Thanks, I'm compiling it now.
> The -6 patch fixes a counter problem when traversing "." or "..", which
> resulted in incorrectly incrementing the counters twice. -6 also fixes
> unmounting, so it should work out-of-the-box (but only unmounts, not
> re-mounts read-only).
Yes, I saw the changes to fs_may_umount. But isn't it good to sync the
device and invalidate inodes after doing the shrink_dcache()? Seems
like there might be dirty inodes and it would be better to sync before
unmounting.
Will report back soon on any problems ...
-Bill