[PATCH] files: files struct with RCU
[linux-3.10.git] / kernel / exit.c
1 /*
2  *  linux/kernel/exit.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 #include <linux/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/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
32 #include <asm/uaccess.h>
33 #include <asm/unistd.h>
34 #include <asm/pgtable.h>
35 #include <asm/mmu_context.h>
36
37 extern void sem_exit (void);
38 extern struct task_struct *child_reaper;
39
40 int getrusage(struct task_struct *, int, struct rusage __user *);
41
42 static void exit_mm(struct task_struct * tsk);
43
44 static void __unhash_process(struct task_struct *p)
45 {
46         nr_threads--;
47         detach_pid(p, PIDTYPE_PID);
48         detach_pid(p, PIDTYPE_TGID);
49         if (thread_group_leader(p)) {
50                 detach_pid(p, PIDTYPE_PGID);
51                 detach_pid(p, PIDTYPE_SID);
52                 if (p->pid)
53                         __get_cpu_var(process_counts)--;
54         }
55
56         REMOVE_LINKS(p);
57 }
58
59 void release_task(struct task_struct * p)
60 {
61         int zap_leader;
62         task_t *leader;
63         struct dentry *proc_dentry;
64
65 repeat: 
66         atomic_dec(&p->user->processes);
67         spin_lock(&p->proc_lock);
68         proc_dentry = proc_pid_unhash(p);
69         write_lock_irq(&tasklist_lock);
70         if (unlikely(p->ptrace))
71                 __ptrace_unlink(p);
72         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
73         __exit_signal(p);
74         __exit_sighand(p);
75         /*
76          * Note that the fastpath in sys_times depends on __exit_signal having
77          * updated the counters before a task is removed from the tasklist of
78          * the process by __unhash_process.
79          */
80         __unhash_process(p);
81
82         /*
83          * If we are the last non-leader member of the thread
84          * group, and the leader is zombie, then notify the
85          * group leader's parent process. (if it wants notification.)
86          */
87         zap_leader = 0;
88         leader = p->group_leader;
89         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
90                 BUG_ON(leader->exit_signal == -1);
91                 do_notify_parent(leader, leader->exit_signal);
92                 /*
93                  * If we were the last child thread and the leader has
94                  * exited already, and the leader's parent ignores SIGCHLD,
95                  * then we are the one who should release the leader.
96                  *
97                  * do_notify_parent() will have marked it self-reaping in
98                  * that case.
99                  */
100                 zap_leader = (leader->exit_signal == -1);
101         }
102
103         sched_exit(p);
104         write_unlock_irq(&tasklist_lock);
105         spin_unlock(&p->proc_lock);
106         proc_pid_flush(proc_dentry);
107         release_thread(p);
108         put_task_struct(p);
109
110         p = leader;
111         if (unlikely(zap_leader))
112                 goto repeat;
113 }
114
115 /* we are using it only for SMP init */
116
117 void unhash_process(struct task_struct *p)
118 {
119         struct dentry *proc_dentry;
120
121         spin_lock(&p->proc_lock);
122         proc_dentry = proc_pid_unhash(p);
123         write_lock_irq(&tasklist_lock);
124         __unhash_process(p);
125         write_unlock_irq(&tasklist_lock);
126         spin_unlock(&p->proc_lock);
127         proc_pid_flush(proc_dentry);
128 }
129
130 /*
131  * This checks not only the pgrp, but falls back on the pid if no
132  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
133  * without this...
134  */
135 int session_of_pgrp(int pgrp)
136 {
137         struct task_struct *p;
138         int sid = -1;
139
140         read_lock(&tasklist_lock);
141         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
142                 if (p->signal->session > 0) {
143                         sid = p->signal->session;
144                         goto out;
145                 }
146         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
147         p = find_task_by_pid(pgrp);
148         if (p)
149                 sid = p->signal->session;
150 out:
151         read_unlock(&tasklist_lock);
152         
153         return sid;
154 }
155
156 /*
157  * Determine if a process group is "orphaned", according to the POSIX
158  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
159  * by terminal-generated stop signals.  Newly orphaned process groups are
160  * to receive a SIGHUP and a SIGCONT.
161  *
162  * "I ask you, have you ever known what it is to be an orphan?"
163  */
164 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
165 {
166         struct task_struct *p;
167         int ret = 1;
168
169         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
170                 if (p == ignored_task
171                                 || p->exit_state
172                                 || p->real_parent->pid == 1)
173                         continue;
174                 if (process_group(p->real_parent) != pgrp
175                             && p->real_parent->signal->session == p->signal->session) {
176                         ret = 0;
177                         break;
178                 }
179         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
180         return ret;     /* (sighing) "Often!" */
181 }
182
183 int is_orphaned_pgrp(int pgrp)
184 {
185         int retval;
186
187         read_lock(&tasklist_lock);
188         retval = will_become_orphaned_pgrp(pgrp, NULL);
189         read_unlock(&tasklist_lock);
190
191         return retval;
192 }
193
194 static inline int has_stopped_jobs(int pgrp)
195 {
196         int retval = 0;
197         struct task_struct *p;
198
199         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
200                 if (p->state != TASK_STOPPED)
201                         continue;
202
203                 /* If p is stopped by a debugger on a signal that won't
204                    stop it, then don't count p as stopped.  This isn't
205                    perfect but it's a good approximation.  */
206                 if (unlikely (p->ptrace)
207                     && p->exit_code != SIGSTOP
208                     && p->exit_code != SIGTSTP
209                     && p->exit_code != SIGTTOU
210                     && p->exit_code != SIGTTIN)
211                         continue;
212
213                 retval = 1;
214                 break;
215         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
216         return retval;
217 }
218
219 /**
220  * reparent_to_init - Reparent the calling kernel thread to the init task.
221  *
222  * If a kernel thread is launched as a result of a system call, or if
223  * it ever exits, it should generally reparent itself to init so that
224  * it is correctly cleaned up on exit.
225  *
226  * The various task state such as scheduling policy and priority may have
227  * been inherited from a user process, so we reset them to sane values here.
228  *
229  * NOTE that reparent_to_init() gives the caller full capabilities.
230  */
231 static inline void reparent_to_init(void)
232 {
233         write_lock_irq(&tasklist_lock);
234
235         ptrace_unlink(current);
236         /* Reparent to init */
237         REMOVE_LINKS(current);
238         current->parent = child_reaper;
239         current->real_parent = child_reaper;
240         SET_LINKS(current);
241
242         /* Set the exit signal to SIGCHLD so we signal init on exit */
243         current->exit_signal = SIGCHLD;
244
245         if ((current->policy == SCHED_NORMAL) && (task_nice(current) < 0))
246                 set_user_nice(current, 0);
247         /* cpus_allowed? */
248         /* rt_priority? */
249         /* signals? */
250         security_task_reparent_to_init(current);
251         memcpy(current->signal->rlim, init_task.signal->rlim,
252                sizeof(current->signal->rlim));
253         atomic_inc(&(INIT_USER->__count));
254         write_unlock_irq(&tasklist_lock);
255         switch_uid(INIT_USER);
256 }
257
258 void __set_special_pids(pid_t session, pid_t pgrp)
259 {
260         struct task_struct *curr = current;
261
262         if (curr->signal->session != session) {
263                 detach_pid(curr, PIDTYPE_SID);
264                 curr->signal->session = session;
265                 attach_pid(curr, PIDTYPE_SID, session);
266         }
267         if (process_group(curr) != pgrp) {
268                 detach_pid(curr, PIDTYPE_PGID);
269                 curr->signal->pgrp = pgrp;
270                 attach_pid(curr, PIDTYPE_PGID, pgrp);
271         }
272 }
273
274 void set_special_pids(pid_t session, pid_t pgrp)
275 {
276         write_lock_irq(&tasklist_lock);
277         __set_special_pids(session, pgrp);
278         write_unlock_irq(&tasklist_lock);
279 }
280
281 /*
282  * Let kernel threads use this to say that they
283  * allow a certain signal (since daemonize() will
284  * have disabled all of them by default).
285  */
286 int allow_signal(int sig)
287 {
288         if (!valid_signal(sig) || sig < 1)
289                 return -EINVAL;
290
291         spin_lock_irq(&current->sighand->siglock);
292         sigdelset(&current->blocked, sig);
293         if (!current->mm) {
294                 /* Kernel threads handle their own signals.
295                    Let the signal code know it'll be handled, so
296                    that they don't get converted to SIGKILL or
297                    just silently dropped */
298                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
299         }
300         recalc_sigpending();
301         spin_unlock_irq(&current->sighand->siglock);
302         return 0;
303 }
304
305 EXPORT_SYMBOL(allow_signal);
306
307 int disallow_signal(int sig)
308 {
309         if (!valid_signal(sig) || sig < 1)
310                 return -EINVAL;
311
312         spin_lock_irq(&current->sighand->siglock);
313         sigaddset(&current->blocked, sig);
314         recalc_sigpending();
315         spin_unlock_irq(&current->sighand->siglock);
316         return 0;
317 }
318
319 EXPORT_SYMBOL(disallow_signal);
320
321 /*
322  *      Put all the gunge required to become a kernel thread without
323  *      attached user resources in one place where it belongs.
324  */
325
326 void daemonize(const char *name, ...)
327 {
328         va_list args;
329         struct fs_struct *fs;
330         sigset_t blocked;
331
332         va_start(args, name);
333         vsnprintf(current->comm, sizeof(current->comm), name, args);
334         va_end(args);
335
336         /*
337          * If we were started as result of loading a module, close all of the
338          * user space pages.  We don't need them, and if we didn't close them
339          * they would be locked into memory.
340          */
341         exit_mm(current);
342
343         set_special_pids(1, 1);
344         down(&tty_sem);
345         current->signal->tty = NULL;
346         up(&tty_sem);
347
348         /* Block and flush all signals */
349         sigfillset(&blocked);
350         sigprocmask(SIG_BLOCK, &blocked, NULL);
351         flush_signals(current);
352
353         /* Become as one with the init task */
354
355         exit_fs(current);       /* current->fs->count--; */
356         fs = init_task.fs;
357         current->fs = fs;
358         atomic_inc(&fs->count);
359         exit_files(current);
360         current->files = init_task.files;
361         atomic_inc(&current->files->count);
362
363         reparent_to_init();
364 }
365
366 EXPORT_SYMBOL(daemonize);
367
368 static inline void close_files(struct files_struct * files)
369 {
370         int i, j;
371         struct fdtable *fdt;
372
373         j = 0;
374         fdt = files_fdtable(files);
375         for (;;) {
376                 unsigned long set;
377                 i = j * __NFDBITS;
378                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
379                         break;
380                 set = fdt->open_fds->fds_bits[j++];
381                 while (set) {
382                         if (set & 1) {
383                                 struct file * file = xchg(&fdt->fd[i], NULL);
384                                 if (file)
385                                         filp_close(file, files);
386                         }
387                         i++;
388                         set >>= 1;
389                 }
390         }
391 }
392
393 struct files_struct *get_files_struct(struct task_struct *task)
394 {
395         struct files_struct *files;
396
397         task_lock(task);
398         files = task->files;
399         if (files)
400                 atomic_inc(&files->count);
401         task_unlock(task);
402
403         return files;
404 }
405
406 void fastcall put_files_struct(struct files_struct *files)
407 {
408         struct fdtable *fdt;
409
410         if (atomic_dec_and_test(&files->count)) {
411                 close_files(files);
412                 /*
413                  * Free the fd and fdset arrays if we expanded them.
414                  * If the fdtable was embedded, pass files for freeing
415                  * at the end of the RCU grace period. Otherwise,
416                  * you can free files immediately.
417                  */
418                 fdt = files_fdtable(files);
419                 if (fdt == &files->fdtab)
420                         fdt->free_files = files;
421                 else
422                         kmem_cache_free(files_cachep, files);
423                 free_fdtable(fdt);
424         }
425 }
426
427 EXPORT_SYMBOL(put_files_struct);
428
429 static inline void __exit_files(struct task_struct *tsk)
430 {
431         struct files_struct * files = tsk->files;
432
433         if (files) {
434                 task_lock(tsk);
435                 tsk->files = NULL;
436                 task_unlock(tsk);
437                 put_files_struct(files);
438         }
439 }
440
441 void exit_files(struct task_struct *tsk)
442 {
443         __exit_files(tsk);
444 }
445
446 static inline void __put_fs_struct(struct fs_struct *fs)
447 {
448         /* No need to hold fs->lock if we are killing it */
449         if (atomic_dec_and_test(&fs->count)) {
450                 dput(fs->root);
451                 mntput(fs->rootmnt);
452                 dput(fs->pwd);
453                 mntput(fs->pwdmnt);
454                 if (fs->altroot) {
455                         dput(fs->altroot);
456                         mntput(fs->altrootmnt);
457                 }
458                 kmem_cache_free(fs_cachep, fs);
459         }
460 }
461
462 void put_fs_struct(struct fs_struct *fs)
463 {
464         __put_fs_struct(fs);
465 }
466
467 static inline void __exit_fs(struct task_struct *tsk)
468 {
469         struct fs_struct * fs = tsk->fs;
470
471         if (fs) {
472                 task_lock(tsk);
473                 tsk->fs = NULL;
474                 task_unlock(tsk);
475                 __put_fs_struct(fs);
476         }
477 }
478
479 void exit_fs(struct task_struct *tsk)
480 {
481         __exit_fs(tsk);
482 }
483
484 EXPORT_SYMBOL_GPL(exit_fs);
485
486 /*
487  * Turn us into a lazy TLB process if we
488  * aren't already..
489  */
490 static void exit_mm(struct task_struct * tsk)
491 {
492         struct mm_struct *mm = tsk->mm;
493
494         mm_release(tsk, mm);
495         if (!mm)
496                 return;
497         /*
498          * Serialize with any possible pending coredump.
499          * We must hold mmap_sem around checking core_waiters
500          * and clearing tsk->mm.  The core-inducing thread
501          * will increment core_waiters for each thread in the
502          * group with ->mm != NULL.
503          */
504         down_read(&mm->mmap_sem);
505         if (mm->core_waiters) {
506                 up_read(&mm->mmap_sem);
507                 down_write(&mm->mmap_sem);
508                 if (!--mm->core_waiters)
509                         complete(mm->core_startup_done);
510                 up_write(&mm->mmap_sem);
511
512                 wait_for_completion(&mm->core_done);
513                 down_read(&mm->mmap_sem);
514         }
515         atomic_inc(&mm->mm_count);
516         if (mm != tsk->active_mm) BUG();
517         /* more a memory barrier than a real lock */
518         task_lock(tsk);
519         tsk->mm = NULL;
520         up_read(&mm->mmap_sem);
521         enter_lazy_tlb(mm, current);
522         task_unlock(tsk);
523         mmput(mm);
524 }
525
526 static inline void choose_new_parent(task_t *p, task_t *reaper, task_t *child_reaper)
527 {
528         /*
529          * Make sure we're not reparenting to ourselves and that
530          * the parent is not a zombie.
531          */
532         BUG_ON(p == reaper || reaper->exit_state >= EXIT_ZOMBIE);
533         p->real_parent = reaper;
534 }
535
536 static inline void reparent_thread(task_t *p, task_t *father, int traced)
537 {
538         /* We don't want people slaying init.  */
539         if (p->exit_signal != -1)
540                 p->exit_signal = SIGCHLD;
541
542         if (p->pdeath_signal)
543                 /* We already hold the tasklist_lock here.  */
544                 group_send_sig_info(p->pdeath_signal, (void *) 0, p);
545
546         /* Move the child from its dying parent to the new one.  */
547         if (unlikely(traced)) {
548                 /* Preserve ptrace links if someone else is tracing this child.  */
549                 list_del_init(&p->ptrace_list);
550                 if (p->parent != p->real_parent)
551                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
552         } else {
553                 /* If this child is being traced, then we're the one tracing it
554                  * anyway, so let go of it.
555                  */
556                 p->ptrace = 0;
557                 list_del_init(&p->sibling);
558                 p->parent = p->real_parent;
559                 list_add_tail(&p->sibling, &p->parent->children);
560
561                 /* If we'd notified the old parent about this child's death,
562                  * also notify the new parent.
563                  */
564                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
565                     thread_group_empty(p))
566                         do_notify_parent(p, p->exit_signal);
567                 else if (p->state == TASK_TRACED) {
568                         /*
569                          * If it was at a trace stop, turn it into
570                          * a normal stop since it's no longer being
571                          * traced.
572                          */
573                         ptrace_untrace(p);
574                 }
575         }
576
577         /*
578          * process group orphan check
579          * Case ii: Our child is in a different pgrp
580          * than we are, and it was the only connection
581          * outside, so the child pgrp is now orphaned.
582          */
583         if ((process_group(p) != process_group(father)) &&
584             (p->signal->session == father->signal->session)) {
585                 int pgrp = process_group(p);
586
587                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
588                         __kill_pg_info(SIGHUP, (void *)1, pgrp);
589                         __kill_pg_info(SIGCONT, (void *)1, pgrp);
590                 }
591         }
592 }
593
594 /*
595  * When we die, we re-parent all our children.
596  * Try to give them to another thread in our thread
597  * group, and if no such member exists, give it to
598  * the global child reaper process (ie "init")
599  */
600 static inline void forget_original_parent(struct task_struct * father,
601                                           struct list_head *to_release)
602 {
603         struct task_struct *p, *reaper = father;
604         struct list_head *_p, *_n;
605
606         do {
607                 reaper = next_thread(reaper);
608                 if (reaper == father) {
609                         reaper = child_reaper;
610                         break;
611                 }
612         } while (reaper->exit_state);
613
614         /*
615          * There are only two places where our children can be:
616          *
617          * - in our child list
618          * - in our ptraced child list
619          *
620          * Search them and reparent children.
621          */
622         list_for_each_safe(_p, _n, &father->children) {
623                 int ptrace;
624                 p = list_entry(_p,struct task_struct,sibling);
625
626                 ptrace = p->ptrace;
627
628                 /* if father isn't the real parent, then ptrace must be enabled */
629                 BUG_ON(father != p->real_parent && !ptrace);
630
631                 if (father == p->real_parent) {
632                         /* reparent with a reaper, real father it's us */
633                         choose_new_parent(p, reaper, child_reaper);
634                         reparent_thread(p, father, 0);
635                 } else {
636                         /* reparent ptraced task to its real parent */
637                         __ptrace_unlink (p);
638                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
639                             thread_group_empty(p))
640                                 do_notify_parent(p, p->exit_signal);
641                 }
642
643                 /*
644                  * if the ptraced child is a zombie with exit_signal == -1
645                  * we must collect it before we exit, or it will remain
646                  * zombie forever since we prevented it from self-reap itself
647                  * while it was being traced by us, to be able to see it in wait4.
648                  */
649                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
650                         list_add(&p->ptrace_list, to_release);
651         }
652         list_for_each_safe(_p, _n, &father->ptrace_children) {
653                 p = list_entry(_p,struct task_struct,ptrace_list);
654                 choose_new_parent(p, reaper, child_reaper);
655                 reparent_thread(p, father, 1);
656         }
657 }
658
659 /*
660  * Send signals to all our closest relatives so that they know
661  * to properly mourn us..
662  */
663 static void exit_notify(struct task_struct *tsk)
664 {
665         int state;
666         struct task_struct *t;
667         struct list_head ptrace_dead, *_p, *_n;
668
669         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
670             && !thread_group_empty(tsk)) {
671                 /*
672                  * This occurs when there was a race between our exit
673                  * syscall and a group signal choosing us as the one to
674                  * wake up.  It could be that we are the only thread
675                  * alerted to check for pending signals, but another thread
676                  * should be woken now to take the signal since we will not.
677                  * Now we'll wake all the threads in the group just to make
678                  * sure someone gets all the pending signals.
679                  */
680                 read_lock(&tasklist_lock);
681                 spin_lock_irq(&tsk->sighand->siglock);
682                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
683                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
684                                 recalc_sigpending_tsk(t);
685                                 if (signal_pending(t))
686                                         signal_wake_up(t, 0);
687                         }
688                 spin_unlock_irq(&tsk->sighand->siglock);
689                 read_unlock(&tasklist_lock);
690         }
691
692         write_lock_irq(&tasklist_lock);
693
694         /*
695          * This does two things:
696          *
697          * A.  Make init inherit all the child processes
698          * B.  Check to see if any process groups have become orphaned
699          *      as a result of our exiting, and if they have any stopped
700          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
701          */
702
703         INIT_LIST_HEAD(&ptrace_dead);
704         forget_original_parent(tsk, &ptrace_dead);
705         BUG_ON(!list_empty(&tsk->children));
706         BUG_ON(!list_empty(&tsk->ptrace_children));
707
708         /*
709          * Check to see if any process groups have become orphaned
710          * as a result of our exiting, and if they have any stopped
711          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
712          *
713          * Case i: Our father is in a different pgrp than we are
714          * and we were the only connection outside, so our pgrp
715          * is about to become orphaned.
716          */
717          
718         t = tsk->real_parent;
719         
720         if ((process_group(t) != process_group(tsk)) &&
721             (t->signal->session == tsk->signal->session) &&
722             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
723             has_stopped_jobs(process_group(tsk))) {
724                 __kill_pg_info(SIGHUP, (void *)1, process_group(tsk));
725                 __kill_pg_info(SIGCONT, (void *)1, process_group(tsk));
726         }
727
728         /* Let father know we died 
729          *
730          * Thread signals are configurable, but you aren't going to use
731          * that to send signals to arbitary processes. 
732          * That stops right now.
733          *
734          * If the parent exec id doesn't match the exec id we saved
735          * when we started then we know the parent has changed security
736          * domain.
737          *
738          * If our self_exec id doesn't match our parent_exec_id then
739          * we have changed execution domain as these two values started
740          * the same after a fork.
741          *      
742          */
743         
744         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
745             ( tsk->parent_exec_id != t->self_exec_id  ||
746               tsk->self_exec_id != tsk->parent_exec_id)
747             && !capable(CAP_KILL))
748                 tsk->exit_signal = SIGCHLD;
749
750
751         /* If something other than our normal parent is ptracing us, then
752          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
753          * only has special meaning to our real parent.
754          */
755         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
756                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
757                 do_notify_parent(tsk, signal);
758         } else if (tsk->ptrace) {
759                 do_notify_parent(tsk, SIGCHLD);
760         }
761
762         state = EXIT_ZOMBIE;
763         if (tsk->exit_signal == -1 &&
764             (likely(tsk->ptrace == 0) ||
765              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
766                 state = EXIT_DEAD;
767         tsk->exit_state = state;
768
769         write_unlock_irq(&tasklist_lock);
770
771         list_for_each_safe(_p, _n, &ptrace_dead) {
772                 list_del_init(_p);
773                 t = list_entry(_p,struct task_struct,ptrace_list);
774                 release_task(t);
775         }
776
777         /* If the process is dead, release it - nobody will wait for it */
778         if (state == EXIT_DEAD)
779                 release_task(tsk);
780
781         /* PF_DEAD causes final put_task_struct after we schedule. */
782         preempt_disable();
783         tsk->flags |= PF_DEAD;
784 }
785
786 fastcall NORET_TYPE void do_exit(long code)
787 {
788         struct task_struct *tsk = current;
789         int group_dead;
790
791         profile_task_exit(tsk);
792
793         WARN_ON(atomic_read(&tsk->fs_excl));
794
795         if (unlikely(in_interrupt()))
796                 panic("Aiee, killing interrupt handler!");
797         if (unlikely(!tsk->pid))
798                 panic("Attempted to kill the idle task!");
799         if (unlikely(tsk->pid == 1))
800                 panic("Attempted to kill init!");
801         if (tsk->io_context)
802                 exit_io_context();
803
804         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
805                 current->ptrace_message = code;
806                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
807         }
808
809         /*
810          * We're taking recursive faults here in do_exit. Safest is to just
811          * leave this task alone and wait for reboot.
812          */
813         if (unlikely(tsk->flags & PF_EXITING)) {
814                 printk(KERN_ALERT
815                         "Fixing recursive fault but reboot is needed!\n");
816                 set_current_state(TASK_UNINTERRUPTIBLE);
817                 schedule();
818         }
819
820         tsk->flags |= PF_EXITING;
821
822         /*
823          * Make sure we don't try to process any timer firings
824          * while we are already exiting.
825          */
826         tsk->it_virt_expires = cputime_zero;
827         tsk->it_prof_expires = cputime_zero;
828         tsk->it_sched_expires = 0;
829
830         if (unlikely(in_atomic()))
831                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
832                                 current->comm, current->pid,
833                                 preempt_count());
834
835         acct_update_integrals(tsk);
836         update_mem_hiwater(tsk);
837         group_dead = atomic_dec_and_test(&tsk->signal->live);
838         if (group_dead) {
839                 del_timer_sync(&tsk->signal->real_timer);
840                 acct_process(code);
841         }
842         exit_mm(tsk);
843
844         exit_sem(tsk);
845         __exit_files(tsk);
846         __exit_fs(tsk);
847         exit_namespace(tsk);
848         exit_thread();
849         cpuset_exit(tsk);
850         exit_keys(tsk);
851
852         if (group_dead && tsk->signal->leader)
853                 disassociate_ctty(1);
854
855         module_put(tsk->thread_info->exec_domain->module);
856         if (tsk->binfmt)
857                 module_put(tsk->binfmt->module);
858
859         tsk->exit_code = code;
860         exit_notify(tsk);
861 #ifdef CONFIG_NUMA
862         mpol_free(tsk->mempolicy);
863         tsk->mempolicy = NULL;
864 #endif
865
866         BUG_ON(!(current->flags & PF_DEAD));
867         schedule();
868         BUG();
869         /* Avoid "noreturn function does return".  */
870         for (;;) ;
871 }
872
873 EXPORT_SYMBOL_GPL(do_exit);
874
875 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
876 {
877         if (comp)
878                 complete(comp);
879         
880         do_exit(code);
881 }
882
883 EXPORT_SYMBOL(complete_and_exit);
884
885 asmlinkage long sys_exit(int error_code)
886 {
887         do_exit((error_code&0xff)<<8);
888 }
889
890 task_t fastcall *next_thread(const task_t *p)
891 {
892         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
893 }
894
895 EXPORT_SYMBOL(next_thread);
896
897 /*
898  * Take down every thread in the group.  This is called by fatal signals
899  * as well as by sys_exit_group (below).
900  */
901 NORET_TYPE void
902 do_group_exit(int exit_code)
903 {
904         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
905
906         if (current->signal->flags & SIGNAL_GROUP_EXIT)
907                 exit_code = current->signal->group_exit_code;
908         else if (!thread_group_empty(current)) {
909                 struct signal_struct *const sig = current->signal;
910                 struct sighand_struct *const sighand = current->sighand;
911                 read_lock(&tasklist_lock);
912                 spin_lock_irq(&sighand->siglock);
913                 if (sig->flags & SIGNAL_GROUP_EXIT)
914                         /* Another thread got here before we took the lock.  */
915                         exit_code = sig->group_exit_code;
916                 else {
917                         sig->flags = SIGNAL_GROUP_EXIT;
918                         sig->group_exit_code = exit_code;
919                         zap_other_threads(current);
920                 }
921                 spin_unlock_irq(&sighand->siglock);
922                 read_unlock(&tasklist_lock);
923         }
924
925         do_exit(exit_code);
926         /* NOTREACHED */
927 }
928
929 /*
930  * this kills every thread in the thread group. Note that any externally
931  * wait4()-ing process will get the correct exit code - even if this
932  * thread is not the thread group leader.
933  */
934 asmlinkage void sys_exit_group(int error_code)
935 {
936         do_group_exit((error_code & 0xff) << 8);
937 }
938
939 static int eligible_child(pid_t pid, int options, task_t *p)
940 {
941         if (pid > 0) {
942                 if (p->pid != pid)
943                         return 0;
944         } else if (!pid) {
945                 if (process_group(p) != process_group(current))
946                         return 0;
947         } else if (pid != -1) {
948                 if (process_group(p) != -pid)
949                         return 0;
950         }
951
952         /*
953          * Do not consider detached threads that are
954          * not ptraced:
955          */
956         if (p->exit_signal == -1 && !p->ptrace)
957                 return 0;
958
959         /* Wait for all children (clone and not) if __WALL is set;
960          * otherwise, wait for clone children *only* if __WCLONE is
961          * set; otherwise, wait for non-clone children *only*.  (Note:
962          * A "clone" child here is one that reports to its parent
963          * using a signal other than SIGCHLD.) */
964         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
965             && !(options & __WALL))
966                 return 0;
967         /*
968          * Do not consider thread group leaders that are
969          * in a non-empty thread group:
970          */
971         if (current->tgid != p->tgid && delay_group_leader(p))
972                 return 2;
973
974         if (security_task_wait(p))
975                 return 0;
976
977         return 1;
978 }
979
980 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
981                                int why, int status,
982                                struct siginfo __user *infop,
983                                struct rusage __user *rusagep)
984 {
985         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
986         put_task_struct(p);
987         if (!retval)
988                 retval = put_user(SIGCHLD, &infop->si_signo);
989         if (!retval)
990                 retval = put_user(0, &infop->si_errno);
991         if (!retval)
992                 retval = put_user((short)why, &infop->si_code);
993         if (!retval)
994                 retval = put_user(pid, &infop->si_pid);
995         if (!retval)
996                 retval = put_user(uid, &infop->si_uid);
997         if (!retval)
998                 retval = put_user(status, &infop->si_status);
999         if (!retval)
1000                 retval = pid;
1001         return retval;
1002 }
1003
1004 /*
1005  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1006  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1007  * the lock and this task is uninteresting.  If we return nonzero, we have
1008  * released the lock and the system call should return.
1009  */
1010 static int wait_task_zombie(task_t *p, int noreap,
1011                             struct siginfo __user *infop,
1012                             int __user *stat_addr, struct rusage __user *ru)
1013 {
1014         unsigned long state;
1015         int retval;
1016         int status;
1017
1018         if (unlikely(noreap)) {
1019                 pid_t pid = p->pid;
1020                 uid_t uid = p->uid;
1021                 int exit_code = p->exit_code;
1022                 int why, status;
1023
1024                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1025                         return 0;
1026                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1027                         return 0;
1028                 get_task_struct(p);
1029                 read_unlock(&tasklist_lock);
1030                 if ((exit_code & 0x7f) == 0) {
1031                         why = CLD_EXITED;
1032                         status = exit_code >> 8;
1033                 } else {
1034                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1035                         status = exit_code & 0x7f;
1036                 }
1037                 return wait_noreap_copyout(p, pid, uid, why,
1038                                            status, infop, ru);
1039         }
1040
1041         /*
1042          * Try to move the task's state to DEAD
1043          * only one thread is allowed to do this:
1044          */
1045         state = xchg(&p->exit_state, EXIT_DEAD);
1046         if (state != EXIT_ZOMBIE) {
1047                 BUG_ON(state != EXIT_DEAD);
1048                 return 0;
1049         }
1050         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1051                 /*
1052                  * This can only happen in a race with a ptraced thread
1053                  * dying on another processor.
1054                  */
1055                 return 0;
1056         }
1057
1058         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1059                 /*
1060                  * The resource counters for the group leader are in its
1061                  * own task_struct.  Those for dead threads in the group
1062                  * are in its signal_struct, as are those for the child
1063                  * processes it has previously reaped.  All these
1064                  * accumulate in the parent's signal_struct c* fields.
1065                  *
1066                  * We don't bother to take a lock here to protect these
1067                  * p->signal fields, because they are only touched by
1068                  * __exit_signal, which runs with tasklist_lock
1069                  * write-locked anyway, and so is excluded here.  We do
1070                  * need to protect the access to p->parent->signal fields,
1071                  * as other threads in the parent group can be right
1072                  * here reaping other children at the same time.
1073                  */
1074                 spin_lock_irq(&p->parent->sighand->siglock);
1075                 p->parent->signal->cutime =
1076                         cputime_add(p->parent->signal->cutime,
1077                         cputime_add(p->utime,
1078                         cputime_add(p->signal->utime,
1079                                     p->signal->cutime)));
1080                 p->parent->signal->cstime =
1081                         cputime_add(p->parent->signal->cstime,
1082                         cputime_add(p->stime,
1083                         cputime_add(p->signal->stime,
1084                                     p->signal->cstime)));
1085                 p->parent->signal->cmin_flt +=
1086                         p->min_flt + p->signal->min_flt + p->signal->cmin_flt;
1087                 p->parent->signal->cmaj_flt +=
1088                         p->maj_flt + p->signal->maj_flt + p->signal->cmaj_flt;
1089                 p->parent->signal->cnvcsw +=
1090                         p->nvcsw + p->signal->nvcsw + p->signal->cnvcsw;
1091                 p->parent->signal->cnivcsw +=
1092                         p->nivcsw + p->signal->nivcsw + p->signal->cnivcsw;
1093                 spin_unlock_irq(&p->parent->sighand->siglock);
1094         }
1095
1096         /*
1097          * Now we are sure this task is interesting, and no other
1098          * thread can reap it because we set its state to EXIT_DEAD.
1099          */
1100         read_unlock(&tasklist_lock);
1101
1102         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1103         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1104                 ? p->signal->group_exit_code : p->exit_code;
1105         if (!retval && stat_addr)
1106                 retval = put_user(status, stat_addr);
1107         if (!retval && infop)
1108                 retval = put_user(SIGCHLD, &infop->si_signo);
1109         if (!retval && infop)
1110                 retval = put_user(0, &infop->si_errno);
1111         if (!retval && infop) {
1112                 int why;
1113
1114                 if ((status & 0x7f) == 0) {
1115                         why = CLD_EXITED;
1116                         status >>= 8;
1117                 } else {
1118                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1119                         status &= 0x7f;
1120                 }
1121                 retval = put_user((short)why, &infop->si_code);
1122                 if (!retval)
1123                         retval = put_user(status, &infop->si_status);
1124         }
1125         if (!retval && infop)
1126                 retval = put_user(p->pid, &infop->si_pid);
1127         if (!retval && infop)
1128                 retval = put_user(p->uid, &infop->si_uid);
1129         if (retval) {
1130                 // TODO: is this safe?
1131                 p->exit_state = EXIT_ZOMBIE;
1132                 return retval;
1133         }
1134         retval = p->pid;
1135         if (p->real_parent != p->parent) {
1136                 write_lock_irq(&tasklist_lock);
1137                 /* Double-check with lock held.  */
1138                 if (p->real_parent != p->parent) {
1139                         __ptrace_unlink(p);
1140                         // TODO: is this safe?
1141                         p->exit_state = EXIT_ZOMBIE;
1142                         /*
1143                          * If this is not a detached task, notify the parent.
1144                          * If it's still not detached after that, don't release
1145                          * it now.
1146                          */
1147                         if (p->exit_signal != -1) {
1148                                 do_notify_parent(p, p->exit_signal);
1149                                 if (p->exit_signal != -1)
1150                                         p = NULL;
1151                         }
1152                 }
1153                 write_unlock_irq(&tasklist_lock);
1154         }
1155         if (p != NULL)
1156                 release_task(p);
1157         BUG_ON(!retval);
1158         return retval;
1159 }
1160
1161 /*
1162  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1163  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1164  * the lock and this task is uninteresting.  If we return nonzero, we have
1165  * released the lock and the system call should return.
1166  */
1167 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1168                              struct siginfo __user *infop,
1169                              int __user *stat_addr, struct rusage __user *ru)
1170 {
1171         int retval, exit_code;
1172
1173         if (!p->exit_code)
1174                 return 0;
1175         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1176             p->signal && p->signal->group_stop_count > 0)
1177                 /*
1178                  * A group stop is in progress and this is the group leader.
1179                  * We won't report until all threads have stopped.
1180                  */
1181                 return 0;
1182
1183         /*
1184          * Now we are pretty sure this task is interesting.
1185          * Make sure it doesn't get reaped out from under us while we
1186          * give up the lock and then examine it below.  We don't want to
1187          * keep holding onto the tasklist_lock while we call getrusage and
1188          * possibly take page faults for user memory.
1189          */
1190         get_task_struct(p);
1191         read_unlock(&tasklist_lock);
1192
1193         if (unlikely(noreap)) {
1194                 pid_t pid = p->pid;
1195                 uid_t uid = p->uid;
1196                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1197
1198                 exit_code = p->exit_code;
1199                 if (unlikely(!exit_code) ||
1200                     unlikely(p->state > TASK_STOPPED))
1201                         goto bail_ref;
1202                 return wait_noreap_copyout(p, pid, uid,
1203                                            why, (exit_code << 8) | 0x7f,
1204                                            infop, ru);
1205         }
1206
1207         write_lock_irq(&tasklist_lock);
1208
1209         /*
1210          * This uses xchg to be atomic with the thread resuming and setting
1211          * it.  It must also be done with the write lock held to prevent a
1212          * race with the EXIT_ZOMBIE case.
1213          */
1214         exit_code = xchg(&p->exit_code, 0);
1215         if (unlikely(p->exit_state)) {
1216                 /*
1217                  * The task resumed and then died.  Let the next iteration
1218                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1219                  * already be zero here if it resumed and did _exit(0).
1220                  * The task itself is dead and won't touch exit_code again;
1221                  * other processors in this function are locked out.
1222                  */
1223                 p->exit_code = exit_code;
1224                 exit_code = 0;
1225         }
1226         if (unlikely(exit_code == 0)) {
1227                 /*
1228                  * Another thread in this function got to it first, or it
1229                  * resumed, or it resumed and then died.
1230                  */
1231                 write_unlock_irq(&tasklist_lock);
1232 bail_ref:
1233                 put_task_struct(p);
1234                 /*
1235                  * We are returning to the wait loop without having successfully
1236                  * removed the process and having released the lock. We cannot
1237                  * continue, since the "p" task pointer is potentially stale.
1238                  *
1239                  * Return -EAGAIN, and do_wait() will restart the loop from the
1240                  * beginning. Do _not_ re-acquire the lock.
1241                  */
1242                 return -EAGAIN;
1243         }
1244
1245         /* move to end of parent's list to avoid starvation */
1246         remove_parent(p);
1247         add_parent(p, p->parent);
1248
1249         write_unlock_irq(&tasklist_lock);
1250
1251         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1252         if (!retval && stat_addr)
1253                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1254         if (!retval && infop)
1255                 retval = put_user(SIGCHLD, &infop->si_signo);
1256         if (!retval && infop)
1257                 retval = put_user(0, &infop->si_errno);
1258         if (!retval && infop)
1259                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1260                                           ? CLD_TRAPPED : CLD_STOPPED),
1261                                   &infop->si_code);
1262         if (!retval && infop)
1263                 retval = put_user(exit_code, &infop->si_status);
1264         if (!retval && infop)
1265                 retval = put_user(p->pid, &infop->si_pid);
1266         if (!retval && infop)
1267                 retval = put_user(p->uid, &infop->si_uid);
1268         if (!retval)
1269                 retval = p->pid;
1270         put_task_struct(p);
1271
1272         BUG_ON(!retval);
1273         return retval;
1274 }
1275
1276 /*
1277  * Handle do_wait work for one task in a live, non-stopped state.
1278  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1279  * the lock and this task is uninteresting.  If we return nonzero, we have
1280  * released the lock and the system call should return.
1281  */
1282 static int wait_task_continued(task_t *p, int noreap,
1283                                struct siginfo __user *infop,
1284                                int __user *stat_addr, struct rusage __user *ru)
1285 {
1286         int retval;
1287         pid_t pid;
1288         uid_t uid;
1289
1290         if (unlikely(!p->signal))
1291                 return 0;
1292
1293         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1294                 return 0;
1295
1296         spin_lock_irq(&p->sighand->siglock);
1297         /* Re-check with the lock held.  */
1298         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1299                 spin_unlock_irq(&p->sighand->siglock);
1300                 return 0;
1301         }
1302         if (!noreap)
1303                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1304         spin_unlock_irq(&p->sighand->siglock);
1305
1306         pid = p->pid;
1307         uid = p->uid;
1308         get_task_struct(p);
1309         read_unlock(&tasklist_lock);
1310
1311         if (!infop) {
1312                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1313                 put_task_struct(p);
1314                 if (!retval && stat_addr)
1315                         retval = put_user(0xffff, stat_addr);
1316                 if (!retval)
1317                         retval = p->pid;
1318         } else {
1319                 retval = wait_noreap_copyout(p, pid, uid,
1320                                              CLD_CONTINUED, SIGCONT,
1321                                              infop, ru);
1322                 BUG_ON(retval == 0);
1323         }
1324
1325         return retval;
1326 }
1327
1328
1329 static inline int my_ptrace_child(struct task_struct *p)
1330 {
1331         if (!(p->ptrace & PT_PTRACED))
1332                 return 0;
1333         if (!(p->ptrace & PT_ATTACHED))
1334                 return 1;
1335         /*
1336          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1337          * we are the attacher.  If we are the real parent, this is a race
1338          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1339          * which we have to switch the parent links, but has already set
1340          * the flags in p->ptrace.
1341          */
1342         return (p->parent != p->real_parent);
1343 }
1344
1345 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1346                     int __user *stat_addr, struct rusage __user *ru)
1347 {
1348         DECLARE_WAITQUEUE(wait, current);
1349         struct task_struct *tsk;
1350         int flag, retval;
1351
1352         add_wait_queue(&current->signal->wait_chldexit,&wait);
1353 repeat:
1354         /*
1355          * We will set this flag if we see any child that might later
1356          * match our criteria, even if we are not able to reap it yet.
1357          */
1358         flag = 0;
1359         current->state = TASK_INTERRUPTIBLE;
1360         read_lock(&tasklist_lock);
1361         tsk = current;
1362         do {
1363                 struct task_struct *p;
1364                 struct list_head *_p;
1365                 int ret;
1366
1367                 list_for_each(_p,&tsk->children) {
1368                         p = list_entry(_p,struct task_struct,sibling);
1369
1370                         ret = eligible_child(pid, options, p);
1371                         if (!ret)
1372                                 continue;
1373
1374                         switch (p->state) {
1375                         case TASK_TRACED:
1376                                 if (!my_ptrace_child(p))
1377                                         continue;
1378                                 /*FALLTHROUGH*/
1379                         case TASK_STOPPED:
1380                                 /*
1381                                  * It's stopped now, so it might later
1382                                  * continue, exit, or stop again.
1383                                  */
1384                                 flag = 1;
1385                                 if (!(options & WUNTRACED) &&
1386                                     !my_ptrace_child(p))
1387                                         continue;
1388                                 retval = wait_task_stopped(p, ret == 2,
1389                                                            (options & WNOWAIT),
1390                                                            infop,
1391                                                            stat_addr, ru);
1392                                 if (retval == -EAGAIN)
1393                                         goto repeat;
1394                                 if (retval != 0) /* He released the lock.  */
1395                                         goto end;
1396                                 break;
1397                         default:
1398                         // case EXIT_DEAD:
1399                                 if (p->exit_state == EXIT_DEAD)
1400                                         continue;
1401                         // case EXIT_ZOMBIE:
1402                                 if (p->exit_state == EXIT_ZOMBIE) {
1403                                         /*
1404                                          * Eligible but we cannot release
1405                                          * it yet:
1406                                          */
1407                                         if (ret == 2)
1408                                                 goto check_continued;
1409                                         if (!likely(options & WEXITED))
1410                                                 continue;
1411                                         retval = wait_task_zombie(
1412                                                 p, (options & WNOWAIT),
1413                                                 infop, stat_addr, ru);
1414                                         /* He released the lock.  */
1415                                         if (retval != 0)
1416                                                 goto end;
1417                                         break;
1418                                 }
1419 check_continued:
1420                                 /*
1421                                  * It's running now, so it might later
1422                                  * exit, stop, or stop and then continue.
1423                                  */
1424                                 flag = 1;
1425                                 if (!unlikely(options & WCONTINUED))
1426                                         continue;
1427                                 retval = wait_task_continued(
1428                                         p, (options & WNOWAIT),
1429                                         infop, stat_addr, ru);
1430                                 if (retval != 0) /* He released the lock.  */
1431                                         goto end;
1432                                 break;
1433                         }
1434                 }
1435                 if (!flag) {
1436                         list_for_each(_p, &tsk->ptrace_children) {
1437                                 p = list_entry(_p, struct task_struct,
1438                                                 ptrace_list);
1439                                 if (!eligible_child(pid, options, p))
1440                                         continue;
1441                                 flag = 1;
1442                                 break;
1443                         }
1444                 }
1445                 if (options & __WNOTHREAD)
1446                         break;
1447                 tsk = next_thread(tsk);
1448                 if (tsk->signal != current->signal)
1449                         BUG();
1450         } while (tsk != current);
1451
1452         read_unlock(&tasklist_lock);
1453         if (flag) {
1454                 retval = 0;
1455                 if (options & WNOHANG)
1456                         goto end;
1457                 retval = -ERESTARTSYS;
1458                 if (signal_pending(current))
1459                         goto end;
1460                 schedule();
1461                 goto repeat;
1462         }
1463         retval = -ECHILD;
1464 end:
1465         current->state = TASK_RUNNING;
1466         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1467         if (infop) {
1468                 if (retval > 0)
1469                 retval = 0;
1470                 else {
1471                         /*
1472                          * For a WNOHANG return, clear out all the fields
1473                          * we would set so the user can easily tell the
1474                          * difference.
1475                          */
1476                         if (!retval)
1477                                 retval = put_user(0, &infop->si_signo);
1478                         if (!retval)
1479                                 retval = put_user(0, &infop->si_errno);
1480                         if (!retval)
1481                                 retval = put_user(0, &infop->si_code);
1482                         if (!retval)
1483                                 retval = put_user(0, &infop->si_pid);
1484                         if (!retval)
1485                                 retval = put_user(0, &infop->si_uid);
1486                         if (!retval)
1487                                 retval = put_user(0, &infop->si_status);
1488                 }
1489         }
1490         return retval;
1491 }
1492
1493 asmlinkage long sys_waitid(int which, pid_t pid,
1494                            struct siginfo __user *infop, int options,
1495                            struct rusage __user *ru)
1496 {
1497         long ret;
1498
1499         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1500                 return -EINVAL;
1501         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1502                 return -EINVAL;
1503
1504         switch (which) {
1505         case P_ALL:
1506                 pid = -1;
1507                 break;
1508         case P_PID:
1509                 if (pid <= 0)
1510                         return -EINVAL;
1511                 break;
1512         case P_PGID:
1513                 if (pid <= 0)
1514                         return -EINVAL;
1515                 pid = -pid;
1516                 break;
1517         default:
1518                 return -EINVAL;
1519         }
1520
1521         ret = do_wait(pid, options, infop, NULL, ru);
1522
1523         /* avoid REGPARM breakage on x86: */
1524         prevent_tail_call(ret);
1525         return ret;
1526 }
1527
1528 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1529                           int options, struct rusage __user *ru)
1530 {
1531         long ret;
1532
1533         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1534                         __WNOTHREAD|__WCLONE|__WALL))
1535                 return -EINVAL;
1536         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1537
1538         /* avoid REGPARM breakage on x86: */
1539         prevent_tail_call(ret);
1540         return ret;
1541 }
1542
1543 #ifdef __ARCH_WANT_SYS_WAITPID
1544
1545 /*
1546  * sys_waitpid() remains for compatibility. waitpid() should be
1547  * implemented by calling sys_wait4() from libc.a.
1548  */
1549 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1550 {
1551         return sys_wait4(pid, stat_addr, options, NULL);
1552 }
1553
1554 #endif