Latest / CyberPodcast / Cracking the Tor Vault with a Congestion Attack | Paper Breakdown #5 | CyberPodcast

Transcript

Cracking the Tor Vault with a Congestion Attack | Paper Breakdown #5 | CyberPodcast · CyberPodcast

• 0:00
  Welcome to The Explainer. Today, we're cracking open a
• 0:03
  genuinely fascinating academic paper that details a practical
• 0:07
  and frankly devastating attack on the Tor anonymity network.
• 0:10
  You know, for years Tor has been considered the absolute gold
• 0:13
  standard for online privacy. It's this impenetrable digital
• 0:17
  vault shielding activists, journalists, and everyday users
• 0:20
  from surveillance. But what actually happens when
• 0:23
  researchers figure out a way to crack that vault wide open?
• 0:26
  Let's find out. So let's just dive right into
• 0:28
  the big question here. How on earth could someone break
• 0:31
  into a highly secure digital vault without using massive
• 0:34
  computing power? I mean, usually attacking a
• 0:36
  global network requires immense resources, right?
• 0:39
  We're talking nation state level server farms or massive
• 0:42
  sprawling bot Nets. But the researchers in this
• 0:45
  paper figured out a way to expose an anonymous Tor user
• 0:48
  using incredibly low bandwidth. They didn't kick the front door
• 0:51
  down. Instead, they found a clever way
• 0:53
  to use the building's own architecture against itself.
• 0:55
  Section 1 The Vault The promise of Tor.
• 0:59
  To pull off this digital heist, we obviously need the floor plan
• 1:02
  first. Now Tor protects us using
• 1:04
  something called onion routing. Think of it like this.
• 1:07
  Your data is a letter, and before you send it, you lock it
• 1:10
  in three nested lock boxes. The system routes your data
• 1:13
  through a chain of volunteer relays, which is usually 3, and
• 1:16
  each relay only peels off one layer of encryption.
• 1:19
  That means no single server knows both who you are and what
• 1:22
  website you're visiting. The first note knows your IP
• 1:25
  address, sure, but not your destination.
• 1:27
  The last note knows your destination, but absolutely not
• 1:30
  who you are. It is a beautifully elegant
• 1:32
  system. Moving on to Section 2.
• 1:35
  The failed breaches. Why old attacks failed.
• 1:39
  So back in 2005, researchers Murdoch and Danes actually did
• 1:43
  successfully attack Tor. But just look at the staggering
• 1:46
  difference between the network back then and what it is today.
• 1:49
  In 2005, attackers only had to sift through 13 lightly loaded
• 1:52
  relays to find a specific user's path.
• 1:55
  It was literally like finding a needle in a thimble.
• 1:57
  But the modern network, studied in this paper, it had over 1500
• 2:00
  heavily loaded relays. If you try to blindly flood that
• 2:03
  many modern servers to track someone, you're going to need a
• 2:05
  tremendous amount of bandwidth. Plus, the natural background
• 2:08
  noise of thousands of other users creates way, way too many
• 2:12
  false positives. Basically, brute force simply
• 2:14
  doesn't scale anymore. Which brings us to Section 3,
• 2:17
  the blueprint, A new vulnerability discovered.
• 2:21
  Since brute force is off the table, the researchers had to
• 2:24
  find a structural flaw. And they found a blueprint for
• 2:27
  an attack that relies on three specific baked in features of
• 2:30
  Tor. First up, Tor routers don't
• 2:32
  introduce artificial delays. They process data as fast as
• 2:35
  humanly possible, which makes the latency highly predictable.
• 2:39
  Second, the IP addresses of the routers are publicly known in a
• 2:42
  directory. And third, well, this is the
• 2:45
  fatal flaw. At the time of this research,
• 2:47
  Tor allowed entirely unrestricted path links.
• 2:50
  Users usually choose a chain of three relays, but the protocol
• 2:53
  didn't actually stop someone from building a circuit that was
• 2:55
  20 or 30 hops long, looping back on itself over and over again.
• 3:00
  Section 4. Gathering the crew setting up
• 3:03
  the attack. So how do they do it?
• 3:06
  Well, the attack builds across 3 distinct phases.
• 3:09
  Stage 1. Injecting an insider into the
• 3:12
  victim's browser, Stage 2, establishing A baseline to know
• 3:16
  exactly what normal looks like, and stage 3, Hitting the
• 3:20
  suspected target with a highly concentrated traffic jam.
• 3:23
  Let's walk through exactly how they pull this off in the real
• 3:25
  world. What makes this initial
• 3:27
  compromise so incredibly elegant is just how simple it is.
• 3:31
  The attacker doesn't even need to hack into your computer.
• 3:34
  No, they simply operate a Tor exit node, which is the final
• 3:37
  server in the Tor chain that actually connects to the
• 3:40
  regular, everyday Internet. They just sit there waiting for
• 3:43
  a victim's traffic to pass through.
• 3:45
  When it does, they secretly inject a tiny, seemingly
• 3:48
  harmless piece of JavaScript code into an unencrypted HTML
• 3:52
  response that's heading back to the victim's browser.
• 3:54
  And because most Tor users don't disable JavaScript because doing
• 3:58
  so would break almost all of the modern web, the victim's browser
• 4:01
  happily just executes it. So our insider here is that
• 4:04
  injected JavaScript, and it acts exactly like a ticking clock.
• 4:08
  It's programmed to send a tiny HTTP request, basically pinging
• 4:12
  the attacker's server every single second.
• 4:15
  The victim's browser unwittingly starts pulsing out these signals
• 4:18
  right back to the attacker. This allows them to track the
• 4:21
  baseline network latency. They just sit back and measure
• 4:24
  ping one second, ping one second.
• 4:28
  They now know exactly what the victim's connection looks like
• 4:30
  when it's perfectly calm and completely undisturbed.
• 4:34
  All right, Section 5, the heist executing the congestion attack.
• 4:39
  Now this is where the real magic happens.
• 4:41
  By exploiting that unrestricted path length flaw we talked about
• 4:44
  a minute ago, the attacker builds a massive looping circuit
• 4:47
  through the Tor network. Instead of the standard 3 hops,
• 4:50
  they might build a path 24 hops long, and they intentionally
• 4:54
  route the circuit back through one specific suspected target
• 4:57
  relay. Let's say it's the guard node,
• 4:59
  the very first entry point the victim might be using.
• 5:01
  If the attacker pushes just one MB of data into this looping
• 5:04
  circuit, the target router ends up processing it over and over
• 5:08
  and over again. This artificially amplifies the
• 5:11
  attacker's low bandwidth into a completely crushing load on that
• 5:14
  specific server. And understand why this actually
• 5:16
  breaks the anonymity. We really need to look at Tors
• 5:19
  internal mechanics. Tor packages data into fixed
• 5:22
  size 512 byte cells and a process them using a round Robin
• 5:26
  queue. Think of it kind of like a
• 5:27
  dealer passing out cards to players at a poker table.
• 5:30
  Under normal conditions, the router handles traffic smoothly.
• 5:33
  It goes from QA to QB to QC. The victim's one second ping
• 5:37
  slip right through without any delay whatsoever.
• 5:40
  But watch what happens when the attacker's looping roundabout of
• 5:42
  data hits that specific router. The whole system just completely
• 5:46
  chokes. Look at how violently the queues
• 5:48
  stack up here. Suddenly, the dealer has to pass
• 5:51
  cards to 20 dummy players controlled by the attacker
• 5:53
  before it can even deal a single card to the victim.
• 5:55
  If the attacker guessed the right router, the victims
• 5:58
  perfectly timed one second pings are suddenly trapped in this
• 6:00
  completely manufactured traffic jam.
• 6:02
  Section 6. The getaway, The unsolved
• 6:05
  anonymity problem. The heist is done.
• 6:07
  But how do they know they actually got the right guy?
• 6:10
  Well, they run a statistical analysis on the arrival times of
• 6:13
  those JavaScript pings. Remember that baseline we
• 6:16
  established earlier? If the attacker attacks a random
• 6:19
  router that the victim isn't using, the pings just keep
• 6:21
  arriving exactly one second apart.
• 6:23
  The histogram data looks completely vertical, but if they
• 6:26
  attack the exact router the victim is using, the ping
• 6:29
  suddenly arrive late, scattered totally randomly by the traffic
• 6:32
  jam. The latency histogram flattens
• 6:34
  out horizontally with extremely high delays.
• 6:37
  The researchers actually used a modified chi square test to
• 6:40
  mathematically prove this shift. Basically, when the graph goes
• 6:43
  horizontal, bingo, the attacker has successfully unmasked the
• 6:47
  victim's entry node. And honestly, this illustrates A
• 6:50
  genuinely shocking conclusion from the researchers.
• 6:52
  Just read this quote. Longer circuits do not guarantee
• 6:55
  stronger anonymity. I mean, that completely subverts
• 6:58
  what retty much every privacy advocate assumed was just common
• 7:01
  sense, right? We intuitively think more layers
• 7:04
  of encryption, more hops naturally equals more safety.
• 7:08
  But as this paper absolutely proves, if the protocol allows
• 7:11
  attackers to manipulate those long paths to amplify their own
• 7:14
  bandwidth, extra length actually becomes a deadly weapon against
• 7:18
  the network itself. Now, as a direct result of this
• 7:20
  paper, the Tor developers did patch the infinite loop flaw.
• 7:24
  They hard cap paths to a maximum of eight hops.
• 7:27
  But here is the truly chilling part.
• 7:30
  The authors note that a complete defense against this kind of
• 7:33
  active congestion based timing attack, it remains completely
• 7:37
  undiscovered. As long as networks drive for
• 7:40
  low latency, an attacker who can measure time can potentially
• 7:43
  track you. It really leaves us all to
• 7:45
  wonder in this constant high stakes cat and mouse game of
• 7:49
  cybersecurity, just how invincible is our online
• 7:52
  anonymity? Truly, thanks so much for
• 7:54
  joining me on this deep dive. And remember, always keep
• 7:56
  questioning the systems that protect you.