Re: p1: additional security considerations

[Amos Jeffries <[e-mail]>]

On 23/04/2013 6:55 p.m., Willy Tarreau wrote:
> On Tue, Apr 23, 2013 at 04:17:22PM +1000, Mark Nottingham wrote:
>> On 23/04/2013, at 4:15 PM, Willy Tarreau <[e-mail]> wrote:
>>
>>> On Tue, Apr 23, 2013 at 04:02:22PM +1000, Mark Nottingham wrote:
>>>> Just wondering if we need to explicitly point out the security considerations
>>>> around the following:
>>>>
>>>> * Message routing -- it's somewhat common AIUI for intermediaries to only
>>>> route on the Host header, for performance reasons; i.e., they do not
>>>> reconstruct the effective request URI (as required by p1 5.5). I know there's
>>>> a theoretical risk here, but is there a real-world risk that we should point
>>>> out?
>>> I see no particular risk since the Host header field is mandatory. Also in
>>> practice, intermediaries which "route" requests tend to be very close to
>>> the servers, at places where the security considerations are very specific
>>> to the environment and explicitly covered in this intermediary's configuration.
>> That's what I was wondering. What concerned me was that people deploy load
>> balancers in front of proxies, and virus scanners, etc. I don't have a
>> specific attack in mind, it just feels like there probably is one.
> At least in my experience, when deploying a load balancer in front of a proxy
> farm or anti-virus farm, parts of the URI are used more than the Host header
> field. For example, you can have an LB which decides that requests for file
> ending in ".mpg" do not pass through the virus scanner and go directly to the
> proxy, but in return the content-type must absolutely match "video/mpeg"
> otherwise they're blocked (it's just an example).
>
> That's why I think that the security considerations are much more of a global
> thing in such deployments than just a matter of correctly relying on the Host
> header field.
>
> Willy
>
>

Yes. There are several vulnerabilities which Host enables or enhances. 
But they are problems which the next-hop is best placed to deal with and 
should be protecting itself against regardless of the presence of a 
routing intermediary. Even that unicity issue Willy pointed at is 
something the origin server or a cache is best placed to reject.

Amos