eventmachine 1.2.0.dev.2-x64-mingw32

data/ext/ed.cpp

@@ -0,0 +1,2111 @@
+/*****************************************************************************
+
+$Id$
+
+File:     ed.cpp
+Date:     06Apr06
+
+Copyright (C) 2006-07 by Francis Cianfrocca. All Rights Reserved.
+Gmail: blackhedd
+
+This program is free software; you can redistribute it and/or modify
+it under the terms of either: 1) the GNU General Public License
+as published by the Free Software Foundation; either version 2 of the
+License, or (at your option) any later version; or 2) Ruby's License.
+
+See the file COPYING for complete licensing information.
+
+*****************************************************************************/
+
+#include "project.h"
+
+
+
+/********************
+SetSocketNonblocking
+********************/
+
+bool SetSocketNonblocking (SOCKET sd)
+{
+	#ifdef OS_UNIX
+	int val = fcntl (sd, F_GETFL, 0);
+	return (fcntl (sd, F_SETFL, val | O_NONBLOCK) != SOCKET_ERROR) ? true : false;
+	#endif
+
+	#ifdef OS_WIN32
+	#ifdef BUILD_FOR_RUBY
+	// 14Jun09 Ruby provides its own wrappers for ioctlsocket. On 1.8 this is a simple wrapper,
+	// however, 1.9 keeps its own state about the socket.
+	// NOTE: F_GETFL is not supported
+	return (fcntl (sd, F_SETFL, O_NONBLOCK) == 0) ? true : false;
+	#else
+	unsigned long one = 1;
+	return (ioctlsocket (sd, FIONBIO, &one) == 0) ? true : false;
+	#endif
+	#endif
+}
+
+/************
+SetFdCloexec
+************/
+
+#ifdef OS_UNIX
+bool SetFdCloexec (int fd)
+{
+	int flags = fcntl(fd, F_GETFD, 0);
+	assert (flags >= 0);
+	flags |= FD_CLOEXEC;
+	return (fcntl(fd, F_SETFD, FD_CLOEXEC) == 0) ? true : false;
+}
+#else
+bool SetFdCloexec (int fd UNUSED)
+{
+	return true;
+}
+#endif
+
+/****************************************
+EventableDescriptor::EventableDescriptor
+****************************************/
+
+EventableDescriptor::EventableDescriptor (SOCKET sd, EventMachine_t *em):
+	bCloseNow (false),
+	bCloseAfterWriting (false),
+	MySocket (sd),
+	bAttached (false),
+	bWatchOnly (false),
+	EventCallback (NULL),
+	bCallbackUnbind (true),
+	UnbindReasonCode (0),
+	ProxyTarget(NULL),
+	ProxiedFrom(NULL),
+	ProxiedBytes(0),
+	MaxOutboundBufSize(0),
+	MyEventMachine (em),
+	PendingConnectTimeout(20000000),
+	InactivityTimeout (0),
+	bPaused (false)
+{
+	/* There are three ways to close a socket, all of which should
+	 * automatically signal to the event machine that this object
+	 * should be removed from the polling scheduler.
+	 * First is a hard close, intended for bad errors or possible
+	 * security violations. It immediately closes the connection
+	 * and puts this object into an error state.
+	 * Second is to set bCloseNow, which will cause the event machine
+	 * to delete this object (and thus close the connection in our
+	 * destructor) the next chance it gets. bCloseNow also inhibits
+	 * the writing of new data on the socket (but not necessarily
+	 * the reading of new data).
+	 * The third way is to set bCloseAfterWriting, which inhibits
+	 * the writing of new data and converts to bCloseNow as soon
+	 * as everything in the outbound queue has been written.
+	 * bCloseAfterWriting is really for use only by protocol handlers
+	 * (for example, HTTP writes an HTML page and then closes the
+	 * connection). All of the error states we generate internally
+	 * cause an immediate close to be scheduled, which may have the
+	 * effect of discarding outbound data.
+	 */
+
+	if (sd == INVALID_SOCKET)
+		throw std::runtime_error ("bad eventable descriptor");
+	if (MyEventMachine == NULL)
+		throw std::runtime_error ("bad em in eventable descriptor");
+	CreatedAt = MyEventMachine->GetCurrentLoopTime();
+
+	#ifdef HAVE_EPOLL
+	EpollEvent.events = 0;
+	EpollEvent.data.ptr = this;
+	#endif
+	LastActivity = MyEventMachine->GetCurrentLoopTime();
+}
+
+
+/*****************************************
+EventableDescriptor::~EventableDescriptor
+*****************************************/
+
+EventableDescriptor::~EventableDescriptor()
+{
+	if (NextHeartbeat)
+		MyEventMachine->ClearHeartbeat(NextHeartbeat, this);
+	if (EventCallback && bCallbackUnbind)
+		(*EventCallback)(GetBinding(), EM_CONNECTION_UNBOUND, NULL, UnbindReasonCode);
+	if (ProxiedFrom) {
+		(*EventCallback)(ProxiedFrom->GetBinding(), EM_PROXY_TARGET_UNBOUND, NULL, 0);
+		ProxiedFrom->StopProxy();
+	}
+	MyEventMachine->NumCloseScheduled--;
+	StopProxy();
+	Close();
+}
+
+
+/*************************************
+EventableDescriptor::SetEventCallback
+*************************************/
+
+void EventableDescriptor::SetEventCallback (EMCallback cb)
+{
+	EventCallback = cb;
+}
+
+
+/**************************
+EventableDescriptor::Close
+**************************/
+
+void EventableDescriptor::Close()
+{
+	/* EventMachine relies on the fact that when close(fd)
+	 * is called that the fd is removed from any
+	 * epoll event queues.
+	 *
+	 * However, this is not *always* the behavior of close(fd)
+	 *
+	 * See man 4 epoll Q6/A6 and then consider what happens
+	 * when using pipes with eventmachine.
+	 * (As is often done when communicating with a subprocess)
+	 *
+	 * The pipes end up looking like:
+	 *
+	 * ls -l /proc/<pid>/fd
+	 * ...
+	 * lr-x------ 1 root root 64 2011-08-19 21:31 3 -> pipe:[940970]
+	 * l-wx------ 1 root root 64 2011-08-19 21:31 4 -> pipe:[940970]
+	 *
+	 * This meets the critera from man 4 epoll Q6/A4 for not
+	 * removing fds from epoll event queues until all fds
+	 * that reference the underlying file have been removed.
+	 *
+	 * If the EventableDescriptor associated with fd 3 is deleted,
+	 * its dtor will call EventableDescriptor::Close(),
+	 * which will call ::close(int fd).
+	 *
+	 * However, unless the EventableDescriptor associated with fd 4 is
+	 * also deleted before the next call to epoll_wait, events may fire
+	 * for fd 3 that were registered with an already deleted
+	 * EventableDescriptor.
+	 *
+	 * Therefore, it is necessary to notify EventMachine that
+	 * the fd associated with this EventableDescriptor is
+	 * closing.
+	 *
+	 * EventMachine also never closes fds for STDIN, STDOUT and 
+	 * STDERR (0, 1 & 2)
+	 */
+
+	// Close the socket right now. Intended for emergencies.
+	if (MySocket != INVALID_SOCKET) {
+		MyEventMachine->Deregister (this);
+		
+		// Do not close STDIN, STDOUT, STDERR
+		if (MySocket > 2 && !bAttached) {
+			shutdown (MySocket, 1);
+			close (MySocket);
+		}
+		
+		MySocket = INVALID_SOCKET;
+	}
+}
+
+
+/*********************************
+EventableDescriptor::ShouldDelete
+*********************************/
+
+bool EventableDescriptor::ShouldDelete()
+{
+	/* For use by a socket manager, which needs to know if this object
+	 * should be removed from scheduling events and deleted.
+	 * Has an immediate close been scheduled, or are we already closed?
+	 * If either of these are the case, return true. In theory, the manager will
+	 * then delete us, which in turn will make sure the socket is closed.
+	 * Note, if bCloseAfterWriting is true, we check a virtual method to see
+	 * if there is outbound data to write, and only request a close if there is none.
+	 */
+
+	return ((MySocket == INVALID_SOCKET) || bCloseNow || (bCloseAfterWriting && (GetOutboundDataSize() <= 0)));
+}
+
+
+/**********************************
+EventableDescriptor::ScheduleClose
+**********************************/
+
+void EventableDescriptor::ScheduleClose (bool after_writing)
+{
+	if (IsCloseScheduled())
+		return;
+	MyEventMachine->NumCloseScheduled++;
+	// KEEP THIS SYNCHRONIZED WITH ::IsCloseScheduled.
+	if (after_writing)
+		bCloseAfterWriting = true;
+	else
+		bCloseNow = true;
+}
+
+
+/*************************************
+EventableDescriptor::IsCloseScheduled
+*************************************/
+
+bool EventableDescriptor::IsCloseScheduled()
+{
+	// KEEP THIS SYNCHRONIZED WITH ::ScheduleClose.
+	return (bCloseNow || bCloseAfterWriting);
+}
+
+
+/*******************************
+EventableDescriptor::StartProxy
+*******************************/
+
+void EventableDescriptor::StartProxy(const uintptr_t to, const unsigned long bufsize, const unsigned long length)
+{
+	EventableDescriptor *ed = dynamic_cast <EventableDescriptor*> (Bindable_t::GetObject (to));
+	if (ed) {
+		StopProxy();
+		ProxyTarget = ed;
+		BytesToProxy = length;
+		ProxiedBytes = 0;
+		ed->SetProxiedFrom(this, bufsize);
+		return;
+	}
+	throw std::runtime_error ("Tried to proxy to an invalid descriptor");
+}
+
+
+/******************************
+EventableDescriptor::StopProxy
+******************************/
+
+void EventableDescriptor::StopProxy()
+{
+	if (ProxyTarget) {
+		ProxyTarget->SetProxiedFrom(NULL, 0);
+		ProxyTarget = NULL;
+	}
+}
+
+
+/***********************************
+EventableDescriptor::SetProxiedFrom
+***********************************/
+
+void EventableDescriptor::SetProxiedFrom(EventableDescriptor *from, const unsigned long bufsize)
+{
+	if (from != NULL && ProxiedFrom != NULL)
+		throw std::runtime_error ("Tried to proxy to a busy target");
+
+	ProxiedFrom = from;
+	MaxOutboundBufSize = bufsize;
+}
+
+
+/********************************************
+EventableDescriptor::_GenericInboundDispatch
+********************************************/
+
+void EventableDescriptor::_GenericInboundDispatch(const char *buf, unsigned long size)
+{
+	assert(EventCallback);
+
+	if (ProxyTarget) {
+		if (BytesToProxy > 0) {
+			unsigned long proxied = min(BytesToProxy, size);
+			ProxyTarget->SendOutboundData(buf, proxied);
+			ProxiedBytes += (unsigned long) proxied;
+			BytesToProxy -= proxied;
+			if (BytesToProxy == 0) {
+				StopProxy();
+				(*EventCallback)(GetBinding(), EM_PROXY_COMPLETED, NULL, 0);
+				if (proxied < size) {
+					(*EventCallback)(GetBinding(), EM_CONNECTION_READ, buf + proxied, size - proxied);
+				}
+			}
+		} else {
+			ProxyTarget->SendOutboundData(buf, size);
+			ProxiedBytes += size;
+		}
+	} else {
+		(*EventCallback)(GetBinding(), EM_CONNECTION_READ, buf, size);
+	}
+}
+
+
+/*********************************************
+EventableDescriptor::GetPendingConnectTimeout
+*********************************************/
+
+uint64_t EventableDescriptor::GetPendingConnectTimeout()
+{
+	return PendingConnectTimeout / 1000;
+}
+
+
+/*********************************************
+EventableDescriptor::SetPendingConnectTimeout
+*********************************************/
+
+int EventableDescriptor::SetPendingConnectTimeout (uint64_t value)
+{
+	if (value > 0) {
+		PendingConnectTimeout = value * 1000;
+		MyEventMachine->QueueHeartbeat(this);
+		return 1;
+	}
+	return 0;
+}
+
+
+/*************************************
+EventableDescriptor::GetNextHeartbeat
+*************************************/
+
+uint64_t EventableDescriptor::GetNextHeartbeat()
+{
+	if (NextHeartbeat)
+		MyEventMachine->ClearHeartbeat(NextHeartbeat, this);
+
+	NextHeartbeat = 0;
+
+	if (!ShouldDelete()) {
+		uint64_t time_til_next = InactivityTimeout;
+		if (IsConnectPending()) {
+			if (time_til_next == 0 || PendingConnectTimeout < time_til_next)
+				time_til_next = PendingConnectTimeout;
+		}
+		if (time_til_next == 0)
+			return 0;
+		NextHeartbeat = time_til_next + MyEventMachine->GetRealTime();
+	}
+
+	return NextHeartbeat;
+}
+
+
+/******************************************
+ConnectionDescriptor::ConnectionDescriptor
+******************************************/
+
+ConnectionDescriptor::ConnectionDescriptor (SOCKET sd, EventMachine_t *em):
+	EventableDescriptor (sd, em),
+	bConnectPending (false),
+	bNotifyReadable (false),
+	bNotifyWritable (false),
+	bReadAttemptedAfterClose (false),
+	bWriteAttemptedAfterClose (false),
+	OutboundDataSize (0),
+	#ifdef WITH_SSL
+	SslBox (NULL),
+	bHandshakeSignaled (false),
+	bSslVerifyPeer (false),
+	bSslPeerAccepted(false),
+	#endif
+	#ifdef HAVE_KQUEUE
+	bGotExtraKqueueEvent(false),
+	#endif
+	bIsServer (false)
+{
+	// 22Jan09: Moved ArmKqueueWriter into SetConnectPending() to fix assertion failure in _WriteOutboundData()
+	//  5May09: Moved EPOLLOUT into SetConnectPending() so it doesn't happen for attached read pipes
+}
+
+
+/*******************************************
+ConnectionDescriptor::~ConnectionDescriptor
+*******************************************/
+
+ConnectionDescriptor::~ConnectionDescriptor()
+{
+	// Run down any stranded outbound data.
+	for (size_t i=0; i < OutboundPages.size(); i++)
+		OutboundPages[i].Free();
+
+	#ifdef WITH_SSL
+	if (SslBox)
+		delete SslBox;
+	#endif
+}
+
+
+/***********************************
+ConnectionDescriptor::_UpdateEvents
+************************************/
+
+void ConnectionDescriptor::_UpdateEvents()
+{
+	_UpdateEvents(true, true);
+}
+
+void ConnectionDescriptor::_UpdateEvents(bool read, bool write)
+{
+	if (MySocket == INVALID_SOCKET)
+		return;
+
+	if (!read && !write)
+		return;
+
+	#ifdef HAVE_EPOLL
+	unsigned int old = EpollEvent.events;
+
+	if (read) {
+		if (SelectForRead())
+			EpollEvent.events |= EPOLLIN;
+		else
+			EpollEvent.events &= ~EPOLLIN;
+	}
+
+	if (write) {
+		if (SelectForWrite())
+			EpollEvent.events |= EPOLLOUT;
+		else
+			EpollEvent.events &= ~EPOLLOUT;
+	}
+
+	if (old != EpollEvent.events)
+		MyEventMachine->Modify (this);
+	#endif
+
+	#ifdef HAVE_KQUEUE
+	if (read && SelectForRead())
+		MyEventMachine->ArmKqueueReader (this);
+	bKqueueArmWrite = SelectForWrite();
+	if (write && bKqueueArmWrite)
+		MyEventMachine->Modify (this);
+	#endif
+}
+
+/***************************************
+ConnectionDescriptor::SetConnectPending
+****************************************/
+
+void ConnectionDescriptor::SetConnectPending(bool f)
+{
+	bConnectPending = f;
+	MyEventMachine->QueueHeartbeat(this);
+	_UpdateEvents();
+}
+
+
+/**********************************
+ConnectionDescriptor::SetAttached
+***********************************/
+
+void ConnectionDescriptor::SetAttached(bool state)
+{
+	bAttached = state;
+}
+
+
+/**********************************
+ConnectionDescriptor::SetWatchOnly
+***********************************/
+
+void ConnectionDescriptor::SetWatchOnly(bool watching)
+{
+	bWatchOnly = watching;
+	_UpdateEvents();
+}
+
+
+/*********************************
+ConnectionDescriptor::HandleError
+*********************************/
+
+void ConnectionDescriptor::HandleError()
+{
+	if (bWatchOnly) {
+		// An EPOLLHUP | EPOLLIN condition will call Read() before HandleError(), in which case the
+		// socket is already detached and invalid, so we don't need to do anything.
+		if (MySocket == INVALID_SOCKET) return;
+
+		// HandleError() is called on WatchOnly descriptors by the epoll reactor
+		// when it gets a EPOLLERR | EPOLLHUP. Usually this would show up as a readable and
+		// writable event on other reactors, so we have to fire those events ourselves.
+		if (bNotifyReadable) Read();
+		if (bNotifyWritable) Write();
+	} else {
+		ScheduleClose (false);
+	}
+}
+
+
+/***********************************
+ConnectionDescriptor::ScheduleClose
+***********************************/
+
+void ConnectionDescriptor::ScheduleClose (bool after_writing)
+{
+	if (bWatchOnly)
+		throw std::runtime_error ("cannot close 'watch only' connections");
+
+	EventableDescriptor::ScheduleClose(after_writing);
+}
+
+
+/***************************************
+ConnectionDescriptor::SetNotifyReadable
+****************************************/
+
+void ConnectionDescriptor::SetNotifyReadable(bool readable)
+{
+	if (!bWatchOnly)
+		throw std::runtime_error ("notify_readable must be on 'watch only' connections");
+
+	bNotifyReadable = readable;
+	_UpdateEvents(true, false);
+}
+
+
+/***************************************
+ConnectionDescriptor::SetNotifyWritable
+****************************************/
+
+void ConnectionDescriptor::SetNotifyWritable(bool writable)
+{
+	if (!bWatchOnly)
+		throw std::runtime_error ("notify_writable must be on 'watch only' connections");
+
+	bNotifyWritable = writable;
+	_UpdateEvents(false, true);
+}
+
+
+/**************************************
+ConnectionDescriptor::SendOutboundData
+**************************************/
+
+int ConnectionDescriptor::SendOutboundData (const char *data, unsigned long length)
+{
+	if (bWatchOnly)
+		throw std::runtime_error ("cannot send data on a 'watch only' connection");
+
+	if (ProxiedFrom && MaxOutboundBufSize && (unsigned int)(GetOutboundDataSize() + length) > MaxOutboundBufSize)
+		ProxiedFrom->Pause();
+
+	#ifdef WITH_SSL
+	if (SslBox) {
+		if (length > 0) {
+			unsigned long writed = 0;
+			char *p = (char*)data;
+
+			while (writed < length) {
+				int to_write = SSLBOX_INPUT_CHUNKSIZE;
+				int remaining = length - writed;
+
+				if (remaining < SSLBOX_INPUT_CHUNKSIZE)
+					to_write = remaining;
+
+				int w = SslBox->PutPlaintext (p, to_write);
+				if (w < 0) {
+					ScheduleClose (false);
+				}else
+					_DispatchCiphertext();
+
+				p += to_write;
+				writed += to_write;
+			}
+		}
+		// TODO: What's the correct return value?
+		return 1; // That's a wild guess, almost certainly wrong.
+	}
+	else
+	#endif
+		return _SendRawOutboundData (data, length);
+}
+
+
+
+/******************************************
+ConnectionDescriptor::_SendRawOutboundData
+******************************************/
+
+int ConnectionDescriptor::_SendRawOutboundData (const char *data, unsigned long length)
+{
+	/* This internal method is called to schedule bytes that
+	 * will be sent out to the remote peer.
+	 * It's not directly accessed by the caller, who hits ::SendOutboundData,
+	 * which may or may not filter or encrypt the caller's data before
+	 * sending it here.
+	 */
+
+	// Highly naive and incomplete implementation.
+	// There's no throttle for runaways (which should abort only this connection
+	// and not the whole process), and no coalescing of small pages.
+	// (Well, not so bad, small pages are coalesced in ::Write)
+
+	if (IsCloseScheduled())
+		return 0;
+	// 25Mar10: Ignore 0 length packets as they are not meaningful in TCP (as opposed to UDP)
+	// and can cause the assert(nbytes>0) to fail when OutboundPages has a bunch of 0 length pages.
+	if (length == 0)
+		return 0;
+
+	if (!data && (length > 0))
+		throw std::runtime_error ("bad outbound data");
+	char *buffer = (char *) malloc (length + 1);
+	if (!buffer)
+		throw std::runtime_error ("no allocation for outbound data");
+
+	memcpy (buffer, data, length);
+	buffer [length] = 0;
+	OutboundPages.push_back (OutboundPage (buffer, length));
+	OutboundDataSize += length;
+
+	_UpdateEvents(false, true);
+
+	return length;
+}
+
+
+
+/***********************************
+ConnectionDescriptor::SelectForRead
+***********************************/
+
+bool ConnectionDescriptor::SelectForRead()
+{
+	/* A connection descriptor is always scheduled for read,
+	 * UNLESS it's in a pending-connect state.
+	 * On Linux, unlike Unix, a nonblocking socket on which
+	 * connect has been called, does NOT necessarily select
+	 * both readable and writable in case of error.
+	 * The socket will select writable when the disposition
+	 * of the connect is known. On the other hand, a socket
+	 * which successfully connects and selects writable may
+	 * indeed have some data available on it, so it will
+	 * select readable in that case, violating expectations!
+	 * So we will not poll for readability until the socket
+	 * is known to be in a connected state.
+	 */
+
+	if (bPaused)
+		return false;
+	else if (bConnectPending)
+		return false;
+	else if (bWatchOnly)
+		return bNotifyReadable ? true : false;
+	else
+		return true;
+}
+
+
+/************************************
+ConnectionDescriptor::SelectForWrite
+************************************/
+
+bool ConnectionDescriptor::SelectForWrite()
+{
+	/* Cf the notes under SelectForRead.
+	 * In a pending-connect state, we ALWAYS select for writable.
+	 * In a normal state, we only select for writable when we
+	 * have outgoing data to send.
+	 */
+
+	if (bPaused)
+		return false;
+	else if (bConnectPending)
+		return true;
+	else if (bWatchOnly)
+		return bNotifyWritable ? true : false;
+	else
+		return (GetOutboundDataSize() > 0);
+}
+
+/***************************
+ConnectionDescriptor::Pause
+***************************/
+
+bool ConnectionDescriptor::Pause()
+{
+	if (bWatchOnly)
+		throw std::runtime_error ("cannot pause/resume 'watch only' connections, set notify readable/writable instead");
+
+	bool old = bPaused;
+	bPaused = true;
+	_UpdateEvents();
+	return old == false;
+}
+
+/****************************
+ConnectionDescriptor::Resume
+****************************/
+
+bool ConnectionDescriptor::Resume()
+{
+	if (bWatchOnly)
+		throw std::runtime_error ("cannot pause/resume 'watch only' connections, set notify readable/writable instead");
+
+	bool old = bPaused;
+	bPaused = false;
+	_UpdateEvents();
+	return old == true;
+}
+
+/**************************
+ConnectionDescriptor::Read
+**************************/
+
+void ConnectionDescriptor::Read()
+{
+	/* Read and dispatch data on a socket that has selected readable.
+	 * It's theoretically possible to get and dispatch incoming data on
+	 * a socket that has already been scheduled for closing or close-after-writing.
+	 * In those cases, we'll leave it up the to protocol handler to "do the
+	 * right thing" (which probably means to ignore the incoming data).
+	 *
+	 * 22Aug06: Chris Ochs reports that on FreeBSD, it's possible to come
+	 * here with the socket already closed, after the process receives
+	 * a ctrl-C signal (not sure if that's TERM or INT on BSD). The application
+	 * was one in which network connections were doing a lot of interleaved reads
+	 * and writes.
+	 * Since we always write before reading (in order to keep the outbound queues
+	 * as light as possible), I think what happened is that an interrupt caused
+	 * the socket to be closed in ConnectionDescriptor::Write. We'll then
+	 * come here in the same pass through the main event loop, and won't get
+	 * cleaned up until immediately after.
+	 * We originally asserted that the socket was valid when we got here.
+	 * To deal properly with the possibility that we are closed when we get here,
+	 * I removed the assert. HOWEVER, the potential for an infinite loop scares me,
+	 * so even though this is really clunky, I added a flag to assert that we never
+	 * come here more than once after being closed. (FCianfrocca)
+	 */
+
+	SOCKET sd = GetSocket();
+	//assert (sd != INVALID_SOCKET); (original, removed 22Aug06)
+	if (sd == INVALID_SOCKET) {
+		assert (!bReadAttemptedAfterClose);
+		bReadAttemptedAfterClose = true;
+		return;
+	}
+
+	if (bWatchOnly) {
+		if (bNotifyReadable && EventCallback)
+			(*EventCallback)(GetBinding(), EM_CONNECTION_NOTIFY_READABLE, NULL, 0);
+		return;
+	}
+
+	LastActivity = MyEventMachine->GetCurrentLoopTime();
+
+	int total_bytes_read = 0;
+	char readbuffer [16 * 1024 + 1];
+
+	for (int i=0; i < 10; i++) {
+		// Don't read just one buffer and then move on. This is faster
+		// if there is a lot of incoming.
+		// But don't read indefinitely. Give other sockets a chance to run.
+		// NOTICE, we're reading one less than the buffer size.
+		// That's so we can put a guard byte at the end of what we send
+		// to user code.
+		
+
+		int r = read (sd, readbuffer, sizeof(readbuffer) - 1);
+#ifdef OS_WIN32
+		int e = WSAGetLastError();
+#else
+		int e = errno;
+#endif
+		//cerr << "<R:" << r << ">";
+
+		if (r > 0) {
+			total_bytes_read += r;
+
+			// Add a null-terminator at the the end of the buffer
+			// that we will send to the callback.
+			// DO NOT EVER CHANGE THIS. We want to explicitly allow users
+			// to be able to depend on this behavior, so they will have
+			// the option to do some things faster. Additionally it's
+			// a security guard against buffer overflows.
+			readbuffer [r] = 0;
+			_DispatchInboundData (readbuffer, r);
+			if (bPaused)
+				break;
+		}
+		else if (r == 0) {
+			break;
+		}
+		else {
+			#ifdef OS_UNIX
+			if ((e != EINPROGRESS) && (e != EWOULDBLOCK) && (e != EAGAIN) && (e != EINTR)) {
+			#endif
+			#ifdef OS_WIN32
+			if ((e != WSAEINPROGRESS) && (e != WSAEWOULDBLOCK)) {
+			#endif
+				// 26Mar11: Previously, all read errors were assumed to be EWOULDBLOCK and ignored.
+				// Now, instead, we call Close() on errors like ECONNRESET and ENOTCONN.
+				UnbindReasonCode = e;
+				Close();
+				break;
+			} else {
+				// Basically a would-block, meaning we've read everything there is to read.
+				break;
+			}
+		}
+
+	}
+
+
+	if (total_bytes_read == 0) {
+		// If we read no data on a socket that selected readable,
+		// it generally means the other end closed the connection gracefully.
+		ScheduleClose (false);
+		//bCloseNow = true;
+	}
+
+}
+
+
+
+/******************************************
+ConnectionDescriptor::_DispatchInboundData
+******************************************/
+
+#ifdef WITH_SSL
+void ConnectionDescriptor::_DispatchInboundData (const char *buffer, unsigned long size)
+{
+	if (SslBox) {
+		SslBox->PutCiphertext (buffer, size);
+
+		int s;
+		char B [2048];
+		while ((s = SslBox->GetPlaintext (B, sizeof(B) - 1)) > 0) {
+			_CheckHandshakeStatus();
+			B [s] = 0;
+			_GenericInboundDispatch(B, s);
+		}
+
+		// If our SSL handshake had a problem, shut down the connection.
+		if (s == -2) {
+			#ifdef OS_UNIX
+			UnbindReasonCode = EPROTO;
+			#endif
+			#ifdef OS_WIN32
+			UnbindReasonCode = WSAECONNABORTED;
+			#endif
+			ScheduleClose(false);
+			return;
+		}
+
+		_CheckHandshakeStatus();
+		_DispatchCiphertext();
+	}
+	else {
+		_GenericInboundDispatch(buffer, size);
+	}
+}
+#else
+void ConnectionDescriptor::_DispatchInboundData (const char *buffer, unsigned long size)
+{
+	_GenericInboundDispatch(buffer, size);
+}
+#endif
+
+
+
+/*******************************************
+ConnectionDescriptor::_CheckHandshakeStatus
+*******************************************/
+
+void ConnectionDescriptor::_CheckHandshakeStatus()
+{
+	#ifdef WITH_SSL
+	if (SslBox && (!bHandshakeSignaled) && SslBox->IsHandshakeCompleted()) {
+		bHandshakeSignaled = true;
+		if (EventCallback)
+			(*EventCallback)(GetBinding(), EM_SSL_HANDSHAKE_COMPLETED, NULL, 0);
+	}
+	#endif
+}
+
+
+
+/***************************
+ConnectionDescriptor::Write
+***************************/
+
+void ConnectionDescriptor::Write()
+{
+	/* A socket which is in a pending-connect state will select
+	 * writable when the disposition of the connect is known.
+	 * At that point, check to be sure there are no errors,
+	 * and if none, then promote the socket out of the pending
+	 * state.
+	 * TODO: I haven't figured out how Windows signals errors on
+	 * unconnected sockets. Maybe it does the untraditional but
+	 * logical thing and makes the socket selectable for error.
+	 * If so, it's unsupported here for the time being, and connect
+	 * errors will have to be caught by the timeout mechanism.
+	 */
+
+	if (bConnectPending) {
+		int error;
+		socklen_t len;
+		len = sizeof(error);
+		#ifdef OS_UNIX
+		int o = getsockopt (GetSocket(), SOL_SOCKET, SO_ERROR, &error, &len);
+		#endif
+		#ifdef OS_WIN32
+		int o = getsockopt (GetSocket(), SOL_SOCKET, SO_ERROR, (char*)&error, &len);
+		#endif
+		if ((o == 0) && (error == 0)) {
+			if (EventCallback)
+				(*EventCallback)(GetBinding(), EM_CONNECTION_COMPLETED, "", 0);
+
+			// 5May09: Moved epoll/kqueue read/write arming into SetConnectPending, so it can be called
+			// from EventMachine_t::AttachFD as well.
+			SetConnectPending (false);
+		}
+		else {
+			if (o == 0)
+				UnbindReasonCode = error;
+			ScheduleClose (false);
+			//bCloseNow = true;
+		}
+	}
+	else {
+
+		if (bNotifyWritable) {
+			if (EventCallback)
+				(*EventCallback)(GetBinding(), EM_CONNECTION_NOTIFY_WRITABLE, NULL, 0);
+
+			_UpdateEvents(false, true);
+			return;
+		}
+
+		assert(!bWatchOnly);
+
+		/* 5May09: Kqueue bugs on OSX cause one extra writable event to fire even though we're using
+		   EV_ONESHOT. We ignore this extra event once, but only the first time. If it happens again,
+		   we should fall through to the assert(nbytes>0) failure to catch any EM bugs which might cause
+		   ::Write to be called in a busy-loop.
+		*/
+		#ifdef HAVE_KQUEUE
+		if (MyEventMachine->GetPoller() == Poller_Kqueue) {
+			if (OutboundDataSize == 0 && !bGotExtraKqueueEvent) {
+				bGotExtraKqueueEvent = true;
+				return;
+			} else if (OutboundDataSize > 0) {
+				bGotExtraKqueueEvent = false;
+			}
+		}
+		#endif
+
+		_WriteOutboundData();
+	}
+}
+
+
+/****************************************
+ConnectionDescriptor::_WriteOutboundData
+****************************************/
+
+void ConnectionDescriptor::_WriteOutboundData()
+{
+	/* This is a helper function called by ::Write.
+	 * It's possible for a socket to select writable and then no longer
+	 * be writable by the time we get around to writing. The kernel might
+	 * have used up its available output buffers between the select call
+	 * and when we get here. So this condition is not an error.
+	 *
+	 * 20Jul07, added the same kind of protection against an invalid socket
+	 * that is at the top of ::Read. Not entirely how this could happen in 
+	 * real life (connection-reset from the remote peer, perhaps?), but I'm
+	 * doing it to address some reports of crashing under heavy loads.
+	 */
+
+	SOCKET sd = GetSocket();
+	//assert (sd != INVALID_SOCKET);
+	if (sd == INVALID_SOCKET) {
+		assert (!bWriteAttemptedAfterClose);
+		bWriteAttemptedAfterClose = true;
+		return;
+	}
+
+	LastActivity = MyEventMachine->GetCurrentLoopTime();
+	size_t nbytes = 0;
+
+	#ifdef HAVE_WRITEV
+	int iovcnt = OutboundPages.size();
+	// Max of 16 outbound pages at a time
+	if (iovcnt > 16) iovcnt = 16;
+
+	iovec iov[16];
+
+	for(int i = 0; i < iovcnt; i++){
+		OutboundPage *op = &(OutboundPages[i]);
+		#ifdef CC_SUNWspro
+		// TODO: The void * cast works fine on Solaris 11, but
+		// I don't know at what point that changed from older Solaris.
+		iov[i].iov_base = (char *)(op->Buffer + op->Offset);
+		#else
+		iov[i].iov_base = (void *)(op->Buffer + op->Offset);
+		#endif
+		iov[i].iov_len	= op->Length - op->Offset;
+
+		nbytes += iov[i].iov_len;
+	}
+	#else
+	char output_buffer [16 * 1024];
+
+	while ((OutboundPages.size() > 0) && (nbytes < sizeof(output_buffer))) {
+		OutboundPage *op = &(OutboundPages[0]);
+		if ((nbytes + op->Length - op->Offset) < sizeof (output_buffer)) {
+			memcpy (output_buffer + nbytes, op->Buffer + op->Offset, op->Length - op->Offset);
+			nbytes += (op->Length - op->Offset);
+			op->Free();
+			OutboundPages.pop_front();
+		}
+		else {
+			int len = sizeof(output_buffer) - nbytes;
+			memcpy (output_buffer + nbytes, op->Buffer + op->Offset, len);
+			op->Offset += len;
+			nbytes += len;
+		}
+	}
+	#endif
+
+	// We should never have gotten here if there were no data to write,
+	// so assert that as a sanity check.
+	// Don't bother to make sure nbytes is less than output_buffer because
+	// if it were we probably would have crashed already.
+	assert (nbytes > 0);
+
+	assert (GetSocket() != INVALID_SOCKET);
+	#ifdef HAVE_WRITEV
+	int bytes_written = writev (GetSocket(), iov, iovcnt);
+	#else
+	int bytes_written = write (GetSocket(), output_buffer, nbytes);
+	#endif
+
+	bool err = false;
+#ifdef OS_WIN32
+	int e = WSAGetLastError();
+#else
+	int e = errno;
+#endif
+	if (bytes_written < 0) {
+		err = true;
+		bytes_written = 0;
+	}
+
+	assert (bytes_written >= 0);
+	OutboundDataSize -= bytes_written;
+
+	if (ProxiedFrom && MaxOutboundBufSize && (unsigned int)GetOutboundDataSize() < MaxOutboundBufSize && ProxiedFrom->IsPaused())
+		ProxiedFrom->Resume();
+
+	#ifdef HAVE_WRITEV
+	if (!err) {
+		unsigned int sent = bytes_written;
+		deque<OutboundPage>::iterator op = OutboundPages.begin();
+
+		for (int i = 0; i < iovcnt; i++) {
+			if (iov[i].iov_len <= sent) {
+				// Sent this page in full, free it.
+				op->Free();
+				OutboundPages.pop_front();
+
+				sent -= iov[i].iov_len;
+			} else {
+				// Sent part (or none) of this page, increment offset to send the remainder
+				op->Offset += sent;
+				break;
+			}
+
+			// Shouldn't be possible run out of pages before the loop ends
+			assert(op != OutboundPages.end());
+			*op++;
+		}
+	}
+	#else
+	if ((size_t)bytes_written < nbytes) {
+		int len = nbytes - bytes_written;
+		char *buffer = (char*) malloc (len + 1);
+		if (!buffer)
+			throw std::runtime_error ("bad alloc throwing back data");
+		memcpy (buffer, output_buffer + bytes_written, len);
+		buffer [len] = 0;
+		OutboundPages.push_front (OutboundPage (buffer, len));
+	}
+	#endif
+
+	_UpdateEvents(false, true);
+
+	if (err) {
+		#ifdef OS_UNIX
+		if ((e != EINPROGRESS) && (e != EWOULDBLOCK) && (e != EINTR)) {
+		#endif
+		#ifdef OS_WIN32
+		if ((e != WSAEINPROGRESS) && (e != WSAEWOULDBLOCK)) {
+		#endif
+			UnbindReasonCode = e;
+			Close();
+		}
+	}
+}
+
+
+/***************************************
+ConnectionDescriptor::ReportErrorStatus
+***************************************/
+
+int ConnectionDescriptor::ReportErrorStatus()
+{
+	if (MySocket == INVALID_SOCKET) {
+		return -1;
+	}
+
+	int error;
+	socklen_t len;
+	len = sizeof(error);
+	#ifdef OS_UNIX
+	int o = getsockopt (GetSocket(), SOL_SOCKET, SO_ERROR, &error, &len);
+	#endif
+	#ifdef OS_WIN32
+	int o = getsockopt (GetSocket(), SOL_SOCKET, SO_ERROR, (char*)&error, &len);
+	#endif
+	if ((o == 0) && (error == 0))
+		return 0;
+	else if (o == 0)
+		return error;
+	else
+		return -1;
+}
+
+
+/******************************
+ConnectionDescriptor::StartTls
+******************************/
+
+#ifdef WITH_SSL
+void ConnectionDescriptor::StartTls()
+{
+	if (SslBox)
+		throw std::runtime_error ("SSL/TLS already running on connection");
+
+	SslBox = new SslBox_t (bIsServer, PrivateKeyFilename, CertChainFilename, bSslVerifyPeer, bSslFailIfNoPeerCert, SniHostName, CipherList, EcdhCurve, DhParam, Protocols, GetBinding());
+	_DispatchCiphertext();
+
+}
+#else
+void ConnectionDescriptor::StartTls()
+{
+	throw std::runtime_error ("Encryption not available on this event-machine");
+}
+#endif
+
+
+/*********************************
+ConnectionDescriptor::SetTlsParms
+*********************************/
+
+#ifdef WITH_SSL
+void ConnectionDescriptor::SetTlsParms (const char *privkey_filename, const char *certchain_filename, bool verify_peer, bool fail_if_no_peer_cert, const char *sni_hostname, const char *cipherlist, const char *ecdh_curve, const char *dhparam, int protocols)
+{
+	if (SslBox)
+		throw std::runtime_error ("call SetTlsParms before calling StartTls");
+	if (privkey_filename && *privkey_filename)
+		PrivateKeyFilename = privkey_filename;
+	if (certchain_filename && *certchain_filename)
+		CertChainFilename = certchain_filename;
+	bSslVerifyPeer     = verify_peer;
+	bSslFailIfNoPeerCert = fail_if_no_peer_cert;
+
+	if (sni_hostname && *sni_hostname)
+		SniHostName = sni_hostname;
+	if (cipherlist && *cipherlist)
+		CipherList = cipherlist;
+	if (ecdh_curve && *ecdh_curve)
+		EcdhCurve = ecdh_curve;
+	if (dhparam && *dhparam)
+		DhParam = dhparam;
+
+	Protocols = protocols;
+}
+#else
+void ConnectionDescriptor::SetTlsParms (const char *privkey_filename UNUSED, const char *certchain_filename UNUSED, bool verify_peer UNUSED, bool fail_if_no_peer_cert UNUSED, const char *sni_hostname UNUSED, const char *cipherlist UNUSED, const char *ecdh_curve UNUSED, const char *dhparam UNUSED, int protocols UNUSED)
+{
+	throw std::runtime_error ("Encryption not available on this event-machine");
+}
+#endif
+
+
+/*********************************
+ConnectionDescriptor::GetPeerCert
+*********************************/
+
+#ifdef WITH_SSL
+X509 *ConnectionDescriptor::GetPeerCert()
+{
+	if (!SslBox)
+		throw std::runtime_error ("SSL/TLS not running on this connection");
+	return SslBox->GetPeerCert();
+}
+#endif
+
+
+/*********************************
+ConnectionDescriptor::GetCipherBits
+*********************************/
+
+#ifdef WITH_SSL
+int ConnectionDescriptor::GetCipherBits()
+{
+	if (!SslBox)
+		throw std::runtime_error ("SSL/TLS not running on this connection");
+	return SslBox->GetCipherBits();
+}
+#endif
+
+
+/*********************************
+ConnectionDescriptor::GetCipherName
+*********************************/
+
+#ifdef WITH_SSL
+const char *ConnectionDescriptor::GetCipherName()
+{
+	if (!SslBox)
+		throw std::runtime_error ("SSL/TLS not running on this connection");
+	return SslBox->GetCipherName();
+}
+#endif
+
+
+/*********************************
+ConnectionDescriptor::GetCipherProtocol
+*********************************/
+
+#ifdef WITH_SSL
+const char *ConnectionDescriptor::GetCipherProtocol()
+{
+	if (!SslBox)
+		throw std::runtime_error ("SSL/TLS not running on this connection");
+	return SslBox->GetCipherProtocol();
+}
+#endif
+
+
+/*********************************
+ConnectionDescriptor::GetSNIHostname
+*********************************/
+
+#ifdef WITH_SSL
+const char *ConnectionDescriptor::GetSNIHostname()
+{
+	if (!SslBox)
+		throw std::runtime_error ("SSL/TLS not running on this connection");
+	return SslBox->GetSNIHostname();
+}
+#endif
+
+
+/***********************************
+ConnectionDescriptor::VerifySslPeer
+***********************************/
+
+#ifdef WITH_SSL
+bool ConnectionDescriptor::VerifySslPeer(const char *cert)
+{
+	bSslPeerAccepted = false;
+
+	if (EventCallback)
+		(*EventCallback)(GetBinding(), EM_SSL_VERIFY, cert, strlen(cert));
+
+	return bSslPeerAccepted;
+}
+#endif
+
+
+/***********************************
+ConnectionDescriptor::AcceptSslPeer
+***********************************/
+
+#ifdef WITH_SSL
+void ConnectionDescriptor::AcceptSslPeer()
+{
+	bSslPeerAccepted = true;
+}
+#endif
+
+
+/*****************************************
+ConnectionDescriptor::_DispatchCiphertext
+*****************************************/
+
+#ifdef WITH_SSL
+void ConnectionDescriptor::_DispatchCiphertext()
+{
+	assert (SslBox);
+
+
+	char BigBuf [SSLBOX_OUTPUT_CHUNKSIZE];
+	bool did_work;
+
+	do {
+		did_work = false;
+
+		// try to drain ciphertext
+		while (SslBox->CanGetCiphertext()) {
+			int r = SslBox->GetCiphertext (BigBuf, sizeof(BigBuf));
+			assert (r > 0);
+			_SendRawOutboundData (BigBuf, r);
+			did_work = true;
+		}
+
+		// Pump the SslBox, in case it has queued outgoing plaintext
+		// This will return >0 if data was written,
+		// 0 if no data was written, and <0 if there was a fatal error.
+		bool pump;
+		do {
+			pump = false;
+			int w = SslBox->PutPlaintext (NULL, 0);
+			if (w > 0) {
+				did_work = true;
+				pump = true;
+			}
+			else if (w < 0)
+				ScheduleClose (false);
+		} while (pump);
+
+		// try to put plaintext. INCOMPLETE, doesn't belong here?
+		// In SendOutboundData, we're spooling plaintext directly
+		// into SslBox. That may be wrong, we may need to buffer it
+		// up here! 
+		/*
+		const char *ptr;
+		int ptr_length;
+		while (OutboundPlaintext.GetPage (&ptr, &ptr_length)) {
+			assert (ptr && (ptr_length > 0));
+			int w = SslMachine.PutPlaintext (ptr, ptr_length);
+			if (w > 0) {
+				OutboundPlaintext.DiscardBytes (w);
+				did_work = true;
+			}
+			else
+				break;
+		}
+		*/
+
+	} while (did_work);
+
+}
+#endif
+
+
+
+/*******************************
+ConnectionDescriptor::Heartbeat
+*******************************/
+
+void ConnectionDescriptor::Heartbeat()
+{
+	/* Only allow a certain amount of time to go by while waiting
+	 * for a pending connect. If it expires, then kill the socket.
+	 * For a connected socket, close it if its inactivity timer
+	 * has expired.
+	 */
+
+	if (bConnectPending) {
+		if ((MyEventMachine->GetCurrentLoopTime() - CreatedAt) >= PendingConnectTimeout) {
+			UnbindReasonCode = ETIMEDOUT;
+			ScheduleClose (false);
+			//bCloseNow = true;
+		}
+	}
+	else {
+		if (InactivityTimeout && ((MyEventMachine->GetCurrentLoopTime() - LastActivity) >= InactivityTimeout)) {
+			UnbindReasonCode = ETIMEDOUT;
+			ScheduleClose (false);
+			//bCloseNow = true;
+		}
+	}
+}
+
+
+/****************************************
+LoopbreakDescriptor::LoopbreakDescriptor
+****************************************/
+
+LoopbreakDescriptor::LoopbreakDescriptor (SOCKET sd, EventMachine_t *parent_em):
+	EventableDescriptor (sd, parent_em)
+{
+	/* This is really bad and ugly. Change someday if possible.
+	 * We have to know about an event-machine (probably the one that owns us),
+	 * so we can pass newly-created connections to it.
+	 */
+
+	bCallbackUnbind = false;
+
+	#ifdef HAVE_EPOLL
+	EpollEvent.events = EPOLLIN;
+	#endif
+	#ifdef HAVE_KQUEUE
+	MyEventMachine->ArmKqueueReader (this);
+	#endif
+}
+
+
+
+
+/*************************
+LoopbreakDescriptor::Read
+*************************/
+
+void LoopbreakDescriptor::Read()
+{
+	// TODO, refactor, this code is probably in the wrong place.
+	assert (MyEventMachine);
+	MyEventMachine->_ReadLoopBreaker();
+}
+
+
+/**************************
+LoopbreakDescriptor::Write
+**************************/
+
+void LoopbreakDescriptor::Write()
+{
+	// Why are we here?
+	throw std::runtime_error ("bad code path in loopbreak");
+}
+
+/**************************************
+AcceptorDescriptor::AcceptorDescriptor
+**************************************/
+
+AcceptorDescriptor::AcceptorDescriptor (SOCKET sd, EventMachine_t *parent_em):
+	EventableDescriptor (sd, parent_em)
+{
+	#ifdef HAVE_EPOLL
+	EpollEvent.events = EPOLLIN;
+	#endif
+	#ifdef HAVE_KQUEUE
+	MyEventMachine->ArmKqueueReader (this);
+	#endif
+}
+
+
+/***************************************
+AcceptorDescriptor::~AcceptorDescriptor
+***************************************/
+
+AcceptorDescriptor::~AcceptorDescriptor()
+{
+}
+
+/****************************************
+STATIC: AcceptorDescriptor::StopAcceptor
+****************************************/
+
+void AcceptorDescriptor::StopAcceptor (const uintptr_t binding)
+{
+	// TODO: This is something of a hack, or at least it's a static method of the wrong class.
+	AcceptorDescriptor *ad = dynamic_cast <AcceptorDescriptor*> (Bindable_t::GetObject (binding));
+	if (ad)
+		ad->ScheduleClose (false);
+	else
+		throw std::runtime_error ("failed to close nonexistent acceptor");
+}
+
+
+/************************
+AcceptorDescriptor::Read
+************************/
+
+void AcceptorDescriptor::Read()
+{
+	/* Accept up to a certain number of sockets on the listening connection.
+	 * Don't try to accept all that are present, because this would allow a DoS attack
+	 * in which no data were ever read or written. We should accept more than one,
+	 * if available, to keep the partially accepted sockets from backing up in the kernel.
+	 */
+
+	/* Make sure we use non-blocking i/o on the acceptor socket, since we're selecting it
+	 * for readability. According to Stevens UNP, it's possible for an acceptor to select readable
+	 * and then block when we call accept. For example, the other end resets the connection after
+	 * the socket selects readable and before we call accept. The kernel will remove the dead
+	 * socket from the accept queue. If the accept queue is now empty, accept will block.
+	 */
+
+
+	struct sockaddr_in pin;
+	socklen_t addrlen = sizeof (pin);
+	int accept_count = EventMachine_t::GetSimultaneousAcceptCount();
+
+	for (int i=0; i < accept_count; i++) {
+#if defined(HAVE_CONST_SOCK_CLOEXEC) && defined(HAVE_ACCEPT4)
+		SOCKET sd = accept4 (GetSocket(), (struct sockaddr*)&pin, &addrlen, SOCK_CLOEXEC);
+		if (sd == INVALID_SOCKET) {
+			// We may be running in a kernel where
+			// SOCK_CLOEXEC is not supported - fall back:
+			sd = accept (GetSocket(), (struct sockaddr*)&pin, &addrlen);
+		}
+#else
+		SOCKET sd = accept (GetSocket(), (struct sockaddr*)&pin, &addrlen);
+#endif
+		if (sd == INVALID_SOCKET) {
+			// This breaks the loop when we've accepted everything on the kernel queue,
+			// up to 10 new connections. But what if the *first* accept fails?
+			// Does that mean anything serious is happening, beyond the situation
+			// described in the note above?
+			break;
+		}
+
+		// Set the newly-accepted socket non-blocking and to close on exec.
+		// On Windows, this may fail because, weirdly, Windows inherits the non-blocking
+		// attribute that we applied to the acceptor socket into the accepted one.
+		if (!SetFdCloexec(sd) || !SetSocketNonblocking (sd)) {
+		//int val = fcntl (sd, F_GETFL, 0);
+		//if (fcntl (sd, F_SETFL, val | O_NONBLOCK) == -1) {
+			shutdown (sd, 1);
+			close (sd);
+			continue;
+		}
+
+		// Disable slow-start (Nagle algorithm). Eventually make this configurable.
+		int one = 1;
+		setsockopt (sd, IPPROTO_TCP, TCP_NODELAY, (char*) &one, sizeof(one));
+
+
+		ConnectionDescriptor *cd = new ConnectionDescriptor (sd, MyEventMachine);
+		if (!cd)
+			throw std::runtime_error ("no newly accepted connection");
+		cd->SetServerMode();
+		if (EventCallback) {
+			(*EventCallback) (GetBinding(), EM_CONNECTION_ACCEPTED, NULL, cd->GetBinding());
+		}
+		#ifdef HAVE_EPOLL
+		cd->GetEpollEvent()->events = 0;
+		if (cd->SelectForRead())
+			cd->GetEpollEvent()->events |= EPOLLIN;
+		if (cd->SelectForWrite())
+			cd->GetEpollEvent()->events |= EPOLLOUT;
+		#endif
+		assert (MyEventMachine);
+		MyEventMachine->Add (cd);
+		#ifdef HAVE_KQUEUE
+		bKqueueArmWrite = cd->SelectForWrite();
+		if (bKqueueArmWrite)
+			MyEventMachine->Modify (cd);
+		if (cd->SelectForRead())
+			MyEventMachine->ArmKqueueReader (cd);
+		#endif
+	}
+
+}
+
+
+/*************************
+AcceptorDescriptor::Write
+*************************/
+
+void AcceptorDescriptor::Write()
+{
+	// Why are we here?
+	throw std::runtime_error ("bad code path in acceptor");
+}
+
+
+/*****************************
+AcceptorDescriptor::Heartbeat
+*****************************/
+
+void AcceptorDescriptor::Heartbeat()
+{
+	// No-op
+}
+
+
+/*******************************
+AcceptorDescriptor::GetSockname
+*******************************/
+
+bool AcceptorDescriptor::GetSockname (struct sockaddr *s, socklen_t *len)
+{
+	bool ok = false;
+	if (s) {
+		int gp = getsockname (GetSocket(), s, len);
+		if (gp == 0)
+			ok = true;
+	}
+	return ok;
+}
+
+
+
+/**************************************
+DatagramDescriptor::DatagramDescriptor
+**************************************/
+
+DatagramDescriptor::DatagramDescriptor (SOCKET sd, EventMachine_t *parent_em):
+	EventableDescriptor (sd, parent_em),
+	OutboundDataSize (0)
+{
+	memset (&ReturnAddress, 0, sizeof(ReturnAddress));
+
+	/* Provisionally added 19Oct07. All datagram sockets support broadcasting.
+	 * Until now, sending to a broadcast address would give EACCES (permission denied)
+	 * on systems like Linux and BSD that require the SO_BROADCAST socket-option in order
+	 * to accept a packet to a broadcast address. Solaris doesn't require it. I think
+	 * Windows DOES require it but I'm not sure.
+	 *
+	 * Ruby does NOT do what we're doing here. In Ruby, you have to explicitly set SO_BROADCAST
+	 * on a UDP socket in order to enable broadcasting. The reason for requiring the option
+	 * in the first place is so that applications don't send broadcast datagrams by mistake.
+	 * I imagine that could happen if a user of an application typed in an address that happened
+	 * to be a broadcast address on that particular subnet.
+	 *
+	 * This is provisional because someone may eventually come up with a good reason not to
+	 * do it for all UDP sockets. If that happens, then we'll need to add a usercode-level API
+	 * to set the socket option, just like Ruby does. AND WE'LL ALSO BREAK CODE THAT DOESN'T
+	 * EXPLICITLY SET THE OPTION.
+	 */
+
+	int oval = 1;
+	setsockopt (GetSocket(), SOL_SOCKET, SO_BROADCAST, (char*)&oval, sizeof(oval));
+
+	#ifdef HAVE_EPOLL
+	EpollEvent.events = EPOLLIN;
+	#endif
+	#ifdef HAVE_KQUEUE
+	MyEventMachine->ArmKqueueReader (this);
+	#endif
+}
+
+
+/***************************************
+DatagramDescriptor::~DatagramDescriptor
+***************************************/
+
+DatagramDescriptor::~DatagramDescriptor()
+{
+	// Run down any stranded outbound data.
+	for (size_t i=0; i < OutboundPages.size(); i++)
+		OutboundPages[i].Free();
+}
+
+
+/*****************************
+DatagramDescriptor::Heartbeat
+*****************************/
+
+void DatagramDescriptor::Heartbeat()
+{
+	// Close it if its inactivity timer has expired.
+
+	if (InactivityTimeout && ((MyEventMachine->GetCurrentLoopTime() - LastActivity) >= InactivityTimeout))
+		ScheduleClose (false);
+		//bCloseNow = true;
+}
+
+
+/************************
+DatagramDescriptor::Read
+************************/
+
+void DatagramDescriptor::Read()
+{
+	SOCKET sd = GetSocket();
+	assert (sd != INVALID_SOCKET);
+	LastActivity = MyEventMachine->GetCurrentLoopTime();
+
+	// This is an extremely large read buffer.
+	// In many cases you wouldn't expect to get any more than 4K.
+	char readbuffer [16 * 1024];
+
+	for (int i=0; i < 10; i++) {
+		// Don't read just one buffer and then move on. This is faster
+		// if there is a lot of incoming.
+		// But don't read indefinitely. Give other sockets a chance to run.
+		// NOTICE, we're reading one less than the buffer size.
+		// That's so we can put a guard byte at the end of what we send
+		// to user code.
+
+		struct sockaddr_in sin;
+		socklen_t slen = sizeof (sin);
+		memset (&sin, 0, slen);
+
+		int r = recvfrom (sd, readbuffer, sizeof(readbuffer) - 1, 0, (struct sockaddr*)&sin, &slen);
+		//cerr << "<R:" << r << ">";
+
+		// In UDP, a zero-length packet is perfectly legal.
+		if (r >= 0) {
+
+			// Add a null-terminator at the the end of the buffer
+			// that we will send to the callback.
+			// DO NOT EVER CHANGE THIS. We want to explicitly allow users
+			// to be able to depend on this behavior, so they will have
+			// the option to do some things faster. Additionally it's
+			// a security guard against buffer overflows.
+			readbuffer [r] = 0;
+
+
+			// Set up a "temporary" return address so that callers can "reply" to us
+			// from within the callback we are about to invoke. That means that ordinary
+			// calls to "send_data_to_connection" (which is of course misnamed in this
+			// case) will result in packets being sent back to the same place that sent
+			// us this one.
+			// There is a different call (evma_send_datagram) for cases where the caller
+			// actually wants to send a packet somewhere else.
+
+			memset (&ReturnAddress, 0, sizeof(ReturnAddress));
+			memcpy (&ReturnAddress, &sin, slen);
+
+			_GenericInboundDispatch(readbuffer, r);
+
+		}
+		else {
+			// Basically a would-block, meaning we've read everything there is to read.
+			break;
+		}
+
+	}
+
+
+}
+
+
+/*************************
+DatagramDescriptor::Write
+*************************/
+
+void DatagramDescriptor::Write()
+{
+	/* It's possible for a socket to select writable and then no longer
+	 * be writable by the time we get around to writing. The kernel might
+	 * have used up its available output buffers between the select call
+	 * and when we get here. So this condition is not an error.
+	 * This code is very reminiscent of ConnectionDescriptor::_WriteOutboundData,
+	 * but differs in the that the outbound data pages (received from the
+	 * user) are _message-structured._ That is, we send each of them out
+	 * one message at a time.
+	 * TODO, we are currently suppressing the EMSGSIZE error!!!
+	 */
+
+	SOCKET sd = GetSocket();
+	assert (sd != INVALID_SOCKET);
+	LastActivity = MyEventMachine->GetCurrentLoopTime();
+
+	assert (OutboundPages.size() > 0);
+
+	// Send out up to 10 packets, then cycle the machine.
+	for (int i = 0; i < 10; i++) {
+		if (OutboundPages.size() <= 0)
+			break;
+		OutboundPage *op = &(OutboundPages[0]);
+
+		// The nasty cast to (char*) is needed because Windows is brain-dead.
+		int s = sendto (sd, (char*)op->Buffer, op->Length, 0, (struct sockaddr*)&(op->From),
+		               (op->From.sin6_family == AF_INET6 ? sizeof (struct sockaddr_in6) : sizeof (struct sockaddr_in)));
+#ifdef OS_WIN32
+		int e = WSAGetLastError();
+#else
+		int e = errno;
+#endif
+
+		OutboundDataSize -= op->Length;
+		op->Free();
+		OutboundPages.pop_front();
+
+		if (s == SOCKET_ERROR) {
+			#ifdef OS_UNIX
+			if ((e != EINPROGRESS) && (e != EWOULDBLOCK) && (e != EINTR)) {
+			#endif
+			#ifdef OS_WIN32
+			if ((e != WSAEINPROGRESS) && (e != WSAEWOULDBLOCK)) {
+			#endif
+				UnbindReasonCode = e;
+				Close();
+				break;
+			}
+		}
+	}
+
+	#ifdef HAVE_EPOLL
+	EpollEvent.events = EPOLLIN;
+	if (SelectForWrite())
+		EpollEvent.events |= EPOLLOUT;
+	assert (MyEventMachine);
+	MyEventMachine->Modify (this);
+	#endif
+	#ifdef HAVE_KQUEUE
+	bKqueueArmWrite = SelectForWrite();
+	assert (MyEventMachine);
+	MyEventMachine->Modify (this);
+	#endif
+}
+
+
+/**********************************
+DatagramDescriptor::SelectForWrite
+**********************************/
+
+bool DatagramDescriptor::SelectForWrite()
+{
+	/* Changed 15Nov07, per bug report by Mark Zvillius.
+	 * The outbound data size will be zero if there are zero-length outbound packets,
+	 * so we now select writable in case the outbound page buffer is not empty.
+	 * Note that the superclass ShouldDelete method still checks for outbound data size,
+	 * which may be wrong.
+	 */
+	//return (GetOutboundDataSize() > 0); (Original)
+	return (OutboundPages.size() > 0);
+}
+
+
+/************************************
+DatagramDescriptor::SendOutboundData
+************************************/
+
+int DatagramDescriptor::SendOutboundData (const char *data, unsigned long length)
+{
+	// This is almost an exact clone of ConnectionDescriptor::_SendRawOutboundData.
+	// That means most of it could be factored to a common ancestor. Note that
+	// empty datagrams are meaningful, which isn't the case for TCP streams.
+
+	if (IsCloseScheduled())
+		return 0;
+
+	if (!data && (length > 0))
+		throw std::runtime_error ("bad outbound data");
+	char *buffer = (char *) malloc (length + 1);
+	if (!buffer)
+		throw std::runtime_error ("no allocation for outbound data");
+	memcpy (buffer, data, length);
+	buffer [length] = 0;
+	OutboundPages.push_back (OutboundPage (buffer, length, ReturnAddress));
+	OutboundDataSize += length;
+
+	#ifdef HAVE_EPOLL
+	EpollEvent.events = (EPOLLIN | EPOLLOUT);
+	assert (MyEventMachine);
+	MyEventMachine->Modify (this);
+	#endif
+	#ifdef HAVE_KQUEUE
+	bKqueueArmWrite = true;
+	assert (MyEventMachine);
+	MyEventMachine->Modify (this);
+	#endif
+
+	return length;
+}
+
+
+/****************************************
+DatagramDescriptor::SendOutboundDatagram
+****************************************/
+
+int DatagramDescriptor::SendOutboundDatagram (const char *data, unsigned long length, const char *address, int port)
+{
+	// This is an exact clone of ConnectionDescriptor::SendOutboundData.
+	// That means it needs to move to a common ancestor.
+	// TODO: Refactor this so there's no overlap with SendOutboundData.
+
+	if (IsCloseScheduled())
+	//if (bCloseNow || bCloseAfterWriting)
+		return 0;
+
+	if (!address || !*address || !port)
+		return 0;
+
+	struct sockaddr_in6 addr_here;
+	size_t addr_here_len = sizeof addr_here;
+	if (!EventMachine_t::name2address (address, port, (struct sockaddr *)&addr_here, &addr_here_len))
+		return -1;
+
+	if (!data && (length > 0))
+		throw std::runtime_error ("bad outbound data");
+	char *buffer = (char *) malloc (length + 1);
+	if (!buffer)
+		throw std::runtime_error ("no allocation for outbound data");
+	memcpy (buffer, data, length);
+	buffer [length] = 0;
+	OutboundPages.push_back (OutboundPage (buffer, length, addr_here));
+	OutboundDataSize += length;
+
+	#ifdef HAVE_EPOLL
+	EpollEvent.events = (EPOLLIN | EPOLLOUT);
+	assert (MyEventMachine);
+	MyEventMachine->Modify (this);
+	#endif
+	#ifdef HAVE_KQUEUE
+	bKqueueArmWrite = true;
+	assert (MyEventMachine);
+	MyEventMachine->Modify (this);
+	#endif
+
+	return length;
+}
+
+
+/*********************************
+ConnectionDescriptor::GetPeername
+*********************************/
+
+bool ConnectionDescriptor::GetPeername (struct sockaddr *s, socklen_t *len)
+{
+	bool ok = false;
+	if (s) {
+		int gp = getpeername (GetSocket(), s, len);
+		if (gp == 0)
+			ok = true;
+	}
+	return ok;
+}
+
+/*********************************
+ConnectionDescriptor::GetSockname
+*********************************/
+
+bool ConnectionDescriptor::GetSockname (struct sockaddr *s, socklen_t *len)
+{
+	bool ok = false;
+	if (s) {
+		int gp = getsockname (GetSocket(), s, len);
+		if (gp == 0)
+			ok = true;
+	}
+	return ok;
+}
+
+
+/**********************************************
+ConnectionDescriptor::GetCommInactivityTimeout
+**********************************************/
+
+uint64_t ConnectionDescriptor::GetCommInactivityTimeout()
+{
+	return InactivityTimeout / 1000;
+}
+
+
+/**********************************************
+ConnectionDescriptor::SetCommInactivityTimeout
+**********************************************/
+
+int ConnectionDescriptor::SetCommInactivityTimeout (uint64_t value)
+{
+	InactivityTimeout = value * 1000;
+	MyEventMachine->QueueHeartbeat(this);
+	return 1;
+}
+
+/*******************************
+DatagramDescriptor::GetPeername
+*******************************/
+
+bool DatagramDescriptor::GetPeername (struct sockaddr *s, socklen_t *len)
+{
+	bool ok = false;
+	if (s) {
+		*len = sizeof(struct sockaddr);
+		memset (s, 0, sizeof(struct sockaddr));
+		memcpy (s, &ReturnAddress, sizeof(ReturnAddress));
+		ok = true;
+	}
+	return ok;
+}
+
+/*******************************
+DatagramDescriptor::GetSockname
+*******************************/
+
+bool DatagramDescriptor::GetSockname (struct sockaddr *s, socklen_t *len)
+{
+	bool ok = false;
+	if (s) {
+		int gp = getsockname (GetSocket(), s, len);
+		if (gp == 0)
+			ok = true;
+	}
+	return ok;
+}
+
+
+
+/********************************************
+DatagramDescriptor::GetCommInactivityTimeout
+********************************************/
+
+uint64_t DatagramDescriptor::GetCommInactivityTimeout()
+{
+	return InactivityTimeout / 1000;
+}
+
+/********************************************
+DatagramDescriptor::SetCommInactivityTimeout
+********************************************/
+
+int DatagramDescriptor::SetCommInactivityTimeout (uint64_t value)
+{
+	if (value > 0) {
+		InactivityTimeout = value * 1000;
+		MyEventMachine->QueueHeartbeat(this);
+		return 1;
+	}
+	return 0;
+}
+
+
+/************************************
+InotifyDescriptor::InotifyDescriptor
+*************************************/
+
+InotifyDescriptor::InotifyDescriptor (EventMachine_t *em):
+	EventableDescriptor(0, em)
+{
+	bCallbackUnbind = false;
+
+	#ifndef HAVE_INOTIFY
+	throw std::runtime_error("no inotify support on this system");
+	#else
+
+	int fd = inotify_init();
+	if (fd == -1) {
+		char buf[200];
+		snprintf (buf, sizeof(buf)-1, "unable to create inotify descriptor: %s", strerror(errno));
+		throw std::runtime_error (buf);
+	}
+
+	MySocket = fd;
+	SetSocketNonblocking(MySocket);
+	#ifdef HAVE_EPOLL
+	EpollEvent.events = EPOLLIN;
+	#endif
+
+	#endif
+}
+
+
+/*************************************
+InotifyDescriptor::~InotifyDescriptor
+**************************************/
+
+InotifyDescriptor::~InotifyDescriptor()
+{
+	close(MySocket);
+	MySocket = INVALID_SOCKET;
+}
+
+/***********************
+InotifyDescriptor::Read
+************************/
+
+void InotifyDescriptor::Read()
+{
+	assert (MyEventMachine);
+	MyEventMachine->_ReadInotifyEvents();
+}
+
+
+/************************
+InotifyDescriptor::Write
+*************************/
+
+void InotifyDescriptor::Write()
+{
+	throw std::runtime_error("bad code path in inotify");
+}