if_bridge looses ARP packets with Intel NICs (bnxt driver)

Started by striker2150, Today at 08:07:09 PM

Previous topic - Next topic
Upstream FreeBSD has a bug which causes ARP messages to be lost when using a bridge (i.e., will not be sent out via bnxt device if).
The issue seems to be that ARP packages are not padded and therefore are shorter than the minimum ethernet frame length.
This causes the bnxt driver (and other Intel drivers) to silently drop the packet.

There is already bug report 296652 in the official FreeBSD Bugzilla.
This report already includes a patch that looks fine to me.
However, the patch is available since early July 2026 but has not been patched to the official kernel.

Is there a possibility to merge this fix for OPNsense 26.7 until FreeBSD gets it fixed in their own git repository?

Just for the record: bnxt(4) is Broadcom, not Intel.
Deciso DEC750
People who think they know everything are a great annoyance to those of us who do. (Isaac Asimov)

That's a strange issue. I'd've thought the MAC would always pad short packets, as 64B has been the standard for... 50 years? Looking at it initially, I thought it might be a carrier extension issue, but apparently not.

I haven't noticed ARP issues (other than proxy issues, which do not result from lost requests). I do not currently have any Broadcom devices in use - just Intel and Realtek at the moment. I imagine the particular switch might be an issue as well.