Re: fairly complex nested filesystems question

[chris barry <Christopher.Barry@qlogic.com>]

On Fri, 2007-12-21 at 23:37 +0900, Ian Kent wrote:
> On Fri, 2007-12-21 at 09:13 -0500, chris barry wrote:
> > On Fri, 2007-12-21 at 20:44 +0900, Ian Kent wrote:
> > > On Thu, 2007-12-20 at 18:29 -0500, chris barry wrote:
> > > > Greetings,
> > > > 
> > > > I'm trying to create a nested filesystem structure that combines an
> > > > automounted nfs home directory, with bind mounts from the local machine.
> > > > 
> > > > The structure will look like this:
> > > > /
> > > > |-home
> > > >    `-user1        <-(nfs somehost:/home/user1)
> > > >        |-myftp    <-(bind /var/ftp/users/user1)
> > > >        `-pub_ftp  <-(bind /var/ftp/pub)
> > > > 
> > > > The entire structure needs to get automounted as a group on login, and
> > > > then is chrooted as users ftp into this box.
> > > > 
> > > > My auto.master looks like this:
> > > > /home   /etc/auto.home --timeout=30
> > > > 
> > > > My auto.home looks like this:
> > > > *              -fstype=nfs,rw,intr   10.20.30.40:/home/& \
> > > > /home/&/myftp  -fstype=file          :/var/ftp/users/& \
> > > > /home/&/ftp    -fstype=file          :/var/ftp/pub
> > > 
> > > This is not valid for a couple of reasons.
> > > First the * will match every key and so nothing else will be seen.
> > > The second and third entries aren't valid map entries for an indirect
> > > map either. Even if they were then the & can't be used on the left hand
> > > side of a map entry. This is because the & is replaced by the key that
> > > has been matched and so can't be used within the key itself. The -fstype
> > > should refer to a valid filesystem but I don't think file is a valid
> > > filesystem. If you're intending this to indicate a bind mount then just
> > > leave out the -fstype=file and leave the :<path>. 
> > > 
> > > Ian
> > > 
> > 
> > Ian,
> > 
> > Thanks so much for your response. I got that one wrong totally wrong. 
> > 
> > Given the drawing and intent though, is the idea itself even possible,
> > or is it simply not workable in any way? If it /is/ possible, any idea
> > on how it should be done? From your explanation, I'm (taking a blind
> > stab here) thinking something like:
> > 
> > *    -fstype=nfs,rw,intr   10.20.30.40:/home/& \
> >      ./myftp                 :/var/ftp/users/& \
> >      ./ftp                   :/var/ftp/pub
> 
> Hehe, that's not quite right either.
> Sorry for the curt e-mail, I'm a bit loaded up at the moment.
> 
> We didn't get the autofs version you're using for this either.
> 
> That's interesting use of the wildcard key, I'm not sure whether that
> would work but, in theory, it just might.
> 
> So a map would need to be something like:
> 
> *	/	-fstype=nfs,rw,intr   10.20.30.40:/home/& \
> 	/myftp  	               :/var/ftp/users/& \
> 	/ftp    	               :/var/ftp/pub
> 
> The initial "/" (the root of the multi-mount) is allowed to be omitted
> as an accepted shortcut. The "." isn't legal the way it is specified
> above.
> 
> Strictly speaking the mount-point directories (such
> as /home/<someuser>/myftp and /home/<someuser>/ftp) must already exist
> as autofs doesn't, and often can't (and should'nt be able to) create
> directories on a remote server as it's operating as root.
> 
> Ian
> 
> 
(putting back on the list)

Thanks Ian! I'll give that a go.

Cheers,
-C