nscd (8) daemon caches negative responses for infinite time.
Parameter 'negative-time-to-live' in /etc/nscd.conf doesn't correct this behavior.

Fix: Supplied below corrects nscd behaviour on 9.1-RELEASE-p4 and HEAD:



if (read_response->error_code == -2) {
-               read_response->error_code = 0;
-               read_response->data = NULL;
-               read_response->data_size = 0;
+               read_response->data = malloc(
+                   read_response->data_size);
+               assert(read_response != NULL);
+               read_response->error_code = cache_read(neg_c_entry,
+                   read_request->cache_key,
+                   read_request->cache_key_size,
+                   read_response->data,
+                   &read_response->data_size);
            }
        }
        configuration_unlock_entry(qstate->config_entry, CELT_NEGATIVE);--gWC5PrrXOzg1QvWcCoB7krostOmUDgByPCTXeREr3OSdbDs2
Content-Type: text/plain; name="file.diff"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment; filename="file.diff"

--- usr.sbin/nscd/query.c.orig  2013-08-22 11:59:03.921956888 +0400
+++ usr.sbin/nscd/query.c   2013-08-22 12:00:01.515957178 +0400
@@ -743,9 +743,14 @@
                &read_response->data_size);
How-To-Repeat: Try set 'negative-time-to-live hosts 1' in /etc/nscd.conf and 'hosts: files cache dns' in /etc/nsswitch.conf.
Restart nscd: '/etc/rc.d/nscd onerestart'.
Now we can see output of 'tcpdump udp and port 53' and repeat many times: 'ping -c1 some.nonexistent'.
Only first time there will be actual dns search (with expected negative response).