[PATCH] proc: Rewrite the proc dentry flush on exit optimization
[linux-2.6] / kernel / exit.c
1 /*
2  *  linux/kernel/exit.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 #include <linux/config.h>
8 #include <linux/mm.h>
9 #include <linux/slab.h>
10 #include <linux/interrupt.h>
11 #include <linux/smp_lock.h>
12 #include <linux/module.h>
13 #include <linux/capability.h>
14 #include <linux/completion.h>
15 #include <linux/personality.h>
16 #include <linux/tty.h>
17 #include <linux/namespace.h>
18 #include <linux/key.h>
19 #include <linux/security.h>
20 #include <linux/cpu.h>
21 #include <linux/acct.h>
22 #include <linux/file.h>
23 #include <linux/binfmts.h>
24 #include <linux/ptrace.h>
25 #include <linux/profile.h>
26 #include <linux/mount.h>
27 #include <linux/proc_fs.h>
28 #include <linux/mempolicy.h>
29 #include <linux/cpuset.h>
30 #include <linux/syscalls.h>
31 #include <linux/signal.h>
32 #include <linux/posix-timers.h>
33 #include <linux/cn_proc.h>
34 #include <linux/mutex.h>
35 #include <linux/futex.h>
36 #include <linux/compat.h>
37 #include <linux/pipe_fs_i.h>
38 #include <linux/audit.h> /* for audit_free() */
39 #include <linux/resource.h>
40
41 #include <asm/uaccess.h>
42 #include <asm/unistd.h>
43 #include <asm/pgtable.h>
44 #include <asm/mmu_context.h>
45
46 extern void sem_exit (void);
47 extern struct task_struct *child_reaper;
48
49 static void exit_mm(struct task_struct * tsk);
50
51 static void __unhash_process(struct task_struct *p)
52 {
53         nr_threads--;
54         detach_pid(p, PIDTYPE_PID);
55         if (thread_group_leader(p)) {
56                 detach_pid(p, PIDTYPE_PGID);
57                 detach_pid(p, PIDTYPE_SID);
58
59                 list_del_rcu(&p->tasks);
60                 __get_cpu_var(process_counts)--;
61         }
62         list_del_rcu(&p->thread_group);
63         remove_parent(p);
64 }
65
66 /*
67  * This function expects the tasklist_lock write-locked.
68  */
69 static void __exit_signal(struct task_struct *tsk)
70 {
71         struct signal_struct *sig = tsk->signal;
72         struct sighand_struct *sighand;
73
74         BUG_ON(!sig);
75         BUG_ON(!atomic_read(&sig->count));
76
77         rcu_read_lock();
78         sighand = rcu_dereference(tsk->sighand);
79         spin_lock(&sighand->siglock);
80
81         posix_cpu_timers_exit(tsk);
82         if (atomic_dec_and_test(&sig->count))
83                 posix_cpu_timers_exit_group(tsk);
84         else {
85                 /*
86                  * If there is any task waiting for the group exit
87                  * then notify it:
88                  */
89                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
90                         wake_up_process(sig->group_exit_task);
91                         sig->group_exit_task = NULL;
92                 }
93                 if (tsk == sig->curr_target)
94                         sig->curr_target = next_thread(tsk);
95                 /*
96                  * Accumulate here the counters for all threads but the
97                  * group leader as they die, so they can be added into
98                  * the process-wide totals when those are taken.
99                  * The group leader stays around as a zombie as long
100                  * as there are other threads.  When it gets reaped,
101                  * the exit.c code will add its counts into these totals.
102                  * We won't ever get here for the group leader, since it
103                  * will have been the last reference on the signal_struct.
104                  */
105                 sig->utime = cputime_add(sig->utime, tsk->utime);
106                 sig->stime = cputime_add(sig->stime, tsk->stime);
107                 sig->min_flt += tsk->min_flt;
108                 sig->maj_flt += tsk->maj_flt;
109                 sig->nvcsw += tsk->nvcsw;
110                 sig->nivcsw += tsk->nivcsw;
111                 sig->sched_time += tsk->sched_time;
112                 sig = NULL; /* Marker for below. */
113         }
114
115         __unhash_process(tsk);
116
117         tsk->signal = NULL;
118         tsk->sighand = NULL;
119         spin_unlock(&sighand->siglock);
120         rcu_read_unlock();
121
122         __cleanup_sighand(sighand);
123         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
124         flush_sigqueue(&tsk->pending);
125         if (sig) {
126                 flush_sigqueue(&sig->shared_pending);
127                 __cleanup_signal(sig);
128         }
129 }
130
131 static void delayed_put_task_struct(struct rcu_head *rhp)
132 {
133         put_task_struct(container_of(rhp, struct task_struct, rcu));
134 }
135
136 void release_task(struct task_struct * p)
137 {
138         int zap_leader;
139         task_t *leader;
140 repeat:
141         atomic_dec(&p->user->processes);
142         write_lock_irq(&tasklist_lock);
143         ptrace_unlink(p);
144         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
145         __exit_signal(p);
146
147         /*
148          * If we are the last non-leader member of the thread
149          * group, and the leader is zombie, then notify the
150          * group leader's parent process. (if it wants notification.)
151          */
152         zap_leader = 0;
153         leader = p->group_leader;
154         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
155                 BUG_ON(leader->exit_signal == -1);
156                 do_notify_parent(leader, leader->exit_signal);
157                 /*
158                  * If we were the last child thread and the leader has
159                  * exited already, and the leader's parent ignores SIGCHLD,
160                  * then we are the one who should release the leader.
161                  *
162                  * do_notify_parent() will have marked it self-reaping in
163                  * that case.
164                  */
165                 zap_leader = (leader->exit_signal == -1);
166         }
167
168         sched_exit(p);
169         write_unlock_irq(&tasklist_lock);
170         proc_flush_task(p);
171         release_thread(p);
172         call_rcu(&p->rcu, delayed_put_task_struct);
173
174         p = leader;
175         if (unlikely(zap_leader))
176                 goto repeat;
177 }
178
179 /*
180  * This checks not only the pgrp, but falls back on the pid if no
181  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
182  * without this...
183  */
184 int session_of_pgrp(int pgrp)
185 {
186         struct task_struct *p;
187         int sid = -1;
188
189         read_lock(&tasklist_lock);
190         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
191                 if (p->signal->session > 0) {
192                         sid = p->signal->session;
193                         goto out;
194                 }
195         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
196         p = find_task_by_pid(pgrp);
197         if (p)
198                 sid = p->signal->session;
199 out:
200         read_unlock(&tasklist_lock);
201         
202         return sid;
203 }
204
205 /*
206  * Determine if a process group is "orphaned", according to the POSIX
207  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
208  * by terminal-generated stop signals.  Newly orphaned process groups are
209  * to receive a SIGHUP and a SIGCONT.
210  *
211  * "I ask you, have you ever known what it is to be an orphan?"
212  */
213 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
214 {
215         struct task_struct *p;
216         int ret = 1;
217
218         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
219                 if (p == ignored_task
220                                 || p->exit_state
221                                 || p->real_parent->pid == 1)
222                         continue;
223                 if (process_group(p->real_parent) != pgrp
224                             && p->real_parent->signal->session == p->signal->session) {
225                         ret = 0;
226                         break;
227                 }
228         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
229         return ret;     /* (sighing) "Often!" */
230 }
231
232 int is_orphaned_pgrp(int pgrp)
233 {
234         int retval;
235
236         read_lock(&tasklist_lock);
237         retval = will_become_orphaned_pgrp(pgrp, NULL);
238         read_unlock(&tasklist_lock);
239
240         return retval;
241 }
242
243 static int has_stopped_jobs(int pgrp)
244 {
245         int retval = 0;
246         struct task_struct *p;
247
248         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
249                 if (p->state != TASK_STOPPED)
250                         continue;
251
252                 /* If p is stopped by a debugger on a signal that won't
253                    stop it, then don't count p as stopped.  This isn't
254                    perfect but it's a good approximation.  */
255                 if (unlikely (p->ptrace)
256                     && p->exit_code != SIGSTOP
257                     && p->exit_code != SIGTSTP
258                     && p->exit_code != SIGTTOU
259                     && p->exit_code != SIGTTIN)
260                         continue;
261
262                 retval = 1;
263                 break;
264         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
265         return retval;
266 }
267
268 /**
269  * reparent_to_init - Reparent the calling kernel thread to the init task.
270  *
271  * If a kernel thread is launched as a result of a system call, or if
272  * it ever exits, it should generally reparent itself to init so that
273  * it is correctly cleaned up on exit.
274  *
275  * The various task state such as scheduling policy and priority may have
276  * been inherited from a user process, so we reset them to sane values here.
277  *
278  * NOTE that reparent_to_init() gives the caller full capabilities.
279  */
280 static void reparent_to_init(void)
281 {
282         write_lock_irq(&tasklist_lock);
283
284         ptrace_unlink(current);
285         /* Reparent to init */
286         remove_parent(current);
287         current->parent = child_reaper;
288         current->real_parent = child_reaper;
289         add_parent(current);
290
291         /* Set the exit signal to SIGCHLD so we signal init on exit */
292         current->exit_signal = SIGCHLD;
293
294         if ((current->policy == SCHED_NORMAL ||
295                         current->policy == SCHED_BATCH)
296                                 && (task_nice(current) < 0))
297                 set_user_nice(current, 0);
298         /* cpus_allowed? */
299         /* rt_priority? */
300         /* signals? */
301         security_task_reparent_to_init(current);
302         memcpy(current->signal->rlim, init_task.signal->rlim,
303                sizeof(current->signal->rlim));
304         atomic_inc(&(INIT_USER->__count));
305         write_unlock_irq(&tasklist_lock);
306         switch_uid(INIT_USER);
307 }
308
309 void __set_special_pids(pid_t session, pid_t pgrp)
310 {
311         struct task_struct *curr = current->group_leader;
312
313         if (curr->signal->session != session) {
314                 detach_pid(curr, PIDTYPE_SID);
315                 curr->signal->session = session;
316                 attach_pid(curr, PIDTYPE_SID, session);
317         }
318         if (process_group(curr) != pgrp) {
319                 detach_pid(curr, PIDTYPE_PGID);
320                 curr->signal->pgrp = pgrp;
321                 attach_pid(curr, PIDTYPE_PGID, pgrp);
322         }
323 }
324
325 void set_special_pids(pid_t session, pid_t pgrp)
326 {
327         write_lock_irq(&tasklist_lock);
328         __set_special_pids(session, pgrp);
329         write_unlock_irq(&tasklist_lock);
330 }
331
332 /*
333  * Let kernel threads use this to say that they
334  * allow a certain signal (since daemonize() will
335  * have disabled all of them by default).
336  */
337 int allow_signal(int sig)
338 {
339         if (!valid_signal(sig) || sig < 1)
340                 return -EINVAL;
341
342         spin_lock_irq(&current->sighand->siglock);
343         sigdelset(&current->blocked, sig);
344         if (!current->mm) {
345                 /* Kernel threads handle their own signals.
346                    Let the signal code know it'll be handled, so
347                    that they don't get converted to SIGKILL or
348                    just silently dropped */
349                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
350         }
351         recalc_sigpending();
352         spin_unlock_irq(&current->sighand->siglock);
353         return 0;
354 }
355
356 EXPORT_SYMBOL(allow_signal);
357
358 int disallow_signal(int sig)
359 {
360         if (!valid_signal(sig) || sig < 1)
361                 return -EINVAL;
362
363         spin_lock_irq(&current->sighand->siglock);
364         sigaddset(&current->blocked, sig);
365         recalc_sigpending();
366         spin_unlock_irq(&current->sighand->siglock);
367         return 0;
368 }
369
370 EXPORT_SYMBOL(disallow_signal);
371
372 /*
373  *      Put all the gunge required to become a kernel thread without
374  *      attached user resources in one place where it belongs.
375  */
376
377 void daemonize(const char *name, ...)
378 {
379         va_list args;
380         struct fs_struct *fs;
381         sigset_t blocked;
382
383         va_start(args, name);
384         vsnprintf(current->comm, sizeof(current->comm), name, args);
385         va_end(args);
386
387         /*
388          * If we were started as result of loading a module, close all of the
389          * user space pages.  We don't need them, and if we didn't close them
390          * they would be locked into memory.
391          */
392         exit_mm(current);
393
394         set_special_pids(1, 1);
395         mutex_lock(&tty_mutex);
396         current->signal->tty = NULL;
397         mutex_unlock(&tty_mutex);
398
399         /* Block and flush all signals */
400         sigfillset(&blocked);
401         sigprocmask(SIG_BLOCK, &blocked, NULL);
402         flush_signals(current);
403
404         /* Become as one with the init task */
405
406         exit_fs(current);       /* current->fs->count--; */
407         fs = init_task.fs;
408         current->fs = fs;
409         atomic_inc(&fs->count);
410         exit_namespace(current);
411         current->namespace = init_task.namespace;
412         get_namespace(current->namespace);
413         exit_files(current);
414         current->files = init_task.files;
415         atomic_inc(&current->files->count);
416
417         reparent_to_init();
418 }
419
420 EXPORT_SYMBOL(daemonize);
421
422 static void close_files(struct files_struct * files)
423 {
424         int i, j;
425         struct fdtable *fdt;
426
427         j = 0;
428
429         /*
430          * It is safe to dereference the fd table without RCU or
431          * ->file_lock because this is the last reference to the
432          * files structure.
433          */
434         fdt = files_fdtable(files);
435         for (;;) {
436                 unsigned long set;
437                 i = j * __NFDBITS;
438                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
439                         break;
440                 set = fdt->open_fds->fds_bits[j++];
441                 while (set) {
442                         if (set & 1) {
443                                 struct file * file = xchg(&fdt->fd[i], NULL);
444                                 if (file)
445                                         filp_close(file, files);
446                         }
447                         i++;
448                         set >>= 1;
449                 }
450         }
451 }
452
453 struct files_struct *get_files_struct(struct task_struct *task)
454 {
455         struct files_struct *files;
456
457         task_lock(task);
458         files = task->files;
459         if (files)
460                 atomic_inc(&files->count);
461         task_unlock(task);
462
463         return files;
464 }
465
466 void fastcall put_files_struct(struct files_struct *files)
467 {
468         struct fdtable *fdt;
469
470         if (atomic_dec_and_test(&files->count)) {
471                 close_files(files);
472                 /*
473                  * Free the fd and fdset arrays if we expanded them.
474                  * If the fdtable was embedded, pass files for freeing
475                  * at the end of the RCU grace period. Otherwise,
476                  * you can free files immediately.
477                  */
478                 fdt = files_fdtable(files);
479                 if (fdt == &files->fdtab)
480                         fdt->free_files = files;
481                 else
482                         kmem_cache_free(files_cachep, files);
483                 free_fdtable(fdt);
484         }
485 }
486
487 EXPORT_SYMBOL(put_files_struct);
488
489 static inline void __exit_files(struct task_struct *tsk)
490 {
491         struct files_struct * files = tsk->files;
492
493         if (files) {
494                 task_lock(tsk);
495                 tsk->files = NULL;
496                 task_unlock(tsk);
497                 put_files_struct(files);
498         }
499 }
500
501 void exit_files(struct task_struct *tsk)
502 {
503         __exit_files(tsk);
504 }
505
506 static inline void __put_fs_struct(struct fs_struct *fs)
507 {
508         /* No need to hold fs->lock if we are killing it */
509         if (atomic_dec_and_test(&fs->count)) {
510                 dput(fs->root);
511                 mntput(fs->rootmnt);
512                 dput(fs->pwd);
513                 mntput(fs->pwdmnt);
514                 if (fs->altroot) {
515                         dput(fs->altroot);
516                         mntput(fs->altrootmnt);
517                 }
518                 kmem_cache_free(fs_cachep, fs);
519         }
520 }
521
522 void put_fs_struct(struct fs_struct *fs)
523 {
524         __put_fs_struct(fs);
525 }
526
527 static inline void __exit_fs(struct task_struct *tsk)
528 {
529         struct fs_struct * fs = tsk->fs;
530
531         if (fs) {
532                 task_lock(tsk);
533                 tsk->fs = NULL;
534                 task_unlock(tsk);
535                 __put_fs_struct(fs);
536         }
537 }
538
539 void exit_fs(struct task_struct *tsk)
540 {
541         __exit_fs(tsk);
542 }
543
544 EXPORT_SYMBOL_GPL(exit_fs);
545
546 /*
547  * Turn us into a lazy TLB process if we
548  * aren't already..
549  */
550 static void exit_mm(struct task_struct * tsk)
551 {
552         struct mm_struct *mm = tsk->mm;
553
554         mm_release(tsk, mm);
555         if (!mm)
556                 return;
557         /*
558          * Serialize with any possible pending coredump.
559          * We must hold mmap_sem around checking core_waiters
560          * and clearing tsk->mm.  The core-inducing thread
561          * will increment core_waiters for each thread in the
562          * group with ->mm != NULL.
563          */
564         down_read(&mm->mmap_sem);
565         if (mm->core_waiters) {
566                 up_read(&mm->mmap_sem);
567                 down_write(&mm->mmap_sem);
568                 if (!--mm->core_waiters)
569                         complete(mm->core_startup_done);
570                 up_write(&mm->mmap_sem);
571
572                 wait_for_completion(&mm->core_done);
573                 down_read(&mm->mmap_sem);
574         }
575         atomic_inc(&mm->mm_count);
576         BUG_ON(mm != tsk->active_mm);
577         /* more a memory barrier than a real lock */
578         task_lock(tsk);
579         tsk->mm = NULL;
580         up_read(&mm->mmap_sem);
581         enter_lazy_tlb(mm, current);
582         task_unlock(tsk);
583         mmput(mm);
584 }
585
586 static inline void choose_new_parent(task_t *p, task_t *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 reparent_thread(task_t *p, task_t *father, int traced)
597 {
598         /* We don't want people slaying init.  */
599         if (p->exit_signal != -1)
600                 p->exit_signal = SIGCHLD;
601
602         if (p->pdeath_signal)
603                 /* We already hold the tasklist_lock here.  */
604                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
605
606         /* Move the child from its dying parent to the new one.  */
607         if (unlikely(traced)) {
608                 /* Preserve ptrace links if someone else is tracing this child.  */
609                 list_del_init(&p->ptrace_list);
610                 if (p->parent != p->real_parent)
611                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
612         } else {
613                 /* If this child is being traced, then we're the one tracing it
614                  * anyway, so let go of it.
615                  */
616                 p->ptrace = 0;
617                 remove_parent(p);
618                 p->parent = p->real_parent;
619                 add_parent(p);
620
621                 /* If we'd notified the old parent about this child's death,
622                  * also notify the new parent.
623                  */
624                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
625                     thread_group_empty(p))
626                         do_notify_parent(p, p->exit_signal);
627                 else if (p->state == TASK_TRACED) {
628                         /*
629                          * If it was at a trace stop, turn it into
630                          * a normal stop since it's no longer being
631                          * traced.
632                          */
633                         ptrace_untrace(p);
634                 }
635         }
636
637         /*
638          * process group orphan check
639          * Case ii: Our child is in a different pgrp
640          * than we are, and it was the only connection
641          * outside, so the child pgrp is now orphaned.
642          */
643         if ((process_group(p) != process_group(father)) &&
644             (p->signal->session == father->signal->session)) {
645                 int pgrp = process_group(p);
646
647                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
648                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
649                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
650                 }
651         }
652 }
653
654 /*
655  * When we die, we re-parent all our children.
656  * Try to give them to another thread in our thread
657  * group, and if no such member exists, give it to
658  * the global child reaper process (ie "init")
659  */
660 static void forget_original_parent(struct task_struct * father,
661                                           struct list_head *to_release)
662 {
663         struct task_struct *p, *reaper = father;
664         struct list_head *_p, *_n;
665
666         do {
667                 reaper = next_thread(reaper);
668                 if (reaper == father) {
669                         reaper = child_reaper;
670                         break;
671                 }
672         } while (reaper->exit_state);
673
674         /*
675          * There are only two places where our children can be:
676          *
677          * - in our child list
678          * - in our ptraced child list
679          *
680          * Search them and reparent children.
681          */
682         list_for_each_safe(_p, _n, &father->children) {
683                 int ptrace;
684                 p = list_entry(_p,struct task_struct,sibling);
685
686                 ptrace = p->ptrace;
687
688                 /* if father isn't the real parent, then ptrace must be enabled */
689                 BUG_ON(father != p->real_parent && !ptrace);
690
691                 if (father == p->real_parent) {
692                         /* reparent with a reaper, real father it's us */
693                         choose_new_parent(p, reaper);
694                         reparent_thread(p, father, 0);
695                 } else {
696                         /* reparent ptraced task to its real parent */
697                         __ptrace_unlink (p);
698                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
699                             thread_group_empty(p))
700                                 do_notify_parent(p, p->exit_signal);
701                 }
702
703                 /*
704                  * if the ptraced child is a zombie with exit_signal == -1
705                  * we must collect it before we exit, or it will remain
706                  * zombie forever since we prevented it from self-reap itself
707                  * while it was being traced by us, to be able to see it in wait4.
708                  */
709                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
710                         list_add(&p->ptrace_list, to_release);
711         }
712         list_for_each_safe(_p, _n, &father->ptrace_children) {
713                 p = list_entry(_p,struct task_struct,ptrace_list);
714                 choose_new_parent(p, reaper);
715                 reparent_thread(p, father, 1);
716         }
717 }
718
719 /*
720  * Send signals to all our closest relatives so that they know
721  * to properly mourn us..
722  */
723 static void exit_notify(struct task_struct *tsk)
724 {
725         int state;
726         struct task_struct *t;
727         struct list_head ptrace_dead, *_p, *_n;
728
729         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
730             && !thread_group_empty(tsk)) {
731                 /*
732                  * This occurs when there was a race between our exit
733                  * syscall and a group signal choosing us as the one to
734                  * wake up.  It could be that we are the only thread
735                  * alerted to check for pending signals, but another thread
736                  * should be woken now to take the signal since we will not.
737                  * Now we'll wake all the threads in the group just to make
738                  * sure someone gets all the pending signals.
739                  */
740                 read_lock(&tasklist_lock);
741                 spin_lock_irq(&tsk->sighand->siglock);
742                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
743                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
744                                 recalc_sigpending_tsk(t);
745                                 if (signal_pending(t))
746                                         signal_wake_up(t, 0);
747                         }
748                 spin_unlock_irq(&tsk->sighand->siglock);
749                 read_unlock(&tasklist_lock);
750         }
751
752         write_lock_irq(&tasklist_lock);
753
754         /*
755          * This does two things:
756          *
757          * A.  Make init inherit all the child processes
758          * B.  Check to see if any process groups have become orphaned
759          *      as a result of our exiting, and if they have any stopped
760          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
761          */
762
763         INIT_LIST_HEAD(&ptrace_dead);
764         forget_original_parent(tsk, &ptrace_dead);
765         BUG_ON(!list_empty(&tsk->children));
766         BUG_ON(!list_empty(&tsk->ptrace_children));
767
768         /*
769          * Check to see if any process groups have become orphaned
770          * as a result of our exiting, and if they have any stopped
771          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
772          *
773          * Case i: Our father is in a different pgrp than we are
774          * and we were the only connection outside, so our pgrp
775          * is about to become orphaned.
776          */
777          
778         t = tsk->real_parent;
779         
780         if ((process_group(t) != process_group(tsk)) &&
781             (t->signal->session == tsk->signal->session) &&
782             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
783             has_stopped_jobs(process_group(tsk))) {
784                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
785                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
786         }
787
788         /* Let father know we died 
789          *
790          * Thread signals are configurable, but you aren't going to use
791          * that to send signals to arbitary processes. 
792          * That stops right now.
793          *
794          * If the parent exec id doesn't match the exec id we saved
795          * when we started then we know the parent has changed security
796          * domain.
797          *
798          * If our self_exec id doesn't match our parent_exec_id then
799          * we have changed execution domain as these two values started
800          * the same after a fork.
801          *      
802          */
803         
804         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
805             ( tsk->parent_exec_id != t->self_exec_id  ||
806               tsk->self_exec_id != tsk->parent_exec_id)
807             && !capable(CAP_KILL))
808                 tsk->exit_signal = SIGCHLD;
809
810
811         /* If something other than our normal parent is ptracing us, then
812          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
813          * only has special meaning to our real parent.
814          */
815         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
816                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
817                 do_notify_parent(tsk, signal);
818         } else if (tsk->ptrace) {
819                 do_notify_parent(tsk, SIGCHLD);
820         }
821
822         state = EXIT_ZOMBIE;
823         if (tsk->exit_signal == -1 &&
824             (likely(tsk->ptrace == 0) ||
825              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
826                 state = EXIT_DEAD;
827         tsk->exit_state = state;
828
829         write_unlock_irq(&tasklist_lock);
830
831         list_for_each_safe(_p, _n, &ptrace_dead) {
832                 list_del_init(_p);
833                 t = list_entry(_p,struct task_struct,ptrace_list);
834                 release_task(t);
835         }
836
837         /* If the process is dead, release it - nobody will wait for it */
838         if (state == EXIT_DEAD)
839                 release_task(tsk);
840 }
841
842 fastcall NORET_TYPE void do_exit(long code)
843 {
844         struct task_struct *tsk = current;
845         int group_dead;
846
847         profile_task_exit(tsk);
848
849         WARN_ON(atomic_read(&tsk->fs_excl));
850
851         if (unlikely(in_interrupt()))
852                 panic("Aiee, killing interrupt handler!");
853         if (unlikely(!tsk->pid))
854                 panic("Attempted to kill the idle task!");
855         if (unlikely(tsk == child_reaper))
856                 panic("Attempted to kill init!");
857
858         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
859                 current->ptrace_message = code;
860                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
861         }
862
863         /*
864          * We're taking recursive faults here in do_exit. Safest is to just
865          * leave this task alone and wait for reboot.
866          */
867         if (unlikely(tsk->flags & PF_EXITING)) {
868                 printk(KERN_ALERT
869                         "Fixing recursive fault but reboot is needed!\n");
870                 if (tsk->io_context)
871                         exit_io_context();
872                 set_current_state(TASK_UNINTERRUPTIBLE);
873                 schedule();
874         }
875
876         tsk->flags |= PF_EXITING;
877
878         if (unlikely(in_atomic()))
879                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
880                                 current->comm, current->pid,
881                                 preempt_count());
882
883         acct_update_integrals(tsk);
884         if (tsk->mm) {
885                 update_hiwater_rss(tsk->mm);
886                 update_hiwater_vm(tsk->mm);
887         }
888         group_dead = atomic_dec_and_test(&tsk->signal->live);
889         if (group_dead) {
890                 hrtimer_cancel(&tsk->signal->real_timer);
891                 exit_itimers(tsk->signal);
892         }
893         acct_collect(code, group_dead);
894         if (unlikely(tsk->robust_list))
895                 exit_robust_list(tsk);
896 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
897         if (unlikely(tsk->compat_robust_list))
898                 compat_exit_robust_list(tsk);
899 #endif
900         if (unlikely(tsk->audit_context))
901                 audit_free(tsk);
902         exit_mm(tsk);
903
904         if (group_dead)
905                 acct_process();
906         exit_sem(tsk);
907         __exit_files(tsk);
908         __exit_fs(tsk);
909         exit_namespace(tsk);
910         exit_thread();
911         cpuset_exit(tsk);
912         exit_keys(tsk);
913
914         if (group_dead && tsk->signal->leader)
915                 disassociate_ctty(1);
916
917         module_put(task_thread_info(tsk)->exec_domain->module);
918         if (tsk->binfmt)
919                 module_put(tsk->binfmt->module);
920
921         tsk->exit_code = code;
922         proc_exit_connector(tsk);
923         exit_notify(tsk);
924 #ifdef CONFIG_NUMA
925         mpol_free(tsk->mempolicy);
926         tsk->mempolicy = NULL;
927 #endif
928         /*
929          * If DEBUG_MUTEXES is on, make sure we are holding no locks:
930          */
931         mutex_debug_check_no_locks_held(tsk);
932
933         if (tsk->io_context)
934                 exit_io_context();
935
936         if (tsk->splice_pipe)
937                 __free_pipe_info(tsk->splice_pipe);
938
939         /* PF_DEAD causes final put_task_struct after we schedule. */
940         preempt_disable();
941         BUG_ON(tsk->flags & PF_DEAD);
942         tsk->flags |= PF_DEAD;
943
944         schedule();
945         BUG();
946         /* Avoid "noreturn function does return".  */
947         for (;;) ;
948 }
949
950 EXPORT_SYMBOL_GPL(do_exit);
951
952 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
953 {
954         if (comp)
955                 complete(comp);
956         
957         do_exit(code);
958 }
959
960 EXPORT_SYMBOL(complete_and_exit);
961
962 asmlinkage long sys_exit(int error_code)
963 {
964         do_exit((error_code&0xff)<<8);
965 }
966
967 /*
968  * Take down every thread in the group.  This is called by fatal signals
969  * as well as by sys_exit_group (below).
970  */
971 NORET_TYPE void
972 do_group_exit(int exit_code)
973 {
974         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
975
976         if (current->signal->flags & SIGNAL_GROUP_EXIT)
977                 exit_code = current->signal->group_exit_code;
978         else if (!thread_group_empty(current)) {
979                 struct signal_struct *const sig = current->signal;
980                 struct sighand_struct *const sighand = current->sighand;
981                 spin_lock_irq(&sighand->siglock);
982                 if (sig->flags & SIGNAL_GROUP_EXIT)
983                         /* Another thread got here before we took the lock.  */
984                         exit_code = sig->group_exit_code;
985                 else {
986                         sig->group_exit_code = exit_code;
987                         zap_other_threads(current);
988                 }
989                 spin_unlock_irq(&sighand->siglock);
990         }
991
992         do_exit(exit_code);
993         /* NOTREACHED */
994 }
995
996 /*
997  * this kills every thread in the thread group. Note that any externally
998  * wait4()-ing process will get the correct exit code - even if this
999  * thread is not the thread group leader.
1000  */
1001 asmlinkage void sys_exit_group(int error_code)
1002 {
1003         do_group_exit((error_code & 0xff) << 8);
1004 }
1005
1006 static int eligible_child(pid_t pid, int options, task_t *p)
1007 {
1008         if (pid > 0) {
1009                 if (p->pid != pid)
1010                         return 0;
1011         } else if (!pid) {
1012                 if (process_group(p) != process_group(current))
1013                         return 0;
1014         } else if (pid != -1) {
1015                 if (process_group(p) != -pid)
1016                         return 0;
1017         }
1018
1019         /*
1020          * Do not consider detached threads that are
1021          * not ptraced:
1022          */
1023         if (p->exit_signal == -1 && !p->ptrace)
1024                 return 0;
1025
1026         /* Wait for all children (clone and not) if __WALL is set;
1027          * otherwise, wait for clone children *only* if __WCLONE is
1028          * set; otherwise, wait for non-clone children *only*.  (Note:
1029          * A "clone" child here is one that reports to its parent
1030          * using a signal other than SIGCHLD.) */
1031         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1032             && !(options & __WALL))
1033                 return 0;
1034         /*
1035          * Do not consider thread group leaders that are
1036          * in a non-empty thread group:
1037          */
1038         if (current->tgid != p->tgid && delay_group_leader(p))
1039                 return 2;
1040
1041         if (security_task_wait(p))
1042                 return 0;
1043
1044         return 1;
1045 }
1046
1047 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
1048                                int why, int status,
1049                                struct siginfo __user *infop,
1050                                struct rusage __user *rusagep)
1051 {
1052         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1053         put_task_struct(p);
1054         if (!retval)
1055                 retval = put_user(SIGCHLD, &infop->si_signo);
1056         if (!retval)
1057                 retval = put_user(0, &infop->si_errno);
1058         if (!retval)
1059                 retval = put_user((short)why, &infop->si_code);
1060         if (!retval)
1061                 retval = put_user(pid, &infop->si_pid);
1062         if (!retval)
1063                 retval = put_user(uid, &infop->si_uid);
1064         if (!retval)
1065                 retval = put_user(status, &infop->si_status);
1066         if (!retval)
1067                 retval = pid;
1068         return retval;
1069 }
1070
1071 /*
1072  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1073  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1074  * the lock and this task is uninteresting.  If we return nonzero, we have
1075  * released the lock and the system call should return.
1076  */
1077 static int wait_task_zombie(task_t *p, int noreap,
1078                             struct siginfo __user *infop,
1079                             int __user *stat_addr, struct rusage __user *ru)
1080 {
1081         unsigned long state;
1082         int retval;
1083         int status;
1084
1085         if (unlikely(noreap)) {
1086                 pid_t pid = p->pid;
1087                 uid_t uid = p->uid;
1088                 int exit_code = p->exit_code;
1089                 int why, status;
1090
1091                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1092                         return 0;
1093                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1094                         return 0;
1095                 get_task_struct(p);
1096                 read_unlock(&tasklist_lock);
1097                 if ((exit_code & 0x7f) == 0) {
1098                         why = CLD_EXITED;
1099                         status = exit_code >> 8;
1100                 } else {
1101                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1102                         status = exit_code & 0x7f;
1103                 }
1104                 return wait_noreap_copyout(p, pid, uid, why,
1105                                            status, infop, ru);
1106         }
1107
1108         /*
1109          * Try to move the task's state to DEAD
1110          * only one thread is allowed to do this:
1111          */
1112         state = xchg(&p->exit_state, EXIT_DEAD);
1113         if (state != EXIT_ZOMBIE) {
1114                 BUG_ON(state != EXIT_DEAD);
1115                 return 0;
1116         }
1117         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1118                 /*
1119                  * This can only happen in a race with a ptraced thread
1120                  * dying on another processor.
1121                  */
1122                 return 0;
1123         }
1124
1125         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1126                 struct signal_struct *psig;
1127                 struct signal_struct *sig;
1128
1129                 /*
1130                  * The resource counters for the group leader are in its
1131                  * own task_struct.  Those for dead threads in the group
1132                  * are in its signal_struct, as are those for the child
1133                  * processes it has previously reaped.  All these
1134                  * accumulate in the parent's signal_struct c* fields.
1135                  *
1136                  * We don't bother to take a lock here to protect these
1137                  * p->signal fields, because they are only touched by
1138                  * __exit_signal, which runs with tasklist_lock
1139                  * write-locked anyway, and so is excluded here.  We do
1140                  * need to protect the access to p->parent->signal fields,
1141                  * as other threads in the parent group can be right
1142                  * here reaping other children at the same time.
1143                  */
1144                 spin_lock_irq(&p->parent->sighand->siglock);
1145                 psig = p->parent->signal;
1146                 sig = p->signal;
1147                 psig->cutime =
1148                         cputime_add(psig->cutime,
1149                         cputime_add(p->utime,
1150                         cputime_add(sig->utime,
1151                                     sig->cutime)));
1152                 psig->cstime =
1153                         cputime_add(psig->cstime,
1154                         cputime_add(p->stime,
1155                         cputime_add(sig->stime,
1156                                     sig->cstime)));
1157                 psig->cmin_flt +=
1158                         p->min_flt + sig->min_flt + sig->cmin_flt;
1159                 psig->cmaj_flt +=
1160                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1161                 psig->cnvcsw +=
1162                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1163                 psig->cnivcsw +=
1164                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1165                 spin_unlock_irq(&p->parent->sighand->siglock);
1166         }
1167
1168         /*
1169          * Now we are sure this task is interesting, and no other
1170          * thread can reap it because we set its state to EXIT_DEAD.
1171          */
1172         read_unlock(&tasklist_lock);
1173
1174         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1175         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1176                 ? p->signal->group_exit_code : p->exit_code;
1177         if (!retval && stat_addr)
1178                 retval = put_user(status, stat_addr);
1179         if (!retval && infop)
1180                 retval = put_user(SIGCHLD, &infop->si_signo);
1181         if (!retval && infop)
1182                 retval = put_user(0, &infop->si_errno);
1183         if (!retval && infop) {
1184                 int why;
1185
1186                 if ((status & 0x7f) == 0) {
1187                         why = CLD_EXITED;
1188                         status >>= 8;
1189                 } else {
1190                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1191                         status &= 0x7f;
1192                 }
1193                 retval = put_user((short)why, &infop->si_code);
1194                 if (!retval)
1195                         retval = put_user(status, &infop->si_status);
1196         }
1197         if (!retval && infop)
1198                 retval = put_user(p->pid, &infop->si_pid);
1199         if (!retval && infop)
1200                 retval = put_user(p->uid, &infop->si_uid);
1201         if (retval) {
1202                 // TODO: is this safe?
1203                 p->exit_state = EXIT_ZOMBIE;
1204                 return retval;
1205         }
1206         retval = p->pid;
1207         if (p->real_parent != p->parent) {
1208                 write_lock_irq(&tasklist_lock);
1209                 /* Double-check with lock held.  */
1210                 if (p->real_parent != p->parent) {
1211                         __ptrace_unlink(p);
1212                         // TODO: is this safe?
1213                         p->exit_state = EXIT_ZOMBIE;
1214                         /*
1215                          * If this is not a detached task, notify the parent.
1216                          * If it's still not detached after that, don't release
1217                          * it now.
1218                          */
1219                         if (p->exit_signal != -1) {
1220                                 do_notify_parent(p, p->exit_signal);
1221                                 if (p->exit_signal != -1)
1222                                         p = NULL;
1223                         }
1224                 }
1225                 write_unlock_irq(&tasklist_lock);
1226         }
1227         if (p != NULL)
1228                 release_task(p);
1229         BUG_ON(!retval);
1230         return retval;
1231 }
1232
1233 /*
1234  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1235  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1236  * the lock and this task is uninteresting.  If we return nonzero, we have
1237  * released the lock and the system call should return.
1238  */
1239 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1240                              struct siginfo __user *infop,
1241                              int __user *stat_addr, struct rusage __user *ru)
1242 {
1243         int retval, exit_code;
1244
1245         if (!p->exit_code)
1246                 return 0;
1247         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1248             p->signal && p->signal->group_stop_count > 0)
1249                 /*
1250                  * A group stop is in progress and this is the group leader.
1251                  * We won't report until all threads have stopped.
1252                  */
1253                 return 0;
1254
1255         /*
1256          * Now we are pretty sure this task is interesting.
1257          * Make sure it doesn't get reaped out from under us while we
1258          * give up the lock and then examine it below.  We don't want to
1259          * keep holding onto the tasklist_lock while we call getrusage and
1260          * possibly take page faults for user memory.
1261          */
1262         get_task_struct(p);
1263         read_unlock(&tasklist_lock);
1264
1265         if (unlikely(noreap)) {
1266                 pid_t pid = p->pid;
1267                 uid_t uid = p->uid;
1268                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1269
1270                 exit_code = p->exit_code;
1271                 if (unlikely(!exit_code) ||
1272                     unlikely(p->state & TASK_TRACED))
1273                         goto bail_ref;
1274                 return wait_noreap_copyout(p, pid, uid,
1275                                            why, (exit_code << 8) | 0x7f,
1276                                            infop, ru);
1277         }
1278
1279         write_lock_irq(&tasklist_lock);
1280
1281         /*
1282          * This uses xchg to be atomic with the thread resuming and setting
1283          * it.  It must also be done with the write lock held to prevent a
1284          * race with the EXIT_ZOMBIE case.
1285          */
1286         exit_code = xchg(&p->exit_code, 0);
1287         if (unlikely(p->exit_state)) {
1288                 /*
1289                  * The task resumed and then died.  Let the next iteration
1290                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1291                  * already be zero here if it resumed and did _exit(0).
1292                  * The task itself is dead and won't touch exit_code again;
1293                  * other processors in this function are locked out.
1294                  */
1295                 p->exit_code = exit_code;
1296                 exit_code = 0;
1297         }
1298         if (unlikely(exit_code == 0)) {
1299                 /*
1300                  * Another thread in this function got to it first, or it
1301                  * resumed, or it resumed and then died.
1302                  */
1303                 write_unlock_irq(&tasklist_lock);
1304 bail_ref:
1305                 put_task_struct(p);
1306                 /*
1307                  * We are returning to the wait loop without having successfully
1308                  * removed the process and having released the lock. We cannot
1309                  * continue, since the "p" task pointer is potentially stale.
1310                  *
1311                  * Return -EAGAIN, and do_wait() will restart the loop from the
1312                  * beginning. Do _not_ re-acquire the lock.
1313                  */
1314                 return -EAGAIN;
1315         }
1316
1317         /* move to end of parent's list to avoid starvation */
1318         remove_parent(p);
1319         add_parent(p);
1320
1321         write_unlock_irq(&tasklist_lock);
1322
1323         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1324         if (!retval && stat_addr)
1325                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1326         if (!retval && infop)
1327                 retval = put_user(SIGCHLD, &infop->si_signo);
1328         if (!retval && infop)
1329                 retval = put_user(0, &infop->si_errno);
1330         if (!retval && infop)
1331                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1332                                           ? CLD_TRAPPED : CLD_STOPPED),
1333                                   &infop->si_code);
1334         if (!retval && infop)
1335                 retval = put_user(exit_code, &infop->si_status);
1336         if (!retval && infop)
1337                 retval = put_user(p->pid, &infop->si_pid);
1338         if (!retval && infop)
1339                 retval = put_user(p->uid, &infop->si_uid);
1340         if (!retval)
1341                 retval = p->pid;
1342         put_task_struct(p);
1343
1344         BUG_ON(!retval);
1345         return retval;
1346 }
1347
1348 /*
1349  * Handle do_wait work for one task in a live, non-stopped state.
1350  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1351  * the lock and this task is uninteresting.  If we return nonzero, we have
1352  * released the lock and the system call should return.
1353  */
1354 static int wait_task_continued(task_t *p, int noreap,
1355                                struct siginfo __user *infop,
1356                                int __user *stat_addr, struct rusage __user *ru)
1357 {
1358         int retval;
1359         pid_t pid;
1360         uid_t uid;
1361
1362         if (unlikely(!p->signal))
1363                 return 0;
1364
1365         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1366                 return 0;
1367
1368         spin_lock_irq(&p->sighand->siglock);
1369         /* Re-check with the lock held.  */
1370         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1371                 spin_unlock_irq(&p->sighand->siglock);
1372                 return 0;
1373         }
1374         if (!noreap)
1375                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1376         spin_unlock_irq(&p->sighand->siglock);
1377
1378         pid = p->pid;
1379         uid = p->uid;
1380         get_task_struct(p);
1381         read_unlock(&tasklist_lock);
1382
1383         if (!infop) {
1384                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1385                 put_task_struct(p);
1386                 if (!retval && stat_addr)
1387                         retval = put_user(0xffff, stat_addr);
1388                 if (!retval)
1389                         retval = p->pid;
1390         } else {
1391                 retval = wait_noreap_copyout(p, pid, uid,
1392                                              CLD_CONTINUED, SIGCONT,
1393                                              infop, ru);
1394                 BUG_ON(retval == 0);
1395         }
1396
1397         return retval;
1398 }
1399
1400
1401 static inline int my_ptrace_child(struct task_struct *p)
1402 {
1403         if (!(p->ptrace & PT_PTRACED))
1404                 return 0;
1405         if (!(p->ptrace & PT_ATTACHED))
1406                 return 1;
1407         /*
1408          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1409          * we are the attacher.  If we are the real parent, this is a race
1410          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1411          * which we have to switch the parent links, but has already set
1412          * the flags in p->ptrace.
1413          */
1414         return (p->parent != p->real_parent);
1415 }
1416
1417 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1418                     int __user *stat_addr, struct rusage __user *ru)
1419 {
1420         DECLARE_WAITQUEUE(wait, current);
1421         struct task_struct *tsk;
1422         int flag, retval;
1423
1424         add_wait_queue(&current->signal->wait_chldexit,&wait);
1425 repeat:
1426         /*
1427          * We will set this flag if we see any child that might later
1428          * match our criteria, even if we are not able to reap it yet.
1429          */
1430         flag = 0;
1431         current->state = TASK_INTERRUPTIBLE;
1432         read_lock(&tasklist_lock);
1433         tsk = current;
1434         do {
1435                 struct task_struct *p;
1436                 struct list_head *_p;
1437                 int ret;
1438
1439                 list_for_each(_p,&tsk->children) {
1440                         p = list_entry(_p,struct task_struct,sibling);
1441
1442                         ret = eligible_child(pid, options, p);
1443                         if (!ret)
1444                                 continue;
1445
1446                         switch (p->state) {
1447                         case TASK_TRACED:
1448                                 /*
1449                                  * When we hit the race with PTRACE_ATTACH,
1450                                  * we will not report this child.  But the
1451                                  * race means it has not yet been moved to
1452                                  * our ptrace_children list, so we need to
1453                                  * set the flag here to avoid a spurious ECHILD
1454                                  * when the race happens with the only child.
1455                                  */
1456                                 flag = 1;
1457                                 if (!my_ptrace_child(p))
1458                                         continue;
1459                                 /*FALLTHROUGH*/
1460                         case TASK_STOPPED:
1461                                 /*
1462                                  * It's stopped now, so it might later
1463                                  * continue, exit, or stop again.
1464                                  */
1465                                 flag = 1;
1466                                 if (!(options & WUNTRACED) &&
1467                                     !my_ptrace_child(p))
1468                                         continue;
1469                                 retval = wait_task_stopped(p, ret == 2,
1470                                                            (options & WNOWAIT),
1471                                                            infop,
1472                                                            stat_addr, ru);
1473                                 if (retval == -EAGAIN)
1474                                         goto repeat;
1475                                 if (retval != 0) /* He released the lock.  */
1476                                         goto end;
1477                                 break;
1478                         default:
1479                         // case EXIT_DEAD:
1480                                 if (p->exit_state == EXIT_DEAD)
1481                                         continue;
1482                         // case EXIT_ZOMBIE:
1483                                 if (p->exit_state == EXIT_ZOMBIE) {
1484                                         /*
1485                                          * Eligible but we cannot release
1486                                          * it yet:
1487                                          */
1488                                         if (ret == 2)
1489                                                 goto check_continued;
1490                                         if (!likely(options & WEXITED))
1491                                                 continue;
1492                                         retval = wait_task_zombie(
1493                                                 p, (options & WNOWAIT),
1494                                                 infop, stat_addr, ru);
1495                                         /* He released the lock.  */
1496                                         if (retval != 0)
1497                                                 goto end;
1498                                         break;
1499                                 }
1500 check_continued:
1501                                 /*
1502                                  * It's running now, so it might later
1503                                  * exit, stop, or stop and then continue.
1504                                  */
1505                                 flag = 1;
1506                                 if (!unlikely(options & WCONTINUED))
1507                                         continue;
1508                                 retval = wait_task_continued(
1509                                         p, (options & WNOWAIT),
1510                                         infop, stat_addr, ru);
1511                                 if (retval != 0) /* He released the lock.  */
1512                                         goto end;
1513                                 break;
1514                         }
1515                 }
1516                 if (!flag) {
1517                         list_for_each(_p, &tsk->ptrace_children) {
1518                                 p = list_entry(_p, struct task_struct,
1519                                                 ptrace_list);
1520                                 if (!eligible_child(pid, options, p))
1521                                         continue;
1522                                 flag = 1;
1523                                 break;
1524                         }
1525                 }
1526                 if (options & __WNOTHREAD)
1527                         break;
1528                 tsk = next_thread(tsk);
1529                 BUG_ON(tsk->signal != current->signal);
1530         } while (tsk != current);
1531
1532         read_unlock(&tasklist_lock);
1533         if (flag) {
1534                 retval = 0;
1535                 if (options & WNOHANG)
1536                         goto end;
1537                 retval = -ERESTARTSYS;
1538                 if (signal_pending(current))
1539                         goto end;
1540                 schedule();
1541                 goto repeat;
1542         }
1543         retval = -ECHILD;
1544 end:
1545         current->state = TASK_RUNNING;
1546         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1547         if (infop) {
1548                 if (retval > 0)
1549                 retval = 0;
1550                 else {
1551                         /*
1552                          * For a WNOHANG return, clear out all the fields
1553                          * we would set so the user can easily tell the
1554                          * difference.
1555                          */
1556                         if (!retval)
1557                                 retval = put_user(0, &infop->si_signo);
1558                         if (!retval)
1559                                 retval = put_user(0, &infop->si_errno);
1560                         if (!retval)
1561                                 retval = put_user(0, &infop->si_code);
1562                         if (!retval)
1563                                 retval = put_user(0, &infop->si_pid);
1564                         if (!retval)
1565                                 retval = put_user(0, &infop->si_uid);
1566                         if (!retval)
1567                                 retval = put_user(0, &infop->si_status);
1568                 }
1569         }
1570         return retval;
1571 }
1572
1573 asmlinkage long sys_waitid(int which, pid_t pid,
1574                            struct siginfo __user *infop, int options,
1575                            struct rusage __user *ru)
1576 {
1577         long ret;
1578
1579         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1580                 return -EINVAL;
1581         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1582                 return -EINVAL;
1583
1584         switch (which) {
1585         case P_ALL:
1586                 pid = -1;
1587                 break;
1588         case P_PID:
1589                 if (pid <= 0)
1590                         return -EINVAL;
1591                 break;
1592         case P_PGID:
1593                 if (pid <= 0)
1594                         return -EINVAL;
1595                 pid = -pid;
1596                 break;
1597         default:
1598                 return -EINVAL;
1599         }
1600
1601         ret = do_wait(pid, options, infop, NULL, ru);
1602
1603         /* avoid REGPARM breakage on x86: */
1604         prevent_tail_call(ret);
1605         return ret;
1606 }
1607
1608 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1609                           int options, struct rusage __user *ru)
1610 {
1611         long ret;
1612
1613         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1614                         __WNOTHREAD|__WCLONE|__WALL))
1615                 return -EINVAL;
1616         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1617
1618         /* avoid REGPARM breakage on x86: */
1619         prevent_tail_call(ret);
1620         return ret;
1621 }
1622
1623 #ifdef __ARCH_WANT_SYS_WAITPID
1624
1625 /*
1626  * sys_waitpid() remains for compatibility. waitpid() should be
1627  * implemented by calling sys_wait4() from libc.a.
1628  */
1629 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1630 {
1631         return sys_wait4(pid, stat_addr, options, NULL);
1632 }
1633
1634 #endif