[PATCH] namespaces: add nsproxy
[linux-3.10.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/smp_lock.h>
11 #include <linux/module.h>
12 #include <linux/capability.h>
13 #include <linux/completion.h>
14 #include <linux/personality.h>
15 #include <linux/tty.h>
16 #include <linux/namespace.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/binfmts.h>
24 #include <linux/nsproxy.h>
25 #include <linux/ptrace.h>
26 #include <linux/profile.h>
27 #include <linux/mount.h>
28 #include <linux/proc_fs.h>
29 #include <linux/mempolicy.h>
30 #include <linux/taskstats_kern.h>
31 #include <linux/delayacct.h>
32 #include <linux/cpuset.h>
33 #include <linux/syscalls.h>
34 #include <linux/signal.h>
35 #include <linux/posix-timers.h>
36 #include <linux/cn_proc.h>
37 #include <linux/mutex.h>
38 #include <linux/futex.h>
39 #include <linux/compat.h>
40 #include <linux/pipe_fs_i.h>
41 #include <linux/audit.h> /* for audit_free() */
42 #include <linux/resource.h>
43 #include <linux/blkdev.h>
44
45 #include <asm/uaccess.h>
46 #include <asm/unistd.h>
47 #include <asm/pgtable.h>
48 #include <asm/mmu_context.h>
49
50 extern void sem_exit (void);
51 extern struct task_struct *child_reaper;
52
53 static void exit_mm(struct task_struct * tsk);
54
55 static void __unhash_process(struct task_struct *p)
56 {
57         nr_threads--;
58         detach_pid(p, PIDTYPE_PID);
59         if (thread_group_leader(p)) {
60                 detach_pid(p, PIDTYPE_PGID);
61                 detach_pid(p, PIDTYPE_SID);
62
63                 list_del_rcu(&p->tasks);
64                 __get_cpu_var(process_counts)--;
65         }
66         list_del_rcu(&p->thread_group);
67         remove_parent(p);
68 }
69
70 /*
71  * This function expects the tasklist_lock write-locked.
72  */
73 static void __exit_signal(struct task_struct *tsk)
74 {
75         struct signal_struct *sig = tsk->signal;
76         struct sighand_struct *sighand;
77
78         BUG_ON(!sig);
79         BUG_ON(!atomic_read(&sig->count));
80
81         rcu_read_lock();
82         sighand = rcu_dereference(tsk->sighand);
83         spin_lock(&sighand->siglock);
84
85         posix_cpu_timers_exit(tsk);
86         if (atomic_dec_and_test(&sig->count))
87                 posix_cpu_timers_exit_group(tsk);
88         else {
89                 /*
90                  * If there is any task waiting for the group exit
91                  * then notify it:
92                  */
93                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
94                         wake_up_process(sig->group_exit_task);
95                         sig->group_exit_task = NULL;
96                 }
97                 if (tsk == sig->curr_target)
98                         sig->curr_target = next_thread(tsk);
99                 /*
100                  * Accumulate here the counters for all threads but the
101                  * group leader as they die, so they can be added into
102                  * the process-wide totals when those are taken.
103                  * The group leader stays around as a zombie as long
104                  * as there are other threads.  When it gets reaped,
105                  * the exit.c code will add its counts into these totals.
106                  * We won't ever get here for the group leader, since it
107                  * will have been the last reference on the signal_struct.
108                  */
109                 sig->utime = cputime_add(sig->utime, tsk->utime);
110                 sig->stime = cputime_add(sig->stime, tsk->stime);
111                 sig->min_flt += tsk->min_flt;
112                 sig->maj_flt += tsk->maj_flt;
113                 sig->nvcsw += tsk->nvcsw;
114                 sig->nivcsw += tsk->nivcsw;
115                 sig->sched_time += tsk->sched_time;
116                 sig = NULL; /* Marker for below. */
117         }
118
119         __unhash_process(tsk);
120
121         tsk->signal = NULL;
122         tsk->sighand = NULL;
123         spin_unlock(&sighand->siglock);
124         rcu_read_unlock();
125
126         __cleanup_sighand(sighand);
127         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
128         flush_sigqueue(&tsk->pending);
129         if (sig) {
130                 flush_sigqueue(&sig->shared_pending);
131                 __cleanup_signal(sig);
132         }
133 }
134
135 static void delayed_put_task_struct(struct rcu_head *rhp)
136 {
137         put_task_struct(container_of(rhp, struct task_struct, rcu));
138 }
139
140 void release_task(struct task_struct * p)
141 {
142         struct task_struct *leader;
143         int zap_leader;
144 repeat:
145         atomic_dec(&p->user->processes);
146         write_lock_irq(&tasklist_lock);
147         ptrace_unlink(p);
148         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
149         __exit_signal(p);
150
151         /*
152          * If we are the last non-leader member of the thread
153          * group, and the leader is zombie, then notify the
154          * group leader's parent process. (if it wants notification.)
155          */
156         zap_leader = 0;
157         leader = p->group_leader;
158         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
159                 BUG_ON(leader->exit_signal == -1);
160                 do_notify_parent(leader, leader->exit_signal);
161                 /*
162                  * If we were the last child thread and the leader has
163                  * exited already, and the leader's parent ignores SIGCHLD,
164                  * then we are the one who should release the leader.
165                  *
166                  * do_notify_parent() will have marked it self-reaping in
167                  * that case.
168                  */
169                 zap_leader = (leader->exit_signal == -1);
170         }
171
172         sched_exit(p);
173         write_unlock_irq(&tasklist_lock);
174         proc_flush_task(p);
175         release_thread(p);
176         call_rcu(&p->rcu, delayed_put_task_struct);
177
178         p = leader;
179         if (unlikely(zap_leader))
180                 goto repeat;
181 }
182
183 /*
184  * This checks not only the pgrp, but falls back on the pid if no
185  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
186  * without this...
187  */
188 int session_of_pgrp(int pgrp)
189 {
190         struct task_struct *p;
191         int sid = -1;
192
193         read_lock(&tasklist_lock);
194         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
195                 if (p->signal->session > 0) {
196                         sid = p->signal->session;
197                         goto out;
198                 }
199         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
200         p = find_task_by_pid(pgrp);
201         if (p)
202                 sid = p->signal->session;
203 out:
204         read_unlock(&tasklist_lock);
205         
206         return sid;
207 }
208
209 /*
210  * Determine if a process group is "orphaned", according to the POSIX
211  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
212  * by terminal-generated stop signals.  Newly orphaned process groups are
213  * to receive a SIGHUP and a SIGCONT.
214  *
215  * "I ask you, have you ever known what it is to be an orphan?"
216  */
217 static int will_become_orphaned_pgrp(int pgrp, struct task_struct *ignored_task)
218 {
219         struct task_struct *p;
220         int ret = 1;
221
222         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
223                 if (p == ignored_task
224                                 || p->exit_state
225                                 || is_init(p->real_parent))
226                         continue;
227                 if (process_group(p->real_parent) != pgrp
228                             && p->real_parent->signal->session == p->signal->session) {
229                         ret = 0;
230                         break;
231                 }
232         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
233         return ret;     /* (sighing) "Often!" */
234 }
235
236 int is_orphaned_pgrp(int pgrp)
237 {
238         int retval;
239
240         read_lock(&tasklist_lock);
241         retval = will_become_orphaned_pgrp(pgrp, NULL);
242         read_unlock(&tasklist_lock);
243
244         return retval;
245 }
246
247 static int has_stopped_jobs(int pgrp)
248 {
249         int retval = 0;
250         struct task_struct *p;
251
252         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
253                 if (p->state != TASK_STOPPED)
254                         continue;
255                 retval = 1;
256                 break;
257         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
258         return retval;
259 }
260
261 /**
262  * reparent_to_init - Reparent the calling kernel thread to the init task.
263  *
264  * If a kernel thread is launched as a result of a system call, or if
265  * it ever exits, it should generally reparent itself to init so that
266  * it is correctly cleaned up on exit.
267  *
268  * The various task state such as scheduling policy and priority may have
269  * been inherited from a user process, so we reset them to sane values here.
270  *
271  * NOTE that reparent_to_init() gives the caller full capabilities.
272  */
273 static void reparent_to_init(void)
274 {
275         write_lock_irq(&tasklist_lock);
276
277         ptrace_unlink(current);
278         /* Reparent to init */
279         remove_parent(current);
280         current->parent = child_reaper;
281         current->real_parent = child_reaper;
282         add_parent(current);
283
284         /* Set the exit signal to SIGCHLD so we signal init on exit */
285         current->exit_signal = SIGCHLD;
286
287         if (!has_rt_policy(current) && (task_nice(current) < 0))
288                 set_user_nice(current, 0);
289         /* cpus_allowed? */
290         /* rt_priority? */
291         /* signals? */
292         security_task_reparent_to_init(current);
293         memcpy(current->signal->rlim, init_task.signal->rlim,
294                sizeof(current->signal->rlim));
295         atomic_inc(&(INIT_USER->__count));
296         write_unlock_irq(&tasklist_lock);
297         switch_uid(INIT_USER);
298 }
299
300 void __set_special_pids(pid_t session, pid_t pgrp)
301 {
302         struct task_struct *curr = current->group_leader;
303
304         if (curr->signal->session != session) {
305                 detach_pid(curr, PIDTYPE_SID);
306                 curr->signal->session = session;
307                 attach_pid(curr, PIDTYPE_SID, session);
308         }
309         if (process_group(curr) != pgrp) {
310                 detach_pid(curr, PIDTYPE_PGID);
311                 curr->signal->pgrp = pgrp;
312                 attach_pid(curr, PIDTYPE_PGID, pgrp);
313         }
314 }
315
316 void set_special_pids(pid_t session, pid_t pgrp)
317 {
318         write_lock_irq(&tasklist_lock);
319         __set_special_pids(session, pgrp);
320         write_unlock_irq(&tasklist_lock);
321 }
322
323 /*
324  * Let kernel threads use this to say that they
325  * allow a certain signal (since daemonize() will
326  * have disabled all of them by default).
327  */
328 int allow_signal(int sig)
329 {
330         if (!valid_signal(sig) || sig < 1)
331                 return -EINVAL;
332
333         spin_lock_irq(&current->sighand->siglock);
334         sigdelset(&current->blocked, sig);
335         if (!current->mm) {
336                 /* Kernel threads handle their own signals.
337                    Let the signal code know it'll be handled, so
338                    that they don't get converted to SIGKILL or
339                    just silently dropped */
340                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
341         }
342         recalc_sigpending();
343         spin_unlock_irq(&current->sighand->siglock);
344         return 0;
345 }
346
347 EXPORT_SYMBOL(allow_signal);
348
349 int disallow_signal(int sig)
350 {
351         if (!valid_signal(sig) || sig < 1)
352                 return -EINVAL;
353
354         spin_lock_irq(&current->sighand->siglock);
355         sigaddset(&current->blocked, sig);
356         recalc_sigpending();
357         spin_unlock_irq(&current->sighand->siglock);
358         return 0;
359 }
360
361 EXPORT_SYMBOL(disallow_signal);
362
363 /*
364  *      Put all the gunge required to become a kernel thread without
365  *      attached user resources in one place where it belongs.
366  */
367
368 void daemonize(const char *name, ...)
369 {
370         va_list args;
371         struct fs_struct *fs;
372         sigset_t blocked;
373
374         va_start(args, name);
375         vsnprintf(current->comm, sizeof(current->comm), name, args);
376         va_end(args);
377
378         /*
379          * If we were started as result of loading a module, close all of the
380          * user space pages.  We don't need them, and if we didn't close them
381          * they would be locked into memory.
382          */
383         exit_mm(current);
384
385         set_special_pids(1, 1);
386         mutex_lock(&tty_mutex);
387         current->signal->tty = NULL;
388         mutex_unlock(&tty_mutex);
389
390         /* Block and flush all signals */
391         sigfillset(&blocked);
392         sigprocmask(SIG_BLOCK, &blocked, NULL);
393         flush_signals(current);
394
395         /* Become as one with the init task */
396
397         exit_fs(current);       /* current->fs->count--; */
398         fs = init_task.fs;
399         current->fs = fs;
400         atomic_inc(&fs->count);
401
402         exit_namespace(current);
403         exit_task_namespaces(current);
404         current->namespace = init_task.namespace;
405         current->nsproxy = init_task.nsproxy;
406         get_namespace(current->namespace);
407         get_task_namespaces(current);
408
409         exit_files(current);
410         current->files = init_task.files;
411         atomic_inc(&current->files->count);
412
413         reparent_to_init();
414 }
415
416 EXPORT_SYMBOL(daemonize);
417
418 static void close_files(struct files_struct * files)
419 {
420         int i, j;
421         struct fdtable *fdt;
422
423         j = 0;
424
425         /*
426          * It is safe to dereference the fd table without RCU or
427          * ->file_lock because this is the last reference to the
428          * files structure.
429          */
430         fdt = files_fdtable(files);
431         for (;;) {
432                 unsigned long set;
433                 i = j * __NFDBITS;
434                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
435                         break;
436                 set = fdt->open_fds->fds_bits[j++];
437                 while (set) {
438                         if (set & 1) {
439                                 struct file * file = xchg(&fdt->fd[i], NULL);
440                                 if (file)
441                                         filp_close(file, files);
442                         }
443                         i++;
444                         set >>= 1;
445                 }
446         }
447 }
448
449 struct files_struct *get_files_struct(struct task_struct *task)
450 {
451         struct files_struct *files;
452
453         task_lock(task);
454         files = task->files;
455         if (files)
456                 atomic_inc(&files->count);
457         task_unlock(task);
458
459         return files;
460 }
461
462 void fastcall put_files_struct(struct files_struct *files)
463 {
464         struct fdtable *fdt;
465
466         if (atomic_dec_and_test(&files->count)) {
467                 close_files(files);
468                 /*
469                  * Free the fd and fdset arrays if we expanded them.
470                  * If the fdtable was embedded, pass files for freeing
471                  * at the end of the RCU grace period. Otherwise,
472                  * you can free files immediately.
473                  */
474                 fdt = files_fdtable(files);
475                 if (fdt == &files->fdtab)
476                         fdt->free_files = files;
477                 else
478                         kmem_cache_free(files_cachep, files);
479                 free_fdtable(fdt);
480         }
481 }
482
483 EXPORT_SYMBOL(put_files_struct);
484
485 void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
486 {
487         struct files_struct *old;
488
489         old = tsk->files;
490         task_lock(tsk);
491         tsk->files = files;
492         task_unlock(tsk);
493         put_files_struct(old);
494 }
495 EXPORT_SYMBOL(reset_files_struct);
496
497 static inline void __exit_files(struct task_struct *tsk)
498 {
499         struct files_struct * files = tsk->files;
500
501         if (files) {
502                 task_lock(tsk);
503                 tsk->files = NULL;
504                 task_unlock(tsk);
505                 put_files_struct(files);
506         }
507 }
508
509 void exit_files(struct task_struct *tsk)
510 {
511         __exit_files(tsk);
512 }
513
514 static inline void __put_fs_struct(struct fs_struct *fs)
515 {
516         /* No need to hold fs->lock if we are killing it */
517         if (atomic_dec_and_test(&fs->count)) {
518                 dput(fs->root);
519                 mntput(fs->rootmnt);
520                 dput(fs->pwd);
521                 mntput(fs->pwdmnt);
522                 if (fs->altroot) {
523                         dput(fs->altroot);
524                         mntput(fs->altrootmnt);
525                 }
526                 kmem_cache_free(fs_cachep, fs);
527         }
528 }
529
530 void put_fs_struct(struct fs_struct *fs)
531 {
532         __put_fs_struct(fs);
533 }
534
535 static inline void __exit_fs(struct task_struct *tsk)
536 {
537         struct fs_struct * fs = tsk->fs;
538
539         if (fs) {
540                 task_lock(tsk);
541                 tsk->fs = NULL;
542                 task_unlock(tsk);
543                 __put_fs_struct(fs);
544         }
545 }
546
547 void exit_fs(struct task_struct *tsk)
548 {
549         __exit_fs(tsk);
550 }
551
552 EXPORT_SYMBOL_GPL(exit_fs);
553
554 /*
555  * Turn us into a lazy TLB process if we
556  * aren't already..
557  */
558 static void exit_mm(struct task_struct * tsk)
559 {
560         struct mm_struct *mm = tsk->mm;
561
562         mm_release(tsk, mm);
563         if (!mm)
564                 return;
565         /*
566          * Serialize with any possible pending coredump.
567          * We must hold mmap_sem around checking core_waiters
568          * and clearing tsk->mm.  The core-inducing thread
569          * will increment core_waiters for each thread in the
570          * group with ->mm != NULL.
571          */
572         down_read(&mm->mmap_sem);
573         if (mm->core_waiters) {
574                 up_read(&mm->mmap_sem);
575                 down_write(&mm->mmap_sem);
576                 if (!--mm->core_waiters)
577                         complete(mm->core_startup_done);
578                 up_write(&mm->mmap_sem);
579
580                 wait_for_completion(&mm->core_done);
581                 down_read(&mm->mmap_sem);
582         }
583         atomic_inc(&mm->mm_count);
584         BUG_ON(mm != tsk->active_mm);
585         /* more a memory barrier than a real lock */
586         task_lock(tsk);
587         tsk->mm = NULL;
588         up_read(&mm->mmap_sem);
589         enter_lazy_tlb(mm, current);
590         task_unlock(tsk);
591         mmput(mm);
592 }
593
594 static inline void
595 choose_new_parent(struct task_struct *p, struct task_struct *reaper)
596 {
597         /*
598          * Make sure we're not reparenting to ourselves and that
599          * the parent is not a zombie.
600          */
601         BUG_ON(p == reaper || reaper->exit_state);
602         p->real_parent = reaper;
603 }
604
605 static void
606 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
607 {
608         /* We don't want people slaying init.  */
609         if (p->exit_signal != -1)
610                 p->exit_signal = SIGCHLD;
611
612         if (p->pdeath_signal)
613                 /* We already hold the tasklist_lock here.  */
614                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
615
616         /* Move the child from its dying parent to the new one.  */
617         if (unlikely(traced)) {
618                 /* Preserve ptrace links if someone else is tracing this child.  */
619                 list_del_init(&p->ptrace_list);
620                 if (p->parent != p->real_parent)
621                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
622         } else {
623                 /* If this child is being traced, then we're the one tracing it
624                  * anyway, so let go of it.
625                  */
626                 p->ptrace = 0;
627                 remove_parent(p);
628                 p->parent = p->real_parent;
629                 add_parent(p);
630
631                 /* If we'd notified the old parent about this child's death,
632                  * also notify the new parent.
633                  */
634                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
635                     thread_group_empty(p))
636                         do_notify_parent(p, p->exit_signal);
637                 else if (p->state == TASK_TRACED) {
638                         /*
639                          * If it was at a trace stop, turn it into
640                          * a normal stop since it's no longer being
641                          * traced.
642                          */
643                         ptrace_untrace(p);
644                 }
645         }
646
647         /*
648          * process group orphan check
649          * Case ii: Our child is in a different pgrp
650          * than we are, and it was the only connection
651          * outside, so the child pgrp is now orphaned.
652          */
653         if ((process_group(p) != process_group(father)) &&
654             (p->signal->session == father->signal->session)) {
655                 int pgrp = process_group(p);
656
657                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
658                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
659                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
660                 }
661         }
662 }
663
664 /*
665  * When we die, we re-parent all our children.
666  * Try to give them to another thread in our thread
667  * group, and if no such member exists, give it to
668  * the global child reaper process (ie "init")
669  */
670 static void
671 forget_original_parent(struct task_struct *father, struct list_head *to_release)
672 {
673         struct task_struct *p, *reaper = father;
674         struct list_head *_p, *_n;
675
676         do {
677                 reaper = next_thread(reaper);
678                 if (reaper == father) {
679                         reaper = child_reaper;
680                         break;
681                 }
682         } while (reaper->exit_state);
683
684         /*
685          * There are only two places where our children can be:
686          *
687          * - in our child list
688          * - in our ptraced child list
689          *
690          * Search them and reparent children.
691          */
692         list_for_each_safe(_p, _n, &father->children) {
693                 int ptrace;
694                 p = list_entry(_p, struct task_struct, sibling);
695
696                 ptrace = p->ptrace;
697
698                 /* if father isn't the real parent, then ptrace must be enabled */
699                 BUG_ON(father != p->real_parent && !ptrace);
700
701                 if (father == p->real_parent) {
702                         /* reparent with a reaper, real father it's us */
703                         choose_new_parent(p, reaper);
704                         reparent_thread(p, father, 0);
705                 } else {
706                         /* reparent ptraced task to its real parent */
707                         __ptrace_unlink (p);
708                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
709                             thread_group_empty(p))
710                                 do_notify_parent(p, p->exit_signal);
711                 }
712
713                 /*
714                  * if the ptraced child is a zombie with exit_signal == -1
715                  * we must collect it before we exit, or it will remain
716                  * zombie forever since we prevented it from self-reap itself
717                  * while it was being traced by us, to be able to see it in wait4.
718                  */
719                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
720                         list_add(&p->ptrace_list, to_release);
721         }
722         list_for_each_safe(_p, _n, &father->ptrace_children) {
723                 p = list_entry(_p, struct task_struct, ptrace_list);
724                 choose_new_parent(p, reaper);
725                 reparent_thread(p, father, 1);
726         }
727 }
728
729 /*
730  * Send signals to all our closest relatives so that they know
731  * to properly mourn us..
732  */
733 static void exit_notify(struct task_struct *tsk)
734 {
735         int state;
736         struct task_struct *t;
737         struct list_head ptrace_dead, *_p, *_n;
738
739         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
740             && !thread_group_empty(tsk)) {
741                 /*
742                  * This occurs when there was a race between our exit
743                  * syscall and a group signal choosing us as the one to
744                  * wake up.  It could be that we are the only thread
745                  * alerted to check for pending signals, but another thread
746                  * should be woken now to take the signal since we will not.
747                  * Now we'll wake all the threads in the group just to make
748                  * sure someone gets all the pending signals.
749                  */
750                 read_lock(&tasklist_lock);
751                 spin_lock_irq(&tsk->sighand->siglock);
752                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
753                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
754                                 recalc_sigpending_tsk(t);
755                                 if (signal_pending(t))
756                                         signal_wake_up(t, 0);
757                         }
758                 spin_unlock_irq(&tsk->sighand->siglock);
759                 read_unlock(&tasklist_lock);
760         }
761
762         write_lock_irq(&tasklist_lock);
763
764         /*
765          * This does two things:
766          *
767          * A.  Make init inherit all the child processes
768          * B.  Check to see if any process groups have become orphaned
769          *      as a result of our exiting, and if they have any stopped
770          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
771          */
772
773         INIT_LIST_HEAD(&ptrace_dead);
774         forget_original_parent(tsk, &ptrace_dead);
775         BUG_ON(!list_empty(&tsk->children));
776         BUG_ON(!list_empty(&tsk->ptrace_children));
777
778         /*
779          * Check to see if any process groups have become orphaned
780          * as a result of our exiting, and if they have any stopped
781          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
782          *
783          * Case i: Our father is in a different pgrp than we are
784          * and we were the only connection outside, so our pgrp
785          * is about to become orphaned.
786          */
787          
788         t = tsk->real_parent;
789         
790         if ((process_group(t) != process_group(tsk)) &&
791             (t->signal->session == tsk->signal->session) &&
792             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
793             has_stopped_jobs(process_group(tsk))) {
794                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
795                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
796         }
797
798         /* Let father know we died 
799          *
800          * Thread signals are configurable, but you aren't going to use
801          * that to send signals to arbitary processes. 
802          * That stops right now.
803          *
804          * If the parent exec id doesn't match the exec id we saved
805          * when we started then we know the parent has changed security
806          * domain.
807          *
808          * If our self_exec id doesn't match our parent_exec_id then
809          * we have changed execution domain as these two values started
810          * the same after a fork.
811          *      
812          */
813         
814         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
815             ( tsk->parent_exec_id != t->self_exec_id  ||
816               tsk->self_exec_id != tsk->parent_exec_id)
817             && !capable(CAP_KILL))
818                 tsk->exit_signal = SIGCHLD;
819
820
821         /* If something other than our normal parent is ptracing us, then
822          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
823          * only has special meaning to our real parent.
824          */
825         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
826                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
827                 do_notify_parent(tsk, signal);
828         } else if (tsk->ptrace) {
829                 do_notify_parent(tsk, SIGCHLD);
830         }
831
832         state = EXIT_ZOMBIE;
833         if (tsk->exit_signal == -1 &&
834             (likely(tsk->ptrace == 0) ||
835              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
836                 state = EXIT_DEAD;
837         tsk->exit_state = state;
838
839         write_unlock_irq(&tasklist_lock);
840
841         list_for_each_safe(_p, _n, &ptrace_dead) {
842                 list_del_init(_p);
843                 t = list_entry(_p, struct task_struct, ptrace_list);
844                 release_task(t);
845         }
846
847         /* If the process is dead, release it - nobody will wait for it */
848         if (state == EXIT_DEAD)
849                 release_task(tsk);
850 }
851
852 fastcall NORET_TYPE void do_exit(long code)
853 {
854         struct task_struct *tsk = current;
855         struct taskstats *tidstats;
856         int group_dead;
857         unsigned int mycpu;
858
859         profile_task_exit(tsk);
860
861         WARN_ON(atomic_read(&tsk->fs_excl));
862
863         if (unlikely(in_interrupt()))
864                 panic("Aiee, killing interrupt handler!");
865         if (unlikely(!tsk->pid))
866                 panic("Attempted to kill the idle task!");
867         if (unlikely(tsk == child_reaper))
868                 panic("Attempted to kill init!");
869
870         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
871                 current->ptrace_message = code;
872                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
873         }
874
875         /*
876          * We're taking recursive faults here in do_exit. Safest is to just
877          * leave this task alone and wait for reboot.
878          */
879         if (unlikely(tsk->flags & PF_EXITING)) {
880                 printk(KERN_ALERT
881                         "Fixing recursive fault but reboot is needed!\n");
882                 if (tsk->io_context)
883                         exit_io_context();
884                 set_current_state(TASK_UNINTERRUPTIBLE);
885                 schedule();
886         }
887
888         tsk->flags |= PF_EXITING;
889
890         if (unlikely(in_atomic()))
891                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
892                                 current->comm, current->pid,
893                                 preempt_count());
894
895         taskstats_exit_alloc(&tidstats, &mycpu);
896
897         acct_update_integrals(tsk);
898         if (tsk->mm) {
899                 update_hiwater_rss(tsk->mm);
900                 update_hiwater_vm(tsk->mm);
901         }
902         group_dead = atomic_dec_and_test(&tsk->signal->live);
903         if (group_dead) {
904                 hrtimer_cancel(&tsk->signal->real_timer);
905                 exit_itimers(tsk->signal);
906         }
907         acct_collect(code, group_dead);
908         if (unlikely(tsk->robust_list))
909                 exit_robust_list(tsk);
910 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
911         if (unlikely(tsk->compat_robust_list))
912                 compat_exit_robust_list(tsk);
913 #endif
914         if (unlikely(tsk->audit_context))
915                 audit_free(tsk);
916         taskstats_exit_send(tsk, tidstats, group_dead, mycpu);
917         taskstats_exit_free(tidstats);
918
919         exit_mm(tsk);
920
921         if (group_dead)
922                 acct_process();
923         exit_sem(tsk);
924         __exit_files(tsk);
925         __exit_fs(tsk);
926         exit_namespace(tsk);
927         exit_task_namespaces(tsk);
928         exit_thread();
929         cpuset_exit(tsk);
930         exit_keys(tsk);
931
932         if (group_dead && tsk->signal->leader)
933                 disassociate_ctty(1);
934
935         module_put(task_thread_info(tsk)->exec_domain->module);
936         if (tsk->binfmt)
937                 module_put(tsk->binfmt->module);
938
939         tsk->exit_code = code;
940         proc_exit_connector(tsk);
941         exit_notify(tsk);
942 #ifdef CONFIG_NUMA
943         mpol_free(tsk->mempolicy);
944         tsk->mempolicy = NULL;
945 #endif
946         /*
947          * This must happen late, after the PID is not
948          * hashed anymore:
949          */
950         if (unlikely(!list_empty(&tsk->pi_state_list)))
951                 exit_pi_state_list(tsk);
952         if (unlikely(current->pi_state_cache))
953                 kfree(current->pi_state_cache);
954         /*
955          * Make sure we are holding no locks:
956          */
957         debug_check_no_locks_held(tsk);
958
959         if (tsk->io_context)
960                 exit_io_context();
961
962         if (tsk->splice_pipe)
963                 __free_pipe_info(tsk->splice_pipe);
964
965         preempt_disable();
966         /* causes final put_task_struct in finish_task_switch(). */
967         tsk->state = TASK_DEAD;
968
969         schedule();
970         BUG();
971         /* Avoid "noreturn function does return".  */
972         for (;;)
973                 cpu_relax();    /* For when BUG is null */
974 }
975
976 EXPORT_SYMBOL_GPL(do_exit);
977
978 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
979 {
980         if (comp)
981                 complete(comp);
982
983         do_exit(code);
984 }
985
986 EXPORT_SYMBOL(complete_and_exit);
987
988 asmlinkage long sys_exit(int error_code)
989 {
990         do_exit((error_code&0xff)<<8);
991 }
992
993 /*
994  * Take down every thread in the group.  This is called by fatal signals
995  * as well as by sys_exit_group (below).
996  */
997 NORET_TYPE void
998 do_group_exit(int exit_code)
999 {
1000         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1001
1002         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1003                 exit_code = current->signal->group_exit_code;
1004         else if (!thread_group_empty(current)) {
1005                 struct signal_struct *const sig = current->signal;
1006                 struct sighand_struct *const sighand = current->sighand;
1007                 spin_lock_irq(&sighand->siglock);
1008                 if (sig->flags & SIGNAL_GROUP_EXIT)
1009                         /* Another thread got here before we took the lock.  */
1010                         exit_code = sig->group_exit_code;
1011                 else {
1012                         sig->group_exit_code = exit_code;
1013                         zap_other_threads(current);
1014                 }
1015                 spin_unlock_irq(&sighand->siglock);
1016         }
1017
1018         do_exit(exit_code);
1019         /* NOTREACHED */
1020 }
1021
1022 /*
1023  * this kills every thread in the thread group. Note that any externally
1024  * wait4()-ing process will get the correct exit code - even if this
1025  * thread is not the thread group leader.
1026  */
1027 asmlinkage void sys_exit_group(int error_code)
1028 {
1029         do_group_exit((error_code & 0xff) << 8);
1030 }
1031
1032 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1033 {
1034         if (pid > 0) {
1035                 if (p->pid != pid)
1036                         return 0;
1037         } else if (!pid) {
1038                 if (process_group(p) != process_group(current))
1039                         return 0;
1040         } else if (pid != -1) {
1041                 if (process_group(p) != -pid)
1042                         return 0;
1043         }
1044
1045         /*
1046          * Do not consider detached threads that are
1047          * not ptraced:
1048          */
1049         if (p->exit_signal == -1 && !p->ptrace)
1050                 return 0;
1051
1052         /* Wait for all children (clone and not) if __WALL is set;
1053          * otherwise, wait for clone children *only* if __WCLONE is
1054          * set; otherwise, wait for non-clone children *only*.  (Note:
1055          * A "clone" child here is one that reports to its parent
1056          * using a signal other than SIGCHLD.) */
1057         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1058             && !(options & __WALL))
1059                 return 0;
1060         /*
1061          * Do not consider thread group leaders that are
1062          * in a non-empty thread group:
1063          */
1064         if (delay_group_leader(p))
1065                 return 2;
1066
1067         if (security_task_wait(p))
1068                 return 0;
1069
1070         return 1;
1071 }
1072
1073 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1074                                int why, int status,
1075                                struct siginfo __user *infop,
1076                                struct rusage __user *rusagep)
1077 {
1078         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1079
1080         put_task_struct(p);
1081         if (!retval)
1082                 retval = put_user(SIGCHLD, &infop->si_signo);
1083         if (!retval)
1084                 retval = put_user(0, &infop->si_errno);
1085         if (!retval)
1086                 retval = put_user((short)why, &infop->si_code);
1087         if (!retval)
1088                 retval = put_user(pid, &infop->si_pid);
1089         if (!retval)
1090                 retval = put_user(uid, &infop->si_uid);
1091         if (!retval)
1092                 retval = put_user(status, &infop->si_status);
1093         if (!retval)
1094                 retval = pid;
1095         return retval;
1096 }
1097
1098 /*
1099  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1100  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1101  * the lock and this task is uninteresting.  If we return nonzero, we have
1102  * released the lock and the system call should return.
1103  */
1104 static int wait_task_zombie(struct task_struct *p, int noreap,
1105                             struct siginfo __user *infop,
1106                             int __user *stat_addr, struct rusage __user *ru)
1107 {
1108         unsigned long state;
1109         int retval;
1110         int status;
1111
1112         if (unlikely(noreap)) {
1113                 pid_t pid = p->pid;
1114                 uid_t uid = p->uid;
1115                 int exit_code = p->exit_code;
1116                 int why, status;
1117
1118                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1119                         return 0;
1120                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1121                         return 0;
1122                 get_task_struct(p);
1123                 read_unlock(&tasklist_lock);
1124                 if ((exit_code & 0x7f) == 0) {
1125                         why = CLD_EXITED;
1126                         status = exit_code >> 8;
1127                 } else {
1128                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1129                         status = exit_code & 0x7f;
1130                 }
1131                 return wait_noreap_copyout(p, pid, uid, why,
1132                                            status, infop, ru);
1133         }
1134
1135         /*
1136          * Try to move the task's state to DEAD
1137          * only one thread is allowed to do this:
1138          */
1139         state = xchg(&p->exit_state, EXIT_DEAD);
1140         if (state != EXIT_ZOMBIE) {
1141                 BUG_ON(state != EXIT_DEAD);
1142                 return 0;
1143         }
1144         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1145                 /*
1146                  * This can only happen in a race with a ptraced thread
1147                  * dying on another processor.
1148                  */
1149                 return 0;
1150         }
1151
1152         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1153                 struct signal_struct *psig;
1154                 struct signal_struct *sig;
1155
1156                 /*
1157                  * The resource counters for the group leader are in its
1158                  * own task_struct.  Those for dead threads in the group
1159                  * are in its signal_struct, as are those for the child
1160                  * processes it has previously reaped.  All these
1161                  * accumulate in the parent's signal_struct c* fields.
1162                  *
1163                  * We don't bother to take a lock here to protect these
1164                  * p->signal fields, because they are only touched by
1165                  * __exit_signal, which runs with tasklist_lock
1166                  * write-locked anyway, and so is excluded here.  We do
1167                  * need to protect the access to p->parent->signal fields,
1168                  * as other threads in the parent group can be right
1169                  * here reaping other children at the same time.
1170                  */
1171                 spin_lock_irq(&p->parent->sighand->siglock);
1172                 psig = p->parent->signal;
1173                 sig = p->signal;
1174                 psig->cutime =
1175                         cputime_add(psig->cutime,
1176                         cputime_add(p->utime,
1177                         cputime_add(sig->utime,
1178                                     sig->cutime)));
1179                 psig->cstime =
1180                         cputime_add(psig->cstime,
1181                         cputime_add(p->stime,
1182                         cputime_add(sig->stime,
1183                                     sig->cstime)));
1184                 psig->cmin_flt +=
1185                         p->min_flt + sig->min_flt + sig->cmin_flt;
1186                 psig->cmaj_flt +=
1187                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1188                 psig->cnvcsw +=
1189                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1190                 psig->cnivcsw +=
1191                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1192                 spin_unlock_irq(&p->parent->sighand->siglock);
1193         }
1194
1195         /*
1196          * Now we are sure this task is interesting, and no other
1197          * thread can reap it because we set its state to EXIT_DEAD.
1198          */
1199         read_unlock(&tasklist_lock);
1200
1201         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1202         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1203                 ? p->signal->group_exit_code : p->exit_code;
1204         if (!retval && stat_addr)
1205                 retval = put_user(status, stat_addr);
1206         if (!retval && infop)
1207                 retval = put_user(SIGCHLD, &infop->si_signo);
1208         if (!retval && infop)
1209                 retval = put_user(0, &infop->si_errno);
1210         if (!retval && infop) {
1211                 int why;
1212
1213                 if ((status & 0x7f) == 0) {
1214                         why = CLD_EXITED;
1215                         status >>= 8;
1216                 } else {
1217                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1218                         status &= 0x7f;
1219                 }
1220                 retval = put_user((short)why, &infop->si_code);
1221                 if (!retval)
1222                         retval = put_user(status, &infop->si_status);
1223         }
1224         if (!retval && infop)
1225                 retval = put_user(p->pid, &infop->si_pid);
1226         if (!retval && infop)
1227                 retval = put_user(p->uid, &infop->si_uid);
1228         if (retval) {
1229                 // TODO: is this safe?
1230                 p->exit_state = EXIT_ZOMBIE;
1231                 return retval;
1232         }
1233         retval = p->pid;
1234         if (p->real_parent != p->parent) {
1235                 write_lock_irq(&tasklist_lock);
1236                 /* Double-check with lock held.  */
1237                 if (p->real_parent != p->parent) {
1238                         __ptrace_unlink(p);
1239                         // TODO: is this safe?
1240                         p->exit_state = EXIT_ZOMBIE;
1241                         /*
1242                          * If this is not a detached task, notify the parent.
1243                          * If it's still not detached after that, don't release
1244                          * it now.
1245                          */
1246                         if (p->exit_signal != -1) {
1247                                 do_notify_parent(p, p->exit_signal);
1248                                 if (p->exit_signal != -1)
1249                                         p = NULL;
1250                         }
1251                 }
1252                 write_unlock_irq(&tasklist_lock);
1253         }
1254         if (p != NULL)
1255                 release_task(p);
1256         BUG_ON(!retval);
1257         return retval;
1258 }
1259
1260 /*
1261  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1262  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1263  * the lock and this task is uninteresting.  If we return nonzero, we have
1264  * released the lock and the system call should return.
1265  */
1266 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1267                              int noreap, struct siginfo __user *infop,
1268                              int __user *stat_addr, struct rusage __user *ru)
1269 {
1270         int retval, exit_code;
1271
1272         if (!p->exit_code)
1273                 return 0;
1274         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1275             p->signal && p->signal->group_stop_count > 0)
1276                 /*
1277                  * A group stop is in progress and this is the group leader.
1278                  * We won't report until all threads have stopped.
1279                  */
1280                 return 0;
1281
1282         /*
1283          * Now we are pretty sure this task is interesting.
1284          * Make sure it doesn't get reaped out from under us while we
1285          * give up the lock and then examine it below.  We don't want to
1286          * keep holding onto the tasklist_lock while we call getrusage and
1287          * possibly take page faults for user memory.
1288          */
1289         get_task_struct(p);
1290         read_unlock(&tasklist_lock);
1291
1292         if (unlikely(noreap)) {
1293                 pid_t pid = p->pid;
1294                 uid_t uid = p->uid;
1295                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1296
1297                 exit_code = p->exit_code;
1298                 if (unlikely(!exit_code) ||
1299                     unlikely(p->state & TASK_TRACED))
1300                         goto bail_ref;
1301                 return wait_noreap_copyout(p, pid, uid,
1302                                            why, (exit_code << 8) | 0x7f,
1303                                            infop, ru);
1304         }
1305
1306         write_lock_irq(&tasklist_lock);
1307
1308         /*
1309          * This uses xchg to be atomic with the thread resuming and setting
1310          * it.  It must also be done with the write lock held to prevent a
1311          * race with the EXIT_ZOMBIE case.
1312          */
1313         exit_code = xchg(&p->exit_code, 0);
1314         if (unlikely(p->exit_state)) {
1315                 /*
1316                  * The task resumed and then died.  Let the next iteration
1317                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1318                  * already be zero here if it resumed and did _exit(0).
1319                  * The task itself is dead and won't touch exit_code again;
1320                  * other processors in this function are locked out.
1321                  */
1322                 p->exit_code = exit_code;
1323                 exit_code = 0;
1324         }
1325         if (unlikely(exit_code == 0)) {
1326                 /*
1327                  * Another thread in this function got to it first, or it
1328                  * resumed, or it resumed and then died.
1329                  */
1330                 write_unlock_irq(&tasklist_lock);
1331 bail_ref:
1332                 put_task_struct(p);
1333                 /*
1334                  * We are returning to the wait loop without having successfully
1335                  * removed the process and having released the lock. We cannot
1336                  * continue, since the "p" task pointer is potentially stale.
1337                  *
1338                  * Return -EAGAIN, and do_wait() will restart the loop from the
1339                  * beginning. Do _not_ re-acquire the lock.
1340                  */
1341                 return -EAGAIN;
1342         }
1343
1344         /* move to end of parent's list to avoid starvation */
1345         remove_parent(p);
1346         add_parent(p);
1347
1348         write_unlock_irq(&tasklist_lock);
1349
1350         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1351         if (!retval && stat_addr)
1352                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1353         if (!retval && infop)
1354                 retval = put_user(SIGCHLD, &infop->si_signo);
1355         if (!retval && infop)
1356                 retval = put_user(0, &infop->si_errno);
1357         if (!retval && infop)
1358                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1359                                           ? CLD_TRAPPED : CLD_STOPPED),
1360                                   &infop->si_code);
1361         if (!retval && infop)
1362                 retval = put_user(exit_code, &infop->si_status);
1363         if (!retval && infop)
1364                 retval = put_user(p->pid, &infop->si_pid);
1365         if (!retval && infop)
1366                 retval = put_user(p->uid, &infop->si_uid);
1367         if (!retval)
1368                 retval = p->pid;
1369         put_task_struct(p);
1370
1371         BUG_ON(!retval);
1372         return retval;
1373 }
1374
1375 /*
1376  * Handle do_wait work for one task in a live, non-stopped state.
1377  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1378  * the lock and this task is uninteresting.  If we return nonzero, we have
1379  * released the lock and the system call should return.
1380  */
1381 static int wait_task_continued(struct task_struct *p, int noreap,
1382                                struct siginfo __user *infop,
1383                                int __user *stat_addr, struct rusage __user *ru)
1384 {
1385         int retval;
1386         pid_t pid;
1387         uid_t uid;
1388
1389         if (unlikely(!p->signal))
1390                 return 0;
1391
1392         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1393                 return 0;
1394
1395         spin_lock_irq(&p->sighand->siglock);
1396         /* Re-check with the lock held.  */
1397         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1398                 spin_unlock_irq(&p->sighand->siglock);
1399                 return 0;
1400         }
1401         if (!noreap)
1402                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1403         spin_unlock_irq(&p->sighand->siglock);
1404
1405         pid = p->pid;
1406         uid = p->uid;
1407         get_task_struct(p);
1408         read_unlock(&tasklist_lock);
1409
1410         if (!infop) {
1411                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1412                 put_task_struct(p);
1413                 if (!retval && stat_addr)
1414                         retval = put_user(0xffff, stat_addr);
1415                 if (!retval)
1416                         retval = p->pid;
1417         } else {
1418                 retval = wait_noreap_copyout(p, pid, uid,
1419                                              CLD_CONTINUED, SIGCONT,
1420                                              infop, ru);
1421                 BUG_ON(retval == 0);
1422         }
1423
1424         return retval;
1425 }
1426
1427
1428 static inline int my_ptrace_child(struct task_struct *p)
1429 {
1430         if (!(p->ptrace & PT_PTRACED))
1431                 return 0;
1432         if (!(p->ptrace & PT_ATTACHED))
1433                 return 1;
1434         /*
1435          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1436          * we are the attacher.  If we are the real parent, this is a race
1437          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1438          * which we have to switch the parent links, but has already set
1439          * the flags in p->ptrace.
1440          */
1441         return (p->parent != p->real_parent);
1442 }
1443
1444 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1445                     int __user *stat_addr, struct rusage __user *ru)
1446 {
1447         DECLARE_WAITQUEUE(wait, current);
1448         struct task_struct *tsk;
1449         int flag, retval;
1450
1451         add_wait_queue(&current->signal->wait_chldexit,&wait);
1452 repeat:
1453         /*
1454          * We will set this flag if we see any child that might later
1455          * match our criteria, even if we are not able to reap it yet.
1456          */
1457         flag = 0;
1458         current->state = TASK_INTERRUPTIBLE;
1459         read_lock(&tasklist_lock);
1460         tsk = current;
1461         do {
1462                 struct task_struct *p;
1463                 struct list_head *_p;
1464                 int ret;
1465
1466                 list_for_each(_p,&tsk->children) {
1467                         p = list_entry(_p, struct task_struct, sibling);
1468
1469                         ret = eligible_child(pid, options, p);
1470                         if (!ret)
1471                                 continue;
1472
1473                         switch (p->state) {
1474                         case TASK_TRACED:
1475                                 /*
1476                                  * When we hit the race with PTRACE_ATTACH,
1477                                  * we will not report this child.  But the
1478                                  * race means it has not yet been moved to
1479                                  * our ptrace_children list, so we need to
1480                                  * set the flag here to avoid a spurious ECHILD
1481                                  * when the race happens with the only child.
1482                                  */
1483                                 flag = 1;
1484                                 if (!my_ptrace_child(p))
1485                                         continue;
1486                                 /*FALLTHROUGH*/
1487                         case TASK_STOPPED:
1488                                 /*
1489                                  * It's stopped now, so it might later
1490                                  * continue, exit, or stop again.
1491                                  */
1492                                 flag = 1;
1493                                 if (!(options & WUNTRACED) &&
1494                                     !my_ptrace_child(p))
1495                                         continue;
1496                                 retval = wait_task_stopped(p, ret == 2,
1497                                                            (options & WNOWAIT),
1498                                                            infop,
1499                                                            stat_addr, ru);
1500                                 if (retval == -EAGAIN)
1501                                         goto repeat;
1502                                 if (retval != 0) /* He released the lock.  */
1503                                         goto end;
1504                                 break;
1505                         default:
1506                         // case EXIT_DEAD:
1507                                 if (p->exit_state == EXIT_DEAD)
1508                                         continue;
1509                         // case EXIT_ZOMBIE:
1510                                 if (p->exit_state == EXIT_ZOMBIE) {
1511                                         /*
1512                                          * Eligible but we cannot release
1513                                          * it yet:
1514                                          */
1515                                         if (ret == 2)
1516                                                 goto check_continued;
1517                                         if (!likely(options & WEXITED))
1518                                                 continue;
1519                                         retval = wait_task_zombie(
1520                                                 p, (options & WNOWAIT),
1521                                                 infop, stat_addr, ru);
1522                                         /* He released the lock.  */
1523                                         if (retval != 0)
1524                                                 goto end;
1525                                         break;
1526                                 }
1527 check_continued:
1528                                 /*
1529                                  * It's running now, so it might later
1530                                  * exit, stop, or stop and then continue.
1531                                  */
1532                                 flag = 1;
1533                                 if (!unlikely(options & WCONTINUED))
1534                                         continue;
1535                                 retval = wait_task_continued(
1536                                         p, (options & WNOWAIT),
1537                                         infop, stat_addr, ru);
1538                                 if (retval != 0) /* He released the lock.  */
1539                                         goto end;
1540                                 break;
1541                         }
1542                 }
1543                 if (!flag) {
1544                         list_for_each(_p, &tsk->ptrace_children) {
1545                                 p = list_entry(_p, struct task_struct,
1546                                                 ptrace_list);
1547                                 if (!eligible_child(pid, options, p))
1548                                         continue;
1549                                 flag = 1;
1550                                 break;
1551                         }
1552                 }
1553                 if (options & __WNOTHREAD)
1554                         break;
1555                 tsk = next_thread(tsk);
1556                 BUG_ON(tsk->signal != current->signal);
1557         } while (tsk != current);
1558
1559         read_unlock(&tasklist_lock);
1560         if (flag) {
1561                 retval = 0;
1562                 if (options & WNOHANG)
1563                         goto end;
1564                 retval = -ERESTARTSYS;
1565                 if (signal_pending(current))
1566                         goto end;
1567                 schedule();
1568                 goto repeat;
1569         }
1570         retval = -ECHILD;
1571 end:
1572         current->state = TASK_RUNNING;
1573         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1574         if (infop) {
1575                 if (retval > 0)
1576                 retval = 0;
1577                 else {
1578                         /*
1579                          * For a WNOHANG return, clear out all the fields
1580                          * we would set so the user can easily tell the
1581                          * difference.
1582                          */
1583                         if (!retval)
1584                                 retval = put_user(0, &infop->si_signo);
1585                         if (!retval)
1586                                 retval = put_user(0, &infop->si_errno);
1587                         if (!retval)
1588                                 retval = put_user(0, &infop->si_code);
1589                         if (!retval)
1590                                 retval = put_user(0, &infop->si_pid);
1591                         if (!retval)
1592                                 retval = put_user(0, &infop->si_uid);
1593                         if (!retval)
1594                                 retval = put_user(0, &infop->si_status);
1595                 }
1596         }
1597         return retval;
1598 }
1599
1600 asmlinkage long sys_waitid(int which, pid_t pid,
1601                            struct siginfo __user *infop, int options,
1602                            struct rusage __user *ru)
1603 {
1604         long ret;
1605
1606         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1607                 return -EINVAL;
1608         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1609                 return -EINVAL;
1610
1611         switch (which) {
1612         case P_ALL:
1613                 pid = -1;
1614                 break;
1615         case P_PID:
1616                 if (pid <= 0)
1617                         return -EINVAL;
1618                 break;
1619         case P_PGID:
1620                 if (pid <= 0)
1621                         return -EINVAL;
1622                 pid = -pid;
1623                 break;
1624         default:
1625                 return -EINVAL;
1626         }
1627
1628         ret = do_wait(pid, options, infop, NULL, ru);
1629
1630         /* avoid REGPARM breakage on x86: */
1631         prevent_tail_call(ret);
1632         return ret;
1633 }
1634
1635 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1636                           int options, struct rusage __user *ru)
1637 {
1638         long ret;
1639
1640         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1641                         __WNOTHREAD|__WCLONE|__WALL))
1642                 return -EINVAL;
1643         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1644
1645         /* avoid REGPARM breakage on x86: */
1646         prevent_tail_call(ret);
1647         return ret;
1648 }
1649
1650 #ifdef __ARCH_WANT_SYS_WAITPID
1651
1652 /*
1653  * sys_waitpid() remains for compatibility. waitpid() should be
1654  * implemented by calling sys_wait4() from libc.a.
1655  */
1656 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1657 {
1658         return sys_wait4(pid, stat_addr, options, NULL);
1659 }
1660
1661 #endif