A vulnerability in the Linux kernel vhost subsystem has been disclosed, allowing a local user to read and write kernel memory and potentially gain root access, across many kernel versions. The flaw, tracked as CVE-2026-74580 and named Vghost, was identified by Tencent Zhuque Lab. Tencent released a fix on 3 August and published a detailed write-up on 12 October with working demonstrations of both a local-root and a guest-to-host escape. There is no public evidence of exploitation to date.
The bug affects Linux kernels from 4.11 up to 7.2-rc4. Tencent reproduced the full local-user-to-root chain on five distributions (Ubuntu 26.04, RHEL 9.8, Amazon Linux 2023, Microsoft Azure Linux 3.0, Oracle Linux 10.2). In Ubuntu, the vhost device node was restricted to the kvm group, while on the other four images the node was world-readable (0666), enabling local access by any user.
The vulnerability arises from the vhost memory-cache mechanism used to speed up virtio device operations. A cache hit is treated as proof of a previously checked region, and the cache is cleared only under certain conditions. When a device IOTLB is attached and VHOST_SET_VRING_ADDR or VHOST_SET_VRING_NUM commands adjust ring parameters while the queue is live, the new addresses can be accepted without immediate validation.
Once the cache is populated, translations may be used without rechecking, allowing the kernel to copy data to or from a pointer outside the mapped region and enabling kernel read/write. The second, guest-to-host path requires an IOMMU-enabled setup and can overwrite a function pointer in QEMU, giving a guest control of the host process. The published fix (commit de845981da67) clears the cache whenever the ring is reconfigured.
Upstream fixes are in place, with first fixed upstream stable releases listed for 5.10.y, 5.15.y, 6.1.y, 6.6.y, 6.12.y, 6.18.y, and 7.1.y/7.2. Until patched, the guidance is to restrict access to /dev/vhost-net and /dev/vhost-vsock, and treat membership in the relevant group as a privilege.