/*
* Run a prolog or epilog script (does NOT drop privileges)
* name IN: class of program (prolog, epilog, etc.),
* path IN: pathname of program to run
* job_id IN: info on associated job
* max_wait IN: maximum time to wait in seconds, -1 for no limit
* env IN: environment variables to use on exec, sets minimal environment
* if NULL
* uid IN: user ID of job owner
* RET 0 on success, -1 on failure.
*/
static int
_run_one_script(const char *name, const char *path, uint32_t job_id,
int max_wait, char **env, uid_t uid)
{
int status;
pid_t cpid;
xassert(env);
if (path == NULL || path[0] == '\0')
return 0;
if (job_id) {
debug("[job %u] attempting to run %s [%s]",
job_id, name, path);
} else
debug("attempting to run %s [%s]", name, path);
if (access(path, R_OK | X_OK) < 0) {
error("Can not run %s [%s]: %m", name, path);
return -1;
}
if ((cpid = fork()) < 0) {
error ("executing %s: fork: %m", name);
return -1;
}
if (cpid == 0) {
char *argv[2];
/* container_g_add_pid needs to be called in the
forked process part of the fork to avoid a race
condition where if this process makes a file or
detacts itself from a child before we add the pid
to the container in the parent of the fork.
*/
if (container_g_add_pid(job_id, getpid(), getuid())
!= SLURM_SUCCESS)
error("container_g_add_pid(%u): %m", job_id);
argv[0] = (char *)xstrdup(path);
argv[1] = NULL;
setpgid(0, 0);
execve(path, argv, env);
error("execve(%s): %m", path);
exit(127);
}
if (waitpid_timeout(name, cpid, &status, max_wait) < 0)
return (-1);
return status;
}
static int _call_external_program(stepd_step_rec_t *job)
{
int status, rc, opt;
pid_t cpid;
int max_wait = 300; /* seconds */
int time_remaining;
if ((job->state != SLURMSTEPD_STEP_RUNNING) ||
program_name == NULL || program_name[0] == '\0')
return 0;
debug("step_terminate_monitor: unkillable after %d sec, calling: %s",
timeout, program_name);
if (access(program_name, R_OK | X_OK) < 0) {
debug("step_terminate_monitor not running %s: %m",
program_name);
return 0;
}
if ((cpid = fork()) < 0) {
error("step_terminate_monitor executing %s: fork: %m",
program_name);
return -1;
}
if (cpid == 0) {
/* child */
char *argv[2];
char buf[16];
/* container_g_add_pid needs to be called in the
forked process part of the fork to avoid a race
condition where if this process makes a file or
detacts itself from a child before we add the pid
to the container in the parent of the fork.
*/
if (container_g_add_pid(recorded_jobid, getpid(), getuid())
!= SLURM_SUCCESS)
error("container_g_add_pid(%u): %m", recorded_jobid);
snprintf(buf, 16, "%u", recorded_jobid);
setenv("SLURM_JOBID", buf, 1);
setenv("SLURM_JOB_ID", buf, 1);
snprintf(buf, 16, "%u", recorded_stepid);
setenv("SLURM_STEPID", buf, 1);
setenv("SLURM_STEP_ID", buf, 1);
argv[0] = program_name;
argv[1] = NULL;
setpgid(0, 0);
execv(program_name, argv);
error("step_terminate_monitor execv(): %m");
exit(127);
}
opt = WNOHANG;
time_remaining = max_wait;
while (1) {
rc = waitpid(cpid, &status, opt);
if (rc < 0) {
if (errno == EINTR)
continue;
/* waitpid may very well fail under normal conditions
because the wait3() in mgr.c:_wait_for_any_task()
may have reaped the return code. */
return 0;
} else if (rc == 0) {
sleep(1);
if ((--time_remaining) == 0) {
error("step_terminate_monitor: %s still running"
" after %d seconds. Killing.",
program_name, max_wait);
killpg(cpid, SIGKILL);
opt = 0;
}
} else {
return status;
}
}
/* NOTREACHED */
}