XML-Signature XPath Filter 2.0

XML Signature [XML-DSig] recommends a standard means for specifying information content to be digitally signed and for representing the resulting digital signatures in XML. Some applications require the ability to specify a subset of a given XML document as the information content to be signed. The XML Signature specification meets this requirement with the XPath transform. However, this transform can be difficult to implement efficiently with existing technologies. This specification defines a new XML Signature transform to facilitate the development of efficient document subsetting technologies that interoperate under similar performance profiles.

Status of this document

1. Introduction

The XML Recommendation [XML] specifies the syntax of a class of objects called XML documents. The Namespaces in XML Recommendation [XML-NS] specifies additional syntax and semantics for XML documents. The XML Signature Recommendation [XML-DSig] defines standard means for specifying information content to be digitally signed, including the ability to select a portion of an XML document to be signed using an XPath transform.

This specification describes a new signature filter transform that, like the XPath transform [XML-DSig, section 6.6.3], provides a method for computing a portion of a document to be signed. In the interest of simplifying the creation of efficient implementations, the architecture of this transform is not based on evaluating an XPath expression for every node of the XML parse tree (as defined by the XPath data model [XPath]). Instead, a sequence of XPath expressions are used to select the roots of document subtrees - location sets, in the language of XPointer - which are combined using set intersection, subtraction and union, and then used to filter the input node set. The principal differences from the XPath transform are:

As with the original XPath transform, the primary purpose of this transform is to ensure that only specifically defined changes to the input XML document are permitted after the signature is affixed. This can be done by excluding precisely those nodes that are allowed to change once the signature is affixed, and including all other input nodes in the output. It is the responsibility of the signature filter transform author to ensure that nodes are not excluded which could affect the interpretation of the transform output in the application context.

Consider the motivating scenario where an application wishes to affix two enveloped signatures signatures to the document; any other change to the document must cause the signatures to be invalid. When the application creates the first signature that signature is automatically omitted from its own digest calculations. However, it will also be necessary to exclude the subsequent (second) signature element from the digest calculations of the first signature. This specification can be used to efficiently satisfy this requirement using the set subtraction operation.

This transform also supports the ability to specify a set of nodes that will be included in a signature, with all non-specified nodes being excluded. This formulation is useful for isolating a portion of a document, such as a chapter of a document, or a payload in a protocol message, and can be expressed using the set intersection operation.

Complete familiarity with the first XML Signature XPath Transform [XML-DSig, section 6.6.3] is required.

2. Terminology

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [Keywords].

The XPath 1.0 Recommendation [XPath] defines the term node-set as "(an unordered collection of nodes without duplicates)" and specifies a data model for representing an input XML document as a set of nodes of various types (element, attribute, namespace, text, comment, processing instruction, and root).

An input document is the document that contains all the nodes available to processing by this transform. A document subset is a portion of an XML document indicated by an XPath node-set, which may not include all of the nodes in the document. For example, the input node set is a collection of XPath nodes from the input document that is passed as a parameter to this transform. A subtree rooted by a given node is a document subset containing the given node and every node having the given node as an ancestor. Subtree expansion is the process of expanding a node set to include all subtrees rooted at any node in the node set. For example, the subtree expansion of a node set consisting of just a single element node would be a node set containing that element, its attribute nodes, namespace nodes, and all its descendants; including the attribute nodes and namespaces nodes of any descendant elements.

The XML Signature Recommendation [XML-DSig] defines a reference as a sequence of steps performed to obtain an octet stream to be digitally signed. A transform is an identified algorithm to be used as a step in the reference processing model. A transform takes an octet stream or XPath node-set as input, and it produces an octet stream or XPath node-set as output (the reference processing model automatically converts the final output to an octet stream if it is an XPath node-set).

3. Specification of Signature Filter Transform

The transform operates by computing a node set that is used to filter the input node set: The output node set consists of only those nodes in both the input node set and the filter node set. In other words, the output node set is the intersection of the input node set and the computed filter node set.

The filter node set is computed by evaluating a sequence of XPath expressions and combining their results. A node set is initially computed containing the entire input document. In sequence, each XPath expression is then evaluated, subtree-expanded, and then used to transform the filter node set according to a specified set operation; intersection, subtraction, or union. After all XPaths have been applied, the resulting node set is used as the filter node set.

3.1 Algorithm Identifier

The XML Signature Recommendation [XML-DSig] uses a URI [URI] to identify each algorithm to be performed when creating or validating a signature. The signature filter transform is identified as follows:

3.2 Syntax of Signature Filter Transform

The signature filter transform shall be represented by a sequence of one or more elements named XPath. The content of XPath is character data containing an XPath expression. The XPath has an attribute named Filter whose possible values are intersect, subtract, and union. The Filter attribute indicates the set operation that is performed with the resulting node set when computing the filter node set. The following is an example of markup for a signature filter that signs the entire input node set except for elements with identifier foo and bar (and all nodes with one of those elements as an ancestor):

3.3 Input and Evaluation Context of Signature Filter Transform

The input required by this transform is an XPath node-set over the input document. If the input document is an octet stream, then the application MUST convert the octet stream to an XPath node-set that contains all of the document nodes (including comment nodes). The evaluation context for the XPath expressions in the filter transform will be:

The here() function returns a node-set containing the attribute or processing instruction node or the parent element of the text node that directly bears the XPath expression. In this transform, this will be the XPath element. This expression results in an error if the containing XPath expression does not appear in the same XML document against which the XPath expression is being evaluated.

3.4 Processing Model of Signature Filter Transform

Using the aforementioned evaluation context, the signature filter transform evaluates the XPath expressions appearing in the character content of the XPath elements and uses these to compute a filter node set F, which is then used to filter the input node set I resulting in an output node set O:

Initialize the filter node set F to consist of all nodes in the input document.
Iterate through each XPath expression, X, in sequence, and update the filter node set F as follows:

Evaluate the XPath expression X. The result is a node set S.
Compute the set S' consisting of all nodes in the input document that are either present in S or that have an ancestor in S. This is equal to the union of all the document subtrees rooted by a node in S.
If the Filter attribute value is intersect, then compute the intersection of the selected subtrees, S', with the filter node set F. The result will include only those nodes that are in both the filter node set and the selected subtrees: F' = F INTERSECT S'
If the Filter attribute value is subtract, then compute the subtraction of the selected subtrees, S', from the filter node set F. The result will include only those nodes that are in the filter node set, but not the selected subtrees: F' = F - S'
Otherwise, if the Filter attribute value is union, then compute the union the selected subtrees, S', with the filter node set F. The result will include all those nodes that are in either the filter node set, the selected subtrees, or both: F' = F UNION S'
Update the filter node set F to be the new node set F'.

Finally, after applying all the XPath expressions, compute the output node set O to be the intersection of the computed filter node set, F, with the input node set, I. The result will include all nodes from the input node set that are also in the filter node set: O = I INTERSECT F
An empty input node set will always result in an empty output node set.

In this processing model, the conversion from a subtree interpretation of the XPath expressions to a node-set containing all nodes that must be used during the set operation, along with actual performance of the set operation, is described explicitly. Implementors SHOULD observe that it is possible to compute the effective result of this operation in a single pass through the input document without performing subtree expansion or any set operations:

For each XPath expression X, in sequence, evaluate the expression and store the resulting node set, S, along with the associated set operation.
Prepend a node set consisting of just the document node, along with the operation union.
Create a new, empty filter node set.
Iterate through the input document in document order, adding each node that is encountered to the filter node set F if a flag Z is true. This flag is computed as follows:

Any time a node is encountered that is in any evaluated node set S, update Z to be true if and only if the node is present in any subtree-expanded union node set and all subsequent subtree-expanded intersect node sets but no subsequent subtree-expanded subtract node sets, or false otherwise. If there are no subsequent intersect or subtract node sets, then that part of the test is automatically passed.
Presence in a subtree-expanded node set can be efficiently determined without actually expanding the node set, by simply maintaining a stack or count that identifies whether any nodes from that node set are an ancestor of the node being processed.
The initial value of Z is irrelevant because it will be automatically computed when the first document node is processed.

Implementers MAY further observe that, if this transform is followed by a canonicalization operation, the described filter computation can be efficiently commingled with the document-order canonicalization processing.

4. Examples of Signature Filter Transform

The example below illustrates one way to create an enveloped signature with the signature filter transform. The function here() identifies the XPath element, and the subsequent location path obtains the nearest ancestor Signature element. Due to the subtract value of the Filter attribute, the output of the signature filter transform is a node-set containing every node from the input node set except the nodes in the subtree rooted by the Signature element containing the example signature filter transform below.

A suitable signature reference URI to use with this subtract filter would be URI="" (the entire signature document, without comments), URI="#xpointer(/)" (the entire signature document, with comments) or any same-document reference that includes the signature itself.

An example of an intersect filter is a signature that co-signs another signature. In this example, a Signature element identified by PrimaryBorrowSig must be signed. The XPath expression obtains the element node, and the transform expands the output node-set to contain all nodes from the input node set that are also in the subtree rooted by the element node.

   <XPath Filter="intersect"
    xmlns="http://www.w3.org/2002/06/xmldsig-filter2">
      id("PrimaryBorrowerSig")
   </XPath>

This type of intersect filter is useful for efficiently signing subsets of a document, whether this is the same document as the signature or an external document. For example, if the signature reference URI is URI="document.xml", then this document will be automatically parsed and just the identified element and its descendants will be signed.

Union filters, by themselves are of no particular use: The initial filter node set consists of the entire input document; any union with this will have no effect, so the output of the transform will be identical to the input. The union operation is intended to follow a subtract operation, to allow a subtree to be removed, with the exception of a lower subtree which is still included in the output.

Consider the following document which contains a same-document enveloped signature reference with an XPath filter containing three XPath operations:

   <Document>
     <ToBeSigned>
       <!-- comment -->
       <Data />
       <NotToBeSigned>
         <ReallyToBeSigned>
           <!-- comment -->
           <Data />
         </ReallyToBeSigned>
       </NotToBeSigned>
     </ToBeSigned>
     <ToBeSigned>
       <Data />
       <NotToBeSigned>
         <Data />
       </NotToBeSigned>
     </ToBeSigned>
     <dsig:Signature
      xmlns:dsig="http://www.w3.org/2000/09/xmldsig#"
      xmlns:dsig-xpath="http://www.w3.org/2002/06/xmldsig-filter2">
       <dsig:SignedInfo>
         ...
         <dsig:Reference URI="">
           <dsig:Transforms>
             <dsig:Transform 
              Algorithm="http://www.w3.org/2002/06/xmldsig-filter2">
               <dsig-xpath:XPath Filter="intersect"> //ToBeSigned </dsig-xpath:XPath>
               <dsig-xpath:XPath Filter="subtract"> //NotToBeSigned </dsig-xpath:XPath>
               <dsig-xpath:XPath Filter="union"> //ReallyToBeSigned </dsig-xpath:XPath>
             </dsig:Transform>
           </dsig:Transforms>
           ...
         </dsig:Reference>
       </dsig:SignedInfo>
       ...
     </dsig:Signature>
   </Document>

The intersect operation computes the intersection of the XPath-selected subtrees with the filter node set. In this case, the filter node set initially contains the entire input document, and the XPath expression evaluates to the two ToBeSigned elements; these are expanded to include all their descendents and intersected with the filter node set, resulting in the following:

     <ToBeSigned>
       <!-- comment -->
       <Data />
       <NotToBeSigned>
         <ReallyToBeSigned>
           <!-- comment -->
           <Data />
         </ReallyToBeSigned>
       </NotToBeSigned>
     </ToBeSigned><ToBeSigned>
       <Data />
       <NotToBeSigned>
         <Data />
       </NotToBeSigned>
     </ToBeSigned>

The subtract filter computes the subtraction of the XPath-selected subtrees from the filter node set. In this case, the XPath expression evaluates to the two NotToBeSigned elements; these are expanded to include all their descendents and subtracted from the filter node set:

   <ToBeSigned>
       <!-- comment -->
       <Data />
       
     </ToBeSigned><ToBeSigned>
       <Data />
      
   </ToBeSigned>

Next, the union filter computes the union of the XPath-selected subtrees with the filter node set. In this case, the XPath expression evaluates to the ReallyToBeSigned element; this is expanded to include all its descendents and added to the filter node set:

   <ToBeSigned>
       <!-- comment -->
       <Data />
       <ReallyToBeSigned>
           <!-- comment -->
           <Data />
         </ReallyToBeSigned>
     </ToBeSigned><ToBeSigned>
       <Data />
      
   </ToBeSigned>

Finally, this resulting filter node set is used to transform the input node set. In this example, the input node set is the entire document, with comments removed. The transformed node set will thus be all those nodes from the input document, less comments, that are also in the filter node set:

   <ToBeSigned>
       
       <Data />
       <ReallyToBeSigned>
           
           <Data />
         </ReallyToBeSigned>
     </ToBeSigned><ToBeSigned>
       <Data />
      
   </ToBeSigned>

Note that the result contains no nodes that were not in the input node set. Although the filter node set included comments, these were not present in the input node set so they are not present in the output node set.

This signature filter does not provide any increased capability over the original XPath transform. For example, this reference could be replicated using the XPath transform as follows.

   <dsig:Reference URI="">
     <dsig:Transforms>
       <dsig:Transform 
        Algorithm="http://www.w3.org/TR/1999/REC-xpath-19991116">
         <dsig:XPath>
           (ancestor-or-self::ToBeSigned and
            not (ancestor-or-self::NotToBeSigned))
           or ancestor-or-self::ReallyToBeSigned
         </dsig:XPath>
       </dsig:Transform>
     </dsig:Transforms>
     ...
   </dsig:Reference>

The advantage of the signature filter transform over the XPath transform is that the latter requires evaluation of a potentially-complex expression against every node in the input set, which has proved costly in practice for many useful operations, whereas the former requires evaluation of simple XPath expressions and then the execution of some basic set operations or their equivalent, which can be implemented significantly more efficiently.

5. References

6. Acknowledgements (Informative)

The following people provided valuable feedback that improved the quality of this specification: