c8d0485578be00a4d46b3b6823f161d0018b3778
[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         /*
644          * The task_lock protects c->mm from changing.
645          * We always want mm->owner->mm == mm
646          */
647         task_lock(c);
648         /*
649          * Delay read_unlock() till we have the task_lock()
650          * to ensure that c does not slip away underneath us
651          */
652         read_unlock(&tasklist_lock);
653         if (c->mm != mm) {
654                 task_unlock(c);
655                 put_task_struct(c);
656                 goto retry;
657         }
658         cgroup_mm_owner_callbacks(mm->owner, c);
659         mm->owner = c;
660         task_unlock(c);
661         put_task_struct(c);
662 }
663 #endif /* CONFIG_MM_OWNER */
664
665 /*
666  * Turn us into a lazy TLB process if we
667  * aren't already..
668  */
669 static void exit_mm(struct task_struct * tsk)
670 {
671         struct mm_struct *mm = tsk->mm;
672         struct core_state *core_state;
673
674         mm_release(tsk, mm);
675         if (!mm)
676                 return;
677         /*
678          * Serialize with any possible pending coredump.
679          * We must hold mmap_sem around checking core_state
680          * and clearing tsk->mm.  The core-inducing thread
681          * will increment ->nr_threads for each thread in the
682          * group with ->mm != NULL.
683          */
684         down_read(&mm->mmap_sem);
685         core_state = mm->core_state;
686         if (core_state) {
687                 struct core_thread self;
688                 up_read(&mm->mmap_sem);
689
690                 self.task = tsk;
691                 self.next = xchg(&core_state->dumper.next, &self);
692                 /*
693                  * Implies mb(), the result of xchg() must be visible
694                  * to core_state->dumper.
695                  */
696                 if (atomic_dec_and_test(&core_state->nr_threads))
697                         complete(&core_state->startup);
698
699                 for (;;) {
700                         set_task_state(tsk, TASK_UNINTERRUPTIBLE);
701                         if (!self.task) /* see coredump_finish() */
702                                 break;
703                         schedule();
704                 }
705                 __set_task_state(tsk, TASK_RUNNING);
706                 down_read(&mm->mmap_sem);
707         }
708         atomic_inc(&mm->mm_count);
709         BUG_ON(mm != tsk->active_mm);
710         /* more a memory barrier than a real lock */
711         task_lock(tsk);
712         tsk->mm = NULL;
713         up_read(&mm->mmap_sem);
714         enter_lazy_tlb(mm, current);
715         /* We don't want this task to be frozen prematurely */
716         clear_freeze_flag(tsk);
717         task_unlock(tsk);
718         mm_update_next_owner(mm);
719         mmput(mm);
720 }
721
722 /*
723  * Return nonzero if @parent's children should reap themselves.
724  *
725  * Called with write_lock_irq(&tasklist_lock) held.
726  */
727 static int ignoring_children(struct task_struct *parent)
728 {
729         int ret;
730         struct sighand_struct *psig = parent->sighand;
731         unsigned long flags;
732         spin_lock_irqsave(&psig->siglock, flags);
733         ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
734                (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
735         spin_unlock_irqrestore(&psig->siglock, flags);
736         return ret;
737 }
738
739 /*
740  * Detach all tasks we were using ptrace on.
741  * Any that need to be release_task'd are put on the @dead list.
742  *
743  * Called with write_lock(&tasklist_lock) held.
744  */
745 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
746 {
747         struct task_struct *p, *n;
748         int ign = -1;
749
750         list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
751                 __ptrace_unlink(p);
752
753                 if (p->exit_state != EXIT_ZOMBIE)
754                         continue;
755
756                 /*
757                  * If it's a zombie, our attachedness prevented normal
758                  * parent notification or self-reaping.  Do notification
759                  * now if it would have happened earlier.  If it should
760                  * reap itself, add it to the @dead list.  We can't call
761                  * release_task() here because we already hold tasklist_lock.
762                  *
763                  * If it's our own child, there is no notification to do.
764                  * But if our normal children self-reap, then this child
765                  * was prevented by ptrace and we must reap it now.
766                  */
767                 if (!task_detached(p) && thread_group_empty(p)) {
768                         if (!same_thread_group(p->real_parent, parent))
769                                 do_notify_parent(p, p->exit_signal);
770                         else {
771                                 if (ign < 0)
772                                         ign = ignoring_children(parent);
773                                 if (ign)
774                                         p->exit_signal = -1;
775                         }
776                 }
777
778                 if (task_detached(p)) {
779                         /*
780                          * Mark it as in the process of being reaped.
781                          */
782                         p->exit_state = EXIT_DEAD;
783                         list_add(&p->ptrace_entry, dead);
784                 }
785         }
786 }
787
788 /*
789  * Finish up exit-time ptrace cleanup.
790  *
791  * Called without locks.
792  */
793 static void ptrace_exit_finish(struct task_struct *parent,
794                                struct list_head *dead)
795 {
796         struct task_struct *p, *n;
797
798         BUG_ON(!list_empty(&parent->ptraced));
799
800         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
801                 list_del_init(&p->ptrace_entry);
802                 release_task(p);
803         }
804 }
805
806 static void reparent_thread(struct task_struct *p, struct task_struct *father)
807 {
808         if (p->pdeath_signal)
809                 /* We already hold the tasklist_lock here.  */
810                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
811
812         list_move_tail(&p->sibling, &p->real_parent->children);
813
814         /* If this is a threaded reparent there is no need to
815          * notify anyone anything has happened.
816          */
817         if (same_thread_group(p->real_parent, father))
818                 return;
819
820         /* We don't want people slaying init.  */
821         if (!task_detached(p))
822                 p->exit_signal = SIGCHLD;
823
824         /* If we'd notified the old parent about this child's death,
825          * also notify the new parent.
826          */
827         if (!ptrace_reparented(p) &&
828             p->exit_state == EXIT_ZOMBIE &&
829             !task_detached(p) && thread_group_empty(p))
830                 do_notify_parent(p, p->exit_signal);
831
832         kill_orphaned_pgrp(p, father);
833 }
834
835 /*
836  * When we die, we re-parent all our children.
837  * Try to give them to another thread in our thread
838  * group, and if no such member exists, give it to
839  * the child reaper process (ie "init") in our pid
840  * space.
841  */
842 static struct task_struct *find_new_reaper(struct task_struct *father)
843 {
844         struct pid_namespace *pid_ns = task_active_pid_ns(father);
845         struct task_struct *thread;
846
847         thread = father;
848         while_each_thread(father, thread) {
849                 if (thread->flags & PF_EXITING)
850                         continue;
851                 if (unlikely(pid_ns->child_reaper == father))
852                         pid_ns->child_reaper = thread;
853                 return thread;
854         }
855
856         if (unlikely(pid_ns->child_reaper == father)) {
857                 write_unlock_irq(&tasklist_lock);
858                 if (unlikely(pid_ns == &init_pid_ns))
859                         panic("Attempted to kill init!");
860
861                 zap_pid_ns_processes(pid_ns);
862                 write_lock_irq(&tasklist_lock);
863                 /*
864                  * We can not clear ->child_reaper or leave it alone.
865                  * There may by stealth EXIT_DEAD tasks on ->children,
866                  * forget_original_parent() must move them somewhere.
867                  */
868                 pid_ns->child_reaper = init_pid_ns.child_reaper;
869         }
870
871         return pid_ns->child_reaper;
872 }
873
874 static void forget_original_parent(struct task_struct *father)
875 {
876         struct task_struct *p, *n, *reaper;
877         LIST_HEAD(ptrace_dead);
878
879         write_lock_irq(&tasklist_lock);
880         reaper = find_new_reaper(father);
881         /*
882          * First clean up ptrace if we were using it.
883          */
884         ptrace_exit(father, &ptrace_dead);
885
886         list_for_each_entry_safe(p, n, &father->children, sibling) {
887                 p->real_parent = reaper;
888                 if (p->parent == father) {
889                         BUG_ON(p->ptrace);
890                         p->parent = p->real_parent;
891                 }
892                 reparent_thread(p, father);
893         }
894
895         write_unlock_irq(&tasklist_lock);
896         BUG_ON(!list_empty(&father->children));
897
898         ptrace_exit_finish(father, &ptrace_dead);
899 }
900
901 /*
902  * Send signals to all our closest relatives so that they know
903  * to properly mourn us..
904  */
905 static void exit_notify(struct task_struct *tsk, int group_dead)
906 {
907         int signal;
908         void *cookie;
909
910         /*
911          * This does two things:
912          *
913          * A.  Make init inherit all the child processes
914          * B.  Check to see if any process groups have become orphaned
915          *      as a result of our exiting, and if they have any stopped
916          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
917          */
918         forget_original_parent(tsk);
919         exit_task_namespaces(tsk);
920
921         write_lock_irq(&tasklist_lock);
922         if (group_dead)
923                 kill_orphaned_pgrp(tsk->group_leader, NULL);
924
925         /* Let father know we died
926          *
927          * Thread signals are configurable, but you aren't going to use
928          * that to send signals to arbitary processes.
929          * That stops right now.
930          *
931          * If the parent exec id doesn't match the exec id we saved
932          * when we started then we know the parent has changed security
933          * domain.
934          *
935          * If our self_exec id doesn't match our parent_exec_id then
936          * we have changed execution domain as these two values started
937          * the same after a fork.
938          */
939         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
940             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
941              tsk->self_exec_id != tsk->parent_exec_id) &&
942             !capable(CAP_KILL))
943                 tsk->exit_signal = SIGCHLD;
944
945         signal = tracehook_notify_death(tsk, &cookie, group_dead);
946         if (signal >= 0)
947                 signal = do_notify_parent(tsk, signal);
948
949         tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
950
951         /* mt-exec, de_thread() is waiting for us */
952         if (thread_group_leader(tsk) &&
953             tsk->signal->group_exit_task &&
954             tsk->signal->notify_count < 0)
955                 wake_up_process(tsk->signal->group_exit_task);
956
957         write_unlock_irq(&tasklist_lock);
958
959         tracehook_report_death(tsk, signal, cookie, group_dead);
960
961         /* If the process is dead, release it - nobody will wait for it */
962         if (signal == DEATH_REAP)
963                 release_task(tsk);
964 }
965
966 #ifdef CONFIG_DEBUG_STACK_USAGE
967 static void check_stack_usage(void)
968 {
969         static DEFINE_SPINLOCK(low_water_lock);
970         static int lowest_to_date = THREAD_SIZE;
971         unsigned long free;
972
973         free = stack_not_used(current);
974
975         if (free >= lowest_to_date)
976                 return;
977
978         spin_lock(&low_water_lock);
979         if (free < lowest_to_date) {
980                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
981                                 "left\n",
982                                 current->comm, free);
983                 lowest_to_date = free;
984         }
985         spin_unlock(&low_water_lock);
986 }
987 #else
988 static inline void check_stack_usage(void) {}
989 #endif
990
991 NORET_TYPE void do_exit(long code)
992 {
993         struct task_struct *tsk = current;
994         int group_dead;
995
996         profile_task_exit(tsk);
997
998         WARN_ON(atomic_read(&tsk->fs_excl));
999
1000         if (unlikely(in_interrupt()))
1001                 panic("Aiee, killing interrupt handler!");
1002         if (unlikely(!tsk->pid))
1003                 panic("Attempted to kill the idle task!");
1004
1005         tracehook_report_exit(&code);
1006
1007         /*
1008          * We're taking recursive faults here in do_exit. Safest is to just
1009          * leave this task alone and wait for reboot.
1010          */
1011         if (unlikely(tsk->flags & PF_EXITING)) {
1012                 printk(KERN_ALERT
1013                         "Fixing recursive fault but reboot is needed!\n");
1014                 /*
1015                  * We can do this unlocked here. The futex code uses
1016                  * this flag just to verify whether the pi state
1017                  * cleanup has been done or not. In the worst case it
1018                  * loops once more. We pretend that the cleanup was
1019                  * done as there is no way to return. Either the
1020                  * OWNER_DIED bit is set by now or we push the blocked
1021                  * task into the wait for ever nirwana as well.
1022                  */
1023                 tsk->flags |= PF_EXITPIDONE;
1024                 if (tsk->io_context)
1025                         exit_io_context();
1026                 set_current_state(TASK_UNINTERRUPTIBLE);
1027                 schedule();
1028         }
1029
1030         exit_signals(tsk);  /* sets PF_EXITING */
1031         /*
1032          * tsk->flags are checked in the futex code to protect against
1033          * an exiting task cleaning up the robust pi futexes.
1034          */
1035         smp_mb();
1036         spin_unlock_wait(&tsk->pi_lock);
1037
1038         if (unlikely(in_atomic()))
1039                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1040                                 current->comm, task_pid_nr(current),
1041                                 preempt_count());
1042
1043         acct_update_integrals(tsk);
1044         if (tsk->mm) {
1045                 update_hiwater_rss(tsk->mm);
1046                 update_hiwater_vm(tsk->mm);
1047         }
1048         group_dead = atomic_dec_and_test(&tsk->signal->live);
1049         if (group_dead) {
1050                 hrtimer_cancel(&tsk->signal->real_timer);
1051                 exit_itimers(tsk->signal);
1052         }
1053         acct_collect(code, group_dead);
1054 #ifdef CONFIG_FUTEX
1055         if (unlikely(tsk->robust_list))
1056                 exit_robust_list(tsk);
1057 #ifdef CONFIG_COMPAT
1058         if (unlikely(tsk->compat_robust_list))
1059                 compat_exit_robust_list(tsk);
1060 #endif
1061 #endif
1062         if (group_dead)
1063                 tty_audit_exit();
1064         if (unlikely(tsk->audit_context))
1065                 audit_free(tsk);
1066
1067         tsk->exit_code = code;
1068         taskstats_exit(tsk, group_dead);
1069
1070         exit_mm(tsk);
1071
1072         if (group_dead)
1073                 acct_process();
1074         exit_sem(tsk);
1075         exit_files(tsk);
1076         exit_fs(tsk);
1077         check_stack_usage();
1078         exit_thread();
1079         cgroup_exit(tsk, 1);
1080         exit_keys(tsk);
1081
1082         if (group_dead && tsk->signal->leader)
1083                 disassociate_ctty(1);
1084
1085         module_put(task_thread_info(tsk)->exec_domain->module);
1086         if (tsk->binfmt)
1087                 module_put(tsk->binfmt->module);
1088
1089         proc_exit_connector(tsk);
1090         exit_notify(tsk, group_dead);
1091 #ifdef CONFIG_NUMA
1092         mpol_put(tsk->mempolicy);
1093         tsk->mempolicy = NULL;
1094 #endif
1095 #ifdef CONFIG_FUTEX
1096         /*
1097          * This must happen late, after the PID is not
1098          * hashed anymore:
1099          */
1100         if (unlikely(!list_empty(&tsk->pi_state_list)))
1101                 exit_pi_state_list(tsk);
1102         if (unlikely(current->pi_state_cache))
1103                 kfree(current->pi_state_cache);
1104 #endif
1105         /*
1106          * Make sure we are holding no locks:
1107          */
1108         debug_check_no_locks_held(tsk);
1109         /*
1110          * We can do this unlocked here. The futex code uses this flag
1111          * just to verify whether the pi state cleanup has been done
1112          * or not. In the worst case it loops once more.
1113          */
1114         tsk->flags |= PF_EXITPIDONE;
1115
1116         if (tsk->io_context)
1117                 exit_io_context();
1118
1119         if (tsk->splice_pipe)
1120                 __free_pipe_info(tsk->splice_pipe);
1121
1122         preempt_disable();
1123         /* causes final put_task_struct in finish_task_switch(). */
1124         tsk->state = TASK_DEAD;
1125
1126         schedule();
1127         BUG();
1128         /* Avoid "noreturn function does return".  */
1129         for (;;)
1130                 cpu_relax();    /* For when BUG is null */
1131 }
1132
1133 EXPORT_SYMBOL_GPL(do_exit);
1134
1135 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1136 {
1137         if (comp)
1138                 complete(comp);
1139
1140         do_exit(code);
1141 }
1142
1143 EXPORT_SYMBOL(complete_and_exit);
1144
1145 asmlinkage long sys_exit(int error_code)
1146 {
1147         do_exit((error_code&0xff)<<8);
1148 }
1149
1150 /*
1151  * Take down every thread in the group.  This is called by fatal signals
1152  * as well as by sys_exit_group (below).
1153  */
1154 NORET_TYPE void
1155 do_group_exit(int exit_code)
1156 {
1157         struct signal_struct *sig = current->signal;
1158
1159         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1160
1161         if (signal_group_exit(sig))
1162                 exit_code = sig->group_exit_code;
1163         else if (!thread_group_empty(current)) {
1164                 struct sighand_struct *const sighand = current->sighand;
1165                 spin_lock_irq(&sighand->siglock);
1166                 if (signal_group_exit(sig))
1167                         /* Another thread got here before we took the lock.  */
1168                         exit_code = sig->group_exit_code;
1169                 else {
1170                         sig->group_exit_code = exit_code;
1171                         sig->flags = SIGNAL_GROUP_EXIT;
1172                         zap_other_threads(current);
1173                 }
1174                 spin_unlock_irq(&sighand->siglock);
1175         }
1176
1177         do_exit(exit_code);
1178         /* NOTREACHED */
1179 }
1180
1181 /*
1182  * this kills every thread in the thread group. Note that any externally
1183  * wait4()-ing process will get the correct exit code - even if this
1184  * thread is not the thread group leader.
1185  */
1186 asmlinkage void sys_exit_group(int error_code)
1187 {
1188         do_group_exit((error_code & 0xff) << 8);
1189 }
1190
1191 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1192 {
1193         struct pid *pid = NULL;
1194         if (type == PIDTYPE_PID)
1195                 pid = task->pids[type].pid;
1196         else if (type < PIDTYPE_MAX)
1197                 pid = task->group_leader->pids[type].pid;
1198         return pid;
1199 }
1200
1201 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1202                           struct task_struct *p)
1203 {
1204         int err;
1205
1206         if (type < PIDTYPE_MAX) {
1207                 if (task_pid_type(p, type) != pid)
1208                         return 0;
1209         }
1210
1211         /* Wait for all children (clone and not) if __WALL is set;
1212          * otherwise, wait for clone children *only* if __WCLONE is
1213          * set; otherwise, wait for non-clone children *only*.  (Note:
1214          * A "clone" child here is one that reports to its parent
1215          * using a signal other than SIGCHLD.) */
1216         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1217             && !(options & __WALL))
1218                 return 0;
1219
1220         err = security_task_wait(p);
1221         if (err)
1222                 return err;
1223
1224         return 1;
1225 }
1226
1227 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1228                                int why, int status,
1229                                struct siginfo __user *infop,
1230                                struct rusage __user *rusagep)
1231 {
1232         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1233
1234         put_task_struct(p);
1235         if (!retval)
1236                 retval = put_user(SIGCHLD, &infop->si_signo);
1237         if (!retval)
1238                 retval = put_user(0, &infop->si_errno);
1239         if (!retval)
1240                 retval = put_user((short)why, &infop->si_code);
1241         if (!retval)
1242                 retval = put_user(pid, &infop->si_pid);
1243         if (!retval)
1244                 retval = put_user(uid, &infop->si_uid);
1245         if (!retval)
1246                 retval = put_user(status, &infop->si_status);
1247         if (!retval)
1248                 retval = pid;
1249         return retval;
1250 }
1251
1252 /*
1253  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1254  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1255  * the lock and this task is uninteresting.  If we return nonzero, we have
1256  * released the lock and the system call should return.
1257  */
1258 static int wait_task_zombie(struct task_struct *p, int options,
1259                             struct siginfo __user *infop,
1260                             int __user *stat_addr, struct rusage __user *ru)
1261 {
1262         unsigned long state;
1263         int retval, status, traced;
1264         pid_t pid = task_pid_vnr(p);
1265
1266         if (!likely(options & WEXITED))
1267                 return 0;
1268
1269         if (unlikely(options & WNOWAIT)) {
1270                 uid_t uid = p->uid;
1271                 int exit_code = p->exit_code;
1272                 int why, status;
1273
1274                 get_task_struct(p);
1275                 read_unlock(&tasklist_lock);
1276                 if ((exit_code & 0x7f) == 0) {
1277                         why = CLD_EXITED;
1278                         status = exit_code >> 8;
1279                 } else {
1280                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1281                         status = exit_code & 0x7f;
1282                 }
1283                 return wait_noreap_copyout(p, pid, uid, why,
1284                                            status, infop, ru);
1285         }
1286
1287         /*
1288          * Try to move the task's state to DEAD
1289          * only one thread is allowed to do this:
1290          */
1291         state = xchg(&p->exit_state, EXIT_DEAD);
1292         if (state != EXIT_ZOMBIE) {
1293                 BUG_ON(state != EXIT_DEAD);
1294                 return 0;
1295         }
1296
1297         traced = ptrace_reparented(p);
1298
1299         if (likely(!traced)) {
1300                 struct signal_struct *psig;
1301                 struct signal_struct *sig;
1302
1303                 /*
1304                  * The resource counters for the group leader are in its
1305                  * own task_struct.  Those for dead threads in the group
1306                  * are in its signal_struct, as are those for the child
1307                  * processes it has previously reaped.  All these
1308                  * accumulate in the parent's signal_struct c* fields.
1309                  *
1310                  * We don't bother to take a lock here to protect these
1311                  * p->signal fields, because they are only touched by
1312                  * __exit_signal, which runs with tasklist_lock
1313                  * write-locked anyway, and so is excluded here.  We do
1314                  * need to protect the access to p->parent->signal fields,
1315                  * as other threads in the parent group can be right
1316                  * here reaping other children at the same time.
1317                  */
1318                 spin_lock_irq(&p->parent->sighand->siglock);
1319                 psig = p->parent->signal;
1320                 sig = p->signal;
1321                 psig->cutime =
1322                         cputime_add(psig->cutime,
1323                         cputime_add(p->utime,
1324                         cputime_add(sig->utime,
1325                                     sig->cutime)));
1326                 psig->cstime =
1327                         cputime_add(psig->cstime,
1328                         cputime_add(p->stime,
1329                         cputime_add(sig->stime,
1330                                     sig->cstime)));
1331                 psig->cgtime =
1332                         cputime_add(psig->cgtime,
1333                         cputime_add(p->gtime,
1334                         cputime_add(sig->gtime,
1335                                     sig->cgtime)));
1336                 psig->cmin_flt +=
1337                         p->min_flt + sig->min_flt + sig->cmin_flt;
1338                 psig->cmaj_flt +=
1339                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1340                 psig->cnvcsw +=
1341                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1342                 psig->cnivcsw +=
1343                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1344                 psig->cinblock +=
1345                         task_io_get_inblock(p) +
1346                         sig->inblock + sig->cinblock;
1347                 psig->coublock +=
1348                         task_io_get_oublock(p) +
1349                         sig->oublock + sig->coublock;
1350                 task_io_accounting_add(&psig->ioac, &p->ioac);
1351                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1352                 spin_unlock_irq(&p->parent->sighand->siglock);
1353         }
1354
1355         /*
1356          * Now we are sure this task is interesting, and no other
1357          * thread can reap it because we set its state to EXIT_DEAD.
1358          */
1359         read_unlock(&tasklist_lock);
1360
1361         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1362         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1363                 ? p->signal->group_exit_code : p->exit_code;
1364         if (!retval && stat_addr)
1365                 retval = put_user(status, stat_addr);
1366         if (!retval && infop)
1367                 retval = put_user(SIGCHLD, &infop->si_signo);
1368         if (!retval && infop)
1369                 retval = put_user(0, &infop->si_errno);
1370         if (!retval && infop) {
1371                 int why;
1372
1373                 if ((status & 0x7f) == 0) {
1374                         why = CLD_EXITED;
1375                         status >>= 8;
1376                 } else {
1377                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1378                         status &= 0x7f;
1379                 }
1380                 retval = put_user((short)why, &infop->si_code);
1381                 if (!retval)
1382                         retval = put_user(status, &infop->si_status);
1383         }
1384         if (!retval && infop)
1385                 retval = put_user(pid, &infop->si_pid);
1386         if (!retval && infop)
1387                 retval = put_user(p->uid, &infop->si_uid);
1388         if (!retval)
1389                 retval = pid;
1390
1391         if (traced) {
1392                 write_lock_irq(&tasklist_lock);
1393                 /* We dropped tasklist, ptracer could die and untrace */
1394                 ptrace_unlink(p);
1395                 /*
1396                  * If this is not a detached task, notify the parent.
1397                  * If it's still not detached after that, don't release
1398                  * it now.
1399                  */
1400                 if (!task_detached(p)) {
1401                         do_notify_parent(p, p->exit_signal);
1402                         if (!task_detached(p)) {
1403                                 p->exit_state = EXIT_ZOMBIE;
1404                                 p = NULL;
1405                         }
1406                 }
1407                 write_unlock_irq(&tasklist_lock);
1408         }
1409         if (p != NULL)
1410                 release_task(p);
1411
1412         return retval;
1413 }
1414
1415 /*
1416  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1417  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1418  * the lock and this task is uninteresting.  If we return nonzero, we have
1419  * released the lock and the system call should return.
1420  */
1421 static int wait_task_stopped(int ptrace, struct task_struct *p,
1422                              int options, struct siginfo __user *infop,
1423                              int __user *stat_addr, struct rusage __user *ru)
1424 {
1425         int retval, exit_code, why;
1426         uid_t uid = 0; /* unneeded, required by compiler */
1427         pid_t pid;
1428
1429         if (!(options & WUNTRACED))
1430                 return 0;
1431
1432         exit_code = 0;
1433         spin_lock_irq(&p->sighand->siglock);
1434
1435         if (unlikely(!task_is_stopped_or_traced(p)))
1436                 goto unlock_sig;
1437
1438         if (!ptrace && p->signal->group_stop_count > 0)
1439                 /*
1440                  * A group stop is in progress and this is the group leader.
1441                  * We won't report until all threads have stopped.
1442                  */
1443                 goto unlock_sig;
1444
1445         exit_code = p->exit_code;
1446         if (!exit_code)
1447                 goto unlock_sig;
1448
1449         if (!unlikely(options & WNOWAIT))
1450                 p->exit_code = 0;
1451
1452         uid = p->uid;
1453 unlock_sig:
1454         spin_unlock_irq(&p->sighand->siglock);
1455         if (!exit_code)
1456                 return 0;
1457
1458         /*
1459          * Now we are pretty sure this task is interesting.
1460          * Make sure it doesn't get reaped out from under us while we
1461          * give up the lock and then examine it below.  We don't want to
1462          * keep holding onto the tasklist_lock while we call getrusage and
1463          * possibly take page faults for user memory.
1464          */
1465         get_task_struct(p);
1466         pid = task_pid_vnr(p);
1467         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1468         read_unlock(&tasklist_lock);
1469
1470         if (unlikely(options & WNOWAIT))
1471                 return wait_noreap_copyout(p, pid, uid,
1472                                            why, exit_code,
1473                                            infop, ru);
1474
1475         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1476         if (!retval && stat_addr)
1477                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1478         if (!retval && infop)
1479                 retval = put_user(SIGCHLD, &infop->si_signo);
1480         if (!retval && infop)
1481                 retval = put_user(0, &infop->si_errno);
1482         if (!retval && infop)
1483                 retval = put_user((short)why, &infop->si_code);
1484         if (!retval && infop)
1485                 retval = put_user(exit_code, &infop->si_status);
1486         if (!retval && infop)
1487                 retval = put_user(pid, &infop->si_pid);
1488         if (!retval && infop)
1489                 retval = put_user(uid, &infop->si_uid);
1490         if (!retval)
1491                 retval = pid;
1492         put_task_struct(p);
1493
1494         BUG_ON(!retval);
1495         return retval;
1496 }
1497
1498 /*
1499  * Handle do_wait work for one task in a live, non-stopped state.
1500  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1501  * the lock and this task is uninteresting.  If we return nonzero, we have
1502  * released the lock and the system call should return.
1503  */
1504 static int wait_task_continued(struct task_struct *p, int options,
1505                                struct siginfo __user *infop,
1506                                int __user *stat_addr, struct rusage __user *ru)
1507 {
1508         int retval;
1509         pid_t pid;
1510         uid_t uid;
1511
1512         if (!unlikely(options & WCONTINUED))
1513                 return 0;
1514
1515         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1516                 return 0;
1517
1518         spin_lock_irq(&p->sighand->siglock);
1519         /* Re-check with the lock held.  */
1520         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1521                 spin_unlock_irq(&p->sighand->siglock);
1522                 return 0;
1523         }
1524         if (!unlikely(options & WNOWAIT))
1525                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1526         spin_unlock_irq(&p->sighand->siglock);
1527
1528         pid = task_pid_vnr(p);
1529         uid = p->uid;
1530         get_task_struct(p);
1531         read_unlock(&tasklist_lock);
1532
1533         if (!infop) {
1534                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1535                 put_task_struct(p);
1536                 if (!retval && stat_addr)
1537                         retval = put_user(0xffff, stat_addr);
1538                 if (!retval)
1539                         retval = pid;
1540         } else {
1541                 retval = wait_noreap_copyout(p, pid, uid,
1542                                              CLD_CONTINUED, SIGCONT,
1543                                              infop, ru);
1544                 BUG_ON(retval == 0);
1545         }
1546
1547         return retval;
1548 }
1549
1550 /*
1551  * Consider @p for a wait by @parent.
1552  *
1553  * -ECHILD should be in *@notask_error before the first call.
1554  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1555  * Returns zero if the search for a child should continue;
1556  * then *@notask_error is 0 if @p is an eligible child,
1557  * or another error from security_task_wait(), or still -ECHILD.
1558  */
1559 static int wait_consider_task(struct task_struct *parent, int ptrace,
1560                               struct task_struct *p, int *notask_error,
1561                               enum pid_type type, struct pid *pid, int options,
1562                               struct siginfo __user *infop,
1563                               int __user *stat_addr, struct rusage __user *ru)
1564 {
1565         int ret = eligible_child(type, pid, options, p);
1566         if (!ret)
1567                 return ret;
1568
1569         if (unlikely(ret < 0)) {
1570                 /*
1571                  * If we have not yet seen any eligible child,
1572                  * then let this error code replace -ECHILD.
1573                  * A permission error will give the user a clue
1574                  * to look for security policy problems, rather
1575                  * than for mysterious wait bugs.
1576                  */
1577                 if (*notask_error)
1578                         *notask_error = ret;
1579         }
1580
1581         if (likely(!ptrace) && unlikely(p->ptrace)) {
1582                 /*
1583                  * This child is hidden by ptrace.
1584                  * We aren't allowed to see it now, but eventually we will.
1585                  */
1586                 *notask_error = 0;
1587                 return 0;
1588         }
1589
1590         if (p->exit_state == EXIT_DEAD)
1591                 return 0;
1592
1593         /*
1594          * We don't reap group leaders with subthreads.
1595          */
1596         if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1597                 return wait_task_zombie(p, options, infop, stat_addr, ru);
1598
1599         /*
1600          * It's stopped or running now, so it might
1601          * later continue, exit, or stop again.
1602          */
1603         *notask_error = 0;
1604
1605         if (task_is_stopped_or_traced(p))
1606                 return wait_task_stopped(ptrace, p, options,
1607                                          infop, stat_addr, ru);
1608
1609         return wait_task_continued(p, options, infop, stat_addr, ru);
1610 }
1611
1612 /*
1613  * Do the work of do_wait() for one thread in the group, @tsk.
1614  *
1615  * -ECHILD should be in *@notask_error before the first call.
1616  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1617  * Returns zero if the search for a child should continue; then
1618  * *@notask_error is 0 if there were any eligible children,
1619  * or another error from security_task_wait(), or still -ECHILD.
1620  */
1621 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1622                           enum pid_type type, struct pid *pid, int options,
1623                           struct siginfo __user *infop, int __user *stat_addr,
1624                           struct rusage __user *ru)
1625 {
1626         struct task_struct *p;
1627
1628         list_for_each_entry(p, &tsk->children, sibling) {
1629                 /*
1630                  * Do not consider detached threads.
1631                  */
1632                 if (!task_detached(p)) {
1633                         int ret = wait_consider_task(tsk, 0, p, notask_error,
1634                                                      type, pid, options,
1635                                                      infop, stat_addr, ru);
1636                         if (ret)
1637                                 return ret;
1638                 }
1639         }
1640
1641         return 0;
1642 }
1643
1644 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1645                           enum pid_type type, struct pid *pid, int options,
1646                           struct siginfo __user *infop, int __user *stat_addr,
1647                           struct rusage __user *ru)
1648 {
1649         struct task_struct *p;
1650
1651         /*
1652          * Traditionally we see ptrace'd stopped tasks regardless of options.
1653          */
1654         options |= WUNTRACED;
1655
1656         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1657                 int ret = wait_consider_task(tsk, 1, p, notask_error,
1658                                              type, pid, options,
1659                                              infop, stat_addr, ru);
1660                 if (ret)
1661                         return ret;
1662         }
1663
1664         return 0;
1665 }
1666
1667 static long do_wait(enum pid_type type, struct pid *pid, int options,
1668                     struct siginfo __user *infop, int __user *stat_addr,
1669                     struct rusage __user *ru)
1670 {
1671         DECLARE_WAITQUEUE(wait, current);
1672         struct task_struct *tsk;
1673         int retval;
1674
1675         add_wait_queue(&current->signal->wait_chldexit,&wait);
1676 repeat:
1677         /*
1678          * If there is nothing that can match our critiera just get out.
1679          * We will clear @retval to zero if we see any child that might later
1680          * match our criteria, even if we are not able to reap it yet.
1681          */
1682         retval = -ECHILD;
1683         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1684                 goto end;
1685
1686         current->state = TASK_INTERRUPTIBLE;
1687         read_lock(&tasklist_lock);
1688         tsk = current;
1689         do {
1690                 int tsk_result = do_wait_thread(tsk, &retval,
1691                                                 type, pid, options,
1692                                                 infop, stat_addr, ru);
1693                 if (!tsk_result)
1694                         tsk_result = ptrace_do_wait(tsk, &retval,
1695                                                     type, pid, options,
1696                                                     infop, stat_addr, ru);
1697                 if (tsk_result) {
1698                         /*
1699                          * tasklist_lock is unlocked and we have a final result.
1700                          */
1701                         retval = tsk_result;
1702                         goto end;
1703                 }
1704
1705                 if (options & __WNOTHREAD)
1706                         break;
1707                 tsk = next_thread(tsk);
1708                 BUG_ON(tsk->signal != current->signal);
1709         } while (tsk != current);
1710         read_unlock(&tasklist_lock);
1711
1712         if (!retval && !(options & WNOHANG)) {
1713                 retval = -ERESTARTSYS;
1714                 if (!signal_pending(current)) {
1715                         schedule();
1716                         goto repeat;
1717                 }
1718         }
1719
1720 end:
1721         current->state = TASK_RUNNING;
1722         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1723         if (infop) {
1724                 if (retval > 0)
1725                         retval = 0;
1726                 else {
1727                         /*
1728                          * For a WNOHANG return, clear out all the fields
1729                          * we would set so the user can easily tell the
1730                          * difference.
1731                          */
1732                         if (!retval)
1733                                 retval = put_user(0, &infop->si_signo);
1734                         if (!retval)
1735                                 retval = put_user(0, &infop->si_errno);
1736                         if (!retval)
1737                                 retval = put_user(0, &infop->si_code);
1738                         if (!retval)
1739                                 retval = put_user(0, &infop->si_pid);
1740                         if (!retval)
1741                                 retval = put_user(0, &infop->si_uid);
1742                         if (!retval)
1743                                 retval = put_user(0, &infop->si_status);
1744                 }
1745         }
1746         return retval;
1747 }
1748
1749 asmlinkage long sys_waitid(int which, pid_t upid,
1750                            struct siginfo __user *infop, int options,
1751                            struct rusage __user *ru)
1752 {
1753         struct pid *pid = NULL;
1754         enum pid_type type;
1755         long ret;
1756
1757         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1758                 return -EINVAL;
1759         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1760                 return -EINVAL;
1761
1762         switch (which) {
1763         case P_ALL:
1764                 type = PIDTYPE_MAX;
1765                 break;
1766         case P_PID:
1767                 type = PIDTYPE_PID;
1768                 if (upid <= 0)
1769                         return -EINVAL;
1770                 break;
1771         case P_PGID:
1772                 type = PIDTYPE_PGID;
1773                 if (upid <= 0)
1774                         return -EINVAL;
1775                 break;
1776         default:
1777                 return -EINVAL;
1778         }
1779
1780         if (type < PIDTYPE_MAX)
1781                 pid = find_get_pid(upid);
1782         ret = do_wait(type, pid, options, infop, NULL, ru);
1783         put_pid(pid);
1784
1785         /* avoid REGPARM breakage on x86: */
1786         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1787         return ret;
1788 }
1789
1790 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1791                           int options, struct rusage __user *ru)
1792 {
1793         struct pid *pid = NULL;
1794         enum pid_type type;
1795         long ret;
1796
1797         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1798                         __WNOTHREAD|__WCLONE|__WALL))
1799                 return -EINVAL;
1800
1801         if (upid == -1)
1802                 type = PIDTYPE_MAX;
1803         else if (upid < 0) {
1804                 type = PIDTYPE_PGID;
1805                 pid = find_get_pid(-upid);
1806         } else if (upid == 0) {
1807                 type = PIDTYPE_PGID;
1808                 pid = get_pid(task_pgrp(current));
1809         } else /* upid > 0 */ {
1810                 type = PIDTYPE_PID;
1811                 pid = find_get_pid(upid);
1812         }
1813
1814         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1815         put_pid(pid);
1816
1817         /* avoid REGPARM breakage on x86: */
1818         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1819         return ret;
1820 }
1821
1822 #ifdef __ARCH_WANT_SYS_WAITPID
1823
1824 /*
1825  * sys_waitpid() remains for compatibility. waitpid() should be
1826  * implemented by calling sys_wait4() from libc.a.
1827  */
1828 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1829 {
1830         return sys_wait4(pid, stat_addr, options, NULL);
1831 }
1832
1833 #endif