Yes, that is a fair summary. More precisely, a short successful virtio-scsi response causes the guest kernel to cache bytes that were not supplied by the device and expose them through the world-readable VPD sysfs attribute.
The guest-kernel heap disclosure is reproducible. I agree that its security classification depends on whether the virtio-scsi device/backend is considered trusted in the relevant threat model. Would validating and propagating the actual virtqueue used length be worthwhile as a robustness fix? Regards, sungbyeongchan
