On 08/01/2015 13:46, Rémy Maucherat wrote:
> 2015-01-08 14:10 GMT+01:00 <ma...@apache.org>:
>
>> @@ -766,32 +776,120 @@ public class Nio2Endpoint extends Abstra
>> this.writeCompletionHandler = new CompletionHandler<Integer,
>> ByteBuffer>() {
>> @Override
>> public void completed(Integer nBytes, ByteBuffer
>> attachment) {
>> - if (nBytes.intValue() < 0) {
>> - failed(new EOFException(), attachment);
>> - } else if (attachment.hasRemaining()) {
>> - getSocket().write(attachment, getTimeout(),
>> - TimeUnit.MILLISECONDS, attachment,
>> writeCompletionHandler);
>> - } else {
>> - writePending.release();
>> - if (!Nio2Endpoint.isInline()) {
>> -
>> getEndpoint().processSocket(Nio2SocketWrapper.this,
>> SocketStatus.OPEN_WRITE, false);
>> + boolean notify = false;
>> + synchronized (writeCompletionHandler) {
>> + if (nBytes.intValue() < 0) {
>> + failed(new
>> EOFException(sm.getString("iob.failedwrite")), attachment);
>> + } else if
>> (Nio2SocketWrapper.this.bufferedWrites.size() > 0) {
>> + // Continue writing data using a gathering
>> write
>> + ArrayList<ByteBuffer> arrayList = new
>> ArrayList<>();
>> + if (attachment.hasRemaining()) {
>> + arrayList.add(attachment);
>> + }
>> + for (ByteBufferHolder buffer :
>> Nio2SocketWrapper.this.bufferedWrites) {
>> + buffer.flip();
>> + arrayList.add(buffer.getBuf());
>> + }
>> + Nio2SocketWrapper.this.bufferedWrites.clear();
>> + ByteBuffer[] array = arrayList.toArray(new
>> ByteBuffer[arrayList.size()]);
>> +
>> Nio2SocketWrapper.this.getSocket().write(array, 0, array.length,
>> + Nio2SocketWrapper.this.getTimeout(),
>> TimeUnit.MILLISECONDS,
>> + array,
>> gatheringWriteCompletionHandler);
>> + } else if (attachment.hasRemaining()) {
>> + // Regular write
>> +
>> Nio2SocketWrapper.this.getSocket().write(attachment,
>> Nio2SocketWrapper.this.getTimeout(),
>> + TimeUnit.MILLISECONDS, attachment,
>> writeCompletionHandler);
>> + } else {
>> + // All data has been written
>> + if (interest && !Nio2Endpoint.isInline()) {
>> + interest = false;
>> + notify = true;
>> + }
>> + writePending.release();
>> }
>> }
>>
> There's a big deadlock risk if the recursion if incomplete and error
> processing are inside the sync.
I came across some of those while doing the refactoring. I think - as of
r1650289 - they are addressed but it is possible I missed something. If
you have a particular sequence in mind then I'd be happy to take another
look.
Mark
>
> Rémy
>
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org
For additional commands, e-mail: dev-h...@tomcat.apache.org
