Fail HTLC backwards on unrevoked remote commitment tx broadcast
authorMatt Corallo <git@bluematt.me>
Fri, 21 Dec 2018 20:16:46 +0000 (15:16 -0500)
committerMatt Corallo <git@bluematt.me>
Tue, 1 Jan 2019 00:56:32 +0000 (19:56 -0500)
src/ln/channel.rs
src/ln/channelmonitor.rs
src/ln/functional_tests.rs

index afecac4297d9097002893973b6bc74f1a5748573..ddfb809ccdeaf6f7b542c0c08e2850f7646a80cf 100644 (file)
@@ -315,6 +315,9 @@ pub(super) struct Channel {
        funding_tx_confirmations: u64,
 
        their_dust_limit_satoshis: u64,
+       #[cfg(test)]
+       pub(super) our_dust_limit_satoshis: u64,
+       #[cfg(not(test))]
        our_dust_limit_satoshis: u64,
        their_max_htlc_value_in_flight_msat: u64,
        //get_our_max_htlc_value_in_flight_msat(): u64,
@@ -346,7 +349,7 @@ pub(super) struct Channel {
        logger: Arc<Logger>,
 }
 
-const OUR_MAX_HTLCS: u16 = 5; //TODO
+const OUR_MAX_HTLCS: u16 = 50; //TODO
 /// Confirmation count threshold at which we close a channel. Ideally we'd keep the channel around
 /// on ice until the funding transaction gets more confirmations, but the LN protocol doesn't
 /// really allow for this, so instead we're stuck closing it out at that point.
index cdd0aad1139734ab29d5b8609b345f85214cccb2..13cdfd2fe5d75433c5c28db097af4c12c0965de1 100644 (file)
@@ -1247,6 +1247,42 @@ impl ChannelMonitor {
 
                        log_trace!(self, "Got broadcast of non-revoked remote commitment transaction {}", commitment_txid);
 
+                       // TODO: We really should only fail backwards after our revocation claims have been
+                       // confirmed, but we also need to do more other tracking of in-flight pre-confirm
+                       // on-chain claims, so we can do that at the same time.
+                       macro_rules! check_htlc_fails {
+                               ($txid: expr, $commitment_tx: expr, $id: tt) => {
+                                       if let Some(&(_, ref latest_outpoints)) = self.remote_claimable_outpoints.get(&$txid) {
+                                               $id: for &(ref payment_hash, ref source, _) in latest_outpoints.iter() {
+                                                       // Check if the HTLC is present in the commitment transaction that was
+                                                       // broadcast, but not if it was below the dust limit, which we should
+                                                       // fail backwards immediately as there is no way for us to learn the
+                                                       // payment_preimage.
+                                                       // Note that if the dust limit were allowed to change between
+                                                       // commitment transactions we'd want to be check whether *any*
+                                                       // broadcastable commitment transaction has the HTLC in it, but it
+                                                       // cannot currently change after channel initialization, so we don't
+                                                       // need to here.
+                                                       for &(_, ref broadcast_source, ref output_idx) in per_commitment_data.1.iter() {
+                                                               if output_idx.is_some() && source == broadcast_source {
+                                                                       continue $id;
+                                                               }
+                                                       }
+                                                       log_trace!(self, "Failing HTLC with payment_hash {} from {} remote commitment tx due to broadcast of remote commitment transaction", log_bytes!(payment_hash.0), $commitment_tx);
+                                                       htlc_updated.push(((*source).clone(), None, payment_hash.clone()));
+                                               }
+                                       }
+                               }
+                       }
+                       if let Storage::Local { ref current_remote_commitment_txid, ref prev_remote_commitment_txid, .. } = self.key_storage {
+                               if let &Some(ref txid) = current_remote_commitment_txid {
+                                       check_htlc_fails!(txid, "current", 'current_loop);
+                               }
+                               if let &Some(ref txid) = prev_remote_commitment_txid {
+                                       check_htlc_fails!(txid, "previous", 'prev_loop);
+                               }
+                       }
+
                        if let Some(revocation_points) = self.their_cur_revocation_points {
                                let revocation_point_option =
                                        if revocation_points.0 == commitment_number { Some(&revocation_points.1) }
@@ -1407,10 +1443,6 @@ impl ChannelMonitor {
                                                output: spend_tx.output[0].clone(),
                                        });
                                        txn_to_broadcast.push(spend_tx);
-
-                                       // TODO: We need to fail back HTLCs that were't included in the broadcast
-                                       // commitment transaction, either because they didn't meet dust or because a
-                                       // stale (but not yet revoked) commitment transaction was broadcast!
                                }
                        }
                }
index 3087f76b84b580422502bcdc7464c3d7dbe4ec7d..9331d2680c3c4e53cd68feff5da9535862eb9b91 100644 (file)
@@ -5697,6 +5697,245 @@ fn test_dynamic_spendable_outputs_local_htlc_success_tx() {
        check_spends!(spend_txn[1], node_txn[2].clone());
 }
 
+fn do_test_fail_backwards_unrevoked_remote_announce(deliver_last_raa: bool, announce_latest: bool) {
+       // Test that we fail backwards the full set of HTLCs we need to when remote broadcasts an
+       // unrevoked commitment transaction.
+       // This includes HTLCs which were below the dust threshold as well as HTLCs which were awaiting
+       // a remote RAA before they could be failed backwards (and combinations thereof).
+       // We also test duplicate-hash HTLCs by adding two nodes on each side of the target nodes which
+       // use the same payment hashes.
+       // Thus, we use a six-node network:
+       //
+       // A \         / E
+       //    - C - D -
+       // B /         \ F
+       // And test where C fails back to A/B when D announces its latest commitment transaction
+       let nodes = create_network(6);
+
+       create_announced_chan_between_nodes(&nodes, 0, 2);
+       create_announced_chan_between_nodes(&nodes, 1, 2);
+       let chan = create_announced_chan_between_nodes(&nodes, 2, 3);
+       create_announced_chan_between_nodes(&nodes, 3, 4);
+       create_announced_chan_between_nodes(&nodes, 3, 5);
+
+       // Rebalance and check output sanity...
+       send_payment(&nodes[0], &[&nodes[2], &nodes[3], &nodes[4]], 500000);
+       send_payment(&nodes[1], &[&nodes[2], &nodes[3], &nodes[5]], 500000);
+       assert_eq!(nodes[3].node.channel_state.lock().unwrap().by_id.get(&chan.2).unwrap().last_local_commitment_txn[0].output.len(), 2);
+
+       let ds_dust_limit = nodes[3].node.channel_state.lock().unwrap().by_id.get(&chan.2).unwrap().our_dust_limit_satoshis;
+       // 0th HTLC:
+       let (_, payment_hash_1) = route_payment(&nodes[0], &[&nodes[2], &nodes[3], &nodes[4]], ds_dust_limit*1000); // not added < dust limit + HTLC tx fee
+       // 1st HTLC:
+       let (_, payment_hash_2) = route_payment(&nodes[0], &[&nodes[2], &nodes[3], &nodes[4]], ds_dust_limit*1000); // not added < dust limit + HTLC tx fee
+       let route = nodes[1].router.get_route(&nodes[5].node.get_our_node_id(), None, &Vec::new(), ds_dust_limit*1000, TEST_FINAL_CLTV).unwrap();
+       // 2nd HTLC:
+       send_along_route_with_hash(&nodes[1], route.clone(), &[&nodes[2], &nodes[3], &nodes[5]], ds_dust_limit*1000, payment_hash_1); // not added < dust limit + HTLC tx fee
+       // 3rd HTLC:
+       send_along_route_with_hash(&nodes[1], route, &[&nodes[2], &nodes[3], &nodes[5]], ds_dust_limit*1000, payment_hash_2); // not added < dust limit + HTLC tx fee
+       // 4th HTLC:
+       let (_, payment_hash_3) = route_payment(&nodes[0], &[&nodes[2], &nodes[3], &nodes[4]], 1000000);
+       // 5th HTLC:
+       let (_, payment_hash_4) = route_payment(&nodes[0], &[&nodes[2], &nodes[3], &nodes[4]], 1000000);
+       let route = nodes[1].router.get_route(&nodes[5].node.get_our_node_id(), None, &Vec::new(), 1000000, TEST_FINAL_CLTV).unwrap();
+       // 6th HTLC:
+       send_along_route_with_hash(&nodes[1], route.clone(), &[&nodes[2], &nodes[3], &nodes[5]], 1000000, payment_hash_3);
+       // 7th HTLC:
+       send_along_route_with_hash(&nodes[1], route, &[&nodes[2], &nodes[3], &nodes[5]], 1000000, payment_hash_4);
+
+       // 8th HTLC:
+       let (_, payment_hash_5) = route_payment(&nodes[0], &[&nodes[2], &nodes[3], &nodes[4]], 1000000);
+       // 9th HTLC:
+       let route = nodes[1].router.get_route(&nodes[5].node.get_our_node_id(), None, &Vec::new(), ds_dust_limit*1000, TEST_FINAL_CLTV).unwrap();
+       send_along_route_with_hash(&nodes[1], route, &[&nodes[2], &nodes[3], &nodes[5]], ds_dust_limit*1000, payment_hash_5); // not added < dust limit + HTLC tx fee
+
+       // 10th HTLC:
+       let (_, payment_hash_6) = route_payment(&nodes[0], &[&nodes[2], &nodes[3], &nodes[4]], ds_dust_limit*1000); // not added < dust limit + HTLC tx fee
+       // 11th HTLC:
+       let route = nodes[1].router.get_route(&nodes[5].node.get_our_node_id(), None, &Vec::new(), 1000000, TEST_FINAL_CLTV).unwrap();
+       send_along_route_with_hash(&nodes[1], route, &[&nodes[2], &nodes[3], &nodes[5]], 1000000, payment_hash_6);
+
+       // Double-check that six of the new HTLC were added
+       // We now have six HTLCs pending over the dust limit and six HTLCs under the dust limit (ie,
+       // with to_local and to_remote outputs, 8 outputs and 6 HTLCs not included).
+       assert_eq!(nodes[3].node.channel_state.lock().unwrap().by_id.get(&chan.2).unwrap().last_local_commitment_txn.len(), 1);
+       assert_eq!(nodes[3].node.channel_state.lock().unwrap().by_id.get(&chan.2).unwrap().last_local_commitment_txn[0].output.len(), 8);
+
+       // Now fail back three of the over-dust-limit and three of the under-dust-limit payments in one go.
+       assert!(nodes[4].node.fail_htlc_backwards(&payment_hash_1, ds_dust_limit*1000));
+       assert!(nodes[4].node.fail_htlc_backwards(&payment_hash_3, 1000000));
+       assert!(nodes[4].node.fail_htlc_backwards(&payment_hash_5, 1000000));
+       assert!(nodes[4].node.fail_htlc_backwards(&payment_hash_6, ds_dust_limit*1000));
+       check_added_monitors!(nodes[4], 0);
+       expect_pending_htlcs_forwardable!(nodes[4]);
+       check_added_monitors!(nodes[4], 1);
+
+       let four_removes = get_htlc_update_msgs!(nodes[4], nodes[3].node.get_our_node_id());
+       nodes[3].node.handle_update_fail_htlc(&nodes[4].node.get_our_node_id(), &four_removes.update_fail_htlcs[0]).unwrap();
+       nodes[3].node.handle_update_fail_htlc(&nodes[4].node.get_our_node_id(), &four_removes.update_fail_htlcs[1]).unwrap();
+       nodes[3].node.handle_update_fail_htlc(&nodes[4].node.get_our_node_id(), &four_removes.update_fail_htlcs[2]).unwrap();
+       nodes[3].node.handle_update_fail_htlc(&nodes[4].node.get_our_node_id(), &four_removes.update_fail_htlcs[3]).unwrap();
+       commitment_signed_dance!(nodes[3], nodes[4], four_removes.commitment_signed, false);
+
+       assert!(nodes[5].node.fail_htlc_backwards(&payment_hash_2, ds_dust_limit*1000));
+       assert!(nodes[5].node.fail_htlc_backwards(&payment_hash_4, 1000000));
+       check_added_monitors!(nodes[5], 0);
+       expect_pending_htlcs_forwardable!(nodes[5]);
+       check_added_monitors!(nodes[5], 1);
+
+       let two_removes = get_htlc_update_msgs!(nodes[5], nodes[3].node.get_our_node_id());
+       nodes[3].node.handle_update_fail_htlc(&nodes[5].node.get_our_node_id(), &two_removes.update_fail_htlcs[0]).unwrap();
+       nodes[3].node.handle_update_fail_htlc(&nodes[5].node.get_our_node_id(), &two_removes.update_fail_htlcs[1]).unwrap();
+       commitment_signed_dance!(nodes[3], nodes[5], two_removes.commitment_signed, false);
+
+       let ds_prev_commitment_tx = nodes[3].node.channel_state.lock().unwrap().by_id.get(&chan.2).unwrap().last_local_commitment_txn.clone();
+
+       expect_pending_htlcs_forwardable!(nodes[3]);
+       check_added_monitors!(nodes[3], 1);
+       let six_removes = get_htlc_update_msgs!(nodes[3], nodes[2].node.get_our_node_id());
+       nodes[2].node.handle_update_fail_htlc(&nodes[3].node.get_our_node_id(), &six_removes.update_fail_htlcs[0]).unwrap();
+       nodes[2].node.handle_update_fail_htlc(&nodes[3].node.get_our_node_id(), &six_removes.update_fail_htlcs[1]).unwrap();
+       nodes[2].node.handle_update_fail_htlc(&nodes[3].node.get_our_node_id(), &six_removes.update_fail_htlcs[2]).unwrap();
+       nodes[2].node.handle_update_fail_htlc(&nodes[3].node.get_our_node_id(), &six_removes.update_fail_htlcs[3]).unwrap();
+       nodes[2].node.handle_update_fail_htlc(&nodes[3].node.get_our_node_id(), &six_removes.update_fail_htlcs[4]).unwrap();
+       nodes[2].node.handle_update_fail_htlc(&nodes[3].node.get_our_node_id(), &six_removes.update_fail_htlcs[5]).unwrap();
+       if deliver_last_raa {
+               commitment_signed_dance!(nodes[2], nodes[3], six_removes.commitment_signed, false);
+       } else {
+               let _cs_last_raa = commitment_signed_dance!(nodes[2], nodes[3], six_removes.commitment_signed, false, true, false, true);
+       }
+
+       // D's latest commitment transaction now contains 1st + 2nd + 9th HTLCs (implicitly, they're
+       // below the dust limit) and the 5th + 6th + 11th HTLCs. It has failed back the 0th, 3rd, 4th,
+       // 7th, 8th, and 10th, but as we haven't yet delivered the final RAA to C, the fails haven't
+       // propagated back to A/B yet (and D has two unrevoked commitment transactions).
+       //
+       // We now broadcast the latest commitment transaction, which *should* result in failures for
+       // the 0th, 1st, 2nd, 3rd, 4th, 7th, 8th, 9th, and 10th HTLCs, ie all the below-dust HTLCs and
+       // the non-broadcast above-dust HTLCs.
+       //
+       // Alternatively, we may broadcast the previous commitment transaction, which should only
+       // result in failures for the below-dust HTLCs, ie the 0th, 1st, 2nd, 3rd, 9th, and 10th HTLCs.
+       let ds_last_commitment_tx = nodes[3].node.channel_state.lock().unwrap().by_id.get(&chan.2).unwrap().last_local_commitment_txn.clone();
+
+       let header = BlockHeader { version: 0x20000000, prev_blockhash: Default::default(), merkle_root: Default::default(), time: 42, bits: 42, nonce: 42 };
+       if announce_latest {
+               nodes[2].chain_monitor.block_connected_checked(&header, 1, &[&ds_last_commitment_tx[0]], &[1; 1]);
+       } else {
+               nodes[2].chain_monitor.block_connected_checked(&header, 1, &[&ds_prev_commitment_tx[0]], &[1; 1]);
+       }
+       check_closed_broadcast!(nodes[2]);
+       expect_pending_htlcs_forwardable!(nodes[2]);
+       check_added_monitors!(nodes[2], 2);
+
+       let cs_msgs = nodes[2].node.get_and_clear_pending_msg_events();
+       assert_eq!(cs_msgs.len(), 2);
+       let mut a_done = false;
+       for msg in cs_msgs {
+               match msg {
+                       MessageSendEvent::UpdateHTLCs { ref node_id, ref updates } => {
+                               // Both under-dust HTLCs and the one above-dust HTLC that we had already failed
+                               // should be failed-backwards here.
+                               let target = if *node_id == nodes[0].node.get_our_node_id() {
+                                       assert_eq!(updates.update_fail_htlcs.len(), if announce_latest { 5 } else { 3 });
+                                       assert!(!a_done);
+                                       a_done = true;
+                                       &nodes[0]
+                               } else {
+                                       assert_eq!(*node_id, nodes[1].node.get_our_node_id());
+                                       assert_eq!(updates.update_fail_htlcs.len(), if announce_latest { 4 } else { 3 });
+                                       &nodes[1]
+                               };
+                               target.node.handle_update_fail_htlc(&nodes[2].node.get_our_node_id(), &updates.update_fail_htlcs[0]).unwrap();
+                               target.node.handle_update_fail_htlc(&nodes[2].node.get_our_node_id(), &updates.update_fail_htlcs[1]).unwrap();
+                               target.node.handle_update_fail_htlc(&nodes[2].node.get_our_node_id(), &updates.update_fail_htlcs[2]).unwrap();
+                               if announce_latest {
+                                       target.node.handle_update_fail_htlc(&nodes[2].node.get_our_node_id(), &updates.update_fail_htlcs[3]).unwrap();
+                                       if *node_id == nodes[0].node.get_our_node_id() {
+                                               target.node.handle_update_fail_htlc(&nodes[2].node.get_our_node_id(), &updates.update_fail_htlcs[4]).unwrap();
+                                       }
+                               }
+                               commitment_signed_dance!(target, nodes[2], updates.commitment_signed, false, true);
+                       },
+                       _ => panic!("Unexpected event"),
+               }
+       }
+
+       let as_events = nodes[0].node.get_and_clear_pending_events();
+       assert_eq!(as_events.len(), if announce_latest { 5 } else { 3 });
+       let mut as_failds = HashSet::new();
+       for event in as_events.iter() {
+               if let &Event::PaymentFailed { ref payment_hash, ref rejected_by_dest, .. } = event {
+                       assert!(as_failds.insert(*payment_hash));
+                       if *payment_hash != payment_hash_2 {
+                               assert_eq!(*rejected_by_dest, deliver_last_raa);
+                       } else {
+                               assert!(!rejected_by_dest);
+                       }
+               } else { panic!("Unexpected event"); }
+       }
+       assert!(as_failds.contains(&payment_hash_1));
+       assert!(as_failds.contains(&payment_hash_2));
+       if announce_latest {
+               assert!(as_failds.contains(&payment_hash_3));
+               assert!(as_failds.contains(&payment_hash_5));
+       }
+       assert!(as_failds.contains(&payment_hash_6));
+
+       let bs_events = nodes[1].node.get_and_clear_pending_events();
+       assert_eq!(bs_events.len(), if announce_latest { 4 } else { 3 });
+       let mut bs_failds = HashSet::new();
+       for event in bs_events.iter() {
+               if let &Event::PaymentFailed { ref payment_hash, ref rejected_by_dest, .. } = event {
+                       assert!(bs_failds.insert(*payment_hash));
+                       if *payment_hash != payment_hash_1 && *payment_hash != payment_hash_5 {
+                               assert_eq!(*rejected_by_dest, deliver_last_raa);
+                       } else {
+                               assert!(!rejected_by_dest);
+                       }
+               } else { panic!("Unexpected event"); }
+       }
+       assert!(bs_failds.contains(&payment_hash_1));
+       assert!(bs_failds.contains(&payment_hash_2));
+       if announce_latest {
+               assert!(bs_failds.contains(&payment_hash_4));
+       }
+       assert!(bs_failds.contains(&payment_hash_5));
+
+       // For each HTLC which was not failed-back by normal process (ie deliver_last_raa), we should
+       // get a PaymentFailureNetworkUpdate. A should have gotten 4 HTLCs which were failed-back due
+       // to unknown-preimage-etc, B should have gotten 2. Thus, in the
+       // announce_latest && deliver_last_raa case, we should have 5-4=1 and 4-2=2
+       // PaymentFailureNetworkUpdates.
+       let as_msg_events = nodes[0].node.get_and_clear_pending_msg_events();
+       assert_eq!(as_msg_events.len(), if deliver_last_raa { 1 } else if !announce_latest { 3 } else { 5 });
+       let bs_msg_events = nodes[1].node.get_and_clear_pending_msg_events();
+       assert_eq!(bs_msg_events.len(), if deliver_last_raa { 2 } else if !announce_latest { 3 } else { 4 });
+       for event in as_msg_events.iter().chain(bs_msg_events.iter()) {
+               match event {
+                       &MessageSendEvent::PaymentFailureNetworkUpdate { .. } => {},
+                       _ => panic!("Unexpected event"),
+               }
+       }
+}
+
+#[test]
+fn test_fail_backwards_latest_remote_announce_a() {
+       do_test_fail_backwards_unrevoked_remote_announce(false, true);
+}
+
+#[test]
+fn test_fail_backwards_latest_remote_announce_b() {
+       do_test_fail_backwards_unrevoked_remote_announce(true, true);
+}
+
+#[test]
+fn test_fail_backwards_previous_remote_announce() {
+       do_test_fail_backwards_unrevoked_remote_announce(false, false);
+       // Note that true, true doesn't make sense as it implies we announce a revoked state, which is
+       // tested for in test_commitment_revoked_fail_backward_exhaustive()
+}
+
 #[test]
 fn test_dynamic_spendable_outputs_local_htlc_timeout_tx() {
        let nodes = create_network(2);