On 08/01/2012 11:05 PM, Maru Newby wrote:
Hi Adam,
I apologize if my questions were answered before. I wasn't aware that
what I perceive as a very serious security concern was openly
discussed. The arguments against revocation support, as you've
described them, seem to be:
- it's complicated/messy/expensive to implement and/or execute
- Kerberos doesn't need it, so why would we?
I'm not sure why either of these arguments would justify the potential
security hole that a lack of revocation represents, but I suppose a
'short enough' token lifespan could minimize that hole. But how short
a span are you suggesting as being acceptable?
The delay between when a user's access permissions change (whether
roles, password or even account deactivation) and when the ticket
reflects that change is my concern. The default in Keystone has been
24h, which is clearly too long. Something on the order of 5 minutes
would be ideal, but then ticket issuance could become the bottleneck.
Validity that's much longer could be a real problem, though. Maybe
not at the cloud administration level, but for a given project I can
imagine someone being fired and their access being revoked. How long
is an acceptable period for that ticket to still be valid? How much
damage could be done by someone who should no longer have access to an
account if their access cannot be revoked, by anyone, at all?
I realize that I had been thinking about the revocation list as
something that needs to be broadcast. This is certainly not the case.
A much better approach would be for the Keystone server to have a list
of revoked tokens exposed in an URL. Then, as service like Glance or
Nova can query the Revocation list on a simple schedule. The time out
would be configurable, of course.
There is a question about what to do if the keystone server cannot be
reached during that interval. Since the current behavior is for
authentication to fail, I suppose we would continue doing that, but
also wait a random amount of time and then requery the Keystone server.
In the future, I would like to make the set of Keystone servers a
configurable list, and the policy for revocation checking should be able
to vary per server: some Keystone servers in a federated approach might
not be accessible. In those cases, it might be necessary for one
Keystone server to proxy the revocation list for another server.
Let me know if this scheme makes sense to you. If so, we can write it
up as an additional blueprint. It should not be that hard to implement.
I'm hearing that you, as the implementer of this feature, don't
consider the lack of revocation to be an issue. What am I missing?
Is support for revocation so repugnant that the potential security
hole is preferable? I can see that from a developer's perspective,
but I don't understand why someone deploying Keystone wouldn't avoid
PKI tokens until revocation support became available.
Thanks,
Maru
On 2012-08-01, at 9:47 PM, Adam Young wrote:
On 08/01/2012 09:19 PM, Maru Newby wrote:
I see that support for PKI Signed Tokens has been added to Keystone
without support for token revocation. I tried to raise this issue
on the bug report:
https://bugs.launchpad.net/keystone/+bug/1003962/comments/4
And the review:
https://review.openstack.org/#/c/7754/
I'm curious as to whether anybody shares my concern and if there is
a specific reason why nobody responded to my question as to why
revocation is not required for this new token scheme. Anybody?
It was discussed back when I wrote the Blueprint. While it is
possible to do revocations with PKI, it is expensive and requires a
lot of extra checking. Revocation is a policy decision, and the
assumption is that people that are going to use PKI tokens are
comfortable with out revocation. Kerberos service tickets have the
same limitation, and Kerberos has been in deployment that way for
close to 25 years.
Assuming that PKI ticket lifespan is short enough, revocation should
not be required. What will be tricky is to balance the needs of long
lived tokens (delayed operations, long running operations) against
the needs for reasonable token timeout.
PKI Token revocation would look like CRLs in the Certificate world.
While they are used, they are clunky. Each time a token gets
revoked, a blast message would have to go out to all registered
parties informing them of the revocation. Keystone does not yet have
a message queue interface, so doing that is prohibitive in the first
implementation.
Note that users can get disabled, and token chaining will no longer
work: you won't be able to use a token to get a new token from Keystone.
Thanks,
Maru
_______________________________________________
Mailing list:https://launchpad.net/~openstack
Post to :openstack@lists.launchpad.net
Unsubscribe :https://launchpad.net/~openstack
More help :https://help.launchpad.net/ListHelp
_______________________________________________
Mailing list: https://launchpad.net/~openstack
<https://launchpad.net/%7Eopenstack>
Post to : openstack@lists.launchpad.net
<mailto:openstack@lists.launchpad.net>
Unsubscribe : https://launchpad.net/~openstack
<https://launchpad.net/%7Eopenstack>
More help : https://help.launchpad.net/ListHelp
_______________________________________________
Mailing list: https://launchpad.net/~openstack
Post to : openstack@lists.launchpad.net
Unsubscribe : https://launchpad.net/~openstack
More help : https://help.launchpad.net/ListHelp
