IdSan: An identity-based memory sanitizer for fuzzing binaries

Abstract: Most memory sanitizers work by instrumenting the program at compile time. There are only a handful of memory sanitizers that can sanitize a binary program without source code. Most are location-based, and are therefore unable to detect overflows of global variables or variables on the stack. In this paper we introduce an identity-based memory sanitizer for binary AArch64 programs which does not need access to the source code. It is able to detect overflows of stack- and global variables if the user provides some annotations or DWARF debugging information is available, as well as dynamically allocated memory.