reparent_thread: fix the "is it traced" check
[linux-2.6.git] / kernel / exit.c
1 /*
2  *  linux/kernel/exit.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 #include <linux/mm.h>
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/mnt_namespace.h>
16 #include <linux/iocontext.h>
17 #include <linux/key.h>
18 #include <linux/security.h>
19 #include <linux/cpu.h>
20 #include <linux/acct.h>
21 #include <linux/tsacct_kern.h>
22 #include <linux/file.h>
23 #include <linux/fdtable.h>
24 #include <linux/binfmts.h>
25 #include <linux/nsproxy.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/ptrace.h>
28 #include <linux/profile.h>
29 #include <linux/mount.h>
30 #include <linux/proc_fs.h>
31 #include <linux/kthread.h>
32 #include <linux/mempolicy.h>
33 #include <linux/taskstats_kern.h>
34 #include <linux/delayacct.h>
35 #include <linux/freezer.h>
36 #include <linux/cgroup.h>
37 #include <linux/syscalls.h>
38 #include <linux/signal.h>
39 #include <linux/posix-timers.h>
40 #include <linux/cn_proc.h>
41 #include <linux/mutex.h>
42 #include <linux/futex.h>
43 #include <linux/pipe_fs_i.h>
44 #include <linux/audit.h> /* for audit_free() */
45 #include <linux/resource.h>
46 #include <linux/blkdev.h>
47 #include <linux/task_io_accounting_ops.h>
48 #include <linux/tracehook.h>
49 #include <linux/init_task.h>
50 #include <trace/sched.h>
51
52 #include <asm/uaccess.h>
53 #include <asm/unistd.h>
54 #include <asm/pgtable.h>
55 #include <asm/mmu_context.h>
56 #include "cred-internals.h"
57
58 DEFINE_TRACE(sched_process_free);
59 DEFINE_TRACE(sched_process_exit);
60 DEFINE_TRACE(sched_process_wait);
61
62 static void exit_mm(struct task_struct * tsk);
63
64 static inline int task_detached(struct task_struct *p)
65 {
66         return p->exit_signal == -1;
67 }
68
69 static void __unhash_process(struct task_struct *p)
70 {
71         nr_threads--;
72         detach_pid(p, PIDTYPE_PID);
73         if (thread_group_leader(p)) {
74                 detach_pid(p, PIDTYPE_PGID);
75                 detach_pid(p, PIDTYPE_SID);
76
77                 list_del_rcu(&p->tasks);
78                 __get_cpu_var(process_counts)--;
79         }
80         list_del_rcu(&p->thread_group);
81         list_del_init(&p->sibling);
82 }
83
84 /*
85  * This function expects the tasklist_lock write-locked.
86  */
87 static void __exit_signal(struct task_struct *tsk)
88 {
89         struct signal_struct *sig = tsk->signal;
90         struct sighand_struct *sighand;
91
92         BUG_ON(!sig);
93         BUG_ON(!atomic_read(&sig->count));
94
95         sighand = rcu_dereference(tsk->sighand);
96         spin_lock(&sighand->siglock);
97
98         posix_cpu_timers_exit(tsk);
99         if (atomic_dec_and_test(&sig->count))
100                 posix_cpu_timers_exit_group(tsk);
101         else {
102                 /*
103                  * If there is any task waiting for the group exit
104                  * then notify it:
105                  */
106                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
107                         wake_up_process(sig->group_exit_task);
108
109                 if (tsk == sig->curr_target)
110                         sig->curr_target = next_thread(tsk);
111                 /*
112                  * Accumulate here the counters for all threads but the
113                  * group leader as they die, so they can be added into
114                  * the process-wide totals when those are taken.
115                  * The group leader stays around as a zombie as long
116                  * as there are other threads.  When it gets reaped,
117                  * the exit.c code will add its counts into these totals.
118                  * We won't ever get here for the group leader, since it
119                  * will have been the last reference on the signal_struct.
120                  */
121                 sig->utime = cputime_add(sig->utime, task_utime(tsk));
122                 sig->stime = cputime_add(sig->stime, task_stime(tsk));
123                 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
124                 sig->min_flt += tsk->min_flt;
125                 sig->maj_flt += tsk->maj_flt;
126                 sig->nvcsw += tsk->nvcsw;
127                 sig->nivcsw += tsk->nivcsw;
128                 sig->inblock += task_io_get_inblock(tsk);
129                 sig->oublock += task_io_get_oublock(tsk);
130                 task_io_accounting_add(&sig->ioac, &tsk->ioac);
131                 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
132                 sig = NULL; /* Marker for below. */
133         }
134
135         __unhash_process(tsk);
136
137         /*
138          * Do this under ->siglock, we can race with another thread
139          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
140          */
141         flush_sigqueue(&tsk->pending);
142
143         tsk->signal = NULL;
144         tsk->sighand = NULL;
145         spin_unlock(&sighand->siglock);
146
147         __cleanup_sighand(sighand);
148         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
149         if (sig) {
150                 flush_sigqueue(&sig->shared_pending);
151                 taskstats_tgid_free(sig);
152                 /*
153                  * Make sure ->signal can't go away under rq->lock,
154                  * see account_group_exec_runtime().
155                  */
156                 task_rq_unlock_wait(tsk);
157                 __cleanup_signal(sig);
158         }
159 }
160
161 static void delayed_put_task_struct(struct rcu_head *rhp)
162 {
163         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
164
165         trace_sched_process_free(tsk);
166         put_task_struct(tsk);
167 }
168
169
170 void release_task(struct task_struct * p)
171 {
172         struct task_struct *leader;
173         int zap_leader;
174 repeat:
175         tracehook_prepare_release_task(p);
176         /* don't need to get the RCU readlock here - the process is dead and
177          * can't be modifying its own credentials */
178         atomic_dec(&__task_cred(p)->user->processes);
179
180         proc_flush_task(p);
181         write_lock_irq(&tasklist_lock);
182         tracehook_finish_release_task(p);
183         __exit_signal(p);
184
185         /*
186          * If we are the last non-leader member of the thread
187          * group, and the leader is zombie, then notify the
188          * group leader's parent process. (if it wants notification.)
189          */
190         zap_leader = 0;
191         leader = p->group_leader;
192         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
193                 BUG_ON(task_detached(leader));
194                 do_notify_parent(leader, leader->exit_signal);
195                 /*
196                  * If we were the last child thread and the leader has
197                  * exited already, and the leader's parent ignores SIGCHLD,
198                  * then we are the one who should release the leader.
199                  *
200                  * do_notify_parent() will have marked it self-reaping in
201                  * that case.
202                  */
203                 zap_leader = task_detached(leader);
204
205                 /*
206                  * This maintains the invariant that release_task()
207                  * only runs on a task in EXIT_DEAD, just for sanity.
208                  */
209                 if (zap_leader)
210                         leader->exit_state = EXIT_DEAD;
211         }
212
213         write_unlock_irq(&tasklist_lock);
214         release_thread(p);
215         call_rcu(&p->rcu, delayed_put_task_struct);
216
217         p = leader;
218         if (unlikely(zap_leader))
219                 goto repeat;
220 }
221
222 /*
223  * This checks not only the pgrp, but falls back on the pid if no
224  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
225  * without this...
226  *
227  * The caller must hold rcu lock or the tasklist lock.
228  */
229 struct pid *session_of_pgrp(struct pid *pgrp)
230 {
231         struct task_struct *p;
232         struct pid *sid = NULL;
233
234         p = pid_task(pgrp, PIDTYPE_PGID);
235         if (p == NULL)
236                 p = pid_task(pgrp, PIDTYPE_PID);
237         if (p != NULL)
238                 sid = task_session(p);
239
240         return sid;
241 }
242
243 /*
244  * Determine if a process group is "orphaned", according to the POSIX
245  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
246  * by terminal-generated stop signals.  Newly orphaned process groups are
247  * to receive a SIGHUP and a SIGCONT.
248  *
249  * "I ask you, have you ever known what it is to be an orphan?"
250  */
251 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
252 {
253         struct task_struct *p;
254
255         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
256                 if ((p == ignored_task) ||
257                     (p->exit_state && thread_group_empty(p)) ||
258                     is_global_init(p->real_parent))
259                         continue;
260
261                 if (task_pgrp(p->real_parent) != pgrp &&
262                     task_session(p->real_parent) == task_session(p))
263                         return 0;
264         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
265
266         return 1;
267 }
268
269 int is_current_pgrp_orphaned(void)
270 {
271         int retval;
272
273         read_lock(&tasklist_lock);
274         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
275         read_unlock(&tasklist_lock);
276
277         return retval;
278 }
279
280 static int has_stopped_jobs(struct pid *pgrp)
281 {
282         int retval = 0;
283         struct task_struct *p;
284
285         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
286                 if (!task_is_stopped(p))
287                         continue;
288                 retval = 1;
289                 break;
290         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
291         return retval;
292 }
293
294 /*
295  * Check to see if any process groups have become orphaned as
296  * a result of our exiting, and if they have any stopped jobs,
297  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
298  */
299 static void
300 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
301 {
302         struct pid *pgrp = task_pgrp(tsk);
303         struct task_struct *ignored_task = tsk;
304
305         if (!parent)
306                  /* exit: our father is in a different pgrp than
307                   * we are and we were the only connection outside.
308                   */
309                 parent = tsk->real_parent;
310         else
311                 /* reparent: our child is in a different pgrp than
312                  * we are, and it was the only connection outside.
313                  */
314                 ignored_task = NULL;
315
316         if (task_pgrp(parent) != pgrp &&
317             task_session(parent) == task_session(tsk) &&
318             will_become_orphaned_pgrp(pgrp, ignored_task) &&
319             has_stopped_jobs(pgrp)) {
320                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
321                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
322         }
323 }
324
325 /**
326  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
327  *
328  * If a kernel thread is launched as a result of a system call, or if
329  * it ever exits, it should generally reparent itself to kthreadd so it
330  * isn't in the way of other processes and is correctly cleaned up on exit.
331  *
332  * The various task state such as scheduling policy and priority may have
333  * been inherited from a user process, so we reset them to sane values here.
334  *
335  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
336  */
337 static void reparent_to_kthreadd(void)
338 {
339         write_lock_irq(&tasklist_lock);
340
341         ptrace_unlink(current);
342         /* Reparent to init */
343         current->real_parent = current->parent = kthreadd_task;
344         list_move_tail(&current->sibling, &current->real_parent->children);
345
346         /* Set the exit signal to SIGCHLD so we signal init on exit */
347         current->exit_signal = SIGCHLD;
348
349         if (task_nice(current) < 0)
350                 set_user_nice(current, 0);
351         /* cpus_allowed? */
352         /* rt_priority? */
353         /* signals? */
354         memcpy(current->signal->rlim, init_task.signal->rlim,
355                sizeof(current->signal->rlim));
356
357         atomic_inc(&init_cred.usage);
358         commit_creds(&init_cred);
359         write_unlock_irq(&tasklist_lock);
360 }
361
362 void __set_special_pids(struct pid *pid)
363 {
364         struct task_struct *curr = current->group_leader;
365         pid_t nr = pid_nr(pid);
366
367         if (task_session(curr) != pid) {
368                 change_pid(curr, PIDTYPE_SID, pid);
369                 set_task_session(curr, nr);
370         }
371         if (task_pgrp(curr) != pid) {
372                 change_pid(curr, PIDTYPE_PGID, pid);
373                 set_task_pgrp(curr, nr);
374         }
375 }
376
377 static void set_special_pids(struct pid *pid)
378 {
379         write_lock_irq(&tasklist_lock);
380         __set_special_pids(pid);
381         write_unlock_irq(&tasklist_lock);
382 }
383
384 /*
385  * Let kernel threads use this to say that they
386  * allow a certain signal (since daemonize() will
387  * have disabled all of them by default).
388  */
389 int allow_signal(int sig)
390 {
391         if (!valid_signal(sig) || sig < 1)
392                 return -EINVAL;
393
394         spin_lock_irq(&current->sighand->siglock);
395         sigdelset(&current->blocked, sig);
396         if (!current->mm) {
397                 /* Kernel threads handle their own signals.
398                    Let the signal code know it'll be handled, so
399                    that they don't get converted to SIGKILL or
400                    just silently dropped */
401                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
402         }
403         recalc_sigpending();
404         spin_unlock_irq(&current->sighand->siglock);
405         return 0;
406 }
407
408 EXPORT_SYMBOL(allow_signal);
409
410 int disallow_signal(int sig)
411 {
412         if (!valid_signal(sig) || sig < 1)
413                 return -EINVAL;
414
415         spin_lock_irq(&current->sighand->siglock);
416         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
417         recalc_sigpending();
418         spin_unlock_irq(&current->sighand->siglock);
419         return 0;
420 }
421
422 EXPORT_SYMBOL(disallow_signal);
423
424 /*
425  *      Put all the gunge required to become a kernel thread without
426  *      attached user resources in one place where it belongs.
427  */
428
429 void daemonize(const char *name, ...)
430 {
431         va_list args;
432         struct fs_struct *fs;
433         sigset_t blocked;
434
435         va_start(args, name);
436         vsnprintf(current->comm, sizeof(current->comm), name, args);
437         va_end(args);
438
439         /*
440          * If we were started as result of loading a module, close all of the
441          * user space pages.  We don't need them, and if we didn't close them
442          * they would be locked into memory.
443          */
444         exit_mm(current);
445         /*
446          * We don't want to have TIF_FREEZE set if the system-wide hibernation
447          * or suspend transition begins right now.
448          */
449         current->flags |= (PF_NOFREEZE | PF_KTHREAD);
450
451         if (current->nsproxy != &init_nsproxy) {
452                 get_nsproxy(&init_nsproxy);
453                 switch_task_namespaces(current, &init_nsproxy);
454         }
455         set_special_pids(&init_struct_pid);
456         proc_clear_tty(current);
457
458         /* Block and flush all signals */
459         sigfillset(&blocked);
460         sigprocmask(SIG_BLOCK, &blocked, NULL);
461         flush_signals(current);
462
463         /* Become as one with the init task */
464
465         exit_fs(current);       /* current->fs->count--; */
466         fs = init_task.fs;
467         current->fs = fs;
468         atomic_inc(&fs->count);
469
470         exit_files(current);
471         current->files = init_task.files;
472         atomic_inc(&current->files->count);
473
474         reparent_to_kthreadd();
475 }
476
477 EXPORT_SYMBOL(daemonize);
478
479 static void close_files(struct files_struct * files)
480 {
481         int i, j;
482         struct fdtable *fdt;
483
484         j = 0;
485
486         /*
487          * It is safe to dereference the fd table without RCU or
488          * ->file_lock because this is the last reference to the
489          * files structure.
490          */
491         fdt = files_fdtable(files);
492         for (;;) {
493                 unsigned long set;
494                 i = j * __NFDBITS;
495                 if (i >= fdt->max_fds)
496                         break;
497                 set = fdt->open_fds->fds_bits[j++];
498                 while (set) {
499                         if (set & 1) {
500                                 struct file * file = xchg(&fdt->fd[i], NULL);
501                                 if (file) {
502                                         filp_close(file, files);
503                                         cond_resched();
504                                 }
505                         }
506                         i++;
507                         set >>= 1;
508                 }
509         }
510 }
511
512 struct files_struct *get_files_struct(struct task_struct *task)
513 {
514         struct files_struct *files;
515
516         task_lock(task);
517         files = task->files;
518         if (files)
519                 atomic_inc(&files->count);
520         task_unlock(task);
521
522         return files;
523 }
524
525 void put_files_struct(struct files_struct *files)
526 {
527         struct fdtable *fdt;
528
529         if (atomic_dec_and_test(&files->count)) {
530                 close_files(files);
531                 /*
532                  * Free the fd and fdset arrays if we expanded them.
533                  * If the fdtable was embedded, pass files for freeing
534                  * at the end of the RCU grace period. Otherwise,
535                  * you can free files immediately.
536                  */
537                 fdt = files_fdtable(files);
538                 if (fdt != &files->fdtab)
539                         kmem_cache_free(files_cachep, files);
540                 free_fdtable(fdt);
541         }
542 }
543
544 void reset_files_struct(struct files_struct *files)
545 {
546         struct task_struct *tsk = current;
547         struct files_struct *old;
548
549         old = tsk->files;
550         task_lock(tsk);
551         tsk->files = files;
552         task_unlock(tsk);
553         put_files_struct(old);
554 }
555
556 void exit_files(struct task_struct *tsk)
557 {
558         struct files_struct * files = tsk->files;
559
560         if (files) {
561                 task_lock(tsk);
562                 tsk->files = NULL;
563                 task_unlock(tsk);
564                 put_files_struct(files);
565         }
566 }
567
568 void put_fs_struct(struct fs_struct *fs)
569 {
570         /* No need to hold fs->lock if we are killing it */
571         if (atomic_dec_and_test(&fs->count)) {
572                 path_put(&fs->root);
573                 path_put(&fs->pwd);
574                 kmem_cache_free(fs_cachep, fs);
575         }
576 }
577
578 void exit_fs(struct task_struct *tsk)
579 {
580         struct fs_struct * fs = tsk->fs;
581
582         if (fs) {
583                 task_lock(tsk);
584                 tsk->fs = NULL;
585                 task_unlock(tsk);
586                 put_fs_struct(fs);
587         }
588 }
589
590 EXPORT_SYMBOL_GPL(exit_fs);
591
592 #ifdef CONFIG_MM_OWNER
593 /*
594  * Task p is exiting and it owned mm, lets find a new owner for it
595  */
596 static inline int
597 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
598 {
599         /*
600          * If there are other users of the mm and the owner (us) is exiting
601          * we need to find a new owner to take on the responsibility.
602          */
603         if (atomic_read(&mm->mm_users) <= 1)
604                 return 0;
605         if (mm->owner != p)
606                 return 0;
607         return 1;
608 }
609
610 void mm_update_next_owner(struct mm_struct *mm)
611 {
612         struct task_struct *c, *g, *p = current;
613
614 retry:
615         if (!mm_need_new_owner(mm, p))
616                 return;
617
618         read_lock(&tasklist_lock);
619         /*
620          * Search in the children
621          */
622         list_for_each_entry(c, &p->children, sibling) {
623                 if (c->mm == mm)
624                         goto assign_new_owner;
625         }
626
627         /*
628          * Search in the siblings
629          */
630         list_for_each_entry(c, &p->parent->children, sibling) {
631                 if (c->mm == mm)
632                         goto assign_new_owner;
633         }
634
635         /*
636          * Search through everything else. We should not get
637          * here often
638          */
639         do_each_thread(g, c) {
640                 if (c->mm == mm)
641                         goto assign_new_owner;
642         } while_each_thread(g, c);
643
644         read_unlock(&tasklist_lock);
645         /*
646          * We found no owner yet mm_users > 1: this implies that we are
647          * most likely racing with swapoff (try_to_unuse()) or /proc or
648          * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
649          */
650         mm->owner = NULL;
651         return;
652
653 assign_new_owner:
654         BUG_ON(c == p);
655         get_task_struct(c);
656         /*
657          * The task_lock protects c->mm from changing.
658          * We always want mm->owner->mm == mm
659          */
660         task_lock(c);
661         /*
662          * Delay read_unlock() till we have the task_lock()
663          * to ensure that c does not slip away underneath us
664          */
665         read_unlock(&tasklist_lock);
666         if (c->mm != mm) {
667                 task_unlock(c);
668                 put_task_struct(c);
669                 goto retry;
670         }
671         mm->owner = c;
672         task_unlock(c);
673         put_task_struct(c);
674 }
675 #endif /* CONFIG_MM_OWNER */
676
677 /*
678  * Turn us into a lazy TLB process if we
679  * aren't already..
680  */
681 static void exit_mm(struct task_struct * tsk)
682 {
683         struct mm_struct *mm = tsk->mm;
684         struct core_state *core_state;
685
686         mm_release(tsk, mm);
687         if (!mm)
688                 return;
689         /*
690          * Serialize with any possible pending coredump.
691          * We must hold mmap_sem around checking core_state
692          * and clearing tsk->mm.  The core-inducing thread
693          * will increment ->nr_threads for each thread in the
694          * group with ->mm != NULL.
695          */
696         down_read(&mm->mmap_sem);
697         core_state = mm->core_state;
698         if (core_state) {
699                 struct core_thread self;
700                 up_read(&mm->mmap_sem);
701
702                 self.task = tsk;
703                 self.next = xchg(&core_state->dumper.next, &self);
704                 /*
705                  * Implies mb(), the result of xchg() must be visible
706                  * to core_state->dumper.
707                  */
708                 if (atomic_dec_and_test(&core_state->nr_threads))
709                         complete(&core_state->startup);
710
711                 for (;;) {
712                         set_task_state(tsk, TASK_UNINTERRUPTIBLE);
713                         if (!self.task) /* see coredump_finish() */
714                                 break;
715                         schedule();
716                 }
717                 __set_task_state(tsk, TASK_RUNNING);
718                 down_read(&mm->mmap_sem);
719         }
720         atomic_inc(&mm->mm_count);
721         BUG_ON(mm != tsk->active_mm);
722         /* more a memory barrier than a real lock */
723         task_lock(tsk);
724         tsk->mm = NULL;
725         up_read(&mm->mmap_sem);
726         enter_lazy_tlb(mm, current);
727         /* We don't want this task to be frozen prematurely */
728         clear_freeze_flag(tsk);
729         task_unlock(tsk);
730         mm_update_next_owner(mm);
731         mmput(mm);
732 }
733
734 /*
735  * Called with irqs disabled, returns true if childs should reap themselves.
736  */
737 static int ignoring_children(struct sighand_struct *sigh)
738 {
739         int ret;
740         spin_lock(&sigh->siglock);
741         ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
742               (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
743         spin_unlock(&sigh->siglock);
744         return ret;
745 }
746
747 /* Returns nonzero if the tracee should be released. */
748 int __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
749 {
750         __ptrace_unlink(p);
751
752         if (p->exit_state != EXIT_ZOMBIE)
753                 return 0;
754         /*
755          * If it's a zombie, our attachedness prevented normal
756          * parent notification or self-reaping.  Do notification
757          * now if it would have happened earlier.  If it should
758          * reap itself we return true.
759          *
760          * If it's our own child, there is no notification to do.
761          * But if our normal children self-reap, then this child
762          * was prevented by ptrace and we must reap it now.
763          */
764         if (!task_detached(p) && thread_group_empty(p)) {
765                 if (!same_thread_group(p->real_parent, tracer))
766                         do_notify_parent(p, p->exit_signal);
767                 else if (ignoring_children(tracer->sighand))
768                         p->exit_signal = -1;
769         }
770
771         if (!task_detached(p))
772                 return 0;
773
774         /* Mark it as in the process of being reaped. */
775         p->exit_state = EXIT_DEAD;
776         return 1;
777 }
778
779 /*
780  * Detach all tasks we were using ptrace on.
781  * Any that need to be release_task'd are put on the @dead list.
782  *
783  * Called with write_lock(&tasklist_lock) held.
784  */
785 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
786 {
787         struct task_struct *p, *n;
788
789         list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
790                 if (__ptrace_detach(parent, p))
791                         list_add(&p->ptrace_entry, dead);
792         }
793 }
794
795 /*
796  * Finish up exit-time ptrace cleanup.
797  *
798  * Called without locks.
799  */
800 static void ptrace_exit_finish(struct task_struct *parent,
801                                struct list_head *dead)
802 {
803         struct task_struct *p, *n;
804
805         BUG_ON(!list_empty(&parent->ptraced));
806
807         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
808                 list_del_init(&p->ptrace_entry);
809                 release_task(p);
810         }
811 }
812
813 static void reparent_thread(struct task_struct *p, struct task_struct *father)
814 {
815         if (p->pdeath_signal)
816                 /* We already hold the tasklist_lock here.  */
817                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
818
819         list_move_tail(&p->sibling, &p->real_parent->children);
820
821         if (task_detached(p))
822                 return;
823         /* If this is a threaded reparent there is no need to
824          * notify anyone anything has happened.
825          */
826         if (same_thread_group(p->real_parent, father))
827                 return;
828
829         /* We don't want people slaying init.  */
830         p->exit_signal = SIGCHLD;
831
832         /* If we'd notified the old parent about this child's death,
833          * also notify the new parent.
834          */
835         if (!p->ptrace &&
836             p->exit_state == EXIT_ZOMBIE && thread_group_empty(p))
837                 do_notify_parent(p, p->exit_signal);
838
839         kill_orphaned_pgrp(p, father);
840 }
841
842 /*
843  * When we die, we re-parent all our children.
844  * Try to give them to another thread in our thread
845  * group, and if no such member exists, give it to
846  * the child reaper process (ie "init") in our pid
847  * space.
848  */
849 static struct task_struct *find_new_reaper(struct task_struct *father)
850 {
851         struct pid_namespace *pid_ns = task_active_pid_ns(father);
852         struct task_struct *thread;
853
854         thread = father;
855         while_each_thread(father, thread) {
856                 if (thread->flags & PF_EXITING)
857                         continue;
858                 if (unlikely(pid_ns->child_reaper == father))
859                         pid_ns->child_reaper = thread;
860                 return thread;
861         }
862
863         if (unlikely(pid_ns->child_reaper == father)) {
864                 write_unlock_irq(&tasklist_lock);
865                 if (unlikely(pid_ns == &init_pid_ns))
866                         panic("Attempted to kill init!");
867
868                 zap_pid_ns_processes(pid_ns);
869                 write_lock_irq(&tasklist_lock);
870                 /*
871                  * We can not clear ->child_reaper or leave it alone.
872                  * There may by stealth EXIT_DEAD tasks on ->children,
873                  * forget_original_parent() must move them somewhere.
874                  */
875                 pid_ns->child_reaper = init_pid_ns.child_reaper;
876         }
877
878         return pid_ns->child_reaper;
879 }
880
881 static void forget_original_parent(struct task_struct *father)
882 {
883         struct task_struct *p, *n, *reaper;
884         LIST_HEAD(ptrace_dead);
885
886         write_lock_irq(&tasklist_lock);
887         reaper = find_new_reaper(father);
888         /*
889          * First clean up ptrace if we were using it.
890          */
891         ptrace_exit(father, &ptrace_dead);
892
893         list_for_each_entry_safe(p, n, &father->children, sibling) {
894                 p->real_parent = reaper;
895                 if (p->parent == father) {
896                         BUG_ON(p->ptrace);
897                         p->parent = p->real_parent;
898                 }
899                 reparent_thread(p, father);
900         }
901
902         write_unlock_irq(&tasklist_lock);
903         BUG_ON(!list_empty(&father->children));
904
905         ptrace_exit_finish(father, &ptrace_dead);
906 }
907
908 /*
909  * Send signals to all our closest relatives so that they know
910  * to properly mourn us..
911  */
912 static void exit_notify(struct task_struct *tsk, int group_dead)
913 {
914         int signal;
915         void *cookie;
916
917         /*
918          * This does two things:
919          *
920          * A.  Make init inherit all the child processes
921          * B.  Check to see if any process groups have become orphaned
922          *      as a result of our exiting, and if they have any stopped
923          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
924          */
925         forget_original_parent(tsk);
926         exit_task_namespaces(tsk);
927
928         write_lock_irq(&tasklist_lock);
929         if (group_dead)
930                 kill_orphaned_pgrp(tsk->group_leader, NULL);
931
932         /* Let father know we died
933          *
934          * Thread signals are configurable, but you aren't going to use
935          * that to send signals to arbitary processes.
936          * That stops right now.
937          *
938          * If the parent exec id doesn't match the exec id we saved
939          * when we started then we know the parent has changed security
940          * domain.
941          *
942          * If our self_exec id doesn't match our parent_exec_id then
943          * we have changed execution domain as these two values started
944          * the same after a fork.
945          */
946         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
947             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
948              tsk->self_exec_id != tsk->parent_exec_id) &&
949             !capable(CAP_KILL))
950                 tsk->exit_signal = SIGCHLD;
951
952         signal = tracehook_notify_death(tsk, &cookie, group_dead);
953         if (signal >= 0)
954                 signal = do_notify_parent(tsk, signal);
955
956         tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
957
958         /* mt-exec, de_thread() is waiting for us */
959         if (thread_group_leader(tsk) &&
960             tsk->signal->group_exit_task &&
961             tsk->signal->notify_count < 0)
962                 wake_up_process(tsk->signal->group_exit_task);
963
964         write_unlock_irq(&tasklist_lock);
965
966         tracehook_report_death(tsk, signal, cookie, group_dead);
967
968         /* If the process is dead, release it - nobody will wait for it */
969         if (signal == DEATH_REAP)
970                 release_task(tsk);
971 }
972
973 #ifdef CONFIG_DEBUG_STACK_USAGE
974 static void check_stack_usage(void)
975 {
976         static DEFINE_SPINLOCK(low_water_lock);
977         static int lowest_to_date = THREAD_SIZE;
978         unsigned long free;
979
980         free = stack_not_used(current);
981
982         if (free >= lowest_to_date)
983                 return;
984
985         spin_lock(&low_water_lock);
986         if (free < lowest_to_date) {
987                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
988                                 "left\n",
989                                 current->comm, free);
990                 lowest_to_date = free;
991         }
992         spin_unlock(&low_water_lock);
993 }
994 #else
995 static inline void check_stack_usage(void) {}
996 #endif
997
998 NORET_TYPE void do_exit(long code)
999 {
1000         struct task_struct *tsk = current;
1001         int group_dead;
1002
1003         profile_task_exit(tsk);
1004
1005         WARN_ON(atomic_read(&tsk->fs_excl));
1006
1007         if (unlikely(in_interrupt()))
1008                 panic("Aiee, killing interrupt handler!");
1009         if (unlikely(!tsk->pid))
1010                 panic("Attempted to kill the idle task!");
1011
1012         tracehook_report_exit(&code);
1013
1014         /*
1015          * We're taking recursive faults here in do_exit. Safest is to just
1016          * leave this task alone and wait for reboot.
1017          */
1018         if (unlikely(tsk->flags & PF_EXITING)) {
1019                 printk(KERN_ALERT
1020                         "Fixing recursive fault but reboot is needed!\n");
1021                 /*
1022                  * We can do this unlocked here. The futex code uses
1023                  * this flag just to verify whether the pi state
1024                  * cleanup has been done or not. In the worst case it
1025                  * loops once more. We pretend that the cleanup was
1026                  * done as there is no way to return. Either the
1027                  * OWNER_DIED bit is set by now or we push the blocked
1028                  * task into the wait for ever nirwana as well.
1029                  */
1030                 tsk->flags |= PF_EXITPIDONE;
1031                 set_current_state(TASK_UNINTERRUPTIBLE);
1032                 schedule();
1033         }
1034
1035         exit_signals(tsk);  /* sets PF_EXITING */
1036         /*
1037          * tsk->flags are checked in the futex code to protect against
1038          * an exiting task cleaning up the robust pi futexes.
1039          */
1040         smp_mb();
1041         spin_unlock_wait(&tsk->pi_lock);
1042
1043         if (unlikely(in_atomic()))
1044                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1045                                 current->comm, task_pid_nr(current),
1046                                 preempt_count());
1047
1048         acct_update_integrals(tsk);
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         if (group_dead)
1057                 tty_audit_exit();
1058         if (unlikely(tsk->audit_context))
1059                 audit_free(tsk);
1060
1061         tsk->exit_code = code;
1062         taskstats_exit(tsk, group_dead);
1063
1064         exit_mm(tsk);
1065
1066         if (group_dead)
1067                 acct_process();
1068         trace_sched_process_exit(tsk);
1069
1070         exit_sem(tsk);
1071         exit_files(tsk);
1072         exit_fs(tsk);
1073         check_stack_usage();
1074         exit_thread();
1075         cgroup_exit(tsk, 1);
1076
1077         if (group_dead && tsk->signal->leader)
1078                 disassociate_ctty(1);
1079
1080         module_put(task_thread_info(tsk)->exec_domain->module);
1081         if (tsk->binfmt)
1082                 module_put(tsk->binfmt->module);
1083
1084         proc_exit_connector(tsk);
1085         exit_notify(tsk, group_dead);
1086 #ifdef CONFIG_NUMA
1087         mpol_put(tsk->mempolicy);
1088         tsk->mempolicy = NULL;
1089 #endif
1090 #ifdef CONFIG_FUTEX
1091         /*
1092          * This must happen late, after the PID is not
1093          * hashed anymore:
1094          */
1095         if (unlikely(!list_empty(&tsk->pi_state_list)))
1096                 exit_pi_state_list(tsk);
1097         if (unlikely(current->pi_state_cache))
1098                 kfree(current->pi_state_cache);
1099 #endif
1100         /*
1101          * Make sure we are holding no locks:
1102          */
1103         debug_check_no_locks_held(tsk);
1104         /*
1105          * We can do this unlocked here. The futex code uses this flag
1106          * just to verify whether the pi state cleanup has been done
1107          * or not. In the worst case it loops once more.
1108          */
1109         tsk->flags |= PF_EXITPIDONE;
1110
1111         if (tsk->io_context)
1112                 exit_io_context();
1113
1114         if (tsk->splice_pipe)
1115                 __free_pipe_info(tsk->splice_pipe);
1116
1117         preempt_disable();
1118         /* causes final put_task_struct in finish_task_switch(). */
1119         tsk->state = TASK_DEAD;
1120         schedule();
1121         BUG();
1122         /* Avoid "noreturn function does return".  */
1123         for (;;)
1124                 cpu_relax();    /* For when BUG is null */
1125 }
1126
1127 EXPORT_SYMBOL_GPL(do_exit);
1128
1129 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1130 {
1131         if (comp)
1132                 complete(comp);
1133
1134         do_exit(code);
1135 }
1136
1137 EXPORT_SYMBOL(complete_and_exit);
1138
1139 SYSCALL_DEFINE1(exit, int, error_code)
1140 {
1141         do_exit((error_code&0xff)<<8);
1142 }
1143
1144 /*
1145  * Take down every thread in the group.  This is called by fatal signals
1146  * as well as by sys_exit_group (below).
1147  */
1148 NORET_TYPE void
1149 do_group_exit(int exit_code)
1150 {
1151         struct signal_struct *sig = current->signal;
1152
1153         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1154
1155         if (signal_group_exit(sig))
1156                 exit_code = sig->group_exit_code;
1157         else if (!thread_group_empty(current)) {
1158                 struct sighand_struct *const sighand = current->sighand;
1159                 spin_lock_irq(&sighand->siglock);
1160                 if (signal_group_exit(sig))
1161                         /* Another thread got here before we took the lock.  */
1162                         exit_code = sig->group_exit_code;
1163                 else {
1164                         sig->group_exit_code = exit_code;
1165                         sig->flags = SIGNAL_GROUP_EXIT;
1166                         zap_other_threads(current);
1167                 }
1168                 spin_unlock_irq(&sighand->siglock);
1169         }
1170
1171         do_exit(exit_code);
1172         /* NOTREACHED */
1173 }
1174
1175 /*
1176  * this kills every thread in the thread group. Note that any externally
1177  * wait4()-ing process will get the correct exit code - even if this
1178  * thread is not the thread group leader.
1179  */
1180 SYSCALL_DEFINE1(exit_group, int, error_code)
1181 {
1182         do_group_exit((error_code & 0xff) << 8);
1183         /* NOTREACHED */
1184         return 0;
1185 }
1186
1187 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1188 {
1189         struct pid *pid = NULL;
1190         if (type == PIDTYPE_PID)
1191                 pid = task->pids[type].pid;
1192         else if (type < PIDTYPE_MAX)
1193                 pid = task->group_leader->pids[type].pid;
1194         return pid;
1195 }
1196
1197 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1198                           struct task_struct *p)
1199 {
1200         int err;
1201
1202         if (type < PIDTYPE_MAX) {
1203                 if (task_pid_type(p, type) != pid)
1204                         return 0;
1205         }
1206
1207         /* Wait for all children (clone and not) if __WALL is set;
1208          * otherwise, wait for clone children *only* if __WCLONE is
1209          * set; otherwise, wait for non-clone children *only*.  (Note:
1210          * A "clone" child here is one that reports to its parent
1211          * using a signal other than SIGCHLD.) */
1212         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1213             && !(options & __WALL))
1214                 return 0;
1215
1216         err = security_task_wait(p);
1217         if (err)
1218                 return err;
1219
1220         return 1;
1221 }
1222
1223 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1224                                int why, int status,
1225                                struct siginfo __user *infop,
1226                                struct rusage __user *rusagep)
1227 {
1228         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1229
1230         put_task_struct(p);
1231         if (!retval)
1232                 retval = put_user(SIGCHLD, &infop->si_signo);
1233         if (!retval)
1234                 retval = put_user(0, &infop->si_errno);
1235         if (!retval)
1236                 retval = put_user((short)why, &infop->si_code);
1237         if (!retval)
1238                 retval = put_user(pid, &infop->si_pid);
1239         if (!retval)
1240                 retval = put_user(uid, &infop->si_uid);
1241         if (!retval)
1242                 retval = put_user(status, &infop->si_status);
1243         if (!retval)
1244                 retval = pid;
1245         return retval;
1246 }
1247
1248 /*
1249  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1250  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1251  * the lock and this task is uninteresting.  If we return nonzero, we have
1252  * released the lock and the system call should return.
1253  */
1254 static int wait_task_zombie(struct task_struct *p, int options,
1255                             struct siginfo __user *infop,
1256                             int __user *stat_addr, struct rusage __user *ru)
1257 {
1258         unsigned long state;
1259         int retval, status, traced;
1260         pid_t pid = task_pid_vnr(p);
1261         uid_t uid = __task_cred(p)->uid;
1262
1263         if (!likely(options & WEXITED))
1264                 return 0;
1265
1266         if (unlikely(options & WNOWAIT)) {
1267                 int exit_code = p->exit_code;
1268                 int why, status;
1269
1270                 get_task_struct(p);
1271                 read_unlock(&tasklist_lock);
1272                 if ((exit_code & 0x7f) == 0) {
1273                         why = CLD_EXITED;
1274                         status = exit_code >> 8;
1275                 } else {
1276                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1277                         status = exit_code & 0x7f;
1278                 }
1279                 return wait_noreap_copyout(p, pid, uid, why,
1280                                            status, infop, ru);
1281         }
1282
1283         /*
1284          * Try to move the task's state to DEAD
1285          * only one thread is allowed to do this:
1286          */
1287         state = xchg(&p->exit_state, EXIT_DEAD);
1288         if (state != EXIT_ZOMBIE) {
1289                 BUG_ON(state != EXIT_DEAD);
1290                 return 0;
1291         }
1292
1293         traced = ptrace_reparented(p);
1294
1295         if (likely(!traced)) {
1296                 struct signal_struct *psig;
1297                 struct signal_struct *sig;
1298                 struct task_cputime cputime;
1299
1300                 /*
1301                  * The resource counters for the group leader are in its
1302                  * own task_struct.  Those for dead threads in the group
1303                  * are in its signal_struct, as are those for the child
1304                  * processes it has previously reaped.  All these
1305                  * accumulate in the parent's signal_struct c* fields.
1306                  *
1307                  * We don't bother to take a lock here to protect these
1308                  * p->signal fields, because they are only touched by
1309                  * __exit_signal, which runs with tasklist_lock
1310                  * write-locked anyway, and so is excluded here.  We do
1311                  * need to protect the access to p->parent->signal fields,
1312                  * as other threads in the parent group can be right
1313                  * here reaping other children at the same time.
1314                  *
1315                  * We use thread_group_cputime() to get times for the thread
1316                  * group, which consolidates times for all threads in the
1317                  * group including the group leader.
1318                  */
1319                 thread_group_cputime(p, &cputime);
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(cputime.utime,
1326                                     sig->cutime));
1327                 psig->cstime =
1328                         cputime_add(psig->cstime,
1329                         cputime_add(cputime.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(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 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1416 {
1417         if (ptrace) {
1418                 if (task_is_stopped_or_traced(p))
1419                         return &p->exit_code;
1420         } else {
1421                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1422                         return &p->signal->group_exit_code;
1423         }
1424         return NULL;
1425 }
1426
1427 /*
1428  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1429  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1430  * the lock and this task is uninteresting.  If we return nonzero, we have
1431  * released the lock and the system call should return.
1432  */
1433 static int wait_task_stopped(int ptrace, struct task_struct *p,
1434                              int options, struct siginfo __user *infop,
1435                              int __user *stat_addr, struct rusage __user *ru)
1436 {
1437         int retval, exit_code, *p_code, why;
1438         uid_t uid = 0; /* unneeded, required by compiler */
1439         pid_t pid;
1440
1441         if (!(options & WUNTRACED))
1442                 return 0;
1443
1444         exit_code = 0;
1445         spin_lock_irq(&p->sighand->siglock);
1446
1447         p_code = task_stopped_code(p, ptrace);
1448         if (unlikely(!p_code))
1449                 goto unlock_sig;
1450
1451         exit_code = *p_code;
1452         if (!exit_code)
1453                 goto unlock_sig;
1454
1455         if (!unlikely(options & WNOWAIT))
1456                 *p_code = 0;
1457
1458         /* don't need the RCU readlock here as we're holding a spinlock */
1459         uid = __task_cred(p)->uid;
1460 unlock_sig:
1461         spin_unlock_irq(&p->sighand->siglock);
1462         if (!exit_code)
1463                 return 0;
1464
1465         /*
1466          * Now we are pretty sure this task is interesting.
1467          * Make sure it doesn't get reaped out from under us while we
1468          * give up the lock and then examine it below.  We don't want to
1469          * keep holding onto the tasklist_lock while we call getrusage and
1470          * possibly take page faults for user memory.
1471          */
1472         get_task_struct(p);
1473         pid = task_pid_vnr(p);
1474         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1475         read_unlock(&tasklist_lock);
1476
1477         if (unlikely(options & WNOWAIT))
1478                 return wait_noreap_copyout(p, pid, uid,
1479                                            why, exit_code,
1480                                            infop, ru);
1481
1482         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1483         if (!retval && stat_addr)
1484                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1485         if (!retval && infop)
1486                 retval = put_user(SIGCHLD, &infop->si_signo);
1487         if (!retval && infop)
1488                 retval = put_user(0, &infop->si_errno);
1489         if (!retval && infop)
1490                 retval = put_user((short)why, &infop->si_code);
1491         if (!retval && infop)
1492                 retval = put_user(exit_code, &infop->si_status);
1493         if (!retval && infop)
1494                 retval = put_user(pid, &infop->si_pid);
1495         if (!retval && infop)
1496                 retval = put_user(uid, &infop->si_uid);
1497         if (!retval)
1498                 retval = pid;
1499         put_task_struct(p);
1500
1501         BUG_ON(!retval);
1502         return retval;
1503 }
1504
1505 /*
1506  * Handle do_wait work for one task in a live, non-stopped state.
1507  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1508  * the lock and this task is uninteresting.  If we return nonzero, we have
1509  * released the lock and the system call should return.
1510  */
1511 static int wait_task_continued(struct task_struct *p, int options,
1512                                struct siginfo __user *infop,
1513                                int __user *stat_addr, struct rusage __user *ru)
1514 {
1515         int retval;
1516         pid_t pid;
1517         uid_t uid;
1518
1519         if (!unlikely(options & WCONTINUED))
1520                 return 0;
1521
1522         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1523                 return 0;
1524
1525         spin_lock_irq(&p->sighand->siglock);
1526         /* Re-check with the lock held.  */
1527         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1528                 spin_unlock_irq(&p->sighand->siglock);
1529                 return 0;
1530         }
1531         if (!unlikely(options & WNOWAIT))
1532                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1533         uid = __task_cred(p)->uid;
1534         spin_unlock_irq(&p->sighand->siglock);
1535
1536         pid = task_pid_vnr(p);
1537         get_task_struct(p);
1538         read_unlock(&tasklist_lock);
1539
1540         if (!infop) {
1541                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1542                 put_task_struct(p);
1543                 if (!retval && stat_addr)
1544                         retval = put_user(0xffff, stat_addr);
1545                 if (!retval)
1546                         retval = pid;
1547         } else {
1548                 retval = wait_noreap_copyout(p, pid, uid,
1549                                              CLD_CONTINUED, SIGCONT,
1550                                              infop, ru);
1551                 BUG_ON(retval == 0);
1552         }
1553
1554         return retval;
1555 }
1556
1557 /*
1558  * Consider @p for a wait by @parent.
1559  *
1560  * -ECHILD should be in *@notask_error before the first call.
1561  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1562  * Returns zero if the search for a child should continue;
1563  * then *@notask_error is 0 if @p is an eligible child,
1564  * or another error from security_task_wait(), or still -ECHILD.
1565  */
1566 static int wait_consider_task(struct task_struct *parent, int ptrace,
1567                               struct task_struct *p, int *notask_error,
1568                               enum pid_type type, struct pid *pid, int options,
1569                               struct siginfo __user *infop,
1570                               int __user *stat_addr, struct rusage __user *ru)
1571 {
1572         int ret = eligible_child(type, pid, options, p);
1573         if (!ret)
1574                 return ret;
1575
1576         if (unlikely(ret < 0)) {
1577                 /*
1578                  * If we have not yet seen any eligible child,
1579                  * then let this error code replace -ECHILD.
1580                  * A permission error will give the user a clue
1581                  * to look for security policy problems, rather
1582                  * than for mysterious wait bugs.
1583                  */
1584                 if (*notask_error)
1585                         *notask_error = ret;
1586         }
1587
1588         if (likely(!ptrace) && unlikely(p->ptrace)) {
1589                 /*
1590                  * This child is hidden by ptrace.
1591                  * We aren't allowed to see it now, but eventually we will.
1592                  */
1593                 *notask_error = 0;
1594                 return 0;
1595         }
1596
1597         if (p->exit_state == EXIT_DEAD)
1598                 return 0;
1599
1600         /*
1601          * We don't reap group leaders with subthreads.
1602          */
1603         if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1604                 return wait_task_zombie(p, options, infop, stat_addr, ru);
1605
1606         /*
1607          * It's stopped or running now, so it might
1608          * later continue, exit, or stop again.
1609          */
1610         *notask_error = 0;
1611
1612         if (task_stopped_code(p, ptrace))
1613                 return wait_task_stopped(ptrace, p, options,
1614                                          infop, stat_addr, ru);
1615
1616         return wait_task_continued(p, options, infop, stat_addr, ru);
1617 }
1618
1619 /*
1620  * Do the work of do_wait() for one thread in the group, @tsk.
1621  *
1622  * -ECHILD should be in *@notask_error before the first call.
1623  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1624  * Returns zero if the search for a child should continue; then
1625  * *@notask_error is 0 if there were any eligible children,
1626  * or another error from security_task_wait(), or still -ECHILD.
1627  */
1628 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1629                           enum pid_type type, struct pid *pid, int options,
1630                           struct siginfo __user *infop, int __user *stat_addr,
1631                           struct rusage __user *ru)
1632 {
1633         struct task_struct *p;
1634
1635         list_for_each_entry(p, &tsk->children, sibling) {
1636                 /*
1637                  * Do not consider detached threads.
1638                  */
1639                 if (!task_detached(p)) {
1640                         int ret = wait_consider_task(tsk, 0, p, notask_error,
1641                                                      type, pid, options,
1642                                                      infop, stat_addr, ru);
1643                         if (ret)
1644                                 return ret;
1645                 }
1646         }
1647
1648         return 0;
1649 }
1650
1651 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1652                           enum pid_type type, struct pid *pid, int options,
1653                           struct siginfo __user *infop, int __user *stat_addr,
1654                           struct rusage __user *ru)
1655 {
1656         struct task_struct *p;
1657
1658         /*
1659          * Traditionally we see ptrace'd stopped tasks regardless of options.
1660          */
1661         options |= WUNTRACED;
1662
1663         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1664                 int ret = wait_consider_task(tsk, 1, p, notask_error,
1665                                              type, pid, options,
1666                                              infop, stat_addr, ru);
1667                 if (ret)
1668                         return ret;
1669         }
1670
1671         return 0;
1672 }
1673
1674 static long do_wait(enum pid_type type, struct pid *pid, int options,
1675                     struct siginfo __user *infop, int __user *stat_addr,
1676                     struct rusage __user *ru)
1677 {
1678         DECLARE_WAITQUEUE(wait, current);
1679         struct task_struct *tsk;
1680         int retval;
1681
1682         trace_sched_process_wait(pid);
1683
1684         add_wait_queue(&current->signal->wait_chldexit,&wait);
1685 repeat:
1686         /*
1687          * If there is nothing that can match our critiera just get out.
1688          * We will clear @retval to zero if we see any child that might later
1689          * match our criteria, even if we are not able to reap it yet.
1690          */
1691         retval = -ECHILD;
1692         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1693                 goto end;
1694
1695         current->state = TASK_INTERRUPTIBLE;
1696         read_lock(&tasklist_lock);
1697         tsk = current;
1698         do {
1699                 int tsk_result = do_wait_thread(tsk, &retval,
1700                                                 type, pid, options,
1701                                                 infop, stat_addr, ru);
1702                 if (!tsk_result)
1703                         tsk_result = ptrace_do_wait(tsk, &retval,
1704                                                     type, pid, options,
1705                                                     infop, stat_addr, ru);
1706                 if (tsk_result) {
1707                         /*
1708                          * tasklist_lock is unlocked and we have a final result.
1709                          */
1710                         retval = tsk_result;
1711                         goto end;
1712                 }
1713
1714                 if (options & __WNOTHREAD)
1715                         break;
1716                 tsk = next_thread(tsk);
1717                 BUG_ON(tsk->signal != current->signal);
1718         } while (tsk != current);
1719         read_unlock(&tasklist_lock);
1720
1721         if (!retval && !(options & WNOHANG)) {
1722                 retval = -ERESTARTSYS;
1723                 if (!signal_pending(current)) {
1724                         schedule();
1725                         goto repeat;
1726                 }
1727         }
1728
1729 end:
1730         current->state = TASK_RUNNING;
1731         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1732         if (infop) {
1733                 if (retval > 0)
1734                         retval = 0;
1735                 else {
1736                         /*
1737                          * For a WNOHANG return, clear out all the fields
1738                          * we would set so the user can easily tell the
1739                          * difference.
1740                          */
1741                         if (!retval)
1742                                 retval = put_user(0, &infop->si_signo);
1743                         if (!retval)
1744                                 retval = put_user(0, &infop->si_errno);
1745                         if (!retval)
1746                                 retval = put_user(0, &infop->si_code);
1747                         if (!retval)
1748                                 retval = put_user(0, &infop->si_pid);
1749                         if (!retval)
1750                                 retval = put_user(0, &infop->si_uid);
1751                         if (!retval)
1752                                 retval = put_user(0, &infop->si_status);
1753                 }
1754         }
1755         return retval;
1756 }
1757
1758 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1759                 infop, int, options, struct rusage __user *, ru)
1760 {
1761         struct pid *pid = NULL;
1762         enum pid_type type;
1763         long ret;
1764
1765         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1766                 return -EINVAL;
1767         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1768                 return -EINVAL;
1769
1770         switch (which) {
1771         case P_ALL:
1772                 type = PIDTYPE_MAX;
1773                 break;
1774         case P_PID:
1775                 type = PIDTYPE_PID;
1776                 if (upid <= 0)
1777                         return -EINVAL;
1778                 break;
1779         case P_PGID:
1780                 type = PIDTYPE_PGID;
1781                 if (upid <= 0)
1782                         return -EINVAL;
1783                 break;
1784         default:
1785                 return -EINVAL;
1786         }
1787
1788         if (type < PIDTYPE_MAX)
1789                 pid = find_get_pid(upid);
1790         ret = do_wait(type, pid, options, infop, NULL, ru);
1791         put_pid(pid);
1792
1793         /* avoid REGPARM breakage on x86: */
1794         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1795         return ret;
1796 }
1797
1798 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1799                 int, options, struct rusage __user *, ru)
1800 {
1801         struct pid *pid = NULL;
1802         enum pid_type type;
1803         long ret;
1804
1805         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1806                         __WNOTHREAD|__WCLONE|__WALL))
1807                 return -EINVAL;
1808
1809         if (upid == -1)
1810                 type = PIDTYPE_MAX;
1811         else if (upid < 0) {
1812                 type = PIDTYPE_PGID;
1813                 pid = find_get_pid(-upid);
1814         } else if (upid == 0) {
1815                 type = PIDTYPE_PGID;
1816                 pid = get_pid(task_pgrp(current));
1817         } else /* upid > 0 */ {
1818                 type = PIDTYPE_PID;
1819                 pid = find_get_pid(upid);
1820         }
1821
1822         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1823         put_pid(pid);
1824
1825         /* avoid REGPARM breakage on x86: */
1826         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1827         return ret;
1828 }
1829
1830 #ifdef __ARCH_WANT_SYS_WAITPID
1831
1832 /*
1833  * sys_waitpid() remains for compatibility. waitpid() should be
1834  * implemented by calling sys_wait4() from libc.a.
1835  */
1836 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1837 {
1838         return sys_wait4(pid, stat_addr, options, NULL);
1839 }
1840
1841 #endif