]> git.bitcoin.ninja Git - rust-lightning/commitdiff
Add Channel support for monitor-update-failed pausing
authorMatt Corallo <git@bluematt.me>
Wed, 17 Oct 2018 22:19:55 +0000 (18:19 -0400)
committerMatt Corallo <git@bluematt.me>
Tue, 23 Oct 2018 20:03:30 +0000 (16:03 -0400)
src/ln/channel.rs

index b5fa5c04c05ba60b721eee5dee1cc5b7b4d2f611..6b1511387d61d0048c8e8b1b29b81378df81c1b3 100644 (file)
@@ -248,28 +248,32 @@ enum ChannelState {
        /// "disconnected" and no updates are allowed until after we've done a channel_reestablish
        /// dance.
        PeerDisconnected = (1 << 7),
+       /// Flag which is set on ChannelFunded and FundingSent indicating the user has told us they
+       /// failed to update our ChannelMonitor somewhere and we should pause sending any outbound
+       /// messages until they've managed to do so.
+       MonitorUpdateFailed = (1 << 8),
        /// Flag which implies that we have sent a commitment_signed but are awaiting the responding
        /// revoke_and_ack message. During this time period, we can't generate new commitment_signed
        /// messages as then we will be unable to determine which HTLCs they included in their
        /// revoke_and_ack implicit ACK, so instead we have to hold them away temporarily to be sent
        /// later.
        /// Flag is set on ChannelFunded.
-       AwaitingRemoteRevoke = (1 << 8),
+       AwaitingRemoteRevoke = (1 << 9),
        /// Flag which is set on ChannelFunded or FundingSent after receiving a shutdown message from
        /// the remote end. If set, they may not add any new HTLCs to the channel, and we are expected
        /// to respond with our own shutdown message when possible.
-       RemoteShutdownSent = (1 << 9),
+       RemoteShutdownSent = (1 << 10),
        /// Flag which is set on ChannelFunded or FundingSent after sending a shutdown message. At this
        /// point, we may not add any new HTLCs to the channel.
        /// TODO: Investigate some kind of timeout mechanism by which point the remote end must provide
        /// us their shutdown.
-       LocalShutdownSent = (1 << 10),
+       LocalShutdownSent = (1 << 11),
        /// We've successfully negotiated a closing_signed dance. At this point ChannelManager is about
        /// to drop us, but we store this anyway.
-       ShutdownComplete = 2048,
+       ShutdownComplete = 4096,
 }
 const BOTH_SIDES_SHUTDOWN_MASK: u32 = (ChannelState::LocalShutdownSent as u32 | ChannelState::RemoteShutdownSent as u32);
-const MULTI_STATE_FLAGS: u32 = (BOTH_SIDES_SHUTDOWN_MASK | ChannelState::PeerDisconnected as u32);
+const MULTI_STATE_FLAGS: u32 = (BOTH_SIDES_SHUTDOWN_MASK | ChannelState::PeerDisconnected as u32 | ChannelState::MonitorUpdateFailed as u32);
 
 const INITIAL_COMMITMENT_NUMBER: u64 = (1 << 48) - 1;
 
@@ -306,6 +310,12 @@ pub(super) struct Channel {
        pending_outbound_htlcs: Vec<OutboundHTLCOutput>,
        holding_cell_htlc_updates: Vec<HTLCUpdateAwaitingACK>,
 
+       monitor_pending_revoke_and_ack: bool,
+       monitor_pending_commitment_signed: bool,
+       monitor_pending_order: Option<RAACommitmentOrder>,
+       monitor_pending_forwards: Vec<(PendingForwardHTLCInfo, u64)>,
+       monitor_pending_failures: Vec<(HTLCSource, [u8; 32], HTLCFailReason)>,
+
        // pending_update_fee is filled when sending and receiving update_fee
        // For outbound channel, feerate_per_kw is updated with the value from
        // pending_update_fee when revoke_and_ack is received
@@ -509,6 +519,12 @@ impl Channel {
                        next_remote_htlc_id: 0,
                        channel_update_count: 1,
 
+                       monitor_pending_revoke_and_ack: false,
+                       monitor_pending_commitment_signed: false,
+                       monitor_pending_order: None,
+                       monitor_pending_forwards: Vec::new(),
+                       monitor_pending_failures: Vec::new(),
+
                        #[cfg(debug_assertions)]
                        max_commitment_tx_output_local: ::std::sync::Mutex::new((channel_value_satoshis * 1000 - push_msat, push_msat)),
                        #[cfg(debug_assertions)]
@@ -666,6 +682,12 @@ impl Channel {
                        next_remote_htlc_id: 0,
                        channel_update_count: 1,
 
+                       monitor_pending_revoke_and_ack: false,
+                       monitor_pending_commitment_signed: false,
+                       monitor_pending_order: None,
+                       monitor_pending_forwards: Vec::new(),
+                       monitor_pending_failures: Vec::new(),
+
                        #[cfg(debug_assertions)]
                        max_commitment_tx_output_local: ::std::sync::Mutex::new((msg.push_msat, msg.funding_satoshis * 1000 - msg.push_msat)),
                        #[cfg(debug_assertions)]
@@ -1166,7 +1188,7 @@ impl Channel {
                // can claim it even if the channel hits the chain before we see their next commitment.
                self.channel_monitor.provide_payment_preimage(&payment_hash_calc, &payment_preimage_arg);
 
-               if (self.channel_state & (ChannelState::AwaitingRemoteRevoke as u32 | ChannelState::PeerDisconnected as u32)) != 0 {
+               if (self.channel_state & (ChannelState::AwaitingRemoteRevoke as u32 | ChannelState::PeerDisconnected as u32 | ChannelState::MonitorUpdateFailed as u32)) != 0 {
                        for pending_update in self.holding_cell_htlc_updates.iter() {
                                match pending_update {
                                        &HTLCUpdateAwaitingACK::ClaimHTLC { htlc_id, .. } => {
@@ -1243,7 +1265,7 @@ impl Channel {
                }
 
                // Now update local state:
-               if (self.channel_state & (ChannelState::AwaitingRemoteRevoke as u32 | ChannelState::PeerDisconnected as u32)) != 0 {
+               if (self.channel_state & (ChannelState::AwaitingRemoteRevoke as u32 | ChannelState::PeerDisconnected as u32 | ChannelState::MonitorUpdateFailed as u32)) != 0 {
                        for pending_update in self.holding_cell_htlc_updates.iter() {
                                match pending_update {
                                        &HTLCUpdateAwaitingACK::ClaimHTLC { htlc_id, .. } => {
@@ -1461,11 +1483,13 @@ impl Channel {
                if self.channel_state & (ChannelState::PeerDisconnected as u32) == ChannelState::PeerDisconnected as u32 {
                        return Err(ChannelError::Close("Peer sent funding_locked when we needed a channel_reestablish"));
                }
-               let non_shutdown_state = self.channel_state & (!BOTH_SIDES_SHUTDOWN_MASK);
+
+               let non_shutdown_state = self.channel_state & (!MULTI_STATE_FLAGS);
+
                if non_shutdown_state == ChannelState::FundingSent as u32 {
                        self.channel_state |= ChannelState::TheirFundingLocked as u32;
                } else if non_shutdown_state == (ChannelState::FundingSent as u32 | ChannelState::OurFundingLocked as u32) {
-                       self.channel_state = ChannelState::ChannelFunded as u32 | (self.channel_state & BOTH_SIDES_SHUTDOWN_MASK);
+                       self.channel_state = ChannelState::ChannelFunded as u32 | (self.channel_state & MULTI_STATE_FLAGS);
                        self.channel_update_count += 1;
                } else if self.channel_state & (ChannelState::ChannelFunded as u32) != 0 &&
                                // Note that funding_signed/funding_created will have decremented both by 1!
@@ -1685,6 +1709,11 @@ impl Channel {
                                }
                        }
                }
+               if self.channel_state & (ChannelState::MonitorUpdateFailed as u32) == 0 {
+                       // This is a response to our post-monitor-failed unfreeze messages, so we can clear the
+                       // monitor_pending_order requirement as we won't re-send the monitor_pending messages.
+                       self.monitor_pending_order = None;
+               }
 
                self.channel_monitor.provide_latest_local_commitment_tx_info(local_commitment_tx.0, local_keys, self.feerate_per_kw, htlcs_and_sigs);
 
@@ -1708,6 +1737,12 @@ impl Channel {
                self.last_local_commitment_txn = new_local_commitment_txn;
                self.received_commitment_while_awaiting_raa = (self.channel_state & (ChannelState::AwaitingRemoteRevoke as u32)) != 0;
 
+               if (self.channel_state & ChannelState::MonitorUpdateFailed as u32) != 0 {
+                       self.monitor_pending_revoke_and_ack = true;
+                       self.monitor_pending_commitment_signed |= need_our_commitment;
+                       return Err(HandleError{err: "Previous monitor update failure prevented generation of RAA", action: Some(ErrorAction::IgnoreError)});
+               }
+
                let (our_commitment_signed, monitor_update) = if need_our_commitment && (self.channel_state & (ChannelState::AwaitingRemoteRevoke as u32)) == 0 {
                        // If we're AwaitingRemoteRevoke we can't send a new commitment here, but that's ok -
                        // we'll send one right away when we get the revoke_and_ack when we
@@ -1726,6 +1761,7 @@ impl Channel {
        /// Used to fulfill holding_cell_htlcs when we get a remote ack (or implicitly get it by them
        /// fulfilling or failing the last pending HTLC)
        fn free_holding_cell_htlcs(&mut self) -> Result<Option<(msgs::CommitmentUpdate, ChannelMonitor)>, HandleError> {
+               assert_eq!(self.channel_state & ChannelState::MonitorUpdateFailed as u32, 0);
                if self.holding_cell_htlc_updates.len() != 0 || self.holding_cell_update_fee.is_some() {
                        let mut htlc_updates = Vec::new();
                        mem::swap(&mut htlc_updates, &mut self.holding_cell_htlc_updates);
@@ -1827,6 +1863,7 @@ impl Channel {
                if self.channel_state & (ChannelState::PeerDisconnected as u32) == ChannelState::PeerDisconnected as u32 {
                        return Err(HandleError{err: "Peer sent revoke_and_ack when we needed a channel_reestablish", action: Some(msgs::ErrorAction::SendErrorMessage{msg: msgs::ErrorMessage{data: "Peer sent revoke_and_ack when we needed a channel_reestablish".to_string(), channel_id: msg.channel_id}})});
                }
+
                if let Some(their_prev_commitment_point) = self.their_prev_commitment_point {
                        if PublicKey::from_secret_key(&self.secp_ctx, &secp_call!(SecretKey::from_slice(&self.secp_ctx, &msg.per_commitment_secret), "Peer provided an invalid per_commitment_secret", self.channel_id())) != their_prev_commitment_point {
                                return Err(HandleError{err: "Got a revoke commitment secret which didn't correspond to their current pubkey", action: None});
@@ -1843,6 +1880,11 @@ impl Channel {
                self.their_cur_commitment_point = Some(msg.next_per_commitment_point);
                self.cur_remote_commitment_transaction_number -= 1;
                self.received_commitment_while_awaiting_raa = false;
+               if self.channel_state & (ChannelState::MonitorUpdateFailed as u32) == 0 {
+                       // This is a response to our post-monitor-failed unfreeze messages, so we can clear the
+                       // monitor_pending_order requirement as we won't re-send the monitor_pending messages.
+                       self.monitor_pending_order = None;
+               }
 
                let mut to_forward_infos = Vec::new();
                let mut revoked_htlcs = Vec::new();
@@ -1934,6 +1976,17 @@ impl Channel {
                        }
                }
 
+               if (self.channel_state & ChannelState::MonitorUpdateFailed as u32) == ChannelState::MonitorUpdateFailed as u32 {
+                       // We can't actually generate a new commitment transaction (incl by freeing holding
+                       // cells) while we can't update the monitor, so we just return what we have.
+                       if require_commitment {
+                               self.monitor_pending_commitment_signed = true;
+                       }
+                       self.monitor_pending_forwards.append(&mut to_forward_infos);
+                       self.monitor_pending_failures.append(&mut revoked_htlcs);
+                       return Ok((None, Vec::new(), Vec::new(), self.channel_monitor.clone()));
+               }
+
                match self.free_holding_cell_htlcs()? {
                        Some(mut commitment_update) => {
                                commitment_update.0.update_fail_htlcs.reserve(update_fail_htlcs.len());
@@ -1976,7 +2029,7 @@ impl Channel {
                        panic!("Cannot update fee until channel is fully established and we haven't started shutting down");
                }
                if !self.is_live() {
-                       panic!("Cannot update fee while peer is disconnected (ChannelManager should have caught this)");
+                       panic!("Cannot update fee while peer is disconnected/we're awaiting a monitor update (ChannelManager should have caught this)");
                }
 
                if (self.channel_state & (ChannelState::AwaitingRemoteRevoke as u32)) == (ChannelState::AwaitingRemoteRevoke as u32) {
@@ -2068,6 +2121,60 @@ impl Channel {
                outbound_drops
        }
 
+       /// Indicates that a ChannelMonitor update failed to be stored by the client and further
+       /// updates are partially paused.
+       /// This must be called immediately after the call which generated the ChannelMonitor update
+       /// which failed, with the order argument set to the type of call it represented (ie a
+       /// commitment update or a revoke_and_ack generation). The messages which were generated from
+       /// that original call must *not* have been sent to the remote end, and must instead have been
+       /// dropped. They will be regenerated when monitor_updating_restored is called.
+       pub fn monitor_update_failed(&mut self, order: RAACommitmentOrder) {
+               assert_eq!(self.channel_state & ChannelState::MonitorUpdateFailed as u32, 0);
+               match order {
+                       RAACommitmentOrder::CommitmentFirst => {
+                               self.monitor_pending_revoke_and_ack = false;
+                               self.monitor_pending_commitment_signed = true;
+                       },
+                       RAACommitmentOrder::RevokeAndACKFirst => {
+                               self.monitor_pending_revoke_and_ack = true;
+                               self.monitor_pending_commitment_signed = false;
+                       },
+               }
+               self.monitor_pending_order = Some(order);
+               self.channel_state |= ChannelState::MonitorUpdateFailed as u32;
+       }
+
+       /// Indicates that the latest ChannelMonitor update has been committed by the client
+       /// successfully and we should restore normal operation. Returns messages which should be sent
+       /// to the remote side.
+       pub fn monitor_updating_restored(&mut self) -> (Option<msgs::RevokeAndACK>, Option<msgs::CommitmentUpdate>, RAACommitmentOrder, Vec<(PendingForwardHTLCInfo, u64)>, Vec<(HTLCSource, [u8; 32], HTLCFailReason)>) {
+               assert_eq!(self.channel_state & ChannelState::MonitorUpdateFailed as u32, ChannelState::MonitorUpdateFailed as u32);
+               self.channel_state &= !(ChannelState::MonitorUpdateFailed as u32);
+
+               let mut forwards = Vec::new();
+               mem::swap(&mut forwards, &mut self.monitor_pending_forwards);
+               let mut failures = Vec::new();
+               mem::swap(&mut failures, &mut self.monitor_pending_failures);
+
+               if self.channel_state & (ChannelState::PeerDisconnected as u32) != 0 {
+                       // Leave monitor_pending_order so we can order our channel_reestablish responses
+                       self.monitor_pending_revoke_and_ack = false;
+                       self.monitor_pending_commitment_signed = false;
+                       return (None, None, RAACommitmentOrder::RevokeAndACKFirst, forwards, failures);
+               }
+
+               let raa = if self.monitor_pending_revoke_and_ack {
+                       Some(self.get_last_revoke_and_ack())
+               } else { None };
+               let commitment_update = if self.monitor_pending_commitment_signed {
+                       Some(self.get_last_commitment_update())
+               } else { None };
+
+               self.monitor_pending_revoke_and_ack = false;
+               self.monitor_pending_commitment_signed = false;
+               (raa, commitment_update, self.monitor_pending_order.clone().unwrap(), forwards, failures)
+       }
+
        pub fn update_fee(&mut self, fee_estimator: &FeeEstimator, msg: &msgs::UpdateFee) -> Result<(), ChannelError> {
                if self.channel_outbound {
                        return Err(ChannelError::Close("Non-funding remote tried to update channel fee"));
@@ -2171,7 +2278,12 @@ impl Channel {
                        // Note that if we need to repeat our FundingLocked we'll do that in the next if block.
                        None
                } else if msg.next_remote_commitment_number == (INITIAL_COMMITMENT_NUMBER - 1) - self.cur_local_commitment_transaction_number {
-                       Some(self.get_last_revoke_and_ack())
+                       if self.channel_state & (ChannelState::MonitorUpdateFailed as u32) != 0 {
+                               self.monitor_pending_revoke_and_ack = true;
+                               None
+                       } else {
+                               Some(self.get_last_revoke_and_ack())
+                       }
                } else {
                        return Err(ChannelError::Close("Peer attempted to reestablish channel with a very old local commitment transaction"));
                };
@@ -2183,6 +2295,7 @@ impl Channel {
                let our_next_remote_commitment_number = INITIAL_COMMITMENT_NUMBER - self.cur_remote_commitment_transaction_number + if (self.channel_state & ChannelState::AwaitingRemoteRevoke as u32) != 0 { 1 } else { 0 };
 
                let resend_funding_locked = if msg.next_local_commitment_number == 1 && INITIAL_COMMITMENT_NUMBER - self.cur_local_commitment_transaction_number == 1 {
+                       // We should never have to worry about MonitorUpdateFailed resending FundingLocked
                        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);
                        Some(msgs::FundingLocked {
@@ -2191,11 +2304,11 @@ impl Channel {
                        })
                } else { None };
 
-               let order = if self.received_commitment_while_awaiting_raa {
-                       RAACommitmentOrder::CommitmentFirst
-               } else {
-                       RAACommitmentOrder::RevokeAndACKFirst
-               };
+               let order = self.monitor_pending_order.clone().unwrap_or(if self.received_commitment_while_awaiting_raa {
+                               RAACommitmentOrder::CommitmentFirst
+                       } else {
+                               RAACommitmentOrder::RevokeAndACKFirst
+                       });
 
                if msg.next_local_commitment_number == our_next_remote_commitment_number {
                        if required_revoke.is_some() {
@@ -2204,7 +2317,8 @@ impl Channel {
                                log_debug!(self, "Reconnected channel {} with no loss", log_bytes!(self.channel_id()));
                        }
 
-                       if (self.channel_state & (ChannelState::AwaitingRemoteRevoke as u32)) == 0 {
+                       if (self.channel_state & (ChannelState::AwaitingRemoteRevoke as u32 | ChannelState::MonitorUpdateFailed as u32)) == 0 &&
+                                       self.monitor_pending_order.is_none() { // monitor_pending_order indicates we're waiting on a response to a unfreeze
                                // 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
@@ -2232,6 +2346,14 @@ impl Channel {
                                log_debug!(self, "Reconnected channel {} with only lost remote commitment tx", log_bytes!(self.channel_id()));
                        }
 
+                       // If monitor_pending_order is set, it must be CommitmentSigned if we have no RAA
+                       debug_assert!(self.monitor_pending_order != Some(RAACommitmentOrder::RevokeAndACKFirst) || required_revoke.is_some());
+
+                       if self.channel_state & (ChannelState::MonitorUpdateFailed as u32) != 0 {
+                               self.monitor_pending_commitment_signed = true;
+                               return Ok((resend_funding_locked, None, None, None, order));
+                       }
+
                        return Ok((resend_funding_locked, required_revoke, Some(self.get_last_commitment_update()), None, order));
                } else {
                        return Err(ChannelError::Close("Peer attempted to reestablish channel with a very old remote commitment transaction"));
@@ -2569,7 +2691,13 @@ impl Channel {
        /// is_usable() and considers things like the channel being temporarily disabled.
        /// Allowed in any state (including after shutdown)
        pub fn is_live(&self) -> bool {
-               self.is_usable() && (self.channel_state & (ChannelState::PeerDisconnected as u32) == 0)
+               self.is_usable() && (self.channel_state & (ChannelState::PeerDisconnected as u32 | ChannelState::MonitorUpdateFailed as u32) == 0)
+       }
+
+       /// Returns true if this channel has been marked as awaiting a monitor update to move forward.
+       /// Allowed in any state (including after shutdown)
+       pub fn is_awaiting_monitor_update(&self) -> bool {
+               (self.channel_state & ChannelState::MonitorUpdateFailed as u32) != 0
        }
 
        /// Returns true if funding_created was sent/received.
@@ -2883,14 +3011,14 @@ impl Channel {
                        return Err(HandleError{err: "Cannot send less than their minimum HTLC value", action: None});
                }
 
-               if (self.channel_state & (ChannelState::PeerDisconnected as u32)) == (ChannelState::PeerDisconnected as u32) {
+               if (self.channel_state & (ChannelState::PeerDisconnected as u32 | ChannelState::MonitorUpdateFailed as u32)) != 0 {
                        // Note that this should never really happen, if we're !is_live() on receipt of an
                        // incoming HTLC for relay will result in us rejecting the HTLC and we won't allow
                        // the user to send directly into a !is_live() channel. However, if we
                        // disconnected during the time the previous hop was doing the commitment dance we may
                        // end up getting here after the forwarding delay. In any case, returning an
                        // IgnoreError will get ChannelManager to do the right thing and fail backwards now.
-                       return Err(HandleError{err: "Cannot send an HTLC while disconnected", action: Some(ErrorAction::IgnoreError)});
+                       return Err(HandleError{err: "Cannot send an HTLC while disconnected/frozen for channel monitor update", action: Some(ErrorAction::IgnoreError)});
                }
 
                let (outbound_htlc_count, htlc_outbound_value_msat) = self.get_outbound_pending_htlc_stats();
@@ -2972,6 +3100,9 @@ impl Channel {
                if (self.channel_state & (ChannelState::PeerDisconnected as u32)) == (ChannelState::PeerDisconnected as u32) {
                        panic!("Cannot create commitment tx while disconnected, as send_htlc will have returned an Err so a send_commitment precondition has been violated");
                }
+               if (self.channel_state & (ChannelState::MonitorUpdateFailed as u32)) == (ChannelState::PeerDisconnected as u32) {
+                       panic!("Cannot create commitment tx while awaiting monitor update unfreeze, as send_htlc will have returned an Err so a send_commitment precondition has been violated");
+               }
                let mut have_updates = self.pending_update_fee.is_some();
                for htlc in self.pending_outbound_htlcs.iter() {
                        if let OutboundHTLCState::LocalAnnounced(_) = htlc.state {
@@ -3080,8 +3211,8 @@ impl Channel {
                        }
                }
                assert_eq!(self.channel_state & ChannelState::ShutdownComplete as u32, 0);
-               if self.channel_state & (ChannelState::PeerDisconnected as u32) == ChannelState::PeerDisconnected as u32 {
-                       return Err(APIError::ChannelUnavailable{err: "Cannot begin shutdown while peer is disconnected, maybe force-close instead?"});
+               if self.channel_state & (ChannelState::PeerDisconnected as u32 | ChannelState::MonitorUpdateFailed as u32) != 0 {
+                       return Err(APIError::ChannelUnavailable{err: "Cannot begin shutdown while peer is disconnected or we're waiting on a monitor update, maybe force-close instead?"});
                }
 
                let our_closing_script = self.get_closing_scriptpubkey();