@@ -563,6 +563,7 @@ struct ClaimablePayments {
563563/// usually because we're running pre-full-init. They are handled immediately once we detect we are
564564/// running normally, and specifically must be processed before any other non-background
565565/// [`ChannelMonitorUpdate`]s are applied.
566+ #[derive(Debug)]
566567enum BackgroundEvent {
567568 /// Handle a ChannelMonitorUpdate which closes the channel or for an already-closed channel.
568569 /// This is only separated from [`Self::MonitorUpdateRegeneratedOnStartup`] as the
@@ -5381,8 +5382,11 @@ where
53815382 for htlc in sources.drain(..) {
53825383 if let Err((pk, err)) = self.claim_funds_from_hop(
53835384 htlc.prev_hop, payment_preimage,
5384- |_| Some(MonitorUpdateCompletionAction::PaymentClaimed { payment_hash }))
5385- {
5385+ |_, definitely_duplicate| {
5386+ debug_assert!(!definitely_duplicate, "We shouldn't claim duplicatively from a payment");
5387+ Some(MonitorUpdateCompletionAction::PaymentClaimed { payment_hash })
5388+ }
5389+ ) {
53865390 if let msgs::ErrorAction::IgnoreError = err.err.action {
53875391 // We got a temporary failure updating monitor, but will claim the
53885392 // HTLC when the monitor updating is restored (or on chain).
@@ -5410,7 +5414,7 @@ where
54105414 }
54115415 }
54125416
5413- fn claim_funds_from_hop<ComplFunc: FnOnce(Option<u64>) -> Option<MonitorUpdateCompletionAction>>(&self,
5417+ fn claim_funds_from_hop<ComplFunc: FnOnce(Option<u64>, bool ) -> Option<MonitorUpdateCompletionAction>>(&self,
54145418 prev_hop: HTLCPreviousHopData, payment_preimage: PaymentPreimage, completion_action: ComplFunc)
54155419 -> Result<(), (PublicKey, MsgHandleErrInternal)> {
54165420 //TODO: Delay the claimed_funds relaying just like we do outbound relay!
@@ -5420,6 +5424,11 @@ where
54205424 // `BackgroundEvent`s.
54215425 let during_init = !self.background_events_processed_since_startup.load(Ordering::Acquire);
54225426
5427+ // As we may call handle_monitor_update_completion_actions in rather rare cases, check that
5428+ // the required mutexes are not held before we start.
5429+ debug_assert_ne!(self.pending_events.held_by_thread(), LockHeldState::HeldByThread);
5430+ debug_assert_ne!(self.claimable_payments.held_by_thread(), LockHeldState::HeldByThread);
5431+
54235432 {
54245433 let per_peer_state = self.per_peer_state.read().unwrap();
54255434 let chan_id = prev_hop.outpoint.to_channel_id();
@@ -5441,25 +5450,70 @@ where
54415450 let counterparty_node_id = chan.context.get_counterparty_node_id();
54425451 let fulfill_res = chan.get_update_fulfill_htlc_and_commit(prev_hop.htlc_id, payment_preimage, &self.logger);
54435452
5444- if let UpdateFulfillCommitFetch::NewClaim { htlc_value_msat, monitor_update } = fulfill_res {
5445- if let Some(action) = completion_action(Some(htlc_value_msat)) {
5446- log_trace!(self.logger, "Tracking monitor update completion action for channel {}: {:?}",
5447- chan_id, action);
5448- peer_state.monitor_update_blocked_actions.entry(chan_id).or_insert(Vec::new()).push(action);
5453+ match fulfill_res {
5454+ UpdateFulfillCommitFetch::NewClaim { htlc_value_msat, monitor_update } => {
5455+ if let Some(action) = completion_action(Some(htlc_value_msat), false) {
5456+ log_trace!(self.logger, "Tracking monitor update completion action for channel {}: {:?}",
5457+ chan_id, action);
5458+ peer_state.monitor_update_blocked_actions.entry(chan_id).or_insert(Vec::new()).push(action);
5459+ }
5460+ if !during_init {
5461+ handle_new_monitor_update!(self, prev_hop.outpoint, monitor_update, peer_state_lock,
5462+ peer_state, per_peer_state, chan);
5463+ } else {
5464+ // If we're running during init we cannot update a monitor directly -
5465+ // they probably haven't actually been loaded yet. Instead, push the
5466+ // monitor update as a background event.
5467+ self.pending_background_events.lock().unwrap().push(
5468+ BackgroundEvent::MonitorUpdateRegeneratedOnStartup {
5469+ counterparty_node_id,
5470+ funding_txo: prev_hop.outpoint,
5471+ update: monitor_update.clone(),
5472+ });
5473+ }
54495474 }
5450- if !during_init {
5451- handle_new_monitor_update!(self, prev_hop.outpoint, monitor_update, peer_state_lock,
5452- peer_state, per_peer_state, chan);
5453- } else {
5454- // If we're running during init we cannot update a monitor directly -
5455- // they probably haven't actually been loaded yet. Instead, push the
5456- // monitor update as a background event.
5457- self.pending_background_events.lock().unwrap().push(
5458- BackgroundEvent::MonitorUpdateRegeneratedOnStartup {
5459- counterparty_node_id,
5460- funding_txo: prev_hop.outpoint,
5461- update: monitor_update.clone(),
5462- });
5475+ UpdateFulfillCommitFetch::DuplicateClaim {} => {
5476+ let action = if let Some(action) = completion_action(None, true) {
5477+ action
5478+ } else {
5479+ return Ok(());
5480+ };
5481+ mem::drop(peer_state_lock);
5482+
5483+ log_trace!(self.logger, "Completing monitor update completion action for channel {} as claim was redundant: {:?}",
5484+ chan_id, action);
5485+ let (node_id, funding_outpoint, blocker) =
5486+ if let MonitorUpdateCompletionAction::FreeOtherChannelImmediately {
5487+ downstream_counterparty_node_id: node_id,
5488+ downstream_funding_outpoint: funding_outpoint,
5489+ blocking_action: blocker,
5490+ } = action {
5491+ (node_id, funding_outpoint, blocker)
5492+ } else {
5493+ debug_assert!(false,
5494+ "Duplicate claims should always free another channel immediately");
5495+ return Ok(());
5496+ };
5497+ if let Some(peer_state_mtx) = per_peer_state.get(&node_id) {
5498+ let mut peer_state = peer_state_mtx.lock().unwrap();
5499+ if let Some(blockers) = peer_state
5500+ .actions_blocking_raa_monitor_updates
5501+ .get_mut(&funding_outpoint.to_channel_id())
5502+ {
5503+ let mut found_blocker = false;
5504+ blockers.retain(|iter| {
5505+ // Note that we could actually be blocked, in
5506+ // which case we need to only remove the one
5507+ // blocker which was added duplicatively.
5508+ let first_blocker = !found_blocker;
5509+ if *iter == blocker { found_blocker = true; }
5510+ *iter != blocker || !first_blocker
5511+ });
5512+ debug_assert!(found_blocker);
5513+ }
5514+ } else {
5515+ debug_assert!(false);
5516+ }
54635517 }
54645518 }
54655519 }
@@ -5507,7 +5561,7 @@ where
55075561 // `ChannelMonitor` we've provided the above update to. Instead, note that `Event`s are
55085562 // generally always allowed to be duplicative (and it's specifically noted in
55095563 // `PaymentForwarded`).
5510- self.handle_monitor_update_completion_actions(completion_action(None));
5564+ self.handle_monitor_update_completion_actions(completion_action(None, false ));
55115565 Ok(())
55125566 }
55135567
@@ -5537,13 +5591,84 @@ where
55375591 HTLCSource::PreviousHopData(hop_data) => {
55385592 let prev_outpoint = hop_data.outpoint;
55395593 let completed_blocker = RAAMonitorUpdateBlockingAction::from_prev_hop_data(&hop_data);
5594+ #[cfg(debug_assertions)]
5595+ let claiming_chan_funding_outpoint = hop_data.outpoint;
55405596 let res = self.claim_funds_from_hop(hop_data, payment_preimage,
5541- |htlc_claim_value_msat| {
5542- if let Some(forwarded_htlc_value) = forwarded_htlc_value_msat {
5543- let fee_earned_msat = if let Some(claimed_htlc_value) = htlc_claim_value_msat {
5544- Some(claimed_htlc_value - forwarded_htlc_value)
5545- } else { None };
5597+ |htlc_claim_value_msat, definitely_duplicate| {
5598+ let chan_to_release =
5599+ if let Some(node_id) = next_channel_counterparty_node_id {
5600+ Some((node_id, next_channel_outpoint, completed_blocker))
5601+ } else {
5602+ // We can only get `None` here if we are processing a
5603+ // `ChannelMonitor`-originated event, in which case we
5604+ // don't care about ensuring we wake the downstream
5605+ // channel's monitor updating - the channel is already
5606+ // closed.
5607+ None
5608+ };
55465609
5610+ if definitely_duplicate && startup_replay {
5611+ // On startup we may get redundant claims which are related to
5612+ // monitor updates still in flight. In that case, we shouldn't
5613+ // immediately free, but instead let that monitor update complete
5614+ // in the background.
5615+ #[cfg(debug_assertions)] {
5616+ let background_events = self.pending_background_events.lock().unwrap();
5617+ // There should be a `BackgroundEvent` pending...
5618+ assert!(background_events.iter().any(|ev| {
5619+ match ev {
5620+ // to apply a monitor update that blocked the claiming channel,
5621+ BackgroundEvent::MonitorUpdateRegeneratedOnStartup {
5622+ funding_txo, update, ..
5623+ } => {
5624+ if *funding_txo == claiming_chan_funding_outpoint {
5625+ assert!(update.updates.iter().any(|upd|
5626+ if let ChannelMonitorUpdateStep::PaymentPreimage {
5627+ payment_preimage: update_preimage
5628+ } = upd {
5629+ payment_preimage == *update_preimage
5630+ } else { false }
5631+ ), "{:?}", update);
5632+ true
5633+ } else { false }
5634+ },
5635+ // or the channel we'd unblock is already closed,
5636+ BackgroundEvent::ClosedMonitorUpdateRegeneratedOnStartup(
5637+ (funding_txo, monitor_update)
5638+ ) => {
5639+ if *funding_txo == next_channel_outpoint {
5640+ assert_eq!(monitor_update.updates.len(), 1);
5641+ assert!(matches!(
5642+ monitor_update.updates[0],
5643+ ChannelMonitorUpdateStep::ChannelForceClosed { .. }
5644+ ));
5645+ true
5646+ } else { false }
5647+ },
5648+ // or the monitor update has completed and will unblock
5649+ // immediately once we get going.
5650+ BackgroundEvent::MonitorUpdatesComplete {
5651+ channel_id, ..
5652+ } =>
5653+ *channel_id == claiming_chan_funding_outpoint.to_channel_id(),
5654+ }
5655+ }), "{:?}", *background_events);
5656+ }
5657+ None
5658+ } else if definitely_duplicate {
5659+ if let Some(other_chan) = chan_to_release {
5660+ Some(MonitorUpdateCompletionAction::FreeOtherChannelImmediately {
5661+ downstream_counterparty_node_id: other_chan.0,
5662+ downstream_funding_outpoint: other_chan.1,
5663+ blocking_action: other_chan.2,
5664+ })
5665+ } else { None }
5666+ } else {
5667+ let fee_earned_msat = if let Some(forwarded_htlc_value) = forwarded_htlc_value_msat {
5668+ if let Some(claimed_htlc_value) = htlc_claim_value_msat {
5669+ Some(claimed_htlc_value - forwarded_htlc_value)
5670+ } else { None }
5671+ } else { None };
55475672 Some(MonitorUpdateCompletionAction::EmitEventAndFreeOtherChannel {
55485673 event: events::Event::PaymentForwarded {
55495674 fee_earned_msat,
@@ -5552,19 +5677,9 @@ where
55525677 next_channel_id: Some(next_channel_outpoint.to_channel_id()),
55535678 outbound_amount_forwarded_msat: forwarded_htlc_value_msat,
55545679 },
5555- downstream_counterparty_and_funding_outpoint:
5556- if let Some(node_id) = next_channel_counterparty_node_id {
5557- Some((node_id, next_channel_outpoint, completed_blocker))
5558- } else {
5559- // We can only get `None` here if we are processing a
5560- // `ChannelMonitor`-originated event, in which case we
5561- // don't care about ensuring we wake the downstream
5562- // channel's monitor updating - the channel is already
5563- // closed.
5564- None
5565- },
5680+ downstream_counterparty_and_funding_outpoint: chan_to_release,
55665681 })
5567- } else { None }
5682+ }
55685683 });
55695684 if let Err((pk, err)) = res {
55705685 let result: Result<(), _> = Err(err);
@@ -5580,6 +5695,10 @@ where
55805695 }
55815696
55825697 fn handle_monitor_update_completion_actions<I: IntoIterator<Item=MonitorUpdateCompletionAction>>(&self, actions: I) {
5698+ debug_assert_ne!(self.pending_events.held_by_thread(), LockHeldState::HeldByThread);
5699+ debug_assert_ne!(self.claimable_payments.held_by_thread(), LockHeldState::HeldByThread);
5700+ debug_assert_ne!(self.per_peer_state.held_by_thread(), LockHeldState::HeldByThread);
5701+
55835702 for action in actions.into_iter() {
55845703 match action {
55855704 MonitorUpdateCompletionAction::PaymentClaimed { payment_hash } => {
0 commit comments