Dmtcp exitest r527

configure

@@ -1,3 +1,4 @@
+
 #! /bin/sh
 # Guess values for system-dependent variables and create Makefiles.
 # Generated by GNU Autoconf 2.65 for DMTCP 1.1.4.

dmtcp/jalib/jsocket.cpp

@@ -188,6 +188,7 @@ ssize_t jalib::JSocket::readAll ( char* buf, size_t len )
         JWARNING ( retval >= 0 ) ( tmp_sockfd ) ( JASSERT_ERRNO ).Text ( "select() failed" );
         return -1;
       }
+      JTRACE("SELECT retval = -1")(strerror(errno));
     }
     else if ( retval )
     {

dmtcp/src/dmtcpworker.cpp

@@ -367,6 +367,14 @@ void dmtcp::DmtcpWorker::CleanupWorker()
   _coordinatorSocket.close();
 }
 
+
+void dmtcp::DmtcpWorker::interruptCkpthread()
+{
+    while( pthread_mutex_trylock(&destroyDmtcpWorker) == EBUSY){
+      killCkpthread();    
+    }
+}
+
 //called after user main()
 dmtcp::DmtcpWorker::~DmtcpWorker()
 {
@@ -393,7 +401,7 @@ dmtcp::DmtcpWorker::~DmtcpWorker()
      */
     JTRACE ( "exit() in progress, disconnecting from dmtcp coordinator" );
     _coordinatorSocket.close();
-    JASSERT(pthread_mutex_lock(&destroyDmtcpWorker)==0)(JASSERT_ERRNO);
+    interruptCkpthread();
   }
   CleanupWorker();
 }
@@ -1019,3 +1027,8 @@ void __attribute__ ((weak)) dmtcp::initializeMtcpEngine()
 {
   JASSERT(false).Text("should not be called");
 }
+
+void __attribute__ ((weak)) dmtcp::killCkpthread()
+{
+  JASSERT(false).Text("should not be called");
+}

dmtcp/src/dmtcpworker.h

@@ -86,6 +86,8 @@ namespace dmtcp
       static void setExitInProgress() { _exitInProgress = true; };
       static bool exitInProgress() { return _exitInProgress; };
 
+      void interruptCkpthread();
+
       void connectToCoordinator(bool doHanshaking=true);
       // np > -1 means it is restarting process that have np processes in its computation group
       // np == -1 means it is new pure process, so coordinator needs to generate compGroup ID for it

dmtcp/src/mtcpinterface.cpp

@@ -92,6 +92,7 @@ extern "C"
           void ( *write_ckpt_prefix ) ( int fd ),
           void ( *write_tid_maps) ());
   typedef int ( *t_mtcp_ok ) ( void );
+  typedef void ( *t_mtcp_kill_ckpthread ) ( void );
 }
 
 static void callbackSleepBetweenCheckpoint ( int sec )
@@ -202,6 +203,15 @@ void dmtcp::initializeMtcpEngine()
   JTRACE ( "mtcp_init complete" ) ( UniquePid::checkpointFilename() );
 }
 
+
+
+void dmtcp::killCkpthread()
+{
+  t_mtcp_kill_ckpthread kill_ckpthread = 
+    (t_mtcp_kill_ckpthread) _get_mtcp_symbol( "kill_ckpthread" );
+  kill_ckpthread();
+}
+
 #ifdef PID_VIRTUALIZATION
 struct ThreadArg {
   int ( *fn ) ( void *arg );

dmtcp/src/mtcpinterface.h

@@ -26,6 +26,7 @@ namespace dmtcp
 {
 
   void initializeMtcpEngine();
+  void killCkpthread();    
 
   void shutdownMtcpEngineOnFork();
 

dmtcp/src/workerhijack.cpp

@@ -60,5 +60,6 @@ void __attribute__ ((weak)) dmtcp::initializeMtcpEngine()
 {
   JASSERT(false).Text("should not be called");
 }
+
 dmtcp::DmtcpWorker& dmtcp::DmtcpWorker::instance() { return theInstance; }
 

mtcp/mtcp.c

@@ -1146,6 +1146,25 @@ int mtcp_no (void)
     }
   }
 }
+
+
+void kill_ckpthread (void)
+{
+  int tid;
+  Thread *thread;
+
+  lock_threads ();
+  for (thread = threads; thread != NULL; thread = thread -> next) {
+    if ( mtcp_state_value(&thread -> state) == ST_CKPNTHREAD ) {
+      unlk_threads ();
+      printf("\n\n\nKill checkpinthread, tid=%d\n\n\n",thread->tid);
+      mtcp_sys_kernel_tkill(thread -> tid, STOPSIGNAL);
+      return;
+    }
+  }
+  unlk_threads ();
+}
+
 
 /********************************************************************************************************************************/
 /*																*/
@@ -1180,6 +1199,7 @@ static void *checkpointhread (void *dummy)
   ckpthread = getcurrenthread ();
   save_sig_state (ckpthread);
   save_tls_state (ckpthread);
+
   /* Release user thread after we've initialized. */
   sem_post(&sem_start);
   if (getcontext (&(ckpthread -> savctx)) < 0) mtcp_abort ();
@@ -2139,12 +2159,19 @@ static void stopthisthread (int signum)
 #define STDERR_FD 826
 #define LOG_FD 826
 
+
+
   DPRINTF (("mtcp stopthisthread*: tid %d returns to %p\n",
             mtcp_sys_kernel_gettid (), __builtin_return_address (0)));
 
   setup_sig_handler ();  // re-establish in case of another STOPSIGNAL so we don't abort by default
 
   thread = getcurrenthread ();                                              // see which thread this is
+
+  if(  mtcp_state_value(&thread -> state) == ST_CKPNTHREAD ){
+    return ;
+  }
+
   if (0 && thread == motherofall) {
     void *buffer[BT_SIZE];
     int nptrs;
@@ -2308,7 +2335,6 @@ static void wait_for_all_restored (void)
 /********************************************************************************************************************************/
 
 static void save_sig_state (Thread *thisthread)
-
 {
   int i;
   sigset_t blockall;
@@ -2331,6 +2357,11 @@ static void save_sig_state (Thread *thisthread)
       }
     }
   }
+
+  sigdelset(&blockall,STOPSIGNAL);
+  if (_real_sigprocmask (SIG_SETMASK, &blockall, &(thisthread -> sigblockmask)) < 0) {
+    mtcp_abort ();
+  }
 }
 
 /********************************************************************************************************************************/