OASIS Mailing List ArchivesView the OASIS mailing list archive below
or browse/search using MarkMail.

 


Help: OASIS Mailing Lists Help | MarkMail Help

uddi-spec message

[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]


Subject: RE: [uddi-spec] Proposal 16: breaking the containment model


Please see inline.
 


From: Andrew Hately [mailto:hately@us.ibm.com]
Sent: Monday, April 12, 2004 11:33 PM
To: uddi-spec@lists.oasis-open.org
Subject: RE: [uddi-spec] Proposal 16: breaking the containment model


The registry cannot add transforms since the effect of them is applied prior to calculating the signature. 
DF: Well, then that will not work :)
 
What I have to document is a set of well known transforms for UDDI and what they mean as far as the signed content.  As you point out, this does not change much in the specification and is intended to enable different publishers to modify different pieces of signed data under the existing containment model.  The only reason to document a few well known transforms is that inclusion of  a Transform (beyond enveloped transform and canonicalazation) is something that is often discouraged, due to the effect that transforms exclude and modify the orignal data being signed.  If the transform and the effect of the transform is not well known, it should be considered a potential security hole when verifying the signature.
 
DF: If we keep the containment model unchanged for V.Next, then the registry can be held accountable for making sure any signature applied to an entity conforms to a known set of transforms, which should be based on the granularity of our ACL expressions.  In fact, there should probably be exactly one valid set of transforms per each entity type.  That combined transform should filter out all contained entities that have their own signature element.  If a signature does not have the required transform, then the signed entity should be rejected by the registry as non-conformant to signature rules.

I agree that there are other attractive alternatives to the current model which involve breaking containment, but I believe we need to document several approaches to access control, since there is a high implementation cost (associated with changing the data model.  Of larger concern than the implementation cost is the stability of UDDI as a technology.  This should force us to examine all proposed solutions to see if there is an alternative that fits within the some constraints imposed by the UDDI V3 specification and data model. 
DF: Very good points. Granted, for each design change (rather than extension) to the V3 spec we should seek to balance its worth against the discontinuity we create.

I'm not sure if it's reflected in the minutes, but there are several proposals where we should continue to develop at least two solutions, one which uses only the existing structures and containment and one which introduces new structures and a new containment model.  We can weigh the benefits/costs of each approach when all of the solution alternatives are developed. 
DF: I didn't see the containment-free proposal posted by or assigned to anyone.  If there is interest in pursuing that, then I can volunteer to document one option.  Is proposal 16 the right place for it?

Regards,

Andrew Hately
IBM Austin
UDDI Development, Emerging Technologies
Lotus Notes: Andrew Hately/Austin/IBM@IBMUS
Internet: hately@us.ibm.com
(512) 838-2866,  t/l 678-2866



"Daniel Feygin" <feygin@unitspace.com>

04/11/2004 11:07 AM

To
"'Rogers, Tony'" <Tony.Rogers@ca.com>, <uddi-spec@lists.oasis-open.org>
cc
Subject
RE: [uddi-spec] Proposal 16: breaking the containment model





I was reading too much into it.  It is obvious that if we split entity ownership across multiple publishers, then each publisher would sign just the part of the entity that (s)he owns.
 
But I do not see why this singular issue cannot be accommodated within the existing containment model.  All we have to do is specify that signature is computed based upon entity content excluding contained entities (services, bindings and potentially contacts).  Contained entities need to have their own signatures produced by their respective owners.  These signatures would validate the contained entities' inclusion in their containing entity because their content includes the key of the containing entity.  Whether a publisher can submit a contained entity for inclusion in some other entity should be governed by the containing entity's ACL.
 
As far as spec goes, it should say "all children of the element being signed are included in the generation of the signature unless they have their own signature element," so all children endowed with signature are automatically excluded.  This is different from current spec text, which states that "all children of the element being signed are included in the generation of the signature unless first excluded by application of a transform."  This appears to presume that all of the entity's contained entities are owned by one publisher.
 
The registry itself can add the required transforms (that filter out contained keyed entities) to signatures that do not have such transforms.
 
I still believe that the publisherAssertion containment-free approach is a better alternative, because it supports greater flexibility and has better controls at the same time.
 
Daniel


From: Rogers, Tony [mailto:Tony.Rogers@ca.com]
Sent:
Sunday, April 11, 2004 1:59 AM
To:
Daniel Feygin; uddi-spec@lists.oasis-open.org
Subject:
RE: [uddi-spec] Proposal 16: breaking the containment model


That phrase "signature transforms to allow signature compartmentalization" just means writing transforms so that the signature in a business just signs the business, and not any of the contained objects, and the same for a service. That way we can suppress a binding or a service without disrupting the signature on the service or business which contains it. This was an important reason for breaking containment.
-----Original Message-----
From:
Daniel Feygin [mailto:feygin@unitspace.com]
Sent:
Sat 10-Apr-04 3:11
To:
uddi-spec@lists.oasis-open.org
Cc:

Subject:
[uddi-spec] Proposal 16: breaking the containment model

I have read the FTF minutes' discussion of proposal 16 and have these
thoughts on the matter.

First, I need to admit that I don't understand what is meant by "signature
transforms to allow signature compartmentalization" and how that would work.
It sounds like something that has the potential to make signatures work
within the framework currently proposed for requirement 16.  However I see
another option of how the concept of containment might be transformed in
V.Next to support ACL granularity and limit their impact on invalidation of
signatures.

My thoughts on this center around extending the use of publisherAssertions
to provide the mechanism to link all types of keyed entities to each other.
This would allow us to do away with containment for all keyed entities and
thereby make it easier to satisfy these requirements:
- filtering out search results inaccessible in a particular query;
- completely separating maintenance of different entities;
- supporting service projections (although they can now be deprecated if we
choose to allow multi-homed services/bindings);
- both publishers control the "inclusion";
- signing the relationship can be supported by adding two signatures ("from"
and "to" publishers') to the publisherAssertion structure

This solution would entail publishing canonical tModels to represent the
relationships between businesses, services, bindings and contacts.  It may
also provide a way to redesign isOwnedBy and isReplacedBy type of solutions
that currently rely on keyedReferences in lieu of publisherAssertion support
of uddiKey (vs. businessKey).

This would simplify the rather complicated visibility rules discussed in the
minutes.  With this proposal, it seems that they can be collapsed to just
one: if the user does not have access to one of the entities linked by the
publisherAssertion, then that publisherAssertion is invisible to the user.
Of course, this is in addition to V3 publisherAssertion visibility
constraints.  I don't really see a plausible way to reconcile ACLs with
keyedReferences (to hide keyedReferences with invisible tModelKeys), since -
unlike publisherAssertions - they are embedded inside an entity and their
exclusion would inevitably break the signature.  Perhaps we can add a rule
that by signing an entity, the publisher makes the whole entity invisible to
all inquirers who have at least one part of the entity hidden from them.
This is less of an issue if publisherAssertion linking is used, because
references to serviceKeys and bindingKeys become external to the content of
the entity.

The nice thing about this approach is that appears to simplify
implementation by reusing existing schema providing a uniform design for all
links across entities.  Requirement 27 would also be solved by this.

Daniel


To unsubscribe from this mailing list (and be removed from the roster of the OASIS TC), go to
http://www.oasis-open.org/apps/org/workgroup/uddi-spec/members/leave_workgroup.php.



[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]