RE: Protocol: modification items may violate schema

[Timothy Hahn <hahnt@us.ibm.com>]

I suspect that not all pathological
cases were considered when the sentence was written.

Thinking through this, I'd say that
in light of "be lenient in what you accept and strict in what you
emit", I think it would acceptable for an implementation to ALLOW
such a case as is posed below - though I agree with you that supporting
such behavior can be problematic.

Of course, you'd want an implementation
to not go overboard on such support (thinking in terms of denial of service
types of "attacks" by sending servers valild LDAP protocol elements
for the server to "chew on" but which don't really do anything
useful - i.e. "silliness").  But I think this would result
in an UNWILLING_TO_PERFORM error code rather than an SCHEMA_VIOLATION error
code.

Regards,
Tim Hahn

Internet: hahnt@us.ibm.com
Internal: Timothy Hahn/Durham/IBM@IBMUS
phone: 919.224.1565     tie-line: 8/687.1565
fax: 919.224.2540

"Jim Sermersheim" <jimse@novell.com> Sent by: owner-ietf-ldapbis@OpenLDAP.org 09/05/2003 07:18 PM

To:        <ietf-ldapbis@OpenLDAP.org>         cc:                 Subject:        RE: Protocol: modification items may violate schema

Yes, this is what I'm trying to understand. Where RFC2251
says "violate directory schema" does it really mean *any* schema
violation, or is it talking only about those that would cause objectClassViolation?
Jim

>>> "Howard Chu" <hyc@highlandsun.com> 9/5/03
5:09:52 PM >>>
 
 
-----Original Message-----
From: owner-ietf-ldapbis@OpenLDAP.org [mailto:owner-ietf-ldapbis@OpenLDAP.org]On
Behalf Of Jim Sermersheim
Sent: Friday, September 05, 2003 3:00 PM
To: ietf-ldapbis@OpenLDAP.org
Subject: Re: Protocol: modification items may violate schema

Let me pose some silliness and wonder at how to deal with
it.
 
An attribute called X is of syntax integer
 
A modify adds the value "1#qrs#5" to X, then
deletes the value "1#xyz#5" from X
 
The server tries to evaluate the two values being updated
on X. Neither of them are integers, thus it can't use its normal compare
routines to check for equality. I suppose the server could (in the event
of syntax violations) fail over to using an exact octet match.
 
While the server is gathering up these modifications (in
order to later commit them), I guess it can't hold the attribute values
in a syntax-based data structure, because it can't rely on the values adhering
to the syntax.
 
I think there's something even more ugly, but I can't
think of it now...
 
Jim

>>> Timothy Hahn <hahnt@us.ibm.com> 9/5/03 1:25:28 PM >>>

Jim, 

I had always interepreted this as indicating that even if "silly things"
are done in some of the modificationItems, that the server shouldn't necessarily
fail the modification because an "intermediate result" was not
compliant to the schema definitions. 

Re-stated, an entry in the process of being modified (within an atomic
operation) is allowed to be non-compliant to the schema, but by the end
of applying all the modifications (at the end of the atomic operation)
the entry MUST be schema compliant. 

That's how I always interpreted it. 

Regards,
Tim Hahn

Internet: hahnt@us.ibm.com
Internal: Timothy Hahn/Durham/IBM@IBMUS
phone: 919.224.1565     tie-line: 8/687.1565
fax: 919.224.2540

"Jim Sermersheim" <jimse@novell.com> Sent by: owner-ietf-ldapbis@OpenLDAP.org 09/05/2003 02:45 PM

To:        <ietf-ldapbis@OpenLDAP.org>        cc:        "Duane Buss" <DBuss@novell.com>, "Vithalprasad Gaitonde" <GVithalprasad@novell.com>, "Joseph Cook" <JCook@novell.com>, "Jim Schnitter" <JSchnitter@novell.com>, "Mike Cronquist" <MCRONQUIST@novell.com>, "Richard Ellis" <RELLIS@novell.com>, "Steve McLain" <SMCLAIN@novell.com>, "Parameswaran S" <SPARAMESH@novell.com>, "Susan Perrin" <SPERRIN@novell.com>        Subject:        Protocol: modification items may violate schema

All,

An issue has been raised regarding the language in 4.6:

"While indi! vidual! modifications may violate the directory schema,
the
resulting entry after the entire list of modifications is performed MUST
conform to the requirements of the directory schema."

While I believe most server implementors understand this to mean that
while any one modification item may cause an object class violation, the
entire set of modifications must be evaluated an no error returned if
the net effect would not violate schema.

For example, in one test it is expected that given the entry
(singleValAttr is a SINGLE-VALUE attribute):

dn: dc=example
objectclass: x
singleValAttr: val1

the following modify should not error:
dn: dc=example
changetype: modify
add: singleValAttr
singleValAttr: val2
-
delete: singleValAttr
singleValAttr: val1

While it's obvious that the net effect is to have a single value (val2)
in singleValAttr, because the order is backward, a server doesn't like
it.
! 
! Other examples might include:
- Deleting a distinguishe! d ! name v a lue and re-adding the same value
- Adding an attribute value of invalid syntax and then removing it.
- Adding a value that would produce contraintViolation, then removing
it.

Regarding the statement above: I've always read this as a requirement
on the data being presented to the server--that at least the resulting
entry MUST conform. I haven't (until recently) thought about it as a
requirement that the server allow silly things to be present in a list
of modification items.

What do others feel? Is there a need for clarity here? Are these kinds
of tests valid indicators of standards-compliance?

Jim

I'm surprised this discussion is still going on. X.511 is pretty clear on this - the result must conform, but it is necessary to allow intermediate steps to violate the schema

On the other hand, Jim's example with the bogus values for Add/Delete of an Integer attribute should fail immediately due to Syntax violation. The point is that temporary ObjectClass violations are allowed, so long as the final object conforms to the requirements of its ObjectClass, but other violations return errors immediately.

Also, from one of Jim's other examples: removing an attribute or value that is part of the entry's RDN is explicitly not allowed.

   -- Howard Chu
 Chief Architect, Symas Corp.       Director, Highland Sun
 http://www.symas.com               http://highlandsun.com/hyc
 Symas: Premier OpenSource Development and Support

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature