kdbus and kernel keys

[Lennart Poettering]

On Thu, 16.01.14 12:34, Simon McVittie (simon.mcvittie at collabora.co.uk) wrote:

> 
> On 15/01/14 20:10, Kay Sievers wrote:
> > On Wed, Jan 15, 2014 at 8:38 PM, David Howells <dhowells at redhat.com> wrote:
> >> Can it be made possible to pass kernel keys around (ie. ones created with the
> >> add_key() syscall) with kdbus messages?
> >
> > Isn't that a bit like file descriptor passing what you look for?
> 
> Is there any possibility of the kernel letting these keys be passed
> around as fds? Then D-Bus, and anything else that uses fd-passing to
> transfer capabilities[1], automatically gets that feature with no
> additional design, maintenance or feature-discovery needed in D-Bus
> implementations. That seems a more graceful way to do it than
> introducing a new D-Bus type or header, #ifdef tests in implementations
> to see whether the OS/libc support the necessary underlying feature,
> runtime tests in implementations to see whether the running kernel
> supports it, and API to query support.
> 
> (You do need that whole mess to be able to ask "does this D-Bus
> implementation support fd-passing?", but we already have that - there
> seems no point in doing it again.)
> 
> I'm very much in favour of the "everything is a fd" trend in
> Linux<->userland API design - eventfd, kdbus memfds, inotify etc. - for
> that sort of reason.

Yeah, I fully agree with Simon. I mean, even DRM graphics objects
nowadays use fds as handles, and so on. In fact, afaik the keyring is
now backed by a memory file system to make it swappable, these days. It
would be great if this could be built on to allow fds as handles for
keys, which could then naturally be exchanged via both AF_UNIX and
kdbus...

Lennart

-- 
Lennart Poettering, Red Hat