On Fri, Jan 11, 2008 at 04:02:28PM -0600, Jon McLachlan wrote: > (please correct me if I'm incorrect but...) > > if the adversary controls your entry-guard (which is nearly impossible > to detect and considered a 'strong' adversary) > if the adversary controls input to your tunnel (like text in an email, > which is easy) > and, if you do not use end to end encryption, > > Then, the adversary can perform traffic analysis on the exit node, and > the adversary can easily discover your true ip. >
The mantra statement I've been using since '96 is: Onion routing protects against traffic analysis not traffic confirmation. Roughly, if the adversary controls both ends of the connection (entry and exit node, entry ISP and exit node, entry node and destination web page, or any variant of these or similar combinations) they will be able to pretty trivially confirm by packet counting and timing that a stream at both ends is linked and "discover your true ip". If they only control one of those ends, they cannot do such linking. There are caveats and subtleties, e.g., website fingerprinting can have some limited statistical success for Tor circuits if only the entry side is known. And there are other subtleties. But it remains rough approximation of the true picture. HTH, Paul
