Re: [PATCH net-next v4] vxlan: try to send a packet normally if local bypass fails

[Vladimir Nikishkin <vladimir@nikishkin.pw>]

Nikolay Aleksandrov <razor@blackwall.org> writes:

> On 22/03/2023 10:26, Ido Schimmel wrote:
>> On Wed, Mar 22, 2023 at 03:04:14PM +0800, Vladimir Nikishkin wrote:
>>> --- a/Documentation/networking/vxlan.rst
>>> +++ b/Documentation/networking/vxlan.rst
>>> @@ -86,3 +86,16 @@ offloaded ports can be interrogated with `ethtool`::
>>>        Types: geneve, vxlan-gpe
>>>        Entries (1):
>>>            port 1230, vxlan-gpe
>>> +
>>> +=================
>>> +Sysctls
>>> +=================
>>> +
>>> +One sysctl influences the behaviour of the vxlan driver.
>>> +
>>> + - `vxlan.disable_local_bypass`
>>> +
>>> +If set to 1, and if there is a packet destined to the local address, for which the
>>> +driver cannot find a corresponding vni, it is forwarded to the userspace networking
>>> +stack. This is useful if there is some userspace UDP tunnel waiting for such
>>> +packets.
>> 
>> Hi,
>> 
>> I don't believe sysctl is the right interface for this. VXLAN options
>> are usually / always added as netlink attributes. See ip-link man page
>> under "VXLAN Type Support".
>> 
>> Also, please add a selftest under tools/testing/selftests/net/. We
>> already have a bunch of VXLAN tests that you can use as a reference.
>> 
>> Thanks
>
> Right, that is what I meant when I suggested making it optional.
> Sorry for not being explicit. Please use vxlan netlink attributes.
>
> Cheers,
>  Nik

Sorry for misunderstanding. I have send two patches to the mailing list,
one for the kernel, and one for the ip-link. If this is the correct way
to do it, I will write a test case. I wouldn't want to test something
which is incorrect by design. Sorry for so many wrong attempts.

-- 
Your sincerely,
Vladimir Nikishkin (MiEr, lockywolf)
(Laptop)
--
Fastmail.