static void runCoordinatorCmd(char c,
int *coordCmdStatus = NULL,
int *numPeers = NULL,
int *isRunning = NULL)
{
_dmtcp_lock();
{
CoordinatorAPI coordinatorAPI;
dmtcp_disable_ckpt();
coordinatorAPI.connectAndSendUserCommand(c, coordCmdStatus, numPeers,
isRunning);
dmtcp_enable_ckpt();
}
_dmtcp_unlock();
}
int main ( int argc, char** argv )
{
bool quiet = false;
dmtcp::string interval = "";
dmtcp::string request = "h";
initializeJalib();
Util::initializeLogFile();
//process args
shift;
while(argc>0){
dmtcp::string s = argv[0];
if((s=="--help" || s=="-h") && argc==1){
fprintf(stderr, theUsage, "");
return 1;
} else if ((s=="--version") && argc==1){
JASSERT_STDERR << DMTCP_VERSION_AND_COPYRIGHT_INFO;
return 1;
}else if(argc>1 && (s == "-h" || s == "--host")){
setenv(ENV_VAR_NAME_HOST, argv[1], 1);
shift; shift;
}else if(argc>1 && (s == "-p" || s == "--port")){
setenv(ENV_VAR_NAME_PORT, argv[1], 1);
shift; shift;
}else if(s == "--quiet"){
quiet = true;
shift;
}else if(s == "h" || s == "-h" || s == "--help" || s == "?"){
fprintf(stderr, theUsage, "");
return 1;
}else{ // else it's a request
char* cmd = argv[0];
//ignore leading dashes
while(*cmd == '-') cmd++;
s = cmd;
if(*cmd == 'b' && *(cmd+1) != 'c'){
// If blocking ckpt, next letter must be 'c'; else print the usage
fprintf(stderr, theUsage, "");
return 1;
} else if (*cmd == 's' || *cmd == 'i' || *cmd == 'c' || *cmd == 'b'
|| *cmd == 'f' || *cmd == 'k' || *cmd == 'q') {
request = s;
if (*cmd == 'i') {
if (isdigit(cmd[1])) { // if -i5, for example
interval = cmd+1;
} else { // else -i 5
if (argc == 1) {
fprintf(stderr, theUsage, "");
return 1;
}
interval = argv[1];
shift;
}
}
shift;
}else{
fprintf(stderr, theUsage, "");
return 1;
}
}
}
if (! quiet)
printf( "DMTCP-" PACKAGE_VERSION " (+ MTCP), Copyright (C) 2006-2011"
" Jason Ansel, Michael Rieker,\n"
" Kapil Arya, and Gene Cooperman\n"
"This program comes with ABSOLUTELY NO WARRANTY.\n"
"This is free software, and you are welcome to redistribute it\n"
"under certain conditions; see COPYING file for details.\n"
"(Use flag \"--quiet\" to hide this message.)\n\n");
int coordErrorCode = CoordinatorAPI::NOERROR;
int numPeers;
int isRunning;
CoordinatorAPI coordinatorAPI;
char *cmd = (char *)request.c_str();
switch (*cmd) {
case 'h':
fprintf(stderr, theUsage, "");
return 1;
case 'i':
setenv(ENV_VAR_CKPT_INTR, interval.c_str(), 1);
coordinatorAPI.connectAndSendUserCommand(*cmd, &coordErrorCode);
printf("Interval changed to %s\n", interval.c_str());
break;
case 'b':
// blocking prefix
coordinatorAPI.connectAndSendUserCommand(*cmd, &coordErrorCode);
// actual command
coordinatorAPI.connectAndSendUserCommand(*(cmd+1), &coordErrorCode);
break;
case 's':
coordinatorAPI.connectAndSendUserCommand(*cmd, &coordErrorCode,
&numPeers, &isRunning);
case 'c':
case 'f':
case 'k':
case 'q':
coordinatorAPI.connectAndSendUserCommand(*cmd, &coordErrorCode);
break;
}
//check for error
if (coordErrorCode != CoordinatorAPI::NOERROR) {
switch(coordErrorCode){
case CoordinatorAPI::ERROR_COORDINATOR_NOT_FOUND:
if (getenv("DMTCP_PORT"))
fprintf(stderr, "Coordinator not found. Please check port and host.\n");
else
fprintf(stderr,
"Coordinator not found. Try specifying port with \'--port\'.\n");
break;
case CoordinatorAPI::ERROR_INVALID_COMMAND:
fprintf(stderr,
"Unknown command: %c, try 'dmtcp_command --help'\n", *cmd);
break;
case CoordinatorAPI::ERROR_NOT_RUNNING_STATE:
fprintf(stderr, "Error, computation not in running state."
" Either a checkpoint is\n"
" currently happening or there are no connected processes.\n");
break;
default:
fprintf(stderr, "Unknown error\n");
break;
}
return 2;
}
if(*cmd == 's'){
if (getenv(ENV_VAR_NAME_HOST))
printf(" Host: %s\n", getenv(ENV_VAR_NAME_HOST));
printf(" Port: %s\n", getenv(ENV_VAR_NAME_PORT));
printf("Status...\n");
printf("NUM_PEERS=%d\n", numPeers);
printf("RUNNING=%s\n", (isRunning?"yes":"no"));
}
return 0;
}
int main ( int argc, char** argv )
{
string interval = "";
string request = "h";
initializeJalib();
// No need to initialize the log file.
// Util::initializeLogFile();
//process args
shift;
while(argc>0){
string s = argv[0];
if((s=="--help" || s=="-h") && argc==1){
printf("%s", theUsage);
return 1;
} else if ((s=="--version") && argc==1){
printf("%s", DMTCP_VERSION_AND_COPYRIGHT_INFO);
return 1;
}else if(argc>1 && (s == "-h" || s == "--coord-host" || s == "--host")){
setenv(ENV_VAR_NAME_HOST, argv[1], 1);
shift; shift;
} else if (argc>1 && (s == "-p" || s == "--coord-port" || s == "--port")) {
setenv(ENV_VAR_NAME_PORT, argv[1], 1);
shift; shift;
} else if (argv[0][0] == '-' && argv[0][1] == 'p' &&
isdigit(argv[0][2])) { // else if -p0, for example
setenv(ENV_VAR_NAME_PORT, argv[0]+2, 1);
shift;
}else if(s == "h" || s == "-h" || s == "--help" || s == "?"){
fprintf(stderr, theUsage, "");
return 1;
}else{ // else it's a request
char* cmd = argv[0];
//ignore leading dashes
while(*cmd == '-') cmd++;
s = cmd;
if((*cmd == 'b' || *cmd == 'x') && *(cmd+1) != 'c'){
// If blocking ckpt, next letter must be 'c'; else print the usage
fprintf(stderr, theUsage, "");
return 1;
} else if (*cmd == 's' || *cmd == 'i' || *cmd == 'c' || *cmd == 'b' ||
*cmd == 'x' || *cmd == 'k' || *cmd == 'q') {
request = s;
if (*cmd == 'i') {
if (isdigit(cmd[1])) { // if -i5, for example
interval = cmd+1;
} else { // else -i 5
if (argc == 1) {
fprintf(stderr, theUsage, "");
return 1;
}
interval = argv[1];
shift;
}
}
shift;
}else{
fprintf(stderr, theUsage, "");
return 1;
}
}
}
int coordCmdStatus = CoordCmdStatus::NOERROR;
int numPeers;
int isRunning;
int ckptInterval;
CoordinatorAPI coordinatorAPI;
char *cmd = (char *)request.c_str();
switch (*cmd) {
case 'h':
fprintf(stderr, theUsage, "");
return 1;
case 'i':
setenv(ENV_VAR_CKPT_INTR, interval.c_str(), 1);
coordinatorAPI.connectAndSendUserCommand(*cmd, &coordCmdStatus);
printf("Interval changed to %s\n", interval.c_str());
break;
case 'b':
case 'x':
// blocking prefix
coordinatorAPI.connectAndSendUserCommand(*cmd, &coordCmdStatus);
// actual command
coordinatorAPI.connectAndSendUserCommand(*(cmd+1), &coordCmdStatus);
break;
case 's':
coordinatorAPI.connectAndSendUserCommand(*cmd, &coordCmdStatus,
&numPeers, &isRunning, &ckptInterval);
case 'c':
case 'k':
case 'q':
coordinatorAPI.connectAndSendUserCommand(*cmd, &coordCmdStatus);
break;
}
//check for error
if (coordCmdStatus != CoordCmdStatus::NOERROR) {
switch(coordCmdStatus){
case CoordCmdStatus::ERROR_COORDINATOR_NOT_FOUND:
if (getenv("DMTCP_COORD_PORT") || getenv("DMTCP_PORT"))
fprintf(stderr, "Coordinator not found. Please check port and host.\n");
else
fprintf(stderr,
"Coordinator not found. Try specifying port with \'--port\'.\n");
break;
case CoordCmdStatus::ERROR_INVALID_COMMAND:
fprintf(stderr,
"Unknown command: %c, try 'dmtcp_command --help'\n", *cmd);
break;
case CoordCmdStatus::ERROR_NOT_RUNNING_STATE:
fprintf(stderr, "Error, computation not in running state."
" Either a checkpoint is\n"
" currently happening or there are no connected processes.\n");
break;
default:
fprintf(stderr, "Unknown error\n");
break;
}
return 2;
}
#define QUOTE(arg) #arg
#define STRINGIFY(arg) QUOTE(arg)
if(*cmd == 's'){
printf("Coordinator:\n");
char *host = getenv(ENV_VAR_NAME_HOST);
if (host == NULL) host = getenv("DMTCP_HOST"); // deprecated
printf(" Host: %s\n", (host ? host : "localhost"));
char *port = getenv(ENV_VAR_NAME_PORT);
if (port == NULL) port = getenv("DMTCP_PORT"); // deprecated
printf(" Port: %s\n",
(port ? port : STRINGIFY(DEFAULT_PORT) " (default port)"));
printf("Status...\n");
printf(" NUM_PEERS=%d\n", numPeers);
printf(" RUNNING=%s\n", (isRunning?"yes":"no"));
if (ckptInterval) {
printf(" CKPT_INTERVAL=%d\n", ckptInterval);
} else {
printf(" CKPT_INTERVAL=0 (checkpoint manually)\n");
}
}
return 0;
}
void RestoreTarget::CreateProcess(CoordinatorAPI& coordinatorAPI,
SlidingFdTable& slidingFd)
{
//change UniquePid
UniquePid::resetOnFork(upid());
//UniquePid::ThisProcess(true) = _conToFd.upid();
Util::initializeLogFile(procname());
JTRACE("Creating process during restart") (upid()) (procname());
JTRACE("")(getpid())(getppid())(getsid(0));
ProcessInfo &pInfo = _processInfo;
pid_t psid = pInfo.sid();
JTRACE("Restore /proc/self/* fds");
ConnectionList& connections = ConnectionList::instance();
ConnectionList::iterator it;
for (it = connections.begin(); it != connections.end(); ++it) {
dmtcp::Connection *con = it->second;
if (con->subType() == FileConnection::FILE_PROCFS) {
dmtcp::FileConnection *filecon = (dmtcp::FileConnection*) con;
char buf[32];
dmtcp::vector<int> fds;
fds.push_back(slidingFd.getFdFor(con->id()));
sprintf(buf, "/proc/%d/", pInfo.pid());
if (dmtcp::Util::strStartsWith(filecon->filePath(), buf)) {
filecon->restore(fds);
}
}
}
if (!isSessionLeader()) {
// Restore Group information
restoreGroup(slidingFd);
// If process is not session leader, restore it and all children.
t_iterator it = _children.begin();
for (; it != _children.end(); it++) {
JTRACE ("Forking Child Process") ((*it)->upid());
pid_t cid = fork();
if (cid == 0) {
(*it)->CreateProcess (coordinatorAPI, slidingFd);
JASSERT (false) . Text ("Unreachable");
}
JASSERT (cid > 0);
}
} else {
// Process is session leader.
// There may be not setsid-ed children.
for (t_iterator it = _children.begin(); it != _children.end(); it++) {
s_iterator sit = (*it)->getSmap().find(psid);
JTRACE("Restore processes that were created before their parent called setsid()");
if (sit == (*it)->getSmap().end()) {
JTRACE ("Forking Child Process") ((*it)->upid());
pid_t cid = fork();
if (cid == 0) {
(*it)->CreateProcess (coordinatorAPI, slidingFd);
JASSERT (false) . Text ("Unreachable");
}
JASSERT (cid > 0);
}
}
pid_t nsid = setsid();
JTRACE("change SID")(nsid);
// Restore Group information
restoreGroup(slidingFd);
for (t_iterator it = _children.begin(); it != _children.end(); it++) {
JTRACE("Restore processes that was created after their parent called setsid()");
s_iterator sit = (*it)->getSmap().find(psid);
if (sit != (*it)->getSmap().end()) {
JTRACE ("Forking Child Process") ((*it)->upid());
pid_t cid = fork();
if (cid == 0) {
(*it)->CreateProcess (coordinatorAPI, slidingFd);
JASSERT (false) . Text ("Unreachable");
}
JASSERT (cid> 0);
}
}
for (t_iterator it = _roots.begin() ; it != _roots.end(); it++) {
JTRACE ("Forking Dependent Root Process") ((*it)->upid());
pid_t cid;
if ((cid = fork())) {
waitpid(cid, NULL, 0);
} else {
if (fork())
exit(0);
(*it)->CreateProcess(coordinatorAPI, slidingFd);
JASSERT (false) . Text("Unreachable");
}
}
}
bool isTheGroupLeader = isGroupLeader(); // Calls JTRACE;avoid recursion
JTRACE("Child and dependent root processes forked, restoring process")
(upid())(getpid())(isTheGroupLeader);
//Reconnect to dmtcp_coordinator
WorkerState::setCurrentState (WorkerState::RESTARTING);
coordinatorAPI.connectToCoordinator();
coordinatorAPI.sendCoordinatorHandshake(procname(), _processInfo.compGroup());
coordinatorAPI.recvCoordinatorHandshake();
//restart targets[i]
dupAllSockets (slidingFd);
mtcpRestart();
JASSERT (false).Text ("unreachable");
}