1. anonymous routing and mix nets (Tor) Yongdae Kim Significant fraction of these slides are borrowed from CS155 at Stanford 1

2. Anonymous web browsing  Why? 1. Discuss health issues or financial matters anonymously 2. Bypass Internet censorship in parts of the world 3. Conceal interaction with gambling sites 4. Law enforcement  Two goals: ▹ Hide user identity from target web site: (1), (4) ▹ Hide browsing pattern from employer or ISP: (2), (3)  Stronger goal: mutual anonymity (e.g. remailers) 2

3. Current state of the world I  ISPs tracking customer browsing habits: ▹ Sell information to advertisers ▹ Embed targeted ads in web pages (1.3%) »Example: MetroFi (free wireless) [Web Tripwires: Reis et al. 2008]  Several technologies used for tracking at ISP: ▹ NebuAd, Phorm, Front Porch ▹ Bring together advertisers, publishers, and ISPs »At ISP: inject targeted ads into non-SSL pages  Tracking technologies at enterprise networks: ▹ Vontu (symantec), Tablus (RSA), Vericept 3

4. Current state of the world II  EU directive 2006/24/EC: 3 year data retention ▹ For ALL traffic, requires EU ISPs to record: »Sufficient information to identify endpoints (both legal entities and natural persons) »Session duration … but not session contents ▹ Make available to law enforcement »… but penalties for transfer or other access to data  For info on US privacy on the net: ▹ “privacy on the line” by W. Diffie and S. Landau 4

5. Part 1: network-layer privacy Goals: Hide user’s IP address from target web site Hide browsing destinations from network

6. 1 st attempt: anonymizing proxy 6 HTTPS:// anonymizer.com ? URL=target User 1 User 2 User 3 anonymizer.com Web 1 Web 2 Web 3 SSL HTTP

7. Anonymizing proxy: security  Monitoring ONE link: eavesdropper gets nothing  Monitoring TWO links: ▹ Eavesdropper can do traffic analysis ▹ More difficult if lots of traffic through proxy  Trust: proxy is a single point of failure ▹ Can be corrupt or subpoenaed »Example: The Church of Scientology vs. anon.penet.fi  Protocol issues: ▹ Long-lived cookies make connections to site linkable 7

8. How proxy works  Proxy rewrites all links in response from web site ▹ Updated links point to anonymizer.com »Ensures all subsequent clicks are anonymized  Proxy rewrites/removes cookies and some HTTP headers  Proxy IP address: ▹ if a single address, could be blocked by site or ISP ▹ anonymizer.com consists of >20,000 addresses »Globally distributed, registered to multiple domains »Note: chinese firewall blocks ALL anonymizer.com addresses  Other issues: attacks (click fraud) through proxy 8

9. 2 nd Attempt: MIX nets Goal: no single point of failure

10. MIX nets [Chaum’81]  Every router has public/private key pair ▹ Sender knows all public keys  To send packet: ▹ Pick random route: R2  R3  R6  srvr ▹ Onion packet: 10 E pk 2 ( R 3, E pk 3 ( R 6, R3R3 R5R5 R4R4 R1R1 R2R2 R6R6 E pk 6 ( srvr, msg) msg srvr

11. Eavesdropper’s view at a single MIX  Eavesdropper observes incoming and outgoing traffic  Crypto prevents linking input/output pairs ▹ Assuming enough packets in incoming batch ▹ If variable length packets then must pad all to max len  Note: router is stateless 11 user 1 user 2 user 3 RiRi batch

12. Performance  Main benefit: ▹ Privacy as long as at least one honest router on path  Problems: ▹ High latency (lots of public key ops) »Inappropriate for interactive sessions »May be OK for email (e.g. Babel system) ▹ No forward security 12 R3R3 R2R2 R6R6 srvr

13. 3 rd Attempt: Tor MIX circuit-based method Goals:privacy as long as one honest router on path, and reasonable performance

14. The Tor design  Trusted directory contains list of Tor routers  User’s machine preemptively creates a circuit ▹ Used for many TCP streams ▹ New circuit is created once a minute 14 R1R1 R2R2 R3R3 R4R4 srvr 1 srvr 2 R5R5 R6R6 stream 1 stream 2 one minute later

15. Creating circuits 15 R1R1 R2R2 TLS encrypted Create C 1 D-H key exchange K1K1 K1K1 Relay C 1 Extend R 2 D-H key exchange K2K2 K2K2 Extend R 2

16. Once circuit is created  User has shared key with each router in circuit  Routers only know ID of successor and predecessor 16 R1R1 R2R2 R3R3 R4R4 K 1, K 2, K 3, K 4 K1K1 K2K2 K3K3 K4K4

17. Sending Data 17 R1R1 R2R2 Relay C 1 Begin site:80 Relay C 2 Begin site:80 TCP handshake Relay C 1 data HTTP GET Relay C 2 data HTTP GET HTTP GET K1K1 K2K2 resp Relay C 2 data resp Relay C 1 data resp

18. Complete View 18

19. Properties  Performance: ▹ Fast connection time: circuit is pre- established ▹ Traffic encrypted with AES: no pub-key on traffic  Tor crypto: ▹ provides end-to-end integrity for traffic ▹ Forward secrecy via TLS  Downside: ▹ Routers must maintain state per circuit ▹ Each router can link multiple streams via CircuitID »all steams in one minute interval share same CircuitID 19

20. Privoxy  Tor only provides network level privacy ▹ No application-level privacy »e.g. mail progs add “From: email-addr” to outgoing mail  Privoxy: ▹ Web proxy for browser-level privacy ▹ Removes/modifies cookies ▹ Other web page filtering 20

21. Anonymity attacks: watermarking  Goal: R 1 and R 3 want to test if user is communicating with server  Basic idea: ▹ R 1 and R 3 share sequence:  1,  2, …,  n  {-10,…,10} ▹ R 1 : introduce inter-packet delay to packets leaving R 1 and bound for R 2. Packet i delayed by  i (ms) ▹ Detect signal at R 3 21 R1R1 R2R2 R3R3

22. Anonymity attacks: congestion  Main idea: R 8 can send Tor traffic to R 1 and measure load on R 1  Exploit: malicious server wants to identify user ▹ Server sends burst of packets to user every 10 seconds ▹ R 8 identifies when bursts are received at R 1 Follow packets from R 1 to discover user’s ID 22 R1R1 R2R2 R3R3 R8R8