From e1f7d4e22703331b40df46551cecb3884f7e5179 Mon Sep 17 00:00:00 2001 From: Matt Corallo Date: Mon, 13 Jan 2020 16:10:30 -0500 Subject: [PATCH] Fix deadlock in handle_error!() when we have HTLCs to fail-back. This partially reverts 933ae3470309f21ef7537ffbcdc42070d60e1e74, though note that 933ae3470309f21ef7537ffbcdc42070d60e1e74 fixed a similar deadlock while introducing this one. If we have HTLCs to fail backwards, handle_error!() will call finish_force_close() which will attempt to lock channel_state while it is locked at the original caller. --- lightning/src/ln/channelmanager.rs | 157 ++++++++++------------------- 1 file changed, 51 insertions(+), 106 deletions(-) diff --git a/lightning/src/ln/channelmanager.rs b/lightning/src/ln/channelmanager.rs index 770ef3dbb..a62004e6e 100644 --- a/lightning/src/ln/channelmanager.rs +++ b/lightning/src/ln/channelmanager.rs @@ -493,21 +493,38 @@ pub enum PaymentSendFailure { } macro_rules! handle_error { - ($self: ident, $internal: expr, $their_node_id: expr, $locked_channel_state: expr) => { + ($self: ident, $internal: expr, $their_node_id: expr) => { match $internal { Ok(msg) => Ok(msg), Err(MsgHandleErrInternal { err, shutdown_finish }) => { + let mut channel_state = None; if let Some((shutdown_res, update_option)) = shutdown_finish { $self.finish_force_close_channel(shutdown_res); if let Some(update) = update_option { - $locked_channel_state.pending_msg_events.push(events::MessageSendEvent::BroadcastChannelUpdate { + channel_state = Some($self.channel_state.lock().unwrap()); + channel_state.as_mut().unwrap().pending_msg_events.push(events::MessageSendEvent::BroadcastChannelUpdate { msg: update }); } } + #[cfg(debug_assertions)] + { + // In testing, we always lock here to ensure there are no deadlocks where we + // were holding the lock coming into the macro but didn't catch it because we + // didn't generate an action and didn't have any HTLCs to fail backwards in the + // finish_force_close_channel. + if channel_state.is_none() { + channel_state = Some($self.channel_state.lock().unwrap()); + } + } log_error!($self, "{}", err.err); if let msgs::ErrorAction::IgnoreError = err.action { - } else { $locked_channel_state.pending_msg_events.push(events::MessageSendEvent::HandleError { node_id: $their_node_id, action: err.action.clone() }); } + } else { + if channel_state.is_none() { + channel_state = Some($self.channel_state.lock().unwrap()); + } + channel_state.as_mut().unwrap().pending_msg_events.push(events::MessageSendEvent::HandleError { node_id: $their_node_id, action: err.action.clone() }); + } // Return error in case higher-API need one Err(err) }, @@ -1260,8 +1277,8 @@ impl ChannelManager where M::T let _ = self.total_consistency_lock.read().unwrap(); - let mut channel_lock = self.channel_state.lock().unwrap(); let err: Result<(), _> = loop { + let mut channel_lock = self.channel_state.lock().unwrap(); let id = match channel_lock.short_to_id.get(&path.first().unwrap().short_channel_id) { None => check_res_push!(Err(APIError::ChannelUnavailable{err: "No channel available with first hop!"})), Some(id) => id.clone(), @@ -1311,7 +1328,7 @@ impl ChannelManager where M::T continue 'path_loop; }; - match handle_error!(self, err, path.first().unwrap().pubkey, channel_lock) { + match handle_error!(self, err, path.first().unwrap().pubkey) { Ok(_) => unreachable!(), Err(e) => { check_res_push!(Err(APIError::ChannelUnavailable { err: e.err })); @@ -1353,8 +1370,7 @@ impl ChannelManager where M::T let _ = self.total_consistency_lock.read().unwrap(); let (mut chan, msg, chan_monitor) = { - let mut channel_state = self.channel_state.lock().unwrap(); - let (res, chan) = match channel_state.by_id.remove(temporary_channel_id) { + let (res, chan) = match self.channel_state.lock().unwrap().by_id.remove(temporary_channel_id) { Some(mut chan) => { (chan.get_outbound_funding_created(funding_txo) .map_err(|e| if let ChannelError::Close(msg) = e { @@ -1364,7 +1380,7 @@ impl ChannelManager where M::T }, None => return }; - match handle_error!(self, res, chan.get_their_node_id(), channel_state) { + match handle_error!(self, res, chan.get_their_node_id()) { Ok(funding_msg) => { (chan, funding_msg.0, funding_msg.1) }, @@ -1376,12 +1392,9 @@ impl ChannelManager where M::T if let Err(e) = self.monitor.add_update_monitor(chan_monitor.get_funding_txo().unwrap(), chan_monitor) { match e { ChannelMonitorUpdateErr::PermanentFailure => { - { - let mut channel_state = self.channel_state.lock().unwrap(); - match handle_error!(self, Err(MsgHandleErrInternal::from_finish_shutdown("ChannelMonitor storage failure", *temporary_channel_id, chan.force_shutdown(), None)), chan.get_their_node_id(), channel_state) { - Err(_) => { return; }, - Ok(()) => unreachable!(), - } + match handle_error!(self, Err(MsgHandleErrInternal::from_finish_shutdown("ChannelMonitor storage failure", *temporary_channel_id, chan.force_shutdown(), None)), chan.get_their_node_id()) { + Err(_) => { return; }, + Ok(()) => unreachable!(), } }, ChannelMonitorUpdateErr::TemporaryFailure => { @@ -1599,10 +1612,8 @@ impl ChannelManager where M::T }, ChannelError::CloseDelayBroadcast { .. } => { panic!("Wait is only generated on receipt of channel_reestablish, which is handled by try_chan_entry, we don't bother to support it here"); } }; - match handle_error!(self, err, their_node_id, channel_state) { - Ok(_) => unreachable!(), - Err(_) => { continue; }, - } + handle_errors.push((their_node_id, err)); + continue; } }; if let Err(e) = self.monitor.add_update_monitor(monitor.get_funding_txo().unwrap(), monitor) { @@ -1695,11 +1706,8 @@ impl ChannelManager where M::T self.fail_htlc_backwards_internal(self.channel_state.lock().unwrap(), htlc_source, &payment_hash, failure_reason); } - if handle_errors.len() > 0 { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - for (their_node_id, err) in handle_errors.drain(..) { - let _ = handle_error!(self, err, their_node_id, channel_state_lock); - } + for (their_node_id, err) in handle_errors.drain(..) { + let _ = handle_error!(self, err, their_node_id); } if new_events.is_empty() { return } @@ -1962,7 +1970,8 @@ impl ChannelManager where M::T return; }; - let _ = handle_error!(self, err, their_node_id, channel_state_lock); + mem::drop(channel_state_lock); + let _ = handle_error!(self, err, their_node_id); } /// Gets the node_id held by this ChannelManager @@ -2693,9 +2702,9 @@ impl ChannelManager where M::T #[doc(hidden)] pub fn update_fee(&self, channel_id: [u8;32], feerate_per_kw: u64) -> Result<(), APIError> { let _ = self.total_consistency_lock.read().unwrap(); - let mut channel_state_lock = self.channel_state.lock().unwrap(); let their_node_id; let err: Result<(), _> = loop { + let mut channel_state_lock = self.channel_state.lock().unwrap(); let channel_state = &mut *channel_state_lock; match channel_state.by_id.entry(channel_id) { @@ -2734,7 +2743,7 @@ impl ChannelManager where M::T return Ok(()) }; - match handle_error!(self, err, their_node_id, channel_state_lock) { + match handle_error!(self, err, their_node_id) { Ok(_) => unreachable!(), Err(e) => { Err(APIError::APIMisuseError { err: e.err })} } @@ -2934,146 +2943,82 @@ impl ChainListener for ChannelM impl ChannelMessageHandler for ChannelManager where M::Target: ManyChannelMonitor { fn handle_open_channel(&self, their_node_id: &PublicKey, their_features: InitFeatures, msg: &msgs::OpenChannel) { let _ = self.total_consistency_lock.read().unwrap(); - let res = self.internal_open_channel(their_node_id, their_features, msg); - if res.is_err() { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - let _ = handle_error!(self, res, *their_node_id, channel_state_lock); - } + let _ = handle_error!(self, self.internal_open_channel(their_node_id, their_features, msg), *their_node_id); } fn handle_accept_channel(&self, their_node_id: &PublicKey, their_features: InitFeatures, msg: &msgs::AcceptChannel) { let _ = self.total_consistency_lock.read().unwrap(); - let res = self.internal_accept_channel(their_node_id, their_features, msg); - if res.is_err() { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - let _ = handle_error!(self, res, *their_node_id, channel_state_lock); - } + let _ = handle_error!(self, self.internal_accept_channel(their_node_id, their_features, msg), *their_node_id); } fn handle_funding_created(&self, their_node_id: &PublicKey, msg: &msgs::FundingCreated) { let _ = self.total_consistency_lock.read().unwrap(); - let res = self.internal_funding_created(their_node_id, msg); - if res.is_err() { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - let _ = handle_error!(self, res, *their_node_id, channel_state_lock); - } + let _ = handle_error!(self, self.internal_funding_created(their_node_id, msg), *their_node_id); } fn handle_funding_signed(&self, their_node_id: &PublicKey, msg: &msgs::FundingSigned) { let _ = self.total_consistency_lock.read().unwrap(); - let res = self.internal_funding_signed(their_node_id, msg); - if res.is_err() { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - let _ = handle_error!(self, res, *their_node_id, channel_state_lock); - } + let _ = handle_error!(self, self.internal_funding_signed(their_node_id, msg), *their_node_id); } fn handle_funding_locked(&self, their_node_id: &PublicKey, msg: &msgs::FundingLocked) { let _ = self.total_consistency_lock.read().unwrap(); - let res = self.internal_funding_locked(their_node_id, msg); - if res.is_err() { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - let _ = handle_error!(self, res, *their_node_id, channel_state_lock); - } + let _ = handle_error!(self, self.internal_funding_locked(their_node_id, msg), *their_node_id); } fn handle_shutdown(&self, their_node_id: &PublicKey, msg: &msgs::Shutdown) { let _ = self.total_consistency_lock.read().unwrap(); - let res = self.internal_shutdown(their_node_id, msg); - if res.is_err() { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - let _ = handle_error!(self, res, *their_node_id, channel_state_lock); - } + let _ = handle_error!(self, self.internal_shutdown(their_node_id, msg), *their_node_id); } fn handle_closing_signed(&self, their_node_id: &PublicKey, msg: &msgs::ClosingSigned) { let _ = self.total_consistency_lock.read().unwrap(); - let res = self.internal_closing_signed(their_node_id, msg); - if res.is_err() { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - let _ = handle_error!(self, res, *their_node_id, channel_state_lock); - } + let _ = handle_error!(self, self.internal_closing_signed(their_node_id, msg), *their_node_id); } fn handle_update_add_htlc(&self, their_node_id: &PublicKey, msg: &msgs::UpdateAddHTLC) { let _ = self.total_consistency_lock.read().unwrap(); - let res = self.internal_update_add_htlc(their_node_id, msg); - if res.is_err() { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - let _ = handle_error!(self, res, *their_node_id, channel_state_lock); - } + let _ = handle_error!(self, self.internal_update_add_htlc(their_node_id, msg), *their_node_id); } fn handle_update_fulfill_htlc(&self, their_node_id: &PublicKey, msg: &msgs::UpdateFulfillHTLC) { let _ = self.total_consistency_lock.read().unwrap(); - let res = self.internal_update_fulfill_htlc(their_node_id, msg); - if res.is_err() { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - let _ = handle_error!(self, res, *their_node_id, channel_state_lock); - } + let _ = handle_error!(self, self.internal_update_fulfill_htlc(their_node_id, msg), *their_node_id); } fn handle_update_fail_htlc(&self, their_node_id: &PublicKey, msg: &msgs::UpdateFailHTLC) { let _ = self.total_consistency_lock.read().unwrap(); - let res = self.internal_update_fail_htlc(their_node_id, msg); - if res.is_err() { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - let _ = handle_error!(self, res, *their_node_id, channel_state_lock); - } + let _ = handle_error!(self, self.internal_update_fail_htlc(their_node_id, msg), *their_node_id); } fn handle_update_fail_malformed_htlc(&self, their_node_id: &PublicKey, msg: &msgs::UpdateFailMalformedHTLC) { let _ = self.total_consistency_lock.read().unwrap(); - let res = self.internal_update_fail_malformed_htlc(their_node_id, msg); - if res.is_err() { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - let _ = handle_error!(self, res, *their_node_id, channel_state_lock); - } + let _ = handle_error!(self, self.internal_update_fail_malformed_htlc(their_node_id, msg), *their_node_id); } fn handle_commitment_signed(&self, their_node_id: &PublicKey, msg: &msgs::CommitmentSigned) { let _ = self.total_consistency_lock.read().unwrap(); - let res = self.internal_commitment_signed(their_node_id, msg); - if res.is_err() { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - let _ = handle_error!(self, res, *their_node_id, channel_state_lock); - } + let _ = handle_error!(self, self.internal_commitment_signed(their_node_id, msg), *their_node_id); } fn handle_revoke_and_ack(&self, their_node_id: &PublicKey, msg: &msgs::RevokeAndACK) { let _ = self.total_consistency_lock.read().unwrap(); - let res = self.internal_revoke_and_ack(their_node_id, msg); - if res.is_err() { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - let _ = handle_error!(self, res, *their_node_id, channel_state_lock); - } + let _ = handle_error!(self, self.internal_revoke_and_ack(their_node_id, msg), *their_node_id); } fn handle_update_fee(&self, their_node_id: &PublicKey, msg: &msgs::UpdateFee) { let _ = self.total_consistency_lock.read().unwrap(); - let res = self.internal_update_fee(their_node_id, msg); - if res.is_err() { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - let _ = handle_error!(self, res, *their_node_id, channel_state_lock); - } + let _ = handle_error!(self, self.internal_update_fee(their_node_id, msg), *their_node_id); } fn handle_announcement_signatures(&self, their_node_id: &PublicKey, msg: &msgs::AnnouncementSignatures) { let _ = self.total_consistency_lock.read().unwrap(); - let res = self.internal_announcement_signatures(their_node_id, msg); - if res.is_err() { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - let _ = handle_error!(self, res, *their_node_id, channel_state_lock); - } + let _ = handle_error!(self, self.internal_announcement_signatures(their_node_id, msg), *their_node_id); } fn handle_channel_reestablish(&self, their_node_id: &PublicKey, msg: &msgs::ChannelReestablish) { let _ = self.total_consistency_lock.read().unwrap(); - let res = self.internal_channel_reestablish(their_node_id, msg); - if res.is_err() { - let mut channel_state_lock = self.channel_state.lock().unwrap(); - let _ = handle_error!(self, res, *their_node_id, channel_state_lock); - } + let _ = 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) { -- 2.39.5