Re: [TLS] Verify data in the RI extension?

[Yoav Nir <[e-mail]>]

On Nov 26, 2009, at 10:28 AM, Stefan Santesson wrote:

> Eric and all,
> 
> On 09-11-25 11:35 PM, "Eric Rescorla" <[e-mail]> wrote:
> 
>> What about not putting the verify data in the extension?
>> I think that issue is separable from the question of how we signal
>> support, so would like to deal with it separately.
> 
> I would like to elaborate my support for an RI with absent verify data.
> 
> The major argument I have seen for including verify data in the RI extension
> is that it would allow us to preserve the current Finished calculation. This
> will allow us to use more of the current implementation code and provide a
> more clean design.
> 
> I think these to a major extent are valid arguments, but I find the opposing
> arguments even more convincing. These are:
> 
> 
> 1) Empty RI It's good security design
> 
> If you want to make sure two parties agree on a value that they have derived
> from a previous state, in order to bind to the previous state, then why
> would you exchange that value before doing the check?
> 
> On the contrary, I find it to be good security design not to exchange that
> value as it:
> 
> 1) Reduce information leakage to an attacker
> 2) Increase the security assumptions that can be made based on a proof of
> matching values (good Finished messages).
> 
> Even if there are no attack today to exploit the information leakage and
> exchanging the values allows secure design, I still find it valuable to
> apply the best possible security design to reduce the chance of future
> problems. 

Agree

> 2) Better migration path towards future
> 
> If we would design TLS 1.3, then would we send this data in the handshake,
> or would we not?
> 
> I suspect we would not, due to argument 1)
> Exchanging this value serves no purpose other than saving a few lines of
> code changes from TLS 1.2.
> 
> Consequently, if we design an updated Finished calculation now, we can keep
> that in TLS 1.3 and just remove the signaling we now add for SSLv3 -> TLS
> 1.2.

Agree

> 3) Less likely to cause implementation problems
> 
> This is just my guessing based on what I have heard.
> But if the client verify data is provided in RI, then it is not enough to
> recognize the extension. It is now also necessary to parse it, extract the
> data and to use it in a critical security check.
> 
> I can't really judge how likely this is to cause more implementation
> problems or even implementations that appears to do the security check
> against previous handshakes, but actually just feeds it into the finished
> calculation blindly.
> 
> But it feels like good design to eliminate/reduce this type of problems if
> possible. Especially if it is better security design at reasonable cost.

As far as implementation complexity, it's a wash. You have to have some long-term "state", which now must include the old verify_data. There's no way around that.

It's not that much different to add this to the calculation of the Finished message, either in concatenation to the handshake messages, or instead of the finished label, as it is to compare it to the contents of the RI extension.

There is the risk that an implementer who hasn't been following this list might fail to verify the data in the extension. Googling 'security bug "fails to verify"' returns 34,000 hits, suggesting that this is a valid concern.