Description of problem:
The E/W routed traffic between two networks (not GW networks) cannot handle networks with different MTU sizes. The flag "ovn_emit_need_to_frag" is configured in the Neutron configuration (although this is affecting only to the GW ports).

Even if the packets sent have DF (Do-Not-Fragment bit) unset, the ICMP packets are not reaching the port in the network with lower MTU.

Version-Release number of selected component (if applicable):

• RHEL 9.0
• OSP17.1
• ovn-nbctl 22.12.1
• Open vSwitch Library 3.0.90
• DB Schema 7.0.0

How reproducible:
See steps to reproduce.

Steps to Reproduce:
1. Deploy an OpenStack environment, using any OVN (geneve, vxlan) backend type: https://rhos-ci-jenkins.lab.eng.tlv2.redhat.com/view/QE/view/OSP17.1/
2. Create two networks and one router between them. The MTUs must be different
$ openstack network create --mtu 1500 net1
$ openstack network create --mtu 1200 net2
$ openstack subnet create --network net1 --subnet-range 10.0.201.0/28 snet1
$ openstack subnet create --network net2 --subnet-range 10.0.202.0/28 snet2
$ openstack router create router1
$ openstack router add subnet router1 snet1
$ openstack router add subnet router1 snet2
3. Spawn two VMs, each one in a different network
$ openstack server create --network net1 --flavor cirros --image cirros server1
$ openstack server create --network net2 --flavor cirros --image cirros server2
4. Access the VMs and ping from server1 (MTU 1500) to server2 (MTU 1200). Use the "ping -M dont" flag to unset the DF flag.

Actual results:
The datapath flow is received in the compute node of server2: https://paste.opendev.org/show/bVhsu1RdXnF7tquQN4Hg/. However, the TAP interface of server2 does not receive any packet. When the MTU is 1200 or lower, this TAP interface receives the ping packets; e.g.: https://paste.opendev.org/show/bdblIQENJeUp2kWio7Ac/

Expected results:
OVN should fragment the traffic and sent the fragments to the 1200 TAP interface, that should be received by server2.

Additional info:
—