int main(int argc, char** argv)
{
char *tmpdir_arg = NULL;
char *ckptdir_arg = NULL;
initializeJalib();
if (!getenv(ENV_VAR_QUIET)) {
setenv(ENV_VAR_QUIET, "0", 0);
}
if (getenv(ENV_VAR_DISABLE_STRICT_CHECKING)) {
noStrictChecking = true;
}
if (getenv(ENV_VAR_CHECKPOINT_DIR)) {
ckptdir_arg = getenv(ENV_VAR_CHECKPOINT_DIR);
}
if (argc == 1) {
printf("%s", DMTCP_VERSION_AND_COPYRIGHT_INFO);
printf("(For help: %s --help)\n\n", argv[0]);
return DMTCP_FAIL_RC;
}
//process args
shift;
while (true) {
string s = argc>0 ? argv[0] : "--help";
if (s == "--help" && argc == 1) {
printf("%s", theUsage);
return DMTCP_FAIL_RC;
} else if ((s == "--version") && argc == 1) {
printf("%s", DMTCP_VERSION_AND_COPYRIGHT_INFO);
return DMTCP_FAIL_RC;
} else if (s == "-j" || s == "--join") {
allowedModes = COORD_JOIN;
shift;
} else if (s == "--new-coordinator") {
allowedModes = COORD_NEW;
shift;
} else if (s == "--no-strict-checking") {
noStrictChecking = true;
shift;
} else if (s == "-i" || s == "--interval") {
setenv(ENV_VAR_CKPT_INTR, argv[1], 1);
shift; shift;
} else if (argv[0][0] == '-' && argv[0][1] == 'i' &&
isdigit(argv[0][2])) { // else if -i5, for example
setenv(ENV_VAR_CKPT_INTR, argv[0]+2, 1);
shift;
} 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 (argc>1 && s == "--port-file"){
thePortFile = argv[1];
shift; shift;
} else if (argc > 1 && (s == "-c" || s == "--ckptdir")) {
ckptdir_arg = argv[1];
shift; shift;
} else if (argc > 1 && (s == "-t" || s == "--tmpdir")) {
tmpdir_arg = argv[1];
shift; shift;
} else if (s == "-q" || s == "--quiet") {
*getenv(ENV_VAR_QUIET) = *getenv(ENV_VAR_QUIET) + 1;
// Just in case a non-standard version of setenv is being used:
setenv(ENV_VAR_QUIET, getenv(ENV_VAR_QUIET), 1);
shift;
} else if ((s.length() > 2 && s.substr(0, 2) == "--") ||
(s.length() > 1 && s.substr(0, 1) == "-")) {
printf("Invalid Argument\n%s", theUsage);
return DMTCP_FAIL_RC;
} else if (argc > 1 && s == "--") {
shift;
break;
} else {
break;
}
}
tmpDir = Util::calcTmpDir(tmpdir_arg);
if (ckptdir_arg) {
setNewCkptDir(ckptdir_arg);
}
jassert_quiet = *getenv(ENV_VAR_QUIET) - '0';
//make sure JASSERT initializes now, rather than during restart
Util::initializeLogFile(tmpDir);
if (!noStrictChecking && jassert_quiet < 2 &&
(getuid() == 0 || geteuid() == 0)) {
JASSERT_STDERR <<
"WARNING: Running dmtcp_restart as root can be dangerous.\n"
" An unknown checkpoint image or bugs in DMTCP may lead to unforeseen\n"
" consequences. Continuing as root ....\n";
}
JTRACE("New dmtcp_restart process; _argc_ ckpt images") (argc);
bool doAbort = false;
for (; argc > 0; shift) {
string restorename(argv[0]);
struct stat buf;
int rc = stat(restorename.c_str(), &buf);
if (Util::strEndsWith(restorename, "_files")) {
continue;
} else if (!Util::strEndsWith(restorename, ".dmtcp")) {
JNOTE("File doesn't have .dmtcp extension. Check Usage.") (restorename);
// Don't test for --quiet here. We're aborting. We need to say why.
JASSERT_STDERR << theUsage;
doAbort = true;
} else if (rc == -1) {
char error_msg[1024];
sprintf(error_msg, "\ndmtcp_restart: ckpt image %s", restorename.c_str());
perror(error_msg);
doAbort = true;
} else if (buf.st_uid != getuid() && !noStrictChecking) {
/*Could also run if geteuid() matches*/
printf("\nProcess uid (%d) doesn't match uid (%d) of\n" \
"checkpoint image (%s).\n" \
"This is dangerous. Aborting for security reasons.\n" \
"If you still want to do this, then re-run dmtcp_restart\n" \
" with the --no-strict-checking flag.\n",
getuid(), buf.st_uid, restorename.c_str());
doAbort = true;
}
if (doAbort) {
exit(DMTCP_FAIL_RC);
}
JTRACE("Will restart ckpt image") (argv[0]);
RestoreTarget *t = new RestoreTarget(argv[0]);
targets[t->upid()] = t;
}
// Prepare list of independent process tree roots
RestoreTargetMap::iterator i;
for (i = targets.begin(); i != targets.end(); i++) {
RestoreTarget *t1 = i->second;
if (t1->isRootOfProcessTree()) {
RestoreTargetMap::iterator j;
for (j = targets.begin(); j != targets.end(); j++) {
RestoreTarget *t2 = j->second;
if (t1 == t2) continue;
if (t1->sid() == t2->pid()) {
break;
}
}
if (j == targets.end()) {
independentProcessTreeRoots[t1->upid()] = t1;
}
}
}
JASSERT(independentProcessTreeRoots.size() > 0)
.Text("There must be at least one process tree that doesn't have\n"
" a different process as session leader.");
WorkerState::setCurrentState(WorkerState::RESTARTING);
/* Try to find non-orphaned process in independent procs list */
RestoreTarget *t;
bool foundNonOrphan = false;
RestoreTargetMap::iterator it;
int size = independentProcessTreeRoots.size();
printf("size = %d\n", size);
for (it = independentProcessTreeRoots.begin();
it != independentProcessTreeRoots.end();
it++) {
t = it->second;
if ( !t->isOrphan() ) {
foundNonOrphan = true;
break;
}
}
JASSERT(t->pid() != 0);
JASSERT(!t->noCoordinator() || allowedModes == COORD_ANY)
.Text("Process had no coordinator prior to checkpoint;\n"
" but either --join or --new-coordinator was specified.");
if( foundNonOrphan ){
t->createProcess(true);
} else {
/* we were unable to find any non-orphaned procs.
* pick the first one and orphan it */
t = independentProcessTreeRoots.begin()->second;
t->createOrphanedProcess(true);
}
JASSERT(false).Text("unreachable");
return -1;
}
int main(int argc, char** argv)
{
bool autoStartCoordinator=true;
bool isRestart = true;
dmtcp::CoordinatorAPI::CoordinatorMode allowedModes =
dmtcp::CoordinatorAPI::COORD_ANY;
initializeJalib();
if (!getenv(ENV_VAR_QUIET)) {
setenv(ENV_VAR_QUIET, "0", 0);
}
if (argc == 1) {
JASSERT_STDERR << DMTCP_VERSION_AND_COPYRIGHT_INFO;
JASSERT_STDERR << "(For help: " << argv[0] << " --help)\n\n";
return DMTCP_FAIL_RC;
}
//process args
shift;
while (true) {
dmtcp::string s = argc>0 ? argv[0] : "--help";
if (s == "--help" && argc == 1) {
JASSERT_STDERR << theUsage;
return DMTCP_FAIL_RC;
} else if ((s == "--version") && argc == 1) {
JASSERT_STDERR << DMTCP_VERSION_AND_COPYRIGHT_INFO;
return DMTCP_FAIL_RC;
} else if (s == "--no-check") {
autoStartCoordinator = false;
shift;
} else if (s == "-j" || s == "--join") {
allowedModes = dmtcp::CoordinatorAPI::COORD_JOIN;
shift;
} else if (s == "-n" || s == "--new") {
allowedModes = dmtcp::CoordinatorAPI::COORD_NEW;
shift;
} else if (s == "--new-coordinator") {
allowedModes = dmtcp::CoordinatorAPI::COORD_FORCE_NEW;
shift;
} else if (s == "-b" || s == "--batch") {
allowedModes = dmtcp::CoordinatorAPI::COORD_BATCH;
shift;
} else if (s == "-i" || s == "--interval" ||
(s.c_str()[0] == '-' && s.c_str()[1] == 'i' &&
isdigit(s.c_str()[2]))) {
if (isdigit(s.c_str()[2])) { // if -i5, for example
setenv(ENV_VAR_CKPT_INTR, s.c_str()+2, 1);
shift;
} else { // else -i 5
setenv(ENV_VAR_CKPT_INTR, argv[1], 1);
shift; shift;
}
} 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 (argc > 1 && (s == "-t" || s == "--tmpdir")) {
setenv(ENV_VAR_TMPDIR, argv[1], 1);
shift; shift;
} else if (s == "-q" || s == "--quiet") {
*getenv(ENV_VAR_QUIET) = *getenv(ENV_VAR_QUIET) + 1;
// Just in case a non-standard version of setenv is being used:
setenv(ENV_VAR_QUIET, getenv(ENV_VAR_QUIET), 1);
shift;
} else if ((s.length() > 2 && s.substr(0, 2) == "--") ||
(s.length() > 1 && s.substr(0, 1) == "-")) {
JASSERT_STDERR << "Invalid Argument\n";
JASSERT_STDERR << theUsage;
return DMTCP_FAIL_RC;
} else if (argc > 1 && s == "--") {
shift;
break;
} else {
break;
}
}
dmtcp::UniquePid::setTmpDir(getenv(ENV_VAR_TMPDIR));
dmtcpTmpDir = dmtcp::UniquePid::getTmpDir();
jassert_quiet = *getenv(ENV_VAR_QUIET) - '0';
//make sure JASSERT initializes now, rather than during restart
Util::initializeLogFile();
if (jassert_quiet == 0)
JASSERT_STDERR << DMTCP_BANNER;
JTRACE("New dmtcp_restart process; _argc_ ckpt images") (argc);
bool doAbort = false;
for (; argc > 0; shift) {
dmtcp::string restorename(argv[0]);
struct stat buf;
int rc = stat(restorename.c_str(), &buf);
if (Util::strEndsWith(restorename, "_files")) {
continue;
} else if (!Util::strEndsWith(restorename, ".dmtcp")) {
JNOTE("File doesn't have .dmtcp extension. Check Usage.")
(restorename);
JASSERT_STDERR << theUsage;
doAbort = true;
} else if (rc == -1) {
char error_msg[1024];
sprintf(error_msg, "\ndmtcp_restart: ckpt image %s", restorename.c_str());
perror(error_msg);
doAbort = true;
} else if (buf.st_uid != getuid()) { /*Could also run if geteuid() matches*/
printf("\nProcess uid (%d) doesn't match uid (%d) of\n" \
"checkpoint image (%s).\n" \
"This is dangerous. Aborting for security reasons.\n" \
"If you still want to do this (at your own risk),\n" \
" then modify dmtcp/src/%s:%d and re-compile.\n",
getuid(), buf.st_uid, restorename.c_str(), __FILE__, __LINE__ - 6);
doAbort = true;
}
if (doAbort) {
exit(DMTCP_FAIL_RC);
}
JTRACE("Will restart ckpt image") (argv[0]);
RestoreTarget *t = new RestoreTarget(argv[0]);
targets[t->upid()] = t;
}
// Prepare list of independent process tree roots
RestoreTargetMap::iterator i;
for (i = targets.begin(); i != targets.end(); i++) {
RestoreTarget *t1 = i->second;
if (t1->isRootOfProcessTree()) {
RestoreTargetMap::iterator j;
for (j = targets.begin(); j != targets.end(); j++) {
RestoreTarget *t2 = j->second;
if (t1 == t2) continue;
if (t1->sid() == t2->pid()) {
break;
}
}
if (j == targets.end()) {
independentProcessTreeRoots[t1->upid()] = t1;
}
}
}
JASSERT(independentProcessTreeRoots.size() > 0)
.Text("There must atleast one process tree which doesn't have a different "
"process as session leader.");
if (autoStartCoordinator) {
dmtcp::CoordinatorAPI::startCoordinatorIfNeeded(allowedModes,
isRestart);
}
RestoreTarget *t = independentProcessTreeRoots.begin()->second;
JASSERT(t->pid() != 0);
t->createProcess(true);
JASSERT(false).Text("unreachable");
return -1;
}
void createProcess(bool createIndependentRootProcesses = false)
{
UniquePid::ThisProcess() = _pInfo.upid();
UniquePid::ParentProcess() = _pInfo.uppid();
Util::initializeLogFile(_pInfo.procname());
if (createIndependentRootProcesses) {
DmtcpUniqueProcessId compId = _pInfo.compGroup().upid();
CoordinatorInfo coordInfo;
struct in_addr localIPAddr;
if (_pInfo.noCoordinator()) {
allowedModes = COORD_NONE;
}
// dmtcp_restart sets ENV_VAR_NAME_HOST/PORT, even if cmd line flag used
const char *host = NULL;
int port = UNINITIALIZED_PORT;
CoordinatorAPI::getCoordHostAndPort(allowedModes, &host, &port);
// FIXME: We will use the new HOST and PORT here, but after restart,,
// we will use the old HOST and PORT from the ckpt image.
CoordinatorAPI::instance().connectToCoordOnRestart(allowedModes,
_pInfo.procname(),
_pInfo.compGroup(),
_pInfo.numPeers(),
&coordInfo,
host,
port,
&localIPAddr);
// If port was 0, we'll get new random port when coordinator starts up.
CoordinatorAPI::getCoordHostAndPort(allowedModes, &host, &port);
Util::writeCoordPortToFile(port, thePortFile.c_str());
string installDir =
jalib::Filesystem::DirName(jalib::Filesystem::GetProgramDir());
#if defined(__i386__) || defined(__arm__)
if (Util::strEndsWith(installDir, "/lib/dmtcp/32")) {
// If dmtcp_launch was compiled for 32 bits in 64-bit O/S, then note:
// DMTCP_ROOT/bin/dmtcp_launch is a symbolic link to:
// DMTCP_ROOT/bin/dmtcp_launch/lib/dmtcp/32/bin
// GetProgramDir() followed the link. So, need to remove the suffix.
char *str = const_cast<char*>(installDir.c_str());
str[strlen(str) - strlen("/lib/dmtcp/32")] = '\0';
installDir = str;
}
#endif
/* We need to initialize SharedData here to make sure that it is
* initialized with the correct coordinator timestamp. The coordinator
* timestamp is updated only during postCkpt callback. However, the
* SharedData area may be initialized earlier (for example, while
* recreating threads), causing it to use *older* timestamp.
*/
SharedData::initialize(tmpDir.c_str(),
installDir.c_str(),
&compId,
&coordInfo,
&localIPAddr);
Util::prepareDlsymWrapper();
}
JTRACE("Creating process during restart") (upid()) (_pInfo.procname());
RestoreTargetMap::iterator it;
for (it = targets.begin(); it != targets.end(); it++) {
RestoreTarget *t = it->second;
if (_pInfo.upid() == t->_pInfo.upid()) {
continue;
} else if (_pInfo.isChild(t->upid()) &&
t->_pInfo.sid() != _pInfo.pid()) {
t->createDependentChildProcess();
}
}
if (createIndependentRootProcesses) {
RestoreTargetMap::iterator it;
for (it = independentProcessTreeRoots.begin();
it != independentProcessTreeRoots.end();
it++) {
RestoreTarget *t = it->second;
if (t != this) {
t->createDependentNonChildProcess();
}
}
}
// If we were the session leader, become one now.
if (_pInfo.sid() == _pInfo.pid()) {
if (getsid(0) != _pInfo.pid()) {
JWARNING(setsid() != -1) (getsid(0)) (JASSERT_ERRNO)
.Text("Failed to restore this process as session leader.");
}
}
// Now recreate processes with sid == _pid
for (it = targets.begin(); it != targets.end(); it++) {
RestoreTarget *t = it->second;
if (_pInfo.upid() == t->_pInfo.upid()) {
continue;
} else if (t->_pInfo.sid() == _pInfo.pid()) {
if (_pInfo.isChild(t->upid())) {
t->createDependentChildProcess();
} else if (t->isRootOfProcessTree()) {
t->createDependentNonChildProcess();
}
}
}
// Now close all open fds except _fd;
for (it = targets.begin(); it != targets.end(); it++) {
RestoreTarget *t = it->second;
if (t != this) {
close(t->fd());
}
}
string ckptDir = jalib::Filesystem::GetDeviceName(PROTECTED_CKPT_DIR_FD);
if (ckptDir.length() == 0) {
// Create the ckpt-dir fd so that the restarted process can know about
// the abs-path of ckpt-image.
string dirName = jalib::Filesystem::DirName(_path);
int dirfd = open(dirName.c_str(), O_RDONLY);
JASSERT(dirfd != -1) (JASSERT_ERRNO);
if (dirfd != PROTECTED_CKPT_DIR_FD) {
JASSERT(dup2(dirfd, PROTECTED_CKPT_DIR_FD) == PROTECTED_CKPT_DIR_FD);
close(dirfd);
}
}
if (!createIndependentRootProcesses) {
// dmtcp_restart sets ENV_VAR_NAME_HOST/PORT, even if cmd line flag used
const char *host = NULL;
int port = UNINITIALIZED_PORT;
int *port_p = &port;
CoordinatorAPI::getCoordHostAndPort(allowedModes, &host, port_p);
CoordinatorAPI::instance().connectToCoordOnRestart(allowedModes,
_pInfo.procname(),
_pInfo.compGroup(),
_pInfo.numPeers(),
NULL,
host,
port,
NULL);
}
setEnvironFd();
int is32bitElf = 0;
#if defined(__x86_64__) || defined(__aarch64__)
is32bitElf = (_pInfo.elfType() == ProcessInfo::Elf_32);
#elif defined(__i386__) || defined(__arm__)
is32bitElf = true;
#endif
runMtcpRestart(is32bitElf, _fd, &_pInfo);
JASSERT ( false ).Text ( "unreachable" );
}
void createProcess(bool createIndependentRootProcesses = false)
{
//change UniquePid
UniquePid::resetOnFork(upid());
dmtcp::Util::initializeLogFile(_pInfo.procname());
JTRACE("Creating process during restart") (upid()) (_pInfo.procname());
RestoreTargetMap::iterator it;
for (it = targets.begin(); it != targets.end(); it++) {
RestoreTarget *t = it->second;
if (_pInfo.upid() == t->_pInfo.upid()) {
continue;
} else if (_pInfo.isChild(t->upid()) &&
t->_pInfo.sid() != _pInfo.pid()) {
t->createDependentProcess(true);
}
}
if (createIndependentRootProcesses) {
RestoreTargetMap::iterator it;
for (it = independentProcessTreeRoots.begin();
it != independentProcessTreeRoots.end();
it++) {
RestoreTarget *t = it->second;
if (t != this) {
t->createDependentProcess(false);
}
}
}
// If we were the session leader, become one now.
if (_pInfo.sid() == _pInfo.pid()) {
if (getsid(0) != _pInfo.pid()) {
JWARNING(setsid() != -1) (getsid(0)) (JASSERT_ERRNO)
.Text("Failed to restore this process as session leader.");
}
}
// Now recreate processes with sid == _pid
for (it = targets.begin(); it != targets.end(); it++) {
RestoreTarget *t = it->second;
if (_pInfo.upid() == t->_pInfo.upid()) {
continue;
} else if (t->_pInfo.sid() == _pInfo.pid()) {
t->createDependentProcess(_pInfo.isChild(t->upid()));
}
}
// Now close all open fds except _fd;
for (it = targets.begin(); it != targets.end(); it++) {
RestoreTarget *t = it->second;
if (t != this) {
close(t->fd());
}
}
// Create the ckpt-dir fd so that the restarted process can know about
// the abs-path of ckpt-image.
dmtcp::string deviceName = jalib::Filesystem::GetDeviceName(_fd);
dmtcp::string dirName = jalib::Filesystem::DirName(deviceName);
int dirfd = open(dirName.c_str(), O_RDONLY);
JASSERT(dirfd != -1) (JASSERT_ERRNO);
if (dirfd != PROTECTED_CKPT_DIR_FD) {
JASSERT(dup2(dirfd, PROTECTED_CKPT_DIR_FD) == PROTECTED_CKPT_DIR_FD);
close(dirfd);
}
dmtcp::CoordinatorAPI coordinatorAPI;
coordinatorAPI.connectToCoordinator();
dmtcp::Util::runMtcpRestore(_path.c_str(), _fd, _extDecompPid,
_pInfo.argvSize(), _pInfo.envSize());
JASSERT ( false ).Text ( "unreachable" );
}