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