f9dfedd94af0c088c417b4c425956e47bf00f7ee
[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/fs_struct.h>
50 #include <linux/init_task.h>
51 #include <trace/sched.h>
52
53 #include <asm/uaccess.h>
54 #include <asm/unistd.h>
55 #include <asm/pgtable.h>
56 #include <asm/mmu_context.h>
57 #include "cred-internals.h"
58
59 DEFINE_TRACE(sched_process_free);
60 DEFINE_TRACE(sched_process_exit);
61 DEFINE_TRACE(sched_process_wait);
62
63 static void exit_mm(struct task_struct * tsk);
64
65 static void __unhash_process(struct task_struct *p)
66 {
67         nr_threads--;
68         detach_pid(p, PIDTYPE_PID);
69         if (thread_group_leader(p)) {
70                 detach_pid(p, PIDTYPE_PGID);
71                 detach_pid(p, PIDTYPE_SID);
72
73                 list_del_rcu(&p->tasks);
74                 __get_cpu_var(process_counts)--;
75         }
76         list_del_rcu(&p->thread_group);
77         list_del_init(&p->sibling);
78 }
79
80 /*
81  * This function expects the tasklist_lock write-locked.
82  */
83 static void __exit_signal(struct task_struct *tsk)
84 {
85         struct signal_struct *sig = tsk->signal;
86         struct sighand_struct *sighand;
87
88         BUG_ON(!sig);
89         BUG_ON(!atomic_read(&sig->count));
90
91         sighand = rcu_dereference(tsk->sighand);
92         spin_lock(&sighand->siglock);
93
94         posix_cpu_timers_exit(tsk);
95         if (atomic_dec_and_test(&sig->count))
96                 posix_cpu_timers_exit_group(tsk);
97         else {
98                 /*
99                  * If there is any task waiting for the group exit
100                  * then notify it:
101                  */
102                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
103                         wake_up_process(sig->group_exit_task);
104
105                 if (tsk == sig->curr_target)
106                         sig->curr_target = next_thread(tsk);
107                 /*
108                  * Accumulate here the counters for all threads but the
109                  * group leader as they die, so they can be added into
110                  * the process-wide totals when those are taken.
111                  * The group leader stays around as a zombie as long
112                  * as there are other threads.  When it gets reaped,
113                  * the exit.c code will add its counts into these totals.
114                  * We won't ever get here for the group leader, since it
115                  * will have been the last reference on the signal_struct.
116                  */
117                 sig->utime = cputime_add(sig->utime, task_utime(tsk));
118                 sig->stime = cputime_add(sig->stime, task_stime(tsk));
119                 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
120                 sig->min_flt += tsk->min_flt;
121                 sig->maj_flt += tsk->maj_flt;
122                 sig->nvcsw += tsk->nvcsw;
123                 sig->nivcsw += tsk->nivcsw;
124                 sig->inblock += task_io_get_inblock(tsk);
125                 sig->oublock += task_io_get_oublock(tsk);
126                 task_io_accounting_add(&sig->ioac, &tsk->ioac);
127                 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
128                 sig = NULL; /* Marker for below. */
129         }
130
131         __unhash_process(tsk);
132
133         /*
134          * Do this under ->siglock, we can race with another thread
135          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
136          */
137         flush_sigqueue(&tsk->pending);
138
139         tsk->signal = NULL;
140         tsk->sighand = NULL;
141         spin_unlock(&sighand->siglock);
142
143         __cleanup_sighand(sighand);
144         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
145         if (sig) {
146                 flush_sigqueue(&sig->shared_pending);
147                 taskstats_tgid_free(sig);
148                 /*
149                  * Make sure ->signal can't go away under rq->lock,
150                  * see account_group_exec_runtime().
151                  */
152                 task_rq_unlock_wait(tsk);
153                 __cleanup_signal(sig);
154         }
155 }
156
157 static void delayed_put_task_struct(struct rcu_head *rhp)
158 {
159         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
160
161 #ifdef CONFIG_PERF_COUNTERS
162         WARN_ON_ONCE(!list_empty(&tsk->perf_counter_ctx.counter_list));
163 #endif
164         trace_sched_process_free(tsk);
165         put_task_struct(tsk);
166 }
167
168
169 void release_task(struct task_struct * p)
170 {
171         struct task_struct *leader;
172         int zap_leader;
173 repeat:
174         tracehook_prepare_release_task(p);
175         /* don't need to get the RCU readlock here - the process is dead and
176          * can't be modifying its own credentials */
177         atomic_dec(&__task_cred(p)->user->processes);
178
179         proc_flush_task(p);
180
181         write_lock_irq(&tasklist_lock);
182         tracehook_finish_release_task(p);
183         __exit_signal(p);
184
185         /*
186          * If we are the last non-leader member of the thread
187          * group, and the leader is zombie, then notify the
188          * group leader's parent process. (if it wants notification.)
189          */
190         zap_leader = 0;
191         leader = p->group_leader;
192         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
193                 BUG_ON(task_detached(leader));
194                 do_notify_parent(leader, leader->exit_signal);
195                 /*
196                  * If we were the last child thread and the leader has
197                  * exited already, and the leader's parent ignores SIGCHLD,
198                  * then we are the one who should release the leader.
199                  *
200                  * do_notify_parent() will have marked it self-reaping in
201                  * that case.
202                  */
203                 zap_leader = task_detached(leader);
204
205                 /*
206                  * This maintains the invariant that release_task()
207                  * only runs on a task in EXIT_DEAD, just for sanity.
208                  */
209                 if (zap_leader)
210                         leader->exit_state = EXIT_DEAD;
211         }
212
213         write_unlock_irq(&tasklist_lock);
214         release_thread(p);
215         call_rcu(&p->rcu, delayed_put_task_struct);
216
217         p = leader;
218         if (unlikely(zap_leader))
219                 goto repeat;
220 }
221
222 /*
223  * This checks not only the pgrp, but falls back on the pid if no
224  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
225  * without this...
226  *
227  * The caller must hold rcu lock or the tasklist lock.
228  */
229 struct pid *session_of_pgrp(struct pid *pgrp)
230 {
231         struct task_struct *p;
232         struct pid *sid = NULL;
233
234         p = pid_task(pgrp, PIDTYPE_PGID);
235         if (p == NULL)
236                 p = pid_task(pgrp, PIDTYPE_PID);
237         if (p != NULL)
238                 sid = task_session(p);
239
240         return sid;
241 }
242
243 /*
244  * Determine if a process group is "orphaned", according to the POSIX
245  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
246  * by terminal-generated stop signals.  Newly orphaned process groups are
247  * to receive a SIGHUP and a SIGCONT.
248  *
249  * "I ask you, have you ever known what it is to be an orphan?"
250  */
251 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
252 {
253         struct task_struct *p;
254
255         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
256                 if ((p == ignored_task) ||
257                     (p->exit_state && thread_group_empty(p)) ||
258                     is_global_init(p->real_parent))
259                         continue;
260
261                 if (task_pgrp(p->real_parent) != pgrp &&
262                     task_session(p->real_parent) == task_session(p))
263                         return 0;
264         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
265
266         return 1;
267 }
268
269 int is_current_pgrp_orphaned(void)
270 {
271         int retval;
272
273         read_lock(&tasklist_lock);
274         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
275         read_unlock(&tasklist_lock);
276
277         return retval;
278 }
279
280 static int has_stopped_jobs(struct pid *pgrp)
281 {
282         int retval = 0;
283         struct task_struct *p;
284
285         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
286                 if (!task_is_stopped(p))
287                         continue;
288                 retval = 1;
289                 break;
290         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
291         return retval;
292 }
293
294 /*
295  * Check to see if any process groups have become orphaned as
296  * a result of our exiting, and if they have any stopped jobs,
297  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
298  */
299 static void
300 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
301 {
302         struct pid *pgrp = task_pgrp(tsk);
303         struct task_struct *ignored_task = tsk;
304
305         if (!parent)
306                  /* exit: our father is in a different pgrp than
307                   * we are and we were the only connection outside.
308                   */
309                 parent = tsk->real_parent;
310         else
311                 /* reparent: our child is in a different pgrp than
312                  * we are, and it was the only connection outside.
313                  */
314                 ignored_task = NULL;
315
316         if (task_pgrp(parent) != pgrp &&
317             task_session(parent) == task_session(tsk) &&
318             will_become_orphaned_pgrp(pgrp, ignored_task) &&
319             has_stopped_jobs(pgrp)) {
320                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
321                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
322         }
323 }
324
325 /**
326  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
327  *
328  * If a kernel thread is launched as a result of a system call, or if
329  * it ever exits, it should generally reparent itself to kthreadd so it
330  * isn't in the way of other processes and is correctly cleaned up on exit.
331  *
332  * The various task state such as scheduling policy and priority may have
333  * been inherited from a user process, so we reset them to sane values here.
334  *
335  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
336  */
337 static void reparent_to_kthreadd(void)
338 {
339         write_lock_irq(&tasklist_lock);
340
341         ptrace_unlink(current);
342         /* Reparent to init */
343         current->real_parent = current->parent = kthreadd_task;
344         list_move_tail(&current->sibling, &current->real_parent->children);
345
346         /* Set the exit signal to SIGCHLD so we signal init on exit */
347         current->exit_signal = SIGCHLD;
348
349         if (task_nice(current) < 0)
350                 set_user_nice(current, 0);
351         /* cpus_allowed? */
352         /* rt_priority? */
353         /* signals? */
354         memcpy(current->signal->rlim, init_task.signal->rlim,
355                sizeof(current->signal->rlim));
356
357         atomic_inc(&init_cred.usage);
358         commit_creds(&init_cred);
359         write_unlock_irq(&tasklist_lock);
360 }
361
362 void __set_special_pids(struct pid *pid)
363 {
364         struct task_struct *curr = current->group_leader;
365
366         if (task_session(curr) != pid)
367                 change_pid(curr, PIDTYPE_SID, pid);
368
369         if (task_pgrp(curr) != pid)
370                 change_pid(curr, PIDTYPE_PGID, pid);
371 }
372
373 static void set_special_pids(struct pid *pid)
374 {
375         write_lock_irq(&tasklist_lock);
376         __set_special_pids(pid);
377         write_unlock_irq(&tasklist_lock);
378 }
379
380 /*
381  * Let kernel threads use this to say that they
382  * allow a certain signal (since daemonize() will
383  * have disabled all of them by default).
384  */
385 int allow_signal(int sig)
386 {
387         if (!valid_signal(sig) || sig < 1)
388                 return -EINVAL;
389
390         spin_lock_irq(&current->sighand->siglock);
391         sigdelset(&current->blocked, sig);
392         if (!current->mm) {
393                 /* Kernel threads handle their own signals.
394                    Let the signal code know it'll be handled, so
395                    that they don't get converted to SIGKILL or
396                    just silently dropped */
397                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
398         }
399         recalc_sigpending();
400         spin_unlock_irq(&current->sighand->siglock);
401         return 0;
402 }
403
404 EXPORT_SYMBOL(allow_signal);
405
406 int disallow_signal(int sig)
407 {
408         if (!valid_signal(sig) || sig < 1)
409                 return -EINVAL;
410
411         spin_lock_irq(&current->sighand->siglock);
412         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
413         recalc_sigpending();
414         spin_unlock_irq(&current->sighand->siglock);
415         return 0;
416 }
417
418 EXPORT_SYMBOL(disallow_signal);
419
420 /*
421  *      Put all the gunge required to become a kernel thread without
422  *      attached user resources in one place where it belongs.
423  */
424
425 void daemonize(const char *name, ...)
426 {
427         va_list args;
428         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         daemonize_fs_struct();
461         exit_files(current);
462         current->files = init_task.files;
463         atomic_inc(&current->files->count);
464
465         reparent_to_kthreadd();
466 }
467
468 EXPORT_SYMBOL(daemonize);
469
470 static void close_files(struct files_struct * files)
471 {
472         int i, j;
473         struct fdtable *fdt;
474
475         j = 0;
476
477         /*
478          * It is safe to dereference the fd table without RCU or
479          * ->file_lock because this is the last reference to the
480          * files structure.
481          */
482         fdt = files_fdtable(files);
483         for (;;) {
484                 unsigned long set;
485                 i = j * __NFDBITS;
486                 if (i >= fdt->max_fds)
487                         break;
488                 set = fdt->open_fds->fds_bits[j++];
489                 while (set) {
490                         if (set & 1) {
491                                 struct file * file = xchg(&fdt->fd[i], NULL);
492                                 if (file) {
493                                         filp_close(file, files);
494                                         cond_resched();
495                                 }
496                         }
497                         i++;
498                         set >>= 1;
499                 }
500         }
501 }
502
503 struct files_struct *get_files_struct(struct task_struct *task)
504 {
505         struct files_struct *files;
506
507         task_lock(task);
508         files = task->files;
509         if (files)
510                 atomic_inc(&files->count);
511         task_unlock(task);
512
513         return files;
514 }
515
516 void put_files_struct(struct files_struct *files)
517 {
518         struct fdtable *fdt;
519
520         if (atomic_dec_and_test(&files->count)) {
521                 close_files(files);
522                 /*
523                  * Free the fd and fdset arrays if we expanded them.
524                  * If the fdtable was embedded, pass files for freeing
525                  * at the end of the RCU grace period. Otherwise,
526                  * you can free files immediately.
527                  */
528                 fdt = files_fdtable(files);
529                 if (fdt != &files->fdtab)
530                         kmem_cache_free(files_cachep, files);
531                 free_fdtable(fdt);
532         }
533 }
534
535 void reset_files_struct(struct files_struct *files)
536 {
537         struct task_struct *tsk = current;
538         struct files_struct *old;
539
540         old = tsk->files;
541         task_lock(tsk);
542         tsk->files = files;
543         task_unlock(tsk);
544         put_files_struct(old);
545 }
546
547 void exit_files(struct task_struct *tsk)
548 {
549         struct files_struct * files = tsk->files;
550
551         if (files) {
552                 task_lock(tsk);
553                 tsk->files = NULL;
554                 task_unlock(tsk);
555                 put_files_struct(files);
556         }
557 }
558
559 #ifdef CONFIG_MM_OWNER
560 /*
561  * Task p is exiting and it owned mm, lets find a new owner for it
562  */
563 static inline int
564 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
565 {
566         /*
567          * If there are other users of the mm and the owner (us) is exiting
568          * we need to find a new owner to take on the responsibility.
569          */
570         if (atomic_read(&mm->mm_users) <= 1)
571                 return 0;
572         if (mm->owner != p)
573                 return 0;
574         return 1;
575 }
576
577 void mm_update_next_owner(struct mm_struct *mm)
578 {
579         struct task_struct *c, *g, *p = current;
580
581 retry:
582         if (!mm_need_new_owner(mm, p))
583                 return;
584
585         read_lock(&tasklist_lock);
586         /*
587          * Search in the children
588          */
589         list_for_each_entry(c, &p->children, sibling) {
590                 if (c->mm == mm)
591                         goto assign_new_owner;
592         }
593
594         /*
595          * Search in the siblings
596          */
597         list_for_each_entry(c, &p->parent->children, sibling) {
598                 if (c->mm == mm)
599                         goto assign_new_owner;
600         }
601
602         /*
603          * Search through everything else. We should not get
604          * here often
605          */
606         do_each_thread(g, c) {
607                 if (c->mm == mm)
608                         goto assign_new_owner;
609         } while_each_thread(g, c);
610
611         read_unlock(&tasklist_lock);
612         /*
613          * We found no owner yet mm_users > 1: this implies that we are
614          * most likely racing with swapoff (try_to_unuse()) or /proc or
615          * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
616          */
617         mm->owner = NULL;
618         return;
619
620 assign_new_owner:
621         BUG_ON(c == p);
622         get_task_struct(c);
623         /*
624          * The task_lock protects c->mm from changing.
625          * We always want mm->owner->mm == mm
626          */
627         task_lock(c);
628         /*
629          * Delay read_unlock() till we have the task_lock()
630          * to ensure that c does not slip away underneath us
631          */
632         read_unlock(&tasklist_lock);
633         if (c->mm != mm) {
634                 task_unlock(c);
635                 put_task_struct(c);
636                 goto retry;
637         }
638         mm->owner = c;
639         task_unlock(c);
640         put_task_struct(c);
641 }
642 #endif /* CONFIG_MM_OWNER */
643
644 /*
645  * Turn us into a lazy TLB process if we
646  * aren't already..
647  */
648 static void exit_mm(struct task_struct * tsk)
649 {
650         struct mm_struct *mm = tsk->mm;
651         struct core_state *core_state;
652
653         mm_release(tsk, mm);
654         if (!mm)
655                 return;
656         /*
657          * Serialize with any possible pending coredump.
658          * We must hold mmap_sem around checking core_state
659          * and clearing tsk->mm.  The core-inducing thread
660          * will increment ->nr_threads for each thread in the
661          * group with ->mm != NULL.
662          */
663         down_read(&mm->mmap_sem);
664         core_state = mm->core_state;
665         if (core_state) {
666                 struct core_thread self;
667                 up_read(&mm->mmap_sem);
668
669                 self.task = tsk;
670                 self.next = xchg(&core_state->dumper.next, &self);
671                 /*
672                  * Implies mb(), the result of xchg() must be visible
673                  * to core_state->dumper.
674                  */
675                 if (atomic_dec_and_test(&core_state->nr_threads))
676                         complete(&core_state->startup);
677
678                 for (;;) {
679                         set_task_state(tsk, TASK_UNINTERRUPTIBLE);
680                         if (!self.task) /* see coredump_finish() */
681                                 break;
682                         schedule();
683                 }
684                 __set_task_state(tsk, TASK_RUNNING);
685                 down_read(&mm->mmap_sem);
686         }
687         atomic_inc(&mm->mm_count);
688         BUG_ON(mm != tsk->active_mm);
689         /* more a memory barrier than a real lock */
690         task_lock(tsk);
691         tsk->mm = NULL;
692         up_read(&mm->mmap_sem);
693         enter_lazy_tlb(mm, current);
694         /* We don't want this task to be frozen prematurely */
695         clear_freeze_flag(tsk);
696         task_unlock(tsk);
697         mm_update_next_owner(mm);
698         mmput(mm);
699 }
700
701 /*
702  * When we die, we re-parent all our children.
703  * Try to give them to another thread in our thread
704  * group, and if no such member exists, give it to
705  * the child reaper process (ie "init") in our pid
706  * space.
707  */
708 static struct task_struct *find_new_reaper(struct task_struct *father)
709 {
710         struct pid_namespace *pid_ns = task_active_pid_ns(father);
711         struct task_struct *thread;
712
713         thread = father;
714         while_each_thread(father, thread) {
715                 if (thread->flags & PF_EXITING)
716                         continue;
717                 if (unlikely(pid_ns->child_reaper == father))
718                         pid_ns->child_reaper = thread;
719                 return thread;
720         }
721
722         if (unlikely(pid_ns->child_reaper == father)) {
723                 write_unlock_irq(&tasklist_lock);
724                 if (unlikely(pid_ns == &init_pid_ns))
725                         panic("Attempted to kill init!");
726
727                 zap_pid_ns_processes(pid_ns);
728                 write_lock_irq(&tasklist_lock);
729                 /*
730                  * We can not clear ->child_reaper or leave it alone.
731                  * There may by stealth EXIT_DEAD tasks on ->children,
732                  * forget_original_parent() must move them somewhere.
733                  */
734                 pid_ns->child_reaper = init_pid_ns.child_reaper;
735         }
736
737         return pid_ns->child_reaper;
738 }
739
740 /*
741 * Any that need to be release_task'd are put on the @dead list.
742  */
743 static void reparent_thread(struct task_struct *father, struct task_struct *p,
744                                 struct list_head *dead)
745 {
746         if (p->pdeath_signal)
747                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
748
749         list_move_tail(&p->sibling, &p->real_parent->children);
750
751         if (task_detached(p))
752                 return;
753         /*
754          * If this is a threaded reparent there is no need to
755          * notify anyone anything has happened.
756          */
757         if (same_thread_group(p->real_parent, father))
758                 return;
759
760         /* We don't want people slaying init.  */
761         p->exit_signal = SIGCHLD;
762
763         /* If it has exited notify the new parent about this child's death. */
764         if (!p->ptrace &&
765             p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
766                 do_notify_parent(p, p->exit_signal);
767                 if (task_detached(p)) {
768                         p->exit_state = EXIT_DEAD;
769                         list_move_tail(&p->sibling, dead);
770                 }
771         }
772
773         kill_orphaned_pgrp(p, father);
774 }
775
776 static void forget_original_parent(struct task_struct *father)
777 {
778         struct task_struct *p, *n, *reaper;
779         LIST_HEAD(dead_children);
780
781         exit_ptrace(father);
782
783         write_lock_irq(&tasklist_lock);
784         reaper = find_new_reaper(father);
785
786         list_for_each_entry_safe(p, n, &father->children, sibling) {
787                 p->real_parent = reaper;
788                 if (p->parent == father) {
789                         BUG_ON(p->ptrace);
790                         p->parent = p->real_parent;
791                 }
792                 reparent_thread(father, p, &dead_children);
793         }
794         write_unlock_irq(&tasklist_lock);
795
796         BUG_ON(!list_empty(&father->children));
797
798         list_for_each_entry_safe(p, n, &dead_children, sibling) {
799                 list_del_init(&p->sibling);
800                 release_task(p);
801         }
802 }
803
804 /*
805  * Send signals to all our closest relatives so that they know
806  * to properly mourn us..
807  */
808 static void exit_notify(struct task_struct *tsk, int group_dead)
809 {
810         int signal;
811         void *cookie;
812
813         /*
814          * This does two things:
815          *
816          * A.  Make init inherit all the child processes
817          * B.  Check to see if any process groups have become orphaned
818          *      as a result of our exiting, and if they have any stopped
819          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
820          */
821         forget_original_parent(tsk);
822         exit_task_namespaces(tsk);
823
824         write_lock_irq(&tasklist_lock);
825         if (group_dead)
826                 kill_orphaned_pgrp(tsk->group_leader, NULL);
827
828         /* Let father know we died
829          *
830          * Thread signals are configurable, but you aren't going to use
831          * that to send signals to arbitary processes.
832          * That stops right now.
833          *
834          * If the parent exec id doesn't match the exec id we saved
835          * when we started then we know the parent has changed security
836          * domain.
837          *
838          * If our self_exec id doesn't match our parent_exec_id then
839          * we have changed execution domain as these two values started
840          * the same after a fork.
841          */
842         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
843             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
844              tsk->self_exec_id != tsk->parent_exec_id))
845                 tsk->exit_signal = SIGCHLD;
846
847         signal = tracehook_notify_death(tsk, &cookie, group_dead);
848         if (signal >= 0)
849                 signal = do_notify_parent(tsk, signal);
850
851         tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
852
853         /* mt-exec, de_thread() is waiting for us */
854         if (thread_group_leader(tsk) &&
855             tsk->signal->group_exit_task &&
856             tsk->signal->notify_count < 0)
857                 wake_up_process(tsk->signal->group_exit_task);
858
859         write_unlock_irq(&tasklist_lock);
860
861         tracehook_report_death(tsk, signal, cookie, group_dead);
862
863         /* If the process is dead, release it - nobody will wait for it */
864         if (signal == DEATH_REAP)
865                 release_task(tsk);
866 }
867
868 #ifdef CONFIG_DEBUG_STACK_USAGE
869 static void check_stack_usage(void)
870 {
871         static DEFINE_SPINLOCK(low_water_lock);
872         static int lowest_to_date = THREAD_SIZE;
873         unsigned long free;
874
875         free = stack_not_used(current);
876
877         if (free >= lowest_to_date)
878                 return;
879
880         spin_lock(&low_water_lock);
881         if (free < lowest_to_date) {
882                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
883                                 "left\n",
884                                 current->comm, free);
885                 lowest_to_date = free;
886         }
887         spin_unlock(&low_water_lock);
888 }
889 #else
890 static inline void check_stack_usage(void) {}
891 #endif
892
893 NORET_TYPE void do_exit(long code)
894 {
895         struct task_struct *tsk = current;
896         int group_dead;
897
898         profile_task_exit(tsk);
899
900         WARN_ON(atomic_read(&tsk->fs_excl));
901
902         if (unlikely(in_interrupt()))
903                 panic("Aiee, killing interrupt handler!");
904         if (unlikely(!tsk->pid))
905                 panic("Attempted to kill the idle task!");
906
907         tracehook_report_exit(&code);
908
909         /*
910          * We're taking recursive faults here in do_exit. Safest is to just
911          * leave this task alone and wait for reboot.
912          */
913         if (unlikely(tsk->flags & PF_EXITING)) {
914                 printk(KERN_ALERT
915                         "Fixing recursive fault but reboot is needed!\n");
916                 /*
917                  * We can do this unlocked here. The futex code uses
918                  * this flag just to verify whether the pi state
919                  * cleanup has been done or not. In the worst case it
920                  * loops once more. We pretend that the cleanup was
921                  * done as there is no way to return. Either the
922                  * OWNER_DIED bit is set by now or we push the blocked
923                  * task into the wait for ever nirwana as well.
924                  */
925                 tsk->flags |= PF_EXITPIDONE;
926                 set_current_state(TASK_UNINTERRUPTIBLE);
927                 schedule();
928         }
929
930         exit_irq_thread();
931
932         exit_signals(tsk);  /* sets PF_EXITING */
933         /*
934          * tsk->flags are checked in the futex code to protect against
935          * an exiting task cleaning up the robust pi futexes.
936          */
937         smp_mb();
938         spin_unlock_wait(&tsk->pi_lock);
939
940         if (unlikely(in_atomic()))
941                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
942                                 current->comm, task_pid_nr(current),
943                                 preempt_count());
944
945         acct_update_integrals(tsk);
946
947         group_dead = atomic_dec_and_test(&tsk->signal->live);
948         if (group_dead) {
949                 hrtimer_cancel(&tsk->signal->real_timer);
950                 exit_itimers(tsk->signal);
951         }
952         acct_collect(code, group_dead);
953         if (group_dead)
954                 tty_audit_exit();
955         if (unlikely(tsk->audit_context))
956                 audit_free(tsk);
957
958         tsk->exit_code = code;
959         taskstats_exit(tsk, group_dead);
960
961         exit_mm(tsk);
962
963         if (group_dead)
964                 acct_process();
965         trace_sched_process_exit(tsk);
966
967         exit_sem(tsk);
968         exit_files(tsk);
969         exit_fs(tsk);
970         check_stack_usage();
971         exit_thread();
972         cgroup_exit(tsk, 1);
973
974         if (group_dead && tsk->signal->leader)
975                 disassociate_ctty(1);
976
977         module_put(task_thread_info(tsk)->exec_domain->module);
978         if (tsk->binfmt)
979                 module_put(tsk->binfmt->module);
980
981         proc_exit_connector(tsk);
982
983         /*
984          * Flush inherited counters to the parent - before the parent
985          * gets woken up by child-exit notifications.
986          */
987         perf_counter_exit_task(tsk);
988
989         exit_notify(tsk, group_dead);
990 #ifdef CONFIG_NUMA
991         mpol_put(tsk->mempolicy);
992         tsk->mempolicy = NULL;
993 #endif
994 #ifdef CONFIG_FUTEX
995         if (unlikely(!list_empty(&tsk->pi_state_list)))
996                 exit_pi_state_list(tsk);
997         if (unlikely(current->pi_state_cache))
998                 kfree(current->pi_state_cache);
999 #endif
1000         /*
1001          * Make sure we are holding no locks:
1002          */
1003         debug_check_no_locks_held(tsk);
1004         /*
1005          * We can do this unlocked here. The futex code uses this flag
1006          * just to verify whether the pi state cleanup has been done
1007          * or not. In the worst case it loops once more.
1008          */
1009         tsk->flags |= PF_EXITPIDONE;
1010
1011         if (tsk->io_context)
1012                 exit_io_context();
1013
1014         if (tsk->splice_pipe)
1015                 __free_pipe_info(tsk->splice_pipe);
1016
1017         preempt_disable();
1018         /* causes final put_task_struct in finish_task_switch(). */
1019         tsk->state = TASK_DEAD;
1020         schedule();
1021         BUG();
1022         /* Avoid "noreturn function does return".  */
1023         for (;;)
1024                 cpu_relax();    /* For when BUG is null */
1025 }
1026
1027 EXPORT_SYMBOL_GPL(do_exit);
1028
1029 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1030 {
1031         if (comp)
1032                 complete(comp);
1033
1034         do_exit(code);
1035 }
1036
1037 EXPORT_SYMBOL(complete_and_exit);
1038
1039 SYSCALL_DEFINE1(exit, int, error_code)
1040 {
1041         do_exit((error_code&0xff)<<8);
1042 }
1043
1044 /*
1045  * Take down every thread in the group.  This is called by fatal signals
1046  * as well as by sys_exit_group (below).
1047  */
1048 NORET_TYPE void
1049 do_group_exit(int exit_code)
1050 {
1051         struct signal_struct *sig = current->signal;
1052
1053         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1054
1055         if (signal_group_exit(sig))
1056                 exit_code = sig->group_exit_code;
1057         else if (!thread_group_empty(current)) {
1058                 struct sighand_struct *const sighand = current->sighand;
1059                 spin_lock_irq(&sighand->siglock);
1060                 if (signal_group_exit(sig))
1061                         /* Another thread got here before we took the lock.  */
1062                         exit_code = sig->group_exit_code;
1063                 else {
1064                         sig->group_exit_code = exit_code;
1065                         sig->flags = SIGNAL_GROUP_EXIT;
1066                         zap_other_threads(current);
1067                 }
1068                 spin_unlock_irq(&sighand->siglock);
1069         }
1070
1071         do_exit(exit_code);
1072         /* NOTREACHED */
1073 }
1074
1075 /*
1076  * this kills every thread in the thread group. Note that any externally
1077  * wait4()-ing process will get the correct exit code - even if this
1078  * thread is not the thread group leader.
1079  */
1080 SYSCALL_DEFINE1(exit_group, int, error_code)
1081 {
1082         do_group_exit((error_code & 0xff) << 8);
1083         /* NOTREACHED */
1084         return 0;
1085 }
1086
1087 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1088 {
1089         struct pid *pid = NULL;
1090         if (type == PIDTYPE_PID)
1091                 pid = task->pids[type].pid;
1092         else if (type < PIDTYPE_MAX)
1093                 pid = task->group_leader->pids[type].pid;
1094         return pid;
1095 }
1096
1097 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1098                           struct task_struct *p)
1099 {
1100         int err;
1101
1102         if (type < PIDTYPE_MAX) {
1103                 if (task_pid_type(p, type) != pid)
1104                         return 0;
1105         }
1106
1107         /* Wait for all children (clone and not) if __WALL is set;
1108          * otherwise, wait for clone children *only* if __WCLONE is
1109          * set; otherwise, wait for non-clone children *only*.  (Note:
1110          * A "clone" child here is one that reports to its parent
1111          * using a signal other than SIGCHLD.) */
1112         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1113             && !(options & __WALL))
1114                 return 0;
1115
1116         err = security_task_wait(p);
1117         if (err)
1118                 return err;
1119
1120         return 1;
1121 }
1122
1123 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1124                                int why, int status,
1125                                struct siginfo __user *infop,
1126                                struct rusage __user *rusagep)
1127 {
1128         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1129
1130         put_task_struct(p);
1131         if (!retval)
1132                 retval = put_user(SIGCHLD, &infop->si_signo);
1133         if (!retval)
1134                 retval = put_user(0, &infop->si_errno);
1135         if (!retval)
1136                 retval = put_user((short)why, &infop->si_code);
1137         if (!retval)
1138                 retval = put_user(pid, &infop->si_pid);
1139         if (!retval)
1140                 retval = put_user(uid, &infop->si_uid);
1141         if (!retval)
1142                 retval = put_user(status, &infop->si_status);
1143         if (!retval)
1144                 retval = pid;
1145         return retval;
1146 }
1147
1148 /*
1149  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1150  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1151  * the lock and this task is uninteresting.  If we return nonzero, we have
1152  * released the lock and the system call should return.
1153  */
1154 static int wait_task_zombie(struct task_struct *p, int options,
1155                             struct siginfo __user *infop,
1156                             int __user *stat_addr, struct rusage __user *ru)
1157 {
1158         unsigned long state;
1159         int retval, status, traced;
1160         pid_t pid = task_pid_vnr(p);
1161         uid_t uid = __task_cred(p)->uid;
1162
1163         if (!likely(options & WEXITED))
1164                 return 0;
1165
1166         if (unlikely(options & WNOWAIT)) {
1167                 int exit_code = p->exit_code;
1168                 int why, status;
1169
1170                 get_task_struct(p);
1171                 read_unlock(&tasklist_lock);
1172                 if ((exit_code & 0x7f) == 0) {
1173                         why = CLD_EXITED;
1174                         status = exit_code >> 8;
1175                 } else {
1176                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1177                         status = exit_code & 0x7f;
1178                 }
1179                 return wait_noreap_copyout(p, pid, uid, why,
1180                                            status, infop, ru);
1181         }
1182
1183         /*
1184          * Try to move the task's state to DEAD
1185          * only one thread is allowed to do this:
1186          */
1187         state = xchg(&p->exit_state, EXIT_DEAD);
1188         if (state != EXIT_ZOMBIE) {
1189                 BUG_ON(state != EXIT_DEAD);
1190                 return 0;
1191         }
1192
1193         traced = ptrace_reparented(p);
1194
1195         if (likely(!traced)) {
1196                 struct signal_struct *psig;
1197                 struct signal_struct *sig;
1198                 struct task_cputime cputime;
1199
1200                 /*
1201                  * The resource counters for the group leader are in its
1202                  * own task_struct.  Those for dead threads in the group
1203                  * are in its signal_struct, as are those for the child
1204                  * processes it has previously reaped.  All these
1205                  * accumulate in the parent's signal_struct c* fields.
1206                  *
1207                  * We don't bother to take a lock here to protect these
1208                  * p->signal fields, because they are only touched by
1209                  * __exit_signal, which runs with tasklist_lock
1210                  * write-locked anyway, and so is excluded here.  We do
1211                  * need to protect the access to p->parent->signal fields,
1212                  * as other threads in the parent group can be right
1213                  * here reaping other children at the same time.
1214                  *
1215                  * We use thread_group_cputime() to get times for the thread
1216                  * group, which consolidates times for all threads in the
1217                  * group including the group leader.
1218                  */
1219                 thread_group_cputime(p, &cputime);
1220                 spin_lock_irq(&p->parent->sighand->siglock);
1221                 psig = p->parent->signal;
1222                 sig = p->signal;
1223                 psig->cutime =
1224                         cputime_add(psig->cutime,
1225                         cputime_add(cputime.utime,
1226                                     sig->cutime));
1227                 psig->cstime =
1228                         cputime_add(psig->cstime,
1229                         cputime_add(cputime.stime,
1230                                     sig->cstime));
1231                 psig->cgtime =
1232                         cputime_add(psig->cgtime,
1233                         cputime_add(p->gtime,
1234                         cputime_add(sig->gtime,
1235                                     sig->cgtime)));
1236                 psig->cmin_flt +=
1237                         p->min_flt + sig->min_flt + sig->cmin_flt;
1238                 psig->cmaj_flt +=
1239                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1240                 psig->cnvcsw +=
1241                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1242                 psig->cnivcsw +=
1243                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1244                 psig->cinblock +=
1245                         task_io_get_inblock(p) +
1246                         sig->inblock + sig->cinblock;
1247                 psig->coublock +=
1248                         task_io_get_oublock(p) +
1249                         sig->oublock + sig->coublock;
1250                 task_io_accounting_add(&psig->ioac, &p->ioac);
1251                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1252                 spin_unlock_irq(&p->parent->sighand->siglock);
1253         }
1254
1255         /*
1256          * Now we are sure this task is interesting, and no other
1257          * thread can reap it because we set its state to EXIT_DEAD.
1258          */
1259         read_unlock(&tasklist_lock);
1260
1261         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1262         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1263                 ? p->signal->group_exit_code : p->exit_code;
1264         if (!retval && stat_addr)
1265                 retval = put_user(status, stat_addr);
1266         if (!retval && infop)
1267                 retval = put_user(SIGCHLD, &infop->si_signo);
1268         if (!retval && infop)
1269                 retval = put_user(0, &infop->si_errno);
1270         if (!retval && infop) {
1271                 int why;
1272
1273                 if ((status & 0x7f) == 0) {
1274                         why = CLD_EXITED;
1275                         status >>= 8;
1276                 } else {
1277                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1278                         status &= 0x7f;
1279                 }
1280                 retval = put_user((short)why, &infop->si_code);
1281                 if (!retval)
1282                         retval = put_user(status, &infop->si_status);
1283         }
1284         if (!retval && infop)
1285                 retval = put_user(pid, &infop->si_pid);
1286         if (!retval && infop)
1287                 retval = put_user(uid, &infop->si_uid);
1288         if (!retval)
1289                 retval = pid;
1290
1291         if (traced) {
1292                 write_lock_irq(&tasklist_lock);
1293                 /* We dropped tasklist, ptracer could die and untrace */
1294                 ptrace_unlink(p);
1295                 /*
1296                  * If this is not a detached task, notify the parent.
1297                  * If it's still not detached after that, don't release
1298                  * it now.
1299                  */
1300                 if (!task_detached(p)) {
1301                         do_notify_parent(p, p->exit_signal);
1302                         if (!task_detached(p)) {
1303                                 p->exit_state = EXIT_ZOMBIE;
1304                                 p = NULL;
1305                         }
1306                 }
1307                 write_unlock_irq(&tasklist_lock);
1308         }
1309         if (p != NULL)
1310                 release_task(p);
1311
1312         return retval;
1313 }
1314
1315 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1316 {
1317         if (ptrace) {
1318                 if (task_is_stopped_or_traced(p))
1319                         return &p->exit_code;
1320         } else {
1321                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1322                         return &p->signal->group_exit_code;
1323         }
1324         return NULL;
1325 }
1326
1327 /*
1328  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1329  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1330  * the lock and this task is uninteresting.  If we return nonzero, we have
1331  * released the lock and the system call should return.
1332  */
1333 static int wait_task_stopped(int ptrace, struct task_struct *p,
1334                              int options, struct siginfo __user *infop,
1335                              int __user *stat_addr, struct rusage __user *ru)
1336 {
1337         int retval, exit_code, *p_code, why;
1338         uid_t uid = 0; /* unneeded, required by compiler */
1339         pid_t pid;
1340
1341         if (!(options & WUNTRACED))
1342                 return 0;
1343
1344         exit_code = 0;
1345         spin_lock_irq(&p->sighand->siglock);
1346
1347         p_code = task_stopped_code(p, ptrace);
1348         if (unlikely(!p_code))
1349                 goto unlock_sig;
1350
1351         exit_code = *p_code;
1352         if (!exit_code)
1353                 goto unlock_sig;
1354
1355         if (!unlikely(options & WNOWAIT))
1356                 *p_code = 0;
1357
1358         /* don't need the RCU readlock here as we're holding a spinlock */
1359         uid = __task_cred(p)->uid;
1360 unlock_sig:
1361         spin_unlock_irq(&p->sighand->siglock);
1362         if (!exit_code)
1363                 return 0;
1364
1365         /*
1366          * Now we are pretty sure this task is interesting.
1367          * Make sure it doesn't get reaped out from under us while we
1368          * give up the lock and then examine it below.  We don't want to
1369          * keep holding onto the tasklist_lock while we call getrusage and
1370          * possibly take page faults for user memory.
1371          */
1372         get_task_struct(p);
1373         pid = task_pid_vnr(p);
1374         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1375         read_unlock(&tasklist_lock);
1376
1377         if (unlikely(options & WNOWAIT))
1378                 return wait_noreap_copyout(p, pid, uid,
1379                                            why, exit_code,
1380                                            infop, ru);
1381
1382         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1383         if (!retval && stat_addr)
1384                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1385         if (!retval && infop)
1386                 retval = put_user(SIGCHLD, &infop->si_signo);
1387         if (!retval && infop)
1388                 retval = put_user(0, &infop->si_errno);
1389         if (!retval && infop)
1390                 retval = put_user((short)why, &infop->si_code);
1391         if (!retval && infop)
1392                 retval = put_user(exit_code, &infop->si_status);
1393         if (!retval && infop)
1394                 retval = put_user(pid, &infop->si_pid);
1395         if (!retval && infop)
1396                 retval = put_user(uid, &infop->si_uid);
1397         if (!retval)
1398                 retval = pid;
1399         put_task_struct(p);
1400
1401         BUG_ON(!retval);
1402         return retval;
1403 }
1404
1405 /*
1406  * Handle do_wait work for one task in a live, non-stopped state.
1407  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1408  * the lock and this task is uninteresting.  If we return nonzero, we have
1409  * released the lock and the system call should return.
1410  */
1411 static int wait_task_continued(struct task_struct *p, int options,
1412                                struct siginfo __user *infop,
1413                                int __user *stat_addr, struct rusage __user *ru)
1414 {
1415         int retval;
1416         pid_t pid;
1417         uid_t uid;
1418
1419         if (!unlikely(options & WCONTINUED))
1420                 return 0;
1421
1422         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1423                 return 0;
1424
1425         spin_lock_irq(&p->sighand->siglock);
1426         /* Re-check with the lock held.  */
1427         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1428                 spin_unlock_irq(&p->sighand->siglock);
1429                 return 0;
1430         }
1431         if (!unlikely(options & WNOWAIT))
1432                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1433         uid = __task_cred(p)->uid;
1434         spin_unlock_irq(&p->sighand->siglock);
1435
1436         pid = task_pid_vnr(p);
1437         get_task_struct(p);
1438         read_unlock(&tasklist_lock);
1439
1440         if (!infop) {
1441                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1442                 put_task_struct(p);
1443                 if (!retval && stat_addr)
1444                         retval = put_user(0xffff, stat_addr);
1445                 if (!retval)
1446                         retval = pid;
1447         } else {
1448                 retval = wait_noreap_copyout(p, pid, uid,
1449                                              CLD_CONTINUED, SIGCONT,
1450                                              infop, ru);
1451                 BUG_ON(retval == 0);
1452         }
1453
1454         return retval;
1455 }
1456
1457 /*
1458  * Consider @p for a wait by @parent.
1459  *
1460  * -ECHILD should be in *@notask_error before the first call.
1461  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1462  * Returns zero if the search for a child should continue;
1463  * then *@notask_error is 0 if @p is an eligible child,
1464  * or another error from security_task_wait(), or still -ECHILD.
1465  */
1466 static int wait_consider_task(struct task_struct *parent, int ptrace,
1467                               struct task_struct *p, int *notask_error,
1468                               enum pid_type type, struct pid *pid, int options,
1469                               struct siginfo __user *infop,
1470                               int __user *stat_addr, struct rusage __user *ru)
1471 {
1472         int ret = eligible_child(type, pid, options, p);
1473         if (!ret)
1474                 return ret;
1475
1476         if (unlikely(ret < 0)) {
1477                 /*
1478                  * If we have not yet seen any eligible child,
1479                  * then let this error code replace -ECHILD.
1480                  * A permission error will give the user a clue
1481                  * to look for security policy problems, rather
1482                  * than for mysterious wait bugs.
1483                  */
1484                 if (*notask_error)
1485                         *notask_error = ret;
1486         }
1487
1488         if (likely(!ptrace) && unlikely(p->ptrace)) {
1489                 /*
1490                  * This child is hidden by ptrace.
1491                  * We aren't allowed to see it now, but eventually we will.
1492                  */
1493                 *notask_error = 0;
1494                 return 0;
1495         }
1496
1497         if (p->exit_state == EXIT_DEAD)
1498                 return 0;
1499
1500         /*
1501          * We don't reap group leaders with subthreads.
1502          */
1503         if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1504                 return wait_task_zombie(p, options, infop, stat_addr, ru);
1505
1506         /*
1507          * It's stopped or running now, so it might
1508          * later continue, exit, or stop again.
1509          */
1510         *notask_error = 0;
1511
1512         if (task_stopped_code(p, ptrace))
1513                 return wait_task_stopped(ptrace, p, options,
1514                                          infop, stat_addr, ru);
1515
1516         return wait_task_continued(p, options, infop, stat_addr, ru);
1517 }
1518
1519 /*
1520  * Do the work of do_wait() for one thread in the group, @tsk.
1521  *
1522  * -ECHILD should be in *@notask_error before the first call.
1523  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1524  * Returns zero if the search for a child should continue; then
1525  * *@notask_error is 0 if there were any eligible children,
1526  * or another error from security_task_wait(), or still -ECHILD.
1527  */
1528 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1529                           enum pid_type type, struct pid *pid, int options,
1530                           struct siginfo __user *infop, int __user *stat_addr,
1531                           struct rusage __user *ru)
1532 {
1533         struct task_struct *p;
1534
1535         list_for_each_entry(p, &tsk->children, sibling) {
1536                 /*
1537                  * Do not consider detached threads.
1538                  */
1539                 if (!task_detached(p)) {
1540                         int ret = wait_consider_task(tsk, 0, p, notask_error,
1541                                                      type, pid, options,
1542                                                      infop, stat_addr, ru);
1543                         if (ret)
1544                                 return ret;
1545                 }
1546         }
1547
1548         return 0;
1549 }
1550
1551 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1552                           enum pid_type type, struct pid *pid, int options,
1553                           struct siginfo __user *infop, int __user *stat_addr,
1554                           struct rusage __user *ru)
1555 {
1556         struct task_struct *p;
1557
1558         /*
1559          * Traditionally we see ptrace'd stopped tasks regardless of options.
1560          */
1561         options |= WUNTRACED;
1562
1563         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1564                 int ret = wait_consider_task(tsk, 1, p, notask_error,
1565                                              type, pid, options,
1566                                              infop, stat_addr, ru);
1567                 if (ret)
1568                         return ret;
1569         }
1570
1571         return 0;
1572 }
1573
1574 static long do_wait(enum pid_type type, struct pid *pid, int options,
1575                     struct siginfo __user *infop, int __user *stat_addr,
1576                     struct rusage __user *ru)
1577 {
1578         DECLARE_WAITQUEUE(wait, current);
1579         struct task_struct *tsk;
1580         int retval;
1581
1582         trace_sched_process_wait(pid);
1583
1584         add_wait_queue(&current->signal->wait_chldexit,&wait);
1585 repeat:
1586         /*
1587          * If there is nothing that can match our critiera just get out.
1588          * We will clear @retval to zero if we see any child that might later
1589          * match our criteria, even if we are not able to reap it yet.
1590          */
1591         retval = -ECHILD;
1592         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1593                 goto end;
1594
1595         current->state = TASK_INTERRUPTIBLE;
1596         read_lock(&tasklist_lock);
1597         tsk = current;
1598         do {
1599                 int tsk_result = do_wait_thread(tsk, &retval,
1600                                                 type, pid, options,
1601                                                 infop, stat_addr, ru);
1602                 if (!tsk_result)
1603                         tsk_result = ptrace_do_wait(tsk, &retval,
1604                                                     type, pid, options,
1605                                                     infop, stat_addr, ru);
1606                 if (tsk_result) {
1607                         /*
1608                          * tasklist_lock is unlocked and we have a final result.
1609                          */
1610                         retval = tsk_result;
1611                         goto end;
1612                 }
1613
1614                 if (options & __WNOTHREAD)
1615                         break;
1616                 tsk = next_thread(tsk);
1617                 BUG_ON(tsk->signal != current->signal);
1618         } while (tsk != current);
1619         read_unlock(&tasklist_lock);
1620
1621         if (!retval && !(options & WNOHANG)) {
1622                 retval = -ERESTARTSYS;
1623                 if (!signal_pending(current)) {
1624                         schedule();
1625                         goto repeat;
1626                 }
1627         }
1628
1629 end:
1630         current->state = TASK_RUNNING;
1631         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1632         if (infop) {
1633                 if (retval > 0)
1634                         retval = 0;
1635                 else {
1636                         /*
1637                          * For a WNOHANG return, clear out all the fields
1638                          * we would set so the user can easily tell the
1639                          * difference.
1640                          */
1641                         if (!retval)
1642                                 retval = put_user(0, &infop->si_signo);
1643                         if (!retval)
1644                                 retval = put_user(0, &infop->si_errno);
1645                         if (!retval)
1646                                 retval = put_user(0, &infop->si_code);
1647                         if (!retval)
1648                                 retval = put_user(0, &infop->si_pid);
1649                         if (!retval)
1650                                 retval = put_user(0, &infop->si_uid);
1651                         if (!retval)
1652                                 retval = put_user(0, &infop->si_status);
1653                 }
1654         }
1655         return retval;
1656 }
1657
1658 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1659                 infop, int, options, struct rusage __user *, ru)
1660 {
1661         struct pid *pid = NULL;
1662         enum pid_type type;
1663         long ret;
1664
1665         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1666                 return -EINVAL;
1667         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1668                 return -EINVAL;
1669
1670         switch (which) {
1671         case P_ALL:
1672                 type = PIDTYPE_MAX;
1673                 break;
1674         case P_PID:
1675                 type = PIDTYPE_PID;
1676                 if (upid <= 0)
1677                         return -EINVAL;
1678                 break;
1679         case P_PGID:
1680                 type = PIDTYPE_PGID;
1681                 if (upid <= 0)
1682                         return -EINVAL;
1683                 break;
1684         default:
1685                 return -EINVAL;
1686         }
1687
1688         if (type < PIDTYPE_MAX)
1689                 pid = find_get_pid(upid);
1690         ret = do_wait(type, pid, options, infop, NULL, ru);
1691         put_pid(pid);
1692
1693         /* avoid REGPARM breakage on x86: */
1694         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1695         return ret;
1696 }
1697
1698 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1699                 int, options, struct rusage __user *, ru)
1700 {
1701         struct pid *pid = NULL;
1702         enum pid_type type;
1703         long ret;
1704
1705         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1706                         __WNOTHREAD|__WCLONE|__WALL))
1707                 return -EINVAL;
1708
1709         if (upid == -1)
1710                 type = PIDTYPE_MAX;
1711         else if (upid < 0) {
1712                 type = PIDTYPE_PGID;
1713                 pid = find_get_pid(-upid);
1714         } else if (upid == 0) {
1715                 type = PIDTYPE_PGID;
1716                 pid = get_task_pid(current, PIDTYPE_PGID);
1717         } else /* upid > 0 */ {
1718                 type = PIDTYPE_PID;
1719                 pid = find_get_pid(upid);
1720         }
1721
1722         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1723         put_pid(pid);
1724
1725         /* avoid REGPARM breakage on x86: */
1726         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1727         return ret;
1728 }
1729
1730 #ifdef __ARCH_WANT_SYS_WAITPID
1731
1732 /*
1733  * sys_waitpid() remains for compatibility. waitpid() should be
1734  * implemented by calling sys_wait4() from libc.a.
1735  */
1736 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1737 {
1738         return sys_wait4(pid, stat_addr, options, NULL);
1739 }
1740
1741 #endif