> I don’t see how you get real security from this approach unless there’s some integration with Intel ME or SGX or other CPU-side ‘secure’ environments, but then you wouldn’t need the TPM to start with.
fTPMs are basically implemented within (or closely working together with) Intel ME reps. AMD PSP.
I'm not sure what you're considering in the emulator case. Either the PCR values are going to be different or the TPM is going to be different, and in both cases that means you're not going to receive the decrypted secret.
this is a question for an TPM expert. I'm a novice at this so take my reasoning with a grain of salt.
A software only emulation shouldn't fool the TPM since part of the secure boot process ties the hash of some PCR banks to the firmware, bootloader and kernel booted, so if you were to modify them in a way that allows you see the key, then it TPM wouldn't be able to produce the correct decryption key. I'm not sure if windows uses those PCR banks to secure bitlocker, but on other OSes you can.
I'm guessing that a hardware mitm would be possible with a discrete tpm, unless you use an aditional factor to boot as it usually recommended to prevent evil maid or cold boot attacks.
Your intuition is sound. The only way for the OS to prove it's identity it to have a secret only it knows, and prove to the TPM it knows it. TPM's do support that, but in this case the OS has nowhere that is robustly secure to store the secret.
Windows could store a obfuscate secret on disk, but it doesn't bother. To be fair, there probably isn't much point - it someone is willing to go to this much work, then it's very likely they would be willing to invest the additional effort to break the obfuscation.
This still gives you a level of protection you wouldn't have without the TPM. The disk can only be read when the TPM is present - so someone stealing disk, or just taking a copy of it won't get them very far. One place that's useful in cloud environments. If the cloud provided replaces a fail disk and doesn't wipe the old one - it's still useless unless someone unless they know what motherboard it was paid with.
Still, I think that's an anti-feature for a laptop. It means if the motherboard fails you've lost the data on the disk even though it's perfectly fine, and indeed that is the case with bitlocker. If you protected the disk with a password you entered on boot up, then it immune to this sort of attack, and you move it between machines. Win, win. That's what I do. But, I don't use Windows to do it.
> Or even if the OS verifies the TPM’s identity somehow
That can be solved using attestation. Sort of. Attestation is just a secret TPM knows, signed by the manufacturer. Windows only deals with TPM's from manufacturers it trusts, and presumably a emulated one wouldn't be one of them. You can't modify Windows to accept any manufacturer because of secure boot.
But I'd lay long odds Windows doesn't do this sort of verification.
But, I ask as someone unfamiliar with TPM, how do authenticated sessions work? How does the OS prove its identity to the TPM in a way an attacker couldn’t spoof in a real MitM attack? Any secrets or keys stored by the OS side would have to reside unencrypted on disk, since it doesn’t have an encryption key yet. Or even if the OS verifies the TPM’s identity somehow, even if this is done in a way that it can’t be worked around just by modifying some files on disk, what stops the attacker from running the same routine in an emulator?
I don’t see how you get real security from this approach unless there’s some integration with Intel ME or SGX or other CPU-side ‘secure’ environments, but then you wouldn’t need the TPM to start with.