INTERNET-DRAFT                        Tatyana Ryutov
CAT Working Group                     Clifford Neuman
Expires February 1999                 USC/Information Sciences Institute                     
draft-ietf-cat-acc-cntrl-frmw-00.txt  August 07, 1998                                                            

Access Control Framework for Distributed Applications

0. Status Of this Document This document is an Internet-Draft. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." To view the entire list of current Internet-Drafts, please check the "1id-abstracts.txt" listing contained in the Internet-Drafts Shadow Directories on ftp.is.co.za (Africa), ftp.nordu.net (Northern Europe), ftp.nis.garr.it (Southern Europe), munnari.oz.au (Pacific Rim), ftp.ietf.org (US East Coast), or ftp.isi.edu (US West Coast). 1. Abstract This document describes a unified model to support authorization in a wide range of applications, including metacomputing, remote printing, video conference, and any other application which will require interactions between entities across autonomous security domains. The document proposes requirements for the support of: - flexible and expressive mechanism for representing and evaluating authorization policies - uniform authorization service interface for facilitating access control decisions for applications and requesting access control information about a particular resource. This specification defines structures and their uses at a level independent of underlying mechanism and programming language environment. This document is accompanied by a second one describing the details of proposed structures and services along with bindings for C language environments. This document is to be found in draft-ietf-cat-gaa-cbind-00.txt. 2. Introduction The variety of services available on the Internet continues to increase and new classes of applications such as metacomputing, remote printing, Ryutov/Neuman Expires February 1999 video conference are evolving. These applications will require interactions between entities across autonomous security domains. The distributed nature of the system, consisting of mutually suspicious security domains, requires a mechanism, which provides fine-grained access control to resources. For example, access control requirements of a remote printing application may include: - authorized individual users and organizations - time availability, e.g. time of the day or day of the week - restrictions on resources consumed by the clients, e.g. maximum job size, maximum number of pages per job - required confidentiality/integrity message protection - accounting for consumed resources Access control requirements of large-scale multicast application [4], e.g. corporate video conference may include: - authorized individual users and organizations - host properties; users on slow hosts or hosts running the wrong OS will be denied communication - payment & charging Some of the security requirements are common across different applications, while others are more individual. Access control policies can be formulated in many ways. Administrators Of each domain might use domain-specific policy syntax and heterogeneous implementations of the policies. It is necessary to define a particular framework applicable for a wide range of systems and applications, which will allow to discuss specific requirements for the representation and evaluation of security policies. The focus of this framework is based on the following two abstractions: 1) uniform mechanism for representation and evaluation of security policies It should be capable of implementing a number of different security policies, based on diverse authorization models, which can coexist in distributed system. Standardizing the way that applications define their security requirements provides the means for integration of local and distributed security policies and translation of security policies across multiple authorization models. The mechanism should support the common authorization requirements but provide the means to defining and integration with application or organization specific policies as well. Applications should not need to re-implement the basic authorization functions in an application-specific manner. Ryutov/Neuman Expires February 1999 2) Generic Authorization and Access control Application Program Interface (GAA API) A common API will facilitate authorization decisions for applications. An application invokes API functions to determine if a requested operation or set of operation are authorized or if additional checks are necessary. The API will support the needs of most applications, thus not forcing the developers to design their own authorization mechanisms. The API will allow better integration of multiple mechanisms with application servers. The GSS API can be used by the GAA API to obtain principal's identity see section 13. Section 14.3 gives an extended example how the GAA API can be used by applications. 3. Glossary OBJECT (RESOURCE) entity that has to be protected e.g. hosts, processes, files, and devices such as printers and faxes. SUBJECT entity that can initiate requests to an object e.g. individual users, hosts, applications and groups. PRINCIPAL identity associated with a subject as a result of some unspecified authentication protocol. It can refer to a person, group, host, and application. Several principals can be associated with the same subject. SECURITY POLICY the set of rules that govern can access to objects ACCESS RIGHT (OPERATION, PERMISSION) a particular type of access to a protected object e.g. read, write, and execute RESTRICTION (CONDITION) a specific policy allowing an operation to be performed on an object This policy can have two meanings: 1) descriptive An operation is allowed if certain condition is satisfied. For instance, a policy may require concurrence of two Principals to perform some operation. If participation of both principals can be proved then this policy is satisfied. 2) prescriptive An operation will be allowed if certain restrictive policy Ryutov/Neuman Expires February 1999 is enforced. For example, a process will be authorized to run on a host if the memory usage limits that a process can occupy in main memory satisfies certain constraints. DELEGATION is the ability of a principal to give to another principal limited authority to act on it's behalf. CREDENTIAL a statement of identity, group membership and non-membership, privilege attribute and transfer of privilege encoded in certificates. DISCRETIONARY ACCESS CONTROL (DAC) A means of restricting access to objects based on the identity of subjects and/or groups to which they belong. The controls are discretionary in the sense that a subject with a certain access permission is capable of passing that permission (perhaps indirectly) on to any other subject(unless restrained by mandatory access control). MANDATORY ACCESS CONTROL (MAC) A means of restricting access to objects based on the sensitivity (as represented by a label) of the information contained in the objects and the formal authorization (i.e. clearance) of subjects to access information of such sensitivity. 4. Security models to be supported It is intended that the framework will support a range of security Models. Security requirements for computer system can be based on different security models: 1) discretionary access control (DAC) DAC policies can be based on open or closed world models. a) closed world model based on implicit denial of all rights. Authorizations are granted by an explicit listing of positive access rights. b) open world model based on implicit granting of all rights and listing of only negative authorizations. There could be hybrid approaches allowing a mix of negative and Positive authorizations. Authorization conflicts can be resolved according to application-specific rules. 2) mandatory access control model (MAC) 5. Access Control Lists (ACLs) Ryutov/Neuman Expires February 1999 ACL is more appropriate management abstraction (comparing to capabilities)for distributed system environment. It allows administrators of a security domain to specify security policies with respect to resources that it controls. ACL provides convenient review what subjects authorized what modes of access. Another advantage is the ease in which access can be revoked. The owner of the object can simply remove or modify any ACL entry to revoke or change type of access granted to any subject. ACL-based approach more readily supports groups of subjects. 6. Objects The purpose of access control is protecting objects from unauthorized access. The kinds of objects to be protected are specific to the application to which the authorization model is applied and are not included into the authorization model. The objects that need to be protected include files, directories, network connections, hosts and auxiliary devices, e.g. printers and faxes. An authorization mechanism should support these different kinds of objects in a uniform manner. Same ACL structure should be used to specify access policies for different kinds of objects. Object names should be drawn from the application-specific name space and must be opaque to the authorization mechanism. 7. ACL format This section describes external (user-level) format for the ACL. The presented ACL format is intended as an example of ACL specification language usable by different applications to express their security policies. The ACL format in this framework extends the conventional ACL concept in two ways: 1) by using conditional authorization as an extension to authorization policies, implemented as restrictions on authentication and authorization credentials 2) by enabling the syntactic specification of various authentication policies Identification of a subject should identify the authentication method to be used in identifying the subject. Flexibility in describing a subject's identity is important because of various authentication mechanisms coexisting in a distributed system, which may not share a common name space. An ACL consists of a set of ACL entries. Each ACL entry represents access control policies directly associated with a principal. It specifies a principal, a list of principals (aggregated principal), or a group of principals, together with a set of granted and/or denied access rights and optional conditions associated with the rights. We use the Backus-Naur Form to denote the elements of our ACL language. Ryutov/Neuman Expires February 1999 Square brackets, [ ], denote optional items and curly brackets, {}, surround that can repeat zero or more times. A vertical line, | separates alternatives. Items inside double quotes are the terminal symbols. The wild-card symbol "*" has the same meaning as in the UNIX environment. ACL is specified according to the following format: acl ::= {acl_entry} acl_entry ::= principal {principal} "<" positive_access_rights ">"{condition} {"<" positive_access_rights ">"{condition}}";"| "<" negative_access_rights ">" ";" 7.1 Principals The authorization framework should support the following kinds of subjects: 1) USER identifies a person, e.g. authenticated user name. 2) HOST identifies a subject as a machine from which request to access the object was originated, e.g., an IP address or host DNS name or host public key. 3) APPLICATION identifies a certain program that has its own associated identity (principal),e.g., a checksum or certified name. This can be useful to grant access to a certain application, e.g. payment program, that is trusted to be written correctly and perform only it's intended purpose. 4) GROUP identifies a group of subjects. The kind of subjects (individual user, host or application) composing the group is opaque to the authorization mechanism. 5) ANYBODY represents any subject regardless of authentication. This may be useful for setting the default policies. Principals can be aggregated into a single entry when the same set of access rights and conditions applies to all of them. Roles can be represented in the framework using the principal types listed above and privilege constrain conditions, see section 7.3.1.1. The framework should support multiple existing principal naming methods. Different administrative domains might use different authentication mechanisms, each having a particular syntax for specification of principals. For example, an application may use Kerberos V5 as an authentication service. Kerberos V5 provides Ryutov/Neuman Expires February 1999 secret-key based authentication and the format of the Kerberos V5 principal name is user_name/instance@realm. Other domains may use DCE to obtain the user's identity credentials, usually identified by a User ID and Group ID. Another domain might use client authentication in SSL, based on public-key cryptography, where principals are identified by a global name, syntactically tied to the X.500 directory. The syntax of a principal ID is defined according to the underlying security mechanism. It is tagged to identify the name space. This specification relies upon the underlying authentication mechanism to provide the principal identities tagged with a type of the mechanism used. GAA API will compare the provided principal identity and one from ACL entries for equality. Note that the model enables the syntactic specification of multiple authentication policies, but it does not translate between heterogeneous authentication mechanisms. The framework should support various strengths of user authentication mechanisms. ACL may have entries associated with a different principals identifying same subject using different authentication methods. Authentication in distributed systems can be accomplished using weak methods, such as authentication by assertion or password-based authentication, as well as stronger methods based on cryptography, such as Kerberos, DASS, SSL. Specification of weaker authentication methods including network address, host name or username will allow the GAA API to be used with any existing application that does not have support for strong authentication. A subject may be granted a different set of rights, depending on the strength of the authentication method used for identification. For example, access rights granted to a subject identified by the IP address or DNS name can be more restricted compared to the rights granted to the same subject if stronger authentication method, such as Kerberos, was used. The principal is specified according to the following format: principal ::= principal_type sec_mech principal_ID | "ANYBODY" principal_type ::= "HOST" | "USER" | "GROUP" | "APPLICATION" sec_mech ::= alphanumeric_string principal_ID ::= alphanumeric_string Examples of principal specifications are: ANYBODY USER kerberos.v5 kot@ORG.EDU HOST IPaddress 164.67.21.82 Ryutov/Neuman Expires February 1999 GROUP DCE 7 APPLICATION checksum 0x77AA45 7.2 Access rights The proposed framework should support different types of access rights. ACL entry may contain multiple access rights because it is common to grant or deny a multiple access rights to the same set of principals. It must be possible to specify which principals or groups of principals are authorized for specific operations, as well as which principals are explicitly denied authorizations for protected object. Conditions placed on positive access rights have the goal of restricting the granted rights. The meaning of conditions on negative (denied) access rights is unclear. We intend to investigate this issue, however, for the time being, we require that: 1) A single ACL entry must not specify both positive and negative rights. 2) If an ACL entry specifies negative rights, it must not have any conditions placed on the denied rights. All operations defined on the object are grouped by type of access to the object they represent, and named using a tag. For a file the following operations are defined: FILE : read FILE : write FILE : execute However, in a bank application, an object might also be a customer account, and the following set of operation might be defined: ACCOUNT : deposit ACCOUNT : withdraw ACCOUNT : transfer Access rights names are from application-specific name space and opaque to the authorization mechanism. Internally a tagged bit vector represents access rights. Each bit in the vector corresponds to an access right. If ACL entry type is GRANT, then a bit is set if the corresponding right is granted. If ACL entry type is DENY, then a bit is set if the corresponding right is denied. The tag indicates how the bits in the bit vector are to be interpreted, for example, for the set of rights, associated with the tag FILE the first bit should be interpreted as read, while for the set associated with tag ACCOUNT, the same bit should be interpreted as deposit. Access rights are specified using the format: Ryutov/Neuman Expires February 1999 positive_access_rights ::= tag ":" value { tag ":" value } negative_access_rights ::= tag ":" "-" value { tag ":" "-" value } tag ::= alphanumeric_string value ::= alphanumeric_string 7.3 Conditions Authorization in distributed system, consisting of mutually suspicious security domains, requires fine-grained control over the conditions within which rights are granted. Individual requirements of evolving applications require access control decisions to depend on the state of other objects within a system, e.g. system load, time, client's host properties. The proposed framework should provide a means for applications to specify any necessary conditions on authorized rights in a uniform manner. Conditions are placed in ACLs, as well as in identity, group membership and authorization credentials. Conditions carried in the credentials are evaluated by the access control framework in addition to the conditions in the matching ACL entry. Conditions can be categorized as generic or specific. A condition is generic if it is evaluated by the access control model. Specific conditions are application-dependent and evaluated by the application. Conditions specify the type-specific policies under which an operation can be performed on an object. A condition is interpreted according to its type. The format used for specifying conditions is as follows: condition ::= type ":" value type ::= alphanumeric_string value ::= alphanumeric_string | alphanumeric_string {alphanumeric_string} "," 7.3.1 Generic conditions Issues that should be recognized in expressing generic conditions: 1) whether a condition is common across different applications 2) degree to which it can actually be enforced. Certainly, security Ryutov/Neuman Expires February 1999 policies can be written that express conditions that are impossible to implement. For instance, a policy that allows a given access only if some undecidable problem is solved would not be possible to implement. Thus, security policy expression requires some knowledge of the degree to which a given procedure is solvable. 2) whether a condition can be evaluated by the access control framework. If it requires interactions on application-specific level, then this condition should be evaluated by the application. 7.3.1.1 Proposed generic conditions that will appear in both ACLs and credentials The following list of conditions should not be considered exhaustive. a) time Time periods for which access is granted, e.g. time of day or day of the week b) location Location of the principal. Authorization is granted to the principals residing in specific hosts or domains. c) network connection Granting or denying authorization for specific routes. d) message protection constraints 1) required confidentiality message protection This condition specifies a mechanism, or a set of mechanisms to be used in confidentiality message protection. 2) required integrity message protection This condition specifies a mechanism, or a set of mechanisms to be used in integrity message protection. e) privilege constraints In general, a principal may belong to more than one group. By default, principal operates with the union of privileges of all groups to which it belongs, as well as all of his individual privileges. In assigning privileges, one can choose to: 1) have the subject operate with the privilege of only one group at a time. This can be used to reduce privileges as a protection against accidents. E.g. a person is a member of two groups: PROGRAMMERS and SYSTEM_MANAGERS. The person may act with the privileges of the group PROGRAMMERS most of the time, and enable privileges of the SYSTEM_MANAGERS group only on occasion. Ryutov/Neuman Expires February 1999 2) have the subject operate with privileges of several specified groupsat a time 3) endorsement Concurrence of N of M (N <= M) principals to perform some operation. f) multi-level security constraints Examples of using these conditions can be found in section 13.2. g) payment This restriction can be specified in two ways: 1) Specifies currency and amount that must be paid prior access to an object will be granted, e.g. payment : $20. 2) Specifies currency and a formula for calculating the amount that must be paid for consumed resources, e.g. payment : $0.5_per_MB. The amount of consumed resources must be known to the authorization framework in order to evaluate the payment condition. A valid certificate, stating the amount and currency paid must be presented to the authorization framework. j) quota This restriction specifies a limit. It limits the quantity of a resource that can be consumed or obtained. For example, a remote printing application may use quota : 10_MB. This condition specifies maximum job size, expressed in mega bytes, that can be sent to a printer. k) strength of authentication Specifies the authentication mechanism or set of suitable mechanisms, used to authenticate a user, e.g. sec_mech : kerberos.V5. This restriction may be useful to specify default access, e.g. ANYBODY < FILE:read FILE:write > sec_mech : DCE ; This condition means that anyone, who has been authenticated by DCE has read and write access to the object. l) attributes of subjects This class of conditions defines a set of attributes that must be Possessed by subjects in order to get access to the object. Examples: Ryutov/Neuman Expires February 1999 A large-scale multicast application can use condition OS : SUN_Solaris_V2. This will restrict hosts running the wrong OS from to participating in a communication. Access will be granted if valid certificate, stating the OS running on the host will be presented to the authorization framework. A metacomputing application can specify condition application_endorser : Globus This will restrict applications that can be run on the node to only ones, certified by Globus authority. A filtering application can use condition age : >=18 This will restrict access to some sites for underage users. The access will be authorized if valid certificate, stating the age of the user, is presented to the authorization framework. 7.3.1.2 Generic conditions appearing only in credential The authorization framework to make access control decision will evaluate these conditions along with conditions in the matching ACL entries. The following list of conditions should not be considered exhaustive. a) grantee This restriction specifies a list of principals (users, groups, applications or hosts) authorized to exercise the credential. This condition will not appear in ACLs. b) issued for Restriction specifies set of servers authorized to accept the public-key credentials which otherwise verifiable by and exercisable on all servers. c) group membership This restriction specifies that the user is a member of only the listed groups. d) group non-membership This restriction specifies that the user is NOT a member of the listed groups. This may be useful if authorization specification explicitly denies access to a specific group. e) authorized This restriction specifies a complete list of objects which may be accessed using the rights granted by the credential and, optionally, a list of operations that may be performed on each object. This condition usually appears in credentials used as capabilities or in credentials, returned by an authorization server. Ryutov/Neuman Expires February 1999 f) accept N times This condition specifies that the credential should be accepted no more than N times. COMMENTS ON THE ADDITIONAL TYPES OF GENERIC CONDITIONS ARE SOLICITED. 7.3.2 Application-specific condition If generic conditions are not sufficient for expressing application-specific security policies, applications specify their own conditions. Anything that can be expressed as type:value alphanumeric string can be a condition. The application must provide a means for evaluation of the application-specific conditions. 8. ACL evaluation The ACL language we presented supports authorization models based on the closed world model, when all rights are implicitly denied. Authorizations are granted by an explicit listing of positive access rights. The framework also supports negative authorizations. If one allows both negative and positive authorizations in individual or group entries, inconsistencies must be resolved according to different resolution rules. The design approach we adopted allows the ordered interpretation of ACLs. An ordered evaluation approach is easier to implement as it allows only partial evaluation of ACL and resolves the authorization conflicts. Evaluation of ordered ACL starts from the first to the last in the List of ACL entries. Evaluation stops either when all requested access rights have been granted by one or more ACL entries, or when any one of the requested access rights has been denied by one of the ACL entries. ACL entries containing principals that do not match the current Subject identity and the identities that are associated with the subject, stored in the security context, e.g. group memberships and delegated credentials, have no effect on the outcome of the evaluation. The resolution of inconsistent authorization is based on ordering. The ACL entries that already have been examined take precedence over new authorizations. Conflicts may arise when more then one entry applies. For example, one matching entry corresponds to a principal specifying individual subject (user, host or application), and the second entry matching a certain group name. In this case, one would expect the entry for the individual subject to be placed before the entry for the group, on the assumption that the policy expressed by the individual subject entry is an exception to the policy expressed by the group entry. When several ACL entries with different conditions apply, the access Ryutov/Neuman Expires February 1999 is based on the conditions in the first matching entry found. For example, if these two entries apply: USER kerberos.v5 tom@ORG.EDU < FILE : read > time_window : 6AM-8PM , time_day : Mon-Fri ; GROUP kerberos.v5 admin@ORG.EDU < FILE : read > time_window : 9AM-6PM ; Then the resulting conditions to be checked by the authorization framework are: time_window:6AM-8PM and time_day : Mon-Fri The ACL entry must have either only positive or only negative access rights. ACL entries specifying negative rights must not have any conditions. Other ACL interpretations are possible, such as unordered. 9. Security context The security context is a GAA API data structure, which is passed as an argument to the GAA API. It stores information relevant to access control policy, e.g. authentication and authorization credentials presented or used by the peer entity (usually the client of the request), connection state information. All credentials are in the GAA API internal format. The context consists of: 1) Identity Verified authentication information, such as principal ID for a Particular security mechanism. To determine which entries apply, the GAA API checks if the specified principal ID appears in an ACL entry that is paired with a privilege for the type of access requested. 3) Authorized credentials This type of credentials is used when individuals grant delegated Credential or generate a capability. 3) Group membership This type of credentials specifies that the grantee is a member of only listed groups. 4) Group non-membership This type of credentials specifies that the grantee is NOT a member of listed groups. 5) Attributes This type of credentials contains miscellaneous attributes attached to the grantee, e.g. age of the grantee, grantee security clearance. 6) Unevaluated Credentials Evaluation of the acquired credentials can be deferred till the credential is needed to perform the operation. Ryutov/Neuman Expires February 1999 7) Evaluation and Retrieval Functions for Upcalls These functions are called to evaluate application-specific conditions, to request additional credentials and verify them. The GSS API is an example of how this is filled in. 8) Connection State Information Containts a mechanism-specific representation of per-connection context, some of the data stored here include keyblocks, addresses. 10. Credential evaluation Credentials are parsed to the GAA API internal format and placed into the GAA API security context. When evaluating an ACL, the necessary credentials are looked for in the security context. Example Assume the following ACL for the file doc.txt is stored in the authorization data base: USER kerberos.v5 tom@ORG.EDU < FILE : read > ; GROUP kerberos.v5 admin@ORG.EDU < FILE : read FILE : write > ; USER kerberos.v5 joe@ORG.EDU < FILE : write> ; Assume the following credentials are stored in the security context associated with the user Tom: a) identity credential grantee: USER kerberos.v5 tom@ORG.EDU conditions: time_window : 06/07/98 19:49:21 06/08/98 05:49:19 b) group membership credential member of: GROUP kerberos.v5 admin@ORG.EDU conditions: privilege:constrained c) unevaluated credential cred_type: GAA_AUTHORIZED mech_type: DCE grantor: USER DCE 88 grantee: USER DCE 99 mech_spec_cred: [DCE-specific credential] c) authorized credential grantor: USER kerberos.v5 joe@ORG.EDU grantee: USER kerberos.v5 tom@ORG.EDU objects: doc.txt Ryutov/Neuman Expires February 1999 Operations: FILE:write conditions: location : *.org.edu Let's consider a request from a user Tom who is connecting from the ORG.EDU domain to write to the file doc.txt on 06/07/98 20.10.01. In evaluation the ACL the first entry does not grant the required operation, the second entry grants it The evaluation function will check the security context for the group membership credential. The proper credential is found, however, there is a condition privilege:constrained. It means that Tom can use this privilege only if logged in as an administrator. Evaluation continues. The third entry grants the requested operation. The evaluation function will look for authorized credential for USER kerberos.v5 tom@ORG.EDU issued by USER kerberos.v5 joe@ORG.EDU. The appropriate authorized credential is found and it grants the requested operation. The condition location:*org.edu is satisfied, so the requested access will be granted. 11. Architecture The major components of the architecture are: Authentication mechanisms (perhaps involving an authentication server) perform authentication of users and supply them with initial credentials. A group server is trusted to maintain and provide group membership information. A group is a convenient method to associate a name with a set of principals for access control purposes. A group server issues group membership and non-membership certificates. When a connection is established with an application server, these certificates are evaluated (evaluation may be deferred until needed) the results are placed into the GAA API security context. They are checked by the GAA API when making authorization decisions. The application calls the GAA API routines to check authorization against the application authorization model. These routines obtain access control list entries from local files, distributed authorization servers, and from credentials provided by the user, combining local and distributed authorizations under a single API according to the requirements of the application. Delegation is supported through inclusion of delegation credentials. Mechanism for delegation such as those supported by restricted proxies [1] in the security context, where they are available for use by authentication and authorization mechanisms used for subsequent connections from the server (now acting as an intermediary) to another server. Conditions can be embedded in authorization credentials or certificates. The restrictions or conditions carried in a certificate are evaluated by the GAA API in addition to the restrictions in the Ryutov/Neuman Expires February 1999 matching ACL entry. 12. Generic Authorization and Access API The GAA API is built into applications through a library. It is used by applications to decide whether a subject is authorized to perform particular operations on an object. In this section we provide a brief description of the main GAA API routines. 12.1 GAA API functions The gaa_get_object_eacl function is called before other GAA API routines which require a handle to an object ACL to identify ACLs on which to operate. Input: o Reference to the object to be accessed The identifier for the object is from an application-dependent name space, it can be represented as unique object identifier, or symbolic name local to the application. o Pointer to application-specific authorization database o Upcall function for the retrieval of the object ACL. The application maintains authorization information in a form understood by the application. It can be stored in a file, database, directory service or in some other way. The upcall function provided for the GAA API translates this information into the internal representation understood by the GAA API. Output: o Mechanism-specific status code o A handle to the list of ACLs associated with the protected object. The gaa_check_authorization function tells the application server whether the requested operation is authorized, or if additional application-specific checks are required. Input: o A handle to the object ACL, returned by gaa_get_object_eacl o Principal's security context o Operations for authorization This argument is optional, it indicates operations to be performed. Ryutov/Neuman Expires February 1999 o GAA API options structure This argument describes the behavior of the GAA API and specifies how the other arguments should be interpreted. For example, type of the ACL: ordered or unordered. Depending on this type corresponding ACL evaluation algorithm will be used be the GAA API. Output: YES (indicating authorization) is returned if all requested operationsare authorized. NO (indicating denial of authorization) is returned if at least one operation is not authorized. MAYBE (indicating a need for application-specific checks) is returned if there are some unevaluated conditions and additional application-specific checks are needed, or continuous evaluation is required. o Mechanism-specific status code o Detailed answer Ryutov/Neuman Expires February 1999 Detailed answer contains: o Authorization valid time period. The time period during which the authorization is granted is returned as condition to be checked by the application. Expiration time is calculated by GAA API, based on: time-related conditions in the object ACL matching entries o Restrictions in the authentication, authorization and delegated credentials. o List of all authorized rights and corresponding conditions, if any. Each condition is marked as evaluated or not evaluated, if evaluated marked as met or not met. o Information about additional security attributes required. Additional credentials might be required from clients to Perform certain operations, e.g. group membership or delegated credentials. If no operation was specified as an input, a list of authorized rights and corresponding conditions, if any, is returned. This allows application to discover access control policies associated with the target object. The application must understand the conditions that are returned unevaluated or it must reject the request. If understood, the application checks the conditions against Ryutov/Neuman Expires February 1999 information about the request, the target object, or environmental conditions to determine whether the conditions are met. Enforcement of the returned conditions is up to the application. 13. Creation of the GAA API security context Prior to calling the gaa_check_authorization function, the application must obtain the authenticated principal's identity and store it in the security context. This context may be constructed from credentials obtained from different mechanisms, e.g. GSS API, Kerberos or others. Figure 1 shows the flow of control: Application calls requesting the principal's identity (1); the request and the verification of the principal's identity credentials take place (2, 3); The principal's authentication credentials are placed in the security context (4a) and returns it to the application (4); the application calls the GAA API (5); the security context, containing the verified principal's identity is being passed into the GAA API (5a). |-----------------------| 5 |----------------------| | Application |------------>| GAA API | |----------|-^----------| |-----------^----------| 1 | | 4 5a | |----------V-|----------| 4a - - - - - - |- - - - - - - | Security API |------------>| Security Context | |----------|-^----------| - - - - - - - - - - - - - 2 | | 3 |----------V-|----------| | Security Server | |-----------------------| Figure 1 This scenario places a heavy burden on the application programmer to provide the integration of the security mechanism with the application. A second scenario is to obtain the authentication credentials from a transport protocol that already has the security context integrated with it. For example, the application can call SSL, and authenticated RPC, or the Asynchronous Reliable Delivery Protocol (ARDP), a communications protocol which handles several security services including authentication, integrity and payment. In this case, it is the implementation of the transport mechanism (usually written by someone other than the application programmer) which calls the security API requesting principal's identity, and which constructs the security context. The principal's authentication information is placed into the security context and is passed to the GAA API. When additional security attributes are required for the requested operation, the list of required attributes is returned to be obtained by the application. The Ryutov/Neuman Expires February 1999 application may provide GAA API with an upcall function for requesting required additional credentials. The credentials pulled by the GAA API are verified and added to the security context by the upcall function. A reference to the upcall function is passed to the GAA API as part of the security context, and it is added to the security context by the application or transport. 14. Emulation of various security models The following are examples of using proposed authorization framework to express various access control models. 14.1 Open world model exampl The open world model is practical when access to an object is granted to a number of principals, and access denied to a few principals. Instead of listing all principals and groups of principals with corresponding positive rights, it is more convenient to make access default and specify only principals with the negative rights. This will save space and makes evaluation of ACL more efficient. The open world model, which is based on implicit granting of all rights, and listing of only negative authorizations can be represented in this framework by including ANYBODY * as the final entry in an ACL (if ordered ACL interpretation is used). This will grant everybody all rights regardless of authentication. Denial of rights is then specified using negative rights in entries earlier in the ACL. 14.2 Mandatory Security security models It is intended, that proposed framework will allow incorporation of multi-level security models. Mandatory policies govern access on the basis of classification of subjects and objects in the system. These subjects and objects are assigned sensitivity labels that Denote their hierarchical sensitivity and need-to-know attributes. Specifically, a security label consists of two components: a hierarchical security level and a possibly empty set of nonhierarchical security categories. For example, in the military, the set of levels consists of Unclassified, Confidential, Secret, and Top secret. Set of categories may consist of NATO, NASA, NOFORN. In commercial environments, these levels might be Restricted, Proprietary, Sensitive, and Public. Set of categories: Department1, Department2 and Department3. Security label that denotes the security sensitivity of a subject is called CLEARANCE. Ryutov/Neuman Expires February 1999 Security label that denotes the security sensitivity of an object is called CLASSIFICATION. A security label DOMINATES another when its security level component is grater then or equal to the other's security level component and when its set of security categories is a superset of the other's security categories component. For example, security label Top Secret/NATO,NASA dominates security label Confidential/NASA. Neither of the following two security labels dominates the other: Top Secret/NATO,NASA and Secret/NOFORN. Mandatory Confidentiality Rules: Read down: A subject's security label must dominate the security label of the object being read. Write up: Subject's security label must be dominated by the Label of the object being written. Mandatory Integrity Rules: The hierarchy of the labels is based on the integrity rather then disclosure-oriented security. Read up: A subject's integrity label must be dominated by the Integrity label of the object being read. Write down: Subject's integrity label must dominate the label of the object being written. Implementation of both Mandatory Confidentiality and Integrity rules can be based on a single security level for both confidentiality and integrity. This would result in a read-equal and write-equal rules. The drawback is reduced flexibility of the resulting system. 14.2.1 Implementing the mandatory security within proposed authorization framewok This section describes labeling requirements for objects, subjects and access rights. 14.2.2 Labeling of objects and subjects Objects and subjects are assigned security labels. There are three label classes: a) Confidentiality labels, e.g. Top_Secret/NASA, Sensitive/Department2 b) Integrity labels, e.g. High_integrity, Low_integrity Ryutov/Neuman Expires February 1999 c) Single security labels for both confidentiality and integrity, e.g. Top_Secret/NASA, Unclassified (assume that the first label denotes high integrity level, whereas the second one denotes low integrity level). Newly created objects are labeled by the security label of the subject that created the object, in order to prevent possible information flows between the security levels, which violates mandatory security policies. The creator of the object must not have the ability to change any attribute of the object or change access permissions to the object. Only the authorized system administration authority is the owner of the all objects in the security domain, including newly created, and have full control over the object security attributes as well as control permissions to it. There may be no mandatory controls on accessing objects categorized as Unclassified or Public, e.g. public bulletin boards. Access to these objects may be restricted by DAC mechanisms. To prove eligibility to access an object, a subject has to present a valid credential, stating subject's security label. 14.2.3 Labeling of access right All access rights are divided into read-class and write-class. Appropriate rules are applied to each class. The rules are expressed through generic conditions. 14.2.4 Generic conditions for read-class access rights a) conf_read_equal : cofidentiality_label This condition specifies that a subject, wishing to get read-class access to the object has to have security clearance equal to the one, specified in the cofidentiality_label field. b) conf_read_below : cofidentiality_label This condition is used to enforce "read down" mandatory confidentiality rule. It specifies that a subject, wishing to get read-class access to the object has to have security clearance no less the one, specified in the cofidentiality_label field. c) integr_read_equal : integrity_label This condition specifies that a subject, wishing to get read-class access to the object has to have security clearance equal to the one, specified in the integrity_label field. d) integr_read_above : integrity_label Ryutov/Neuman Expires February 1999 This condition is used to enforce "read up" mandatory integrity rule. It specifies that a subject, wishing to get read-class access to the object has to have integrity clearance no less then the one, specified in the integrity_label field. 14.2.5 Generic conditions for write-class access rights a) conf_write_equal : cofidentiality_label This condition is used to enforce restricted "write up" mandatory confidentiality rule to avoid "blind writes" [?]. It specifies that a subject, wishing to get write-class access to the object has to have security clearance equal to the one, specified in the cofidentiality_label field. b) conf_write_above : cofidentiality_label This condition is used to enforce "write up" mandatory confidentiality rule. It specifies that a subject, wishing to get write-class access to the object has to have security clearance equal or greater then the one, specified in the cofidentiality_label field. c) integr_write_equal : integrity_label This condition specifies that a subject, wishing to get write-class access to the object has to have security clearance equal to the one, specified in the integrity_label field. d) integr_write_below : integrity_label This condition is used to enforce "write down" mandatory integrity rule. It specifies that a subject, wishing to get write-class access to the object has to have integrity clearance equal or greater then the one, specified in the integrity_label field. 14.2.6. Example Assume file doc.txt has classification Sensitive/Departmen1 and integrity label Medium, then ACL for this file can be specified as: ANYBODY FILE:read conf_read_below:Sensitive/Department1 integr_read_above:Medium ; ANYBODY FILE:write conf_write_above:Sensitive/Department1 integr_write_below:Medium ; Note that the example above allows everybody in the distributed system to get read or write access to the file doc.txt if valid credential stating appropriate security label attribute is presented. This pose a requirements for security labels to be unique across Different security domains. This may not be easily satisfied. There should be a way to restrict scope of the security labels to a particular administrative domain. A possible solution can be specifying an additional condition, e.g. Ryutov/Neuman Expires February 1999 location:*.org.com. Another solution may be using wildcard symbol in the principal's ID specification instead of ANYBODY), e.g. USER Kerberos.V5 *@ORG.COM. 14.3 Extended example: simple Printer Manager applicatio To illustrate the flow of control involved in use of GAA API by application servers we describe a simple Printer Manager application, where protected objects are printers. The Printer Manager accepts requests from users to access printers and invokes the GAA API routines to make authorization decisions, under the assumption that the administrator of the resources has specified the local policy regarding the use of the resources by means of ACL files. These files are stored in an Authorization Database, maintained by the Printer Manager. 14.3.1 Conditions Administrators will be more willing to grant access to the printers if they can restrict the access to the resources to users or organizations they trust. Further, the administrators should be able to specify time availability, restrictions on resources consumed by the clients and accounting for the consumed resources. To specify these limits, the Printer Manager uses generic and specific ACL conditions. The following are the examples of generic conditions, that might be used by the Printer manager application to express security policies: a) Time window, expressed as a time interval within which access to printer is allowed. Example: time_window : 8AM-6PM b) Time window, expressed as days of week when access to the printer is allowed. Example: time_day : Monday-Friday c) Location, expressed as hosts from which the access is allowed. Example: location : DNS_*_ORG.EDU d) Payment required prior to submitting print job Example: payment : $0.05/page e) Quota, such as job size, expressed as maximum size in mega bytes. Example: quota : 10_MB The following are the examples of application-specific conditions, that might be used by the Printer manager application to express security policies: a) Printer load, expressed as maximum number of jobs that allowed to be submitted to a printer. Example: printer_load : 20 Ryutov/Neuman Expires February 1999 b) While a job is waiting to be printed or has been started but not yet completed, the original submitter shall be able to cancel the job entirely. The condition specifies who can perform this operation. Example: who : owner 14.3.2 Authorization walk-through Here we present two authorization scenarios to demonstrate the use of the authorization framework for the case of printing a document. Assume Kerberos V5 is used for principal authentication. Assume that printer ps12a has the following ordered ACL, stored in the Printer Manager Authorization Database: USER kerberos.v5 tom@ORG.EDU < PRINTER : submit_print_job > time_window : 8AM-8PM printer_load : 20 ; GROUP kerberos.v5 operators@ORG.EDU USER kerberos.v5 john@ORG.EDU < PRINTER : * > < DEVICE : * > ; ANYBODY < PRINTER: view_printer_capabilities > ; Let's consider a request from user Tom who is connecting from the ORG.EDU domain to print a document on the printer ps_12a on Monday at 7:30 PM. When a client process running on the behalf of user "Tom" contacts a Printer Manager with the request to submit_print_job to printer "ps_12a", the Printer Manager first calls gaa_get_object_eacl function to obtain a handle to the ACL of printer "ps_12a". The upcall function for retrieving the ACL for the specified object from the authorization database system is passed to the GAA API and is being called by gaa_get_object_eacl, which returns the ACL handle. The printer manager must place the principal's authenticated identity in the security context to pass into the gaa_check_authorization function. This context may be constructed according to the first or second scenario, described in Section 12. If Tom is authenticated successfully, then the verified identity credential is placed into the security context, specifying Tom as the Kerberos principal tom@ORG.EDU. The gaa_check_authorization function is called by the Printer Manager, which asks if "Tom" is authorized to submit_print_job to printer "ps_12a". In evaluating the ACL, the first entry applies. It grants the requested operation, but there two conditions that must be evaluated. The first condition time_window : 8AM-8PM is generic and is evaluated directly by the GAA API. Since, the request was issued on Monday at 7:30 PM this condition is satisfied. The second condition printer_load : 20 is Printer Manager-specific. If the security context passed by Printer Manager defined a condition Ryutov/Neuman Expires February 1999 evaluation function for upcall, then this function is invoked and if this condition is met then the final answer is YES (authorized) and detailed answer contains: Authorization expiration time : 8PM Assume that authentication credential has expiration time 9PM. Allowed operation: submit_print_job List of conditions: time_window : 8AM-8PM printer_load : 20 Both conditions are marked as evaluated and met. During the execution of the task the Printer Manager is enforcing the limits imposed on the local resources and authorization time. If the corresponding upcall function was not passed to the GAA API, the answer is MAYBE and detailed answer contains: Authorization expiration time : 8PM Allowed operation: submit_print_job List of conditions: time_window : 8AM-8PM printer_load : 20 The first condition is marked as evaluated and met; the second condition is marked as not evaluated and must be checked by the Printer Manager. Next, we present an authorization scenario where additional credentials are needed. Let's consider a request from the same user Tom to change priority of the job he has successfully submitted on the printer ps_12a on Monday at 7:31 PM. The Printer Manager calls the gaa_check_authorization function with the request for user "Tom" to change_print_job_attributes on printer "ps_12a". In ACL evaluation, the first entry applies but does not grant this operation. The temporary answer is NO (not authorized). The second entry grants this permission. If the security context defines a credential retrieval function for upcall, then this function is invoked and if either a group "operators" membership credential or delegated credential from user "John" for "Tom" is obtained, then the final answer is YES. If the credential retrieval upcall function was not passed to the GAA API, the answer is NO. 15. References [1] B.C. Neuman. Proxy-based authorization and accounting for distributed systems. Proceedings of the 13th International Conference on Distributed Computing Systems, Pittsburgh, May 1993. [2] B.C. Neuman and Theodore Ts'o. Kerberos: An authentication service for computer networks. IEEE Communications Magazine, pages 33-38, September 1994 Ryutov/Neuman Expires February 1999 [3] M. Blaze, J. Feigenbaum and J. Lacy. Decentralized Trust Management. in Proc. IEEE Symp. on Security and Privacy, IEEE Computer Press, Los Angeles, pages 164-173, 1996. [4] Large Scale Multicast Applications (lsma) working group. Taxonomy of Communication Requirements for Large-scale Multicast Applications. Internet draft. [5] Department of Defence National Computer Security Center. Department of Defence Trusted Computer system Evaluation Criteria, December 1985. DoD 5200.28-STD 16. Acknowledgments Gene Tsudik, Brian Tung, Bapa Rao, Ilia Ovsiannikov and the Xerox IS team have contributed to discussion of ideas and material in this draft. 17. Authors' Addresses Tatyana Ryutov Clifford Neuman USC/Information Sciences Institute 4676 Admiralty Way Suite 1001 Marina del Rey, CA 90292-6695 Phone: +1 310 822 1511 E-Mail: {tryutov, bcn}@isi.edu Ryutov/Neuman Expires February 1999