1 files changed, 234 insertions, 0 deletions
diff --git a/src/transport/libp2p/relays.rs b/src/transport/libp2p/relays.rs
new file mode 100644
index 0000000..3463408
--- /dev/null
+++ b/src/transport/libp2p/relays.rs
@@ -0,0 +1,234 @@
+//! Reaching peers past NATs: picking relays and keeping slots on them, upgrading
+//! relayed connections to direct ones, retrying hole punches, and finding peers on the
+//! DHT.
+
+use super::*;
+
+/// How a node relates to relays.
+pub(super) struct Relaying {
+    /// It relays for others.
+    pub(super) serving: bool,
+    /// Public relays may be picked, after members.
+    pub(super) public: bool,
+    pub(super) 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.
+pub(super) 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 (configured ones first, then listed
+/// ones, members, public ones), and lets those picked go once AutoNAT says this node can
+/// be dialled.
+pub(super) 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, configured, listed, members, public) = {
+        let mut book = lock(book);
+        let now = tokio::time::Instant::now();
+        book.failed_relays.retain(|_, until| now < *until);
+        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.named(Named::Configured)),
+            usable(&book.named(Named::Listed)),
+            usable(&book.relays),
+            public,
+        )
+    };
+    let mut using: Vec<PeerId> = picked.values().copied().collect();
+    // Each kind for what the ones before left.
+    for (kind, offers) in [
+        (Named::Configured.as_str(), configured),
+        (Named::Listed.as_str(), listed),
+        ("member", members),
+        ("public", public),
+    ] {
+        for (relay, address) in relay_check(relaying.local, direct, nat, &offers, &using) {
+            using.push(relay);
+            match swarm.listen_on(address.clone().with(Protocol::P2pCircuit)) {
+                Ok(listener) => {
+                    info!(relay = %Addressed(SCHEME, &relay), address = %address, kind, "Picked a relay");
+                    picked.insert(listener, relay);
+                }
+                Err(error) => {
+                    debug!(relay = %Addressed(SCHEME, &relay), kind, error = %error, "Couldn't use a relay");
+                }
+            }
+        }
+    }
+}
+
+/// Dials `peer` directly if it's connected only through a relay and is known elsewhere
+/// too (the member list, identify, mDNS), since the first connection, made through a
+/// relay (a relayed seed), would otherwise be kept for good: a peer connected isn't
+/// dialled again, and hole punching needs both sides to listen. The direct connection,
+/// once open, wins over the relayed one (`Book::duplicates`).
+pub(super) fn upgrade(
+    swarm: &mut Swarm<Behaviour>,
+    book: &Mutex<Book>,
+    peer: PeerId,
+    local: PeerId,
+) {
+    let direct: Vec<Multiaddr> = {
+        let mut book = lock(book);
+        if book.connection(&peer) != Some(Connection::Relayed) {
+            return;
+        }
+        let direct: Vec<Multiaddr> = book
+            .addresses
+            .get(&peer)
+            .map(|known| dialable(known, local))
+            .unwrap_or_default()
+            .into_iter()
+            .filter(|address| !is_relayed(address) && !ignored(&book.ignored, address))
+            .collect();
+        if direct.is_empty() {
+            return;
+        }
+        let now = tokio::time::Instant::now();
+        let (due, wait) = book
+            .upgrades
+            .get(&peer)
+            .copied()
+            .unwrap_or((now, UPGRADE_RETRY));
+        if now < due {
+            return;
+        }
+        book.upgrades
+            .insert(peer, (now + wait, (wait * 2).min(RELAY_PAUSE)));
+        direct
+    };
+    debug!(peer = %Addressed(SCHEME, &peer), addresses = ?direct, "Connected only through a relay; dialling directly");
+    let _ = swarm.dial(
+        DialOpts::peer_id(peer)
+            .addresses(direct)
+            .condition(PeerCondition::NotDialing)
+            .build(),
+    );
+}
+
+/// Tries hole punching again with the peers it failed with that are due (`punching`).
+/// libp2p punches as a relayed connection opens, started by the side that was dialled
+/// through the relay; that side is where a punch's failure is told, so it is this one.
+/// A new relayed connection starts a new punch: we dial the peer through a relay if it
+/// has a slot on one; else we close our relayed connections to it, and its next message
+/// (a probe, within a minute) dials us through ours again. Peers connected directly by
+/// now are dropped; those not connected now are kept, failures and all.
+pub(super) fn retry_punches(swarm: &mut Swarm<Behaviour>, book: &Mutex<Book>, local: PeerId) {
+    let now = tokio::time::Instant::now();
+    let mut close: Vec<ConnectionId> = Vec::new();
+    let due: Vec<(PeerId, usize, Vec<Multiaddr>)> = {
+        let mut book = lock(book);
+        let book = &mut *book;
+        let mut due = Vec::new();
+        book.punching.retain(|peer, (failed, next)| {
+            match book_connection(&book.connections, peer) {
+                Some(Connection::Direct) => return false,
+                // Kept, failures and all: one we closed to have it dial anew comes back.
+                None => return true,
+                Some(Connection::Relayed) if now < *next => return true,
+                Some(Connection::Relayed) => {}
+            }
+            let through: Vec<Multiaddr> = book
+                .addresses
+                .get(peer)
+                .map(|known| dialable(known, local))
+                .unwrap_or_default()
+                .into_iter()
+                .filter(is_relayed)
+                .collect();
+            // Until its punch fails again (or the peer tries from its side).
+            *next = now + PUNCH_RETRIES[(*failed).min(PUNCH_RETRIES.len() - 1)];
+            if through.is_empty() {
+                close.extend(
+                    book.connections
+                        .iter()
+                        .filter(|(_, (connected, relayed, _))| connected == peer && *relayed)
+                        .map(|(id, _)| *id),
+                );
+            } else {
+                due.push((*peer, *failed, through));
+            }
+            true
+        });
+        due
+    };
+    for connection in close {
+        debug!(connection = %connection, "Hole punching again: closing a relayed connection, for the peer to dial anew");
+        swarm.close_connection(connection);
+    }
+    for (peer, failed, through) in due {
+        debug!(peer = %Addressed(SCHEME, &peer), failed, "Hole punching again, through a new relayed connection");
+        let _ = swarm.dial(
+            DialOpts::peer_id(peer)
+                .addresses(through)
+                .condition(PeerCondition::Always)
+                .build(),
+        );
+    }
+}
+
+/// Asks the DHT (with `--libp2p-public`) where `peer` is, for the next attempt.
+pub(super) 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);
+    }
+}