add identity layer

1 files changed, 510 insertions, 0 deletions

diff --git a/src/transport/identity.rs b/src/transport/identity.rs
new file mode 100644
index 0000000..712e1f9
--- /dev/null
+++ b/src/transport/identity.rs
@@ -0,0 +1,510 @@
+//! The identity layer: between the node and a transport whose identities are addresses
+//! (Mycelium's public keys), so that the rest of the node deals in node IDs
+//! (`crate::identity`) only, whatever transport reaches a peer.
+//!
+//! A node proves who it is to an address with WHO (which asks back) or ME (the answer):
+//! its public node key, and a signature over its own address and the recipient's
+//! (`signed`). The transport authenticates the sender's address, so a proof is good for
+//! that pair of addresses only: replayed from another address, or to another recipient,
+//! it fails. Once a node proved itself at an address, messages from there are its.
+//!
+//! * A message from an address no node proved itself at is dropped, and the address is
+//!   asked WHO (at most every `ASK_INTERVAL`): after a restart, the first message from
+//!   a peer brings its proof back, and the peer's node retries what was lost.
+//! * A seed (`resolve_seed`) is an address, or something the transport resolves to one;
+//!   it is asked WHO, and its ME says which node is there.
+//! * Where other nodes are comes from the member list and HELLO_REPLYs
+//!   (`learn_addresses`): hints, not proofs. A message to a node goes to the address it
+//!   proved itself at; with only hints, those are asked WHO, and the send is refused
+//!   until one answers (the node retries refused sends).
+//!
+//! Addresses are `<scheme>:<transport identity>`, e.g. `mycelium:<public key>`.
+
+use std::collections::HashMap;
+use std::sync::{Arc, Mutex, MutexGuard};
+use std::time::Duration;
+
+use anyhow::{Result, anyhow, bail};
+use async_trait::async_trait;
+use tokio::sync::oneshot;
+use tokio::time::Instant;
+use tracing::{debug, info};
+
+use crate::identity::{self, NodeKey};
+use crate::locks::MutexExt;
+use crate::transport::traffic::Traffic;
+use crate::transport::{
+    EncodedMessage, HandlerFn, IdentityProof, MessageHandler, MessageKind, ParsedMessage, Transport,
+};
+
+pub(crate) const WHO_LABEL: &[u8] = b"continuum/who/1";
+pub(crate) const ME_LABEL: &[u8] = b"continuum/me/1";
+/// An address is asked WHO at most this often, unless a seed is being resolved.
+const ASK_INTERVAL: Duration = Duration::from_secs(5);
+/// How long resolving a seed waits for its ME.
+const RESOLVE_TIMEOUT: Duration = Duration::from_secs(10);
+/// Addresses kept per node from hints: about one per transport.
+const MAX_HINTS: usize = 8;
+/// Addresses with a proven node, nodes with hints, and addresses asked, each at most:
+/// anyone can send us proofs (keys are free), so the books are bounded.
+const MAX_ENTRIES: usize = 4096;
+
+/// What a node under the identity layer needs (`node::NodeConfig::identity`).
+#[derive(Clone)]
+pub(crate) struct IdentityConfig {
+    pub(crate) key: Arc<NodeKey>,
+    /// The transport's address scheme, e.g. `mycelium`.
+    pub(crate) scheme: &'static str,
+}
+
+/// The bytes a WHO (`label`: `WHO_LABEL`) or ME proof signs: the label, then the
+/// sender's and the recipient's address, each with its length.
+pub(crate) fn signed(label: &[u8], from: &str, to: &str) -> Vec<u8> {
+    let mut bytes = label.to_vec();
+    for address in [from, to] {
+        bytes.extend_from_slice(&(address.len() as u32).to_be_bytes());
+        bytes.extend_from_slice(address.as_bytes());
+    }
+    bytes
+}
+
+pub(crate) struct IdentityTransport {
+    shared: Arc<Shared>,
+}
+
+struct Shared {
+    inner: Arc<dyn Transport>,
+    scheme: &'static str,
+    key: Arc<NodeKey>,
+    node_id: String,
+    /// WHO and ME are counted here, under the address (`traffic`).
+    traffic: Arc<Traffic>,
+    // A std mutex: every access is short and never spans an await.
+    book: Mutex<Book>,
+}
+
+/// Who is where. Keys are transport identities (an address without its scheme).
+#[derive(Default)]
+struct Book {
+    /// The node that proved itself at a transport identity, and when.
+    at: HashMap<String, (String, Instant)>,
+    /// The transport identity each node last proved itself at.
+    proven: HashMap<String, String>,
+    /// Where others say a node is.
+    hints: HashMap<String, Vec<String>>,
+    /// When each transport identity was last asked WHO.
+    asked: HashMap<String, Instant>,
+    /// Seeds being resolved: who waits for a ME from where.
+    waiting: HashMap<String, Vec<oneshot::Sender<String>>>,
+}
+
+impl Book {
+    /// `node` proved itself at `target`. Returns whether that is news.
+    fn bind(&mut self, target: &str, node: &str, now: Instant) -> bool {
+        let news = self.at.get(target).is_none_or(|(known, _)| known != node);
+        // One address per node and one node per address: a node that moved, or an
+        // address taken over by another node, leaves no stale entry behind.
+        if let Some(previous) = self.proven.insert(node.to_owned(), target.to_owned())
+            && previous != target
+        {
+            self.at.remove(&previous);
+        }
+        if let Some((previous, _)) = self.at.insert(target.to_owned(), (node.to_owned(), now))
+            && previous != node
+            && self.proven.get(&previous).is_some_and(|at| at == target)
+        {
+            self.proven.remove(&previous);
+        }
+        if self.at.len() > MAX_ENTRIES
+            && let Some(oldest) = self
+                .at
+                .iter()
+                .min_by_key(|(_, (_, at))| *at)
+                .map(|(target, _)| target.clone())
+            && let Some((node, _)) = self.at.remove(&oldest)
+        {
+            self.proven.remove(&node);
+        }
+        for waiter in self.waiting.remove(target).into_iter().flatten() {
+            let _ = waiter.send(node.to_owned());
+        }
+        news
+    }
+
+    /// Whether to ask `target` WHO now (always, if `force`), noting that it was asked.
+    fn ask(&mut self, target: &str, now: Instant, force: bool) -> bool {
+        if !force
+            && self
+                .asked
+                .get(target)
+                .is_some_and(|at| now.duration_since(*at) < ASK_INTERVAL)
+        {
+            return false;
+        }
+        if self.asked.len() >= MAX_ENTRIES {
+            self.asked
+                .retain(|_, at| now.duration_since(*at) < ASK_INTERVAL);
+        }
+        self.asked.insert(target.to_owned(), now);
+        true
+    }
+}
+
+impl IdentityTransport {
+    pub(crate) fn new(
+        inner: Arc<dyn Transport>,
+        config: IdentityConfig,
+        traffic: Arc<Traffic>,
+    ) -> Self {
+        let node_id = config.key.id();
+        Self {
+            shared: Arc::new(Shared {
+                inner,
+                scheme: config.scheme,
+                key: config.key,
+                node_id,
+                traffic,
+                book: Mutex::default(),
+            }),
+        }
+    }
+}
+
+impl Shared {
+    fn lock(&self) -> MutexGuard<'_, Book> {
+        self.book.lock_or_recover()
+    }
+
+    fn address(&self, target: &str) -> String {
+        format!("{}:{target}", self.scheme)
+    }
+
+    /// The transport identity in `address`, if it is one of this transport's.
+    fn target_of<'a>(&self, address: &'a str) -> Option<&'a str> {
+        address
+            .strip_prefix(self.scheme)
+            .and_then(|rest| rest.strip_prefix(':'))
+            .filter(|target| self.inner.is_peer_id(target))
+    }
+
+    fn proof(&self, label: &[u8], target: &str) -> IdentityProof {
+        let message = signed(
+            label,
+            &self.address(self.inner.local_id()),
+            &self.address(target),
+        );
+        IdentityProof {
+            key: self.key.public(),
+            signature: self.key.sign(&message),
+        }
+    }
+
+    /// Notes the node `proof` proves to be at `sender`; None if it proves nothing.
+    fn prove(&self, sender: &str, label: &[u8], proof: &IdentityProof) -> Option<String> {
+        let message = signed(
+            label,
+            &self.address(sender),
+            &self.address(self.inner.local_id()),
+        );
+        if !identity::verify(&proof.key, &message, &proof.signature) {
+            debug!(from = %self.address(sender), "Ignoring an identity proof that doesn't hold");
+            return None;
+        }
+        let node = identity::node_id_of(&proof.key);
+        if node == self.node_id {
+            debug!(from = %self.address(sender), "Ignoring a proof of our own node key");
+            return None;
+        }
+        if self.lock().bind(sender, &node, Instant::now()) {
+            debug!(node = %node, address = %self.address(sender), "Node proved its address");
+        }
+        Some(node)
+    }
+
+    /// Asks `target` WHO, unless it was asked recently (`force`: anyway).
+    fn ask(self: &Arc<Self>, target: &str, force: bool) {
+        if !self.lock().ask(target, Instant::now(), force) {
+            return;
+        }
+        let shared = Arc::clone(self);
+        let target = target.to_owned();
+        tokio::spawn(async move { shared.send_proof(MessageKind::Who, &target).await });
+    }
+
+    /// Sends a WHO or ME (`kind`) to `target`.
+    async fn send_proof(&self, kind: MessageKind, target: &str) {
+        let message = if kind == MessageKind::Who {
+            ParsedMessage::Who(self.proof(WHO_LABEL, target))
+        } else {
+            ParsedMessage::Me(self.proof(ME_LABEL, target))
+        };
+        let encoded = match self.inner.encode(target, &message) {
+            Ok(encoded) => encoded,
+            Err(error) => {
+                debug!(to = %target, error = %format!("{error:#}"), "Couldn't encode an identity proof");
+                return;
+            }
+        };
+        let address = self.address(target);
+        match self.inner.send_encoded(&encoded).await {
+            Ok(()) => self
+                .traffic
+                .sent(&address, encoded.kind, encoded.payload_len),
+            Err(error) => {
+                self.traffic
+                    .refused(&address, encoded.kind, encoded.payload_len);
+                debug!(to = %address, error = %format!("{error:#}"), "Couldn't send an identity proof");
+            }
+        }
+    }
+
+    async fn receive(
+        self: Arc<Self>,
+        sender: String,
+        message: ParsedMessage,
+        payload_len: usize,
+        handler: &HandlerFn,
+    ) -> Result<()> {
+        match message {
+            ParsedMessage::Who(proof) => {
+                self.traffic
+                    .received(&self.address(&sender), MessageKind::Who, payload_len);
+                if self.prove(&sender, WHO_LABEL, &proof).is_some() {
+                    let shared = Arc::clone(&self);
+                    tokio::spawn(async move { shared.send_proof(MessageKind::Me, &sender).await });
+                }
+                Ok(())
+            }
+            ParsedMessage::Me(proof) => {
+                self.traffic
+                    .received(&self.address(&sender), MessageKind::Me, payload_len);
+                self.prove(&sender, ME_LABEL, &proof);
+                Ok(())
+            }
+            message => {
+                let node = self.lock().at.get(&sender).map(|(node, _)| node.clone());
+                match node {
+                    Some(node) => handler(node, message, payload_len).await,
+                    None => {
+                        debug!(
+                            from = %self.address(&sender),
+                            kind = ?message.kind(),
+                            "Dropping a message from an address no node proved itself at; asking who is there"
+                        );
+                        self.ask(&sender, false);
+                        Ok(())
+                    }
+                }
+            }
+        }
+    }
+}
+
+#[async_trait]
+impl Transport for IdentityTransport {
+    fn local_id(&self) -> &str {
+        &self.shared.node_id
+    }
+
+    fn is_peer_id(&self, id: &str) -> bool {
+        identity::is_node_id(id)
+    }
+
+    fn delivery_limit(&self) -> Duration {
+        self.shared.inner.delivery_limit()
+    }
+
+    fn local_addresses(&self) -> Vec<String> {
+        vec![self.shared.address(self.shared.inner.local_id())]
+    }
+
+    fn addresses_of(&self, peer: &str) -> Vec<String> {
+        let book = self.shared.lock();
+        match book.proven.get(peer) {
+            Some(target) => vec![self.shared.address(target)],
+            None => book
+                .hints
+                .get(peer)
+                .into_iter()
+                .flatten()
+                .map(|target| self.shared.address(target))
+                .collect(),
+        }
+    }
+
+    fn learn_addresses(&self, peer: &str, addresses: &[String]) {
+        if peer == self.shared.node_id || !identity::is_node_id(peer) {
+            return;
+        }
+        let local = self.shared.inner.local_id();
+        let mut targets: Vec<String> = addresses
+            .iter()
+            .filter_map(|address| self.shared.target_of(address))
+            .filter(|target| *target != local)
+            .map(str::to_owned)
+            .collect();
+        targets.dedup();
+        targets.truncate(MAX_HINTS);
+        if targets.is_empty() {
+            return;
+        }
+        let mut book = self.shared.lock();
+        if book.hints.len() < MAX_ENTRIES || book.hints.contains_key(peer) {
+            book.hints.insert(peer.to_owned(), targets);
+        }
+    }
+
+    /// A node ID resolves to itself if this node knows where it is; anything else is
+    /// an address (`<scheme>:...`) or what the transport resolves to one, asked WHO.
+    async fn resolve_seed(&self, seed: &str) -> Result<String> {
+        if identity::is_node_id(seed) {
+            let book = self.shared.lock();
+            if book.proven.contains_key(seed) || book.hints.contains_key(seed) {
+                return Ok(seed.to_owned());
+            }
+            bail!("a node ID alone doesn't say where that node is; give its address");
+        }
+        let target = match self.shared.target_of(seed) {
+            Some(target) => target.to_owned(),
+            None => self.shared.inner.resolve_seed(seed).await?,
+        };
+        if target == self.shared.inner.local_id() {
+            bail!("{seed} is this node");
+        }
+        let receiver = {
+            let mut book = self.shared.lock();
+            if let Some((node, _)) = book.at.get(&target) {
+                return Ok(node.clone());
+            }
+            let (sender, receiver) = oneshot::channel();
+            let waiting = book.waiting.entry(target.clone()).or_default();
+            waiting.retain(|waiter| !waiter.is_closed());
+            waiting.push(sender);
+            receiver
+        };
+        self.shared.ask(&target, true);
+        match tokio::time::timeout(RESOLVE_TIMEOUT, receiver).await {
+            Ok(Ok(node)) => {
+                info!(seed = %seed, node = %node, "Seed is node");
+                Ok(node)
+            }
+            _ => Err(anyhow!(
+                "no node answered at {} within {RESOLVE_TIMEOUT:?}",
+                self.shared.address(&target)
+            )),
+        }
+    }
+
+    /// For the address `peer` proved itself at, or else one it is said to be at (a send
+    /// there is refused until it proves itself, `send_encoded`).
+    fn encode(&self, peer: &str, message: &ParsedMessage) -> Result<EncodedMessage> {
+        let target = {
+            let book = self.shared.lock();
+            book.proven
+                .get(peer)
+                .or_else(|| book.hints.get(peer).and_then(|hints| hints.first()))
+                .cloned()
+        };
+        let Some(target) = target else {
+            bail!("no address known for {peer}");
+        };
+        let mut encoded = self.shared.inner.encode(&target, message)?;
+        encoded.peer = peer.to_owned();
+        Ok(encoded)
+    }
+
+    async fn send_encoded(&self, message: &EncodedMessage) -> Result<()> {
+        let hints = {
+            let book = self.shared.lock();
+            if book.proven.get(&message.peer) == Some(&message.address) {
+                None
+            } else {
+                Some(book.hints.get(&message.peer).cloned().unwrap_or_default())
+            }
+        };
+        if let Some(hints) = hints {
+            for target in hints {
+                self.shared.ask(&target, false);
+            }
+            bail!(
+                "{} hasn't proven where it is yet; asked where it is said to be",
+                message.peer
+            );
+        }
+        let mut inner = message.clone();
+        inner.peer = message.address.clone();
+        self.shared.inner.send_encoded(&inner).await
+    }
+
+    async fn start_consumer(&self, handler: MessageHandler) -> Result<()> {
+        let handler: Arc<HandlerFn> = Arc::from(handler);
+        let shared = Arc::clone(&self.shared);
+        self.shared
+            .inner
+            .start_consumer(Box::new(move |sender, message, payload_len| {
+                let shared = Arc::clone(&shared);
+                let handler = Arc::clone(&handler);
+                Box::pin(async move {
+                    shared
+                        .receive(sender, message, payload_len, &*handler)
+                        .await
+                })
+            }))
+            .await
+    }
+
+    async fn wait_for_exit(&self) -> std::io::Result<std::process::ExitStatus> {
+        self.shared.inner.wait_for_exit().await
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn signed_bytes_say_both_addresses() {
+        // A length before each address: "a:bc" to "d" isn't "a:b" to "cd".
+        assert_ne!(
+            signed(WHO_LABEL, "a:bc", "d"),
+            signed(WHO_LABEL, "a:b", "cd")
+        );
+        assert_ne!(signed(WHO_LABEL, "a", "b"), signed(ME_LABEL, "a", "b"));
+        assert_ne!(signed(WHO_LABEL, "a", "b"), signed(WHO_LABEL, "b", "a"));
+    }
+
+    #[test]
+    fn one_address_per_node_and_one_node_per_address() {
+        let now = Instant::now();
+        let mut book = Book::default();
+        assert!(book.bind("x", "n1", now));
+        assert!(!book.bind("x", "n1", now));
+        // n1 moved to y: x is no longer n1's.
+        assert!(book.bind("y", "n1", now));
+        assert!(!book.at.contains_key("x"));
+        assert_eq!(book.proven["n1"], "y");
+        // n2 took over y: n1 is nowhere proven.
+        assert!(book.bind("y", "n2", now));
+        assert!(!book.proven.contains_key("n1"));
+        assert_eq!(book.at["y"].0, "n2");
+    }
+
+    #[test]
+    fn asking_is_rate_limited() {
+        let now = Instant::now();
+        let mut book = Book::default();
+        assert!(book.ask("x", now, false));
+        assert!(!book.ask("x", now, false));
+        assert!(book.ask("x", now, true));
+        assert!(book.ask("x", now + ASK_INTERVAL, false));
+    }
+
+    #[tokio::test]
+    async fn a_waiting_seed_learns_the_node() {
+        let mut book = Book::default();
+        let (sender, receiver) = oneshot::channel();
+        book.waiting.insert("x".into(), vec![sender]);
+        book.bind("x", "n1", Instant::now());
+        assert_eq!(receiver.await.unwrap(), "n1");
+    }
+}