Re: [MMUSIC] 4572 update: forbid weak hashes?

[Roman Shpount <[e-mail]>]

On Mon, Apr 25, 2016 at 3:56 PM, Paul Kyzivat <[e-mail]> wrote:

> On 4/25/16 3:26 PM, Christer Holmberg wrote:
>
>> So, does anyone object to the proposal: no matter how many certificates
>> (and associated fingerprints) the SDP contains, as long as there is
>> **one** hash (that the receiver supports and consider secure enough)
>> matches the certificate, things are good.
>>
>
> The possibility that the fingerprints might cover multiple certs, for
> multiple connections, messes things up.
>
> IIUC, *if* there is only one connection to be protected then all the
> fingerprints must be fingerprints of the cert for that one. And in that
> case if you check one and it fails to match then there is indeed an error,
> even if another one matches.
>
> Should it be acceptable to consider this case a failure? Or is this too
> complicated?
>

There is still a use case when an offer is sent but it is not yet known
which of the several media processors will be used when connection is
actually established. Such offer will contain fingerprints for certificates
on all the media processors, but only one of them will match. This is true
even for a single connection (bundle).

>
> One possibility would be to state that the minimum check, that MUST be
> applied, is that at least one fingerprint MUST match every cert covered by
> the SDP, and that an optional check (MAY be applied) is that *each*
> fingerprint match one of the certs covered by the SDP. (The optional check
> is something that a formal tester/verifier might perform.) It would be a
> requirement that the sender ensure that the harder test succeed if the
> recipient applies it.
>

We might add another SDP parameter for this or extend fingerprint in a
backwards compatible way to indicate some sort of certificate group, but
this is making things more complicated. Multiple people indicated that this
is out of scope for this document revision.


> I can live with just the minimum being stated.
>

We are mostly trying to document what people thought was already supported
in maximum backwards compatible manner. I think minimum match is the only
way.
_____________
Roman Shpount