1 /*
2 * Copyright 2004 The WebRTC Project Authors. All rights reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #include "p2p/base/pseudo_tcp.h"
12
13 #include <errno.h>
14 #include <stdio.h>
15 #include <string.h>
16
17 #include <algorithm>
18 #include <cstdint>
19 #include <memory>
20 #include <set>
21
22 #include "rtc_base/byte_buffer.h"
23 #include "rtc_base/byte_order.h"
24 #include "rtc_base/checks.h"
25 #include "rtc_base/logging.h"
26 #include "rtc_base/numerics/safe_minmax.h"
27 #include "rtc_base/socket.h"
28 #include "rtc_base/time_utils.h"
29
30 // The following logging is for detailed (packet-level) analysis only.
31 #define _DBG_NONE 0
32 #define _DBG_NORMAL 1
33 #define _DBG_VERBOSE 2
34 #define _DEBUGMSG _DBG_NONE
35
36 namespace cricket {
37
38 //////////////////////////////////////////////////////////////////////
39 // Network Constants
40 //////////////////////////////////////////////////////////////////////
41
42 // Standard MTUs
43 const uint16_t PACKET_MAXIMUMS[] = {
44 65535, // Theoretical maximum, Hyperchannel
45 32000, // Nothing
46 17914, // 16Mb IBM Token Ring
47 8166, // IEEE 802.4
48 // 4464, // IEEE 802.5 (4Mb max)
49 4352, // FDDI
50 // 2048, // Wideband Network
51 2002, // IEEE 802.5 (4Mb recommended)
52 // 1536, // Expermental Ethernet Networks
53 // 1500, // Ethernet, Point-to-Point (default)
54 1492, // IEEE 802.3
55 1006, // SLIP, ARPANET
56 // 576, // X.25 Networks
57 // 544, // DEC IP Portal
58 // 512, // NETBIOS
59 508, // IEEE 802/Source-Rt Bridge, ARCNET
60 296, // Point-to-Point (low delay)
61 // 68, // Official minimum
62 0, // End of list marker
63 };
64
65 const uint32_t MAX_PACKET = 65535;
66 // Note: we removed lowest level because packet overhead was larger!
67 const uint32_t MIN_PACKET = 296;
68
69 const uint32_t IP_HEADER_SIZE = 20; // (+ up to 40 bytes of options?)
70 const uint32_t UDP_HEADER_SIZE = 8;
71 // TODO(?): Make JINGLE_HEADER_SIZE transparent to this code?
72 const uint32_t JINGLE_HEADER_SIZE = 64; // when relay framing is in use
73
74 // Default size for receive and send buffer.
75 const uint32_t DEFAULT_RCV_BUF_SIZE = 60 * 1024;
76 const uint32_t DEFAULT_SND_BUF_SIZE = 90 * 1024;
77
78 //////////////////////////////////////////////////////////////////////
79 // Global Constants and Functions
80 //////////////////////////////////////////////////////////////////////
81 //
82 // 0 1 2 3
83 // 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
84 // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
85 // 0 | Conversation Number |
86 // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
87 // 4 | Sequence Number |
88 // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
89 // 8 | Acknowledgment Number |
90 // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
91 // | | |U|A|P|R|S|F| |
92 // 12 | Control | |R|C|S|S|Y|I| Window |
93 // | | |G|K|H|T|N|N| |
94 // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
95 // 16 | Timestamp sending |
96 // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
97 // 20 | Timestamp receiving |
98 // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
99 // 24 | data |
100 // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
101 //
102 //////////////////////////////////////////////////////////////////////
103
104 #define PSEUDO_KEEPALIVE 0
105
106 const uint32_t HEADER_SIZE = 24;
107 const uint32_t PACKET_OVERHEAD =
108 HEADER_SIZE + UDP_HEADER_SIZE + IP_HEADER_SIZE + JINGLE_HEADER_SIZE;
109
110 const uint32_t MIN_RTO =
111 250; // 250 ms (RFC1122, Sec 4.2.3.1 "fractions of a second")
112 const uint32_t DEF_RTO = 3000; // 3 seconds (RFC1122, Sec 4.2.3.1)
113 const uint32_t MAX_RTO = 60000; // 60 seconds
114 const uint32_t DEF_ACK_DELAY = 100; // 100 milliseconds
115
116 const uint8_t FLAG_CTL = 0x02;
117 const uint8_t FLAG_RST = 0x04;
118
119 const uint8_t CTL_CONNECT = 0;
120
121 // TCP options.
122 const uint8_t TCP_OPT_EOL = 0; // End of list.
123 const uint8_t TCP_OPT_NOOP = 1; // No-op.
124 const uint8_t TCP_OPT_MSS = 2; // Maximum segment size.
125 const uint8_t TCP_OPT_WND_SCALE = 3; // Window scale factor.
126
127 const long DEFAULT_TIMEOUT =
128 4000; // If there are no pending clocks, wake up every 4 seconds
129 const long CLOSED_TIMEOUT =
130 60 * 1000; // If the connection is closed, once per minute
131
132 #if PSEUDO_KEEPALIVE
133 // !?! Rethink these times
134 const uint32_t IDLE_PING =
135 20 *
136 1000; // 20 seconds (note: WinXP SP2 firewall udp timeout is 90 seconds)
137 const uint32_t IDLE_TIMEOUT = 90 * 1000; // 90 seconds;
138 #endif // PSEUDO_KEEPALIVE
139
140 //////////////////////////////////////////////////////////////////////
141 // Helper Functions
142 //////////////////////////////////////////////////////////////////////
143
long_to_bytes(uint32_t val,void * buf)144 inline void long_to_bytes(uint32_t val, void* buf) {
145 *static_cast<uint32_t*>(buf) = rtc::HostToNetwork32(val);
146 }
147
short_to_bytes(uint16_t val,void * buf)148 inline void short_to_bytes(uint16_t val, void* buf) {
149 *static_cast<uint16_t*>(buf) = rtc::HostToNetwork16(val);
150 }
151
bytes_to_long(const void * buf)152 inline uint32_t bytes_to_long(const void* buf) {
153 return rtc::NetworkToHost32(*static_cast<const uint32_t*>(buf));
154 }
155
bytes_to_short(const void * buf)156 inline uint16_t bytes_to_short(const void* buf) {
157 return rtc::NetworkToHost16(*static_cast<const uint16_t*>(buf));
158 }
159
160 //////////////////////////////////////////////////////////////////////
161 // Debugging Statistics
162 //////////////////////////////////////////////////////////////////////
163
164 #if 0 // Not used yet
165
166 enum Stat {
167 S_SENT_PACKET, // All packet sends
168 S_RESENT_PACKET, // All packet sends that are retransmits
169 S_RECV_PACKET, // All packet receives
170 S_RECV_NEW, // All packet receives that are too new
171 S_RECV_OLD, // All packet receives that are too old
172 S_NUM_STATS
173 };
174
175 const char* const STAT_NAMES[S_NUM_STATS] = {
176 "snt",
177 "snt-r",
178 "rcv"
179 "rcv-n",
180 "rcv-o"
181 };
182
183 int g_stats[S_NUM_STATS];
184 inline void Incr(Stat s) { ++g_stats[s]; }
185 void ReportStats() {
186 char buffer[256];
187 size_t len = 0;
188 for (int i = 0; i < S_NUM_STATS; ++i) {
189 len += snprintf(buffer, arraysize(buffer), "%s%s:%d",
190 (i == 0) ? "" : ",", STAT_NAMES[i], g_stats[i]);
191 g_stats[i] = 0;
192 }
193 RTC_LOG(LS_INFO) << "Stats[" << buffer << "]";
194 }
195
196 #endif
197
198 //////////////////////////////////////////////////////////////////////
199 // PseudoTcp
200 //////////////////////////////////////////////////////////////////////
201
Now()202 uint32_t PseudoTcp::Now() {
203 #if 0 // Use this to synchronize timers with logging timestamps (easier debug)
204 return static_cast<uint32_t>(rtc::TimeSince(StartTime()));
205 #else
206 return rtc::Time32();
207 #endif
208 }
209
PseudoTcp(IPseudoTcpNotify * notify,uint32_t conv)210 PseudoTcp::PseudoTcp(IPseudoTcpNotify* notify, uint32_t conv)
211 : m_notify(notify),
212 m_shutdown(SD_NONE),
213 m_error(0),
214 m_rbuf_len(DEFAULT_RCV_BUF_SIZE),
215 m_rbuf(m_rbuf_len),
216 m_sbuf_len(DEFAULT_SND_BUF_SIZE),
217 m_sbuf(m_sbuf_len) {
218 // Sanity check on buffer sizes (needed for OnTcpWriteable notification logic)
219 RTC_DCHECK(m_rbuf_len + MIN_PACKET < m_sbuf_len);
220
221 uint32_t now = Now();
222
223 m_state = TCP_LISTEN;
224 m_conv = conv;
225 m_rcv_wnd = m_rbuf_len;
226 m_rwnd_scale = m_swnd_scale = 0;
227 m_snd_nxt = 0;
228 m_snd_wnd = 1;
229 m_snd_una = m_rcv_nxt = 0;
230 m_bReadEnable = true;
231 m_bWriteEnable = false;
232 m_t_ack = 0;
233
234 m_msslevel = 0;
235 m_largest = 0;
236 RTC_DCHECK(MIN_PACKET > PACKET_OVERHEAD);
237 m_mss = MIN_PACKET - PACKET_OVERHEAD;
238 m_mtu_advise = MAX_PACKET;
239
240 m_rto_base = 0;
241
242 m_cwnd = 2 * m_mss;
243 m_ssthresh = m_rbuf_len;
244 m_lastrecv = m_lastsend = m_lasttraffic = now;
245 m_bOutgoing = false;
246
247 m_dup_acks = 0;
248 m_recover = 0;
249
250 m_ts_recent = m_ts_lastack = 0;
251
252 m_rx_rto = DEF_RTO;
253 m_rx_srtt = m_rx_rttvar = 0;
254
255 m_use_nagling = true;
256 m_ack_delay = DEF_ACK_DELAY;
257 m_support_wnd_scale = true;
258 }
259
~PseudoTcp()260 PseudoTcp::~PseudoTcp() {}
261
Connect()262 int PseudoTcp::Connect() {
263 if (m_state != TCP_LISTEN) {
264 m_error = EINVAL;
265 return -1;
266 }
267
268 m_state = TCP_SYN_SENT;
269 RTC_LOG(LS_INFO) << "State: TCP_SYN_SENT";
270
271 queueConnectMessage();
272 attemptSend();
273
274 return 0;
275 }
276
NotifyMTU(uint16_t mtu)277 void PseudoTcp::NotifyMTU(uint16_t mtu) {
278 m_mtu_advise = mtu;
279 if (m_state == TCP_ESTABLISHED) {
280 adjustMTU();
281 }
282 }
283
NotifyClock(uint32_t now)284 void PseudoTcp::NotifyClock(uint32_t now) {
285 if (m_state == TCP_CLOSED)
286 return;
287
288 // Check if it's time to retransmit a segment
289 if (m_rto_base && (rtc::TimeDiff32(m_rto_base + m_rx_rto, now) <= 0)) {
290 if (m_slist.empty()) {
291 RTC_NOTREACHED();
292 } else {
293 // Note: (m_slist.front().xmit == 0)) {
294 // retransmit segments
295 #if _DEBUGMSG >= _DBG_NORMAL
296 RTC_LOG(LS_INFO) << "timeout retransmit (rto: " << m_rx_rto
297 << ") (rto_base: " << m_rto_base << ") (now: " << now
298 << ") (dup_acks: " << static_cast<unsigned>(m_dup_acks)
299 << ")";
300 #endif // _DEBUGMSG
301 if (!transmit(m_slist.begin(), now)) {
302 closedown(ECONNABORTED);
303 return;
304 }
305
306 uint32_t nInFlight = m_snd_nxt - m_snd_una;
307 m_ssthresh = std::max(nInFlight / 2, 2 * m_mss);
308 // RTC_LOG(LS_INFO) << "m_ssthresh: " << m_ssthresh << " nInFlight: " <<
309 // nInFlight << " m_mss: " << m_mss;
310 m_cwnd = m_mss;
311
312 // Back off retransmit timer. Note: the limit is lower when connecting.
313 uint32_t rto_limit = (m_state < TCP_ESTABLISHED) ? DEF_RTO : MAX_RTO;
314 m_rx_rto = std::min(rto_limit, m_rx_rto * 2);
315 m_rto_base = now;
316 }
317 }
318
319 // Check if it's time to probe closed windows
320 if ((m_snd_wnd == 0) && (rtc::TimeDiff32(m_lastsend + m_rx_rto, now) <= 0)) {
321 if (rtc::TimeDiff32(now, m_lastrecv) >= 15000) {
322 closedown(ECONNABORTED);
323 return;
324 }
325
326 // probe the window
327 packet(m_snd_nxt - 1, 0, 0, 0);
328 m_lastsend = now;
329
330 // back off retransmit timer
331 m_rx_rto = std::min(MAX_RTO, m_rx_rto * 2);
332 }
333
334 // Check if it's time to send delayed acks
335 if (m_t_ack && (rtc::TimeDiff32(m_t_ack + m_ack_delay, now) <= 0)) {
336 packet(m_snd_nxt, 0, 0, 0);
337 }
338
339 #if PSEUDO_KEEPALIVE
340 // Check for idle timeout
341 if ((m_state == TCP_ESTABLISHED) &&
342 (TimeDiff32(m_lastrecv + IDLE_TIMEOUT, now) <= 0)) {
343 closedown(ECONNABORTED);
344 return;
345 }
346
347 // Check for ping timeout (to keep udp mapping open)
348 if ((m_state == TCP_ESTABLISHED) &&
349 (TimeDiff32(m_lasttraffic + (m_bOutgoing ? IDLE_PING * 3 / 2 : IDLE_PING),
350 now) <= 0)) {
351 packet(m_snd_nxt, 0, 0, 0);
352 }
353 #endif // PSEUDO_KEEPALIVE
354 }
355
NotifyPacket(const char * buffer,size_t len)356 bool PseudoTcp::NotifyPacket(const char* buffer, size_t len) {
357 if (len > MAX_PACKET) {
358 RTC_LOG_F(WARNING) << "packet too large";
359 return false;
360 }
361 return parse(reinterpret_cast<const uint8_t*>(buffer), uint32_t(len));
362 }
363
GetNextClock(uint32_t now,long & timeout)364 bool PseudoTcp::GetNextClock(uint32_t now, long& timeout) {
365 return clock_check(now, timeout);
366 }
367
GetOption(Option opt,int * value)368 void PseudoTcp::GetOption(Option opt, int* value) {
369 if (opt == OPT_NODELAY) {
370 *value = m_use_nagling ? 0 : 1;
371 } else if (opt == OPT_ACKDELAY) {
372 *value = m_ack_delay;
373 } else if (opt == OPT_SNDBUF) {
374 *value = m_sbuf_len;
375 } else if (opt == OPT_RCVBUF) {
376 *value = m_rbuf_len;
377 } else {
378 RTC_NOTREACHED();
379 }
380 }
SetOption(Option opt,int value)381 void PseudoTcp::SetOption(Option opt, int value) {
382 if (opt == OPT_NODELAY) {
383 m_use_nagling = value == 0;
384 } else if (opt == OPT_ACKDELAY) {
385 m_ack_delay = value;
386 } else if (opt == OPT_SNDBUF) {
387 RTC_DCHECK(m_state == TCP_LISTEN);
388 resizeSendBuffer(value);
389 } else if (opt == OPT_RCVBUF) {
390 RTC_DCHECK(m_state == TCP_LISTEN);
391 resizeReceiveBuffer(value);
392 } else {
393 RTC_NOTREACHED();
394 }
395 }
396
GetCongestionWindow() const397 uint32_t PseudoTcp::GetCongestionWindow() const {
398 return m_cwnd;
399 }
400
GetBytesInFlight() const401 uint32_t PseudoTcp::GetBytesInFlight() const {
402 return m_snd_nxt - m_snd_una;
403 }
404
GetBytesBufferedNotSent() const405 uint32_t PseudoTcp::GetBytesBufferedNotSent() const {
406 return static_cast<uint32_t>(m_snd_una + m_sbuf.GetBuffered() - m_snd_nxt);
407 }
408
GetRoundTripTimeEstimateMs() const409 uint32_t PseudoTcp::GetRoundTripTimeEstimateMs() const {
410 return m_rx_srtt;
411 }
412
413 //
414 // IPStream Implementation
415 //
416
Recv(char * buffer,size_t len)417 int PseudoTcp::Recv(char* buffer, size_t len) {
418 if (m_state != TCP_ESTABLISHED) {
419 m_error = ENOTCONN;
420 return SOCKET_ERROR;
421 }
422
423 size_t read = 0;
424 if (!m_rbuf.Read(buffer, len, &read)) {
425 m_bReadEnable = true;
426 m_error = EWOULDBLOCK;
427 return SOCKET_ERROR;
428 }
429
430 size_t available_space = 0;
431 m_rbuf.GetWriteRemaining(&available_space);
432
433 if (uint32_t(available_space) - m_rcv_wnd >=
434 std::min<uint32_t>(m_rbuf_len / 2, m_mss)) {
435 // TODO(jbeda): !?! Not sure about this was closed business
436 bool bWasClosed = (m_rcv_wnd == 0);
437 m_rcv_wnd = static_cast<uint32_t>(available_space);
438
439 if (bWasClosed) {
440 attemptSend(sfImmediateAck);
441 }
442 }
443
444 return static_cast<int>(read);
445 }
446
Send(const char * buffer,size_t len)447 int PseudoTcp::Send(const char* buffer, size_t len) {
448 if (m_state != TCP_ESTABLISHED) {
449 m_error = ENOTCONN;
450 return SOCKET_ERROR;
451 }
452
453 size_t available_space = 0;
454 m_sbuf.GetWriteRemaining(&available_space);
455
456 if (!available_space) {
457 m_bWriteEnable = true;
458 m_error = EWOULDBLOCK;
459 return SOCKET_ERROR;
460 }
461
462 int written = queue(buffer, uint32_t(len), false);
463 attemptSend();
464 return written;
465 }
466
Close(bool force)467 void PseudoTcp::Close(bool force) {
468 RTC_LOG_F(LS_VERBOSE) << "(" << (force ? "true" : "false") << ")";
469 m_shutdown = force ? SD_FORCEFUL : SD_GRACEFUL;
470 }
471
GetError()472 int PseudoTcp::GetError() {
473 return m_error;
474 }
475
476 //
477 // Internal Implementation
478 //
479
queue(const char * data,uint32_t len,bool bCtrl)480 uint32_t PseudoTcp::queue(const char* data, uint32_t len, bool bCtrl) {
481 size_t available_space = 0;
482 m_sbuf.GetWriteRemaining(&available_space);
483
484 if (len > static_cast<uint32_t>(available_space)) {
485 RTC_DCHECK(!bCtrl);
486 len = static_cast<uint32_t>(available_space);
487 }
488
489 // We can concatenate data if the last segment is the same type
490 // (control v. regular data), and has not been transmitted yet
491 if (!m_slist.empty() && (m_slist.back().bCtrl == bCtrl) &&
492 (m_slist.back().xmit == 0)) {
493 m_slist.back().len += len;
494 } else {
495 SSegment sseg(static_cast<uint32_t>(m_snd_una + m_sbuf.GetBuffered()), len,
496 bCtrl);
497 m_slist.push_back(sseg);
498 }
499
500 size_t written = 0;
501 m_sbuf.Write(data, len, &written);
502 return static_cast<uint32_t>(written);
503 }
504
packet(uint32_t seq,uint8_t flags,uint32_t offset,uint32_t len)505 IPseudoTcpNotify::WriteResult PseudoTcp::packet(uint32_t seq,
506 uint8_t flags,
507 uint32_t offset,
508 uint32_t len) {
509 RTC_DCHECK(HEADER_SIZE + len <= MAX_PACKET);
510
511 uint32_t now = Now();
512
513 std::unique_ptr<uint8_t[]> buffer(new uint8_t[MAX_PACKET]);
514 long_to_bytes(m_conv, buffer.get());
515 long_to_bytes(seq, buffer.get() + 4);
516 long_to_bytes(m_rcv_nxt, buffer.get() + 8);
517 buffer[12] = 0;
518 buffer[13] = flags;
519 short_to_bytes(static_cast<uint16_t>(m_rcv_wnd >> m_rwnd_scale),
520 buffer.get() + 14);
521
522 // Timestamp computations
523 long_to_bytes(now, buffer.get() + 16);
524 long_to_bytes(m_ts_recent, buffer.get() + 20);
525 m_ts_lastack = m_rcv_nxt;
526
527 if (len) {
528 size_t bytes_read = 0;
529 bool result =
530 m_sbuf.ReadOffset(buffer.get() + HEADER_SIZE, len, offset, &bytes_read);
531 RTC_DCHECK(result);
532 RTC_DCHECK(static_cast<uint32_t>(bytes_read) == len);
533 }
534
535 #if _DEBUGMSG >= _DBG_VERBOSE
536 RTC_LOG(LS_INFO) << "<-- <CONV=" << m_conv
537 << "><FLG=" << static_cast<unsigned>(flags)
538 << "><SEQ=" << seq << ":" << seq + len
539 << "><ACK=" << m_rcv_nxt << "><WND=" << m_rcv_wnd
540 << "><TS=" << (now % 10000)
541 << "><TSR=" << (m_ts_recent % 10000) << "><LEN=" << len
542 << ">";
543 #endif // _DEBUGMSG
544
545 IPseudoTcpNotify::WriteResult wres = m_notify->TcpWritePacket(
546 this, reinterpret_cast<char*>(buffer.get()), len + HEADER_SIZE);
547 // Note: When len is 0, this is an ACK packet. We don't read the return value
548 // for those, and thus we won't retry. So go ahead and treat the packet as a
549 // success (basically simulate as if it were dropped), which will prevent our
550 // timers from being messed up.
551 if ((wres != IPseudoTcpNotify::WR_SUCCESS) && (0 != len))
552 return wres;
553
554 m_t_ack = 0;
555 if (len > 0) {
556 m_lastsend = now;
557 }
558 m_lasttraffic = now;
559 m_bOutgoing = true;
560
561 return IPseudoTcpNotify::WR_SUCCESS;
562 }
563
parse(const uint8_t * buffer,uint32_t size)564 bool PseudoTcp::parse(const uint8_t* buffer, uint32_t size) {
565 if (size < HEADER_SIZE)
566 return false;
567
568 Segment seg;
569 seg.conv = bytes_to_long(buffer);
570 seg.seq = bytes_to_long(buffer + 4);
571 seg.ack = bytes_to_long(buffer + 8);
572 seg.flags = buffer[13];
573 seg.wnd = bytes_to_short(buffer + 14);
574
575 seg.tsval = bytes_to_long(buffer + 16);
576 seg.tsecr = bytes_to_long(buffer + 20);
577
578 seg.data = reinterpret_cast<const char*>(buffer) + HEADER_SIZE;
579 seg.len = size - HEADER_SIZE;
580
581 #if _DEBUGMSG >= _DBG_VERBOSE
582 RTC_LOG(LS_INFO) << "--> <CONV=" << seg.conv
583 << "><FLG=" << static_cast<unsigned>(seg.flags)
584 << "><SEQ=" << seg.seq << ":" << seg.seq + seg.len
585 << "><ACK=" << seg.ack << "><WND=" << seg.wnd
586 << "><TS=" << (seg.tsval % 10000)
587 << "><TSR=" << (seg.tsecr % 10000) << "><LEN=" << seg.len
588 << ">";
589 #endif // _DEBUGMSG
590
591 return process(seg);
592 }
593
clock_check(uint32_t now,long & nTimeout)594 bool PseudoTcp::clock_check(uint32_t now, long& nTimeout) {
595 if (m_shutdown == SD_FORCEFUL)
596 return false;
597
598 if ((m_shutdown == SD_GRACEFUL) &&
599 ((m_state != TCP_ESTABLISHED) ||
600 ((m_sbuf.GetBuffered() == 0) && (m_t_ack == 0)))) {
601 return false;
602 }
603
604 if (m_state == TCP_CLOSED) {
605 nTimeout = CLOSED_TIMEOUT;
606 return true;
607 }
608
609 nTimeout = DEFAULT_TIMEOUT;
610
611 if (m_t_ack) {
612 nTimeout = std::min<int32_t>(nTimeout,
613 rtc::TimeDiff32(m_t_ack + m_ack_delay, now));
614 }
615 if (m_rto_base) {
616 nTimeout = std::min<int32_t>(nTimeout,
617 rtc::TimeDiff32(m_rto_base + m_rx_rto, now));
618 }
619 if (m_snd_wnd == 0) {
620 nTimeout = std::min<int32_t>(nTimeout,
621 rtc::TimeDiff32(m_lastsend + m_rx_rto, now));
622 }
623 #if PSEUDO_KEEPALIVE
624 if (m_state == TCP_ESTABLISHED) {
625 nTimeout = std::min<int32_t>(
626 nTimeout,
627 rtc::TimeDiff32(
628 m_lasttraffic + (m_bOutgoing ? IDLE_PING * 3 / 2 : IDLE_PING),
629 now));
630 }
631 #endif // PSEUDO_KEEPALIVE
632 return true;
633 }
634
process(Segment & seg)635 bool PseudoTcp::process(Segment& seg) {
636 // If this is the wrong conversation, send a reset!?! (with the correct
637 // conversation?)
638 if (seg.conv != m_conv) {
639 // if ((seg.flags & FLAG_RST) == 0) {
640 // packet(tcb, seg.ack, 0, FLAG_RST, 0, 0);
641 //}
642 RTC_LOG_F(LS_ERROR) << "wrong conversation";
643 return false;
644 }
645
646 uint32_t now = Now();
647 m_lasttraffic = m_lastrecv = now;
648 m_bOutgoing = false;
649
650 if (m_state == TCP_CLOSED) {
651 // !?! send reset?
652 RTC_LOG_F(LS_ERROR) << "closed";
653 return false;
654 }
655
656 // Check if this is a reset segment
657 if (seg.flags & FLAG_RST) {
658 closedown(ECONNRESET);
659 return false;
660 }
661
662 // Check for control data
663 bool bConnect = false;
664 if (seg.flags & FLAG_CTL) {
665 if (seg.len == 0) {
666 RTC_LOG_F(LS_ERROR) << "Missing control code";
667 return false;
668 } else if (seg.data[0] == CTL_CONNECT) {
669 bConnect = true;
670
671 // TCP options are in the remainder of the payload after CTL_CONNECT.
672 parseOptions(&seg.data[1], seg.len - 1);
673
674 if (m_state == TCP_LISTEN) {
675 m_state = TCP_SYN_RECEIVED;
676 RTC_LOG(LS_INFO) << "State: TCP_SYN_RECEIVED";
677 // m_notify->associate(addr);
678 queueConnectMessage();
679 } else if (m_state == TCP_SYN_SENT) {
680 m_state = TCP_ESTABLISHED;
681 RTC_LOG(LS_INFO) << "State: TCP_ESTABLISHED";
682 adjustMTU();
683 if (m_notify) {
684 m_notify->OnTcpOpen(this);
685 }
686 // notify(evOpen);
687 }
688 } else {
689 RTC_LOG_F(LS_WARNING) << "Unknown control code: " << seg.data[0];
690 return false;
691 }
692 }
693
694 // Update timestamp
695 if ((seg.seq <= m_ts_lastack) && (m_ts_lastack < seg.seq + seg.len)) {
696 m_ts_recent = seg.tsval;
697 }
698
699 // Check if this is a valuable ack
700 if ((seg.ack > m_snd_una) && (seg.ack <= m_snd_nxt)) {
701 // Calculate round-trip time
702 if (seg.tsecr) {
703 int32_t rtt = rtc::TimeDiff32(now, seg.tsecr);
704 if (rtt >= 0) {
705 if (m_rx_srtt == 0) {
706 m_rx_srtt = rtt;
707 m_rx_rttvar = rtt / 2;
708 } else {
709 uint32_t unsigned_rtt = static_cast<uint32_t>(rtt);
710 uint32_t abs_err = unsigned_rtt > m_rx_srtt
711 ? unsigned_rtt - m_rx_srtt
712 : m_rx_srtt - unsigned_rtt;
713 m_rx_rttvar = (3 * m_rx_rttvar + abs_err) / 4;
714 m_rx_srtt = (7 * m_rx_srtt + rtt) / 8;
715 }
716 m_rx_rto = rtc::SafeClamp(m_rx_srtt + rtc::SafeMax(1, 4 * m_rx_rttvar),
717 MIN_RTO, MAX_RTO);
718 #if _DEBUGMSG >= _DBG_VERBOSE
719 RTC_LOG(LS_INFO) << "rtt: " << rtt << " srtt: " << m_rx_srtt
720 << " rto: " << m_rx_rto;
721 #endif // _DEBUGMSG
722 } else {
723 RTC_LOG(LS_WARNING) << "rtt < 0";
724 }
725 }
726
727 m_snd_wnd = static_cast<uint32_t>(seg.wnd) << m_swnd_scale;
728
729 uint32_t nAcked = seg.ack - m_snd_una;
730 m_snd_una = seg.ack;
731
732 m_rto_base = (m_snd_una == m_snd_nxt) ? 0 : now;
733
734 m_sbuf.ConsumeReadData(nAcked);
735
736 for (uint32_t nFree = nAcked; nFree > 0;) {
737 RTC_DCHECK(!m_slist.empty());
738 if (nFree < m_slist.front().len) {
739 m_slist.front().len -= nFree;
740 nFree = 0;
741 } else {
742 if (m_slist.front().len > m_largest) {
743 m_largest = m_slist.front().len;
744 }
745 nFree -= m_slist.front().len;
746 m_slist.pop_front();
747 }
748 }
749
750 if (m_dup_acks >= 3) {
751 if (m_snd_una >= m_recover) { // NewReno
752 uint32_t nInFlight = m_snd_nxt - m_snd_una;
753 m_cwnd = std::min(m_ssthresh, nInFlight + m_mss); // (Fast Retransmit)
754 #if _DEBUGMSG >= _DBG_NORMAL
755 RTC_LOG(LS_INFO) << "exit recovery";
756 #endif // _DEBUGMSG
757 m_dup_acks = 0;
758 } else {
759 #if _DEBUGMSG >= _DBG_NORMAL
760 RTC_LOG(LS_INFO) << "recovery retransmit";
761 #endif // _DEBUGMSG
762 if (!transmit(m_slist.begin(), now)) {
763 closedown(ECONNABORTED);
764 return false;
765 }
766 m_cwnd += m_mss - std::min(nAcked, m_cwnd);
767 }
768 } else {
769 m_dup_acks = 0;
770 // Slow start, congestion avoidance
771 if (m_cwnd < m_ssthresh) {
772 m_cwnd += m_mss;
773 } else {
774 m_cwnd += std::max<uint32_t>(1, m_mss * m_mss / m_cwnd);
775 }
776 }
777 } else if (seg.ack == m_snd_una) {
778 // !?! Note, tcp says don't do this... but otherwise how does a closed
779 // window become open?
780 m_snd_wnd = static_cast<uint32_t>(seg.wnd) << m_swnd_scale;
781
782 // Check duplicate acks
783 if (seg.len > 0) {
784 // it's a dup ack, but with a data payload, so don't modify m_dup_acks
785 } else if (m_snd_una != m_snd_nxt) {
786 m_dup_acks += 1;
787 if (m_dup_acks == 3) { // (Fast Retransmit)
788 #if _DEBUGMSG >= _DBG_NORMAL
789 RTC_LOG(LS_INFO) << "enter recovery";
790 RTC_LOG(LS_INFO) << "recovery retransmit";
791 #endif // _DEBUGMSG
792 if (!transmit(m_slist.begin(), now)) {
793 closedown(ECONNABORTED);
794 return false;
795 }
796 m_recover = m_snd_nxt;
797 uint32_t nInFlight = m_snd_nxt - m_snd_una;
798 m_ssthresh = std::max(nInFlight / 2, 2 * m_mss);
799 // RTC_LOG(LS_INFO) << "m_ssthresh: " << m_ssthresh << " nInFlight: "
800 // << nInFlight << " m_mss: " << m_mss;
801 m_cwnd = m_ssthresh + 3 * m_mss;
802 } else if (m_dup_acks > 3) {
803 m_cwnd += m_mss;
804 }
805 } else {
806 m_dup_acks = 0;
807 }
808 }
809
810 // !?! A bit hacky
811 if ((m_state == TCP_SYN_RECEIVED) && !bConnect) {
812 m_state = TCP_ESTABLISHED;
813 RTC_LOG(LS_INFO) << "State: TCP_ESTABLISHED";
814 adjustMTU();
815 if (m_notify) {
816 m_notify->OnTcpOpen(this);
817 }
818 // notify(evOpen);
819 }
820
821 // If we make room in the send queue, notify the user
822 // The goal it to make sure we always have at least enough data to fill the
823 // window. We'd like to notify the app when we are halfway to that point.
824 const uint32_t kIdealRefillSize = (m_sbuf_len + m_rbuf_len) / 2;
825 if (m_bWriteEnable &&
826 static_cast<uint32_t>(m_sbuf.GetBuffered()) < kIdealRefillSize) {
827 m_bWriteEnable = false;
828 if (m_notify) {
829 m_notify->OnTcpWriteable(this);
830 }
831 // notify(evWrite);
832 }
833
834 // Conditions were acks must be sent:
835 // 1) Segment is too old (they missed an ACK) (immediately)
836 // 2) Segment is too new (we missed a segment) (immediately)
837 // 3) Segment has data (so we need to ACK!) (delayed)
838 // ... so the only time we don't need to ACK, is an empty segment that points
839 // to rcv_nxt!
840
841 SendFlags sflags = sfNone;
842 if (seg.seq != m_rcv_nxt) {
843 sflags = sfImmediateAck; // (Fast Recovery)
844 } else if (seg.len != 0) {
845 if (m_ack_delay == 0) {
846 sflags = sfImmediateAck;
847 } else {
848 sflags = sfDelayedAck;
849 }
850 }
851 #if _DEBUGMSG >= _DBG_NORMAL
852 if (sflags == sfImmediateAck) {
853 if (seg.seq > m_rcv_nxt) {
854 RTC_LOG_F(LS_INFO) << "too new";
855 } else if (seg.seq + seg.len <= m_rcv_nxt) {
856 RTC_LOG_F(LS_INFO) << "too old";
857 }
858 }
859 #endif // _DEBUGMSG
860
861 // Adjust the incoming segment to fit our receive buffer
862 if (seg.seq < m_rcv_nxt) {
863 uint32_t nAdjust = m_rcv_nxt - seg.seq;
864 if (nAdjust < seg.len) {
865 seg.seq += nAdjust;
866 seg.data += nAdjust;
867 seg.len -= nAdjust;
868 } else {
869 seg.len = 0;
870 }
871 }
872
873 size_t available_space = 0;
874 m_rbuf.GetWriteRemaining(&available_space);
875
876 if ((seg.seq + seg.len - m_rcv_nxt) >
877 static_cast<uint32_t>(available_space)) {
878 uint32_t nAdjust =
879 seg.seq + seg.len - m_rcv_nxt - static_cast<uint32_t>(available_space);
880 if (nAdjust < seg.len) {
881 seg.len -= nAdjust;
882 } else {
883 seg.len = 0;
884 }
885 }
886
887 bool bIgnoreData = (seg.flags & FLAG_CTL) || (m_shutdown != SD_NONE);
888 bool bNewData = false;
889
890 if (seg.len > 0) {
891 bool bRecover = false;
892 if (bIgnoreData) {
893 if (seg.seq == m_rcv_nxt) {
894 m_rcv_nxt += seg.len;
895 // If we received a data segment out of order relative to a control
896 // segment, then we wrote it into the receive buffer at an offset (see
897 // "WriteOffset") below. So we need to advance the position in the
898 // buffer to avoid corrupting data. See bugs.webrtc.org/9208
899 //
900 // We advance the position in the buffer by N bytes by acting like we
901 // wrote N bytes and then immediately read them. We can only do this if
902 // there's not already data ready to read, but this should always be
903 // true in the problematic scenario, since control frames are always
904 // sent first in the stream.
905 if (m_rbuf.GetBuffered() == 0) {
906 m_rbuf.ConsumeWriteBuffer(seg.len);
907 m_rbuf.ConsumeReadData(seg.len);
908 // After shifting the position in the buffer, we may have
909 // out-of-order packets ready to be recovered.
910 bRecover = true;
911 }
912 }
913 } else {
914 uint32_t nOffset = seg.seq - m_rcv_nxt;
915
916 if (!m_rbuf.WriteOffset(seg.data, seg.len, nOffset, NULL)) {
917 // Ignore incoming packets outside of the receive window.
918 return false;
919 }
920
921 if (seg.seq == m_rcv_nxt) {
922 m_rbuf.ConsumeWriteBuffer(seg.len);
923 m_rcv_nxt += seg.len;
924 m_rcv_wnd -= seg.len;
925 bNewData = true;
926 // May be able to recover packets previously received out-of-order
927 // now.
928 bRecover = true;
929 } else {
930 #if _DEBUGMSG >= _DBG_NORMAL
931 RTC_LOG(LS_INFO) << "Saving " << seg.len << " bytes (" << seg.seq
932 << " -> " << seg.seq + seg.len << ")";
933 #endif // _DEBUGMSG
934 RSegment rseg;
935 rseg.seq = seg.seq;
936 rseg.len = seg.len;
937 RList::iterator it = m_rlist.begin();
938 while ((it != m_rlist.end()) && (it->seq < rseg.seq)) {
939 ++it;
940 }
941 m_rlist.insert(it, rseg);
942 }
943 }
944 if (bRecover) {
945 RList::iterator it = m_rlist.begin();
946 while ((it != m_rlist.end()) && (it->seq <= m_rcv_nxt)) {
947 if (it->seq + it->len > m_rcv_nxt) {
948 sflags = sfImmediateAck; // (Fast Recovery)
949 uint32_t nAdjust = (it->seq + it->len) - m_rcv_nxt;
950 #if _DEBUGMSG >= _DBG_NORMAL
951 RTC_LOG(LS_INFO) << "Recovered " << nAdjust << " bytes (" << m_rcv_nxt
952 << " -> " << m_rcv_nxt + nAdjust << ")";
953 #endif // _DEBUGMSG
954 m_rbuf.ConsumeWriteBuffer(nAdjust);
955 m_rcv_nxt += nAdjust;
956 m_rcv_wnd -= nAdjust;
957 bNewData = true;
958 }
959 it = m_rlist.erase(it);
960 }
961 }
962 }
963
964 attemptSend(sflags);
965
966 // If we have new data, notify the user
967 if (bNewData && m_bReadEnable) {
968 m_bReadEnable = false;
969 if (m_notify) {
970 m_notify->OnTcpReadable(this);
971 }
972 // notify(evRead);
973 }
974
975 return true;
976 }
977
transmit(const SList::iterator & seg,uint32_t now)978 bool PseudoTcp::transmit(const SList::iterator& seg, uint32_t now) {
979 if (seg->xmit >= ((m_state == TCP_ESTABLISHED) ? 15 : 30)) {
980 RTC_LOG_F(LS_VERBOSE) << "too many retransmits";
981 return false;
982 }
983
984 uint32_t nTransmit = std::min(seg->len, m_mss);
985
986 while (true) {
987 uint32_t seq = seg->seq;
988 uint8_t flags = (seg->bCtrl ? FLAG_CTL : 0);
989 IPseudoTcpNotify::WriteResult wres =
990 packet(seq, flags, seg->seq - m_snd_una, nTransmit);
991
992 if (wres == IPseudoTcpNotify::WR_SUCCESS)
993 break;
994
995 if (wres == IPseudoTcpNotify::WR_FAIL) {
996 RTC_LOG_F(LS_VERBOSE) << "packet failed";
997 return false;
998 }
999
1000 RTC_DCHECK(wres == IPseudoTcpNotify::WR_TOO_LARGE);
1001
1002 while (true) {
1003 if (PACKET_MAXIMUMS[m_msslevel + 1] == 0) {
1004 RTC_LOG_F(LS_VERBOSE) << "MTU too small";
1005 return false;
1006 }
1007 // !?! We need to break up all outstanding and pending packets and then
1008 // retransmit!?!
1009
1010 m_mss = PACKET_MAXIMUMS[++m_msslevel] - PACKET_OVERHEAD;
1011 m_cwnd = 2 * m_mss; // I added this... haven't researched actual formula
1012 if (m_mss < nTransmit) {
1013 nTransmit = m_mss;
1014 break;
1015 }
1016 }
1017 #if _DEBUGMSG >= _DBG_NORMAL
1018 RTC_LOG(LS_INFO) << "Adjusting mss to " << m_mss << " bytes";
1019 #endif // _DEBUGMSG
1020 }
1021
1022 if (nTransmit < seg->len) {
1023 RTC_LOG_F(LS_VERBOSE) << "mss reduced to " << m_mss;
1024
1025 SSegment subseg(seg->seq + nTransmit, seg->len - nTransmit, seg->bCtrl);
1026 // subseg.tstamp = seg->tstamp;
1027 subseg.xmit = seg->xmit;
1028 seg->len = nTransmit;
1029
1030 SList::iterator next = seg;
1031 m_slist.insert(++next, subseg);
1032 }
1033
1034 if (seg->xmit == 0) {
1035 m_snd_nxt += seg->len;
1036 }
1037 seg->xmit += 1;
1038 // seg->tstamp = now;
1039 if (m_rto_base == 0) {
1040 m_rto_base = now;
1041 }
1042
1043 return true;
1044 }
1045
attemptSend(SendFlags sflags)1046 void PseudoTcp::attemptSend(SendFlags sflags) {
1047 uint32_t now = Now();
1048
1049 if (rtc::TimeDiff32(now, m_lastsend) > static_cast<long>(m_rx_rto)) {
1050 m_cwnd = m_mss;
1051 }
1052
1053 #if _DEBUGMSG
1054 bool bFirst = true;
1055 #endif // _DEBUGMSG
1056
1057 while (true) {
1058 uint32_t cwnd = m_cwnd;
1059 if ((m_dup_acks == 1) || (m_dup_acks == 2)) { // Limited Transmit
1060 cwnd += m_dup_acks * m_mss;
1061 }
1062 uint32_t nWindow = std::min(m_snd_wnd, cwnd);
1063 uint32_t nInFlight = m_snd_nxt - m_snd_una;
1064 uint32_t nUseable = (nInFlight < nWindow) ? (nWindow - nInFlight) : 0;
1065
1066 size_t snd_buffered = m_sbuf.GetBuffered();
1067 uint32_t nAvailable =
1068 std::min(static_cast<uint32_t>(snd_buffered) - nInFlight, m_mss);
1069
1070 if (nAvailable > nUseable) {
1071 if (nUseable * 4 < nWindow) {
1072 // RFC 813 - avoid SWS
1073 nAvailable = 0;
1074 } else {
1075 nAvailable = nUseable;
1076 }
1077 }
1078
1079 #if _DEBUGMSG >= _DBG_VERBOSE
1080 if (bFirst) {
1081 size_t available_space = 0;
1082 m_sbuf.GetWriteRemaining(&available_space);
1083
1084 bFirst = false;
1085 RTC_LOG(LS_INFO) << "[cwnd: " << m_cwnd << " nWindow: " << nWindow
1086 << " nInFlight: " << nInFlight
1087 << " nAvailable: " << nAvailable
1088 << " nQueued: " << snd_buffered
1089 << " nEmpty: " << available_space
1090 << " ssthresh: " << m_ssthresh << "]";
1091 }
1092 #endif // _DEBUGMSG
1093
1094 if (nAvailable == 0) {
1095 if (sflags == sfNone)
1096 return;
1097
1098 // If this is an immediate ack, or the second delayed ack
1099 if ((sflags == sfImmediateAck) || m_t_ack) {
1100 packet(m_snd_nxt, 0, 0, 0);
1101 } else {
1102 m_t_ack = Now();
1103 }
1104 return;
1105 }
1106
1107 // Nagle's algorithm.
1108 // If there is data already in-flight, and we haven't a full segment of
1109 // data ready to send then hold off until we get more to send, or the
1110 // in-flight data is acknowledged.
1111 if (m_use_nagling && (m_snd_nxt > m_snd_una) && (nAvailable < m_mss)) {
1112 return;
1113 }
1114
1115 // Find the next segment to transmit
1116 SList::iterator it = m_slist.begin();
1117 while (it->xmit > 0) {
1118 ++it;
1119 RTC_DCHECK(it != m_slist.end());
1120 }
1121 SList::iterator seg = it;
1122
1123 // If the segment is too large, break it into two
1124 if (seg->len > nAvailable) {
1125 SSegment subseg(seg->seq + nAvailable, seg->len - nAvailable, seg->bCtrl);
1126 seg->len = nAvailable;
1127 m_slist.insert(++it, subseg);
1128 }
1129
1130 if (!transmit(seg, now)) {
1131 RTC_LOG_F(LS_VERBOSE) << "transmit failed";
1132 // TODO(?): consider closing socket
1133 return;
1134 }
1135
1136 sflags = sfNone;
1137 }
1138 }
1139
closedown(uint32_t err)1140 void PseudoTcp::closedown(uint32_t err) {
1141 RTC_LOG(LS_INFO) << "State: TCP_CLOSED";
1142 m_state = TCP_CLOSED;
1143 if (m_notify) {
1144 m_notify->OnTcpClosed(this, err);
1145 }
1146 // notify(evClose, err);
1147 }
1148
adjustMTU()1149 void PseudoTcp::adjustMTU() {
1150 // Determine our current mss level, so that we can adjust appropriately later
1151 for (m_msslevel = 0; PACKET_MAXIMUMS[m_msslevel + 1] > 0; ++m_msslevel) {
1152 if (static_cast<uint16_t>(PACKET_MAXIMUMS[m_msslevel]) <= m_mtu_advise) {
1153 break;
1154 }
1155 }
1156 m_mss = m_mtu_advise - PACKET_OVERHEAD;
1157 // !?! Should we reset m_largest here?
1158 #if _DEBUGMSG >= _DBG_NORMAL
1159 RTC_LOG(LS_INFO) << "Adjusting mss to " << m_mss << " bytes";
1160 #endif // _DEBUGMSG
1161 // Enforce minimums on ssthresh and cwnd
1162 m_ssthresh = std::max(m_ssthresh, 2 * m_mss);
1163 m_cwnd = std::max(m_cwnd, m_mss);
1164 }
1165
isReceiveBufferFull() const1166 bool PseudoTcp::isReceiveBufferFull() const {
1167 size_t available_space = 0;
1168 m_rbuf.GetWriteRemaining(&available_space);
1169 return !available_space;
1170 }
1171
disableWindowScale()1172 void PseudoTcp::disableWindowScale() {
1173 m_support_wnd_scale = false;
1174 }
1175
queueConnectMessage()1176 void PseudoTcp::queueConnectMessage() {
1177 rtc::ByteBufferWriter buf;
1178
1179 buf.WriteUInt8(CTL_CONNECT);
1180 if (m_support_wnd_scale) {
1181 buf.WriteUInt8(TCP_OPT_WND_SCALE);
1182 buf.WriteUInt8(1);
1183 buf.WriteUInt8(m_rwnd_scale);
1184 }
1185 m_snd_wnd = static_cast<uint32_t>(buf.Length());
1186 queue(buf.Data(), static_cast<uint32_t>(buf.Length()), true);
1187 }
1188
parseOptions(const char * data,uint32_t len)1189 void PseudoTcp::parseOptions(const char* data, uint32_t len) {
1190 std::set<uint8_t> options_specified;
1191
1192 // See http://www.freesoft.org/CIE/Course/Section4/8.htm for
1193 // parsing the options list.
1194 rtc::ByteBufferReader buf(data, len);
1195 while (buf.Length()) {
1196 uint8_t kind = TCP_OPT_EOL;
1197 buf.ReadUInt8(&kind);
1198
1199 if (kind == TCP_OPT_EOL) {
1200 // End of option list.
1201 break;
1202 } else if (kind == TCP_OPT_NOOP) {
1203 // No op.
1204 continue;
1205 }
1206
1207 // Length of this option.
1208 RTC_DCHECK(len != 0);
1209 uint8_t opt_len = 0;
1210 buf.ReadUInt8(&opt_len);
1211
1212 // Content of this option.
1213 if (opt_len <= buf.Length()) {
1214 applyOption(kind, buf.Data(), opt_len);
1215 buf.Consume(opt_len);
1216 } else {
1217 RTC_LOG(LS_ERROR) << "Invalid option length received.";
1218 return;
1219 }
1220 options_specified.insert(kind);
1221 }
1222
1223 if (options_specified.find(TCP_OPT_WND_SCALE) == options_specified.end()) {
1224 RTC_LOG(LS_WARNING) << "Peer doesn't support window scaling";
1225
1226 if (m_rwnd_scale > 0) {
1227 // Peer doesn't support TCP options and window scaling.
1228 // Revert receive buffer size to default value.
1229 resizeReceiveBuffer(DEFAULT_RCV_BUF_SIZE);
1230 m_swnd_scale = 0;
1231 }
1232 }
1233 }
1234
applyOption(char kind,const char * data,uint32_t len)1235 void PseudoTcp::applyOption(char kind, const char* data, uint32_t len) {
1236 if (kind == TCP_OPT_MSS) {
1237 RTC_LOG(LS_WARNING) << "Peer specified MSS option which is not supported.";
1238 // TODO(?): Implement.
1239 } else if (kind == TCP_OPT_WND_SCALE) {
1240 // Window scale factor.
1241 // http://www.ietf.org/rfc/rfc1323.txt
1242 if (len != 1) {
1243 RTC_LOG_F(WARNING) << "Invalid window scale option received.";
1244 return;
1245 }
1246 applyWindowScaleOption(data[0]);
1247 }
1248 }
1249
applyWindowScaleOption(uint8_t scale_factor)1250 void PseudoTcp::applyWindowScaleOption(uint8_t scale_factor) {
1251 m_swnd_scale = scale_factor;
1252 }
1253
resizeSendBuffer(uint32_t new_size)1254 void PseudoTcp::resizeSendBuffer(uint32_t new_size) {
1255 m_sbuf_len = new_size;
1256 m_sbuf.SetCapacity(new_size);
1257 }
1258
resizeReceiveBuffer(uint32_t new_size)1259 void PseudoTcp::resizeReceiveBuffer(uint32_t new_size) {
1260 uint8_t scale_factor = 0;
1261
1262 // Determine the scale factor such that the scaled window size can fit
1263 // in a 16-bit unsigned integer.
1264 while (new_size > 0xFFFF) {
1265 ++scale_factor;
1266 new_size >>= 1;
1267 }
1268
1269 // Determine the proper size of the buffer.
1270 new_size <<= scale_factor;
1271 bool result = m_rbuf.SetCapacity(new_size);
1272
1273 // Make sure the new buffer is large enough to contain data in the old
1274 // buffer. This should always be true because this method is called either
1275 // before connection is established or when peers are exchanging connect
1276 // messages.
1277 RTC_DCHECK(result);
1278 m_rbuf_len = new_size;
1279 m_rwnd_scale = scale_factor;
1280 m_ssthresh = new_size;
1281
1282 size_t available_space = 0;
1283 m_rbuf.GetWriteRemaining(&available_space);
1284 m_rcv_wnd = static_cast<uint32_t>(available_space);
1285 }
1286
LockedFifoBuffer(size_t size)1287 PseudoTcp::LockedFifoBuffer::LockedFifoBuffer(size_t size)
1288 : buffer_(new char[size]),
1289 buffer_length_(size),
1290 data_length_(0),
1291 read_position_(0) {}
1292
~LockedFifoBuffer()1293 PseudoTcp::LockedFifoBuffer::~LockedFifoBuffer() {}
1294
GetBuffered() const1295 size_t PseudoTcp::LockedFifoBuffer::GetBuffered() const {
1296 webrtc::MutexLock lock(&mutex_);
1297 return data_length_;
1298 }
1299
SetCapacity(size_t size)1300 bool PseudoTcp::LockedFifoBuffer::SetCapacity(size_t size) {
1301 webrtc::MutexLock lock(&mutex_);
1302 if (data_length_ > size)
1303 return false;
1304
1305 if (size != buffer_length_) {
1306 char* buffer = new char[size];
1307 const size_t copy = data_length_;
1308 const size_t tail_copy = std::min(copy, buffer_length_ - read_position_);
1309 memcpy(buffer, &buffer_[read_position_], tail_copy);
1310 memcpy(buffer + tail_copy, &buffer_[0], copy - tail_copy);
1311 buffer_.reset(buffer);
1312 read_position_ = 0;
1313 buffer_length_ = size;
1314 }
1315
1316 return true;
1317 }
1318
ReadOffset(void * buffer,size_t bytes,size_t offset,size_t * bytes_read)1319 bool PseudoTcp::LockedFifoBuffer::ReadOffset(void* buffer,
1320 size_t bytes,
1321 size_t offset,
1322 size_t* bytes_read) {
1323 webrtc::MutexLock lock(&mutex_);
1324 return ReadOffsetLocked(buffer, bytes, offset, bytes_read);
1325 }
1326
WriteOffset(const void * buffer,size_t bytes,size_t offset,size_t * bytes_written)1327 bool PseudoTcp::LockedFifoBuffer::WriteOffset(const void* buffer,
1328 size_t bytes,
1329 size_t offset,
1330 size_t* bytes_written) {
1331 webrtc::MutexLock lock(&mutex_);
1332 return WriteOffsetLocked(buffer, bytes, offset, bytes_written);
1333 }
1334
Read(void * buffer,size_t bytes,size_t * bytes_read)1335 bool PseudoTcp::LockedFifoBuffer::Read(void* buffer,
1336 size_t bytes,
1337 size_t* bytes_read) {
1338 webrtc::MutexLock lock(&mutex_);
1339 size_t copy = 0;
1340 if (!ReadOffsetLocked(buffer, bytes, 0, ©))
1341 return false;
1342
1343 // If read was successful then adjust the read position and number of
1344 // bytes buffered.
1345 read_position_ = (read_position_ + copy) % buffer_length_;
1346 data_length_ -= copy;
1347 if (bytes_read)
1348 *bytes_read = copy;
1349
1350 return true;
1351 }
1352
Write(const void * buffer,size_t bytes,size_t * bytes_written)1353 bool PseudoTcp::LockedFifoBuffer::Write(const void* buffer,
1354 size_t bytes,
1355 size_t* bytes_written) {
1356 webrtc::MutexLock lock(&mutex_);
1357 size_t copy = 0;
1358 if (!WriteOffsetLocked(buffer, bytes, 0, ©))
1359 return false;
1360
1361 // If write was successful then adjust the number of readable bytes.
1362 data_length_ += copy;
1363 if (bytes_written) {
1364 *bytes_written = copy;
1365 }
1366
1367 return true;
1368 }
1369
ConsumeReadData(size_t size)1370 void PseudoTcp::LockedFifoBuffer::ConsumeReadData(size_t size) {
1371 webrtc::MutexLock lock(&mutex_);
1372 RTC_DCHECK(size <= data_length_);
1373 read_position_ = (read_position_ + size) % buffer_length_;
1374 data_length_ -= size;
1375 }
1376
ConsumeWriteBuffer(size_t size)1377 void PseudoTcp::LockedFifoBuffer::ConsumeWriteBuffer(size_t size) {
1378 webrtc::MutexLock lock(&mutex_);
1379 RTC_DCHECK(size <= buffer_length_ - data_length_);
1380 data_length_ += size;
1381 }
1382
GetWriteRemaining(size_t * size) const1383 bool PseudoTcp::LockedFifoBuffer::GetWriteRemaining(size_t* size) const {
1384 webrtc::MutexLock lock(&mutex_);
1385 *size = buffer_length_ - data_length_;
1386 return true;
1387 }
1388
ReadOffsetLocked(void * buffer,size_t bytes,size_t offset,size_t * bytes_read)1389 bool PseudoTcp::LockedFifoBuffer::ReadOffsetLocked(void* buffer,
1390 size_t bytes,
1391 size_t offset,
1392 size_t* bytes_read) {
1393 if (offset >= data_length_)
1394 return false;
1395
1396 const size_t available = data_length_ - offset;
1397 const size_t read_position = (read_position_ + offset) % buffer_length_;
1398 const size_t copy = std::min(bytes, available);
1399 const size_t tail_copy = std::min(copy, buffer_length_ - read_position);
1400 char* const p = static_cast<char*>(buffer);
1401 memcpy(p, &buffer_[read_position], tail_copy);
1402 memcpy(p + tail_copy, &buffer_[0], copy - tail_copy);
1403
1404 if (bytes_read)
1405 *bytes_read = copy;
1406
1407 return true;
1408 }
1409
WriteOffsetLocked(const void * buffer,size_t bytes,size_t offset,size_t * bytes_written)1410 bool PseudoTcp::LockedFifoBuffer::WriteOffsetLocked(const void* buffer,
1411 size_t bytes,
1412 size_t offset,
1413 size_t* bytes_written) {
1414 if (data_length_ + offset >= buffer_length_)
1415 return false;
1416
1417 const size_t available = buffer_length_ - data_length_ - offset;
1418 const size_t write_position =
1419 (read_position_ + data_length_ + offset) % buffer_length_;
1420 const size_t copy = std::min(bytes, available);
1421 const size_t tail_copy = std::min(copy, buffer_length_ - write_position);
1422 const char* const p = static_cast<const char*>(buffer);
1423 memcpy(&buffer_[write_position], p, tail_copy);
1424 memcpy(&buffer_[0], p + tail_copy, copy - tail_copy);
1425
1426 if (bytes_written)
1427 *bytes_written = copy;
1428
1429 return true;
1430 }
1431
1432 } // namespace cricket
1433