Lean on the holding cell when batch-forwarding/failing HTLCs

TheBlueMatt · TheBlueMatt · commit 3395a94146c2 · 2022-11-30T03:56:22.000Z
When we batch HTLC updates, we currently do the explicit queueing
plus the commitment generation in the `ChannelManager`. This is a
bit strange as its ultimately really a `Channel` responsibility to
generate commitments at the correct time, with the abstraction
leaking into `ChannelManager` with the `send_htlc` and
`get_update_fail_htlc` method docs having clear comments about
how `send_commitment` MUST be called prior to calling other
`Channel` methods.

Luckily `Channel` already has an update queue - the holding cell.
Thus, we can trivially rewrite the batch update logic as inserting
the desired updates into the holding cell and then asking all
channels to clear their holding cells.
diff --git a/lightning/src/ln/channel.rs b/lightning/src/ln/channel.rs
@@ -1948,7 +1948,20 @@ impl<Signer: Sign> Channel<Signer> {
 	/// before we fail backwards.
 	/// If we do fail twice, we debug_assert!(false) and return Ok(None). Thus, will always return
 	/// Ok(_) if debug assertions are turned on or preconditions are met.
-	pub fn get_update_fail_htlc<L: Deref>(&mut self, htlc_id_arg: u64, err_packet: msgs::OnionErrorPacket, logger: &L) -> Result<Option<msgs::UpdateFailHTLC>, ChannelError> where L::Target: Logger {
+	pub fn queue_fail_htlc<L: Deref>(&mut self, htlc_id_arg: u64, err_packet: msgs::OnionErrorPacket, logger: &L)
+	-> Result<(), ChannelError> where L::Target: Logger {
+		self.fail_htlc(htlc_id_arg, err_packet, true, logger)
+			.map(|msg_opt| assert!(msg_opt.is_none(), "We forced holding cell?"))
+	}
+
+	/// We can only have one resolution per HTLC. In some cases around reconnect, we may fulfill
+	/// an HTLC more than once or fulfill once and then attempt to fail after reconnect. We cannot,
+	/// however, fail more than once as we wait for an upstream failure to be irrevocably committed
+	/// before we fail backwards.
+	/// If we do fail twice, we debug_assert!(false) and return Ok(None). Thus, will always return
+	/// Ok(_) if debug assertions are turned on or preconditions are met.
+	fn fail_htlc<L: Deref>(&mut self, htlc_id_arg: u64, err_packet: msgs::OnionErrorPacket, mut force_holding_cell: bool, logger: &L)
+	-> Result<Option<msgs::UpdateFailHTLC>, ChannelError> where L::Target: Logger {
 		if (self.channel_state & (ChannelState::ChannelReady as u32)) != (ChannelState::ChannelReady as u32) {
 			panic!("Was asked to fail an HTLC when channel was not in an operational state");
 		}
@@ -1986,8 +1999,13 @@ impl<Signer: Sign> Channel<Signer> {
 			return Ok(None);
 		}
 
-		// Now update local state:
 		if (self.channel_state & (ChannelState::AwaitingRemoteRevoke as u32 | ChannelState::PeerDisconnected as u32 | ChannelState::MonitorUpdateInProgress as u32)) != 0 {
+			debug_assert!(force_holding_cell, "We don't expect to need to use the holding cell if we weren't trying to");
+			force_holding_cell = true;
+		}
+
+		// Now update local state:
+		if force_holding_cell {
 			for pending_update in self.holding_cell_htlc_updates.iter() {
 				match pending_update {
 					&HTLCUpdateAwaitingACK::ClaimHTLC { htlc_id, .. } => {
@@ -3173,7 +3191,7 @@ impl<Signer: Sign> Channel<Signer> {
 				// to rebalance channels.
 				match &htlc_update {
 					&HTLCUpdateAwaitingACK::AddHTLC {amount_msat, cltv_expiry, ref payment_hash, ref source, ref onion_routing_packet, ..} => {
-						match self.send_htlc(amount_msat, *payment_hash, cltv_expiry, source.clone(), onion_routing_packet.clone(), logger) {
+						match self.send_htlc(amount_msat, *payment_hash, cltv_expiry, source.clone(), onion_routing_packet.clone(), false, logger) {
 							Ok(update_add_msg_option) => update_add_htlcs.push(update_add_msg_option.unwrap()),
 							Err(e) => {
 								match e {
@@ -3209,13 +3227,13 @@ impl<Signer: Sign> Channel<Signer> {
 						monitor_update.updates.append(&mut additional_monitor_update.updates);
 					},
 					&HTLCUpdateAwaitingACK::FailHTLC { htlc_id, ref err_packet } => {
-						match self.get_update_fail_htlc(htlc_id, err_packet.clone(), logger) {
+						match self.fail_htlc(htlc_id, err_packet.clone(), false, logger) {
 							Ok(update_fail_msg_option) => {
 								// If an HTLC failure was previously added to the holding cell (via
-								// `get_update_fail_htlc`) then generating the fail message itself
-								// must not fail - we should never end up in a state where we
-								// double-fail an HTLC or fail-then-claim an HTLC as it indicates
-								// we didn't wait for a full revocation before failing.
+								// `fail_htlc`) then generating the fail message itself must not
+								// fail - we should never end up in a state where we double-fail an
+								// HTLC or fail-then-claim an HTLC as it indicates we didn't wait
+								// for a full revocation before failing.
 								update_fail_htlcs.push(update_fail_msg_option.unwrap())
 							},
 							Err(e) => {
@@ -5470,6 +5488,19 @@ impl<Signer: Sign> Channel<Signer> {
 
 	// Send stuff to our remote peers:
 
+	/// Queues up an outbound HTLC to send by placing it in the holding cell. You should call
+	/// `maybe_free_holding_cell_htlcs` in order to actually generate and send the commitment
+	/// update.
+	///
+	/// If an Err is returned, it's a ChannelError::Ignore!
+	pub fn queue_htlc<L: Deref>(&mut self, amount_msat: u64, payment_hash: PaymentHash, cltv_expiry: u32, source: HTLCSource,
+		onion_routing_packet: msgs::OnionPacket, logger: &L)
+	-> Result<(), ChannelError> where L::Target: Logger {
+		self.send_htlc(amount_msat, payment_hash, cltv_expiry, source, onion_routing_packet, true, logger)
+			.map(|msg_opt| assert!(msg_opt.is_none(), "We forced holding cell?"))
+
+	}
+
 	/// Adds a pending outbound HTLC to this channel, note that you probably want
 	/// send_htlc_and_commit instead cause you'll want both messages at once.
 	///
@@ -5485,7 +5516,9 @@ impl<Signer: Sign> Channel<Signer> {
 	/// You MUST call send_commitment prior to calling any other methods on this Channel!
 	///
 	/// If an Err is returned, it's a ChannelError::Ignore!
-	pub fn send_htlc<L: Deref>(&mut self, amount_msat: u64, payment_hash: PaymentHash, cltv_expiry: u32, source: HTLCSource, onion_routing_packet: msgs::OnionPacket, logger: &L) -> Result<Option<msgs::UpdateAddHTLC>, ChannelError> where L::Target: Logger {
+	fn send_htlc<L: Deref>(&mut self, amount_msat: u64, payment_hash: PaymentHash, cltv_expiry: u32, source: HTLCSource,
+		onion_routing_packet: msgs::OnionPacket, mut force_holding_cell: bool, logger: &L)
+	-> Result<Option<msgs::UpdateAddHTLC>, ChannelError> where L::Target: Logger {
 		if (self.channel_state & (ChannelState::ChannelReady as u32 | BOTH_SIDES_SHUTDOWN_MASK)) != (ChannelState::ChannelReady as u32) {
 			return Err(ChannelError::Ignore("Cannot send HTLC until channel is fully established and we haven't started shutting down".to_owned()));
 		}
@@ -5580,8 +5613,12 @@ impl<Signer: Sign> Channel<Signer> {
 			return Err(ChannelError::Ignore(format!("Cannot send value that would put our balance under counterparty-announced channel reserve value ({})", chan_reserve_msat)));
 		}
 
-		// Now update local state:
 		if (self.channel_state & (ChannelState::AwaitingRemoteRevoke as u32 | ChannelState::MonitorUpdateInProgress as u32)) != 0 {
+			force_holding_cell = true;
+		}
+
+		// Now update local state:
+		if force_holding_cell {
 			self.holding_cell_htlc_updates.push(HTLCUpdateAwaitingACK::AddHTLC {
 				amount_msat,
 				payment_hash,
@@ -5774,7 +5811,7 @@ impl<Signer: Sign> Channel<Signer> {
 	/// Shorthand for calling send_htlc() followed by send_commitment(), see docs on those for
 	/// more info.
 	pub fn send_htlc_and_commit<L: Deref>(&mut self, amount_msat: u64, payment_hash: PaymentHash, cltv_expiry: u32, source: HTLCSource, onion_routing_packet: msgs::OnionPacket, logger: &L) -> Result<Option<(msgs::UpdateAddHTLC, msgs::CommitmentSigned, ChannelMonitorUpdate)>, ChannelError> where L::Target: Logger {
-		match self.send_htlc(amount_msat, payment_hash, cltv_expiry, source, onion_routing_packet, logger)? {
+		match self.send_htlc(amount_msat, payment_hash, cltv_expiry, source, onion_routing_packet, false, logger)? {
 			Some(update_add_htlc) => {
 				let (commitment_signed, monitor_update) = self.send_commitment_no_status_check(logger)?;
 				Ok(Some((update_add_htlc, commitment_signed, monitor_update)))
diff --git a/lightning/src/ln/channelmanager.rs b/lightning/src/ln/channelmanager.rs
@@ -3033,7 +3033,6 @@ impl<M: Deref, T: Deref, K: Deref, F: Deref, L: Deref> ChannelManager<M, T, K, F
 		let mut new_events = Vec::new();
 		let mut failed_forwards = Vec::new();
 		let mut phantom_receives: Vec<(u64, OutPoint, u128, Vec<(PendingHTLCInfo, u64)>)> = Vec::new();
-		let mut handle_errors = Vec::new();
 		{
 			let mut forward_htlcs = HashMap::new();
 			mem::swap(&mut forward_htlcs, &mut self.forward_htlcs.lock().unwrap());
@@ -3149,8 +3148,6 @@ impl<M: Deref, T: Deref, K: Deref, F: Deref, L: Deref> ChannelManager<M, T, K, F
 							continue;
 						},
 						hash_map::Entry::Occupied(mut chan) => {
-							let mut add_htlc_msgs = Vec::new();
-							let mut fail_htlc_msgs = Vec::new();
 							for forward_info in pending_forwards.drain(..) {
 								match forward_info {
 									HTLCForwardInfo::AddHTLC(PendingAddHTLCInfo {
@@ -3169,7 +3166,7 @@ impl<M: Deref, T: Deref, K: Deref, F: Deref, L: Deref> ChannelManager<M, T, K, F
 											// Phantom payments are only PendingHTLCRouting::Receive.
 											phantom_shared_secret: None,
 										});
-										match chan.get_mut().send_htlc(outgoing_amt_msat, payment_hash, outgoing_cltv_value, htlc_source.clone(), onion_packet, &self.logger) {
+										match chan.get_mut().queue_htlc(outgoing_amt_msat, payment_hash, outgoing_cltv_value, htlc_source.clone(), onion_packet, &self.logger) {
 											Err(e) => {
 												if let ChannelError::Ignore(msg) = e {
 													log_trace!(self.logger, "Failed to forward HTLC with payment_hash {}: {}", log_bytes!(payment_hash.0), msg);
@@ -3183,28 +3180,15 @@ impl<M: Deref, T: Deref, K: Deref, F: Deref, L: Deref> ChannelManager<M, T, K, F
 												));
 												continue;
 											},
-											Ok(update_add) => {
-												match update_add {
-													Some(msg) => { add_htlc_msgs.push(msg); },
-													None => {
-														// Nothing to do here...we're waiting on a remote
-														// revoke_and_ack before we can add anymore HTLCs. The Channel
-														// will automatically handle building the update_add_htlc and
-														// commitment_signed messages when we can.
-														// TODO: Do some kind of timer to set the channel as !is_live()
-														// as we don't really want others relying on us relaying through
-														// this channel currently :/.
-													}
-												}
-											}
+											Ok(()) => { },
 										}
 									},
 									HTLCForwardInfo::AddHTLC { .. } => {
 										panic!("short_channel_id != 0 should imply any pending_forward entries are of type Forward");
 									},
 									HTLCForwardInfo::FailHTLC { htlc_id, err_packet } => {
 										log_trace!(self.logger, "Failing HTLC back to channel with short id {} (backward HTLC ID {}) after delay", short_chan_id, htlc_id);
-										match chan.get_mut().get_update_fail_htlc(htlc_id, err_packet, &self.logger) {
+										match chan.get_mut().queue_fail_htlc(htlc_id, err_packet, &self.logger) {
 											Err(e) => {
 												if let ChannelError::Ignore(msg) = e {
 													log_trace!(self.logger, "Failed to fail HTLC with ID {} backwards to short_id {}: {}", htlc_id, short_chan_id, msg);
@@ -3216,65 +3200,11 @@ impl<M: Deref, T: Deref, K: Deref, F: Deref, L: Deref> ChannelManager<M, T, K, F
 												// the chain and sending the HTLC-Timeout is their problem.
 												continue;
 											},
-											Ok(Some(msg)) => { fail_htlc_msgs.push(msg); },
-											Ok(None) => {
-												// Nothing to do here...we're waiting on a remote
-												// revoke_and_ack before we can update the commitment
-												// transaction. The Channel will automatically handle
-												// building the update_fail_htlc and commitment_signed
-												// messages when we can.
-												// We don't need any kind of timer here as they should fail
-												// the channel onto the chain if they can't get our
-												// update_fail_htlc in time, it's not our problem.
-											}
+											Ok(()) => { },
 										}
 									},
 								}
 							}
-
-							if !add_htlc_msgs.is_empty() || !fail_htlc_msgs.is_empty() {
-								let (commitment_msg, monitor_update) = match chan.get_mut().send_commitment(&self.logger) {
-									Ok(res) => res,
-									Err(e) => {
-										// We surely failed send_commitment due to bad keys, in that case
-										// close channel and then send error message to peer.
-										let counterparty_node_id = chan.get().get_counterparty_node_id();
-										let err: Result<(), _>  = match e {
-											ChannelError::Ignore(_) | ChannelError::Warn(_) => {
-												panic!("Stated return value requirements in send_commitment() were not met");
-											}
-											ChannelError::Close(msg) => {
-												log_trace!(self.logger, "Closing channel {} due to Close-required error: {}", log_bytes!(chan.key()[..]), msg);
-												let mut channel = remove_channel!(self, chan);
-												// ChannelClosed event is generated by handle_error for us.
-												Err(MsgHandleErrInternal::from_finish_shutdown(msg, channel.channel_id(), channel.get_user_id(), channel.force_shutdown(true), self.get_channel_update_for_broadcast(&channel).ok()))
-											},
-										};
-										handle_errors.push((counterparty_node_id, err));
-										continue;
-									}
-								};
-								match self.chain_monitor.update_channel(chan.get().get_funding_txo().unwrap(), monitor_update) {
-									ChannelMonitorUpdateStatus::Completed => {},
-									e => {
-										handle_errors.push((chan.get().get_counterparty_node_id(), handle_monitor_update_res!(self, e, chan, RAACommitmentOrder::CommitmentFirst, false, true)));
-										continue;
-									}
-								}
-								log_debug!(self.logger, "Forwarding HTLCs resulted in a commitment update with {} HTLCs added and {} HTLCs failed for channel {}",
-									add_htlc_msgs.len(), fail_htlc_msgs.len(), log_bytes!(chan.get().channel_id()));
-								channel_state.pending_msg_events.push(events::MessageSendEvent::UpdateHTLCs {
-									node_id: chan.get().get_counterparty_node_id(),
-									updates: msgs::CommitmentUpdate {
-										update_add_htlcs: add_htlc_msgs,
-										update_fulfill_htlcs: Vec::new(),
-										update_fail_htlcs: fail_htlc_msgs,
-										update_fail_malformed_htlcs: Vec::new(),
-										update_fee: None,
-										commitment_signed: commitment_msg,
-									},
-								});
-							}
 						}
 					}
 				} else {
@@ -3478,9 +3408,11 @@ impl<M: Deref, T: Deref, K: Deref, F: Deref, L: Deref> ChannelManager<M, T, K, F
 		}
 		self.forward_htlcs(&mut phantom_receives);
 
-		for (counterparty_node_id, err) in handle_errors.drain(..) {
-			let _ = handle_error!(self, err, counterparty_node_id);
-		}
+		// Freeing the holding cell here is relatively redundant - in practice we'll do it when we
+		// next get a `get_and_clear_pending_msg_events` call, but some tests rely on it, and it's
+		// nice to do the work now if we can rather than while we're trying to get messages in the
+		// network stack.
+		self.check_free_holding_cells();
 
 		if new_events.is_empty() { return }
 		let mut events = self.pending_events.lock().unwrap();
