Abstract ▼ AI-Helper

Container escape, which exploits vulnerabilities in the shared kernel to break container isolation, is a severe security threat in cloud-native computing. To alleviate the threat, we should allow the minimum number of system calls required by individual containers, but figuring out which system call...

Container escape, which exploits vulnerabilities in the shared kernel to break container isolation, is a severe security threat in cloud-native computing. To alleviate the threat, we should allow the minimum number of system calls required by individual containers, but figuring out which system calls an arbitrary container will need is a challenging problem. This paper presents Prof-gen that automatically creates a restrictive system call policy using static binary analysis and dynamic analysis without any prior knowledge. The tool only requires a container image and a run command. We compared the created system call policy with the results of Confine, a recent study for container attack surface reduction. For 120 official images, Prof-gen reduced the attack surface by 20.2% compared to Confine. All the test containers that applied the profile generated in the application-specific tests ran without failure.