On 3/4/22 8:06 AM, Borislav Petkov wrote: > On Thu, Mar 03, 2022 at 10:47:20AM -0600, Brijesh Singh wrote: >> I did not fail on !req.cert_len, because my read of the GHCB spec says that >> additional data (certificate blob) is optional. A user could call >> SNP_GET_EXT_REPORT without asking for the extended certificate. In this >> case, SNP_GET_EXT_REPORT == SNP_GET_REPORT. >> >> Text from the GHCB spec section 4.1.8 >> --------------- >> https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdeveloper.amd.com%2Fwp-content%2Fresources%2F56421.pdf&data=04%7C01%7Cbrijesh.singh%40amd.com%7C10f73f9bd6524f9f9c4e08d9fde843d5%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637819996563585169%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=GS3iyUCAOmaPuxN4sRi%2B%2FFTsRGF7QC1jISc%2B25c4XtY%3D&reserved=0 >> >> The SNP Extended Guest Request NAE event is very similar to the SNP Guest >> Request NAE event. The difference is related to the additional data that can >> be returned based on the guest request. Any SNP Guest Request that does not >> support returning additional data must execute as if invoked as an SNP Guest >> Request. >> -------------- > Sorry, it is still not clear to me how > > "without asking for the extended certificate" == !req.certs_len > > That's not explained in the help text either. And ->certs_len is part of > the input structure but nowhere does it say that when that thing is 0, > the request will be downgraded to a SNP_GET_REPORT. The decision to downgrade will be done by the hypervisor. See the GHCB spec (page 36), if RBX (aka number of certificate data pages) is zero then section SNP_GET_EXT_REQUEST section (4.1.8) say that hypervisor must treat it as SNP_GET_REQUEST. > > How is the user of this supposed to know? Depending on which ioctl user want to use for querying the attestation report, she need to look at the SNP/GHCB specification for more details. The blob contains header that application need to parse to get to the actual certificate. The header is defined in the spec. From kernel driver point-of-view, all these are opaque data. > > And regardless, you can still streamline the code as in the example I > gave so that it is clear which values are checked and for which does the > request get failed... Will do. thanks
