W3C home > Mailing lists > Public > ietf-http-wg@w3.org > April to June 2009

Re: absolute URI in request to origin servers which are also proxies.

From: Adrien de Croy <adrien@qbik.com>
Date: Tue, 02 Jun 2009 15:15:13 +1200
Message-ID: <4A249941.7000207@qbik.com>
To: David Morris <dwm@xpasc.com>
CC: HTTP Working Group <ietf-http-wg@w3.org>


David Morris wrote:
>
>
> On Tue, 2 Jun 2009, Adrien de Croy wrote:
>
>>
>> I think you'll find there are a lot of proxy products which also act 
>> as origin servers for whatever reason.  Admin sites, or even reverse 
>> proxying are common examples of this.
>>
>> There were semantic differences in HTTP/1.0 between "proxy" and 
>> "server" requests - e.g. those which contained an absoluteURI or not.
>>
>> This distinction was removed in 1.1, by requiring origin servers to 
>> accept absoluteURI in a request.
>>
>> however the distinction is still required in cases such as auth. 
>> Configuration doesn't make any difference for this.  the proxy / 
>> server can no longer tell what the appropriate response should be.
>>
>> Take a simple example.  Proxy at A is also a server for 
>> www.example.com, but requires authentication for access to that site 
>> and also requires proxy users to auth.  Proxy A receives a request 
>> from a client on the internal network for
>>
>> GET http://www.example.com HTTP/1.1
>> Host: www.example.com
>> etc
>>
>> Should the proxy send back  a 407 or a 401?  should it send 
>> Proxy-Authenticate or WWW-Authenticate?
>>
>> How should it tell what to do?
>>
>> This isn't a configuration issue.
>
> Yes it is ... in this example, it would see the request as local 
> content since it was recognized as such and perform WWW-Authenticate, 
> nothing else
> would be sensible.
OK, so we send back a 401, with WWW-Authenticate

But then what does the client do if it thought it was talking to a 
proxy?  I guess there's no problem, since auth will work through a proxy 
already...

So IOW the handling of the request is based on the proxy / servers 
knowledge of which sites it's an O-S for.

>
> It must be able to recognize local content and it's own host name.
understood.

What about the case where a proxy offers authenticated proxy services 
over the internet but is also an origin server.  If a request to an 
origin server can contain an absolute URI then the proxy can no longer 
tell whether it's being asked to be a proxy for that domain, or be a 
server for that domain.   If it does host the domain, then it can assume 
it's being an origin server without problem, but what if it doesn't host 
it?  Should it then presume it's being asked to be a proxy?

if it sends back a 407 (which could be reasonable), then a UA may choke 
on that if it thought it was making a request to an origin server.

I guess this isn't a big issue... but if relative URI requests die out, 
then this could become bigger.  Which makes me wonder, why after over 10 
years of RFC2616 doesn't any UA send absoluteURI in requests except when 
configured to talk to a proxy?

>
>
>>
>> Adrien
>>
>> David Morris wrote:
>>>
>>> On Tue, 2 Jun 2009, Adrien de Croy wrote:
>>>
>>>> There is an ambiguity brought about by the requirements of RFC2616 
>>>> to require origin servers to support absolute URI in requests 
>>>> (RFC2616 s 19.6.1.1).
>>>>
>>>> If you consider the case of a proxy which is also an origin server, 
>>>> then you get a problem when the proxy receives a request for a URI 
>>>> for which the proxy is an origin server, and that request contains 
>>>> an absolute URI, and the proxy requires the client to be 
>>>> authenticated.
>>>>
>>>> In such cases, it's impossible to tell whether to send a 
>>>> Proxy-Authenticate, or a WWW-Authenticate header in the response, 
>>>> since you can't tell from the syntax of the request whether the 
>>>> client believes it is talking to a proxy or an origin server.
>>>
>>> Huh ... a proxy which can't differentiate based on local 
>>> configuration whether content is local or proxied is fundamentally 
>>> flawed and I suspect has the potential for security issues. The 
>>> shape of the URL should not influence how the server differentiates 
>>> handling.
>>>
>>>> I've never seen a client send an absoluteURI in a request to an 
>>>> origin server, so the obvious solution is to ignore RFC 19.6.1.1, 
>>>> and base the decision on the semantics of the request.
>>>
>>> NO ... the decision must be made based on the configuration of the 
>>> proxy software.  Having the client able to influence the choice 
>>> makes no sense.
>>>
>>>> However that doesn't seem like a great option if clients are one 
>>>> day going to start send requests to origin servers with absolute URIs.
>>>>
>>>> Should this requirement be deprecated?  Otherwise how should this 
>>>> case be handled?  A proxy that can also be an origin server needs 
>>>> to know
>>>
>>> It doesn't need to know what the client believes. I frankly think it 
>>> foolish to combine the functions in a single software installation, But
>>> the combined server needs external configuration to know what it is 
>>> willing to do for the client.
>>>
>>
>> -- 
>> Adrien de Croy - WinGate Proxy Server - http://www.wingate.com
>>
>

-- 
Adrien de Croy - WinGate Proxy Server - http://www.wingate.com
Received on Tuesday, 2 June 2009 03:12:28 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Friday, 27 April 2012 06:51:03 GMT