[Ksummit-2012-discuss] Proposed discussion: Kernelising D-BUS

Arnd Bergmann arnd at arndb.de
Fri Jun 22 12:33:24 UTC 2012


On Thursday 21 June 2012, David Howells wrote:
> I propose that D-BUS should be done by the kernel rather than by a userspace
> daemon.  The problem is that how does the userspace daemon handle security?
> How does it guarantee to a packet recipient who the author of a message was?
> 
> Further, if service management gets kernelised and D-BUS is selected for
> message passing, how does the kernel interact with it?
> 
> Now, it's entirely possible that the security requirements are at least
> partially solved.  Possibly by putting labels on endpoints or using wibbly bits
> of AF_UNIX black magic.

I think this topic needs to take into account the other established
method we have that is like DBUS: The android binder.

The code we have living in drivers/staging/binder is not anywhere near
the quality we want to have for something that exists as the main
IPC method, but at the same time a lot of people agree we should have
something in the mainline kernel that can be used by Android user space
as a replacement for binder, without having to do too many changes to
existing Android applications.

Whatever we put into the kernel to solve this situation, I would rather
not have both ipc/dbus.c and ipc/binder.c provide the same functionality
but instead have something that can do both.

	Arnd


More information about the Ksummit-2012-discuss mailing list