ae2b92be5faec1efa73beefb63304a22d030fc16
[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/iocontext.h>
17 #include <linux/key.h>
18 #include <linux/security.h>
19 #include <linux/cpu.h>
20 #include <linux/acct.h>
21 #include <linux/tsacct_kern.h>
22 #include <linux/file.h>
23 #include <linux/fdtable.h>
24 #include <linux/binfmts.h>
25 #include <linux/nsproxy.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/ptrace.h>
28 #include <linux/profile.h>
29 #include <linux/mount.h>
30 #include <linux/proc_fs.h>
31 #include <linux/kthread.h>
32 #include <linux/mempolicy.h>
33 #include <linux/taskstats_kern.h>
34 #include <linux/delayacct.h>
35 #include <linux/freezer.h>
36 #include <linux/cgroup.h>
37 #include <linux/syscalls.h>
38 #include <linux/signal.h>
39 #include <linux/posix-timers.h>
40 #include <linux/cn_proc.h>
41 #include <linux/mutex.h>
42 #include <linux/futex.h>
43 #include <linux/compat.h>
44 #include <linux/pipe_fs_i.h>
45 #include <linux/audit.h> /* for audit_free() */
46 #include <linux/resource.h>
47 #include <linux/blkdev.h>
48 #include <linux/task_io_accounting_ops.h>
49 #include <linux/tracehook.h>
50 #include <trace/sched.h>
51
52 #include <asm/uaccess.h>
53 #include <asm/unistd.h>
54 #include <asm/pgtable.h>
55 #include <asm/mmu_context.h>
56
57 static void exit_mm(struct task_struct * tsk);
58
59 static inline int task_detached(struct task_struct *p)
60 {
61         return p->exit_signal == -1;
62 }
63
64 static void __unhash_process(struct task_struct *p)
65 {
66         nr_threads--;
67         detach_pid(p, PIDTYPE_PID);
68         if (thread_group_leader(p)) {
69                 detach_pid(p, PIDTYPE_PGID);
70                 detach_pid(p, PIDTYPE_SID);
71
72                 list_del_rcu(&p->tasks);
73                 __get_cpu_var(process_counts)--;
74         }
75         list_del_rcu(&p->thread_group);
76         list_del_init(&p->sibling);
77 }
78
79 /*
80  * This function expects the tasklist_lock write-locked.
81  */
82 static void __exit_signal(struct task_struct *tsk)
83 {
84         struct signal_struct *sig = tsk->signal;
85         struct sighand_struct *sighand;
86
87         BUG_ON(!sig);
88         BUG_ON(!atomic_read(&sig->count));
89
90         sighand = rcu_dereference(tsk->sighand);
91         spin_lock(&sighand->siglock);
92
93         posix_cpu_timers_exit(tsk);
94         if (atomic_dec_and_test(&sig->count))
95                 posix_cpu_timers_exit_group(tsk);
96         else {
97                 /*
98                  * If there is any task waiting for the group exit
99                  * then notify it:
100                  */
101                 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
102                         wake_up_process(sig->group_exit_task);
103
104                 if (tsk == sig->curr_target)
105                         sig->curr_target = next_thread(tsk);
106                 /*
107                  * Accumulate here the counters for all threads but the
108                  * group leader as they die, so they can be added into
109                  * the process-wide totals when those are taken.
110                  * The group leader stays around as a zombie as long
111                  * as there are other threads.  When it gets reaped,
112                  * the exit.c code will add its counts into these totals.
113                  * We won't ever get here for the group leader, since it
114                  * will have been the last reference on the signal_struct.
115                  */
116                 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
117                 sig->min_flt += tsk->min_flt;
118                 sig->maj_flt += tsk->maj_flt;
119                 sig->nvcsw += tsk->nvcsw;
120                 sig->nivcsw += tsk->nivcsw;
121                 sig->inblock += task_io_get_inblock(tsk);
122                 sig->oublock += task_io_get_oublock(tsk);
123                 task_io_accounting_add(&sig->ioac, &tsk->ioac);
124                 sig = NULL; /* Marker for below. */
125         }
126
127         __unhash_process(tsk);
128
129         /*
130          * Do this under ->siglock, we can race with another thread
131          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
132          */
133         flush_sigqueue(&tsk->pending);
134
135         tsk->signal = NULL;
136         tsk->sighand = NULL;
137         spin_unlock(&sighand->siglock);
138
139         __cleanup_sighand(sighand);
140         clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
141         if (sig) {
142                 flush_sigqueue(&sig->shared_pending);
143                 taskstats_tgid_free(sig);
144                 /*
145                  * Make sure ->signal can't go away under rq->lock,
146                  * see account_group_exec_runtime().
147                  */
148                 task_rq_unlock_wait(tsk);
149                 __cleanup_signal(sig);
150         }
151 }
152
153 static void delayed_put_task_struct(struct rcu_head *rhp)
154 {
155         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
156
157         trace_sched_process_free(tsk);
158         put_task_struct(tsk);
159 }
160
161
162 void release_task(struct task_struct * p)
163 {
164         struct task_struct *leader;
165         int zap_leader;
166 repeat:
167         tracehook_prepare_release_task(p);
168         atomic_dec(&p->user->processes);
169         proc_flush_task(p);
170         write_lock_irq(&tasklist_lock);
171         tracehook_finish_release_task(p);
172         __exit_signal(p);
173
174         /*
175          * If we are the last non-leader member of the thread
176          * group, and the leader is zombie, then notify the
177          * group leader's parent process. (if it wants notification.)
178          */
179         zap_leader = 0;
180         leader = p->group_leader;
181         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
182                 BUG_ON(task_detached(leader));
183                 do_notify_parent(leader, leader->exit_signal);
184                 /*
185                  * If we were the last child thread and the leader has
186                  * exited already, and the leader's parent ignores SIGCHLD,
187                  * then we are the one who should release the leader.
188                  *
189                  * do_notify_parent() will have marked it self-reaping in
190                  * that case.
191                  */
192                 zap_leader = task_detached(leader);
193
194                 /*
195                  * This maintains the invariant that release_task()
196                  * only runs on a task in EXIT_DEAD, just for sanity.
197                  */
198                 if (zap_leader)
199                         leader->exit_state = EXIT_DEAD;
200         }
201
202         write_unlock_irq(&tasklist_lock);
203         release_thread(p);
204         call_rcu(&p->rcu, delayed_put_task_struct);
205
206         p = leader;
207         if (unlikely(zap_leader))
208                 goto repeat;
209 }
210
211 /*
212  * This checks not only the pgrp, but falls back on the pid if no
213  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
214  * without this...
215  *
216  * The caller must hold rcu lock or the tasklist lock.
217  */
218 struct pid *session_of_pgrp(struct pid *pgrp)
219 {
220         struct task_struct *p;
221         struct pid *sid = NULL;
222
223         p = pid_task(pgrp, PIDTYPE_PGID);
224         if (p == NULL)
225                 p = pid_task(pgrp, PIDTYPE_PID);
226         if (p != NULL)
227                 sid = task_session(p);
228
229         return sid;
230 }
231
232 /*
233  * Determine if a process group is "orphaned", according to the POSIX
234  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
235  * by terminal-generated stop signals.  Newly orphaned process groups are
236  * to receive a SIGHUP and a SIGCONT.
237  *
238  * "I ask you, have you ever known what it is to be an orphan?"
239  */
240 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
241 {
242         struct task_struct *p;
243
244         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
245                 if ((p == ignored_task) ||
246                     (p->exit_state && thread_group_empty(p)) ||
247                     is_global_init(p->real_parent))
248                         continue;
249
250                 if (task_pgrp(p->real_parent) != pgrp &&
251                     task_session(p->real_parent) == task_session(p))
252                         return 0;
253         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
254
255         return 1;
256 }
257
258 int is_current_pgrp_orphaned(void)
259 {
260         int retval;
261
262         read_lock(&tasklist_lock);
263         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
264         read_unlock(&tasklist_lock);
265
266         return retval;
267 }
268
269 static int has_stopped_jobs(struct pid *pgrp)
270 {
271         int retval = 0;
272         struct task_struct *p;
273
274         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
275                 if (!task_is_stopped(p))
276                         continue;
277                 retval = 1;
278                 break;
279         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
280         return retval;
281 }
282
283 /*
284  * Check to see if any process groups have become orphaned as
285  * a result of our exiting, and if they have any stopped jobs,
286  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
287  */
288 static void
289 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
290 {
291         struct pid *pgrp = task_pgrp(tsk);
292         struct task_struct *ignored_task = tsk;
293
294         if (!parent)
295                  /* exit: our father is in a different pgrp than
296                   * we are and we were the only connection outside.
297                   */
298                 parent = tsk->real_parent;
299         else
300                 /* reparent: our child is in a different pgrp than
301                  * we are, and it was the only connection outside.
302                  */
303                 ignored_task = NULL;
304
305         if (task_pgrp(parent) != pgrp &&
306             task_session(parent) == task_session(tsk) &&
307             will_become_orphaned_pgrp(pgrp, ignored_task) &&
308             has_stopped_jobs(pgrp)) {
309                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
310                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
311         }
312 }
313
314 /**
315  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
316  *
317  * If a kernel thread is launched as a result of a system call, or if
318  * it ever exits, it should generally reparent itself to kthreadd so it
319  * isn't in the way of other processes and is correctly cleaned up on exit.
320  *
321  * The various task state such as scheduling policy and priority may have
322  * been inherited from a user process, so we reset them to sane values here.
323  *
324  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
325  */
326 static void reparent_to_kthreadd(void)
327 {
328         write_lock_irq(&tasklist_lock);
329
330         ptrace_unlink(current);
331         /* Reparent to init */
332         current->real_parent = current->parent = kthreadd_task;
333         list_move_tail(&current->sibling, &current->real_parent->children);
334
335         /* Set the exit signal to SIGCHLD so we signal init on exit */
336         current->exit_signal = SIGCHLD;
337
338         if (task_nice(current) < 0)
339                 set_user_nice(current, 0);
340         /* cpus_allowed? */
341         /* rt_priority? */
342         /* signals? */
343         security_task_reparent_to_init(current);
344         memcpy(current->signal->rlim, init_task.signal->rlim,
345                sizeof(current->signal->rlim));
346         atomic_inc(&(INIT_USER->__count));
347         write_unlock_irq(&tasklist_lock);
348         switch_uid(INIT_USER);
349 }
350
351 void __set_special_pids(struct pid *pid)
352 {
353         struct task_struct *curr = current->group_leader;
354         pid_t nr = pid_nr(pid);
355
356         if (task_session(curr) != pid) {
357                 change_pid(curr, PIDTYPE_SID, pid);
358                 set_task_session(curr, nr);
359         }
360         if (task_pgrp(curr) != pid) {
361                 change_pid(curr, PIDTYPE_PGID, pid);
362                 set_task_pgrp(curr, nr);
363         }
364 }
365
366 static void set_special_pids(struct pid *pid)
367 {
368         write_lock_irq(&tasklist_lock);
369         __set_special_pids(pid);
370         write_unlock_irq(&tasklist_lock);
371 }
372
373 /*
374  * Let kernel threads use this to say that they
375  * allow a certain signal (since daemonize() will
376  * have disabled all of them by default).
377  */
378 int allow_signal(int sig)
379 {
380         if (!valid_signal(sig) || sig < 1)
381                 return -EINVAL;
382
383         spin_lock_irq(&current->sighand->siglock);
384         sigdelset(&current->blocked, sig);
385         if (!current->mm) {
386                 /* Kernel threads handle their own signals.
387                    Let the signal code know it'll be handled, so
388                    that they don't get converted to SIGKILL or
389                    just silently dropped */
390                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
391         }
392         recalc_sigpending();
393         spin_unlock_irq(&current->sighand->siglock);
394         return 0;
395 }
396
397 EXPORT_SYMBOL(allow_signal);
398
399 int disallow_signal(int sig)
400 {
401         if (!valid_signal(sig) || sig < 1)
402                 return -EINVAL;
403
404         spin_lock_irq(&current->sighand->siglock);
405         current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
406         recalc_sigpending();
407         spin_unlock_irq(&current->sighand->siglock);
408         return 0;
409 }
410
411 EXPORT_SYMBOL(disallow_signal);
412
413 /*
414  *      Put all the gunge required to become a kernel thread without
415  *      attached user resources in one place where it belongs.
416  */
417
418 void daemonize(const char *name, ...)
419 {
420         va_list args;
421         struct fs_struct *fs;
422         sigset_t blocked;
423
424         va_start(args, name);
425         vsnprintf(current->comm, sizeof(current->comm), name, args);
426         va_end(args);
427
428         /*
429          * If we were started as result of loading a module, close all of the
430          * user space pages.  We don't need them, and if we didn't close them
431          * they would be locked into memory.
432          */
433         exit_mm(current);
434         /*
435          * We don't want to have TIF_FREEZE set if the system-wide hibernation
436          * or suspend transition begins right now.
437          */
438         current->flags |= (PF_NOFREEZE | PF_KTHREAD);
439
440         if (current->nsproxy != &init_nsproxy) {
441                 get_nsproxy(&init_nsproxy);
442                 switch_task_namespaces(current, &init_nsproxy);
443         }
444         set_special_pids(&init_struct_pid);
445         proc_clear_tty(current);
446
447         /* Block and flush all signals */
448         sigfillset(&blocked);
449         sigprocmask(SIG_BLOCK, &blocked, NULL);
450         flush_signals(current);
451
452         /* Become as one with the init task */
453
454         exit_fs(current);       /* current->fs->count--; */
455         fs = init_task.fs;
456         current->fs = fs;
457         atomic_inc(&fs->count);
458
459         exit_files(current);
460         current->files = init_task.files;
461         atomic_inc(&current->files->count);
462
463         reparent_to_kthreadd();
464 }
465
466 EXPORT_SYMBOL(daemonize);
467
468 static void close_files(struct files_struct * files)
469 {
470         int i, j;
471         struct fdtable *fdt;
472
473         j = 0;
474
475         /*
476          * It is safe to dereference the fd table without RCU or
477          * ->file_lock because this is the last reference to the
478          * files structure.
479          */
480         fdt = files_fdtable(files);
481         for (;;) {
482                 unsigned long set;
483                 i = j * __NFDBITS;
484                 if (i >= fdt->max_fds)
485                         break;
486                 set = fdt->open_fds->fds_bits[j++];
487                 while (set) {
488                         if (set & 1) {
489                                 struct file * file = xchg(&fdt->fd[i], NULL);
490                                 if (file) {
491                                         filp_close(file, files);
492                                         cond_resched();
493                                 }
494                         }
495                         i++;
496                         set >>= 1;
497                 }
498         }
499 }
500
501 struct files_struct *get_files_struct(struct task_struct *task)
502 {
503         struct files_struct *files;
504
505         task_lock(task);
506         files = task->files;
507         if (files)
508                 atomic_inc(&files->count);
509         task_unlock(task);
510
511         return files;
512 }
513
514 void put_files_struct(struct files_struct *files)
515 {
516         struct fdtable *fdt;
517
518         if (atomic_dec_and_test(&files->count)) {
519                 close_files(files);
520                 /*
521                  * Free the fd and fdset arrays if we expanded them.
522                  * If the fdtable was embedded, pass files for freeing
523                  * at the end of the RCU grace period. Otherwise,
524                  * you can free files immediately.
525                  */
526                 fdt = files_fdtable(files);
527                 if (fdt != &files->fdtab)
528                         kmem_cache_free(files_cachep, files);
529                 free_fdtable(fdt);
530         }
531 }
532
533 void reset_files_struct(struct files_struct *files)
534 {
535         struct task_struct *tsk = current;
536         struct files_struct *old;
537
538         old = tsk->files;
539         task_lock(tsk);
540         tsk->files = files;
541         task_unlock(tsk);
542         put_files_struct(old);
543 }
544
545 void exit_files(struct task_struct *tsk)
546 {
547         struct files_struct * files = tsk->files;
548
549         if (files) {
550                 task_lock(tsk);
551                 tsk->files = NULL;
552                 task_unlock(tsk);
553                 put_files_struct(files);
554         }
555 }
556
557 void put_fs_struct(struct fs_struct *fs)
558 {
559         /* No need to hold fs->lock if we are killing it */
560         if (atomic_dec_and_test(&fs->count)) {
561                 path_put(&fs->root);
562                 path_put(&fs->pwd);
563                 kmem_cache_free(fs_cachep, fs);
564         }
565 }
566
567 void exit_fs(struct task_struct *tsk)
568 {
569         struct fs_struct * fs = tsk->fs;
570
571         if (fs) {
572                 task_lock(tsk);
573                 tsk->fs = NULL;
574                 task_unlock(tsk);
575                 put_fs_struct(fs);
576         }
577 }
578
579 EXPORT_SYMBOL_GPL(exit_fs);
580
581 #ifdef CONFIG_MM_OWNER
582 /*
583  * Task p is exiting and it owned mm, lets find a new owner for it
584  */
585 static inline int
586 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
587 {
588         /*
589          * If there are other users of the mm and the owner (us) is exiting
590          * we need to find a new owner to take on the responsibility.
591          */
592         if (atomic_read(&mm->mm_users) <= 1)
593                 return 0;
594         if (mm->owner != p)
595                 return 0;
596         return 1;
597 }
598
599 void mm_update_next_owner(struct mm_struct *mm)
600 {
601         struct task_struct *c, *g, *p = current;
602
603 retry:
604         if (!mm_need_new_owner(mm, p))
605                 return;
606
607         read_lock(&tasklist_lock);
608         /*
609          * Search in the children
610          */
611         list_for_each_entry(c, &p->children, sibling) {
612                 if (c->mm == mm)
613                         goto assign_new_owner;
614         }
615
616         /*
617          * Search in the siblings
618          */
619         list_for_each_entry(c, &p->parent->children, sibling) {
620                 if (c->mm == mm)
621                         goto assign_new_owner;
622         }
623
624         /*
625          * Search through everything else. We should not get
626          * here often
627          */
628         do_each_thread(g, c) {
629                 if (c->mm == mm)
630                         goto assign_new_owner;
631         } while_each_thread(g, c);
632
633         read_unlock(&tasklist_lock);
634         /*
635          * We found no owner yet mm_users > 1: this implies that we are
636          * most likely racing with swapoff (try_to_unuse()) or /proc or
637          * ptrace or page migration (get_task_mm()).  Mark owner as NULL,
638          * so that subsystems can understand the callback and take action.
639          */
640         down_write(&mm->mmap_sem);
641         cgroup_mm_owner_callbacks(mm->owner, NULL);
642         mm->owner = NULL;
643         up_write(&mm->mmap_sem);
644         return;
645
646 assign_new_owner:
647         BUG_ON(c == p);
648         get_task_struct(c);
649         read_unlock(&tasklist_lock);
650         down_write(&mm->mmap_sem);
651         /*
652          * The task_lock protects c->mm from changing.
653          * We always want mm->owner->mm == mm
654          */
655         task_lock(c);
656         if (c->mm != mm) {
657                 task_unlock(c);
658                 up_write(&mm->mmap_sem);
659                 put_task_struct(c);
660                 goto retry;
661         }
662         cgroup_mm_owner_callbacks(mm->owner, c);
663         mm->owner = c;
664         task_unlock(c);
665         up_write(&mm->mmap_sem);
666         put_task_struct(c);
667 }
668 #endif /* CONFIG_MM_OWNER */
669
670 /*
671  * Turn us into a lazy TLB process if we
672  * aren't already..
673  */
674 static void exit_mm(struct task_struct * tsk)
675 {
676         struct mm_struct *mm = tsk->mm;
677         struct core_state *core_state;
678
679         mm_release(tsk, mm);
680         if (!mm)
681                 return;
682         /*
683          * Serialize with any possible pending coredump.
684          * We must hold mmap_sem around checking core_state
685          * and clearing tsk->mm.  The core-inducing thread
686          * will increment ->nr_threads for each thread in the
687          * group with ->mm != NULL.
688          */
689         down_read(&mm->mmap_sem);
690         core_state = mm->core_state;
691         if (core_state) {
692                 struct core_thread self;
693                 up_read(&mm->mmap_sem);
694
695                 self.task = tsk;
696                 self.next = xchg(&core_state->dumper.next, &self);
697                 /*
698                  * Implies mb(), the result of xchg() must be visible
699                  * to core_state->dumper.
700                  */
701                 if (atomic_dec_and_test(&core_state->nr_threads))
702                         complete(&core_state->startup);
703
704                 for (;;) {
705                         set_task_state(tsk, TASK_UNINTERRUPTIBLE);
706                         if (!self.task) /* see coredump_finish() */
707                                 break;
708                         schedule();
709                 }
710                 __set_task_state(tsk, TASK_RUNNING);
711                 down_read(&mm->mmap_sem);
712         }
713         atomic_inc(&mm->mm_count);
714         BUG_ON(mm != tsk->active_mm);
715         /* more a memory barrier than a real lock */
716         task_lock(tsk);
717         tsk->mm = NULL;
718         up_read(&mm->mmap_sem);
719         enter_lazy_tlb(mm, current);
720         /* We don't want this task to be frozen prematurely */
721         clear_freeze_flag(tsk);
722         task_unlock(tsk);
723         mm_update_next_owner(mm);
724         mmput(mm);
725 }
726
727 /*
728  * Return nonzero if @parent's children should reap themselves.
729  *
730  * Called with write_lock_irq(&tasklist_lock) held.
731  */
732 static int ignoring_children(struct task_struct *parent)
733 {
734         int ret;
735         struct sighand_struct *psig = parent->sighand;
736         unsigned long flags;
737         spin_lock_irqsave(&psig->siglock, flags);
738         ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
739                (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
740         spin_unlock_irqrestore(&psig->siglock, flags);
741         return ret;
742 }
743
744 /*
745  * Detach all tasks we were using ptrace on.
746  * Any that need to be release_task'd are put on the @dead list.
747  *
748  * Called with write_lock(&tasklist_lock) held.
749  */
750 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
751 {
752         struct task_struct *p, *n;
753         int ign = -1;
754
755         list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
756                 __ptrace_unlink(p);
757
758                 if (p->exit_state != EXIT_ZOMBIE)
759                         continue;
760
761                 /*
762                  * If it's a zombie, our attachedness prevented normal
763                  * parent notification or self-reaping.  Do notification
764                  * now if it would have happened earlier.  If it should
765                  * reap itself, add it to the @dead list.  We can't call
766                  * release_task() here because we already hold tasklist_lock.
767                  *
768                  * If it's our own child, there is no notification to do.
769                  * But if our normal children self-reap, then this child
770                  * was prevented by ptrace and we must reap it now.
771                  */
772                 if (!task_detached(p) && thread_group_empty(p)) {
773                         if (!same_thread_group(p->real_parent, parent))
774                                 do_notify_parent(p, p->exit_signal);
775                         else {
776                                 if (ign < 0)
777                                         ign = ignoring_children(parent);
778                                 if (ign)
779                                         p->exit_signal = -1;
780                         }
781                 }
782
783                 if (task_detached(p)) {
784                         /*
785                          * Mark it as in the process of being reaped.
786                          */
787                         p->exit_state = EXIT_DEAD;
788                         list_add(&p->ptrace_entry, dead);
789                 }
790         }
791 }
792
793 /*
794  * Finish up exit-time ptrace cleanup.
795  *
796  * Called without locks.
797  */
798 static void ptrace_exit_finish(struct task_struct *parent,
799                                struct list_head *dead)
800 {
801         struct task_struct *p, *n;
802
803         BUG_ON(!list_empty(&parent->ptraced));
804
805         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
806                 list_del_init(&p->ptrace_entry);
807                 release_task(p);
808         }
809 }
810
811 static void reparent_thread(struct task_struct *p, struct task_struct *father)
812 {
813         if (p->pdeath_signal)
814                 /* We already hold the tasklist_lock here.  */
815                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
816
817         list_move_tail(&p->sibling, &p->real_parent->children);
818
819         /* If this is a threaded reparent there is no need to
820          * notify anyone anything has happened.
821          */
822         if (same_thread_group(p->real_parent, father))
823                 return;
824
825         /* We don't want people slaying init.  */
826         if (!task_detached(p))
827                 p->exit_signal = SIGCHLD;
828
829         /* If we'd notified the old parent about this child's death,
830          * also notify the new parent.
831          */
832         if (!ptrace_reparented(p) &&
833             p->exit_state == EXIT_ZOMBIE &&
834             !task_detached(p) && thread_group_empty(p))
835                 do_notify_parent(p, p->exit_signal);
836
837         kill_orphaned_pgrp(p, father);
838 }
839
840 /*
841  * When we die, we re-parent all our children.
842  * Try to give them to another thread in our thread
843  * group, and if no such member exists, give it to
844  * the child reaper process (ie "init") in our pid
845  * space.
846  */
847 static struct task_struct *find_new_reaper(struct task_struct *father)
848 {
849         struct pid_namespace *pid_ns = task_active_pid_ns(father);
850         struct task_struct *thread;
851
852         thread = father;
853         while_each_thread(father, thread) {
854                 if (thread->flags & PF_EXITING)
855                         continue;
856                 if (unlikely(pid_ns->child_reaper == father))
857                         pid_ns->child_reaper = thread;
858                 return thread;
859         }
860
861         if (unlikely(pid_ns->child_reaper == father)) {
862                 write_unlock_irq(&tasklist_lock);
863                 if (unlikely(pid_ns == &init_pid_ns))
864                         panic("Attempted to kill init!");
865
866                 zap_pid_ns_processes(pid_ns);
867                 write_lock_irq(&tasklist_lock);
868                 /*
869                  * We can not clear ->child_reaper or leave it alone.
870                  * There may by stealth EXIT_DEAD tasks on ->children,
871                  * forget_original_parent() must move them somewhere.
872                  */
873                 pid_ns->child_reaper = init_pid_ns.child_reaper;
874         }
875
876         return pid_ns->child_reaper;
877 }
878
879 static void forget_original_parent(struct task_struct *father)
880 {
881         struct task_struct *p, *n, *reaper;
882         LIST_HEAD(ptrace_dead);
883
884         write_lock_irq(&tasklist_lock);
885         reaper = find_new_reaper(father);
886         /*
887          * First clean up ptrace if we were using it.
888          */
889         ptrace_exit(father, &ptrace_dead);
890
891         list_for_each_entry_safe(p, n, &father->children, sibling) {
892                 p->real_parent = reaper;
893                 if (p->parent == father) {
894                         BUG_ON(p->ptrace);
895                         p->parent = p->real_parent;
896                 }
897                 reparent_thread(p, father);
898         }
899
900         write_unlock_irq(&tasklist_lock);
901         BUG_ON(!list_empty(&father->children));
902
903         ptrace_exit_finish(father, &ptrace_dead);
904 }
905
906 /*
907  * Send signals to all our closest relatives so that they know
908  * to properly mourn us..
909  */
910 static void exit_notify(struct task_struct *tsk, int group_dead)
911 {
912         int signal;
913         void *cookie;
914
915         /*
916          * This does two things:
917          *
918          * A.  Make init inherit all the child processes
919          * B.  Check to see if any process groups have become orphaned
920          *      as a result of our exiting, and if they have any stopped
921          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
922          */
923         forget_original_parent(tsk);
924         exit_task_namespaces(tsk);
925
926         write_lock_irq(&tasklist_lock);
927         if (group_dead)
928                 kill_orphaned_pgrp(tsk->group_leader, NULL);
929
930         /* Let father know we died
931          *
932          * Thread signals are configurable, but you aren't going to use
933          * that to send signals to arbitary processes.
934          * That stops right now.
935          *
936          * If the parent exec id doesn't match the exec id we saved
937          * when we started then we know the parent has changed security
938          * domain.
939          *
940          * If our self_exec id doesn't match our parent_exec_id then
941          * we have changed execution domain as these two values started
942          * the same after a fork.
943          */
944         if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
945             (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
946              tsk->self_exec_id != tsk->parent_exec_id) &&
947             !capable(CAP_KILL))
948                 tsk->exit_signal = SIGCHLD;
949
950         signal = tracehook_notify_death(tsk, &cookie, group_dead);
951         if (signal >= 0)
952                 signal = do_notify_parent(tsk, signal);
953
954         tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
955
956         /* mt-exec, de_thread() is waiting for us */
957         if (thread_group_leader(tsk) &&
958             tsk->signal->group_exit_task &&
959             tsk->signal->notify_count < 0)
960                 wake_up_process(tsk->signal->group_exit_task);
961
962         write_unlock_irq(&tasklist_lock);
963
964         tracehook_report_death(tsk, signal, cookie, group_dead);
965
966         /* If the process is dead, release it - nobody will wait for it */
967         if (signal == DEATH_REAP)
968                 release_task(tsk);
969 }
970
971 #ifdef CONFIG_DEBUG_STACK_USAGE
972 static void check_stack_usage(void)
973 {
974         static DEFINE_SPINLOCK(low_water_lock);
975         static int lowest_to_date = THREAD_SIZE;
976         unsigned long *n = end_of_stack(current);
977         unsigned long free;
978
979         while (*n == 0)
980                 n++;
981         free = (unsigned long)n - (unsigned long)end_of_stack(current);
982
983         if (free >= lowest_to_date)
984                 return;
985
986         spin_lock(&low_water_lock);
987         if (free < lowest_to_date) {
988                 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
989                                 "left\n",
990                                 current->comm, free);
991                 lowest_to_date = free;
992         }
993         spin_unlock(&low_water_lock);
994 }
995 #else
996 static inline void check_stack_usage(void) {}
997 #endif
998
999 NORET_TYPE void do_exit(long code)
1000 {
1001         struct task_struct *tsk = current;
1002         int group_dead;
1003
1004         profile_task_exit(tsk);
1005
1006         WARN_ON(atomic_read(&tsk->fs_excl));
1007
1008         if (unlikely(in_interrupt()))
1009                 panic("Aiee, killing interrupt handler!");
1010         if (unlikely(!tsk->pid))
1011                 panic("Attempted to kill the idle task!");
1012
1013         tracehook_report_exit(&code);
1014
1015         /*
1016          * We're taking recursive faults here in do_exit. Safest is to just
1017          * leave this task alone and wait for reboot.
1018          */
1019         if (unlikely(tsk->flags & PF_EXITING)) {
1020                 printk(KERN_ALERT
1021                         "Fixing recursive fault but reboot is needed!\n");
1022                 /*
1023                  * We can do this unlocked here. The futex code uses
1024                  * this flag just to verify whether the pi state
1025                  * cleanup has been done or not. In the worst case it
1026                  * loops once more. We pretend that the cleanup was
1027                  * done as there is no way to return. Either the
1028                  * OWNER_DIED bit is set by now or we push the blocked
1029                  * task into the wait for ever nirwana as well.
1030                  */
1031                 tsk->flags |= PF_EXITPIDONE;
1032                 if (tsk->io_context)
1033                         exit_io_context();
1034                 set_current_state(TASK_UNINTERRUPTIBLE);
1035                 schedule();
1036         }
1037
1038         exit_signals(tsk);  /* sets PF_EXITING */
1039         /*
1040          * tsk->flags are checked in the futex code to protect against
1041          * an exiting task cleaning up the robust pi futexes.
1042          */
1043         smp_mb();
1044         spin_unlock_wait(&tsk->pi_lock);
1045
1046         if (unlikely(in_atomic()))
1047                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1048                                 current->comm, task_pid_nr(current),
1049                                 preempt_count());
1050
1051         acct_update_integrals(tsk);
1052         if (tsk->mm) {
1053                 update_hiwater_rss(tsk->mm);
1054                 update_hiwater_vm(tsk->mm);
1055         }
1056         group_dead = atomic_dec_and_test(&tsk->signal->live);
1057         if (group_dead) {
1058                 hrtimer_cancel(&tsk->signal->real_timer);
1059                 exit_itimers(tsk->signal);
1060         }
1061         acct_collect(code, group_dead);
1062 #ifdef CONFIG_FUTEX
1063         if (unlikely(tsk->robust_list))
1064                 exit_robust_list(tsk);
1065 #ifdef CONFIG_COMPAT
1066         if (unlikely(tsk->compat_robust_list))
1067                 compat_exit_robust_list(tsk);
1068 #endif
1069 #endif
1070         if (group_dead)
1071                 tty_audit_exit();
1072         if (unlikely(tsk->audit_context))
1073                 audit_free(tsk);
1074
1075         tsk->exit_code = code;
1076         taskstats_exit(tsk, group_dead);
1077
1078         exit_mm(tsk);
1079
1080         if (group_dead)
1081                 acct_process();
1082         trace_sched_process_exit(tsk);
1083
1084         exit_sem(tsk);
1085         exit_files(tsk);
1086         exit_fs(tsk);
1087         check_stack_usage();
1088         exit_thread();
1089         cgroup_exit(tsk, 1);
1090         exit_keys(tsk);
1091
1092         if (group_dead && tsk->signal->leader)
1093                 disassociate_ctty(1);
1094
1095         module_put(task_thread_info(tsk)->exec_domain->module);
1096         if (tsk->binfmt)
1097                 module_put(tsk->binfmt->module);
1098
1099         proc_exit_connector(tsk);
1100         exit_notify(tsk, group_dead);
1101 #ifdef CONFIG_NUMA
1102         mpol_put(tsk->mempolicy);
1103         tsk->mempolicy = NULL;
1104 #endif
1105 #ifdef CONFIG_FUTEX
1106         /*
1107          * This must happen late, after the PID is not
1108          * hashed anymore:
1109          */
1110         if (unlikely(!list_empty(&tsk->pi_state_list)))
1111                 exit_pi_state_list(tsk);
1112         if (unlikely(current->pi_state_cache))
1113                 kfree(current->pi_state_cache);
1114 #endif
1115         /*
1116          * Make sure we are holding no locks:
1117          */
1118         debug_check_no_locks_held(tsk);
1119         /*
1120          * We can do this unlocked here. The futex code uses this flag
1121          * just to verify whether the pi state cleanup has been done
1122          * or not. In the worst case it loops once more.
1123          */
1124         tsk->flags |= PF_EXITPIDONE;
1125
1126         if (tsk->io_context)
1127                 exit_io_context();
1128
1129         if (tsk->splice_pipe)
1130                 __free_pipe_info(tsk->splice_pipe);
1131
1132         preempt_disable();
1133         /* causes final put_task_struct in finish_task_switch(). */
1134         tsk->state = TASK_DEAD;
1135
1136         schedule();
1137         BUG();
1138         /* Avoid "noreturn function does return".  */
1139         for (;;)
1140                 cpu_relax();    /* For when BUG is null */
1141 }
1142
1143 EXPORT_SYMBOL_GPL(do_exit);
1144
1145 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1146 {
1147         if (comp)
1148                 complete(comp);
1149
1150         do_exit(code);
1151 }
1152
1153 EXPORT_SYMBOL(complete_and_exit);
1154
1155 asmlinkage long sys_exit(int error_code)
1156 {
1157         do_exit((error_code&0xff)<<8);
1158 }
1159
1160 /*
1161  * Take down every thread in the group.  This is called by fatal signals
1162  * as well as by sys_exit_group (below).
1163  */
1164 NORET_TYPE void
1165 do_group_exit(int exit_code)
1166 {
1167         struct signal_struct *sig = current->signal;
1168
1169         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1170
1171         if (signal_group_exit(sig))
1172                 exit_code = sig->group_exit_code;
1173         else if (!thread_group_empty(current)) {
1174                 struct sighand_struct *const sighand = current->sighand;
1175                 spin_lock_irq(&sighand->siglock);
1176                 if (signal_group_exit(sig))
1177                         /* Another thread got here before we took the lock.  */
1178                         exit_code = sig->group_exit_code;
1179                 else {
1180                         sig->group_exit_code = exit_code;
1181                         sig->flags = SIGNAL_GROUP_EXIT;
1182                         zap_other_threads(current);
1183                 }
1184                 spin_unlock_irq(&sighand->siglock);
1185         }
1186
1187         do_exit(exit_code);
1188         /* NOTREACHED */
1189 }
1190
1191 /*
1192  * this kills every thread in the thread group. Note that any externally
1193  * wait4()-ing process will get the correct exit code - even if this
1194  * thread is not the thread group leader.
1195  */
1196 asmlinkage void sys_exit_group(int error_code)
1197 {
1198         do_group_exit((error_code & 0xff) << 8);
1199 }
1200
1201 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1202 {
1203         struct pid *pid = NULL;
1204         if (type == PIDTYPE_PID)
1205                 pid = task->pids[type].pid;
1206         else if (type < PIDTYPE_MAX)
1207                 pid = task->group_leader->pids[type].pid;
1208         return pid;
1209 }
1210
1211 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1212                           struct task_struct *p)
1213 {
1214         int err;
1215
1216         if (type < PIDTYPE_MAX) {
1217                 if (task_pid_type(p, type) != pid)
1218                         return 0;
1219         }
1220
1221         /* Wait for all children (clone and not) if __WALL is set;
1222          * otherwise, wait for clone children *only* if __WCLONE is
1223          * set; otherwise, wait for non-clone children *only*.  (Note:
1224          * A "clone" child here is one that reports to its parent
1225          * using a signal other than SIGCHLD.) */
1226         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1227             && !(options & __WALL))
1228                 return 0;
1229
1230         err = security_task_wait(p);
1231         if (err)
1232                 return err;
1233
1234         return 1;
1235 }
1236
1237 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1238                                int why, int status,
1239                                struct siginfo __user *infop,
1240                                struct rusage __user *rusagep)
1241 {
1242         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1243
1244         put_task_struct(p);
1245         if (!retval)
1246                 retval = put_user(SIGCHLD, &infop->si_signo);
1247         if (!retval)
1248                 retval = put_user(0, &infop->si_errno);
1249         if (!retval)
1250                 retval = put_user((short)why, &infop->si_code);
1251         if (!retval)
1252                 retval = put_user(pid, &infop->si_pid);
1253         if (!retval)
1254                 retval = put_user(uid, &infop->si_uid);
1255         if (!retval)
1256                 retval = put_user(status, &infop->si_status);
1257         if (!retval)
1258                 retval = pid;
1259         return retval;
1260 }
1261
1262 /*
1263  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1264  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1265  * the lock and this task is uninteresting.  If we return nonzero, we have
1266  * released the lock and the system call should return.
1267  */
1268 static int wait_task_zombie(struct task_struct *p, int options,
1269                             struct siginfo __user *infop,
1270                             int __user *stat_addr, struct rusage __user *ru)
1271 {
1272         unsigned long state;
1273         int retval, status, traced;
1274         pid_t pid = task_pid_vnr(p);
1275
1276         if (!likely(options & WEXITED))
1277                 return 0;
1278
1279         if (unlikely(options & WNOWAIT)) {
1280                 uid_t uid = p->uid;
1281                 int exit_code = p->exit_code;
1282                 int why, status;
1283
1284                 get_task_struct(p);
1285                 read_unlock(&tasklist_lock);
1286                 if ((exit_code & 0x7f) == 0) {
1287                         why = CLD_EXITED;
1288                         status = exit_code >> 8;
1289                 } else {
1290                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1291                         status = exit_code & 0x7f;
1292                 }
1293                 return wait_noreap_copyout(p, pid, uid, why,
1294                                            status, infop, ru);
1295         }
1296
1297         /*
1298          * Try to move the task's state to DEAD
1299          * only one thread is allowed to do this:
1300          */
1301         state = xchg(&p->exit_state, EXIT_DEAD);
1302         if (state != EXIT_ZOMBIE) {
1303                 BUG_ON(state != EXIT_DEAD);
1304                 return 0;
1305         }
1306
1307         traced = ptrace_reparented(p);
1308
1309         if (likely(!traced)) {
1310                 struct signal_struct *psig;
1311                 struct signal_struct *sig;
1312                 struct task_cputime cputime;
1313
1314                 /*
1315                  * The resource counters for the group leader are in its
1316                  * own task_struct.  Those for dead threads in the group
1317                  * are in its signal_struct, as are those for the child
1318                  * processes it has previously reaped.  All these
1319                  * accumulate in the parent's signal_struct c* fields.
1320                  *
1321                  * We don't bother to take a lock here to protect these
1322                  * p->signal fields, because they are only touched by
1323                  * __exit_signal, which runs with tasklist_lock
1324                  * write-locked anyway, and so is excluded here.  We do
1325                  * need to protect the access to p->parent->signal fields,
1326                  * as other threads in the parent group can be right
1327                  * here reaping other children at the same time.
1328                  *
1329                  * We use thread_group_cputime() to get times for the thread
1330                  * group, which consolidates times for all threads in the
1331                  * group including the group leader.
1332                  */
1333                 spin_lock_irq(&p->parent->sighand->siglock);
1334                 psig = p->parent->signal;
1335                 sig = p->signal;
1336                 thread_group_cputime(p, &cputime);
1337                 psig->cutime =
1338                         cputime_add(psig->cutime,
1339                         cputime_add(cputime.utime,
1340                                     sig->cutime));
1341                 psig->cstime =
1342                         cputime_add(psig->cstime,
1343                         cputime_add(cputime.stime,
1344                                     sig->cstime));
1345                 psig->cgtime =
1346                         cputime_add(psig->cgtime,
1347                         cputime_add(p->gtime,
1348                         cputime_add(sig->gtime,
1349                                     sig->cgtime)));
1350                 psig->cmin_flt +=
1351                         p->min_flt + sig->min_flt + sig->cmin_flt;
1352                 psig->cmaj_flt +=
1353                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1354                 psig->cnvcsw +=
1355                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1356                 psig->cnivcsw +=
1357                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1358                 psig->cinblock +=
1359                         task_io_get_inblock(p) +
1360                         sig->inblock + sig->cinblock;
1361                 psig->coublock +=
1362                         task_io_get_oublock(p) +
1363                         sig->oublock + sig->coublock;
1364                 task_io_accounting_add(&psig->ioac, &p->ioac);
1365                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1366                 spin_unlock_irq(&p->parent->sighand->siglock);
1367         }
1368
1369         /*
1370          * Now we are sure this task is interesting, and no other
1371          * thread can reap it because we set its state to EXIT_DEAD.
1372          */
1373         read_unlock(&tasklist_lock);
1374
1375         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1376         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1377                 ? p->signal->group_exit_code : p->exit_code;
1378         if (!retval && stat_addr)
1379                 retval = put_user(status, stat_addr);
1380         if (!retval && infop)
1381                 retval = put_user(SIGCHLD, &infop->si_signo);
1382         if (!retval && infop)
1383                 retval = put_user(0, &infop->si_errno);
1384         if (!retval && infop) {
1385                 int why;
1386
1387                 if ((status & 0x7f) == 0) {
1388                         why = CLD_EXITED;
1389                         status >>= 8;
1390                 } else {
1391                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1392                         status &= 0x7f;
1393                 }
1394                 retval = put_user((short)why, &infop->si_code);
1395                 if (!retval)
1396                         retval = put_user(status, &infop->si_status);
1397         }
1398         if (!retval && infop)
1399                 retval = put_user(pid, &infop->si_pid);
1400         if (!retval && infop)
1401                 retval = put_user(p->uid, &infop->si_uid);
1402         if (!retval)
1403                 retval = pid;
1404
1405         if (traced) {
1406                 write_lock_irq(&tasklist_lock);
1407                 /* We dropped tasklist, ptracer could die and untrace */
1408                 ptrace_unlink(p);
1409                 /*
1410                  * If this is not a detached task, notify the parent.
1411                  * If it's still not detached after that, don't release
1412                  * it now.
1413                  */
1414                 if (!task_detached(p)) {
1415                         do_notify_parent(p, p->exit_signal);
1416                         if (!task_detached(p)) {
1417                                 p->exit_state = EXIT_ZOMBIE;
1418                                 p = NULL;
1419                         }
1420                 }
1421                 write_unlock_irq(&tasklist_lock);
1422         }
1423         if (p != NULL)
1424                 release_task(p);
1425
1426         return retval;
1427 }
1428
1429 /*
1430  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1431  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1432  * the lock and this task is uninteresting.  If we return nonzero, we have
1433  * released the lock and the system call should return.
1434  */
1435 static int wait_task_stopped(int ptrace, struct task_struct *p,
1436                              int options, struct siginfo __user *infop,
1437                              int __user *stat_addr, struct rusage __user *ru)
1438 {
1439         int retval, exit_code, why;
1440         uid_t uid = 0; /* unneeded, required by compiler */
1441         pid_t pid;
1442
1443         if (!(options & WUNTRACED))
1444                 return 0;
1445
1446         exit_code = 0;
1447         spin_lock_irq(&p->sighand->siglock);
1448
1449         if (unlikely(!task_is_stopped_or_traced(p)))
1450                 goto unlock_sig;
1451
1452         if (!ptrace && p->signal->group_stop_count > 0)
1453                 /*
1454                  * A group stop is in progress and this is the group leader.
1455                  * We won't report until all threads have stopped.
1456                  */
1457                 goto unlock_sig;
1458
1459         exit_code = p->exit_code;
1460         if (!exit_code)
1461                 goto unlock_sig;
1462
1463         if (!unlikely(options & WNOWAIT))
1464                 p->exit_code = 0;
1465
1466         uid = p->uid;
1467 unlock_sig:
1468         spin_unlock_irq(&p->sighand->siglock);
1469         if (!exit_code)
1470                 return 0;
1471
1472         /*
1473          * Now we are pretty sure this task is interesting.
1474          * Make sure it doesn't get reaped out from under us while we
1475          * give up the lock and then examine it below.  We don't want to
1476          * keep holding onto the tasklist_lock while we call getrusage and
1477          * possibly take page faults for user memory.
1478          */
1479         get_task_struct(p);
1480         pid = task_pid_vnr(p);
1481         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1482         read_unlock(&tasklist_lock);
1483
1484         if (unlikely(options & WNOWAIT))
1485                 return wait_noreap_copyout(p, pid, uid,
1486                                            why, exit_code,
1487                                            infop, ru);
1488
1489         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1490         if (!retval && stat_addr)
1491                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1492         if (!retval && infop)
1493                 retval = put_user(SIGCHLD, &infop->si_signo);
1494         if (!retval && infop)
1495                 retval = put_user(0, &infop->si_errno);
1496         if (!retval && infop)
1497                 retval = put_user((short)why, &infop->si_code);
1498         if (!retval && infop)
1499                 retval = put_user(exit_code, &infop->si_status);
1500         if (!retval && infop)
1501                 retval = put_user(pid, &infop->si_pid);
1502         if (!retval && infop)
1503                 retval = put_user(uid, &infop->si_uid);
1504         if (!retval)
1505                 retval = pid;
1506         put_task_struct(p);
1507
1508         BUG_ON(!retval);
1509         return retval;
1510 }
1511
1512 /*
1513  * Handle do_wait work for one task in a live, non-stopped state.
1514  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1515  * the lock and this task is uninteresting.  If we return nonzero, we have
1516  * released the lock and the system call should return.
1517  */
1518 static int wait_task_continued(struct task_struct *p, int options,
1519                                struct siginfo __user *infop,
1520                                int __user *stat_addr, struct rusage __user *ru)
1521 {
1522         int retval;
1523         pid_t pid;
1524         uid_t uid;
1525
1526         if (!unlikely(options & WCONTINUED))
1527                 return 0;
1528
1529         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1530                 return 0;
1531
1532         spin_lock_irq(&p->sighand->siglock);
1533         /* Re-check with the lock held.  */
1534         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1535                 spin_unlock_irq(&p->sighand->siglock);
1536                 return 0;
1537         }
1538         if (!unlikely(options & WNOWAIT))
1539                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1540         spin_unlock_irq(&p->sighand->siglock);
1541
1542         pid = task_pid_vnr(p);
1543         uid = p->uid;
1544         get_task_struct(p);
1545         read_unlock(&tasklist_lock);
1546
1547         if (!infop) {
1548                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1549                 put_task_struct(p);
1550                 if (!retval && stat_addr)
1551                         retval = put_user(0xffff, stat_addr);
1552                 if (!retval)
1553                         retval = pid;
1554         } else {
1555                 retval = wait_noreap_copyout(p, pid, uid,
1556                                              CLD_CONTINUED, SIGCONT,
1557                                              infop, ru);
1558                 BUG_ON(retval == 0);
1559         }
1560
1561         return retval;
1562 }
1563
1564 /*
1565  * Consider @p for a wait by @parent.
1566  *
1567  * -ECHILD should be in *@notask_error before the first call.
1568  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1569  * Returns zero if the search for a child should continue;
1570  * then *@notask_error is 0 if @p is an eligible child,
1571  * or another error from security_task_wait(), or still -ECHILD.
1572  */
1573 static int wait_consider_task(struct task_struct *parent, int ptrace,
1574                               struct task_struct *p, int *notask_error,
1575                               enum pid_type type, struct pid *pid, int options,
1576                               struct siginfo __user *infop,
1577                               int __user *stat_addr, struct rusage __user *ru)
1578 {
1579         int ret = eligible_child(type, pid, options, p);
1580         if (!ret)
1581                 return ret;
1582
1583         if (unlikely(ret < 0)) {
1584                 /*
1585                  * If we have not yet seen any eligible child,
1586                  * then let this error code replace -ECHILD.
1587                  * A permission error will give the user a clue
1588                  * to look for security policy problems, rather
1589                  * than for mysterious wait bugs.
1590                  */
1591                 if (*notask_error)
1592                         *notask_error = ret;
1593         }
1594
1595         if (likely(!ptrace) && unlikely(p->ptrace)) {
1596                 /*
1597                  * This child is hidden by ptrace.
1598                  * We aren't allowed to see it now, but eventually we will.
1599                  */
1600                 *notask_error = 0;
1601                 return 0;
1602         }
1603
1604         if (p->exit_state == EXIT_DEAD)
1605                 return 0;
1606
1607         /*
1608          * We don't reap group leaders with subthreads.
1609          */
1610         if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1611                 return wait_task_zombie(p, options, infop, stat_addr, ru);
1612
1613         /*
1614          * It's stopped or running now, so it might
1615          * later continue, exit, or stop again.
1616          */
1617         *notask_error = 0;
1618
1619         if (task_is_stopped_or_traced(p))
1620                 return wait_task_stopped(ptrace, p, options,
1621                                          infop, stat_addr, ru);
1622
1623         return wait_task_continued(p, options, infop, stat_addr, ru);
1624 }
1625
1626 /*
1627  * Do the work of do_wait() for one thread in the group, @tsk.
1628  *
1629  * -ECHILD should be in *@notask_error before the first call.
1630  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1631  * Returns zero if the search for a child should continue; then
1632  * *@notask_error is 0 if there were any eligible children,
1633  * or another error from security_task_wait(), or still -ECHILD.
1634  */
1635 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1636                           enum pid_type type, struct pid *pid, int options,
1637                           struct siginfo __user *infop, int __user *stat_addr,
1638                           struct rusage __user *ru)
1639 {
1640         struct task_struct *p;
1641
1642         list_for_each_entry(p, &tsk->children, sibling) {
1643                 /*
1644                  * Do not consider detached threads.
1645                  */
1646                 if (!task_detached(p)) {
1647                         int ret = wait_consider_task(tsk, 0, p, notask_error,
1648                                                      type, pid, options,
1649                                                      infop, stat_addr, ru);
1650                         if (ret)
1651                                 return ret;
1652                 }
1653         }
1654
1655         return 0;
1656 }
1657
1658 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1659                           enum pid_type type, struct pid *pid, int options,
1660                           struct siginfo __user *infop, int __user *stat_addr,
1661                           struct rusage __user *ru)
1662 {
1663         struct task_struct *p;
1664
1665         /*
1666          * Traditionally we see ptrace'd stopped tasks regardless of options.
1667          */
1668         options |= WUNTRACED;
1669
1670         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1671                 int ret = wait_consider_task(tsk, 1, p, notask_error,
1672                                              type, pid, options,
1673                                              infop, stat_addr, ru);
1674                 if (ret)
1675                         return ret;
1676         }
1677
1678         return 0;
1679 }
1680
1681 static long do_wait(enum pid_type type, struct pid *pid, int options,
1682                     struct siginfo __user *infop, int __user *stat_addr,
1683                     struct rusage __user *ru)
1684 {
1685         DECLARE_WAITQUEUE(wait, current);
1686         struct task_struct *tsk;
1687         int retval;
1688
1689         trace_sched_process_wait(pid);
1690
1691         add_wait_queue(&current->signal->wait_chldexit,&wait);
1692 repeat:
1693         /*
1694          * If there is nothing that can match our critiera just get out.
1695          * We will clear @retval to zero if we see any child that might later
1696          * match our criteria, even if we are not able to reap it yet.
1697          */
1698         retval = -ECHILD;
1699         if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1700                 goto end;
1701
1702         current->state = TASK_INTERRUPTIBLE;
1703         read_lock(&tasklist_lock);
1704         tsk = current;
1705         do {
1706                 int tsk_result = do_wait_thread(tsk, &retval,
1707                                                 type, pid, options,
1708                                                 infop, stat_addr, ru);
1709                 if (!tsk_result)
1710                         tsk_result = ptrace_do_wait(tsk, &retval,
1711                                                     type, pid, options,
1712                                                     infop, stat_addr, ru);
1713                 if (tsk_result) {
1714                         /*
1715                          * tasklist_lock is unlocked and we have a final result.
1716                          */
1717                         retval = tsk_result;
1718                         goto end;
1719                 }
1720
1721                 if (options & __WNOTHREAD)
1722                         break;
1723                 tsk = next_thread(tsk);
1724                 BUG_ON(tsk->signal != current->signal);
1725         } while (tsk != current);
1726         read_unlock(&tasklist_lock);
1727
1728         if (!retval && !(options & WNOHANG)) {
1729                 retval = -ERESTARTSYS;
1730                 if (!signal_pending(current)) {
1731                         schedule();
1732                         goto repeat;
1733                 }
1734         }
1735
1736 end:
1737         current->state = TASK_RUNNING;
1738         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1739         if (infop) {
1740                 if (retval > 0)
1741                         retval = 0;
1742                 else {
1743                         /*
1744                          * For a WNOHANG return, clear out all the fields
1745                          * we would set so the user can easily tell the
1746                          * difference.
1747                          */
1748                         if (!retval)
1749                                 retval = put_user(0, &infop->si_signo);
1750                         if (!retval)
1751                                 retval = put_user(0, &infop->si_errno);
1752                         if (!retval)
1753                                 retval = put_user(0, &infop->si_code);
1754                         if (!retval)
1755                                 retval = put_user(0, &infop->si_pid);
1756                         if (!retval)
1757                                 retval = put_user(0, &infop->si_uid);
1758                         if (!retval)
1759                                 retval = put_user(0, &infop->si_status);
1760                 }
1761         }
1762         return retval;
1763 }
1764
1765 asmlinkage long sys_waitid(int which, pid_t upid,
1766                            struct siginfo __user *infop, int options,
1767                            struct rusage __user *ru)
1768 {
1769         struct pid *pid = NULL;
1770         enum pid_type type;
1771         long ret;
1772
1773         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1774                 return -EINVAL;
1775         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1776                 return -EINVAL;
1777
1778         switch (which) {
1779         case P_ALL:
1780                 type = PIDTYPE_MAX;
1781                 break;
1782         case P_PID:
1783                 type = PIDTYPE_PID;
1784                 if (upid <= 0)
1785                         return -EINVAL;
1786                 break;
1787         case P_PGID:
1788                 type = PIDTYPE_PGID;
1789                 if (upid <= 0)
1790                         return -EINVAL;
1791                 break;
1792         default:
1793                 return -EINVAL;
1794         }
1795
1796         if (type < PIDTYPE_MAX)
1797                 pid = find_get_pid(upid);
1798         ret = do_wait(type, pid, options, infop, NULL, ru);
1799         put_pid(pid);
1800
1801         /* avoid REGPARM breakage on x86: */
1802         asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1803         return ret;
1804 }
1805
1806 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1807                           int options, struct rusage __user *ru)
1808 {
1809         struct pid *pid = NULL;
1810         enum pid_type type;
1811         long ret;
1812
1813         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1814                         __WNOTHREAD|__WCLONE|__WALL))
1815                 return -EINVAL;
1816
1817         if (upid == -1)
1818                 type = PIDTYPE_MAX;
1819         else if (upid < 0) {
1820                 type = PIDTYPE_PGID;
1821                 pid = find_get_pid(-upid);
1822         } else if (upid == 0) {
1823                 type = PIDTYPE_PGID;
1824                 pid = get_pid(task_pgrp(current));
1825         } else /* upid > 0 */ {
1826                 type = PIDTYPE_PID;
1827                 pid = find_get_pid(upid);
1828         }
1829
1830         ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1831         put_pid(pid);
1832
1833         /* avoid REGPARM breakage on x86: */
1834         asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1835         return ret;
1836 }
1837
1838 #ifdef __ARCH_WANT_SYS_WAITPID
1839
1840 /*
1841  * sys_waitpid() remains for compatibility. waitpid() should be
1842  * implemented by calling sys_wait4() from libc.a.
1843  */
1844 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1845 {
1846         return sys_wait4(pid, stat_addr, options, NULL);
1847 }
1848
1849 #endif