Merge pull request #3422 from zkxuerb/feat/banned-peers

Spam connections and forged block locators fix

node/bft/src/gateway.rs

@@ -65,6 +65,8 @@ use futures::SinkExt;
 use indexmap::{IndexMap, IndexSet};
 use parking_lot::{Mutex, RwLock};
 use rand::seq::{IteratorRandom, SliceRandom};
+#[cfg(not(any(test, feature = "test")))]
+use std::net::IpAddr;
 use std::{collections::HashSet, future::Future, io, net::SocketAddr, sync::Arc, time::Duration};
 use tokio::{
     net::TcpStream,
@@ -89,6 +91,12 @@ const MIN_CONNECTED_VALIDATORS: usize = 175;
 /// The maximum number of validators to send in a validators response event.
 const MAX_VALIDATORS_TO_SEND: usize = 200;
 
+/// The minimum permitted interval between connection attempts for an IP; anything shorter is considered malicious.
+#[cfg(not(any(test, feature = "test")))]
+const CONNECTION_ATTEMPTS_SINCE_SECS: i64 = 10;
+/// The amount of time an IP address is prohibited from connecting.
+const IP_BAN_TIME_IN_SECS: u64 = 300;
+
 /// Part of the Gateway API that deals with networking.
 /// This is a separate trait to allow for easier testing/mocking.
 #[async_trait]
@@ -460,6 +468,18 @@ impl<N: Network> Gateway<N> {
         Ok(())
     }
 
+    /// Check whether the given IP address is currently banned.
+    #[cfg(not(any(test, feature = "test")))]
+    fn is_ip_banned(&self, ip: IpAddr) -> bool {
+        self.tcp.banned_peers().is_ip_banned(&ip)
+    }
+
+    /// Insert or update a banned IP.
+    #[cfg(not(any(test, feature = "test")))]
+    fn update_ip_ban(&self, ip: IpAddr) {
+        self.tcp.banned_peers().update_ip_ban(ip);
+    }
+
     #[cfg(feature = "metrics")]
     fn update_metrics(&self) {
         metrics::gauge(metrics::bft::CONNECTED, self.connected_peers.read().len() as f64);
@@ -885,6 +905,8 @@ impl<N: Network> Gateway<N> {
         self.handle_unauthorized_validators();
         // If the number of connected validators is less than the minimum, send a `ValidatorsRequest`.
         self.handle_min_connected_validators();
+        // Unban any addresses whose ban time has expired.
+        self.handle_banned_ips();
     }
 
     /// Logs the connected validators.
@@ -965,6 +987,11 @@ impl<N: Network> Gateway<N> {
             }
         }
     }
+
+    // Remove addresses whose ban time has expired.
+    fn handle_banned_ips(&self) {
+        self.tcp.banned_peers().remove_old_bans(IP_BAN_TIME_IN_SECS);
+    }
 }
 
 #[async_trait]
@@ -1125,6 +1152,26 @@ impl<N: Network> Handshake for Gateway<N> {
         // Perform the handshake.
         let peer_addr = connection.addr();
         let peer_side = connection.side();
+
+        // Check (or impose) IP-level bans.
+        #[cfg(not(any(test, feature = "test")))]
+        if self.dev().is_none() && peer_side == ConnectionSide::Initiator {
+            // If the IP is already banned reject the connection.
+            if self.is_ip_banned(peer_addr.ip()) {
+                trace!("{CONTEXT} Gateway rejected a connection request from banned IP '{}'", peer_addr.ip());
+                return Err(error(format!("'{}' is a banned IP address", peer_addr.ip())));
+            }
+
+            let num_attempts = self.cache.insert_inbound_connection(peer_addr.ip(), CONNECTION_ATTEMPTS_SINCE_SECS);
+
+            debug!("Number of connection attempts from '{}': {}", peer_addr.ip(), num_attempts);
+            if num_attempts > MAX_CONNECTION_ATTEMPTS {
+                self.update_ip_ban(peer_addr.ip());
+                trace!("{CONTEXT} Gateway rejected a consecutive connection request from IP '{}'", peer_addr.ip());
+                return Err(error(format!("'{}' appears to be spamming connections", peer_addr.ip())));
+            }
+        }
+
         let stream = self.borrow_stream(&mut connection);
 
         // If this is an inbound connection, we log it, but don't know the listening address yet.

node/bft/src/sync/mod.rs

@@ -24,6 +24,7 @@ use crate::{
 use snarkos_node_bft_events::{CertificateRequest, CertificateResponse, Event};
 use snarkos_node_bft_ledger_service::LedgerService;
 use snarkos_node_sync::{BlockSync, BlockSyncMode, locators::BlockLocators};
+use snarkos_node_tcp::P2P;
 use snarkvm::{
     console::{network::Network, types::Field},
     ledger::{authority::Authority, block::Block, narwhal::BatchCertificate},
@@ -67,7 +68,7 @@ impl<N: Network> Sync<N> {
     /// Initializes a new sync instance.
     pub fn new(gateway: Gateway<N>, storage: Storage<N>, ledger: Arc<dyn LedgerService<N>>) -> Self {
         // Initialize the block sync module.
-        let block_sync = BlockSync::new(BlockSyncMode::Gateway, ledger.clone());
+        let block_sync = BlockSync::new(BlockSyncMode::Gateway, ledger.clone(), gateway.tcp().clone());
         // Return the sync instance.
         Self {
             gateway,

node/router/src/handshake.rs

@@ -101,6 +101,26 @@ impl<N: Network> Router<N> {
             Some(peer_addr)
         };
 
+        // Check (or impose) IP-level bans.
+        #[cfg(not(any(test, feature = "test")))]
+        if !self.is_dev() && peer_side == ConnectionSide::Initiator {
+            // If the IP is already banned reject the connection.
+            if self.is_ip_banned(peer_addr.ip()) {
+                trace!("Rejected a connection request from banned IP '{}'", peer_addr.ip());
+                return Err(error(format!("'{}' is a banned IP address", peer_addr.ip())));
+            }
+
+            let num_attempts =
+                self.cache.insert_inbound_connection(peer_addr.ip(), Router::<N>::CONNECTION_ATTEMPTS_SINCE_SECS);
+
+            debug!("Number of connection attempts from '{}': {}", peer_addr.ip(), num_attempts);
+            if num_attempts > Router::<N>::MAX_CONNECTION_ATTEMPTS {
+                self.update_ip_ban(peer_addr.ip());
+                trace!("Rejected a consecutive connection request from IP '{}'", peer_addr.ip());
+                return Err(error(format!("'{}' appears to be spamming connections", peer_addr.ip())));
+            }
+        }
+
         // Perform the handshake; we pass on a mutable reference to peer_ip in case the process is broken at any point in time.
         let handshake_result = if peer_side == ConnectionSide::Responder {
             self.handshake_inner_initiator(peer_addr, &mut peer_ip, stream, genesis_header, restrictions_id).await

node/router/src/heartbeat.rs

@@ -43,6 +43,8 @@ pub trait Heartbeat<N: Network>: Outbound<N> {
     const MAXIMUM_NUMBER_OF_PEERS: usize = 21;
     /// The maximum number of provers to maintain connections with.
     const MAXIMUM_NUMBER_OF_PROVERS: usize = Self::MAXIMUM_NUMBER_OF_PEERS / 4;
+    /// The amount of time an IP address is prohibited from connecting.
+    const IP_BAN_TIME_IN_SECS: u64 = 300;
 
     /// Handles the heartbeat request.
     fn heartbeat(&self) {
@@ -61,6 +63,8 @@ pub trait Heartbeat<N: Network>: Outbound<N> {
         self.handle_trusted_peers();
         // Keep the puzzle request up to date.
         self.handle_puzzle_request();
+        // Unban any addresses whose ban time has expired.
+        self.handle_banned_ips();
     }
 
     /// TODO (howardwu): Consider checking minimum number of validators, to exclude clients and provers.
@@ -238,6 +242,7 @@ pub trait Heartbeat<N: Network>: Outbound<N> {
             for peer_ip in self.router().candidate_peers().into_iter().choose_multiple(rng, num_deficient) {
                 self.router().connect(peer_ip);
             }
+
             if self.router().allow_external_peers() {
                 // Request more peers from the connected peers.
                 for peer_ip in self.router().connected_peers().into_iter().choose_multiple(rng, 3) {
@@ -298,4 +303,9 @@ pub trait Heartbeat<N: Network>: Outbound<N> {
     fn handle_puzzle_request(&self) {
         // No-op
     }
+
+    // Remove addresses whose ban time has expired.
+    fn handle_banned_ips(&self) {
+        self.tcp().banned_peers().remove_old_bans(Self::IP_BAN_TIME_IN_SECS);
+    }
 }

node/router/src/lib.rs

@@ -41,11 +41,14 @@ pub use routing::*;
 
 use crate::messages::NodeType;
 use snarkos_account::Account;
-use snarkos_node_tcp::{Config, Tcp, is_bogon_ip, is_unspecified_or_broadcast_ip};
+use snarkos_node_tcp::{Config, P2P, Tcp, is_bogon_ip, is_unspecified_or_broadcast_ip};
+
 use snarkvm::prelude::{Address, Network, PrivateKey, ViewKey};
 
 use anyhow::{Result, bail};
 use parking_lot::{Mutex, RwLock};
+#[cfg(not(any(test, feature = "test")))]
+use std::net::IpAddr;
 use std::{
     collections::{HashMap, HashSet, hash_map::Entry},
     future::Future,
@@ -103,10 +106,16 @@ pub struct InnerRouter<N: Network> {
 }
 
 impl<N: Network> Router<N> {
+    /// The minimum permitted interval between connection attempts for an IP; anything shorter is considered malicious.
+    #[cfg(not(any(test, feature = "test")))]
+    const CONNECTION_ATTEMPTS_SINCE_SECS: i64 = 10;
     /// The maximum number of candidate peers permitted to be stored in the node.
     const MAXIMUM_CANDIDATE_PEERS: usize = 10_000;
     /// The maximum number of connection failures permitted by an inbound connecting peer.
     const MAXIMUM_CONNECTION_FAILURES: usize = 5;
+    /// The maximum amount of connection attempts withing a 10 second threshold
+    #[cfg(not(any(test, feature = "test")))]
+    const MAX_CONNECTION_ATTEMPTS: usize = 10;
     /// The duration in seconds after which a connected peer is considered inactive or
     /// disconnected if no message has been received in the meantime.
     const RADIO_SILENCE_IN_SECS: u64 = 150; // 2.5 minutes
@@ -390,7 +399,8 @@ impl<N: Network> Router<N> {
 
     /// Returns the list of candidate peers.
     pub fn candidate_peers(&self) -> HashSet<SocketAddr> {
-        self.candidate_peers.read().clone()
+        let banned_ips = self.tcp().banned_peers().get_banned_ips();
+        self.candidate_peers.read().iter().filter(|peer| !banned_ips.contains(&peer.ip())).copied().collect()
     }
 
     /// Returns the list of restricted peers.
@@ -439,6 +449,18 @@ impl<N: Network> Router<N> {
         }
     }
 
+    /// Check whether the given IP address is currently banned.
+    #[cfg(not(any(test, feature = "test")))]
+    fn is_ip_banned(&self, ip: IpAddr) -> bool {
+        self.tcp.banned_peers().is_ip_banned(&ip)
+    }
+
+    /// Insert or update a banned IP.
+    #[cfg(not(any(test, feature = "test")))]
+    fn update_ip_ban(&self, ip: IpAddr) {
+        self.tcp.banned_peers().update_ip_ban(ip);
+    }
+
     /// Returns the list of metrics for the connected peers.
     pub fn connected_metrics(&self) -> Vec<(SocketAddr, NodeType)> {
         self.connected_peers.read().iter().map(|(ip, peer)| (*ip, peer.node_type())).collect()

node/src/client/mod.rs

@@ -106,8 +106,6 @@ impl<N: Network, C: ConsensusStorage<N>> Client<N, C> {
 
         // Initialize the ledger service.
         let ledger_service = Arc::new(CoreLedgerService::<N, C>::new(ledger.clone(), shutdown.clone()));
-        // Initialize the sync module.
-        let sync = BlockSync::new(BlockSyncMode::Router, ledger_service.clone());
         // Determine if the client should allow external peers.
         let allow_external_peers = true;
 
@@ -123,6 +121,10 @@ impl<N: Network, C: ConsensusStorage<N>> Client<N, C> {
             matches!(storage_mode, StorageMode::Development(_)),
         )
         .await?;
+
+        // Initialize the sync module.
+        let sync = BlockSync::new(BlockSyncMode::Router, ledger_service.clone(), router.tcp().clone());
+
         // Initialize the node.
         let mut node = Self {
             ledger: ledger.clone(),

node/src/prover/mod.rs

@@ -100,8 +100,6 @@ impl<N: Network, C: ConsensusStorage<N>> Prover<N, C> {
 
         // Initialize the ledger service.
         let ledger_service = Arc::new(ProverLedgerService::new());
-        // Initialize the sync module.
-        let sync = BlockSync::new(BlockSyncMode::Router, ledger_service.clone());
         // Determine if the prover should allow external peers.
         let allow_external_peers = true;
         // Determine if the prover should rotate external peers.
@@ -119,6 +117,10 @@ impl<N: Network, C: ConsensusStorage<N>> Prover<N, C> {
             matches!(storage_mode, StorageMode::Development(_)),
         )
         .await?;
+
+        // Initialize the sync module.
+        let sync = BlockSync::new(BlockSyncMode::Router, ledger_service.clone(), router.tcp().clone());
+
         // Compute the maximum number of puzzle instances.
         let max_puzzle_instances = num_cpus::get().saturating_sub(2).clamp(1, 6);
         // Initialize the node.

node/src/validator/mod.rs

@@ -107,12 +107,10 @@ impl<N: Network, C: ConsensusStorage<N>> Validator<N, C> {
 
         // Initialize the ledger service.
         let ledger_service = Arc::new(CoreLedgerService::new(ledger.clone(), shutdown.clone()));
-        // Initialize the sync module.
-        let sync = BlockSync::new(BlockSyncMode::Gateway, ledger_service.clone());
 
         // Initialize the consensus.
         let mut consensus =
-            Consensus::new(account.clone(), ledger_service, bft_ip, trusted_validators, storage_mode.clone())?;
+            Consensus::new(account.clone(), ledger_service.clone(), bft_ip, trusted_validators, storage_mode.clone())?;
         // Initialize the primary channels.
         let (primary_sender, primary_receiver) = init_primary_channels::<N>();
         // Start the consensus.
@@ -133,6 +131,9 @@ impl<N: Network, C: ConsensusStorage<N>> Validator<N, C> {
         )
         .await?;
 
+        // Initialize the sync module.
+        let sync = BlockSync::new(BlockSyncMode::Gateway, ledger_service, router.tcp().clone());
+
         // Initialize the node.
         let mut node = Self {
             ledger: ledger.clone(),

node/sync/Cargo.toml

@@ -67,6 +67,10 @@ version = "=3.0.0"
 path = "locators"
 version = "=3.0.0"
 
+[dependencies.snarkos-node-tcp]
+path = "../tcp"
+version = "=3.0.0"
+
 [dependencies.snarkvm]
 workspace = true