GPU VulnDB

Database/NVIDIA / GPU stack

Linux kernel vmwgfx: surface array_size taken from userspace without a shader-model bound

CVE-2026-68446NVIDIA / GPU stackcurated

Impact

vmw_surface_metadata::array_size arrives from userspace and was not validated against the limits the available Shader Model imposes. vmwgfx is the guest DRM driver for Linux VMs on VMware ESXi, which in most GPU estates means the management, build and utility VMs sitting alongside the accelerator hosts rather than the GPU nodes themselves. NVD rates the result 7.8 local with full C/I/A - guest-kernel compromise reachable by an unprivileged user inside the VM, so a foothold in a low-value utility VM becomes root in that VM. Guests with no virtual SVGA device, or where vmwgfx never loads, are not exposed.

Who can reach it

Unprivileged local user inside a VMware guest holding an open /dev/dri render node. No remote path, and nothing to exploit if the vmwgfx module is not loaded.

What to do

Update the guest kernel to a stable release carrying the validation fix and reboot the guest - this is guest-side work, not an ESXi host patch. For headless server VMs that have no need for a virtual display, blacklisting vmwgfx removes the attack surface without a kernel update.

References

This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.