Latest / CyberPodcast / RAPTOR: How Internet Routing Attacks Can Break Tor | Paper Breakdown #17 | CyberPodcast
Transcript
- 0:00Hey everyone, welcome to this explainer.
- 0:02Today we're diving into something absolutely wild in the
- 0:05world of network security. We're going to break down
- 0:08exactly how the fundamental routing infrastructure of the
- 0:11Internet itself can actually be exploited to completely bypass
- 0:14towards cryptographic anonymity. So gear up for a deep technical
- 0:18dive that is definitely going to change how you look at online
- 0:21privacy. All right, really quick.
- 0:23Here's our road map for today. We'll start with the anonymity
- 0:26myth, then breakdown the Raptor attacks themselves, cover
- 0:29asymmetric traffic analysis, look at exploiting BGP dynamics,
- 0:33shift over to active interception, and finally, we'll
- 0:35wrap up with the defense countermeasures.
- 0:38OK, Section 1, the anonymity myth, the baseline of Tor
- 0:41routing. Let's set the stage here.
- 0:43Think about this for a second. How can you deanonymize a
- 0:46completely secure, encrypted network without actually
- 0:50breaking the encryption itself? You know, we basically just
- 0:53assume that SSL or TLS guarantees absolute privacy.
- 0:56But actually, scratch that, the application layer is really only
- 1:00half the story. To really get this, we have to
- 1:02understand what an autonomous system or AS actually is.
- 1:06This is the crucial technical vocabulary for everything we're
- 1:08about to discuss. Think of an AS as a massive
- 1:11network or cluster of networks, usually run by a major Internet
- 1:14service provider. These systems talk to each other
- 1:17using BGP, the Border Gateway protocol.
- 1:20Basically, BGP is the internet's GPS.
- 1:23It's the underlying logic deciding the exact physical
- 1:25paths that your data packets take around the globe.
- 1:28Moving on to Section 2, Raptor attacks, the disruption.
- 1:32This is where things get really interesting.
- 1:35So traditional traffic analysis attacks are what we call
- 1:38symmetric. That means an adversary has to
- 1:40physically observe the traffic on both ends of the path at the
- 1:43exact same time. So from the user to Tor, and
- 1:46then from Tor to the final server.
- 1:49Honestly, that's incredibly difficult to pull off in the
- 1:51real world. But these novel Raptor attacks,
- 1:54they're total game changer. They introduce brand new BGP
- 1:57based vectors, proving that symmetric attacks are really
- 2:00just the tip of the iceberg compared to asymmetric analysis,
- 2:04BGP churn and active hijacks. Section 3.
- 2:07Asymmetric traffic analysis, the technical deep dive.
- 2:11Let's look at the actual mechanics of how this works.
- 2:14Now you might be wondering, doesn't Tor's encryption stop
- 2:17this while the payload is encrypted?
- 2:19Yes, but the underlying TCP headers?
- 2:21Those are just sitting right there in plain text.
- 2:24By observing those unencrypted headers, an attacker can simply
- 2:26extract the TCP sequence and acknowledgement numbers.
- 2:29It's like reading the tracking numbers on a perfectly sealed
- 2:32envelope. By correlating those endpoints
- 2:34over time, the cumulative TCP acknowledgements leak massive
- 2:37amounts of usable data. They can perfectly track the
- 2:40connection even though they have zero clue what the actual
- 2:42message says. And the result of this
- 2:45correlation technique? It's honestly staggering.
- 2:48Experimental data shows that this asymmetric traffic analysis
- 2:52can deanonymize a Tor user with 95% accuracy.
- 2:56And get this, that comes with literally zero false positives.
- 3:01All right, Section 4. Exploiting BGP dynamics.
- 3:05The threat of natural churn. Let's see how the Internet
- 3:08itself helps the attacker. You see, Internet routing isn't
- 3:12static. It's actually inherently
- 3:14chaotic. We call this phenomenon BGP
- 3:17churn. Natural routing changes are
- 3:19happening constantly due to routine link failures or ISP
- 3:22policy shifts. Because of this, an increasing
- 3:25number of Tor users are automatically exposed to AS
- 3:28level adversaries over time. Because these BGP paths are
- 3:31dynamically shifting around, a passive attacker doesn't even
- 3:34have to lift a finger, they literally just have to sit there
- 3:36and wait, and eventually the Internet will just naturally
- 3:39route that traffic right through their surveillance nodes.
- 3:42To put that into perspective, just factoring in the routing
- 3:45asymmetry increases an attacker's surveillance capacity
- 3:48by 50%, but when you add in that natural chaotic BGP churn, their
- 3:53capability grows by a full 100% over the course of a single
- 3:57month. That is a massive expansion of
- 3:59threat surface for doing absolutely nothing.
- 4:02Section 5. Active interception and
- 4:04hijacking, escalating the attack.
- 4:07This is where the adversary stops being passive.
- 4:10The crucial distinction to make here is between a BGP hijack and
- 4:13ABGP interception. In a traditional hijack, the
- 4:17attacker forcefully reroutes the traffic, but the connection just
- 4:20drops or gets black holed. It's noisy interception, on the
- 4:23other hand, is so much more insidious.
- 4:26The traffic is actively routed to the attacker for analysis,
- 4:29and then here's the kicker, it is silently forwarded back to
- 4:32the legitimate tour guard relay. It keeps the connection
- 4:35perfectly alive, the client gets the anonymized, and nobody
- 4:38notices a single thing. So how bad is the exposure here?
- 4:42Over 90% of Tor relays are highly vulnerable to these
- 4:45specific active manipulations. 9 out of 10, and it all boils down
- 4:49to the specific way their IP prefixes are currently
- 4:52configured at that fundamental BGP protocol level.
- 4:56Finally, section 6 defense countermeasures The patch.
- 5:00How do we actually fight back against this?
- 5:03To catch these attacks, we absolutely need a 2 pronged
- 5:07detection framework. On the theoretical side, we need
- 5:10control plane monitoring. This means analyzing massive BGP
- 5:14routing tables via services like route views, using specific time
- 5:18and frequency heuristics to spot anomalies.
- 5:21But that's not enough on its own.
- 5:23On the physical side, we need data plane monitoring.
- 5:25This involves running constant trace route tests globally,
- 5:28using hundreds of Planet Lab nodes to detect the actual
- 5:32physical shifts in the data path.
- 5:34But detection is one thing, prevention is another.
- 5:37The ultimate fix for Tor operators is surprisingly
- 5:40straightforward. Configure the relays to
- 5:43advertise exactly slash 24 prefixes.
- 5:46Right now, most relays use prefixes shorter than a slash
- 5:4924. Attackers exploit this because
- 5:51in Internet routing, the most specific route always wins.
- 5:54However, autonomous systems universally filter out anything
- 5:57longer than a slash 24. So if an operator advertises
- 6:00exactly that slash 24 prefix, it completely prevents the hijack.
- 6:04It mathematically blocks the attacker's ability to broadcast
- 6:07a more specific overriding route.
- 6:09Boom, attack mitigated. So what do the Raptor attacks
- 6:13ultimately teach us? Well, they prove that encryption
- 6:15alone just isn't enough. If fundamental Internet routing
- 6:18protocols can effortlessly bypass our best application
- 6:21level encryption, are any of our current anonymity systems truly
- 6:25secure without real network level awareness?
- 6:27Security doesn't exist in a vacuum, completely disconnected
- 6:30from the actual tracks the data travels on.
- 6:32I'll leave you to chew on that one.
- 6:34Thanks so much for joining me for this explainer and as always
- 6:36keep exploring.