TCP fast open support (linux only). Enabled with --enable-tcp-fastopen configure option.

configure

@@ -744,6 +744,7 @@ with_gnu_ld
 with_sysroot
 enable_libtool_lock
 enable_rpath
+enable_tcp_fastopen
 with_libidn
 with_libldns
 with_libunbound
@@ -1380,6 +1381,7 @@ Optional Features:
                           optimize for fast installation [default=yes]
   --disable-libtool-lock  avoid locking (might break parallel builds)
   --disable-rpath         disable hardcoded rpath (default=enabled)
+  --enable-tcp-fastopen   Enable TCP Fast Open
 
 Optional Packages:
   --with-PACKAGE[=ARG]    use PACKAGE [ARG=yes]
@@ -10788,6 +10790,28 @@ if test "x$enable_rpath" = xno; then
 fi
 
 
+# Check whether --enable-tcp-fastopen was given.
+if test "${enable_tcp_fastopen+set}" = set; then :
+  enableval=$enable_tcp_fastopen;
+fi
+
+case "$enable_tcp_fastopen" in
+	yes)
+		# TODO: Check if it is available (linux only)
+		{ $as_echo "$as_me:${as_lineno-$LINENO}: TCP Fast Open enabled" >&5
+$as_echo "$as_me: TCP Fast Open enabled" >&6;}
+
+cat >>confdefs.h <<_ACEOF
+#define USE_TCP_FASTOPEN 1
+_ACEOF
+
+		;;
+	no|*)
+		{ $as_echo "$as_me:${as_lineno-$LINENO}: TCP Fast Open not enabled" >&5
+$as_echo "$as_me: TCP Fast Open not enabled" >&6;}
+		;;
+esac
+
 # search to set include and library paths right
 # find libidn
 

configure.ac

@@ -79,6 +79,18 @@ fi
 ])
 ACX_ARG_RPATH
 
+AC_ARG_ENABLE(tcp-fastopen, AC_HELP_STRING([--enable-tcp-fastopen], [Enable TCP Fast Open]))
+case "$enable_tcp_fastopen" in
+	yes)
+		# TODO: Check if it is available (linux only)
+		AC_MSG_NOTICE([TCP Fast Open enabled])
+		AC_DEFINE_UNQUOTED([USE_TCP_FASTOPEN], [1], [Define this to enable TCP fast open.])
+		;;
+	no|*)
+		AC_MSG_NOTICE([TCP Fast Open not enabled])
+		;;
+esac
+
 # search to set include and library paths right
 # find libidn
 AC_ARG_WITH(libidn, AS_HELP_STRING([--with-libidn=pathname],

src/config.h.in

@@ -130,6 +130,9 @@
 /* Needed for sync stub resolver functions */
 #undef USE_MINI_EVENT
 
+/* Define this to enable TCP fast open. */
+#undef USE_TCP_FASTOPEN
+
 /* Define for Solaris 2.5.1 so the uint32_t typedef from <sys/synch.h>,
    <pthread.h>, or <semaphore.h> is not used. If the typedef were allowed, the
    #define below would cause a syntax error. */

src/stub.c

@@ -802,11 +802,29 @@ stub_tcp_write(int fd, getdns_tcp_state *tcp, getdns_network_req *netreq)
 		/* We have an initialized packet buffer.
 		 * Lets see how much of it we can write
 		 */
+#if defined (USE_TCP_FASTOPEN) && defined(MSG_FASTOPEN)
+		/* We use sendto() here which will do both a connect and send */
+		written = sendto(fd, pkt, pkt_len + 2, MSG_FASTOPEN,
+					(struct sockaddr *)&(netreq->upstream->addr),
+					netreq->upstream->addr_len);
+		/* If pipelining we will find that the connection is already up so 
+		   just fall back to a 'normal' write. */
+		if (written == -1 && errno == EISCONN) 
+			written = write(fd, pkt, pkt_len + 2);
+
+		if ((written == -1 && (errno == EAGAIN ||
+		                       errno == EWOULDBLOCK ||
+		/* Add the error case where the connection is in progress which is when
+		   a cookie is not available (e.g. when doing the first request to an
+		   upstream). We must let the handshake complete since non-blocking. */
+		                       errno == EINPROGRESS)) ||
+		     written  < pkt_len + 2) {
+#else
 		written = write(fd, pkt, pkt_len + 2);
 		if ((written == -1 && (errno == EAGAIN ||
 		                       errno == EWOULDBLOCK)) ||
 		     written  < pkt_len + 2) {
-
+#endif
 			/* We couldn't write the whole packet.
 			 * We have to return with STUB_TCP_AGAIN, but if
 			 * the packet was on the stack only, we have to copy
@@ -834,7 +852,7 @@ stub_tcp_write(int fd, getdns_tcp_state *tcp, getdns_network_req *netreq)
 
 	} else {/* if (! tcp->write_buf) */
 
-		/* Coming back from an earlier unfinished write.
+		/* Coming back from an earlier unfinished write or handshake.
 		 * Try to send remaining data */
 		written = write(fd, tcp->write_buf     + tcp->written,
 		                    tcp->write_buf_len - tcp->written);
@@ -998,12 +1016,16 @@ priv_getdns_submit_stub_request(getdns_network_req *netreq)
 			return GETDNS_RETURN_GENERIC_ERROR;
 		
 		getdns_sock_nonblock(netreq->fd);
+#if defined (USE_TCP_FASTOPEN) && defined(MSG_FASTOPEN)
+		/* Leave the connect to the later call to sendto() */
+#else
 		if (connect(netreq->fd, (struct sockaddr *)&upstream->addr,
 		    upstream->addr_len) == -1 && errno != EINPROGRESS) {
 
 			close(netreq->fd);
 			return GETDNS_RETURN_GENERIC_ERROR;
 		}
+#endif
 		netreq->upstream = upstream;
 
 		GETDNS_SCHEDULE_EVENT(
@@ -1025,6 +1047,9 @@ priv_getdns_submit_stub_request(getdns_network_req *netreq)
 				return GETDNS_RETURN_GENERIC_ERROR;
 			
 			getdns_sock_nonblock(upstream->fd);
+#if defined (USE_TCP_FASTOPEN) && defined(MSG_FASTOPEN)
+		/* Leave the connect to the later call to sendto() */
+#else
 			if (connect(upstream->fd,
 			    (struct sockaddr *)&upstream->addr,
 			    upstream->addr_len) == -1 && errno != EINPROGRESS){
@@ -1033,6 +1058,7 @@ priv_getdns_submit_stub_request(getdns_network_req *netreq)
 				upstream->fd = -1;
 				return GETDNS_RETURN_GENERIC_ERROR;
 			}
+#endif
 			/* Attach to the global event loop
 			 * so it can do it's own scheduling
 			 */