e0f6e1892fb9315b0291943cf12a39d6feb810c6
[linux-2.6.git] / kernel / exit.c
1 /*
2  *  linux/kernel/exit.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 #include <linux/mm.h>
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/mnt_namespace.h>
16 #include <linux/iocontext.h>
17 #include <linux/key.h>
18 #include <linux/security.h>
19 #include <linux/cpu.h>
20 #include <linux/acct.h>
21 #include <linux/tsacct_kern.h>
22 #include <linux/file.h>
23 #include <linux/fdtable.h>
24 #include <linux/binfmts.h>
25 #include <linux/nsproxy.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/ptrace.h>
28 #include <linux/profile.h>
29 #include <linux/mount.h>
30 #include <linux/proc_fs.h>
31 #include <linux/kthread.h>
32 #include <linux/mempolicy.h>
33 #include <linux/taskstats_kern.h>
34 #include <linux/delayacct.h>
35 #include <linux/freezer.h>
36 #include <linux/cgroup.h>
37 #include <linux/syscalls.h>
38 #include <linux/signal.h>
39 #include <linux/posix-timers.h>
40 #include <linux/cn_proc.h>
41 #include <linux/mutex.h>
42 #include <linux/futex.h>
43 #include <linux/compat.h>
44 #include <linux/pipe_fs_i.h>
45 #include <linux/audit.h> /* for audit_free() */
46 #include <linux/resource.h>
47 #include <linux/blkdev.h>
48 #include <linux/task_io_accounting_ops.h>
49 #include <linux/tracehook.h>
50 #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
57 static void exit_mm(struct task_struct * tsk);
58
59 static inline int task_detached(struct task_struct *p)
60 {
61         return p->exit_signal == -1;
62 }
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->gtime = cputime_add(sig->gtime, task_gtime(tsk));
117                 sig->min_flt += tsk->min_flt;
118                 sig->maj_flt += tsk->maj_flt;
119                 sig->nvcsw += tsk->nvcsw;
120                 sig->nivcsw += tsk->nivcsw;
121                 sig->inblock += task_io_get_inblock(tsk);
122                 sig->oublock += task_io_get_oublock(tsk);
123                 task_io_accounting_add(&sig->ioac, &tsk->ioac);
124                 sig = NULL; /* Marker for below. */
125         }
126
127         __unhash_process(tsk);
128
129         /*
130          * Do this under ->siglock, we can race with another thread
131          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
132          */
133         flush_sigqueue(&tsk->pending);
134
135         tsk->signal = NULL;
136         tsk->sighand = NULL;
137         spin_unlock(&sighand->siglock);
138
139         __cleanup_sighand(sighand);
140         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
141         if (sig) {
142                 flush_sigqueue(&sig->shared_pending);
143                 taskstats_tgid_free(sig);
144                 __cleanup_signal(sig);
145         }
146 }
147
148 static void delayed_put_task_struct(struct rcu_head *rhp)
149 {
150         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
151
152         trace_sched_process_free(tsk);
153         put_task_struct(tsk);
154 }
155
156
157 void release_task(struct task_struct * p)
158 {
159         struct task_struct *leader;
160         int zap_leader;
161 repeat:
162         tracehook_prepare_release_task(p);
163         atomic_dec(&p->cred->user->processes);
164         proc_flush_task(p);
165         write_lock_irq(&tasklist_lock);
166         tracehook_finish_release_task(p);
167         __exit_signal(p);
168
169         /*
170          * If we are the last non-leader member of the thread
171          * group, and the leader is zombie, then notify the
172          * group leader's parent process. (if it wants notification.)
173          */
174         zap_leader = 0;
175         leader = p->group_leader;
176         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
177                 BUG_ON(task_detached(leader));
178                 do_notify_parent(leader, leader->exit_signal);
179                 /*
180                  * If we were the last child thread and the leader has
181                  * exited already, and the leader's parent ignores SIGCHLD,
182                  * then we are the one who should release the leader.
183                  *
184                  * do_notify_parent() will have marked it self-reaping in
185                  * that case.
186                  */
187                 zap_leader = task_detached(leader);
188
189                 /*
190                  * This maintains the invariant that release_task()
191                  * only runs on a task in EXIT_DEAD, just for sanity.
192                  */
193                 if (zap_leader)
194                         leader->exit_state = EXIT_DEAD;
195         }
196
197         write_unlock_irq(&tasklist_lock);
198         release_thread(p);
199         call_rcu(&p->rcu, delayed_put_task_struct);
200
201         p = leader;
202         if (unlikely(zap_leader))
203                 goto repeat;
204 }
205
206 /*
207  * This checks not only the pgrp, but falls back on the pid if no
208  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
209  * without this...
210  *
211  * The caller must hold rcu lock or the tasklist lock.
212  */
213 struct pid *session_of_pgrp(struct pid *pgrp)
214 {
215         struct task_struct *p;
216         struct pid *sid = NULL;
217
218         p = pid_task(pgrp, PIDTYPE_PGID);
219         if (p == NULL)
220                 p = pid_task(pgrp, PIDTYPE_PID);
221         if (p != NULL)
222                 sid = task_session(p);
223
224         return sid;
225 }
226
227 /*
228  * Determine if a process group is "orphaned", according to the POSIX
229  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
230  * by terminal-generated stop signals.  Newly orphaned process groups are
231  * to receive a SIGHUP and a SIGCONT.
232  *
233  * "I ask you, have you ever known what it is to be an orphan?"
234  */
235 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
236 {
237         struct task_struct *p;
238
239         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
240                 if ((p == ignored_task) ||
241                     (p->exit_state && thread_group_empty(p)) ||
242                     is_global_init(p->real_parent))
243                         continue;
244
245                 if (task_pgrp(p->real_parent) != pgrp &&
246                     task_session(p->real_parent) == task_session(p))
247                         return 0;
248         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
249
250         return 1;
251 }
252
253 int is_current_pgrp_orphaned(void)
254 {
255         int retval;
256
257         read_lock(&tasklist_lock);
258         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
259         read_unlock(&tasklist_lock);
260
261         return retval;
262 }
263
264 static int has_stopped_jobs(struct pid *pgrp)
265 {
266         int retval = 0;
267         struct task_struct *p;
268
269         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
270                 if (!task_is_stopped(p))
271                         continue;
272                 retval = 1;
273                 break;
274         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
275         return retval;
276 }
277
278 /*
279  * Check to see if any process groups have become orphaned as
280  * a result of our exiting, and if they have any stopped jobs,
281  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
282  */
283 static void
284 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
285 {
286         struct pid *pgrp = task_pgrp(tsk);
287         struct task_struct *ignored_task = tsk;
288
289         if (!parent)
290                  /* exit: our father is in a different pgrp than
291                   * we are and we were the only connection outside.
292                   */
293                 parent = tsk->real_parent;
294         else
295                 /* reparent: our child is in a different pgrp than
296                  * we are, and it was the only connection outside.
297                  */
298                 ignored_task = NULL;
299
300         if (task_pgrp(parent) != pgrp &&
301             task_session(parent) == task_session(tsk) &&
302             will_become_orphaned_pgrp(pgrp, ignored_task) &&
303             has_stopped_jobs(pgrp)) {
304                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
305                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
306         }
307 }
308
309 /**
310  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
311  *
312  * If a kernel thread is launched as a result of a system call, or if
313  * it ever exits, it should generally reparent itself to kthreadd so it
314  * isn't in the way of other processes and is correctly cleaned up on exit.
315  *
316  * The various task state such as scheduling policy and priority may have
317  * been inherited from a user process, so we reset them to sane values here.
318  *
319  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
320  */
321 static void reparent_to_kthreadd(void)
322 {
323         write_lock_irq(&tasklist_lock);
324
325         ptrace_unlink(current);
326         /* Reparent to init */
327         current->real_parent = current->parent = kthreadd_task;
328         list_move_tail(&current->sibling, &current->real_parent->children);
329
330         /* Set the exit signal to SIGCHLD so we signal init on exit */
331         current->exit_signal = SIGCHLD;
332
333         if (task_nice(current) < 0)
334                 set_user_nice(current, 0);
335         /* cpus_allowed? */
336         /* rt_priority? */
337         /* signals? */
338         security_task_reparent_to_init(current);
339         memcpy(current->signal->rlim, init_task.signal->rlim,
340                sizeof(current->signal->rlim));
341         atomic_inc(&(INIT_USER->__count));
342         write_unlock_irq(&tasklist_lock);
343         switch_uid(INIT_USER);
344 }
345
346 void __set_special_pids(struct pid *pid)
347 {
348         struct task_struct *curr = current->group_leader;
349         pid_t nr = pid_nr(pid);
350
351         if (task_session(curr) != pid) {
352                 change_pid(curr, PIDTYPE_SID, pid);
353                 set_task_session(curr, nr);
354         }
355         if (task_pgrp(curr) != pid) {
356                 change_pid(curr, PIDTYPE_PGID, pid);
357                 set_task_pgrp(curr, nr);
358         }
359 }
360
361 static void set_special_pids(struct pid *pid)
362 {
363         write_lock_irq(&tasklist_lock);
364         __set_special_pids(pid);
365         write_unlock_irq(&tasklist_lock);
366 }
367
368 /*
369  * Let kernel threads use this to say that they
370  * allow a certain signal (since daemonize() will
371  * have disabled all of them by default).
372  */
373 int allow_signal(int sig)
374 {
375         if (!valid_signal(sig) || sig < 1)
376                 return -EINVAL;
377
378         spin_lock_irq(&current->sighand->siglock);
379         sigdelset(&current->blocked, sig);
380         if (!current->mm) {
381                 /* Kernel threads handle their own signals.
382                    Let the signal code know it'll be handled, so
383                    that they don't get converted to SIGKILL or
384                    just silently dropped */
385                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
386         }
387         recalc_sigpending();
388         spin_unlock_irq(&current->sighand->siglock);
389         return 0;
390 }
391
392 EXPORT_SYMBOL(allow_signal);
393
394 int disallow_signal(int sig)
395 {
396         if (!valid_signal(sig) || sig < 1)
397                 return -EINVAL;
398
399         spin_lock_irq(&current->sighand->siglock);
400         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
401         recalc_sigpending();
402         spin_unlock_irq(&current->sighand->siglock);
403         return 0;
404 }
405
406 EXPORT_SYMBOL(disallow_signal);
407
408 /*
409  *      Put all the gunge required to become a kernel thread without
410  *      attached user resources in one place where it belongs.
411  */
412
413 void daemonize(const char *name, ...)
414 {
415         va_list args;
416         struct fs_struct *fs;
417         sigset_t blocked;
418
419         va_start(args, name);
420         vsnprintf(current->comm, sizeof(current->comm), name, args);
421         va_end(args);
422
423         /*
424          * If we were started as result of loading a module, close all of the
425          * user space pages.  We don't need them, and if we didn't close them
426          * they would be locked into memory.
427          */
428         exit_mm(current);
429         /*
430          * We don't want to have TIF_FREEZE set if the system-wide hibernation
431          * or suspend transition begins right now.
432          */
433         current->flags |= (PF_NOFREEZE | PF_KTHREAD);
434
435         if (current->nsproxy != &init_nsproxy) {
436                 get_nsproxy(&init_nsproxy);
437                 switch_task_namespaces(current, &init_nsproxy);
438         }
439         set_special_pids(&init_struct_pid);
440         proc_clear_tty(current);
441
442         /* Block and flush all signals */
443         sigfillset(&blocked);
444         sigprocmask(SIG_BLOCK, &blocked, NULL);
445         flush_signals(current);
446
447         /* Become as one with the init task */
448
449         exit_fs(current);       /* current->fs->count--; */
450         fs = init_task.fs;
451         current->fs = fs;
452         atomic_inc(&fs->count);
453
454         exit_files(current);
455         current->files = init_task.files;
456         atomic_inc(&current->files->count);
457
458         reparent_to_kthreadd();
459 }
460
461 EXPORT_SYMBOL(daemonize);
462
463 static void close_files(struct files_struct * files)
464 {
465         int i, j;
466         struct fdtable *fdt;
467
468         j = 0;
469
470         /*
471          * It is safe to dereference the fd table without RCU or
472          * ->file_lock because this is the last reference to the
473          * files structure.
474          */
475         fdt = files_fdtable(files);
476         for (;;) {
477                 unsigned long set;
478                 i = j * __NFDBITS;
479                 if (i >= fdt->max_fds)
480                         break;
481                 set = fdt->open_fds->fds_bits[j++];
482                 while (set) {
483                         if (set & 1) {
484                                 struct file * file = xchg(&fdt->fd[i], NULL);
485                                 if (file) {
486                                         filp_close(file, files);
487                                         cond_resched();
488                                 }
489                         }
490                         i++;
491                         set >>= 1;
492                 }
493         }
494 }
495
496 struct files_struct *get_files_struct(struct task_struct *task)
497 {
498         struct files_struct *files;
499
500         task_lock(task);
501         files = task->files;
502         if (files)
503                 atomic_inc(&files->count);
504         task_unlock(task);
505
506         return files;
507 }
508
509 void put_files_struct(struct files_struct *files)
510 {
511         struct fdtable *fdt;
512
513         if (atomic_dec_and_test(&files->count)) {
514                 close_files(files);
515                 /*
516                  * Free the fd and fdset arrays if we expanded them.
517                  * If the fdtable was embedded, pass files for freeing
518                  * at the end of the RCU grace period. Otherwise,
519                  * you can free files immediately.
520                  */
521                 fdt = files_fdtable(files);
522                 if (fdt != &files->fdtab)
523                         kmem_cache_free(files_cachep, files);
524                 free_fdtable(fdt);
525         }
526 }
527
528 void reset_files_struct(struct files_struct *files)
529 {
530         struct task_struct *tsk = current;
531         struct files_struct *old;
532
533         old = tsk->files;
534         task_lock(tsk);
535         tsk->files = files;
536         task_unlock(tsk);
537         put_files_struct(old);
538 }
539
540 void exit_files(struct task_struct *tsk)
541 {
542         struct files_struct * files = tsk->files;
543
544         if (files) {
545                 task_lock(tsk);
546                 tsk->files = NULL;
547                 task_unlock(tsk);
548                 put_files_struct(files);
549         }
550 }
551
552 void put_fs_struct(struct fs_struct *fs)
553 {
554         /* No need to hold fs->lock if we are killing it */
555         if (atomic_dec_and_test(&fs->count)) {
556                 path_put(&fs->root);
557                 path_put(&fs->pwd);
558                 kmem_cache_free(fs_cachep, fs);
559         }
560 }
561
562 void exit_fs(struct task_struct *tsk)
563 {
564         struct fs_struct * fs = tsk->fs;
565
566         if (fs) {
567                 task_lock(tsk);
568                 tsk->fs = NULL;
569                 task_unlock(tsk);
570                 put_fs_struct(fs);
571         }
572 }
573
574 EXPORT_SYMBOL_GPL(exit_fs);
575
576 #ifdef CONFIG_MM_OWNER
577 /*
578  * Task p is exiting and it owned mm, lets find a new owner for it
579  */
580 static inline int
581 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
582 {
583         /*
584          * If there are other users of the mm and the owner (us) is exiting
585          * we need to find a new owner to take on the responsibility.
586          */
587         if (atomic_read(&mm->mm_users) <= 1)
588                 return 0;
589         if (mm->owner != p)
590                 return 0;
591         return 1;
592 }
593
594 void mm_update_next_owner(struct mm_struct *mm)
595 {
596         struct task_struct *c, *g, *p = current;
597
598 retry:
599         if (!mm_need_new_owner(mm, p))
600                 return;
601
602         read_lock(&tasklist_lock);
603         /*
604          * Search in the children
605          */
606         list_for_each_entry(c, &p->children, sibling) {
607                 if (c->mm == mm)
608                         goto assign_new_owner;
609         }
610
611         /*
612          * Search in the siblings
613          */
614         list_for_each_entry(c, &p->parent->children, sibling) {
615                 if (c->mm == mm)
616                         goto assign_new_owner;
617         }
618
619         /*
620          * Search through everything else. We should not get
621          * here often
622          */
623         do_each_thread(g, c) {
624                 if (c->mm == mm)
625                         goto assign_new_owner;
626         } while_each_thread(g, c);
627
628         read_unlock(&tasklist_lock);
629         /*
630          * We found no owner yet mm_users > 1: this implies that we are
631          * most likely racing with swapoff (try_to_unuse()) or /proc or
632          * ptrace or page migration (get_task_mm()).  Mark owner as NULL,
633          * so that subsystems can understand the callback and take action.
634          */
635         down_write(&mm->mmap_sem);
636         cgroup_mm_owner_callbacks(mm->owner, NULL);
637         mm->owner = NULL;
638         up_write(&mm->mmap_sem);
639         return;
640
641 assign_new_owner:
642         BUG_ON(c == p);
643         get_task_struct(c);
644         read_unlock(&tasklist_lock);
645         down_write(&mm->mmap_sem);
646         /*
647          * The task_lock protects c->mm from changing.
648          * We always want mm->owner->mm == mm
649          */
650         task_lock(c);
651         if (c->mm != mm) {
652                 task_unlock(c);
653                 up_write(&mm->mmap_sem);
654                 put_task_struct(c);
655                 goto retry;
656         }
657         cgroup_mm_owner_callbacks(mm->owner, c);
658         mm->owner = c;
659         task_unlock(c);
660         up_write(&mm->mmap_sem);
661         put_task_struct(c);
662 }
663 #endif /* CONFIG_MM_OWNER */
664
665 /*
666  * Turn us into a lazy TLB process if we
667  * aren't already..
668  */
669 static void exit_mm(struct task_struct * tsk)
670 {
671         struct mm_struct *mm = tsk->mm;
672         struct core_state *core_state;
673
674         mm_release(tsk, mm);
675         if (!mm)
676                 return;
677         /*
678          * Serialize with any possible pending coredump.
679          * We must hold mmap_sem around checking core_state
680          * and clearing tsk->mm.  The core-inducing thread
681          * will increment ->nr_threads for each thread in the
682          * group with ->mm != NULL.
683          */
684         down_read(&mm->mmap_sem);
685         core_state = mm->core_state;
686         if (core_state) {
687                 struct core_thread self;
688                 up_read(&mm->mmap_sem);
689
690                 self.task = tsk;
691                 self.next = xchg(&core_state->dumper.next, &self);
692                 /*
693                  * Implies mb(), the result of xchg() must be visible
694                  * to core_state->dumper.
695                  */
696                 if (atomic_dec_and_test(&core_state->nr_threads))
697                         complete(&core_state->startup);
698
699                 for (;;) {
700                         set_task_state(tsk, TASK_UNINTERRUPTIBLE);
701                         if (!self.task) /* see coredump_finish() */
702                                 break;
703                         schedule();
704                 }
705                 __set_task_state(tsk, TASK_RUNNING);
706                 down_read(&mm->mmap_sem);
707         }
708         atomic_inc(&mm->mm_count);
709         BUG_ON(mm != tsk->active_mm);
710         /* more a memory barrier than a real lock */
711         task_lock(tsk);
712         tsk->mm = NULL;
713         up_read(&mm->mmap_sem);
714         enter_lazy_tlb(mm, current);
715         /* We don't want this task to be frozen prematurely */
716         clear_freeze_flag(tsk);
717         task_unlock(tsk);
718         mm_update_next_owner(mm);
719         mmput(mm);
720 }
721
722 /*
723  * Return nonzero if @parent's children should reap themselves.
724  *
725  * Called with write_lock_irq(&tasklist_lock) held.
726  */
727 static int ignoring_children(struct task_struct *parent)
728 {
729         int ret;
730         struct sighand_struct *psig = parent->sighand;
731         unsigned long flags;
732         spin_lock_irqsave(&psig->siglock, flags);
733         ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
734                (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
735         spin_unlock_irqrestore(&psig->siglock, flags);
736         return ret;
737 }
738
739 /*
740  * Detach all tasks we were using ptrace on.
741  * Any that need to be release_task'd are put on the @dead list.
742  *
743  * Called with write_lock(&tasklist_lock) held.
744  */
745 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
746 {
747         struct task_struct *p, *n;
748         int ign = -1;
749
750         list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
751                 __ptrace_unlink(p);
752
753                 if (p->exit_state != EXIT_ZOMBIE)
754                         continue;
755
756                 /*
757                  * If it's a zombie, our attachedness prevented normal
758                  * parent notification or self-reaping.  Do notification
759                  * now if it would have happened earlier.  If it should
760                  * reap itself, add it to the @dead list.  We can't call
761                  * release_task() here because we already hold tasklist_lock.
762                  *
763                  * If it's our own child, there is no notification to do.
764                  * But if our normal children self-reap, then this child
765                  * was prevented by ptrace and we must reap it now.
766                  */
767                 if (!task_detached(p) && thread_group_empty(p)) {
768                         if (!same_thread_group(p->real_parent, parent))
769                                 do_notify_parent(p, p->exit_signal);
770                         else {
771                                 if (ign < 0)
772                                         ign = ignoring_children(parent);
773                                 if (ign)
774                                         p->exit_signal = -1;
775                         }
776                 }
777
778                 if (task_detached(p)) {
779                         /*
780                          * Mark it as in the process of being reaped.
781                          */
782                         p->exit_state = EXIT_DEAD;
783                         list_add(&p->ptrace_entry, dead);
784                 }
785         }
786 }
787
788 /*
789  * Finish up exit-time ptrace cleanup.
790  *
791  * Called without locks.
792  */
793 static void ptrace_exit_finish(struct task_struct *parent,
794                                struct list_head *dead)
795 {
796         struct task_struct *p, *n;
797
798         BUG_ON(!list_empty(&parent->ptraced));
799
800         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
801                 list_del_init(&p->ptrace_entry);
802                 release_task(p);
803         }
804 }
805
806 static void reparent_thread(struct task_struct *p, struct task_struct *father)
807 {
808         if (p->pdeath_signal)
809                 /* We already hold the tasklist_lock here.  */
810                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
811
812         list_move_tail(&p->sibling, &p->real_parent->children);
813
814         /* If this is a threaded reparent there is no need to
815          * notify anyone anything has happened.
816          */
817         if (same_thread_group(p->real_parent, father))
818                 return;
819
820         /* We don't want people slaying init.  */
821         if (!task_detached(p))
822                 p->exit_signal = SIGCHLD;
823
824         /* If we'd notified the old parent about this child's death,
825          * also notify the new parent.
826          */
827         if (!ptrace_reparented(p) &&
828             p->exit_state == EXIT_ZOMBIE &&
829             !task_detached(p) && thread_group_empty(p))
830                 do_notify_parent(p, p->exit_signal);
831
832         kill_orphaned_pgrp(p, father);
833 }
834
835 /*
836  * When we die, we re-parent all our children.
837  * Try to give them to another thread in our thread
838  * group, and if no such member exists, give it to
839  * the child reaper process (ie "init") in our pid
840  * space.
841  */
842 static struct task_struct *find_new_reaper(struct task_struct *father)
843 {
844         struct pid_namespace *pid_ns = task_active_pid_ns(father);
845         struct task_struct *thread;
846
847         thread = father;
848         while_each_thread(father, thread) {
849                 if (thread->flags & PF_EXITING)
850                         continue;
851                 if (unlikely(pid_ns->child_reaper == father))
852                         pid_ns->child_reaper = thread;
853                 return thread;
854         }
855
856         if (unlikely(pid_ns->child_reaper == father)) {
857                 write_unlock_irq(&tasklist_lock);
858                 if (unlikely(pid_ns == &init_pid_ns))
859                         panic("Attempted to kill init!");
860
861                 zap_pid_ns_processes(pid_ns);
862                 write_lock_irq(&tasklist_lock);
863                 /*
864                  * We can not clear ->child_reaper or leave it alone.
865                  * There may by stealth EXIT_DEAD tasks on ->children,
866                  * forget_original_parent() must move them somewhere.
867                  */
868                 pid_ns->child_reaper = init_pid_ns.child_reaper;
869         }
870
871         return pid_ns->child_reaper;
872 }
873
874 static void forget_original_parent(struct task_struct *father)
875 {
876         struct task_struct *p, *n, *reaper;
877         LIST_HEAD(ptrace_dead);
878
879         write_lock_irq(&tasklist_lock);
880         reaper = find_new_reaper(father);
881         /*
882          * First clean up ptrace if we were using it.
883          */
884         ptrace_exit(father, &ptrace_dead);
885
886         list_for_each_entry_safe(p, n, &father->children, sibling) {
887                 p->real_parent = reaper;
888                 if (p->parent == father) {
889                         BUG_ON(p->ptrace);
890                         p->parent = p->real_parent;
891                 }
892                 reparent_thread(p, father);
893         }
894
895         write_unlock_irq(&tasklist_lock);
896         BUG_ON(!list_empty(&father->children));
897
898         ptrace_exit_finish(father, &ptrace_dead);
899 }
900
901 /*
902  * Send signals to all our closest relatives so that they know
903  * to properly mourn us..
904  */
905 static void exit_notify(struct task_struct *tsk, int group_dead)
906 {
907         int signal;
908         void *cookie;
909
910         /*
911          * This does two things:
912          *
913          * A.  Make init inherit all the child processes
914          * B.  Check to see if any process groups have become orphaned
915          *      as a result of our exiting, and if they have any stopped
916          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
917          */
918         forget_original_parent(tsk);
919         exit_task_namespaces(tsk);
920
921         write_lock_irq(&tasklist_lock);
922         if (group_dead)
923                 kill_orphaned_pgrp(tsk->group_leader, NULL);
924
925         /* Let father know we died
926          *
927          * Thread signals are configurable, but you aren't going to use
928          * that to send signals to arbitary processes.
929          * That stops right now.
930          *
931          * If the parent exec id doesn't match the exec id we saved
932          * when we started then we know the parent has changed security
933          * domain.
934          *
935          * If our self_exec id doesn't match our parent_exec_id then
936          * we have changed execution domain as these two values started
937          * the same after a fork.
938          */
939         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
940             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
941              tsk->self_exec_id != tsk->parent_exec_id) &&
942             !capable(CAP_KILL))
943                 tsk->exit_signal = SIGCHLD;
944
945         signal = tracehook_notify_death(tsk, &cookie, group_dead);
946         if (signal >= 0)
947                 signal = do_notify_parent(tsk, signal);
948
949         tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
950
951         /* mt-exec, de_thread() is waiting for us */
952         if (thread_group_leader(tsk) &&
953             tsk->signal->group_exit_task &&
954             tsk->signal->notify_count < 0)
955                 wake_up_process(tsk->signal->group_exit_task);
956
957         write_unlock_irq(&tasklist_lock);
958
959         tracehook_report_death(tsk, signal, cookie, group_dead);
960
961         /* If the process is dead, release it - nobody will wait for it */
962         if (signal == DEATH_REAP)
963                 release_task(tsk);
964 }
965
966 #ifdef CONFIG_DEBUG_STACK_USAGE
967 static void check_stack_usage(void)
968 {
969         static DEFINE_SPINLOCK(low_water_lock);
970         static int lowest_to_date = THREAD_SIZE;
971         unsigned long *n = end_of_stack(current);
972         unsigned long free;
973
974         while (*n == 0)
975                 n++;
976         free = (unsigned long)n - (unsigned long)end_of_stack(current);
977
978         if (free >= lowest_to_date)
979                 return;
980
981         spin_lock(&low_water_lock);
982         if (free < lowest_to_date) {
983                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
984                                 "left\n",
985                                 current->comm, free);
986                 lowest_to_date = free;
987         }
988         spin_unlock(&low_water_lock);
989 }
990 #else
991 static inline void check_stack_usage(void) {}
992 #endif
993
994 NORET_TYPE void do_exit(long code)
995 {
996         struct task_struct *tsk = current;
997         int group_dead;
998
999         profile_task_exit(tsk);
1000
1001         WARN_ON(atomic_read(&tsk->fs_excl));
1002
1003         if (unlikely(in_interrupt()))
1004                 panic("Aiee, killing interrupt handler!");
1005         if (unlikely(!tsk->pid))
1006                 panic("Attempted to kill the idle task!");
1007
1008         tracehook_report_exit(&code);
1009
1010         /*
1011          * We're taking recursive faults here in do_exit. Safest is to just
1012          * leave this task alone and wait for reboot.
1013          */
1014         if (unlikely(tsk->flags & PF_EXITING)) {
1015                 printk(KERN_ALERT
1016                         "Fixing recursive fault but reboot is needed!\n");
1017                 /*
1018                  * We can do this unlocked here. The futex code uses
1019                  * this flag just to verify whether the pi state
1020                  * cleanup has been done or not. In the worst case it
1021                  * loops once more. We pretend that the cleanup was
1022                  * done as there is no way to return. Either the
1023                  * OWNER_DIED bit is set by now or we push the blocked
1024                  * task into the wait for ever nirwana as well.
1025                  */
1026                 tsk->flags |= PF_EXITPIDONE;
1027                 if (tsk->io_context)
1028                         exit_io_context();
1029                 set_current_state(TASK_UNINTERRUPTIBLE);
1030                 schedule();
1031         }
1032
1033         exit_signals(tsk);  /* sets PF_EXITING */
1034         /*
1035          * tsk->flags are checked in the futex code to protect against
1036          * an exiting task cleaning up the robust pi futexes.
1037          */
1038         smp_mb();
1039         spin_unlock_wait(&tsk->pi_lock);
1040
1041         if (unlikely(in_atomic()))
1042                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1043                                 current->comm, task_pid_nr(current),
1044                                 preempt_count());
1045
1046         acct_update_integrals(tsk);
1047         if (tsk->mm) {
1048                 update_hiwater_rss(tsk->mm);
1049                 update_hiwater_vm(tsk->mm);
1050         }
1051         group_dead = atomic_dec_and_test(&tsk->signal->live);
1052         if (group_dead) {
1053                 hrtimer_cancel(&tsk->signal->real_timer);
1054                 exit_itimers(tsk->signal);
1055         }
1056         acct_collect(code, group_dead);
1057 #ifdef CONFIG_FUTEX
1058         if (unlikely(tsk->robust_list))
1059                 exit_robust_list(tsk);
1060 #ifdef CONFIG_COMPAT
1061         if (unlikely(tsk->compat_robust_list))
1062                 compat_exit_robust_list(tsk);
1063 #endif
1064 #endif
1065         if (group_dead)
1066                 tty_audit_exit();
1067         if (unlikely(tsk->audit_context))
1068                 audit_free(tsk);
1069
1070         tsk->exit_code = code;
1071         taskstats_exit(tsk, group_dead);
1072
1073         exit_mm(tsk);
1074
1075         if (group_dead)
1076                 acct_process();
1077         trace_sched_process_exit(tsk);
1078
1079         exit_sem(tsk);
1080         exit_files(tsk);
1081         exit_fs(tsk);
1082         check_stack_usage();
1083         exit_thread();
1084         cgroup_exit(tsk, 1);
1085         exit_keys(tsk);
1086
1087         if (group_dead && tsk->signal->leader)
1088                 disassociate_ctty(1);
1089
1090         module_put(task_thread_info(tsk)->exec_domain->module);
1091         if (tsk->binfmt)
1092                 module_put(tsk->binfmt->module);
1093
1094         proc_exit_connector(tsk);
1095         exit_notify(tsk, group_dead);
1096 #ifdef CONFIG_NUMA
1097         mpol_put(tsk->mempolicy);
1098         tsk->mempolicy = NULL;
1099 #endif
1100 #ifdef CONFIG_FUTEX
1101         /*
1102          * This must happen late, after the PID is not
1103          * hashed anymore:
1104          */
1105         if (unlikely(!list_empty(&tsk->pi_state_list)))
1106                 exit_pi_state_list(tsk);
1107         if (unlikely(current->pi_state_cache))
1108                 kfree(current->pi_state_cache);
1109 #endif
1110         /*
1111          * Make sure we are holding no locks:
1112          */
1113         debug_check_no_locks_held(tsk);
1114         /*
1115          * We can do this unlocked here. The futex code uses this flag
1116          * just to verify whether the pi state cleanup has been done
1117          * or not. In the worst case it loops once more.
1118          */
1119         tsk->flags |= PF_EXITPIDONE;
1120
1121         if (tsk->io_context)
1122                 exit_io_context();
1123
1124         if (tsk->splice_pipe)
1125                 __free_pipe_info(tsk->splice_pipe);
1126
1127         preempt_disable();
1128         /* causes final put_task_struct in finish_task_switch(). */
1129         tsk->state = TASK_DEAD;
1130
1131         schedule();
1132         BUG();
1133         /* Avoid "noreturn function does return".  */
1134         for (;;)
1135                 cpu_relax();    /* For when BUG is null */
1136 }
1137
1138 EXPORT_SYMBOL_GPL(do_exit);
1139
1140 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1141 {
1142         if (comp)
1143                 complete(comp);
1144
1145         do_exit(code);
1146 }
1147
1148 EXPORT_SYMBOL(complete_and_exit);
1149
1150 asmlinkage long sys_exit(int error_code)
1151 {
1152         do_exit((error_code&0xff)<<8);
1153 }
1154
1155 /*
1156  * Take down every thread in the group.  This is called by fatal signals
1157  * as well as by sys_exit_group (below).
1158  */
1159 NORET_TYPE void
1160 do_group_exit(int exit_code)
1161 {
1162         struct signal_struct *sig = current->signal;
1163
1164         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1165
1166         if (signal_group_exit(sig))
1167                 exit_code = sig->group_exit_code;
1168         else if (!thread_group_empty(current)) {
1169                 struct sighand_struct *const sighand = current->sighand;
1170                 spin_lock_irq(&sighand->siglock);
1171                 if (signal_group_exit(sig))
1172                         /* Another thread got here before we took the lock.  */
1173                         exit_code = sig->group_exit_code;
1174                 else {
1175                         sig->group_exit_code = exit_code;
1176                         sig->flags = SIGNAL_GROUP_EXIT;
1177                         zap_other_threads(current);
1178                 }
1179                 spin_unlock_irq(&sighand->siglock);
1180         }
1181
1182         do_exit(exit_code);
1183         /* NOTREACHED */
1184 }
1185
1186 /*
1187  * this kills every thread in the thread group. Note that any externally
1188  * wait4()-ing process will get the correct exit code - even if this
1189  * thread is not the thread group leader.
1190  */
1191 asmlinkage void sys_exit_group(int error_code)
1192 {
1193         do_group_exit((error_code & 0xff) << 8);
1194 }
1195
1196 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1197 {
1198         struct pid *pid = NULL;
1199         if (type == PIDTYPE_PID)
1200                 pid = task->pids[type].pid;
1201         else if (type < PIDTYPE_MAX)
1202                 pid = task->group_leader->pids[type].pid;
1203         return pid;
1204 }
1205
1206 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1207                           struct task_struct *p)
1208 {
1209         int err;
1210
1211         if (type < PIDTYPE_MAX) {
1212                 if (task_pid_type(p, type) != pid)
1213                         return 0;
1214         }
1215
1216         /* Wait for all children (clone and not) if __WALL is set;
1217          * otherwise, wait for clone children *only* if __WCLONE is
1218          * set; otherwise, wait for non-clone children *only*.  (Note:
1219          * A "clone" child here is one that reports to its parent
1220          * using a signal other than SIGCHLD.) */
1221         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1222             && !(options & __WALL))
1223                 return 0;
1224
1225         err = security_task_wait(p);
1226         if (err)
1227                 return err;
1228
1229         return 1;
1230 }
1231
1232 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1233                                int why, int status,
1234                                struct siginfo __user *infop,
1235                                struct rusage __user *rusagep)
1236 {
1237         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1238
1239         put_task_struct(p);
1240         if (!retval)
1241                 retval = put_user(SIGCHLD, &infop->si_signo);
1242         if (!retval)
1243                 retval = put_user(0, &infop->si_errno);
1244         if (!retval)
1245                 retval = put_user((short)why, &infop->si_code);
1246         if (!retval)
1247                 retval = put_user(pid, &infop->si_pid);
1248         if (!retval)
1249                 retval = put_user(uid, &infop->si_uid);
1250         if (!retval)
1251                 retval = put_user(status, &infop->si_status);
1252         if (!retval)
1253                 retval = pid;
1254         return retval;
1255 }
1256
1257 /*
1258  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1259  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1260  * the lock and this task is uninteresting.  If we return nonzero, we have
1261  * released the lock and the system call should return.
1262  */
1263 static int wait_task_zombie(struct task_struct *p, int options,
1264                             struct siginfo __user *infop,
1265                             int __user *stat_addr, struct rusage __user *ru)
1266 {
1267         unsigned long state;
1268         int retval, status, traced;
1269         pid_t pid = task_pid_vnr(p);
1270
1271         if (!likely(options & WEXITED))
1272                 return 0;
1273
1274         if (unlikely(options & WNOWAIT)) {
1275                 uid_t uid = p->cred->uid;
1276                 int exit_code = p->exit_code;
1277                 int why, status;
1278
1279                 get_task_struct(p);
1280                 read_unlock(&tasklist_lock);
1281                 if ((exit_code & 0x7f) == 0) {
1282                         why = CLD_EXITED;
1283                         status = exit_code >> 8;
1284                 } else {
1285                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1286                         status = exit_code & 0x7f;
1287                 }
1288                 return wait_noreap_copyout(p, pid, uid, why,
1289                                            status, infop, ru);
1290         }
1291
1292         /*
1293          * Try to move the task's state to DEAD
1294          * only one thread is allowed to do this:
1295          */
1296         state = xchg(&p->exit_state, EXIT_DEAD);
1297         if (state != EXIT_ZOMBIE) {
1298                 BUG_ON(state != EXIT_DEAD);
1299                 return 0;
1300         }
1301
1302         traced = ptrace_reparented(p);
1303
1304         if (likely(!traced)) {
1305                 struct signal_struct *psig;
1306                 struct signal_struct *sig;
1307                 struct task_cputime cputime;
1308
1309                 /*
1310                  * The resource counters for the group leader are in its
1311                  * own task_struct.  Those for dead threads in the group
1312                  * are in its signal_struct, as are those for the child
1313                  * processes it has previously reaped.  All these
1314                  * accumulate in the parent's signal_struct c* fields.
1315                  *
1316                  * We don't bother to take a lock here to protect these
1317                  * p->signal fields, because they are only touched by
1318                  * __exit_signal, which runs with tasklist_lock
1319                  * write-locked anyway, and so is excluded here.  We do
1320                  * need to protect the access to p->parent->signal fields,
1321                  * as other threads in the parent group can be right
1322                  * here reaping other children at the same time.
1323                  *
1324                  * We use thread_group_cputime() to get times for the thread
1325                  * group, which consolidates times for all threads in the
1326                  * group including the group leader.
1327                  */
1328                 spin_lock_irq(&p->parent->sighand->siglock);
1329                 psig = p->parent->signal;
1330                 sig = p->signal;
1331                 thread_group_cputime(p, &cputime);
1332                 psig->cutime =
1333                         cputime_add(psig->cutime,
1334                         cputime_add(cputime.utime,
1335                                     sig->cutime));
1336                 psig->cstime =
1337                         cputime_add(psig->cstime,
1338                         cputime_add(cputime.stime,
1339                                     sig->cstime));
1340                 psig->cgtime =
1341                         cputime_add(psig->cgtime,
1342                         cputime_add(p->gtime,
1343                         cputime_add(sig->gtime,
1344                                     sig->cgtime)));
1345                 psig->cmin_flt +=
1346                         p->min_flt + sig->min_flt + sig->cmin_flt;
1347                 psig->cmaj_flt +=
1348                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1349                 psig->cnvcsw +=
1350                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1351                 psig->cnivcsw +=
1352                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1353                 psig->cinblock +=
1354                         task_io_get_inblock(p) +
1355                         sig->inblock + sig->cinblock;
1356                 psig->coublock +=
1357                         task_io_get_oublock(p) +
1358                         sig->oublock + sig->coublock;
1359                 task_io_accounting_add(&psig->ioac, &p->ioac);
1360                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1361                 spin_unlock_irq(&p->parent->sighand->siglock);
1362         }
1363
1364         /*
1365          * Now we are sure this task is interesting, and no other
1366          * thread can reap it because we set its state to EXIT_DEAD.
1367          */
1368         read_unlock(&tasklist_lock);
1369
1370         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1371         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1372                 ? p->signal->group_exit_code : p->exit_code;
1373         if (!retval && stat_addr)
1374                 retval = put_user(status, stat_addr);
1375         if (!retval && infop)
1376                 retval = put_user(SIGCHLD, &infop->si_signo);
1377         if (!retval && infop)
1378                 retval = put_user(0, &infop->si_errno);
1379         if (!retval && infop) {
1380                 int why;
1381
1382                 if ((status & 0x7f) == 0) {
1383                         why = CLD_EXITED;
1384                         status >>= 8;
1385                 } else {
1386                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1387                         status &= 0x7f;
1388                 }
1389                 retval = put_user((short)why, &infop->si_code);
1390                 if (!retval)
1391                         retval = put_user(status, &infop->si_status);
1392         }
1393         if (!retval && infop)
1394                 retval = put_user(pid, &infop->si_pid);
1395         if (!retval && infop)
1396                 retval = put_user(p->cred->uid, &infop->si_uid);
1397         if (!retval)
1398                 retval = pid;
1399
1400         if (traced) {
1401                 write_lock_irq(&tasklist_lock);
1402                 /* We dropped tasklist, ptracer could die and untrace */
1403                 ptrace_unlink(p);
1404                 /*
1405                  * If this is not a detached task, notify the parent.
1406                  * If it's still not detached after that, don't release
1407                  * it now.
1408                  */
1409                 if (!task_detached(p)) {
1410                         do_notify_parent(p, p->exit_signal);
1411                         if (!task_detached(p)) {
1412                                 p->exit_state = EXIT_ZOMBIE;
1413                                 p = NULL;
1414                         }
1415                 }
1416                 write_unlock_irq(&tasklist_lock);
1417         }
1418         if (p != NULL)
1419                 release_task(p);
1420
1421         return retval;
1422 }
1423
1424 /*
1425  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1426  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1427  * the lock and this task is uninteresting.  If we return nonzero, we have
1428  * released the lock and the system call should return.
1429  */
1430 static int wait_task_stopped(int ptrace, struct task_struct *p,
1431                              int options, struct siginfo __user *infop,
1432                              int __user *stat_addr, struct rusage __user *ru)
1433 {
1434         int retval, exit_code, why;
1435         uid_t uid = 0; /* unneeded, required by compiler */
1436         pid_t pid;
1437
1438         if (!(options & WUNTRACED))
1439                 return 0;
1440
1441         exit_code = 0;
1442         spin_lock_irq(&p->sighand->siglock);
1443
1444         if (unlikely(!task_is_stopped_or_traced(p)))
1445                 goto unlock_sig;
1446
1447         if (!ptrace && p->signal->group_stop_count > 0)
1448                 /*
1449                  * A group stop is in progress and this is the group leader.
1450                  * We won't report until all threads have stopped.
1451                  */
1452                 goto unlock_sig;
1453
1454         exit_code = p->exit_code;
1455         if (!exit_code)
1456                 goto unlock_sig;
1457
1458         if (!unlikely(options & WNOWAIT))
1459                 p->exit_code = 0;
1460
1461         uid = p->cred->uid;
1462 unlock_sig:
1463         spin_unlock_irq(&p->sighand->siglock);
1464         if (!exit_code)
1465                 return 0;
1466
1467         /*
1468          * Now we are pretty sure this task is interesting.
1469          * Make sure it doesn't get reaped out from under us while we
1470          * give up the lock and then examine it below.  We don't want to
1471          * keep holding onto the tasklist_lock while we call getrusage and
1472          * possibly take page faults for user memory.
1473          */
1474         get_task_struct(p);
1475         pid = task_pid_vnr(p);
1476         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1477         read_unlock(&tasklist_lock);
1478
1479         if (unlikely(options & WNOWAIT))
1480                 return wait_noreap_copyout(p, pid, uid,
1481                                            why, exit_code,
1482                                            infop, ru);
1483
1484         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1485         if (!retval && stat_addr)
1486                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1487         if (!retval && infop)
1488                 retval = put_user(SIGCHLD, &infop->si_signo);
1489         if (!retval && infop)
1490                 retval = put_user(0, &infop->si_errno);
1491         if (!retval && infop)
1492                 retval = put_user((short)why, &infop->si_code);
1493         if (!retval && infop)
1494                 retval = put_user(exit_code, &infop->si_status);
1495         if (!retval && infop)
1496                 retval = put_user(pid, &infop->si_pid);
1497         if (!retval && infop)
1498                 retval = put_user(uid, &infop->si_uid);
1499         if (!retval)
1500                 retval = pid;
1501         put_task_struct(p);
1502
1503         BUG_ON(!retval);
1504         return retval;
1505 }
1506
1507 /*
1508  * Handle do_wait work for one task in a live, non-stopped state.
1509  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1510  * the lock and this task is uninteresting.  If we return nonzero, we have
1511  * released the lock and the system call should return.
1512  */
1513 static int wait_task_continued(struct task_struct *p, int options,
1514                                struct siginfo __user *infop,
1515                                int __user *stat_addr, struct rusage __user *ru)
1516 {
1517         int retval;
1518         pid_t pid;
1519         uid_t uid;
1520
1521         if (!unlikely(options & WCONTINUED))
1522                 return 0;
1523
1524         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1525                 return 0;
1526
1527         spin_lock_irq(&p->sighand->siglock);
1528         /* Re-check with the lock held.  */
1529         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1530                 spin_unlock_irq(&p->sighand->siglock);
1531                 return 0;
1532         }
1533         if (!unlikely(options & WNOWAIT))
1534                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1535         spin_unlock_irq(&p->sighand->siglock);
1536
1537         pid = task_pid_vnr(p);
1538         uid = p->cred->uid;
1539         get_task_struct(p);
1540         read_unlock(&tasklist_lock);
1541
1542         if (!infop) {
1543                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1544                 put_task_struct(p);
1545                 if (!retval && stat_addr)
1546                         retval = put_user(0xffff, stat_addr);
1547                 if (!retval)
1548                         retval = pid;
1549         } else {
1550                 retval = wait_noreap_copyout(p, pid, uid,
1551                                              CLD_CONTINUED, SIGCONT,
1552                                              infop, ru);
1553                 BUG_ON(retval == 0);
1554         }
1555
1556         return retval;
1557 }
1558
1559 /*
1560  * Consider @p for a wait by @parent.
1561  *
1562  * -ECHILD should be in *@notask_error before the first call.
1563  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1564  * Returns zero if the search for a child should continue;
1565  * then *@notask_error is 0 if @p is an eligible child,
1566  * or another error from security_task_wait(), or still -ECHILD.
1567  */
1568 static int wait_consider_task(struct task_struct *parent, int ptrace,
1569                               struct task_struct *p, int *notask_error,
1570                               enum pid_type type, struct pid *pid, int options,
1571                               struct siginfo __user *infop,
1572                               int __user *stat_addr, struct rusage __user *ru)
1573 {
1574         int ret = eligible_child(type, pid, options, p);
1575         if (!ret)
1576                 return ret;
1577
1578         if (unlikely(ret < 0)) {
1579                 /*
1580                  * If we have not yet seen any eligible child,
1581                  * then let this error code replace -ECHILD.
1582                  * A permission error will give the user a clue
1583                  * to look for security policy problems, rather
1584                  * than for mysterious wait bugs.
1585                  */
1586                 if (*notask_error)
1587                         *notask_error = ret;
1588         }
1589
1590         if (likely(!ptrace) && unlikely(p->ptrace)) {
1591                 /*
1592                  * This child is hidden by ptrace.
1593                  * We aren't allowed to see it now, but eventually we will.
1594                  */
1595                 *notask_error = 0;
1596                 return 0;
1597         }
1598
1599         if (p->exit_state == EXIT_DEAD)
1600                 return 0;
1601
1602         /*
1603          * We don't reap group leaders with subthreads.
1604          */
1605         if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1606                 return wait_task_zombie(p, options, infop, stat_addr, ru);
1607
1608         /*
1609          * It's stopped or running now, so it might
1610          * later continue, exit, or stop again.
1611          */
1612         *notask_error = 0;
1613
1614         if (task_is_stopped_or_traced(p))
1615                 return wait_task_stopped(ptrace, p, options,
1616                                          infop, stat_addr, ru);
1617
1618         return wait_task_continued(p, options, infop, stat_addr, ru);
1619 }
1620
1621 /*
1622  * Do the work of do_wait() for one thread in the group, @tsk.
1623  *
1624  * -ECHILD should be in *@notask_error before the first call.
1625  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1626  * Returns zero if the search for a child should continue; then
1627  * *@notask_error is 0 if there were any eligible children,
1628  * or another error from security_task_wait(), or still -ECHILD.
1629  */
1630 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1631                           enum pid_type type, struct pid *pid, int options,
1632                           struct siginfo __user *infop, int __user *stat_addr,
1633                           struct rusage __user *ru)
1634 {
1635         struct task_struct *p;
1636
1637         list_for_each_entry(p, &tsk->children, sibling) {
1638                 /*
1639                  * Do not consider detached threads.
1640                  */
1641                 if (!task_detached(p)) {
1642                         int ret = wait_consider_task(tsk, 0, p, notask_error,
1643                                                      type, pid, options,
1644                                                      infop, stat_addr, ru);
1645                         if (ret)
1646                                 return ret;
1647                 }
1648         }
1649
1650         return 0;
1651 }
1652
1653 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1654                           enum pid_type type, struct pid *pid, int options,
1655                           struct siginfo __user *infop, int __user *stat_addr,
1656                           struct rusage __user *ru)
1657 {
1658         struct task_struct *p;
1659
1660         /*
1661          * Traditionally we see ptrace'd stopped tasks regardless of options.
1662          */
1663         options |= WUNTRACED;
1664
1665         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1666                 int ret = wait_consider_task(tsk, 1, p, notask_error,
1667                                              type, pid, options,
1668                                              infop, stat_addr, ru);
1669                 if (ret)
1670                         return ret;
1671         }
1672
1673         return 0;
1674 }
1675
1676 static long do_wait(enum pid_type type, struct pid *pid, int options,
1677                     struct siginfo __user *infop, int __user *stat_addr,
1678                     struct rusage __user *ru)
1679 {
1680         DECLARE_WAITQUEUE(wait, current);
1681         struct task_struct *tsk;
1682         int retval;
1683
1684         trace_sched_process_wait(pid);
1685
1686         add_wait_queue(&current->signal->wait_chldexit,&wait);
1687 repeat:
1688         /*
1689          * If there is nothing that can match our critiera just get out.
1690          * We will clear @retval to zero if we see any child that might later
1691          * match our criteria, even if we are not able to reap it yet.
1692          */
1693         retval = -ECHILD;
1694         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1695                 goto end;
1696
1697         current->state = TASK_INTERRUPTIBLE;
1698         read_lock(&tasklist_lock);
1699         tsk = current;
1700         do {
1701                 int tsk_result = do_wait_thread(tsk, &retval,
1702                                                 type, pid, options,
1703                                                 infop, stat_addr, ru);
1704                 if (!tsk_result)
1705                         tsk_result = ptrace_do_wait(tsk, &retval,
1706                                                     type, pid, options,
1707                                                     infop, stat_addr, ru);
1708                 if (tsk_result) {
1709                         /*
1710                          * tasklist_lock is unlocked and we have a final result.
1711                          */
1712                         retval = tsk_result;
1713                         goto end;
1714                 }
1715
1716                 if (options & __WNOTHREAD)
1717                         break;
1718                 tsk = next_thread(tsk);
1719                 BUG_ON(tsk->signal != current->signal);
1720         } while (tsk != current);
1721         read_unlock(&tasklist_lock);
1722
1723         if (!retval && !(options & WNOHANG)) {
1724                 retval = -ERESTARTSYS;
1725                 if (!signal_pending(current)) {
1726                         schedule();
1727                         goto repeat;
1728                 }
1729         }
1730
1731 end:
1732         current->state = TASK_RUNNING;
1733         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1734         if (infop) {
1735                 if (retval > 0)
1736                         retval = 0;
1737                 else {
1738                         /*
1739                          * For a WNOHANG return, clear out all the fields
1740                          * we would set so the user can easily tell the
1741                          * difference.
1742                          */
1743                         if (!retval)
1744                                 retval = put_user(0, &infop->si_signo);
1745                         if (!retval)
1746                                 retval = put_user(0, &infop->si_errno);
1747                         if (!retval)
1748                                 retval = put_user(0, &infop->si_code);
1749                         if (!retval)
1750                                 retval = put_user(0, &infop->si_pid);
1751                         if (!retval)
1752                                 retval = put_user(0, &infop->si_uid);
1753                         if (!retval)
1754                                 retval = put_user(0, &infop->si_status);
1755                 }
1756         }
1757         return retval;
1758 }
1759
1760 asmlinkage long sys_waitid(int which, pid_t upid,
1761                            struct siginfo __user *infop, int options,
1762                            struct rusage __user *ru)
1763 {
1764         struct pid *pid = NULL;
1765         enum pid_type type;
1766         long ret;
1767
1768         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1769                 return -EINVAL;
1770         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1771                 return -EINVAL;
1772
1773         switch (which) {
1774         case P_ALL:
1775                 type = PIDTYPE_MAX;
1776                 break;
1777         case P_PID:
1778                 type = PIDTYPE_PID;
1779                 if (upid <= 0)
1780                         return -EINVAL;
1781                 break;
1782         case P_PGID:
1783                 type = PIDTYPE_PGID;
1784                 if (upid <= 0)
1785                         return -EINVAL;
1786                 break;
1787         default:
1788                 return -EINVAL;
1789         }
1790
1791         if (type < PIDTYPE_MAX)
1792                 pid = find_get_pid(upid);
1793         ret = do_wait(type, pid, options, infop, NULL, ru);
1794         put_pid(pid);
1795
1796         /* avoid REGPARM breakage on x86: */
1797         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1798         return ret;
1799 }
1800
1801 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1802                           int options, struct rusage __user *ru)
1803 {
1804         struct pid *pid = NULL;
1805         enum pid_type type;
1806         long ret;
1807
1808         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1809                         __WNOTHREAD|__WCLONE|__WALL))
1810                 return -EINVAL;
1811
1812         if (upid == -1)
1813                 type = PIDTYPE_MAX;
1814         else if (upid < 0) {
1815                 type = PIDTYPE_PGID;
1816                 pid = find_get_pid(-upid);
1817         } else if (upid == 0) {
1818                 type = PIDTYPE_PGID;
1819                 pid = get_pid(task_pgrp(current));
1820         } else /* upid > 0 */ {
1821                 type = PIDTYPE_PID;
1822                 pid = find_get_pid(upid);
1823         }
1824
1825         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1826         put_pid(pid);
1827
1828         /* avoid REGPARM breakage on x86: */
1829         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1830         return ret;
1831 }
1832
1833 #ifdef __ARCH_WANT_SYS_WAITPID
1834
1835 /*
1836  * sys_waitpid() remains for compatibility. waitpid() should be
1837  * implemented by calling sys_wait4() from libc.a.
1838  */
1839 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1840 {
1841         return sys_wait4(pid, stat_addr, options, NULL);
1842 }
1843
1844 #endif