xref: /haiku/src/add-ons/kernel/network/protocols/tcp/TCPEndpoint.cpp (revision 084e24d0bf3808808e2bf58d4d65f493bd2b8f49)
1 /*
2  * Copyright 2006-2010, Haiku, Inc. All Rights Reserved.
3  * Distributed under the terms of the MIT License.
4  *
5  * Authors:
6  *		Andrew Galante, haiku.galante@gmail.com
7  *		Axel Dörfler, axeld@pinc-software.de
8  *		Hugo Santos, hugosantos@gmail.com
9  */
10 
11 
12 #include "TCPEndpoint.h"
13 
14 #include <netinet/in.h>
15 #include <netinet/ip.h>
16 #include <netinet/tcp.h>
17 #include <new>
18 #include <signal.h>
19 #include <stdlib.h>
20 #include <string.h>
21 #include <stdint.h>
22 
23 #include <KernelExport.h>
24 #include <Select.h>
25 
26 #include <net_buffer.h>
27 #include <net_datalink.h>
28 #include <net_stat.h>
29 #include <NetBufferUtilities.h>
30 #include <NetUtilities.h>
31 
32 #include <lock.h>
33 #include <tracing.h>
34 #include <util/AutoLock.h>
35 #include <util/list.h>
36 
37 #include "EndpointManager.h"
38 
39 
40 // References:
41 //	- RFC 793 - Transmission Control Protocol
42 //	- RFC 813 - Window and Acknowledgement Strategy in TCP
43 //	- RFC 1337 - TIME_WAIT Assassination Hazards in TCP
44 //
45 // Things incomplete in this implementation:
46 //	- TCP Extensions for High Performance, RFC 1323 - RTTM, PAWS
47 //	- Congestion Control, RFC 5681
48 //	- Limited Transit, RFC 3042
49 //	- SACK, Selective Acknowledgment; RFC 2018, RFC 2883, RFC 6675
50 //	- NewReno Modification to TCP's Fast Recovery, RFC 2582
51 //
52 // Things this implementation currently doesn't implement:
53 //	- Explicit Congestion Notification (ECN), RFC 3168
54 //	- SYN-Cache
55 //	- Forward RTO-Recovery, RFC 4138
56 //	- Time-Wait hash instead of keeping sockets alive
57 
58 #define PrintAddress(address) \
59 	AddressString(Domain(), address, true).Data()
60 
61 //#define TRACE_TCP
62 //#define PROBE_TCP
63 
64 #ifdef TRACE_TCP
65 // the space before ', ##args' is important in order for this to work with cpp 2.95
66 #	define TRACE(format, args...)	dprintf("%" B_PRId32 ": TCP [%" \
67 		B_PRIdBIGTIME "] %p (%12s) " format "\n", find_thread(NULL), \
68 		system_time(), this, name_for_state(fState) , ##args)
69 #else
70 #	define TRACE(args...)			do { } while (0)
71 #endif
72 
73 #ifdef PROBE_TCP
74 #	define PROBE(buffer, window) \
75 	dprintf("TCP PROBE %" B_PRIdBIGTIME " %s %s %" B_PRIu32 " snxt %" B_PRIu32 \
76 		" suna %" B_PRIu32 " cw %" B_PRIu32 " sst %" B_PRIu32 " win %" \
77 		B_PRIu32 " swin %" B_PRIu32 " smax-suna %" B_PRIu32 " savail %" \
78 		B_PRIuSIZE " sqused %" B_PRIuSIZE " rto %" B_PRIdBIGTIME "\n", \
79 		system_time(), PrintAddress(buffer->source), \
80 		PrintAddress(buffer->destination), buffer->size, fSendNext.Number(), \
81 		fSendUnacknowledged.Number(), fCongestionWindow, fSlowStartThreshold, \
82 		window, fSendWindow, (fSendMax - fSendUnacknowledged).Number(), \
83 		fSendQueue.Available(fSendNext), fSendQueue.Used(), fRetransmitTimeout)
84 #else
85 #	define PROBE(buffer, window)	do { } while (0)
86 #endif
87 
88 #if TCP_TRACING
89 namespace TCPTracing {
90 
91 class Receive : public AbstractTraceEntry {
92 public:
93 	Receive(TCPEndpoint* endpoint, tcp_segment_header& segment, uint32 window,
94 			net_buffer* buffer)
95 		:
96 		fEndpoint(endpoint),
97 		fBuffer(buffer),
98 		fBufferSize(buffer->size),
99 		fSequence(segment.sequence),
100 		fAcknowledge(segment.acknowledge),
101 		fWindow(window),
102 		fState(endpoint->State()),
103 		fFlags(segment.flags)
104 	{
105 		Initialized();
106 	}
107 
108 	virtual void AddDump(TraceOutput& out)
109 	{
110 		out.Print("tcp:%p (%12s) receive buffer %p (%" B_PRIu32 " bytes), "
111 			"flags %#" B_PRIx8 ", seq %" B_PRIu32 ", ack %" B_PRIu32
112 			", wnd %" B_PRIu32, fEndpoint, name_for_state(fState), fBuffer,
113 			fBufferSize, fFlags, fSequence, fAcknowledge, fWindow);
114 	}
115 
116 protected:
117 	TCPEndpoint*	fEndpoint;
118 	net_buffer*		fBuffer;
119 	uint32			fBufferSize;
120 	uint32			fSequence;
121 	uint32			fAcknowledge;
122 	uint32			fWindow;
123 	tcp_state		fState;
124 	uint8			fFlags;
125 };
126 
127 class Send : public AbstractTraceEntry {
128 public:
129 	Send(TCPEndpoint* endpoint, tcp_segment_header& segment, net_buffer* buffer,
130 			tcp_sequence firstSequence, tcp_sequence lastSequence)
131 		:
132 		fEndpoint(endpoint),
133 		fBuffer(buffer),
134 		fBufferSize(buffer->size),
135 		fSequence(segment.sequence),
136 		fAcknowledge(segment.acknowledge),
137 		fFirstSequence(firstSequence.Number()),
138 		fLastSequence(lastSequence.Number()),
139 		fState(endpoint->State()),
140 		fFlags(segment.flags)
141 	{
142 		Initialized();
143 	}
144 
145 	virtual void AddDump(TraceOutput& out)
146 	{
147 		out.Print("tcp:%p (%12s) send buffer %p (%" B_PRIu32 " bytes), "
148 			"flags %#" B_PRIx8 ", seq %" B_PRIu32 ", ack %" B_PRIu32
149 			", first %" B_PRIu32 ", last %" B_PRIu32, fEndpoint,
150 			name_for_state(fState), fBuffer, fBufferSize, fFlags, fSequence,
151 			fAcknowledge, fFirstSequence, fLastSequence);
152 	}
153 
154 protected:
155 	TCPEndpoint*	fEndpoint;
156 	net_buffer*		fBuffer;
157 	uint32			fBufferSize;
158 	uint32			fSequence;
159 	uint32			fAcknowledge;
160 	uint32			fFirstSequence;
161 	uint32			fLastSequence;
162 	tcp_state		fState;
163 	uint8			fFlags;
164 };
165 
166 class State : public AbstractTraceEntry {
167 public:
168 	State(TCPEndpoint* endpoint)
169 		:
170 		fEndpoint(endpoint),
171 		fState(endpoint->State())
172 	{
173 		Initialized();
174 	}
175 
176 	virtual void AddDump(TraceOutput& out)
177 	{
178 		out.Print("tcp:%p (%12s) state change", fEndpoint,
179 			name_for_state(fState));
180 	}
181 
182 protected:
183 	TCPEndpoint*	fEndpoint;
184 	tcp_state		fState;
185 };
186 
187 class Spawn : public AbstractTraceEntry {
188 public:
189 	Spawn(TCPEndpoint* listeningEndpoint, TCPEndpoint* spawnedEndpoint)
190 		:
191 		fListeningEndpoint(listeningEndpoint),
192 		fSpawnedEndpoint(spawnedEndpoint)
193 	{
194 		Initialized();
195 	}
196 
197 	virtual void AddDump(TraceOutput& out)
198 	{
199 		out.Print("tcp:%p spawns %p", fListeningEndpoint, fSpawnedEndpoint);
200 	}
201 
202 protected:
203 	TCPEndpoint*	fListeningEndpoint;
204 	TCPEndpoint*	fSpawnedEndpoint;
205 };
206 
207 class Error : public AbstractTraceEntry {
208 public:
209 	Error(TCPEndpoint* endpoint, const char* error, int32 line)
210 		:
211 		fEndpoint(endpoint),
212 		fLine(line),
213 		fError(error),
214 		fState(endpoint->State())
215 	{
216 		Initialized();
217 	}
218 
219 	virtual void AddDump(TraceOutput& out)
220 	{
221 		out.Print("tcp:%p (%12s) error at line %" B_PRId32 ": %s", fEndpoint,
222 			name_for_state(fState), fLine, fError);
223 	}
224 
225 protected:
226 	TCPEndpoint*	fEndpoint;
227 	int32			fLine;
228 	const char*		fError;
229 	tcp_state		fState;
230 };
231 
232 class TimerSet : public AbstractTraceEntry {
233 public:
234 	TimerSet(TCPEndpoint* endpoint, const char* which, bigtime_t timeout)
235 		:
236 		fEndpoint(endpoint),
237 		fWhich(which),
238 		fTimeout(timeout),
239 		fState(endpoint->State())
240 	{
241 		Initialized();
242 	}
243 
244 	virtual void AddDump(TraceOutput& out)
245 	{
246 		out.Print("tcp:%p (%12s) %s timer set to %" B_PRIdBIGTIME, fEndpoint,
247 			name_for_state(fState), fWhich, fTimeout);
248 	}
249 
250 protected:
251 	TCPEndpoint*	fEndpoint;
252 	const char*		fWhich;
253 	bigtime_t		fTimeout;
254 	tcp_state		fState;
255 };
256 
257 class TimerTriggered : public AbstractTraceEntry {
258 public:
259 	TimerTriggered(TCPEndpoint* endpoint, const char* which)
260 		:
261 		fEndpoint(endpoint),
262 		fWhich(which),
263 		fState(endpoint->State())
264 	{
265 		Initialized();
266 	}
267 
268 	virtual void AddDump(TraceOutput& out)
269 	{
270 		out.Print("tcp:%p (%12s) %s timer triggered", fEndpoint,
271 			name_for_state(fState), fWhich);
272 	}
273 
274 protected:
275 	TCPEndpoint*	fEndpoint;
276 	const char*		fWhich;
277 	tcp_state		fState;
278 };
279 
280 class APICall : public AbstractTraceEntry {
281 public:
282 	APICall(TCPEndpoint* endpoint, const char* which)
283 		:
284 		fEndpoint(endpoint),
285 		fWhich(which),
286 		fState(endpoint->State())
287 	{
288 		Initialized();
289 	}
290 
291 	virtual void AddDump(TraceOutput& out)
292 	{
293 		out.Print("tcp:%p (%12s) api call: %s", fEndpoint,
294 			name_for_state(fState), fWhich);
295 	}
296 
297 protected:
298 	TCPEndpoint*	fEndpoint;
299 	const char*		fWhich;
300 	tcp_state		fState;
301 };
302 
303 }	// namespace TCPTracing
304 
305 #	define T(x)	new(std::nothrow) TCPTracing::x
306 #else
307 #	define T(x)
308 #endif	// TCP_TRACING
309 
310 
311 // constants for the fFlags field
312 enum {
313 	FLAG_OPTION_WINDOW_SCALE	= 0x01,
314 	FLAG_OPTION_TIMESTAMP		= 0x02,
315 	// TODO: Should FLAG_NO_RECEIVE apply as well to received connections?
316 	//       That is, what is expected from accept() after a shutdown()
317 	//       is performed on a listen()ing socket.
318 	FLAG_NO_RECEIVE				= 0x04,
319 	FLAG_CLOSED					= 0x08,
320 	FLAG_DELETE_ON_CLOSE		= 0x10,
321 	FLAG_LOCAL					= 0x20,
322 	FLAG_RECOVERY				= 0x40,
323 	FLAG_OPTION_SACK_PERMITTED	= 0x80,
324 };
325 
326 
327 static const int kTimestampFactor = 1000;
328 	// conversion factor between usec system time and msec tcp time
329 
330 
331 static inline bigtime_t
332 absolute_timeout(bigtime_t timeout)
333 {
334 	if (timeout == 0 || timeout == B_INFINITE_TIMEOUT)
335 		return timeout;
336 
337 	return timeout + system_time();
338 }
339 
340 
341 static inline status_t
342 posix_error(status_t error)
343 {
344 	if (error == B_TIMED_OUT)
345 		return B_WOULD_BLOCK;
346 
347 	return error;
348 }
349 
350 
351 static inline bool
352 in_window(const tcp_sequence& sequence, const tcp_sequence& receiveNext,
353 	uint32 receiveWindow)
354 {
355 	return sequence >= receiveNext && sequence < (receiveNext + receiveWindow);
356 }
357 
358 
359 static inline bool
360 segment_in_sequence(const tcp_segment_header& segment, int size,
361 	const tcp_sequence& receiveNext, uint32 receiveWindow)
362 {
363 	tcp_sequence sequence(segment.sequence);
364 	if (size == 0) {
365 		if (receiveWindow == 0)
366 			return sequence == receiveNext;
367 		return in_window(sequence, receiveNext, receiveWindow);
368 	} else {
369 		if (receiveWindow == 0)
370 			return false;
371 		return in_window(sequence, receiveNext, receiveWindow)
372 			|| in_window(sequence + size - 1, receiveNext, receiveWindow);
373 	}
374 }
375 
376 
377 static inline bool
378 is_writable(tcp_state state)
379 {
380 	return state == ESTABLISHED || state == FINISH_RECEIVED;
381 }
382 
383 
384 static inline bool
385 is_establishing(tcp_state state)
386 {
387 	return state == SYNCHRONIZE_SENT || state == SYNCHRONIZE_RECEIVED;
388 }
389 
390 
391 static inline uint32 tcp_now()
392 {
393 	return system_time() / kTimestampFactor;
394 }
395 
396 
397 static inline uint32 tcp_diff_timestamp(uint32 base)
398 {
399 	uint32 now = tcp_now();
400 
401 	if (now > base)
402 		return now - base;
403 
404 	return now + UINT_MAX - base;
405 }
406 
407 
408 static inline bool
409 state_needs_finish(int32 state)
410 {
411 	return state == WAIT_FOR_FINISH_ACKNOWLEDGE
412 		|| state == FINISH_SENT || state == CLOSING;
413 }
414 
415 
416 //	#pragma mark -
417 
418 
419 TCPEndpoint::TCPEndpoint(net_socket* socket)
420 	:
421 	ProtocolSocket(socket),
422 	fManager(NULL),
423 	fOptions(0),
424 	fSendWindowShift(0),
425 	fReceiveWindowShift(0),
426 	fSendUnacknowledged(0),
427 	fSendNext(0),
428 	fSendMax(0),
429 	fSendUrgentOffset(0),
430 	fSendWindow(0),
431 	fSendMaxWindow(0),
432 	fSendMaxSegmentSize(TCP_DEFAULT_MAX_SEGMENT_SIZE),
433 	fSendMaxSegments(0),
434 	fSendQueue(socket->send.buffer_size),
435 	fInitialSendSequence(0),
436 	fPreviousHighestAcknowledge(0),
437 	fDuplicateAcknowledgeCount(0),
438 	fPreviousFlightSize(0),
439 	fRecover(0),
440 	fRoute(NULL),
441 	fReceiveNext(0),
442 	fReceiveMaxAdvertised(0),
443 	fReceiveWindow(socket->receive.buffer_size),
444 	fReceiveMaxSegmentSize(TCP_DEFAULT_MAX_SEGMENT_SIZE),
445 	fReceiveQueue(socket->receive.buffer_size),
446 	fSmoothedRoundTripTime(0),
447 	fRoundTripVariation(0),
448 	fSendTime(0),
449 	fRoundTripStartSequence(0),
450 	fRetransmitTimeout(TCP_INITIAL_RTT),
451 	fReceivedTimestamp(0),
452 	fCongestionWindow(0),
453 	fSlowStartThreshold(0),
454 	fState(CLOSED),
455 	fFlags(FLAG_OPTION_WINDOW_SCALE | FLAG_OPTION_TIMESTAMP | FLAG_OPTION_SACK_PERMITTED)
456 {
457 	// TODO: to be replaced with a real read/write locking strategy!
458 	mutex_init(&fLock, "tcp lock");
459 
460 	fReceiveCondition.Init(this, "tcp receive");
461 	fSendCondition.Init(this, "tcp send");
462 
463 	gStackModule->init_timer(&fPersistTimer, TCPEndpoint::_PersistTimer, this);
464 	gStackModule->init_timer(&fRetransmitTimer, TCPEndpoint::_RetransmitTimer,
465 		this);
466 	gStackModule->init_timer(&fDelayedAcknowledgeTimer,
467 		TCPEndpoint::_DelayedAcknowledgeTimer, this);
468 	gStackModule->init_timer(&fTimeWaitTimer, TCPEndpoint::_TimeWaitTimer,
469 		this);
470 
471 	T(APICall(this, "constructor"));
472 }
473 
474 
475 TCPEndpoint::~TCPEndpoint()
476 {
477 	mutex_lock(&fLock);
478 
479 	T(APICall(this, "destructor"));
480 
481 	_CancelConnectionTimers();
482 	gStackModule->cancel_timer(&fTimeWaitTimer);
483 	T(TimerSet(this, "time-wait", -1));
484 
485 	if (fManager != NULL) {
486 		fManager->Unbind(this);
487 		put_endpoint_manager(fManager);
488 	}
489 
490 	mutex_destroy(&fLock);
491 
492 	// we need to wait for all timers to return
493 	gStackModule->wait_for_timer(&fRetransmitTimer);
494 	gStackModule->wait_for_timer(&fPersistTimer);
495 	gStackModule->wait_for_timer(&fDelayedAcknowledgeTimer);
496 	gStackModule->wait_for_timer(&fTimeWaitTimer);
497 
498 	gDatalinkModule->put_route(Domain(), fRoute);
499 }
500 
501 
502 status_t
503 TCPEndpoint::InitCheck() const
504 {
505 	return B_OK;
506 }
507 
508 
509 //	#pragma mark - protocol API
510 
511 
512 status_t
513 TCPEndpoint::Open()
514 {
515 	TRACE("Open()");
516 	T(APICall(this, "open"));
517 
518 	status_t status = ProtocolSocket::Open();
519 	if (status < B_OK)
520 		return status;
521 
522 	fManager = get_endpoint_manager(Domain());
523 	if (fManager == NULL)
524 		return EAFNOSUPPORT;
525 
526 	return B_OK;
527 }
528 
529 
530 status_t
531 TCPEndpoint::Close()
532 {
533 	MutexLocker locker(fLock);
534 
535 	TRACE("Close()");
536 	T(APICall(this, "close"));
537 
538 	if (fState == LISTEN)
539 		delete_sem(fAcceptSemaphore);
540 
541 	if (fState == SYNCHRONIZE_SENT || fState == LISTEN) {
542 		// TODO: what about linger in case of SYNCHRONIZE_SENT?
543 		fState = CLOSED;
544 		T(State(this));
545 		return B_OK;
546 	}
547 
548 	// handle linger with zero timeout
549 	if ((socket->options & SO_LINGER) != 0 && socket->linger == 0) {
550 		fState = CLOSED;
551 		T(State(this));
552 		return _SendQueued(true);
553 	}
554 
555 	status_t status = _Disconnect(true);
556 	if (status != B_OK)
557 		return status;
558 
559 	if ((socket->options & SO_LINGER) != 0) {
560 		TRACE("Close(): Lingering for %i secs", socket->linger);
561 
562 		bigtime_t maximum = absolute_timeout(socket->linger * 1000000LL);
563 
564 		while (fSendQueue.Used() > 0) {
565 			status = _WaitForCondition(fSendCondition, locker, maximum);
566 			if (status == B_TIMED_OUT || status == B_WOULD_BLOCK)
567 				break;
568 			else if (status < B_OK)
569 				return status;
570 		}
571 
572 		TRACE("Close(): after waiting, the SendQ was left with %" B_PRIuSIZE
573 			" bytes.", fSendQueue.Used());
574 	}
575 	return B_OK;
576 }
577 
578 
579 void
580 TCPEndpoint::Free()
581 {
582 	MutexLocker _(fLock);
583 
584 	TRACE("Free()");
585 	T(APICall(this, "free"));
586 
587 	if (fState <= SYNCHRONIZE_SENT)
588 		return;
589 
590 	// we are only interested in the timer, not in changing state
591 	_EnterTimeWait();
592 
593 	fFlags |= FLAG_CLOSED;
594 	if ((fFlags & FLAG_DELETE_ON_CLOSE) == 0) {
595 		// we'll be freed later when the 2MSL timer expires
596 		gSocketModule->acquire_socket(socket);
597 	}
598 }
599 
600 
601 /*!	Creates and sends a synchronize packet to /a address, and then waits
602 	until the connection has been established or refused.
603 */
604 status_t
605 TCPEndpoint::Connect(const sockaddr* address)
606 {
607 	if (!AddressModule()->is_same_family(address))
608 		return EAFNOSUPPORT;
609 
610 	MutexLocker locker(fLock);
611 
612 	TRACE("Connect() on address %s", PrintAddress(address));
613 	T(APICall(this, "connect"));
614 
615 	if (gStackModule->is_restarted_syscall()) {
616 		bigtime_t timeout = gStackModule->restore_syscall_restart_timeout();
617 		status_t status = _WaitForEstablished(locker, timeout);
618 		TRACE("  Connect(): Connection complete: %s (timeout was %"
619 			B_PRIdBIGTIME ")", strerror(status), timeout);
620 		return posix_error(status);
621 	}
622 
623 	// Can only call connect() from CLOSED or LISTEN states
624 	// otherwise endpoint is considered already connected
625 	if (fState == LISTEN) {
626 		// this socket is about to connect; remove pending connections in the backlog
627 		gSocketModule->set_max_backlog(socket, 0);
628 	} else if (fState == ESTABLISHED) {
629 		return EISCONN;
630 	} else if (fState != CLOSED)
631 		return EALREADY;
632 
633 	// consider destination address INADDR_ANY as INADDR_LOOPBACK
634 	sockaddr_storage _address;
635 	if (AddressModule()->is_empty_address(address, false)) {
636 		AddressModule()->get_loopback_address((sockaddr *)&_address);
637 		// for IPv4 and IPv6 the port is at the same offset
638 		((sockaddr_in &)_address).sin_port = ((sockaddr_in *)address)->sin_port;
639 		address = (sockaddr *)&_address;
640 	}
641 
642 	status_t status = _PrepareSendPath(address);
643 	if (status < B_OK)
644 		return status;
645 
646 	TRACE("  Connect(): starting 3-way handshake...");
647 
648 	fState = SYNCHRONIZE_SENT;
649 	T(State(this));
650 
651 	// send SYN
652 	status = _SendAcknowledge();
653 	if (status != B_OK) {
654 		_Close();
655 		return status;
656 	}
657 
658 	// If we are running over Loopback, after _SendQueued() returns we
659 	// may be in ESTABLISHED already.
660 	if (fState == ESTABLISHED) {
661 		TRACE("  Connect() completed after _SendQueued()");
662 		return B_OK;
663 	}
664 
665 	// wait until 3-way handshake is complete (if needed)
666 	bigtime_t timeout = min_c(socket->send.timeout, TCP_CONNECTION_TIMEOUT);
667 	if (timeout == 0) {
668 		// we're a non-blocking socket
669 		TRACE("  Connect() delayed, return EINPROGRESS");
670 		return EINPROGRESS;
671 	}
672 
673 	bigtime_t absoluteTimeout = absolute_timeout(timeout);
674 	gStackModule->store_syscall_restart_timeout(absoluteTimeout);
675 
676 	status = _WaitForEstablished(locker, absoluteTimeout);
677 	TRACE("  Connect(): Connection complete: %s (timeout was %" B_PRIdBIGTIME
678 		")", strerror(status), timeout);
679 	return posix_error(status);
680 }
681 
682 
683 status_t
684 TCPEndpoint::Accept(struct net_socket** _acceptedSocket)
685 {
686 	MutexLocker locker(fLock);
687 
688 	TRACE("Accept()");
689 	T(APICall(this, "accept"));
690 
691 	status_t status;
692 	bigtime_t timeout = absolute_timeout(socket->receive.timeout);
693 	if (gStackModule->is_restarted_syscall())
694 		timeout = gStackModule->restore_syscall_restart_timeout();
695 	else
696 		gStackModule->store_syscall_restart_timeout(timeout);
697 
698 	do {
699 		locker.Unlock();
700 
701 		status = acquire_sem_etc(fAcceptSemaphore, 1, B_ABSOLUTE_TIMEOUT
702 			| B_CAN_INTERRUPT, timeout);
703 		if (status != B_OK) {
704 			if (status == B_TIMED_OUT && socket->receive.timeout == 0)
705 				return B_WOULD_BLOCK;
706 
707 			return status;
708 		}
709 
710 		locker.Lock();
711 		status = gSocketModule->dequeue_connected(socket, _acceptedSocket);
712 #ifdef TRACE_TCP
713 		if (status == B_OK)
714 			TRACE("  Accept() returning %p", (*_acceptedSocket)->first_protocol);
715 #endif
716 	} while (status != B_OK);
717 
718 	return status;
719 }
720 
721 
722 status_t
723 TCPEndpoint::Bind(const sockaddr *address)
724 {
725 	if (address == NULL)
726 		return B_BAD_VALUE;
727 
728 	MutexLocker lock(fLock);
729 
730 	TRACE("Bind() on address %s", PrintAddress(address));
731 	T(APICall(this, "bind"));
732 
733 	if (fState != CLOSED)
734 		return EISCONN;
735 
736 	return fManager->Bind(this, address);
737 }
738 
739 
740 status_t
741 TCPEndpoint::Unbind(struct sockaddr *address)
742 {
743 	MutexLocker _(fLock);
744 
745 	TRACE("Unbind()");
746 	T(APICall(this, "unbind"));
747 
748 	return fManager->Unbind(this);
749 }
750 
751 
752 status_t
753 TCPEndpoint::Listen(int count)
754 {
755 	MutexLocker _(fLock);
756 
757 	TRACE("Listen()");
758 	T(APICall(this, "listen"));
759 
760 	if (fState != CLOSED && fState != LISTEN)
761 		return B_BAD_VALUE;
762 
763 	if (fState == CLOSED) {
764 		fAcceptSemaphore = create_sem(0, "tcp accept");
765 		if (fAcceptSemaphore < B_OK)
766 			return ENOBUFS;
767 
768 		status_t status = fManager->SetPassive(this);
769 		if (status != B_OK) {
770 			delete_sem(fAcceptSemaphore);
771 			fAcceptSemaphore = -1;
772 			return status;
773 		}
774 	}
775 
776 	gSocketModule->set_max_backlog(socket, count);
777 
778 	fState = LISTEN;
779 	T(State(this));
780 	return B_OK;
781 }
782 
783 
784 status_t
785 TCPEndpoint::Shutdown(int direction)
786 {
787 	MutexLocker lock(fLock);
788 
789 	TRACE("Shutdown(%i)", direction);
790 	T(APICall(this, "shutdown"));
791 
792 	if (direction == SHUT_RD || direction == SHUT_RDWR)
793 		fFlags |= FLAG_NO_RECEIVE;
794 
795 	if (direction == SHUT_WR || direction == SHUT_RDWR) {
796 		// TODO: That's not correct. After read/write shutting down the socket
797 		// one should still be able to read previously arrived data.
798 		_Disconnect(false);
799 	}
800 
801 	return B_OK;
802 }
803 
804 
805 /*!	Puts data contained in \a buffer into send buffer */
806 status_t
807 TCPEndpoint::SendData(net_buffer *buffer)
808 {
809 	MutexLocker lock(fLock);
810 
811 	TRACE("SendData(buffer %p, size %" B_PRIu32 ", flags %#" B_PRIx32
812 		") [total %" B_PRIuSIZE " bytes, has %" B_PRIuSIZE "]", buffer,
813 		buffer->size, buffer->flags, fSendQueue.Size(), fSendQueue.Free());
814 	T(APICall(this, "senddata"));
815 
816 	const uint32 flags = buffer->flags;
817 	if ((flags & ~(MSG_DONTWAIT | MSG_OOB | MSG_EOF)) != 0)
818 		return EOPNOTSUPP;
819 
820 	if (fState == CLOSED)
821 		return ENOTCONN;
822 	if (fState == LISTEN)
823 		return EDESTADDRREQ;
824 	if (!is_writable(fState) && !is_establishing(fState))
825 		return EPIPE;
826 
827 	size_t left = buffer->size;
828 
829 	bigtime_t timeout = 0;
830 	if ((flags & MSG_DONTWAIT) == 0) {
831 		timeout = absolute_timeout(socket->send.timeout);
832 		if (gStackModule->is_restarted_syscall())
833 			timeout = gStackModule->restore_syscall_restart_timeout();
834 		else
835 			gStackModule->store_syscall_restart_timeout(timeout);
836 	}
837 
838 	while (left > 0) {
839 		while (fSendQueue.Free() < socket->send.low_water_mark) {
840 			// initiate a send before waiting
841 			_SendQueued();
842 
843 			// wait until enough space is available
844 			status_t status = _WaitForCondition(fSendCondition, lock, timeout);
845 			if (status < B_OK) {
846 				TRACE("  SendData() returning %s (%d)",
847 					strerror(posix_error(status)), (int)posix_error(status));
848 				return posix_error(status);
849 			}
850 
851 			if (!is_writable(fState) && !is_establishing(fState))
852 				return EPIPE;
853 		}
854 
855 		size_t size = fSendQueue.Free();
856 		if (size < left) {
857 			// we need to split the original buffer
858 			net_buffer* clone = gBufferModule->clone(buffer, false);
859 				// TODO: add offset/size parameter to net_buffer::clone() or
860 				// even a move_data() function, as this is a bit inefficient
861 			if (clone == NULL)
862 				return ENOBUFS;
863 
864 			status_t status = gBufferModule->trim(clone, size);
865 			if (status != B_OK) {
866 				gBufferModule->free(clone);
867 				return status;
868 			}
869 
870 			gBufferModule->remove_header(buffer, size);
871 			left -= size;
872 			fSendQueue.Add(clone);
873 		} else {
874 			left -= buffer->size;
875 			fSendQueue.Add(buffer);
876 		}
877 	}
878 
879 	TRACE("  SendData(): %" B_PRIuSIZE " bytes used.", fSendQueue.Used());
880 
881 	bool force = false;
882 	if ((flags & MSG_OOB) != 0) {
883 		fSendUrgentOffset = fSendQueue.LastSequence();
884 			// RFC 961 specifies that the urgent offset points to the last
885 			// byte of urgent data. However, this is commonly implemented as
886 			// here, ie. it points to the first byte after the urgent data.
887 		force = true;
888 	}
889 	if ((flags & MSG_EOF) != 0)
890 		_Disconnect(false);
891 
892 	_SendQueued(force);
893 
894 	return B_OK;
895 }
896 
897 
898 ssize_t
899 TCPEndpoint::SendAvailable()
900 {
901 	MutexLocker locker(fLock);
902 
903 	ssize_t available;
904 
905 	if (is_writable(fState))
906 		available = fSendQueue.Free();
907 	else if (is_establishing(fState))
908 		available = 0;
909 	else
910 		available = EPIPE;
911 
912 	TRACE("SendAvailable(): %" B_PRIdSSIZE, available);
913 	T(APICall(this, "sendavailable"));
914 	return available;
915 }
916 
917 
918 status_t
919 TCPEndpoint::FillStat(net_stat *stat)
920 {
921 	MutexLocker _(fLock);
922 
923 	strlcpy(stat->state, name_for_state(fState), sizeof(stat->state));
924 	stat->receive_queue_size = fReceiveQueue.Available();
925 	stat->send_queue_size = fSendQueue.Used();
926 
927 	return B_OK;
928 }
929 
930 
931 status_t
932 TCPEndpoint::ReadData(size_t numBytes, uint32 flags, net_buffer** _buffer)
933 {
934 	if ((flags & ~(MSG_DONTWAIT | MSG_WAITALL | MSG_PEEK)) != 0)
935 		return EOPNOTSUPP;
936 
937 	MutexLocker locker(fLock);
938 
939 	TRACE("ReadData(%" B_PRIuSIZE " bytes, flags %#" B_PRIx32 ")", numBytes,
940 		flags);
941 	T(APICall(this, "readdata"));
942 
943 	*_buffer = NULL;
944 
945 	if (fState == CLOSED) {
946 		if (socket->error != B_OK)
947 			return socket->error;
948 		return ENOTCONN;
949 	}
950 
951 	bigtime_t timeout = 0;
952 	if ((flags & MSG_DONTWAIT) == 0) {
953 		timeout = absolute_timeout(socket->receive.timeout);
954 		if (gStackModule->is_restarted_syscall())
955 			timeout = gStackModule->restore_syscall_restart_timeout();
956 		else
957 			gStackModule->store_syscall_restart_timeout(timeout);
958 	}
959 
960 	if (fState == SYNCHRONIZE_SENT || fState == SYNCHRONIZE_RECEIVED) {
961 		if (flags & MSG_DONTWAIT)
962 			return B_WOULD_BLOCK;
963 
964 		status_t status = _WaitForEstablished(locker, timeout);
965 		if (status < B_OK)
966 			return posix_error(status);
967 	}
968 
969 	size_t dataNeeded = socket->receive.low_water_mark;
970 
971 	// When MSG_WAITALL is set then the function should block
972 	// until the full amount of data can be returned.
973 	if (flags & MSG_WAITALL)
974 		dataNeeded = numBytes;
975 
976 	// TODO: add support for urgent data (MSG_OOB)
977 
978 	while (true) {
979 		if (fState == CLOSING || fState == WAIT_FOR_FINISH_ACKNOWLEDGE
980 			|| fState == TIME_WAIT) {
981 			// ``Connection closing''.
982 			if (fReceiveQueue.Available() > 0)
983 				break;
984 			return B_OK;
985 		}
986 
987 		if (fReceiveQueue.Available() > 0) {
988 			if (fReceiveQueue.Available() >= dataNeeded
989 				|| (fReceiveQueue.PushedData() > 0
990 					&& fReceiveQueue.PushedData() >= fReceiveQueue.Available()))
991 				break;
992 		} else if (fState == FINISH_RECEIVED) {
993 			// ``If no text is awaiting delivery, the RECEIVE will
994 			//   get a Connection closing''.
995 			return B_OK;
996 		}
997 
998 		if (timeout == 0)
999 			return B_WOULD_BLOCK;
1000 
1001 		if ((fFlags & FLAG_NO_RECEIVE) != 0)
1002 			return B_OK;
1003 
1004 		status_t status = _WaitForCondition(fReceiveCondition, locker, timeout);
1005 		if (status < B_OK) {
1006 			// The Open Group base specification mentions that EINTR should be
1007 			// returned if the recv() is interrupted before _any data_ is
1008 			// available. So we actually check if there is data, and if so,
1009 			// push it to the user.
1010 			if ((status == B_TIMED_OUT || status == B_INTERRUPTED)
1011 				&& fReceiveQueue.Available() > 0)
1012 				break;
1013 
1014 			return posix_error(status);
1015 		}
1016 	}
1017 
1018 	TRACE("  ReadData(): %" B_PRIuSIZE " are available.",
1019 		fReceiveQueue.Available());
1020 
1021 	if (numBytes < fReceiveQueue.Available())
1022 		fReceiveCondition.NotifyAll();
1023 
1024 	bool clone = (flags & MSG_PEEK) != 0;
1025 
1026 	ssize_t receivedBytes = fReceiveQueue.Get(numBytes, !clone, _buffer);
1027 
1028 	TRACE("  ReadData(): %" B_PRIuSIZE " bytes kept.",
1029 		fReceiveQueue.Available());
1030 
1031 	if (fReceiveQueue.Available() == 0 && fState == FINISH_RECEIVED)
1032 		socket->receive.low_water_mark = 0;
1033 
1034 	// if we are opening the window, check if we should send an ACK
1035 	if (!clone)
1036 		_SendAcknowledge();
1037 
1038 	return receivedBytes;
1039 }
1040 
1041 
1042 ssize_t
1043 TCPEndpoint::ReadAvailable()
1044 {
1045 	MutexLocker locker(fLock);
1046 
1047 	TRACE("ReadAvailable(): %" B_PRIdSSIZE, _AvailableData());
1048 	T(APICall(this, "readavailable"));
1049 
1050 	return _AvailableData();
1051 }
1052 
1053 
1054 status_t
1055 TCPEndpoint::SetSendBufferSize(size_t length)
1056 {
1057 	MutexLocker _(fLock);
1058 	fSendQueue.SetMaxBytes(length);
1059 	return B_OK;
1060 }
1061 
1062 
1063 status_t
1064 TCPEndpoint::SetReceiveBufferSize(size_t length)
1065 {
1066 	MutexLocker _(fLock);
1067 	fReceiveQueue.SetMaxBytes(length);
1068 	return B_OK;
1069 }
1070 
1071 
1072 status_t
1073 TCPEndpoint::GetOption(int option, void* _value, int* _length)
1074 {
1075 	if (*_length != sizeof(int))
1076 		return B_BAD_VALUE;
1077 
1078 	int* value = (int*)_value;
1079 
1080 	switch (option) {
1081 		case TCP_NODELAY:
1082 			if ((fOptions & TCP_NODELAY) != 0)
1083 				*value = 1;
1084 			else
1085 				*value = 0;
1086 			return B_OK;
1087 
1088 		case TCP_MAXSEG:
1089 			*value = fReceiveMaxSegmentSize;
1090 			return B_OK;
1091 
1092 		default:
1093 			return B_BAD_VALUE;
1094 	}
1095 }
1096 
1097 
1098 status_t
1099 TCPEndpoint::SetOption(int option, const void* _value, int length)
1100 {
1101 	if (option != TCP_NODELAY)
1102 		return B_BAD_VALUE;
1103 
1104 	if (length != sizeof(int))
1105 		return B_BAD_VALUE;
1106 
1107 	const int* value = (const int*)_value;
1108 
1109 	MutexLocker _(fLock);
1110 	if (*value)
1111 		fOptions |= TCP_NODELAY;
1112 	else
1113 		fOptions &= ~TCP_NODELAY;
1114 
1115 	return B_OK;
1116 }
1117 
1118 
1119 //	#pragma mark - misc
1120 
1121 
1122 bool
1123 TCPEndpoint::IsBound() const
1124 {
1125 	return !LocalAddress().IsEmpty(true);
1126 }
1127 
1128 
1129 bool
1130 TCPEndpoint::IsLocal() const
1131 {
1132 	return (fFlags & FLAG_LOCAL) != 0;
1133 }
1134 
1135 
1136 status_t
1137 TCPEndpoint::DelayedAcknowledge()
1138 {
1139 	// ACKs "MUST" be generated within 500ms of the first unACKed packet, and
1140 	// "SHOULD" be for at least every second full-size segment. (RFC 5681 § 4.2)
1141 
1142 	bigtime_t delay = TCP_DELAYED_ACKNOWLEDGE_TIMEOUT;
1143 	if ((fReceiveNext - fLastAcknowledgeSent) >= (fReceiveMaxSegmentSize * 2)) {
1144 		// Trigger an immediate timeout rather than invoking Send directly,
1145 		// allowing multiple invocations to be coalesced.
1146 		delay = 0;
1147 	} else if (gStackModule->is_timer_active(&fDelayedAcknowledgeTimer)) {
1148 		return B_OK;
1149 	}
1150 
1151 	gStackModule->set_timer(&fDelayedAcknowledgeTimer, delay);
1152 	T(TimerSet(this, "delayed ack", TCP_DELAYED_ACKNOWLEDGE_TIMEOUT));
1153 	return B_OK;
1154 }
1155 
1156 
1157 void
1158 TCPEndpoint::_StartPersistTimer()
1159 {
1160 	gStackModule->set_timer(&fPersistTimer, TCP_PERSIST_TIMEOUT);
1161 	T(TimerSet(this, "persist", TCP_PERSIST_TIMEOUT));
1162 }
1163 
1164 
1165 void
1166 TCPEndpoint::_EnterTimeWait()
1167 {
1168 	TRACE("_EnterTimeWait()");
1169 
1170 	if (fState == TIME_WAIT) {
1171 		_CancelConnectionTimers();
1172 	}
1173 
1174 	_UpdateTimeWait();
1175 }
1176 
1177 
1178 void
1179 TCPEndpoint::_UpdateTimeWait()
1180 {
1181 	gStackModule->set_timer(&fTimeWaitTimer, TCP_MAX_SEGMENT_LIFETIME << 1);
1182 	T(TimerSet(this, "time-wait", TCP_MAX_SEGMENT_LIFETIME << 1));
1183 }
1184 
1185 
1186 void
1187 TCPEndpoint::_CancelConnectionTimers()
1188 {
1189 	gStackModule->cancel_timer(&fRetransmitTimer);
1190 	T(TimerSet(this, "retransmit", -1));
1191 	gStackModule->cancel_timer(&fPersistTimer);
1192 	T(TimerSet(this, "persist", -1));
1193 	gStackModule->cancel_timer(&fDelayedAcknowledgeTimer);
1194 	T(TimerSet(this, "delayed ack", -1));
1195 }
1196 
1197 
1198 /*!	Sends the FIN flag to the peer when the connection is still open.
1199 	Moves the endpoint to the next state depending on where it was.
1200 */
1201 status_t
1202 TCPEndpoint::_Disconnect(bool closing)
1203 {
1204 	tcp_state previousState = fState;
1205 
1206 	if (fState == SYNCHRONIZE_RECEIVED || fState == ESTABLISHED)
1207 		fState = FINISH_SENT;
1208 	else if (fState == FINISH_RECEIVED)
1209 		fState = WAIT_FOR_FINISH_ACKNOWLEDGE;
1210 	else
1211 		return B_OK;
1212 
1213 	T(State(this));
1214 
1215 	status_t status = _SendQueued();
1216 	if (status != B_OK) {
1217 		fState = previousState;
1218 		T(State(this));
1219 		return status;
1220 	}
1221 
1222 	return B_OK;
1223 }
1224 
1225 
1226 void
1227 TCPEndpoint::_MarkEstablished()
1228 {
1229 	fState = ESTABLISHED;
1230 	T(State(this));
1231 
1232 	gSocketModule->set_connected(socket);
1233 	if (gSocketModule->has_parent(socket))
1234 		release_sem_etc(fAcceptSemaphore, 1, B_DO_NOT_RESCHEDULE);
1235 
1236 	fSendCondition.NotifyAll();
1237 	gSocketModule->notify(socket, B_SELECT_WRITE, fSendQueue.Free());
1238 }
1239 
1240 
1241 status_t
1242 TCPEndpoint::_WaitForEstablished(MutexLocker &locker, bigtime_t timeout)
1243 {
1244 	// TODO: Checking for CLOSED seems correct, but breaks several neon tests.
1245 	// When investigating this, also have a look at _Close() and _HandleReset().
1246 	while (fState < ESTABLISHED/* && fState != CLOSED*/) {
1247 		if (socket->error != B_OK)
1248 			return socket->error;
1249 
1250 		status_t status = _WaitForCondition(fSendCondition, locker, timeout);
1251 		if (status < B_OK)
1252 			return status;
1253 	}
1254 
1255 	return B_OK;
1256 }
1257 
1258 
1259 //	#pragma mark - receive
1260 
1261 
1262 void
1263 TCPEndpoint::_Close()
1264 {
1265 	_CancelConnectionTimers();
1266 	fState = CLOSED;
1267 	T(State(this));
1268 
1269 	fFlags |= FLAG_DELETE_ON_CLOSE;
1270 
1271 	fSendCondition.NotifyAll();
1272 	_NotifyReader();
1273 
1274 	if (gSocketModule->has_parent(socket)) {
1275 		// We still have a parent - obviously, we haven't been accepted yet,
1276 		// so no one could ever close us.
1277 		_CancelConnectionTimers();
1278 		gSocketModule->set_aborted(socket);
1279 	}
1280 }
1281 
1282 
1283 void
1284 TCPEndpoint::_HandleReset(status_t error)
1285 {
1286 	socket->error = error;
1287 	_Close();
1288 
1289 	gSocketModule->notify(socket, B_SELECT_WRITE, error);
1290 	gSocketModule->notify(socket, B_SELECT_ERROR, error);
1291 }
1292 
1293 
1294 void
1295 TCPEndpoint::_DuplicateAcknowledge(tcp_segment_header &segment)
1296 {
1297 	if (fDuplicateAcknowledgeCount == 0)
1298 		fPreviousFlightSize = (fSendMax - fSendUnacknowledged).Number();
1299 
1300 	if (++fDuplicateAcknowledgeCount < 3) {
1301 		if (fSendQueue.Available(fSendMax) != 0  && fSendWindow != 0) {
1302 			fSendNext = fSendMax;
1303 			fCongestionWindow += fDuplicateAcknowledgeCount * fSendMaxSegmentSize;
1304 			_SendQueued();
1305 			TRACE("_DuplicateAcknowledge(): packet sent under limited transmit on receipt of dup ack");
1306 			fCongestionWindow -= fDuplicateAcknowledgeCount * fSendMaxSegmentSize;
1307 		}
1308 	}
1309 
1310 	if (fDuplicateAcknowledgeCount == 3) {
1311 		if ((segment.acknowledge - 1) > fRecover || (fCongestionWindow > fSendMaxSegmentSize &&
1312 			(fSendUnacknowledged - fPreviousHighestAcknowledge) <= 4 * fSendMaxSegmentSize)) {
1313 			fFlags |= FLAG_RECOVERY;
1314 			fRecover = fSendMax.Number() - 1;
1315 			fSlowStartThreshold = max_c(fPreviousFlightSize / 2, 2 * fSendMaxSegmentSize);
1316 			fCongestionWindow = fSlowStartThreshold + 3 * fSendMaxSegmentSize;
1317 			fSendNext = segment.acknowledge;
1318 			_SendQueued();
1319 			TRACE("_DuplicateAcknowledge(): packet sent under fast restransmit on the receipt of 3rd dup ack");
1320 		}
1321 	} else if (fDuplicateAcknowledgeCount > 3) {
1322 		uint32 flightSize = (fSendMax - fSendUnacknowledged).Number();
1323 		if ((fDuplicateAcknowledgeCount - 3) * fSendMaxSegmentSize <= flightSize)
1324 			fCongestionWindow += fSendMaxSegmentSize;
1325 		if (fSendQueue.Available(fSendMax) != 0) {
1326 			fSendNext = fSendMax;
1327 			_SendQueued();
1328 		}
1329 	}
1330 }
1331 
1332 
1333 void
1334 TCPEndpoint::_UpdateTimestamps(tcp_segment_header& segment,
1335 	size_t segmentLength)
1336 {
1337 	if (fFlags & FLAG_OPTION_TIMESTAMP) {
1338 		tcp_sequence sequence(segment.sequence);
1339 
1340 		if (fLastAcknowledgeSent >= sequence
1341 			&& fLastAcknowledgeSent < (sequence + segmentLength))
1342 			fReceivedTimestamp = segment.timestamp_value;
1343 	}
1344 }
1345 
1346 
1347 ssize_t
1348 TCPEndpoint::_AvailableData() const
1349 {
1350 	// TODO: Refer to the FLAG_NO_RECEIVE comment above regarding
1351 	//       the application of FLAG_NO_RECEIVE in listen()ing
1352 	//       sockets.
1353 	if (fState == LISTEN)
1354 		return gSocketModule->count_connected(socket);
1355 	if (fState == SYNCHRONIZE_SENT)
1356 		return 0;
1357 
1358 	ssize_t availableData = fReceiveQueue.Available();
1359 
1360 	if (availableData == 0 && !_ShouldReceive())
1361 		return ENOTCONN;
1362 	if (availableData == 0 && (fState == FINISH_RECEIVED || fState == WAIT_FOR_FINISH_ACKNOWLEDGE))
1363 		return ESHUTDOWN;
1364 	return availableData;
1365 }
1366 
1367 
1368 void
1369 TCPEndpoint::_NotifyReader()
1370 {
1371 	fReceiveCondition.NotifyAll();
1372 	gSocketModule->notify(socket, B_SELECT_READ, _AvailableData());
1373 }
1374 
1375 
1376 bool
1377 TCPEndpoint::_AddData(tcp_segment_header& segment, net_buffer* buffer)
1378 {
1379 	if ((segment.flags & TCP_FLAG_FINISH) != 0) {
1380 		// Remember the position of the finish received flag
1381 		fFinishReceived = true;
1382 		fFinishReceivedAt = segment.sequence + buffer->size;
1383 	}
1384 
1385 	fReceiveQueue.Add(buffer, segment.sequence);
1386 	fReceiveNext = fReceiveQueue.NextSequence();
1387 
1388 	if (fFinishReceived) {
1389 		// Set or reset the finish flag on the current segment
1390 		if (fReceiveNext < fFinishReceivedAt)
1391 			segment.flags &= ~TCP_FLAG_FINISH;
1392 		else
1393 			segment.flags |= TCP_FLAG_FINISH;
1394 	}
1395 
1396 	TRACE("  _AddData(): adding data, receive next = %" B_PRIu32 ". Now have %"
1397 		B_PRIuSIZE " bytes.", fReceiveNext.Number(), fReceiveQueue.Available());
1398 
1399 	if ((segment.flags & TCP_FLAG_PUSH) != 0)
1400 		fReceiveQueue.SetPushPointer();
1401 
1402 	return fReceiveQueue.Available() > 0;
1403 }
1404 
1405 
1406 void
1407 TCPEndpoint::_PrepareReceivePath(tcp_segment_header& segment)
1408 {
1409 	fInitialReceiveSequence = segment.sequence;
1410 	fFinishReceived = false;
1411 
1412 	// count the received SYN
1413 	segment.sequence++;
1414 
1415 	fReceiveNext = segment.sequence;
1416 	fReceiveQueue.SetInitialSequence(segment.sequence);
1417 
1418 	if ((fOptions & TCP_NOOPT) == 0) {
1419 		if (segment.max_segment_size > 0) {
1420 			// The maximum size of a segment that a TCP endpoint really sends,
1421 			// the "effective send MSS", MUST be the smaller of the send MSS and
1422 			// the largest transmission size permitted by the IP layer:
1423 			fSendMaxSegmentSize = min_c(segment.max_segment_size,
1424 				_MaxSegmentSize(*PeerAddress()));
1425 		}
1426 
1427 		if (segment.options & TCP_HAS_WINDOW_SCALE) {
1428 			fFlags |= FLAG_OPTION_WINDOW_SCALE;
1429 			fSendWindowShift = segment.window_shift;
1430 		} else {
1431 			fFlags &= ~FLAG_OPTION_WINDOW_SCALE;
1432 			fReceiveWindowShift = 0;
1433 		}
1434 
1435 		if (segment.options & TCP_HAS_TIMESTAMPS) {
1436 			fFlags |= FLAG_OPTION_TIMESTAMP;
1437 			fReceivedTimestamp = segment.timestamp_value;
1438 		} else
1439 			fFlags &= ~FLAG_OPTION_TIMESTAMP;
1440 
1441 		if ((segment.options & TCP_SACK_PERMITTED) == 0)
1442 			fFlags &= ~FLAG_OPTION_SACK_PERMITTED;
1443 	}
1444 
1445 	if (fSendMaxSegmentSize > 2190)
1446 		fCongestionWindow = 2 * fSendMaxSegmentSize;
1447 	else if (fSendMaxSegmentSize > 1095)
1448 		fCongestionWindow = 3 * fSendMaxSegmentSize;
1449 	else
1450 		fCongestionWindow = 4 * fSendMaxSegmentSize;
1451 
1452 	fSendMaxSegments = fCongestionWindow / fSendMaxSegmentSize;
1453 	fSlowStartThreshold = (uint32)segment.advertised_window << fSendWindowShift;
1454 }
1455 
1456 
1457 bool
1458 TCPEndpoint::_ShouldReceive() const
1459 {
1460 	if ((fFlags & FLAG_NO_RECEIVE) != 0)
1461 		return false;
1462 
1463 	return fState == ESTABLISHED || fState == FINISH_SENT
1464 		|| fState == FINISH_ACKNOWLEDGED || fState == FINISH_RECEIVED;
1465 }
1466 
1467 
1468 int32
1469 TCPEndpoint::_Spawn(TCPEndpoint* parent, tcp_segment_header& segment,
1470 	net_buffer* buffer)
1471 {
1472 	MutexLocker _(fLock);
1473 
1474 	// TODO error checking
1475 	ProtocolSocket::Open();
1476 
1477 	fState = SYNCHRONIZE_RECEIVED;
1478 	T(Spawn(parent, this));
1479 
1480 	fManager = parent->fManager;
1481 
1482 	LocalAddress().SetTo(buffer->destination);
1483 	PeerAddress().SetTo(buffer->source);
1484 
1485 	TRACE("Spawn()");
1486 
1487 	// TODO: proper error handling!
1488 	if (fManager->BindChild(this) != B_OK) {
1489 		T(Error(this, "binding failed", __LINE__));
1490 		return DROP;
1491 	}
1492 	if (_PrepareSendPath(*PeerAddress()) != B_OK) {
1493 		T(Error(this, "prepare send faild", __LINE__));
1494 		return DROP;
1495 	}
1496 
1497 	fOptions = parent->fOptions;
1498 	fAcceptSemaphore = parent->fAcceptSemaphore;
1499 
1500 	_PrepareReceivePath(segment);
1501 
1502 	// send SYN+ACK
1503 	if (_SendAcknowledge() != B_OK) {
1504 		T(Error(this, "sending failed", __LINE__));
1505 		return DROP;
1506 	}
1507 
1508 	segment.flags &= ~TCP_FLAG_SYNCHRONIZE;
1509 		// we handled this flag now, it must not be set for further processing
1510 
1511 	return _Receive(segment, buffer);
1512 }
1513 
1514 
1515 int32
1516 TCPEndpoint::_ListenReceive(tcp_segment_header& segment, net_buffer* buffer)
1517 {
1518 	TRACE("ListenReceive()");
1519 
1520 	// Essentially, we accept only TCP_FLAG_SYNCHRONIZE in this state,
1521 	// but the error behaviour differs
1522 	if (segment.flags & TCP_FLAG_RESET)
1523 		return DROP;
1524 	if (segment.flags & TCP_FLAG_ACKNOWLEDGE)
1525 		return DROP | RESET;
1526 	if ((segment.flags & TCP_FLAG_SYNCHRONIZE) == 0)
1527 		return DROP;
1528 
1529 	// TODO: drop broadcast/multicast
1530 
1531 	// spawn new endpoint for accept()
1532 	net_socket* newSocket;
1533 	if (gSocketModule->spawn_pending_socket(socket, &newSocket) < B_OK) {
1534 		T(Error(this, "spawning failed", __LINE__));
1535 		return DROP;
1536 	}
1537 
1538 	return ((TCPEndpoint *)newSocket->first_protocol)->_Spawn(this,
1539 		segment, buffer);
1540 }
1541 
1542 
1543 int32
1544 TCPEndpoint::_SynchronizeSentReceive(tcp_segment_header &segment,
1545 	net_buffer *buffer)
1546 {
1547 	TRACE("_SynchronizeSentReceive()");
1548 
1549 	if ((segment.flags & TCP_FLAG_ACKNOWLEDGE) != 0
1550 		&& (fInitialSendSequence >= segment.acknowledge
1551 			|| fSendMax < segment.acknowledge))
1552 		return DROP | RESET;
1553 
1554 	if (segment.flags & TCP_FLAG_RESET) {
1555 		_HandleReset(ECONNREFUSED);
1556 		return DROP;
1557 	}
1558 
1559 	if ((segment.flags & TCP_FLAG_SYNCHRONIZE) == 0)
1560 		return DROP;
1561 
1562 	fSendUnacknowledged = segment.acknowledge;
1563 	_PrepareReceivePath(segment);
1564 
1565 	if (segment.flags & TCP_FLAG_ACKNOWLEDGE) {
1566 		_MarkEstablished();
1567 	} else {
1568 		// simultaneous open
1569 		fState = SYNCHRONIZE_RECEIVED;
1570 		T(State(this));
1571 	}
1572 
1573 	segment.flags &= ~TCP_FLAG_SYNCHRONIZE;
1574 		// we handled this flag now, it must not be set for further processing
1575 
1576 	return _Receive(segment, buffer) | IMMEDIATE_ACKNOWLEDGE;
1577 }
1578 
1579 
1580 int32
1581 TCPEndpoint::_Receive(tcp_segment_header& segment, net_buffer* buffer)
1582 {
1583 	// PAWS processing takes precedence over regular TCP acceptability check
1584 	if ((fFlags & FLAG_OPTION_TIMESTAMP) != 0 && (segment.flags & TCP_FLAG_RESET) == 0) {
1585 		if ((segment.options & TCP_HAS_TIMESTAMPS) == 0)
1586 			return DROP;
1587 		if ((int32)(fReceivedTimestamp - segment.timestamp_value) > 0
1588 			&& (fReceivedTimestamp - segment.timestamp_value) <= INT32_MAX)
1589 			return DROP | IMMEDIATE_ACKNOWLEDGE;
1590 	}
1591 
1592 	uint32 advertisedWindow = segment.AdvertisedWindow(fSendWindowShift);
1593 	size_t segmentLength = buffer->size;
1594 
1595 	// First, handle the most common case for uni-directional data transfer
1596 	// (known as header prediction - the segment must not change the window,
1597 	// and must be the expected sequence, and contain no control flags)
1598 
1599 	if (fState == ESTABLISHED
1600 		&& segment.AcknowledgeOnly()
1601 		&& fReceiveNext == segment.sequence
1602 		&& advertisedWindow > 0 && advertisedWindow == fSendWindow
1603 		&& fSendNext == fSendMax) {
1604 		_UpdateTimestamps(segment, segmentLength);
1605 
1606 		if (segmentLength == 0) {
1607 			// this is a pure acknowledge segment - we're on the sending end
1608 			if (fSendUnacknowledged < segment.acknowledge
1609 				&& fSendMax >= segment.acknowledge) {
1610 				_Acknowledged(segment);
1611 				return DROP;
1612 			}
1613 		} else if (segment.acknowledge == fSendUnacknowledged
1614 			&& fReceiveQueue.IsContiguous()
1615 			&& fReceiveQueue.Free() >= segmentLength
1616 			&& (fFlags & FLAG_NO_RECEIVE) == 0) {
1617 			if (_AddData(segment, buffer))
1618 				_NotifyReader();
1619 
1620 			return KEEP | ((segment.flags & TCP_FLAG_PUSH) != 0
1621 				? IMMEDIATE_ACKNOWLEDGE : ACKNOWLEDGE);
1622 		}
1623 	}
1624 
1625 	// The fast path was not applicable, so we continue with the standard
1626 	// processing of the incoming segment
1627 
1628 	ASSERT(fState != SYNCHRONIZE_SENT && fState != LISTEN);
1629 
1630 	if (fState != CLOSED && fState != TIME_WAIT) {
1631 		// Check sequence number
1632 		if (!segment_in_sequence(segment, segmentLength, fReceiveNext,
1633 				fReceiveWindow)) {
1634 			TRACE("  Receive(): segment out of window, next: %" B_PRIu32
1635 				" wnd: %" B_PRIu32, fReceiveNext.Number(), fReceiveWindow);
1636 			if ((segment.flags & TCP_FLAG_RESET) != 0) {
1637 				// TODO: this doesn't look right - review!
1638 				return DROP;
1639 			}
1640 			return DROP | IMMEDIATE_ACKNOWLEDGE;
1641 		}
1642 	}
1643 
1644 	if ((segment.flags & TCP_FLAG_RESET) != 0) {
1645 		// Is this a valid reset?
1646 		// We generally ignore resets in time wait state (see RFC 1337)
1647 		if (fLastAcknowledgeSent <= segment.sequence
1648 			&& tcp_sequence(segment.sequence) < (fLastAcknowledgeSent
1649 				+ fReceiveWindow)
1650 			&& fState != TIME_WAIT) {
1651 			status_t error;
1652 			if (fState == SYNCHRONIZE_RECEIVED)
1653 				error = ECONNREFUSED;
1654 			else if (fState == CLOSING || fState == WAIT_FOR_FINISH_ACKNOWLEDGE)
1655 				error = ENOTCONN;
1656 			else
1657 				error = ECONNRESET;
1658 
1659 			_HandleReset(error);
1660 		}
1661 
1662 		return DROP;
1663 	}
1664 
1665 	if ((segment.flags & TCP_FLAG_SYNCHRONIZE) != 0
1666 		|| (fState == SYNCHRONIZE_RECEIVED
1667 			&& (fInitialReceiveSequence > segment.sequence
1668 				|| ((segment.flags & TCP_FLAG_ACKNOWLEDGE) != 0
1669 					&& (fSendUnacknowledged > segment.acknowledge
1670 						|| fSendMax < segment.acknowledge))))) {
1671 		// reset the connection - either the initial SYN was faulty, or we
1672 		// received a SYN within the data stream
1673 		return DROP | RESET;
1674 	}
1675 
1676 	// TODO: Check this! Why do we advertize a window outside of what we should
1677 	// buffer?
1678 	fReceiveWindow = max_c(fReceiveQueue.Free(), fReceiveWindow);
1679 		// the window must not shrink
1680 
1681 	// trim buffer to be within the receive window
1682 	int32 drop = (int32)(fReceiveNext - segment.sequence).Number();
1683 	if (drop > 0) {
1684 		if ((uint32)drop > buffer->size
1685 			|| ((uint32)drop == buffer->size
1686 				&& (segment.flags & TCP_FLAG_FINISH) == 0)) {
1687 			// don't accidently remove a FIN we shouldn't remove
1688 			segment.flags &= ~TCP_FLAG_FINISH;
1689 			drop = buffer->size;
1690 		}
1691 
1692 		// remove duplicate data at the start
1693 		TRACE("* remove %" B_PRId32 " bytes from the start", drop);
1694 		gBufferModule->remove_header(buffer, drop);
1695 		segment.sequence += drop;
1696 	}
1697 
1698 	int32 action = KEEP;
1699 
1700 	// immediately acknowledge out-of-order segment to trigger fast-retransmit at the sender
1701 	if (drop != 0)
1702 		action |= IMMEDIATE_ACKNOWLEDGE;
1703 
1704 	drop = (int32)(segment.sequence + buffer->size
1705 		- (fReceiveNext + fReceiveWindow)).Number();
1706 	if (drop > 0) {
1707 		// remove data exceeding our window
1708 		if ((uint32)drop >= buffer->size) {
1709 			// if we can accept data, or the segment is not what we'd expect,
1710 			// drop the segment (an immediate acknowledge is always triggered)
1711 			if (fReceiveWindow != 0 || segment.sequence != fReceiveNext)
1712 				return DROP | IMMEDIATE_ACKNOWLEDGE;
1713 
1714 			action |= IMMEDIATE_ACKNOWLEDGE;
1715 		}
1716 
1717 		if ((segment.flags & TCP_FLAG_FINISH) != 0) {
1718 			// we need to remove the finish, too, as part of the data
1719 			drop--;
1720 		}
1721 
1722 		segment.flags &= ~(TCP_FLAG_FINISH | TCP_FLAG_PUSH);
1723 		TRACE("* remove %" B_PRId32 " bytes from the end", drop);
1724 		gBufferModule->remove_trailer(buffer, drop);
1725 	}
1726 
1727 #ifdef TRACE_TCP
1728 	if (advertisedWindow > fSendWindow) {
1729 		TRACE("  Receive(): Window update %" B_PRIu32 " -> %" B_PRIu32,
1730 			fSendWindow, advertisedWindow);
1731 	}
1732 #endif
1733 
1734 	if (fSendWindow < fSendMaxSegmentSize
1735 			&& advertisedWindow >= fSendMaxSegmentSize) {
1736 		// Our current send window is less than a segment wide, and the new one
1737 		// is larger, so trigger a send in case there's anything to be sent.
1738 		action |= SEND_QUEUED;
1739 	}
1740 
1741 	fSendWindow = advertisedWindow;
1742 	if (advertisedWindow > fSendMaxWindow)
1743 		fSendMaxWindow = advertisedWindow;
1744 
1745 	// Then look at the acknowledgement for any updates
1746 
1747 	if ((segment.flags & TCP_FLAG_ACKNOWLEDGE) != 0) {
1748 		// process acknowledged data
1749 		if (fState == SYNCHRONIZE_RECEIVED)
1750 			_MarkEstablished();
1751 
1752 		if (fSendMax < segment.acknowledge)
1753 			return DROP | IMMEDIATE_ACKNOWLEDGE;
1754 
1755 		if (segment.acknowledge == fSendUnacknowledged) {
1756 			if (buffer->size == 0 && advertisedWindow == fSendWindow
1757 				&& (segment.flags & TCP_FLAG_FINISH) == 0 && fSendUnacknowledged != fSendMax) {
1758 				TRACE("Receive(): duplicate ack!");
1759 				_DuplicateAcknowledge(segment);
1760 			}
1761 		} else if (segment.acknowledge < fSendUnacknowledged) {
1762 			return DROP;
1763 		} else {
1764 			// this segment acknowledges in flight data
1765 
1766 			if (fDuplicateAcknowledgeCount >= 3) {
1767 				// deflate the window.
1768 				if (segment.acknowledge > fRecover) {
1769 					uint32 flightSize = (fSendMax - fSendUnacknowledged).Number();
1770 					fCongestionWindow = min_c(fSlowStartThreshold,
1771 						max_c(flightSize, fSendMaxSegmentSize) + fSendMaxSegmentSize);
1772 					fFlags &= ~FLAG_RECOVERY;
1773 				}
1774 			}
1775 
1776 			if (fSendMax == segment.acknowledge)
1777 				TRACE("Receive(): all inflight data ack'd!");
1778 
1779 			if (segment.acknowledge > fSendQueue.LastSequence()
1780 					&& fState > ESTABLISHED) {
1781 				TRACE("Receive(): FIN has been acknowledged!");
1782 
1783 				switch (fState) {
1784 					case FINISH_SENT:
1785 						fState = FINISH_ACKNOWLEDGED;
1786 						T(State(this));
1787 						break;
1788 					case CLOSING:
1789 						fState = TIME_WAIT;
1790 						T(State(this));
1791 						_EnterTimeWait();
1792 						return DROP;
1793 					case WAIT_FOR_FINISH_ACKNOWLEDGE:
1794 						_Close();
1795 						break;
1796 
1797 					default:
1798 						break;
1799 				}
1800 			}
1801 
1802 			if (fState != CLOSED) {
1803 				tcp_sequence last = fLastAcknowledgeSent;
1804 				_Acknowledged(segment);
1805 				// we just sent an acknowledge, remove from action
1806 				if (last < fLastAcknowledgeSent)
1807 					action &= ~IMMEDIATE_ACKNOWLEDGE;
1808 			}
1809 		}
1810 	}
1811 
1812 	if (segment.flags & TCP_FLAG_URGENT) {
1813 		if (fState == ESTABLISHED || fState == FINISH_SENT
1814 			|| fState == FINISH_ACKNOWLEDGED) {
1815 			// TODO: Handle urgent data:
1816 			//  - RCV.UP <- max(RCV.UP, SEG.UP)
1817 			//  - signal the user that urgent data is available (SIGURG)
1818 		}
1819 	}
1820 
1821 	bool notify = false;
1822 
1823 	// The buffer may be freed if its data is added to the queue, so cache
1824 	// the size as we still need it later.
1825 	const uint32 bufferSize = buffer->size;
1826 
1827 	if ((bufferSize > 0 || (segment.flags & TCP_FLAG_FINISH) != 0)
1828 		&& _ShouldReceive())
1829 		notify = _AddData(segment, buffer);
1830 	else {
1831 		if ((fFlags & FLAG_NO_RECEIVE) != 0)
1832 			fReceiveNext += buffer->size;
1833 
1834 		action = (action & ~KEEP) | DROP;
1835 	}
1836 
1837 	if ((segment.flags & TCP_FLAG_FINISH) != 0) {
1838 		segmentLength++;
1839 		if (fState != CLOSED && fState != LISTEN && fState != SYNCHRONIZE_SENT) {
1840 			TRACE("Receive(): peer is finishing connection!");
1841 			fReceiveNext++;
1842 			notify = true;
1843 
1844 			// FIN implies PUSH
1845 			fReceiveQueue.SetPushPointer();
1846 
1847 			// we'll reply immediately to the FIN if we are not
1848 			// transitioning to TIME WAIT so we immediatly ACK it.
1849 			action |= IMMEDIATE_ACKNOWLEDGE;
1850 
1851 			// other side is closing connection; change states
1852 			switch (fState) {
1853 				case ESTABLISHED:
1854 				case SYNCHRONIZE_RECEIVED:
1855 					fState = FINISH_RECEIVED;
1856 					T(State(this));
1857 					break;
1858 				case FINISH_SENT:
1859 					// simultaneous close
1860 					fState = CLOSING;
1861 					T(State(this));
1862 					break;
1863 				case FINISH_ACKNOWLEDGED:
1864 					fState = TIME_WAIT;
1865 					T(State(this));
1866 					_EnterTimeWait();
1867 					break;
1868 				case TIME_WAIT:
1869 					_UpdateTimeWait();
1870 					break;
1871 
1872 				default:
1873 					break;
1874 			}
1875 		}
1876 	}
1877 
1878 	if (notify)
1879 		_NotifyReader();
1880 
1881 	if (bufferSize > 0 || (segment.flags & TCP_FLAG_SYNCHRONIZE) != 0)
1882 		action |= ACKNOWLEDGE;
1883 
1884 	_UpdateTimestamps(segment, segmentLength);
1885 
1886 	TRACE("Receive() Action %" B_PRId32, action);
1887 
1888 	return action;
1889 }
1890 
1891 
1892 int32
1893 TCPEndpoint::SegmentReceived(tcp_segment_header& segment, net_buffer* buffer)
1894 {
1895 	MutexLocker locker(fLock);
1896 
1897 	TRACE("SegmentReceived(): buffer %p (%" B_PRIu32 " bytes) address %s "
1898 		"to %s flags %#" B_PRIx8 ", seq %" B_PRIu32 ", ack %" B_PRIu32
1899 		", wnd %" B_PRIu32, buffer, buffer->size, PrintAddress(buffer->source),
1900 		PrintAddress(buffer->destination), segment.flags, segment.sequence,
1901 		segment.acknowledge,
1902 		(uint32)segment.advertised_window << fSendWindowShift);
1903 	T(Receive(this, segment,
1904 		(uint32)segment.advertised_window << fSendWindowShift, buffer));
1905 	int32 segmentAction = DROP;
1906 
1907 	switch (fState) {
1908 		case LISTEN:
1909 			segmentAction = _ListenReceive(segment, buffer);
1910 			break;
1911 
1912 		case SYNCHRONIZE_SENT:
1913 			segmentAction = _SynchronizeSentReceive(segment, buffer);
1914 			break;
1915 
1916 		case SYNCHRONIZE_RECEIVED:
1917 		case ESTABLISHED:
1918 		case FINISH_RECEIVED:
1919 		case WAIT_FOR_FINISH_ACKNOWLEDGE:
1920 		case FINISH_SENT:
1921 		case FINISH_ACKNOWLEDGED:
1922 		case CLOSING:
1923 		case TIME_WAIT:
1924 		case CLOSED:
1925 			segmentAction = _Receive(segment, buffer);
1926 			break;
1927 	}
1928 
1929 	// process acknowledge action as asked for by the *Receive() method
1930 	if (segmentAction & IMMEDIATE_ACKNOWLEDGE)
1931 		_SendAcknowledge(true);
1932 	else if (segmentAction & ACKNOWLEDGE)
1933 		DelayedAcknowledge();
1934 
1935 	if (segmentAction & SEND_QUEUED)
1936 		_SendQueued();
1937 
1938 	if ((fFlags & (FLAG_CLOSED | FLAG_DELETE_ON_CLOSE))
1939 			== (FLAG_CLOSED | FLAG_DELETE_ON_CLOSE)) {
1940 
1941 		locker.Unlock();
1942 		if (gSocketModule->release_socket(socket))
1943 			segmentAction |= DELETED_ENDPOINT;
1944 	}
1945 
1946 	return segmentAction;
1947 }
1948 
1949 
1950 //	#pragma mark - send
1951 
1952 
1953 tcp_segment_header
1954 TCPEndpoint::_PrepareSendSegment()
1955 {
1956 	// we don't set FLAG_FINISH here, instead we do it
1957 	// conditionally below depending if we are sending
1958 	// the last bytes of the send queue.
1959 
1960 	uint8 flags = 0;
1961 	switch (fState) {
1962 		case CLOSED:
1963 			flags = TCP_FLAG_RESET | TCP_FLAG_ACKNOWLEDGE;
1964 			break;
1965 
1966 		case SYNCHRONIZE_SENT:
1967 			flags = TCP_FLAG_SYNCHRONIZE;
1968 			break;
1969 
1970 		case SYNCHRONIZE_RECEIVED:
1971 			flags = TCP_FLAG_SYNCHRONIZE | TCP_FLAG_ACKNOWLEDGE;
1972 			break;
1973 
1974 		case ESTABLISHED:
1975 		case FINISH_RECEIVED:
1976 		case FINISH_ACKNOWLEDGED:
1977 		case TIME_WAIT:
1978 		case WAIT_FOR_FINISH_ACKNOWLEDGE:
1979 		case FINISH_SENT:
1980 		case CLOSING:
1981 			flags = TCP_FLAG_ACKNOWLEDGE;
1982 
1983 		default:
1984 			break;
1985 	}
1986 
1987 	tcp_segment_header segment(flags);
1988 
1989 	if ((fOptions & TCP_NOOPT) == 0) {
1990 		if ((fFlags & FLAG_OPTION_TIMESTAMP) != 0) {
1991 			segment.options |= TCP_HAS_TIMESTAMPS;
1992 			segment.timestamp_reply = fReceivedTimestamp;
1993 			segment.timestamp_value = tcp_now();
1994 		}
1995 
1996 		if ((segment.flags & TCP_FLAG_SYNCHRONIZE) != 0
1997 				&& fSendNext == fInitialSendSequence) {
1998 			// add connection establishment options
1999 			segment.max_segment_size = fReceiveMaxSegmentSize;
2000 			if (fFlags & FLAG_OPTION_WINDOW_SCALE) {
2001 				segment.options |= TCP_HAS_WINDOW_SCALE;
2002 				segment.window_shift = fReceiveWindowShift;
2003 			}
2004 			if ((fFlags & FLAG_OPTION_SACK_PERMITTED) != 0)
2005 				segment.options |= TCP_SACK_PERMITTED;
2006 		}
2007 
2008 		if (!fReceiveQueue.IsContiguous()
2009 				&& (fFlags & FLAG_OPTION_SACK_PERMITTED) != 0) {
2010 			segment.options |= TCP_HAS_SACK;
2011 			int maxSackCount = MAX_SACK_BLKS
2012 				- ((fFlags & FLAG_OPTION_TIMESTAMP) != 0) ? 1 : 0;
2013 			memset(segment.sacks, 0, sizeof(segment.sacks));
2014 			segment.sackCount = fReceiveQueue.PopulateSackInfo(fReceiveNext,
2015 				maxSackCount, segment.sacks);
2016 		}
2017 	}
2018 
2019 	size_t availableBytes = fReceiveQueue.Free();
2020 	// window size must remain same for duplicate acknowledgements
2021 	if (!fReceiveQueue.IsContiguous())
2022 		availableBytes = (fReceiveMaxAdvertised - fReceiveNext).Number();
2023 	segment.SetAdvertisedWindow(availableBytes, fReceiveWindowShift);
2024 
2025 	segment.acknowledge = fReceiveNext.Number();
2026 
2027 	// Process urgent data
2028 	if (fSendUrgentOffset > fSendNext) {
2029 		segment.flags |= TCP_FLAG_URGENT;
2030 		segment.urgent_offset = (fSendUrgentOffset - fSendNext).Number();
2031 	} else {
2032 		fSendUrgentOffset = fSendUnacknowledged.Number();
2033 			// Keep urgent offset updated, so that it doesn't reach into our
2034 			// send window on overlap
2035 		segment.urgent_offset = 0;
2036 	}
2037 
2038 	return segment;
2039 }
2040 
2041 
2042 status_t
2043 TCPEndpoint::_PrepareAndSend(tcp_segment_header& segment, net_buffer* buffer,
2044 	bool isRetransmit)
2045 {
2046 	LocalAddress().CopyTo(buffer->source);
2047 	PeerAddress().CopyTo(buffer->destination);
2048 
2049 	uint32 size = buffer->size, segmentLength = size;
2050 	segment.sequence = fSendNext.Number();
2051 
2052 	TRACE("_PrepareAndSend(): buffer %p (%" B_PRIu32 " bytes) address %s to "
2053 		"%s flags %#" B_PRIx8 ", seq %" B_PRIu32 ", ack %" B_PRIu32
2054 		", rwnd %" B_PRIu16 ", cwnd %" B_PRIu32 ", ssthresh %" B_PRIu32
2055 		", len %" B_PRIu32 ", first %" B_PRIu32 ", last %" B_PRIu32,
2056 		buffer, buffer->size, PrintAddress(buffer->source),
2057 		PrintAddress(buffer->destination), segment.flags, segment.sequence,
2058 		segment.acknowledge, segment.advertised_window,
2059 		fCongestionWindow, fSlowStartThreshold, segmentLength,
2060 		fSendQueue.FirstSequence().Number(),
2061 		fSendQueue.LastSequence().Number());
2062 	T(Send(this, segment, buffer, fSendQueue.FirstSequence(),
2063 		fSendQueue.LastSequence()));
2064 
2065 	PROBE(buffer, sendWindow);
2066 
2067 	status_t status = add_tcp_header(AddressModule(), segment, buffer);
2068 	if (status != B_OK) {
2069 		gBufferModule->free(buffer);
2070 		return status;
2071 	}
2072 
2073 	if (segment.flags & TCP_FLAG_SYNCHRONIZE) {
2074 		segment.options &= ~TCP_HAS_WINDOW_SCALE;
2075 		segment.max_segment_size = 0;
2076 		size++;
2077 	}
2078 
2079 	if (segment.flags & TCP_FLAG_FINISH)
2080 		size++;
2081 
2082 	status = next->module->send_routed_data(next, fRoute, buffer);
2083 	if (status < B_OK) {
2084 		gBufferModule->free(buffer);
2085 		return status;
2086 	}
2087 
2088 	fSendNext += size;
2089 	if (fSendMax < fSendNext)
2090 		fSendMax = fSendNext;
2091 
2092 	fReceiveMaxAdvertised = fReceiveNext + segment.AdvertisedWindow(fReceiveWindowShift);
2093 
2094 	if (segmentLength != 0 && fState == ESTABLISHED)
2095 		--fSendMaxSegments;
2096 
2097 	if (fSendTime == 0 && !isRetransmit
2098 			&& (segmentLength != 0 || (segment.flags & TCP_FLAG_SYNCHRONIZE) != 0)) {
2099 		fSendTime = tcp_now();
2100 		fRoundTripStartSequence = segment.sequence;
2101 	}
2102 
2103 	if (segment.flags & TCP_FLAG_ACKNOWLEDGE) {
2104 		fLastAcknowledgeSent = segment.acknowledge;
2105 		gStackModule->cancel_timer(&fDelayedAcknowledgeTimer);
2106 	}
2107 
2108 	return B_OK;
2109 }
2110 
2111 
2112 inline bool
2113 TCPEndpoint::_ShouldSendSegment(tcp_segment_header& segment, uint32 length,
2114 	uint32 segmentMaxSize, uint32 flightSize)
2115 {
2116 	if (fState == ESTABLISHED && fSendMaxSegments == 0)
2117 		return false;
2118 
2119 	if (length > 0) {
2120 		// Avoid the silly window syndrome - we only send a segment in case:
2121 		// - we have a full segment to send, or
2122 		// - we're at the end of our buffer queue, or
2123 		// - the buffer is at least larger than half of the maximum send window,
2124 		//   or
2125 		// - we're retransmitting data
2126 		if (length == segmentMaxSize
2127 			|| (fOptions & TCP_NODELAY) != 0
2128 			|| tcp_sequence(fSendNext + length) == fSendQueue.LastSequence()
2129 			|| (fSendMaxWindow > 0 && length >= fSendMaxWindow / 2))
2130 			return true;
2131 	}
2132 
2133 	// check if we need to send a window update to the peer
2134 	if (segment.advertised_window > 0) {
2135 		// correct the window to take into account what already has been advertised
2136 		uint32 window = segment.AdvertisedWindow(fReceiveWindowShift)
2137 			- (fReceiveMaxAdvertised - fReceiveNext).Number();
2138 
2139 		// if we can advertise a window larger than twice the maximum segment
2140 		// size, or half the maximum buffer size we send a window update
2141 		if (window >= (fReceiveMaxSegmentSize << 1)
2142 			|| window >= (socket->receive.buffer_size >> 1))
2143 			return true;
2144 	}
2145 
2146 	if ((segment.flags & (TCP_FLAG_SYNCHRONIZE | TCP_FLAG_FINISH
2147 			| TCP_FLAG_RESET)) != 0)
2148 		return true;
2149 
2150 	// We do have urgent data pending
2151 	if (fSendUrgentOffset > fSendNext)
2152 		return true;
2153 
2154 	// there is no reason to send a segment just now
2155 	return false;
2156 }
2157 
2158 
2159 status_t
2160 TCPEndpoint::_SendAcknowledge(bool force)
2161 {
2162 	if (fRoute == NULL || fState == LISTEN)
2163 		return B_ERROR;
2164 
2165 	tcp_segment_header segment = _PrepareSendSegment();
2166 
2167 	// Is there actually anything to do?
2168 	if (!force && fState == ESTABLISHED
2169 			&& fLastAcknowledgeSent == fReceiveNext
2170 			&& fReceiveQueue.IsContiguous()
2171 			&& !_ShouldSendSegment(segment, 0, 0, 0))
2172 		return B_OK;
2173 
2174 	net_buffer* buffer = gBufferModule->create(256);
2175 	if (buffer == NULL)
2176 		return B_NO_MEMORY;
2177 
2178 	return _PrepareAndSend(segment, buffer, false);
2179 }
2180 
2181 
2182 /*!	Sends one or more TCP segments with the data waiting in the queue. */
2183 status_t
2184 TCPEndpoint::_SendQueued(bool force)
2185 {
2186 	if (fRoute == NULL || fState < ESTABLISHED)
2187 		return B_ERROR;
2188 
2189 	tcp_segment_header segment = _PrepareSendSegment();
2190 
2191 	uint32 sendWindow = fSendWindow;
2192 	if (fCongestionWindow > 0 && fCongestionWindow < sendWindow)
2193 		sendWindow = fCongestionWindow;
2194 
2195 	// fSendUnacknowledged
2196 	//  |    fSendNext      fSendMax
2197 	//  |        |              |
2198 	//  v        v              v
2199 	//  -----------------------------------
2200 	//  | effective window           |
2201 	//  -----------------------------------
2202 
2203 	// Flight size represents the window of data which is currently in the
2204 	// ether. We should never send data such as the flight size becomes larger
2205 	// than the effective window. Note however that the effective window may be
2206 	// reduced (by congestion for instance), so at some point in time flight
2207 	// size may be larger than the currently calculated window.
2208 
2209 	uint32 flightSize = (fSendMax - fSendUnacknowledged).Number();
2210 	uint32 consumedWindow = (fSendNext - fSendUnacknowledged).Number();
2211 
2212 	if (consumedWindow > sendWindow) {
2213 		sendWindow = 0;
2214 		// TODO: enter persist state? try to get a window update.
2215 	} else
2216 		sendWindow -= consumedWindow;
2217 
2218 	uint32 length = min_c(fSendQueue.Available(fSendNext), sendWindow);
2219 	if (length == 0) {
2220 		// Nothing to send.
2221 		return B_OK;
2222 	}
2223 
2224 	bool shouldStartRetransmitTimer = fSendNext == fSendUnacknowledged;
2225 	bool retransmit = fSendNext < fSendMax;
2226 
2227 	if (fDuplicateAcknowledgeCount != 0) {
2228 		// send at most 1 SMSS of data when under limited transmit, fast transmit/recovery
2229 		length = min_c(length, fSendMaxSegmentSize);
2230 	}
2231 
2232 	do {
2233 		uint32 segmentMaxSize = fSendMaxSegmentSize
2234 			- tcp_options_length(segment);
2235 		uint32 segmentLength = min_c(length, segmentMaxSize);
2236 
2237 		if ((fSendNext + segmentLength) == fSendQueue.LastSequence() && !force) {
2238 			if (state_needs_finish(fState))
2239 				segment.flags |= TCP_FLAG_FINISH;
2240 			if (length > 0)
2241 				segment.flags |= TCP_FLAG_PUSH;
2242 		}
2243 
2244 		// Determine if we should really send this segment
2245 		if (!force && !retransmit && !_ShouldSendSegment(segment, segmentLength,
2246 				segmentMaxSize, flightSize)) {
2247 			if (fSendQueue.Available()
2248 				&& !gStackModule->is_timer_active(&fPersistTimer)
2249 				&& !gStackModule->is_timer_active(&fRetransmitTimer))
2250 				_StartPersistTimer();
2251 			break;
2252 		}
2253 
2254 		net_buffer *buffer = gBufferModule->create(256);
2255 		if (buffer == NULL)
2256 			return B_NO_MEMORY;
2257 
2258 		status_t status = B_OK;
2259 		if (segmentLength > 0)
2260 			status = fSendQueue.Get(buffer, fSendNext, segmentLength);
2261 		if (status < B_OK) {
2262 			gBufferModule->free(buffer);
2263 			return status;
2264 		}
2265 
2266 		sendWindow -= buffer->size;
2267 
2268 		status = _PrepareAndSend(segment, buffer, retransmit);
2269 		if (status != B_OK)
2270 			return status;
2271 
2272 		if (shouldStartRetransmitTimer) {
2273 			TRACE("starting initial retransmit timer of: %" B_PRIdBIGTIME,
2274 				fRetransmitTimeout);
2275 			gStackModule->set_timer(&fRetransmitTimer, fRetransmitTimeout);
2276 			T(TimerSet(this, "retransmit", fRetransmitTimeout));
2277 			shouldStartRetransmitTimer = false;
2278 		}
2279 
2280 		length -= segmentLength;
2281 		segment.flags &= ~(TCP_FLAG_SYNCHRONIZE | TCP_FLAG_RESET
2282 			| TCP_FLAG_FINISH);
2283 
2284 		if (retransmit)
2285 			break;
2286 
2287 	} while (length > 0);
2288 
2289 	return B_OK;
2290 }
2291 
2292 
2293 int
2294 TCPEndpoint::_MaxSegmentSize(const sockaddr* address) const
2295 {
2296 	return next->module->get_mtu(next, address) - sizeof(tcp_header);
2297 }
2298 
2299 
2300 status_t
2301 TCPEndpoint::_PrepareSendPath(const sockaddr* peer)
2302 {
2303 	if (fRoute == NULL) {
2304 		fRoute = gDatalinkModule->get_route(Domain(), peer);
2305 		if (fRoute == NULL)
2306 			return ENETUNREACH;
2307 
2308 		if ((fRoute->flags & RTF_LOCAL) != 0)
2309 			fFlags |= FLAG_LOCAL;
2310 	}
2311 
2312 	// make sure connection does not already exist
2313 	status_t status = fManager->SetConnection(this, *LocalAddress(), peer,
2314 		fRoute->interface_address->local);
2315 	if (status < B_OK)
2316 		return status;
2317 
2318 	fInitialSendSequence = system_time() >> 4;
2319 	fSendNext = fInitialSendSequence;
2320 	fSendUnacknowledged = fInitialSendSequence;
2321 	fSendMax = fInitialSendSequence;
2322 	fSendUrgentOffset = fInitialSendSequence;
2323 	fRecover = fInitialSendSequence.Number();
2324 
2325 	// we are counting the SYN here
2326 	fSendQueue.SetInitialSequence(fSendNext + 1);
2327 
2328 	fReceiveMaxSegmentSize = _MaxSegmentSize(peer);
2329 
2330 	// Compute the window shift we advertise to our peer - if it doesn't support
2331 	// this option, this will be reset to 0 (when its SYN is received)
2332 	fReceiveWindowShift = 0;
2333 	while (fReceiveWindowShift < TCP_MAX_WINDOW_SHIFT
2334 		&& (0xffffUL << fReceiveWindowShift) < socket->receive.buffer_size) {
2335 		fReceiveWindowShift++;
2336 	}
2337 
2338 	return B_OK;
2339 }
2340 
2341 
2342 void
2343 TCPEndpoint::_Acknowledged(tcp_segment_header& segment)
2344 {
2345 	TRACE("_Acknowledged(): ack %" B_PRIu32 "; uack %" B_PRIu32 "; next %"
2346 		B_PRIu32 "; max %" B_PRIu32, segment.acknowledge,
2347 		fSendUnacknowledged.Number(), fSendNext.Number(), fSendMax.Number());
2348 
2349 	ASSERT(fSendUnacknowledged <= segment.acknowledge);
2350 
2351 	if (fSendUnacknowledged < segment.acknowledge) {
2352 		fSendQueue.RemoveUntil(segment.acknowledge);
2353 
2354 		uint32 bytesAcknowledged = segment.acknowledge - fSendUnacknowledged.Number();
2355 		fPreviousHighestAcknowledge = fSendUnacknowledged;
2356 		fSendUnacknowledged = segment.acknowledge;
2357 		uint32 flightSize = (fSendMax - fSendUnacknowledged).Number();
2358 		int32 expectedSamples = flightSize / (fSendMaxSegmentSize << 1);
2359 
2360 		if (fPreviousHighestAcknowledge > fSendUnacknowledged) {
2361 			// need to update the recover variable upon a sequence wraparound
2362 			fRecover = segment.acknowledge - 1;
2363 		}
2364 
2365 		// the acknowledgment of the SYN/ACK MUST NOT increase the size of the congestion window
2366 		if (fSendUnacknowledged != fInitialSendSequence) {
2367 			if (fCongestionWindow < fSlowStartThreshold)
2368 				fCongestionWindow += min_c(bytesAcknowledged, fSendMaxSegmentSize);
2369 			else {
2370 				uint32 increment = fSendMaxSegmentSize * fSendMaxSegmentSize;
2371 
2372 				if (increment < fCongestionWindow)
2373 					increment = 1;
2374 				else
2375 					increment /= fCongestionWindow;
2376 
2377 				fCongestionWindow += increment;
2378 			}
2379 
2380 			fSendMaxSegments = UINT32_MAX;
2381 		}
2382 
2383 		if ((fFlags & FLAG_RECOVERY) != 0) {
2384 			fSendNext = fSendUnacknowledged;
2385 			_SendQueued();
2386 			fCongestionWindow -= bytesAcknowledged;
2387 
2388 			if (bytesAcknowledged > fSendMaxSegmentSize)
2389 				fCongestionWindow += fSendMaxSegmentSize;
2390 
2391 			fSendNext = fSendMax;
2392 		} else
2393 			fDuplicateAcknowledgeCount = 0;
2394 
2395 		if (fSendNext < fSendUnacknowledged)
2396 			fSendNext = fSendUnacknowledged;
2397 
2398 		if (fFlags & FLAG_OPTION_TIMESTAMP) {
2399 			_UpdateRoundTripTime(tcp_diff_timestamp(segment.timestamp_reply),
2400 				expectedSamples > 0 ? expectedSamples : 1);
2401 		} else if (fSendTime != 0 && fRoundTripStartSequence < segment.acknowledge) {
2402 			_UpdateRoundTripTime(tcp_diff_timestamp(fSendTime), 1);
2403 			fSendTime = 0;
2404 		}
2405 
2406 		if (fSendUnacknowledged == fSendMax) {
2407 			TRACE("all acknowledged, cancelling retransmission timer.");
2408 			gStackModule->cancel_timer(&fRetransmitTimer);
2409 			T(TimerSet(this, "retransmit", -1));
2410 		} else {
2411 			TRACE("data acknowledged, resetting retransmission timer to: %"
2412 				B_PRIdBIGTIME, fRetransmitTimeout);
2413 			gStackModule->set_timer(&fRetransmitTimer, fRetransmitTimeout);
2414 			T(TimerSet(this, "retransmit", fRetransmitTimeout));
2415 		}
2416 
2417 		if (is_writable(fState)) {
2418 			// notify threads waiting on the socket to become writable again
2419 			fSendCondition.NotifyAll();
2420 			gSocketModule->notify(socket, B_SELECT_WRITE, fSendQueue.Free());
2421 		}
2422 	}
2423 
2424 	// if there is data left to be sent, send it now
2425 	if (fSendQueue.Used() > 0)
2426 		_SendQueued();
2427 }
2428 
2429 
2430 void
2431 TCPEndpoint::_Retransmit()
2432 {
2433 	TRACE("Retransmit()");
2434 
2435 	if (fState < ESTABLISHED) {
2436 		fRetransmitTimeout = TCP_SYN_RETRANSMIT_TIMEOUT;
2437 		fCongestionWindow = fSendMaxSegmentSize;
2438 	} else {
2439 		_ResetSlowStart();
2440 		fDuplicateAcknowledgeCount = 0;
2441 		// Do exponential back off of the retransmit timeout
2442 		fRetransmitTimeout *= 2;
2443 		if (fRetransmitTimeout > TCP_MAX_RETRANSMIT_TIMEOUT)
2444 			fRetransmitTimeout = TCP_MAX_RETRANSMIT_TIMEOUT;
2445 	}
2446 
2447 	fSendNext = fSendUnacknowledged;
2448 	_SendQueued();
2449 
2450 	fRecover = fSendNext.Number() - 1;
2451 	if ((fFlags & FLAG_RECOVERY) != 0)
2452 		fFlags &= ~FLAG_RECOVERY;
2453 }
2454 
2455 
2456 void
2457 TCPEndpoint::_UpdateRoundTripTime(int32 roundTripTime, int32 expectedSamples)
2458 {
2459 	if (fSmoothedRoundTripTime == 0) {
2460 		fSmoothedRoundTripTime = roundTripTime;
2461 		fRoundTripVariation = roundTripTime / 2;
2462 		fRetransmitTimeout = (fSmoothedRoundTripTime + max_c(100, fRoundTripVariation * 4))
2463 				* kTimestampFactor;
2464 	} else {
2465 		int32 delta = fSmoothedRoundTripTime - roundTripTime;
2466 		if (delta < 0)
2467 			delta = -delta;
2468 		fRoundTripVariation += (delta - fRoundTripVariation) / (expectedSamples * 4);
2469 		fSmoothedRoundTripTime += (roundTripTime - fSmoothedRoundTripTime) / (expectedSamples * 8);
2470 		fRetransmitTimeout = (fSmoothedRoundTripTime + max_c(100, fRoundTripVariation * 4))
2471 			* kTimestampFactor;
2472 	}
2473 
2474 	if (fRetransmitTimeout > TCP_MAX_RETRANSMIT_TIMEOUT)
2475 		fRetransmitTimeout = TCP_MAX_RETRANSMIT_TIMEOUT;
2476 
2477 	if (fRetransmitTimeout < TCP_MIN_RETRANSMIT_TIMEOUT)
2478 		fRetransmitTimeout = TCP_MIN_RETRANSMIT_TIMEOUT;
2479 
2480 	TRACE("  RTO is now %" B_PRIdBIGTIME " (after rtt %" B_PRId32 "ms)",
2481 		fRetransmitTimeout, roundTripTime);
2482 }
2483 
2484 
2485 void
2486 TCPEndpoint::_ResetSlowStart()
2487 {
2488 	fSlowStartThreshold = max_c((fSendMax - fSendUnacknowledged).Number() / 2,
2489 		2 * fSendMaxSegmentSize);
2490 	fCongestionWindow = fSendMaxSegmentSize;
2491 }
2492 
2493 
2494 //	#pragma mark - timer
2495 
2496 
2497 /*static*/ void
2498 TCPEndpoint::_RetransmitTimer(net_timer* timer, void* _endpoint)
2499 {
2500 	TCPEndpoint* endpoint = (TCPEndpoint*)_endpoint;
2501 	T(TimerTriggered(endpoint, "retransmit"));
2502 
2503 	MutexLocker locker(endpoint->fLock);
2504 	if (!locker.IsLocked() || gStackModule->is_timer_active(timer))
2505 		return;
2506 
2507 	endpoint->_Retransmit();
2508 }
2509 
2510 
2511 /*static*/ void
2512 TCPEndpoint::_PersistTimer(net_timer* timer, void* _endpoint)
2513 {
2514 	TCPEndpoint* endpoint = (TCPEndpoint*)_endpoint;
2515 	T(TimerTriggered(endpoint, "persist"));
2516 
2517 	MutexLocker locker(endpoint->fLock);
2518 	if (!locker.IsLocked())
2519 		return;
2520 
2521 	// the timer might not have been canceled early enough
2522 	if (endpoint->State() == CLOSED)
2523 		return;
2524 
2525 	endpoint->_SendQueued(true);
2526 }
2527 
2528 
2529 /*static*/ void
2530 TCPEndpoint::_DelayedAcknowledgeTimer(net_timer* timer, void* _endpoint)
2531 {
2532 	TCPEndpoint* endpoint = (TCPEndpoint*)_endpoint;
2533 	T(TimerTriggered(endpoint, "delayed ack"));
2534 
2535 	MutexLocker locker(endpoint->fLock);
2536 	if (!locker.IsLocked())
2537 		return;
2538 
2539 	// the timer might not have been canceled early enough
2540 	if (endpoint->State() == CLOSED)
2541 		return;
2542 
2543 	endpoint->_SendAcknowledge();
2544 }
2545 
2546 
2547 /*static*/ void
2548 TCPEndpoint::_TimeWaitTimer(net_timer* timer, void* _endpoint)
2549 {
2550 	TCPEndpoint* endpoint = (TCPEndpoint*)_endpoint;
2551 	T(TimerTriggered(endpoint, "time-wait"));
2552 
2553 	MutexLocker locker(endpoint->fLock);
2554 	if (!locker.IsLocked())
2555 		return;
2556 
2557 	if ((endpoint->fFlags & FLAG_CLOSED) == 0) {
2558 		endpoint->fFlags |= FLAG_DELETE_ON_CLOSE;
2559 		return;
2560 	}
2561 
2562 	locker.Unlock();
2563 
2564 	gSocketModule->release_socket(endpoint->socket);
2565 }
2566 
2567 
2568 /*static*/ status_t
2569 TCPEndpoint::_WaitForCondition(ConditionVariable& condition,
2570 	MutexLocker& locker, bigtime_t timeout)
2571 {
2572 	ConditionVariableEntry entry;
2573 	condition.Add(&entry);
2574 
2575 	locker.Unlock();
2576 	status_t result = entry.Wait(B_ABSOLUTE_TIMEOUT | B_CAN_INTERRUPT, timeout);
2577 	locker.Lock();
2578 
2579 	return result;
2580 }
2581 
2582 
2583 //	#pragma mark -
2584 
2585 
2586 void
2587 TCPEndpoint::Dump() const
2588 {
2589 	kprintf("TCP endpoint %p\n", this);
2590 	kprintf("  state: %s\n", name_for_state(fState));
2591 	kprintf("  flags: 0x%" B_PRIx32 "\n", fFlags);
2592 #if KDEBUG
2593 	kprintf("  lock: { %p, holder: %" B_PRId32 " }\n", &fLock, fLock.holder);
2594 #endif
2595 	kprintf("  accept sem: %" B_PRId32 "\n", fAcceptSemaphore);
2596 	kprintf("  options: 0x%" B_PRIx32 "\n", (uint32)fOptions);
2597 	kprintf("  send\n");
2598 	kprintf("    window shift: %" B_PRIu8 "\n", fSendWindowShift);
2599 	kprintf("    unacknowledged: %" B_PRIu32 "\n",
2600 		fSendUnacknowledged.Number());
2601 	kprintf("    next: %" B_PRIu32 "\n", fSendNext.Number());
2602 	kprintf("    max: %" B_PRIu32 "\n", fSendMax.Number());
2603 	kprintf("    urgent offset: %" B_PRIu32 "\n", fSendUrgentOffset.Number());
2604 	kprintf("    window: %" B_PRIu32 "\n", fSendWindow);
2605 	kprintf("    max window: %" B_PRIu32 "\n", fSendMaxWindow);
2606 	kprintf("    max segment size: %" B_PRIu32 "\n", fSendMaxSegmentSize);
2607 	kprintf("    queue: %" B_PRIuSIZE " / %" B_PRIuSIZE "\n", fSendQueue.Used(),
2608 		fSendQueue.Size());
2609 #if DEBUG_TCP_BUFFER_QUEUE
2610 	fSendQueue.Dump();
2611 #endif
2612 	kprintf("    last acknowledge sent: %" B_PRIu32 "\n",
2613 		fLastAcknowledgeSent.Number());
2614 	kprintf("    initial sequence: %" B_PRIu32 "\n",
2615 		fInitialSendSequence.Number());
2616 	kprintf("  receive\n");
2617 	kprintf("    window shift: %" B_PRIu8 "\n", fReceiveWindowShift);
2618 	kprintf("    next: %" B_PRIu32 "\n", fReceiveNext.Number());
2619 	kprintf("    max advertised: %" B_PRIu32 "\n",
2620 		fReceiveMaxAdvertised.Number());
2621 	kprintf("    window: %" B_PRIu32 "\n", fReceiveWindow);
2622 	kprintf("    max segment size: %" B_PRIu32 "\n", fReceiveMaxSegmentSize);
2623 	kprintf("    queue: %" B_PRIuSIZE " / %" B_PRIuSIZE "\n",
2624 		fReceiveQueue.Available(), fReceiveQueue.Size());
2625 #if DEBUG_TCP_BUFFER_QUEUE
2626 	fReceiveQueue.Dump();
2627 #endif
2628 	kprintf("    initial sequence: %" B_PRIu32 "\n",
2629 		fInitialReceiveSequence.Number());
2630 	kprintf("    duplicate acknowledge count: %" B_PRIu32 "\n",
2631 		fDuplicateAcknowledgeCount);
2632 	kprintf("  smoothed round trip time: %" B_PRId32 " (deviation %" B_PRId32 ")\n",
2633 		fSmoothedRoundTripTime, fRoundTripVariation);
2634 	kprintf("  retransmit timeout: %" B_PRId64 "\n", fRetransmitTimeout);
2635 	kprintf("  congestion window: %" B_PRIu32 "\n", fCongestionWindow);
2636 	kprintf("  slow start threshold: %" B_PRIu32 "\n", fSlowStartThreshold);
2637 }
2638 
2639