firebird/src/remote/inet.cpp

/*
 *	PROGRAM:	JRD Remote Interface/Server
 *	MODULE:		inet.cpp
 *	DESCRIPTION:	TCP/UCP/IP Communications module.
 *
 * The contents of this file are subject to the Interbase Public
 * License Version 1.0 (the "License"); you may not use this file
 * except in compliance with the License. You may obtain a copy
 * of the License at http://www.Inprise.com/IPL.html
 *
 * Software distributed under the License is distributed on an
 * "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, either express
 * or implied. See the License for the specific language governing
 * rights and limitations under the License.
 *
 * The Original Code was created by Inprise Corporation
 * and its predecessors. Portions created by Inprise Corporation are
 * Copyright (C) Inprise Corporation.
 *
 * All Rights Reserved.
 * Contributor(s): ______________________________________.
 *
 * 2002.02.15 Sean Leyne - Code Cleanup, removed obsolete "EPSON" port
 * 2002.02.15 Sean Leyne - Code Cleanup, removed obsolete "XENIX" port
 * 2002.02.15 Sean Leyne - Code Cleanup, removed obsolete "IMP" port
 * 2002.02.15 Sean Leyne - Code Cleanup, removed obsolete "NCR3000" port
 *
 * 2002-02-23 Sean Leyne - Code Cleanup, removed old M88K and NCR3000 port
 *
 * 2002.10.27 Sean Leyne - Code Cleanup, removed obsolete "UNIXWARE" port
 * 2002.10.27 Sean Leyne - Code Cleanup, removed obsolete "Ultrix/MIPS" port
 *
 * 2002.10.28 Sean Leyne - Completed removal of obsolete "DGUX" port
 *
 * 2002.10.29 Sean Leyne - Removed obsolete "Netware" port
 *
 * 2002.10.30 Sean Leyne - Removed support for obsolete "PC_PLATFORM" define
 * 2002.10.30 Sean Leyne - Code Cleanup, removed obsolete "SUN3_3" port
 * 2005.04.01 Konstantin Kuznetsov - allow setting NoNagle option in Classic
 *
 */

#include "firebird.h"
#include <stdio.h>
#ifdef HAVE_SYS_TYPES_H
#include <sys/types.h>
#endif
#include <errno.h>
#include <string.h>
#include <stdlib.h>
#include "../common/file_params.h"
#include <stdarg.h>

#include "../common/classes/timestamp.h"
#include "../common/classes/init.h"
#include "../common/ThreadStart.h"
#include "../common/os/os_utils.h"

#ifdef HAVE_PWD_H
#include <pwd.h>
#endif

#ifdef HAVE_SYS_PARAM_H
#include <sys/param.h>
#endif

#ifdef HAVE_GRP_H
#include <grp.h>
#endif

#ifdef HAVE_SYS_SOCKET_H
#include <sys/socket.h> // for socket()
#endif

#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif



#ifdef	WIN_NT
#define FD_SETSIZE 2048
#endif

#ifndef WIN_NT
#include <netinet/tcp.h>
#include <netinet/in.h>
#include <netdb.h>
#include <arpa/inet.h>
#include <sys/wait.h>

#if defined(HAVE_POLL_H)
#include <poll.h>
#elif defined(HAVE_SYS_SELECT_H)
#include <sys/select.h>
#endif

#endif // !WIN_NT

const int INET_RETRY_CALL = 5;

#include "../remote/remote.h"
#include "../remote/SockAddr.h"
#include "ibase.h"
#include "../remote/inet_proto.h"
#include "../remote/proto_proto.h"
#include "../remote/remot_proto.h"
#include "../yvalve/gds_proto.h"
#include "../common/isc_proto.h"
#include "../common/isc_f_proto.h"
#include "../common/os/isc_i_proto.h"
#include "../common/config/config.h"
#include "../common/utils_proto.h"
#include "../common/classes/ClumpletWriter.h"

// Please review. Maybe not needed. See H_ERRNO in common.h.
#if defined HPUX
extern int h_errno;
#endif

using namespace Firebird;


#ifdef WIN_NT

#include <fcntl.h>
#include <process.h>
#include <signal.h>
#include "../utilities/install/install_nt.h"
#include <mstcpip.h>

#ifndef SIO_LOOPBACK_FAST_PATH
#define SIO_LOOPBACK_FAST_PATH              _WSAIOW(IOC_VENDOR,16)
#endif

#define INET_RETRY_ERRNO	WSAEINPROGRESS
#define INET_ADDR_IN_USE	WSAEADDRINUSE
#define sleep(seconds)  Sleep ((seconds) * 1000)
const int NOTASOCKET = WSAENOTSOCK;

#else // WIN_NT

#ifndef INET_ADDR_IN_USE
#define INET_ADDR_IN_USE EADDRINUSE
#endif
#ifndef INET_RETRY_ERRNO
#define INET_RETRY_ERRNO TRY_AGAIN
#endif
const int NOTASOCKET = EBADF;

#endif // WIN_NT

static void SOCLOSE(SOCKET& socket)
{
	SOCKET s = socket;
	if (s != INVALID_SOCKET)
	{
		socket = INVALID_SOCKET;
#ifdef WIN_NT
		closesocket(s);
#else
		close(s);
#endif
	}
}

#ifndef ENOBUFS
#define ENOBUFS	0
#endif

#ifndef FB_SEND_FLAGS
#define FB_SEND_FLAGS 0
#endif

#ifndef FB_SETOPT_FLAGS
#define FB_SETOPT_FLAGS 0
#endif

//
//#define DEBUG	1
//

#ifdef DEBUG
#ifdef HAVE_SYS_TIMEB_H
# include <sys/timeb.h>
#endif
const int TRACE_packets		= 1 << 0;	// bit 0
const int TRACE_operations	= 1 << 1;	// bit 1
const int TRACE_summary		= 1 << 2;	// bit 2

static int INET_trace = TRACE_summary | TRACE_packets | TRACE_operations;
static time_t INET_start_time = 0;
SLONG INET_force_error = -1;	// simulate a network error
static ULONG INET_count_send = 0;
static ULONG INET_count_recv = 0;
static ULONG INET_bytes_send = 0;
static ULONG INET_bytes_recv = 0;

static ULONG inet_debug_timer()
{
/**************************************
 *
 *	i n e t _ d e b u g _ t i m e r
 *
 **************************************
 *
 * Functional description
 *	Utility function used in DEBUG mode only to put a timestamp
 *	since start of connection on each debug output message.
 *
 *	This has been implemented and tested on SOLARIS, and may
 *	need tweeking on any other platform where DEBUG is needed.
 *
 **************************************/
#ifdef HAVE_GETTIMEOFDAY
	struct timeval tv;
	GETTIMEOFDAY(&tv);
	return (tv.tv_sec * 1000 + tv.tv_usec - INET_start_time);
#else
	struct timeb now;
	ftime(&now);
	return (now.time * 1000 + now.millitm - INET_start_time);
#endif // HAVE_GETTIMEOFDAY
}
#endif // DEBUG

const ULONG MAX_DATA_LW		= 1448;		// Low  Water mark
const ULONG MAX_DATA_HW		= 32768;	// High Water mark
const ULONG DEF_MAX_DATA	= 8192;

//const int MAXHOSTLEN		= 64;

const int SELECT_TIMEOUT	= 60;		// Dispatch thread select timeout (sec)

class Select
{
#ifdef HAVE_POLL
private:
	static const int SEL_INIT_EVENTS = POLLIN;
	static const int SEL_CHECK_MASK = POLLIN;

	pollfd* getPollFd(int n)
	{
		FB_SIZE_T pos;
		if (slct_poll.find(n, pos))
			return &slct_poll[pos];

		return nullptr;
	}
#endif

public:
#ifdef HAVE_POLL
	Select()
		: slct_time(0), slct_count(0), slct_poll(*getDefaultMemoryPool()),
		  slct_ready(*getDefaultMemoryPool())
	{ }

	explicit Select(Firebird::MemoryPool& pool)
		: slct_time(0), slct_count(0), slct_poll(pool), slct_ready(pool)
	{ }
#else
	Select()
		: slct_time(0), slct_count(0), slct_width(0)
	{
		memset(&slct_fdset, 0, sizeof slct_fdset);
	}

	explicit Select(Firebird::MemoryPool& /*pool*/)
		: slct_time(0), slct_count(0), slct_width(0)
	{
		memset(&slct_fdset, 0, sizeof slct_fdset);
	}
#endif

	enum HandleState {SEL_BAD, SEL_DISCONNECTED, SEL_NO_DATA, SEL_READY};

	// set first port to check for readyness
	void checkStart(RemPortPtr& port)
	{
		slct_main = port;
		slct_port = port;
#ifdef WIRE_COMPRESS_SUPPORT
		slct_zport = nullptr;
#endif
	}

	// get port to check for readyness
	// assume port_mutex is locked
	HandleState checkNext(RemPortPtr& port)
	{
#ifdef WIRE_COMPRESS_SUPPORT
		if (slct_zport)
		{
			if (slct_zport->port_z_data &&
				(slct_zport->port_state != rem_port::DISCONNECTED))
			{
				port = slct_zport;
				slct_zport = nullptr;	// Will be set again by select_multi() if needed
				return SEL_READY;
			}

			slct_zport = nullptr;
		}
#endif

		if (slct_port && slct_port->port_state == rem_port::DISCONNECTED)
		{
			// restart from main port
			slct_port = nullptr;
			if (slct_main && slct_main->port_state == rem_port::DISCONNECTED)
				slct_main = nullptr;

			slct_port = slct_main;
		}

		port = slct_port;
		if (!slct_port)
			return SEL_NO_DATA;

#ifdef WIRE_COMPRESS_SUPPORT
		if (slct_port->port_z_data)
			return SEL_READY;
#endif

		slct_port = slct_port->port_next;
		return ok(port);
	}

	void setZDataPort(RemPortPtr& port)
	{
#ifdef WIRE_COMPRESS_SUPPORT
		slct_zport = port;
#endif
	}

	HandleState ok(const rem_port* port)
	{
#ifdef WIRE_COMPRESS_SUPPORT
		if (port->port_z_data)
			return SEL_READY;
#endif
		SOCKET n = port->port_handle;
#if defined(WIN_NT)
		if (FD_ISSET(n, &slct_fdset))
		{
			unset(n);
			return SEL_READY;
		}
		return SEL_NO_DATA;
#elif defined(HAVE_POLL)
		pollfd* pf = nullptr;
		FB_SIZE_T pos;
		if (slct_ready.find(n, pos))
			pf = slct_ready[pos];

		if (pf)
		{
			HandleState ret = pf->events & SEL_CHECK_MASK ? SEL_READY : SEL_NO_DATA;
			pf->events = 0;		// unset
			return ret;
		}
		return n < 0 ? (port->port_flags & PORT_disconnect ? SEL_DISCONNECTED : SEL_BAD) : SEL_NO_DATA;
#else
		if (n < 0 || n >= FD_SETSIZE)
			return port->port_flags & PORT_disconnect ? SEL_DISCONNECTED : SEL_BAD;

		if (n < slct_width && FD_ISSET(n, &slct_fdset))
		{
			unset(n);
			return SEL_READY;
		}

		return SEL_NO_DATA;
#endif
	}

	void unset(SOCKET handle)
	{
#if defined(HAVE_POLL)
		pollfd* pf = getPollFd(handle);
		if (pf)
		{
			pf->events = 0;
		}
#else
		FD_CLR(handle, &slct_fdset);
		--slct_count;
#endif
	}

	void set(SOCKET handle)
	{
#ifdef HAVE_POLL
		FB_SIZE_T pos;
		if (slct_poll.find(handle, pos))
		{
			slct_poll[pos].events = SEL_INIT_EVENTS;
		}
		else
		{
			pollfd f;
			f.fd = handle;
			f.events = SEL_INIT_EVENTS;
			slct_poll.insert(pos, f);
		}
#else
		FD_SET(handle, &slct_fdset);
#ifdef WIN_NT
		++slct_width;
#else
		slct_width = MAX(slct_width, handle + 1);
#endif // WIN_NT
#endif // HAVE_POLL
	}

	void clear()
	{
		slct_count = 0;
#if defined(HAVE_POLL)
		slct_poll.clear();
#else
		slct_width = 0;
		FD_ZERO(&slct_fdset);
#endif
		slct_main = nullptr;
		slct_port = nullptr;
#ifdef WIRE_COMPRESS_SUPPORT
		slct_zport = nullptr;
#endif
	}

	void select(timeval* timeout)
	{
#ifdef HAVE_POLL
		slct_ready.clear();
		bool hasRequest = false;
		pollfd* const end = slct_poll.end();
		for (pollfd* pf = slct_poll.begin(); pf < end; ++pf)
		{
			pf->revents = pf->events;
			if (pf->events & SEL_CHECK_MASK)
				hasRequest = true;
		}

		if (!hasRequest)
		{
			errno = NOTASOCKET;
			slct_count = -1;
			return;
		}

		int milliseconds = timeout ? timeout->tv_sec * 1000 + timeout->tv_usec / 1000 : -1;
		slct_count = ::poll(slct_poll.begin(), slct_poll.getCount(), milliseconds);

		if (slct_count >= 0)	// in case of error return revents may contain something bad
		{
			for (pollfd* pf = slct_poll.begin(); pf < end; ++pf)
			{
				pf->events = pf->revents;
				if (pf->revents & SEL_CHECK_MASK)
					slct_ready.add(pf);
			}
		}
#else
#ifdef WIN_NT
		slct_count = ::select(FD_SETSIZE, &slct_fdset, NULL, NULL, timeout);
#else
		slct_count = ::select(slct_width, &slct_fdset, NULL, NULL, timeout);
#endif // WIN_NT
#endif // HAVE_POLL
	}

	int getCount()
	{
		return slct_count;
	}

	time_t	slct_time;

private:
	int		slct_count;
#ifdef HAVE_POLL
	class PollToFD
	{
	public:
		static int generate(const pollfd* p) { return p->fd; };
		static int generate(const pollfd& p) { return p.fd; };
	};

	SortedArray<pollfd, InlineStorage<pollfd, 8>, int, PollToFD>  slct_poll;
	SortedArray<pollfd*, InlineStorage<pollfd*, 8>, int, PollToFD>  slct_ready;
#else
	int		slct_width;
	fd_set	slct_fdset;
#endif
	RemPortPtr slct_main;	// first port to check for readyness
	RemPortPtr slct_port;	// next port to check for readyness
#ifdef WIRE_COMPRESS_SUPPORT
	RemPortPtr slct_zport;	// port with some compressed data remaining in the buffer
#endif
};

static bool		accept_connection(rem_port*, const P_CNCT*);
#ifdef HAVE_SETITIMER
static void		alarm_handler(int);
#endif
static rem_port*		alloc_port(rem_port*, const USHORT = 0);
static rem_port*		aux_connect(rem_port*, PACKET*);
static void				abort_aux_connection(rem_port*);
static rem_port*		aux_request(rem_port*, PACKET*);

#if !defined(WIN_NT)
static THREAD_ENTRY_DECLARE waitThread(THREAD_ENTRY_PARAM);

static GlobalPtr<Mutex> waitThreadMutex;
static unsigned int procCount = 0;
#endif // WIN_NT

static void		disconnect(rem_port*);
static void		force_close(rem_port*);
static int		cleanup_ports(const int, const int, void*);

#ifdef NO_FORK
static int		fork();
#endif

typedef Array<SOCKET> SocketsArray;

#ifdef WIN_NT
static int		wsaExitHandler(const int, const int, void*);
static int		fork(SOCKET, USHORT);
static THREAD_ENTRY_DECLARE forkThread(THREAD_ENTRY_PARAM);

static GlobalPtr<Mutex> forkMutex;
static HANDLE forkEvent = INVALID_HANDLE_VALUE;
static bool forkThreadStarted = false;

static SocketsArray* forkSockets;
#endif

static void		get_peer_info(rem_port*);

static void		inet_gen_error(bool, rem_port*, const Arg::StatusVector& v);
static void		inet_error(bool, rem_port*, const TEXT*, ISC_STATUS, int);
static bool		inet_read(RemoteXdr*);
static rem_port*		inet_try_connect(	PACKET*,
									Rdb*,
									const PathName&,
									const TEXT*,
									ClumpletReader&,
									RefPtr<const Config>*,
									const PathName*,
									int);
static bool		inet_write(RemoteXdr*);
static rem_port* listener_socket(rem_port* port, USHORT flag, const addrinfo* pai);

#ifdef DEBUG
static void packet_print(const TEXT*, const UCHAR*, int, ULONG);
#endif

static bool		packet_receive(rem_port*, UCHAR*, SSHORT, SSHORT*);
static bool		packet_receive2(rem_port*, UCHAR*, SSHORT, SSHORT*);
static bool		packet_send(rem_port*, const SCHAR*, SSHORT);
static rem_port*		receive(rem_port*, PACKET *);
static rem_port*		select_accept(rem_port*);

static void		select_port(rem_port*, Select*, RemPortPtr&);
static bool		select_multi(rem_port*, UCHAR* buffer, SSHORT bufsize, SSHORT* length, RemPortPtr&);
static bool		select_wait(rem_port*, Select*);
static int		send_full(rem_port*, PACKET *);
static int		send_partial(rem_port*, PACKET *);

static RemoteXdr*		xdrinet_create(rem_port*, UCHAR *, USHORT, enum xdr_op);
static bool		setNoNagleOption(rem_port*);
static bool		setKeepAlive(SOCKET);
static FPTR_INT	tryStopMainThread = 0;



struct InetXdr : public RemoteXdr
{
	virtual bool_t x_getbytes(SCHAR *, unsigned);		// get some bytes from "
	virtual bool_t x_putbytes(const SCHAR*, unsigned);	// put some bytes to "
};


#define MAXCLIENTS	NOFILE - 10

// Select uses bit masks of file descriptors in longs.

#ifndef NBBY
#define	NBBY	8
#endif

#ifndef NFDBITS
#if !defined(WIN_NT)
#define NFDBITS		(sizeof(SLONG) * NBBY)

#define	FD_SET(n, p)	((p)->fds_bits[(n) / NFDBITS] |= (1 << ((n) % NFDBITS)))
#define	FD_CLR(n, p)	((p)->fds_bits[(n) / NFDBITS] &= ~(1 << ((n) % NFDBITS)))
#define	FD_ISSET(n, p)	((p)->fds_bits[(n) / NFDBITS] & (1 << ((n) % NFDBITS)))
#define FD_ZERO(p)	memset(p, 0, sizeof(*(p)))
#endif
#endif

#ifdef	WIN_NT
#define	INTERRUPT_ERROR(x)	(SYSCALL_INTERRUPTED(x) || (x) == WSAEINTR)
#else
#define	INTERRUPT_ERROR(x)	(SYSCALL_INTERRUPTED(x))
#endif



ULONG INET_remote_buffer;
static GlobalPtr<Mutex> init_mutex;
static volatile bool INET_initialized = false;
static volatile bool INET_shutting_down = false;
static Firebird::GlobalPtr<Select> INET_select;
static rem_port* inet_async_receive = NULL;


static GlobalPtr<Mutex> port_mutex;
static GlobalPtr<PortsCleanup>	inet_ports;
static GlobalPtr<SocketsArray> ports_to_close;


rem_port* INET_analyze(ClntAuthBlock* cBlock,
					   const PathName& file_name,
					   const TEXT* node_name,
					   bool uv_flag,
					   ClumpletReader &dpb,
					   RefPtr<const Config>* config,
					   const PathName* ref_db_name,
					   Firebird::ICryptKeyCallback* cryptCb,
					   int af)
{
/**************************************
 *
 *	I N E T _ a n a l y z e
 *
 **************************************
 *
 * Functional description
 *	File_name is on node_name.
 *	Establish an external connection to node_name.
 *
 *	If a connection is established, return a port block, otherwise
 *	return NULL.
 *
 *	If the "uv_flag" is non-zero, user verification also takes place.
 *
 **************************************/
	// We need to establish a connection to a remote server.  Allocate the necessary
	// blocks and get ready to go.

	Rdb* rdb = FB_NEW Rdb;
	PACKET* packet = &rdb->rdb_packet;

	// Pick up some user identification information
	ClumpletWriter user_id(ClumpletReader::UnTagged, 64000);
	if (cBlock)
	{
		cBlock->extractDataFromPluginTo(user_id);
	}

	string buffer;
	int eff_gid;
	int eff_uid;

	ISC_get_user(&buffer, &eff_uid, &eff_gid);
#ifdef WIN_NT
	// XNET lowercases user names (as it's always case-insensitive in Windows),
	// so let's be consistent and use the same trick for INET as well
	buffer.lower();
#endif
	ISC_systemToUtf8(buffer);
	user_id.insertString(CNCT_user, buffer);

	ISC_get_host(buffer);
	buffer.lower();
	ISC_systemToUtf8(buffer);
	user_id.insertString(CNCT_host, buffer);

	if ((eff_uid == -1) || uv_flag) {
		user_id.insertTag(CNCT_user_verification);
	}
	else
	{
		// Communicate group id info to server, as user maybe running under group
		// id other than default specified in /etc/passwd.

		eff_gid = htonl(eff_gid);
		user_id.insertBytes(CNCT_group, reinterpret_cast<UCHAR*>(&eff_gid), sizeof(eff_gid));
	}

	// Should compression be tried?

	bool compression = config && (*config)->getWireCompression();

	// Establish connection to server
	// If we want user verification, we can't speak anything less than version 7

	P_CNCT*	cnct = &packet->p_cnct;

	cnct->p_cnct_user_id.cstr_length = (ULONG) user_id.getBufferLength();
	cnct->p_cnct_user_id.cstr_address = user_id.getBuffer();

	static const p_cnct::p_cnct_repeat protocols_to_try[] =
	{
		REMOTE_PROTOCOL(PROTOCOL_VERSION10, ptype_lazy_send, 1),
		REMOTE_PROTOCOL(PROTOCOL_VERSION11, ptype_lazy_send, 2),
		REMOTE_PROTOCOL(PROTOCOL_VERSION12, ptype_lazy_send, 3),
		REMOTE_PROTOCOL(PROTOCOL_VERSION13, ptype_lazy_send, 4),
		REMOTE_PROTOCOL(PROTOCOL_VERSION14, ptype_lazy_send, 5),
		REMOTE_PROTOCOL(PROTOCOL_VERSION15, ptype_lazy_send, 6),
		REMOTE_PROTOCOL(PROTOCOL_VERSION16, ptype_lazy_send, 7),
		REMOTE_PROTOCOL(PROTOCOL_VERSION17, ptype_lazy_send, 8),
		REMOTE_PROTOCOL(PROTOCOL_VERSION18, ptype_lazy_send, 9),
		REMOTE_PROTOCOL(PROTOCOL_VERSION19, ptype_lazy_send, 10)
	};
	fb_assert(FB_NELEM(protocols_to_try) <= FB_NELEM(cnct->p_cnct_versions));
	cnct->p_cnct_count = FB_NELEM(protocols_to_try);

	for (size_t i = 0; i < cnct->p_cnct_count; i++) {
		cnct->p_cnct_versions[i] = protocols_to_try[i];
		if (compression && cnct->p_cnct_versions[i].p_cnct_version >= PROTOCOL_VERSION13 &&
			rem_port::checkCompression())
		{
			cnct->p_cnct_versions[i].p_cnct_max_type |= pflag_compress;
		}
	}

	rem_port* port = inet_try_connect(packet, rdb, file_name, node_name, dpb, config, ref_db_name, af);
	P_ACPT* accept;

	for (;;)
	{
		accept = NULL;
		switch (packet->p_operation)
		{
		case op_accept_data:
		case op_cond_accept:
			accept = &packet->p_acpd;
			if (cBlock)
			{
				cBlock->storeDataForPlugin(packet->p_acpd.p_acpt_data.cstr_length,
										   packet->p_acpd.p_acpt_data.cstr_address);
				cBlock->authComplete = packet->p_acpd.p_acpt_authenticated;
				port->addServerKeys(&packet->p_acpd.p_acpt_keys);
				cBlock->resetClnt(&packet->p_acpd.p_acpt_keys);
			}
			break;

		case op_accept:
			if (cBlock)
			{
				cBlock->resetClnt();
			}
			accept = &packet->p_acpt;
			break;

		case op_crypt_key_callback:
			try
			{
				UCharBuffer buf;
				P_CRYPT_CALLBACK* cc = &packet->p_cc;

				if (cryptCb)
				{
					if (cc->p_cc_reply <= 0)
					{
						cc->p_cc_reply = 1;
					}
					UCHAR* reply = buf.getBuffer(cc->p_cc_reply);
					unsigned l = cryptCb->callback(cc->p_cc_data.cstr_length,
						cc->p_cc_data.cstr_address, cc->p_cc_reply, reply);

					REMOTE_free_packet(port, packet, true);
					cc->p_cc_data.cstr_length = l;
					cc->p_cc_data.cstr_address = reply;
				}
				else
				{
					REMOTE_free_packet(port, packet, true);
					cc->p_cc_data.cstr_length = 0;
				}

				packet->p_operation = op_crypt_key_callback;
				cc->p_cc_reply = 0;
				port->send(packet);
				port->receive(packet);
				continue;
			}
			catch (const Exception&)
			{
				disconnect(port);
				delete rdb;
				throw;
			}

		case op_response:
			try
			{
				LocalStatus warning;		// Ignore connect warnings for a while
				CheckStatusWrapper statusWrapper(&warning);
				REMOTE_check_response(&statusWrapper, rdb, packet, false);
			}
			catch (const Exception&)
			{
				disconnect(port);
				delete rdb;
				throw;
			}
			// fall through - response is not a required accept

		default:
			disconnect(port);
			delete rdb;
			Arg::Gds(isc_connect_reject).raise();
			break;
		}

		break;	// Always leave for() loop here
	}

	fb_assert(accept);
	fb_assert(port);
	port->port_protocol = accept->p_acpt_version;

	// once we've decided on a protocol, concatenate the version
	// string to reflect it...
	string temp;
	temp.printf("%s/P%d", port->port_version->str_data, port->port_protocol & FB_PROTOCOL_MASK);
	delete port->port_version;
	port->port_version = REMOTE_make_string(temp.c_str());

	if (accept->p_acpt_architecture == ARCHITECTURE) {
		port->port_flags |= PORT_symmetric;
	}

	bool compress = accept->p_acpt_type & pflag_compress;
	accept->p_acpt_type &= ptype_MASK;

	if (accept->p_acpt_type != ptype_out_of_band) {
		port->port_flags |= PORT_no_oob;
	}

	if (accept->p_acpt_type == ptype_lazy_send) {
		port->port_flags |= PORT_lazy;
	}

	if (compress)
	{
		port->initCompression();
		port->port_flags |= PORT_compressed;
	}

	return port;
}

rem_port* INET_connect(const TEXT* name,
					   PACKET* packet,
					   USHORT flag,
					   ClumpletReader* dpb,
					   RefPtr<const Config>* config,
					   int af)
{
/**************************************
 *
 *	I N E T _ c o n n e c t
 *
 **************************************
 *
 * Functional description
 *	Establish half of a communication link.  If a connect packet is given,
 *	the connection is on behalf of a remote interface.  Otherwise the connect
 *	is for a server process.
 *
 **************************************/
#ifdef DEBUG
	{
		if (INET_trace & TRACE_operations)
		{
			fprintf(stdout, "INET_connect\n");
			fflush(stdout);
		}
		INET_start_time = inet_debug_timer();
		// CVC: I don't see the point in replacing this with fb_utils::readenv().
		const char* p = getenv("INET_force_error");
		if (p != NULL) {
			INET_force_error = atoi(p);
		}
	}
#endif

	rem_port* port = alloc_port(NULL);
	if (config)
	{
		port->port_config = *config;
	}
	REMOTE_get_timeout_params(port, dpb);

	string host;
	string protocol;

	if ((!name || !name[0]) && !packet)
	{
		name = port->getPortConfig()->getRemoteBindAddress();
	}

	if (name)
	{
		host = name;
		const FB_SIZE_T pos = host.find("/");

		if (pos != string::npos)
		{
			protocol = host.substr(pos + 1);
			host = host.substr(0, pos);
		}

		if (host.hasData() && host[0] == '[' && host[host.length() - 1] == ']')
		{
			// host name or address is in brackets, remove them
			host.erase(host.length() - 1);
			host.erase(0, 1);
		}
	}

	if (host.hasData())
	{
		delete port->port_connection;
		port->port_connection = nullptr;
		port->port_connection = REMOTE_make_string(host.c_str());
	}
	else if (packet)
	{
		host = port->port_host->str_data;
	}

	if (protocol.isEmpty())
	{
		const unsigned short port2 = port->getPortConfig()->getRemoteServicePort();
		if (port2) {
			protocol.printf("%hu", port2);
		}
		else {
			protocol = port->getPortConfig()->getRemoteServiceName();
		}
	}

	// Prepare hints
	const bool ipv6 = os_utils::isIPv6supported();

	struct addrinfo gai_hints;
	memset(&gai_hints, 0, sizeof(gai_hints));
	if (packet)
		gai_hints.ai_family = af;
	else
		gai_hints.ai_family = ((host.hasData() || !ipv6) ? AF_UNSPEC : AF_INET6);
	gai_hints.ai_socktype = SOCK_STREAM;

#if !defined(WIN_NT) && !defined(__clang__)
	gai_hints.ai_protocol = SOL_TCP;
#else
	gai_hints.ai_protocol = IPPROTO_TCP;
#endif

	gai_hints.ai_flags =
#ifndef ANDROID
		((af == AF_UNSPEC) ? AI_V4MAPPED : 0) |
#endif
			AI_ADDRCONFIG | (packet ? 0 : AI_PASSIVE);

	struct AutoAddrInfo
	{
		~AutoAddrInfo()
		{
			if (ptr)
				freeaddrinfo(ptr);
		}

		addrinfo* ptr = nullptr;
	} gai_result;

	const char* host_str = (host.hasData() ? host.c_str() : NULL);
	bool retry_gai;
	int n;

	do
	{
		retry_gai = false;
		n = getaddrinfo(host_str, protocol.c_str(), &gai_hints, &gai_result.ptr);

		if ((n == EAI_FAMILY || (!host_str && n == EAI_NONAME)) &&
			(gai_hints.ai_family == AF_INET6) && (af != AF_INET6))
		{
			// May be on a system without IPv6 support, try IPv4
			gai_hints.ai_family = AF_UNSPEC;
			retry_gai = true;
		}

		if ((n == EAI_SERVICE) && (protocol == FB_SERVICE_NAME))
		{
			// Try hard-wired translation of "gds_db" to "3050"
			protocol.printf("%hu", FB_SERVICE_PORT);
			retry_gai = (protocol != FB_SERVICE_NAME);
		}
	} while (retry_gai);

	if (n)
	{
		gds__log("INET/INET_connect: getaddrinfo(%s,%s) failed: %s",
				host.c_str(), protocol.c_str(), gai_strerror(n));
		inet_gen_error(true, port, Arg::Gds(isc_net_lookup_err) << Arg::Gds(isc_host_unknown));
	}

	for (const addrinfo* pai = gai_result.ptr; pai; pai = pai->ai_next)
	{
		// Allocate a port block and initialize a socket for communications
		port->port_handle = os_utils::socket(pai->ai_family, pai->ai_socktype, pai->ai_protocol);

		if (port->port_handle == INVALID_SOCKET)
		{
			gds__log("socket: error creating socket (family %d, socktype %d, protocol %d",
					pai->ai_family, pai->ai_socktype, pai->ai_protocol);
			continue;
		}

		if (!packet) // server
			return listener_socket(port, flag, pai);

		// client
		if (!setKeepAlive(port->port_handle))
			gds__log("setsockopt: error setting SO_KEEPALIVE");

		if (!setNoNagleOption(port))
			gds__log("setsockopt: error setting TCP_NODELAY");
		else
		{
			n = connect(port->port_handle, pai->ai_addr, pai->ai_addrlen);
			if (n != -1)
			{
				port->port_peer_name = host;
				get_peer_info(port);
				if (send_full(port, packet))
					return port;
			}
		}

		SOCLOSE(port->port_handle);
	}

	// all attempts failed
	if (packet)
		inet_error(true, port, "connect", isc_net_connect_err, 0);
	else
		inet_error(true, port, "listen", isc_net_connect_listen_err, 0);

	return port;
}

static rem_port* listener_socket(rem_port* port, USHORT flag, const addrinfo* pai)
{
/**************************************
 *
 *	l i s t e n e r _ s o c k e t
 *
 **************************************
 *
 * Functional description
 *	Final part of server (listening) socket setup. Sets socket options,
 *	binds the socket and calls listen().
 *  For multi-client server (SuperServer or SuperClassic) return listener
 *  port.
 *  For classic server - accept incoming connections and fork worker
 *  processes, return NULL at exit;
 *  On error throw exception.
 *
 **************************************/

	int ipv6_v6only = port->getPortConfig()->getIPv6V6Only() ? 1 : 0;

	int n = setsockopt(port->port_handle, IPPROTO_IPV6, IPV6_V6ONLY,
				   (SCHAR*) &ipv6_v6only, sizeof(ipv6_v6only));

	if (n == -1)
		gds__log("setsockopt: error setting IPV6_V6ONLY to %d", ipv6_v6only);

	if (flag & SRVR_multi_client)
	{
		struct linger lingerInfo;
		lingerInfo.l_onoff = 0;
		lingerInfo.l_linger = 0;

#ifndef WIN_NT
		// dimitr:	on Windows, lack of SO_REUSEADDR works the same way as it was specified on POSIX,
		//			i.e. it allows binding to a port in a TIME_WAIT/FIN_WAIT state. If this option
		//			is turned on explicitly, then a port can be re-bound regardless of its state,
		//			e.g. while it's listening. This is surely not what we want.

		int optval = TRUE;
		n = setsockopt(port->port_handle, SOL_SOCKET, SO_REUSEADDR,
					   (SCHAR*) &optval, sizeof(optval));
		if (n == -1)
		{
			inet_error(true, port, "setsockopt REUSE", isc_net_connect_listen_err, INET_ERRNO);
		}
#endif

		// Get any values for SO_LINGER so that they can be reset during
		// disconnect.  SO_LINGER should be set by default on the socket

		socklen_t optlen = sizeof(port->port_linger);
		n = getsockopt(port->port_handle, SOL_SOCKET, SO_LINGER,
					   (SCHAR *) & port->port_linger, &optlen);

		if (n != 0)				// getsockopt failed
			port->port_linger.l_onoff = 0;

		n = setsockopt(port->port_handle, SOL_SOCKET, SO_LINGER,
					   (SCHAR *) & lingerInfo, sizeof(lingerInfo));
		if (n == -1)
		{
			inet_error(true, port, "setsockopt LINGER", isc_net_connect_listen_err, INET_ERRNO);
		}
	}
	else
	{
		if (! setKeepAlive(port->port_handle))
		{
			inet_error(true, port, "setsockopt SO_KEEPALIVE", isc_net_connect_listen_err, INET_ERRNO);
		}
	}

	// RS: In linux sockets inherit this option from listener. Previously CLASSIC had no its own listen socket
	// Now it's necessary to respect the option via listen socket.
	if (! setNoNagleOption(port))
	{
		inet_error(true, port, "setsockopt TCP_NODELAY", isc_net_connect_listen_err, INET_ERRNO);
	}

	// On Linux platform, when the server dies the system holds a port
	// for some time (we don't set SO_REUSEADDR for standalone server).
	int retry = -1;
	do
	{
		if (++retry)
			sleep(10);
		n = bind(port->port_handle, pai->ai_addr, pai->ai_addrlen);
	} while (n == -1 && INET_ERRNO == INET_ADDR_IN_USE && retry < INET_RETRY_CALL);

	if (n == -1)
	{
		inet_error(true, port, "bind", isc_net_connect_listen_err, INET_ERRNO);
	}

	n = listen(port->port_handle, SOMAXCONN);

	if (n == -1)
	{
		inet_error(false, port, "listen", isc_net_connect_listen_err, INET_ERRNO);
	}

	inet_ports->registerPort(port);

	if (flag & SRVR_multi_client)
	{
		// Prevent the generation of dummy keepalive packets on the connect port.

		port->port_dummy_packet_interval = 0;
		port->port_dummy_timeout = 0;
		port->port_server_flags |= (SRVR_server | SRVR_multi_client);
		return port;
	}

	while (true)
	{
		SOCKET s = os_utils::accept(port->port_handle, NULL, NULL);
		const int inetErrNo = INET_ERRNO;
		if (s == INVALID_SOCKET)
		{
			if (INET_shutting_down)
				return NULL;
			inet_error(true, port, "accept", isc_net_connect_err, inetErrNo);
		}
#ifdef WIN_NT
		if (flag & SRVR_debug)
#else
		if ((flag & SRVR_debug) || !fork())
#endif
		{
			SOCLOSE(port->port_handle);
			port->port_handle = s;
			port->port_server_flags |= SRVR_server;
			port->port_flags |= PORT_server;
			return port;
		}

#ifdef WIN_NT
		MutexLockGuard forkGuard(forkMutex, FB_FUNCTION);
		if (!forkThreadStarted)
		{
			forkThreadStarted = true;
			forkEvent = CreateEvent(NULL, FALSE, FALSE, NULL);
			forkSockets = FB_NEW SocketsArray(*getDefaultMemoryPool());

			Thread::start(forkThread, (void*) flag, THREAD_medium);
		}
		forkSockets->add(s);
		SetEvent(forkEvent);
#else
		MutexLockGuard guard(waitThreadMutex, FB_FUNCTION);
		if (! procCount++) {
			Thread::start(waitThread, 0, THREAD_medium);
		}

		SOCLOSE(s);
#endif
	}

#ifdef WIN_NT
	MutexLockGuard forkGuard(forkMutex, FB_FUNCTION);
	if (forkThreadStarted)
	{
		SetEvent(forkEvent);
		CloseHandle(forkEvent);

		delete forkSockets;
		forkSockets = NULL;
	}
#endif
	return NULL;
}


rem_port* INET_reconnect(SOCKET handle)
{
/**************************************
 *
 *	I N E T _ r e c o n n e c t
 *
 **************************************
 *
 * Functional description
 *	A communications link has been established by another
 *	process.  We have inheritted the handle.  Set up
 *	a port block.
 *
 **************************************/
	rem_port* const port = alloc_port(NULL);

	port->port_handle = handle;
	port->port_flags |= PORT_server;
	port->port_server_flags |= SRVR_server;

	if (! setKeepAlive(port->port_handle)) {
		gds__log("inet server err: setting KEEPALIVE socket option \n");
	}

	if (! setNoNagleOption(port)) {
		gds__log("inet server err: setting NODELAY socket option \n");
	}

	return port;
}

rem_port* INET_server(SOCKET sock)
{
/**************************************
 *
 *	I N E T _ s e r v e r
 *
 **************************************
 *
 * Functional description
 *	We have been spawned by a master server with a connection
 *	established.  Set up port block with the appropriate socket.
 *
 **************************************/
	rem_port* const port = alloc_port(NULL);
	port->port_flags |= PORT_server;
	port->port_server_flags |= SRVR_server;
	port->port_handle = sock;

	if (! setKeepAlive(port->port_handle)) {
		gds__log("inet server err: setting KEEPALIVE socket option \n");
	}

	if (! setNoNagleOption(port)) {
		gds__log("inet server err: setting NODELAY socket option \n");
	}

	return port;
}

static bool accept_connection(rem_port* port, const P_CNCT* cnct)
{
/**************************************
 *
 *	a c c e p t _ c o n n e c t i o n
 *
 **************************************
 *
 * Functional description
 *	Accept an incoming request for connection.  This is purely a lower
 *	level handshaking function, and does not constitute the server
 *	response for protocol selection.
 *
 **************************************/
	// Default account to "guest" (in theory all packets contain a name)

	string user_name("guest"), host_name;

	// Pick up account and host name, if given

	ClumpletReader id(ClumpletReader::UnTagged,
					  cnct->p_cnct_user_id.cstr_address,
					  cnct->p_cnct_user_id.cstr_length);

	for (id.rewind(); !id.isEof(); id.moveNext())
	{
		switch (id.getClumpTag())
		{
		case CNCT_user:
			id.getString(user_name);
			break;

		case CNCT_host:
			id.getString(host_name);
			break;

		default:
			break;
		}
	}

#ifndef WIN_NT
	{ // scope
		// If the environment variable ISC_INET_SERVER_HOME is set,
		// change the home directory to the specified directory.
		// Note that this will overrule the normal setting of
		// the current directory to the effective user's home directory.
		// This feature was added primarily for testing via remote
		// loopback - but does seem to be of good general use, so
		// is activated for the release version.
		// 1995-February-27 David Schnepper

		PathName home;
		if (fb_utils::readenv("ISC_INET_SERVER_HOME", home))
		{
			if (chdir(home.c_str()))
			{
				gds__log("inet_server: unable to cd to %s errno %d\n", home.c_str(), INET_ERRNO);
				// We continue after the error
			}
		}
	} // end scope
#endif // !WIN_NT

	// store user identity
	port->port_login = port->port_user_name = user_name;
	port->port_peer_name = host_name;

	get_peer_info(port);

	return true;
}


static rem_port* alloc_port(rem_port* const parent, const USHORT flags)
{
/**************************************
 *
 *	a l l o c _ p o r t
 *
 **************************************
 *
 * Functional description
 *	Allocate a port block, link it in to parent (if there is a parent),
 *	and initialize input and output XDR streams.
 *
 **************************************/

	if (!INET_initialized)
	{
		MutexLockGuard guard(init_mutex, FB_FUNCTION);
		if (!INET_initialized)
		{
#ifdef WIN_NT
			static WSADATA wsadata;
			const WORD version = MAKEWORD(2, 0);
			const int wsaError = WSAStartup(version, &wsadata);
			if (wsaError)
			{
				inet_error(false, parent, "WSAStartup", isc_net_init_error, wsaError);
			}
			fb_shutdown_callback(0, wsaExitHandler, fb_shut_finish, 0);
#endif
			INET_remote_buffer = Config::getTcpRemoteBufferSize();
			if (INET_remote_buffer < MAX_DATA_LW || INET_remote_buffer > MAX_DATA_HW)
			{
				INET_remote_buffer = DEF_MAX_DATA;
			}
#ifdef DEBUG
			gds__log(" Info: Remote Buffer Size set to %ld", INET_remote_buffer);
#endif

			fb_shutdown_callback(0, cleanup_ports, fb_shut_postproviders, 0);

			INET_initialized = true;

			// This should go AFTER 'INET_initialized = true' to avoid recursion
			// That order of statements at the first glance appears to be possible cause of races:
			// Someone may pass the if (!INET_initialized) with INET_initialized == true,
			// but inet_async_receive still being NULL. Luckily it's not so awful actually.
			// Async receive is used only by network server and there is no way for it
			// to allocate secondary ports until completion of master port init.
			inet_async_receive = alloc_port(0);
			inet_async_receive->port_flags |= PORT_server;
		}
	}

	rem_port* const port = FB_NEW rem_port(rem_port::INET, INET_remote_buffer * 2);
	REMOTE_get_timeout_params(port, 0);

	TEXT buffer[BUFFER_SMALL];
	gethostname(buffer, sizeof(buffer));

	port->port_host = REMOTE_make_string(buffer);
	port->port_connection = REMOTE_make_string(buffer);
	SNPRINTF(buffer, FB_NELEM(buffer), "tcp (%s)", port->port_host->str_data);
	port->port_version = REMOTE_make_string(buffer);

	port->port_accept = accept_connection;
	port->port_disconnect = disconnect;
	port->port_force_close = force_close;
	port->port_receive_packet = receive;
	port->port_select_multi = select_multi;
	port->port_send_packet = send_full;
	port->port_send_partial = send_partial;
	port->port_connect = aux_connect;
	port->port_abort_aux_connection = abort_aux_connection;
	port->port_request = aux_request;
	port->port_buff_size = (USHORT) INET_remote_buffer;
	port->port_async_receive = inet_async_receive;
	port->port_flags |= flags;

	port->port_send = xdrinet_create(port,
		&port->port_buffer[REM_SEND_OFFSET(INET_remote_buffer)],
		(USHORT) INET_remote_buffer, XDR_ENCODE);

	port->port_receive = xdrinet_create(port,
		&port->port_buffer[REM_RECV_OFFSET(INET_remote_buffer)], 0, XDR_DECODE);

	if (parent && !(parent->port_server_flags & SRVR_thread_per_port))
	{
		MutexLockGuard guard(port_mutex, FB_FUNCTION);
		port->linkParent(parent);
	}

	return port;
}

static void abort_aux_connection(rem_port* port)
{
	if (port->port_flags & PORT_connecting)
	{
		shutdown(port->port_channel, 2);
		SOCLOSE(port->port_channel);
	}
}

static rem_port* aux_connect(rem_port* port, PACKET* packet)
{
/**************************************
 *
 *	a u x _ c o n n e c t
 *
 **************************************
 *
 * Functional description
 *	Try to establish an alternative connection.  Somebody has already
 *	done a successfull connect request ("packet" contains the response).
 *
 **************************************/

	// If this is a server, we're got an auxiliary connection.  Accept it

	if (port->port_server_flags)
	{
		struct timeval timeout;
		timeout.tv_sec = port->port_connect_timeout;
		timeout.tv_usec = 0;

		Select slct;
		slct.set(port->port_channel);

		int inetErrNo = 0;

		while (true)
		{
			slct.select(&timeout);
			const int count = slct.getCount();
			inetErrNo = INET_ERRNO;

			if (count != -1 || !INTERRUPT_ERROR(inetErrNo))
			{
				if (count == 1)
					break;

				const ISC_STATUS error_code =
					(count == 0) ? isc_net_event_connect_timeout : isc_net_event_connect_err;
				int savedError = inetErrNo;
				SOCLOSE(port->port_channel);
				inet_error(false, port, "select", error_code, savedError);
			}
		}

		if (port->port_channel == INVALID_SOCKET)
			return NULL;

		const SOCKET n = os_utils::accept(port->port_channel, NULL, NULL);
		inetErrNo = INET_ERRNO;

		if (n == INVALID_SOCKET)
		{
			int savedError = inetErrNo;
			SOCLOSE(port->port_channel);
			inet_error(false, port, "accept", isc_net_event_connect_err, savedError);
		}

		SOCLOSE(port->port_channel);
		port->port_handle = n;
		port->port_flags |= PORT_async;

		get_peer_info(port);

		return port;
	}

	rem_port* const new_port = alloc_port(port->port_parent,
		(port->port_flags & PORT_no_oob) | PORT_async);
	port->port_async = new_port;
	new_port->port_dummy_packet_interval = port->port_dummy_packet_interval;
	new_port->port_dummy_timeout = new_port->port_dummy_packet_interval;
	P_RESP* response = &packet->p_resp;

	// NJK - Determine address and port to use.
	//
	// The address returned by the server may be incorrect if it is behind a NAT box
	// so we must use the address that was used to connect the main socket, not the
	// address reported by the server.
	//
	// The port number reported by the server is used. For NAT support the port number
	// should be configured to be a fixed port number in the server configuration.

	SockAddr address;
	int status = address.getpeername(port->port_handle);
	if (status != 0)
	{
		int savedError = INET_ERRNO;
		port->auxAcceptError(packet);
		inet_error(false, port, "socket", isc_net_event_connect_err, savedError);
	}

	SockAddr resp_address(response->p_resp_data.cstr_address, response->p_resp_data.cstr_length);
	address.setPort(resp_address.port());

	// Set up new socket

	SOCKET n = os_utils::socket(address.family(), SOCK_STREAM, 0);
	if (n == INVALID_SOCKET)
	{
		int savedError = INET_ERRNO;
		port->auxAcceptError(packet);
		inet_error(false, port, "socket", isc_net_event_connect_err, savedError);
	}

	setKeepAlive(n);

	status = address.connect(n);
	if (status < 0)
	{
		int savedError = INET_ERRNO;
		SOCLOSE(n);
		port->auxAcceptError(packet);
		inet_error(false, port, "connect", isc_net_event_connect_err, savedError);
	}

	new_port->port_handle = n;

	new_port->port_peer_name = port->port_peer_name;
	get_peer_info(new_port);

	return new_port;
}

static rem_port* aux_request( rem_port* port, PACKET* packet)
{
/**************************************
 *
 *	a u x _ r e q u e s t
 *
 **************************************
 *
 * Functional description
 *	A remote interface has requested the server prepare an auxiliary
 *	connection; the server calls aux_request to set up the connection.
 *
 **************************************/

	// listen on (local) address of the original socket
	SockAddr our_address;
	if (our_address.getsockname(port->port_handle) < 0)
	{
		gds__log("INET/aux_request: failed to get local address of the original socket");
		inet_error(false, port, "getsockname", isc_net_event_listen_err, INET_ERRNO);
	}
	unsigned short aux_port = port->getPortConfig()->getRemoteAuxPort();
	our_address.setPort(aux_port); // may be 0

	SOCKET n = os_utils::socket(our_address.family(), SOCK_STREAM, 0);
	if (n == INVALID_SOCKET)
	{
		inet_error(false, port, "socket", isc_net_event_listen_err, INET_ERRNO);
	}

	int optval;
#ifndef WIN_NT
	// dimitr:	on Windows, lack of SO_REUSEADDR works the same way as it was specified on POSIX,
	//			i.e. it allows binding to a port in a TIME_WAIT/FIN_WAIT state. If this option
	//			is turned on explicitly, then a port can be re-bound regardless of its state,
	//			e.g. while it's listening. This is surely not what we want.

	optval = TRUE;
	if (setsockopt(n, SOL_SOCKET, SO_REUSEADDR, (SCHAR*) &optval, sizeof(optval)) < 0)
	{
		inet_error(false, port, "setsockopt REUSE", isc_net_event_listen_err, INET_ERRNO);
	}
#endif

	optval = port->getPortConfig()->getIPv6V6Only() ? 1 : 0;
	// ignore failure, we already have it logged from the main listening port
	setsockopt(n, IPPROTO_IPV6, IPV6_V6ONLY, (SCHAR*) &optval, sizeof(optval));

	if (bind(n, our_address.ptr(), our_address.length()) < 0)
	{
		inet_error(false, port, "bind", isc_net_event_listen_err, INET_ERRNO);
	}

	if (our_address.getsockname(n) < 0)
	{
		inet_error(false, port, "getsockname", isc_net_event_listen_err, INET_ERRNO);
	}

	if (listen(n, 1) < 0)
	{
		inet_error(false, port, "listen", isc_net_event_listen_err, INET_ERRNO);
	}

	rem_port* const new_port = alloc_port(port->port_parent,
		(port->port_flags & PORT_no_oob) | PORT_async | PORT_connecting);
	port->port_async = new_port;
	new_port->port_dummy_packet_interval = port->port_dummy_packet_interval;
	new_port->port_dummy_timeout = new_port->port_dummy_packet_interval;

	new_port->port_server_flags = port->port_server_flags;
	new_port->port_channel = (int) n;

	P_RESP* response = &packet->p_resp;

	SockAddr port_address;

	if (port_address.getsockname(port->port_handle) < 0)
		inet_error(false, port, "getsockname", isc_net_event_listen_err, INET_ERRNO);

	port_address.setPort(our_address.port());

	// CORE-5902: MacOS has sockaddr struct layout different than one found in POSIX/Windows.
	// This prevent usage of events when client or server is MacOS but the other end is not.
	// Here we try to make this case work. However it's not bullet-proof for others platforms and architectures.
	// A proper solution would be to just send the port number in a protocol friendly way.

	bool macOsClient =
		port->port_client_arch == arch_darwin_ppc ||
		port->port_client_arch == arch_darwin_x64 ||
		port->port_client_arch == arch_darwin_ppc64;

	bool macOsServer =
		ARCHITECTURE == arch_darwin_ppc ||
		ARCHITECTURE == arch_darwin_x64 ||
		ARCHITECTURE == arch_darwin_ppc64;

	if (macOsServer && !macOsClient)
		port_address.convertFromMacOsToPosixWindows();
	else if (!macOsServer && macOsClient)
		port_address.convertFromPosixWindowsToMacOs();

	response->p_resp_data.cstr_length = (ULONG) port_address.length();
	memcpy(response->p_resp_data.cstr_address, port_address.ptr(), port_address.length());

	new_port->port_peer_name = port->port_peer_name;
	return new_port;
}


#if !(defined WIN_NT)
static THREAD_ENTRY_DECLARE waitThread(THREAD_ENTRY_PARAM)
{
/**************************************
 *
 *	w a i t T h r e a d
 *
 **************************************
 *
 * Functional description
 *	Waits for processes started by standalone classic server (avoid zombies)
 *
 **************************************/
	while (procCount > 0)
	{
		int rc = wait(0);

		MutexLockGuard guard(waitThreadMutex, FB_FUNCTION);
		if (rc > 0) {
			--procCount;
		}
	}

	return 0;
}
#endif // !defined(WIN_NT)

static void disconnect(rem_port* port)
{
/**************************************
 *
 *	d i s c o n n e c t
 *
 **************************************
 *
 * Functional description
 *	Break a remote connection.
 *
 **************************************/

	// SO_LINGER was turned off on the initial bind when the server was started.
	// This will force a reset to be sent to the client when the socket is closed.
	// We only want this behavior in the case of the server terminating
	// abnormally and not on an orderly shut down.  Because of this, turn the
	// SO_LINGER option back on for the socket.  The result of setsockopt isn't
	// too important at this stage since we are closing the socket anyway.  This
	// is an attempt to return the socket to a state where a graceful shutdown can
	// occur.

	if (port->port_linger.l_onoff)
	{
		setsockopt(port->port_handle, SOL_SOCKET, SO_LINGER,
				   (SCHAR*) &port->port_linger, sizeof(port->port_linger));
	}

	if (port->port_handle != INVALID_SOCKET)
	{
		shutdown(port->port_handle, 2);
	}

	MutexLockGuard guard(port_mutex, FB_FUNCTION);
	if (port->port_state == rem_port::DISCONNECTED)
		return;

	port->port_state = rem_port::DISCONNECTED;
	port->port_flags &= ~PORT_connecting;

	if (port->port_async)
	{
		disconnect(port->port_async);
		port->port_async = NULL;
	}
	port->port_context = NULL;

	// hvlad: delay closing of the server sockets to prevent its reuse
	// by another (newly accepted) port until next select() call. See
	// also select_wait() function.
	const bool delayClose = (port->port_server_flags && port->port_parent);

	// If this is a sub-port, unlink it from its parent
	port->unlinkParent();

	inet_ports->unRegisterPort(port);

	if (delayClose)
	{
		if (port->port_handle != INVALID_SOCKET)
			ports_to_close->push(port->port_handle);

		if (port->port_channel != INVALID_SOCKET)
			ports_to_close->push(port->port_channel);
	}
	else
	{
		SOCLOSE(port->port_handle);
		SOCLOSE(port->port_channel);
	}

	if (port->port_thread_guard && port->port_events_thread && !port->port_events_threadId.isCurrent())
		port->port_thread_guard->setWait(port->port_events_thread);
	else
		port->releasePort();

#ifdef DEBUG
	if (INET_trace & TRACE_summary)
	{
		fprintf(stdout, "INET_count_send = %u packets\n", INET_count_send);
		fprintf(stdout, "INET_bytes_send = %u bytes\n", INET_bytes_send);
		fprintf(stdout, "INET_count_recv = %u packets\n", INET_count_recv);
		fprintf(stdout, "INET_bytes_recv = %u bytes\n", INET_bytes_recv);
		fflush(stdout);
	}
#endif

	return;
}


static void force_close(rem_port* port)
{
/**************************************
 *
 *	f o r c e _ c l o s e
 *
 **************************************
 *
 * Functional description
 *	Forcebly close remote connection.
 *
 **************************************/

	if (port->port_async)
		abort_aux_connection(port->port_async);

	if (port->port_state != rem_port::PENDING)
		return;

	RefMutexGuard guard(*port->port_write_sync, FB_FUNCTION);

	port->port_state = rem_port::BROKEN;
	if (port->port_handle != INVALID_SOCKET)
	{
		shutdown(port->port_handle, 2);
		SOCLOSE(port->port_handle);
	}
}


static int cleanup_ports(const int, const int, void* /*arg*/)
{
/**************************************
 *
 *	c l e a n u p _ p o r t s
 *
 **************************************
 *
 * Functional description
 *	Shutdown all active connections
 *	to allow correct shutdown.
 *
 **************************************/
	INET_shutting_down = true;

	inet_ports->closePorts();

	while (ports_to_close->hasData())
	{
		SOCKET s = ports_to_close->pop();
		SOCLOSE(s);
	}

	return 0;
}


#ifdef NO_FORK
static int fork()
{
/**************************************
 *
 *	f o r k		( N O _ F O R K )
 *
 **************************************
 *
 * Functional description
 *	Hmmm.
 *
 **************************************/

	return 1;
}
#endif

#ifdef WIN_NT
static int wsaExitHandler(const int, const int, void*)
{
/**************************************
 *
 *	w s a E x i t H a n d l e r
 *
 **************************************
 *
 * Functional description
 *	Cleanup WSA.
 *
 **************************************/
	SleepEx(0, FALSE);	// let select in other thread(s) shutdown gracefully
	WSACleanup();
	return 0;
}


static int fork(SOCKET old_handle, USHORT flag)
{
/**************************************
 *
 *	f o r k		( W I N _ N T )
 *
 **************************************
 *
 * Functional description
 *	Create a child process.
 *
 **************************************/
	TEXT name[MAXPATHLEN];
	GetModuleFileName(NULL, name, sizeof(name));

	HANDLE new_handle;
	if (!DuplicateHandle(GetCurrentProcess(), (HANDLE) old_handle,
						 GetCurrentProcess(), &new_handle,
						 0, TRUE, DUPLICATE_SAME_ACCESS))
	{
		gds__log("INET/inet_error: fork/DuplicateHandle errno = %d", GetLastError());
		return 0;
	}

	string cmdLine;
	cmdLine.printf("%s -i -h %" HANDLEFORMAT"@%" ULONGFORMAT, name, new_handle, GetCurrentProcessId());

	STARTUPINFO start_crud;
	start_crud.cb = sizeof(STARTUPINFO);
	start_crud.lpReserved = NULL;
	start_crud.lpReserved2 = NULL;
	start_crud.cbReserved2 = 0;
	start_crud.lpDesktop = NULL;
	start_crud.lpTitle = NULL;
	start_crud.dwFlags = STARTF_FORCEOFFFEEDBACK;

	PROCESS_INFORMATION pi;
	if (CreateProcess(NULL, cmdLine.begin(), NULL, NULL, FALSE,
					  (flag & SRVR_high_priority ?
						 HIGH_PRIORITY_CLASS | DETACHED_PROCESS :
						 NORMAL_PRIORITY_CLASS | DETACHED_PROCESS),
					  NULL, NULL, &start_crud, &pi))
	{
		CloseHandle(pi.hThread);
		CloseHandle(pi.hProcess);
		// hvlad: child process will close our handle of just accepted socket
		return 1;
	}

	gds__log("INET/inet_error: fork/CreateProcess errno = %d", GetLastError());
	CloseHandle(new_handle);
	return 0;
}


THREAD_ENTRY_DECLARE forkThread(THREAD_ENTRY_PARAM arg)
{
	const USHORT flag = (USHORT)(U_IPTR) arg;

	while (!INET_shutting_down)
	{
		if (WaitForSingleObject(forkEvent, INFINITE) != WAIT_OBJECT_0)
			break;

		while (!INET_shutting_down)
		{
			SOCKET s = 0;
			{	// scope
				MutexLockGuard forkGuard(forkMutex, FB_FUNCTION);

				if (!forkSockets || forkSockets->getCount() == 0)
					break;

				s = (*forkSockets)[0];
				forkSockets->remove((FB_SIZE_T) 0);
			}
			fork(s, flag);
			SOCLOSE(s);
		}
	}

	return 0;
}


// Windows does not have an inet_aton function.
bool inet_aton(const char* name, in_addr* address)
{
	address->s_addr = inet_addr(name);
	return address->s_addr != INADDR_NONE;
}
#endif


static rem_port* receive( rem_port* main_port, PACKET * packet)
{
/**************************************
 *
 *	r e c e i v e
 *
 **************************************
 *
 * Functional description
 *	Receive a message from a port or clients of a port.  If the process
 *	is a server and a connection request comes in, generate a new port
 *	block for the client.
 *
 **************************************/

	// loop as long as we are receiving dummy packets, just
	// throwing them away--note that if we are a server we won't
	// be receiving them, but it is better to check for them at
	// this level rather than try to catch them in all places where
	// this routine is called

	do {
		if (!xdr_protocol(main_port->port_receive, packet))
		{
			packet->p_operation = main_port->port_partial_data ? op_partial : op_exit;
			if (packet->p_operation == op_exit)
				main_port->port_state = rem_port::BROKEN;

			main_port->port_partial_data = false;
			break;
		}
#ifdef DEBUG
		{
			static ULONG op_rec_count = 0;
			op_rec_count++;
			if (INET_trace & TRACE_operations)
			{
				fprintf(stdout, "%04u: OP Recd %5u opcode %d\n",
						   inet_debug_timer(),
						   op_rec_count, packet->p_operation);
				fflush(stdout);
			}
		}
#endif
	} while (packet->p_operation == op_dummy);

	return main_port;
}

static bool select_multi(rem_port* main_port, UCHAR* buffer, SSHORT bufsize, SSHORT* length,
						 RemPortPtr& port)
{
/**************************************
 *
 *	s e l e c t _ m u l t i
 *
 **************************************
 *
 * Functional description
 *	Receive an IP packet from a port or clients of a port.
 *  Used only by the multiclient server on main server's port.
 *	If a connection request comes in, generate a new port
 *	block for the client.
 *
 **************************************/

// This code is used to test error handling in main server loop
#ifdef NEVERDEF
	static int dummyCnt = 0;
	if (++dummyCnt % 64 == 0)
		(Arg::Gds(isc_random) << "Simulated select_multi error").raise();
#endif

	for (;;)
	{
		select_port(main_port, &INET_select, port);
		if (port == main_port && (port->port_server_flags & SRVR_multi_client))
		{
			if (INET_shutting_down)
			{
				if (main_port->port_state == rem_port::PENDING)
				{
					main_port->port_state = rem_port::BROKEN;

					shutdown(main_port->port_handle, 2);
					SOCLOSE(main_port->port_handle);
				}
			}
			else if ((port = select_accept(main_port)))
			{
				if (!REMOTE_inflate(port, packet_receive, buffer, bufsize, length))
				{
					*length = 0;
				}
#ifdef WIRE_COMPRESS_SUPPORT
				if (port->port_z_data)
					INET_select->setZDataPort(port);
#endif
				return (*length) ? true : false;
			}

			continue;
		}
		if (port)
		{
			if (port->port_dummy_timeout < 0)
			{
				port->port_dummy_timeout = port->port_dummy_packet_interval;
				if (port->port_flags & PORT_async)
					continue;
				*length = 0;
				return true;
			}

			if (!REMOTE_inflate(port, packet_receive, buffer, bufsize, length))
			{
				if (port->port_flags & (PORT_disconnect | PORT_connecting))
				{
					continue;
				}
				*length = 0;
			}
#ifdef WIRE_COMPRESS_SUPPORT
			if (port->port_z_data)
				INET_select->setZDataPort(port);
#endif
			return (*length) ? true : false;
		}
		if (!select_wait(main_port, &INET_select))
		{
			port = NULL;
			return false;
		}
	}
}

static rem_port* select_accept( rem_port* main_port)
{
/**************************************
 *
 *	s e l e c t _ a c c e p t
 *
 **************************************
 *
 * Functional description
 *	Accept a new connection request.
 *
 **************************************/

	rem_port* const port = alloc_port(main_port);
	inet_ports->registerPort(port);

	port->port_handle = os_utils::accept(main_port->port_handle, NULL, NULL);
	if (port->port_handle == INVALID_SOCKET)
	{
		inet_error(true, port, "accept", isc_net_connect_err, INET_ERRNO);
	}

	setKeepAlive(port->port_handle);

	port->port_flags |= PORT_server;

	if (main_port->port_server_flags & SRVR_thread_per_port)
	{
		port->port_server_flags = (SRVR_server | SRVR_inet | SRVR_thread_per_port);
		return port;
	}

	return 0;
}

static void select_port(rem_port* main_port, Select* selct, RemPortPtr& port)
{
/**************************************
 *
 *	s e l e c t _ p o r t
 *
 **************************************
 *
 * Functional description
 *	Select a descriptor that is ready to read
 *	and return the port block. Additionally,
 *	check if a port's keepalive timer has
 *	expired and return the port block so that
 *	a keepalive packet can be queued. Return
 *	NULL if	none are active.
 *
 **************************************/

	MutexLockGuard guard(port_mutex, FB_FUNCTION);
	while (true)
	{
		Select::HandleState result = selct->checkNext(port);
		if (!port)
			return;

		switch (result)
		{
		case Select::SEL_BAD:
			if (port->port_state == rem_port::BROKEN || (port->port_flags & PORT_connecting))
				continue;
			if (port->port_flags & PORT_async)
				continue;
			return;

		case Select::SEL_DISCONNECTED:
			continue;

		case Select::SEL_READY:
			port->port_dummy_timeout = port->port_dummy_packet_interval;
			return;

		default:
			break;
		}

		if (port->port_dummy_timeout < 0)
			return;
	}
}

static bool select_wait( rem_port* main_port, Select* selct)
{
/**************************************
 *
 *	s e l e c t _ w a i t
 *
 **************************************
 *
 * Functional description
 *	Select interesting descriptors from
 *	port blocks and wait for something
 *	to read from them.
 *
 **************************************/
	struct timeval timeout;
	bool checkPorts = false;

	for (;;)
	{
		selct->clear();
		bool found = false;

		// Use the time interval between select() calls to expire
		// keepalive timers on all ports.

		time_t delta_time;
		if (selct->slct_time)
		{
			delta_time = time(NULL) - selct->slct_time;
			selct->slct_time += delta_time;
		}
		else
		{
			delta_time = 0;
			selct->slct_time = time(NULL);
		}

		{ // port_mutex scope
			MutexLockGuard guard(port_mutex, FB_FUNCTION);

			while (ports_to_close->hasData())
			{
				SOCKET s = ports_to_close->pop();
				SOCLOSE(s);
			}

			for (rem_port* port = main_port; port; port = port->port_next)
			{
				if (port->port_state == rem_port::PENDING &&
					// don't wait on still listening (not connected) async port
					!(port->port_handle == INVALID_SOCKET && (port->port_flags & PORT_async)))
 				{
					// Adjust down the port's keepalive timer.

					if (port->port_dummy_packet_interval)
					{
						port->port_dummy_timeout -= delta_time;
					}

					if (checkPorts)
					{
						// select() returned EBADF\WSAENOTSOCK - we have a broken socket
						// in current fdset. Search and return it to caller to close
						// broken connection correctly

						struct linger lngr;
						socklen_t optlen = sizeof(lngr);
						const bool badSocket =
#ifdef WIN_NT
							false;
#else
							(port->port_handle < 0 || port->port_handle >= FD_SETSIZE);
#endif

						if (badSocket || getsockopt(port->port_handle,
								SOL_SOCKET, SO_LINGER, (SCHAR*) &lngr, &optlen) != 0)
						{
							if (badSocket || INET_ERRNO == NOTASOCKET)
							{
								// not a socket, strange !
								gds__log("INET/select_wait: found \"not a socket\" socket : %" HANDLEFORMAT,
										 port->port_handle);

								// this will lead to receive() which will break bad connection
								selct->clear();
								if (!badSocket)
								{
									selct->set(port->port_handle);
								}
								return true;
							}
						}
					}

					// if process is shuting down - don't listen on main port
					if (!INET_shutting_down || port != main_port)
					{
						selct->set(port->port_handle);
						found = true;
					}
				}
			}
			checkPorts = false;
		} // port_mutex scope

		if (!found)
		{
			if (!INET_shutting_down && (main_port->port_server_flags & SRVR_multi_client))
				gds__log("INET/select_wait: client rundown complete, server exiting");

			return false;
		}

		for (;;)
		{
			// Before waiting for incoming packet, check for server shutdown
			if (tryStopMainThread && tryStopMainThread())
			{
				// this is not server port any more
				main_port->port_server_flags &= ~SRVR_multi_client;
				return false;
			}

			// Some platforms change the timeout in the select call.
			// Reset timeout for each iteration to avoid problems.
			timeout.tv_sec = SELECT_TIMEOUT;
			timeout.tv_usec = 0;

			selct->select(&timeout);
			const int inetErrNo = INET_ERRNO;

			//if (INET_shutting_down) {
			//	return false;
			//}

			if (selct->getCount() != -1)
			{
				RemPortPtr p(main_port);
				selct->checkStart(p);

				// if selct->slct_count is zero it means that we timed out of
				// select with nothing to read or accept, so clear the fd_set
				// bit as this value is undefined on some platforms (eg. HP-UX),
				// when the select call times out. Once these bits are cleared
				// they can be used in select_port()
				if (selct->getCount() == 0)
				{
					MutexLockGuard guard(port_mutex, FB_FUNCTION);
					for (rem_port* port = main_port; port; port = port->port_next)
					{
						selct->unset(port->port_handle);
					}
				}
				return true;
			}
			if (INTERRUPT_ERROR(inetErrNo))
				continue;
			if (inetErrNo == NOTASOCKET)
			{
				checkPorts = true;
				break;
			}

			gds__log("INET/select_wait: select failed, errno = %d", inetErrNo);
			return false;
		}	// for (;;)
	}
}

static int send_full( rem_port* port, PACKET * packet)
{
/**************************************
 *
 *	s e n d _ f u l l
 *
 **************************************
 *
 * Functional description
 *	Send a packet across a port to another process.
 *
 **************************************/

	if (!xdr_protocol(port->port_send, packet))
		return false;

#ifdef DEBUG
	{ // scope
		static ULONG op_sent_count = 0;
		op_sent_count++;
		if (INET_trace & TRACE_operations)
		{
			fprintf(stdout, "%05u: OP Sent %5u opcode %d\n", inet_debug_timer(),
				op_sent_count, packet->p_operation);
			fflush(stdout);
		}
	} // end scope
#endif

	return REMOTE_deflate(port->port_send, inet_write, packet_send, true);
}

static int send_partial( rem_port* port, PACKET * packet)
{
/**************************************
 *
 *	s e n d _ p a r t i a l
 *
 **************************************
 *
 * Functional description
 *	Send a packet across a port to another process.
 *
 **************************************/

#ifdef DEBUG
	{ // scope
		static ULONG op_sentp_count = 0;
		op_sentp_count++;
		if (INET_trace & TRACE_operations)
		{
			fprintf(stdout, "%05u: OP Sent %5u opcode %d (partial)\n", inet_debug_timer(),
				op_sentp_count, packet->p_operation);
			fflush(stdout);
		}
	} // end scope
#endif

	return xdr_protocol(port->port_send, packet);
}


RemoteXdr* xdrinet_create(rem_port* port, UCHAR* buffer, USHORT length, enum xdr_op x_op)
{
/**************************************
 *
 *	x d r i n e t _ c r e a t e
 *
 **************************************
 *
 * Functional description
 *	Initialize an XDR stream.
 *
 **************************************/

	RemoteXdr* xdrs = FB_NEW InetXdr;

	xdrs->x_public = port;
	xdrs->create(reinterpret_cast<SCHAR*>(buffer), length, x_op);

	return xdrs;
}

#ifdef HAVE_SETITIMER
static void alarm_handler( int x)
{
/**************************************
 *
 *	a l a r m _ h a n d l e r
 *
 **************************************
 *
 * Functional description
 *	Handle an alarm clock interrupt.  If we were waiting on
 *	a semaphone, zap it.
 *
 **************************************/
}
#endif

void get_peer_info(rem_port* port)
{
/**************************************
*
*	g e t _ p e e r _ i n f o
*
**************************************
*
* Functional description
*	Port just connected. Obtain some info about connection and peer.
*
**************************************/
	port->port_protocol_id = "TCPv4";

	SockAddr address;
	if (address.getpeername(port->port_handle) == 0)
	{
		address.unmapV4();	// convert mapped IPv4 to regular IPv4
		char host[64];		// 32 digits, 7 colons, 1 trailing null byte
		char serv[16];
		int nameinfo = getnameinfo(address.ptr(), address.length(), host, sizeof(host),
			serv, sizeof(serv), NI_NUMERICHOST | NI_NUMERICSERV);

		if (!nameinfo)
			port->port_address.printf("%s/%s", host, serv);

		if (address.family() == AF_INET6)
			port->port_protocol_id = "TCPv6";
	}
}


static void inet_gen_error(bool releasePort, rem_port* port, const Arg::StatusVector& v)
{
/**************************************
 *
 *	i n e t _ g e n _ e r r o r
 *
 **************************************
 *
 * Functional description
 *	An error has occurred.  Mark the port as broken.
 *	Format the status vector if there is one and
 *	save the status vector strings in a permanent place.
 *
 **************************************/
	port->port_state = rem_port::BROKEN;

	string node_name(port->port_connection ? port->port_connection->str_data : "(unknown)");

	if (releasePort)
	{
		disconnect(port);
	}

	Arg::Gds error(isc_network_error);
	error << Arg::Str(node_name) << v;
	error.raise();
}


bool_t InetXdr::x_getbytes(SCHAR* buff, unsigned bytecount)
{
/**************************************
 *
 *	i n e t _ g e t b y t e s
 *
 **************************************
 *
 * Functional description
 *	Get a bunch of bytes from a memory stream if it fits.
 *
 **************************************/
	if (x_public->port_flags & PORT_server)
		return REMOTE_getbytes(this, buff, bytecount);

	// Use memcpy to optimize bulk transfers.

	while (bytecount > sizeof(ISC_QUAD))
	{
		if (x_handy >= bytecount)
		{
			memcpy(buff, x_private, bytecount);
			x_private += bytecount;
			x_handy -= bytecount;
			return TRUE;
		}

		if (x_handy > 0)
		{
			memcpy(buff, x_private, x_handy);
			x_private += x_handy;
			buff += x_handy;
			bytecount -= x_handy;
			x_handy = 0;
		}

		if (!inet_read(this))
			return FALSE;
	}

	// Scalar values and bulk transfer remainder fall thru
	// to be moved byte-by-byte to avoid memcpy setup costs.

	if (!bytecount)
		return TRUE;

	if (x_handy >= bytecount)
	{
		x_handy -= bytecount;
		while (bytecount--)
			*buff++ = *x_private++;

		return TRUE;
	}

	while (bytecount--)
	{
		if (x_handy == 0 && !inet_read(this))
			return FALSE;
		*buff++ = *x_private++;
		--x_handy;
	}

	return TRUE;
}


static void inet_error(bool releasePort, rem_port* port, const TEXT* function, ISC_STATUS operation, int status)
{
/**************************************
 *
 *	i n e t _ e r r o r
 *
 **************************************
 *
 * Functional description
 *	An I/O error has occurred.  Call
 * 	inet_gen_error with the appropriate args
 *	to format the status vector if any.
 *
 **************************************/
	if (status)
	{
		if (port->port_state == rem_port::PENDING)
		{
			string err;
			err.printf("INET/inet_error: %s errno = %d", function, status);

			if (port->port_peer_name.hasData() || port->port_address.hasData())
			{
				err.append(port->port_flags & PORT_async ? ", aux " : ", ");
				err.append(port->port_server_flags ? "client" : "server");

				if (port->port_peer_name.hasData())
				{
					err.append(" host = ");
					err.append(port->port_peer_name);
				}

				if (port->port_address.hasData())
				{
					if (port->port_peer_name.hasData())
						err.append(",");

					err.append(" address = ");
					err.append(port->port_address);
				}
			}

			if (port->port_user_name.hasData())
			{
				err.append(", user = ");
				err.append(port->port_user_name);
			}

			// Address could contain percent sign inside, therefore make
			// sure error string not used as printf format string.
			gds__log("%s", err.c_str());
		}

		inet_gen_error(releasePort, port, Arg::Gds(operation) << SYS_ERR(status));
	}
	else
	{
		// No status value, just format the basic arguments.
		inet_gen_error(releasePort, port, Arg::Gds(operation));
	}
}

bool_t InetXdr::x_putbytes(const SCHAR* buff, unsigned bytecount)
{
/**************************************
 *
 *	i n e t _ p u t b y t e s
 *
 **************************************
 *
 * Functional description
 *	Put a bunch of bytes to a memory stream if it fits.
 *
 **************************************/

	// Use memcpy to optimize bulk transfers.

	while (bytecount > sizeof(ISC_QUAD))
	{
		if (x_handy >= bytecount)
		{
			memcpy(x_private, buff, bytecount);
			x_private += bytecount;
			x_handy -= bytecount;
			return TRUE;
		}

		if (x_handy > 0)
		{
			memcpy(x_private, buff, x_handy);
			x_private += x_handy;
			buff += x_handy;
			bytecount -= x_handy;
			x_handy = 0;
		}

		if (!REMOTE_deflate(this, inet_write, packet_send, false))
		{
			return FALSE;
		}
	}

	// Scalar values and bulk transfer remainder fall thru
	// to be moved byte-by-byte to avoid memcpy setup costs.

	if (!bytecount)
		return TRUE;

	if (x_handy >= bytecount)
	{
		x_handy -= bytecount;
		while (bytecount--)
			*x_private++ = *buff++;

		return TRUE;
	}

	while (bytecount--)
	{
		if (x_handy == 0 && !REMOTE_deflate(this, inet_write, packet_send, false))
			return FALSE;
		--x_handy;
		*x_private++ = *buff++;
	}

	return TRUE;
}

static bool inet_read( RemoteXdr* xdrs)
{
/**************************************
 *
 *	i n e t _ r e a d
 *
 **************************************
 *
 * Functional description
 *	Read a buffer full of data.  If we receive a bad packet,
 *	send the moral equivalent of a NAK and retry.  ACK all
 *	partial packets.  Don't ACK the last packet -- the next
 *	message sent will handle this.
 *
 **************************************/
	rem_port* port = xdrs->x_public;
	char* p = xdrs->x_base;
	const char* const end = p + INET_remote_buffer;

	// If buffer is not completely empty, slide down what's left

	if (xdrs->x_handy > 0)
	{
		memmove(p, xdrs->x_private, xdrs->x_handy);
		p += xdrs->x_handy;
	}

	SSHORT length = end - p;
	port->port_z_data = false;
	if (!REMOTE_inflate(port, packet_receive2, (UCHAR*)p, length, &length))
		return false;
	p += length;

	xdrs->x_handy = (SCHAR *) p - xdrs->x_base;
	xdrs->x_private = xdrs->x_base;

	return true;
}

static bool packet_receive2(rem_port* port, UCHAR* p, SSHORT bufSize, SSHORT* length)
{
	*length = 0;

	while (true)
	{
		SSHORT l = bufSize - *length;
		if (!packet_receive(port, p + *length, l, &l))
			return false;

		if (l >= 0)
		{
			*length += l;
			break;
		}

		*length -= l;
		if (!packet_send(port, 0, 0))
			return false;
	}

	return true;
}

static rem_port* inet_try_connect(PACKET* packet,
								  Rdb* rdb,
								  const PathName& file_name,
								  const TEXT* node_name,
								  ClumpletReader& dpb,
								  RefPtr<const Config>* config,
								  const PathName* ref_db_name,
								  int af)
{
/**************************************
 *
 *	i n e t _ t r y _ c o n n e c t
 *
 **************************************
 *
 * Functional description
 *	Given a packet with formatted protocol infomation,
 *	set header information and try the connection.
 *
 *	If a connection is established, return a port block, otherwise
 *	return NULL.
 *
 **************************************/
	P_CNCT* cnct = &packet->p_cnct;
	packet->p_operation = op_connect;
	cnct->p_cnct_operation = 0;
	cnct->p_cnct_cversion = CONNECT_VERSION3;
	cnct->p_cnct_client = ARCHITECTURE;

	const PathName& cnct_file(ref_db_name ? (*ref_db_name) : file_name);
	cnct->p_cnct_file.cstr_length = (ULONG) cnct_file.length();
	cnct->p_cnct_file.cstr_address = reinterpret_cast<const UCHAR*>(cnct_file.c_str());

	// If we can't talk to a server, punt.  Let somebody else generate
	// an error.  status_vector will have the network error info.

	rem_port* port = NULL;
	try
	{
		port = INET_connect(node_name, packet, false, &dpb, config, af);
	}
	catch (const Exception&)
	{
		delete rdb;
		throw;
	}

	// Get response packet from server.

	rdb->rdb_port = port;
	port->port_context = rdb;
	if (!port->receive(packet))
	{
		rdb->rdb_port = NULL;
		delete rdb;
		inet_error(true, port, "receive in try_connect", isc_net_connect_err, INET_ERRNO);
	}

	return port;
}

static bool inet_write(RemoteXdr* xdrs)
{
/**************************************
 *
 *	i n e t _ w r i t e
 *
 **************************************
 *
 * Functional description
 *	Write a buffer full of data.
 *
 **************************************/
	// Encode the data portion of the packet

	rem_port* port = xdrs->x_public;
	const char* p = xdrs->x_base;
	USHORT length = xdrs->x_private - p;

	// Send data in manageable hunks.  If a packet is partial, indicate
	// that with a negative length.  A positive length marks the end.

	while (length)
	{
		const SSHORT l = (SSHORT) MIN(length, INET_remote_buffer);
		length -= l;
		if (!packet_send(port, p, (SSHORT) (length ? -l : l)))
			return false;
		p += l;
	}

	xdrs->x_private = xdrs->x_base;
	xdrs->x_handy = INET_remote_buffer;

	return true;

}

#ifdef DEBUG
static void packet_print(const TEXT* string, const UCHAR* packet, int length, ULONG counter)
{
/**************************************
 *
 *	p a c k e t _ p r i n t
 *
 **************************************
 *
 * Functional description
 *	Print a summary of packet.
 *
 **************************************/
	int sum = 0;
	for (int l = length; l > 0; --l)
		sum += *packet++;

	fprintf(stdout, "%05u:    PKT %s\t(%u): length = %4d, checksum = %d\n",
			   inet_debug_timer(), string, counter, length, sum);
	fflush(stdout);
}
#endif

static bool packet_receive(rem_port* port, UCHAR* buffer, SSHORT buffer_length, SSHORT* length)
{
/**************************************
 *
 *	p a c k e t _ r e c e i v e
 *
 **************************************
 *
 * Functional description
 *	Receive a packet and pass on it's goodness.  If it's good,
 *	return true and the reported length of the packet, and update
 *	the receive sequence number.  If it's bad, return false.  If it's
 *	a duplicate message, just ignore it.
 *
 **************************************/

	if (port->port_flags & PORT_disconnect) {
		return false;
	}

	timeval timeout;
	timeout.tv_usec = 0;
	timeval* time_ptr = NULL;

	if (port->port_protocol == 0)
	{
		// If the protocol is 0 we are still in the process of establishing
		// a connection. Add a time out to the wait.
		timeout.tv_sec = port->port_connect_timeout;
		time_ptr = &timeout;
	}
	else if (port->port_dummy_packet_interval > 0)
	{
		// Set the time interval for sending dummy packets to the client
		timeout.tv_sec = port->port_dummy_packet_interval;
		time_ptr = &timeout;
	}

	// On Linux systems (and possibly others too) select will eventually
	// change timout values so save it here for later reuse.
	// Thanks to Brad Pepers who reported this bug  FSG 3 MAY 2001
	const timeval savetime = timeout;

	const SOCKET ph = port->port_handle;
	if (ph == INVALID_SOCKET)
	{
		const bool releasePort = (port->port_flags & PORT_server);
		if (!(port->port_flags & PORT_disconnect) && releasePort)
			inet_error(true, port, "invalid socket in packet_receive", isc_net_read_err, EINVAL);

		return false;
	}

	// Unsed to send a dummy packet, but too big to be defined in the loop.
	PACKET packet;

	int n = 0;
	int inetErrNo;
	LocalStatus ls;
	CheckStatusWrapper st(&ls);

	for (;;)
	{
		// Implement an error-detection protocol to ensure that the client
		// is still there.  Use the select() call with a timeout to wait on
		// the connection for an incoming packet.  If none comes within a
		// suitable time interval, write a dummy packet on the connection.
		// If the client is not there, an error will be returned on the write.
		// If the client is there, the dummy packet will be ignored by all
		// InterBase clients V4 or greater.  This protocol will detect when
		// clients are lost abnormally through reboot or network disconnect.

		// Don't send op_dummy packets on aux port; the server won't
		// read them because it only writes to aux ports.

		if ( !(port->port_flags & PORT_async) )
		{
			Select slct;
			slct.set(ph);

			int slct_count;
			for (;;)
			{
				slct.select(time_ptr);
				slct_count = slct.getCount();
				inetErrNo = INET_ERRNO;

				// restore original timeout value FSG 3 MAY 2001
				timeout = savetime;

				if (slct_count != -1 || !INTERRUPT_ERROR(inetErrNo))
				{
					break;
				}
			}

			if (slct_count == -1)
			{
				if (!(port->port_flags & PORT_disconnect))
				{
					try
					{
						inet_error(false, port, "select in packet_receive", isc_net_read_err, inetErrNo);
					}
					catch (const Exception&) { }
				}
				return false;
			}

			if (!slct_count)
			{
				if (port->port_protocol == 0)
					return false;

#ifdef DEBUG
				if (INET_trace & TRACE_operations)
				{
					fprintf(stdout, "%05u: OP Sent: op_dummy\n", inet_debug_timer());
					fflush(stdout);
				}
#endif
				packet.p_operation = op_dummy;
				if (!send_full(port, &packet))
				{
					return false;
				}
				continue;
			}
		}

		n = recv(port->port_handle, reinterpret_cast<char*>(buffer), buffer_length, 0);
		inetErrNo = INET_ERRNO;

		// decrypt
		if (n > 0 && port->port_crypt_plugin)
		{
			port->port_crypt_plugin->decrypt(&st, n, buffer, buffer);
			if (st.getState() & Firebird::IStatus::STATE_ERRORS)
			{
				status_exception::raise(&st);
			}
		}

		if (n != -1 || !INTERRUPT_ERROR(inetErrNo))
			break;
	}

	if (n <= 0 && (port->port_flags & PORT_disconnect)) {
		return false;
	}

	if (n == -1)
	{
		try
		{
			inet_error(false, port, "read", isc_net_read_err, inetErrNo);
		}
		catch (const Exception&) { }
		return false;
	}

	if (!n)
	{
		port->port_state = rem_port::BROKEN;
		return false;
	}

#ifdef DEBUG
	{ // scope
		INET_count_recv++;
		INET_bytes_recv += n;
		if (INET_trace & TRACE_packets)
			packet_print("receive", buffer, n, INET_count_recv);
		INET_force_error--;
		if (INET_force_error == 0)
		{
			INET_force_error = 1;
			try
			{
				inet_error(false, port, "simulated error - read", isc_net_read_err);
			}
			catch (const Exception&) { }
			return false;
		}
	} // end scope
#endif

	port->port_rcv_packets++;
	port->port_rcv_bytes += n;

	*length = n;

	return true;
}


static bool packet_send( rem_port* port, const SCHAR* buffer, SSHORT buffer_length)
{
/**************************************
 *
 *	p a c k e t _ s e n d
 *
 **************************************
 *
 * Functional description
 *	Send some data on it's way.
 *
 **************************************/

	SSHORT length = buffer_length;
	const char* data = buffer;

	// encrypt
	HalfStaticArray<char, BUFFER_TINY> b;
	if (port->port_crypt_plugin && port->port_crypt_complete)
	{
		LocalStatus ls;
		CheckStatusWrapper st(&ls);

		char* d = b.getBuffer(buffer_length);
		port->port_crypt_plugin->encrypt(&st, buffer_length, data, d);
		if (st.getState() & Firebird::IStatus::STATE_ERRORS)
		{
			status_exception::raise(&st);
		}

		data = d;
	}

	while (length)
	{
#ifdef DEBUG
		if (INET_trace & TRACE_operations)
		{
			fprintf(stdout, "Before Send\n");
			fflush(stdout);
		}
#endif
		SSHORT n = send(port->port_handle, data, length, FB_SEND_FLAGS);
#if COMPRESS_DEBUG > 1
		fprintf(stderr, "send(%d, %p, %d, FB_SEND_FLAGS) == %d\n", port->port_handle, data, length, n);
#endif
#ifdef DEBUG
		if (INET_trace & TRACE_operations)
		{
			fprintf(stdout, "After Send n is %d\n", n);
			fflush(stdout);
		}
#endif
		if (n == length) {
			break;
		}

		if (n == -1)
		{
			if (INTERRUPT_ERROR(INET_ERRNO)) {
				continue;
			}

			try
			{
				inet_error(false, port, "send", isc_net_write_err, INET_ERRNO);
			}
			catch (const Exception&) { }
			return false;
		}

		data += n;
		length -= n;
	}

#ifdef HAVE_SETITIMER
	struct itimerval internal_timer, client_timer;
	struct sigaction internal_handler, client_handler;
#endif // HAVE_SETITIMER

	if ((port->port_flags & PORT_async) && !(port->port_flags & PORT_no_oob))
	{
		int count = 0;
		SSHORT n;
		int inetErrNo = 0;
		const char* b = buffer;
		while ((n = send(port->port_handle, b, 1, MSG_OOB | FB_SEND_FLAGS)) == -1 &&
				(INET_ERRNO == ENOBUFS || INTERRUPT_ERROR(INET_ERRNO)))
		{
			inetErrNo = INET_ERRNO;

			if (count++ > 20) {
				break;
			}

#ifndef HAVE_SETITIMER
#ifdef WIN_NT
			SleepEx(50, TRUE);
#else
			sleep(1);
#endif
		} // end of while() loop for systems without setitimer.
#else // HAVE_SETITIMER
			if (count == 1)
			{
				// Wait in a loop until the lock becomes available

				internal_timer.it_interval.tv_sec = 0;
				internal_timer.it_interval.tv_usec = 0;
				internal_timer.it_value.tv_sec = 0;
				internal_timer.it_value.tv_usec = 0;
				setitimer(ITIMER_REAL, &internal_timer, &client_timer);
				internal_handler.sa_handler = alarm_handler;
				sigemptyset(&internal_handler.sa_mask);
				internal_handler.sa_flags = SA_RESTART;
				sigaction(SIGALRM, &internal_handler, &client_handler);
			}

			internal_timer.it_value.tv_sec = 0;
			internal_timer.it_value.tv_usec = 50000;
			setitimer(ITIMER_REAL, &internal_timer, NULL);
			pause();
		} // end of while() loop for systems with setitimer

		if (count)
		{
			// Restore user's outstanding alarm request and handler

			internal_timer.it_value.tv_sec = 0;
			internal_timer.it_value.tv_usec = 0;
			setitimer(ITIMER_REAL, &internal_timer, NULL);
			sigaction(SIGALRM, &client_handler, NULL);
			setitimer(ITIMER_REAL, &client_timer, NULL);
		}
#endif // HAVE_SETITIMER

		if (n == -1)
		{
			try
			{
				inet_error(false, port, "send/oob", isc_net_write_err, inetErrNo);
			}
			catch (const Exception&) { }
			return false;
		}
	}

#ifdef DEBUG
	{ // scope
		INET_count_send++;
		INET_bytes_send += buffer_length;
		if (INET_trace & TRACE_packets)
			packet_print("send", (const UCHAR*) buffer, buffer_length, INET_count_send);
		INET_force_error--;
		if (INET_force_error == 0)
		{
			INET_force_error = 1;
			try
			{
				inet_error(false, port, "simulated error - send", isc_net_write_err, 0);
			}
			catch (const Exception&) { }
			return false;
		}
	} // end scope
#endif

	port->port_snd_packets++;
	port->port_snd_bytes += buffer_length;

	return true;
}

static bool setNoNagleOption(rem_port* port)
{
/**************************************
 *
 *      s e t N o N a g l e O p t i o n
 *
 **************************************
 *
 * Functional description
 *      Set TCP_NODELAY, return false
 *		in case of unexpected error
 *
 **************************************/
	if (port->getPortConfig()->getTcpNoNagle())
	{
		int optval = TRUE;
		int n = setsockopt(port->port_handle, IPPROTO_TCP, TCP_NODELAY,
						   (SCHAR*) &optval, sizeof(optval));

		if (n == -1)
		{
			return false;
		}
	}
	return true;
}

static bool setKeepAlive(SOCKET s)
{
/**************************************
 *
 *      s e t K e e p A l i v e
 *
 **************************************
 *
 * Functional description
 *      Set SO_KEEPALIVE, return false
 *		in case of unexpected error
 *
 **************************************/
	int optval = 1;
	int n = setsockopt(s, SOL_SOCKET, SO_KEEPALIVE,
					   (SCHAR*) &optval, sizeof(optval));
	return n != -1;
}

void setStopMainThread(FPTR_INT func)
{
/**************************************
 *
 *   s e t S t o p M a i n T h r e a d
 *
 **************************************
 *
 * Functional description
 *	Set function called by main thread
 *	in order to check for shutdown.
 *
 **************************************/
	tryStopMainThread = func;
}

namespace os_utils
{

// force socket descriptor to have SOCK_CLOEXEC set
SOCKET socket(int domain, int type, int protocol)
{
#ifdef WIN_NT
	return ::socket(domain, type, protocol);
#else
	int fd;
#if HAVE_DECL_SOCK_CLOEXEC
	do {
		fd = ::socket(domain, type | SOCK_CLOEXEC, protocol);
	} while (fd < 0 && SYSCALL_INTERRUPTED(errno));

	if (fd < 0 && errno == EINVAL)	// probably SOCK_CLOEXEC not accepted
#endif
	{
		do {
			fd = ::socket(domain, type, protocol);
		} while (fd < 0 && SYSCALL_INTERRUPTED(errno));
	}

	setCloseOnExec(fd);
	return fd;
#endif
}

// force socket descriptor to have SOCK_CLOEXEC set
SOCKET accept(SOCKET sockfd, struct sockaddr* addr, socklen_t* addrlen)
{
#ifdef WIN_NT
	return ::accept(sockfd, addr, addrlen);
#else
	int fd;
#if defined(HAVE_ACCEPT4) && HAVE_DECL_SOCK_CLOEXEC
	do {
		fd = ::accept4(sockfd, addr, addrlen, SOCK_CLOEXEC);
	} while (fd < 0 && SYSCALL_INTERRUPTED(errno));

	if (fd < 0 && errno == EINVAL)	// probably SOCK_CLOEXEC not accepted
#endif
	{
		do {
			fd = ::accept(sockfd, addr, addrlen);
		} while (fd < 0 && SYSCALL_INTERRUPTED(errno));
	}

	setCloseOnExec(fd);
	return fd;
#endif
}

} // namespace os_utils