Yes, with a TPM. A TPM (2.0) can seal secrets and only release it when a machine fulfills certain configuration and state requirements (saved into registers called PCRs).
For example: make the decryption key one part dependant on a passphrase you memorized (to not only rely on a TPM), and one part on something saved in a TPM. If you select the correct PCRs when saving the latter, and your TPM works as advertised (and doesn’t offer an easy way to eavesdrop/fool it), removing the battery would make the TPM not release the secret (if removing the battery even still works on modern machines).
However, this depends on having a unified kernel image, having configured dm-verity and maybe more stuff I don’t recall right now. Probably should also make sure you don’t allow Microsoft’s Secure Boot keys and instead only your own. I hope this will get easier in the future, but I know SystemD is actively developing useful tools for that (e.g. ukify).
That all doesn’t mean the critique of TPMs (intransparent, proprietary) is invalid. Maybe we’ll have OpenTitan based TPMs at some point?
I think Heads (osresearch.net) uses security keys as a kind of substitute TPM, however that only works if you replace your - supported - PCs firmware with it.
I don’t know too much about how this works in particular, so I can’t really compare it. safeboot.dev recommends Heads where possible, which I understand is partly due to safeboot relying on proprietary firmware implementations, while Heads uses libre software for the most part. Sadly the Heads firmware only supports older models/CPUs, which afaik don’t receive (all) microcode updates, including one which weakens the IOMMU.
Yes, with a TPM. A TPM (2.0) can seal secrets and only release it when a machine fulfills certain configuration and state requirements (saved into registers called PCRs).
For example: make the decryption key one part dependant on a passphrase you memorized (to not only rely on a TPM), and one part on something saved in a TPM. If you select the correct PCRs when saving the latter, and your TPM works as advertised (and doesn’t offer an easy way to eavesdrop/fool it), removing the battery would make the TPM not release the secret (if removing the battery even still works on modern machines).
However, this depends on having a unified kernel image, having configured dm-verity and maybe more stuff I don’t recall right now. Probably should also make sure you don’t allow Microsoft’s Secure Boot keys and instead only your own. I hope this will get easier in the future, but I know SystemD is actively developing useful tools for that (e.g. ukify).
That all doesn’t mean the critique of TPMs (intransparent, proprietary) is invalid. Maybe we’ll have OpenTitan based TPMs at some point?
can’t it be done with a security key, like yubikey or similar?
I think Heads (osresearch.net) uses security keys as a kind of substitute TPM, however that only works if you replace your - supported - PCs firmware with it.
I don’t know too much about how this works in particular, so I can’t really compare it. safeboot.dev recommends Heads where possible, which I understand is partly due to safeboot relying on proprietary firmware implementations, while Heads uses libre software for the most part. Sadly the Heads firmware only supports older models/CPUs, which afaik don’t receive (all) microcode updates, including one which weakens the IOMMU.