Comment 5 for bug 11561

On Fri, Dec 31, 2004 at 10:03:53PM +0100, Marco d'Itri wrote:
> On Dec 31, Daniel Jacobowitz <email address hidden> wrote:
>
> > If devfs is mounted, is there any point in making udev try to take over
> > /dev? I had to uninstall udev and gnome-desktop-environment to get it to
> Not sure. So far it has been enabled on devfs systems too, but only
> after a reboot, and I do not see a good reason to not do it.
> I do not understand why postinst fails on your system, apparently it
> happens when "ln /tmp/$TEMPDEV/fd /proc/self/fd" is run.
>
> > It leaves my devfs moved to /.dev, a messed up tmpfs mounted on /dev (I'm
> > not sure why - it returned "/dev: Not a directory") and a garbage tmpfs
> > mounted in /tmp/fileOaRTAz. This means no ptys available.
> I can't see how this could be possible, the commands which do this are
> executed *after* ln:
>
> # -n is used because mount gets very confused by --bind (see #282205)
> mount -n --bind /dev/pts/ $TEMPDEV/pts/
> mount -n --bind /dev/shm/ $TEMPDEV/shm/
> mount -n --bind /dev /.dev
> mount -n --move $TEMPDEV /dev
>
> I think that your description of the events is not fully correct.
> Do you mind trying again? :-)

It appears to be a problem with tmpfs in 2.6.10-rc1, maybe. Witness:

root@nevyn:/home/drow# mount -t tmpfs tmpfs /mnt/spazmount
root@nevyn:/home/drow# ls /mnt/spazmount
ls: /mnt/spazmount: Not a directory

Aha:
# CONFIG_TMPFS is not set

I don't know why the tmpfs mount succeeds even though tmpfs is
disabled. tmpfs also appears in /proc/filesystems. That's really
nasty! It has something to do with the fact that tmpfs is used for
in-kernel filesystems.

Should we be detecting this problem?

--
Daniel Jacobowitz