directory-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Alex Karasulu" <>
Subject [ApacheDS] Comments on Stored Procedure Implementation (was Re: [Roadmap]Apache Directory Server 2.0 Roadmap proposal)
Date Fri, 28 Sep 2007 19:19:17 GMT
On 9/28/07, Stefan Zoerner <> wrote:
> Alex Karasulu wrote:
> >     * Support for Scripting languages for Stored Procedures (e.g.
> Groovy)
> >
> >
> > This also would be really cool to have.  We just need more participation
> > on getting stored procedures
> > implemented properly.  It would be great if you could lend a hand if it
> > interests you of course.
> It seems that it is not much code we talk about. I am interested.

This is great Stefan.

Stefan you and others insterested in this stuff may not be interested in
these details below
or how they will be managed by the stored procedure subsystem in the
server.  Wanting to
do it with groovy should not have to force you to have to deal with these
details.  So please
feel free to skip the material below.  However at a minimum understanding
these problems
and their impact on you may help you a bit.

Incidentally I want to just list some of the issues we presently have on the
table regarding our
stored procedure implementation:

1). Security: Rights and Identity (owner verses invoker)

Stored procedures can be invoked by anyone if access to the SP is allowed.
The visibility of SP entries to
invoker principals can be controlled with ACI.  However if the invoker is
different from the owner then with
who's identity and authorization permissions do we execute the procedure?
This is a big question we've been
discussing for some time.  We can continue to discuss this on this thread if
people like but IMO the SP should
execute with the authorization permissions of the owner with the identity of
the invoker for auditing purposes.

How to do this is entirely another question that our implementation must

2). Security: Sandboxing SP Code

SP code is code which needs to be sandboxed so code with lesser rights does
not have the ability to
execute internal code with administrator like rights.  Today SPs in the
server can do whatever they
want no matter who invokes them.  This is why they are turned off by default
with the server as an
insecure experimental feature.  To take this to the next level we need to
sandbox SP code and allow
access to internal methods based on the rights and identity of the user.
This is a really difficult problem
to solve elegantly in ApacheDS as it stands today but it is a must.

3). SP Invocation: Procedure Qualification

Right now we have an ExtendedOperation used to invoke stored procedures
using qualified names based
on the namespace of the procedure language.  This is a questionable
mechanism since it requires the user
to know what the language of the implementation of the procedure is and how
to qualify the procedure in that
language.  If a universal call specification is devised it can link language
independent invocation requests to
the proper language specific method.  It would then be the job of the writer
to add this call spec and not the
job of the user to track which language the stored procedure was implemented

4). SP Invocation: Parameter Bindings

This is connected to 3.  The same extended operation allows invokers to
provide parameters to the stored
procedure.  It would not be fun to have SPs without parameters.  This opens
up a slew of problems for us.
First we need to define now a language independent means to specify
parameters.  ASN.1 does this nicely
so we can leverage ASN.1 primitive types at first.  ApacheDS can easily
transform these types into the types
expected by the implementation language of the SP.  This however requires
some primitive type bindings and
the code to translate them.

Right now the extended operation works with primitive Java types and
serialized objects (I think but Ersin can
confirm).  This makes it so a C client cannot really invoke a SP implemented
in Java unless it knows how to
serialize java objects into a byte buffer.  This is not really feasible so
we need to specify some type bindings
and use that instead of serializing objects.  We cannot expect invokers from
python, groovy, php, c, perl etc
to have to know anything about Java.  Invocation should not have be language
dependent to allow for maximum

I'm adding these details btw to a the first draft (version 00) of an IETF
draft submission.


View raw message