Fix foreground process group on restart

     * Sometimes 'dmtcp_launch' is called from inside a shell script
       (e.g., for a batch job).  In this case, on restart, the foreground
       process group leader is the restarted process, and yet the
       process group leader is the parent (presumably the login shell).
       So, the foreground process group leader is not the leader of a
       process group.  To fix this, we set the process group leader to
       also be the leader of the foreground process group.

src/terminal.cpp

@@ -7,6 +7,7 @@
 #endif  // ifdef HAS_PR_SET_PTRACER
 #include "../jalib/jassert.h"
 #include "config.h"
+#include "syscallwrappers.h"
 #include "dmtcp.h"
 
 /*************************************************************************
@@ -60,18 +61,57 @@ safe_tcsetattr(int fd, int optional_actions, const struct termios *termios_p)
   return 0;
 }
 
-// FIXME: Handle Virtual Pids
+static int
+get_parent_pid(int pid)
+{
+  char buf[8192];
+  // /proc/PID/stat is:  PID (COMMAND) STATE PPID ...
+  snprintf(buf, sizeof(buf), "/proc/%d/stat", pid);
+  FILE* fp = fopen(buf, "r");
+  JASSERT(fp != NULL)(pid)(JASSERT_ERRNO).Text("fopen");
+  int rc = -1;
+  if (fp) {
+    size_t size = fread(buf, 1, sizeof(buf), fp);
+    if (size > 0 && strstr(buf, ") ")) {
+      rc = strtol(strstr(strstr(buf, ") ")+2, " "), NULL, 10);
+    }
+  }
+  JASSERT(rc != -1)(pid).Text("parent pid not found in /proc/*/stat");
+  return rc;
+}
+
+// See:  info libc -> Job Control -> Implementing a Shell ->
+//                                               Foreground and Background
 static void
 restore_term_settings()
 {
   if (saved_termios_exists) {
     /* First check if we are in foreground. If not, skip this and print
      *   warning.  If we try to call tcsetattr in background, we will hang up.
+     * In some shells, the shell itself will be a pgrp that includes is
+     *   first child process, instead of the first child forming a pgrp.
      */
-    int foreground = (tcgetpgrp(STDIN_FILENO) == getpgrp());
+    int foreground = (_real_tcgetpgrp(STDIN_FILENO) == _libc_getpgrp() ||
+                      get_parent_pid(_real_tcgetpgrp(STDIN_FILENO)) ==
+                        _libc_getpgrp());
     JTRACE("restore terminal attributes, check foreground status first")
-      (foreground);
+          (foreground);
     if (foreground) {
+      if (get_parent_pid(_real_tcgetpgrp(STDIN_FILENO)) == _libc_getpgrp()) {
+        /* At ckpt time, maybe a shell script called dmtcp_launch.
+         *   Now, at restart, we are running directly under the login shell,
+         *   and we are foreground process group leader, not the login shell.
+         *   Normally, the foreground process group leader is also
+         *   a process group leader.  But this broke, during restart.
+         *   So, we will set a new foregroup process group leader to fix this.
+         */
+        tcsetpgrp(0, getpgrp()); // Set login shell (not under DMTCP control)
+                                 //   to be the foreground process, instead of
+                                 //   this process.
+        JTRACE("The process group ID of this process is the parent of\n"
+               "  the process group ID for the foreground process group.\n"
+               "Make current process group the new foreground process group.");
+      }
       if ((!isatty(STDIN_FILENO)
            || safe_tcsetattr(STDIN_FILENO, TCSANOW, &saved_termios) == -1)) {
         JWARNING(false).Text("failed to restore terminal");

test/misc/README

@@ -4,15 +4,11 @@ Some of these special cases could be combined into autotest.py, and
 
 Examples of special cases are:
 * All CPU architectures:  x86, x86_64, arm, aarch64
-* The two major branches of distros:  Debian/Ubuntu and
-                                          RedHat/Fedora/CoreOS
+* The three major branches of distros:  Debian/Ubuntu, and
+                                          RedHat/Fedora/CoreOS, OpenSUSE/SUSE
 * 32-bit Linux:  native, --enable-m32, and multi-arch  (see multi-arch.sh)
-* Builds of DMTCP with other compilers:  icc LLVM/clang
-* musl libc (and Alpine Linux) (?)
 * 'make install' for native 64-bit and each of the 32-bit Linux combinations
 * for file in test/plugin/*; do (cd $file && make check); done
-* dmtcp_launch invoked in extra shell:  #!/bin/sh / dmtcp_launth test/dmtcp1
-                                        [ and test restart: foreground issue ]
 * forked checkpointing
 * fast restart
 * --no-gzip
@@ -30,9 +26,10 @@ Examples of special cases are:
   *  Network:  InfiniBand (CM and UD), TCP
   +  Shared filesystems:  NFS, Lustre
 * openmp and Intel TBB
-* MANA for MPI and CRAC for CUDA
-* SOON:  OpenGL (?)
+* MIC
+* SOON:  OpenGL
 * Widely used commercial software:  Matlab, ...
 * Non-default plugins:
-  modify-env, path-virt
+  --ptrace, modify-env
   test/plugin/{applic-inititated-ckdpt,applic-delayed-ckpt}
+* Builds of DMTCP with other compilers:  icc LLVM/Clang

util/gdb-dmtcp-utils

@@ -169,7 +169,7 @@ class ShowFilenameAtAddress(gdb.Command):
         if getpid() == 0:
           gdb.execute('print "Process not yet started"', False, False)
         else:
-          memory_region = "%s (r-x): 0x%x-0x%x" % \
+          memory_region = "%s (r[-w]x): 0x%x-0x%x" % \
                           memory_region_at_address(memory_address)
           gdb.execute('print "' + memory_region + '"', False, False)
 # This will add the new gdb command: show-filename-at-address MEMORY_ADDRESS
@@ -352,6 +352,7 @@ def procmap_filename_address(line):
   return (triple[-1],) + \
          tuple([int("0x"+elt, 16) for elt in triple[0].split("-")])
 def is_text_segment(procmap_line):
+  print(procmap_line)
   return "r-x" in procmap_line and \
          (procmap_line == "" or procmap_line.isspace() or '/' in procmap_line)
 def memory_regions():