[dhcwg] Re: Comments on draft-ietf-dhc-pktc-kerb-tckt-00.txt

[Ken Raeburn <[e-mail]>]

Sam Hartman <[e-mail]> writes:
> Hi.  I'm not subscribed to the list, so please copy me on responses.
> I'm sending these comments because a review of the option from a
> Kerberos viewpoint was requested.

Same here....

> The rationale for this draft claims that it will be used among other
> reasons to cancel subscriber service.  The presence of a Kerberos
> ticket should not be used to imply authorization to use a service;
> this is incompatible both with RFC 1510 and with current Kerberos
> drafts.  Even if the authorization data field is used to convey
> authorization along with authentication, service tickets with should
> not be issued with lifetimes longer than are acceptable.  I.E. you
> should only issue a ticket with a lifetime as long as you are willing
> to have authorization cached.

Agreed.  The security model should not rely on the CCD behaving
itself nicely.

Even if you want long-lived tickets, presumably to reduce the
frequency with which public-key PKINIT operations need to be done, and
if you want them to carry authorization data, you could use renewable
tickets or server-based hot-lists to close out a customer's access
fairly quickly.  Renewable tickets require repeated communication
with the KDC to get new valid tickets with later expiration times,
but they don't require the same public-key operations, and thus
should be cheaper in CPU time.

>   The option if it exists should serve to invalidate caches to
> deal with operational problems such as emergency rekeying of services
> or testing,

These do seem like valid reasons.

>   and the only effect of failing to honor this option should
> be a temporary loss of service.

If the issue is that the old service key has been exposed, failing to
honor this option could result in transmitting arbitrary data
"securely" to an intruder, or in a way that an intruder may be able to
read it.  I don't think there's much that can be done about that; it's
comparable to choosing not to check a certificate revocation list.

> The format of this option is insufficiently extensible.  Many of the
> sixteen code points for available services have already been used and
> experiences shows new Kerberos services tend to be deployed over time.
> Why would you want to selectively invalidate some but not all tickets?
> If this option exists, it seems like providing a mechanism to
> completely invalidate the cache should be sufficient.

Completely invalidating the cache means starting over with public key
ops, and reauthenticating to unaffected services that may be in active
use at the time.  There may be some benefit to invalidating only
certain tickets, in cases like emergency rekeying of a single
compromised service.

I won't speak to the extensibility issue, not knowing much about the
Packet Cable context.


> Finally security considerations need to discuss the disadvantages of
> storing persistent Kerberos tickets.  Other deployments of Kerberos
> are moving away from doing this and instead storing tickets in
> volatile memory.  It would be excellent if the write up of this issue
> in the security considerations section gave numbers to justify the
> need for persistent tickets, although requiring this would be
> excessive.

Keeping Kerberos tickets for the long term is probably not a great
idea, though since the CCD is presumably storing a public key pair
anyways, the usual reasons concerning limited exposure after a
compromise of the device don't apply as much.  The difficulty of
revocation is a more interesting issue.

However, since I'd expect the CCD is likely to reconnect and
reauthenticate to the system on power-up, avoidance of a huge server
load spike after a power outage by retaining tickets for a while would
be a good thing, especially with PKINIT.  (As would randomized delays,
if the retained tickets are found to have already expired at power-up
time.)


Oh, and an editorial issue: "Persist" is not a transitive verb in any
English-language dictionary I've checked; it doesn't take an object.
"The CCD persists Kerberos tickets" is just wrong.  "Persist" is what
the tickets do; it's not done to them.  Perhaps "retain" or "store"
should be used, maybe specifically mentioning persistent storage?
("Persistent", instead of "persisted", is fine as an adjective.)

Ken
_______________________________________________
dhcwg mailing list
[e-mail]
https://www1.ietf.org/mailman/listinfo/dhcwg