Fixed syntax error
Notes about not working with 4.x
|Deletions are marked like this.||Additions are marked like this.|
|Line 28:||Line 28:|
|* As it turns out, the entire authentication handling is server-first oriented. There is no way to integrate client-first unless the internal code is rewritten. See [[http://mail-archives.apache.org/mod_mbox/hc-dev/201504.mbox/%3C552C2B3B.9060003%40apache.org%3E|discussion]].|
This page tracks the design goals of the complete redesign of the GSS-based authentication in HttpClient. Namely, implementation decisions, known issues, questions, testing, etc. All code will be developed in a separate branch.
Implemenation decisions are comprised of several blocks like interface implementations, exception handling, logging, etc.
AuthSchemeProvider: merely a factory for creating AuthScheme instances. Implemenation will be GSSBasedSchemeProvider. It will take in one argument, the OID string of the desired authentication mechanism or simply the AuthScheme name.
AuthSchemeBase (implements ContextAwareAuthScheme): the implementation GSSBasedScheme will take in one argument, the OID string of the desired authentication mechanism or simply the AuthScheme. It will internally maintain a stateful GSSContext for the authentication against a target or a proxy. Since the implementation itself does not know when it will be nulled and garbage collected, it will maintain its state internally and release the GSSContext immediately upon successful completion or the first failure. This implemenation will not be threadsafe.
Credentials: this will be GSSBasedCredentials and will take in a GSSCredential. Useful if not the default GSSCredential will be used. It is also necessary to create a GSSPrincipal class which will wrap the GSSName from the credential.
Response token is not handed over to GSSBasedScheme, thus authentication can never be completed. It is highly likely that the HttpAuthenticator needs to be changed. There must be a notion of isClientFirst just as in SASL (RFC 4422, section 5, 2a).
A CredentialsProvider with an fake item must be set otherwise authentication is not triggered.
As it turns out, the entire authentication handling is server-first oriented. There is no way to integrate client-first unless the internal code is rewritten. See discussion.
Testing is comprised of two sections: unit tests and integration tests.
It has to be determined how one can reasonably mock GSS objects to test the new implementation.
Integeration tests will be performed in a corporate environment with the following setup:
- Client OS: Windows 7, RHEL 6, FreeBSD 9.x, HP-UX 11.31
- Java runtime: 1.6 and 1.7 from Oracle, OpenJDK and HP
Target servers: Microsoft Forefront TMG (HTTP proxy) (SSPI), Microsoft IIS 7.5/8.0 (SSPI), Apache Web Server 2.2.x with mod_spnego (MIT Kerberos) and Apache Tomcat 6 with Tomcat Authnz SPNEGO AD (JGSS).
Not all combinations can be tested.
Concrete requests are still open.
Why does MalformedChallengeException not extend AuthenticationException though it is documented for authentication purposes?
OK: MalformedChallengeException signals syntax violation of some sort presenting the client from understanding the challenge whereas AuthenticationException signals inability or unwillingness to respond to the challenge. To me these are different type of issues, but I am open to changing it in 5.0.
The name of ChallengeState is quite confusing. Where is the state? This is merely a ChallengeHostType.
OK: We can deprecate it and replace with AuthCounterpartType or some such in 4.5.
Can a ContextAwareAuthScheme instance be reused?
OK: It can be re-used and will automatically be re-used within the same context (requests executed with the same instance of HttpContext).
Can an HttpContext be used concurrently?
MO: in theory, yes. See Javadoc of HttpContext.
Update documentation of DefaultUserTokenHandler