/* * libjingle * Copyright 2004--2005, Google Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are met: * * 1. Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright notice, * this list of conditions and the following disclaimer in the documentation * and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include "talk/base/virtualsocketserver.h" #include #include #include #include #include #include "talk/base/common.h" #include "talk/base/logging.h" #include "talk/base/physicalsocketserver.h" #include "talk/base/socketaddresspair.h" #include "talk/base/thread.h" #include "talk/base/timeutils.h" namespace talk_base { #ifdef WIN32 const in_addr kInitialNextIPv4 = { {0x01, 0, 0, 0} }; #else // This value is entirely arbitrary, hence the lack of concern about endianness. const in_addr kInitialNextIPv4 = { 0x01000000 }; #endif // Starts at ::2 so as to not cause confusion with ::1. const in6_addr kInitialNextIPv6 = { { { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2 } } }; const uint16 kFirstEphemeralPort = 49152; const uint16 kLastEphemeralPort = 65535; const uint16 kEphemeralPortCount = kLastEphemeralPort - kFirstEphemeralPort + 1; const uint32 kDefaultNetworkCapacity = 64 * 1024; const uint32 kDefaultTcpBufferSize = 32 * 1024; const uint32 UDP_HEADER_SIZE = 28; // IP + UDP headers const uint32 TCP_HEADER_SIZE = 40; // IP + TCP headers const uint32 TCP_MSS = 1400; // Maximum segment size // Note: The current algorithm doesn't work for sample sizes smaller than this. const int NUM_SAMPLES = 1000; enum { MSG_ID_PACKET, MSG_ID_CONNECT, MSG_ID_DISCONNECT, }; // Packets are passed between sockets as messages. We copy the data just like // the kernel does. class Packet : public MessageData { public: Packet(const char* data, size_t size, const SocketAddress& from) : size_(size), consumed_(0), from_(from) { ASSERT(NULL != data); data_ = new char[size_]; memcpy(data_, data, size_); } virtual ~Packet() { delete[] data_; } const char* data() const { return data_ + consumed_; } size_t size() const { return size_ - consumed_; } const SocketAddress& from() const { return from_; } // Remove the first size bytes from the data. void Consume(size_t size) { ASSERT(size + consumed_ < size_); consumed_ += size; } private: char* data_; size_t size_, consumed_; SocketAddress from_; }; struct MessageAddress : public MessageData { explicit MessageAddress(const SocketAddress& a) : addr(a) { } SocketAddress addr; }; // Implements the socket interface using the virtual network. Packets are // passed as messages using the message queue of the socket server. class VirtualSocket : public AsyncSocket, public MessageHandler { public: VirtualSocket(VirtualSocketServer* server, int family, int type, bool async) : server_(server), family_(family), type_(type), async_(async), state_(CS_CLOSED), error_(0), listen_queue_(NULL), write_enabled_(false), network_size_(0), recv_buffer_size_(0), bound_(false), was_any_(false) { ASSERT((type_ == SOCK_DGRAM) || (type_ == SOCK_STREAM)); ASSERT(async_ || (type_ != SOCK_STREAM)); // We only support async streams } virtual ~VirtualSocket() { Close(); for (RecvBuffer::iterator it = recv_buffer_.begin(); it != recv_buffer_.end(); ++it) { delete *it; } } virtual SocketAddress GetLocalAddress() const { return local_addr_; } virtual SocketAddress GetRemoteAddress() const { return remote_addr_; } // Used by server sockets to set the local address without binding. void SetLocalAddress(const SocketAddress& addr) { local_addr_ = addr; } virtual int Bind(const SocketAddress& addr) { if (!local_addr_.IsNil()) { error_ = EINVAL; return -1; } local_addr_ = addr; int result = server_->Bind(this, &local_addr_); if (result != 0) { local_addr_.Clear(); error_ = EADDRINUSE; } else { bound_ = true; was_any_ = addr.IsAnyIP(); } return result; } virtual int Connect(const SocketAddress& addr) { return InitiateConnect(addr, true); } virtual int Close() { if (!local_addr_.IsNil() && bound_) { // Remove from the binding table. server_->Unbind(local_addr_, this); bound_ = false; } if (SOCK_STREAM == type_) { // Cancel pending sockets if (listen_queue_) { while (!listen_queue_->empty()) { SocketAddress addr = listen_queue_->front(); // Disconnect listening socket. server_->Disconnect(server_->LookupBinding(addr)); listen_queue_->pop_front(); } delete listen_queue_; listen_queue_ = NULL; } // Disconnect stream sockets if (CS_CONNECTED == state_) { // Disconnect remote socket, check if it is a child of a server socket. VirtualSocket* socket = server_->LookupConnection(local_addr_, remote_addr_); if (!socket) { // Not a server socket child, then see if it is bound. // TODO: If this is indeed a server socket that has no // children this will cause the server socket to be // closed. This might lead to unexpected results, how to fix this? socket = server_->LookupBinding(remote_addr_); } server_->Disconnect(socket); // Remove mapping for both directions. server_->RemoveConnection(remote_addr_, local_addr_); server_->RemoveConnection(local_addr_, remote_addr_); } // Cancel potential connects MessageList msgs; if (server_->msg_queue_) { server_->msg_queue_->Clear(this, MSG_ID_CONNECT, &msgs); } for (MessageList::iterator it = msgs.begin(); it != msgs.end(); ++it) { ASSERT(NULL != it->pdata); MessageAddress* data = static_cast(it->pdata); // Lookup remote side. VirtualSocket* socket = server_->LookupConnection(local_addr_, data->addr); if (socket) { // Server socket, remote side is a socket retreived by // accept. Accepted sockets are not bound so we will not // find it by looking in the bindings table. server_->Disconnect(socket); server_->RemoveConnection(local_addr_, data->addr); } else { server_->Disconnect(server_->LookupBinding(data->addr)); } delete data; } // Clear incoming packets and disconnect messages if (server_->msg_queue_) { server_->msg_queue_->Clear(this); } } state_ = CS_CLOSED; local_addr_.Clear(); remote_addr_.Clear(); return 0; } virtual int Send(const void *pv, size_t cb) { if (CS_CONNECTED != state_) { error_ = ENOTCONN; return -1; } if (SOCK_DGRAM == type_) { return SendUdp(pv, cb, remote_addr_); } else { return SendTcp(pv, cb); } } virtual int SendTo(const void *pv, size_t cb, const SocketAddress& addr) { if (SOCK_DGRAM == type_) { return SendUdp(pv, cb, addr); } else { if (CS_CONNECTED != state_) { error_ = ENOTCONN; return -1; } return SendTcp(pv, cb); } } virtual int Recv(void *pv, size_t cb) { SocketAddress addr; return RecvFrom(pv, cb, &addr); } virtual int RecvFrom(void *pv, size_t cb, SocketAddress *paddr) { // If we don't have a packet, then either error or wait for one to arrive. if (recv_buffer_.empty()) { if (async_) { error_ = EAGAIN; return -1; } while (recv_buffer_.empty()) { Message msg; server_->msg_queue_->Get(&msg); server_->msg_queue_->Dispatch(&msg); } } // Return the packet at the front of the queue. Packet* packet = recv_buffer_.front(); size_t data_read = _min(cb, packet->size()); memcpy(pv, packet->data(), data_read); *paddr = packet->from(); if (data_read < packet->size()) { packet->Consume(data_read); } else { recv_buffer_.pop_front(); delete packet; } if (SOCK_STREAM == type_) { bool was_full = (recv_buffer_size_ == server_->recv_buffer_capacity_); recv_buffer_size_ -= data_read; if (was_full) { VirtualSocket* sender = server_->LookupBinding(remote_addr_); ASSERT(NULL != sender); server_->SendTcp(sender); } } return static_cast(data_read); } virtual int Listen(int backlog) { ASSERT(SOCK_STREAM == type_); ASSERT(CS_CLOSED == state_); if (local_addr_.IsNil()) { error_ = EINVAL; return -1; } ASSERT(NULL == listen_queue_); listen_queue_ = new ListenQueue; state_ = CS_CONNECTING; return 0; } virtual VirtualSocket* Accept(SocketAddress *paddr) { if (NULL == listen_queue_) { error_ = EINVAL; return NULL; } while (!listen_queue_->empty()) { VirtualSocket* socket = new VirtualSocket(server_, AF_INET, type_, async_); // Set the new local address to the same as this server socket. socket->SetLocalAddress(local_addr_); // Sockets made from a socket that 'was Any' need to inherit that. socket->set_was_any(was_any_); SocketAddress remote_addr(listen_queue_->front()); int result = socket->InitiateConnect(remote_addr, false); listen_queue_->pop_front(); if (result != 0) { delete socket; continue; } socket->CompleteConnect(remote_addr, false); if (paddr) { *paddr = remote_addr; } return socket; } error_ = EWOULDBLOCK; return NULL; } virtual int GetError() const { return error_; } virtual void SetError(int error) { error_ = error; } virtual ConnState GetState() const { return state_; } virtual int GetOption(Option opt, int* value) { OptionsMap::const_iterator it = options_map_.find(opt); if (it == options_map_.end()) { return -1; } *value = it->second; return 0; // 0 is success to emulate getsockopt() } virtual int SetOption(Option opt, int value) { options_map_[opt] = value; return 0; // 0 is success to emulate setsockopt() } virtual int EstimateMTU(uint16* mtu) { if (CS_CONNECTED != state_) return ENOTCONN; else return 65536; } void OnMessage(Message *pmsg) { if (pmsg->message_id == MSG_ID_PACKET) { //ASSERT(!local_addr_.IsAny()); ASSERT(NULL != pmsg->pdata); Packet* packet = static_cast(pmsg->pdata); recv_buffer_.push_back(packet); if (async_) { SignalReadEvent(this); } } else if (pmsg->message_id == MSG_ID_CONNECT) { ASSERT(NULL != pmsg->pdata); MessageAddress* data = static_cast(pmsg->pdata); if (listen_queue_ != NULL) { listen_queue_->push_back(data->addr); if (async_) { SignalReadEvent(this); } } else if ((SOCK_STREAM == type_) && (CS_CONNECTING == state_)) { CompleteConnect(data->addr, true); } else { LOG(LS_VERBOSE) << "Socket at " << local_addr_ << " is not listening"; server_->Disconnect(server_->LookupBinding(data->addr)); } delete data; } else if (pmsg->message_id == MSG_ID_DISCONNECT) { ASSERT(SOCK_STREAM == type_); if (CS_CLOSED != state_) { int error = (CS_CONNECTING == state_) ? ECONNREFUSED : 0; state_ = CS_CLOSED; remote_addr_.Clear(); if (async_) { SignalCloseEvent(this, error); } } } else { ASSERT(false); } } bool was_any() { return was_any_; } void set_was_any(bool was_any) { was_any_ = was_any; } private: struct NetworkEntry { size_t size; uint32 done_time; }; typedef std::deque ListenQueue; typedef std::deque NetworkQueue; typedef std::vector SendBuffer; typedef std::list RecvBuffer; typedef std::map OptionsMap; int InitiateConnect(const SocketAddress& addr, bool use_delay) { if (!remote_addr_.IsNil()) { error_ = (CS_CONNECTED == state_) ? EISCONN : EINPROGRESS; return -1; } if (local_addr_.IsNil()) { // If there's no local address set, grab a random one in the correct AF. int result = 0; if (addr.ipaddr().family() == AF_INET) { result = Bind(SocketAddress("0.0.0.0", 0)); } else if (addr.ipaddr().family() == AF_INET6) { result = Bind(SocketAddress("::", 0)); } if (result != 0) { return result; } } if (type_ == SOCK_DGRAM) { remote_addr_ = addr; state_ = CS_CONNECTED; } else { int result = server_->Connect(this, addr, use_delay); if (result != 0) { error_ = EHOSTUNREACH; return -1; } state_ = CS_CONNECTING; } return 0; } void CompleteConnect(const SocketAddress& addr, bool notify) { ASSERT(CS_CONNECTING == state_); remote_addr_ = addr; state_ = CS_CONNECTED; server_->AddConnection(remote_addr_, local_addr_, this); if (async_ && notify) { SignalConnectEvent(this); } } int SendUdp(const void* pv, size_t cb, const SocketAddress& addr) { // If we have not been assigned a local port, then get one. if (local_addr_.IsNil()) { local_addr_ = EmptySocketAddressWithFamily(addr.ipaddr().family()); int result = server_->Bind(this, &local_addr_); if (result != 0) { local_addr_.Clear(); error_ = EADDRINUSE; return result; } } // Send the data in a message to the appropriate socket. return server_->SendUdp(this, static_cast(pv), cb, addr); } int SendTcp(const void* pv, size_t cb) { size_t capacity = server_->send_buffer_capacity_ - send_buffer_.size(); if (0 == capacity) { write_enabled_ = true; error_ = EWOULDBLOCK; return -1; } size_t consumed = _min(cb, capacity); const char* cpv = static_cast(pv); send_buffer_.insert(send_buffer_.end(), cpv, cpv + consumed); server_->SendTcp(this); return static_cast(consumed); } VirtualSocketServer* server_; int family_; int type_; bool async_; ConnState state_; int error_; SocketAddress local_addr_; SocketAddress remote_addr_; // Pending sockets which can be Accepted ListenQueue* listen_queue_; // Data which tcp has buffered for sending SendBuffer send_buffer_; bool write_enabled_; // Critical section to protect the recv_buffer and queue_ CriticalSection crit_; // Network model that enforces bandwidth and capacity constraints NetworkQueue network_; size_t network_size_; // Data which has been received from the network RecvBuffer recv_buffer_; // The amount of data which is in flight or in recv_buffer_ size_t recv_buffer_size_; // Is this socket bound? bool bound_; // When we bind a socket to Any, VSS's Bind gives it another address. For // dual-stack sockets, we want to distinguish between sockets that were // explicitly given a particular address and sockets that had one picked // for them by VSS. bool was_any_; // Store the options that are set OptionsMap options_map_; friend class VirtualSocketServer; }; VirtualSocketServer::VirtualSocketServer(SocketServer* ss) : server_(ss), server_owned_(false), msg_queue_(NULL), stop_on_idle_(false), network_delay_(Time()), next_ipv4_(kInitialNextIPv4), next_ipv6_(kInitialNextIPv6), next_port_(kFirstEphemeralPort), bindings_(new AddressMap()), connections_(new ConnectionMap()), bandwidth_(0), network_capacity_(kDefaultNetworkCapacity), send_buffer_capacity_(kDefaultTcpBufferSize), recv_buffer_capacity_(kDefaultTcpBufferSize), delay_mean_(0), delay_stddev_(0), delay_samples_(NUM_SAMPLES), delay_dist_(NULL), drop_prob_(0.0) { if (!server_) { server_ = new PhysicalSocketServer(); server_owned_ = true; } UpdateDelayDistribution(); } VirtualSocketServer::~VirtualSocketServer() { delete bindings_; delete connections_; delete delay_dist_; if (server_owned_) { delete server_; } } IPAddress VirtualSocketServer::GetNextIP(int family) { if (family == AF_INET) { IPAddress next_ip(next_ipv4_); next_ipv4_.s_addr = HostToNetwork32(NetworkToHost32(next_ipv4_.s_addr) + 1); return next_ip; } else if (family == AF_INET6) { IPAddress next_ip(next_ipv6_); uint32* as_ints = reinterpret_cast(&next_ipv6_.s6_addr); as_ints[3] += 1; return next_ip; } return IPAddress(); } uint16 VirtualSocketServer::GetNextPort() { uint16 port = next_port_; if (next_port_ < kLastEphemeralPort) { ++next_port_; } else { next_port_ = kFirstEphemeralPort; } return port; } Socket* VirtualSocketServer::CreateSocket(int type) { return CreateSocket(AF_INET, type); } Socket* VirtualSocketServer::CreateSocket(int family, int type) { return CreateSocketInternal(family, type); } AsyncSocket* VirtualSocketServer::CreateAsyncSocket(int type) { return CreateAsyncSocket(AF_INET, type); } AsyncSocket* VirtualSocketServer::CreateAsyncSocket(int family, int type) { return CreateSocketInternal(family, type); } VirtualSocket* VirtualSocketServer::CreateSocketInternal(int family, int type) { return new VirtualSocket(this, family, type, true); } void VirtualSocketServer::SetMessageQueue(MessageQueue* msg_queue) { msg_queue_ = msg_queue; if (msg_queue_) { msg_queue_->SignalQueueDestroyed.connect(this, &VirtualSocketServer::OnMessageQueueDestroyed); } } bool VirtualSocketServer::Wait(int cmsWait, bool process_io) { ASSERT(msg_queue_ == Thread::Current()); if (stop_on_idle_ && Thread::Current()->empty()) { return false; } return socketserver()->Wait(cmsWait, process_io); } void VirtualSocketServer::WakeUp() { socketserver()->WakeUp(); } bool VirtualSocketServer::ProcessMessagesUntilIdle() { ASSERT(msg_queue_ == Thread::Current()); stop_on_idle_ = true; while (!msg_queue_->empty()) { Message msg; if (msg_queue_->Get(&msg, kForever)) { msg_queue_->Dispatch(&msg); } } stop_on_idle_ = false; return !msg_queue_->IsQuitting(); } int VirtualSocketServer::Bind(VirtualSocket* socket, const SocketAddress& addr) { ASSERT(NULL != socket); // Address must be completely specified at this point ASSERT(!IPIsUnspec(addr.ipaddr())); ASSERT(addr.port() != 0); // Normalize the address (turns v6-mapped addresses into v4-addresses). SocketAddress normalized(addr.ipaddr().Normalized(), addr.port()); AddressMap::value_type entry(normalized, socket); return bindings_->insert(entry).second ? 0 : -1; } int VirtualSocketServer::Bind(VirtualSocket* socket, SocketAddress* addr) { ASSERT(NULL != socket); if (IPIsAny(addr->ipaddr())) { addr->SetIP(GetNextIP(addr->ipaddr().family())); } else if (!IPIsUnspec(addr->ipaddr())) { addr->SetIP(addr->ipaddr().Normalized()); } else { ASSERT(false); } if (addr->port() == 0) { for (int i = 0; i < kEphemeralPortCount; ++i) { addr->SetPort(GetNextPort()); if (bindings_->find(*addr) == bindings_->end()) { break; } } } return Bind(socket, *addr); } VirtualSocket* VirtualSocketServer::LookupBinding(const SocketAddress& addr) { SocketAddress normalized(addr.ipaddr().Normalized(), addr.port()); AddressMap::iterator it = bindings_->find(normalized); return (bindings_->end() != it) ? it->second : NULL; } int VirtualSocketServer::Unbind(const SocketAddress& addr, VirtualSocket* socket) { SocketAddress normalized(addr.ipaddr().Normalized(), addr.port()); ASSERT((*bindings_)[normalized] == socket); bindings_->erase(bindings_->find(normalized)); return 0; } void VirtualSocketServer::AddConnection(const SocketAddress& local, const SocketAddress& remote, VirtualSocket* remote_socket) { // Add this socket pair to our routing table. This will allow // multiple clients to connect to the same server address. SocketAddress local_normalized(local.ipaddr().Normalized(), local.port()); SocketAddress remote_normalized(remote.ipaddr().Normalized(), remote.port()); SocketAddressPair address_pair(local_normalized, remote_normalized); connections_->insert(std::pair(address_pair, remote_socket)); } VirtualSocket* VirtualSocketServer::LookupConnection( const SocketAddress& local, const SocketAddress& remote) { SocketAddress local_normalized(local.ipaddr().Normalized(), local.port()); SocketAddress remote_normalized(remote.ipaddr().Normalized(), remote.port()); SocketAddressPair address_pair(local_normalized, remote_normalized); ConnectionMap::iterator it = connections_->find(address_pair); return (connections_->end() != it) ? it->second : NULL; } void VirtualSocketServer::RemoveConnection(const SocketAddress& local, const SocketAddress& remote) { SocketAddress local_normalized(local.ipaddr().Normalized(), local.port()); SocketAddress remote_normalized(remote.ipaddr().Normalized(), remote.port()); SocketAddressPair address_pair(local_normalized, remote_normalized); connections_->erase(address_pair); } static double Random() { return static_cast(rand()) / RAND_MAX; } int VirtualSocketServer::Connect(VirtualSocket* socket, const SocketAddress& remote_addr, bool use_delay) { uint32 delay = use_delay ? GetRandomTransitDelay() : 0; VirtualSocket* remote = LookupBinding(remote_addr); if (!CanInteractWith(socket, remote)) { LOG(LS_INFO) << "Address family mismatch between " << socket->GetLocalAddress() << " and " << remote_addr; return -1; } if (remote != NULL) { SocketAddress addr = socket->GetLocalAddress(); msg_queue_->PostDelayed(delay, remote, MSG_ID_CONNECT, new MessageAddress(addr)); } else { LOG(LS_INFO) << "No one listening at " << remote_addr; msg_queue_->PostDelayed(delay, socket, MSG_ID_DISCONNECT); } return 0; } bool VirtualSocketServer::Disconnect(VirtualSocket* socket) { if (socket) { // Remove the mapping. msg_queue_->Post(socket, MSG_ID_DISCONNECT); return true; } return false; } int VirtualSocketServer::SendUdp(VirtualSocket* socket, const char* data, size_t data_size, const SocketAddress& remote_addr) { // See if we want to drop this packet. if (Random() < drop_prob_) { LOG(LS_VERBOSE) << "Dropping packet: bad luck"; return static_cast(data_size); } VirtualSocket* recipient = LookupBinding(remote_addr); if (!recipient) { // Make a fake recipient for address family checking. scoped_ptr dummy_socket( CreateSocketInternal(AF_INET, SOCK_DGRAM)); dummy_socket->SetLocalAddress(remote_addr); if (!CanInteractWith(socket, dummy_socket.get())) { LOG(LS_VERBOSE) << "Incompatible address families: " << socket->GetLocalAddress() << " and " << remote_addr; return -1; } LOG(LS_VERBOSE) << "No one listening at " << remote_addr; return static_cast(data_size); } if (!CanInteractWith(socket, recipient)) { LOG(LS_VERBOSE) << "Incompatible address families: " << socket->GetLocalAddress() << " and " << remote_addr; return -1; } CritScope cs(&socket->crit_); uint32 cur_time = Time(); PurgeNetworkPackets(socket, cur_time); // Determine whether we have enough bandwidth to accept this packet. To do // this, we need to update the send queue. Once we know it's current size, // we know whether we can fit this packet. // // NOTE: There are better algorithms for maintaining such a queue (such as // "Derivative Random Drop"); however, this algorithm is a more accurate // simulation of what a normal network would do. size_t packet_size = data_size + UDP_HEADER_SIZE; if (socket->network_size_ + packet_size > network_capacity_) { LOG(LS_VERBOSE) << "Dropping packet: network capacity exceeded"; return static_cast(data_size); } AddPacketToNetwork(socket, recipient, cur_time, data, data_size, UDP_HEADER_SIZE, false); return static_cast(data_size); } void VirtualSocketServer::SendTcp(VirtualSocket* socket) { // TCP can't send more data than will fill up the receiver's buffer. // We track the data that is in the buffer plus data in flight using the // recipient's recv_buffer_size_. Anything beyond that must be stored in the // sender's buffer. We will trigger the buffered data to be sent when data // is read from the recv_buffer. // Lookup the local/remote pair in the connections table. VirtualSocket* recipient = LookupConnection(socket->local_addr_, socket->remote_addr_); if (!recipient) { LOG(LS_VERBOSE) << "Sending data to no one."; return; } CritScope cs(&socket->crit_); uint32 cur_time = Time(); PurgeNetworkPackets(socket, cur_time); while (true) { size_t available = recv_buffer_capacity_ - recipient->recv_buffer_size_; size_t max_data_size = _min(available, TCP_MSS - TCP_HEADER_SIZE); size_t data_size = _min(socket->send_buffer_.size(), max_data_size); if (0 == data_size) break; AddPacketToNetwork(socket, recipient, cur_time, &socket->send_buffer_[0], data_size, TCP_HEADER_SIZE, true); recipient->recv_buffer_size_ += data_size; size_t new_buffer_size = socket->send_buffer_.size() - data_size; // Avoid undefined access beyond the last element of the vector. // This only happens when new_buffer_size is 0. if (data_size < socket->send_buffer_.size()) { // memmove is required for potentially overlapping source/destination. memmove(&socket->send_buffer_[0], &socket->send_buffer_[data_size], new_buffer_size); } socket->send_buffer_.resize(new_buffer_size); } if (socket->write_enabled_ && (socket->send_buffer_.size() < send_buffer_capacity_)) { socket->write_enabled_ = false; socket->SignalWriteEvent(socket); } } void VirtualSocketServer::AddPacketToNetwork(VirtualSocket* sender, VirtualSocket* recipient, uint32 cur_time, const char* data, size_t data_size, size_t header_size, bool ordered) { VirtualSocket::NetworkEntry entry; entry.size = data_size + header_size; sender->network_size_ += entry.size; uint32 send_delay = SendDelay(static_cast(sender->network_size_)); entry.done_time = cur_time + send_delay; sender->network_.push_back(entry); // Find the delay for crossing the many virtual hops of the network. uint32 transit_delay = GetRandomTransitDelay(); // Post the packet as a message to be delivered (on our own thread) Packet* p = new Packet(data, data_size, sender->local_addr_); uint32 ts = TimeAfter(send_delay + transit_delay); if (ordered) { // Ensure that new packets arrive after previous ones // TODO: consider ordering on a per-socket basis, since this // introduces artifical delay. ts = TimeMax(ts, network_delay_); } msg_queue_->PostAt(ts, recipient, MSG_ID_PACKET, p); network_delay_ = TimeMax(ts, network_delay_); } void VirtualSocketServer::PurgeNetworkPackets(VirtualSocket* socket, uint32 cur_time) { while (!socket->network_.empty() && (socket->network_.front().done_time <= cur_time)) { ASSERT(socket->network_size_ >= socket->network_.front().size); socket->network_size_ -= socket->network_.front().size; socket->network_.pop_front(); } } uint32 VirtualSocketServer::SendDelay(uint32 size) { if (bandwidth_ == 0) return 0; else return 1000 * size / bandwidth_; } #if 0 void PrintFunction(std::vector >* f) { return; double sum = 0; for (uint32 i = 0; i < f->size(); ++i) { std::cout << (*f)[i].first << '\t' << (*f)[i].second << std::endl; sum += (*f)[i].second; } if (!f->empty()) { const double mean = sum / f->size(); double sum_sq_dev = 0; for (uint32 i = 0; i < f->size(); ++i) { double dev = (*f)[i].second - mean; sum_sq_dev += dev * dev; } std::cout << "Mean = " << mean << " StdDev = " << sqrt(sum_sq_dev / f->size()) << std::endl; } } #endif // void VirtualSocketServer::UpdateDelayDistribution() { Function* dist = CreateDistribution(delay_mean_, delay_stddev_, delay_samples_); // We take a lock just to make sure we don't leak memory. { CritScope cs(&delay_crit_); delete delay_dist_; delay_dist_ = dist; } } static double PI = 4 * atan(1.0); static double Normal(double x, double mean, double stddev) { double a = (x - mean) * (x - mean) / (2 * stddev * stddev); return exp(-a) / (stddev * sqrt(2 * PI)); } #if 0 // static unused gives a warning static double Pareto(double x, double min, double k) { if (x < min) return 0; else return k * std::pow(min, k) / std::pow(x, k+1); } #endif VirtualSocketServer::Function* VirtualSocketServer::CreateDistribution( uint32 mean, uint32 stddev, uint32 samples) { Function* f = new Function(); if (0 == stddev) { f->push_back(Point(mean, 1.0)); } else { double start = 0; if (mean >= 4 * static_cast(stddev)) start = mean - 4 * static_cast(stddev); double end = mean + 4 * static_cast(stddev); for (uint32 i = 0; i < samples; i++) { double x = start + (end - start) * i / (samples - 1); double y = Normal(x, mean, stddev); f->push_back(Point(x, y)); } } return Resample(Invert(Accumulate(f)), 0, 1, samples); } uint32 VirtualSocketServer::GetRandomTransitDelay() { size_t index = rand() % delay_dist_->size(); double delay = (*delay_dist_)[index].second; //LOG_F(LS_INFO) << "random[" << index << "] = " << delay; return static_cast(delay); } struct FunctionDomainCmp { bool operator()(const VirtualSocketServer::Point& p1, const VirtualSocketServer::Point& p2) { return p1.first < p2.first; } bool operator()(double v1, const VirtualSocketServer::Point& p2) { return v1 < p2.first; } bool operator()(const VirtualSocketServer::Point& p1, double v2) { return p1.first < v2; } }; VirtualSocketServer::Function* VirtualSocketServer::Accumulate(Function* f) { ASSERT(f->size() >= 1); double v = 0; for (Function::size_type i = 0; i < f->size() - 1; ++i) { double dx = (*f)[i + 1].first - (*f)[i].first; double avgy = ((*f)[i + 1].second + (*f)[i].second) / 2; (*f)[i].second = v; v = v + dx * avgy; } (*f)[f->size()-1].second = v; return f; } VirtualSocketServer::Function* VirtualSocketServer::Invert(Function* f) { for (Function::size_type i = 0; i < f->size(); ++i) std::swap((*f)[i].first, (*f)[i].second); std::sort(f->begin(), f->end(), FunctionDomainCmp()); return f; } VirtualSocketServer::Function* VirtualSocketServer::Resample( Function* f, double x1, double x2, uint32 samples) { Function* g = new Function(); for (size_t i = 0; i < samples; i++) { double x = x1 + (x2 - x1) * i / (samples - 1); double y = Evaluate(f, x); g->push_back(Point(x, y)); } delete f; return g; } double VirtualSocketServer::Evaluate(Function* f, double x) { Function::iterator iter = std::lower_bound(f->begin(), f->end(), x, FunctionDomainCmp()); if (iter == f->begin()) { return (*f)[0].second; } else if (iter == f->end()) { ASSERT(f->size() >= 1); return (*f)[f->size() - 1].second; } else if (iter->first == x) { return iter->second; } else { double x1 = (iter - 1)->first; double y1 = (iter - 1)->second; double x2 = iter->first; double y2 = iter->second; return y1 + (y2 - y1) * (x - x1) / (x2 - x1); } } bool VirtualSocketServer::CanInteractWith(VirtualSocket* local, VirtualSocket* remote) { if (!local || !remote) { return false; } IPAddress local_ip = local->GetLocalAddress().ipaddr(); IPAddress remote_ip = remote->GetLocalAddress().ipaddr(); IPAddress local_normalized = local_ip.Normalized(); IPAddress remote_normalized = remote_ip.Normalized(); // Check if the addresses are the same family after Normalization (turns // mapped IPv6 address into IPv4 addresses). // This will stop unmapped V6 addresses from talking to mapped V6 addresses. if (local_normalized.family() == remote_normalized.family()) { return true; } // If ip1 is IPv4 and ip2 is :: and ip2 is not IPV6_V6ONLY. int remote_v6_only = 0; remote->GetOption(Socket::OPT_IPV6_V6ONLY, &remote_v6_only); if (local_ip.family() == AF_INET && !remote_v6_only && IPIsAny(remote_ip)) { return true; } // Same check, backwards. int local_v6_only = 0; local->GetOption(Socket::OPT_IPV6_V6ONLY, &local_v6_only); if (remote_ip.family() == AF_INET && !local_v6_only && IPIsAny(local_ip)) { return true; } // Check to see if either socket was explicitly bound to IPv6-any. // These sockets can talk with anyone. if (local_ip.family() == AF_INET6 && local->was_any()) { return true; } if (remote_ip.family() == AF_INET6 && remote->was_any()) { return true; } return false; } } // namespace talk_base