This week's "why does anything in this world even work?" thing: Linux kernel arp handling.

In the default configuration it's super promiscuous, and doesn't care any about little details like "does this arp entry I have just learned even belong to the subnet on that interface" or "does this arp request match that inbound interface"...

That's great (not), but usually doesn't break (outside of stateful firewallsin the path - different topic), right until you have interfaces with different MTUs...

Follow

git.kernel.org/pub/scm/linux/k has fun comments like "This may seem wrong but it usually makes sense, because it increases the chance of successful communication."

Now, until very recently Openstack kept the default kernel configuration - but when you use VXLAN, you usually have a higher MTU on the VXLAN transport interface at the very least... What could possibly go wrong?

Sign in to participate in the conversation
INFRa Mastodon

This Mastodon instance is not open for public registration. Site administrator is Alexander Bochmann.

Contact email: ab+mastodon@infra.de