Moritz Bechler

It took me literally tens of hours to figure out how to do SPNEGO proxy authentication for JAVAs builtin HTTP routines. So let me share my results:

1. Documentation sucks as hell


2. Documentation sucks as hell, and there is plenty of it


3. JAVAs implementation of GSSAPI and Kerberos:
   a) fail to establish a security context with my MIT krb services (I was neither able to authenticate agains mod_auth_kerb and my squid negotiate helper),
   b) are a nightmare to configure and
   c) seem to be unable (without really ugly hacks) to obtain the credentials from MIT's default credential cache (are not using KRB5CCNAME env)


4. Fortunatly JAVA SE6 has a so called native Mode for JGSS - which you need to find first - see http://java.sun.com/javase/6/docs/technotes/guides/security/jgss/jgss-features.html - this setting makes JAVA use your GSSAPI Library instead of their fucked upenterpricy library

To make your Java Applications integrate with your well-established Kerberos SSO just set

export _JAVA_OPTIONS="-Dsun.security.jgss.native=true -Dsun.security.jgss.lib=/usr/lib/libgssapi_krb5.so -Dhttp.auth.preference=spnego -Dhttp.proxyHost= -Dhttp.proxyPort="

(Location of your gssapi library might differ)

During the last days I had some time to work on different things that were in the working queue for quite some time:

First of all, I had the chance to look at the possibilities for OpenAFS web-administration again. As I might have written before, the libadmin library is such a mess in terms of documentation and I really did not want to dig into this - so I wanted to try out what was possible using JAFS. Until now there was no support for use in non-KAS cells but then - some weeks ago - there was a post on openafs-dev about using it in a K5 cell (and also fixing the code for compiling with java 1.5) so this was the chance to try it out. It really took some time to get it kind of working - but finally (after patching out the various kas_ calls that were still in the code - which made the code block almost infinitly, and some other patches to make it compile) it did. After that I was almost unable to belive that the php-java-bridge was working perfectly out of the box. So I could take phpSATk build some objects and definitions around the interface (this unfortunatly was a bit more complicated than the other things I build so farr because I had to do the right type casts in the wrapper) and hell yeah ... now it seems to work. And I updated my php_afs extension to only take a kerberos ticket (from php_krb5) and perform "aklog"-ging to the afs cell to have the right token in kernel.

This might be a pretty much glued together solution (and there are some problems like error handling which are not that nice), but yes - IT WORKS!!

The other thing I had been working on is the ICAP server (which still needs some unique name ) which made some great progress. It now supports previews and persistent connections, has the base for a parser to extract text tokens out of the source for content filtering and the score management is to be replaced by a real solutions soon. Squid3 is (marked) stable now - I do have it installed in a production environment already and the only thing on the bug list right now is a annonying ICAP bug to be reported) - so it might be a good idea to get it working soon. (Help is always appreciated)

I have not had time to work on phpSATk for quite a while (there will be some progress soon ... promised), the squid negotiate helper is kind of ready (I have been using it in a test environment for some time now), there will soon be a publicly available version (the impatient might have a look at the svn version) and my PHP CRL patch will propably included in PHP HEAD as soon as I manage to put together some test cases and possibly 5.3 afterwards.

But the project I spent the most time on in the last weeks was building a ICAP based web filtering engine. In the past I felt like all existing (open source) web filters have major shortcomings:

- completely relying on URL black-/whitelists totally sucks - the number of false positives and false negatives is extremely high, higher quality blacklists are expensive, you never know what the political/commercial/whatever interests of people/institutions putting sites on these blacklists are.
- the only free content filtering solution I know about is dansguardian which relies on proxy chaining what sucks when it comes to authentication. This approach also is not as flexible as I'd like it to. The licensing terms are imho a bit too restrictive.
- All solutions I know are not really configurable at run time. In production use I need to possibility to make online changes to the black/whitelists and or wordlists without causing connections to disrupt and/or increased latency (I'd consider writing to configuration files and/or black-/whitelists from a webapp inacceptable).

so I'm trying to build a solution having the following properties:
- uses the ICAP standard (squid3 is coming...)
- will be probably licensed under GPL
- will be a hybrid solution combining the results of content analysis and url filtering
- content analysis should include reliable word/phrase matching as well as parsing of PICS tags.
- will be based on scores (one for each category) which will be used to match a profile of allowed sites
- contains efficient "database engines" for the different datatypes used - each of them manageable in real-time through an RPC (XMLRPC for now) interface.
- should be extremely fast (it already uses threads and asynchronous i/o) and scaleable
- ... will integrate with phpSATk as administrative interface


So far I have a working prototype which can already deny/allow access based on: server ip (this is really useful for some popular sites which have hundreds of aliases), host/domainname and regular expressions. Both parsing PICS tags as well as reliable and fast phrase/word matching are very hard to implement - so maybe this will need some time until I can show something working.

I'll announce this thingy (and the negotiate/GSS helper) to the squid users/dev soon so maybe somebody volunteers to contribute.

Update: squid-3.0-stabe1 is released ... its time to have this thingy working ... the icap protocol implementation is kind of feature complete now (previews, persistent connection are implemented)

Kerberos has been the de-facto industry standard for Single-Sign-On for many years but not yet been widely adapted for intranet/web-applications. Firefox supports GSSAPI (on Linux/Unix and Windows - using MIT Kerberos for Windows (KfW)) and SSPI (Windows) for Kerberos authentication quite a while, usable since version 1.5.

Because Microsoft also uses NTLM for SSO purposes they invented a GSSAPI pseudo-mechanism named SPNEGO to do negotiating of which of the protocols to use - fortunately MIT kerberbos (since version 1.5) supports SPNEGO (thanks to SUN donating a implementation) so we can use MITs GSSAPI library for the server side. SPNEGO is yet only supported on Linux platforms (KfW 3.1 includes a krb library of version 1.4.5 - 3.2 which is currently in beta status will have 1.6+) but at least Mozilla will automatically fall back to plain kerberos authentication which in case of a MIT server side will perfectly work.

Update: KfW 3.2 has been released today - SPNEGO now works on Windows.

I have not yet had a chance to experiment in a SSPI environment and there are rumors about Microsoft not implementing their own protocol (SPNEGO) correctly, so I cannot really say whether it is compatible.

There is a (not yet standardized) extension to the HTTP/1.1 specification which specifies the "HTTP negotiate" authentication method. We try to use it for SSO on web-applications and web-resources.

I wrote a PHP extension which gives server side support for this kind of authentication. Continue reading for directions on how to use it...

I finally managed to get the krb5 extension for PHP ready for some kind of release, since the last time I wrote about it I felt the urge to implement yet another feature: Negotiate authentication using GSSAPI. This enables real kerberos single sign on support for webapplications in PHP and works in Mozilla (tested on Windows and Linux - using MIT kerberos/KfW) and should work on IE, too, but I have not yet had the chance to test using SSPI. Nice to have

I'll post some article on this feature in the next days.

I now have copied the missing headers from the the mit-krb5 distribution, so the source is not needed anymore.


You can get it here -
php_krb5-beta.tar.gz - and maybe some day through PECL.

Here comes - for my and possibly your convenience - a combined patch against ext/openssl for PKCS#12 and CRL support. It also fixes my patch (missing the header file) and Marc Dellings part (missing some TSRM macros).

ext-openssl.patch

Update: Marc Delling's patch has made it into the distribution, updated patch against current -dev.