[RFC,PATCH 0/4] Dynamic Pseudo Root

[Neil Brown]

On Friday December 7, bfields at fieldses.org wrote:
> On Fri, Dec 07, 2007 at 05:23:37PM -0500, Steve Dickson wrote:
> > A concern I have with this implement, is how scalable it is? Do server 
> > export hundreds or even thousands of directories? I don't know...  If they do, 
> > then creating a mount point for each export probably will not scale very well. 
> 
> Yeah, we'll need some sort of "automounting" ability.  Neil proposed
> using special symlinks of some kind.

I do vaguely remember saying something about symlinks in this
context.  I don't remember exactly why, and it was probably a silly
idea....

However I suspect it also isn't necessary any more.
Some months ago linux gained a concept of share mount trees and mount
propagation. 
The result - as I understand it, which is only superficially - is that
you can --bind mount a directory from A to B, and anything else that
appears under A will automagically appear under B.  Does this address
the potential problem that you are seeing (I'm not sure from the above
exactly what your perceived issue is)?

As for scalability - the best thing to do is try it.  As I understand
it, Linux --bind mounts are quite light-weight and having thousands of
them may not be a problem at all.

> 
> I assume the only reason that would happen is by accident.  But we
> should make sure it doesn't make things any worse than it has to.

If we need to ensure there is only one mountd (which sounds like a
good idea anyway), it is hardly difficult to achieve that (lock files
etc).

The idea of having mountd in a private namespace is not one that
appeals to me.  It is complexity from the perspective of the sysadmin
and that should be avoided if possible.
So if there was a clear need for a separate namespace, then I wouldn't
object.  But until that need were demonstrated (and documented:-) I
wouldn't be keen on the idea.

NeilBrown