Re: Static vxlan configuration fails - probably because of wrong order in rc

On 8/28/25 09:47, Patrick M. Hausen wrote:
> Hi!
> 
>> Am 26.08.2025 um 20:02 schrieb John Baldwin <[email protected]>:
>>
>> On 8/22/25 17:30, Patrick M. Hausen wrote:
>>> Hi,
>>>> Am 22.08.2025 um 21:02 schrieb John Baldwin <[email protected]>:
>>>>
>>>> A verbose dmesg of the two cases would probably be helpful.
>>> I set verbose_loading="YES", then booted two times, once without, once
>>> with if_smsc_load="YES".
>>> --------
>>> $ diff dmesg-not-loaded.txt dmesg-loaded.txt
>>> 10a11
>>>> module smsc already present!
>>> 12c13
>>> < avail memory = 945016832 (901 MB)
>>> ---
>>>> avail memory = 944844800 (901 MB)
>>> 178a180
>>>> vxlan0: Ethernet address: 58:9c:fc:00:82:82
>>> 182d183
>>> < vxlan0: Ethernet address: 58:9c:fc:00:82:82
>>> --------
>>
>> That's not the full dmesg, just a diff.
> 
> I was reluctant to spam the list with two full dmesg outputs if just a small part
> is probably relevant. Thought a diff would catch it.
> 
> Anyway I'll add both as an attachment - hope that passes the list server.

Yep, and the context around the last diff (the vlxan) does show it is intertwined
with smsc0 initializing.  That said, I'm still not sure why it's not always failing
TBH.  In both cases vxlan0 reports its MAC address before ue0 has even attached.
It might be useful to look at dmesg -a output to see what console messages from
userspace are firing (e.g. when the rc.d scripts run), but I doubt that will really
tell us anything new.

>> A more general fix would be to add support for a 'vxlans_<IF>'
>> variable similar to `vlans_<IF>' which would be a list of child vxlan
>> interfaces.   You would then just need `vxlans_ue0="vxlan0"` instead of
>> cloned_interfaces="vxlan0" in rc.conf.
> 
> That's actually the most interesting suggestion because shell scripting is what
> I can do and I am familiar with that mechanism having used e.g. vlans_igb0...
> for years. I'll try to implement and submit that possibly in the coming week.
> OTOH there are still two EuroBSCCon presentations to be done, atm. :-)

I think this would also perhaps fix your original root bug you mentioned at the
start of the thread that vxlans are not created when an interface is brought up.

-- 
John Baldwin