[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [Elist Home]
Subject: [xacml] Updated list of Change Requests
I suggest that requests that have received heavy discussion on the list should be re-submitted in their entirety, incorporating any corrections and comments. Title: Change Requests Author: Anne Anderson Version: 1.3, 02/09/06 (yy/mm/dd) Original Source: /net/labeast.east/files2/east/info/projects/isrg/xacml/docs/SCCS/s.ChangeRequests.txt 0001. [Daniel] Add AttributeValue element to support sequence-of values http://lists.oasis-open.org/archives/xacml/200208/msg00047.html http://lists.oasis-open.org/archives/xacml/200208/msg00140.html STATUS: POSTPONED (to discussion of functions #23) Currently <attributeValue> element is an extension of anyType, with specified DataType. XACML functions arguments are, in general, a sequence of typed elements - as it can be returned by a designator from context. While most accept only single values (sequence of size 1), some set operation functions have an arbitrary length sequence as an argument. We need to provide syntax to express a sequence of values as a literal argument in set operations (like union, member_of, intersection..) I propose adding element Value, of anyType, and defining <AttributeValueType> as sequence of <Value>, with DataType attribute, applicable to all elements in the sequence. It supports 0 size of the sequence, to support empty set in operations <xs:element name="Value" type="xs:anyType"/> <xs:element name="AttributeValue"/> <xs:complexType name="AttributeValueType"> <xs:sequence minOccurs="0" maxOccurs="unbounded"> <xs:element ref="Value"/> </xs:sequence> <xs:attribute name="name" type="xs:anyURI" use="required"/> </xs:complexType> ------------------------------------- >2. Adding a separate construct to XACML policy schema to represent >sequences of attribute values, such that you cannot create sequeences of >sequences. As an alternative to my proposal to modify schema element <attributevalue>, I propose to define standard core function collection <type>-sequence of undefined arity, that takes arguments of singleton <type> kind and return a sequence<type> value, a sequence of its argument values. In this case, when a literal sequence definition is needed, this funciton can be used. No schema modification is needed. The drawback is that it can not be used to specify an argument of sequence<type> kind for predicate functions that can be used in <Match> elements. 0002. [Anne] Add mandatory action-id attribute STATUS: APPROVED 8/29 (QUORUM) Create a reserved attribute identifier "urn:oasis:names:tc:xacml:1.0:action:action-id". Make inclusion of an <Attribute> with AttributeId of this identifier mandatory under the <Action> element of the <Request> context. Change minOccurs for <Attribute> under <Action> 1. Rationale: We had previously decided that <Action> would have a single string value that would be the action id. Now we need a specific AttributeId for this. This is consistent with the way resource-id is handled. It provides a consistent, interoperable way of specifying the action. The <DataType> of the <Attribute> can specify whether the action value is a string or URI. 0003. [Anne] Add optional action-namespace attribute STATUS: APPROVED 8/29 (QUORUM) Create a reserved attribute identifier "urn:oasis:names:tc:xacml:1.0:action:action-namespace". Make inclusion of an <Attribute> with AttributeId of this identifier optional under the <Action> element of the <Request> context. Rationale: We previously decided that an <Action> value might be associated with a specific namespace, and that an XML attribute was needed to express this. 0004. [Anne] Add optional action:implied-action identifier STATUS: APPROVED 8/29 (QUORUM) Create a reserved identifier "urn:oasis:names:tc:xacml:1.0:action:implied-action" to represent the value of an action that is implied by the <Resource> Rationale: We have agreed to this concept, but have not reserved an identifier for it. 0005. [Anne] Change <Result> ResourceURI xml attribute to ResourceId STATUS: APPROVED 8/29 (QUORUM) Rationale: Since the Request <Resource> identifier is now called resource-id, and can be of any data type, the <Result> should be consistent. 0006. [Anne] Add missing-attribute identifier for StatusCode STATUS: APPROVED 8/29 (QUORUM) Rationale: We have values for ok, processing-error, and syntax-error. Although we discussed the use case for missing attributes extensively, we have not defined a standard identifier for this status. 0007. [Anne] Make context Resource Attribute minoccurs=1 STATUS: REJECTED 9/5 (NO QUORUM, BUT ANNE CONCURS) Current value is minOccurs=0 maxOccurs=unbounded. Change this to minOccurs=1 maxOccurs=unbounded. Rationale: Since Resource MUST contain a resource-id attribute, minimum value should be 1. 0008. [Anne] list mandatory vs. non-mandatory functions http://lists.oasis-open.org/archives/xacml/200208/msg00069.html STATUS: POSTPONED Add a list of all functions, with an indicator as to whether the function is mandatory to implement, to the Conformance Section. This list should be updated based on the final resolution of CR#23. Rationale: I forgot to include these when I created the original list. Text to be inserted: [complete list of functions in e-mail] 0009. [Daniel] Function naming convention http://lists.oasis-open.org/archives/xacml/200208/msg00071.html STATUS: REJECTED 8/29 (QUORUM) Change function names to blah_blah Rationale: Dashes are not always supported. 0010. [Anne] allow more than two arguments to "add" http://lists.oasis-open.org/archives/xacml/200208/msg00090.html STATUS: APPROVED 8/29 (QUORUM) Allow function:integer-add and function:decimal-add to take more than two arguments. Rationale: This saves having to write nested statements for these simple operations. 0011. [Carlisle] state *Match is matched against AttributeValue http://lists.oasis-open.org/archives/xacml/200208/msg00094.html STATUS: APPROVED 8/29 (QUORUM) In section 5.7 (Element <SubjectMatch>), I think we need to specify that whatever is returned by SubjectAttributeDesignator or AttributeSelector is matched against the value carried in AttributeValue. The same is true for <ResourceMatch> and <ActionMatch> (sections 5.10 and 5.13). Rationale: This behaviour is not spelled out anywhere. (In fact, AttributeValue is not even described in the "list of elements contained in <SubjectMatch>" discussion; it only appears in the schema fragment.) 0012. [Carlisle] <AttributeSelector> in <SubjectMatch> should be [optional] http://lists.oasis-open.org/archives/xacml/200208/msg00094.html STATUS: APPROVED 8/29 (QUORUM) Make <AttributeSelector> in <SubjectMatch> [optional] rather than [required]. Rationale: Also, in the "list of elements contained in <SubjectMatch>" discussion, why is <AttributeSelector> described as [required]? I thought it was optional. 0013. [Carlisle] <SubjectMatch> in <SubjectAttributeDesignatorWhere> min/max http://lists.oasis-open.org/archives/xacml/200208/msg00094.html STATUS: APPROVED 9/5 (NO QUORUM) Make minOccurs="1" and maxOccurs="unbounded" for <SubjectMatch> in <SubjectAttributeDesignatorWhere>. Rationale: In section 5.26 (Element <SubjectAttributeDesignatorWhere>), why does the element <SubjectMatch> have minOccurs="0"? If this element was omitted, you wouldn't use SubjectAttributeDesignatorWhere; you would just use SubjectAttributeDesignator. In any case, the verbal description says that SubjectMatch has [Any Number], so I suspect that minOccurs="0" in the schema should be changed to maxOccurs="unbounded". 0014. [Carlisle] 10. "Security and Privacy Considerations" http://lists.oasis-open.org/archives/xacml/200208/msg00096.html STATUS: APPROVED 9/5 (NO QUORUM) Change title of Section 10 to "Security and Privacy Considerations". Rationale: Should this be called "Security and Privacy Considerations" instead of just "Security and Privacy"? 0015. [Carlisle] 10. "Statement Level Confidentiality": rules/policies http://lists.oasis-open.org/archives/xacml/200208/msg00096.html STATUS: NOT YET CONSIDERED In the 10. "Statement Level Confidentiality" section, 1st paragraph, say "... a PRP only needs access to the target elements in order to find the appropriate policies". Rationale: Currently says: "... a PRP only needs access to the target elements in order to find the appropriate rules". Should this say "rules/policies", or just "policies", instead of "rules"? 0016. [Carlisle] 10. "Policy Integrity": rule/policy http://lists.oasis-open.org/archives/xacml/200208/msg00096.html STATUS: NOT YET CONSIDERED In the 10. "Policy Integrity" section, 4th paragraph, say "The PDP SHOULD NOT request a policy based on who signed the policy...". Rationale: In the "Policy Integrity" section, 4th paragraph, it currently says "The PDP SHOULD NOT request a rule based on who signed the rule...". Should both occurrences of "rule" be "policy"? 0017. [Carlisle] 10. "Resource Matching": NotApplicable treated as Permit http://lists.oasis-open.org/archives/xacml/200208/msg00096.html STATUS: NOT YET CONSIDERED In 10. "Resource Matching" section, say, 'although the policy result of "Not Applicable" is treated as equivalent to "Permit" by many web servers, this behavior is not recommended.' Rationale: In the "Resource Matching" section, 1st paragraph, it says "... the policy result of "Not Applicable" is treated as equivalent to "Permit" as is common in many web servers". I'm a bit surprised that this is true (although I probably shouldn't be!). In any case, we probably don't want to encourage this behaviour. Should we simply not mention this, or should we at least say that this behaviour is not recommended? Hal's rejoinder: First of all, this is intended as a cautionary example. However, this is a real world problem that affects my product, and all of our compeitors. I can show you security holes posted to BUGTRAQ, for example, where this occurred. We have struggled with it for years. The Not Applicable == Permit behavior is not only standard for web servers, it is embedded in the Java Servlet and JSP specifications. The logic is that only a small portion of the resources on a public web server are protected, so people want to only specify rules for those that are. If the namespace is not conveniencely laid out (as most are not) it may require a number of rules just to specify all the pages that are not protected. 0018. [Carlisle] C.3 use "First Applicable" http://lists.oasis-open.org/archives/xacml/200208/msg00099.html STATUS: NOT YET CONSIDERED Call Section C.3, "First Applicable" rather than "First Applicable Rule Combining Algorithm". In first sentence of C.3, change "First Determinate" rule-combining algorithm to "First Applicable". Rationale: Section C.3, should this be called "First Applicable" rather than "First Applicable Rule Combining Algorithm"? This would be more consistent with both C.1 and C.2, and would also be more consistent with the fact that C.3 contains both a rule combining algorithm and a policy combining algorithm. The first sentence of C.3 calls it the "First Determinate" rule-combining algorithm; this should be changed to "First Applicable". 0019. [Carlisle] C. consistent error behavior http://lists.oasis-open.org/archives/xacml/200208/msg00099.html STATUS: NOT YET CONSIDERED Policy combiner should halt and return Indeterminate when an error is encountered. Rationale: The behaviour specified in the policy-combining algorithm when an error is encountered is different from the behaviour specified in the rule-combining algorithm when an error is encountered (the rule combiner says halt and return Indeterminate, whereas the policy combiner says to keep looking for an applicable policy). Is this what we wanted? More importantly, might the policy combiner behaviour not lead to different answers for the same inputs? For example, say there are two policies that are to be combined using this algorithm. Given a particular set of input values, the first policy would return a decision of "Permit" and the second policy would return a decision of "Deny". Now we give all the inputs to two different PDPs. The first PDP retrieves the first policy, gets an answer of "Permit", and returns this to the PEP. The second PDP has trouble retrieving the first policy for whatever reason and, according to the combining algorithm, retrieves the second policy; it then returns a "Deny" to the PEP. Isn't this the sort of result we want to avoid? Wouldn't the behaviour specified in the rule combining algorithm be preferable (that way, the first PDP would return "Permit" and the second would return "Indeterminate", which seems fine to me)? 0020. [Michiharu] B.10: Resource Attributes scope value http://lists.oasis-open.org/archives/xacml/200208/msg00112.html STATUS: NOT YET CONSIDERED In B.10, the identifier indicates the scope of the request with regard to the resource. When this attribute is specified in the request, the value MUST be either 'Immediate', 'Children', or 'Descendant'. 0021. [Michiharu] 8.1 Operational Model: PDP description change http://lists.oasis-open.org/archives/xacml/200208/msg00112.html STATUS: NOT YET CONSIDERED 8.1 Policy Decision Point (PDP) Given a valid XACML "Policy" or a "PolicySet", a compliant XACML PDP MUST evaluate that statement in accordance to the semantics specified in Section 4,5, and 6 when applied to a specific input context. The PDP MUST return an output context, with one value of "Permit", "Deny", "Indeterminate", or "NotApplicable". If a permit is returned, the PEP permit access to the requested resource. If a denial is returned, the PEP denies access to the requested resource. If a permit with one or more obligations is returned, the PEP permits access provided that every obligations are fulfilled successfully. If a denial with one or more obligations is returned, the PEP denies access but still fulfills the obligations. In each case, when fulfilling obligations failed, the PEP SHOULD raise an error. How the error is raised is out of the scope of XACML. In any case, the PDP can return additional information in the status code element in the response context. For 'Permit' decision, it MAY specify which rules are used in decision making. If an indeterminate is returned, it means that the PDP could not make decision due to some reason. The PDP MAY return decision of "indeterminate" with a status code of "urn:oasis:names:tc:xacml:1.0:missing-attribute", signifying that more information is needed. In this case, the decision MAY list the names of any attributes of the subject and the resource that are needed by the PDP to refine its decision. A PEP MAY resubmit a refined request context in response to a decision of "indeterminate" with a status code of "missing-attribute" by adding attribute values for the attribute names that are listed in the response. When the PDP returns an decision of "indeterminate", with a status code of "missing-attribute", a PDP MUST NOT list the names of any attribute of the subject or the resource of the request for which values were already supplied in the request. Note, this requirement forces the PDP to eventually return a decision of "permit", "deny", or "indeterminate" with some other reason, in response to successively-refined requests. If not applicable is returned, it means that the PDP's policy does not cover the request, implying that the PEP should ask another PDP. XACML does not assume how top-level XACML policies should be configured. For example, a top-level policy might be a 'Policy' element containing a target element that matches every request, or it might be a 'Policy' element containing a target element that matches only a specific subject. 0022. [Michiharu] 8.1 Operational Model: New sections http://lists.oasis-open.org/archives/xacml/200208/msg00112.html STATUS: NOT YET CONSIDERED 8.2 Hierarchical Resource It is often the case that a target resource is organized as a hierarchy (e.g. file system, XML document). Some applications may require access to an entire subtree of the resource. XACML allows the PEP (or Context Handler) to specify whether the access is just for a single resource or for a subtree below the specified resource. The latter is equivalent to repeating a single request for the entire subtree. When a request context contains a resource attribute of 'urn:oasis:names:tc:xacml:1.0:resource:scope' with a value of 'Immediate', or does not contain that attribute in the context, then it means that the access is just for a single resource specified by 'ResourceId' attribute. When 'urn:oasis:names:tc:xacml:1.0:resource:scope' attribute specifies a value of 'Children', it means that the access is for both a specified resource and its children resources. When 'urn:oasis:names:tc:xacml:1.0:resource:scope' attribute specifies a value of 'Descendant', it means that the access is for both a specified resource and all the descendant resources. In the case of 'Children' and 'Descendant', the access decision may include multiple results for the multiple resources. XACML response can contain multiple result elements. In such case, the status element SHOULD be included only in the first result element (the remaining result elements SHOULD NOT include the status element). Note that the method how PDP finds out whether the resource is hierarchically organized or not is out of the scope of the XACML. 8.3 Propagation through Data Hierarchy When the resource is hierarchically organized, it is often the case that an access control rule associated to a certain node propagates down to the descendant nodes. The XACML core rule combining algorithm does not support such propagation with regard to access control rules. Policy writers who need propagation MUST implement their own local algorithm and specify that algorithm ID in RuleCombiningAlgId in policy element. 0023. [Daniel, Polar] function names, types, semantics http://lists.oasis-open.org/archives/xacml/200208/msg00181.html STATUS: NOT YET CONSIDERED Daniel and Polar have developed a list of functions and their semantics, attached to the above e-mail. (XACML_functions0.8.DOC) CR#8 should be updated based on the final list of functions to show which are mandatory-to-implement. 0024. [Anne] Make "gregorian" functions non-mandatory http://lists.oasis-open.org/archives/xacml/200208/msg00137.html STATUS: NOT YET CONSIDERED Text Changes: Section 11.1.4(?) [Conformance] Functions Remove "M" from left column of urn:oasis:names:tc:xacml:1.0:function:gregorian-equal Rationale: We already support xs:date, so support for Gregorian is icing on the cake. Also, I haven't found a standard for "gregorian", so it is hard to develop tests for it. 0025. [Anne] remove NOTATION from supported types; remove assoc. functions http://lists.oasis-open.org/archives/xacml/200208/msg00138.html STATUS: NOT YET CONSIDERED Text Changes: Section 11. Conformance, DataTypes, remove: line listing "xs:NOTATION" Appendix A.1. Functions, remove: function:NOTATION-equal function:NOTATION-not-equal Rationale: XML Schema says "It is an error for NOTATION to be used directly in a schema. Only datatypes that are derived from NOTATION by specifying a value for enumeration can be used in a schema. For compatibility NOTATION should be used only on attributes." Also, QName-equal and QName-not-equal can be used to compare such data types if necessary. I'm on thin ground here, since I don't know enough about XML Schema to be sure I am interpreting the impact of this correctly. 0026. [Michiharu] Add XPath functions as non-mandatory http://lists.oasis-open.org/archives/xacml/200208/msg00142.html STATUS: NOT YET CONSIDERED [See referenced e-mail for function definitions] I propose to include the following functions (all relevant to XPath) in the spec as non-mandatory to implement functions. The original proposal (a little different from this proposal) was posted on 29th July and it described more in detail. http://lists.oasis-open.org/archives/xacml/200207/msg00162.html function:general-string-equal xs:boolean object(*1) object A = B function:node-boolean xs:boolean object function:node-match xs:boolean object object ======================== (*1) "object" type is defined in Introduction section of [2] as: "... object, which has one of the following four basic types: node-set (an unordered collection of nodes without duplicates) boolean (true or false) number (a floating-point number) string (a sequence of UCS characters)" [1] XQuery 1.0 and XPath 2.0 Data Model, http://www.w3.org/TR/xquery-operators/ [2] XPath 1.0, http://www.w3.org/TR/xpath 0027. [Michiharu] Change resource-uri to resource-id http://lists.oasis-open.org/archives/xacml/200208/msg00143.html (A1) STATUS: NOT YET CONSIDERED urn:oasis:names:tc:xacml:1.0:resource:resource-uri ==> urn:oasis:names:tc:xacml:1.0:resource:resource-id 0028. [Michiharu] Add identifier for resource:syntax http://lists.oasis-open.org/archives/xacml/200208/msg00143.html (B1) STATUS: NOT YET CONSIDERED urn:oasis:names:tc:xacml:1.0:resource:syntax This identifier indicates the syntax of the resource-id (urn:oasis:names:tc:xacml:1.0:resource:resource-id). The valid values are either 'String', 'URI' or 'XPointer'. If this attribute is omitted, 'String' syntax is assumed. 'URI' indicates that the syntax of the resource-id is URI format. 'XPointer' indicates that the syntax of the resource-id is XPointer format proposed in http://www.w3.org/TR/2001/CR-xptr-20010911/. 0029. [Michiharu] Add identifier for resource:scope http://lists.oasis-open.org/archives/xacml/200208/msg00143.html (B2) STATUS: NOT YET CONSIDERED urn:oasis:names:tc:xacml:1.0:resource:scope This identifier indicates the scope of the resource. The valid values are either 'Immediate', 'Children', or 'Descendant'. When 'Immediate', it indicates that the scope of the resource is a target value specified by urn:oasis:names:tc:xacml:1.0:resource:resource-id attribute. When 'Children', it indicates that the scope of the resource is a target value specified by urn:oasis:names:tc:xacml:1.0:resource:resource-id attribute and its children resource. When 'Descendant', it indicates that the scope of the resource is a target value specified by urn:oasis:names:tc:xacml:1.0:resource:resource-id attribute and all the descendant resources. 0030. [Michiharu] Add identifier for resource:target-namespace http://lists.oasis-open.org/archives/xacml/200208/msg00143.html (B3) STATUS: NOT YET CONSIDERED urn:oasis:names:tc:xacml:1.0:resource:target-namespace This identifier indicates the target-namespace of the requested XML document, that is a namespace URI associated with the root element of the XML document. 0031. [Michiharu] Schema change of Defaults element http://lists.oasis-open.org/archives/xacml/200208/msg00144.html STATUS: NOT YET CONSIDERED I propose to change the XPathVersion element and to add a new "XPathNamespace" element as follows: (1) Schema change request <xs:element name="XPathVersion" type="xs:string" substitutionGroup ="xacml:AbstractDefaults"/> ==> <xs:element name="XPathVersion" type="xs:anyURI" substitutionGroup ="xacml:AbstractDefaults"/> (2) Schema addition request See #0032 for change to XPathNamespace schema. (3) Text change request In Section 5.3, Element <PolicySetDefaults>, line 1487-1489, <AbstractDefaults>[Any Number] This is the head of substitution group to specify default parameters. The elements in this substitution group defined at this time are <XPathVersion> element and <XPathNamespace> element. 0032. [Michiharu] Schema change of AttributeSelector http://lists.oasis-open.org/archives/xacml/200208/msg00145.html http://lists.oasis-open.org/archives/xacml/200208/msg00146.html (example) http://lists.oasis-open.org/archives/xacml/200208/msg00149.html (correction) http://lists.oasis-open.org/archives/xacml/200208/msg00150.html (Simon) STATUS: NOT YET CONSIDERED. NOTE: we need a current statement of the complete proposal, taking into account comments and corrections to date. I propose to change the AttributeSelector element. The reason of this change is described in a separate mail titled "[xacml] AttributeSelector example". <xs:complexType name="AttributeSelectorType"> <xs:attribute name="RequestContextPath" type="xs:anyURI" use="required"/> <xs:attribute name="DataType" type="xs:anyURI" use="required"/> <xs:attribute name="XPathVersion" type="xs:anyURI" use="optional" default ="http://www.w3.org/TR/1999/Rec-xpath-19991116"/> </xs:complexType> ==> <xs:complexType name="AttributeSelectorType"> <xs:element ref="xacml:XPathNamespace" minOccurs="0" maxOccurs ="unbounded"/> <xs:choice> <xs:attribute name="RequestContextPath" type="xs:anyURI" use ="optional"/> <xs:attribute name="RequestContextId" type="xs:string" use="optional"/> </xs:choice> <xs:attribute name="DataType" type="xs:anyURI" use="optional"/> <xs:attribute name="XPathVersion" type="xs:anyURI" use="optional" default ="http://www.w3.org/TR/1999/Rec-xpath-19991116"/> </xs:complexType> <xs:element name="XPathNamespace" type="xacml:XPathNamespaceType" substitutionGroup="xacml:AbstractDefaults"/> <xs:complexType name="XPathNamespaceType"> <xs:attribute name="NamespaceURI" type="xs:anyURI"/> <xs:attribute name="Prefix" type="xs:string" use="optional"/> </xs:complexType> Scope of the XPathNamespace for AttributeSelector element: 1. AttributeSelector element that includes XPathNamespace element, XPathNamespace elements in PolicyDefaults or PolicySetDefaults elements that include the AttributeSelector element. For the precedence, 1. XPathNamespace elements in AttributeSelector take precedence over XPathNamespace elements in PolicyDefaults in Policy element.. 2. XPathNamespace elements in PolicyDefaults in Policy take precedence over XPathNamespace elements in PolicySetDefaults in PolicySet element.. 3. If there are two or more identical prefixes are specified under an AttributeSelector, a PolicyDefaults or a PolicySetDefaults elements, the last prefix takes precedence over the previous prefixes. Others: 1. Global xmlns attribute is not used for resolving namespace-prefix pair specified in XPath expression. 2. If no XPathNamespace element is found in valid scope, it means no namespace-prefix pair is defined. 3. If Prefix attribute is missing, it means that default namespace is defined. 0033. [Michiharu] Example two to separate document http://lists.oasis-open.org/archives/xacml/200208/msg00147.html STATUS: NOT YET CONSIDERED. After looking at line 589-1102, I think that Section 3.2 policy specification in Example two is too complicated for average readers. While my preference is to hold this example in the spec, most readers would feel that XACML is difficult to read and understand when they try to go through this example. Besides, this example requires expertise about XPath syntax and its data model that is basically different from this specification. Honestly speaking, it is a little difficult even for me... :-). So my suggestion is to move this section to a separate document (say Primer document or Use Case document). 0034. [Polar] Target Match Semantics Section 4 & 5 http://lists.oasis-open.org/archives/xacml/200208/msg00188.html [long set of specific text changes included in e-mail] http://lists.oasis-open.org/archives/xacml/200208/msg00189.html (correction) STATUS: NOT YET CONSIDERED. The semantics for the Target need to be updated. It doesn't seem to describe how "Match" or "No-Match" are derived at. The document mentions logical-AND and logical-OR, and they are not defined for Match and No-Match. Also, we need to resolve the discovery and issue of the inconsistency between the evaluation of Target and Condition with respect to Indeterminate and NotApplicable, now that we are making progress on the "functions" document. I feel that if we refer to the Target evaluations with the same semantics as Condition, we can leverage the use of True, False, and Indeterminate that we have already defined, and then we can use the notions of conjunctive sequence and disjunctive sequence as combinators of our boolean values (true, false, indeterminate) with the normative specifications of our functions "function:and" and "function:or" for the combining rules. This will take care of the normative handling of error conditions in the evaluation logic. When the XACML Data Types, Functions, and Semantics gets put in the document, I suggest making the following changes. Some changes are editorial. Note: There is some issue about <Target> and <Condition> of where they say that they possible to be "empty", but I'm not sure if that means "omitted" (i.e. minOccurs="0"), or <Target/>, or <Condition/>. 0035. [Michiharu] AttributeSelectorIndirect http://lists.oasis-open.org/archives/xacml/200208/msg00190.html http://lists.oasis-open.org/archives/xacml/200208/msg00201.html (followup) http://lists.oasis-open.org/archives/xacml/200209/msg00016.html (more) http://lists.oasis-open.org/archives/xacml/200209/msg00024.html (Polar) http://lists.oasis-open.org/archives/xacml/200209/msg00030.html (Michiharu) http://lists.oasis-open.org/archives/xacml/200209/msg00031.html (Polar) http://lists.oasis-open.org/archives/xacml/200209/msg00035.html (Daniel) http://lists.oasis-open.org/archives/xacml/200209/msg00036.html (Daniel) http://lists.oasis-open.org/archives/xacml/200209/msg00040.html (Polar) http://lists.oasis-open.org/archives/xacml/200209/msg00042.html (Daniel) STATUS: NOT YET CONSIDERED. NOTE: we need a restatement of the current complete proposal, taking into account comments and corrections to date. Based on the discussion on Monday call, Simon and I agreed to changing the schema to support an AttributeSelectorIndirect element to retrieve a XPath expression from the context. I wonder whether the name of this element is appropriate or not. <xs:complexType name="AttributeSelectorBaseType"> <xs:element ref="xacml:XPathNamespace" minOccurs="0" maxOccurs ="unbounded"/> <xs:attribute name="DataType" type="xs:anyURI" use="optional"/> <xs:attribute name="XPathVersion" type="xs:anyURI" use="optional" default ="http://www.w3.org/TR/1999/Rec-xpath-19991116"/> </xs:complexType> <xs:complexType name="AttributeSelectorType"> <xs:complexContent> <xs:extension base="AttributeSelectorBaseType"> <xs:attribute name="RequestContextPath" type="xs:string" use ="required"/> </xs:extension> </xs:complexContent> </xs:complexType> <xs:complexType name="AttributeSelectorIndirectType"> <xs:extension base="AttributeSelectorBaseType"> <xs:choice> <xs:element ref="xacml:SubjectAttributeDesignator"/> <xs:element ref="xacml:ResourceAttributeDesignator"/> <xs:element ref="xacml:ActionAttributeDesignator"/> <xs:element ref="xacml:EnvironmentAttributeDesignator"/> </xs:choice> </xs:extension> </xs:complexType> <xs:element name="AttributeSelector" type="AttributeSelectorType"/> <xs:element name="AttributeSelectorIndirect" type ="AttributeSelectorIndirectType"/> <xs:element name="XPathNamespace" type="xacml:XPathNamespaceType" substitutionGroup="xacml:AbstractDefaults"/> <xs:complexType name="XPathNamespaceType"> <xs:attribute name="NamespaceURI" type="xs:anyURI"/> <xs:attribute name="Prefix" type="xs:string" use="optional"/> </xs:complexType> Sample text for this element: 5.28. Element <AttributeSelectorIndirect> The <AttributeSelectorIndirect> element is a free-form pointing device into the <xacml-context:Request> element using an attribute designator. The actual xpath expression is retrieved from the context pointed by that attribute designator. Other than that, the semantics is the same with <AttributeSelector> element. Support for the <AttributeSelectorIndirect> element is OPTIONAL. The <AttributeSelectorIndirect> element is of AttributeSelectorIndirectType complex type. The <AttributeSelectorIndirect> element has either one of the following elements: SubjectAttributeDesignator ResourceAttributeDesignator ActionAttributeDesignator EnvironmentAttributeDesignator 0036. [Michiharu] DataType attribute required? http://lists.oasis-open.org/archives/xacml/200208/msg00192.html STATUS: NOT YET CONSIDERED. DataType attribute in AttributeDesignatorType is 'required'. But in most cases, AttributeDesignator is used below the matching function (e.g. string-match) and that function is type-specific. So I think "DataType" attribute is not required. I propose to change it to "optional". <xs:complexType name="AttributeDesignatorType"> <xs:attribute name="AttributeId" type="xs:anyURI" use="required"/> <xs:attribute name="DataType" type="xs:anyURI" use="required"/> <xs:attribute name="Issuer" type="xs:anyURI" use="optional"/> </xs:complexType> [should be] <xs:complexType name="AttributeDesignatorType"> <xs:attribute name="AttributeId" type="xs:anyURI" use="required"/> <xs:attribute name="DataType" type="xs:anyURI" use="optional"/> <xs:attribute name="Issuer" type="xs:anyURI" use="optional"/> </xs:complexType> [Polar] Taking this approach to the extreme, means you really need a typechecking compiler that not only "checks" the type correctness of the policy, but to WILL HAVE TO "infer" the types of the elements, which is a little bit of a harder problem. However, in this simple type system (i.e. no recursive, abstract types), it's not much harder. So, if you go that far, you might as well eliminate DataType attribute altogether from the schema, as it doesn't buy you anything but possible confusion. As as Michiharu says "In most cases...". However, I would say "In all cases...." since we can see every expression only appears in a Target or Condition. The top level expressions in a Target or a Condition in XACML must result in xs:boolean, and as Michiharu states, the types of all elements in the expression can be "inferred" from the functions applied to them. I would propose to remove the DataType attribute, in this case. [Daniel] That is correct. When we started discussion about functions, at least my personal idea was that if you specify data type within arguments - you can overload function names (so when you see add(integer, integer) you know which add to use). But then we decided to have argument type embedded into function names, so now we have all string-sequence and decimal-member-of. Type of the argument is indeed redundant now - as the external source is not aware of XACML types and PDP gets everything encoded as strings anyway.. But this will work as long as we do not allow any type conversions ever. If we decide to specify that any functions taking a decimal (numeric) can also take an integer, and promote it to decimal, you may get yourself an ambiguous situation with extensions. If I remember correctly - that was exactly the reason - at some point we have proposed implicit promotion between numeric datatypes.. I think no much harm is done by having DataType embedded. Simplifies some things.. May be more useful in the future.. [Polar] Such is the thingy with "optional" attributes. If we don't need them now, why have them. You can always "add" an optional attribute to the schema later, unless there is some non-standard use for it. Otherwise if it exists, it just holds redundant information. [Daniel] ..which raises another issue. Do we want to allow the same named attribute to be declared of different types in different rules. So that "AccountNumber" is used as integer in one rule, and as <date> in another.. Having "DataType" specified, we may specify that within one PolicySet, same AttributeId should be associated with one and only one data type.. [Polar] Still, you can infer the type information from the functions applied in the expression it appears in.
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [Elist Home]
Powered by eList eXpress LLC