Re: XFS allows expansions, but no contraction?

[Andi Kleen <ak@xxxxxxx>]

> Probably an extension of the xfs_fsr approach would be the way to
> go. One major issue is it is impossible to do this without changing
> inode numbers. Some applications rely on inode numbers remaining a
> constant.

But none I know do this over umount/remount (ok except for NFS stateless 
filehandles, but that can be cured by remounting on the client too) 

So when you limit yourself to doing the shrinking offline (which I think
would be ok for most users) then this would be no problem.

I guess adding hot-shrinking would be a locking nightmare anyways,
because one would need to add checks everywhere in the fs that the blocks
it just allocated didn't go away ...

-Andi