+static pid_t
+fork_and_wait_for_startup(int *fdp)
+{
+ int fds[2];
+ pid_t pid;
+
+ if (pipe(fds) < 0) {
+ ovs_fatal(errno, "pipe failed");
+ }
+
+ pid = fork();
+ if (pid > 0) {
+ /* Running in parent process. */
+ char c;
+
+ close(fds[1]);
+ fatal_signal_fork();
+ if (read(fds[0], &c, 1) != 1) {
+ int retval;
+ int status;
+
+ do {
+ retval = waitpid(pid, &status, 0);
+ } while (retval == -1 && errno == EINTR);
+
+ if (retval == pid
+ && WIFEXITED(status)
+ && WEXITSTATUS(status)) {
+ /* Child exited with an error. Convey the same error to
+ * our parent process as a courtesy. */
+ exit(WEXITSTATUS(status));
+ }
+
+ ovs_fatal(errno, "fork child failed to signal startup");
+ }
+ close(fds[0]);
+ *fdp = -1;
+ } else if (!pid) {
+ /* Running in child process. */
+ close(fds[0]);
+ time_postfork();
+ lockfile_postfork();
+ *fdp = fds[1];
+ } else {
+ ovs_fatal(errno, "could not fork");
+ }
+
+ return pid;
+}
+
+static void
+fork_notify_startup(int fd)
+{
+ if (fd != -1) {
+ size_t bytes_written;
+ int error;
+
+ error = write_fully(fd, "", 1, &bytes_written);
+ if (error) {
+ ovs_fatal(error, "could not write to pipe");
+ }
+
+ close(fd);
+ }
+}
+
+static bool
+should_restart(int status)
+{
+ if (WIFSIGNALED(status)) {
+ static const int error_signals[] = {
+ SIGABRT, SIGALRM, SIGBUS, SIGFPE, SIGILL, SIGPIPE, SIGSEGV,
+ SIGXCPU, SIGXFSZ
+ };
+
+ size_t i;
+
+ for (i = 0; i < ARRAY_SIZE(error_signals); i++) {
+ if (error_signals[i] == WTERMSIG(status)) {
+ return true;
+ }
+ }
+ }
+ return false;
+}
+
+static void
+monitor_daemon(pid_t daemon_pid)
+{
+ /* XXX Should log daemon's stderr output at startup time. */
+ const char *saved_program_name;
+ time_t last_restart;
+ char *status_msg;
+ int crashes;
+
+ saved_program_name = program_name;
+ program_name = xasprintf("monitor(%s)", program_name);
+ status_msg = xstrdup("healthy");
+ last_restart = TIME_MIN;
+ crashes = 0;
+ for (;;) {
+ int retval;
+ int status;
+
+ proctitle_set("%s: monitoring pid %lu (%s)",
+ saved_program_name, (unsigned long int) daemon_pid,
+ status_msg);
+
+ do {
+ retval = waitpid(daemon_pid, &status, 0);
+ } while (retval == -1 && errno == EINTR);
+
+ if (retval == -1) {
+ ovs_fatal(errno, "waitpid failed");
+ } else if (retval == daemon_pid) {
+ char *s = process_status_msg(status);
+ if (should_restart(status)) {
+ free(status_msg);
+ status_msg = xasprintf("%d crashes: pid %lu died, %s",
+ ++crashes,
+ (unsigned long int) daemon_pid, s);
+ free(s);
+
+ if (WCOREDUMP(status)) {
+ /* Disable further core dumps to save disk space. */
+ struct rlimit r;
+
+ r.rlim_cur = 0;
+ r.rlim_max = 0;
+ if (setrlimit(RLIMIT_CORE, &r) == -1) {
+ VLOG_WARN("failed to disable core dumps: %s",
+ strerror(errno));
+ }
+ }
+
+ /* Throttle restarts to no more than once every 10 seconds. */
+ if (time(NULL) < last_restart + 10) {
+ VLOG_WARN("%s, waiting until 10 seconds since last "
+ "restart", status_msg);
+ for (;;) {
+ time_t now = time(NULL);
+ time_t wakeup = last_restart + 10;
+ if (now >= wakeup) {
+ break;
+ }
+ sleep(wakeup - now);
+ }
+ }
+ last_restart = time(NULL);
+
+ VLOG_ERR("%s, restarting", status_msg);
+ daemon_pid = fork_and_wait_for_startup(&daemonize_fd);
+ if (!daemon_pid) {
+ break;
+ }
+ } else {
+ VLOG_INFO("pid %lu died, %s, exiting",
+ (unsigned long int) daemon_pid, s);
+ free(s);
+ exit(0);
+ }
+ }
+ }
+ free(status_msg);
+
+ /* Running in new daemon process. */
+ proctitle_restore();
+ free((char *) program_name);
+ program_name = saved_program_name;
+}
+
+/* Close stdin, stdout, stderr. If we're started from e.g. an SSH session,
+ * then this keeps us from holding that session open artificially. */
+static void
+close_standard_fds(void)
+{
+ int null_fd = get_null_fd();
+ if (null_fd >= 0) {
+ dup2(null_fd, STDIN_FILENO);
+ dup2(null_fd, STDOUT_FILENO);
+ dup2(null_fd, STDERR_FILENO);
+ }
+}
+