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