Re: [apps-discuss] Looking at Webfinger

["Paul E. Jones" <[e-mail]>]

Guys,

 

John, I understand the pain of getting TLS records properly installed.  It's
no small task, for sure.  I am hoping DNSSEC and RFC 6698 might help with
this.  That sounds scary, but it's really fairly simple to configure.  Most
importantly, it can be easily automated.  I can't very well automate telling
somebody to go to a CA and buy a certificate . wait, first generate a CSR
and then. it's messy.  But, it's presently a necessary evil.

 

What I am hearing is that we must allow the host-meta{.json} resource to be
relocated.  That relocation could be via HTTP redirects (already mandatory)
or something else.  The only other mechanisms that seem extremely trivial to
me are:

 

1) TXT records (e.g., _webfinger.packetizer.com IN TXT
"https://packetizer.webfinger-provider.com/")

2) A records (e.g., webfinger.packetizer.com IN A 10.1.2.3)

3) CNAME records (e.g., webfinger.packetizer.com IN CNAME
packetizer.webfinger-provider.com.)

 

I'm not a fan of any of those, but querying "webfinger.packetizer.com" seems
like it would be the easiest alternative (meaning one could use either A or
CNAME records).

 

So, the client would issue these queries in this order

1)      https://packetizer.com/.well-known/host-meta.json

2)      https://webfinger.packetizer.com/.well-known/host-meta.json

3)      http://packetizer.com/.well-known/host-meta.json

4)      http://webfinger.packetizer.com/.well-known/host-meta.json

 

Note the ordering to ensure that TLS is attempted before insecure
connections.

 

Is this something we really need to do?  If so, it also suggests that other
services rooted in /.well-known are going to present similar issues.

 

Paul

 

From: John Panzer [mailto:[e-mail]] 
Sent: Wednesday, August 29, 2012 2:13 PM
To: John Bradley
Cc: Paul E. Jones; Mark Nottingham; IETF Apps Discuss
Subject: Re: [apps-discuss] Looking at Webfinger

 

Based on experience, I'd prefer to avoid things that depend on the bare
("naked") domain (example.com instead of www.example.com or
webfinger.example.com).  I could write up the reasons but it'd take some
time.  Unfortunately, webfinger as originally spec'd requires this.

 

If we were to start over, and get to vote for whatever I wanted, and were
going to allow DNS records at all, I would probably vote for something like
webfinger.example.com as a special magic name (with the actual name chosen
so as not to collide with any existing DNS entries).  Any normal hosting
mechanisms will work here, including A records and CNAMEs and HTTP
redirects, but I'd also require everything be done via TLS with correctly
signed certificates for that subdomain (argh!  the pain!).

 

Organizationally, this would mean that any part of the organization that can
stand up a separate SSL service on a new subdomain can provide webfinger
services.  

On Wed, Aug 29, 2012 at 10:57 AM, John Bradley <[e-mail]> wrote:

I am not the best person to represent Google's needs.

 

However as I understand it Google hosts applications such as email documents
openID for tens of thousands of domains.

Google themselves don't control the DNS.

 

The people using the service generally add some MX records for
aspmx.l.google.com. and a Cname for mail.example.com to ghs.google.com.

 

The A record for the bare domain typically points off to some Content
management site the company uses for their web pages.

 

I think this is probably typical of Yahoo's mail hosting services and
others.   

 

The service hosing the email/authentication/openID is not the one that
controls the web server for company.

 

Saying the CMS venders will provide WebFinger services doesn't seem all that
likely, especially in virtual hosting environments.

 

Getting a typical company to do anything more than enter a cname for
webfinger.example.org is wildly optimistic.

 

I am entirely open to Ideas on this.   However the previous solution of
having every RP check with google first to see if they host the email and
provide the XRDS seems horribly flawed to me.

 

I would like to see a workable solution at the discovery layer that
accommodates how people deploy there sites.

 

I think Bill suggested at one point using the MX record to find the
webfinger host.  That has a bunch of problems I would prefer to avoid.

 

John B.

 

On 2012-08-29, at 1:36 PM, "Paul E. Jones" <[e-mail]> wrote:





John,

 

Well, we need to figure out how to address this.

 

Would it be reasonable to redirect requests from /.well-known/host-meta.json
and /.well-known/host-meta to Google?

 

Are there other services or files under /.well-known that Google's customers
would not want Google to host?  If they were OK with Google's servers
responding to anything , then one could put an A (or CNAME) record in place
for  <http://example.com> example.com that points to Google.

 

Not being familiar with what Google offers, I'm a bit challenged to
understand exactly what is and is not possible.

 

Paul

 

From: John Bradley [mailto:[e-mail]] 
Sent: Wednesday, August 29, 2012 10:14 AM
To: Paul E. Jones
Cc: 'George Fletcher'; 'Mark Nottingham'; 'IETF Apps Discuss'
Subject: Re: [apps-discuss] Looking at Webfinger

 

There mite be a A record but that typically goes off to some virtual web
hosting company and not the email service provider.

 

I think I have also heard William say this is a problem for Yahoo.

 

Google was not able to get people to deploy XRDS for hosted domains.   They
came up with a proprietary extension to openID discovery to make hosted
google apps domains work with some subset of RP.

 

The problem is that the company hosting a small businesses website is
unlikely to provide the web finger infrastructure and there is no way for
the email/openID provider to do it without their cooperation.

 

Adding a A record rather than a CNAME is generally not a good idea if it can
be avoided.   In the event of the provider changing an IP address it breaks
all the customers if they have used A records, but that is separate issue.  

 

You can set up webfinger on your web server and manage it.   It just won't
work for large numbers of people as we have it now.  

 

If webfinger won't work for Google Apps for Domains and other hosted
services like that then It will significantly impact adoption in my opinion.

 

We will also need to work around that for Connect.  We don't want another
proprietary work around with the security problems that can entail.

 

John B.

 

On 2012-08-28, at 11:37 PM, "Paul E. Jones" < <mailto:[e-mail]>
[e-mail]> wrote:

 

John,

 

If Google is hosting the domain or any other service provider, wouldn't
there still be an A record for the domain (e.g., <http://packetizer.com>
packetizer.com)?  I know there is for virtually every web hosting company
I've used.  It seems like this might just be one more hosted service Google
could provide to its customers, no?

 

I do not know exactly how this hosted service works, but what's hosted?  I
assume it's just email.  If web, then I see no issue.  If only email, then
the user just needs to have MX records pointing to Google and an A record
pointing to whatever service runs the WebFinger service.

 

In any case, if they can add a CNAME or MX record, I think we can get them
to add an A record.  I think it would be far more challenging for SMBs to
add a host like  <http://webfinger.example.com> webfinger.example.com.  That
would still require an A record and a service provider capable of supporting
it.

 

Paul

 

From: John Bradley [mailto:ve7jtb@ <http://ve7jtb.com> ve7jtb.com] 
Sent: Tuesday, August 28, 2012 3:29 PM
To: Paul E. Jones
Cc: 'George Fletcher'; 'Mark Nottingham'; 'IETF Apps Discuss'
Subject: Re: [apps-discuss] Looking at Webfinger

 

There are cases where there are hosted domains (Google etc) that may not
have a http host for the domain name used in the users email address.  

 

There may be merit to having a  <http://webfinger.example.com>
webfinger.example.com fallback where the client can't reach the .well-known
for the primary host.

 

I know that some sort of SRV record would be the correct way to do it, but
in the real world SMB don't enter SRV records even if there DNS provider
support them.

The most you can get them to do is add a CNAME or MX record.

 

Supporting these sorts of domains somehow is a important issue.

 

John B.

 

On 2012-08-28, at 3:17 PM, "Paul E. Jones" < <mailto:[e-mail]>
[e-mail]> wrote:





George,

 

I believe it might be useful to introduce those who break your WebFinger
server to Louisville Slugger. :)

 

Your pain is understood, but I do not see a way to avoid it.  We could
introduce something in DNS, but that would also present challenges.  No
matter where we "root" the discovery process, there is a potential somebody
could break it.  It could be rooted somewhere other than the root of the
domain (e.g.,  <http://webfinger.example.com> webfinger.example.com), but we
either need to decide in advance of such a location or introduce a way to
discovery the discovery resources.

 

Do you have a suggestion that would make this less likely to be broken?

 

Paul

 

From:  <mailto:[e-mail]> [e-mail]
[mailto:apps- <mailto:[e-mail]> [e-mail]] On
Behalf Of George Fletcher
Sent: Tuesday, August 28, 2012 11:09 AM
To: Mark Nottingham
Cc: IETF Apps Discuss
Subject: Re: [apps-discuss] Looking at Webfinger

 

Way "late to the party" but I can speak from experience that deployment can
be a real issue in some environments. It's all really straight forward in a
small company or an environment where the identity team "owns" the root
domain (e.g.  <http://example.com> example.com). However, if an entire other
group in a large organization "owns" the root domain (home page for the
site) and the identity team then needs to get them to make changes, the
deployment solution gets brittle pretty quick. I've had our webfinger
support broken at least twice because the "other" team didn't know that
certain configs were required:)

Also, installing the "dynamic pluming" can be more problematic is these
cases. It is possible to get apache rewrite rules or netscaler magic in
place to make it work, it's just a more brittle deployment architecture.

Thanks,
George

On 7/4/12 6:58 PM, John Panzer wrote:

Mark -- Of course I was speaking about practical realities of typical web
server administration and deployment.  In practical terms, adding a new
mod_rewrite rule or moral equivalent is going to be easier than adding a new
PHP script that connects to a database.  The latter is just always going to
be a much higher bar.

 

And, something that returns per-user data is generally going to need a
dynamic service of some kind, unless your site has just a handful of users
and you don't mind going through a publishing exercise each time you add or
change a user...

 

None of this has anything to do with the interface, just deployment
realities.  And in reality all of this is going to need a dynamic service
somewhere for each non-trivial site, this is all just a question of how to
hook it up.




--
John Panzer / Google
 <mailto:[e-mail]> [e-mail] /
<http://www.abstractioneer.org/> abstractioneer.org / @jpanzer

 

 

On Wed, Jul 4, 2012 at 3:36 PM, Mark Nottingham < <mailto:[e-mail]>
[e-mail]> wrote:

On 05/07/2012, at 8:16 AM, John Panzer wrote:

> Just as a historical note.  The envisioned usage of host-meta was
originally to avoid a specification which mandated a particular dynamic URL
API at a particular /.well-known endpoint (because it may not be feasible to
do that across all organizations and deployments).  The host-meta document
itself would be highly cacheable and so wouldn't incur an additional network
trip in the common case.
>
> Having a 3xx redirect is a reasonable alternative that allows a similar
escape hatch via something like mod_rewrite, albeit at the cost of needing
an additional network trip each time.  Since a deployment can always avoid
the 3xx redirect with additional dynamic plumbing behind the well-known
endpoint, I don't think that's a horrible thing.
>
> An application-level redirect would be almost equivalent to an HTTP
redirect, but then there are two ways to do the same thing.  If _only_ an
application-level redirect is allowed, then you have to have at least a
minimal dynamic service at the well-known endpoint (no more mod_rewrite).
But the whole reason for this is to avoid the requirement for a dynamic
service behind well-known...

"dynamic" and "static" are properties of the implementation, not the
interface. HTTP doesn't require that any particular URL be "dynamic";
anything can, with the right metadata, be cached (and indeed, I've cached
many, many things with the wrong metadata, because of silly site operators
and their ideas about "dynamic").

Now, if people want to target a particular implementation that makes it
easier to serve a particular style of URL without writing code, fine, but
let's not confuse things.

E.g., a URL like

 <http://example.com/.well-known/user/bob>
http://example.com/.well-known/user/bob

is easy to serve in pretty much any way you like with Apache.

I'm also going to push back on the "it may not be feasible to do that across
all organizations and deployments" motivation. This is a race to the bottom.
The trick is to make it accessible enough to get sufficient traction to pull
everyone along, without pandering to *everyone*'s requirements.

Regards,


--
Mark Nottingham    <http://www.mnot.net/> http://www.mnot.net/





 







_______________________________________________
 
 
apps-discuss mailing list
 <mailto:[e-mail]> [e-mail]
 <https://www.ietf.org/mailman/listinfo/apps-discuss>
https://www.ietf.org/mailman/listinfo/apps-discuss

 

_______________________________________________
apps-discuss mailing list
 <mailto:[e-mail]> [e-mail]
 <https://www.ietf.org/mailman/listinfo/apps-discuss>
https://www.ietf.org/mailman/listinfo/apps-discuss

 


_______________________________________________
apps-discuss mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/apps-discuss