Ph.D. Defence Notice - Jon Shahen

Thursday, July 23, 2020 2:00 pm - 2:00 pm EDT (GMT -04:00)
Candidate: Jon Shahen
Title: Solving Problems in Access Control, Cloud Computing, Logic Locking, and Ethereum Smart Contracts with Cook/Turing Reductions
Date: July 23, 2020
Time: 2:00 PM
Supervisor(s): Tripunitara, Mahesh

This thesis addresses computer security problems in: Access Control, Ethereum Smart Contracts, Cloud VM Scheduling, and Logic Locking. These problems are solved using polynomially timed reductions to 2 complexity classes: PSPACE-Complete and NP-Complete. This thesis is divided into 2 parts, problems reduced to: Model Checking (PSPACE-Complete) and Integer Linear Programming (ILP) (NP-Complete). The PSPACE-Complete problems are: Safety Analysis of Administrative Temporal Role Based Access Control (ATRBAC) Policies, and Safety Analysis of Ethereum Smart Contracts. The NP-Complete problems are: Minimizing Information Leakage in Virtual Machine (VM) Cloud Environments using VM Migrations, and Attacking Logic Locked Circuits using a Reduction to Integer Linear Programming (ILP).

In Chapter 3, I create the Cree Administrative Temporal Role Based Access Control (ATRBAC)-Safety solver. Which is a reduction from ATRBAC-Safety to Model Checking. I create 4 general performance techniques which can be utilized in any ATRBAC-Safety solver.

1.      Polynomial Time Solving, which is able to solve specific archetypes of ATRBAC-Safety policies using a polynomial timed algorithm.
2.      Static Pruning, which includes 2 methods for reducing the size of the policy without effecting the result of the safety query.
3.      Abstraction Refinement, which can increase the speed for reachable safety queries by only solving a subset of the original policy.
4.      Bound Estimation, which creates a bound on the number of steps from the initial state, where a satisfying state must exist. This is directly used by the model checker's bounded model checking mode, but can be utilized by any solver with a bound limiting parameter.

In Chapter 4, I analyze ATRBAC-Safety policies to identify some of the ``sources of complexity'' which make solving ATRBAC-Safety policies difficult. I provide analysis of the sources of complexity that exists in the previously published datasets [128,90,54]. I perform analysis of Cree's performance techniques on the previous datasets. I create 2 new datasets, which are shown to be hard instances of ATRBAC-Safety. I analyze the new datasets to show how they achieve this hardness and how they differ from each other and the previous datasets.

In Chapter 5, I create a novel reduction from a Reduced-Solidity Smart Contract, subset of available Solidity features, to Model Checking. This reduction reduces Reduced-Solidity Smart Contract into a Finite State Machine and then reduces to an instance of a Model Checking problem. This provides the ability to test smart contracts published on the Ethereum blockchain and test if there exists bugs or malicious code. I perform empirical analysis on select Smart contracts.

In Chapter 6, I create 2 methods for generating instances of ATRBAC policies into Solidity Smart Contracts. The first method is the Generic ATRBAC Smart Contract. This method requires no modification before deployment. After deployed the owner is able to create, and maintain, the policy using special access functions. The special action functions are automated with code that converts an ATRBAC policy into a series of transactions the owner can run. The second method is the Baked ATRBAC Smart Contract. This method takes an ATRBAC policy and reduces it to a Smart Contract instance with no special access functions. The smart contract can then be deployed by anyone, and that person will have no special access. I perform an empirical analysis on the setup costs, transaction costs, and security each provides.

In Chapter 7, I create a new reduction from Minimizing Information Leakage via Virtual Machine (VM) Migrations to Integer Linear Programming (ILP). I compare a polynomial algorithm by Moon et. al. [71], my ILP reduction, and a reduction to CNF-SAT that is not included in this thesis. The polynomial method is faster, but the problem is NP-Complete thus that solution must have sacrificed something to obtain the polynomial time speed (unless P = NP). I show instances in which the polynomial time algorithm does not produce the minimum total information leakage, but the ILP and CNF-SAT reductions are able to. In addition to this, I show that Total Information Leakage also has a security vulnerability for non-zero information leakage using the <R,C> model. I propose an alternative method to Total Information Leakage, called Max Client-to-Client Information Leakage, which removes the vulnerability at the cost of increased total information leakage.

In Chapter 8, I create a reduction from the Key Recovery Attack on Logic Locked Circuits to Integer Linear Programming (ILP). This is a recreation of the ``SAT Attack'' using ILP. I provide an empirical analysis of the ILP attack and compare it to the SAT-Attack. I show that ``ILP Attack'' is a viable attack, thus future claims of ``SAT-Attack Resistant Logic Locking Techniques'' need to also show resistance to all potential NP-Complete attacks.