On this page (6 sections)

Ubuntu has released security notice USN-9003-1 for the Linux kernel built for NVIDIA systems. It covers Ubuntu 26.04 LTS and 24.04 LTS and corrects a large number of kernel issues across many subsystems. The one it describes by name is CVE-2023-20585, a flaw in how some AMD processors check the Reverse Map Table (RMP) when the IOMMU touches host buffers.

If you run Ubuntu servers on NVIDIA-specific kernels, especially hosts that run confidential virtual machines on AMD hardware, this is an update to schedule. Most ordinary websites on standard Ubuntu kernels are not running these packages.

What the notice says

Ubuntu's notice has two parts. The first describes CVE-2023-20585: some AMD processors did not properly perform RMP checks when the IOMMU accessed certain host buffers. Ubuntu says a local attacker with hypervisor access could possibly use this to trigger an out-of-bounds condition and compromise the integrity of SEV-SNP guest memory.

The second part is a broad statement that several other security issues were found in the Linux kernel and that an attacker could possibly use them to compromise the system. The notice lists affected areas, including x86, ARM64, GPU drivers, network drivers, file systems, USB, the BPF subsystem and many more. The text available to us names no CVE identifiers for those other fixes, so this article cannot describe them individually.

The notice text we reviewed does not give fixed package version numbers. Use your package manager, or the full notice page, to see the exact version your release should install.

Understanding CVE-2023-20585

SEV-SNP is an AMD feature that protects virtual machine memory from the host that runs it. The RMP is a table the processor uses to track which memory belongs to which guest, so that even the hypervisor cannot tamper with protected guest pages. The weakness, filed as CWE-788 (memory access after end of buffer), is that the checks were insufficient when the IOMMU accessed certain host buffers. The CVE record describes the result as a potential loss of confidential guest integrity.

The GitHub advisory lists the CVSS v4 base score as 5.6 with these traits: local attack vector, high attack complexity, no attack requirements, high privileges required, no user interaction, and high impact on integrity only. In plain terms, this is not a remote attack on a web server. The attacker needs high privileges and a compromised hypervisor first. The AMD bulletin is AMD-SB-3016.

Who is realistically exposed

  • Hosts running the linux-nvidia-7.0 or linux-nvidia-bos kernel packages on Ubuntu 26.04 LTS or 24.04 LTS.
  • Operators of AMD-based virtualization hosts that rely on SEV-SNP confidential guests, since the described impact is on SEV-SNP guest memory integrity.
  • Teams running the other, unlisted kernel fixes in the same notice, whose impact the available text only describes as possible system compromise.

A typical shared hosting account or a site on a standard cloud VM is unlikely to be directly affected by the AMD issue. The kernel update still matters because it bundles many other fixes.

How to check your exposure

First see which kernel is running. This read-only command prints the running kernel release; NVIDIA-specific kernels usually say so in the name.

Show the running kernel release (Ubuntu)Shell
uname -r

Next, list whether the affected packages are installed. This read-only query prints a line for each matching package with its installed version, or nothing if none are present.

List installed NVIDIA kernel packages (Ubuntu)Shell
dpkg -l | grep -E 'linux-nvidia'

Check your Ubuntu release too, since the notice covers 26.04 LTS and 24.04 LTS only.

Show the Ubuntu release (Ubuntu)Shell
lsb_release -a

What to do now

Caution. Kernel updates take effect only after a reboot. Schedule it in a maintenance window, drain or migrate virtual machines first, and confirm you can reach the host out of band if it fails to come back.

This updates the package index and upgrades installed packages, including the kernel packages, on Ubuntu. Review the list apt shows before confirming.

Apply available updates (Ubuntu)Shell
sudo apt update && sudo apt upgrade

After the reboot, run the uname command again and compare the result with the new package version.

Reboot to load the new kernel (Ubuntu)Shell
sudo reboot
  1. Find hosts running linux-nvidia-7.0 or linux-nvidia-bos.
  2. Read the full USN-9003-1 page for the fixed versions for your release.
  3. Run apt update and apt upgrade on those hosts.
  4. Reboot in a planned window and confirm the new kernel with uname -r.
  5. For AMD SEV-SNP hosts, also check AMD-SB-3016 for firmware or microcode guidance.
  6. Restrict who has hypervisor-level access, since the flaw assumes a privileged, compromised hypervisor.

Monitoring afterwards

Kernel patching is about the host layer, so keep an inventory of kernel versions across your fleet and re-check it after each Ubuntu notice. Outside the host, watch your public-facing services for unexpected changes such as certificate or DNS changes, which Webforta Monitor can track. For the wider picture of keeping servers current, see small business website security.

The sources reviewed do not say whether CVE-2023-20585 or the other kernel issues are exploited in the wild. Treat that as unconfirmed rather than assuming either way.

Sources

This article is based on the public sources below. Details can change as vendors and researchers publish more; check the original advisories before acting on version numbers.

  1. CVE-2023-20585 - GitHub Advisory Databasegithub.com
  2. USN-9003-1: Linux kernel (NVIDIA) vulnerabilities | Ubuntu security notices | Ubuntuubuntu.com ·
  • Ubuntu
  • Linux kernel
  • AMD SEV-SNP
  • CVE-2023-20585
  • patching