Hi,
Thanks for interesting logs, it is possible that this is a kernel bug, the implementation is pretty fresh and there were few issues with it in the past.
I was beginning to suspect the same. What would be a good place for the bug report? Some kernel teams seem to prefer Bugzilla and others their mailing lists, and I have a very mixed track record for getting any reaction at all from kernel bug reports (as in, about half my reports never got any reply).
You could try modified configurations (e.g. multihop instead of direct, real ethernet interface (or dummy interface) instead of tun interface) to see if the issue is related to some of these variants.
Unfortunately I only control one side of the BGP session here, and I don't have any BGP experience so I wouldn't even know how to go about doing such experiments. Kind regards, Ralf