Re: mountpoint visibility

[aurelien naldi <aurelien resus univ-mrs fr>]

Le mercredi 15 décembre 2004 à 10:11 +0100, Alexander Larsson a écrit :
> On Mon, 2004-12-13 at 14:32 +0100, aurelien naldi wrote:
> > hello,
> > 
> > I'd like to start a discussion about visibility of mountpoint.
> > 
> > The main point is that I'm not sure users should see devices mounted by
> > other users.
> > 
> > this is not a common case, personnal computers are generally used by ONE
> > person at a time but let's think about a XDMCP server (I'm working on
> > such a system). When using a thin X client as workstation, the main
> > discomfort is to be unable to use removable media like floppy or usb
> > disk. That's why we are working on making them avaible as transparantly
> > as possible : on the thin client a script is launched by udev and inform
> > the XDMCP server of devices being plugged, the XDMCP server will then
> > pmount them through nbd as the user logged from this thin client.
> > 
> > Everything can be like on a real single system, except that if 10 users
> > have their usbdisk mounted, then they will see all mountpoint and it
> > becomes quite annoying.
> > 
> > the attached (yet very simple) patch to gnome-vfs prevents this: the
> > mountpoint is marked as visible ONLY if the user is the owner of the
> > mountpoint
> > 
> > it's not intended to be a perfect solution, I'd just like to discuss on
> > it, and get some feedback.
> > of course I secretly hope this feature will make it into gnome-vfs (or
> > somewhere else in gnome if it's not the right place?), so that I could
> > later upgrade to gnome2.10 without having to patch gnome-vfs again and
> > again :))
> 
> In general i think this makes sense. If someone else mounts it its
> unlikely that you're interested in the mount. However, I'm not sure the
> check is good enough in general. I mean, what about non-fat mounts that
> actually store the uid/gid, then stat:ing it might not give you your
> uid. 
> 
> Also, i'm a bit afraid of doing actual stats on random locations, it has
> historically caused problems on things like NFS mounts that are down
> (check the fork and stat horror we do in gnome-vfs-unix-mounts.c) .
> Perhaps its safe in this case though...

thanks for your interest, as I said I'm pretty unsure about the
implementation, BTW we are going to use it in this state for now (in our
system this can't fail on a disconnected NFS). I really hope to see this
integrated anyway (we are using an "old" gnome2.6 for now but we will
probably upgrade with ubuntu hoary). I can rework the patch to avoid
waiting for NFS timeout if you think it's necessary

to make it fully working we also need to change pmount policy, do you
know where to propose this kind of patch ?


PS: I subscribed to the list, no need to Cc me

regards
-- 
aurelien