Track peer-disconnection in Channel and handle channel_reestablish
authorMatt Corallo <git@bluematt.me>
Sat, 8 Sep 2018 20:02:46 +0000 (16:02 -0400)
committerMatt Corallo <git@bluematt.me>
Sat, 15 Sep 2018 14:53:16 +0000 (10:53 -0400)
src/ln/channel.rs
src/ln/channelmanager.rs

index 3c605859dff193385c53a50fc137a737ea92f4d8..d11447a865983fb88dbdbdcf3e7cba82dcfc3e9c 100644 (file)
@@ -1826,7 +1826,9 @@ impl Channel {
        /// HTLCs that we intended to add but haven't as we were waiting on a remote revoke.
        /// Returns the set of PendingHTLCStatuses from remote uncommitted HTLCs (which we're
        /// implicitly dropping) and the payment_hashes of HTLCs we tried to add but are dropping.
-       pub fn remove_uncommitted_htlcs(&mut self) -> Vec<(HTLCSource, [u8; 32])> {
+       /// No further message handling calls may be made until a channel_reestablish dance has
+       /// completed.
+       pub fn remove_uncommitted_htlcs_and_mark_paused(&mut self) -> Vec<(HTLCSource, [u8; 32])> {
                let mut outbound_drops = Vec::new();
 
                assert_eq!(self.channel_state & ChannelState::ShutdownComplete as u32, 0);
@@ -1835,12 +1837,14 @@ impl Channel {
                        return outbound_drops;
                }
 
+               let mut inbound_drop_count = 0;
                self.pending_inbound_htlcs.retain(|htlc| {
                        match htlc.state {
                                InboundHTLCState::RemoteAnnounced => {
                                        // They sent us an update_add_htlc but we never got the commitment_signed.
                                        // We'll tell them what commitment_signed we're expecting next and they'll drop
                                        // this HTLC accordingly
+                                       inbound_drop_count += 1;
                                        false
                                },
                                InboundHTLCState::AwaitingRemoteRevokeToAnnounce|InboundHTLCState::AwaitingAnnouncedRemoteRevoke => {
@@ -1879,6 +1883,8 @@ impl Channel {
                                &HTLCUpdateAwaitingACK::ClaimHTLC {..} | &HTLCUpdateAwaitingACK::FailHTLC {..} => true,
                        }
                });
+               self.channel_state |= ChannelState::PeerDisconnected as u32;
+               log_debug!(self, "Peer disconnection resulted in {} remote-announced HTLC drops and {} waiting-to-locally-announced HTLC drops on channel {}", outbound_drops.len(), inbound_drop_count, log_bytes!(self.channel_id()));
                outbound_drops
        }
 
@@ -1895,6 +1901,83 @@ impl Channel {
                Ok(())
        }
 
+       /// May panic if some calls other than message-handling calls (which will all Err immediately)
+       /// have been called between remove_uncommitted_htlcs_and_mark_paused and this call.
+       pub fn channel_reestablish(&mut self, msg: &msgs::ChannelReestablish) -> Result<(Option<msgs::FundingLocked>, Option<msgs::RevokeAndACK>, Option<msgs::CommitmentUpdate>, Option<ChannelMonitor>), HandleError> {
+               if self.channel_state & (ChannelState::PeerDisconnected as u32) == 0 {
+                       return Err(HandleError{err: "Peer sent a loose channel_reestablish not after reconnect", action: Some(msgs::ErrorAction::SendErrorMessage{msg: msgs::ErrorMessage{data: "Peer sent a loose channel_reestablish not after reconnect".to_string(), channel_id: msg.channel_id}})});
+               }
+
+               if msg.next_local_commitment_number == 0 || msg.next_local_commitment_number >= 0xffffffffffff ||
+                               msg.next_remote_commitment_number == 0 || msg.next_remote_commitment_number >= 0xffffffffffff {
+                       return Err(HandleError{err: "Peer send garbage channel_reestablish", action: Some(msgs::ErrorAction::SendErrorMessage{msg: msgs::ErrorMessage{data: "Peer send garbage channel_reestablish".to_string(), channel_id: msg.channel_id}})});
+               }
+
+               // Go ahead and unmark PeerDisconnected as various calls we may make check for it (and all
+               // remaining cases either succeed or ErrorMessage-fail).
+               self.channel_state &= !(ChannelState::PeerDisconnected as u32);
+
+               let mut required_revoke = None;
+               if msg.next_remote_commitment_number == 0xffffffffffff - self.cur_local_commitment_transaction_number {
+               } else if msg.next_remote_commitment_number == 0xfffffffffffe - self.cur_local_commitment_transaction_number {
+                       let next_per_commitment_point = PublicKey::from_secret_key(&self.secp_ctx, &self.build_local_commitment_secret(self.cur_local_commitment_transaction_number));
+                       let per_commitment_secret = chan_utils::build_commitment_secret(self.local_keys.commitment_seed, self.cur_local_commitment_transaction_number + 2);
+                       required_revoke = Some(msgs::RevokeAndACK {
+                               channel_id: self.channel_id,
+                               per_commitment_secret,
+                               next_per_commitment_point,
+                       });
+               } else {
+                       return Err(HandleError{err: "Peer attempted to reestablish channel with a very old local commitment transaction", action: Some(msgs::ErrorAction::SendErrorMessage{msg: msgs::ErrorMessage{data: "Peer attempted to reestablish channel with a very old remote commitment transaction".to_string(), channel_id: msg.channel_id}})});
+               }
+
+               if msg.next_local_commitment_number == 0xffffffffffff - self.cur_remote_commitment_transaction_number {
+                       if msg.next_remote_commitment_number == 0xffffffffffff - self.cur_local_commitment_transaction_number {
+                               log_debug!(self, "Reconnected channel {} with no lost commitment txn", log_bytes!(self.channel_id()));
+                               if msg.next_local_commitment_number == 1 && msg.next_remote_commitment_number == 1 {
+                                       let next_per_commitment_secret = self.build_local_commitment_secret(self.cur_local_commitment_transaction_number);
+                                       let next_per_commitment_point = PublicKey::from_secret_key(&self.secp_ctx, &next_per_commitment_secret);
+                                       return Ok((Some(msgs::FundingLocked {
+                                               channel_id: self.channel_id(),
+                                               next_per_commitment_point: next_per_commitment_point,
+                                       }), None, None, None));
+                               }
+                       }
+
+                       if (self.channel_state & (ChannelState::AwaitingRemoteRevoke as u32)) == 0 {
+                               // We're up-to-date and not waiting on a remote revoke (if we are our
+                               // channel_reestablish should result in them sending a revoke_and_ack), but we may
+                               // have received some updates while we were disconnected. Free the holding cell
+                               // now!
+                               match self.free_holding_cell_htlcs() {
+                                       Err(e) => {
+                                               if let &Some(msgs::ErrorAction::DisconnectPeer{msg: Some(_)}) = &e.action {
+                                               } else if let &Some(msgs::ErrorAction::SendErrorMessage{msg: _}) = &e.action {
+                                               } else {
+                                                       panic!("Got non-channel-failing result from free_holding_cell_htlcs");
+                                               }
+                                               return Err(e);
+                                       },
+                                       Ok(Some((commitment_update, channel_monitor))) => return Ok((None, required_revoke, Some(commitment_update), Some(channel_monitor))),
+                                       Ok(None) => return Ok((None, required_revoke, None, None)),
+                               }
+                       } else {
+                               return Ok((None, required_revoke, None, None));
+                       }
+               } else if msg.next_local_commitment_number == 0xfffffffffffe - self.cur_remote_commitment_transaction_number {
+                       return Ok((None, required_revoke,
+                                       Some(msgs::CommitmentUpdate {
+                                               update_add_htlcs: Vec::new(),
+                                               update_fulfill_htlcs: Vec::new(),
+                                               update_fail_htlcs: Vec::new(),
+                                               update_fail_malformed_htlcs: Vec::new(),
+                                               commitment_signed: self.send_commitment_no_state_update().expect("It looks like we failed to re-generate a commitment_signed we had previously sent?").0,
+                                       }), None));
+               } else {
+                       return Err(HandleError{err: "Peer attempted to reestablish channel with a very old remote commitment transaction", action: Some(msgs::ErrorAction::SendErrorMessage{msg: msgs::ErrorMessage{data: "Peer attempted to reestablish channel with a very old remote commitment transaction".to_string(), channel_id: msg.channel_id}})});
+               }
+       }
+
        pub fn shutdown(&mut self, fee_estimator: &FeeEstimator, msg: &msgs::Shutdown) -> Result<(Option<msgs::Shutdown>, Option<msgs::ClosingSigned>, Vec<(HTLCSource, [u8; 32])>), HandleError> {
                if self.channel_state & (ChannelState::PeerDisconnected as u32) == ChannelState::PeerDisconnected as u32 {
                        return Err(HandleError{err: "Peer sent shutdown when we needed a channel_reestablish", action: Some(msgs::ErrorAction::SendErrorMessage{msg: msgs::ErrorMessage{data: "Peer sent shutdown when we needed a channel_reestablish".to_string(), channel_id: msg.channel_id}})});
@@ -2167,6 +2250,11 @@ impl Channel {
                res as u32
        }
 
+       /// Returns true if we've ever received a message from the remote end for this Channel
+       pub fn have_received_message(&self) -> bool {
+               self.channel_state > (ChannelState::OurInitSent as u32)
+       }
+
        /// Returns true if this channel is fully established and not known to be closing.
        /// Allowed in any state (including after shutdown)
        pub fn is_usable(&self) -> bool {
@@ -2457,6 +2545,18 @@ impl Channel {
                Ok((msg, sig))
        }
 
+       /// May panic if called on a channel that wasn't immediately-previously
+       /// self.remove_uncommitted_htlcs_and_mark_paused()'d
+       pub fn get_channel_reestablish(&self) -> msgs::ChannelReestablish {
+               assert_eq!(self.channel_state & ChannelState::PeerDisconnected as u32, ChannelState::PeerDisconnected as u32);
+               msgs::ChannelReestablish {
+                       channel_id: self.channel_id(),
+                       next_local_commitment_number: 0xffffffffffff - self.cur_local_commitment_transaction_number,
+                       next_remote_commitment_number: 0xffffffffffff - self.cur_remote_commitment_transaction_number,
+                       data_loss_protect: None,
+               }
+       }
+
 
        // Send stuff to our remote peers:
 
index c1f09c50eaf6e219750633b7adf37fff71f576f4..87dc059a6e2744e9c346f5f84f6410792f821f47 100644 (file)
@@ -1902,7 +1902,27 @@ impl ChannelManager {
                Ok(())
        }
 
-
+       fn internal_channel_reestablish(&self, their_node_id: &PublicKey, msg: &msgs::ChannelReestablish) -> Result<(Option<msgs::FundingLocked>, Option<msgs::RevokeAndACK>, Option<msgs::CommitmentUpdate>), MsgHandleErrInternal> {
+               let (res, chan_monitor) = {
+                       let mut channel_state = self.channel_state.lock().unwrap();
+                       match channel_state.by_id.get_mut(&msg.channel_id) {
+                               Some(chan) => {
+                                       if chan.get_their_node_id() != *their_node_id {
+                                               return Err(MsgHandleErrInternal::send_err_msg_no_close("Got a message for a channel from the wrong node!", msg.channel_id));
+                                       }
+                                       let (funding_locked, revoke_and_ack, commitment_update, channel_monitor) = chan.channel_reestablish(msg).map_err(|e| MsgHandleErrInternal::from_maybe_close(e))?;
+                                       (Ok((funding_locked, revoke_and_ack, commitment_update)), channel_monitor)
+                               },
+                               None => return Err(MsgHandleErrInternal::send_err_msg_no_close("Failed to find corresponding channel", msg.channel_id))
+                       }
+               };
+               if let Some(monitor) = chan_monitor {
+                       if let Err(_e) = self.monitor.add_update_monitor(monitor.get_funding_txo().unwrap(), monitor) {
+                               unimplemented!();
+                       }
+               }
+               res
+       }
 }
 
 impl events::EventsProvider for ChannelManager {
@@ -2124,7 +2144,7 @@ impl ChannelMessageHandler for ChannelManager {
        }
 
        fn handle_channel_reestablish(&self, their_node_id: &PublicKey, msg: &msgs::ChannelReestablish) -> Result<(Option<msgs::FundingLocked>, Option<msgs::RevokeAndACK>, Option<msgs::CommitmentUpdate>), HandleError> {
-               Ok((None, None, None))
+               handle_error!(self, self.internal_channel_reestablish(their_node_id, msg), their_node_id)
        }
 
        fn peer_disconnected(&self, their_node_id: &PublicKey, no_connection_possible: bool) {
@@ -2156,7 +2176,7 @@ impl ChannelMessageHandler for ChannelManager {
                                channel_state.by_id.retain(|_, chan| {
                                        if chan.get_their_node_id() == *their_node_id {
                                                //TODO: mark channel disabled (and maybe announce such after a timeout).
-                                               let failed_adds = chan.remove_uncommitted_htlcs();
+                                               let failed_adds = chan.remove_uncommitted_htlcs_and_mark_paused();
                                                if !failed_adds.is_empty() {
                                                        let chan_update = self.get_channel_update(&chan).map(|u| u.encode_with_len()).unwrap(); // Cannot add/recv HTLCs before we have a short_id so unwrap is safe
                                                        failed_payments.push((chan_update, failed_adds));
@@ -2188,8 +2208,25 @@ impl ChannelMessageHandler for ChannelManager {
                }
        }
 
-       fn peer_connected(&self, _their_node_id: &PublicKey) -> Vec<msgs::ChannelReestablish> {
-               Vec::new()
+       fn peer_connected(&self, their_node_id: &PublicKey) -> Vec<msgs::ChannelReestablish> {
+               let mut res = Vec::new();
+               let mut channel_state = self.channel_state.lock().unwrap();
+               channel_state.by_id.retain(|_, chan| {
+                       if chan.get_their_node_id() == *their_node_id {
+                               if !chan.have_received_message() {
+                                       // If we created this (outbound) channel while we were disconnected from the
+                                       // peer we probably failed to send the open_channel message, which is now
+                                       // lost. We can't have had anything pending related to this channel, so we just
+                                       // drop it.
+                                       false
+                               } else {
+                                       res.push(chan.get_channel_reestablish());
+                                       true
+                               }
+                       } else { true }
+               });
+               //TODO: Also re-broadcast announcement_signatures
+               res
        }
 
        fn handle_error(&self, their_node_id: &PublicKey, msg: &msgs::ErrorMessage) {