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>
28 #include <inttypes.h>
29 #include <linux/if_ether.h>
30 #include <net/if.h>
31 #include <net/if_arp.h>
32 #include <netinet/ip.h>
33 #include <poll.h>
34 #include <stdio.h>
35 #include <sys/eventfd.h>
36 #include <sys/ioctl.h>
37 #include <sys/stat.h>
38 #include <sys/types.h>
39 #include <sys/socket.h>
40
41 #include <common_time/local_clock.h>
42 #include <binder/IPCThreadState.h>
43 #include <binder/ProcessState.h>
44 #include <utils/Timers.h>
45
46 #include "common_clock_service.h"
47 #include "common_time_config_service.h"
48 #include "common_time_server.h"
49 #include "common_time_server_packets.h"
50 #include "clock_recovery.h"
51 #include "common_clock.h"
52
53 #define MAX_INT ((int)0x7FFFFFFF)
54
55 namespace android {
56
57 const char* CommonTimeServer::kDefaultMasterElectionAddr = "255.255.255.255";
58 const uint16_t CommonTimeServer::kDefaultMasterElectionPort = 8886;
59 const uint64_t CommonTimeServer::kDefaultSyncGroupID = 1;
60 const uint8_t CommonTimeServer::kDefaultMasterPriority = 1;
61 const uint32_t CommonTimeServer::kDefaultMasterAnnounceIntervalMs = 10000;
62 const uint32_t CommonTimeServer::kDefaultSyncRequestIntervalMs = 1000;
63 const uint32_t CommonTimeServer::kDefaultPanicThresholdUsec = 50000;
64 const bool CommonTimeServer::kDefaultAutoDisable = true;
65 const int CommonTimeServer::kSetupRetryTimeoutMs = 30000;
66 const int64_t CommonTimeServer::kNoGoodDataPanicThresholdUsec = 600000000ll;
67 const uint32_t CommonTimeServer::kRTTDiscardPanicThreshMultiplier = 5;
68
69 // timeout value representing an infinite timeout
70 const int CommonTimeServer::kInfiniteTimeout = -1;
71
72 /*** Initial state constants ***/
73
74 // number of WhoIsMaster attempts sent before giving up
75 const int CommonTimeServer::kInitial_NumWhoIsMasterRetries = 6;
76
77 // timeout used when waiting for a response to a WhoIsMaster request
78 const int CommonTimeServer::kInitial_WhoIsMasterTimeoutMs = 500;
79
80 /*** Client state constants ***/
81
82 // number of sync requests that can fail before a client assumes its master
83 // is dead
84 const int CommonTimeServer::kClient_NumSyncRequestRetries = 10;
85
86 /*** Master state constants ***/
87
88 /*** Ronin state constants ***/
89
90 // number of WhoIsMaster attempts sent before declaring ourselves master
91 const int CommonTimeServer::kRonin_NumWhoIsMasterRetries = 20;
92
93 // timeout used when waiting for a response to a WhoIsMaster request
94 const int CommonTimeServer::kRonin_WhoIsMasterTimeoutMs = 500;
95
96 /*** WaitForElection state constants ***/
97
98 // how long do we wait for an announcement from a master before
99 // trying another election?
100 const int CommonTimeServer::kWaitForElection_TimeoutMs = 12500;
101
CommonTimeServer()102 CommonTimeServer::CommonTimeServer()
103 : Thread(false)
104 , mState(ICommonClock::STATE_INITIAL)
105 , mClockRecovery(&mLocalClock, &mCommonClock)
106 , mSocket(-1)
107 , mLastPacketRxLocalTime(0)
108 , mTimelineID(ICommonClock::kInvalidTimelineID)
109 , mClockSynced(false)
110 , mCommonClockHasClients(false)
111 , mStateChangeLog("Recent State Change Events", 30)
112 , mElectionLog("Recent Master Election Traffic", 30)
113 , mBadPktLog("Recent Bad Packet RX Info", 8)
114 , mInitial_WhoIsMasterRequestTimeouts(0)
115 , mClient_MasterDeviceID(0)
116 , mClient_MasterDevicePriority(0)
117 , mRonin_WhoIsMasterRequestTimeouts(0) {
118 // zero out sync stats
119 resetSyncStats();
120
121 // Setup the master election endpoint to use the default.
122 struct sockaddr_in* meep =
123 reinterpret_cast<struct sockaddr_in*>(&mMasterElectionEP);
124 memset(&mMasterElectionEP, 0, sizeof(mMasterElectionEP));
125 inet_aton(kDefaultMasterElectionAddr, &meep->sin_addr);
126 meep->sin_family = AF_INET;
127 meep->sin_port = htons(kDefaultMasterElectionPort);
128
129 // Zero out the master endpoint.
130 memset(&mMasterEP, 0, sizeof(mMasterEP));
131 mMasterEPValid = false;
132 mBindIfaceValid = false;
133 setForceLowPriority(false);
134
135 // Set all remaining configuration parameters to their defaults.
136 mDeviceID = 0;
137 mSyncGroupID = kDefaultSyncGroupID;
138 mMasterPriority = kDefaultMasterPriority;
139 mMasterAnnounceIntervalMs = kDefaultMasterAnnounceIntervalMs;
140 mSyncRequestIntervalMs = kDefaultSyncRequestIntervalMs;
141 mPanicThresholdUsec = kDefaultPanicThresholdUsec;
142 mAutoDisable = kDefaultAutoDisable;
143
144 // Create the eventfd we will use to signal our thread to wake up when
145 // needed.
146 mWakeupThreadFD = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
147
148 // seed the random number generator (used to generated timeline IDs)
149 srand48(static_cast<unsigned int>(systemTime()));
150 }
151
~CommonTimeServer()152 CommonTimeServer::~CommonTimeServer() {
153 shutdownThread();
154
155 // No need to grab the lock here. We are in the destructor; if the the user
156 // has a thread in any of the APIs while the destructor is being called,
157 // there is a threading problem a the application level we cannot reasonably
158 // do anything about.
159 cleanupSocket_l();
160
161 if (mWakeupThreadFD >= 0) {
162 close(mWakeupThreadFD);
163 mWakeupThreadFD = -1;
164 }
165 }
166
startServices()167 bool CommonTimeServer::startServices() {
168 // start the ICommonClock service
169 mICommonClock = CommonClockService::instantiate(*this);
170 if (mICommonClock == NULL)
171 return false;
172
173 // start the ICommonTimeConfig service
174 mICommonTimeConfig = CommonTimeConfigService::instantiate(*this);
175 if (mICommonTimeConfig == NULL)
176 return false;
177
178 return true;
179 }
180
threadLoop()181 bool CommonTimeServer::threadLoop() {
182 // Register our service interfaces.
183 if (!startServices())
184 return false;
185
186 // Hold the lock while we are in the main thread loop. It will release the
187 // lock when it blocks, and hold the lock at all other times.
188 mLock.lock();
189 runStateMachine_l();
190 mLock.unlock();
191
192 IPCThreadState::self()->stopProcess();
193 return false;
194 }
195
runStateMachine_l()196 bool CommonTimeServer::runStateMachine_l() {
197 if (!mLocalClock.initCheck())
198 return false;
199
200 if (!mCommonClock.init(mLocalClock.getLocalFreq()))
201 return false;
202
203 // Enter the initial state.
204 becomeInitial("startup");
205
206 // run the state machine
207 while (!exitPending()) {
208 struct pollfd pfds[2];
209 int rc, timeout;
210 int eventCnt = 0;
211 int64_t wakeupTime;
212 uint32_t t1, t2;
213 bool needHandleTimeout = false;
214
215 // We are always interested in our wakeup FD.
216 pfds[eventCnt].fd = mWakeupThreadFD;
217 pfds[eventCnt].events = POLLIN;
218 pfds[eventCnt].revents = 0;
219 eventCnt++;
220
221 // If we have a valid socket, then we are interested in what it has to
222 // say as well.
223 if (mSocket >= 0) {
224 pfds[eventCnt].fd = mSocket;
225 pfds[eventCnt].events = POLLIN;
226 pfds[eventCnt].revents = 0;
227 eventCnt++;
228 }
229
230 t1 = static_cast<uint32_t>(mCurTimeout.msecTillTimeout());
231 t2 = static_cast<uint32_t>(mClockRecovery.applyRateLimitedSlew());
232 timeout = static_cast<int>(t1 < t2 ? t1 : t2);
233
234 // Note, we were holding mLock when this function was called. We
235 // release it only while we are blocking and hold it at all other times.
236 mLock.unlock();
237 rc = poll(pfds, eventCnt, timeout);
238 wakeupTime = mLocalClock.getLocalTime();
239 mLock.lock();
240
241 // Is it time to shutdown? If so, don't hesitate... just do it.
242 if (exitPending())
243 break;
244
245 // Did the poll fail? This should never happen and is fatal if it does.
246 if (rc < 0) {
247 ALOGE("%s:%d poll failed", __PRETTY_FUNCTION__, __LINE__);
248 return false;
249 }
250
251 if (rc == 0) {
252 needHandleTimeout = !mCurTimeout.msecTillTimeout();
253 if (needHandleTimeout)
254 mCurTimeout.setTimeout(kInfiniteTimeout);
255 }
256
257 // Were we woken up on purpose? If so, clear the eventfd with a read.
258 if (pfds[0].revents)
259 clearPendingWakeupEvents_l();
260
261 // Is out bind address dirty? If so, clean up our socket (if any).
262 // Alternatively, do we have an active socket but should be auto
263 // disabled? If so, release the socket and enter the proper sync state.
264 bool droppedSocket = false;
265 if (mBindIfaceDirty || ((mSocket >= 0) && shouldAutoDisable())) {
266 cleanupSocket_l();
267 mBindIfaceDirty = false;
268 droppedSocket = true;
269 }
270
271 // Do we not have a socket but should have one? If so, try to set one
272 // up.
273 if ((mSocket < 0) && mBindIfaceValid && !shouldAutoDisable()) {
274 if (setupSocket_l()) {
275 // Success! We are now joining a new network (either coming
276 // from no network, or coming from a potentially different
277 // network). Force our priority to be lower so that we defer to
278 // any other masters which may already be on the network we are
279 // joining. Later, when we enter either the client or the
280 // master state, we will clear this flag and go back to our
281 // normal election priority.
282 setForceLowPriority(true);
283 switch (mState) {
284 // If we were in initial (whether we had a immediately
285 // before this network or not) we want to simply reset the
286 // system and start again. Forcing a transition from
287 // INITIAL to INITIAL should do the job.
288 case CommonClockService::STATE_INITIAL:
289 becomeInitial("bound interface");
290 break;
291
292 // If we were in the master state, then either we were the
293 // master in a no-network situation, or we were the master
294 // of a different network and have moved to a new interface.
295 // In either case, immediately transition to Ronin at low
296 // priority. If there is no one in the network we just
297 // joined, we will become master soon enough. If there is,
298 // we want to be certain to defer master status to the
299 // existing timeline currently running on the network.
300 //
301 case CommonClockService::STATE_MASTER:
302 becomeRonin("leaving networkless mode");
303 break;
304
305 // If we were in any other state (CLIENT, RONIN, or
306 // WAIT_FOR_ELECTION) then we must be moving from one
307 // network to another. We have lost our old master;
308 // transition to RONIN in an attempt to find a new master.
309 // If there are none out there, we will just assume
310 // responsibility for the timeline we used to be a client
311 // of.
312 default:
313 becomeRonin("bound interface");
314 break;
315 }
316 } else {
317 // That's odd... we failed to set up our socket. This could be
318 // due to some transient network change which will work itself
319 // out shortly; schedule a retry attempt in the near future.
320 mCurTimeout.setTimeout(kSetupRetryTimeoutMs);
321 }
322
323 // One way or the other, we don't have any data to process at this
324 // point (since we just tried to bulid a new socket). Loop back
325 // around and wait for the next thing to do.
326 continue;
327 } else if (droppedSocket) {
328 // We just lost our socket, and for whatever reason (either no
329 // config, or auto disable engaged) we are not supposed to rebuild
330 // one at this time. We are not going to rebuild our socket until
331 // something about our config/auto-disabled status changes, so we
332 // are basically in network-less mode. If we are already in either
333 // INITIAL or MASTER, just stay there until something changes. If
334 // we are in any other state (CLIENT, RONIN or WAIT_FOR_ELECTION),
335 // then transition to either INITIAL or MASTER depending on whether
336 // or not our timeline is valid.
337 mStateChangeLog.log(ANDROID_LOG_INFO, LOG_TAG,
338 "Entering networkless mode interface is %s, "
339 "shouldAutoDisable = %s",
340 mBindIfaceValid ? "valid" : "invalid",
341 shouldAutoDisable() ? "true" : "false");
342 if ((mState != ICommonClock::STATE_INITIAL) &&
343 (mState != ICommonClock::STATE_MASTER)) {
344 if (mTimelineID == ICommonClock::kInvalidTimelineID)
345 becomeInitial("network-less mode");
346 else
347 becomeMaster("network-less mode");
348 }
349
350 continue;
351 }
352
353 // Time to handle the timeouts?
354 if (needHandleTimeout) {
355 if (!handleTimeout())
356 ALOGE("handleTimeout failed");
357 continue;
358 }
359
360 // Does our socket have data for us (assuming we still have one, we
361 // may have RXed a packet at the same time as a config change telling us
362 // to shut our socket down)? If so, process its data.
363 if ((mSocket >= 0) && (eventCnt > 1) && (pfds[1].revents)) {
364 mLastPacketRxLocalTime = wakeupTime;
365 if (!handlePacket())
366 ALOGE("handlePacket failed");
367 }
368 }
369
370 cleanupSocket_l();
371 return true;
372 }
373
clearPendingWakeupEvents_l()374 void CommonTimeServer::clearPendingWakeupEvents_l() {
375 int64_t tmp;
376 read(mWakeupThreadFD, &tmp, sizeof(tmp));
377 }
378
wakeupThread_l()379 void CommonTimeServer::wakeupThread_l() {
380 int64_t tmp = 1;
381 write(mWakeupThreadFD, &tmp, sizeof(tmp));
382 }
383
cleanupSocket_l()384 void CommonTimeServer::cleanupSocket_l() {
385 if (mSocket >= 0) {
386 close(mSocket);
387 mSocket = -1;
388 }
389 }
390
shutdownThread()391 void CommonTimeServer::shutdownThread() {
392 // Flag the work thread for shutdown.
393 this->requestExit();
394
395 // Signal the thread in case its sleeping.
396 mLock.lock();
397 wakeupThread_l();
398 mLock.unlock();
399
400 // Wait for the thread to exit.
401 this->join();
402 }
403
setupSocket_l()404 bool CommonTimeServer::setupSocket_l() {
405 int rc;
406 bool ret_val = false;
407 struct sockaddr_in* ipv4_addr = NULL;
408 char masterElectionEPStr[64];
409 const int one = 1;
410
411 // This should never be needed, but if we happened to have an old socket
412 // lying around, be sure to not leak it before proceeding.
413 cleanupSocket_l();
414
415 // If we don't have a valid endpoint to bind to, then how did we get here in
416 // the first place? Regardless, we know that we are going to fail to bind,
417 // so don't even try.
418 if (!mBindIfaceValid)
419 return false;
420
421 sockaddrToString(mMasterElectionEP, true, masterElectionEPStr,
422 sizeof(masterElectionEPStr));
423 mStateChangeLog.log(ANDROID_LOG_INFO, LOG_TAG,
424 "Building socket :: bind = %s master election = %s",
425 mBindIface.string(), masterElectionEPStr);
426
427 // TODO: add proper support for IPv6. Right now, we block IPv6 addresses at
428 // the configuration interface level.
429 if (AF_INET != mMasterElectionEP.ss_family) {
430 mStateChangeLog.log(ANDROID_LOG_WARN, LOG_TAG,
431 "TODO: add proper IPv6 support");
432 goto bailout;
433 }
434
435 // open a UDP socket for the timeline serivce
436 mSocket = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
437 if (mSocket < 0) {
438 mStateChangeLog.log(ANDROID_LOG_ERROR, LOG_TAG,
439 "Failed to create socket (errno = %d)", errno);
440 goto bailout;
441 }
442
443 // Bind to the selected interface using Linux's spiffy SO_BINDTODEVICE.
444 struct ifreq ifr;
445 memset(&ifr, 0, sizeof(ifr));
446 snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "%s", mBindIface.string());
447 ifr.ifr_name[sizeof(ifr.ifr_name) - 1] = 0;
448 rc = setsockopt(mSocket, SOL_SOCKET, SO_BINDTODEVICE,
449 (void *)&ifr, sizeof(ifr));
450 if (rc) {
451 mStateChangeLog.log(ANDROID_LOG_ERROR, LOG_TAG,
452 "Failed to bind socket at to interface %s "
453 "(errno = %d)", ifr.ifr_name, errno);
454 goto bailout;
455 }
456
457 // Bind our socket to INADDR_ANY and the master election port. The
458 // interface binding we made using SO_BINDTODEVICE should limit us to
459 // traffic only on the interface we are interested in. We need to bind to
460 // INADDR_ANY and the specific master election port in order to be able to
461 // receive both unicast traffic and master election multicast traffic with
462 // just a single socket.
463 struct sockaddr_in bindAddr;
464 ipv4_addr = reinterpret_cast<struct sockaddr_in*>(&mMasterElectionEP);
465 memcpy(&bindAddr, ipv4_addr, sizeof(bindAddr));
466 bindAddr.sin_addr.s_addr = INADDR_ANY;
467 rc = bind(mSocket,
468 reinterpret_cast<const sockaddr *>(&bindAddr),
469 sizeof(bindAddr));
470 if (rc) {
471 mStateChangeLog.log(ANDROID_LOG_ERROR, LOG_TAG,
472 "Failed to bind socket to port %hu (errno = %d)",
473 ntohs(bindAddr.sin_port), errno);
474 goto bailout;
475 }
476
477 if (0xE0000000 == (ntohl(ipv4_addr->sin_addr.s_addr) & 0xF0000000)) {
478 // If our master election endpoint is a multicast address, be sure to join
479 // the multicast group.
480 struct ip_mreq mreq;
481 mreq.imr_multiaddr = ipv4_addr->sin_addr;
482 mreq.imr_interface.s_addr = htonl(INADDR_ANY);
483 rc = setsockopt(mSocket, IPPROTO_IP, IP_ADD_MEMBERSHIP,
484 &mreq, sizeof(mreq));
485 if (rc == -1) {
486 ALOGE("Failed to join multicast group at %s. (errno = %d)",
487 masterElectionEPStr, errno);
488 goto bailout;
489 }
490
491 // disable loopback of multicast packets
492 const int zero = 0;
493 rc = setsockopt(mSocket, IPPROTO_IP, IP_MULTICAST_LOOP,
494 &zero, sizeof(zero));
495 if (rc == -1) {
496 mStateChangeLog.log(ANDROID_LOG_ERROR, LOG_TAG,
497 "Failed to disable multicast loopback "
498 "(errno = %d)", errno);
499 goto bailout;
500 }
501 } else
502 if (ntohl(ipv4_addr->sin_addr.s_addr) == 0xFFFFFFFF) {
503 // If the master election address is the broadcast address, then enable
504 // the broadcast socket option
505 rc = setsockopt(mSocket, SOL_SOCKET, SO_BROADCAST, &one, sizeof(one));
506 if (rc == -1) {
507 mStateChangeLog.log(ANDROID_LOG_ERROR, LOG_TAG,
508 "Failed to enable broadcast (errno = %d)",
509 errno);
510 goto bailout;
511 }
512 } else {
513 // If the master election address is neither broadcast, nor multicast,
514 // then we are misconfigured. The config API layer should prevent this
515 // from ever happening.
516 goto bailout;
517 }
518
519 // Set the TTL of sent packets to 1. (Time protocol sync should never leave
520 // the local subnet)
521 rc = setsockopt(mSocket, IPPROTO_IP, IP_TTL, &one, sizeof(one));
522 if (rc == -1) {
523 mStateChangeLog.log(ANDROID_LOG_ERROR, LOG_TAG,
524 "Failed to set TTL to %d (errno = %d)", one, errno);
525 goto bailout;
526 }
527
528 // get the device's unique ID
529 if (!assignDeviceID())
530 goto bailout;
531
532 ret_val = true;
533
534 bailout:
535 if (!ret_val)
536 cleanupSocket_l();
537 return ret_val;
538 }
539
540 // generate a unique device ID that can be used for arbitration
assignDeviceID()541 bool CommonTimeServer::assignDeviceID() {
542 if (!mBindIfaceValid)
543 return false;
544
545 struct ifreq ifr;
546 memset(&ifr, 0, sizeof(ifr));
547 ifr.ifr_addr.sa_family = AF_INET;
548 strlcpy(ifr.ifr_name, mBindIface.string(), IFNAMSIZ);
549
550 int rc = ioctl(mSocket, SIOCGIFHWADDR, &ifr);
551 if (rc) {
552 ALOGE("%s:%d ioctl failed", __PRETTY_FUNCTION__, __LINE__);
553 return false;
554 }
555
556 if (ifr.ifr_addr.sa_family != ARPHRD_ETHER) {
557 ALOGE("%s:%d got non-Ethernet address", __PRETTY_FUNCTION__, __LINE__);
558 return false;
559 }
560
561 mDeviceID = 0;
562 for (int i = 0; i < ETH_ALEN; i++) {
563 mDeviceID = (mDeviceID << 8) | ifr.ifr_hwaddr.sa_data[i];
564 }
565
566 return true;
567 }
568
569 // generate a new timeline ID
assignTimelineID()570 void CommonTimeServer::assignTimelineID() {
571 do {
572 mTimelineID = (static_cast<uint64_t>(lrand48()) << 32)
573 | static_cast<uint64_t>(lrand48());
574 } while (mTimelineID == ICommonClock::kInvalidTimelineID);
575 }
576
577 // Select a preference between the device IDs of two potential masters.
578 // Returns true if the first ID wins, or false if the second ID wins.
arbitrateMaster(uint64_t deviceID1,uint8_t devicePrio1,uint64_t deviceID2,uint8_t devicePrio2)579 bool CommonTimeServer::arbitrateMaster(
580 uint64_t deviceID1, uint8_t devicePrio1,
581 uint64_t deviceID2, uint8_t devicePrio2) {
582 return ((devicePrio1 > devicePrio2) ||
583 ((devicePrio1 == devicePrio2) && (deviceID1 > deviceID2)));
584 }
585
hexDumpToString(const uint8_t * src,size_t src_len,char * dst,size_t dst_len)586 static void hexDumpToString(const uint8_t* src, size_t src_len,
587 char* dst, size_t dst_len) {
588 size_t offset = 0;
589 size_t i;
590
591 for (i = 0; (i < src_len) && (offset < dst_len); ++i) {
592 int res;
593 if (0 == (i % 16)) {
594 res = snprintf(dst + offset, dst_len - offset, "\n%04zx :", i);
595 if (res < 0)
596 break;
597 offset += res;
598 if (offset >= dst_len)
599 break;
600 }
601
602 res = snprintf(dst + offset, dst_len - offset, " %02x", src[i]);
603 if (res < 0)
604 break;
605 offset += res;
606 }
607
608 dst[dst_len - 1] = 0;
609 }
610
handlePacket()611 bool CommonTimeServer::handlePacket() {
612 uint8_t buf[256];
613 struct sockaddr_storage srcAddr;
614 socklen_t srcAddrLen = sizeof(srcAddr);
615
616 ssize_t recvBytes = recvfrom(
617 mSocket, buf, sizeof(buf), 0,
618 reinterpret_cast<sockaddr *>(&srcAddr), &srcAddrLen);
619
620 if (recvBytes < 0) {
621 mBadPktLog.log(ANDROID_LOG_ERROR, LOG_TAG, "recvfrom failed (%s)",
622 strerror(errno));
623 return false;
624 }
625
626 UniversalTimeServicePacket pkt;
627 if (pkt.deserializePacket(buf, recvBytes, mSyncGroupID) < 0) {
628 char hex[256];
629 char srcEPStr[64];
630
631 hexDumpToString(buf, static_cast<size_t>(recvBytes), hex, sizeof(hex));
632 sockaddrToString(srcAddr, true, srcEPStr, sizeof(srcEPStr));
633
634 mBadPktLog.log("Failed to parse %d byte packet from %s.%s",
635 recvBytes, srcEPStr, hex);
636 return false;
637 }
638
639 bool result;
640 switch (pkt.packetType) {
641 case TIME_PACKET_WHO_IS_MASTER_REQUEST:
642 result = handleWhoIsMasterRequest(&pkt.p.who_is_master_request,
643 srcAddr);
644 break;
645
646 case TIME_PACKET_WHO_IS_MASTER_RESPONSE:
647 result = handleWhoIsMasterResponse(&pkt.p.who_is_master_response,
648 srcAddr);
649 break;
650
651 case TIME_PACKET_SYNC_REQUEST:
652 result = handleSyncRequest(&pkt.p.sync_request, srcAddr);
653 break;
654
655 case TIME_PACKET_SYNC_RESPONSE:
656 result = handleSyncResponse(&pkt.p.sync_response, srcAddr);
657 break;
658
659 case TIME_PACKET_MASTER_ANNOUNCEMENT:
660 result = handleMasterAnnouncement(&pkt.p.master_announcement,
661 srcAddr);
662 break;
663
664 default: {
665 char srcEPStr[64];
666 sockaddrToString(srcAddr, true, srcEPStr, sizeof(srcEPStr));
667
668 mBadPktLog.log(ANDROID_LOG_WARN, LOG_TAG,
669 "unknown packet type (%d) from %s",
670 pkt.packetType, srcEPStr);
671
672 result = false;
673 } break;
674 }
675
676 return result;
677 }
678
handleTimeout()679 bool CommonTimeServer::handleTimeout() {
680 // If we have no socket, then this must be a timeout to retry socket setup.
681 if (mSocket < 0)
682 return true;
683
684 switch (mState) {
685 case ICommonClock::STATE_INITIAL:
686 return handleTimeoutInitial();
687 case ICommonClock::STATE_CLIENT:
688 return handleTimeoutClient();
689 case ICommonClock::STATE_MASTER:
690 return handleTimeoutMaster();
691 case ICommonClock::STATE_RONIN:
692 return handleTimeoutRonin();
693 case ICommonClock::STATE_WAIT_FOR_ELECTION:
694 return handleTimeoutWaitForElection();
695 }
696
697 return false;
698 }
699
handleTimeoutInitial()700 bool CommonTimeServer::handleTimeoutInitial() {
701 if (++mInitial_WhoIsMasterRequestTimeouts ==
702 kInitial_NumWhoIsMasterRetries) {
703 // none of our attempts to discover a master succeeded, so make
704 // this device the master
705 return becomeMaster("initial timeout");
706 } else {
707 // retry the WhoIsMaster request
708 return sendWhoIsMasterRequest();
709 }
710 }
711
handleTimeoutClient()712 bool CommonTimeServer::handleTimeoutClient() {
713 if (shouldPanicNotGettingGoodData())
714 return becomeInitial("timeout panic, no good data");
715
716 if (mClient_SyncRequestPending) {
717 mClient_SyncRequestPending = false;
718
719 if (++mClient_SyncRequestTimeouts < kClient_NumSyncRequestRetries) {
720 // a sync request has timed out, so retry
721 return sendSyncRequest();
722 } else {
723 // The master has failed to respond to a sync request for too many
724 // times in a row. Assume the master is dead and start electing
725 // a new master.
726 return becomeRonin("master not responding");
727 }
728 } else {
729 // initiate the next sync request
730 return sendSyncRequest();
731 }
732 }
733
handleTimeoutMaster()734 bool CommonTimeServer::handleTimeoutMaster() {
735 // send another announcement from the master
736 return sendMasterAnnouncement();
737 }
738
handleTimeoutRonin()739 bool CommonTimeServer::handleTimeoutRonin() {
740 if (++mRonin_WhoIsMasterRequestTimeouts == kRonin_NumWhoIsMasterRetries) {
741 // no other master is out there, so we won the election
742 return becomeMaster("no better masters detected");
743 } else {
744 return sendWhoIsMasterRequest();
745 }
746 }
747
handleTimeoutWaitForElection()748 bool CommonTimeServer::handleTimeoutWaitForElection() {
749 return becomeRonin("timeout waiting for election conclusion");
750 }
751
handleWhoIsMasterRequest(const WhoIsMasterRequestPacket * request,const sockaddr_storage & srcAddr)752 bool CommonTimeServer::handleWhoIsMasterRequest(
753 const WhoIsMasterRequestPacket* request,
754 const sockaddr_storage& srcAddr) {
755 // Skip our own messages which come back via broadcast loopback.
756 if (request->senderDeviceID == mDeviceID)
757 return true;
758
759 char srcEPStr[64];
760 sockaddrToString(srcAddr, true, srcEPStr, sizeof(srcEPStr));
761 mElectionLog.log("RXed WhoIs master request while in state %s. "
762 "src %s reqTID %016llx ourTID %016llx",
763 stateToString(mState), srcEPStr,
764 request->timelineID, mTimelineID);
765
766 if (mState == ICommonClock::STATE_MASTER) {
767 // is this request related to this master's timeline?
768 if (request->timelineID != ICommonClock::kInvalidTimelineID &&
769 request->timelineID != mTimelineID)
770 return true;
771
772 WhoIsMasterResponsePacket pkt;
773 pkt.initHeader(mTimelineID, mSyncGroupID);
774 pkt.deviceID = mDeviceID;
775 pkt.devicePriority = effectivePriority();
776
777 mElectionLog.log("TXing WhoIs master resp to %s while in state %s. "
778 "ourTID %016llx ourGID %016llx ourDID %016llx "
779 "ourPrio %u",
780 srcEPStr, stateToString(mState),
781 mTimelineID, mSyncGroupID,
782 pkt.deviceID, pkt.devicePriority);
783
784 uint8_t buf[256];
785 ssize_t bufSz = pkt.serializePacket(buf, sizeof(buf));
786 if (bufSz < 0)
787 return false;
788
789 ssize_t sendBytes = sendto(
790 mSocket, buf, bufSz, 0,
791 reinterpret_cast<const sockaddr *>(&srcAddr),
792 sizeof(srcAddr));
793 if (sendBytes == -1) {
794 ALOGE("%s:%d sendto failed", __PRETTY_FUNCTION__, __LINE__);
795 return false;
796 }
797 } else if (mState == ICommonClock::STATE_RONIN) {
798 // if we hear a WhoIsMaster request from another device following
799 // the same timeline and that device wins arbitration, then we will stop
800 // trying to elect ourselves master and will instead wait for an
801 // announcement from the election winner
802 if (request->timelineID != mTimelineID)
803 return true;
804
805 if (arbitrateMaster(request->senderDeviceID,
806 request->senderDevicePriority,
807 mDeviceID,
808 effectivePriority()))
809 return becomeWaitForElection("would lose election");
810
811 return true;
812 } else if (mState == ICommonClock::STATE_INITIAL) {
813 // If a group of devices booted simultaneously (e.g. after a power
814 // outage) and all of them are in the initial state and there is no
815 // master, then each device may time out and declare itself master at
816 // the same time. To avoid this, listen for
817 // WhoIsMaster(InvalidTimeline) requests from peers. If we would lose
818 // arbitration against that peer, reset our timeout count so that the
819 // peer has a chance to become master before we time out.
820 if (request->timelineID == ICommonClock::kInvalidTimelineID &&
821 arbitrateMaster(request->senderDeviceID,
822 request->senderDevicePriority,
823 mDeviceID,
824 effectivePriority())) {
825 mInitial_WhoIsMasterRequestTimeouts = 0;
826 }
827 }
828
829 return true;
830 }
831
handleWhoIsMasterResponse(const WhoIsMasterResponsePacket * response,const sockaddr_storage & srcAddr)832 bool CommonTimeServer::handleWhoIsMasterResponse(
833 const WhoIsMasterResponsePacket* response,
834 const sockaddr_storage& srcAddr) {
835 // Skip our own messages which come back via broadcast loopback.
836 if (response->deviceID == mDeviceID)
837 return true;
838
839 char srcEPStr[64];
840 sockaddrToString(srcAddr, true, srcEPStr, sizeof(srcEPStr));
841 mElectionLog.log("RXed WhoIs master response while in state %s. "
842 "src %s respTID %016llx respDID %016llx respPrio %u "
843 "ourTID %016llx",
844 stateToString(mState), srcEPStr,
845 response->timelineID,
846 response->deviceID,
847 static_cast<uint32_t>(response->devicePriority),
848 mTimelineID);
849
850 if (mState == ICommonClock::STATE_INITIAL || mState == ICommonClock::STATE_RONIN) {
851 return becomeClient(srcAddr,
852 response->deviceID,
853 response->devicePriority,
854 response->timelineID,
855 "heard whois response");
856 } else if (mState == ICommonClock::STATE_CLIENT) {
857 // if we get multiple responses because there are multiple devices
858 // who believe that they are master, then follow the master that
859 // wins arbitration
860 if (arbitrateMaster(response->deviceID,
861 response->devicePriority,
862 mClient_MasterDeviceID,
863 mClient_MasterDevicePriority)) {
864 return becomeClient(srcAddr,
865 response->deviceID,
866 response->devicePriority,
867 response->timelineID,
868 "heard whois response");
869 }
870 }
871
872 return true;
873 }
874
handleSyncRequest(const SyncRequestPacket * request,const sockaddr_storage & srcAddr)875 bool CommonTimeServer::handleSyncRequest(const SyncRequestPacket* request,
876 const sockaddr_storage& srcAddr) {
877 SyncResponsePacket pkt;
878 pkt.initHeader(mTimelineID, mSyncGroupID);
879
880 if ((mState == ICommonClock::STATE_MASTER) &&
881 (mTimelineID == request->timelineID)) {
882 int64_t rxLocalTime = mLastPacketRxLocalTime;
883 int64_t rxCommonTime;
884
885 // If we are master on an actual network and have actual clients, then
886 // we are no longer low priority.
887 setForceLowPriority(false);
888
889 if (OK != mCommonClock.localToCommon(rxLocalTime, &rxCommonTime)) {
890 return false;
891 }
892
893 int64_t txLocalTime = mLocalClock.getLocalTime();;
894 int64_t txCommonTime;
895 if (OK != mCommonClock.localToCommon(txLocalTime, &txCommonTime)) {
896 return false;
897 }
898
899 pkt.nak = 0;
900 pkt.clientTxLocalTime = request->clientTxLocalTime;
901 pkt.masterRxCommonTime = rxCommonTime;
902 pkt.masterTxCommonTime = txCommonTime;
903 } else {
904 pkt.nak = 1;
905 pkt.clientTxLocalTime = 0;
906 pkt.masterRxCommonTime = 0;
907 pkt.masterTxCommonTime = 0;
908 }
909
910 uint8_t buf[256];
911 ssize_t bufSz = pkt.serializePacket(buf, sizeof(buf));
912 if (bufSz < 0)
913 return false;
914
915 ssize_t sendBytes = sendto(
916 mSocket, &buf, bufSz, 0,
917 reinterpret_cast<const sockaddr *>(&srcAddr),
918 sizeof(srcAddr));
919 if (sendBytes == -1) {
920 ALOGE("%s:%d sendto failed", __PRETTY_FUNCTION__, __LINE__);
921 return false;
922 }
923
924 return true;
925 }
926
handleSyncResponse(const SyncResponsePacket * response,const sockaddr_storage & srcAddr)927 bool CommonTimeServer::handleSyncResponse(
928 const SyncResponsePacket* response,
929 const sockaddr_storage& srcAddr) {
930 if (mState != ICommonClock::STATE_CLIENT)
931 return true;
932
933 assert(mMasterEPValid);
934 if (!sockaddrMatch(srcAddr, mMasterEP, true)) {
935 char srcEP[64], expectedEP[64];
936 sockaddrToString(srcAddr, true, srcEP, sizeof(srcEP));
937 sockaddrToString(mMasterEP, true, expectedEP, sizeof(expectedEP));
938 ALOGI("Dropping sync response from unexpected address."
939 " Expected %s Got %s", expectedEP, srcEP);
940 return true;
941 }
942
943 if (response->nak) {
944 // if our master is no longer accepting requests, then we need to find
945 // a new master
946 return becomeRonin("master NAK'ed");
947 }
948
949 mClient_SyncRequestPending = 0;
950 mClient_SyncRequestTimeouts = 0;
951 mClient_PacketRTTLog.logRX(response->clientTxLocalTime,
952 mLastPacketRxLocalTime);
953
954 bool result;
955 if (!(mClient_SyncRespsRXedFromCurMaster++)) {
956 // the first request/response exchange between a client and a master
957 // may take unusually long due to ARP, so discard it.
958 result = true;
959 } else {
960 int64_t clientTxLocalTime = response->clientTxLocalTime;
961 int64_t clientRxLocalTime = mLastPacketRxLocalTime;
962 int64_t masterTxCommonTime = response->masterTxCommonTime;
963 int64_t masterRxCommonTime = response->masterRxCommonTime;
964
965 int64_t rtt = (clientRxLocalTime - clientTxLocalTime);
966 int64_t avgLocal = (clientTxLocalTime + clientRxLocalTime) >> 1;
967 int64_t avgCommon = (masterTxCommonTime + masterRxCommonTime) >> 1;
968
969 // if the RTT of the packet is significantly larger than the panic
970 // threshold, we should simply discard it. Its better to do nothing
971 // than to take cues from a packet like that.
972 int64_t rttCommon = mCommonClock.localDurationToCommonDuration(rtt);
973 if (rttCommon > (static_cast<int64_t>(mPanicThresholdUsec) *
974 kRTTDiscardPanicThreshMultiplier)) {
975 ALOGV("Dropping sync response with RTT of %" PRId64 " uSec", rttCommon);
976 mClient_ExpiredSyncRespsRXedFromCurMaster++;
977 if (shouldPanicNotGettingGoodData())
978 return becomeInitial("RX panic, no good data");
979 return true;
980 } else {
981 result = mClockRecovery.pushDisciplineEvent(avgLocal, avgCommon, rttCommon);
982 mClient_LastGoodSyncRX = clientRxLocalTime;
983
984 if (result) {
985 // indicate to listeners that we've synced to the common timeline
986 notifyClockSync();
987 } else {
988 ALOGE("Panic! Observed clock sync error is too high to tolerate,"
989 " resetting state machine and starting over.");
990 notifyClockSyncLoss();
991 return becomeInitial("panic");
992 }
993 }
994 }
995
996 mCurTimeout.setTimeout(mSyncRequestIntervalMs);
997 return result;
998 }
999
handleMasterAnnouncement(const MasterAnnouncementPacket * packet,const sockaddr_storage & srcAddr)1000 bool CommonTimeServer::handleMasterAnnouncement(
1001 const MasterAnnouncementPacket* packet,
1002 const sockaddr_storage& srcAddr) {
1003 uint64_t newDeviceID = packet->deviceID;
1004 uint8_t newDevicePrio = packet->devicePriority;
1005 uint64_t newTimelineID = packet->timelineID;
1006
1007 // Skip our own messages which come back via broadcast loopback.
1008 if (newDeviceID == mDeviceID)
1009 return true;
1010
1011 char srcEPStr[64];
1012 sockaddrToString(srcAddr, true, srcEPStr, sizeof(srcEPStr));
1013 mElectionLog.log("RXed master announcement while in state %s. "
1014 "src %s srcDevID %lld srcPrio %u srcTID %016llx",
1015 stateToString(mState), srcEPStr,
1016 newDeviceID, static_cast<uint32_t>(newDevicePrio),
1017 newTimelineID);
1018
1019 if (mState == ICommonClock::STATE_INITIAL ||
1020 mState == ICommonClock::STATE_RONIN ||
1021 mState == ICommonClock::STATE_WAIT_FOR_ELECTION) {
1022 // if we aren't currently following a master, then start following
1023 // this new master
1024 return becomeClient(srcAddr,
1025 newDeviceID,
1026 newDevicePrio,
1027 newTimelineID,
1028 "heard master announcement");
1029 } else if (mState == ICommonClock::STATE_CLIENT) {
1030 // if the new master wins arbitration against our current master,
1031 // then become a client of the new master
1032 if (arbitrateMaster(newDeviceID,
1033 newDevicePrio,
1034 mClient_MasterDeviceID,
1035 mClient_MasterDevicePriority))
1036 return becomeClient(srcAddr,
1037 newDeviceID,
1038 newDevicePrio,
1039 newTimelineID,
1040 "heard master announcement");
1041 } else if (mState == ICommonClock::STATE_MASTER) {
1042 // two masters are competing - if the new one wins arbitration, then
1043 // cease acting as master
1044 if (arbitrateMaster(newDeviceID, newDevicePrio,
1045 mDeviceID, effectivePriority()))
1046 return becomeClient(srcAddr, newDeviceID,
1047 newDevicePrio, newTimelineID,
1048 "heard master announcement");
1049 }
1050
1051 return true;
1052 }
1053
sendWhoIsMasterRequest()1054 bool CommonTimeServer::sendWhoIsMasterRequest() {
1055 assert(mState == ICommonClock::STATE_INITIAL || mState == ICommonClock::STATE_RONIN);
1056
1057 // If we have no socket, then we must be in the unconfigured initial state.
1058 // Don't report any errors, just don't try to send the initial who-is-master
1059 // query. Eventually, our network will either become configured, or we will
1060 // be forced into network-less master mode by higher level code.
1061 if (mSocket < 0) {
1062 assert(mState == ICommonClock::STATE_INITIAL);
1063 return true;
1064 }
1065
1066 bool ret = false;
1067 WhoIsMasterRequestPacket pkt;
1068 pkt.initHeader(mSyncGroupID);
1069 pkt.senderDeviceID = mDeviceID;
1070 pkt.senderDevicePriority = effectivePriority();
1071
1072 uint8_t buf[256];
1073 ssize_t bufSz = pkt.serializePacket(buf, sizeof(buf));
1074 if (bufSz >= 0) {
1075 char dstEPStr[64];
1076 sockaddrToString(mMasterElectionEP, true, dstEPStr, sizeof(dstEPStr));
1077 mElectionLog.log("TXing WhoIs master request to %s while in state %s. "
1078 "ourTID %016llx ourGID %016llx ourDID %016llx "
1079 "ourPrio %u",
1080 dstEPStr, stateToString(mState),
1081 mTimelineID, mSyncGroupID,
1082 pkt.senderDeviceID, pkt.senderDevicePriority);
1083
1084 ssize_t sendBytes = sendto(
1085 mSocket, buf, bufSz, 0,
1086 reinterpret_cast<const sockaddr *>(&mMasterElectionEP),
1087 sizeof(mMasterElectionEP));
1088 if (sendBytes < 0)
1089 ALOGE("WhoIsMaster sendto failed (errno %d)", errno);
1090 ret = true;
1091 }
1092
1093 if (mState == ICommonClock::STATE_INITIAL) {
1094 mCurTimeout.setTimeout(kInitial_WhoIsMasterTimeoutMs);
1095 } else {
1096 mCurTimeout.setTimeout(kRonin_WhoIsMasterTimeoutMs);
1097 }
1098
1099 return ret;
1100 }
1101
sendSyncRequest()1102 bool CommonTimeServer::sendSyncRequest() {
1103 // If we are sending sync requests, then we must be in the client state and
1104 // we must have a socket (when we have no network, we are only supposed to
1105 // be in INITIAL or MASTER)
1106 assert(mState == ICommonClock::STATE_CLIENT);
1107 assert(mSocket >= 0);
1108
1109 bool ret = false;
1110 SyncRequestPacket pkt;
1111 pkt.initHeader(mTimelineID, mSyncGroupID);
1112 pkt.clientTxLocalTime = mLocalClock.getLocalTime();
1113
1114 if (!mClient_FirstSyncTX)
1115 mClient_FirstSyncTX = pkt.clientTxLocalTime;
1116
1117 mClient_PacketRTTLog.logTX(pkt.clientTxLocalTime);
1118
1119 uint8_t buf[256];
1120 ssize_t bufSz = pkt.serializePacket(buf, sizeof(buf));
1121 if (bufSz >= 0) {
1122 ssize_t sendBytes = sendto(
1123 mSocket, buf, bufSz, 0,
1124 reinterpret_cast<const sockaddr *>(&mMasterEP),
1125 sizeof(mMasterEP));
1126 if (sendBytes < 0)
1127 ALOGE("SyncRequest sendto failed (errno %d)", errno);
1128 ret = true;
1129 }
1130
1131 mClient_SyncsSentToCurMaster++;
1132 mCurTimeout.setTimeout(mSyncRequestIntervalMs);
1133 mClient_SyncRequestPending = true;
1134
1135 return ret;
1136 }
1137
sendMasterAnnouncement()1138 bool CommonTimeServer::sendMasterAnnouncement() {
1139 bool ret = false;
1140 assert(mState == ICommonClock::STATE_MASTER);
1141
1142 // If we are being asked to send a master announcement, but we have no
1143 // socket, we must be in network-less master mode. Don't bother to send the
1144 // announcement, and don't bother to schedule a timeout. When the network
1145 // comes up, the work thread will get poked and start the process of
1146 // figuring out who the current master should be.
1147 if (mSocket < 0) {
1148 mCurTimeout.setTimeout(kInfiniteTimeout);
1149 return true;
1150 }
1151
1152 MasterAnnouncementPacket pkt;
1153 pkt.initHeader(mTimelineID, mSyncGroupID);
1154 pkt.deviceID = mDeviceID;
1155 pkt.devicePriority = effectivePriority();
1156
1157 uint8_t buf[256];
1158 ssize_t bufSz = pkt.serializePacket(buf, sizeof(buf));
1159 if (bufSz >= 0) {
1160 char dstEPStr[64];
1161 sockaddrToString(mMasterElectionEP, true, dstEPStr, sizeof(dstEPStr));
1162 mElectionLog.log("TXing Master announcement to %s while in state %s. "
1163 "ourTID %016llx ourGID %016llx ourDID %016llx "
1164 "ourPrio %u",
1165 dstEPStr, stateToString(mState),
1166 mTimelineID, mSyncGroupID,
1167 pkt.deviceID, pkt.devicePriority);
1168
1169 ssize_t sendBytes = sendto(
1170 mSocket, buf, bufSz, 0,
1171 reinterpret_cast<const sockaddr *>(&mMasterElectionEP),
1172 sizeof(mMasterElectionEP));
1173 if (sendBytes < 0)
1174 ALOGE("MasterAnnouncement sendto failed (errno %d)", errno);
1175 ret = true;
1176 }
1177
1178 mCurTimeout.setTimeout(mMasterAnnounceIntervalMs);
1179 return ret;
1180 }
1181
becomeClient(const sockaddr_storage & masterEP,uint64_t masterDeviceID,uint8_t masterDevicePriority,uint64_t timelineID,const char * cause)1182 bool CommonTimeServer::becomeClient(const sockaddr_storage& masterEP,
1183 uint64_t masterDeviceID,
1184 uint8_t masterDevicePriority,
1185 uint64_t timelineID,
1186 const char* cause) {
1187 char newEPStr[64], oldEPStr[64];
1188 sockaddrToString(masterEP, true, newEPStr, sizeof(newEPStr));
1189 sockaddrToString(mMasterEP, mMasterEPValid, oldEPStr, sizeof(oldEPStr));
1190
1191 mStateChangeLog.log(ANDROID_LOG_INFO, LOG_TAG,
1192 "%s --> CLIENT (%s) :%s"
1193 " OldMaster: %02x-%014llx::%016llx::%s"
1194 " NewMaster: %02x-%014llx::%016llx::%s",
1195 stateToString(mState), cause,
1196 (mTimelineID != timelineID) ? " (new timeline)" : "",
1197 mClient_MasterDevicePriority, mClient_MasterDeviceID,
1198 mTimelineID, oldEPStr,
1199 masterDevicePriority, masterDeviceID,
1200 timelineID, newEPStr);
1201
1202 if (mTimelineID != timelineID) {
1203 // start following a new timeline
1204 mTimelineID = timelineID;
1205 mClockRecovery.reset(true, true);
1206 notifyClockSyncLoss();
1207 } else {
1208 // start following a new master on the existing timeline
1209 mClockRecovery.reset(false, true);
1210 }
1211
1212 mMasterEP = masterEP;
1213 mMasterEPValid = true;
1214
1215 // If we are on a real network as a client of a real master, then we should
1216 // no longer force low priority. If our master disappears, we should have
1217 // the high priority bit set during the election to replace the master
1218 // because this group was a real group and not a singleton created in
1219 // networkless mode.
1220 setForceLowPriority(false);
1221
1222 mClient_MasterDeviceID = masterDeviceID;
1223 mClient_MasterDevicePriority = masterDevicePriority;
1224 resetSyncStats();
1225
1226 setState(ICommonClock::STATE_CLIENT);
1227
1228 // add some jitter to when the various clients send their requests
1229 // in order to reduce the likelihood that a group of clients overload
1230 // the master after receiving a master announcement
1231 usleep((lrand48() % 100) * 1000);
1232
1233 return sendSyncRequest();
1234 }
1235
becomeMaster(const char * cause)1236 bool CommonTimeServer::becomeMaster(const char* cause) {
1237 uint64_t oldTimelineID = mTimelineID;
1238 if (mTimelineID == ICommonClock::kInvalidTimelineID) {
1239 // this device has not been following any existing timeline,
1240 // so it will create a new timeline and declare itself master
1241 assert(!mCommonClock.isValid());
1242
1243 // set the common time basis
1244 mCommonClock.setBasis(mLocalClock.getLocalTime(), 0);
1245
1246 // assign an arbitrary timeline iD
1247 assignTimelineID();
1248
1249 // notify listeners that we've created a common timeline
1250 notifyClockSync();
1251 }
1252
1253 mStateChangeLog.log(ANDROID_LOG_INFO, LOG_TAG,
1254 "%s --> MASTER (%s) : %s timeline %016llx",
1255 stateToString(mState), cause,
1256 (oldTimelineID == mTimelineID) ? "taking ownership of"
1257 : "creating new",
1258 mTimelineID);
1259
1260 memset(&mMasterEP, 0, sizeof(mMasterEP));
1261 mMasterEPValid = false;
1262 mClient_MasterDevicePriority = effectivePriority();
1263 mClient_MasterDeviceID = mDeviceID;
1264 mClockRecovery.reset(false, true);
1265 resetSyncStats();
1266
1267 setState(ICommonClock::STATE_MASTER);
1268 return sendMasterAnnouncement();
1269 }
1270
becomeRonin(const char * cause)1271 bool CommonTimeServer::becomeRonin(const char* cause) {
1272 // If we were the client of a given timeline, but had never received even a
1273 // single time sync packet, then we transition back to Initial instead of
1274 // Ronin. If we transition to Ronin and end up becoming the new Master, we
1275 // will be unable to service requests for other clients because we never
1276 // actually knew what time it was. By going to initial, we ensure that
1277 // other clients who know what time it is, but would lose master arbitration
1278 // in the Ronin case, will step up and become the proper new master of the
1279 // old timeline.
1280
1281 char oldEPStr[64];
1282 sockaddrToString(mMasterEP, mMasterEPValid, oldEPStr, sizeof(oldEPStr));
1283 memset(&mMasterEP, 0, sizeof(mMasterEP));
1284 mMasterEPValid = false;
1285
1286 if (mCommonClock.isValid()) {
1287 mStateChangeLog.log(ANDROID_LOG_INFO, LOG_TAG,
1288 "%s --> RONIN (%s) : lost track of previously valid timeline "
1289 "%02x-%014llx::%016llx::%s (%d TXed %d RXed %d RXExpired)",
1290 stateToString(mState), cause,
1291 mClient_MasterDevicePriority, mClient_MasterDeviceID,
1292 mTimelineID, oldEPStr,
1293 mClient_SyncsSentToCurMaster,
1294 mClient_SyncRespsRXedFromCurMaster,
1295 mClient_ExpiredSyncRespsRXedFromCurMaster);
1296
1297 mRonin_WhoIsMasterRequestTimeouts = 0;
1298 setState(ICommonClock::STATE_RONIN);
1299 return sendWhoIsMasterRequest();
1300 } else {
1301 mStateChangeLog.log(ANDROID_LOG_INFO, LOG_TAG,
1302 "%s --> INITIAL (%s) : never synced timeline "
1303 "%02x-%014llx::%016llx::%s (%d TXed %d RXed %d RXExpired)",
1304 stateToString(mState), cause,
1305 mClient_MasterDevicePriority, mClient_MasterDeviceID,
1306 mTimelineID, oldEPStr,
1307 mClient_SyncsSentToCurMaster,
1308 mClient_SyncRespsRXedFromCurMaster,
1309 mClient_ExpiredSyncRespsRXedFromCurMaster);
1310
1311 return becomeInitial("ronin, no timeline");
1312 }
1313 }
1314
becomeWaitForElection(const char * cause)1315 bool CommonTimeServer::becomeWaitForElection(const char* cause) {
1316 mStateChangeLog.log(ANDROID_LOG_INFO, LOG_TAG,
1317 "%s --> WAIT_FOR_ELECTION (%s) : dropping out of election,"
1318 " waiting %d mSec for completion.",
1319 stateToString(mState), cause, kWaitForElection_TimeoutMs);
1320
1321 setState(ICommonClock::STATE_WAIT_FOR_ELECTION);
1322 mCurTimeout.setTimeout(kWaitForElection_TimeoutMs);
1323 return true;
1324 }
1325
becomeInitial(const char * cause)1326 bool CommonTimeServer::becomeInitial(const char* cause) {
1327 mStateChangeLog.log(ANDROID_LOG_INFO, LOG_TAG,
1328 "Entering INITIAL (%s), total reset.",
1329 cause);
1330
1331 setState(ICommonClock::STATE_INITIAL);
1332
1333 // reset clock recovery
1334 mClockRecovery.reset(true, true);
1335
1336 // reset internal state bookkeeping.
1337 mCurTimeout.setTimeout(kInfiniteTimeout);
1338 memset(&mMasterEP, 0, sizeof(mMasterEP));
1339 mMasterEPValid = false;
1340 mLastPacketRxLocalTime = 0;
1341 mTimelineID = ICommonClock::kInvalidTimelineID;
1342 mClockSynced = false;
1343 mInitial_WhoIsMasterRequestTimeouts = 0;
1344 mClient_MasterDeviceID = 0;
1345 mClient_MasterDevicePriority = 0;
1346 mRonin_WhoIsMasterRequestTimeouts = 0;
1347 resetSyncStats();
1348
1349 // send the first request to discover the master
1350 return sendWhoIsMasterRequest();
1351 }
1352
notifyClockSync()1353 void CommonTimeServer::notifyClockSync() {
1354 if (!mClockSynced) {
1355 mClockSynced = true;
1356 mICommonClock->notifyOnTimelineChanged(mTimelineID);
1357 }
1358 }
1359
notifyClockSyncLoss()1360 void CommonTimeServer::notifyClockSyncLoss() {
1361 if (mClockSynced) {
1362 mClockSynced = false;
1363 mICommonClock->notifyOnTimelineChanged(
1364 ICommonClock::kInvalidTimelineID);
1365 }
1366 }
1367
setState(ICommonClock::State s)1368 void CommonTimeServer::setState(ICommonClock::State s) {
1369 mState = s;
1370 }
1371
stateToString(ICommonClock::State s)1372 const char* CommonTimeServer::stateToString(ICommonClock::State s) {
1373 switch(s) {
1374 case ICommonClock::STATE_INITIAL:
1375 return "INITIAL";
1376 case ICommonClock::STATE_CLIENT:
1377 return "CLIENT";
1378 case ICommonClock::STATE_MASTER:
1379 return "MASTER";
1380 case ICommonClock::STATE_RONIN:
1381 return "RONIN";
1382 case ICommonClock::STATE_WAIT_FOR_ELECTION:
1383 return "WAIT_FOR_ELECTION";
1384 default:
1385 return "unknown";
1386 }
1387 }
1388
sockaddrToString(const sockaddr_storage & addr,bool addrValid,char * buf,size_t bufLen)1389 void CommonTimeServer::sockaddrToString(const sockaddr_storage& addr,
1390 bool addrValid,
1391 char* buf, size_t bufLen) {
1392 if (!bufLen || !buf)
1393 return;
1394
1395 if (addrValid) {
1396 switch (addr.ss_family) {
1397 case AF_INET: {
1398 const struct sockaddr_in* sa =
1399 reinterpret_cast<const struct sockaddr_in*>(&addr);
1400 unsigned long a = ntohl(sa->sin_addr.s_addr);
1401 uint16_t p = ntohs(sa->sin_port);
1402 snprintf(buf, bufLen, "%lu.%lu.%lu.%lu:%hu",
1403 ((a >> 24) & 0xFF), ((a >> 16) & 0xFF),
1404 ((a >> 8) & 0xFF), (a & 0xFF), p);
1405 } break;
1406
1407 case AF_INET6: {
1408 const struct sockaddr_in6* sa =
1409 reinterpret_cast<const struct sockaddr_in6*>(&addr);
1410 const uint8_t* a = sa->sin6_addr.s6_addr;
1411 uint16_t p = ntohs(sa->sin6_port);
1412 snprintf(buf, bufLen,
1413 "%02X%02X:%02X%02X:%02X%02X:%02X%02X:"
1414 "%02X%02X:%02X%02X:%02X%02X:%02X%02X port %hd",
1415 a[0], a[1], a[ 2], a[ 3], a[ 4], a[ 5], a[ 6], a[ 7],
1416 a[8], a[9], a[10], a[11], a[12], a[13], a[14], a[15],
1417 p);
1418 } break;
1419
1420 default:
1421 snprintf(buf, bufLen,
1422 "<unknown sockaddr family %d>", addr.ss_family);
1423 break;
1424 }
1425 } else {
1426 snprintf(buf, bufLen, "<none>");
1427 }
1428
1429 buf[bufLen - 1] = 0;
1430 }
1431
sockaddrMatch(const sockaddr_storage & a1,const sockaddr_storage & a2,bool matchAddressOnly)1432 bool CommonTimeServer::sockaddrMatch(const sockaddr_storage& a1,
1433 const sockaddr_storage& a2,
1434 bool matchAddressOnly) {
1435 if (a1.ss_family != a2.ss_family)
1436 return false;
1437
1438 switch (a1.ss_family) {
1439 case AF_INET: {
1440 const struct sockaddr_in* sa1 =
1441 reinterpret_cast<const struct sockaddr_in*>(&a1);
1442 const struct sockaddr_in* sa2 =
1443 reinterpret_cast<const struct sockaddr_in*>(&a2);
1444
1445 if (sa1->sin_addr.s_addr != sa2->sin_addr.s_addr)
1446 return false;
1447
1448 return (matchAddressOnly || (sa1->sin_port == sa2->sin_port));
1449 } break;
1450
1451 case AF_INET6: {
1452 const struct sockaddr_in6* sa1 =
1453 reinterpret_cast<const struct sockaddr_in6*>(&a1);
1454 const struct sockaddr_in6* sa2 =
1455 reinterpret_cast<const struct sockaddr_in6*>(&a2);
1456
1457 if (memcmp(&sa1->sin6_addr, &sa2->sin6_addr, sizeof(sa2->sin6_addr)))
1458 return false;
1459
1460 return (matchAddressOnly || (sa1->sin6_port == sa2->sin6_port));
1461 } break;
1462
1463 // Huh? We don't deal in non-IPv[46] addresses. Not sure how we got
1464 // here, but we don't know how to comapre these addresses and simply
1465 // default to a no-match decision.
1466 default: return false;
1467 }
1468 }
1469
shouldPanicNotGettingGoodData()1470 bool CommonTimeServer::shouldPanicNotGettingGoodData() {
1471 if (mClient_FirstSyncTX) {
1472 int64_t now = mLocalClock.getLocalTime();
1473 int64_t delta = now - (mClient_LastGoodSyncRX
1474 ? mClient_LastGoodSyncRX
1475 : mClient_FirstSyncTX);
1476 int64_t deltaUsec = mCommonClock.localDurationToCommonDuration(delta);
1477
1478 if (deltaUsec >= kNoGoodDataPanicThresholdUsec)
1479 return true;
1480 }
1481
1482 return false;
1483 }
1484
logTX(int64_t txTime)1485 void CommonTimeServer::PacketRTTLog::logTX(int64_t txTime) {
1486 txTimes[wrPtr] = txTime;
1487 rxTimes[wrPtr] = 0;
1488 wrPtr = (wrPtr + 1) % RTT_LOG_SIZE;
1489 if (!wrPtr)
1490 logFull = true;
1491 }
1492
logRX(int64_t txTime,int64_t rxTime)1493 void CommonTimeServer::PacketRTTLog::logRX(int64_t txTime, int64_t rxTime) {
1494 if (!logFull && !wrPtr)
1495 return;
1496
1497 uint32_t i = logFull ? wrPtr : 0;
1498 do {
1499 if (txTimes[i] == txTime) {
1500 rxTimes[i] = rxTime;
1501 break;
1502 }
1503 i = (i + 1) % RTT_LOG_SIZE;
1504 } while (i != wrPtr);
1505 }
1506
1507 } // namespace android
1508