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