Check for timing-out HTLCs in remote unrevoked commitments
authorMatt Corallo <git@bluematt.me>
Fri, 4 Jan 2019 19:38:05 +0000 (14:38 -0500)
committerMatt Corallo <git@bluematt.me>
Sun, 13 Jan 2019 17:59:19 +0000 (12:59 -0500)
This resolves a TODO/issue in would_broadcast_at_height where we
will not fail a channel with HTLCs which time out in remote
broadcastable transactions.

src/ln/channelmonitor.rs
src/ln/functional_tests.rs

index 546e4a68451cc076c440ca9b2ac0902233d2e8ef..c71800e594acde70ce4e2df59b86b2e19bb27d72 100644 (file)
@@ -1568,6 +1568,7 @@ impl ChannelMonitor {
                        if let Some(transaction_output_index) = htlc.transaction_output_index {
                                if let &Some((ref their_sig, ref our_sig)) = sigs {
                                        if htlc.offered {
+                                               log_trace!(self, "Broadcasting HTLC-Timeout transaction against local commitment transactions");
                                                let mut htlc_timeout_tx = chan_utils::build_htlc_transaction(&local_tx.txid, local_tx.feerate_per_kw, self.their_to_self_delay.unwrap(), htlc, &local_tx.delayed_payment_key, &local_tx.revocation_key);
 
                                                htlc_timeout_tx.input[0].witness.push(Vec::new()); // First is the multisig dummy
@@ -1584,6 +1585,7 @@ impl ChannelMonitor {
                                                res.push(htlc_timeout_tx);
                                        } else {
                                                if let Some(payment_preimage) = self.payment_preimages.get(&htlc.payment_hash) {
+                                                       log_trace!(self, "Broadcasting HTLC-Success transaction against local commitment transactions");
                                                        let mut htlc_success_tx = chan_utils::build_htlc_transaction(&local_tx.txid, local_tx.feerate_per_kw, self.their_to_self_delay.unwrap(), htlc, &local_tx.delayed_payment_key, &local_tx.revocation_key);
 
                                                        htlc_success_tx.input[0].witness.push(Vec::new()); // First is the multisig dummy
@@ -1618,6 +1620,7 @@ impl ChannelMonitor {
                // weren't yet included in our commitment transaction(s).
                if let &Some(ref local_tx) = &self.current_local_signed_commitment_tx {
                        if local_tx.txid == commitment_txid {
+                               log_trace!(self, "Got latest local commitment tx broadcast, searching for available HTLCs to claim");
                                match self.key_storage {
                                        Storage::Local { ref delayed_payment_base_key, ref latest_per_commitment_point, .. } => {
                                                let (local_txn, spendable_outputs, watch_outputs) = self.broadcast_by_local_state(local_tx, latest_per_commitment_point, &Some(*delayed_payment_base_key));
@@ -1632,6 +1635,7 @@ impl ChannelMonitor {
                }
                if let &Some(ref local_tx) = &self.prev_local_signed_commitment_tx {
                        if local_tx.txid == commitment_txid {
+                               log_trace!(self, "Got previous local commitment tx broadcast, searching for available HTLCs to claim");
                                match self.key_storage {
                                        Storage::Local { ref delayed_payment_base_key, ref prev_latest_per_commitment_point, .. } => {
                                                let (local_txn, spendable_outputs, watch_outputs) = self.broadcast_by_local_state(local_tx, prev_latest_per_commitment_point, &Some(*delayed_payment_base_key));
@@ -1787,44 +1791,66 @@ impl ChannelMonitor {
        }
 
        pub(super) fn would_broadcast_at_height(&self, height: u32) -> bool {
-               // TODO: We need to consider HTLCs which weren't included in latest local commitment
-               // transaction (or in any of the latest two local commitment transactions). This probably
-               // needs to use the same logic as the revoked-tx-announe logic - checking the last two
-               // remote commitment transactions. This probably has implications for what data we need to
-               // store in local commitment transactions.
+               // We need to consider all HTLCs which are:
+               //  * in any unrevoked remote commitment transaction, as they could broadcast said
+               //    transactions and we'd end up in a race, or
+               //  * are in our latest local commitment transaction, as this is the thing we will
+               //    broadcast if we go on-chain.
                // Note that we consider HTLCs which were below dust threshold here - while they don't
                // strictly imply that we need to fail the channel, we need to go ahead and fail them back
                // to the source, and if we don't fail the channel we will have to ensure that the next
                // updates that peer sends us are update_fails, failing the channel if not. It's probably
                // easier to just fail the channel as this case should be rare enough anyway.
+               macro_rules! scan_commitment {
+                       ($htlcs: expr, $local_tx: expr) => {
+                               for ref htlc in $htlcs {
+                                       // For inbound HTLCs which we know the preimage for, we have to ensure we hit the
+                                       // chain with enough room to claim the HTLC without our counterparty being able to
+                                       // time out the HTLC first.
+                                       // For outbound HTLCs which our counterparty hasn't failed/claimed, our primary
+                                       // concern is being able to claim the corresponding inbound HTLC (on another
+                                       // channel) before it expires. In fact, we don't even really care if our
+                                       // counterparty here claims such an outbound HTLC after it expired as long as we
+                                       // can still claim the corresponding HTLC. Thus, to avoid needlessly hitting the
+                                       // chain when our counterparty is waiting for expiration to off-chain fail an HTLC
+                                       // we give ourselves a few blocks of headroom after expiration before going
+                                       // on-chain for an expired HTLC.
+                                       // Note that, to avoid a potential attack whereby a node delays claiming an HTLC
+                                       // from us until we've reached the point where we go on-chain with the
+                                       // corresponding inbound HTLC, we must ensure that outbound HTLCs go on chain at
+                                       // least CLTV_CLAIM_BUFFER blocks prior to the inbound HTLC.
+                                       //  aka outbound_cltv + HTLC_FAIL_TIMEOUT_BLOCKS == height - CLTV_CLAIM_BUFFER
+                                       //      inbound_cltv == height + CLTV_CLAIM_BUFFER
+                                       //      outbound_cltv + HTLC_FAIL_TIMEOUT_BLOCKS + CLTV_CLAIM_BUFER <= inbound_cltv - CLTV_CLAIM_BUFFER
+                                       //      HTLC_FAIL_TIMEOUT_BLOCKS + 2*CLTV_CLAIM_BUFER <= inbound_cltv - outbound_cltv
+                                       //      HTLC_FAIL_TIMEOUT_BLOCKS + 2*CLTV_CLAIM_BUFER <= CLTV_EXPIRY_DELTA
+                                       let htlc_outbound = $local_tx == htlc.offered;
+                                       if ( htlc_outbound && htlc.cltv_expiry + HTLC_FAIL_TIMEOUT_BLOCKS <= height) ||
+                                          (!htlc_outbound && htlc.cltv_expiry <= height + CLTV_CLAIM_BUFFER && self.payment_preimages.contains_key(&htlc.payment_hash)) {
+                                               log_info!(self, "Force-closing channel due to {} HTLC timeout, HTLC expiry is {}", if htlc_outbound { "outbound" } else { "inbound "}, htlc.cltv_expiry);
+                                               return true;
+                                       }
+                               }
+                       }
+               }
+
                if let Some(ref cur_local_tx) = self.current_local_signed_commitment_tx {
-                       for &(ref htlc, _, _) in cur_local_tx.htlc_outputs.iter() {
-                               // For inbound HTLCs which we know the preimage for, we have to ensure we hit the
-                               // chain with enough room to claim the HTLC without our counterparty being able to
-                               // time out the HTLC first.
-                               // For outbound HTLCs which our counterparty hasn't failed/claimed, our primary
-                               // concern is being able to claim the corresponding inbound HTLC (on another
-                               // channel) before it expires. In fact, we don't even really care if our
-                               // counterparty here claims such an outbound HTLC after it expired as long as we
-                               // can still claim the corresponding HTLC. Thus, to avoid needlessly hitting the
-                               // chain when our counterparty is waiting for expiration to off-chain fail an HTLC
-                               // we give ourselves a few blocks of headroom after expiration before going
-                               // on-chain for an expired HTLC.
-                               // Note that, to avoid a potential attack whereby a node delays claiming an HTLC
-                               // from us until we've reached the point where we go on-chain with the
-                               // corresponding inbound HTLC, we must ensure that outbound HTLCs go on chain at
-                               // least CLTV_CLAIM_BUFFER blocks prior to the inbound HTLC.
-                               //  aka outbound_cltv + HTLC_FAIL_TIMEOUT_BLOCKS == height - CLTV_CLAIM_BUFFER
-                               //      inbound_cltv == height + CLTV_CLAIM_BUFFER
-                               //      outbound_cltv + HTLC_FAIL_TIMEOUT_BLOCKS + CLTV_CLAIM_BUFER <= inbound_cltv - CLTV_CLAIM_BUFFER
-                               //      HTLC_FAIL_TIMEOUT_BLOCKS + 2*CLTV_CLAIM_BUFER <= inbound_cltv - outbound_cltv
-                               //      HTLC_FAIL_TIMEOUT_BLOCKS + 2*CLTV_CLAIM_BUFER <= CLTV_EXPIRY_DELTA
-                               if ( htlc.offered && htlc.cltv_expiry + HTLC_FAIL_TIMEOUT_BLOCKS <= height) ||
-                                  (!htlc.offered && htlc.cltv_expiry <= height + CLTV_CLAIM_BUFFER && self.payment_preimages.contains_key(&htlc.payment_hash)) {
-                                       return true;
+                       scan_commitment!(cur_local_tx.htlc_outputs.iter().map(|&(ref a, _, _)| a), true);
+               }
+
+               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 {
+                               if let Some(ref htlc_outputs) = self.remote_claimable_outpoints.get(txid) {
+                                       scan_commitment!(htlc_outputs.iter().map(|&(ref a, _)| a), false);
+                               }
+                       }
+                       if let &Some(ref txid) = prev_remote_commitment_txid {
+                               if let Some(ref htlc_outputs) = self.remote_claimable_outpoints.get(txid) {
+                                       scan_commitment!(htlc_outputs.iter().map(|&(ref a, _)| a), false);
                                }
                        }
                }
+
                false
        }
 
index ca394fcdf98fda96379630cb2d6c9024e4739f05..de2b42d3fa372e32f1aa5fdf64f051b84a9acb01 100644 (file)
@@ -6084,10 +6084,107 @@ fn do_htlc_claim_local_commitment_only(use_dust: bool) {
        check_closed_broadcast!(nodes[1]);
 }
 
+fn do_htlc_claim_current_remote_commitment_only(use_dust: bool) {
+       let mut nodes = create_network(2);
+       let chan = create_announced_chan_between_nodes(&nodes, 0, 1);
+
+       let route = nodes[0].router.get_route(&nodes[1].node.get_our_node_id(), None, &Vec::new(), if use_dust { 50000 } else { 3000000 }, TEST_FINAL_CLTV).unwrap();
+       let (_, payment_hash) = get_payment_preimage_hash!(nodes[0]);
+       nodes[0].node.send_payment(route, payment_hash).unwrap();
+       check_added_monitors!(nodes[0], 1);
+
+       let _as_update = get_htlc_update_msgs!(nodes[0], nodes[1].node.get_our_node_id());
+
+       // As far as A is concerened, the HTLC is now present only in the latest remote commitment
+       // transaction, however it is not in A's latest local commitment, so we can just broadcast that
+       // to "time out" the HTLC.
+
+       let mut header = BlockHeader { version: 0x20000000, prev_blockhash: Default::default(), merkle_root: Default::default(), time: 42, bits: 42, nonce: 42 };
+       for i in 1..TEST_FINAL_CLTV + HTLC_FAIL_TIMEOUT_BLOCKS + CHAN_CONFIRM_DEPTH + 1 {
+               nodes[0].chain_monitor.block_connected_checked(&header, i, &Vec::new(), &Vec::new());
+               header.prev_blockhash = header.bitcoin_hash();
+       }
+       test_txn_broadcast(&nodes[0], &chan, None, HTLCType::NONE);
+       check_closed_broadcast!(nodes[0]);
+}
+
+fn do_htlc_claim_previous_remote_commitment_only(use_dust: bool, check_revoke_no_close: bool) {
+       let nodes = create_network(3);
+       let chan = create_announced_chan_between_nodes(&nodes, 0, 1);
+
+       // Fail the payment, but don't deliver A's final RAA, resulting in the HTLC only being present
+       // in B's previous (unrevoked) commitment transaction, but none of A's commitment transactions.
+       // Also optionally test that we *don't* fail the channel in case the commitment transaction was
+       // actually revoked.
+       let htlc_value = if use_dust { 50000 } else { 3000000 };
+       let (_, our_payment_hash) = route_payment(&nodes[0], &[&nodes[1]], htlc_value);
+       assert!(nodes[1].node.fail_htlc_backwards(&our_payment_hash, htlc_value));
+       expect_pending_htlcs_forwardable!(nodes[1]);
+       check_added_monitors!(nodes[1], 1);
+
+       let bs_updates = get_htlc_update_msgs!(nodes[1], nodes[0].node.get_our_node_id());
+       nodes[0].node.handle_update_fail_htlc(&nodes[1].node.get_our_node_id(), &bs_updates.update_fail_htlcs[0]).unwrap();
+       nodes[0].node.handle_commitment_signed(&nodes[1].node.get_our_node_id(), &bs_updates.commitment_signed).unwrap();
+       check_added_monitors!(nodes[0], 1);
+       let as_updates = get_revoke_commit_msgs!(nodes[0], nodes[1].node.get_our_node_id());
+       nodes[1].node.handle_revoke_and_ack(&nodes[0].node.get_our_node_id(), &as_updates.0).unwrap();
+       check_added_monitors!(nodes[1], 1);
+       nodes[1].node.handle_commitment_signed(&nodes[0].node.get_our_node_id(), &as_updates.1).unwrap();
+       check_added_monitors!(nodes[1], 1);
+       let bs_revoke_and_ack = get_event_msg!(nodes[1], MessageSendEvent::SendRevokeAndACK, nodes[0].node.get_our_node_id());
+
+       if check_revoke_no_close {
+               nodes[0].node.handle_revoke_and_ack(&nodes[1].node.get_our_node_id(), &bs_revoke_and_ack).unwrap();
+               check_added_monitors!(nodes[0], 1);
+       }
+
+       let mut header = BlockHeader { version: 0x20000000, prev_blockhash: Default::default(), merkle_root: Default::default(), time: 42, bits: 42, nonce: 42 };
+       for i in 1..TEST_FINAL_CLTV + HTLC_FAIL_TIMEOUT_BLOCKS + CHAN_CONFIRM_DEPTH + 1 {
+               nodes[0].chain_monitor.block_connected_checked(&header, i, &Vec::new(), &Vec::new());
+               header.prev_blockhash = header.bitcoin_hash();
+       }
+       if !check_revoke_no_close {
+               test_txn_broadcast(&nodes[0], &chan, None, HTLCType::NONE);
+               check_closed_broadcast!(nodes[0]);
+       } else {
+               let events = nodes[0].node.get_and_clear_pending_events();
+               assert_eq!(events.len(), 1);
+               match events[0] {
+                       Event::PaymentFailed { payment_hash, rejected_by_dest, .. } => {
+                               assert_eq!(payment_hash, our_payment_hash);
+                               assert!(rejected_by_dest);
+                       },
+                       _ => panic!("Unexpected event"),
+               }
+       }
+}
+
+// Test that we close channels on-chain when broadcastable HTLCs reach their timeout window.
+// There are only a few cases to test here:
+//  * its not really normative behavior, but we test that below-dust HTLCs "included" in
+//    broadcastable commitment transactions result in channel closure,
+//  * its included in an unrevoked-but-previous remote commitment transaction,
+//  * its included in the latest remote or local commitment transactions.
+// We test each of the three possible commitment transactions individually and use both dust and
+// non-dust HTLCs.
+// Note that we don't bother testing both outbound and inbound HTLC failures for each case, and we
+// assume they are handled the same across all six cases, as both outbound and inbound failures are
+// tested for at least one of the cases in other tests.
 #[test]
-fn htlc_claim_local_commitment_only() {
+fn htlc_claim_single_commitment_only_a() {
        do_htlc_claim_local_commitment_only(true);
        do_htlc_claim_local_commitment_only(false);
+
+       do_htlc_claim_current_remote_commitment_only(true);
+       do_htlc_claim_current_remote_commitment_only(false);
+}
+
+#[test]
+fn htlc_claim_single_commitment_only_b() {
+       do_htlc_claim_previous_remote_commitment_only(true, false);
+       do_htlc_claim_previous_remote_commitment_only(false, false);
+       do_htlc_claim_previous_remote_commitment_only(true, true);
+       do_htlc_claim_previous_remote_commitment_only(false, true);
 }
 
 fn run_onion_failure_test<F1,F2>(_name: &str, test_case: u8, nodes: &Vec<Node>, route: &Route, payment_hash: &PaymentHash, callback_msg: F1, callback_node: F2, expected_retryable: bool, expected_error_code: Option<u16>, expected_channel_update: Option<HTLCFailChannelUpdate>)