Delay cosign acknowledgments

coordinator/cosign/README.md

@@ -66,7 +66,7 @@ to exhibit the same behavior), yet prevents interaction with it.
 
 If cosigns representing 17% of the non-Serai validators sets by weight are
 detected for distinct blocks at the same position, the protocol halts. An
-explicit latency period of five seconds is enacted after receiving a cosign
+explicit latency period of seventy seconds is enacted after receiving a cosign
 commit for the detection of such an equivocation. This is largely redundant
 given how the Serai blockchain node will presumably have halted itself by this
 time.
@@ -114,8 +114,8 @@ asynchronous network or 11.33% of non-Serai validator sets' stake.
 ### TODO
 
 The Serai node no longer responding to RPC requests upon detecting any
-equivocation, the delayed acknowledgement of cosigns, and the fallback protocol
-where validators individually produce signatures, are not implemented at this
-time. The former means the detection of equivocating cosigns not redundant and
-the latter makes 5.67% of non-Serai validator sets' stake the DoS threshold,
-even without control of an asynchronous network.
+equivocation, and the fallback protocol where validators individually produce
+signatures, are not implemented at this time. The former means the detection of
+equivocating cosigns is not redundant and the latter makes 5.67% of non-Serai
+validator sets' stake the DoS threshold, even without control of an
+asynchronous network.

coordinator/cosign/src/delay.rs

@@ -0,0 +1,55 @@
+use core::future::Future;
+use std::time::{Duration, SystemTime};
+
+use serai_db::*;
+use serai_task::ContinuallyRan;
+
+use crate::evaluator::CosignedBlocks;
+
+/// How often callers should broadcast the cosigns flagged for rebroadcasting.
+pub const BROADCAST_FREQUENCY: Duration = Duration::from_secs(60);
+const SYNCHRONY_EXPECTATION: Duration = Duration::from_secs(10);
+const ACKNOWLEDGEMENT_DELAY: Duration =
+  Duration::from_secs(BROADCAST_FREQUENCY.as_secs() + SYNCHRONY_EXPECTATION.as_secs());
+
+create_db!(
+  SubstrateCosignDelay {
+    // The latest cosigned block number.
+    LatestCosignedBlockNumber: () -> u64,
+  }
+);
+
+/// A task to delay acknowledgement of cosigns.
+pub(crate) struct CosignDelayTask<D: Db> {
+  pub(crate) db: D,
+}
+
+impl<D: Db> ContinuallyRan for CosignDelayTask<D> {
+  fn run_iteration(&mut self) -> impl Send + Future<Output = Result<bool, String>> {
+    async move {
+      let mut made_progress = false;
+      loop {
+        let mut txn = self.db.txn();
+
+        // Receive the next block to mark as cosigned
+        let Some((block_number, time_evaluated)) = CosignedBlocks::try_recv(&mut txn) else {
+          break;
+        };
+        // Calculate when we should mark it as valid
+        let time_valid =
+          SystemTime::UNIX_EPOCH + Duration::from_secs(time_evaluated) + ACKNOWLEDGEMENT_DELAY;
+        // Sleep until then
+        tokio::time::sleep(SystemTime::now().duration_since(time_valid).unwrap_or(Duration::ZERO))
+          .await;
+
+        // Set the cosigned block
+        LatestCosignedBlockNumber::set(&mut txn, &block_number);
+        txn.commit();
+
+        made_progress = true;
+      }
+
+      Ok(made_progress)
+    }
+  }
+}

coordinator/cosign/src/evaluator.rs

@@ -1,4 +1,5 @@
 use core::future::Future;
+use std::time::{Duration, SystemTime};
 
 use serai_db::*;
 use serai_task::ContinuallyRan;
@@ -10,13 +11,18 @@ use crate::{
 
 create_db!(
   SubstrateCosignEvaluator {
-    // The latest cosigned block number.
-    LatestCosignedBlockNumber: () -> u64,
     // The global session currently being evaluated.
     CurrentlyEvaluatedGlobalSession: () -> ([u8; 32], GlobalSession),
   }
 );
 
+db_channel!(
+  SubstrateCosignEvaluatorChannels {
+    // (cosigned block, time cosign was evaluated)
+    CosignedBlocks: () -> (u64, u64),
+  }
+);
+
 // This is a strict function which won't panic, even with a malicious Serai node, so long as:
 // - It's called incrementally
 // - It's only called for block numbers we've completed indexing on within the intend task
@@ -72,8 +78,6 @@ pub(crate) struct CosignEvaluatorTask<D: Db, R: RequestNotableCosigns> {
 impl<D: Db, R: RequestNotableCosigns> ContinuallyRan for CosignEvaluatorTask<D, R> {
   fn run_iteration(&mut self) -> impl Send + Future<Output = Result<bool, String>> {
     async move {
-      let latest_cosigned_block_number = LatestCosignedBlockNumber::get(&self.db).unwrap_or(0);
-
       let mut known_cosign = None;
       let mut made_progress = false;
       loop {
@@ -82,11 +86,6 @@ impl<D: Db, R: RequestNotableCosigns> ContinuallyRan for CosignEvaluatorTask<D,
         else {
           break;
         };
-        // Make sure these two feeds haven't desynchronized somehow
-        // We could remove our `LatestCosignedBlockNumber`, making the latest cosigned block number
-        // the next message in the channel's block number minus one, but that'd only work when the
-        // channel isn't empty
-        assert_eq!(block_number, latest_cosigned_block_number + 1);
 
         match has_events {
           // Because this had notable events, we require an explicit cosign for this block by a
@@ -201,8 +200,17 @@ impl<D: Db, R: RequestNotableCosigns> ContinuallyRan for CosignEvaluatorTask<D,
           HasEvents::No => {}
         }
 
-        // Since we checked we had the necessary cosigns, increment the latest cosigned block
-        LatestCosignedBlockNumber::set(&mut txn, &block_number);
+        // Since we checked we had the necessary cosigns, send it for delay before acknowledgement
+        CosignedBlocks::send(
+          &mut txn,
+          &(
+            block_number,
+            SystemTime::now()
+              .duration_since(SystemTime::UNIX_EPOCH)
+              .unwrap_or(Duration::ZERO)
+              .as_secs(),
+          ),
+        );
         txn.commit();
 
         made_progress = true;

coordinator/cosign/src/lib.rs

@@ -22,7 +22,10 @@ use serai_task::*;
 mod intend;
 /// The evaluator of the cosigns.
 mod evaluator;
-use evaluator::LatestCosignedBlockNumber;
+/// The task to delay acknowledgement of the cosigns.
+mod delay;
+pub use delay::BROADCAST_FREQUENCY;
+use delay::LatestCosignedBlockNumber;
 
 /// The schnorrkel context to used when signing a cosign.
 pub const COSIGN_CONTEXT: &[u8] = b"serai-cosign";
@@ -235,13 +238,18 @@ impl<D: Db> Cosigning<D> {
   ) -> Self {
     let (intend_task, _intend_task_handle) = Task::new();
     let (evaluator_task, evaluator_task_handle) = Task::new();
+    let (delay_task, delay_task_handle) = Task::new();
     tokio::spawn(
       (intend::CosignIntendTask { db: db.clone(), serai })
         .continually_run(intend_task, vec![evaluator_task_handle]),
     );
     tokio::spawn(
       (evaluator::CosignEvaluatorTask { db: db.clone(), request })
-        .continually_run(evaluator_task, tasks_to_run_upon_cosigning),
+        .continually_run(evaluator_task, vec![delay_task_handle]),
+    );
+    tokio::spawn(
+      (delay::CosignDelayTask { db: db.clone() })
+        .continually_run(delay_task, tasks_to_run_upon_cosigning),
     );
     Self { db }
   }
@@ -269,7 +277,7 @@ impl<D: Db> Cosigning<D> {
     cosigns
   }
 
-  /// The cosigns to rebroadcast ever so often.
+  /// The cosigns to rebroadcast every `BROADCAST_FREQUENCY` seconds.
   ///
   /// This will be the most recent cosigns, in case the initial broadcast failed, or the faulty
   /// cosigns, in case of a fault, to induce identification of the fault by others.
@@ -348,6 +356,8 @@ impl<D: Db> Cosigning<D> {
       return Ok(false);
     }
     if !faulty {
+      // This prevents a malicious validator set, on the same chain, from producing a cosign after
+      // their final block, replacing their notable cosign
       if let Some(last_block) = GlobalSessionsLastBlock::get(&self.db, cosign.global_session) {
         if cosign.block_number > last_block {
           // Cosign is for a block after the last block this global session should have signed