Skip to content

Commit 2c8a26c

Browse files
TheBlueMattTheBlueMatt
committed
Don't per_peer_state read locks recursively in monitor updating
When handling a `ChannelMonitor` update via the new `handle_new_monitor_update` macro, we always call the macro with the `per_peer_state` read lock held and have the macro drop the per-peer state lock. Then, when handling the resulting updates, we may take the `per_peer_state` read lock again in another function. In a coming commit, recursive read locks will be disallowed, so we have to drop the `per_peer_state` read lock before calling additional functions in `handle_new_monitor_update`, which we do here.
1 parent 38374dd commit 2c8a26c

File tree

1 file changed

+55
-46
lines changed

1 file changed

+55
-46
lines changed

‎lightning/src/ln/channelmanager.rs‎

Lines changed: 55 additions & 46 deletions
Original file line numberDiff line numberDiff line change
@@ -1417,7 +1417,7 @@ macro_rules! emit_channel_ready_event {
14171417
}
14181418

14191419
macro_rules! handle_monitor_update_completion {
1420-
($self: ident, $update_id: expr, $peer_state_lock: expr, $peer_state: expr, $chan: expr) => { {
1420+
($self: ident, $update_id: expr, $peer_state_lock: expr, $peer_state: expr, $per_peer_state_lock: expr, $chan: expr) => { {
14211421
let mut updates = $chan.monitor_updating_restored(&$self.logger,
14221422
&$self.node_signer, $self.genesis_hash, &$self.default_configuration,
14231423
$self.best_block.read().unwrap().height());
@@ -1450,6 +1450,7 @@ macro_rules! handle_monitor_update_completion {
14501450

14511451
let channel_id = $chan.channel_id();
14521452
core::mem::drop($peer_state_lock);
1453+
core::mem::drop($per_peer_state_lock);
14531454

14541455
$self.handle_monitor_update_completion_actions(update_actions);
14551456

@@ -1465,7 +1466,7 @@ macro_rules! handle_monitor_update_completion {
14651466
}
14661467

14671468
macro_rules! handle_new_monitor_update {
1468-
($self: ident, $update_res: expr, $update_id: expr, $peer_state_lock: expr, $peer_state: expr, $chan: expr, MANUALLY_REMOVING, $remove: expr) => { {
1469+
($self: ident, $update_res: expr, $update_id: expr, $peer_state_lock: expr, $peer_state: expr, $per_peer_state_lock: expr, $chan: expr, MANUALLY_REMOVING, $remove: expr) => { {
14691470
// update_maps_on_chan_removal needs to be able to take id_to_peer, so make sure we can in
14701471
// any case so that it won't deadlock.
14711472
debug_assert!($self.id_to_peer.try_lock().is_ok());
@@ -1492,14 +1493,14 @@ macro_rules! handle_new_monitor_update {
14921493
.update_id == $update_id) &&
14931494
$chan.get_latest_monitor_update_id() == $update_id
14941495
{
1495-
handle_monitor_update_completion!($self, $update_id, $peer_state_lock, $peer_state, $chan);
1496+
handle_monitor_update_completion!($self, $update_id, $peer_state_lock, $peer_state, $per_peer_state_lock, $chan);
14961497
}
14971498
Ok(())
14981499
},
14991500
}
15001501
} };
1501-
($self: ident, $update_res: expr, $update_id: expr, $peer_state_lock: expr, $peer_state: expr, $chan_entry: expr) => {
1502-
handle_new_monitor_update!($self, $update_res, $update_id, $peer_state_lock, $peer_state, $chan_entry.get_mut(), MANUALLY_REMOVING, $chan_entry.remove_entry())
1502+
($self: ident, $update_res: expr, $update_id: expr, $peer_state_lock: expr, $peer_state: expr, $per_peer_state_lock: expr, $chan_entry: expr) => {
1503+
handle_new_monitor_update!($self, $update_res, $update_id, $peer_state_lock, $peer_state, $per_peer_state_lock, $chan_entry.get_mut(), MANUALLY_REMOVING, $chan_entry.remove_entry())
15031504
}
15041505
}
15051506

@@ -1835,7 +1836,7 @@ where
18351836
if let Some(monitor_update) = monitor_update_opt.take() {
18361837
let update_id = monitor_update.update_id;
18371838
let update_res = self.chain_monitor.update_channel(funding_txo_opt.unwrap(), monitor_update);
1838-
break handle_new_monitor_update!(self, update_res, update_id, peer_state_lock, peer_state, chan_entry);
1839+
break handle_new_monitor_update!(self, update_res, update_id, peer_state_lock, peer_state, per_peer_state, chan_entry);
18391840
}
18401841

18411842
if chan_entry.get().is_shutdown() {
@@ -2464,7 +2465,7 @@ where
24642465
Some(monitor_update) => {
24652466
let update_id = monitor_update.update_id;
24662467
let update_res = self.chain_monitor.update_channel(funding_txo, monitor_update);
2467-
if let Err(e) = handle_new_monitor_update!(self, update_res, update_id, peer_state_lock, peer_state, chan) {
2468+
if let Err(e) = handle_new_monitor_update!(self, update_res, update_id, peer_state_lock, peer_state, per_peer_state, chan) {
24682469
break Err(e);
24692470
}
24702471
if update_res == ChannelMonitorUpdateStatus::InProgress {
@@ -3991,7 +3992,8 @@ where
39913992
)
39923993
).unwrap_or(None);
39933994

3994-
if let Some(mut peer_state_lock) = peer_state_opt.take() {
3995+
if peer_state_opt.is_some() {
3996+
let mut peer_state_lock = peer_state_opt.unwrap();
39953997
let peer_state = &mut *peer_state_lock;
39963998
if let hash_map::Entry::Occupied(mut chan) = peer_state.channel_by_id.entry(chan_id) {
39973999
let counterparty_node_id = chan.get().get_counterparty_node_id();
@@ -4006,7 +4008,7 @@ where
40064008
let update_id = monitor_update.update_id;
40074009
let update_res = self.chain_monitor.update_channel(prev_hop.outpoint, monitor_update);
40084010
let res = handle_new_monitor_update!(self, update_res, update_id, peer_state_lock,
4009-
peer_state, chan);
4011+
peer_state, per_peer_state, chan);
40104012
if let Err(e) = res {
40114013
// TODO: This is a *critical* error - we probably updated the outbound edge
40124014
// of the HTLC's monitor with a preimage. We should retry this monitor
@@ -4207,7 +4209,7 @@ where
42074209
if !channel.get().is_awaiting_monitor_update() || channel.get().get_latest_monitor_update_id() != highest_applied_update_id {
42084210
return;
42094211
}
4210-
handle_monitor_update_completion!(self, highest_applied_update_id, peer_state_lock, peer_state, channel.get_mut());
4212+
handle_monitor_update_completion!(self, highest_applied_update_id, peer_state_lock, peer_state, per_peer_state, channel.get_mut());
42114213
}
42124214

42134215
/// Accepts a request to open a channel after a [`Event::OpenChannelRequest`].
@@ -4513,7 +4515,8 @@ where
45134515
let monitor_res = self.chain_monitor.watch_channel(monitor.get_funding_txo().0, monitor);
45144516

45154517
let chan = e.insert(chan);
4516-
let mut res = handle_new_monitor_update!(self, monitor_res, 0, peer_state_lock, peer_state, chan, MANUALLY_REMOVING, { peer_state.channel_by_id.remove(&new_channel_id) });
4518+
let mut res = handle_new_monitor_update!(self, monitor_res, 0, peer_state_lock, peer_state,
4519+
per_peer_state, chan, MANUALLY_REMOVING, { peer_state.channel_by_id.remove(&new_channel_id) });
45174520

45184521
// Note that we reply with the new channel_id in error messages if we gave up on the
45194522
// channel, not the temporary_channel_id. This is compatible with ourselves, but the
@@ -4546,7 +4549,7 @@ where
45464549
let monitor = try_chan_entry!(self,
45474550
chan.get_mut().funding_signed(&msg, best_block, &self.signer_provider, &self.logger), chan);
45484551
let update_res = self.chain_monitor.watch_channel(chan.get().get_funding_txo().unwrap(), monitor);
4549-
let mut res = handle_new_monitor_update!(self, update_res, 0, peer_state_lock, peer_state, chan);
4552+
let mut res = handle_new_monitor_update!(self, update_res, 0, peer_state_lock, peer_state, per_peer_state, chan);
45504553
if let Err(MsgHandleErrInternal { ref mut shutdown_finish, .. }) = res {
45514554
// We weren't able to watch the channel to begin with, so no updates should be made on
45524555
// it. Previously, full_stack_target found an (unreachable) panic when the
@@ -4642,7 +4645,7 @@ where
46424645
if let Some(monitor_update) = monitor_update_opt {
46434646
let update_id = monitor_update.update_id;
46444647
let update_res = self.chain_monitor.update_channel(funding_txo_opt.unwrap(), monitor_update);
4645-
break handle_new_monitor_update!(self, update_res, update_id, peer_state_lock, peer_state, chan_entry);
4648+
break handle_new_monitor_update!(self, update_res, update_id, peer_state_lock, peer_state, per_peer_state, chan_entry);
46464649
}
46474650
break Ok(());
46484651
},
@@ -4834,7 +4837,7 @@ where
48344837
let update_res = self.chain_monitor.update_channel(funding_txo.unwrap(), monitor_update);
48354838
let update_id = monitor_update.update_id;
48364839
handle_new_monitor_update!(self, update_res, update_id, peer_state_lock,
4837-
peer_state, chan)
4840+
peer_state, per_peer_state, chan)
48384841
},
48394842
hash_map::Entry::Vacant(_) => return Err(MsgHandleErrInternal::send_err_msg_no_close(format!("Got a message for a channel from the wrong node! No such channel for the passed counterparty_node_id {}", counterparty_node_id), msg.channel_id))
48404843
}
@@ -4940,21 +4943,20 @@ where
49404943
fn internal_revoke_and_ack(&self, counterparty_node_id: &PublicKey, msg: &msgs::RevokeAndACK) -> Result<(), MsgHandleErrInternal> {
49414944
let (htlcs_to_fail, res) = {
49424945
let per_peer_state = self.per_peer_state.read().unwrap();
4943-
let peer_state_mutex = per_peer_state.get(counterparty_node_id)
4946+
let mut peer_state_lock = per_peer_state.get(counterparty_node_id)
49444947
.ok_or_else(|| {
49454948
debug_assert!(false);
49464949
MsgHandleErrInternal::send_err_msg_no_close(format!("Can't find a peer matching the passed counterparty node_id {}", counterparty_node_id), msg.channel_id)
4947-
})?;
4948-
let mut peer_state_lock = peer_state_mutex.lock().unwrap();
4950+
}).map(|mtx| mtx.lock().unwrap())?;
49494951
let peer_state = &mut *peer_state_lock;
49504952
match peer_state.channel_by_id.entry(msg.channel_id) {
49514953
hash_map::Entry::Occupied(mut chan) => {
49524954
let funding_txo = chan.get().get_funding_txo();
49534955
let (htlcs_to_fail, monitor_update) = try_chan_entry!(self, chan.get_mut().revoke_and_ack(&msg, &self.logger), chan);
49544956
let update_res = self.chain_monitor.update_channel(funding_txo.unwrap(), monitor_update);
49554957
let update_id = monitor_update.update_id;
4956-
let res = handle_new_monitor_update!(self, update_res, update_id, peer_state_lock,
4957-
peer_state, chan);
4958+
let res = handle_new_monitor_update!(self, update_res, update_id,
4959+
peer_state_lock, peer_state, per_peer_state, chan);
49584960
(htlcs_to_fail, res)
49594961
},
49604962
hash_map::Entry::Vacant(_) => return Err(MsgHandleErrInternal::send_err_msg_no_close(format!("Got a message for a channel from the wrong node! No such channel for the passed counterparty_node_id {}", counterparty_node_id), msg.channel_id))
@@ -5211,38 +5213,45 @@ where
52115213
let mut has_monitor_update = false;
52125214
let mut failed_htlcs = Vec::new();
52135215
let mut handle_errors = Vec::new();
5214-
let per_peer_state = self.per_peer_state.read().unwrap();
52155216

5216-
for (_cp_id, peer_state_mutex) in per_peer_state.iter() {
5217-
'chan_loop: loop {
5218-
let mut peer_state_lock = peer_state_mutex.lock().unwrap();
5219-
let peer_state: &mut PeerState<_> = &mut *peer_state_lock;
5220-
for (channel_id, chan) in peer_state.channel_by_id.iter_mut() {
5221-
let counterparty_node_id = chan.get_counterparty_node_id();
5222-
let funding_txo = chan.get_funding_txo();
5223-
let (monitor_opt, holding_cell_failed_htlcs) =
5224-
chan.maybe_free_holding_cell_htlcs(&self.logger);
5225-
if !holding_cell_failed_htlcs.is_empty() {
5226-
failed_htlcs.push((holding_cell_failed_htlcs, *channel_id, counterparty_node_id));
5227-
}
5228-
if let Some(monitor_update) = monitor_opt {
5229-
has_monitor_update = true;
5230-
5231-
let update_res = self.chain_monitor.update_channel(
5232-
funding_txo.expect("channel is live"), monitor_update);
5233-
let update_id = monitor_update.update_id;
5234-
let channel_id: [u8; 32] = *channel_id;
5235-
let res = handle_new_monitor_update!(self, update_res, update_id,
5236-
peer_state_lock, peer_state, chan, MANUALLY_REMOVING,
5237-
peer_state.channel_by_id.remove(&channel_id));
5238-
if res.is_err() {
5239-
handle_errors.push((counterparty_node_id, res));
5217+
// Walk our list of channels and find any that need to update. Note that when we do find an
5218+
// update, if it includes actions that must be taken afterwards, we have to drop the
5219+
// per-peer state lock as well as the top level per_peer_state lock. Thus, we loop until we
5220+
// manage to go through all our peers without finding a single channel to update.
5221+
'peer_loop: loop {
5222+
let per_peer_state = self.per_peer_state.read().unwrap();
5223+
for (_cp_id, peer_state_mutex) in per_peer_state.iter() {
5224+
'chan_loop: loop {
5225+
let mut peer_state_lock = peer_state_mutex.lock().unwrap();
5226+
let peer_state: &mut PeerState<_> = &mut *peer_state_lock;
5227+
for (channel_id, chan) in peer_state.channel_by_id.iter_mut() {
5228+
let counterparty_node_id = chan.get_counterparty_node_id();
5229+
let funding_txo = chan.get_funding_txo();
5230+
let (monitor_opt, holding_cell_failed_htlcs) =
5231+
chan.maybe_free_holding_cell_htlcs(&self.logger);
5232+
if !holding_cell_failed_htlcs.is_empty() {
5233+
failed_htlcs.push((holding_cell_failed_htlcs, *channel_id, counterparty_node_id));
5234+
}
5235+
if let Some(monitor_update) = monitor_opt {
5236+
has_monitor_update = true;
5237+
5238+
let update_res = self.chain_monitor.update_channel(
5239+
funding_txo.expect("channel is live"), monitor_update);
5240+
let update_id = monitor_update.update_id;
5241+
let channel_id: [u8; 32] = *channel_id;
5242+
let res = handle_new_monitor_update!(self, update_res, update_id,
5243+
peer_state_lock, peer_state, per_peer_state, chan, MANUALLY_REMOVING,
5244+
peer_state.channel_by_id.remove(&channel_id));
5245+
if res.is_err() {
5246+
handle_errors.push((counterparty_node_id, res));
5247+
}
5248+
continue 'peer_loop;
52405249
}
5241-
continue 'chan_loop;
52425250
}
5251+
break 'chan_loop;
52435252
}
5244-
break 'chan_loop;
52455253
}
5254+
break 'peer_loop;
52465255
}
52475256

52485257
let has_update = has_monitor_update || !failed_htlcs.is_empty() || !handle_errors.is_empty();

0 commit comments

Comments
 (0)