Revert "[PATCH] namespaces: fix exit race by splitting exit"
[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/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/mnt_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/pid_namespace.h>
26 #include <linux/ptrace.h>
27 #include <linux/profile.h>
28 #include <linux/mount.h>
29 #include <linux/proc_fs.h>
30 #include <linux/mempolicy.h>
31 #include <linux/taskstats_kern.h>
32 #include <linux/delayacct.h>
33 #include <linux/cpuset.h>
34 #include <linux/syscalls.h>
35 #include <linux/signal.h>
36 #include <linux/posix-timers.h>
37 #include <linux/cn_proc.h>
38 #include <linux/mutex.h>
39 #include <linux/futex.h>
40 #include <linux/compat.h>
41 #include <linux/pipe_fs_i.h>
42 #include <linux/audit.h> /* for audit_free() */
43 #include <linux/resource.h>
44 #include <linux/blkdev.h>
45
46 #include <asm/uaccess.h>
47 #include <asm/unistd.h>
48 #include <asm/pgtable.h>
49 #include <asm/mmu_context.h>
50
51 extern void sem_exit (void);
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                 taskstats_tgid_free(sig);
132                 __cleanup_signal(sig);
133         }
134 }
135
136 static void delayed_put_task_struct(struct rcu_head *rhp)
137 {
138         put_task_struct(container_of(rhp, struct task_struct, rcu));
139 }
140
141 void release_task(struct task_struct * p)
142 {
143         struct task_struct *leader;
144         int zap_leader;
145 repeat:
146         atomic_dec(&p->user->processes);
147         write_lock_irq(&tasklist_lock);
148         ptrace_unlink(p);
149         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
150         __exit_signal(p);
151
152         /*
153          * If we are the last non-leader member of the thread
154          * group, and the leader is zombie, then notify the
155          * group leader's parent process. (if it wants notification.)
156          */
157         zap_leader = 0;
158         leader = p->group_leader;
159         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
160                 BUG_ON(leader->exit_signal == -1);
161                 do_notify_parent(leader, leader->exit_signal);
162                 /*
163                  * If we were the last child thread and the leader has
164                  * exited already, and the leader's parent ignores SIGCHLD,
165                  * then we are the one who should release the leader.
166                  *
167                  * do_notify_parent() will have marked it self-reaping in
168                  * that case.
169                  */
170                 zap_leader = (leader->exit_signal == -1);
171         }
172
173         sched_exit(p);
174         write_unlock_irq(&tasklist_lock);
175         proc_flush_task(p);
176         release_thread(p);
177         call_rcu(&p->rcu, delayed_put_task_struct);
178
179         p = leader;
180         if (unlikely(zap_leader))
181                 goto repeat;
182 }
183
184 /*
185  * This checks not only the pgrp, but falls back on the pid if no
186  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
187  * without this...
188  */
189 int session_of_pgrp(int pgrp)
190 {
191         struct task_struct *p;
192         int sid = 0;
193
194         read_lock(&tasklist_lock);
195
196         p = find_task_by_pid_type(PIDTYPE_PGID, pgrp);
197         if (p == NULL)
198                 p = find_task_by_pid(pgrp);
199         if (p != NULL)
200                 sid = process_session(p);
201
202         read_unlock(&tasklist_lock);
203
204         return sid;
205 }
206
207 /*
208  * Determine if a process group is "orphaned", according to the POSIX
209  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
210  * by terminal-generated stop signals.  Newly orphaned process groups are
211  * to receive a SIGHUP and a SIGCONT.
212  *
213  * "I ask you, have you ever known what it is to be an orphan?"
214  */
215 static int will_become_orphaned_pgrp(int pgrp, struct task_struct *ignored_task)
216 {
217         struct task_struct *p;
218         int ret = 1;
219
220         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
221                 if (p == ignored_task
222                                 || p->exit_state
223                                 || is_init(p->real_parent))
224                         continue;
225                 if (process_group(p->real_parent) != pgrp &&
226                     process_session(p->real_parent) == process_session(p)) {
227                         ret = 0;
228                         break;
229                 }
230         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
231         return ret;     /* (sighing) "Often!" */
232 }
233
234 int is_orphaned_pgrp(int pgrp)
235 {
236         int retval;
237
238         read_lock(&tasklist_lock);
239         retval = will_become_orphaned_pgrp(pgrp, NULL);
240         read_unlock(&tasklist_lock);
241
242         return retval;
243 }
244
245 static int has_stopped_jobs(int pgrp)
246 {
247         int retval = 0;
248         struct task_struct *p;
249
250         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
251                 if (p->state != TASK_STOPPED)
252                         continue;
253                 retval = 1;
254                 break;
255         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
256         return retval;
257 }
258
259 /**
260  * reparent_to_init - Reparent the calling kernel thread to the init task
261  * of the pid space that the thread belongs to.
262  *
263  * If a kernel thread is launched as a result of a system call, or if
264  * it ever exits, it should generally reparent itself to init so that
265  * it is correctly cleaned up on exit.
266  *
267  * The various task state such as scheduling policy and priority may have
268  * been inherited from a user process, so we reset them to sane values here.
269  *
270  * NOTE that reparent_to_init() gives the caller full capabilities.
271  */
272 static void reparent_to_init(void)
273 {
274         write_lock_irq(&tasklist_lock);
275
276         ptrace_unlink(current);
277         /* Reparent to init */
278         remove_parent(current);
279         current->parent = child_reaper(current);
280         current->real_parent = child_reaper(current);
281         add_parent(current);
282
283         /* Set the exit signal to SIGCHLD so we signal init on exit */
284         current->exit_signal = SIGCHLD;
285
286         if (!has_rt_policy(current) && (task_nice(current) < 0))
287                 set_user_nice(current, 0);
288         /* cpus_allowed? */
289         /* rt_priority? */
290         /* signals? */
291         security_task_reparent_to_init(current);
292         memcpy(current->signal->rlim, init_task.signal->rlim,
293                sizeof(current->signal->rlim));
294         atomic_inc(&(INIT_USER->__count));
295         write_unlock_irq(&tasklist_lock);
296         switch_uid(INIT_USER);
297 }
298
299 void __set_special_pids(pid_t session, pid_t pgrp)
300 {
301         struct task_struct *curr = current->group_leader;
302
303         if (process_session(curr) != session) {
304                 detach_pid(curr, PIDTYPE_SID);
305                 set_signal_session(curr->signal, session);
306                 attach_pid(curr, PIDTYPE_SID, session);
307         }
308         if (process_group(curr) != pgrp) {
309                 detach_pid(curr, PIDTYPE_PGID);
310                 curr->signal->pgrp = pgrp;
311                 attach_pid(curr, PIDTYPE_PGID, pgrp);
312         }
313 }
314
315 static void set_special_pids(pid_t session, pid_t pgrp)
316 {
317         write_lock_irq(&tasklist_lock);
318         __set_special_pids(session, pgrp);
319         write_unlock_irq(&tasklist_lock);
320 }
321
322 /*
323  * Let kernel threads use this to say that they
324  * allow a certain signal (since daemonize() will
325  * have disabled all of them by default).
326  */
327 int allow_signal(int sig)
328 {
329         if (!valid_signal(sig) || sig < 1)
330                 return -EINVAL;
331
332         spin_lock_irq(&current->sighand->siglock);
333         sigdelset(&current->blocked, sig);
334         if (!current->mm) {
335                 /* Kernel threads handle their own signals.
336                    Let the signal code know it'll be handled, so
337                    that they don't get converted to SIGKILL or
338                    just silently dropped */
339                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
340         }
341         recalc_sigpending();
342         spin_unlock_irq(&current->sighand->siglock);
343         return 0;
344 }
345
346 EXPORT_SYMBOL(allow_signal);
347
348 int disallow_signal(int sig)
349 {
350         if (!valid_signal(sig) || sig < 1)
351                 return -EINVAL;
352
353         spin_lock_irq(&current->sighand->siglock);
354         sigaddset(&current->blocked, sig);
355         recalc_sigpending();
356         spin_unlock_irq(&current->sighand->siglock);
357         return 0;
358 }
359
360 EXPORT_SYMBOL(disallow_signal);
361
362 /*
363  *      Put all the gunge required to become a kernel thread without
364  *      attached user resources in one place where it belongs.
365  */
366
367 void daemonize(const char *name, ...)
368 {
369         va_list args;
370         struct fs_struct *fs;
371         sigset_t blocked;
372
373         va_start(args, name);
374         vsnprintf(current->comm, sizeof(current->comm), name, args);
375         va_end(args);
376
377         /*
378          * If we were started as result of loading a module, close all of the
379          * user space pages.  We don't need them, and if we didn't close them
380          * they would be locked into memory.
381          */
382         exit_mm(current);
383
384         set_special_pids(1, 1);
385         proc_clear_tty(current);
386
387         /* Block and flush all signals */
388         sigfillset(&blocked);
389         sigprocmask(SIG_BLOCK, &blocked, NULL);
390         flush_signals(current);
391
392         /* Become as one with the init task */
393
394         exit_fs(current);       /* current->fs->count--; */
395         fs = init_task.fs;
396         current->fs = fs;
397         atomic_inc(&fs->count);
398
399         exit_task_namespaces(current);
400         current->nsproxy = init_task.nsproxy;
401         get_task_namespaces(current);
402
403         exit_files(current);
404         current->files = init_task.files;
405         atomic_inc(&current->files->count);
406
407         reparent_to_init();
408 }
409
410 EXPORT_SYMBOL(daemonize);
411
412 static void close_files(struct files_struct * files)
413 {
414         int i, j;
415         struct fdtable *fdt;
416
417         j = 0;
418
419         /*
420          * It is safe to dereference the fd table without RCU or
421          * ->file_lock because this is the last reference to the
422          * files structure.
423          */
424         fdt = files_fdtable(files);
425         for (;;) {
426                 unsigned long set;
427                 i = j * __NFDBITS;
428                 if (i >= fdt->max_fds)
429                         break;
430                 set = fdt->open_fds->fds_bits[j++];
431                 while (set) {
432                         if (set & 1) {
433                                 struct file * file = xchg(&fdt->fd[i], NULL);
434                                 if (file)
435                                         filp_close(file, files);
436                         }
437                         i++;
438                         set >>= 1;
439                 }
440         }
441 }
442
443 struct files_struct *get_files_struct(struct task_struct *task)
444 {
445         struct files_struct *files;
446
447         task_lock(task);
448         files = task->files;
449         if (files)
450                 atomic_inc(&files->count);
451         task_unlock(task);
452
453         return files;
454 }
455
456 void fastcall put_files_struct(struct files_struct *files)
457 {
458         struct fdtable *fdt;
459
460         if (atomic_dec_and_test(&files->count)) {
461                 close_files(files);
462                 /*
463                  * Free the fd and fdset arrays if we expanded them.
464                  * If the fdtable was embedded, pass files for freeing
465                  * at the end of the RCU grace period. Otherwise,
466                  * you can free files immediately.
467                  */
468                 fdt = files_fdtable(files);
469                 if (fdt != &files->fdtab)
470                         kmem_cache_free(files_cachep, files);
471                 free_fdtable(fdt);
472         }
473 }
474
475 EXPORT_SYMBOL(put_files_struct);
476
477 void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
478 {
479         struct files_struct *old;
480
481         old = tsk->files;
482         task_lock(tsk);
483         tsk->files = files;
484         task_unlock(tsk);
485         put_files_struct(old);
486 }
487 EXPORT_SYMBOL(reset_files_struct);
488
489 static inline void __exit_files(struct task_struct *tsk)
490 {
491         struct files_struct * files = tsk->files;
492
493         if (files) {
494                 task_lock(tsk);
495                 tsk->files = NULL;
496                 task_unlock(tsk);
497                 put_files_struct(files);
498         }
499 }
500
501 void exit_files(struct task_struct *tsk)
502 {
503         __exit_files(tsk);
504 }
505
506 static inline void __put_fs_struct(struct fs_struct *fs)
507 {
508         /* No need to hold fs->lock if we are killing it */
509         if (atomic_dec_and_test(&fs->count)) {
510                 dput(fs->root);
511                 mntput(fs->rootmnt);
512                 dput(fs->pwd);
513                 mntput(fs->pwdmnt);
514                 if (fs->altroot) {
515                         dput(fs->altroot);
516                         mntput(fs->altrootmnt);
517                 }
518                 kmem_cache_free(fs_cachep, fs);
519         }
520 }
521
522 void put_fs_struct(struct fs_struct *fs)
523 {
524         __put_fs_struct(fs);
525 }
526
527 static inline void __exit_fs(struct task_struct *tsk)
528 {
529         struct fs_struct * fs = tsk->fs;
530
531         if (fs) {
532                 task_lock(tsk);
533                 tsk->fs = NULL;
534                 task_unlock(tsk);
535                 __put_fs_struct(fs);
536         }
537 }
538
539 void exit_fs(struct task_struct *tsk)
540 {
541         __exit_fs(tsk);
542 }
543
544 EXPORT_SYMBOL_GPL(exit_fs);
545
546 /*
547  * Turn us into a lazy TLB process if we
548  * aren't already..
549  */
550 static void exit_mm(struct task_struct * tsk)
551 {
552         struct mm_struct *mm = tsk->mm;
553
554         mm_release(tsk, mm);
555         if (!mm)
556                 return;
557         /*
558          * Serialize with any possible pending coredump.
559          * We must hold mmap_sem around checking core_waiters
560          * and clearing tsk->mm.  The core-inducing thread
561          * will increment core_waiters for each thread in the
562          * group with ->mm != NULL.
563          */
564         down_read(&mm->mmap_sem);
565         if (mm->core_waiters) {
566                 up_read(&mm->mmap_sem);
567                 down_write(&mm->mmap_sem);
568                 if (!--mm->core_waiters)
569                         complete(mm->core_startup_done);
570                 up_write(&mm->mmap_sem);
571
572                 wait_for_completion(&mm->core_done);
573                 down_read(&mm->mmap_sem);
574         }
575         atomic_inc(&mm->mm_count);
576         BUG_ON(mm != tsk->active_mm);
577         /* more a memory barrier than a real lock */
578         task_lock(tsk);
579         tsk->mm = NULL;
580         up_read(&mm->mmap_sem);
581         enter_lazy_tlb(mm, current);
582         task_unlock(tsk);
583         mmput(mm);
584 }
585
586 static inline void
587 choose_new_parent(struct task_struct *p, struct task_struct *reaper)
588 {
589         /*
590          * Make sure we're not reparenting to ourselves and that
591          * the parent is not a zombie.
592          */
593         BUG_ON(p == reaper || reaper->exit_state);
594         p->real_parent = reaper;
595 }
596
597 static void
598 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
599 {
600         if (p->pdeath_signal)
601                 /* We already hold the tasklist_lock here.  */
602                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
603
604         /* Move the child from its dying parent to the new one.  */
605         if (unlikely(traced)) {
606                 /* Preserve ptrace links if someone else is tracing this child.  */
607                 list_del_init(&p->ptrace_list);
608                 if (p->parent != p->real_parent)
609                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
610         } else {
611                 /* If this child is being traced, then we're the one tracing it
612                  * anyway, so let go of it.
613                  */
614                 p->ptrace = 0;
615                 remove_parent(p);
616                 p->parent = p->real_parent;
617                 add_parent(p);
618
619                 if (p->state == TASK_TRACED) {
620                         /*
621                          * If it was at a trace stop, turn it into
622                          * a normal stop since it's no longer being
623                          * traced.
624                          */
625                         ptrace_untrace(p);
626                 }
627         }
628
629         /* If this is a threaded reparent there is no need to
630          * notify anyone anything has happened.
631          */
632         if (p->real_parent->group_leader == father->group_leader)
633                 return;
634
635         /* We don't want people slaying init.  */
636         if (p->exit_signal != -1)
637                 p->exit_signal = SIGCHLD;
638
639         /* If we'd notified the old parent about this child's death,
640          * also notify the new parent.
641          */
642         if (!traced && p->exit_state == EXIT_ZOMBIE &&
643             p->exit_signal != -1 && thread_group_empty(p))
644                 do_notify_parent(p, p->exit_signal);
645
646         /*
647          * process group orphan check
648          * Case ii: Our child is in a different pgrp
649          * than we are, and it was the only connection
650          * outside, so the child pgrp is now orphaned.
651          */
652         if ((process_group(p) != process_group(father)) &&
653             (process_session(p) == process_session(father))) {
654                 int pgrp = process_group(p);
655
656                 if (will_become_orphaned_pgrp(pgrp, NULL) &&
657                     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 child reaper process (ie "init") in our pid
669  * space.
670  */
671 static void
672 forget_original_parent(struct task_struct *father, struct list_head *to_release)
673 {
674         struct task_struct *p, *reaper = father;
675         struct list_head *_p, *_n;
676
677         do {
678                 reaper = next_thread(reaper);
679                 if (reaper == father) {
680                         reaper = child_reaper(father);
681                         break;
682                 }
683         } while (reaper->exit_state);
684
685         /*
686          * There are only two places where our children can be:
687          *
688          * - in our child list
689          * - in our ptraced child list
690          *
691          * Search them and reparent children.
692          */
693         list_for_each_safe(_p, _n, &father->children) {
694                 int ptrace;
695                 p = list_entry(_p, struct task_struct, sibling);
696
697                 ptrace = p->ptrace;
698
699                 /* if father isn't the real parent, then ptrace must be enabled */
700                 BUG_ON(father != p->real_parent && !ptrace);
701
702                 if (father == p->real_parent) {
703                         /* reparent with a reaper, real father it's us */
704                         choose_new_parent(p, reaper);
705                         reparent_thread(p, father, 0);
706                 } else {
707                         /* reparent ptraced task to its real parent */
708                         __ptrace_unlink (p);
709                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
710                             thread_group_empty(p))
711                                 do_notify_parent(p, p->exit_signal);
712                 }
713
714                 /*
715                  * if the ptraced child is a zombie with exit_signal == -1
716                  * we must collect it before we exit, or it will remain
717                  * zombie forever since we prevented it from self-reap itself
718                  * while it was being traced by us, to be able to see it in wait4.
719                  */
720                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
721                         list_add(&p->ptrace_list, to_release);
722         }
723         list_for_each_safe(_p, _n, &father->ptrace_children) {
724                 p = list_entry(_p, struct task_struct, ptrace_list);
725                 choose_new_parent(p, reaper);
726                 reparent_thread(p, father, 1);
727         }
728 }
729
730 /*
731  * Send signals to all our closest relatives so that they know
732  * to properly mourn us..
733  */
734 static void exit_notify(struct task_struct *tsk)
735 {
736         int state;
737         struct task_struct *t;
738         struct list_head ptrace_dead, *_p, *_n;
739
740         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
741             && !thread_group_empty(tsk)) {
742                 /*
743                  * This occurs when there was a race between our exit
744                  * syscall and a group signal choosing us as the one to
745                  * wake up.  It could be that we are the only thread
746                  * alerted to check for pending signals, but another thread
747                  * should be woken now to take the signal since we will not.
748                  * Now we'll wake all the threads in the group just to make
749                  * sure someone gets all the pending signals.
750                  */
751                 read_lock(&tasklist_lock);
752                 spin_lock_irq(&tsk->sighand->siglock);
753                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
754                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
755                                 recalc_sigpending_tsk(t);
756                                 if (signal_pending(t))
757                                         signal_wake_up(t, 0);
758                         }
759                 spin_unlock_irq(&tsk->sighand->siglock);
760                 read_unlock(&tasklist_lock);
761         }
762
763         write_lock_irq(&tasklist_lock);
764
765         /*
766          * This does two things:
767          *
768          * A.  Make init inherit all the child processes
769          * B.  Check to see if any process groups have become orphaned
770          *      as a result of our exiting, and if they have any stopped
771          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
772          */
773
774         INIT_LIST_HEAD(&ptrace_dead);
775         forget_original_parent(tsk, &ptrace_dead);
776         BUG_ON(!list_empty(&tsk->children));
777         BUG_ON(!list_empty(&tsk->ptrace_children));
778
779         /*
780          * Check to see if any process groups have become orphaned
781          * as a result of our exiting, and if they have any stopped
782          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
783          *
784          * Case i: Our father is in a different pgrp than we are
785          * and we were the only connection outside, so our pgrp
786          * is about to become orphaned.
787          */
788          
789         t = tsk->real_parent;
790         
791         if ((process_group(t) != process_group(tsk)) &&
792             (process_session(t) == process_session(tsk)) &&
793             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
794             has_stopped_jobs(process_group(tsk))) {
795                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
796                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
797         }
798
799         /* Let father know we died 
800          *
801          * Thread signals are configurable, but you aren't going to use
802          * that to send signals to arbitary processes. 
803          * That stops right now.
804          *
805          * If the parent exec id doesn't match the exec id we saved
806          * when we started then we know the parent has changed security
807          * domain.
808          *
809          * If our self_exec id doesn't match our parent_exec_id then
810          * we have changed execution domain as these two values started
811          * the same after a fork.
812          *      
813          */
814         
815         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
816             ( tsk->parent_exec_id != t->self_exec_id  ||
817               tsk->self_exec_id != tsk->parent_exec_id)
818             && !capable(CAP_KILL))
819                 tsk->exit_signal = SIGCHLD;
820
821
822         /* If something other than our normal parent is ptracing us, then
823          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
824          * only has special meaning to our real parent.
825          */
826         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
827                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
828                 do_notify_parent(tsk, signal);
829         } else if (tsk->ptrace) {
830                 do_notify_parent(tsk, SIGCHLD);
831         }
832
833         state = EXIT_ZOMBIE;
834         if (tsk->exit_signal == -1 &&
835             (likely(tsk->ptrace == 0) ||
836              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
837                 state = EXIT_DEAD;
838         tsk->exit_state = state;
839
840         write_unlock_irq(&tasklist_lock);
841
842         list_for_each_safe(_p, _n, &ptrace_dead) {
843                 list_del_init(_p);
844                 t = list_entry(_p, struct task_struct, ptrace_list);
845                 release_task(t);
846         }
847
848         /* If the process is dead, release it - nobody will wait for it */
849         if (state == EXIT_DEAD)
850                 release_task(tsk);
851 }
852
853 fastcall NORET_TYPE void do_exit(long code)
854 {
855         struct task_struct *tsk = current;
856         int group_dead;
857
858         profile_task_exit(tsk);
859
860         WARN_ON(atomic_read(&tsk->fs_excl));
861
862         if (unlikely(in_interrupt()))
863                 panic("Aiee, killing interrupt handler!");
864         if (unlikely(!tsk->pid))
865                 panic("Attempted to kill the idle task!");
866         if (unlikely(tsk == child_reaper(tsk))) {
867                 if (tsk->nsproxy->pid_ns != &init_pid_ns)
868                         tsk->nsproxy->pid_ns->child_reaper = init_pid_ns.child_reaper;
869                 else
870                         panic("Attempted to kill init!");
871         }
872
873
874         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
875                 current->ptrace_message = code;
876                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
877         }
878
879         /*
880          * We're taking recursive faults here in do_exit. Safest is to just
881          * leave this task alone and wait for reboot.
882          */
883         if (unlikely(tsk->flags & PF_EXITING)) {
884                 printk(KERN_ALERT
885                         "Fixing recursive fault but reboot is needed!\n");
886                 if (tsk->io_context)
887                         exit_io_context();
888                 set_current_state(TASK_UNINTERRUPTIBLE);
889                 schedule();
890         }
891
892         tsk->flags |= PF_EXITING;
893
894         if (unlikely(in_atomic()))
895                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
896                                 current->comm, current->pid,
897                                 preempt_count());
898
899         acct_update_integrals(tsk);
900         if (tsk->mm) {
901                 update_hiwater_rss(tsk->mm);
902                 update_hiwater_vm(tsk->mm);
903         }
904         group_dead = atomic_dec_and_test(&tsk->signal->live);
905         if (group_dead) {
906                 hrtimer_cancel(&tsk->signal->real_timer);
907                 exit_itimers(tsk->signal);
908         }
909         acct_collect(code, group_dead);
910         if (unlikely(tsk->robust_list))
911                 exit_robust_list(tsk);
912 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
913         if (unlikely(tsk->compat_robust_list))
914                 compat_exit_robust_list(tsk);
915 #endif
916         if (unlikely(tsk->audit_context))
917                 audit_free(tsk);
918
919         taskstats_exit(tsk, group_dead);
920
921         exit_mm(tsk);
922
923         if (group_dead)
924                 acct_process();
925         exit_sem(tsk);
926         __exit_files(tsk);
927         __exit_fs(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         exit_task_namespaces(tsk);
943 #ifdef CONFIG_NUMA
944         mpol_free(tsk->mempolicy);
945         tsk->mempolicy = NULL;
946 #endif
947         /*
948          * This must happen late, after the PID is not
949          * hashed anymore:
950          */
951         if (unlikely(!list_empty(&tsk->pi_state_list)))
952                 exit_pi_state_list(tsk);
953         if (unlikely(current->pi_state_cache))
954                 kfree(current->pi_state_cache);
955         /*
956          * Make sure we are holding no locks:
957          */
958         debug_check_no_locks_held(tsk);
959
960         if (tsk->io_context)
961                 exit_io_context();
962
963         if (tsk->splice_pipe)
964                 __free_pipe_info(tsk->splice_pipe);
965
966         preempt_disable();
967         /* causes final put_task_struct in finish_task_switch(). */
968         tsk->state = TASK_DEAD;
969
970         schedule();
971         BUG();
972         /* Avoid "noreturn function does return".  */
973         for (;;)
974                 cpu_relax();    /* For when BUG is null */
975 }
976
977 EXPORT_SYMBOL_GPL(do_exit);
978
979 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
980 {
981         if (comp)
982                 complete(comp);
983
984         do_exit(code);
985 }
986
987 EXPORT_SYMBOL(complete_and_exit);
988
989 asmlinkage long sys_exit(int error_code)
990 {
991         do_exit((error_code&0xff)<<8);
992 }
993
994 /*
995  * Take down every thread in the group.  This is called by fatal signals
996  * as well as by sys_exit_group (below).
997  */
998 NORET_TYPE void
999 do_group_exit(int exit_code)
1000 {
1001         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1002
1003         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1004                 exit_code = current->signal->group_exit_code;
1005         else if (!thread_group_empty(current)) {
1006                 struct signal_struct *const sig = current->signal;
1007                 struct sighand_struct *const sighand = current->sighand;
1008                 spin_lock_irq(&sighand->siglock);
1009                 if (sig->flags & SIGNAL_GROUP_EXIT)
1010                         /* Another thread got here before we took the lock.  */
1011                         exit_code = sig->group_exit_code;
1012                 else {
1013                         sig->group_exit_code = exit_code;
1014                         zap_other_threads(current);
1015                 }
1016                 spin_unlock_irq(&sighand->siglock);
1017         }
1018
1019         do_exit(exit_code);
1020         /* NOTREACHED */
1021 }
1022
1023 /*
1024  * this kills every thread in the thread group. Note that any externally
1025  * wait4()-ing process will get the correct exit code - even if this
1026  * thread is not the thread group leader.
1027  */
1028 asmlinkage void sys_exit_group(int error_code)
1029 {
1030         do_group_exit((error_code & 0xff) << 8);
1031 }
1032
1033 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1034 {
1035         if (pid > 0) {
1036                 if (p->pid != pid)
1037                         return 0;
1038         } else if (!pid) {
1039                 if (process_group(p) != process_group(current))
1040                         return 0;
1041         } else if (pid != -1) {
1042                 if (process_group(p) != -pid)
1043                         return 0;
1044         }
1045
1046         /*
1047          * Do not consider detached threads that are
1048          * not ptraced:
1049          */
1050         if (p->exit_signal == -1 && !p->ptrace)
1051                 return 0;
1052
1053         /* Wait for all children (clone and not) if __WALL is set;
1054          * otherwise, wait for clone children *only* if __WCLONE is
1055          * set; otherwise, wait for non-clone children *only*.  (Note:
1056          * A "clone" child here is one that reports to its parent
1057          * using a signal other than SIGCHLD.) */
1058         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1059             && !(options & __WALL))
1060                 return 0;
1061         /*
1062          * Do not consider thread group leaders that are
1063          * in a non-empty thread group:
1064          */
1065         if (delay_group_leader(p))
1066                 return 2;
1067
1068         if (security_task_wait(p))
1069                 return 0;
1070
1071         return 1;
1072 }
1073
1074 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1075                                int why, int status,
1076                                struct siginfo __user *infop,
1077                                struct rusage __user *rusagep)
1078 {
1079         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1080
1081         put_task_struct(p);
1082         if (!retval)
1083                 retval = put_user(SIGCHLD, &infop->si_signo);
1084         if (!retval)
1085                 retval = put_user(0, &infop->si_errno);
1086         if (!retval)
1087                 retval = put_user((short)why, &infop->si_code);
1088         if (!retval)
1089                 retval = put_user(pid, &infop->si_pid);
1090         if (!retval)
1091                 retval = put_user(uid, &infop->si_uid);
1092         if (!retval)
1093                 retval = put_user(status, &infop->si_status);
1094         if (!retval)
1095                 retval = pid;
1096         return retval;
1097 }
1098
1099 /*
1100  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1101  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1102  * the lock and this task is uninteresting.  If we return nonzero, we have
1103  * released the lock and the system call should return.
1104  */
1105 static int wait_task_zombie(struct task_struct *p, int noreap,
1106                             struct siginfo __user *infop,
1107                             int __user *stat_addr, struct rusage __user *ru)
1108 {
1109         unsigned long state;
1110         int retval;
1111         int status;
1112
1113         if (unlikely(noreap)) {
1114                 pid_t pid = p->pid;
1115                 uid_t uid = p->uid;
1116                 int exit_code = p->exit_code;
1117                 int why, status;
1118
1119                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1120                         return 0;
1121                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1122                         return 0;
1123                 get_task_struct(p);
1124                 read_unlock(&tasklist_lock);
1125                 if ((exit_code & 0x7f) == 0) {
1126                         why = CLD_EXITED;
1127                         status = exit_code >> 8;
1128                 } else {
1129                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1130                         status = exit_code & 0x7f;
1131                 }
1132                 return wait_noreap_copyout(p, pid, uid, why,
1133                                            status, infop, ru);
1134         }
1135
1136         /*
1137          * Try to move the task's state to DEAD
1138          * only one thread is allowed to do this:
1139          */
1140         state = xchg(&p->exit_state, EXIT_DEAD);
1141         if (state != EXIT_ZOMBIE) {
1142                 BUG_ON(state != EXIT_DEAD);
1143                 return 0;
1144         }
1145         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1146                 /*
1147                  * This can only happen in a race with a ptraced thread
1148                  * dying on another processor.
1149                  */
1150                 return 0;
1151         }
1152
1153         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1154                 struct signal_struct *psig;
1155                 struct signal_struct *sig;
1156
1157                 /*
1158                  * The resource counters for the group leader are in its
1159                  * own task_struct.  Those for dead threads in the group
1160                  * are in its signal_struct, as are those for the child
1161                  * processes it has previously reaped.  All these
1162                  * accumulate in the parent's signal_struct c* fields.
1163                  *
1164                  * We don't bother to take a lock here to protect these
1165                  * p->signal fields, because they are only touched by
1166                  * __exit_signal, which runs with tasklist_lock
1167                  * write-locked anyway, and so is excluded here.  We do
1168                  * need to protect the access to p->parent->signal fields,
1169                  * as other threads in the parent group can be right
1170                  * here reaping other children at the same time.
1171                  */
1172                 spin_lock_irq(&p->parent->sighand->siglock);
1173                 psig = p->parent->signal;
1174                 sig = p->signal;
1175                 psig->cutime =
1176                         cputime_add(psig->cutime,
1177                         cputime_add(p->utime,
1178                         cputime_add(sig->utime,
1179                                     sig->cutime)));
1180                 psig->cstime =
1181                         cputime_add(psig->cstime,
1182                         cputime_add(p->stime,
1183                         cputime_add(sig->stime,
1184                                     sig->cstime)));
1185                 psig->cmin_flt +=
1186                         p->min_flt + sig->min_flt + sig->cmin_flt;
1187                 psig->cmaj_flt +=
1188                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1189                 psig->cnvcsw +=
1190                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1191                 psig->cnivcsw +=
1192                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1193                 spin_unlock_irq(&p->parent->sighand->siglock);
1194         }
1195
1196         /*
1197          * Now we are sure this task is interesting, and no other
1198          * thread can reap it because we set its state to EXIT_DEAD.
1199          */
1200         read_unlock(&tasklist_lock);
1201
1202         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1203         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1204                 ? p->signal->group_exit_code : p->exit_code;
1205         if (!retval && stat_addr)
1206                 retval = put_user(status, stat_addr);
1207         if (!retval && infop)
1208                 retval = put_user(SIGCHLD, &infop->si_signo);
1209         if (!retval && infop)
1210                 retval = put_user(0, &infop->si_errno);
1211         if (!retval && infop) {
1212                 int why;
1213
1214                 if ((status & 0x7f) == 0) {
1215                         why = CLD_EXITED;
1216                         status >>= 8;
1217                 } else {
1218                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1219                         status &= 0x7f;
1220                 }
1221                 retval = put_user((short)why, &infop->si_code);
1222                 if (!retval)
1223                         retval = put_user(status, &infop->si_status);
1224         }
1225         if (!retval && infop)
1226                 retval = put_user(p->pid, &infop->si_pid);
1227         if (!retval && infop)
1228                 retval = put_user(p->uid, &infop->si_uid);
1229         if (retval) {
1230                 // TODO: is this safe?
1231                 p->exit_state = EXIT_ZOMBIE;
1232                 return retval;
1233         }
1234         retval = p->pid;
1235         if (p->real_parent != p->parent) {
1236                 write_lock_irq(&tasklist_lock);
1237                 /* Double-check with lock held.  */
1238                 if (p->real_parent != p->parent) {
1239                         __ptrace_unlink(p);
1240                         // TODO: is this safe?
1241                         p->exit_state = EXIT_ZOMBIE;
1242                         /*
1243                          * If this is not a detached task, notify the parent.
1244                          * If it's still not detached after that, don't release
1245                          * it now.
1246                          */
1247                         if (p->exit_signal != -1) {
1248                                 do_notify_parent(p, p->exit_signal);
1249                                 if (p->exit_signal != -1)
1250                                         p = NULL;
1251                         }
1252                 }
1253                 write_unlock_irq(&tasklist_lock);
1254         }
1255         if (p != NULL)
1256                 release_task(p);
1257         BUG_ON(!retval);
1258         return retval;
1259 }
1260
1261 /*
1262  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1263  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1264  * the lock and this task is uninteresting.  If we return nonzero, we have
1265  * released the lock and the system call should return.
1266  */
1267 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1268                              int noreap, struct siginfo __user *infop,
1269                              int __user *stat_addr, struct rusage __user *ru)
1270 {
1271         int retval, exit_code;
1272
1273         if (!p->exit_code)
1274                 return 0;
1275         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1276             p->signal && p->signal->group_stop_count > 0)
1277                 /*
1278                  * A group stop is in progress and this is the group leader.
1279                  * We won't report until all threads have stopped.
1280                  */
1281                 return 0;
1282
1283         /*
1284          * Now we are pretty sure this task is interesting.
1285          * Make sure it doesn't get reaped out from under us while we
1286          * give up the lock and then examine it below.  We don't want to
1287          * keep holding onto the tasklist_lock while we call getrusage and
1288          * possibly take page faults for user memory.
1289          */
1290         get_task_struct(p);
1291         read_unlock(&tasklist_lock);
1292
1293         if (unlikely(noreap)) {
1294                 pid_t pid = p->pid;
1295                 uid_t uid = p->uid;
1296                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1297
1298                 exit_code = p->exit_code;
1299                 if (unlikely(!exit_code) ||
1300                     unlikely(p->state & TASK_TRACED))
1301                         goto bail_ref;
1302                 return wait_noreap_copyout(p, pid, uid,
1303                                            why, (exit_code << 8) | 0x7f,
1304                                            infop, ru);
1305         }
1306
1307         write_lock_irq(&tasklist_lock);
1308
1309         /*
1310          * This uses xchg to be atomic with the thread resuming and setting
1311          * it.  It must also be done with the write lock held to prevent a
1312          * race with the EXIT_ZOMBIE case.
1313          */
1314         exit_code = xchg(&p->exit_code, 0);
1315         if (unlikely(p->exit_state)) {
1316                 /*
1317                  * The task resumed and then died.  Let the next iteration
1318                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1319                  * already be zero here if it resumed and did _exit(0).
1320                  * The task itself is dead and won't touch exit_code again;
1321                  * other processors in this function are locked out.
1322                  */
1323                 p->exit_code = exit_code;
1324                 exit_code = 0;
1325         }
1326         if (unlikely(exit_code == 0)) {
1327                 /*
1328                  * Another thread in this function got to it first, or it
1329                  * resumed, or it resumed and then died.
1330                  */
1331                 write_unlock_irq(&tasklist_lock);
1332 bail_ref:
1333                 put_task_struct(p);
1334                 /*
1335                  * We are returning to the wait loop without having successfully
1336                  * removed the process and having released the lock. We cannot
1337                  * continue, since the "p" task pointer is potentially stale.
1338                  *
1339                  * Return -EAGAIN, and do_wait() will restart the loop from the
1340                  * beginning. Do _not_ re-acquire the lock.
1341                  */
1342                 return -EAGAIN;
1343         }
1344
1345         /* move to end of parent's list to avoid starvation */
1346         remove_parent(p);
1347         add_parent(p);
1348
1349         write_unlock_irq(&tasklist_lock);
1350
1351         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1352         if (!retval && stat_addr)
1353                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1354         if (!retval && infop)
1355                 retval = put_user(SIGCHLD, &infop->si_signo);
1356         if (!retval && infop)
1357                 retval = put_user(0, &infop->si_errno);
1358         if (!retval && infop)
1359                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1360                                           ? CLD_TRAPPED : CLD_STOPPED),
1361                                   &infop->si_code);
1362         if (!retval && infop)
1363                 retval = put_user(exit_code, &infop->si_status);
1364         if (!retval && infop)
1365                 retval = put_user(p->pid, &infop->si_pid);
1366         if (!retval && infop)
1367                 retval = put_user(p->uid, &infop->si_uid);
1368         if (!retval)
1369                 retval = p->pid;
1370         put_task_struct(p);
1371
1372         BUG_ON(!retval);
1373         return retval;
1374 }
1375
1376 /*
1377  * Handle do_wait work for one task in a live, non-stopped state.
1378  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1379  * the lock and this task is uninteresting.  If we return nonzero, we have
1380  * released the lock and the system call should return.
1381  */
1382 static int wait_task_continued(struct task_struct *p, int noreap,
1383                                struct siginfo __user *infop,
1384                                int __user *stat_addr, struct rusage __user *ru)
1385 {
1386         int retval;
1387         pid_t pid;
1388         uid_t uid;
1389
1390         if (unlikely(!p->signal))
1391                 return 0;
1392
1393         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1394                 return 0;
1395
1396         spin_lock_irq(&p->sighand->siglock);
1397         /* Re-check with the lock held.  */
1398         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1399                 spin_unlock_irq(&p->sighand->siglock);
1400                 return 0;
1401         }
1402         if (!noreap)
1403                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1404         spin_unlock_irq(&p->sighand->siglock);
1405
1406         pid = p->pid;
1407         uid = p->uid;
1408         get_task_struct(p);
1409         read_unlock(&tasklist_lock);
1410
1411         if (!infop) {
1412                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1413                 put_task_struct(p);
1414                 if (!retval && stat_addr)
1415                         retval = put_user(0xffff, stat_addr);
1416                 if (!retval)
1417                         retval = p->pid;
1418         } else {
1419                 retval = wait_noreap_copyout(p, pid, uid,
1420                                              CLD_CONTINUED, SIGCONT,
1421                                              infop, ru);
1422                 BUG_ON(retval == 0);
1423         }
1424
1425         return retval;
1426 }
1427
1428
1429 static inline int my_ptrace_child(struct task_struct *p)
1430 {
1431         if (!(p->ptrace & PT_PTRACED))
1432                 return 0;
1433         if (!(p->ptrace & PT_ATTACHED))
1434                 return 1;
1435         /*
1436          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1437          * we are the attacher.  If we are the real parent, this is a race
1438          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1439          * which we have to switch the parent links, but has already set
1440          * the flags in p->ptrace.
1441          */
1442         return (p->parent != p->real_parent);
1443 }
1444
1445 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1446                     int __user *stat_addr, struct rusage __user *ru)
1447 {
1448         DECLARE_WAITQUEUE(wait, current);
1449         struct task_struct *tsk;
1450         int flag, retval;
1451
1452         add_wait_queue(&current->signal->wait_chldexit,&wait);
1453 repeat:
1454         /*
1455          * We will set this flag if we see any child that might later
1456          * match our criteria, even if we are not able to reap it yet.
1457          */
1458         flag = 0;
1459         current->state = TASK_INTERRUPTIBLE;
1460         read_lock(&tasklist_lock);
1461         tsk = current;
1462         do {
1463                 struct task_struct *p;
1464                 struct list_head *_p;
1465                 int ret;
1466
1467                 list_for_each(_p,&tsk->children) {
1468                         p = list_entry(_p, struct task_struct, sibling);
1469
1470                         ret = eligible_child(pid, options, p);
1471                         if (!ret)
1472                                 continue;
1473
1474                         switch (p->state) {
1475                         case TASK_TRACED:
1476                                 /*
1477                                  * When we hit the race with PTRACE_ATTACH,
1478                                  * we will not report this child.  But the
1479                                  * race means it has not yet been moved to
1480                                  * our ptrace_children list, so we need to
1481                                  * set the flag here to avoid a spurious ECHILD
1482                                  * when the race happens with the only child.
1483                                  */
1484                                 flag = 1;
1485                                 if (!my_ptrace_child(p))
1486                                         continue;
1487                                 /*FALLTHROUGH*/
1488                         case TASK_STOPPED:
1489                                 /*
1490                                  * It's stopped now, so it might later
1491                                  * continue, exit, or stop again.
1492                                  */
1493                                 flag = 1;
1494                                 if (!(options & WUNTRACED) &&
1495                                     !my_ptrace_child(p))
1496                                         continue;
1497                                 retval = wait_task_stopped(p, ret == 2,
1498                                                            (options & WNOWAIT),
1499                                                            infop,
1500                                                            stat_addr, ru);
1501                                 if (retval == -EAGAIN)
1502                                         goto repeat;
1503                                 if (retval != 0) /* He released the lock.  */
1504                                         goto end;
1505                                 break;
1506                         default:
1507                         // case EXIT_DEAD:
1508                                 if (p->exit_state == EXIT_DEAD)
1509                                         continue;
1510                         // case EXIT_ZOMBIE:
1511                                 if (p->exit_state == EXIT_ZOMBIE) {
1512                                         /*
1513                                          * Eligible but we cannot release
1514                                          * it yet:
1515                                          */
1516                                         if (ret == 2)
1517                                                 goto check_continued;
1518                                         if (!likely(options & WEXITED))
1519                                                 continue;
1520                                         retval = wait_task_zombie(
1521                                                 p, (options & WNOWAIT),
1522                                                 infop, stat_addr, ru);
1523                                         /* He released the lock.  */
1524                                         if (retval != 0)
1525                                                 goto end;
1526                                         break;
1527                                 }
1528 check_continued:
1529                                 /*
1530                                  * It's running now, so it might later
1531                                  * exit, stop, or stop and then continue.
1532                                  */
1533                                 flag = 1;
1534                                 if (!unlikely(options & WCONTINUED))
1535                                         continue;
1536                                 retval = wait_task_continued(
1537                                         p, (options & WNOWAIT),
1538                                         infop, stat_addr, ru);
1539                                 if (retval != 0) /* He released the lock.  */
1540                                         goto end;
1541                                 break;
1542                         }
1543                 }
1544                 if (!flag) {
1545                         list_for_each(_p, &tsk->ptrace_children) {
1546                                 p = list_entry(_p, struct task_struct,
1547                                                 ptrace_list);
1548                                 if (!eligible_child(pid, options, p))
1549                                         continue;
1550                                 flag = 1;
1551                                 break;
1552                         }
1553                 }
1554                 if (options & __WNOTHREAD)
1555                         break;
1556                 tsk = next_thread(tsk);
1557                 BUG_ON(tsk->signal != current->signal);
1558         } while (tsk != current);
1559
1560         read_unlock(&tasklist_lock);
1561         if (flag) {
1562                 retval = 0;
1563                 if (options & WNOHANG)
1564                         goto end;
1565                 retval = -ERESTARTSYS;
1566                 if (signal_pending(current))
1567                         goto end;
1568                 schedule();
1569                 goto repeat;
1570         }
1571         retval = -ECHILD;
1572 end:
1573         current->state = TASK_RUNNING;
1574         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1575         if (infop) {
1576                 if (retval > 0)
1577                 retval = 0;
1578                 else {
1579                         /*
1580                          * For a WNOHANG return, clear out all the fields
1581                          * we would set so the user can easily tell the
1582                          * difference.
1583                          */
1584                         if (!retval)
1585                                 retval = put_user(0, &infop->si_signo);
1586                         if (!retval)
1587                                 retval = put_user(0, &infop->si_errno);
1588                         if (!retval)
1589                                 retval = put_user(0, &infop->si_code);
1590                         if (!retval)
1591                                 retval = put_user(0, &infop->si_pid);
1592                         if (!retval)
1593                                 retval = put_user(0, &infop->si_uid);
1594                         if (!retval)
1595                                 retval = put_user(0, &infop->si_status);
1596                 }
1597         }
1598         return retval;
1599 }
1600
1601 asmlinkage long sys_waitid(int which, pid_t pid,
1602                            struct siginfo __user *infop, int options,
1603                            struct rusage __user *ru)
1604 {
1605         long ret;
1606
1607         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1608                 return -EINVAL;
1609         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1610                 return -EINVAL;
1611
1612         switch (which) {
1613         case P_ALL:
1614                 pid = -1;
1615                 break;
1616         case P_PID:
1617                 if (pid <= 0)
1618                         return -EINVAL;
1619                 break;
1620         case P_PGID:
1621                 if (pid <= 0)
1622                         return -EINVAL;
1623                 pid = -pid;
1624                 break;
1625         default:
1626                 return -EINVAL;
1627         }
1628
1629         ret = do_wait(pid, options, infop, NULL, ru);
1630
1631         /* avoid REGPARM breakage on x86: */
1632         prevent_tail_call(ret);
1633         return ret;
1634 }
1635
1636 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1637                           int options, struct rusage __user *ru)
1638 {
1639         long ret;
1640
1641         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1642                         __WNOTHREAD|__WCLONE|__WALL))
1643                 return -EINVAL;
1644         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1645
1646         /* avoid REGPARM breakage on x86: */
1647         prevent_tail_call(ret);
1648         return ret;
1649 }
1650
1651 #ifdef __ARCH_WANT_SYS_WAITPID
1652
1653 /*
1654  * sys_waitpid() remains for compatibility. waitpid() should be
1655  * implemented by calling sys_wait4() from libc.a.
1656  */
1657 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1658 {
1659         return sys_wait4(pid, stat_addr, options, NULL);
1660 }
1661
1662 #endif