Geoff & Rob,
First, my recollection is that we decided that the cert for a
manifest appears only in the manifest data structure, which still
seems like a useful simplification.
So, I suggest that this EE cert appear only in this location, and
this not have its own publication point. That does require that the
cert be used ONLY for manifest validation. In the terminology Geoff
crafted for the manifest I-D, this makes the EE cert for a manifest a
single-use cert, (optionally a sequential use cert) even if it is not
a one-time-use cert.
Geoff's message emphasized reuse of the private key, but Rob's
message emphasized the desire to not have to put the manifest EE cert
on a CRL if a manifest is reissued one or more times before its next
scheduled issuance time.
There was no mention of the validity interval is for this cert in
either Geoff's or Rob's messages.
For a one-time-use cert, the manifest I-D suggests that the EE cert
lifetime match that of the manifest. That will cause each new EE to
be on the CRL until the matching manifest expires, when a manifest is
issued before the nextUpdate time. Presumably this is what triggered
Rob's concern. The manifest I-D does not provide (useful) guidance
for the validity period for a sequential-use EE cert associated with
a manifest. Presumably the validity period will be longer than that
anticipated for the manifest, else it makes no sense to view this as
a sequential-use cert. So, it appears to me that a sequential-use,
single-use cert satisfies Rob's concern about CRL growth.
The primary security rationale for revoking the EE cert for a
manifest when a new manifest is issued is to alert RPs of the
manifest change. The argument is that the revoked EE cert for the
manifest will appear on the CRL, and RPs will thus know not to accept
the old manifest. This argument presumes that RPs will acquire the
new CRL in orders to perform this check. However, if an attacker
replaces the new manifest and new CRL with the older, not yet expired
versions, RPs will not be able to tell the difference, so the cited
rationale is not a good one.
This suggests that a sequential-use EE cert for the manifest is just
as secure as a one-time-use cert, relative to the unscheduled
manifest change case.
Given this observation, I suggest that we mandate that the manifest
nextUpdate value and the nextUpdate value in the CRL that covers the
EE cert used to verify the manifest be the same. Otherwise, if the
CRL is scheduled to be reissued before the manifest, the manifest
will have to be reissued at that time, so that it contains the
current CRL hash value, and we'll create a preventable stale manifest
problem. (I also suggest that this analysis be included in the
manifest I-D.) If CRL and manifest issuance are synchronized, as
suggested above, then a sequential-use EE cert for a manifest can
safely have a lifetime that spans multiple manifests. The primary
argument for one-time-use EE certs for manifests, is the same as for
ROAs, i.e., the underlying private key can be forgotten after use,
making the system somewhat safer as a result.
As for terminology, a CA is an entity, a CA cert is not. We can talk
about the current CRL issued by a CA and verified using the CA's
cert, but a manifest is issued by a CA (through an EE representing
the CA), nit by a CA cert.
Finaly, the document has LOTS of instances where is refers to objects
signed by a given cert. This has to be fixed, as certs are not used
to sign objects; they are used to verify signed objects.
I'l, send Geoff a marked up version of the manifest I-D with some of
the edits based on my comments above.
Steve
_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr