static job *locate_new_job(
int sock, /* I */
char *jobid) /* I (optional) */
{
job *pj;
pj = (job *)GET_NEXT(svr_newjobs);
while (pj != NULL)
{
if ((pj->ji_qs.ji_un.ji_newt.ji_fromsock == -1) ||
((pj->ji_qs.ji_un.ji_newt.ji_fromsock == sock) &&
(pj->ji_qs.ji_un.ji_newt.ji_fromaddr == get_connectaddr(sock))))
{
if ((jobid != NULL) && (*jobid != '\0'))
{
if (!strncmp(pj->ji_qs.ji_jobid, jobid, PBS_MAXSVRJOBID))
{
/* requested job located */
break;
}
}
else if (pj->ji_qs.ji_un.ji_newt.ji_fromsock == -1)
{
/* empty job slot located */
break;
}
else
{
/* matching job slot located */
break;
}
} /* END if ((pj->ji_qs.ji_un.ji_newt.ji_fromsock == -1) || ...) */
pj = (job *)GET_NEXT(pj->ji_alljobs);
} /* END while(pj != NULL) */
/* return job slot located (NULL on FAILURE) */
return(pj);
} /* END locate_new_job() */
void
process_request(int sfds)
{
int rc;
struct batch_request *request;
conn_t *conn;
time_now = time(NULL);
conn = get_conn(sfds);
if (!conn) {
log_event(PBSEVENT_SYSTEM, PBS_EVENTCLASS_REQUEST, LOG_ERR,
"process_request", "did not find socket in connection table");
#ifdef WIN32
(void)closesocket(sfds);
#else
(void)close(sfds);
#endif
return;
}
if ((request = alloc_br(0)) == NULL) {
log_event(PBSEVENT_SYSTEM, PBS_EVENTCLASS_REQUEST, LOG_ERR,
"process_request", "Unable to allocate request structure");
close_conn(sfds);
return;
}
request->rq_conn = sfds;
/*
* Read in the request and decode it to the internal request structure.
*/
if (get_connecthost(sfds, request->rq_host, PBS_MAXHOSTNAME)) {
(void)sprintf(log_buffer, "%s: %lu", msg_reqbadhost,
get_connectaddr(sfds));
log_event(PBSEVENT_DEBUG, PBS_EVENTCLASS_REQUEST, LOG_DEBUG,
"", log_buffer);
req_reject(PBSE_BADHOST, 0, request);
return;
}
#ifndef PBS_MOM
if (conn->cn_active == FromClientDIS) {
rc = dis_request_read(sfds, request);
} else {
log_event(PBSEVENT_SYSTEM, PBS_EVENTCLASS_REQUEST, LOG_ERR,
"process_req", "request on invalid type of connection");
close_conn(sfds);
free_br(request);
return;
}
#else /* PBS_MOM */
rc = dis_request_read(sfds, request);
#endif /* PBS_MOM */
if (rc == -1) { /* End of file */
close_client(sfds);
free_br(request);
return;
} else if ((rc == PBSE_SYSTEM) || (rc == PBSE_INTERNAL)) {
/* read error, likely cannot send reply so just disconnect */
/* ??? not sure about this ??? */
close_client(sfds);
free_br(request);
return;
} else if (rc > 0) {
/*
* request didn't decode, either garbage or unknown
* request type, in ether case, return reject-reply
*/
req_reject(rc, 0, request);
close_client(sfds);
return;
}
#ifndef PBS_MOM
/* If the request is coming on the socket we opened to the */
/* scheduler, change the "user" from "root" to "Scheduler" */
if (find_sched_from_sock(request->rq_conn) != NULL) {
strncpy(request->rq_user, PBS_SCHED_DAEMON_NAME, PBS_MAXUSER);
request->rq_user[PBS_MAXUSER] = '\0';
}
#endif /* PBS_MOM */
(void)sprintf(log_buffer, msg_request, request->rq_type,
request->rq_user, request->rq_host, sfds);
log_event(PBSEVENT_DEBUG2, PBS_EVENTCLASS_REQUEST, LOG_DEBUG,
"", log_buffer);
/* is the request from a host acceptable to the server */
if (request->rq_type == PBS_BATCH_AuthExternal) {
rc = authenticate_external(conn, request);
if (rc == 0)
reply_ack(request);
else if (rc == -2)
req_reject(PBSE_NOSUP, 0, request);
else
req_reject(PBSE_BADCRED, 0, request);
return;
}
#ifndef PBS_MOM
if (server.sv_attr[(int)SRV_ATR_acl_host_enable].at_val.at_long) {
/* acl enabled, check it; always allow myself */
struct pbsnode *isanode = NULL;
if ((server.sv_attr[SRV_ATR_acl_host_moms_enable].at_flags & ATR_VFLAG_SET) &&
(server.sv_attr[(int)SRV_ATR_acl_host_moms_enable].at_val.at_long == 1)) {
isanode = find_nodebyaddr(get_connectaddr(sfds));
if ((isanode != NULL) && (isanode->nd_state & INUSE_DELETED))
isanode = NULL;
}
if (isanode == NULL) {
if ((acl_check(&server.sv_attr[(int)SRV_ATR_acl_hosts],
request->rq_host, ACL_Host) == 0) &&
(strcasecmp(server_host, request->rq_host) != 0)) {
req_reject(PBSE_BADHOST, 0, request);
close_client(sfds);
return;
}
}
}
/*
* determine source (user client or another server) of request.
* set the permissions granted to the client
*/
if (conn->cn_authen & PBS_NET_CONN_FROM_PRIVIL) {
/* request came from another server */
request->rq_fromsvr = 1;
request->rq_perm = ATR_DFLAG_USRD | ATR_DFLAG_USWR |
ATR_DFLAG_OPRD | ATR_DFLAG_OPWR |
ATR_DFLAG_MGRD | ATR_DFLAG_MGWR |
ATR_DFLAG_SvWR;
} else {
/* request not from another server */
request->rq_fromsvr = 0;
/*
* Client must be authenticated by a Authenticate User Request,
* if not, reject request and close connection.
* -- The following is retained for compat with old cmds --
* The exception to this is of course the Connect Request which
* cannot have been authenticated, because it contains the
* needed ticket; so trap it here. Of course, there is no
* prior authentication on the Authenticate User request either,
* but it comes over a reserved port and appears from another
* server, hence is automatically granted authorization.
*/
if (request->rq_type == PBS_BATCH_Connect) {
req_connect(request);
return;
}
if ((conn->cn_authen & PBS_NET_CONN_AUTHENTICATED) ==0) {
rc = PBSE_BADCRED;
} else {
rc = authenticate_user(request, conn);
}
if (rc != 0) {
req_reject(rc, 0, request);
if (rc == PBSE_BADCRED)
close_client(sfds);
return;
}
request->rq_perm =
svr_get_privilege(request->rq_user, request->rq_host);
}
/* if server shutting down, disallow new jobs and new running */
if (server.sv_attr[(int)SRV_ATR_State].at_val.at_long > SV_STATE_RUN) {
switch (request->rq_type) {
case PBS_BATCH_AsyrunJob:
case PBS_BATCH_JobCred:
case PBS_BATCH_UserCred:
case PBS_BATCH_UserMigrate:
case PBS_BATCH_MoveJob:
case PBS_BATCH_QueueJob:
case PBS_BATCH_RunJob:
case PBS_BATCH_StageIn:
case PBS_BATCH_jobscript:
req_reject(PBSE_SVRDOWN, 0, request);
return;
}
}
#else /* THIS CODE FOR MOM ONLY */
/* check connecting host against allowed list of ok clients */
if (!addrfind(conn->cn_addr)) {
req_reject(PBSE_BADHOST, 0, request);
close_client(sfds);
return;
}
request->rq_fromsvr = 1;
request->rq_perm = ATR_DFLAG_USRD | ATR_DFLAG_USWR |
ATR_DFLAG_OPRD | ATR_DFLAG_OPWR |
ATR_DFLAG_MGRD | ATR_DFLAG_MGWR |
ATR_DFLAG_SvWR | ATR_DFLAG_MOM;
#endif
/*
* dispatch the request to the correct processing function.
* The processing function must call reply_send() to free
* the request struture.
*/
dispatch_request(sfds, request);
return;
}
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 process_request(
int sfds) /* file descriptor (socket) to get request */
{
#ifdef PBS_MOM
char *id = "process_request";
#endif
int rc;
struct batch_request *request = NULL;
#ifndef PBS_MOM
char *auth_err = NULL;
#endif
time_now = time(NULL);
request = alloc_br(0);
request->rq_conn = sfds;
/*
* Read in the request and decode it to the internal request structure.
*/
#ifndef PBS_MOM
if (svr_conn[sfds].cn_active == FromClientDIS)
{
#ifdef ENABLE_UNIX_SOCKETS
if ((svr_conn[sfds].cn_socktype & PBS_SOCK_UNIX) &&
(svr_conn[sfds].cn_authen != PBS_NET_CONN_AUTHENTICATED))
{
get_creds(sfds, conn_credent[sfds].username, conn_credent[sfds].hostname);
}
#endif /* END ENABLE_UNIX_SOCKETS */
rc = dis_request_read(sfds, request);
}
else
{
LOG_EVENT(
PBSEVENT_SYSTEM,
PBS_EVENTCLASS_REQUEST,
"process_req",
"request on invalid type of connection");
close_conn(sfds);
free_br(request);
return;
}
#else /* PBS_MOM */
rc = dis_request_read(sfds, request);
#endif /* PBS_MOM */
if (rc == -1)
{
/* FAILURE */
/* premature end of file */
close_client(sfds);
free_br(request);
return;
}
if ((rc == PBSE_SYSTEM) || (rc == PBSE_INTERNAL))
{
/* FAILURE */
/* read error, likely cannot send reply so just disconnect */
/* ??? not sure about this ??? */
close_client(sfds);
free_br(request);
return;
}
if (rc > 0)
{
/* FAILURE */
/*
* request didn't decode, either garbage or unknown
* request type, in either case, return reject-reply
*/
req_reject(rc, 0, request, NULL, "cannot decode message");
close_client(sfds);
return;
}
if (get_connecthost(sfds, request->rq_host, PBS_MAXHOSTNAME) != 0)
{
char tmpLine[1024];
sprintf(log_buffer, "%s: %lu",
pbse_to_txt(PBSE_BADHOST),
get_connectaddr(sfds));
LOG_EVENT(PBSEVENT_DEBUG, PBS_EVENTCLASS_REQUEST, "", log_buffer);
snprintf(tmpLine, sizeof(tmpLine), "cannot determine hostname for connection from %lu",
get_connectaddr(sfds));
req_reject(PBSE_BADHOST, 0, request, NULL, tmpLine);
return;
}
if (LOGLEVEL >= 1)
{
sprintf(
log_buffer,
msg_request,
reqtype_to_txt(request->rq_type),
request->rq_user,
request->rq_host,
sfds);
LOG_EVENT(PBSEVENT_DEBUG2, PBS_EVENTCLASS_REQUEST, "", log_buffer);
}
/* is the request from a host acceptable to the server */
#ifndef PBS_MOM
if (svr_conn[sfds].cn_socktype & PBS_SOCK_UNIX)
{
strcpy(request->rq_host, server_name);
}
if (server.sv_attr[SRV_ATR_acl_host_enable].at_val.at_long)
{
/* acl enabled, check it; always allow myself and nodes */
struct pbsnode *isanode;
isanode = PGetNodeFromAddr(get_connectaddr(sfds));
if ((isanode == NULL) &&
(strcmp(server_host, request->rq_host) != 0) &&
(acl_check(
&server.sv_attr[SRV_ATR_acl_hosts],
request->rq_host,
ACL_Host) == 0))
{
char tmpLine[1024];
snprintf(tmpLine, sizeof(tmpLine), "request not authorized from host %s",
request->rq_host);
req_reject(PBSE_BADHOST, 0, request, NULL, tmpLine);
close_client(sfds);
return;
}
}
/*
* determine source (user client or another server) of request.
* set the permissions granted to the client
*/
if (svr_conn[sfds].cn_authen == PBS_NET_CONN_FROM_PRIVIL)
{
/* request came from another server */
request->rq_fromsvr = 1;
request->rq_perm =
ATR_DFLAG_USRD | ATR_DFLAG_USWR |
ATR_DFLAG_OPRD | ATR_DFLAG_OPWR |
ATR_DFLAG_MGRD | ATR_DFLAG_MGWR |
ATR_DFLAG_SvWR;
}
else
{
/* request not from another server */
request->rq_fromsvr = 0;
/*
* Client must be authenticated by an Authenticate User Request, if not,
* reject request and close connection. -- The following is retained for
* compat with old cmds -- The exception to this is of course the Connect
* Request which cannot have been authenticated, because it contains the
* needed ticket; so trap it here. Of course, there is no prior
* authentication on the Authenticate User request either, but it comes
* over a reserved port and appears from another server, hence is
* automatically granted authentication.
*
* The above is only true with inet sockets. With unix domain sockets, the
* user creds were read before the first dis_request_read call above.
* We automatically granted authentication because we can trust the socket
* creds. Authorization is still granted in svr_get_privilege below
*/
if (request->rq_type == PBS_BATCH_Connect)
{
req_connect(request);
if (svr_conn[sfds].cn_socktype == PBS_SOCK_INET)
return;
}
if (svr_conn[sfds].cn_socktype & PBS_SOCK_UNIX)
{
conn_credent[sfds].timestamp = time_now;
svr_conn[sfds].cn_authen = PBS_NET_CONN_AUTHENTICATED;
}
if (ENABLE_TRUSTED_AUTH == TRUE )
rc = 0; /* bypass the authentication of the user--trust the client completely */
else if (munge_on)
{
/* If munge_on is true we will validate the connection now */
if ( request->rq_type == PBS_BATCH_AltAuthenUser)
{
rc = req_altauthenuser(request);
if (rc == PBSE_NONE)
{
conn_credent[sfds].timestamp = time_now;
svr_conn[sfds].cn_authen = PBS_NET_CONN_AUTHENTICATED;
}
return;
}
else if (svr_conn[sfds].cn_authen != PBS_NET_CONN_AUTHENTICATED)
/* skip checking user if we did not get an authenticated credential */
rc = PBSE_BADCRED;
else
{
rc = authenticate_user(request, &conn_credent[sfds], &auth_err);
}
}
else if (svr_conn[sfds].cn_authen != PBS_NET_CONN_AUTHENTICATED)
rc = PBSE_BADCRED;
else
rc = authenticate_user(request, &conn_credent[sfds], &auth_err);
if (rc != 0)
{
req_reject(rc, 0, request, NULL, auth_err);
if (auth_err != NULL)
free(auth_err);
close_client(sfds);
return;
}
/*
* pbs_mom and checkpoint restart scripts both need the authority to do
* alters and releases on checkpointable jobs. Allow manager permission
* for root on the jobs execution node.
*/
if (((request->rq_type == PBS_BATCH_ModifyJob) ||
(request->rq_type == PBS_BATCH_ReleaseJob)) &&
(strcmp(request->rq_user, PBS_DEFAULT_ADMIN) == 0))
{
job *pjob;
char *dptr;
int skip = FALSE;
char short_host[PBS_MAXHOSTNAME+1];
/* make short host name */
strcpy(short_host, request->rq_host);
if ((dptr = strchr(short_host, '.')) != NULL)
{
*dptr = '\0';
}
if (((pjob = find_job(request->rq_ind.rq_modify.rq_objname)) != (job *)0) &&
(pjob->ji_qs.ji_state == JOB_STATE_RUNNING))
{
if ((pjob->ji_wattr[JOB_ATR_checkpoint].at_flags & ATR_VFLAG_SET) &&
((csv_find_string(pjob->ji_wattr[JOB_ATR_checkpoint].at_val.at_str, "s") != NULL) ||
(csv_find_string(pjob->ji_wattr[JOB_ATR_checkpoint].at_val.at_str, "c") != NULL) ||
(csv_find_string(pjob->ji_wattr[JOB_ATR_checkpoint].at_val.at_str, "enabled") != NULL)) &&
(strstr(pjob->ji_wattr[JOB_ATR_exec_host].at_val.at_str, short_host) != NULL))
{
request->rq_perm = svr_get_privilege(request->rq_user, server_host);
skip = TRUE;
}
}
if (!skip)
{
request->rq_perm = svr_get_privilege(request->rq_user, request->rq_host);
}
}
else
{
request->rq_perm = svr_get_privilege(request->rq_user, request->rq_host);
}
} /* END else (svr_conn[sfds].cn_authen == PBS_NET_CONN_FROM_PRIVIL) */
/* if server shutting down, disallow new jobs and new running */
if (server.sv_attr[SRV_ATR_State].at_val.at_long > SV_STATE_RUN)
{
switch (request->rq_type)
{
case PBS_BATCH_AsyrunJob:
case PBS_BATCH_JobCred:
case PBS_BATCH_MoveJob:
case PBS_BATCH_QueueJob:
case PBS_BATCH_RunJob:
case PBS_BATCH_StageIn:
case PBS_BATCH_jobscript:
req_reject(PBSE_SVRDOWN, 0, request, NULL, NULL);
return;
/*NOTREACHED*/
break;
}
}
#else /* THIS CODE FOR MOM ONLY */
{
/*extern tree *okclients; */
extern void mom_server_update_receive_time_by_ip(u_long ipaddr, const char *cmd);
/* check connecting host against allowed list of ok clients */
if (LOGLEVEL >= 6)
{
sprintf(log_buffer, "request type %s from host %s received",
reqtype_to_txt(request->rq_type),
request->rq_host);
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
id,
log_buffer);
}
/* if (!tfind(svr_conn[sfds].cn_addr, &okclients)) */
if (!AVL_is_in_tree(svr_conn[sfds].cn_addr, 0, okclients))
{
sprintf(log_buffer, "request type %s from host %s rejected (host not authorized)",
reqtype_to_txt(request->rq_type),
request->rq_host);
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
id,
log_buffer);
req_reject(PBSE_BADHOST, 0, request, NULL, "request not authorized");
close_client(sfds);
return;
}
if (LOGLEVEL >= 3)
{
sprintf(log_buffer, "request type %s from host %s allowed",
reqtype_to_txt(request->rq_type),
request->rq_host);
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
id,
log_buffer);
}
mom_server_update_receive_time_by_ip(svr_conn[sfds].cn_addr, reqtype_to_txt(request->rq_type));
} /* END BLOCK */
request->rq_fromsvr = 1;
request->rq_perm =
ATR_DFLAG_USRD | ATR_DFLAG_USWR |
ATR_DFLAG_OPRD | ATR_DFLAG_OPWR |
ATR_DFLAG_MGRD | ATR_DFLAG_MGWR |
ATR_DFLAG_SvWR | ATR_DFLAG_MOM;
#endif /* END else !PBS_MOM */
/*
* dispatch the request to the correct processing function.
* The processing function must call reply_send() to free
* the request struture.
*/
dispatch_request(sfds, request);
return;
} /* END process_request() */
void req_commit(
struct batch_request *preq) /* I */
{
unsigned int momport = 0;
int rc;
job *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", (char *)"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);
alljobs_list.push_back(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,FALSE);
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->ji_qs.ji_jobid, "req_commit:starting job execution");
rc = start_exec(pj);
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
pj->ji_qs.ji_jobid,
"req_commit: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->ji_qs.ji_jobid, tmpLine);
}
reply_text(preq, rc, tmpLine);
}
else
{
reply_sid(preq, pj->ji_wattr[JOB_ATR_session_id].at_val.at_long,BATCH_REPLY_CHOICE_Text);
}
if (multi_mom)
{
momport = pbs_rm_port;
}
job_save(pj, SAVEJOB_FULL, momport);
#ifdef NVIDIA_GPUS
/*
* Does this job have a gpuid assigned?
* if so, then update gpu status
*/
if ((use_nvidia_gpu) &&
((pj->ji_wattr[JOB_ATR_exec_gpus].at_flags & ATR_VFLAG_SET) != 0) &&
(pj->ji_wattr[JOB_ATR_exec_gpus].at_val.at_str != NULL))
{
send_update_soon();
}
#endif /* NVIDIA_GPUS */
/* 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 pbs_attribute "SEND" flag
* was added to handle this process. */
return;
} /* END req_commit() */
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() */
void *mom_process_request(
void *sock_num) /* file descriptor (socket) to get request */
{
int rc;
struct batch_request *request = NULL;
int sfds = *(int *)sock_num;
struct tcp_chan *chan = NULL;
time_now = time(NULL);
if ((request = alloc_br(0)) == NULL)
{
mom_close_client(sfds);
return NULL;
}
request->rq_conn = sfds;
if ((chan = DIS_tcp_setup(sfds)) == NULL)
{
mom_close_client(sfds);
free_br(request);
return NULL;
}
/* Read in the request and decode it to the internal request structure. */
rc = dis_request_read(chan, request);
if (rc == -1)
{
/* FAILURE */
/* premature end of file */
mom_close_client(chan->sock);
free_br(request);
DIS_tcp_cleanup(chan);
return NULL;
}
if ((rc == PBSE_SYSTEM) || (rc == PBSE_INTERNAL) || (rc == PBSE_SOCKET_CLOSE))
{
/* FAILURE */
/* read error, likely cannot send reply so just disconnect */
/* ??? not sure about this ??? */
mom_close_client(chan->sock);
free_br(request);
DIS_tcp_cleanup(chan);
return NULL;
}
if (rc > 0)
{
/* FAILURE */
/*
* request didn't decode, either garbage or unknown
* request type, in either case, return reject-reply
*/
req_reject(rc, 0, request, NULL, "cannot decode message");
mom_close_client(chan->sock);
DIS_tcp_cleanup(chan);
return NULL;
}
if (get_connecthost(chan->sock, request->rq_host, PBS_MAXHOSTNAME) != 0)
{
char tmpLine[MAXLINE];
sprintf(log_buffer, "%s: %lu",
pbse_to_txt(PBSE_BADHOST),
get_connectaddr(chan->sock,FALSE));
log_event(PBSEVENT_DEBUG, PBS_EVENTCLASS_REQUEST, "", log_buffer);
snprintf(tmpLine, sizeof(tmpLine), "cannot determine hostname for connection from %lu",
get_connectaddr(chan->sock,FALSE));
req_reject(PBSE_BADHOST, 0, request, NULL, tmpLine);
mom_close_client(chan->sock);
DIS_tcp_cleanup(chan);
return NULL;
}
if (LOGLEVEL >= 1)
{
sprintf(
log_buffer,
msg_request,
reqtype_to_txt(request->rq_type),
request->rq_user,
request->rq_host,
chan->sock);
log_event(PBSEVENT_DEBUG2, PBS_EVENTCLASS_REQUEST, "", log_buffer);
}
/* is the request from a host acceptable to the server */
{
/*extern tree *okclients; */
extern void mom_server_update_receive_time_by_ip(u_long ipaddr, const char *cmd);
/* check connecting host against allowed list of ok clients */
if (LOGLEVEL >= 6)
{
sprintf(log_buffer, "request type %s from host %s received",
reqtype_to_txt(request->rq_type),
request->rq_host);
log_record(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, __func__, log_buffer);
}
if (!AVL_is_in_tree_no_port_compare(svr_conn[chan->sock].cn_addr, 0, okclients))
{
sprintf(log_buffer, "request type %s from host %s rejected (host not authorized)",
reqtype_to_txt(request->rq_type),
request->rq_host);
log_record(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, __func__, log_buffer);
req_reject(PBSE_BADHOST, 0, request, NULL, "request not authorized");
mom_close_client(chan->sock);
DIS_tcp_cleanup(chan);
return NULL;
}
if (LOGLEVEL >= 3)
{
sprintf(log_buffer, "request type %s from host %s allowed",
reqtype_to_txt(request->rq_type),
request->rq_host);
log_record(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, __func__, log_buffer);
}
mom_server_update_receive_time_by_ip(svr_conn[chan->sock].cn_addr, reqtype_to_txt(request->rq_type));
} /* END BLOCK */
request->rq_fromsvr = 1;
request->rq_perm =
ATR_DFLAG_USRD | ATR_DFLAG_USWR |
ATR_DFLAG_OPRD | ATR_DFLAG_OPWR |
ATR_DFLAG_MGRD | ATR_DFLAG_MGWR |
ATR_DFLAG_SvWR | ATR_DFLAG_MOM;
/*
* dispatch the request to the correct processing function.
* The processing function must call reply_send() to free
* the request struture.
*/
mom_dispatch_request(chan->sock, request);
DIS_tcp_cleanup(chan);
return NULL;
} /* END mom_process_request() */
