netdev-dpdk: Stop relying on vhost-user Tx flags.

vhost-user legacy behavior has been to mark mbuf with Tx offload flags
based on what the virtio-net header contained (but provide no
Rx information, like IP checksum or L4 checksum validity).

OVS current implementation was relying on this behavior so far,
though this clashed in some corner cases, like when a guest sends a
tunneled packet requesting outer udp checksum offload
(in such a case, the packet is marked as requesting UDP checksum).

Changing to the non legacy mode means that it is now OVS responsibility
to set Tx ol_flags (which should be the case already in
miniflow_extract()).

Signed-off-by: David Marchand <[email protected]>

lib/netdev-dpdk.c

@@ -6387,6 +6387,11 @@ netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev)
             vhost_flags |= RTE_VHOST_USER_POSTCOPY_SUPPORT;
         }
 
+        /* Prefer that the vhost library provides Rx ol_flags, rather than let
+         * it mark the packet with Tx ol_flags (which may collide with OVS Tx
+         * offloads decisions). */
+        vhost_flags |= RTE_VHOST_USER_NET_COMPLIANT_OL_FLAGS;
+
         /* Enable External Buffers if TCP Segmentation Offload is enabled. */
         if (enable_tso) {
             vhost_flags |= RTE_VHOST_USER_EXTBUF_SUPPORT;