KubeVirt Resource and Request Handling Flaws

Server-side request forgery in KubeVirt's virt-api port-forward handler allows authenticated Kubernetes users with kubevirt.io:edit permissions to establish arbitrary bidirectional TCP tunnels from virt-api's trusted cluster-internal network position to any routable destination, bypassing NetworkPolicy isolation. The vulnerability arises because virt-api reads the target IP from vmi.Status.Interfaces[0].IP - a value supplied by the QEMU guest agent inside the VM and thus fully controllable by the VM owner when non-masquerade network bindings (bridge or secondary-only) are in use - and passes it directly to net.Dial() without validation. No public exploit or active exploitation has been identified at time of analysis; EPSS data was not available in the provided intelligence.

Symlink-following in KubeVirt's virt-handler permits a container-level attacker to overwrite arbitrary host files and change their ownership. An attacker with access to the virt-launcher container can plant a symlink at the network cache file path expected by the `WriteToCachedFile` function; virt-handler then follows that symlink when calling `os.WriteFile` and `os.Chown`, enabling writes of attacker-influenced JSON content to any path accessible by virt-handler on the host. This represents a partial container-to-host escape - a scope change from the container context to the underlying node - with no public exploit identified at time of analysis.

Unbounded memory allocation in KubeVirt's downward metrics virtio-serial server within Red Hat OpenShift Virtualization 4 allows a local VM guest user to exhaust host-side memory. The virt-handler process - which runs on the Kubernetes node and manages VM lifecycle - uses textproto.Reader.ReadLine() with no read deadline or buffer cap, so a continuous byte stream from the guest causes virt-handler to allocate memory without bound until the Linux OOM killer terminates it. No active exploitation has been confirmed (not in CISA KEV) and no public exploit code has been identified at time of analysis; the CVSS score of 3.8 reflects low severity but the scope-change flag (S:C) correctly captures that the disruption crosses the guest boundary to a privileged host process.