/* * Copyright 2004 The WebRTC Project Authors. All rights reserved. * * Use of this source code is governed by a BSD-style license * that can be found in the LICENSE file in the root of the source * tree. An additional intellectual property rights grant can be found * in the file PATENTS. All contributing project authors may * be found in the AUTHORS file in the root of the source tree. */ #include "p2p/base/pseudo_tcp.h" #include #include #include #include #include #include #include #include "rtc_base/byte_buffer.h" #include "rtc_base/byte_order.h" #include "rtc_base/checks.h" #include "rtc_base/logging.h" #include "rtc_base/numerics/safe_minmax.h" #include "rtc_base/socket.h" #include "rtc_base/time_utils.h" // The following logging is for detailed (packet-level) analysis only. #define _DBG_NONE 0 #define _DBG_NORMAL 1 #define _DBG_VERBOSE 2 #define _DEBUGMSG _DBG_NONE namespace cricket { ////////////////////////////////////////////////////////////////////// // Network Constants ////////////////////////////////////////////////////////////////////// // Standard MTUs const uint16_t PACKET_MAXIMUMS[] = { 65535, // Theoretical maximum, Hyperchannel 32000, // Nothing 17914, // 16Mb IBM Token Ring 8166, // IEEE 802.4 // 4464, // IEEE 802.5 (4Mb max) 4352, // FDDI // 2048, // Wideband Network 2002, // IEEE 802.5 (4Mb recommended) // 1536, // Expermental Ethernet Networks // 1500, // Ethernet, Point-to-Point (default) 1492, // IEEE 802.3 1006, // SLIP, ARPANET // 576, // X.25 Networks // 544, // DEC IP Portal // 512, // NETBIOS 508, // IEEE 802/Source-Rt Bridge, ARCNET 296, // Point-to-Point (low delay) // 68, // Official minimum 0, // End of list marker }; const uint32_t MAX_PACKET = 65535; // Note: we removed lowest level because packet overhead was larger! const uint32_t MIN_PACKET = 296; const uint32_t IP_HEADER_SIZE = 20; // (+ up to 40 bytes of options?) const uint32_t UDP_HEADER_SIZE = 8; // TODO(?): Make JINGLE_HEADER_SIZE transparent to this code? const uint32_t JINGLE_HEADER_SIZE = 64; // when relay framing is in use // Default size for receive and send buffer. const uint32_t DEFAULT_RCV_BUF_SIZE = 60 * 1024; const uint32_t DEFAULT_SND_BUF_SIZE = 90 * 1024; ////////////////////////////////////////////////////////////////////// // Global Constants and Functions ////////////////////////////////////////////////////////////////////// // // 0 1 2 3 // 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 // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ // 0 | Conversation Number | // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ // 4 | Sequence Number | // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ // 8 | Acknowledgment Number | // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ // | | |U|A|P|R|S|F| | // 12 | Control | |R|C|S|S|Y|I| Window | // | | |G|K|H|T|N|N| | // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ // 16 | Timestamp sending | // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ // 20 | Timestamp receiving | // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ // 24 | data | // +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ // ////////////////////////////////////////////////////////////////////// #define PSEUDO_KEEPALIVE 0 const uint32_t HEADER_SIZE = 24; const uint32_t PACKET_OVERHEAD = HEADER_SIZE + UDP_HEADER_SIZE + IP_HEADER_SIZE + JINGLE_HEADER_SIZE; const uint32_t MIN_RTO = 250; // 250 ms (RFC1122, Sec 4.2.3.1 "fractions of a second") const uint32_t DEF_RTO = 3000; // 3 seconds (RFC1122, Sec 4.2.3.1) const uint32_t MAX_RTO = 60000; // 60 seconds const uint32_t DEF_ACK_DELAY = 100; // 100 milliseconds const uint8_t FLAG_CTL = 0x02; const uint8_t FLAG_RST = 0x04; const uint8_t CTL_CONNECT = 0; // TCP options. const uint8_t TCP_OPT_EOL = 0; // End of list. const uint8_t TCP_OPT_NOOP = 1; // No-op. const uint8_t TCP_OPT_MSS = 2; // Maximum segment size. const uint8_t TCP_OPT_WND_SCALE = 3; // Window scale factor. const long DEFAULT_TIMEOUT = 4000; // If there are no pending clocks, wake up every 4 seconds const long CLOSED_TIMEOUT = 60 * 1000; // If the connection is closed, once per minute #if PSEUDO_KEEPALIVE // !?! Rethink these times const uint32_t IDLE_PING = 20 * 1000; // 20 seconds (note: WinXP SP2 firewall udp timeout is 90 seconds) const uint32_t IDLE_TIMEOUT = 90 * 1000; // 90 seconds; #endif // PSEUDO_KEEPALIVE ////////////////////////////////////////////////////////////////////// // Helper Functions ////////////////////////////////////////////////////////////////////// inline void long_to_bytes(uint32_t val, void* buf) { *static_cast(buf) = rtc::HostToNetwork32(val); } inline void short_to_bytes(uint16_t val, void* buf) { *static_cast(buf) = rtc::HostToNetwork16(val); } inline uint32_t bytes_to_long(const void* buf) { return rtc::NetworkToHost32(*static_cast(buf)); } inline uint16_t bytes_to_short(const void* buf) { return rtc::NetworkToHost16(*static_cast(buf)); } ////////////////////////////////////////////////////////////////////// // Debugging Statistics ////////////////////////////////////////////////////////////////////// #if 0 // Not used yet enum Stat { S_SENT_PACKET, // All packet sends S_RESENT_PACKET, // All packet sends that are retransmits S_RECV_PACKET, // All packet receives S_RECV_NEW, // All packet receives that are too new S_RECV_OLD, // All packet receives that are too old S_NUM_STATS }; const char* const STAT_NAMES[S_NUM_STATS] = { "snt", "snt-r", "rcv" "rcv-n", "rcv-o" }; int g_stats[S_NUM_STATS]; inline void Incr(Stat s) { ++g_stats[s]; } void ReportStats() { char buffer[256]; size_t len = 0; for (int i = 0; i < S_NUM_STATS; ++i) { len += snprintf(buffer, arraysize(buffer), "%s%s:%d", (i == 0) ? "" : ",", STAT_NAMES[i], g_stats[i]); g_stats[i] = 0; } RTC_LOG(LS_INFO) << "Stats[" << buffer << "]"; } #endif ////////////////////////////////////////////////////////////////////// // PseudoTcp ////////////////////////////////////////////////////////////////////// uint32_t PseudoTcp::Now() { #if 0 // Use this to synchronize timers with logging timestamps (easier debug) return static_cast(rtc::TimeSince(StartTime())); #else return rtc::Time32(); #endif } PseudoTcp::PseudoTcp(IPseudoTcpNotify* notify, uint32_t conv) : m_notify(notify), m_shutdown(SD_NONE), m_error(0), m_rbuf_len(DEFAULT_RCV_BUF_SIZE), m_rbuf(m_rbuf_len), m_sbuf_len(DEFAULT_SND_BUF_SIZE), m_sbuf(m_sbuf_len) { // Sanity check on buffer sizes (needed for OnTcpWriteable notification logic) RTC_DCHECK(m_rbuf_len + MIN_PACKET < m_sbuf_len); uint32_t now = Now(); m_state = TCP_LISTEN; m_conv = conv; m_rcv_wnd = m_rbuf_len; m_rwnd_scale = m_swnd_scale = 0; m_snd_nxt = 0; m_snd_wnd = 1; m_snd_una = m_rcv_nxt = 0; m_bReadEnable = true; m_bWriteEnable = false; m_t_ack = 0; m_msslevel = 0; m_largest = 0; RTC_DCHECK(MIN_PACKET > PACKET_OVERHEAD); m_mss = MIN_PACKET - PACKET_OVERHEAD; m_mtu_advise = MAX_PACKET; m_rto_base = 0; m_cwnd = 2 * m_mss; m_ssthresh = m_rbuf_len; m_lastrecv = m_lastsend = m_lasttraffic = now; m_bOutgoing = false; m_dup_acks = 0; m_recover = 0; m_ts_recent = m_ts_lastack = 0; m_rx_rto = DEF_RTO; m_rx_srtt = m_rx_rttvar = 0; m_use_nagling = true; m_ack_delay = DEF_ACK_DELAY; m_support_wnd_scale = true; } PseudoTcp::~PseudoTcp() {} int PseudoTcp::Connect() { if (m_state != TCP_LISTEN) { m_error = EINVAL; return -1; } m_state = TCP_SYN_SENT; RTC_LOG(LS_INFO) << "State: TCP_SYN_SENT"; queueConnectMessage(); attemptSend(); return 0; } void PseudoTcp::NotifyMTU(uint16_t mtu) { m_mtu_advise = mtu; if (m_state == TCP_ESTABLISHED) { adjustMTU(); } } void PseudoTcp::NotifyClock(uint32_t now) { if (m_state == TCP_CLOSED) return; // Check if it's time to retransmit a segment if (m_rto_base && (rtc::TimeDiff32(m_rto_base + m_rx_rto, now) <= 0)) { if (m_slist.empty()) { RTC_DCHECK_NOTREACHED(); } else { // Note: (m_slist.front().xmit == 0)) { // retransmit segments #if _DEBUGMSG >= _DBG_NORMAL RTC_LOG(LS_INFO) << "timeout retransmit (rto: " << m_rx_rto << ") (rto_base: " << m_rto_base << ") (now: " << now << ") (dup_acks: " << static_cast(m_dup_acks) << ")"; #endif // _DEBUGMSG if (!transmit(m_slist.begin(), now)) { closedown(ECONNABORTED); return; } uint32_t nInFlight = m_snd_nxt - m_snd_una; m_ssthresh = std::max(nInFlight / 2, 2 * m_mss); // RTC_LOG(LS_INFO) << "m_ssthresh: " << m_ssthresh << " nInFlight: " << // nInFlight << " m_mss: " << m_mss; m_cwnd = m_mss; // Back off retransmit timer. Note: the limit is lower when connecting. uint32_t rto_limit = (m_state < TCP_ESTABLISHED) ? DEF_RTO : MAX_RTO; m_rx_rto = std::min(rto_limit, m_rx_rto * 2); m_rto_base = now; } } // Check if it's time to probe closed windows if ((m_snd_wnd == 0) && (rtc::TimeDiff32(m_lastsend + m_rx_rto, now) <= 0)) { if (rtc::TimeDiff32(now, m_lastrecv) >= 15000) { closedown(ECONNABORTED); return; } // probe the window packet(m_snd_nxt - 1, 0, 0, 0); m_lastsend = now; // back off retransmit timer m_rx_rto = std::min(MAX_RTO, m_rx_rto * 2); } // Check if it's time to send delayed acks if (m_t_ack && (rtc::TimeDiff32(m_t_ack + m_ack_delay, now) <= 0)) { packet(m_snd_nxt, 0, 0, 0); } #if PSEUDO_KEEPALIVE // Check for idle timeout if ((m_state == TCP_ESTABLISHED) && (TimeDiff32(m_lastrecv + IDLE_TIMEOUT, now) <= 0)) { closedown(ECONNABORTED); return; } // Check for ping timeout (to keep udp mapping open) if ((m_state == TCP_ESTABLISHED) && (TimeDiff32(m_lasttraffic + (m_bOutgoing ? IDLE_PING * 3 / 2 : IDLE_PING), now) <= 0)) { packet(m_snd_nxt, 0, 0, 0); } #endif // PSEUDO_KEEPALIVE } bool PseudoTcp::NotifyPacket(const char* buffer, size_t len) { if (len > MAX_PACKET) { RTC_LOG_F(LS_WARNING) << "packet too large"; return false; } return parse(reinterpret_cast(buffer), uint32_t(len)); } bool PseudoTcp::GetNextClock(uint32_t now, long& timeout) { return clock_check(now, timeout); } void PseudoTcp::GetOption(Option opt, int* value) { if (opt == OPT_NODELAY) { *value = m_use_nagling ? 0 : 1; } else if (opt == OPT_ACKDELAY) { *value = m_ack_delay; } else if (opt == OPT_SNDBUF) { *value = m_sbuf_len; } else if (opt == OPT_RCVBUF) { *value = m_rbuf_len; } else { RTC_DCHECK_NOTREACHED(); } } void PseudoTcp::SetOption(Option opt, int value) { if (opt == OPT_NODELAY) { m_use_nagling = value == 0; } else if (opt == OPT_ACKDELAY) { m_ack_delay = value; } else if (opt == OPT_SNDBUF) { RTC_DCHECK(m_state == TCP_LISTEN); resizeSendBuffer(value); } else if (opt == OPT_RCVBUF) { RTC_DCHECK(m_state == TCP_LISTEN); resizeReceiveBuffer(value); } else { RTC_DCHECK_NOTREACHED(); } } uint32_t PseudoTcp::GetCongestionWindow() const { return m_cwnd; } uint32_t PseudoTcp::GetBytesInFlight() const { return m_snd_nxt - m_snd_una; } uint32_t PseudoTcp::GetBytesBufferedNotSent() const { return static_cast(m_snd_una + m_sbuf.GetBuffered() - m_snd_nxt); } uint32_t PseudoTcp::GetRoundTripTimeEstimateMs() const { return m_rx_srtt; } // // IPStream Implementation // int PseudoTcp::Recv(char* buffer, size_t len) { if (m_state != TCP_ESTABLISHED) { m_error = ENOTCONN; return SOCKET_ERROR; } size_t read = 0; if (!m_rbuf.Read(buffer, len, &read)) { m_bReadEnable = true; m_error = EWOULDBLOCK; return SOCKET_ERROR; } size_t available_space = 0; m_rbuf.GetWriteRemaining(&available_space); if (uint32_t(available_space) - m_rcv_wnd >= std::min(m_rbuf_len / 2, m_mss)) { // TODO(jbeda): !?! Not sure about this was closed business bool bWasClosed = (m_rcv_wnd == 0); m_rcv_wnd = static_cast(available_space); if (bWasClosed) { attemptSend(sfImmediateAck); } } return static_cast(read); } int PseudoTcp::Send(const char* buffer, size_t len) { if (m_state != TCP_ESTABLISHED) { m_error = ENOTCONN; return SOCKET_ERROR; } size_t available_space = 0; m_sbuf.GetWriteRemaining(&available_space); if (!available_space) { m_bWriteEnable = true; m_error = EWOULDBLOCK; return SOCKET_ERROR; } int written = queue(buffer, uint32_t(len), false); attemptSend(); return written; } void PseudoTcp::Close(bool force) { RTC_LOG_F(LS_VERBOSE) << "(" << (force ? "true" : "false") << ")"; m_shutdown = force ? SD_FORCEFUL : SD_GRACEFUL; } int PseudoTcp::GetError() { return m_error; } // // Internal Implementation // uint32_t PseudoTcp::queue(const char* data, uint32_t len, bool bCtrl) { size_t available_space = 0; m_sbuf.GetWriteRemaining(&available_space); if (len > static_cast(available_space)) { RTC_DCHECK(!bCtrl); len = static_cast(available_space); } // We can concatenate data if the last segment is the same type // (control v. regular data), and has not been transmitted yet if (!m_slist.empty() && (m_slist.back().bCtrl == bCtrl) && (m_slist.back().xmit == 0)) { m_slist.back().len += len; } else { SSegment sseg(static_cast(m_snd_una + m_sbuf.GetBuffered()), len, bCtrl); m_slist.push_back(sseg); } size_t written = 0; m_sbuf.Write(data, len, &written); return static_cast(written); } IPseudoTcpNotify::WriteResult PseudoTcp::packet(uint32_t seq, uint8_t flags, uint32_t offset, uint32_t len) { RTC_DCHECK(HEADER_SIZE + len <= MAX_PACKET); uint32_t now = Now(); std::unique_ptr buffer(new uint8_t[MAX_PACKET]); long_to_bytes(m_conv, buffer.get()); long_to_bytes(seq, buffer.get() + 4); long_to_bytes(m_rcv_nxt, buffer.get() + 8); buffer[12] = 0; buffer[13] = flags; short_to_bytes(static_cast(m_rcv_wnd >> m_rwnd_scale), buffer.get() + 14); // Timestamp computations long_to_bytes(now, buffer.get() + 16); long_to_bytes(m_ts_recent, buffer.get() + 20); m_ts_lastack = m_rcv_nxt; if (len) { size_t bytes_read = 0; bool result = m_sbuf.ReadOffset(buffer.get() + HEADER_SIZE, len, offset, &bytes_read); RTC_DCHECK(result); RTC_DCHECK(static_cast(bytes_read) == len); } #if _DEBUGMSG >= _DBG_VERBOSE RTC_LOG(LS_INFO) << "<-- (flags) << ">"; #endif // _DEBUGMSG IPseudoTcpNotify::WriteResult wres = m_notify->TcpWritePacket( this, reinterpret_cast(buffer.get()), len + HEADER_SIZE); // Note: When len is 0, this is an ACK packet. We don't read the return value // for those, and thus we won't retry. So go ahead and treat the packet as a // success (basically simulate as if it were dropped), which will prevent our // timers from being messed up. if ((wres != IPseudoTcpNotify::WR_SUCCESS) && (0 != len)) return wres; m_t_ack = 0; if (len > 0) { m_lastsend = now; } m_lasttraffic = now; m_bOutgoing = true; return IPseudoTcpNotify::WR_SUCCESS; } bool PseudoTcp::parse(const uint8_t* buffer, uint32_t size) { if (size < HEADER_SIZE) return false; Segment seg; seg.conv = bytes_to_long(buffer); seg.seq = bytes_to_long(buffer + 4); seg.ack = bytes_to_long(buffer + 8); seg.flags = buffer[13]; seg.wnd = bytes_to_short(buffer + 14); seg.tsval = bytes_to_long(buffer + 16); seg.tsecr = bytes_to_long(buffer + 20); seg.data = reinterpret_cast(buffer) + HEADER_SIZE; seg.len = size - HEADER_SIZE; #if _DEBUGMSG >= _DBG_VERBOSE RTC_LOG(LS_INFO) << "--> (seg.flags) << ">"; #endif // _DEBUGMSG return process(seg); } bool PseudoTcp::clock_check(uint32_t now, long& nTimeout) { if (m_shutdown == SD_FORCEFUL) return false; if ((m_shutdown == SD_GRACEFUL) && ((m_state != TCP_ESTABLISHED) || ((m_sbuf.GetBuffered() == 0) && (m_t_ack == 0)))) { return false; } if (m_state == TCP_CLOSED) { nTimeout = CLOSED_TIMEOUT; return true; } nTimeout = DEFAULT_TIMEOUT; if (m_t_ack) { nTimeout = std::min(nTimeout, rtc::TimeDiff32(m_t_ack + m_ack_delay, now)); } if (m_rto_base) { nTimeout = std::min(nTimeout, rtc::TimeDiff32(m_rto_base + m_rx_rto, now)); } if (m_snd_wnd == 0) { nTimeout = std::min(nTimeout, rtc::TimeDiff32(m_lastsend + m_rx_rto, now)); } #if PSEUDO_KEEPALIVE if (m_state == TCP_ESTABLISHED) { nTimeout = std::min( nTimeout, rtc::TimeDiff32( m_lasttraffic + (m_bOutgoing ? IDLE_PING * 3 / 2 : IDLE_PING), now)); } #endif // PSEUDO_KEEPALIVE return true; } bool PseudoTcp::process(Segment& seg) { // If this is the wrong conversation, send a reset!?! (with the correct // conversation?) if (seg.conv != m_conv) { // if ((seg.flags & FLAG_RST) == 0) { // packet(tcb, seg.ack, 0, FLAG_RST, 0, 0); //} RTC_LOG_F(LS_ERROR) << "wrong conversation"; return false; } uint32_t now = Now(); m_lasttraffic = m_lastrecv = now; m_bOutgoing = false; if (m_state == TCP_CLOSED) { // !?! send reset? RTC_LOG_F(LS_ERROR) << "closed"; return false; } // Check if this is a reset segment if (seg.flags & FLAG_RST) { closedown(ECONNRESET); return false; } // Check for control data bool bConnect = false; if (seg.flags & FLAG_CTL) { if (seg.len == 0) { RTC_LOG_F(LS_ERROR) << "Missing control code"; return false; } else if (seg.data[0] == CTL_CONNECT) { bConnect = true; // TCP options are in the remainder of the payload after CTL_CONNECT. parseOptions(&seg.data[1], seg.len - 1); if (m_state == TCP_LISTEN) { m_state = TCP_SYN_RECEIVED; RTC_LOG(LS_INFO) << "State: TCP_SYN_RECEIVED"; // m_notify->associate(addr); queueConnectMessage(); } else if (m_state == TCP_SYN_SENT) { m_state = TCP_ESTABLISHED; RTC_LOG(LS_INFO) << "State: TCP_ESTABLISHED"; adjustMTU(); if (m_notify) { m_notify->OnTcpOpen(this); } // notify(evOpen); } } else { RTC_LOG_F(LS_WARNING) << "Unknown control code: " << seg.data[0]; return false; } } // Update timestamp if ((seg.seq <= m_ts_lastack) && (m_ts_lastack < seg.seq + seg.len)) { m_ts_recent = seg.tsval; } // Check if this is a valuable ack if ((seg.ack > m_snd_una) && (seg.ack <= m_snd_nxt)) { // Calculate round-trip time if (seg.tsecr) { int32_t rtt = rtc::TimeDiff32(now, seg.tsecr); if (rtt >= 0) { if (m_rx_srtt == 0) { m_rx_srtt = rtt; m_rx_rttvar = rtt / 2; } else { uint32_t unsigned_rtt = static_cast(rtt); uint32_t abs_err = unsigned_rtt > m_rx_srtt ? unsigned_rtt - m_rx_srtt : m_rx_srtt - unsigned_rtt; m_rx_rttvar = (3 * m_rx_rttvar + abs_err) / 4; m_rx_srtt = (7 * m_rx_srtt + rtt) / 8; } m_rx_rto = rtc::SafeClamp(m_rx_srtt + rtc::SafeMax(1, 4 * m_rx_rttvar), MIN_RTO, MAX_RTO); #if _DEBUGMSG >= _DBG_VERBOSE RTC_LOG(LS_INFO) << "rtt: " << rtt << " srtt: " << m_rx_srtt << " rto: " << m_rx_rto; #endif // _DEBUGMSG } else { RTC_LOG(LS_WARNING) << "rtt < 0"; } } m_snd_wnd = static_cast(seg.wnd) << m_swnd_scale; uint32_t nAcked = seg.ack - m_snd_una; m_snd_una = seg.ack; m_rto_base = (m_snd_una == m_snd_nxt) ? 0 : now; m_sbuf.ConsumeReadData(nAcked); for (uint32_t nFree = nAcked; nFree > 0;) { RTC_DCHECK(!m_slist.empty()); if (nFree < m_slist.front().len) { m_slist.front().len -= nFree; nFree = 0; } else { if (m_slist.front().len > m_largest) { m_largest = m_slist.front().len; } nFree -= m_slist.front().len; m_slist.pop_front(); } } if (m_dup_acks >= 3) { if (m_snd_una >= m_recover) { // NewReno uint32_t nInFlight = m_snd_nxt - m_snd_una; m_cwnd = std::min(m_ssthresh, nInFlight + m_mss); // (Fast Retransmit) #if _DEBUGMSG >= _DBG_NORMAL RTC_LOG(LS_INFO) << "exit recovery"; #endif // _DEBUGMSG m_dup_acks = 0; } else { #if _DEBUGMSG >= _DBG_NORMAL RTC_LOG(LS_INFO) << "recovery retransmit"; #endif // _DEBUGMSG if (!transmit(m_slist.begin(), now)) { closedown(ECONNABORTED); return false; } m_cwnd += m_mss - std::min(nAcked, m_cwnd); } } else { m_dup_acks = 0; // Slow start, congestion avoidance if (m_cwnd < m_ssthresh) { m_cwnd += m_mss; } else { m_cwnd += std::max(1, m_mss * m_mss / m_cwnd); } } } else if (seg.ack == m_snd_una) { // !?! Note, tcp says don't do this... but otherwise how does a closed // window become open? m_snd_wnd = static_cast(seg.wnd) << m_swnd_scale; // Check duplicate acks if (seg.len > 0) { // it's a dup ack, but with a data payload, so don't modify m_dup_acks } else if (m_snd_una != m_snd_nxt) { m_dup_acks += 1; if (m_dup_acks == 3) { // (Fast Retransmit) #if _DEBUGMSG >= _DBG_NORMAL RTC_LOG(LS_INFO) << "enter recovery"; RTC_LOG(LS_INFO) << "recovery retransmit"; #endif // _DEBUGMSG if (!transmit(m_slist.begin(), now)) { closedown(ECONNABORTED); return false; } m_recover = m_snd_nxt; uint32_t nInFlight = m_snd_nxt - m_snd_una; m_ssthresh = std::max(nInFlight / 2, 2 * m_mss); // RTC_LOG(LS_INFO) << "m_ssthresh: " << m_ssthresh << " nInFlight: " // << nInFlight << " m_mss: " << m_mss; m_cwnd = m_ssthresh + 3 * m_mss; } else if (m_dup_acks > 3) { m_cwnd += m_mss; } } else { m_dup_acks = 0; } } // !?! A bit hacky if ((m_state == TCP_SYN_RECEIVED) && !bConnect) { m_state = TCP_ESTABLISHED; RTC_LOG(LS_INFO) << "State: TCP_ESTABLISHED"; adjustMTU(); if (m_notify) { m_notify->OnTcpOpen(this); } // notify(evOpen); } // If we make room in the send queue, notify the user // The goal it to make sure we always have at least enough data to fill the // window. We'd like to notify the app when we are halfway to that point. const uint32_t kIdealRefillSize = (m_sbuf_len + m_rbuf_len) / 2; if (m_bWriteEnable && static_cast(m_sbuf.GetBuffered()) < kIdealRefillSize) { m_bWriteEnable = false; if (m_notify) { m_notify->OnTcpWriteable(this); } // notify(evWrite); } // Conditions were acks must be sent: // 1) Segment is too old (they missed an ACK) (immediately) // 2) Segment is too new (we missed a segment) (immediately) // 3) Segment has data (so we need to ACK!) (delayed) // ... so the only time we don't need to ACK, is an empty segment that points // to rcv_nxt! SendFlags sflags = sfNone; if (seg.seq != m_rcv_nxt) { sflags = sfImmediateAck; // (Fast Recovery) } else if (seg.len != 0) { if (m_ack_delay == 0) { sflags = sfImmediateAck; } else { sflags = sfDelayedAck; } } #if _DEBUGMSG >= _DBG_NORMAL if (sflags == sfImmediateAck) { if (seg.seq > m_rcv_nxt) { RTC_LOG_F(LS_INFO) << "too new"; } else if (seg.seq + seg.len <= m_rcv_nxt) { RTC_LOG_F(LS_INFO) << "too old"; } } #endif // _DEBUGMSG // Adjust the incoming segment to fit our receive buffer if (seg.seq < m_rcv_nxt) { uint32_t nAdjust = m_rcv_nxt - seg.seq; if (nAdjust < seg.len) { seg.seq += nAdjust; seg.data += nAdjust; seg.len -= nAdjust; } else { seg.len = 0; } } size_t available_space = 0; m_rbuf.GetWriteRemaining(&available_space); if ((seg.seq + seg.len - m_rcv_nxt) > static_cast(available_space)) { uint32_t nAdjust = seg.seq + seg.len - m_rcv_nxt - static_cast(available_space); if (nAdjust < seg.len) { seg.len -= nAdjust; } else { seg.len = 0; } } bool bIgnoreData = (seg.flags & FLAG_CTL) || (m_shutdown != SD_NONE); bool bNewData = false; if (seg.len > 0) { bool bRecover = false; if (bIgnoreData) { if (seg.seq == m_rcv_nxt) { m_rcv_nxt += seg.len; // If we received a data segment out of order relative to a control // segment, then we wrote it into the receive buffer at an offset (see // "WriteOffset") below. So we need to advance the position in the // buffer to avoid corrupting data. See bugs.webrtc.org/9208 // // We advance the position in the buffer by N bytes by acting like we // wrote N bytes and then immediately read them. We can only do this if // there's not already data ready to read, but this should always be // true in the problematic scenario, since control frames are always // sent first in the stream. if (m_rbuf.GetBuffered() == 0) { m_rbuf.ConsumeWriteBuffer(seg.len); m_rbuf.ConsumeReadData(seg.len); // After shifting the position in the buffer, we may have // out-of-order packets ready to be recovered. bRecover = true; } } } else { uint32_t nOffset = seg.seq - m_rcv_nxt; if (!m_rbuf.WriteOffset(seg.data, seg.len, nOffset, NULL)) { // Ignore incoming packets outside of the receive window. return false; } if (seg.seq == m_rcv_nxt) { m_rbuf.ConsumeWriteBuffer(seg.len); m_rcv_nxt += seg.len; m_rcv_wnd -= seg.len; bNewData = true; // May be able to recover packets previously received out-of-order // now. bRecover = true; } else { #if _DEBUGMSG >= _DBG_NORMAL RTC_LOG(LS_INFO) << "Saving " << seg.len << " bytes (" << seg.seq << " -> " << seg.seq + seg.len << ")"; #endif // _DEBUGMSG RSegment rseg; rseg.seq = seg.seq; rseg.len = seg.len; RList::iterator it = m_rlist.begin(); while ((it != m_rlist.end()) && (it->seq < rseg.seq)) { ++it; } m_rlist.insert(it, rseg); } } if (bRecover) { RList::iterator it = m_rlist.begin(); while ((it != m_rlist.end()) && (it->seq <= m_rcv_nxt)) { if (it->seq + it->len > m_rcv_nxt) { sflags = sfImmediateAck; // (Fast Recovery) uint32_t nAdjust = (it->seq + it->len) - m_rcv_nxt; #if _DEBUGMSG >= _DBG_NORMAL RTC_LOG(LS_INFO) << "Recovered " << nAdjust << " bytes (" << m_rcv_nxt << " -> " << m_rcv_nxt + nAdjust << ")"; #endif // _DEBUGMSG m_rbuf.ConsumeWriteBuffer(nAdjust); m_rcv_nxt += nAdjust; m_rcv_wnd -= nAdjust; bNewData = true; } it = m_rlist.erase(it); } } } attemptSend(sflags); // If we have new data, notify the user if (bNewData && m_bReadEnable) { m_bReadEnable = false; if (m_notify) { m_notify->OnTcpReadable(this); } // notify(evRead); } return true; } bool PseudoTcp::transmit(const SList::iterator& seg, uint32_t now) { if (seg->xmit >= ((m_state == TCP_ESTABLISHED) ? 15 : 30)) { RTC_LOG_F(LS_VERBOSE) << "too many retransmits"; return false; } uint32_t nTransmit = std::min(seg->len, m_mss); while (true) { uint32_t seq = seg->seq; uint8_t flags = (seg->bCtrl ? FLAG_CTL : 0); IPseudoTcpNotify::WriteResult wres = packet(seq, flags, seg->seq - m_snd_una, nTransmit); if (wres == IPseudoTcpNotify::WR_SUCCESS) break; if (wres == IPseudoTcpNotify::WR_FAIL) { RTC_LOG_F(LS_VERBOSE) << "packet failed"; return false; } RTC_DCHECK(wres == IPseudoTcpNotify::WR_TOO_LARGE); while (true) { if (PACKET_MAXIMUMS[m_msslevel + 1] == 0) { RTC_LOG_F(LS_VERBOSE) << "MTU too small"; return false; } // !?! We need to break up all outstanding and pending packets and then // retransmit!?! m_mss = PACKET_MAXIMUMS[++m_msslevel] - PACKET_OVERHEAD; m_cwnd = 2 * m_mss; // I added this... haven't researched actual formula if (m_mss < nTransmit) { nTransmit = m_mss; break; } } #if _DEBUGMSG >= _DBG_NORMAL RTC_LOG(LS_INFO) << "Adjusting mss to " << m_mss << " bytes"; #endif // _DEBUGMSG } if (nTransmit < seg->len) { RTC_LOG_F(LS_VERBOSE) << "mss reduced to " << m_mss; SSegment subseg(seg->seq + nTransmit, seg->len - nTransmit, seg->bCtrl); // subseg.tstamp = seg->tstamp; subseg.xmit = seg->xmit; seg->len = nTransmit; SList::iterator next = seg; m_slist.insert(++next, subseg); } if (seg->xmit == 0) { m_snd_nxt += seg->len; } seg->xmit += 1; // seg->tstamp = now; if (m_rto_base == 0) { m_rto_base = now; } return true; } void PseudoTcp::attemptSend(SendFlags sflags) { uint32_t now = Now(); if (rtc::TimeDiff32(now, m_lastsend) > static_cast(m_rx_rto)) { m_cwnd = m_mss; } #if _DEBUGMSG bool bFirst = true; #endif // _DEBUGMSG while (true) { uint32_t cwnd = m_cwnd; if ((m_dup_acks == 1) || (m_dup_acks == 2)) { // Limited Transmit cwnd += m_dup_acks * m_mss; } uint32_t nWindow = std::min(m_snd_wnd, cwnd); uint32_t nInFlight = m_snd_nxt - m_snd_una; uint32_t nUseable = (nInFlight < nWindow) ? (nWindow - nInFlight) : 0; size_t snd_buffered = m_sbuf.GetBuffered(); uint32_t nAvailable = std::min(static_cast(snd_buffered) - nInFlight, m_mss); if (nAvailable > nUseable) { if (nUseable * 4 < nWindow) { // RFC 813 - avoid SWS nAvailable = 0; } else { nAvailable = nUseable; } } #if _DEBUGMSG >= _DBG_VERBOSE if (bFirst) { size_t available_space = 0; m_sbuf.GetWriteRemaining(&available_space); bFirst = false; RTC_LOG(LS_INFO) << "[cwnd: " << m_cwnd << " nWindow: " << nWindow << " nInFlight: " << nInFlight << " nAvailable: " << nAvailable << " nQueued: " << snd_buffered << " nEmpty: " << available_space << " ssthresh: " << m_ssthresh << "]"; } #endif // _DEBUGMSG if (nAvailable == 0) { if (sflags == sfNone) return; // If this is an immediate ack, or the second delayed ack if ((sflags == sfImmediateAck) || m_t_ack) { packet(m_snd_nxt, 0, 0, 0); } else { m_t_ack = Now(); } return; } // Nagle's algorithm. // If there is data already in-flight, and we haven't a full segment of // data ready to send then hold off until we get more to send, or the // in-flight data is acknowledged. if (m_use_nagling && (m_snd_nxt > m_snd_una) && (nAvailable < m_mss)) { return; } // Find the next segment to transmit SList::iterator it = m_slist.begin(); while (it->xmit > 0) { ++it; RTC_DCHECK(it != m_slist.end()); } SList::iterator seg = it; // If the segment is too large, break it into two if (seg->len > nAvailable) { SSegment subseg(seg->seq + nAvailable, seg->len - nAvailable, seg->bCtrl); seg->len = nAvailable; m_slist.insert(++it, subseg); } if (!transmit(seg, now)) { RTC_LOG_F(LS_VERBOSE) << "transmit failed"; // TODO(?): consider closing socket return; } sflags = sfNone; } } void PseudoTcp::closedown(uint32_t err) { RTC_LOG(LS_INFO) << "State: TCP_CLOSED"; m_state = TCP_CLOSED; if (m_notify) { m_notify->OnTcpClosed(this, err); } // notify(evClose, err); } void PseudoTcp::adjustMTU() { // Determine our current mss level, so that we can adjust appropriately later for (m_msslevel = 0; PACKET_MAXIMUMS[m_msslevel + 1] > 0; ++m_msslevel) { if (static_cast(PACKET_MAXIMUMS[m_msslevel]) <= m_mtu_advise) { break; } } m_mss = m_mtu_advise - PACKET_OVERHEAD; // !?! Should we reset m_largest here? #if _DEBUGMSG >= _DBG_NORMAL RTC_LOG(LS_INFO) << "Adjusting mss to " << m_mss << " bytes"; #endif // _DEBUGMSG // Enforce minimums on ssthresh and cwnd m_ssthresh = std::max(m_ssthresh, 2 * m_mss); m_cwnd = std::max(m_cwnd, m_mss); } bool PseudoTcp::isReceiveBufferFull() const { size_t available_space = 0; m_rbuf.GetWriteRemaining(&available_space); return !available_space; } void PseudoTcp::disableWindowScale() { m_support_wnd_scale = false; } void PseudoTcp::queueConnectMessage() { rtc::ByteBufferWriter buf; buf.WriteUInt8(CTL_CONNECT); if (m_support_wnd_scale) { buf.WriteUInt8(TCP_OPT_WND_SCALE); buf.WriteUInt8(1); buf.WriteUInt8(m_rwnd_scale); } m_snd_wnd = static_cast(buf.Length()); queue(buf.Data(), static_cast(buf.Length()), true); } void PseudoTcp::parseOptions(const char* data, uint32_t len) { std::set options_specified; // See http://www.freesoft.org/CIE/Course/Section4/8.htm for // parsing the options list. rtc::ByteBufferReader buf( rtc::MakeArrayView(reinterpret_cast(data), len)); while (buf.Length()) { uint8_t kind = TCP_OPT_EOL; buf.ReadUInt8(&kind); if (kind == TCP_OPT_EOL) { // End of option list. break; } else if (kind == TCP_OPT_NOOP) { // No op. continue; } // Length of this option. RTC_DCHECK(len != 0); uint8_t opt_len = 0; buf.ReadUInt8(&opt_len); // Content of this option. if (opt_len <= buf.Length()) { applyOption(kind, buf.Data(), opt_len); buf.Consume(opt_len); } else { RTC_LOG(LS_ERROR) << "Invalid option length received."; return; } options_specified.insert(kind); } if (options_specified.find(TCP_OPT_WND_SCALE) == options_specified.end()) { RTC_LOG(LS_WARNING) << "Peer doesn't support window scaling"; if (m_rwnd_scale > 0) { // Peer doesn't support TCP options and window scaling. // Revert receive buffer size to default value. resizeReceiveBuffer(DEFAULT_RCV_BUF_SIZE); m_swnd_scale = 0; } } } void PseudoTcp::applyOption(char kind, const char* data, uint32_t len) { if (kind == TCP_OPT_MSS) { RTC_LOG(LS_WARNING) << "Peer specified MSS option which is not supported."; // TODO(?): Implement. } else if (kind == TCP_OPT_WND_SCALE) { // Window scale factor. // http://www.ietf.org/rfc/rfc1323.txt if (len != 1) { RTC_LOG_F(LS_WARNING) << "Invalid window scale option received."; return; } applyWindowScaleOption(data[0]); } } void PseudoTcp::applyWindowScaleOption(uint8_t scale_factor) { m_swnd_scale = scale_factor; } void PseudoTcp::resizeSendBuffer(uint32_t new_size) { m_sbuf_len = new_size; m_sbuf.SetCapacity(new_size); } void PseudoTcp::resizeReceiveBuffer(uint32_t new_size) { uint8_t scale_factor = 0; // Determine the scale factor such that the scaled window size can fit // in a 16-bit unsigned integer. while (new_size > 0xFFFF) { ++scale_factor; new_size >>= 1; } // Determine the proper size of the buffer. new_size <<= scale_factor; bool result = m_rbuf.SetCapacity(new_size); // Make sure the new buffer is large enough to contain data in the old // buffer. This should always be true because this method is called either // before connection is established or when peers are exchanging connect // messages. RTC_DCHECK(result); m_rbuf_len = new_size; m_rwnd_scale = scale_factor; m_ssthresh = new_size; size_t available_space = 0; m_rbuf.GetWriteRemaining(&available_space); m_rcv_wnd = static_cast(available_space); } PseudoTcp::LockedFifoBuffer::LockedFifoBuffer(size_t size) : buffer_(new char[size]), buffer_length_(size), data_length_(0), read_position_(0) {} PseudoTcp::LockedFifoBuffer::~LockedFifoBuffer() {} size_t PseudoTcp::LockedFifoBuffer::GetBuffered() const { webrtc::MutexLock lock(&mutex_); return data_length_; } bool PseudoTcp::LockedFifoBuffer::SetCapacity(size_t size) { webrtc::MutexLock lock(&mutex_); if (data_length_ > size) return false; if (size != buffer_length_) { char* buffer = new char[size]; const size_t copy = data_length_; const size_t tail_copy = std::min(copy, buffer_length_ - read_position_); memcpy(buffer, &buffer_[read_position_], tail_copy); memcpy(buffer + tail_copy, &buffer_[0], copy - tail_copy); buffer_.reset(buffer); read_position_ = 0; buffer_length_ = size; } return true; } bool PseudoTcp::LockedFifoBuffer::ReadOffset(void* buffer, size_t bytes, size_t offset, size_t* bytes_read) { webrtc::MutexLock lock(&mutex_); return ReadOffsetLocked(buffer, bytes, offset, bytes_read); } bool PseudoTcp::LockedFifoBuffer::WriteOffset(const void* buffer, size_t bytes, size_t offset, size_t* bytes_written) { webrtc::MutexLock lock(&mutex_); return WriteOffsetLocked(buffer, bytes, offset, bytes_written); } bool PseudoTcp::LockedFifoBuffer::Read(void* buffer, size_t bytes, size_t* bytes_read) { webrtc::MutexLock lock(&mutex_); size_t copy = 0; if (!ReadOffsetLocked(buffer, bytes, 0, ©)) return false; // If read was successful then adjust the read position and number of // bytes buffered. read_position_ = (read_position_ + copy) % buffer_length_; data_length_ -= copy; if (bytes_read) *bytes_read = copy; return true; } bool PseudoTcp::LockedFifoBuffer::Write(const void* buffer, size_t bytes, size_t* bytes_written) { webrtc::MutexLock lock(&mutex_); size_t copy = 0; if (!WriteOffsetLocked(buffer, bytes, 0, ©)) return false; // If write was successful then adjust the number of readable bytes. data_length_ += copy; if (bytes_written) { *bytes_written = copy; } return true; } void PseudoTcp::LockedFifoBuffer::ConsumeReadData(size_t size) { webrtc::MutexLock lock(&mutex_); RTC_DCHECK(size <= data_length_); read_position_ = (read_position_ + size) % buffer_length_; data_length_ -= size; } void PseudoTcp::LockedFifoBuffer::ConsumeWriteBuffer(size_t size) { webrtc::MutexLock lock(&mutex_); RTC_DCHECK(size <= buffer_length_ - data_length_); data_length_ += size; } bool PseudoTcp::LockedFifoBuffer::GetWriteRemaining(size_t* size) const { webrtc::MutexLock lock(&mutex_); *size = buffer_length_ - data_length_; return true; } bool PseudoTcp::LockedFifoBuffer::ReadOffsetLocked(void* buffer, size_t bytes, size_t offset, size_t* bytes_read) { if (offset >= data_length_) return false; const size_t available = data_length_ - offset; const size_t read_position = (read_position_ + offset) % buffer_length_; const size_t copy = std::min(bytes, available); const size_t tail_copy = std::min(copy, buffer_length_ - read_position); char* const p = static_cast(buffer); memcpy(p, &buffer_[read_position], tail_copy); memcpy(p + tail_copy, &buffer_[0], copy - tail_copy); if (bytes_read) *bytes_read = copy; return true; } bool PseudoTcp::LockedFifoBuffer::WriteOffsetLocked(const void* buffer, size_t bytes, size_t offset, size_t* bytes_written) { if (data_length_ + offset >= buffer_length_) return false; const size_t available = buffer_length_ - data_length_ - offset; const size_t write_position = (read_position_ + data_length_ + offset) % buffer_length_; const size_t copy = std::min(bytes, available); const size_t tail_copy = std::min(copy, buffer_length_ - write_position); const char* const p = static_cast(buffer); memcpy(&buffer_[write_position], p, tail_copy); memcpy(&buffer_[0], p + tail_copy, copy - tail_copy); if (bytes_written) *bytes_written = copy; return true; } } // namespace cricket