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