Buffer overflow vulnerabilities in CUDA: a preliminary analysis

Abstract

We present a preliminary study of buffer overflow vulnerabilities in CUDA software running on GPUs. We show how an attacker can overrun a buffer to corrupt sensitive data or steer the execution flow by overwriting function pointers, e.g., manipulating the virtual table of a C++ object. In view of a potential mass market diffusion of GPU accelerated software this may be a major concern.

A Survey of Techniques for Improving Security of GPUs

“…Miele [31] discusses ways of causing stack and heap overflows for corrupting sensitive data or changing the execution flow. They discuss two attacks.…”

A Study of Overflow Vulnerabilities on GPUs

“…It is worth noting that the line 12 is only executed in the malicious kernel to initialize the overf buffer. If we set length to 26 and initialize overf with the value 0x590 (address of the malicious function that can be obtained using printf("%p",malicious) or CUDA-GDB [5]), the output at line 18 would be string "Normal". This is because with value 26, we can only overwrite the first 5 pointers in fp (sizeof(buf ) + sizeof(pFdummy) * 5 == 26 ).…”

“…However, setting length to 27 would cause the output at line 18 to be "Attack! ", indicating that fp [5] is successfully overwritten by the address of the malicious function. This example demonstrates that current GPUs have no mechanisms to prevent stack overflow like stack canaries on the CPU counterpart.…”

“…However, all existing studies on GPU security have less considered vulnerabilities that are more extensively studied on CPUs. The paper [5] presents a preliminary study of buffer overflow vulnerabilities in CUDA, but the breadth and depth are limited as compared to our work. For example, we put more emphasis on heap overflows between concurrently running kernels, which we believe deserves special attentions in future research of securing shared GPU access.…”

“…Furthermore, we expect that other traditional software vulnerabilities on CPU platforms would have important implications for GPUs, because of similar language features (CUDA programming language inherits C/C++). Although preliminary experiments has been conducted to show the impact of overflow issues on GPU security [5], much remains unclear considering the wide spectrum of security issues. In this paper, we explore the potential overflow vulnerabilities on GPUs.…”