Obtain machine-id from SMBIOS, DT, or TPM (on bare metal)

[septatrix]

Component

systemd

Is your feature request related to a problem? Please describe

On stateless systems there is currently no way to persist the machine-id even though this may often be desirable. Sometimes systems are only stateless for robustness reasons and would benefit from persistent identification.

An example of this would be the planned support of slow update rollouts with sysupdate. This mechanism would likely be based on some function which uses the machine-id, version, and threshold percentage as an input and checks if the machine-id crosses the threshold¹. If these devices reboot frequently they would get multiple tries to pass the threshold raising their chance to install the update and skewing the rollout percentage.

¹ Something like: (sd_id128_get_machine_app_specific(<version>) & 0xFF) / 0xFF < threshold (I have never calculated with 128 bit numbers but the ratio should be correct. Also UUIDv4 has fixed bits but as we hash it we can disregard that.)

Describe the solution you'd like

Similar to KVM and QEMU, the machine-id should also be tried to be gathered from SMBIOS or the DT on bare metal. Another suggestion was looking up an ID from the TPM.

Describe alternatives you've considered

An addition source for this could have been MAC addresses though these are sometimes also only randomly generated, have a lower bit size, and the drivers/devices may not be available at early boot.

The systemd version you checked that didn't have the feature you are asking for

255

[poettering]

¹ Something like: (sd_id128_get_machine_app_specific(<version>) & 0xFF) / 0xFF < threshold (I have never calculated with 128 bit numbers but the ratio should be correct. Also UUIDv4 has fixed bits but as we hash it we can disregard that.)

128bit arithmetic sucks in C. It's fine to just cut off the first 64bit however, or something like that and then do simple modulo on it

[poettering]

So yeah, I am on board with this idea. Key is that for the SMBIOS/DT thing the behaviour is opt-in, as you only want this in specific setups.

[septatrix]

So yeah, I am on board with this idea. Key is that for the SMBIOS/DT thing the behaviour is opt-in, as you only want this in specific setups.

I would be fine with having this as a special value for sd.machine_id (or a new kernel cmdline arg)