A BIND 9.x implementation configured as a caching name server must restrict recursive queries to only the IP addresses and IP address ranges of known supported clients.

Any host that can query a resolving name server has the potential to poison the server's name cache or take advantage of other vulnerabilities that may be accessed through the query service. The best way to prevent this type of attack is to limit queries to internal hosts, which need to have this service available to them.

To guard against poisoning, name servers authoritative for .mil domains must be separated functionally from name servers that resolve queries on behalf of internal clients. Organizations may achieve this separation by dedicating machines to each function or, if possible, by running two instances of the name server software on the same machine: one for the authoritative function and the other for the resolving function. In this design, each name server process may be bound to a different IP address or network interface to implement the required segregation.