rds: tcp: send handshake ping-probe from passive endpoint

The RDS handshake ping probe added by commit 5916e2c1554f
("RDS: TCP: Enable multipath RDS for TCP") is sent from rds_sendmsg()
before the first data packet is sent to a peer. If the conversation
is not bidirectional  (i.e., one side is always passive and never
invokes rds_sendmsg()) and the passive side restarts its rds_tcp
module, a new HS ping probe needs to be sent, so that the number
of paths can be re-established.

This patch achieves that by sending a HS ping probe from
rds_tcp_accept_one() when c_npaths is 0 (i.e., we have not done
a handshake probe with this peer yet).

Orabug: 26477841

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Reviewed-by: Yuval Shaia <yuval.shaia@oracle.com>

diff --git a/net/rds/rds.h b/net/rds/rds.h
index 8ea9d60c14b8865fe6c95199772b05b8efc2f591..18a8da1e6b71a0cd112742202ff8bfc98dfbb935 100644 (file)
--- a/net/rds/rds.h
+++ b/net/rds/rds.h
@@ -1062,6 +1062,7 @@ void rds_send_drop_acked(struct rds_connection *conn, u64 ack,
 void rds_send_path_drop_acked(struct rds_conn_path *cp, u64 ack,
                              is_acked_func is_acked);
 void rds_send_remove_from_sock(struct list_head *messages, int status);
+void rds_send_ping(struct rds_connection *conn, int cp_index);
 int rds_send_pong(struct rds_conn_path *cp, __be16 dport);
 int rds_send_hb(struct rds_connection *conn, int response);
 struct rds_message *rds_send_get_message(struct rds_connection *,

diff --git a/net/rds/recv.c b/net/rds/recv.c
index bf8c8b55a7a0bfb5a31e8ff2aa1cf2bec393df93..8e8a07795a84e0bb75e6305e226302d0e2a7a16e 100644 (file)
--- a/net/rds/recv.c
+++ b/net/rds/recv.c
@@ -278,6 +278,7 @@ static void rds_recv_hs_exthdrs(struct rds_header *hdr,
        }
        /* if RDS_EXTHDR_NPATHS was not found, default to a single-path */
        conn->c_npaths = max_t(int, conn->c_npaths, 1);
+       conn->c_ping_triggered = 0;
        rds_conn_peer_gen_update(conn, new_peer_gen_num);
 }
 
@@ -295,8 +296,7 @@ static void rds_recv_hs_exthdrs(struct rds_header *hdr,
  *    called after reception of the probe-pong on all mprds_paths.
  *    Otherwise (sender of probe-ping is not the smaller ip addr): just call
  *    rds_conn_path_connect_if_down on the hashed path. (see rule 4)
- * 4. when cp_index > 0, rds_connect_worker must only trigger
- *    a connection if laddr < faddr.
+ * 4. rds_connect_worker must only trigger a connection if laddr < faddr.
  * 5. sender may end up queuing the packet on the cp. will get sent out later.
  *    when connection is completed.
  */
@@ -307,7 +307,7 @@ static void rds_start_mprds(struct rds_connection *conn)
 
        if (conn->c_npaths > 1 &&
            IS_CANONICAL(conn->c_laddr, conn->c_faddr)) {
-               for (i = 1; i < conn->c_npaths; i++) {
+               for (i = 0; i < conn->c_npaths; i++) {
                        cp = &conn->c_path[i];
                        rds_conn_path_connect_if_down(cp);
                }

diff --git a/net/rds/send.c b/net/rds/send.c
index 37c7883822dc6c433b8ea7f4c6b2f0f5e7763f1f..0977678bee58e061f60f0b58705e7cdaecfac0bc 100644 (file)
--- a/net/rds/send.c
+++ b/net/rds/send.c
@@ -1178,8 +1178,6 @@ static inline int rds_rdma_bytes(struct msghdr *msg, size_t *rdma_bytes)
        return 0;
 }
 
-static void rds_send_ping(struct rds_connection *conn);
-
 static int rds_send_mprds_hash(struct rds_sock *rs, struct rds_connection *conn)
 {
        int hash;
@@ -1189,7 +1187,7 @@ static int rds_send_mprds_hash(struct rds_sock *rs, struct rds_connection *conn)
        else
                hash = RDS_MPATH_HASH(rs, conn->c_npaths);
        if (conn->c_npaths == 0 && hash != 0) {
-               rds_send_ping(conn);
+               rds_send_ping(conn, 0);
 
                if (conn->c_npaths == 0) {
                        wait_event_interruptible(conn->c_hs_waitq,
@@ -1725,10 +1723,10 @@ rds_send_pong(struct rds_conn_path *cp, __be16 dport)
 }
 
 void
-rds_send_ping(struct rds_connection *conn)
+rds_send_ping(struct rds_connection *conn, int cp_index)
 {
        unsigned long flags;
-       struct rds_conn_path *cp = &conn->c_path[0];
+       struct rds_conn_path *cp = &conn->c_path[cp_index];
 
        spin_lock_irqsave(&cp->cp_lock, flags);
        if (conn->c_ping_triggered) {
@@ -1737,6 +1735,6 @@ rds_send_ping(struct rds_connection *conn)
        }
        conn->c_ping_triggered = 1;
        spin_unlock_irqrestore(&cp->cp_lock, flags);
-       rds_send_probe(&conn->c_path[0], cpu_to_be16(RDS_FLAG_PROBE_PORT),
-                      0, 0);
+       rds_send_probe(cp, cpu_to_be16(RDS_FLAG_PROBE_PORT), 0, 0);
 }
+EXPORT_SYMBOL_GPL(rds_send_ping);

diff --git a/net/rds/tcp_listen.c b/net/rds/tcp_listen.c
index 6776bb7c7d76b9a31d9f131517c5bbdf754df944..cc98a98c7bd65bb7ee378a211306fad6c38bff71 100644 (file)
--- a/net/rds/tcp_listen.c
+++ b/net/rds/tcp_listen.c
@@ -185,6 +185,8 @@ int rds_tcp_accept_one(struct socket *sock)
        }
        new_sock = NULL;
        ret = 0;
+       if (conn->c_npaths == 0)
+               rds_send_ping(cp->cp_conn, cp->cp_index);
        goto out;
 rst_nsk:
        /* reset the newly returned accept sock and bail.