From: Bill Stewart <bill.stewart@pobox.com>
Date: Sat, 14 Nov 1998 16:39:15 +0800
To: Lucky Green <shamrock@cypherpunks.to>
Subject: Re: Free Email as Anonymous Remailer Re: NPR is at it again...
In-Reply-To: <3.0.5.32.19981111015237.0079f350@idiom.com>
Message-ID: <3.0.5.32.19981114002632.007caa30@idiom.com>
MIME-Version: 1.0
Content-Type: text/plain



At 11:47 PM 11/11/98 +0100, Lucky Green wrote:
>On Wed, 11 Nov 1998, Bill Stewart wrote:
>> An interesting project would be a free low-volume anonymizer cgi for Apache,
>> given the large number of current users and the much larger number
>> of people who will run web servers once they have cable modems.
>How do you do chaining with a cgi?

Looks easy enough to do, if a bit ugly, where "ugly" is somewhat equivalent to
"build yet another local proxy widget to hide the gory details",
though it's not really much uglier than doing a good anonymizer,
and getting details like cookies and Java/script right are harder.

Define "encrypted" as "PGP or something like it".  It may be possible
to gain some efficiencies by using SSL, but not critical.
Take a cgi script and use POST to hand it an encrypted block containing:
	Response-Key:  
	HTTP Request, either vanilla URL or cgi URL with GET or POST data.
	Maybe some digicash
	Maybe some additional data
The script fetches the URL, handing along any data,
packages the response in HTTP reply format, and encrypts it with the 
response key for the client proxy to unpack.

To chain these, have the client nest the requests, doing a
URL that points to another anonymizer script and POSTs an encrypted block.
Eventually you'll get to a non-anonymizing URL;
it may be interesting to include any expected cookies in the block,
so the client can hand them to the destination web server,
or to gain some efficiencies by having the cgi script fetch
any IMG requests, and sending a bundle of HTTP reply packets
instead of just a single one.

The problems - 
- How can easily can you break the system?  
--- Does it leave too many open connections that can be followed?
--- Does the decreasing size of the requests and
	increasing size of responses make it too easy to trace?
--- What other obvious security holes are there?
- Timeouts or other problems?
- Denial of service attacks?

				Thanks! 
					Bill
Bill Stewart, bill.stewart@pobox.com
PGP Fingerprint D454 E202 CBC8 40BF  3C85 B884 0ABE 4639