Re: [TLS] Sending fatal alerts over TCP

[Florian Weimer <[e-mail]>]

* Peter Gutmann:

> Florian Weimer <[e-mail]> writes:
>
>>What is the best way to deal with this? 
>
> If it's the other side that does it then there's not much you can do, if
> you're doing it then you need to wander into a twisty little maze of
> shutdown()/close() passages, all different, buggy, and incompatible:

Yeah, that's about as much fine print as I feared.  Thanks.

Fortunately, I'm only trying to resolve an interoperability problem
between GNUTLS and the OpenJDK implementation.  There seems to be some
confusion about the error reporting for connection failures, and I'm
trying to provide better messages to consumers.  It turns out that both
GNUTLS and OpenJDK properly report the certificate_unknown alert during
handshake failure when talking to themselves implementation.  The
difference between GNUTLS and OpenJDK when talking to an OpenJDK server
is this: the GNUTLS client sends the Certificate and Client Key Exchange
messages in separate TLS records and (in the way we use it) TCP
segments, while the OpenJDK client uses a single TLS record in a single
TCP segment.  As a result, the race leading to reception of an early RST
is rarely lost by the OpenJDK client (if ever).

I changed the GNUTLS client to create just one TCP segment per server
round-trip, and I can now read the TLS alert before the RST segment
arrives.  So the fix was quite trivial after all.

-- 
Florian Weimer                <[e-mail]>
BFK edv-consulting GmbH       http://www.bfk.de/
Kriegsstraße 100              tel: [phone]
D-76133 Karlsruhe             fax: [phone]