use prelude::*;
use io;
use alloc::collections::LinkedList;
-use sync::{Arc, Mutex, MutexGuard, RwLock};
+use sync::{Arc, Mutex, MutexGuard, FairRwLock};
+use core::sync::atomic::{AtomicBool, Ordering};
use core::{cmp, hash, fmt, mem};
use core::ops::Deref;
use core::convert::Infallible;
L::Target: Logger,
CMH::Target: CustomMessageHandler {
message_handler: MessageHandler<CM, RM>,
- peers: RwLock<PeerHolder<Descriptor>>,
+ peers:
FairRwLock<PeerHolder<Descriptor>>,
/// Only add to this set when noise completes.
/// Locked *after* peers. When an item is removed, it must be removed with the `peers` write
/// lock held. Entries may be added with only the `peers` read lock held (though the
/// `Descriptor` value must already exist in `peers`).
node_id_to_descriptor: Mutex<HashMap<PublicKey, Descriptor>>,
+ /// We can only have one thread processing events at once, but we don't usually need the full
+ /// `peers` write lock to do so, so instead we block on this empty mutex when entering
+ /// `process_events`.
+ event_processing_lock: Mutex<()>,
+ /// Because event processing is global and always does all available work before returning,
+ /// there is no reason for us to have many event processors waiting on the lock at once.
+ /// Instead, we limit the total blocked event processors to always exactly one by setting this
+ /// when an event process call is waiting.
+ blocked_event_processors: AtomicBool,
our_node_secret: SecretKey,
ephemeral_key_midstate: Sha256Engine,
custom_message_handler: CMH,
PeerManager {
message_handler,
- peers: RwLock::new(PeerHolder {
+ peers: FairRwLock::new(PeerHolder {
peers: HashMap::new(),
}),
node_id_to_descriptor: Mutex::new(HashMap::new()),
+ event_processing_lock: Mutex::new(()),
+ blocked_event_processors: AtomicBool::new(false),
our_node_secret,
ephemeral_key_midstate,
peer_counter: AtomicCounter::new(),
if peer.pending_read_is_header {
let msg_len = try_potential_handleerror!(peer,
peer.channel_encryptor.decrypt_length_header(&peer.pending_read_buffer[..]));
- peer.pending_read_buffer = Vec::with_capacity(msg_len as usize + 16);
+ if peer.pending_read_buffer.capacity() > 8192 { peer.pending_read_buffer = Vec::new(); }
peer.pending_read_buffer.resize(msg_len as usize + 16, 0);
if msg_len < 2 { // Need at least the message type tag
return Err(PeerHandleError{ no_connection_possible: false });
assert!(msg_data.len() >= 2);
// Reset read buffer
- peer.pending_read_buffer = [0; 18].to_vec();
+ if peer.pending_read_buffer.capacity() > 8192 { peer.pending_read_buffer = Vec::new(); }
+ peer.pending_read_buffer.resize(18, 0);
peer.pending_read_is_header = true;
let mut reader = io::Cursor::new(&msg_data[..]);
/// You don't have to call this function explicitly if you are using [`lightning-net-tokio`]
/// or one of the other clients provided in our language bindings.
///
+ /// Note that if there are any other calls to this function waiting on lock(s) this may return
+ /// without doing any work. All available events that need handling will be handled before the
+ /// other calls return.
+ ///
/// [`send_payment`]: crate::ln::channelmanager::ChannelManager::send_payment
/// [`ChannelManager::process_pending_htlc_forwards`]: crate::ln::channelmanager::ChannelManager::process_pending_htlc_forwards
/// [`send_data`]: SocketDescriptor::send_data
pub fn process_events(&self) {
+ let mut _single_processor_lock = self.event_processing_lock.try_lock();
+ if _single_processor_lock.is_err() {
+ // While we could wake the older sleeper here with a CV and make more even waiting
+ // times, that would be a lot of overengineering for a simple "reduce total waiter
+ // count" goal.
+ match self.blocked_event_processors.compare_exchange(false, true, Ordering::AcqRel, Ordering::Acquire) {
+ Err(val) => {
+ debug_assert!(val, "compare_exchange failed spuriously?");
+ return;
+ },
+ Ok(val) => {
+ debug_assert!(!val, "compare_exchange succeeded spuriously?");
+ // We're the only waiter, as the running process_events may have emptied the
+ // pending events "long" ago and there are new events for us to process, wait until
+ // its done and process any leftover events before returning.
+ _single_processor_lock = Ok(self.event_processing_lock.lock().unwrap());
+ self.blocked_event_processors.store(false, Ordering::Release);
+ }
+ }
+ }
+
+ let mut peers_to_disconnect = HashMap::new();
+ let mut events_generated = self.message_handler.chan_handler.get_and_clear_pending_msg_events();
+ events_generated.append(&mut self.message_handler.route_handler.get_and_clear_pending_msg_events());
+
{
// TODO: There are some DoS attacks here where you can flood someone's outbound send
// buffer by doing things like announcing channels on another node. We should be willing to
// drop optional-ish messages when send buffers get full!
- let mut peers_lock = self.peers.write().unwrap();
- let mut events_generated = self.message_handler.chan_handler.get_and_clear_pending_msg_events();
- events_generated.append(&mut self.message_handler.route_handler.get_and_clear_pending_msg_events());
- let peers = &mut *peers_lock;
+ let peers_lock = self.peers.read().unwrap();
+ let peers = &*peers_lock;
macro_rules! get_peer_for_forwarding {
($node_id: expr) => {
{
- match self.node_id_to_descriptor.lock().unwrap().get($node_id) {
- Some(descriptor) => match peers.peers.get_mut(&descriptor) {
+ if peers_to_disconnect.get($node_id).is_some() {
+ // If we've "disconnected" this peer, do not send to it.
+ continue;
+ }
+ let descriptor_opt = self.node_id_to_descriptor.lock().unwrap().get($node_id).cloned();
+ match descriptor_opt {
+ Some(descriptor) => match peers.peers.get(&descriptor) {
Some(peer_mutex) => {
let peer_lock = peer_mutex.lock().unwrap();
if peer_lock.their_features.is_none() {
}
peer_lock
},
- None => panic!("Inconsistent peers set state!"),
+ None => {
+ debug_assert!(false, "Inconsistent peers set state!");
+ continue;
+ }
},
None => {
continue;
MessageSendEvent::HandleError { ref node_id, ref action } => {
match *action {
msgs::ErrorAction::DisconnectPeer { ref msg } => {
- // Note that since we are holding the peers *write* lock we can
- // remove from node_id_to_descriptor immediately (as no other
- // thread can be holding the peer lock if we have the global write
- // lock).
- if let Some(mut descriptor) = self.node_id_to_descriptor.lock().unwrap().remove(node_id) {
- if let Some(peer_mutex) = peers.peers.remove(&descriptor) {
- if let Some(ref msg) = *msg {
- log_trace!(self.logger, "Handling DisconnectPeer HandleError event in peer_handler for node {} with message {}",
- log_pubkey!(node_id),
- msg.data);
- let mut peer = peer_mutex.lock().unwrap();
- self.enqueue_message(&mut *peer, msg);
- // This isn't guaranteed to work, but if there is enough free
- // room in the send buffer, put the error message there...
- self.do_attempt_write_data(&mut descriptor, &mut *peer);
- } else {
- log_gossip!(self.logger, "Handling DisconnectPeer HandleError event in peer_handler for node {} with no message", log_pubkey!(node_id));
- }
- }
- descriptor.disconnect_socket();
- self.message_handler.chan_handler.peer_disconnected(&node_id, false);
- }
+ // We do not have the peers write lock, so we just store that we're
+ // about to disconenct the peer and do it after we finish
+ // processing most messages.
+ peers_to_disconnect.insert(*node_id, msg.clone());
},
msgs::ErrorAction::IgnoreAndLog(level) => {
log_given_level!(self.logger, level, "Received a HandleError event to be ignored for node {}", log_pubkey!(node_id));
}
for (node_id, msg) in self.custom_message_handler.get_and_clear_pending_msg() {
+ if peers_to_disconnect.get(&node_id).is_some() { continue; }
self.enqueue_message(&mut *get_peer_for_forwarding!(&node_id), &msg);
}
- for (descriptor, peer_mutex) in peers.peers.iter_mut() {
+ for (descriptor, peer_mutex) in peers.peers.iter() {
self.do_attempt_write_data(&mut (*descriptor).clone(), &mut *peer_mutex.lock().unwrap());
}
}
+ if !peers_to_disconnect.is_empty() {
+ let mut peers_lock = self.peers.write().unwrap();
+ let peers = &mut *peers_lock;
+ for (node_id, msg) in peers_to_disconnect.drain() {
+ // Note that since we are holding the peers *write* lock we can
+ // remove from node_id_to_descriptor immediately (as no other
+ // thread can be holding the peer lock if we have the global write
+ // lock).
+
+ if let Some(mut descriptor) = self.node_id_to_descriptor.lock().unwrap().remove(&node_id) {
+ if let Some(peer_mutex) = peers.peers.remove(&descriptor) {
+ if let Some(msg) = msg {
+ log_trace!(self.logger, "Handling DisconnectPeer HandleError event in peer_handler for node {} with message {}",
+ log_pubkey!(node_id),
+ msg.data);
+ let mut peer = peer_mutex.lock().unwrap();
+ self.enqueue_message(&mut *peer, &msg);
+ // This isn't guaranteed to work, but if there is enough free
+ // room in the send buffer, put the error message there...
+ self.do_attempt_write_data(&mut descriptor, &mut *peer);
+ } else {
+ log_trace!(self.logger, "Handling DisconnectPeer HandleError event in peer_handler for node {} with no message", log_pubkey!(node_id));
+ }
+ }
+ descriptor.disconnect_socket();
+ self.message_handler.chan_handler.peer_disconnected(&node_id, false);
+ }
+ }
+ }
}
/// Indicates that the given socket descriptor's connection is now closed.
///
/// [`send_data`]: SocketDescriptor::send_data
pub fn timer_tick_occurred(&self) {
- let mut peers_lock = self.peers.write().unwrap();
+ let mut descriptors_needing_disconnect = Vec::new();
{
- let mut descriptors_needing_disconnect = Vec::new();
- let peer_count = peers_lock.peers.len();
+ let peers_lock = self.peers.read().unwrap();
- peers_lock.peers.retain(|descriptor, peer_mutex| {
+ for (descriptor, peer_mutex) in peers_lock.peers.iter() {
let mut peer = peer_mutex.lock().unwrap();
- let mut do_disconnect_peer = false;
if !peer.channel_encryptor.is_ready_for_encryption() || peer.their_node_id.is_none() {
// The peer needs to complete its handshake before we can exchange messages. We
// give peers one timer tick to complete handshake, reusing
// `awaiting_pong_timer_tick_intervals` to track number of timer ticks taken
// for handshake completion.
if peer.awaiting_pong_timer_tick_intervals != 0 {
- do_disconnect_peer = true;
+ descriptors_needing_disconnect.push(descriptor.clone());
} else {
peer.awaiting_pong_timer_tick_intervals = 1;
- return true;
}
+ continue;
}
if peer.awaiting_pong_timer_tick_intervals == -1 {
// Magic value set in `maybe_send_extra_ping`.
peer.awaiting_pong_timer_tick_intervals = 1;
peer.received_message_since_timer_tick = false;
- return true;
+ continue;
}
- if do_disconnect_peer
- || (peer.awaiting_pong_timer_tick_intervals > 0 && !peer.received_message_since_timer_tick)
+ if (peer.awaiting_pong_timer_tick_intervals > 0 && !peer.received_message_since_timer_tick)
|| peer.awaiting_pong_timer_tick_intervals as u64 >
- MAX_BUFFER_DRAIN_TICK_INTERVALS_PER_PEER as u64 * peer_count as u64
+ MAX_BUFFER_DRAIN_TICK_INTERVALS_PER_PEER as u64 * peers_lock.peers.len() as u64
{
descriptors_needing_disconnect.push(descriptor.clone());
- match peer.their_node_id {
- Some(node_id) => {
- log_trace!(self.logger, "Disconnecting peer with id {} due to ping timeout", node_id);
- self.node_id_to_descriptor.lock().unwrap().remove(&node_id);
- self.message_handler.chan_handler.peer_disconnected(&node_id, false);
- }
- None => {},
- }
- return false;
+ continue;
}
peer.received_message_since_timer_tick = false;
if peer.awaiting_pong_timer_tick_intervals > 0 {
peer.awaiting_pong_timer_tick_intervals += 1;
- return true;
+ continue;
}
peer.awaiting_pong_timer_tick_intervals = 1;
};
self.enqueue_message(&mut *peer, &ping);
self.do_attempt_write_data(&mut (descriptor.clone()), &mut *peer);
+ }
+ }
- true
- });
+ if !descriptors_needing_disconnect.is_empty() {
+ {
+ let mut peers_lock = self.peers.write().unwrap();
+ for descriptor in descriptors_needing_disconnect.iter() {
+ if let Some(peer) = peers_lock.peers.remove(&descriptor) {
+ if let Some(node_id) = peer.lock().unwrap().their_node_id {
+ log_trace!(self.logger, "Disconnecting peer with id {} due to ping timeout", node_id);
+ self.node_id_to_descriptor.lock().unwrap().remove(&node_id);
+ self.message_handler.chan_handler.peer_disconnected(&node_id, false);
+ }
+ }
+ }
+ }
for mut descriptor in descriptors_needing_disconnect.drain(..) {
descriptor.disconnect_socket();
projects / rust-lightning / blobdiff