- Title
- Detecting derivative malware samples using deobfuscation-assisted similarity analysis
- Creator
- Wrench, Peter Mark
- ThesisAdvisor
- Irvin, Barry
- Date
- 2016
- Type
- Thesis
- Type
- Masters
- Type
- MSc
- Identifier
- http://hdl.handle.net/10962/383
- Identifier
- vital:19954
- Description
- The overwhelming popularity of PHP as a hosting platform has made it the language of choice for developers of Remote Access Trojans (RATs or web shells) and other malicious software. These shells are typically used to compromise and monetise web platforms by providing the attacker with basic remote access to the system, including _le transfer, command execution, network reconnaissance, and database connectivity. Once infected, compromised systems can be used to defraud users by hosting phishing sites, performing Distributed Denial of Service attacks, or serving as anonymous platforms for sending spam or other malfeasance. The vast majority of these threats are largely derivative, incorporating core capabilities found in more established RATs such as c99 and r57. Authors of malicious software routinely produce new shell variants by modifying the behaviours of these ubiquitous RATs, either to add desired functionality or to avoid detection by signature-based detection systems. Once these modified shells are eventually identified (or additional functionality is required), the process of shell adaptation begins again. The end result of this iterative process is a web of separate but related shell variants, many of which are at least partially derived from one of the more popular and influential RATs. In response to the problem outlined above, the author set out to design and implement a system capable of circumventing common obfuscation techniques and identifying derivative malware samples in a given collection. To begin with, a decoder component was developed to syntactically deobfuscate and normalise PHP code by detecting and reversing idiomatic obfuscation constructs, and to apply uniform formatting conventions to all system inputs. A unified malware analysis framework, called Viper, was then extended to create a modular similarity analysis system comprised of individual feature extraction modules, modules responsible for batch processing, a matrix module for comparing sample features, and two visualisation modules capable of generating visual representations of shell similarity. The principal conclusion of the research was that the deobfuscation performed by the decoder component prior to analysis dramatically improved the observed levels of similarity between test samples. This in turn allowed the modular similarity analysis system to identify derivative clusters (or families) within a large collection of shells more accurately. Techniques for isolating and re-rendering these clusters were also developed and demonstrated to be effective at increasing the amount of detail available for evaluating the relative magnitudes of the relationships within each cluster.
- Format
- 148 leaves, pdf
- Publisher
- Rhodes University, Faculty of Science, Computer Science
- Language
- English
- Rights
- Wrench, Peter Mark
- Hits: 764
- Visitors: 874
- Downloads: 151
Thumbnail | File | Description | Size | Format | |||
---|---|---|---|---|---|---|---|
View Details | SOURCE1 | Adobe Acrobat PDF | 3 MB | Adobe Acrobat PDF | View Details |