Merge branch 'refactor-node' into refactor-node-next

Carries the leg-embed storage move: PeerMachine.leg replaces the
Node.connections map. Hand-resolved onto the XX surfaces keeping this
line's semantics with the new storage: the msg3 termination arms and
executor swap/rekey-responder teardowns keep their exact cleanup sets
and index-free behavior; the msg2 cross-connection extract takes the
leg before the unconditional machine dispose; msg1 and anonymous leg
births embed the leg at machine insertion; the rekey-vs-establish gate
tests leg-absence. The merge also drops the now-redundant explicit
machine inserts next to the seeded test seam, whose auto-merged
combination clobbered leg-carrying machines.
This commit is contained in:
Johnathan Corgan
2026-07-17 05:07:36 +00:00
11 changed files with 328 additions and 317 deletions

View File

@@ -199,7 +199,6 @@ impl Node {
// (`handle_msg1` L665): the send error text is surfaced
// at the executor point where the failure is now handled.
warn!(link_id = %link, error = %e, "Failed to send msg2");
self.connections.remove(&link);
self.links.remove(&link);
self.addr_to_link
.remove(&(ambient.transport_id, ambient.remote_addr.clone()));
@@ -247,8 +246,7 @@ impl Node {
let wire_msg2 = if ambient.is_outbound {
None
} else {
self.connections
.get(&promote_link)
self.leg(&promote_link)
.and_then(|c| c.handshake_msg2().map(|m| m.to_vec()))
};
@@ -393,10 +391,11 @@ impl Node {
// records the reject but performs no link/index teardown
// and leaves the `pending_outbound` entry for the stale-
// connection reaper. The leg's machine must go with the
// leg, though: `promote_connection` consumed the
// `connections` entry before erring, so the reaper (which
// sweeps `connections`) can never reach this link's
// machine — dropping it here is the only disposal point.
// leg, though: `promote_connection` took the pending
// connection off the machine before erring, so the reaper
// (which sweeps the machines' embedded legs) can never
// reach this link's machine — dropping it here is the
// only disposal point.
warn!(
target: "fips::node::handlers::handshake",
link_id = %promote_link,
@@ -581,27 +580,22 @@ impl Node {
if our_inbound_wins {
// Larger node side: swap to the inbound session so it pairs
// with the peer's kept outbound session.
let inbound_session = match self
.connections
.get_mut(&link)
.and_then(|c| c.take_session())
{
Some(s) => s,
None => {
self.connections.remove(&link);
self.remove_link(&link);
self.remove_peer_machine(link);
self.stats_mut().record_reject(RejectReason::Handshake(
HandshakeReject::BadState,
));
return;
}
};
let inbound_session =
match self.leg_mut(&link).and_then(|c| c.take_session()) {
Some(s) => s,
None => {
self.remove_link(&link);
self.remove_peer_machine(link);
self.stats_mut().record_reject(RejectReason::Handshake(
HandshakeReject::BadState,
));
return;
}
};
if let Some(peer_ref) = self.peers.get_mut(&peer) {
let old_our_index =
peer_ref.replace_session(inbound_session, our_index, their_index);
let Some(transport_id) = peer_ref.transport_id() else {
self.connections.remove(&link);
self.remove_link(&link);
self.remove_peer_machine(link);
self.stats_mut().record_reject(RejectReason::Handshake(
@@ -636,8 +630,8 @@ impl Node {
// Both branches tear down the temporary inbound link fully
// (including its `addr_to_link` mapping) via `remove_link`,
// disposing the leg's machine with it.
self.connections.remove(&link);
// disposing the leg's machine (and its embedded connection)
// with it.
self.remove_link(&link);
self.remove_peer_machine(link);
return;
@@ -677,7 +671,7 @@ impl Node {
// Rekey: process as responder, store new session as pending.
let noise_session = {
let Some(conn) = self.connections.get_mut(&link) else {
let Some(conn) = self.leg_mut(&link) else {
warn!(link_id = %link, "Connection removed during rekey msg3 processing");
self.links.remove(&link);
self.remove_peer_machine(link);
@@ -694,7 +688,6 @@ impl Node {
Some(s) => s,
None => {
warn!("Rekey msg3: no session from handshake");
self.connections.remove(&link);
self.links.remove(&link);
self.remove_peer_machine(link);
self.stats_mut()
@@ -713,13 +706,13 @@ impl Node {
self.peers_by_index
.insert((ambient.transport_id, our_new_index.as_u32()), peer);
// Clean up: remove the temporary connection/link and the leg's
// machine (the established peer keeps its own, keyed by its own
// link). Do NOT remove addr_to_link — the entry must remain
// pointing to the original link so the established peer stays
// routable, so this uses the bare `links.remove` rather than
// the full `remove_link`.
self.connections.remove(&link);
// Clean up: remove the temporary link and the leg's machine
// (dropping its embedded connection; the established peer
// keeps its own machine, keyed by its own link). Do NOT
// remove addr_to_link the entry must remain pointing to
// the original link so the established peer stays routable,
// so this uses the bare `links.remove` rather than the full
// `remove_link`.
self.links.remove(&link);
self.remove_peer_machine(link);

View File

@@ -299,22 +299,21 @@ impl Node {
self.links.insert(link_id, link);
self.addr_to_link.insert(addr_key, link_id);
self.connections.insert(link_id, conn);
// The leg's persistent control machine is born alongside its pending
// Build the msg2 response, storing it on the connection for potential
// resend before the connection is embedded on the machine below.
let wire_msg2 = build_msg2(our_index, header.sender_idx, &msg2_response);
conn.set_handshake_msg2(wire_msg2.clone());
// The leg's persistent control machine is born carrying its pending
// connection, parked at `SentMsg2` awaiting msg3 (identity is unknown
// until then). Inserted before the msg2 send below so no suspension
// point observes a leg without a machine. `handle_msg3` steps this
// same machine; every teardown path disposes it with the leg.
self.peer_machines.insert(
link_id,
PeerMachine::inbound_msg2_sent(link_id, our_index, packet.timestamp_ms),
);
// Build and send msg2 response, storing for potential resend
let wire_msg2 = build_msg2(our_index, header.sender_idx, &msg2_response);
if let Some(conn) = self.connections.get_mut(&link_id) {
conn.set_handshake_msg2(wire_msg2.clone());
}
// point observes a leg in flight without a machine. `handle_msg3`
// steps this same machine; every teardown path disposes it with the
// embedded leg.
let mut machine = PeerMachine::inbound_msg2_sent(link_id, our_index, packet.timestamp_ms);
machine.set_leg(conn);
self.peer_machines.insert(link_id, machine);
if let Some(transport) = self.transports.get(&packet.transport_id) {
match transport.send(&packet.remote_addr, &wire_msg2).await {
@@ -333,8 +332,8 @@ impl Node {
error = %e,
"Failed to send msg2"
);
// Clean up on failure
self.connections.remove(&link_id);
// Clean up on failure (the machine disposal drops the
// embedded connection with it)
self.links.remove(&link_id);
self.addr_to_link
.remove(&(packet.transport_id, packet.remote_addr));
@@ -362,7 +361,7 @@ impl Node {
/// (if already promoted).
fn find_stored_msg2(&self, link_id: LinkId) -> Option<Vec<u8>> {
// Check pending connection first
if let Some(conn) = self.connections.get(&link_id)
if let Some(conn) = self.leg(&link_id)
&& let Some(msg2) = conn.handshake_msg2()
{
return Some(msg2.to_vec());
@@ -412,9 +411,17 @@ impl Node {
};
// Check if this is a rekey msg2: the handshake state is on the
// ActivePeer (not a PeerConnection), so self.connections won't have it.
// Look for a peer with matching rekey_our_index.
if !self.connections.contains_key(&link_id) {
// ActivePeer (not a PeerConnection), so the link's machine — if one
// survives at all — carries no pending connection. A bare machine
// lookup would NOT discriminate here: an established peer's machine
// stays keyed by this link, so the pending connection's presence is
// what marks a fresh establish. Look for a peer with matching
// rekey_our_index.
if self
.peer_machines
.get(&link_id)
.is_none_or(|machine| machine.leg().is_none())
{
let noise_msg2 = &packet.data[header.noise_msg2_offset..];
// Find peer with rekey in progress for this index
@@ -592,7 +599,7 @@ impl Node {
let our_profile = self.node_profile();
let (peer_identity, msg3_bytes, our_index) = {
let Some(conn) = self.connections.get_mut(&link_id) else {
let Some(conn) = self.leg_mut(&link_id) else {
warn!(link_id = %link_id, "Connection removed during msg2 processing");
self.pending_outbound.remove(&key);
self.stats_mut()
@@ -672,8 +679,8 @@ impl Node {
.is_err()
{
self.pending_outbound.remove(&key);
self.connections.remove(&link_id);
// Drop the machine persisted at dial — this leg never promotes.
// Drop the machine persisted at dial — this leg never promotes,
// and its pending connection is dropped with it.
self.remove_peer_machine(link_id);
self.remove_link(&link_id);
self.stats_mut()
@@ -687,10 +694,10 @@ impl Node {
// identified dial misdirected at ourselves lands here too (the
// learned identity overwrites the dial-time expectation and is
// never compared against it). This leg never promotes; its machine
// goes with it. The index, link, and `pending_outbound` entry are
// deliberately NOT freed here (pre-existing shape).
// goes with it (dropping the embedded pending connection). The
// index, link, and `pending_outbound` entry are deliberately NOT
// freed here (pre-existing shape).
debug!(link_id = %link_id, "Discovered self via shared-media beacon, dropping");
self.connections.remove(&link_id);
self.remove_peer_machine(link_id);
self.stats_mut()
.record_reject(RejectReason::Handshake(HandshakeReject::BadState));
@@ -718,7 +725,7 @@ impl Node {
error = %e,
"Failed to send msg3"
);
if let Some(conn) = self.connections.get_mut(&link_id) {
if let Some(conn) = self.leg_mut(&link_id) {
conn.mark_failed();
}
self.stats_mut()
@@ -746,11 +753,19 @@ impl Node {
// This ensures both nodes use the same Noise handshake (the winner's
// outbound = the loser's inbound).
if self.peers.contains_key(&peer_node_addr) {
// The dial-persisted outbound machine is not consumed by the inline
// Swap/Keep resolution below (which mutates the existing promoted peer
// directly, with no machine). Drop it on entry so none of this block's
// exits leave a dangling machine — matching the pre-persistence path,
// Extract the outbound connection from its machine FIRST — the
// machine owns it, so disposing the machine before the take would
// destroy the connection. The dial-persisted outbound machine is
// not consumed by the inline Swap/Keep resolution below (which
// mutates the existing promoted peer directly, with no machine),
// so drop it right after the take — unconditionally, whether or
// not a connection was carried — so none of this block's exits
// leave a dangling machine. Matches the pre-persistence path,
// which created no machine for a cross-connection.
let taken_conn = self
.peer_machines
.get_mut(&link_id)
.and_then(|machine| machine.take_leg());
self.remove_peer_machine(link_id);
let our_outbound_wins = cross_connection_winner(
self.identity().node_addr(),
@@ -758,8 +773,7 @@ impl Node {
true, // this IS our outbound
);
// Extract the outbound connection
let mut conn = match self.connections.remove(&link_id) {
let mut conn = match taken_conn {
Some(c) => c,
None => {
self.pending_outbound.remove(&key);
@@ -923,7 +937,7 @@ impl Node {
}
None => {
// A miss is a state-machine inconsistency (e.g. a test seeding
// `connections`/`pending_outbound` directly): rebuild the
// a connection/`pending_outbound` entry directly): rebuild the
// machine defensively and persist it, so the promotion feedback
// below still finds it and the promoted peer keeps a machine.
debug_assert!(
@@ -946,7 +960,7 @@ impl Node {
// The outbound Msg2 Promote step cancels the two dial-armed handshake
// timers (the machine survives promotion, so they would otherwise linger
// in `peer_timers` until `drive_peer_timers` lazily discards them — the
// promoted leg's `connections` entry is gone and the machine has left
// promoted leg's pending connection is consumed and the machine has left
// `SentMsg1`, so they can no longer fire) and then promotes.
// `PromoteToActive` is still what performs the promotion.
debug_assert_eq!(
@@ -1023,7 +1037,7 @@ impl Node {
let our_profile = self.node_profile();
let (peer_identity, our_index, remote_epoch) = {
// Get the pending connection
let conn = match self.connections.get_mut(&link_id) {
let conn = match self.leg_mut(&link_id) {
Some(c) => c,
None => {
debug!(
@@ -1047,12 +1061,11 @@ impl Node {
error = %e,
"Msg3 processing failed"
);
// Clean up. Capture the index before removing the
// connection; reading it after the remove would always
// return None and leak the allocated index.
let our_idx_to_free =
self.connections.get(&link_id).and_then(|c| c.our_index());
self.connections.remove(&link_id);
// Clean up. Capture the index before disposing the
// machine (and the connection embedded on it); reading
// it after the disposal would always return None and
// leak the allocated index.
let our_idx_to_free = self.leg(&link_id).and_then(|c| c.our_index());
self.remove_link(&link_id);
self.remove_peer_machine(link_id);
if let Some(idx) = our_idx_to_free {
@@ -1070,7 +1083,6 @@ impl Node {
Ok(()) => {}
Err(e) => {
warn!(link_id = %link_id, our_profile = %our_profile, error = %e, "FMP negotiation failed");
self.connections.remove(&link_id);
self.remove_link(&link_id);
self.remove_peer_machine(link_id);
self.stats_mut()
@@ -1085,7 +1097,6 @@ impl Node {
Some(id) => *id,
None => {
warn!("Identity not learned from msg3");
self.connections.remove(&link_id);
self.remove_link(&link_id);
self.remove_peer_machine(link_id);
self.stats_mut()
@@ -1119,7 +1130,7 @@ impl Node {
// and the initiator has a matching session from processing
// msg2. Reason `Other` is used instead of `SecurityViolation`
// to avoid naming the ACL mechanism on the wire.
let reject_info = match self.connections.get_mut(&link_id) {
let reject_info = match self.leg_mut(&link_id) {
Some(conn) => match (conn.their_index(), conn.take_session()) {
(Some(idx), Some(session)) => Some((idx, session)),
_ => None,
@@ -1139,7 +1150,6 @@ impl Node {
)
.await;
}
self.connections.remove(&link_id);
self.remove_link(&link_id);
self.remove_peer_machine(link_id);
self.stats_mut()
@@ -1149,7 +1159,6 @@ impl Node {
if peer_node_addr == *self.identity().node_addr() {
debug!(link_id = %link_id, "Received msg3 from self, dropping");
self.connections.remove(&link_id);
self.remove_link(&link_id);
self.remove_peer_machine(link_id);
self.stats_mut()
@@ -1248,7 +1257,6 @@ impl Node {
// We keep our in-progress rekey and drop their msg3; return the
// msg1-allocated inbound index rather than orphaning it.
let _ = self.index_allocator.free(our_index);
self.connections.remove(&link_id);
self.links.remove(&link_id);
self.remove_peer_machine(link_id);
self.stats_mut()
@@ -1276,7 +1284,6 @@ impl Node {
// Duplicate handshake: the active peer is untouched, so return the
// msg1-allocated inbound index rather than orphaning it.
let _ = self.index_allocator.free(our_index);
self.connections.remove(&link_id);
self.links.remove(&link_id);
self.remove_peer_machine(link_id);
return;
@@ -1380,8 +1387,12 @@ impl Node {
link_id = %link_id,
"Leaf node rejecting additional peer (single-peer enforcement)"
);
// Clean up the connection and its control machine
if let Some(conn) = self.connections.remove(&link_id)
// Clean up the connection (taken off its machine first) and its
// control machine
if let Some(conn) = self
.peer_machines
.get_mut(&link_id)
.and_then(|machine| machine.take_leg())
&& let Some(idx) = conn.our_index()
{
let _ = self.index_allocator.free(idx);
@@ -1391,10 +1402,13 @@ impl Node {
return Err(NodeError::MaxPeersExceeded { max: 1 });
}
// Remove the connection from pending
// Take the pending connection off its control machine. The machine
// survives the promotion (it becomes the active peer's control
// machine), left with no pending connection.
let mut connection = self
.connections
.remove(&link_id)
.peer_machines
.get_mut(&link_id)
.and_then(|machine| machine.take_leg())
.ok_or(NodeError::ConnectionNotFound(link_id))?;
// Verify handshake is complete and extract session
@@ -1595,14 +1609,13 @@ impl Node {
// peer. The outbound will be cleaned up in handle_msg2 or by
// the 30s handshake timeout.
let pending_to_same_peer: Vec<LinkId> = self
.connections
.iter()
.filter(|(_, conn)| {
.connections()
.filter(|conn| {
conn.expected_identity()
.map(|id| *id.node_addr() == peer_node_addr)
.unwrap_or(false)
})
.map(|(lid, _)| *lid)
.map(|conn| conn.link_id())
.collect();
for pending_link_id in &pending_to_same_peer {

View File

@@ -14,11 +14,12 @@ impl LifecycleView for Node {
// `is_failed()` legs are always reaped here (~1s), as before. The
// idle-timeout is reaped here ONLY for legs whose timeout is not already
// driven by a machine `HandshakeTimeout` timer — i.e. inbound legs (IK
// inbound arms none) and machine-less test connections. Outbound legs with
// an armed timer are reaped by `drive_handshake_timeouts`, so excluding
// them here avoids a double reap.
self.connections
// inbound arms none). Outbound legs with an armed timer are reaped by
// `drive_handshake_timeouts`, so excluding them here avoids a double
// reap.
self.peer_machines
.iter()
.filter_map(|(link_id, machine)| machine.leg().map(|conn| (link_id, conn)))
.filter(|(link_id, conn)| {
conn.is_failed()
|| (conn.is_timed_out(now_ms, timeout_ms)
@@ -57,7 +58,7 @@ impl Node {
/// the retry-then-teardown choreography is the pure
/// [`Fmp::poll_timeouts`](crate::proto::fmp::Fmp::poll_timeouts) decision.
pub(in crate::node) fn check_timeouts(&mut self) {
if self.connections.is_empty() {
if self.connection_count() == 0 {
return;
}
@@ -70,7 +71,7 @@ impl Node {
ConnAction::ScheduleRetry { peer } => self.note_handshake_timeout(peer, now_ms),
ConnAction::Teardown { link } => {
// Log before cleanup (needs live connection state).
if let Some(conn) = self.connections.get(&link) {
if let Some(conn) = self.leg(&link) {
let direction = conn.direction();
if conn.is_failed() {
debug!(
@@ -101,14 +102,21 @@ impl Node {
/// the link and address mapping. Does not log — callers provide context-appropriate
/// log messages.
pub(in crate::node) fn cleanup_stale_connection(&mut self, link_id: LinkId, _now_ms: u64) {
let conn = match self.connections.remove(&link_id) {
// Take the connection off its machine BEFORE disposing the machine
// (the machine owns it), keeping it readable for the index/link
// cleanup below. The machine shares the connection's `link_id` and
// lifetime; dropping it here means a reaped handshake leg leaves no
// dangling machine. A no-op for promoted peers — `promote_connection`
// already consumed their connection, so this reaper never runs for
// them.
let conn = match self
.peer_machines
.get_mut(&link_id)
.and_then(|machine| machine.take_leg())
{
Some(c) => c,
None => return,
};
// A dial-persisted outbound control machine shares the connection's
// `link_id` and lifetime; drop it here so a reaped handshake leg leaves
// no dangling machine. A no-op for promoted peers — `promote_connection`
// already removed their connection, so this reaper never runs for them.
self.remove_peer_machine(link_id);
let transport_id = conn.transport_id();
@@ -159,7 +167,7 @@ impl Node {
/// reflex, then `cleanup_stale_connection` (which drops the machine + timers).
///
/// `check_timeouts` keeps reaping everything else — `is_failed()` legs and the
/// idle-timeout of legs without a machine timer (inbound / machine-less).
/// idle-timeout of legs without a machine timer (inbound legs).
fn drive_handshake_timeouts(&mut self, now_ms: u64) {
let timeout_ms = self.config().node.rate_limit.handshake_timeout_secs * 1000;
let timer_links: Vec<LinkId> = self
@@ -169,7 +177,7 @@ impl Node {
.map(|(link, _)| *link)
.collect();
for link in timer_links {
let (reap, retry_peer) = match self.connections.get(&link) {
let (reap, retry_peer) = match self.leg(&link) {
Some(conn) if conn.is_timed_out(now_ms, timeout_ms) => {
let retry_peer = if conn.is_outbound() {
conn.expected_identity().map(|id| *id.node_addr())
@@ -239,7 +247,7 @@ impl Node {
// Skip a resend whose target peer already promoted via the inbound
// cross-connection path — resending msg1 would start a new handshake
// (session mismatch).
if let Some(conn) = self.connections.get(&link)
if let Some(conn) = self.leg(&link)
&& conn
.expected_identity()
.map(|id| self.peers.contains_key(id.node_addr()))
@@ -264,7 +272,7 @@ impl Node {
continue;
}
};
match self.connections.get(&link).and_then(|c| c.handshake_msg1()) {
match self.leg(&link).and_then(|c| c.handshake_msg1()) {
// Armed but the stored wire isn't there yet — leave the timer and
// retry next tick (matches the old candidate filter skipping it).
None => continue,
@@ -296,7 +304,7 @@ impl Node {
continue;
};
let (transport_id, remote_addr) = match self.connections.get(&link) {
let (transport_id, remote_addr) = match self.leg(&link) {
Some(conn) => match (conn.transport_id(), conn.source_addr()) {
(Some(tid), Some(addr)) => (tid, addr.clone()),
_ => continue,

View File

@@ -372,7 +372,7 @@ impl Node {
}
fn is_connecting_to_peer(&self, peer_node_addr: &NodeAddr) -> bool {
self.connections.values().any(|conn| {
self.connections().any(|conn| {
conn.expected_identity()
.map(|id| id.node_addr() == peer_node_addr)
.unwrap_or(false)
@@ -385,7 +385,7 @@ impl Node {
transport_id: TransportId,
remote_addr: &TransportAddr,
) -> bool {
self.connections.values().any(|conn| {
self.connections().any(|conn| {
conn.expected_identity()
.map(|id| id.node_addr() == peer_node_addr)
.unwrap_or(false)
@@ -583,15 +583,16 @@ impl Node {
/// Prepare an outbound Noise msg1 at dial for an IDENTIFIED leg: allocate the
/// session index, run the Noise leaf, frame the wire, arm the shell-side
/// resend, track `pending_outbound`, and persist the connection. Returns
/// `Err` on index-allocation or Noise failure (cleaning the partial leg,
/// including the dial-time control machine), leaving the armed wire on the
/// connection for `send_stored_msg1` to transmit. Does NOT send — so the
/// fallible setup can propagate its error synchronously before any machine
/// drive. Anonymous-discovery legs use the monolithic `start_handshake`
/// instead, whose machine is born after its fallible setup — so this is
/// the only caller path whose Err arms clean up `peer_machines` (the
/// dial-time machine predates the failure).
/// resend, track `pending_outbound`, and embed the prepared connection on
/// the dial-time control machine. Returns `Err` on index-allocation or
/// Noise failure (cleaning the partial leg, including the dial-time control
/// machine), leaving the armed wire on the connection for
/// `send_stored_msg1` to transmit. Does NOT send — so the fallible setup
/// can propagate its error synchronously before any machine drive.
/// Anonymous-discovery legs use the monolithic `start_handshake` instead,
/// whose machine is born after its fallible setup — so this is the only
/// caller path whose Err arms clean up `peer_machines` (the dial-time
/// machine predates the failure).
pub(in crate::node) fn prepare_outbound_msg1(
&mut self,
link_id: LinkId,
@@ -658,7 +659,21 @@ impl Node {
// Track in pending_outbound for msg2 dispatch
self.pending_outbound
.insert((transport_id, our_index.as_u32()), link_id);
self.connections.insert(link_id, connection);
// The dial-born machine (persisted in `initiate_connection`) carries
// the prepared connection from here. Every live caller dialed first,
// so the machine exists; recover with a fresh one if a direct caller
// ever skips the dial.
debug_assert!(
self.peer_machines.contains_key(&link_id),
"outbound msg1 prepared for link {link_id} with no dial-time machine"
);
self.peer_machines
.entry(link_id)
.or_insert_with(|| {
PeerMachine::new_outbound(link_id, Some(peer_identity), current_time_ms)
})
.set_leg(connection);
Ok(())
}
@@ -745,23 +760,21 @@ impl Node {
// Track in pending_outbound for msg2 dispatch
self.pending_outbound
.insert((transport_id, our_index.as_u32()), link_id);
self.connections.insert(link_id, connection);
// The leg's persistent control machine is born alongside its pending
// The leg's persistent control machine is born carrying its pending
// connection, after all fallible setup (the index-alloc and Noise Err
// returns above predate it and need no disposal). Both production
// callers dial anonymously (identified dials persist their machine at
// dial and go through `prepare_outbound_msg1`), so the machine starts
// identity-less; `handle_msg2` crystallizes the identity the XX
// handshake learns. Inserted before the send below so no suspension
// point observes a leg without a machine. The send-failure arm retains
// the failed connection, and the machine stays with it — the stale-
// connection reaper disposes both together. No timers are armed and no
// event is dispatched: this path sends msg1 inline, so the machine
// parks at `Discovered` until msg2.
self.peer_machines.insert(
link_id,
PeerMachine::new_outbound(link_id, peer_identity, current_time_ms),
);
// point observes a leg in flight without a machine. The send-failure
// arm retains the failed connection, and the machine keeps carrying
// it — the stale-connection reaper disposes both together. No timers
// are armed and no event is dispatched: this path sends msg1 inline,
// so the machine parks at `Discovered` until msg2.
let mut machine = PeerMachine::new_outbound(link_id, peer_identity, current_time_ms);
machine.set_leg(connection);
self.peer_machines.insert(link_id, machine);
// Send the wire format handshake message
if let Some(transport) = self.transports.get(&transport_id) {
@@ -782,7 +795,7 @@ impl Node {
);
// Mark connection as failed but don't remove it yet
// The event loop can handle retry logic
if let Some(conn) = self.connections.get_mut(&link_id) {
if let Some(conn) = self.leg_mut(&link_id) {
conn.mark_failed();
}
}
@@ -804,15 +817,11 @@ impl Node {
transport_id: TransportId,
remote_addr: &TransportAddr,
) {
let wire_msg1 = match self
.connections
.get(&link_id)
.and_then(|c| c.handshake_msg1())
{
let wire_msg1 = match self.leg(&link_id).and_then(|c| c.handshake_msg1()) {
Some(w) => w.to_vec(),
None => return,
};
let our_index = self.connections.get(&link_id).and_then(|c| c.our_index());
let our_index = self.leg(&link_id).and_then(|c| c.our_index());
// Send the wire format handshake message
if let Some(transport) = self.transports.get(&transport_id) {
@@ -835,7 +844,7 @@ impl Node {
);
// Mark connection as failed but don't remove it yet
// The event loop can handle retry logic
if let Some(conn) = self.connections.get_mut(&link_id) {
if let Some(conn) = self.leg_mut(&link_id) {
conn.mark_failed();
}
}
@@ -1066,14 +1075,13 @@ impl Node {
);
let now_ms = Self::now_ms();
let stale: Vec<LinkId> = self
.connections
.iter()
.filter(|(_, conn)| {
.connections()
.filter(|conn| {
conn.expected_identity()
.map(|id| id.node_addr() == &peer_addr)
.unwrap_or(false)
})
.map(|(link_id, _)| *link_id)
.map(|conn| conn.link_id())
.collect();
for link_id in stale {
self.cleanup_stale_connection(link_id, now_ms);
@@ -2043,7 +2051,7 @@ impl Node {
info!("Node started:");
info!(" state: {}", self.supervisor.state);
info!(" transports: {}", self.transports.len());
info!(" connections: {}", self.connections.len());
info!(" connections: {}", self.connection_count());
Ok(())
}
@@ -2887,18 +2895,17 @@ impl Node {
/// The `connected` / `connecting` sets gate the floor, retry-dial, overlay,
/// and LAN layers; `in_flight_by_peer` feeds the opportunistic layer's
/// per-peer parallel cap, computed exactly as the deleted
/// `path_candidate_attempt_budget` did: `connections(expected == addr) +
/// `path_candidate_attempt_budget` did: `pending legs(expected == addr) +
/// pending_connects(addr)`. The scalar counts stay unpopulated at the
/// ceiling-only posture (no layer reads them; see [`Observed`]).
pub(in crate::node) fn observe_peering(&self) -> Observed {
let connected: HashSet<NodeAddr> = self.peers.keys().copied().collect();
let connecting: HashSet<NodeAddr> = self
.connections
.values()
.connections()
.filter_map(|conn| conn.expected_identity().map(|id| *id.node_addr()))
.collect();
let mut in_flight_by_peer: HashMap<NodeAddr, usize> = HashMap::new();
for conn in self.connections.values() {
for conn in self.connections() {
if let Some(id) = conn.expected_identity() {
*in_flight_by_peer.entry(*id.node_addr()).or_default() += 1;
}
@@ -2940,9 +2947,11 @@ impl Node {
}
fn outbound_handshake_slots(&self) -> usize {
// Count pending connections (legs), not machines: a dial-born machine
// in the connect window has no connection yet but DOES hold a
// `pending_connects` slot, so counting machines would tally it twice.
let used = self
.connections
.len()
.connection_count()
.saturating_add(self.peering.pending_connects.len());
if self.max_connections() == 0 {
usize::MAX
@@ -2968,8 +2977,7 @@ impl Node {
}
let in_flight_for_peer = self
.connections
.values()
.connections()
.filter(|conn| {
conn.expected_identity()
.map(|identity| identity.node_addr() == peer_node_addr)

View File

@@ -297,7 +297,8 @@ struct PendingConnect {
/// ## Peer Lifecycle
///
/// Peers go through two phases:
/// 1. **Connection phase** (`connections`): Handshake in progress, indexed by LinkId
/// 1. **Connection phase**: Handshake in progress; the pending connection is
/// carried by its per-peer control machine (`peer_machines`), indexed by LinkId
/// 2. **Active phase** (`peers`): Authenticated, indexed by NodeAddr
///
/// The `addr_to_link` map enables dispatching incoming packets to the right
@@ -364,23 +365,17 @@ pub struct Node {
/// rx_loop select arm that feeds `Event::ChildExited` to the supervisor FSM.
child_exit_rx: Option<tokio::sync::mpsc::Receiver<crate::node::lifecycle::supervisor::Child>>,
// === Connections (Handshake Phase) ===
/// Pending connections (handshake in progress).
/// Indexed by LinkId since we don't know the peer's identity yet.
connections: HashMap<LinkId, PeerConnection>,
// === Per-Peer Control Machines ===
/// Per-peer lifecycle control FSMs, keyed by the stable `LinkId` that spans
/// the handshake→active lifetime. A parallel structure introduced by the
/// node-runtime decomposition: `connections`/`peers` stay byte-unchanged (hot
/// path pristine) and are cut over to this machine home path-by-path.
/// Machines are inserted at the identified dial and born `established()`
/// inside `promote_connection` (msg3 decisions run through a throwaway
/// machine), and stepped in production by the handshake handlers, the
/// rekey-cadence and liveness-reap routers, and the lifecycle paths, with
/// the executor (`dataplane/peer_actions.rs`) performing the returned
/// actions. Timer FIRING decisions remain shell-side: `PeerEvent::Timeout`
/// is never dispatched in production.
/// the handshake→active lifetime. Each machine owns its pending handshake
/// connection (the `PeerConnection` leg) while the handshake is in
/// progress — the single LinkId-keyed per-peer map on `Node`; `peers`
/// stays byte-unchanged (hot path pristine).
/// Machines are inserted at dial and inbound msg1, and stepped in production
/// by the handshake handlers, the rekey-cadence and liveness-reap routers,
/// and the lifecycle paths, with the executor (`dataplane/peer_actions.rs`)
/// performing the returned actions. Timer FIRING decisions remain
/// shell-side: `PeerEvent::Timeout` is never dispatched in production.
peer_machines: HashMap<LinkId, PeerMachine>,
/// Per-peer timer store, keyed by `LinkId` then `TimerKind`, holding each
@@ -667,7 +662,6 @@ impl Node {
packet_rx: None,
child_exit_tx: None,
child_exit_rx: None,
connections: HashMap::new(),
peer_machines: HashMap::new(),
peer_timers: HashMap::new(),
peers: HashMap::new(),
@@ -818,7 +812,6 @@ impl Node {
packet_rx: None,
child_exit_tx: None,
child_exit_rx: None,
connections: HashMap::new(),
peer_machines: HashMap::new(),
peer_timers: HashMap::new(),
peers: HashMap::new(),
@@ -1562,7 +1555,7 @@ impl Node {
tun_state: self.tun_state,
tun_name: self.tun_name.clone(),
effective_ipv6_mtu: self.effective_ipv6_mtu(),
connection_count: self.connections.len(),
connection_count: self.connection_count(),
peer_count: self.peers.len(),
link_count: self.links.len(),
transport_count: self.transports.len(),
@@ -2193,7 +2186,10 @@ impl Node {
/// Number of pending connections (handshake in progress).
pub fn connection_count(&self) -> usize {
self.connections.len()
self.peer_machines
.values()
.filter(|machine| machine.leg().is_some())
.count()
}
/// Number of authenticated peers.
@@ -2314,31 +2310,19 @@ impl Node {
self.peer_timers.remove(&link);
}
/// Gates the leg→machine direction of `debug_assert_peer_maps_coherent`.
/// Asserting it requires the convention that every `connections` insert is
/// paired with a `peer_machines` insert before the next await point —
/// which now holds everywhere: identified dials persist their machine
/// before the leg exists (`initiate_connection`), anonymous-discovery legs
/// are born with one inside `start_handshake` alongside the connection
/// insert, inbound legs get theirs in `handle_msg1` at the same point, and
/// the `add_connection` test seam seeds one for directly-inserted legs.
#[cfg(debug_assertions)]
const CHECK_LEGS_HAVE_MACHINES: bool = true;
/// Debug-build coherence sweep over the peer-lifecycle maps, run once per
/// rx-loop tick and invoked directly by unit tests.
///
/// Machine→carrier: every `peer_machines` entry must have a live carrier —
/// a pending connection on the same link, an active peer on it, or a
/// its own embedded pending connection, an active peer on its link, or a
/// pending connect still resolving toward it. A machine with none of these
/// is unreachable by every teardown path and has leaked.
///
/// Leg→machine: every pending connection carries a control machine (gated
/// by `CHECK_LEGS_HAVE_MACHINES` above).
/// is unreachable by every teardown path and has leaked. (The pending
/// connection lives inside the machine, so no separate connection→machine
/// direction exists to check.)
#[cfg(debug_assertions)]
pub(in crate::node) fn debug_assert_peer_maps_coherent(&self) {
for link in self.peer_machines.keys() {
let has_carrier = self.connections.contains_key(link)
for (link, machine) in &self.peer_machines {
let has_carrier = machine.leg().is_some()
|| self.peers.values().any(|peer| peer.link_id() == *link)
|| self
.peering
@@ -2351,22 +2335,13 @@ impl Node {
(no pending connection, active peer, or pending connect)"
);
}
if Self::CHECK_LEGS_HAVE_MACHINES {
for link in self.connections.keys() {
assert!(
self.peer_machines.contains_key(link),
"pending connection on link {link} has no control machine"
);
}
}
}
/// Operator-visible msg1 resend count for a pending handshake `link`, read
/// from the per-peer machine (the counter's home once the resend drive moved
/// off the shell connection). Machine-less connections (inbound legs that
/// never resend, and test-created connections) report 0, matching what the
/// shell connection reported before the counter moved.
/// off the shell connection). A link with no machine reports 0, and inbound
/// machines never resend, matching what the shell connection reported
/// before the counter moved.
pub(crate) fn connection_resend_count(&self, link: LinkId) -> u32 {
self.peer_machines
.get(&link)
@@ -2387,8 +2362,7 @@ impl Node {
.values()
.any(|link| link.transport_id() == transport_id)
|| self
.connections
.values()
.connections()
.any(|conn| conn.transport_id() == Some(transport_id))
|| self
.peers
@@ -2423,58 +2397,86 @@ impl Node {
// === Connection Management (Handshake Phase) ===
/// The pending connection for `link_id`, read through the control machine
/// that carries it.
fn leg(&self, link_id: &LinkId) -> Option<&PeerConnection> {
self.peer_machines
.get(link_id)
.and_then(|machine| machine.leg())
}
/// Mutable access to the pending connection for `link_id`.
fn leg_mut(&mut self, link_id: &LinkId) -> Option<&mut PeerConnection> {
self.peer_machines
.get_mut(link_id)
.and_then(|machine| machine.leg_mut())
}
/// Add a pending connection.
///
/// Also seeds a control machine for the leg when none exists yet, keeping
/// the leg→machine invariant (`debug_assert_peer_maps_coherent`) intact
/// for callers that insert a connection directly rather than through the
/// dial or inbound-msg1 paths, which pair the two inserts themselves.
/// Seeds a control machine for the leg when none exists yet and embeds the
/// connection on it, for callers that insert a connection directly rather
/// than through the dial or inbound-msg1 paths, which build the machine
/// themselves.
pub fn add_connection(&mut self, connection: PeerConnection) -> Result<(), NodeError> {
let link_id = connection.link_id();
if self.connections.contains_key(&link_id) {
if self
.peer_machines
.get(&link_id)
.is_some_and(|machine| machine.leg().is_some())
{
return Err(NodeError::ConnectionAlreadyExists(link_id));
}
if self.max_connections() > 0 && self.connections.len() >= self.max_connections() {
if self.max_connections() > 0 && self.connection_count() >= self.max_connections() {
return Err(NodeError::MaxConnectionsExceeded {
max: self.max_connections(),
});
}
self.peer_machines.entry(link_id).or_insert_with(|| {
let now = connection.started_at();
if connection.is_outbound() {
PeerMachine::new_outbound(link_id, connection.expected_identity().copied(), now)
} else {
PeerMachine::new_inbound(link_id, now)
}
});
self.connections.insert(link_id, connection);
self.peer_machines
.entry(link_id)
.or_insert_with(|| {
let now = connection.started_at();
if connection.is_outbound() {
PeerMachine::new_outbound(link_id, connection.expected_identity().copied(), now)
} else {
PeerMachine::new_inbound(link_id, now)
}
})
.set_leg(connection);
Ok(())
}
/// Get a connection by LinkId.
pub fn get_connection(&self, link_id: &LinkId) -> Option<&PeerConnection> {
self.connections.get(link_id)
self.leg(link_id)
}
/// Get a mutable connection by LinkId.
pub fn get_connection_mut(&mut self, link_id: &LinkId) -> Option<&mut PeerConnection> {
self.connections.get_mut(link_id)
self.leg_mut(link_id)
}
/// Remove a connection, disposing its control machine alongside
/// (the disposal complement of `add_connection`'s machine seeding).
/// (the disposal complement of `add_connection`'s machine seeding). The
/// connection is taken off the machine BEFORE the machine is dropped, so
/// the caller still receives it.
pub fn remove_connection(&mut self, link_id: &LinkId) -> Option<PeerConnection> {
let connection = self
.peer_machines
.get_mut(link_id)
.and_then(|machine| machine.take_leg());
self.remove_peer_machine(*link_id);
self.connections.remove(link_id)
connection
}
/// Iterate over all connections.
pub fn connections(&self) -> impl Iterator<Item = &PeerConnection> {
self.connections.values()
self.peer_machines
.values()
.filter_map(|machine| machine.leg())
}
// === Peer Management (Active Phase) ===
@@ -3044,7 +3046,8 @@ impl Node {
/// Unlike `send_encrypted_link_message`, this does not look the peer up
/// in `self.peers`. It takes the session and wire parameters directly,
/// so it can be used during handshake teardown (before promotion) where
/// the peer state lives in `self.connections` rather than `self.peers`.
/// the peer state lives on the pending connection embedded in its control
/// machine rather than in `self.peers`.
///
/// The inner-header timestamp is set to 0 — the session has just been
/// established and no session-elapsed reference is available yet; this

View File

@@ -147,7 +147,7 @@ pub(crate) struct Observed {
#[allow(dead_code)]
// ceiling-only posture: no layer reads the scalar counts (future set-point)
pub peers: usize,
/// `self.connections.len()`.
/// `self.connection_count()`.
#[allow(dead_code)]
// ceiling-only posture: no layer reads the scalar counts (future set-point)
pub connections: usize,

View File

@@ -76,16 +76,10 @@ async fn test_outbound_msg2_denied_after_acl_reload() {
node_a
.addr_to_link
.insert((transport_id, remote_addr.clone()), link_id_a);
node_a.connections.insert(link_id_a, conn_a);
node_a.add_connection(conn_a).unwrap();
node_a
.pending_outbound
.insert((transport_id, our_index_a.as_u32()), link_id_a);
// Mirror the production dial path: an outbound leg persists its control
// machine at dial.
node_a.peer_machines.insert(
link_id_a,
crate::peer::machine::PeerMachine::new_outbound(link_id_a, Some(peer_b_identity), 1000),
);
let mut conn_b = PeerConnection::inbound(LinkId::new(2), 1000);
let responder_epoch = [0x11; 8];
@@ -202,16 +196,12 @@ async fn test_inbound_msg3_denied_triggers_disconnect() {
node_a
.addr_to_link
.insert((transport_id_a, addr_b.clone()), link_id_a);
node_a.connections.insert(link_id_a, conn_a);
// Mirror the production dial path: the seam seeds the outbound leg's
// control machine and embeds the connection on it.
node_a.add_connection(conn_a).unwrap();
node_a
.pending_outbound
.insert((transport_id_a, our_index_a.as_u32()), link_id_a);
// Mirror the production dial path: an outbound leg persists its control
// machine at dial.
node_a.peer_machines.insert(
link_id_a,
crate::peer::machine::PeerMachine::new_outbound(link_id_a, Some(peer_b_identity), 1000),
);
let transport = node_a.transports.get(&transport_id_a).unwrap();
transport

View File

@@ -81,16 +81,10 @@ async fn test_two_node_handshake_udp() {
Duration::from_millis(100),
);
node_a.links.insert(link_id_a, link_a);
node_a.connections.insert(link_id_a, conn_a);
node_a.add_connection(conn_a).unwrap();
node_a
.pending_outbound
.insert((transport_id_a, our_index_a.as_u32()), link_id_a);
// Mirror the production dial path: an outbound leg persists its control
// machine at dial.
node_a.peer_machines.insert(
link_id_a,
crate::peer::machine::PeerMachine::new_outbound(link_id_a, Some(peer_b_identity), 1000),
);
// Send msg1 from A to B over UDP
let transport = node_a.transports.get(&transport_id_a).unwrap();
@@ -118,7 +112,7 @@ async fn test_two_node_handshake_udp() {
"Node B should have 0 peers after msg1 (XX awaits msg3)"
);
assert_eq!(
node_b.connections.len(),
node_b.connection_count(),
1,
"Node B should have 1 pending connection awaiting msg3"
);
@@ -346,16 +340,10 @@ async fn test_run_rx_loop_handshake() {
Duration::from_millis(100),
);
node_a.links.insert(link_id_a, link_a);
node_a.connections.insert(link_id_a, conn_a);
node_a.add_connection(conn_a).unwrap();
node_a
.pending_outbound
.insert((transport_id_a, our_index_a.as_u32()), link_id_a);
// Mirror the production dial path: an outbound leg persists its control
// machine at dial.
node_a.peer_machines.insert(
link_id_a,
crate::peer::machine::PeerMachine::new_outbound(link_id_a, Some(peer_b_identity), 1000),
);
// Send msg1 from A to B over real UDP
let transport = node_a.transports.get(&transport_id_a).unwrap();
@@ -394,7 +382,7 @@ async fn test_run_rx_loop_handshake() {
"Node B should have 0 peers after rx loop processed msg1 (XX awaits msg3)"
);
assert_eq!(
node_b.connections.len(),
node_b.connection_count(),
1,
"Node B should have 1 pending connection"
);
@@ -445,7 +433,7 @@ async fn test_run_rx_loop_handshake() {
// This test verifies rx_loop correctly dispatches PHASE_MSG1 (Phase 2)
// and PHASE_MSG2 (Phase 3). B still has a pending connection awaiting msg3.
assert_eq!(
node_b.connections.len(),
node_b.connection_count(),
1,
"Node B should still have pending connection awaiting msg3"
);
@@ -540,16 +528,10 @@ async fn test_cross_connection_both_initiate() {
node_a
.addr_to_link
.insert((transport_id_a, remote_addr_b.clone()), link_id_a_out);
node_a.connections.insert(link_id_a_out, conn_a);
node_a.add_connection(conn_a).unwrap();
node_a
.pending_outbound
.insert((transport_id_a, our_index_a.as_u32()), link_id_a_out);
// Mirror the production dial path: an outbound leg persists its control
// machine at dial.
node_a.peer_machines.insert(
link_id_a_out,
crate::peer::machine::PeerMachine::new_outbound(link_id_a_out, Some(peer_b_identity), 1000),
);
// Node B initiates to Node A
let link_id_b_out = node_b.allocate_link_id();
@@ -576,16 +558,10 @@ async fn test_cross_connection_both_initiate() {
node_b
.addr_to_link
.insert((transport_id_b, remote_addr_a.clone()), link_id_b_out);
node_b.connections.insert(link_id_b_out, conn_b);
node_b.add_connection(conn_b).unwrap();
node_b
.pending_outbound
.insert((transport_id_b, our_index_b.as_u32()), link_id_b_out);
// Mirror the production dial path: an outbound leg persists its control
// machine at dial.
node_b.peer_machines.insert(
link_id_b_out,
crate::peer::machine::PeerMachine::new_outbound(link_id_b_out, Some(peer_a_identity), 1000),
);
// Both send msg1 over UDP
let transport = node_a.transports.get(&transport_id_a).unwrap();
@@ -750,7 +726,7 @@ async fn test_stale_connection_cleanup() {
node.links.insert(link_id, link);
node.addr_to_link
.insert((transport_id, remote_addr.clone()), link_id);
node.connections.insert(link_id, conn);
node.add_connection(conn).unwrap();
node.pending_outbound
.insert((transport_id, our_index.as_u32()), link_id);
@@ -828,7 +804,7 @@ async fn test_failed_connection_cleanup() {
node.links.insert(link_id, link);
node.addr_to_link
.insert((transport_id, remote_addr.clone()), link_id);
node.connections.insert(link_id, conn);
node.add_connection(conn).unwrap();
node.pending_outbound
.insert((transport_id, our_index.as_u32()), link_id);
@@ -924,11 +900,11 @@ async fn test_resend_scheduling() {
.insert((transport_id, remote_addr.clone()), link_id);
node.pending_outbound
.insert((transport_id, our_index.as_u32()), link_id);
node.connections.insert(link_id, conn);
// The msg1-resend counter and its due timer live on the per-peer machine.
// Dial it to `SentMsg1` (connectionless: no connect step) and arm its
// retransmit timer at now + 1000ms, mirroring what a real dial arms.
// The msg1-resend counter and its due timer live on the per-peer machine,
// which also carries the pending connection. Dial it to `SentMsg1`
// (connectionless: no connect step) and arm its retransmit timer at
// now + 1000ms, mirroring what a real dial arms.
let mut machine =
crate::peer::machine::PeerMachine::new_outbound(link_id, Some(peer_identity), now_ms);
let _ = machine.step(
@@ -941,6 +917,7 @@ async fn test_resend_scheduling() {
now_ms,
&mut node.index_allocator,
);
machine.set_leg(conn);
node.peer_machines.insert(link_id, machine);
node.peer_timers.entry(link_id).or_default().insert(
crate::peer::machine::TimerKind::HandshakeRetransmit,
@@ -999,11 +976,11 @@ async fn test_handshake_timeout_drive() {
node.links.insert(link_id, link);
node.addr_to_link
.insert((transport_id, remote_addr.clone()), link_id);
node.connections.insert(link_id, conn);
node.pending_outbound
.insert((transport_id, our_index.as_u32()), link_id);
// Machine in SentMsg1 with a HandshakeTimeout timer armed at dial + 30s.
// Machine in SentMsg1, carrying the pending connection, with a
// HandshakeTimeout timer armed at dial + 30s.
let mut machine =
crate::peer::machine::PeerMachine::new_outbound(link_id, Some(peer_identity), dial_ms);
let _ = machine.step(
@@ -1016,6 +993,7 @@ async fn test_handshake_timeout_drive() {
dial_ms,
&mut node.index_allocator,
);
machine.set_leg(conn);
node.peer_machines.insert(link_id, machine);
node.peer_timers.entry(link_id).or_default().insert(
crate::peer::machine::TimerKind::HandshakeTimeout,
@@ -1505,14 +1483,10 @@ async fn drive_to_msg3(
.node
.addr_to_link
.insert((initiator.transport_id, responder.addr.clone()), link_id);
initiator.node.connections.insert(link_id, conn);
// Mirror the production dial path: an identified outbound leg persists
// its control machine at dial, and the promote feedback later
// crystallizes that same machine in place.
initiator.node.peer_machines.insert(
link_id,
crate::peer::machine::PeerMachine::new_outbound(link_id, Some(peer_identity), now_ms),
);
// Mirror the production dial path: the seam seeds the identified outbound
// leg's control machine (carrying the connection) at dial, and the promote
// feedback later crystallizes that same machine in place.
initiator.node.add_connection(conn).unwrap();
initiator
.node
.pending_outbound
@@ -1698,8 +1672,8 @@ async fn test_inbound_machine_born_at_msg1_and_crystallized_at_promote() {
// After msg1 the responder's window leg carries a machine parked at
// `SentMsg2`, seeded with the leg's msg1-allocated index.
assert_eq!(responder.node.connections.len(), 1);
let leg_link = *responder.node.connections.keys().next().unwrap();
assert_eq!(responder.node.connection_count(), 1);
let leg_link = responder.node.connections().next().unwrap().link_id();
let leg_index = responder
.node
.get_connection(&leg_link)
@@ -1777,7 +1751,10 @@ async fn test_msg3_crypto_fail_disposes_leg_machine() {
responder.node.handle_msg3(msg3).await;
assert_eq!(responder.node.peer_count(), 0, "no promotion");
assert!(responder.node.connections.is_empty(), "leg torn down");
assert!(
responder.node.connections().next().is_none(),
"leg torn down"
);
assert!(
responder.node.peer_machines.is_empty(),
"crypto-fail teardown disposes the leg's machine"
@@ -1864,8 +1841,8 @@ async fn test_anonymous_dial_births_identityless_machine_at_leg_birth() {
.await
.expect("anonymous dial");
assert_eq!(initiator.node.connections.len(), 1);
let leg_link = *initiator.node.connections.keys().next().unwrap();
assert_eq!(initiator.node.connection_count(), 1);
let leg_link = initiator.node.connections().next().unwrap().link_id();
let machine = initiator
.node
.peer_machines
@@ -1900,7 +1877,7 @@ async fn test_anonymous_msg2_crystallizes_identity_and_promotes() {
.initiate_connection(initiator.transport_id, responder.addr.clone(), None)
.await
.expect("anonymous dial");
let leg_link = *initiator.node.connections.keys().next().unwrap();
let leg_link = initiator.node.connections().next().unwrap().link_id();
initiator.node.debug_assert_peer_maps_coherent();
// Responder answers msg1 with msg2; the initiator's msg2 processing learns
@@ -1960,7 +1937,7 @@ async fn test_anonymous_self_connect_drop_disposes_machine() {
.initiate_connection(node.transport_id, self_addr, None)
.await
.expect("anonymous self dial");
let leg_link = *node.node.connections.keys().next().unwrap();
let leg_link = node.node.connections().next().unwrap().link_id();
assert_eq!(node.node.peer_machines.len(), 1);
// We answer our own msg1, then our msg2 processing discovers the learned
@@ -1972,7 +1949,7 @@ async fn test_anonymous_self_connect_drop_disposes_machine() {
assert_eq!(node.node.peer_count(), 0, "no promotion");
assert!(
!node.node.connections.contains_key(&leg_link),
node.node.get_connection(&leg_link).is_none(),
"self-connect drop removes the outbound leg"
);
assert!(

View File

@@ -162,14 +162,7 @@ pub(super) async fn initiate_handshake(nodes: &mut [TestNode], i: usize, j: usiz
.node
.addr_to_link
.insert((transport_id, responder_addr.clone()), link_id);
initiator.node.connections.insert(link_id, conn);
// Mirror the production dial path: an identified outbound leg persists
// its control machine at dial, and the promote feedback later
// crystallizes that same machine in place.
initiator.node.peer_machines.insert(
link_id,
crate::peer::machine::PeerMachine::new_outbound(link_id, Some(peer_identity), 1000),
);
initiator.node.add_connection(conn).unwrap();
initiator
.node
.pending_outbound

View File

@@ -137,8 +137,7 @@ async fn test_try_peer_addresses_races_all_concrete_udp_candidates() {
.unwrap();
let mut addrs = node
.connections
.values()
.connections()
.filter_map(|conn| conn.source_addr().and_then(|addr| addr.as_str()))
.collect::<Vec<_>>();
addrs.sort();
@@ -733,7 +732,7 @@ fn test_promote_cleans_up_pending_outbound_to_same_peer() {
node.links.insert(pending_link_id, pending_link);
node.addr_to_link
.insert((transport_id, pending_addr.clone()), pending_link_id);
node.connections.insert(pending_link_id, pending_conn);
node.add_connection(pending_conn).unwrap();
node.pending_outbound
.insert((transport_id, pending_index.as_u32()), pending_link_id);
@@ -1255,8 +1254,7 @@ async fn update_peers_races_new_alternative_without_dropping_active_peer() {
assert_eq!(node.peer_count(), 1, "existing link must stay live");
assert_eq!(node.connection_count(), 1);
assert_eq!(
node.connections
.values()
node.connections()
.next()
.and_then(|conn| conn.source_addr()),
Some(&new_addr)
@@ -2027,16 +2025,12 @@ async fn drive_xx_handshake(
Duration::from_millis(100),
);
node_a.links.insert(link_id_a, link_a);
node_a.connections.insert(link_id_a, conn_a);
// Mirror the production dial path: the seam seeds the outbound leg's
// control machine and embeds the connection on it.
node_a.add_connection(conn_a).unwrap();
node_a
.pending_outbound
.insert((transport_id, our_index_a.as_u32()), link_id_a);
// Mirror the production dial path: an outbound leg persists its control
// machine at dial.
node_a.peer_machines.insert(
link_id_a,
crate::peer::machine::PeerMachine::new_outbound(link_id_a, Some(peer_b_identity), 1000),
);
let transport = node_a.transports.get(&transport_id).unwrap();
transport

View File

@@ -78,6 +78,7 @@
#![allow(dead_code)]
use crate::peer::PeerConnection;
use crate::proto::fmp::{
ConnAction, ConnSnapshot, ConnectionState, EstablishSnapshot, Fmp, InboundDecision,
InboundReject, PeerSnapshot, PromotionResult, RekeyCfg, RekeyResendSnapshot, WireOutcome,
@@ -428,6 +429,12 @@ pub(crate) struct PeerMachine {
/// The crystallized peer node address. Known up front for outbound (from the
/// dial identity); learned at msg3 for inbound (from [`WireOutcome`]).
node_addr: Option<NodeAddr>,
/// The pending handshake connection this machine owns while the leg is in
/// the handshake window. `None` before the connection is built (the dial
/// window) and after promotion consumes it (the machine survives as the
/// active peer's control machine). Pure storage — the machine never reads
/// or drives it; the shell reaches it through the accessors below.
leg: Option<PeerConnection>,
/// Pure handshake-phase bookkeeping (link/direction/indices/transport/
/// stored handshake bytes/epoch). Reused verbatim from the FMP state core.
conn: ConnectionState,
@@ -475,6 +482,7 @@ impl PeerMachine {
link,
identity,
node_addr: identity.map(|id| *id.node_addr()),
leg: None,
conn: match identity {
Some(id) => ConnectionState::outbound(link, id, now),
None => ConnectionState::outbound_anonymous(link, now),
@@ -505,6 +513,7 @@ impl PeerMachine {
link,
identity: None,
node_addr: None,
leg: None,
conn: ConnectionState::inbound(link, now),
remote_epoch: None,
rekey_in_progress: false,
@@ -572,6 +581,7 @@ impl PeerMachine {
link,
identity: Some(identity),
node_addr: Some(addr),
leg: None,
conn,
remote_epoch,
rekey_in_progress: false,
@@ -594,6 +604,28 @@ impl PeerMachine {
self.state
}
/// The pending handshake connection, if this leg is still in the
/// handshake window.
pub(crate) fn leg(&self) -> Option<&PeerConnection> {
self.leg.as_ref()
}
/// Mutable access to the pending handshake connection.
pub(crate) fn leg_mut(&mut self) -> Option<&mut PeerConnection> {
self.leg.as_mut()
}
/// Take the pending handshake connection off the machine (promotion and
/// teardown consume it by value).
pub(crate) fn take_leg(&mut self) -> Option<PeerConnection> {
self.leg.take()
}
/// Embed a pending handshake connection on the machine.
pub(crate) fn set_leg(&mut self, leg: PeerConnection) {
self.leg = Some(leg);
}
/// The index we allocated for this peer's inbound session at msg1. `None`
/// before allocation. The inbound cross-connection / rekey-respond triggers
/// and the executor read this to perform the shell registry surgery with the