[enfs-devel] Stacking file system discussion

[Hudson, T. B. <Tra...@ce...>]

Lee and I discussed stacking file systems this afternoon and arrived at a
decision to avoid the complexity of allowing multiple paths the underlying
vnodes.  This allows us to hold onto vnodes from the underlying file system
without worrying about them getting reclaimed by another path into the
filesystem.

The implementation for mounting is also straight forward -- we use the
argc+argv that is passed into the mount routine to encode the arguments for
this mount, followed by the arguments for the underlying mount.  The
top-level fileystem calls the underlying mount routine on the current vnode
and then covers the vnode with its own filesystem.

This way namei (and v_lookup) traverse the cover list to track down .. from
a top level mount.

The interface for the mount command might look like this:

mount /foo/bar/mountpoint -t layer1 -- -t layer2 -o ro -- -t nfs
server:/export/enfs

This does change the semantics of the mount command, but hopefully in a
manner to make it more flexible.

Trammell
-- 
H: hu...@sw...               240 476 1373
W: Tra...@ce...    240 453 3317