A warning to proxy writers

[Adam Langley]

Writing a Tor HTTP proxy is probably a lot harder than you think, for
lots of reasons.

You would scrub the User-Agent and Referer headers, of course, because
they divide the anonymity set. Yet all that information (and more) is
still around:

http://www.imperialviolet.org/browser-information.html

and it can be returned to the server via XMLHTTPRequest.

Next, any embeds in the HTML can trigger plugins which have their own
proxy settings. Realmedia objects will almost certainly start a
connection to the given server, Flash I don't know about, but I would
guess so. Flash objects can also be used to store cookies which aren't
handled via Cookie headers nor the browser.

If the user doesn't have every protocol proxyied then an image link to
https:// or ftp:// etc could cause a non-Tor connection.

Even PDFs can have embedded Javascript which can causes Acrobat to
make a connection.

You can generally craft malformed HTML or CSS which will only cause
certain browsers/versions of browsers to fetch an image.

It's a really hard problem. The Freenet proxy was very aggressive, but
the advantage was that Freenet websites were designed to work with it
(so, no Javascript etc). The general web, however, isn't designed to
work with Tor.


AGL

-- 
Adam Langley                                      agl at imperialviolet.org
http://www.imperialviolet.org                       (+44) (0)7906 332512
PGP: 9113   256A   CC0F   71A6   4C84   5087   CDA5   52DF   2CB6   3D60