Linux Archive

Linux Archive (
-   Fedora Directory (
-   -   Problems with chaining (was: Sanity check for install approach) (

Jonathan Boulle 07-30-2010 04:18 PM

Problems with chaining (was: Sanity check for install approach)
I've done a colossal amount of reading and experimentation to remedy the ignorance demonstrated in my last email, so let's start again:

I have, in a simple two-node (one supplier, one consumer) setup:
- Consumer set up with one local database (replicated from the supplier), + one database link back to the supplier
- Suffix dc=example using both of these databases, + replication plugin (with repl_chain_on_update) enabled, + suffix request processing set to "use the databases" (i.e. no referrals)
- Database link on the consumer using cn=replication manager,cn=config; an ACI on the supplier of (targetattr = "*")(version 3.0; acl "proxy acl"; allow(proxy) userdn="ldap:///cn=replication manager,cn=config";)

So, on a client which is pointed towards the consumer, let's try modify something:

[bloggsj@client02 ~]$ ldapmodify -x -ZZ -D uid=bloggsj,ou=people,dc=example -w example
dn: uid=bloggsj,ou=People,dc=example
changetype: modify
replace: title
title: Sir

modifying entry "uid=bloggsj,ou=People,dc=example"


On the supplier I see the relevant bind requests (as replication user *and* as the bloggsj user, i.e. successfully proxied), and also that the change succeeds (I turned on some ACL logging):

[30/Jul/2010:17:10:03 +0100] NSACLPlugin - proxied authorization dn is (uid=bloggsj,ou=people,dc=example)
[30/Jul/2010:17:10:03 +0100] NSACLPlugin - #### conn=26 op=2 binddn="cn=replication manager,cn=config"
[30/Jul/2010:17:10:03 +0100] NSACLPlugin - #### conn=26 op=2 binddn="cn=replication manager,cn=config"
[30/Jul/2010:16:56:27 +0100] NSACLPlugin - conn=20 op=2 (main): Allow write on entry(uid=bloggsj,ou=people,dc=example).attr(title ) to proxy (uid=bloggsj,ou=people,dc=example): allowed by aci(51): aciname= "Enable self write for common attributes", acidn="dc=example

Great. Works fine, as expected. But if I try to change the userPassword, for some reason the database link does _not proxy the authorisation_, but instead uses the replication user to try make the change:

[bloggsj@client02 ~]$ ldapmodify -x -ZZ -D uid=bloggsj,ou=people,dc=example -w example
dn: uid=bloggsj,ou=People,dc=example
changetype: modify
replace: userPassword
userPassword: abc123!@#

modifying entry "uid=bloggsj,ou=People,dc=example"
ldapmodify: Insufficient access (50)
additional info: Insufficient 'write' privilege to the 'userPassword' attribute of entry 'uid=bloggsj,ou=people,dc=example'.


On the supplier:

[30/Jul/2010:17:08:13 +0100] NSACLPlugin - #### conn=24 op=2 binddn="cn=replication manager,cn=config"
[30/Jul/2010:16:57:09 +0100] NSACLPlugin - conn=21 op=2 (main): Deny write on entry(uid=bloggsj,ou=people,dc=example).attr(userP assword) to cn=replication manager,cn=config: no aci matched the subject by aci(53): aciname= "self", acidn="dc=example"

Why do userPassword changes not proxy?

I have tried
- enabling/disabling "ACL Plugin" component to chain
- enabling/disabling "password policy" component to chain
- enabling/disabling various LDAP controls to chain
- other attributes (homephone, surname, etc), all of which proxy fine
to no avail.

Rich (or anyone else :-), do you have any ideas? This really looks like a bug to me - I haven't started poking around in the source yet though.

-----Original Message-----
From: [] On Behalf Of Jonathan Boulle
Sent: 29 July 2010 14:54
To: 'General discussion list for the 389 Directory server project.'
Subject: Re: [389-users] Sanity check for install approach

On closer examination of the doc, it appears that chaining updates is only possible when using database links.
However, as I infer, using database links removes the possibility of replication, because the link would pass any modification back to the remote database.
Thus, if you had a consumer configured with a database link back to a supplier, and then set up a replication agreement from the supplier to the consumer, it would be replicating to its own database!
Am I understanding this correctly?

Is there a way to achieve our desired scenario: where no clients can directly access a read-write supplier (i.e. referrals are disabled, because network access is blocked); but they're still able to change their passwords, because the read-only consumer chains the update request back to a supplier?


__________________________________________________ ______________________
In order to protect our email recipients, Betfair Group use SkyScan from
MessageLabs to scan all Incoming and Outgoing mail for viruses.

__________________________________________________ ______________________
389 users mailing list

All times are GMT. The time now is 07:54 AM.

VBulletin, Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Content Relevant URLs by vBSEO ©2007, Crawlability, Inc.