Two New Attacks Show Rogue Cloud Hosts Could Peek Inside Protected AMD Virtual Machines
Two attacks made public this week, named Mior and Skyfall, allow a malicious host to read data inside virtual machines (VMs) protected by AMD SEV-SNP. This technology is designed to keep a server's operator out of a customer's VM. Both attacks need control of the server's hypervisor, and Mior also needs control of the BIOS. The researchers demonstrated both in lab tests on AMD EPYC server processors.
Mior, developed by a team at ETH Zurich, targets the start-up phase. The researchers redirected writes from the chip's security co-processor to a graphics card they controlled, so start-up finished without errors while part of a key memory ownership table stayed uninitialized. That let the hypervisor change the table's protections. The team then switched on debugging for a VM that had it disabled and read and wrote its memory, succeeding in all 1,000 attempts across both test runs. In a second test, they copied a harmless VM's measurement into a malicious VM, so its signed attestation reports described the harmless software instead.
AMD lists SEV firmware for Mior and microcode for Skyfall, both dated August 12. It plans to give server makers platform firmware in November for Skyfall and in December for Mior, with fixes for embedded chips later. Customers are told to get the BIOS update from their server maker.