[CVE-2009-0029] Convert all system calls to return a long
[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->gtime = cputime_add(sig->gtime, task_gtime(tsk));
122                 sig->min_flt += tsk->min_flt;
123                 sig->maj_flt += tsk->maj_flt;
124                 sig->nvcsw += tsk->nvcsw;
125                 sig->nivcsw += tsk->nivcsw;
126                 sig->inblock += task_io_get_inblock(tsk);
127                 sig->oublock += task_io_get_oublock(tsk);
128                 task_io_accounting_add(&sig->ioac, &tsk->ioac);
129                 sig = NULL; /* Marker for below. */
130         }
131
132         __unhash_process(tsk);
133
134         /*
135          * Do this under ->siglock, we can race with another thread
136          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
137          */
138         flush_sigqueue(&tsk->pending);
139
140         tsk->signal = NULL;
141         tsk->sighand = NULL;
142         spin_unlock(&sighand->siglock);
143
144         __cleanup_sighand(sighand);
145         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
146         if (sig) {
147                 flush_sigqueue(&sig->shared_pending);
148                 taskstats_tgid_free(sig);
149                 /*
150                  * Make sure ->signal can't go away under rq->lock,
151                  * see account_group_exec_runtime().
152                  */
153                 task_rq_unlock_wait(tsk);
154                 __cleanup_signal(sig);
155         }
156 }
157
158 static void delayed_put_task_struct(struct rcu_head *rhp)
159 {
160         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
161
162         trace_sched_process_free(tsk);
163         put_task_struct(tsk);
164 }
165
166
167 void release_task(struct task_struct * p)
168 {
169         struct task_struct *leader;
170         int zap_leader;
171 repeat:
172         tracehook_prepare_release_task(p);
173         /* don't need to get the RCU readlock here - the process is dead and
174          * can't be modifying its own credentials */
175         atomic_dec(&__task_cred(p)->user->processes);
176
177         proc_flush_task(p);
178         write_lock_irq(&tasklist_lock);
179         tracehook_finish_release_task(p);
180         __exit_signal(p);
181
182         /*
183          * If we are the last non-leader member of the thread
184          * group, and the leader is zombie, then notify the
185          * group leader's parent process. (if it wants notification.)
186          */
187         zap_leader = 0;
188         leader = p->group_leader;
189         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
190                 BUG_ON(task_detached(leader));
191                 do_notify_parent(leader, leader->exit_signal);
192                 /*
193                  * If we were the last child thread and the leader has
194                  * exited already, and the leader's parent ignores SIGCHLD,
195                  * then we are the one who should release the leader.
196                  *
197                  * do_notify_parent() will have marked it self-reaping in
198                  * that case.
199                  */
200                 zap_leader = task_detached(leader);
201
202                 /*
203                  * This maintains the invariant that release_task()
204                  * only runs on a task in EXIT_DEAD, just for sanity.
205                  */
206                 if (zap_leader)
207                         leader->exit_state = EXIT_DEAD;
208         }
209
210         write_unlock_irq(&tasklist_lock);
211         release_thread(p);
212         call_rcu(&p->rcu, delayed_put_task_struct);
213
214         p = leader;
215         if (unlikely(zap_leader))
216                 goto repeat;
217 }
218
219 /*
220  * This checks not only the pgrp, but falls back on the pid if no
221  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
222  * without this...
223  *
224  * The caller must hold rcu lock or the tasklist lock.
225  */
226 struct pid *session_of_pgrp(struct pid *pgrp)
227 {
228         struct task_struct *p;
229         struct pid *sid = NULL;
230
231         p = pid_task(pgrp, PIDTYPE_PGID);
232         if (p == NULL)
233                 p = pid_task(pgrp, PIDTYPE_PID);
234         if (p != NULL)
235                 sid = task_session(p);
236
237         return sid;
238 }
239
240 /*
241  * Determine if a process group is "orphaned", according to the POSIX
242  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
243  * by terminal-generated stop signals.  Newly orphaned process groups are
244  * to receive a SIGHUP and a SIGCONT.
245  *
246  * "I ask you, have you ever known what it is to be an orphan?"
247  */
248 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
249 {
250         struct task_struct *p;
251
252         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
253                 if ((p == ignored_task) ||
254                     (p->exit_state && thread_group_empty(p)) ||
255                     is_global_init(p->real_parent))
256                         continue;
257
258                 if (task_pgrp(p->real_parent) != pgrp &&
259                     task_session(p->real_parent) == task_session(p))
260                         return 0;
261         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
262
263         return 1;
264 }
265
266 int is_current_pgrp_orphaned(void)
267 {
268         int retval;
269
270         read_lock(&tasklist_lock);
271         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
272         read_unlock(&tasklist_lock);
273
274         return retval;
275 }
276
277 static int has_stopped_jobs(struct pid *pgrp)
278 {
279         int retval = 0;
280         struct task_struct *p;
281
282         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
283                 if (!task_is_stopped(p))
284                         continue;
285                 retval = 1;
286                 break;
287         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
288         return retval;
289 }
290
291 /*
292  * Check to see if any process groups have become orphaned as
293  * a result of our exiting, and if they have any stopped jobs,
294  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
295  */
296 static void
297 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
298 {
299         struct pid *pgrp = task_pgrp(tsk);
300         struct task_struct *ignored_task = tsk;
301
302         if (!parent)
303                  /* exit: our father is in a different pgrp than
304                   * we are and we were the only connection outside.
305                   */
306                 parent = tsk->real_parent;
307         else
308                 /* reparent: our child is in a different pgrp than
309                  * we are, and it was the only connection outside.
310                  */
311                 ignored_task = NULL;
312
313         if (task_pgrp(parent) != pgrp &&
314             task_session(parent) == task_session(tsk) &&
315             will_become_orphaned_pgrp(pgrp, ignored_task) &&
316             has_stopped_jobs(pgrp)) {
317                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
318                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
319         }
320 }
321
322 /**
323  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
324  *
325  * If a kernel thread is launched as a result of a system call, or if
326  * it ever exits, it should generally reparent itself to kthreadd so it
327  * isn't in the way of other processes and is correctly cleaned up on exit.
328  *
329  * The various task state such as scheduling policy and priority may have
330  * been inherited from a user process, so we reset them to sane values here.
331  *
332  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
333  */
334 static void reparent_to_kthreadd(void)
335 {
336         write_lock_irq(&tasklist_lock);
337
338         ptrace_unlink(current);
339         /* Reparent to init */
340         current->real_parent = current->parent = kthreadd_task;
341         list_move_tail(&current->sibling, &current->real_parent->children);
342
343         /* Set the exit signal to SIGCHLD so we signal init on exit */
344         current->exit_signal = SIGCHLD;
345
346         if (task_nice(current) < 0)
347                 set_user_nice(current, 0);
348         /* cpus_allowed? */
349         /* rt_priority? */
350         /* signals? */
351         memcpy(current->signal->rlim, init_task.signal->rlim,
352                sizeof(current->signal->rlim));
353
354         atomic_inc(&init_cred.usage);
355         commit_creds(&init_cred);
356         write_unlock_irq(&tasklist_lock);
357 }
358
359 void __set_special_pids(struct pid *pid)
360 {
361         struct task_struct *curr = current->group_leader;
362         pid_t nr = pid_nr(pid);
363
364         if (task_session(curr) != pid) {
365                 change_pid(curr, PIDTYPE_SID, pid);
366                 set_task_session(curr, nr);
367         }
368         if (task_pgrp(curr) != pid) {
369                 change_pid(curr, PIDTYPE_PGID, pid);
370                 set_task_pgrp(curr, nr);
371         }
372 }
373
374 static void set_special_pids(struct pid *pid)
375 {
376         write_lock_irq(&tasklist_lock);
377         __set_special_pids(pid);
378         write_unlock_irq(&tasklist_lock);
379 }
380
381 /*
382  * Let kernel threads use this to say that they
383  * allow a certain signal (since daemonize() will
384  * have disabled all of them by default).
385  */
386 int allow_signal(int sig)
387 {
388         if (!valid_signal(sig) || sig < 1)
389                 return -EINVAL;
390
391         spin_lock_irq(&current->sighand->siglock);
392         sigdelset(&current->blocked, sig);
393         if (!current->mm) {
394                 /* Kernel threads handle their own signals.
395                    Let the signal code know it'll be handled, so
396                    that they don't get converted to SIGKILL or
397                    just silently dropped */
398                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
399         }
400         recalc_sigpending();
401         spin_unlock_irq(&current->sighand->siglock);
402         return 0;
403 }
404
405 EXPORT_SYMBOL(allow_signal);
406
407 int disallow_signal(int sig)
408 {
409         if (!valid_signal(sig) || sig < 1)
410                 return -EINVAL;
411
412         spin_lock_irq(&current->sighand->siglock);
413         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
414         recalc_sigpending();
415         spin_unlock_irq(&current->sighand->siglock);
416         return 0;
417 }
418
419 EXPORT_SYMBOL(disallow_signal);
420
421 /*
422  *      Put all the gunge required to become a kernel thread without
423  *      attached user resources in one place where it belongs.
424  */
425
426 void daemonize(const char *name, ...)
427 {
428         va_list args;
429         struct fs_struct *fs;
430         sigset_t blocked;
431
432         va_start(args, name);
433         vsnprintf(current->comm, sizeof(current->comm), name, args);
434         va_end(args);
435
436         /*
437          * If we were started as result of loading a module, close all of the
438          * user space pages.  We don't need them, and if we didn't close them
439          * they would be locked into memory.
440          */
441         exit_mm(current);
442         /*
443          * We don't want to have TIF_FREEZE set if the system-wide hibernation
444          * or suspend transition begins right now.
445          */
446         current->flags |= (PF_NOFREEZE | PF_KTHREAD);
447
448         if (current->nsproxy != &init_nsproxy) {
449                 get_nsproxy(&init_nsproxy);
450                 switch_task_namespaces(current, &init_nsproxy);
451         }
452         set_special_pids(&init_struct_pid);
453         proc_clear_tty(current);
454
455         /* Block and flush all signals */
456         sigfillset(&blocked);
457         sigprocmask(SIG_BLOCK, &blocked, NULL);
458         flush_signals(current);
459
460         /* Become as one with the init task */
461
462         exit_fs(current);       /* current->fs->count--; */
463         fs = init_task.fs;
464         current->fs = fs;
465         atomic_inc(&fs->count);
466
467         exit_files(current);
468         current->files = init_task.files;
469         atomic_inc(&current->files->count);
470
471         reparent_to_kthreadd();
472 }
473
474 EXPORT_SYMBOL(daemonize);
475
476 static void close_files(struct files_struct * files)
477 {
478         int i, j;
479         struct fdtable *fdt;
480
481         j = 0;
482
483         /*
484          * It is safe to dereference the fd table without RCU or
485          * ->file_lock because this is the last reference to the
486          * files structure.
487          */
488         fdt = files_fdtable(files);
489         for (;;) {
490                 unsigned long set;
491                 i = j * __NFDBITS;
492                 if (i >= fdt->max_fds)
493                         break;
494                 set = fdt->open_fds->fds_bits[j++];
495                 while (set) {
496                         if (set & 1) {
497                                 struct file * file = xchg(&fdt->fd[i], NULL);
498                                 if (file) {
499                                         filp_close(file, files);
500                                         cond_resched();
501                                 }
502                         }
503                         i++;
504                         set >>= 1;
505                 }
506         }
507 }
508
509 struct files_struct *get_files_struct(struct task_struct *task)
510 {
511         struct files_struct *files;
512
513         task_lock(task);
514         files = task->files;
515         if (files)
516                 atomic_inc(&files->count);
517         task_unlock(task);
518
519         return files;
520 }
521
522 void put_files_struct(struct files_struct *files)
523 {
524         struct fdtable *fdt;
525
526         if (atomic_dec_and_test(&files->count)) {
527                 close_files(files);
528                 /*
529                  * Free the fd and fdset arrays if we expanded them.
530                  * If the fdtable was embedded, pass files for freeing
531                  * at the end of the RCU grace period. Otherwise,
532                  * you can free files immediately.
533                  */
534                 fdt = files_fdtable(files);
535                 if (fdt != &files->fdtab)
536                         kmem_cache_free(files_cachep, files);
537                 free_fdtable(fdt);
538         }
539 }
540
541 void reset_files_struct(struct files_struct *files)
542 {
543         struct task_struct *tsk = current;
544         struct files_struct *old;
545
546         old = tsk->files;
547         task_lock(tsk);
548         tsk->files = files;
549         task_unlock(tsk);
550         put_files_struct(old);
551 }
552
553 void exit_files(struct task_struct *tsk)
554 {
555         struct files_struct * files = tsk->files;
556
557         if (files) {
558                 task_lock(tsk);
559                 tsk->files = NULL;
560                 task_unlock(tsk);
561                 put_files_struct(files);
562         }
563 }
564
565 void put_fs_struct(struct fs_struct *fs)
566 {
567         /* No need to hold fs->lock if we are killing it */
568         if (atomic_dec_and_test(&fs->count)) {
569                 path_put(&fs->root);
570                 path_put(&fs->pwd);
571                 kmem_cache_free(fs_cachep, fs);
572         }
573 }
574
575 void exit_fs(struct task_struct *tsk)
576 {
577         struct fs_struct * fs = tsk->fs;
578
579         if (fs) {
580                 task_lock(tsk);
581                 tsk->fs = NULL;
582                 task_unlock(tsk);
583                 put_fs_struct(fs);
584         }
585 }
586
587 EXPORT_SYMBOL_GPL(exit_fs);
588
589 #ifdef CONFIG_MM_OWNER
590 /*
591  * Task p is exiting and it owned mm, lets find a new owner for it
592  */
593 static inline int
594 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
595 {
596         /*
597          * If there are other users of the mm and the owner (us) is exiting
598          * we need to find a new owner to take on the responsibility.
599          */
600         if (atomic_read(&mm->mm_users) <= 1)
601                 return 0;
602         if (mm->owner != p)
603                 return 0;
604         return 1;
605 }
606
607 void mm_update_next_owner(struct mm_struct *mm)
608 {
609         struct task_struct *c, *g, *p = current;
610
611 retry:
612         if (!mm_need_new_owner(mm, p))
613                 return;
614
615         read_lock(&tasklist_lock);
616         /*
617          * Search in the children
618          */
619         list_for_each_entry(c, &p->children, sibling) {
620                 if (c->mm == mm)
621                         goto assign_new_owner;
622         }
623
624         /*
625          * Search in the siblings
626          */
627         list_for_each_entry(c, &p->parent->children, sibling) {
628                 if (c->mm == mm)
629                         goto assign_new_owner;
630         }
631
632         /*
633          * Search through everything else. We should not get
634          * here often
635          */
636         do_each_thread(g, c) {
637                 if (c->mm == mm)
638                         goto assign_new_owner;
639         } while_each_thread(g, c);
640
641         read_unlock(&tasklist_lock);
642         /*
643          * We found no owner yet mm_users > 1: this implies that we are
644          * most likely racing with swapoff (try_to_unuse()) or /proc or
645          * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
646          */
647         mm->owner = NULL;
648         return;
649
650 assign_new_owner:
651         BUG_ON(c == p);
652         get_task_struct(c);
653         /*
654          * The task_lock protects c->mm from changing.
655          * We always want mm->owner->mm == mm
656          */
657         task_lock(c);
658         /*
659          * Delay read_unlock() till we have the task_lock()
660          * to ensure that c does not slip away underneath us
661          */
662         read_unlock(&tasklist_lock);
663         if (c->mm != mm) {
664                 task_unlock(c);
665                 put_task_struct(c);
666                 goto retry;
667         }
668         mm->owner = c;
669         task_unlock(c);
670         put_task_struct(c);
671 }
672 #endif /* CONFIG_MM_OWNER */
673
674 /*
675  * Turn us into a lazy TLB process if we
676  * aren't already..
677  */
678 static void exit_mm(struct task_struct * tsk)
679 {
680         struct mm_struct *mm = tsk->mm;
681         struct core_state *core_state;
682
683         mm_release(tsk, mm);
684         if (!mm)
685                 return;
686         /*
687          * Serialize with any possible pending coredump.
688          * We must hold mmap_sem around checking core_state
689          * and clearing tsk->mm.  The core-inducing thread
690          * will increment ->nr_threads for each thread in the
691          * group with ->mm != NULL.
692          */
693         down_read(&mm->mmap_sem);
694         core_state = mm->core_state;
695         if (core_state) {
696                 struct core_thread self;
697                 up_read(&mm->mmap_sem);
698
699                 self.task = tsk;
700                 self.next = xchg(&core_state->dumper.next, &self);
701                 /*
702                  * Implies mb(), the result of xchg() must be visible
703                  * to core_state->dumper.
704                  */
705                 if (atomic_dec_and_test(&core_state->nr_threads))
706                         complete(&core_state->startup);
707
708                 for (;;) {
709                         set_task_state(tsk, TASK_UNINTERRUPTIBLE);
710                         if (!self.task) /* see coredump_finish() */
711                                 break;
712                         schedule();
713                 }
714                 __set_task_state(tsk, TASK_RUNNING);
715                 down_read(&mm->mmap_sem);
716         }
717         atomic_inc(&mm->mm_count);
718         BUG_ON(mm != tsk->active_mm);
719         /* more a memory barrier than a real lock */
720         task_lock(tsk);
721         tsk->mm = NULL;
722         up_read(&mm->mmap_sem);
723         enter_lazy_tlb(mm, current);
724         /* We don't want this task to be frozen prematurely */
725         clear_freeze_flag(tsk);
726         task_unlock(tsk);
727         mm_update_next_owner(mm);
728         mmput(mm);
729 }
730
731 /*
732  * Return nonzero if @parent's children should reap themselves.
733  *
734  * Called with write_lock_irq(&tasklist_lock) held.
735  */
736 static int ignoring_children(struct task_struct *parent)
737 {
738         int ret;
739         struct sighand_struct *psig = parent->sighand;
740         unsigned long flags;
741         spin_lock_irqsave(&psig->siglock, flags);
742         ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
743                (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
744         spin_unlock_irqrestore(&psig->siglock, flags);
745         return ret;
746 }
747
748 /*
749  * Detach all tasks we were using ptrace on.
750  * Any that need to be release_task'd are put on the @dead list.
751  *
752  * Called with write_lock(&tasklist_lock) held.
753  */
754 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
755 {
756         struct task_struct *p, *n;
757         int ign = -1;
758
759         list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
760                 __ptrace_unlink(p);
761
762                 if (p->exit_state != EXIT_ZOMBIE)
763                         continue;
764
765                 /*
766                  * If it's a zombie, our attachedness prevented normal
767                  * parent notification or self-reaping.  Do notification
768                  * now if it would have happened earlier.  If it should
769                  * reap itself, add it to the @dead list.  We can't call
770                  * release_task() here because we already hold tasklist_lock.
771                  *
772                  * If it's our own child, there is no notification to do.
773                  * But if our normal children self-reap, then this child
774                  * was prevented by ptrace and we must reap it now.
775                  */
776                 if (!task_detached(p) && thread_group_empty(p)) {
777                         if (!same_thread_group(p->real_parent, parent))
778                                 do_notify_parent(p, p->exit_signal);
779                         else {
780                                 if (ign < 0)
781                                         ign = ignoring_children(parent);
782                                 if (ign)
783                                         p->exit_signal = -1;
784                         }
785                 }
786
787                 if (task_detached(p)) {
788                         /*
789                          * Mark it as in the process of being reaped.
790                          */
791                         p->exit_state = EXIT_DEAD;
792                         list_add(&p->ptrace_entry, dead);
793                 }
794         }
795 }
796
797 /*
798  * Finish up exit-time ptrace cleanup.
799  *
800  * Called without locks.
801  */
802 static void ptrace_exit_finish(struct task_struct *parent,
803                                struct list_head *dead)
804 {
805         struct task_struct *p, *n;
806
807         BUG_ON(!list_empty(&parent->ptraced));
808
809         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
810                 list_del_init(&p->ptrace_entry);
811                 release_task(p);
812         }
813 }
814
815 static void reparent_thread(struct task_struct *p, struct task_struct *father)
816 {
817         if (p->pdeath_signal)
818                 /* We already hold the tasklist_lock here.  */
819                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
820
821         list_move_tail(&p->sibling, &p->real_parent->children);
822
823         /* If this is a threaded reparent there is no need to
824          * notify anyone anything has happened.
825          */
826         if (same_thread_group(p->real_parent, father))
827                 return;
828
829         /* We don't want people slaying init.  */
830         if (!task_detached(p))
831                 p->exit_signal = SIGCHLD;
832
833         /* If we'd notified the old parent about this child's death,
834          * also notify the new parent.
835          */
836         if (!ptrace_reparented(p) &&
837             p->exit_state == EXIT_ZOMBIE &&
838             !task_detached(p) && thread_group_empty(p))
839                 do_notify_parent(p, p->exit_signal);
840
841         kill_orphaned_pgrp(p, father);
842 }
843
844 /*
845  * When we die, we re-parent all our children.
846  * Try to give them to another thread in our thread
847  * group, and if no such member exists, give it to
848  * the child reaper process (ie "init") in our pid
849  * space.
850  */
851 static struct task_struct *find_new_reaper(struct task_struct *father)
852 {
853         struct pid_namespace *pid_ns = task_active_pid_ns(father);
854         struct task_struct *thread;
855
856         thread = father;
857         while_each_thread(father, thread) {
858                 if (thread->flags & PF_EXITING)
859                         continue;
860                 if (unlikely(pid_ns->child_reaper == father))
861                         pid_ns->child_reaper = thread;
862                 return thread;
863         }
864
865         if (unlikely(pid_ns->child_reaper == father)) {
866                 write_unlock_irq(&tasklist_lock);
867                 if (unlikely(pid_ns == &init_pid_ns))
868                         panic("Attempted to kill init!");
869
870                 zap_pid_ns_processes(pid_ns);
871                 write_lock_irq(&tasklist_lock);
872                 /*
873                  * We can not clear ->child_reaper or leave it alone.
874                  * There may by stealth EXIT_DEAD tasks on ->children,
875                  * forget_original_parent() must move them somewhere.
876                  */
877                 pid_ns->child_reaper = init_pid_ns.child_reaper;
878         }
879
880         return pid_ns->child_reaper;
881 }
882
883 static void forget_original_parent(struct task_struct *father)
884 {
885         struct task_struct *p, *n, *reaper;
886         LIST_HEAD(ptrace_dead);
887
888         write_lock_irq(&tasklist_lock);
889         reaper = find_new_reaper(father);
890         /*
891          * First clean up ptrace if we were using it.
892          */
893         ptrace_exit(father, &ptrace_dead);
894
895         list_for_each_entry_safe(p, n, &father->children, sibling) {
896                 p->real_parent = reaper;
897                 if (p->parent == father) {
898                         BUG_ON(p->ptrace);
899                         p->parent = p->real_parent;
900                 }
901                 reparent_thread(p, father);
902         }
903
904         write_unlock_irq(&tasklist_lock);
905         BUG_ON(!list_empty(&father->children));
906
907         ptrace_exit_finish(father, &ptrace_dead);
908 }
909
910 /*
911  * Send signals to all our closest relatives so that they know
912  * to properly mourn us..
913  */
914 static void exit_notify(struct task_struct *tsk, int group_dead)
915 {
916         int signal;
917         void *cookie;
918
919         /*
920          * This does two things:
921          *
922          * A.  Make init inherit all the child processes
923          * B.  Check to see if any process groups have become orphaned
924          *      as a result of our exiting, and if they have any stopped
925          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
926          */
927         forget_original_parent(tsk);
928         exit_task_namespaces(tsk);
929
930         write_lock_irq(&tasklist_lock);
931         if (group_dead)
932                 kill_orphaned_pgrp(tsk->group_leader, NULL);
933
934         /* Let father know we died
935          *
936          * Thread signals are configurable, but you aren't going to use
937          * that to send signals to arbitary processes.
938          * That stops right now.
939          *
940          * If the parent exec id doesn't match the exec id we saved
941          * when we started then we know the parent has changed security
942          * domain.
943          *
944          * If our self_exec id doesn't match our parent_exec_id then
945          * we have changed execution domain as these two values started
946          * the same after a fork.
947          */
948         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
949             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
950              tsk->self_exec_id != tsk->parent_exec_id) &&
951             !capable(CAP_KILL))
952                 tsk->exit_signal = SIGCHLD;
953
954         signal = tracehook_notify_death(tsk, &cookie, group_dead);
955         if (signal >= 0)
956                 signal = do_notify_parent(tsk, signal);
957
958         tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
959
960         /* mt-exec, de_thread() is waiting for us */
961         if (thread_group_leader(tsk) &&
962             tsk->signal->group_exit_task &&
963             tsk->signal->notify_count < 0)
964                 wake_up_process(tsk->signal->group_exit_task);
965
966         write_unlock_irq(&tasklist_lock);
967
968         tracehook_report_death(tsk, signal, cookie, group_dead);
969
970         /* If the process is dead, release it - nobody will wait for it */
971         if (signal == DEATH_REAP)
972                 release_task(tsk);
973 }
974
975 #ifdef CONFIG_DEBUG_STACK_USAGE
976 static void check_stack_usage(void)
977 {
978         static DEFINE_SPINLOCK(low_water_lock);
979         static int lowest_to_date = THREAD_SIZE;
980         unsigned long *n = end_of_stack(current);
981         unsigned long free;
982
983         while (*n == 0)
984                 n++;
985         free = (unsigned long)n - (unsigned long)end_of_stack(current);
986
987         if (free >= lowest_to_date)
988                 return;
989
990         spin_lock(&low_water_lock);
991         if (free < lowest_to_date) {
992                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
993                                 "left\n",
994                                 current->comm, free);
995                 lowest_to_date = free;
996         }
997         spin_unlock(&low_water_lock);
998 }
999 #else
1000 static inline void check_stack_usage(void) {}
1001 #endif
1002
1003 NORET_TYPE void do_exit(long code)
1004 {
1005         struct task_struct *tsk = current;
1006         int group_dead;
1007
1008         profile_task_exit(tsk);
1009
1010         WARN_ON(atomic_read(&tsk->fs_excl));
1011
1012         if (unlikely(in_interrupt()))
1013                 panic("Aiee, killing interrupt handler!");
1014         if (unlikely(!tsk->pid))
1015                 panic("Attempted to kill the idle task!");
1016
1017         tracehook_report_exit(&code);
1018
1019         /*
1020          * We're taking recursive faults here in do_exit. Safest is to just
1021          * leave this task alone and wait for reboot.
1022          */
1023         if (unlikely(tsk->flags & PF_EXITING)) {
1024                 printk(KERN_ALERT
1025                         "Fixing recursive fault but reboot is needed!\n");
1026                 /*
1027                  * We can do this unlocked here. The futex code uses
1028                  * this flag just to verify whether the pi state
1029                  * cleanup has been done or not. In the worst case it
1030                  * loops once more. We pretend that the cleanup was
1031                  * done as there is no way to return. Either the
1032                  * OWNER_DIED bit is set by now or we push the blocked
1033                  * task into the wait for ever nirwana as well.
1034                  */
1035                 tsk->flags |= PF_EXITPIDONE;
1036                 set_current_state(TASK_UNINTERRUPTIBLE);
1037                 schedule();
1038         }
1039
1040         exit_signals(tsk);  /* sets PF_EXITING */
1041         /*
1042          * tsk->flags are checked in the futex code to protect against
1043          * an exiting task cleaning up the robust pi futexes.
1044          */
1045         smp_mb();
1046         spin_unlock_wait(&tsk->pi_lock);
1047
1048         if (unlikely(in_atomic()))
1049                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1050                                 current->comm, task_pid_nr(current),
1051                                 preempt_count());
1052
1053         acct_update_integrals(tsk);
1054
1055         group_dead = atomic_dec_and_test(&tsk->signal->live);
1056         if (group_dead) {
1057                 hrtimer_cancel(&tsk->signal->real_timer);
1058                 exit_itimers(tsk->signal);
1059         }
1060         acct_collect(code, group_dead);
1061         if (group_dead)
1062                 tty_audit_exit();
1063         if (unlikely(tsk->audit_context))
1064                 audit_free(tsk);
1065
1066         tsk->exit_code = code;
1067         taskstats_exit(tsk, group_dead);
1068
1069         exit_mm(tsk);
1070
1071         if (group_dead)
1072                 acct_process();
1073         trace_sched_process_exit(tsk);
1074
1075         exit_sem(tsk);
1076         exit_files(tsk);
1077         exit_fs(tsk);
1078         check_stack_usage();
1079         exit_thread();
1080         cgroup_exit(tsk, 1);
1081
1082         if (group_dead && tsk->signal->leader)
1083                 disassociate_ctty(1);
1084
1085         module_put(task_thread_info(tsk)->exec_domain->module);
1086         if (tsk->binfmt)
1087                 module_put(tsk->binfmt->module);
1088
1089         proc_exit_connector(tsk);
1090         exit_notify(tsk, group_dead);
1091 #ifdef CONFIG_NUMA
1092         mpol_put(tsk->mempolicy);
1093         tsk->mempolicy = NULL;
1094 #endif
1095 #ifdef CONFIG_FUTEX
1096         /*
1097          * This must happen late, after the PID is not
1098          * hashed anymore:
1099          */
1100         if (unlikely(!list_empty(&tsk->pi_state_list)))
1101                 exit_pi_state_list(tsk);
1102         if (unlikely(current->pi_state_cache))
1103                 kfree(current->pi_state_cache);
1104 #endif
1105         /*
1106          * Make sure we are holding no locks:
1107          */
1108         debug_check_no_locks_held(tsk);
1109         /*
1110          * We can do this unlocked here. The futex code uses this flag
1111          * just to verify whether the pi state cleanup has been done
1112          * or not. In the worst case it loops once more.
1113          */
1114         tsk->flags |= PF_EXITPIDONE;
1115
1116         if (tsk->io_context)
1117                 exit_io_context();
1118
1119         if (tsk->splice_pipe)
1120                 __free_pipe_info(tsk->splice_pipe);
1121
1122         preempt_disable();
1123         /* causes final put_task_struct in finish_task_switch(). */
1124         tsk->state = TASK_DEAD;
1125         schedule();
1126         BUG();
1127         /* Avoid "noreturn function does return".  */
1128         for (;;)
1129                 cpu_relax();    /* For when BUG is null */
1130 }
1131
1132 EXPORT_SYMBOL_GPL(do_exit);
1133
1134 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1135 {
1136         if (comp)
1137                 complete(comp);
1138
1139         do_exit(code);
1140 }
1141
1142 EXPORT_SYMBOL(complete_and_exit);
1143
1144 asmlinkage long sys_exit(int error_code)
1145 {
1146         do_exit((error_code&0xff)<<8);
1147 }
1148
1149 /*
1150  * Take down every thread in the group.  This is called by fatal signals
1151  * as well as by sys_exit_group (below).
1152  */
1153 NORET_TYPE void
1154 do_group_exit(int exit_code)
1155 {
1156         struct signal_struct *sig = current->signal;
1157
1158         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1159
1160         if (signal_group_exit(sig))
1161                 exit_code = sig->group_exit_code;
1162         else if (!thread_group_empty(current)) {
1163                 struct sighand_struct *const sighand = current->sighand;
1164                 spin_lock_irq(&sighand->siglock);
1165                 if (signal_group_exit(sig))
1166                         /* Another thread got here before we took the lock.  */
1167                         exit_code = sig->group_exit_code;
1168                 else {
1169                         sig->group_exit_code = exit_code;
1170                         sig->flags = SIGNAL_GROUP_EXIT;
1171                         zap_other_threads(current);
1172                 }
1173                 spin_unlock_irq(&sighand->siglock);
1174         }
1175
1176         do_exit(exit_code);
1177         /* NOTREACHED */
1178 }
1179
1180 /*
1181  * this kills every thread in the thread group. Note that any externally
1182  * wait4()-ing process will get the correct exit code - even if this
1183  * thread is not the thread group leader.
1184  */
1185 asmlinkage long sys_exit_group(int error_code)
1186 {
1187         do_group_exit((error_code & 0xff) << 8);
1188         /* NOTREACHED */
1189         return 0;
1190 }
1191
1192 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1193 {
1194         struct pid *pid = NULL;
1195         if (type == PIDTYPE_PID)
1196                 pid = task->pids[type].pid;
1197         else if (type < PIDTYPE_MAX)
1198                 pid = task->group_leader->pids[type].pid;
1199         return pid;
1200 }
1201
1202 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1203                           struct task_struct *p)
1204 {
1205         int err;
1206
1207         if (type < PIDTYPE_MAX) {
1208                 if (task_pid_type(p, type) != pid)
1209                         return 0;
1210         }
1211
1212         /* Wait for all children (clone and not) if __WALL is set;
1213          * otherwise, wait for clone children *only* if __WCLONE is
1214          * set; otherwise, wait for non-clone children *only*.  (Note:
1215          * A "clone" child here is one that reports to its parent
1216          * using a signal other than SIGCHLD.) */
1217         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1218             && !(options & __WALL))
1219                 return 0;
1220
1221         err = security_task_wait(p);
1222         if (err)
1223                 return err;
1224
1225         return 1;
1226 }
1227
1228 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1229                                int why, int status,
1230                                struct siginfo __user *infop,
1231                                struct rusage __user *rusagep)
1232 {
1233         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1234
1235         put_task_struct(p);
1236         if (!retval)
1237                 retval = put_user(SIGCHLD, &infop->si_signo);
1238         if (!retval)
1239                 retval = put_user(0, &infop->si_errno);
1240         if (!retval)
1241                 retval = put_user((short)why, &infop->si_code);
1242         if (!retval)
1243                 retval = put_user(pid, &infop->si_pid);
1244         if (!retval)
1245                 retval = put_user(uid, &infop->si_uid);
1246         if (!retval)
1247                 retval = put_user(status, &infop->si_status);
1248         if (!retval)
1249                 retval = pid;
1250         return retval;
1251 }
1252
1253 /*
1254  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1255  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1256  * the lock and this task is uninteresting.  If we return nonzero, we have
1257  * released the lock and the system call should return.
1258  */
1259 static int wait_task_zombie(struct task_struct *p, int options,
1260                             struct siginfo __user *infop,
1261                             int __user *stat_addr, struct rusage __user *ru)
1262 {
1263         unsigned long state;
1264         int retval, status, traced;
1265         pid_t pid = task_pid_vnr(p);
1266         uid_t uid = __task_cred(p)->uid;
1267
1268         if (!likely(options & WEXITED))
1269                 return 0;
1270
1271         if (unlikely(options & WNOWAIT)) {
1272                 int exit_code = p->exit_code;
1273                 int why, status;
1274
1275                 get_task_struct(p);
1276                 read_unlock(&tasklist_lock);
1277                 if ((exit_code & 0x7f) == 0) {
1278                         why = CLD_EXITED;
1279                         status = exit_code >> 8;
1280                 } else {
1281                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1282                         status = exit_code & 0x7f;
1283                 }
1284                 return wait_noreap_copyout(p, pid, uid, why,
1285                                            status, infop, ru);
1286         }
1287
1288         /*
1289          * Try to move the task's state to DEAD
1290          * only one thread is allowed to do this:
1291          */
1292         state = xchg(&p->exit_state, EXIT_DEAD);
1293         if (state != EXIT_ZOMBIE) {
1294                 BUG_ON(state != EXIT_DEAD);
1295                 return 0;
1296         }
1297
1298         traced = ptrace_reparented(p);
1299
1300         if (likely(!traced)) {
1301                 struct signal_struct *psig;
1302                 struct signal_struct *sig;
1303                 struct task_cputime cputime;
1304
1305                 /*
1306                  * The resource counters for the group leader are in its
1307                  * own task_struct.  Those for dead threads in the group
1308                  * are in its signal_struct, as are those for the child
1309                  * processes it has previously reaped.  All these
1310                  * accumulate in the parent's signal_struct c* fields.
1311                  *
1312                  * We don't bother to take a lock here to protect these
1313                  * p->signal fields, because they are only touched by
1314                  * __exit_signal, which runs with tasklist_lock
1315                  * write-locked anyway, and so is excluded here.  We do
1316                  * need to protect the access to p->parent->signal fields,
1317                  * as other threads in the parent group can be right
1318                  * here reaping other children at the same time.
1319                  *
1320                  * We use thread_group_cputime() to get times for the thread
1321                  * group, which consolidates times for all threads in the
1322                  * group including the group leader.
1323                  */
1324                 thread_group_cputime(p, &cputime);
1325                 spin_lock_irq(&p->parent->sighand->siglock);
1326                 psig = p->parent->signal;
1327                 sig = p->signal;
1328                 psig->cutime =
1329                         cputime_add(psig->cutime,
1330                         cputime_add(cputime.utime,
1331                                     sig->cutime));
1332                 psig->cstime =
1333                         cputime_add(psig->cstime,
1334                         cputime_add(cputime.stime,
1335                                     sig->cstime));
1336                 psig->cgtime =
1337                         cputime_add(psig->cgtime,
1338                         cputime_add(p->gtime,
1339                         cputime_add(sig->gtime,
1340                                     sig->cgtime)));
1341                 psig->cmin_flt +=
1342                         p->min_flt + sig->min_flt + sig->cmin_flt;
1343                 psig->cmaj_flt +=
1344                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1345                 psig->cnvcsw +=
1346                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1347                 psig->cnivcsw +=
1348                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1349                 psig->cinblock +=
1350                         task_io_get_inblock(p) +
1351                         sig->inblock + sig->cinblock;
1352                 psig->coublock +=
1353                         task_io_get_oublock(p) +
1354                         sig->oublock + sig->coublock;
1355                 task_io_accounting_add(&psig->ioac, &p->ioac);
1356                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1357                 spin_unlock_irq(&p->parent->sighand->siglock);
1358         }
1359
1360         /*
1361          * Now we are sure this task is interesting, and no other
1362          * thread can reap it because we set its state to EXIT_DEAD.
1363          */
1364         read_unlock(&tasklist_lock);
1365
1366         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1367         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1368                 ? p->signal->group_exit_code : p->exit_code;
1369         if (!retval && stat_addr)
1370                 retval = put_user(status, stat_addr);
1371         if (!retval && infop)
1372                 retval = put_user(SIGCHLD, &infop->si_signo);
1373         if (!retval && infop)
1374                 retval = put_user(0, &infop->si_errno);
1375         if (!retval && infop) {
1376                 int why;
1377
1378                 if ((status & 0x7f) == 0) {
1379                         why = CLD_EXITED;
1380                         status >>= 8;
1381                 } else {
1382                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1383                         status &= 0x7f;
1384                 }
1385                 retval = put_user((short)why, &infop->si_code);
1386                 if (!retval)
1387                         retval = put_user(status, &infop->si_status);
1388         }
1389         if (!retval && infop)
1390                 retval = put_user(pid, &infop->si_pid);
1391         if (!retval && infop)
1392                 retval = put_user(uid, &infop->si_uid);
1393         if (!retval)
1394                 retval = pid;
1395
1396         if (traced) {
1397                 write_lock_irq(&tasklist_lock);
1398                 /* We dropped tasklist, ptracer could die and untrace */
1399                 ptrace_unlink(p);
1400                 /*
1401                  * If this is not a detached task, notify the parent.
1402                  * If it's still not detached after that, don't release
1403                  * it now.
1404                  */
1405                 if (!task_detached(p)) {
1406                         do_notify_parent(p, p->exit_signal);
1407                         if (!task_detached(p)) {
1408                                 p->exit_state = EXIT_ZOMBIE;
1409                                 p = NULL;
1410                         }
1411                 }
1412                 write_unlock_irq(&tasklist_lock);
1413         }
1414         if (p != NULL)
1415                 release_task(p);
1416
1417         return retval;
1418 }
1419
1420 /*
1421  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1422  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1423  * the lock and this task is uninteresting.  If we return nonzero, we have
1424  * released the lock and the system call should return.
1425  */
1426 static int wait_task_stopped(int ptrace, struct task_struct *p,
1427                              int options, struct siginfo __user *infop,
1428                              int __user *stat_addr, struct rusage __user *ru)
1429 {
1430         int retval, exit_code, why;
1431         uid_t uid = 0; /* unneeded, required by compiler */
1432         pid_t pid;
1433
1434         if (!(options & WUNTRACED))
1435                 return 0;
1436
1437         exit_code = 0;
1438         spin_lock_irq(&p->sighand->siglock);
1439
1440         if (unlikely(!task_is_stopped_or_traced(p)))
1441                 goto unlock_sig;
1442
1443         if (!ptrace && p->signal->group_stop_count > 0)
1444                 /*
1445                  * A group stop is in progress and this is the group leader.
1446                  * We won't report until all threads have stopped.
1447                  */
1448                 goto unlock_sig;
1449
1450         exit_code = p->exit_code;
1451         if (!exit_code)
1452                 goto unlock_sig;
1453
1454         if (!unlikely(options & WNOWAIT))
1455                 p->exit_code = 0;
1456
1457         /* don't need the RCU readlock here as we're holding a spinlock */
1458         uid = __task_cred(p)->uid;
1459 unlock_sig:
1460         spin_unlock_irq(&p->sighand->siglock);
1461         if (!exit_code)
1462                 return 0;
1463
1464         /*
1465          * Now we are pretty sure this task is interesting.
1466          * Make sure it doesn't get reaped out from under us while we
1467          * give up the lock and then examine it below.  We don't want to
1468          * keep holding onto the tasklist_lock while we call getrusage and
1469          * possibly take page faults for user memory.
1470          */
1471         get_task_struct(p);
1472         pid = task_pid_vnr(p);
1473         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1474         read_unlock(&tasklist_lock);
1475
1476         if (unlikely(options & WNOWAIT))
1477                 return wait_noreap_copyout(p, pid, uid,
1478                                            why, exit_code,
1479                                            infop, ru);
1480
1481         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1482         if (!retval && stat_addr)
1483                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1484         if (!retval && infop)
1485                 retval = put_user(SIGCHLD, &infop->si_signo);
1486         if (!retval && infop)
1487                 retval = put_user(0, &infop->si_errno);
1488         if (!retval && infop)
1489                 retval = put_user((short)why, &infop->si_code);
1490         if (!retval && infop)
1491                 retval = put_user(exit_code, &infop->si_status);
1492         if (!retval && infop)
1493                 retval = put_user(pid, &infop->si_pid);
1494         if (!retval && infop)
1495                 retval = put_user(uid, &infop->si_uid);
1496         if (!retval)
1497                 retval = pid;
1498         put_task_struct(p);
1499
1500         BUG_ON(!retval);
1501         return retval;
1502 }
1503
1504 /*
1505  * Handle do_wait work for one task in a live, non-stopped state.
1506  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1507  * the lock and this task is uninteresting.  If we return nonzero, we have
1508  * released the lock and the system call should return.
1509  */
1510 static int wait_task_continued(struct task_struct *p, int options,
1511                                struct siginfo __user *infop,
1512                                int __user *stat_addr, struct rusage __user *ru)
1513 {
1514         int retval;
1515         pid_t pid;
1516         uid_t uid;
1517
1518         if (!unlikely(options & WCONTINUED))
1519                 return 0;
1520
1521         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1522                 return 0;
1523
1524         spin_lock_irq(&p->sighand->siglock);
1525         /* Re-check with the lock held.  */
1526         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1527                 spin_unlock_irq(&p->sighand->siglock);
1528                 return 0;
1529         }
1530         if (!unlikely(options & WNOWAIT))
1531                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1532         uid = __task_cred(p)->uid;
1533         spin_unlock_irq(&p->sighand->siglock);
1534
1535         pid = task_pid_vnr(p);
1536         get_task_struct(p);
1537         read_unlock(&tasklist_lock);
1538
1539         if (!infop) {
1540                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1541                 put_task_struct(p);
1542                 if (!retval && stat_addr)
1543                         retval = put_user(0xffff, stat_addr);
1544                 if (!retval)
1545                         retval = pid;
1546         } else {
1547                 retval = wait_noreap_copyout(p, pid, uid,
1548                                              CLD_CONTINUED, SIGCONT,
1549                                              infop, ru);
1550                 BUG_ON(retval == 0);
1551         }
1552
1553         return retval;
1554 }
1555
1556 /*
1557  * Consider @p for a wait by @parent.
1558  *
1559  * -ECHILD should be in *@notask_error before the first call.
1560  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1561  * Returns zero if the search for a child should continue;
1562  * then *@notask_error is 0 if @p is an eligible child,
1563  * or another error from security_task_wait(), or still -ECHILD.
1564  */
1565 static int wait_consider_task(struct task_struct *parent, int ptrace,
1566                               struct task_struct *p, int *notask_error,
1567                               enum pid_type type, struct pid *pid, int options,
1568                               struct siginfo __user *infop,
1569                               int __user *stat_addr, struct rusage __user *ru)
1570 {
1571         int ret = eligible_child(type, pid, options, p);
1572         if (!ret)
1573                 return ret;
1574
1575         if (unlikely(ret < 0)) {
1576                 /*
1577                  * If we have not yet seen any eligible child,
1578                  * then let this error code replace -ECHILD.
1579                  * A permission error will give the user a clue
1580                  * to look for security policy problems, rather
1581                  * than for mysterious wait bugs.
1582                  */
1583                 if (*notask_error)
1584                         *notask_error = ret;
1585         }
1586
1587         if (likely(!ptrace) && unlikely(p->ptrace)) {
1588                 /*
1589                  * This child is hidden by ptrace.
1590                  * We aren't allowed to see it now, but eventually we will.
1591                  */
1592                 *notask_error = 0;
1593                 return 0;
1594         }
1595
1596         if (p->exit_state == EXIT_DEAD)
1597                 return 0;
1598
1599         /*
1600          * We don't reap group leaders with subthreads.
1601          */
1602         if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1603                 return wait_task_zombie(p, options, infop, stat_addr, ru);
1604
1605         /*
1606          * It's stopped or running now, so it might
1607          * later continue, exit, or stop again.
1608          */
1609         *notask_error = 0;
1610
1611         if (task_is_stopped_or_traced(p))
1612                 return wait_task_stopped(ptrace, p, options,
1613                                          infop, stat_addr, ru);
1614
1615         return wait_task_continued(p, options, infop, stat_addr, ru);
1616 }
1617
1618 /*
1619  * Do the work of do_wait() for one thread in the group, @tsk.
1620  *
1621  * -ECHILD should be in *@notask_error before the first call.
1622  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1623  * Returns zero if the search for a child should continue; then
1624  * *@notask_error is 0 if there were any eligible children,
1625  * or another error from security_task_wait(), or still -ECHILD.
1626  */
1627 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1628                           enum pid_type type, struct pid *pid, int options,
1629                           struct siginfo __user *infop, int __user *stat_addr,
1630                           struct rusage __user *ru)
1631 {
1632         struct task_struct *p;
1633
1634         list_for_each_entry(p, &tsk->children, sibling) {
1635                 /*
1636                  * Do not consider detached threads.
1637                  */
1638                 if (!task_detached(p)) {
1639                         int ret = wait_consider_task(tsk, 0, p, notask_error,
1640                                                      type, pid, options,
1641                                                      infop, stat_addr, ru);
1642                         if (ret)
1643                                 return ret;
1644                 }
1645         }
1646
1647         return 0;
1648 }
1649
1650 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1651                           enum pid_type type, struct pid *pid, int options,
1652                           struct siginfo __user *infop, int __user *stat_addr,
1653                           struct rusage __user *ru)
1654 {
1655         struct task_struct *p;
1656
1657         /*
1658          * Traditionally we see ptrace'd stopped tasks regardless of options.
1659          */
1660         options |= WUNTRACED;
1661
1662         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1663                 int ret = wait_consider_task(tsk, 1, p, notask_error,
1664                                              type, pid, options,
1665                                              infop, stat_addr, ru);
1666                 if (ret)
1667                         return ret;
1668         }
1669
1670         return 0;
1671 }
1672
1673 static long do_wait(enum pid_type type, struct pid *pid, int options,
1674                     struct siginfo __user *infop, int __user *stat_addr,
1675                     struct rusage __user *ru)
1676 {
1677         DECLARE_WAITQUEUE(wait, current);
1678         struct task_struct *tsk;
1679         int retval;
1680
1681         trace_sched_process_wait(pid);
1682
1683         add_wait_queue(&current->signal->wait_chldexit,&wait);
1684 repeat:
1685         /*
1686          * If there is nothing that can match our critiera just get out.
1687          * We will clear @retval to zero if we see any child that might later
1688          * match our criteria, even if we are not able to reap it yet.
1689          */
1690         retval = -ECHILD;
1691         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1692                 goto end;
1693
1694         current->state = TASK_INTERRUPTIBLE;
1695         read_lock(&tasklist_lock);
1696         tsk = current;
1697         do {
1698                 int tsk_result = do_wait_thread(tsk, &retval,
1699                                                 type, pid, options,
1700                                                 infop, stat_addr, ru);
1701                 if (!tsk_result)
1702                         tsk_result = ptrace_do_wait(tsk, &retval,
1703                                                     type, pid, options,
1704                                                     infop, stat_addr, ru);
1705                 if (tsk_result) {
1706                         /*
1707                          * tasklist_lock is unlocked and we have a final result.
1708                          */
1709                         retval = tsk_result;
1710                         goto end;
1711                 }
1712
1713                 if (options & __WNOTHREAD)
1714                         break;
1715                 tsk = next_thread(tsk);
1716                 BUG_ON(tsk->signal != current->signal);
1717         } while (tsk != current);
1718         read_unlock(&tasklist_lock);
1719
1720         if (!retval && !(options & WNOHANG)) {
1721                 retval = -ERESTARTSYS;
1722                 if (!signal_pending(current)) {
1723                         schedule();
1724                         goto repeat;
1725                 }
1726         }
1727
1728 end:
1729         current->state = TASK_RUNNING;
1730         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1731         if (infop) {
1732                 if (retval > 0)
1733                         retval = 0;
1734                 else {
1735                         /*
1736                          * For a WNOHANG return, clear out all the fields
1737                          * we would set so the user can easily tell the
1738                          * difference.
1739                          */
1740                         if (!retval)
1741                                 retval = put_user(0, &infop->si_signo);
1742                         if (!retval)
1743                                 retval = put_user(0, &infop->si_errno);
1744                         if (!retval)
1745                                 retval = put_user(0, &infop->si_code);
1746                         if (!retval)
1747                                 retval = put_user(0, &infop->si_pid);
1748                         if (!retval)
1749                                 retval = put_user(0, &infop->si_uid);
1750                         if (!retval)
1751                                 retval = put_user(0, &infop->si_status);
1752                 }
1753         }
1754         return retval;
1755 }
1756
1757 asmlinkage long sys_waitid(int which, pid_t upid,
1758                            struct siginfo __user *infop, int options,
1759                            struct rusage __user *ru)
1760 {
1761         struct pid *pid = NULL;
1762         enum pid_type type;
1763         long ret;
1764
1765         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1766                 return -EINVAL;
1767         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1768                 return -EINVAL;
1769
1770         switch (which) {
1771         case P_ALL:
1772                 type = PIDTYPE_MAX;
1773                 break;
1774         case P_PID:
1775                 type = PIDTYPE_PID;
1776                 if (upid <= 0)
1777                         return -EINVAL;
1778                 break;
1779         case P_PGID:
1780                 type = PIDTYPE_PGID;
1781                 if (upid <= 0)
1782                         return -EINVAL;
1783                 break;
1784         default:
1785                 return -EINVAL;
1786         }
1787
1788         if (type < PIDTYPE_MAX)
1789                 pid = find_get_pid(upid);
1790         ret = do_wait(type, pid, options, infop, NULL, ru);
1791         put_pid(pid);
1792
1793         /* avoid REGPARM breakage on x86: */
1794         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1795         return ret;
1796 }
1797
1798 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1799                           int options, struct rusage __user *ru)
1800 {
1801         struct pid *pid = NULL;
1802         enum pid_type type;
1803         long ret;
1804
1805         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1806                         __WNOTHREAD|__WCLONE|__WALL))
1807                 return -EINVAL;
1808
1809         if (upid == -1)
1810                 type = PIDTYPE_MAX;
1811         else if (upid < 0) {
1812                 type = PIDTYPE_PGID;
1813                 pid = find_get_pid(-upid);
1814         } else if (upid == 0) {
1815                 type = PIDTYPE_PGID;
1816                 pid = get_pid(task_pgrp(current));
1817         } else /* upid > 0 */ {
1818                 type = PIDTYPE_PID;
1819                 pid = find_get_pid(upid);
1820         }
1821
1822         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1823         put_pid(pid);
1824
1825         /* avoid REGPARM breakage on x86: */
1826         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1827         return ret;
1828 }
1829
1830 #ifdef __ARCH_WANT_SYS_WAITPID
1831
1832 /*
1833  * sys_waitpid() remains for compatibility. waitpid() should be
1834  * implemented by calling sys_wait4() from libc.a.
1835  */
1836 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1837 {
1838         return sys_wait4(pid, stat_addr, options, NULL);
1839 }
1840
1841 #endif