38 releases (8 breaking)

new 0.9.0 Oct 29, 2024
0.7.0 Aug 28, 2024
0.6.0 Apr 22, 2024
0.4.1 Jan 31, 2024
0.1.0-alpha.7 Jul 21, 2020

#40 in Network programming

Download history 7623/week @ 2024-07-09 12815/week @ 2024-07-16 6965/week @ 2024-07-23 6935/week @ 2024-07-30 6861/week @ 2024-08-06 6258/week @ 2024-08-13 7260/week @ 2024-08-20 7462/week @ 2024-08-27 9365/week @ 2024-09-03 8221/week @ 2024-09-10 10420/week @ 2024-09-17 8891/week @ 2024-09-24 7813/week @ 2024-10-01 10550/week @ 2024-10-08 13356/week @ 2024-10-15 13230/week @ 2024-10-22

46,384 downloads per month
Used in 3 crates

Apache-2.0

640KB
12K SLoC

discv5

Build Status Doc Status Crates Status

Documentation at docs.rs

Overview

This is a rust implementation of the Discovery v5 peer discovery protocol.

Discovery v5 is a protocol designed for encrypted peer discovery. Each peer/node on the network is identified via it's ENR (Ethereum Node Record), which is essentially a signed key-value store containing the node's public key and optionally IP address and port.

Discv5 employs a kademlia-like routing table to store and manage discovered peers and topics. The protocol allows for external IP discovery in NAT environments through regular PING/PONG's with discovered nodes. Nodes return the external IP address that they have received and a simple majority is chosen as our external IP address. If an external IP address is updated, this is produced as an event to notify the swarm (if one is used for this behaviour).

For a simple CLI discovery service see discv5-cli

Usage

A simple example of creating this service is as follows:

   use discv5::{enr, enr::{CombinedKey, NodeId}, TokioExecutor, Discv5, ConfigBuilder};
   use discv5::socket::ListenConfig;
   use std::net::SocketAddr;

   // construct a local ENR
   let enr_key = CombinedKey::generate_secp256k1();
   let enr = enr::Enr::empty(&enr_key).unwrap();

   // build the tokio executor
   let mut runtime = tokio::runtime::Builder::new_multi_thread()
       .thread_name("Discv5-example")
       .enable_all()
       .build()
       .unwrap();

   // configuration for the sockets to listen on
   let listen_config = ListenConfig::Ipv4 {
       ip: Ipv4Addr::UNSPECIFIED,
       port: 9000,
   };

   // default configuration
   let config = ConfigBuilder::new(listen_config).build();

   // construct the discv5 server
   let mut discv5: Discv5 = Discv5::new(enr, enr_key, config).unwrap();

   // In order to bootstrap the routing table an external ENR should be added
   // This can be done via add_enr. I.e.:
   // discv5.add_enr(<ENR>)

   // start the discv5 server
   runtime.block_on(discv5.start());

   // run a find_node query
   runtime.block_on(async {
      let found_nodes = discv5.find_node(NodeId::random()).await.unwrap();
      println!("Found nodes: {:?}", found_nodes);
   });

Addresses in ENRs

An ENR is a signed record which is primarily used in this protocol for identifying and connecting to peers. ENRs have OPTIONAL ip and port fields.

If a node does not know its contactable address (i.e if it is behind a NAT), it should leave these fields empty. This is done for the following reasons:

  1. When we receive an ENR we must decide whether to add it to our local routing table and advertise it to other peers. If a node has put some non-contactable address in the ENR (e.g 127.0.0.1 when connecting to non-local nodes) we cannot use this ENR to contact the node and we therefore do not wish to advertise it to other nodes. Putting a non-contactable address is therefore functionally equivalent to leaving the fields empty.
  2. For every new inbound connection, we do not wish to check that the address given to us in an ENR is contactable. We do not want the scenario, where any peer can give us any address and force us to attempt a connection to arbitrary addresses (to check their validity) as it consumes unnecessary bandwidth and we want to avoid DOS attacks where malicious users spam many nodes attempting them all to send messages to a victim IP.

How this protocol handles advertised IPs in ENRs

To handle the above two cases this protocol filters out and only advertises contactable ENRs. It doesn't make sense for a discovery protocol to advertise non-contactable peers.

This is done in the following way:

  1. If a connecting node provides an ENR without specifying an address (this should be the default case for most nodes behind a NAT, or ones that have just started) we consider this valid. Typically this will occur when a node has yet to determine its external IP address via PONG responses and has not updated its ENR to a contactable address. In this case, we respond to all requests this peer asks for but we do not store or add its ENR to our routing table.
  2. If a peer connects to us with an ENR that specifies an IP address that matches the src address we received the packet from, we consider this peer valid and attempt to add it to our local routing table and therefore may advertise its ENR to others.
  3. If a peer connects to us with an ENR that specifies an IP address that does not match the src socket it connects to us on (e.g 127.0.0.1, or potentially some internal subnet IP that is unreachable from our current network) we consider this ENR invalid and will not add it to our routing table. However we will still respond to discovery requests.

Dependencies

~12–24MB
~314K SLoC