DHCP renew fails with ICMP port unreachable (3/3)

[lrosenman]

Bug Report

Description

talos VM's (all 5 of mine) whine about renew packets not making it. I captured pcaps of all the DHCP (67/68) and ICMP traffic for a period of time to demonstrate, in the tarball are:

1. console log (talosctl logs -f controller-runtime | grep dhcp >file
2. macaddrs.txt showing mac's and IP's of the machines in this test
3. talos.vm.pcap (the pcap from the vm's switch port)
4. talos.switch.pcap (the pcap from the bridge that has also has the other 4 talos nodes)

in the talos.vm.pcap file look at packets 30-34 for what I believe is the renew->ack->ICMP.

logs

talos-dhcp.tar.gz

Environment

• Talos version: [talosctl version --nodes <problematic nodes>] v1.1.0
• Kubernetes version: [kubectl version --short] 1.24.2
• Platform: FreeBSD/bhyve VM's/using sysutils/vm-bhyve to control the VM's.

[lrosenman]

still occurring with 1.1.1 and 1.1.2

[RussianFox]

Hi! I have this problem too.

10.2.64.21: 2023-05-22T14:34:45.368Z DEBUG DHCP RENEW {"component": "controller-runtime", "controller": "network.OperatorSpecController", "operator": "dhcp4", "link": "eth0"} 10.2.64.21: 2023-05-22T14:35:20.370Z WARN request/renew failed {"component": "controller-runtime", "controller": "network.OperatorSpecController", "operator": "dhcp4", "error": "got an error while processing the request: no matching response packet received", "link": "eth0"}

But no packets in network

[smira]

Talos 1.4+ has a completely rewritten DHCP stack, so please specify which version you're seeing it with.

[RussianFox]

v1.4.1

[smira]

It makes sense to supply the full packet dump from the network, otherwise it's hard to say anything. And it's better to open a new issue, as it's not clear whether this is the same one.