Patent Application
20080066160
The same numbers are used throughout the drawings to reference like and/or corresponding aspects, features, and components.
In a described implementation, security scheme 100 can be overlaid and/or integrated with one or more devices 102, which can be comprised of hardware, software, firmware, some combination thereof, and so forth. As illustrated, “d” devices, with “d” being some integer, are interconnected over one or more networks 104. More specifically, device 102(1), device 102(2), device 102(3) . . . device 102(d) are capable of communicating over network 104.
Each device 102 may be any device that is capable of implementing at least a part of security scheme 100. Examples of such devices include, but are not limited to, computers (e.g., a client computer, a server computer, a personal computer, a workstation, a desktop, a laptop, a palm-top, etc.), game machines (e.g., a console, a portable game device, etc.), set-top boxes, televisions, consumer electronics (e.g., DVD player/recorders, camcorders, digital video recorders (DVRs), etc.), personal digital assistants (PDAs), mobile phones, portable media players, some combination thereof, and so forth. An example electronic device is described herein below with particular reference to
Network 104 may be formed from any one or more networks that are linked together and/or overlaid on top of each other. Examples of networks 104 include, but are not limited to, an internet, a telephone network, an Ethernet, a local area network (LAN), a wide area network (WAN), a cable network, a fibre network, a digital subscriber line (DSL) network, a cellular network, a Wi-Fi® network, a WiMAX® network, a virtual private network (VPN), some combination thereof, and so forth. Network 104 may include multiple domains, one or more grid networks, and so forth. Each of these networks or combination of networks may be operating in accordance with any networking standard.
As illustrated, device 102(1) corresponds to a user 106 that is interacting with it. Device 102(2) corresponds to a service 108 that is executing on it. Device 102(3) is associated with a resource 110. Resource 110 may be part of device 102(3) or separate from device 102(3).
User 106, service 108, and a machine such as any given device 102 form a non-exhaustive list of example entities. Entities, from time to time, may wish to access resource 110. Security scheme 100 ensures that entities that are properly authenticated and authorized are permitted to access resource 110 while other entities are prevented from accessing resource 110.
As illustrated, device 102(A) includes two security-related components: a security token 204 and an application 210. Security token 204 includes one or more assertions 206. Device 102(B) includes five security-related components: an authorization context 212, a resource guard 214, an audit log 216, an authorization engine 218, and a security policy 220. Security policy 220 includes a trust and authorization policy 222, an authorization query table 224, and an audit policy 226.
Each device 102 may be configured differently and still be capable of implementing all or a part of security scheme 100. For example, device 102(A) may have multiple security tokens 204 and/or applications 210. As another example, device 102(B) may not include an audit log 216 or an audit policy 226. Other configurations are also possible.
In a described implementation, authority 202 issues security token 204 having assertions 206 to entity 208. Assertions 206 are described herein below, including in the section entitled “Security Policy Assertion Language Example Characteristics”. Entity 208 is therefore associated with security token 204. In operation, entity 208 wishes to use application 210 to access resource 110 by virtue of security token 204.
Resource guard 214 receives requests to access resource 110 and effectively manages the authentication and authorization process with the other security-related components of device 102(B). Trust and authorization policy 222, as its name implies, includes policies directed to trusting entities and authorizing actions within security environment 200. Trust and authorization policy 222 may include, for example, security policy assertions (not explicitly shown in
Authorization context 212 collects assertions 206 from security token 204, which is/are used to authenticate the requesting entity, and security policy assertions from trust and authorization policy 222. These collected assertions in authorization context 212 form an assertion context. Hence, authorization context 212 may include other information in addition to the various assertions.
The assertion context from authorization context 212 and an authorization query from authorization query table 224 are provided to authorization engine 218. Using the assertion context and the authorization query, authorization engine 218 makes an authorization decision. Resource guard 214 responds to the access request based on the authorization decision. Audit log 216 contains audit information such as, for example, identification of the requested resource 110 and/or the algorithmic evaluation logic performed by authorization engine 218.
In a described implementation, entity 208 authenticates* itself to resource guard 214 with a token*, security token 204. Resource guard 214 forwards the token assertions* to authorization context 212. These token assertions are assertions 206 (of
Policy assertions are extracted from trust and authorization policy 222 by security policy 220. The policy assertions may include both trust-related assertions and authorization-related assertions. Security policy 220 forwards the policy assertions* to authorization context 212. Authorization context 212 combines the token assertions and the policy assertions into an assertion context. The assertion context* is provided from authorization context 212 to authorization engine 218 as indicated by the encircled “A”.
An authorization query is ascertained from the authorization query table. Resource guard 214 provides the authorization query (auth. query*) to authorization engine 218. Authorization engine 218 uses the authorization query and the assertion context in an evaluation algorithm to produce an authorization decision. The authorization decision (auth. dcn.*) is returned to resource guard 214. Whether entity 208 is granted access* to resource 110 by resource guard 214 is dependent on the authorization decision. If the authorization decision is affirmative, then access is granted. If, on the other hand, the authorization decision issued by authorization engine 218 is negative, then resource guard 214 does not grant entity 208 access to resource 110.
The authorization process can also be audited using semantics that are complementary to the authorization process. The auditing may entail monitoring of the authorization process and/or the storage of any intermediate and/or final products of, e.g., the evaluation algorithm logically performed by authorization engine 218. To that end, security policy 220 provides to authorization engine 218 an audit policy* from audit policy 226. At least when auditing is requested, an audit record* having audit information may be forwarded from authorization engine 218 to audit log 216. Alternatively, audit information may be routed to audit log 216 via resource guard 214, for example, as part of the authorization decision or separately.
Generally, a device 102 may represent any computer or processing-capable device, such as a client or server device; a workstation or other general computer device; a PDA; a mobile phone; a gaming platform; an entertainment device; one of the devices listed above with reference to
In a described implementation of device 102, I/O interfaces 404 may include (i) a network interface for communicating across network 104, (ii) a display device interface for displaying information on a display screen, (iii) one or more man-machine interfaces, and so forth. Examples of (i) network interfaces include a network card, a modem, one or more ports, and so forth. Examples of (ii) display device interfaces include a graphics driver, a graphics card, a hardware or software driver for a screen or monitor, and so forth. Printing device interfaces may similarly be included as part of I/O interfaces 404. Examples of (iii) man-machine interfaces include those that communicate by wire or wirelessly to man-machine interface devices 402 (e.g., a keyboard, a remote, a mouse or other graphical pointing device, etc.).
Generally, processor 406 is capable of executing, performing, and/or otherwise effectuating processor-executable instructions, such as processor-executable instructions 410. Media 408 is comprised of one or more processor-accessible media. In other words, media 408 may include processor-executable instructions 410 that are executable by processor 406 to effectuate the performance of functions by device 102.
Thus, realizations for security-related implementations may be described in the general context of processor-executable instructions. Generally, processor-executable instructions include routines, programs, applications, coding, modules, protocols, objects, components, metadata and definitions thereof, data structures, application programming interfaces (APIs), schema, etc. that perform and/or enable particular tasks and/or implement particular abstract data types. Processor-executable instructions may be located in separate storage media, executed by different processors, and/or propagated over or extant on various transmission media.
Processor(s) 406 may be implemented using any applicable processing-capable technology. Media 408 may be any available media that is included as part of and/or accessible by device 102. It includes volatile and non-volatile media, removable and non-removable media, and storage and transmission media (e.g., wireless or wired communication channels). For example, media 408 may include an array of disks/flash memory/optical media for longer-term mass storage of processor-executable instructions 410, random access memory (RAM) for shorter-term storing of instructions that are currently being executed, link(s) on network 104 for transmitting communications (e.g., security-related data), and so forth.
As specifically illustrated, media 408 comprises at least processor-executable instructions 410. Generally, processor-executable instructions 410, when executed by processor 406, enable device 102 to perform the various functions described herein, including those actions that are illustrated in the various flow diagrams. By way of example only, processor-executable instructions 410 may include a security token 204, at least one of its assertions 206, an authorization context module 212, a resource guard 214, an audit log 216, an authorization engine 218, a security policy 220 (e.g., a trust and authorization policy 222, an authorization query table 224, and/or an audit policy 226, etc.), some combination thereof, and so forth. Although not explicitly shown in
This section describes example characteristics of an implementation of a security policy assertion language (SecPAL). The SecPAL implementation of this section is described in a relatively informal manner and by way of example only. It has an ability to address a wide spectrum of security policy and security token obligations involved in creating an end-to-end solution. These security policy and security token obligations include, by way of example but not limitation: describing explicit trust relationships; expressing security token issuance policies; providing security tokens containing identities, attributes, capabilities, and/or delegation policies; expressing resource authorization and delegation policies; and so forth.
In a described implementation, SecPAL is a declarative, logic-based language for expressing security in a flexible and tractable manner. It can be comprehensive, and it can provide a uniform mechanism for expressing trust relationships, authorization policies, delegation policies, identity and attribute assertions, capability assertions, revocations, audit requirements, and so forth. This uniformity provides tangible benefits in terms of making the security scheme understandable and analyzable. The uniform mechanism also improves security assurance by allowing one to avoid, or at least significantly curtail, the need for semantic translation and reconciliation between disparate security technologies.
A SecPAL implementation may include any of the following example features: [1] SecPAL can be relatively easy to understand. It may use a definitional syntax that allows its assertions to be read as English-language sentences. Also, its grammar may be restricted such that it requires users to understand only a few subject-verb-object (e.g., subject-verb phrase) constructs with cleanly defined semantics. Finally, the algorithm for evaluating the deducible facts based on a collection of assertions may rely on a small number of relatively simple rules.
[2] SecPAL can leverage industry standard infrastructure in its implementation to ease its adoption and integration into existing systems. For example, an extensible markup language (XML) syntax may be used that is a straightforward mapping from the formal model. This enables use of standard parsers and syntactic correctness validation tools. It also allows use of the W3C XML Digital Signature and Encryption standards for integrity, proof of origin, and confidentiality.
[3] SecPAL may enable distributed policy management by supporting distributed policy authoring and composition. This allows flexible adaptation to different operational models governing where policies, or portions of policies, are authored based on assigned administrative duties. Use of standard approaches to digitally signing and encrypting policy objects allow for their secure distribution. [4] SecPAL enables an efficient and safe evaluation. Simple syntactic checks on the inputs are sufficient to ensure evaluations will terminate and produce correct answers.
[5] SecPAL can provide a complete solution for access control requirements supporting required policies, authorization decisions, auditing, and a public-key infrastructure (PKI) for identity management. In contrast, most other approaches only manage to focus on and address one subset of the spectrum of security issues. [6] SecPAL may be sufficiently expressive for a number of purposes, including, but not limited to, handling the security issues for Grid environments and other types of distributed systems. Extensibility is enabled in ways that maintain the language semantics and evaluation properties while allowing adaptation to the needs of specific systems.
As illustrated at the top row of assertion format 500, an example assertion at a broad level includes: a principal portion 502, a says portion 504, and a claim portion 506. Textually, the broad level of assertion format 500 may be represented by: principal says claim.
At the next row of assertion format 500, claim portion 506 is separated into example constituent parts. Hence, an example claim portion 506 includes: a fact portion 508, an if portion 510, “n” conditional fact1 . . . n portions 508(1 . . . n), and a c portion 512. The subscript “n” represents some integer value. As indicated by legend 524, c portion 512 represents a constraint portion. Although only a single constraint is illustrated, c portion 512 may actually represent multiple constraints (e.g., c1, . . . , cm). The set of conditional fact portions 508(1 . . . n) and constraints 512(1 . . . m) on the right-hand side of if portion 510 may be termed the antecedent.
Textually, claim portion 506 may be represented by: fact if fact1, . . . , factn, c. Hence, the overall assertion format 500 may be represented textually as follows: principal says fact if fact1, . . . , factn, c. However, an assertion may be as simple as: principal says fact. In this abbreviated, three-part version of an assertion, the conditional portion that starts with if portion 510 and extends to c portion 512 is omitted.
Each fact portion 508 may also be further subdivided into its constituent parts. Example constituent parts are: an e portion 514 and a verb phrase portion 516. As indicated by legend 524, e portion 514 represents an expression portion. Textually, a fact portion 508 may be represented by: e verbphrase.
Each e or expression portion 514 may take on one of two example options. These two example expression options are: a constant 514(c) and a variable 514(v). Principals may fall under constants 514(c) and/or variables 514(v).
Each verb phrase portion 516 may also take on one of three example options. These three example verb phrase options are: a predicate portion 518 followed by one or more e1 . . . n portions 514(1 . . . n), a can assert portion 520 followed by a fact portion 508, and an alias portion 522 followed by an expression portion 514. Textually, these three verb phrase options may be represented by: predicate e1 . . . en, can assert fact, and alias e, respectively. The integer “n” may take different values for facts 508(1 . . . n) and expressions 514(1 . . . n).
Generally, SecPAL statements are in the form of assertions made by a security principal. Security principals are typically identified by cryptographic keys so that they can be authenticated across system boundaries. In their simplest form, an assertion states that the principal believes a fact is valid (e.g., as represented by a claim 506 that includes a fact portion 508). They may also state a fact is valid if one or more other facts are valid and some set of conditions are satisfied (e.g., as represented by a claim 506 that extends from a fact portion 508 to an if portion 510 to conditional fact portions 508(1 . . . n) to a c portion 512). There may also be conditional facts 508(1 . . . n) without any constraints 512 and/or constraints 512 without any conditional facts 508(1 . . . n).
In a described implementation, facts are statements about a principal. Four example types of fact statements are described here in this section. First, a fact can state that a principal has the right to exercise an action(s) on a resource with an “action verb”. Example action verbs include, but are not limited to, call, send, read, list, execute, write, modify, append, delete, install, own, and so forth. Resources may be identified by universal resource indicators (URIs) or any other approach.
Second, a fact can express the binding between a principal identifier and one or more attribute(s) using the “possess” verb. Example attributes include, but are not limited to, email name, common name, group name, role title, account name, domain name server/service (DNS) name, internet protocol (IP) address, device name, application name, organization name, service name, account identification/identifier (ID), and so forth. An example third type of fact is that two principal identifiers can be defined to represent the same principal using the “alias” verb.
“Qualifiers” or fact qualifiers may be included as part of any of the above three fact types. Qualifiers enable an assertor to indicate environmental parameters (e.g., time, principal location, etc.) that it believes should hold if the fact is to be considered valid. Such statements may be cleanly separated between the assertor and a relying party's validity checks based on these qualifier values.
An example fourth type of fact is defined by the “can assert” verb. This “can assert” verb provides a flexible and powerful mechanism for expressing trust relationships and delegations. For example, it allows one principal (A) to state its willingness to believe certain types of facts asserted by a second principal (B). For instance, given the assertions “A says B can assert fact0” and “B says fact0”, it can be concluded that A believes fact0 to be valid and therefore it can be deduced that “A says fact0”.
Such trust and delegation assertions may be (i) unbounded and transitive to permit downstream delegation or (ii) bounded to preclude downstream delegation. Although qualifiers can be applied to “can assert” type facts, omitting support for qualifiers to these “can assert” type facts can significantly simplify the semantics and evaluation safety properties of a given security scheme.
In a described implementation, concrete facts can be stated, or policy expressions may be written using variables. The variables are typed and may either be unrestricted (e.g., allowed to match any concrete value of the correct type) or restricted (e.g., required to match a subset of concrete values based on a specified pattern).
Security authorization decisions are based on an evaluation algorithm (e.g., that may be conducted at authorization engine 218) of an authorization query against a collection of assertions (e.g., an assertion context) from applicable security policies (e.g., a security policy 220) and security tokens (e.g., one or more security tokens 204). Authorization queries are logical expressions, which may become quite complex, that combine facts and/or conditions. These logical expressions may include, for example, AND, OR, and/or NOT logical operations on facts, either with or without attendant conditions and/or constraints.
This approach to authorization queries provides a flexible mechanism for defining what must be known and valid before a given action is authorized. Query templates (e.g., from authorization query table 224) form a part of the overall security scheme and allow the appropriate authorization query to be declaratively stated for different types of access requests and other operations/actions.
Security is critical in modern computing systems. Security may be facilitated with an efficient, high assurance, and flexible mechanism for determining authorized access. This is achievable using a policy-based approach. In such systems, policy can change to reflect current needs without having to change the underlying code for making and enforcing access decisions.
Existing approaches fail to fully meet these needs, especially in complex distributed computing environments. Simple rule-based policies, such as ACLs, are very efficient but not sufficiently flexible. More complex rule-based policy systems such as XACML offer more flexibility in return for a less efficient evaluation. Furthermore, such rule-based policy systems still have functional limitations (e.g., the lack of delegation support).
Conventional logic-based policy models generally offer the best functionality, but they are not without other problems. Some of these models do not offer adequate evaluation safety properties. In other words, a given authorization decision algorithm can not be guaranteed to terminate. Some are highly inefficient. Still others are theoretically sound, but they are not considered practical to implement in commercial systems.
What is needed is a logic-based security policy language that is both practical to implement and highly flexible and that has the requisite evaluation safety properties that ensure efficient evaluation as well as guaranteed termination. Described implementations of an example security policy language provide improvements in these areas.
In the sections that follow, an example security language is described. In many locations, the description is precise inasmuch as it provides a security language model that is supportable by a relatively-rigorous logical foundation. The example described implementations that include a rigorous logical foundation are provided to ensure that the theoretical underpinnings are understood. However, any given real-world implementations of a security language may not include any particular (much less all) of the aspects of the rigorous logical descriptions. Thus, the invention of the instant application is to be defined by the claims and not by any particular detailed implementations that are described herein.
Multiple example implementations for the expressions of a security language are provided herein. The implementations are designed such that the resulting security policies can be resolved logically in an efficient and certain manner.
Assertion syntax 604, authorization query syntax 606, and language semantics 608 each contribute to the formation and governance of example security language 602. Security scheme also includes assertion syntax safety checks 610 and authorization query safety checks 612. Assertion syntax safety checks 610 and authorization query safety checks 612 may be integrated into security language 602 to help ensure authorization query tractability.
In section 2 below, example syntax and semantics for assertions of the security language are described. In section 3, example syntax and semantics for authorization queries are described. In section 4, safety principles for the assertions and authorization queries are described. These safety properties, when properly applied, can guarantee that an authorization query evaluation will terminate and can ensure that the evaluation will proceed efficiently.
A core syntax for an example security language is described here. Additional syntax for a security language may provide for the grouping of assertions, for example to delegate a series of rights in a single assertion; however, additions such as these can be reduced to the core syntax. An example implementation of the security language may also enforce a typing discipline for constants, functions, and variables, but such typing enforcement is omitted here to concentrate on the core syntax.
Assertions
Authorization policies are specified as sets of assertions of the form
A says fact if fact1, . . . , factn, c
where the facts range over predicates that state properties on principals, for example that someone has the right to read a file. In an example implementation, assertions may be similar to Horn clauses, with the difference that (1) they are qualified by some principal A who issues and vouches for the asserted claim; (2) facts can be nested, using the keyword can assert, by means of which delegation rights are specified; and (3) variables in the assertion are constrained by c, a first order formula that can express, e.g., temporal, inequality, tree structure, regular expression constraints, and so forth.
The following listing sets forth assertion syntax 604 terminology as used herein:
In a described implementation, variables only range over strings and integers—but not predicates, facts, claims or assertions. A phrase of syntax is ground when it contains no variables. The set PredicateNames may not contain can assert∞, can assert0, and alias. Functions and predicates have a fixed arity. A predicate is a verb phrase with holes for its object parameters. If it has multiple words, these holes may appear at any fixed position within it, e.g. has access from [-] till [-].
In the remainder of the text, we use A, B, C, and D as meta variables for constants, usually for denoting principals. We use a vector notation to denote a (possibly empty) list (or tuple) of items, e.g. we may write ƒ({right arrow over (e)}) for ƒ(e1, . . . , en).
Facts can be nested, as can be seen from the syntax definition above. Nested facts are of the form e1 can assertD1 . . . en can assertDn fact, for some n≧1, where fact is flat, as defined next.
Two definitions are presented below. The first defines a flat fact, and the second defines parts of an assertion.
Definition 2.1. A fact is flat iff can assert does not occur in it, and nested otherwise.
For example, the fact Alice says Bob can read f is flat, but Alice says Charlie can assert0 Bob can read f is not flat.
Definition 2.2. Let A says fact if fact1, . . . , factn, c be an assertion. Then A is its issuer, the facti are its conditional facts, and c is its constraint. The fact may be considered the assertion's primary fact or asserted fact.
Constraints
Constraints range over any constraint domain that extends the example basic constraint domain shown below. Basic constraints include integer inequalities (e.g., for expressing temporal constraints), tree order constraints (e.g., for directories), and regular expressions (e.g., for ad hoc filtering). Examples of basic constraints are set forth below:
Additional constraints can be added without affecting decidability or significantly affecting tractability. However, it is preferred that the validity of ground constraints can be checked in polynomial time to maintain a desired level of tractability.
We use a sugared notation for constraints that can be derived from the basic ones, e.g. False, e1≠e2 or c1 or c2. In assertions, we usually omit the True constraint, and also omit the if when the assertion has no conditional facts.
Semantics
We now give a formal definition of the language semantics 608. We first define the denotation of a ground expression. The denotation of a constant A is just A, so [[A]]=A. The denotation of a function ƒ({right arrow over (e)}) is defined if ({right arrow over (e)}) is ground, and is also a constant, but may depend on the system state as well as [[{right arrow over (e)}]]. For example, [[CurrentTime( )]] presumably returns a different constant when called at different times. However, we assume that a single authorization query evaluation is atomic with respect to system state. That is, even though an expression may be evaluated multiple times, its denotation is the same during a single evaluation.
For a given constraint c, we write |=c iff it is valid. The following defines validity within the basic constraint domain.
In the rest of this document, we refer to a substitution θ as a function mapping variables to constants and variables. Substitutions are extended to constraints, predicates, facts, claims, assertions etc. in the natural way, and are usually written in postfix notation. We write Vars(X) for the set of variables occurring in a phase of syntax X.
In a described implementation, the security language includes three deduction rules. We now present these three deduction rules to capture the semantics of the language. Each rule has a set of premises and a single consequence of the form AC, D|=A says fact where Vars(fact)= and Dε{0, ∞}. Intuitively, the deduction relation holds if the consequence can be derived from the assertion context AC. If the derivation flag D=0, the Derivation Rule (can assert) is not used in the derivation.
Conditional semantics rule 608(1) is as follows:
Delegation semantics rule 608(2), with a “can assert” verb implementation for the general delegation-directive verb, is as follows:
Alias semantics rule 608(3) is as follows:
Rule (cond) allows the deduction of matching assertions in AC with all free variables substituted by constants. In a described strict logical implementation, the conditional facts must be deducible, and the substitution must also make the constraint(s) valid. The delegation flag D is propagated to all conditional facts.
In other words, given an assertion context AC, a delegation flag D, a principal A, and a substitution θ, we can derive “AC, D|=A says fact θ” if all of the following hold:
Rule (can assert) deduces an assertion made by A by combining a can assert assertion made by A and a matching assertion made by B. In a described strict logical implementation, this rule applies when the delegation flag is ∞. The matching assertion made by B must be proved with the delegation flag D obtained from A's can assert assertion.
In other words, given an assertion context AC, a principal A, and a fact fact, we can derive “AC, ∞|=A says fact” if there is a principal B and a delegation flag D such that both of the following hold:
Rule (alias) states that any fact that holds for C also holds for B.
In other words, given an assertion context AC, a delegation flag D, principals A, B, and C, and a verb phrase “verbphrase”, we can derive “AC, D|=A says B verbphrase” if both of the following hold:
The following propositions state basic properties of the deduction relation; they are established by induction on the rules above.
Authorization requests are decided by querying an assertion context (containing local as well as imported assertions). In a described implementation of a logic language, an authorization query may include a collection of atomic queries of the form A says fact and constraints C, combined by logical connectives, including negation. An example authorization query syntax 606 is as follows:
The resulting query language is more expressive than in other logic-based languages where only atomic queries are considered. For example, separation of duties, threshold and denying policies can be expressed by composing atomic queries with negation and constraints. In a described implementation, negation is not allowed within the assertion language, as coupling negation with a recursive language results in semantic ambiguities, and often to higher computational complexity or even undecidability. By restricting the use of negation to the level of authorization queries (rather than adding these features to the assertion language itself), we effectively separate it from recursion, thereby circumventing the problems usually associated with negation.
The semantics of queries is defined by the relation AC, θ|−q. In the following, let ε be the empty substitution. Note that negated queries and constraints are grounded, and that conjunctions are not commutative, as the second query may be instantiated by the outcome of the first query.
e says fact
q1, q2
q1 and AC, θ2
q2θ 1
q1 or q2
q1 or AC, θ
q2
not q
q does not hold and Vars(q) =
c
Given a query q and an authorization context AC, an authorization algorithm should return the set of all substitutions θ such that AC, θ|−q. If the query is ground, the answer set will be either empty (meaning “no”) or be a singleton set containing the empty substitution ε (meaning “yes”). If the query contains variables, then the substitutions in the answer set are all the variable assignments that make the query true.
In the following section, section 4, safety conditions for ensuring that this set of variable assignments is finite and meaningful are described. Section 9 below gives an algorithm for evaluating authorization queries.
Authorization Query Tables
Conceptually, authorization queries are part of the local policy and may be kept separate from imperative code. In a described implementation of a security language, authorization queries belonging to a local assertion context are kept in a single place, termed the authorization query table. The table provides an interface to authorization queries by mapping parameterized method names to queries. Upon a request, the resource guard calls a method (instead of issuing a query directly) that gets mapped by the table to an authorization query, which is then used to query the assertion context.
For example, an authorization query table could contain the mapping:
canAuthorizePayment(requester, payment):
Admin says Alice possesses BankManagerID id,
not(Admin says Alice has initiated Payment47).
The resulting answer set (either an empty set if the request should be denied or a variable assignment for id) is returned to the resource guard who can then enforce the policy in accordance with the authorization decision.
In an example implementation of the security language, the authorization algorithm is required to terminate and return a complete set of answers with respect to the described semantics, no matter what the inputs are. Without further restrictions, this cannot be guaranteed as the constraint domain is not constraint compact. Thus, in a described implementation, relatively simple, purely syntactic restrictions are enforced on assertions and queries.
Definition 4.1. (Assertion safety check 610 (of FIG. 6)): Let A says fact if fact1, . . . , factn, c be an assertion. A variable xεVars(fact) is safe iff xεVars(fact1)∪ . . . ∪Vars(factn).
The assertion A says fact if fact1, . . . , factn, c is safe iff
Some examples of safe assertions are:
A says B can read Foo
A says B can read Foo if B can x y
A says B can read Foo if B can x y, x≠y
A says B can x y if B can x y
A says z can x y if z can x Foo, z can ready
A says B can assert x can y z
Some examples of unsafe assertions are:
A says B can x Foo
A says z can read Foo if B can x y
A says B can read Foo if B can x y, w≠y
A says B can x y if B can assert C can x y
A says w can assert x can y z
The safety condition guarantees that the evaluation of the logic language translation, as described below in section 8, terminates in all cases and is complete. Furthermore, the safety condition enables a simple handling of constraints within the evaluation algorithm of section 8: as long as the query is flat, all variables of a constraint are fully instantiated when conditional predicates are processed, so constraints do not need to be propagated, and there is no need to implement complex constraint solving algorithms.
We now define a safety condition on authorization queries that guarantees that the set of substitutions is finite, given that the assertions of the assertion context is safe. Furthermore, the condition ensures that subqueries of the form not(q) or c will be ground at evaluation time, assuming a left-to-right evaluation rule with propagation for conjunctions, as is defined in section 9.
We first define a deduction relation ∥− with judgments of the form I∥−q:O where q is a query and I, O are sets of variables. Intuitively, the set I represents the variables that are grounded by the context of the query, and O represents the variables that are grounded by the query.
The following deduction relationships provide syntactic checks 612 (of
A safe assertion inference rule is:
A safe conjunction inference rule is:
A safe disjunction inference rule is:
A safe negation inference rule is:
A safe constraint inference rule is:
Definition 4.2. (Authorization query safety 612 (of FIG. 6)): An authorization query q is safe iff there exists a set of variables O such that ∥−q:O.
Checking safety can be done by recursively traversing all subqueries and thereby constructing the set O (which is always uniquely determined by the query and I).
Examples of safe and unsafe queries are provided in the table below: In the following examples, “,” and “or” are left-associative.
In a described implementation, at block 702, the safety of the assertions of the assertion context are checked. For example, definition 4.1 may be applied to each assertion of an assertion context AC. Accordingly, at block 702(1), it is verified that the initial variables of an assertion are safe. For example, part (1) of definition 4.1 may be applied. For instance, the initial variables are the variables of fact (if fact is flat) or the variables of the expression e (if fact is not flat).
At block 702(2), it is verified that any constraint variables are also present somewhere else in the assertion. For example, part (2) of definition 4.1 may be applied. At block 702(3), it is verified that the conditional facts are flat. For example, part (3) of definition 4.1 may be applied.
At block 704, the safety of the authorization query is checked. For example, definition 4.2 may be applied to the authorization query. At block 704(1-5), depending on the syntax of the authorization query, one or more of the safe inference rules is applied to the authorization query to confirm compliance and ensure the query's safety. Example inference rules are: assertion, conjunction, disjunction, negation, and constraint.
If the assertion context fails its check (at block 702) or the authorization query fails its check (at block 704), then the evaluation algorithm may be aborted to avoid attempting to evaluate an assertion context and/or authorization query that will not terminate or that will not be completed. After the safety has been checked successfully (at blocks 702 and 704), then at block 706 the authorization query may be evaluated in conjunction with the assertion context.
In a described implementation of the security language, expiration dates can be expressed as ordinary verb phrase parameters:
An acceptor can then use the date to enforce its own time-based constraints on the validity of the assertion:
Assertions may have to be revoked before their scheduled expiration date. If the revocation is necessitated due to the compromise of an explicitly trusted issuer's cryptographic key, then revocation is done by informing the relying party that the key should no longer be trusted. The relying party then removes the policy expressing the explicit trust in that key. But it is far more common that an issuer needs to revoke an assertion it has made in the past. This could be due to compromise of the cryptographic key associated with the principal whom the assertion is about, some change in the principal's relationships (e.g., change in employment), or malicious behavior on the part of the principal. For instance, the assertion in the example above has to be revoked if Alice drops out of her university.
We assume that every assertion M is associated with an identifier (e.g., a serial number) IDM. Revocation (and delegation of revocation) can then be expressed in a described implementation by revocation assertions with the verb phrase revokes IDM. For example, the revocation assertion
Definition 5.1. (revocation assertion): An assertion is a revocation assertion if it is safe and of the form
Given an assertion context AC and a set of revocation assertions ACrev where AC∩ACrev=, we remove all assertions revoked by ACrev in AC before an authorization query is evaluated. The filtered assertion context is defined by
The condition that AC and ACrev are disjoint means that revocation assertions cannot be revoked (at least not within a described implementation of the security language). Allowing revocation assertions to be revoked by each other causes the same problems and semantic ambiguities as negated body predicates in logic programming Although these problems could be formally overcome, for example by only allowing stratifiable revocation sets or by computing the well-founded model, these approaches are not simple enough to encourage their use in commercial systems.
An example implementation of a security language as described herein above provides a mechanism for expressing security policies in a highly human-understandable form. Moreover, the security language can be checked for safety. However, security policies also need to be evaluated to make authorization decisions.
To that end, existing logic languages can be leveraged. There are existing logic languages that have been developed over time and are thoroughly tested. There are also helpful tools that may be used with these logic languages. An example applicable logic language is Datalog, and specifically constrained Datalog. However, other logic languages may alternatively be utilized.
Techniques and mechanisms are described herein to enable such existing logic languages to be used with a security language. The security language may be translated into a logic language, as described in section 7. Section 8 describes an evaluation algorithm with tabling that is performed in the logic language. The evaluation algorithm can produce a proof graph in the logic language. In section 9, a technique is described for translating the proof graph reflecting the logic language into a proof graph reflecting the security language. Hence, the authorization process, including the evaluation algorithm, may be analyzed in terms of the security language.
Example assertion semantics for a described security language is defined by the three deduction rules of section 2. This semantics is more comprehensible and intuitive than one defined in terms of a translation into some form of formal logic, which has been proposed. Nevertheless, it is useful to be able to efficiently translate security language assertion contexts into equivalent logic language programs. We can then leverage known complexity results (e.g., polynomial data complexity) and use the translated logic language program for query evaluation, which is described in section 8.
We now describe an algorithm for translating a safe assertion context into an equivalent logic language program. By way of example only, the safe assertion context is translated into a constrained Datalog program. In the following, we treat expressions of the form e1 saysk fact as Datalog predicates, where k is either a variable or 0 or ∞. This can be seen as a sugared notation for a predicate where the predicate name is the string concatenation of all infix operators (says, can assert, alias, revokes, and predicate names) occurring in the expression, including subscripts for can assert. The arguments of the predicate are the collected expressions between these infix operators. For example, the expression
Algorithm 7.1. The translation of an assertion context AC proceeds as follows:
In a described implementation, at block 802, a security language assertion with an asserted fact and zero or more conditional facts is extracted from an assertion context. At block 804, it is determined whether the asserted fact is flat. If so, then at block 806 the assertion is translated into a logic language rule having a primary fact and zero or more side conditions. For example, part 1 of Algorithm 7.1 may be performed. The primary fact corresponds to the asserted fact, and the side conditions correspond to the conditional facts. These differing terms are used for the sake of clarity to textually differentiate parts of the logic language program from parts of the security language assertion context.
If, on the other hand, the asserted fact is not determined (at block 804) to be flat, then the security assertion is translated by blocks 808 and 810. If the fact is not flat, then the assertion is a delegation assertion with at least one delegation-directive verb (e.g., can assert, can say, etc.).
At block 808, a logic language rule having a primary fact and zero or more side conditions is added (e.g., to a growing logic language program). For example, part 2 (a) of Algorithm 7.1 may be performed. At block 810, for each delegation-directive verb of the asserted fact of the security language assertion, a logic language rule having an unbounded delegation depth and a fresh variable representing the delegatee is added. More specifically, the fresh variable represents the principal to whom rights are being delegated. For example, part 2 (b) of Algorithm 7.1 may be performed.
At block 812, for each logic language rule having an asserted fact with an expression that has been created (at block 806 or 808/810), another logic language rule having an alias capability and a fresh variable representing the object of the alias is added to the logic language program. For example, part 3 of Algorithm 7.1 may be performed. At block 814, a logic language program segment corresponding to the input assertion of the security language assertion context is thereby produced. The actions of flow diagram 800 are repeated for each assertion in a given assertion context.
An example translation of an assertion is as follows:
is translated into
Definition 7.2. (Consequence operator): The immediate consequence operator TP is a function between sets of ground predicates and is defined as:
Theorem 7.3. (Soundness and completeness): Let P be the Datalog translation of the assertion context AC. A saysD factεTPw() iff AC, D|=A says fact.
In the section above, we show how a set of safe security language assertions can be translated into an equivalent, at least partially safe, constrained Datalog program. In the context of deductive databases, the bottom-up approach is most often used for Datalog evaluation. There, the program's model, or its least fixed-point (which is finite, due to safety), is computed once and for all. This has the advantage that it is a complete, terminating procedure, and query evaluation is fast once the fixed-point has been constructed.
However, bottom-up evaluation is not as suitable for a described implementation of the security language because the assertion context is not constant. In fact, it may be completely different between different requests. Computing the model for every request is not efficient as it results in the evaluation of irrelevant goals. Furthermore, the queries we are interested in are usually fully or partially instantiated; thus, a top-down, goal-directed approach seems more appropriate.
The most widely-known top-down evaluation algorithm is SLD resolution as is used in Prolog. Unfortunately, SLD resolution can run into infinite loops even for safe Datalog programs, if some of the predicates have recursive definitions. The problem remains even if instead of a depth-first a breadth-first search strategy is employed: the looping occurs because the SLD search tree is infinite. Tabling, or memoing, is an approach to guarantee termination by incorporating some bottom-up techniques into a top-down resolution strategy. The basic idea is to prune infinite search trees by keeping a table of encountered subgoals and their answers, and to compute a subgoal only if it is not already in the table.
We describe here a deterministic algorithm based on tabling and apply it to security language assertions that have been translated into a logic language program. Two global tables are kept during the evaluation process.
The first table, Ans table 904(1), maps predicates to sets of ground predicates. If a subgoal P (uninstantiated or partially or fully instantiated) is computed, P is added to the domain of Ans, and at any point in time, Ans(P) returns the set of answers to P (these are ground instances of P) that have been found so far. At the end of the evaluation process, Ans will contain the complete answers to all encountered subgoals including the root goal.
The second table is Wait table 904(2). It maps predicates P to subgoals of the form P; P0:-{right arrow over (P)}, c; S
and a predicate P01. The subgoal can be interpreted as follows: P is the top-level predicate of the proof tree; P0 is the predicate to be solved next; {right arrow over (P)} is the list of predicates that have to be solved after P0; c is a constraint on the predicate variables; and S is an instance of P that keeps track of the instantiations found so far by being continuously narrowed down.
A number of terms and concepts that relate to the evaluation algorithm are defined and/or explained below.
Definition 8.1. A substitution ρ is more general than θ iff there exists a substitution θ′ such that θ=ρθ′.
Definition 8.2. Let P and Q be two predicates. A substitution θ is a unifier of P and Q iff Pθ=Qθ. A substitution θ is a most general unifier of P and Q iff it is more general than any other unifier of P and Q.
Definition 8.3. A variable renaming for a predicate P is a substitution θ such that θ defines a permutation on the set of variables occurring in P.
If two predicates P and Q are unifiable, they also have a most general unifier that is unique up to variable renaming. We denote it by Mgu(P,Q). Finding the most general unifier is relatively simple, but there are more intricate algorithms that run in linear time.
Additional terms and concepts are defined and/or explained below:
Definition 8.4. Let P and Q be two predicates. Then P is an instance of Q iff P=Qθ for some substitution θ. We also say that P is subsumed by Q and write P=>Q
Proposition 8.5 P=>Q iff there exists a most general unifier θ of P and Q, and θ is a variable renaming of P.
The following pseudocode provides an example of an evaluation algorithm. It is explained further below.
P0′
To find the answers to a predicate P with respect to a set of rules and with initially empty Ans and Wait tables (i.e., Dom(A ns)=Dom(Wait)=), call the routine Q
Each of the five modules 902 is described below:
Q
R against Fresh(P) by calling R
.
RP; P0:-{right arrow over (P)}, c; S
and a predicate P′0. The Simplify function in R
If the unification and the satisfiability check succeed, there are two subcases. Either the list {right arrow over (P)} is empty, in which case we know that Sθ is an answer to P, and moreover, it is a ground answer, due to safety. P
P
P
Additional definitions and theorems are described below:
Definition 8.6. Answers(P) is defined as the return value of the call Q
and initially empty Ans and Wait tables 904, if the call terminates. Otherwise, Answers
(P) is undefined.
Theorem 8.7. (termination): Answers(P) is defined and finite for all partially safe queries P.
Theorem 8.8. (soundness and completeness): Let P be a partially safe query. Answers(P) is equal to the set of all ground instances of P that are in TPw().
When testing and troubleshooting security policies, it may be useful to be able to see a justification of an authorization decision. This can be a visual and/or textual representation of the corresponding proof graph constructed according to the semantics rule system in section 2 above. The proof graph produced during an evaluation may also be archived for auditing purposes.
Leaf nodes of logic language proof graph 1000 are represented by rectangular nodes 1002(4), 1002(5), 1002(6), 1002(8), 1002(9), and 1002(10). Non-leaf nodes are represented by circular nodes 1002(1), 1002(2), 1002(3), and 1002(7). The circles are nodes that prove side conditions (e.g., body predicates). The rectangular leaf nodes of logic language proof graph 1000 are rules translated from assertions from the original assertion context. The other nodes are derived statements, which are taken from the answers table (e.g., Ans table 904(1)) if the deterministic tabling algorithm described above is employed in the evaluation.
Logic language proof graph 1000 is separated into example fragments 1006. Specifically, three fragments 1006(1, 2, 3) are shown, but a proof graph may be separated into any number of fragments 1006. Each fragment 1006 forms part of a structure of the overall proof graph. Consequently, each fragment 1006 includes node(s) 1002 and directed edge(s) 1004.
In order for analysis, troubleshooting, and/or auditing, etc. to be performed by someone familiar with the security language, logic language proof graph 1000 may be translated into a corresponding security language proof graph (not separately shown in whole). To effect the translation, a pattern matching process is performed. A logic language fragment 1006 is matched to a corresponding security language fragment.
In a described implementation, the sample security language fragments that may be used in the pattern matching process are each respectively associated with a semantics rule from the security language. Example fragment correspondences between logic language fragments and security language fragments are described below with reference to
In a described specific example implementation, the algorithm presented above with reference to
If during an execution of the Algorithm 7.1 from above, each generated Datalog rule is labeled with the algorithm step at which it was generated (i.e., 1, 2(a), 2(b), or 3), the Datalog proof graph contains sufficient information to be converted into a corresponding security logic proof graph.
In a described implementation, at block 1102, a first proof graph that reflects a logic language during a logical evaluation is produced. For example, a logic language proof graph 1000 may be produced during a logical evaluation. The logical evaluation is effected on a logical language program that is derived from a security language assertion context via a translation.
At block 1104, a fragment of the first proof graph is identified. For example, a fragment 1006 of logic language proof graph 1000 may be identified.
At block 1106, the fragment of the first proof graph reflecting a logic language is translated into a corresponding fragment of a second proof graph reflecting a security language. For example, fragment 1006 of logic language proof graph 1000 that reflects a logical language (e.g., Datalog, etc.) is translated into a corresponding fragment of a second proof graph that reflects a security language (e.g., a policy assertion security language). For instance, a logic language pattern of the fragment of the first proof graph may be matched to a corresponding security language pattern selected from multiple possible security language patterns. These security language patterns may, for example, be associated with semantics rules of the security language.
At block 1108, the identification of a fragment and the translation thereof is repeated for each fragment of the first proof graph. For example, as indicated by next fragment arrow 1112, another fragment 1006 of logic language proof graph 1000 may be identified and translated into a corresponding fragment of a security language proof graph.
At block 1110, the second proof graph reflecting the security language is output once the logic language proof graph has been traversed and translated. The second proof graph includes fragments corresponding to respective fragments of the first proof graph. This second proof graph reflecting the security language can also be archived and/or audited.
Based on the algorithm from the previous section, we can now describe how to evaluate complex authorization queries as defined in section 3. In the following, let AC be an assertion context and its Datalog translation, and let ε be the empty substitution. We define the function AuthAnsAC on authorization queries as follows.
The following theorem shows that AuthAnsAC is an algorithm for evaluation safe authorization queries.
Theorem 9.1. (Finiteness, soundness, and completeness of authorization query evaluation): For all safe assertion contexts AC and safe authorization queries q,
q iff θ ∈ AuthAnsAC (q).
The devices, actions, aspects, features, functions, procedures, modules, data structures, protocols, components, etc. of
Although systems, media, devices, methods, procedures, apparatuses, mechanisms, schemes, approaches, processes, arrangements, and other implementations have been described in language specific to structural, logical, algorithmic, and functional features and/or diagrams, it is to be understood that the invention defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as example forms of implementing the claims.
