Fix uncached CLIENT_RESPONSE'es on stateful transports

[wtoorop]

I noticed dnstap CLIENT_RESPONSE did not contain the response sent to the client when:

• the client is connected on a stateful transport, and
• the response did not come from cache.

When investigating the issue, I noticed that repinfo->c->buffer did not contain the response in this case (second half of comm_point_send_reply).
Only after tcp_req_info_send_reply() was called, the the buffer contains the response.

Is it safe to change this order?

[wcawijngaards]

No, it may be somewhere else, in a queue.

[wcawijngaards]

I think this may be better. Also buffer may no longer be there, because allocated in queue for tcp responses.

index 09023838..049c7ff0 100644
--- a/util/netevent.c
+++ b/util/netevent.c
@@ -3157,7 +3157,7 @@ comm_point_send_reply(struct comm_reply *repinfo)
                if(repinfo->c->tcp_parent->dtenv != NULL &&
                   repinfo->c->tcp_parent->dtenv->log_client_response_messages)
                        dt_msg_send_client_response(repinfo->c->tcp_parent->dtenv,
-                       &repinfo->addr, repinfo->c->type, repinfo->c->buffer);
+                       &repinfo->addr, repinfo->c->type, (repinfo->c->tcp_req_info?repinfo->c->tcp_req_info->spool_buffer:repinfo->c->buffer));
 #endif
                if(repinfo->c->tcp_req_info) {
                        tcp_req_info_send_reply(repinfo->c->tcp_req_info);

[wtoorop]

Yes that works! Thanks! I've done a new commit with your proposed fix.

[wcawijngaards]

Okay let's merge it, then.

[wtoorop]

Do you want me to push the merge button?

[wcawijngaards]

Sure! I don't know who should merge, but I just did.

Good to have this fixed.