1 Secure Sensor Routing A Clean-Slate Approach Bryan Parno, Mark Luk, Evan Gaustad, Adrian Perrig Carnegie Mellon University

2 Sensor Networks Thousands of nodes, each with: –A CPU and ~10 KB of RAM –A radio –Sensors (e.g., heat, motion, sound) –Limited power Communicate via multi-hop routing Applications: burglar alarms, emergency response, industrial uses All require secure routing!

3 Attacks on Routing Inject incorrect routing information or alter setup/update messages Compromise sensors –Provide malicious routing data/messages Suppress (selectively) routing messages Specific attacks: –Blackhole –Wormhole –Replication –Denial of Service –Sybil –Rushing –Slander –Framing

4 Consequences of Routing Attacks Controlling routing allows the attacker to control the network’s communication –Can view, modify, and/or drop messages –Create loops to exhaust legitimate nodes –Prevent or subvert proper network functionality

5 Techniques for Secure Routing Prevention –Harden protocols by restricting participants’ actions –Typically employs cryptography –Only forestalls known attacks Detection & Recovery –Monitor behavior for malicious activity –Eliminate malicious participants –Must be able to distinguish anomalous behavior and accurately assign blame Resilience –Maintain availability even under unpredicted attacks –Provide graceful performance degradation

6 Previous Work Sensor routing –Most assume trusted environment –INSENS only applicable to certain topologies –SIGF requires GPS Other secure routing protocols –Typically rely on a single technique Prevention: S-BGP, Ariadne Detection & Recovery: Watchdog, Pathrater, Secure Traceroute Resilience: INSENS –Inappropriate for resource-constrained sensor nodes Require PKI or excessive amounts of memory, computation or communication

7 Goals Start from a clean-slate Incorporate all three security techniques –Prevention, detection & recovery, and resilience Provide highly secure, highly available point- to-point routing –Necessary in many applications, e.g., Geographic Hash Tables (GHTs), key establishment, etc. Minimize resource utilization

8 Outline Introduction Overview and Assumptions Address and Routing Setup Forwarding Detection and Recovery Simulation and Implementation

9 Our Routing Protocol Architecture Establish routing tables and network addresses –Use prevention techniques to thwart active attackers –Detect and recover from attempts to deviate from the protocol or to launch additional attacks Apply resilient routing techniques to forward packets –Uses the securely established routing tables and network addresses

10 Assumptions Network authority (NA) uses a public/private keypair {K NA, K -1 NA } Each sensor node preloaded with: –Network authority’s public key K NA –Unique ID x –Certificate: Sig(K -1 NA, ID x ) Signature scheme optimizes for verification Intended for networks of primarily stationary sensors

11 Outline Introduction Overview and Assumptions Address and Routing Setup Forwarding Detection and Recovery Simulation and Implementation

12 Address and Route Setup Overview Goal: –Assign a unique network address to each node –Populate each node’s routing table Accomplished with a recursive grouping algorithm –Initially, each sensor constitutes its own group –Groups repeatedly merge until all nodes belong to same group Each time a node’s group merges, the node adds one bit to its network address and one entry to its routing table Node Routing ID Address Table A 0.1 B 0.0 C 1.1 D 1.0 Node IDAddress Routing Table A0.1 B0.0 C1.1 D1.0 Node IDAddress Routing Table A0.1RT A B0.0RT B C1.1RT C D1.0RT D

13 Recursive Grouping Algorithm Groups act in an asynchronous, distributed fashion Each group: –Collects information about its neighbors –Proposes to merge with smallest neighboring group Based on number of nodes in the group Ties broken based on group ID This metric keeps addresses and routing tables small –Mutual proposal triggers merge Entire process is deterministic for a given topology –Limits the damage an attacker can inflict

14 Recursive Grouping Example

15 Calculating Network Addresses Assume G and G’ decide to merge Each node in G independently extends its network address by one bit based on: Nodes in G’ make similar changes

16 Network Addresses Formation

17 Populating Routing Tables Assume G and G’ decide to merge Each node in G records the neighbor from whom it heard about G’ in its current routing table slot G Prefix Next Hop 0.*C 1.0C D G G’

18 Sample Routing Table Prefix Next Hop 0.* *

19 Outline Introduction Overview and Assumptions Address and Routing Setup Forwarding Detection and Recovery Simulation and Implementation

20 Forwarding Basic forwarding similar to area-style forwarding Given a destination network address route towards node with longest matching prefix Path length in logical hops bound by log(n) –A logical hop may require several physical hops

21 Forwarding Example Prefix Next Hop 0.* * Prefix Next Hop 1.* * Prefix Next Hop 1.* * Prefix Next Hop 0.* * Message from to 0.0.0

22 Outline Introduction Overview and Assumptions Address and Routing Setup Forwarding Detection and Recovery –Threats –Detecting Grouping Deviations –Eliminating Malicious Nodes Simulation and Implementation

23 Threats Compromised nodes may lie about group size or ID to subvert route setup Compromised nodes may claim multiple IDs or try to simultaneously group with several other nodes

24 Detecting Grouping Deviations Maintain a Grouping Verification Tree (GVT) for each group during recursive grouping –Prevents attacker from lying about group ID or size –Based on a hash tree construction Before two groups merge, they verify each other’s GVT Integrity of the GVTs insures integrity of the recursive grouping algorithm Final GVT covers all nodes in the network –Can be used to authenticate any node’s network address

25 Background: Hash Trees Employ a one-way hash function H: {0,1}*→{0,1} ρ to create one-way data structures The Merkle Tree is one such data structure –Each internal node calculated as: Parent = H(Child L || Child R ) –Authenticates a leaf node given the root value and nodes along the path to the root

26 Group ID Computation Assume G and G’ decide to merge Each node in G independently calculates the new group ID as:

27 GVT Formation One GVT per group GVT leaves are IDs of nodes in the group Internal nodes represent intermediate group IDs Each node maintains information about its branch of the GVT –Specifically, the group ID and size of each merge partner

28 GVT Verification Before merging, group G verifies the GVT for G’ (and vice versa) G’ announces its group ID (and size) Group G sends a challenge value to G’ The challenge uniquely selects a node in G’ –Chosen node sends its certificate and GVT information to G Nodes in G verify the GVT values ResponderChallenger λ

29 Eliminating Malicious Nodes Legitimate nodes use the Honeybee mechanism to eliminate malicious nodes To revoke malicious node M, legitimate node L broadcasts: –ID L, ID M, and a signature Legitimate nodes revoke M and L –Prevents a compromised node from revoking more than one legitimate node

30 Outline Introduction Overview and Assumptions Address and Routing Setup Forwarding Detection and Recovery Simulation and Implementation

31 Simulations Comparison against Beacon Vector Routing (BVR) protocol [NSDI 2005] –Optimized for efficiency –No security included Experimental Setup: –500 nodes, random deployment, DOI radio model Summary of Results: –Our routing success rate: 100% –Paths longer than shortest path –Distributes overhead evenly throughout network Better than BVR, even in topologies with voids

32 Metric: Path Stretch Stretch = Protocol Path Length / Optimal Path Length Optimistic for BVR: does not include failed BVR routes

33 Metric: Load Distribution - Uniform ~ 168,000 messages

34 Metric: Load Distribution - Irregular ~ 26,000 messages

35 Implementation Developed in NesC on TinyOS using Telos sensor nodes –Source code to be available soon Challenges overcome: –Reliable Broadcast –Asynchronicity –Asymmetric Links Ongoing work to expand the current testbed

36 Other Contributions Techniques for resilient forwarding Duplicate detection Proofs of performance and correctness Implementation details

37 Conclusions Secure sensor routing is an important and difficult problem Most previous techniques assume a trusted environment or use a single security technique We design a protocol incorporating all three security techniques that still compares favorably to insecure protocols

38 Thank you!

39 Drawbacks of Wired Networks Expensive to deploy Expensive to maintain –Upgrade –Replace Wires can introduce failures Wires are costly Wireless networks are more cost effective!

40 Merging Two Groups Assume G and G’ decide to merge Each node in G independently: –Calculates the new group ID: –Extends its network address by one bit according to: –Records the neighbor from whom it heard about G’ in the current routing table slot

41 Duplicate Detection After recursive grouping concludes, each node announces its ID and network address to its neighbors Run a replication detection algorithm [PaPeGl2005] to identify duplicates Detects nodes that: –Claim multiple IDs –Simultaneously group with several other nodes Duplicates are revoked

42 Resilient Forwarding Extend routing tables to facilitate multi-path forwarding During each merge, a node remembers multiple neighbors that announced the merge target –Leverages natural redundancy in the recursive grouping algorithm Prefix Next Hop 0 Next Hop 1 1.*AC 0.1.*CD 0.0.0A-- B

43 Resilient Forwarding Sender includes a direction string  in its packet  =  0 ||  1 || …||  k,  i  {0,1} Forwarding node selects among next hops based on current value of 

44 GVT Verification Before merging, group G verifies the GVT for G’ (and vice versa) G’ announces its group ID and size Group G chooses a challenger node –Challenger creates challenge –Challenger broadcasts the challenge to G and G’ Based on challenge, G’ chooses responder node –Responder sends its certificate and GVT branch information to G Nodes in G verify the GVT values