feat: Use OptimisticTransactionDB for the database engine

Fixes #1131

Co-authored-by: Jade Ellis <jade@ellis.link>

.forgejo/workflows/build-debian.yml

@@ -35,6 +35,7 @@ jobs:
         uses: actions/checkout@v6
         with:
           fetch-depth: 0
+          ref: ${{ github.ref_name }}
 
       - name: Cache Cargo registry
         uses: actions/cache@v4

.forgejo/workflows/build-fedora.yml

@@ -33,6 +33,7 @@ jobs:
         uses: actions/checkout@v6
         with:
           fetch-depth: 0
+          ref: ${{ github.ref_name }}
 
 
       - name: Cache DNF packages

docs/deploying/generic.mdx

@@ -134,7 +134,7 @@ ### Example systemd Unit File
 ## Creating the Continuwuity configuration file
 
 Now you need to create the Continuwuity configuration file in
-`/etc/continuwuity/continuwuity.toml`. You can find an example configuration at
+`/etc/conduwuit/conduwuit.toml`. You can find an example configuration at
 [conduwuit-example.toml](../reference/config.mdx).
 
 **Please take a moment to read the config. You need to change at least the

src/api/client/membership/join.rs

@@ -44,6 +44,7 @@
 	rooms::{
 		state::RoomMutexGuard,
 		state_compressor::{CompressedState, HashSetCompressStateEvent},
+		timeline::pdu_fits,
 	},
 };
 
@@ -573,6 +574,13 @@ async fn join_room_by_id_helper_remote(
 					return state;
 				},
 			};
+			if !pdu_fits(&mut value.clone()) {
+				warn!(
+					"dropping incoming PDU {event_id} in room {room_id} from room join because \
+					 it exceeds 65535 bytes or is otherwise too large."
+				);
+				return state;
+			}
 			services.rooms.outlier.add_pdu_outlier(&event_id, &value);
 			if let Some(state_key) = &pdu.state_key {
 				let shortstatekey = services

src/core/config/mod.rs

@@ -1224,12 +1224,6 @@ pub struct Config {
 	#[serde(default)]
 	pub rocksdb_repair: bool,
 
-	#[serde(default)]
-	pub rocksdb_read_only: bool,
-
-	#[serde(default)]
-	pub rocksdb_secondary: bool,
-
 	/// Enables idle CPU priority for compaction thread. This is not enabled by
 	/// default to prevent compaction from falling too far behind on busy
 	/// systems.

src/core/matrix/state_res/event_auth.rs

@@ -177,11 +177,6 @@ pub async fn auth_check<E, F, Fut>(
 
 	// [synapse] do_sig_check check the event has valid signatures for member events
 
-	// TODO do_size_check is false when called by `iterative_auth_check`
-	// do_size_check is also mostly accomplished by ruma with the exception of
-	// checking event_type, state_key, and json are below a certain size (255 and
-	// 65_536 respectively)
-
 	let sender = incoming_event.sender();
 
 	// Implementation of https://spec.matrix.org/latest/rooms/v1/#authorization-rules

src/database/engine.rs

@@ -19,7 +19,7 @@
 
 use conduwuit::{Err, Result, debug, info, warn};
 use rocksdb::{
-	AsColumnFamilyRef, BoundColumnFamily, DBCommon, DBWithThreadMode, MultiThreaded,
+	AsColumnFamilyRef, BoundColumnFamily, DBCommon, MultiThreaded, OptimisticTransactionDB,
 	WaitForCompactOptions,
 };
 
@@ -33,13 +33,11 @@ pub struct Engine {
 	pub(crate) db: Db,
 	pub(crate) pool: Arc<Pool>,
 	pub(crate) ctx: Arc<Context>,
-	pub(super) read_only: bool,
-	pub(super) secondary: bool,
 	pub(crate) checksums: bool,
 	corks: AtomicU32,
 }
 
-pub(crate) type Db = DBWithThreadMode<MultiThreaded>;
+pub(crate) type Db = OptimisticTransactionDB<MultiThreaded>;
 
 impl Engine {
 	#[tracing::instrument(
@@ -129,14 +127,6 @@ pub fn current_sequence(&self) -> u64 {
 
 		sequence
 	}
-
-	#[inline]
-	#[must_use]
-	pub fn is_read_only(&self) -> bool { self.secondary || self.read_only }
-
-	#[inline]
-	#[must_use]
-	pub fn is_secondary(&self) -> bool { self.secondary }
 }
 
 impl Drop for Engine {

src/database/engine/backup.rs

@@ -12,9 +12,8 @@ pub fn backup(&self) -> Result {
 	let mut engine = self.backup_engine()?;
 	let config = &self.ctx.server.config;
 	if config.database_backups_to_keep > 0 {
-		let flush = !self.is_read_only();
 		engine
-			.create_new_backup_flush(&self.db, flush)
+			.create_new_backup_flush(&self.db, true)
 			.map_err(map_err)?;
 
 		let engine_info = engine.get_backup_info();

src/database/engine/memory_usage.rs

@@ -1,7 +1,7 @@
 use std::fmt::Write;
 
 use conduwuit::{Result, implement};
-use rocksdb::perf::get_memory_usage_stats;
+use rocksdb::perf::MemoryUsageBuilder;
 
 use super::Engine;
 use crate::or_else;
@@ -9,16 +9,21 @@
 #[implement(Engine)]
 pub fn memory_usage(&self) -> Result<String> {
 	let mut res = String::new();
-	let stats = get_memory_usage_stats(Some(&[&self.db]), Some(&[&*self.ctx.row_cache.lock()]))
-		.or_else(or_else)?;
+
+	let mut builder = MemoryUsageBuilder::new().or_else(or_else)?;
+	builder.add_db(&self.db);
+	builder.add_cache(&self.ctx.row_cache.lock());
+
+	let usage = builder.build().or_else(or_else)?;
+
 	let mibs = |input| f64::from(u32::try_from(input / 1024).unwrap_or(0)) / 1024.0;
 	writeln!(
 		res,
 		"Memory buffers: {:.2} MiB\nPending write: {:.2} MiB\nTable readers: {:.2} MiB\nRow \
 		 cache: {:.2} MiB",
-		mibs(stats.mem_table_total),
-		mibs(stats.mem_table_unflushed),
-		mibs(stats.mem_table_readers_total),
+		mibs(usage.approximate_mem_table_total()),
+		mibs(usage.approximate_mem_table_unflushed()),
+		mibs(usage.approximate_mem_table_readers_total()),
 		mibs(u64::try_from(self.ctx.row_cache.lock().get_usage())?),
 	)?;
 

src/database/engine/open.rs

@@ -35,14 +35,7 @@ pub(crate) async fn open(ctx: Arc<Context>, desc: &[Descriptor]) -> Result<Arc<S
 	}
 
 	debug!("Opening database...");
-	let db = if config.rocksdb_read_only {
-		Db::open_cf_descriptors_read_only(&db_opts, path, cfds, false)
-	} else if config.rocksdb_secondary {
-		Db::open_cf_descriptors_as_secondary(&db_opts, path, path, cfds)
-	} else {
-		Db::open_cf_descriptors(&db_opts, path, cfds)
-	}
-	.or_else(or_else)?;
+	let db = Db::open_cf_descriptors(&db_opts, path, cfds).or_else(or_else)?;
 
 	info!(
 		columns = num_cfds,
@@ -55,8 +48,6 @@ pub(crate) async fn open(ctx: Arc<Context>, desc: &[Descriptor]) -> Result<Arc<S
 		db,
 		pool: ctx.pool.clone(),
 		ctx: ctx.clone(),
-		read_only: config.rocksdb_read_only,
-		secondary: config.rocksdb_secondary,
 		checksums: config.rocksdb_checksums,
 		corks: AtomicU32::new(0),
 	}))

src/database/map/insert.rs

@@ -219,7 +219,7 @@ pub fn insert_batch<'a, I, K, V>(&'a self, iter: I)
 	K: AsRef<[u8]> + Sized + Debug + 'a,
 	V: AsRef<[u8]> + Sized + 'a,
 {
-	let mut batch = WriteBatchWithTransaction::<false>::default();
+	let mut batch = WriteBatchWithTransaction::<true>::default();
 	for (key, val) in iter {
 		batch.put_cf(&self.cf(), key.as_ref(), val.as_ref());
 	}

src/database/mod.rs

@@ -77,14 +77,6 @@ pub fn iter(&self) -> impl Iterator<Item = (&MapsKey, &MapsVal)> + Send + '_ {
 
 	#[inline]
 	pub fn keys(&self) -> impl Iterator<Item = &MapsKey> + Send + '_ { self.maps.keys() }
-
-	#[inline]
-	#[must_use]
-	pub fn is_read_only(&self) -> bool { self.db.is_read_only() }
-
-	#[inline]
-	#[must_use]
-	pub fn is_secondary(&self) -> bool { self.db.is_secondary() }
 }
 
 impl Index<&str> for Database {

src/service/emergency/mod.rs

@@ -37,10 +37,6 @@ fn build(args: crate::Args<'_>) -> Result<Arc<Self>> {
 	}
 
 	async fn worker(self: Arc<Self>) -> Result {
-		if self.services.globals.is_read_only() {
-			return Ok(());
-		}
-
 		if self.services.config.ldap.enable {
 			warn!("emergency password feature not available with LDAP enabled.");
 			return Ok(());

src/service/globals/mod.rs

@@ -31,12 +31,13 @@ fn build(args: crate::Args<'_>) -> Result<Arc<Self>> {
 
 		let turn_secret = config.turn_secret_file.as_ref().map_or_else(
 			|| config.turn_secret.clone(),
-			|path| {
-				std::fs::read_to_string(path).unwrap_or_else(|e| {
+			|path| match std::fs::read_to_string(path) {
+				| Ok(secret) => secret.trim().to_owned(),
+				| Err(e) => {
 					error!("Failed to read the TURN secret file: {e}");
 
 					config.turn_secret.clone()
-				})
+				},
 			},
 		);
 
@@ -49,7 +50,7 @@ fn build(args: crate::Args<'_>) -> Result<Arc<Self>> {
 					return config.registration_token.clone();
 				};
 
-				Some(token)
+				Some(token.trim().to_owned())
 			},
 		);
 
@@ -170,7 +171,4 @@ pub fn user_is_local(&self, user_id: &UserId) -> bool {
 	pub fn server_is_ours(&self, server_name: &ServerName) -> bool {
 		server_name == self.server_name()
 	}
-
-	#[inline]
-	pub fn is_read_only(&self) -> bool { self.db.db.is_read_only() }
 }

src/service/migrations.rs

@@ -590,6 +590,10 @@ async fn fix_readreceiptid_readreceipt_duplicates(services: &Services) -> Result
 
 const FIXED_CORRUPT_MSC4133_FIELDS_MARKER: &[u8] = b"fix_corrupt_msc4133_fields";
 async fn fix_corrupt_msc4133_fields(services: &Services) -> Result {
+	// Due to an old bug, some conduwuit databases have `us.cloke.msc4175.tz` user
+	// profile fields with raw strings instead of quoted JSON ones.
+	// This migration fixes that.
+
 	use serde_json::{Value, from_slice};
 	type KeyVal<'a> = ((OwnedUserId, String), &'a [u8]);
 
@@ -606,24 +610,28 @@ async fn fix_corrupt_msc4133_fields(services: &Services) -> Result {
 			async |(mut total, mut fixed),
 			       ((user, key), value): KeyVal<'_>|
 			       -> Result<(usize, usize)> {
-				if let Err(error) = from_slice::<Value>(value) {
-					// Due to an old bug, some conduwuit databases have `us.cloke.msc4175.tz` user
-					// profile fields with raw strings instead of quoted JSON ones.
-					// This migration fixes that.
-					let new_value = if key == "us.cloke.msc4175.tz" {
-						Value::String(String::from_utf8(value.to_vec())?)
-					} else {
-						return Err!(
-							"failed to deserialize msc4133 key {} of user {}: {}",
-							key,
-							user,
-							error
+				match from_slice::<Value>(value) {
+					// corrupted timezone field
+					| Err(_) if key == "us.cloke.msc4175.tz" => {
+						let new_value = Value::String(String::from_utf8(value.to_vec())?);
+						useridprofilekey_value.put((user, key), Json(new_value));
+						fixed = fixed.saturating_add(1);
+					},
+					// corrupted value for some other key
+					| Err(error) => {
+						warn!(
+							"deleting MSC4133 key {} for user {} due to deserialization \
+							 failure: {}",
+							key, user, error
 						);
-					};
-
-					useridprofilekey_value.put((user, key), Json(new_value));
-					fixed = fixed.saturating_add(1);
+						useridprofilekey_value.del((user, key));
+					},
+					// other key with no issues
+					| Ok(_) => {
+						// do nothing
+					},
 				}
+
 				total = total.saturating_add(1);
 
 				Ok((total, fixed))

src/service/rooms/event_handler/handle_incoming_pdu.rs

@@ -14,7 +14,7 @@
 use ruma::{CanonicalJsonValue, EventId, RoomId, ServerName, UserId, events::StateEventType};
 use tracing::debug;
 
-use crate::rooms::timeline::RawPduId;
+use crate::rooms::timeline::{RawPduId, pdu_fits};
 
 /// When receiving an event one needs to:
 /// 0. Check the server is in the room
@@ -62,6 +62,13 @@ pub async fn handle_incoming_pdu<'a>(
 	if let Ok(pdu_id) = self.services.timeline.get_pdu_id(event_id).await {
 		return Ok(Some(pdu_id));
 	}
+	if !pdu_fits(&mut value.clone()) {
+		warn!(
+			"dropping incoming PDU {event_id} in room {room_id} from {origin} because it \
+			 exceeds 65535 bytes or is otherwise too large."
+		);
+		return Err!(Request(TooLarge("PDU is too large")));
+	}
 
 	// 1.1 Check the server is in the room
 	let meta_exists = self.services.metadata.exists(room_id).map(Ok);

src/service/rooms/event_handler/handle_outlier_pdu.rs

@@ -1,7 +1,8 @@
 use std::collections::{BTreeMap, HashMap, hash_map};
 
 use conduwuit::{
-	Err, Event, PduEvent, Result, debug, debug_info, debug_warn, err, implement, state_res, trace,
+	Err, Event, PduEvent, Result, debug, debug_info, debug_warn, err, implement, state_res,
+	trace, warn,
 };
 use futures::future::ready;
 use ruma::{
@@ -10,6 +11,7 @@
 };
 
 use super::{check_room_id, get_room_version_id, to_room_version};
+use crate::rooms::timeline::pdu_fits;
 
 #[implement(super::Service)]
 #[allow(clippy::too_many_arguments)]
@@ -25,6 +27,13 @@ pub(super) async fn handle_outlier_pdu<'a, Pdu>(
 where
 	Pdu: Event + Send + Sync,
 {
+	if !pdu_fits(&mut value.clone()) {
+		warn!(
+			"dropping incoming PDU {event_id} in room {room_id} from {origin} because it \
+			 exceeds 65535 bytes or is otherwise too large."
+		);
+		return Err!(Request(TooLarge("PDU is too large")));
+	}
 	// 1. Remove unsigned field
 	value.remove("unsigned");
 

src/service/rooms/timeline/create.rs

@@ -23,6 +23,40 @@
 
 use super::RoomMutexGuard;
 
+pub fn pdu_fits(owned_obj: &mut CanonicalJsonObject) -> bool {
+	// room IDs, event IDs, senders, types, and state keys must all be <= 255 bytes
+	if let Some(CanonicalJsonValue::String(room_id)) = owned_obj.get("room_id") {
+		if room_id.len() > 255 {
+			return false;
+		}
+	}
+	if let Some(CanonicalJsonValue::String(event_id)) = owned_obj.get("event_id") {
+		if event_id.len() > 255 {
+			return false;
+		}
+	}
+	if let Some(CanonicalJsonValue::String(sender)) = owned_obj.get("sender") {
+		if sender.len() > 255 {
+			return false;
+		}
+	}
+	if let Some(CanonicalJsonValue::String(kind)) = owned_obj.get("type") {
+		if kind.len() > 255 {
+			return false;
+		}
+	}
+	if let Some(CanonicalJsonValue::String(state_key)) = owned_obj.get("state_key") {
+		if state_key.len() > 255 {
+			return false;
+		}
+	}
+	// Now check the full PDU size
+	match serde_json::to_string(owned_obj) {
+		| Ok(s) => s.len() <= 65535,
+		| Err(_) => false,
+	}
+}
+
 #[implement(super::Service)]
 pub async fn create_hash_and_sign_event(
 	&self,
@@ -148,19 +182,6 @@ fn from_evt(
 		}
 	}
 
-	// if event_type != TimelineEventType::RoomCreate && prev_events.is_empty() {
-	// 	return Err!(Request(Unknown("Event incorrectly had zero prev_events.")));
-	// }
-	// if state_key.is_none() && depth.lt(&uint!(2)) {
-	// 	// The first two events in a room are always m.room.create and
-	// m.room.member, 	// so any other events with that same depth are illegal.
-	// 	warn!(
-	// 		"Had unsafe depth {depth} when creating non-state event in {}. Cowardly
-	// aborting", 		room_id.expect("room_id is Some here").as_str()
-	// 	);
-	// 	return Err!(Request(Unknown("Unsafe depth for non-state event.")));
-	// }
-
 	let mut pdu = PduEvent {
 		event_id: ruma::event_id!("$thiswillbefilledinlater").into(),
 		room_id: room_id.map(ToOwned::to_owned),
@@ -269,8 +290,16 @@ fn from_evt(
 	}
 	// Generate event id
 	pdu.event_id = gen_event_id(&pdu_json, &room_version_id)?;
-	// Check with the policy server
 	pdu_json.insert("event_id".into(), CanonicalJsonValue::String(pdu.event_id.clone().into()));
+	// Verify that the *full* PDU isn't over 64KiB.
+	// Ruma only validates that it's under 64KiB before signing and hashing.
+	// Has to be cloned to prevent mutating pdu_json itself :(
+	if !pdu_fits(&mut pdu_json.clone()) {
+		// feckin huge PDU mate
+		return Err!(Request(TooLarge("Message/PDU is too long (exceeds 65535 bytes)")));
+	}
+
+	// Check with the policy server
 	if room_id.is_some() {
 		trace!(
 			"Checking event in room {} with policy server",

src/service/rooms/timeline/mod.rs

@@ -26,7 +26,7 @@
 use serde::Deserialize;
 
 use self::data::Data;
-pub use self::data::PdusIterItem;
+pub use self::{create::pdu_fits, data::PdusIterItem};
 use crate::{
 	Dep, account_data, admin, appservice, globals, pusher, rooms, sending, server_keys, users,
 };

src/service/services.rs

@@ -130,7 +130,7 @@ pub async fn start(self: &Arc<Self>) -> Result<Arc<Self>> {
 
 		// reset dormant online/away statuses to offline, and set the server user as
 		// online
-		if self.server.config.allow_local_presence && !self.db.is_read_only() {
+		if self.server.config.allow_local_presence {
 			self.presence.unset_all_presence().await;
 			_ = self
 				.presence
@@ -146,7 +146,7 @@ pub async fn stop(&self) {
 		info!("Shutting down services...");
 
 		// set the server user as offline
-		if self.server.config.allow_local_presence && !self.db.is_read_only() {
+		if self.server.config.allow_local_presence {
 			_ = self
 				.presence
 				.ping_presence(&self.globals.server_user, &ruma::presence::PresenceState::Offline)

src/service/uiaa/mod.rs

@@ -11,7 +11,7 @@
 use ruma::{
 	CanonicalJsonValue, DeviceId, OwnedDeviceId, OwnedUserId, UserId,
 	api::client::{
-		error::ErrorKind,
+		error::{ErrorKind, StandardErrorBody},
 		uiaa::{AuthData, AuthType, Password, UiaaInfo, UserIdentifier},
 	},
 };
@@ -104,6 +104,7 @@ pub fn create(
 }
 
 #[implement(Service)]
+#[allow(clippy::useless_let_if_seq)]
 pub async fn try_auth(
 	&self,
 	user_id: &UserId,
@@ -163,17 +164,39 @@ pub async fn try_auth(
 			let user_id = user_id_from_username;
 
 			// Check if password is correct
+			let mut password_verified = false;
+
+			// First try local password hash verification
 			if let Ok(hash) = self.services.users.password_hash(&user_id).await {
-				let hash_matches = hash::verify_password(password, &hash).is_ok();
-				if !hash_matches {
-					uiaainfo.auth_error = Some(ruma::api::client::error::StandardErrorBody {
-						kind: ErrorKind::forbidden(),
-						message: "Invalid username or password.".to_owned(),
-					});
-					return Ok((false, uiaainfo));
+				password_verified = hash::verify_password(password, &hash).is_ok();
+			}
+
+			// If local password verification failed, try LDAP authentication
+			#[cfg(feature = "ldap")]
+			if !password_verified && self.services.config.ldap.enable {
+				// Search for user in LDAP to get their DN
+				if let Ok(dns) = self.services.users.search_ldap(&user_id).await {
+					if let Some((user_dn, _is_admin)) = dns.first() {
+						// Try to authenticate with LDAP
+						password_verified = self
+							.services
+							.users
+							.auth_ldap(user_dn, password)
+							.await
+							.is_ok();
+					}
 				}
 			}
 
+			if !password_verified {
+				uiaainfo.auth_error = Some(StandardErrorBody {
+					kind: ErrorKind::forbidden(),
+					message: "Invalid username or password.".to_owned(),
+				});
+
+				return Ok((false, uiaainfo));
+			}
+
 			// Password was correct! Let's add it to `completed`
 			uiaainfo.completed.push(AuthType::Password);
 		},
@@ -197,7 +220,7 @@ pub async fn try_auth(
 				},
 				| Err(e) => {
 					error!("ReCaptcha verification failed: {e:?}");
-					uiaainfo.auth_error = Some(ruma::api::client::error::StandardErrorBody {
+					uiaainfo.auth_error = Some(StandardErrorBody {
 						kind: ErrorKind::forbidden(),
 						message: "ReCaptcha verification failed.".to_owned(),
 					});
@@ -210,7 +233,7 @@ pub async fn try_auth(
 			if tokens.contains(t.token.trim()) {
 				uiaainfo.completed.push(AuthType::RegistrationToken);
 			} else {
-				uiaainfo.auth_error = Some(ruma::api::client::error::StandardErrorBody {
+				uiaainfo.auth_error = Some(StandardErrorBody {
 					kind: ErrorKind::forbidden(),
 					message: "Invalid registration token.".to_owned(),
 				});