Low MTU on native link in fra1

[kasperd]

I am using the fra1 tunnel server, and I have noticed that even though the MSS being negotiated for TCP connections is small enough for the tunnel, I still end up having to rely on PMTU discovery, because the next link after the tunnel has an even smaller MTU. The tunnel has an MTU of 1480 bytes, but the next link from 2001:470:0:69::2 to 2001:470:0:69::1 has an MTU of only 1436 bytes. Judging from reverse DNS those two are located like right next to each other, and I assume it is a native link between them, so why would that link have an MTU of only 1436 bytes?

Do you think this is intentional? And what might be the reason for such a configuration?

[kasperd]

Turns out it is not even consistent. Packets of identical size to the same destination get different results. Sometimes I can send IPv6 packets of 1480 (the maximum permitted by the tunnel), and they make it through to the destination. At other times the tunnel server will report that the MTU of the link to the next hop is only 1436 bytes. That definitely can't be intentional. Right now it behaves fine, I'll file a ticket if it starts behaving weird again.

[kasperd]

The spurious packet too big errors keeps showing up periodically. It is reported on the same link every time it happens, but the actual value being reported as the MTU of that link varies. It affects multiple destinations. For one of the destinations I have verified that the actual PMTU is 1480 bytes. I don't see any differences in the traceroute output between times where the problem is present and times where it is not. So I don't think it is caused by traffic being shifted to a different link.

For now I have this tcpdump command running:
tcpdump -pni eth0 'src 216.66.80.30 && ip[9] == 41 && ip[26] == 0x3a && ip[60] == 2' -s0 -Uw /tmp/too-big
Is there a smarter way to match tunnelled packets with tcpdump than what I am doing? Is there any other data that I could collect, which may be useful to include in a ticket?

[kasperd]

Ticket #1892276 filed.