Description of problem: 
         Configuring automembership plugin with "nsslapd-pluginConfigArea", fails to update entries in the replicated environment.


Version-Release number of selected component (if applicable): 389-ds-base-1.2.10.1-1.el6.x86_64 


How reproducible: Consistently.


Steps to Reproduce:
1. Install latest 389-ds-base(1.2.10) bits and configure four way MMR.
- Replicated Suffix - "dc=replAutoMembers,dc=com".
2. Configure Automembership plugin with "nsslapd-pluginConfigArea" on all the masters.
nsslapd-pluginConfigArea: dc=replAutomembers,dc=com
3. Add plugin configuration entries to "dc=replAutoMembers,dc=com" with regex rules.
Automembers plugin's regex rule:

dn: cn=replsubGroups,cn=autoMembersPlugin,dc=replAutoMembers,dc=com
objectclass: autoMemberDefinition
autoMemberScope: cn=Employees,dc=replAutoMembers,dc=com
autoMemberFilter: objectclass=posixAccount
autoMemberDefaultGroup: cn=SubDef1,dc=replAutoMembers,dc=com
autoMemberDefaultGroup: cn=SubDef2,dc=replAutoMembers,dc=com
autoMemberDefaultGroup: cn=SubDef3,dc=replAutoMembers,dc=com
autoMemberDefaultGroup: cn=SubDef4,dc=replAutoMembers,dc=com
autoMemberDefaultGroup: cn=SubDef5,dc=replAutoMembers,dc=com
autoMemberGroupingAttr: member: dn

dn: cn=Managers,cn=replsubGroups,cn=autoMembersPlugin,dc=replAutoMembers,dc=com
objectclass: autoMemberRegexRule
description: Group placement for Managers
cn: Managers
autoMemberTargetGroup: cn=Managers,cn=replsubGroups,dc=replAutoMembers,dc=com
autoMemberInclusiveRegex: uidNumber=^5..5$
autoMemberInclusiveRegex: gidNumber=^[1-4]..3$
autoMemberInclusiveRegex: nsAdminGroupName=^Manager$|^Supervisor$
autoMemberExclusiveRegex: uidNumber=^999$
autoMemberExclusiveRegex: gidNumber=^[6-8].0$
autoMemberExclusiveRegex: nsAdminGroupName=^Junior$

dn: cn=Contractors,cn=replsubGroups,cn=autoMembersPlugin,dc=replAutoMembers,dc=com
objectclass: autoMemberRegexRule
description: Group placement for Contractors
cn:  Contractors
autoMemberTargetGroup: cn=Contractors,cn=replsubGroups,dc=replAutoMembers,dc=com
autoMemberInclusiveRegex: uidNumber=^8..5$
autoMemberInclusiveRegex: gidNumber=^[5-9]..3$
autoMemberInclusiveRegex: nsAdminGroupName=^Contract|^Temporary$
autoMemberExclusiveRegex: uidNumber=^[1,3,8]99$
autoMemberExclusiveRegex: gidNumber=^[2-4]00$
autoMemberExclusiveRegex: nsAdminGroupName=^Employee$

4. Create all the associated groups and containers before creating user entries.
For eg: Groups - Contractors, Managers, SubDef1, SubDef2, SubDef3, SubDef4 and SubDef5.
Containers: cn=autoMembersPlugin, cn=replsubGroups and cn=Employees

5. Add few user entries(with posixAccount objectClass) to M1 and observe whether the entry is added as a member to the groups. 

Actual results: 
          The first user is added as member to the group but not the subsequent entries. Moreover the entries are not synced over to other Masters. At the same time the attempt to stop/restart the instance failed.

Expected results: 
          Plugin configuration should throw proper error messages if not configured properly.

Additional info:



Nathan's comment for the same problem:

There is a deadlock issue.  It appears that it's related to the way you have the config area overlapping with the data area.  This configuration is causing the plug-in to be called recursively.

The add operation for your test user comes in, which triggers the automembership plugin.  Since the add is in the config area, it first obtains a write lock to reload the config, which it then successfully releases.  It then obtains a read lock to check if the entry being added meets any of the automembership criteria in the config.  The entry does meet the criteria, so an internal operation is kicked off to update the group entry with the new member value (the read lock is still held at this time).  This internal operation is modifying a group entry that unfortunately is inside of your config area.  We need to treat this as a configuration change, so we attempt to get a write lock to reload the config.  Since this thread already holds a read lock, it deadlocks.  This will cause further automembership operations to deadlock the first time they attempt to do anything with the lock, with the eventual possibility of locking up all of the worker threads and making the server unresponsive to clients.

You should file a bug on this, though you should not see the issue if you correct the way you are using the config area.  Please provide exact steps to reproduce the issue in the bug.  I have attached the stack trace.