Difference between revisions of "Projects/HTTP Transport"

Revision as of 12:22, 23 August 2013

Overview

This project intends to add HTTP and HTTPS transport to Kerberos traffic. This change is useful especially for firewall configurations that allow traffic on port 80/443 but not on port 88.

Precedent

Both Heimdal and Microsoft Kerberos have such a technology.

Heimdal

Heimdal has such a mechanism as seen here. It uses a GET request with a base64-encoded version of the UDP traffic. This is not particularly in keeping with the HTTP specification, since a GET request should not incur change to the server. Additionally, Apache has a URL length for GET of about 4000 characters, and requests of this nature have been measured as close to that limit and may exceed it in practice. It uses a separate field in krb5.conf for specification of the http_proxy to be used. There is almost no evidence of this in use in active deployment.

Microsoft

Microsoft has documented their mechanism, MS-KKDCPP, here. It uses POST requests which is much more in keeping with the HTTP specification than GET, and it also specifies HTTPS to be used in all cases, though it is not known whether it will respond to plain HTTP.

Implementation Process

HTTP transport will be implemented first, followed by HTTPS transport. We prefer Microsoft interoperability to Heimdal interoperability because despite the added complexity we feel it to be more in keeping with the HTTP specification. We will implement HTTP and HTTPS transports that correspond to our interpretation of Microsoft's specification. Parallel to implementation, we will attempt to set up a Microsoft KDC with proxying enabled so that we can test our implementation. For HTTPS transportation, we will first implement OpenSSL support, with NSS support to follow.

Implementation Design

• We will expand the definition of the `kdc` field (and related fields including kpasswd_server and admin_server) in krb5.conf to take a URL (optionally including page) in order to allow HTTP or HTTPS transport. The new syntax will look like this:

 kdc = (http|https)://<kdc.addr>[:<port>][/<page>]

Note that the syntax (and parser) does not change in the non-HTTP{,S} cases. In order to facilitate this change, we will need to stop carrying a socktype around in the code that is either SOCK_DGRAM or SOCK_STREAM, and instead carry around our own protocol designator.

• We will need to build against a cryptography library, and to add options to the build system for such. We will include an option to disable HTTPS support (i.e., build against no cryptographic library).

As we have not yet been able to examine traffic from a Microsoft KDC with proxy enabled, our current working protocol for traffic is: POST request to /KdcProxy with body consisting of the base64-encoded version of the Kerberos request, with response from the MS-KKDCP consisting of the base64-encoding of the KDC's reply. Preliminary testing suggests that this is not what Microsoft uses, but the server isn't particularly chatty about errors.

Test Plan

Due to the nature of the changes, it will be extremely difficult to write test cases for the new code. Instead, code has been written to implement the "other side" (on the KDC-end) of HTTP/HTTPS transport as a reference implementation. That code, including its implementation of the client-end of the HTTP/HTTPS pipe, can be found here.

Review

This section documents the review of the project according to Project policy. It is divided into multiple sections. First, approvals should be listed. To list an approval type

#~~~~

(hash mark followed by four tilde characters) on its own line. The next section is for summarizing discussion, which should take place on krbdev@mit.edu. Provide links to the archive at http://mailman.mit.edu/pipermail/krbdev/ if appropriate. Blocking objections can be noted with {{project-block}}.

Approvals

Discussion

From mail from ghudson:

blocking Should remove the tcp: and udp: specifiers from the `kdc` configuration line and write a separate case for http: and https: instead of trying to integrate them

blocking Switch to non-portable socket enumerator to portable protocol enumerator

blocking Remove cut-and-paste between service_tcp_fd and service_http_fd in favor of more careful design