On Mon, 12 Mar 2001, Todd Vierling wrote:

# On Mon, 12 Mar 2001, Greywolf wrote:
#
# : # Additionally, something needs to *read* the ACLs for the purposes of
# : # manipulation and listing, and stashing them in the vnode may not be
# : # appropriate (it might be variably sized, and it's a space waste)
# :
# : ...which is why you have a POINTER to the acl.
#
# That's not quite my point.  Having the acl's stashed in-memory should be
# measured against storing them on-disk; if variably sized, you'll fragment
# kmem, and if fixed-size, you may end up using twice as much memory for
# acl-attached vnodes.  It *might* be more useful to fetch the ACL data on an
# as-needed basis, based on that.

Duh, you said vnode, and I was thinking "inode".  Sorry.

I'd think you'd still end up caching the ACL as part of the file's
metadata; that is, after all, the whole point of the cache; otherwise, you
lose on performance.  This is starting to get complex and we've not even
materialized concept yet, let alone code! :-)

Of course, if the kernel is compiled sans ACLs, nothing gets loaded...

[anyone remember the mythical thing of arrays of pointers to functions
returning pointers to arrays of cans of worms?  I believe we've just
found it!]

# (Just a suggestion.)


				--*greywolf;
--
*BSD: The power to serve, also.