JIGSAW: Protecting resource access by inferring programmer expectations

Processes retrieve a variety of resources, such as files, from the operating system to function. However, securely accessing resources has proven to be a challenging task, accounting for 10-15% of vulnerabilities reported each year. Current defenses address only a subset of these vulnerabilities in ad-hoc and incomplete ways. In this paper, we provide a comprehensive defense against vulnerabilities during resource access. First, we identify a fundamental reason that resource access vulnerabilities exist - a mismatch between programmer expectations and the actual environment the program runs in. To address such mismatches, we propose JIGSAW, a system that can automatically derive programmer expectations and enforce it on the deployment. JIGSAW constructs programmer expectations as a name flow graph, which represents the data flows from the inputs used to construct file pathnames to the retrieval of system resources using those pathnames. We find that whether a program makes any attempt to filter such flows implies expectations about the threats the programmer expects during resource retrieval, the enabling JIGSAW to enforce those expectations. We evaluated JIGSAW on widely-used programs and found that programmers have many implicit expectations. These mismatches led us to discover two previously-unknown vulnerabilities and a default misconfiguration in the Apache webserver. JIGSAW enforces program expectations for approximately 5% overhead for Apache webservers, thus eliminating vulnerabilities due to resource access efficiently and in a principled manner.