Latest / CyberPodcast / Unmasking the Tor Network with an HTTP Attack | Paper Breakdown #4 | CyberPodcast

Transcript

Unmasking the Tor Network with an HTTP Attack | Paper Breakdown #4 | CyberPodcast · CyberPodcast

• 0:00
  Imagine locking yourself inside a windowless, completely
• 0:03
  impenetrable vault. You feel entirely hidden from
• 0:05
  the outside world for years. You know that's exactly how we
• 0:09
  viewed our digital privacy when using advanced anonymizing tools
• 0:12
  on the Internet. But what if the very
• 0:13
  architecture of that vault could somehow be turned against you?
• 0:16
  Welcome to this explainer. Today we're breaking down a
• 0:19
  truly fascinating 2011 research paper by Xiao Gang, Wang Jinjo,
• 0:24
  Loo, Ming Yang, and Jin Ling. They outlined an ingenious,
• 0:27
  totally theoretical, but incredibly potent application
• 0:30
  level attack against the internet's most famous privacy
• 0:32
  shield, the Tor network. This brings us to the core
• 0:36
  question that this entire research paper fundamentally
• 0:39
  challenges. Is true online anonymity
• 0:41
  actually possible now? For a long time, the security
• 0:44
  community thought that if you just encrypted the data well
• 0:46
  enough, the answer was yes, absolutely.
• 0:49
  But these researchers, they discover that you don't always
• 0:51
  need to break the lock to see what's inside the vault.
• 0:53
  Sometimes you just need to watch the footrints leading U to the
• 0:56
  door. We're going to look at exactly
• 0:58
  how this brilliant digital investigative work is done.
• 1:01
  OK, let's dive into this. Here's our blueprint for the
• 1:04
  explainer today. We'll start with the promise of
• 1:06
  privacy, move into how Tor hides traffic, uncover the invisible
• 1:10
  tracker attack, map out how this digital heist is executed, weigh
• 1:14
  accuracy versus stealth, and finally, look at defending the
• 1:18
  Tor network. Part 1, The promise of privacy.
• 1:21
  The impenetrable fortress. To really appreciate the
• 1:25
  brilliance of this attack, we first have to understand the
• 1:27
  sheer strength of the fortress we're dealing with.
• 1:30
  The Tor network is widely considered the absolute gold
• 1:33
  standard for Internet privacy. As the paper defines it, it's a
• 1:37
  circuit based, low latency, anonymous communication system.
• 1:41
  But what does that mean in plain English?
• 1:42
  Basically, it's a network built to anonymize your web traffic in
• 1:45
  real time. It allows you to browse the web,
• 1:48
  send instant messages, use location services, all without
• 1:51
  anyone knowing who you are or where you're connecting from.
• 1:55
  Conventional encryption by itself just isn't enough to hide
• 1:57
  your identity, right? But Tor uses incredibly strong
• 2:00
  layered cryptography distributed across a massive volunteer
• 2:03
  network to completely mask your digital footprint.
• 2:06
  Moving right along to Part 2, how Tor hides traffic mapping
• 2:10
  the terrain. Now, what's really interesting
• 2:12
  about this architecture is the cast of characters involved.
• 2:16
  On one end we have Alice. That's the client.
• 2:18
  Basically you sitting at your computer running a local onion
• 2:21
  proxy. On the far opposite end is Bob.
• 2:24
  He's the server Alice wants to talk to, like a website.
• 2:27
  Now, to get from Alice to Bob, the data doesn't just travel in
• 2:29
  a straight predictable line. Instead, it bounces through
• 2:32
  exactly 3 specialized Roxy servers called Onion routers, an
• 2:36
  entry router, a middle router, and an exit router.
• 2:39
  The way this circuit is built is actually where the Onion and Tor
• 2:42
  gets its name. First, Alice's software picks 3
• 2:45
  routers from a massive directory.
• 2:47
  Then she establishes a secure connection to the entry router.
• 2:51
  And here's the incredibly clever part.
• 2:53
  She then extends that circuit. Layer by layer.
• 2:55
  She negotiates secret encryption keys with the entry router.
• 2:58
  Then through that, she negotiates keys with the middle
• 3:01
  router, and finally through that, negotiates keys with the
• 3:04
  exit router. So the data is wrapped in
• 3:06
  multiple thick layers of encryption.
• 3:08
  Think about it. The entry router knows exactly
• 3:11
  who Alice is, but has no clue what she's saying or who she's
• 3:14
  talking to. The exit router knows who Bob is
• 3:16
  and what the message says, but has absolutely no idea that
• 3:19
  Alice sent it. And the middle router?
• 3:21
  It just blindly shuffles encrypted data back and forth.
• 3:24
  It's an elegant, highly compartmentalized system.
• 3:27
  Which brings us to Part 3, the invisible tracker attack and
• 3:31
  inside job. So the narrative shifts
• 3:35
  dramatically right here. What the researchers realized is
• 3:37
  that you literally do not need to break the encryption from the
• 3:40
  outside if you can just get malicious actors on the inside.
• 3:44
  Listen to this quote directly from the study.
• 3:46
  The anonymity of Tor clients is threatened by potential HTTP
• 3:49
  based application level attack against Tor, which exploits both
• 3:53
  Tord's design features and HTT. PS vulnerability to man in the
• 3:56
  middle attacks. Why is this so utterly
• 3:58
  terrifying? Well, previously to Unmaskator,
• 4:00
  user attackers had to trick them into downloading invasive clunky
• 4:03
  plug insurance like Java or Flash just to leak their true IP
• 4:06
  address. This attack, it doesn't need any
• 4:08
  of that zero plug insurance. Even if you completely disable
• 4:11
  all active content in your browser, this attack is still
• 4:13
  going to find you. It exploits the fundamental way
• 4:15
  the web itself functions. So what is the fundamental
• 4:19
  prerequisite for this heist? The adversary has to control
• 4:22
  both the entry router and the exit router for a user specific
• 4:26
  circuit. Now remember how Tor is run by
• 4:29
  everyday volunteers? Literally anyone can set up a
• 4:32
  router. If an attacker sets up a few
• 4:34
  high bandwidth, highly reliable servers, the Tor network
• 4:37
  naturally prefers them because they're fast.
• 4:39
  The attacker doesn't have to compromise the entire network,
• 4:42
  they just need Alice to randomly pick their malicious router as
• 4:45
  her entry point and their other malicious router as her exit
• 4:48
  point. Once they hold both ends of the
• 4:50
  tunnel, the trap is completely set.
• 4:52
  Part 4 Executing the digital heist, The invisible tripwire.
• 4:57
  And this brilliantly illustrates the exact mechanic of Scheme 1.
• 5:01
  Alice thinks she is talking securely to Bob, the server.
• 5:04
  She requests a web page. Her request travels through the
• 5:06
  network and hits the malicious exit router.
• 5:09
  Because the exit router is the one that actually fetches the
• 5:11
  raw data from Bob, it gets to see the unencrypted request.
• 5:14
  But instead of just passing Bob's real web page back to
• 5:17
  Alice like it's supposed to, the malicious exit router intercepts
• 5:20
  it. It acts as a man in the middle
• 5:22
  and injects a completely forged web page straight back down the
• 5:24
  tunnel to Alice. At the code level, it's devious.
• 5:28
  The malicious exit router injects a series of links for
• 5:31
  tiny 35 byte empty GIF files into the HTML.
• 5:35
  But the real kicker is right at the end of those lines, style
• 5:39
  equals display none. This tiny snippet of code forces
• 5:43
  Alice's web browser to fetch these images, but strictly tells
• 5:46
  it not to display them on the screen.
• 5:49
  Alice is completely oblivious. She's just staring at a blank
• 5:52
  page or a loading screen, but in the background, her browser is
• 5:55
  secretly and frantically requesting these invisible
• 5:58
  images. So why go through all this
• 6:01
  trouble to force Alice's browser to fetch invisible images?
• 6:04
  Well, it all comes down to math. Specifically, this formula.
• 6:08
  Let's say the attacker injected M number of invisible images.
• 6:12
  To fetch those, Alice's computer has to send specific data
• 6:15
  packets, called relay cells, back through the network.
• 6:18
  It's exactly 2 cells forward and three cells backward for every
• 6:21
  single image. This creates a highly specific,
• 6:24
  totally unnatural burst of traffic.
• 6:26
  Exactly 5 * M relay cells. Now remember our malicious entry
• 6:30
  router? It can't read Alice's encrypted
• 6:32
  data, but it can count the cells passing through.
• 6:35
  It's just sitting there, watching the traffic flow.
• 6:37
  Suddenly, boom, it sees that exact 5 M burst of traffic
• 6:41
  coming directly from Alice's IP address in mere milliseconds.
• 6:44
  The entry router syncs its timestamps with the exit router.
• 6:47
  The trap Snap shut. The entry router knows exactly
• 6:49
  who Alice is, and the exit router knows exactly what
• 6:52
  website she is visiting. Her anonymity is completely
• 6:55
  destroyed. Part 5 Accuracy versus stealth
• 6:59
  choosing the weapon. So that's one way to pull it
• 7:02
  off. But attackers actually have
• 7:04
  options. The researchers laid out two
• 7:06
  distinct approaches here. Scheme one, which we just walked
• 7:09
  through, is the forged web page injection.
• 7:11
  It completely hijacks the session, injects a totally fake
• 7:14
  page with those invisible images, forces that traffic
• 7:17
  spike, and then seamlessly redirects Alice back to her real
• 7:20
  page. It is incredibly aggressive,
• 7:22
  fast, and remarkably accurate. But Alice might notice a slight
• 7:25
  split second delay. Then you have Scheme 2 target
• 7:29
  web page modification. Instead of using a totally fake
• 7:32
  page, the exit router grabs the real web page Alice asked for,
• 7:36
  stealthily slips the invisible image tags right into the
• 7:39
  existing code, and sends it to her.
• 7:40
  This is much stealthier. Alice sees exactly what she
• 7:43
  expected to see with 0 redirect delay.
• 7:46
  However, it's inherently less accurate.
• 7:48
  Why? Because the real web page has
• 7:50
  its own text, its own real images and links, which creates
• 7:53
  a bunch of traffic noise that can actually obscure that
• 7:55
  perfect five M burst of relay cells.
• 7:57
  So the crucial point is how these two strategies stack up in
• 8:01
  the real world. The researchers actually built a
• 8:03
  prototype and tested it on the live Tor network.
• 8:07
  For Scheme 1, the detection delay was less than 4 seconds
• 8:10
  and the correlation accuracy was near 100%.
• 8:13
  It honestly didn't matter what the rest of the Internet was
• 8:15
  doing, the signal was loud and clear.
• 8:18
  For Scheme 2, the delay jumped to over 5 seconds, and more
• 8:21
  importantly, its effectiveness decreased the more complex the
• 8:24
  original web page was. If a page had lots of existing
• 8:27
  links, that invisible tracker just got lost in the noise.
• 8:30
  Therefore, an attacker looking for an absolute sure thing is
• 8:33
  almost definitely going to accept the slight delay of
• 8:35
  scheme one to guarantee they unmask their target.
• 8:38
  Finally, Part 6. Defending the Tor network
• 8:42
  securing the perimeter. Now, before we conclude that
• 8:45
  privacy is completely dead and buried, the researchers did
• 8:48
  provide the blueprint for how to secure the perimeter against
• 8:51
  this exact type of digital heist.
• 8:53
  Here are the countermeasures they proposed.
• 8:55
  First, simply deploy more normal, trusted routers.
• 8:58
  If there are thousands of good routers out there, the
• 9:00
  statistical odds of Alice accidentally picking an
• 9:02
  attacker's malicious entry and exit router drops significantly.
• 9:05
  Second, implements strict authentication.
• 9:07
  Don't just let anyone with a fast server automatically become
• 9:10
  a highly trusted node. Third, build browser plug in
• 9:13
  specifically designed to monitor for these unnatural traffic
• 9:16
  bursts. You know, like a browser
• 9:17
  suddenly requesting dozens of tiny invisible files all at the
• 9:20
  same time. But the absolute, undisputed,
• 9:22
  most critical defense here is the last one.
• 9:24
  Utilize HTTPS. You see, this entire attack
• 9:27
  relies heavily on the exit router being able to read and
• 9:30
  alter raw unencrypted HTTPHTML code.
• 9:33
  But if you use HTTPS, the traffic between you and the
• 9:36
  final web server is encrypted end to end.
• 9:39
  Sure, the Tor network still shuffles the data around, but
• 9:41
  the exit router just sees an unreadable block of ciphertext.
• 9:44
  It literally cannot inject those invisible image tags because it
• 9:47
  can't interact with the web pages code in the 1st place.
• 9:49
  And that brings us to our final thought.
• 9:51
  The massive, widespread adoption of HTTPS across the web today
• 9:55
  has largely blunted the specific, incredibly brilliant
• 9:58
  HTTP attack. But it leaves us with a massive,
• 10:01
  provocative question. As long as the Internet relies
• 10:04
  on incredibly complex, interconnected layers, will the
• 10:07
  cat and mouse game of digital privacy ever truly end?
• 10:11
  Every single time a flawless fortress is built, someone
• 10:14
  somewhere is figuring out how to analyze the footprints right
• 10:17
  outside the door. Thank you so much for joining me
• 10:19
  for this explainer. Keep your eyes open, stay
• 10:22
  curious, and keep learning.