Digital Forensics: The Acceleration of Password Cracking

Digitální forenzní analýza: Zrychlení lámání hesel

Cryptographic protection of sensitive data is one of the biggest challenges in digital forensics. A password is both a traditional way of authentication and a pivotal input for creating encryption keys. Therefore, they frequently protect devices, systems, documents, and disks. Forensic experts know that a single password may notably complicate the entire investigation. With suspects unwilling to comply, the only way the investigators can break the protection is password cracking. While its basic principle is relatively simple, the complexity of a single cracking session may be enormous. Serious tasks require to verify billions of candidate passwords and may take days and months to solve. The purpose of the thesis is thereby to explore how to accelerate the cracking process.

I studied methods of distributing the workload across multiple nodes. This way, if done correctly, one can achieve higher cracking performance and shorten the time necessary to resolve a task. To answer what "correctly" means, I analyzed the aspects that influence the actual acceleration of cracking sessions. My research revealed that a distributed attack's efficiency relies upon the attack mode - i.e., how we guess the passwords, cryptographic algorithms involved, concrete technology, and distribution strategy. Therefore, the thesis compares available frameworks for distributed processing and possible schemes of assigning work. For different attack modes, it discusses potential distribution strategies and suggests the most convenient one. I demonstrate the proposed techniques on a proof-of-concept password cracking system, the Fitcrack - built upon the BOINC framework, and using the hashcat tool as a "cracking engine." A series of experiments aim to study the time, performance, and efficiency properties of distributed attacks with Fitcrack. Moreover, they compare the solution with an existing hashcat-based distributed tool - the Hashtopolis.

Another way to accelerate the cracking process is by reducing the number of candidate passwords. Since users prefer strings that are easy to remember, they unwittingly follow a series of common password-creation patterns. Automated processing of leaked user credentials can create a mathematical model of these patterns. Forensic investigators may use such a model to guess passwords more precisely and limit tested candidates' set to the most probable ones. Cracking with probabilistic context-free grammars represents a smart alternative to traditional brute-force and dictionary password guessing. The thesis contributes with a series of enhancements to grammar-based cracking, including the proposal of a novelty parallel and distributed solution. The idea is to distribute sentential forms of partially-generated passwords, which reduces the amount of data necessary to transfer through the network. Solving tasks is thus more efficient and takes less amount of time. A proof-of-concept implementation and a series of practical experiments demonstrate the usability of the proposed techniques.

Forensics, password, cracking, acceleration, GPGPU, BOINC, hashcat, PCFG

HRANICKÝ, Radek. Digital Forensics: The Acceleration of Password Cracking. Brno, 2021. Ph.D. Thesis. Brno University of Technology, Faculty of Information Technology. 2022-04-19. Supervised by Ryšavý Ondřej. Available from: https://www.fit.vut.cz/study/phd-thesis/890/