static int
jobResumeAction (struct jobCard *jp, int sigValue, int suspReason)
{
static char fname[] = "jobResumeAction";
if (jp->jobSpecs.reasons & SUSP_MBD_LOCK) {
return -1;
};
if (jp->jobSpecs.actPid)
return 0;
if (!(jp->jobSpecs.reasons & suspReason))
return -1;
if (logclass & (LC_TRACE | LC_SCHED | LC_EXEC))
ls_syslog(LOG_DEBUG1, "%s: Try to resume job %s with the current reason %d and the triggered reason %d;",
fname, lsb_jobid2str(jp->jobSpecs.jobId), jp->jobSpecs.reasons, suspReason);
if (jobSigStart(jp, sigValue, 0, 0, SIGLOG) < 0)
if (jobsig(jp, 0, FALSE) < 0) {
SBD_SET_STATE(jp, JOB_STAT_EXIT);
return -1;
}
sbdlog_newstatus(jp);
return 0;
}
int
job_resume (struct jobCard *jp)
{
static char fname[] = "job_resume";
int rep;
if (jp->jobSpecs.actPid)
return 0;
if (jobsig(jp, SIGCONT, FALSE) < 0)
return -1;
SBD_SET_STATE(jp, JOB_STAT_RUN);
jp->jobSpecs.reasons = 0;
jp->jobSpecs.subreasons = 0;
rep = status_job (BATCH_STATUS_JOB, jp, jp->jobSpecs.jStatus,
ERR_NO_ERROR);
if (rep < 0)
jp->notReported++;
else {
if (jp->notReported > 0)
jp->notReported = 0;
}
if (logclass & (LC_TRACE | LC_SCHED | LC_EXEC))
ls_syslog(LOG_DEBUG1, "%s: Resume job %s",
fname, lsb_jobid2str(jp->jobSpecs.jobId));
return 0;
}
void
do_sigjob(XDR * xdrs, int chfd, struct LSFHeader * reqHdr)
{
static char fname[] = "do_sigjob()";
char reply_buf[MSGSIZE];
XDR xdrs2;
struct jobSig jobSig;
sbdReplyType reply;
struct jobReply jobReply;
struct LSFHeader replyHdr;
char *replyStruct;
struct jobCard *jp = NULL;
char found = FALSE;
int cc;
int sigValue;
int savedActReasons;
int savedActSubReasons;
struct lsfAuth *auth = NULL;
memset(&jobReply, 0, sizeof(struct jobReply));
if (!xdr_jobSig(xdrs, &jobSig, reqHdr)) {
reply = ERR_BAD_REQ;
ls_syslog(LOG_ERR, I18N_FUNC_FAIL, fname, "xdr_jobSig");
goto Reply1;
}
jobSig.sigValue = sig_decode(jobSig.sigValue);
sigValue = jobSig.sigValue;
if (logclass & LC_SIGNAL)
ls_syslog(LOG_DEBUG, "do_sigJob: sigValue =%d", sigValue);
for (jp = jobQueHead->forw; (jp != jobQueHead); jp = jp->forw) {
if (jp->jobSpecs.jobId != jobSig.jobId)
continue;
found = TRUE;
break;
}
if (found == FALSE) {
reply = ERR_NO_JOB;
jp = NULL;
goto Reply1;
}
if (jobSig.reasons & SUSP_MBD_LOCK) {
jp->jobSpecs.reasons = jobSig.reasons;
jp->jobSpecs.subreasons = jobSig.subReasons;
savedActReasons = jp->actReasons;
savedActSubReasons = jp->actSubReasons;
jp->actReasons = jobSig.reasons;
jp->actSubReasons = jobSig.subReasons;
}
if (jp->postJobStarted) {
reply = ERR_NO_ERROR;
goto Reply1;
}
if (IS_FINISH(jp->jobSpecs.jStatus)) {
reply = ERR_NO_ERROR;
goto Reply1;
}
if (jp->jobSpecs.jobPGid == -1) {
SBD_SET_STATE(jp, JOB_STAT_EXIT);
reply = ERR_NO_ERROR;
goto Reply;
}
if (!JOB_STARTED(jp)) {
if (isSigTerm(sigValue) == TRUE) {
if ((cc = jobSigStart (jp, sigValue, jobSig.actFlags, jobSig.chkPeriod, NO_SIGLOG)) < 0)
reply = ERR_SIG_RETRY;
else
reply = ERR_NO_ERROR;
goto Reply;
}
reply = ERR_SIG_RETRY;
if (logclass & LC_EXEC)
ls_syslog(LOG_DEBUG, "%s: Retry signal %s for job <%s>",
fname, getLsbSigSymbol(sigValue),
lsb_jobid2str(jp->jobSpecs.jobId));
goto Reply1;
}
if (IS_PEND(jp->jobSpecs.jStatus)) {
reply = ERR_SIG_RETRY;
goto Reply1;
}
if (jp->jobSpecs.actPid || (jp->jobSpecs.jStatus & JOB_STAT_MIG)) {
if ((cc = jobSigStart(jp,
sigValue,
jobSig.actFlags,
jobSig.chkPeriod,
NO_SIGLOG)) < 0)
reply = ERR_SIG_RETRY;
else {
jp->jobSpecs.jStatus &= ~JOB_STAT_MIG;
reply = ERR_NO_ERROR;
}
goto Reply;
}
if ((cc = jobSigStart(jp,
sigValue,
jobSig.actFlags,
jobSig.chkPeriod,
NO_SIGLOG)) < 0)
reply = ERR_SIG_RETRY;
else
reply = ERR_NO_ERROR;
Reply:
sbdlog_newstatus(jp);
Reply1:
xdrmem_create(&xdrs2, reply_buf, MSGSIZE, XDR_ENCODE);
initLSFHeader_(&replyHdr);
replyHdr.opCode = reply;
if (reply == ERR_NO_ERROR) {
jobReply.jobPid = jp->jobSpecs.jobPid;
jobReply.actPid = jp->jobSpecs.actPid;
jobReply.jobId = jp->jobSpecs.jobId;
jobReply.jobPGid = jp->jobSpecs.jobPGid;
jobReply.jStatus = jp->jobSpecs.jStatus;
jobReply.reasons = jp->jobSpecs.reasons;
jobReply.actStatus = jp->actStatus;
replyStruct = (char *) &jobReply;
} else {
if (reply != ERR_NO_JOB)
if ((jp != NULL) && (jobSig.reasons & SUSP_MBD_LOCK)) {
jp->actReasons = savedActReasons;
jp->actSubReasons = savedActSubReasons;
}
replyStruct = (char *) 0;
}
if (!xdr_encodeMsg(&xdrs2, replyStruct, &replyHdr, xdr_jobReply, 0, auth)) {
ls_syslog(LOG_ERR, I18N_JOB_FAIL_S_M, fname,
lsb_jobid2str(jp->jobSpecs.jobId), "xdr_jobReply");
relife();
}
if (chanWrite_(chfd, reply_buf, XDR_GETPOS(&xdrs2)) <= 0) {
ls_syslog(LOG_ERR, _i18n_msg_get(ls_catd , NL_SETN, 5821,
"%s: Sending jobReply (len=%d) to master failed for job <%s>: %m"), fname, XDR_GETPOS(&xdrs2), lsb_jobid2str(jobSig.jobId)); /* catgets 5821 */
}
if (jp != NULL)
jp->actStatus = ACT_NO;
xdr_destroy(&xdrs2);
return;
}
static void
chkpntEnd (struct jobCard *jobCard, int w_status, bool_t *freed)
{
static char fname[] = "chkpntEnd()";
int savePid, saveStatus;
if (IS_SUSP(jobCard->jobSpecs.jStatus)
&& !(jobCard->jobSpecs.jStatus & JOB_STAT_MIG))
jobsig(jobCard, SIGSTOP, TRUE);
saveStatus = jobCard->jobSpecs.jStatus;
if (jobCard->jobSpecs.jStatus & JOB_STAT_MIG) {
if (w_status == 0) {
if (!jobCard->missing) {
jobCard->missing = TRUE;
need_checkfinish = TRUE;
return;
} else if (jobCard->notReported == 0)
return;
if (jobCard->cleanupPid == 0) {
if ((jobCard->cleanupPid = rmJobBufFilesPid(jobCard)) > 0)
return;
ls_syslog(LOG_ERR, _i18n_msg_get(ls_catd , NL_SETN, 5709,
"%s: Unable to cleanup migrating job <%s>"), /* catgets 5709 */
fname, lsb_jobid2str(jobCard->jobSpecs.jobId));
}
SBD_SET_STATE(jobCard, JOB_STAT_PEND);
} else {
jobCard->jobSpecs.jStatus &= ~JOB_STAT_MIG;
}
}
savePid = jobCard->jobSpecs.actPid;
if (status_job (BATCH_STATUS_JOB, jobCard, jobCard->jobSpecs.jStatus,
w_status == 0 ? ERR_NO_ERROR :
ERR_SYSACT_FAIL) < 0) {
jobCard->jobSpecs.actPid = savePid;
jobCard->jobSpecs.jStatus = saveStatus;
} else {
jobCard->lastChkpntTime = now;
jobCard->jobSpecs.actPid = 0;
jobCard->actStatus = ACT_NO;
jobCard->jobSpecs.actValue = SIG_NULL;
if (w_status == 0) {
jobCard->migCnt = 1;
}
if (saveStatus & JOB_STAT_MIG) {
if (w_status == 0) {
cleanupMigJob(jobCard);
deallocJobCard(jobCard);
*freed = TRUE;
} else
jobCard->migCnt *= 2;
}
}
}
void
do_newjob(XDR *xdrs, int chfd, struct LSFHeader *reqHdr)
{
static char fname[] = "do_newjob()";
char reply_buf[MSGSIZE];
XDR xdrs2;
struct jobSpecs jobSpecs;
struct jobReply jobReply;
struct jobCard *jp;
sbdReplyType reply;
struct LSFHeader replyHdr;
char *replyStruct;
struct lsfAuth *auth = NULL;
memset(&jobReply, 0, sizeof(struct jobReply));
if (!xdr_jobSpecs(xdrs, &jobSpecs, reqHdr)) {
reply = ERR_BAD_REQ;
ls_syslog(LOG_ERR, I18N_FUNC_FAIL, fname, "xdr_jobSpecs");
goto sendReply;
}
for (jp = jobQueHead->forw; (jp != jobQueHead); jp = jp->forw) {
if (jp->jobSpecs.jobId == jobSpecs.jobId) {
jobReply.jobId = jp->jobSpecs.jobId;
jobReply.jobPid = jp->jobSpecs.jobPid;
jobReply.jobPGid = jp->jobSpecs.jobPGid;
jobReply.jStatus = jp->jobSpecs.jStatus;
reply = ERR_NO_ERROR;
goto sendReply;
}
}
jp = calloc(1, sizeof(struct jobCard));
if (jp == NULL) {
ls_syslog(LOG_ERR, I18N_JOB_FAIL_S_M, fname,
lsb_jobid2str(jobSpecs.jobId), "calloc");
reply = ERR_MEM;
goto sendReply;
}
memcpy((char *) &jp->jobSpecs, (char *) &jobSpecs,
sizeof(struct jobSpecs));
jp->jobSpecs.jStatus &= ~JOB_STAT_MIG;
jp->jobSpecs.startTime = now;
jp->jobSpecs.reasons = 0;
jp->jobSpecs.subreasons = 0;
/* Initialize the core number
*/
jp->core_num = -1;
if (jp->jobSpecs.jAttrib & Q_ATTRIB_EXCLUSIVE) {
if (lockHosts (jp) < 0) {
ls_syslog(LOG_ERR, I18N_JOB_FAIL_S, fname,
lsb_jobid2str(jp->jobSpecs.jobId), "lockHosts");
unlockHosts (jp, jp->jobSpecs.numToHosts);
reply = ERR_LOCK_FAIL;
freeWeek(jp->week);
FREEUP(jp);
goto sendReply;
}
}
jp->runTime = 0;
if (initJobCard(jp, &jobSpecs, (int *)&reply) < 0) {
if (jp->jobSpecs.jAttrib & Q_ATTRIB_EXCLUSIVE) {
unlockHosts (jp, jp->jobSpecs.numToHosts);
}
FREEUP(jp);
goto sendReply;
}
jp->execJobFlag = 0;
if (jp->runTime < 0) {
jp->runTime = 0;
}
jp->execGid = 0;
jp->execUsername[0] = '\0';
jp->jobSpecs.execUid = -1;
jp->jobSpecs.execUsername[0] = '\0';
if (jp->jobSpecs.jobSpoolDir[0] != '\0') {
char *tmp;
if ((tmp = getUnixSpoolDir (jp->jobSpecs.jobSpoolDir)) == NULL) {
jp->jobSpecs.jobSpoolDir[0] = '\0';
}
}
if ((logclass & LC_TRACE) && jp->jobSpecs.jobSpoolDir[0] != 0) {
ls_syslog(LOG_DEBUG,
"%s: the SpoolDir for job <%s> is %s \n",
fname, lsb_jobid2str(jp->jobSpecs.jobId),
jp->jobSpecs.jobSpoolDir);
}
if (jp->jobSpecs.options & SUB_PRE_EXEC)
SBD_SET_STATE(jp, (JOB_STAT_RUN | JOB_STAT_PRE_EXEC))
else
SBD_SET_STATE(jp, JOB_STAT_RUN);
reply = job_exec(jp, chfd);
if (reply != ERR_NO_ERROR) {
ls_syslog(LOG_ERR, I18N_JOB_FAIL_S, fname,
lsb_jobid2str(jp->jobSpecs.jobId), "job_exec");
if (jp->jobSpecs.jAttrib & Q_ATTRIB_EXCLUSIVE) {
unlockHosts (jp, jp->jobSpecs.numToHosts);
}
deallocJobCard(jp);
} else {
jobReply.jobId = jp->jobSpecs.jobId;
jobReply.jobPid = jp->jobSpecs.jobPid;
jobReply.jobPGid = jp->jobSpecs.jobPGid;
jobReply.jStatus = jp->jobSpecs.jStatus;
}
sendReply:
xdr_lsffree(xdr_jobSpecs, (char *)&jobSpecs, reqHdr);
xdrmem_create(&xdrs2, reply_buf, MSGSIZE, XDR_ENCODE);
initLSFHeader_(&replyHdr);
replyHdr.opCode = reply;
replyStruct = (reply == ERR_NO_ERROR) ? (char *) &jobReply : (char *) NULL;
if (!xdr_encodeMsg(&xdrs2, replyStruct, &replyHdr, xdr_jobReply, 0, auth)) {
ls_syslog(LOG_ERR, I18N_FUNC_FAIL, fname, "xdr_jobReply");
lsb_merr(_i18n_msg_get(ls_catd , NL_SETN, 5804,
"Fatal error: xdr_jobReply() failed; sbatchd relifing")); /* catgets 5804 */
relife();
}
if (chanWrite_(chfd, reply_buf, XDR_GETPOS(&xdrs2)) <= 0) {
ls_syslog(LOG_ERR, _i18n_msg_get(ls_catd , NL_SETN, 5805,
"%s: Sending jobReply (len=%d) to master failed: %m"), /* catgets 5805 */
fname, XDR_GETPOS(&xdrs2));
}
xdr_destroy(&xdrs2);
if (reply == ERR_NO_ERROR && !daemonParams[LSB_BSUBI_OLD].paramValue &&
PURE_INTERACTIVE(&jp->jobSpecs)) {
if (status_job (BATCH_STATUS_JOB, jp, jp->jobSpecs.jStatus,
ERR_NO_ERROR) < 0) {
jp->notReported++;
}
}
}
