[Warning: DMARC Fail Email] Re: ipv6 connexion fail - ipv4 OK

[Mike O'Connor]

Hi

On a 1500 link I'm having to use 1280 to get ipv6 to successfully go 
over a wireguard link.

I really think wireguard should be able to fragment and send via 
multiply UDP packets.

wireguard works very well other than this issue, performance is 
extremely good.

Mike

On 28/8/21 2:46 am, Daniel wrote:
> Hi ROman
>
> Le 27/08/2021 à 18:14, Roman Mamedov a écrit :
>> On Thu, 26 Aug 2021 13:14:00 +0200
>> Daniel <tech at tootai.net> wrote:
>>
>>> Correction
>>>
>>> Le 25/08/2021 à 17:25, Daniel a écrit :
>>>> Hi list,
>>>>
>>>> I setup wireguard on a server running Debian 11 and get it to work 
>>>> with
>>>> 2 clients (Debian 11 and Ubuntu 20.04). Clients and server are on
>>>> separate networks, one client behind a FW the other direct on 
>>>> Internet,
>>>> no FW at all (VPS).
>>>>
>>>> With this setup and ipv4 connection to the public IP of the server,
>>>> everything is working as expected, ipv4 as well as ipv6 are passing
>>>> smoothly.
>>>>
>>>> Now I want to connect using the ipv6 address of the wg interface as 
>>>> both
>>>> clients and server have ULA ipv6.
>>> Here is GUA to read.
>>>
>>>> This fail, wg show that connection is
>>>> established but VPN is not usable. It's not a FW problem as I can 
>>>> ssh to
>>>> the ipv6 address, as well as a netcat test from/to server IP -from 
>>>> each
>>>> client- on an UDP port is working properly. Also,
>>>> net.ipv6.conf.all.forwarding=1 is activated in sysctl.conf
>>>>
>>>> All network stuff is done in /etc/network/interfaces which call the
>>>> config file. The ipv6 address of the server is affected _to the
>>>> wireguard interface_ (in ipv4 it's another interface who take care of
>>>> the public address)
>>>>
>>>> Server version is wireguard-tools v1.0.20210223.
>>>>
>>>> If someone have any hint, thanks to share ;)
>> IPv6 requires the in-WG MTU to be 20 bytes less than when running 
>> over IPv4.
>> Try reducing it accordingly.
>
> Tried 1400, 1396 and 1392, problem stay.
>
> Thanks for your help
>