Fix responding to RAAs when monitor updating had already failed 2019-01-test-split-raa-flag-fix
authorMatt Corallo <git@bluematt.me>
Mon, 14 Jan 2019 22:24:52 +0000 (17:24 -0500)
committerMatt Corallo <git@bluematt.me>
Wed, 23 Jan 2019 22:51:46 +0000 (17:51 -0500)
src/ln/chanmon_update_fail_tests.rs
src/ln/channel.rs

index 1e0fe0c1f78a6c4c9ff368b2e288f4bb82d50b41..c915307385f4731f02852e035ca5bbfc68b01013 100644 (file)
@@ -1275,3 +1275,93 @@ fn monitor_failed_no_reestablish_response() {
 
        claim_payment(&nodes[0], &[&nodes[1]], payment_preimage_1);
 }
+
+#[test]
+fn first_message_on_recv_ordering() {
+       // Test that if the initial generator of a monitor-update-frozen state doesn't generate
+       // messages, we're willing to flip the order of response messages if neccessary in resposne to
+       // a commitment_signed which needs to send an RAA first.
+       // At a high level, our goal is to fail monitor updating in response to an RAA which needs no
+       // response and then handle a CS while in the failed state, requiring an RAA followed by a CS
+       // response. To do this, we start routing two payments, with the final RAA for the first being
+       // delivered while B is in AwaitingRAA, hence when we deliver the CS for the second B will
+       // have no pending response but will want to send a RAA/CS (with the updates for the second
+       // payment applied).
+       // Backported from chanmon_fail_consistency fuzz tests as it caught a bug here.
+       let mut nodes = create_network(2);
+       create_announced_chan_between_nodes(&nodes, 0, 1);
+
+       // Route the first payment outbound, holding the last RAA for B until we are set up so that we
+       // can deliver it and fail the monitor update.
+       let route = nodes[0].router.get_route(&nodes[1].node.get_our_node_id(), None, &Vec::new(), 1000000, TEST_FINAL_CLTV).unwrap();
+       let (payment_preimage_1, payment_hash_1) = get_payment_preimage_hash!(nodes[0]);
+       nodes[0].node.send_payment(route, payment_hash_1).unwrap();
+       check_added_monitors!(nodes[0], 1);
+
+       let mut events = nodes[0].node.get_and_clear_pending_msg_events();
+       assert_eq!(events.len(), 1);
+       let payment_event = SendEvent::from_event(events.pop().unwrap());
+       assert_eq!(payment_event.node_id, nodes[1].node.get_our_node_id());
+       nodes[1].node.handle_update_add_htlc(&nodes[0].node.get_our_node_id(), &payment_event.msgs[0]).unwrap();
+       nodes[1].node.handle_commitment_signed(&nodes[0].node.get_our_node_id(), &payment_event.commitment_msg).unwrap();
+       check_added_monitors!(nodes[1], 1);
+       let bs_responses = get_revoke_commit_msgs!(nodes[1], nodes[0].node.get_our_node_id());
+
+       nodes[0].node.handle_revoke_and_ack(&nodes[1].node.get_our_node_id(), &bs_responses.0).unwrap();
+       check_added_monitors!(nodes[0], 1);
+       nodes[0].node.handle_commitment_signed(&nodes[1].node.get_our_node_id(), &bs_responses.1).unwrap();
+       check_added_monitors!(nodes[0], 1);
+
+       let as_raa = get_event_msg!(nodes[0], MessageSendEvent::SendRevokeAndACK, nodes[1].node.get_our_node_id());
+
+       // Route the second payment, generating an update_add_htlc/commitment_signed
+       let route = nodes[0].router.get_route(&nodes[1].node.get_our_node_id(), None, &Vec::new(), 1000000, TEST_FINAL_CLTV).unwrap();
+       let (payment_preimage_2, payment_hash_2) = get_payment_preimage_hash!(nodes[0]);
+       nodes[0].node.send_payment(route, payment_hash_2).unwrap();
+       check_added_monitors!(nodes[0], 1);
+       let mut events = nodes[0].node.get_and_clear_pending_msg_events();
+       assert_eq!(events.len(), 1);
+       let payment_event = SendEvent::from_event(events.pop().unwrap());
+       assert_eq!(payment_event.node_id, nodes[1].node.get_our_node_id());
+
+       *nodes[1].chan_monitor.update_ret.lock().unwrap() = Err(ChannelMonitorUpdateErr::TemporaryFailure);
+
+       // Deliver the final RAA for the first payment, which does not require a response. RAAs
+       // generally require a commitment_signed, so the fact that we're expecting an opposite response
+       // to the next message also tests resetting the delivery order.
+       if let msgs::HandleError { err, action: Some(msgs::ErrorAction::IgnoreError) } = nodes[1].node.handle_revoke_and_ack(&nodes[0].node.get_our_node_id(), &as_raa).unwrap_err() {
+               assert_eq!(err, "Failed to update ChannelMonitor");
+       } else { panic!(); }
+       check_added_monitors!(nodes[1], 1);
+
+       // Now deliver the update_add_htlc/commitment_signed for the second payment, which does need an
+       // RAA/CS response, which should be generated when we call test_restore_channel_monitor (with
+       // the appropriate HTLC acceptance).
+       nodes[1].node.handle_update_add_htlc(&nodes[0].node.get_our_node_id(), &payment_event.msgs[0]).unwrap();
+       if let msgs::HandleError { err, action: Some(msgs::ErrorAction::IgnoreError) } = nodes[1].node.handle_commitment_signed(&nodes[0].node.get_our_node_id(), &payment_event.commitment_msg).unwrap_err() {
+               assert_eq!(err, "Previous monitor update failure prevented generation of RAA");
+       } else { panic!(); }
+
+       *nodes[1].chan_monitor.update_ret.lock().unwrap() = Ok(());
+       nodes[1].node.test_restore_channel_monitor();
+       check_added_monitors!(nodes[1], 1);
+
+       expect_pending_htlcs_forwardable!(nodes[1]);
+       expect_payment_received!(nodes[1], payment_hash_1, 1000000);
+
+       let bs_responses = get_revoke_commit_msgs!(nodes[1], nodes[0].node.get_our_node_id());
+       nodes[0].node.handle_revoke_and_ack(&nodes[1].node.get_our_node_id(), &bs_responses.0).unwrap();
+       check_added_monitors!(nodes[0], 1);
+       nodes[0].node.handle_commitment_signed(&nodes[1].node.get_our_node_id(), &bs_responses.1).unwrap();
+       check_added_monitors!(nodes[0], 1);
+
+       let as_raa = get_event_msg!(nodes[0], MessageSendEvent::SendRevokeAndACK, nodes[1].node.get_our_node_id());
+       nodes[1].node.handle_revoke_and_ack(&nodes[0].node.get_our_node_id(), &as_raa).unwrap();
+       check_added_monitors!(nodes[1], 1);
+
+       expect_pending_htlcs_forwardable!(nodes[1]);
+       expect_payment_received!(nodes[1], payment_hash_2, 1000000);
+
+       claim_payment(&nodes[0], &[&nodes[1]], payment_preimage_1);
+       claim_payment(&nodes[0], &[&nodes[1]], payment_preimage_2);
+}
index 2a8dd924506038bc1f1f0207aa3fd1714ee4e4f3..bb17b51975460e24cbc78dc9f4a35adcf251fc48 100644 (file)
@@ -1790,8 +1790,24 @@ impl Channel {
                self.received_commitment_while_awaiting_raa = (self.channel_state & (ChannelState::AwaitingRemoteRevoke as u32)) != 0;
 
                if (self.channel_state & ChannelState::MonitorUpdateFailed as u32) != 0 {
+                       // In case we initially failed monitor updating without requiring a response, we need
+                       // to make sure the RAA gets sent first.
+                       if !self.monitor_pending_commitment_signed {
+                               self.monitor_pending_order = Some(RAACommitmentOrder::RevokeAndACKFirst);
+                       }
                        self.monitor_pending_revoke_and_ack = true;
-                       self.monitor_pending_commitment_signed |= need_our_commitment;
+                       if need_our_commitment && (self.channel_state & (ChannelState::AwaitingRemoteRevoke as u32)) == 0 {
+                               // If we were going to send a commitment_signed after the RAA, go ahead and do all
+                               // the corresponding HTLC status updates so that get_last_commitment_update
+                               // includes the right HTLCs.
+                               // Note that this generates a monitor update that we ignore! This is OK since we
+                               // won't actually send the commitment_signed that generated the update to the other
+                               // side until the latest monitor has been pulled from us and stored.
+                               self.monitor_pending_commitment_signed = true;
+                               self.send_commitment_no_status_check()?;
+                       }
+                       // TODO: Call maybe_propose_first_closing_signed on restoration (or call it here and
+                       // re-send the message on restoration)
                        return Err(ChannelError::Ignore("Previous monitor update failure prevented generation of RAA"));
                }