[PATCH v2 04/14] xf86Xinput: Modify API for server-managed fd support

[Mark Kettenis]

> Date: Tue, 04 Feb 2014 16:44:07 +0100
> From: Hans de Goede <hdegoede at redhat.com>
> 
> Hi,
> 
> On 02/04/2014 02:12 PM, Mark Kettenis wrote:
> >> Date: Tue, 04 Feb 2014 13:16:21 +0100
> >> From: Hans de Goede <hdegoede at redhat.com>
> >>
> >> The systemd-logind API uses separate major and minor numbers rather
> >> then a dev_t.
> > 
> > I don't think the design of the Xorg driver API should be blindly
> > copied from Linux-specific APIs.
> > 
> > The integration of systemd support in Xorg gives me an uneasy feeling.
> > For us OpenBSD developers systemd is an example of the sort of
> > complexity that we think is responsible for many of the security
> > issues in software.  I would feel much more at ease if it would be
> > firmly jailed in hw/xfree86/os-support/linux.
> > 
> >> Also I don't want to leak the platform specific dev_t type into
> >> public headers.
> > 
> > That is a valid point; having to include <sys/stat.h> everywhere would
> > probably be a bad idea.
> > 
> > Not quite convinced why you need to add major and minor to the Xorg
> > driver API though.  Since you have the file descriptor, you can easily
> > call stat(2) to find out.
> 
> We're getting the file-descriptor from logind on both init and vtenter,
> in order to get if from logind we need to pass it a major + minor, note
> that at this point we don't have a fd yet, so nothing to fstat.

Ah, so udev gives you the major/minor which you then feed to
systemd/logind, which passes back the file descriptor, which is then
handed over to the driver.  Thanks for the explanation.

Still, passing information around in the server inside a struct that's
part of the driver interface doesn't strike me as good design.

On that same subject, putting the systemd_logind_xxx() calls directly
into xf86NewInputDevice() function seems like a bad idea to me as
well.  I can see a use case for server managed file descriptors on
OpenBSD as well as it could be used to hide the privsep implementation
from the drivers.  But would you be happy if I'd sprinkle wscons calls
all over the core X parts?