Re: Fwd: New Version Notification for draft-wouters-edns-tcp-keepalive-00.txt (fwd)

In message <[email protected]>, Mark Andrews writes:
> 
> In message <[email protected]>, Paul Wouters
>  w
> rites:
> > On Sat, 19 Oct 2013, Mark Andrews wrote:
> > 
> > >>> An RFC 5966 server supports persistent TCP but does not advertise this
> > >>> fact.
> > >>
> > >> I hope to have some measurements done in time for Vancouver to see what
> > >> the percentage of queries/servers is where this is a problem. That could
> > >> help assess whether or not we still need a special indication via edns
> > >> or not.
> > >
> > > Why?  Clients should be able to handle the connection being closed
> > > unexpectedly and retry any uncompleted queries.
> > 
> > I'm trying to reduce dns(sec) latency, not adding more repeating tcp
> > failures. If it turns out that most obtained DNS resolvers are useless
> > for TCP than there is a need to give a clear indication to a client
> > that a resolver does work well.
> > 
> > The question is how pervasive are rfc 5966 compliant resolvers out
> > there, and whether there is a need to advertise those who are.
> > 
> > Paul
> 
> All DNS servers should already handle multiple questions over a TCP
> socket.  This is *basic* RFC 1035.  Just send the queries and deal
> with the fact that you may have broken servers. 
> 
> Adding a new option often the worst thing you can do to fix a failure
> to follow specification.  It is definitely not needed to fix this
> problem.
> 
> Just write a RFC with a title like "Common DNS server errors in
> handling queries over TCP" which points out that DNS servers are
> expected to handle multiple queries over TCP.  Then a user has a
> second clue bat to go beat the vendor over the head with.
> 
> Mark
> -- 
> Mark Andrews, ISC
> 1 Seymour St., Dundas Valley, NSW 2117, Australia
> PHONE: +61 2 9871 4742                 INTERNET: [email protected]
> _______________________________________________
> dnsext mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/dnsext

Just so you can have a something to test with I hacked dig to keep
the tcp socket open.  It needs more work to be production quality.

	dig @server +tcp -f querylist

Now I know named handles multiple queries over a socket.
Perhaps others can report what works for them.

Mark

diff --git a/bin/dig/dighost.c b/bin/dig/dighost.c
index 79b11d0..a10742a 100644
--- a/bin/dig/dighost.c
+++ b/bin/dig/dighost.c
@@ -120,7 +120,8 @@ isc_boolean_t
 	usesearch = ISC_FALSE,
 	showsearch = ISC_FALSE,
 	qr = ISC_FALSE,
-	is_dst_up = ISC_FALSE;
+	is_dst_up = ISC_FALSE,
+	keep_open = ISC_TRUE;
 in_port_t port = 53;
 unsigned int timeout = 0;
 unsigned int extrabytes;
@@ -151,6 +152,8 @@ static void		idn_check_result(idn_result_t r, const char *msg);
 int  idnoptions	= 0;
 #endif
 
+isc_socket_t *keep = NULL;
+
 /*%
  * Exit Codes:
  *
@@ -2339,6 +2342,16 @@ send_tcp_connect(dig_query_t *query) {
 	}
 
 	INSIST(query->sock == NULL);
+
+	if (keep != NULL) {
+fprintf(stderr, "REUSING socket\n");
+		sockcount++;
+		isc_socket_attach(keep, &query->sock);
+		query->waiting_connect = ISC_FALSE;
+		launch_next_query(query, ISC_TRUE);
+		goto search;
+	}
+	
 	result = isc_socket_create(socketmgr,
 				   isc_sockaddr_pf(&query->sockaddr),
 				   isc_sockettype_tcp, &query->sock);
@@ -2361,6 +2374,7 @@ send_tcp_connect(dig_query_t *query) {
 	result = isc_socket_connect(query->sock, &query->sockaddr,
 				    global_task, connect_done, query);
 	check_result(result, "isc_socket_connect");
+ search:
 	/*
 	 * If we're at the endgame of a nameserver search, we need to
 	 * immediately bring up all the queries.  Do it here.
@@ -2725,6 +2739,12 @@ connect_done(isc_task_t *task, isc_event_t *event) {
 		UNLOCK_LOOKUP;
 		return;
 	}
+	if (keep_open) {
+fprintf(stderr, "SAVING socket\n");
+		if (keep != NULL)
+			isc_socket_detach(&keep);
+		isc_socket_attach(query->sock, &keep);
+	}
 	launch_next_query(query, ISC_TRUE);
 	isc_event_free(&event);
 	UNLOCK_LOOKUP;
@@ -3541,6 +3561,9 @@ destroy_libs(void) {
 	isc_result_t result;
 #endif
 
+	fprintf(stderr, "CLOSING socket\n");
+	if (keep != 0)
+		isc_socket_detach(&keep);
 	debug("destroy_libs()");
 	if (global_task != NULL) {
 		debug("freeing task");
-- 
Mark Andrews, ISC
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: +61 2 9871 4742                 INTERNET: [email protected]
_______________________________________________
dnsext mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/dnsext