RE: Accountability in AC4CSR

On Thu, 7 Feb 2008, Close, Tyler J. wrote:
> >
> > A hostile client can already do cross-site third party requests.
> 
> But can the hostile client convincingly blame another site for the 
> request?

Yes, of course. The Referer header (which is what is currently used to 
determine who sent the request) can obviously be faked along with 
everything else. Referer-Root is only a subset of Referer -- it has the 
path information removed, so that we can include it without leaking 
privacy-critical information like account IDs which might be in the path 
or CGI parameters of the requesting page.


> That's the new part.

Referer-Root is not new. It's a subset of an existing header.


> A hostile client can send a request that looks like it was sent by an 
> honest client and is the fault of the Referer-Root site.

A hostile client can take a request from party A, change it, send it to 
party B, without ever involving evil party C. It can just _be_ the evil 
party. The only way around this is for parties A and B to use encryption 
or signing from the server side, without trusting the hostile client at 
all. This is the case both today, without Access-Control, and with any 
implementation of Access-Control that I can imagine.

-- 
Ian Hickson               U+1047E                )\._.,--....,'``.    fL
http://ln.hixie.ch/       U+263A                /,   _.. \   _\  ;`._ ,.
Things that are impossible just take longer.   `._.-(,_..'--(,_..'`-.;.'

Received on Thursday, 7 February 2008 23:01:15 UTC