ISC recently discovered an issue in BIND that allows TCP connection slots to be consumed for an indefinite time frame via a specifically crafted TCP stream sent from a client. This issue is present in BIND 9.16.11 to 9.16.26 (including S editions), and 9.18.0.
This issue can only be triggered on BIND servers which have keep-response-order enabled, which is not the default configuration. The keep-response-order option is an ACL block; any hosts which are specified within it will be able to trigger this issue on affected versions. Specifically crafted TCP streams can cause connections to BIND to remain in CLOSE_WAIT status for an indefinite period of time, even after the client has terminated the connection.