Re: [TLS] SCSV vs RI when both specified. Was: Updated draft

[Marsh Ray <[e-mail]>]

Yoav Nir wrote:
> On Dec 21, 2009, at 6:28 PM, Marsh Ray wrote:
> 
>> Blumenthal, Uri - 0662 - MITLL wrote:
>> 
>>> If the protocol spec demands aborting connection, it better have
>>> a damn good reason to do so - and more substantive than "some
>>> Steve decided it doesn't really matter to him if the peers
>>> connect or not".
>> How about "remote endpoint doesn't pass the bozo test"?
> 
> We do not discriminate against bozos.

If just one or two of the major implementations did from the beginning,
then they wouldn't exist in the ecosystem.

> Seriously, servers are there to communicate. Amazon or Google are not
> going to turn away customers because their browsers are a little off.
> That's why they agree to work in SSLv2.

That's their decision, but you cannot let it drive the evolution of the
protocol. Some users take security very very seriously.

It's common to have a trade-off between security and compatibility.

If the protocol spec mandates behavior on the side of security, it
remains secure for everyone who needs security. If A&G preferred
compatibility they can opt to configure their systems that way.

But if the spec sacrifices security for the sake of compatibility, then
no one can put the missing security back in later.

- Marsh