You are both right, about different attacks, and separating the attacks
dissolves the disagreement about the first link.

Brian is right that truncation has to contradict a signed statement rather
than fall to a policy judgment. A defense that depends on the verifier
noticing something missing is not a defense.

Yaron is right that committing chain length and the ordered hop set in the
first link breaks the deployments brokering exists for. In his topology,
subsidiaries insert and remove their own brokers, and the first hop cannot
know the eventual shape of the chain. A commitment scheme that requires
topology knowledge at origination welds the chain to one arrangement of
intermediaries at signing time.

Now split truncation into its three cases. Tail truncation: the
line-of-sight design already defeats it, because each link names the
audience of the next hop, so a chain cut short terminates in a link whose
stated audience is not the verifying AS, and reconstruction fails closed.
Middle removal: also defeated, because link k names its delegate, and
removing hop k breaks the signer continuity between link k-1 and link k+1.
Head truncation, meaning a middle broker re-originating a chain to hide
what sits upstream of it: no commitment scheme defeats this, in either
design, because the adversary is the party signing the first link. That
case is a policy question the terminal AS answers from what it knows about
its own registered clients, which is exactly Brian's point about
operator-set broker expectations, and the reason a self-asserted
client_roles value cannot carry it.

So audience chaining handles everything signatures can handle, operator-set
expectations at the AS handle the one thing they cannot, and downgrade
stays a per-hop subset check either way. The first-link topology commitment
buys nothing those two do not already provide, and it costs the dynamic
chains that are the reason this draft exists instead of the
register-everything countermeasure. Worth writing into the draft which
check defeats which attack, so implementers do not reintroduce the
commitment out of caution.

Kieran Sweeney
_______________________________________________
OAuth mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to