Hi All,

My name's Roger Anderson, work at the University of California, San
Diego as a software engineer, manager of a systems department in our
facilities area. Using OpenSSL to support SSL/TLS protocols on a
facilities management website for campus. Server is cross-platform
C/C++ (Win32/Unix), in-process request handling, CAD and raster
graphics work, and database management with PostgreSQL. Server links
in OpenSSL (recent versions up to 0.9.6c) on Win32, BSD, Linux, etc.

Spending some time this weekend grappling with a bug with Netscape 
6.2 (recent Mozilla builds) and SSL_accept()/SSL_read() internals.
Connects from these clients cause the server-side to wait forever 
in blocking mode and/or timeout in non-blocking mode. To be more
precise, behavior I'm seeing is success with the first accept or
handshake, and problems with the second. Recent posts from Tim,
Daniel and Bodo on openssl-dev "netscape 6.2 crash" provide good
descriptions of the problem. Bodo's comments about blocking on the
first connection seem right on target to me. Also dug up several
related bug reports over at <mozilla.org>, using component "Security
General" for query criteria. Some seem to be source issues with the 
SSL connection, others related to build and config management.

We've got a fair number of users around campus connecting now with
new NS and Mozilla builds on Win32, Unix even VMS on an Alpha if I
remember correctly. I'd very much like to find a workaround on the
server-side to handle these clients. I've been tinkering with calls
in the SSL API: BIO_sock_should_retry(), SSL_renegotiate(),
SSL_do_handshake() and so on. No success though. Need to download
the client-side source, build/debug to see what's going on exactly.
It's been wonderful and fun learning how OpenSSL works, but it's
Saturday afternoon and I'm running out of time.

You guys are way more knowledgable and experienced with these source
packages, and even have some experience with this specific bug. I
wanted to poll whether you think there might be something we can do
on the server-side to handle the problem? Am I wasting time with the
server-side SSL tweaks? Advice on where to look, even whether you 
think a workaround might actually work, is greatly appreciated.

Fallback position is to simply timeout the problem connections and
move on to clients that are able to negotiate the SSL, with fixes
arriving at a later date in updated versions of NS and Mozilla. Just
hoping there's something we can do in the interim to keep these
users connecting...

I'll keep hacking at it for a while longer. Downloading the client
build next. Couple hours left this afternoon. Might be able to
figure something out. Work tomorrow and lots todo in other areas
though. Any pointers (of a non-null variety!) greatly appreciated.

Thanks!

- Roger

Roger Anderson, Director
Campus Planning Data and Systems
University of California, San Diego
______________________________________________________________________
OpenSSL Project                                 http://www.openssl.org
Development Mailing List                       [EMAIL PROTECTED]
Automated List Manager                           [EMAIL PROTECTED]

Reply via email to