Merge branch 'tracing-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[linux-2.6.git] / kernel / exit.c
1 /*
2  *  linux/kernel/exit.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 #include <linux/mm.h>
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/mnt_namespace.h>
16 #include <linux/iocontext.h>
17 #include <linux/key.h>
18 #include <linux/security.h>
19 #include <linux/cpu.h>
20 #include <linux/acct.h>
21 #include <linux/tsacct_kern.h>
22 #include <linux/file.h>
23 #include <linux/fdtable.h>
24 #include <linux/binfmts.h>
25 #include <linux/nsproxy.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/ptrace.h>
28 #include <linux/profile.h>
29 #include <linux/mount.h>
30 #include <linux/proc_fs.h>
31 #include <linux/kthread.h>
32 #include <linux/mempolicy.h>
33 #include <linux/taskstats_kern.h>
34 #include <linux/delayacct.h>
35 #include <linux/freezer.h>
36 #include <linux/cgroup.h>
37 #include <linux/syscalls.h>
38 #include <linux/signal.h>
39 #include <linux/posix-timers.h>
40 #include <linux/cn_proc.h>
41 #include <linux/mutex.h>
42 #include <linux/futex.h>
43 #include <linux/pipe_fs_i.h>
44 #include <linux/audit.h> /* for audit_free() */
45 #include <linux/resource.h>
46 #include <linux/blkdev.h>
47 #include <linux/task_io_accounting_ops.h>
48 #include <linux/tracehook.h>
49 #include <linux/init_task.h>
50 #include <trace/sched.h>
51
52 #include <asm/uaccess.h>
53 #include <asm/unistd.h>
54 #include <asm/pgtable.h>
55 #include <asm/mmu_context.h>
56 #include "cred-internals.h"
57
58 DEFINE_TRACE(sched_process_free);
59 DEFINE_TRACE(sched_process_exit);
60 DEFINE_TRACE(sched_process_wait);
61
62 static void exit_mm(struct task_struct * tsk);
63
64 static inline int task_detached(struct task_struct *p)
65 {
66         return p->exit_signal == -1;
67 }
68
69 static void __unhash_process(struct task_struct *p)
70 {
71         nr_threads--;
72         detach_pid(p, PIDTYPE_PID);
73         if (thread_group_leader(p)) {
74                 detach_pid(p, PIDTYPE_PGID);
75                 detach_pid(p, PIDTYPE_SID);
76
77                 list_del_rcu(&p->tasks);
78                 __get_cpu_var(process_counts)--;
79         }
80         list_del_rcu(&p->thread_group);
81         list_del_init(&p->sibling);
82 }
83
84 /*
85  * This function expects the tasklist_lock write-locked.
86  */
87 static void __exit_signal(struct task_struct *tsk)
88 {
89         struct signal_struct *sig = tsk->signal;
90         struct sighand_struct *sighand;
91
92         BUG_ON(!sig);
93         BUG_ON(!atomic_read(&sig->count));
94
95         sighand = rcu_dereference(tsk->sighand);
96         spin_lock(&sighand->siglock);
97
98         posix_cpu_timers_exit(tsk);
99         if (atomic_dec_and_test(&sig->count))
100                 posix_cpu_timers_exit_group(tsk);
101         else {
102                 /*
103                  * If there is any task waiting for the group exit
104                  * then notify it:
105                  */
106                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
107                         wake_up_process(sig->group_exit_task);
108
109                 if (tsk == sig->curr_target)
110                         sig->curr_target = next_thread(tsk);
111                 /*
112                  * Accumulate here the counters for all threads but the
113                  * group leader as they die, so they can be added into
114                  * the process-wide totals when those are taken.
115                  * The group leader stays around as a zombie as long
116                  * as there are other threads.  When it gets reaped,
117                  * the exit.c code will add its counts into these totals.
118                  * We won't ever get here for the group leader, since it
119                  * will have been the last reference on the signal_struct.
120                  */
121                 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
122                 sig->min_flt += tsk->min_flt;
123                 sig->maj_flt += tsk->maj_flt;
124                 sig->nvcsw += tsk->nvcsw;
125                 sig->nivcsw += tsk->nivcsw;
126                 sig->inblock += task_io_get_inblock(tsk);
127                 sig->oublock += task_io_get_oublock(tsk);
128                 task_io_accounting_add(&sig->ioac, &tsk->ioac);
129                 sig = NULL; /* Marker for below. */
130         }
131
132         __unhash_process(tsk);
133
134         /*
135          * Do this under ->siglock, we can race with another thread
136          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
137          */
138         flush_sigqueue(&tsk->pending);
139
140         tsk->signal = NULL;
141         tsk->sighand = NULL;
142         spin_unlock(&sighand->siglock);
143
144         __cleanup_sighand(sighand);
145         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
146         if (sig) {
147                 flush_sigqueue(&sig->shared_pending);
148                 taskstats_tgid_free(sig);
149                 /*
150                  * Make sure ->signal can't go away under rq->lock,
151                  * see account_group_exec_runtime().
152                  */
153                 task_rq_unlock_wait(tsk);
154                 __cleanup_signal(sig);
155         }
156 }
157
158 static void delayed_put_task_struct(struct rcu_head *rhp)
159 {
160         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
161
162         trace_sched_process_free(tsk);
163         put_task_struct(tsk);
164 }
165
166
167 void release_task(struct task_struct * p)
168 {
169         struct task_struct *leader;
170         int zap_leader;
171 repeat:
172         tracehook_prepare_release_task(p);
173         /* don't need to get the RCU readlock here - the process is dead and
174          * can't be modifying its own credentials */
175         atomic_dec(&__task_cred(p)->user->processes);
176
177         proc_flush_task(p);
178         write_lock_irq(&tasklist_lock);
179         tracehook_finish_release_task(p);
180         __exit_signal(p);
181
182         /*
183          * If we are the last non-leader member of the thread
184          * group, and the leader is zombie, then notify the
185          * group leader's parent process. (if it wants notification.)
186          */
187         zap_leader = 0;
188         leader = p->group_leader;
189         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
190                 BUG_ON(task_detached(leader));
191                 do_notify_parent(leader, leader->exit_signal);
192                 /*
193                  * If we were the last child thread and the leader has
194                  * exited already, and the leader's parent ignores SIGCHLD,
195                  * then we are the one who should release the leader.
196                  *
197                  * do_notify_parent() will have marked it self-reaping in
198                  * that case.
199                  */
200                 zap_leader = task_detached(leader);
201
202                 /*
203                  * This maintains the invariant that release_task()
204                  * only runs on a task in EXIT_DEAD, just for sanity.
205                  */
206                 if (zap_leader)
207                         leader->exit_state = EXIT_DEAD;
208         }
209
210         write_unlock_irq(&tasklist_lock);
211         release_thread(p);
212         call_rcu(&p->rcu, delayed_put_task_struct);
213
214         p = leader;
215         if (unlikely(zap_leader))
216                 goto repeat;
217 }
218
219 /*
220  * This checks not only the pgrp, but falls back on the pid if no
221  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
222  * without this...
223  *
224  * The caller must hold rcu lock or the tasklist lock.
225  */
226 struct pid *session_of_pgrp(struct pid *pgrp)
227 {
228         struct task_struct *p;
229         struct pid *sid = NULL;
230
231         p = pid_task(pgrp, PIDTYPE_PGID);
232         if (p == NULL)
233                 p = pid_task(pgrp, PIDTYPE_PID);
234         if (p != NULL)
235                 sid = task_session(p);
236
237         return sid;
238 }
239
240 /*
241  * Determine if a process group is "orphaned", according to the POSIX
242  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
243  * by terminal-generated stop signals.  Newly orphaned process groups are
244  * to receive a SIGHUP and a SIGCONT.
245  *
246  * "I ask you, have you ever known what it is to be an orphan?"
247  */
248 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
249 {
250         struct task_struct *p;
251
252         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
253                 if ((p == ignored_task) ||
254                     (p->exit_state && thread_group_empty(p)) ||
255                     is_global_init(p->real_parent))
256                         continue;
257
258                 if (task_pgrp(p->real_parent) != pgrp &&
259                     task_session(p->real_parent) == task_session(p))
260                         return 0;
261         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
262
263         return 1;
264 }
265
266 int is_current_pgrp_orphaned(void)
267 {
268         int retval;
269
270         read_lock(&tasklist_lock);
271         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
272         read_unlock(&tasklist_lock);
273
274         return retval;
275 }
276
277 static int has_stopped_jobs(struct pid *pgrp)
278 {
279         int retval = 0;
280         struct task_struct *p;
281
282         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
283                 if (!task_is_stopped(p))
284                         continue;
285                 retval = 1;
286                 break;
287         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
288         return retval;
289 }
290
291 /*
292  * Check to see if any process groups have become orphaned as
293  * a result of our exiting, and if they have any stopped jobs,
294  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
295  */
296 static void
297 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
298 {
299         struct pid *pgrp = task_pgrp(tsk);
300         struct task_struct *ignored_task = tsk;
301
302         if (!parent)
303                  /* exit: our father is in a different pgrp than
304                   * we are and we were the only connection outside.
305                   */
306                 parent = tsk->real_parent;
307         else
308                 /* reparent: our child is in a different pgrp than
309                  * we are, and it was the only connection outside.
310                  */
311                 ignored_task = NULL;
312
313         if (task_pgrp(parent) != pgrp &&
314             task_session(parent) == task_session(tsk) &&
315             will_become_orphaned_pgrp(pgrp, ignored_task) &&
316             has_stopped_jobs(pgrp)) {
317                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
318                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
319         }
320 }
321
322 /**
323  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
324  *
325  * If a kernel thread is launched as a result of a system call, or if
326  * it ever exits, it should generally reparent itself to kthreadd so it
327  * isn't in the way of other processes and is correctly cleaned up on exit.
328  *
329  * The various task state such as scheduling policy and priority may have
330  * been inherited from a user process, so we reset them to sane values here.
331  *
332  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
333  */
334 static void reparent_to_kthreadd(void)
335 {
336         write_lock_irq(&tasklist_lock);
337
338         ptrace_unlink(current);
339         /* Reparent to init */
340         current->real_parent = current->parent = kthreadd_task;
341         list_move_tail(&current->sibling, &current->real_parent->children);
342
343         /* Set the exit signal to SIGCHLD so we signal init on exit */
344         current->exit_signal = SIGCHLD;
345
346         if (task_nice(current) < 0)
347                 set_user_nice(current, 0);
348         /* cpus_allowed? */
349         /* rt_priority? */
350         /* signals? */
351         memcpy(current->signal->rlim, init_task.signal->rlim,
352                sizeof(current->signal->rlim));
353
354         atomic_inc(&init_cred.usage);
355         commit_creds(&init_cred);
356         write_unlock_irq(&tasklist_lock);
357 }
358
359 void __set_special_pids(struct pid *pid)
360 {
361         struct task_struct *curr = current->group_leader;
362         pid_t nr = pid_nr(pid);
363
364         if (task_session(curr) != pid) {
365                 change_pid(curr, PIDTYPE_SID, pid);
366                 set_task_session(curr, nr);
367         }
368         if (task_pgrp(curr) != pid) {
369                 change_pid(curr, PIDTYPE_PGID, pid);
370                 set_task_pgrp(curr, nr);
371         }
372 }
373
374 static void set_special_pids(struct pid *pid)
375 {
376         write_lock_irq(&tasklist_lock);
377         __set_special_pids(pid);
378         write_unlock_irq(&tasklist_lock);
379 }
380
381 /*
382  * Let kernel threads use this to say that they
383  * allow a certain signal (since daemonize() will
384  * have disabled all of them by default).
385  */
386 int allow_signal(int sig)
387 {
388         if (!valid_signal(sig) || sig < 1)
389                 return -EINVAL;
390
391         spin_lock_irq(&current->sighand->siglock);
392         sigdelset(&current->blocked, sig);
393         if (!current->mm) {
394                 /* Kernel threads handle their own signals.
395                    Let the signal code know it'll be handled, so
396                    that they don't get converted to SIGKILL or
397                    just silently dropped */
398                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
399         }
400         recalc_sigpending();
401         spin_unlock_irq(&current->sighand->siglock);
402         return 0;
403 }
404
405 EXPORT_SYMBOL(allow_signal);
406
407 int disallow_signal(int sig)
408 {
409         if (!valid_signal(sig) || sig < 1)
410                 return -EINVAL;
411
412         spin_lock_irq(&current->sighand->siglock);
413         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
414         recalc_sigpending();
415         spin_unlock_irq(&current->sighand->siglock);
416         return 0;
417 }
418
419 EXPORT_SYMBOL(disallow_signal);
420
421 /*
422  *      Put all the gunge required to become a kernel thread without
423  *      attached user resources in one place where it belongs.
424  */
425
426 void daemonize(const char *name, ...)
427 {
428         va_list args;
429         struct fs_struct *fs;
430         sigset_t blocked;
431
432         va_start(args, name);
433         vsnprintf(current->comm, sizeof(current->comm), name, args);
434         va_end(args);
435
436         /*
437          * If we were started as result of loading a module, close all of the
438          * user space pages.  We don't need them, and if we didn't close them
439          * they would be locked into memory.
440          */
441         exit_mm(current);
442         /*
443          * We don't want to have TIF_FREEZE set if the system-wide hibernation
444          * or suspend transition begins right now.
445          */
446         current->flags |= (PF_NOFREEZE | PF_KTHREAD);
447
448         if (current->nsproxy != &init_nsproxy) {
449                 get_nsproxy(&init_nsproxy);
450                 switch_task_namespaces(current, &init_nsproxy);
451         }
452         set_special_pids(&init_struct_pid);
453         proc_clear_tty(current);
454
455         /* Block and flush all signals */
456         sigfillset(&blocked);
457         sigprocmask(SIG_BLOCK, &blocked, NULL);
458         flush_signals(current);
459
460         /* Become as one with the init task */
461
462         exit_fs(current);       /* current->fs->count--; */
463         fs = init_task.fs;
464         current->fs = fs;
465         atomic_inc(&fs->count);
466
467         exit_files(current);
468         current->files = init_task.files;
469         atomic_inc(&current->files->count);
470
471         reparent_to_kthreadd();
472 }
473
474 EXPORT_SYMBOL(daemonize);
475
476 static void close_files(struct files_struct * files)
477 {
478         int i, j;
479         struct fdtable *fdt;
480
481         j = 0;
482
483         /*
484          * It is safe to dereference the fd table without RCU or
485          * ->file_lock because this is the last reference to the
486          * files structure.
487          */
488         fdt = files_fdtable(files);
489         for (;;) {
490                 unsigned long set;
491                 i = j * __NFDBITS;
492                 if (i >= fdt->max_fds)
493                         break;
494                 set = fdt->open_fds->fds_bits[j++];
495                 while (set) {
496                         if (set & 1) {
497                                 struct file * file = xchg(&fdt->fd[i], NULL);
498                                 if (file) {
499                                         filp_close(file, files);
500                                         cond_resched();
501                                 }
502                         }
503                         i++;
504                         set >>= 1;
505                 }
506         }
507 }
508
509 struct files_struct *get_files_struct(struct task_struct *task)
510 {
511         struct files_struct *files;
512
513         task_lock(task);
514         files = task->files;
515         if (files)
516                 atomic_inc(&files->count);
517         task_unlock(task);
518
519         return files;
520 }
521
522 void put_files_struct(struct files_struct *files)
523 {
524         struct fdtable *fdt;
525
526         if (atomic_dec_and_test(&files->count)) {
527                 close_files(files);
528                 /*
529                  * Free the fd and fdset arrays if we expanded them.
530                  * If the fdtable was embedded, pass files for freeing
531                  * at the end of the RCU grace period. Otherwise,
532                  * you can free files immediately.
533                  */
534                 fdt = files_fdtable(files);
535                 if (fdt != &files->fdtab)
536                         kmem_cache_free(files_cachep, files);
537                 free_fdtable(fdt);
538         }
539 }
540
541 void reset_files_struct(struct files_struct *files)
542 {
543         struct task_struct *tsk = current;
544         struct files_struct *old;
545
546         old = tsk->files;
547         task_lock(tsk);
548         tsk->files = files;
549         task_unlock(tsk);
550         put_files_struct(old);
551 }
552
553 void exit_files(struct task_struct *tsk)
554 {
555         struct files_struct * files = tsk->files;
556
557         if (files) {
558                 task_lock(tsk);
559                 tsk->files = NULL;
560                 task_unlock(tsk);
561                 put_files_struct(files);
562         }
563 }
564
565 void put_fs_struct(struct fs_struct *fs)
566 {
567         /* No need to hold fs->lock if we are killing it */
568         if (atomic_dec_and_test(&fs->count)) {
569                 path_put(&fs->root);
570                 path_put(&fs->pwd);
571                 kmem_cache_free(fs_cachep, fs);
572         }
573 }
574
575 void exit_fs(struct task_struct *tsk)
576 {
577         struct fs_struct * fs = tsk->fs;
578
579         if (fs) {
580                 task_lock(tsk);
581                 tsk->fs = NULL;
582                 task_unlock(tsk);
583                 put_fs_struct(fs);
584         }
585 }
586
587 EXPORT_SYMBOL_GPL(exit_fs);
588
589 #ifdef CONFIG_MM_OWNER
590 /*
591  * Task p is exiting and it owned mm, lets find a new owner for it
592  */
593 static inline int
594 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
595 {
596         /*
597          * If there are other users of the mm and the owner (us) is exiting
598          * we need to find a new owner to take on the responsibility.
599          */
600         if (atomic_read(&mm->mm_users) <= 1)
601                 return 0;
602         if (mm->owner != p)
603                 return 0;
604         return 1;
605 }
606
607 void mm_update_next_owner(struct mm_struct *mm)
608 {
609         struct task_struct *c, *g, *p = current;
610
611 retry:
612         if (!mm_need_new_owner(mm, p))
613                 return;
614
615         read_lock(&tasklist_lock);
616         /*
617          * Search in the children
618          */
619         list_for_each_entry(c, &p->children, sibling) {
620                 if (c->mm == mm)
621                         goto assign_new_owner;
622         }
623
624         /*
625          * Search in the siblings
626          */
627         list_for_each_entry(c, &p->parent->children, sibling) {
628                 if (c->mm == mm)
629                         goto assign_new_owner;
630         }
631
632         /*
633          * Search through everything else. We should not get
634          * here often
635          */
636         do_each_thread(g, c) {
637                 if (c->mm == mm)
638                         goto assign_new_owner;
639         } while_each_thread(g, c);
640
641         read_unlock(&tasklist_lock);
642         /*
643          * We found no owner yet mm_users > 1: this implies that we are
644          * most likely racing with swapoff (try_to_unuse()) or /proc or
645          * ptrace or page migration (get_task_mm()).  Mark owner as NULL,
646          * so that subsystems can understand the callback and take action.
647          */
648         down_write(&mm->mmap_sem);
649         cgroup_mm_owner_callbacks(mm->owner, NULL);
650         mm->owner = NULL;
651         up_write(&mm->mmap_sem);
652         return;
653
654 assign_new_owner:
655         BUG_ON(c == p);
656         get_task_struct(c);
657         read_unlock(&tasklist_lock);
658         down_write(&mm->mmap_sem);
659         /*
660          * The task_lock protects c->mm from changing.
661          * We always want mm->owner->mm == mm
662          */
663         task_lock(c);
664         if (c->mm != mm) {
665                 task_unlock(c);
666                 up_write(&mm->mmap_sem);
667                 put_task_struct(c);
668                 goto retry;
669         }
670         cgroup_mm_owner_callbacks(mm->owner, c);
671         mm->owner = c;
672         task_unlock(c);
673         up_write(&mm->mmap_sem);
674         put_task_struct(c);
675 }
676 #endif /* CONFIG_MM_OWNER */
677
678 /*
679  * Turn us into a lazy TLB process if we
680  * aren't already..
681  */
682 static void exit_mm(struct task_struct * tsk)
683 {
684         struct mm_struct *mm = tsk->mm;
685         struct core_state *core_state;
686
687         mm_release(tsk, mm);
688         if (!mm)
689                 return;
690         /*
691          * Serialize with any possible pending coredump.
692          * We must hold mmap_sem around checking core_state
693          * and clearing tsk->mm.  The core-inducing thread
694          * will increment ->nr_threads for each thread in the
695          * group with ->mm != NULL.
696          */
697         down_read(&mm->mmap_sem);
698         core_state = mm->core_state;
699         if (core_state) {
700                 struct core_thread self;
701                 up_read(&mm->mmap_sem);
702
703                 self.task = tsk;
704                 self.next = xchg(&core_state->dumper.next, &self);
705                 /*
706                  * Implies mb(), the result of xchg() must be visible
707                  * to core_state->dumper.
708                  */
709                 if (atomic_dec_and_test(&core_state->nr_threads))
710                         complete(&core_state->startup);
711
712                 for (;;) {
713                         set_task_state(tsk, TASK_UNINTERRUPTIBLE);
714                         if (!self.task) /* see coredump_finish() */
715                                 break;
716                         schedule();
717                 }
718                 __set_task_state(tsk, TASK_RUNNING);
719                 down_read(&mm->mmap_sem);
720         }
721         atomic_inc(&mm->mm_count);
722         BUG_ON(mm != tsk->active_mm);
723         /* more a memory barrier than a real lock */
724         task_lock(tsk);
725         tsk->mm = NULL;
726         up_read(&mm->mmap_sem);
727         enter_lazy_tlb(mm, current);
728         /* We don't want this task to be frozen prematurely */
729         clear_freeze_flag(tsk);
730         task_unlock(tsk);
731         mm_update_next_owner(mm);
732         mmput(mm);
733 }
734
735 /*
736  * Return nonzero if @parent's children should reap themselves.
737  *
738  * Called with write_lock_irq(&tasklist_lock) held.
739  */
740 static int ignoring_children(struct task_struct *parent)
741 {
742         int ret;
743         struct sighand_struct *psig = parent->sighand;
744         unsigned long flags;
745         spin_lock_irqsave(&psig->siglock, flags);
746         ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
747                (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
748         spin_unlock_irqrestore(&psig->siglock, flags);
749         return ret;
750 }
751
752 /*
753  * Detach all tasks we were using ptrace on.
754  * Any that need to be release_task'd are put on the @dead list.
755  *
756  * Called with write_lock(&tasklist_lock) held.
757  */
758 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
759 {
760         struct task_struct *p, *n;
761         int ign = -1;
762
763         list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
764                 __ptrace_unlink(p);
765
766                 if (p->exit_state != EXIT_ZOMBIE)
767                         continue;
768
769                 /*
770                  * If it's a zombie, our attachedness prevented normal
771                  * parent notification or self-reaping.  Do notification
772                  * now if it would have happened earlier.  If it should
773                  * reap itself, add it to the @dead list.  We can't call
774                  * release_task() here because we already hold tasklist_lock.
775                  *
776                  * If it's our own child, there is no notification to do.
777                  * But if our normal children self-reap, then this child
778                  * was prevented by ptrace and we must reap it now.
779                  */
780                 if (!task_detached(p) && thread_group_empty(p)) {
781                         if (!same_thread_group(p->real_parent, parent))
782                                 do_notify_parent(p, p->exit_signal);
783                         else {
784                                 if (ign < 0)
785                                         ign = ignoring_children(parent);
786                                 if (ign)
787                                         p->exit_signal = -1;
788                         }
789                 }
790
791                 if (task_detached(p)) {
792                         /*
793                          * Mark it as in the process of being reaped.
794                          */
795                         p->exit_state = EXIT_DEAD;
796                         list_add(&p->ptrace_entry, dead);
797                 }
798         }
799 }
800
801 /*
802  * Finish up exit-time ptrace cleanup.
803  *
804  * Called without locks.
805  */
806 static void ptrace_exit_finish(struct task_struct *parent,
807                                struct list_head *dead)
808 {
809         struct task_struct *p, *n;
810
811         BUG_ON(!list_empty(&parent->ptraced));
812
813         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
814                 list_del_init(&p->ptrace_entry);
815                 release_task(p);
816         }
817 }
818
819 static void reparent_thread(struct task_struct *p, struct task_struct *father)
820 {
821         if (p->pdeath_signal)
822                 /* We already hold the tasklist_lock here.  */
823                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
824
825         list_move_tail(&p->sibling, &p->real_parent->children);
826
827         /* If this is a threaded reparent there is no need to
828          * notify anyone anything has happened.
829          */
830         if (same_thread_group(p->real_parent, father))
831                 return;
832
833         /* We don't want people slaying init.  */
834         if (!task_detached(p))
835                 p->exit_signal = SIGCHLD;
836
837         /* If we'd notified the old parent about this child's death,
838          * also notify the new parent.
839          */
840         if (!ptrace_reparented(p) &&
841             p->exit_state == EXIT_ZOMBIE &&
842             !task_detached(p) && thread_group_empty(p))
843                 do_notify_parent(p, p->exit_signal);
844
845         kill_orphaned_pgrp(p, father);
846 }
847
848 /*
849  * When we die, we re-parent all our children.
850  * Try to give them to another thread in our thread
851  * group, and if no such member exists, give it to
852  * the child reaper process (ie "init") in our pid
853  * space.
854  */
855 static struct task_struct *find_new_reaper(struct task_struct *father)
856 {
857         struct pid_namespace *pid_ns = task_active_pid_ns(father);
858         struct task_struct *thread;
859
860         thread = father;
861         while_each_thread(father, thread) {
862                 if (thread->flags & PF_EXITING)
863                         continue;
864                 if (unlikely(pid_ns->child_reaper == father))
865                         pid_ns->child_reaper = thread;
866                 return thread;
867         }
868
869         if (unlikely(pid_ns->child_reaper == father)) {
870                 write_unlock_irq(&tasklist_lock);
871                 if (unlikely(pid_ns == &init_pid_ns))
872                         panic("Attempted to kill init!");
873
874                 zap_pid_ns_processes(pid_ns);
875                 write_lock_irq(&tasklist_lock);
876                 /*
877                  * We can not clear ->child_reaper or leave it alone.
878                  * There may by stealth EXIT_DEAD tasks on ->children,
879                  * forget_original_parent() must move them somewhere.
880                  */
881                 pid_ns->child_reaper = init_pid_ns.child_reaper;
882         }
883
884         return pid_ns->child_reaper;
885 }
886
887 static void forget_original_parent(struct task_struct *father)
888 {
889         struct task_struct *p, *n, *reaper;
890         LIST_HEAD(ptrace_dead);
891
892         write_lock_irq(&tasklist_lock);
893         reaper = find_new_reaper(father);
894         /*
895          * First clean up ptrace if we were using it.
896          */
897         ptrace_exit(father, &ptrace_dead);
898
899         list_for_each_entry_safe(p, n, &father->children, sibling) {
900                 p->real_parent = reaper;
901                 if (p->parent == father) {
902                         BUG_ON(p->ptrace);
903                         p->parent = p->real_parent;
904                 }
905                 reparent_thread(p, father);
906         }
907
908         write_unlock_irq(&tasklist_lock);
909         BUG_ON(!list_empty(&father->children));
910
911         ptrace_exit_finish(father, &ptrace_dead);
912 }
913
914 /*
915  * Send signals to all our closest relatives so that they know
916  * to properly mourn us..
917  */
918 static void exit_notify(struct task_struct *tsk, int group_dead)
919 {
920         int signal;
921         void *cookie;
922
923         /*
924          * This does two things:
925          *
926          * A.  Make init inherit all the child processes
927          * B.  Check to see if any process groups have become orphaned
928          *      as a result of our exiting, and if they have any stopped
929          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
930          */
931         forget_original_parent(tsk);
932         exit_task_namespaces(tsk);
933
934         write_lock_irq(&tasklist_lock);
935         if (group_dead)
936                 kill_orphaned_pgrp(tsk->group_leader, NULL);
937
938         /* Let father know we died
939          *
940          * Thread signals are configurable, but you aren't going to use
941          * that to send signals to arbitary processes.
942          * That stops right now.
943          *
944          * If the parent exec id doesn't match the exec id we saved
945          * when we started then we know the parent has changed security
946          * domain.
947          *
948          * If our self_exec id doesn't match our parent_exec_id then
949          * we have changed execution domain as these two values started
950          * the same after a fork.
951          */
952         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
953             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
954              tsk->self_exec_id != tsk->parent_exec_id) &&
955             !capable(CAP_KILL))
956                 tsk->exit_signal = SIGCHLD;
957
958         signal = tracehook_notify_death(tsk, &cookie, group_dead);
959         if (signal >= 0)
960                 signal = do_notify_parent(tsk, signal);
961
962         tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
963
964         /* mt-exec, de_thread() is waiting for us */
965         if (thread_group_leader(tsk) &&
966             tsk->signal->group_exit_task &&
967             tsk->signal->notify_count < 0)
968                 wake_up_process(tsk->signal->group_exit_task);
969
970         write_unlock_irq(&tasklist_lock);
971
972         tracehook_report_death(tsk, signal, cookie, group_dead);
973
974         /* If the process is dead, release it - nobody will wait for it */
975         if (signal == DEATH_REAP)
976                 release_task(tsk);
977 }
978
979 #ifdef CONFIG_DEBUG_STACK_USAGE
980 static void check_stack_usage(void)
981 {
982         static DEFINE_SPINLOCK(low_water_lock);
983         static int lowest_to_date = THREAD_SIZE;
984         unsigned long *n = end_of_stack(current);
985         unsigned long free;
986
987         while (*n == 0)
988                 n++;
989         free = (unsigned long)n - (unsigned long)end_of_stack(current);
990
991         if (free >= lowest_to_date)
992                 return;
993
994         spin_lock(&low_water_lock);
995         if (free < lowest_to_date) {
996                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
997                                 "left\n",
998                                 current->comm, free);
999                 lowest_to_date = free;
1000         }
1001         spin_unlock(&low_water_lock);
1002 }
1003 #else
1004 static inline void check_stack_usage(void) {}
1005 #endif
1006
1007 NORET_TYPE void do_exit(long code)
1008 {
1009         struct task_struct *tsk = current;
1010         int group_dead;
1011
1012         profile_task_exit(tsk);
1013
1014         WARN_ON(atomic_read(&tsk->fs_excl));
1015
1016         if (unlikely(in_interrupt()))
1017                 panic("Aiee, killing interrupt handler!");
1018         if (unlikely(!tsk->pid))
1019                 panic("Attempted to kill the idle task!");
1020
1021         tracehook_report_exit(&code);
1022
1023         /*
1024          * We're taking recursive faults here in do_exit. Safest is to just
1025          * leave this task alone and wait for reboot.
1026          */
1027         if (unlikely(tsk->flags & PF_EXITING)) {
1028                 printk(KERN_ALERT
1029                         "Fixing recursive fault but reboot is needed!\n");
1030                 /*
1031                  * We can do this unlocked here. The futex code uses
1032                  * this flag just to verify whether the pi state
1033                  * cleanup has been done or not. In the worst case it
1034                  * loops once more. We pretend that the cleanup was
1035                  * done as there is no way to return. Either the
1036                  * OWNER_DIED bit is set by now or we push the blocked
1037                  * task into the wait for ever nirwana as well.
1038                  */
1039                 tsk->flags |= PF_EXITPIDONE;
1040                 if (tsk->io_context)
1041                         exit_io_context();
1042                 set_current_state(TASK_UNINTERRUPTIBLE);
1043                 schedule();
1044         }
1045
1046         exit_signals(tsk);  /* sets PF_EXITING */
1047         /*
1048          * tsk->flags are checked in the futex code to protect against
1049          * an exiting task cleaning up the robust pi futexes.
1050          */
1051         smp_mb();
1052         spin_unlock_wait(&tsk->pi_lock);
1053
1054         if (unlikely(in_atomic()))
1055                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1056                                 current->comm, task_pid_nr(current),
1057                                 preempt_count());
1058
1059         acct_update_integrals(tsk);
1060         if (tsk->mm) {
1061                 update_hiwater_rss(tsk->mm);
1062                 update_hiwater_vm(tsk->mm);
1063         }
1064         group_dead = atomic_dec_and_test(&tsk->signal->live);
1065         if (group_dead) {
1066                 hrtimer_cancel(&tsk->signal->real_timer);
1067                 exit_itimers(tsk->signal);
1068         }
1069         acct_collect(code, group_dead);
1070         if (group_dead)
1071                 tty_audit_exit();
1072         if (unlikely(tsk->audit_context))
1073                 audit_free(tsk);
1074
1075         tsk->exit_code = code;
1076         taskstats_exit(tsk, group_dead);
1077
1078         exit_mm(tsk);
1079
1080         if (group_dead)
1081                 acct_process();
1082         trace_sched_process_exit(tsk);
1083
1084         exit_sem(tsk);
1085         exit_files(tsk);
1086         exit_fs(tsk);
1087         check_stack_usage();
1088         exit_thread();
1089         cgroup_exit(tsk, 1);
1090
1091         if (group_dead && tsk->signal->leader)
1092                 disassociate_ctty(1);
1093
1094         module_put(task_thread_info(tsk)->exec_domain->module);
1095         if (tsk->binfmt)
1096                 module_put(tsk->binfmt->module);
1097
1098         proc_exit_connector(tsk);
1099         exit_notify(tsk, group_dead);
1100 #ifdef CONFIG_NUMA
1101         mpol_put(tsk->mempolicy);
1102         tsk->mempolicy = NULL;
1103 #endif
1104 #ifdef CONFIG_FUTEX
1105         /*
1106          * This must happen late, after the PID is not
1107          * hashed anymore:
1108          */
1109         if (unlikely(!list_empty(&tsk->pi_state_list)))
1110                 exit_pi_state_list(tsk);
1111         if (unlikely(current->pi_state_cache))
1112                 kfree(current->pi_state_cache);
1113 #endif
1114         /*
1115          * Make sure we are holding no locks:
1116          */
1117         debug_check_no_locks_held(tsk);
1118         /*
1119          * We can do this unlocked here. The futex code uses this flag
1120          * just to verify whether the pi state cleanup has been done
1121          * or not. In the worst case it loops once more.
1122          */
1123         tsk->flags |= PF_EXITPIDONE;
1124
1125         if (tsk->io_context)
1126                 exit_io_context();
1127
1128         if (tsk->splice_pipe)
1129                 __free_pipe_info(tsk->splice_pipe);
1130
1131         preempt_disable();
1132         /* causes final put_task_struct in finish_task_switch(). */
1133         tsk->state = TASK_DEAD;
1134         schedule();
1135         BUG();
1136         /* Avoid "noreturn function does return".  */
1137         for (;;)
1138                 cpu_relax();    /* For when BUG is null */
1139 }
1140
1141 EXPORT_SYMBOL_GPL(do_exit);
1142
1143 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1144 {
1145         if (comp)
1146                 complete(comp);
1147
1148         do_exit(code);
1149 }
1150
1151 EXPORT_SYMBOL(complete_and_exit);
1152
1153 asmlinkage long sys_exit(int error_code)
1154 {
1155         do_exit((error_code&0xff)<<8);
1156 }
1157
1158 /*
1159  * Take down every thread in the group.  This is called by fatal signals
1160  * as well as by sys_exit_group (below).
1161  */
1162 NORET_TYPE void
1163 do_group_exit(int exit_code)
1164 {
1165         struct signal_struct *sig = current->signal;
1166
1167         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1168
1169         if (signal_group_exit(sig))
1170                 exit_code = sig->group_exit_code;
1171         else if (!thread_group_empty(current)) {
1172                 struct sighand_struct *const sighand = current->sighand;
1173                 spin_lock_irq(&sighand->siglock);
1174                 if (signal_group_exit(sig))
1175                         /* Another thread got here before we took the lock.  */
1176                         exit_code = sig->group_exit_code;
1177                 else {
1178                         sig->group_exit_code = exit_code;
1179                         sig->flags = SIGNAL_GROUP_EXIT;
1180                         zap_other_threads(current);
1181                 }
1182                 spin_unlock_irq(&sighand->siglock);
1183         }
1184
1185         do_exit(exit_code);
1186         /* NOTREACHED */
1187 }
1188
1189 /*
1190  * this kills every thread in the thread group. Note that any externally
1191  * wait4()-ing process will get the correct exit code - even if this
1192  * thread is not the thread group leader.
1193  */
1194 asmlinkage void sys_exit_group(int error_code)
1195 {
1196         do_group_exit((error_code & 0xff) << 8);
1197 }
1198
1199 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1200 {
1201         struct pid *pid = NULL;
1202         if (type == PIDTYPE_PID)
1203                 pid = task->pids[type].pid;
1204         else if (type < PIDTYPE_MAX)
1205                 pid = task->group_leader->pids[type].pid;
1206         return pid;
1207 }
1208
1209 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1210                           struct task_struct *p)
1211 {
1212         int err;
1213
1214         if (type < PIDTYPE_MAX) {
1215                 if (task_pid_type(p, type) != pid)
1216                         return 0;
1217         }
1218
1219         /* Wait for all children (clone and not) if __WALL is set;
1220          * otherwise, wait for clone children *only* if __WCLONE is
1221          * set; otherwise, wait for non-clone children *only*.  (Note:
1222          * A "clone" child here is one that reports to its parent
1223          * using a signal other than SIGCHLD.) */
1224         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1225             && !(options & __WALL))
1226                 return 0;
1227
1228         err = security_task_wait(p);
1229         if (err)
1230                 return err;
1231
1232         return 1;
1233 }
1234
1235 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1236                                int why, int status,
1237                                struct siginfo __user *infop,
1238                                struct rusage __user *rusagep)
1239 {
1240         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1241
1242         put_task_struct(p);
1243         if (!retval)
1244                 retval = put_user(SIGCHLD, &infop->si_signo);
1245         if (!retval)
1246                 retval = put_user(0, &infop->si_errno);
1247         if (!retval)
1248                 retval = put_user((short)why, &infop->si_code);
1249         if (!retval)
1250                 retval = put_user(pid, &infop->si_pid);
1251         if (!retval)
1252                 retval = put_user(uid, &infop->si_uid);
1253         if (!retval)
1254                 retval = put_user(status, &infop->si_status);
1255         if (!retval)
1256                 retval = pid;
1257         return retval;
1258 }
1259
1260 /*
1261  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1262  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1263  * the lock and this task is uninteresting.  If we return nonzero, we have
1264  * released the lock and the system call should return.
1265  */
1266 static int wait_task_zombie(struct task_struct *p, int options,
1267                             struct siginfo __user *infop,
1268                             int __user *stat_addr, struct rusage __user *ru)
1269 {
1270         unsigned long state;
1271         int retval, status, traced;
1272         pid_t pid = task_pid_vnr(p);
1273         uid_t uid = __task_cred(p)->uid;
1274
1275         if (!likely(options & WEXITED))
1276                 return 0;
1277
1278         if (unlikely(options & WNOWAIT)) {
1279                 int exit_code = p->exit_code;
1280                 int why, status;
1281
1282                 get_task_struct(p);
1283                 read_unlock(&tasklist_lock);
1284                 if ((exit_code & 0x7f) == 0) {
1285                         why = CLD_EXITED;
1286                         status = exit_code >> 8;
1287                 } else {
1288                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1289                         status = exit_code & 0x7f;
1290                 }
1291                 return wait_noreap_copyout(p, pid, uid, why,
1292                                            status, infop, ru);
1293         }
1294
1295         /*
1296          * Try to move the task's state to DEAD
1297          * only one thread is allowed to do this:
1298          */
1299         state = xchg(&p->exit_state, EXIT_DEAD);
1300         if (state != EXIT_ZOMBIE) {
1301                 BUG_ON(state != EXIT_DEAD);
1302                 return 0;
1303         }
1304
1305         traced = ptrace_reparented(p);
1306
1307         if (likely(!traced)) {
1308                 struct signal_struct *psig;
1309                 struct signal_struct *sig;
1310                 struct task_cputime cputime;
1311
1312                 /*
1313                  * The resource counters for the group leader are in its
1314                  * own task_struct.  Those for dead threads in the group
1315                  * are in its signal_struct, as are those for the child
1316                  * processes it has previously reaped.  All these
1317                  * accumulate in the parent's signal_struct c* fields.
1318                  *
1319                  * We don't bother to take a lock here to protect these
1320                  * p->signal fields, because they are only touched by
1321                  * __exit_signal, which runs with tasklist_lock
1322                  * write-locked anyway, and so is excluded here.  We do
1323                  * need to protect the access to p->parent->signal fields,
1324                  * as other threads in the parent group can be right
1325                  * here reaping other children at the same time.
1326                  *
1327                  * We use thread_group_cputime() to get times for the thread
1328                  * group, which consolidates times for all threads in the
1329                  * group including the group leader.
1330                  */
1331                 spin_lock_irq(&p->parent->sighand->siglock);
1332                 psig = p->parent->signal;
1333                 sig = p->signal;
1334                 thread_group_cputime(p, &cputime);
1335                 psig->cutime =
1336                         cputime_add(psig->cutime,
1337                         cputime_add(cputime.utime,
1338                                     sig->cutime));
1339                 psig->cstime =
1340                         cputime_add(psig->cstime,
1341                         cputime_add(cputime.stime,
1342                                     sig->cstime));
1343                 psig->cgtime =
1344                         cputime_add(psig->cgtime,
1345                         cputime_add(p->gtime,
1346                         cputime_add(sig->gtime,
1347                                     sig->cgtime)));
1348                 psig->cmin_flt +=
1349                         p->min_flt + sig->min_flt + sig->cmin_flt;
1350                 psig->cmaj_flt +=
1351                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1352                 psig->cnvcsw +=
1353                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1354                 psig->cnivcsw +=
1355                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1356                 psig->cinblock +=
1357                         task_io_get_inblock(p) +
1358                         sig->inblock + sig->cinblock;
1359                 psig->coublock +=
1360                         task_io_get_oublock(p) +
1361                         sig->oublock + sig->coublock;
1362                 task_io_accounting_add(&psig->ioac, &p->ioac);
1363                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1364                 spin_unlock_irq(&p->parent->sighand->siglock);
1365         }
1366
1367         /*
1368          * Now we are sure this task is interesting, and no other
1369          * thread can reap it because we set its state to EXIT_DEAD.
1370          */
1371         read_unlock(&tasklist_lock);
1372
1373         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1374         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1375                 ? p->signal->group_exit_code : p->exit_code;
1376         if (!retval && stat_addr)
1377                 retval = put_user(status, stat_addr);
1378         if (!retval && infop)
1379                 retval = put_user(SIGCHLD, &infop->si_signo);
1380         if (!retval && infop)
1381                 retval = put_user(0, &infop->si_errno);
1382         if (!retval && infop) {
1383                 int why;
1384
1385                 if ((status & 0x7f) == 0) {
1386                         why = CLD_EXITED;
1387                         status >>= 8;
1388                 } else {
1389                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1390                         status &= 0x7f;
1391                 }
1392                 retval = put_user((short)why, &infop->si_code);
1393                 if (!retval)
1394                         retval = put_user(status, &infop->si_status);
1395         }
1396         if (!retval && infop)
1397                 retval = put_user(pid, &infop->si_pid);
1398         if (!retval && infop)
1399                 retval = put_user(uid, &infop->si_uid);
1400         if (!retval)
1401                 retval = pid;
1402
1403         if (traced) {
1404                 write_lock_irq(&tasklist_lock);
1405                 /* We dropped tasklist, ptracer could die and untrace */
1406                 ptrace_unlink(p);
1407                 /*
1408                  * If this is not a detached task, notify the parent.
1409                  * If it's still not detached after that, don't release
1410                  * it now.
1411                  */
1412                 if (!task_detached(p)) {
1413                         do_notify_parent(p, p->exit_signal);
1414                         if (!task_detached(p)) {
1415                                 p->exit_state = EXIT_ZOMBIE;
1416                                 p = NULL;
1417                         }
1418                 }
1419                 write_unlock_irq(&tasklist_lock);
1420         }
1421         if (p != NULL)
1422                 release_task(p);
1423
1424         return retval;
1425 }
1426
1427 /*
1428  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1429  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1430  * the lock and this task is uninteresting.  If we return nonzero, we have
1431  * released the lock and the system call should return.
1432  */
1433 static int wait_task_stopped(int ptrace, struct task_struct *p,
1434                              int options, struct siginfo __user *infop,
1435                              int __user *stat_addr, struct rusage __user *ru)
1436 {
1437         int retval, exit_code, why;
1438         uid_t uid = 0; /* unneeded, required by compiler */
1439         pid_t pid;
1440
1441         if (!(options & WUNTRACED))
1442                 return 0;
1443
1444         exit_code = 0;
1445         spin_lock_irq(&p->sighand->siglock);
1446
1447         if (unlikely(!task_is_stopped_or_traced(p)))
1448                 goto unlock_sig;
1449
1450         if (!ptrace && p->signal->group_stop_count > 0)
1451                 /*
1452                  * A group stop is in progress and this is the group leader.
1453                  * We won't report until all threads have stopped.
1454                  */
1455                 goto unlock_sig;
1456
1457         exit_code = p->exit_code;
1458         if (!exit_code)
1459                 goto unlock_sig;
1460
1461         if (!unlikely(options & WNOWAIT))
1462                 p->exit_code = 0;
1463
1464         /* don't need the RCU readlock here as we're holding a spinlock */
1465         uid = __task_cred(p)->uid;
1466 unlock_sig:
1467         spin_unlock_irq(&p->sighand->siglock);
1468         if (!exit_code)
1469                 return 0;
1470
1471         /*
1472          * Now we are pretty sure this task is interesting.
1473          * Make sure it doesn't get reaped out from under us while we
1474          * give up the lock and then examine it below.  We don't want to
1475          * keep holding onto the tasklist_lock while we call getrusage and
1476          * possibly take page faults for user memory.
1477          */
1478         get_task_struct(p);
1479         pid = task_pid_vnr(p);
1480         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1481         read_unlock(&tasklist_lock);
1482
1483         if (unlikely(options & WNOWAIT))
1484                 return wait_noreap_copyout(p, pid, uid,
1485                                            why, exit_code,
1486                                            infop, ru);
1487
1488         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1489         if (!retval && stat_addr)
1490                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1491         if (!retval && infop)
1492                 retval = put_user(SIGCHLD, &infop->si_signo);
1493         if (!retval && infop)
1494                 retval = put_user(0, &infop->si_errno);
1495         if (!retval && infop)
1496                 retval = put_user((short)why, &infop->si_code);
1497         if (!retval && infop)
1498                 retval = put_user(exit_code, &infop->si_status);
1499         if (!retval && infop)
1500                 retval = put_user(pid, &infop->si_pid);
1501         if (!retval && infop)
1502                 retval = put_user(uid, &infop->si_uid);
1503         if (!retval)
1504                 retval = pid;
1505         put_task_struct(p);
1506
1507         BUG_ON(!retval);
1508         return retval;
1509 }
1510
1511 /*
1512  * Handle do_wait work for one task in a live, non-stopped state.
1513  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1514  * the lock and this task is uninteresting.  If we return nonzero, we have
1515  * released the lock and the system call should return.
1516  */
1517 static int wait_task_continued(struct task_struct *p, int options,
1518                                struct siginfo __user *infop,
1519                                int __user *stat_addr, struct rusage __user *ru)
1520 {
1521         int retval;
1522         pid_t pid;
1523         uid_t uid;
1524
1525         if (!unlikely(options & WCONTINUED))
1526                 return 0;
1527
1528         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1529                 return 0;
1530
1531         spin_lock_irq(&p->sighand->siglock);
1532         /* Re-check with the lock held.  */
1533         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1534                 spin_unlock_irq(&p->sighand->siglock);
1535                 return 0;
1536         }
1537         if (!unlikely(options & WNOWAIT))
1538                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1539         uid = __task_cred(p)->uid;
1540         spin_unlock_irq(&p->sighand->siglock);
1541
1542         pid = task_pid_vnr(p);
1543         get_task_struct(p);
1544         read_unlock(&tasklist_lock);
1545
1546         if (!infop) {
1547                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1548                 put_task_struct(p);
1549                 if (!retval && stat_addr)
1550                         retval = put_user(0xffff, stat_addr);
1551                 if (!retval)
1552                         retval = pid;
1553         } else {
1554                 retval = wait_noreap_copyout(p, pid, uid,
1555                                              CLD_CONTINUED, SIGCONT,
1556                                              infop, ru);
1557                 BUG_ON(retval == 0);
1558         }
1559
1560         return retval;
1561 }
1562
1563 /*
1564  * Consider @p for a wait by @parent.
1565  *
1566  * -ECHILD should be in *@notask_error before the first call.
1567  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1568  * Returns zero if the search for a child should continue;
1569  * then *@notask_error is 0 if @p is an eligible child,
1570  * or another error from security_task_wait(), or still -ECHILD.
1571  */
1572 static int wait_consider_task(struct task_struct *parent, int ptrace,
1573                               struct task_struct *p, int *notask_error,
1574                               enum pid_type type, struct pid *pid, int options,
1575                               struct siginfo __user *infop,
1576                               int __user *stat_addr, struct rusage __user *ru)
1577 {
1578         int ret = eligible_child(type, pid, options, p);
1579         if (!ret)
1580                 return ret;
1581
1582         if (unlikely(ret < 0)) {
1583                 /*
1584                  * If we have not yet seen any eligible child,
1585                  * then let this error code replace -ECHILD.
1586                  * A permission error will give the user a clue
1587                  * to look for security policy problems, rather
1588                  * than for mysterious wait bugs.
1589                  */
1590                 if (*notask_error)
1591                         *notask_error = ret;
1592         }
1593
1594         if (likely(!ptrace) && unlikely(p->ptrace)) {
1595                 /*
1596                  * This child is hidden by ptrace.
1597                  * We aren't allowed to see it now, but eventually we will.
1598                  */
1599                 *notask_error = 0;
1600                 return 0;
1601         }
1602
1603         if (p->exit_state == EXIT_DEAD)
1604                 return 0;
1605
1606         /*
1607          * We don't reap group leaders with subthreads.
1608          */
1609         if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1610                 return wait_task_zombie(p, options, infop, stat_addr, ru);
1611
1612         /*
1613          * It's stopped or running now, so it might
1614          * later continue, exit, or stop again.
1615          */
1616         *notask_error = 0;
1617
1618         if (task_is_stopped_or_traced(p))
1619                 return wait_task_stopped(ptrace, p, options,
1620                                          infop, stat_addr, ru);
1621
1622         return wait_task_continued(p, options, infop, stat_addr, ru);
1623 }
1624
1625 /*
1626  * Do the work of do_wait() for one thread in the group, @tsk.
1627  *
1628  * -ECHILD should be in *@notask_error before the first call.
1629  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1630  * Returns zero if the search for a child should continue; then
1631  * *@notask_error is 0 if there were any eligible children,
1632  * or another error from security_task_wait(), or still -ECHILD.
1633  */
1634 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1635                           enum pid_type type, struct pid *pid, int options,
1636                           struct siginfo __user *infop, int __user *stat_addr,
1637                           struct rusage __user *ru)
1638 {
1639         struct task_struct *p;
1640
1641         list_for_each_entry(p, &tsk->children, sibling) {
1642                 /*
1643                  * Do not consider detached threads.
1644                  */
1645                 if (!task_detached(p)) {
1646                         int ret = wait_consider_task(tsk, 0, p, notask_error,
1647                                                      type, pid, options,
1648                                                      infop, stat_addr, ru);
1649                         if (ret)
1650                                 return ret;
1651                 }
1652         }
1653
1654         return 0;
1655 }
1656
1657 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1658                           enum pid_type type, struct pid *pid, int options,
1659                           struct siginfo __user *infop, int __user *stat_addr,
1660                           struct rusage __user *ru)
1661 {
1662         struct task_struct *p;
1663
1664         /*
1665          * Traditionally we see ptrace'd stopped tasks regardless of options.
1666          */
1667         options |= WUNTRACED;
1668
1669         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1670                 int ret = wait_consider_task(tsk, 1, p, notask_error,
1671                                              type, pid, options,
1672                                              infop, stat_addr, ru);
1673                 if (ret)
1674                         return ret;
1675         }
1676
1677         return 0;
1678 }
1679
1680 static long do_wait(enum pid_type type, struct pid *pid, int options,
1681                     struct siginfo __user *infop, int __user *stat_addr,
1682                     struct rusage __user *ru)
1683 {
1684         DECLARE_WAITQUEUE(wait, current);
1685         struct task_struct *tsk;
1686         int retval;
1687
1688         trace_sched_process_wait(pid);
1689
1690         add_wait_queue(&current->signal->wait_chldexit,&wait);
1691 repeat:
1692         /*
1693          * If there is nothing that can match our critiera just get out.
1694          * We will clear @retval to zero if we see any child that might later
1695          * match our criteria, even if we are not able to reap it yet.
1696          */
1697         retval = -ECHILD;
1698         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1699                 goto end;
1700
1701         current->state = TASK_INTERRUPTIBLE;
1702         read_lock(&tasklist_lock);
1703         tsk = current;
1704         do {
1705                 int tsk_result = do_wait_thread(tsk, &retval,
1706                                                 type, pid, options,
1707                                                 infop, stat_addr, ru);
1708                 if (!tsk_result)
1709                         tsk_result = ptrace_do_wait(tsk, &retval,
1710                                                     type, pid, options,
1711                                                     infop, stat_addr, ru);
1712                 if (tsk_result) {
1713                         /*
1714                          * tasklist_lock is unlocked and we have a final result.
1715                          */
1716                         retval = tsk_result;
1717                         goto end;
1718                 }
1719
1720                 if (options & __WNOTHREAD)
1721                         break;
1722                 tsk = next_thread(tsk);
1723                 BUG_ON(tsk->signal != current->signal);
1724         } while (tsk != current);
1725         read_unlock(&tasklist_lock);
1726
1727         if (!retval && !(options & WNOHANG)) {
1728                 retval = -ERESTARTSYS;
1729                 if (!signal_pending(current)) {
1730                         schedule();
1731                         goto repeat;
1732                 }
1733         }
1734
1735 end:
1736         current->state = TASK_RUNNING;
1737         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1738         if (infop) {
1739                 if (retval > 0)
1740                         retval = 0;
1741                 else {
1742                         /*
1743                          * For a WNOHANG return, clear out all the fields
1744                          * we would set so the user can easily tell the
1745                          * difference.
1746                          */
1747                         if (!retval)
1748                                 retval = put_user(0, &infop->si_signo);
1749                         if (!retval)
1750                                 retval = put_user(0, &infop->si_errno);
1751                         if (!retval)
1752                                 retval = put_user(0, &infop->si_code);
1753                         if (!retval)
1754                                 retval = put_user(0, &infop->si_pid);
1755                         if (!retval)
1756                                 retval = put_user(0, &infop->si_uid);
1757                         if (!retval)
1758                                 retval = put_user(0, &infop->si_status);
1759                 }
1760         }
1761         return retval;
1762 }
1763
1764 asmlinkage long sys_waitid(int which, pid_t upid,
1765                            struct siginfo __user *infop, int options,
1766                            struct rusage __user *ru)
1767 {
1768         struct pid *pid = NULL;
1769         enum pid_type type;
1770         long ret;
1771
1772         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1773                 return -EINVAL;
1774         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1775                 return -EINVAL;
1776
1777         switch (which) {
1778         case P_ALL:
1779                 type = PIDTYPE_MAX;
1780                 break;
1781         case P_PID:
1782                 type = PIDTYPE_PID;
1783                 if (upid <= 0)
1784                         return -EINVAL;
1785                 break;
1786         case P_PGID:
1787                 type = PIDTYPE_PGID;
1788                 if (upid <= 0)
1789                         return -EINVAL;
1790                 break;
1791         default:
1792                 return -EINVAL;
1793         }
1794
1795         if (type < PIDTYPE_MAX)
1796                 pid = find_get_pid(upid);
1797         ret = do_wait(type, pid, options, infop, NULL, ru);
1798         put_pid(pid);
1799
1800         /* avoid REGPARM breakage on x86: */
1801         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1802         return ret;
1803 }
1804
1805 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1806                           int options, struct rusage __user *ru)
1807 {
1808         struct pid *pid = NULL;
1809         enum pid_type type;
1810         long ret;
1811
1812         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1813                         __WNOTHREAD|__WCLONE|__WALL))
1814                 return -EINVAL;
1815
1816         if (upid == -1)
1817                 type = PIDTYPE_MAX;
1818         else if (upid < 0) {
1819                 type = PIDTYPE_PGID;
1820                 pid = find_get_pid(-upid);
1821         } else if (upid == 0) {
1822                 type = PIDTYPE_PGID;
1823                 pid = get_pid(task_pgrp(current));
1824         } else /* upid > 0 */ {
1825                 type = PIDTYPE_PID;
1826                 pid = find_get_pid(upid);
1827         }
1828
1829         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1830         put_pid(pid);
1831
1832         /* avoid REGPARM breakage on x86: */
1833         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1834         return ret;
1835 }
1836
1837 #ifdef __ARCH_WANT_SYS_WAITPID
1838
1839 /*
1840  * sys_waitpid() remains for compatibility. waitpid() should be
1841  * implemented by calling sys_wait4() from libc.a.
1842  */
1843 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1844 {
1845         return sys_wait4(pid, stat_addr, options, NULL);
1846 }
1847
1848 #endif