Re: [Utopia] gnome-mount 0.3 is out
- From: Kay Sievers <kay sievers vrfy org>
- To: David Zeuthen <david fubar dk>
- Cc: utopia-list gnome org, Martin Pitt <martin pitt ubuntu com>
- Subject: Re: [Utopia] gnome-mount 0.3 is out
- Date: Thu, 12 Jan 2006 17:17:08 +0100
On Thu, Jan 12, 2006 at 11:03:45AM -0500, David Zeuthen wrote:
> On Thu, 2006-01-12 at 16:18 +0100, Kay Sievers wrote:
> > > But today all privileged access to a device happens in a helper called
> > > out from hald, no? That's in hald/linux2/probing and elsewhere.
> >
> > Yes, most of it. Who will start add-ons that need privileges?
>
> The helper process of hald running will be running as uid 0.
So, what will spawn a addon for a new detected device, if the main
daemon does not have the needed privileges?
> > > There may be a few cases (reading battery info comes to mind) where we
> > > need to clean this up too; that's all part of the work of separating
> > > hald into two processes - the unprivileged one handling D-BUS requests
> > > and the uid 0 one that executes helpers. Don't you think this is nicer,
> > > we get less code running at uid 0 which is always good even if there are
> > > no real threats (still I'm waiting for Martin to point those out).
> >
> > Sure, it's nicer, I never disagreed, but I didn't see a convincing
> > proposal that still works as expected. If we are going to do that than
> > we should reconsider my old proposal, to make the hal device store
> > generic and not part of the device handling. You didn't like the ipc
> > overhead that time, which is what this privilege split model will
> > introduce anyway.
>
> Well we can still do that at some point if you convince me it's a good
> idea ;-)
It's still valid, sure. If we go doing all ipc, that sounds obvouis to
me to do it.
> > I'm still can't really imagine, what will be left to the "unpriv.
> > main daemon" if we put everything into external privileged processes.
> > Then we can just rip out the store and make it generic for other
> > subsytems too, to put in their objects too and use all the nice
> > infrastructure like fdi files, callouts, ...
>
> Along the way that might be nice. But one task left for the unprivileged
> daemon is also reading sysfs stuff.
Yeah, you can spawn a setuid binary for every file to read. :)
Kay
[
Date Prev][
Date Next] [
Thread Prev][
Thread Next]
[
Thread Index]
[
Date Index]
[
Author Index]