client.GetQuote() returns error "inappropriate ioctl for device"

[shuk777]

Hi, there.

I tried tools/attest/attest.go but got error. However, I was able to get quote with Intel official tdx-tools.

If u need more info plz let me know.

[deeglaze]

What’s your output for uname -a? The driver has been in flux, so we’d need the kernel version.

[shuk777]

Thanks for ur reply @deeglaze . It's Linux tdx-test 6.2.16-v5.0.mvp40-generic #tdx SMP PREEMPT_DYNAMIC Mon Nov 6 11:00:15 UTC 2023 x86_64 x86_64 x86_64 GNU/Linux, both host and guest

[msanft]

Hey @deeglaze,

i have been running into the same issue. v0.3.1 returns an empty report for me when querying through the configfs, and the device doesn't seem to be supported anymore in this version. I tried to run v0.2.2, which should be the last version supporting the device, and client.GetRawQuote() returns "inappropriate ioctl for device".

Guest kernel is 6.7.0, host is 6.8.0-1004-intel.

[deeglaze]

tdx-guest is not an upstream driver. This repo started as a means to test attestation with proposed patches. Now that configfs-tsm has been merged upstream, the configfs quote provider should be the only one that works. Still, this all depends on how your distro is managing their kernel patches and built config. Which distro are you using? We test the public images on GCE that are marked with TDX_CAPABLE guestOsFeatures.

[msanft]

tdx-guest is not an upstream driver. This repo started as a means to test attestation with proposed patches. Now that configfs-tsm has been merged upstream, the configfs quote provider should be the only one that works. Still, this all depends on how your distro is managing their kernel patches and built config. Which distro are you using? We test the public images on GCE that are marked with TDX_CAPABLE guestOsFeatures.

Yeah, I know. I don't think my issue specifically revolves around go-tdx-guest. I can manually query a report on the machine through the ioctl, so I think the reason for v0.2.2 not working is some API differential between the library and the driver since then.

Regarding the configfs, I'll see and try to debug this. I think this is a kernel / distro issue, as manually querying a report through the configfs yields an empty report too. But I think this is nothing you, as a library author, should invest any time in.