fix(stateres): Correctly fetch missing auth events for incoming PDUs

src/core/matrix/state_res/mod.rs

@@ -36,7 +36,7 @@ pub use self::{
 	room_version::RoomVersion,
 };
 use crate::{
-	debug, debug_error,
+	debug, debug_error, err,
 	matrix::{Event, StateKey},
 	state_res::room_version::StateResolutionVersion,
 	trace,
@@ -319,8 +319,19 @@ where
 			path.pop();
 			continue;
 		}
-		let evt = fetch_event(event_id.clone()).await?;
+		trace!(event_id = event_id.as_str(), "fetching event for its auth events");
-		stack.push(evt.auth_events().map(ToOwned::to_owned).collect());
+		let evt = fetch_event(event_id.clone()).await;
+		if evt.is_none() {
+			err!("could not fetch event {} to calculate conflicted subgraph", event_id);
+			path.pop();
+			continue;
+		}
+		stack.push(
+			evt.expect("checked")
+				.auth_events()
+				.map(ToOwned::to_owned)
+				.collect(),
+		);
 		seen.insert(event_id);
 	}
 	Some(subgraph)

src/service/rooms/event_handler/fetch_and_handle_outliers.rs

@@ -4,9 +4,8 @@ use std::{
 };
 
 use conduwuit::{
-	Event, PduEvent, debug, debug_error, debug_warn, implement,
+	Event, PduEvent, debug, debug_warn, implement, matrix::event::gen_event_id_canonical_json,
-	matrix::event::gen_event_id_canonical_json, trace, utils::continue_exponential_backoff_secs,
+	trace, utils::continue_exponential_backoff_secs, warn,
-	warn,
 };
 use ruma::{
 	CanonicalJsonValue, EventId, OwnedEventId, RoomId, ServerName,
@@ -52,12 +51,14 @@ where
 	};
 
 	let mut events_with_auth_events = Vec::with_capacity(events.clone().count());
+	trace!("Fetching {} outlier pdus", events.clone().count());
 
 	for id in events {
 		// a. Look in the main timeline (pduid_pdu tree)
 		// b. Look at outlier pdu tree
 		// (get_pdu_json checks both)
 		if let Ok(local_pdu) = self.services.timeline.get_pdu(id).await {
+			trace!("Found {id} in main timeline or outlier tree");
 			events_with_auth_events.push((id.to_owned(), Some(local_pdu), vec![]));
 			continue;
 		}
@@ -104,7 +105,7 @@ where
 				continue;
 			}
 
-			debug!("Fetching {next_id} over federation.");
+			debug!("Fetching {next_id} over federation from {origin}.");
 			match self
 				.services
 				.sending
@@ -115,7 +116,7 @@ where
 				.await
 			{
 				| Ok(res) => {
-					debug!("Got {next_id} over federation");
+					debug!("Got {next_id} over federation from {origin}");
 					let Ok(room_version_id) = get_room_version_id(create_event) else {
 						back_off((*next_id).to_owned());
 						continue;
@@ -145,6 +146,9 @@ where
 								auth_event.clone().into(),
 							) {
 								| Ok(auth_event) => {
+									trace!(
+										"Found auth event id {auth_event} for event {next_id}"
+									);
 									todo_auth_events.push_back(auth_event);
 								},
 								| _ => {
@@ -160,7 +164,7 @@ where
 					events_all.insert(next_id);
 				},
 				| Err(e) => {
-					debug_error!("Failed to fetch event {next_id}: {e}");
+					warn!("Failed to fetch auth event {next_id} from {origin}: {e}");
 					back_off((*next_id).to_owned());
 				},
 			}
@@ -175,7 +179,7 @@ where
 		// b. Look at outlier pdu tree
 		// (get_pdu_json checks both)
 		if let Some(local_pdu) = local_pdu {
-			trace!("Found {id} in db");
+			trace!("Found {id} in main timeline or outlier tree");
 			pdus.push((local_pdu.clone(), None));
 		}
 
@@ -201,6 +205,7 @@ where
 				}
 			}
 
+			trace!("Handling outlier {next_id}");
 			match Box::pin(self.handle_outlier_pdu(
 				origin,
 				create_event,
@@ -213,6 +218,7 @@ where
 			{
 				| Ok((pdu, json)) =>
 					if next_id == *id {
+						trace!("Handled outlier {next_id} (original request)");
 						pdus.push((pdu, Some(json)));
 					},
 				| Err(e) => {
@@ -222,6 +228,6 @@ where
 			}
 		}
 	}
-
+	trace!("Fetched and handled {} outlier pdus", pdus.len());
 	pdus
 }

src/service/rooms/event_handler/handle_outlier_pdu.rs

@@ -1,11 +1,12 @@
 use std::collections::{BTreeMap, HashMap, hash_map};
 
 use conduwuit::{
-	Err, Event, PduEvent, Result, debug, debug_info, err, implement, state_res, trace, warn,
+	Err, Event, PduEvent, Result, debug, debug_info, debug_warn, err, implement, state_res, trace,
 };
 use futures::future::ready;
 use ruma::{
-	CanonicalJsonObject, CanonicalJsonValue, EventId, RoomId, ServerName, events::StateEventType,
+	CanonicalJsonObject, CanonicalJsonValue, EventId, OwnedEventId, RoomId, ServerName,
+	events::StateEventType,
 };
 
 use super::{check_room_id, get_room_version_id, to_room_version};
@@ -74,36 +75,73 @@ where
 
 	check_room_id(room_id, &pdu_event)?;
 
-	if !auth_events_known {
+	// Fetch all auth events
-		// 4. fetch any missing auth events doing all checks listed here starting at 1.
+	let mut auth_events: HashMap<OwnedEventId, PduEvent> = HashMap::new();
-		//    These are not timeline events
+
-		// 5. Reject "due to auth events" if can't get all the auth events or some of
+	for aid in pdu_event.auth_events() {
-		//    the auth events are also rejected "due to auth events"
+		if let Ok(auth_event) = self.services.timeline.get_pdu(aid).await {
-		// NOTE: Step 5 is not applied anymore because it failed too often
+			check_room_id(room_id, &auth_event)?;
-		debug!("Fetching auth events");
+			trace!("Found auth event {aid} for outlier event {event_id} locally");
-		Box::pin(self.fetch_and_handle_outliers(
+			auth_events.insert(aid.to_owned(), auth_event);
-			origin,
+		} else {
-			pdu_event.auth_events(),
+			debug_warn!("Could not find auth event {aid} for outlier event {event_id} locally");
-			create_event,
+		}
-			room_id,
+	}
-		))
+
-		.await;
+	// Fetch any missing ones & reject invalid ones
+	let missing_auth_events = if auth_events_known {
+		pdu_event
+			.auth_events()
+			.filter(|id| !auth_events.contains_key(*id))
+			.collect::<Vec<_>>()
+	} else {
+		pdu_event.auth_events().collect::<Vec<_>>()
+	};
+	if !missing_auth_events.is_empty() || !auth_events_known {
+		debug_info!(
+			"Fetching {} missing auth events for outlier event {event_id}",
+			missing_auth_events.len()
+		);
+		for (pdu, _) in self
+			.fetch_and_handle_outliers(
+				origin,
+				missing_auth_events.iter().copied(),
+				create_event,
+				room_id,
+			)
+			.await
+		{
+			auth_events.insert(pdu.event_id().to_owned(), pdu);
+		}
+	} else {
+		debug!("No missing auth events for outlier event {event_id}");
+	}
+	// reject if we are still missing some
+	let still_missing = pdu_event
+		.auth_events()
+		.filter(|id| !auth_events.contains_key(*id))
+		.collect::<Vec<_>>();
+	if !still_missing.is_empty() {
+		return Err!(Request(InvalidParam(
+			"Could not fetch all auth events for outlier event {event_id}, still missing: \
+			 {still_missing:?}"
+		)));
 	}
 
 	// 6. Reject "due to auth events" if the event doesn't pass auth based on the
 	//    auth events
 	debug!("Checking based on auth events");
+	let mut auth_events_by_key: HashMap<_, _> = HashMap::with_capacity(auth_events.len());
 	// Build map of auth events
-	let mut auth_events = HashMap::with_capacity(pdu_event.auth_events().count());
 	for id in pdu_event.auth_events() {
-		let Ok(auth_event) = self.services.timeline.get_pdu(id).await else {
+		let auth_event = auth_events
-			warn!("Could not find auth event {id}");
+			.get(id)
-			continue;
+			.expect("we just checked that we have all auth events")
-		};
+			.to_owned();
 
 		check_room_id(room_id, &auth_event)?;
 
-		match auth_events.entry((
+		match auth_events_by_key.entry((
 			auth_event.kind.to_string().into(),
 			auth_event
 				.state_key
@@ -123,7 +161,7 @@ where
 
 	// The original create event must be in the auth events
 	if !matches!(
-		auth_events.get(&(StateEventType::RoomCreate, String::new().into())),
+		auth_events_by_key.get(&(StateEventType::RoomCreate, String::new().into())),
 		Some(_) | None
 	) {
 		return Err!(Request(InvalidParam("Incoming event refers to wrong create event.")));
@@ -131,7 +169,7 @@ where
 
 	let state_fetch = |ty: &StateEventType, sk: &str| {
 		let key = (ty.to_owned(), sk.into());
-		ready(auth_events.get(&key).map(ToOwned::to_owned))
+		ready(auth_events_by_key.get(&key).map(ToOwned::to_owned))
 	};
 
 	let auth_check = state_res::event_auth::auth_check(