130 likes | 316 Views
Toward a Framework for Preventing Side-Channel Attacks in Wireless Networks. Jeff Pang. WPA/VPN/SSHTunnel. The problem. http…ali…@bb!t. divx…data…bcd. Packet sizes, timing :. Packet contents:. time. time. → [probably generated by alice ’s laptop]
E N D
Toward a Framework for Preventing Side-Channel Attacks in Wireless Networks Jeff Pang
WPA/VPN/SSHTunnel The problem http…ali…@bb!t divx…data…bcd Packet sizes, timing: Packet contents: time time → [probably generated by alice’s laptop] → [probably keystrokes: wh!teR@bb!t] ← [probably webpage: http://rightwing-politics.net/madhatter/blog.html] ← [probably watching video: Wonderland.DivX.avi] → [probably speaking: Spanish] ← [probably using a p2p application] → user login: alice → password: wh!teR@bb!t ← webpage: http://rightwing-politics.net/madhatter/blog.html ← streaming video: Wonderland.DivX.avi → voip: “¿Cómo es usted, somberero loco?” ← p2p download: QueenOfHearts.mp3
The problem • Adversary • Passive eavesdropper • Packet contents appear random • Can only determine packet size, time, direction • Packet sizes and timing can reveal sensitive information • Passwords used [Song ’01] • Webpages visited [Sun ’02] • Videos watched [Saponas ’07] • Languages spoken (over VoIP) [Wright ’07] • Identity (e.g., broadcast packet sizes) [Pang ’07]
Problem 1: packet size 100 Example: Broadcast packet sizes used as a fingerprint 250 • Set of packet sizes reveals: • identity: >35% accuracy (< 1 hour of traffic) • webpage: 76% accuracy (< 1 min of traffic) • voip language: 66% accuracy (3 min of traffic) • Usual countermeasures: • Pad packets to [almost] same size 500 300 200 120 Broadcast transmission sizes Broadcast transmission sizes time time
Problem 2: packet timing • Inter-packet spacing reveals: • Keystrokes: 50x faster password cracking time • Countermeasures: • [near] constant bit rate cover traffic
Problem 3: size evolution over time DFT • Fourier transform/HMM on packet size evolution: • video: 66% accuracy (10 min of traffic) • application type: 76% accuracy (10 min of traffic) • Usual countermeasures: • Send at [near] constant bit rate Example: DFT of VBR videos as fingerprints ≈
Previous solutions code modification Language-based information flow security [Volpano ’96, Agat ’00, Meyers ‘99] Proposal: Framework to implement select countermeasures • Enable overhead / privacy protection trade-off • Similar to signature-based anti-virus and IDS Application transparency knowledge of traffic patterns Specific attack countermeasures [Timmerman ’99, Smart ‘00] Trace-based cover traffic [Newman-Wolfe ‘92, Guan ‘01] Status quo Naïve cover traffic opaque overhead none all potential Information prevented from leaking
Part I: Rule-based masking • Example: masking packet sizes • 100 400 250 400 300 400 200 400 120 400 Input transmissions Output transmissions time time Input transmissions • Masking rules: “output size independent of input size” • Performance constraints: “minimize delay”
System overview • definitionMasking rulesPerf. constraints • outputOutput traffic profile
Primary challenges • definition: masking rule language • Must be flexible enough for real countermeasures • Describe packet size, inter-packet spacing • Describe sequences, frequencies, periodicity • Describe multiple time granularities • Must be uniform enough to enable rule composition • e.g. break up all packets so they have uniform size express all rules in terms of inter-packet spacing • output: satisfying multiple masking rules • Must handle infeasible constraints gracefully • Allow the rule language to describe some slack • e.g. “make output as independent as possible of input”
Design questions App 1 App 2 App 3 • Where to apply rules? • per flow: • Can use some flows to cover for others • Assumes flows (mostly) independent • on all outbound traffic • Takes into account flow dependencies • Harder to make application-specific rules • combination of both • Requires explicit declaration of dependencies • How opaque should traffic be? • e.g., treat TCP flows as a unit? • Possibly avoid adverse end-to-end interactions • Don’t inspect packet contents at all • Simpler to analyze, implement rules, hide RTT/bottleneck network
Part II: Learning masking rules • APs learn location dependent rule parameters • Traffic profiles become location rather than user dependent • Mimic local traffic patterns to customize overhead • Challenges: • How to learn parameters over time • How to minimize performance impact of adversarial clients learner learner learner home masking rules airport masking rules starbucks masking rules input traffic profiles input traffic profiles input traffic profiles
Summary • Side channels reveal encrypted data • Wireless makes attacks easy to perform • Attacks discovered after apps deployed • Need temporary “patches” afterward • Proposed rule-based masking • Primary challenges: rule language, composition