[CAP] Then Again... (was Re: CAP Security Using Digital Signatures)

Thu Mar 12 09:08:08 PDT 2009

Thinking about it a bit more... am I mistaken, or wouldn't a simple  
way to implement a "sign the blob" approach be just to use a "null"  
canonicalization in otherwise-normal XMLSIG?

As I understand it, the reason for C14N is to deal with variations in  
whitespace, attribute order and such that can occur when XML is parsed  
and then re-serialized.  So we try to factor out every possible change  
that might occur in those processes... with, it appears, limited  
success.

But to simply forward/publish someone else's CAP alert, seems like  
passing it along precisely, byte-for-byte, would be relatively  
simple... and (as the Gutmann paper points out) much more auditable.   
Of course that wouldn't prevent any node from parsing the alert and  
acting on the basis of the contents.  It would only mean that if that  
node decides to pass the alert to other nodes, it must give them the  
precise version it received.

In other words, if C14N is murky bathwater, maybe we could dispense  
with it without needing to toss the baby too.  All we'd need would be  
a clarification in the OASIS spec (and an implementers' convention  
till then) that canonicalization SHALL NOT be applied during signing.   
(And to fix the schema, of course.)

Or am I missing something here?

- Art

On Mar 12, 2009, at 3/12/09 8:32 AM, Art Botterell wrote:

> Interesting link, Matt.  Sounds like an object lesson on why we  
> should resist the temptation to standardize what we don't yet  
> understand.
>
> I was thinking we might wind up proposing an erratum to OASIS to fix  
> the schema issue, but hadn't appreciated that cannonicalization was  
> still proving so intractable.  Although that article is from 2004, I  
> take it the C14N situation hasn't improved.  So maybe it would make  
> more sense to identify (and demonstrate!) alternate approaches that  
> could be fed back into the standard.
>
> My concern is that if we don't address the end-to-end signature  
> problem as a community there might not be a business incentive for  
> any particular implementer to design for that level of  
> interoperability.  And while the OASIS process usually does a good  
> job of refining and ratifying contributed designs, it seems not to  
> be as effective as a framework for developing those designs in the  
> first place.
>
> - Art
>
>
>
> On Mar 12, 2009, at 3/12/09 5:59 AM, matt hoffman wrote:
>
>> I did a proof-of-concept implementation of this on a previous DHS  
>> system. Although, as you say, there was no way of verifying it with  
>> other providers or consumers at the time.
>> However, a couple of points that I recall:
>>
>> a.) The document mentions that the signature element can be a child  
>> of "alert", but unless I'm mistaken it is not specified in the  
>> schema.  So messages containing signatures fail schema  
>> validation ... I had to explicitly add it to the schema.
>>
>> b.) The signatures are delicate, and tricky:  http://www.cs.auckland.ac.nz/~pgut001/pubs/xmlsec.txt
>>
>> Unfortunately I no longer have access to that code, but I'd be  
>> happy to help others if they're looking into it.
>>
>>
>> - matt
>>
>> On Wed, Mar 11, 2009 at 11:20 PM, Art Botterell <acb at incident.com>  
>> wrote:
>> Friends -
>>
>> We're moving rapidly toward an important threshold in CAP  
>> implementations.  So far, most CAP-based systems have been self- 
>> contained, single vendor/implementer arrangements.  But soon we're  
>> going to need to "federate" CAP traffic among multiple  
>> interoperable systems.  And that has important implications for  
>> security.
>>
>> Most current systems use a trusted-link/trusted-host mode based on  
>> encrypted network links and password access control at a central  
>> server.  That's a familiar Web 1.0 approach and it works fine for  
>> "single-hop" implementations.  But it has a major drawback: As soon  
>> as messages are forwarded from one server to another, a security  
>> compromise anywhere can compromise the authentication and integrity  
>> of all CAP messages downstream.
>>
>> The alternative, of course, is to apply digital signatures to CAP  
>> messages at their origin, and to verify them at receiving points.   
>> That way, it doesn't matter if the links or intervening nodes are  
>> secure or not; any recipient can verify independently that the  
>> message is a) from who it says it's from, and b) hasn't been  
>> modified in transit.
>>
>> There's a standard way of doing this for XML, as cited in the CAP  
>> Specification:
>>
>> >3.3.2.1 Digital Signatures
>> >The alert element of a CAP Alert Message MAY have an Enveloped  
>> Signature, as described by XML Signature and
>> >Syntax Processing [XMLSIG]. Other XML signature mechanisms MUST  
>> NOT be used in CAP Alert Messages.  Processors
>> >MUST NOT reject a CAP Alert Message containing such a signature  
>> simply because they are not capable of verifying
>> >it; they MUST continue processing and MAY inform the user of their  
>> failure to validate the signature.
>>
>> But I'm not aware of anyone that's implemented it yet... partly  
>> because it hasn't been necessary in stand-alone systems, and partly  
>> because it involves a type of programming a lot of folks haven't  
>> had occasion to explore yet.
>>
>> But ultimately, it's going to be essential for interoperability.   
>> So I'd be interested in hearing, has anyone implemented XMLSIG on  
>> CAP yet?  And would anyone be interested in doing some  
>> interoperability experiments?  The standard is there, the  
>> technology is there (in Java and a number of other languages) and I  
>> see the requirement bearing down on us quickly.
>>
>> What say?
>>
>> - Art
>>
>>
>> _______________________________________________
>> This list is for public discussion of the Common Alerting  
>> Protocol.  This list is NOT part of the formal record of the OASIS  
>> Emergency Management TC.  Comments for the OASIS record should be  
>> posted using the form at http://www.oasis-open.org/committees/comments/form.php?wg_abbrev=emergency
>> CAP-list mailing list
>> CAP-list at lists.incident.com
>> http://lists.incident.com/mailman/listinfo/cap-list
>>
>> This list is not for announcements, advertising or advocacy of any  
>> particular program or product other than the CAP itself.
>>
>
> _______________________________________________
> This list is for public discussion of the Common Alerting Protocol.   
> This list is NOT part of the formal record of the OASIS Emergency  
> Management TC.  Comments for the OASIS record should be posted using  
> the form at http://www.oasis-open.org/committees/comments/form.php?wg_abbrev=emergency
> CAP-list mailing list
> CAP-list at lists.incident.com
> http://lists.incident.com/mailman/listinfo/cap-list
>
> This list is not for announcements, advertising or advocacy of any  
> particular program or product other than the CAP itself.