Privilege Escalation and Denial of Service in NVIDIA GPU Driver for Linux
Overview
NVIDIA released a security update for a high-severity vulnerability in the kernel mode layer of their GPU driver for Linux. The vulnerability, tracked as CVE-2024-0090, could allow a local attacker with basic user privileges to cause a denial of service (DoS) or escalate their privileges to root. The flaw resides in the driver's handling of specific IOCTL (Input/Output Control) calls. By crafting a malicious input to a vulnerable IOCTL handler, an attacker can trigger a use-after-free condition, leading to kernel memory corruption. In a DoS scenario, this would crash the host system, impacting all users and processes. In a privilege escalation scenario, the memory corruption could be exploited to overwrite kernel data structures, allowing the attacker to execute arbitrary code with kernel-level privileges. This vulnerability is particularly critical in multi-tenant AI/ML environments where multiple users or containerized workloads share the same physical GPU resources on a single host. A malicious user or a compromised container could exploit this vulnerability to escape its sandbox and gain full control over the underlying host machine, thereby accessing data from other tenants' workloads and compromising the entire system.
Affected Systems
Testing Guide
1. Check your currently installed NVIDIA driver version using the `nvidia-smi` command. 2. Cross-reference the installed version with the "Affected Software" versions listed in the official NVIDIA Security Bulletin for this CVE. 3. If your driver version is listed as vulnerable, the system is affected. 4. It is not recommended to attempt to trigger the vulnerability, as it can cause a system crash. The presence of a vulnerable version is sufficient confirmation.
Mitigation Steps
1. **Update NVIDIA Drivers.** Update to the latest driver version provided by NVIDIA or your hardware vendor. Refer to the NVIDIA security bulletin for the specific patched versions for your driver branch. 2. **Restrict User Access.** In multi-tenant environments, avoid giving users or containers direct access to the GPU device nodes (`/dev/nvidia*`) if not strictly necessary. 3. **Use Sandboxing Technologies.** Run GPU-accelerated workloads inside hardened containers with technologies like gVisor or Kata Containers to provide an additional layer of kernel isolation, which can mitigate the impact of a successful driver exploit. 4. **Monitor for Malicious Activity.** Implement host-based intrusion detection systems (HIDS) to monitor for anomalous system calls and potential privilege escalation attempts.
Patch Details
NVIDIA released patched drivers in February 2026. For example, updating the 535 branch to version 535.154.05 or newer mitigates this issue. Check the bulletin for your specific branch.