]> git.bitcoin.ninja Git - rust-lightning/commitdiff
Avoid writing `ChannelManager` when hitting lnd bug 6039 2024-03-no-wake-on-shutdown-121
authorMatt Corallo <git@bluematt.me>
Thu, 14 Mar 2024 19:47:07 +0000 (19:47 +0000)
committerMatt Corallo <git@bluematt.me>
Thu, 14 Mar 2024 20:08:47 +0000 (20:08 +0000)
When we hit lnd bug 6039, we end up sending error messages to peers
in a loop. This should be fine, but because we used the generic
`PersistenceNotifierGuard::notify_on_drop` lock above the specific
handling, we end up writing `ChannelManager` every time we manage a
round-trip to our peer.

This can add up quite quickly, and isn't actually changing, so we
really need to avoid writing the `ChannelManager` in this case.

lightning/src/ln/channelmanager.rs

index 46002f442966f647f0f2e0b4992e6848d46fee55..2728240c719d27be15b28fbd77bf9840d8df38be 100644 (file)
@@ -9026,8 +9026,6 @@ where
        }
 
        fn handle_error(&self, counterparty_node_id: &PublicKey, msg: &msgs::ErrorMessage) {
-               let _persistence_guard = PersistenceNotifierGuard::notify_on_drop(self);
-
                match &msg.data as &str {
                        "cannot co-op close channel w/ active htlcs"|
                        "link failed to shutdown" =>
@@ -9040,34 +9038,45 @@ where
                                // We're not going to bother handling this in a sensible way, instead simply
                                // repeating the Shutdown message on repeat until morale improves.
                                if !msg.channel_id.is_zero() {
-                                       let per_peer_state = self.per_peer_state.read().unwrap();
-                                       let peer_state_mutex_opt = per_peer_state.get(counterparty_node_id);
-                                       if peer_state_mutex_opt.is_none() { return; }
-                                       let mut peer_state = peer_state_mutex_opt.unwrap().lock().unwrap();
-                                       if let Some(ChannelPhase::Funded(chan)) = peer_state.channel_by_id.get(&msg.channel_id) {
-                                               if let Some(msg) = chan.get_outbound_shutdown() {
-                                                       peer_state.pending_msg_events.push(events::MessageSendEvent::SendShutdown {
-                                                               node_id: *counterparty_node_id,
-                                                               msg,
-                                                       });
-                                               }
-                                               peer_state.pending_msg_events.push(events::MessageSendEvent::HandleError {
-                                                       node_id: *counterparty_node_id,
-                                                       action: msgs::ErrorAction::SendWarningMessage {
-                                                               msg: msgs::WarningMessage {
-                                                                       channel_id: msg.channel_id,
-                                                                       data: "You appear to be exhibiting LND bug 6039, we'll keep sending you shutdown messages until you handle them correctly".to_owned()
-                                                               },
-                                                               log_level: Level::Trace,
+                                       PersistenceNotifierGuard::optionally_notify(
+                                               self,
+                                               || -> NotifyOption {
+                                                       let per_peer_state = self.per_peer_state.read().unwrap();
+                                                       let peer_state_mutex_opt = per_peer_state.get(counterparty_node_id);
+                                                       if peer_state_mutex_opt.is_none() { return NotifyOption::SkipPersistNoEvents; }
+                                                       let mut peer_state = peer_state_mutex_opt.unwrap().lock().unwrap();
+                                                       if let Some(ChannelPhase::Funded(chan)) = peer_state.channel_by_id.get(&msg.channel_id) {
+                                                               if let Some(msg) = chan.get_outbound_shutdown() {
+                                                                       peer_state.pending_msg_events.push(events::MessageSendEvent::SendShutdown {
+                                                                               node_id: *counterparty_node_id,
+                                                                               msg,
+                                                                       });
+                                                               }
+                                                               peer_state.pending_msg_events.push(events::MessageSendEvent::HandleError {
+                                                                       node_id: *counterparty_node_id,
+                                                                       action: msgs::ErrorAction::SendWarningMessage {
+                                                                               msg: msgs::WarningMessage {
+                                                                                       channel_id: msg.channel_id,
+                                                                                       data: "You appear to be exhibiting LND bug 6039, we'll keep sending you shutdown messages until you handle them correctly".to_owned()
+                                                                               },
+                                                                               log_level: Level::Trace,
+                                                                       }
+                                                               });
+                                                               // This can happen in a fairly tight loop, so we absolutely cannot trigger
+                                                               // a `ChannelManager` write here.
+                                                               return NotifyOption::SkipPersistHandleEvents;
                                                        }
-                                               });
-                                       }
+                                                       NotifyOption::SkipPersistNoEvents
+                                               }
+                                       );
                                }
                                return;
                        }
                        _ => {}
                }
 
+               let _persistence_guard = PersistenceNotifierGuard::notify_on_drop(self);
+
                if msg.channel_id.is_zero() {
                        let channel_ids: Vec<ChannelId> = {
                                let per_peer_state = self.per_peer_state.read().unwrap();