[cabfpub] CNAME-based validation

Wed Sep 7 18:51:01 UTC 2016

Jeremy,

It seems that there are two different discussions here that are getting conflated.

The first is allowing someone to confirm control of a standard domain by using a CNAME record instead of a TXT record.  The process would be very similar: prepend a defined string to the domain and set the canonical name to something containing a random value.

The second is authentication with one level of indirection.  I think that is already mostly covered as the Authorization Domain Name can be the result of dereferencing a CNAME pointer.  So if you have:

shop.example.com IN CNAME webstorehosting.webpki.com.

You can run the validation for webstorehosting.webpki.com using constructed email to domain contact, agreed-upon change to a website, DNS change, test certificate or TLS using random number.  If this passes, then shop.example.com is validated.

Does that not solve your issue?

Thansk,
Peter

> On Sep 6, 2016, at 5:07 PM, Jeremy Rowley <jeremy.rowley at digicert.com> wrote:
> 
> Hi Ryan, 
>  
> Pardon my DNS ignorance, but how does the server know which CNAME record to use in this case? If shop.domain.comCNAMEs to digicert.com and *.domain.com CNAMEs to google.com, then how do you know which will resolve? There isn’t a weight/ordering preference for CNAME like there is TXT records.  
>  
> As for how to resolve this, what if the CA generated two random numbers, holding one of the numbers in reserve. Immediately after verifying that <rnd1>.domain.com CNAME points to a validated domain, the CA can then check <rnd2>.domain.com to see if it resolves. If it resolves successfully, a wildcard DNS is in place and the domain is not validated. If the domain does not resolve, a wildcard DNS is not present and the domain is considered validated. Does this resolve your concern?
>  
> Jeremy
>  
> From: Ryan Sleevi [mailto:sleevi at google.com] 
> Sent: Friday, September 2, 2016 7:40 PM
> To: Jeremy Rowley <jeremy.rowley at digicert.com>
> Cc: Peter Bowen <pzb at amzn.com>; public at cabforum.org
> Subject: Re: [cabfpub] CNAME-based validation
>  
> Hi Jeremy,
>  
> I'm afraid you may have misread my example.
>  
> shop.example.com is a CNAME to paymentprovider.example.org (curse our lack of distinct TLDs). It does not resolve to example.net
>  
> On Fri, Sep 2, 2016 at 6:29 PM, Jeremy Rowley <jeremy.rowley at digicert.com> wrote:
> Here’s the steps:
> 1)      CA verifies example.net using any one of the methods permitted
> 
> 2)      CA receives request for shop.example.com
> 
> 3)     CA retrieves CNAME for shop.example.com (permitted under Authorization Domain Name definition: “The CA may use the FQDN returned from a DNS CNAME lookup as the FQDN for the purposes of domain validation”)
> 
> 4)     CNAME causes resolution to example.net. 
> 
> 5)     Example.net is verified under step 1, permitting issuance of the certificate 
> 
>  
> 
>  
> 
>  
> From: Ryan Sleevi [mailto:sleevi at google.com] 
> Sent: Friday, September 2, 2016 5:12 PM
> To: Jeremy Rowley <jeremy.rowley at digicert.com>
> Cc: Peter Bowen <pzb at amzn.com>; public at cabforum.org
> 
> Subject: Re: [cabfpub] CNAME-based validation
>  
> Right, and I'm not sure that (2) sufficiently establishes that example.com is authorizing the request, given wildcards.
>  
> I think this would be a similar concern with, say, following HTTP open redirects and saying a "200 OK" code authorizes the request - wellllll, not really.
>  
> I'm curious if you could expand more why you believe other methods would permit a cert to be issued in the presence of this Wildcard DNS.
>  
> As a concrete example:
> example.com A [my host]
> *.example.com CNAME example.net
> shop.example.com CNAME paymentprovider.example.org
>  
> Under this scenario, could you explain what ways that the [<rnd>.example.com] would be able to issue a cert for either example.com or shop.example.com ? Perhaps I'm just missing the implications of the existing validation methods.
>  
>  
>  
>  
> On Fri, Sep 2, 2016 at 3:33 PM, Jeremy Rowley <jeremy.rowley at digicert.com> wrote:
> Yes. Those are the two steps I am proposing.
> 
> -----Original Message-----
> From: Peter Bowen [mailto:pzb at amzn.com]
> Sent: Friday, September 2, 2016 4:31 PM
> To: Jeremy Rowley <jeremy.rowley at digicert.com>
> Cc: Ryan Sleevi <sleevi at google.com>; public at cabforum.org
> Subject: Re: [cabfpub] CNAME-based validation
> 
> I think you are talking about two different things.
> 
> Ryan is concerned that a customer has an existing record that is “*.example.com CNAME vanityblogs.example.net”.  If you say “if you can make a record show up at _309654fddb59444d8efd6d2cc98881d5.example.com you are validated” that is bad.  Instead, you are proposing a two prong validation test:
> 
> 1) Confirm control of vanityblogs.example.net.
> 2) Make _309654fddb59444d8efd6d2cc98881d5.example.com point to vanityblogs.example.net.
> 
> You are proposing that passing these both would confirm control of example.com, right?  And this would allow getting a certificate for shop.example.com, correct?
> 
> Thanks,
> Peter
> 
> > On Sep 2, 2016, at 3:25 PM, Jeremy Rowley <jeremy.rowley at digicert.com> wrote:
> >
> > We are talking about the same thing (wildcard DNS records).
> >
> > Examples:
> >
> > sleevi.example.com CNAME vdomain.com
> > *.example.com CNAME vdomain.com
> > <rnd>.example.com CNAME vdomain.com
> >
> > The BRs permit verification of each of these by establishing control over vdomain.com (under the definition of authorization domain name).
> >
> > If *.example.com can be verified this way, what is different between verifying *.example.com and <rnd>.example.com to verify all sub domains of example.com? All the RND does is make it so the website doesn’t have to point to vdomain.com.
> >
> > Jeremy
> >
> > From: public-bounces at cabforum.org [mailto:public-bounces at cabforum.org]
> > On Behalf Of Ryan Sleevi
> > Sent: Friday, September 2, 2016 4:05 PM
> > To: Jeremy Rowley <jeremy.rowley at digicert.com>
> > Cc: public at cabforum.org
> > Subject: Re: [cabfpub] CNAME-based validation
> >
> > Jeremy,
> >
> > Perhaps it wasn't clear, I wasn't speaking of wildcard certificates, but wildcard DNS rules, in which all requests for a given subdomain return a preconfigured record type. While for TXT and CAA records this is quite uncommon, it's exceedingly common to have CNAME records.
> >
> > That is, both <rnd>.example.com and sleevi.example.com may both CNAME to example.com, by virtue of of the host putting a rule of "*.example.com 3600 CNAME example.com"
> >
> > I am attempting to assert that placing the <rnd> in the subdomain is insufficient proof of authorization, and is meaningfully and tangibly different than the proof of control demonstrated in 3.2.2.4.7.
> >
> > As I read your wording, it suggests the following:
> > CA looks up <rnd>.example.com
> > <rnd>.example.com points to example.com CA sees it previously issued a
> > certificate for example.com using one of the other methods CA issues
> > certificate for <rnd>.example.com
> >
> > That concerns me.
> >
> > Peter's rewording suggests the inverse:
> > CA looks up _certvalidation.example.com _certvalidation.example.com
> > points (CNAMEs) to <rnd>.validation.[nameofca].com CA issues
> > certificate for example.com
> >
> > This is much less concerning.
> >
> > Could you help clarify which you intend, and for what names/purposes?
> >
> >
> > On Fri, Sep 2, 2016 at 2:52 PM, Jeremy Rowley <jeremy.rowley at digicert.com> wrote:
> > Wildcard domains are already allowed. We can verify Wildcard DNS because a CNAME for *.domain.com is pointing to a record previously verified. This verification method is permitted under the definition of Authorization Domain Name (where the FQDN returned by a CNAME lookup can be used to verify the requested FQDN). Although <rnd>.domain.com isn’t necessarily distinguishable from *.domain.com, the validation ends up being the same because either its considered an Authorized Domain Name (under the definition) or it was validated as a random value in this new method.
> >
> > For example:
> >
> > *.domain.com -> dcv.example.com (validated under the Authorized Domain
> > Name section) <rnd>.domain.com ->validation.example.com (validated
> > under this new section)
> >
> > Because each is validated properly, tracking which exact section was used in the validation isn’t necessary.
> >
> > Jeremy
> >
> > From: Ryan Sleevi [mailto:sleevi at google.com]
> > Sent: Friday, September 2, 2016 3:28 PM
> > To: Jeremy Rowley <jeremy.rowley at digicert.com>
> > Cc: public at cabforum.org
> > Subject: Re: [cabfpub] CNAME-based validation
> >
> >
> > _______________________________________________
> > Public mailing list
> > Public at cabforum.org
> > https://cabforum.org/mailman/listinfo/public
> 