In message <[EMAIL PROTECTED]>, Shaya Potter writes:
> On Tue, 2005-11-01 at 19:34 -0600, Brian Harring wrote:
> > On Tue, Nov 01, 2005 at 10:50:38PM +0000, Nix wrote:
> > > One moan: unionfs doesn't seem willing to cross mount points, which
> > > eliminates one major potential use:
> > > 
> > > mount -t unionfs -o dirs=/tmp/rwdir=rw:/=ro none /tmp/unionfs
> > > chroot /tmp/unionfs
> > > # do stuff, say a make install, in perfect safety
> > Crossing mount points doesn't bother me too much, what I've had issues 
> > with is any chroot'd unionfs / results in compilation failures I've 
> > not been able to track down (both 1.0.14 and 1.1.1).
> 
> i didn't see the message originally. 
> 
> No Fist based file system really crosses mount points (This came up
> years ago on the fistgen list).

Not quite.  It's an option if you want to cross or not:

- of you don't cross, you get unique and persistent inode numbers.

- if you do cross, you get persistent inodes, but they are not unique b/c
  each file system has its own inode number space.  So the solution here is
  to use iunique() to generate unique inode numbers, but now they're not
  persistent (which messes up NFS exports and various unix tools that depend
  on inode numbers).

The ultimate solution is what Dave implemented for Unionfs: a persistent
mapping of uniquely generated inode numbers, independent of the actual inums
of the files and file systems below.

Erez.
_______________________________________________
unionfs mailing list
[email protected]
http://www.fsl.cs.sunysb.edu/mailman/listinfo/unionfs

Reply via email to