Abstract

This talk evaluates pointer tainting, an incarnation of Dynamic Information Flow Tracking (DIFT). Pointer tainting has been used for two main purposes: detection of privacy-breaching malware (e.g., trojan keyloggers obtaining the characters typed by a user), and detection of memory corruption attacks against non-control data (e.g., a buffer over?ow that modi?es a user’s privilege level). The technique is considered one of the only methods for detecting them in unmodi?ed binaries. Unfortunately, almost all of the incarnations of pointer tainting are ?awed. We found that pointer tainting generates itself the conditions for false positives. We analyse the problems in detail and investigate various ways to improve the technique. Most have serious drawbacks in that they are either impractical (and incur many false pos- itives still), and/or cripple the technique’s ability to detect attacks. We argue that depending on architecture and operating system, pointer tainting may have some value in detecting memory corruption attacks (albeit with false negatives and not on the popular x86 architecture), but it is not suitable for automated detecting of privacy-breaching malware such as keyloggers.

Bio: Asia Slowinska is a third-year PhD student at the Vrije Universiteit Amsterdam. Her research concerns intrusion detection, signature generation, and honeypots. Currently she’s interning with MSRC .