Hello Pavel:

You'll be pleased to learn that we've addressed this issue in our upcoming maintenance release, 4.3.6 due out July 31st, and in 4.4.0 release date is TBD.  

Your patch as supplied makes both the server and client perform "lazy" inits of the DNS client context/ports. This means that errors in server configuration, such as bad addresses for ddns-local-address<4/6> would not be detected until the first attempt to do an update. Furthermore, the error would recur continually with each subsequent attempt to do an update. In the current code, such an error is treated as fatal and server exits.

So we extended your patch with a new context creation flag, DHCP_DNS_CLIENT_LAZY_INIT, which enables the DNS client init code to be skipped in dhcp_context_create(). The server code does not set this flag, while the client code does.

In addition, the server code was modified to skip the second call to dhcp_context_create() with DHCP_CONTEXT_POST_DB on, unless the ddns-update-style is NOT "none". This allows the server to only init the DNS client if DDNS updating is globally enabled. This is in keeping with the man page recommendation for using "ddns-update-style none" to globally shut off DDNS updating (ddns-updates can be specified at multiple levels).

Lastly, the call setting the DHCP_CONTEDXT_POST_DB flag was removed from the call to dhcp_context_create() in dhcrelay as the relay has nothing to do with DDNS.  Rather than always open them, dhcrelay will now never open them.

Customarily, we recognize contributors in our release notes. If you'd like to be so recognized please let me know.  Thank you for your submission.

Sincerely,

Thomas Markwalder
ISC Software Engineering