diff options
Diffstat (limited to 'src/transport/libp2p.rs')
| -rw-r--r-- | src/transport/libp2p.rs | 1836 |
1 files changed, 1836 insertions, 0 deletions
diff --git a/src/transport/libp2p.rs b/src/transport/libp2p.rs new file mode 100644 index 0000000..30fdb23 --- /dev/null +++ b/src/transport/libp2p.rs @@ -0,0 +1,1836 @@ +//! libp2p as a transport (`--transports libp2p`, in builds with the `libp2p` feature): +//! direct connections over TCP (Noise, Yamux) and QUIC, to addresses this node knows. +//! +//! A node's transport identity is its libp2p peer ID; where it is, its listen addresses +//! (multiaddrs ending in the peer ID, `locations`). Seeds are such multiaddrs, or ones +//! without the peer ID, which a connection finds out. Messages go one per stream of the +//! `/continuum/1` protocol: their length (4 bytes, big endian), then the message as +//! `wire` encodes it. A send is done once the stream is written and closed; a connection +//! that can't be made, or a stream that can't be written, refuses it. +//! +//! Connected peers tell each other where they listen (libp2p's identify protocol), so a +//! node learns the addresses of peers that dialled it. A node publishes where it can be +//! reached (`local_addresses`): the addresses it announces (`--libp2p-announce`, e.g. a +//! public address forwarded to it), or else those it listens on, loopback left out. +//! +//! A node nobody can dial (behind NAT, or a closed port) can still be reached through a +//! relay: a node of the swarm that relays (`--libp2p-relay`), on which it reserves a slot +//! (`--libp2p-relays`); its address through the relay (`<relay>/p2p-circuit/p2p/<it>`) +//! is published with the others. Over a relayed connection, both sides then try to +//! connect directly (libp2p's DCUtR, hole punching); the relay carries the traffic until +//! that works, or for good if it doesn't, within its limits (`RELAY_CIRCUIT_DURATION`, +//! `RELAY_CIRCUIT_BYTES`; a circuit that ends is made again on the next send). +//! +//! A relay says so where it publishes its addresses: next to each, the same address +//! ending in `/p2p-circuit` (where to reserve a slot), which other nodes find in the +//! member list. A node that needs a relay and has none configured picks up to +//! `AUTO_RELAYS` of those (`relay_check`). It needs one when it has no address others +//! could dial, or when AutoNAT says it can't be dialled: AutoNAT asks connected peers +//! (nodes of the swarm) to dial it back. Once AutoNAT says it can be dialled, it lets +//! the relays it picked go. +//! +//! With `--libp2p-mdns`, a node finds libp2p peers on its LAN (multicast DNS). Their +//! LAN addresses are noted, and those that turn out to speak continuum (identify says +//! so) are seeds while the node joins (`discovered_seeds`): nodes on one LAN find each +//! other without configured seeds. +//! +//! It uses public libp2p infrastructure only with `--libp2p-public`: then it joins the +//! IPFS DHT (Kademlia, through the IPFS bootstrap nodes), where it finds addresses of +//! peers the member list doesn't have, and public peers that relay; those are picked +//! after members that relay. Public relays keep circuits short (by default 2 minutes and +//! 128 KiB): enough to punch holes through NATs, not to carry a swarm's traffic. Without +//! it, relays are those configured, or members that say they relay. + +use std::collections::HashMap; +use std::path::Path; +use std::str::FromStr; +use std::sync::{Arc, Mutex, MutexGuard}; +use std::time::Duration; + +use ::libp2p::multiaddr::Protocol; +use ::libp2p::swarm::dial_opts::{DialOpts, PeerCondition}; +use ::libp2p::swarm::{ConnectionId, SwarmEvent}; +use ::libp2p::{Multiaddr, PeerId, StreamProtocol, Swarm, identity::Keypair}; +use anyhow::{Context, Result, anyhow, bail}; +use async_trait::async_trait; +use futures::{AsyncReadExt, AsyncWriteExt, StreamExt}; +use tokio::sync::{Semaphore, mpsc, oneshot, watch}; +use tracing::{debug, info, trace, warn}; + +use crate::locks::MutexExt; +use crate::transport::stats::{ + Connections, DialFailures, HolePunches, Libp2pStats, Reachability, RelaySlots, TransportStats, +}; +use crate::transport::{ + Connection, EncodedMessage, MessageHandler, ParsedMessage, Transport, wire, +}; + +const PROTOCOL: StreamProtocol = StreamProtocol::new("/continuum/1"); +/// The largest message read: anyone who can connect can send anything, so this bounds +/// what one stream costs. Sync messages above 16 chunks travel as blobs. +const MAX_MESSAGE: usize = 128 << 20; +/// How long a connection may take to come up. +const CONNECT_TIMEOUT: Duration = Duration::from_secs(10); +/// A connection nobody uses is closed after this; neighbours keep theirs busy. +const IDLE_TIMEOUT: Duration = Duration::from_secs(300); +/// How long a stream may take to open and to carry a message, before what its size adds +/// (`stream_timeout`). +const STREAM_TIMEOUT: Duration = Duration::from_secs(10); +/// The slowest a message is expected to go, in bytes per second. +const MIN_RATE: u64 = 1 << 20; +/// Inbound streams read at once. +const READS: usize = 32; +/// Addresses kept per peer, and peers with addresses at most. +const ADDRESSES_PER_PEER: usize = 8; +const MAX_PEERS: usize = 4096; + +/// A relayed connection (circuit) through this node lasts at most this long and carries +/// at most this much: generous, since the swarm's own relays may carry sync traffic for +/// good where hole punching fails (libp2p's defaults, 2 minutes and 128 KiB, are for +/// the punching only). +const RELAY_CIRCUIT_DURATION: Duration = Duration::from_secs(3600); +const RELAY_CIRCUIT_BYTES: u64 = 1 << 30; +/// Circuits through this node at once, and per peer. +const RELAY_CIRCUITS: usize = 64; +const RELAY_CIRCUITS_PER_PEER: usize = 8; +/// A reservation on a relay that ended (the relay restarted, say) is made again after +/// this. +const RELAY_RETRY: Duration = Duration::from_secs(30); +/// Relays picked among the members, at most, and how often a node checks whether it +/// needs them. +const AUTO_RELAYS: usize = 2; +const RELAY_CHECK: Duration = Duration::from_secs(30); +/// A relay that lost the slot before granting it (it doesn't answer, or another peer +/// answers at its address: a relay restarted without its libp2p key) isn't picked again +/// for this long. +const RELAY_PAUSE: Duration = Duration::from_secs(600); +/// A peer nobody mentioned for this long (the member list, identify, mDNS, a +/// connection) is forgotten, unless connected: mostly old peer IDs of nodes restarted +/// without their libp2p key, whose new ones the member list has since. Members are +/// mentioned every few seconds (`swarm::topology`). Checked every `PRUNE_EVERY`. +const STALE_AFTER: Duration = Duration::from_secs(600); +const PRUNE_EVERY: Duration = Duration::from_secs(60); + +/// How long a message of `bytes` may take on a stream. +fn stream_timeout(bytes: usize) -> Duration { + STREAM_TIMEOUT + Duration::from_millis(bytes as u64 * 1000 / MIN_RATE) +} + +/// What a libp2p transport is made with. +pub(crate) struct Libp2pConfig { + pub(crate) key: Keypair, + /// Multiaddrs to listen on. + pub(crate) listen: Vec<Multiaddr>, + /// Multiaddrs to publish instead of the listen addresses, if any. + pub(crate) announce: Vec<Multiaddr>, + /// Relays to reserve a slot on (multiaddrs; without the relay's peer ID, a connection + /// finds it out). + pub(crate) relays: Vec<Multiaddr>, + /// Relay for other nodes. + pub(crate) relay: bool, + /// Find peers on the LAN (mDNS). + pub(crate) mdns: bool, + /// Use public infrastructure: the IPFS DHT, public relays. + pub(crate) public: bool, + /// libp2p's in-memory transport instead of TCP and QUIC: for tests. + pub(crate) memory: bool, +} + +impl Libp2pConfig { + /// The key in `path` (32 raw bytes, an Ed25519 secret key), or a new one each start + /// without a file: the peer ID only says where a node is, so a new one costs nothing + /// but the WHO that finds it again (`transport::identity`). + pub(crate) fn load_key(path: Option<&Path>) -> Result<Keypair> { + let Some(path) = path else { + return Ok(Keypair::generate_ed25519()); + }; + let mut bytes = std::fs::read(path) + .with_context(|| format!("Reading the libp2p key from {}", path.display()))?; + crate::swarm::auth::warn_if_readable_by_others(path, "libp2p key"); + if bytes.len() != 32 { + bail!( + "The libp2p key in {} has {} bytes; it has 32", + path.display(), + bytes.len() + ); + } + Keypair::ed25519_from_bytes(&mut bytes) + .map_err(|error| anyhow!("The libp2p key in {}: {error}", path.display())) + } +} + +enum Command { + /// Connect to `peer` (at `addresses`) unless connected. + Connect { + peer: PeerId, + addresses: Vec<Multiaddr>, + done: oneshot::Sender<Result<(), String>>, + }, + /// Connect to `address` to find out which peer is there. + Identify { + address: Multiaddr, + done: oneshot::Sender<Result<PeerId, String>>, + }, + /// See whether relays among the members are needed now (one was just learned of). + CheckRelays, + /// Reserve a slot on the relay at `relay` (again). `configured`: the address as + /// configured, for later attempts (without a peer ID, the relay may since have + /// another). + Reserve { + relay: Multiaddr, + configured: Multiaddr, + }, +} + +/// Where peers are, and where we listen. +#[derive(Default)] +struct Book { + addresses: HashMap<PeerId, Vec<Multiaddr>>, + listening: Vec<Multiaddr>, + /// `Libp2pConfig::announce`. + announced: Vec<Multiaddr>, + /// `Libp2pConfig::relay`: this node relays, and says so. + serving: bool, + /// Members that say they relay, and where (with the relay's peer ID). + relays: HashMap<PeerId, Multiaddr>, + /// With `Libp2pConfig::public`: other peers that relay (identify says so), and + /// where. + public_relays: HashMap<PeerId, Multiaddr>, + /// Relays picked that never granted a slot, and when (`RELAY_PAUSE`). + failed_relays: HashMap<PeerId, tokio::time::Instant>, + /// Relays that granted us a slot we hold. + slots: std::collections::HashSet<PeerId>, + /// Open connections: to whom, and whether through a relay. + connections: HashMap<ConnectionId, (PeerId, bool)>, + /// What AutoNAT says, and the address it confirmed while public. + reachability: Reachability, + public_address: Option<Multiaddr>, + /// Since start (`stats`). + counts: Counts, + /// When each peer was last mentioned (`STALE_AFTER`). + seen: HashMap<PeerId, tokio::time::Instant>, + /// Peers found on the LAN (mDNS) not known yet to speak continuum. + lan: std::collections::HashSet<PeerId>, + /// Peers that speak continuum (identify says so). + continuum: std::collections::HashSet<PeerId>, + /// Peers found on the LAN that speak continuum, with their address: seeds not + /// handed out yet (`discovered_seeds`). + discovered: Vec<Multiaddr>, +} + +/// What `Libp2pStats` counts since start. +#[derive(Default)] +struct Counts { + slots_granted: u64, + slots_lost: u64, + slots_not_granted: u64, + hole_punches: HolePunches, + dial_failures: DialFailures, +} + +/// Counts a dial that failed with `error` (not one that wasn't needed). +fn count_dial_failure(failures: &mut DialFailures, error: &::libp2p::swarm::DialError) { + use ::libp2p::swarm::DialError; + match error { + DialError::DialPeerConditionFalse(_) => {} + DialError::Transport(_) => failures.unreachable += 1, + DialError::WrongPeerId { .. } => failures.wrong_peer_id += 1, + DialError::NoAddresses => failures.no_addresses += 1, + _ => failures.other += 1, + } +} + +impl Book { + /// How we are connected to `peer`: directly if any connection is direct. + fn connection(&self, peer: &PeerId) -> Option<Connection> { + let mut found = None; + for (connected, relayed) in self.connections.values() { + if connected == peer { + if !relayed { + return Some(Connection::Direct); + } + found = Some(Connection::Relayed); + } + } + found + } + + fn stats(&self) -> Libp2pStats { + let mut connections = Connections::default(); + for (_, relayed) in self.connections.values() { + if *relayed { + connections.relayed += 1; + } else { + connections.direct += 1; + } + } + Libp2pStats { + reachability: self.reachability, + public_address: self.public_address.as_ref().map(Multiaddr::to_string), + connections, + relay_slots: RelaySlots { + held: self.slots.len(), + paused: self + .failed_relays + .values() + .filter(|at| at.elapsed() < RELAY_PAUSE) + .count(), + granted: self.counts.slots_granted, + lost: self.counts.slots_lost, + not_granted: self.counts.slots_not_granted, + }, + hole_punches: self.counts.hole_punches, + dial_failures: self.counts.dial_failures, + } + } + + /// `peer` was mentioned just now. + fn touch(&mut self, peer: PeerId) { + self.seen.insert(peer, tokio::time::Instant::now()); + } + + /// Forgets the peers not mentioned within `STALE_AFTER` before `now`, but those + /// `connected`. Returns how many. + fn prune(&mut self, now: tokio::time::Instant, connected: impl Fn(&PeerId) -> bool) -> usize { + let stale: Vec<PeerId> = self + .addresses + .keys() + .chain(self.relays.keys()) + .chain(self.public_relays.keys()) + .chain(self.continuum.iter()) + .chain(self.lan.iter()) + .filter(|peer| { + !connected(peer) + && self + .seen + .get(peer) + .is_none_or(|at| now.saturating_duration_since(*at) >= STALE_AFTER) + }) + .copied() + .collect::<std::collections::HashSet<_>>() + .into_iter() + .collect(); + for peer in &stale { + self.addresses.remove(peer); + self.relays.remove(peer); + self.public_relays.remove(peer); + self.continuum.remove(peer); + self.lan.remove(peer); + self.failed_relays.remove(peer); + self.seen.remove(peer); + } + stale.len() + } + + /// `peer`, found on the LAN, speaks continuum: a seed, at the address we know first. + fn discover(&mut self, peer: PeerId) { + let Some(address) = self.addresses.get(&peer).and_then(|known| known.first()) else { + return; + }; + if let Ok(address) = address.clone().with_p2p(peer) + && !self.discovered.contains(&address) + && self.discovered.len() < MAX_PEERS + { + self.discovered.push(address); + } + } + + /// Where others could dial us directly: the announced addresses, or else those we + /// listen on (`published`), without the peer ID. + fn direct(&self) -> Vec<Multiaddr> { + if !self.announced.is_empty() { + return self.announced.clone(); + } + let direct: Vec<Multiaddr> = self + .listening + .iter() + .filter(|address| !is_relayed(address)) + .cloned() + .collect(); + published(&direct) + } +} + +/// Whether `address` only reaches this host (or this link): no use to other hosts, and +/// dialled by one, it reaches that host itself. +fn is_local(address: &Multiaddr) -> bool { + match address.iter().next() { + Some(Protocol::Ip4(ip)) => ip.is_loopback() || ip.is_link_local() || ip.is_unspecified(), + Some(Protocol::Ip6(ip)) => { + ip.is_loopback() || ip.is_unicast_link_local() || ip.is_unspecified() + } + _ => false, + } +} + +/// The addresses of `all` worth publishing: all but local ones, unless they're all +/// local (a test on one host). +fn published(all: &[Multiaddr]) -> Vec<Multiaddr> { + let reachable: Vec<Multiaddr> = all + .iter() + .filter(|address| !is_local(address)) + .cloned() + .collect(); + if reachable.is_empty() { + all.to_vec() + } else { + reachable + } +} + +impl Book { + /// Whether all of `addresses` are known of `peer` (`remember`). + fn knows(&self, peer: PeerId, addresses: &[Multiaddr]) -> bool { + let Some(known) = self.addresses.get(&peer) else { + return false; + }; + addresses.iter().all(|address| { + let mut address = address.clone(); + if let Some(Protocol::P2p(_)) = address.iter().last() { + address.pop(); + } + known.contains(&address) + }) + } + + /// `peer` is at `address` (without its `/p2p` part): kept first. + fn remember(&mut self, peer: PeerId, mut address: Multiaddr) { + if let Some(Protocol::P2p(_)) = address.iter().last() { + address.pop(); + } + if !self.addresses.contains_key(&peer) && self.addresses.len() >= MAX_PEERS { + return; + } + self.touch(peer); + let addresses = self.addresses.entry(peer).or_default(); + addresses.retain(|known| *known != address); + addresses.insert(0, address); + addresses.truncate(ADDRESSES_PER_PEER); + } +} + +/// Whether `address` goes through a relay. +fn is_relayed(address: &Multiaddr) -> bool { + address + .iter() + .any(|protocol| matches!(protocol, Protocol::P2pCircuit)) +} + +/// The relay `address` goes through, if it goes through one: the peer before +/// `/p2p-circuit`. +fn relay_of(address: &Multiaddr) -> Option<PeerId> { + let mut relay = None; + for protocol in address.iter() { + match protocol { + Protocol::P2p(peer) => relay = Some(peer), + Protocol::P2pCircuit => return relay, + _ => {} + } + } + None +} + +/// The addresses of `known` this node can dial: not those through itself as the relay +/// (a relay reaches the nodes it relays for only while they are connected to it). +fn dialable(known: &[Multiaddr], local: PeerId) -> Vec<Multiaddr> { + known + .iter() + .filter(|address| relay_of(address) != Some(local)) + .cloned() + .collect() +} + +/// `address` with the peer ID it ends in, if it ends in one. +fn split_peer(address: &Multiaddr) -> Option<PeerId> { + match address.iter().last() { + Some(Protocol::P2p(peer)) => Some(peer), + _ => None, + } +} + +pub(crate) struct Libp2pTransport { + local_id: String, + peer_id: PeerId, + control: libp2p_stream::Control, + commands: mpsc::UnboundedSender<Command>, + book: Arc<Mutex<Book>>, + incoming: tokio::sync::Mutex<Option<libp2p_stream::IncomingStreams>>, + stopped: watch::Receiver<bool>, +} + +fn lock(book: &Mutex<Book>) -> MutexGuard<'_, Book> { + book.lock_or_recover() +} + +impl Libp2pTransport { + /// Builds the swarm, listens, and runs it in a task of its own. + pub(crate) fn start(config: Libp2pConfig) -> Result<Self> { + let peer_id = config.key.public().to_peer_id(); + let features = Features::of(&config); + let mut swarm = if config.memory { + memory_swarm(config.key, features)? + } else { + network_swarm(config.key, features)? + }; + for address in config.listen { + swarm + .listen_on(address.clone()) + .with_context(|| format!("Listening on {address}"))?; + } + for address in &config.announce { + // Identify tells peers these. + swarm.add_external_address(address.clone()); + } + let mut control = swarm.behaviour().stream.new_control(); + let incoming = control + .accept(PROTOCOL) + .map_err(|error| anyhow!("Accepting {PROTOCOL}: {error}"))?; + let configured_relays = !config.relays.is_empty(); + let book = Arc::new(Mutex::new(Book { + announced: config.announce, + serving: config.relay, + ..Book::default() + })); + let (commands, receiver) = mpsc::unbounded_channel(); + for relay in config.relays { + let _ = commands.send(Command::Reserve { + configured: relay.clone(), + relay, + }); + } + let (stopping, stopped) = watch::channel(false); + tokio::spawn(run( + swarm, + receiver, + commands.clone(), + Arc::clone(&book), + Relaying { + serving: config.relay, + configured: configured_relays, + public: config.public, + local: peer_id, + }, + stopping, + )); + info!(peer_id = %peer_id, "libp2p transport started"); + Ok(Self { + local_id: peer_id.to_string(), + peer_id, + control, + commands, + book, + incoming: tokio::sync::Mutex::new(Some(incoming)), + stopped, + }) + } + + /// Connected to `peer`, if it can be had in time. + async fn connect(&self, peer: PeerId) -> Result<()> { + let addresses = lock(&self.book) + .addresses + .get(&peer) + .map(|known| dialable(known, self.peer_id)) + .unwrap_or_default(); + let (done, connected) = oneshot::channel(); + self.commands + .send(Command::Connect { + peer, + addresses, + done, + }) + .map_err(|_| anyhow!("the libp2p swarm stopped"))?; + match tokio::time::timeout(CONNECT_TIMEOUT, connected).await { + Ok(Ok(Ok(()))) => Ok(()), + Ok(Ok(Err(error))) => bail!("couldn't connect to {peer}: {error}"), + Ok(Err(_)) => bail!("the libp2p swarm stopped"), + Err(_) => bail!("connecting to {peer} took longer than {CONNECT_TIMEOUT:?}"), + } + } +} + +/// The protocol version identify says (informational: nothing checks it). +const IDENTIFY_VERSION: &str = "/continuum/1"; + +#[derive(::libp2p::swarm::NetworkBehaviour)] +struct Behaviour { + stream: libp2p_stream::Behaviour, + identify: ::libp2p::identify::Behaviour, + relay_client: ::libp2p::relay::client::Behaviour, + /// Only with `Libp2pConfig::relay`. + relay: ::libp2p::swarm::behaviour::toggle::Toggle<::libp2p::relay::Behaviour>, + dcutr: ::libp2p::dcutr::Behaviour, + autonat: ::libp2p::autonat::Behaviour, + /// Only with `Libp2pConfig::mdns`. + mdns: ::libp2p::swarm::behaviour::toggle::Toggle<::libp2p::mdns::tokio::Behaviour>, + /// Only with `Libp2pConfig::public`: the IPFS DHT. + kad: ::libp2p::swarm::behaviour::toggle::Toggle< + ::libp2p::kad::Behaviour<::libp2p::kad::store::MemoryStore>, + >, +} + +/// What the behaviour has besides what every node has. +#[derive(Clone, Copy)] +struct Features { + relay: bool, + mdns: bool, + public: bool, +} + +impl Features { + fn of(config: &Libp2pConfig) -> Self { + Self { + relay: config.relay, + mdns: config.mdns, + public: config.public, + } + } +} + +/// The IPFS DHT's protocol, and its bootstrap nodes (all at `IPFS_BOOTSTRAP_ADDRESS`). +const IPFS_DHT: StreamProtocol = StreamProtocol::new("/ipfs/kad/1.0.0"); +const IPFS_BOOTSTRAP_ADDRESS: &str = "/dnsaddr/bootstrap.libp2p.io"; +const IPFS_BOOTSTRAP_PEERS: [&str; 4] = [ + "QmNnooDu7bfjPFoTZYxMNLWUQJyrVwtbZg5gBMjTezGAJN", + "QmQCU2EcMqAqQPR2i9bChDtGNJchTbq5TbXJJ16u19uLTa", + "QmbLHAnMoJPWSCR5Zhtx6BHJX9KiKNN6tpvbUcqanj75Nb", + "QmcZf59bWwK5XFi76CZX8cbJ4BhTzzA3gU1ZjYZcYW3dwt", +]; +/// How often a node on the DHT refreshes its routing table. +const DHT_BOOTSTRAP: Duration = Duration::from_secs(300); + +impl Behaviour { + fn new( + key: &Keypair, + relay_client: ::libp2p::relay::client::Behaviour, + features: Features, + ) -> std::io::Result<Self> { + let Features { + relay, + mdns, + public, + } = features; + let peer_id = key.public().to_peer_id(); + let kad = public.then(|| { + let mut dht = ::libp2p::kad::Behaviour::with_config( + peer_id, + ::libp2p::kad::store::MemoryStore::new(peer_id), + ::libp2p::kad::Config::new(IPFS_DHT), + ); + let bootstrap: Multiaddr = IPFS_BOOTSTRAP_ADDRESS.parse().expect("a multiaddr"); + for peer in IPFS_BOOTSTRAP_PEERS { + let peer: PeerId = peer.parse().expect("a peer ID"); + let _ = dht.add_address(&peer, bootstrap.clone()); + } + dht + }); + let identify = ::libp2p::identify::Config::new(IDENTIFY_VERSION.to_owned(), key.public()) + .with_agent_version(format!("continuum/{}", env!("CARGO_PKG_VERSION"))); + let relay = relay.then(|| { + ::libp2p::relay::Behaviour::new( + peer_id, + ::libp2p::relay::Config { + max_circuits: RELAY_CIRCUITS, + max_circuits_per_peer: RELAY_CIRCUITS_PER_PEER, + max_circuit_duration: RELAY_CIRCUIT_DURATION, + max_circuit_bytes: RELAY_CIRCUIT_BYTES, + ..Default::default() + }, + ) + }); + let mdns = match mdns { + true => Some(::libp2p::mdns::tokio::Behaviour::new( + ::libp2p::mdns::Config::default(), + peer_id, + )?), + false => None, + }; + Ok(Self { + stream: libp2p_stream::Behaviour::new(), + identify: ::libp2p::identify::Behaviour::new(identify), + relay_client, + relay: relay.into(), + dcutr: ::libp2p::dcutr::Behaviour::new(peer_id), + autonat: ::libp2p::autonat::Behaviour::new( + peer_id, + ::libp2p::autonat::Config { + // Swarms on private networks (a LAN, WireGuard) count too. + only_global_ips: false, + ..Default::default() + }, + ), + mdns: mdns.into(), + kad: kad.into(), + }) + } +} + +fn network_swarm(key: Keypair, features: Features) -> Result<Swarm<Behaviour>> { + Ok(::libp2p::SwarmBuilder::with_existing_identity(key) + .with_tokio() + .with_tcp( + ::libp2p::tcp::Config::default(), + ::libp2p::noise::Config::new, + ::libp2p::yamux::Config::default, + )? + .with_quic() + .with_dns()? + .with_relay_client( + ::libp2p::noise::Config::new, + ::libp2p::yamux::Config::default, + )? + .with_behaviour(|key, relay_client| { + Behaviour::new(key, relay_client, features) + .map_err(|error| Box::new(error) as Box<dyn std::error::Error + Send + Sync>) + })? + .with_swarm_config(|config| config.with_idle_connection_timeout(IDLE_TIMEOUT)) + .build()) +} + +fn memory_swarm(key: Keypair, features: Features) -> Result<Swarm<Behaviour>> { + use ::libp2p::Transport as _; + use ::libp2p::core::transport::{MemoryTransport, OrTransport}; + use ::libp2p::core::upgrade::Version; + + let peer_id = key.public().to_peer_id(); + let (relay_transport, relay_client) = ::libp2p::relay::client::new(peer_id); + let transport = OrTransport::new(relay_transport, MemoryTransport::default()) + .boxed() + .upgrade(Version::V1) + .authenticate(::libp2p::noise::Config::new(&key)?) + .multiplex(::libp2p::yamux::Config::default()) + .timeout(CONNECT_TIMEOUT) + .boxed(); + Ok(Swarm::new( + transport, + Behaviour::new(&key, relay_client, features)?, + peer_id, + ::libp2p::swarm::Config::with_tokio_executor().with_idle_connection_timeout(IDLE_TIMEOUT), + )) +} + +/// Drives the swarm: dials what `commands` ask for and notes where we listen, until the +/// transport is dropped. +/// How a node relates to relays. +struct Relaying { + /// It relays for others. + serving: bool, + /// Relays to use are configured: it doesn't pick any. + configured: bool, + /// Public relays may be picked, after members. + public: bool, + local: PeerId, +} + +/// The relays to pick among `offers`, if this node needs relays, keeping `using`: a +/// node needs them when it has no address others could dial (`direct`), or AutoNAT says +/// it can't be dialled; up to `AUTO_RELAYS`, in an order of its own (the lowest hash of +/// both peer IDs), so that not every node picks the same. +fn relay_check( + local: PeerId, + direct: bool, + nat: &::libp2p::autonat::NatStatus, + offers: &HashMap<PeerId, Multiaddr>, + using: &[PeerId], +) -> Vec<(PeerId, Multiaddr)> { + use ::libp2p::autonat::NatStatus; + let needed = match nat { + NatStatus::Private => true, + NatStatus::Public(_) => false, + NatStatus::Unknown => !direct, + }; + if !needed || using.len() >= AUTO_RELAYS { + return Vec::new(); + } + let score = |relay: &PeerId| { + use std::hash::{Hash, Hasher}; + let mut hasher = std::collections::hash_map::DefaultHasher::new(); + (local, relay).hash(&mut hasher); + hasher.finish() + }; + let mut candidates: Vec<(PeerId, Multiaddr)> = offers + .iter() + .filter(|(relay, _)| **relay != local && !using.contains(relay)) + .map(|(relay, address)| (*relay, address.clone())) + .collect(); + candidates.sort_by_key(|(relay, _)| score(relay)); + candidates.truncate(AUTO_RELAYS - using.len()); + candidates +} + +/// Reserves slots on the relays `relay_check` picks among the members, and lets those +/// picked go once AutoNAT says this node can be dialled. +fn pick_relays( + swarm: &mut Swarm<Behaviour>, + book: &Mutex<Book>, + relaying: &Relaying, + nat: &::libp2p::autonat::NatStatus, + picked: &mut HashMap<::libp2p::core::transport::ListenerId, PeerId>, +) { + if nat.is_public() && !picked.is_empty() { + info!("Reachable directly (AutoNAT); letting the relays picked go"); + let mut book = lock(book); + for (listener, relay) in std::mem::take(picked) { + swarm.remove_listener(listener); + book.slots.remove(&relay); + } + return; + } + let (direct, offers, public) = { + let mut book = lock(book); + book.failed_relays + .retain(|_, at| at.elapsed() < RELAY_PAUSE); + let failed = book.failed_relays.clone(); + let usable = |offers: &HashMap<PeerId, Multiaddr>| -> HashMap<PeerId, Multiaddr> { + offers + .iter() + .filter(|(relay, _)| !failed.contains_key(relay)) + .map(|(relay, address)| (*relay, address.clone())) + .collect() + }; + let public = if relaying.public { + usable(&book.public_relays) + } else { + HashMap::new() + }; + (!book.direct().is_empty(), usable(&book.relays), public) + }; + let mut using: Vec<PeerId> = picked.values().copied().collect(); + // Members first, then public ones for what's left. + let mut wanted = relay_check(relaying.local, direct, nat, &offers, &using); + using.extend(wanted.iter().map(|(relay, _)| *relay)); + wanted.extend(relay_check(relaying.local, direct, nat, &public, &using)); + for (relay, address) in wanted { + match swarm.listen_on(address.clone().with(Protocol::P2pCircuit)) { + Ok(listener) => { + info!(relay = %relay, address = %address, "Picked a member as libp2p relay"); + picked.insert(listener, relay); + } + Err(error) => { + debug!(relay = %relay, error = %error, "Couldn't use a member as libp2p relay"); + } + } + } +} + +/// Asks the DHT (with `--libp2p-public`) where `peer` is, for the next attempt. +fn find_on_dht(swarm: &mut Swarm<Behaviour>, peer: PeerId) { + if let Some(dht) = swarm.behaviour_mut().kad.as_mut() { + dht.get_closest_peers(peer); + } +} + +/// `again`: to schedule commands. +async fn run( + mut swarm: Swarm<Behaviour>, + mut commands: mpsc::UnboundedReceiver<Command>, + again: mpsc::UnboundedSender<Command>, + book: Arc<Mutex<Book>>, + relaying: Relaying, + stopping: watch::Sender<bool>, +) { + let serving = relaying.serving; + let mut nat = ::libp2p::autonat::NatStatus::Unknown; + // Relays picked among the members, by listener. + let mut picked: HashMap<::libp2p::core::transport::ListenerId, PeerId> = HashMap::new(); + let mut check = tokio::time::interval(RELAY_CHECK); + check.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay); + let mut dht_bootstrap = tokio::time::interval(DHT_BOOTSTRAP); + dht_bootstrap.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay); + // Relays we listen through, by listener: as configured, and their peer. + let mut reservations: HashMap<::libp2p::core::transport::ListenerId, (Multiaddr, PeerId)> = + HashMap::new(); + let reserve_later = |configured: Multiaddr| { + let again = again.clone(); + tokio::spawn(async move { + tokio::time::sleep(RELAY_RETRY).await; + let _ = again.send(Command::Reserve { + relay: configured.clone(), + configured, + }); + }); + }; + let mut connecting: HashMap<PeerId, Vec<oneshot::Sender<Result<(), String>>>> = HashMap::new(); + let mut identifying: HashMap< + ConnectionId, + (Multiaddr, oneshot::Sender<Result<PeerId, String>>), + > = HashMap::new(); + let mut prune = tokio::time::interval(PRUNE_EVERY); + prune.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay); + loop { + tokio::select! { + _ = prune.tick() => { + let forgotten = lock(&book).prune(tokio::time::Instant::now(), |peer| swarm.is_connected(peer)); + if forgotten > 0 { + debug!(forgotten, "Forgot libp2p peers nobody mentions any more"); + } + } + _ = dht_bootstrap.tick(), if relaying.public => { + if let Some(dht) = swarm.behaviour_mut().kad.as_mut() + && let Err(error) = dht.bootstrap() + { + warn!(error = %error, "Couldn't join the IPFS DHT"); + } + } + _ = check.tick(), if !relaying.configured => { + pick_relays(&mut swarm, &book, &relaying, &nat, &mut picked); + } + command = commands.recv() => match command { + None => break, + Some(Command::CheckRelays) => { + if !relaying.configured { + pick_relays(&mut swarm, &book, &relaying, &nat, &mut picked); + } + } + Some(Command::Connect { peer, addresses, done }) => { + if swarm.is_connected(&peer) { + let _ = done.send(Ok(())); + continue; + } + let waiting = connecting.entry(peer).or_default(); + waiting.push(done); + if waiting.len() > 1 { + continue; // Already dialing. + } + let mut opts = DialOpts::peer_id(peer) + .addresses(addresses) + .condition(PeerCondition::DisconnectedAndNotDialing); + if relaying.public { + // Addresses the DHT knows too. + opts = opts.extend_addresses_through_behaviour(); + } + if let Err(error) = swarm.dial(opts.build()) { + count_dial_failure(&mut lock(&book).counts.dial_failures, &error); + for done in connecting.remove(&peer).into_iter().flatten() { + let _ = done.send(Err(error.to_string())); + } + find_on_dht(&mut swarm, peer); + } + } + Some(Command::Reserve { relay, configured }) if split_peer(&relay).is_none() => { + // Which peer relays there first. + let again = again.clone(); + tokio::spawn(async move { + let (done, identified) = oneshot::channel(); + if again + .send(Command::Identify { address: relay.clone(), done }) + .is_err() + { + return; + } + match tokio::time::timeout(CONNECT_TIMEOUT, identified).await { + Ok(Ok(Ok(peer))) => { + if let Ok(relay) = relay.with_p2p(peer) { + let _ = again.send(Command::Reserve { relay, configured }); + } + } + _ => { + warn!(relay = %relay, "Couldn't reach a libp2p relay; trying again later"); + tokio::time::sleep(RELAY_RETRY).await; + let _ = again.send(Command::Reserve { relay, configured }); + } + } + }); + } + Some(Command::Reserve { relay, configured }) => { + let Some(relay_peer) = split_peer(&relay) else { + continue; // The arm above takes those. + }; + match swarm.listen_on(relay.clone().with(Protocol::P2pCircuit)) { + Ok(listener) => { + debug!(relay = %relay, "Reserving a slot on a libp2p relay"); + reservations.insert(listener, (configured, relay_peer)); + } + Err(error) => { + warn!(relay = %relay, error = %error, "Couldn't use a libp2p relay; trying again later"); + reserve_later(configured); + } + } + } + Some(Command::Identify { address, done }) => { + let opts = DialOpts::unknown_peer_id().address(address.clone()).build(); + let id = opts.connection_id(); + match swarm.dial(opts) { + Ok(()) => { + identifying.insert(id, (address, done)); + } + Err(error) => { + let _ = done.send(Err(error.to_string())); + } + } + } + }, + event = swarm.select_next_some() => match event { + SwarmEvent::ConnectionEstablished { peer_id, connection_id, endpoint, .. } => { + { + let mut book = lock(&book); + book.touch(peer_id); + let relayed = endpoint.is_relayed(); + book.connections.insert(connection_id, (peer_id, relayed)); + debug!( + peer = %peer_id, + connection = %connection_id, + dialled = endpoint.is_dialer(), + address = %endpoint.get_remote_address(), + relayed, + "libp2p connection opened" + ); + } + for done in connecting.remove(&peer_id).into_iter().flatten() { + let _ = done.send(Ok(())); + } + if let Some((address, done)) = identifying.remove(&connection_id) { + lock(&book).remember(peer_id, address); + let _ = done.send(Ok(peer_id)); + } + } + SwarmEvent::ConnectionClosed { peer_id, connection_id, cause, .. } => { + lock(&book).connections.remove(&connection_id); + debug!( + peer = %peer_id, + connection = %connection_id, + cause = ?cause, + "libp2p connection closed" + ); + } + SwarmEvent::OutgoingConnectionError { connection_id, peer_id, error } => { + count_dial_failure(&mut lock(&book).counts.dial_failures, &error); + if let Some((_, done)) = identifying.remove(&connection_id) { + let _ = done.send(Err(error.to_string())); + } + if let Some(peer) = peer_id + && !swarm.is_connected(&peer) + { + if let Some(waiting) = connecting.remove(&peer) { + for done in waiting { + let _ = done.send(Err(error.to_string())); + } + find_on_dht(&mut swarm, peer); + } + } + } + SwarmEvent::NewListenAddr { address, .. } => { + info!(address = %address, "libp2p listening"); + // A relay tells those reserving where it is: its external addresses. + if serving && !is_local(&address) && !is_relayed(&address) { + swarm.add_external_address(address.clone()); + } + let mut book = lock(&book); + if !book.listening.contains(&address) { + book.listening.push(address); + } + } + SwarmEvent::ExpiredListenAddr { address, .. } => { + info!(address = %address, "libp2p no longer listening"); + if serving { + swarm.remove_external_address(&address); + } + lock(&book).listening.retain(|known| *known != address); + } + SwarmEvent::ListenerClosed { listener_id, addresses, reason } => { + lock(&book) + .listening + .retain(|known| !addresses.contains(known)); + if let Some((relay, relay_peer)) = reservations.remove(&listener_id) { + { + let mut book = lock(&book); + if book.slots.remove(&relay_peer) { + book.counts.slots_lost += 1; + } else { + book.counts.slots_not_granted += 1; + } + } + warn!(relay = %relay, reason = ?reason, "Lost the slot on a libp2p relay; reserving again later"); + reserve_later(relay); + } + // A relay picked among the members: the next check picks again, another + // one if this one never granted the slot. + if let Some(relay) = picked.remove(&listener_id) { + let lost = { + let mut book = lock(&book); + let lost = book.slots.remove(&relay); + if lost { + book.counts.slots_lost += 1; + } else { + book.counts.slots_not_granted += 1; + book.failed_relays.insert(relay, tokio::time::Instant::now()); + } + lost + }; + if lost { + debug!(relay = %relay, reason = ?reason, "Lost the slot on a member relaying"); + } else { + info!( + relay = %relay, + pause = ?RELAY_PAUSE, + "A relay didn't grant a slot (unreachable, or another peer at its address); picking others for a while" + ); + } + } + } + SwarmEvent::ListenerError { error, .. } => { + warn!(error = %error, "libp2p listener failed"); + } + SwarmEvent::Behaviour(BehaviourEvent::RelayClient( + ::libp2p::relay::client::Event::ReservationReqAccepted { relay_peer_id, renewal, .. }, + )) => { + { + let mut book = lock(&book); + book.slots.insert(relay_peer_id); + book.failed_relays.remove(&relay_peer_id); + if !renewal { + book.counts.slots_granted += 1; + } + } + if !renewal { + info!(relay = %relay_peer_id, "Reachable through a libp2p relay"); + } + } + SwarmEvent::Behaviour(BehaviourEvent::Autonat( + ::libp2p::autonat::Event::StatusChanged { old, new }, + )) => { + info!(old = ?old, new = ?new, "libp2p reachability (AutoNAT)"); + { + use ::libp2p::autonat::NatStatus; + let mut book = lock(&book); + (book.reachability, book.public_address) = match &new { + NatStatus::Unknown => (Reachability::Unknown, None), + NatStatus::Private => (Reachability::Private, None), + NatStatus::Public(address) => { + (Reachability::Public, Some(address.clone())) + } + }; + } + nat = new; + } + SwarmEvent::Behaviour(BehaviourEvent::Relay(event)) => { + debug!(event = ?event, "libp2p relay"); + } + SwarmEvent::Behaviour(BehaviourEvent::Dcutr(event)) => match event.result { + Ok(_) => { + lock(&book).counts.hole_punches.succeeded += 1; + info!(peer = %event.remote_peer_id, "Connected directly to a peer behind a relay"); + } + Err(error) => { + lock(&book).counts.hole_punches.failed += 1; + debug!(peer = %event.remote_peer_id, error = %error, "Couldn't connect directly to a peer behind a relay; the relay carries on"); + } + }, + SwarmEvent::Behaviour(BehaviourEvent::Identify( + ::libp2p::identify::Event::Received { connection_id, peer_id, info }, + )) => { + let mut book = lock(&book); + let listen = published(&info.listen_addrs); + // Every new connection identifies again: logged when it says something new. + if book.knows(peer_id, &listen) { + trace!(peer = %peer_id, connection = %connection_id, "libp2p peer identified again"); + } else { + debug!( + peer = %peer_id, + connection = %connection_id, + listen = ?info.listen_addrs, + observed = %info.observed_addr, + agent = %info.agent_version, + "libp2p peer identified" + ); + } + // Last first: `remember` puts each in front. + for address in listen.into_iter().rev() { + book.remember(peer_id, address); + } + if relaying.public + && info.protocols.contains(&::libp2p::relay::HOP_PROTOCOL_NAME) + && (book.public_relays.len() < MAX_PEERS + || book.public_relays.contains_key(&peer_id)) + && let Some(address) = book + .addresses + .get(&peer_id) + .and_then(|known| known.first()) + .and_then(|address| address.clone().with_p2p(peer_id).ok()) + { + book.public_relays.insert(peer_id, address); + } + if info.protocols.contains(&PROTOCOL) { + book.continuum.insert(peer_id); + if book.lan.remove(&peer_id) { + book.discover(peer_id); + } + } + } + SwarmEvent::Behaviour(BehaviourEvent::Mdns( + ::libp2p::mdns::Event::Discovered(found), + )) => { + for (peer, address) in found { + if peer == relaying.local { + continue; + } + debug!(peer = %peer, address = %address, "libp2p peer on the LAN"); + let continuum = { + let mut book = lock(&book); + book.remember(peer, address.clone()); + book.continuum.contains(&peer) + }; + if continuum { + lock(&book).discover(peer); + } else if lock(&book).lan.len() < MAX_PEERS { + // Connected, identify says whether it speaks continuum. + lock(&book).lan.insert(peer); + let _ = swarm.dial( + DialOpts::peer_id(peer) + .addresses(vec![address]) + .condition(PeerCondition::DisconnectedAndNotDialing) + .build(), + ); + } + } + } + _ => {} + }, + } + } + let _ = stopping.send(true); +} + +/// One message from an inbound stream. +async fn read_message(mut stream: ::libp2p::Stream) -> Result<Vec<u8>> { + let mut length = [0u8; 4]; + tokio::time::timeout(STREAM_TIMEOUT, stream.read_exact(&mut length)) + .await + .map_err(|_| anyhow!("no length within {STREAM_TIMEOUT:?}"))??; + let length = u32::from_be_bytes(length) as usize; + if length > MAX_MESSAGE { + bail!("a message of {length} bytes; at most {MAX_MESSAGE} are read"); + } + let mut body = vec![0u8; length]; + tokio::time::timeout(stream_timeout(length), stream.read_exact(&mut body)) + .await + .map_err(|_| { + anyhow!( + "{length} bytes took longer than {:?}", + stream_timeout(length) + ) + })??; + Ok(body) +} + +#[async_trait] +impl Transport for Libp2pTransport { + fn local_id(&self) -> &str { + &self.local_id + } + + fn is_peer_id(&self, id: &str) -> bool { + PeerId::from_str(id).is_ok() + } + + /// The longest a message may take on a stream. + fn delivery_limit(&self) -> Duration { + stream_timeout(MAX_MESSAGE) + } + + /// Where we can be reached, each with our peer ID: the announced addresses, or else + /// those we listen on (`published`), and those through relays. A relay adds each + /// direct one again, ending in `/p2p-circuit`: where to reserve a slot. + fn local_addresses(&self) -> Vec<String> { + let book = lock(&self.book); + let direct: Vec<Multiaddr> = book + .direct() + .into_iter() + .filter_map(|address| address.with_p2p(self.peer_id).ok()) + .collect(); + let relayed = book + .listening + .iter() + .filter(|address| is_relayed(address)) + .filter_map(|address| address.clone().with_p2p(self.peer_id).ok()); + let offers: Vec<Multiaddr> = if book.serving { + direct + .iter() + .map(|address| address.clone().with(Protocol::P2pCircuit)) + .collect() + } else { + Vec::new() + }; + direct + .into_iter() + .chain(relayed) + .chain(offers) + .map(|address| address.to_string()) + .collect() + } + + /// A peer ID, or a multiaddr ending in one (noted as where that peer is). + fn locate(&self, target: &str) -> Option<String> { + if let Ok(peer) = PeerId::from_str(target) { + return Some(peer.to_string()); + } + let mut address = Multiaddr::from_str(target).ok()?; + // A relay saying where to reserve a slot: `<address>/p2p/<relay>/p2p-circuit`. + if let Some(Protocol::P2pCircuit) = address.iter().last() { + address.pop(); + let relay = split_peer(&address)?; + if relay != self.peer_id { + let new = { + let mut book = lock(&self.book); + let new = book.relays.get(&relay) != Some(&address); + if new && (book.relays.len() < MAX_PEERS || book.relays.contains_key(&relay)) { + book.relays.insert(relay, address.clone()); + } + book.remember(relay, address); + new + }; + if new { + let _ = self.commands.send(Command::CheckRelays); + } + } + return Some(relay.to_string()); + } + let peer = split_peer(&address)?; + if peer != self.peer_id { + lock(&self.book).remember(peer, address); + } + Some(peer.to_string()) + } + + fn connection(&self, target: &str) -> Option<Connection> { + let peer = PeerId::from_str(target).ok()?; + lock(&self.book).connection(&peer) + } + + fn stats(&self) -> TransportStats { + TransportStats { + libp2p: Some(lock(&self.book).stats()), + ..TransportStats::default() + } + } + + /// Nodes found on the LAN (mDNS) that speak continuum. + fn discovered_seeds(&self) -> Vec<String> { + std::mem::take(&mut lock(&self.book).discovered) + .iter() + .map(Multiaddr::to_string) + .collect() + } + + /// Where `target` is, each address with its peer ID. + fn locations(&self, target: &str) -> Vec<String> { + let Ok(peer) = PeerId::from_str(target) else { + return Vec::new(); + }; + lock(&self.book) + .addresses + .get(&peer) + .into_iter() + .flatten() + .filter_map(|address| address.clone().with_p2p(peer).ok()) + .map(|address| address.to_string()) + .collect() + } + + /// A multiaddr ending in a peer ID is that peer; one without is connected to, to + /// find out. A bare peer ID resolves if this transport knows where it is. + async fn resolve_seed(&self, seed: &str) -> Result<String> { + if let Ok(peer) = PeerId::from_str(seed) { + if lock(&self.book).addresses.contains_key(&peer) { + return Ok(peer.to_string()); + } + bail!("no address known for libp2p peer {peer}; give a multiaddr"); + } + let address = + Multiaddr::from_str(seed).with_context(|| format!("{seed} is no multiaddr"))?; + if let Some(peer) = self.locate(seed) { + return Ok(peer); + } + let (done, identified) = oneshot::channel(); + self.commands + .send(Command::Identify { address, done }) + .map_err(|_| anyhow!("the libp2p swarm stopped"))?; + match tokio::time::timeout(CONNECT_TIMEOUT, identified).await { + Ok(Ok(Ok(peer))) => Ok(peer.to_string()), + Ok(Ok(Err(error))) => bail!("couldn't connect to {seed}: {error}"), + Ok(Err(_)) => bail!("the libp2p swarm stopped"), + Err(_) => bail!("connecting to {seed} took longer than {CONNECT_TIMEOUT:?}"), + } + } + + fn encode(&self, peer: &str, message: &ParsedMessage) -> Result<EncodedMessage> { + let body = wire::encode(message)?; + Ok(EncodedMessage { + peer: peer.to_owned(), + address: peer.to_owned(), + payload_len: body.len(), + body: body.into(), + kind: message.kind(), + message: None, + }) + } + + async fn send_encoded(&self, message: &EncodedMessage) -> Result<()> { + let peer = PeerId::from_str(&message.peer) + .with_context(|| format!("{} is no libp2p peer ID", message.peer))?; + self.connect(peer).await?; + let mut control = self.control.clone(); + let body = message.body.clone(); + let write = async move { + let mut stream = control.open_stream(peer, PROTOCOL).await?; + stream.write_all(&(body.len() as u32).to_be_bytes()).await?; + stream.write_all(&body).await?; + stream.close().await?; + Ok::<(), anyhow::Error>(()) + }; + let timeout = stream_timeout(message.body.len()); + tokio::time::timeout(timeout, write) + .await + .map_err(|_| anyhow!("sending to {peer} took longer than {timeout:?}"))? + .with_context(|| format!("Sending to {peer}")) + } + + /// Reads inbound streams (`READS` at once) and handles their messages one at a + /// time, in the order they were read. + async fn start_consumer(&self, handler: MessageHandler) -> Result<()> { + let Some(mut incoming) = self.incoming.lock().await.take() else { + bail!("libp2p consumer already started"); + }; + let (sender, mut received) = mpsc::channel::<(PeerId, Vec<u8>)>(READS); + let reads = Arc::new(Semaphore::new(READS)); + tokio::spawn(async move { + while let Some((peer, stream)) = incoming.next().await { + let Ok(permit) = Arc::clone(&reads).acquire_owned().await else { + break; + }; + let sender = sender.clone(); + tokio::spawn(async move { + match read_message(stream).await { + Ok(body) => { + let _ = sender.send((peer, body)).await; + } + Err(error) => { + debug!(peer = %peer, error = %format!("{error:#}"), "Dropping an unreadable libp2p stream"); + } + } + drop(permit); + }); + } + }); + while let Some((peer, body)) = received.recv().await { + match wire::decode(&body) { + Ok(message) => { + if let Err(error) = handler(peer.to_string(), message, body.len()).await { + warn!(peer = %peer, error = %format!("{error:#}"), "Failed to process inbound message"); + } + } + Err(error) => { + debug!(peer = %peer, error = %error, "Dropping a libp2p message that doesn't decode"); + } + } + } + Ok(()) + } + + async fn wait_for_exit(&self) -> std::io::Result<std::process::ExitStatus> { + let mut stopped = self.stopped.clone(); + let _ = stopped.wait_for(|stopped| *stopped).await; + Err(std::io::Error::other("the libp2p swarm stopped")) + } +} + +#[cfg(test)] +mod tests { + use std::sync::atomic::{AtomicU64, Ordering}; + + use super::*; + use crate::document::replica::Epoch; + use crate::transport::{HeadsSummary, Status}; + + /// Memory ports are process-wide. + static NEXT_PORT: AtomicU64 = AtomicU64::new(40_000); + + /// A node on the in-memory transport, listening on a port of its own (or nowhere), + /// and its address with its peer ID (or just its peer ID). + fn memory_node_with( + listen: bool, + relays: Vec<Multiaddr>, + relay: bool, + ) -> (Libp2pTransport, Multiaddr) { + let port = NEXT_PORT.fetch_add(1, Ordering::Relaxed); + let key = Keypair::generate_ed25519(); + let address: Multiaddr = format!("/memory/{port}").parse().unwrap(); + let peer = key.public().to_peer_id(); + let transport = Libp2pTransport::start(Libp2pConfig { + key, + listen: if listen { + vec![address.clone()] + } else { + Vec::new() + }, + announce: Vec::new(), + relays, + relay, + mdns: false, + public: false, + memory: true, + }) + .unwrap(); + let address = if listen { + address.with_p2p(peer).unwrap() + } else { + Multiaddr::empty().with(Protocol::P2p(peer)) + }; + (transport, address) + } + + fn memory_node() -> (Libp2pTransport, Multiaddr) { + memory_node_with(true, Vec::new(), false) + } + + fn status() -> ParsedMessage { + ParsedMessage::Status(Status { + heads: HeadsSummary::of(Vec::new()), + epoch: Epoch::GENESIS, + epoch_seq: 3, + ready: true, + syncs_with_you: false, + }) + } + + /// Starts `transport`'s consumer, forwarding what it receives. + fn received( + transport: Arc<Libp2pTransport>, + ) -> mpsc::UnboundedReceiver<(String, ParsedMessage)> { + let (sender, receiver) = mpsc::unbounded_channel(); + tokio::spawn(async move { + transport + .start_consumer(Box::new(move |from, message, _| { + let sender = sender.clone(); + Box::pin(async move { + let _ = sender.send((from, message)); + Ok(()) + }) + })) + .await + }); + receiver + } + + #[tokio::test] + async fn messages_both_ways_over_one_connection() { + let (a, a_address) = memory_node(); + let (b, _) = memory_node(); + let (a, b) = (Arc::new(a), Arc::new(b)); + let mut at_a = received(Arc::clone(&a)); + let mut at_b = received(Arc::clone(&b)); + + assert_eq!( + b.resolve_seed(&a_address.to_string()).await.unwrap(), + a.local_id() + ); + b.send_message(a.local_id(), &status()).await.unwrap(); + let (from, message) = at_a.recv().await.unwrap(); + assert_eq!(from, b.local_id()); + assert!(matches!(message, ParsedMessage::Status(status) if status.epoch_seq == 3)); + + // a knows no address of b, but b's connection is there. + a.send_message(b.local_id(), &status()).await.unwrap(); + let (from, _) = at_b.recv().await.unwrap(); + assert_eq!(from, a.local_id()); + + assert_eq!(a.connection(b.local_id()), Some(Connection::Direct)); + assert_eq!(b.connection(a.local_id()), Some(Connection::Direct)); + let stats = a.stats().libp2p.unwrap(); + assert!(stats.connections.direct >= 1); + assert_eq!(stats.connections.relayed, 0); + } + + #[tokio::test] + async fn a_seed_without_a_peer_id_is_connected_to_to_find_out() { + let (a, a_address) = memory_node(); + let (b, _) = memory_node(); + let mut bare = a_address.clone(); + bare.pop(); + assert_eq!( + b.resolve_seed(&bare.to_string()).await.unwrap(), + a.local_id() + ); + // And b now knows where a is. + assert_eq!(b.locations(a.local_id()), [a_address.to_string()]); + } + + #[tokio::test] + async fn listen_addresses_carry_the_peer_id() { + let (a, a_address) = memory_node(); + let deadline = tokio::time::Instant::now() + Duration::from_secs(5); + while a.local_addresses().is_empty() && tokio::time::Instant::now() < deadline { + tokio::time::sleep(Duration::from_millis(10)).await; + } + assert_eq!(a.local_addresses(), [a_address.to_string()]); + assert_eq!( + a.locate(&a_address.to_string()).as_deref(), + Some(a.local_id()) + ); + } + + /// Waits up to 5 s for `condition`. + async fn eventually(mut condition: impl FnMut() -> bool) -> bool { + let deadline = tokio::time::Instant::now() + Duration::from_secs(5); + while !condition() { + if tokio::time::Instant::now() >= deadline { + return false; + } + tokio::time::sleep(Duration::from_millis(10)).await; + } + true + } + + /// A node learns where a peer that dialled it listens (identify), and can then + /// pass it on. + #[tokio::test] + async fn a_dialled_node_learns_where_the_dialler_listens() { + let (a, a_address) = memory_node(); + let (b, b_address) = memory_node(); + let (a, b) = (Arc::new(a), Arc::new(b)); + let _at_a = received(Arc::clone(&a)); + b.resolve_seed(&a_address.to_string()).await.unwrap(); + b.send_message(a.local_id(), &status()).await.unwrap(); + let b_id = b.local_id().to_owned(); + assert!( + eventually(|| a.locations(&b_id) == [b_address.to_string()]).await, + "a knows {:?} of b", + a.locations(&b_id) + ); + } + + #[tokio::test] + async fn announced_addresses_are_published_instead() { + let key = Keypair::generate_ed25519(); + let peer = key.public().to_peer_id(); + let port = NEXT_PORT.fetch_add(1, Ordering::Relaxed); + let announced: Multiaddr = "/ip4/203.0.113.7/tcp/9661".parse().unwrap(); + let transport = Libp2pTransport::start(Libp2pConfig { + key, + listen: vec![format!("/memory/{port}").parse().unwrap()], + announce: vec![announced.clone()], + relays: Vec::new(), + relay: false, + mdns: false, + public: false, + memory: true, + }) + .unwrap(); + assert_eq!( + transport.local_addresses(), + [announced.with_p2p(peer).unwrap().to_string()] + ); + } + + #[test] + fn local_addresses_are_left_out_unless_there_are_only_local_ones() { + let parse = |text: &str| text.parse::<Multiaddr>().unwrap(); + let loopback = parse("/ip4/127.0.0.1/tcp/1"); + let link = parse("/ip6/fe80::1/tcp/1"); + let lan = parse("/ip4/192.168.1.2/tcp/1"); + let wireguard = parse("/ip6/fd00::2/tcp/1"); + assert_eq!( + published(&[ + loopback.clone(), + link.clone(), + lan.clone(), + wireguard.clone() + ]), + [lan, wireguard] + ); + assert_eq!( + published(&[loopback.clone(), link.clone()]), + [loopback, link] + ); + } + + /// b listens nowhere but through the relay r; a reaches it there. + #[tokio::test] + async fn a_node_nobody_can_dial_is_reached_through_a_relay() { + let (_r, r_address) = memory_node_with(true, Vec::new(), true); + let (b, _) = memory_node_with(false, vec![r_address.clone()], false); + let b = Arc::new(b); + let mut at_b = received(Arc::clone(&b)); + assert!( + eventually(|| !b.local_addresses().is_empty()).await, + "b got no slot on the relay" + ); + let relayed = b.local_addresses()[0].clone(); + assert!( + relayed.starts_with(&format!("{r_address}/p2p-circuit/")), + "{relayed}" + ); + + assert!(eventually(|| b.stats().libp2p.unwrap().relay_slots.held == 1).await); + let slots = b.stats().libp2p.unwrap().relay_slots; + assert_eq!((slots.granted, slots.not_granted), (1, 0)); + + let (a, _) = memory_node(); + assert_eq!(a.resolve_seed(&relayed).await.unwrap(), b.local_id()); + a.send_message(b.local_id(), &status()).await.unwrap(); + let (from, _) = at_b.recv().await.unwrap(); + assert_eq!(from, a.local_id()); + assert!(a.connection(b.local_id()).is_some()); + } + + fn peer() -> PeerId { + Keypair::generate_ed25519().public().to_peer_id() + } + + #[test] + fn relays_are_picked_when_needed() { + use ::libp2p::autonat::NatStatus; + let local = peer(); + let offers: HashMap<PeerId, Multiaddr> = (0..4) + .map(|port| (peer(), format!("/memory/{port}").parse().unwrap())) + .collect(); + let check = |direct, nat: &NatStatus, using: &[PeerId]| { + relay_check(local, direct, nat, &offers, using) + }; + // Dialable, or not known yet with an address to dial: none. + assert!(check(true, &NatStatus::Public(Multiaddr::empty()), &[]).is_empty()); + assert!(check(false, &NatStatus::Public(Multiaddr::empty()), &[]).is_empty()); + assert!(check(true, &NatStatus::Unknown, &[]).is_empty()); + // No address to dial, or AutoNAT says it can't be dialled: some. + assert_eq!(check(false, &NatStatus::Unknown, &[]).len(), AUTO_RELAYS); + let picked = check(true, &NatStatus::Private, &[]); + assert_eq!(picked.len(), AUTO_RELAYS); + // One in use: the next in the same order. + let more = check(true, &NatStatus::Private, &[picked[0].0]); + assert_eq!(more.len(), AUTO_RELAYS - 1); + assert_eq!(more[0].0, picked[1].0); + let all: Vec<PeerId> = picked.iter().map(|(relay, _)| *relay).collect(); + assert!(check(true, &NatStatus::Private, &all).is_empty()); + // Never itself. + let only_us = HashMap::from([(local, Multiaddr::empty())]); + assert!(relay_check(local, false, &NatStatus::Private, &only_us, &[]).is_empty()); + } + + /// A relay says so in its addresses; b, which nobody can dial, learns of it (as from + /// the member list) and becomes reachable through it. + #[tokio::test] + async fn a_member_that_relays_is_picked() { + let (r, _) = memory_node_with(true, Vec::new(), true); + assert!(eventually(|| !r.local_addresses().is_empty()).await); + let offer = r + .local_addresses() + .into_iter() + .find(|address| address.ends_with("/p2p-circuit")) + .expect("the relay offers itself"); + let (b, _) = memory_node_with(false, Vec::new(), false); + assert_eq!(b.locate(&offer).as_deref(), Some(r.local_id())); + assert!( + eventually(|| b + .local_addresses() + .iter() + .any(|address| address.contains("/p2p-circuit/"))) + .await, + "b isn't reachable through r: {:?}", + b.local_addresses() + ); + } + + /// A peer on the LAN that speaks continuum becomes a seed once, at its address. + #[test] + fn lan_peers_become_seeds_once() { + let mut book = Book::default(); + let lan = peer(); + let address: Multiaddr = "/ip4/192.168.1.5/tcp/9661".parse().unwrap(); + book.discover(lan); // No address known: nothing. + assert!(book.discovered.is_empty()); + book.remember(lan, address.clone()); + book.discover(lan); + book.discover(lan); + assert_eq!(book.discovered, [address.with_p2p(lan).unwrap()]); + } + + #[test] + fn known_addresses() { + let mut book = Book::default(); + let (relay, us) = (peer(), peer()); + let direct: Multiaddr = "/ip6/fd00::1/tcp/9661".parse().unwrap(); + let relayed: Multiaddr = format!("/ip6/fd00::2/tcp/9661/p2p/{relay}/p2p-circuit/p2p/{us}") + .parse() + .unwrap(); + assert!(!book.knows(us, &[direct.clone()])); + book.remember(us, relayed.clone()); + book.remember(us, direct.clone()); + // In any order, with or without the peer ID. + assert!(book.knows(us, &[relayed.clone(), direct.clone()])); + assert!(book.knows(us, &[direct.clone().with_p2p(us).unwrap()])); + assert!(!book.knows(us, &["/ip6/fd00::3/tcp/9661".parse().unwrap()])); + } + + /// Peers nobody mentions any more are forgotten, but connected ones. + #[test] + fn stale_peers_are_forgotten() { + let start = tokio::time::Instant::now(); + let (old, current, connected) = (peer(), peer(), peer()); + let address: Multiaddr = "/ip6/fd00::1/tcp/9661".parse().unwrap(); + let mut book = Book::default(); + for peer in [old, current, connected] { + book.remember(peer, address.clone()); + } + book.relays.insert(old, address.clone()); + book.continuum.insert(old); + let later = start + STALE_AFTER; + // `current` was mentioned again since. + book.seen.insert(current, later); + assert_eq!(book.prune(later, |peer| *peer == connected), 1); + assert!(!book.addresses.contains_key(&old)); + assert!(!book.relays.contains_key(&old) && !book.continuum.contains(&old)); + assert!(book.addresses.contains_key(¤t)); + assert!(book.addresses.contains_key(&connected)); + assert_eq!(book.prune(later, |_| false), 1); // `connected`, once it isn't. + } + + /// A relay doesn't dial the nodes it relays for through itself. + #[test] + fn no_circuits_through_ourselves() { + let (us, other_relay, target) = (peer(), peer(), peer()); + let through = |relay: PeerId| -> Multiaddr { + format!("/ip6/fd00::1/tcp/9661/p2p/{relay}/p2p-circuit") + .parse() + .unwrap() + }; + let direct: Multiaddr = "/ip6/fd00::2/tcp/9661".parse().unwrap(); + assert_eq!(relay_of(&through(us)), Some(us)); + assert_eq!(relay_of(&direct), None); + assert_eq!(relay_of(&direct.clone().with_p2p(target).unwrap()), None); + assert_eq!( + dialable(&[through(us), through(other_relay), direct.clone()], us), + [through(other_relay), direct] + ); + } + + #[tokio::test] + async fn unknown_peers_are_refused() { + let (b, _) = memory_node(); + let stranger = Keypair::generate_ed25519() + .public() + .to_peer_id() + .to_string(); + assert!(b.send_message(&stranger, &status()).await.is_err()); + assert!(b.resolve_seed(&stranger).await.is_err()); + assert_eq!(b.connection(&stranger), None); + } + + #[test] + fn a_direct_connection_wins_over_a_relayed_one() { + let mut book = Book::default(); + let (relayed_only, both) = (peer(), peer()); + let mut id = 0; + let mut connect = |book: &mut Book, peer: PeerId, relayed: bool| { + id += 1; + book.connections + .insert(ConnectionId::new_unchecked(id), (peer, relayed)); + }; + connect(&mut book, relayed_only, true); + connect(&mut book, both, true); + connect(&mut book, both, false); + assert_eq!(book.connection(&relayed_only), Some(Connection::Relayed)); + assert_eq!(book.connection(&both), Some(Connection::Direct)); + assert_eq!(book.connection(&peer()), None); + let connections = book.stats().connections; + assert_eq!((connections.direct, connections.relayed), (1, 2)); + } +} |