Skip to content

Commit 951174a

Browse files
authored
Merge pull request #3196 from TheBlueMatt/2024-07-monitor-ordering
Ensure `ChannelMonitorUpdate`s are ordered with full monitor writes
2 parents fafa655 + 036c31c commit 951174a

File tree

1 file changed

+19
-3
lines changed

1 file changed

+19
-3
lines changed

lightning/src/chain/chainmonitor.rs

+19-3
Original file line numberDiff line numberDiff line change
@@ -167,10 +167,17 @@ struct MonitorHolder<ChannelSigner: EcdsaChannelSigner> {
167167
monitor: ChannelMonitor<ChannelSigner>,
168168
/// The full set of pending monitor updates for this Channel.
169169
///
170-
/// Note that this lock must be held during updates to prevent a race where we call
171-
/// update_persisted_channel, the user returns a
170+
/// Note that this lock must be held from [`ChannelMonitor::update_monitor`] through to
171+
/// [`Persist::update_persisted_channel`] to prevent a race where we call
172+
/// [`Persist::update_persisted_channel`], the user returns a
172173
/// [`ChannelMonitorUpdateStatus::InProgress`], and then calls channel_monitor_updated
173174
/// immediately, racing our insertion of the pending update into the contained Vec.
175+
///
176+
/// This also avoids a race where we update a [`ChannelMonitor`], then while connecting a block
177+
/// persist a full [`ChannelMonitor`] prior to persisting the [`ChannelMonitorUpdate`]. This
178+
/// could cause users to have a full [`ChannelMonitor`] on disk as well as a
179+
/// [`ChannelMonitorUpdate`] which was already applied. While this isn't an issue for the
180+
/// LDK-provided update-based [`Persist`], its somewhat surprising for users so we avoid it.
174181
pending_monitor_updates: Mutex<Vec<u64>>,
175182
}
176183

@@ -322,6 +329,11 @@ where C::Target: chain::Filter,
322329
let has_pending_claims = monitor_state.monitor.has_pending_claims();
323330
if has_pending_claims || get_partition_key(funding_outpoint) % partition_factor == 0 {
324331
log_trace!(logger, "Syncing Channel Monitor for channel {}", log_funding_info!(monitor));
332+
// Even though we don't track monitor updates from chain-sync as pending, we still want
333+
// updates per-channel to be well-ordered so that users don't see a
334+
// `ChannelMonitorUpdate` after a channel persist for a channel with the same
335+
// `latest_update_id`.
336+
let _pending_monitor_updates = monitor_state.pending_monitor_updates.lock().unwrap();
325337
match self.persister.update_persisted_channel(*funding_outpoint, None, monitor) {
326338
ChannelMonitorUpdateStatus::Completed =>
327339
log_trace!(logger, "Finished syncing Channel Monitor for channel {} for block-data",
@@ -796,10 +808,14 @@ where C::Target: chain::Filter,
796808
let monitor = &monitor_state.monitor;
797809
let logger = WithChannelMonitor::from(&self.logger, &monitor, None);
798810
log_trace!(logger, "Updating ChannelMonitor to id {} for channel {}", update.update_id, log_funding_info!(monitor));
811+
812+
// We hold a `pending_monitor_updates` lock through `update_monitor` to ensure we
813+
// have well-ordered updates from the users' point of view. See the
814+
// `pending_monitor_updates` docs for more.
815+
let mut pending_monitor_updates = monitor_state.pending_monitor_updates.lock().unwrap();
799816
let update_res = monitor.update_monitor(update, &self.broadcaster, &self.fee_estimator, &self.logger);
800817

801818
let update_id = update.update_id;
802-
let mut pending_monitor_updates = monitor_state.pending_monitor_updates.lock().unwrap();
803819
let persist_res = if update_res.is_err() {
804820
// Even if updating the monitor returns an error, the monitor's state will
805821
// still be changed. Therefore, we should persist the updated monitor despite the error.

0 commit comments

Comments
 (0)