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