On 7 January 2013 14:04, Carl Wallace <[email protected]> wrote:
> Here are a few questions/comments on draft-laurie-pki-sunlight-05.
>
> - In section 2.1.4, should algorithm requirements be stated using RFC2119
> language?

Yes.

> - In section 3.2, why would multiple SCTs be included in a handshake - for
> sct version diversity, for log signer diversity, for certification path
> diversity, something else?

To increase the chance that one of the SCTs is valid (i.e. the log is
still accepted by the client).

>  Is there any required relationship between sct
> included in a handshake and certificates returned in the handshake?

I'm not sure what you mean? The SCT has to be for the end entity certificate.

>
> - In section 4, what key is subject to the "one-to-one" mapping - the
> server's TLS key or the log signing key?  I am assuming these are
> different keys and it means the log signing key, but this could made more
> clear.

Yes.

> In either case, why is this mapping necessary?  Given only the key
> ID appears in the sct, verification occurs with no knowledge of which <log
> server> prefix was employed.  Why impose a different requirement for these
> API calls?  The requirement interferes with mirroring log contents at a
> different <log server> prefix.

Hmmm. Good point.

> - In 4.1, if multiple clients submit the same certificate, is the same sct
> signature returned to each?

This is addressed in 3. "If the log has previously seen the
certificate, it MAY return the same SCT as it returned before"

> What if the path is different for different
> submissions?

Only the end entity certificate is used to make the SCT.

>  What should occur if the certificate contains log data
> already?

Nothing special, I think? i.e. it is just part of the data in the certificate.

>  Should it matter to a server operator if a certificate is
> already present in the log when he attempts to add the certificate to a
> log (probably not but this may be worth noting)?

I guess that depends what the provenance of the cert is, not sure I
have anything intelligent to say about that.

> - In 4.1, what is the output when the log refuses to log a certificate
> presented by a client?  Generally, all of the section 4 sections would
> benefit from discussion of error returns.

Already addressed.

> - In section 7.2, if a server operator wants to make sure there are no
> other certificates for names of interest, must the entire log be
> monitored?

Yes.

>  It'd be nice if there were a client message that took a domain
> name as input.

That is a service a monitor could provide to its clients.

> Of course, this could probably be achieved by sticking an
> X.500 or LDAP interface on the log instead:-)  Should a server operator
> take any steps if it finds a new sct that chains to a different root than
> the sct it includes in TLS handshakes?

I don't know, but I also don't think CT is a chain discovery
mechanism, just an end entity cert discovery mechanism.

> - In section 9, what does "new log" mean?  Doesn't the spec already
> support key rollover by simply including a different path in the <log
> server> prefix?  Why not just use that for key rollover and define a "log"
> as the <log server>/LogID pair?  If you are thinking about supporting
> multiple keys under the same <log server> prefix, the one-to-one mapping
> requirement in section 4 should be removed.

The note is there in response to a previous comment - in short, we
haven't really decided what to do, but my inclination was as you say -
define the log by its log id (I don't really see what the cost is for
a log server operator to start a new log compared to key rollover -
if, indeed, they even get a second chance).

> - I agree with JeffH regarding lack of TLS client (at a minimum)
> verification steps.

Agreed,.

> - It would be helpful if each type of client with different functionality
> (log, TLS, audit, monitor) were introduced early in the document and if
> unqualified "client" references were identified by type.

Its a bit tricky, since that would cause forward references to how
they perform their function - which would,. no doubt, lead to
suggestions that we should put the messages and log structure earlier
in the document...

Client references have been qualified anyway.

> - Having a client message to retrieve roots recognized by the log would be
> nice, to avoid having a client repeatedly attempt to submit certificates
> to the log that will be rejected.

Good idea, I can add one.

>
>
>
> _______________________________________________
> therightkey mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/therightkey
_______________________________________________
therightkey mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/therightkey

Reply via email to