Year: 2026
Type: article
Abstract: Abortive functions are commonly employed in the Windows kernel to prevent further execution on critical failures. However, some non-abortive functions bear confusing names that mislead developers into believing they will abort. we denoted them as pseudo-abortive functions. Invoking pseudo-abortive functions to handle critical errors introduces pseudo-abortive bugs . These bugs form a distinct class of error-handling weaknesses that have not been systematically modeled, rooted in the semantic mismatch between the caller's intent (i.e., requiring abortion) and the callee's functionality (i.e., no abortion). Detecting such bugs is challenging in binaries , as callers’ intent and callees’ semantics are obscure. To this end, we present MisAbort , a novel binary analysis technique that detects pseudo-abortive bugs in Windows by analyzing callee functionality and caller intent. MisAbort first classifies Windows kernel functions as abortive or non-abortive callees. For non-abortive callees, it isolates suspicious callsites via rule-based caller intent filtering. For these filtered callsites, it then applies a customized machine language model to infer callers’ abortive intent. Finally, MisAbort uses a caller-callee bug assessment to assess the influence of intent-functionality mismatches. We implement a prototype of MisAbort and evaluate it on Windows kernel binaries. MisAbort identifies 85 newly confirmed pseudo-abortive bugs with 13 CNNVD IDs assigned.
DOI: 10.1145/3862596
Language: English
Subfield: Software
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
Award: G8414988738
OpenAlex ID: W7235584596
