Mitch:
What I interpreted from Dave's e-mail is as follows:
1. For operational security and operational workload reasons, we may not
want to bring the off-line Root on-line to generate the CRL too often.
2. For security reasons, we also want the Root CRL to have a relatively
current nextUpdate (vice one month or one year from now).
3. To meet these two, the root CA generates a bunch of CRLs in advance with
thisUpdate not of the current time, but desired times in the future and
nextUpdate also of various desired times in the future.
4. One can put these CRLs on portable media and control them under the same
physical and procedural controls that Root CA and associated private key
activation is controlled.
5. Most of the time, a CA will not be revoked and hence the proper portable
medium can be carried to the repository and CRL published.
6. If a CA is revoked, root CA will be booted up, all the portable media
will be erased and root CA will generate a bunch of new CRLs using step 3.
This does not sound technically complex to me. It does require some
procedural controls for additional items, namely portable media for the
future CRLs.
-----Original Message-----
From: owner-ietf-pkix@xxxxxxxxxxxx [mailto:owner-ietf-pkix@xxxxxxxxxxxx] On
Behalf Of Mitchell Arnone
Sent: Monday, December 30, 2002 8:13 PM
To: David P. Kemp; ietf-pkix@xxxxxxx
Subject: Re: Offline Root CA with valid CRL hierachie
Dave,
This sounds like a very complex technical solution that circumvents proper
procedures designed to maintain a significantly high level of
assurance. The Root CA is the root of trust and therefore must live up to
the highest level of assurance attested to by a particular PKI.
CRLs must be generated on the date and at the time they become
effective. Predicting contents of a CRL and generating them in advance
leaves much room and opportunity to render a PKI untrustworthy.
A CRL can be generated by an off-line Root CA where the next update can be
one day, one month or even one year from the date contained in
"thisUpdate". Depending on this setting, and assuming that no subordinate
CA certificates have been revoked prior to the normally scheduled update
period, the off-line Root CA can be brought on-line allowing it to generate
and publish a fresh CRL. In the event that a subordinate CA certificate is
revoked, obviously the Root CA must brought on-line to do so and
immediately following the revocation, the Root CA can generate and publish
an updated CRL. This is the proper way to handle an Off-line Root CA.
The burden of obtaining a current CRL is on the Relying Party where it must
periodically refresh its CRL regardless of whether the "nextUpdate" has
been reached or not. A Relying Party can flush its cache at prescribed
intervals and then retrieve a new (updated) CRL as required.
Mitch
At 11:01 AM 12/30/2002, David P. Kemp wrote:
>Because revocation of a signing CA should be very rare event, the
>offline root can take advantage of a "long pipeline" to minimize the
>human effort required to transport disks of CRLs to an online
>repository. The root could pre-generate, say, one week's worth of
>empty daily CRLs (where nextUpdate = thisUpdate + 1 day) and put the 7
>CRLs on media. The repository would then be responsible for doling out
>the correct CRL each day and not publishing any CRL for which
>thisUpdate is in the future.
>
>In the event of a CA compromise, the pipeline would need to be flushed
>- future CRLs in the repository would be destroyed and the root CA
>would generate a new batch of CRLs on media.
>
>This assumes that compromise of the online repository (allowing access
>to CRLs up to 7 days in the future) is a much less serious event than
>compromise of the offline root CA. I believe that assumption is valid
>and supportable.
>
>Dave
>
>
>
>
>Al Arsenault wrote:
> >
> > Okay, I'm a little confused here. This problem isn't really all
> > that hard to solve. My comments/questions are in-line, prefaced by
> > my initials,
> "AWA".
> >
> > ----- Original Message -----
> > From: "Haaino Beljaars" <beljaars@xxxxxxxxxxxx>
> > To: "Mark Scherling" <markscherling@xxxxxxx>
> > Cc: <ietf-pkix@xxxxxxx>
> > Sent: Monday, December 23, 2002 6:36 AM
> > Subject: Re: Offline Root CA with valid CRL hierachie
> >
> > >
> > > > You can still publish a CRL for an off-line CA. You will need
> > > > to
> > > establish
> > > > a fairly long expiry time for the CRL if you do not plan to
> > > > bring
> the CA
> > > > on-line often. In many cases the Root CA is off-line and only
> > > > used to
> > > issue
> > > > new intermediary CAs or revoke intermediary CAs. You will need
> > > > to
> > > establish
> > > > a very good procedure for startup and shutdown of the root CA
> > > > (two
> > person
> > > > control, locked in a safe, two person combination on the safe,
> > documenting
> > > > each time the CA is removed, etc. ) The reason for documenting
> > > > the
> > > process
> > > > is for audit purposes.
> > > >
> > > > You will also need to document in your CP that the CA is
> > > > off-line and
> > that
> > > > the onus is on the relying party to verify that an intermediary
> > > > CA is
> > > still
> > > > valid.
> > >
> >
> > AWA: Let's first clarify terms. What, exactly, do you mean by
> > "off-line CA"? To me, that term means that the CA is not connected
> > to a larger network (specifically, it's not connected to the
> > Internet or larger telecom network through which it can be
> > accessed). It's still running, locked in its secure facility
> > somewhere. That is, start-up/shut-down is not an
> issue;
> > it's already running. Is that what you mean, or is there something
> > else?
> >
> > > I agree that you still publish a valid CRL for an offline Root CA.
> > > The problem is the following,
> > > example: I issue an CRL with the Root CA with a validity of for
> > > example a month, and after issuing the CRL I take it offline.
> > > During that month, for example after
> > two
> > > weeks, one of the
> > > intermediate CA's is compromised and I have to revoke that CA.
> > > According to the specs I can only issue a CRL which has a validity
> > > time
> > that
> > > starts after
> > > current one.
> >
> > AWA: I'm not sure you're interpreting RFC 3280 correctly.
> > Paragraph 5.1.2.5 starts:
> >
> > 5.1.2.5 Next Update
> >
> > This field indicates the date by which the next CRL will be issued.
> > The next CRL could be issued before the indicated date, but it will
> > not be issued any later than the indicated date. CRL issuers SHOULD
> > issue CRLs with a nextUpdate time equal to or later than all previous
> > CRLs. nextUpdate may be encoded as UTCTime or GeneralizedTime.
> >
> > That is, you can only issue a CRL whose "thisUpdate" time is later
> > than the previous one, and whose nextUpdate time is equal to or
> > later than those on previous CRLs (although that latter part is not
> > required). There's no reason why you have to wait until the current
> > CRL expires to issue the new one.
> >
> > >This means in practise that I have a valid intermediate CA for
> > >over two weeks, but in reality that intermediate CA is revoked.
> > >How can I let everybody
> > know
> > > during that
> > > two week period that the intermediate CA is compromised, taking in
> > > account:"Conforming applications are NOT REQUIRED to support
> > > processing of delta CRLs,
> > indirect
> > > CRLs, or CRLs
> > > with a scope other than all certificates issued by one CA"?
> > >
> >
> > AWA: Okay, here's how I've solved this problem in the past. The
> > root CA, while off-line, is running in its secure facility. When it
> > is necessary to revoke an intermediate CA, the root CA is accessed
> > to create a new CRL. This CRL is dumped to some medium, either
> > burned onto a CD or put onto a floppy. This medium is then
> > hand-carried (the term we used to use was
> "sent
> > via sneaker-net") to a server that is on the network. It can be
> > either
> made
> > available as a CRL from, e.g., an LDAP repository; or it can be
> > provided to an OCSP responder.
> >
> > The rest is up to your revocation policy. If you rely only on CRLs,
> and you
> > allow users to cache CRLs until the end of their validity period
> > (e.g., until their "nextUpdate" time), then yes, you have in your
> > scenario a two-week window of vulnerability where users think an
> > intermediate CA is good but it's really not. That's a risk you take
> > by relying only on static CRLs with that long a validity period. (Of
> > course, one would hope that revocation of an intermediate CA would
> > be an extremely rare event, so the overall system risk/cost tradeoff
> > may be acceptable, but...)
> >
> > On the other hand, you can use an OCSP-based system, where users
> > query the responder and will discover that the intermediate CA has
> > been revoked the first time they query its status. OR you can use
> > CRLs with a shorter life cycle. Depending on what you're willing to
> > pay in "human capital" costs, there's no reason you can't have the
> > off-line CA issue a new CRL -
> generally
> > identical to the previous one - as often as you want; e.g., every
> > day,
> every
> > 4 hours, whatever. It just costs you to have the human move the
> > medium - floppy, CD - from the off-line CA to the network-connected
> repository. It's
> > not that hard.
> >
> > > Best regards,
> > > Haaino
> > >
> >
> > Hope this helps.
> >
> > Al Arsenault
> > Chief Security Architect
> > Diversinet Corp.
***********************************************************
Mitchell Arnone
Managing Consultant
SchlumbergerSema
Technical Consulting Practice, Northeast Region
Network & Infrastructure Solutions
marnone@xxxxxxxxxxxxxxxxxxxxxx
www.slb.com/nws
35 Waterview Blvd.
Suite 210
Parsippany, NJ 07054-1200
USA
Phone +1 410-579-8691
Mobile +1 443-864-1590