Implement server-side rate limits for Key Images on FLR

fog/ledger/connection/src/lib.rs

@@ -20,4 +20,4 @@ mod untrusted;
 pub use untrusted::FogUntrustedLedgerGrpcClient;
 
 mod router_client;
-pub use router_client::LedgerGrpcClient;
+pub use router_client::{Error as RouterClientError, LedgerGrpcClient};

fog/ledger/connection/src/router_client.rs

@@ -47,7 +47,7 @@ pub struct LedgerGrpcClient {
     response_receiver: ClientDuplexReceiver<LedgerResponse>,
 
     /// Low-lever ledger API client
-    _client: LedgerApiClient,
+    client: LedgerApiClient,
 }
 
 impl LedgerGrpcClient {
@@ -76,14 +76,27 @@ impl LedgerGrpcClient {
         Self {
             logger,
             attest_cipher: None,
-            _client: client,
+            client,
             request_sender,
             response_receiver,
             uri,
             verifier,
         }
     }
 
+    /// Need ability to reconnect in case of rate limiting
+    pub fn reconnect(&mut self) {
+        self.deattest();
+
+        let (request_sender, response_receiver) = self
+            .client
+            .request()
+            .expect("Could not retrieve grpc sender and receiver.");
+
+        self.request_sender = request_sender;
+        self.response_receiver = response_receiver;
+    }
+
     fn is_attested(&self) -> bool {
         self.attest_cipher.is_some()
     }
@@ -136,7 +149,6 @@ impl LedgerGrpcClient {
         &mut self,
         key_images: &[KeyImage],
     ) -> Result<CheckKeyImagesResponse, Error> {
-        log::trace!(self.logger, "Check key images was called");
         if !self.is_attested() {
             let verification_report = self.attest().await;
             verification_report?;
@@ -202,7 +214,8 @@ impl LedgerGrpcClient {
 
 impl Drop for LedgerGrpcClient {
     fn drop(&mut self) {
-        block_on(self.request_sender.close()).expect("Couldn't close the router request sender");
+        // closing streams that have received an error result will fail, but that's OK
+        let _ = block_on(self.request_sender.close());
     }
 }
 

fog/ledger/server/Cargo.toml

@@ -55,6 +55,7 @@ mc-fog-uri = { path = "../../uri" }
 clap = { version = "4.1", features = ["derive", "env"] }
 displaydoc = { version = "0.2", default-features = false }
 futures = "0.3"
+governor = "0.5"
 grpcio = "0.12.1"
 itertools = "0.10"
 lazy_static = "1.4"

fog/ledger/server/src/config.rs

@@ -9,10 +9,10 @@ use clap::Parser;
 use mc_attest_core::ProviderId;
 use mc_common::ResponderId;
 use mc_fog_uri::{FogLedgerUri, KeyImageStoreUri};
-use mc_util_parse::parse_duration_in_seconds;
+use mc_util_parse::{parse_duration_in_milliseconds, parse_duration_in_seconds};
 use mc_util_uri::AdminUri;
 use serde::Serialize;
-use std::{path::PathBuf, str::FromStr, time::Duration};
+use std::{num::NonZeroU32, path::PathBuf, str::FromStr, time::Duration};
 
 /// Configuration parameters for the Fog Ledger Router service.
 #[derive(Clone, Parser, Serialize)]
@@ -65,6 +65,14 @@ pub struct LedgerRouterConfig {
     #[clap(long, default_value = "86400", value_parser = parse_duration_in_seconds, env = "MC_CLIENT_AUTH_TOKEN_MAX_LIFETIME")]
     pub client_auth_token_max_lifetime: Duration,
 
+    /// Rate limiting burst period, in milliseconds. Defaults to 10000 (10 sec)
+    #[clap(long, default_value = "10000", value_parser = parse_duration_in_milliseconds, env = "MC_RATE_LIMIT_BURST_PERIOD")]
+    pub rate_limit_burst_period: Duration,
+
+    /// Rate limiting maximum burst. Defaults to 80 requests.
+    #[clap(long, default_value = "80", env = "MC_RATE_LIMIT_MAX_BURST")]
+    pub rate_limit_max_burst: NonZeroU32,
+
     /// Path to ledger db (lmdb)
     #[clap(long, env = "MC_LEDGER_DB")]
     pub ledger_db: PathBuf,

fog/ledger/server/src/router_handlers.rs

@@ -5,6 +5,11 @@ use crate::{
     SVC_COUNTERS,
 };
 use futures::{future::try_join_all, SinkExt, TryStreamExt};
+use governor::{
+    clock::{Clock, DefaultClock},
+    state::keyed::DefaultKeyedStateStore,
+    RateLimiter,
+};
 use grpcio::{ChannelBuilder, DuplexSink, RequestStream, RpcStatus, WriteFlags};
 use mc_attest_api::attest;
 use mc_attest_enclave_api::{EnclaveMessage, NonceSession};
@@ -21,7 +26,9 @@ use mc_fog_api::{
 };
 use mc_fog_ledger_enclave::LedgerEnclaveProxy;
 use mc_fog_uri::{ConnectionUri, KeyImageStoreUri};
-use mc_util_grpc::{rpc_invalid_arg_error, ConnectionUriGrpcioChannel, ResponseStatus};
+use mc_util_grpc::{
+    rpc_invalid_arg_error, rpc_resource_exhausted_error, ConnectionUriGrpcioChannel, ResponseStatus,
+};
 use mc_util_metrics::GrpcMethodName;
 use mc_util_telemetry::{create_context, tracer, BoxedTracer, FutureExt, Tracer};
 use std::{collections::BTreeMap, str::FromStr, sync::Arc};
@@ -35,6 +42,7 @@ pub async fn handle_requests<E>(
     mut requests: RequestStream<LedgerRequest>,
     mut responses: DuplexSink<LedgerResponse>,
     query_retries: usize,
+    rate_limit_context: Arc<RateLimiter<Vec<u8>, DefaultKeyedStateStore<Vec<u8>>, DefaultClock>>,
     logger: Logger,
 ) -> Result<(), grpcio::Error>
 where
@@ -52,6 +60,7 @@ where
             shard_clients.clone(),
             enclave.clone(),
             query_retries,
+            rate_limit_context.clone(),
             logger.clone(),
         )
         .await;
@@ -76,6 +85,7 @@ pub async fn handle_request<E>(
     shard_clients: Vec<Arc<KeyImageStoreApiClient>>,
     enclave: E,
     query_retries: usize,
+    rate_limit_context: Arc<RateLimiter<Vec<u8>, DefaultKeyedStateStore<Vec<u8>>, DefaultClock>>,
     logger: Logger,
 ) -> Result<LedgerResponse, RpcStatus>
 where
@@ -87,16 +97,33 @@ where
             tracer.in_span("auth", |_cx| handle_auth_request(enclave, request, logger))
         }
         Some(LedgerRequest_oneof_request_data::check_key_images(request)) => {
-            handle_query_request(
-                request,
-                enclave,
-                shard_clients,
-                query_retries,
-                logger,
-                &tracer,
-            )
-            .with_context(create_context(&tracer, "check_key_images"))
-            .await
+            match rate_limit_context.check_key(&request.channel_id) {
+                Ok(()) => {
+                    handle_query_request(
+                        request,
+                        enclave,
+                        shard_clients,
+                        query_retries,
+                        logger,
+                        &tracer,
+                    )
+                    .with_context(create_context(&tracer, "check_key_images"))
+                    .await
+                }
+                Err(not_until) => {
+                    let rpc_status = rpc_resource_exhausted_error(
+                        "Key image rate limit exceeded",
+                        format!(
+                            "Try again in {} milliseconds",
+                            not_until
+                                .wait_time_from(DefaultClock::default().now())
+                                .as_millis()
+                        ),
+                        &logger,
+                    );
+                    Err(rpc_status)
+                }
+            }
         }
         None => {
             let rpc_status = rpc_invalid_arg_error(

fog/ledger/server/src/router_server.rs

@@ -99,6 +99,8 @@ where
             enclave.clone(),
             ledger_store_grpc_clients.clone(),
             config.query_retries,
+            config.rate_limit_burst_period,
+            config.rate_limit_max_burst,
             logger.clone(),
         );