Ephemeral key reuse is a performance feature of TLS stacks, at least it
used to be. See e.g.
https://mailarchive.ietf.org/arch/browse/cfrg/?gbt=1&index=oMS8YfLOWSBlHMFm7pBeJobafrA
, where DJB described one product and supported reuse. I know of another
big vendor who did the same.

I think that the new language is OK, but I am surprised that there was
enough support for such a tighter language. In any case, punishing servers
for reuse would be unwise IMO.

On Tue, Mar 17, 2026 at 8:41 PM Viktor Dukhovni <[email protected]>
wrote:

> On Tue, Mar 17, 2026 at 08:56:32AM -0700, Eric Rescorla wrote:
>
> > On Tue, Mar 17, 2026 at 8:01 AM Viktor Dukhovni <[email protected]>
> > wrote:
> >
> > > On Tue, Mar 17, 2026 at 01:09:22PM +0000, Ben Schwartz wrote:
> > >
> > > > The support for this change is clearly overwhelming, but I'm still
> > > > confused about the motivation.
> > >
> > > FWIW, I just don't have the energy to object to every well-meaning, but
> > > counterproductive proposal.  And it can be uncomfortable to uphold a
> > > minority view...
> > >
> > > I agree that reuse of keyshares across multiple connections should
> > > generally be avoided, which is the status-quo in RFC8446, but there are
> > > sometimes just exceptions.  An unenforceable MUST NOT may feel like
> > > progress, but it may do more harm than good.
> > >
> >
> > "May" is doing a lot of work here.
> >
> > Do you have some actual substantive argument to offer?
>
> The specification already correctly discourages reuse, changing this to
> a MUST, especially when enforcment on the receiving end is neither very
> practical, nor wise, rather like a feel-good exercise.  And though I
> don't have a compelling example of reuse in mind, if a particular client
> stack or deployment finds sound reasons to reuse a keyshare across
> multiple connections (perhaps closely spaced in time, or happy eyeballs,
> or some other just reason) I don't think it is appropriate to put
> roadblocks in their way.
>
> The problem this proposed "MUST" is solving is essentially non-existent,
> mainstream TLS software stacks and deployments already don't use
> long-term keys for ephemeral key exchange.  I do however recall (I hope
> accurately, apologies if this is fiction) that Microsoft's TLS may under
> some conditions reuse ephemeral keys briefly for connections that are
> closely spaced in time.  There's nothing fundamentally wrong with that,
> assuming they've done their homework.
>
> --
>     Viktor.  🇺🇦 Слава Україні!
>
> _______________________________________________
> TLS mailing list -- [email protected]
> To unsubscribe send an email to [email protected]
>
_______________________________________________
TLS mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to