Dave CROCKER wrote:
> 
> 
> Stephen Farrell wrote:
>> The putative attack would be to sign with foo-sha256 where foo is some
>> algorithm that a verifier supports and such that the foo key encoding
>> could
>> accept an rsa p= value as input and such that that combination allows the
>> attacker to forge signatures.
>>
>> That's fairly theoretical, but there'd be a bit of work to go check that
>> there is in fact no realistic situation where such confusion could
>> arise. So
>> on the whole, for me, keeping the k= just makes it easier to get
>> things right
>> now and in future.
> 
> 
> If out-of-band algorithm/key-type registration like this is not a
> regular "protection" mechanism provided in existing, related
> crypto-based schemes:

I think it is common mechanism. Normally its in a certificate for
PKI based applications or as part of negotiation in e.g. TLS etc.
XMLDSIG also includes the same information in its data structures.

>      1) Why should it be only in DKIM?

So I don't think that this is the case.

> 
>      2) Won't it need expert vetting by the security community to
> validate that the threat is real and the protection is sufficient?

Well, 4871 got last called, secdir review etc so it got as much
security review as most things. I don't think this is the kind of
thing where we'd need to get cfrg input.

> I have been repeatedly taught by the security community to be extremely
> cautious about casual assumptions of what will fix a theoretical
> exposure.  So, for example, there have been cases where encrypting twice
> using the same algorithm reduces protection rather than raising it. 
> That ain't intuitive.
> 
> Same concern here.

Well I guess both concerns are theoretical really. Personally,
I think having these fields in the key record reduces the need
for security analysis which is a good enough reason for me.
Of course, others might disagree, but I've not heard any
security concern about including them.

> Having this information maintained in the DNS isn't free and isn't
> certain to be safe.

Fair enough. Though I don't understand the additional expense
of k= and h= in key records - I'd have thought it might even
help maintaining those records, but since I don't maintain any
DNS stuff I may be wrong.

I do agree that key records being unsigned in the DNS is a risk,
but its one we accepted at the start of the WG.

S.


> 
> d/
_______________________________________________
NOTE WELL: This list operates according to 
http://mipassoc.org/dkim/ietf-list-rules.html

Reply via email to