int
GLExecPrivSepHelper::create_process(const char* path,
ArgList& args,
Env& env,
const char* iwd,
int job_std_fds[3],
const char* std_file_names[3],
int nice_inc,
size_t* core_size_ptr,
int reaper_id,
int dc_job_opts,
FamilyInfo* family_info,
int *,
MyString *error_msg)
{
ASSERT(m_initialized);
if (!proxy_valid_right_now()) {
dprintf(D_ALWAYS, "GLExecPrivSepHelper::create_process: not invoking glexec since the proxy is not valid!\n");
if( error_msg ) {
error_msg->formatstr_cat("The job proxy is invalid.");
}
return -1;
}
// make a copy of std FDs so we're not messing w/ our caller's
// memory
int std_fds[3] = {-1, -1, -1};
ArgList modified_args;
modified_args.AppendArg(m_run_script);
modified_args.AppendArg(m_glexec);
modified_args.AppendArg(m_proxy);
modified_args.AppendArg(m_sandbox);
modified_args.AppendArg(m_wrapper_script);
for (int i = 0; i < 3; i++) {
modified_args.AppendArg((job_std_fds == NULL || job_std_fds[i] == -1) ?
std_file_names[i] : "-");
}
modified_args.AppendArg(path);
for (int i = 1; i < args.Count(); i++) {
modified_args.AppendArg(args.GetArg(i));
}
int glexec_errors = 0;
while(1) {
// setup a UNIX domain socket for communicating with
// condor_glexec_wrapper (see comment above feed_wrapper()
// for details
//
int sock_fds[2];
if (socketpair(PF_UNIX, SOCK_STREAM, 0, sock_fds) == -1)
{
dprintf(D_ALWAYS,
"GLEXEC: socketpair error: %s\n",
strerror(errno));
return false;
}
std_fds[0] = sock_fds[1];
// now create a pipe for receiving diagnostic stdout/stderr from glexec
int glexec_out_fds[2];
if (pipe(glexec_out_fds) < 0) {
dprintf(D_ALWAYS,
"GLEXEC: pipe() error: %s\n",
strerror(errno));
close(sock_fds[0]);
close(sock_fds[1]);
return false;
}
std_fds[1] = glexec_out_fds[1];
std_fds[2] = std_fds[1]; // collect glexec stderr and stdout together
FamilyInfo fi;
FamilyInfo* fi_ptr = (family_info != NULL) ? family_info : &fi;
MyString proxy_path;
proxy_path.formatstr("%s.condor/%s", m_sandbox, m_proxy);
fi_ptr->glexec_proxy = proxy_path.Value();
// At the very least, we need to pass the condor daemon's
// X509_USER_PROXY to condor_glexec_run. Currently, we just
// pass all daemon environment. We do _not_ run
// condor_glexec_run in the job environment, because that
// would be a security risk and would serve no purpose, since
// glexec cleanses the environment anyway.
dc_job_opts &= ~(DCJOBOPT_NO_ENV_INHERIT);
int pid = daemonCore->Create_Process(m_run_script.Value(),
modified_args,
PRIV_USER_FINAL,
reaper_id,
FALSE,
FALSE,
NULL,
iwd,
fi_ptr,
NULL,
std_fds,
NULL,
nice_inc,
NULL,
dc_job_opts,
core_size_ptr,
NULL,
NULL,
error_msg);
// close our handle to glexec's end of the diagnostic output pipe
close(glexec_out_fds[1]);
MyString glexec_error_msg;
int glexec_rc = 0;
int ret_val = feed_wrapper(pid, sock_fds, env, dc_job_opts, job_std_fds, glexec_out_fds[0], &glexec_error_msg, &glexec_rc);
// if not closed in feed_wrapper, close the glexec error pipe now
if( glexec_out_fds[0] != -1 ) {
close(glexec_out_fds[0]);
}
// Unlike the other glexec operations where we handle
// glexec retry inside the helper script,
// condor_glexec_run cannot handle retry for us, because
// it must exec glexec rather than spawning it off in a
// new process. Therefore, we handle here retries in case
// of transient errors.
if( ret_val != 0 ) {
return ret_val; // success
}
bool retry = true;
if( glexec_rc != 202 && glexec_rc != 203 ) {
// Either a non-transient glexec error, or some other
// non-glexec error.
retry = false;
}
else {
// This _could_ be a transient glexec issue, such as a
// communication error with GUMS, so retry up to some
// limit.
glexec_errors += 1;
if( glexec_errors > m_glexec_retries ) {
retry = false;
}
}
if( !retry ) {
// return the most recent glexec error output
if( error_msg ) {
error_msg->formatstr_cat(glexec_error_msg.Value());
}
return 0;
}
// truncated exponential backoff
int delay_rand = 1 + (get_random_int() % glexec_errors) % 100;
int delay = m_glexec_retry_delay * delay_rand;
dprintf(D_ALWAYS,"Glexec exited with status %d; retrying in %d seconds.\n",
glexec_rc, delay );
sleep(delay);
// now try again ...
}
// should never get here
return 0;
}
int
GLExecPrivSepHelper::create_process(const char* path,
ArgList& args,
Env& env,
const char* iwd,
int job_std_fds[3],
const char* std_file_names[3],
int nice_inc,
size_t* core_size_ptr,
int reaper_id,
int dc_job_opts,
FamilyInfo* family_info,
int *,
MyString *error_msg)
{
ASSERT(m_initialized);
if (!proxy_valid_right_now()) {
dprintf(D_ALWAYS, "GLExecPrivSepHelper::create_process: not invoking glexec since the proxy is not valid!\n");
return -1;
}
// make a copy of std FDs so we're not messing w/ our caller's
// memory
int std_fds[3] = {-1, -1, -1};
ArgList modified_args;
modified_args.AppendArg(m_run_script);
modified_args.AppendArg(m_glexec);
modified_args.AppendArg(m_proxy);
modified_args.AppendArg(m_sandbox);
modified_args.AppendArg(m_wrapper_script);
for (int i = 0; i < 3; i++) {
modified_args.AppendArg((job_std_fds == NULL || job_std_fds[i] == -1) ?
std_file_names[i] : "-");
}
modified_args.AppendArg(path);
for (int i = 1; i < args.Count(); i++) {
modified_args.AppendArg(args.GetArg(i));
}
// setup a UNIX domain socket for communicating with
// condor_glexec_wrapper (see comment above feed_wrapper()
// for details
//
int sock_fds[2];
if (socketpair(PF_UNIX, SOCK_STREAM, 0, sock_fds) == -1)
{
dprintf(D_ALWAYS,
"GLEXEC: socketpair error: %s\n",
strerror(errno));
return false;
}
std_fds[0] = sock_fds[1];
// now create a pipe for receiving diagnostic stdout/stderr from glexec
int glexec_out_fds[2];
if (pipe(glexec_out_fds) < 0) {
dprintf(D_ALWAYS,
"GLEXEC: pipe() error: %s\n",
strerror(errno));
close(sock_fds[0]);
close(sock_fds[1]);
return false;
}
std_fds[1] = glexec_out_fds[1];
std_fds[2] = std_fds[1]; // collect glexec stderr and stdout together
FamilyInfo fi;
FamilyInfo* fi_ptr = (family_info != NULL) ? family_info : &fi;
MyString proxy_path;
proxy_path.formatstr("%s.condor/%s", m_sandbox, m_proxy);
fi_ptr->glexec_proxy = proxy_path.Value();
// At the very least, we need to pass the condor daemon's
// X509_USER_PROXY to condor_glexec_run. Currently, we just
// pass all daemon environment. We do _not_ run
// condor_glexec_run in the job environment, because that
// would be a security risk and would serve no purpose, since
// glexec cleanses the environment anyway.
dc_job_opts &= ~(DCJOBOPT_NO_ENV_INHERIT);
int pid = daemonCore->Create_Process(m_run_script.Value(),
modified_args,
PRIV_USER_FINAL,
reaper_id,
FALSE,
NULL,
iwd,
fi_ptr,
NULL,
std_fds,
NULL,
nice_inc,
NULL,
dc_job_opts,
core_size_ptr,
NULL,
NULL,
error_msg);
// close our handle to glexec's end of the diagnostic output pipe
close(glexec_out_fds[1]);
int ret_val = feed_wrapper(pid, sock_fds, env, dc_job_opts, job_std_fds, glexec_out_fds[0], error_msg);
// if not closed in feed_wrapper, close the glexec error pipe now
if( glexec_out_fds[0] != -1 ) {
close(glexec_out_fds[0]);
}
return ret_val;
}
