'Machine ID' underspecified?

[Simon McVittie]

On Mon, 07 Dec 2020 at 08:10:52 +1300, Lawrence D'Oliveiro wrote:
> Seems like D-Bus is trying to hide the difference between IPC within a
> single machine versus that across different machines.

Yes and no. One of the original use-cases for D-Bus involved sharing
the session bus via TCP on a trusted LAN, in conjunction with
network-transparent X11 over TCP and an NFS-shared home directory.

This was designed in a more innocent time when trusted LANs were
considered to be something that was generally achievable and would not
immediately get subverted by connecting an untrusted laptop, phone,
Raspberry Pi, games console, smart fridge or similar, and the current
D-Bus maintainers consider it to be a bad idea.

> Dare I say this
> seems a bit ... misguided? Maybe this was part of the original concept
> for D-Bus, but as far as I know all the current use cases are
> restricted to IPC within a single machine.

I agree that this was misguided, at least for the two well-known buses.

The session bus is local to a single machine, except in special
circumstances (analogous to how X11 is local-only 99% of the time, with
the remainder being forwarded via ssh if you're paying any attention
to security).

The system bus is local to a single machine by definition.

It is possible (although rare) to use D-Bus in ways other than the
two well-known buses, and some of those could conceivably make sense
to use over TCP, although anyone doing this should be aware that D-Bus
over TCP has no integrity or confidentiality protection and has led to
a 1.4 million vehicle recall[1] (presumably rather expensive) on at least
one occasion.

    smcv

[1] https://www.wired.com/2015/07/hackers-remotely-kill-jeep-highway/