---
title: "Critical Linux KVM Flaw Let Guests Access Host Memory"
date: 2026-11-20
author: "Sofia Ramirez"
featured_image: "https://sqmagazine.co.uk/wp-content/uploads/2026/11/linux-kvm-flaw-digital-ocean.jpg"
categories:
  - name: "Cybersecurity"
    url: "/cybersecurity.md"
tags:
  - name: "News"
    url: "/tag/news.md"
---

# Critical Linux KVM Flaw Let Guests Access Host Memory

DigitalOcean disclosed on November 19 that Linux KVM flaw CVE-2026-99312 let a researcher read about 127 GB of host RAM from inside a Droplet. The bug hits only AMD hosts with nested virtualization switched on.

## What Happened?

- Terra Security researcher Oren Yomtov found the flaw and reported it through DigitalOcean’s bug bounty program.
- DigitalOcean live-patched its hypervisor fleet without a full reboot cycle, so Droplet customers have nothing to install.
- Upstream stable kernel 7.3.5 and long-term kernels 6.18.64 and 6.12.118 already carry the fix.
- Self-hosted KVM operators on AMD need a patched host kernel and a reboot, or nested virtualization turned off.

## A stale pointer turned freed memory into a page table

KVM confines each guest with a nested page table that maps guest addresses to real host memory. The processor finds that table through a pointer stored for each virtual CPU.

When a reset signal called INIT pulled a virtual CPU out of a nested VM, KVM never refreshed that pointer. It still aimed at a page KVM had already freed. Once the kernel reused the page, the guest could fill it with entries of its choosing, and the processor obeyed them.

Those forged entries could name any physical page on the host. On the test Droplet, the first data back was the host’s kernel version banner. Terra ran the write test only on its own bare-metal AMD server with no other tenants. There, the guest placed 32 chosen bytes into a live host kernel page.

Both tests began with root inside a researcher-controlled VM, a foothold any attacker would also need on a vulnerable host.

The flaw grew out of a fix. A 2022 patch for [CVE-2022-3344](https://nvd.nist.gov/vuln/detail/CVE-2022-3344), Terra says, switched a reset path back to the outer VM’s control block but missed the page-table pointer. The error then sat in the kernel for just under four years. Terra’s [PatchSlop research](https://www.terra.security/blog/how-security-patches-became-maps-for-attackers) found that **3.2%** of kernel CVEs were introduced by the fix for another CVE.

> 🔴 CRITICAL · Linux kernel: 12 new CVEs Worst (9.8): out-of-bounds read in Ceph snap trace bounds check (CVE-2026-99304) 👉 Fix: see advisories https://t.co/9Sa3bgBudY
> 
> — cve.now (@cvedotnow) [November 19, 2026](https://x.com/cvedotnow/status/2123419106938413163?ref_src=twsrc%5Etfw)

## Distribution kernels trail the upstream fix

DigitalOcean validated the exploit on its hosts and treated it as critical. Linux KVM maintainers, told on October 21, merged the [fix](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=c7c288d178cb), which reached stable kernels on November 17. DigitalOcean rebuilt its [live patch](https://www.digitalocean.com/blog/cve-2026-99312-kvm-patch) from that code and told Droplet customers: “**No action needed on your part.**“

The vulnerable code entered the 6.1 kernel in 2022 and also landed in 5.15.81 and 6.0.11, so older distribution kernels fall in range. Packaged fixes from the major [Linux distributions](https://sqmagazine.co.uk/linux-distro-market-share-statistics/) lag. Debian’s security tracker lists its releases as vulnerable except unstable, and Ubuntu has not shipped a fixed kernel package.

Severity scores split. The **Linux kernel CVE** team rates the flaw **8.8** out of 10, with high impact on confidentiality, integrity, and availability. Red Hat’s [preliminary score](https://access.redhat.com/security/cve/CVE-2026-99312) is 5.6, Moderate, and counts availability impact alone. That fits poorly with a demonstrated read of host memory.

## Nested virtualization keeps exposing KVM hosts

**VirtFusion**, a virtualization vendor tracking the flaw, says nested virtualization is on by default in the kvm\_amd module. Unpatched hosts stay exposed unless someone changed the default. **Januscape (CVE-2026-53359)** in July and **Zapscape (CVE-2026-64561)** in August also needed nested virtualization. Many cloud providers Terra surveyed escaped this bug because they had not enabled the feature on AMD.

Operators can check exposure now:

- **Read /sys/module/kvm\_amd/parameters/nested. A value of 1 or Y means nested virtualization is on.**
- **Install a host kernel that carries the fix, then reboot. Patching guest operating systems changes nothing.**
- **If the update must wait, add options kvm\_amd nested=0 to the host’s module configuration and reboot.**
- **Confirm with uname -r, then reread the nested setting.**

The nested setting can’t change while VMs run, and a new kernel does nothing until the host restarts, so the reboot is the deadline. Switching nested virtualization off also stops tenants running their own VMs, so warn anyone who relies on it first. No exploit code is public, and VirtFusion reports no use outside research. Nobody has published a way to detect a past attempt, so every exposed host leaves one question open: did a tenant find this first?