Re: [rtcweb] JSEP fingerprint hash requirements

[Eric Rescorla <[e-mail]>]

On Thu, Oct 17, 2013 at 11:54 AM, Martin Thomson
<[e-mail]>wrote:

> On 17 October 2013 01:37, Kevin Dempsey <[e-mail]> wrote:
> > 1) does the fingerprinh hash need to match the certificate
>
> Yes.  Without that, you've got no binding between signaling and media
> path, which is bad.
>
> > 2) do webrtc compatible endpoints need to handle hashes 'weaker' than
> > sha-256
>
> No.  RFC 4572 is clear:
>    A certificate fingerprint MUST be computed using the same one-way
>    hash function as is used in the certificate's signature algorithm.
>
> That means that you need to generate the certificate with a hash that
> is strong enough.


I'm not sure this was a sensible rule on 4572, but I don't think it's
particularly harmful. Here's my analysis: there are two kinds of
certs in play here:

- self-signed certs (most likely)
- CA-signed certs (less likely)

In the case of self-signed certs, you can control your digest. In the case
of
CA signed certs, someone is presumably going to verify the signature
and so we would expect the digest used to form that signature to be
reasonably strong.


 > 3) are there any rules for handling multiple fingerprints?
>
> RFC 4572 is silent on that, unless I missed something, which I
> probably did.  The only plausible choice given the above statement
> from 4572 is to suggest that multiple a=fingerprint values indicate
> alternative certificates.
>
> That should probably be written down, of course.
>

Agreed.

-Ekr