RFC 2743
Generic Security Service Application Program Interface Version 2, Update 1
Network Working Group J. Linn Request for Comments: 2743 RSA Laboratories Obsoletes: 2078 January 2000 Category: Standards Track Generic Security Service Application Program Interface Version 2, Update 1 Status of this Memo This document specifies an Internet standards track protocol for the Internet community, and requests discussion and suggestions for improvements. Please refer to the current edition of the "Internet Official Protocol Standards" (STD 1) for the standardization state and status of this protocol. Distribution of this memo is unlimited. Copyright Notice Copyright (C) The Internet Society (2000). All Rights Reserved.Abstract
The Generic Security Service Application Program Interface (GSS-API), Version 2, as defined in [RFC-2078], provides security services to callers in a generic fashion, supportable with a range of underlying mechanisms and technologies and hence allowing source-level portability of applications to different environments. This specification defines GSS-API services and primitives at a level independent of underlying mechanism and programming language environment, and is to be complemented by other, related specifications: documents defining specific parameter bindings for particular language environments documents defining token formats, protocols, and procedures to be implemented in order to realize GSS-API services atop particular security mechanisms This memo obsoletes [RFC-2078], making specific, incremental changes in response to implementation experience and liaison requests. It is intended, therefore, that this memo or a successor version thereto will become the basis for subsequent progression of the GSS-API specification on the standards track.
TABLE OF CONTENTS
1: GSS-API Characteristics and Concepts . . . . . . . . . . . . 4 1.1: GSS-API Constructs . . . . . . . . . . . . . . . . . . . . 6 1.1.1: Credentials . . . . . . . . . . . . . . . . . . . . . . 6 1.1.1.1: Credential Constructs and Concepts . . . . . . . . . . 6 1.1.1.2: Credential Management . . . . . . . . . . . . . . . . 7 1.1.1.3: Default Credential Resolution . . . . . . . . . . . . 8 1.1.2: Tokens . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.1.3: Security Contexts . . . . . . . . . . . . . . . . . . . 11 1.1.4: Mechanism Types . . . . . . . . . . . . . . . . . . . . 12 1.1.5: Naming . . . . . . . . . . . . . . . . . . . . . . . . 13 1.1.6: Channel Bindings . . . . . . . . . . . . . . . . . . . 16 1.2: GSS-API Features and Issues . . . . . . . . . . . . . . . 17 1.2.1: Status Reporting and Optional Service Support . . . . 17 1.2.1.1: Status Reporting . . . . . . . . . . . . . . . . . . . 17 1.2.1.2: Optional Service Support . . . . . . . . . . . . . . . 19 1.2.2: Per-Message Security Service Availability . . . . . . . 20 1.2.3: Per-Message Replay Detection and Sequencing . . . . . . 21 1.2.4: Quality of Protection . . . . . . . . . . . . . . . . . 24 1.2.5: Anonymity Support . . . . . . . . . . . . . . . . . . . 25 1.2.6: Initialization . . . . . . . . . . . . . . . . . . . . . 25 1.2.7: Per-Message Protection During Context Establishment . . 26 1.2.8: Implementation Robustness . . . . . . . . . . . . . . . 27 1.2.9: Delegation . . . . . . . . . . . . . . . . . . . . . . . 28 1.2.10: Interprocess Context Transfer . . . . . . . . . . . . . 28 2: Interface Descriptions . . . . . . . . . . . . . . . . . . 29 2.1: Credential management calls . . . . . . . . . . . . . . . 31 2.1.1: GSS_Acquire_cred call . . . . . . . . . . . . . . . . . 31 2.1.2: GSS_Release_cred call . . . . . . . . . . . . . . . . . 34 2.1.3: GSS_Inquire_cred call . . . . . . . . . . . . . . . . . 35 2.1.4: GSS_Add_cred call . . . . . . . . . . . . . . . . . . . 37 2.1.5: GSS_Inquire_cred_by_mech call . . . . . . . . . . . . . 40 2.2: Context-level calls . . . . . . . . . . . . . . . . . . . 41 2.2.1: GSS_Init_sec_context call . . . . . . . . . . . . . . . 42 2.2.2: GSS_Accept_sec_context call . . . . . . . . . . . . . . 49 2.2.3: GSS_Delete_sec_context call . . . . . . . . . . . . . . 53 2.2.4: GSS_Process_context_token call . . . . . . . . . . . . 54 2.2.5: GSS_Context_time call . . . . . . . . . . . . . . . . . 55 2.2.6: GSS_Inquire_context call . . . . . . . . . . . . . . . 56 2.2.7: GSS_Wrap_size_limit call . . . . . . . . . . . . . . . 57 2.2.8: GSS_Export_sec_context call . . . . . . . . . . . . . . 59 2.2.9: GSS_Import_sec_context call . . . . . . . . . . . . . . 61 2.3: Per-message calls . . . . . . . . . . . . . . . . . . . . 62 2.3.1: GSS_GetMIC call . . . . . . . . . . . . . . . . . . . . 63 2.3.2: GSS_VerifyMIC call . . . . . . . . . . . . . . . . . . 64 2.3.3: GSS_Wrap call . . . . . . . . . . . . . . . . . . . . . 65 2.3.4: GSS_Unwrap call . . . . . . . . . . . . . . . . . . . . 66
2.4: Support calls . . . . . . . . . . . . . . . . . . . . . . 68 2.4.1: GSS_Display_status call . . . . . . . . . . . . . . . . 68 2.4.2: GSS_Indicate_mechs call . . . . . . . . . . . . . . . . 69 2.4.3: GSS_Compare_name call . . . . . . . . . . . . . . . . . 70 2.4.4: GSS_Display_name call . . . . . . . . . . . . . . . . . 71 2.4.5: GSS_Import_name call . . . . . . . . . . . . . . . . . 72 2.4.6: GSS_Release_name call . . . . . . . . . . . . . . . . . 73 2.4.7: GSS_Release_buffer call . . . . . . . . . . . . . . . . 74 2.4.8: GSS_Release_OID_set call . . . . . . . . . . . . . . . 74 2.4.9: GSS_Create_empty_OID_set call . . . . . . . . . . . . . 75 2.4.10: GSS_Add_OID_set_member call . . . . . . . . . . . . . . 76 2.4.11: GSS_Test_OID_set_member call . . . . . . . . . . . . . 76 2.4.12: GSS_Inquire_names_for_mech call . . . . . . . . . . . . 77 2.4.13: GSS_Inquire_mechs_for_name call . . . . . . . . . . . . 77 2.4.14: GSS_Canonicalize_name call . . . . . . . . . . . . . . 78 2.4.15: GSS_Export_name call . . . . . . . . . . . . . . . . . 79 2.4.16: GSS_Duplicate_name call . . . . . . . . . . . . . . . . 80 3: Data Structure Definitions for GSS-V2 Usage . . . . . . . . 81 3.1: Mechanism-Independent Token Format . . . . . . . . . . . . 81 3.2: Mechanism-Independent Exported Name Object Format . . . . 84 4: Name Type Definitions . . . . . . . . . . . . . . . . . . . 85 4.1: Host-Based Service Name Form . . . . . . . . . . . . . . . 85 4.2: User Name Form . . . . . . . . . . . . . . . . . . . . . . 86 4.3: Machine UID Form . . . . . . . . . . . . . . . . . . . . . 87 4.4: String UID Form . . . . . . . . . . . . . . . . . . . . . 87 4.5: Anonymous Nametype . . . . . . . . . . . . . . . . . . . . 87 4.6: GSS_C_NO_OID . . . . . . . . . . . . . . . . . . . . . . . 88 4.7: Exported Name Object . . . . . . . . . . . . . . . . . . . 88 4.8: GSS_C_NO_NAME . . . . . . . . . . . . . . . . . . . . . . 88 5: Mechanism-Specific Example Scenarios . . . . . . . . . . . 88 5.1: Kerberos V5, single-TGT . . . . . . . . . . . . . . . . . 89 5.2: Kerberos V5, double-TGT . . . . . . . . . . . . . . . . . 89 5.3: X.509 Authentication Framework . . . . . . . . . . . . . 90 6: Security Considerations . . . . . . . . . . . . . . . . . . 91 7: Related Activities . . . . . . . . . . . . . . . . . . . . 92 8: Referenced Documents . . . . . . . . . . . . . . . . . . . 93 Appendix A: Mechanism Design Constraints . . . . . . . . . . . 94 Appendix B: Compatibility with GSS-V1 . . . . . . . . . . . . . 94 Appendix C: Changes Relative to RFC-2078 . . . . . . . . . . . 96 Author's Address . . . . . . . . . . . . . . . . . . . . . . .100 Full Copyright Statement . . . . . . . . . . . . . . . . . . .101
1: GSS-API Characteristics and Concepts
GSS-API operates in the following paradigm. A typical GSS-API caller is itself a communications protocol, calling on GSS-API in order to protect its communications with authentication, integrity, and/or confidentiality security services. A GSS-API caller accepts tokens provided to it by its local GSS-API implementation and transfers the tokens to a peer on a remote system; that peer passes the received tokens to its local GSS-API implementation for processing. The security services available through GSS-API in this fashion are implementable (and have been implemented) over a range of underlying mechanisms based on secret-key and public-key cryptographic technologies. The GSS-API separates the operations of initializing a security context between peers, achieving peer entity authentication (GSS_Init_sec_context() and GSS_Accept_sec_context() calls), from the operations of providing per-message data origin authentication and data integrity protection (GSS_GetMIC() and GSS_VerifyMIC() calls) for messages subsequently transferred in conjunction with that context. (The definition for the peer entity authentication service, and other definitions used in this document, corresponds to that provided in [ISO-7498-2].) When establishing a security context, the GSS-API enables a context initiator to optionally permit its credentials to be delegated, meaning that the context acceptor may initiate further security contexts on behalf of the initiating caller. Per-message GSS_Wrap() and GSS_Unwrap() calls provide the data origin authentication and data integrity services which GSS_GetMIC() and GSS_VerifyMIC() offer, and also support selection of confidentiality services as a caller option. Additional calls provide supportive functions to the GSS-API's users. The following paragraphs provide an example illustrating the dataflows involved in use of the GSS-API by a client and server in a mechanism-independent fashion, establishing a security context and transferring a protected message. The example assumes that credential acquisition has already been completed. The example also assumes that the underlying authentication technology is capable of authenticating a client to a server using elements carried within a single token, and of authenticating the server to the client (mutual authentication) with a single returned token; this assumption holds for some presently-documented CAT mechanisms but is not necessarily true for other cryptographic technologies and associated protocols. The client calls GSS_Init_sec_context() to establish a security context to the server identified by targ_name, and elects to set the mutual_req_flag so that mutual authentication is performed in the course of context establishment. GSS_Init_sec_context() returns an
output_token to be passed to the server, and indicates GSS_S_CONTINUE_NEEDED status pending completion of the mutual authentication sequence. Had mutual_req_flag not been set, the initial call to GSS_Init_sec_context() would have returned GSS_S_COMPLETE status. The client sends the output_token to the server. The server passes the received token as the input_token parameter to GSS_Accept_sec_context(). GSS_Accept_sec_context indicates GSS_S_COMPLETE status, provides the client's authenticated identity in the src_name result, and provides an output_token to be passed to the client. The server sends the output_token to the client. The client passes the received token as the input_token parameter to a successor call to GSS_Init_sec_context(), which processes data included in the token in order to achieve mutual authentication from the client's viewpoint. This call to GSS_Init_sec_context() returns GSS_S_COMPLETE status, indicating successful mutual authentication and the completion of context establishment for this example. The client generates a data message and passes it to GSS_Wrap(). GSS_Wrap() performs data origin authentication, data integrity, and (optionally) confidentiality processing on the message and encapsulates the result into output_message, indicating GSS_S_COMPLETE status. The client sends the output_message to the server. The server passes the received message to GSS_Unwrap(). GSS_Unwrap() inverts the encapsulation performed by GSS_Wrap(), deciphers the message if the optional confidentiality feature was applied, and validates the data origin authentication and data integrity checking quantities. GSS_Unwrap() indicates successful validation by returning GSS_S_COMPLETE status along with the resultant output_message. For purposes of this example, we assume that the server knows by out-of-band means that this context will have no further use after one protected message is transferred from client to server. Given this premise, the server now calls GSS_Delete_sec_context() to flush context-level information. Optionally, the server-side application may provide a token buffer to GSS_Delete_sec_context(), to receive a context_token to be transferred to the client in order to request that client-side context-level information be deleted. If a context_token is transferred, the client passes the context_token to GSS_Process_context_token(), which returns GSS_S_COMPLETE status after deleting context-level information at the client system.
The GSS-API design assumes and addresses several basic goals,
including:
Mechanism independence: The GSS-API defines an interface to
cryptographically implemented strong authentication and other
security services at a generic level which is independent of
particular underlying mechanisms. For example, GSS-API-provided
services have been implemented using secret-key technologies
(e.g., Kerberos, per [RFC-1964]) and with public-key approaches
(e.g., SPKM, per [RFC-2025]).
Protocol environment independence: The GSS-API is independent of
the communications protocol suites with which it is employed,
permitting use in a broad range of protocol environments. In
appropriate environments, an intermediate implementation "veneer"
which is oriented to a particular communication protocol may be
interposed between applications which call that protocol and the
GSS-API (e.g., as defined in [RFC-2203] for Open Network Computing
Remote Procedure Call (RPC)), thereby invoking GSS-API facilities
in conjunction with that protocol's communications invocations.
Protocol association independence: The GSS-API's security context
construct is independent of communications protocol association
constructs. This characteristic allows a single GSS-API
implementation to be utilized by a variety of invoking protocol
modules on behalf of those modules' calling applications. GSS-API
services can also be invoked directly by applications, wholly
independent of protocol associations.
Suitability to a range of implementation placements: GSS-API
clients are not constrained to reside within any Trusted Computing
Base (TCB) perimeter defined on a system where the GSS-API is
implemented; security services are specified in a manner suitable
to both intra-TCB and extra-TCB callers.
1.1: GSS-API Constructs
This section describes the basic elements comprising the GSS-API.
1.1.1: Credentials
1.1.1.1: Credential Constructs and Concepts
Credentials provide the prerequisites which permit GSS-API peers to
establish security contexts with each other. A caller may designate
that the credential elements which are to be applied for context
initiation or acceptance be selected by default. Alternately, those
GSS-API callers which need to make explicit selection of particular
credentials structures may make references to those credentials
through GSS-API-provided credential handles ("cred_handles"). In all
cases, callers' credential references are indirect, mediated by GSS-
API implementations and not requiring callers to access the selected
credential elements.
A single credential structure may be used to initiate outbound
contexts and to accept inbound contexts. Callers needing to operate
in only one of these modes may designate this fact when credentials
are acquired for use, allowing underlying mechanisms to optimize
their processing and storage requirements. The credential elements
defined by a particular mechanism may contain multiple cryptographic
keys, e.g., to enable authentication and message encryption to be
performed with different algorithms.
A GSS-API credential structure may contain multiple credential
elements, each containing mechanism-specific information for a
particular underlying mechanism (mech_type), but the set of elements
within a given credential structure represent a common entity. A
credential structure's contents will vary depending on the set of
mech_types supported by a particular GSS-API implementation. Each
credential element identifies the data needed by its mechanism in
order to establish contexts on behalf of a particular principal, and
may contain separate credential references for use in context
initiation and context acceptance. Multiple credential elements
within a given credential having overlapping combinations of
mechanism, usage mode, and validity period are not permitted.
Commonly, a single mech_type will be used for all security contexts
established by a particular initiator to a particular target. A major
motivation for supporting credential sets representing multiple
mech_types is to allow initiators on systems which are equipped to
handle multiple types to initiate contexts to targets on other
systems which can accommodate only a subset of the set supported at
the initiator's system.
1.1.1.2: Credential Management
It is the responsibility of underlying system-specific mechanisms and
OS functions below the GSS-API to ensure that the ability to acquire
and use credentials associated with a given identity is constrained
to appropriate processes within a system. This responsibility should
be taken seriously by implementors, as the ability for an entity to
utilize a principal's credentials is equivalent to the entity's
ability to successfully assert that principal's identity.
Once a set of GSS-API credentials is established, the transferability
of that credentials set to other processes or analogous constructs
within a system is a local matter, not defined by the GSS-API. An
example local policy would be one in which any credentials received
as a result of login to a given user account, or of delegation of
rights to that account, are accessible by, or transferable to,
processes running under that account.
The credential establishment process (particularly when performed on
behalf of users rather than server processes) is likely to require
access to passwords or other quantities which should be protected
locally and exposed for the shortest time possible. As a result, it
will often be appropriate for preliminary credential establishment to
be performed through local means at user login time, with the
result(s) cached for subsequent reference. These preliminary
credentials would be set aside (in a system-specific fashion) for
subsequent use, either:
to be accessed by an invocation of the GSS-API GSS_Acquire_cred()
call, returning an explicit handle to reference that credential
to comprise default credential elements to be installed, and to be
used when default credential behavior is requested on behalf of a
process
1.1.1.3: Default Credential Resolution
The GSS_Init_sec_context() and GSS_Accept_sec_context() routines
allow the value GSS_C_NO_CREDENTIAL to be specified as their
credential handle parameter. This special credential handle
indicates a desire by the application to act as a default principal.
In support of application portability, support for the default
resolution behavior described below for initiator credentials
(GSS_Init_sec_context() usage) is mandated; support for the default
resolution behavior described below for acceptor credentials
(GSS_Accept_sec_context() usage) is recommended. If default
credential resolution fails, GSS_S_NO_CRED status is to be returned.
GSS_Init_sec_context:
(i) If there is only a single principal capable of initiating
security contexts that the application is authorized to act on
behalf of, then that principal shall be used, otherwise
(ii) If the platform maintains a concept of a default network-
identity, and if the application is authorized to act on behalf
of that identity for the purpose of initiating security
contexts, then the principal corresponding to that identity
shall be used, otherwise
(iii) If the platform maintains a concept of a default local
identity, and provides a means to map local identities into
network-identities, and if the application is authorized to act
on behalf of the network-identity image of the default local
identity for the purpose of initiating security contexts, then
the principal corresponding to that identity shall be used,
otherwise
(iv) A user-configurable default identity should be used.
GSS_Accept_sec_context:
(i) If there is only a single authorized principal identity
capable of accepting security contexts, then that principal
shall be used, otherwise
(ii) If the mechanism can determine the identity of the target
principal by examining the context-establishment token, and if
the accepting application is authorized to act as that
principal for the purpose of accepting security contexts, then
that principal identity shall be used, otherwise
(iii) If the mechanism supports context acceptance by any
principal, and mutual authentication was not requested, any
principal that the application is authorized to accept security
contexts under may be used, otherwise
(iv) A user-configurable default identity shall be used.
The purpose of the above rules is to allow security contexts to be
established by both initiator and acceptor using the default behavior
wherever possible. Applications requesting default behavior are
likely to be more portable across mechanisms and platforms than those
that use GSS_Acquire_cred() to request a specific identity.
1.1.2: Tokens
Tokens are data elements transferred between GSS-API callers, and are
divided into two classes. Context-level tokens are exchanged in order
to establish and manage a security context between peers. Per-message
tokens relate to an established context and are exchanged to provide
protective security services (i.e., data origin authentication, integrity, and optional confidentiality) for corresponding data messages. The first context-level token obtained from GSS_Init_sec_context() is required to indicate at its very beginning a globally-interpretable mechanism identifier, i.e., an Object Identifier (OID) of the security mechanism. The remaining part of this token as well as the whole content of all other tokens are specific to the particular underlying mechanism used to support the GSS-API. Section 3.1 of this document provides, for designers of GSS-API mechanisms, the description of the header of the first context-level token which is then followed by mechanism-specific information. Tokens' contents are opaque from the viewpoint of GSS-API callers. They are generated within the GSS-API implementation at an end system, provided to a GSS-API caller to be transferred to the peer GSS-API caller at a remote end system, and processed by the GSS-API implementation at that remote end system. Context-level tokens may be output by GSS-API calls (and should be transferred to GSS-API peers) whether or not the calls' status indicators indicate successful completion. Per-message tokens, in contrast, are to be returned only upon successful completion of per- message calls. Zero-length tokens are never returned by GSS routines for transfer to a peer. Token transfer may take place in an in-band manner, integrated into the same protocol stream used by the GSS-API callers for other data transfers, or in an out-of-band manner across a logically separate channel. Different GSS-API tokens are used for different purposes (e.g., context initiation, context acceptance, protected message data on an established context), and it is the responsibility of a GSS-API caller receiving tokens to distinguish their types, associate them with corresponding security contexts, and pass them to appropriate GSS-API processing routines. Depending on the caller protocol environment, this distinction may be accomplished in several ways. The following examples illustrate means through which tokens' types may be distinguished: - implicit tagging based on state information (e.g., all tokens on a new association are considered to be context establishment tokens until context establishment is completed, at which point all tokens are considered to be wrapped data objects for that context),
- explicit tagging at the caller protocol level,
- a hybrid of these approaches.
Commonly, the encapsulated data within a token includes internal
mechanism-specific tagging information, enabling mechanism-level
processing modules to distinguish tokens used within the mechanism
for different purposes. Such internal mechanism-level tagging is
recommended to mechanism designers, and enables mechanisms to
determine whether a caller has passed a particular token for
processing by an inappropriate GSS-API routine.
Development of GSS-API mechanisms based on a particular underlying
cryptographic technique and protocol (i.e., conformant to a specific
GSS-API mechanism definition) does not necessarily imply that GSS-API
callers using that GSS-API mechanism will be able to interoperate
with peers invoking the same technique and protocol outside the GSS-
API paradigm, or with peers implementing a different GSS-API
mechanism based on the same underlying technology. The format of
GSS-API tokens defined in conjunction with a particular mechanism,
and the techniques used to integrate those tokens into callers'
protocols, may not be interoperable with the tokens used by non-GSS-
API callers of the same underlying technique.
1.1.3: Security Contexts
Security contexts are established between peers, using credentials
established locally in conjunction with each peer or received by
peers via delegation. Multiple contexts may exist simultaneously
between a pair of peers, using the same or different sets of
credentials. Coexistence of multiple contexts using different
credentials allows graceful rollover when credentials expire.
Distinction among multiple contexts based on the same credentials
serves applications by distinguishing different message streams in a
security sense.
The GSS-API is independent of underlying protocols and addressing
structure, and depends on its callers to transport GSS-API-provided
data elements. As a result of these factors, it is a caller
responsibility to parse communicated messages, separating GSS-API-
related data elements from caller-provided data. The GSS-API is
independent of connection vs. connectionless orientation of the
underlying communications service.
No correlation between security context and communications protocol
association is dictated. (The optional channel binding facility,
discussed in Section 1.1.6 of this document, represents an
intentional exception to this rule, supporting additional protection
features within GSS-API supporting mechanisms.) This separation allows the GSS-API to be used in a wide range of communications environments, and also simplifies the calling sequences of the individual calls. In many cases (depending on underlying security protocol, associated mechanism, and availability of cached information), the state information required for context setup can be sent concurrently with initial signed user data, without interposing additional message exchanges. Messages may be protected and transferred in both directions on an established GSS-API security context concurrently; protection of messages in one direction does not interfere with protection of messages in the reverse direction. GSS-API implementations are expected to retain inquirable context data on a context until the context is released by a caller, even after the context has expired, although underlying cryptographic data elements may be deleted after expiration in order to limit their exposure.1.1.4: Mechanism Types
In order to successfully establish a security context with a target peer, it is necessary to identify an appropriate underlying mechanism type (mech_type) which both initiator and target peers support. The definition of a mechanism embodies not only the use of a particular cryptographic technology (or a hybrid or choice among alternative cryptographic technologies), but also definition of the syntax and semantics of data element exchanges which that mechanism will employ in order to support security services. It is recommended that callers initiating contexts specify the "default" mech_type value, allowing system-specific functions within or invoked by the GSS-API implementation to select the appropriate mech_type, but callers may direct that a particular mech_type be employed when necessary. For GSS-API purposes, the phrase "negotiating mechanism" refers to a mechanism which itself performs negotiation in order to select a concrete mechanism which is shared between peers and is then used for context establishment. Only those mechanisms which are defined in their specifications as negotiating mechanisms are to yield selected mechanisms with different identifier values than the value which is input by a GSS-API caller, except for the case of a caller requesting the "default" mech_type. The means for identifying a shared mech_type to establish a security context with a peer will vary in different environments and circumstances; examples include (but are not limited to):
use of a fixed mech_type, defined by configuration, within an
environment
syntactic convention on a target-specific basis, through
examination of a target's name lookup of a target's name in a
naming service or other database in order to identify mech_types
supported by that target
explicit negotiation between GSS-API callers in advance of
security context setup
use of a negotiating mechanism
When transferred between GSS-API peers, mech_type specifiers (per
Section 3 of this document, represented as Object Identifiers (OIDs))
serve to qualify the interpretation of associated tokens. (The
structure and encoding of Object Identifiers is defined in [ISOIEC-
8824] and [ISOIEC-8825].) Use of hierarchically structured OIDs
serves to preclude ambiguous interpretation of mech_type specifiers.
The OID representing the DASS ([RFC-1507]) MechType, for example, is
1.3.12.2.1011.7.5, and that of the Kerberos V5 mechanism ([RFC-
1964]), having been advanced to the level of Proposed Standard, is
1.2.840.113554.1.2.2.
1.1.5: Naming
The GSS-API avoids prescribing naming structures, treating the names
which are transferred across the interface in order to initiate and
accept security contexts as opaque objects. This approach supports
the GSS-API's goal of implementability atop a range of underlying
security mechanisms, recognizing the fact that different mechanisms
process and authenticate names which are presented in different
forms. Generalized services offering translation functions among
arbitrary sets of naming environments are outside the scope of the
GSS-API; availability and use of local conversion functions to
translate among the naming formats supported within a given end
system is anticipated.
Different classes of name representations are used in conjunction
with different GSS-API parameters:
- Internal form (denoted in this document by INTERNAL NAME),
opaque to callers and defined by individual GSS-API
implementations. GSS-API implementations supporting multiple
namespace types must maintain internal tags to disambiguate the
interpretation of particular names. A Mechanism Name (MN) is a
special case of INTERNAL NAME, guaranteed to contain elements
corresponding to one and only one mechanism; calls which are
guaranteed to emit MNs or which require MNs as input are so
identified within this specification.
- Contiguous string ("flat") form (denoted in this document by
OCTET STRING); accompanied by OID tags identifying the namespace
to which they correspond. Depending on tag value, flat names may
or may not be printable strings for direct acceptance from and
presentation to users. Tagging of flat names allows GSS-API
callers and underlying GSS-API mechanisms to disambiguate name
types and to determine whether an associated name's type is one
which they are capable of processing, avoiding aliasing problems
which could result from misinterpreting a name of one type as a
name of another type.
- The GSS-API Exported Name Object, a special case of flat name
designated by a reserved OID value, carries a canonicalized form
of a name suitable for binary comparisons.
In addition to providing means for names to be tagged with types,
this specification defines primitives to support a level of naming
environment independence for certain calling applications. To provide
basic services oriented towards the requirements of callers which
need not themselves interpret the internal syntax and semantics of
names, GSS-API calls for name comparison (GSS_Compare_name()),
human-readable display (GSS_Display_name()), input conversion
(GSS_Import_name()), internal name deallocation (GSS_Release_name()),
and internal name duplication (GSS_Duplicate_name()) functions are
defined. (It is anticipated that these proposed GSS-API calls will be
implemented in many end systems based on system-specific name
manipulation primitives already extant within those end systems;
inclusion within the GSS-API is intended to offer GSS-API callers a
portable means to perform specific operations, supportive of
authorization and audit requirements, on authenticated names.)
GSS_Import_name() implementations can, where appropriate, support
more than one printable syntax corresponding to a given namespace
(e.g., alternative printable representations for X.500 Distinguished
Names), allowing flexibility for their callers to select among
alternative representations. GSS_Display_name() implementations
output a printable syntax selected as appropriate to their
operational environments; this selection is a local matter. Callers
desiring portability across alternative printable syntaxes should
refrain from implementing comparisons based on printable name forms
and should instead use the GSS_Compare_name() call to determine
whether or not one internal-format name matches another.
When used in large access control lists, the overhead of invoking GSS_Import_name() and GSS_Compare_name() on each name from the ACL may be prohibitive. As an alternative way of supporting this case, GSS-API defines a special form of the contiguous string name which may be compared directly (e.g., with memcmp()). Contiguous names suitable for comparison are generated by the GSS_Export_name() routine, which requires an MN as input. Exported names may be re- imported by the GSS_Import_name() routine, and the resulting internal name will also be an MN. The symbolic constant GSS_C_NT_EXPORT_NAME identifies the "export name" type. Structurally, an exported name object consists of a header containing an OID identifying the mechanism that authenticated the name, and a trailer containing the name itself, where the syntax of the trailer is defined by the individual mechanism specification. The precise format of an exported name is defined in Section 3.2 of this specification. Note that the results obtained by using GSS_Compare_name() will in general be different from those obtained by invoking GSS_Canonicalize_name() and GSS_Export_name(), and then comparing the exported names. The first series of operations determines whether two (unauthenticated) names identify the same principal; the second whether a particular mechanism would authenticate them as the same principal. These two operations will in general give the same results only for MNs. The following diagram illustrates the intended dataflow among name- related GSS-API processing routines.
GSS-API library defaults
|
|
V text, for
text --------------> internal_name (IN) -----------> display only
import_name() / display_name()
/
/
/
accept_sec_context() /
| /
| /
| / canonicalize_name()
| /
| /
| /
| /
| /
| |
V V <---------------------
single mechanism import_name() exported name: flat
internal_name (MN) binary "blob" usable
----------------------> for access control
export_name()
1.1.6: Channel Bindings
The GSS-API accommodates the concept of caller-provided channel
binding ("chan_binding") information. Channel bindings are used to
strengthen the quality with which peer entity authentication is
provided during context establishment, by limiting the scope within
which an intercepted context establishment token can be reused by an
attacker. Specifically, they enable GSS-API callers to bind the
establishment of a security context to relevant characteristics
(e.g., addresses, transformed representations of encryption keys) of
the underlying communications channel, of protection mechanisms
applied to that communications channel, and to application-specific
data.
The caller initiating a security context must determine the
appropriate channel binding values to provide as input to the
GSS_Init_sec_context() call, and consistent values must be provided
to GSS_Accept_sec_context() by the context's target, in order for
both peers' GSS-API mechanisms to validate that received tokens
possess correct channel-related characteristics. Use or non-use of
the GSS-API channel binding facility is a caller option. GSS-API
mechanisms can operate in an environment where NULL channel bindings
are presented; mechanism implementors are encouraged, but not
required, to make use of caller-provided channel binding data within their mechanisms. Callers should not assume that underlying mechanisms provide confidentiality protection for channel binding information. When non-NULL channel bindings are provided by callers, certain mechanisms can offer enhanced security value by interpreting the bindings' content (rather than simply representing those bindings, or integrity check values computed on them, within tokens) and will therefore depend on presentation of specific data in a defined format. To this end, agreements among mechanism implementors are defining conventional interpretations for the contents of channel binding arguments, including address specifiers (with content dependent on communications protocol environment) for context initiators and acceptors. (These conventions are being incorporated in GSS-API mechanism specifications and into the GSS-API C language bindings specification.) In order for GSS-API callers to be portable across multiple mechanisms and achieve the full security functionality which each mechanism can provide, it is strongly recommended that GSS-API callers provide channel bindings consistent with these conventions and those of the networking environment in which they operate.1.2: GSS-API Features and Issues
This section describes aspects of GSS-API operations, of the security services which the GSS-API provides, and provides commentary on design issues.1.2.1: Status Reporting and Optional Service Support
1.2.1.1: Status Reporting
Each GSS-API call provides two status return values. Major_status values provide a mechanism-independent indication of call status (e.g., GSS_S_COMPLETE, GSS_S_FAILURE, GSS_S_CONTINUE_NEEDED), sufficient to drive normal control flow within the caller in a generic fashion. Table 1 summarizes the defined major_status return codes in tabular fashion. Sequencing-related informatory major_status codes (GSS_S_DUPLICATE_TOKEN, GSS_S_OLD_TOKEN, GSS_S_UNSEQ_TOKEN, and GSS_S_GAP_TOKEN) can be indicated in conjunction with either GSS_S_COMPLETE or GSS_S_FAILURE status for GSS-API per-message calls. For context establishment calls, these sequencing-related codes will be indicated only in conjunction with GSS_S_FAILURE status (never in
conjunction with GSS_S_COMPLETE or GSS_S_CONTINUE_NEEDED), and,
therefore, always correspond to fatal failures if encountered during
the context establishment phase.
Table 1: GSS-API Major Status Codes
FATAL ERROR CODES
GSS_S_BAD_BINDINGS channel binding mismatch
GSS_S_BAD_MECH unsupported mechanism requested
GSS_S_BAD_NAME invalid name provided
GSS_S_BAD_NAMETYPE name of unsupported type provided
GSS_S_BAD_STATUS invalid input status selector
GSS_S_BAD_SIG token had invalid integrity check
GSS_S_BAD_MIC preferred alias for GSS_S_BAD_SIG
GSS_S_CONTEXT_EXPIRED specified security context expired
GSS_S_CREDENTIALS_EXPIRED expired credentials detected
GSS_S_DEFECTIVE_CREDENTIAL defective credential detected
GSS_S_DEFECTIVE_TOKEN defective token detected
GSS_S_FAILURE failure, unspecified at GSS-API
level
GSS_S_NO_CONTEXT no valid security context specified
GSS_S_NO_CRED no valid credentials provided
GSS_S_BAD_QOP unsupported QOP value
GSS_S_UNAUTHORIZED operation unauthorized
GSS_S_UNAVAILABLE operation unavailable
GSS_S_DUPLICATE_ELEMENT duplicate credential element requested
GSS_S_NAME_NOT_MN name contains multi-mechanism elements
INFORMATORY STATUS CODES
GSS_S_COMPLETE normal completion
GSS_S_CONTINUE_NEEDED continuation call to routine
required
GSS_S_DUPLICATE_TOKEN duplicate per-message token
detected
GSS_S_OLD_TOKEN timed-out per-message token
detected
GSS_S_UNSEQ_TOKEN reordered (early) per-message token
detected
GSS_S_GAP_TOKEN skipped predecessor token(s)
detected
Minor_status provides more detailed status information which may
include status codes specific to the underlying security mechanism.
Minor_status values are not specified in this document.
GSS_S_CONTINUE_NEEDED major_status returns, and optional message outputs, are provided in GSS_Init_sec_context() and GSS_Accept_sec_context() calls so that different mechanisms' employment of different numbers of messages within their authentication sequences need not be reflected in separate code paths within calling applications. Instead, such cases are accommodated with sequences of continuation calls to GSS_Init_sec_context() and GSS_Accept_sec_context(). The same facility is used to encapsulate mutual authentication within the GSS-API's context initiation calls. For mech_types which require interactions with third-party servers in order to establish a security context, GSS-API context establishment calls may block pending completion of such third-party interactions. On the other hand, no GSS-API calls pend on serialized interactions with GSS-API peer entities. As a result, local GSS-API status returns cannot reflect unpredictable or asynchronous exceptions occurring at remote peers, and reflection of such status information is a caller responsibility outside the GSS-API.1.2.1.2: Optional Service Support
A context initiator may request various optional services at context establishment time. Each of these services is requested by setting a flag in the req_flags input parameter to GSS_Init_sec_context(). The optional services currently defined are: - Delegation - The (usually temporary) transfer of rights from initiator to acceptor, enabling the acceptor to authenticate itself as an agent of the initiator. - Mutual Authentication - In addition to the initiator authenticating its identity to the context acceptor, the context acceptor should also authenticate itself to the initiator. - Replay detection - In addition to providing message integrity services, GSS_GetMIC() and GSS_Wrap() should include message numbering information to enable GSS_VerifyMIC() and GSS_Unwrap() to detect if a message has been duplicated. - Out-of-sequence detection - In addition to providing message integrity services, GSS_GetMIC() and GSS_Wrap() should include message sequencing information to enable GSS_VerifyMIC() and GSS_Unwrap() to detect if a message has been received out of sequence.
- Anonymous authentication - The establishment of the security
context should not reveal the initiator's identity to the context
acceptor.
- Available per-message confidentiality - requests that per-
message confidentiality services be available on the context.
- Available per-message integrity - requests that per-message
integrity services be available on the context.
Any currently undefined bits within such flag arguments should be
ignored by GSS-API implementations when presented by an application,
and should be set to zero when returned to the application by the
GSS-API implementation.
Some mechanisms may not support all optional services, and some
mechanisms may only support some services in conjunction with others.
Both GSS_Init_sec_context() and GSS_Accept_sec_context() inform the
applications which services will be available from the context when
the establishment phase is complete, via the ret_flags output
parameter. In general, if the security mechanism is capable of
providing a requested service, it should do so, even if additional
services must be enabled in order to provide the requested service.
If the mechanism is incapable of providing a requested service, it
should proceed without the service, leaving the application to abort
the context establishment process if it considers the requested
service to be mandatory.
Some mechanisms may specify that support for some services is
optional, and that implementors of the mechanism need not provide it.
This is most commonly true of the confidentiality service, often
because of legal restrictions on the use of data-encryption, but may
apply to any of the services. Such mechanisms are required to send
at least one token from acceptor to initiator during context
establishment when the initiator indicates a desire to use such a
service, so that the initiating GSS-API can correctly indicate
whether the service is supported by the acceptor's GSS-API.
1.2.2: Per-Message Security Service Availability
When a context is established, two flags are returned to indicate the
set of per-message protection security services which will be
available on the context:
the integ_avail flag indicates whether per-message integrity and
data origin authentication services are available
the conf_avail flag indicates whether per-message confidentiality
services are available, and will never be returned TRUE unless the
integ_avail flag is also returned TRUE
GSS-API callers desiring per-message security services should check
the values of these flags at context establishment time, and must be
aware that a returned FALSE value for integ_avail means that
invocation of GSS_GetMIC() or GSS_Wrap() primitives on the associated
context will apply no cryptographic protection to user data messages.
The GSS-API per-message integrity and data origin authentication
services provide assurance to a receiving caller that protection was
applied to a message by the caller's peer on the security context,
corresponding to the entity named at context initiation. The GSS-API
per-message confidentiality service provides assurance to a sending
caller that the message's content is protected from access by
entities other than the context's named peer.
The GSS-API per-message protection service primitives, as the
category name implies, are oriented to operation at the granularity
of protocol data units. They perform cryptographic operations on the
data units, transfer cryptographic control information in tokens,
and, in the case of GSS_Wrap(), encapsulate the protected data unit.
As such, these primitives are not oriented to efficient data
protection for stream-paradigm protocols (e.g., Telnet) if
cryptography must be applied on an octet-by-octet basis.
1.2.3: Per-Message Replay Detection and Sequencing
Certain underlying mech_types offer support for replay detection
and/or sequencing of messages transferred on the contexts they
support. These optionally-selectable protection features are distinct
from replay detection and sequencing features applied to the context
establishment operation itself; the presence or absence of context-
level replay or sequencing features is wholly a function of the
underlying mech_type's capabilities, and is not selected or omitted
as a caller option.
The caller initiating a context provides flags (replay_det_req_flag
and sequence_req_flag) to specify whether the use of per-message
replay detection and sequencing features is desired on the context
being established. The GSS-API implementation at the initiator system
can determine whether these features are supported (and whether they
are optionally selectable) as a function of the selected mechanism,
without need for bilateral negotiation with the target. When enabled,
these features provide recipients with indicators as a result of
GSS-API processing of incoming messages, identifying whether those
messages were detected as duplicates or out-of-sequence. Detection of
such events does not prevent a suspect message from being provided to
a recipient; the appropriate course of action on a suspect message is
a matter of caller policy.
The semantics of the replay detection and sequencing services applied
to received messages, as visible across the interface which the GSS-
API provides to its clients, are as follows:
When replay_det_state is TRUE, the possible major_status returns for
well-formed and correctly signed messages are as follows:
1. GSS_S_COMPLETE, without concurrent indication of
GSS_S_DUPLICATE_TOKEN or GSS_S_OLD_TOKEN, indicates that the
message was within the window (of time or sequence space) allowing
replay events to be detected, and that the message was not a
replay of a previously-processed message within that window.
2. GSS_S_DUPLICATE_TOKEN indicates that the cryptographic
checkvalue on the received message was correct, but that the
message was recognized as a duplicate of a previously-processed
message. In addition to identifying duplicated tokens originated
by a context's peer, this status may also be used to identify
reflected copies of locally-generated tokens; it is recommended
that mechanism designers include within their protocols facilities
to detect and report such tokens.
3. GSS_S_OLD_TOKEN indicates that the cryptographic checkvalue on
the received message was correct, but that the message is too old
to be checked for duplication.
When sequence_state is TRUE, the possible major_status returns for
well-formed and correctly signed messages are as follows:
1. GSS_S_COMPLETE, without concurrent indication of
GSS_S_DUPLICATE_TOKEN, GSS_S_OLD_TOKEN, GSS_S_UNSEQ_TOKEN, or
GSS_S_GAP_TOKEN, indicates that the message was within the window
(of time or sequence space) allowing replay events to be detected,
that the message was not a replay of a previously-processed
message within that window, and that no predecessor sequenced
messages are missing relative to the last received message (if
any) processed on the context with a correct cryptographic
checkvalue.
2. GSS_S_DUPLICATE_TOKEN indicates that the integrity check value
on the received message was correct, but that the message was
recognized as a duplicate of a previously-processed message. In
addition to identifying duplicated tokens originated by a
context's peer, this status may also be used to identify reflected
copies of locally-generated tokens; it is recommended that
mechanism designers include within their protocols facilities to
detect and report such tokens.
3. GSS_S_OLD_TOKEN indicates that the integrity check value on the
received message was correct, but that the token is too old to be
checked for duplication.
4. GSS_S_UNSEQ_TOKEN indicates that the cryptographic checkvalue
on the received message was correct, but that it is earlier in a
sequenced stream than a message already processed on the context.
[Note: Mechanisms can be architected to provide a stricter form of
sequencing service, delivering particular messages to recipients
only after all predecessor messages in an ordered stream have been
delivered. This type of support is incompatible with the GSS-API
paradigm in which recipients receive all messages, whether in
order or not, and provide them (one at a time, without intra-GSS-
API message buffering) to GSS-API routines for validation. GSS-
API facilities provide supportive functions, aiding clients to
achieve strict message stream integrity in an efficient manner in
conjunction with sequencing provisions in communications
protocols, but the GSS-API does not offer this level of message
stream integrity service by itself.]
5. GSS_S_GAP_TOKEN indicates that the cryptographic checkvalue on
the received message was correct, but that one or more predecessor
sequenced messages have not been successfully processed relative
to the last received message (if any) processed on the context
with a correct cryptographic checkvalue.
As the message stream integrity features (especially sequencing) may
interfere with certain applications' intended communications
paradigms, and since support for such features is likely to be
resource intensive, it is highly recommended that mech_types
supporting these features allow them to be activated selectively on
initiator request when a context is established. A context initiator
and target are provided with corresponding indicators
(replay_det_state and sequence_state), signifying whether these
features are active on a given context.
An example mech_type supporting per-message replay detection could
(when replay_det_state is TRUE) implement the feature as follows: The
underlying mechanism would insert timestamps in data elements output
by GSS_GetMIC() and GSS_Wrap(), and would maintain (within a time-
limited window) a cache (qualified by originator-recipient pair)
identifying received data elements processed by GSS_VerifyMIC() and
GSS_Unwrap(). When this feature is active, exception status returns
(GSS_S_DUPLICATE_TOKEN, GSS_S_OLD_TOKEN) will be provided when
GSS_VerifyMIC() or GSS_Unwrap() is presented with a message which is either a detected duplicate of a prior message or which is too old to validate against a cache of recently received messages.1.2.4: Quality of Protection
Some mech_types provide their users with fine granularity control over the means used to provide per-message protection, allowing callers to trade off security processing overhead dynamically against the protection requirements of particular messages. A per-message quality-of-protection parameter (analogous to quality-of-service, or QOS) selects among different QOP options supported by that mechanism. On context establishment for a multi-QOP mech_type, context-level data provides the prerequisite data for a range of protection qualities. It is expected that the majority of callers will not wish to exert explicit mechanism-specific QOP control and will therefore request selection of a default QOP. Definitions of, and choices among, non- default QOP values are mechanism-specific, and no ordered sequences of QOP values can be assumed equivalent across different mechanisms. Meaningful use of non-default QOP values demands that callers be familiar with the QOP definitions of an underlying mechanism or mechanisms, and is therefore a non-portable construct. The GSS_S_BAD_QOP major_status value is defined in order to indicate that a provided QOP value is unsupported for a security context, most likely because that value is unrecognized by the underlying mechanism. In the interests of interoperability, mechanisms which allow optional support of particular QOP values shall satisfy one of the following conditions. Either: (i) All implementations of the mechanism are required to be capable of processing messages protected using any QOP value, regardless of whether they can apply protection corresponding to that QOP, or (ii) The set of mutually-supported receiver QOP values must be determined during context establishment, and messages may be protected by either peer using only QOP values from this mutually-supported set. NOTE: (i) is just a special-case of (ii), where implementations are required to support all QOP values on receipt.
1.2.5: Anonymity Support
In certain situations or environments, an application may wish to authenticate a peer and/or protect communications using GSS-API per- message services without revealing its own identity. For example, consider an application which provides read access to a research database, and which permits queries by arbitrary requestors. A client of such a service might wish to authenticate the service, to establish trust in the information received from it, but might not wish to disclose its identity to the service for privacy reasons. In ordinary GSS-API usage, a context initiator's identity is made available to the context acceptor as part of the context establishment process. To provide for anonymity support, a facility (input anon_req_flag to GSS_Init_sec_context()) is provided through which context initiators may request that their identity not be provided to the context acceptor. Mechanisms are not required to honor this request, but a caller will be informed (via returned anon_state indicator from GSS_Init_sec_context()) whether or not the request is honored. Note that authentication as the anonymous principal does not necessarily imply that credentials are not required in order to establish a context. Section 4.5 of this document defines the Object Identifier value used to identify an anonymous principal. Four possible combinations of anon_state and mutual_state are possible, with the following results: anon_state == FALSE, mutual_state == FALSE: initiator authenticated to target. anon_state == FALSE, mutual_state == TRUE: initiator authenticated to target, target authenticated to initiator. anon_state == TRUE, mutual_state == FALSE: initiator authenticated as anonymous principal to target. anon_state == TRUE, mutual_state == TRUE: initiator authenticated as anonymous principal to target, target authenticated to initiator.1.2.6: Initialization
No initialization calls (i.e., calls which must be invoked prior to invocation of other facilities in the interface) are defined in GSS- API. As an implication of this fact, GSS-API implementations must themselves be self-initializing.
1.2.7: Per-Message Protection During Context Establishment
A facility is defined in GSS-V2 to enable protection and buffering of data messages for later transfer while a security context's establishment is in GSS_S_CONTINUE_NEEDED status, to be used in cases where the caller side already possesses the necessary session key to enable this processing. Specifically, a new state Boolean, called prot_ready_state, is added to the set of information returned by GSS_Init_sec_context(), GSS_Accept_sec_context(), and GSS_Inquire_context(). For context establishment calls, this state Boolean is valid and interpretable when the associated major_status is either GSS_S_CONTINUE_NEEDED, or GSS_S_COMPLETE. Callers of GSS-API (both initiators and acceptors) can assume that per-message protection (via GSS_Wrap(), GSS_Unwrap(), GSS_GetMIC() and GSS_VerifyMIC()) is available and ready for use if either: prot_ready_state == TRUE, or major_status == GSS_S_COMPLETE, though mutual authentication (if requested) cannot be guaranteed until GSS_S_COMPLETE is returned. Callers making use of per-message protection services in advance of GSS_S_COMPLETE status should be aware of the possibility that a subsequent context establishment step may fail, and that certain context data (e.g., mech_type) as returned for subsequent calls may change. This approach achieves full, transparent backward compatibility for GSS-API V1 callers, who need not even know of the existence of prot_ready_state, and who will get the expected behavior from GSS_S_COMPLETE, but who will not be able to use per-message protection before GSS_S_COMPLETE is returned. It is not a requirement that GSS-V2 mechanisms ever return TRUE prot_ready_state before completion of context establishment (indeed, some mechanisms will not evolve usable message protection keys, especially at the context acceptor, before context establishment is complete). It is expected but not required that GSS-V2 mechanisms will return TRUE prot_ready_state upon completion of context establishment if they support per-message protection at all (however GSS-V2 applications should not assume that TRUE prot_ready_state will always be returned together with the GSS_S_COMPLETE major_status, since GSS-V2 implementations may continue to support GSS-V1 mechanism code, which will never return TRUE prot_ready_state). When prot_ready_state is returned TRUE, mechanisms shall also set those context service indicator flags (deleg_state, mutual_state, replay_det_state, sequence_state, anon_state, trans_state, conf_avail, integ_avail) which represent facilities confirmed, at that time, to be available on the context being established. In
situations where prot_ready_state is returned before GSS_S_COMPLETE, it is possible that additional facilities may be confirmed and subsequently indicated when GSS_S_COMPLETE is returned.1.2.8: Implementation Robustness
This section recommends aspects of GSS-API implementation behavior in the interests of overall robustness. Invocation of GSS-API calls is to incur no undocumented side effects visible at the GSS-API level. If a token is presented for processing on a GSS-API security context and that token generates a fatal error in processing or is otherwise determined to be invalid for that context, the context's state should not be disrupted for purposes of processing subsequent valid tokens. Certain local conditions at a GSS-API implementation (e.g., unavailability of memory) may preclude, temporarily or permanently, the successful processing of tokens on a GSS-API security context, typically generating GSS_S_FAILURE major_status returns along with locally-significant minor_status. For robust operation under such conditions, the following recommendations are made: Failing calls should free any memory they allocate, so that callers may retry without causing further loss of resources. Failure of an individual call on an established context should not preclude subsequent calls from succeeding on the same context. Whenever possible, it should be possible for GSS_Delete_sec_context() calls to be successfully processed even if other calls cannot succeed, thereby enabling context-related resources to be released. A failure of GSS_GetMIC() or GSS_Wrap() due to an attempt to use an unsupported QOP will not interfere with context validity, nor shall such a failure impact the ability of the application to subsequently invoke GSS_GetMIC() or GSS_Wrap() using a supported QOP. Any state information concerning sequencing of outgoing messages shall be unchanged by an unsuccessful call of GSS_GetMIC() or GSS_Wrap().
1.2.9: Delegation
The GSS-API allows delegation to be controlled by the initiating application via a Boolean parameter to GSS_Init_sec_context(), the routine that establishes a security context. Some mechanisms do not support delegation, and for such mechanisms attempts by an application to enable delegation are ignored. The acceptor of a security context for which the initiator enabled delegation will receive (via the delegated_cred_handle parameter of GSS_Accept_sec_context()) a credential handle that contains the delegated identity, and this credential handle may be used to initiate subsequent GSS-API security contexts as an agent or delegate of the initiator. If the original initiator's identity is "A" and the delegate's identity is "B", then, depending on the underlying mechanism, the identity embodied by the delegated credential may be either "A" or "B acting for A". For many mechanisms that support delegation, a simple Boolean does not provide enough control. Examples of additional aspects of delegation control that a mechanism might provide to an application are duration of delegation, network addresses from which delegation is valid, and constraints on the tasks that may be performed by a delegate. Such controls are presently outside the scope of the GSS- API. GSS-API implementations supporting mechanisms offering additional controls should provide extension routines that allow these controls to be exercised (perhaps by modifying the initiator's GSS-API credential prior to its use in establishing a context). However, the simple delegation control provided by GSS-API should always be able to over-ride other mechanism-specific delegation controls; if the application instructs GSS_Init_sec_context() that delegation is not desired, then the implementation must not permit delegation to occur. This is an exception to the general rule that a mechanism may enable services even if they are not requested; delegation may only be provided at the explicit request of the application.1.2.10: Interprocess Context Transfer
GSS-API V2 provides routines (GSS_Export_sec_context() and GSS_Import_sec_context()) which allow a security context to be transferred between processes on a single machine. The most common use for such a feature is a client-server design where the server is implemented as a single process that accepts incoming security contexts, which then launches child processes to deal with the data on these contexts. In such a design, the child processes must have access to the security context data structure created within the
parent by its call to GSS_Accept_sec_context() so that they can use per-message protection services and delete the security context when the communication session ends. Since the security context data structure is expected to contain sequencing information, it is impractical in general to share a context between processes. Thus GSS-API provides a call (GSS_Export_sec_context()) that the process which currently owns the context can call to declare that it has no intention to use the context subsequently, and to create an inter-process token containing information needed by the adopting process to successfully import the context. After successful completion of this call, the original security context is made inaccessible to the calling process by GSS- API, and any context handles referring to this context are no longer valid. The originating process transfers the inter-process token to the adopting process, which passes it to GSS_Import_sec_context(), and a fresh context handle is created such that it is functionally identical to the original context. The inter-process token may contain sensitive data from the original security context (including cryptographic keys). Applications using inter-process tokens to transfer security contexts must take appropriate steps to protect these tokens in transit. Implementations are not required to support the inter-process transfer of security contexts. The ability to transfer a security context is indicated when the context is created, by GSS_Init_sec_context() or GSS_Accept_sec_context() indicating a TRUE trans_state return value.
(page 29 continued on part 2)
