Memory controller: OOM handling
[linux-2.6.git] / kernel / exit.c
1 /*
2  *  linux/kernel/exit.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 #include <linux/mm.h>
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/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 static void exit_mm(struct task_struct * tsk);
54
55 static void __unhash_process(struct task_struct *p)
56 {
57         nr_threads--;
58         detach_pid(p, PIDTYPE_PID);
59         if (thread_group_leader(p)) {
60                 detach_pid(p, PIDTYPE_PGID);
61                 detach_pid(p, PIDTYPE_SID);
62
63                 list_del_rcu(&p->tasks);
64                 __get_cpu_var(process_counts)--;
65         }
66         list_del_rcu(&p->thread_group);
67         remove_parent(p);
68 }
69
70 /*
71  * This function expects the tasklist_lock write-locked.
72  */
73 static void __exit_signal(struct task_struct *tsk)
74 {
75         struct signal_struct *sig = tsk->signal;
76         struct sighand_struct *sighand;
77
78         BUG_ON(!sig);
79         BUG_ON(!atomic_read(&sig->count));
80
81         rcu_read_lock();
82         sighand = rcu_dereference(tsk->sighand);
83         spin_lock(&sighand->siglock);
84
85         posix_cpu_timers_exit(tsk);
86         if (atomic_dec_and_test(&sig->count))
87                 posix_cpu_timers_exit_group(tsk);
88         else {
89                 /*
90                  * If there is any task waiting for the group exit
91                  * then notify it:
92                  */
93                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
94                         wake_up_process(sig->group_exit_task);
95
96                 if (tsk == sig->curr_target)
97                         sig->curr_target = next_thread(tsk);
98                 /*
99                  * Accumulate here the counters for all threads but the
100                  * group leader as they die, so they can be added into
101                  * the process-wide totals when those are taken.
102                  * The group leader stays around as a zombie as long
103                  * as there are other threads.  When it gets reaped,
104                  * the exit.c code will add its counts into these totals.
105                  * We won't ever get here for the group leader, since it
106                  * will have been the last reference on the signal_struct.
107                  */
108                 sig->utime = cputime_add(sig->utime, tsk->utime);
109                 sig->stime = cputime_add(sig->stime, tsk->stime);
110                 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
111                 sig->min_flt += tsk->min_flt;
112                 sig->maj_flt += tsk->maj_flt;
113                 sig->nvcsw += tsk->nvcsw;
114                 sig->nivcsw += tsk->nivcsw;
115                 sig->inblock += task_io_get_inblock(tsk);
116                 sig->oublock += task_io_get_oublock(tsk);
117                 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
118                 sig = NULL; /* Marker for below. */
119         }
120
121         __unhash_process(tsk);
122
123         tsk->signal = NULL;
124         tsk->sighand = NULL;
125         spin_unlock(&sighand->siglock);
126         rcu_read_unlock();
127
128         __cleanup_sighand(sighand);
129         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
130         flush_sigqueue(&tsk->pending);
131         if (sig) {
132                 flush_sigqueue(&sig->shared_pending);
133                 taskstats_tgid_free(sig);
134                 __cleanup_signal(sig);
135         }
136 }
137
138 static void delayed_put_task_struct(struct rcu_head *rhp)
139 {
140         put_task_struct(container_of(rhp, struct task_struct, rcu));
141 }
142
143 void release_task(struct task_struct * p)
144 {
145         struct task_struct *leader;
146         int zap_leader;
147 repeat:
148         atomic_dec(&p->user->processes);
149         proc_flush_task(p);
150         write_lock_irq(&tasklist_lock);
151         ptrace_unlink(p);
152         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
153         __exit_signal(p);
154
155         /*
156          * If we are the last non-leader member of the thread
157          * group, and the leader is zombie, then notify the
158          * group leader's parent process. (if it wants notification.)
159          */
160         zap_leader = 0;
161         leader = p->group_leader;
162         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
163                 BUG_ON(leader->exit_signal == -1);
164                 do_notify_parent(leader, leader->exit_signal);
165                 /*
166                  * If we were the last child thread and the leader has
167                  * exited already, and the leader's parent ignores SIGCHLD,
168                  * then we are the one who should release the leader.
169                  *
170                  * do_notify_parent() will have marked it self-reaping in
171                  * that case.
172                  */
173                 zap_leader = (leader->exit_signal == -1);
174         }
175
176         write_unlock_irq(&tasklist_lock);
177         release_thread(p);
178         call_rcu(&p->rcu, delayed_put_task_struct);
179
180         p = leader;
181         if (unlikely(zap_leader))
182                 goto repeat;
183 }
184
185 /*
186  * This checks not only the pgrp, but falls back on the pid if no
187  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
188  * without this...
189  *
190  * The caller must hold rcu lock or the tasklist lock.
191  */
192 struct pid *session_of_pgrp(struct pid *pgrp)
193 {
194         struct task_struct *p;
195         struct pid *sid = NULL;
196
197         p = pid_task(pgrp, PIDTYPE_PGID);
198         if (p == NULL)
199                 p = pid_task(pgrp, PIDTYPE_PID);
200         if (p != NULL)
201                 sid = task_session(p);
202
203         return sid;
204 }
205
206 /*
207  * Determine if a process group is "orphaned", according to the POSIX
208  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
209  * by terminal-generated stop signals.  Newly orphaned process groups are
210  * to receive a SIGHUP and a SIGCONT.
211  *
212  * "I ask you, have you ever known what it is to be an orphan?"
213  */
214 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
215 {
216         struct task_struct *p;
217         int ret = 1;
218
219         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
220                 if (p == ignored_task
221                                 || p->exit_state
222                                 || is_global_init(p->real_parent))
223                         continue;
224                 if (task_pgrp(p->real_parent) != pgrp &&
225                     task_session(p->real_parent) == task_session(p)) {
226                         ret = 0;
227                         break;
228                 }
229         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
230         return ret;     /* (sighing) "Often!" */
231 }
232
233 int is_current_pgrp_orphaned(void)
234 {
235         int retval;
236
237         read_lock(&tasklist_lock);
238         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
239         read_unlock(&tasklist_lock);
240
241         return retval;
242 }
243
244 static int has_stopped_jobs(struct pid *pgrp)
245 {
246         int retval = 0;
247         struct task_struct *p;
248
249         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
250                 if (!task_is_stopped(p))
251                         continue;
252                 retval = 1;
253                 break;
254         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
255         return retval;
256 }
257
258 /**
259  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
260  *
261  * If a kernel thread is launched as a result of a system call, or if
262  * it ever exits, it should generally reparent itself to kthreadd so it
263  * isn't in the way of other processes and is correctly cleaned up on exit.
264  *
265  * The various task state such as scheduling policy and priority may have
266  * been inherited from a user process, so we reset them to sane values here.
267  *
268  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
269  */
270 static void reparent_to_kthreadd(void)
271 {
272         write_lock_irq(&tasklist_lock);
273
274         ptrace_unlink(current);
275         /* Reparent to init */
276         remove_parent(current);
277         current->real_parent = current->parent = kthreadd_task;
278         add_parent(current);
279
280         /* Set the exit signal to SIGCHLD so we signal init on exit */
281         current->exit_signal = SIGCHLD;
282
283         if (task_nice(current) < 0)
284                 set_user_nice(current, 0);
285         /* cpus_allowed? */
286         /* rt_priority? */
287         /* signals? */
288         security_task_reparent_to_init(current);
289         memcpy(current->signal->rlim, init_task.signal->rlim,
290                sizeof(current->signal->rlim));
291         atomic_inc(&(INIT_USER->__count));
292         write_unlock_irq(&tasklist_lock);
293         switch_uid(INIT_USER);
294 }
295
296 void __set_special_pids(pid_t session, pid_t pgrp)
297 {
298         struct task_struct *curr = current->group_leader;
299
300         if (task_session_nr(curr) != session) {
301                 detach_pid(curr, PIDTYPE_SID);
302                 set_task_session(curr, session);
303                 attach_pid(curr, PIDTYPE_SID, find_pid(session));
304         }
305         if (task_pgrp_nr(curr) != pgrp) {
306                 detach_pid(curr, PIDTYPE_PGID);
307                 set_task_pgrp(curr, pgrp);
308                 attach_pid(curr, PIDTYPE_PGID, find_pid(pgrp));
309         }
310 }
311
312 static void set_special_pids(pid_t session, pid_t pgrp)
313 {
314         write_lock_irq(&tasklist_lock);
315         __set_special_pids(session, pgrp);
316         write_unlock_irq(&tasklist_lock);
317 }
318
319 /*
320  * Let kernel threads use this to say that they
321  * allow a certain signal (since daemonize() will
322  * have disabled all of them by default).
323  */
324 int allow_signal(int sig)
325 {
326         if (!valid_signal(sig) || sig < 1)
327                 return -EINVAL;
328
329         spin_lock_irq(&current->sighand->siglock);
330         sigdelset(&current->blocked, sig);
331         if (!current->mm) {
332                 /* Kernel threads handle their own signals.
333                    Let the signal code know it'll be handled, so
334                    that they don't get converted to SIGKILL or
335                    just silently dropped */
336                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
337         }
338         recalc_sigpending();
339         spin_unlock_irq(&current->sighand->siglock);
340         return 0;
341 }
342
343 EXPORT_SYMBOL(allow_signal);
344
345 int disallow_signal(int sig)
346 {
347         if (!valid_signal(sig) || sig < 1)
348                 return -EINVAL;
349
350         spin_lock_irq(&current->sighand->siglock);
351         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
352         recalc_sigpending();
353         spin_unlock_irq(&current->sighand->siglock);
354         return 0;
355 }
356
357 EXPORT_SYMBOL(disallow_signal);
358
359 /*
360  *      Put all the gunge required to become a kernel thread without
361  *      attached user resources in one place where it belongs.
362  */
363
364 void daemonize(const char *name, ...)
365 {
366         va_list args;
367         struct fs_struct *fs;
368         sigset_t blocked;
369
370         va_start(args, name);
371         vsnprintf(current->comm, sizeof(current->comm), name, args);
372         va_end(args);
373
374         /*
375          * If we were started as result of loading a module, close all of the
376          * user space pages.  We don't need them, and if we didn't close them
377          * they would be locked into memory.
378          */
379         exit_mm(current);
380         /*
381          * We don't want to have TIF_FREEZE set if the system-wide hibernation
382          * or suspend transition begins right now.
383          */
384         current->flags |= PF_NOFREEZE;
385
386         set_special_pids(1, 1);
387         proc_clear_tty(current);
388
389         /* Block and flush all signals */
390         sigfillset(&blocked);
391         sigprocmask(SIG_BLOCK, &blocked, NULL);
392         flush_signals(current);
393
394         /* Become as one with the init task */
395
396         exit_fs(current);       /* current->fs->count--; */
397         fs = init_task.fs;
398         current->fs = fs;
399         atomic_inc(&fs->count);
400
401         if (current->nsproxy != init_task.nsproxy) {
402                 get_nsproxy(init_task.nsproxy);
403                 switch_task_namespaces(current, init_task.nsproxy);
404         }
405
406         exit_files(current);
407         current->files = init_task.files;
408         atomic_inc(&current->files->count);
409
410         reparent_to_kthreadd();
411 }
412
413 EXPORT_SYMBOL(daemonize);
414
415 static void close_files(struct files_struct * files)
416 {
417         int i, j;
418         struct fdtable *fdt;
419
420         j = 0;
421
422         /*
423          * It is safe to dereference the fd table without RCU or
424          * ->file_lock because this is the last reference to the
425          * files structure.
426          */
427         fdt = files_fdtable(files);
428         for (;;) {
429                 unsigned long set;
430                 i = j * __NFDBITS;
431                 if (i >= fdt->max_fds)
432                         break;
433                 set = fdt->open_fds->fds_bits[j++];
434                 while (set) {
435                         if (set & 1) {
436                                 struct file * file = xchg(&fdt->fd[i], NULL);
437                                 if (file) {
438                                         filp_close(file, files);
439                                         cond_resched();
440                                 }
441                         }
442                         i++;
443                         set >>= 1;
444                 }
445         }
446 }
447
448 struct files_struct *get_files_struct(struct task_struct *task)
449 {
450         struct files_struct *files;
451
452         task_lock(task);
453         files = task->files;
454         if (files)
455                 atomic_inc(&files->count);
456         task_unlock(task);
457
458         return files;
459 }
460
461 void fastcall put_files_struct(struct files_struct *files)
462 {
463         struct fdtable *fdt;
464
465         if (atomic_dec_and_test(&files->count)) {
466                 close_files(files);
467                 /*
468                  * Free the fd and fdset arrays if we expanded them.
469                  * If the fdtable was embedded, pass files for freeing
470                  * at the end of the RCU grace period. Otherwise,
471                  * you can free files immediately.
472                  */
473                 fdt = files_fdtable(files);
474                 if (fdt != &files->fdtab)
475                         kmem_cache_free(files_cachep, files);
476                 free_fdtable(fdt);
477         }
478 }
479
480 EXPORT_SYMBOL(put_files_struct);
481
482 void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
483 {
484         struct files_struct *old;
485
486         old = tsk->files;
487         task_lock(tsk);
488         tsk->files = files;
489         task_unlock(tsk);
490         put_files_struct(old);
491 }
492 EXPORT_SYMBOL(reset_files_struct);
493
494 static void __exit_files(struct task_struct *tsk)
495 {
496         struct files_struct * files = tsk->files;
497
498         if (files) {
499                 task_lock(tsk);
500                 tsk->files = NULL;
501                 task_unlock(tsk);
502                 put_files_struct(files);
503         }
504 }
505
506 void exit_files(struct task_struct *tsk)
507 {
508         __exit_files(tsk);
509 }
510
511 static void __put_fs_struct(struct fs_struct *fs)
512 {
513         /* No need to hold fs->lock if we are killing it */
514         if (atomic_dec_and_test(&fs->count)) {
515                 dput(fs->root);
516                 mntput(fs->rootmnt);
517                 dput(fs->pwd);
518                 mntput(fs->pwdmnt);
519                 if (fs->altroot) {
520                         dput(fs->altroot);
521                         mntput(fs->altrootmnt);
522                 }
523                 kmem_cache_free(fs_cachep, fs);
524         }
525 }
526
527 void put_fs_struct(struct fs_struct *fs)
528 {
529         __put_fs_struct(fs);
530 }
531
532 static void __exit_fs(struct task_struct *tsk)
533 {
534         struct fs_struct * fs = tsk->fs;
535
536         if (fs) {
537                 task_lock(tsk);
538                 tsk->fs = NULL;
539                 task_unlock(tsk);
540                 __put_fs_struct(fs);
541         }
542 }
543
544 void exit_fs(struct task_struct *tsk)
545 {
546         __exit_fs(tsk);
547 }
548
549 EXPORT_SYMBOL_GPL(exit_fs);
550
551 /*
552  * Turn us into a lazy TLB process if we
553  * aren't already..
554  */
555 static void exit_mm(struct task_struct * tsk)
556 {
557         struct mm_struct *mm = tsk->mm;
558
559         mm_release(tsk, mm);
560         if (!mm)
561                 return;
562         /*
563          * Serialize with any possible pending coredump.
564          * We must hold mmap_sem around checking core_waiters
565          * and clearing tsk->mm.  The core-inducing thread
566          * will increment core_waiters for each thread in the
567          * group with ->mm != NULL.
568          */
569         down_read(&mm->mmap_sem);
570         if (mm->core_waiters) {
571                 up_read(&mm->mmap_sem);
572                 down_write(&mm->mmap_sem);
573                 if (!--mm->core_waiters)
574                         complete(mm->core_startup_done);
575                 up_write(&mm->mmap_sem);
576
577                 wait_for_completion(&mm->core_done);
578                 down_read(&mm->mmap_sem);
579         }
580         atomic_inc(&mm->mm_count);
581         BUG_ON(mm != tsk->active_mm);
582         /* more a memory barrier than a real lock */
583         task_lock(tsk);
584         tsk->mm = NULL;
585         up_read(&mm->mmap_sem);
586         enter_lazy_tlb(mm, current);
587         /* We don't want this task to be frozen prematurely */
588         clear_freeze_flag(tsk);
589         task_unlock(tsk);
590         mmput(mm);
591 }
592
593 static void
594 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
595 {
596         if (p->pdeath_signal)
597                 /* We already hold the tasklist_lock here.  */
598                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
599
600         /* Move the child from its dying parent to the new one.  */
601         if (unlikely(traced)) {
602                 /* Preserve ptrace links if someone else is tracing this child.  */
603                 list_del_init(&p->ptrace_list);
604                 if (p->parent != p->real_parent)
605                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
606         } else {
607                 /* If this child is being traced, then we're the one tracing it
608                  * anyway, so let go of it.
609                  */
610                 p->ptrace = 0;
611                 remove_parent(p);
612                 p->parent = p->real_parent;
613                 add_parent(p);
614
615                 if (task_is_traced(p)) {
616                         /*
617                          * If it was at a trace stop, turn it into
618                          * a normal stop since it's no longer being
619                          * traced.
620                          */
621                         ptrace_untrace(p);
622                 }
623         }
624
625         /* If this is a threaded reparent there is no need to
626          * notify anyone anything has happened.
627          */
628         if (p->real_parent->group_leader == father->group_leader)
629                 return;
630
631         /* We don't want people slaying init.  */
632         if (p->exit_signal != -1)
633                 p->exit_signal = SIGCHLD;
634
635         /* If we'd notified the old parent about this child's death,
636          * also notify the new parent.
637          */
638         if (!traced && p->exit_state == EXIT_ZOMBIE &&
639             p->exit_signal != -1 && thread_group_empty(p))
640                 do_notify_parent(p, p->exit_signal);
641
642         /*
643          * process group orphan check
644          * Case ii: Our child is in a different pgrp
645          * than we are, and it was the only connection
646          * outside, so the child pgrp is now orphaned.
647          */
648         if ((task_pgrp(p) != task_pgrp(father)) &&
649             (task_session(p) == task_session(father))) {
650                 struct pid *pgrp = task_pgrp(p);
651
652                 if (will_become_orphaned_pgrp(pgrp, NULL) &&
653                     has_stopped_jobs(pgrp)) {
654                         __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
655                         __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
656                 }
657         }
658 }
659
660 /*
661  * When we die, we re-parent all our children.
662  * Try to give them to another thread in our thread
663  * group, and if no such member exists, give it to
664  * the child reaper process (ie "init") in our pid
665  * space.
666  */
667 static void forget_original_parent(struct task_struct *father)
668 {
669         struct task_struct *p, *n, *reaper = father;
670         struct list_head ptrace_dead;
671
672         INIT_LIST_HEAD(&ptrace_dead);
673
674         write_lock_irq(&tasklist_lock);
675
676         do {
677                 reaper = next_thread(reaper);
678                 if (reaper == father) {
679                         reaper = task_child_reaper(father);
680                         break;
681                 }
682         } while (reaper->flags & PF_EXITING);
683
684         /*
685          * There are only two places where our children can be:
686          *
687          * - in our child list
688          * - in our ptraced child list
689          *
690          * Search them and reparent children.
691          */
692         list_for_each_entry_safe(p, n, &father->children, sibling) {
693                 int ptrace;
694
695                 ptrace = p->ptrace;
696
697                 /* if father isn't the real parent, then ptrace must be enabled */
698                 BUG_ON(father != p->real_parent && !ptrace);
699
700                 if (father == p->real_parent) {
701                         /* reparent with a reaper, real father it's us */
702                         p->real_parent = reaper;
703                         reparent_thread(p, father, 0);
704                 } else {
705                         /* reparent ptraced task to its real parent */
706                         __ptrace_unlink (p);
707                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
708                             thread_group_empty(p))
709                                 do_notify_parent(p, p->exit_signal);
710                 }
711
712                 /*
713                  * if the ptraced child is a zombie with exit_signal == -1
714                  * we must collect it before we exit, or it will remain
715                  * zombie forever since we prevented it from self-reap itself
716                  * while it was being traced by us, to be able to see it in wait4.
717                  */
718                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
719                         list_add(&p->ptrace_list, &ptrace_dead);
720         }
721
722         list_for_each_entry_safe(p, n, &father->ptrace_children, ptrace_list) {
723                 p->real_parent = reaper;
724                 reparent_thread(p, father, 1);
725         }
726
727         write_unlock_irq(&tasklist_lock);
728         BUG_ON(!list_empty(&father->children));
729         BUG_ON(!list_empty(&father->ptrace_children));
730
731         list_for_each_entry_safe(p, n, &ptrace_dead, ptrace_list) {
732                 list_del_init(&p->ptrace_list);
733                 release_task(p);
734         }
735
736 }
737
738 /*
739  * Send signals to all our closest relatives so that they know
740  * to properly mourn us..
741  */
742 static void exit_notify(struct task_struct *tsk)
743 {
744         int state;
745         struct task_struct *t;
746         struct pid *pgrp;
747
748         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
749             && !thread_group_empty(tsk)) {
750                 /*
751                  * This occurs when there was a race between our exit
752                  * syscall and a group signal choosing us as the one to
753                  * wake up.  It could be that we are the only thread
754                  * alerted to check for pending signals, but another thread
755                  * should be woken now to take the signal since we will not.
756                  * Now we'll wake all the threads in the group just to make
757                  * sure someone gets all the pending signals.
758                  */
759                 spin_lock_irq(&tsk->sighand->siglock);
760                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
761                         if (!signal_pending(t) && !(t->flags & PF_EXITING))
762                                 recalc_sigpending_and_wake(t);
763                 spin_unlock_irq(&tsk->sighand->siglock);
764         }
765
766         /*
767          * This does two things:
768          *
769          * A.  Make init inherit all the child processes
770          * B.  Check to see if any process groups have become orphaned
771          *      as a result of our exiting, and if they have any stopped
772          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
773          */
774         forget_original_parent(tsk);
775         exit_task_namespaces(tsk);
776
777         write_lock_irq(&tasklist_lock);
778         /*
779          * Check to see if any process groups have become orphaned
780          * as a result of our exiting, and if they have any stopped
781          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
782          *
783          * Case i: Our father is in a different pgrp than we are
784          * and we were the only connection outside, so our pgrp
785          * is about to become orphaned.
786          */
787         t = tsk->real_parent;
788
789         pgrp = task_pgrp(tsk);
790         if ((task_pgrp(t) != pgrp) &&
791             (task_session(t) == task_session(tsk)) &&
792             will_become_orphaned_pgrp(pgrp, tsk) &&
793             has_stopped_jobs(pgrp)) {
794                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
795                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
796         }
797
798         /* Let father know we died
799          *
800          * Thread signals are configurable, but you aren't going to use
801          * that to send signals to arbitary processes.
802          * That stops right now.
803          *
804          * If the parent exec id doesn't match the exec id we saved
805          * when we started then we know the parent has changed security
806          * domain.
807          *
808          * If our self_exec id doesn't match our parent_exec_id then
809          * we have changed execution domain as these two values started
810          * the same after a fork.
811          */
812         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
813             ( tsk->parent_exec_id != t->self_exec_id  ||
814               tsk->self_exec_id != tsk->parent_exec_id)
815             && !capable(CAP_KILL))
816                 tsk->exit_signal = SIGCHLD;
817
818
819         /* If something other than our normal parent is ptracing us, then
820          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
821          * only has special meaning to our real parent.
822          */
823         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
824                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
825                 do_notify_parent(tsk, signal);
826         } else if (tsk->ptrace) {
827                 do_notify_parent(tsk, SIGCHLD);
828         }
829
830         state = EXIT_ZOMBIE;
831         if (tsk->exit_signal == -1 && likely(!tsk->ptrace))
832                 state = EXIT_DEAD;
833         tsk->exit_state = state;
834
835         if (thread_group_leader(tsk) &&
836             tsk->signal->notify_count < 0 &&
837             tsk->signal->group_exit_task)
838                 wake_up_process(tsk->signal->group_exit_task);
839
840         write_unlock_irq(&tasklist_lock);
841
842         /* If the process is dead, release it - nobody will wait for it */
843         if (state == EXIT_DEAD)
844                 release_task(tsk);
845 }
846
847 #ifdef CONFIG_DEBUG_STACK_USAGE
848 static void check_stack_usage(void)
849 {
850         static DEFINE_SPINLOCK(low_water_lock);
851         static int lowest_to_date = THREAD_SIZE;
852         unsigned long *n = end_of_stack(current);
853         unsigned long free;
854
855         while (*n == 0)
856                 n++;
857         free = (unsigned long)n - (unsigned long)end_of_stack(current);
858
859         if (free >= lowest_to_date)
860                 return;
861
862         spin_lock(&low_water_lock);
863         if (free < lowest_to_date) {
864                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
865                                 "left\n",
866                                 current->comm, free);
867                 lowest_to_date = free;
868         }
869         spin_unlock(&low_water_lock);
870 }
871 #else
872 static inline void check_stack_usage(void) {}
873 #endif
874
875 static inline void exit_child_reaper(struct task_struct *tsk)
876 {
877         if (likely(tsk->group_leader != task_child_reaper(tsk)))
878                 return;
879
880         if (tsk->nsproxy->pid_ns == &init_pid_ns)
881                 panic("Attempted to kill init!");
882
883         /*
884          * @tsk is the last thread in the 'cgroup-init' and is exiting.
885          * Terminate all remaining processes in the namespace and reap them
886          * before exiting @tsk.
887          *
888          * Note that @tsk (last thread of cgroup-init) may not necessarily
889          * be the child-reaper (i.e main thread of cgroup-init) of the
890          * namespace i.e the child_reaper may have already exited.
891          *
892          * Even after a child_reaper exits, we let it inherit orphaned children,
893          * because, pid_ns->child_reaper remains valid as long as there is
894          * at least one living sub-thread in the cgroup init.
895
896          * This living sub-thread of the cgroup-init will be notified when
897          * a child inherited by the 'child-reaper' exits (do_notify_parent()
898          * uses __group_send_sig_info()). Further, when reaping child processes,
899          * do_wait() iterates over children of all living sub threads.
900
901          * i.e even though 'child_reaper' thread is listed as the parent of the
902          * orphaned children, any living sub-thread in the cgroup-init can
903          * perform the role of the child_reaper.
904          */
905         zap_pid_ns_processes(tsk->nsproxy->pid_ns);
906 }
907
908 fastcall NORET_TYPE void do_exit(long code)
909 {
910         struct task_struct *tsk = current;
911         int group_dead;
912
913         profile_task_exit(tsk);
914
915         WARN_ON(atomic_read(&tsk->fs_excl));
916
917         if (unlikely(in_interrupt()))
918                 panic("Aiee, killing interrupt handler!");
919         if (unlikely(!tsk->pid))
920                 panic("Attempted to kill the idle task!");
921
922         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
923                 current->ptrace_message = code;
924                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
925         }
926
927         /*
928          * We're taking recursive faults here in do_exit. Safest is to just
929          * leave this task alone and wait for reboot.
930          */
931         if (unlikely(tsk->flags & PF_EXITING)) {
932                 printk(KERN_ALERT
933                         "Fixing recursive fault but reboot is needed!\n");
934                 /*
935                  * We can do this unlocked here. The futex code uses
936                  * this flag just to verify whether the pi state
937                  * cleanup has been done or not. In the worst case it
938                  * loops once more. We pretend that the cleanup was
939                  * done as there is no way to return. Either the
940                  * OWNER_DIED bit is set by now or we push the blocked
941                  * task into the wait for ever nirwana as well.
942                  */
943                 tsk->flags |= PF_EXITPIDONE;
944                 if (tsk->io_context)
945                         exit_io_context();
946                 set_current_state(TASK_UNINTERRUPTIBLE);
947                 schedule();
948         }
949
950         tsk->flags |= PF_EXITING;
951         /*
952          * tsk->flags are checked in the futex code to protect against
953          * an exiting task cleaning up the robust pi futexes.
954          */
955         smp_mb();
956         spin_unlock_wait(&tsk->pi_lock);
957
958         if (unlikely(in_atomic()))
959                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
960                                 current->comm, task_pid_nr(current),
961                                 preempt_count());
962
963         acct_update_integrals(tsk);
964         if (tsk->mm) {
965                 update_hiwater_rss(tsk->mm);
966                 update_hiwater_vm(tsk->mm);
967         }
968         group_dead = atomic_dec_and_test(&tsk->signal->live);
969         if (group_dead) {
970                 exit_child_reaper(tsk);
971                 hrtimer_cancel(&tsk->signal->real_timer);
972                 exit_itimers(tsk->signal);
973         }
974         acct_collect(code, group_dead);
975 #ifdef CONFIG_FUTEX
976         if (unlikely(tsk->robust_list))
977                 exit_robust_list(tsk);
978 #ifdef CONFIG_COMPAT
979         if (unlikely(tsk->compat_robust_list))
980                 compat_exit_robust_list(tsk);
981 #endif
982 #endif
983         if (group_dead)
984                 tty_audit_exit();
985         if (unlikely(tsk->audit_context))
986                 audit_free(tsk);
987
988         tsk->exit_code = code;
989         taskstats_exit(tsk, group_dead);
990
991         exit_mm(tsk);
992
993         if (group_dead)
994                 acct_process();
995         exit_sem(tsk);
996         __exit_files(tsk);
997         __exit_fs(tsk);
998         check_stack_usage();
999         exit_thread();
1000         cgroup_exit(tsk, 1);
1001         exit_keys(tsk);
1002
1003         if (group_dead && tsk->signal->leader)
1004                 disassociate_ctty(1);
1005
1006         module_put(task_thread_info(tsk)->exec_domain->module);
1007         if (tsk->binfmt)
1008                 module_put(tsk->binfmt->module);
1009
1010         proc_exit_connector(tsk);
1011         exit_notify(tsk);
1012 #ifdef CONFIG_NUMA
1013         mpol_free(tsk->mempolicy);
1014         tsk->mempolicy = NULL;
1015 #endif
1016 #ifdef CONFIG_FUTEX
1017         /*
1018          * This must happen late, after the PID is not
1019          * hashed anymore:
1020          */
1021         if (unlikely(!list_empty(&tsk->pi_state_list)))
1022                 exit_pi_state_list(tsk);
1023         if (unlikely(current->pi_state_cache))
1024                 kfree(current->pi_state_cache);
1025 #endif
1026         /*
1027          * Make sure we are holding no locks:
1028          */
1029         debug_check_no_locks_held(tsk);
1030         /*
1031          * We can do this unlocked here. The futex code uses this flag
1032          * just to verify whether the pi state cleanup has been done
1033          * or not. In the worst case it loops once more.
1034          */
1035         tsk->flags |= PF_EXITPIDONE;
1036
1037         if (tsk->io_context)
1038                 exit_io_context();
1039
1040         if (tsk->splice_pipe)
1041                 __free_pipe_info(tsk->splice_pipe);
1042
1043         preempt_disable();
1044         /* causes final put_task_struct in finish_task_switch(). */
1045         tsk->state = TASK_DEAD;
1046
1047         schedule();
1048         BUG();
1049         /* Avoid "noreturn function does return".  */
1050         for (;;)
1051                 cpu_relax();    /* For when BUG is null */
1052 }
1053
1054 EXPORT_SYMBOL_GPL(do_exit);
1055
1056 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1057 {
1058         if (comp)
1059                 complete(comp);
1060
1061         do_exit(code);
1062 }
1063
1064 EXPORT_SYMBOL(complete_and_exit);
1065
1066 asmlinkage long sys_exit(int error_code)
1067 {
1068         do_exit((error_code&0xff)<<8);
1069 }
1070
1071 /*
1072  * Take down every thread in the group.  This is called by fatal signals
1073  * as well as by sys_exit_group (below).
1074  */
1075 NORET_TYPE void
1076 do_group_exit(int exit_code)
1077 {
1078         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1079
1080         if (current->signal->flags & SIGNAL_GROUP_EXIT)
1081                 exit_code = current->signal->group_exit_code;
1082         else if (!thread_group_empty(current)) {
1083                 struct signal_struct *const sig = current->signal;
1084                 struct sighand_struct *const sighand = current->sighand;
1085                 spin_lock_irq(&sighand->siglock);
1086                 if (signal_group_exit(sig))
1087                         /* Another thread got here before we took the lock.  */
1088                         exit_code = sig->group_exit_code;
1089                 else {
1090                         sig->group_exit_code = exit_code;
1091                         sig->flags = SIGNAL_GROUP_EXIT;
1092                         zap_other_threads(current);
1093                 }
1094                 spin_unlock_irq(&sighand->siglock);
1095         }
1096
1097         do_exit(exit_code);
1098         /* NOTREACHED */
1099 }
1100
1101 /*
1102  * this kills every thread in the thread group. Note that any externally
1103  * wait4()-ing process will get the correct exit code - even if this
1104  * thread is not the thread group leader.
1105  */
1106 asmlinkage void sys_exit_group(int error_code)
1107 {
1108         do_group_exit((error_code & 0xff) << 8);
1109 }
1110
1111 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1112 {
1113         int err;
1114         struct pid_namespace *ns;
1115
1116         ns = current->nsproxy->pid_ns;
1117         if (pid > 0) {
1118                 if (task_pid_nr_ns(p, ns) != pid)
1119                         return 0;
1120         } else if (!pid) {
1121                 if (task_pgrp_nr_ns(p, ns) != task_pgrp_vnr(current))
1122                         return 0;
1123         } else if (pid != -1) {
1124                 if (task_pgrp_nr_ns(p, ns) != -pid)
1125                         return 0;
1126         }
1127
1128         /*
1129          * Do not consider detached threads that are
1130          * not ptraced:
1131          */
1132         if (p->exit_signal == -1 && !p->ptrace)
1133                 return 0;
1134
1135         /* Wait for all children (clone and not) if __WALL is set;
1136          * otherwise, wait for clone children *only* if __WCLONE is
1137          * set; otherwise, wait for non-clone children *only*.  (Note:
1138          * A "clone" child here is one that reports to its parent
1139          * using a signal other than SIGCHLD.) */
1140         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1141             && !(options & __WALL))
1142                 return 0;
1143         /*
1144          * Do not consider thread group leaders that are
1145          * in a non-empty thread group:
1146          */
1147         if (delay_group_leader(p))
1148                 return 2;
1149
1150         err = security_task_wait(p);
1151         if (err)
1152                 return err;
1153
1154         return 1;
1155 }
1156
1157 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1158                                int why, int status,
1159                                struct siginfo __user *infop,
1160                                struct rusage __user *rusagep)
1161 {
1162         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1163
1164         put_task_struct(p);
1165         if (!retval)
1166                 retval = put_user(SIGCHLD, &infop->si_signo);
1167         if (!retval)
1168                 retval = put_user(0, &infop->si_errno);
1169         if (!retval)
1170                 retval = put_user((short)why, &infop->si_code);
1171         if (!retval)
1172                 retval = put_user(pid, &infop->si_pid);
1173         if (!retval)
1174                 retval = put_user(uid, &infop->si_uid);
1175         if (!retval)
1176                 retval = put_user(status, &infop->si_status);
1177         if (!retval)
1178                 retval = pid;
1179         return retval;
1180 }
1181
1182 /*
1183  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1184  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1185  * the lock and this task is uninteresting.  If we return nonzero, we have
1186  * released the lock and the system call should return.
1187  */
1188 static int wait_task_zombie(struct task_struct *p, int noreap,
1189                             struct siginfo __user *infop,
1190                             int __user *stat_addr, struct rusage __user *ru)
1191 {
1192         unsigned long state;
1193         int retval, status, traced;
1194         struct pid_namespace *ns;
1195
1196         ns = current->nsproxy->pid_ns;
1197
1198         if (unlikely(noreap)) {
1199                 pid_t pid = task_pid_nr_ns(p, ns);
1200                 uid_t uid = p->uid;
1201                 int exit_code = p->exit_code;
1202                 int why, status;
1203
1204                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1205                         return 0;
1206                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1207                         return 0;
1208                 get_task_struct(p);
1209                 read_unlock(&tasklist_lock);
1210                 if ((exit_code & 0x7f) == 0) {
1211                         why = CLD_EXITED;
1212                         status = exit_code >> 8;
1213                 } else {
1214                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1215                         status = exit_code & 0x7f;
1216                 }
1217                 return wait_noreap_copyout(p, pid, uid, why,
1218                                            status, infop, ru);
1219         }
1220
1221         /*
1222          * Try to move the task's state to DEAD
1223          * only one thread is allowed to do this:
1224          */
1225         state = xchg(&p->exit_state, EXIT_DEAD);
1226         if (state != EXIT_ZOMBIE) {
1227                 BUG_ON(state != EXIT_DEAD);
1228                 return 0;
1229         }
1230
1231         /* traced means p->ptrace, but not vice versa */
1232         traced = (p->real_parent != p->parent);
1233
1234         if (likely(!traced)) {
1235                 struct signal_struct *psig;
1236                 struct signal_struct *sig;
1237
1238                 /*
1239                  * The resource counters for the group leader are in its
1240                  * own task_struct.  Those for dead threads in the group
1241                  * are in its signal_struct, as are those for the child
1242                  * processes it has previously reaped.  All these
1243                  * accumulate in the parent's signal_struct c* fields.
1244                  *
1245                  * We don't bother to take a lock here to protect these
1246                  * p->signal fields, because they are only touched by
1247                  * __exit_signal, which runs with tasklist_lock
1248                  * write-locked anyway, and so is excluded here.  We do
1249                  * need to protect the access to p->parent->signal fields,
1250                  * as other threads in the parent group can be right
1251                  * here reaping other children at the same time.
1252                  */
1253                 spin_lock_irq(&p->parent->sighand->siglock);
1254                 psig = p->parent->signal;
1255                 sig = p->signal;
1256                 psig->cutime =
1257                         cputime_add(psig->cutime,
1258                         cputime_add(p->utime,
1259                         cputime_add(sig->utime,
1260                                     sig->cutime)));
1261                 psig->cstime =
1262                         cputime_add(psig->cstime,
1263                         cputime_add(p->stime,
1264                         cputime_add(sig->stime,
1265                                     sig->cstime)));
1266                 psig->cgtime =
1267                         cputime_add(psig->cgtime,
1268                         cputime_add(p->gtime,
1269                         cputime_add(sig->gtime,
1270                                     sig->cgtime)));
1271                 psig->cmin_flt +=
1272                         p->min_flt + sig->min_flt + sig->cmin_flt;
1273                 psig->cmaj_flt +=
1274                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1275                 psig->cnvcsw +=
1276                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1277                 psig->cnivcsw +=
1278                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1279                 psig->cinblock +=
1280                         task_io_get_inblock(p) +
1281                         sig->inblock + sig->cinblock;
1282                 psig->coublock +=
1283                         task_io_get_oublock(p) +
1284                         sig->oublock + sig->coublock;
1285                 spin_unlock_irq(&p->parent->sighand->siglock);
1286         }
1287
1288         /*
1289          * Now we are sure this task is interesting, and no other
1290          * thread can reap it because we set its state to EXIT_DEAD.
1291          */
1292         read_unlock(&tasklist_lock);
1293
1294         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1295         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1296                 ? p->signal->group_exit_code : p->exit_code;
1297         if (!retval && stat_addr)
1298                 retval = put_user(status, stat_addr);
1299         if (!retval && infop)
1300                 retval = put_user(SIGCHLD, &infop->si_signo);
1301         if (!retval && infop)
1302                 retval = put_user(0, &infop->si_errno);
1303         if (!retval && infop) {
1304                 int why;
1305
1306                 if ((status & 0x7f) == 0) {
1307                         why = CLD_EXITED;
1308                         status >>= 8;
1309                 } else {
1310                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1311                         status &= 0x7f;
1312                 }
1313                 retval = put_user((short)why, &infop->si_code);
1314                 if (!retval)
1315                         retval = put_user(status, &infop->si_status);
1316         }
1317         if (!retval && infop)
1318                 retval = put_user(task_pid_nr_ns(p, ns), &infop->si_pid);
1319         if (!retval && infop)
1320                 retval = put_user(p->uid, &infop->si_uid);
1321         if (!retval)
1322                 retval = task_pid_nr_ns(p, ns);
1323
1324         if (traced) {
1325                 write_lock_irq(&tasklist_lock);
1326                 /* We dropped tasklist, ptracer could die and untrace */
1327                 ptrace_unlink(p);
1328                 /*
1329                  * If this is not a detached task, notify the parent.
1330                  * If it's still not detached after that, don't release
1331                  * it now.
1332                  */
1333                 if (p->exit_signal != -1) {
1334                         do_notify_parent(p, p->exit_signal);
1335                         if (p->exit_signal != -1) {
1336                                 p->exit_state = EXIT_ZOMBIE;
1337                                 p = NULL;
1338                         }
1339                 }
1340                 write_unlock_irq(&tasklist_lock);
1341         }
1342         if (p != NULL)
1343                 release_task(p);
1344
1345         return retval;
1346 }
1347
1348 /*
1349  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1350  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1351  * the lock and this task is uninteresting.  If we return nonzero, we have
1352  * released the lock and the system call should return.
1353  */
1354 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1355                              int noreap, struct siginfo __user *infop,
1356                              int __user *stat_addr, struct rusage __user *ru)
1357 {
1358         int retval, exit_code;
1359         pid_t pid;
1360
1361         if (!p->exit_code)
1362                 return 0;
1363         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1364             p->signal->group_stop_count > 0)
1365                 /*
1366                  * A group stop is in progress and this is the group leader.
1367                  * We won't report until all threads have stopped.
1368                  */
1369                 return 0;
1370
1371         /*
1372          * Now we are pretty sure this task is interesting.
1373          * Make sure it doesn't get reaped out from under us while we
1374          * give up the lock and then examine it below.  We don't want to
1375          * keep holding onto the tasklist_lock while we call getrusage and
1376          * possibly take page faults for user memory.
1377          */
1378         pid = task_pid_nr_ns(p, current->nsproxy->pid_ns);
1379         get_task_struct(p);
1380         read_unlock(&tasklist_lock);
1381
1382         if (unlikely(noreap)) {
1383                 uid_t uid = p->uid;
1384                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1385
1386                 exit_code = p->exit_code;
1387                 if (unlikely(!exit_code) || unlikely(p->exit_state))
1388                         goto bail_ref;
1389                 return wait_noreap_copyout(p, pid, uid,
1390                                            why, exit_code,
1391                                            infop, ru);
1392         }
1393
1394         write_lock_irq(&tasklist_lock);
1395
1396         /*
1397          * This uses xchg to be atomic with the thread resuming and setting
1398          * it.  It must also be done with the write lock held to prevent a
1399          * race with the EXIT_ZOMBIE case.
1400          */
1401         exit_code = xchg(&p->exit_code, 0);
1402         if (unlikely(p->exit_state)) {
1403                 /*
1404                  * The task resumed and then died.  Let the next iteration
1405                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1406                  * already be zero here if it resumed and did _exit(0).
1407                  * The task itself is dead and won't touch exit_code again;
1408                  * other processors in this function are locked out.
1409                  */
1410                 p->exit_code = exit_code;
1411                 exit_code = 0;
1412         }
1413         if (unlikely(exit_code == 0)) {
1414                 /*
1415                  * Another thread in this function got to it first, or it
1416                  * resumed, or it resumed and then died.
1417                  */
1418                 write_unlock_irq(&tasklist_lock);
1419 bail_ref:
1420                 put_task_struct(p);
1421                 /*
1422                  * We are returning to the wait loop without having successfully
1423                  * removed the process and having released the lock. We cannot
1424                  * continue, since the "p" task pointer is potentially stale.
1425                  *
1426                  * Return -EAGAIN, and do_wait() will restart the loop from the
1427                  * beginning. Do _not_ re-acquire the lock.
1428                  */
1429                 return -EAGAIN;
1430         }
1431
1432         /* move to end of parent's list to avoid starvation */
1433         remove_parent(p);
1434         add_parent(p);
1435
1436         write_unlock_irq(&tasklist_lock);
1437
1438         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1439         if (!retval && stat_addr)
1440                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1441         if (!retval && infop)
1442                 retval = put_user(SIGCHLD, &infop->si_signo);
1443         if (!retval && infop)
1444                 retval = put_user(0, &infop->si_errno);
1445         if (!retval && infop)
1446                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1447                                           ? CLD_TRAPPED : CLD_STOPPED),
1448                                   &infop->si_code);
1449         if (!retval && infop)
1450                 retval = put_user(exit_code, &infop->si_status);
1451         if (!retval && infop)
1452                 retval = put_user(pid, &infop->si_pid);
1453         if (!retval && infop)
1454                 retval = put_user(p->uid, &infop->si_uid);
1455         if (!retval)
1456                 retval = pid;
1457         put_task_struct(p);
1458
1459         BUG_ON(!retval);
1460         return retval;
1461 }
1462
1463 /*
1464  * Handle do_wait work for one task in a live, non-stopped state.
1465  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1466  * the lock and this task is uninteresting.  If we return nonzero, we have
1467  * released the lock and the system call should return.
1468  */
1469 static int wait_task_continued(struct task_struct *p, int noreap,
1470                                struct siginfo __user *infop,
1471                                int __user *stat_addr, struct rusage __user *ru)
1472 {
1473         int retval;
1474         pid_t pid;
1475         uid_t uid;
1476         struct pid_namespace *ns;
1477
1478         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1479                 return 0;
1480
1481         spin_lock_irq(&p->sighand->siglock);
1482         /* Re-check with the lock held.  */
1483         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1484                 spin_unlock_irq(&p->sighand->siglock);
1485                 return 0;
1486         }
1487         if (!noreap)
1488                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1489         spin_unlock_irq(&p->sighand->siglock);
1490
1491         ns = current->nsproxy->pid_ns;
1492         pid = task_pid_nr_ns(p, ns);
1493         uid = p->uid;
1494         get_task_struct(p);
1495         read_unlock(&tasklist_lock);
1496
1497         if (!infop) {
1498                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1499                 put_task_struct(p);
1500                 if (!retval && stat_addr)
1501                         retval = put_user(0xffff, stat_addr);
1502                 if (!retval)
1503                         retval = task_pid_nr_ns(p, ns);
1504         } else {
1505                 retval = wait_noreap_copyout(p, pid, uid,
1506                                              CLD_CONTINUED, SIGCONT,
1507                                              infop, ru);
1508                 BUG_ON(retval == 0);
1509         }
1510
1511         return retval;
1512 }
1513
1514
1515 static inline int my_ptrace_child(struct task_struct *p)
1516 {
1517         if (!(p->ptrace & PT_PTRACED))
1518                 return 0;
1519         if (!(p->ptrace & PT_ATTACHED))
1520                 return 1;
1521         /*
1522          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1523          * we are the attacher.  If we are the real parent, this is a race
1524          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1525          * which we have to switch the parent links, but has already set
1526          * the flags in p->ptrace.
1527          */
1528         return (p->parent != p->real_parent);
1529 }
1530
1531 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1532                     int __user *stat_addr, struct rusage __user *ru)
1533 {
1534         DECLARE_WAITQUEUE(wait, current);
1535         struct task_struct *tsk;
1536         int flag, retval;
1537         int allowed, denied;
1538
1539         add_wait_queue(&current->signal->wait_chldexit,&wait);
1540 repeat:
1541         /*
1542          * We will set this flag if we see any child that might later
1543          * match our criteria, even if we are not able to reap it yet.
1544          */
1545         flag = 0;
1546         allowed = denied = 0;
1547         current->state = TASK_INTERRUPTIBLE;
1548         read_lock(&tasklist_lock);
1549         tsk = current;
1550         do {
1551                 struct task_struct *p;
1552                 int ret;
1553
1554                 list_for_each_entry(p, &tsk->children, sibling) {
1555                         ret = eligible_child(pid, options, p);
1556                         if (!ret)
1557                                 continue;
1558
1559                         if (unlikely(ret < 0)) {
1560                                 denied = ret;
1561                                 continue;
1562                         }
1563                         allowed = 1;
1564
1565                         if (task_is_stopped_or_traced(p)) {
1566                                 /*
1567                                  * It's stopped now, so it might later
1568                                  * continue, exit, or stop again.
1569                                  *
1570                                  * When we hit the race with PTRACE_ATTACH, we
1571                                  * will not report this child.  But the race
1572                                  * means it has not yet been moved to our
1573                                  * ptrace_children list, so we need to set the
1574                                  * flag here to avoid a spurious ECHILD when
1575                                  * the race happens with the only child.
1576                                  */
1577                                 flag = 1;
1578
1579                                 if (!my_ptrace_child(p)) {
1580                                         if (task_is_traced(p))
1581                                                 continue;
1582                                         if (!(options & WUNTRACED))
1583                                                 continue;
1584                                 }
1585
1586                                 retval = wait_task_stopped(p, ret == 2,
1587                                                 (options & WNOWAIT), infop,
1588                                                 stat_addr, ru);
1589                                 if (retval == -EAGAIN)
1590                                         goto repeat;
1591                                 if (retval != 0) /* He released the lock.  */
1592                                         goto end;
1593                         } else if (p->exit_state == EXIT_ZOMBIE) {
1594                                 /*
1595                                  * Eligible but we cannot release it yet:
1596                                  */
1597                                 if (ret == 2)
1598                                         goto check_continued;
1599                                 if (!likely(options & WEXITED))
1600                                         continue;
1601                                 retval = wait_task_zombie(p,
1602                                                 (options & WNOWAIT), infop,
1603                                                 stat_addr, ru);
1604                                 /* He released the lock.  */
1605                                 if (retval != 0)
1606                                         goto end;
1607                         } else if (p->exit_state != EXIT_DEAD) {
1608 check_continued:
1609                                 /*
1610                                  * It's running now, so it might later
1611                                  * exit, stop, or stop and then continue.
1612                                  */
1613                                 flag = 1;
1614                                 if (!unlikely(options & WCONTINUED))
1615                                         continue;
1616                                 retval = wait_task_continued(p,
1617                                                 (options & WNOWAIT), infop,
1618                                                 stat_addr, ru);
1619                                 if (retval != 0) /* He released the lock.  */
1620                                         goto end;
1621                         }
1622                 }
1623                 if (!flag) {
1624                         list_for_each_entry(p, &tsk->ptrace_children,
1625                                             ptrace_list) {
1626                                 if (!eligible_child(pid, options, p))
1627                                         continue;
1628                                 flag = 1;
1629                                 break;
1630                         }
1631                 }
1632                 if (options & __WNOTHREAD)
1633                         break;
1634                 tsk = next_thread(tsk);
1635                 BUG_ON(tsk->signal != current->signal);
1636         } while (tsk != current);
1637
1638         read_unlock(&tasklist_lock);
1639         if (flag) {
1640                 retval = 0;
1641                 if (options & WNOHANG)
1642                         goto end;
1643                 retval = -ERESTARTSYS;
1644                 if (signal_pending(current))
1645                         goto end;
1646                 schedule();
1647                 goto repeat;
1648         }
1649         retval = -ECHILD;
1650         if (unlikely(denied) && !allowed)
1651                 retval = denied;
1652 end:
1653         current->state = TASK_RUNNING;
1654         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1655         if (infop) {
1656                 if (retval > 0)
1657                 retval = 0;
1658                 else {
1659                         /*
1660                          * For a WNOHANG return, clear out all the fields
1661                          * we would set so the user can easily tell the
1662                          * difference.
1663                          */
1664                         if (!retval)
1665                                 retval = put_user(0, &infop->si_signo);
1666                         if (!retval)
1667                                 retval = put_user(0, &infop->si_errno);
1668                         if (!retval)
1669                                 retval = put_user(0, &infop->si_code);
1670                         if (!retval)
1671                                 retval = put_user(0, &infop->si_pid);
1672                         if (!retval)
1673                                 retval = put_user(0, &infop->si_uid);
1674                         if (!retval)
1675                                 retval = put_user(0, &infop->si_status);
1676                 }
1677         }
1678         return retval;
1679 }
1680
1681 asmlinkage long sys_waitid(int which, pid_t pid,
1682                            struct siginfo __user *infop, int options,
1683                            struct rusage __user *ru)
1684 {
1685         long ret;
1686
1687         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1688                 return -EINVAL;
1689         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1690                 return -EINVAL;
1691
1692         switch (which) {
1693         case P_ALL:
1694                 pid = -1;
1695                 break;
1696         case P_PID:
1697                 if (pid <= 0)
1698                         return -EINVAL;
1699                 break;
1700         case P_PGID:
1701                 if (pid <= 0)
1702                         return -EINVAL;
1703                 pid = -pid;
1704                 break;
1705         default:
1706                 return -EINVAL;
1707         }
1708
1709         ret = do_wait(pid, options, infop, NULL, ru);
1710
1711         /* avoid REGPARM breakage on x86: */
1712         prevent_tail_call(ret);
1713         return ret;
1714 }
1715
1716 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1717                           int options, struct rusage __user *ru)
1718 {
1719         long ret;
1720
1721         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1722                         __WNOTHREAD|__WCLONE|__WALL))
1723                 return -EINVAL;
1724         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1725
1726         /* avoid REGPARM breakage on x86: */
1727         prevent_tail_call(ret);
1728         return ret;
1729 }
1730
1731 #ifdef __ARCH_WANT_SYS_WAITPID
1732
1733 /*
1734  * sys_waitpid() remains for compatibility. waitpid() should be
1735  * implemented by calling sys_wait4() from libc.a.
1736  */
1737 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1738 {
1739         return sys_wait4(pid, stat_addr, options, NULL);
1740 }
1741
1742 #endif