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

[Paul Kyzivat <[e-mail]>]

On 4/8/16 3:14 PM, Roman Shpount wrote:
> I just wanted to re-iterate the question. We have two options:
>
> 1. If m= line has multiple fingerprints with hash functions X, Y, Z, at
> least one fingerprint with each function must match DTLS association for
> the media to be accepted. For example, if m= line has two fingerprints
> with hash function SHA-1, and two with SHA-512, DTLS association
> certificate must match at least one SHA-1 and one SHA-512 fingerprint.
>
> 2. If m= line has multiple fingerprints with hash functions X, Y, Z, at
> least one fingerprint with any hash function which is considered secure
> enough must match DTLS association for the media to be accepted. For
> example, if m= line has two fingerprints with hash function SHA-1, and
> two with SHA-512, DTLS association certificate must match one of either
> SHA-1 or SHA-512 fingerprints.
>
> Option 1 has a benefit of being more secure since it provides the
> combined secure of individual hash functions. It does put additional
> restrictions how equipment can be combined into producing a single m=
> line. For instance if fingerprints from two servers are combined into a
> single m= line and it is not known in advance which one will be used,
> both servers will need to always support the same hash functions and be
> updated at the same time.
>
> Option 2 is less secure since it is only as secure as the least secure
> hash. On the other hand option 2 allows to be more flexible in how
> equipment is combined to form the single m= line. Two servers in above
> example, for instance, can support different hash functions.
>
> I would ask the group to state their preferences.
>
> I prefer option 2 but can live with option 1.

I don't have a preference. But I will observe that a problem here arises 
because fingerprints for multiple DTLS associations don't identify which 
association they are intended for. Obviously the source of the 
fingerprints has this information. Discarding that info is complicating 
the algorithm.

	Thanks,
	Paul