The "Zapscape" Vulnerability Explained: How a KVM Flaw Allows Complete Linux Host Takeover

In a critical development for cloud infrastructure security, a new Linux kernel vulnerability dubbed "Zapscape" (CVE-2026-64561) has been disclosed. This severe flaw allows an attacker with kernel privileges inside an L1 guest virtual machine (VM) to escape KVM isolation and execute arbitrary code on the host system with root privileges. The risk is particularly high when nested virtualization is exposed to untrusted guests.

The Technical Anatomy of Zapscape

At its core, Zapscape is a stale-root check ordering flaw in KVM's shadow-MMU bookkeeping that leads to a use-after-free (CWE-825). During a guest-triggered page fault handling, KVM can reclaim MMU pages and invalidate the shadow MMU root page that is still being utilized by the fault-handling path. Because the path fails to re-verify the root, KVM blindly continues operations under the invalidated root.

Child shadow pages are subsequently created, inheriting this invalid state, which eventually results in the page being freed while stale list references remain. This creates a dangling link and enables a post-free write primitive.

Exploitation and Proof of Concept

Security researcher Hyunwoo Kim, who discovered the bug, demonstrated a public Proof-of-Concept (PoC) targeting AMD nested SVM/NPT on Linux 7.1.3. The PoC successfully builds a full exploit chain that writes a root-owned file named /Zapscape directly on the vulnerable KVM host.

While AMD systems are natively vulnerable, Intel systems require both EPT page-walk lengths 4 and 5 to be explicitly exposed to the L1 guest for the exploit to work.

Mitigation and Patch Details

Red Hat assigned a preliminary CVSS score of 7.0 to the vulnerability. It affects Linux kernels from 5.9 upwards. The upstream fix, merged as commit 2abd5287f083, addresses the logic error by moving the stale-root check after the make_mmu_pages_available() function. Administrators running KVM hosts are strongly advised to update to fixed stable kernels (such as 6.6.148 or 7.1.6) immediately.