RDS: TCP: Avoid rds connection churn from rogue SYNs
authorSowmini Varadhan <sowmini.varadhan@oracle.com>
Wed, 18 May 2016 17:06:24 +0000 (10:06 -0700)
committerDavid S. Miller <davem@davemloft.net>
Fri, 20 May 2016 23:19:57 +0000 (19:19 -0400)
When a rogue SYN is received after the connection arbitration
algorithm has converged, the incoming SYN should not needlessly
quiesce the transmit path, and it should not result in needless
TCP connection resets due to re-execution of the connection
arbitration logic.

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/rds/tcp_listen.c

index 094a8cae0ef1d18ab7e10634b33db57461e47390..4bf4befe5066a0084a7430cc39f9fd0c08a9a20e 100644 (file)
@@ -132,11 +132,13 @@ int rds_tcp_accept_one(struct socket *sock)
                 * so we must quiesce any send threads before resetting
                 * c_transport_data.
                 */
-               wait_event(conn->c_waitq,
-                          !test_bit(RDS_IN_XMIT, &conn->c_flags));
-               if (ntohl(inet->inet_saddr) < ntohl(inet->inet_daddr)) {
+               if (ntohl(inet->inet_saddr) < ntohl(inet->inet_daddr) ||
+                   !conn->c_outgoing) {
                        goto rst_nsk;
-               } else if (rs_tcp->t_sock) {
+               } else {
+                       atomic_set(&conn->c_state, RDS_CONN_CONNECTING);
+                       wait_event(conn->c_waitq,
+                                  !test_bit(RDS_IN_XMIT, &conn->c_flags));
                        rds_tcp_restore_callbacks(rs_tcp->t_sock, rs_tcp);
                        conn->c_outgoing = 0;
                }