9p: eliminate callback complexity
[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/compat.h>
44 #include <linux/pipe_fs_i.h>
45 #include <linux/audit.h> /* for audit_free() */
46 #include <linux/resource.h>
47 #include <linux/blkdev.h>
48 #include <linux/task_io_accounting_ops.h>
49 #include <linux/tracehook.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 static void exit_mm(struct task_struct * tsk);
57
58 static inline int task_detached(struct task_struct *p)
59 {
60         return p->exit_signal == -1;
61 }
62
63 static void __unhash_process(struct task_struct *p)
64 {
65         nr_threads--;
66         detach_pid(p, PIDTYPE_PID);
67         if (thread_group_leader(p)) {
68                 detach_pid(p, PIDTYPE_PGID);
69                 detach_pid(p, PIDTYPE_SID);
70
71                 list_del_rcu(&p->tasks);
72                 __get_cpu_var(process_counts)--;
73         }
74         list_del_rcu(&p->thread_group);
75         list_del_init(&p->sibling);
76 }
77
78 /*
79  * This function expects the tasklist_lock write-locked.
80  */
81 static void __exit_signal(struct task_struct *tsk)
82 {
83         struct signal_struct *sig = tsk->signal;
84         struct sighand_struct *sighand;
85
86         BUG_ON(!sig);
87         BUG_ON(!atomic_read(&sig->count));
88
89         sighand = rcu_dereference(tsk->sighand);
90         spin_lock(&sighand->siglock);
91
92         posix_cpu_timers_exit(tsk);
93         if (atomic_dec_and_test(&sig->count))
94                 posix_cpu_timers_exit_group(tsk);
95         else {
96                 /*
97                  * If there is any task waiting for the group exit
98                  * then notify it:
99                  */
100                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
101                         wake_up_process(sig->group_exit_task);
102
103                 if (tsk == sig->curr_target)
104                         sig->curr_target = next_thread(tsk);
105                 /*
106                  * Accumulate here the counters for all threads but the
107                  * group leader as they die, so they can be added into
108                  * the process-wide totals when those are taken.
109                  * The group leader stays around as a zombie as long
110                  * as there are other threads.  When it gets reaped,
111                  * the exit.c code will add its counts into these totals.
112                  * We won't ever get here for the group leader, since it
113                  * will have been the last reference on the signal_struct.
114                  */
115                 sig->utime = cputime_add(sig->utime, task_utime(tsk));
116                 sig->stime = cputime_add(sig->stime, task_stime(tsk));
117                 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
118                 sig->min_flt += tsk->min_flt;
119                 sig->maj_flt += tsk->maj_flt;
120                 sig->nvcsw += tsk->nvcsw;
121                 sig->nivcsw += tsk->nivcsw;
122                 sig->inblock += task_io_get_inblock(tsk);
123                 sig->oublock += task_io_get_oublock(tsk);
124                 task_io_accounting_add(&sig->ioac, &tsk->ioac);
125                 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
126                 sig = NULL; /* Marker for below. */
127         }
128
129         __unhash_process(tsk);
130
131         /*
132          * Do this under ->siglock, we can race with another thread
133          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
134          */
135         flush_sigqueue(&tsk->pending);
136
137         tsk->signal = NULL;
138         tsk->sighand = NULL;
139         spin_unlock(&sighand->siglock);
140
141         __cleanup_sighand(sighand);
142         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
143         if (sig) {
144                 flush_sigqueue(&sig->shared_pending);
145                 taskstats_tgid_free(sig);
146                 __cleanup_signal(sig);
147         }
148 }
149
150 static void delayed_put_task_struct(struct rcu_head *rhp)
151 {
152         put_task_struct(container_of(rhp, struct task_struct, rcu));
153 }
154
155
156 void release_task(struct task_struct * p)
157 {
158         struct task_struct *leader;
159         int zap_leader;
160 repeat:
161         tracehook_prepare_release_task(p);
162         atomic_dec(&p->user->processes);
163         proc_flush_task(p);
164         write_lock_irq(&tasklist_lock);
165         tracehook_finish_release_task(p);
166         __exit_signal(p);
167
168         /*
169          * If we are the last non-leader member of the thread
170          * group, and the leader is zombie, then notify the
171          * group leader's parent process. (if it wants notification.)
172          */
173         zap_leader = 0;
174         leader = p->group_leader;
175         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
176                 BUG_ON(task_detached(leader));
177                 do_notify_parent(leader, leader->exit_signal);
178                 /*
179                  * If we were the last child thread and the leader has
180                  * exited already, and the leader's parent ignores SIGCHLD,
181                  * then we are the one who should release the leader.
182                  *
183                  * do_notify_parent() will have marked it self-reaping in
184                  * that case.
185                  */
186                 zap_leader = task_detached(leader);
187
188                 /*
189                  * This maintains the invariant that release_task()
190                  * only runs on a task in EXIT_DEAD, just for sanity.
191                  */
192                 if (zap_leader)
193                         leader->exit_state = EXIT_DEAD;
194         }
195
196         write_unlock_irq(&tasklist_lock);
197         release_thread(p);
198         call_rcu(&p->rcu, delayed_put_task_struct);
199
200         p = leader;
201         if (unlikely(zap_leader))
202                 goto repeat;
203 }
204
205 /*
206  * This checks not only the pgrp, but falls back on the pid if no
207  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
208  * without this...
209  *
210  * The caller must hold rcu lock or the tasklist lock.
211  */
212 struct pid *session_of_pgrp(struct pid *pgrp)
213 {
214         struct task_struct *p;
215         struct pid *sid = NULL;
216
217         p = pid_task(pgrp, PIDTYPE_PGID);
218         if (p == NULL)
219                 p = pid_task(pgrp, PIDTYPE_PID);
220         if (p != NULL)
221                 sid = task_session(p);
222
223         return sid;
224 }
225
226 /*
227  * Determine if a process group is "orphaned", according to the POSIX
228  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
229  * by terminal-generated stop signals.  Newly orphaned process groups are
230  * to receive a SIGHUP and a SIGCONT.
231  *
232  * "I ask you, have you ever known what it is to be an orphan?"
233  */
234 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
235 {
236         struct task_struct *p;
237
238         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
239                 if ((p == ignored_task) ||
240                     (p->exit_state && thread_group_empty(p)) ||
241                     is_global_init(p->real_parent))
242                         continue;
243
244                 if (task_pgrp(p->real_parent) != pgrp &&
245                     task_session(p->real_parent) == task_session(p))
246                         return 0;
247         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
248
249         return 1;
250 }
251
252 int is_current_pgrp_orphaned(void)
253 {
254         int retval;
255
256         read_lock(&tasklist_lock);
257         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
258         read_unlock(&tasklist_lock);
259
260         return retval;
261 }
262
263 static int has_stopped_jobs(struct pid *pgrp)
264 {
265         int retval = 0;
266         struct task_struct *p;
267
268         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
269                 if (!task_is_stopped(p))
270                         continue;
271                 retval = 1;
272                 break;
273         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
274         return retval;
275 }
276
277 /*
278  * Check to see if any process groups have become orphaned as
279  * a result of our exiting, and if they have any stopped jobs,
280  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
281  */
282 static void
283 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
284 {
285         struct pid *pgrp = task_pgrp(tsk);
286         struct task_struct *ignored_task = tsk;
287
288         if (!parent)
289                  /* exit: our father is in a different pgrp than
290                   * we are and we were the only connection outside.
291                   */
292                 parent = tsk->real_parent;
293         else
294                 /* reparent: our child is in a different pgrp than
295                  * we are, and it was the only connection outside.
296                  */
297                 ignored_task = NULL;
298
299         if (task_pgrp(parent) != pgrp &&
300             task_session(parent) == task_session(tsk) &&
301             will_become_orphaned_pgrp(pgrp, ignored_task) &&
302             has_stopped_jobs(pgrp)) {
303                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
304                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
305         }
306 }
307
308 /**
309  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
310  *
311  * If a kernel thread is launched as a result of a system call, or if
312  * it ever exits, it should generally reparent itself to kthreadd so it
313  * isn't in the way of other processes and is correctly cleaned up on exit.
314  *
315  * The various task state such as scheduling policy and priority may have
316  * been inherited from a user process, so we reset them to sane values here.
317  *
318  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
319  */
320 static void reparent_to_kthreadd(void)
321 {
322         write_lock_irq(&tasklist_lock);
323
324         ptrace_unlink(current);
325         /* Reparent to init */
326         current->real_parent = current->parent = kthreadd_task;
327         list_move_tail(&current->sibling, &current->real_parent->children);
328
329         /* Set the exit signal to SIGCHLD so we signal init on exit */
330         current->exit_signal = SIGCHLD;
331
332         if (task_nice(current) < 0)
333                 set_user_nice(current, 0);
334         /* cpus_allowed? */
335         /* rt_priority? */
336         /* signals? */
337         security_task_reparent_to_init(current);
338         memcpy(current->signal->rlim, init_task.signal->rlim,
339                sizeof(current->signal->rlim));
340         atomic_inc(&(INIT_USER->__count));
341         write_unlock_irq(&tasklist_lock);
342         switch_uid(INIT_USER);
343 }
344
345 void __set_special_pids(struct pid *pid)
346 {
347         struct task_struct *curr = current->group_leader;
348         pid_t nr = pid_nr(pid);
349
350         if (task_session(curr) != pid) {
351                 change_pid(curr, PIDTYPE_SID, pid);
352                 set_task_session(curr, nr);
353         }
354         if (task_pgrp(curr) != pid) {
355                 change_pid(curr, PIDTYPE_PGID, pid);
356                 set_task_pgrp(curr, nr);
357         }
358 }
359
360 static void set_special_pids(struct pid *pid)
361 {
362         write_lock_irq(&tasklist_lock);
363         __set_special_pids(pid);
364         write_unlock_irq(&tasklist_lock);
365 }
366
367 /*
368  * Let kernel threads use this to say that they
369  * allow a certain signal (since daemonize() will
370  * have disabled all of them by default).
371  */
372 int allow_signal(int sig)
373 {
374         if (!valid_signal(sig) || sig < 1)
375                 return -EINVAL;
376
377         spin_lock_irq(&current->sighand->siglock);
378         sigdelset(&current->blocked, sig);
379         if (!current->mm) {
380                 /* Kernel threads handle their own signals.
381                    Let the signal code know it'll be handled, so
382                    that they don't get converted to SIGKILL or
383                    just silently dropped */
384                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
385         }
386         recalc_sigpending();
387         spin_unlock_irq(&current->sighand->siglock);
388         return 0;
389 }
390
391 EXPORT_SYMBOL(allow_signal);
392
393 int disallow_signal(int sig)
394 {
395         if (!valid_signal(sig) || sig < 1)
396                 return -EINVAL;
397
398         spin_lock_irq(&current->sighand->siglock);
399         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
400         recalc_sigpending();
401         spin_unlock_irq(&current->sighand->siglock);
402         return 0;
403 }
404
405 EXPORT_SYMBOL(disallow_signal);
406
407 /*
408  *      Put all the gunge required to become a kernel thread without
409  *      attached user resources in one place where it belongs.
410  */
411
412 void daemonize(const char *name, ...)
413 {
414         va_list args;
415         struct fs_struct *fs;
416         sigset_t blocked;
417
418         va_start(args, name);
419         vsnprintf(current->comm, sizeof(current->comm), name, args);
420         va_end(args);
421
422         /*
423          * If we were started as result of loading a module, close all of the
424          * user space pages.  We don't need them, and if we didn't close them
425          * they would be locked into memory.
426          */
427         exit_mm(current);
428         /*
429          * We don't want to have TIF_FREEZE set if the system-wide hibernation
430          * or suspend transition begins right now.
431          */
432         current->flags |= (PF_NOFREEZE | PF_KTHREAD);
433
434         if (current->nsproxy != &init_nsproxy) {
435                 get_nsproxy(&init_nsproxy);
436                 switch_task_namespaces(current, &init_nsproxy);
437         }
438         set_special_pids(&init_struct_pid);
439         proc_clear_tty(current);
440
441         /* Block and flush all signals */
442         sigfillset(&blocked);
443         sigprocmask(SIG_BLOCK, &blocked, NULL);
444         flush_signals(current);
445
446         /* Become as one with the init task */
447
448         exit_fs(current);       /* current->fs->count--; */
449         fs = init_task.fs;
450         current->fs = fs;
451         atomic_inc(&fs->count);
452
453         exit_files(current);
454         current->files = init_task.files;
455         atomic_inc(&current->files->count);
456
457         reparent_to_kthreadd();
458 }
459
460 EXPORT_SYMBOL(daemonize);
461
462 static void close_files(struct files_struct * files)
463 {
464         int i, j;
465         struct fdtable *fdt;
466
467         j = 0;
468
469         /*
470          * It is safe to dereference the fd table without RCU or
471          * ->file_lock because this is the last reference to the
472          * files structure.
473          */
474         fdt = files_fdtable(files);
475         for (;;) {
476                 unsigned long set;
477                 i = j * __NFDBITS;
478                 if (i >= fdt->max_fds)
479                         break;
480                 set = fdt->open_fds->fds_bits[j++];
481                 while (set) {
482                         if (set & 1) {
483                                 struct file * file = xchg(&fdt->fd[i], NULL);
484                                 if (file) {
485                                         filp_close(file, files);
486                                         cond_resched();
487                                 }
488                         }
489                         i++;
490                         set >>= 1;
491                 }
492         }
493 }
494
495 struct files_struct *get_files_struct(struct task_struct *task)
496 {
497         struct files_struct *files;
498
499         task_lock(task);
500         files = task->files;
501         if (files)
502                 atomic_inc(&files->count);
503         task_unlock(task);
504
505         return files;
506 }
507
508 void put_files_struct(struct files_struct *files)
509 {
510         struct fdtable *fdt;
511
512         if (atomic_dec_and_test(&files->count)) {
513                 close_files(files);
514                 /*
515                  * Free the fd and fdset arrays if we expanded them.
516                  * If the fdtable was embedded, pass files for freeing
517                  * at the end of the RCU grace period. Otherwise,
518                  * you can free files immediately.
519                  */
520                 fdt = files_fdtable(files);
521                 if (fdt != &files->fdtab)
522                         kmem_cache_free(files_cachep, files);
523                 free_fdtable(fdt);
524         }
525 }
526
527 void reset_files_struct(struct files_struct *files)
528 {
529         struct task_struct *tsk = current;
530         struct files_struct *old;
531
532         old = tsk->files;
533         task_lock(tsk);
534         tsk->files = files;
535         task_unlock(tsk);
536         put_files_struct(old);
537 }
538
539 void exit_files(struct task_struct *tsk)
540 {
541         struct files_struct * files = tsk->files;
542
543         if (files) {
544                 task_lock(tsk);
545                 tsk->files = NULL;
546                 task_unlock(tsk);
547                 put_files_struct(files);
548         }
549 }
550
551 void put_fs_struct(struct fs_struct *fs)
552 {
553         /* No need to hold fs->lock if we are killing it */
554         if (atomic_dec_and_test(&fs->count)) {
555                 path_put(&fs->root);
556                 path_put(&fs->pwd);
557                 kmem_cache_free(fs_cachep, fs);
558         }
559 }
560
561 void exit_fs(struct task_struct *tsk)
562 {
563         struct fs_struct * fs = tsk->fs;
564
565         if (fs) {
566                 task_lock(tsk);
567                 tsk->fs = NULL;
568                 task_unlock(tsk);
569                 put_fs_struct(fs);
570         }
571 }
572
573 EXPORT_SYMBOL_GPL(exit_fs);
574
575 #ifdef CONFIG_MM_OWNER
576 /*
577  * Task p is exiting and it owned mm, lets find a new owner for it
578  */
579 static inline int
580 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
581 {
582         /*
583          * If there are other users of the mm and the owner (us) is exiting
584          * we need to find a new owner to take on the responsibility.
585          */
586         if (atomic_read(&mm->mm_users) <= 1)
587                 return 0;
588         if (mm->owner != p)
589                 return 0;
590         return 1;
591 }
592
593 void mm_update_next_owner(struct mm_struct *mm)
594 {
595         struct task_struct *c, *g, *p = current;
596
597 retry:
598         if (!mm_need_new_owner(mm, p))
599                 return;
600
601         read_lock(&tasklist_lock);
602         /*
603          * Search in the children
604          */
605         list_for_each_entry(c, &p->children, sibling) {
606                 if (c->mm == mm)
607                         goto assign_new_owner;
608         }
609
610         /*
611          * Search in the siblings
612          */
613         list_for_each_entry(c, &p->parent->children, sibling) {
614                 if (c->mm == mm)
615                         goto assign_new_owner;
616         }
617
618         /*
619          * Search through everything else. We should not get
620          * here often
621          */
622         do_each_thread(g, c) {
623                 if (c->mm == mm)
624                         goto assign_new_owner;
625         } while_each_thread(g, c);
626
627         read_unlock(&tasklist_lock);
628         /*
629          * We found no owner yet mm_users > 1: this implies that we are
630          * most likely racing with swapoff (try_to_unuse()) or /proc or
631          * ptrace or page migration (get_task_mm()).  Mark owner as NULL,
632          * so that subsystems can understand the callback and take action.
633          */
634         down_write(&mm->mmap_sem);
635         cgroup_mm_owner_callbacks(mm->owner, NULL);
636         mm->owner = NULL;
637         up_write(&mm->mmap_sem);
638         return;
639
640 assign_new_owner:
641         BUG_ON(c == p);
642         get_task_struct(c);
643         read_unlock(&tasklist_lock);
644         down_write(&mm->mmap_sem);
645         /*
646          * The task_lock protects c->mm from changing.
647          * We always want mm->owner->mm == mm
648          */
649         task_lock(c);
650         if (c->mm != mm) {
651                 task_unlock(c);
652                 up_write(&mm->mmap_sem);
653                 put_task_struct(c);
654                 goto retry;
655         }
656         cgroup_mm_owner_callbacks(mm->owner, c);
657         mm->owner = c;
658         task_unlock(c);
659         up_write(&mm->mmap_sem);
660         put_task_struct(c);
661 }
662 #endif /* CONFIG_MM_OWNER */
663
664 /*
665  * Turn us into a lazy TLB process if we
666  * aren't already..
667  */
668 static void exit_mm(struct task_struct * tsk)
669 {
670         struct mm_struct *mm = tsk->mm;
671         struct core_state *core_state;
672
673         mm_release(tsk, mm);
674         if (!mm)
675                 return;
676         /*
677          * Serialize with any possible pending coredump.
678          * We must hold mmap_sem around checking core_state
679          * and clearing tsk->mm.  The core-inducing thread
680          * will increment ->nr_threads for each thread in the
681          * group with ->mm != NULL.
682          */
683         down_read(&mm->mmap_sem);
684         core_state = mm->core_state;
685         if (core_state) {
686                 struct core_thread self;
687                 up_read(&mm->mmap_sem);
688
689                 self.task = tsk;
690                 self.next = xchg(&core_state->dumper.next, &self);
691                 /*
692                  * Implies mb(), the result of xchg() must be visible
693                  * to core_state->dumper.
694                  */
695                 if (atomic_dec_and_test(&core_state->nr_threads))
696                         complete(&core_state->startup);
697
698                 for (;;) {
699                         set_task_state(tsk, TASK_UNINTERRUPTIBLE);
700                         if (!self.task) /* see coredump_finish() */
701                                 break;
702                         schedule();
703                 }
704                 __set_task_state(tsk, TASK_RUNNING);
705                 down_read(&mm->mmap_sem);
706         }
707         atomic_inc(&mm->mm_count);
708         BUG_ON(mm != tsk->active_mm);
709         /* more a memory barrier than a real lock */
710         task_lock(tsk);
711         tsk->mm = NULL;
712         up_read(&mm->mmap_sem);
713         enter_lazy_tlb(mm, current);
714         /* We don't want this task to be frozen prematurely */
715         clear_freeze_flag(tsk);
716         task_unlock(tsk);
717         mm_update_next_owner(mm);
718         mmput(mm);
719 }
720
721 /*
722  * Return nonzero if @parent's children should reap themselves.
723  *
724  * Called with write_lock_irq(&tasklist_lock) held.
725  */
726 static int ignoring_children(struct task_struct *parent)
727 {
728         int ret;
729         struct sighand_struct *psig = parent->sighand;
730         unsigned long flags;
731         spin_lock_irqsave(&psig->siglock, flags);
732         ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
733                (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
734         spin_unlock_irqrestore(&psig->siglock, flags);
735         return ret;
736 }
737
738 /*
739  * Detach all tasks we were using ptrace on.
740  * Any that need to be release_task'd are put on the @dead list.
741  *
742  * Called with write_lock(&tasklist_lock) held.
743  */
744 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
745 {
746         struct task_struct *p, *n;
747         int ign = -1;
748
749         list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
750                 __ptrace_unlink(p);
751
752                 if (p->exit_state != EXIT_ZOMBIE)
753                         continue;
754
755                 /*
756                  * If it's a zombie, our attachedness prevented normal
757                  * parent notification or self-reaping.  Do notification
758                  * now if it would have happened earlier.  If it should
759                  * reap itself, add it to the @dead list.  We can't call
760                  * release_task() here because we already hold tasklist_lock.
761                  *
762                  * If it's our own child, there is no notification to do.
763                  * But if our normal children self-reap, then this child
764                  * was prevented by ptrace and we must reap it now.
765                  */
766                 if (!task_detached(p) && thread_group_empty(p)) {
767                         if (!same_thread_group(p->real_parent, parent))
768                                 do_notify_parent(p, p->exit_signal);
769                         else {
770                                 if (ign < 0)
771                                         ign = ignoring_children(parent);
772                                 if (ign)
773                                         p->exit_signal = -1;
774                         }
775                 }
776
777                 if (task_detached(p)) {
778                         /*
779                          * Mark it as in the process of being reaped.
780                          */
781                         p->exit_state = EXIT_DEAD;
782                         list_add(&p->ptrace_entry, dead);
783                 }
784         }
785 }
786
787 /*
788  * Finish up exit-time ptrace cleanup.
789  *
790  * Called without locks.
791  */
792 static void ptrace_exit_finish(struct task_struct *parent,
793                                struct list_head *dead)
794 {
795         struct task_struct *p, *n;
796
797         BUG_ON(!list_empty(&parent->ptraced));
798
799         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
800                 list_del_init(&p->ptrace_entry);
801                 release_task(p);
802         }
803 }
804
805 static void reparent_thread(struct task_struct *p, struct task_struct *father)
806 {
807         if (p->pdeath_signal)
808                 /* We already hold the tasklist_lock here.  */
809                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
810
811         list_move_tail(&p->sibling, &p->real_parent->children);
812
813         /* If this is a threaded reparent there is no need to
814          * notify anyone anything has happened.
815          */
816         if (same_thread_group(p->real_parent, father))
817                 return;
818
819         /* We don't want people slaying init.  */
820         if (!task_detached(p))
821                 p->exit_signal = SIGCHLD;
822
823         /* If we'd notified the old parent about this child's death,
824          * also notify the new parent.
825          */
826         if (!ptrace_reparented(p) &&
827             p->exit_state == EXIT_ZOMBIE &&
828             !task_detached(p) && thread_group_empty(p))
829                 do_notify_parent(p, p->exit_signal);
830
831         kill_orphaned_pgrp(p, father);
832 }
833
834 /*
835  * When we die, we re-parent all our children.
836  * Try to give them to another thread in our thread
837  * group, and if no such member exists, give it to
838  * the child reaper process (ie "init") in our pid
839  * space.
840  */
841 static struct task_struct *find_new_reaper(struct task_struct *father)
842 {
843         struct pid_namespace *pid_ns = task_active_pid_ns(father);
844         struct task_struct *thread;
845
846         thread = father;
847         while_each_thread(father, thread) {
848                 if (thread->flags & PF_EXITING)
849                         continue;
850                 if (unlikely(pid_ns->child_reaper == father))
851                         pid_ns->child_reaper = thread;
852                 return thread;
853         }
854
855         if (unlikely(pid_ns->child_reaper == father)) {
856                 write_unlock_irq(&tasklist_lock);
857                 if (unlikely(pid_ns == &init_pid_ns))
858                         panic("Attempted to kill init!");
859
860                 zap_pid_ns_processes(pid_ns);
861                 write_lock_irq(&tasklist_lock);
862                 /*
863                  * We can not clear ->child_reaper or leave it alone.
864                  * There may by stealth EXIT_DEAD tasks on ->children,
865                  * forget_original_parent() must move them somewhere.
866                  */
867                 pid_ns->child_reaper = init_pid_ns.child_reaper;
868         }
869
870         return pid_ns->child_reaper;
871 }
872
873 static void forget_original_parent(struct task_struct *father)
874 {
875         struct task_struct *p, *n, *reaper;
876         LIST_HEAD(ptrace_dead);
877
878         write_lock_irq(&tasklist_lock);
879         reaper = find_new_reaper(father);
880         /*
881          * First clean up ptrace if we were using it.
882          */
883         ptrace_exit(father, &ptrace_dead);
884
885         list_for_each_entry_safe(p, n, &father->children, sibling) {
886                 p->real_parent = reaper;
887                 if (p->parent == father) {
888                         BUG_ON(p->ptrace);
889                         p->parent = p->real_parent;
890                 }
891                 reparent_thread(p, father);
892         }
893
894         write_unlock_irq(&tasklist_lock);
895         BUG_ON(!list_empty(&father->children));
896
897         ptrace_exit_finish(father, &ptrace_dead);
898 }
899
900 /*
901  * Send signals to all our closest relatives so that they know
902  * to properly mourn us..
903  */
904 static void exit_notify(struct task_struct *tsk, int group_dead)
905 {
906         int signal;
907         void *cookie;
908
909         /*
910          * This does two things:
911          *
912          * A.  Make init inherit all the child processes
913          * B.  Check to see if any process groups have become orphaned
914          *      as a result of our exiting, and if they have any stopped
915          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
916          */
917         forget_original_parent(tsk);
918         exit_task_namespaces(tsk);
919
920         write_lock_irq(&tasklist_lock);
921         if (group_dead)
922                 kill_orphaned_pgrp(tsk->group_leader, NULL);
923
924         /* Let father know we died
925          *
926          * Thread signals are configurable, but you aren't going to use
927          * that to send signals to arbitary processes.
928          * That stops right now.
929          *
930          * If the parent exec id doesn't match the exec id we saved
931          * when we started then we know the parent has changed security
932          * domain.
933          *
934          * If our self_exec id doesn't match our parent_exec_id then
935          * we have changed execution domain as these two values started
936          * the same after a fork.
937          */
938         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
939             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
940              tsk->self_exec_id != tsk->parent_exec_id) &&
941             !capable(CAP_KILL))
942                 tsk->exit_signal = SIGCHLD;
943
944         signal = tracehook_notify_death(tsk, &cookie, group_dead);
945         if (signal >= 0)
946                 signal = do_notify_parent(tsk, signal);
947
948         tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
949
950         /* mt-exec, de_thread() is waiting for us */
951         if (thread_group_leader(tsk) &&
952             tsk->signal->group_exit_task &&
953             tsk->signal->notify_count < 0)
954                 wake_up_process(tsk->signal->group_exit_task);
955
956         write_unlock_irq(&tasklist_lock);
957
958         tracehook_report_death(tsk, signal, cookie, group_dead);
959
960         /* If the process is dead, release it - nobody will wait for it */
961         if (signal == DEATH_REAP)
962                 release_task(tsk);
963 }
964
965 #ifdef CONFIG_DEBUG_STACK_USAGE
966 static void check_stack_usage(void)
967 {
968         static DEFINE_SPINLOCK(low_water_lock);
969         static int lowest_to_date = THREAD_SIZE;
970         unsigned long *n = end_of_stack(current);
971         unsigned long free;
972
973         while (*n == 0)
974                 n++;
975         free = (unsigned long)n - (unsigned long)end_of_stack(current);
976
977         if (free >= lowest_to_date)
978                 return;
979
980         spin_lock(&low_water_lock);
981         if (free < lowest_to_date) {
982                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
983                                 "left\n",
984                                 current->comm, free);
985                 lowest_to_date = free;
986         }
987         spin_unlock(&low_water_lock);
988 }
989 #else
990 static inline void check_stack_usage(void) {}
991 #endif
992
993 NORET_TYPE void do_exit(long code)
994 {
995         struct task_struct *tsk = current;
996         int group_dead;
997
998         profile_task_exit(tsk);
999
1000         WARN_ON(atomic_read(&tsk->fs_excl));
1001
1002         if (unlikely(in_interrupt()))
1003                 panic("Aiee, killing interrupt handler!");
1004         if (unlikely(!tsk->pid))
1005                 panic("Attempted to kill the idle task!");
1006
1007         tracehook_report_exit(&code);
1008
1009         /*
1010          * We're taking recursive faults here in do_exit. Safest is to just
1011          * leave this task alone and wait for reboot.
1012          */
1013         if (unlikely(tsk->flags & PF_EXITING)) {
1014                 printk(KERN_ALERT
1015                         "Fixing recursive fault but reboot is needed!\n");
1016                 /*
1017                  * We can do this unlocked here. The futex code uses
1018                  * this flag just to verify whether the pi state
1019                  * cleanup has been done or not. In the worst case it
1020                  * loops once more. We pretend that the cleanup was
1021                  * done as there is no way to return. Either the
1022                  * OWNER_DIED bit is set by now or we push the blocked
1023                  * task into the wait for ever nirwana as well.
1024                  */
1025                 tsk->flags |= PF_EXITPIDONE;
1026                 if (tsk->io_context)
1027                         exit_io_context();
1028                 set_current_state(TASK_UNINTERRUPTIBLE);
1029                 schedule();
1030         }
1031
1032         exit_signals(tsk);  /* sets PF_EXITING */
1033         /*
1034          * tsk->flags are checked in the futex code to protect against
1035          * an exiting task cleaning up the robust pi futexes.
1036          */
1037         smp_mb();
1038         spin_unlock_wait(&tsk->pi_lock);
1039
1040         if (unlikely(in_atomic()))
1041                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1042                                 current->comm, task_pid_nr(current),
1043                                 preempt_count());
1044
1045         acct_update_integrals(tsk);
1046         if (tsk->mm) {
1047                 update_hiwater_rss(tsk->mm);
1048                 update_hiwater_vm(tsk->mm);
1049         }
1050         group_dead = atomic_dec_and_test(&tsk->signal->live);
1051         if (group_dead) {
1052                 hrtimer_cancel(&tsk->signal->real_timer);
1053                 exit_itimers(tsk->signal);
1054         }
1055         acct_collect(code, group_dead);
1056 #ifdef CONFIG_FUTEX
1057         if (unlikely(tsk->robust_list))
1058                 exit_robust_list(tsk);
1059 #ifdef CONFIG_COMPAT
1060         if (unlikely(tsk->compat_robust_list))
1061                 compat_exit_robust_list(tsk);
1062 #endif
1063 #endif
1064         if (group_dead)
1065                 tty_audit_exit();
1066         if (unlikely(tsk->audit_context))
1067                 audit_free(tsk);
1068
1069         tsk->exit_code = code;
1070         taskstats_exit(tsk, group_dead);
1071
1072         exit_mm(tsk);
1073
1074         if (group_dead)
1075                 acct_process();
1076         exit_sem(tsk);
1077         exit_files(tsk);
1078         exit_fs(tsk);
1079         check_stack_usage();
1080         exit_thread();
1081         cgroup_exit(tsk, 1);
1082         exit_keys(tsk);
1083
1084         if (group_dead && tsk->signal->leader)
1085                 disassociate_ctty(1);
1086
1087         module_put(task_thread_info(tsk)->exec_domain->module);
1088         if (tsk->binfmt)
1089                 module_put(tsk->binfmt->module);
1090
1091         proc_exit_connector(tsk);
1092         exit_notify(tsk, group_dead);
1093 #ifdef CONFIG_NUMA
1094         mpol_put(tsk->mempolicy);
1095         tsk->mempolicy = NULL;
1096 #endif
1097 #ifdef CONFIG_FUTEX
1098         /*
1099          * This must happen late, after the PID is not
1100          * hashed anymore:
1101          */
1102         if (unlikely(!list_empty(&tsk->pi_state_list)))
1103                 exit_pi_state_list(tsk);
1104         if (unlikely(current->pi_state_cache))
1105                 kfree(current->pi_state_cache);
1106 #endif
1107         /*
1108          * Make sure we are holding no locks:
1109          */
1110         debug_check_no_locks_held(tsk);
1111         /*
1112          * We can do this unlocked here. The futex code uses this flag
1113          * just to verify whether the pi state cleanup has been done
1114          * or not. In the worst case it loops once more.
1115          */
1116         tsk->flags |= PF_EXITPIDONE;
1117
1118         if (tsk->io_context)
1119                 exit_io_context();
1120
1121         if (tsk->splice_pipe)
1122                 __free_pipe_info(tsk->splice_pipe);
1123
1124         preempt_disable();
1125         /* causes final put_task_struct in finish_task_switch(). */
1126         tsk->state = TASK_DEAD;
1127
1128         schedule();
1129         BUG();
1130         /* Avoid "noreturn function does return".  */
1131         for (;;)
1132                 cpu_relax();    /* For when BUG is null */
1133 }
1134
1135 EXPORT_SYMBOL_GPL(do_exit);
1136
1137 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1138 {
1139         if (comp)
1140                 complete(comp);
1141
1142         do_exit(code);
1143 }
1144
1145 EXPORT_SYMBOL(complete_and_exit);
1146
1147 asmlinkage long sys_exit(int error_code)
1148 {
1149         do_exit((error_code&0xff)<<8);
1150 }
1151
1152 /*
1153  * Take down every thread in the group.  This is called by fatal signals
1154  * as well as by sys_exit_group (below).
1155  */
1156 NORET_TYPE void
1157 do_group_exit(int exit_code)
1158 {
1159         struct signal_struct *sig = current->signal;
1160
1161         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1162
1163         if (signal_group_exit(sig))
1164                 exit_code = sig->group_exit_code;
1165         else if (!thread_group_empty(current)) {
1166                 struct sighand_struct *const sighand = current->sighand;
1167                 spin_lock_irq(&sighand->siglock);
1168                 if (signal_group_exit(sig))
1169                         /* Another thread got here before we took the lock.  */
1170                         exit_code = sig->group_exit_code;
1171                 else {
1172                         sig->group_exit_code = exit_code;
1173                         sig->flags = SIGNAL_GROUP_EXIT;
1174                         zap_other_threads(current);
1175                 }
1176                 spin_unlock_irq(&sighand->siglock);
1177         }
1178
1179         do_exit(exit_code);
1180         /* NOTREACHED */
1181 }
1182
1183 /*
1184  * this kills every thread in the thread group. Note that any externally
1185  * wait4()-ing process will get the correct exit code - even if this
1186  * thread is not the thread group leader.
1187  */
1188 asmlinkage void sys_exit_group(int error_code)
1189 {
1190         do_group_exit((error_code & 0xff) << 8);
1191 }
1192
1193 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1194 {
1195         struct pid *pid = NULL;
1196         if (type == PIDTYPE_PID)
1197                 pid = task->pids[type].pid;
1198         else if (type < PIDTYPE_MAX)
1199                 pid = task->group_leader->pids[type].pid;
1200         return pid;
1201 }
1202
1203 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1204                           struct task_struct *p)
1205 {
1206         int err;
1207
1208         if (type < PIDTYPE_MAX) {
1209                 if (task_pid_type(p, type) != pid)
1210                         return 0;
1211         }
1212
1213         /* Wait for all children (clone and not) if __WALL is set;
1214          * otherwise, wait for clone children *only* if __WCLONE is
1215          * set; otherwise, wait for non-clone children *only*.  (Note:
1216          * A "clone" child here is one that reports to its parent
1217          * using a signal other than SIGCHLD.) */
1218         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1219             && !(options & __WALL))
1220                 return 0;
1221
1222         err = security_task_wait(p);
1223         if (err)
1224                 return err;
1225
1226         return 1;
1227 }
1228
1229 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1230                                int why, int status,
1231                                struct siginfo __user *infop,
1232                                struct rusage __user *rusagep)
1233 {
1234         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1235
1236         put_task_struct(p);
1237         if (!retval)
1238                 retval = put_user(SIGCHLD, &infop->si_signo);
1239         if (!retval)
1240                 retval = put_user(0, &infop->si_errno);
1241         if (!retval)
1242                 retval = put_user((short)why, &infop->si_code);
1243         if (!retval)
1244                 retval = put_user(pid, &infop->si_pid);
1245         if (!retval)
1246                 retval = put_user(uid, &infop->si_uid);
1247         if (!retval)
1248                 retval = put_user(status, &infop->si_status);
1249         if (!retval)
1250                 retval = pid;
1251         return retval;
1252 }
1253
1254 /*
1255  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1256  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1257  * the lock and this task is uninteresting.  If we return nonzero, we have
1258  * released the lock and the system call should return.
1259  */
1260 static int wait_task_zombie(struct task_struct *p, int options,
1261                             struct siginfo __user *infop,
1262                             int __user *stat_addr, struct rusage __user *ru)
1263 {
1264         unsigned long state;
1265         int retval, status, traced;
1266         pid_t pid = task_pid_vnr(p);
1267
1268         if (!likely(options & WEXITED))
1269                 return 0;
1270
1271         if (unlikely(options & WNOWAIT)) {
1272                 uid_t uid = p->uid;
1273                 int exit_code = p->exit_code;
1274                 int why, status;
1275
1276                 get_task_struct(p);
1277                 read_unlock(&tasklist_lock);
1278                 if ((exit_code & 0x7f) == 0) {
1279                         why = CLD_EXITED;
1280                         status = exit_code >> 8;
1281                 } else {
1282                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1283                         status = exit_code & 0x7f;
1284                 }
1285                 return wait_noreap_copyout(p, pid, uid, why,
1286                                            status, infop, ru);
1287         }
1288
1289         /*
1290          * Try to move the task's state to DEAD
1291          * only one thread is allowed to do this:
1292          */
1293         state = xchg(&p->exit_state, EXIT_DEAD);
1294         if (state != EXIT_ZOMBIE) {
1295                 BUG_ON(state != EXIT_DEAD);
1296                 return 0;
1297         }
1298
1299         traced = ptrace_reparented(p);
1300
1301         if (likely(!traced)) {
1302                 struct signal_struct *psig;
1303                 struct signal_struct *sig;
1304
1305                 /*
1306                  * The resource counters for the group leader are in its
1307                  * own task_struct.  Those for dead threads in the group
1308                  * are in its signal_struct, as are those for the child
1309                  * processes it has previously reaped.  All these
1310                  * accumulate in the parent's signal_struct c* fields.
1311                  *
1312                  * We don't bother to take a lock here to protect these
1313                  * p->signal fields, because they are only touched by
1314                  * __exit_signal, which runs with tasklist_lock
1315                  * write-locked anyway, and so is excluded here.  We do
1316                  * need to protect the access to p->parent->signal fields,
1317                  * as other threads in the parent group can be right
1318                  * here reaping other children at the same time.
1319                  */
1320                 spin_lock_irq(&p->parent->sighand->siglock);
1321                 psig = p->parent->signal;
1322                 sig = p->signal;
1323                 psig->cutime =
1324                         cputime_add(psig->cutime,
1325                         cputime_add(p->utime,
1326                         cputime_add(sig->utime,
1327                                     sig->cutime)));
1328                 psig->cstime =
1329                         cputime_add(psig->cstime,
1330                         cputime_add(p->stime,
1331                         cputime_add(sig->stime,
1332                                     sig->cstime)));
1333                 psig->cgtime =
1334                         cputime_add(psig->cgtime,
1335                         cputime_add(p->gtime,
1336                         cputime_add(sig->gtime,
1337                                     sig->cgtime)));
1338                 psig->cmin_flt +=
1339                         p->min_flt + sig->min_flt + sig->cmin_flt;
1340                 psig->cmaj_flt +=
1341                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1342                 psig->cnvcsw +=
1343                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1344                 psig->cnivcsw +=
1345                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1346                 psig->cinblock +=
1347                         task_io_get_inblock(p) +
1348                         sig->inblock + sig->cinblock;
1349                 psig->coublock +=
1350                         task_io_get_oublock(p) +
1351                         sig->oublock + sig->coublock;
1352                 task_io_accounting_add(&psig->ioac, &p->ioac);
1353                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1354                 spin_unlock_irq(&p->parent->sighand->siglock);
1355         }
1356
1357         /*
1358          * Now we are sure this task is interesting, and no other
1359          * thread can reap it because we set its state to EXIT_DEAD.
1360          */
1361         read_unlock(&tasklist_lock);
1362
1363         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1364         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1365                 ? p->signal->group_exit_code : p->exit_code;
1366         if (!retval && stat_addr)
1367                 retval = put_user(status, stat_addr);
1368         if (!retval && infop)
1369                 retval = put_user(SIGCHLD, &infop->si_signo);
1370         if (!retval && infop)
1371                 retval = put_user(0, &infop->si_errno);
1372         if (!retval && infop) {
1373                 int why;
1374
1375                 if ((status & 0x7f) == 0) {
1376                         why = CLD_EXITED;
1377                         status >>= 8;
1378                 } else {
1379                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1380                         status &= 0x7f;
1381                 }
1382                 retval = put_user((short)why, &infop->si_code);
1383                 if (!retval)
1384                         retval = put_user(status, &infop->si_status);
1385         }
1386         if (!retval && infop)
1387                 retval = put_user(pid, &infop->si_pid);
1388         if (!retval && infop)
1389                 retval = put_user(p->uid, &infop->si_uid);
1390         if (!retval)
1391                 retval = pid;
1392
1393         if (traced) {
1394                 write_lock_irq(&tasklist_lock);
1395                 /* We dropped tasklist, ptracer could die and untrace */
1396                 ptrace_unlink(p);
1397                 /*
1398                  * If this is not a detached task, notify the parent.
1399                  * If it's still not detached after that, don't release
1400                  * it now.
1401                  */
1402                 if (!task_detached(p)) {
1403                         do_notify_parent(p, p->exit_signal);
1404                         if (!task_detached(p)) {
1405                                 p->exit_state = EXIT_ZOMBIE;
1406                                 p = NULL;
1407                         }
1408                 }
1409                 write_unlock_irq(&tasklist_lock);
1410         }
1411         if (p != NULL)
1412                 release_task(p);
1413
1414         return retval;
1415 }
1416
1417 /*
1418  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1419  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1420  * the lock and this task is uninteresting.  If we return nonzero, we have
1421  * released the lock and the system call should return.
1422  */
1423 static int wait_task_stopped(int ptrace, struct task_struct *p,
1424                              int options, struct siginfo __user *infop,
1425                              int __user *stat_addr, struct rusage __user *ru)
1426 {
1427         int retval, exit_code, why;
1428         uid_t uid = 0; /* unneeded, required by compiler */
1429         pid_t pid;
1430
1431         if (!(options & WUNTRACED))
1432                 return 0;
1433
1434         exit_code = 0;
1435         spin_lock_irq(&p->sighand->siglock);
1436
1437         if (unlikely(!task_is_stopped_or_traced(p)))
1438                 goto unlock_sig;
1439
1440         if (!ptrace && p->signal->group_stop_count > 0)
1441                 /*
1442                  * A group stop is in progress and this is the group leader.
1443                  * We won't report until all threads have stopped.
1444                  */
1445                 goto unlock_sig;
1446
1447         exit_code = p->exit_code;
1448         if (!exit_code)
1449                 goto unlock_sig;
1450
1451         if (!unlikely(options & WNOWAIT))
1452                 p->exit_code = 0;
1453
1454         uid = p->uid;
1455 unlock_sig:
1456         spin_unlock_irq(&p->sighand->siglock);
1457         if (!exit_code)
1458                 return 0;
1459
1460         /*
1461          * Now we are pretty sure this task is interesting.
1462          * Make sure it doesn't get reaped out from under us while we
1463          * give up the lock and then examine it below.  We don't want to
1464          * keep holding onto the tasklist_lock while we call getrusage and
1465          * possibly take page faults for user memory.
1466          */
1467         get_task_struct(p);
1468         pid = task_pid_vnr(p);
1469         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1470         read_unlock(&tasklist_lock);
1471
1472         if (unlikely(options & WNOWAIT))
1473                 return wait_noreap_copyout(p, pid, uid,
1474                                            why, exit_code,
1475                                            infop, ru);
1476
1477         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1478         if (!retval && stat_addr)
1479                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1480         if (!retval && infop)
1481                 retval = put_user(SIGCHLD, &infop->si_signo);
1482         if (!retval && infop)
1483                 retval = put_user(0, &infop->si_errno);
1484         if (!retval && infop)
1485                 retval = put_user((short)why, &infop->si_code);
1486         if (!retval && infop)
1487                 retval = put_user(exit_code, &infop->si_status);
1488         if (!retval && infop)
1489                 retval = put_user(pid, &infop->si_pid);
1490         if (!retval && infop)
1491                 retval = put_user(uid, &infop->si_uid);
1492         if (!retval)
1493                 retval = pid;
1494         put_task_struct(p);
1495
1496         BUG_ON(!retval);
1497         return retval;
1498 }
1499
1500 /*
1501  * Handle do_wait work for one task in a live, non-stopped state.
1502  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1503  * the lock and this task is uninteresting.  If we return nonzero, we have
1504  * released the lock and the system call should return.
1505  */
1506 static int wait_task_continued(struct task_struct *p, int options,
1507                                struct siginfo __user *infop,
1508                                int __user *stat_addr, struct rusage __user *ru)
1509 {
1510         int retval;
1511         pid_t pid;
1512         uid_t uid;
1513
1514         if (!unlikely(options & WCONTINUED))
1515                 return 0;
1516
1517         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1518                 return 0;
1519
1520         spin_lock_irq(&p->sighand->siglock);
1521         /* Re-check with the lock held.  */
1522         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1523                 spin_unlock_irq(&p->sighand->siglock);
1524                 return 0;
1525         }
1526         if (!unlikely(options & WNOWAIT))
1527                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1528         spin_unlock_irq(&p->sighand->siglock);
1529
1530         pid = task_pid_vnr(p);
1531         uid = p->uid;
1532         get_task_struct(p);
1533         read_unlock(&tasklist_lock);
1534
1535         if (!infop) {
1536                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1537                 put_task_struct(p);
1538                 if (!retval && stat_addr)
1539                         retval = put_user(0xffff, stat_addr);
1540                 if (!retval)
1541                         retval = pid;
1542         } else {
1543                 retval = wait_noreap_copyout(p, pid, uid,
1544                                              CLD_CONTINUED, SIGCONT,
1545                                              infop, ru);
1546                 BUG_ON(retval == 0);
1547         }
1548
1549         return retval;
1550 }
1551
1552 /*
1553  * Consider @p for a wait by @parent.
1554  *
1555  * -ECHILD should be in *@notask_error before the first call.
1556  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1557  * Returns zero if the search for a child should continue;
1558  * then *@notask_error is 0 if @p is an eligible child,
1559  * or another error from security_task_wait(), or still -ECHILD.
1560  */
1561 static int wait_consider_task(struct task_struct *parent, int ptrace,
1562                               struct task_struct *p, int *notask_error,
1563                               enum pid_type type, struct pid *pid, int options,
1564                               struct siginfo __user *infop,
1565                               int __user *stat_addr, struct rusage __user *ru)
1566 {
1567         int ret = eligible_child(type, pid, options, p);
1568         if (!ret)
1569                 return ret;
1570
1571         if (unlikely(ret < 0)) {
1572                 /*
1573                  * If we have not yet seen any eligible child,
1574                  * then let this error code replace -ECHILD.
1575                  * A permission error will give the user a clue
1576                  * to look for security policy problems, rather
1577                  * than for mysterious wait bugs.
1578                  */
1579                 if (*notask_error)
1580                         *notask_error = ret;
1581         }
1582
1583         if (likely(!ptrace) && unlikely(p->ptrace)) {
1584                 /*
1585                  * This child is hidden by ptrace.
1586                  * We aren't allowed to see it now, but eventually we will.
1587                  */
1588                 *notask_error = 0;
1589                 return 0;
1590         }
1591
1592         if (p->exit_state == EXIT_DEAD)
1593                 return 0;
1594
1595         /*
1596          * We don't reap group leaders with subthreads.
1597          */
1598         if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1599                 return wait_task_zombie(p, options, infop, stat_addr, ru);
1600
1601         /*
1602          * It's stopped or running now, so it might
1603          * later continue, exit, or stop again.
1604          */
1605         *notask_error = 0;
1606
1607         if (task_is_stopped_or_traced(p))
1608                 return wait_task_stopped(ptrace, p, options,
1609                                          infop, stat_addr, ru);
1610
1611         return wait_task_continued(p, options, infop, stat_addr, ru);
1612 }
1613
1614 /*
1615  * Do the work of do_wait() for one thread in the group, @tsk.
1616  *
1617  * -ECHILD should be in *@notask_error before the first call.
1618  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1619  * Returns zero if the search for a child should continue; then
1620  * *@notask_error is 0 if there were any eligible children,
1621  * or another error from security_task_wait(), or still -ECHILD.
1622  */
1623 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1624                           enum pid_type type, struct pid *pid, int options,
1625                           struct siginfo __user *infop, int __user *stat_addr,
1626                           struct rusage __user *ru)
1627 {
1628         struct task_struct *p;
1629
1630         list_for_each_entry(p, &tsk->children, sibling) {
1631                 /*
1632                  * Do not consider detached threads.
1633                  */
1634                 if (!task_detached(p)) {
1635                         int ret = wait_consider_task(tsk, 0, p, notask_error,
1636                                                      type, pid, options,
1637                                                      infop, stat_addr, ru);
1638                         if (ret)
1639                                 return ret;
1640                 }
1641         }
1642
1643         return 0;
1644 }
1645
1646 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1647                           enum pid_type type, struct pid *pid, int options,
1648                           struct siginfo __user *infop, int __user *stat_addr,
1649                           struct rusage __user *ru)
1650 {
1651         struct task_struct *p;
1652
1653         /*
1654          * Traditionally we see ptrace'd stopped tasks regardless of options.
1655          */
1656         options |= WUNTRACED;
1657
1658         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1659                 int ret = wait_consider_task(tsk, 1, p, notask_error,
1660                                              type, pid, options,
1661                                              infop, stat_addr, ru);
1662                 if (ret)
1663                         return ret;
1664         }
1665
1666         return 0;
1667 }
1668
1669 static long do_wait(enum pid_type type, struct pid *pid, int options,
1670                     struct siginfo __user *infop, int __user *stat_addr,
1671                     struct rusage __user *ru)
1672 {
1673         DECLARE_WAITQUEUE(wait, current);
1674         struct task_struct *tsk;
1675         int retval;
1676
1677         add_wait_queue(&current->signal->wait_chldexit,&wait);
1678 repeat:
1679         /*
1680          * If there is nothing that can match our critiera just get out.
1681          * We will clear @retval to zero if we see any child that might later
1682          * match our criteria, even if we are not able to reap it yet.
1683          */
1684         retval = -ECHILD;
1685         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1686                 goto end;
1687
1688         current->state = TASK_INTERRUPTIBLE;
1689         read_lock(&tasklist_lock);
1690         tsk = current;
1691         do {
1692                 int tsk_result = do_wait_thread(tsk, &retval,
1693                                                 type, pid, options,
1694                                                 infop, stat_addr, ru);
1695                 if (!tsk_result)
1696                         tsk_result = ptrace_do_wait(tsk, &retval,
1697                                                     type, pid, options,
1698                                                     infop, stat_addr, ru);
1699                 if (tsk_result) {
1700                         /*
1701                          * tasklist_lock is unlocked and we have a final result.
1702                          */
1703                         retval = tsk_result;
1704                         goto end;
1705                 }
1706
1707                 if (options & __WNOTHREAD)
1708                         break;
1709                 tsk = next_thread(tsk);
1710                 BUG_ON(tsk->signal != current->signal);
1711         } while (tsk != current);
1712         read_unlock(&tasklist_lock);
1713
1714         if (!retval && !(options & WNOHANG)) {
1715                 retval = -ERESTARTSYS;
1716                 if (!signal_pending(current)) {
1717                         schedule();
1718                         goto repeat;
1719                 }
1720         }
1721
1722 end:
1723         current->state = TASK_RUNNING;
1724         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1725         if (infop) {
1726                 if (retval > 0)
1727                         retval = 0;
1728                 else {
1729                         /*
1730                          * For a WNOHANG return, clear out all the fields
1731                          * we would set so the user can easily tell the
1732                          * difference.
1733                          */
1734                         if (!retval)
1735                                 retval = put_user(0, &infop->si_signo);
1736                         if (!retval)
1737                                 retval = put_user(0, &infop->si_errno);
1738                         if (!retval)
1739                                 retval = put_user(0, &infop->si_code);
1740                         if (!retval)
1741                                 retval = put_user(0, &infop->si_pid);
1742                         if (!retval)
1743                                 retval = put_user(0, &infop->si_uid);
1744                         if (!retval)
1745                                 retval = put_user(0, &infop->si_status);
1746                 }
1747         }
1748         return retval;
1749 }
1750
1751 asmlinkage long sys_waitid(int which, pid_t upid,
1752                            struct siginfo __user *infop, int options,
1753                            struct rusage __user *ru)
1754 {
1755         struct pid *pid = NULL;
1756         enum pid_type type;
1757         long ret;
1758
1759         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1760                 return -EINVAL;
1761         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1762                 return -EINVAL;
1763
1764         switch (which) {
1765         case P_ALL:
1766                 type = PIDTYPE_MAX;
1767                 break;
1768         case P_PID:
1769                 type = PIDTYPE_PID;
1770                 if (upid <= 0)
1771                         return -EINVAL;
1772                 break;
1773         case P_PGID:
1774                 type = PIDTYPE_PGID;
1775                 if (upid <= 0)
1776                         return -EINVAL;
1777                 break;
1778         default:
1779                 return -EINVAL;
1780         }
1781
1782         if (type < PIDTYPE_MAX)
1783                 pid = find_get_pid(upid);
1784         ret = do_wait(type, pid, options, infop, NULL, ru);
1785         put_pid(pid);
1786
1787         /* avoid REGPARM breakage on x86: */
1788         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1789         return ret;
1790 }
1791
1792 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1793                           int options, struct rusage __user *ru)
1794 {
1795         struct pid *pid = NULL;
1796         enum pid_type type;
1797         long ret;
1798
1799         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1800                         __WNOTHREAD|__WCLONE|__WALL))
1801                 return -EINVAL;
1802
1803         if (upid == -1)
1804                 type = PIDTYPE_MAX;
1805         else if (upid < 0) {
1806                 type = PIDTYPE_PGID;
1807                 pid = find_get_pid(-upid);
1808         } else if (upid == 0) {
1809                 type = PIDTYPE_PGID;
1810                 pid = get_pid(task_pgrp(current));
1811         } else /* upid > 0 */ {
1812                 type = PIDTYPE_PID;
1813                 pid = find_get_pid(upid);
1814         }
1815
1816         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1817         put_pid(pid);
1818
1819         /* avoid REGPARM breakage on x86: */
1820         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1821         return ret;
1822 }
1823
1824 #ifdef __ARCH_WANT_SYS_WAITPID
1825
1826 /*
1827  * sys_waitpid() remains for compatibility. waitpid() should be
1828  * implemented by calling sys_wait4() from libc.a.
1829  */
1830 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1831 {
1832         return sys_wait4(pid, stat_addr, options, NULL);
1833 }
1834
1835 #endif