1496 lines
54 KiB
C++
Raw Normal View History

/*
* Copyright (C) 2012 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* A service that exchanges time synchronization information between
* a master that defines a timeline and clients that follow the timeline.
*/
#define LOG_TAG "common_time"
#include <utils/Log.h>
#include <arpa/inet.h>
#include <assert.h>
#include <fcntl.h>
#include <linux/if_ether.h>
#include <net/if.h>
#include <net/if_arp.h>
#include <netinet/ip.h>
#include <poll.h>
#include <stdio.h>
#include <sys/eventfd.h>
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <common_time/local_clock.h>
#include <binder/IPCThreadState.h>
#include <binder/ProcessState.h>
#include <utils/Timers.h>
#include "common_clock_service.h"
#include "common_time_config_service.h"
#include "common_time_server.h"
#include "common_time_server_packets.h"
#include "clock_recovery.h"
#include "common_clock.h"
#define MAX_INT ((int)0x7FFFFFFF)
namespace android {
const char* CommonTimeServer::kDefaultMasterElectionAddr = "239.195.128.88";
const uint16_t CommonTimeServer::kDefaultMasterElectionPort = 8887;
const uint64_t CommonTimeServer::kDefaultSyncGroupID = 0;
const uint8_t CommonTimeServer::kDefaultMasterPriority = 1;
const uint32_t CommonTimeServer::kDefaultMasterAnnounceIntervalMs = 10000;
const uint32_t CommonTimeServer::kDefaultSyncRequestIntervalMs = 1000;
const uint32_t CommonTimeServer::kDefaultPanicThresholdUsec = 50000;
const bool CommonTimeServer::kDefaultAutoDisable = true;
const int CommonTimeServer::kSetupRetryTimeoutMs = 30000;
const int64_t CommonTimeServer::kNoGoodDataPanicThresholdUsec = 600000000ll;
const uint32_t CommonTimeServer::kRTTDiscardPanicThreshMultiplier = 5;
// timeout value representing an infinite timeout
const int CommonTimeServer::kInfiniteTimeout = -1;
/*** Initial state constants ***/
// number of WhoIsMaster attempts sent before giving up
const int CommonTimeServer::kInitial_NumWhoIsMasterRetries = 6;
// timeout used when waiting for a response to a WhoIsMaster request
const int CommonTimeServer::kInitial_WhoIsMasterTimeoutMs = 500;
/*** Client state constants ***/
// number of sync requests that can fail before a client assumes its master
// is dead
Make common_time more deferential when coming out of networkless mode. Addresses issues seen in bug 6260139. This is a really tough bug to repro, but there is no doubt that it is happening occasionally on our super huge A@H subnet. I have collected data all weekend; the failure did not occur, but I got enough to have a theoretical sequence of events which could trigger this behavior. The sequence goes like this. 1) A network is running and happy with a timeline master M, maintaining timeline X. 2) Device B boots, but its network is taking a long time to come up. After 60 seconds of waiting for the network to come up, device B goes into networkless master mode and creates timeline Y. 3) Device B's network comes up. It immediately sends a master announcement saying that it is the current low-priority master of timeline Y (its low priority because it has never had any real clients) 4) Master M ignores B because B is low priority. 5) Device C boots and sends out a who is master request. It is a race between M and A to see who will respond first. In this case, A responds first. 6) C sends B a request which B receives. B now has its first client and is now high priority. In this scenario, B matches M in all aspects of the priority ranking function, including winning the tie breaker (larger MAC address when interpreted as a 48 bit integer) 7) M sends its master announcement; it is ignored by B since B now wins in the ranking function vs M. 8) Finally, B sends its next master announcement. M sees it, realizes that there is a higher priority master out there (looks like a bridged network scenario to M). M gives up master status along with timeline X. The clients of M become clients of B and move from timeline X to timeline Y (something which should only be needed during an actual network bridging event) This change has a few different things meant to severely minimize the chance that this can happen. First, and the most important change, is that networkless masters do not immediately announce themselves as masters on the network they are joining. Instead, they transition into Ronin to discover any pre-existing masters on the network. If there are no masters out there, the device will simply transition back to master and continue to maintain the timeline it had in networkless mode. In the scenario above, however, B should discover M and become its client, preserving the established timeline X. Second, any time a device experienced an interface reconfiguration (including coming out of networkless mode), it clears its high priority bit. This is a good thing. The bit used to get set again any time... 1) The device is master and receives a client request. 2) The device becomes a client of another master on the network. 3) The device becomes a master. Number 3 in this list is a mistake. The high priority bit should only be set for devices during master election which have been participating in a timeline which has been used by multiple devices. We know that this is the case when we are master and receive a request. We also know that this is the case when we hear from a master and decide to become its client. Simply becoming a master should not make us high priority. This behavior has been removed. Third, timeouts have been adjusted just for some extra "stickyness" when it comes to master status. Clients now say in the Ronin state for up to 10 seconds looking for a master sending up to 20 discovery requests, instead of only 3 seconds (sending 6 requests). The wait-for-election timeout has been adjusted up from 5 seconds to 12.5 seconds to track the longer election cycle as well. Also, while in steady-state, clients will now wait until 10 packets (10 seconds) have not been answered by its master before giving up and dropping into Ronin. Change-Id: I438b39f31265e34d6719d4adfa9e8b95a2afc188 Signed-off-by: John Grossman <johngro@google.com>
2012-04-09 11:26:16 -07:00
const int CommonTimeServer::kClient_NumSyncRequestRetries = 10;
/*** Master state constants ***/
/*** Ronin state constants ***/
// number of WhoIsMaster attempts sent before declaring ourselves master
Make common_time more deferential when coming out of networkless mode. Addresses issues seen in bug 6260139. This is a really tough bug to repro, but there is no doubt that it is happening occasionally on our super huge A@H subnet. I have collected data all weekend; the failure did not occur, but I got enough to have a theoretical sequence of events which could trigger this behavior. The sequence goes like this. 1) A network is running and happy with a timeline master M, maintaining timeline X. 2) Device B boots, but its network is taking a long time to come up. After 60 seconds of waiting for the network to come up, device B goes into networkless master mode and creates timeline Y. 3) Device B's network comes up. It immediately sends a master announcement saying that it is the current low-priority master of timeline Y (its low priority because it has never had any real clients) 4) Master M ignores B because B is low priority. 5) Device C boots and sends out a who is master request. It is a race between M and A to see who will respond first. In this case, A responds first. 6) C sends B a request which B receives. B now has its first client and is now high priority. In this scenario, B matches M in all aspects of the priority ranking function, including winning the tie breaker (larger MAC address when interpreted as a 48 bit integer) 7) M sends its master announcement; it is ignored by B since B now wins in the ranking function vs M. 8) Finally, B sends its next master announcement. M sees it, realizes that there is a higher priority master out there (looks like a bridged network scenario to M). M gives up master status along with timeline X. The clients of M become clients of B and move from timeline X to timeline Y (something which should only be needed during an actual network bridging event) This change has a few different things meant to severely minimize the chance that this can happen. First, and the most important change, is that networkless masters do not immediately announce themselves as masters on the network they are joining. Instead, they transition into Ronin to discover any pre-existing masters on the network. If there are no masters out there, the device will simply transition back to master and continue to maintain the timeline it had in networkless mode. In the scenario above, however, B should discover M and become its client, preserving the established timeline X. Second, any time a device experienced an interface reconfiguration (including coming out of networkless mode), it clears its high priority bit. This is a good thing. The bit used to get set again any time... 1) The device is master and receives a client request. 2) The device becomes a client of another master on the network. 3) The device becomes a master. Number 3 in this list is a mistake. The high priority bit should only be set for devices during master election which have been participating in a timeline which has been used by multiple devices. We know that this is the case when we are master and receive a request. We also know that this is the case when we hear from a master and decide to become its client. Simply becoming a master should not make us high priority. This behavior has been removed. Third, timeouts have been adjusted just for some extra "stickyness" when it comes to master status. Clients now say in the Ronin state for up to 10 seconds looking for a master sending up to 20 discovery requests, instead of only 3 seconds (sending 6 requests). The wait-for-election timeout has been adjusted up from 5 seconds to 12.5 seconds to track the longer election cycle as well. Also, while in steady-state, clients will now wait until 10 packets (10 seconds) have not been answered by its master before giving up and dropping into Ronin. Change-Id: I438b39f31265e34d6719d4adfa9e8b95a2afc188 Signed-off-by: John Grossman <johngro@google.com>
2012-04-09 11:26:16 -07:00
const int CommonTimeServer::kRonin_NumWhoIsMasterRetries = 20;
// timeout used when waiting for a response to a WhoIsMaster request
const int CommonTimeServer::kRonin_WhoIsMasterTimeoutMs = 500;
/*** WaitForElection state constants ***/
// how long do we wait for an announcement from a master before
// trying another election?
Make common_time more deferential when coming out of networkless mode. Addresses issues seen in bug 6260139. This is a really tough bug to repro, but there is no doubt that it is happening occasionally on our super huge A@H subnet. I have collected data all weekend; the failure did not occur, but I got enough to have a theoretical sequence of events which could trigger this behavior. The sequence goes like this. 1) A network is running and happy with a timeline master M, maintaining timeline X. 2) Device B boots, but its network is taking a long time to come up. After 60 seconds of waiting for the network to come up, device B goes into networkless master mode and creates timeline Y. 3) Device B's network comes up. It immediately sends a master announcement saying that it is the current low-priority master of timeline Y (its low priority because it has never had any real clients) 4) Master M ignores B because B is low priority. 5) Device C boots and sends out a who is master request. It is a race between M and A to see who will respond first. In this case, A responds first. 6) C sends B a request which B receives. B now has its first client and is now high priority. In this scenario, B matches M in all aspects of the priority ranking function, including winning the tie breaker (larger MAC address when interpreted as a 48 bit integer) 7) M sends its master announcement; it is ignored by B since B now wins in the ranking function vs M. 8) Finally, B sends its next master announcement. M sees it, realizes that there is a higher priority master out there (looks like a bridged network scenario to M). M gives up master status along with timeline X. The clients of M become clients of B and move from timeline X to timeline Y (something which should only be needed during an actual network bridging event) This change has a few different things meant to severely minimize the chance that this can happen. First, and the most important change, is that networkless masters do not immediately announce themselves as masters on the network they are joining. Instead, they transition into Ronin to discover any pre-existing masters on the network. If there are no masters out there, the device will simply transition back to master and continue to maintain the timeline it had in networkless mode. In the scenario above, however, B should discover M and become its client, preserving the established timeline X. Second, any time a device experienced an interface reconfiguration (including coming out of networkless mode), it clears its high priority bit. This is a good thing. The bit used to get set again any time... 1) The device is master and receives a client request. 2) The device becomes a client of another master on the network. 3) The device becomes a master. Number 3 in this list is a mistake. The high priority bit should only be set for devices during master election which have been participating in a timeline which has been used by multiple devices. We know that this is the case when we are master and receive a request. We also know that this is the case when we hear from a master and decide to become its client. Simply becoming a master should not make us high priority. This behavior has been removed. Third, timeouts have been adjusted just for some extra "stickyness" when it comes to master status. Clients now say in the Ronin state for up to 10 seconds looking for a master sending up to 20 discovery requests, instead of only 3 seconds (sending 6 requests). The wait-for-election timeout has been adjusted up from 5 seconds to 12.5 seconds to track the longer election cycle as well. Also, while in steady-state, clients will now wait until 10 packets (10 seconds) have not been answered by its master before giving up and dropping into Ronin. Change-Id: I438b39f31265e34d6719d4adfa9e8b95a2afc188 Signed-off-by: John Grossman <johngro@google.com>
2012-04-09 11:26:16 -07:00
const int CommonTimeServer::kWaitForElection_TimeoutMs = 12500;
CommonTimeServer::CommonTimeServer()
: Thread(false)
, mState(ICommonClock::STATE_INITIAL)
, mClockRecovery(&mLocalClock, &mCommonClock)
, mSocket(-1)
, mLastPacketRxLocalTime(0)
, mTimelineID(ICommonClock::kInvalidTimelineID)
, mClockSynced(false)
, mCommonClockHasClients(false)
, mStateChangeLog("Recent State Change Events", 30)
, mElectionLog("Recent Master Election Traffic", 30)
, mBadPktLog("Recent Bad Packet RX Info", 8)
, mInitial_WhoIsMasterRequestTimeouts(0)
, mClient_MasterDeviceID(0)
, mClient_MasterDevicePriority(0)
, mRonin_WhoIsMasterRequestTimeouts(0) {
// zero out sync stats
resetSyncStats();
// Setup the master election endpoint to use the default.
struct sockaddr_in* meep =
reinterpret_cast<struct sockaddr_in*>(&mMasterElectionEP);
memset(&mMasterElectionEP, 0, sizeof(mMasterElectionEP));
inet_aton(kDefaultMasterElectionAddr, &meep->sin_addr);
meep->sin_family = AF_INET;
meep->sin_port = htons(kDefaultMasterElectionPort);
// Zero out the master endpoint.
memset(&mMasterEP, 0, sizeof(mMasterEP));
mMasterEPValid = false;
mBindIfaceValid = false;
setForceLowPriority(false);
// Set all remaining configuration parameters to their defaults.
mDeviceID = 0;
mSyncGroupID = kDefaultSyncGroupID;
mMasterPriority = kDefaultMasterPriority;
mMasterAnnounceIntervalMs = kDefaultMasterAnnounceIntervalMs;
mSyncRequestIntervalMs = kDefaultSyncRequestIntervalMs;
mPanicThresholdUsec = kDefaultPanicThresholdUsec;
mAutoDisable = kDefaultAutoDisable;
// Create the eventfd we will use to signal our thread to wake up when
// needed.
mWakeupThreadFD = eventfd(0, EFD_NONBLOCK);
// seed the random number generator (used to generated timeline IDs)
srand48(static_cast<unsigned int>(systemTime()));
}
CommonTimeServer::~CommonTimeServer() {
shutdownThread();
// No need to grab the lock here. We are in the destructor; if the the user
// has a thread in any of the APIs while the destructor is being called,
// there is a threading problem a the application level we cannot reasonably
// do anything about.
cleanupSocket_l();
if (mWakeupThreadFD >= 0) {
close(mWakeupThreadFD);
mWakeupThreadFD = -1;
}
}
bool CommonTimeServer::startServices() {
// start the ICommonClock service
mICommonClock = CommonClockService::instantiate(*this);
if (mICommonClock == NULL)
return false;
// start the ICommonTimeConfig service
mICommonTimeConfig = CommonTimeConfigService::instantiate(*this);
if (mICommonTimeConfig == NULL)
return false;
return true;
}
bool CommonTimeServer::threadLoop() {
// Register our service interfaces.
if (!startServices())
return false;
// Hold the lock while we are in the main thread loop. It will release the
// lock when it blocks, and hold the lock at all other times.
mLock.lock();
runStateMachine_l();
mLock.unlock();
IPCThreadState::self()->stopProcess();
return false;
}
bool CommonTimeServer::runStateMachine_l() {
if (!mLocalClock.initCheck())
return false;
if (!mCommonClock.init(mLocalClock.getLocalFreq()))
return false;
// Enter the initial state.
becomeInitial("startup");
// run the state machine
while (!exitPending()) {
struct pollfd pfds[2];
Fix for bug 6691452 Hand merge from ics-aah > Fix for bug 6691452 : DO NOT MERGE > > As it so happens, there seem to be panels out there who disapprove of > sudden changes in their HDMI clock rate. In particular, Sony LCD > panels made from around 2010-2011 (including the Sony GTV panel) seem > to dislike this behavior. When exposed to a large jump in the clock > rate (say from -100pmm to +100ppm in about 30mSec), they seem to > panic, blank their audio and video, and then resync. The whole > panic process takes about 2 seconds. > > The HDMI spec says that its clock jitter requirements are defined by > their differential signalling eye diagram requirements relative to an > "Ideal Recovery Clock" (see section 4.2.3.1 of the HDMI 1.3a spec). > Basically, if you pass the eye diagram tests, you pass the clock > jitter requirements. We have determined in lab that even being > extremely aggressive in our VCXO rate changes does not come even close > to violating the HDMI eye diagrams. Its just this era of Sony panels > which seem to be upset by this behavior. > > One way or the other, experiments which the GTV devices have seemed to > indicate that a full range sweep of the VCXO done in 10mSec steps over > anything faster than 190mSec can cause trouble. Adding a healthy > degree of margin to this finding, the fix is to limit the rate of VCXO > control change such that it never goes at a rate faster than > FullRange/300mSec. > > Change flagged as do not merge due to the code structure changes to master. > This will need to be merged by hand. > > Signed-off-by: John Grossman <johngro@google.com> > Change-Id: Ibfd361fe1cc2cbd4909489e3317fb12e005c6a75 Change-Id: If62f791c826f1145262a6b546b1dc1f9776c37d8 Signed-off-by: John Grossman <johngro@google.com>
2012-06-26 12:50:28 -07:00
int rc, timeout;
int eventCnt = 0;
int64_t wakeupTime;
Fix for bug 6691452 Hand merge from ics-aah > Fix for bug 6691452 : DO NOT MERGE > > As it so happens, there seem to be panels out there who disapprove of > sudden changes in their HDMI clock rate. In particular, Sony LCD > panels made from around 2010-2011 (including the Sony GTV panel) seem > to dislike this behavior. When exposed to a large jump in the clock > rate (say from -100pmm to +100ppm in about 30mSec), they seem to > panic, blank their audio and video, and then resync. The whole > panic process takes about 2 seconds. > > The HDMI spec says that its clock jitter requirements are defined by > their differential signalling eye diagram requirements relative to an > "Ideal Recovery Clock" (see section 4.2.3.1 of the HDMI 1.3a spec). > Basically, if you pass the eye diagram tests, you pass the clock > jitter requirements. We have determined in lab that even being > extremely aggressive in our VCXO rate changes does not come even close > to violating the HDMI eye diagrams. Its just this era of Sony panels > which seem to be upset by this behavior. > > One way or the other, experiments which the GTV devices have seemed to > indicate that a full range sweep of the VCXO done in 10mSec steps over > anything faster than 190mSec can cause trouble. Adding a healthy > degree of margin to this finding, the fix is to limit the rate of VCXO > control change such that it never goes at a rate faster than > FullRange/300mSec. > > Change flagged as do not merge due to the code structure changes to master. > This will need to be merged by hand. > > Signed-off-by: John Grossman <johngro@google.com> > Change-Id: Ibfd361fe1cc2cbd4909489e3317fb12e005c6a75 Change-Id: If62f791c826f1145262a6b546b1dc1f9776c37d8 Signed-off-by: John Grossman <johngro@google.com>
2012-06-26 12:50:28 -07:00
uint32_t t1, t2;
bool needHandleTimeout = false;
// We are always interested in our wakeup FD.
pfds[eventCnt].fd = mWakeupThreadFD;
pfds[eventCnt].events = POLLIN;
pfds[eventCnt].revents = 0;
eventCnt++;
// If we have a valid socket, then we are interested in what it has to
// say as well.
if (mSocket >= 0) {
pfds[eventCnt].fd = mSocket;
pfds[eventCnt].events = POLLIN;
pfds[eventCnt].revents = 0;
eventCnt++;
}
Fix for bug 6691452 Hand merge from ics-aah > Fix for bug 6691452 : DO NOT MERGE > > As it so happens, there seem to be panels out there who disapprove of > sudden changes in their HDMI clock rate. In particular, Sony LCD > panels made from around 2010-2011 (including the Sony GTV panel) seem > to dislike this behavior. When exposed to a large jump in the clock > rate (say from -100pmm to +100ppm in about 30mSec), they seem to > panic, blank their audio and video, and then resync. The whole > panic process takes about 2 seconds. > > The HDMI spec says that its clock jitter requirements are defined by > their differential signalling eye diagram requirements relative to an > "Ideal Recovery Clock" (see section 4.2.3.1 of the HDMI 1.3a spec). > Basically, if you pass the eye diagram tests, you pass the clock > jitter requirements. We have determined in lab that even being > extremely aggressive in our VCXO rate changes does not come even close > to violating the HDMI eye diagrams. Its just this era of Sony panels > which seem to be upset by this behavior. > > One way or the other, experiments which the GTV devices have seemed to > indicate that a full range sweep of the VCXO done in 10mSec steps over > anything faster than 190mSec can cause trouble. Adding a healthy > degree of margin to this finding, the fix is to limit the rate of VCXO > control change such that it never goes at a rate faster than > FullRange/300mSec. > > Change flagged as do not merge due to the code structure changes to master. > This will need to be merged by hand. > > Signed-off-by: John Grossman <johngro@google.com> > Change-Id: Ibfd361fe1cc2cbd4909489e3317fb12e005c6a75 Change-Id: If62f791c826f1145262a6b546b1dc1f9776c37d8 Signed-off-by: John Grossman <johngro@google.com>
2012-06-26 12:50:28 -07:00
t1 = static_cast<uint32_t>(mCurTimeout.msecTillTimeout());
t2 = static_cast<uint32_t>(mClockRecovery.applyRateLimitedSlew());
timeout = static_cast<int>(t1 < t2 ? t1 : t2);
// Note, we were holding mLock when this function was called. We
// release it only while we are blocking and hold it at all other times.
mLock.unlock();
Fix for bug 6691452 Hand merge from ics-aah > Fix for bug 6691452 : DO NOT MERGE > > As it so happens, there seem to be panels out there who disapprove of > sudden changes in their HDMI clock rate. In particular, Sony LCD > panels made from around 2010-2011 (including the Sony GTV panel) seem > to dislike this behavior. When exposed to a large jump in the clock > rate (say from -100pmm to +100ppm in about 30mSec), they seem to > panic, blank their audio and video, and then resync. The whole > panic process takes about 2 seconds. > > The HDMI spec says that its clock jitter requirements are defined by > their differential signalling eye diagram requirements relative to an > "Ideal Recovery Clock" (see section 4.2.3.1 of the HDMI 1.3a spec). > Basically, if you pass the eye diagram tests, you pass the clock > jitter requirements. We have determined in lab that even being > extremely aggressive in our VCXO rate changes does not come even close > to violating the HDMI eye diagrams. Its just this era of Sony panels > which seem to be upset by this behavior. > > One way or the other, experiments which the GTV devices have seemed to > indicate that a full range sweep of the VCXO done in 10mSec steps over > anything faster than 190mSec can cause trouble. Adding a healthy > degree of margin to this finding, the fix is to limit the rate of VCXO > control change such that it never goes at a rate faster than > FullRange/300mSec. > > Change flagged as do not merge due to the code structure changes to master. > This will need to be merged by hand. > > Signed-off-by: John Grossman <johngro@google.com> > Change-Id: Ibfd361fe1cc2cbd4909489e3317fb12e005c6a75 Change-Id: If62f791c826f1145262a6b546b1dc1f9776c37d8 Signed-off-by: John Grossman <johngro@google.com>
2012-06-26 12:50:28 -07:00
rc = poll(pfds, eventCnt, timeout);
wakeupTime = mLocalClock.getLocalTime();
mLock.lock();
// Is it time to shutdown? If so, don't hesitate... just do it.
if (exitPending())
break;
// Did the poll fail? This should never happen and is fatal if it does.
if (rc < 0) {
ALOGE("%s:%d poll failed", __PRETTY_FUNCTION__, __LINE__);
return false;
}
Fix for bug 6691452 Hand merge from ics-aah > Fix for bug 6691452 : DO NOT MERGE > > As it so happens, there seem to be panels out there who disapprove of > sudden changes in their HDMI clock rate. In particular, Sony LCD > panels made from around 2010-2011 (including the Sony GTV panel) seem > to dislike this behavior. When exposed to a large jump in the clock > rate (say from -100pmm to +100ppm in about 30mSec), they seem to > panic, blank their audio and video, and then resync. The whole > panic process takes about 2 seconds. > > The HDMI spec says that its clock jitter requirements are defined by > their differential signalling eye diagram requirements relative to an > "Ideal Recovery Clock" (see section 4.2.3.1 of the HDMI 1.3a spec). > Basically, if you pass the eye diagram tests, you pass the clock > jitter requirements. We have determined in lab that even being > extremely aggressive in our VCXO rate changes does not come even close > to violating the HDMI eye diagrams. Its just this era of Sony panels > which seem to be upset by this behavior. > > One way or the other, experiments which the GTV devices have seemed to > indicate that a full range sweep of the VCXO done in 10mSec steps over > anything faster than 190mSec can cause trouble. Adding a healthy > degree of margin to this finding, the fix is to limit the rate of VCXO > control change such that it never goes at a rate faster than > FullRange/300mSec. > > Change flagged as do not merge due to the code structure changes to master. > This will need to be merged by hand. > > Signed-off-by: John Grossman <johngro@google.com> > Change-Id: Ibfd361fe1cc2cbd4909489e3317fb12e005c6a75 Change-Id: If62f791c826f1145262a6b546b1dc1f9776c37d8 Signed-off-by: John Grossman <johngro@google.com>
2012-06-26 12:50:28 -07:00
if (rc == 0) {
needHandleTimeout = !mCurTimeout.msecTillTimeout();
if (needHandleTimeout)
mCurTimeout.setTimeout(kInfiniteTimeout);
}
// Were we woken up on purpose? If so, clear the eventfd with a read.
if (pfds[0].revents)
clearPendingWakeupEvents_l();
// Is out bind address dirty? If so, clean up our socket (if any).
// Alternatively, do we have an active socket but should be auto
// disabled? If so, release the socket and enter the proper sync state.
bool droppedSocket = false;
if (mBindIfaceDirty || ((mSocket >= 0) && shouldAutoDisable())) {
cleanupSocket_l();
mBindIfaceDirty = false;
droppedSocket = true;
}
// Do we not have a socket but should have one? If so, try to set one
// up.
if ((mSocket < 0) && mBindIfaceValid && !shouldAutoDisable()) {
if (setupSocket_l()) {
// Success! We are now joining a new network (either coming
// from no network, or coming from a potentially different
// network). Force our priority to be lower so that we defer to
// any other masters which may already be on the network we are
// joining. Later, when we enter either the client or the
// master state, we will clear this flag and go back to our
// normal election priority.
setForceLowPriority(true);
switch (mState) {
// If we were in initial (whether we had a immediately
// before this network or not) we want to simply reset the
// system and start again. Forcing a transition from
// INITIAL to INITIAL should do the job.
case CommonClockService::STATE_INITIAL:
becomeInitial("bound interface");
break;
// If we were in the master state, then either we were the
// master in a no-network situation, or we were the master
// of a different network and have moved to a new interface.
Make common_time more deferential when coming out of networkless mode. Addresses issues seen in bug 6260139. This is a really tough bug to repro, but there is no doubt that it is happening occasionally on our super huge A@H subnet. I have collected data all weekend; the failure did not occur, but I got enough to have a theoretical sequence of events which could trigger this behavior. The sequence goes like this. 1) A network is running and happy with a timeline master M, maintaining timeline X. 2) Device B boots, but its network is taking a long time to come up. After 60 seconds of waiting for the network to come up, device B goes into networkless master mode and creates timeline Y. 3) Device B's network comes up. It immediately sends a master announcement saying that it is the current low-priority master of timeline Y (its low priority because it has never had any real clients) 4) Master M ignores B because B is low priority. 5) Device C boots and sends out a who is master request. It is a race between M and A to see who will respond first. In this case, A responds first. 6) C sends B a request which B receives. B now has its first client and is now high priority. In this scenario, B matches M in all aspects of the priority ranking function, including winning the tie breaker (larger MAC address when interpreted as a 48 bit integer) 7) M sends its master announcement; it is ignored by B since B now wins in the ranking function vs M. 8) Finally, B sends its next master announcement. M sees it, realizes that there is a higher priority master out there (looks like a bridged network scenario to M). M gives up master status along with timeline X. The clients of M become clients of B and move from timeline X to timeline Y (something which should only be needed during an actual network bridging event) This change has a few different things meant to severely minimize the chance that this can happen. First, and the most important change, is that networkless masters do not immediately announce themselves as masters on the network they are joining. Instead, they transition into Ronin to discover any pre-existing masters on the network. If there are no masters out there, the device will simply transition back to master and continue to maintain the timeline it had in networkless mode. In the scenario above, however, B should discover M and become its client, preserving the established timeline X. Second, any time a device experienced an interface reconfiguration (including coming out of networkless mode), it clears its high priority bit. This is a good thing. The bit used to get set again any time... 1) The device is master and receives a client request. 2) The device becomes a client of another master on the network. 3) The device becomes a master. Number 3 in this list is a mistake. The high priority bit should only be set for devices during master election which have been participating in a timeline which has been used by multiple devices. We know that this is the case when we are master and receive a request. We also know that this is the case when we hear from a master and decide to become its client. Simply becoming a master should not make us high priority. This behavior has been removed. Third, timeouts have been adjusted just for some extra "stickyness" when it comes to master status. Clients now say in the Ronin state for up to 10 seconds looking for a master sending up to 20 discovery requests, instead of only 3 seconds (sending 6 requests). The wait-for-election timeout has been adjusted up from 5 seconds to 12.5 seconds to track the longer election cycle as well. Also, while in steady-state, clients will now wait until 10 packets (10 seconds) have not been answered by its master before giving up and dropping into Ronin. Change-Id: I438b39f31265e34d6719d4adfa9e8b95a2afc188 Signed-off-by: John Grossman <johngro@google.com>
2012-04-09 11:26:16 -07:00
// In either case, immediately transition to Ronin at low
// priority. If there is no one in the network we just
// joined, we will become master soon enough. If there is,
// we want to be certain to defer master status to the
// existing timeline currently running on the network.
//
case CommonClockService::STATE_MASTER:
Make common_time more deferential when coming out of networkless mode. Addresses issues seen in bug 6260139. This is a really tough bug to repro, but there is no doubt that it is happening occasionally on our super huge A@H subnet. I have collected data all weekend; the failure did not occur, but I got enough to have a theoretical sequence of events which could trigger this behavior. The sequence goes like this. 1) A network is running and happy with a timeline master M, maintaining timeline X. 2) Device B boots, but its network is taking a long time to come up. After 60 seconds of waiting for the network to come up, device B goes into networkless master mode and creates timeline Y. 3) Device B's network comes up. It immediately sends a master announcement saying that it is the current low-priority master of timeline Y (its low priority because it has never had any real clients) 4) Master M ignores B because B is low priority. 5) Device C boots and sends out a who is master request. It is a race between M and A to see who will respond first. In this case, A responds first. 6) C sends B a request which B receives. B now has its first client and is now high priority. In this scenario, B matches M in all aspects of the priority ranking function, including winning the tie breaker (larger MAC address when interpreted as a 48 bit integer) 7) M sends its master announcement; it is ignored by B since B now wins in the ranking function vs M. 8) Finally, B sends its next master announcement. M sees it, realizes that there is a higher priority master out there (looks like a bridged network scenario to M). M gives up master status along with timeline X. The clients of M become clients of B and move from timeline X to timeline Y (something which should only be needed during an actual network bridging event) This change has a few different things meant to severely minimize the chance that this can happen. First, and the most important change, is that networkless masters do not immediately announce themselves as masters on the network they are joining. Instead, they transition into Ronin to discover any pre-existing masters on the network. If there are no masters out there, the device will simply transition back to master and continue to maintain the timeline it had in networkless mode. In the scenario above, however, B should discover M and become its client, preserving the established timeline X. Second, any time a device experienced an interface reconfiguration (including coming out of networkless mode), it clears its high priority bit. This is a good thing. The bit used to get set again any time... 1) The device is master and receives a client request. 2) The device becomes a client of another master on the network. 3) The device becomes a master. Number 3 in this list is a mistake. The high priority bit should only be set for devices during master election which have been participating in a timeline which has been used by multiple devices. We know that this is the case when we are master and receive a request. We also know that this is the case when we hear from a master and decide to become its client. Simply becoming a master should not make us high priority. This behavior has been removed. Third, timeouts have been adjusted just for some extra "stickyness" when it comes to master status. Clients now say in the Ronin state for up to 10 seconds looking for a master sending up to 20 discovery requests, instead of only 3 seconds (sending 6 requests). The wait-for-election timeout has been adjusted up from 5 seconds to 12.5 seconds to track the longer election cycle as well. Also, while in steady-state, clients will now wait until 10 packets (10 seconds) have not been answered by its master before giving up and dropping into Ronin. Change-Id: I438b39f31265e34d6719d4adfa9e8b95a2afc188 Signed-off-by: John Grossman <johngro@google.com>
2012-04-09 11:26:16 -07:00
becomeRonin("leaving networkless mode");
break;
// If we were in any other state (CLIENT, RONIN, or
// WAIT_FOR_ELECTION) then we must be moving from one
// network to another. We have lost our old master;
// transition to RONIN in an attempt to find a new master.
// If there are none out there, we will just assume
// responsibility for the timeline we used to be a client
// of.
default:
becomeRonin("bound interface");
break;
}
} else {
// That's odd... we failed to set up our socket. This could be
// due to some transient network change which will work itself
// out shortly; schedule a retry attempt in the near future.
mCurTimeout.setTimeout(kSetupRetryTimeoutMs);
}
// One way or the other, we don't have any data to process at this
// point (since we just tried to bulid a new socket). Loop back
// around and wait for the next thing to do.
continue;
} else if (droppedSocket) {
// We just lost our socket, and for whatever reason (either no
// config, or auto disable engaged) we are not supposed to rebuild
// one at this time. We are not going to rebuild our socket until
// something about our config/auto-disabled status changes, so we
// are basically in network-less mode. If we are already in either
// INITIAL or MASTER, just stay there until something changes. If
// we are in any other state (CLIENT, RONIN or WAIT_FOR_ELECTION),
// then transition to either INITIAL or MASTER depending on whether
// or not our timeline is valid.
mStateChangeLog.log(ANDROID_LOG_INFO, LOG_TAG,
"Entering networkless mode interface is %s, "
"shouldAutoDisable = %s",
mBindIfaceValid ? "valid" : "invalid",
shouldAutoDisable() ? "true" : "false");
if ((mState != ICommonClock::STATE_INITIAL) &&
(mState != ICommonClock::STATE_MASTER)) {
if (mTimelineID == ICommonClock::kInvalidTimelineID)
becomeInitial("network-less mode");
else
becomeMaster("network-less mode");
}
continue;
}
Fix for bug 6691452 Hand merge from ics-aah > Fix for bug 6691452 : DO NOT MERGE > > As it so happens, there seem to be panels out there who disapprove of > sudden changes in their HDMI clock rate. In particular, Sony LCD > panels made from around 2010-2011 (including the Sony GTV panel) seem > to dislike this behavior. When exposed to a large jump in the clock > rate (say from -100pmm to +100ppm in about 30mSec), they seem to > panic, blank their audio and video, and then resync. The whole > panic process takes about 2 seconds. > > The HDMI spec says that its clock jitter requirements are defined by > their differential signalling eye diagram requirements relative to an > "Ideal Recovery Clock" (see section 4.2.3.1 of the HDMI 1.3a spec). > Basically, if you pass the eye diagram tests, you pass the clock > jitter requirements. We have determined in lab that even being > extremely aggressive in our VCXO rate changes does not come even close > to violating the HDMI eye diagrams. Its just this era of Sony panels > which seem to be upset by this behavior. > > One way or the other, experiments which the GTV devices have seemed to > indicate that a full range sweep of the VCXO done in 10mSec steps over > anything faster than 190mSec can cause trouble. Adding a healthy > degree of margin to this finding, the fix is to limit the rate of VCXO > control change such that it never goes at a rate faster than > FullRange/300mSec. > > Change flagged as do not merge due to the code structure changes to master. > This will need to be merged by hand. > > Signed-off-by: John Grossman <johngro@google.com> > Change-Id: Ibfd361fe1cc2cbd4909489e3317fb12e005c6a75 Change-Id: If62f791c826f1145262a6b546b1dc1f9776c37d8 Signed-off-by: John Grossman <johngro@google.com>
2012-06-26 12:50:28 -07:00
// Time to handle the timeouts?
if (needHandleTimeout) {
if (!handleTimeout())
ALOGE("handleTimeout failed");
continue;
}
// Does our socket have data for us (assuming we still have one, we
// may have RXed a packet at the same time as a config change telling us
// to shut our socket down)? If so, process its data.
if ((mSocket >= 0) && (eventCnt > 1) && (pfds[1].revents)) {
mLastPacketRxLocalTime = wakeupTime;
if (!handlePacket())
ALOGE("handlePacket failed");
}
}
cleanupSocket_l();
return true;
}
void CommonTimeServer::clearPendingWakeupEvents_l() {
int64_t tmp;
read(mWakeupThreadFD, &tmp, sizeof(tmp));
}
void CommonTimeServer::wakeupThread_l() {
int64_t tmp = 1;
write(mWakeupThreadFD, &tmp, sizeof(tmp));
}
void CommonTimeServer::cleanupSocket_l() {
if (mSocket >= 0) {
close(mSocket);
mSocket = -1;
}
}
void CommonTimeServer::shutdownThread() {
// Flag the work thread for shutdown.
this->requestExit();
// Signal the thread in case its sleeping.
mLock.lock();
wakeupThread_l();
mLock.unlock();
// Wait for the thread to exit.
this->join();
}
bool CommonTimeServer::setupSocket_l() {
int rc;
bool ret_val = false;
struct sockaddr_in* ipv4_addr = NULL;
char masterElectionEPStr[64];
const int one = 1;
// This should never be needed, but if we happened to have an old socket
// lying around, be sure to not leak it before proceeding.
cleanupSocket_l();
// If we don't have a valid endpoint to bind to, then how did we get here in
// the first place? Regardless, we know that we are going to fail to bind,
// so don't even try.
if (!mBindIfaceValid)
return false;
sockaddrToString(mMasterElectionEP, true, masterElectionEPStr,
sizeof(masterElectionEPStr));
mStateChangeLog.log(ANDROID_LOG_INFO, LOG_TAG,
"Building socket :: bind = %s master election = %s",
mBindIface.string(), masterElectionEPStr);
// TODO: add proper support for IPv6. Right now, we block IPv6 addresses at
// the configuration interface level.
if (AF_INET != mMasterElectionEP.ss_family) {
mStateChangeLog.log(ANDROID_LOG_WARN, LOG_TAG,
"TODO: add proper IPv6 support");
goto bailout;
}
// open a UDP socket for the timeline serivce
mSocket = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
if (mSocket < 0) {
mStateChangeLog.log(ANDROID_LOG_ERROR, LOG_TAG,
"Failed to create socket (errno = %d)", errno);
goto bailout;
}
// Bind to the selected interface using Linux's spiffy SO_BINDTODEVICE.
struct ifreq ifr;
memset(&ifr, 0, sizeof(ifr));
snprintf(ifr.ifr_name, sizeof(ifr.ifr_name), "%s", mBindIface.string());
ifr.ifr_name[sizeof(ifr.ifr_name) - 1] = 0;
rc = setsockopt(mSocket, SOL_SOCKET, SO_BINDTODEVICE,
(void *)&ifr, sizeof(ifr));
if (rc) {
mStateChangeLog.log(ANDROID_LOG_ERROR, LOG_TAG,
"Failed to bind socket at to interface %s "
"(errno = %d)", ifr.ifr_name, errno);
goto bailout;
}
// Bind our socket to INADDR_ANY and the master election port. The
// interface binding we made using SO_BINDTODEVICE should limit us to
// traffic only on the interface we are interested in. We need to bind to
// INADDR_ANY and the specific master election port in order to be able to
// receive both unicast traffic and master election multicast traffic with
// just a single socket.
struct sockaddr_in bindAddr;
ipv4_addr = reinterpret_cast<struct sockaddr_in*>(&mMasterElectionEP);
memcpy(&bindAddr, ipv4_addr, sizeof(bindAddr));
bindAddr.sin_addr.s_addr = INADDR_ANY;
rc = bind(mSocket,
reinterpret_cast<const sockaddr *>(&bindAddr),
sizeof(bindAddr));
if (rc) {
mStateChangeLog.log(ANDROID_LOG_ERROR, LOG_TAG,
"Failed to bind socket to port %hu (errno = %d)",
ntohs(bindAddr.sin_port), errno);
goto bailout;
}
if (0xE0000000 == (ntohl(ipv4_addr->sin_addr.s_addr) & 0xF0000000)) {
// If our master election endpoint is a multicast address, be sure to join
// the multicast group.
struct ip_mreq mreq;
mreq.imr_multiaddr = ipv4_addr->sin_addr;
mreq.imr_interface.s_addr = htonl(INADDR_ANY);
rc = setsockopt(mSocket, IPPROTO_IP, IP_ADD_MEMBERSHIP,
&mreq, sizeof(mreq));
if (rc == -1) {
ALOGE("Failed to join multicast group at %s. (errno = %d)",
masterElectionEPStr, errno);
goto bailout;
}
// disable loopback of multicast packets
const int zero = 0;
rc = setsockopt(mSocket, IPPROTO_IP, IP_MULTICAST_LOOP,
&zero, sizeof(zero));
if (rc == -1) {
mStateChangeLog.log(ANDROID_LOG_ERROR, LOG_TAG,
"Failed to disable multicast loopback "
"(errno = %d)", errno);
goto bailout;
}
} else
if (ntohl(ipv4_addr->sin_addr.s_addr) == 0xFFFFFFFF) {
// If the master election address is the broadcast address, then enable
// the broadcast socket option
rc = setsockopt(mSocket, SOL_SOCKET, SO_BROADCAST, &one, sizeof(one));
if (rc == -1) {
mStateChangeLog.log(ANDROID_LOG_ERROR, LOG_TAG,
"Failed to enable broadcast (errno = %d)",
errno);
goto bailout;
}
} else {
// If the master election address is neither broadcast, nor multicast,
// then we are misconfigured. The config API layer should prevent this
// from ever happening.
goto bailout;
}
// Set the TTL of sent packets to 1. (Time protocol sync should never leave
// the local subnet)
rc = setsockopt(mSocket, IPPROTO_IP, IP_TTL, &one, sizeof(one));
if (rc == -1) {
mStateChangeLog.log(ANDROID_LOG_ERROR, LOG_TAG,
"Failed to set TTL to %d (errno = %d)", one, errno);
goto bailout;
}
// get the device's unique ID
if (!assignDeviceID())
goto bailout;
ret_val = true;
bailout:
if (!ret_val)
cleanupSocket_l();
return ret_val;
}
// generate a unique device ID that can be used for arbitration
bool CommonTimeServer::assignDeviceID() {
if (!mBindIfaceValid)
return false;
struct ifreq ifr;
memset(&ifr, 0, sizeof(ifr));
ifr.ifr_addr.sa_family = AF_INET;
strlcpy(ifr.ifr_name, mBindIface.string(), IFNAMSIZ);
int rc = ioctl(mSocket, SIOCGIFHWADDR, &ifr);
if (rc) {
ALOGE("%s:%d ioctl failed", __PRETTY_FUNCTION__, __LINE__);
return false;
}
if (ifr.ifr_addr.sa_family != ARPHRD_ETHER) {
ALOGE("%s:%d got non-Ethernet address", __PRETTY_FUNCTION__, __LINE__);
return false;
}
mDeviceID = 0;
for (int i = 0; i < ETH_ALEN; i++) {
mDeviceID = (mDeviceID << 8) | ifr.ifr_hwaddr.sa_data[i];
}
return true;
}
// generate a new timeline ID
void CommonTimeServer::assignTimelineID() {
do {
mTimelineID = (static_cast<uint64_t>(lrand48()) << 32)
| static_cast<uint64_t>(lrand48());
} while (mTimelineID == ICommonClock::kInvalidTimelineID);
}
// Select a preference between the device IDs of two potential masters.
// Returns true if the first ID wins, or false if the second ID wins.
bool CommonTimeServer::arbitrateMaster(
uint64_t deviceID1, uint8_t devicePrio1,
uint64_t deviceID2, uint8_t devicePrio2) {
return ((devicePrio1 > devicePrio2) ||
((devicePrio1 == devicePrio2) && (deviceID1 > deviceID2)));
}
static void hexDumpToString(const uint8_t* src, size_t src_len,
char* dst, size_t dst_len) {
size_t offset = 0;
size_t i;
for (i = 0; (i < src_len) && (offset < dst_len); ++i) {
int res;
if (0 == (i % 16)) {
res = snprintf(dst + offset, dst_len - offset, "\n%04x :", i);
if (res < 0)
break;
offset += res;
if (offset >= dst_len)
break;
}
res = snprintf(dst + offset, dst_len - offset, " %02x", src[i]);
if (res < 0)
break;
offset += res;
}
dst[dst_len - 1] = 0;
}
bool CommonTimeServer::handlePacket() {
uint8_t buf[256];
struct sockaddr_storage srcAddr;
socklen_t srcAddrLen = sizeof(srcAddr);
ssize_t recvBytes = recvfrom(
mSocket, buf, sizeof(buf), 0,
reinterpret_cast<const sockaddr *>(&srcAddr), &srcAddrLen);
if (recvBytes < 0) {
mBadPktLog.log(ANDROID_LOG_ERROR, LOG_TAG,
"recvfrom failed (res %d, errno %d)",
recvBytes, errno);
return false;
}
UniversalTimeServicePacket pkt;
if (pkt.deserializePacket(buf, recvBytes, mSyncGroupID) < 0) {
char hex[256];
char srcEPStr[64];
hexDumpToString(buf, static_cast<size_t>(recvBytes), hex, sizeof(hex));
sockaddrToString(srcAddr, true, srcEPStr, sizeof(srcEPStr));
mBadPktLog.log("Failed to parse %d byte packet from %s.%s",
recvBytes, srcEPStr, hex);
return false;
}
bool result;
switch (pkt.packetType) {
case TIME_PACKET_WHO_IS_MASTER_REQUEST:
result = handleWhoIsMasterRequest(&pkt.p.who_is_master_request,
srcAddr);
break;
case TIME_PACKET_WHO_IS_MASTER_RESPONSE:
result = handleWhoIsMasterResponse(&pkt.p.who_is_master_response,
srcAddr);
break;
case TIME_PACKET_SYNC_REQUEST:
result = handleSyncRequest(&pkt.p.sync_request, srcAddr);
break;
case TIME_PACKET_SYNC_RESPONSE:
result = handleSyncResponse(&pkt.p.sync_response, srcAddr);
break;
case TIME_PACKET_MASTER_ANNOUNCEMENT:
result = handleMasterAnnouncement(&pkt.p.master_announcement,
srcAddr);
break;
default: {
char srcEPStr[64];
sockaddrToString(srcAddr, true, srcEPStr, sizeof(srcEPStr));
mBadPktLog.log(ANDROID_LOG_WARN, LOG_TAG,
"unknown packet type (%d) from %s",
pkt.packetType, srcEPStr);
result = false;
} break;
}
return result;
}
bool CommonTimeServer::handleTimeout() {
// If we have no socket, then this must be a timeout to retry socket setup.
if (mSocket < 0)
return true;
switch (mState) {
case ICommonClock::STATE_INITIAL:
return handleTimeoutInitial();
case ICommonClock::STATE_CLIENT:
return handleTimeoutClient();
case ICommonClock::STATE_MASTER:
return handleTimeoutMaster();
case ICommonClock::STATE_RONIN:
return handleTimeoutRonin();
case ICommonClock::STATE_WAIT_FOR_ELECTION:
return handleTimeoutWaitForElection();
}
return false;
}
bool CommonTimeServer::handleTimeoutInitial() {
if (++mInitial_WhoIsMasterRequestTimeouts ==
kInitial_NumWhoIsMasterRetries) {
// none of our attempts to discover a master succeeded, so make
// this device the master
return becomeMaster("initial timeout");
} else {
// retry the WhoIsMaster request
return sendWhoIsMasterRequest();
}
}
bool CommonTimeServer::handleTimeoutClient() {
if (shouldPanicNotGettingGoodData())
return becomeInitial("timeout panic, no good data");
if (mClient_SyncRequestPending) {
mClient_SyncRequestPending = false;
if (++mClient_SyncRequestTimeouts < kClient_NumSyncRequestRetries) {
// a sync request has timed out, so retry
return sendSyncRequest();
} else {
// The master has failed to respond to a sync request for too many
// times in a row. Assume the master is dead and start electing
// a new master.
return becomeRonin("master not responding");
}
} else {
// initiate the next sync request
return sendSyncRequest();
}
}
bool CommonTimeServer::handleTimeoutMaster() {
// send another announcement from the master
return sendMasterAnnouncement();
}
bool CommonTimeServer::handleTimeoutRonin() {
if (++mRonin_WhoIsMasterRequestTimeouts == kRonin_NumWhoIsMasterRetries) {
// no other master is out there, so we won the election
return becomeMaster("no better masters detected");
} else {
return sendWhoIsMasterRequest();
}
}
bool CommonTimeServer::handleTimeoutWaitForElection() {
return becomeRonin("timeout waiting for election conclusion");
}
bool CommonTimeServer::handleWhoIsMasterRequest(
const WhoIsMasterRequestPacket* request,
const sockaddr_storage& srcAddr) {
char srcEPStr[64];
sockaddrToString(srcAddr, true, srcEPStr, sizeof(srcEPStr));
mElectionLog.log("RXed WhoIs master request while in state %s. "
"src %s reqTID %016llx ourTID %016llx",
stateToString(mState), srcEPStr,
request->timelineID, mTimelineID);
if (mState == ICommonClock::STATE_MASTER) {
// is this request related to this master's timeline?
if (request->timelineID != ICommonClock::kInvalidTimelineID &&
request->timelineID != mTimelineID)
return true;
WhoIsMasterResponsePacket pkt;
pkt.initHeader(mTimelineID, mSyncGroupID);
pkt.deviceID = mDeviceID;
pkt.devicePriority = effectivePriority();
mElectionLog.log("TXing WhoIs master resp to %s while in state %s. "
"ourTID %016llx ourGID %016llx ourDID %016llx "
"ourPrio %u",
srcEPStr, stateToString(mState),
mTimelineID, mSyncGroupID,
pkt.deviceID, pkt.devicePriority);
uint8_t buf[256];
ssize_t bufSz = pkt.serializePacket(buf, sizeof(buf));
if (bufSz < 0)
return false;
ssize_t sendBytes = sendto(
mSocket, buf, bufSz, 0,
reinterpret_cast<const sockaddr *>(&srcAddr),
sizeof(srcAddr));
if (sendBytes == -1) {
ALOGE("%s:%d sendto failed", __PRETTY_FUNCTION__, __LINE__);
return false;
}
} else if (mState == ICommonClock::STATE_RONIN) {
// if we hear a WhoIsMaster request from another device following
// the same timeline and that device wins arbitration, then we will stop
// trying to elect ourselves master and will instead wait for an
// announcement from the election winner
if (request->timelineID != mTimelineID)
return true;
if (arbitrateMaster(request->senderDeviceID,
request->senderDevicePriority,
mDeviceID,
effectivePriority()))
return becomeWaitForElection("would lose election");
return true;
} else if (mState == ICommonClock::STATE_INITIAL) {
// If a group of devices booted simultaneously (e.g. after a power
// outage) and all of them are in the initial state and there is no
// master, then each device may time out and declare itself master at
// the same time. To avoid this, listen for
// WhoIsMaster(InvalidTimeline) requests from peers. If we would lose
// arbitration against that peer, reset our timeout count so that the
// peer has a chance to become master before we time out.
if (request->timelineID == ICommonClock::kInvalidTimelineID &&
arbitrateMaster(request->senderDeviceID,
request->senderDevicePriority,
mDeviceID,
effectivePriority())) {
mInitial_WhoIsMasterRequestTimeouts = 0;
}
}
return true;
}
bool CommonTimeServer::handleWhoIsMasterResponse(
const WhoIsMasterResponsePacket* response,
const sockaddr_storage& srcAddr) {
char srcEPStr[64];
sockaddrToString(srcAddr, true, srcEPStr, sizeof(srcEPStr));
mElectionLog.log("RXed WhoIs master response while in state %s. "
"src %s respTID %016llx respDID %016llx respPrio %u "
"ourTID %016llx",
stateToString(mState), srcEPStr,
response->timelineID,
response->deviceID,
static_cast<uint32_t>(response->devicePriority),
mTimelineID);
if (mState == ICommonClock::STATE_INITIAL || mState == ICommonClock::STATE_RONIN) {
return becomeClient(srcAddr,
response->deviceID,
response->devicePriority,
response->timelineID,
"heard whois response");
} else if (mState == ICommonClock::STATE_CLIENT) {
// if we get multiple responses because there are multiple devices
// who believe that they are master, then follow the master that
// wins arbitration
if (arbitrateMaster(response->deviceID,
response->devicePriority,
mClient_MasterDeviceID,
mClient_MasterDevicePriority)) {
return becomeClient(srcAddr,
response->deviceID,
response->devicePriority,
response->timelineID,
"heard whois response");
}
}
return true;
}
bool CommonTimeServer::handleSyncRequest(const SyncRequestPacket* request,
const sockaddr_storage& srcAddr) {
SyncResponsePacket pkt;
pkt.initHeader(mTimelineID, mSyncGroupID);
if ((mState == ICommonClock::STATE_MASTER) &&
(mTimelineID == request->timelineID)) {
int64_t rxLocalTime = mLastPacketRxLocalTime;
int64_t rxCommonTime;
// If we are master on an actual network and have actual clients, then
// we are no longer low priority.
setForceLowPriority(false);
if (OK != mCommonClock.localToCommon(rxLocalTime, &rxCommonTime)) {
return false;
}
int64_t txLocalTime = mLocalClock.getLocalTime();;
int64_t txCommonTime;
if (OK != mCommonClock.localToCommon(txLocalTime, &txCommonTime)) {
return false;
}
pkt.nak = 0;
pkt.clientTxLocalTime = request->clientTxLocalTime;
pkt.masterRxCommonTime = rxCommonTime;
pkt.masterTxCommonTime = txCommonTime;
} else {
pkt.nak = 1;
pkt.clientTxLocalTime = 0;
pkt.masterRxCommonTime = 0;
pkt.masterTxCommonTime = 0;
}
uint8_t buf[256];
ssize_t bufSz = pkt.serializePacket(buf, sizeof(buf));
if (bufSz < 0)
return false;
ssize_t sendBytes = sendto(
mSocket, &buf, bufSz, 0,
reinterpret_cast<const sockaddr *>(&srcAddr),
sizeof(srcAddr));
if (sendBytes == -1) {
ALOGE("%s:%d sendto failed", __PRETTY_FUNCTION__, __LINE__);
return false;
}
return true;
}
bool CommonTimeServer::handleSyncResponse(
const SyncResponsePacket* response,
const sockaddr_storage& srcAddr) {
if (mState != ICommonClock::STATE_CLIENT)
return true;
assert(mMasterEPValid);
if (!sockaddrMatch(srcAddr, mMasterEP, true)) {
char srcEP[64], expectedEP[64];
sockaddrToString(srcAddr, true, srcEP, sizeof(srcEP));
sockaddrToString(mMasterEP, true, expectedEP, sizeof(expectedEP));
ALOGI("Dropping sync response from unexpected address."
" Expected %s Got %s", expectedEP, srcEP);
return true;
}
if (response->nak) {
// if our master is no longer accepting requests, then we need to find
// a new master
return becomeRonin("master NAK'ed");
}
mClient_SyncRequestPending = 0;
mClient_SyncRequestTimeouts = 0;
mClient_PacketRTTLog.logRX(response->clientTxLocalTime,
mLastPacketRxLocalTime);
bool result;
if (!(mClient_SyncRespsRXedFromCurMaster++)) {
// the first request/response exchange between a client and a master
// may take unusually long due to ARP, so discard it.
result = true;
} else {
int64_t clientTxLocalTime = response->clientTxLocalTime;
int64_t clientRxLocalTime = mLastPacketRxLocalTime;
int64_t masterTxCommonTime = response->masterTxCommonTime;
int64_t masterRxCommonTime = response->masterRxCommonTime;
int64_t rtt = (clientRxLocalTime - clientTxLocalTime);
int64_t avgLocal = (clientTxLocalTime + clientRxLocalTime) >> 1;
int64_t avgCommon = (masterTxCommonTime + masterRxCommonTime) >> 1;
// if the RTT of the packet is significantly larger than the panic
// threshold, we should simply discard it. Its better to do nothing
// than to take cues from a packet like that.
int rttCommon = mCommonClock.localDurationToCommonDuration(rtt);
if (rttCommon > (static_cast<int64_t>(mPanicThresholdUsec) *
kRTTDiscardPanicThreshMultiplier)) {
ALOGV("Dropping sync response with RTT of %lld uSec", rttCommon);
mClient_ExpiredSyncRespsRXedFromCurMaster++;
if (shouldPanicNotGettingGoodData())
return becomeInitial("RX panic, no good data");
} else {
result = mClockRecovery.pushDisciplineEvent(avgLocal, avgCommon, rttCommon);
mClient_LastGoodSyncRX = clientRxLocalTime;
if (result) {
// indicate to listeners that we've synced to the common timeline
notifyClockSync();
} else {
ALOGE("Panic! Observed clock sync error is too high to tolerate,"
" resetting state machine and starting over.");
notifyClockSyncLoss();
return becomeInitial("panic");
}
}
}
mCurTimeout.setTimeout(mSyncRequestIntervalMs);
return result;
}
bool CommonTimeServer::handleMasterAnnouncement(
const MasterAnnouncementPacket* packet,
const sockaddr_storage& srcAddr) {
uint64_t newDeviceID = packet->deviceID;
uint8_t newDevicePrio = packet->devicePriority;
uint64_t newTimelineID = packet->timelineID;
char srcEPStr[64];
sockaddrToString(srcAddr, true, srcEPStr, sizeof(srcEPStr));
mElectionLog.log("RXed master announcement while in state %s. "
"src %s srcDevID %lld srcPrio %u srcTID %016llx",
stateToString(mState), srcEPStr,
newDeviceID, static_cast<uint32_t>(newDevicePrio),
newTimelineID);
if (mState == ICommonClock::STATE_INITIAL ||
mState == ICommonClock::STATE_RONIN ||
mState == ICommonClock::STATE_WAIT_FOR_ELECTION) {
// if we aren't currently following a master, then start following
// this new master
return becomeClient(srcAddr,
newDeviceID,
newDevicePrio,
newTimelineID,
"heard master announcement");
} else if (mState == ICommonClock::STATE_CLIENT) {
// if the new master wins arbitration against our current master,
// then become a client of the new master
if (arbitrateMaster(newDeviceID,
newDevicePrio,
mClient_MasterDeviceID,
mClient_MasterDevicePriority))
return becomeClient(srcAddr,
newDeviceID,
newDevicePrio,
newTimelineID,
"heard master announcement");
} else if (mState == ICommonClock::STATE_MASTER) {
// two masters are competing - if the new one wins arbitration, then
// cease acting as master
if (arbitrateMaster(newDeviceID, newDevicePrio,
mDeviceID, effectivePriority()))
return becomeClient(srcAddr, newDeviceID,
newDevicePrio, newTimelineID,
"heard master announcement");
}
return true;
}
bool CommonTimeServer::sendWhoIsMasterRequest() {
assert(mState == ICommonClock::STATE_INITIAL || mState == ICommonClock::STATE_RONIN);
// If we have no socket, then we must be in the unconfigured initial state.
// Don't report any errors, just don't try to send the initial who-is-master
// query. Eventually, our network will either become configured, or we will
// be forced into network-less master mode by higher level code.
if (mSocket < 0) {
assert(mState == ICommonClock::STATE_INITIAL);
return true;
}
bool ret = false;
WhoIsMasterRequestPacket pkt;
pkt.initHeader(mSyncGroupID);
pkt.senderDeviceID = mDeviceID;
pkt.senderDevicePriority = effectivePriority();
uint8_t buf[256];
ssize_t bufSz = pkt.serializePacket(buf, sizeof(buf));
if (bufSz >= 0) {
char dstEPStr[64];
sockaddrToString(mMasterElectionEP, true, dstEPStr, sizeof(dstEPStr));
mElectionLog.log("TXing WhoIs master request to %s while in state %s. "
"ourTID %016llx ourGID %016llx ourDID %016llx "
"ourPrio %u",
dstEPStr, stateToString(mState),
mTimelineID, mSyncGroupID,
pkt.senderDeviceID, pkt.senderDevicePriority);
ssize_t sendBytes = sendto(
mSocket, buf, bufSz, 0,
reinterpret_cast<const sockaddr *>(&mMasterElectionEP),
sizeof(mMasterElectionEP));
if (sendBytes < 0)
ALOGE("WhoIsMaster sendto failed (errno %d)", errno);
ret = true;
}
if (mState == ICommonClock::STATE_INITIAL) {
mCurTimeout.setTimeout(kInitial_WhoIsMasterTimeoutMs);
} else {
mCurTimeout.setTimeout(kRonin_WhoIsMasterTimeoutMs);
}
return ret;
}
bool CommonTimeServer::sendSyncRequest() {
// If we are sending sync requests, then we must be in the client state and
// we must have a socket (when we have no network, we are only supposed to
// be in INITIAL or MASTER)
assert(mState == ICommonClock::STATE_CLIENT);
assert(mSocket >= 0);
bool ret = false;
SyncRequestPacket pkt;
pkt.initHeader(mTimelineID, mSyncGroupID);
pkt.clientTxLocalTime = mLocalClock.getLocalTime();
if (!mClient_FirstSyncTX)
mClient_FirstSyncTX = pkt.clientTxLocalTime;
mClient_PacketRTTLog.logTX(pkt.clientTxLocalTime);
uint8_t buf[256];
ssize_t bufSz = pkt.serializePacket(buf, sizeof(buf));
if (bufSz >= 0) {
ssize_t sendBytes = sendto(
mSocket, buf, bufSz, 0,
reinterpret_cast<const sockaddr *>(&mMasterEP),
sizeof(mMasterEP));
if (sendBytes < 0)
ALOGE("SyncRequest sendto failed (errno %d)", errno);
ret = true;
}
mClient_SyncsSentToCurMaster++;
mCurTimeout.setTimeout(mSyncRequestIntervalMs);
mClient_SyncRequestPending = true;
return ret;
}
bool CommonTimeServer::sendMasterAnnouncement() {
bool ret = false;
assert(mState == ICommonClock::STATE_MASTER);
// If we are being asked to send a master announcement, but we have no
// socket, we must be in network-less master mode. Don't bother to send the
// announcement, and don't bother to schedule a timeout. When the network
// comes up, the work thread will get poked and start the process of
// figuring out who the current master should be.
if (mSocket < 0) {
mCurTimeout.setTimeout(kInfiniteTimeout);
return true;
}
MasterAnnouncementPacket pkt;
pkt.initHeader(mTimelineID, mSyncGroupID);
pkt.deviceID = mDeviceID;
pkt.devicePriority = effectivePriority();
uint8_t buf[256];
ssize_t bufSz = pkt.serializePacket(buf, sizeof(buf));
if (bufSz >= 0) {
char dstEPStr[64];
sockaddrToString(mMasterElectionEP, true, dstEPStr, sizeof(dstEPStr));
mElectionLog.log("TXing Master announcement to %s while in state %s. "
"ourTID %016llx ourGID %016llx ourDID %016llx "
"ourPrio %u",
dstEPStr, stateToString(mState),
mTimelineID, mSyncGroupID,
pkt.deviceID, pkt.devicePriority);
ssize_t sendBytes = sendto(
mSocket, buf, bufSz, 0,
reinterpret_cast<const sockaddr *>(&mMasterElectionEP),
sizeof(mMasterElectionEP));
if (sendBytes < 0)
ALOGE("MasterAnnouncement sendto failed (errno %d)", errno);
ret = true;
}
mCurTimeout.setTimeout(mMasterAnnounceIntervalMs);
return ret;
}
bool CommonTimeServer::becomeClient(const sockaddr_storage& masterEP,
uint64_t masterDeviceID,
uint8_t masterDevicePriority,
uint64_t timelineID,
const char* cause) {
char newEPStr[64], oldEPStr[64];
sockaddrToString(masterEP, true, newEPStr, sizeof(newEPStr));
sockaddrToString(mMasterEP, mMasterEPValid, oldEPStr, sizeof(oldEPStr));
mStateChangeLog.log(ANDROID_LOG_INFO, LOG_TAG,
"%s --> CLIENT (%s) :%s"
" OldMaster: %02x-%014llx::%016llx::%s"
" NewMaster: %02x-%014llx::%016llx::%s",
stateToString(mState), cause,
(mTimelineID != timelineID) ? " (new timeline)" : "",
mClient_MasterDevicePriority, mClient_MasterDeviceID,
mTimelineID, oldEPStr,
masterDevicePriority, masterDeviceID,
timelineID, newEPStr);
if (mTimelineID != timelineID) {
// start following a new timeline
mTimelineID = timelineID;
mClockRecovery.reset(true, true);
notifyClockSyncLoss();
} else {
// start following a new master on the existing timeline
mClockRecovery.reset(false, true);
}
mMasterEP = masterEP;
mMasterEPValid = true;
Make common_time more deferential when coming out of networkless mode. Addresses issues seen in bug 6260139. This is a really tough bug to repro, but there is no doubt that it is happening occasionally on our super huge A@H subnet. I have collected data all weekend; the failure did not occur, but I got enough to have a theoretical sequence of events which could trigger this behavior. The sequence goes like this. 1) A network is running and happy with a timeline master M, maintaining timeline X. 2) Device B boots, but its network is taking a long time to come up. After 60 seconds of waiting for the network to come up, device B goes into networkless master mode and creates timeline Y. 3) Device B's network comes up. It immediately sends a master announcement saying that it is the current low-priority master of timeline Y (its low priority because it has never had any real clients) 4) Master M ignores B because B is low priority. 5) Device C boots and sends out a who is master request. It is a race between M and A to see who will respond first. In this case, A responds first. 6) C sends B a request which B receives. B now has its first client and is now high priority. In this scenario, B matches M in all aspects of the priority ranking function, including winning the tie breaker (larger MAC address when interpreted as a 48 bit integer) 7) M sends its master announcement; it is ignored by B since B now wins in the ranking function vs M. 8) Finally, B sends its next master announcement. M sees it, realizes that there is a higher priority master out there (looks like a bridged network scenario to M). M gives up master status along with timeline X. The clients of M become clients of B and move from timeline X to timeline Y (something which should only be needed during an actual network bridging event) This change has a few different things meant to severely minimize the chance that this can happen. First, and the most important change, is that networkless masters do not immediately announce themselves as masters on the network they are joining. Instead, they transition into Ronin to discover any pre-existing masters on the network. If there are no masters out there, the device will simply transition back to master and continue to maintain the timeline it had in networkless mode. In the scenario above, however, B should discover M and become its client, preserving the established timeline X. Second, any time a device experienced an interface reconfiguration (including coming out of networkless mode), it clears its high priority bit. This is a good thing. The bit used to get set again any time... 1) The device is master and receives a client request. 2) The device becomes a client of another master on the network. 3) The device becomes a master. Number 3 in this list is a mistake. The high priority bit should only be set for devices during master election which have been participating in a timeline which has been used by multiple devices. We know that this is the case when we are master and receive a request. We also know that this is the case when we hear from a master and decide to become its client. Simply becoming a master should not make us high priority. This behavior has been removed. Third, timeouts have been adjusted just for some extra "stickyness" when it comes to master status. Clients now say in the Ronin state for up to 10 seconds looking for a master sending up to 20 discovery requests, instead of only 3 seconds (sending 6 requests). The wait-for-election timeout has been adjusted up from 5 seconds to 12.5 seconds to track the longer election cycle as well. Also, while in steady-state, clients will now wait until 10 packets (10 seconds) have not been answered by its master before giving up and dropping into Ronin. Change-Id: I438b39f31265e34d6719d4adfa9e8b95a2afc188 Signed-off-by: John Grossman <johngro@google.com>
2012-04-09 11:26:16 -07:00
// If we are on a real network as a client of a real master, then we should
// no longer force low priority. If our master disappears, we should have
// the high priority bit set during the election to replace the master
// because this group was a real group and not a singleton created in
// networkless mode.
setForceLowPriority(false);
mClient_MasterDeviceID = masterDeviceID;
mClient_MasterDevicePriority = masterDevicePriority;
resetSyncStats();
setState(ICommonClock::STATE_CLIENT);
// add some jitter to when the various clients send their requests
// in order to reduce the likelihood that a group of clients overload
// the master after receiving a master announcement
usleep((lrand48() % 100) * 1000);
return sendSyncRequest();
}
bool CommonTimeServer::becomeMaster(const char* cause) {
uint64_t oldTimelineID = mTimelineID;
if (mTimelineID == ICommonClock::kInvalidTimelineID) {
// this device has not been following any existing timeline,
// so it will create a new timeline and declare itself master
assert(!mCommonClock.isValid());
// set the common time basis
mCommonClock.setBasis(mLocalClock.getLocalTime(), 0);
// assign an arbitrary timeline iD
assignTimelineID();
// notify listeners that we've created a common timeline
notifyClockSync();
}
mStateChangeLog.log(ANDROID_LOG_INFO, LOG_TAG,
"%s --> MASTER (%s) : %s timeline %016llx",
stateToString(mState), cause,
(oldTimelineID == mTimelineID) ? "taking ownership of"
: "creating new",
mTimelineID);
memset(&mMasterEP, 0, sizeof(mMasterEP));
mMasterEPValid = false;
mClient_MasterDevicePriority = effectivePriority();
mClient_MasterDeviceID = mDeviceID;
mClockRecovery.reset(false, true);
resetSyncStats();
setState(ICommonClock::STATE_MASTER);
return sendMasterAnnouncement();
}
bool CommonTimeServer::becomeRonin(const char* cause) {
// If we were the client of a given timeline, but had never received even a
// single time sync packet, then we transition back to Initial instead of
// Ronin. If we transition to Ronin and end up becoming the new Master, we
// will be unable to service requests for other clients because we never
// actually knew what time it was. By going to initial, we ensure that
// other clients who know what time it is, but would lose master arbitration
// in the Ronin case, will step up and become the proper new master of the
// old timeline.
char oldEPStr[64];
sockaddrToString(mMasterEP, mMasterEPValid, oldEPStr, sizeof(oldEPStr));
memset(&mMasterEP, 0, sizeof(mMasterEP));
mMasterEPValid = false;
if (mCommonClock.isValid()) {
mStateChangeLog.log(ANDROID_LOG_INFO, LOG_TAG,
"%s --> RONIN (%s) : lost track of previously valid timeline "
"%02x-%014llx::%016llx::%s (%d TXed %d RXed %d RXExpired)",
stateToString(mState), cause,
mClient_MasterDevicePriority, mClient_MasterDeviceID,
mTimelineID, oldEPStr,
mClient_SyncsSentToCurMaster,
mClient_SyncRespsRXedFromCurMaster,
mClient_ExpiredSyncRespsRXedFromCurMaster);
mRonin_WhoIsMasterRequestTimeouts = 0;
setState(ICommonClock::STATE_RONIN);
return sendWhoIsMasterRequest();
} else {
mStateChangeLog.log(ANDROID_LOG_INFO, LOG_TAG,
"%s --> INITIAL (%s) : never synced timeline "
"%02x-%014llx::%016llx::%s (%d TXed %d RXed %d RXExpired)",
stateToString(mState), cause,
mClient_MasterDevicePriority, mClient_MasterDeviceID,
mTimelineID, oldEPStr,
mClient_SyncsSentToCurMaster,
mClient_SyncRespsRXedFromCurMaster,
mClient_ExpiredSyncRespsRXedFromCurMaster);
return becomeInitial("ronin, no timeline");
}
}
bool CommonTimeServer::becomeWaitForElection(const char* cause) {
mStateChangeLog.log(ANDROID_LOG_INFO, LOG_TAG,
"%s --> WAIT_FOR_ELECTION (%s) : dropping out of election,"
" waiting %d mSec for completion.",
stateToString(mState), cause, kWaitForElection_TimeoutMs);
setState(ICommonClock::STATE_WAIT_FOR_ELECTION);
mCurTimeout.setTimeout(kWaitForElection_TimeoutMs);
return true;
}
bool CommonTimeServer::becomeInitial(const char* cause) {
mStateChangeLog.log(ANDROID_LOG_INFO, LOG_TAG,
"Entering INITIAL (%s), total reset.",
cause);
setState(ICommonClock::STATE_INITIAL);
// reset clock recovery
mClockRecovery.reset(true, true);
// reset internal state bookkeeping.
mCurTimeout.setTimeout(kInfiniteTimeout);
memset(&mMasterEP, 0, sizeof(mMasterEP));
mMasterEPValid = false;
mLastPacketRxLocalTime = 0;
mTimelineID = ICommonClock::kInvalidTimelineID;
mClockSynced = false;
mInitial_WhoIsMasterRequestTimeouts = 0;
mClient_MasterDeviceID = 0;
mClient_MasterDevicePriority = 0;
mRonin_WhoIsMasterRequestTimeouts = 0;
resetSyncStats();
// send the first request to discover the master
return sendWhoIsMasterRequest();
}
void CommonTimeServer::notifyClockSync() {
if (!mClockSynced) {
mClockSynced = true;
mICommonClock->notifyOnTimelineChanged(mTimelineID);
}
}
void CommonTimeServer::notifyClockSyncLoss() {
if (mClockSynced) {
mClockSynced = false;
mICommonClock->notifyOnTimelineChanged(
ICommonClock::kInvalidTimelineID);
}
}
void CommonTimeServer::setState(ICommonClock::State s) {
mState = s;
}
const char* CommonTimeServer::stateToString(ICommonClock::State s) {
switch(s) {
case ICommonClock::STATE_INITIAL:
return "INITIAL";
case ICommonClock::STATE_CLIENT:
return "CLIENT";
case ICommonClock::STATE_MASTER:
return "MASTER";
case ICommonClock::STATE_RONIN:
return "RONIN";
case ICommonClock::STATE_WAIT_FOR_ELECTION:
return "WAIT_FOR_ELECTION";
default:
return "unknown";
}
}
void CommonTimeServer::sockaddrToString(const sockaddr_storage& addr,
bool addrValid,
char* buf, size_t bufLen) {
if (!bufLen || !buf)
return;
if (addrValid) {
switch (addr.ss_family) {
case AF_INET: {
const struct sockaddr_in* sa =
reinterpret_cast<const struct sockaddr_in*>(&addr);
unsigned long a = ntohl(sa->sin_addr.s_addr);
uint16_t p = ntohs(sa->sin_port);
snprintf(buf, bufLen, "%lu.%lu.%lu.%lu:%hu",
((a >> 24) & 0xFF), ((a >> 16) & 0xFF),
((a >> 8) & 0xFF), (a & 0xFF), p);
} break;
case AF_INET6: {
const struct sockaddr_in6* sa =
reinterpret_cast<const struct sockaddr_in6*>(&addr);
const uint8_t* a = sa->sin6_addr.s6_addr;
uint16_t p = ntohs(sa->sin6_port);
snprintf(buf, bufLen,
"%02X%02X:%02X%02X:%02X%02X:%02X%02X:"
"%02X%02X:%02X%02X:%02X%02X:%02X%02X port %hd",
a[0], a[1], a[ 2], a[ 3], a[ 4], a[ 5], a[ 6], a[ 7],
a[8], a[9], a[10], a[11], a[12], a[13], a[14], a[15],
p);
} break;
default:
snprintf(buf, bufLen,
"<unknown sockaddr family %d>", addr.ss_family);
break;
}
} else {
snprintf(buf, bufLen, "<none>");
}
buf[bufLen - 1] = 0;
}
bool CommonTimeServer::sockaddrMatch(const sockaddr_storage& a1,
const sockaddr_storage& a2,
bool matchAddressOnly) {
if (a1.ss_family != a2.ss_family)
return false;
switch (a1.ss_family) {
case AF_INET: {
const struct sockaddr_in* sa1 =
reinterpret_cast<const struct sockaddr_in*>(&a1);
const struct sockaddr_in* sa2 =
reinterpret_cast<const struct sockaddr_in*>(&a2);
if (sa1->sin_addr.s_addr != sa2->sin_addr.s_addr)
return false;
return (matchAddressOnly || (sa1->sin_port == sa2->sin_port));
} break;
case AF_INET6: {
const struct sockaddr_in6* sa1 =
reinterpret_cast<const struct sockaddr_in6*>(&a1);
const struct sockaddr_in6* sa2 =
reinterpret_cast<const struct sockaddr_in6*>(&a2);
if (memcmp(&sa1->sin6_addr, &sa2->sin6_addr, sizeof(sa2->sin6_addr)))
return false;
return (matchAddressOnly || (sa1->sin6_port == sa2->sin6_port));
} break;
// Huh? We don't deal in non-IPv[46] addresses. Not sure how we got
// here, but we don't know how to comapre these addresses and simply
// default to a no-match decision.
default: return false;
}
}
bool CommonTimeServer::shouldPanicNotGettingGoodData() {
if (mClient_FirstSyncTX) {
int64_t now = mLocalClock.getLocalTime();
int64_t delta = now - (mClient_LastGoodSyncRX
? mClient_LastGoodSyncRX
: mClient_FirstSyncTX);
int64_t deltaUsec = mCommonClock.localDurationToCommonDuration(delta);
if (deltaUsec >= kNoGoodDataPanicThresholdUsec)
return true;
}
return false;
}
void CommonTimeServer::PacketRTTLog::logTX(int64_t txTime) {
txTimes[wrPtr] = txTime;
rxTimes[wrPtr] = 0;
wrPtr = (wrPtr + 1) % RTT_LOG_SIZE;
if (!wrPtr)
logFull = true;
}
void CommonTimeServer::PacketRTTLog::logRX(int64_t txTime, int64_t rxTime) {
if (!logFull && !wrPtr)
return;
uint32_t i = logFull ? wrPtr : 0;
do {
if (txTimes[i] == txTime) {
rxTimes[i] = rxTime;
break;
}
i = (i + 1) % RTT_LOG_SIZE;
} while (i != wrPtr);
}
} // namespace android