Discovery and Traversal of Security Gateways Alwyn E. Goodloe University of Pennsylvania

advertisement
Discovery and Traversal
of Security Gateways
Alwyn E. Goodloe
University of Pennsylvania
Contessa NS
Protocol eXchange
June 10, 2005
History of Routing Protocols
In early days of ARPANet


Few nodes
Routing tables manually configured at each
node by local system admin
Centralized Management an Alternative


Network manager knows topology and
handles everything
Tools can help, but still difficult
Drawbacks
Managers must know topology
Managers control who gets to play

Can not just go and add or delete a node
Hard to see how the Internet would have
grown to present size had either of these
schemes been adopted.
Dynamic Routing Protocols
Routing tables are updated as part of
protocol
Adapts to changing topology and growth
Theory



Convergence in the face of changes
Correctness
Efficiency of underlying protocols
Security Gateways
Located at cutpoints in the network
Possess an inside and an outside
Nodes on the inside constitute its domain
Gateways control what traffic can enter
and leave a domain
Single Gateway
r1
g
a
c
r2
r3
b
Network
Network as Graph
Gateway Hierarchy
A
A’
D’
D
Traversing Gateways
High-level policies at the gateways
determine which users can communicate
with members of its domain
To enforce policies, gateways authenticate
packets using cryptographic tunnels

Security Associations (IPsec)
Packet filters determine which packets go
in which association
Industrial Practice
Gateways are usually configured using
command line interfaces
Moving to centralized management

Tool support: Solsoft Policy server
Drawbacks same as for routers

Inflexible in the face of changing topology
Want protocols to dynamically find
gateways and set up associations
Set Up Protocol Requirements
Discover gateways along path

Send out distinguished control packets
Negotiate trust relationship based on high-level
policy
Set up associations using some key-exchange
protocol (IKE, JFK)
Install packet filters (low-level policies) on the
gateways that are derived from/compatible with
high-level policies
Discovery protocols are a special class of
signaling protocol
Do People Really Want This
Cisco’s Tunnel Endpoint Discovery (TED)
Protocol performs discovery



Limited. Assumes two gateways.
Built into high-end security gateways
Indicates industrial demand
IETF’s IP Security Policy (IPSP) group

Charter says they will develop a discovery
protocol
Need For Theory
We have designed several protocols for
setting up collections of IPsec tunnels

Sectrace, L3A (WITS 05)
Each had subtle flaws that were
uncovered by formal analysis
Want a formalism and theory for
developing such signaling protocols

Like SPI-Calculus and MSR for crypto
protocols
Tunnel Calculus
Key-Exchange as abstract building-block


Not concerned with the cryptography
Terminates with associations and policies
properly set up
Captures essential details of the network

Contrasts with process algebras that abstract
away from network
Built in layers
Layers
Discovery
Establishment
Trust Negotiation
Security Processing
Packet Routing
Example
a
g
b
Establishment
Authenticate
Negotiate
Discovery
Discovery
Negotiation
Establishment
Authenticate
Establishment
Encryption
Establishment Layer
A
B
Req(spi-a, request)
SADB AB
SPDB
Rep(spi-a, spi-b, request)
SADB AB
SPDB
SADB BA
SPDB
SADB BA
SPDB
Trust Negotiation
When discovery packet destined for node
B arrives at a gateway G, how does


G know if it should allow the set up
The initiator know that B is inside of G’s
domain
These questions need to be settled by
high-level policy
This must be known before establishment
begins
Trust Management
Need to discover, access, process high level
policy
Work in progress
Related works





Security Policy Protocol (SSP) IETF IPSP
SPKI/SDSI
PolicyMaker/KeyNote
QCM/SD3
….
Borrow ideas and abstract away details
Security Processing Layer
Abstraction of IPsec
Security Associations (SA) – Define
cryptographic transforms


Abstract away the cryptography
Tunnel mode
Packet P(a,b,y) in association cd:I
P(c,d,S(I,P(a,b,y))
Association Database (SADB)
Security Processing Layer Contd
Packet filters called security policies direct
traffic into SAs
Security Policy Database (SPDB)

SPDB-IN and SPDB-Out
Must model the processing of packets!

Headers added and removed in accordance with
policy
Each packet that enters the system must
undergo processing
Outgoing packets processed before sent down
to routing layer
IPsec example
i1
i2
G
A
B
i3
P(A,B,y)
AB:[(AB)(AG)]
AB:[(AG)]
P(A,B,S(i3,P(A,B,y)))
P(A,G,S(i1,P(A,B,S(i3,P(A,B,y)))))
AB:[(GB)]
AB:[(AB)(GB)]
P(A,B,y)
P(G,B,S(i2,P(A,B,S(i3,P(A,B,y))
Routing Layer
Network topology induced by forwarding
tables
Routers only route



Packet p arrives @ r.
Lookup next hop in table.
Send packet to next hop
Secure nodes do IPsec processing

All packets that arrive are sent up to be
processed by security layer
Formalism
Based on multiset rewriting and equational
logic
Very basic logic


Control flow must be explicit
Each rule may execute concurrently unless
constrained
State must be explicitly passed among rules


MSR’s L-Predicates
Our resumption terms <…..>
Routing Grammar
Routing Layer Rules
Security Processing Grammar
Nesting a packet
Output Rule
Safety/Liveness Properties
Safety:If a tunnel if formed, then a proper
set of credentials exist
Liveness: Given some global policy, the
two parties should be able to communicate
assuming everything is in the right place
Still working on formalizing these
Future Work
Dissertation will flush out the details of
each layer




Executable models in Maude
Proofs of properties
Work on the theorems
Trust negotiation layer
Contessa NS People
Carl A. Gunter
Mark-Oliver Stehr
Alwyn Goodloe
Matthew Jacobs
Gaurav Shah
Michael McDougall
Gual Agha
Michael Greenwald
Sanjeev Khanna
Jose Meseguer
Koushik Sen
Prasanna Thati
Download