xref: /haiku/src/add-ons/kernel/network/protocols/tcp/tcp.cpp (revision 60a6f1d5d7a8715cd3897dd0b626f2e4a64984a8)
1 /*
2  * Copyright 2006-2009, Haiku, Inc. All Rights Reserved.
3  * Distributed under the terms of the MIT License.
4  *
5  * Authors:
6  *		Axel Dörfler, axeld@pinc-software.de
7  *		Andrew Galante, haiku.galante@gmail.com
8  *		Hugo Santos, hugosantos@gmail.com
9  */
10 
11 
12 #include "EndpointManager.h"
13 #include "TCPEndpoint.h"
14 #include "tcp.h"
15 
16 #include <net_protocol.h>
17 #include <net_stat.h>
18 
19 #include <KernelExport.h>
20 #include <util/list.h>
21 
22 #include <netinet/in.h>
23 #include <netinet/ip.h>
24 #include <new>
25 #include <stdlib.h>
26 #include <string.h>
27 
28 #include <lock.h>
29 #include <util/AutoLock.h>
30 
31 #include <NetBufferUtilities.h>
32 #include <NetUtilities.h>
33 
34 
35 //#define TRACE_TCP
36 #ifdef TRACE_TCP
37 #	define TRACE(x) dprintf x
38 #else
39 #	define TRACE(x)
40 #endif
41 
42 
43 typedef NetBufferField<uint16, offsetof(tcp_header, checksum)> TCPChecksumField;
44 
45 
46 net_buffer_module_info *gBufferModule;
47 net_datalink_module_info *gDatalinkModule;
48 net_socket_module_info *gSocketModule;
49 net_stack_module_info *gStackModule;
50 
51 
52 static EndpointManager* sEndpointManagers[AF_MAX];
53 static rw_lock sEndpointManagersLock;
54 
55 
56 // The TCP header length is at most 64 bytes.
57 static const int kMaxOptionSize = 64 - sizeof(tcp_header);
58 
59 
60 /*!	Returns an endpoint manager for the specified domain, if any.
61 	You need to hold the sEndpointManagersLock when calling this function.
62 */
63 static inline EndpointManager*
64 endpoint_manager_for_locked(int family)
65 {
66 	if (family >= AF_MAX || family < 0)
67 		return NULL;
68 
69 	return sEndpointManagers[family];
70 }
71 
72 
73 /*!	Returns an endpoint manager for the specified domain, if any */
74 static inline EndpointManager*
75 endpoint_manager_for(net_domain* domain)
76 {
77 	ReadLocker _(sEndpointManagersLock);
78 
79 	return endpoint_manager_for_locked(domain->family);
80 }
81 
82 
83 static inline void
84 bump_option(tcp_option *&option, size_t &length)
85 {
86 	if (option->kind <= TCP_OPTION_NOP) {
87 		length++;
88 		option = (tcp_option *)((uint8 *)option + 1);
89 	} else {
90 		length += option->length;
91 		option = (tcp_option *)((uint8 *)option + option->length);
92 	}
93 }
94 
95 
96 static inline size_t
97 add_options(tcp_segment_header &segment, uint8 *buffer, size_t bufferSize)
98 {
99 	tcp_option *option = (tcp_option *)buffer;
100 	size_t length = 0;
101 
102 	if (segment.max_segment_size > 0 && length + 8 <= bufferSize) {
103 		option->kind = TCP_OPTION_MAX_SEGMENT_SIZE;
104 		option->length = 4;
105 		option->max_segment_size = htons(segment.max_segment_size);
106 		bump_option(option, length);
107 	}
108 
109 	if ((segment.options & TCP_HAS_TIMESTAMPS) != 0
110 		&& length + 12 <= bufferSize) {
111 		// two NOPs so the timestamps get aligned to a 4 byte boundary
112 		option->kind = TCP_OPTION_NOP;
113 		bump_option(option, length);
114 		option->kind = TCP_OPTION_NOP;
115 		bump_option(option, length);
116 		option->kind = TCP_OPTION_TIMESTAMP;
117 		option->length = 10;
118 		option->timestamp.value = htonl(segment.timestamp_value);
119 		option->timestamp.reply = htonl(segment.timestamp_reply);
120 		bump_option(option, length);
121 	}
122 
123 	if ((segment.options & TCP_HAS_WINDOW_SCALE) != 0
124 		&& length + 4 <= bufferSize) {
125 		// insert one NOP so that the subsequent data is aligned on a 4 byte boundary
126 		option->kind = TCP_OPTION_NOP;
127 		bump_option(option, length);
128 
129 		option->kind = TCP_OPTION_WINDOW_SHIFT;
130 		option->length = 3;
131 		option->window_shift = segment.window_shift;
132 		bump_option(option, length);
133 	}
134 
135 	if ((segment.options & TCP_SACK_PERMITTED) != 0
136 		&& length + 2 <= bufferSize) {
137 		option->kind = TCP_OPTION_SACK_PERMITTED;
138 		option->length = 2;
139 		bump_option(option, length);
140 	}
141 
142 	if (segment.sack_count > 0) {
143 		int sackCount = ((int)(bufferSize - length) - 4) / sizeof(tcp_sack);
144 		if (sackCount > segment.sack_count)
145 			sackCount = segment.sack_count;
146 
147 		if (sackCount > 0) {
148 			option->kind = TCP_OPTION_NOP;
149 			bump_option(option, length);
150 			option->kind = TCP_OPTION_NOP;
151 			bump_option(option, length);
152 			option->kind = TCP_OPTION_SACK;
153 			option->length = 2 + sackCount * sizeof(tcp_sack);
154 			memcpy(option->sack, segment.sacks, sackCount * sizeof(tcp_sack));
155 			bump_option(option, length);
156 		}
157 	}
158 
159 	if ((length & 3) == 0) {
160 		// options completely fill out the option space
161 		return length;
162 	}
163 
164 	option->kind = TCP_OPTION_END;
165 	return (length + 3) & ~3;
166 		// bump to a multiple of 4 length
167 }
168 
169 
170 static void
171 process_options(tcp_segment_header &segment, net_buffer *buffer, size_t size)
172 {
173 	if (size == 0)
174 		return;
175 
176 	tcp_option *option;
177 
178 	uint8 optionsBuffer[kMaxOptionSize];
179 	if (gBufferModule->direct_access(buffer, sizeof(tcp_header), size,
180 			(void **)&option) != B_OK) {
181 		if ((size_t)size > sizeof(optionsBuffer)) {
182 			dprintf("Ignoring TCP options larger than expected.\n");
183 			return;
184 		}
185 
186 		gBufferModule->read(buffer, sizeof(tcp_header), optionsBuffer, size);
187 		option = (tcp_option *)optionsBuffer;
188 	}
189 
190 	while (size > 0) {
191 		int32 length = -1;
192 
193 		switch (option->kind) {
194 			case TCP_OPTION_END:
195 			case TCP_OPTION_NOP:
196 				length = 1;
197 				break;
198 			case TCP_OPTION_MAX_SEGMENT_SIZE:
199 				if (option->length == 4 && size >= 4)
200 					segment.max_segment_size = ntohs(option->max_segment_size);
201 				break;
202 			case TCP_OPTION_WINDOW_SHIFT:
203 				if (option->length == 3 && size >= 3) {
204 					segment.options |= TCP_HAS_WINDOW_SCALE;
205 					segment.window_shift = option->window_shift;
206 				}
207 				break;
208 			case TCP_OPTION_TIMESTAMP:
209 				if (option->length == 10 && size >= 10) {
210 					segment.options |= TCP_HAS_TIMESTAMPS;
211 					segment.timestamp_value = ntohl(option->timestamp.value);
212 					segment.timestamp_reply =
213 						ntohl(option->timestamp.reply);
214 				}
215 				break;
216 			case TCP_OPTION_SACK_PERMITTED:
217 				if (option->length == 2 && size >= 2)
218 					segment.options |= TCP_SACK_PERMITTED;
219 				break;
220 		}
221 
222 		if (length < 0) {
223 			length = option->length;
224 			if (length == 0 || length > (ssize_t)size)
225 				break;
226 		}
227 
228 		option = (tcp_option *)((uint8 *)option + length);
229 		size -= length;
230 	}
231 }
232 
233 
234 #if 0
235 static void
236 dump_tcp_header(tcp_header &header)
237 {
238 	dprintf("  source port: %u\n", ntohs(header.source_port));
239 	dprintf("  dest port: %u\n", ntohs(header.destination_port));
240 	dprintf("  sequence: %lu\n", header.Sequence());
241 	dprintf("  ack: %lu\n", header.Acknowledge());
242 	dprintf("  flags: %s%s%s%s%s%s\n", (header.flags & TCP_FLAG_FINISH) ? "FIN " : "",
243 		(header.flags & TCP_FLAG_SYNCHRONIZE) ? "SYN " : "",
244 		(header.flags & TCP_FLAG_RESET) ? "RST " : "",
245 		(header.flags & TCP_FLAG_PUSH) ? "PUSH " : "",
246 		(header.flags & TCP_FLAG_ACKNOWLEDGE) ? "ACK " : "",
247 		(header.flags & TCP_FLAG_URGENT) ? "URG " : "");
248 	dprintf("  window: %u\n", header.AdvertisedWindow());
249 	dprintf("  urgent offset: %u\n", header.UrgentOffset());
250 }
251 #endif
252 
253 
254 static int
255 dump_endpoints(int argc, char** argv)
256 {
257 	for (int i = 0; i < AF_MAX; i++) {
258 		EndpointManager* manager = sEndpointManagers[i];
259 		if (manager != NULL)
260 			manager->Dump();
261 	}
262 
263 	return 0;
264 }
265 
266 
267 static int
268 dump_endpoint(int argc, char** argv)
269 {
270 	if (argc < 2) {
271 		kprintf("usage: tcp_endpoint [address]\n");
272 		return 0;
273 	}
274 
275 	TCPEndpoint* endpoint = (TCPEndpoint*)parse_expression(argv[1]);
276 	endpoint->Dump();
277 
278 	return 0;
279 }
280 
281 
282 //	#pragma mark - internal API
283 
284 
285 /*!	Creates a new endpoint manager for the specified domain, or returns
286 	an existing one for this domain.
287 */
288 EndpointManager*
289 get_endpoint_manager(net_domain* domain)
290 {
291 	// See if there is one already
292 	EndpointManager* endpointManager = endpoint_manager_for(domain);
293 	if (endpointManager != NULL)
294 		return endpointManager;
295 
296 	WriteLocker _(sEndpointManagersLock);
297 
298 	endpointManager = endpoint_manager_for_locked(domain->family);
299 	if (endpointManager != NULL)
300 		return endpointManager;
301 
302 	// There is no endpoint manager for this domain yet, so we need
303 	// to create one.
304 
305 	endpointManager = new(std::nothrow) EndpointManager(domain);
306 	if (endpointManager == NULL)
307 		return NULL;
308 
309 	if (endpointManager->Init() != B_OK) {
310 		delete endpointManager;
311 		return NULL;
312 	}
313 
314 	sEndpointManagers[domain->family] = endpointManager;
315 	return endpointManager;
316 }
317 
318 
319 void
320 put_endpoint_manager(EndpointManager* endpointManager)
321 {
322 	// TODO: we may want to use reference counting instead of only discarding
323 	// them on unload. But since there is likely only IPv4/v6 there is not much
324 	// point to it.
325 }
326 
327 
328 const char*
329 name_for_state(tcp_state state)
330 {
331 	switch (state) {
332 		case CLOSED:
333 			return "closed";
334 		case LISTEN:
335 			return "listen";
336 		case SYNCHRONIZE_SENT:
337 			return "syn-sent";
338 		case SYNCHRONIZE_RECEIVED:
339 			return "syn-received";
340 		case ESTABLISHED:
341 			return "established";
342 
343 		// peer closes the connection
344 		case FINISH_RECEIVED:
345 			return "close-wait";
346 		case WAIT_FOR_FINISH_ACKNOWLEDGE:
347 			return "last-ack";
348 
349 		// we close the connection
350 		case FINISH_SENT:
351 			return "fin-wait1";
352 		case FINISH_ACKNOWLEDGED:
353 			return "fin-wait2";
354 		case CLOSING:
355 			return "closing";
356 
357 		case TIME_WAIT:
358 			return "time-wait";
359 	}
360 
361 	return "-";
362 }
363 
364 
365 /*!	Constructs a TCP header on \a buffer with the specified values
366 	for \a flags, \a seq \a ack and \a advertisedWindow.
367 */
368 status_t
369 add_tcp_header(net_address_module_info* addressModule,
370 	tcp_segment_header& segment, net_buffer* buffer)
371 {
372 	buffer->protocol = IPPROTO_TCP;
373 
374 	uint8 optionsBuffer[kMaxOptionSize];
375 	uint32 optionsLength = add_options(segment, optionsBuffer,
376 		sizeof(optionsBuffer));
377 
378 	NetBufferPrepend<tcp_header> bufferHeader(buffer,
379 		sizeof(tcp_header) + optionsLength);
380 	if (bufferHeader.Status() != B_OK)
381 		return bufferHeader.Status();
382 
383 	tcp_header& header = bufferHeader.Data();
384 
385 	header.source_port = addressModule->get_port(buffer->source);
386 	header.destination_port = addressModule->get_port(buffer->destination);
387 	header.sequence = htonl(segment.sequence);
388 	header.acknowledge = (segment.flags & TCP_FLAG_ACKNOWLEDGE)
389 		? htonl(segment.acknowledge) : 0;
390 	header.reserved = 0;
391 	header.header_length = (sizeof(tcp_header) + optionsLength) >> 2;
392 	header.flags = segment.flags;
393 	header.advertised_window = htons(segment.advertised_window);
394 	header.checksum = 0;
395 	header.urgent_offset = htons(segment.urgent_offset);
396 
397 	// we must detach before calculating the checksum as we may
398 	// not have a contiguous buffer.
399 	bufferHeader.Sync();
400 
401 	if (optionsLength > 0) {
402 		gBufferModule->write(buffer, sizeof(tcp_header), optionsBuffer,
403 			optionsLength);
404 	}
405 
406 	TRACE(("add_tcp_header(): buffer %p, flags 0x%x, seq %lu, ack %lu, up %u, "
407 		"win %u\n", buffer, segment.flags, segment.sequence,
408 		segment.acknowledge, segment.urgent_offset, segment.advertised_window));
409 
410 	*TCPChecksumField(buffer) = Checksum::PseudoHeader(addressModule,
411 		gBufferModule, buffer, IPPROTO_TCP);
412 
413 	return B_OK;
414 }
415 
416 
417 size_t
418 tcp_options_length(tcp_segment_header& segment)
419 {
420 	size_t length = 0;
421 
422 	if (segment.max_segment_size > 0)
423 		length += 4;
424 
425 	if (segment.options & TCP_HAS_TIMESTAMPS)
426 		length += 12;
427 
428 	if (segment.options & TCP_HAS_WINDOW_SCALE)
429 		length += 4;
430 
431 	if (segment.options & TCP_SACK_PERMITTED)
432 		length += 2;
433 
434 	if (segment.sack_count > 0) {
435 		int sackCount = min_c((int)((kMaxOptionSize - length - 4)
436 			/ sizeof(tcp_sack)), segment.sack_count);
437 		if (sackCount > 0)
438 			length += 4 + sackCount * sizeof(tcp_sack);
439 	}
440 
441 	if ((length & 3) == 0)
442 		return length;
443 
444 	return (length + 3) & ~3;
445 }
446 
447 
448 //	#pragma mark - protocol API
449 
450 
451 net_protocol*
452 tcp_init_protocol(net_socket* socket)
453 {
454 	socket->send.buffer_size = 32768;
455 		// override net_socket default
456 
457 	TCPEndpoint* protocol = new (std::nothrow) TCPEndpoint(socket);
458 	if (protocol == NULL)
459 		return NULL;
460 
461 	if (protocol->InitCheck() != B_OK) {
462 		delete protocol;
463 		return NULL;
464 	}
465 
466 	TRACE(("Creating new TCPEndpoint: %p\n", protocol));
467 	socket->protocol = IPPROTO_TCP;
468 	return protocol;
469 }
470 
471 
472 status_t
473 tcp_uninit_protocol(net_protocol* protocol)
474 {
475 	TRACE(("Deleting TCPEndpoint: %p\n", protocol));
476 	delete (TCPEndpoint*)protocol;
477 	return B_OK;
478 }
479 
480 
481 status_t
482 tcp_open(net_protocol* protocol)
483 {
484 	return ((TCPEndpoint*)protocol)->Open();
485 }
486 
487 
488 status_t
489 tcp_close(net_protocol* protocol)
490 {
491 	return ((TCPEndpoint*)protocol)->Close();
492 }
493 
494 
495 status_t
496 tcp_free(net_protocol* protocol)
497 {
498 	((TCPEndpoint*)protocol)->Free();
499 	return B_OK;
500 }
501 
502 
503 status_t
504 tcp_connect(net_protocol* protocol, const struct sockaddr* address)
505 {
506 	return ((TCPEndpoint*)protocol)->Connect(address);
507 }
508 
509 
510 status_t
511 tcp_accept(net_protocol* protocol, struct net_socket** _acceptedSocket)
512 {
513 	return ((TCPEndpoint*)protocol)->Accept(_acceptedSocket);
514 }
515 
516 
517 status_t
518 tcp_control(net_protocol* _protocol, int level, int option, void* value,
519 	size_t* _length)
520 {
521 	TCPEndpoint* protocol = (TCPEndpoint*)_protocol;
522 
523 	if ((level & LEVEL_MASK) == IPPROTO_TCP) {
524 		if (option == NET_STAT_SOCKET)
525 			return protocol->FillStat((net_stat*)value);
526 	}
527 
528 	return protocol->next->module->control(protocol->next, level, option,
529 		value, _length);
530 }
531 
532 
533 status_t
534 tcp_getsockopt(net_protocol* _protocol, int level, int option, void* value,
535 	int* _length)
536 {
537 	TCPEndpoint* protocol = (TCPEndpoint*)_protocol;
538 
539 	if (level == IPPROTO_TCP)
540 		return protocol->GetOption(option, value, _length);
541 
542 	return protocol->next->module->getsockopt(protocol->next, level, option,
543 		value, _length);
544 }
545 
546 
547 status_t
548 tcp_setsockopt(net_protocol* _protocol, int level, int option,
549 	const void* _value, int length)
550 {
551 	TCPEndpoint* protocol = (TCPEndpoint*)_protocol;
552 
553 	if (level == SOL_SOCKET) {
554 		if (option == SO_SNDBUF || option == SO_RCVBUF) {
555 			if (length != sizeof(int))
556 				return B_BAD_VALUE;
557 
558 			status_t status;
559 			const int* value = (const int*)_value;
560 
561 			if (option == SO_SNDBUF)
562 				status = protocol->SetSendBufferSize(*value);
563 			else
564 				status = protocol->SetReceiveBufferSize(*value);
565 
566 			if (status < B_OK)
567 				return status;
568 		}
569 	} else if (level == IPPROTO_TCP)
570 		return protocol->SetOption(option, _value, length);
571 
572 	return protocol->next->module->setsockopt(protocol->next, level, option,
573 		_value, length);
574 }
575 
576 
577 status_t
578 tcp_bind(net_protocol* protocol, const struct sockaddr* address)
579 {
580 	return ((TCPEndpoint*)protocol)->Bind(address);
581 }
582 
583 
584 status_t
585 tcp_unbind(net_protocol* protocol, struct sockaddr* address)
586 {
587 	return ((TCPEndpoint*)protocol)->Unbind(address);
588 }
589 
590 
591 status_t
592 tcp_listen(net_protocol* protocol, int count)
593 {
594 	return ((TCPEndpoint*)protocol)->Listen(count);
595 }
596 
597 
598 status_t
599 tcp_shutdown(net_protocol* protocol, int direction)
600 {
601 	return ((TCPEndpoint*)protocol)->Shutdown(direction);
602 }
603 
604 
605 status_t
606 tcp_send_data(net_protocol* protocol, net_buffer* buffer)
607 {
608 	return ((TCPEndpoint*)protocol)->SendData(buffer);
609 }
610 
611 
612 status_t
613 tcp_send_routed_data(net_protocol* protocol, struct net_route* route,
614 	net_buffer* buffer)
615 {
616 	// TCP never sends routed data
617 	return B_ERROR;
618 }
619 
620 
621 ssize_t
622 tcp_send_avail(net_protocol* protocol)
623 {
624 	return ((TCPEndpoint*)protocol)->SendAvailable();
625 }
626 
627 
628 status_t
629 tcp_read_data(net_protocol* protocol, size_t numBytes, uint32 flags,
630 	net_buffer** _buffer)
631 {
632 	return ((TCPEndpoint*)protocol)->ReadData(numBytes, flags, _buffer);
633 }
634 
635 
636 ssize_t
637 tcp_read_avail(net_protocol* protocol)
638 {
639 	return ((TCPEndpoint*)protocol)->ReadAvailable();
640 }
641 
642 
643 struct net_domain*
644 tcp_get_domain(net_protocol* protocol)
645 {
646 	return protocol->next->module->get_domain(protocol->next);
647 }
648 
649 
650 size_t
651 tcp_get_mtu(net_protocol* protocol, const struct sockaddr* address)
652 {
653 	return protocol->next->module->get_mtu(protocol->next, address);
654 }
655 
656 
657 status_t
658 tcp_receive_data(net_buffer* buffer)
659 {
660 	TRACE(("TCP: Received buffer %p\n", buffer));
661 
662 	if (buffer->interface_address == NULL
663 		|| buffer->interface_address->domain == NULL)
664 		return B_ERROR;
665 
666 	net_domain* domain = buffer->interface_address->domain;
667 	net_address_module_info* addressModule = domain->address_module;
668 
669 	NetBufferHeaderReader<tcp_header> bufferHeader(buffer);
670 	if (bufferHeader.Status() < B_OK)
671 		return bufferHeader.Status();
672 
673 	tcp_header& header = bufferHeader.Data();
674 
675 	uint16 headerLength = header.HeaderLength();
676 	if (headerLength < sizeof(tcp_header))
677 		return B_BAD_DATA;
678 
679 	if (Checksum::PseudoHeader(addressModule, gBufferModule, buffer,
680 			IPPROTO_TCP) != 0)
681 		return B_BAD_DATA;
682 
683 	addressModule->set_port(buffer->source, header.source_port);
684 	addressModule->set_port(buffer->destination, header.destination_port);
685 
686 	TRACE(("  Looking for: peer %s, local %s\n",
687 		AddressString(domain, buffer->source, true).Data(),
688 		AddressString(domain, buffer->destination, true).Data()));
689 	//dump_tcp_header(header);
690 	//gBufferModule->dump(buffer);
691 
692 	tcp_segment_header segment(header.flags);
693 	segment.sequence = header.Sequence();
694 	segment.acknowledge = header.Acknowledge();
695 	segment.advertised_window = header.AdvertisedWindow();
696 	segment.urgent_offset = header.UrgentOffset();
697 	process_options(segment, buffer, headerLength - sizeof(tcp_header));
698 
699 	bufferHeader.Remove(headerLength);
700 		// we no longer need to keep the header around
701 
702 	EndpointManager* endpointManager = endpoint_manager_for(domain);
703 	if (endpointManager == NULL) {
704 		TRACE(("  No endpoint manager!\n"));
705 		return B_ERROR;
706 	}
707 
708 	int32 segmentAction = DROP;
709 
710 	TCPEndpoint* endpoint = endpointManager->FindConnection(
711 		buffer->destination, buffer->source);
712 	if (endpoint != NULL) {
713 		segmentAction = endpoint->SegmentReceived(segment, buffer);
714 
715 		// There are some states in which the socket could have been deleted
716 		// while handling a segment. If this flag is set in segmentAction
717 		// then we know the socket has been freed and can skip releasing
718 		// the reference acquired in EndpointManager::FindConnection()
719 		// above.
720 		if ((segmentAction & DELETED_ENDPOINT) == 0)
721 			gSocketModule->release_socket(endpoint->socket);
722 	} else if ((segment.flags & TCP_FLAG_RESET) == 0)
723 		segmentAction = DROP | RESET;
724 
725 	if ((segmentAction & RESET) != 0) {
726 		// send reset
727 		endpointManager->ReplyWithReset(segment, buffer);
728 	}
729 	if ((segmentAction & DROP) != 0)
730 		gBufferModule->free(buffer);
731 
732 	return B_OK;
733 }
734 
735 
736 status_t
737 tcp_error_received(net_error error, net_buffer* data)
738 {
739 	return B_ERROR;
740 }
741 
742 
743 status_t
744 tcp_error_reply(net_protocol* protocol, net_buffer* cause, net_error error,
745 	net_error_data* errorData)
746 {
747 	return B_ERROR;
748 }
749 
750 
751 //	#pragma mark -
752 
753 
754 static status_t
755 tcp_init()
756 {
757 	rw_lock_init(&sEndpointManagersLock, "endpoint managers");
758 
759 	status_t status = gStackModule->register_domain_protocols(AF_INET,
760 		SOCK_STREAM, 0,
761 		"network/protocols/tcp/v1",
762 		"network/protocols/ipv4/v1",
763 		NULL);
764 	if (status < B_OK)
765 		return status;
766 	status = gStackModule->register_domain_protocols(AF_INET6,
767 		SOCK_STREAM, 0,
768 		"network/protocols/tcp/v1",
769 		"network/protocols/ipv6/v1",
770 		NULL);
771 	if (status < B_OK)
772 		return status;
773 
774 	status = gStackModule->register_domain_protocols(AF_INET, SOCK_STREAM,
775 		IPPROTO_TCP,
776 		"network/protocols/tcp/v1",
777 		"network/protocols/ipv4/v1",
778 		NULL);
779 	if (status < B_OK)
780 		return status;
781 	status = gStackModule->register_domain_protocols(AF_INET6, SOCK_STREAM,
782 		IPPROTO_TCP,
783 		"network/protocols/tcp/v1",
784 		"network/protocols/ipv6/v1",
785 		NULL);
786 	if (status < B_OK)
787 		return status;
788 
789 	status = gStackModule->register_domain_receiving_protocol(AF_INET,
790 		IPPROTO_TCP, "network/protocols/tcp/v1");
791 	if (status < B_OK)
792 		return status;
793 	status = gStackModule->register_domain_receiving_protocol(AF_INET6,
794 		IPPROTO_TCP, "network/protocols/tcp/v1");
795 	if (status < B_OK)
796 		return status;
797 
798 	add_debugger_command("tcp_endpoints", dump_endpoints,
799 		"lists all open TCP endpoints");
800 	add_debugger_command("tcp_endpoint", dump_endpoint,
801 		"dumps a TCP endpoint internal state");
802 
803 	return B_OK;
804 }
805 
806 
807 static status_t
808 tcp_uninit()
809 {
810 	remove_debugger_command("tcp_endpoint", dump_endpoint);
811 	remove_debugger_command("tcp_endpoints", dump_endpoints);
812 
813 	rw_lock_destroy(&sEndpointManagersLock);
814 
815 	for (int i = 0; i < AF_MAX; i++) {
816 		delete sEndpointManagers[i];
817 	}
818 
819 	return B_OK;
820 }
821 
822 
823 static status_t
824 tcp_std_ops(int32 op, ...)
825 {
826 	switch (op) {
827 		case B_MODULE_INIT:
828 			return tcp_init();
829 
830 		case B_MODULE_UNINIT:
831 			return tcp_uninit();
832 
833 		default:
834 			return B_ERROR;
835 	}
836 }
837 
838 
839 net_protocol_module_info sTCPModule = {
840 	{
841 		"network/protocols/tcp/v1",
842 		0,
843 		tcp_std_ops
844 	},
845 	0,
846 
847 	tcp_init_protocol,
848 	tcp_uninit_protocol,
849 	tcp_open,
850 	tcp_close,
851 	tcp_free,
852 	tcp_connect,
853 	tcp_accept,
854 	tcp_control,
855 	tcp_getsockopt,
856 	tcp_setsockopt,
857 	tcp_bind,
858 	tcp_unbind,
859 	tcp_listen,
860 	tcp_shutdown,
861 	tcp_send_data,
862 	tcp_send_routed_data,
863 	tcp_send_avail,
864 	tcp_read_data,
865 	tcp_read_avail,
866 	tcp_get_domain,
867 	tcp_get_mtu,
868 	tcp_receive_data,
869 	NULL,		// deliver_data()
870 	tcp_error_received,
871 	tcp_error_reply,
872 	NULL,		// add_ancillary_data()
873 	NULL,		// process_ancillary_data()
874 	NULL,		// process_ancillary_data_no_container()
875 	NULL,		// send_data_no_buffer()
876 	NULL		// read_data_no_buffer()
877 };
878 
879 module_dependency module_dependencies[] = {
880 	{NET_STACK_MODULE_NAME, (module_info **)&gStackModule},
881 	{NET_BUFFER_MODULE_NAME, (module_info **)&gBufferModule},
882 	{NET_DATALINK_MODULE_NAME, (module_info **)&gDatalinkModule},
883 	{NET_SOCKET_MODULE_NAME, (module_info **)&gSocketModule},
884 	{}
885 };
886 
887 module_info *modules[] = {
888 	(module_info *)&sTCPModule,
889 	NULL
890 };
891