- Nvidia confirms a new bug in Container Toolkit, and GPU Operator
- The bug allows malicious actors to execute code remotely
- A fix was already deployed, so patch now
The Nvidia Container Toolkit for Linux, a set of tools that allows devs to build and run GPU-accelerated containers using Docker, or other container runtimes, carries a vulnerability that allows threat actors to gain access to the host file system and thus execute malicious code remotely, run denial of service attacks, escalate privileges, steal sensitive information, or tamper with the victim’s data.
The company confirmed the news in a security advisory, noting both the Nvidia Container Toolkit, and Nvidia GPU Operator (a Kubernetes-native solution that automates the deployment, management, and monitoring of Nvidia GPU resources in a Kubernetes cluster) are vulnerable to the bug which is being tracked as CVE-2025-23359.
It was assigned a severity score of 8.3, and was said to affect all versions of Container Toolkit up to and including 1.17.3, and all versions up to and including 24.9.1 of GPU Operator.
Patch bypass
The bugs were fixed in versions 1.17.4 and 24.9.2 respectively. It is also worth mentioning that the flaw is only present on Linux, and does not impact use cases where CDI is used.
Cybersecurity researchers from Wiz claim this is actually a bypass for another vulnerability. Apparently, the previous bug is tracked as CVE-2024-0132, and has a 9.0 severity score, making it critical, as it could allow malicious actors to mount the host’s root file system into a container, granting them free access to virtually anything. What’s more, the access can be used to launch privileged containers and achieve full host compromise.
Nvidia says the issue was fixed in September 2024, and to address the issue, users are advised to apply the released patches, and make sure not to disable the “–no-cntlibs” flag in production environments, it was said.
Via The Hacker News
Leave a comment