Additional points of clarification: I only use good ol' route(8) for routing (i.e, no pf(4) magic is used). This is possible since my VPS provider actually routes me the /48 and /29 IPv6 and IPv4 blocks respectively.
My comment "running OpenBSD 7.9 for a few months" was a comment on the belief that this issue is not caused by something in OpenBSD since I'd presumably experience issues when I upgraded. While I have run syspatch(8), nothing in the errata suggests this would be affected. Unfortunately due to not having logs that go back far enough, I don't know when the issue crept up. I'm confident it happened last month though. Why? Because I receive a billing e-mail from my VPS provider on the first of the month. I received such an e-mail on August 1st, but I didn't receive such e-mail on September 1st. It was a day or two later that I actually noticed the problem after I attempted to e-mail someone that uses Zoho Mail. I've been using OpenBSD 7.9 since June; and as stated, OpenBSD in general with this setup for over five years. I don't think ixl(4) is the issue either. Not only am I able to use a 1500 MTU on the interface connected to my ISP, but I have no issue with fragmentation when using vlan(4)s in rdomain 0 that use the same ixl(4) for the parent interface as the vlan(4)s in rdomain 1. Is there something weird with how wg(4) works in rdomain 1 using rtable 0 to establish the WireGuard tunnel? Something weird configuring it as the default route via: !route -qn -T1 add -inet6 -net default -link -iface -static wg0 !route -qn -T1 add -inet -net default -link -iface -static wg0 in hostname.if(5)? By "weird", it requires an additional 12 bytes of headers? Not sure why I wouldn't have experienced this problem before. Admittedly, I don't use my e-mail server for much; so it's possible I've never had a large enough payload to send/receive to experience MTU issues. I find that somewhat unlikely, but it's certainly possible. I do use TLS 1.3; and when an MTU larger than 1408 is defined on Server 1's wg(4), openssl s_client -6 -tls1_3 -crlf -starttls smtp mail.philomathiclife.com:25 hangs on CONNECTED(00000003) when run on Server 2. This aligns with the eventual connection timeouts that were causing issues for external MTAs. Maybe the certificates I was using for the past 5 years were never large enough to cause this hang; and Let's Encrypt recently started issuing a larger chain. I do use short-lived certificates, so the chain changes every other day. Interesting Gmail and Outlook still worked.
