cifs: add mount parameter tcpnodelay

Although corking and uncorking the socket (which cifs.ko already
does) should usually have the desired benefit, using the new
tcpnodelay mount option causes tcp_sock_set_nodelay() to be set
on the socket which may be useful in order to ensure that we don't
ever have cases where the network stack is waiting on sending an
SMB request until multiple SMB requests have been added to the
send queue (since this could lead to long latencies).

To enable it simply append "tcpnodelay" it to the mount options

Reviewed-by: Paulo Alcantara (SUSE) <pc@cjr.nz>
Signed-off-by: Steve French <stfrench@microsoft.com>

fs/cifs/fs_context.c

@@ -116,6 +116,7 @@ const struct fs_parameter_spec smb3_fs_parameters[] = {
 	fsparam_flag("nosharesock", Opt_nosharesock),
 	fsparam_flag_no("persistenthandles", Opt_persistent),
 	fsparam_flag_no("resilienthandles", Opt_resilient),
+	fsparam_flag_no("tcpnodelay", Opt_tcp_nodelay),
 	fsparam_flag("domainauto", Opt_domainauto),
 	fsparam_flag("rdma", Opt_rdma),
 	fsparam_flag("modesid", Opt_modesid),
@@ -1389,6 +1390,13 @@ static int smb3_fs_context_parse_param(struct fs_context *fc,
 			}
 		}
 		break;
+	case Opt_tcp_nodelay:
+		/* tcp nodelay should not usually be needed since we CORK/UNCORK the socket */
+		if (result.negated)
+			ctx->sockopt_tcp_nodelay = false;
+		else
+			ctx->sockopt_tcp_nodelay = true;
+		break;
 	case Opt_domainauto:
 		ctx->domainauto = true;
 		break;

fs/cifs/fs_context.h

@@ -98,6 +98,7 @@ enum cifs_param {
 	Opt_nosharesock,
 	Opt_persistent,
 	Opt_resilient,
+	Opt_tcp_nodelay,
 	Opt_domainauto,
 	Opt_rdma,
 	Opt_modesid,