int req_signaljob(
void *vp) /* I */
{
struct batch_request *preq = (struct batch_request *)vp;
job *pjob;
int rc;
char log_buf[LOCAL_LOG_BUF_SIZE];
struct batch_request *dup_req = NULL;
/* preq free'd in error cases */
if ((pjob = chk_job_request(preq->rq_ind.rq_signal.rq_jid, preq)) == 0)
{
return(PBSE_NONE);
}
/* the job must be running */
if (pjob->ji_qs.ji_state != JOB_STATE_RUNNING)
{
req_reject(PBSE_BADSTATE, 0, preq, NULL, NULL);
unlock_ji_mutex(pjob, __func__, (char *)"1", LOGLEVEL);
return(PBSE_NONE);
}
/* Special pseudo signals for suspend and resume require op/mgr */
if (!strcmp(preq->rq_ind.rq_signal.rq_signame, SIG_RESUME) ||
!strcmp(preq->rq_ind.rq_signal.rq_signame, SIG_SUSPEND))
{
if ((preq->rq_perm & (ATR_DFLAG_OPRD | ATR_DFLAG_OPWR | ATR_DFLAG_MGRD | ATR_DFLAG_MGWR)) == 0)
{
/* for suspend/resume, must be mgr/op */
req_reject(PBSE_PERM, 0, preq, NULL, NULL);
unlock_ji_mutex(pjob, __func__, (char *)"2", LOGLEVEL);
return(PBSE_NONE);
}
}
/* save job ptr for post_signal_req() */
preq->rq_extra = strdup(pjob->ji_qs.ji_jobid);
/* FIXME: need a race-free check for available free subnodes before
* resuming a suspended job */
#ifdef DONOTSUSPINTJOB
/* interactive jobs don't resume correctly so don't allow a suspend */
if (!strcmp(preq->rq_ind.rq_signal.rq_signame, SIG_SUSPEND) &&
(pjob->ji_wattr[JOB_ATR_interactive].at_flags & ATR_VFLAG_SET) &&
(pjob->ji_wattr[JOB_ATR_interactive].at_val.at_long > 0))
{
req_reject(PBSE_JOBTYPE, 0, preq, NULL, NULL);
unlock_ji_mutex(pjob, __func__, (char *)"3", LOGLEVEL);
return(PBSE_NONE);
}
#endif
if (LOGLEVEL >= 6)
{
sprintf(log_buf, "relaying signal request to mom %lu", pjob->ji_qs.ji_un.ji_exect.ji_momaddr);
log_record(PBSEVENT_SCHED,PBS_EVENTCLASS_REQUEST,"req_signaljob",log_buf);
}
/* send reply for asynchronous suspend */
if (preq->rq_type == PBS_BATCH_AsySignalJob)
{
reply_ack(preq);
preq->rq_noreply = TRUE;
}
/* pass the request on to MOM */
if ((rc = copy_batchrequest(&dup_req, preq, 0, -1)) != 0)
{
req_reject(rc, 0, preq, NULL, "can not allocate memory");
unlock_ji_mutex(pjob, __func__, (char *)"4", LOGLEVEL);
}
/* The dup_req is freed in relay_to_mom (failure)
* or in issue_Drequest (success) */
else
{
rc = relay_to_mom(&pjob, dup_req, NULL);
if (pjob != NULL)
unlock_ji_mutex(pjob, __func__, (char *)"4", LOGLEVEL);
if (rc != PBSE_NONE)
{
free_br(dup_req);
req_reject(rc, 0, preq, NULL, NULL); /* unable to get to MOM */
}
else
{
post_signal_req(dup_req);
free_br(preq);
}
}
/* If successful we ack after mom replies to us, we pick up in post_signal_req() */
return(PBSE_NONE);
} /* END req_signaljob() */
int req_signaljob(
batch_request *preq) /* I */
{
job *pjob;
int rc;
char log_buf[LOCAL_LOG_BUF_SIZE];
batch_request *dup_req = NULL;
/* preq free'd in error cases */
if ((pjob = chk_job_request(preq->rq_ind.rq_signal.rq_jid, preq)) == 0)
{
return(PBSE_NONE);
}
mutex_mgr job_mutex(pjob->ji_mutex, true);
/* the job must be running */
if (pjob->ji_qs.ji_state != JOB_STATE_RUNNING)
{
req_reject(PBSE_BADSTATE, 0, preq, NULL, NULL);
return(PBSE_NONE);
}
/* Special pseudo signals for suspend and resume require op/mgr */
if (!strcmp(preq->rq_ind.rq_signal.rq_signame, SIG_RESUME) ||
!strcmp(preq->rq_ind.rq_signal.rq_signame, SIG_SUSPEND))
{
if ((preq->rq_perm & (ATR_DFLAG_OPRD | ATR_DFLAG_OPWR | ATR_DFLAG_MGRD | ATR_DFLAG_MGWR)) == 0)
{
/* for suspend/resume, must be mgr/op */
req_reject(PBSE_PERM, 0, preq, NULL, NULL);
return(PBSE_NONE);
}
}
/* save job ptr for post_signal_req() */
preq->rq_extra = strdup(pjob->ji_qs.ji_jobid);
/* FIXME: need a race-free check for available free subnodes before
* resuming a suspended job */
#ifdef DONOTSUSPINTJOB
/* interactive jobs don't resume correctly so don't allow a suspend */
if (!strcmp(preq->rq_ind.rq_signal.rq_signame, SIG_SUSPEND) &&
(pjob->ji_wattr[JOB_ATR_interactive].at_flags & ATR_VFLAG_SET) &&
(pjob->ji_wattr[JOB_ATR_interactive].at_val.at_long > 0))
{
req_reject(PBSE_JOBTYPE, 0, preq, NULL, NULL);
return(PBSE_NONE);
}
#endif
if (LOGLEVEL >= 6)
{
char ipstr[128];
sprintf(log_buf, "relaying signal request to mom %s", netaddr_long(pjob->ji_qs.ji_un.ji_exect.ji_momaddr,ipstr));
log_record(PBSEVENT_SCHED,PBS_EVENTCLASS_REQUEST,"req_signaljob",log_buf);
}
/* send reply for asynchronous suspend */
if (preq->rq_type == PBS_BATCH_AsySignalJob)
{
/* reply_ack will free preq. We need to copy it before we call reply_ack */
batch_request *new_preq;
new_preq = duplicate_request(preq, -1);
if (new_preq == NULL)
{
sprintf(log_buf, "failed to duplicate batch request");
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, __func__, log_buf);
return(PBSE_MEM_MALLOC);
}
get_batch_request_id(new_preq);
reply_ack(new_preq);
preq->rq_noreply = TRUE;
}
/* pass the request on to MOM */
if ((dup_req = duplicate_request(preq)) == NULL)
{
req_reject(PBSE_SYSTEM, 0, preq, NULL, "can not allocate memory");
}
/* The dup_req is freed in relay_to_mom (failure)
* or in issue_Drequest (success) */
else
{
rc = relay_to_mom(&pjob, dup_req, NULL);
if (pjob != NULL)
job_mutex.unlock();
else
job_mutex.set_unlock_on_exit(false);
if (rc != PBSE_NONE)
{
free_br(dup_req);
req_reject(rc, 0, preq, NULL, NULL); /* unable to get to MOM */
}
else
{
post_signal_req(dup_req);
free_br(preq);
}
}
/* If successful we ack after mom replies to us, we pick up in post_signal_req() */
return(PBSE_NONE);
} /* END req_signaljob() */