[PATCH] move __exit_signal() to kernel/exit.c
[pandora-kernel.git] / kernel / exit.c
1 /*
2  *  linux/kernel/exit.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 #include <linux/config.h>
8 #include <linux/mm.h>
9 #include <linux/slab.h>
10 #include <linux/interrupt.h>
11 #include <linux/smp_lock.h>
12 #include <linux/module.h>
13 #include <linux/capability.h>
14 #include <linux/completion.h>
15 #include <linux/personality.h>
16 #include <linux/tty.h>
17 #include <linux/namespace.h>
18 #include <linux/key.h>
19 #include <linux/security.h>
20 #include <linux/cpu.h>
21 #include <linux/acct.h>
22 #include <linux/file.h>
23 #include <linux/binfmts.h>
24 #include <linux/ptrace.h>
25 #include <linux/profile.h>
26 #include <linux/mount.h>
27 #include <linux/proc_fs.h>
28 #include <linux/mempolicy.h>
29 #include <linux/cpuset.h>
30 #include <linux/syscalls.h>
31 #include <linux/signal.h>
32 #include <linux/posix-timers.h>
33 #include <linux/cn_proc.h>
34 #include <linux/mutex.h>
35 #include <linux/futex.h>
36 #include <linux/compat.h>
37
38 #include <asm/uaccess.h>
39 #include <asm/unistd.h>
40 #include <asm/pgtable.h>
41 #include <asm/mmu_context.h>
42
43 extern void sem_exit (void);
44 extern struct task_struct *child_reaper;
45
46 int getrusage(struct task_struct *, int, struct rusage __user *);
47
48 static void exit_mm(struct task_struct * tsk);
49
50 static void __unhash_process(struct task_struct *p)
51 {
52         nr_threads--;
53         detach_pid(p, PIDTYPE_PID);
54         detach_pid(p, PIDTYPE_TGID);
55         if (thread_group_leader(p)) {
56                 detach_pid(p, PIDTYPE_PGID);
57                 detach_pid(p, PIDTYPE_SID);
58
59                 list_del_init(&p->tasks);
60                 __get_cpu_var(process_counts)--;
61         }
62
63         remove_parent(p);
64 }
65
66 /*
67  * This function expects the tasklist_lock write-locked.
68  */
69 static void __exit_signal(struct task_struct *tsk)
70 {
71         struct signal_struct *sig = tsk->signal;
72         struct sighand_struct *sighand;
73
74         BUG_ON(!sig);
75         BUG_ON(!atomic_read(&sig->count));
76
77         rcu_read_lock();
78         sighand = rcu_dereference(tsk->sighand);
79         spin_lock(&sighand->siglock);
80
81         posix_cpu_timers_exit(tsk);
82         if (atomic_dec_and_test(&sig->count))
83                 posix_cpu_timers_exit_group(tsk);
84         else {
85                 /*
86                  * If there is any task waiting for the group exit
87                  * then notify it:
88                  */
89                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
90                         wake_up_process(sig->group_exit_task);
91                         sig->group_exit_task = NULL;
92                 }
93                 if (tsk == sig->curr_target)
94                         sig->curr_target = next_thread(tsk);
95                 /*
96                  * Accumulate here the counters for all threads but the
97                  * group leader as they die, so they can be added into
98                  * the process-wide totals when those are taken.
99                  * The group leader stays around as a zombie as long
100                  * as there are other threads.  When it gets reaped,
101                  * the exit.c code will add its counts into these totals.
102                  * We won't ever get here for the group leader, since it
103                  * will have been the last reference on the signal_struct.
104                  */
105                 sig->utime = cputime_add(sig->utime, tsk->utime);
106                 sig->stime = cputime_add(sig->stime, tsk->stime);
107                 sig->min_flt += tsk->min_flt;
108                 sig->maj_flt += tsk->maj_flt;
109                 sig->nvcsw += tsk->nvcsw;
110                 sig->nivcsw += tsk->nivcsw;
111                 sig->sched_time += tsk->sched_time;
112                 sig = NULL; /* Marker for below. */
113         }
114
115         tsk->signal = NULL;
116         cleanup_sighand(tsk);
117         spin_unlock(&sighand->siglock);
118         rcu_read_unlock();
119
120         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
121         flush_sigqueue(&tsk->pending);
122         if (sig) {
123                 flush_sigqueue(&sig->shared_pending);
124                 __cleanup_signal(sig);
125         }
126 }
127
128 void release_task(struct task_struct * p)
129 {
130         int zap_leader;
131         task_t *leader;
132         struct dentry *proc_dentry;
133
134 repeat:
135         atomic_dec(&p->user->processes);
136         spin_lock(&p->proc_lock);
137         proc_dentry = proc_pid_unhash(p);
138         write_lock_irq(&tasklist_lock);
139         ptrace_unlink(p);
140         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
141         __exit_signal(p);
142         /*
143          * Note that the fastpath in sys_times depends on __exit_signal having
144          * updated the counters before a task is removed from the tasklist of
145          * the process by __unhash_process.
146          */
147         __unhash_process(p);
148
149         /*
150          * If we are the last non-leader member of the thread
151          * group, and the leader is zombie, then notify the
152          * group leader's parent process. (if it wants notification.)
153          */
154         zap_leader = 0;
155         leader = p->group_leader;
156         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
157                 BUG_ON(leader->exit_signal == -1);
158                 do_notify_parent(leader, leader->exit_signal);
159                 /*
160                  * If we were the last child thread and the leader has
161                  * exited already, and the leader's parent ignores SIGCHLD,
162                  * then we are the one who should release the leader.
163                  *
164                  * do_notify_parent() will have marked it self-reaping in
165                  * that case.
166                  */
167                 zap_leader = (leader->exit_signal == -1);
168         }
169
170         sched_exit(p);
171         write_unlock_irq(&tasklist_lock);
172         spin_unlock(&p->proc_lock);
173         proc_pid_flush(proc_dentry);
174         release_thread(p);
175         put_task_struct(p);
176
177         p = leader;
178         if (unlikely(zap_leader))
179                 goto repeat;
180 }
181
182 /*
183  * This checks not only the pgrp, but falls back on the pid if no
184  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
185  * without this...
186  */
187 int session_of_pgrp(int pgrp)
188 {
189         struct task_struct *p;
190         int sid = -1;
191
192         read_lock(&tasklist_lock);
193         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
194                 if (p->signal->session > 0) {
195                         sid = p->signal->session;
196                         goto out;
197                 }
198         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
199         p = find_task_by_pid(pgrp);
200         if (p)
201                 sid = p->signal->session;
202 out:
203         read_unlock(&tasklist_lock);
204         
205         return sid;
206 }
207
208 /*
209  * Determine if a process group is "orphaned", according to the POSIX
210  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
211  * by terminal-generated stop signals.  Newly orphaned process groups are
212  * to receive a SIGHUP and a SIGCONT.
213  *
214  * "I ask you, have you ever known what it is to be an orphan?"
215  */
216 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
217 {
218         struct task_struct *p;
219         int ret = 1;
220
221         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
222                 if (p == ignored_task
223                                 || p->exit_state
224                                 || p->real_parent->pid == 1)
225                         continue;
226                 if (process_group(p->real_parent) != pgrp
227                             && p->real_parent->signal->session == p->signal->session) {
228                         ret = 0;
229                         break;
230                 }
231         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
232         return ret;     /* (sighing) "Often!" */
233 }
234
235 int is_orphaned_pgrp(int pgrp)
236 {
237         int retval;
238
239         read_lock(&tasklist_lock);
240         retval = will_become_orphaned_pgrp(pgrp, NULL);
241         read_unlock(&tasklist_lock);
242
243         return retval;
244 }
245
246 static int has_stopped_jobs(int pgrp)
247 {
248         int retval = 0;
249         struct task_struct *p;
250
251         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
252                 if (p->state != TASK_STOPPED)
253                         continue;
254
255                 /* If p is stopped by a debugger on a signal that won't
256                    stop it, then don't count p as stopped.  This isn't
257                    perfect but it's a good approximation.  */
258                 if (unlikely (p->ptrace)
259                     && p->exit_code != SIGSTOP
260                     && p->exit_code != SIGTSTP
261                     && p->exit_code != SIGTTOU
262                     && p->exit_code != SIGTTIN)
263                         continue;
264
265                 retval = 1;
266                 break;
267         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
268         return retval;
269 }
270
271 /**
272  * reparent_to_init - Reparent the calling kernel thread to the init task.
273  *
274  * If a kernel thread is launched as a result of a system call, or if
275  * it ever exits, it should generally reparent itself to init so that
276  * it is correctly cleaned up on exit.
277  *
278  * The various task state such as scheduling policy and priority may have
279  * been inherited from a user process, so we reset them to sane values here.
280  *
281  * NOTE that reparent_to_init() gives the caller full capabilities.
282  */
283 static void reparent_to_init(void)
284 {
285         write_lock_irq(&tasklist_lock);
286
287         ptrace_unlink(current);
288         /* Reparent to init */
289         remove_parent(current);
290         current->parent = child_reaper;
291         current->real_parent = child_reaper;
292         add_parent(current);
293
294         /* Set the exit signal to SIGCHLD so we signal init on exit */
295         current->exit_signal = SIGCHLD;
296
297         if ((current->policy == SCHED_NORMAL ||
298                         current->policy == SCHED_BATCH)
299                                 && (task_nice(current) < 0))
300                 set_user_nice(current, 0);
301         /* cpus_allowed? */
302         /* rt_priority? */
303         /* signals? */
304         security_task_reparent_to_init(current);
305         memcpy(current->signal->rlim, init_task.signal->rlim,
306                sizeof(current->signal->rlim));
307         atomic_inc(&(INIT_USER->__count));
308         write_unlock_irq(&tasklist_lock);
309         switch_uid(INIT_USER);
310 }
311
312 void __set_special_pids(pid_t session, pid_t pgrp)
313 {
314         struct task_struct *curr = current->group_leader;
315
316         if (curr->signal->session != session) {
317                 detach_pid(curr, PIDTYPE_SID);
318                 curr->signal->session = session;
319                 attach_pid(curr, PIDTYPE_SID, session);
320         }
321         if (process_group(curr) != pgrp) {
322                 detach_pid(curr, PIDTYPE_PGID);
323                 curr->signal->pgrp = pgrp;
324                 attach_pid(curr, PIDTYPE_PGID, pgrp);
325         }
326 }
327
328 void set_special_pids(pid_t session, pid_t pgrp)
329 {
330         write_lock_irq(&tasklist_lock);
331         __set_special_pids(session, pgrp);
332         write_unlock_irq(&tasklist_lock);
333 }
334
335 /*
336  * Let kernel threads use this to say that they
337  * allow a certain signal (since daemonize() will
338  * have disabled all of them by default).
339  */
340 int allow_signal(int sig)
341 {
342         if (!valid_signal(sig) || sig < 1)
343                 return -EINVAL;
344
345         spin_lock_irq(&current->sighand->siglock);
346         sigdelset(&current->blocked, sig);
347         if (!current->mm) {
348                 /* Kernel threads handle their own signals.
349                    Let the signal code know it'll be handled, so
350                    that they don't get converted to SIGKILL or
351                    just silently dropped */
352                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
353         }
354         recalc_sigpending();
355         spin_unlock_irq(&current->sighand->siglock);
356         return 0;
357 }
358
359 EXPORT_SYMBOL(allow_signal);
360
361 int disallow_signal(int sig)
362 {
363         if (!valid_signal(sig) || sig < 1)
364                 return -EINVAL;
365
366         spin_lock_irq(&current->sighand->siglock);
367         sigaddset(&current->blocked, sig);
368         recalc_sigpending();
369         spin_unlock_irq(&current->sighand->siglock);
370         return 0;
371 }
372
373 EXPORT_SYMBOL(disallow_signal);
374
375 /*
376  *      Put all the gunge required to become a kernel thread without
377  *      attached user resources in one place where it belongs.
378  */
379
380 void daemonize(const char *name, ...)
381 {
382         va_list args;
383         struct fs_struct *fs;
384         sigset_t blocked;
385
386         va_start(args, name);
387         vsnprintf(current->comm, sizeof(current->comm), name, args);
388         va_end(args);
389
390         /*
391          * If we were started as result of loading a module, close all of the
392          * user space pages.  We don't need them, and if we didn't close them
393          * they would be locked into memory.
394          */
395         exit_mm(current);
396
397         set_special_pids(1, 1);
398         mutex_lock(&tty_mutex);
399         current->signal->tty = NULL;
400         mutex_unlock(&tty_mutex);
401
402         /* Block and flush all signals */
403         sigfillset(&blocked);
404         sigprocmask(SIG_BLOCK, &blocked, NULL);
405         flush_signals(current);
406
407         /* Become as one with the init task */
408
409         exit_fs(current);       /* current->fs->count--; */
410         fs = init_task.fs;
411         current->fs = fs;
412         atomic_inc(&fs->count);
413         exit_namespace(current);
414         current->namespace = init_task.namespace;
415         get_namespace(current->namespace);
416         exit_files(current);
417         current->files = init_task.files;
418         atomic_inc(&current->files->count);
419
420         reparent_to_init();
421 }
422
423 EXPORT_SYMBOL(daemonize);
424
425 static void close_files(struct files_struct * files)
426 {
427         int i, j;
428         struct fdtable *fdt;
429
430         j = 0;
431
432         /*
433          * It is safe to dereference the fd table without RCU or
434          * ->file_lock because this is the last reference to the
435          * files structure.
436          */
437         fdt = files_fdtable(files);
438         for (;;) {
439                 unsigned long set;
440                 i = j * __NFDBITS;
441                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
442                         break;
443                 set = fdt->open_fds->fds_bits[j++];
444                 while (set) {
445                         if (set & 1) {
446                                 struct file * file = xchg(&fdt->fd[i], NULL);
447                                 if (file)
448                                         filp_close(file, files);
449                         }
450                         i++;
451                         set >>= 1;
452                 }
453         }
454 }
455
456 struct files_struct *get_files_struct(struct task_struct *task)
457 {
458         struct files_struct *files;
459
460         task_lock(task);
461         files = task->files;
462         if (files)
463                 atomic_inc(&files->count);
464         task_unlock(task);
465
466         return files;
467 }
468
469 void fastcall put_files_struct(struct files_struct *files)
470 {
471         struct fdtable *fdt;
472
473         if (atomic_dec_and_test(&files->count)) {
474                 close_files(files);
475                 /*
476                  * Free the fd and fdset arrays if we expanded them.
477                  * If the fdtable was embedded, pass files for freeing
478                  * at the end of the RCU grace period. Otherwise,
479                  * you can free files immediately.
480                  */
481                 fdt = files_fdtable(files);
482                 if (fdt == &files->fdtab)
483                         fdt->free_files = files;
484                 else
485                         kmem_cache_free(files_cachep, files);
486                 free_fdtable(fdt);
487         }
488 }
489
490 EXPORT_SYMBOL(put_files_struct);
491
492 static inline void __exit_files(struct task_struct *tsk)
493 {
494         struct files_struct * files = tsk->files;
495
496         if (files) {
497                 task_lock(tsk);
498                 tsk->files = NULL;
499                 task_unlock(tsk);
500                 put_files_struct(files);
501         }
502 }
503
504 void exit_files(struct task_struct *tsk)
505 {
506         __exit_files(tsk);
507 }
508
509 static inline void __put_fs_struct(struct fs_struct *fs)
510 {
511         /* No need to hold fs->lock if we are killing it */
512         if (atomic_dec_and_test(&fs->count)) {
513                 dput(fs->root);
514                 mntput(fs->rootmnt);
515                 dput(fs->pwd);
516                 mntput(fs->pwdmnt);
517                 if (fs->altroot) {
518                         dput(fs->altroot);
519                         mntput(fs->altrootmnt);
520                 }
521                 kmem_cache_free(fs_cachep, fs);
522         }
523 }
524
525 void put_fs_struct(struct fs_struct *fs)
526 {
527         __put_fs_struct(fs);
528 }
529
530 static inline void __exit_fs(struct task_struct *tsk)
531 {
532         struct fs_struct * fs = tsk->fs;
533
534         if (fs) {
535                 task_lock(tsk);
536                 tsk->fs = NULL;
537                 task_unlock(tsk);
538                 __put_fs_struct(fs);
539         }
540 }
541
542 void exit_fs(struct task_struct *tsk)
543 {
544         __exit_fs(tsk);
545 }
546
547 EXPORT_SYMBOL_GPL(exit_fs);
548
549 /*
550  * Turn us into a lazy TLB process if we
551  * aren't already..
552  */
553 static void exit_mm(struct task_struct * tsk)
554 {
555         struct mm_struct *mm = tsk->mm;
556
557         mm_release(tsk, mm);
558         if (!mm)
559                 return;
560         /*
561          * Serialize with any possible pending coredump.
562          * We must hold mmap_sem around checking core_waiters
563          * and clearing tsk->mm.  The core-inducing thread
564          * will increment core_waiters for each thread in the
565          * group with ->mm != NULL.
566          */
567         down_read(&mm->mmap_sem);
568         if (mm->core_waiters) {
569                 up_read(&mm->mmap_sem);
570                 down_write(&mm->mmap_sem);
571                 if (!--mm->core_waiters)
572                         complete(mm->core_startup_done);
573                 up_write(&mm->mmap_sem);
574
575                 wait_for_completion(&mm->core_done);
576                 down_read(&mm->mmap_sem);
577         }
578         atomic_inc(&mm->mm_count);
579         if (mm != tsk->active_mm) BUG();
580         /* more a memory barrier than a real lock */
581         task_lock(tsk);
582         tsk->mm = NULL;
583         up_read(&mm->mmap_sem);
584         enter_lazy_tlb(mm, current);
585         task_unlock(tsk);
586         mmput(mm);
587 }
588
589 static inline void choose_new_parent(task_t *p, task_t *reaper)
590 {
591         /*
592          * Make sure we're not reparenting to ourselves and that
593          * the parent is not a zombie.
594          */
595         BUG_ON(p == reaper || reaper->exit_state);
596         p->real_parent = reaper;
597 }
598
599 static void reparent_thread(task_t *p, task_t *father, int traced)
600 {
601         /* We don't want people slaying init.  */
602         if (p->exit_signal != -1)
603                 p->exit_signal = SIGCHLD;
604
605         if (p->pdeath_signal)
606                 /* We already hold the tasklist_lock here.  */
607                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
608
609         /* Move the child from its dying parent to the new one.  */
610         if (unlikely(traced)) {
611                 /* Preserve ptrace links if someone else is tracing this child.  */
612                 list_del_init(&p->ptrace_list);
613                 if (p->parent != p->real_parent)
614                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
615         } else {
616                 /* If this child is being traced, then we're the one tracing it
617                  * anyway, so let go of it.
618                  */
619                 p->ptrace = 0;
620                 remove_parent(p);
621                 p->parent = p->real_parent;
622                 add_parent(p);
623
624                 /* If we'd notified the old parent about this child's death,
625                  * also notify the new parent.
626                  */
627                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
628                     thread_group_empty(p))
629                         do_notify_parent(p, p->exit_signal);
630                 else if (p->state == TASK_TRACED) {
631                         /*
632                          * If it was at a trace stop, turn it into
633                          * a normal stop since it's no longer being
634                          * traced.
635                          */
636                         ptrace_untrace(p);
637                 }
638         }
639
640         /*
641          * process group orphan check
642          * Case ii: Our child is in a different pgrp
643          * than we are, and it was the only connection
644          * outside, so the child pgrp is now orphaned.
645          */
646         if ((process_group(p) != process_group(father)) &&
647             (p->signal->session == father->signal->session)) {
648                 int pgrp = process_group(p);
649
650                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
651                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
652                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
653                 }
654         }
655 }
656
657 /*
658  * When we die, we re-parent all our children.
659  * Try to give them to another thread in our thread
660  * group, and if no such member exists, give it to
661  * the global child reaper process (ie "init")
662  */
663 static void forget_original_parent(struct task_struct * father,
664                                           struct list_head *to_release)
665 {
666         struct task_struct *p, *reaper = father;
667         struct list_head *_p, *_n;
668
669         do {
670                 reaper = next_thread(reaper);
671                 if (reaper == father) {
672                         reaper = child_reaper;
673                         break;
674                 }
675         } while (reaper->exit_state);
676
677         /*
678          * There are only two places where our children can be:
679          *
680          * - in our child list
681          * - in our ptraced child list
682          *
683          * Search them and reparent children.
684          */
685         list_for_each_safe(_p, _n, &father->children) {
686                 int ptrace;
687                 p = list_entry(_p,struct task_struct,sibling);
688
689                 ptrace = p->ptrace;
690
691                 /* if father isn't the real parent, then ptrace must be enabled */
692                 BUG_ON(father != p->real_parent && !ptrace);
693
694                 if (father == p->real_parent) {
695                         /* reparent with a reaper, real father it's us */
696                         choose_new_parent(p, reaper);
697                         reparent_thread(p, father, 0);
698                 } else {
699                         /* reparent ptraced task to its real parent */
700                         __ptrace_unlink (p);
701                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
702                             thread_group_empty(p))
703                                 do_notify_parent(p, p->exit_signal);
704                 }
705
706                 /*
707                  * if the ptraced child is a zombie with exit_signal == -1
708                  * we must collect it before we exit, or it will remain
709                  * zombie forever since we prevented it from self-reap itself
710                  * while it was being traced by us, to be able to see it in wait4.
711                  */
712                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
713                         list_add(&p->ptrace_list, to_release);
714         }
715         list_for_each_safe(_p, _n, &father->ptrace_children) {
716                 p = list_entry(_p,struct task_struct,ptrace_list);
717                 choose_new_parent(p, reaper);
718                 reparent_thread(p, father, 1);
719         }
720 }
721
722 /*
723  * Send signals to all our closest relatives so that they know
724  * to properly mourn us..
725  */
726 static void exit_notify(struct task_struct *tsk)
727 {
728         int state;
729         struct task_struct *t;
730         struct list_head ptrace_dead, *_p, *_n;
731
732         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
733             && !thread_group_empty(tsk)) {
734                 /*
735                  * This occurs when there was a race between our exit
736                  * syscall and a group signal choosing us as the one to
737                  * wake up.  It could be that we are the only thread
738                  * alerted to check for pending signals, but another thread
739                  * should be woken now to take the signal since we will not.
740                  * Now we'll wake all the threads in the group just to make
741                  * sure someone gets all the pending signals.
742                  */
743                 read_lock(&tasklist_lock);
744                 spin_lock_irq(&tsk->sighand->siglock);
745                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
746                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
747                                 recalc_sigpending_tsk(t);
748                                 if (signal_pending(t))
749                                         signal_wake_up(t, 0);
750                         }
751                 spin_unlock_irq(&tsk->sighand->siglock);
752                 read_unlock(&tasklist_lock);
753         }
754
755         write_lock_irq(&tasklist_lock);
756
757         /*
758          * This does two things:
759          *
760          * A.  Make init inherit all the child processes
761          * B.  Check to see if any process groups have become orphaned
762          *      as a result of our exiting, and if they have any stopped
763          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
764          */
765
766         INIT_LIST_HEAD(&ptrace_dead);
767         forget_original_parent(tsk, &ptrace_dead);
768         BUG_ON(!list_empty(&tsk->children));
769         BUG_ON(!list_empty(&tsk->ptrace_children));
770
771         /*
772          * Check to see if any process groups have become orphaned
773          * as a result of our exiting, and if they have any stopped
774          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
775          *
776          * Case i: Our father is in a different pgrp than we are
777          * and we were the only connection outside, so our pgrp
778          * is about to become orphaned.
779          */
780          
781         t = tsk->real_parent;
782         
783         if ((process_group(t) != process_group(tsk)) &&
784             (t->signal->session == tsk->signal->session) &&
785             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
786             has_stopped_jobs(process_group(tsk))) {
787                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
788                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
789         }
790
791         /* Let father know we died 
792          *
793          * Thread signals are configurable, but you aren't going to use
794          * that to send signals to arbitary processes. 
795          * That stops right now.
796          *
797          * If the parent exec id doesn't match the exec id we saved
798          * when we started then we know the parent has changed security
799          * domain.
800          *
801          * If our self_exec id doesn't match our parent_exec_id then
802          * we have changed execution domain as these two values started
803          * the same after a fork.
804          *      
805          */
806         
807         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
808             ( tsk->parent_exec_id != t->self_exec_id  ||
809               tsk->self_exec_id != tsk->parent_exec_id)
810             && !capable(CAP_KILL))
811                 tsk->exit_signal = SIGCHLD;
812
813
814         /* If something other than our normal parent is ptracing us, then
815          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
816          * only has special meaning to our real parent.
817          */
818         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
819                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
820                 do_notify_parent(tsk, signal);
821         } else if (tsk->ptrace) {
822                 do_notify_parent(tsk, SIGCHLD);
823         }
824
825         state = EXIT_ZOMBIE;
826         if (tsk->exit_signal == -1 &&
827             (likely(tsk->ptrace == 0) ||
828              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
829                 state = EXIT_DEAD;
830         tsk->exit_state = state;
831
832         write_unlock_irq(&tasklist_lock);
833
834         list_for_each_safe(_p, _n, &ptrace_dead) {
835                 list_del_init(_p);
836                 t = list_entry(_p,struct task_struct,ptrace_list);
837                 release_task(t);
838         }
839
840         /* If the process is dead, release it - nobody will wait for it */
841         if (state == EXIT_DEAD)
842                 release_task(tsk);
843 }
844
845 fastcall NORET_TYPE void do_exit(long code)
846 {
847         struct task_struct *tsk = current;
848         int group_dead;
849
850         profile_task_exit(tsk);
851
852         WARN_ON(atomic_read(&tsk->fs_excl));
853
854         if (unlikely(in_interrupt()))
855                 panic("Aiee, killing interrupt handler!");
856         if (unlikely(!tsk->pid))
857                 panic("Attempted to kill the idle task!");
858         if (unlikely(tsk == child_reaper))
859                 panic("Attempted to kill init!");
860
861         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
862                 current->ptrace_message = code;
863                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
864         }
865
866         /*
867          * We're taking recursive faults here in do_exit. Safest is to just
868          * leave this task alone and wait for reboot.
869          */
870         if (unlikely(tsk->flags & PF_EXITING)) {
871                 printk(KERN_ALERT
872                         "Fixing recursive fault but reboot is needed!\n");
873                 if (tsk->io_context)
874                         exit_io_context();
875                 set_current_state(TASK_UNINTERRUPTIBLE);
876                 schedule();
877         }
878
879         tsk->flags |= PF_EXITING;
880
881         /*
882          * Make sure we don't try to process any timer firings
883          * while we are already exiting.
884          */
885         tsk->it_virt_expires = cputime_zero;
886         tsk->it_prof_expires = cputime_zero;
887         tsk->it_sched_expires = 0;
888
889         if (unlikely(in_atomic()))
890                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
891                                 current->comm, current->pid,
892                                 preempt_count());
893
894         acct_update_integrals(tsk);
895         if (tsk->mm) {
896                 update_hiwater_rss(tsk->mm);
897                 update_hiwater_vm(tsk->mm);
898         }
899         group_dead = atomic_dec_and_test(&tsk->signal->live);
900         if (group_dead) {
901                 hrtimer_cancel(&tsk->signal->real_timer);
902                 exit_itimers(tsk->signal);
903                 acct_process(code);
904         }
905         if (unlikely(tsk->robust_list))
906                 exit_robust_list(tsk);
907 #ifdef CONFIG_COMPAT
908         if (unlikely(tsk->compat_robust_list))
909                 compat_exit_robust_list(tsk);
910 #endif
911         exit_mm(tsk);
912
913         exit_sem(tsk);
914         __exit_files(tsk);
915         __exit_fs(tsk);
916         exit_namespace(tsk);
917         exit_thread();
918         cpuset_exit(tsk);
919         exit_keys(tsk);
920
921         if (group_dead && tsk->signal->leader)
922                 disassociate_ctty(1);
923
924         module_put(task_thread_info(tsk)->exec_domain->module);
925         if (tsk->binfmt)
926                 module_put(tsk->binfmt->module);
927
928         tsk->exit_code = code;
929         proc_exit_connector(tsk);
930         exit_notify(tsk);
931 #ifdef CONFIG_NUMA
932         mpol_free(tsk->mempolicy);
933         tsk->mempolicy = NULL;
934 #endif
935         /*
936          * If DEBUG_MUTEXES is on, make sure we are holding no locks:
937          */
938         mutex_debug_check_no_locks_held(tsk);
939
940         if (tsk->io_context)
941                 exit_io_context();
942
943         /* PF_DEAD causes final put_task_struct after we schedule. */
944         preempt_disable();
945         BUG_ON(tsk->flags & PF_DEAD);
946         tsk->flags |= PF_DEAD;
947
948         schedule();
949         BUG();
950         /* Avoid "noreturn function does return".  */
951         for (;;) ;
952 }
953
954 EXPORT_SYMBOL_GPL(do_exit);
955
956 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
957 {
958         if (comp)
959                 complete(comp);
960         
961         do_exit(code);
962 }
963
964 EXPORT_SYMBOL(complete_and_exit);
965
966 asmlinkage long sys_exit(int error_code)
967 {
968         do_exit((error_code&0xff)<<8);
969 }
970
971 task_t fastcall *next_thread(const task_t *p)
972 {
973         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
974 }
975
976 EXPORT_SYMBOL(next_thread);
977
978 /*
979  * Take down every thread in the group.  This is called by fatal signals
980  * as well as by sys_exit_group (below).
981  */
982 NORET_TYPE void
983 do_group_exit(int exit_code)
984 {
985         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
986
987         if (current->signal->flags & SIGNAL_GROUP_EXIT)
988                 exit_code = current->signal->group_exit_code;
989         else if (!thread_group_empty(current)) {
990                 struct signal_struct *const sig = current->signal;
991                 struct sighand_struct *const sighand = current->sighand;
992                 read_lock(&tasklist_lock);
993                 spin_lock_irq(&sighand->siglock);
994                 if (sig->flags & SIGNAL_GROUP_EXIT)
995                         /* Another thread got here before we took the lock.  */
996                         exit_code = sig->group_exit_code;
997                 else {
998                         sig->group_exit_code = exit_code;
999                         zap_other_threads(current);
1000                 }
1001                 spin_unlock_irq(&sighand->siglock);
1002                 read_unlock(&tasklist_lock);
1003         }
1004
1005         do_exit(exit_code);
1006         /* NOTREACHED */
1007 }
1008
1009 /*
1010  * this kills every thread in the thread group. Note that any externally
1011  * wait4()-ing process will get the correct exit code - even if this
1012  * thread is not the thread group leader.
1013  */
1014 asmlinkage void sys_exit_group(int error_code)
1015 {
1016         do_group_exit((error_code & 0xff) << 8);
1017 }
1018
1019 static int eligible_child(pid_t pid, int options, task_t *p)
1020 {
1021         if (pid > 0) {
1022                 if (p->pid != pid)
1023                         return 0;
1024         } else if (!pid) {
1025                 if (process_group(p) != process_group(current))
1026                         return 0;
1027         } else if (pid != -1) {
1028                 if (process_group(p) != -pid)
1029                         return 0;
1030         }
1031
1032         /*
1033          * Do not consider detached threads that are
1034          * not ptraced:
1035          */
1036         if (p->exit_signal == -1 && !p->ptrace)
1037                 return 0;
1038
1039         /* Wait for all children (clone and not) if __WALL is set;
1040          * otherwise, wait for clone children *only* if __WCLONE is
1041          * set; otherwise, wait for non-clone children *only*.  (Note:
1042          * A "clone" child here is one that reports to its parent
1043          * using a signal other than SIGCHLD.) */
1044         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1045             && !(options & __WALL))
1046                 return 0;
1047         /*
1048          * Do not consider thread group leaders that are
1049          * in a non-empty thread group:
1050          */
1051         if (current->tgid != p->tgid && delay_group_leader(p))
1052                 return 2;
1053
1054         if (security_task_wait(p))
1055                 return 0;
1056
1057         return 1;
1058 }
1059
1060 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
1061                                int why, int status,
1062                                struct siginfo __user *infop,
1063                                struct rusage __user *rusagep)
1064 {
1065         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1066         put_task_struct(p);
1067         if (!retval)
1068                 retval = put_user(SIGCHLD, &infop->si_signo);
1069         if (!retval)
1070                 retval = put_user(0, &infop->si_errno);
1071         if (!retval)
1072                 retval = put_user((short)why, &infop->si_code);
1073         if (!retval)
1074                 retval = put_user(pid, &infop->si_pid);
1075         if (!retval)
1076                 retval = put_user(uid, &infop->si_uid);
1077         if (!retval)
1078                 retval = put_user(status, &infop->si_status);
1079         if (!retval)
1080                 retval = pid;
1081         return retval;
1082 }
1083
1084 /*
1085  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1086  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1087  * the lock and this task is uninteresting.  If we return nonzero, we have
1088  * released the lock and the system call should return.
1089  */
1090 static int wait_task_zombie(task_t *p, int noreap,
1091                             struct siginfo __user *infop,
1092                             int __user *stat_addr, struct rusage __user *ru)
1093 {
1094         unsigned long state;
1095         int retval;
1096         int status;
1097
1098         if (unlikely(noreap)) {
1099                 pid_t pid = p->pid;
1100                 uid_t uid = p->uid;
1101                 int exit_code = p->exit_code;
1102                 int why, status;
1103
1104                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1105                         return 0;
1106                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1107                         return 0;
1108                 get_task_struct(p);
1109                 read_unlock(&tasklist_lock);
1110                 if ((exit_code & 0x7f) == 0) {
1111                         why = CLD_EXITED;
1112                         status = exit_code >> 8;
1113                 } else {
1114                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1115                         status = exit_code & 0x7f;
1116                 }
1117                 return wait_noreap_copyout(p, pid, uid, why,
1118                                            status, infop, ru);
1119         }
1120
1121         /*
1122          * Try to move the task's state to DEAD
1123          * only one thread is allowed to do this:
1124          */
1125         state = xchg(&p->exit_state, EXIT_DEAD);
1126         if (state != EXIT_ZOMBIE) {
1127                 BUG_ON(state != EXIT_DEAD);
1128                 return 0;
1129         }
1130         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1131                 /*
1132                  * This can only happen in a race with a ptraced thread
1133                  * dying on another processor.
1134                  */
1135                 return 0;
1136         }
1137
1138         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1139                 struct signal_struct *psig;
1140                 struct signal_struct *sig;
1141
1142                 /*
1143                  * The resource counters for the group leader are in its
1144                  * own task_struct.  Those for dead threads in the group
1145                  * are in its signal_struct, as are those for the child
1146                  * processes it has previously reaped.  All these
1147                  * accumulate in the parent's signal_struct c* fields.
1148                  *
1149                  * We don't bother to take a lock here to protect these
1150                  * p->signal fields, because they are only touched by
1151                  * __exit_signal, which runs with tasklist_lock
1152                  * write-locked anyway, and so is excluded here.  We do
1153                  * need to protect the access to p->parent->signal fields,
1154                  * as other threads in the parent group can be right
1155                  * here reaping other children at the same time.
1156                  */
1157                 spin_lock_irq(&p->parent->sighand->siglock);
1158                 psig = p->parent->signal;
1159                 sig = p->signal;
1160                 psig->cutime =
1161                         cputime_add(psig->cutime,
1162                         cputime_add(p->utime,
1163                         cputime_add(sig->utime,
1164                                     sig->cutime)));
1165                 psig->cstime =
1166                         cputime_add(psig->cstime,
1167                         cputime_add(p->stime,
1168                         cputime_add(sig->stime,
1169                                     sig->cstime)));
1170                 psig->cmin_flt +=
1171                         p->min_flt + sig->min_flt + sig->cmin_flt;
1172                 psig->cmaj_flt +=
1173                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1174                 psig->cnvcsw +=
1175                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1176                 psig->cnivcsw +=
1177                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1178                 spin_unlock_irq(&p->parent->sighand->siglock);
1179         }
1180
1181         /*
1182          * Now we are sure this task is interesting, and no other
1183          * thread can reap it because we set its state to EXIT_DEAD.
1184          */
1185         read_unlock(&tasklist_lock);
1186
1187         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1188         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1189                 ? p->signal->group_exit_code : p->exit_code;
1190         if (!retval && stat_addr)
1191                 retval = put_user(status, stat_addr);
1192         if (!retval && infop)
1193                 retval = put_user(SIGCHLD, &infop->si_signo);
1194         if (!retval && infop)
1195                 retval = put_user(0, &infop->si_errno);
1196         if (!retval && infop) {
1197                 int why;
1198
1199                 if ((status & 0x7f) == 0) {
1200                         why = CLD_EXITED;
1201                         status >>= 8;
1202                 } else {
1203                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1204                         status &= 0x7f;
1205                 }
1206                 retval = put_user((short)why, &infop->si_code);
1207                 if (!retval)
1208                         retval = put_user(status, &infop->si_status);
1209         }
1210         if (!retval && infop)
1211                 retval = put_user(p->pid, &infop->si_pid);
1212         if (!retval && infop)
1213                 retval = put_user(p->uid, &infop->si_uid);
1214         if (retval) {
1215                 // TODO: is this safe?
1216                 p->exit_state = EXIT_ZOMBIE;
1217                 return retval;
1218         }
1219         retval = p->pid;
1220         if (p->real_parent != p->parent) {
1221                 write_lock_irq(&tasklist_lock);
1222                 /* Double-check with lock held.  */
1223                 if (p->real_parent != p->parent) {
1224                         __ptrace_unlink(p);
1225                         // TODO: is this safe?
1226                         p->exit_state = EXIT_ZOMBIE;
1227                         /*
1228                          * If this is not a detached task, notify the parent.
1229                          * If it's still not detached after that, don't release
1230                          * it now.
1231                          */
1232                         if (p->exit_signal != -1) {
1233                                 do_notify_parent(p, p->exit_signal);
1234                                 if (p->exit_signal != -1)
1235                                         p = NULL;
1236                         }
1237                 }
1238                 write_unlock_irq(&tasklist_lock);
1239         }
1240         if (p != NULL)
1241                 release_task(p);
1242         BUG_ON(!retval);
1243         return retval;
1244 }
1245
1246 /*
1247  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1248  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1249  * the lock and this task is uninteresting.  If we return nonzero, we have
1250  * released the lock and the system call should return.
1251  */
1252 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1253                              struct siginfo __user *infop,
1254                              int __user *stat_addr, struct rusage __user *ru)
1255 {
1256         int retval, exit_code;
1257
1258         if (!p->exit_code)
1259                 return 0;
1260         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1261             p->signal && p->signal->group_stop_count > 0)
1262                 /*
1263                  * A group stop is in progress and this is the group leader.
1264                  * We won't report until all threads have stopped.
1265                  */
1266                 return 0;
1267
1268         /*
1269          * Now we are pretty sure this task is interesting.
1270          * Make sure it doesn't get reaped out from under us while we
1271          * give up the lock and then examine it below.  We don't want to
1272          * keep holding onto the tasklist_lock while we call getrusage and
1273          * possibly take page faults for user memory.
1274          */
1275         get_task_struct(p);
1276         read_unlock(&tasklist_lock);
1277
1278         if (unlikely(noreap)) {
1279                 pid_t pid = p->pid;
1280                 uid_t uid = p->uid;
1281                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1282
1283                 exit_code = p->exit_code;
1284                 if (unlikely(!exit_code) ||
1285                     unlikely(p->state & TASK_TRACED))
1286                         goto bail_ref;
1287                 return wait_noreap_copyout(p, pid, uid,
1288                                            why, (exit_code << 8) | 0x7f,
1289                                            infop, ru);
1290         }
1291
1292         write_lock_irq(&tasklist_lock);
1293
1294         /*
1295          * This uses xchg to be atomic with the thread resuming and setting
1296          * it.  It must also be done with the write lock held to prevent a
1297          * race with the EXIT_ZOMBIE case.
1298          */
1299         exit_code = xchg(&p->exit_code, 0);
1300         if (unlikely(p->exit_state)) {
1301                 /*
1302                  * The task resumed and then died.  Let the next iteration
1303                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1304                  * already be zero here if it resumed and did _exit(0).
1305                  * The task itself is dead and won't touch exit_code again;
1306                  * other processors in this function are locked out.
1307                  */
1308                 p->exit_code = exit_code;
1309                 exit_code = 0;
1310         }
1311         if (unlikely(exit_code == 0)) {
1312                 /*
1313                  * Another thread in this function got to it first, or it
1314                  * resumed, or it resumed and then died.
1315                  */
1316                 write_unlock_irq(&tasklist_lock);
1317 bail_ref:
1318                 put_task_struct(p);
1319                 /*
1320                  * We are returning to the wait loop without having successfully
1321                  * removed the process and having released the lock. We cannot
1322                  * continue, since the "p" task pointer is potentially stale.
1323                  *
1324                  * Return -EAGAIN, and do_wait() will restart the loop from the
1325                  * beginning. Do _not_ re-acquire the lock.
1326                  */
1327                 return -EAGAIN;
1328         }
1329
1330         /* move to end of parent's list to avoid starvation */
1331         remove_parent(p);
1332         add_parent(p);
1333
1334         write_unlock_irq(&tasklist_lock);
1335
1336         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1337         if (!retval && stat_addr)
1338                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1339         if (!retval && infop)
1340                 retval = put_user(SIGCHLD, &infop->si_signo);
1341         if (!retval && infop)
1342                 retval = put_user(0, &infop->si_errno);
1343         if (!retval && infop)
1344                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1345                                           ? CLD_TRAPPED : CLD_STOPPED),
1346                                   &infop->si_code);
1347         if (!retval && infop)
1348                 retval = put_user(exit_code, &infop->si_status);
1349         if (!retval && infop)
1350                 retval = put_user(p->pid, &infop->si_pid);
1351         if (!retval && infop)
1352                 retval = put_user(p->uid, &infop->si_uid);
1353         if (!retval)
1354                 retval = p->pid;
1355         put_task_struct(p);
1356
1357         BUG_ON(!retval);
1358         return retval;
1359 }
1360
1361 /*
1362  * Handle do_wait work for one task in a live, non-stopped state.
1363  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1364  * the lock and this task is uninteresting.  If we return nonzero, we have
1365  * released the lock and the system call should return.
1366  */
1367 static int wait_task_continued(task_t *p, int noreap,
1368                                struct siginfo __user *infop,
1369                                int __user *stat_addr, struct rusage __user *ru)
1370 {
1371         int retval;
1372         pid_t pid;
1373         uid_t uid;
1374
1375         if (unlikely(!p->signal))
1376                 return 0;
1377
1378         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1379                 return 0;
1380
1381         spin_lock_irq(&p->sighand->siglock);
1382         /* Re-check with the lock held.  */
1383         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1384                 spin_unlock_irq(&p->sighand->siglock);
1385                 return 0;
1386         }
1387         if (!noreap)
1388                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1389         spin_unlock_irq(&p->sighand->siglock);
1390
1391         pid = p->pid;
1392         uid = p->uid;
1393         get_task_struct(p);
1394         read_unlock(&tasklist_lock);
1395
1396         if (!infop) {
1397                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1398                 put_task_struct(p);
1399                 if (!retval && stat_addr)
1400                         retval = put_user(0xffff, stat_addr);
1401                 if (!retval)
1402                         retval = p->pid;
1403         } else {
1404                 retval = wait_noreap_copyout(p, pid, uid,
1405                                              CLD_CONTINUED, SIGCONT,
1406                                              infop, ru);
1407                 BUG_ON(retval == 0);
1408         }
1409
1410         return retval;
1411 }
1412
1413
1414 static inline int my_ptrace_child(struct task_struct *p)
1415 {
1416         if (!(p->ptrace & PT_PTRACED))
1417                 return 0;
1418         if (!(p->ptrace & PT_ATTACHED))
1419                 return 1;
1420         /*
1421          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1422          * we are the attacher.  If we are the real parent, this is a race
1423          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1424          * which we have to switch the parent links, but has already set
1425          * the flags in p->ptrace.
1426          */
1427         return (p->parent != p->real_parent);
1428 }
1429
1430 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1431                     int __user *stat_addr, struct rusage __user *ru)
1432 {
1433         DECLARE_WAITQUEUE(wait, current);
1434         struct task_struct *tsk;
1435         int flag, retval;
1436
1437         add_wait_queue(&current->signal->wait_chldexit,&wait);
1438 repeat:
1439         /*
1440          * We will set this flag if we see any child that might later
1441          * match our criteria, even if we are not able to reap it yet.
1442          */
1443         flag = 0;
1444         current->state = TASK_INTERRUPTIBLE;
1445         read_lock(&tasklist_lock);
1446         tsk = current;
1447         do {
1448                 struct task_struct *p;
1449                 struct list_head *_p;
1450                 int ret;
1451
1452                 list_for_each(_p,&tsk->children) {
1453                         p = list_entry(_p,struct task_struct,sibling);
1454
1455                         ret = eligible_child(pid, options, p);
1456                         if (!ret)
1457                                 continue;
1458
1459                         switch (p->state) {
1460                         case TASK_TRACED:
1461                                 /*
1462                                  * When we hit the race with PTRACE_ATTACH,
1463                                  * we will not report this child.  But the
1464                                  * race means it has not yet been moved to
1465                                  * our ptrace_children list, so we need to
1466                                  * set the flag here to avoid a spurious ECHILD
1467                                  * when the race happens with the only child.
1468                                  */
1469                                 flag = 1;
1470                                 if (!my_ptrace_child(p))
1471                                         continue;
1472                                 /*FALLTHROUGH*/
1473                         case TASK_STOPPED:
1474                                 /*
1475                                  * It's stopped now, so it might later
1476                                  * continue, exit, or stop again.
1477                                  */
1478                                 flag = 1;
1479                                 if (!(options & WUNTRACED) &&
1480                                     !my_ptrace_child(p))
1481                                         continue;
1482                                 retval = wait_task_stopped(p, ret == 2,
1483                                                            (options & WNOWAIT),
1484                                                            infop,
1485                                                            stat_addr, ru);
1486                                 if (retval == -EAGAIN)
1487                                         goto repeat;
1488                                 if (retval != 0) /* He released the lock.  */
1489                                         goto end;
1490                                 break;
1491                         default:
1492                         // case EXIT_DEAD:
1493                                 if (p->exit_state == EXIT_DEAD)
1494                                         continue;
1495                         // case EXIT_ZOMBIE:
1496                                 if (p->exit_state == EXIT_ZOMBIE) {
1497                                         /*
1498                                          * Eligible but we cannot release
1499                                          * it yet:
1500                                          */
1501                                         if (ret == 2)
1502                                                 goto check_continued;
1503                                         if (!likely(options & WEXITED))
1504                                                 continue;
1505                                         retval = wait_task_zombie(
1506                                                 p, (options & WNOWAIT),
1507                                                 infop, stat_addr, ru);
1508                                         /* He released the lock.  */
1509                                         if (retval != 0)
1510                                                 goto end;
1511                                         break;
1512                                 }
1513 check_continued:
1514                                 /*
1515                                  * It's running now, so it might later
1516                                  * exit, stop, or stop and then continue.
1517                                  */
1518                                 flag = 1;
1519                                 if (!unlikely(options & WCONTINUED))
1520                                         continue;
1521                                 retval = wait_task_continued(
1522                                         p, (options & WNOWAIT),
1523                                         infop, stat_addr, ru);
1524                                 if (retval != 0) /* He released the lock.  */
1525                                         goto end;
1526                                 break;
1527                         }
1528                 }
1529                 if (!flag) {
1530                         list_for_each(_p, &tsk->ptrace_children) {
1531                                 p = list_entry(_p, struct task_struct,
1532                                                 ptrace_list);
1533                                 if (!eligible_child(pid, options, p))
1534                                         continue;
1535                                 flag = 1;
1536                                 break;
1537                         }
1538                 }
1539                 if (options & __WNOTHREAD)
1540                         break;
1541                 tsk = next_thread(tsk);
1542                 if (tsk->signal != current->signal)
1543                         BUG();
1544         } while (tsk != current);
1545
1546         read_unlock(&tasklist_lock);
1547         if (flag) {
1548                 retval = 0;
1549                 if (options & WNOHANG)
1550                         goto end;
1551                 retval = -ERESTARTSYS;
1552                 if (signal_pending(current))
1553                         goto end;
1554                 schedule();
1555                 goto repeat;
1556         }
1557         retval = -ECHILD;
1558 end:
1559         current->state = TASK_RUNNING;
1560         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1561         if (infop) {
1562                 if (retval > 0)
1563                 retval = 0;
1564                 else {
1565                         /*
1566                          * For a WNOHANG return, clear out all the fields
1567                          * we would set so the user can easily tell the
1568                          * difference.
1569                          */
1570                         if (!retval)
1571                                 retval = put_user(0, &infop->si_signo);
1572                         if (!retval)
1573                                 retval = put_user(0, &infop->si_errno);
1574                         if (!retval)
1575                                 retval = put_user(0, &infop->si_code);
1576                         if (!retval)
1577                                 retval = put_user(0, &infop->si_pid);
1578                         if (!retval)
1579                                 retval = put_user(0, &infop->si_uid);
1580                         if (!retval)
1581                                 retval = put_user(0, &infop->si_status);
1582                 }
1583         }
1584         return retval;
1585 }
1586
1587 asmlinkage long sys_waitid(int which, pid_t pid,
1588                            struct siginfo __user *infop, int options,
1589                            struct rusage __user *ru)
1590 {
1591         long ret;
1592
1593         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1594                 return -EINVAL;
1595         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1596                 return -EINVAL;
1597
1598         switch (which) {
1599         case P_ALL:
1600                 pid = -1;
1601                 break;
1602         case P_PID:
1603                 if (pid <= 0)
1604                         return -EINVAL;
1605                 break;
1606         case P_PGID:
1607                 if (pid <= 0)
1608                         return -EINVAL;
1609                 pid = -pid;
1610                 break;
1611         default:
1612                 return -EINVAL;
1613         }
1614
1615         ret = do_wait(pid, options, infop, NULL, ru);
1616
1617         /* avoid REGPARM breakage on x86: */
1618         prevent_tail_call(ret);
1619         return ret;
1620 }
1621
1622 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1623                           int options, struct rusage __user *ru)
1624 {
1625         long ret;
1626
1627         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1628                         __WNOTHREAD|__WCLONE|__WALL))
1629                 return -EINVAL;
1630         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1631
1632         /* avoid REGPARM breakage on x86: */
1633         prevent_tail_call(ret);
1634         return ret;
1635 }
1636
1637 #ifdef __ARCH_WANT_SYS_WAITPID
1638
1639 /*
1640  * sys_waitpid() remains for compatibility. waitpid() should be
1641  * implemented by calling sys_wait4() from libc.a.
1642  */
1643 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1644 {
1645         return sys_wait4(pid, stat_addr, options, NULL);
1646 }
1647
1648 #endif