Ting: Measuring and Exploiting Latencies Between All Tor Nodes. Frank Cangialosi Dave Levin Neil Spring University of Maryland

Size: px

Start display at page:

Download "Ting: Measuring and Exploiting Latencies Between All Tor Nodes. Frank Cangialosi Dave Levin Neil Spring University of Maryland"

Julius Warner
6 years ago
Views:

1 Ting: Measuring and Eploiting Latencies Between All Tor Nodes Frank Cangialosi Dave Levin Neil Spring Universit of Marland 1

2 Measuring latencies Ping Measurement Host Eternal Host Limited to the nodes we control 2

3 Measuring latencies Ping Measurement Host Eternal Host To gain broader insight, we can: 1. Control more nodes? 2. Estimate latencies? Limited to the nodes we control 3

4 Measuring latencies Ping Goal Measurement Host Eternal Host Limited to the nodes we control Latenc between arbitrar nodes 4

5 King technique [Gummadi et al, 2002] ns ns 5

6 King technique [Gummadi et al, 2002] ns ns 5

7 King technique [Gummadi et al, 2002] Measure the RTT of recursive DNS queries Q ns ns 5

8 King technique [Gummadi et al, 2002] Measure the RTT of recursive DNS queries A Q ns Q A ns 5

9 King technique [Gummadi et al, 2002] Measure the RTT of recursive DNS queries ns ns 5

10 King technique [Gummadi et al, 2002] Measure the RTT of recursive DNS queries Ping ns ns 5

11 King technique [Gummadi et al, 2002] Measure the RTT of recursive DNS queries Ping Subtract latencies ns ns 5

12 King technique [Gummadi et al, 2002] Measure the RTT of recursive DNS queries Subtract latencies ns ns 5

13 King technique [Gummadi et al, 2002] Measure the RTT of recursive DNS queries Subtract latencies King does not directl measure the path between two hosts ns ns 5

14 King technique [Gummadi et al, 2002] Measure the RTT of recursive DNS queries Subtract latencies King does not directl measure the path between two hosts ns ns 5

15 King technique [Gummadi et al, 2002] Measure the RTT of recursive DNS queries Subtract latencies King does not directl measure the path between two hosts ns ns 5

16 King technique [Gummadi et al, 2002] Measure the RTT of recursive DNS queries Subtract latencies King does not directl measure the path between two hosts ns ns Onl 3% support recursive queries 5

17 TING a tool for measuring latenc between arbitrar Tor nodes 1 Accurate measures the full path between end hosts 2 Practical does not require modification of end hosts 6

18 What is Tor? Anonmit-enabling overla network Packets routed through series of relas, called a circuit Client Destination Clients choose their own circuits 7

19 What is Tor? Anonmit-enabling overla network Packets routed through series of relas, called a circuit Client Destination Clients choose their own circuits 7

20 Wh Tor? Large 6,536 relas in 77 countries (on October 29, 2015) Diverse 5,520 /24s, ~61% residential networks Growing [ 8

21 Tor-specific constraints 9

22 Tor-specific constraints 9

23 Tor-specific constraints 9

24 Tor-specific constraints Ping 9

25 Tor-specific constraints Ping Subtract latencies? 9

26 Tor-specific constraints Tor traffic ma be treated differentl 9

27 Tor-specific constraints Tor traffic ma be treated differentl % ICMP slower 0.6 CDF % TCP slower min(icmp) - min(tcp) (msec) 10

28 Tor-specific constraints Tor traffic ma be treated differentl 11

29 Tor-specific constraints Tor traffic ma be treated differentl 11

30 Tor-specific constraints Tor traffic ma be treated differentl Tor Cannot create one-hop circuits 11

31 Tor-specific constraints Tor traffic ma be treated differentl Tor Cannot create one-hop circuits 12

32 Tor-specific constraints Tor traffic ma be treated differentl Cannot create one-hop circuits 13

33 Tor-specific constraints Tor traffic ma be treated differentl Cannot create one-hop circuits 14

34 Tor-specific constraints Tor traffic ma be treated differentl Cannot create one-hop circuits Must account for forwarding delas F 14

35 Tor-specific constraints Tor traffic ma be treated differentl Cannot create one-hop circuits Must account for forwarding delas F Queuing / Scheduling Encrption & Decrption Contet Switches 14

36 Tor-specific constraints Tor traffic ma be treated differentl Cannot create one-hop circuits Must account for forwarding delas F F F F Queuing / Scheduling Encrption & Decrption Contet Switches 14

37 Tor-specific constraints Tor traffic ma be treated differentl Cannot create one-hop circuits Must account for forwarding delas 15

38 Ting technique 16

39 Measurement Host s d Summar 17

40 Measurement Host s d Measure full path between and Summar w z 18

41 Measurement Host s d Measure full path between and Summar w z 18

42 Measurement Host s d Measure full path between and Summar w z 18

43 Measurement Host s d Measure full path between and Summar w z F F F F F F F F 18

44 Measurement Host s d Measure full path between and Summar w z F F F F F F F F 18

45 Measurement Host s d Isolate RTT between client and Summar w z F F F F 19

46 Measurement Host s d Isolate RTT between client and Summar w z F F F F 19

47 Measurement Host s d Isolate RTT between client and Summar w z F F F F F F 19

48 Measurement Host s d Isolate RTT between client and Summar w z F F F F F F 19

49 Measurement Host s d Isolate RTT between client and Summar w z F F F F F F 19

50 Measurement Host s d Isolate RTT between client and Summar w z F F F F F F 20

51 Measurement Host s d Isolate RTT between client and Summar w z Subtract latencies F F F F F F 20

52 Measurement Host s d Isolate RTT between client and Summar w z F F F F 20

53 Measurement Host s d Isolate RTT between client and Summar w z F F F 21

54 Measurement Host s d Isolate RTT between client and Summar w z F F F 21

55 Measurement Host s d Isolate RTT between client and Summar w z F F F F F 21

56 Measurement Host s d Isolate RTT between client and Summar w z F F F F F 21

57 Measurement Host s d Isolate RTT between client and Summar w z F F F F F 21

58 Measurement Host s d Isolate RTT between client and Summar w z F F F F F 22

59 Measurement Host s w d z Isolate RTT between client and Summar Subtract latencies F F F F F 22

60 1 2 3 Isolate RTT between client and Summar F F 22

61 1 2 3 = RTT(,) + F + F Summar F F 23

62 1 2 3 s d s d s d w z w z w z 24

63 1 2 3 min ( X * ) min ( X * ) min ( X * ) s d s d s d w z w z w z Minimum of multiple, independent samples of each circuit 25

64 Ting evaluation Implemented Ting using the Stem Tor controller No modifications to the Tor client How well does Ting work? How accurate? How man samples? How consistent? 26

65 Ting evaluation How well does Ting work? How accurate? How man samples? How consistent? 31 Tor relas 930 Total pairs 27

66 How accurate is Ting? Fraction of PlanetLab Pairs All Pairs Measured / Real 28

67 How accurate is Ting? Fraction of PlanetLab Pairs Ideal All Pairs Measured / Real 28

68 How accurate is Ting? Fraction of PlanetLab Pairs Ideal All Pairs Measured / Real 28

69 How accurate is Ting? Fraction of PlanetLab Pairs Ideal All Pairs Measured / Real Ting estimates tpicall within 10% of real 28

70 How accurate is Ting? Fraction of PlanetLab Pairs < 50 ms ms ms > 250 ms Ideal Measured / Real Ting estimates tpicall within 10% of real 29

71 How accurate is Ting? Fraction of PlanetLab Pairs < 50 ms ms ms > 250 ms Ideal Measured / Real Ting estimates tpicall within 10% of real 29

72 How accurate is Ting? Fraction of PlanetLab Pairs < 50 ms ms ms > 250 ms Ideal Measured / Real Relative error lower at higher latencies Forwarding delas create an additive error RTT(,) + F + F Ting estimates tpicall within 10% of real 29

73 How man samples does Ting need? Cumulative Fraction of Fraction Pairs Measured Min Within 1ms Within 1% Within 5% Within 10% Cumulative Samples Per Circuit 30

74 How man samples does Ting need? Cumulative Fraction of Fraction Pairs Measured Min Within 1ms Within 1% Within 5% Within 10% Cumulative Samples Per Circuit 30

75 How man samples does Ting need? Fraction of Pairs Onl ~15 samples needed to reach values within 5% of min Within 5% Cumulative Samples Per Circuit Ting remains accurate with few samples 31

76 How consistent are Ting measurements? 30 pairs of real Tor relas, measured once an hour over a week Latenc (ms) Pairs (Sorted in Increasing Order of Median Latenc) 32

77 How consistent are Ting measurements? 30 pairs of real Tor relas, measured once an hour over a week 50% var b < 5ms over a week Ting measurements need not be updated often

78 Evaluation summar How well does Ting work? Tpicall within 10% of real latenc Remains accurate with few samples Var b onl a few ms over time Ting s stabilit and accurac permit collection of an all-pairs RTT dataset

79 Evaluation summar How well does Ting work? Tpicall within 10% of real latenc Remains accurate with few samples Var b onl a few ms over time Ting s stabilit and accurac permit collection of an all-pairs RTT dataset

80 All-pairs RTT dataset 50 Tor relas outside of our control 1,225 pairs in all-pairs RTT dataset Geographic distribution Latenc distribution CDF Latenc (msec) 35

81 Applications 1 Speeding up deanonmization of Tor circuits 2 Improving Tor s path selection algorithm 3 Gain insight into non-tor nodes 36

82 Applications 1 Speeding up deanonmization of Tor circuits 37

83 Classic traffic-analsis attack [Murdoch and Danezis, 2005] Client Malicious Destination 38

84 Classic traffic-analsis attack [Murdoch and Danezis, 2005] Client Malicious Destination 38

85 Classic traffic-analsis attack [Murdoch and Danezis, 2005] Client Malicious Destination 38

86 Classic traffic-analsis attack [Murdoch and Danezis, 2005] Attacker s Goal: find all nodes in the circuit Client Malicious Destination 38

87 Classic traffic-analsis attack [Murdoch and Danezis, 2005] Attacker s Goal: find all nodes in the circuit Client Malicious Destination Probe 38

88 Classic traffic-analsis attack [Murdoch and Danezis, 2005] Attacker s Goal: find all nodes in the circuit Client Malicious Destination Probe 38

89 Classic traffic-analsis attack [Murdoch and Danezis, 2005] Attacker s Goal: find all nodes in the circuit Client Malicious Destination Probe Time- and bandwidth-intensive 38

90 Classic traffic-analsis attack [Murdoch and Danezis, 2005] Attacker s Goal: find all nodes in the circuit Client Malicious Destination Probe Time- and bandwidth-intensive Done in an unguided fashion 38

91 Classic traffic-analsis attack [Murdoch and Danezis, 2005] Attacker s Goal: find all nodes in the circuit Client Malicious Destination Probe Time- and bandwidth-intensive Done in an unguided fashion 38

92 Classic traffic-analsis attack CDF RTT-unaware Ignore too-large RTTs + informed target selection Fraction of nodes tested 39

93 Classic traffic-analsis attack CDF RTT-unaware Ignore too-large RTTs + informed target selection Fraction of nodes tested 39

94 Faster deanonmization with Ting Appl what the attacker knows about latencies 40

95 Faster deanonmization with Ting Appl what the attacker knows about latencies e2e RTT 40

96 Faster deanonmization with Ting Appl what the attacker knows about latencies e2e RTT 40 Directl measured

97 Faster deanonmization with Ting Appl what the attacker knows about latencies e2e RTT Pre-measured with Ting 40 Directl measured

98 Faster deanonmization with Ting Appl what the attacker knows about latencies e2e RTT Client s RTT to the entr node is unknown Pre-measured with Ting 40 Directl measured

99 Faster deanonmization with Ting Appl what the attacker knows about latencies e2e RTT Client s RTT to the entr node is unknown Pre-measured with Ting 40 Directl measured

100 Faster deanonmization with Ting Reason about what the client entr RTT would have to be e2e RTT Client s RTT to the entr node is unknown Pre-measured with Ting 40 Directl measured

101 Faster deanonmization with Ting Reason about what the client entr RTT would have to be Client s RTT to the entr node is unknown Pre-measured with Ting 41 Directl measured

102 Faster deanonmization with Ting Reason about what the client entr RTT would have to be Client s RTT to the entr node is unknown Pre-measured with Ting 41 Directl measured

103 Ruling out nodes without probing them Reason about what the client entr RTT would have to be Client s RTT to the entr node is unknown Pre-measured with Ting 42 Directl measured

104 Ruling out nodes without probing them Reason about what the client entr RTT would have to be Client s RTT to the entr node is unknown Pre-measured with Ting 42 Directl measured

105 Ruling out nodes without probing them Reason about what the client entr RTT would have to be Client s RTT to the entr node is unknown Pre-measured with Ting 42 Directl measured

106 Ruling out nodes without probing them Reason about what the client entr RTT would have to be Client s RTT would have to be negative Client s RTT to the entr node is unknown Pre-measured with Ting 42 Directl measured

107 Ruling out nodes without probing them Reason about what the client entr RTT would have to be Too-Large RTT Client s RTT would have to be negative Client s RTT to the entr node is unknown Pre-measured with Ting 42 Directl measured

108 Ruling out too-large RTTs CDF RTT-unaware Ignore too-large RTTs + informed target selection Fraction of nodes tested 43

109 Ruling out too-large RTTs CDF RTT-unaware Ignore too-large RTTs + informed target selection Fraction of nodes tested 43

110 Informed target selection Probe nodes according to probabilit that the are on the circuit Too-Large RTT Client s RTT would have to be negative Client s RTT to the entr node is unknown Pre-measured with Ting 44 Directl measured

111 Informed target selection Probe the more likel circuits first Client s RTT to the entr node is unknown Pre-measured with Ting 45 Directl measured

112 Informed target selection Probe the more likel circuits first Client s RTT to the entr node is unknown Pre-measured with Ting 45 Directl measured

113 Informed target selection Probe the more likel circuits first Client s RTT to the entr node is unknown Pre-measured with Ting 45 Directl measured

114 Informed target selection Probe the more likel circuits first Average Ting RTT Client s RTT to the entr node is unknown Pre-measured with Ting 45 Directl measured

115 Informed target selection Probe the more likel circuits first Average Ting RTT Client s RTT to the entr node is unknown Pre-measured with Ting 45 Directl measured

116 Informed target selection Probe the more likel circuits first Score Lower score implies higher likelihood Client s RTT to the entr node is unknown Pre-measured with Ting 46 Directl measured

117 Informed target selection Probe the more likel circuits first Score Lower score implies higher likelihood Client s RTT to the entr node is unknown Pre-measured with Ting 46 Directl measured

118 Informed target selection Probe the more likel circuits first Score Lower score implies higher likelihood Client s RTT to the entr node is unknown Pre-measured with Ting 47 Directl measured

119 Informed target selection Probe the more likel circuits first Score Lower score implies higher likelihood Client s RTT to the entr node is unknown Pre-measured with Ting 47 Directl measured

120 Faster deanonmization with Ting CDF RTT-unaware Ignore too-large RTTs + informed target selection Fraction of nodes tested Informed target selection decreases search time b a median of

121 Faster deanonmization with Ting CDF RTT-unaware Ignore too-large RTTs + informed target selection Fraction of nodes tested Informed target selection decreases search time b a median of

122 Faster deanonmization with Ting CDF RTT-unaware Ignore too-large RTTs + informed target selection Fraction of nodes tested Informed target selection decreases search time b a median of

123 Summar We lack a practical tool for measuring the latenc between two arbitrar hosts TING measures the latenc between Tor nodes is fast, accurate, and practical Source code and data available at: 49

124 Implementation Ting Client Test Relas Language: Pthon Tor Controller: Stem Tor patched SLOC: ~400 Tor (latest) PublishDescriptors 0 Restricted Eit Polic Uptime: > 1 month 50

Pinpointing Anomalies in Large-Scale Traceroute Measurements. Romain Fontugne & Kenjiro Cho November 10, 2016

Pinpointing Anomalies in Large-Scale Traceroute Measurements Romain Fontugne & Kenjiro Cho November 10, 2016 1 / 31 This presentation On going research work conducted at IIJ-II In collaboration with: Emile