removed subshelling from veth between containers

[steiler]

Added MTU capability
Had to remove concurrency -> further investigation required

This is still work in progress, fixing #68

[hellt]

@steiler when I was researching the topic, I found out that the CNI repo is the best one to draw inspiration from on how to create p2p links. https://github.com/containernetworking/plugins

They have a good explanation how dangerous is it to use netns switching with goroutines - https://github.com/containernetworking/plugins/tree/master/pkg/ns

For our cause of p2p links creation we should browse this repo and see how they create veth pairs, instead of using vishavanda/netns package directly

and maybe even a better reference will be that project and their APIs - https://github.com/redhat-nfvpe/koko/blob/8a6435ee27d70ab2fdf42b05023ec8d0a75276b1/api/koko_api.go#L73

[karimra]

@steiler here is a blog post about locking goroutines to a thread and thus to a network namespace,
https://www.weave.works/blog/linux-namespaces-golang-followup
this is probably there reason you had to disable concurrency

[steiler]

I've tested this now and it works fine for me.

This should implicitly also resolve #197 and lays foundation for #196 which is already available as a field in the internal struct LinkAttributes which just needs to be filled. hence just the front-end needs implementation.

[hellt]

@steiler it seems this PR doesn't address #197? In that issue the prime focus is on the single-interfaces linux containers (eth0) which we do not explicitly create
This is created by docker

[steiler]

@hellt you're right. It wasn't clear to me that the default docker bridge was meant.
Any additional feedback on this pull request?

[hellt]

I think I will get to it starting next year
I still want to see if reusing the koko code will make it cleaner

[hellt]

Changes are made in #209