whitespace fixes: task exit handling
[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/mm.h>
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/mnt_namespace.h>
16 #include <linux/key.h>
17 #include <linux/security.h>
18 #include <linux/cpu.h>
19 #include <linux/acct.h>
20 #include <linux/tsacct_kern.h>
21 #include <linux/file.h>
22 #include <linux/binfmts.h>
23 #include <linux/nsproxy.h>
24 #include <linux/pid_namespace.h>
25 #include <linux/ptrace.h>
26 #include <linux/profile.h>
27 #include <linux/mount.h>
28 #include <linux/proc_fs.h>
29 #include <linux/kthread.h>
30 #include <linux/mempolicy.h>
31 #include <linux/taskstats_kern.h>
32 #include <linux/delayacct.h>
33 #include <linux/freezer.h>
34 #include <linux/cgroup.h>
35 #include <linux/syscalls.h>
36 #include <linux/signal.h>
37 #include <linux/posix-timers.h>
38 #include <linux/cn_proc.h>
39 #include <linux/mutex.h>
40 #include <linux/futex.h>
41 #include <linux/compat.h>
42 #include <linux/pipe_fs_i.h>
43 #include <linux/audit.h> /* for audit_free() */
44 #include <linux/resource.h>
45 #include <linux/blkdev.h>
46 #include <linux/task_io_accounting_ops.h>
47
48 #include <asm/uaccess.h>
49 #include <asm/unistd.h>
50 #include <asm/pgtable.h>
51 #include <asm/mmu_context.h>
52
53 extern void sem_exit (void);
54
55 static void exit_mm(struct task_struct * tsk);
56
57 static void __unhash_process(struct task_struct *p)
58 {
59         nr_threads--;
60         detach_pid(p, PIDTYPE_PID);
61         if (thread_group_leader(p)) {
62                 detach_pid(p, PIDTYPE_PGID);
63                 detach_pid(p, PIDTYPE_SID);
64
65                 list_del_rcu(&p->tasks);
66                 __get_cpu_var(process_counts)--;
67         }
68         list_del_rcu(&p->thread_group);
69         remove_parent(p);
70 }
71
72 /*
73  * This function expects the tasklist_lock write-locked.
74  */
75 static void __exit_signal(struct task_struct *tsk)
76 {
77         struct signal_struct *sig = tsk->signal;
78         struct sighand_struct *sighand;
79
80         BUG_ON(!sig);
81         BUG_ON(!atomic_read(&sig->count));
82
83         rcu_read_lock();
84         sighand = rcu_dereference(tsk->sighand);
85         spin_lock(&sighand->siglock);
86
87         posix_cpu_timers_exit(tsk);
88         if (atomic_dec_and_test(&sig->count))
89                 posix_cpu_timers_exit_group(tsk);
90         else {
91                 /*
92                  * If there is any task waiting for the group exit
93                  * then notify it:
94                  */
95                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
96                         wake_up_process(sig->group_exit_task);
97
98                 if (tsk == sig->curr_target)
99                         sig->curr_target = next_thread(tsk);
100                 /*
101                  * Accumulate here the counters for all threads but the
102                  * group leader as they die, so they can be added into
103                  * the process-wide totals when those are taken.
104                  * The group leader stays around as a zombie as long
105                  * as there are other threads.  When it gets reaped,
106                  * the exit.c code will add its counts into these totals.
107                  * We won't ever get here for the group leader, since it
108                  * will have been the last reference on the signal_struct.
109                  */
110                 sig->utime = cputime_add(sig->utime, tsk->utime);
111                 sig->stime = cputime_add(sig->stime, tsk->stime);
112                 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
113                 sig->min_flt += tsk->min_flt;
114                 sig->maj_flt += tsk->maj_flt;
115                 sig->nvcsw += tsk->nvcsw;
116                 sig->nivcsw += tsk->nivcsw;
117                 sig->inblock += task_io_get_inblock(tsk);
118                 sig->oublock += task_io_get_oublock(tsk);
119                 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
120                 sig = NULL; /* Marker for below. */
121         }
122
123         __unhash_process(tsk);
124
125         tsk->signal = NULL;
126         tsk->sighand = NULL;
127         spin_unlock(&sighand->siglock);
128         rcu_read_unlock();
129
130         __cleanup_sighand(sighand);
131         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
132         flush_sigqueue(&tsk->pending);
133         if (sig) {
134                 flush_sigqueue(&sig->shared_pending);
135                 taskstats_tgid_free(sig);
136                 __cleanup_signal(sig);
137         }
138 }
139
140 static void delayed_put_task_struct(struct rcu_head *rhp)
141 {
142         put_task_struct(container_of(rhp, struct task_struct, rcu));
143 }
144
145 void release_task(struct task_struct * p)
146 {
147         struct task_struct *leader;
148         int zap_leader;
149 repeat:
150         atomic_dec(&p->user->processes);
151         write_lock_irq(&tasklist_lock);
152         ptrace_unlink(p);
153         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
154         __exit_signal(p);
155
156         /*
157          * If we are the last non-leader member of the thread
158          * group, and the leader is zombie, then notify the
159          * group leader's parent process. (if it wants notification.)
160          */
161         zap_leader = 0;
162         leader = p->group_leader;
163         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
164                 BUG_ON(leader->exit_signal == -1);
165                 do_notify_parent(leader, leader->exit_signal);
166                 /*
167                  * If we were the last child thread and the leader has
168                  * exited already, and the leader's parent ignores SIGCHLD,
169                  * then we are the one who should release the leader.
170                  *
171                  * do_notify_parent() will have marked it self-reaping in
172                  * that case.
173                  */
174                 zap_leader = (leader->exit_signal == -1);
175         }
176
177         write_unlock_irq(&tasklist_lock);
178         proc_flush_task(p);
179         release_thread(p);
180         call_rcu(&p->rcu, delayed_put_task_struct);
181
182         p = leader;
183         if (unlikely(zap_leader))
184                 goto repeat;
185 }
186
187 /*
188  * This checks not only the pgrp, but falls back on the pid if no
189  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
190  * without this...
191  *
192  * The caller must hold rcu lock or the tasklist lock.
193  */
194 struct pid *session_of_pgrp(struct pid *pgrp)
195 {
196         struct task_struct *p;
197         struct pid *sid = NULL;
198
199         p = pid_task(pgrp, PIDTYPE_PGID);
200         if (p == NULL)
201                 p = pid_task(pgrp, PIDTYPE_PID);
202         if (p != NULL)
203                 sid = task_session(p);
204
205         return sid;
206 }
207
208 /*
209  * Determine if a process group is "orphaned", according to the POSIX
210  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
211  * by terminal-generated stop signals.  Newly orphaned process groups are
212  * to receive a SIGHUP and a SIGCONT.
213  *
214  * "I ask you, have you ever known what it is to be an orphan?"
215  */
216 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
217 {
218         struct task_struct *p;
219         int ret = 1;
220
221         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
222                 if (p == ignored_task
223                                 || p->exit_state
224                                 || is_global_init(p->real_parent))
225                         continue;
226                 if (task_pgrp(p->real_parent) != pgrp &&
227                     task_session(p->real_parent) == task_session(p)) {
228                         ret = 0;
229                         break;
230                 }
231         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
232         return ret;     /* (sighing) "Often!" */
233 }
234
235 int is_current_pgrp_orphaned(void)
236 {
237         int retval;
238
239         read_lock(&tasklist_lock);
240         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
241         read_unlock(&tasklist_lock);
242
243         return retval;
244 }
245
246 static int has_stopped_jobs(struct pid *pgrp)
247 {
248         int retval = 0;
249         struct task_struct *p;
250
251         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
252                 if (p->state != TASK_STOPPED)
253                         continue;
254                 retval = 1;
255                 break;
256         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
257         return retval;
258 }
259
260 /**
261  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
262  *
263  * If a kernel thread is launched as a result of a system call, or if
264  * it ever exits, it should generally reparent itself to kthreadd so it
265  * isn't in the way of other processes and is correctly cleaned up on exit.
266  *
267  * The various task state such as scheduling policy and priority may have
268  * been inherited from a user process, so we reset them to sane values here.
269  *
270  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
271  */
272 static void reparent_to_kthreadd(void)
273 {
274         write_lock_irq(&tasklist_lock);
275
276         ptrace_unlink(current);
277         /* Reparent to init */
278         remove_parent(current);
279         current->real_parent = current->parent = kthreadd_task;
280         add_parent(current);
281
282         /* Set the exit signal to SIGCHLD so we signal init on exit */
283         current->exit_signal = SIGCHLD;
284
285         if (task_nice(current) < 0)
286                 set_user_nice(current, 0);
287         /* cpus_allowed? */
288         /* rt_priority? */
289         /* signals? */
290         security_task_reparent_to_init(current);
291         memcpy(current->signal->rlim, init_task.signal->rlim,
292                sizeof(current->signal->rlim));
293         atomic_inc(&(INIT_USER->__count));
294         write_unlock_irq(&tasklist_lock);
295         switch_uid(INIT_USER);
296 }
297
298 void __set_special_pids(pid_t session, pid_t pgrp)
299 {
300         struct task_struct *curr = current->group_leader;
301
302         if (task_session_nr(curr) != session) {
303                 detach_pid(curr, PIDTYPE_SID);
304                 set_task_session(curr, session);
305                 attach_pid(curr, PIDTYPE_SID, find_pid(session));
306         }
307         if (task_pgrp_nr(curr) != pgrp) {
308                 detach_pid(curr, PIDTYPE_PGID);
309                 curr->signal->pgrp = pgrp;
310                 attach_pid(curr, PIDTYPE_PGID, find_pid(pgrp));
311         }
312 }
313
314 static void set_special_pids(pid_t session, pid_t pgrp)
315 {
316         write_lock_irq(&tasklist_lock);
317         __set_special_pids(session, pgrp);
318         write_unlock_irq(&tasklist_lock);
319 }
320
321 /*
322  * Let kernel threads use this to say that they
323  * allow a certain signal (since daemonize() will
324  * have disabled all of them by default).
325  */
326 int allow_signal(int sig)
327 {
328         if (!valid_signal(sig) || sig < 1)
329                 return -EINVAL;
330
331         spin_lock_irq(&current->sighand->siglock);
332         sigdelset(&current->blocked, sig);
333         if (!current->mm) {
334                 /* Kernel threads handle their own signals.
335                    Let the signal code know it'll be handled, so
336                    that they don't get converted to SIGKILL or
337                    just silently dropped */
338                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
339         }
340         recalc_sigpending();
341         spin_unlock_irq(&current->sighand->siglock);
342         return 0;
343 }
344
345 EXPORT_SYMBOL(allow_signal);
346
347 int disallow_signal(int sig)
348 {
349         if (!valid_signal(sig) || sig < 1)
350                 return -EINVAL;
351
352         spin_lock_irq(&current->sighand->siglock);
353         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
354         recalc_sigpending();
355         spin_unlock_irq(&current->sighand->siglock);
356         return 0;
357 }
358
359 EXPORT_SYMBOL(disallow_signal);
360
361 /*
362  *      Put all the gunge required to become a kernel thread without
363  *      attached user resources in one place where it belongs.
364  */
365
366 void daemonize(const char *name, ...)
367 {
368         va_list args;
369         struct fs_struct *fs;
370         sigset_t blocked;
371
372         va_start(args, name);
373         vsnprintf(current->comm, sizeof(current->comm), name, args);
374         va_end(args);
375
376         /*
377          * If we were started as result of loading a module, close all of the
378          * user space pages.  We don't need them, and if we didn't close them
379          * they would be locked into memory.
380          */
381         exit_mm(current);
382         /*
383          * We don't want to have TIF_FREEZE set if the system-wide hibernation
384          * or suspend transition begins right now.
385          */
386         current->flags |= PF_NOFREEZE;
387
388         set_special_pids(1, 1);
389         proc_clear_tty(current);
390
391         /* Block and flush all signals */
392         sigfillset(&blocked);
393         sigprocmask(SIG_BLOCK, &blocked, NULL);
394         flush_signals(current);
395
396         /* Become as one with the init task */
397
398         exit_fs(current);       /* current->fs->count--; */
399         fs = init_task.fs;
400         current->fs = fs;
401         atomic_inc(&fs->count);
402
403         if (current->nsproxy != init_task.nsproxy) {
404                 get_nsproxy(init_task.nsproxy);
405                 switch_task_namespaces(current, init_task.nsproxy);
406         }
407
408         exit_files(current);
409         current->files = init_task.files;
410         atomic_inc(&current->files->count);
411
412         reparent_to_kthreadd();
413 }
414
415 EXPORT_SYMBOL(daemonize);
416
417 static void close_files(struct files_struct * files)
418 {
419         int i, j;
420         struct fdtable *fdt;
421
422         j = 0;
423
424         /*
425          * It is safe to dereference the fd table without RCU or
426          * ->file_lock because this is the last reference to the
427          * files structure.
428          */
429         fdt = files_fdtable(files);
430         for (;;) {
431                 unsigned long set;
432                 i = j * __NFDBITS;
433                 if (i >= fdt->max_fds)
434                         break;
435                 set = fdt->open_fds->fds_bits[j++];
436                 while (set) {
437                         if (set & 1) {
438                                 struct file * file = xchg(&fdt->fd[i], NULL);
439                                 if (file) {
440                                         filp_close(file, files);
441                                         cond_resched();
442                                 }
443                         }
444                         i++;
445                         set >>= 1;
446                 }
447         }
448 }
449
450 struct files_struct *get_files_struct(struct task_struct *task)
451 {
452         struct files_struct *files;
453
454         task_lock(task);
455         files = task->files;
456         if (files)
457                 atomic_inc(&files->count);
458         task_unlock(task);
459
460         return files;
461 }
462
463 void fastcall put_files_struct(struct files_struct *files)
464 {
465         struct fdtable *fdt;
466
467         if (atomic_dec_and_test(&files->count)) {
468                 close_files(files);
469                 /*
470                  * Free the fd and fdset arrays if we expanded them.
471                  * If the fdtable was embedded, pass files for freeing
472                  * at the end of the RCU grace period. Otherwise,
473                  * you can free files immediately.
474                  */
475                 fdt = files_fdtable(files);
476                 if (fdt != &files->fdtab)
477                         kmem_cache_free(files_cachep, files);
478                 free_fdtable(fdt);
479         }
480 }
481
482 EXPORT_SYMBOL(put_files_struct);
483
484 void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
485 {
486         struct files_struct *old;
487
488         old = tsk->files;
489         task_lock(tsk);
490         tsk->files = files;
491         task_unlock(tsk);
492         put_files_struct(old);
493 }
494 EXPORT_SYMBOL(reset_files_struct);
495
496 static inline void __exit_files(struct task_struct *tsk)
497 {
498         struct files_struct * files = tsk->files;
499
500         if (files) {
501                 task_lock(tsk);
502                 tsk->files = NULL;
503                 task_unlock(tsk);
504                 put_files_struct(files);
505         }
506 }
507
508 void exit_files(struct task_struct *tsk)
509 {
510         __exit_files(tsk);
511 }
512
513 static inline void __put_fs_struct(struct fs_struct *fs)
514 {
515         /* No need to hold fs->lock if we are killing it */
516         if (atomic_dec_and_test(&fs->count)) {
517                 dput(fs->root);
518                 mntput(fs->rootmnt);
519                 dput(fs->pwd);
520                 mntput(fs->pwdmnt);
521                 if (fs->altroot) {
522                         dput(fs->altroot);
523                         mntput(fs->altrootmnt);
524                 }
525                 kmem_cache_free(fs_cachep, fs);
526         }
527 }
528
529 void put_fs_struct(struct fs_struct *fs)
530 {
531         __put_fs_struct(fs);
532 }
533
534 static inline void __exit_fs(struct task_struct *tsk)
535 {
536         struct fs_struct * fs = tsk->fs;
537
538         if (fs) {
539                 task_lock(tsk);
540                 tsk->fs = NULL;
541                 task_unlock(tsk);
542                 __put_fs_struct(fs);
543         }
544 }
545
546 void exit_fs(struct task_struct *tsk)
547 {
548         __exit_fs(tsk);
549 }
550
551 EXPORT_SYMBOL_GPL(exit_fs);
552
553 /*
554  * Turn us into a lazy TLB process if we
555  * aren't already..
556  */
557 static void exit_mm(struct task_struct * tsk)
558 {
559         struct mm_struct *mm = tsk->mm;
560
561         mm_release(tsk, mm);
562         if (!mm)
563                 return;
564         /*
565          * Serialize with any possible pending coredump.
566          * We must hold mmap_sem around checking core_waiters
567          * and clearing tsk->mm.  The core-inducing thread
568          * will increment core_waiters for each thread in the
569          * group with ->mm != NULL.
570          */
571         down_read(&mm->mmap_sem);
572         if (mm->core_waiters) {
573                 up_read(&mm->mmap_sem);
574                 down_write(&mm->mmap_sem);
575                 if (!--mm->core_waiters)
576                         complete(mm->core_startup_done);
577                 up_write(&mm->mmap_sem);
578
579                 wait_for_completion(&mm->core_done);
580                 down_read(&mm->mmap_sem);
581         }
582         atomic_inc(&mm->mm_count);
583         BUG_ON(mm != tsk->active_mm);
584         /* more a memory barrier than a real lock */
585         task_lock(tsk);
586         tsk->mm = NULL;
587         up_read(&mm->mmap_sem);
588         enter_lazy_tlb(mm, current);
589         /* We don't want this task to be frozen prematurely */
590         clear_freeze_flag(tsk);
591         task_unlock(tsk);
592         mmput(mm);
593 }
594
595 static void
596 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
597 {
598         if (p->pdeath_signal)
599                 /* We already hold the tasklist_lock here.  */
600                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
601
602         /* Move the child from its dying parent to the new one.  */
603         if (unlikely(traced)) {
604                 /* Preserve ptrace links if someone else is tracing this child.  */
605                 list_del_init(&p->ptrace_list);
606                 if (p->parent != p->real_parent)
607                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
608         } else {
609                 /* If this child is being traced, then we're the one tracing it
610                  * anyway, so let go of it.
611                  */
612                 p->ptrace = 0;
613                 remove_parent(p);
614                 p->parent = p->real_parent;
615                 add_parent(p);
616
617                 if (p->state == TASK_TRACED) {
618                         /*
619                          * If it was at a trace stop, turn it into
620                          * a normal stop since it's no longer being
621                          * traced.
622                          */
623                         ptrace_untrace(p);
624                 }
625         }
626
627         /* If this is a threaded reparent there is no need to
628          * notify anyone anything has happened.
629          */
630         if (p->real_parent->group_leader == father->group_leader)
631                 return;
632
633         /* We don't want people slaying init.  */
634         if (p->exit_signal != -1)
635                 p->exit_signal = SIGCHLD;
636
637         /* If we'd notified the old parent about this child's death,
638          * also notify the new parent.
639          */
640         if (!traced && p->exit_state == EXIT_ZOMBIE &&
641             p->exit_signal != -1 && thread_group_empty(p))
642                 do_notify_parent(p, p->exit_signal);
643
644         /*
645          * process group orphan check
646          * Case ii: Our child is in a different pgrp
647          * than we are, and it was the only connection
648          * outside, so the child pgrp is now orphaned.
649          */
650         if ((task_pgrp(p) != task_pgrp(father)) &&
651             (task_session(p) == task_session(father))) {
652                 struct pid *pgrp = task_pgrp(p);
653
654                 if (will_become_orphaned_pgrp(pgrp, NULL) &&
655                     has_stopped_jobs(pgrp)) {
656                         __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
657                         __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
658                 }
659         }
660 }
661
662 /*
663  * When we die, we re-parent all our children.
664  * Try to give them to another thread in our thread
665  * group, and if no such member exists, give it to
666  * the child reaper process (ie "init") in our pid
667  * space.
668  */
669 static void
670 forget_original_parent(struct task_struct *father, struct list_head *to_release)
671 {
672         struct task_struct *p, *n, *reaper = father;
673
674         do {
675                 reaper = next_thread(reaper);
676                 if (reaper == father) {
677                         reaper = task_child_reaper(father);
678                         break;
679                 }
680         } while (reaper->exit_state);
681
682         /*
683          * There are only two places where our children can be:
684          *
685          * - in our child list
686          * - in our ptraced child list
687          *
688          * Search them and reparent children.
689          */
690         list_for_each_entry_safe(p, n, &father->children, sibling) {
691                 int ptrace;
692
693                 ptrace = p->ptrace;
694
695                 /* if father isn't the real parent, then ptrace must be enabled */
696                 BUG_ON(father != p->real_parent && !ptrace);
697
698                 if (father == p->real_parent) {
699                         /* reparent with a reaper, real father it's us */
700                         p->real_parent = reaper;
701                         reparent_thread(p, father, 0);
702                 } else {
703                         /* reparent ptraced task to its real parent */
704                         __ptrace_unlink (p);
705                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
706                             thread_group_empty(p))
707                                 do_notify_parent(p, p->exit_signal);
708                 }
709
710                 /*
711                  * if the ptraced child is a zombie with exit_signal == -1
712                  * we must collect it before we exit, or it will remain
713                  * zombie forever since we prevented it from self-reap itself
714                  * while it was being traced by us, to be able to see it in wait4.
715                  */
716                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
717                         list_add(&p->ptrace_list, to_release);
718         }
719         list_for_each_entry_safe(p, n, &father->ptrace_children, ptrace_list) {
720                 p->real_parent = reaper;
721                 reparent_thread(p, father, 1);
722         }
723 }
724
725 /*
726  * Send signals to all our closest relatives so that they know
727  * to properly mourn us..
728  */
729 static void exit_notify(struct task_struct *tsk)
730 {
731         int state;
732         struct task_struct *t;
733         struct list_head ptrace_dead, *_p, *_n;
734         struct pid *pgrp;
735
736         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
737             && !thread_group_empty(tsk)) {
738                 /*
739                  * This occurs when there was a race between our exit
740                  * syscall and a group signal choosing us as the one to
741                  * wake up.  It could be that we are the only thread
742                  * alerted to check for pending signals, but another thread
743                  * should be woken now to take the signal since we will not.
744                  * Now we'll wake all the threads in the group just to make
745                  * sure someone gets all the pending signals.
746                  */
747                 spin_lock_irq(&tsk->sighand->siglock);
748                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
749                         if (!signal_pending(t) && !(t->flags & PF_EXITING))
750                                 recalc_sigpending_and_wake(t);
751                 spin_unlock_irq(&tsk->sighand->siglock);
752         }
753
754         write_lock_irq(&tasklist_lock);
755
756         /*
757          * This does two things:
758          *
759          * A.  Make init inherit all the child processes
760          * B.  Check to see if any process groups have become orphaned
761          *      as a result of our exiting, and if they have any stopped
762          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
763          */
764
765         INIT_LIST_HEAD(&ptrace_dead);
766         forget_original_parent(tsk, &ptrace_dead);
767         BUG_ON(!list_empty(&tsk->children));
768         BUG_ON(!list_empty(&tsk->ptrace_children));
769
770         /*
771          * Check to see if any process groups have become orphaned
772          * as a result of our exiting, and if they have any stopped
773          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
774          *
775          * Case i: Our father is in a different pgrp than we are
776          * and we were the only connection outside, so our pgrp
777          * is about to become orphaned.
778          */
779         t = tsk->real_parent;
780
781         pgrp = task_pgrp(tsk);
782         if ((task_pgrp(t) != pgrp) &&
783             (task_session(t) == task_session(tsk)) &&
784             will_become_orphaned_pgrp(pgrp, tsk) &&
785             has_stopped_jobs(pgrp)) {
786                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
787                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
788         }
789
790         /* Let father know we died
791          *
792          * Thread signals are configurable, but you aren't going to use
793          * that to send signals to arbitary processes.
794          * That stops right now.
795          *
796          * If the parent exec id doesn't match the exec id we saved
797          * when we started then we know the parent has changed security
798          * domain.
799          *
800          * If our self_exec id doesn't match our parent_exec_id then
801          * we have changed execution domain as these two values started
802          * the same after a fork.
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 && likely(!tsk->ptrace))
824                 state = EXIT_DEAD;
825         tsk->exit_state = state;
826
827         if (thread_group_leader(tsk) &&
828             tsk->signal->notify_count < 0 &&
829             tsk->signal->group_exit_task)
830                 wake_up_process(tsk->signal->group_exit_task);
831
832         write_unlock_irq(&tasklist_lock);
833
834         list_for_each_safe(_p, _n, &ptrace_dead) {
835                 list_del_init(_p);
836                 t = list_entry(_p, struct task_struct, ptrace_list);
837                 release_task(t);
838         }
839
840         /* If the process is dead, release it - nobody will wait for it */
841         if (state == EXIT_DEAD)
842                 release_task(tsk);
843 }
844
845 #ifdef CONFIG_DEBUG_STACK_USAGE
846 static void check_stack_usage(void)
847 {
848         static DEFINE_SPINLOCK(low_water_lock);
849         static int lowest_to_date = THREAD_SIZE;
850         unsigned long *n = end_of_stack(current);
851         unsigned long free;
852
853         while (*n == 0)
854                 n++;
855         free = (unsigned long)n - (unsigned long)end_of_stack(current);
856
857         if (free >= lowest_to_date)
858                 return;
859
860         spin_lock(&low_water_lock);
861         if (free < lowest_to_date) {
862                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
863                                 "left\n",
864                                 current->comm, free);
865                 lowest_to_date = free;
866         }
867         spin_unlock(&low_water_lock);
868 }
869 #else
870 static inline void check_stack_usage(void) {}
871 #endif
872
873 static inline void exit_child_reaper(struct task_struct *tsk)
874 {
875         if (likely(tsk->group_leader != task_child_reaper(tsk)))
876                 return;
877
878         panic("Attempted to kill init!");
879 }
880
881 fastcall NORET_TYPE void do_exit(long code)
882 {
883         struct task_struct *tsk = current;
884         int group_dead;
885
886         profile_task_exit(tsk);
887
888         WARN_ON(atomic_read(&tsk->fs_excl));
889
890         if (unlikely(in_interrupt()))
891                 panic("Aiee, killing interrupt handler!");
892         if (unlikely(!tsk->pid))
893                 panic("Attempted to kill the idle task!");
894
895         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
896                 current->ptrace_message = code;
897                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
898         }
899
900         /*
901          * We're taking recursive faults here in do_exit. Safest is to just
902          * leave this task alone and wait for reboot.
903          */
904         if (unlikely(tsk->flags & PF_EXITING)) {
905                 printk(KERN_ALERT
906                         "Fixing recursive fault but reboot is needed!\n");
907                 /*
908                  * We can do this unlocked here. The futex code uses
909                  * this flag just to verify whether the pi state
910                  * cleanup has been done or not. In the worst case it
911                  * loops once more. We pretend that the cleanup was
912                  * done as there is no way to return. Either the
913                  * OWNER_DIED bit is set by now or we push the blocked
914                  * task into the wait for ever nirwana as well.
915                  */
916                 tsk->flags |= PF_EXITPIDONE;
917                 if (tsk->io_context)
918                         exit_io_context();
919                 set_current_state(TASK_UNINTERRUPTIBLE);
920                 schedule();
921         }
922
923         tsk->flags |= PF_EXITING;
924         /*
925          * tsk->flags are checked in the futex code to protect against
926          * an exiting task cleaning up the robust pi futexes.
927          */
928         smp_mb();
929         spin_unlock_wait(&tsk->pi_lock);
930
931         if (unlikely(in_atomic()))
932                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
933                                 current->comm, current->pid,
934                                 preempt_count());
935
936         acct_update_integrals(tsk);
937         if (tsk->mm) {
938                 update_hiwater_rss(tsk->mm);
939                 update_hiwater_vm(tsk->mm);
940         }
941         group_dead = atomic_dec_and_test(&tsk->signal->live);
942         if (group_dead) {
943                 exit_child_reaper(tsk);
944                 hrtimer_cancel(&tsk->signal->real_timer);
945                 exit_itimers(tsk->signal);
946         }
947         acct_collect(code, group_dead);
948 #ifdef CONFIG_FUTEX
949         if (unlikely(tsk->robust_list))
950                 exit_robust_list(tsk);
951 #ifdef CONFIG_COMPAT
952         if (unlikely(tsk->compat_robust_list))
953                 compat_exit_robust_list(tsk);
954 #endif
955 #endif
956         if (group_dead)
957                 tty_audit_exit();
958         if (unlikely(tsk->audit_context))
959                 audit_free(tsk);
960
961         tsk->exit_code = code;
962         taskstats_exit(tsk, group_dead);
963
964         exit_mm(tsk);
965
966         if (group_dead)
967                 acct_process();
968         exit_sem(tsk);
969         __exit_files(tsk);
970         __exit_fs(tsk);
971         check_stack_usage();
972         exit_thread();
973         cgroup_exit(tsk, 1);
974         exit_keys(tsk);
975
976         if (group_dead && tsk->signal->leader)
977                 disassociate_ctty(1);
978
979         module_put(task_thread_info(tsk)->exec_domain->module);
980         if (tsk->binfmt)
981                 module_put(tsk->binfmt->module);
982
983         proc_exit_connector(tsk);
984         exit_task_namespaces(tsk);
985         exit_notify(tsk);
986 #ifdef CONFIG_NUMA
987         mpol_free(tsk->mempolicy);
988         tsk->mempolicy = NULL;
989 #endif
990 #ifdef CONFIG_FUTEX
991         /*
992          * This must happen late, after the PID is not
993          * hashed anymore:
994          */
995         if (unlikely(!list_empty(&tsk->pi_state_list)))
996                 exit_pi_state_list(tsk);
997         if (unlikely(current->pi_state_cache))
998                 kfree(current->pi_state_cache);
999 #endif
1000         /*
1001          * Make sure we are holding no locks:
1002          */
1003         debug_check_no_locks_held(tsk);
1004         /*
1005          * We can do this unlocked here. The futex code uses this flag
1006          * just to verify whether the pi state cleanup has been done
1007          * or not. In the worst case it loops once more.
1008          */
1009         tsk->flags |= PF_EXITPIDONE;
1010
1011         if (tsk->io_context)
1012                 exit_io_context();
1013
1014         if (tsk->splice_pipe)
1015                 __free_pipe_info(tsk->splice_pipe);
1016
1017         preempt_disable();
1018         /* causes final put_task_struct in finish_task_switch(). */
1019         tsk->state = TASK_DEAD;
1020
1021         schedule();
1022         BUG();
1023         /* Avoid "noreturn function does return".  */
1024         for (;;)
1025                 cpu_relax();    /* For when BUG is null */
1026 }
1027
1028 EXPORT_SYMBOL_GPL(do_exit);
1029
1030 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1031 {
1032         if (comp)
1033                 complete(comp);
1034
1035         do_exit(code);
1036 }
1037
1038 EXPORT_SYMBOL(complete_and_exit);
1039
1040 asmlinkage long sys_exit(int error_code)
1041 {
1042         do_exit((error_code&0xff)<<8);
1043 }
1044
1045 /*
1046  * Take down every thread in the group.  This is called by fatal signals
1047  * as well as by sys_exit_group (below).
1048  */
1049 NORET_TYPE void
1050 do_group_exit(int exit_code)
1051 {
1052         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1053
1054         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1055                 exit_code = current->signal->group_exit_code;
1056         else if (!thread_group_empty(current)) {
1057                 struct signal_struct *const sig = current->signal;
1058                 struct sighand_struct *const sighand = current->sighand;
1059                 spin_lock_irq(&sighand->siglock);
1060                 if (sig->flags & SIGNAL_GROUP_EXIT)
1061                         /* Another thread got here before we took the lock.  */
1062                         exit_code = sig->group_exit_code;
1063                 else {
1064                         sig->group_exit_code = exit_code;
1065                         zap_other_threads(current);
1066                 }
1067                 spin_unlock_irq(&sighand->siglock);
1068         }
1069
1070         do_exit(exit_code);
1071         /* NOTREACHED */
1072 }
1073
1074 /*
1075  * this kills every thread in the thread group. Note that any externally
1076  * wait4()-ing process will get the correct exit code - even if this
1077  * thread is not the thread group leader.
1078  */
1079 asmlinkage void sys_exit_group(int error_code)
1080 {
1081         do_group_exit((error_code & 0xff) << 8);
1082 }
1083
1084 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1085 {
1086         int err;
1087
1088         if (pid > 0) {
1089                 if (p->pid != pid)
1090                         return 0;
1091         } else if (!pid) {
1092                 if (task_pgrp_nr(p) != task_pgrp_nr(current))
1093                         return 0;
1094         } else if (pid != -1) {
1095                 if (task_pgrp_nr(p) != -pid)
1096                         return 0;
1097         }
1098
1099         /*
1100          * Do not consider detached threads that are
1101          * not ptraced:
1102          */
1103         if (p->exit_signal == -1 && !p->ptrace)
1104                 return 0;
1105
1106         /* Wait for all children (clone and not) if __WALL is set;
1107          * otherwise, wait for clone children *only* if __WCLONE is
1108          * set; otherwise, wait for non-clone children *only*.  (Note:
1109          * A "clone" child here is one that reports to its parent
1110          * using a signal other than SIGCHLD.) */
1111         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1112             && !(options & __WALL))
1113                 return 0;
1114         /*
1115          * Do not consider thread group leaders that are
1116          * in a non-empty thread group:
1117          */
1118         if (delay_group_leader(p))
1119                 return 2;
1120
1121         err = security_task_wait(p);
1122         if (err)
1123                 return err;
1124
1125         return 1;
1126 }
1127
1128 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1129                                int why, int status,
1130                                struct siginfo __user *infop,
1131                                struct rusage __user *rusagep)
1132 {
1133         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1134
1135         put_task_struct(p);
1136         if (!retval)
1137                 retval = put_user(SIGCHLD, &infop->si_signo);
1138         if (!retval)
1139                 retval = put_user(0, &infop->si_errno);
1140         if (!retval)
1141                 retval = put_user((short)why, &infop->si_code);
1142         if (!retval)
1143                 retval = put_user(pid, &infop->si_pid);
1144         if (!retval)
1145                 retval = put_user(uid, &infop->si_uid);
1146         if (!retval)
1147                 retval = put_user(status, &infop->si_status);
1148         if (!retval)
1149                 retval = pid;
1150         return retval;
1151 }
1152
1153 /*
1154  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1155  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1156  * the lock and this task is uninteresting.  If we return nonzero, we have
1157  * released the lock and the system call should return.
1158  */
1159 static int wait_task_zombie(struct task_struct *p, int noreap,
1160                             struct siginfo __user *infop,
1161                             int __user *stat_addr, struct rusage __user *ru)
1162 {
1163         unsigned long state;
1164         int retval, status, traced;
1165
1166         if (unlikely(noreap)) {
1167                 pid_t pid = p->pid;
1168                 uid_t uid = p->uid;
1169                 int exit_code = p->exit_code;
1170                 int why, status;
1171
1172                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1173                         return 0;
1174                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1175                         return 0;
1176                 get_task_struct(p);
1177                 read_unlock(&tasklist_lock);
1178                 if ((exit_code & 0x7f) == 0) {
1179                         why = CLD_EXITED;
1180                         status = exit_code >> 8;
1181                 } else {
1182                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1183                         status = exit_code & 0x7f;
1184                 }
1185                 return wait_noreap_copyout(p, pid, uid, why,
1186                                            status, infop, ru);
1187         }
1188
1189         /*
1190          * Try to move the task's state to DEAD
1191          * only one thread is allowed to do this:
1192          */
1193         state = xchg(&p->exit_state, EXIT_DEAD);
1194         if (state != EXIT_ZOMBIE) {
1195                 BUG_ON(state != EXIT_DEAD);
1196                 return 0;
1197         }
1198
1199         /* traced means p->ptrace, but not vice versa */
1200         traced = (p->real_parent != p->parent);
1201
1202         if (likely(!traced)) {
1203                 struct signal_struct *psig;
1204                 struct signal_struct *sig;
1205
1206                 /*
1207                  * The resource counters for the group leader are in its
1208                  * own task_struct.  Those for dead threads in the group
1209                  * are in its signal_struct, as are those for the child
1210                  * processes it has previously reaped.  All these
1211                  * accumulate in the parent's signal_struct c* fields.
1212                  *
1213                  * We don't bother to take a lock here to protect these
1214                  * p->signal fields, because they are only touched by
1215                  * __exit_signal, which runs with tasklist_lock
1216                  * write-locked anyway, and so is excluded here.  We do
1217                  * need to protect the access to p->parent->signal fields,
1218                  * as other threads in the parent group can be right
1219                  * here reaping other children at the same time.
1220                  */
1221                 spin_lock_irq(&p->parent->sighand->siglock);
1222                 psig = p->parent->signal;
1223                 sig = p->signal;
1224                 psig->cutime =
1225                         cputime_add(psig->cutime,
1226                         cputime_add(p->utime,
1227                         cputime_add(sig->utime,
1228                                     sig->cutime)));
1229                 psig->cstime =
1230                         cputime_add(psig->cstime,
1231                         cputime_add(p->stime,
1232                         cputime_add(sig->stime,
1233                                     sig->cstime)));
1234                 psig->cgtime =
1235                         cputime_add(psig->cgtime,
1236                         cputime_add(p->gtime,
1237                         cputime_add(sig->gtime,
1238                                     sig->cgtime)));
1239                 psig->cmin_flt +=
1240                         p->min_flt + sig->min_flt + sig->cmin_flt;
1241                 psig->cmaj_flt +=
1242                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1243                 psig->cnvcsw +=
1244                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1245                 psig->cnivcsw +=
1246                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1247                 psig->cinblock +=
1248                         task_io_get_inblock(p) +
1249                         sig->inblock + sig->cinblock;
1250                 psig->coublock +=
1251                         task_io_get_oublock(p) +
1252                         sig->oublock + sig->coublock;
1253                 spin_unlock_irq(&p->parent->sighand->siglock);
1254         }
1255
1256         /*
1257          * Now we are sure this task is interesting, and no other
1258          * thread can reap it because we set its state to EXIT_DEAD.
1259          */
1260         read_unlock(&tasklist_lock);
1261
1262         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1263         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1264                 ? p->signal->group_exit_code : p->exit_code;
1265         if (!retval && stat_addr)
1266                 retval = put_user(status, stat_addr);
1267         if (!retval && infop)
1268                 retval = put_user(SIGCHLD, &infop->si_signo);
1269         if (!retval && infop)
1270                 retval = put_user(0, &infop->si_errno);
1271         if (!retval && infop) {
1272                 int why;
1273
1274                 if ((status & 0x7f) == 0) {
1275                         why = CLD_EXITED;
1276                         status >>= 8;
1277                 } else {
1278                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1279                         status &= 0x7f;
1280                 }
1281                 retval = put_user((short)why, &infop->si_code);
1282                 if (!retval)
1283                         retval = put_user(status, &infop->si_status);
1284         }
1285         if (!retval && infop)
1286                 retval = put_user(p->pid, &infop->si_pid);
1287         if (!retval && infop)
1288                 retval = put_user(p->uid, &infop->si_uid);
1289         if (!retval)
1290                 retval = p->pid;
1291
1292         if (traced) {
1293                 write_lock_irq(&tasklist_lock);
1294                 /* We dropped tasklist, ptracer could die and untrace */
1295                 ptrace_unlink(p);
1296                 /*
1297                  * If this is not a detached task, notify the parent.
1298                  * If it's still not detached after that, don't release
1299                  * it now.
1300                  */
1301                 if (p->exit_signal != -1) {
1302                         do_notify_parent(p, p->exit_signal);
1303                         if (p->exit_signal != -1) {
1304                                 p->exit_state = EXIT_ZOMBIE;
1305                                 p = NULL;
1306                         }
1307                 }
1308                 write_unlock_irq(&tasklist_lock);
1309         }
1310         if (p != NULL)
1311                 release_task(p);
1312
1313         return retval;
1314 }
1315
1316 /*
1317  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1318  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1319  * the lock and this task is uninteresting.  If we return nonzero, we have
1320  * released the lock and the system call should return.
1321  */
1322 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1323                              int noreap, struct siginfo __user *infop,
1324                              int __user *stat_addr, struct rusage __user *ru)
1325 {
1326         int retval, exit_code;
1327
1328         if (!p->exit_code)
1329                 return 0;
1330         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1331             p->signal->group_stop_count > 0)
1332                 /*
1333                  * A group stop is in progress and this is the group leader.
1334                  * We won't report until all threads have stopped.
1335                  */
1336                 return 0;
1337
1338         /*
1339          * Now we are pretty sure this task is interesting.
1340          * Make sure it doesn't get reaped out from under us while we
1341          * give up the lock and then examine it below.  We don't want to
1342          * keep holding onto the tasklist_lock while we call getrusage and
1343          * possibly take page faults for user memory.
1344          */
1345         get_task_struct(p);
1346         read_unlock(&tasklist_lock);
1347
1348         if (unlikely(noreap)) {
1349                 pid_t pid = p->pid;
1350                 uid_t uid = p->uid;
1351                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1352
1353                 exit_code = p->exit_code;
1354                 if (unlikely(!exit_code) ||
1355                     unlikely(p->state & TASK_TRACED))
1356                         goto bail_ref;
1357                 return wait_noreap_copyout(p, pid, uid,
1358                                            why, (exit_code << 8) | 0x7f,
1359                                            infop, ru);
1360         }
1361
1362         write_lock_irq(&tasklist_lock);
1363
1364         /*
1365          * This uses xchg to be atomic with the thread resuming and setting
1366          * it.  It must also be done with the write lock held to prevent a
1367          * race with the EXIT_ZOMBIE case.
1368          */
1369         exit_code = xchg(&p->exit_code, 0);
1370         if (unlikely(p->exit_state)) {
1371                 /*
1372                  * The task resumed and then died.  Let the next iteration
1373                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1374                  * already be zero here if it resumed and did _exit(0).
1375                  * The task itself is dead and won't touch exit_code again;
1376                  * other processors in this function are locked out.
1377                  */
1378                 p->exit_code = exit_code;
1379                 exit_code = 0;
1380         }
1381         if (unlikely(exit_code == 0)) {
1382                 /*
1383                  * Another thread in this function got to it first, or it
1384                  * resumed, or it resumed and then died.
1385                  */
1386                 write_unlock_irq(&tasklist_lock);
1387 bail_ref:
1388                 put_task_struct(p);
1389                 /*
1390                  * We are returning to the wait loop without having successfully
1391                  * removed the process and having released the lock. We cannot
1392                  * continue, since the "p" task pointer is potentially stale.
1393                  *
1394                  * Return -EAGAIN, and do_wait() will restart the loop from the
1395                  * beginning. Do _not_ re-acquire the lock.
1396                  */
1397                 return -EAGAIN;
1398         }
1399
1400         /* move to end of parent's list to avoid starvation */
1401         remove_parent(p);
1402         add_parent(p);
1403
1404         write_unlock_irq(&tasklist_lock);
1405
1406         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1407         if (!retval && stat_addr)
1408                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1409         if (!retval && infop)
1410                 retval = put_user(SIGCHLD, &infop->si_signo);
1411         if (!retval && infop)
1412                 retval = put_user(0, &infop->si_errno);
1413         if (!retval && infop)
1414                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1415                                           ? CLD_TRAPPED : CLD_STOPPED),
1416                                   &infop->si_code);
1417         if (!retval && infop)
1418                 retval = put_user(exit_code, &infop->si_status);
1419         if (!retval && infop)
1420                 retval = put_user(p->pid, &infop->si_pid);
1421         if (!retval && infop)
1422                 retval = put_user(p->uid, &infop->si_uid);
1423         if (!retval)
1424                 retval = p->pid;
1425         put_task_struct(p);
1426
1427         BUG_ON(!retval);
1428         return retval;
1429 }
1430
1431 /*
1432  * Handle do_wait work for one task in a live, non-stopped state.
1433  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1434  * the lock and this task is uninteresting.  If we return nonzero, we have
1435  * released the lock and the system call should return.
1436  */
1437 static int wait_task_continued(struct task_struct *p, int noreap,
1438                                struct siginfo __user *infop,
1439                                int __user *stat_addr, struct rusage __user *ru)
1440 {
1441         int retval;
1442         pid_t pid;
1443         uid_t uid;
1444
1445         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1446                 return 0;
1447
1448         spin_lock_irq(&p->sighand->siglock);
1449         /* Re-check with the lock held.  */
1450         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1451                 spin_unlock_irq(&p->sighand->siglock);
1452                 return 0;
1453         }
1454         if (!noreap)
1455                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1456         spin_unlock_irq(&p->sighand->siglock);
1457
1458         pid = p->pid;
1459         uid = p->uid;
1460         get_task_struct(p);
1461         read_unlock(&tasklist_lock);
1462
1463         if (!infop) {
1464                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1465                 put_task_struct(p);
1466                 if (!retval && stat_addr)
1467                         retval = put_user(0xffff, stat_addr);
1468                 if (!retval)
1469                         retval = p->pid;
1470         } else {
1471                 retval = wait_noreap_copyout(p, pid, uid,
1472                                              CLD_CONTINUED, SIGCONT,
1473                                              infop, ru);
1474                 BUG_ON(retval == 0);
1475         }
1476
1477         return retval;
1478 }
1479
1480
1481 static inline int my_ptrace_child(struct task_struct *p)
1482 {
1483         if (!(p->ptrace & PT_PTRACED))
1484                 return 0;
1485         if (!(p->ptrace & PT_ATTACHED))
1486                 return 1;
1487         /*
1488          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1489          * we are the attacher.  If we are the real parent, this is a race
1490          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1491          * which we have to switch the parent links, but has already set
1492          * the flags in p->ptrace.
1493          */
1494         return (p->parent != p->real_parent);
1495 }
1496
1497 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1498                     int __user *stat_addr, struct rusage __user *ru)
1499 {
1500         DECLARE_WAITQUEUE(wait, current);
1501         struct task_struct *tsk;
1502         int flag, retval;
1503         int allowed, denied;
1504
1505         add_wait_queue(&current->signal->wait_chldexit,&wait);
1506 repeat:
1507         /*
1508          * We will set this flag if we see any child that might later
1509          * match our criteria, even if we are not able to reap it yet.
1510          */
1511         flag = 0;
1512         allowed = denied = 0;
1513         current->state = TASK_INTERRUPTIBLE;
1514         read_lock(&tasklist_lock);
1515         tsk = current;
1516         do {
1517                 struct task_struct *p;
1518                 int ret;
1519
1520                 list_for_each_entry(p, &tsk->children, sibling) {
1521                         ret = eligible_child(pid, options, p);
1522                         if (!ret)
1523                                 continue;
1524
1525                         if (unlikely(ret < 0)) {
1526                                 denied = ret;
1527                                 continue;
1528                         }
1529                         allowed = 1;
1530
1531                         switch (p->state) {
1532                         case TASK_TRACED:
1533                                 /*
1534                                  * When we hit the race with PTRACE_ATTACH,
1535                                  * we will not report this child.  But the
1536                                  * race means it has not yet been moved to
1537                                  * our ptrace_children list, so we need to
1538                                  * set the flag here to avoid a spurious ECHILD
1539                                  * when the race happens with the only child.
1540                                  */
1541                                 flag = 1;
1542                                 if (!my_ptrace_child(p))
1543                                         continue;
1544                                 /*FALLTHROUGH*/
1545                         case TASK_STOPPED:
1546                                 /*
1547                                  * It's stopped now, so it might later
1548                                  * continue, exit, or stop again.
1549                                  */
1550                                 flag = 1;
1551                                 if (!(options & WUNTRACED) &&
1552                                     !my_ptrace_child(p))
1553                                         continue;
1554                                 retval = wait_task_stopped(p, ret == 2,
1555                                                            (options & WNOWAIT),
1556                                                            infop,
1557                                                            stat_addr, ru);
1558                                 if (retval == -EAGAIN)
1559                                         goto repeat;
1560                                 if (retval != 0) /* He released the lock.  */
1561                                         goto end;
1562                                 break;
1563                         default:
1564                         // case EXIT_DEAD:
1565                                 if (p->exit_state == EXIT_DEAD)
1566                                         continue;
1567                         // case EXIT_ZOMBIE:
1568                                 if (p->exit_state == EXIT_ZOMBIE) {
1569                                         /*
1570                                          * Eligible but we cannot release
1571                                          * it yet:
1572                                          */
1573                                         if (ret == 2)
1574                                                 goto check_continued;
1575                                         if (!likely(options & WEXITED))
1576                                                 continue;
1577                                         retval = wait_task_zombie(
1578                                                 p, (options & WNOWAIT),
1579                                                 infop, stat_addr, ru);
1580                                         /* He released the lock.  */
1581                                         if (retval != 0)
1582                                                 goto end;
1583                                         break;
1584                                 }
1585 check_continued:
1586                                 /*
1587                                  * It's running now, so it might later
1588                                  * exit, stop, or stop and then continue.
1589                                  */
1590                                 flag = 1;
1591                                 if (!unlikely(options & WCONTINUED))
1592                                         continue;
1593                                 retval = wait_task_continued(
1594                                         p, (options & WNOWAIT),
1595                                         infop, stat_addr, ru);
1596                                 if (retval != 0) /* He released the lock.  */
1597                                         goto end;
1598                                 break;
1599                         }
1600                 }
1601                 if (!flag) {
1602                         list_for_each_entry(p, &tsk->ptrace_children,
1603                                             ptrace_list) {
1604                                 if (!eligible_child(pid, options, p))
1605                                         continue;
1606                                 flag = 1;
1607                                 break;
1608                         }
1609                 }
1610                 if (options & __WNOTHREAD)
1611                         break;
1612                 tsk = next_thread(tsk);
1613                 BUG_ON(tsk->signal != current->signal);
1614         } while (tsk != current);
1615
1616         read_unlock(&tasklist_lock);
1617         if (flag) {
1618                 retval = 0;
1619                 if (options & WNOHANG)
1620                         goto end;
1621                 retval = -ERESTARTSYS;
1622                 if (signal_pending(current))
1623                         goto end;
1624                 schedule();
1625                 goto repeat;
1626         }
1627         retval = -ECHILD;
1628         if (unlikely(denied) && !allowed)
1629                 retval = denied;
1630 end:
1631         current->state = TASK_RUNNING;
1632         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1633         if (infop) {
1634                 if (retval > 0)
1635                 retval = 0;
1636                 else {
1637                         /*
1638                          * For a WNOHANG return, clear out all the fields
1639                          * we would set so the user can easily tell the
1640                          * difference.
1641                          */
1642                         if (!retval)
1643                                 retval = put_user(0, &infop->si_signo);
1644                         if (!retval)
1645                                 retval = put_user(0, &infop->si_errno);
1646                         if (!retval)
1647                                 retval = put_user(0, &infop->si_code);
1648                         if (!retval)
1649                                 retval = put_user(0, &infop->si_pid);
1650                         if (!retval)
1651                                 retval = put_user(0, &infop->si_uid);
1652                         if (!retval)
1653                                 retval = put_user(0, &infop->si_status);
1654                 }
1655         }
1656         return retval;
1657 }
1658
1659 asmlinkage long sys_waitid(int which, pid_t pid,
1660                            struct siginfo __user *infop, int options,
1661                            struct rusage __user *ru)
1662 {
1663         long ret;
1664
1665         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1666                 return -EINVAL;
1667         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1668                 return -EINVAL;
1669
1670         switch (which) {
1671         case P_ALL:
1672                 pid = -1;
1673                 break;
1674         case P_PID:
1675                 if (pid <= 0)
1676                         return -EINVAL;
1677                 break;
1678         case P_PGID:
1679                 if (pid <= 0)
1680                         return -EINVAL;
1681                 pid = -pid;
1682                 break;
1683         default:
1684                 return -EINVAL;
1685         }
1686
1687         ret = do_wait(pid, options, infop, NULL, ru);
1688
1689         /* avoid REGPARM breakage on x86: */
1690         prevent_tail_call(ret);
1691         return ret;
1692 }
1693
1694 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1695                           int options, struct rusage __user *ru)
1696 {
1697         long ret;
1698
1699         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1700                         __WNOTHREAD|__WCLONE|__WALL))
1701                 return -EINVAL;
1702         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1703
1704         /* avoid REGPARM breakage on x86: */
1705         prevent_tail_call(ret);
1706         return ret;
1707 }
1708
1709 #ifdef __ARCH_WANT_SYS_WAITPID
1710
1711 /*
1712  * sys_waitpid() remains for compatibility. waitpid() should be
1713  * implemented by calling sys_wait4() from libc.a.
1714  */
1715 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1716 {
1717         return sys_wait4(pid, stat_addr, options, NULL);
1718 }
1719
1720 #endif