Tempest: Temporal Dynamics in Anonymity Systems Ryan Wails U.S. - - PowerPoint PPT Presentation

▶
tempest temporal dynamics in anonymity systems
SMART_READER_LITE
LIVE PREVIEW

Tempest: Temporal Dynamics in Anonymity Systems Ryan Wails U.S. - - PowerPoint PPT Presentation

Sep 17, 2023 194 likes •906 views

Tempest: Temporal Dynamics in Anonymity Systems Ryan Wails U.S. Naval Research Laboratory Yixin Sun Princeton University Aaron Johnson U.S. Naval Research Laboratory Mung Chiang Princeton University Prateek Mittal Princeton University

slide-1
SLIDE 1

Tempest: Temporal Dynamics in Anonymity Systems

Ryan Wails U.S. Naval Research Laboratory Yixin Sun Princeton University Aaron Johnson U.S. Naval Research Laboratory Mung Chiang Princeton University Prateek Mittal Princeton University

PETS ’18 25 July 2018

slide-2
SLIDE 2

Introduction

slide-3
SLIDE 3

Motivation

  • Tor (USENIX 2004)
  • DeNASA (PETS 2016)
  • Counter-RAPTOR (S&P 2017)
  • TAPS (NDSS 2017)
  • LAP (S&P 2012)
  • HORNET (CCS 2015)
  • Dovetail (PETS 2014)
  • PHI (PETS 2017)

Prior work: static security analyses

1

slide-4
SLIDE 4

Our Contribution: Three Temporal Dynamics

2

slide-5
SLIDE 5

Our Contribution: Three Temporal Dynamics

  • 1. Client Mobility: Clients move over time

2

slide-6
SLIDE 6

Our Contribution: Three Temporal Dynamics

  • 1. Client Mobility: Clients move over time
  • 2. User Behavior: Users make many connections over time

2

slide-7
SLIDE 7

Our Contribution: Three Temporal Dynamics

  • 1. Client Mobility: Clients move over time
  • 2. User Behavior: Users make many connections over time
  • 3. Routing Changes: Internet routes change over time

2

slide-8
SLIDE 8

Temporal Dynamics & Anonymity Systems

Client Mobility User Behavior Routing Changes Tor Tor DeNASA Counter-RAPTOR TAPS Legend: Vulnerability Resistance

3

slide-9
SLIDE 9

Temporal Dynamics & Anonymity Systems

Client Mobility User Behavior Routing Changes Tor Tor DeNASA Counter-RAPTOR TAPS Net-Layer HORNET LAP Dovetail PHI Legend: Vulnerability Resistance

3

slide-10
SLIDE 10

Temporal Dynamics & Anonymity Systems

Client Mobility User Behavior Routing Changes Tor Known Known Tor DeNASA Counter-RAPTOR Known TAPS Known Net-Layer HORNET LAP Dovetail PHI Legend: Vulnerability Resistance

3

slide-11
SLIDE 11

Temporal Dynamics & Anonymity Systems

Client Mobility User Behavior Routing Changes Tor Novel Known Known Tor DeNASA Novel Novel Counter-RAPTOR Novel Novel Known TAPS Known Novel Net-Layer HORNET Novel Novel LAP Novel Novel Dovetail Novel PHI Novel Legend: Vulnerability Resistance

3

slide-12
SLIDE 12

Outline

  • 1. Client Mobility & Tor
  • 2. User Behavior & DeNASA

4

slide-13
SLIDE 13

Brief Tor Background

slide-14
SLIDE 14

Tor Background

guard middle exit

  • 1. Clients use only one guard for a few months
  • 2. Relays are selected with probability prop. to bandwidth

5

slide-15
SLIDE 15

Client Mobility & Tor

slide-16
SLIDE 16

Client Mobility Example

Ryan as an example...

6

slide-17
SLIDE 17

Client Mobility Example

Connected to Tor from 1) Home Which networks (ASes) saw my ingress Tor traffic?

7

slide-18
SLIDE 18

Client Mobility Example

Connected to Tor from 1) Home Which networks (ASes) saw my ingress Tor traffic? Verizon Tata LeaseWeb

7

slide-19
SLIDE 19

Client Mobility Example

Connected to Tor from 1) Home 2) Coffee Shop Which networks (ASes) saw my ingress Tor traffic? Verizon Tata LeaseWeb

7

slide-20
SLIDE 20

Client Mobility Example

Connected to Tor from 1) Home 2) Coffee Shop Which networks (ASes) saw my ingress Tor traffic? Verizon Tata LeaseWeb Comcast Telia

7

slide-21
SLIDE 21

Client Mobility Example

Connected to Tor from 1) Home 2) Coffee Shop 3) Mobile Hotspot Which networks (ASes) saw my ingress Tor traffic? Verizon Tata LeaseWeb Comcast Telia

7

slide-22
SLIDE 22

Client Mobility Example

Connected to Tor from 1) Home 2) Coffee Shop 3) Mobile Hotspot Which networks (ASes) saw my ingress Tor traffic? Verizon Tata LeaseWeb Comcast Telia AT&T

7

slide-23
SLIDE 23

Client Mobility Example

Connected to Tor from 1) Home 2) Coffee Shop 3) Mobile Hotspot 4) Airport Which networks (ASes) saw my ingress Tor traffic? Verizon Tata LeaseWeb Comcast Telia AT&T

7

slide-24
SLIDE 24

Client Mobility Example

Connected to Tor from 1) Home 2) Coffee Shop 3) Mobile Hotspot 4) Airport Which networks (ASes) saw my ingress Tor traffic? Verizon Tata LeaseWeb Comcast Telia AT&T Zayo

7

slide-25
SLIDE 25

Client Mobility Example

Connected to Tor from 1) Home 2) Coffee Shop 3) Mobile Hotspot 4) Airport 5) Hotel Which networks (ASes) saw my ingress Tor traffic? Verizon Tata LeaseWeb Comcast Telia AT&T Zayo

7

slide-26
SLIDE 26

Client Mobility Example

Connected to Tor from 1) Home 2) Coffee Shop 3) Mobile Hotspot 4) Airport 5) Hotel Which networks (ASes) saw my ingress Tor traffic? Verizon Tata LeaseWeb Comcast Telia AT&T Zayo TelefonicaEspaña Telxius

7

slide-27
SLIDE 27

Client Mobility Example

Connected to Tor from 1) Home 2) Coffee Shop 3) Mobile Hotspot 4) Airport 5) Hotel 6) CCCB Which networks (ASes) saw my ingress Tor traffic? Verizon Tata LeaseWeb Comcast Telia AT&T Zayo TelefonicaEspaña Telxius

7

slide-28
SLIDE 28

Client Mobility Example

Connected to Tor from 1) Home 2) Coffee Shop 3) Mobile Hotspot 4) Airport 5) Hotel 6) CCCB Which networks (ASes) saw my ingress Tor traffic? Verizon Tata LeaseWeb Comcast Telia AT&T Zayo TelefonicaEspaña Telxius OrangeEspaña

7

slide-29
SLIDE 29

Client Mobility Example

During travel from United States to Spain:

  • Connected to Tor from 6 different locations
  • Exposed my traffic to additional 7 ASes (3.3× increase)

8

slide-30
SLIDE 30

Adversary Model

  • Adversary compromises a single AS
  • Passive
  • Goal: observe client-guard traffic

9

slide-31
SLIDE 31

Mobility Analysis

  • 1. How mobile are some clients?
  • 2. Does mobility weaken system security?

10

slide-32
SLIDE 32

Mobility Datasets

  • 1. Foursquare (F)
  • 270,000 users
  • 18 months (Apr 2012 – Sep 2013)
  • 2. Gowalla (G)
  • 100,000 users
  • 20 months (Feb 2009 – Oct 2010)

# Countries 2 3 4 5 6 ≥ 7 Users F 40145 13179 5649 2708 1490 2574 Users G 17884 4557 1694 705 305 299 Q1 Days F 48 120 195 228 248 245 Q1 Days G 7 31 56 77 103 125

11

slide-33
SLIDE 33

Mobility & Tor

  • Assume each user connects from most popular Tor AS in

each country.

  • Compute average probability that largest 50 ASes

compromise client-guard path.

12

slide-34
SLIDE 34

5 10 15 20 25 0.00 0.05 0.10 0.15 0.20 0.25 Vulnerability of Mobile Tor Users to Client−Guard Compromise

  • Num. Countries

Mean Probability of Compromise

  • ●
  • ●
  • ●
  • ●
  • ● ● ●
  • Gowalla

Foursquare

Points denote median user, shade shows range

13

slide-35
SLIDE 35

Takeaways

  • 1. Many clients are mobile!
  • 2. Mobility can reduce system security.

14

slide-36
SLIDE 36

User Behavior & DeNASA (PETS 2016)

slide-37
SLIDE 37

DeNASA

The DeNASA “g-select” algorithm: Do not select guards where suspects AS1299 (Telia) or AS3356 (Level 3) are on the client-guard link.

15

slide-38
SLIDE 38

DeNASA Example

AS44 g1 g2 g3 Pr(g1) = 0.25 Pr(g2) = 0.25 Pr(g3) = 0.5

✭✭✭✭✭✭ ✭ ❤❤❤❤❤❤ ❤

Pr(g1) = 0.25

✭✭✭✭✭✭ ✭ ❤❤❤❤❤❤ ❤

Pr(g2) = 0.25

✭✭✭✭✭✭ ✭ ❤❤❤❤❤❤ ❤

Pr(g3) = 0.5 AS1299 Pr(g1) = 0 Pr(g2) = 0.33 Pr(g3) = 0.67

16

slide-39
SLIDE 39

DeNASA Example

AS44 g1 g2 g3 Pr(g1) = 0.25 Pr(g2) = 0.25 Pr(g3) = 0.5

✭✭✭✭✭✭ ✭ ❤❤❤❤❤❤ ❤

Pr(g1) = 0.25

✭✭✭✭✭✭ ✭ ❤❤❤❤❤❤ ❤

Pr(g2) = 0.25

✭✭✭✭✭✭ ✭ ❤❤❤❤❤❤ ❤

Pr(g3) = 0.5 AS1299 Pr(g1) = 0 Pr(g2) = 0.33 Pr(g3) = 0.67

16

slide-40
SLIDE 40

DeNASA Example

AS44 g1 g2 g3 Pr(g1) = 0.25 Pr(g2) = 0.25 Pr(g3) = 0.5

✭✭✭✭✭✭ ✭ ❤❤❤❤❤❤ ❤

Pr(g1) = 0.25

✭✭✭✭✭✭ ✭ ❤❤❤❤❤❤ ❤

Pr(g2) = 0.25

✭✭✭✭✭✭ ✭ ❤❤❤❤❤❤ ❤

Pr(g3) = 0.5 AS1299 Pr(g1) = 0 Pr(g2) = 0.33 Pr(g3) = 0.67

16

slide-41
SLIDE 41

DeNASA Example

g-select leaks location information!

17

slide-42
SLIDE 42

DeNASA Example

AS44 g1 g2 g3 AS1299 AS88 AS1299 AS3356 Pr(g3 | AS44) = 0.67 Pr(g3 | AS88) = 1

18

slide-43
SLIDE 43

DeNASA Example

AS44 g1 g2 g3 AS1299 AS88 AS1299 AS3356 Pr(g3 | AS44) = 0.67 Pr(g3 | AS88) = 1

18

slide-44
SLIDE 44

DeNASA Example

AS44 g1 g2 g3 AS1299 AS88 AS1299 AS3356 Pr(g3 | AS44) = 0.67 Pr(g3 | AS88) = 1

18

slide-45
SLIDE 45

DeNASA Example

g3 middle exit Pr(AS44 | g3) = 0.4 Pr(AS88 | g3) = 0.6

19

slide-46
SLIDE 46

Our Tempest Attack

Leak worsens over time! Pr(AS88 | G1 ∧ G2 ∧ · · · ∧ GN) ≫ Pr(AS44 | G1 ∧ G2 ∧ · · · ∧ GN) How can the adversary learn a client’s guard history?

20

slide-47
SLIDE 47

Adversary Model

  • Adversary runs a destination and some relays
  • Passive
  • Goal: learn client AS

21

slide-48
SLIDE 48

Discovering guards over time

AS?? johndoe1 g2 g1 g3 m1 m2 m3 e1 e2 Day 1 Day 2 johndoe1 is using g1! Day 3 Day 4 johndoe1 is now using g3!

22

slide-49
SLIDE 49

Discovering guards over time

AS?? johndoe1 g2 g1 g3 m1 m2 m3 e1 e2 Day 1 Day 2 johndoe1 is using g1! Day 3 Day 4 johndoe1 is now using g3!

22

slide-50
SLIDE 50

Discovering guards over time

AS?? johndoe1 g2 g1 g3 m1 m2 m3 e1 e2 Day 1 Day 2 johndoe1 is using g1! Day 3 Day 4 johndoe1 is now using g3!

22

slide-51
SLIDE 51

Discovering guards over time

AS?? johndoe1 g2 g1 g3 m1 m2 m3 e1 e2 Day 1 Day 2 johndoe1 is using g1! Day 3 Day 4 johndoe1 is now using g3!

22

slide-52
SLIDE 52

Discovering guards over time

AS?? johndoe1 g2 g1 g3 m1 m2 m3 e1 e2 Day 1 Day 2 johndoe1 is using g1! Day 3 Day 4 johndoe1 is now using g3!

22

slide-53
SLIDE 53

Discovering guards over time

AS?? johndoe1 g2 g1 g3 m1 m2 m3 e1 e2 Day 1 Day 2 johndoe1 is using g1! Day 3 Day 4 johndoe1 is now using g3!

22

slide-54
SLIDE 54

Discovering guards over time

AS?? johndoe1 g2 g1 g3 m1 m2 m3 e1 e2 Day 1 Day 2 johndoe1 is using g1! Day 3 Day 4 johndoe1 is now using g3!

22

slide-55
SLIDE 55

Discovering guards over time

AS?? johndoe1 g2 g1 g3 m1 m2 m3 e1 e2 Day 1 Day 2 johndoe1 is using g1! Day 3 Day 4 johndoe1 is now using g3!

22

slide-56
SLIDE 56

Discovering guards over time

AS?? johndoe1 g2 g1 g3 m1 m2 m3 e1 e2 Day 1 Day 2 johndoe1 is using g1! Day 3 Day 4 johndoe1 is now using g3!

22

slide-57
SLIDE 57

Discovering guards over time

AS?? johndoe1 g2 g1 g3 m1 m2 m3 e1 e2 Day 1 Day 2 johndoe1 is using g1! Day 3 Day 4 johndoe1 is now using g3!

22

slide-58
SLIDE 58

Discovering guards over time

AS?? johndoe1 g2 g1 g3 m1 m2 m3 e1 e2 Day 1 Day 2 johndoe1 is using g1! Day 3 Day 4 johndoe1 is now using g3!

22

slide-59
SLIDE 59

Our Tempest Attack

Note:

  • 1. Many other known guard discovery attacks.
  • 2. Other ways to link client connections.

23

slide-60
SLIDE 60

Our Tempest Attack

Adversary then computes posterior location distribution: Pr(AS1 | g1 ∧ g3) Pr(AS2 | g1 ∧ g3) . . . Pr(AS60000 | g1 ∧ g3)

24

slide-61
SLIDE 61

Attack Evaluation

  • 1. Adversary starts with uniform prior over ∼ 60K ASes

25

slide-62
SLIDE 62

Attack Evaluation

  • 1. Adversary starts with uniform prior over ∼ 60K ASes
  • 2. Identified 10 “leaky” client locations

25

slide-63
SLIDE 63

Attack Evaluation

  • 1. Adversary starts with uniform prior over ∼ 60K ASes
  • 2. Identified 10 “leaky” client locations
  • 3. Simulated a client making up to 6 guard selections

25

slide-64
SLIDE 64

Attack Evaluation

  • 1. Adversary starts with uniform prior over ∼ 60K ASes
  • 2. Identified 10 “leaky” client locations
  • 3. Simulated a client making up to 6 guard selections
  • 4. Collected 100 samples for each location

25

slide-65
SLIDE 65

Attack Evaluation

  • 1. Adversary starts with uniform prior over ∼ 60K ASes
  • 2. Identified 10 “leaky” client locations
  • 3. Simulated a client making up to 6 guard selections
  • 4. Collected 100 samples for each location
  • 5. Computed average posterior entropy after adversary

makes x guard observations

25

slide-66
SLIDE 66

1 2 3 4 5 6 5 10 15 Inference of DeNASA clients' locations

  • Num. Observed Guards

Entropy (Bits)

  • AS1759

AS199348 AS25089 AS57217

Points show average entropy after x guard observations

26

slide-67
SLIDE 67

Takeaways

  • 1. Small leaks can quickly become significant.
  • 2. Important to consider the worst-case.

27

slide-68
SLIDE 68

Conclusion

slide-69
SLIDE 69

Temporal Dynamics & Anonymity Systems

Client Mobility User Behavior Routing Changes Tor Novel Known Known Tor DeNASA Novel Novel Counter-RAPTOR Novel Novel Known TAPS Known Novel Net-Layer HORNET Novel Novel LAP Novel Novel Dovetail Novel PHI Novel Legend: Vulnerability Resistance

28

slide-70
SLIDE 70

Future Work

  • 1. Explicitly accounting for temporal dynamics
  • 2. Considering the long-lived adversary
  • 3. Capturing time in evaluations and formalization

29

slide-71
SLIDE 71

Thank you!

30