[PATCH] sanitize unshare_files/reset_files_struct
[safe/jmp/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 /*
561  * Turn us into a lazy TLB process if we
562  * aren't already..
563  */
564 static void exit_mm(struct task_struct * tsk)
565 {
566         struct mm_struct *mm = tsk->mm;
567
568         mm_release(tsk, mm);
569         if (!mm)
570                 return;
571         /*
572          * Serialize with any possible pending coredump.
573          * We must hold mmap_sem around checking core_waiters
574          * and clearing tsk->mm.  The core-inducing thread
575          * will increment core_waiters for each thread in the
576          * group with ->mm != NULL.
577          */
578         down_read(&mm->mmap_sem);
579         if (mm->core_waiters) {
580                 up_read(&mm->mmap_sem);
581                 down_write(&mm->mmap_sem);
582                 if (!--mm->core_waiters)
583                         complete(mm->core_startup_done);
584                 up_write(&mm->mmap_sem);
585
586                 wait_for_completion(&mm->core_done);
587                 down_read(&mm->mmap_sem);
588         }
589         atomic_inc(&mm->mm_count);
590         BUG_ON(mm != tsk->active_mm);
591         /* more a memory barrier than a real lock */
592         task_lock(tsk);
593         tsk->mm = NULL;
594         up_read(&mm->mmap_sem);
595         enter_lazy_tlb(mm, current);
596         /* We don't want this task to be frozen prematurely */
597         clear_freeze_flag(tsk);
598         task_unlock(tsk);
599         mmput(mm);
600 }
601
602 static void
603 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
604 {
605         if (p->pdeath_signal)
606                 /* We already hold the tasklist_lock here.  */
607                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
608
609         /* Move the child from its dying parent to the new one.  */
610         if (unlikely(traced)) {
611                 /* Preserve ptrace links if someone else is tracing this child.  */
612                 list_del_init(&p->ptrace_list);
613                 if (p->parent != p->real_parent)
614                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
615         } else {
616                 /* If this child is being traced, then we're the one tracing it
617                  * anyway, so let go of it.
618                  */
619                 p->ptrace = 0;
620                 remove_parent(p);
621                 p->parent = p->real_parent;
622                 add_parent(p);
623
624                 if (task_is_traced(p)) {
625                         /*
626                          * If it was at a trace stop, turn it into
627                          * a normal stop since it's no longer being
628                          * traced.
629                          */
630                         ptrace_untrace(p);
631                 }
632         }
633
634         /* If this is a threaded reparent there is no need to
635          * notify anyone anything has happened.
636          */
637         if (p->real_parent->group_leader == father->group_leader)
638                 return;
639
640         /* We don't want people slaying init.  */
641         if (p->exit_signal != -1)
642                 p->exit_signal = SIGCHLD;
643
644         /* If we'd notified the old parent about this child's death,
645          * also notify the new parent.
646          */
647         if (!traced && p->exit_state == EXIT_ZOMBIE &&
648             p->exit_signal != -1 && thread_group_empty(p))
649                 do_notify_parent(p, p->exit_signal);
650
651         kill_orphaned_pgrp(p, father);
652 }
653
654 /*
655  * When we die, we re-parent all our children.
656  * Try to give them to another thread in our thread
657  * group, and if no such member exists, give it to
658  * the child reaper process (ie "init") in our pid
659  * space.
660  */
661 static void forget_original_parent(struct task_struct *father)
662 {
663         struct task_struct *p, *n, *reaper = father;
664         struct list_head ptrace_dead;
665
666         INIT_LIST_HEAD(&ptrace_dead);
667
668         write_lock_irq(&tasklist_lock);
669
670         do {
671                 reaper = next_thread(reaper);
672                 if (reaper == father) {
673                         reaper = task_child_reaper(father);
674                         break;
675                 }
676         } while (reaper->flags & PF_EXITING);
677
678         /*
679          * There are only two places where our children can be:
680          *
681          * - in our child list
682          * - in our ptraced child list
683          *
684          * Search them and reparent children.
685          */
686         list_for_each_entry_safe(p, n, &father->children, sibling) {
687                 int ptrace;
688
689                 ptrace = p->ptrace;
690
691                 /* if father isn't the real parent, then ptrace must be enabled */
692                 BUG_ON(father != p->real_parent && !ptrace);
693
694                 if (father == p->real_parent) {
695                         /* reparent with a reaper, real father it's us */
696                         p->real_parent = reaper;
697                         reparent_thread(p, father, 0);
698                 } else {
699                         /* reparent ptraced task to its real parent */
700                         __ptrace_unlink (p);
701                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
702                             thread_group_empty(p))
703                                 do_notify_parent(p, p->exit_signal);
704                 }
705
706                 /*
707                  * if the ptraced child is a zombie with exit_signal == -1
708                  * we must collect it before we exit, or it will remain
709                  * zombie forever since we prevented it from self-reap itself
710                  * while it was being traced by us, to be able to see it in wait4.
711                  */
712                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
713                         list_add(&p->ptrace_list, &ptrace_dead);
714         }
715
716         list_for_each_entry_safe(p, n, &father->ptrace_children, ptrace_list) {
717                 p->real_parent = reaper;
718                 reparent_thread(p, father, 1);
719         }
720
721         write_unlock_irq(&tasklist_lock);
722         BUG_ON(!list_empty(&father->children));
723         BUG_ON(!list_empty(&father->ptrace_children));
724
725         list_for_each_entry_safe(p, n, &ptrace_dead, ptrace_list) {
726                 list_del_init(&p->ptrace_list);
727                 release_task(p);
728         }
729
730 }
731
732 /*
733  * Send signals to all our closest relatives so that they know
734  * to properly mourn us..
735  */
736 static void exit_notify(struct task_struct *tsk, int group_dead)
737 {
738         int state;
739
740         /*
741          * This does two things:
742          *
743          * A.  Make init inherit all the child processes
744          * B.  Check to see if any process groups have become orphaned
745          *      as a result of our exiting, and if they have any stopped
746          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
747          */
748         forget_original_parent(tsk);
749         exit_task_namespaces(tsk);
750
751         write_lock_irq(&tasklist_lock);
752         if (group_dead)
753                 kill_orphaned_pgrp(tsk->group_leader, NULL);
754
755         /* Let father know we died
756          *
757          * Thread signals are configurable, but you aren't going to use
758          * that to send signals to arbitary processes.
759          * That stops right now.
760          *
761          * If the parent exec id doesn't match the exec id we saved
762          * when we started then we know the parent has changed security
763          * domain.
764          *
765          * If our self_exec id doesn't match our parent_exec_id then
766          * we have changed execution domain as these two values started
767          * the same after a fork.
768          */
769         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
770             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
771              tsk->self_exec_id != tsk->parent_exec_id)
772             && !capable(CAP_KILL))
773                 tsk->exit_signal = SIGCHLD;
774
775
776         /* If something other than our normal parent is ptracing us, then
777          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
778          * only has special meaning to our real parent.
779          */
780         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
781                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
782                 do_notify_parent(tsk, signal);
783         } else if (tsk->ptrace) {
784                 do_notify_parent(tsk, SIGCHLD);
785         }
786
787         state = EXIT_ZOMBIE;
788         if (tsk->exit_signal == -1 && likely(!tsk->ptrace))
789                 state = EXIT_DEAD;
790         tsk->exit_state = state;
791
792         if (thread_group_leader(tsk) &&
793             tsk->signal->notify_count < 0 &&
794             tsk->signal->group_exit_task)
795                 wake_up_process(tsk->signal->group_exit_task);
796
797         write_unlock_irq(&tasklist_lock);
798
799         /* If the process is dead, release it - nobody will wait for it */
800         if (state == EXIT_DEAD)
801                 release_task(tsk);
802 }
803
804 #ifdef CONFIG_DEBUG_STACK_USAGE
805 static void check_stack_usage(void)
806 {
807         static DEFINE_SPINLOCK(low_water_lock);
808         static int lowest_to_date = THREAD_SIZE;
809         unsigned long *n = end_of_stack(current);
810         unsigned long free;
811
812         while (*n == 0)
813                 n++;
814         free = (unsigned long)n - (unsigned long)end_of_stack(current);
815
816         if (free >= lowest_to_date)
817                 return;
818
819         spin_lock(&low_water_lock);
820         if (free < lowest_to_date) {
821                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
822                                 "left\n",
823                                 current->comm, free);
824                 lowest_to_date = free;
825         }
826         spin_unlock(&low_water_lock);
827 }
828 #else
829 static inline void check_stack_usage(void) {}
830 #endif
831
832 static inline void exit_child_reaper(struct task_struct *tsk)
833 {
834         if (likely(tsk->group_leader != task_child_reaper(tsk)))
835                 return;
836
837         if (tsk->nsproxy->pid_ns == &init_pid_ns)
838                 panic("Attempted to kill init!");
839
840         /*
841          * @tsk is the last thread in the 'cgroup-init' and is exiting.
842          * Terminate all remaining processes in the namespace and reap them
843          * before exiting @tsk.
844          *
845          * Note that @tsk (last thread of cgroup-init) may not necessarily
846          * be the child-reaper (i.e main thread of cgroup-init) of the
847          * namespace i.e the child_reaper may have already exited.
848          *
849          * Even after a child_reaper exits, we let it inherit orphaned children,
850          * because, pid_ns->child_reaper remains valid as long as there is
851          * at least one living sub-thread in the cgroup init.
852
853          * This living sub-thread of the cgroup-init will be notified when
854          * a child inherited by the 'child-reaper' exits (do_notify_parent()
855          * uses __group_send_sig_info()). Further, when reaping child processes,
856          * do_wait() iterates over children of all living sub threads.
857
858          * i.e even though 'child_reaper' thread is listed as the parent of the
859          * orphaned children, any living sub-thread in the cgroup-init can
860          * perform the role of the child_reaper.
861          */
862         zap_pid_ns_processes(tsk->nsproxy->pid_ns);
863 }
864
865 NORET_TYPE void do_exit(long code)
866 {
867         struct task_struct *tsk = current;
868         int group_dead;
869
870         profile_task_exit(tsk);
871
872         WARN_ON(atomic_read(&tsk->fs_excl));
873
874         if (unlikely(in_interrupt()))
875                 panic("Aiee, killing interrupt handler!");
876         if (unlikely(!tsk->pid))
877                 panic("Attempted to kill the idle task!");
878
879         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
880                 current->ptrace_message = code;
881                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
882         }
883
884         /*
885          * We're taking recursive faults here in do_exit. Safest is to just
886          * leave this task alone and wait for reboot.
887          */
888         if (unlikely(tsk->flags & PF_EXITING)) {
889                 printk(KERN_ALERT
890                         "Fixing recursive fault but reboot is needed!\n");
891                 /*
892                  * We can do this unlocked here. The futex code uses
893                  * this flag just to verify whether the pi state
894                  * cleanup has been done or not. In the worst case it
895                  * loops once more. We pretend that the cleanup was
896                  * done as there is no way to return. Either the
897                  * OWNER_DIED bit is set by now or we push the blocked
898                  * task into the wait for ever nirwana as well.
899                  */
900                 tsk->flags |= PF_EXITPIDONE;
901                 if (tsk->io_context)
902                         exit_io_context();
903                 set_current_state(TASK_UNINTERRUPTIBLE);
904                 schedule();
905         }
906
907         exit_signals(tsk);  /* sets PF_EXITING */
908         /*
909          * tsk->flags are checked in the futex code to protect against
910          * an exiting task cleaning up the robust pi futexes.
911          */
912         smp_mb();
913         spin_unlock_wait(&tsk->pi_lock);
914
915         if (unlikely(in_atomic()))
916                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
917                                 current->comm, task_pid_nr(current),
918                                 preempt_count());
919
920         acct_update_integrals(tsk);
921         if (tsk->mm) {
922                 update_hiwater_rss(tsk->mm);
923                 update_hiwater_vm(tsk->mm);
924         }
925         group_dead = atomic_dec_and_test(&tsk->signal->live);
926         if (group_dead) {
927                 exit_child_reaper(tsk);
928                 hrtimer_cancel(&tsk->signal->real_timer);
929                 exit_itimers(tsk->signal);
930         }
931         acct_collect(code, group_dead);
932 #ifdef CONFIG_FUTEX
933         if (unlikely(tsk->robust_list))
934                 exit_robust_list(tsk);
935 #ifdef CONFIG_COMPAT
936         if (unlikely(tsk->compat_robust_list))
937                 compat_exit_robust_list(tsk);
938 #endif
939 #endif
940         if (group_dead)
941                 tty_audit_exit();
942         if (unlikely(tsk->audit_context))
943                 audit_free(tsk);
944
945         tsk->exit_code = code;
946         taskstats_exit(tsk, group_dead);
947
948         exit_mm(tsk);
949
950         if (group_dead)
951                 acct_process();
952         exit_sem(tsk);
953         exit_files(tsk);
954         exit_fs(tsk);
955         check_stack_usage();
956         exit_thread();
957         cgroup_exit(tsk, 1);
958         exit_keys(tsk);
959
960         if (group_dead && tsk->signal->leader)
961                 disassociate_ctty(1);
962
963         module_put(task_thread_info(tsk)->exec_domain->module);
964         if (tsk->binfmt)
965                 module_put(tsk->binfmt->module);
966
967         proc_exit_connector(tsk);
968         exit_notify(tsk, group_dead);
969 #ifdef CONFIG_NUMA
970         mpol_free(tsk->mempolicy);
971         tsk->mempolicy = NULL;
972 #endif
973 #ifdef CONFIG_FUTEX
974         /*
975          * This must happen late, after the PID is not
976          * hashed anymore:
977          */
978         if (unlikely(!list_empty(&tsk->pi_state_list)))
979                 exit_pi_state_list(tsk);
980         if (unlikely(current->pi_state_cache))
981                 kfree(current->pi_state_cache);
982 #endif
983         /*
984          * Make sure we are holding no locks:
985          */
986         debug_check_no_locks_held(tsk);
987         /*
988          * We can do this unlocked here. The futex code uses this flag
989          * just to verify whether the pi state cleanup has been done
990          * or not. In the worst case it loops once more.
991          */
992         tsk->flags |= PF_EXITPIDONE;
993
994         if (tsk->io_context)
995                 exit_io_context();
996
997         if (tsk->splice_pipe)
998                 __free_pipe_info(tsk->splice_pipe);
999
1000         preempt_disable();
1001         /* causes final put_task_struct in finish_task_switch(). */
1002         tsk->state = TASK_DEAD;
1003
1004         schedule();
1005         BUG();
1006         /* Avoid "noreturn function does return".  */
1007         for (;;)
1008                 cpu_relax();    /* For when BUG is null */
1009 }
1010
1011 EXPORT_SYMBOL_GPL(do_exit);
1012
1013 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1014 {
1015         if (comp)
1016                 complete(comp);
1017
1018         do_exit(code);
1019 }
1020
1021 EXPORT_SYMBOL(complete_and_exit);
1022
1023 asmlinkage long sys_exit(int error_code)
1024 {
1025         do_exit((error_code&0xff)<<8);
1026 }
1027
1028 /*
1029  * Take down every thread in the group.  This is called by fatal signals
1030  * as well as by sys_exit_group (below).
1031  */
1032 NORET_TYPE void
1033 do_group_exit(int exit_code)
1034 {
1035         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1036
1037         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1038                 exit_code = current->signal->group_exit_code;
1039         else if (!thread_group_empty(current)) {
1040                 struct signal_struct *const sig = current->signal;
1041                 struct sighand_struct *const sighand = current->sighand;
1042                 spin_lock_irq(&sighand->siglock);
1043                 if (signal_group_exit(sig))
1044                         /* Another thread got here before we took the lock.  */
1045                         exit_code = sig->group_exit_code;
1046                 else {
1047                         sig->group_exit_code = exit_code;
1048                         sig->flags = SIGNAL_GROUP_EXIT;
1049                         zap_other_threads(current);
1050                 }
1051                 spin_unlock_irq(&sighand->siglock);
1052         }
1053
1054         do_exit(exit_code);
1055         /* NOTREACHED */
1056 }
1057
1058 /*
1059  * this kills every thread in the thread group. Note that any externally
1060  * wait4()-ing process will get the correct exit code - even if this
1061  * thread is not the thread group leader.
1062  */
1063 asmlinkage void sys_exit_group(int error_code)
1064 {
1065         do_group_exit((error_code & 0xff) << 8);
1066 }
1067
1068 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1069 {
1070         struct pid *pid = NULL;
1071         if (type == PIDTYPE_PID)
1072                 pid = task->pids[type].pid;
1073         else if (type < PIDTYPE_MAX)
1074                 pid = task->group_leader->pids[type].pid;
1075         return pid;
1076 }
1077
1078 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1079                           struct task_struct *p)
1080 {
1081         int err;
1082
1083         if (type < PIDTYPE_MAX) {
1084                 if (task_pid_type(p, type) != pid)
1085                         return 0;
1086         }
1087
1088         /*
1089          * Do not consider detached threads that are
1090          * not ptraced:
1091          */
1092         if (p->exit_signal == -1 && !p->ptrace)
1093                 return 0;
1094
1095         /* Wait for all children (clone and not) if __WALL is set;
1096          * otherwise, wait for clone children *only* if __WCLONE is
1097          * set; otherwise, wait for non-clone children *only*.  (Note:
1098          * A "clone" child here is one that reports to its parent
1099          * using a signal other than SIGCHLD.) */
1100         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1101             && !(options & __WALL))
1102                 return 0;
1103
1104         err = security_task_wait(p);
1105         if (likely(!err))
1106                 return 1;
1107
1108         if (type != PIDTYPE_PID)
1109                 return 0;
1110         /* This child was explicitly requested, abort */
1111         read_unlock(&tasklist_lock);
1112         return err;
1113 }
1114
1115 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1116                                int why, int status,
1117                                struct siginfo __user *infop,
1118                                struct rusage __user *rusagep)
1119 {
1120         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1121
1122         put_task_struct(p);
1123         if (!retval)
1124                 retval = put_user(SIGCHLD, &infop->si_signo);
1125         if (!retval)
1126                 retval = put_user(0, &infop->si_errno);
1127         if (!retval)
1128                 retval = put_user((short)why, &infop->si_code);
1129         if (!retval)
1130                 retval = put_user(pid, &infop->si_pid);
1131         if (!retval)
1132                 retval = put_user(uid, &infop->si_uid);
1133         if (!retval)
1134                 retval = put_user(status, &infop->si_status);
1135         if (!retval)
1136                 retval = pid;
1137         return retval;
1138 }
1139
1140 /*
1141  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1142  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1143  * the lock and this task is uninteresting.  If we return nonzero, we have
1144  * released the lock and the system call should return.
1145  */
1146 static int wait_task_zombie(struct task_struct *p, int noreap,
1147                             struct siginfo __user *infop,
1148                             int __user *stat_addr, struct rusage __user *ru)
1149 {
1150         unsigned long state;
1151         int retval, status, traced;
1152         pid_t pid = task_pid_vnr(p);
1153
1154         if (unlikely(noreap)) {
1155                 uid_t uid = p->uid;
1156                 int exit_code = p->exit_code;
1157                 int why, status;
1158
1159                 get_task_struct(p);
1160                 read_unlock(&tasklist_lock);
1161                 if ((exit_code & 0x7f) == 0) {
1162                         why = CLD_EXITED;
1163                         status = exit_code >> 8;
1164                 } else {
1165                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1166                         status = exit_code & 0x7f;
1167                 }
1168                 return wait_noreap_copyout(p, pid, uid, why,
1169                                            status, infop, ru);
1170         }
1171
1172         /*
1173          * Try to move the task's state to DEAD
1174          * only one thread is allowed to do this:
1175          */
1176         state = xchg(&p->exit_state, EXIT_DEAD);
1177         if (state != EXIT_ZOMBIE) {
1178                 BUG_ON(state != EXIT_DEAD);
1179                 return 0;
1180         }
1181
1182         /* traced means p->ptrace, but not vice versa */
1183         traced = (p->real_parent != p->parent);
1184
1185         if (likely(!traced)) {
1186                 struct signal_struct *psig;
1187                 struct signal_struct *sig;
1188
1189                 /*
1190                  * The resource counters for the group leader are in its
1191                  * own task_struct.  Those for dead threads in the group
1192                  * are in its signal_struct, as are those for the child
1193                  * processes it has previously reaped.  All these
1194                  * accumulate in the parent's signal_struct c* fields.
1195                  *
1196                  * We don't bother to take a lock here to protect these
1197                  * p->signal fields, because they are only touched by
1198                  * __exit_signal, which runs with tasklist_lock
1199                  * write-locked anyway, and so is excluded here.  We do
1200                  * need to protect the access to p->parent->signal fields,
1201                  * as other threads in the parent group can be right
1202                  * here reaping other children at the same time.
1203                  */
1204                 spin_lock_irq(&p->parent->sighand->siglock);
1205                 psig = p->parent->signal;
1206                 sig = p->signal;
1207                 psig->cutime =
1208                         cputime_add(psig->cutime,
1209                         cputime_add(p->utime,
1210                         cputime_add(sig->utime,
1211                                     sig->cutime)));
1212                 psig->cstime =
1213                         cputime_add(psig->cstime,
1214                         cputime_add(p->stime,
1215                         cputime_add(sig->stime,
1216                                     sig->cstime)));
1217                 psig->cgtime =
1218                         cputime_add(psig->cgtime,
1219                         cputime_add(p->gtime,
1220                         cputime_add(sig->gtime,
1221                                     sig->cgtime)));
1222                 psig->cmin_flt +=
1223                         p->min_flt + sig->min_flt + sig->cmin_flt;
1224                 psig->cmaj_flt +=
1225                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1226                 psig->cnvcsw +=
1227                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1228                 psig->cnivcsw +=
1229                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1230                 psig->cinblock +=
1231                         task_io_get_inblock(p) +
1232                         sig->inblock + sig->cinblock;
1233                 psig->coublock +=
1234                         task_io_get_oublock(p) +
1235                         sig->oublock + sig->coublock;
1236                 spin_unlock_irq(&p->parent->sighand->siglock);
1237         }
1238
1239         /*
1240          * Now we are sure this task is interesting, and no other
1241          * thread can reap it because we set its state to EXIT_DEAD.
1242          */
1243         read_unlock(&tasklist_lock);
1244
1245         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1246         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1247                 ? p->signal->group_exit_code : p->exit_code;
1248         if (!retval && stat_addr)
1249                 retval = put_user(status, stat_addr);
1250         if (!retval && infop)
1251                 retval = put_user(SIGCHLD, &infop->si_signo);
1252         if (!retval && infop)
1253                 retval = put_user(0, &infop->si_errno);
1254         if (!retval && infop) {
1255                 int why;
1256
1257                 if ((status & 0x7f) == 0) {
1258                         why = CLD_EXITED;
1259                         status >>= 8;
1260                 } else {
1261                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1262                         status &= 0x7f;
1263                 }
1264                 retval = put_user((short)why, &infop->si_code);
1265                 if (!retval)
1266                         retval = put_user(status, &infop->si_status);
1267         }
1268         if (!retval && infop)
1269                 retval = put_user(pid, &infop->si_pid);
1270         if (!retval && infop)
1271                 retval = put_user(p->uid, &infop->si_uid);
1272         if (!retval)
1273                 retval = pid;
1274
1275         if (traced) {
1276                 write_lock_irq(&tasklist_lock);
1277                 /* We dropped tasklist, ptracer could die and untrace */
1278                 ptrace_unlink(p);
1279                 /*
1280                  * If this is not a detached task, notify the parent.
1281                  * If it's still not detached after that, don't release
1282                  * it now.
1283                  */
1284                 if (p->exit_signal != -1) {
1285                         do_notify_parent(p, p->exit_signal);
1286                         if (p->exit_signal != -1) {
1287                                 p->exit_state = EXIT_ZOMBIE;
1288                                 p = NULL;
1289                         }
1290                 }
1291                 write_unlock_irq(&tasklist_lock);
1292         }
1293         if (p != NULL)
1294                 release_task(p);
1295
1296         return retval;
1297 }
1298
1299 /*
1300  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1301  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1302  * the lock and this task is uninteresting.  If we return nonzero, we have
1303  * released the lock and the system call should return.
1304  */
1305 static int wait_task_stopped(struct task_struct *p,
1306                              int noreap, struct siginfo __user *infop,
1307                              int __user *stat_addr, struct rusage __user *ru)
1308 {
1309         int retval, exit_code, why;
1310         uid_t uid = 0; /* unneeded, required by compiler */
1311         pid_t pid;
1312
1313         exit_code = 0;
1314         spin_lock_irq(&p->sighand->siglock);
1315
1316         if (unlikely(!task_is_stopped_or_traced(p)))
1317                 goto unlock_sig;
1318
1319         if (!(p->ptrace & PT_PTRACED) && p->signal->group_stop_count > 0)
1320                 /*
1321                  * A group stop is in progress and this is the group leader.
1322                  * We won't report until all threads have stopped.
1323                  */
1324                 goto unlock_sig;
1325
1326         exit_code = p->exit_code;
1327         if (!exit_code)
1328                 goto unlock_sig;
1329
1330         if (!noreap)
1331                 p->exit_code = 0;
1332
1333         uid = p->uid;
1334 unlock_sig:
1335         spin_unlock_irq(&p->sighand->siglock);
1336         if (!exit_code)
1337                 return 0;
1338
1339         /*
1340          * Now we are pretty sure this task is interesting.
1341          * Make sure it doesn't get reaped out from under us while we
1342          * give up the lock and then examine it below.  We don't want to
1343          * keep holding onto the tasklist_lock while we call getrusage and
1344          * possibly take page faults for user memory.
1345          */
1346         get_task_struct(p);
1347         pid = task_pid_vnr(p);
1348         why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1349         read_unlock(&tasklist_lock);
1350
1351         if (unlikely(noreap))
1352                 return wait_noreap_copyout(p, pid, uid,
1353                                            why, exit_code,
1354                                            infop, ru);
1355
1356         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1357         if (!retval && stat_addr)
1358                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1359         if (!retval && infop)
1360                 retval = put_user(SIGCHLD, &infop->si_signo);
1361         if (!retval && infop)
1362                 retval = put_user(0, &infop->si_errno);
1363         if (!retval && infop)
1364                 retval = put_user((short)why, &infop->si_code);
1365         if (!retval && infop)
1366                 retval = put_user(exit_code, &infop->si_status);
1367         if (!retval && infop)
1368                 retval = put_user(pid, &infop->si_pid);
1369         if (!retval && infop)
1370                 retval = put_user(uid, &infop->si_uid);
1371         if (!retval)
1372                 retval = pid;
1373         put_task_struct(p);
1374
1375         BUG_ON(!retval);
1376         return retval;
1377 }
1378
1379 /*
1380  * Handle do_wait work for one task in a live, non-stopped state.
1381  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1382  * the lock and this task is uninteresting.  If we return nonzero, we have
1383  * released the lock and the system call should return.
1384  */
1385 static int wait_task_continued(struct task_struct *p, int noreap,
1386                                struct siginfo __user *infop,
1387                                int __user *stat_addr, struct rusage __user *ru)
1388 {
1389         int retval;
1390         pid_t pid;
1391         uid_t uid;
1392
1393         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1394                 return 0;
1395
1396         spin_lock_irq(&p->sighand->siglock);
1397         /* Re-check with the lock held.  */
1398         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1399                 spin_unlock_irq(&p->sighand->siglock);
1400                 return 0;
1401         }
1402         if (!noreap)
1403                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1404         spin_unlock_irq(&p->sighand->siglock);
1405
1406         pid = task_pid_vnr(p);
1407         uid = p->uid;
1408         get_task_struct(p);
1409         read_unlock(&tasklist_lock);
1410
1411         if (!infop) {
1412                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1413                 put_task_struct(p);
1414                 if (!retval && stat_addr)
1415                         retval = put_user(0xffff, stat_addr);
1416                 if (!retval)
1417                         retval = pid;
1418         } else {
1419                 retval = wait_noreap_copyout(p, pid, uid,
1420                                              CLD_CONTINUED, SIGCONT,
1421                                              infop, ru);
1422                 BUG_ON(retval == 0);
1423         }
1424
1425         return retval;
1426 }
1427
1428 static long do_wait(enum pid_type type, struct pid *pid, int options,
1429                     struct siginfo __user *infop, int __user *stat_addr,
1430                     struct rusage __user *ru)
1431 {
1432         DECLARE_WAITQUEUE(wait, current);
1433         struct task_struct *tsk;
1434         int flag, retval;
1435
1436         add_wait_queue(&current->signal->wait_chldexit,&wait);
1437 repeat:
1438         /* If there is nothing that can match our critier just get out */
1439         retval = -ECHILD;
1440         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1441                 goto end;
1442
1443         /*
1444          * We will set this flag if we see any child that might later
1445          * match our criteria, even if we are not able to reap it yet.
1446          */
1447         flag = retval = 0;
1448         current->state = TASK_INTERRUPTIBLE;
1449         read_lock(&tasklist_lock);
1450         tsk = current;
1451         do {
1452                 struct task_struct *p;
1453
1454                 list_for_each_entry(p, &tsk->children, sibling) {
1455                         int ret = eligible_child(type, pid, options, p);
1456                         if (!ret)
1457                                 continue;
1458
1459                         if (unlikely(ret < 0)) {
1460                                 retval = ret;
1461                         } else if (task_is_stopped_or_traced(p)) {
1462                                 /*
1463                                  * It's stopped now, so it might later
1464                                  * continue, exit, or stop again.
1465                                  */
1466                                 flag = 1;
1467                                 if (!(p->ptrace & PT_PTRACED) &&
1468                                     !(options & WUNTRACED))
1469                                         continue;
1470
1471                                 retval = wait_task_stopped(p,
1472                                                 (options & WNOWAIT), infop,
1473                                                 stat_addr, ru);
1474                         } else if (p->exit_state == EXIT_ZOMBIE &&
1475                                         !delay_group_leader(p)) {
1476                                 /*
1477                                  * We don't reap group leaders with subthreads.
1478                                  */
1479                                 if (!likely(options & WEXITED))
1480                                         continue;
1481                                 retval = wait_task_zombie(p,
1482                                                 (options & WNOWAIT), infop,
1483                                                 stat_addr, ru);
1484                         } else if (p->exit_state != EXIT_DEAD) {
1485                                 /*
1486                                  * It's running now, so it might later
1487                                  * exit, stop, or stop and then continue.
1488                                  */
1489                                 flag = 1;
1490                                 if (!unlikely(options & WCONTINUED))
1491                                         continue;
1492                                 retval = wait_task_continued(p,
1493                                                 (options & WNOWAIT), infop,
1494                                                 stat_addr, ru);
1495                         }
1496                         if (retval != 0) /* tasklist_lock released */
1497                                 goto end;
1498                 }
1499                 if (!flag) {
1500                         list_for_each_entry(p, &tsk->ptrace_children,
1501                                                                 ptrace_list) {
1502                                 flag = eligible_child(type, pid, options, p);
1503                                 if (!flag)
1504                                         continue;
1505                                 if (likely(flag > 0))
1506                                         break;
1507                                 retval = flag;
1508                                 goto end;
1509                         }
1510                 }
1511                 if (options & __WNOTHREAD)
1512                         break;
1513                 tsk = next_thread(tsk);
1514                 BUG_ON(tsk->signal != current->signal);
1515         } while (tsk != current);
1516         read_unlock(&tasklist_lock);
1517
1518         if (flag) {
1519                 if (options & WNOHANG)
1520                         goto end;
1521                 retval = -ERESTARTSYS;
1522                 if (signal_pending(current))
1523                         goto end;
1524                 schedule();
1525                 goto repeat;
1526         }
1527         retval = -ECHILD;
1528 end:
1529         current->state = TASK_RUNNING;
1530         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1531         if (infop) {
1532                 if (retval > 0)
1533                         retval = 0;
1534                 else {
1535                         /*
1536                          * For a WNOHANG return, clear out all the fields
1537                          * we would set so the user can easily tell the
1538                          * difference.
1539                          */
1540                         if (!retval)
1541                                 retval = put_user(0, &infop->si_signo);
1542                         if (!retval)
1543                                 retval = put_user(0, &infop->si_errno);
1544                         if (!retval)
1545                                 retval = put_user(0, &infop->si_code);
1546                         if (!retval)
1547                                 retval = put_user(0, &infop->si_pid);
1548                         if (!retval)
1549                                 retval = put_user(0, &infop->si_uid);
1550                         if (!retval)
1551                                 retval = put_user(0, &infop->si_status);
1552                 }
1553         }
1554         return retval;
1555 }
1556
1557 asmlinkage long sys_waitid(int which, pid_t upid,
1558                            struct siginfo __user *infop, int options,
1559                            struct rusage __user *ru)
1560 {
1561         struct pid *pid = NULL;
1562         enum pid_type type;
1563         long ret;
1564
1565         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1566                 return -EINVAL;
1567         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1568                 return -EINVAL;
1569
1570         switch (which) {
1571         case P_ALL:
1572                 type = PIDTYPE_MAX;
1573                 break;
1574         case P_PID:
1575                 type = PIDTYPE_PID;
1576                 if (upid <= 0)
1577                         return -EINVAL;
1578                 break;
1579         case P_PGID:
1580                 type = PIDTYPE_PGID;
1581                 if (upid <= 0)
1582                         return -EINVAL;
1583                 break;
1584         default:
1585                 return -EINVAL;
1586         }
1587
1588         if (type < PIDTYPE_MAX)
1589                 pid = find_get_pid(upid);
1590         ret = do_wait(type, pid, options, infop, NULL, ru);
1591         put_pid(pid);
1592
1593         /* avoid REGPARM breakage on x86: */
1594         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1595         return ret;
1596 }
1597
1598 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1599                           int options, struct rusage __user *ru)
1600 {
1601         struct pid *pid = NULL;
1602         enum pid_type type;
1603         long ret;
1604
1605         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1606                         __WNOTHREAD|__WCLONE|__WALL))
1607                 return -EINVAL;
1608
1609         if (upid == -1)
1610                 type = PIDTYPE_MAX;
1611         else if (upid < 0) {
1612                 type = PIDTYPE_PGID;
1613                 pid = find_get_pid(-upid);
1614         } else if (upid == 0) {
1615                 type = PIDTYPE_PGID;
1616                 pid = get_pid(task_pgrp(current));
1617         } else /* upid > 0 */ {
1618                 type = PIDTYPE_PID;
1619                 pid = find_get_pid(upid);
1620         }
1621
1622         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1623         put_pid(pid);
1624
1625         /* avoid REGPARM breakage on x86: */
1626         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1627         return ret;
1628 }
1629
1630 #ifdef __ARCH_WANT_SYS_WAITPID
1631
1632 /*
1633  * sys_waitpid() remains for compatibility. waitpid() should be
1634  * implemented by calling sys_wait4() from libc.a.
1635  */
1636 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1637 {
1638         return sys_wait4(pid, stat_addr, options, NULL);
1639 }
1640
1641 #endif