pids: de_thread: don't clear session/pgrp pids for the old leader

Based on Eric W. Biederman's idea.

Unless task == current, without tasklist_lock held task_session()/task_pgrp()
can return NULL if the caller races with de_thread() which switches the group
leader.

Change transfer_pid() to not clear old->pids[type].pid for the old leader.
This means that its .pid can point to "nowhere", but this is already true for
sub-threads, and the old leader is not group_leader() any longer.  IOW, with
or without this change we can't trust task's special pids unless it is the
group leader.

With this change the following code

	rcu_read_lock();
	task = find_task_by_xxx();
	do_something(task_pgrp(task), task_session(task));
	rcu_read_unlock();

can't race with exec and hit the NULL pid.
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc:  "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kernel/pid.c

@@ -354,7 +354,6 @@ void transfer_pid(struct task_struct *old, struct task_struct *new,
 {
 	new->pids[type].pid = old->pids[type].pid;
 	hlist_replace_rcu(&old->pids[type].node, &new->pids[type].node);
-	old->pids[type].pid = NULL;
 }
 
 struct task_struct *pid_task(struct pid *pid, enum pid_type type)