#3451: Improve length field prepending on bungee -> server connection

Use alternative implementation of Varint21LengthFieldPrepender on bungee -> server connection for improved speed - it uses separate buffer to prepend the length to avoid copying large data buffer.
Not applied bungee -> client because encrypting 1-5 bytes of length separately through expensive jni call could make it not worth (not measured).

protocol/src/main/java/net/md_5/bungee/protocol/Varint21LengthFieldExtraBufPrepender.java

@@ -0,0 +1,25 @@
+package net.md_5.bungee.protocol;
+
+import io.netty.buffer.ByteBuf;
+import io.netty.channel.ChannelHandler;
+import io.netty.channel.ChannelHandlerContext;
+import io.netty.handler.codec.MessageToMessageEncoder;
+import java.util.List;
+
+/**
+ * Prepend length of the message as a Varint21 using an extra buffer for the length, avoiding copying packet data
+ */
+@ChannelHandler.Sharable
+public class Varint21LengthFieldExtraBufPrepender extends MessageToMessageEncoder<ByteBuf>
+{
+
+    @Override
+    protected void encode(ChannelHandlerContext ctx, ByteBuf msg, List<Object> out) throws Exception
+    {
+        int bodyLen = msg.readableBytes();
+        ByteBuf lenBuf = ctx.alloc().ioBuffer( Varint21LengthFieldPrepender.varintSize( bodyLen ) );
+        DefinedPacket.writeVarInt( bodyLen, lenBuf );
+        out.add( lenBuf );
+        out.add( msg.retain() );
+    }
+}

protocol/src/main/java/net/md_5/bungee/protocol/Varint21LengthFieldPrepender.java

@@ -5,6 +5,9 @@
 import io.netty.channel.ChannelHandlerContext;
 import io.netty.handler.codec.MessageToByteEncoder;
 
+/**
+ * Prepend length of the message as a Varint21 by writing length and data to a new buffer
+ */
 @ChannelHandler.Sharable
 public class Varint21LengthFieldPrepender extends MessageToByteEncoder<ByteBuf>
 {
@@ -20,7 +23,7 @@ protected void encode(ChannelHandlerContext ctx, ByteBuf msg, ByteBuf out) throw
         out.writeBytes( msg );
     }
 
-    private static int varintSize(int paramInt)
+    static int varintSize(int paramInt)
     {
         if ( ( paramInt & 0xFFFFFF80 ) == 0 )
         {

proxy/src/main/java/net/md_5/bungee/BungeeServerInfo.java

@@ -182,7 +182,7 @@ public void operationComplete(ChannelFuture future) throws Exception
         new Bootstrap()
                 .channel( PipelineUtils.getChannel( socketAddress ) )
                 .group( BungeeCord.getInstance().eventLoops )
-                .handler( PipelineUtils.BASE )
+                .handler( PipelineUtils.BASE_SERVERSIDE )
                 .option( ChannelOption.CONNECT_TIMEOUT_MILLIS, BungeeCord.getInstance().getConfig().getRemotePingTimeout() )
                 .remoteAddress( socketAddress )
                 .connect()

proxy/src/main/java/net/md_5/bungee/UserConnection.java

@@ -320,7 +320,7 @@ public void connect(final ServerConnectRequest request)
             @Override
             protected void initChannel(Channel ch) throws Exception
             {
-                PipelineUtils.BASE.initChannel( ch );
+                PipelineUtils.BASE_SERVERSIDE.initChannel( ch );
                 ch.pipeline().addAfter( PipelineUtils.FRAME_DECODER, PipelineUtils.PACKET_DECODER, new MinecraftDecoder( Protocol.HANDSHAKE, false, getPendingConnection().getVersion() ) );
                 ch.pipeline().addAfter( PipelineUtils.FRAME_PREPENDER, PipelineUtils.PACKET_ENCODER, new MinecraftEncoder( Protocol.HANDSHAKE, false, getPendingConnection().getVersion() ) );
                 ch.pipeline().get( HandlerBoss.class ).setHandler( new ServerConnector( bungee, UserConnection.this, target ) );

proxy/src/main/java/net/md_5/bungee/netty/PipelineUtils.java

@@ -30,6 +30,8 @@
 import java.util.concurrent.ThreadFactory;
 import java.util.concurrent.TimeUnit;
 import java.util.logging.Level;
+import lombok.AllArgsConstructor;
+import lombok.NoArgsConstructor;
 import net.md_5.bungee.BungeeCord;
 import net.md_5.bungee.Util;
 import net.md_5.bungee.api.ProxyServer;
@@ -42,6 +44,7 @@
 import net.md_5.bungee.protocol.MinecraftEncoder;
 import net.md_5.bungee.protocol.Protocol;
 import net.md_5.bungee.protocol.Varint21FrameDecoder;
+import net.md_5.bungee.protocol.Varint21LengthFieldExtraBufPrepender;
 import net.md_5.bungee.protocol.Varint21LengthFieldPrepender;
 
 public class PipelineUtils
@@ -82,9 +85,11 @@ protected void initChannel(Channel ch) throws Exception
             }
         }
     };
-    public static final Base BASE = new Base();
+    public static final Base BASE = new Base( false );
+    public static final Base BASE_SERVERSIDE = new Base( true );
     private static final KickStringWriter legacyKicker = new KickStringWriter();
     private static final Varint21LengthFieldPrepender framePrepender = new Varint21LengthFieldPrepender();
+    private static final Varint21LengthFieldExtraBufPrepender serverFramePrepender = new Varint21LengthFieldExtraBufPrepender();
     public static final String TIMEOUT_HANDLER = "timeout";
     public static final String PACKET_DECODER = "packet-decoder";
     public static final String PACKET_ENCODER = "packet-encoder";
@@ -152,9 +157,13 @@ public static Class<? extends DatagramChannel> getDatagramChannel()
     private static final int HIGH_MARK = Integer.getInteger( "net.md_5.bungee.high_mark", 2 << 20 ); // 2 mb
     private static final WriteBufferWaterMark MARK = new WriteBufferWaterMark( LOW_MARK, HIGH_MARK );
 
+    @NoArgsConstructor // for backwards compatibility
+    @AllArgsConstructor
     public static final class Base extends ChannelInitializer<Channel>
     {
 
+        private boolean toServer = false;
+
         @Override
         public void initChannel(Channel ch) throws Exception
         {
@@ -170,7 +179,9 @@ public void initChannel(Channel ch) throws Exception
 
             ch.pipeline().addLast( FRAME_DECODER, new Varint21FrameDecoder() );
             ch.pipeline().addLast( TIMEOUT_HANDLER, new ReadTimeoutHandler( BungeeCord.getInstance().config.getTimeout(), TimeUnit.MILLISECONDS ) );
-            ch.pipeline().addLast( FRAME_PREPENDER, framePrepender );
+            // No encryption bungee -> server, therefore use extra buffer to avoid copying everything for length prepending
+            // Not used bungee -> client as header would need to be encrypted separately through expensive JNI call
+            ch.pipeline().addLast( FRAME_PREPENDER, ( toServer ) ? serverFramePrepender : framePrepender );
 
             ch.pipeline().addLast( BOSS_HANDLER, new HandlerBoss() );
         }