Users Get Routed: Traffic Correlation on Tor by Realistic Adversaries

Aaron Johnson1 Chris Wacek2 Rob Jansen1 Micah Sherr2 Paul Syverson1
1 U.S.

Naval Research Laboratory, Washington, DC 2 Georgetown University, Washington, DC

MPI-SWS July 29, 2013

1

Summary: What is Tor?

Summary: What is Tor?

Tor is a system for anonymous communication.

3

Summary: What is Tor?

Tor is a system for anonymous communication. ^ popular

Over 500000 daily users and 2.4GiB/s aggregate 4

Summary: Who uses Tor?

Summary: Who uses Tor?

Individuals avoiding censorship Individuals avoiding surveillance Journalists protecting themselves or sources Law enforcement during investigations Intelligence analysts for gathering data

Summary: Tors Big Problem

Summary: Tors Big Problem

Summary: Tors Big Problem

Summary: Tors Big Problem

10

Summary: Tors Big Problem

Traffic Correlation Attack

11

Summary: Tors Big Problem

Traffic Correlation Attack

Congestion attacks Throughput attacks Latency leaks Website fingerprinting Application-layer leaks Denial-of-Service attacks
12

Summary: Our Contributions

13

Summary: Our Contributions

1. Empirical analysis of traffic correlation threat 2. Develop adversary framework and security metrics 3. Develop analysis methodology and tools
14

Overview
Summary Tor Background Tor Security Analysis
o Adversary Framework o Security Metrics o Evaluation Methodology o Node Adversary Analysis o Link Adversary Analysis

Future Work
15

Overview
Summary Tor Background Tor Security Analysis
o Adversary Framework o Security Metrics o Evaluation Methodology o Node Adversary Analysis o Link Adversary Analysis

Future Work
16

Background: Onion Routing

Users

Onion Routers

Destinations

17

Background: Onion Routing

Users

Onion Routers

Destinations

18

Background: Onion Routing

Users

Onion Routers

Destinations

19

Background: Onion Routing

Users

Onion Routers

Destinations

20

Background: Onion Routing

Users

Onion Routers

Destinations

21

Background: Using Circuits

22

Background: Using Circuits

1. Clients begin all circuits with a selected guard.

23

Background: Using Circuits

1. Clients begin all circuits with a selected guard. 2. Relays define individual exit policies.

24

Background: Using Circuits

1. Clients begin all circuits with a selected guard. 2. Relays define individual exit policies. 3. Clients multiplex streams over a circuit.
25

Background: Using Circuits

1. 2. 3. 4.

Clients begin all circuits with a selected guard. Relays define individual exit policies. Clients multiplex streams over a circuit. New circuits replace existing ones periodically.
26

Overview
Summary Tor Background Tor Security Analysis
o Adversary Framework o Security Metrics o Evaluation Methodology o Node Adversary Analysis o Link Adversary Analysis

Future Work
27

Adversary Framework

28

Adversary Framework

29

Adversary Framework

30

Adversary Framework

31

Adversary Framework
Resource Types Relays Bandwidth Autonomous Systems (ASes) Internet Exchange Points (IXPs) Money
32

Adversary Framework
Resource Types Relays Bandwidth Autonomous Systems (ASes) Internet Exchange Points (IXPs) Money
Resource Endowment Destination host 5% Tor bandwidth Source AS Equinix IXPs

33

Adversary Framework
Resource Types Relays Bandwidth Autonomous Systems (ASes) Internet Exchange Points (IXPs) Money
Resource Endowment Destination host 5% Tor bandwidth Source AS Equinix IXPs Goal Target a given users communication Compromise as much traffic as possible Learn who uses Tor Learn what Tor is used for
34

Overview
Summary Tor Background Tor Security Analysis
o Adversary Framework o Security Metrics o Evaluation Methodology o Node Adversary Analysis o Link Adversary Analysis

Future Work
35

Security Metrics

Prior metrics

Security Metrics

Prior metrics 1. Probability of choosing bad guard and exit

a. c2 / n2 : Adversary controls c of n relays b. ge : g guard and e exit BW fractions are bad

Security Metrics

Prior metrics 1. Probability of choosing bad guard and exit

a. c2 / n2 : Adversary controls c of n relays b. ge : g guard and e exit BW fractions are bad

2. Probability some AS/IXP exists on both entry and exit paths (i.e. path independence)

Security Metrics

Prior metrics 1. Probability of choosing bad guard and exit

a. c2 / n2 : Adversary controls c of n relays b. ge : g guard and e exit BW fractions are bad

2. Probability some AS/IXP exists on both entry and exit paths (i.e. path independence) 3. gt : Probability of choosing malicious guard within time t

Security Metrics
Principles 1. Probability distribution 2. Measure on human timescales 3. Based on adversaries

40

Security Metrics
Principles 1. Probability distribution 2. Measure on human timescales 3. Based on adversaries
Metrics 1. Probability distribution of time until first path compromise 2. Probability distribution of number of path compromises for a given user over given time period
41

Overview
Background Onion Routing Security Analysis
o Problem: Traffic correlation o Adversary Model o Security Metrics o Evaluation Methodology o Node Adversary Analysis o Link Adversary Analysis

Future Work
42

TorPS: The Tor Path Simulator

Network Model

Relay statuses Streams StreamCircuit Client Software mappings Model

43

User Model

TorPS: The Tor Path Simulator

Network Model

Relay statuses Streams StreamCircuit Client Software mappings Model

44

User Model

TorPS: User Model

Gmail/GChat Gcal/GDocs Facebook Web search IRC BitTorrent

20-minute traces

45

TorPS: User Model

Gmail/GChat Gcal/GDocs

Typical
Facebook Web search IRC BitTorrent

20-minute traces

46

TorPS: User Model

Gmail/GChat Gcal/GDocs

Session schedule One session at 9:00, 12:00, 15:00, and 18:00 Su-Sa

Typical
Facebook Web search IRC BitTorrent

Repeated sessions 8:00-17:00, M-F Repeated sessions 0:00-6:00, Sa-Su

47

20-minute traces

TorPS: User Model

Gmail/GChat Gcal/GDocs Facebook Web search IRC BitTorrent

Session schedule One session at 9:00, 12:00, 15:00, and 18:00 Su-Sa

Typical Worst Port (6523) Best Port Repeated sessions (443)

8:00-17:00, M-F Repeated sessions 0:00-6:00, Sa-Su
48

20-minute traces

TorPS: User Model

Rank 1 2 3 65312 65313 65314 Port # 8300 6523 26 993 80 443 Exit BW % 19.8 20.1 25.3 89.8 90.1 93.0 LongLived Yes Yes No No No No Application iTunes? Gobby (SMTP+1) IMAP SSL HTTP HTTPS

Default-accept ports by exit capacity.

49

TorPS: User Model

Model Typical IRC BitTorrent WorstPort BestPorst Streams/wee k 2632 135 6768 2632 2632 IPs 205 1 171 205 205 Ports (#s) 2 (80, 443) 1 (6697) 118 1 (6523) 1 (443)

User model stream activity

50

TorPS: The Tor Path Simulator

Network Model

Relay statuses Streams StreamCircuit Client Software mappings Model

51

User Model

TorPS: The Tor Path Simulator

Network Model

metrics.torproject.org
Hourly consensuses Monthly server descriptors archive

52

TorPS: The Tor Path Simulator

Network Model

Relay statuses Streams StreamCircuit Client Software mappings Model

53

User Model

TorPS: The Tor Path Simulator

Client Software Model

Reimplemented path selection in Python Based on current Tor stable version (0.2.3.25) Major path selection features include
Bandwidth weighting Exit policies Guards and guard rotation Hibernation /16 and family conflicts
54

Omits effects of network performance

Overview
Background Onion Routing Security Analysis
o Problem: Traffic correlation o Adversary Model o Security Metrics o Evaluation Methodology o Node Adversary Analysis o Link Adversary Analysis

Future Work
55

Node Adversary

100 MiB/s total bandwidth

Relay Type Any Guard only Exit only Guard & Exit Number 2646 670 403 272 Bandwidth (GiB/s) 3.10 1.25 0.30 0.98

Rank 1 2 3 4 5

Bandwidth Family (MiB/s) 260.5 115.7 107.8 95.3 80.5 torservers.net Chaos Computer Club DFRI Team Cymru Paint
56

Tor relay capacity, 3/31/13

Top Tor families, 3/31/13

Node Adversary

100 MiB/s total bandwidth

Probability to compromise at least one stream and rate of compromise, 10/12 3/13.

57

Node Adversary

100 MiB/s total bandwidth 83.3 MiB/s guard,16.7 MiB/s exit

58

Node Adversary Results

Time to first compromised stream, 10/12 3/13

Fraction compromised streams, 10/12 3/13

59

Node Adversary Results

Time to first compromised guard, 10/12 3/13

Fraction streams with compromised guard, 10/12 3/13

60

Node Adversary Results

Time to first compromised exit, 10/12 3/13

Fraction compromised exits, 10/12 3/13

61

Node Adversary Results

Time to first compromised circuit, 10/12-3/13

62

Overview
Background Onion Routing Security Analysis
o Problem: Traffic correlation o Adversary Model o Security Metrics o Evaluation Methodology o Node Adversary Analysis o Link Adversary Analysis

Future Work
63

Link Adversary

64

Link Adversary
AS 6
AS1 AS2

AS8
AS 6
AS3 AS4

AS 7
AS5

1. Autonomous Systems (ASes)

65

Link Adversary
AS 6
AS1 AS2

AS8
AS 6
AS3 AS4

AS 7
AS5

1. Autonomous Systems (ASes) 2. Internet Exchange Points (IXPs)

66

Link Adversary
AS 6
AS1 AS2

AS8
AS 6
AS3 AS4

AS 7
AS5

1. Autonomous Systems (ASes) 2. Internet Exchange Points (IXPs) 3. Adversary has fixed location

67

Link Adversary
AS 6
AS1 AS2

AS8
AS 6
AS3 AS4

AS 7
AS5

1. Autonomous Systems (ASes) 2. Internet Exchange Points (IXPs) 3. Adversary has fixed location

68

Link Adversary
AS 6
AS1 AS2

AS8
AS 6
AS3 AS4

AS 7
AS5

1. 2. 3. 4.

Autonomous Systems (ASes) Internet Exchange Points (IXPs) Adversary has fixed location Adversary may control multiple entities
a. Top ASes b. IXP organizations
69

Link Adversary
Client locations Top 5 non-Chinese source ASes in Tor (Edman&Syverson 09)
AS# 3320 3209 3269 Description Deutsche Telekom AG Arcor Telecom Italia Country Germany Germany Italy Germany

13184 HanseNet Telekommunikation 6805

Telefonica Deutschland Germany ID 3356 1299 6939 286 DE-CIX Description Level 3 Communications TeliaNet Global Hurricane Electric DE-CIX Frankfurt DE-CIX

AS/IXP Locations Ranked for client location by frequency on entry or exit paths Exclude src/dst ASes Top k ASes /top IXP organization

Type AS AS AS IXP IXP Org.

Example: Adversary locations for BitTorrent client in AS 3320

70

Link Adversary
IXP organizations obtained by manual clustering based on PeerDB and PCH.

# 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

IXP Organization Equinix PTTMetro PIPE NIXI XChangePoint MAE/VERIZON Netnod Any2 PIX JPNAP DE-CIX AEPROVI Vietnam NorthWestIX Terremark Telx NorrNod ECIX JPIX

Size Country 26 8 6 6 5 5 5 4 4 3 2 2 2 2 2 2 2 2 2 global Brazil Australia India global global Sweden US Canada Japan Germany Equador Vietnam Montana, US global US Sweden Germany Japan
71

IXP organizations ranked by size

Link Adversary

Adversary controls one AS, Time to first compromised stream, 1/13 3/13 Best: most secure client AS Worst: least secure client AS

Adversary controls one AS, Fraction compromised streams, 1/13 3/13 Best: most secure client AS Worst: least secure client AS

72

Link Adversary
Adversary controls IXP organization, Time to first compromised stream, 1/13 3/13, Best: most secure client AS Worst: least secure client AS

Adversary controls top ASes, Time to first compromised stream, 1/13 3/13, Only best client AS
73

Overview
Background Onion Routing Security Analysis
o Problem: Traffic correlation o Adversary Model o Security Metrics o Evaluation Methodology o Node Adversary Analysis o Link Adversary Analysis

Future Work
74

Future Work
1. Extending analysis 2. Improving guard selection 3. Using trust-based path selection to protect against traffic correlation 4. Dealing with incomplete and inaccurate AS and IXP maps 5. Include Tors performance-based pathselection features in TorPS
75