Export limits, and a role for every peer where routes are passed on

What other networks were told of rested on one thing: the Only to
Customer mark that routes from providers, peers and route servers get as
they come in. A route without it, from a session without a role or by a
mistake anywhere along the way, went to every peer, from every router of
ours, and nothing of ours stood in the way.

export_limits are a second check, independent of the roles, the marks
and the export policy. For peers above or beside us (PROVIDER, PEER,
ROUTE_SERVER, and any in another AS without a role):

- prefixes: only routes for these go out, learned or our own. With our
  and our customers' space listed, a leak of anyone else's routes goes
  nowhere above us.
- max_prefixes: if more routes were to go out than this, none of them
  does. The session ends with a Cease / Administrative Shutdown whose
  Shutdown Communication says why (RFC 9003), and the peer is held down
  for its import.hold_down_seconds, so that it is told nothing rather
  than too much. The whole table is counted before any of it is sent.

They are set at the top and hold for every such peer that has none of its
own; unlike `export`, a peer's own export policy does not drop them. A
peer's own export_limits take their place, and an empty one lifts them.
Customers, route server clients and our own routers are not limited.

A configuration in which learned routes are passed on and a peer in
another AS has no role is now refused, as is accept_unconfigured_peers
with learned routes: without a role, a peer's routes go to everyone and
it is told of everything. This used to be a warning.

At startup, a warning for each peer the limits apply to that has none,
and for each originated route the limits leave out. A metric,
bgp_export_limit_exceeded_total, counts the sessions ended for it.

The interop lab has a phase in which a customer leaks: a route outside
its space never reaches BIRD, and when more were to go than the limit,
BIRD's session ends and its log has our reason.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

crates/integration-tests/src/test_server.rs

@@ -72,6 +72,7 @@ impl TestServer {
             rpki: None,
             kernel: None,
             export: Default::default(),
+            export_limits: None,
             control: None,
             graceful_restart: Default::default(),
         };

crates/integration-tests/tests/add_path.rs

@@ -42,6 +42,7 @@ async fn server_offering(
         c.peers = vec![PeerConfig {
             graceful_restart: Default::default(),
             groups: Vec::new(),
+            export_limits: None,
             reflector_client: false,
             name: "route-server-client".into(),
             endpoint: SocketAddr::new(Ipv4Addr::LOCALHOST.into(), 1).into(),

crates/integration-tests/tests/admission.rs

@@ -32,6 +32,7 @@ fn peer_at(endpoint: SocketAddr) -> PeerConfig {
         role: None,
         require_role: false,
         export: Default::default(),
+        export_limits: None,
     }
 }
 

crates/integration-tests/tests/aspa.rs

@@ -59,6 +59,7 @@ async fn server_with(
         c.peers = vec![PeerConfig {
             graceful_restart: Default::default(),
             groups: Vec::new(),
+            export_limits: None,
             reflector_client: false,
             name: "r1".into(),
             endpoint: endpoint.into(),

crates/integration-tests/tests/collision.rs

@@ -38,6 +38,7 @@ async fn server_dialing(listener: &TcpListener) -> Result<TestServer> {
         c.peers = vec![PeerConfig {
             graceful_restart: Default::default(),
             groups: Vec::new(),
+            export_limits: None,
             reflector_client: false,
             name: "r1".into(),
             endpoint: endpoint.into(),

crates/integration-tests/tests/debugging.rs

@@ -169,6 +169,7 @@ async fn test_refused_routes_can_be_seen_and_counted() -> Result<()> {
         c.peers = vec![PeerConfig {
             graceful_restart: Default::default(),
             groups: Vec::new(),
+            export_limits: None,
             reflector_client: false,
             name: "refused-demo".into(),
             endpoint: endpoint.into(),

crates/integration-tests/tests/dialer.rs

@@ -29,6 +29,7 @@ async fn server_dialing(listener: &TcpListener, remote_asn: u32) -> Result<TestS
         c.peers = vec![PeerConfig {
             graceful_restart: Default::default(),
             groups: Vec::new(),
+            export_limits: None,
             reflector_client: false,
             name: "r1".into(),
             endpoint: endpoint.into(),
@@ -190,6 +191,7 @@ async fn test_inbound_from_configured_address_uses_config() -> Result<()> {
         c.peers = vec![PeerConfig {
             graceful_restart: Default::default(),
             groups: Vec::new(),
+            export_limits: None,
             reflector_client: false,
             name: "r1".into(),
             endpoint: SocketAddr::new(Ipv4Addr::LOCALHOST.into(), 1).into(),
@@ -245,6 +247,7 @@ async fn test_local_address_is_used_as_the_source() -> Result<()> {
         c.peers = vec![PeerConfig {
             graceful_restart: Default::default(),
             groups: Vec::new(),
+            export_limits: None,
             reflector_client: false,
             name: "r1".into(),
             endpoint: endpoint.into(),
@@ -269,6 +272,7 @@ async fn test_local_address_is_used_as_the_source() -> Result<()> {
         c.peers = vec![PeerConfig {
             graceful_restart: Default::default(),
             groups: Vec::new(),
+            export_limits: None,
             reflector_client: false,
             name: "r1".into(),
             endpoint: endpoint.into(),

crates/integration-tests/tests/export_limits.rs

@@ -0,0 +1,187 @@
+//! Export limits, end to end: what a peer above or beside us is told is
+//! held to the prefixes they list and to as many routes as they allow,
+//! whatever else would send it more.
+
+use std::collections::BTreeSet;
+use std::net::Ipv4Addr;
+use std::sync::Arc;
+use std::time::Duration;
+
+use bgp_integration_tests::default_policy;
+use bgp_integration_tests::test_peer::TestPeer;
+use bgp_integration_tests::test_server::TestServer;
+use bgp_packet::errors::{BgpError, CeaseSubcode};
+use bgp_packet::message::Message;
+use bgp_packet::open::{BgpRole, Capability};
+use bgp_primitives::constants::AfiSafi;
+use bgp_primitives::ip_prefix::IpPrefix;
+use bgp_rib::LocalRoute;
+use bgp_server::config::PeerConfig;
+use bgp_server::family::FamilySet;
+use bgp_server::policy::{ExportLimits, PeerRole, PrefixSet};
+use eyre::{Result, bail};
+use tokio::net::TcpListener;
+
+const LOCAL_ASN: u32 = 1299;
+const PEER_ASN: u32 = 3333;
+const TIMEOUT: Duration = Duration::from_secs(5);
+
+fn prefix(s: &str) -> IpPrefix {
+    IpPrefix::try_from(s).unwrap()
+}
+
+/// What a test holds on to: the server, and the listener its dialling
+/// goes to.
+struct Setup {
+    server: TestServer,
+    _listener: TcpListener,
+}
+
+/// A server that originates `originated`, with one configured peer at the
+/// address the test connects from, which is a peer of ours held to
+/// `limits`.
+async fn server_with(originated: &[&str], limits: ExportLimits) -> Result<Setup> {
+    let listener = TcpListener::bind((Ipv4Addr::LOCALHOST, 0)).await?;
+    let endpoint = listener.local_addr()?;
+    let originate = originated
+        .iter()
+        .map(|p| LocalRoute::new(prefix(p), Default::default()))
+        .collect();
+    let server = TestServer::start_with(LOCAL_ASN, |c| {
+        c.accept_unconfigured_peers = false;
+        c.originate = originate;
+        c.peers = vec![PeerConfig {
+            graceful_restart: Default::default(),
+            groups: Vec::new(),
+            export_limits: Some(Arc::new(limits)),
+            reflector_client: false,
+            name: "r1".into(),
+            endpoint: endpoint.into(),
+            remote_asn: PEER_ASN,
+            hold_time_secs: 90,
+            local_address: None,
+            families: FamilySet::supported(),
+            add_path: None,
+            import: default_policy(),
+            role: Some(PeerRole::Peer),
+            require_role: false,
+            export: Default::default(),
+        }];
+    })
+    .await?;
+    Ok(Setup {
+        server,
+        _listener: listener,
+    })
+}
+
+/// A session with the server, as its peer.
+async fn established(server: &TestServer) -> Result<TestPeer> {
+    let mut peer = TestPeer::connect(server.addr).await?;
+    let capabilities = vec![
+        Capability::FourByteAsn { asn: PEER_ASN },
+        Capability::multiprotocol(AfiSafi::IPV4_UNICAST),
+        Capability::multiprotocol(AfiSafi::IPV6_UNICAST),
+        Capability::BgpRole {
+            role: BgpRole::Peer,
+        },
+    ];
+    peer.handshake_with(PEER_ASN, Ipv4Addr::new(10, 0, 0, 2), 90, capabilities)
+        .await?;
+    Ok(peer)
+}
+
+/// What the peer has been told once it has been quiet for a while: the
+/// prefixes it holds, and the NOTIFICATION that ended the session, if one
+/// did.
+async fn told(peer: &mut TestPeer) -> (BTreeSet<String>, Option<(BgpError, Vec<u8>)>) {
+    let mut routes = BTreeSet::new();
+    while let Ok(Ok(message)) = tokio::time::timeout(Duration::from_millis(500), peer.recv()).await
+    {
+        match message {
+            Message::Update(update) => {
+                for route in &update.withdrawn_routes {
+                    routes.remove(&route.prefix.to_string());
+                }
+                for route in &update.announced_routes {
+                    routes.insert(route.prefix.to_string());
+                }
+            }
+            Message::Notification(notification) => {
+                return (routes, Some((notification.error, notification.data)));
+            }
+            _ => {}
+        }
+    }
+    (routes, None)
+}
+
+fn allowing(prefixes: &[&str]) -> Option<PrefixSet> {
+    let mut set = PrefixSet::default();
+    for p in prefixes {
+        let p = prefix(p);
+        set.allow(p, p.length());
+    }
+    Some(set)
+}
+
+/// A route of ours whose prefix is not in the peer's export limits is not
+/// announced to it; the one that is, is.
+#[tokio::test]
+async fn test_a_route_outside_the_export_limits_is_not_announced() -> Result<()> {
+    let limits = ExportLimits {
+        prefixes: allowing(&["193.0.0.0/21"]),
+        max_prefixes: None,
+    };
+    let Setup { server, .. } = server_with(&["193.0.0.0/21", "198.51.100.0/24"], limits).await?;
+    let mut peer = established(&server).await?;
+    let (routes, ended) = told(&mut peer).await;
+    assert_eq!(routes, BTreeSet::from(["193.0.0.0/21".to_string()]));
+    assert_eq!(ended, None);
+    Ok(())
+}
+
+/// One route more than the export limit allows is not announced: the
+/// session ends with an Administrative Shutdown that says why (RFC 9003),
+/// and the peer is kept down, so that it is told nothing rather than too
+/// much.
+#[tokio::test]
+async fn test_going_over_the_export_limit_ends_the_session_and_keeps_the_peer_down() -> Result<()> {
+    let limits = ExportLimits {
+        prefixes: None,
+        max_prefixes: Some(2),
+    };
+    let Setup { server, .. } = server_with(&["193.0.0.0/21", "193.0.8.0/21"], limits).await?;
+    let mut peer = established(&server).await?;
+    let (routes, ended) = told(&mut peer).await;
+    assert_eq!(routes.len(), 2);
+    assert_eq!(ended, None);
+
+    server
+        .local_rib
+        .announce(LocalRoute::new(prefix("193.0.16.0/21"), Default::default()))
+        .await?;
+    let (more, ended) = told(&mut peer).await;
+    assert_eq!(more, BTreeSet::new(), "nothing more is announced");
+    let Some((error, data)) = ended else {
+        bail!("the session was not ended");
+    };
+    assert_eq!(error, BgpError::Cease(CeaseSubcode::AdminShutdown));
+    let text = String::from_utf8(data[1..].to_vec())?;
+    assert!(
+        text.starts_with("export limit of 2 routes reached"),
+        "{text}"
+    );
+    assert!(peer.is_closed().await);
+
+    // The peer is kept down: a new connection is closed unanswered.
+    let mut again = TestPeer::connect(server.addr).await?;
+    assert!(
+        matches!(
+            tokio::time::timeout(TIMEOUT, again.recv()).await,
+            Ok(Err(_))
+        ),
+        "a held-down peer gets no session"
+    );
+    Ok(())
+}

crates/integration-tests/tests/import_policy.rs

@@ -44,6 +44,7 @@ async fn server_with(import: ImportPolicy) -> Result<(TestServer, TcpListener)>
         c.peers = vec![PeerConfig {
             graceful_restart: Default::default(),
             groups: Vec::new(),
+            export_limits: None,
             reflector_client: false,
             name: "r1".into(),
             endpoint: endpoint.into(),

crates/integration-tests/tests/negotiation.rs

@@ -185,6 +185,7 @@ async fn test_no_family_in_common_refuses_the_session() -> Result<()> {
         c.peers = vec![PeerConfig {
             graceful_restart: Default::default(),
             groups: Vec::new(),
+            export_limits: None,
             reflector_client: false,
             name: "v6-only".into(),
             endpoint: SocketAddr::new(Ipv4Addr::LOCALHOST.into(), 1).into(),

crates/integration-tests/tests/roles.rs

@@ -50,6 +50,7 @@ async fn server_with(role: Option<PeerRole>, require_role: bool) -> Result<Setup
         c.peers = vec![PeerConfig {
             graceful_restart: Default::default(),
             groups: Vec::new(),
+            export_limits: None,
             reflector_client: false,
             name: "r1".into(),
             endpoint: endpoint.into(),

crates/integration-tests/tests/rpki.rs

@@ -69,6 +69,7 @@ async fn server_with(
         c.peers = vec![PeerConfig {
             graceful_restart: Default::default(),
             groups: Vec::new(),
+            export_limits: None,
             reflector_client: false,
             name: "r1".into(),
             endpoint: endpoint.into(),

crates/server/proto/config.proto

@@ -107,6 +107,37 @@ message BgpConfig {
   // back (RFC 4456). Give routers that serve the same clients the same
   // cluster ID. Defaults to router_id.
   string cluster_id = 15;
+
+  // Limits on what goes to other networks that are not our customers: to
+  // every peer whose role is PROVIDER, PEER or ROUTE_SERVER, and to one
+  // in another AS that has no role. Not to customers, route server
+  // clients or routers of our own AS. A peer's own `export_limits` takes
+  // the place of these.
+  //
+  // They hold whatever the routes are and wherever they came from, and
+  // whatever `export` says: a check on everything else, in case it goes
+  // wrong. With the roles, a route learned from a provider or a peer is
+  // passed on to customers only; these keep it from going further even
+  // if a route is not marked as it should be.
+  ExportLimits export_limits = 16;
+}
+
+// Limits on what is announced to a peer, which hold whatever `export`
+// and the routes say.
+message ExportLimits {
+  // Only routes for these prefixes are announced: our own address space
+  // and our customers', say, with max_length as long as the longest
+  // prefix announced from it. Routes for other prefixes, learned or
+  // originated, are not. Empty allows any.
+  repeated PrefixRange prefixes = 1;
+
+  // The most routes that are announced to the peer at a time. If more
+  // were to be, none of them is: the session is ended with a Cease /
+  // Administrative Shutdown NOTIFICATION that says why (RFC 9003), and the
+  // peer is kept down for its import.hold_down_seconds, as when it sends
+  // us too many. Set it somewhat above what is expected. Zero is no
+  // limit.
+  uint32 max_prefixes = 2;
 }
 
 // Graceful restart (RFC 4724): BGP restarting without the routes going.
@@ -322,6 +353,12 @@ message Peer {
   // told of learned routes by `export` and the roles alone. The peer's
   // `export` still has to pass on learned routes.
   bool route_reflector_client = 14;
+
+  // Limits on what is announced to the peer, in the place of the
+  // top-level export_limits. They apply to the same peers as those do: to
+  // one whose role is PROVIDER, PEER or ROUTE_SERVER, or that is in
+  // another AS and has no role.
+  ExportLimits export_limits = 15;
 }
 
 // What is announced to a peer.

crates/server/src/advertise.rs

@@ -15,6 +15,11 @@
 //! The advertiser itself only starts an exporter when a session comes up
 //! and stops it when the session goes.
 //!
+//! To a peer above or beside us, the export limits are a check on all of
+//! this ([`ExportLimits`]): routes for prefixes outside them are not
+//! announced, whatever else says they are to be, and if more routes were
+//! to be announced than they allow, none of them is and the session ends.
+//!
 //! To a peer that agreed to graceful restart (RFC 4724) an exporter sends
 //! an End-of-RIB marker for each address family once the peer has all our
 //! routes: at once after the originated ones, or, where learned routes are
@@ -42,13 +47,14 @@ use tokio::sync::broadcast::Receiver;
 use tokio::sync::broadcast::error::RecvError;
 use tokio::sync::{Notify, OwnedSemaphorePermit, Semaphore, watch};
 use tokio::task::JoinHandle;
-use tracing::{debug, info, warn};
+use tracing::{debug, error, info, warn};
 
+use crate::error::SessionError;
 use crate::export::{self, Egress, Outgoing, Reflection, Source};
 use crate::family::FamilySet;
 use crate::marks::Control;
 use crate::metrics::ROUTES_EXPORTED;
-use crate::policy::PeerRole;
+use crate::policy::{ExportLimits, PeerRole};
 use crate::session::{PeerId, SessionClosed, SessionEvent, SessionHandle, SessionRegistry};
 
 /// Starts the advertiser. It hears of every session that comes up after
@@ -497,6 +503,7 @@ impl Exporter {
         }
 
         let session = &self.session;
+        let limits = session.export_limits.as_deref();
         let egress = Egress {
             towards,
             peer_asn: session.peer.asn,
@@ -559,7 +566,8 @@ impl Exporter {
                         // that, and not as someone else's route for it.
                         Some(source)
                             if session.families.carries(&prefix)
-                                && !self.announced.contains(&prefix) =>
+                                && !self.announced.contains(&prefix)
+                                && limits.is_none_or(|limits| limits.allows(&prefix)) =>
                         {
                             learned.dressed(&attrs, source, prefix.address_family(), &egress)
                         }
@@ -612,6 +620,16 @@ impl Exporter {
                 withdrawn[version_index(prefix.address_family())].push(prefix.into());
             }
         }
+        // What the peer would have once this goes out: none of it goes if
+        // that is more than it may.
+        let routes = self.announced.len()
+            + learned.announced.len()
+            + learned.stale.as_ref().map_or(0, HashSet::len);
+        if let Some(limits) = limits
+            && limits.exceeded_by(routes)
+        {
+            return Err(over_limit(session, limits, routes));
+        }
         for (version, routes) in [IpVersion::Ipv4, IpVersion::Ipv6]
             .into_iter()
             .zip(withdrawn)
@@ -673,6 +691,25 @@ impl Exporter {
             );
             return Ok(());
         }
+        if let Some(limits) = session.export_limits.as_deref() {
+            if !limits.allows(&route.prefix) {
+                debug!(
+                    peer = %session.peer,
+                    prefix = %route.prefix,
+                    "Not announcing route: its prefix is not in the peer's export limits"
+                );
+                return Ok(());
+            }
+            let learned = self.learned.as_ref().map_or(0, |learned| {
+                learned.announced.len() + learned.stale.as_ref().map_or(0, HashSet::len)
+            });
+            let routes = self.announced.len()
+                + usize::from(!self.announced.contains(&route.prefix))
+                + learned;
+            if limits.exceeded_by(routes) {
+                return Err(over_limit(session, limits, routes));
+            }
+        }
         let towards = self.towards();
         let Some(mut update) = build_announce(route, &towards) else {
             debug!(
@@ -728,6 +765,22 @@ impl Exporter {
     }
 }
 
+/// Ends `session` because `routes` routes were to be announced on it, more
+/// than `limits` allow; none of them has been.
+fn over_limit(session: &SessionHandle, limits: &ExportLimits, routes: usize) -> Finished {
+    let limit = limits.max_prefixes.unwrap_or_default();
+    error!(
+        peer = %session.peer,
+        name = session.name.as_deref().unwrap_or_default(),
+        routes,
+        limit,
+        "More routes were to be advertised to the peer than its export limit allows; advertising \
+         none of them, ending the session and keeping the peer down"
+    );
+    session.end(SessionError::ExportLimitExceeded { limit, routes });
+    Finished
+}
+
 /// What about a session decides how a route is announced on it.
 #[derive(Debug, Clone, Copy)]
 pub struct Towards {
@@ -1579,4 +1632,203 @@ mod tests {
         let mut second = fixture.session(1, 4);
         second.read_until_it_has(&routes).await;
     }
+
+    // -- Export limits --
+
+    use crate::policy::{ExportLimits, ExportPolicy, PrefixSet};
+
+    /// A table with routes from a customer for `prefixes`, an advertiser
+    /// over it with `originated` routes of our own, and a session to a
+    /// peer of ours that is told of learned routes, within `limits`.
+    async fn limited(
+        prefixes: &[IpPrefix],
+        originated: &[IpPrefix],
+        limits: ExportLimits,
+    ) -> (
+        mpsc::Receiver<Message>,
+        mpsc::UnboundedReceiver<SessionError>,
+        SessionGuard,
+    ) {
+        let registry = SessionRegistry::new();
+        let rib = Arc::new(RibTable::new(4));
+        let local_rib = LocalRib::spawn();
+        for prefix in originated {
+            let route = LocalRoute::new(*prefix, LocalRouteAttrs::default());
+            local_rib.announce(route).await.unwrap();
+        }
+        spawn_advertiser(
+            local_rib,
+            Arc::clone(&rib),
+            registry.clone(),
+            65000,
+            None,
+            Ipv4Addr::new(192, 0, 2, 1),
+            watch::channel(true).1,
+        );
+        let peer = |n: u8, asn| PeerInfo {
+            asn,
+            router_id: Ipv4Addr::new(10, 0, 0, n),
+            address: IpAddr::V4(Ipv4Addr::new(192, 0, 2, n)),
+            interface: 0,
+            ibgp: false,
+            reflector_client: false,
+        };
+        // A customer's routes carry no Only to Customer mark: as far as
+        // the roles go, they are for everyone.
+        let customer = rib.add_peer(peer(1, 64496)).unwrap();
+        let came = [
+            PathAttribute::Origin(OriginPathAttribute::IGP),
+            AsPathAttribute::from_asns(vec![64496]),
+            PathAttribute::NextHop(NextHopPathAttribute(Ipv4Addr::new(192, 0, 2, 1))),
+        ];
+        let attrs = rib.attrs().intern(&came, None).unwrap();
+        let nlri: Vec<Nlri> = prefixes.iter().copied().map(Nlri::from).collect();
+        rib.announce(customer, &attrs, &nlri);
+
+        let negotiated = Negotiated {
+            peer_asn: 64497,
+            ibgp: false,
+            peer_identifier: Ipv4Addr::new(10, 0, 0, 2),
+            hold_time_secs: 90,
+            four_octet_asn: true,
+            route_refresh: false,
+            families: FamilySet::supported(),
+            add_path_receive: FamilySet::default(),
+            add_path_send: FamilySet::default(),
+            role_confirmed: false,
+            extended_next_hop: false,
+            graceful_restart: None,
+        };
+        let rib_peer = rib.add_peer(peer(2, 64497)).unwrap();
+        let (tx, outbound) = mpsc::channel(1024);
+        let mut handle = SessionHandle::new(
+            PeerId {
+                addr: (Ipv4Addr::new(192, 0, 2, 2), 179).into(),
+                asn: 64497,
+            },
+            None,
+            Direction::Inbound,
+            (Ipv4Addr::new(192, 0, 2, 100), 179).into(),
+            &negotiated,
+            rib_peer,
+            tx,
+        );
+        handle.role = Some(PeerRole::Peer);
+        handle.export = Arc::new(ExportPolicy {
+            learned_routes: true,
+            ..ExportPolicy::default()
+        });
+        handle.export_limits = Some(Arc::new(limits));
+        let ending = handle.listen_for_end();
+        (outbound, ending, registry.register(handle))
+    }
+
+    /// The routes announced on a session's queue until it has been quiet
+    /// for a while.
+    async fn announced(outbound: &mut mpsc::Receiver<Message>) -> HashSet<IpPrefix> {
+        let mut routes = HashSet::new();
+        while let Ok(Some(message)) =
+            tokio::time::timeout(Duration::from_millis(500), outbound.recv()).await
+        {
+            if let Message::Update(update) = message {
+                for nlri in &update.withdrawn_routes {
+                    routes.remove(&nlri.prefix);
+                }
+                routes.extend(update.announced_routes.iter().map(|nlri| nlri.prefix));
+                if let Some(PathAttribute::MpReachNlri(reach)) = update
+                    .path_attributes
+                    .iter()
+                    .find(|a| matches!(a, PathAttribute::MpReachNlri(_)))
+                {
+                    routes.extend(reach.prefixes.iter().map(|nlri| nlri.prefix));
+                }
+            }
+        }
+        routes
+    }
+
+    fn allowing(prefixes: &[IpPrefix]) -> Option<PrefixSet> {
+        let mut set = PrefixSet::default();
+        for prefix in prefixes {
+            set.allow(*prefix, prefix.length());
+        }
+        Some(set)
+    }
+
+    /// Only routes for the prefixes in the export limits go to a peer,
+    /// learned or our own, though nothing else keeps the others from it:
+    /// a customer's routes are for everyone, as far as the roles go.
+    #[tokio::test]
+    async fn routes_outside_the_export_limits_do_not_go_to_a_peer() {
+        let limits = ExportLimits {
+            prefixes: allowing(&[nth_prefix(2), nth_prefix(11)]),
+            max_prefixes: None,
+        };
+        let (mut outbound, mut ending, _registration) = limited(
+            &[nth_prefix(1), nth_prefix(2), nth_prefix(3)],
+            &[nth_prefix(10), nth_prefix(11)],
+            limits,
+        )
+        .await;
+        assert_eq!(
+            announced(&mut outbound).await,
+            HashSet::from([nth_prefix(2), nth_prefix(11)])
+        );
+        assert!(ending.try_recv().is_err(), "the session goes on");
+    }
+
+    /// When more routes would go to a peer than its export limit allows,
+    /// none of them does, and the session is ended: a leak becomes a
+    /// session that is down, and not a table spread across the Internet.
+    #[tokio::test]
+    async fn more_routes_than_the_export_limit_end_the_session_and_none_goes() {
+        let prefixes: Vec<IpPrefix> = (1..=5).map(nth_prefix).collect();
+        let limits = ExportLimits {
+            prefixes: None,
+            max_prefixes: Some(3),
+        };
+        let (mut outbound, mut ending, _registration) = limited(&prefixes, &[], limits).await;
+        let why = tokio::time::timeout(Duration::from_secs(10), ending.recv())
+            .await
+            .expect("the session is asked to end")
+            .expect("by the exporter");
+        assert!(
+            matches!(
+                why,
+                SessionError::ExportLimitExceeded {
+                    limit: 3,
+                    routes: 5
+                }
+            ),
+            "{why:?}"
+        );
+        assert_eq!(announced(&mut outbound).await, HashSet::new());
+    }
+
+    /// Our own routes count too, and none goes past the limit: those that
+    /// fit go, and the one that does not ends the session.
+    #[tokio::test]
+    async fn originated_routes_count_towards_the_export_limit() {
+        let originated: Vec<IpPrefix> = (10..13).map(nth_prefix).collect();
+        let limits = ExportLimits {
+            prefixes: None,
+            max_prefixes: Some(2),
+        };
+        let (mut outbound, mut ending, _registration) = limited(&[], &originated, limits).await;
+        let why = tokio::time::timeout(Duration::from_secs(10), ending.recv())
+            .await
+            .expect("the session is asked to end")
+            .expect("by the exporter");
+        assert!(
+            matches!(
+                why,
+                SessionError::ExportLimitExceeded {
+                    limit: 2,
+                    routes: 3
+                }
+            ),
+            "{why:?}"
+        );
+        assert_eq!(announced(&mut outbound).await.len(), 2);
+    }
 }

crates/server/src/config.rs

@@ -16,7 +16,9 @@ use crate::address::{Endpoint, PeerAddress};
 use crate::error::{ConfigError, ExportProblem, ImportProblem, OriginateProblem, PeerProblem};
 use crate::family::FamilySet;
 use crate::marks::{Control, Marks, Patterns, Scrub};
-use crate::policy::{ExportPolicy, ImportPolicy, ImportRule, PeerRole, PrefixSet, Reflection};
+use crate::policy::{
+    ExportLimits, ExportPolicy, ImportPolicy, ImportRule, PeerRole, PrefixSet, Reflection,
+};
 use crate::proto::config::{self as proto, BgpConfig, OriginatedRoute, Peer};
 use crate::restart::GracefulRestartConfig;
 
@@ -59,6 +61,9 @@ pub struct Config {
     pub rpki: Option<RpkiConfig>,
     /// What a peer is told that has no `export` of its own.
     pub export: Arc<ExportPolicy>,
+    /// Limits on what goes to a peer that has none of its own, if it is
+    /// one they are for ([`ExportLimits::apply_to`]).
+    pub export_limits: Option<Arc<ExportLimits>>,
     /// The communities by which routes say who is told of them, if they
     /// are honoured.
     pub control: Option<Control>,
@@ -92,6 +97,9 @@ pub struct PeerConfig {
     pub require_role: bool,
     /// What the peer is told, and how.
     pub export: Arc<ExportPolicy>,
+    /// Limits on what the peer is told, if it is one they are for
+    /// ([`ExportLimits::apply_to`]): its own, or the top-level ones.
+    pub export_limits: Option<Arc<ExportLimits>>,
     /// Graceful restart with the peer (RFC 4724).
     pub graceful_restart: GracefulRestartConfig,
     /// The groups the session is in, for control communities to name it
@@ -147,6 +155,9 @@ impl Config {
             Some(export) => parse_export(export)?,
             None => ExportPolicy::default(),
         });
+        let export_limits = parse_export_limits(proto.export_limits.as_ref())
+            .map_err(ConfigError::ExportLimits)?
+            .map(Arc::new);
         let graceful_restart = match proto.graceful_restart.as_ref() {
             Some(own) => GracefulRestartConfig::default().overridden_by(own)?,
             None => GracefulRestartConfig::default(),
@@ -158,7 +169,8 @@ impl Config {
                 peer: name.clone(),
                 problem,
             };
-            let peer = parse_peer(peer, &export, graceful_restart).map_err(rejected)?;
+            let peer =
+                parse_peer(peer, &export, &export_limits, graceful_restart).map_err(rejected)?;
             // RFC 9234 is about what networks are to each other.
             if peer.role.is_some() && peer.remote_asn == proto.local_asn {
                 return Err(rejected(PeerProblem::RoleInOwnAs));
@@ -179,6 +191,25 @@ impl Config {
             }
             peers.push(peer);
         }
+        // Where routes learned from one peer go on to others, which go
+        // where is decided by the peers' roles: a peer in another AS
+        // without one would have its routes go to everyone, and be told
+        // of everything.
+        let learned_go_on =
+            export.learned_routes || peers.iter().any(|peer| peer.export.learned_routes);
+        if learned_go_on
+            && let Some(peer) = peers
+                .iter()
+                .find(|peer| peer.remote_asn != proto.local_asn && peer.role.is_none())
+        {
+            return Err(ConfigError::Peer {
+                peer: peer.name.clone(),
+                problem: PeerProblem::NoRole,
+            });
+        }
+        if proto.accept_unconfigured_peers && export.learned_routes {
+            return Err(ConfigError::UnconfiguredPeersWithoutRole);
+        }
 
         let mut originate: Vec<LocalRoute> = Vec::with_capacity(proto.originate.len());
         for entry in &proto.originate {
@@ -257,6 +288,7 @@ impl Config {
             kernel: proto.kernel.as_ref().map(parse_kernel),
             rpki,
             export,
+            export_limits,
             control: proto.control_communities.then_some(Control {
                 local_asn: proto.local_asn,
             }),
@@ -296,9 +328,29 @@ fn parse_export(export: &proto::ExportPolicy) -> Result<ExportPolicy, ExportProb
     })
 }
 
+/// Limits on what is announced; `None` if there are none, as when the
+/// message is left out or says nothing.
+fn parse_export_limits(
+    limits: Option<&proto::ExportLimits>,
+) -> Result<Option<ExportLimits>, ImportProblem> {
+    let Some(limits) = limits else {
+        return Ok(None);
+    };
+    let limits = ExportLimits {
+        prefixes: if limits.prefixes.is_empty() {
+            None
+        } else {
+            Some(parse_prefixes(&limits.prefixes)?)
+        },
+        max_prefixes: Some(limits.max_prefixes).filter(|limit| *limit > 0),
+    };
+    Ok((limits != ExportLimits::default()).then_some(limits))
+}
+
 fn parse_peer(
     peer: Peer,
     export: &Arc<ExportPolicy>,
+    export_limits: &Option<Arc<ExportLimits>>,
     graceful_restart: GracefulRestartConfig,
 ) -> Result<PeerConfig, PeerProblem> {
     let ep = peer.endpoint.as_ref().ok_or(PeerProblem::MissingEndpoint)?;
@@ -392,6 +444,12 @@ fn parse_peer(
             Some(own) => Arc::new(parse_export(own)?),
             None => Arc::clone(export),
         },
+        export_limits: match peer.export_limits.as_ref() {
+            Some(own) => parse_export_limits(Some(own))
+                .map_err(PeerProblem::ExportLimits)?
+                .map(Arc::new),
+            None => export_limits.clone(),
+        },
         graceful_restart: match peer.graceful_restart.as_ref() {
             Some(own) => graceful_restart.overridden_by(own)?,
             None => graceful_restart,
@@ -1235,4 +1293,111 @@ mod tests {
                 .accept_rpki_invalid
         );
     }
+
+    /// Where learned routes are passed on, every peer in another AS has
+    /// to have a role: without one, its routes would go to everyone and it
+    /// would be told of everything. Where they are not, or for our own
+    /// routers, there is nothing for a role to decide.
+    #[test]
+    fn a_peer_in_another_as_needs_a_role_where_learned_routes_go_on() {
+        let peer = |body: &str| {
+            format!("peers {{ name: \"p\" endpoint {{ address: \"192.0.2.2\" }} {body} }}")
+        };
+        let learned = "export { learned_routes: true }";
+        assert_eq!(
+            rejected(&format!("{BASE}{learned}{}", peer("remote_asn: 64496"))),
+            ConfigError::Peer {
+                peer: "p".into(),
+                problem: PeerProblem::NoRole
+            }
+        );
+        // The peer's own export passes them on; or another peer's does.
+        assert!(matches!(
+            rejected(&format!(
+                "{BASE}{}",
+                peer(&format!("remote_asn: 64496 {learned}"))
+            )),
+            ConfigError::Peer {
+                problem: PeerProblem::NoRole,
+                ..
+            }
+        ));
+        assert!(matches!(
+            rejected(&format!(
+                "{BASE}{}
+                 peers {{ name: \"c\" endpoint {{ address: \"192.0.2.3\" }} remote_asn: 64497
+                         role: CUSTOMER {learned} }}",
+                peer("remote_asn: 64496")
+            )),
+            ConfigError::Peer {
+                problem: PeerProblem::NoRole,
+                ..
+            }
+        ));
+        parse(&format!(
+            "{BASE}{learned}{}",
+            peer("remote_asn: 64496 role: PEER")
+        ))
+        .unwrap();
+        parse(&format!("{BASE}{learned}{}", peer("remote_asn: 65000"))).unwrap();
+        parse(&format!("{BASE}{}", peer("remote_asn: 64496"))).unwrap();
+        // Peers that are not configured have no role either.
+        assert_eq!(
+            rejected(&format!("{BASE}{learned} accept_unconfigured_peers: true")),
+            ConfigError::UnconfiguredPeersWithoutRole
+        );
+        parse(&format!("{BASE}accept_unconfigured_peers: true")).unwrap();
+    }
+
+    /// The export limits at the top hold for every peer that has none of
+    /// its own; a peer's own take their place, and an empty one lifts
+    /// them. Which peers they apply to is the role's to say, when a
+    /// session comes up.
+    #[test]
+    fn export_limits_are_the_top_levels_or_a_peers_own() {
+        let config = parse(&format!(
+            "{BASE}export_limits {{
+                 prefixes {{ prefix: \"192.0.2.0/24\" }}
+                 prefixes {{ prefix: \"2001:db8::/32\" max_length: 48 }}
+                 max_prefixes: 10
+             }}
+             peers {{ name: \"a\" endpoint {{ address: \"192.0.2.2\" }} remote_asn: 64496
+                     role: PEER }}
+             peers {{ name: \"b\" endpoint {{ address: \"192.0.2.3\" }} remote_asn: 64497
+                     role: PROVIDER export_limits {{ max_prefixes: 50 }} }}
+             peers {{ name: \"c\" endpoint {{ address: \"192.0.2.4\" }} remote_asn: 64498
+                     role: PEER export_limits {{ }} }}"
+        ))
+        .unwrap();
+        let prefix = |text: &str| IpPrefix::try_from(text).unwrap();
+        let top = config.export_limits.as_deref().expect("top-level limits");
+        assert_eq!(top.max_prefixes, Some(10));
+        assert!(top.allows(&prefix("192.0.2.0/24")));
+        assert!(!top.allows(&prefix("192.0.2.0/25")));
+        assert!(top.allows(&prefix("2001:db8:1::/48")));
+        assert!(!top.allows(&prefix("2001:db8:1::/49")));
+        assert!(!top.allows(&prefix("198.51.100.0/24")));
+        assert_eq!(config.peers[0].export_limits.as_deref(), Some(top));
+        let own = config.peers[1].export_limits.as_deref().expect("its own");
+        assert_eq!(own.max_prefixes, Some(50));
+        assert!(own.allows(&prefix("198.51.100.0/24")));
+        assert_eq!(config.peers[2].export_limits, None);
+
+        assert!(matches!(
+            rejected(&format!(
+                "{BASE}export_limits {{ prefixes {{ prefix: \"192.0.2.0/33\" }} }}"
+            )),
+            ConfigError::ExportLimits(ImportProblem::Prefix { .. })
+        ));
+        assert!(matches!(
+            rejected(&format!(
+                "{BASE}peers {{ name: \"a\" endpoint {{ address: \"192.0.2.2\" }} remote_asn: 1
+                 export_limits {{ prefixes {{ prefix: \"192.0.2.0/24\" max_length: 16 }} }} }}"
+            )),
+            ConfigError::Peer {
+                problem: PeerProblem::ExportLimits(ImportProblem::MaxLength { .. }),
+                ..
+            }
+        ));
+    }
 }

crates/server/src/connection.rs

@@ -42,11 +42,13 @@ use crate::error::SessionError;
 use crate::family::FamilySet;
 use crate::fsm::{Action, Event, Fsm, FsmConfig, Negotiated, Received};
 use crate::metrics::{
-    MESSAGES_RECEIVED, MESSAGES_SENT, PEER_ROUTES, PREFIX_LIMIT_EXCEEDED,
+    EXPORT_LIMIT_EXCEEDED, MESSAGES_RECEIVED, MESSAGES_SENT, PEER_ROUTES, PREFIX_LIMIT_EXCEEDED,
     SESSIONS_ESTABLISHED_TOTAL, UPDATE_ERRORS, message_type_name,
 };
 use crate::panic::contained;
-use crate::policy::{self, ExportPolicy, HeldDown, HoldDowns, ImportPolicy, PeerRole};
+use crate::policy::{
+    self, ExportLimits, ExportPolicy, HeldDown, HoldDowns, ImportPolicy, PeerRole,
+};
 use crate::reach::{Reach, neighbor_at};
 use crate::refusals::{PeerRefusals, RefusalLog};
 use crate::restart::{self, Agreed, GracefulRestartConfig, Leave, Restarts};
@@ -114,6 +116,8 @@ pub struct PeerSettings {
     pub require_role: bool,
     /// What the peer is told, and how.
     pub export: Arc<ExportPolicy>,
+    /// Limits on what the peer is told, if it is one they are for.
+    pub export_limits: Option<Arc<ExportLimits>>,
     /// The groups the session is in, for control communities.
     pub groups: Arc<[u32]>,
     /// Whether the peer is a route reflector client of ours (RFC 4456).
@@ -129,6 +133,7 @@ impl PeerSettings {
     pub fn unconfigured(
         hold_time_secs: u16,
         export: Arc<ExportPolicy>,
+        export_limits: Option<Arc<ExportLimits>>,
         graceful_restart: GracefulRestartConfig,
     ) -> Self {
         Self {
@@ -141,6 +146,7 @@ impl PeerSettings {
             role: None,
             require_role: false,
             export,
+            export_limits,
             groups: Arc::default(),
             reflector_client: false,
             graceful_restart,
@@ -160,6 +166,7 @@ impl From<&PeerConfig> for PeerSettings {
             role: peer.role,
             require_role: peer.require_role,
             export: Arc::clone(&peer.export),
+            export_limits: peer.export_limits.clone(),
             groups: peer.groups.as_slice().into(),
             reflector_client: peer.reflector_client,
             graceful_restart: peer.graceful_restart,
@@ -215,6 +222,19 @@ pub async fn run_session(
             .with_label_values(&[session.refusals.label()])
             .inc();
     }
+    if matches!(result, Err(SessionError::ExportLimitExceeded { .. })) {
+        // Whatever made us want to send too much will be there on the
+        // next session too. Until it is put right, the peer is told of
+        // nothing rather than of too much.
+        hold_downs.hold(
+            peer_addr.into(),
+            session.import.hold_down,
+            HeldDown::OverExportLimit,
+        );
+        EXPORT_LIMIT_EXCEEDED
+            .with_label_values(&[session.refusals.label()])
+            .inc();
+    }
     // A configured peer whose sessions keep coming up and failing is
     // slowed down. Not one that is nobody in particular: such peers are
     // only taken on a test network, where many share an address.
@@ -304,6 +324,8 @@ struct Established {
     rib_peer: PeerIndex,
     /// Messages the rest of the server queued for the peer.
     outbound: mpsc::Receiver<Message>,
+    /// Where the rest of the server asks for the session to end, and why.
+    ending: mpsc::UnboundedReceiver<SessionError>,
     /// The routes learned from the peer, which go when the session does.
     routes: PeerRoutes,
     /// Whose routes the import policy is judging.
@@ -575,6 +597,8 @@ struct Session {
     role: Option<PeerRole>,
     /// What the peer is told, and how.
     export: Arc<ExportPolicy>,
+    /// Limits on what the peer is told, if it is one they are for.
+    export_limits: Option<Arc<ExportLimits>>,
     /// The groups the session is in, for control communities.
     groups: Arc<[u32]>,
     /// Whether the peer is a route reflector client of ours (RFC 4456).
@@ -641,6 +665,7 @@ impl Session {
             import: peer.import,
             role: peer.role,
             export: peer.export,
+            export_limits: peer.export_limits,
             groups: peer.groups,
             reflector_client: peer.reflector_client,
             said: Said::new(),
@@ -946,6 +971,12 @@ impl Session {
         handle.role = self.role;
         handle.role_confirmed = negotiated.role_confirmed;
         handle.export = Arc::clone(&self.export);
+        // To a network above or beside us, or one we know nothing of.
+        handle.export_limits = self
+            .export_limits
+            .clone()
+            .filter(|_| ExportLimits::apply_to(self.role, negotiated.ibgp));
+        let ending = handle.listen_for_end();
         handle.groups = Arc::clone(&self.groups);
         handle.reflector_client = self.reflector_client && negotiated.ibgp;
         // Routes from other networks are validated, where there is data
@@ -987,6 +1018,7 @@ impl Session {
             peer_id,
             rib_peer,
             outbound,
+            ending,
             routes: PeerRoutes::new(Arc::clone(&self.server.rib), rib_peer),
             from: policy::RouteSource {
                 peer_asn: negotiated.peer_asn,
@@ -1264,6 +1296,7 @@ fn sweep(
 async fn from_server(established: &mut Option<Established>) -> Wake {
     let Some(Established {
         outbound,
+        ending,
         validation,
         stale,
         swept,
@@ -1285,6 +1318,7 @@ async fn from_server(established: &mut Option<Established>) -> Wake {
         }
     };
     tokio::select! {
+        Some(error) = ending.recv() => Wake::Event(Event::LocalFailure(error)),
         Some(msg) = outbound.recv() => Wake::Event(Event::Outbound(msg)),
         Some(suspects) = invalidated => Wake::Revalidate(suspects),
         Some(outcome) = swept.1.recv() => Wake::Swept(outcome),
@@ -1524,7 +1558,7 @@ mod tests {
             peer_addr,
             Direction::Inbound,
             server,
-            PeerSettings::unconfigured(90, Arc::default(), GracefulRestartConfig::default()),
+            PeerSettings::unconfigured(90, Arc::default(), None, GracefulRestartConfig::default()),
         )
         .unwrap();
 
@@ -1575,7 +1609,7 @@ mod tests {
             peer_addr,
             Direction::Inbound,
             server,
-            PeerSettings::unconfigured(90, Arc::default(), GracefulRestartConfig::default()),
+            PeerSettings::unconfigured(90, Arc::default(), None, GracefulRestartConfig::default()),
         )
         .unwrap();
         session.send_hold_time = Duration::from_millis(100);

crates/server/src/error.rs

@@ -82,6 +82,13 @@ pub enum SessionError {
     /// down for a while after.
     #[error("the peer went over its limit of {limit} routes")]
     PrefixLimitExceeded { limit: u32 },
+    /// We were to announce the peer more routes than its export limit
+    /// allows, and did not. It is kept down for a while after.
+    #[error(
+        "{routes} routes were to be advertised to the peer, over its export limit of {limit}; \
+         none of them was"
+    )]
+    ExportLimitExceeded { limit: u32, routes: usize },
     /// The peer took nothing we sent for this long (RFC 9687): it has
     /// stopped reading, and there is no telling it so.
     #[error("the peer accepted no data for {0:?}")]
@@ -117,6 +124,17 @@ impl SessionError {
             Self::Collision => BgpError::Cease(CeaseSubcode::ConnCollisionResolution),
             Self::TableFull(_) => BgpError::Cease(CeaseSubcode::OutOfResources),
             Self::PrefixLimitExceeded { .. } => BgpError::Cease(CeaseSubcode::MaxPrefixes),
+            // Nothing is wrong at the peer's end, as far as we know; its
+            // operators are told what is at ours.
+            Self::ExportLimitExceeded { limit, .. } => {
+                return Some(NotificationMessage {
+                    error: BgpError::Cease(CeaseSubcode::AdminShutdown),
+                    data: shutdown_communication(&format!(
+                        "export limit of {limit} routes reached; no more routes were sent, \
+                         and the session is held down for a while"
+                    )),
+                });
+            }
             // The session will be set up again; nothing is wrong at the
             // peer's end.
             Self::Panicked(_) => BgpError::Cease(CeaseSubcode::AdminReset),
@@ -136,6 +154,19 @@ impl SessionError {
     }
 }
 
+/// A Shutdown Communication (RFC 9003 §2): text for the peer's operators,
+/// at most 255 octets of UTF-8, after its length.
+fn shutdown_communication(text: &str) -> Vec<u8> {
+    let mut end = text.len().min(255);
+    while !text.is_char_boundary(end) {
+        end -= 1;
+    }
+    let mut data = Vec::with_capacity(end + 1);
+    data.push(end as u8);
+    data.extend_from_slice(&text.as_bytes()[..end]);
+    data
+}
+
 /// Why the server could not start, or could not go on.
 #[derive(Debug, Error)]
 pub enum ServerError {
@@ -180,6 +211,14 @@ pub enum ConfigError {
     RpkiNoCaches,
     #[error("export: {0}")]
     Export(#[from] ExportProblem),
+    #[error("export_limits: {0}")]
+    ExportLimits(ImportProblem),
+    #[error(
+        "accept_unconfigured_peers is set and export.learned_routes passes learned routes on: \
+         a peer that is not configured has no role, so the routes it sends would go to every \
+         peer, and every route to it"
+    )]
+    UnconfiguredPeersWithoutRole,
     #[error("rpki: the port of cache {0:?} is required")]
     RpkiCachePort(String),
     #[error("graceful_restart: {0}")]
@@ -226,6 +265,14 @@ pub enum PeerProblem {
     Group,
     #[error("route_reflector_client is set for a peer in another AS; it is for routers of our own")]
     ReflectorClientOutsideAs,
+    #[error(
+        "role is not set, and learned routes are passed on (export.learned_routes): without a \
+         role, the routes the peer sends would go to every other peer, and it would be told of \
+         every route. Set role to PROVIDER, CUSTOMER, PEER, ROUTE_SERVER or ROUTE_SERVER_CLIENT"
+    )]
+    NoRole,
+    #[error("export_limits: {0}")]
+    ExportLimits(ImportProblem),
     #[error("import: {0}")]
     Import(#[from] ImportProblem),
     #[error("export: {0}")]
@@ -356,5 +403,31 @@ mod tests {
             notification(SessionError::Io(io::ErrorKind::BrokenPipe.into())),
             None
         );
+        // Going over an export limit is ours to explain: an Administrative
+        // Shutdown, with the reason for the peer's operators (RFC 9003).
+        let (error, data) = notification(SessionError::ExportLimitExceeded {
+            limit: 50,
+            routes: 900_000,
+        })
+        .unwrap();
+        assert_eq!(error, BgpError::Cease(CeaseSubcode::AdminShutdown));
+        let text = std::str::from_utf8(&data[1..]).unwrap();
+        assert_eq!(usize::from(data[0]), text.len());
+        assert!(
+            text.starts_with("export limit of 50 routes reached"),
+            "{text}"
+        );
+    }
+
+    /// A Shutdown Communication is at most 255 octets, and whole
+    /// characters (RFC 9003 §2).
+    #[test]
+    fn a_shutdown_communication_is_cut_to_fit() {
+        let long = "é".repeat(200);
+        let data = shutdown_communication(&long);
+        assert_eq!(usize::from(data[0]), data.len() - 1);
+        assert_eq!(data[0], 254);
+        assert!(std::str::from_utf8(&data[1..]).is_ok());
+        assert_eq!(shutdown_communication(""), [0]);
     }
 }

crates/server/src/metrics.rs

@@ -87,6 +87,15 @@ pub static PREFIX_LIMIT_EXCEEDED: LazyLock<IntCounterVec> = LazyLock::new(|| {
     .expect("metric registration")
 });
 
+pub static EXPORT_LIMIT_EXCEEDED: LazyLock<IntCounterVec> = LazyLock::new(|| {
+    register_int_counter_vec!(
+        "bgp_export_limit_exceeded_total",
+        "Sessions ended because more routes were to be announced to the peer than its export limit allows",
+        &["peer"]
+    )
+    .expect("metric registration")
+});
+
 pub static RPKI_VRPS: LazyLock<IntGaugeVec> = LazyLock::new(|| {
     register_int_gauge_vec!(
         "bgp_rpki_vrps",
@@ -206,6 +215,7 @@ pub fn init() {
     LazyLock::force(&PEER_ROUTES);
     LazyLock::force(&PEER_MAX_PREFIXES);
     LazyLock::force(&PREFIX_LIMIT_EXCEEDED);
+    LazyLock::force(&EXPORT_LIMIT_EXCEEDED);
     LazyLock::force(&RPKI_VRPS);
     LazyLock::force(&RPKI_ASPAS);
     LazyLock::force(&RPKI_UPDATES);

crates/server/src/policy.rs

@@ -269,6 +269,40 @@ pub struct ExportPolicy {
     pub prepend: u8,
 }
 
+/// Limits on what is announced to a network that is not our customer,
+/// which hold whatever the export policy and the routes say: a check on
+/// both, in case either goes wrong.
+#[derive(Debug, Clone, Default, PartialEq, Eq)]
+pub struct ExportLimits {
+    /// The prefixes that may be announced; `None` for any.
+    pub prefixes: Option<PrefixSet>,
+    /// The most routes announced at a time.
+    pub max_prefixes: Option<u32>,
+}
+
+impl ExportLimits {
+    /// Whether they are for a session to a peer that is `role` to us, in
+    /// another AS or not (`ibgp`): one above or beside us, or one we know
+    /// nothing of. Customers, route server clients and our own routers
+    /// are told of whatever they are to be.
+    pub fn apply_to(role: Option<PeerRole>, ibgp: bool) -> bool {
+        !ibgp && role.is_none_or(PeerRole::is_above_or_beside)
+    }
+
+    /// Whether a route for `prefix` may be announced.
+    pub fn allows(&self, prefix: &IpPrefix) -> bool {
+        self.prefixes
+            .as_ref()
+            .is_none_or(|prefixes| prefixes.allows(prefix))
+    }
+
+    /// Whether `routes` routes at a time are more than may be announced.
+    pub fn exceeded_by(&self, routes: usize) -> bool {
+        self.max_prefixes
+            .is_some_and(|limit| routes > limit as usize)
+    }
+}
+
 /// Why a route was not taken.
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
 pub enum Rejection {
@@ -881,6 +915,8 @@ impl PrefixSet {
 pub enum HeldDown {
     /// It went over its limit of routes.
     OverLimit,
+    /// We were to announce it more routes than its export limit allows.
+    OverExportLimit,
     /// Its sessions keep coming up and failing.
     Flapping,
 }
@@ -889,6 +925,9 @@ impl HeldDown {
     pub fn as_str(self) -> &'static str {
         match self {
             Self::OverLimit => "held down after going over its limit of routes",
+            Self::OverExportLimit => {
+                "held down after more routes were to be advertised to it than its export limit allows"
+            }
             Self::Flapping => "held down because its sessions keep failing",
         }
     }
@@ -1333,6 +1372,31 @@ mod tests {
         assert!(!Peer.may_be_sent_only_to_customer_routes());
     }
 
+    /// Export limits are for networks above or beside us, and for one we
+    /// know nothing of; not for customers, route server clients or our own
+    /// routers, which are told of what they are to be.
+    #[test]
+    fn export_limits_are_for_those_above_or_beside_us() {
+        use PeerRole::*;
+        for role in [Provider, Peer, RouteServer] {
+            assert!(ExportLimits::apply_to(Some(role), false), "{role:?}");
+        }
+        assert!(ExportLimits::apply_to(None, false));
+        for role in [Customer, RouteServerClient] {
+            assert!(!ExportLimits::apply_to(Some(role), false), "{role:?}");
+        }
+        assert!(!ExportLimits::apply_to(None, true));
+
+        let limits = ExportLimits {
+            prefixes: None,
+            max_prefixes: Some(2),
+        };
+        assert!(!limits.exceeded_by(2));
+        assert!(limits.exceeded_by(3));
+        assert!(limits.allows(&IpPrefix::try_from("203.0.113.0/24").unwrap()));
+        assert!(!ExportLimits::default().exceeded_by(usize::MAX));
+    }
+
     #[test]
     fn the_path_is_read_with_its_sets() {
         let mut out = vec![99];

crates/server/src/server.rs

@@ -27,7 +27,7 @@ use crate::dialer::spawn_dialers;
 use crate::error::ServerError;
 use crate::metrics::{CONNECTIONS_REFUSED, PEER_MAX_PREFIXES};
 use crate::panic;
-use crate::policy::HoldDowns;
+use crate::policy::{ExportLimits, HoldDowns};
 use crate::reach;
 use crate::refusals::RefusalLog;
 use crate::restart::{Leave, Restarts};
@@ -210,6 +210,43 @@ impl BgpServer {
                 );
             }
         }
+        // The export limits that hold for some peer, each once, with the
+        // peers they hold for.
+        let mut limited: Vec<(&ExportLimits, Vec<&str>)> = Vec::new();
+        for peer in &self.config.peers {
+            let ibgp = peer.remote_asn == self.config.local_asn;
+            if !ExportLimits::apply_to(peer.role, ibgp) {
+                continue;
+            }
+            let Some(limits) = peer.export_limits.as_deref() else {
+                warn!(
+                    peer = %peer.name,
+                    "Nothing limits what this peer is told (export_limits)"
+                );
+                continue;
+            };
+            match limited
+                .iter_mut()
+                .find(|(seen, _)| std::ptr::eq(*seen, limits))
+            {
+                Some((_, names)) => names.push(&peer.name),
+                None => limited.push((limits, vec![&peer.name])),
+            }
+        }
+        // A route of our own that the limits leave out is likely a mistake
+        // in one or the other.
+        for route in &self.config.originate {
+            for (limits, names) in &limited {
+                if !limits.allows(&route.prefix) {
+                    warn!(
+                        prefix = %route.prefix,
+                        peers = names.join(", "),
+                        "An originated route is not announced to these peers: export_limits.prefixes \
+                         leaves it out"
+                    );
+                }
+            }
+        }
         if self.config.accept_unconfigured_peers {
             warn!("Taking sessions and routes from any host (accept_unconfigured_peers)");
         }
@@ -421,6 +458,7 @@ fn admit(
         None if config.accept_unconfigured_peers => PeerSettings::unconfigured(
             config.hold_time_secs,
             Arc::clone(&config.export),
+            config.export_limits.clone(),
             config.graceful_restart,
         ),
         None => return Err(Refusal::NotAPeer),

crates/server/src/session.rs

@@ -18,10 +18,11 @@ use tokio::sync::{broadcast, mpsc, watch};
 
 use crate::address::{Endpoint, PeerAddress, socket_text};
 use crate::collision::Direction;
+use crate::error::SessionError;
 use crate::family::FamilySet;
 use crate::fsm::Negotiated;
 use crate::metrics::{PEER_UP, SESSIONS_ESTABLISHED};
-use crate::policy::{ExportPolicy, PeerRole};
+use crate::policy::{ExportLimits, ExportPolicy, PeerRole};
 use crate::restart::{Agreed, Leave};
 use crate::rpki::Subject;
 use crate::sync::IgnorePoisonRw;
@@ -165,6 +166,8 @@ pub struct SessionHandle {
     pub counters: Arc<SessionCounters>,
     /// What the peer is told, and how.
     pub export: Arc<ExportPolicy>,
+    /// Limits on what the peer is told, where there are any for it.
+    pub export_limits: Option<Arc<ExportLimits>>,
     /// The groups the session is in, for control communities.
     pub groups: Arc<[u32]>,
     /// Whether the peer is a route reflector client of ours (RFC 4456).
@@ -172,6 +175,8 @@ pub struct SessionHandle {
     /// The peer's BGP Identifier.
     pub peer_router_id: Ipv4Addr,
     outbound: mpsc::Sender<Message>,
+    /// Where the session is asked to end, if it listens for that.
+    ending: Option<mpsc::UnboundedSender<SessionError>>,
     /// Which of the RPKI's checks the peer's routes are subject to, if
     /// any, and where the session is told to look at its routes again.
     validation: Option<(Option<Subject>, mpsc::UnboundedSender<Vec<Nlri>>)>,
@@ -209,14 +214,34 @@ impl SessionHandle {
             graceful_restart: negotiated.graceful_restart.clone(),
             counters: Arc::default(),
             export: Arc::default(),
+            export_limits: None,
             groups: Arc::default(),
             reflector_client: false,
             peer_router_id: negotiated.peer_identifier,
             outbound,
+            ending: None,
             validation: None,
         }
     }
 
+    /// Has the session listen for being asked to end: what is asked with
+    /// [`SessionHandle::end`] arrives at the receiver returned.
+    pub(crate) fn listen_for_end(&mut self) -> mpsc::UnboundedReceiver<SessionError> {
+        let (tx, rx) = mpsc::unbounded_channel();
+        self.ending = Some(tx);
+        rx
+    }
+
+    /// Asks the session to end because of `error`: it tells the peer with
+    /// the NOTIFICATION for it, and closes. What was queued before may go
+    /// out first; the caller queues nothing more.
+    pub(crate) fn end(&self, error: SessionError) {
+        if let Some(ending) = &self.ending {
+            // Fails only if the session has ended already.
+            let _ = ending.send(error);
+        }
+    }
+
     /// Marks the session's routes as subject to validation with the
     /// RPKI's data. What is sent with [`SessionHandle::revalidate`]
     /// arrives at the receiver returned.