| 11 Aug 2021 |
 Mic92 | I think for out-of-tree modules one could have a build hook | 14:25:07 |
 andi- | and the signing key is a (separate) output of the actual kernel build? | 14:25:30 |
| Mic92 | Yes. It could be actually a build hook in the kernel. Every out-of-tree kernel module already has this as a depedency | 14:27:54 |
| Mic92 | *dependency | 14:27:57 |
 @grahamc:nixos.org | I'd love to see development in that area, it'd be a bit tricky to know you're supposed to have access to the signing key | 15:44:08 |
| Mic92 | I just stumbled over this features for the first time when modifying some runc hypervisor. | 15:48:29 |
| andi- | we could have a disallowedRequisites = [ kernel.signingKey ]; as very minimal "safety" against having the key world readable on the system (by accident). That is obviously not a silver bullet. Everyone that can build software against the systems nixpkgs checkout could generate properly signed modules and given that it would have to be deterministic you could probably just generate the key "offline" on another box.. | 15:52:46 |
| andi- | What exactly are we gaining again? :D | 15:52:55 |
| Mic92 | So would need an activation phase that signs all keys afterwards? | 16:10:35 |
| 13 Aug 2021 |
| @grahamc:nixos.org | the work I'm doing around secureboot support is based on a more involved bootloader "install" step which could support signing modules | 19:47:39 |
| 18 Aug 2021 |
| Mic92 | https://github.com/NixOS/nixpkgs/pull/134577 | 05:30:59 |
| @grahamc:nixos.org | tpm2_unseal -c ${dev.tpm2KeyFile.persistentObject} -p ${dev.tpm2KeyFile.authString} > /crypt-ramfs/tpm/unsealed
| 14:47:23 |
| @grahamc:nixos.org | I'm thinking this should be starting an auth session (I think that is the right term) and using the session key for subsequent calls so that the channel with the TPM is all encrypted | 14:48:23 |
