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