bit is set by now or we push the blocked
952                  * task into the wait for ever nirwana as well.
953                  */
954                 tsk->flags |= PF_EXITPIDONE;
955                 set_current_state(TASK_UNINTERRUPTIBLE);
956                 schedule();
957         }
958
959         exit_signals(tsk);  /* sets PF_EXITING */
960         /*
961          * tsk->flags are checked in the futex code to protect against
962          * an exiting task cleaning up the robust pi futexes.
963          */
964         smp_mb();
965         spin_unlock_wait(&tsk->pi_lock);
966
967         if (unlikely(in_atomic()))
968                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
969                                 current->comm, task_pid_nr(current),
970                                 preempt_count());
971
972         acct_update_integrals(tsk);
973
974         group_dead = atomic_dec_and_test(&tsk->signal->live);
975         if (group_dead) {
976                 hrtimer_cancel(&tsk->signal->real_timer);
977                 exit_itimers(tsk->signal);
978         }
979         acct_collect(code, group_dead);
980         if (group_dead)
981                 tty_audit_exit();
982         if (unlikely(tsk->audit_context))
983                 audit_free(tsk);
984
985         tsk->exit_code = code;
986         taskstats_exit(tsk, group_dead);
987
988         exit_mm(tsk);
989
990         if (group_dead)
991                 acct_process();
992         trace_sched_process_exit(tsk);
993
994         exit_sem(tsk);
995         exit_files(tsk);
996         exit_fs(tsk);
997         check_stack_usage();
998         exit_thread();
999         cgroup_exit(tsk, 1);
1000
1001         if (group_dead && tsk->signal->leader)
1002                 disassociate_ctty(1);
1003
1004         module_put(task_thread_info(tsk)->exec_domain->module);
1005         if (tsk->binfmt)
1006                 module_put(tsk->binfmt->module);
1007
1008         proc_exit_connector(tsk);
1009         exit_notify(tsk, group_dead);
1010 #ifdef CONFIG_NUMA
1011         mpol_put(tsk->mempolicy);
1012         tsk->mempolicy = NULL;
1013 #endif
1014 #ifdef CONFIG_FUTEX
1015         /*
1016          * This must happen late, after the PID is not
1017          * hashed anymore:
1018          */
1019         if (unlikely(!list_empty(&tsk->pi_state_list)))
1020                 exit_pi_state_list(tsk);
1021         if (unlikely(current->pi_state_cache))
1022                 kfree(current->pi_state_cache);
1023 #endif
1024         /*
1025          * Make sure we are holding no locks:
1026          */
1027         debug_check_no_locks_held(tsk);
1028         /*
1029          * We can do this unlocked here. The futex code uses this flag
1030          * just to verify whether the pi state cleanup has been done
1031          * or not. In the worst case it loops once more.
1032          */
1033         tsk->flags |= PF_EXITPIDONE;
1034
1035         if (tsk->io_context)
1036                 exit_io_context();
1037
1038         if (tsk->splice_pipe)
1039                 __free_pipe_info(tsk->splice_pipe);
1040
1041         preempt_disable();
1042         /* causes final put_task_struct in finish_task_switch(). */
1043         tsk->state = TASK_DEAD;
1044         schedule();
1045         BUG();
1046         /* Avoid "noreturn function does return".  */
1047         for (;;)
1048                 cpu_relax();    /* For when BUG is null */
1049 }
1050
1051 EXPORT_SYMBOL_GPL(do_exit);
1052
1053 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1054 {
1055         if (comp)
1056                 complete(comp);
1057
1058         do_exit(code);
1059 }
1060
1061 EXPORT_SYMBOL(complete_and_exit);
1062
1063 SYSCALL_DEFINE1(exit, int, error_code)
1064 {
1065         do_exit((error_code&0xff)<<8);
1066 }
1067
1068 /*
1069  * Take down every thread in the group.  This is called by fatal signals
1070  * as well as by sys_exit_group (below).
1071  */
1072 NORET_TYPE void
1073 do_group_exit(int exit_code)
1074 {
1075         struct signal_struct *sig = current->signal;
1076
1077         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1078
1079         if (signal_group_exit(sig))
1080                 exit_code = sig->group_exit_code;
1081         else if (!thread_group_empty(current)) {
1082                 struct sighand_struct *const sighand = current->sighand;
1083                 spin_lock_irq(&sighand->siglock);
1084                 if (signal_group_exit(sig))
1085                         /* Another thread got here before we took the lock.  */
1086                         exit_code = sig->group_exit_code;
1087                 else {
1088                         sig->group_exit_code = exit_code;
1089                         sig->flags = SIGNAL_GROUP_EXIT;
1090                         zap_other_threads(current);
1091                 }
1092                 spin_unlock_irq(&sighand->siglock);
1093         }
1094
1095         do_exit(exit_code);
1096         /* NOTREACHED */
1097 }
1098
1099 /*
1100  * this kills every thread in the thread group. Note that any externally
1101  * wait4()-ing process will get the correct exit code - even if this
1102  * thread is not the thread group leader.
1103  */
1104 SYSCALL_DEFINE1(exit_group, int, error_code)
1105 {
1106         do_group_exit((error_code & 0xff) << 8);
1107         /* NOTREACHED */
1108         return 0;
1109 }
1110
1111 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1112 {
1113         struct pid *pid = NULL;
1114         if (type == PIDTYPE_PID)
1115                 pid = task->pids[type].pid;
1116         else if (type < PIDTYPE_MAX)
1117                 pid = task->group_leader->pids[type].pid;
1118         return pid;
1119 }
1120
1121 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1122                           struct task_struct *p)
1123 {
1124         int err;
1125
1126         if (type < PIDTYPE_MAX) {
1127                 if (task_pid_type(p, type) != pid)
1128                         return 0;
1129         }
1130
1131         /* Wait for all children (clone and not) if __WALL is set;
1132          * otherwise, wait for clone children *only* if __WCLONE is
1133          * set; otherwise, wait for non-clone children *only*.  (Note:
1134          * A "clone" child here is one that reports to its parent
1135          * using a signal other than SIGCHLD.) */
1136         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1137             && !(options & __WALL))
1138                 return 0;
1139
1140         err = security_task_wait(p);
1141         if (err)
1142                 return err;
1143
1144         return 1;
1145 }
1146
1147 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1148                                int why, int status,
1149                                struct siginfo __user *infop,
1150                                struct rusage __user *rusagep)
1151 {
1152         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1153
1154         put_task_struct(p);
1155         if (!retval)
1156                 retval = put_user(SIGCHLD, &infop->si_signo);
1157         if (!retval)
1158                 retval = put_user(0, &infop->si_errno);
1159         if (!retval)
1160                 retval = put_user((short)why, &infop->si_code);
1161         if (!retval)
1162                 retval = put_user(pid, &infop->si_pid);
1163         if (!retval)
1164                 retval = put_user(uid, &infop->si_uid);
1165         if (!retval)
1166                 retval = put_user(status, &infop->si_status);
1167         if (!retval)
1168                 retval = pid;
1169         return retval;
1170 }
1171
1172 /*
1173  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1174  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1175  * the lock and this task is uninteresting.  If we return nonzero, we have
1176  * released the lock and the system call should return.
1177  */
1178 static int wait_task_zombie(struct task_struct *p, int options,
1179                             struct siginfo __user *infop,
1180                             int __user *stat_addr, struct rusage __user *ru)
1181 {
1182         unsigned long state;
1183         int retval, status, traced;
1184         pid_t pid = task_pid_vnr(p);
1185         uid_t uid = __task_cred(p)->uid;
1186
1187         if (!likely(options & WEXITED))
1188                 return 0;
1189
1190         if (unlikely(options & WNOWAIT)) {
1191                 int exit_code = p->exit_code;
1192                 int why, status;
1193
1194                 get_task_struct(p);
1195                 read_unlock(&tasklist_lock);
1196                 if ((exit_code & 0x7f) == 0) {
1197                         why = CLD_EXITED;
1198                         status = exit_code >> 8;
1199                 } else {
1200                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1201                         status = exit_code & 0x7f;
1202                 }
1203                 return wait_noreap_copyout(p, pid, uid, why,
1204                                            status, infop, ru);
1205         }
1206
1207         /*
1208          * Try to move the task's state to DEAD
1209          * only one thread is allowed to do this:
1210          */
1211         state = xchg(&p->exit_state, EXIT_DEAD);
1212         if (state != EXIT_ZOMBIE) {
1213                 BUG_ON(state != EXIT_DEAD);
1214                 return 0;
1215         }
1216
1217         traced = ptrace_reparented(p);
1218
1219         if (likely(!traced)) {
1220                 struct signal_struct *psig;
1221                 struct signal_struct *sig;
1222                 struct task_cputime cputime;
1223
1224                 /*
1225                  * The resource counters for the group leader are in its
1226                  * own task_struct.  Those for dead threads in the group
1227                  * are in its signal_struct, as are those for the child
1228                  * processes it has previously reaped.  All these
1229                  * accumulate in the parent's signal_struct c* fields.
1230                  *
1231                  * We don't bother to take a lock here to protect these
1232                  * p->signal fields, because they are only touched by
1233                  * __exit_signal, which runs with tasklist_lock
1234                  * write-locked anyway, and so is excluded here.  We do
1235                  * need to protect the access to p->parent->signal fields,
1236                  * as other threads in the parent group can be right
1237                  * here reaping other children at the same time.
1238                  *
1239                  * We use thread_group_cputime() to get times for the thread
1240                  * group, which consolidates times for all threads in the
1241                  * group including the group leader.
1242                  */
1243                 thread_group_cputime(p, &cputime);
1244                 spin_lock_irq(&p->parent->sighand->siglock);
1245                 psig = p->parent->signal;
1246                 sig = p->signal;
1247                 psig->cutime =
1248                         cputime_add(psig->cutime,
1249                         cputime_add(cputime.utime,
1250                                     sig->cutime));
1251                 psig->cstime =
1252                         cputime_add(psig->cstime,
1253                         cputime_add(cputime.stime,
1254                                     sig->cstime));
1255                 psig->cgtime =
1256                         cputime_add(psig->cgtime,
1257                         cputime_add(p->gtime,
1258                         cputime_add(sig->gtime,
1259                                     sig->cgtime)));
1260                 psig->cmin_flt +=
1261                         p->min_flt + sig->min_flt + sig->cmin_flt;
1262                 psig->cmaj_flt +=
1263                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1264                 psig->cnvcsw +=
1265                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1266                 psig->cnivcsw +=
1267                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1268                 psig->cinblock +=
1269                         task_io_get_inblock(p) +
1270                         sig->inblock + sig->cinblock;
1271                 psig->coublock +=
1272                         task_io_get_oublock(p) +
1273                         sig->oublock + sig->coublock;
1274                 task_io_accounting_add(&psig->ioac, &p->ioac);
1275                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1276                 spin_unlock_irq(&p->parent->sighand->siglock);
1277         }
1278
1279         /*
1280          * Now we are sure this task is interesting, and no other
1281          * thread can reap it because we set its state to EXIT_DEAD.
1282          */
1283         read_unlock(&tasklist_lock);
1284
1285         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1286         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1287                 ? p->signal->group_exit_code : p->exit_code;
1288         if (!retval && stat_addr)
1289                 retval = put_user(status, stat_addr);
1290         if (!retval && infop)
1291                 retval = put_user(SIGCHLD, &infop->si_signo);
1292         if (!retval && infop)
1293                 retval = put_user(0, &infop->si_errno);
1294         if (!retval && infop) {
1295                 int why;
1296
1297                 if ((status & 0x7f) == 0) {
1298                         why = CLD_EXITED;
1299                         status >>= 8;
1300                 } else {
1301                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1302                         status &= 0x7f;
1303                 }
1304                 retval = put_user((short)why, &infop->si_code);
1305                 if (!retval)
1306                         retval = put_user(status, &infop->si_status);
1307         }
1308         if (!retval && infop)
1309                 retval = put_user(pid, &infop->si_pid);
1310         if (!retval && infop)
1311                 retval = put_user(uid, &infop->si_uid);
1312         if (!retval)
1313                 retval = pid;
1314
1315         if (traced) {
1316                 write_lock_irq(&tasklist_lock);
1317                 /* We dropped tasklist, ptracer could die and untrace */
1318                 ptrace_unlink(p);
1319                 /*
1320                  * If this is not a detached task, notify the parent.
1321                  * If it's still not detached after that, don't release
1322                  * it now.
1323                  */
1324                 if (!task_detached(p)) {
1325                         do_notify_parent(p, p->exit_signal);
1326                         if (!task_detached(p)) {
1327                                 p->exit_state = EXIT_ZOMBIE;
1328                                 p = NULL;
1329                         }
1330                 }
1331                 write_unlock_irq(&tasklist_lock);
1332         }
1333         if (p != NULL)
1334                 release_task(p);
1335
1336         return retval;
1337 }
1338
1339 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1340 {
1341         if (ptrace) {
1342                 if (task_is_stopped_or_traced(p))
1343                         return &p->exit_code;
1344         } else {
1345                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1346                         return &p->signal->group_exit_code;
1347         }
1348         return NULL;
1349 }
1350
1351 /*
1352  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1353  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1354  * the lock and this task is uninteresting.  If we return nonzero, we have
1355  * released the lock and the system call should return.
1356  */
1357 static int wait_task_stopped(int ptrace, struct task_struct *p,
1358                              int options, struct siginfo __user *infop,
1359                              int __user *stat_addr, struct rusage __user *ru)
1360 {
1361         int retval, exit_code, *p_code, why;
1362         uid_t uid = 0; /* unneeded, required by compiler */
1363         pid_t pid;
1364
1365         if (!(options & WUNTRACED))
1366                 return 0;
1367
1368         exit_code = 0;
1369         spin_lock_irq(&p->sighand->siglock);
1370
1371         p_code = task_stopped_code(p, ptrace);
1372         if (unlikely(!p_code))
1373                 goto unlock_sig;
1374
1375         exit_code = *p_code;
1376         if (!exit_code)
1377                 goto unlock_sig;
1378
1379         if (!unlikely(options & WNOWAIT))
1380                 *p_code = 0;
1381
1382         /* don't need the RCU readlock here as we're holding a spinlock */
1383         uid = __task_cred(p)->uid;
1384 unlock_sig:
1385         spin_unlock_irq(&p->sighand->siglock);
1386         if (!exit_code)
1387                 return 0;
1388
1389         /*
1390          * Now we are pretty sure this task is interesting.
1391          * Make sure it doesn't get reaped out from under us while we
1392          * give up the lock and then examine it below.  We don't want to
1393          * keep holding onto the tasklist_lock while we call getrusage and
1394          * possibly take page faults for user memory.
1395          */
1396         get_task_struct(p);
1397         pid = task_pid_vnr(p);
1398         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1399         read_unlock(&tasklist_lock);
1400
1401         if (unlikely(options & WNOWAIT))
1402                 return wait_noreap_copyout(p, pid, uid,
1403                                            why, exit_code,
1404                                            infop, ru);
1405
1406         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1407         if (!retval && stat_addr)
1408                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1409         if (!retval && infop)
1410                 retval = put_user(SIGCHLD, &infop->si_signo);
1411         if (!retval && infop)
1412                 retval = put_user(0, &infop->si_errno);
1413         if (!retval && infop)
1414                 retval = put_user((short)why, &infop->si_code);
1415         if (!retval && infop)
1416                 retval = put_user(exit_code, &infop->si_status);
1417         if (!retval && infop)
1418                 retval = put_user(pid, &infop->si_pid);
1419         if (!retval && infop)
1420                 retval = put_user(uid, &infop->si_uid);
1421         if (!retval)
1422                 retval = pid;
1423         put_task_struct(p);
1424
1425         BUG_ON(!retval);
1426         return retval;
1427 }
1428
1429 /*
1430  * Handle do_wait work for one task in a live, non-stopped state.
1431  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1432  * the lock and this task is uninteresting.  If we return nonzero, we have
1433  * released the lock and the system call should return.
1434  */
1435 static int wait_task_continued(struct task_struct *p, int options,
1436                                struct siginfo __user *infop,
1437                                int __user *stat_addr, struct rusage __user *ru)
1438 {
1439         int retval;
1440         pid_t pid;
1441         uid_t uid;
1442
1443         if (!unlikely(options & WCONTINUED))
1444                 return 0;
1445
1446         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1447                 return 0;
1448
1449         spin_lock_irq(&p->sighand->siglock);
1450         /* Re-check with the lock held.  */
1451         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1452                 spin_unlock_irq(&p->sighand->siglock);
1453                 return 0;
1454         }
1455         if (!unlikely(options & WNOWAIT))
1456                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1457         uid = __task_cred(p)->uid;
1458         spin_unlock_irq(&p->sighand->siglock);
1459
1460         pid = task_pid_vnr(p);
1461         get_task_struct(p);
1462         read_unlock(&tasklist_lock);
1463
1464         if (!infop) {
1465                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1466                 put_task_struct(p);
1467                 if (!retval && stat_addr)
1468                         retval = put_user(0xffff, stat_addr);
1469                 if (!retval)
1470                         retval = pid;
1471         } else {
1472                 retval = wait_noreap_copyout(p, pid, uid,
1473                                              CLD_CONTINUED, SIGCONT,
1474                                              infop, ru);
1475                 BUG_ON(retval == 0);
1476         }
1477
1478         return retval;
1479 }
1480
1481 /*
1482  * Consider @p for a wait by @parent.
1483  *
1484  * -ECHILD should be in *@notask_error before the first call.
1485  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1486  * Returns zero if the search for a child should continue;
1487  * then *@notask_error is 0 if @p is an eligible child,
1488  * or another error from security_task_wait(), or still -ECHILD.
1489  */
1490 static int wait_consider_task(struct task_struct *parent, int ptrace,
1491                               struct task_struct *p, int *notask_error,
1492                               enum pid_type type, struct pid *pid, int options,
1493                               struct siginfo __user *infop,
1494                               int __user *stat_addr, struct rusage __user *ru)
1495 {
1496         int ret = eligible_child(type, pid, options, p);
1497         if (!ret)
1498                 return ret;
1499
1500         if (unlikely(ret < 0)) {
1501                 /*
1502                  * If we have not yet seen any eligible child,
1503                  * then let this error code replace -ECHILD.
1504                  * A permission error will give the user a clue
1505                  * to look for security policy problems, rather
1506                  * than for mysterious wait bugs.
1507                  */
1508                 if (*notask_error)
1509                         *notask_error = ret;
1510         }
1511
1512         if (likely(!ptrace) && unlikely(p->ptrace)) {
1513                 /*
1514                  * This child is hidden by ptrace.
1515                  * We aren't allowed to see it now, but eventually we will.
1516                  */
1517                 *notask_error = 0;
1518                 return 0;
1519         }
1520
1521         if (p->exit_state == EXIT_DEAD)
1522                 return 0;
1523
1524         /*
1525          * We don't reap group leaders with subthreads.
1526          */
1527         if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1528                 return wait_task_zombie(p, options, infop, stat_addr, ru);
1529
1530         /*
1531          * It's stopped or running now, so it might
1532          * later continue, exit, or stop again.
1533          */
1534         *notask_error = 0;
1535
1536         if (task_stopped_code(p, ptrace))
1537                 return wait_task_stopped(ptrace, p, options,
1538                                          infop, stat_addr, ru);
1539
1540         return wait_task_continued(p, options, infop, stat_addr, ru);
1541 }
1542
1543 /*
1544  * Do the work of do_wait() for one thread in the group, @tsk.
1545  *
1546  * -ECHILD should be in *@notask_error before the first call.
1547  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1548  * Returns zero if the search for a child should continue; then
1549  * *@notask_error is 0 if there were any eligible children,
1550  * or another error from security_task_wait(), or still -ECHILD.
1551  */
1552 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1553                           enum pid_type type, struct pid *pid, int options,
1554                           struct siginfo __user *infop, int __user *stat_addr,
1555                           struct rusage __user *ru)
1556 {
1557         struct task_struct *p;
1558
1559         list_for_each_entry(p, &tsk->children, sibling) {
1560                 /*
1561                  * Do not consider detached threads.
1562                  */
1563                 if (!task_detached(p)) {
1564                         int ret = wait_consider_task(tsk, 0, p, notask_error,
1565                                                      type, pid, options,
1566                                                      infop, stat_addr, ru);
1567                         if (ret)
1568                                 return ret;
1569                 }
1570         }
1571
1572         return 0;
1573 }
1574
1575 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1576                           enum pid_type type, struct pid *pid, int options,
1577                           struct siginfo __user *infop, int __user *stat_addr,
1578                           struct rusage __user *ru)
1579 {
1580         struct task_struct *p;
1581
1582         /*
1583          * Traditionally we see ptrace'd stopped tasks regardless of options.
1584          */
1585         options |= WUNTRACED;
1586
1587         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1588                 int ret = wait_consider_task(tsk, 1, p, notask_error,
1589                                              type, pid, options,
1590                                              infop, stat_addr, ru);
1591                 if (ret)
1592                         return ret;
1593         }
1594
1595         return 0;
1596 }
1597
1598 static long do_wait(enum pid_type type, struct pid *pid, int options,
1599                     struct siginfo __user *infop, int __user *stat_addr,
1600                     struct rusage __user *ru)
1601 {
1602         DECLARE_WAITQUEUE(wait, current);
1603         struct task_struct *tsk;
1604         int retval;
1605
1606         trace_sched_process_wait(pid);
1607
1608         add_wait_queue(&current->signal->wait_chldexit,&wait);
1609 repeat:
1610         /*
1611          * If there is nothing that can match our critiera just get out.
1612          * We will clear @retval to zero if we see any child that might later
1613          * match our criteria, even if we are not able to reap it yet.
1614          */
1615         retval = -ECHILD;
1616         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1617                 goto end;
1618
1619         current->state = TASK_INTERRUPTIBLE;
1620         read_lock(&tasklist_lock);
1621         tsk = current;
1622         do {
1623                 int tsk_result = do_wait_thread(tsk, &retval,
1624                                                 type, pid, options,
1625                                                 infop, stat_addr, ru);
1626                 if (!tsk_result)
1627                         tsk_result = ptrace_do_wait(tsk, &retval,
1628                                                     type, pid, options,
1629                                                     infop, stat_addr, ru);
1630                 if (tsk_result) {
1631                         /*
1632                          * tasklist_lock is unlocked and we have a final result.
1633                          */
1634                         retval = tsk_result;
1635                         goto end;
1636                 }
1637
1638                 if (options & __WNOTHREAD)
1639                         break;
1640                 tsk = next_thread(tsk);
1641                 BUG_ON(tsk->signal != current->signal);
1642         } while (tsk != current);
1643         read_unlock(&tasklist_lock);
1644
1645         if (!retval && !(options & WNOHANG)) {
1646                 retval = -ERESTARTSYS;
1647                 if (!signal_pending(current)) {
1648                         schedule();
1649                         goto repeat;
1650                 }
1651         }
1652
1653 end:
1654         current->state = TASK_RUNNING;
1655         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1656         if (infop) {
1657                 if (retval > 0)
1658                         retval = 0;
1659                 else {
1660                         /*
1661                          * For a WNOHANG return, clear out all the fields
1662                          * we would set so the user can easily tell the
1663                          * difference.
1664                          */
1665                         if (!retval)
1666                                 retval = put_user(0, &infop->si_signo);
1667                         if (!retval)
1668                                 retval = put_user(0, &infop->si_errno);
1669                         if (!retval)
1670                                 retval = put_user(0, &infop->si_code);
1671                         if (!retval)
1672                                 retval = put_user(0, &infop->si_pid);
1673                         if (!retval)
1674                                 retval = put_user(0, &infop->si_uid);
1675                         if (!retval)
1676                                 retval = put_user(0, &infop->si_status);
1677                 }
1678         }
1679         return retval;
1680 }
1681
1682 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1683                 infop, int, options, struct rusage __user *, ru)
1684 {
1685         struct pid *pid = NULL;
1686         enum pid_type type;
1687         long ret;
1688
1689         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1690                 return -EINVAL;
1691         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1692                 return -EINVAL;
1693
1694         switch (which) {
1695         case P_ALL:
1696                 type = PIDTYPE_MAX;
1697                 break;
1698         case P_PID:
1699                 type = PIDTYPE_PID;
1700                 if (upid <= 0)
1701                         return -EINVAL;
1702                 break;
1703         case P_PGID:
1704                 type = PIDTYPE_PGID;
1705                 if (upid <= 0)
1706                         return -EINVAL;
1707                 break;
1708         default:
1709                 return -EINVAL;
1710         }
1711
1712         if (type < PIDTYPE_MAX)
1713                 pid = find_get_pid(upid);
1714         ret = do_wait(type, pid, options, infop, NULL, ru);
1715         put_pid(pid);
1716
1717         /* avoid REGPARM breakage on x86: */
1718         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1719         return ret;
1720 }
1721
1722 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1723                 int, options, struct rusage __user *, ru)
1724 {
1725         struct pid *pid = NULL;
1726         enum pid_type type;
1727         long ret;
1728
1729         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1730                         __WNOTHREAD|__WCLONE|__WALL))
1731                 return -EINVAL;
1732
1733         if (upid == -1)
1734                 type = PIDTYPE_MAX;
1735         else if (upid < 0) {
1736                 type = PIDTYPE_PGID;
1737                 pid = find_get_pid(-upid);
1738         } else if (upid == 0) {
1739                 type = PIDTYPE_PGID;
1740                 pid = get_task_pid(current, PIDTYPE_PGID);
1741         } else /* upid > 0 */ {
1742                 type = PIDTYPE_PID;
1743                 pid = find_get_pid(upid);
1744         }
1745
1746         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1747         put_pid(pid);
1748
1749         /* avoid REGPARM breakage on x86: */
1750         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1751         return ret;
1752 }
1753
1754 #ifdef __ARCH_WANT_SYS_WAITPID
1755
1756 /*
1757  * sys_waitpid() remains for compatibility. waitpid() should be
1758  * implemented by calling sys_wait4() from libc.a.
1759  */
1760 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1761 {
1762         return sys_wait4(pid, stat_addr, options, NULL);
1763 }
1764
1765 #endif