Most state-of-the-art approaches of securing XML documents are based on a partial annotation of an XML tree with security labels which are later propagated to unlabeled nodes of the XML so that the resulting labeling is full (i.e. defined for every XML node). The first contribution of this paper is an investigation of possible alternatives for policy definition that lead to a fully annotated XML. We provide a classification of policies using different options of security label propagation and conflict resolution. Our second contribution is a generalized algorithm that constructs a full DTD annotation (from the the partial one) w.r.t. the policy classification. Finally, we discuss the query rewriting approach for our model of XML security views.