Merge git://git.kernel.org/pub/scm/linux/kernel/git/wim/linux-2.6-watchdog
[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         proc_flush_task(p);
152         write_lock_irq(&tasklist_lock);
153         ptrace_unlink(p);
154         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
155         __exit_signal(p);
156
157         /*
158          * If we are the last non-leader member of the thread
159          * group, and the leader is zombie, then notify the
160          * group leader's parent process. (if it wants notification.)
161          */
162         zap_leader = 0;
163         leader = p->group_leader;
164         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
165                 BUG_ON(leader->exit_signal == -1);
166                 do_notify_parent(leader, leader->exit_signal);
167                 /*
168                  * If we were the last child thread and the leader has
169                  * exited already, and the leader's parent ignores SIGCHLD,
170                  * then we are the one who should release the leader.
171                  *
172                  * do_notify_parent() will have marked it self-reaping in
173                  * that case.
174                  */
175                 zap_leader = (leader->exit_signal == -1);
176         }
177
178         write_unlock_irq(&tasklist_lock);
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                 set_task_pgrp(curr, 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 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 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 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 forget_original_parent(struct task_struct *father)
670 {
671         struct task_struct *p, *n, *reaper = father;
672         struct list_head ptrace_dead;
673
674         INIT_LIST_HEAD(&ptrace_dead);
675
676         write_lock_irq(&tasklist_lock);
677
678         do {
679                 reaper = next_thread(reaper);
680                 if (reaper == father) {
681                         reaper = task_child_reaper(father);
682                         break;
683                 }
684         } while (reaper->flags & PF_EXITING);
685
686         /*
687          * There are only two places where our children can be:
688          *
689          * - in our child list
690          * - in our ptraced child list
691          *
692          * Search them and reparent children.
693          */
694         list_for_each_entry_safe(p, n, &father->children, sibling) {
695                 int ptrace;
696
697                 ptrace = p->ptrace;
698
699                 /* if father isn't the real parent, then ptrace must be enabled */
700                 BUG_ON(father != p->real_parent && !ptrace);
701
702                 if (father == p->real_parent) {
703                         /* reparent with a reaper, real father it's us */
704                         p->real_parent = reaper;
705                         reparent_thread(p, father, 0);
706                 } else {
707                         /* reparent ptraced task to its real parent */
708                         __ptrace_unlink (p);
709                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
710                             thread_group_empty(p))
711                                 do_notify_parent(p, p->exit_signal);
712                 }
713
714                 /*
715                  * if the ptraced child is a zombie with exit_signal == -1
716                  * we must collect it before we exit, or it will remain
717                  * zombie forever since we prevented it from self-reap itself
718                  * while it was being traced by us, to be able to see it in wait4.
719                  */
720                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
721                         list_add(&p->ptrace_list, &ptrace_dead);
722         }
723
724         list_for_each_entry_safe(p, n, &father->ptrace_children, ptrace_list) {
725                 p->real_parent = reaper;
726                 reparent_thread(p, father, 1);
727         }
728
729         write_unlock_irq(&tasklist_lock);
730         BUG_ON(!list_empty(&father->children));
731         BUG_ON(!list_empty(&father->ptrace_children));
732
733         list_for_each_entry_safe(p, n, &ptrace_dead, ptrace_list) {
734                 list_del_init(&p->ptrace_list);
735                 release_task(p);
736         }
737
738 }
739
740 /*
741  * Send signals to all our closest relatives so that they know
742  * to properly mourn us..
743  */
744 static void exit_notify(struct task_struct *tsk)
745 {
746         int state;
747         struct task_struct *t;
748         struct pid *pgrp;
749
750         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
751             && !thread_group_empty(tsk)) {
752                 /*
753                  * This occurs when there was a race between our exit
754                  * syscall and a group signal choosing us as the one to
755                  * wake up.  It could be that we are the only thread
756                  * alerted to check for pending signals, but another thread
757                  * should be woken now to take the signal since we will not.
758                  * Now we'll wake all the threads in the group just to make
759                  * sure someone gets all the pending signals.
760                  */
761                 spin_lock_irq(&tsk->sighand->siglock);
762                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
763                         if (!signal_pending(t) && !(t->flags & PF_EXITING))
764                                 recalc_sigpending_and_wake(t);
765                 spin_unlock_irq(&tsk->sighand->siglock);
766         }
767
768         /*
769          * This does two things:
770          *
771          * A.  Make init inherit all the child processes
772          * B.  Check to see if any process groups have become orphaned
773          *      as a result of our exiting, and if they have any stopped
774          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
775          */
776         forget_original_parent(tsk);
777         exit_task_namespaces(tsk);
778
779         write_lock_irq(&tasklist_lock);
780         /*
781          * Check to see if any process groups have become orphaned
782          * as a result of our exiting, and if they have any stopped
783          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
784          *
785          * Case i: Our father is in a different pgrp than we are
786          * and we were the only connection outside, so our pgrp
787          * is about to become orphaned.
788          */
789         t = tsk->real_parent;
790
791         pgrp = task_pgrp(tsk);
792         if ((task_pgrp(t) != pgrp) &&
793             (task_session(t) == task_session(tsk)) &&
794             will_become_orphaned_pgrp(pgrp, tsk) &&
795             has_stopped_jobs(pgrp)) {
796                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
797                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
798         }
799
800         /* Let father know we died
801          *
802          * Thread signals are configurable, but you aren't going to use
803          * that to send signals to arbitary processes.
804          * That stops right now.
805          *
806          * If the parent exec id doesn't match the exec id we saved
807          * when we started then we know the parent has changed security
808          * domain.
809          *
810          * If our self_exec id doesn't match our parent_exec_id then
811          * we have changed execution domain as these two values started
812          * the same after a fork.
813          */
814         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
815             ( tsk->parent_exec_id != t->self_exec_id  ||
816               tsk->self_exec_id != tsk->parent_exec_id)
817             && !capable(CAP_KILL))
818                 tsk->exit_signal = SIGCHLD;
819
820
821         /* If something other than our normal parent is ptracing us, then
822          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
823          * only has special meaning to our real parent.
824          */
825         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
826                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
827                 do_notify_parent(tsk, signal);
828         } else if (tsk->ptrace) {
829                 do_notify_parent(tsk, SIGCHLD);
830         }
831
832         state = EXIT_ZOMBIE;
833         if (tsk->exit_signal == -1 && likely(!tsk->ptrace))
834                 state = EXIT_DEAD;
835         tsk->exit_state = state;
836
837         if (thread_group_leader(tsk) &&
838             tsk->signal->notify_count < 0 &&
839             tsk->signal->group_exit_task)
840                 wake_up_process(tsk->signal->group_exit_task);
841
842         write_unlock_irq(&tasklist_lock);
843
844         /* If the process is dead, release it - nobody will wait for it */
845         if (state == EXIT_DEAD)
846                 release_task(tsk);
847 }
848
849 #ifdef CONFIG_DEBUG_STACK_USAGE
850 static void check_stack_usage(void)
851 {
852         static DEFINE_SPINLOCK(low_water_lock);
853         static int lowest_to_date = THREAD_SIZE;
854         unsigned long *n = end_of_stack(current);
855         unsigned long free;
856
857         while (*n == 0)
858                 n++;
859         free = (unsigned long)n - (unsigned long)end_of_stack(current);
860
861         if (free >= lowest_to_date)
862                 return;
863
864         spin_lock(&low_water_lock);
865         if (free < lowest_to_date) {
866                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
867                                 "left\n",
868                                 current->comm, free);
869                 lowest_to_date = free;
870         }
871         spin_unlock(&low_water_lock);
872 }
873 #else
874 static inline void check_stack_usage(void) {}
875 #endif
876
877 static inline void exit_child_reaper(struct task_struct *tsk)
878 {
879         if (likely(tsk->group_leader != task_child_reaper(tsk)))
880                 return;
881
882         if (tsk->nsproxy->pid_ns == &init_pid_ns)
883                 panic("Attempted to kill init!");
884
885         /*
886          * @tsk is the last thread in the 'cgroup-init' and is exiting.
887          * Terminate all remaining processes in the namespace and reap them
888          * before exiting @tsk.
889          *
890          * Note that @tsk (last thread of cgroup-init) may not necessarily
891          * be the child-reaper (i.e main thread of cgroup-init) of the
892          * namespace i.e the child_reaper may have already exited.
893          *
894          * Even after a child_reaper exits, we let it inherit orphaned children,
895          * because, pid_ns->child_reaper remains valid as long as there is
896          * at least one living sub-thread in the cgroup init.
897
898          * This living sub-thread of the cgroup-init will be notified when
899          * a child inherited by the 'child-reaper' exits (do_notify_parent()
900          * uses __group_send_sig_info()). Further, when reaping child processes,
901          * do_wait() iterates over children of all living sub threads.
902
903          * i.e even though 'child_reaper' thread is listed as the parent of the
904          * orphaned children, any living sub-thread in the cgroup-init can
905          * perform the role of the child_reaper.
906          */
907         zap_pid_ns_processes(tsk->nsproxy->pid_ns);
908 }
909
910 fastcall NORET_TYPE void do_exit(long code)
911 {
912         struct task_struct *tsk = current;
913         int group_dead;
914
915         profile_task_exit(tsk);
916
917         WARN_ON(atomic_read(&tsk->fs_excl));
918
919         if (unlikely(in_interrupt()))
920                 panic("Aiee, killing interrupt handler!");
921         if (unlikely(!tsk->pid))
922                 panic("Attempted to kill the idle task!");
923
924         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
925                 current->ptrace_message = code;
926                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
927         }
928
929         /*
930          * We're taking recursive faults here in do_exit. Safest is to just
931          * leave this task alone and wait for reboot.
932          */
933         if (unlikely(tsk->flags & PF_EXITING)) {
934                 printk(KERN_ALERT
935                         "Fixing recursive fault but reboot is needed!\n");
936                 /*
937                  * We can do this unlocked here. The futex code uses
938                  * this flag just to verify whether the pi state
939                  * cleanup has been done or not. In the worst case it
940                  * loops once more. We pretend that the cleanup was
941                  * done as there is no way to return. Either the
942                  * OWNER_DIED bit is set by now or we push the blocked
943                  * task into the wait for ever nirwana as well.
944                  */
945                 tsk->flags |= PF_EXITPIDONE;
946                 if (tsk->io_context)
947                         exit_io_context();
948                 set_current_state(TASK_UNINTERRUPTIBLE);
949                 schedule();
950         }
951
952         tsk->flags |= PF_EXITING;
953         /*
954          * tsk->flags are checked in the futex code to protect against
955          * an exiting task cleaning up the robust pi futexes.
956          */
957         smp_mb();
958         spin_unlock_wait(&tsk->pi_lock);
959
960         if (unlikely(in_atomic()))
961                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
962                                 current->comm, task_pid_nr(current),
963                                 preempt_count());
964
965         acct_update_integrals(tsk);
966         if (tsk->mm) {
967                 update_hiwater_rss(tsk->mm);
968                 update_hiwater_vm(tsk->mm);
969         }
970         group_dead = atomic_dec_and_test(&tsk->signal->live);
971         if (group_dead) {
972                 exit_child_reaper(tsk);
973                 hrtimer_cancel(&tsk->signal->real_timer);
974                 exit_itimers(tsk->signal);
975         }
976         acct_collect(code, group_dead);
977 #ifdef CONFIG_FUTEX
978         if (unlikely(tsk->robust_list))
979                 exit_robust_list(tsk);
980 #ifdef CONFIG_COMPAT
981         if (unlikely(tsk->compat_robust_list))
982                 compat_exit_robust_list(tsk);
983 #endif
984 #endif
985         if (group_dead)
986                 tty_audit_exit();
987         if (unlikely(tsk->audit_context))
988                 audit_free(tsk);
989
990         tsk->exit_code = code;
991         taskstats_exit(tsk, group_dead);
992
993         exit_mm(tsk);
994
995         if (group_dead)
996                 acct_process();
997         exit_sem(tsk);
998         __exit_files(tsk);
999         __exit_fs(tsk);
1000         check_stack_usage();
1001         exit_thread();
1002         cgroup_exit(tsk, 1);
1003         exit_keys(tsk);
1004
1005         if (group_dead && tsk->signal->leader)
1006                 disassociate_ctty(1);
1007
1008         module_put(task_thread_info(tsk)->exec_domain->module);
1009         if (tsk->binfmt)
1010                 module_put(tsk->binfmt->module);
1011
1012         proc_exit_connector(tsk);
1013         exit_notify(tsk);
1014 #ifdef CONFIG_NUMA
1015         mpol_free(tsk->mempolicy);
1016         tsk->mempolicy = NULL;
1017 #endif
1018 #ifdef CONFIG_FUTEX
1019         /*
1020          * This must happen late, after the PID is not
1021          * hashed anymore:
1022          */
1023         if (unlikely(!list_empty(&tsk->pi_state_list)))
1024                 exit_pi_state_list(tsk);
1025         if (unlikely(current->pi_state_cache))
1026                 kfree(current->pi_state_cache);
1027 #endif
1028         /*
1029          * Make sure we are holding no locks:
1030          */
1031         debug_check_no_locks_held(tsk);
1032         /*
1033          * We can do this unlocked here. The futex code uses this flag
1034          * just to verify whether the pi state cleanup has been done
1035          * or not. In the worst case it loops once more.
1036          */
1037         tsk->flags |= PF_EXITPIDONE;
1038
1039         if (tsk->io_context)
1040                 exit_io_context();
1041
1042         if (tsk->splice_pipe)
1043                 __free_pipe_info(tsk->splice_pipe);
1044
1045         preempt_disable();
1046         /* causes final put_task_struct in finish_task_switch(). */
1047         tsk->state = TASK_DEAD;
1048
1049         schedule();
1050         BUG();
1051         /* Avoid "noreturn function does return".  */
1052         for (;;)
1053                 cpu_relax();    /* For when BUG is null */
1054 }
1055
1056 EXPORT_SYMBOL_GPL(do_exit);
1057
1058 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1059 {
1060         if (comp)
1061                 complete(comp);
1062
1063         do_exit(code);
1064 }
1065
1066 EXPORT_SYMBOL(complete_and_exit);
1067
1068 asmlinkage long sys_exit(int error_code)
1069 {
1070         do_exit((error_code&0xff)<<8);
1071 }
1072
1073 /*
1074  * Take down every thread in the group.  This is called by fatal signals
1075  * as well as by sys_exit_group (below).
1076  */
1077 NORET_TYPE void
1078 do_group_exit(int exit_code)
1079 {
1080         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1081
1082         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1083                 exit_code = current->signal->group_exit_code;
1084         else if (!thread_group_empty(current)) {
1085                 struct signal_struct *const sig = current->signal;
1086                 struct sighand_struct *const sighand = current->sighand;
1087                 spin_lock_irq(&sighand->siglock);
1088                 if (sig->flags & SIGNAL_GROUP_EXIT)
1089                         /* Another thread got here before we took the lock.  */
1090                         exit_code = sig->group_exit_code;
1091                 else {
1092                         sig->group_exit_code = exit_code;
1093                         zap_other_threads(current);
1094                 }
1095                 spin_unlock_irq(&sighand->siglock);
1096         }
1097
1098         do_exit(exit_code);
1099         /* NOTREACHED */
1100 }
1101
1102 /*
1103  * this kills every thread in the thread group. Note that any externally
1104  * wait4()-ing process will get the correct exit code - even if this
1105  * thread is not the thread group leader.
1106  */
1107 asmlinkage void sys_exit_group(int error_code)
1108 {
1109         do_group_exit((error_code & 0xff) << 8);
1110 }
1111
1112 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1113 {
1114         int err;
1115         struct pid_namespace *ns;
1116
1117         ns = current->nsproxy->pid_ns;
1118         if (pid > 0) {
1119                 if (task_pid_nr_ns(p, ns) != pid)
1120                         return 0;
1121         } else if (!pid) {
1122                 if (task_pgrp_nr_ns(p, ns) != task_pgrp_vnr(current))
1123                         return 0;
1124         } else if (pid != -1) {
1125                 if (task_pgrp_nr_ns(p, ns) != -pid)
1126                         return 0;
1127         }
1128
1129         /*
1130          * Do not consider detached threads that are
1131          * not ptraced:
1132          */
1133         if (p->exit_signal == -1 && !p->ptrace)
1134                 return 0;
1135
1136         /* Wait for all children (clone and not) if __WALL is set;
1137          * otherwise, wait for clone children *only* if __WCLONE is
1138          * set; otherwise, wait for non-clone children *only*.  (Note:
1139          * A "clone" child here is one that reports to its parent
1140          * using a signal other than SIGCHLD.) */
1141         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1142             && !(options & __WALL))
1143                 return 0;
1144         /*
1145          * Do not consider thread group leaders that are
1146          * in a non-empty thread group:
1147          */
1148         if (delay_group_leader(p))
1149                 return 2;
1150
1151         err = security_task_wait(p);
1152         if (err)
1153                 return err;
1154
1155         return 1;
1156 }
1157
1158 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1159                                int why, int status,
1160                                struct siginfo __user *infop,
1161                                struct rusage __user *rusagep)
1162 {
1163         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1164
1165         put_task_struct(p);
1166         if (!retval)
1167                 retval = put_user(SIGCHLD, &infop->si_signo);
1168         if (!retval)
1169                 retval = put_user(0, &infop->si_errno);
1170         if (!retval)
1171                 retval = put_user((short)why, &infop->si_code);
1172         if (!retval)
1173                 retval = put_user(pid, &infop->si_pid);
1174         if (!retval)
1175                 retval = put_user(uid, &infop->si_uid);
1176         if (!retval)
1177                 retval = put_user(status, &infop->si_status);
1178         if (!retval)
1179                 retval = pid;
1180         return retval;
1181 }
1182
1183 /*
1184  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1185  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1186  * the lock and this task is uninteresting.  If we return nonzero, we have
1187  * released the lock and the system call should return.
1188  */
1189 static int wait_task_zombie(struct task_struct *p, int noreap,
1190                             struct siginfo __user *infop,
1191                             int __user *stat_addr, struct rusage __user *ru)
1192 {
1193         unsigned long state;
1194         int retval, status, traced;
1195         struct pid_namespace *ns;
1196
1197         ns = current->nsproxy->pid_ns;
1198
1199         if (unlikely(noreap)) {
1200                 pid_t pid = task_pid_nr_ns(p, ns);
1201                 uid_t uid = p->uid;
1202                 int exit_code = p->exit_code;
1203                 int why, status;
1204
1205                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1206                         return 0;
1207                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1208                         return 0;
1209                 get_task_struct(p);
1210                 read_unlock(&tasklist_lock);
1211                 if ((exit_code & 0x7f) == 0) {
1212                         why = CLD_EXITED;
1213                         status = exit_code >> 8;
1214                 } else {
1215                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1216                         status = exit_code & 0x7f;
1217                 }
1218                 return wait_noreap_copyout(p, pid, uid, why,
1219                                            status, infop, ru);
1220         }
1221
1222         /*
1223          * Try to move the task's state to DEAD
1224          * only one thread is allowed to do this:
1225          */
1226         state = xchg(&p->exit_state, EXIT_DEAD);
1227         if (state != EXIT_ZOMBIE) {
1228                 BUG_ON(state != EXIT_DEAD);
1229                 return 0;
1230         }
1231
1232         /* traced means p->ptrace, but not vice versa */
1233         traced = (p->real_parent != p->parent);
1234
1235         if (likely(!traced)) {
1236                 struct signal_struct *psig;
1237                 struct signal_struct *sig;
1238
1239                 /*
1240                  * The resource counters for the group leader are in its
1241                  * own task_struct.  Those for dead threads in the group
1242                  * are in its signal_struct, as are those for the child
1243                  * processes it has previously reaped.  All these
1244                  * accumulate in the parent's signal_struct c* fields.
1245                  *
1246                  * We don't bother to take a lock here to protect these
1247                  * p->signal fields, because they are only touched by
1248                  * __exit_signal, which runs with tasklist_lock
1249                  * write-locked anyway, and so is excluded here.  We do
1250                  * need to protect the access to p->parent->signal fields,
1251                  * as other threads in the parent group can be right
1252                  * here reaping other children at the same time.
1253                  */
1254                 spin_lock_irq(&p->parent->sighand->siglock);
1255                 psig = p->parent->signal;
1256                 sig = p->signal;
1257                 psig->cutime =
1258                         cputime_add(psig->cutime,
1259                         cputime_add(p->utime,
1260                         cputime_add(sig->utime,
1261                                     sig->cutime)));
1262                 psig->cstime =
1263                         cputime_add(psig->cstime,
1264                         cputime_add(p->stime,
1265                         cputime_add(sig->stime,
1266                                     sig->cstime)));
1267                 psig->cgtime =
1268                         cputime_add(psig->cgtime,
1269                         cputime_add(p->gtime,
1270                         cputime_add(sig->gtime,
1271                                     sig->cgtime)));
1272                 psig->cmin_flt +=
1273                         p->min_flt + sig->min_flt + sig->cmin_flt;
1274                 psig->cmaj_flt +=
1275                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1276                 psig->cnvcsw +=
1277                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1278                 psig->cnivcsw +=
1279                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1280                 psig->cinblock +=
1281                         task_io_get_inblock(p) +
1282                         sig->inblock + sig->cinblock;
1283                 psig->coublock +=
1284                         task_io_get_oublock(p) +
1285                         sig->oublock + sig->coublock;
1286                 spin_unlock_irq(&p->parent->sighand->siglock);
1287         }
1288
1289         /*
1290          * Now we are sure this task is interesting, and no other
1291          * thread can reap it because we set its state to EXIT_DEAD.
1292          */
1293         read_unlock(&tasklist_lock);
1294
1295         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1296         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1297                 ? p->signal->group_exit_code : p->exit_code;
1298         if (!retval && stat_addr)
1299                 retval = put_user(status, stat_addr);
1300         if (!retval && infop)
1301                 retval = put_user(SIGCHLD, &infop->si_signo);
1302         if (!retval && infop)
1303                 retval = put_user(0, &infop->si_errno);
1304         if (!retval && infop) {
1305                 int why;
1306
1307                 if ((status & 0x7f) == 0) {
1308                         why = CLD_EXITED;
1309                         status >>= 8;
1310                 } else {
1311                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1312                         status &= 0x7f;
1313                 }
1314                 retval = put_user((short)why, &infop->si_code);
1315                 if (!retval)
1316                         retval = put_user(status, &infop->si_status);
1317         }
1318         if (!retval && infop)
1319                 retval = put_user(task_pid_nr_ns(p, ns), &infop->si_pid);
1320         if (!retval && infop)
1321                 retval = put_user(p->uid, &infop->si_uid);
1322         if (!retval)
1323                 retval = task_pid_nr_ns(p, ns);
1324
1325         if (traced) {
1326                 write_lock_irq(&tasklist_lock);
1327                 /* We dropped tasklist, ptracer could die and untrace */
1328                 ptrace_unlink(p);
1329                 /*
1330                  * If this is not a detached task, notify the parent.
1331                  * If it's still not detached after that, don't release
1332                  * it now.
1333                  */
1334                 if (p->exit_signal != -1) {
1335                         do_notify_parent(p, p->exit_signal);
1336                         if (p->exit_signal != -1) {
1337                                 p->exit_state = EXIT_ZOMBIE;
1338                                 p = NULL;
1339                         }
1340                 }
1341                 write_unlock_irq(&tasklist_lock);
1342         }
1343         if (p != NULL)
1344                 release_task(p);
1345
1346         return retval;
1347 }
1348
1349 /*
1350  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1351  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1352  * the lock and this task is uninteresting.  If we return nonzero, we have
1353  * released the lock and the system call should return.
1354  */
1355 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1356                              int noreap, struct siginfo __user *infop,
1357                              int __user *stat_addr, struct rusage __user *ru)
1358 {
1359         int retval, exit_code;
1360         pid_t pid;
1361
1362         if (!p->exit_code)
1363                 return 0;
1364         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1365             p->signal->group_stop_count > 0)
1366                 /*
1367                  * A group stop is in progress and this is the group leader.
1368                  * We won't report until all threads have stopped.
1369                  */
1370                 return 0;
1371
1372         /*
1373          * Now we are pretty sure this task is interesting.
1374          * Make sure it doesn't get reaped out from under us while we
1375          * give up the lock and then examine it below.  We don't want to
1376          * keep holding onto the tasklist_lock while we call getrusage and
1377          * possibly take page faults for user memory.
1378          */
1379         pid = task_pid_nr_ns(p, current->nsproxy->pid_ns);
1380         get_task_struct(p);
1381         read_unlock(&tasklist_lock);
1382
1383         if (unlikely(noreap)) {
1384                 uid_t uid = p->uid;
1385                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1386
1387                 exit_code = p->exit_code;
1388                 if (unlikely(!exit_code) || unlikely(p->exit_state))
1389                         goto bail_ref;
1390                 return wait_noreap_copyout(p, pid, uid,
1391                                            why, exit_code,
1392                                            infop, ru);
1393         }
1394
1395         write_lock_irq(&tasklist_lock);
1396
1397         /*
1398          * This uses xchg to be atomic with the thread resuming and setting
1399          * it.  It must also be done with the write lock held to prevent a
1400          * race with the EXIT_ZOMBIE case.
1401          */
1402         exit_code = xchg(&p->exit_code, 0);
1403         if (unlikely(p->exit_state)) {
1404                 /*
1405                  * The task resumed and then died.  Let the next iteration
1406                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1407                  * already be zero here if it resumed and did _exit(0).
1408                  * The task itself is dead and won't touch exit_code again;
1409                  * other processors in this function are locked out.
1410                  */
1411                 p->exit_code = exit_code;
1412                 exit_code = 0;
1413         }
1414         if (unlikely(exit_code == 0)) {
1415                 /*
1416                  * Another thread in this function got to it first, or it
1417                  * resumed, or it resumed and then died.
1418                  */
1419                 write_unlock_irq(&tasklist_lock);
1420 bail_ref:
1421                 put_task_struct(p);
1422                 /*
1423                  * We are returning to the wait loop without having successfully
1424                  * removed the process and having released the lock. We cannot
1425                  * continue, since the "p" task pointer is potentially stale.
1426                  *
1427                  * Return -EAGAIN, and do_wait() will restart the loop from the
1428                  * beginning. Do _not_ re-acquire the lock.
1429                  */
1430                 return -EAGAIN;
1431         }
1432
1433         /* move to end of parent's list to avoid starvation */
1434         remove_parent(p);
1435         add_parent(p);
1436
1437         write_unlock_irq(&tasklist_lock);
1438
1439         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1440         if (!retval && stat_addr)
1441                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1442         if (!retval && infop)
1443                 retval = put_user(SIGCHLD, &infop->si_signo);
1444         if (!retval && infop)
1445                 retval = put_user(0, &infop->si_errno);
1446         if (!retval && infop)
1447                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1448                                           ? CLD_TRAPPED : CLD_STOPPED),
1449                                   &infop->si_code);
1450         if (!retval && infop)
1451                 retval = put_user(exit_code, &infop->si_status);
1452         if (!retval && infop)
1453                 retval = put_user(pid, &infop->si_pid);
1454         if (!retval && infop)
1455                 retval = put_user(p->uid, &infop->si_uid);
1456         if (!retval)
1457                 retval = pid;
1458         put_task_struct(p);
1459
1460         BUG_ON(!retval);
1461         return retval;
1462 }
1463
1464 /*
1465  * Handle do_wait work for one task in a live, non-stopped state.
1466  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1467  * the lock and this task is uninteresting.  If we return nonzero, we have
1468  * released the lock and the system call should return.
1469  */
1470 static int wait_task_continued(struct task_struct *p, int noreap,
1471                                struct siginfo __user *infop,
1472                                int __user *stat_addr, struct rusage __user *ru)
1473 {
1474         int retval;
1475         pid_t pid;
1476         uid_t uid;
1477         struct pid_namespace *ns;
1478
1479         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1480                 return 0;
1481
1482         spin_lock_irq(&p->sighand->siglock);
1483         /* Re-check with the lock held.  */
1484         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1485                 spin_unlock_irq(&p->sighand->siglock);
1486                 return 0;
1487         }
1488         if (!noreap)
1489                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1490         spin_unlock_irq(&p->sighand->siglock);
1491
1492         ns = current->nsproxy->pid_ns;
1493         pid = task_pid_nr_ns(p, ns);
1494         uid = p->uid;
1495         get_task_struct(p);
1496         read_unlock(&tasklist_lock);
1497
1498         if (!infop) {
1499                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1500                 put_task_struct(p);
1501                 if (!retval && stat_addr)
1502                         retval = put_user(0xffff, stat_addr);
1503                 if (!retval)
1504                         retval = task_pid_nr_ns(p, ns);
1505         } else {
1506                 retval = wait_noreap_copyout(p, pid, uid,
1507                                              CLD_CONTINUED, SIGCONT,
1508                                              infop, ru);
1509                 BUG_ON(retval == 0);
1510         }
1511
1512         return retval;
1513 }
1514
1515
1516 static inline int my_ptrace_child(struct task_struct *p)
1517 {
1518         if (!(p->ptrace & PT_PTRACED))
1519                 return 0;
1520         if (!(p->ptrace & PT_ATTACHED))
1521                 return 1;
1522         /*
1523          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1524          * we are the attacher.  If we are the real parent, this is a race
1525          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1526          * which we have to switch the parent links, but has already set
1527          * the flags in p->ptrace.
1528          */
1529         return (p->parent != p->real_parent);
1530 }
1531
1532 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1533                     int __user *stat_addr, struct rusage __user *ru)
1534 {
1535         DECLARE_WAITQUEUE(wait, current);
1536         struct task_struct *tsk;
1537         int flag, retval;
1538         int allowed, denied;
1539
1540         add_wait_queue(&current->signal->wait_chldexit,&wait);
1541 repeat:
1542         /*
1543          * We will set this flag if we see any child that might later
1544          * match our criteria, even if we are not able to reap it yet.
1545          */
1546         flag = 0;
1547         allowed = denied = 0;
1548         current->state = TASK_INTERRUPTIBLE;
1549         read_lock(&tasklist_lock);
1550         tsk = current;
1551         do {
1552                 struct task_struct *p;
1553                 int ret;
1554
1555                 list_for_each_entry(p, &tsk->children, sibling) {
1556                         ret = eligible_child(pid, options, p);
1557                         if (!ret)
1558                                 continue;
1559
1560                         if (unlikely(ret < 0)) {
1561                                 denied = ret;
1562                                 continue;
1563                         }
1564                         allowed = 1;
1565
1566                         switch (p->state) {
1567                         case TASK_TRACED:
1568                                 /*
1569                                  * When we hit the race with PTRACE_ATTACH,
1570                                  * we will not report this child.  But the
1571                                  * race means it has not yet been moved to
1572                                  * our ptrace_children list, so we need to
1573                                  * set the flag here to avoid a spurious ECHILD
1574                                  * when the race happens with the only child.
1575                                  */
1576                                 flag = 1;
1577                                 if (!my_ptrace_child(p))
1578                                         continue;
1579                                 /*FALLTHROUGH*/
1580                         case TASK_STOPPED:
1581                                 /*
1582                                  * It's stopped now, so it might later
1583                                  * continue, exit, or stop again.
1584                                  */
1585                                 flag = 1;
1586                                 if (!(options & WUNTRACED) &&
1587                                     !my_ptrace_child(p))
1588                                         continue;
1589                                 retval = wait_task_stopped(p, ret == 2,
1590                                                            (options & WNOWAIT),
1591                                                            infop,
1592                                                            stat_addr, ru);
1593                                 if (retval == -EAGAIN)
1594                                         goto repeat;
1595                                 if (retval != 0) /* He released the lock.  */
1596                                         goto end;
1597                                 break;
1598                         default:
1599                         // case EXIT_DEAD:
1600                                 if (p->exit_state == EXIT_DEAD)
1601                                         continue;
1602                         // case EXIT_ZOMBIE:
1603                                 if (p->exit_state == EXIT_ZOMBIE) {
1604                                         /*
1605                                          * Eligible but we cannot release
1606                                          * it yet:
1607                                          */
1608                                         if (ret == 2)
1609                                                 goto check_continued;
1610                                         if (!likely(options & WEXITED))
1611                                                 continue;
1612                                         retval = wait_task_zombie(
1613                                                 p, (options & WNOWAIT),
1614                                                 infop, stat_addr, ru);
1615                                         /* He released the lock.  */
1616                                         if (retval != 0)
1617                                                 goto end;
1618                                         break;
1619                                 }
1620 check_continued:
1621                                 /*
1622                                  * It's running now, so it might later
1623                                  * exit, stop, or stop and then continue.
1624                                  */
1625                                 flag = 1;
1626                                 if (!unlikely(options & WCONTINUED))
1627                                         continue;
1628                                 retval = wait_task_continued(
1629                                         p, (options & WNOWAIT),
1630                                         infop, stat_addr, ru);
1631                                 if (retval != 0) /* He released the lock.  */
1632                                         goto end;
1633                                 break;
1634                         }
1635                 }
1636                 if (!flag) {
1637                         list_for_each_entry(p, &tsk->ptrace_children,
1638                                             ptrace_list) {
1639                                 if (!eligible_child(pid, options, p))
1640                                         continue;
1641                                 flag = 1;
1642                                 break;
1643                         }
1644                 }
1645                 if (options & __WNOTHREAD)
1646                         break;
1647                 tsk = next_thread(tsk);
1648                 BUG_ON(tsk->signal != current->signal);
1649         } while (tsk != current);
1650
1651         read_unlock(&tasklist_lock);
1652         if (flag) {
1653                 retval = 0;
1654                 if (options & WNOHANG)
1655                         goto end;
1656                 retval = -ERESTARTSYS;
1657                 if (signal_pending(current))
1658                         goto end;
1659                 schedule();
1660                 goto repeat;
1661         }
1662         retval = -ECHILD;
1663         if (unlikely(denied) && !allowed)
1664                 retval = denied;
1665 end:
1666         current->state = TASK_RUNNING;
1667         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1668         if (infop) {
1669                 if (retval > 0)
1670                 retval = 0;
1671                 else {
1672                         /*
1673                          * For a WNOHANG return, clear out all the fields
1674                          * we would set so the user can easily tell the
1675                          * difference.
1676                          */
1677                         if (!retval)
1678                                 retval = put_user(0, &infop->si_signo);
1679                         if (!retval)
1680                                 retval = put_user(0, &infop->si_errno);
1681                         if (!retval)
1682                                 retval = put_user(0, &infop->si_code);
1683                         if (!retval)
1684                                 retval = put_user(0, &infop->si_pid);
1685                         if (!retval)
1686                                 retval = put_user(0, &infop->si_uid);
1687                         if (!retval)
1688                                 retval = put_user(0, &infop->si_status);
1689                 }
1690         }
1691         return retval;
1692 }
1693
1694 asmlinkage long sys_waitid(int which, pid_t pid,
1695                            struct siginfo __user *infop, int options,
1696                            struct rusage __user *ru)
1697 {
1698         long ret;
1699
1700         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1701                 return -EINVAL;
1702         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1703                 return -EINVAL;
1704
1705         switch (which) {
1706         case P_ALL:
1707                 pid = -1;
1708                 break;
1709         case P_PID:
1710                 if (pid <= 0)
1711                         return -EINVAL;
1712                 break;
1713         case P_PGID:
1714                 if (pid <= 0)
1715                         return -EINVAL;
1716                 pid = -pid;
1717                 break;
1718         default:
1719                 return -EINVAL;
1720         }
1721
1722         ret = do_wait(pid, options, infop, NULL, ru);
1723
1724         /* avoid REGPARM breakage on x86: */
1725         prevent_tail_call(ret);
1726         return ret;
1727 }
1728
1729 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1730                           int options, struct rusage __user *ru)
1731 {
1732         long ret;
1733
1734         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1735                         __WNOTHREAD|__WCLONE|__WALL))
1736                 return -EINVAL;
1737         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1738
1739         /* avoid REGPARM breakage on x86: */
1740         prevent_tail_call(ret);
1741         return ret;
1742 }
1743
1744 #ifdef __ARCH_WANT_SYS_WAITPID
1745
1746 /*
1747  * sys_waitpid() remains for compatibility. waitpid() should be
1748  * implemented by calling sys_wait4() from libc.a.
1749  */
1750 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1751 {
1752         return sys_wait4(pid, stat_addr, options, NULL);
1753 }
1754
1755 #endif