After fixing a problem where mutexes were held when destructing

them, Boost 1.54.0 is no longer a problem.

CollectAgent/simplemqttserverthread.cpp

@@ -10,12 +10,12 @@
 using namespace std;
 using namespace boost::system;
 
-static boost::mutex threadMtx;
-
 #ifdef SimpleMQTTVerbose
 static boost::mutex coutMtx;
 #endif
 
+static boost::mutex launchMtx;
+
 SimpleMQTTServerThread::SimpleMQTTServerThread()
 {
   /*
@@ -25,11 +25,21 @@ SimpleMQTTServerThread::SimpleMQTTServerThread()
    *
    * Special care must be taken not to call the run function
    * before the child has gone through its constructor. Thus,
-   * we lock a mutex that will cause the 'launch' function to
-   * wait until the mutex is released by the child's
-   * constructor.
+   * we lock a mutex (launchMtx) that will cause the 'launch'
+   * function to  wait until the mutex is released by the
+   * child's constructor.
+   *
+   * A second mutex (cleanupMtx) is held to allow others to
+   * wait for a thread to return from its child-specific
+   * 'run' function. This is necessary for cases where the
+   * child class has local objects that can only be destroyed
+   * after the 'run' function has finished (to avoid access
+   * to these objects in the 'run'function after destroying
+   * them). In such cases, the child class destructor can try
+   * to obtain the mutex in the destructor.
    */
-  threadMtx.lock();
+  launchMtx.lock();
+  cleanupMtx.lock();
 
   terminate = false;
   if (pthread_create(&t, NULL, launch, this) != 0) {
@@ -76,14 +86,19 @@ void* SimpleMQTTServerThread::launch(void *selfPtr)
    * before the constructor of the child class has finished.
    * The lock is released at the end of the child's constructor.
    */
-  threadMtx.lock();
-  threadMtx.unlock();
+  launchMtx.lock();
+  launchMtx.unlock();
 
   /*
    * Call the child's run function...
    */
   ((SimpleMQTTServerThread*)selfPtr)->run();
 
+  /*
+   * The thread will terminate. Unlock the cleanupMtx...
+   */
+  ((SimpleMQTTServerThread*)selfPtr)->cleanupMtx.unlock();
+
   return NULL;
 }
 
@@ -210,7 +225,7 @@ SimpleMQTTServerAcceptThread::SimpleMQTTServerAcceptThread(int listenSock, Simpl
    * Release the lock to indicate that the constructor has
    * finished. This causes the launcher to call the run function.
    */
-  threadMtx.unlock();
+  launchMtx.unlock();
 }
 
 SimpleMQTTServerAcceptThread::~SimpleMQTTServerAcceptThread()
@@ -423,10 +438,21 @@ SimpleMQTTServerMessageThread::SimpleMQTTServerMessageThread(SimpleMQTTMessageCa
    * Release the lock to indicate that the constructor has
    * finished. This causes the launcher to call the run function.
    */
-  threadMtx.unlock();
+  launchMtx.unlock();
 }
 
 SimpleMQTTServerMessageThread::~SimpleMQTTServerMessageThread()
 {
+  /*
+   * When this destructor is called externally, we are likely
+   * holding the fdsMtx mutex and waiting in poll().
+   * Terminate the thread from here and acquire cleanupMtx to
+   * ensure that the thread's 'run' function has terminated in
+   * order to avoid fdsMtx to be accessed after destruction.
+   */
+  terminate = true;
+
+  cleanupMtx.lock();
+  cleanupMtx.unlock();
 }
 

CollectAgent/simplemqttserverthread.h

@@ -14,6 +14,8 @@ protected:
   pthread_t t;
   bool terminate;
 
+  boost::mutex cleanupMtx;
+
   static void* launch(void* thisPtr);
   static void yield();
   virtual void run() = 0;

Makefile

@@ -5,7 +5,7 @@ DISTFILES = thrift-0.9.0.tar.gz;https://dist.apache.org/repos/dist/release/thrif
 		apache-cassandra-1.2.6.tar.gz;http://mirror.netcologne.de/apache.org/cassandra/1.2.6/apache-cassandra-1.2.6-bin.tar.gz \
 		mosquitto-1.1.3.tar.gz;http://mosquitto.org/files/source/mosquitto-1.1.3.tar.gz \
 		rudeconfig-5.0.5.tar.gz;http://rudeserver.com/config/download/rudeconfig-5.0.5.tar.gz \
-		boost_1_53_0.tar.gz;http://netcologne.dl.sourceforge.net/project/boost/boost/1.53.0/boost_1_53_0.tar.gz
+		boost_1_54_0.tar.gz;http://netcologne.dl.sourceforge.net/project/boost/boost/1.54.0/boost_1_54_0.tar.gz
 FETCH = wget -c --no-check-certificate -O -
 MAKETHREADS ?= 4