Latest / CyberPodcast / How Low-Cost Traffic Analysis Can Break Tor | Paper Breakdown #16 | CyberPodcast

Transcript

How Low-Cost Traffic Analysis Can Break Tor | Paper Breakdown #16 | CyberPodcast · CyberPodcast

• 0:00
  Welcome everyone, to this explainer.
• 0:02
  Today we are taking on a genuinely fascinating technical
• 0:05
  autopsy of one of the most famous privacy tools on the
• 0:08
  entire Internet, the Tor network.
• 0:10
  We're going to look right under the hood of a system design for
• 0:13
  absolute, unbreakable privacy. Using a groundbreaking research
• 0:17
  paper that somehow managed to safely dissect exactly what kind
• 0:21
  of traffic is actually flowing through these hidden channels,
• 0:23
  we are getting deeply technical. Today.
• 0:26
  We're breaking down network architecture protocol
• 0:28
  vulnerabilities and looking at some honestly brilliant
• 0:30
  detective work by network researchers.
• 0:32
  OK, let's dive into this. So here is our road map for this
• 0:36
  technical deep dive. One, how Tor works, 2.
• 0:40
  Deep packet inspection, 3. The BitTorrent anomaly, 4.
• 0:44
  The hijacking technique, 5. Tor as Sox proxy.
• 0:48
  And finally, six network geography.
• 0:51
  All right, Section 1, How Tor works.
• 0:54
  Let's establish that baseline architecture.
• 0:57
  As you probably know, the network relies on a concept
• 0:59
  called onion routing. When a client wants to
• 1:02
  communicate over Tor, they don't just connect directly to their
• 1:04
  destination, right? Instead, they select a circuit
• 1:07
  of intermediate nodes, typically three of them, an entry node, a
• 1:10
  middle node, and an exit node. Now, the absolute genius here is
• 1:15
  the encryption. Actually, think of it exactly
• 1:17
  like a set of nesting dolls. The original message is
• 1:20
  encrypted 3 times over in layers as the data passes through the
• 1:24
  circuit. Each node only holds the
• 1:26
  cryptographic key to peel back one single layer of encryption.
• 1:30
  Therefore, each intermediate node only knows the identity of
• 1:33
  the node immediately before it and the one right after it.
• 1:36
  It is only the final node in the chain, the exit node, that can
• 1:40
  strip away the very last layer of encryption to recover the
• 1:43
  original message and send it out to the open Internet.
• 1:46
  Moving on to Section 2, deep acket inspection, or how we
• 1:50
  actually analyze the payload, because that final exit node is
• 1:54
  a literal bridge to the open Internet.
• 1:56
  Well, it's the only place in the whole architecture where the
• 1:59
  actual traffic payload can be observed.
• 2:02
  To look at this, the researchers deployed a technique called Deep
• 2:04
  Acket Inspection, or DPI. Unlike your simple port based
• 2:08
  classification, which really just looks at the outer envelope
• 2:11
  of a data packet, DPI digs deep inside the packet.
• 2:15
  It analyzes both the header and the payload content to see what
• 2:18
  application is running. Now, to do this safely and most
• 2:22
  importantly, ethically, the researchers set up six of their
• 2:25
  very own global Tor exit nodes. As traffic left their nodes,
• 2:29
  they used a modified open source tool called Open DPI to
• 2:32
  categorize the data. Crucially, and this is a big
• 2:35
  deal for maintaining strict privacy standards, they
• 2:37
  explicitly anonymized and stripped away all original
• 2:40
  destination IP addresses on the fly before they extracted any
• 2:44
  payload data whatsoever. Which brings us to Section 3,
• 2:48
  the BitTorrent anomaly and the unmasking of unknown traffic.
• 2:53
  With this DPI engine running safely on their exit nodes for a
• 2:56
  23 day period, they collected massive amounts of data and
• 3:00
  almost immediately a major technical anomaly OED up.
• 3:03
  When we look at the traffic size they observed standard web
• 3:06
  browsing, Http://traffic took up the largest chunk at about
• 3:09
  36.4%. Unencorrected clear text
• 3:12
  BitTorrent traffic made-up almost 25% of the bandwidth,
• 3:16
  which by the way, is a massive load for a network design for
• 3:18
  low latency web browsing. But so the crucial point is that
• 3:22
  a massive 25.47% of the total network bandwidth was entirely
• 3:27
  unknown. The DPI engine literally
• 3:29
  couldn't identify it and get this despite taking up 1/4 of
• 3:33
  the network's entire data capacity.
• 3:35
  This unknown traffic was generated by less than 6% of the
• 3:38
  total network connections. But before we solve the mystery
• 3:41
  of that huge block of unknown traffic, let's take a really
• 3:44
  quick look at the Http://web traffic they actually could
• 3:47
  identify. Now you might assume this high
• 3:49
  latency, heavily encrypted network would be used primarily
• 3:52
  for simple text based communication, but the data
• 3:55
  proves otherwise. Applications make up 18% and
• 3:58
  flash and video make up 13.5%. This is a highly selfish,
• 4:02
  bandwidth heavy use of the network.
• 4:03
  It severely punishes users who actually need Tor for critical
• 4:06
  low latency safety. People are out there actively
• 4:09
  pushing rich media through a network that inherently
• 4:11
  penalizes latency by routing data through three separate
• 4:14
  global nodes. Let's get to the good stuff in
• 4:17
  Section 4, the hijacking technique.
• 4:20
  To figure out what was hiding in those unidentified packets, the
• 4:23
  researchers had to get creative. They strongly suspected the
• 4:26
  unknown traffic was encrypted BitTorrent, but they needed
• 4:28
  solid proof to get it. We quickly need to understand
• 4:31
  the standard BitTorrent protocol flow.
• 4:34
  When a client, let's call her Alice, wants to download a file,
• 4:37
  she first sends an announce message to a central server
• 4:40
  called a tracker. The tracker responds with a list
• 4:42
  of IP addresses of other peers who already have pieces of the
• 4:45
  file. Alice then reaches out to those
• 4:48
  peers directly, initiates A handshake, and begins
• 4:50
  downloading the actual data. Now, within that initial
• 4:54
  announce message Alice sends to the tracker, there is a critical
• 4:57
  piece of data called an info hash.
• 4:59
  This is a unique cryptographic identifier for the specific
• 5:03
  torrent file she wants. The structural vulnerability
• 5:06
  here is that historically this specific communication with the
• 5:09
  tracker and the info hash itself is sent completely unencrypted
• 5:13
  over the network, even if the subsequent peer-to-peer download
• 5:16
  is fully encrypted. The researchers exploited this
• 5:19
  exact unencrypted tracker communication.
• 5:22
  How? By using a classic man in the
• 5:24
  middle attack directly at their exit notes.
• 5:27
  And this brilliantly illustrates their technical creativity.
• 5:30
  When a Tor user asked a tracker for peers, the researchers
• 5:33
  intercepted the trackers unencrypted response.
• 5:36
  They rewrote the peer list on the fly, injecting the IP
• 5:39
  address of a machine they controlled right at the very
• 5:42
  top. The Tor user would then
• 5:43
  unwittingly initiate their handshake with the researchers
• 5:46
  controlled machine. By receiving that direct
• 5:49
  handshake, the researchers machine could instantly detect
• 5:51
  if the incoming connection was using BitTorrent protocol
• 5:54
  encryption. The results?
• 5:56
  They proved that a staggering five, 2.78% of BitTorrent
• 5:59
  handshakes were encrypted. That massive 25% block of
• 6:03
  unknown traffic was indeed encrypted.
• 6:05
  Peer-to-peer file sharing mystery solved.
• 6:07
  By taking those unencrypted info hashes and resolving them
• 6:10
  against public torrent databases, the researchers could
• 6:13
  see exactly what was driving this massive network load and
• 6:16
  look at this word cloud. The top 30 most popular
• 6:19
  downloads were heavily dominated by copyright protected media.
• 6:22
  Blockbuster films like Sherlock Holmes, massive video games like
• 6:25
  Mass Effect, popular television shows.
• 6:27
  The data irrefutably proved that a really significant portion of
• 6:31
  the Tor network's bandwidth is just being swallowed up by users
• 6:34
  anonymously sharing copyrighted entertainment.
• 6:37
  On to Section 5. Tor as SOCKS proxy or how users
• 6:41
  are completely bypassing the architecture.
• 6:44
  So heavy peer-to-peer file sharing isn't the only way the
• 6:47
  network architecture is being manipulated.
• 6:49
  As we discussed, Tors anonymity relies on a three hop circuit,
• 6:53
  but the researchers discovered that thousands of clients were
• 6:56
  intentionally bypassing this completely.
• 6:59
  Instead of a three hop circuit, users were creating a one hop
• 7:01
  tunnel directly to the exit node.
• 7:04
  By doing this, the user is trading their deep multi layer
• 7:07
  anonymity for a high speed encrypted single HO connection.
• 7:11
  Essentially, they aren't treating the Tor exit node as an
• 7:13
  anonymous router at all, but simply as a free encryted SOCS
• 7:16
  proxy to bypass cororate firewalls or hide their I with
• 7:20
  minimal latency. The researchers figured out a
• 7:23
  really clever mathematical way to diagnose this one hop abuse.
• 7:26
  In a normal 3 hop circuit, the exit node receives a specific
• 7:30
  control message called a relay under score begin cell.
• 7:33
  This cell instructs the exit node to connect to the final
• 7:36
  website, and it usually comes from the middle node in the
• 7:38
  circuit. Because the IP addresses of
• 7:40
  legitimate Tor middle nodes are public knowledge, the logic
• 7:43
  becomes super simple. If an exit node receives a relay
• 7:45
  under score begin control cell directly from an IP that is not
• 7:49
  on that public list. Well, it mathematically exposes
• 7:52
  that the user has bypassed the network and established a direct
• 7:55
  one hop tunnel. During their test, they found
• 7:57
  over 150 unique hosts continuously abusing their exit
• 8:01
  nodes in this exact manner. Finally, let's hit section 6,
• 8:05
  the geography of the network. Where are these clients actually
• 8:09
  coming from? When you think of Tor, you
• 8:11
  probably picture whistleblowers or political dissonance and
• 8:14
  heavily censored regions. But by analyzing the entry
• 8:16
  points, the data tells a very different story.
• 8:19
  The heaviest usage is actually concentrated in Western nations.
• 8:23
  Germany leads at nearly 15%, followed by the United States at
• 8:26
  nearly 13%. The researchers know that this
• 8:29
  high concentration in Western nations strongly correlates with
• 8:32
  countries that have aggressive anti piracy and strict copyright
• 8:35
  enforcement laws, which perfectly aligns with our
• 8:38
  earlier discovery regarding the massive volume of BIT torrent
• 8:41
  media piracy. Now, to help users in countries
• 8:43
  that actively block public Tor nodes, the network provides
• 8:46
  bridges, unlisted entry points. To prevent governments from
• 8:50
  finding and blocking these bridges, they're supposed to be
• 8:52
  handed out very sparingly. But the researchers found a
• 8:55
  glaring technical flaw in the distribution system.
• 8:58
  The server handing out bridge IPS lacked a simple capture to
• 9:01
  verify human users, and incredibly, it didn't even block
• 9:04
  requests coming from Tor exit nodes themselves.
• 9:06
  Because of this massive architectural oversight, the
• 9:09
  researchers were able to write an automated script that used
• 9:11
  the Tor network itself to rapidly harvest thousands of
• 9:14
  hidden bridge IP addresses. This proved that a determined
• 9:17
  adversary could easily map and block these secret network
• 9:20
  entrances. So this deep technical autopsy
• 9:23
  reveals a really complex ecosystem.
• 9:26
  A network built for ultimate privacy is simultaneously
• 9:29
  carrying massive loads of encrypted piracy, acting as a
• 9:32
  hijacked proxy service, and suffering from structural
• 9:35
  distribution flaws. It leaves us with a fascinating
• 9:38
  architectural paradox to ponder today.
• 9:40
  When you engineer a system that prioritizes absolute,
• 9:43
  unbreakable anonymity above all else, does that design
• 9:46
  inherently guarantee the technical abuse of the network?
• 9:49
  It is a deeply complex trade off between privacy, freedom, and
• 9:53
  systemic misuse. Thank you so much for joining me
• 9:56
  on this explainer and keep questioning the technology
• 9:58
  behind the curtain.