From 70fc25ad6e71a99b5802eb92b95c26407acbe990 Mon Sep 17 00:00:00 2001
From: Michael Brown <mcb30@ipxe.org>
Date: Wed, 25 Jan 2017 14:48:24 +0000
Subject: [PATCH] [netdevice] Limit MTU by hardware maximum frame length

Separate out the concept of "hardware maximum supported frame length"
and "configured link MTU", and limit the latter according to the
former.

In networks where the DHCP-supplied link MTU is inconsistent with the
hardware or driver capabilities (e.g. a network using jumbo frames),
this will result in iPXE advertising a TCP MSS consistent with a size
that can actually be received.

Note that the term "MTU" is typically used to refer to the maximum
length excluding the link-layer headers; we adopt this usage.

Signed-off-by: Michael Brown <mcb30@ipxe.org>
---
 src/include/ipxe/netdevice.h |  9 ++++++++-
 src/net/infiniband/xsigo.c   |  1 +
 src/net/netdev_settings.c    | 22 +++++++++++++++-------
 src/net/netdevice.c          |  6 ++++++
 src/net/tcpip.c              |  3 +--
 5 files changed, 31 insertions(+), 10 deletions(-)

diff --git a/src/include/ipxe/netdevice.h b/src/include/ipxe/netdevice.h
index a1d207ff..d498ab69 100644
--- a/src/include/ipxe/netdevice.h
+++ b/src/include/ipxe/netdevice.h
@@ -397,9 +397,16 @@ struct net_device {
 	struct retry_timer link_block;
 	/** Maximum packet length
 	 *
-	 * This length includes any link-layer headers.
+	 * This is the maximum packet length (including any link-layer
+	 * headers) supported by the hardware.
 	 */
 	size_t max_pkt_len;
+	/** Maximum transmission unit length
+	 *
+	 * This is the maximum transmission unit length (excluding any
+	 * link-layer headers) configured for the link.
+	 */
+	size_t mtu;
 	/** TX packet queue */
 	struct list_head tx_queue;
 	/** Deferred TX packet queue */
diff --git a/src/net/infiniband/xsigo.c b/src/net/infiniband/xsigo.c
index 91b7b71f..0ee753c3 100644
--- a/src/net/infiniband/xsigo.c
+++ b/src/net/infiniband/xsigo.c
@@ -323,6 +323,7 @@ static int xve_update_mtu ( struct xsigo_nic *xve, struct eoib_device *eoib,
 	 * not the EoIB header.
 	 */
 	netdev->max_pkt_len = ( mtu + sizeof ( struct ethhdr ) );
+	netdev->mtu = mtu;
 	DBGC ( xve, "XVE %s has MTU %zd\n", xve->name, mtu );
 
 	return 0;
diff --git a/src/net/netdev_settings.c b/src/net/netdev_settings.c
index 67a45bed..c54288d4 100644
--- a/src/net/netdev_settings.c
+++ b/src/net/netdev_settings.c
@@ -393,7 +393,8 @@ static int apply_netdev_settings ( void ) {
 	struct net_device *netdev;
 	struct settings *settings;
 	struct ll_protocol *ll_protocol;
-	size_t old_max_pkt_len;
+	size_t max_mtu;
+	size_t old_mtu;
 	size_t mtu;
 	int rc;
 
@@ -410,18 +411,25 @@ static int apply_netdev_settings ( void ) {
 		if ( ! mtu )
 			continue;
 
-		/* Update maximum packet length */
+		/* Limit MTU to maximum supported by hardware */
 		ll_protocol = netdev->ll_protocol;
-		old_max_pkt_len = netdev->max_pkt_len;
-		netdev->max_pkt_len = ( mtu + ll_protocol->ll_header_len );
-		if ( netdev->max_pkt_len != old_max_pkt_len ) {
+		max_mtu = ( netdev->max_pkt_len - ll_protocol->ll_header_len );
+		if ( mtu > max_mtu ) {
+			DBGC ( netdev, "NETDEV %s cannot support MTU %zd (max "
+			       "%zd)\n", netdev->name, mtu, max_mtu );
+			mtu = max_mtu;
+		}
+
+		/* Update maximum packet length */
+		old_mtu = netdev->mtu;
+		netdev->mtu = mtu;
+		if ( mtu != old_mtu ) {
 			DBGC ( netdev, "NETDEV %s MTU is %zd\n",
 			       netdev->name, mtu );
 		}
 
 		/* Close and reopen network device if MTU has increased */
-		if ( netdev_is_open ( netdev ) &&
-		     ( netdev->max_pkt_len > old_max_pkt_len ) ) {
+		if ( netdev_is_open ( netdev ) && ( mtu > old_mtu ) ) {
 			netdev_close ( netdev );
 			if ( ( rc = netdev_open ( netdev ) ) != 0 ) {
 				DBGC ( netdev, "NETDEV %s could not reopen: "
diff --git a/src/net/netdevice.c b/src/net/netdevice.c
index 9df21196..41ece77f 100644
--- a/src/net/netdevice.c
+++ b/src/net/netdevice.c
@@ -663,6 +663,12 @@ int register_netdev ( struct net_device *netdev ) {
 		ll_protocol->init_addr ( netdev->hw_addr, netdev->ll_addr );
 	}
 
+	/* Set MTU, if not already set */
+	if ( ! netdev->mtu ) {
+		netdev->mtu = ( netdev->max_pkt_len -
+				ll_protocol->ll_header_len );
+	}
+
 	/* Reject network devices that are already available via a
 	 * different hardware device.
 	 */
diff --git a/src/net/tcpip.c b/src/net/tcpip.c
index c9e4ee78..cc7d0200 100644
--- a/src/net/tcpip.c
+++ b/src/net/tcpip.c
@@ -144,8 +144,7 @@ size_t tcpip_mtu ( struct sockaddr_tcpip *st_dest ) {
 		return 0;
 
 	/* Calculate MTU */
-	mtu = ( netdev->max_pkt_len - netdev->ll_protocol->ll_header_len -
-		tcpip_net->header_len );
+	mtu = ( netdev->mtu - tcpip_net->header_len );
 
 	return mtu;
 }