Cloudanix VulnDB

CVE-2022-21821

High · 7.8

CVSS

7.8

EPSS

2.1%

Published

Mar 29, 2022

CISA due date

—

Cloudanix analysis

Schedule a fix SSVC: Track

A problem in an NVIDIA graphics programming tool can let an attacker take over a computer if someone runs the tool on a specially made bad file.

Exploitation
Requires User Interaction
Attacker needs
User Interaction
Complexity
Low
Impact
Remote Code Execution
Component
Application
Class
Memory Corruption
Remediation
Upgrade
Effort
Medium
Patch risk
Low

What it actually does

Integer overflow in cuobjdump triggers memory corruption that can be leveraged for arbitrary code execution when a malformed file is processed.

Why it matters to the business

Successful exploitation on developer or build systems can result in full host compromise, data theft, or denial of service; the issue affects widely deployed NVIDIA CUDA Toolkit installations on Linux and Windows with no listed fixed versions.

How an attack unfolds

An attacker prepares a specially crafted corrupted binary, delivers it to a target user (via download or shared storage), and relies on the victim executing cuobjdump locally against the file, causing the integer overflow to corrupt memory and achieve remote code execution.

How to fix it

Update the NVIDIA CUDA Toolkit to the newest available release that resolves the integer overflow in cuobjdump; until then, avoid executing cuobjdump against files from untrusted sources and restrict its use to isolated analysis environments.

How to confirm the fix worked

Verify the installed CUDA Toolkit version against the latest NVIDIA release notes, run cuobjdump in a controlled sandbox against known-good and known-bad test files, and confirm the process does not exhibit unexpected memory access or crashes.

Typical attack path

  1. Create a malformed file that triggers the integer overflow in cuobjdump
  2. Distribute the file to the intended victim through common channels
  3. Victim runs cuobjdump against the file on a Linux or Windows system
  4. Integer overflow leads to memory corruption and attacker-controlled code execution

How to detect it

Search endpoint logs and process creation events for invocations of cuobjdump on files originating outside trusted repositories; monitor for unexpected child processes, crashes, or memory anomalies immediately following cuobjdump execution.

  • cuobjdump
  • process execution of cuobjdump on files with unusual extensions or from downloads folder

What your security team should do

  • Prioritize patching NVIDIA CUDA Toolkit on all developer and CI systems
  • Implement application allow-listing or sandboxing for cuobjdump execution
  • Add cuobjdump to file-type and process monitoring rules in endpoint detection tools
  • Educate users about the risks of running analysis tools on untrusted binaries

The record

NVIDIA CUDA Toolkit SDK contains an integer overflow vulnerability in cuobjdump.To exploit this vulnerability, a remote attacker would require a local user to download a specially crafted, corrupted file and locally execute cuobjdump against the file. Such an attack may lead to remote code execution that causes complete denial of service and an impact on data confidentiality and integrity.

Affected packages

Ecosystem Package Introduced Fixed
Debian:11 nvidia-cuda-toolkit 0 No fix
Debian:12 nvidia-cuda-toolkit 0 11.6.2-2
Debian:13 nvidia-cuda-toolkit 0 11.6.2-2
Ubuntu:16.04:LTS nvidia-cuda-toolkit 0 No fix
Ubuntu:18.04:LTS nvidia-cuda-toolkit 0 No fix
Ubuntu:20.04:LTS nvidia-cuda-toolkit 0 No fix
Ubuntu:22.04:LTS nvidia-cuda-toolkit 0 No fix
Ubuntu:24.04:LTS nvidia-cuda-toolkit 0 No fix
Ubuntu:24.10 nvidia-cuda-toolkit 0 No fix
Ubuntu:25.04 nvidia-cuda-toolkit 0 No fix
Ubuntu:25.10 nvidia-cuda-toolkit 0 No fix
Ubuntu:Pro:16.04:LTS nvidia-cuda-toolkit 0 No fix
Ubuntu:Pro:18.04:LTS nvidia-cuda-toolkit 0 No fix
Ubuntu:Pro:20.04:LTS nvidia-cuda-toolkit 0 No fix

Related vulnerabilities