Latest / CyberPodcast / How Tor Hidden Services Can Be Abused by Botnets | Paper Breakdown #6 | CyberPodcast

Transcript

How Tor Hidden Services Can Be Abused by Botnets | Paper Breakdown #6 | CyberPodcast · CyberPodcast

• 0:00
  All right, let's jump right in. Welcome to this explainer.
• 0:03
  Today we are unpacking an absolute cyber medical mystery.
• 0:06
  We're going back to the day the Tor unanimity network literally
• 0:09
  almost buckled under its own weight, and we're going to teach
• 0:12
  you the core technical concepts behind exactly what happened.
• 0:16
  We'll look at how a massive botnet choked the entire system.
• 0:19
  And more importantly, we're going to explore the ingenious
• 0:22
  engineering proposals designed to fix it.
• 0:25
  OK, let's dive into this. Here is our road map.
• 0:28
  First, we're going to diagnose the technical illness with a
• 0:31
  2013 Tor crisis and then understand Tor hidden services.
• 0:36
  After we find the botnet bottleneck, we'll systematically
• 0:39
  evaluate some treatments covering throttling bots and
• 0:42
  guard nodes, reusing circuits and traffic isolation before
• 0:45
  finally looking at the future of Tor security.
• 0:48
  So Section 1, the 2013 Tor crisis.
• 0:52
  Picture this It's August 2013, and in a span of just three
• 0:56
  weeks, the estimated daily Tor clients absolutely skyrocketed.
• 1:00
  We're talking a jump from under 1,000,000 to nearly 6,000,000
• 1:03
  users. Now this massive sudden spike
• 1:06
  wasn't some organic surge in privacy awareness, you know, it
• 1:09
  was an infection. The culprit was a massive click
• 1:12
  fraud botnet known as Movado or Cefnet.
• 1:15
  And here's the real kicker. This botnet was actually running
• 1:18
  its command and control server directly as a Tor hidden
• 1:21
  service. That's essentially the master
• 1:24
  brain issuing instructions, and it forced every single infected
• 1:27
  machine across the globe to communicate with it entirely
• 1:29
  through the Tor network. Now, the tangible symptom of
• 1:32
  this user spike was immediate, and it was painful for regular
• 1:35
  users. The whole network slowed to an
• 1:38
  absolute crawl. To put that into perspective,
• 1:40
  the time it took to download a tiny 50 KB file literally
• 1:44
  doubled, jumping from 1.5 seconds to 3.0 seconds.
• 1:48
  But here is the real mystery, and it's a weird one.
• 1:51
  The actual amount of data traffic moving through the
• 1:53
  network didn't change all that dramatically.
• 1:55
  The pipe wasn't full of data, So what was causing the blockage?
• 1:59
  That brings us to Section 2, understanding Tor hidden
• 2:03
  services. To understand the disease, we
• 2:06
  first have to learn the baseline mechanics of how Tor hidden
• 2:09
  services actually operate. Let's breakdown this complex 5
• 2:12
  step process. First, the service picks its
• 2:15
  identity and secret keys. Second, it builds encrypted
• 2:18
  circuits to at least three Tor relays, asking them to act as
• 2:22
  introduction points. Third, it publishes A
• 2:25
  descriptor. You can think of this like a
• 2:26
  listing in a decentralized anonymous phone book. 4th, a
• 2:30
  client retrieves this listing and builds 2 circuits, 1 to an
• 2:33
  introduction point and one to a randomly selected rendezvous
• 2:36
  point. Finally, Step 5, the hidden
• 2:39
  service builds its own circuit back to that rendezvous point.
• 2:42
  It's like 2 spies agreeing to meet at a secure, neutral
• 2:44
  location to exchange information.
• 2:47
  The crazy part? Connecting just one single bot
• 2:49
  to its master requires building up to six separate circuits.
• 2:52
  Moving right along to Section 3, the botnet bottleneck.
• 2:56
  So let's define the core issue here, which is something called
• 2:59
  an onion skin. To build an anonymous encrypted
• 3:02
  tunnel, Tor uses a key exchange protocol called Diffie Hellman.
• 3:06
  In simple terms, this is a complex mathematical handshake.
• 3:09
  It allows 2 strangers to agree on a secret lock and key without
• 3:13
  anyone who's listening in being able to figure it out.
• 3:16
  The very first message in this exchange is the onion skin.
• 3:19
  Decryting these cryptographic layers requires heavy, complex
• 3:22
  math. Now multiply that math by
• 3:24
  millions of bots trying to build 6 circuits each.
• 3:27
  The result? The Tor relays became completely
• 3:29
  CPU bound, their processors just flat out maxed out.
• 3:33
  And this brilliantly illustrates the catastrophic chain reaction
• 3:37
  that followed. Because the relays were CPU
• 3:39
  bound, incoming onion skins were forced to just sit and wait in
• 3:43
  decryption queues. Because they waited too long,
• 3:45
  they naturally began to time out.
• 3:47
  Because they timed out, the entire partial circuit failed.
• 3:50
  And because the circuit failed, what did the bot do?
• 3:52
  It just tried again, injecting more onion skins into the
• 3:55
  network. It was a vicious, cascading
• 3:57
  feedback loop. The system was literally choking
• 4:00
  on its own failed connections. All right, Section 4, throttling
• 4:04
  bots and guard nodes. Put your network engineer hat on
• 4:08
  for a second. Your first instinct might be to
• 4:10
  just charge a cost to build a circuit, right?
• 4:12
  Just stop the spam. But simple cost throttling tools
• 4:15
  fail spectacularly in an anonymous network.
• 4:18
  If you use proof of work. You know, making computers solve
• 4:20
  math puzzles. Bot Nets actually have massive
• 4:22
  CPU resources, often way more than regular users.
• 4:27
  A clever bot master can even use a chain proving attack where a
• 4:30
  previous bot solves the puzzle for the next bot connecting.
• 4:33
  OK, well what about captchas to prove you're human?
• 4:36
  Well, bot Nets are headless, meaning they have no screen or
• 4:38
  user interface. But actually, so are many
• 4:41
  legitimate Tor servers. You can't serve a visual capture
• 4:43
  to a headless server. Plus in an anonymous network you
• 4:46
  can't easily blacklist an IP address.
• 4:48
  Bots can just guess captchas endlessly until they eventually
• 4:51
  slip through. So if we can't throttle the
• 4:53
  cost, what if we throttle at the entry guard node?
• 4:57
  This is a really interesting concept called assigned guards.
• 5:01
  Basically, regular clients get a low rate limit.
• 5:03
  Say they're only allowed to build 30 circuits per hour.
• 5:06
  If a legitimate server needs to be busier, they can use a
• 5:09
  verifiable cryptographic token to get a higher limit without
• 5:12
  flooding the network. Notice the simulation data we're
• 5:15
  looking at here. That throttle traffic line in
• 5:18
  blue successfully isolates the network from that chaotic red
• 5:21
  spike of botnet traffic. The network stays completely
• 5:24
  stable. Which brings us to Section 5,
• 5:28
  reusing circuits and traffic isolation.
• 5:31
  Take a look at the green light on this graph.
• 5:33
  Remember how we set a single timeout cause Tor to destroy an
• 5:36
  entire multi hop circuit and start completely over?
• 5:39
  Well, researchers asked, what if we just reuse the failed partial
• 5:42
  circuit if a connection times out on the third hop?
• 5:45
  Actually, scratch that, just truncate it back to the second
• 5:47
  hop and try extending again from there.
• 5:50
  The math shows that this incredibly simple reuse could
• 5:53
  reduce the overall cryptographic load on the network by a massive
• 5:56
  10 to 30%. Another proposed treatment is
• 5:59
  hidden service isolation. This introduces a brand new type
• 6:03
  of data packet called a No HS Create Cell, which stands for No
• 6:06
  Hidden Service. The goal here is pure
• 6:09
  segregation. If a circuit is built with this
• 6:11
  new cell, the relays will silently drop any hidden service
• 6:14
  traffic trying to use it. This guarantees that normal
• 6:16
  everyday users checking their e-mail aren't competing for Hwy.
• 6:19
  space with millions of bots all trying to reach a hidden
• 6:21
  service. Finally, Section 6, the future
• 6:25
  of Tor security. So here's a thought to leave you
• 6:29
  with. The absolutely crucial take away
• 6:31
  is that in cybersecurity, there is literally always a trade off.
• 6:35
  It's a fascinating dilemma, really.
• 6:37
  If we isolate hidden service traffic to protect the network's
• 6:40
  overall speed, we inadvertently make that specific traffic stand
• 6:43
  out. Will these mitigation tools
• 6:45
  eventually erode anonymity for the legitimate users relying on
• 6:48
  them? Think about it.
• 6:49
  If relays can clearly distinguish hidden service
• 6:52
  traffic from normal traffic, it logically shrinks the crowd that
• 6:54
  those users can hide within. It's a delegate ongoing balance
• 6:58
  between keeping the system alive and preserving absolute
• 7:01
  anonymity. Which side of the scale carries
• 7:03
  more weight? That is the ever evolving puzzle
• 7:05
  of network security. Thank you so much for joining me
• 7:08
  to explore the mechanics behind this incredible event.
• 7:11
  Until next time, keep learning.