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