Revisiting Address Space Randomization

摘要

Address space randomization is believed to be a strong defense against memory error exploits. Many code and data objects in a potentially vulnerable program and the system could be randomized, including those on the stack and heap, base address of code, order of functions, PLT, GOT, etc. Randomizing these code and data objects is believed to be effective in obfuscating the addresses in memory to obscure locations of code and data objects. However, attacking techniques have advanced since the introduction of address space randomization. In particular, return-oriented programming has made attacks without injected code much more powerful than what they were before. Keeping this new attacking technique in mind, in this paper, we revisit address space randomization and analyze the effectiveness of randomizing various code and data objects. We show that randomizing certain code and data objects has become much less effective. Typically, randomizing the base and order of functions in shared libraries and randomizing the location and order of entries in PLT and GOT do not introduce significant difficulty to attacks using return-oriented programming. We propose a more general version of such attacks than what was introduced before, and point out weaknesses of a previously proposed fix. We argue that address space randomization was introduced without considering such attacks and a simple fix probably does not exist.