RPi associating two IPs with its one and only wifi interface

Many ISPs - especially cookie-cutter consumer-oriented ones - do not support customers using IPv6. (Although it might be argued that that is not very different from the level of support they provide for IPv4)

TNP is thinking of the ‘predictable interface names’ scheme, which is a misnomer since predicting the name depends on information you don’t necessarily have readily to hand. Really it is about keeping the mapping between names and specific hardware interfaces stable, which is valuable in some situations and inconvenient in others.

IMHO they should have taken a lesson from filesystem naming, and allowed interfaces to be identified by their attributes, rather than depending on a unique name.

What attributes would you use?

If your system is running NetworkManager, it is the culprit.

In my RasPi3B+ router, I disabled and stopped NetworkManager. systemd-networkd is perfectly capable to handle the DHCP client duties.

After this, you have to create the needed interface and network descriptions into /etc/systemd/network, and that's it.

The two links in the directory are fine, do not mess with them.

Er no. On my HP laptop the interface is something like (runs to laptop and returns) "wlo1"...

The raspi-config allows you to choose between simple names (such as wlan0) and unique names (such as wl0sp3 on my laptop). With unique names you can and another identical device, but the original one's name wont change no matter what order they are enumerated in.

---druck

Given that all of my systems have been running NetworkManager for many years, and that I have only seen this happen once, I'm having a hard time seeing why you can make such a definitive statement. Care to elaborate?

I run a number of systemd-free systems, and having as much commonality as possibly reduces admin time. So, for me, I would prefer to stay away from switching network configuration tools if at all possible. YMMV.

Jim

On my Thinkpad its wlp3s0. Those two couldn't possibly cause any confusion for anyone, I'm sure!

Curiously

curl -4

Last thing I receive in IPv6 pcap from Greenend says [TCP Previous segment not captured]

Curious. The curl -6 command works for me (and comes in like a rocket!) and the link also works in Firefox, and I know that IPv6 is preferred.

That's the first instance I've hear of something failing to work correctly in IPv6 with no obvious explanation.

David

It works for me from a couple of locations (one with native IPv6 and the other via a Hurricance Electric tunnel). traceroute -6 might reveal where your problem is.

AFAICT that just means you started tracing part way through a session.

Actually mine is wlp3s0 too, I remembered it wrong!

Also as someone else mentioned, it's actually predicable names rather than unique names.

---druck

Yes, it would be.

I wasn't thinking when I replied, but wl0sp3 isn't in the right format.

Yes, for some strange value of "predictable" that requires you to know the naming scheme and also to know the hardware rather more intimately than most people usually have any need or desire to do.

See:

I had a look on my system, to see where network interface names come from. I found those names used inside /sys/class/net, which is low-level information directly from the Linux kernel, not obtained from udev rules or systemd or any other such userland source.

I had smililar problem when I upgraded to Bullseye. My solution was deleted connman package.

Refer to

Thanks both of you, that gave me confidence to investigate further.

I'm guessing it is due to a split TCP packet. TLS asks for a change of cipher, which is a big packet, and then it all goes wrong.

Anyway, setting MTU=1444 fixes the problem. Internet tells me that the root problem is because the firewall is blocking ICMP too big, breaking Path MTU Discovery (PMTUD) . I can't see where, so now I have to understand that.

Thanks for the pointer, but that wasn't it.

The system in question (as well as my other systems) doesn't (respectively, don't) have connman installed.

(I have a feeling I might not have named a piece of software "connman", even though I can imagine where they got that name.)

Jim

I actually do that. I route my IPTV boxes out via an alternate interface due to some stupid contractual issues. So all I did was add routing rules with ip -6 rule add from $addr table Magic and all the Magic table has is a defaultroute out via the other interface. Same as IPv4. But maybe your policy routing is something different?

For sure this would be a problem if the IPv6 addresses were changing all the time but they haven't.

Some routers will let you use the source mac address in routing rules which nicely overcomes the problem with varying IPv6 addresses.

John