On 18/11/2026 2:33 am, Viktor Dukhovni wrote:
Taking a close look at the current DKIM2 I see some opportunities for
simplification.
Indeed.
I am reluctantly willing to take it as given that replay of already
signed messages is a practically significant issue for at least some of
the larger providers, and that this warrants signing their mail with
DKIM2, allowing receiving systems to detect unauthorised replays.
That said, much of the complexity of the design is a result of the
design also attempting to authenticate bounces of messages that
might have been modified in transit. This major redesign of
the reverse path is, at least for me, a bridge too far. Is this
absolutely necessary? The cure looks worse than the disease.
Agreed.
This is entirely unnecessary to prevent replay and backscatter. Once you
can cryptographically verify the intent for a particular e-mail to be
sent from and to any given recipient (including a bounce), the route it
takes to get there is irrelevant.
So I'd like to see DKIM2 slimmed down, to just make the minimal
requisite forward-path changes, and leave the reverse path alone.
I'd like to see this removed too, as this changes signing flow
significantly and adds no benefit for the majority of implementers.
It's not technically feasible to record all changes within the headers.
The authors have acknowledged this isn't practical in some cases and
even described attempts to explain large changes (which may result from
minor edits) as "dumb". Creating a standard to do x, then creating a
guide/clause telling you not to use it, shows there are inherent flaws
with the approach that need addressing.
With that in mind, I'd like to suggest that email systems that have no
need for the additional anti-replay protection be able to opt-out of
that feature set entirely, and should be able to sign just the
message-content, sans envelope, when that meets their needs. This
substantially reduces complexity, and supports continued delivery of
multi-recipient mail without disclosing Bcc recipients in the message
payload.
Anti-replay protection is the single most important feature of DKIM2, so
I don't agree with being able to opt out of it. I do agree, however,
that we need to address the regressions for signers that are not capable
of splitting e-mail to Bcc recipients. This includes on-premise systems
that relay through another host and client-side signing.
I think we can have a more flexible design that still supports the
ostensible real needs of the large providers, without unnecessarily
burdening the entire ecosystem with the same concerns..
Agreed. The real need, which benefits everyone, is replay protection
andĀ backscatter prevention.
If you remove the over-engineered chain of custody and acknowledge that
change tracking isn't practical, you're left with something not entirely
unlike DKIM. As I have said many times, you can add replay protection
and backscatter prevention to DKIM without introducing breaking changes,
yet this isĀ something the Working Group has avoided exploring in favour
of an ARC replacement outlined in a flawed motivations document.
Regards,
R. Latimer
_______________________________________________
Ietf-dkim mailing list -- [email protected]
To unsubscribe send an email to [email protected]