int
relay_to_mom2(job *pjob, struct batch_request *request,
void (*func)(struct work_task *), struct work_task **ppwt)
{
int rc;
int conn; /* a client style connection handle */
pbs_net_t momaddr;
unsigned int momport;
struct work_task *pwt;
int prot;
mominfo_t *pmom = 0;
pbs_list_head *mom_tasklist_ptr = NULL;
momaddr = pjob->ji_qs.ji_un.ji_exect.ji_momaddr;
momport = pjob->ji_qs.ji_un.ji_exect.ji_momport;
if (pbs_conf.pbs_use_tcp == 1) {
prot = PROT_RPP;
pmom = tfind2((unsigned long) momaddr, momport, &ipaddrs);
if (!pmom || (((mom_svrinfo_t *) (pmom->mi_data))->msr_state & INUSE_DOWN)) {
return (PBSE_NORELYMOM);
}
mom_tasklist_ptr = &(((mom_svrinfo_t *) (pmom->mi_data))->msr_deferred_cmds);
} else {
prot = PROT_TCP;
}
conn = svr_connect(momaddr, momport, process_Dreply, ToServerDIS, prot);
if (conn < 0) {
log_event(PBSEVENT_ERROR, PBS_EVENTCLASS_REQUEST, LOG_WARNING, "", msg_norelytomom);
return (PBSE_NORELYMOM);
}
request->rq_orgconn = request->rq_conn; /* save client socket */
pbs_errno = 0;
rc = issue_Drequest(conn, request, func, &pwt, prot);
if ((rc == 0) && (func != release_req)) {
/* work-task entry job related rpp, link to the job's list */
append_link(&pjob->ji_svrtask, &pwt->wt_linkobj, pwt);
if (prot == PROT_RPP)
append_link(mom_tasklist_ptr, &pwt->wt_linkobj2, pwt); /* if rpp, link to mom list as well */
}
if (ppwt != NULL)
*ppwt = pwt;
/*
* We do not want req_reject() to send non PBSE error numbers.
* Check for internal errors and when found return PBSE_SYSTEM.
*/
if ((rc != 0) && (pbs_errno == 0))
return (PBSE_SYSTEM);
else
return (rc);
}
/**
* @brief
* process the reply received for a request issued to
* another server via issue_request()
*
* Reads the reply from the RPP stream and executes the work task associated
* with the RPP reply message. The RPP request for which this reply arrived
* is matched by comparing the msgid of the reply with the msgid of the work
* tasks stored in the msr_deferred_cmds list of the mom for this stream.
*
* @param[in] handle - RPP handle on which reply/close arrived
*
* @return void
*/
void
process_DreplyRPP(int handle)
{
struct work_task *ptask;
int rc;
struct batch_request *request;
struct batch_reply *reply;
char *msgid = NULL;
mominfo_t *pmom = 0;
if ((pmom = tfind2((u_long) handle, 0, &streams)) == NULL)
return;
DIS_rpp_reset();
/* find the work task for the socket, it will point us to the request */
msgid = disrst(handle, &rc);
if (!msgid || rc) { /* rpp connection actually broke, cull all pending requests */
while ((ptask = (struct work_task *)GET_NEXT((((mom_svrinfo_t *) (pmom->mi_data))->msr_deferred_cmds)))) {
/* no need to compare wt_event with handle, since the
* task list is for this mom and so it will always match
*/
if (ptask->wt_type == WORK_Deferred_Reply) {
request = ptask->wt_parm1;
if (request) {
request->rq_reply.brp_code = rc;
request->rq_reply.brp_choice = BATCH_REPLY_CHOICE_NULL;
}
}
ptask->wt_aux = PBSE_NORELYMOM;
pbs_errno = PBSE_NORELYMOM;
if (ptask->wt_event2)
free(ptask->wt_event2);
dispatch_task(ptask);
}
} else {
/* we read msgid fine, so proceed to match it and process the respective task */
/* get the task list */
ptask = (struct work_task *)GET_NEXT((((mom_svrinfo_t *) (pmom->mi_data))->msr_deferred_cmds));
while (ptask) {
char *cmd_msgid = ptask->wt_event2;
if (strcmp(cmd_msgid, msgid) == 0) {
if (ptask->wt_type == WORK_Deferred_Reply)
request = ptask->wt_parm1;
else
request = NULL;
if (!request) {
if ((reply = (struct batch_reply *) malloc(sizeof(struct batch_reply))) == 0) {
delete_task(ptask);
free(cmd_msgid);
log_err(errno, msg_daemonname, "Out of memory creating batch reply");
return;
}
(void) memset(reply, 0, sizeof(struct batch_reply));
} else {
reply = &request->rq_reply;
}
/* read and decode the reply */
if ((rc = DIS_reply_read(handle, reply, 1)) != 0) {
reply->brp_code = rc;
reply->brp_choice = BATCH_REPLY_CHOICE_NULL;
ptask->wt_aux = PBSE_NORELYMOM;
pbs_errno = PBSE_NORELYMOM;
} else {
ptask->wt_aux = reply->brp_code;
pbs_errno = reply->brp_code;
}
ptask->wt_parm3 = reply; /* set the reply in case callback fn uses without having a preq */
dispatch_task(ptask);
if (!request)
PBSD_FreeReply(reply);
free(cmd_msgid);
break;
}
ptask = (struct work_task *) GET_NEXT(ptask->wt_linkobj2);
}
free(msgid); /* the msgid read should be free after use in matching */
}
}
/**
* @brief
* Send execution job on connected rpp stream.
*
* @param[in] jobp - pointer to the job being sent
* @param[in] hostaddr - the address of host to send job to, host byte order
* @param[in] port - the destination port, host byte order
* @param[in] request - The batch request associated with this send job call
*
* @return int
* @retval 2 : success
* @retval -1 : failure (pbs_errno set to error number)
*
*/
int
send_job_exec(job *jobp, pbs_net_t hostaddr, int port, struct batch_request *request)
{
pbs_list_head attrl;
attribute *pattr;
mominfo_t *pmom = NULL;
int stream = -1;
int encode_type;
char *destin = jobp->ji_qs.ji_destin;
int i;
size_t credlen = 0;
char *credbuf = NULL;
char job_id[PBS_MAXSVRJOBID + 1];
struct attropl *pqjatr; /* list (single) of attropl for quejob */
int rc;
int rpp = 1;
char *jobid = NULL;
char *script = NULL;
char *msgid = NULL;
char *dup_msgid = NULL;
struct work_task *ptask = NULL;
/* if job has a script read it from database */
if (jobp->ji_qs.ji_svrflags & JOB_SVFLG_SCRIPT) {
/*
* copy the job script from database to a temp file
* PBSD_jscript works with a file
* delete it at the end of the send
*/
if ((script = svr_load_jobscript(jobp)) == NULL) {
pbs_errno = PBSE_SYSTEM;
snprintf(log_buffer, sizeof(log_buffer),
"Failed to load job script for job %s",
jobp->ji_qs.ji_jobid);
log_err(pbs_errno, "send_job", log_buffer);
goto send_err;
}
}
stream = svr_connect(hostaddr, port, NULL, ToServerDIS, rpp);
if (stream < 0) {
log_event(PBSEVENT_ERROR, PBS_EVENTCLASS_REQUEST, LOG_WARNING, "", "Could not connect to Mom");
goto send_err;
}
pmom = tfind2((unsigned long) jobp->ji_qs.ji_un.ji_exect.ji_momaddr,
jobp->ji_qs.ji_un.ji_exect.ji_momport,
&ipaddrs);
if (!pmom || (((mom_svrinfo_t *)(pmom->mi_data))->msr_state & INUSE_DOWN))
goto send_err;
CLEAR_HEAD(attrl);
resc_access_perm = ATR_DFLAG_MOM;
encode_type = ATR_ENCODE_MOM;
pattr = jobp->ji_wattr;
for (i = 0; i < (int) JOB_ATR_LAST; i++) {
if ((job_attr_def + i)->at_flags & resc_access_perm) {
(void)(job_attr_def + i)->at_encode(pattr + i, &attrl,
(job_attr_def + i)->at_name, (char *) 0, encode_type,
NULL);
}
}
attrl_fixlink(&attrl);
/* save the job id for when after we purge the job */
/* read any credential file */
(void)get_credential(pmom->mi_host, jobp, PBS_GC_BATREQ, &credbuf, &credlen);
(void) strcpy(job_id, jobp->ji_qs.ji_jobid);
pbs_errno = 0;
pqjatr = &((svrattrl *) GET_NEXT(attrl))->al_atopl;
jobid = PBSD_queuejob(stream, jobp->ji_qs.ji_jobid, destin, pqjatr, (char *) 0, rpp, &msgid);
free_attrlist(&attrl);
if (jobid == NULL)
goto send_err;
rpp_add_close_func(stream, process_DreplyRPP); /* register a close handler */
/* adding msgid to deferred list, dont free msgid */
if ((ptask = add_mom_deferred_list(stream, pmom, post_sendmom, msgid, request, jobp)) == NULL)
goto send_err;
/* add to pjob->svrtask list so its automatically cleared when job is purged */
append_link(&jobp->ji_svrtask, &ptask->wt_linkobj, ptask);
/* we cannot use the same msgid, since it is not part of the preq,
* make a dup of it, and we can freely free it
*/
if ((dup_msgid = strdup(msgid)) == NULL)
goto send_err;
/*
* henceforth use the same msgid, since we mean to say all this is
* part of a single logical request to the mom
* and we will be hanging off one request to be answered to finally
*/
if (jobp->ji_qs.ji_svrflags & JOB_SVFLG_SCRIPT) {
if (PBSD_jscript_direct(stream, script, rpp, &dup_msgid) != 0)
goto send_err;
}
free(script);
script = NULL;
if (credlen > 0) {
rc = PBSD_jcred(stream, jobp->ji_extended.ji_ext.ji_credtype, credbuf, credlen, rpp, &dup_msgid);
if (credbuf)
free(credbuf);
if (rc != 0)
goto send_err;
}
if ((jobp->ji_qs.ji_svrflags & JOB_SVFLG_HASRUN)
&& (hostaddr != pbs_server_addr)) {
if ((move_job_file(stream, jobp, StdOut, rpp, &dup_msgid) != 0) ||
(move_job_file(stream, jobp, StdErr, rpp, &dup_msgid) != 0) ||
(move_job_file(stream, jobp, Chkpt, rpp, &dup_msgid) != 0))
goto send_err;
}
if (PBSD_commit(stream, job_id, rpp, &dup_msgid) != 0)
goto send_err;
free(dup_msgid); /* free this as it is not part of any work task */
return 2;
send_err:
if (dup_msgid)
free(dup_msgid);
if (script)
free(script);
if (ptask) {
if (ptask->wt_event2)
free(ptask->wt_event2);
delete_task(ptask);
}
sprintf(log_buffer, "send of job to %s failed error = %d", destin, pbs_errno);
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, LOG_INFO, jobp->ji_qs.ji_jobid, log_buffer);
return (-1);
}
