int main(int argc, char **argv) {
int i, cleanupd_fd;
struct tempfile *stdout_log, *stderr_log, *singularity_debug;
struct image_object image;
pid_t child;
siginfo_t siginfo;
struct stat filestat;
singularity_config_init();
singularity_suid_init();
singularity_priv_init();
singularity_registry_init();
singularity_priv_drop();
singularity_runtime_autofs();
singularity_registry_set("UNSHARE_PID", "1");
singularity_registry_set("UNSHARE_IPC", "1");
if ( singularity_registry_get("INSTANCE_BOOT") != NULL ) {
singularity_registry_set("CONTAIN", "1");
}
singularity_cleanupd();
if ( singularity_registry_get("WRITABLE") != NULL ) {
singularity_message(VERBOSE3, "Instantiating writable container image object\n");
image = singularity_image_init(singularity_registry_get("IMAGE"), O_RDWR);
} else {
singularity_message(VERBOSE3, "Instantiating read only container image object\n");
image = singularity_image_init(singularity_registry_get("IMAGE"), O_RDONLY);
}
singularity_runtime_ns(SR_NS_ALL);
singularity_sessiondir();
singularity_image_mount(&image, CONTAINER_MOUNTDIR);
action_ready();
singularity_runtime_overlayfs();
singularity_runtime_mounts();
singularity_runtime_files();
/* After this point, we are running as PID 1 inside PID NS */
singularity_message(DEBUG, "Preparing sinit daemon\n");
singularity_registry_set("ROOTFS", CONTAINER_FINALDIR);
singularity_daemon_init();
singularity_message(DEBUG, "Entering chroot environment\n");
singularity_runtime_enter();
singularity_priv_drop_perm();
if ( envclean() != 0 ) {
singularity_message(ERROR, "Failed sanitizing the environment\n");
ABORT(255);
}
if ( chdir("/") < 0 ) {
singularity_message(ERROR, "Can't change directory to /\n");
}
setsid();
umask(0);
cleanupd_fd = atoi(singularity_registry_get("CLEANUPD_FD"));
if ( singularity_registry_get("INSTANCE_BOOT") != NULL ) {
int pipes[2];
if ( pipe2(pipes, O_CLOEXEC) < 0 ) {
singularity_signal_go_ahead(255);
return(0);
}
if ( fork() == 0 ) {
/* wait a broken pipe which mean exec success */
struct pollfd pfd;
pfd.fd = pipes[0];
pfd.events = POLLRDHUP;
close(pipes[1]);
while( poll(&pfd, 1, 1000) >= 0 ) {
if ( pfd.revents == POLLHUP ) break;
}
singularity_signal_go_ahead(0);
/* wait /sbin/init install signal handler */
usleep(20000);
return(0);
} else {
close(pipes[0]);
if ( is_exec("/sbin/init") == 0 ) {
argv[1] = NULL;
if ( execv("/sbin/init", argv) < 0 ) { // Flawfinder: ignore
singularity_message(ERROR, "Exec of /sbin/init failed\n");
}
} else {
singularity_message(ERROR, "/sbin/init not present in container\n");
}
/* send exit status and implicitly kill polling child */
singularity_signal_go_ahead(255);
return(0);
}
}
/* set program name */
if ( prctl(PR_SET_NAME, "sinit", 0, 0, 0) < 0 ) {
singularity_message(ERROR, "Failed to set program name\n");
ABORT(255);
}
singularity_install_signal_handler();
/* Close all open fd's that may be present besides daemon info file fd */
singularity_message(DEBUG, "Closing open fd's\n");
for( i = sysconf(_SC_OPEN_MAX); i > 2; i-- ) {
if ( i != cleanupd_fd ) {
if ( fstat(i, &filestat) == 0 ) {
if ( S_ISFIFO(filestat.st_mode) != 0 ) {
continue;
}
}
close(i);
}
}
singularity_debug = make_logfile("singularity-debug");
stdout_log = make_logfile("stdout");
stderr_log = make_logfile("stderr");
for( i = 0; i <= 2; i++ ) {
close(i);
}
if ( chdir("/") < 0 ) {
singularity_message(ERROR, "Can't change directory to /\n");
}
setsid();
umask(0);
/* set program name */
if ( prctl(PR_SET_NAME, "sinit", 0, 0, 0) < 0 ) {
singularity_message(ERROR, "Failed to set program name\n");
ABORT(255);
}
child = fork();
if ( child == 0 ) {
/* Make standard output and standard error files to log stdout & stderr into */
if ( stdout_log != NULL ) {
if ( -1 == dup2(stdout_log->fd, 1) ) {
singularity_message(ERROR, "Unable to dup2(): %s\n", strerror(errno));
ABORT(255);
}
}
if ( stderr_log != NULL ) {
if ( -1 == dup2(stderr_log->fd, 2) ) {
singularity_message(ERROR, "Unable to dup2(): %s\n", strerror(errno));
ABORT(255);
}
}
/* Unblock signals and execute startscript */
singularity_unblock_signals();
if ( is_exec("/.singularity.d/actions/start") == 0 ) {
singularity_message(DEBUG, "Exec'ing /.singularity.d/actions/start\n");
if ( execv("/.singularity.d/actions/start", argv) < 0 ) { // Flawfinder: ignore
singularity_message(ERROR, "Failed to execv() /.singularity.d/actions/start: %s\n", strerror(errno));
ABORT(CHILD_FAILED);
}
} else {
singularity_message(VERBOSE, "Instance start script not found\n");
kill(1, SIGCONT);
}
} else if ( child > 0 ) {
if ( singularity_debug != NULL ) {
if ( -1 == dup2(singularity_debug->fd, 2) ) {
singularity_message(ERROR, "Unable to dup2(): %s\n", strerror(errno));
ABORT(255);
}
}
singularity_message(DEBUG, "Waiting for signals\n");
/* send a SIGALRM if start script doesn't send SIGCONT within 1 seconds */
alarm(1);
while (1) {
if ( singularity_handle_signals(&siginfo) < 0 ) {
singularity_signal_go_ahead(255);
break;
}
if ( siginfo.si_signo == SIGCHLD ) {
singularity_message(DEBUG, "Child exited\n");
if ( siginfo.si_pid == 2 && siginfo.si_status == CHILD_FAILED ) {
singularity_signal_go_ahead(CHILD_FAILED);
break;
}
} else if ( siginfo.si_signo == SIGCONT && siginfo.si_pid == 2 ) {
/* start script correctly exec */
singularity_signal_go_ahead(0);
started = 1;
} else if ( siginfo.si_signo == SIGALRM && started == 0 ) {
/* don't receive SIGCONT, start script modified/replaced ? */
singularity_message(ERROR, "Start script doesn't send SIGCONT\n");
singularity_signal_go_ahead(255);
break;
}
}
} else {
singularity_message(ERROR, "Failed to execute start script\n");
singularity_signal_go_ahead(255);
}
return(0);
}
static int setup_container(spank_t spank)
{
int rc;
struct image_object image;
char *command = NULL;
if ((rc = setup_container_environment(spank)) != 0) { return rc; }
/*
* Ugg, singularity_* calls tend to call ABORT(255), which translates to
* exit(255), all over the place. The slurm SPANK hook API may not
* expect such sudden death of the pending slurm task. I've left
* a bunch of following "return rc;" commented out, as the failure
* conditions from singularity_* calls isn't clear to me.
*/
// Before we do anything, check privileges and drop permission
singularity_priv_init();
singularity_priv_drop();
singularity_message(VERBOSE, "Running Slurm/Singularity integration "
"plugin\n");
if ((rc = singularity_config_init()) != 0) {
return rc;
}
singularity_priv_init();
//TODO singularity_suid_init(argv);
singularity_registry_init();
singularity_priv_userns();
singularity_priv_drop();
singularity_cleanupd();
singularity_runtime_ns(SR_NS_ALL);
singularity_sessiondir();
image = singularity_image_init(singularity_registry_get("IMAGE"));
if ( singularity_registry_get("WRITABLE") == NULL ) {
singularity_image_open(&image, O_RDONLY);
} else {
singularity_image_open(&image, O_RDWR);
}
singularity_image_check(&image);
singularity_image_bind(&image);
singularity_image_mount(&image, singularity_runtime_rootfs(NULL));
action_ready(singularity_runtime_rootfs(NULL));
singularity_runtime_overlayfs();
singularity_runtime_mounts();
singularity_runtime_files();
singularity_runtime_enter();
singularity_runtime_environment();
singularity_priv_drop_perm();
if ((rc = setup_container_cwd()) < 0) {
singularity_message(ERROR, "Could not obtain current directory.\n");
return rc;
}
envar_set("SINGULARITY_CONTAINER", singularity_image_name(&image), 1); // Legacy PS1 support
envar_set("SINGULARITY_NAME", singularity_image_name(&image), 1);
envar_set("SINGULARITY_SHELL", singularity_registry_get("SHELL"), 1);
command = singularity_registry_get("COMMAND");
singularity_message(LOG, "USER=%s, IMAGE='%s', COMMAND='%s'\n", singularity_priv_getuser(), singularity_image_name(&image), singularity_registry_get("COMMAND"));
// At this point, the current process is in the runtime container environment.
// Return control flow back to Slurm: when execv is invoked, it'll be done from
// within the container.
return 0;
}
