| 1 | // Copyright (C) Microsoft Corporation. All rights reserved. |
| 2 | |
| 3 | #include <arpa/inet.h> |
| 4 | #include <sys/epoll.h> |
| 5 | #include <netinet/in.h> |
| 6 | #include <sys/socket.h> |
| 7 | #include "RuntimeErrorWithSourceLocation.h" |
| 8 | #include "DnsServer.h" |
| 9 | #include "Syscall.h" |
| 10 | #include "util.h" |
| 11 | |
| 12 | // Port used by DNS server |
| 13 | constexpr int c_dnsServerPort = 53; |
| 14 | // Max number of events to be returned by epoll_wait() |
| 15 | constexpr int c_epollWaitMaxEvents = 100; |
| 16 | // Maximum size of DNS over UDP requests is 4096 bytes (max size is reached for EDNS UDP requests) |
| 17 | constexpr int c_maxUdpDnsBufferSize = 4096; |
| 18 | // Maximum time to wait for a tunneled UDP DNS response |
| 19 | constexpr auto c_udpRequestTimeout = std::chrono::seconds{60}; |
| 20 | // Max number of pending connections in the TCP listen queue |
| 21 | constexpr int c_maxListenBacklog = 1000; |
| 22 | |
| 23 | DnsServer::DnsServer(DnsTunnelingCallback&& tunnelDnsRequest) : m_tunnelDnsRequest(std::move(tunnelDnsRequest)) |
| 24 | { |
| 25 | } |
| 26 | |
| 27 | DnsServer::~DnsServer() noexcept |
| 28 | { |
| 29 | Stop(); |
| 30 | } |
| 31 | |
| 32 | void DnsServer::Start(const std::string& ipAddress) noexcept |
| 33 | try |
| 34 | { |
| 35 | // Create epoll handler fd. 0 represents default flags |
| 36 | m_epollFd = Syscall(epoll_create1, 0); |
| 37 | |
| 38 | StartUdpDnsServer(ipAddress); |
| 39 | StartTcpDnsServer(ipAddress); |
| 40 | |
| 41 | // Create and register the shutdown pipe with epoll |
| 42 | m_shutdownServerLoopPipe = wil::unique_pipe::create(0); |
| 43 | |
| 44 | epoll_event event{}; |
| 45 | event.events = EPOLLIN; |
| 46 | event.data.fd = m_shutdownServerLoopPipe.read().get(); |
| 47 | Syscall(epoll_ctl, m_epollFd.get(), EPOLL_CTL_ADD, m_shutdownServerLoopPipe.read().get(), &event); |
| 48 | |
| 49 | // Start server loop |
| 50 | m_serverThread = std::thread([this]() { ServerLoop(); }); |
| 51 | } |
| 52 | CATCH_LOG() |
| 53 | |
| 54 | void DnsServer::StartUdpDnsServer(const std::string& ipAddress) noexcept |
| 55 | try |
| 56 | { |
| 57 | sockaddr_in serverAddr{}; |
| 58 | |
| 59 | serverAddr.sin_family = AF_INET; |
| 60 | Syscall(inet_pton, AF_INET, ipAddress.c_str(), &serverAddr.sin_addr); |
| 61 | serverAddr.sin_port = htons(c_dnsServerPort); |
| 62 | |
| 63 | // Create IPv4 UDP socket |
| 64 | m_udpSocket = Syscall(socket, AF_INET, SOCK_DGRAM | SOCK_NONBLOCK, 0); |
| 65 | |
| 66 | // Bind socket |
| 67 | Syscall(bind, m_udpSocket.get(), reinterpret_cast<sockaddr*>(&serverAddr), sizeof(serverAddr)); |
| 68 | |
| 69 | // Configure epoll to track the UDP socket. EPOLLIN is used to get epoll notifications |
| 70 | // whenever there is data available to be read from the socket |
| 71 | epoll_event event{}; |
| 72 | event.events = EPOLLIN; |
| 73 | event.data.fd = m_udpSocket.get(); |
| 74 | Syscall(epoll_ctl, m_epollFd.get(), EPOLL_CTL_ADD, m_udpSocket.get(), &event); |
| 75 | |
| 76 | GNS_LOG_INFO("Successfully started UDP server on IP {}", ipAddress.c_str()); |
| 77 | } |
| 78 | CATCH_LOG() |
| 79 | |
| 80 | void DnsServer::StartTcpDnsServer(const std::string& ipAddress) noexcept |
| 81 | try |
| 82 | { |
| 83 | sockaddr_in serverAddr{}; |
| 84 | |
| 85 | serverAddr.sin_family = AF_INET; |
| 86 | Syscall(inet_pton, AF_INET, ipAddress.c_str(), &serverAddr.sin_addr); |
| 87 | serverAddr.sin_port = htons(c_dnsServerPort); |
| 88 | |
| 89 | // Create IPv4 TCP socket |
| 90 | m_tcpListenSocket = Syscall(socket, AF_INET, SOCK_STREAM | SOCK_NONBLOCK, 0); |
| 91 | |
| 92 | // Bind socket |
| 93 | Syscall(bind, m_tcpListenSocket.get(), reinterpret_cast<sockaddr*>(&serverAddr), sizeof(serverAddr)); |
| 94 | |
| 95 | // Listen for incoming connections |
| 96 | Syscall(listen, m_tcpListenSocket.get(), c_maxListenBacklog); |
| 97 | |
| 98 | // Configure epoll to track the TCP listening socket. EPOLLIN is used to get epoll notifications |
| 99 | // whenever there is a new incoming TCP connection. |
| 100 | epoll_event event{}; |
| 101 | event.events = EPOLLIN; |
| 102 | event.data.fd = m_tcpListenSocket.get(); |
| 103 | Syscall(epoll_ctl, m_epollFd.get(), EPOLL_CTL_ADD, m_tcpListenSocket.get(), &event); |
| 104 | |
| 105 | GNS_LOG_INFO("Successfully started TCP server on IP {}", ipAddress.c_str()); |
| 106 | } |
| 107 | CATCH_LOG(); |
| 108 | |
| 109 | void DnsServer::HandleUdpDnsResponse(const gsl::span<gsl::byte> dnsBuffer, const LX_GNS_DNS_CLIENT_IDENTIFIER& dnsClientIdentifier) noexcept |
| 110 | try |
| 111 | { |
| 112 | GNS_LOG_INFO("New UDP DNS response DNS buffer size: {}, UDP request id: {}", dnsBuffer.size(), dnsClientIdentifier.DnsClientId); |
| 113 | |
| 114 | std::scoped_lock<std::mutex> lock{m_udpLock}; |
| 115 | |
| 116 | auto it = m_udpRequests.find(dnsClientIdentifier.DnsClientId); |
| 117 | if (it == m_udpRequests.end()) |
| 118 | { |
| 119 | GNS_LOG_ERROR("Received a response for a UDP request that is not tracked, UDP request id: {}", dnsClientIdentifier.DnsClientId); |
| 120 | return; |
| 121 | } |
| 122 | |
| 123 | // Stop tracking the request, irrespective of the DNS response being successfully sent |
| 124 | const auto removeDnsRequest = wil::scope_exit([&] { |
| 125 | m_udpRequestExpirations.erase(it->second.m_expiration); |
| 126 | m_udpRequests.erase(it); |
| 127 | }); |
| 128 | |
| 129 | sockaddr_in& remoteAddr = it->second.m_remoteAddress; |
| 130 | |
| 131 | // Send DNS response buffer back to the Linux DNS client |
| 132 | int bufferSize = dnsBuffer.size(); |
| 133 | int totalBytesSent = 0; |
| 134 | |
| 135 | while (totalBytesSent < bufferSize) |
| 136 | { |
| 137 | int bytesSent = Syscall( |
| 138 | sendto, m_udpSocket.get(), dnsBuffer.data() + totalBytesSent, bufferSize - totalBytesSent, 0, reinterpret_cast<sockaddr*>(&remoteAddr), sizeof(remoteAddr)); |
| 139 | totalBytesSent += bytesSent; |
| 140 | } |
| 141 | } |
| 142 | CATCH_LOG() |
| 143 | |
| 144 | void DnsServer::HandleTcpDnsResponse(const gsl::span<gsl::byte> dnsBuffer, const LX_GNS_DNS_CLIENT_IDENTIFIER& dnsClientIdentifier) noexcept |
| 145 | try |
| 146 | { |
| 147 | GNS_LOG_INFO( |
| 148 | "New TCP DNS response " |
| 149 | "DNS buffer size: {}, TCP connection id: {}", |
| 150 | dnsBuffer.size(), |
| 151 | dnsClientIdentifier.DnsClientId); |
| 152 | |
| 153 | std::scoped_lock<std::mutex> lock{m_tcpLock}; |
| 154 | |
| 155 | auto it = m_tcpConnectionContexts.find(dnsClientIdentifier.DnsClientId); |
| 156 | if (it == m_tcpConnectionContexts.end()) |
| 157 | { |
| 158 | GNS_LOG_ERROR("Received a response for an untracked TCP connection id: {}", dnsClientIdentifier.DnsClientId); |
| 159 | return; |
| 160 | } |
| 161 | |
| 162 | auto tcpConnection = it->second->m_tcpConnection.get(); |
| 163 | |
| 164 | // Send DNS response buffer back to the Linux DNS client. |
| 165 | // |
| 166 | // Note: there might be more DNS requests sent on the same TCP connection. The DNS protocol allows sending the responses in a |
| 167 | // different order than the order of the corresponding DNS requests. |
| 168 | int bufferSize = dnsBuffer.size(); |
| 169 | int totalBytesSent = 0; |
| 170 | |
| 171 | while (totalBytesSent < bufferSize) |
| 172 | { |
| 173 | int bytesSent = Syscall(write, tcpConnection, dnsBuffer.data() + totalBytesSent, bufferSize - totalBytesSent); |
| 174 | totalBytesSent += bytesSent; |
| 175 | } |
| 176 | } |
| 177 | CATCH_LOG() |
| 178 | |
| 179 | void DnsServer::HandleNewTcpConnection() noexcept |
| 180 | try |
| 181 | { |
| 182 | std::scoped_lock<std::mutex> lock{m_tcpLock}; |
| 183 | |
| 184 | // Accept new connection. Mark connection socket as non-blocking |
| 185 | wil::unique_fd connectionFd = Syscall(accept4, m_tcpListenSocket.get(), nullptr, nullptr, SOCK_NONBLOCK); |
| 186 | |
| 187 | // Get next connection id. If value reaches UINT_MAX + 1 it will be automatically reset to 0 |
| 188 | const auto connectionId = m_currentTcpConnectionId++; |
| 189 | |
| 190 | // Track the new connection |
| 191 | auto [it, _] = m_tcpConnectionContexts.emplace( |
| 192 | connectionId, std::make_unique<DnsServer::TcpConnectionContext>(connectionId, std::move(connectionFd))); |
| 193 | auto& localContext = it->second; |
| 194 | |
| 195 | auto removeContextOnError = wil::scope_exit([&] { m_tcpConnectionContexts.erase(connectionId); }); |
| 196 | |
| 197 | // Register the new connection with epoll. EPOLLIN is used to get epoll notifications |
| 198 | // whenever there is new data on the TCP connection. |
| 199 | epoll_event event{}; |
| 200 | event.events = EPOLLIN; |
| 201 | event.data.ptr = localContext.get(); |
| 202 | Syscall(epoll_ctl, m_epollFd.get(), EPOLL_CTL_ADD, localContext->m_tcpConnection.get(), &event); |
| 203 | |
| 204 | removeContextOnError.release(); |
| 205 | } |
| 206 | CATCH_LOG(); |
| 207 | |
| 208 | void DnsServer::HandleNewTcpData(TcpConnectionContext* context) noexcept |
| 209 | try |
| 210 | { |
| 211 | std::vector<gsl::byte> dnsRequest; |
| 212 | uint32_t tcpConnectionId{}; |
| 213 | |
| 214 | // Scoped m_tcpLock |
| 215 | { |
| 216 | std::scoped_lock<std::mutex> lock{m_tcpLock}; |
| 217 | |
| 218 | // In case of any failure reading data, close the connection and stop tracking it. |
| 219 | // Note: Closing the connection automatically unregisters it from epoll. |
| 220 | auto removeConnectionOnError = wil::scope_exit([&] { m_tcpConnectionContexts.erase(context->m_connectionId); }); |
| 221 | |
| 222 | // Read the remaining bytes of the current DNS request |
| 223 | int bytesReceived = Syscall( |
| 224 | recv, |
| 225 | context->m_tcpConnection.get(), |
| 226 | context->m_currentDnsRequest.data() + context->m_currentRequestOffset, |
| 227 | context->m_currentDnsRequest.size() - context->m_currentRequestOffset, |
| 228 | 0); |
| 229 | |
| 230 | // 0 bytes received indicates connection was closed by the TCP client |
| 231 | if (bytesReceived == 0) |
| 232 | { |
| 233 | return; |
| 234 | } |
| 235 | |
| 236 | context->m_currentRequestOffset += bytesReceived; |
| 237 | |
| 238 | if (context->m_currentRequestOffset == context->m_currentDnsRequest.size()) |
| 239 | { |
| 240 | // We read the 2 bytes that represent the DNS request length |
| 241 | // Resize buffer to fit the entire DNS request (2 bytes storing the request length + the actual DNS request) |
| 242 | if (context->m_currentDnsRequest.size() == c_byteCountTcpRequestLength) |
| 243 | { |
| 244 | uint16_t dnsRequestLength = 0; |
| 245 | memcpy(&dnsRequestLength, context->m_currentDnsRequest.data(), c_byteCountTcpRequestLength); |
| 246 | // The request length is stored in network byte order |
| 247 | dnsRequestLength = ntohs(dnsRequestLength); |
| 248 | |
| 249 | context->m_currentDnsRequest.resize(c_byteCountTcpRequestLength + dnsRequestLength); |
| 250 | } |
| 251 | // We read a full DNS request |
| 252 | else |
| 253 | { |
| 254 | // Move request to a local variable |
| 255 | dnsRequest = std::move(context->m_currentDnsRequest); |
| 256 | tcpConnectionId = context->m_connectionId; |
| 257 | |
| 258 | // Reset state to prepare for the next DNS request on the connection (if any) |
| 259 | context->m_currentRequestOffset = 0; |
| 260 | context->m_currentDnsRequest.resize(c_byteCountTcpRequestLength); |
| 261 | } |
| 262 | } |
| 263 | |
| 264 | removeConnectionOnError.release(); |
| 265 | } |
| 266 | |
| 267 | if (!dnsRequest.empty()) |
| 268 | { |
| 269 | // Tunnel request to Windows |
| 270 | LX_GNS_DNS_CLIENT_IDENTIFIER dnsClientIdentifier{}; |
| 271 | dnsClientIdentifier.DnsClientId = tcpConnectionId; |
| 272 | dnsClientIdentifier.Protocol = IPPROTO_TCP; |
| 273 | |
| 274 | GNS_LOG_INFO("New TCP DNS request DNS buffer size: {}, TCP connection id: {}", dnsRequest.size(), dnsClientIdentifier.DnsClientId); |
| 275 | |
| 276 | m_tunnelDnsRequest(gsl::make_span(dnsRequest), dnsClientIdentifier); |
| 277 | } |
| 278 | } |
| 279 | CATCH_LOG(); |
| 280 | |
| 281 | void DnsServer::HandleDnsResponse(const gsl::span<gsl::byte> dnsBuffer, const LX_GNS_DNS_CLIENT_IDENTIFIER& dnsClientIdentifier) noexcept |
| 282 | try |
| 283 | { |
| 284 | switch (dnsClientIdentifier.Protocol) |
| 285 | { |
| 286 | case IPPROTO_UDP: |
| 287 | { |
| 288 | HandleUdpDnsResponse(dnsBuffer, dnsClientIdentifier); |
| 289 | break; |
| 290 | } |
| 291 | case IPPROTO_TCP: |
| 292 | { |
| 293 | HandleTcpDnsResponse(dnsBuffer, dnsClientIdentifier); |
| 294 | break; |
| 295 | } |
| 296 | |
| 297 | default: |
| 298 | { |
| 299 | GNS_LOG_ERROR("Unexpected DNS protocol {}", dnsClientIdentifier.Protocol); |
| 300 | break; |
| 301 | } |
| 302 | } |
| 303 | } |
| 304 | CATCH_LOG() |
| 305 | |
| 306 | int DnsServer::ExpireUdpRequestsAndGetTimeout() noexcept |
| 307 | { |
| 308 | std::scoped_lock<std::mutex> lock{m_udpLock}; |
| 309 | const auto now = std::chrono::steady_clock::now(); |
| 310 | |
| 311 | while (!m_udpRequestExpirations.empty() && m_udpRequestExpirations.front().first <= now) |
| 312 | { |
| 313 | m_udpRequests.erase(m_udpRequestExpirations.front().second); |
| 314 | m_udpRequestExpirations.pop_front(); |
| 315 | } |
| 316 | |
| 317 | if (m_udpRequestExpirations.empty()) |
| 318 | { |
| 319 | return -1; |
| 320 | } |
| 321 | |
| 322 | return static_cast<int>(std::chrono::ceil<std::chrono::milliseconds>(m_udpRequestExpirations.front().first - now).count()); |
| 323 | } |
| 324 | |
| 325 | void DnsServer::ServerLoop() noexcept |
| 326 | { |
| 327 | UtilSetThreadName("DnsServer"); |
| 328 | |
| 329 | epoll_event events[c_epollWaitMaxEvents]; |
| 330 | memset(events, 0, sizeof(events)); |
| 331 | |
| 332 | for (;;) |
| 333 | { |
| 334 | try |
| 335 | { |
| 336 | // A fixed number of events is requested from epoll_wait (c_epollWaitMaxEvents). In case the number of ready events is |
| 337 | // greater than c_epollWaitMaxEvents, epoll will round-robin through the ready events until we get a notification for all of them. |
| 338 | const auto timeout = ExpireUdpRequestsAndGetTimeout(); |
| 339 | size_t numReadyEvents = Syscall(epoll_wait, m_epollFd.get(), events, c_epollWaitMaxEvents, timeout); |
| 340 | |
| 341 | // No event |
| 342 | if (numReadyEvents == 0) |
| 343 | { |
| 344 | continue; |
| 345 | } |
| 346 | |
| 347 | for (size_t index = 0; index < numReadyEvents; index++) |
| 348 | { |
| 349 | // Notification for the shutdown pipe == the server needs to exit |
| 350 | if (events[index].data.fd == m_shutdownServerLoopPipe.read().get()) |
| 351 | { |
| 352 | return; |
| 353 | } |
| 354 | // Notification for the listen socket == a new incoming TCP connection |
| 355 | else if (events[index].data.fd == m_tcpListenSocket.get()) |
| 356 | { |
| 357 | HandleNewTcpConnection(); |
| 358 | } |
| 359 | // Notification for the UDP socket == There is data to be read from the UDP socket, indicating a new DNS request was received |
| 360 | else if (events[index].data.fd == m_udpSocket.get()) |
| 361 | { |
| 362 | HandleUdpDnsRequest(); |
| 363 | } |
| 364 | // Other notifications == new data was received on one of the active TCP connections |
| 365 | else |
| 366 | { |
| 367 | HandleNewTcpData(static_cast<TcpConnectionContext*>(events[index].data.ptr)); |
| 368 | } |
| 369 | } |
| 370 | } |
| 371 | CATCH_LOG() |
| 372 | } |
| 373 | } |
| 374 | |
| 375 | void DnsServer::HandleUdpDnsRequest() noexcept |
| 376 | try |
| 377 | { |
| 378 | static std::array<gsl::byte, c_maxUdpDnsBufferSize> s_dnsBuffer; |
| 379 | |
| 380 | gsl::span<gsl::byte> dnsRequest; |
| 381 | uint32_t udpRequestId{}; |
| 382 | |
| 383 | // Scoped m_udpLock |
| 384 | { |
| 385 | std::scoped_lock<std::mutex> lock{m_udpLock}; |
| 386 | |
| 387 | // Since we only configure an IPv4 DNS server in Linux, we expect all Linux DNS clients to use IPv4 addresses |
| 388 | sockaddr_in remoteAddr{}; |
| 389 | socklen_t remoteAddrLen = sizeof(remoteAddr); |
| 390 | |
| 391 | // Read the DNS request |
| 392 | int bytesReceived = Syscall( |
| 393 | recvfrom, m_udpSocket.get(), s_dnsBuffer.data(), c_maxUdpDnsBufferSize, 0, reinterpret_cast<sockaddr*>(&remoteAddr), &remoteAddrLen); |
| 394 | |
| 395 | if (bytesReceived == 0) |
| 396 | { |
| 397 | GNS_LOG_ERROR("recvfrom returned 0 bytes"); |
| 398 | return; |
| 399 | } |
| 400 | |
| 401 | // Get next request id. If value reaches UINT_MAX + 1 it will be automatically reset to 0 |
| 402 | const auto requestId = m_currentUdpRequestId++; |
| 403 | |
| 404 | GNS_LOG_INFO( |
| 405 | "New UDP DNS request DNS client IP: {}, DNS client port {}, DNS buffer size: {}, UDP request id: {}", |
| 406 | Address::FromBinary(AF_INET, 0, &remoteAddr.sin_addr).Addr().c_str(), |
| 407 | ntohs(remoteAddr.sin_port), |
| 408 | bytesReceived, |
| 409 | requestId); |
| 410 | |
| 411 | // Move request to a local variable |
| 412 | dnsRequest = std::move(gsl::make_span(s_dnsBuffer).subspan(0, bytesReceived)); |
| 413 | udpRequestId = requestId; |
| 414 | |
| 415 | // Track the request |
| 416 | const auto expiration = std::chrono::steady_clock::now() + c_udpRequestTimeout; |
| 417 | const auto expirationIt = m_udpRequestExpirations.emplace(m_udpRequestExpirations.end(), expiration, requestId); |
| 418 | auto removeExpirationOnError = wil::scope_exit([&] { m_udpRequestExpirations.erase(expirationIt); }); |
| 419 | |
| 420 | const auto [_, inserted] = m_udpRequests.emplace(requestId, UdpRequestContext{remoteAddr, expirationIt}); |
| 421 | THROW_UNEXPECTED_IF(!inserted); |
| 422 | |
| 423 | removeExpirationOnError.release(); |
| 424 | } |
| 425 | |
| 426 | if (!dnsRequest.empty()) |
| 427 | { |
| 428 | auto removeRequestOnError = wil::scope_exit([&] { |
| 429 | std::scoped_lock<std::mutex> lock{m_udpLock}; |
| 430 | const auto it = m_udpRequests.find(udpRequestId); |
| 431 | if (it != m_udpRequests.end()) |
| 432 | { |
| 433 | m_udpRequestExpirations.erase(it->second.m_expiration); |
| 434 | m_udpRequests.erase(it); |
| 435 | } |
| 436 | }); |
| 437 | |
| 438 | // Tunnel request to Windows |
| 439 | LX_GNS_DNS_CLIENT_IDENTIFIER dnsClientIdentifier{}; |
| 440 | dnsClientIdentifier.Protocol = IPPROTO_UDP; |
| 441 | dnsClientIdentifier.DnsClientId = udpRequestId; |
| 442 | |
| 443 | m_tunnelDnsRequest(dnsRequest, dnsClientIdentifier); |
| 444 | |
| 445 | removeRequestOnError.release(); |
| 446 | } |
| 447 | } |
| 448 | CATCH_LOG() |
| 449 | |
| 450 | void DnsServer::Stop() noexcept |
| 451 | try |
| 452 | { |
| 453 | GNS_LOG_INFO("stopping DNS server"); |
| 454 | |
| 455 | // Signal the server loop to stop by closing the write fd of the pipe |
| 456 | m_shutdownServerLoopPipe.write().reset(); |
| 457 | |
| 458 | if (m_serverThread.joinable()) |
| 459 | { |
| 460 | m_serverThread.join(); |
| 461 | } |
| 462 | } |
| 463 | CATCH_LOG() |