reparent_thread: fix a zombie leak if /sbin/init ignores SIGCHLD
[linux-2.6.git] / kernel / exit.c
1 /*
2  *  linux/kernel/exit.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 #include <linux/mm.h>
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/mnt_namespace.h>
16 #include <linux/iocontext.h>
17 #include <linux/key.h>
18 #include <linux/security.h>
19 #include <linux/cpu.h>
20 #include <linux/acct.h>
21 #include <linux/tsacct_kern.h>
22 #include <linux/file.h>
23 #include <linux/fdtable.h>
24 #include <linux/binfmts.h>
25 #include <linux/nsproxy.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/ptrace.h>
28 #include <linux/profile.h>
29 #include <linux/mount.h>
30 #include <linux/proc_fs.h>
31 #include <linux/kthread.h>
32 #include <linux/mempolicy.h>
33 #include <linux/taskstats_kern.h>
34 #include <linux/delayacct.h>
35 #include <linux/freezer.h>
36 #include <linux/cgroup.h>
37 #include <linux/syscalls.h>
38 #include <linux/signal.h>
39 #include <linux/posix-timers.h>
40 #include <linux/cn_proc.h>
41 #include <linux/mutex.h>
42 #include <linux/futex.h>
43 #include <linux/pipe_fs_i.h>
44 #include <linux/audit.h> /* for audit_free() */
45 #include <linux/resource.h>
46 #include <linux/blkdev.h>
47 #include <linux/task_io_accounting_ops.h>
48 #include <linux/tracehook.h>
49 #include <linux/init_task.h>
50 #include <trace/sched.h>
51
52 #include <asm/uaccess.h>
53 #include <asm/unistd.h>
54 #include <asm/pgtable.h>
55 #include <asm/mmu_context.h>
56 #include "cred-internals.h"
57
58 DEFINE_TRACE(sched_process_free);
59 DEFINE_TRACE(sched_process_exit);
60 DEFINE_TRACE(sched_process_wait);
61
62 static void exit_mm(struct task_struct * tsk);
63
64 static inline int task_detached(struct task_struct *p)
65 {
66         return p->exit_signal == -1;
67 }
68
69 static void __unhash_process(struct task_struct *p)
70 {
71         nr_threads--;
72         detach_pid(p, PIDTYPE_PID);
73         if (thread_group_leader(p)) {
74                 detach_pid(p, PIDTYPE_PGID);
75                 detach_pid(p, PIDTYPE_SID);
76
77                 list_del_rcu(&p->tasks);
78                 __get_cpu_var(process_counts)--;
79         }
80         list_del_rcu(&p->thread_group);
81         list_del_init(&p->sibling);
82 }
83
84 /*
85  * This function expects the tasklist_lock write-locked.
86  */
87 static void __exit_signal(struct task_struct *tsk)
88 {
89         struct signal_struct *sig = tsk->signal;
90         struct sighand_struct *sighand;
91
92         BUG_ON(!sig);
93         BUG_ON(!atomic_read(&sig->count));
94
95         sighand = rcu_dereference(tsk->sighand);
96         spin_lock(&sighand->siglock);
97
98         posix_cpu_timers_exit(tsk);
99         if (atomic_dec_and_test(&sig->count))
100                 posix_cpu_timers_exit_group(tsk);
101         else {
102                 /*
103                  * If there is any task waiting for the group exit
104                  * then notify it:
105                  */
106                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
107                         wake_up_process(sig->group_exit_task);
108
109                 if (tsk == sig->curr_target)
110                         sig->curr_target = next_thread(tsk);
111                 /*
112                  * Accumulate here the counters for all threads but the
113                  * group leader as they die, so they can be added into
114                  * the process-wide totals when those are taken.
115                  * The group leader stays around as a zombie as long
116                  * as there are other threads.  When it gets reaped,
117                  * the exit.c code will add its counts into these totals.
118                  * We won't ever get here for the group leader, since it
119                  * will have been the last reference on the signal_struct.
120                  */
121                 sig->utime = cputime_add(sig->utime, task_utime(tsk));
122                 sig->stime = cputime_add(sig->stime, task_stime(tsk));
123                 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
124                 sig->min_flt += tsk->min_flt;
125                 sig->maj_flt += tsk->maj_flt;
126                 sig->nvcsw += tsk->nvcsw;
127                 sig->nivcsw += tsk->nivcsw;
128                 sig->inblock += task_io_get_inblock(tsk);
129                 sig->oublock += task_io_get_oublock(tsk);
130                 task_io_accounting_add(&sig->ioac, &tsk->ioac);
131                 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
132                 sig = NULL; /* Marker for below. */
133         }
134
135         __unhash_process(tsk);
136
137         /*
138          * Do this under ->siglock, we can race with another thread
139          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
140          */
141         flush_sigqueue(&tsk->pending);
142
143         tsk->signal = NULL;
144         tsk->sighand = NULL;
145         spin_unlock(&sighand->siglock);
146
147         __cleanup_sighand(sighand);
148         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
149         if (sig) {
150                 flush_sigqueue(&sig->shared_pending);
151                 taskstats_tgid_free(sig);
152                 /*
153                  * Make sure ->signal can't go away under rq->lock,
154                  * see account_group_exec_runtime().
155                  */
156                 task_rq_unlock_wait(tsk);
157                 __cleanup_signal(sig);
158         }
159 }
160
161 static void delayed_put_task_struct(struct rcu_head *rhp)
162 {
163         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
164
165         trace_sched_process_free(tsk);
166         put_task_struct(tsk);
167 }
168
169
170 void release_task(struct task_struct * p)
171 {
172         struct task_struct *leader;
173         int zap_leader;
174 repeat:
175         tracehook_prepare_release_task(p);
176         /* don't need to get the RCU readlock here - the process is dead and
177          * can't be modifying its own credentials */
178         atomic_dec(&__task_cred(p)->user->processes);
179
180         proc_flush_task(p);
181         write_lock_irq(&tasklist_lock);
182         tracehook_finish_release_task(p);
183         __exit_signal(p);
184
185         /*
186          * If we are the last non-leader member of the thread
187          * group, and the leader is zombie, then notify the
188          * group leader's parent process. (if it wants notification.)
189          */
190         zap_leader = 0;
191         leader = p->group_leader;
192         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
193                 BUG_ON(task_detached(leader));
194                 do_notify_parent(leader, leader->exit_signal);
195                 /*
196                  * If we were the last child thread and the leader has
197                  * exited already, and the leader's parent ignores SIGCHLD,
198                  * then we are the one who should release the leader.
199                  *
200                  * do_notify_parent() will have marked it self-reaping in
201                  * that case.
202                  */
203                 zap_leader = task_detached(leader);
204
205                 /*
206                  * This maintains the invariant that release_task()
207                  * only runs on a task in EXIT_DEAD, just for sanity.
208                  */
209                 if (zap_leader)
210                         leader->exit_state = EXIT_DEAD;
211         }
212
213         write_unlock_irq(&tasklist_lock);
214         release_thread(p);
215         call_rcu(&p->rcu, delayed_put_task_struct);
216
217         p = leader;
218         if (unlikely(zap_leader))
219                 goto repeat;
220 }
221
222 /*
223  * This checks not only the pgrp, but falls back on the pid if no
224  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
225  * without this...
226  *
227  * The caller must hold rcu lock or the tasklist lock.
228  */
229 struct pid *session_of_pgrp(struct pid *pgrp)
230 {
231         struct task_struct *p;
232         struct pid *sid = NULL;
233
234         p = pid_task(pgrp, PIDTYPE_PGID);
235         if (p == NULL)
236                 p = pid_task(pgrp, PIDTYPE_PID);
237         if (p != NULL)
238                 sid = task_session(p);
239
240         return sid;
241 }
242
243 /*
244  * Determine if a process group is "orphaned", according to the POSIX
245  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
246  * by terminal-generated stop signals.  Newly orphaned process groups are
247  * to receive a SIGHUP and a SIGCONT.
248  *
249  * "I ask you, have you ever known what it is to be an orphan?"
250  */
251 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
252 {
253         struct task_struct *p;
254
255         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
256                 if ((p == ignored_task) ||
257                     (p->exit_state && thread_group_empty(p)) ||
258                     is_global_init(p->real_parent))
259                         continue;
260
261                 if (task_pgrp(p->real_parent) != pgrp &&
262                     task_session(p->real_parent) == task_session(p))
263                         return 0;
264         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
265
266         return 1;
267 }
268
269 int is_current_pgrp_orphaned(void)
270 {
271         int retval;
272
273         read_lock(&tasklist_lock);
274         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
275         read_unlock(&tasklist_lock);
276
277         return retval;
278 }
279
280 static int has_stopped_jobs(struct pid *pgrp)
281 {
282         int retval = 0;
283         struct task_struct *p;
284
285         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
286                 if (!task_is_stopped(p))
287                         continue;
288                 retval = 1;
289                 break;
290         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
291         return retval;
292 }
293
294 /*
295  * Check to see if any process groups have become orphaned as
296  * a result of our exiting, and if they have any stopped jobs,
297  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
298  */
299 static void
300 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
301 {
302         struct pid *pgrp = task_pgrp(tsk);
303         struct task_struct *ignored_task = tsk;
304
305         if (!parent)
306                  /* exit: our father is in a different pgrp than
307                   * we are and we were the only connection outside.
308                   */
309                 parent = tsk->real_parent;
310         else
311                 /* reparent: our child is in a different pgrp than
312                  * we are, and it was the only connection outside.
313                  */
314                 ignored_task = NULL;
315
316         if (task_pgrp(parent) != pgrp &&
317             task_session(parent) == task_session(tsk) &&
318             will_become_orphaned_pgrp(pgrp, ignored_task) &&
319             has_stopped_jobs(pgrp)) {
320                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
321                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
322         }
323 }
324
325 /**
326  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
327  *
328  * If a kernel thread is launched as a result of a system call, or if
329  * it ever exits, it should generally reparent itself to kthreadd so it
330  * isn't in the way of other processes and is correctly cleaned up on exit.
331  *
332  * The various task state such as scheduling policy and priority may have
333  * been inherited from a user process, so we reset them to sane values here.
334  *
335  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
336  */
337 static void reparent_to_kthreadd(void)
338 {
339         write_lock_irq(&tasklist_lock);
340
341         ptrace_unlink(current);
342         /* Reparent to init */
343         current->real_parent = current->parent = kthreadd_task;
344         list_move_tail(&current->sibling, &current->real_parent->children);
345
346         /* Set the exit signal to SIGCHLD so we signal init on exit */
347         current->exit_signal = SIGCHLD;
348
349         if (task_nice(current) < 0)
350                 set_user_nice(current, 0);
351         /* cpus_allowed? */
352         /* rt_priority? */
353         /* signals? */
354         memcpy(current->signal->rlim, init_task.signal->rlim,
355                sizeof(current->signal->rlim));
356
357         atomic_inc(&init_cred.usage);
358         commit_creds(&init_cred);
359         write_unlock_irq(&tasklist_lock);
360 }
361
362 void __set_special_pids(struct pid *pid)
363 {
364         struct task_struct *curr = current->group_leader;
365         pid_t nr = pid_nr(pid);
366
367         if (task_session(curr) != pid) {
368                 change_pid(curr, PIDTYPE_SID, pid);
369                 set_task_session(curr, nr);
370         }
371         if (task_pgrp(curr) != pid) {
372                 change_pid(curr, PIDTYPE_PGID, pid);
373                 set_task_pgrp(curr, nr);
374         }
375 }
376
377 static void set_special_pids(struct pid *pid)
378 {
379         write_lock_irq(&tasklist_lock);
380         __set_special_pids(pid);
381         write_unlock_irq(&tasklist_lock);
382 }
383
384 /*
385  * Let kernel threads use this to say that they
386  * allow a certain signal (since daemonize() will
387  * have disabled all of them by default).
388  */
389 int allow_signal(int sig)
390 {
391         if (!valid_signal(sig) || sig < 1)
392                 return -EINVAL;
393
394         spin_lock_irq(&current->sighand->siglock);
395         sigdelset(&current->blocked, sig);
396         if (!current->mm) {
397                 /* Kernel threads handle their own signals.
398                    Let the signal code know it'll be handled, so
399                    that they don't get converted to SIGKILL or
400                    just silently dropped */
401                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
402         }
403         recalc_sigpending();
404         spin_unlock_irq(&current->sighand->siglock);
405         return 0;
406 }
407
408 EXPORT_SYMBOL(allow_signal);
409
410 int disallow_signal(int sig)
411 {
412         if (!valid_signal(sig) || sig < 1)
413                 return -EINVAL;
414
415         spin_lock_irq(&current->sighand->siglock);
416         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
417         recalc_sigpending();
418         spin_unlock_irq(&current->sighand->siglock);
419         return 0;
420 }
421
422 EXPORT_SYMBOL(disallow_signal);
423
424 /*
425  *      Put all the gunge required to become a kernel thread without
426  *      attached user resources in one place where it belongs.
427  */
428
429 void daemonize(const char *name, ...)
430 {
431         va_list args;
432         struct fs_struct *fs;
433         sigset_t blocked;
434
435         va_start(args, name);
436         vsnprintf(current->comm, sizeof(current->comm), name, args);
437         va_end(args);
438
439         /*
440          * If we were started as result of loading a module, close all of the
441          * user space pages.  We don't need them, and if we didn't close them
442          * they would be locked into memory.
443          */
444         exit_mm(current);
445         /*
446          * We don't want to have TIF_FREEZE set if the system-wide hibernation
447          * or suspend transition begins right now.
448          */
449         current->flags |= (PF_NOFREEZE | PF_KTHREAD);
450
451         if (current->nsproxy != &init_nsproxy) {
452                 get_nsproxy(&init_nsproxy);
453                 switch_task_namespaces(current, &init_nsproxy);
454         }
455         set_special_pids(&init_struct_pid);
456         proc_clear_tty(current);
457
458         /* Block and flush all signals */
459         sigfillset(&blocked);
460         sigprocmask(SIG_BLOCK, &blocked, NULL);
461         flush_signals(current);
462
463         /* Become as one with the init task */
464
465         exit_fs(current);       /* current->fs->count--; */
466         fs = init_task.fs;
467         current->fs = fs;
468         atomic_inc(&fs->count);
469
470         exit_files(current);
471         current->files = init_task.files;
472         atomic_inc(&current->files->count);
473
474         reparent_to_kthreadd();
475 }
476
477 EXPORT_SYMBOL(daemonize);
478
479 static void close_files(struct files_struct * files)
480 {
481         int i, j;
482         struct fdtable *fdt;
483
484         j = 0;
485
486         /*
487          * It is safe to dereference the fd table without RCU or
488          * ->file_lock because this is the last reference to the
489          * files structure.
490          */
491         fdt = files_fdtable(files);
492         for (;;) {
493                 unsigned long set;
494                 i = j * __NFDBITS;
495                 if (i >= fdt->max_fds)
496                         break;
497                 set = fdt->open_fds->fds_bits[j++];
498                 while (set) {
499                         if (set & 1) {
500                                 struct file * file = xchg(&fdt->fd[i], NULL);
501                                 if (file) {
502                                         filp_close(file, files);
503                                         cond_resched();
504                                 }
505                         }
506                         i++;
507                         set >>= 1;
508                 }
509         }
510 }
511
512 struct files_struct *get_files_struct(struct task_struct *task)
513 {
514         struct files_struct *files;
515
516         task_lock(task);
517         files = task->files;
518         if (files)
519                 atomic_inc(&files->count);
520         task_unlock(task);
521
522         return files;
523 }
524
525 void put_files_struct(struct files_struct *files)
526 {
527         struct fdtable *fdt;
528
529         if (atomic_dec_and_test(&files->count)) {
530                 close_files(files);
531                 /*
532                  * Free the fd and fdset arrays if we expanded them.
533                  * If the fdtable was embedded, pass files for freeing
534                  * at the end of the RCU grace period. Otherwise,
535                  * you can free files immediately.
536                  */
537                 fdt = files_fdtable(files);
538                 if (fdt != &files->fdtab)
539                         kmem_cache_free(files_cachep, files);
540                 free_fdtable(fdt);
541         }
542 }
543
544 void reset_files_struct(struct files_struct *files)
545 {
546         struct task_struct *tsk = current;
547         struct files_struct *old;
548
549         old = tsk->files;
550         task_lock(tsk);
551         tsk->files = files;
552         task_unlock(tsk);
553         put_files_struct(old);
554 }
555
556 void exit_files(struct task_struct *tsk)
557 {
558         struct files_struct * files = tsk->files;
559
560         if (files) {
561                 task_lock(tsk);
562                 tsk->files = NULL;
563                 task_unlock(tsk);
564                 put_files_struct(files);
565         }
566 }
567
568 void put_fs_struct(struct fs_struct *fs)
569 {
570         /* No need to hold fs->lock if we are killing it */
571         if (atomic_dec_and_test(&fs->count)) {
572                 path_put(&fs->root);
573                 path_put(&fs->pwd);
574                 kmem_cache_free(fs_cachep, fs);
575         }
576 }
577
578 void exit_fs(struct task_struct *tsk)
579 {
580         struct fs_struct * fs = tsk->fs;
581
582         if (fs) {
583                 task_lock(tsk);
584                 tsk->fs = NULL;
585                 task_unlock(tsk);
586                 put_fs_struct(fs);
587         }
588 }
589
590 EXPORT_SYMBOL_GPL(exit_fs);
591
592 #ifdef CONFIG_MM_OWNER
593 /*
594  * Task p is exiting and it owned mm, lets find a new owner for it
595  */
596 static inline int
597 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
598 {
599         /*
600          * If there are other users of the mm and the owner (us) is exiting
601          * we need to find a new owner to take on the responsibility.
602          */
603         if (atomic_read(&mm->mm_users) <= 1)
604                 return 0;
605         if (mm->owner != p)
606                 return 0;
607         return 1;
608 }
609
610 void mm_update_next_owner(struct mm_struct *mm)
611 {
612         struct task_struct *c, *g, *p = current;
613
614 retry:
615         if (!mm_need_new_owner(mm, p))
616                 return;
617
618         read_lock(&tasklist_lock);
619         /*
620          * Search in the children
621          */
622         list_for_each_entry(c, &p->children, sibling) {
623                 if (c->mm == mm)
624                         goto assign_new_owner;
625         }
626
627         /*
628          * Search in the siblings
629          */
630         list_for_each_entry(c, &p->parent->children, sibling) {
631                 if (c->mm == mm)
632                         goto assign_new_owner;
633         }
634
635         /*
636          * Search through everything else. We should not get
637          * here often
638          */
639         do_each_thread(g, c) {
640                 if (c->mm == mm)
641                         goto assign_new_owner;
642         } while_each_thread(g, c);
643
644         read_unlock(&tasklist_lock);
645         /*
646          * We found no owner yet mm_users > 1: this implies that we are
647          * most likely racing with swapoff (try_to_unuse()) or /proc or
648          * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
649          */
650         mm->owner = NULL;
651         return;
652
653 assign_new_owner:
654         BUG_ON(c == p);
655         get_task_struct(c);
656         /*
657          * The task_lock protects c->mm from changing.
658          * We always want mm->owner->mm == mm
659          */
660         task_lock(c);
661         /*
662          * Delay read_unlock() till we have the task_lock()
663          * to ensure that c does not slip away underneath us
664          */
665         read_unlock(&tasklist_lock);
666         if (c->mm != mm) {
667                 task_unlock(c);
668                 put_task_struct(c);
669                 goto retry;
670         }
671         mm->owner = c;
672         task_unlock(c);
673         put_task_struct(c);
674 }
675 #endif /* CONFIG_MM_OWNER */
676
677 /*
678  * Turn us into a lazy TLB process if we
679  * aren't already..
680  */
681 static void exit_mm(struct task_struct * tsk)
682 {
683         struct mm_struct *mm = tsk->mm;
684         struct core_state *core_state;
685
686         mm_release(tsk, mm);
687         if (!mm)
688                 return;
689         /*
690          * Serialize with any possible pending coredump.
691          * We must hold mmap_sem around checking core_state
692          * and clearing tsk->mm.  The core-inducing thread
693          * will increment ->nr_threads for each thread in the
694          * group with ->mm != NULL.
695          */
696         down_read(&mm->mmap_sem);
697         core_state = mm->core_state;
698         if (core_state) {
699                 struct core_thread self;
700                 up_read(&mm->mmap_sem);
701
702                 self.task = tsk;
703                 self.next = xchg(&core_state->dumper.next, &self);
704                 /*
705                  * Implies mb(), the result of xchg() must be visible
706                  * to core_state->dumper.
707                  */
708                 if (atomic_dec_and_test(&core_state->nr_threads))
709                         complete(&core_state->startup);
710
711                 for (;;) {
712                         set_task_state(tsk, TASK_UNINTERRUPTIBLE);
713                         if (!self.task) /* see coredump_finish() */
714                                 break;
715                         schedule();
716                 }
717                 __set_task_state(tsk, TASK_RUNNING);
718                 down_read(&mm->mmap_sem);
719         }
720         atomic_inc(&mm->mm_count);
721         BUG_ON(mm != tsk->active_mm);
722         /* more a memory barrier than a real lock */
723         task_lock(tsk);
724         tsk->mm = NULL;
725         up_read(&mm->mmap_sem);
726         enter_lazy_tlb(mm, current);
727         /* We don't want this task to be frozen prematurely */
728         clear_freeze_flag(tsk);
729         task_unlock(tsk);
730         mm_update_next_owner(mm);
731         mmput(mm);
732 }
733
734 /*
735  * Called with irqs disabled, returns true if childs should reap themselves.
736  */
737 static int ignoring_children(struct sighand_struct *sigh)
738 {
739         int ret;
740         spin_lock(&sigh->siglock);
741         ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
742               (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
743         spin_unlock(&sigh->siglock);
744         return ret;
745 }
746
747 /* Returns nonzero if the tracee should be released. */
748 int __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
749 {
750         __ptrace_unlink(p);
751
752         if (p->exit_state != EXIT_ZOMBIE)
753                 return 0;
754         /*
755          * If it's a zombie, our attachedness prevented normal
756          * parent notification or self-reaping.  Do notification
757          * now if it would have happened earlier.  If it should
758          * reap itself we return true.
759          *
760          * If it's our own child, there is no notification to do.
761          * But if our normal children self-reap, then this child
762          * was prevented by ptrace and we must reap it now.
763          */
764         if (!task_detached(p) && thread_group_empty(p)) {
765                 if (!same_thread_group(p->real_parent, tracer))
766                         do_notify_parent(p, p->exit_signal);
767                 else if (ignoring_children(tracer->sighand))
768                         p->exit_signal = -1;
769         }
770
771         if (!task_detached(p))
772                 return 0;
773
774         /* Mark it as in the process of being reaped. */
775         p->exit_state = EXIT_DEAD;
776         return 1;
777 }
778
779 /*
780  * Detach all tasks we were using ptrace on.
781  * Any that need to be release_task'd are put on the @dead list.
782  *
783  * Called with write_lock(&tasklist_lock) held.
784  */
785 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
786 {
787         struct task_struct *p, *n;
788
789         list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
790                 if (__ptrace_detach(parent, p))
791                         list_add(&p->ptrace_entry, dead);
792         }
793 }
794
795 /*
796  * Finish up exit-time ptrace cleanup.
797  *
798  * Called without locks.
799  */
800 static void ptrace_exit_finish(struct task_struct *parent,
801                                struct list_head *dead)
802 {
803         struct task_struct *p, *n;
804
805         BUG_ON(!list_empty(&parent->ptraced));
806
807         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
808                 list_del_init(&p->ptrace_entry);
809                 release_task(p);
810         }
811 }
812
813 /* Returns nonzero if the child should be released. */
814 static int reparent_thread(struct task_struct *p, struct task_struct *father)
815 {
816         int dead;
817
818         if (p->pdeath_signal)
819                 /* We already hold the tasklist_lock here.  */
820                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
821
822         list_move_tail(&p->sibling, &p->real_parent->children);
823
824         if (task_detached(p))
825                 return 0;
826         /* If this is a threaded reparent there is no need to
827          * notify anyone anything has happened.
828          */
829         if (same_thread_group(p->real_parent, father))
830                 return 0;
831
832         /* We don't want people slaying init.  */
833         p->exit_signal = SIGCHLD;
834
835         /* If we'd notified the old parent about this child's death,
836          * also notify the new parent.
837          */
838         dead = 0;
839         if (!p->ptrace &&
840             p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
841                 do_notify_parent(p, p->exit_signal);
842                 if (task_detached(p)) {
843                         p->exit_state = EXIT_DEAD;
844                         dead = 1;
845                 }
846         }
847
848         kill_orphaned_pgrp(p, father);
849
850         return dead;
851 }
852
853 /*
854  * When we die, we re-parent all our children.
855  * Try to give them to another thread in our thread
856  * group, and if no such member exists, give it to
857  * the child reaper process (ie "init") in our pid
858  * space.
859  */
860 static struct task_struct *find_new_reaper(struct task_struct *father)
861 {
862         struct pid_namespace *pid_ns = task_active_pid_ns(father);
863         struct task_struct *thread;
864
865         thread = father;
866         while_each_thread(father, thread) {
867                 if (thread->flags & PF_EXITING)
868                         continue;
869                 if (unlikely(pid_ns->child_reaper == father))
870                         pid_ns->child_reaper = thread;
871                 return thread;
872         }
873
874         if (unlikely(pid_ns->child_reaper == father)) {
875                 write_unlock_irq(&tasklist_lock);
876                 if (unlikely(pid_ns == &init_pid_ns))
877                         panic("Attempted to kill init!");
878
879                 zap_pid_ns_processes(pid_ns);
880                 write_lock_irq(&tasklist_lock);
881                 /*
882                  * We can not clear ->child_reaper or leave it alone.
883                  * There may by stealth EXIT_DEAD tasks on ->children,
884                  * forget_original_parent() must move them somewhere.
885                  */
886                 pid_ns->child_reaper = init_pid_ns.child_reaper;
887         }
888
889         return pid_ns->child_reaper;
890 }
891
892 static void forget_original_parent(struct task_struct *father)
893 {
894         struct task_struct *p, *n, *reaper;
895         LIST_HEAD(ptrace_dead);
896
897         write_lock_irq(&tasklist_lock);
898         reaper = find_new_reaper(father);
899         /*
900          * First clean up ptrace if we were using it.
901          */
902         ptrace_exit(father, &ptrace_dead);
903
904         list_for_each_entry_safe(p, n, &father->children, sibling) {
905                 p->real_parent = reaper;
906                 if (p->parent == father) {
907                         BUG_ON(p->ptrace);
908                         p->parent = p->real_parent;
909                 }
910                 if (reparent_thread(p, father))
911                         list_add(&p->ptrace_entry, &ptrace_dead);;
912         }
913
914         write_unlock_irq(&tasklist_lock);
915         BUG_ON(!list_empty(&father->children));
916
917         ptrace_exit_finish(father, &ptrace_dead);
918 }
919
920 /*
921  * Send signals to all our closest relatives so that they know
922  * to properly mourn us..
923  */
924 static void exit_notify(struct task_struct *tsk, int group_dead)
925 {
926         int signal;
927         void *cookie;
928
929         /*
930          * This does two things:
931          *
932          * A.  Make init inherit all the child processes
933          * B.  Check to see if any process groups have become orphaned
934          *      as a result of our exiting, and if they have any stopped
935          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
936          */
937         forget_original_parent(tsk);
938         exit_task_namespaces(tsk);
939
940         write_lock_irq(&tasklist_lock);
941         if (group_dead)
942                 kill_orphaned_pgrp(tsk->group_leader, NULL);
943
944         /* Let father know we died
945          *
946          * Thread signals are configurable, but you aren't going to use
947          * that to send signals to arbitary processes.
948          * That stops right now.
949          *
950          * If the parent exec id doesn't match the exec id we saved
951          * when we started then we know the parent has changed security
952          * domain.
953          *
954          * If our self_exec id doesn't match our parent_exec_id then
955          * we have changed execution domain as these two values started
956          * the same after a fork.
957          */
958         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
959             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
960              tsk->self_exec_id != tsk->parent_exec_id) &&
961             !capable(CAP_KILL))
962                 tsk->exit_signal = SIGCHLD;
963
964         signal = tracehook_notify_death(tsk, &cookie, group_dead);
965         if (signal >= 0)
966                 signal = do_notify_parent(tsk, signal);
967
968         tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
969
970         /* mt-exec, de_thread() is waiting for us */
971         if (thread_group_leader(tsk) &&
972             tsk->signal->group_exit_task &&
973             tsk->signal->notify_count < 0)
974                 wake_up_process(tsk->signal->group_exit_task);
975
976         write_unlock_irq(&tasklist_lock);
977
978         tracehook_report_death(tsk, signal, cookie, group_dead);
979
980         /* If the process is dead, release it - nobody will wait for it */
981         if (signal == DEATH_REAP)
982                 release_task(tsk);
983 }
984
985 #ifdef CONFIG_DEBUG_STACK_USAGE
986 static void check_stack_usage(void)
987 {
988         static DEFINE_SPINLOCK(low_water_lock);
989         static int lowest_to_date = THREAD_SIZE;
990         unsigned long free;
991
992         free = stack_not_used(current);
993
994         if (free >= lowest_to_date)
995                 return;
996
997         spin_lock(&low_water_lock);
998         if (free < lowest_to_date) {
999                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
1000                                 "left\n",
1001                                 current->comm, free);
1002                 lowest_to_date = free;
1003         }
1004         spin_unlock(&low_water_lock);
1005 }
1006 #else
1007 static inline void check_stack_usage(void) {}
1008 #endif
1009
1010 NORET_TYPE void do_exit(long code)
1011 {
1012         struct task_struct *tsk = current;
1013         int group_dead;
1014
1015         profile_task_exit(tsk);
1016
1017         WARN_ON(atomic_read(&tsk->fs_excl));
1018
1019         if (unlikely(in_interrupt()))
1020                 panic("Aiee, killing interrupt handler!");
1021         if (unlikely(!tsk->pid))
1022                 panic("Attempted to kill the idle task!");
1023
1024         tracehook_report_exit(&code);
1025
1026         /*
1027          * We're taking recursive faults here in do_exit. Safest is to just
1028          * leave this task alone and wait for reboot.
1029          */
1030         if (unlikely(tsk->flags & PF_EXITING)) {
1031                 printk(KERN_ALERT
1032                         "Fixing recursive fault but reboot is needed!\n");
1033                 /*
1034                  * We can do this unlocked here. The futex code uses
1035                  * this flag just to verify whether the pi state
1036                  * cleanup has been done or not. In the worst case it
1037                  * loops once more. We pretend that the cleanup was
1038                  * done as there is no way to return. Either the
1039                  * OWNER_DIED bit is set by now or we push the blocked
1040                  * task into the wait for ever nirwana as well.
1041                  */
1042                 tsk->flags |= PF_EXITPIDONE;
1043                 set_current_state(TASK_UNINTERRUPTIBLE);
1044                 schedule();
1045         }
1046
1047         exit_signals(tsk);  /* sets PF_EXITING */
1048         /*
1049          * tsk->flags are checked in the futex code to protect against
1050          * an exiting task cleaning up the robust pi futexes.
1051          */
1052         smp_mb();
1053         spin_unlock_wait(&tsk->pi_lock);
1054
1055         if (unlikely(in_atomic()))
1056                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1057                                 current->comm, task_pid_nr(current),
1058                                 preempt_count());
1059
1060         acct_update_integrals(tsk);
1061
1062         group_dead = atomic_dec_and_test(&tsk->signal->live);
1063         if (group_dead) {
1064                 hrtimer_cancel(&tsk->signal->real_timer);
1065                 exit_itimers(tsk->signal);
1066         }
1067         acct_collect(code, group_dead);
1068         if (group_dead)
1069                 tty_audit_exit();
1070         if (unlikely(tsk->audit_context))
1071                 audit_free(tsk);
1072
1073         tsk->exit_code = code;
1074         taskstats_exit(tsk, group_dead);
1075
1076         exit_mm(tsk);
1077
1078         if (group_dead)
1079                 acct_process();
1080         trace_sched_process_exit(tsk);
1081
1082         exit_sem(tsk);
1083         exit_files(tsk);
1084         exit_fs(tsk);
1085         check_stack_usage();
1086         exit_thread();
1087         cgroup_exit(tsk, 1);
1088
1089         if (group_dead && tsk->signal->leader)
1090                 disassociate_ctty(1);
1091
1092         module_put(task_thread_info(tsk)->exec_domain->module);
1093         if (tsk->binfmt)
1094                 module_put(tsk->binfmt->module);
1095
1096         proc_exit_connector(tsk);
1097         exit_notify(tsk, group_dead);
1098 #ifdef CONFIG_NUMA
1099         mpol_put(tsk->mempolicy);
1100         tsk->mempolicy = NULL;
1101 #endif
1102 #ifdef CONFIG_FUTEX
1103         /*
1104          * This must happen late, after the PID is not
1105          * hashed anymore:
1106          */
1107         if (unlikely(!list_empty(&tsk->pi_state_list)))
1108                 exit_pi_state_list(tsk);
1109         if (unlikely(current->pi_state_cache))
1110                 kfree(current->pi_state_cache);
1111 #endif
1112         /*
1113          * Make sure we are holding no locks:
1114          */
1115         debug_check_no_locks_held(tsk);
1116         /*
1117          * We can do this unlocked here. The futex code uses this flag
1118          * just to verify whether the pi state cleanup has been done
1119          * or not. In the worst case it loops once more.
1120          */
1121         tsk->flags |= PF_EXITPIDONE;
1122
1123         if (tsk->io_context)
1124                 exit_io_context();
1125
1126         if (tsk->splice_pipe)
1127                 __free_pipe_info(tsk->splice_pipe);
1128
1129         preempt_disable();
1130         /* causes final put_task_struct in finish_task_switch(). */
1131         tsk->state = TASK_DEAD;
1132         schedule();
1133         BUG();
1134         /* Avoid "noreturn function does return".  */
1135         for (;;)
1136                 cpu_relax();    /* For when BUG is null */
1137 }
1138
1139 EXPORT_SYMBOL_GPL(do_exit);
1140
1141 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1142 {
1143         if (comp)
1144                 complete(comp);
1145
1146         do_exit(code);
1147 }
1148
1149 EXPORT_SYMBOL(complete_and_exit);
1150
1151 SYSCALL_DEFINE1(exit, int, error_code)
1152 {
1153         do_exit((error_code&0xff)<<8);
1154 }
1155
1156 /*
1157  * Take down every thread in the group.  This is called by fatal signals
1158  * as well as by sys_exit_group (below).
1159  */
1160 NORET_TYPE void
1161 do_group_exit(int exit_code)
1162 {
1163         struct signal_struct *sig = current->signal;
1164
1165         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1166
1167         if (signal_group_exit(sig))
1168                 exit_code = sig->group_exit_code;
1169         else if (!thread_group_empty(current)) {
1170                 struct sighand_struct *const sighand = current->sighand;
1171                 spin_lock_irq(&sighand->siglock);
1172                 if (signal_group_exit(sig))
1173                         /* Another thread got here before we took the lock.  */
1174                         exit_code = sig->group_exit_code;
1175                 else {
1176                         sig->group_exit_code = exit_code;
1177                         sig->flags = SIGNAL_GROUP_EXIT;
1178                         zap_other_threads(current);
1179                 }
1180                 spin_unlock_irq(&sighand->siglock);
1181         }
1182
1183         do_exit(exit_code);
1184         /* NOTREACHED */
1185 }
1186
1187 /*
1188  * this kills every thread in the thread group. Note that any externally
1189  * wait4()-ing process will get the correct exit code - even if this
1190  * thread is not the thread group leader.
1191  */
1192 SYSCALL_DEFINE1(exit_group, int, error_code)
1193 {
1194         do_group_exit((error_code & 0xff) << 8);
1195         /* NOTREACHED */
1196         return 0;
1197 }
1198
1199 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1200 {
1201         struct pid *pid = NULL;
1202         if (type == PIDTYPE_PID)
1203                 pid = task->pids[type].pid;
1204         else if (type < PIDTYPE_MAX)
1205                 pid = task->group_leader->pids[type].pid;
1206         return pid;
1207 }
1208
1209 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1210                           struct task_struct *p)
1211 {
1212         int err;
1213
1214         if (type < PIDTYPE_MAX) {
1215                 if (task_pid_type(p, type) != pid)
1216                         return 0;
1217         }
1218
1219         /* Wait for all children (clone and not) if __WALL is set;
1220          * otherwise, wait for clone children *only* if __WCLONE is
1221          * set; otherwise, wait for non-clone children *only*.  (Note:
1222          * A "clone" child here is one that reports to its parent
1223          * using a signal other than SIGCHLD.) */
1224         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1225             && !(options & __WALL))
1226                 return 0;
1227
1228         err = security_task_wait(p);
1229         if (err)
1230                 return err;
1231
1232         return 1;
1233 }
1234
1235 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1236                                int why, int status,
1237                                struct siginfo __user *infop,
1238                                struct rusage __user *rusagep)
1239 {
1240         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1241
1242         put_task_struct(p);
1243         if (!retval)
1244                 retval = put_user(SIGCHLD, &infop->si_signo);
1245         if (!retval)
1246                 retval = put_user(0, &infop->si_errno);
1247         if (!retval)
1248                 retval = put_user((short)why, &infop->si_code);
1249         if (!retval)
1250                 retval = put_user(pid, &infop->si_pid);
1251         if (!retval)
1252                 retval = put_user(uid, &infop->si_uid);
1253         if (!retval)
1254                 retval = put_user(status, &infop->si_status);
1255         if (!retval)
1256                 retval = pid;
1257         return retval;
1258 }
1259
1260 /*
1261  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1262  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1263  * the lock and this task is uninteresting.  If we return nonzero, we have
1264  * released the lock and the system call should return.
1265  */
1266 static int wait_task_zombie(struct task_struct *p, int options,
1267                             struct siginfo __user *infop,
1268                             int __user *stat_addr, struct rusage __user *ru)
1269 {
1270         unsigned long state;
1271         int retval, status, traced;
1272         pid_t pid = task_pid_vnr(p);
1273         uid_t uid = __task_cred(p)->uid;
1274
1275         if (!likely(options & WEXITED))
1276                 return 0;
1277
1278         if (unlikely(options & WNOWAIT)) {
1279                 int exit_code = p->exit_code;
1280                 int why, status;
1281
1282                 get_task_struct(p);
1283                 read_unlock(&tasklist_lock);
1284                 if ((exit_code & 0x7f) == 0) {
1285                         why = CLD_EXITED;
1286                         status = exit_code >> 8;
1287                 } else {
1288                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1289                         status = exit_code & 0x7f;
1290                 }
1291                 return wait_noreap_copyout(p, pid, uid, why,
1292                                            status, infop, ru);
1293         }
1294
1295         /*
1296          * Try to move the task's state to DEAD
1297          * only one thread is allowed to do this:
1298          */
1299         state = xchg(&p->exit_state, EXIT_DEAD);
1300         if (state != EXIT_ZOMBIE) {
1301                 BUG_ON(state != EXIT_DEAD);
1302                 return 0;
1303         }
1304
1305         traced = ptrace_reparented(p);
1306
1307         if (likely(!traced)) {
1308                 struct signal_struct *psig;
1309                 struct signal_struct *sig;
1310                 struct task_cputime cputime;
1311
1312                 /*
1313                  * The resource counters for the group leader are in its
1314                  * own task_struct.  Those for dead threads in the group
1315                  * are in its signal_struct, as are those for the child
1316                  * processes it has previously reaped.  All these
1317                  * accumulate in the parent's signal_struct c* fields.
1318                  *
1319                  * We don't bother to take a lock here to protect these
1320                  * p->signal fields, because they are only touched by
1321                  * __exit_signal, which runs with tasklist_lock
1322                  * write-locked anyway, and so is excluded here.  We do
1323                  * need to protect the access to p->parent->signal fields,
1324                  * as other threads in the parent group can be right
1325                  * here reaping other children at the same time.
1326                  *
1327                  * We use thread_group_cputime() to get times for the thread
1328                  * group, which consolidates times for all threads in the
1329                  * group including the group leader.
1330                  */
1331                 thread_group_cputime(p, &cputime);
1332                 spin_lock_irq(&p->parent->sighand->siglock);
1333                 psig = p->parent->signal;
1334                 sig = p->signal;
1335                 psig->cutime =
1336                         cputime_add(psig->cutime,
1337                         cputime_add(cputime.utime,
1338                                     sig->cutime));
1339                 psig->cstime =
1340                         cputime_add(psig->cstime,
1341                         cputime_add(cputime.stime,
1342                                     sig->cstime));
1343                 psig->cgtime =
1344                         cputime_add(psig->cgtime,
1345                         cputime_add(p->gtime,
1346                         cputime_add(sig->gtime,
1347                                     sig->cgtime)));
1348                 psig->cmin_flt +=
1349                         p->min_flt + sig->min_flt + sig->cmin_flt;
1350                 psig->cmaj_flt +=
1351                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1352                 psig->cnvcsw +=
1353                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1354                 psig->cnivcsw +=
1355                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1356                 psig->cinblock +=
1357                         task_io_get_inblock(p) +
1358                         sig->inblock + sig->cinblock;
1359                 psig->coublock +=
1360                         task_io_get_oublock(p) +
1361                         sig->oublock + sig->coublock;
1362                 task_io_accounting_add(&psig->ioac, &p->ioac);
1363                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1364                 spin_unlock_irq(&p->parent->sighand->siglock);
1365         }
1366
1367         /*
1368          * Now we are sure this task is interesting, and no other
1369          * thread can reap it because we set its state to EXIT_DEAD.
1370          */
1371         read_unlock(&tasklist_lock);
1372
1373         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1374         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1375                 ? p->signal->group_exit_code : p->exit_code;
1376         if (!retval && stat_addr)
1377                 retval = put_user(status, stat_addr);
1378         if (!retval && infop)
1379                 retval = put_user(SIGCHLD, &infop->si_signo);
1380         if (!retval && infop)
1381                 retval = put_user(0, &infop->si_errno);
1382         if (!retval && infop) {
1383                 int why;
1384
1385                 if ((status & 0x7f) == 0) {
1386                         why = CLD_EXITED;
1387                         status >>= 8;
1388                 } else {
1389                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1390                         status &= 0x7f;
1391                 }
1392                 retval = put_user((short)why, &infop->si_code);
1393                 if (!retval)
1394                         retval = put_user(status, &infop->si_status);
1395         }
1396         if (!retval && infop)
1397                 retval = put_user(pid, &infop->si_pid);
1398         if (!retval && infop)
1399                 retval = put_user(uid, &infop->si_uid);
1400         if (!retval)
1401                 retval = pid;
1402
1403         if (traced) {
1404                 write_lock_irq(&tasklist_lock);
1405                 /* We dropped tasklist, ptracer could die and untrace */
1406                 ptrace_unlink(p);
1407                 /*
1408                  * If this is not a detached task, notify the parent.
1409                  * If it's still not detached after that, don't release
1410                  * it now.
1411                  */
1412                 if (!task_detached(p)) {
1413                         do_notify_parent(p, p->exit_signal);
1414                         if (!task_detached(p)) {
1415                                 p->exit_state = EXIT_ZOMBIE;
1416                                 p = NULL;
1417                         }
1418                 }
1419                 write_unlock_irq(&tasklist_lock);
1420         }
1421         if (p != NULL)
1422                 release_task(p);
1423
1424         return retval;
1425 }
1426
1427 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1428 {
1429         if (ptrace) {
1430                 if (task_is_stopped_or_traced(p))
1431                         return &p->exit_code;
1432         } else {
1433                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1434                         return &p->signal->group_exit_code;
1435         }
1436         return NULL;
1437 }
1438
1439 /*
1440  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1441  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1442  * the lock and this task is uninteresting.  If we return nonzero, we have
1443  * released the lock and the system call should return.
1444  */
1445 static int wait_task_stopped(int ptrace, struct task_struct *p,
1446                              int options, struct siginfo __user *infop,
1447                              int __user *stat_addr, struct rusage __user *ru)
1448 {
1449         int retval, exit_code, *p_code, why;
1450         uid_t uid = 0; /* unneeded, required by compiler */
1451         pid_t pid;
1452
1453         if (!(options & WUNTRACED))
1454                 return 0;
1455
1456         exit_code = 0;
1457         spin_lock_irq(&p->sighand->siglock);
1458
1459         p_code = task_stopped_code(p, ptrace);
1460         if (unlikely(!p_code))
1461                 goto unlock_sig;
1462
1463         exit_code = *p_code;
1464         if (!exit_code)
1465                 goto unlock_sig;
1466
1467         if (!unlikely(options & WNOWAIT))
1468                 *p_code = 0;
1469
1470         /* don't need the RCU readlock here as we're holding a spinlock */
1471         uid = __task_cred(p)->uid;
1472 unlock_sig:
1473         spin_unlock_irq(&p->sighand->siglock);
1474         if (!exit_code)
1475                 return 0;
1476
1477         /*
1478          * Now we are pretty sure this task is interesting.
1479          * Make sure it doesn't get reaped out from under us while we
1480          * give up the lock and then examine it below.  We don't want to
1481          * keep holding onto the tasklist_lock while we call getrusage and
1482          * possibly take page faults for user memory.
1483          */
1484         get_task_struct(p);
1485         pid = task_pid_vnr(p);
1486         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1487         read_unlock(&tasklist_lock);
1488
1489         if (unlikely(options & WNOWAIT))
1490                 return wait_noreap_copyout(p, pid, uid,
1491                                            why, exit_code,
1492                                            infop, ru);
1493
1494         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1495         if (!retval && stat_addr)
1496                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1497         if (!retval && infop)
1498                 retval = put_user(SIGCHLD, &infop->si_signo);
1499         if (!retval && infop)
1500                 retval = put_user(0, &infop->si_errno);
1501         if (!retval && infop)
1502                 retval = put_user((short)why, &infop->si_code);
1503         if (!retval && infop)
1504                 retval = put_user(exit_code, &infop->si_status);
1505         if (!retval && infop)
1506                 retval = put_user(pid, &infop->si_pid);
1507         if (!retval && infop)
1508                 retval = put_user(uid, &infop->si_uid);
1509         if (!retval)
1510                 retval = pid;
1511         put_task_struct(p);
1512
1513         BUG_ON(!retval);
1514         return retval;
1515 }
1516
1517 /*
1518  * Handle do_wait work for one task in a live, non-stopped state.
1519  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1520  * the lock and this task is uninteresting.  If we return nonzero, we have
1521  * released the lock and the system call should return.
1522  */
1523 static int wait_task_continued(struct task_struct *p, int options,
1524                                struct siginfo __user *infop,
1525                                int __user *stat_addr, struct rusage __user *ru)
1526 {
1527         int retval;
1528         pid_t pid;
1529         uid_t uid;
1530
1531         if (!unlikely(options & WCONTINUED))
1532                 return 0;
1533
1534         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1535                 return 0;
1536
1537         spin_lock_irq(&p->sighand->siglock);
1538         /* Re-check with the lock held.  */
1539         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1540                 spin_unlock_irq(&p->sighand->siglock);
1541                 return 0;
1542         }
1543         if (!unlikely(options & WNOWAIT))
1544                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1545         uid = __task_cred(p)->uid;
1546         spin_unlock_irq(&p->sighand->siglock);
1547
1548         pid = task_pid_vnr(p);
1549         get_task_struct(p);
1550         read_unlock(&tasklist_lock);
1551
1552         if (!infop) {
1553                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1554                 put_task_struct(p);
1555                 if (!retval && stat_addr)
1556                         retval = put_user(0xffff, stat_addr);
1557                 if (!retval)
1558                         retval = pid;
1559         } else {
1560                 retval = wait_noreap_copyout(p, pid, uid,
1561                                              CLD_CONTINUED, SIGCONT,
1562                                              infop, ru);
1563                 BUG_ON(retval == 0);
1564         }
1565
1566         return retval;
1567 }
1568
1569 /*
1570  * Consider @p for a wait by @parent.
1571  *
1572  * -ECHILD should be in *@notask_error before the first call.
1573  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1574  * Returns zero if the search for a child should continue;
1575  * then *@notask_error is 0 if @p is an eligible child,
1576  * or another error from security_task_wait(), or still -ECHILD.
1577  */
1578 static int wait_consider_task(struct task_struct *parent, int ptrace,
1579                               struct task_struct *p, int *notask_error,
1580                               enum pid_type type, struct pid *pid, int options,
1581                               struct siginfo __user *infop,
1582                               int __user *stat_addr, struct rusage __user *ru)
1583 {
1584         int ret = eligible_child(type, pid, options, p);
1585         if (!ret)
1586                 return ret;
1587
1588         if (unlikely(ret < 0)) {
1589                 /*
1590                  * If we have not yet seen any eligible child,
1591                  * then let this error code replace -ECHILD.
1592                  * A permission error will give the user a clue
1593                  * to look for security policy problems, rather
1594                  * than for mysterious wait bugs.
1595                  */
1596                 if (*notask_error)
1597                         *notask_error = ret;
1598         }
1599
1600         if (likely(!ptrace) && unlikely(p->ptrace)) {
1601                 /*
1602                  * This child is hidden by ptrace.
1603                  * We aren't allowed to see it now, but eventually we will.
1604                  */
1605                 *notask_error = 0;
1606                 return 0;
1607         }
1608
1609         if (p->exit_state == EXIT_DEAD)
1610                 return 0;
1611
1612         /*
1613          * We don't reap group leaders with subthreads.
1614          */
1615         if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1616                 return wait_task_zombie(p, options, infop, stat_addr, ru);
1617
1618         /*
1619          * It's stopped or running now, so it might
1620          * later continue, exit, or stop again.
1621          */
1622         *notask_error = 0;
1623
1624         if (task_stopped_code(p, ptrace))
1625                 return wait_task_stopped(ptrace, p, options,
1626                                          infop, stat_addr, ru);
1627
1628         return wait_task_continued(p, options, infop, stat_addr, ru);
1629 }
1630
1631 /*
1632  * Do the work of do_wait() for one thread in the group, @tsk.
1633  *
1634  * -ECHILD should be in *@notask_error before the first call.
1635  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1636  * Returns zero if the search for a child should continue; then
1637  * *@notask_error is 0 if there were any eligible children,
1638  * or another error from security_task_wait(), or still -ECHILD.
1639  */
1640 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1641                           enum pid_type type, struct pid *pid, int options,
1642                           struct siginfo __user *infop, int __user *stat_addr,
1643                           struct rusage __user *ru)
1644 {
1645         struct task_struct *p;
1646
1647         list_for_each_entry(p, &tsk->children, sibling) {
1648                 /*
1649                  * Do not consider detached threads.
1650                  */
1651                 if (!task_detached(p)) {
1652                         int ret = wait_consider_task(tsk, 0, p, notask_error,
1653                                                      type, pid, options,
1654                                                      infop, stat_addr, ru);
1655                         if (ret)
1656                                 return ret;
1657                 }
1658         }
1659
1660         return 0;
1661 }
1662
1663 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1664                           enum pid_type type, struct pid *pid, int options,
1665                           struct siginfo __user *infop, int __user *stat_addr,
1666                           struct rusage __user *ru)
1667 {
1668         struct task_struct *p;
1669
1670         /*
1671          * Traditionally we see ptrace'd stopped tasks regardless of options.
1672          */
1673         options |= WUNTRACED;
1674
1675         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1676                 int ret = wait_consider_task(tsk, 1, p, notask_error,
1677                                              type, pid, options,
1678                                              infop, stat_addr, ru);
1679                 if (ret)
1680                         return ret;
1681         }
1682
1683         return 0;
1684 }
1685
1686 static long do_wait(enum pid_type type, struct pid *pid, int options,
1687                     struct siginfo __user *infop, int __user *stat_addr,
1688                     struct rusage __user *ru)
1689 {
1690         DECLARE_WAITQUEUE(wait, current);
1691         struct task_struct *tsk;
1692         int retval;
1693
1694         trace_sched_process_wait(pid);
1695
1696         add_wait_queue(&current->signal->wait_chldexit,&wait);
1697 repeat:
1698         /*
1699          * If there is nothing that can match our critiera just get out.
1700          * We will clear @retval to zero if we see any child that might later
1701          * match our criteria, even if we are not able to reap it yet.
1702          */
1703         retval = -ECHILD;
1704         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1705                 goto end;
1706
1707         current->state = TASK_INTERRUPTIBLE;
1708         read_lock(&tasklist_lock);
1709         tsk = current;
1710         do {
1711                 int tsk_result = do_wait_thread(tsk, &retval,
1712                                                 type, pid, options,
1713                                                 infop, stat_addr, ru);
1714                 if (!tsk_result)
1715                         tsk_result = ptrace_do_wait(tsk, &retval,
1716                                                     type, pid, options,
1717                                                     infop, stat_addr, ru);
1718                 if (tsk_result) {
1719                         /*
1720                          * tasklist_lock is unlocked and we have a final result.
1721                          */
1722                         retval = tsk_result;
1723                         goto end;
1724                 }
1725
1726                 if (options & __WNOTHREAD)
1727                         break;
1728                 tsk = next_thread(tsk);
1729                 BUG_ON(tsk->signal != current->signal);
1730         } while (tsk != current);
1731         read_unlock(&tasklist_lock);
1732
1733         if (!retval && !(options & WNOHANG)) {
1734                 retval = -ERESTARTSYS;
1735                 if (!signal_pending(current)) {
1736                         schedule();
1737                         goto repeat;
1738                 }
1739         }
1740
1741 end:
1742         current->state = TASK_RUNNING;
1743         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1744         if (infop) {
1745                 if (retval > 0)
1746                         retval = 0;
1747                 else {
1748                         /*
1749                          * For a WNOHANG return, clear out all the fields
1750                          * we would set so the user can easily tell the
1751                          * difference.
1752                          */
1753                         if (!retval)
1754                                 retval = put_user(0, &infop->si_signo);
1755                         if (!retval)
1756                                 retval = put_user(0, &infop->si_errno);
1757                         if (!retval)
1758                                 retval = put_user(0, &infop->si_code);
1759                         if (!retval)
1760                                 retval = put_user(0, &infop->si_pid);
1761                         if (!retval)
1762                                 retval = put_user(0, &infop->si_uid);
1763                         if (!retval)
1764                                 retval = put_user(0, &infop->si_status);
1765                 }
1766         }
1767         return retval;
1768 }
1769
1770 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1771                 infop, int, options, struct rusage __user *, ru)
1772 {
1773         struct pid *pid = NULL;
1774         enum pid_type type;
1775         long ret;
1776
1777         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1778                 return -EINVAL;
1779         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1780                 return -EINVAL;
1781
1782         switch (which) {
1783         case P_ALL:
1784                 type = PIDTYPE_MAX;
1785                 break;
1786         case P_PID:
1787                 type = PIDTYPE_PID;
1788                 if (upid <= 0)
1789                         return -EINVAL;
1790                 break;
1791         case P_PGID:
1792                 type = PIDTYPE_PGID;
1793                 if (upid <= 0)
1794                         return -EINVAL;
1795                 break;
1796         default:
1797                 return -EINVAL;
1798         }
1799
1800         if (type < PIDTYPE_MAX)
1801                 pid = find_get_pid(upid);
1802         ret = do_wait(type, pid, options, infop, NULL, ru);
1803         put_pid(pid);
1804
1805         /* avoid REGPARM breakage on x86: */
1806         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1807         return ret;
1808 }
1809
1810 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1811                 int, options, struct rusage __user *, ru)
1812 {
1813         struct pid *pid = NULL;
1814         enum pid_type type;
1815         long ret;
1816
1817         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1818                         __WNOTHREAD|__WCLONE|__WALL))
1819                 return -EINVAL;
1820
1821         if (upid == -1)
1822                 type = PIDTYPE_MAX;
1823         else if (upid < 0) {
1824                 type = PIDTYPE_PGID;
1825                 pid = find_get_pid(-upid);
1826         } else if (upid == 0) {
1827                 type = PIDTYPE_PGID;
1828                 pid = get_pid(task_pgrp(current));
1829         } else /* upid > 0 */ {
1830                 type = PIDTYPE_PID;
1831                 pid = find_get_pid(upid);
1832         }
1833
1834         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1835         put_pid(pid);
1836
1837         /* avoid REGPARM breakage on x86: */
1838         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1839         return ret;
1840 }
1841
1842 #ifdef __ARCH_WANT_SYS_WAITPID
1843
1844 /*
1845  * sys_waitpid() remains for compatibility. waitpid() should be
1846  * implemented by calling sys_wait4() from libc.a.
1847  */
1848 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1849 {
1850         return sys_wait4(pid, stat_addr, options, NULL);
1851 }
1852
1853 #endif