Signed envelopes & routing records

[yusefnapora]

This implements the signed envelopes and routing records described in libp2p/specs#217.

I've also updated the peerstore implementation to fit the new API & will open a PR there in a sec.

Travis correctly marks this as a breaking API change, since the new methods on the peerstore make existing implementations no longer fit the interface. I think this means I need to bump the minor version?

I still feel like a Go newbie, so don't be shy about review feedback :)

Some TODO items:

• filter out local addresses when making a routing record for a Host
• ensure envelope type hint has valid multicodec prefix (right now it accepts any byte string)
  • maybe? we'd have to keep this updated as new multicodecs are added

[Stebalien]

I like it.

[vyzo]

good work, mostly nits; the important one is that we use gogo protobuf instead of the golang stuff.

[yusefnapora]

I've been playing around with interfaces a bit, and there is a way to add this functionality without making a breaking API change.

Instead of adding the new methods to the peerstore.AddrBook interface, we can define a new CertifiedAddrBook interface that just contains the new methods (AddCertifiedAddrs, CertifiedAddrs and SignedRoutingState).

To avoid the breaking change, we can't embed the new interface into the main Peerstore interface though, so callers would have to type assert to check if the peerstore supports signed addresses, e.g.

p := myHost.Peerstore()
if cab, ok := p.(CertifiedAddrBook); ok {
  cab.AddCertifiedAddrs(signedRoutingState, ttl)
}

This is more awkward to use though, and I'm probably overthinking this...

[vyzo]

Definitely overthinking this! I think it's ok to just extend the interface in this case.

[raulk]

Some feedback. I'm gonna rejig a few things, and open a PR against your branch so we can discuss alternative options for some things.

[raulk]

ensure envelope type hint has valid multicodec prefix (right now it accepts any byte string)

• maybe? we'd have to keep this updated as new multicodecs are added

Yeah, I wouldn't hardcode an enumeration in the protobuf.

[raulk]

This is more awkward to use though, and I'm probably overthinking this...

Note that the using component would only do that once, and it would safe a reference to the CertifiedAddrBook.

[yusefnapora]

Thanks for the feedback @raulk

I spent some time today pulling in your branch and am refactoring a few things that I can push tomorrow.

I also think I've found a way around the implementation problem I was having with a separate CertifiedAddrBook interface, so I can make that change here once I verify that it works in the peerstore implementation.

[yusefnapora]

Hey @raulk, I pulled in your review branch and added a few commits based on your feedback.

The main changes are:

• Defined a separate CertifiedAddrBook and a helper method GetCertifiedAddrBook(AddrBook) (CertifiedAddrBook, bool) to do the upcasting.
  • yay for API compatibility & the elusive green checkmark :)
• Renamed the record from RoutingStateRecord to PeerRecord and moved to the peer package.
• Removed the embedded reference to the SignedEnvelope from the PeerRecord. It's now just a representation of the envelope payload, with helpers to produce a SignedEnvelope and unmarshal a PeerRecord from a SignedEnvelope.
• The AddCertifiedAddrs method on CertifiedAddrBook now takes a generic SignedEnvelope as an argument, and unpacks the PeerRecord payload. Likewise, CertifiedAddrBook.SignedPeerRecord(peerId) returns a SignedEnvelope.
• Moved the Seq field from the record to the SignedEnvelope and added to the signature calculation
  • will need to update the spec

What do you think? I've updated my -peerstore and go-libp2p branches locally to reflect the changes & will push after some cleanup if you think we're on the right track here.

[raulk]

Initial feedback. Still going through the rest, but posting this quickly so we can discuss the proposed API changes and build up the rest of the review on top of that discussion ;-)

[raulk]

@yusefnapora really, really great work here! I am especially enjoying the thoughtful structure and godocs ❤️

[raulk]

This is looking close to the finish line. Just a few comments from my first review that went unaddressed, and a proposal to further refine the Record interface and simplify the overall solution.

[vyzo]

LGTM

[raulk]

🎉🎉🎉🎉🎉🎉🎉🎉

So excited to merge this! Excellent work, @yusefnapora! 🥇

🎉🎉🎉🎉🎉🎉🎉🎉

[aarshkshah1992]

record.Envelope*

[aarshkshah1992]

@yusefnapora Why do we use the RawPayload here instead of using a field of Record type since any payload that goes into the Envelope will always be a Record.

[aarshkshah1992]

EvtLocalPeerRecordUpdated*