When Encryption is Not Enough Privacy Attacks in Content- Centric - PowerPoint PPT Presentation

When Encryption is Not Enough Privacy Attacks in Content- Centric Networking ACM ICN 2017 1

Privacy with IP GET /a/b/c C S RESPONSE: <data> ACM ICN 2017 2

Privacy with IP secure channel GET /a/b/c C S RESPONSE: <data> What’s revealed? • Source and destination addresses and port # • Timing • Packet sizes ACM ICN 2017 3

Privacy with CCN Interest: /a/b/c C P Content: <data> ACM ICN 2017 4

Privacy with CCN encrypted name? Interest: /a/b/c C P Content: <data> encrypted content? What’s revealed? • Consumer and producer locations • Timing • Packet sizes • Producer identity • Interest name (and equality) • … ACM ICN 2017 5

Motivating Question What can an adversary do with interest equality alone? ACM ICN 2017 6

Database Security SELECT * FROM SECRET_TABLE WHERE NAME = <secret> C DB [secret record] ACM ICN 2017 7

Database Security SELECT * FROM SECRET_TABLE WHERE NAME = <secret> C DB [secret record] ACM ICN 2017 8

Eavesdropping Attack C C NAME = 0x1234… DB C ACM ICN 2017 9

Eavesdropping Attack C 0x1234… 0x4356… 0x4356… 0x1234… C NAME = 0x1234… DB 0x1234… 0x1234… 0x1234… 0x9981… 0x9981… 0x9271… C 0x3233… … ACM ICN 2017 10

Empirical Frequency Counts 0x1234… 0x4356… 6 0x4356… 0x1234… 5 0x1234… 0x1234… 0x1234… 4 0x9981… 0x9981… 3 0x9271… 0x3233… 2 … 1 0 0 x1234… 0 x9981… 0x4536 0x9271 0x3233 Count ACM ICN 2017 11

Auxiliary Popularity Info 0.6 0.5 0.4 0.3 0.2 0.1 0 Item 1 Item 2 Item 3 Item 4 Item 5 Popularity ACM ICN 2017 12

Frequency Analysis Attack 6 Count 4 2 0 0 x1234… 0 x9981… 0x4536 0x9271 0x3233 0.6 Popularity 0.5 0.4 0.3 0.2 0.1 0 Item 1 Item 2 Item 3 Item 4 Item 5 ACM ICN 2017 13

Frequency Analysis Attack 6 Count 4 2 0 0 x1234… 0 x9981… 0x4536 0x9271 0x3233 0.6 Popularity 0.5 0.4 0.3 0.2 0.1 0 Item 1 Item 2 Item 3 Item 4 Item 5 ACM ICN 2017 14

CCN as a Content Database Request for encrypted content Get <secret> C Network [Secret Content] Application data P Encrypted data items C ACM ICN 2017 15

CCN as a Content Database Request for encrypted content P ∈ P Get <secret> P C [Secret Content] C ∈ C ACM ICN 2017 16

Relevant Distributions • Real popularity distribution D R ( P ) D A • Auxiliary information distribution A ( P ) • Empirical frequency distribution D E ( C ) ACM ICN 2017 17

Global Eavesdropping Adversary • Nefarious ISPs, nation states, etc. • Questions: – To what extent does auxiliary information accuracy matter? – To what extent does universe size matter? ACM ICN 2017 18

Topology Consumer Edge Router Core Router ACM ICN 2017 19

Different Auxiliary and Popularity Information ACM ICN 2017 20

Matching Auxiliary and Popularity Information ACM ICN 2017 21

Takeaway ∆ ( D A A ( P ) , D R ( P )) ≈ 0 . 0 ∆ ( D E ( C ) , D A A ( P )) ≈ 0 . 0 ACM ICN 2017 22

Auxiliary Information Gap 0.000 0.005 0.010 0.015 0.020 0.025 0.030 0.035 0.040 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Match Ratio ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Length of Simulation [s] ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 10000 ● ● ● ● ● 8000 6000 ● 4000 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 2000 ● ● 0 0.0 0.2 0.4 0.6 0.8 1.0 Statistical Distance ACM ICN 2017 23

Content Universe Size ● ● ● ● ● ● 1.0 ● ● 0.8 Length of Simulation [s] Match Ratio 0.6 ● ● ● ● ● 10000 ● ● ● ● ● ● ● 0.4 ● ● ● ● ● ● ● ● ● ● 8000 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 6000 ● ● ● ● ● ● ● ● ● ● ● 0.2 ● ● ● ● ● ● 4000 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 2000 ● ● ● ● ● ● ● ● ● 0.0 ● ● ● ● ● ● ● 0 ● 0 200 400 600 800 1000 Sample Size ACM ICN 2017 24

Takeaway Auxiliary information accuracy is not as important as sample size ACM ICN 2017 25

Distributed Adversary • Access point, enterprise network middlebox, compromised transit router, etc. • Questions: – Where does the adversary have the best chance at succeeding? – To what extent does caching dampen attack efficacy? – Can content replication (across different producers) help? ACM ICN 2017 26

Edge vs Inner Router ACM ICN 2017 27

Cache Presence ACM ICN 2017 28

Replication ACM ICN 2017 29

Probing for Popularity • What does do if it has no popularity information? ACM ICN 2017 30

Probing for Popularity • What does do if it has no popularity information? • Exploit caches to learn popularity – Assumes plaintext and ciphertext equivalents are fetched with equal distributions ACM ICN 2017 31

Probing Algorithm ACM ICN 2017 32

Probe Results (S = 50) ACM ICN 2017 33

Probe Results (S = 100) ACM ICN 2017 34

Summary • Caching both helps and hurts privacy • Content replication helps bypass adversaries • Preventing namespace enumeration is key to mitigating the attack ACM ICN 2017 35

Future Work • Expand simulator and widen experiments • Analytically quantify the attack match percentage given distributions, network topologies, and cache hit probabilities • Study attack on CDNs today ACM ICN 2017 36

/this/is/the/end/ version=0x00/chunk=0x01/PID=0x02 Questions? ACM ICN 2017 37