getport: Restore historical TCP connect timeout error code
authorChuck Lever <chuck.lever@oracle.com>
Tue, 14 Jul 2009 20:26:42 +0000 (16:26 -0400)
committerSteve Dickson <steved@redhat.com>
Tue, 14 Jul 2009 20:26:42 +0000 (16:26 -0400)
The latest versions of mount.nfs appear not to fall back to
UDP if TCP isn't available on the server.

Our new nfs_getport() implementation is missing a bit of logic
from the original mount getport() implementation.  Without it,
nfs_probe_port() sees a TCP connect timeout as a permanent error,
so it fails immediately instead of attempting to try again with
UDP.

Similar changes for our new ping API (see the old clnt_ping()
function, which is still in utils/mount/network.c).

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Steve Dickson <steved@redhat.com>
support/nfs/getport.c

index aa9c154..e39f809 100644 (file)
@@ -66,6 +66,23 @@ static const rpcvers_t default_rpcb_version = PMAPVERS;
 #endif /* !HAVE_LIBTIRPC */
 
 /*
+ * Historical: Map TCP connect timeouts to timeout
+ * error code used by UDP.
+ */
+static void
+nfs_gp_map_tcp_errorcodes(const unsigned short protocol)
+{
+       if (protocol != IPPROTO_TCP)
+               return;
+
+       switch (rpc_createerr.cf_error.re_errno) {
+       case ETIMEDOUT:
+               rpc_createerr.cf_stat = RPC_TIMEDOUT;
+               break;
+       }
+}
+
+/*
  * There's no easy way to tell how the local system's networking
  * and rpcbind is configured (ie. whether we want to use IPv6 or
  * IPv4 loopback to contact RPC services on the local host).  We
@@ -188,10 +205,13 @@ static CLIENT *nfs_gp_get_rpcbclient(struct sockaddr *sap,
                NULL,
        };
        rpcprog_t rpcb_prog = nfs_getrpcbyname(RPCBPROG, rpcb_pgmtbl);
+       CLIENT *clnt;
 
        nfs_gp_set_port(sap, nfs_gp_get_rpcb_port(transport));
-       return nfs_get_rpcclient(sap, salen, transport, rpcb_prog,
-                                       version, timeout);
+       clnt = nfs_get_rpcclient(sap, salen, transport, rpcb_prog,
+                                                       version, timeout);
+       nfs_gp_map_tcp_errorcodes(transport);
+       return clnt;
 }
 
 /*
@@ -634,6 +654,7 @@ int nfs_rpc_ping(const struct sockaddr *sap, const socklen_t salen,
                                                program, version, &tout);
        if (client != NULL) {
                result = nfs_gp_ping(client, tout);
+               nfs_gp_map_tcp_errorcodes(protocol);
                CLNT_DESTROY(client);
        }
 
@@ -757,6 +778,7 @@ int nfs_getport_ping(struct sockaddr *sap, const socklen_t salen,
                                                program, version, &timeout);
                if (client != NULL) {
                        result = nfs_gp_ping(client, timeout);
+                       nfs_gp_map_tcp_errorcodes(protocol);
                        CLNT_DESTROY(client);
                }
        }