Merge pull request dmtcp#719 from karya0/exit_fix

Simplified dmtcp_initialize and exit handling

include/util.h

@@ -148,7 +148,6 @@ void initializeLogFile(const char *tmpDir,
                        const char *procname = "",
                        const char *preLogPath = "");
 
-void prepareDlsymWrapper();
 void adjustRlimitStack();
 void runMtcpRestore(int is32bitElf,
                     const char *path,

src/constants.h

@@ -116,12 +116,6 @@
 #define ENV_VAR_SCREENDIR               "SCREENDIR"
 #define ENV_VAR_DISABLE_STRICT_CHECKING "DMTCP_DISABLE_STRICT_CHECKING"
 
-#define GLIBC_BASE_FUNC                 isalnum
-
-#define LIBDL_BASE_FUNC                 dlinfo
-#define LIBDL_BASE_FUNC_STR             "dlinfo"
-#define ENV_VAR_DLSYM_OFFSET            "DMTCP_DLSYM_OFFSET"
-#define ENV_VAR_DLSYM_OFFSET_M32        "DMTCP_DLSYM_OFFSET_M32"
 #define ENV_VAR_REMOTE_SHELL_CMD        "DMTCP_REMOTE_SHELL_CMD"
 
 // this list should be kept up to date with all "protected" environment vars
@@ -144,8 +138,6 @@
   ENV_VAR_DL_PLUGIN,                  \
   ENV_VAR_SIGCKPT,                    \
   ENV_VAR_SCREENDIR,                  \
-  ENV_VAR_DLSYM_OFFSET,               \
-  ENV_VAR_DLSYM_OFFSET_M32,           \
   ENV_VAR_VIRTUAL_PID,                \
   ENV_VAR_SKIP_WRITING_TEXT_SEGMENTS, \
   ENV_DELTACOMPRESSION

src/coordinatorapi.cpp

@@ -93,11 +93,6 @@ eventHook(DmtcpEvent_t event, DmtcpEventData_t *data)
       init();
       break;
 
-    case DMTCP_EVENT_EXIT:
-      JTRACE("exit() in progress, disconnecting from dmtcp coordinator");
-      closeConnection();
-      break;
-
   default:
     break;
   }
@@ -371,20 +366,38 @@ recvMsgFromCoordinatorRaw(int fd, DmtcpMessage *msg, void **extraData)
     sem_launch_first_time = false;
   }
 
-  if (Util::readAll(fd, msg, sizeof(*msg)) != sizeof(*msg)) {
+  // Read into a temporary buffer in case the process exits after reading the
+  // message but before receiving the extradata.
+  DmtcpMessage tmpMsg;
+  if (Util::readAll(fd, &tmpMsg, sizeof(tmpMsg)) != sizeof(tmpMsg)) {
     // Perhaps the process is exit()'ing.
     return;
   }
 
-  if (msg->extraBytes > 0) {
+  if (tmpMsg.extraBytes > 0) {
     JASSERT(extraData != NULL);
 
     // Caller must free this buffer
-    void *buf = JALLOC_HELPER_MALLOC(msg->extraBytes);
-    JASSERT(Util::readAll(fd, (char*) buf, msg->extraBytes) == msg->extraBytes);
-    JASSERT(extraData != NULL);
+    void *buf = JALLOC_HELPER_MALLOC(tmpMsg.extraBytes);
+    if (Util::readAll(fd, buf, tmpMsg.extraBytes) != tmpMsg.extraBytes) {
+      JALLOC_HELPER_FREE(buf);
+      return;
+    }
+
     *extraData = buf;
   }
+
+  // All is well, return the received message.
+  *msg = tmpMsg;
+
+  // TODO(Kapil): Distinguish between DMT_KILL_PEER that arrives during
+  // checkpoint-phase (potentially due to a stuck computation that the user
+  // wants to kill) vs. normal runtime.
+  // TODO(Kapil): Consider generating an EXIT event for plugins.
+  if (msg->isValid() && msg->type == DMT_KILL_PEER) {
+    JTRACE("Received KILL message from coordinator, exiting");
+    _exit(0);
+  }
 }
 
 void sendMsgToCoordinator(DmtcpMessage msg, const void *extraData, size_t len)
@@ -414,10 +427,6 @@ void waitForBarrier(const string& barrierId)
   recvMsgFromCoordinator(&msg, (void**)&extraData);
 
   msg.assertValid();
-  if (msg.type == DMT_KILL_PEER) {
-    JTRACE("Received KILL message from coordinator, exiting");
-    _exit(0);
-  }
 
   JASSERT(msg.type == DMT_BARRIER_RELEASED) (msg.type);
   JASSERT(extraData != NULL);
@@ -549,10 +558,7 @@ sendRecvHandshake(int fd,
 
   recvMsgFromCoordinatorRaw(fd, &msg);
   msg.assertValid();
-  if (msg.type == DMT_KILL_PEER) {
-    JTRACE("Received KILL message from coordinator, exiting");
-    _real_exit(0);
-  }
+
   if (msg.type == DMT_REJECT_NOT_RUNNING) {
     JASSERT(false)
     .Text("Connection rejected by the coordinator.\n"

src/dmtcp_dlsym.cpp

@@ -67,7 +67,7 @@
 #endif
 #include <dlfcn.h>
 
-#include "dmtcp_dlsym.h"
+#include "dmtcp.h"
 #include "jassert.h"
 #include "config.h"
 

src/dmtcp_launch.cpp

@@ -606,9 +606,6 @@ main(int argc, char **argv)
                          &coordInfo,
                          &localIPAddr);
 
-  // Set DLSYM_OFFSET env var(s).
-  Util::prepareDlsymWrapper();
-
   setLDPreloadLibs(is32bitElf);
 
   // run the user program

src/dmtcp_restart.cpp

@@ -35,7 +35,6 @@
 #include "../jalib/jfilesystem.h"
 #include "constants.h"
 #include "coordinatorapi.h"
-#include "dmtcp_dlsym.h"
 #include "processinfo.h"
 #include "shareddata.h"
 #include "uniquepid.h"
@@ -312,11 +311,10 @@ class RestoreTarget
                                &compId,
                                &coordInfo,
                                &localIPAddr);
+
         Util::initializeLogFile(SharedData::getTmpDir().c_str(),
                                 _pInfo.procname().c_str(),
                                 NULL);
-
-        Util::prepareDlsymWrapper();
       }
 
       JTRACE("Creating process during restart") (upid()) (_pInfo.procname());

src/dmtcpplugin.cpp

@@ -427,39 +427,6 @@ dmtcp_get_ptrace_fd(void)
   return PROTECTED_PTRACE_FD;
 }
 
-LIB_PRIVATE int32_t dmtcp_dlsym_offset = -1;
-typedef void * (*dlsym_fnptr_t) (void *handle, const char *symbol);
-EXTERNC void *
-dmtcp_get_libc_dlsym_addr(void)
-{
-  static dlsym_fnptr_t _libc_dlsym_fnptr = NULL;
-
-#ifndef CONFIG_M32
-  const char *evar = ENV_VAR_DLSYM_OFFSET;
-#else // ifndef CONFIG_M32
-  const char *evar = ENV_VAR_DLSYM_OFFSET_M32;
-#endif // ifndef CONFIG_M32
-
-  if (_libc_dlsym_fnptr == NULL) {
-    if (getenv(evar) == NULL) {
-      fprintf(stderr,
-              "%s:%d DMTCP Internal Error: Env var DMTCP_DLSYM_OFFSET not"
-              " set.\n"
-              "      Aborting.\n\n",
-              __FILE__,
-              __LINE__);
-      abort();
-    }
-
-    dmtcp_dlsym_offset = (int32_t)strtol(getenv(evar), NULL, 10);
-
-    _libc_dlsym_fnptr = (dlsym_fnptr_t)((char *)&LIBDL_BASE_FUNC +
-                                        dmtcp_dlsym_offset);
-  }
-
-  return (void *)_libc_dlsym_fnptr;
-}
-
 EXTERNC void
 dmtcp_block_ckpt_signal(void)
 {