"roaming" between source ports does not work

[Ivan Labáth]

Hello,

are you sure changing of source port is the issue?
Seems like something that would be reported a long
time ago.

Wireguard handshake fails, if your timestamps aren't
monotonically increasing - maybe this is the issue?

For confirmation - does connection fail on wg restart without
a device power cycle, or if you change the source port
while the tunnel is running?

If your device is power cycling on a schedule, without a RTC,
you should arrange an increasing nonce/time, if you can save
data, maybe use NTP or a workaround may be to remove and
re-add the peer on the server on a compatible schedule, 

Regards,
Ivan


On Sun, Nov 08, 2020 at 11:00:30PM +0100, Matthias May wrote:
> Hi
> 
> == Premise
> * I've recently implemented support for wireguard in our LTE-router.
> 
> == Source Environment
> * The basis is OpenWRT.
> * Used versions:
> * On the client/initiator:
>  * wg
>   * 1.0.20200908
>   * ad33b2d2267a37e0f65c97e65e7d4d926d5aef7d530c251b63fbf919048eead9
>  * wg-tools
>   * 1.0.20200827
>   * 51bc85e33a5b3cf353786ae64b0f1216d7a871447f058b6137f793eb0f53b7fd
> * On the server/responder:
>  * Debian stretch (9.13), installed from repository
>  * deb http://deb.debian.org/debian/ unstable main
>  * # wg --version
>  * wireguard-tools v1.0.20200827
>  * I don't really know what the version of the build dkms is
> 
> == Issue
> * We've implemented an automated test that seems to have a problem.
>  * Each night, the device is configured to connect to the debian box.
>   * This works fine the first time.
>   * However it doesn't work anymore after this first time.
> 
> == Observerion
> When the "client" connects the first time, wg-output on the "server"
> looks like this:
> > interface: wg1
> >   public key: 7GxCG4m+6Kf4wjJ9vbQaGFASLGXLB5ddPWgBYw4gOk8=
> >   private key: (hidden)
> >   listening port: 51821
> >
> > peer: fizBdi/YkdzFLaq6Hnq+OZaGmbJBYC15QSP1Mik/EFU=
> >   endpoint: 172.29.42.230:38442
> >   allowed ips: 10.0.41.3/32
> >   latest handshake: 44 seconds ago
> >   transfer: 8.01 MiB received, 7.96 MiB sent
> 
> and on the "client:
> > interface: wg1
> >   public key: fizBdi/YkdzFLaq6Hnq+OZaGmbJBYC15QSP1Mik/EFU=
> >   private key: (hidden)
> >   listening port: 38442
> >
> > peer: 7GxCG4m+6Kf4wjJ9vbQaGFASLGXLB5ddPWgBYw4gOk8=
> >   endpoint: 172.29.60.13:51821
> >   allowed ips: 10.0.41.0/24
> >   latest handshake: 1 minute, 3 seconds ago
> >   transfer: 187.06 KiB received, 189.96 KiB sent
> 
> Ports and IPs match, everything works.
> 
> However on the second run of the test:
> On the "server" still:
> > peer: fizBdi/YkdzFLaq6Hnq+OZaGmbJBYC15QSP1Mik/EFU=
> >   endpoint: 172.29.42.230:38442
> >   allowed ips: 10.0.41.3/32
> >   latest handshake: 4 minutes, 52 seconds ago
> >   transfer: 8.05 MiB received, 7.99 MiB sent
> 
> But the "client" shows:
> > interface: wg1
> >   public key: fizBdi/YkdzFLaq6Hnq+OZaGmbJBYC15QSP1Mik/EFU=
> >   private key: (hidden)
> >   listening port: 47858
> 
> The client device has been restarted in between.
> 
> Since the listen-port is set to 0, it obviously has now a new,
> different, source-port.
> The server doesn't pick this up.
> Since peers may roam between IPs, i was under the impression, that it
> would also roam between ports.
> 
> 
> Is this working as intended?
> If yes: How should the configuration look like to support clients doing
> a power-cycle?
> 
> 
> I'm aware, that i could set a static port on the client, but this won't
> work when going through NAT with port-scrambling.
> So i don't really have control over the source-port of the connection
> anyway.
> I suppose this would also apply when a router/firewall inbetween has
> some aggressive killing of states where the keepalive is not fast
> enough, and source-port scrambling is done.
> 
> But the main usecase i'm looking at here is: restart of a device.
> 
> BR
> Matthias