Re: [hybi] Extensibility mechanisms?

[Adam Barth <[e-mail]>]

On Mon, Jul 26, 2010 at 1:58 AM, Greg Wilkins <[e-mail]> wrote:
> your analysis of the cross protocol security issues continue to be very
> informative.

Thanks.  Glad you're finding it useful.

> But why does it have to be in the tone of "the WHATWG guys are the only ones
> that understand anything and there is no need to change anything"?   Perhaps
> you are reacting against the harsh tone that is sometimes directed towards
> the WHATWG guys in this WG and feeling like you need to defend them.  Well
> I'm sorry about that tone, but there has been significant stonewalling from
> that quarter and there is a lot of frustration here that many other concerns
> raised are being rejected.

I'm not sure why you consider me "one of the WHATWG guys."  I
participate about equally in WHATWG, W3C, and IETF.  At the moment I'm
more involved with IETF than the other two.

> So thanks for expanding on some of the details of the cross protocol
> attacks, and you have rightly drawn attention to deficiencies in some of the
> simplistic counter proposals.
> However, the bit that is missing is how your analysis relates to the current
> proposed solution with it's rigid content, space counting, random bytes
> without frames etc.   Surely that cannot be the only solution that addresses
> these issues?   Can't we try to find a solution that both reduces the cross
> protocol issues, but is also less objectional to most here in the WG?

I'm sure there are other solutions.  For example, I proposed a
handshake a while back that isn't rigid at all.  If you feel the
current proposal doesn't meet your requirements, please feel free to
propose something else.

> To harden a websocket server against attacks from other protocols, there are
> many things that framing can help with.  For example, we could shutdown
> connections after any bad frames (we are on reliable transports so no need
> to forgive bad frames) and/or have a simple frame checksum.
>
> To ensure that websocket clients are themselves not tools to attack other
> protocols we could also have less minimalistic framing. Lengths and
> checksums are going to be more of an obstacle to attackers than simple
> sentinel framing.
>
> I have yet to see a good description of an attack vector that is cut off by
> the rigid handshake, the space counting or the unframed bytes that could not
> be addressed by far simpler and less objectional means.
>
> regards
>
>
>
>
>