Re: Implementation support for SSH_MSG_UNIMPLEMENTED

["Joel N. Weber II" <ietf-secsh%joelweber.com@localhost>]

> >Are there any implementations which do not respond with SSH_MSG_UNIMPLEMENTED
> >to unknown packet types during the key exchange phase of the protocol?
>
> I don't, but that can be fixed if it becomes a critical requirement of the
> protocol.  The reason I don't is that I always send the minimal amount of info
> in error returns for any protocol I do (SSH/SSL/CMP/TSP/RTCS/OCSP/etc), which
> has saved me from at least two attacks on SSL and probably attacks on other
> protocols as well.  In other words if the protocol requires a certain response
> in order to function I'll do it, but if it's merely a nicety for debugging,
> I'll send the most generic response I can get away with.

The text of the protocol spec says that sending SSH_MSG_UNIMPLEMENTED
is mandatory, if I recall correctly.  Not implementing this according
to the spec makes interoperability harder when new features get added
later.

SSH_MSG_UNIMPLEMENTED is already fairly generic.