On 3/23/22 3:36 PM, Tyler Fanelli wrote: > This an RFC discussing a new API, virDomainGetSevAttestationReport (along with a > virsh command "domgetsevreport"), with initial QEMU support via the > "query-sev-attestation-report" QAPI mechanism. "query-sev-attestation-report" is > supplied a base64-encoded 16 byte "mnonce" string as input, with a purpose of > being embedded into the attestation report to provide protection. > > My main point of concern is the design/communication of the virTypedParameterPtr > exchanged between the client and libvirtd and how they interact together, as I > have seen no other API follow the method I used. Namely, the same > virTypedParameterPtr is used for both input _AND_ output. The same > virTypedParameterPtr containing the original mnonce string inputted to the API is > also used to contain the attestation report upon being returned from the API. > > This contrasts with much of the APIs I've noticed, which use a > virTypedParameterPtr for either input or output, but not both. > > This patch is not final, as I still would like some human-readable outputting > and storage of the attestation report. > > Looking for thoughts on the design of this API, as well as suggested > improvements. The proposed API name contains Sev, when all the existing domain APIs have generic names: virDomainGetLaunchSecurityInfo, virDomainSetLaunchSecurityState. I was thinking it would be nice to avoid that Sev specific bit. So I looked at upcoming SNP and TDX qemu patches to see if they add any qmp commands that take input and return a lot of output. Then maybe it would make sense to name this something like virDomainGetLaunchSecurityInfoWithParams which we could use more in the future. qemu SNP patches: https://github.com/AMDESE/qemu/tree/snp-v3 - Only extend existing query-sev and query-sev-capabilities qemu TDX patches: https://github.com/intel/qemu-tdx - Adds query-tdx and query-tdx-capabilities, both with no input So, that doesn't help. After that, my question is, what query-sev-attestation-report adds. The kernel commit explains what it is: https://github.com/torvalds/linux/commit/2c07ded06427dd3339278487a1413d5e478f05f9 > The SEV FW version >= 0.23 added a new command that can be used to query > the attestation report containing the SHA-256 digest of the guest memory > encrypted through the KVM_SEV_LAUNCH_UPDATE_{DATA, VMSA} commands and > sign the report with the Platform Endorsement Key (PEK). > > See the SEV FW API spec section 6.8 for more details. > > Note there already exist a command (KVM_SEV_LAUNCH_MEASURE) that can be > used to get the SHA-256 digest. The main difference between the > KVM_SEV_LAUNCH_MEASURE and KVM_SEV_ATTESTATION_REPORT is that the latter > can be called while the guest is running and the measurement value is > signed with PEK. So it is LAUNCH_MEASURE, available at VM runtime, signed with an extra key. qemu caches the LAUNCH_MEASURE value at VM startup and lets us query it with qmp any time, so I don't think runtime access matters. Maybe the extra key signing is a security fix or something. I haven't figured it out. But as is it's not clear what this buys us over the launch measurement we already report with virDomainGetLaunchSecurityInfo If we figure out what the point of this is, IMO we can more easily reason about whether it makes sense to add a Sev specific libvirt API, and whether we need virTypedParams for both input and output. For example if the API really is specific to this one and only KVM ioctl/QMP command, we could hardcode the parameters and skip the virTypedParams question entirely. CCing danpb for his thoughts - Cole
