root/dmtcp_restart.cpp

DEFINITIONS

1. fd
2. upid
3. sid
4. isRootOfProcessTree
5. procname
6. compGroup
7. numPeers
8. noCoordinator
9. restoreGroup
10. createDependentChildProcess
11. createDependentNonChildProcess
12. createProcess
13. runMtcpRestart
14. readCkptHeader
15. first_char
16. open_ckpt_to_read
17. openCkptFileToRead
18. setEnvironFd
19. setNewCkptDir
20. main

   1 /****************************************************************************
   2  *   Copyright (C) 2006-2013 by Jason Ansel, Kapil Arya, and Gene Cooperman *
   3  *   jansel@csail.mit.edu, kapil@ccs.neu.edu, gene@ccs.neu.edu              *
   4  *                                                                          *
   5  *  This file is part of DMTCP.                                             *
   6  *                                                                          *
   7  *  DMTCP is free software: you can redistribute it and/or                  *
   8  *  modify it under the terms of the GNU Lesser General Public License as   *
   9  *  published by the Free Software Foundation, either version 3 of the      *
  10  *  License, or (at your option) any later version.                         *
  11  *                                                                          *
  12  *  DMTCP is distributed in the hope that it will be useful,                *
  13  *  but WITHOUT ANY WARRANTY; without even the implied warranty of          *
  14  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the           *
  15  *  GNU Lesser General Public License for more details.                     *
  16  *                                                                          *
  17  *  You should have received a copy of the GNU Lesser General Public        *
  18  *  License along with DMTCP:dmtcp/src.  If not, see                        *
  19  *  <http://www.gnu.org/licenses/>.                                         *
  20  ****************************************************************************/
  22 #include <stdio.h>
  23 #include <sys/stat.h>
  24 #include <sys/fcntl.h>
  25 #include <sys/wait.h>
  26 #include <sys/mman.h>
  27 #include <limits.h>
  28 #include <elf.h>
  29 #include "config.h"
  30 #ifdef HAS_PR_SET_PTRACER
  31 # include <sys/prctl.h>
  32 #endif
  34 #include "constants.h"
  35 #include "coordinatorapi.h"
  36 #include "util.h"
  37 #include "uniquepid.h"
  38 #include "processinfo.h"
  39 #include "shareddata.h"
  40 #include  "../jalib/jassert.h"
  41 #include  "../jalib/jfilesystem.h"
  43 #define BINARY_NAME "dmtcp_restart"
  44 #define MTCP_RESTART_BINARY "mtcp_restart"
  46 using namespace dmtcp;
  48 // Copied from mtcp/mtcp_restart.c.
  49 #define DMTCP_MAGIC_FIRST 'D'
  50 #define GZIP_FIRST 037
  51 #ifdef HBICT_DELTACOMP
  52 #define HBICT_FIRST 'H'
  53 #endif
  55 static void setEnvironFd();
string tmpDir = "/DMTCP/Uninitialized/Tmp/Dir";
  59 // gcc-4.3.4 -Wformat=2 issues false positives for warnings unless the format
  60 // string has at least one format specifier with corresponding format argument.
  61 // Ubuntu 9.01 uses -Wformat=2 by default.
  62 static const char* theUsage =
  63   "Usage: dmtcp_restart [OPTIONS] <ckpt1.dmtcp> [ckpt2.dmtcp...]\n\n"
  64   "Restart processes from a checkpoint image.\n\n"
  65   "Connecting to the DMTCP Coordinator:\n"
  66   "  -h, --coord-host HOSTNAME (environment variable DMTCP_COORD_HOST)\n"
  67   "              Hostname where dmtcp_coordinator is run (default: localhost)\n"
  68   "  -p, --coord-port PORT_NUM (environment variable DMTCP_COORD_PORT)\n"
  69   "              Port where dmtcp_coordinator is run (default: 7779)\n"
  70   "  --port-file FILENAME\n"
  71   "              File to write listener port number.\n"
  72   "              (Useful with '--port 0', which is used to assign a random port)\n"
  73   "  -j, --join\n"
  74   "              Join an existing coordinator, raise error if one doesn't\n"
  75   "              already exist\n"
  76   "  --new-coordinator\n"
  77   "              Create a new coordinator at the given port. Fail if one\n"
  78   "              already exists on the given port. The port can be specified\n"
  79   "              with --coord-port, or with environment variable DMTCP_COORD_PORT.\n"
  80   "              If no port is specified, start coordinator at a random port\n"
  81   "              (same as specifying port '0').\n"
  82   "  -i, --interval SECONDS (environment variable DMTCP_CHECKPOINT_INTERVAL)\n"
  83   "              Time in seconds between automatic checkpoints.\n"
  84   "              0 implies never (manual ckpt only); if not set and no env var,\n"
  85   "              use default value set in dmtcp_coordinator or dmtcp_command.\n"
  86   "              Not allowed if --join is specified\n"
  87   "\n"
  88   "Other options:\n"
  89   "  --run-as-root\n"
  90   "              Allow root to run dmtcp_restart and disable uid checking.\n"
  91   "              (default: disabled)\n"
  92   "  --no-strict-uid-checking\n"
  93   "              Disable uid checking for the checkpoint image. This allows\n"
  94   "              the checkpoint image to be restarted by a different user\n"
  95   "              than the one that created it.\n"
  96   "              (environment variable DMTCP_DISABLE_UID_CHECKING)\n"
  97   "  --ckptdir (environment variable DMTCP_CHECKPOINT_DIR):\n"
  98   "              Directory to store checkpoint images\n"
  99   "              (default: use the same directory used in previous checkpoint)\n"
 100   "  --tmpdir PATH (environment variable DMTCP_TMPDIR)\n"
 101   "              Directory to store temporary files (default: $TMDPIR or /tmp)\n"
 102   "  -q, --quiet (or set environment variable DMTCP_QUIET = 0, 1, or 2)\n"
 103   "              Skip NOTE messages; if given twice, also skip WARNINGs\n"
 104   "  --help\n"
 105   "              Print this message and exit.\n"
 106   "  --version\n"
 107   "              Print version information and exit.\n"
 108   "\n"
HELP_AND_CONTACT_INFO
 110   "\n"
 111 ;
 113 class RestoreTarget;
 115 typedef map<UniquePid, RestoreTarget*> RestoreTargetMap;
RestoreTargetMap targets;
RestoreTargetMap independentProcessTreeRoots;
 118 bool noStrictUIDChecking = false;
 119 bool runAsRoot = false;
 120 static string thePortFile;
CoordinatorMode allowedModes = COORD_ANY;
 123 static void setEnvironFd();
 124 static void runMtcpRestart(int is32bitElf, int fd, ProcessInfo *pInfo);
 125 static int readCkptHeader(const string& path, ProcessInfo *pInfo);
 126 static int openCkptFileToRead(const string& path);
 128 class RestoreTarget
 129 {
 130   public:
 131     RestoreTarget(const string& path)
 132       : _path(path)
 133     {
JASSERT(jalib::Filesystem::FileExists(_path)) (_path)
 135         .Text ( "checkpoint file missing" );
_fd = readCkptHeader(_path, &_pInfo);
JTRACE("restore target") (_path) (_pInfo.numPeers()) (_pInfo.compGroup());
 139     }
 141     int fd() const { return _fd; }
 142     const UniquePid& upid() const { return _pInfo.upid(); }
pid_t pid() const { return _pInfo.pid(); }
pid_t sid() const { return _pInfo.sid(); }
 145     bool isRootOfProcessTree() const {
 146       return _pInfo.isRootOfProcessTree();
 147     }
string procname() { return _pInfo.procname(); }
UniquePid compGroup() { return _pInfo.compGroup(); }
 150     int numPeers() { return _pInfo.numPeers(); }
 151     bool noCoordinator() { return _pInfo.noCoordinator(); }
 153     void restoreGroup()
 154     {
 155       if (_pInfo.isGroupLeader()) {
 156         // create new Group where this process becomes a leader
JTRACE("Create new Group.");
setpgid(0, 0);
 159       }
 160     }
 162     void createDependentChildProcess()
 163     {
pid_t pid = fork();
JASSERT(pid != -1);
 166       if (pid != 0) {
 167         return;
 168       }
createProcess();
 170     }
 172     void createDependentNonChildProcess()
 173     {
pid_t pid = fork();
JASSERT(pid != -1);
 176       if (pid == 0) {
pid_t gchild = fork();
JASSERT(gchild != -1);
 179         if (gchild != 0) {
exit(0);
 181         }
createProcess();
 183       } else {
JASSERT(waitpid(pid, NULL, 0) == pid);
 185       }
 186     }
 188     void createProcess(bool createIndependentRootProcesses = false)
 189     {
UniquePid::ThisProcess() = _pInfo.upid();
UniquePid::ParentProcess() = _pInfo.uppid();
Util::initializeLogFile(_pInfo.procname());
 194       if (createIndependentRootProcesses) {
DmtcpUniqueProcessId compId = _pInfo.compGroup().upid();
CoordinatorInfo coordInfo;
 197         struct in_addr localIPAddr;
 198         if (_pInfo.noCoordinator()) {
allowedModes = COORD_NONE;
 200         }
 202         // dmtcp_restart sets ENV_VAR_NAME_HOST/PORT, even if cmd line flag used
 203         const char *host = NULL;
 204         int port = UNINITIALIZED_PORT;
Util::getCoordHostAndPort(allowedModes, &host, &port);
 206         // FIXME:  We will use the new HOST and PORT here, but after restart,,
 207         //           we will use the old HOST and PORT from the ckpt image.
CoordinatorAPI::instance().connectToCoordOnRestart(allowedModes,
_pInfo.procname(),
_pInfo.compGroup(),
_pInfo.numPeers(),
 212                                                            &coordInfo,
host,
port,
 215                                                            &localIPAddr);
 216         // If port was 0, we'll get new random port when coordinator starts up.
Util::getCoordHostAndPort(allowedModes, &host, &port);
Util::writeCoordPortToFile(port, thePortFile.c_str());
string installDir =
jalib::Filesystem::DirName(jalib::Filesystem::GetProgramDir());
 223 #if defined(__i386__) || defined(__arm__)
 224         if (Util::strEndsWith(installDir, "/lib/dmtcp/32")) {
 225           // If dmtcp_launch was compiled for 32 bits in 64-bit O/S, then note:
 226           // DMTCP_ROOT/bin/dmtcp_launch is a symbolic link to:
 227           //    DMTCP_ROOT/bin/dmtcp_launch/lib/dmtcp/32/bin
 228           // GetProgramDir() followed the link.  So, need to remove the suffix.
 229           char *str = const_cast<char*>(installDir.c_str());
str[strlen(str) - strlen("/lib/dmtcp/32")] = '\0';
installDir = str;
 232         }
 233 #endif
 235         /* We need to initialize SharedData here to make sure that it is
 236          * initialized with the correct coordinator timestamp.  The coordinator
 237          * timestamp is updated only during postCkpt callback. However, the
 238          * SharedData area may be initialized earlier (for example, while
 239          * recreating threads), causing it to use *older* timestamp.
 240          */
SharedData::initialize(tmpDir.c_str(),
installDir.c_str(),
 243                                &compId,
 244                                &coordInfo,
 245                                &localIPAddr);
Util::prepareDlsymWrapper();
 248       }
JTRACE("Creating process during restart") (upid()) (_pInfo.procname());
RestoreTargetMap::iterator it;
 253       for (it = targets.begin(); it != targets.end(); it++) {
RestoreTarget *t = it->second;
 255         if (_pInfo.upid() == t->_pInfo.upid()) {
 256           continue;
 257         } else if (_pInfo.isChild(t->upid()) &&
t->_pInfo.sid() != _pInfo.pid()) {
t->createDependentChildProcess();
 260         }
 261       }
 263       if (createIndependentRootProcesses) {
RestoreTargetMap::iterator it;
 265         for (it = independentProcessTreeRoots.begin();
it != independentProcessTreeRoots.end();
it++) {
RestoreTarget *t = it->second;
 269           if (t != this) {
t->createDependentNonChildProcess();
 271           }
 272         }
 273       }
 275       // If we were the session leader, become one now.
 276       if (_pInfo.sid() == _pInfo.pid()) {
 277         if (getsid(0) != _pInfo.pid()) {
JWARNING(setsid() != -1) (getsid(0)) (JASSERT_ERRNO)
 279             .Text("Failed to restore this process as session leader.");
 280         }
 281       }
 283       // Now recreate processes with sid == _pid
 284       for (it = targets.begin(); it != targets.end(); it++) {
RestoreTarget *t = it->second;
 286         if (_pInfo.upid() == t->_pInfo.upid()) {
 287           continue;
 288         } else if (t->_pInfo.sid() == _pInfo.pid()) {
 289           if (_pInfo.isChild(t->upid())) {
t->createDependentChildProcess();
 291           } else if (t->isRootOfProcessTree()) {
t->createDependentNonChildProcess();
 293           }
 294         }
 295       }
 297       // Now close all open fds except _fd;
 298       for (it = targets.begin(); it != targets.end(); it++) {
RestoreTarget *t = it->second;
 300         if (t != this) {
close(t->fd());
 302         }
 303       }
string ckptDir = jalib::Filesystem::GetDeviceName(PROTECTED_CKPT_DIR_FD);
 306       if (ckptDir.length() == 0) {
 307         // Create the ckpt-dir fd so that the restarted process can know about
 308         // the abs-path of ckpt-image.
string dirName = jalib::Filesystem::DirName(_path);
 310         int dirfd = open(dirName.c_str(), O_RDONLY);
JASSERT(dirfd != -1) (JASSERT_ERRNO);
 312         if (dirfd != PROTECTED_CKPT_DIR_FD) {
JASSERT(dup2(dirfd, PROTECTED_CKPT_DIR_FD) == PROTECTED_CKPT_DIR_FD);
close(dirfd);
 315         }
 316       }
 318       if (!createIndependentRootProcesses) {
 319         // dmtcp_restart sets ENV_VAR_NAME_HOST/PORT, even if cmd line flag used
 320         const char *host = NULL;
 321         int port = UNINITIALIZED_PORT;
 322         int *port_p = &port;
Util::getCoordHostAndPort(allowedModes, &host, port_p);
CoordinatorAPI::instance().connectToCoordOnRestart(allowedModes,
_pInfo.procname(),
_pInfo.compGroup(),
_pInfo.numPeers(),
NULL,
host,
port,
NULL);
 332       }
setEnvironFd();
 335       int is32bitElf = 0;
 337 #if defined(__x86_64__) || defined(__aarch64__)
is32bitElf = (_pInfo.elfType() == ProcessInfo::Elf_32);
 339 #elif defined(__i386__) || defined(__arm__)
is32bitElf = true;
 341 #endif
runMtcpRestart(is32bitElf, _fd, &_pInfo);
JASSERT ( false ).Text ( "unreachable" );
 345     }
 347   private:
string _path;
ProcessInfo _pInfo;
 350     int _fd;
 351 };
 353 static void runMtcpRestart(int is32bitElf, int fd, ProcessInfo *pInfo)
 354 {
 355   char fdBuf[8];
 356   char stderrFdBuf[8];
sprintf(fdBuf, "%d", fd);
sprintf(stderrFdBuf, "%d", PROTECTED_STDERR_FD);
 360 #ifdef HAS_PR_SET_PTRACER
 361   if (getenv("DMTCP_GDB_ATTACH_ON_RESTART")) {
JNOTE("\n     *******************************************************\n"
 363           "     *** Environment variable, DMTCP_GDB_ATTACH_ON_RESTART is set\n"
 364           "     *** You can attach to the running process as follows:\n"
 365           "     ***     gdb _PROGRAM_NAME_ PID  [See below for PID.]\n"
 366           "     *** NOTE:  This mode can be a security risk.\n"
 367           "     ***        Do not set the env. variable normally.\n"
 368           "     *******************************************************")
 369          (getpid());
prctl(PR_SET_PTRACER, PR_SET_PTRACER_ANY, 0, 0, 0); // Allow 'gdb attach'
 371   }
 372 #endif
 374   static string mtcprestart = Util::getPath ("mtcp_restart");
 376 #if defined(CONFIG_M32)
 377   if (is32bitElf) {
mtcprestart = Util::getPath("mtcp_restart-32", is32bitElf);
 379   }
 380 #endif
 382   char* const newArgs[] = {
 383     (char*) mtcprestart.c_str(),
 384     const_cast<char*> ("--fd"), fdBuf,
 385     const_cast<char*> ("--stderr-fd"), stderrFdBuf,
NULL
 387   };
execve (newArgs[0], newArgs, environ);
JASSERT (false) (newArgs[0]) (newArgs[1]) (JASSERT_ERRNO)
 391     .Text ("exec() failed");
 392 }
 394 // ************************ For reading checkpoint files *****************
 396 int readCkptHeader(const string& path, ProcessInfo *pInfo)
 397 {
 398   int fd = openCkptFileToRead(path);
 399   const size_t len = strlen(DMTCP_FILE_HEADER);
jalib::JBinarySerializeReaderRaw rdr("", fd);
pInfo->serialize(rdr);
size_t numRead = len + rdr.bytes();
 405   // We must read in multiple of PAGE_SIZE
 406   const ssize_t pagesize = Util::pageSize();
ssize_t remaining = pagesize - (numRead % pagesize);
 408   char buf[remaining];
JASSERT(Util::readAll(fd, buf, remaining) == remaining);
 410   return fd;
 411 }
 413 static char first_char(const char *filename)
 414 {
 415   int fd, rc;
 416   char c;
fd = open(filename, O_RDONLY);
JASSERT(fd >= 0) (filename) .Text("ERROR: Cannot open filename");
rc = read(fd, &c, 1);
JASSERT(rc == 1) (filename) .Text("ERROR: Error reading from filename");
close(fd);
 425   return c;
 426 }
 428 // Copied from mtcp/mtcp_restart.c.
 429 // Let's keep this code close to MTCP code to avoid maintenance problems.
 430 // MTCP code in:  mtcp/mtcp_restart.c:open_ckpt_to_read()
 431 // A previous version tried to replace this with popen, causing a regression:
 432 //   (no call to pclose, and possibility of using a wrong fd).
 433 // Returns fd;
 434 static int open_ckpt_to_read(const char *filename)
 435 {
 436   int fd;
 437   int fds[2];
 438   char fc;
 439   const char *decomp_path;
 440   const char **decomp_args;
 441   const char *gzip_path = "gzip";
 442   static const char * gzip_args[] = {
 443     const_cast<char*> ("gzip"),
 444     const_cast<char*> ("-d"),
 445     const_cast<char*> ("-"),
NULL
 447   };
 448 #ifdef HBICT_DELTACOMP
 449   const char *hbict_path = const_cast<char*> ("hbict");
 450   static const char *hbict_args[] = {
 451     const_cast<char*> ("hbict"),
 452     const_cast<char*> ("-r"),
NULL
 454   };
 455 #endif
pid_t cpid;
fc = first_char(filename);
fd = open(filename, O_RDONLY);
JASSERT(fd>=0)(filename).Text("Failed to open file.");
 462   if (fc == DMTCP_MAGIC_FIRST) { /* no compression */
 463     return fd;
 464   }
 465   else if (fc == GZIP_FIRST
 466 #ifdef HBICT_DELTACOMP
 467            || fc == HBICT_FIRST
 468 #endif
 469           ) {
 470     if (fc == GZIP_FIRST) {
decomp_path = gzip_path;
decomp_args = gzip_args;
 473     }
 474 #ifdef HBICT_DELTACOMP
 475     else {
decomp_path = hbict_path;
decomp_args = hbict_args;
 478     }
 479 #endif
JASSERT(pipe(fds) != -1) (filename)
 482       .Text("Cannot create pipe to execute gunzip to decompress ckpt file!");
cpid = fork();
JASSERT(cpid != -1)
 487       .Text("ERROR: Cannot fork to execute gunzip to decompress ckpt file!");
 488     if (cpid > 0) { /* parent process */
JTRACE("created child process to uncompress checkpoint file") (cpid);
close(fd);
close(fds[1]);
 492       // Wait for child process
JASSERT(waitpid(cpid, NULL, 0) == cpid);
 494       return fds[0];
 495     } else { /* child process */
 496       /* Fork a grandchild process and kill the parent. This way the grandchild
 497        * process never becomes a zombie.
 498        *
 499        * Sometimes dmtcp_restart is called with multiple ckpt images. In that
 500        * situation, the dmtcp_restart process creates gzip processes and only
 501        * later forks mtcp_restart processes. The gzip processes can not be
 502        * wait()'d upon by the corresponding mtcp_restart processes because
 503        * their parent is the original dmtcp_restart process and thus they
 504        * become zombie.
 505        */
cpid = fork();
JASSERT(cpid != -1);
 508       if (cpid > 0) {
 509         // Use _exit() instead of exit() to avoid popping atexit() handlers
 510         // registered by the parent process.
_exit(0);
 512       }
 514       // Grandchild process
JTRACE ( "child process, will exec into external de-compressor");
fd = dup(dup(dup(fd)));
fds[1] = dup(fds[1]);
close(fds[0]);
JASSERT(fd != -1);
JASSERT(dup2(fd, STDIN_FILENO) == STDIN_FILENO);
close(fd);
JASSERT(dup2(fds[1], STDOUT_FILENO) == STDOUT_FILENO);
close(fds[1]);
execvp(decomp_path, (char **)decomp_args);
JASSERT(decomp_path!=NULL) (decomp_path)
 526         .Text("Failed to launch gzip.");
 527       /* should not get here */
JASSERT(false)
 529         .Text("Decompression failed!  No restoration will be performed!");
 530     }
 531   } else { /* invalid magic number */
JASSERT(false)
 533       .Text("ERROR: Invalid magic number in this checkpoint file!");
 534   }
 535   return -1;
 536 }
 538 // See comments above for open_ckpt_to_read()
 539 int openCkptFileToRead(const string& path)
 540 {
 541   char buf[1024];
 542   int fd = open_ckpt_to_read(path.c_str());
 543   // The rest of this function is for compatibility with original definition.
JASSERT(fd >= 0) (path) .Text("Failed to open file.");
 545   const int len = strlen(DMTCP_FILE_HEADER);
JASSERT(read(fd, buf, len) == len)(path) .Text("read() failed");
 547   if (strncmp(buf, DMTCP_FILE_HEADER, len) == 0) {
JTRACE("opened checkpoint file [uncompressed]")(path);
 549   } else {
close(fd);
fd = open_ckpt_to_read(path.c_str()); /* Re-open from beginning */
JASSERT(fd >= 0) (path) .Text("Failed to open file.");
 553   }
 554   return fd;
 555 }
 556 // ************************ End of for reading checkpoint files *************
 559 static void setEnvironFd()
 560 {
 561   char envFile[PATH_MAX];
sprintf(envFile, "%s/envFile.XXXXXX", tmpDir.c_str());
 563   int fd = mkstemp(envFile);
JASSERT(fd != -1) (envFile) (JASSERT_ERRNO);
JASSERT(unlink(envFile) == 0) (envFile) (JASSERT_ERRNO);
JASSERT(dup2(fd, PROTECTED_ENVIRON_FD) == PROTECTED_ENVIRON_FD)
 567     (JASSERT_ERRNO);
JASSERT(close(fd) == 0);
fd = PROTECTED_ENVIRON_FD;
 571   char **env = environ;
 572   while (*env != NULL) {
Util::writeAll(fd, *env, strlen(*env) + 1); // Also write null character
env++;
 575   }
Util::writeAll(fd, *env, 1); // Write final null character
 577 }
 579 static void setNewCkptDir(char *path)
 580 {
 581   struct stat st;
 582   if (stat(path, &st) == -1) {
JASSERT(mkdir(path, S_IRWXU) == 0 || errno == EEXIST)
 584       (JASSERT_ERRNO) (path)
 585       .Text("Error creating checkpoint directory");
JASSERT(0 == access(path, X_OK|W_OK)) (path)
 587       .Text("ERROR: Missing execute- or write-access to checkpoint dir");
 588   } else {
JASSERT(S_ISDIR(st.st_mode)) (path) .Text("ckptdir not a directory");
 590   }
 592   int fd = open(path, O_RDONLY);
JASSERT(fd != -1) (path);
JASSERT(dup2(fd, PROTECTED_CKPT_DIR_FD) == PROTECTED_CKPT_DIR_FD)
 595     (fd) (path);
 596   if (fd != PROTECTED_CKPT_DIR_FD) {
close(fd);
 598   }
 599 }
 601 //shift args
 602 #define shift argc--,argv++
 604 int main(int argc, char** argv)
 605 {
 606   char *tmpdir_arg = NULL;
initializeJalib();
 610   if (!getenv(ENV_VAR_QUIET)) {
setenv(ENV_VAR_QUIET, "0", 0);
 612   }
 614   if (getenv(ENV_VAR_DISABLE_UID_CHECKING)) {
noStrictUIDChecking = true;
 616   }
 618   if (argc == 1) {
printf("%s", DMTCP_VERSION_AND_COPYRIGHT_INFO);
printf("(For help: %s --help)\n\n", argv[0]);
 621     return DMTCP_FAIL_RC;
 622   }
 624   //process args
shift;
 626   while (true) {
string s = argc>0 ? argv[0] : "--help";
 628     if (s == "--help" && argc == 1) {
printf("%s", theUsage);
 630       return DMTCP_FAIL_RC;
 631     } else if ((s == "--version") && argc == 1) {
printf("%s", DMTCP_VERSION_AND_COPYRIGHT_INFO);
 633       return DMTCP_FAIL_RC;
 634     } else if (s == "-j" || s == "--join") {
allowedModes = COORD_JOIN;
shift;
 637     } else if (s == "--new-coordinator") {
allowedModes = COORD_NEW;
shift;
 640     } else if (s == "--run-as-root") {
runAsRoot = true;
shift;
 643     } else if (s == "--no-strict-uid-checking") {
noStrictUIDChecking = true;
shift;
 646     } else if (s == "-i" || s == "--interval") {
setenv(ENV_VAR_CKPT_INTR, argv[1], 1);
shift; shift;
 649     } 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;
 653     } else if (argc > 1 && (s == "-h" || s == "--coord-host" || s == "--host")){
setenv(ENV_VAR_NAME_HOST, argv[1], 1);
shift; shift;
 656     } else if (argc>1 && (s == "-p" || s == "--coord-port" || s == "--port")) {
setenv(ENV_VAR_NAME_PORT, argv[1], 1);
shift; shift;
 659     } 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;
 663     } else if (argc>1 && s == "--port-file"){
thePortFile = argv[1];
shift; shift;
 666     } else if (argc > 1 && (s == "-c" || s == "--ckptdir")) {
setNewCkptDir(argv[1]);
shift; shift;
 669     } else if (argc > 1 && (s == "-t" || s == "--tmpdir")) {
tmpdir_arg = argv[1];
shift; shift;
 672     } else if (s == "-q" || s == "--quiet") {
 673       *getenv(ENV_VAR_QUIET) = *getenv(ENV_VAR_QUIET) + 1;
 674       // Just in case a non-standard version of setenv is being used:
setenv(ENV_VAR_QUIET, getenv(ENV_VAR_QUIET), 1);
shift;
 677     } else if ((s.length() > 2 && s.substr(0, 2) == "--") ||
 678                (s.length() > 1 && s.substr(0, 1) == "-")) {
printf("Invalid Argument\n%s", theUsage);
 680       return DMTCP_FAIL_RC;
 681     } else if (argc > 1 && s == "--") {
shift;
 683       break;
 684     } else {
 685       break;
 686     }
 687   }
tmpDir = Util::calcTmpDir(tmpdir_arg);
jassert_quiet = *getenv(ENV_VAR_QUIET) - '0';
 693   //make sure JASSERT initializes now, rather than during restart
Util::initializeLogFile(tmpDir);
 696   if (!runAsRoot && (getuid() == 0 || geteuid() == 0)) {
JASSERT_STDERR <<
 698       "Running dmtcp_restart as root is dangerous.  Aborting.\n"
 699       "If you still want to do this (at your own risk), then use\n" \
 700       "    dmtcp_restart --run-as-root\n";
exit(0);
 702   }
JTRACE("New dmtcp_restart process; _argc_ ckpt images") (argc);
 706   bool doAbort = false;
 707   for (; argc > 0; shift) {
string restorename(argv[0]);
 709     struct stat buf;
 710     int rc = stat(restorename.c_str(), &buf);
 711     if (Util::strEndsWith(restorename, "_files")) {
 712       continue;
 713     } else if (!Util::strEndsWith(restorename, ".dmtcp")) {
JNOTE("File doesn't have .dmtcp extension. Check Usage.")
 715         (restorename);
JASSERT_STDERR << theUsage;
doAbort = true;
 718     } else if (rc == -1) {
 719       char error_msg[1024];
sprintf(error_msg, "\ndmtcp_restart: ckpt image %s", restorename.c_str());
perror(error_msg);
doAbort = true;
 723     } else if (buf.st_uid != getuid() && !noStrictUIDChecking && !runAsRoot) {
 724       /*Could also run if geteuid() matches*/
printf("\nProcess uid (%d) doesn't match uid (%d) of\n" \
 726              "checkpoint image (%s).\n" \
 727              "This is dangerous.  Aborting for security reasons.\n" \
 728              "If you still want to do this (at your own risk),\n" \
 729              "  then modify dmtcp/src/%s:%d and re-compile.\n",
getuid(), buf.st_uid, restorename.c_str(), __FILE__, __LINE__ - 7);
doAbort = true;
 732     }
 733     if (doAbort) {
exit(DMTCP_FAIL_RC);
 735     }
JTRACE("Will restart ckpt image") (argv[0]);
RestoreTarget *t = new RestoreTarget(argv[0]);
targets[t->upid()] = t;
 740   }
 742   // Prepare list of independent process tree roots
RestoreTargetMap::iterator i;
 744   for (i = targets.begin(); i != targets.end(); i++) {
RestoreTarget *t1 = i->second;
 746     if (t1->isRootOfProcessTree()) {
RestoreTargetMap::iterator j;
 748       for (j = targets.begin(); j != targets.end(); j++) {
RestoreTarget *t2 = j->second;
 750         if (t1 == t2) continue;
 751         if (t1->sid() == t2->pid()) {
 752           break;
 753         }
 754       }
 755       if (j == targets.end()) {
independentProcessTreeRoots[t1->upid()] = t1;
 757       }
 758     }
 759   }
JASSERT(independentProcessTreeRoots.size() > 0)
 761     .Text("There must be at least one process tree that doesn't have\n"
 762           "  a different process as session leader.");
WorkerState::setCurrentState(WorkerState::RESTARTING);
RestoreTarget *t = independentProcessTreeRoots.begin()->second;
JASSERT(t->pid() != 0);
JASSERT(!t->noCoordinator() || allowedModes == COORD_ANY)
 769     .Text("Process had no coordinator prior to checkpoint;\n"
 770           "  but either --join or --new-coordinator was specified.");
t->createProcess(true);
JASSERT(false).Text("unreachable");
 773   return -1;
 774 }