[BusyBox] Endless "umount" corner cases: judgement call time...

Aurelien Jacobs aurel at gnuage.org
Fri Aug 5 01:29:23 UTC 2005


On Thu, 4 Aug 2005 08:59:00 -0500
Rob Landley <rob at landley.net> wrote:

> Okay, with 2.6 you can now happily mount the same block device in
> several  places and it acts a bit like multiple --bind mounts (they'll
> all share the  same superblock so nothing gets corrupted).
> 
> So, if you mount /dev/hda4 in three places and then umount
> /dev/hda4... Does  this mean all three places should be unmounted?  Or
> just the most recent one?   (Note you can specify a directory you want
> unmounted, and this is now a  different thing than specifying the
> block devices.  Unmounts now always try  to work on directories
> because the current internal logic that tries to deal  with a --bind
> mount and convert that to a block device is going to get really 
> confused.  Not that I'm ever quite sure what it expected to do
> umounting an  NFS mount or ramfs anyway...)
> 
> Anyway, I'm leaning towards "umount /dev/hda4" umounts all instances
> of hda4,  because if you wanted to specify umounting a specific
> directory you've have  indicated the directory.  But I'd like a show
> of hands about this.  (It  wasn't an issue back before bind mounts
> existed because you couldn't mount a  block device in two places and
> expect it to work back then.)

What util-linux mount version do is just unmounting the most recent one.
Now, wether you want to respect what this kind of "standard" version do
is another question.

Aurel



More information about the busybox mailing list