snet: add extension header support to DataplanePath and PacketInfo

[marcodermatt]

For the implementation of the FABRID protocol in scionlab (netsec-ethz#168) it was necessary to access Hop-by-Hop and End-to-End extension headers during serialization and decoding of the packet.

We added a new SetExtensions function to the DataplanePath interface, which is called in packet.go. For existing DataplanePaths (onehop, scion, etc.) this function just returns nil.

// DataplanePath is an abstract representation of a SCION dataplane path.
type DataplanePath interface {
	// SetPath sets the path in the SCION header. It assumes that all the fields
	// except the path and path type are set correctly.
	SetPath(scion *slayers.SCION) error
	// SetExtensions sets the HBH and E2E extension headers. It is called directly after
	// the SetPath function.
	SetExtensions(s *slayers.SCION, p *PacketInfo) error
}

The complete code changes can be found in this commit (marcodermatt@576373e) which is independent of FABRID.

This would also allow implementing EPIC-HP as an HBH-option (#4099).

[amdfxlucas]

I think that's a great idea! It would facilitate the development of new experimental SCION features a lot. Such as ID-INT or multicast

[jiceatscion]

Does not seem to add a lot of cost or risk. I have no objection. Opinions? Other opinions?

[amdfxlucas]

I think that's a great idea! It would facilitate the development of new experimental SCION features a lot. Such as ID-INT or multicast

Found yet another interesting reference/use case SPIO SCION Path Information Option

[oncilla]

Without having looked at the concrete usage examples. This feels like the wrong API.
Extensions are not really a property that is defined by a path.

While I understand that some extensions require path information, I don't really see that the path should be used to define the extensions that are attached to a SCION packet.

[amdfxlucas]

isn't this just a naming issue ?! Rename 'DataplanePath' into sth. more aptly like 'PCFSSetter' and we are good ?!

[oncilla]

In the same spirit, you could argue to just add another field to snet.UDPAddr to set the extensions and "we are good".

What do you gain by tightly coupling the path and extensions in one abstraction?