Authentication errors on NLB cluster

I configred a 2 node NLB cluster to load balance Remote Desktop Session Hosts with Windows Server 2008 R2. These were virtual servers running on VMWare so I selected to use multicast mode for the cluster. The cluster IP (.3) correctly resolved to the cluster multicast address with ARP. The cluster formed and converged successfully.

After a few minutes I could no longer access node 2 (.2) from any other computer. The error given by Windows was “The target account name is incorrect”. This is a Kerberos authentication error (KRB5KRB_AP_ERR_MODIFIED)., which means there is a mismatch between the name you entered in your request and the name of the server you are actually contacting. In other words, the name of the machine is different from the name you specified, yet you were sent to that machine by the networking stack. I also experienced another error when trying to remotely administering node 2; The RPC server is unavailable. How could this be?

Turns out the reason is IPv6 6to4. When both NLB nodes were given the same official IPv4 address as a secondary address on their adapters (the cluster address), they also both configured the corresponding IPv6 address for that IPv4 address on their 6to4 adapters. Both machines had the same IPv6 6to4 address. Once they both registered this in DNS the game was up. Whenever I typed in the name of node 2 in an UNC path it randomly resolved to the IPv6 address of node 1. When the connection was attempted on node 1 the actual name of node 1 was different from what was specified in the Kerberos ticket that the connecting machine got from the Domain Controller, thus producing the “mismatch” error, and in turn the RPC error.

To work around the error I disabled 6to4 on both NLB nodes:

netsh interface ipv6 6to4 set state disabled

This is not a perfect solution and I would much rather find a way to disable the registration of 6to4 addresses in DNS. I will have to look into this further…

Another approach would be to disable strict name checking on the hosts, in effect disabling the requirement that the name of the machine must match the name in the Kerberos ticket, but that is a major change with some serious security issues attached to it. Also the effect of having two machines with the same 6to4 address is unknown to me.

By the way, you might not see this error in your network or lab, since 6to4 addresses are only configured for machines with official IPv4 addresses. In other words, if you use any of the RFC1918 addresses (10/8, 172.16/12 or 192.168/16) you will not have 6to4 addresses.

One thought on “Authentication errors on NLB cluster”

  1. Holy moly, I’ve been looking for this solution! Thank you! Our darn NLB cluster for the Exchange 2010 CAS array has been cause me to pull out what little hair I have left.

    When I disabled IPv6, the cluster built right, but Exchange wouldn’t work right. With IPv6 enabled, the cluster would always throw these communication inconsistencies like Kerberos, RPC, and DNS problems.

    I tried so many things, but this actually ended up being a pretty elegant way to fix it. After disabling the 6TO4 adapters, re-adding the 2nd server to the cluster, and then restarting a couple Exchange services…all the errors went away.

Leave a Reply

Your email address will not be published. Required fields are marked *