Allow claiming a payment if a channel with an HTLC has closed
authorMatt Corallo <git@bluematt.me>
Tue, 4 Apr 2023 04:12:55 +0000 (04:12 +0000)
committerMatt Corallo <git@bluematt.me>
Tue, 4 Apr 2023 23:08:46 +0000 (23:08 +0000)
Previously, LDK would refuse to claim a payment if a channel on
which the payment was received had been closed between when the
HTLC was received and when we went to claim it. This makes sense in
the payment case - why pay an on-chain fee to claim the HTLC when
presumably the sender may retry later. Long ago it also reduced
total code in the claim pipeline.

However, this doesn't make sense if you're trying to do an atomic
swap or some other protocol that requires atomicity with some other
action - if your money got claimed elsewhere you need to be able to
claim the HTLC in lightning no matter what. Further, this is an
over-optimization - there should be a very, very low likelihood
that a channel closes between when we receive the last HTLC for a
payment and the user goes to claim the payment. Since we now have
code to handle this anyway we should allow it.

Fixes #2017.

lightning/src/ln/channelmanager.rs
lightning/src/ln/payment_tests.rs

index c2ff9da67f38cc169c048ca7279952109f346d3d..8b536c5dfc6464fc99e73460ec82ff86809c7fc1 100644 (file)
@@ -3981,21 +3981,10 @@ where
                };
                debug_assert!(!sources.is_empty());
 
-               // If we are claiming an MPP payment, we check that all channels which contain a claimable
-               // HTLC still exist. While this isn't guaranteed to remain true if a channel closes while
-               // we're claiming (or even after we claim, before the commitment update dance completes),
-               // it should be a relatively rare race, and we'd rather not claim HTLCs that require us to
-               // go on-chain (and lose the on-chain fee to do so) than just reject the payment.
-               //
-               // Note that we'll still always get our funds - as long as the generated
-               // `ChannelMonitorUpdate` makes it out to the relevant monitor we can claim on-chain.
-               //
-               // If we find an HTLC which we would need to claim but for which we do not have a
-               // channel, we will fail all parts of the MPP payment. While we could wait and see if
-               // the sender retries the already-failed path(s), it should be a pretty rare case where
-               // we got all the HTLCs and then a channel closed while we were waiting for the user to
-               // provide the preimage, so worrying too much about the optimal handling isn't worth
-               // it.
+               // Just in case one HTLC has been failed between when we generated the `PaymentClaimable`
+               // and when we got here we need to check that the amount we're about to claim matches the
+               // amount we told the user in the last `PaymentClaimable`. We also do a sanity-check that
+               // the MPP parts all have the same `total_msat`.
                let mut claimable_amt_msat = 0;
                let mut prev_total_msat = None;
                let mut expected_amt_msat = None;
@@ -4003,28 +3992,6 @@ where
                let mut errs = Vec::new();
                let per_peer_state = self.per_peer_state.read().unwrap();
                for htlc in sources.iter() {
-                       let (counterparty_node_id, chan_id) = match self.short_to_chan_info.read().unwrap().get(&htlc.prev_hop.short_channel_id) {
-                               Some((cp_id, chan_id)) => (cp_id.clone(), chan_id.clone()),
-                               None => {
-                                       valid_mpp = false;
-                                       break;
-                               }
-                       };
-
-                       let peer_state_mutex_opt = per_peer_state.get(&counterparty_node_id);
-                       if peer_state_mutex_opt.is_none() {
-                               valid_mpp = false;
-                               break;
-                       }
-
-                       let mut peer_state_lock = peer_state_mutex_opt.unwrap().lock().unwrap();
-                       let peer_state = &mut *peer_state_lock;
-
-                       if peer_state.channel_by_id.get(&chan_id).is_none() {
-                               valid_mpp = false;
-                               break;
-                       }
-
                        if prev_total_msat.is_some() && prev_total_msat != Some(htlc.total_msat) {
                                log_error!(self.logger, "Somehow ended up with an MPP payment with different expected total amounts - this should not be reachable!");
                                debug_assert!(false);
index 1ce0cc0345869dd9a392ebf78944c124d471aefc..fb50f5a4a38d4d2629e21954679a6c850b420a58 100644 (file)
@@ -12,7 +12,7 @@
 //! payments thereafter.
 
 use crate::chain::{ChannelMonitorUpdateStatus, Confirm, Listen, Watch};
-use crate::chain::channelmonitor::{ANTI_REORG_DELAY, LATENCY_GRACE_PERIOD_BLOCKS};
+use crate::chain::channelmonitor::{ANTI_REORG_DELAY, HTLC_FAIL_BACK_BUFFER, LATENCY_GRACE_PERIOD_BLOCKS};
 use crate::chain::keysinterface::EntropySource;
 use crate::chain::transaction::OutPoint;
 use crate::events::{ClosureReason, Event, HTLCDestination, MessageSendEvent, MessageSendEventsProvider, PathFailure};
@@ -23,7 +23,7 @@ use crate::ln::msgs;
 use crate::ln::msgs::ChannelMessageHandler;
 use crate::ln::outbound_payment::Retry;
 use crate::routing::gossip::{EffectiveCapacity, RoutingFees};
-use crate::routing::router::{get_route, PaymentParameters, Route, RouteHint, RouteHintHop, RouteHop, RouteParameters};
+use crate::routing::router::{get_route, PaymentParameters, Route, Router, RouteHint, RouteHintHop, RouteHop, RouteParameters};
 use crate::routing::scoring::ChannelUsage;
 use crate::util::test_utils;
 use crate::util::errors::APIError;
@@ -2838,3 +2838,138 @@ fn no_missing_sent_on_midpoint_reload() {
        do_no_missing_sent_on_midpoint_reload(false);
        do_no_missing_sent_on_midpoint_reload(true);
 }
+
+fn do_claim_from_closed_chan(fail_payment: bool) {
+       // Previously, LDK would refuse to claim a payment if a channel on which the payment was
+       // received had been closed between when the HTLC was received and when we went to claim it.
+       // This makes sense in the payment case - why pay an on-chain fee to claim the HTLC when
+       // presumably the sender may retry later. Long ago it also reduced total code in the claim
+       // pipeline.
+       //
+       // However, this doesn't make sense if you're trying to do an atomic swap or some other
+       // protocol that requires atomicity with some other action - if your money got claimed
+       // elsewhere you need to be able to claim the HTLC in lightning no matter what. Further, this
+       // is an over-optimization - there should be a very, very low likelihood that a channel closes
+       // between when we receive the last HTLC for a payment and the user goes to claim the payment.
+       // Since we now have code to handle this anyway we should allow it.
+
+       // Build 4 nodes and send an MPP payment across two paths. By building a route manually set the
+       // CLTVs on the paths to different value resulting in a different claim deadline.
+       let chanmon_cfgs = create_chanmon_cfgs(4);
+       let node_cfgs = create_node_cfgs(4, &chanmon_cfgs);
+       let node_chanmgrs = create_node_chanmgrs(4, &node_cfgs, &[None, None, None, None]);
+       let mut nodes = create_network(4, &node_cfgs, &node_chanmgrs);
+
+       create_announced_chan_between_nodes(&nodes, 0, 1);
+       create_announced_chan_between_nodes_with_value(&nodes, 0, 2, 1_000_000, 0);
+       let chan_bd = create_announced_chan_between_nodes_with_value(&nodes, 1, 3, 1_000_000, 0).2;
+       create_announced_chan_between_nodes(&nodes, 2, 3);
+
+       let (payment_preimage, payment_hash, payment_secret) = get_payment_preimage_hash!(nodes[3]);
+       let mut route_params = RouteParameters {
+               payment_params: PaymentParameters::from_node_id(nodes[3].node.get_our_node_id(), TEST_FINAL_CLTV)
+                       .with_features(nodes[1].node.invoice_features()),
+               final_value_msat: 10_000_000,
+       };
+       let mut route = nodes[0].router.find_route(&nodes[0].node.get_our_node_id(), &route_params,
+               None, &nodes[0].node.compute_inflight_htlcs()).unwrap();
+       // Make sure the route is ordered as the B->D path before C->D
+       route.paths.sort_by(|a, _| if a[0].pubkey == nodes[1].node.get_our_node_id() {
+               std::cmp::Ordering::Less } else { std::cmp::Ordering::Greater });
+
+       // Note that we add an extra 1 in the send pipeline to compensate for any blocks found while
+       // the HTLC is being relayed.
+       route.paths[0][1].cltv_expiry_delta = TEST_FINAL_CLTV + 8;
+       route.paths[1][1].cltv_expiry_delta = TEST_FINAL_CLTV + 12;
+       let final_cltv = nodes[0].best_block_info().1 + TEST_FINAL_CLTV + 8 + 1;
+
+       nodes[0].router.expect_find_route(route_params.clone(), Ok(route.clone()));
+       nodes[0].node.send_payment_with_retry(payment_hash, &Some(payment_secret),
+               PaymentId(payment_hash.0), route_params.clone(), Retry::Attempts(1)).unwrap();
+       check_added_monitors(&nodes[0], 2);
+       let mut send_msgs = nodes[0].node.get_and_clear_pending_msg_events();
+       send_msgs.sort_by(|a, _| {
+               let a_node_id =
+                       if let MessageSendEvent::UpdateHTLCs { node_id, .. } = a { node_id } else { panic!() };
+               let node_b_id = nodes[1].node.get_our_node_id();
+               if *a_node_id == node_b_id { std::cmp::Ordering::Less } else { std::cmp::Ordering::Greater }
+       });
+
+       assert_eq!(send_msgs.len(), 2);
+       pass_along_path(&nodes[0], &[&nodes[1], &nodes[3]], 10_000_000,
+               payment_hash, Some(payment_secret), send_msgs.remove(0), false, None);
+       pass_along_path(&nodes[0], &[&nodes[2], &nodes[3]], 10_000_000,
+               payment_hash, Some(payment_secret), send_msgs.remove(0), true, None);
+
+       // Ensure that the claim_deadline is correct, with the payment failing at exactly the given
+       // height.
+       connect_blocks(&nodes[3], final_cltv - HTLC_FAIL_BACK_BUFFER - nodes[3].best_block_info().1
+               - if fail_payment { 0 } else { 2 });
+       if fail_payment {
+               // We fail the HTLC on the A->B->D path first as it expires 4 blocks earlier. We go ahead
+               // and expire both immediately, though, by connecting another 4 blocks.
+               let reason = HTLCDestination::FailedPayment { payment_hash };
+               expect_pending_htlcs_forwardable_and_htlc_handling_failed!(&nodes[3], [reason.clone()]);
+               connect_blocks(&nodes[3], 4);
+               expect_pending_htlcs_forwardable_and_htlc_handling_failed!(&nodes[3], [reason]);
+               pass_failed_payment_back(&nodes[0], &[&[&nodes[1], &nodes[3]], &[&nodes[2], &nodes[3]]], false, payment_hash);
+       } else {
+               nodes[1].node.force_close_broadcasting_latest_txn(&chan_bd, &nodes[3].node.get_our_node_id()).unwrap();
+               check_closed_event(&nodes[1], 1, ClosureReason::HolderForceClosed, false);
+               check_closed_broadcast(&nodes[1], 1, true);
+               let bs_tx = nodes[1].tx_broadcaster.txn_broadcasted.lock().unwrap().split_off(0);
+               assert_eq!(bs_tx.len(), 1);
+
+               mine_transaction(&nodes[3], &bs_tx[0]);
+               check_added_monitors(&nodes[3], 1);
+               check_closed_broadcast(&nodes[3], 1, true);
+               check_closed_event(&nodes[3], 1, ClosureReason::CommitmentTxConfirmed, false);
+
+               nodes[3].node.claim_funds(payment_preimage);
+               check_added_monitors(&nodes[3], 2);
+               expect_payment_claimed!(nodes[3], payment_hash, 10_000_000);
+
+               let ds_tx = nodes[3].tx_broadcaster.txn_broadcasted.lock().unwrap().split_off(0);
+               assert_eq!(ds_tx.len(), 1);
+               check_spends!(&ds_tx[0], &bs_tx[0]);
+
+               mine_transactions(&nodes[1], &[&bs_tx[0], &ds_tx[0]]);
+               check_added_monitors(&nodes[1], 1);
+               expect_payment_forwarded!(nodes[1], nodes[0], nodes[3], Some(1000), false, true);
+
+               let bs_claims = nodes[1].node.get_and_clear_pending_msg_events();
+               check_added_monitors(&nodes[1], 1);
+               assert_eq!(bs_claims.len(), 1);
+               if let MessageSendEvent::UpdateHTLCs { updates, .. } = &bs_claims[0] {
+                       nodes[0].node.handle_update_fulfill_htlc(&nodes[1].node.get_our_node_id(), &updates.update_fulfill_htlcs[0]);
+                       commitment_signed_dance!(nodes[0], nodes[1], updates.commitment_signed, false, true);
+               } else { panic!(); }
+
+               expect_payment_sent!(nodes[0], payment_preimage);
+
+               let ds_claim_msgs = nodes[3].node.get_and_clear_pending_msg_events();
+               assert_eq!(ds_claim_msgs.len(), 1);
+               let cs_claim_msgs = if let MessageSendEvent::UpdateHTLCs { updates, .. } = &ds_claim_msgs[0] {
+                       nodes[2].node.handle_update_fulfill_htlc(&nodes[3].node.get_our_node_id(), &updates.update_fulfill_htlcs[0]);
+                       let cs_claim_msgs = nodes[2].node.get_and_clear_pending_msg_events();
+                       check_added_monitors(&nodes[2], 1);
+                       commitment_signed_dance!(nodes[2], nodes[3], updates.commitment_signed, false, true);
+                       expect_payment_forwarded!(nodes[2], nodes[0], nodes[3], Some(1000), false, false);
+                       cs_claim_msgs
+               } else { panic!(); };
+
+               assert_eq!(cs_claim_msgs.len(), 1);
+               if let MessageSendEvent::UpdateHTLCs { updates, .. } = &cs_claim_msgs[0] {
+                       nodes[0].node.handle_update_fulfill_htlc(&nodes[2].node.get_our_node_id(), &updates.update_fulfill_htlcs[0]);
+                       commitment_signed_dance!(nodes[0], nodes[2], updates.commitment_signed, false, true);
+               } else { panic!(); }
+
+               expect_payment_path_successful!(nodes[0]);
+       }
+}
+
+#[test]
+fn claim_from_closed_chan() {
+       do_claim_from_closed_chan(true);
+       do_claim_from_closed_chan(false);
+}