On Mon, Mar 16, 2026 at 11:47 AM Loganaden Velvindron <[email protected]>
wrote:

>
>
> On Mon, 16 Mar 2026, 21:20 Eric Rescorla, <[email protected]> wrote:
>
>>
>>
>> On Mon, Mar 16, 2026 at 10:15 AM Nico Williams <[email protected]>
>> wrote:
>>
>>> On Mon, Mar 16, 2026 at 02:17:31PM +0000, Ben Schwartz wrote:
>>> > I don't understand this.  All secrets derived from ECDH also depend on
>>> > the (hashed) handshake transcript, including the randoms, so the
>>> > resulting shared secrets will never be duplicated between connections.
>>> > What am I missing?
>>>
>>> You right that there is enough entropy with the small nonces (which have
>>> to be kept small), though not in the rest of the handshake.
>>>
>>> Without enforcement of the non-reuse requirement, I'm not finding the
>>> new MUST very credible.
>>>
>>
>> This specification, like many specifications, is full of unenforceable
>> MUSTs.
>> The purpose of this one is to tell implementations not to do something
>> unwise.
>>
>>
>> An optional enforcement requirement would make this requirement more
>>> credible, in the form of a replay cache, might help, but performant
>>> replay caches are quite difficult to build.
>>>
>>> IOW, w/o an optional replay cache enforcement mechanism I think this
>>> SHOULD->MUST is just cosmetic.
>>>
>>
>> As stated in the PR, there might be implementations which reused keys
>> and were previously compliant. Enforcing this MUST would break interop
>> with them, so we can't do that.
>>
> Could there be willingness to fix the issue for those tls implementations ?
>

We don't have an inventory of those implementations, so we'll just
see bustage in the field.

I suppose we could make this a SHOULD NOT, but I think we really
need to make clear that enforcement is a risky behavior.

-Ekr
_______________________________________________
TLS mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to