Title: From Memory Safety to Semantic Bugs: Automated Vulnerability Discovery Across the Java Ecosystem
Date: November 23 (Monday)
Time: 11:30am EST
Location: (Virtual) https://teams.microsoft.com/meet/213271552961371?p=xRW3btFyzmnSnGsRhi
Fabian Fleischer
Ph.D. Candidate - School of Cybersecurity and Privacy
Georgia Institute of Technology
Committee Members
Dr. Taesoo Kim (Advisor) - School of Cybersecurity and Privacy, Georgia Institute of Technology
Dr. Brendan Saltaformaggio - School of Cybersecurity and Privacy, Georgia Institute of Technology
Dr. Saman Zonouz - School of Cybersecurity and Privacy, Georgia Institute of Technology
Dr. Sukarno Mertoguno - School of Cybersecurity and Privacy, Georgia Institute of Technology
Dr. Greg Heon - Armadin, Inc.
Abstract
Security bugs in the Java and Android ecosystem span a wide range, from memory corruption in native libraries, through established weakness classes such as unsafe deserialization, to flaws in authorization logic. The higher a bug class sits in this range, the harder it is to tell automatically whether an execution reveals a bug at all. This dissertation argues that automated vulnerability discovery can cover the whole range by supplying, at each level, the bug oracle that level lacks, and that this oracle becomes more expensive to obtain as the bug class becomes more semantic.
MALintent fuzzes the Intent handlers of closed-source Android applications, and JNIFuzz generates harnesses for the native code behind them. Here the language itself defines what a bug is, so the challenge is to call the code the way the application does. Gondar targets security-sensitive sinks in Java applications. It pairs a coverage-guided fuzzer with two LLM agents, one that reaches a sink and one that crafts an input that triggers it, and finds three times as many vulnerabilities as Jazzer.
The defense focuses on AuthGap, which detects authorization flaws in Java data systems. For authorization, no general oracle exists: the weakness class does not say which operations of a given system require which checks. AuthGap builds the oracle for each system from the system’s own redundancy. Where several paths reach the same sensitive effect and all but one enforce a check, the exception is the bug. Because each system places its checks differently, supporting a new system takes a short profile rather than new code. Across 14 Java data systems, AuthGap found 11 new authorization vulnerabilities, including 3 CVEs and 2 pending.