void dmtcp::FileConnList::remapShmMaps()
{
for (size_t i = 0; i < shmAreas.size(); i++) {
Util::ProcMapsArea *area = &shmAreas[i];
FileConnection *fileCon = shmAreaConn[i];
int fd = fileCon->getFds()[0];
JTRACE("Restoring shared memory area") (area->name) (area->addr);
void *addr = _real_mmap(area->addr, area->size, area->prot,
MAP_FIXED | area->flags,
fd, area->offset);
JASSERT(addr != MAP_FAILED) (area->flags) (area->prot) (JASSERT_ERRNO) .Text("mmap failed");
_real_close(fd);
processClose(fd);
}
shmAreas.clear();
shmAreaConn.clear();
}
void dmtcp::FileConnList::prepareShmList()
{
Util::ProcMapsArea area;
int mapsfd = _real_open("/proc/self/maps", O_RDONLY, 0);
JASSERT(mapsfd != -1) (JASSERT_ERRNO);
shmAreas.clear();
shmAreaConn.clear();
while (Util::readProcMapsLine(mapsfd, &area)) {
if ((area.flags & MAP_SHARED) && area.prot != 0) {
if (strstr(area.name, "ptraceSharedInfo") != NULL ||
strstr(area.name, "dmtcpPidMap") != NULL ||
strstr(area.name, "dmtcpSharedArea") != NULL ||
strstr(area.name, "dmtcpSharedArea") != NULL ||
strstr(area.name, "synchronization-log") != NULL ||
strstr(area.name, "synchronization-read-log") != NULL) {
continue;
}
if (jalib::Filesystem::FileExists(area.name)) {
if (_real_access(area.name, W_OK) == 0) {
JTRACE("Will checkpoint shared memory area") (area.name);
int flags = Util::memProtToOpenFlags(area.prot);
int fd = _real_open(area.name, flags, 0);
JASSERT(fd != -1) (JASSERT_ERRNO) (area.name);
FileConnection *fileConn =
new FileConnection(area.name, flags, 0, FileConnection::FILE_SHM);
add(fd, fileConn);
shmAreas.push_back(area);
shmAreaConn.push_back(fileConn);
/* Instead of unmapping the shared memory area, we make it
* non-readable. This way mtcp will skip the region while at the same
* time, we prevent JALLOC arena to grow over it.
*
* By munmapping the area, a bug was observed on CCIS linux with
* 'make check-java'. Once the region was unmapped, the JALLOC arena
* grew over it. During restart, the JALLOC'd area was reclaimed for
* remapping the shm file without informing JALLOC. Finally, during
* the second checkpoint cycle, the area was again unmapped and later
* JALLOC tried to access it, causing a SIGSEGV.
*/
JASSERT(_real_mmap(area.addr, area.size, PROT_NONE,
MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED,
-1, 0) != MAP_FAILED) (JASSERT_ERRNO);
} else {
JTRACE("Will not checkpoint shared memory area") (area.name);
}
} else {
// TODO: Shared memory areas with unlinked backing files.
#if 0
JASSERT(Util::strEndsWith(area.name, DELETED_FILE_SUFFIX)) (area.name);
if (Util::strStartsWith(area.name, DEV_ZERO_DELETED_STR) ||
Util::strStartsWith(area.name, DEV_NULL_DELETED_STR)) {
JWARNING(false) (area.name)
.Text("Ckpt/Restart of Anon Shared memory not supported.");
} else {
JTRACE("Will recreate shm file on restart.") (area.name);
//shmAreas[area] = NULL;
}
#endif
}
}
}
_real_close(mapsfd);
}
bool dmtcp::DmtcpWorker::waitForStage2bCheckpoint()
{
waitForCoordinatorMsg ( "PEER_LOOKUP", DMT_DO_PEER_LOOKUP );
JTRACE ( "Looking up Socket Peers..." );
theTcpConnections.clear();
theCheckpointState->preCheckpointPeerLookup(theTcpConnections);
sendPeerLookupRequest(theTcpConnections);
JTRACE ( "Done Socket Peer Lookup" );
WorkerState::setCurrentState ( WorkerState::PEER_LOOKUP_COMPLETE );
{
dmtcp::DmtcpMessage msg;
msg.type = DMT_OK;
msg.state = WorkerState::currentState();
_coordinatorSocket << msg;
JTRACE ( "waiting for DRAIN/RESUME message" );
do {
msg.poison();
_coordinatorSocket >> msg;
msg.assertValid();
if ( msg.type == DMT_KILL_PEER ) {
JTRACE ( "Received KILL message from coordinator, exiting" );
_exit ( 0 );
}
JTRACE ( "received message" ) (msg.type );
if ( msg.type != DMT_UNKNOWN_PEER )
break;
JTRACE ("received DMT_UNKNOWN_PEER message") (msg.conId);
TcpConnection* con =
(TcpConnection*) &( ConnectionList::instance() [msg.conId] );
con->markExternal();
externalTcpConnections.push_back(msg.conId);
_waitingForExternalSocketsToClose = true;
} while ( msg.type == DMT_UNKNOWN_PEER );
JASSERT ( msg.type == DMT_DO_DRAIN || msg.type == DMT_DO_RESUME )
( msg.type );
ConnectionList& connections = ConnectionList::instance();
// Tcp Accept and Connect connection with PeerType UNKNOWN should be marked as INTERNAL
for ( ConnectionList::iterator i = connections.begin()
; i!= connections.end()
; ++i )
{
Connection* con = i->second;
if ( con->conType() == Connection::TCP ) {
TcpConnection* tcpCon = (TcpConnection *) con;
if ( (tcpCon->tcpType() == TcpConnection::TCP_ACCEPT ||
tcpCon->tcpType() == TcpConnection::TCP_CONNECT) &&
tcpCon->peerType() == TcpConnection::PEER_UNKNOWN )
tcpCon->markInternal();
}
}
if ( msg.type == DMT_DO_RESUME ) {
JTRACE ( "Peer Lookup not complete, skipping checkpointing \n\n\n\n\n");
return false;
}
JASSERT (msg.type == DMT_DO_DRAIN);
}
}