2 Mar 2010 23:52
Re: Preliminary patch for libkadm5srv hooks
Russ Allbery <rra <at> stanford.edu>
2010-03-02 22:52:00 GMT
2010-03-02 22:52:00 GMT
Love Hörnquist Åstrand <lha <at> kth.se> writes:
> Yes, the idea was to have stacking, but do it though the lib/hdb/hdb.c
> layer where is doing all the processing, so that the individual layers
> are not ware of each other.
> I'm don't know how reasonable it is or how to handle errors just that
> api seemed very much the same.
Well, it's a very tiny subset. For my purposes, even including deletions
and renames, I'd only want to implement:
hdb_store
hdb_remove
hdb_password
The semantics are also a little wrong in that I'd like to distinguish
between a modify that only updates flags, a creation of a new entry, and a
password change, which hdb_store doesn't really let me do. It would also
be nice to have a principal rename as a first-class operation rather than
as a fetch, remove, and store, which I assume is how it's implemented now
(I haven't gone and looked). And as mentioned, for an hdb_store that
creates a new principal, I'd need the actual password and not just the
keys.
Stacked databases are a nice general solution that could be made to solve
this problem as well, but they'd be a bit more complicated and I think the
way in which the rest of Heimdal calls into the database layer would have
to be changed to provide the password rather than the keys in more places.
(Continue reading)
RSS Feed