void req_commit(
struct batch_request *preq) /* I */
{
job *pj;
pj = locate_new_job(preq->rq_conn, preq->rq_ind.rq_commit);
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
(pj != NULL) ? pj->ji_qs.ji_jobid : "NULL",
"committing job");
}
if (pj == NULL)
{
req_reject(PBSE_UNKJOBID, 0, preq, NULL, NULL);
return;
}
if (pj->ji_qs.ji_substate != JOB_SUBSTATE_TRANSICM)
{
log_err(errno, "req_commit", "cannot commit job in unexpected state");
req_reject(PBSE_IVALREQ, 0, preq, NULL, NULL);
return;
}
/* move job from new job list to "all" job list, set to running state */
delete_link(&pj->ji_alljobs);
append_link(&svr_alljobs, &pj->ji_alljobs, pj);
/*
** Set JOB_SVFLG_HERE to indicate that this is Mother Superior.
*/
pj->ji_qs.ji_svrflags |= JOB_SVFLG_HERE;
pj->ji_qs.ji_state = JOB_STATE_RUNNING;
pj->ji_qs.ji_substate = JOB_SUBSTATE_PRERUN;
pj->ji_qs.ji_un_type = JOB_UNION_TYPE_MOM;
pj->ji_qs.ji_un.ji_momt.ji_svraddr = get_connectaddr(preq->rq_conn);
pj->ji_qs.ji_un.ji_momt.ji_exitstat = 0;
/* For MOM - start up the job (blocks) */
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
(pj != NULL) ? pj->ji_qs.ji_jobid : "NULL",
"starting job execution");
}
start_exec(pj);
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
(pj != NULL) ? pj->ji_qs.ji_jobid : "NULL",
"job execution started");
}
/* if start request fails, reply with failure string */
if (pj->ji_qs.ji_substate == JOB_SUBSTATE_EXITING)
{
char tmpLine[1024];
if ((pj->ji_hosts != NULL) &&
(pj->ji_nodekill >= 0) &&
(pj->ji_hosts[pj->ji_nodekill].hn_host != NULL))
{
sprintf(tmpLine, "start failed on node %s",
pj->ji_hosts[pj->ji_nodekill].hn_host);
}
else
{
sprintf(tmpLine, "start failed on unknown node");
}
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
(pj != NULL) ? pj->ji_qs.ji_jobid : "NULL",
tmpLine);
}
reply_text(preq, 0, tmpLine);
}
else
{
reply_jobid(preq, pj->ji_qs.ji_jobid, BATCH_REPLY_CHOICE_Commit);
}
job_save(pj, SAVEJOB_FULL);
/* NOTE: we used to flag JOB_ATR_errpath, JOB_ATR_outpath,
* JOB_ATR_session_id, and JOB_ATR_altid as modified at this point to make sure
* pbs_server got these attr values. This worked fine before TORQUE modified
* job launched into an async process. At 2.0.0p6, a new attribute "SEND" flag
* was added to handle this process. */
return;
} /* END req_commit() */
void req_quejob(
struct batch_request *preq) /* ptr to the decoded request */
{
char *id = "req_quejob";
char basename[PBS_JOBBASE + 1];
int created_here = 0;
int index;
char *jid;
attribute_def *pdef;
job *pj;
svrattrl *psatl;
int rc;
int sock = preq->rq_conn;
int IsCheckpoint = 0;
/* set basic (user) level access permission */
resc_access_perm = ATR_DFLAG_USWR | ATR_DFLAG_Creat;
if (PBSNodeCheckProlog)
{
check_state(1);
mom_server_all_update_stat();
if (internal_state & INUSE_DOWN)
{
req_reject(PBSE_MOMREJECT,0,preq,NULL,NULL);
return;
}
}
if (preq->rq_fromsvr)
{
/* from another server - accept the extra attributes */
resc_access_perm |= ATR_DFLAG_MGWR | ATR_DFLAG_SvWR | ATR_DFLAG_MOM;
jid = preq->rq_ind.rq_queuejob.rq_jid;
}
else
{
/* request must be from server */
log_err(errno, id, "request not from server");
req_reject(PBSE_IVALREQ, 0, preq, NULL, "request not received from server");
return;
}
/* does job already exist, check both old and new jobs */
if ((pj = find_job(jid)) == NULL)
{
pj = (job *)GET_NEXT(svr_newjobs);
while (pj != NULL)
{
if (!strcmp(pj->ji_qs.ji_jobid, jid))
break;
pj = (job *)GET_NEXT(pj->ji_alljobs);
}
}
/*
* New job ...
*
* for MOM - rather than make up a hashname, we use the name sent
* to us by the server as an attribute.
*/
psatl = (svrattrl *)GET_NEXT(preq->rq_ind.rq_queuejob.rq_attr);
while (psatl != NULL)
{
if (!strcmp(psatl->al_name,ATTR_hashname))
{
strcpy(basename,psatl->al_value);
break;
}
psatl = (svrattrl *)GET_NEXT(psatl->al_link);
}
if (pj != NULL)
{
/* newly queued job already exists */
if (pj->ji_qs.ji_substate == JOB_SUBSTATE_RUNNING)
{
/* FAILURE - job exists and is running */
log_err(errno,id,"cannot queue new job, job exists and is running");
req_reject(PBSE_JOBEXIST,0,preq,NULL,"job is running");
return;
}
/* if checkpointed, then keep old and skip rest of process */
if (pj->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_FILE)
{
IsCheckpoint = 1;
} /* END if (pj->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_FILE) */
else
{
/* unlink job from svr_alljobs since it will be placed on newjobs */
delete_link(&pj->ji_alljobs);
}
} /* END if (pj != NULL) */
else
{
/* if not already here, allocate job struct */
if ((pj = job_alloc()) == NULL)
{
/* FAILURE */
req_reject(PBSE_SYSTEM, 0, preq, NULL, "cannot allocate new job structure");
return;
}
} /* END else (pj != NULL) */
if (IsCheckpoint == 0)
{
strcpy(pj->ji_qs.ji_jobid,jid);
strcpy(pj->ji_qs.ji_fileprefix,basename);
pj->ji_modified = 1;
pj->ji_qs.ji_svrflags = created_here;
pj->ji_qs.ji_un_type = JOB_UNION_TYPE_NEW;
}
/* decode attributes from request into job structure */
psatl = (svrattrl *)GET_NEXT(preq->rq_ind.rq_queuejob.rq_attr);
while (psatl != NULL)
{
if (IsCheckpoint == 1)
{
if (strcmp(psatl->al_name,ATTR_checkpoint_name) &&
strcmp(psatl->al_name,ATTR_v))
{
psatl = (svrattrl *)GET_NEXT(psatl->al_link);
continue;
}
}
/* identify the attribute by name */
index = find_attr(job_attr_def,psatl->al_name,JOB_ATR_LAST);
if (index < 0)
{
/* FAILURE */
/* didn`t recognize the name */
job_purge(pj); /* CRI - 12/20/2004 */
reply_badattr(PBSE_NOATTR,1,psatl,preq);
return;
}
pdef = &job_attr_def[index];
/* Is attribute not writeable by manager or by a server? */
if ((pdef->at_flags & resc_access_perm) == 0)
{
/* FAILURE */
job_purge(pj);
reply_badattr(PBSE_ATTRRO,1,psatl,preq);
return;
}
/* decode attribute */
if (!strcmp(psatl->al_name,ATTR_v))
{
rc = decode_arst_merge(
&pj->ji_wattr[index],
psatl->al_name,
psatl->al_resc,
psatl->al_value);
}
else
{
rc = pdef->at_decode(
&pj->ji_wattr[index],
psatl->al_name,
psatl->al_resc,
psatl->al_value);
}
if (rc != 0)
{
/* FAILURE */
/* all errors are fatal for MOM */
job_purge(pj);
reply_badattr(rc,1,psatl,preq);
return;
}
if (psatl->al_op == DFLT)
{
if (psatl->al_resc)
{
resource *presc;
resource_def *prdef;
prdef = find_resc_def(svr_resc_def,psatl->al_resc,svr_resc_size);
if (prdef == NULL)
{
job_purge(pj);
reply_badattr(rc,1,psatl, preq);
return;
}
presc = find_resc_entry(&pj->ji_wattr[index],prdef);
if (presc != NULL)
presc->rs_value.at_flags |= ATR_VFLAG_DEFLT;
}
else
{
pj->ji_wattr[index].at_flags |= ATR_VFLAG_DEFLT;
}
} /* END if (psatl->al_op == DFLT) */
psatl = (svrattrl *)GET_NEXT(psatl->al_link);
} /* END while (psatl != NULL) */
if (IsCheckpoint == 1)
{
pj->ji_qs.ji_substate = JOB_SUBSTATE_TRANSIN;
if (reply_jobid(preq,pj->ji_qs.ji_jobid,BATCH_REPLY_CHOICE_Queue) == 0)
{
delete_link(&pj->ji_alljobs);
append_link(&svr_newjobs,&pj->ji_alljobs,pj);
pj->ji_qs.ji_un_type = JOB_UNION_TYPE_NEW;
pj->ji_qs.ji_un.ji_newt.ji_fromsock = sock;
pj->ji_qs.ji_un.ji_newt.ji_fromaddr = get_connectaddr(sock);
pj->ji_qs.ji_un.ji_newt.ji_scriptsz = 0;
/* Per Eric R., req_mvjobfile was giving error in open_std_file,
showed up as fishy error message */
if (pj->ji_grpcache != NULL)
{
free(pj->ji_grpcache);
pj->ji_grpcache = NULL;
}
}
else
{
close_conn(sock);
}
/* SUCCESS */
return;
}
/* set remaining job structure elements */
pj->ji_qs.ji_state = JOB_STATE_TRANSIT;
pj->ji_qs.ji_substate = JOB_SUBSTATE_TRANSIN;
pj->ji_wattr[(int)JOB_ATR_mtime].at_val.at_long = (long)time_now;
pj->ji_wattr[(int)JOB_ATR_mtime].at_flags |= ATR_VFLAG_SET;
pj->ji_qs.ji_un_type = JOB_UNION_TYPE_NEW;
pj->ji_qs.ji_un.ji_newt.ji_fromsock = sock;
pj->ji_qs.ji_un.ji_newt.ji_fromaddr = get_connectaddr(sock);
pj->ji_qs.ji_un.ji_newt.ji_scriptsz = 0;
/* acknowledge the request with the job id */
if (reply_jobid(preq, pj->ji_qs.ji_jobid, BATCH_REPLY_CHOICE_Queue) != 0)
{
/* reply failed, purge the job and close the connection */
close_conn(sock);
job_purge(pj);
return;
}
/* link job into server's new jobs list request */
append_link(&svr_newjobs, &pj->ji_alljobs, pj);
return;
} /* END req_quejob() */
void req_rdytocommit(
struct batch_request *preq) /* I */
{
job *pj;
int sock = preq->rq_conn;
int OrigState;
int OrigSState;
char OrigSChar;
long OrigFlags;
pj = locate_new_job(sock, preq->rq_ind.rq_rdytocommit);
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
(pj != NULL) ? pj->ji_qs.ji_jobid : "NULL",
"ready to commit job");
}
if (pj == NULL)
{
log_err(errno, "req_rdytocommit", "unknown job id");
req_reject(PBSE_UNKJOBID, 0, preq, NULL, NULL);
/* FAILURE */
return;
}
if (pj->ji_qs.ji_substate != JOB_SUBSTATE_TRANSIN)
{
log_err(errno, "req_rdytocommit", "cannot commit job in unexpected state");
req_reject(PBSE_IVALREQ, 0, preq, NULL, NULL);
/* FAILURE */
return;
}
OrigState = pj->ji_qs.ji_state;
OrigSState = pj->ji_qs.ji_substate;
OrigSChar = pj->ji_wattr[(int)JOB_ATR_state].at_val.at_char;
OrigFlags = pj->ji_wattr[(int)JOB_ATR_state].at_flags;
pj->ji_qs.ji_state = JOB_STATE_TRANSIT;
pj->ji_qs.ji_substate = JOB_SUBSTATE_TRANSICM;
pj->ji_wattr[(int)JOB_ATR_state].at_val.at_char = 'T';
pj->ji_wattr[(int)JOB_ATR_state].at_flags |= ATR_VFLAG_SET;
if (job_save(pj, SAVEJOB_NEW) == -1)
{
char tmpLine[1024];
sprintf(tmpLine, "cannot save job - errno=%d - %s",
errno,
strerror(errno));
log_err(errno, "req_rdytocommit", tmpLine);
/* commit failed, backoff state changes */
pj->ji_qs.ji_state = OrigState;
pj->ji_qs.ji_substate = OrigSState;
pj->ji_wattr[(int)JOB_ATR_state].at_val.at_char = OrigSChar;
pj->ji_wattr[(int)JOB_ATR_state].at_flags = OrigFlags;
req_reject(PBSE_SYSTEM, 0, preq, NULL, tmpLine);
/* FAILURE */
return;
}
/* acknowledge the request with the job id */
if (reply_jobid(preq, pj->ji_qs.ji_jobid, BATCH_REPLY_CHOICE_RdytoCom) != 0)
{
/* reply failed, purge the job and close the connection */
sprintf(log_buffer, "cannot report jobid - errno=%d - %s",
errno,
strerror(errno));
log_err(errno, "req_rdytocommit", log_buffer);
close_conn(sock);
job_purge(pj);
/* FAILURE */
return;
}
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
(pj != NULL) ? pj->ji_qs.ji_jobid : "NULL",
"ready to commit job completed");
}
return;
} /* END req_rdytocommit() */
void mom_req_quejob(
batch_request *preq) /* ptr to the decoded request */
{
char basename[PBS_JOBBASE + 1];
int created_here = 0;
int index;
char *jid;
attribute_def *pdef;
job *pj;
svrattrl *psatl;
int rc;
int sock = preq->rq_conn;
int IsCheckpoint = 0;
/* set basic (user) level access permission */
int resc_access_perm = ATR_DFLAG_USWR | ATR_DFLAG_Creat;
memset(basename, 0, sizeof(basename));
if (PBSNodeCheckProlog)
{
check_state(1);
if (internal_state & INUSE_DOWN)
{
req_reject(PBSE_BADMOMSTATE, 0, preq, NULL, NULL);
return;
}
}
if (reject_job_submit == TRUE)
{
req_reject(-1, 0, preq, NULL, "This mom is configured not to run jobs");
return;
}
if (preq->rq_fromsvr)
{
/* from another server - accept the extra attributes */
resc_access_perm |= ATR_DFLAG_MGWR | ATR_DFLAG_SvWR | ATR_DFLAG_MOM;
jid = preq->rq_ind.rq_queuejob.rq_jid;
}
else
{
/* request must be from server */
log_err(errno, __func__, (char *)"request not from server");
req_reject(PBSE_IVALREQ, 0, preq, NULL, "request not received from server");
return;
}
/* does job already exist, check both old and new jobs */
if ((pj = mom_find_job(jid)) == NULL)
{
pj = (job *)GET_NEXT(svr_newjobs);
while (pj != NULL)
{
if (!strcmp(pj->ji_qs.ji_jobid, jid))
break;
pj = (job *)GET_NEXT(pj->ji_alljobs);
}
}
/*
* New job ...
*
* for MOM - rather than make up a hashname, we use the name sent
* to us by the server as an pbs_attribute.
*/
psatl = (svrattrl *)GET_NEXT(preq->rq_ind.rq_queuejob.rq_attr);
while (psatl != NULL)
{
if (!strcmp(psatl->al_name,ATTR_hashname))
{
snprintf(basename, sizeof(basename), "%s", psatl->al_value);
break;
}
psatl = (svrattrl *)GET_NEXT(psatl->al_link);
}
if (basename[0] == '\0')
snprintf(basename, sizeof(basename), "%s", jid);
if (pj != NULL)
{
/* newly queued job already exists */
if (pj->ji_qs.ji_substate == JOB_SUBSTATE_RUNNING)
{
/* FAILURE - job exists and is running */
log_err(errno, __func__, (char *)"cannot queue new job, job exists and is running");
req_reject(PBSE_JOBEXIST, 0, preq, NULL, "job is running");
return;
}
/* if checkpointed, then keep old and skip rest of process */
if (pj->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_FILE)
{
IsCheckpoint = 1;
} /* END if (pj->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_FILE) */
else
{
/* reject the job. It is already working here. */
sprintf(log_buffer, "Job already exists. State: %d substate: %d", pj->ji_qs.ji_state, pj->ji_qs.ji_substate);
log_err(-1, __func__, log_buffer);
sprintf(log_buffer, "Job %s already on mom", pj->ji_qs.ji_jobid);
req_reject(PBSE_JOBEXIST, 0, preq, NULL, log_buffer);
return;
}
} /* END if (pj != NULL) */
else
{
/* if not already here, allocate job struct */
if ((pj = job_alloc()) == NULL)
{
/* FAILURE */
req_reject(PBSE_MEM_MALLOC, 0, preq, NULL, "cannot allocate new job structure");
return;
}
} /* END else (pj != NULL) */
if (IsCheckpoint == 0)
{
strcpy(pj->ji_qs.ji_jobid,jid);
strcpy(pj->ji_qs.ji_fileprefix,basename);
pj->ji_modified = 1;
pj->ji_qs.ji_svrflags = created_here;
pj->ji_qs.ji_un_type = JOB_UNION_TYPE_NEW;
/* changing the union type overwrites the euid for the job, and if
* ji_grpcache is set this potentially allows jobs to run as root. Unsetting
* ji_grpcache fixes this problem --dbeer */
if (pj->ji_grpcache != NULL)
{
free(pj->ji_grpcache);
pj->ji_grpcache = NULL;
}
}
/* decode attributes from request into job structure */
psatl = (svrattrl *)GET_NEXT(preq->rq_ind.rq_queuejob.rq_attr);
while (psatl != NULL)
{
if (IsCheckpoint == 1)
{
if (strcmp(psatl->al_name,ATTR_checkpoint_name) &&
strcmp(psatl->al_name,ATTR_v))
{
psatl = (svrattrl *)GET_NEXT(psatl->al_link);
continue;
}
}
/* identify the pbs_attribute by name */
index = find_attr(job_attr_def,psatl->al_name,JOB_ATR_LAST);
if (index < 0)
{
/* FAILURE */
/* didn`t recognize the name */
mom_job_purge(pj); /* CRI - 12/20/2004 */
reply_badattr(PBSE_NOATTR, 1, psatl, preq);
return;
}
pdef = &job_attr_def[index];
/* Is pbs_attribute not writeable by manager or by a server? */
if ((pdef->at_flags & resc_access_perm) == 0)
{
/* FAILURE */
mom_job_purge(pj);
reply_badattr(PBSE_ATTRRO, 1, psatl, preq);
return;
}
/* decode pbs_attribute */
if (!strcmp(psatl->al_name,ATTR_v))
{
rc = decode_arst_merge(
&pj->ji_wattr[index],
psatl->al_name,
psatl->al_resc,
psatl->al_value);
}
else
{
rc = pdef->at_decode(
&pj->ji_wattr[index],
psatl->al_name,
psatl->al_resc,
psatl->al_value,
resc_access_perm);
}
if (rc != 0)
{
/* FAILURE */
/* all errors are fatal for MOM */
mom_job_purge(pj);
reply_badattr(rc, 1, psatl, preq);
return;
}
if (psatl->al_op == DFLT)
{
if (psatl->al_resc)
{
resource *presc;
resource_def *prdef;
prdef = find_resc_def(svr_resc_def,psatl->al_resc,svr_resc_size);
if (prdef == NULL)
{
mom_job_purge(pj);
reply_badattr(rc, 1, psatl, preq);
return;
}
presc = find_resc_entry(&pj->ji_wattr[index],prdef);
if (presc != NULL)
presc->rs_value.at_flags |= ATR_VFLAG_DEFLT;
}
else
{
pj->ji_wattr[index].at_flags |= ATR_VFLAG_DEFLT;
}
} /* END if (psatl->al_op == DFLT) */
psatl = (svrattrl *)GET_NEXT(psatl->al_link);
} /* END while (psatl != NULL) */
if (IsCheckpoint == 1)
{
pj->ji_qs.ji_substate = JOB_SUBSTATE_TRANSIN;
if (reply_jobid(preq,pj->ji_qs.ji_jobid,BATCH_REPLY_CHOICE_Queue) == 0)
{
remove_from_job_list(pj);
append_link(&svr_newjobs,&pj->ji_alljobs,pj);
if (pj->ji_grpcache != NULL)
{
free(pj->ji_grpcache);
pj->ji_grpcache = NULL;
}
pj->ji_qs.ji_un_type = JOB_UNION_TYPE_NEW;
pj->ji_qs.ji_un.ji_newt.ji_fromsock = sock;
pj->ji_qs.ji_un.ji_newt.ji_fromaddr = get_connectaddr(sock,FALSE);
pj->ji_qs.ji_un.ji_newt.ji_scriptsz = 0;
/* Per Eric R., req_mvjobfile was giving error in open_std_file,
showed up as fishy error message */
if (pj->ji_grpcache != NULL)
{
free(pj->ji_grpcache);
pj->ji_grpcache = NULL;
}
}
else
{
close_conn(sock, FALSE);
}
/* SUCCESS */
return;
}
/* set remaining job structure elements */
pj->ji_qs.ji_state = JOB_STATE_TRANSIT;
pj->ji_qs.ji_substate = JOB_SUBSTATE_TRANSIN;
pj->ji_wattr[JOB_ATR_mtime].at_val.at_long = (long)time_now;
pj->ji_wattr[JOB_ATR_mtime].at_flags |= ATR_VFLAG_SET;
if (pj->ji_grpcache != NULL)
{
free(pj->ji_grpcache);
pj->ji_grpcache = NULL;
}
pj->ji_qs.ji_un_type = JOB_UNION_TYPE_NEW;
pj->ji_qs.ji_un.ji_newt.ji_fromsock = sock;
pj->ji_qs.ji_un.ji_newt.ji_fromaddr = get_connectaddr(sock,FALSE);
pj->ji_qs.ji_un.ji_newt.ji_scriptsz = 0;
/* acknowledge the request with the job id */
if (reply_jobid(preq, pj->ji_qs.ji_jobid, BATCH_REPLY_CHOICE_Queue) != 0)
{
/* reply failed, purge the job and close the connection */
// call mom_job_purge first so that double-frees don't happen
// when the on_close function is called
mom_job_purge(pj);
close_conn(sock, FALSE);
return;
}
/* link job into server's new jobs list request */
append_link(&svr_newjobs, &pj->ji_alljobs, pj);
return;
} /* END mom_req_quejob() */
