tracepoints: add DECLARE_TRACE() and DEFINE_TRACE()
[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 DEFINE_TRACE(sched_process_free);
58 DEFINE_TRACE(sched_process_exit);
59 DEFINE_TRACE(sched_process_wait);
60
61 static void exit_mm(struct task_struct * tsk);
62
63 static inline int task_detached(struct task_struct *p)
64 {
65         return p->exit_signal == -1;
66 }
67
68 static void __unhash_process(struct task_struct *p)
69 {
70         nr_threads--;
71         detach_pid(p, PIDTYPE_PID);
72         if (thread_group_leader(p)) {
73                 detach_pid(p, PIDTYPE_PGID);
74                 detach_pid(p, PIDTYPE_SID);
75
76                 list_del_rcu(&p->tasks);
77                 __get_cpu_var(process_counts)--;
78         }
79         list_del_rcu(&p->thread_group);
80         list_del_init(&p->sibling);
81 }
82
83 /*
84  * This function expects the tasklist_lock write-locked.
85  */
86 static void __exit_signal(struct task_struct *tsk)
87 {
88         struct signal_struct *sig = tsk->signal;
89         struct sighand_struct *sighand;
90
91         BUG_ON(!sig);
92         BUG_ON(!atomic_read(&sig->count));
93
94         sighand = rcu_dereference(tsk->sighand);
95         spin_lock(&sighand->siglock);
96
97         posix_cpu_timers_exit(tsk);
98         if (atomic_dec_and_test(&sig->count))
99                 posix_cpu_timers_exit_group(tsk);
100         else {
101                 /*
102                  * If there is any task waiting for the group exit
103                  * then notify it:
104                  */
105                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
106                         wake_up_process(sig->group_exit_task);
107
108                 if (tsk == sig->curr_target)
109                         sig->curr_target = next_thread(tsk);
110                 /*
111                  * Accumulate here the counters for all threads but the
112                  * group leader as they die, so they can be added into
113                  * the process-wide totals when those are taken.
114                  * The group leader stays around as a zombie as long
115                  * as there are other threads.  When it gets reaped,
116                  * the exit.c code will add its counts into these totals.
117                  * We won't ever get here for the group leader, since it
118                  * will have been the last reference on the signal_struct.
119                  */
120                 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
121                 sig->min_flt += tsk->min_flt;
122                 sig->maj_flt += tsk->maj_flt;
123                 sig->nvcsw += tsk->nvcsw;
124                 sig->nivcsw += tsk->nivcsw;
125                 sig->inblock += task_io_get_inblock(tsk);
126                 sig->oublock += task_io_get_oublock(tsk);
127                 task_io_accounting_add(&sig->ioac, &tsk->ioac);
128                 sig = NULL; /* Marker for below. */
129         }
130
131         __unhash_process(tsk);
132
133         /*
134          * Do this under ->siglock, we can race with another thread
135          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
136          */
137         flush_sigqueue(&tsk->pending);
138
139         tsk->signal = NULL;
140         tsk->sighand = NULL;
141         spin_unlock(&sighand->siglock);
142
143         __cleanup_sighand(sighand);
144         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
145         if (sig) {
146                 flush_sigqueue(&sig->shared_pending);
147                 taskstats_tgid_free(sig);
148                 /*
149                  * Make sure ->signal can't go away under rq->lock,
150                  * see account_group_exec_runtime().
151                  */
152                 task_rq_unlock_wait(tsk);
153                 __cleanup_signal(sig);
154         }
155 }
156
157 static void delayed_put_task_struct(struct rcu_head *rhp)
158 {
159         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
160
161         trace_sched_process_free(tsk);
162         put_task_struct(tsk);
163 }
164
165
166 void release_task(struct task_struct * p)
167 {
168         struct task_struct *leader;
169         int zap_leader;
170 repeat:
171         tracehook_prepare_release_task(p);
172         atomic_dec(&p->user->processes);
173         proc_flush_task(p);
174         write_lock_irq(&tasklist_lock);
175         tracehook_finish_release_task(p);
176         __exit_signal(p);
177
178         /*
179          * If we are the last non-leader member of the thread
180          * group, and the leader is zombie, then notify the
181          * group leader's parent process. (if it wants notification.)
182          */
183         zap_leader = 0;
184         leader = p->group_leader;
185         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
186                 BUG_ON(task_detached(leader));
187                 do_notify_parent(leader, leader->exit_signal);
188                 /*
189                  * If we were the last child thread and the leader has
190                  * exited already, and the leader's parent ignores SIGCHLD,
191                  * then we are the one who should release the leader.
192                  *
193                  * do_notify_parent() will have marked it self-reaping in
194                  * that case.
195                  */
196                 zap_leader = task_detached(leader);
197
198                 /*
199                  * This maintains the invariant that release_task()
200                  * only runs on a task in EXIT_DEAD, just for sanity.
201                  */
202                 if (zap_leader)
203                         leader->exit_state = EXIT_DEAD;
204         }
205
206         write_unlock_irq(&tasklist_lock);
207         release_thread(p);
208         call_rcu(&p->rcu, delayed_put_task_struct);
209
210         p = leader;
211         if (unlikely(zap_leader))
212                 goto repeat;
213 }
214
215 /*
216  * This checks not only the pgrp, but falls back on the pid if no
217  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
218  * without this...
219  *
220  * The caller must hold rcu lock or the tasklist lock.
221  */
222 struct pid *session_of_pgrp(struct pid *pgrp)
223 {
224         struct task_struct *p;
225         struct pid *sid = NULL;
226
227         p = pid_task(pgrp, PIDTYPE_PGID);
228         if (p == NULL)
229                 p = pid_task(pgrp, PIDTYPE_PID);
230         if (p != NULL)
231                 sid = task_session(p);
232
233         return sid;
234 }
235
236 /*
237  * Determine if a process group is "orphaned", according to the POSIX
238  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
239  * by terminal-generated stop signals.  Newly orphaned process groups are
240  * to receive a SIGHUP and a SIGCONT.
241  *
242  * "I ask you, have you ever known what it is to be an orphan?"
243  */
244 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
245 {
246         struct task_struct *p;
247
248         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
249                 if ((p == ignored_task) ||
250                     (p->exit_state && thread_group_empty(p)) ||
251                     is_global_init(p->real_parent))
252                         continue;
253
254                 if (task_pgrp(p->real_parent) != pgrp &&
255                     task_session(p->real_parent) == task_session(p))
256                         return 0;
257         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
258
259         return 1;
260 }
261
262 int is_current_pgrp_orphaned(void)
263 {
264         int retval;
265
266         read_lock(&tasklist_lock);
267         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
268         read_unlock(&tasklist_lock);
269
270         return retval;
271 }
272
273 static int has_stopped_jobs(struct pid *pgrp)
274 {
275         int retval = 0;
276         struct task_struct *p;
277
278         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
279                 if (!task_is_stopped(p))
280                         continue;
281                 retval = 1;
282                 break;
283         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
284         return retval;
285 }
286
287 /*
288  * Check to see if any process groups have become orphaned as
289  * a result of our exiting, and if they have any stopped jobs,
290  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
291  */
292 static void
293 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
294 {
295         struct pid *pgrp = task_pgrp(tsk);
296         struct task_struct *ignored_task = tsk;
297
298         if (!parent)
299                  /* exit: our father is in a different pgrp than
300                   * we are and we were the only connection outside.
301                   */
302                 parent = tsk->real_parent;
303         else
304                 /* reparent: our child is in a different pgrp than
305                  * we are, and it was the only connection outside.
306                  */
307                 ignored_task = NULL;
308
309         if (task_pgrp(parent) != pgrp &&
310             task_session(parent) == task_session(tsk) &&
311             will_become_orphaned_pgrp(pgrp, ignored_task) &&
312             has_stopped_jobs(pgrp)) {
313                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
314                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
315         }
316 }
317
318 /**
319  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
320  *
321  * If a kernel thread is launched as a result of a system call, or if
322  * it ever exits, it should generally reparent itself to kthreadd so it
323  * isn't in the way of other processes and is correctly cleaned up on exit.
324  *
325  * The various task state such as scheduling policy and priority may have
326  * been inherited from a user process, so we reset them to sane values here.
327  *
328  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
329  */
330 static void reparent_to_kthreadd(void)
331 {
332         write_lock_irq(&tasklist_lock);
333
334         ptrace_unlink(current);
335         /* Reparent to init */
336         current->real_parent = current->parent = kthreadd_task;
337         list_move_tail(&current->sibling, &current->real_parent->children);
338
339         /* Set the exit signal to SIGCHLD so we signal init on exit */
340         current->exit_signal = SIGCHLD;
341
342         if (task_nice(current) < 0)
343                 set_user_nice(current, 0);
344         /* cpus_allowed? */
345         /* rt_priority? */
346         /* signals? */
347         security_task_reparent_to_init(current);
348         memcpy(current->signal->rlim, init_task.signal->rlim,
349                sizeof(current->signal->rlim));
350         atomic_inc(&(INIT_USER->__count));
351         write_unlock_irq(&tasklist_lock);
352         switch_uid(INIT_USER);
353 }
354
355 void __set_special_pids(struct pid *pid)
356 {
357         struct task_struct *curr = current->group_leader;
358         pid_t nr = pid_nr(pid);
359
360         if (task_session(curr) != pid) {
361                 change_pid(curr, PIDTYPE_SID, pid);
362                 set_task_session(curr, nr);
363         }
364         if (task_pgrp(curr) != pid) {
365                 change_pid(curr, PIDTYPE_PGID, pid);
366                 set_task_pgrp(curr, nr);
367         }
368 }
369
370 static void set_special_pids(struct pid *pid)
371 {
372         write_lock_irq(&tasklist_lock);
373         __set_special_pids(pid);
374         write_unlock_irq(&tasklist_lock);
375 }
376
377 /*
378  * Let kernel threads use this to say that they
379  * allow a certain signal (since daemonize() will
380  * have disabled all of them by default).
381  */
382 int allow_signal(int sig)
383 {
384         if (!valid_signal(sig) || sig < 1)
385                 return -EINVAL;
386
387         spin_lock_irq(&current->sighand->siglock);
388         sigdelset(&current->blocked, sig);
389         if (!current->mm) {
390                 /* Kernel threads handle their own signals.
391                    Let the signal code know it'll be handled, so
392                    that they don't get converted to SIGKILL or
393                    just silently dropped */
394                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
395         }
396         recalc_sigpending();
397         spin_unlock_irq(&current->sighand->siglock);
398         return 0;
399 }
400
401 EXPORT_SYMBOL(allow_signal);
402
403 int disallow_signal(int sig)
404 {
405         if (!valid_signal(sig) || sig < 1)
406                 return -EINVAL;
407
408         spin_lock_irq(&current->sighand->siglock);
409         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
410         recalc_sigpending();
411         spin_unlock_irq(&current->sighand->siglock);
412         return 0;
413 }
414
415 EXPORT_SYMBOL(disallow_signal);
416
417 /*
418  *      Put all the gunge required to become a kernel thread without
419  *      attached user resources in one place where it belongs.
420  */
421
422 void daemonize(const char *name, ...)
423 {
424         va_list args;
425         struct fs_struct *fs;
426         sigset_t blocked;
427
428         va_start(args, name);
429         vsnprintf(current->comm, sizeof(current->comm), name, args);
430         va_end(args);
431
432         /*
433          * If we were started as result of loading a module, close all of the
434          * user space pages.  We don't need them, and if we didn't close them
435          * they would be locked into memory.
436          */
437         exit_mm(current);
438         /*
439          * We don't want to have TIF_FREEZE set if the system-wide hibernation
440          * or suspend transition begins right now.
441          */
442         current->flags |= (PF_NOFREEZE | PF_KTHREAD);
443
444         if (current->nsproxy != &init_nsproxy) {
445                 get_nsproxy(&init_nsproxy);
446                 switch_task_namespaces(current, &init_nsproxy);
447         }
448         set_special_pids(&init_struct_pid);
449         proc_clear_tty(current);
450
451         /* Block and flush all signals */
452         sigfillset(&blocked);
453         sigprocmask(SIG_BLOCK, &blocked, NULL);
454         flush_signals(current);
455
456         /* Become as one with the init task */
457
458         exit_fs(current);       /* current->fs->count--; */
459         fs = init_task.fs;
460         current->fs = fs;
461         atomic_inc(&fs->count);
462
463         exit_files(current);
464         current->files = init_task.files;
465         atomic_inc(&current->files->count);
466
467         reparent_to_kthreadd();
468 }
469
470 EXPORT_SYMBOL(daemonize);
471
472 static void close_files(struct files_struct * files)
473 {
474         int i, j;
475         struct fdtable *fdt;
476
477         j = 0;
478
479         /*
480          * It is safe to dereference the fd table without RCU or
481          * ->file_lock because this is the last reference to the
482          * files structure.
483          */
484         fdt = files_fdtable(files);
485         for (;;) {
486                 unsigned long set;
487                 i = j * __NFDBITS;
488                 if (i >= fdt->max_fds)
489                         break;
490                 set = fdt->open_fds->fds_bits[j++];
491                 while (set) {
492                         if (set & 1) {
493                                 struct file * file = xchg(&fdt->fd[i], NULL);
494                                 if (file) {
495                                         filp_close(file, files);
496                                         cond_resched();
497                                 }
498                         }
499                         i++;
500                         set >>= 1;
501                 }
502         }
503 }
504
505 struct files_struct *get_files_struct(struct task_struct *task)
506 {
507         struct files_struct *files;
508
509         task_lock(task);
510         files = task->files;
511         if (files)
512                 atomic_inc(&files->count);
513         task_unlock(task);
514
515         return files;
516 }
517
518 void put_files_struct(struct files_struct *files)
519 {
520         struct fdtable *fdt;
521
522         if (atomic_dec_and_test(&files->count)) {
523                 close_files(files);
524                 /*
525                  * Free the fd and fdset arrays if we expanded them.
526                  * If the fdtable was embedded, pass files for freeing
527                  * at the end of the RCU grace period. Otherwise,
528                  * you can free files immediately.
529                  */
530                 fdt = files_fdtable(files);
531                 if (fdt != &files->fdtab)
532                         kmem_cache_free(files_cachep, files);
533                 free_fdtable(fdt);
534         }
535 }
536
537 void reset_files_struct(struct files_struct *files)
538 {
539         struct task_struct *tsk = current;
540         struct files_struct *old;
541
542         old = tsk->files;
543         task_lock(tsk);
544         tsk->files = files;
545         task_unlock(tsk);
546         put_files_struct(old);
547 }
548
549 void exit_files(struct task_struct *tsk)
550 {
551         struct files_struct * files = tsk->files;
552
553         if (files) {
554                 task_lock(tsk);
555                 tsk->files = NULL;
556                 task_unlock(tsk);
557                 put_files_struct(files);
558         }
559 }
560
561 void put_fs_struct(struct fs_struct *fs)
562 {
563         /* No need to hold fs->lock if we are killing it */
564         if (atomic_dec_and_test(&fs->count)) {
565                 path_put(&fs->root);
566                 path_put(&fs->pwd);
567                 kmem_cache_free(fs_cachep, fs);
568         }
569 }
570
571 void exit_fs(struct task_struct *tsk)
572 {
573         struct fs_struct * fs = tsk->fs;
574
575         if (fs) {
576                 task_lock(tsk);
577                 tsk->fs = NULL;
578                 task_unlock(tsk);
579                 put_fs_struct(fs);
580         }
581 }
582
583 EXPORT_SYMBOL_GPL(exit_fs);
584
585 #ifdef CONFIG_MM_OWNER
586 /*
587  * Task p is exiting and it owned mm, lets find a new owner for it
588  */
589 static inline int
590 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
591 {
592         /*
593          * If there are other users of the mm and the owner (us) is exiting
594          * we need to find a new owner to take on the responsibility.
595          */
596         if (atomic_read(&mm->mm_users) <= 1)
597                 return 0;
598         if (mm->owner != p)
599                 return 0;
600         return 1;
601 }
602
603 void mm_update_next_owner(struct mm_struct *mm)
604 {
605         struct task_struct *c, *g, *p = current;
606
607 retry:
608         if (!mm_need_new_owner(mm, p))
609                 return;
610
611         read_lock(&tasklist_lock);
612         /*
613          * Search in the children
614          */
615         list_for_each_entry(c, &p->children, sibling) {
616                 if (c->mm == mm)
617                         goto assign_new_owner;
618         }
619
620         /*
621          * Search in the siblings
622          */
623         list_for_each_entry(c, &p->parent->children, sibling) {
624                 if (c->mm == mm)
625                         goto assign_new_owner;
626         }
627
628         /*
629          * Search through everything else. We should not get
630          * here often
631          */
632         do_each_thread(g, c) {
633                 if (c->mm == mm)
634                         goto assign_new_owner;
635         } while_each_thread(g, c);
636
637         read_unlock(&tasklist_lock);
638         /*
639          * We found no owner yet mm_users > 1: this implies that we are
640          * most likely racing with swapoff (try_to_unuse()) or /proc or
641          * ptrace or page migration (get_task_mm()).  Mark owner as NULL,
642          * so that subsystems can understand the callback and take action.
643          */
644         down_write(&mm->mmap_sem);
645         cgroup_mm_owner_callbacks(mm->owner, NULL);
646         mm->owner = NULL;
647         up_write(&mm->mmap_sem);
648         return;
649
650 assign_new_owner:
651         BUG_ON(c == p);
652         get_task_struct(c);
653         read_unlock(&tasklist_lock);
654         down_write(&mm->mmap_sem);
655         /*
656          * The task_lock protects c->mm from changing.
657          * We always want mm->owner->mm == mm
658          */
659         task_lock(c);
660         if (c->mm != mm) {
661                 task_unlock(c);
662                 up_write(&mm->mmap_sem);
663                 put_task_struct(c);
664                 goto retry;
665         }
666         cgroup_mm_owner_callbacks(mm->owner, c);
667         mm->owner = c;
668         task_unlock(c);
669         up_write(&mm->mmap_sem);
670         put_task_struct(c);
671 }
672 #endif /* CONFIG_MM_OWNER */
673
674 /*
675  * Turn us into a lazy TLB process if we
676  * aren't already..
677  */
678 static void exit_mm(struct task_struct * tsk)
679 {
680         struct mm_struct *mm = tsk->mm;
681         struct core_state *core_state;
682
683         mm_release(tsk, mm);
684         if (!mm)
685                 return;
686         /*
687          * Serialize with any possible pending coredump.
688          * We must hold mmap_sem around checking core_state
689          * and clearing tsk->mm.  The core-inducing thread
690          * will increment ->nr_threads for each thread in the
691          * group with ->mm != NULL.
692          */
693         down_read(&mm->mmap_sem);
694         core_state = mm->core_state;
695         if (core_state) {
696                 struct core_thread self;
697                 up_read(&mm->mmap_sem);
698
699                 self.task = tsk;
700                 self.next = xchg(&core_state->dumper.next, &self);
701                 /*
702                  * Implies mb(), the result of xchg() must be visible
703                  * to core_state->dumper.
704                  */
705                 if (atomic_dec_and_test(&core_state->nr_threads))
706                         complete(&core_state->startup);
707
708                 for (;;) {
709                         set_task_state(tsk, TASK_UNINTERRUPTIBLE);
710                         if (!self.task) /* see coredump_finish() */
711                                 break;
712                         schedule();
713                 }
714                 __set_task_state(tsk, TASK_RUNNING);
715                 down_read(&mm->mmap_sem);
716         }
717         atomic_inc(&mm->mm_count);
718         BUG_ON(mm != tsk->active_mm);
719         /* more a memory barrier than a real lock */
720         task_lock(tsk);
721         tsk->mm = NULL;
722         up_read(&mm->mmap_sem);
723         enter_lazy_tlb(mm, current);
724         /* We don't want this task to be frozen prematurely */
725         clear_freeze_flag(tsk);
726         task_unlock(tsk);
727         mm_update_next_owner(mm);
728         mmput(mm);
729 }
730
731 /*
732  * Return nonzero if @parent's children should reap themselves.
733  *
734  * Called with write_lock_irq(&tasklist_lock) held.
735  */
736 static int ignoring_children(struct task_struct *parent)
737 {
738         int ret;
739         struct sighand_struct *psig = parent->sighand;
740         unsigned long flags;
741         spin_lock_irqsave(&psig->siglock, flags);
742         ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
743                (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
744         spin_unlock_irqrestore(&psig->siglock, flags);
745         return ret;
746 }
747
748 /*
749  * Detach all tasks we were using ptrace on.
750  * Any that need to be release_task'd are put on the @dead list.
751  *
752  * Called with write_lock(&tasklist_lock) held.
753  */
754 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
755 {
756         struct task_struct *p, *n;
757         int ign = -1;
758
759         list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
760                 __ptrace_unlink(p);
761
762                 if (p->exit_state != EXIT_ZOMBIE)
763                         continue;
764
765                 /*
766                  * If it's a zombie, our attachedness prevented normal
767                  * parent notification or self-reaping.  Do notification
768                  * now if it would have happened earlier.  If it should
769                  * reap itself, add it to the @dead list.  We can't call
770                  * release_task() here because we already hold tasklist_lock.
771                  *
772                  * If it's our own child, there is no notification to do.
773                  * But if our normal children self-reap, then this child
774                  * was prevented by ptrace and we must reap it now.
775                  */
776                 if (!task_detached(p) && thread_group_empty(p)) {
777                         if (!same_thread_group(p->real_parent, parent))
778                                 do_notify_parent(p, p->exit_signal);
779                         else {
780                                 if (ign < 0)
781                                         ign = ignoring_children(parent);
782                                 if (ign)
783                                         p->exit_signal = -1;
784                         }
785                 }
786
787                 if (task_detached(p)) {
788                         /*
789                          * Mark it as in the process of being reaped.
790                          */
791                         p->exit_state = EXIT_DEAD;
792                         list_add(&p->ptrace_entry, dead);
793                 }
794         }
795 }
796
797 /*
798  * Finish up exit-time ptrace cleanup.
799  *
800  * Called without locks.
801  */
802 static void ptrace_exit_finish(struct task_struct *parent,
803                                struct list_head *dead)
804 {
805         struct task_struct *p, *n;
806
807         BUG_ON(!list_empty(&parent->ptraced));
808
809         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
810                 list_del_init(&p->ptrace_entry);
811                 release_task(p);
812         }
813 }
814
815 static void reparent_thread(struct task_struct *p, struct task_struct *father)
816 {
817         if (p->pdeath_signal)
818                 /* We already hold the tasklist_lock here.  */
819                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
820
821         list_move_tail(&p->sibling, &p->real_parent->children);
822
823         /* If this is a threaded reparent there is no need to
824          * notify anyone anything has happened.
825          */
826         if (same_thread_group(p->real_parent, father))
827                 return;
828
829         /* We don't want people slaying init.  */
830         if (!task_detached(p))
831                 p->exit_signal = SIGCHLD;
832
833         /* If we'd notified the old parent about this child's death,
834          * also notify the new parent.
835          */
836         if (!ptrace_reparented(p) &&
837             p->exit_state == EXIT_ZOMBIE &&
838             !task_detached(p) && thread_group_empty(p))
839                 do_notify_parent(p, p->exit_signal);
840
841         kill_orphaned_pgrp(p, father);
842 }
843
844 /*
845  * When we die, we re-parent all our children.
846  * Try to give them to another thread in our thread
847  * group, and if no such member exists, give it to
848  * the child reaper process (ie "init") in our pid
849  * space.
850  */
851 static struct task_struct *find_new_reaper(struct task_struct *father)
852 {
853         struct pid_namespace *pid_ns = task_active_pid_ns(father);
854         struct task_struct *thread;
855
856         thread = father;
857         while_each_thread(father, thread) {
858                 if (thread->flags & PF_EXITING)
859                         continue;
860                 if (unlikely(pid_ns->child_reaper == father))
861                         pid_ns->child_reaper = thread;
862                 return thread;
863         }
864
865         if (unlikely(pid_ns->child_reaper == father)) {
866                 write_unlock_irq(&tasklist_lock);
867                 if (unlikely(pid_ns == &init_pid_ns))
868                         panic("Attempted to kill init!");
869
870                 zap_pid_ns_processes(pid_ns);
871                 write_lock_irq(&tasklist_lock);
872                 /*
873                  * We can not clear ->child_reaper or leave it alone.
874                  * There may by stealth EXIT_DEAD tasks on ->children,
875                  * forget_original_parent() must move them somewhere.
876                  */
877                 pid_ns->child_reaper = init_pid_ns.child_reaper;
878         }
879
880         return pid_ns->child_reaper;
881 }
882
883 static void forget_original_parent(struct task_struct *father)
884 {
885         struct task_struct *p, *n, *reaper;
886         LIST_HEAD(ptrace_dead);
887
888         write_lock_irq(&tasklist_lock);
889         reaper = find_new_reaper(father);
890         /*
891          * First clean up ptrace if we were using it.
892          */
893         ptrace_exit(father, &ptrace_dead);
894
895         list_for_each_entry_safe(p, n, &father->children, sibling) {
896                 p->real_parent = reaper;
897                 if (p->parent == father) {
898                         BUG_ON(p->ptrace);
899                         p->parent = p->real_parent;
900                 }
901                 reparent_thread(p, father);
902         }
903
904         write_unlock_irq(&tasklist_lock);
905         BUG_ON(!list_empty(&father->children));
906
907         ptrace_exit_finish(father, &ptrace_dead);
908 }
909
910 /*
911  * Send signals to all our closest relatives so that they know
912  * to properly mourn us..
913  */
914 static void exit_notify(struct task_struct *tsk, int group_dead)
915 {
916         int signal;
917         void *cookie;
918
919         /*
920          * This does two things:
921          *
922          * A.  Make init inherit all the child processes
923          * B.  Check to see if any process groups have become orphaned
924          *      as a result of our exiting, and if they have any stopped
925          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
926          */
927         forget_original_parent(tsk);
928         exit_task_namespaces(tsk);
929
930         write_lock_irq(&tasklist_lock);
931         if (group_dead)
932                 kill_orphaned_pgrp(tsk->group_leader, NULL);
933
934         /* Let father know we died
935          *
936          * Thread signals are configurable, but you aren't going to use
937          * that to send signals to arbitary processes.
938          * That stops right now.
939          *
940          * If the parent exec id doesn't match the exec id we saved
941          * when we started then we know the parent has changed security
942          * domain.
943          *
944          * If our self_exec id doesn't match our parent_exec_id then
945          * we have changed execution domain as these two values started
946          * the same after a fork.
947          */
948         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
949             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
950              tsk->self_exec_id != tsk->parent_exec_id) &&
951             !capable(CAP_KILL))
952                 tsk->exit_signal = SIGCHLD;
953
954         signal = tracehook_notify_death(tsk, &cookie, group_dead);
955         if (signal >= 0)
956                 signal = do_notify_parent(tsk, signal);
957
958         tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
959
960         /* mt-exec, de_thread() is waiting for us */
961         if (thread_group_leader(tsk) &&
962             tsk->signal->group_exit_task &&
963             tsk->signal->notify_count < 0)
964                 wake_up_process(tsk->signal->group_exit_task);
965
966         write_unlock_irq(&tasklist_lock);
967
968         tracehook_report_death(tsk, signal, cookie, group_dead);
969
970         /* If the process is dead, release it - nobody will wait for it */
971         if (signal == DEATH_REAP)
972                 release_task(tsk);
973 }
974
975 #ifdef CONFIG_DEBUG_STACK_USAGE
976 static void check_stack_usage(void)
977 {
978         static DEFINE_SPINLOCK(low_water_lock);
979         static int lowest_to_date = THREAD_SIZE;
980         unsigned long *n = end_of_stack(current);
981         unsigned long free;
982
983         while (*n == 0)
984                 n++;
985         free = (unsigned long)n - (unsigned long)end_of_stack(current);
986
987         if (free >= lowest_to_date)
988                 return;
989
990         spin_lock(&low_water_lock);
991         if (free < lowest_to_date) {
992                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
993                                 "left\n",
994                                 current->comm, free);
995                 lowest_to_date = free;
996         }
997         spin_unlock(&low_water_lock);
998 }
999 #else
1000 static inline void check_stack_usage(void) {}
1001 #endif
1002
1003 NORET_TYPE void do_exit(long code)
1004 {
1005         struct task_struct *tsk = current;
1006         int group_dead;
1007
1008         profile_task_exit(tsk);
1009
1010         WARN_ON(atomic_read(&tsk->fs_excl));
1011
1012         if (unlikely(in_interrupt()))
1013                 panic("Aiee, killing interrupt handler!");
1014         if (unlikely(!tsk->pid))
1015                 panic("Attempted to kill the idle task!");
1016
1017         tracehook_report_exit(&code);
1018
1019         /*
1020          * We're taking recursive faults here in do_exit. Safest is to just
1021          * leave this task alone and wait for reboot.
1022          */
1023         if (unlikely(tsk->flags & PF_EXITING)) {
1024                 printk(KERN_ALERT
1025                         "Fixing recursive fault but reboot is needed!\n");
1026                 /*
1027                  * We can do this unlocked here. The futex code uses
1028                  * this flag just to verify whether the pi state
1029                  * cleanup has been done or not. In the worst case it
1030                  * loops once more. We pretend that the cleanup was
1031                  * done as there is no way to return. Either the
1032                  * OWNER_DIED bit is set by now or we push the blocked
1033                  * task into the wait for ever nirwana as well.
1034                  */
1035                 tsk->flags |= PF_EXITPIDONE;
1036                 if (tsk->io_context)
1037                         exit_io_context();
1038                 set_current_state(TASK_UNINTERRUPTIBLE);
1039                 schedule();
1040         }
1041
1042         exit_signals(tsk);  /* sets PF_EXITING */
1043         /*
1044          * tsk->flags are checked in the futex code to protect against
1045          * an exiting task cleaning up the robust pi futexes.
1046          */
1047         smp_mb();
1048         spin_unlock_wait(&tsk->pi_lock);
1049
1050         if (unlikely(in_atomic()))
1051                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1052                                 current->comm, task_pid_nr(current),
1053                                 preempt_count());
1054
1055         acct_update_integrals(tsk);
1056         if (tsk->mm) {
1057                 update_hiwater_rss(tsk->mm);
1058                 update_hiwater_vm(tsk->mm);
1059         }
1060         group_dead = atomic_dec_and_test(&tsk->signal->live);
1061         if (group_dead) {
1062                 hrtimer_cancel(&tsk->signal->real_timer);
1063                 exit_itimers(tsk->signal);
1064         }
1065         acct_collect(code, group_dead);
1066 #ifdef CONFIG_FUTEX
1067         if (unlikely(tsk->robust_list))
1068                 exit_robust_list(tsk);
1069 #ifdef CONFIG_COMPAT
1070         if (unlikely(tsk->compat_robust_list))
1071                 compat_exit_robust_list(tsk);
1072 #endif
1073 #endif
1074         if (group_dead)
1075                 tty_audit_exit();
1076         if (unlikely(tsk->audit_context))
1077                 audit_free(tsk);
1078
1079         tsk->exit_code = code;
1080         taskstats_exit(tsk, group_dead);
1081
1082         exit_mm(tsk);
1083
1084         if (group_dead)
1085                 acct_process();
1086         trace_sched_process_exit(tsk);
1087
1088         exit_sem(tsk);
1089         exit_files(tsk);
1090         exit_fs(tsk);
1091         check_stack_usage();
1092         exit_thread();
1093         cgroup_exit(tsk, 1);
1094         exit_keys(tsk);
1095
1096         if (group_dead && tsk->signal->leader)
1097                 disassociate_ctty(1);
1098
1099         module_put(task_thread_info(tsk)->exec_domain->module);
1100         if (tsk->binfmt)
1101                 module_put(tsk->binfmt->module);
1102
1103         proc_exit_connector(tsk);
1104         exit_notify(tsk, group_dead);
1105 #ifdef CONFIG_NUMA
1106         mpol_put(tsk->mempolicy);
1107         tsk->mempolicy = NULL;
1108 #endif
1109 #ifdef CONFIG_FUTEX
1110         /*
1111          * This must happen late, after the PID is not
1112          * hashed anymore:
1113          */
1114         if (unlikely(!list_empty(&tsk->pi_state_list)))
1115                 exit_pi_state_list(tsk);
1116         if (unlikely(current->pi_state_cache))
1117                 kfree(current->pi_state_cache);
1118 #endif
1119         /*
1120          * Make sure we are holding no locks:
1121          */
1122         debug_check_no_locks_held(tsk);
1123         /*
1124          * We can do this unlocked here. The futex code uses this flag
1125          * just to verify whether the pi state cleanup has been done
1126          * or not. In the worst case it loops once more.
1127          */
1128         tsk->flags |= PF_EXITPIDONE;
1129
1130         if (tsk->io_context)
1131                 exit_io_context();
1132
1133         if (tsk->splice_pipe)
1134                 __free_pipe_info(tsk->splice_pipe);
1135
1136         preempt_disable();
1137         /* causes final put_task_struct in finish_task_switch(). */
1138         tsk->state = TASK_DEAD;
1139
1140         schedule();
1141         BUG();
1142         /* Avoid "noreturn function does return".  */
1143         for (;;)
1144                 cpu_relax();    /* For when BUG is null */
1145 }
1146
1147 EXPORT_SYMBOL_GPL(do_exit);
1148
1149 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1150 {
1151         if (comp)
1152                 complete(comp);
1153
1154         do_exit(code);
1155 }
1156
1157 EXPORT_SYMBOL(complete_and_exit);
1158
1159 asmlinkage long sys_exit(int error_code)
1160 {
1161         do_exit((error_code&0xff)<<8);
1162 }
1163
1164 /*
1165  * Take down every thread in the group.  This is called by fatal signals
1166  * as well as by sys_exit_group (below).
1167  */
1168 NORET_TYPE void
1169 do_group_exit(int exit_code)
1170 {
1171         struct signal_struct *sig = current->signal;
1172
1173         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1174
1175         if (signal_group_exit(sig))
1176                 exit_code = sig->group_exit_code;
1177         else if (!thread_group_empty(current)) {
1178                 struct sighand_struct *const sighand = current->sighand;
1179                 spin_lock_irq(&sighand->siglock);
1180                 if (signal_group_exit(sig))
1181                         /* Another thread got here before we took the lock.  */
1182                         exit_code = sig->group_exit_code;
1183                 else {
1184                         sig->group_exit_code = exit_code;
1185                         sig->flags = SIGNAL_GROUP_EXIT;
1186                         zap_other_threads(current);
1187                 }
1188                 spin_unlock_irq(&sighand->siglock);
1189         }
1190
1191         do_exit(exit_code);
1192         /* NOTREACHED */
1193 }
1194
1195 /*
1196  * this kills every thread in the thread group. Note that any externally
1197  * wait4()-ing process will get the correct exit code - even if this
1198  * thread is not the thread group leader.
1199  */
1200 asmlinkage void sys_exit_group(int error_code)
1201 {
1202         do_group_exit((error_code & 0xff) << 8);
1203 }
1204
1205 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1206 {
1207         struct pid *pid = NULL;
1208         if (type == PIDTYPE_PID)
1209                 pid = task->pids[type].pid;
1210         else if (type < PIDTYPE_MAX)
1211                 pid = task->group_leader->pids[type].pid;
1212         return pid;
1213 }
1214
1215 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1216                           struct task_struct *p)
1217 {
1218         int err;
1219
1220         if (type < PIDTYPE_MAX) {
1221                 if (task_pid_type(p, type) != pid)
1222                         return 0;
1223         }
1224
1225         /* Wait for all children (clone and not) if __WALL is set;
1226          * otherwise, wait for clone children *only* if __WCLONE is
1227          * set; otherwise, wait for non-clone children *only*.  (Note:
1228          * A "clone" child here is one that reports to its parent
1229          * using a signal other than SIGCHLD.) */
1230         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1231             && !(options & __WALL))
1232                 return 0;
1233
1234         err = security_task_wait(p);
1235         if (err)
1236                 return err;
1237
1238         return 1;
1239 }
1240
1241 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1242                                int why, int status,
1243                                struct siginfo __user *infop,
1244                                struct rusage __user *rusagep)
1245 {
1246         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1247
1248         put_task_struct(p);
1249         if (!retval)
1250                 retval = put_user(SIGCHLD, &infop->si_signo);
1251         if (!retval)
1252                 retval = put_user(0, &infop->si_errno);
1253         if (!retval)
1254                 retval = put_user((short)why, &infop->si_code);
1255         if (!retval)
1256                 retval = put_user(pid, &infop->si_pid);
1257         if (!retval)
1258                 retval = put_user(uid, &infop->si_uid);
1259         if (!retval)
1260                 retval = put_user(status, &infop->si_status);
1261         if (!retval)
1262                 retval = pid;
1263         return retval;
1264 }
1265
1266 /*
1267  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1268  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1269  * the lock and this task is uninteresting.  If we return nonzero, we have
1270  * released the lock and the system call should return.
1271  */
1272 static int wait_task_zombie(struct task_struct *p, int options,
1273                             struct siginfo __user *infop,
1274                             int __user *stat_addr, struct rusage __user *ru)
1275 {
1276         unsigned long state;
1277         int retval, status, traced;
1278         pid_t pid = task_pid_vnr(p);
1279
1280         if (!likely(options & WEXITED))
1281                 return 0;
1282
1283         if (unlikely(options & WNOWAIT)) {
1284                 uid_t uid = p->uid;
1285                 int exit_code = p->exit_code;
1286                 int why, status;
1287
1288                 get_task_struct(p);
1289                 read_unlock(&tasklist_lock);
1290                 if ((exit_code & 0x7f) == 0) {
1291                         why = CLD_EXITED;
1292                         status = exit_code >> 8;
1293                 } else {
1294                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1295                         status = exit_code & 0x7f;
1296                 }
1297                 return wait_noreap_copyout(p, pid, uid, why,
1298                                            status, infop, ru);
1299         }
1300
1301         /*
1302          * Try to move the task's state to DEAD
1303          * only one thread is allowed to do this:
1304          */
1305         state = xchg(&p->exit_state, EXIT_DEAD);
1306         if (state != EXIT_ZOMBIE) {
1307                 BUG_ON(state != EXIT_DEAD);
1308                 return 0;
1309         }
1310
1311         traced = ptrace_reparented(p);
1312
1313         if (likely(!traced)) {
1314                 struct signal_struct *psig;
1315                 struct signal_struct *sig;
1316                 struct task_cputime cputime;
1317
1318                 /*
1319                  * The resource counters for the group leader are in its
1320                  * own task_struct.  Those for dead threads in the group
1321                  * are in its signal_struct, as are those for the child
1322                  * processes it has previously reaped.  All these
1323                  * accumulate in the parent's signal_struct c* fields.
1324                  *
1325                  * We don't bother to take a lock here to protect these
1326                  * p->signal fields, because they are only touched by
1327                  * __exit_signal, which runs with tasklist_lock
1328                  * write-locked anyway, and so is excluded here.  We do
1329                  * need to protect the access to p->parent->signal fields,
1330                  * as other threads in the parent group can be right
1331                  * here reaping other children at the same time.
1332                  *
1333                  * We use thread_group_cputime() to get times for the thread
1334                  * group, which consolidates times for all threads in the
1335                  * group including the group leader.
1336                  */
1337                 spin_lock_irq(&p->parent->sighand->siglock);
1338                 psig = p->parent->signal;
1339                 sig = p->signal;
1340                 thread_group_cputime(p, &cputime);
1341                 psig->cutime =
1342                         cputime_add(psig->cutime,
1343                         cputime_add(cputime.utime,
1344                                     sig->cutime));
1345                 psig->cstime =
1346                         cputime_add(psig->cstime,
1347                         cputime_add(cputime.stime,
1348                                     sig->cstime));
1349                 psig->cgtime =
1350                         cputime_add(psig->cgtime,
1351                         cputime_add(p->gtime,
1352                         cputime_add(sig->gtime,
1353                                     sig->cgtime)));
1354                 psig->cmin_flt +=
1355                         p->min_flt + sig->min_flt + sig->cmin_flt;
1356                 psig->cmaj_flt +=
1357                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1358                 psig->cnvcsw +=
1359                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1360                 psig->cnivcsw +=
1361                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1362                 psig->cinblock +=
1363                         task_io_get_inblock(p) +
1364                         sig->inblock + sig->cinblock;
1365                 psig->coublock +=
1366                         task_io_get_oublock(p) +
1367                         sig->oublock + sig->coublock;
1368                 task_io_accounting_add(&psig->ioac, &p->ioac);
1369                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1370                 spin_unlock_irq(&p->parent->sighand->siglock);
1371         }
1372
1373         /*
1374          * Now we are sure this task is interesting, and no other
1375          * thread can reap it because we set its state to EXIT_DEAD.
1376          */
1377         read_unlock(&tasklist_lock);
1378
1379         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1380         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1381                 ? p->signal->group_exit_code : p->exit_code;
1382         if (!retval && stat_addr)
1383                 retval = put_user(status, stat_addr);
1384         if (!retval && infop)
1385                 retval = put_user(SIGCHLD, &infop->si_signo);
1386         if (!retval && infop)
1387                 retval = put_user(0, &infop->si_errno);
1388         if (!retval && infop) {
1389                 int why;
1390
1391                 if ((status & 0x7f) == 0) {
1392                         why = CLD_EXITED;
1393                         status >>= 8;
1394                 } else {
1395                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1396                         status &= 0x7f;
1397                 }
1398                 retval = put_user((short)why, &infop->si_code);
1399                 if (!retval)
1400                         retval = put_user(status, &infop->si_status);
1401         }
1402         if (!retval && infop)
1403                 retval = put_user(pid, &infop->si_pid);
1404         if (!retval && infop)
1405                 retval = put_user(p->uid, &infop->si_uid);
1406         if (!retval)
1407                 retval = pid;
1408
1409         if (traced) {
1410                 write_lock_irq(&tasklist_lock);
1411                 /* We dropped tasklist, ptracer could die and untrace */
1412                 ptrace_unlink(p);
1413                 /*
1414                  * If this is not a detached task, notify the parent.
1415                  * If it's still not detached after that, don't release
1416                  * it now.
1417                  */
1418                 if (!task_detached(p)) {
1419                         do_notify_parent(p, p->exit_signal);
1420                         if (!task_detached(p)) {
1421                                 p->exit_state = EXIT_ZOMBIE;
1422                                 p = NULL;
1423                         }
1424                 }
1425                 write_unlock_irq(&tasklist_lock);
1426         }
1427         if (p != NULL)
1428                 release_task(p);
1429
1430         return retval;
1431 }
1432
1433 /*
1434  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1435  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1436  * the lock and this task is uninteresting.  If we return nonzero, we have
1437  * released the lock and the system call should return.
1438  */
1439 static int wait_task_stopped(int ptrace, struct task_struct *p,
1440                              int options, struct siginfo __user *infop,
1441                              int __user *stat_addr, struct rusage __user *ru)
1442 {
1443         int retval, exit_code, why;
1444         uid_t uid = 0; /* unneeded, required by compiler */
1445         pid_t pid;
1446
1447         if (!(options & WUNTRACED))
1448                 return 0;
1449
1450         exit_code = 0;
1451         spin_lock_irq(&p->sighand->siglock);
1452
1453         if (unlikely(!task_is_stopped_or_traced(p)))
1454                 goto unlock_sig;
1455
1456         if (!ptrace && p->signal->group_stop_count > 0)
1457                 /*
1458                  * A group stop is in progress and this is the group leader.
1459                  * We won't report until all threads have stopped.
1460                  */
1461                 goto unlock_sig;
1462
1463         exit_code = p->exit_code;
1464         if (!exit_code)
1465                 goto unlock_sig;
1466
1467         if (!unlikely(options & WNOWAIT))
1468                 p->exit_code = 0;
1469
1470         uid = p->uid;
1471 unlock_sig:
1472         spin_unlock_irq(&p->sighand->siglock);
1473         if (!exit_code)
1474                 return 0;
1475
1476         /*
1477          * Now we are pretty sure this task is interesting.
1478          * Make sure it doesn't get reaped out from under us while we
1479          * give up the lock and then examine it below.  We don't want to
1480          * keep holding onto the tasklist_lock while we call getrusage and
1481          * possibly take page faults for user memory.
1482          */
1483         get_task_struct(p);
1484         pid = task_pid_vnr(p);
1485         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1486         read_unlock(&tasklist_lock);
1487
1488         if (unlikely(options & WNOWAIT))
1489                 return wait_noreap_copyout(p, pid, uid,
1490                                            why, exit_code,
1491                                            infop, ru);
1492
1493         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1494         if (!retval && stat_addr)
1495                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1496         if (!retval && infop)
1497                 retval = put_user(SIGCHLD, &infop->si_signo);
1498         if (!retval && infop)
1499                 retval = put_user(0, &infop->si_errno);
1500         if (!retval && infop)
1501                 retval = put_user((short)why, &infop->si_code);
1502         if (!retval && infop)
1503                 retval = put_user(exit_code, &infop->si_status);
1504         if (!retval && infop)
1505                 retval = put_user(pid, &infop->si_pid);
1506         if (!retval && infop)
1507                 retval = put_user(uid, &infop->si_uid);
1508         if (!retval)
1509                 retval = pid;
1510         put_task_struct(p);
1511
1512         BUG_ON(!retval);
1513         return retval;
1514 }
1515
1516 /*
1517  * Handle do_wait work for one task in a live, non-stopped state.
1518  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1519  * the lock and this task is uninteresting.  If we return nonzero, we have
1520  * released the lock and the system call should return.
1521  */
1522 static int wait_task_continued(struct task_struct *p, int options,
1523                                struct siginfo __user *infop,
1524                                int __user *stat_addr, struct rusage __user *ru)
1525 {
1526         int retval;
1527         pid_t pid;
1528         uid_t uid;
1529
1530         if (!unlikely(options & WCONTINUED))
1531                 return 0;
1532
1533         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1534                 return 0;
1535
1536         spin_lock_irq(&p->sighand->siglock);
1537         /* Re-check with the lock held.  */
1538         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1539                 spin_unlock_irq(&p->sighand->siglock);
1540                 return 0;
1541         }
1542         if (!unlikely(options & WNOWAIT))
1543                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1544         spin_unlock_irq(&p->sighand->siglock);
1545
1546         pid = task_pid_vnr(p);
1547         uid = p->uid;
1548         get_task_struct(p);
1549         read_unlock(&tasklist_lock);
1550
1551         if (!infop) {
1552                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1553                 put_task_struct(p);
1554                 if (!retval && stat_addr)
1555                         retval = put_user(0xffff, stat_addr);
1556                 if (!retval)
1557                         retval = pid;
1558         } else {
1559                 retval = wait_noreap_copyout(p, pid, uid,
1560                                              CLD_CONTINUED, SIGCONT,
1561                                              infop, ru);
1562                 BUG_ON(retval == 0);
1563         }
1564
1565         return retval;
1566 }
1567
1568 /*
1569  * Consider @p for a wait by @parent.
1570  *
1571  * -ECHILD should be in *@notask_error before the first call.
1572  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1573  * Returns zero if the search for a child should continue;
1574  * then *@notask_error is 0 if @p is an eligible child,
1575  * or another error from security_task_wait(), or still -ECHILD.
1576  */
1577 static int wait_consider_task(struct task_struct *parent, int ptrace,
1578                               struct task_struct *p, int *notask_error,
1579                               enum pid_type type, struct pid *pid, int options,
1580                               struct siginfo __user *infop,
1581                               int __user *stat_addr, struct rusage __user *ru)
1582 {
1583         int ret = eligible_child(type, pid, options, p);
1584         if (!ret)
1585                 return ret;
1586
1587         if (unlikely(ret < 0)) {
1588                 /*
1589                  * If we have not yet seen any eligible child,
1590                  * then let this error code replace -ECHILD.
1591                  * A permission error will give the user a clue
1592                  * to look for security policy problems, rather
1593                  * than for mysterious wait bugs.
1594                  */
1595                 if (*notask_error)
1596                         *notask_error = ret;
1597         }
1598
1599         if (likely(!ptrace) && unlikely(p->ptrace)) {
1600                 /*
1601                  * This child is hidden by ptrace.
1602                  * We aren't allowed to see it now, but eventually we will.
1603                  */
1604                 *notask_error = 0;
1605                 return 0;
1606         }
1607
1608         if (p->exit_state == EXIT_DEAD)
1609                 return 0;
1610
1611         /*
1612          * We don't reap group leaders with subthreads.
1613          */
1614         if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1615                 return wait_task_zombie(p, options, infop, stat_addr, ru);
1616
1617         /*
1618          * It's stopped or running now, so it might
1619          * later continue, exit, or stop again.
1620          */
1621         *notask_error = 0;
1622
1623         if (task_is_stopped_or_traced(p))
1624                 return wait_task_stopped(ptrace, p, options,
1625                                          infop, stat_addr, ru);
1626
1627         return wait_task_continued(p, options, infop, stat_addr, ru);
1628 }
1629
1630 /*
1631  * Do the work of do_wait() for one thread in the group, @tsk.
1632  *
1633  * -ECHILD should be in *@notask_error before the first call.
1634  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1635  * Returns zero if the search for a child should continue; then
1636  * *@notask_error is 0 if there were any eligible children,
1637  * or another error from security_task_wait(), or still -ECHILD.
1638  */
1639 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1640                           enum pid_type type, struct pid *pid, int options,
1641                           struct siginfo __user *infop, int __user *stat_addr,
1642                           struct rusage __user *ru)
1643 {
1644         struct task_struct *p;
1645
1646         list_for_each_entry(p, &tsk->children, sibling) {
1647                 /*
1648                  * Do not consider detached threads.
1649                  */
1650                 if (!task_detached(p)) {
1651                         int ret = wait_consider_task(tsk, 0, p, notask_error,
1652                                                      type, pid, options,
1653                                                      infop, stat_addr, ru);
1654                         if (ret)
1655                                 return ret;
1656                 }
1657         }
1658
1659         return 0;
1660 }
1661
1662 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1663                           enum pid_type type, struct pid *pid, int options,
1664                           struct siginfo __user *infop, int __user *stat_addr,
1665                           struct rusage __user *ru)
1666 {
1667         struct task_struct *p;
1668
1669         /*
1670          * Traditionally we see ptrace'd stopped tasks regardless of options.
1671          */
1672         options |= WUNTRACED;
1673
1674         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1675                 int ret = wait_consider_task(tsk, 1, p, notask_error,
1676                                              type, pid, options,
1677                                              infop, stat_addr, ru);
1678                 if (ret)
1679                         return ret;
1680         }
1681
1682         return 0;
1683 }
1684
1685 static long do_wait(enum pid_type type, struct pid *pid, int options,
1686                     struct siginfo __user *infop, int __user *stat_addr,
1687                     struct rusage __user *ru)
1688 {
1689         DECLARE_WAITQUEUE(wait, current);
1690         struct task_struct *tsk;
1691         int retval;
1692
1693         trace_sched_process_wait(pid);
1694
1695         add_wait_queue(&current->signal->wait_chldexit,&wait);
1696 repeat:
1697         /*
1698          * If there is nothing that can match our critiera just get out.
1699          * We will clear @retval to zero if we see any child that might later
1700          * match our criteria, even if we are not able to reap it yet.
1701          */
1702         retval = -ECHILD;
1703         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1704                 goto end;
1705
1706         current->state = TASK_INTERRUPTIBLE;
1707         read_lock(&tasklist_lock);
1708         tsk = current;
1709         do {
1710                 int tsk_result = do_wait_thread(tsk, &retval,
1711                                                 type, pid, options,
1712                                                 infop, stat_addr, ru);
1713                 if (!tsk_result)
1714                         tsk_result = ptrace_do_wait(tsk, &retval,
1715                                                     type, pid, options,
1716                                                     infop, stat_addr, ru);
1717                 if (tsk_result) {
1718                         /*
1719                          * tasklist_lock is unlocked and we have a final result.
1720                          */
1721                         retval = tsk_result;
1722                         goto end;
1723                 }
1724
1725                 if (options & __WNOTHREAD)
1726                         break;
1727                 tsk = next_thread(tsk);
1728                 BUG_ON(tsk->signal != current->signal);
1729         } while (tsk != current);
1730         read_unlock(&tasklist_lock);
1731
1732         if (!retval && !(options & WNOHANG)) {
1733                 retval = -ERESTARTSYS;
1734                 if (!signal_pending(current)) {
1735                         schedule();
1736                         goto repeat;
1737                 }
1738         }
1739
1740 end:
1741         current->state = TASK_RUNNING;
1742         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1743         if (infop) {
1744                 if (retval > 0)
1745                         retval = 0;
1746                 else {
1747                         /*
1748                          * For a WNOHANG return, clear out all the fields
1749                          * we would set so the user can easily tell the
1750                          * difference.
1751                          */
1752                         if (!retval)
1753                                 retval = put_user(0, &infop->si_signo);
1754                         if (!retval)
1755                                 retval = put_user(0, &infop->si_errno);
1756                         if (!retval)
1757                                 retval = put_user(0, &infop->si_code);
1758                         if (!retval)
1759                                 retval = put_user(0, &infop->si_pid);
1760                         if (!retval)
1761                                 retval = put_user(0, &infop->si_uid);
1762                         if (!retval)
1763                                 retval = put_user(0, &infop->si_status);
1764                 }
1765         }
1766         return retval;
1767 }
1768
1769 asmlinkage long sys_waitid(int which, pid_t upid,
1770                            struct siginfo __user *infop, int options,
1771                            struct rusage __user *ru)
1772 {
1773         struct pid *pid = NULL;
1774         enum pid_type type;
1775         long ret;
1776
1777         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1778                 return -EINVAL;
1779         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1780                 return -EINVAL;
1781
1782         switch (which) {
1783         case P_ALL:
1784                 type = PIDTYPE_MAX;
1785                 break;
1786         case P_PID:
1787                 type = PIDTYPE_PID;
1788                 if (upid <= 0)
1789                         return -EINVAL;
1790                 break;
1791         case P_PGID:
1792                 type = PIDTYPE_PGID;
1793                 if (upid <= 0)
1794                         return -EINVAL;
1795                 break;
1796         default:
1797                 return -EINVAL;
1798         }
1799
1800         if (type < PIDTYPE_MAX)
1801                 pid = find_get_pid(upid);
1802         ret = do_wait(type, pid, options, infop, NULL, ru);
1803         put_pid(pid);
1804
1805         /* avoid REGPARM breakage on x86: */
1806         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1807         return ret;
1808 }
1809
1810 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1811                           int options, struct rusage __user *ru)
1812 {
1813         struct pid *pid = NULL;
1814         enum pid_type type;
1815         long ret;
1816
1817         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1818                         __WNOTHREAD|__WCLONE|__WALL))
1819                 return -EINVAL;
1820
1821         if (upid == -1)
1822                 type = PIDTYPE_MAX;
1823         else if (upid < 0) {
1824                 type = PIDTYPE_PGID;
1825                 pid = find_get_pid(-upid);
1826         } else if (upid == 0) {
1827                 type = PIDTYPE_PGID;
1828                 pid = get_pid(task_pgrp(current));
1829         } else /* upid > 0 */ {
1830                 type = PIDTYPE_PID;
1831                 pid = find_get_pid(upid);
1832         }
1833
1834         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1835         put_pid(pid);
1836
1837         /* avoid REGPARM breakage on x86: */
1838         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1839         return ret;
1840 }
1841
1842 #ifdef __ARCH_WANT_SYS_WAITPID
1843
1844 /*
1845  * sys_waitpid() remains for compatibility. waitpid() should be
1846  * implemented by calling sys_wait4() from libc.a.
1847  */
1848 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1849 {
1850         return sys_wait4(pid, stat_addr, options, NULL);
1851 }
1852
1853 #endif