Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2012 The Android Open Source Project |
| 3 | * |
| 4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | * you may not use this file except in compliance with the License. |
| 6 | * You may obtain a copy of the License at |
| 7 | * |
| 8 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | * |
| 10 | * Unless required by applicable law or agreed to in writing, software |
| 11 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | * See the License for the specific language governing permissions and |
| 14 | * limitations under the License. |
| 15 | */ |
| 16 | |
| 17 | /* |
| 18 | * A service that exchanges time synchronization information between |
| 19 | * a master that defines a timeline and clients that follow the timeline. |
| 20 | */ |
| 21 | |
| 22 | #define LOG_TAG "common_time" |
| 23 | #include <utils/Log.h> |
| 24 | |
| 25 | #include <arpa/inet.h> |
| 26 | #include <assert.h> |
| 27 | #include <fcntl.h> |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 28 | #include <linux/if_ether.h> |
| 29 | #include <net/if.h> |
| 30 | #include <net/if_arp.h> |
| 31 | #include <netinet/ip.h> |
| 32 | #include <poll.h> |
| 33 | #include <stdio.h> |
| 34 | #include <sys/eventfd.h> |
| 35 | #include <sys/ioctl.h> |
| 36 | #include <sys/stat.h> |
| 37 | #include <sys/types.h> |
| 38 | #include <sys/socket.h> |
| 39 | |
| 40 | #include <common_time/local_clock.h> |
| 41 | #include <binder/IPCThreadState.h> |
| 42 | #include <binder/ProcessState.h> |
| 43 | #include <utils/Timers.h> |
| 44 | |
| 45 | #include "common_clock_service.h" |
| 46 | #include "common_time_config_service.h" |
| 47 | #include "common_time_server.h" |
| 48 | #include "common_time_server_packets.h" |
| 49 | #include "clock_recovery.h" |
| 50 | #include "common_clock.h" |
| 51 | |
John Grossman | b8525e9 | 2012-02-16 14:53:24 -0800 | [diff] [blame] | 52 | #define MAX_INT ((int)0x7FFFFFFF) |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 53 | |
| 54 | namespace android { |
| 55 | |
| 56 | const char* CommonTimeServer::kDefaultMasterElectionAddr = "239.195.128.88"; |
| 57 | const uint16_t CommonTimeServer::kDefaultMasterElectionPort = 8887; |
| 58 | const uint64_t CommonTimeServer::kDefaultSyncGroupID = 0; |
| 59 | const uint8_t CommonTimeServer::kDefaultMasterPriority = 1; |
| 60 | const uint32_t CommonTimeServer::kDefaultMasterAnnounceIntervalMs = 10000; |
| 61 | const uint32_t CommonTimeServer::kDefaultSyncRequestIntervalMs = 1000; |
| 62 | const uint32_t CommonTimeServer::kDefaultPanicThresholdUsec = 50000; |
| 63 | const bool CommonTimeServer::kDefaultAutoDisable = true; |
| 64 | const int CommonTimeServer::kSetupRetryTimeoutMs = 30000; |
| 65 | const int64_t CommonTimeServer::kNoGoodDataPanicThresholdUsec = 600000000ll; |
| 66 | const uint32_t CommonTimeServer::kRTTDiscardPanicThreshMultiplier = 5; |
| 67 | |
| 68 | // timeout value representing an infinite timeout |
| 69 | const int CommonTimeServer::kInfiniteTimeout = -1; |
| 70 | |
| 71 | /*** Initial state constants ***/ |
| 72 | |
| 73 | // number of WhoIsMaster attempts sent before giving up |
| 74 | const int CommonTimeServer::kInitial_NumWhoIsMasterRetries = 6; |
| 75 | |
| 76 | // timeout used when waiting for a response to a WhoIsMaster request |
| 77 | const int CommonTimeServer::kInitial_WhoIsMasterTimeoutMs = 500; |
| 78 | |
| 79 | /*** Client state constants ***/ |
| 80 | |
| 81 | // number of sync requests that can fail before a client assumes its master |
| 82 | // is dead |
John Grossman | e1d6c08 | 2012-04-09 11:26:16 -0700 | [diff] [blame] | 83 | const int CommonTimeServer::kClient_NumSyncRequestRetries = 10; |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 84 | |
| 85 | /*** Master state constants ***/ |
| 86 | |
| 87 | /*** Ronin state constants ***/ |
| 88 | |
| 89 | // number of WhoIsMaster attempts sent before declaring ourselves master |
John Grossman | e1d6c08 | 2012-04-09 11:26:16 -0700 | [diff] [blame] | 90 | const int CommonTimeServer::kRonin_NumWhoIsMasterRetries = 20; |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 91 | |
| 92 | // timeout used when waiting for a response to a WhoIsMaster request |
| 93 | const int CommonTimeServer::kRonin_WhoIsMasterTimeoutMs = 500; |
| 94 | |
| 95 | /*** WaitForElection state constants ***/ |
| 96 | |
| 97 | // how long do we wait for an announcement from a master before |
| 98 | // trying another election? |
John Grossman | e1d6c08 | 2012-04-09 11:26:16 -0700 | [diff] [blame] | 99 | const int CommonTimeServer::kWaitForElection_TimeoutMs = 12500; |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 100 | |
| 101 | CommonTimeServer::CommonTimeServer() |
| 102 | : Thread(false) |
| 103 | , mState(ICommonClock::STATE_INITIAL) |
| 104 | , mClockRecovery(&mLocalClock, &mCommonClock) |
| 105 | , mSocket(-1) |
| 106 | , mLastPacketRxLocalTime(0) |
| 107 | , mTimelineID(ICommonClock::kInvalidTimelineID) |
| 108 | , mClockSynced(false) |
| 109 | , mCommonClockHasClients(false) |
| 110 | , mInitial_WhoIsMasterRequestTimeouts(0) |
| 111 | , mClient_MasterDeviceID(0) |
| 112 | , mClient_MasterDevicePriority(0) |
| 113 | , mRonin_WhoIsMasterRequestTimeouts(0) { |
| 114 | // zero out sync stats |
| 115 | resetSyncStats(); |
| 116 | |
| 117 | // Setup the master election endpoint to use the default. |
| 118 | struct sockaddr_in* meep = |
| 119 | reinterpret_cast<struct sockaddr_in*>(&mMasterElectionEP); |
| 120 | memset(&mMasterElectionEP, 0, sizeof(mMasterElectionEP)); |
| 121 | inet_aton(kDefaultMasterElectionAddr, &meep->sin_addr); |
| 122 | meep->sin_family = AF_INET; |
| 123 | meep->sin_port = htons(kDefaultMasterElectionPort); |
| 124 | |
| 125 | // Zero out the master endpoint. |
| 126 | memset(&mMasterEP, 0, sizeof(mMasterEP)); |
| 127 | mMasterEPValid = false; |
| 128 | mBindIfaceValid = false; |
| 129 | setForceLowPriority(false); |
| 130 | |
| 131 | // Set all remaining configuration parameters to their defaults. |
| 132 | mDeviceID = 0; |
| 133 | mSyncGroupID = kDefaultSyncGroupID; |
| 134 | mMasterPriority = kDefaultMasterPriority; |
| 135 | mMasterAnnounceIntervalMs = kDefaultMasterAnnounceIntervalMs; |
| 136 | mSyncRequestIntervalMs = kDefaultSyncRequestIntervalMs; |
| 137 | mPanicThresholdUsec = kDefaultPanicThresholdUsec; |
| 138 | mAutoDisable = kDefaultAutoDisable; |
| 139 | |
| 140 | // Create the eventfd we will use to signal our thread to wake up when |
| 141 | // needed. |
| 142 | mWakeupThreadFD = eventfd(0, EFD_NONBLOCK); |
| 143 | |
| 144 | // seed the random number generator (used to generated timeline IDs) |
| 145 | srand48(static_cast<unsigned int>(systemTime())); |
| 146 | } |
| 147 | |
| 148 | CommonTimeServer::~CommonTimeServer() { |
| 149 | shutdownThread(); |
| 150 | |
| 151 | // No need to grab the lock here. We are in the destructor; if the the user |
| 152 | // has a thread in any of the APIs while the destructor is being called, |
| 153 | // there is a threading problem a the application level we cannot reasonably |
| 154 | // do anything about. |
| 155 | cleanupSocket_l(); |
| 156 | |
| 157 | if (mWakeupThreadFD >= 0) { |
| 158 | close(mWakeupThreadFD); |
| 159 | mWakeupThreadFD = -1; |
| 160 | } |
| 161 | } |
| 162 | |
| 163 | bool CommonTimeServer::startServices() { |
| 164 | // start the ICommonClock service |
| 165 | mICommonClock = CommonClockService::instantiate(*this); |
| 166 | if (mICommonClock == NULL) |
| 167 | return false; |
| 168 | |
| 169 | // start the ICommonTimeConfig service |
| 170 | mICommonTimeConfig = CommonTimeConfigService::instantiate(*this); |
| 171 | if (mICommonTimeConfig == NULL) |
| 172 | return false; |
| 173 | |
| 174 | return true; |
| 175 | } |
| 176 | |
| 177 | bool CommonTimeServer::threadLoop() { |
| 178 | // Register our service interfaces. |
| 179 | if (!startServices()) |
| 180 | return false; |
| 181 | |
| 182 | // Hold the lock while we are in the main thread loop. It will release the |
| 183 | // lock when it blocks, and hold the lock at all other times. |
| 184 | mLock.lock(); |
| 185 | runStateMachine_l(); |
| 186 | mLock.unlock(); |
| 187 | |
| 188 | IPCThreadState::self()->stopProcess(); |
| 189 | return false; |
| 190 | } |
| 191 | |
| 192 | bool CommonTimeServer::runStateMachine_l() { |
| 193 | if (!mLocalClock.initCheck()) |
| 194 | return false; |
| 195 | |
| 196 | if (!mCommonClock.init(mLocalClock.getLocalFreq())) |
| 197 | return false; |
| 198 | |
| 199 | // Enter the initial state. |
| 200 | becomeInitial("startup"); |
| 201 | |
| 202 | // run the state machine |
| 203 | while (!exitPending()) { |
| 204 | struct pollfd pfds[2]; |
John Grossman | c7f57c6 | 2012-06-26 12:50:28 -0700 | [diff] [blame] | 205 | int rc, timeout; |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 206 | int eventCnt = 0; |
| 207 | int64_t wakeupTime; |
John Grossman | c7f57c6 | 2012-06-26 12:50:28 -0700 | [diff] [blame] | 208 | uint32_t t1, t2; |
| 209 | bool needHandleTimeout = false; |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 210 | |
| 211 | // We are always interested in our wakeup FD. |
| 212 | pfds[eventCnt].fd = mWakeupThreadFD; |
| 213 | pfds[eventCnt].events = POLLIN; |
| 214 | pfds[eventCnt].revents = 0; |
| 215 | eventCnt++; |
| 216 | |
| 217 | // If we have a valid socket, then we are interested in what it has to |
| 218 | // say as well. |
| 219 | if (mSocket >= 0) { |
| 220 | pfds[eventCnt].fd = mSocket; |
| 221 | pfds[eventCnt].events = POLLIN; |
| 222 | pfds[eventCnt].revents = 0; |
| 223 | eventCnt++; |
| 224 | } |
| 225 | |
John Grossman | c7f57c6 | 2012-06-26 12:50:28 -0700 | [diff] [blame] | 226 | t1 = static_cast<uint32_t>(mCurTimeout.msecTillTimeout()); |
| 227 | t2 = static_cast<uint32_t>(mClockRecovery.applyRateLimitedSlew()); |
| 228 | timeout = static_cast<int>(t1 < t2 ? t1 : t2); |
| 229 | |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 230 | // Note, we were holding mLock when this function was called. We |
| 231 | // release it only while we are blocking and hold it at all other times. |
| 232 | mLock.unlock(); |
John Grossman | c7f57c6 | 2012-06-26 12:50:28 -0700 | [diff] [blame] | 233 | rc = poll(pfds, eventCnt, timeout); |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 234 | wakeupTime = mLocalClock.getLocalTime(); |
| 235 | mLock.lock(); |
| 236 | |
| 237 | // Is it time to shutdown? If so, don't hesitate... just do it. |
| 238 | if (exitPending()) |
| 239 | break; |
| 240 | |
| 241 | // Did the poll fail? This should never happen and is fatal if it does. |
| 242 | if (rc < 0) { |
| 243 | ALOGE("%s:%d poll failed", __PRETTY_FUNCTION__, __LINE__); |
| 244 | return false; |
| 245 | } |
| 246 | |
John Grossman | c7f57c6 | 2012-06-26 12:50:28 -0700 | [diff] [blame] | 247 | if (rc == 0) { |
| 248 | needHandleTimeout = !mCurTimeout.msecTillTimeout(); |
| 249 | if (needHandleTimeout) |
| 250 | mCurTimeout.setTimeout(kInfiniteTimeout); |
| 251 | } |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 252 | |
| 253 | // Were we woken up on purpose? If so, clear the eventfd with a read. |
| 254 | if (pfds[0].revents) |
| 255 | clearPendingWakeupEvents_l(); |
| 256 | |
| 257 | // Is out bind address dirty? If so, clean up our socket (if any). |
| 258 | // Alternatively, do we have an active socket but should be auto |
| 259 | // disabled? If so, release the socket and enter the proper sync state. |
| 260 | bool droppedSocket = false; |
| 261 | if (mBindIfaceDirty || ((mSocket >= 0) && shouldAutoDisable())) { |
| 262 | cleanupSocket_l(); |
| 263 | mBindIfaceDirty = false; |
| 264 | droppedSocket = true; |
| 265 | } |
| 266 | |
| 267 | // Do we not have a socket but should have one? If so, try to set one |
| 268 | // up. |
| 269 | if ((mSocket < 0) && mBindIfaceValid && !shouldAutoDisable()) { |
| 270 | if (setupSocket_l()) { |
| 271 | // Success! We are now joining a new network (either coming |
| 272 | // from no network, or coming from a potentially different |
| 273 | // network). Force our priority to be lower so that we defer to |
| 274 | // any other masters which may already be on the network we are |
| 275 | // joining. Later, when we enter either the client or the |
| 276 | // master state, we will clear this flag and go back to our |
| 277 | // normal election priority. |
| 278 | setForceLowPriority(true); |
| 279 | switch (mState) { |
| 280 | // If we were in initial (whether we had a immediately |
| 281 | // before this network or not) we want to simply reset the |
| 282 | // system and start again. Forcing a transition from |
| 283 | // INITIAL to INITIAL should do the job. |
| 284 | case CommonClockService::STATE_INITIAL: |
| 285 | becomeInitial("bound interface"); |
| 286 | break; |
| 287 | |
| 288 | // If we were in the master state, then either we were the |
| 289 | // master in a no-network situation, or we were the master |
| 290 | // of a different network and have moved to a new interface. |
John Grossman | e1d6c08 | 2012-04-09 11:26:16 -0700 | [diff] [blame] | 291 | // In either case, immediately transition to Ronin at low |
| 292 | // priority. If there is no one in the network we just |
| 293 | // joined, we will become master soon enough. If there is, |
| 294 | // we want to be certain to defer master status to the |
| 295 | // existing timeline currently running on the network. |
| 296 | // |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 297 | case CommonClockService::STATE_MASTER: |
John Grossman | e1d6c08 | 2012-04-09 11:26:16 -0700 | [diff] [blame] | 298 | becomeRonin("leaving networkless mode"); |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 299 | break; |
| 300 | |
| 301 | // If we were in any other state (CLIENT, RONIN, or |
| 302 | // WAIT_FOR_ELECTION) then we must be moving from one |
| 303 | // network to another. We have lost our old master; |
| 304 | // transition to RONIN in an attempt to find a new master. |
| 305 | // If there are none out there, we will just assume |
| 306 | // responsibility for the timeline we used to be a client |
| 307 | // of. |
| 308 | default: |
| 309 | becomeRonin("bound interface"); |
| 310 | break; |
| 311 | } |
| 312 | } else { |
| 313 | // That's odd... we failed to set up our socket. This could be |
| 314 | // due to some transient network change which will work itself |
| 315 | // out shortly; schedule a retry attempt in the near future. |
| 316 | mCurTimeout.setTimeout(kSetupRetryTimeoutMs); |
| 317 | } |
| 318 | |
| 319 | // One way or the other, we don't have any data to process at this |
| 320 | // point (since we just tried to bulid a new socket). Loop back |
| 321 | // around and wait for the next thing to do. |
| 322 | continue; |
| 323 | } else if (droppedSocket) { |
| 324 | // We just lost our socket, and for whatever reason (either no |
| 325 | // config, or auto disable engaged) we are not supposed to rebuild |
| 326 | // one at this time. We are not going to rebuild our socket until |
| 327 | // something about our config/auto-disabled status changes, so we |
| 328 | // are basically in network-less mode. If we are already in either |
| 329 | // INITIAL or MASTER, just stay there until something changes. If |
| 330 | // we are in any other state (CLIENT, RONIN or WAIT_FOR_ELECTION), |
| 331 | // then transition to either INITIAL or MASTER depending on whether |
| 332 | // or not our timeline is valid. |
| 333 | ALOGI("Entering networkless mode interface is %s, " |
| 334 | "shouldAutoDisable = %s", |
| 335 | mBindIfaceValid ? "valid" : "invalid", |
| 336 | shouldAutoDisable() ? "true" : "false"); |
| 337 | if ((mState != ICommonClock::STATE_INITIAL) && |
| 338 | (mState != ICommonClock::STATE_MASTER)) { |
| 339 | if (mTimelineID == ICommonClock::kInvalidTimelineID) |
| 340 | becomeInitial("network-less mode"); |
| 341 | else |
| 342 | becomeMaster("network-less mode"); |
| 343 | } |
| 344 | |
| 345 | continue; |
| 346 | } |
| 347 | |
John Grossman | c7f57c6 | 2012-06-26 12:50:28 -0700 | [diff] [blame] | 348 | // Time to handle the timeouts? |
| 349 | if (needHandleTimeout) { |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 350 | if (!handleTimeout()) |
| 351 | ALOGE("handleTimeout failed"); |
| 352 | continue; |
| 353 | } |
| 354 | |
| 355 | // Does our socket have data for us (assuming we still have one, we |
| 356 | // may have RXed a packet at the same time as a config change telling us |
| 357 | // to shut our socket down)? If so, process its data. |
| 358 | if ((mSocket >= 0) && (eventCnt > 1) && (pfds[1].revents)) { |
| 359 | mLastPacketRxLocalTime = wakeupTime; |
| 360 | if (!handlePacket()) |
| 361 | ALOGE("handlePacket failed"); |
| 362 | } |
| 363 | } |
| 364 | |
| 365 | cleanupSocket_l(); |
| 366 | return true; |
| 367 | } |
| 368 | |
| 369 | void CommonTimeServer::clearPendingWakeupEvents_l() { |
| 370 | int64_t tmp; |
| 371 | read(mWakeupThreadFD, &tmp, sizeof(tmp)); |
| 372 | } |
| 373 | |
| 374 | void CommonTimeServer::wakeupThread_l() { |
| 375 | int64_t tmp = 1; |
| 376 | write(mWakeupThreadFD, &tmp, sizeof(tmp)); |
| 377 | } |
| 378 | |
| 379 | void CommonTimeServer::cleanupSocket_l() { |
| 380 | if (mSocket >= 0) { |
| 381 | close(mSocket); |
| 382 | mSocket = -1; |
| 383 | } |
| 384 | } |
| 385 | |
| 386 | void CommonTimeServer::shutdownThread() { |
| 387 | // Flag the work thread for shutdown. |
| 388 | this->requestExit(); |
| 389 | |
| 390 | // Signal the thread in case its sleeping. |
| 391 | mLock.lock(); |
| 392 | wakeupThread_l(); |
| 393 | mLock.unlock(); |
| 394 | |
| 395 | // Wait for the thread to exit. |
| 396 | this->join(); |
| 397 | } |
| 398 | |
| 399 | bool CommonTimeServer::setupSocket_l() { |
| 400 | int rc; |
| 401 | bool ret_val = false; |
| 402 | struct sockaddr_in* ipv4_addr = NULL; |
| 403 | char masterElectionEPStr[64]; |
| 404 | const int one = 1; |
| 405 | |
| 406 | // This should never be needed, but if we happened to have an old socket |
| 407 | // lying around, be sure to not leak it before proceeding. |
| 408 | cleanupSocket_l(); |
| 409 | |
| 410 | // If we don't have a valid endpoint to bind to, then how did we get here in |
| 411 | // the first place? Regardless, we know that we are going to fail to bind, |
| 412 | // so don't even try. |
| 413 | if (!mBindIfaceValid) |
| 414 | return false; |
| 415 | |
| 416 | sockaddrToString(mMasterElectionEP, true, masterElectionEPStr, |
| 417 | sizeof(masterElectionEPStr)); |
| 418 | ALOGI("Building socket :: bind = %s master election = %s", |
| 419 | mBindIface.string(), masterElectionEPStr); |
| 420 | |
| 421 | // TODO: add proper support for IPv6. Right now, we block IPv6 addresses at |
| 422 | // the configuration interface level. |
| 423 | if (AF_INET != mMasterElectionEP.ss_family) { |
| 424 | ALOGW("TODO: add proper IPv6 support"); |
| 425 | goto bailout; |
| 426 | } |
| 427 | |
| 428 | // open a UDP socket for the timeline serivce |
| 429 | mSocket = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP); |
| 430 | if (mSocket < 0) { |
| 431 | ALOGE("Failed to create socket (errno = %d)", errno); |
| 432 | goto bailout; |
| 433 | } |
| 434 | |
| 435 | // Bind to the selected interface using Linux's spiffy SO_BINDTODEVICE. |
| 436 | struct ifreq ifr; |
| 437 | memset(&ifr, 0, sizeof(ifr)); |
| 438 | snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "%s", mBindIface.string()); |
| 439 | ifr.ifr_name[sizeof(ifr.ifr_name) - 1] = 0; |
| 440 | rc = setsockopt(mSocket, SOL_SOCKET, SO_BINDTODEVICE, |
| 441 | (void *)&ifr, sizeof(ifr)); |
| 442 | if (rc) { |
| 443 | ALOGE("Failed to bind socket at to interface %s (errno = %d)", |
| 444 | ifr.ifr_name, errno); |
| 445 | goto bailout; |
| 446 | } |
| 447 | |
| 448 | // Bind our socket to INADDR_ANY and the master election port. The |
| 449 | // interface binding we made using SO_BINDTODEVICE should limit us to |
| 450 | // traffic only on the interface we are interested in. We need to bind to |
| 451 | // INADDR_ANY and the specific master election port in order to be able to |
| 452 | // receive both unicast traffic and master election multicast traffic with |
| 453 | // just a single socket. |
| 454 | struct sockaddr_in bindAddr; |
| 455 | ipv4_addr = reinterpret_cast<struct sockaddr_in*>(&mMasterElectionEP); |
| 456 | memcpy(&bindAddr, ipv4_addr, sizeof(bindAddr)); |
| 457 | bindAddr.sin_addr.s_addr = INADDR_ANY; |
| 458 | rc = bind(mSocket, |
| 459 | reinterpret_cast<const sockaddr *>(&bindAddr), |
| 460 | sizeof(bindAddr)); |
| 461 | if (rc) { |
| 462 | ALOGE("Failed to bind socket to port %hu (errno = %d)", |
| 463 | ntohs(bindAddr.sin_port), errno); |
| 464 | goto bailout; |
| 465 | } |
| 466 | |
| 467 | if (0xE0000000 == (ntohl(ipv4_addr->sin_addr.s_addr) & 0xF0000000)) { |
| 468 | // If our master election endpoint is a multicast address, be sure to join |
| 469 | // the multicast group. |
| 470 | struct ip_mreq mreq; |
| 471 | mreq.imr_multiaddr = ipv4_addr->sin_addr; |
| 472 | mreq.imr_interface.s_addr = htonl(INADDR_ANY); |
| 473 | rc = setsockopt(mSocket, IPPROTO_IP, IP_ADD_MEMBERSHIP, |
| 474 | &mreq, sizeof(mreq)); |
| 475 | if (rc == -1) { |
| 476 | ALOGE("Failed to join multicast group at %s. (errno = %d)", |
| 477 | masterElectionEPStr, errno); |
| 478 | goto bailout; |
| 479 | } |
| 480 | |
| 481 | // disable loopback of multicast packets |
| 482 | const int zero = 0; |
| 483 | rc = setsockopt(mSocket, IPPROTO_IP, IP_MULTICAST_LOOP, |
| 484 | &zero, sizeof(zero)); |
| 485 | if (rc == -1) { |
| 486 | ALOGE("Failed to disable multicast loopback (errno = %d)", errno); |
| 487 | goto bailout; |
| 488 | } |
| 489 | } else |
Jason Simmons | db63260 | 2012-07-17 15:48:53 -0700 | [diff] [blame^] | 490 | if (ntohl(ipv4_addr->sin_addr.s_addr) == 0xFFFFFFFF) { |
| 491 | // If the master election address is the broadcast address, then enable |
| 492 | // the broadcast socket option |
| 493 | const int one = 1; |
| 494 | rc = setsockopt(mSocket, SOL_SOCKET, SO_BROADCAST, &one, sizeof(one)); |
| 495 | if (rc == -1) { |
| 496 | ALOGE("Failed to enable broadcast (errno = %d)", errno); |
| 497 | goto bailout; |
| 498 | } |
| 499 | } else { |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 500 | // If the master election address is neither broadcast, nor multicast, |
| 501 | // then we are misconfigured. The config API layer should prevent this |
| 502 | // from ever happening. |
| 503 | goto bailout; |
| 504 | } |
| 505 | |
| 506 | // Set the TTL of sent packets to 1. (Time protocol sync should never leave |
| 507 | // the local subnet) |
| 508 | rc = setsockopt(mSocket, IPPROTO_IP, IP_TTL, &one, sizeof(one)); |
| 509 | if (rc == -1) { |
| 510 | ALOGE("Failed to set TTL to %d (errno = %d)", one, errno); |
| 511 | goto bailout; |
| 512 | } |
| 513 | |
| 514 | // get the device's unique ID |
| 515 | if (!assignDeviceID()) |
| 516 | goto bailout; |
| 517 | |
| 518 | ret_val = true; |
| 519 | |
| 520 | bailout: |
| 521 | if (!ret_val) |
| 522 | cleanupSocket_l(); |
| 523 | return ret_val; |
| 524 | } |
| 525 | |
| 526 | // generate a unique device ID that can be used for arbitration |
| 527 | bool CommonTimeServer::assignDeviceID() { |
| 528 | if (!mBindIfaceValid) |
| 529 | return false; |
| 530 | |
| 531 | struct ifreq ifr; |
| 532 | memset(&ifr, 0, sizeof(ifr)); |
| 533 | ifr.ifr_addr.sa_family = AF_INET; |
| 534 | strlcpy(ifr.ifr_name, mBindIface.string(), IFNAMSIZ); |
| 535 | |
| 536 | int rc = ioctl(mSocket, SIOCGIFHWADDR, &ifr); |
| 537 | if (rc) { |
| 538 | ALOGE("%s:%d ioctl failed", __PRETTY_FUNCTION__, __LINE__); |
| 539 | return false; |
| 540 | } |
| 541 | |
| 542 | if (ifr.ifr_addr.sa_family != ARPHRD_ETHER) { |
| 543 | ALOGE("%s:%d got non-Ethernet address", __PRETTY_FUNCTION__, __LINE__); |
| 544 | return false; |
| 545 | } |
| 546 | |
| 547 | mDeviceID = 0; |
| 548 | for (int i = 0; i < ETH_ALEN; i++) { |
| 549 | mDeviceID = (mDeviceID << 8) | ifr.ifr_hwaddr.sa_data[i]; |
| 550 | } |
| 551 | |
| 552 | return true; |
| 553 | } |
| 554 | |
| 555 | // generate a new timeline ID |
| 556 | void CommonTimeServer::assignTimelineID() { |
| 557 | do { |
| 558 | mTimelineID = (static_cast<uint64_t>(lrand48()) << 32) |
| 559 | | static_cast<uint64_t>(lrand48()); |
| 560 | } while (mTimelineID == ICommonClock::kInvalidTimelineID); |
| 561 | } |
| 562 | |
| 563 | // Select a preference between the device IDs of two potential masters. |
| 564 | // Returns true if the first ID wins, or false if the second ID wins. |
| 565 | bool CommonTimeServer::arbitrateMaster( |
| 566 | uint64_t deviceID1, uint8_t devicePrio1, |
| 567 | uint64_t deviceID2, uint8_t devicePrio2) { |
| 568 | return ((devicePrio1 > devicePrio2) || |
| 569 | ((devicePrio1 == devicePrio2) && (deviceID1 > deviceID2))); |
| 570 | } |
| 571 | |
| 572 | bool CommonTimeServer::handlePacket() { |
| 573 | uint8_t buf[256]; |
| 574 | struct sockaddr_storage srcAddr; |
| 575 | socklen_t srcAddrLen = sizeof(srcAddr); |
| 576 | |
| 577 | ssize_t recvBytes = recvfrom( |
| 578 | mSocket, buf, sizeof(buf), 0, |
| 579 | reinterpret_cast<const sockaddr *>(&srcAddr), &srcAddrLen); |
| 580 | |
| 581 | if (recvBytes < 0) { |
| 582 | ALOGE("%s:%d recvfrom failed", __PRETTY_FUNCTION__, __LINE__); |
| 583 | return false; |
| 584 | } |
| 585 | |
| 586 | UniversalTimeServicePacket pkt; |
| 587 | recvBytes = pkt.deserializePacket(buf, recvBytes, mSyncGroupID); |
| 588 | if (recvBytes < 0) |
| 589 | return false; |
| 590 | |
| 591 | bool result; |
| 592 | switch (pkt.packetType) { |
| 593 | case TIME_PACKET_WHO_IS_MASTER_REQUEST: |
| 594 | result = handleWhoIsMasterRequest(&pkt.p.who_is_master_request, |
| 595 | srcAddr); |
| 596 | break; |
| 597 | |
| 598 | case TIME_PACKET_WHO_IS_MASTER_RESPONSE: |
| 599 | result = handleWhoIsMasterResponse(&pkt.p.who_is_master_response, |
| 600 | srcAddr); |
| 601 | break; |
| 602 | |
| 603 | case TIME_PACKET_SYNC_REQUEST: |
| 604 | result = handleSyncRequest(&pkt.p.sync_request, srcAddr); |
| 605 | break; |
| 606 | |
| 607 | case TIME_PACKET_SYNC_RESPONSE: |
| 608 | result = handleSyncResponse(&pkt.p.sync_response, srcAddr); |
| 609 | break; |
| 610 | |
| 611 | case TIME_PACKET_MASTER_ANNOUNCEMENT: |
| 612 | result = handleMasterAnnouncement(&pkt.p.master_announcement, |
| 613 | srcAddr); |
| 614 | break; |
| 615 | |
| 616 | default: { |
| 617 | ALOGD("%s:%d unknown packet type(%d)", |
| 618 | __PRETTY_FUNCTION__, __LINE__, pkt.packetType); |
| 619 | result = false; |
| 620 | } break; |
| 621 | } |
| 622 | |
| 623 | return result; |
| 624 | } |
| 625 | |
| 626 | bool CommonTimeServer::handleTimeout() { |
| 627 | // If we have no socket, then this must be a timeout to retry socket setup. |
| 628 | if (mSocket < 0) |
| 629 | return true; |
| 630 | |
| 631 | switch (mState) { |
| 632 | case ICommonClock::STATE_INITIAL: |
| 633 | return handleTimeoutInitial(); |
| 634 | case ICommonClock::STATE_CLIENT: |
| 635 | return handleTimeoutClient(); |
| 636 | case ICommonClock::STATE_MASTER: |
| 637 | return handleTimeoutMaster(); |
| 638 | case ICommonClock::STATE_RONIN: |
| 639 | return handleTimeoutRonin(); |
| 640 | case ICommonClock::STATE_WAIT_FOR_ELECTION: |
| 641 | return handleTimeoutWaitForElection(); |
| 642 | } |
| 643 | |
| 644 | return false; |
| 645 | } |
| 646 | |
| 647 | bool CommonTimeServer::handleTimeoutInitial() { |
| 648 | if (++mInitial_WhoIsMasterRequestTimeouts == |
| 649 | kInitial_NumWhoIsMasterRetries) { |
| 650 | // none of our attempts to discover a master succeeded, so make |
| 651 | // this device the master |
| 652 | return becomeMaster("initial timeout"); |
| 653 | } else { |
| 654 | // retry the WhoIsMaster request |
| 655 | return sendWhoIsMasterRequest(); |
| 656 | } |
| 657 | } |
| 658 | |
| 659 | bool CommonTimeServer::handleTimeoutClient() { |
| 660 | if (shouldPanicNotGettingGoodData()) |
| 661 | return becomeInitial("timeout panic, no good data"); |
| 662 | |
| 663 | if (mClient_SyncRequestPending) { |
| 664 | mClient_SyncRequestPending = false; |
| 665 | |
| 666 | if (++mClient_SyncRequestTimeouts < kClient_NumSyncRequestRetries) { |
| 667 | // a sync request has timed out, so retry |
| 668 | return sendSyncRequest(); |
| 669 | } else { |
| 670 | // The master has failed to respond to a sync request for too many |
| 671 | // times in a row. Assume the master is dead and start electing |
| 672 | // a new master. |
| 673 | return becomeRonin("master not responding"); |
| 674 | } |
| 675 | } else { |
| 676 | // initiate the next sync request |
| 677 | return sendSyncRequest(); |
| 678 | } |
| 679 | } |
| 680 | |
| 681 | bool CommonTimeServer::handleTimeoutMaster() { |
| 682 | // send another announcement from the master |
| 683 | return sendMasterAnnouncement(); |
| 684 | } |
| 685 | |
| 686 | bool CommonTimeServer::handleTimeoutRonin() { |
| 687 | if (++mRonin_WhoIsMasterRequestTimeouts == kRonin_NumWhoIsMasterRetries) { |
| 688 | // no other master is out there, so we won the election |
| 689 | return becomeMaster("no better masters detected"); |
| 690 | } else { |
| 691 | return sendWhoIsMasterRequest(); |
| 692 | } |
| 693 | } |
| 694 | |
| 695 | bool CommonTimeServer::handleTimeoutWaitForElection() { |
| 696 | return becomeRonin("timeout waiting for election conclusion"); |
| 697 | } |
| 698 | |
| 699 | bool CommonTimeServer::handleWhoIsMasterRequest( |
| 700 | const WhoIsMasterRequestPacket* request, |
| 701 | const sockaddr_storage& srcAddr) { |
| 702 | if (mState == ICommonClock::STATE_MASTER) { |
| 703 | // is this request related to this master's timeline? |
| 704 | if (request->timelineID != ICommonClock::kInvalidTimelineID && |
| 705 | request->timelineID != mTimelineID) |
| 706 | return true; |
| 707 | |
| 708 | WhoIsMasterResponsePacket pkt; |
| 709 | pkt.initHeader(mTimelineID, mSyncGroupID); |
| 710 | pkt.deviceID = mDeviceID; |
| 711 | pkt.devicePriority = effectivePriority(); |
| 712 | |
| 713 | uint8_t buf[256]; |
| 714 | ssize_t bufSz = pkt.serializePacket(buf, sizeof(buf)); |
| 715 | if (bufSz < 0) |
| 716 | return false; |
| 717 | |
| 718 | ssize_t sendBytes = sendto( |
| 719 | mSocket, buf, bufSz, 0, |
| 720 | reinterpret_cast<const sockaddr *>(&srcAddr), |
| 721 | sizeof(srcAddr)); |
| 722 | if (sendBytes == -1) { |
| 723 | ALOGE("%s:%d sendto failed", __PRETTY_FUNCTION__, __LINE__); |
| 724 | return false; |
| 725 | } |
| 726 | } else if (mState == ICommonClock::STATE_RONIN) { |
| 727 | // if we hear a WhoIsMaster request from another device following |
| 728 | // the same timeline and that device wins arbitration, then we will stop |
| 729 | // trying to elect ourselves master and will instead wait for an |
| 730 | // announcement from the election winner |
| 731 | if (request->timelineID != mTimelineID) |
| 732 | return true; |
| 733 | |
| 734 | if (arbitrateMaster(request->senderDeviceID, |
| 735 | request->senderDevicePriority, |
| 736 | mDeviceID, |
| 737 | effectivePriority())) |
| 738 | return becomeWaitForElection("would lose election"); |
| 739 | |
| 740 | return true; |
| 741 | } else if (mState == ICommonClock::STATE_INITIAL) { |
| 742 | // If a group of devices booted simultaneously (e.g. after a power |
| 743 | // outage) and all of them are in the initial state and there is no |
| 744 | // master, then each device may time out and declare itself master at |
| 745 | // the same time. To avoid this, listen for |
| 746 | // WhoIsMaster(InvalidTimeline) requests from peers. If we would lose |
| 747 | // arbitration against that peer, reset our timeout count so that the |
| 748 | // peer has a chance to become master before we time out. |
| 749 | if (request->timelineID == ICommonClock::kInvalidTimelineID && |
| 750 | arbitrateMaster(request->senderDeviceID, |
| 751 | request->senderDevicePriority, |
| 752 | mDeviceID, |
| 753 | effectivePriority())) { |
| 754 | mInitial_WhoIsMasterRequestTimeouts = 0; |
| 755 | } |
| 756 | } |
| 757 | |
| 758 | return true; |
| 759 | } |
| 760 | |
| 761 | bool CommonTimeServer::handleWhoIsMasterResponse( |
| 762 | const WhoIsMasterResponsePacket* response, |
| 763 | const sockaddr_storage& srcAddr) { |
| 764 | if (mState == ICommonClock::STATE_INITIAL || mState == ICommonClock::STATE_RONIN) { |
| 765 | return becomeClient(srcAddr, |
| 766 | response->deviceID, |
| 767 | response->devicePriority, |
| 768 | response->timelineID, |
| 769 | "heard whois response"); |
| 770 | } else if (mState == ICommonClock::STATE_CLIENT) { |
| 771 | // if we get multiple responses because there are multiple devices |
| 772 | // who believe that they are master, then follow the master that |
| 773 | // wins arbitration |
| 774 | if (arbitrateMaster(response->deviceID, |
| 775 | response->devicePriority, |
| 776 | mClient_MasterDeviceID, |
| 777 | mClient_MasterDevicePriority)) { |
| 778 | return becomeClient(srcAddr, |
| 779 | response->deviceID, |
| 780 | response->devicePriority, |
| 781 | response->timelineID, |
| 782 | "heard whois response"); |
| 783 | } |
| 784 | } |
| 785 | |
| 786 | return true; |
| 787 | } |
| 788 | |
| 789 | bool CommonTimeServer::handleSyncRequest(const SyncRequestPacket* request, |
| 790 | const sockaddr_storage& srcAddr) { |
| 791 | SyncResponsePacket pkt; |
| 792 | pkt.initHeader(mTimelineID, mSyncGroupID); |
| 793 | |
| 794 | if ((mState == ICommonClock::STATE_MASTER) && |
| 795 | (mTimelineID == request->timelineID)) { |
| 796 | int64_t rxLocalTime = mLastPacketRxLocalTime; |
| 797 | int64_t rxCommonTime; |
| 798 | |
| 799 | // If we are master on an actual network and have actual clients, then |
| 800 | // we are no longer low priority. |
| 801 | setForceLowPriority(false); |
| 802 | |
| 803 | if (OK != mCommonClock.localToCommon(rxLocalTime, &rxCommonTime)) { |
| 804 | return false; |
| 805 | } |
| 806 | |
| 807 | int64_t txLocalTime = mLocalClock.getLocalTime();; |
| 808 | int64_t txCommonTime; |
| 809 | if (OK != mCommonClock.localToCommon(txLocalTime, &txCommonTime)) { |
| 810 | return false; |
| 811 | } |
| 812 | |
| 813 | pkt.nak = 0; |
| 814 | pkt.clientTxLocalTime = request->clientTxLocalTime; |
| 815 | pkt.masterRxCommonTime = rxCommonTime; |
| 816 | pkt.masterTxCommonTime = txCommonTime; |
| 817 | } else { |
| 818 | pkt.nak = 1; |
| 819 | pkt.clientTxLocalTime = 0; |
| 820 | pkt.masterRxCommonTime = 0; |
| 821 | pkt.masterTxCommonTime = 0; |
| 822 | } |
| 823 | |
| 824 | uint8_t buf[256]; |
| 825 | ssize_t bufSz = pkt.serializePacket(buf, sizeof(buf)); |
| 826 | if (bufSz < 0) |
| 827 | return false; |
| 828 | |
| 829 | ssize_t sendBytes = sendto( |
| 830 | mSocket, &buf, bufSz, 0, |
| 831 | reinterpret_cast<const sockaddr *>(&srcAddr), |
| 832 | sizeof(srcAddr)); |
| 833 | if (sendBytes == -1) { |
| 834 | ALOGE("%s:%d sendto failed", __PRETTY_FUNCTION__, __LINE__); |
| 835 | return false; |
| 836 | } |
| 837 | |
| 838 | return true; |
| 839 | } |
| 840 | |
| 841 | bool CommonTimeServer::handleSyncResponse( |
| 842 | const SyncResponsePacket* response, |
| 843 | const sockaddr_storage& srcAddr) { |
| 844 | if (mState != ICommonClock::STATE_CLIENT) |
| 845 | return true; |
| 846 | |
| 847 | assert(mMasterEPValid); |
| 848 | if (!sockaddrMatch(srcAddr, mMasterEP, true)) { |
| 849 | char srcEP[64], expectedEP[64]; |
| 850 | sockaddrToString(srcAddr, true, srcEP, sizeof(srcEP)); |
| 851 | sockaddrToString(mMasterEP, true, expectedEP, sizeof(expectedEP)); |
| 852 | ALOGI("Dropping sync response from unexpected address." |
| 853 | " Expected %s Got %s", expectedEP, srcEP); |
| 854 | return true; |
| 855 | } |
| 856 | |
| 857 | if (response->nak) { |
| 858 | // if our master is no longer accepting requests, then we need to find |
| 859 | // a new master |
| 860 | return becomeRonin("master NAK'ed"); |
| 861 | } |
| 862 | |
| 863 | mClient_SyncRequestPending = 0; |
| 864 | mClient_SyncRequestTimeouts = 0; |
| 865 | mClient_PacketRTTLog.logRX(response->clientTxLocalTime, |
| 866 | mLastPacketRxLocalTime); |
| 867 | |
| 868 | bool result; |
| 869 | if (!(mClient_SyncRespsRXedFromCurMaster++)) { |
| 870 | // the first request/response exchange between a client and a master |
| 871 | // may take unusually long due to ARP, so discard it. |
| 872 | result = true; |
| 873 | } else { |
| 874 | int64_t clientTxLocalTime = response->clientTxLocalTime; |
| 875 | int64_t clientRxLocalTime = mLastPacketRxLocalTime; |
| 876 | int64_t masterTxCommonTime = response->masterTxCommonTime; |
| 877 | int64_t masterRxCommonTime = response->masterRxCommonTime; |
| 878 | |
| 879 | int64_t rtt = (clientRxLocalTime - clientTxLocalTime); |
| 880 | int64_t avgLocal = (clientTxLocalTime + clientRxLocalTime) >> 1; |
| 881 | int64_t avgCommon = (masterTxCommonTime + masterRxCommonTime) >> 1; |
| 882 | |
| 883 | // if the RTT of the packet is significantly larger than the panic |
| 884 | // threshold, we should simply discard it. Its better to do nothing |
| 885 | // than to take cues from a packet like that. |
| 886 | int rttCommon = mCommonClock.localDurationToCommonDuration(rtt); |
| 887 | if (rttCommon > (static_cast<int64_t>(mPanicThresholdUsec) * |
| 888 | kRTTDiscardPanicThreshMultiplier)) { |
| 889 | ALOGV("Dropping sync response with RTT of %lld uSec", rttCommon); |
| 890 | mClient_ExpiredSyncRespsRXedFromCurMaster++; |
| 891 | if (shouldPanicNotGettingGoodData()) |
| 892 | return becomeInitial("RX panic, no good data"); |
| 893 | } else { |
Kent Ryhorchuk | 11bc45f | 2012-02-13 16:24:29 -0800 | [diff] [blame] | 894 | result = mClockRecovery.pushDisciplineEvent(avgLocal, avgCommon, rttCommon); |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 895 | mClient_LastGoodSyncRX = clientRxLocalTime; |
| 896 | |
| 897 | if (result) { |
| 898 | // indicate to listeners that we've synced to the common timeline |
| 899 | notifyClockSync(); |
| 900 | } else { |
| 901 | ALOGE("Panic! Observed clock sync error is too high to tolerate," |
| 902 | " resetting state machine and starting over."); |
| 903 | notifyClockSyncLoss(); |
| 904 | return becomeInitial("panic"); |
| 905 | } |
| 906 | } |
| 907 | } |
| 908 | |
| 909 | mCurTimeout.setTimeout(mSyncRequestIntervalMs); |
| 910 | return result; |
| 911 | } |
| 912 | |
| 913 | bool CommonTimeServer::handleMasterAnnouncement( |
| 914 | const MasterAnnouncementPacket* packet, |
| 915 | const sockaddr_storage& srcAddr) { |
| 916 | uint64_t newDeviceID = packet->deviceID; |
| 917 | uint8_t newDevicePrio = packet->devicePriority; |
| 918 | uint64_t newTimelineID = packet->timelineID; |
| 919 | |
| 920 | if (mState == ICommonClock::STATE_INITIAL || |
| 921 | mState == ICommonClock::STATE_RONIN || |
| 922 | mState == ICommonClock::STATE_WAIT_FOR_ELECTION) { |
| 923 | // if we aren't currently following a master, then start following |
| 924 | // this new master |
| 925 | return becomeClient(srcAddr, |
| 926 | newDeviceID, |
| 927 | newDevicePrio, |
| 928 | newTimelineID, |
| 929 | "heard master announcement"); |
| 930 | } else if (mState == ICommonClock::STATE_CLIENT) { |
| 931 | // if the new master wins arbitration against our current master, |
| 932 | // then become a client of the new master |
| 933 | if (arbitrateMaster(newDeviceID, |
| 934 | newDevicePrio, |
| 935 | mClient_MasterDeviceID, |
| 936 | mClient_MasterDevicePriority)) |
| 937 | return becomeClient(srcAddr, |
| 938 | newDeviceID, |
| 939 | newDevicePrio, |
| 940 | newTimelineID, |
| 941 | "heard master announcement"); |
| 942 | } else if (mState == ICommonClock::STATE_MASTER) { |
| 943 | // two masters are competing - if the new one wins arbitration, then |
| 944 | // cease acting as master |
| 945 | if (arbitrateMaster(newDeviceID, newDevicePrio, |
| 946 | mDeviceID, effectivePriority())) |
| 947 | return becomeClient(srcAddr, newDeviceID, |
| 948 | newDevicePrio, newTimelineID, |
| 949 | "heard master announcement"); |
| 950 | } |
| 951 | |
| 952 | return true; |
| 953 | } |
| 954 | |
| 955 | bool CommonTimeServer::sendWhoIsMasterRequest() { |
| 956 | assert(mState == ICommonClock::STATE_INITIAL || mState == ICommonClock::STATE_RONIN); |
| 957 | |
| 958 | // If we have no socket, then we must be in the unconfigured initial state. |
| 959 | // Don't report any errors, just don't try to send the initial who-is-master |
| 960 | // query. Eventually, our network will either become configured, or we will |
| 961 | // be forced into network-less master mode by higher level code. |
| 962 | if (mSocket < 0) { |
| 963 | assert(mState == ICommonClock::STATE_INITIAL); |
| 964 | return true; |
| 965 | } |
| 966 | |
| 967 | bool ret = false; |
| 968 | WhoIsMasterRequestPacket pkt; |
| 969 | pkt.initHeader(mSyncGroupID); |
| 970 | pkt.senderDeviceID = mDeviceID; |
| 971 | pkt.senderDevicePriority = effectivePriority(); |
| 972 | |
| 973 | uint8_t buf[256]; |
| 974 | ssize_t bufSz = pkt.serializePacket(buf, sizeof(buf)); |
| 975 | if (bufSz >= 0) { |
| 976 | ssize_t sendBytes = sendto( |
| 977 | mSocket, buf, bufSz, 0, |
| 978 | reinterpret_cast<const sockaddr *>(&mMasterElectionEP), |
| 979 | sizeof(mMasterElectionEP)); |
| 980 | if (sendBytes < 0) |
| 981 | ALOGE("WhoIsMaster sendto failed (errno %d)", errno); |
| 982 | ret = true; |
| 983 | } |
| 984 | |
| 985 | if (mState == ICommonClock::STATE_INITIAL) { |
| 986 | mCurTimeout.setTimeout(kInitial_WhoIsMasterTimeoutMs); |
| 987 | } else { |
| 988 | mCurTimeout.setTimeout(kRonin_WhoIsMasterTimeoutMs); |
| 989 | } |
| 990 | |
| 991 | return ret; |
| 992 | } |
| 993 | |
| 994 | bool CommonTimeServer::sendSyncRequest() { |
| 995 | // If we are sending sync requests, then we must be in the client state and |
| 996 | // we must have a socket (when we have no network, we are only supposed to |
| 997 | // be in INITIAL or MASTER) |
| 998 | assert(mState == ICommonClock::STATE_CLIENT); |
| 999 | assert(mSocket >= 0); |
| 1000 | |
| 1001 | bool ret = false; |
| 1002 | SyncRequestPacket pkt; |
| 1003 | pkt.initHeader(mTimelineID, mSyncGroupID); |
| 1004 | pkt.clientTxLocalTime = mLocalClock.getLocalTime(); |
| 1005 | |
| 1006 | if (!mClient_FirstSyncTX) |
| 1007 | mClient_FirstSyncTX = pkt.clientTxLocalTime; |
| 1008 | |
| 1009 | mClient_PacketRTTLog.logTX(pkt.clientTxLocalTime); |
| 1010 | |
| 1011 | uint8_t buf[256]; |
| 1012 | ssize_t bufSz = pkt.serializePacket(buf, sizeof(buf)); |
| 1013 | if (bufSz >= 0) { |
| 1014 | ssize_t sendBytes = sendto( |
| 1015 | mSocket, buf, bufSz, 0, |
| 1016 | reinterpret_cast<const sockaddr *>(&mMasterEP), |
| 1017 | sizeof(mMasterEP)); |
| 1018 | if (sendBytes < 0) |
| 1019 | ALOGE("SyncRequest sendto failed (errno %d)", errno); |
| 1020 | ret = true; |
| 1021 | } |
| 1022 | |
| 1023 | mClient_SyncsSentToCurMaster++; |
| 1024 | mCurTimeout.setTimeout(mSyncRequestIntervalMs); |
| 1025 | mClient_SyncRequestPending = true; |
| 1026 | |
| 1027 | return ret; |
| 1028 | } |
| 1029 | |
| 1030 | bool CommonTimeServer::sendMasterAnnouncement() { |
| 1031 | bool ret = false; |
| 1032 | assert(mState == ICommonClock::STATE_MASTER); |
| 1033 | |
| 1034 | // If we are being asked to send a master announcement, but we have no |
| 1035 | // socket, we must be in network-less master mode. Don't bother to send the |
| 1036 | // announcement, and don't bother to schedule a timeout. When the network |
| 1037 | // comes up, the work thread will get poked and start the process of |
| 1038 | // figuring out who the current master should be. |
| 1039 | if (mSocket < 0) { |
| 1040 | mCurTimeout.setTimeout(kInfiniteTimeout); |
| 1041 | return true; |
| 1042 | } |
| 1043 | |
| 1044 | MasterAnnouncementPacket pkt; |
| 1045 | pkt.initHeader(mTimelineID, mSyncGroupID); |
| 1046 | pkt.deviceID = mDeviceID; |
| 1047 | pkt.devicePriority = effectivePriority(); |
| 1048 | |
| 1049 | uint8_t buf[256]; |
| 1050 | ssize_t bufSz = pkt.serializePacket(buf, sizeof(buf)); |
| 1051 | if (bufSz >= 0) { |
| 1052 | ssize_t sendBytes = sendto( |
| 1053 | mSocket, buf, bufSz, 0, |
| 1054 | reinterpret_cast<const sockaddr *>(&mMasterElectionEP), |
| 1055 | sizeof(mMasterElectionEP)); |
| 1056 | if (sendBytes < 0) |
| 1057 | ALOGE("MasterAnnouncement sendto failed (errno %d)", errno); |
| 1058 | ret = true; |
| 1059 | } |
| 1060 | |
| 1061 | mCurTimeout.setTimeout(mMasterAnnounceIntervalMs); |
| 1062 | return ret; |
| 1063 | } |
| 1064 | |
| 1065 | bool CommonTimeServer::becomeClient(const sockaddr_storage& masterEP, |
| 1066 | uint64_t masterDeviceID, |
| 1067 | uint8_t masterDevicePriority, |
| 1068 | uint64_t timelineID, |
| 1069 | const char* cause) { |
| 1070 | char newEPStr[64], oldEPStr[64]; |
| 1071 | sockaddrToString(masterEP, true, newEPStr, sizeof(newEPStr)); |
| 1072 | sockaddrToString(mMasterEP, mMasterEPValid, oldEPStr, sizeof(oldEPStr)); |
| 1073 | |
| 1074 | ALOGI("%s --> CLIENT (%s) :%s" |
| 1075 | " OldMaster: %02x-%014llx::%016llx::%s" |
| 1076 | " NewMaster: %02x-%014llx::%016llx::%s", |
| 1077 | stateToString(mState), cause, |
| 1078 | (mTimelineID != timelineID) ? " (new timeline)" : "", |
| 1079 | mClient_MasterDevicePriority, mClient_MasterDeviceID, |
| 1080 | mTimelineID, oldEPStr, |
| 1081 | masterDevicePriority, masterDeviceID, |
| 1082 | timelineID, newEPStr); |
| 1083 | |
| 1084 | if (mTimelineID != timelineID) { |
| 1085 | // start following a new timeline |
| 1086 | mTimelineID = timelineID; |
| 1087 | mClockRecovery.reset(true, true); |
| 1088 | notifyClockSyncLoss(); |
| 1089 | } else { |
| 1090 | // start following a new master on the existing timeline |
| 1091 | mClockRecovery.reset(false, true); |
| 1092 | } |
| 1093 | |
| 1094 | mMasterEP = masterEP; |
| 1095 | mMasterEPValid = true; |
John Grossman | e1d6c08 | 2012-04-09 11:26:16 -0700 | [diff] [blame] | 1096 | |
| 1097 | // If we are on a real network as a client of a real master, then we should |
| 1098 | // no longer force low priority. If our master disappears, we should have |
| 1099 | // the high priority bit set during the election to replace the master |
| 1100 | // because this group was a real group and not a singleton created in |
| 1101 | // networkless mode. |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 1102 | setForceLowPriority(false); |
| 1103 | |
| 1104 | mClient_MasterDeviceID = masterDeviceID; |
| 1105 | mClient_MasterDevicePriority = masterDevicePriority; |
| 1106 | resetSyncStats(); |
| 1107 | |
| 1108 | setState(ICommonClock::STATE_CLIENT); |
| 1109 | |
| 1110 | // add some jitter to when the various clients send their requests |
| 1111 | // in order to reduce the likelihood that a group of clients overload |
| 1112 | // the master after receiving a master announcement |
| 1113 | usleep((lrand48() % 100) * 1000); |
| 1114 | |
| 1115 | return sendSyncRequest(); |
| 1116 | } |
| 1117 | |
| 1118 | bool CommonTimeServer::becomeMaster(const char* cause) { |
| 1119 | uint64_t oldTimelineID = mTimelineID; |
| 1120 | if (mTimelineID == ICommonClock::kInvalidTimelineID) { |
| 1121 | // this device has not been following any existing timeline, |
| 1122 | // so it will create a new timeline and declare itself master |
| 1123 | assert(!mCommonClock.isValid()); |
| 1124 | |
| 1125 | // set the common time basis |
| 1126 | mCommonClock.setBasis(mLocalClock.getLocalTime(), 0); |
| 1127 | |
| 1128 | // assign an arbitrary timeline iD |
| 1129 | assignTimelineID(); |
| 1130 | |
| 1131 | // notify listeners that we've created a common timeline |
| 1132 | notifyClockSync(); |
| 1133 | } |
| 1134 | |
| 1135 | ALOGI("%s --> MASTER (%s) : %s timeline %016llx", |
| 1136 | stateToString(mState), cause, |
| 1137 | (oldTimelineID == mTimelineID) ? "taking ownership of" |
| 1138 | : "creating new", |
| 1139 | mTimelineID); |
| 1140 | |
| 1141 | memset(&mMasterEP, 0, sizeof(mMasterEP)); |
| 1142 | mMasterEPValid = false; |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 1143 | mClient_MasterDevicePriority = effectivePriority(); |
| 1144 | mClient_MasterDeviceID = mDeviceID; |
| 1145 | mClockRecovery.reset(false, true); |
| 1146 | resetSyncStats(); |
| 1147 | |
| 1148 | setState(ICommonClock::STATE_MASTER); |
| 1149 | return sendMasterAnnouncement(); |
| 1150 | } |
| 1151 | |
| 1152 | bool CommonTimeServer::becomeRonin(const char* cause) { |
| 1153 | // If we were the client of a given timeline, but had never received even a |
| 1154 | // single time sync packet, then we transition back to Initial instead of |
| 1155 | // Ronin. If we transition to Ronin and end up becoming the new Master, we |
| 1156 | // will be unable to service requests for other clients because we never |
| 1157 | // actually knew what time it was. By going to initial, we ensure that |
| 1158 | // other clients who know what time it is, but would lose master arbitration |
| 1159 | // in the Ronin case, will step up and become the proper new master of the |
| 1160 | // old timeline. |
| 1161 | |
| 1162 | char oldEPStr[64]; |
| 1163 | sockaddrToString(mMasterEP, mMasterEPValid, oldEPStr, sizeof(oldEPStr)); |
| 1164 | memset(&mMasterEP, 0, sizeof(mMasterEP)); |
| 1165 | mMasterEPValid = false; |
| 1166 | |
| 1167 | if (mCommonClock.isValid()) { |
| 1168 | ALOGI("%s --> RONIN (%s) : lost track of previously valid timeline " |
| 1169 | "%02x-%014llx::%016llx::%s (%d TXed %d RXed %d RXExpired)", |
| 1170 | stateToString(mState), cause, |
| 1171 | mClient_MasterDevicePriority, mClient_MasterDeviceID, |
| 1172 | mTimelineID, oldEPStr, |
| 1173 | mClient_SyncsSentToCurMaster, |
| 1174 | mClient_SyncRespsRXedFromCurMaster, |
| 1175 | mClient_ExpiredSyncRespsRXedFromCurMaster); |
| 1176 | |
| 1177 | mRonin_WhoIsMasterRequestTimeouts = 0; |
| 1178 | setState(ICommonClock::STATE_RONIN); |
| 1179 | return sendWhoIsMasterRequest(); |
| 1180 | } else { |
| 1181 | ALOGI("%s --> INITIAL (%s) : never synced timeline " |
| 1182 | "%02x-%014llx::%016llx::%s (%d TXed %d RXed %d RXExpired)", |
| 1183 | stateToString(mState), cause, |
| 1184 | mClient_MasterDevicePriority, mClient_MasterDeviceID, |
| 1185 | mTimelineID, oldEPStr, |
| 1186 | mClient_SyncsSentToCurMaster, |
| 1187 | mClient_SyncRespsRXedFromCurMaster, |
| 1188 | mClient_ExpiredSyncRespsRXedFromCurMaster); |
| 1189 | |
| 1190 | return becomeInitial("ronin, no timeline"); |
| 1191 | } |
| 1192 | } |
| 1193 | |
| 1194 | bool CommonTimeServer::becomeWaitForElection(const char* cause) { |
| 1195 | ALOGI("%s --> WAIT_FOR_ELECTION (%s) : dropping out of election," |
| 1196 | " waiting %d mSec for completion.", |
| 1197 | stateToString(mState), cause, kWaitForElection_TimeoutMs); |
| 1198 | |
| 1199 | setState(ICommonClock::STATE_WAIT_FOR_ELECTION); |
| 1200 | mCurTimeout.setTimeout(kWaitForElection_TimeoutMs); |
| 1201 | return true; |
| 1202 | } |
| 1203 | |
| 1204 | bool CommonTimeServer::becomeInitial(const char* cause) { |
| 1205 | ALOGI("Entering INITIAL (%s), total reset.", cause); |
| 1206 | |
| 1207 | setState(ICommonClock::STATE_INITIAL); |
| 1208 | |
| 1209 | // reset clock recovery |
| 1210 | mClockRecovery.reset(true, true); |
| 1211 | |
| 1212 | // reset internal state bookkeeping. |
| 1213 | mCurTimeout.setTimeout(kInfiniteTimeout); |
| 1214 | memset(&mMasterEP, 0, sizeof(mMasterEP)); |
| 1215 | mMasterEPValid = false; |
| 1216 | mLastPacketRxLocalTime = 0; |
| 1217 | mTimelineID = ICommonClock::kInvalidTimelineID; |
| 1218 | mClockSynced = false; |
| 1219 | mInitial_WhoIsMasterRequestTimeouts = 0; |
| 1220 | mClient_MasterDeviceID = 0; |
| 1221 | mClient_MasterDevicePriority = 0; |
| 1222 | mRonin_WhoIsMasterRequestTimeouts = 0; |
| 1223 | resetSyncStats(); |
| 1224 | |
| 1225 | // send the first request to discover the master |
| 1226 | return sendWhoIsMasterRequest(); |
| 1227 | } |
| 1228 | |
| 1229 | void CommonTimeServer::notifyClockSync() { |
| 1230 | if (!mClockSynced) { |
| 1231 | mClockSynced = true; |
| 1232 | mICommonClock->notifyOnTimelineChanged(mTimelineID); |
| 1233 | } |
| 1234 | } |
| 1235 | |
| 1236 | void CommonTimeServer::notifyClockSyncLoss() { |
| 1237 | if (mClockSynced) { |
| 1238 | mClockSynced = false; |
| 1239 | mICommonClock->notifyOnTimelineChanged( |
| 1240 | ICommonClock::kInvalidTimelineID); |
| 1241 | } |
| 1242 | } |
| 1243 | |
| 1244 | void CommonTimeServer::setState(ICommonClock::State s) { |
| 1245 | mState = s; |
| 1246 | } |
| 1247 | |
| 1248 | const char* CommonTimeServer::stateToString(ICommonClock::State s) { |
| 1249 | switch(s) { |
| 1250 | case ICommonClock::STATE_INITIAL: |
| 1251 | return "INITIAL"; |
| 1252 | case ICommonClock::STATE_CLIENT: |
| 1253 | return "CLIENT"; |
| 1254 | case ICommonClock::STATE_MASTER: |
| 1255 | return "MASTER"; |
| 1256 | case ICommonClock::STATE_RONIN: |
| 1257 | return "RONIN"; |
| 1258 | case ICommonClock::STATE_WAIT_FOR_ELECTION: |
| 1259 | return "WAIT_FOR_ELECTION"; |
| 1260 | default: |
| 1261 | return "unknown"; |
| 1262 | } |
| 1263 | } |
| 1264 | |
| 1265 | void CommonTimeServer::sockaddrToString(const sockaddr_storage& addr, |
| 1266 | bool addrValid, |
| 1267 | char* buf, size_t bufLen) { |
| 1268 | if (!bufLen || !buf) |
| 1269 | return; |
| 1270 | |
| 1271 | if (addrValid) { |
| 1272 | switch (addr.ss_family) { |
| 1273 | case AF_INET: { |
| 1274 | const struct sockaddr_in* sa = |
| 1275 | reinterpret_cast<const struct sockaddr_in*>(&addr); |
| 1276 | unsigned long a = ntohl(sa->sin_addr.s_addr); |
| 1277 | uint16_t p = ntohs(sa->sin_port); |
| 1278 | snprintf(buf, bufLen, "%lu.%lu.%lu.%lu:%hu", |
| 1279 | ((a >> 24) & 0xFF), ((a >> 16) & 0xFF), |
| 1280 | ((a >> 8) & 0xFF), (a & 0xFF), p); |
| 1281 | } break; |
| 1282 | |
| 1283 | case AF_INET6: { |
| 1284 | const struct sockaddr_in6* sa = |
| 1285 | reinterpret_cast<const struct sockaddr_in6*>(&addr); |
| 1286 | const uint8_t* a = sa->sin6_addr.s6_addr; |
| 1287 | uint16_t p = ntohs(sa->sin6_port); |
| 1288 | snprintf(buf, bufLen, |
| 1289 | "%02X%02X:%02X%02X:%02X%02X:%02X%02X:" |
| 1290 | "%02X%02X:%02X%02X:%02X%02X:%02X%02X port %hd", |
| 1291 | a[0], a[1], a[ 2], a[ 3], a[ 4], a[ 5], a[ 6], a[ 7], |
| 1292 | a[8], a[9], a[10], a[11], a[12], a[13], a[14], a[15], |
| 1293 | p); |
| 1294 | } break; |
| 1295 | |
| 1296 | default: |
| 1297 | snprintf(buf, bufLen, |
| 1298 | "<unknown sockaddr family %d>", addr.ss_family); |
| 1299 | break; |
| 1300 | } |
| 1301 | } else { |
| 1302 | snprintf(buf, bufLen, "<none>"); |
| 1303 | } |
| 1304 | |
| 1305 | buf[bufLen - 1] = 0; |
| 1306 | } |
| 1307 | |
| 1308 | bool CommonTimeServer::sockaddrMatch(const sockaddr_storage& a1, |
| 1309 | const sockaddr_storage& a2, |
| 1310 | bool matchAddressOnly) { |
| 1311 | if (a1.ss_family != a2.ss_family) |
| 1312 | return false; |
| 1313 | |
| 1314 | switch (a1.ss_family) { |
| 1315 | case AF_INET: { |
| 1316 | const struct sockaddr_in* sa1 = |
| 1317 | reinterpret_cast<const struct sockaddr_in*>(&a1); |
| 1318 | const struct sockaddr_in* sa2 = |
| 1319 | reinterpret_cast<const struct sockaddr_in*>(&a2); |
| 1320 | |
| 1321 | if (sa1->sin_addr.s_addr != sa2->sin_addr.s_addr) |
| 1322 | return false; |
| 1323 | |
| 1324 | return (matchAddressOnly || (sa1->sin_port == sa2->sin_port)); |
| 1325 | } break; |
| 1326 | |
| 1327 | case AF_INET6: { |
| 1328 | const struct sockaddr_in6* sa1 = |
| 1329 | reinterpret_cast<const struct sockaddr_in6*>(&a1); |
| 1330 | const struct sockaddr_in6* sa2 = |
| 1331 | reinterpret_cast<const struct sockaddr_in6*>(&a2); |
| 1332 | |
| 1333 | if (memcmp(&sa1->sin6_addr, &sa2->sin6_addr, sizeof(sa2->sin6_addr))) |
| 1334 | return false; |
| 1335 | |
| 1336 | return (matchAddressOnly || (sa1->sin6_port == sa2->sin6_port)); |
| 1337 | } break; |
| 1338 | |
| 1339 | // Huh? We don't deal in non-IPv[46] addresses. Not sure how we got |
| 1340 | // here, but we don't know how to comapre these addresses and simply |
| 1341 | // default to a no-match decision. |
| 1342 | default: return false; |
| 1343 | } |
| 1344 | } |
| 1345 | |
Mike J. Chen | 6c92951 | 2011-08-15 11:59:47 -0700 | [diff] [blame] | 1346 | bool CommonTimeServer::shouldPanicNotGettingGoodData() { |
| 1347 | if (mClient_FirstSyncTX) { |
| 1348 | int64_t now = mLocalClock.getLocalTime(); |
| 1349 | int64_t delta = now - (mClient_LastGoodSyncRX |
| 1350 | ? mClient_LastGoodSyncRX |
| 1351 | : mClient_FirstSyncTX); |
| 1352 | int64_t deltaUsec = mCommonClock.localDurationToCommonDuration(delta); |
| 1353 | |
| 1354 | if (deltaUsec >= kNoGoodDataPanicThresholdUsec) |
| 1355 | return true; |
| 1356 | } |
| 1357 | |
| 1358 | return false; |
| 1359 | } |
| 1360 | |
| 1361 | void CommonTimeServer::PacketRTTLog::logTX(int64_t txTime) { |
| 1362 | txTimes[wrPtr] = txTime; |
| 1363 | rxTimes[wrPtr] = 0; |
| 1364 | wrPtr = (wrPtr + 1) % RTT_LOG_SIZE; |
| 1365 | if (!wrPtr) |
| 1366 | logFull = true; |
| 1367 | } |
| 1368 | |
| 1369 | void CommonTimeServer::PacketRTTLog::logRX(int64_t txTime, int64_t rxTime) { |
| 1370 | if (!logFull && !wrPtr) |
| 1371 | return; |
| 1372 | |
| 1373 | uint32_t i = logFull ? wrPtr : 0; |
| 1374 | do { |
| 1375 | if (txTimes[i] == txTime) { |
| 1376 | rxTimes[i] = rxTime; |
| 1377 | break; |
| 1378 | } |
| 1379 | i = (i + 1) % RTT_LOG_SIZE; |
| 1380 | } while (i != wrPtr); |
| 1381 | } |
| 1382 | |
| 1383 | } // namespace android |