/**
* @brief
* Restart the job from the checkpoint file.
*
* @param[in] ptask - pointer to task
* @param[in] file - filename
*
* @return long
* @retval session id Success
* @retval -1 Error
*/
long
mach_restart(task *ptask, char *file)
{
#if MOM_CHECKPOINT == 1
ckpt_id_t rc;
ash_t momash;
ash_t oldash = 0;
char cvtbuf[20];
cpr_flags = CKPT_NQE;
/* KLUDGE to work-around SGI problem, for some reason ckpt_restart() */
/* passes open file descriptor to /proc to restarted process */
if (pdir)
closedir(pdir);
/* To restart the job with its old ASH, Mom must be in that ASH */
/* When she does the restart. However, before changing to that */
/* ASH, Mom must put herself in a new ASH all by herself, otherwise */
/* she will take other system daemons with her into the job's ASH */
momash = getash();
newarraysess(); /* isolate Mom in a ASH by herself */
if (ptask->ti_job->ji_globid != NULL) {
/* now get job's old ASH and set it */
sscanf(ptask->ti_job->ji_globid, "%llx", &oldash);
if (setash(oldash) == -1) {
DBPRT(("setash failed before restart, errno = %d", errno))
}
}
int
set_job(job *pjob, struct startjob_rtn *sjr)
{
extern char noglobid[];
char cvtbuf[20];
if (pjob->ji_globid == NULL ||
strcmp(pjob->ji_globid, noglobid) == 0) {
if (newarraysess() == -1) {
sprintf(log_buffer, "newarraysess failed %d", errno);
return (-2);
}
sjr->sj_ash = getash();
}
else {
if (sscanf(pjob->ji_globid, "%llx", &sjr->sj_ash) != 1) {
sprintf(log_buffer, "invalid global id %s",
pjob->ji_globid);
return (-2);
}
if (syssgi(SGI_JOINARRAYSESS, -1, &sjr->sj_ash) == -1) {
sprintf(log_buffer, "cannot set ash %llx err %d",
sjr->sj_ash, errno);
return (-2);
}
}
if (set_sgi_proj(pjob->ji_wattr[(int)JOB_ATR_euser].at_val.at_str,
&pjob->ji_wattr[(int)JOB_ATR_account]) < 0) {
(void)sprintf(log_buffer, "Invalid project id");
return (-3);
}
return (sjr->sj_session = setsid());
}
void
irix_setusercontext(struct passwd *pw)
{
#ifdef WITH_IRIX_PROJECT
prid_t projid;
#endif
#ifdef WITH_IRIX_JOBS
jid_t jid = 0;
#elif defined(WITH_IRIX_ARRAY)
int jid = 0;
#endif
#ifdef WITH_IRIX_JOBS
jid = jlimit_startjob(pw->pw_name, pw->pw_uid, "interactive");
if (jid == -1)
fatal("Failed to create job container: %.100s",
strerror(errno));
#endif /* WITH_IRIX_JOBS */
#ifdef WITH_IRIX_ARRAY
/* initialize array session */
if (jid == 0 && newarraysess() != 0)
fatal("Failed to set up new array session: %.100s",
strerror(errno));
#endif /* WITH_IRIX_ARRAY */
#ifdef WITH_IRIX_PROJECT
/* initialize irix project info */
if ((projid = getdfltprojuser(pw->pw_name)) == -1) {
debug("Failed to get project id, using projid 0");
projid = 0;
}
if (setprid(projid))
fatal("Failed to initialize project %d for %s: %.100s",
(int)projid, pw->pw_name, strerror(errno));
#endif /* WITH_IRIX_PROJECT */
#ifdef WITH_IRIX_AUDIT
if (sysconf(_SC_AUDIT)) {
debug("Setting sat id to %d", (int) pw->pw_uid);
if (satsetid(pw->pw_uid))
debug("error setting satid: %.100s", strerror(errno));
}
#endif /* WITH_IRIX_AUDIT */
}
void setosjobid(pid_t sid, gid_t *add_grp_id_ptr, struct passwd *pw)
{
FILE *fp=NULL;
shepherd_trace("setosjobid: uid = "pid_t_fmt", euid = "pid_t_fmt, getuid(), geteuid());
# if defined(SOLARIS) || defined(ALPHA) || defined(LINUX) || defined(FREEBSD) || defined(DARWIN)
/* Read SgeId from config-File and create Addgrpid-File */
{
char *cp;
if ((cp = search_conf_val("add_grp_id")))
*add_grp_id_ptr = atol(cp);
else
*add_grp_id_ptr = 0;
}
if ((fp = fopen("addgrpid", "w")) == NULL) {
shepherd_error(1, "can't open \"addgrpid\" file");
}
fprintf(fp, gid_t_fmt"\n", *add_grp_id_ptr);
FCLOSE(fp);
# elif defined(HP1164) || defined(AIX)
{
if ((fp = fopen("addgrpid", "w")) == NULL) {
shepherd_error(1, "can't open \"addgrpid\" file");
}
fprintf(fp, pid_t_fmt"\n", getpgrp());
FCLOSE(fp);
}
# else
{
char osjobid[100];
if ((fp = fopen("osjobid", "w")) == NULL) {
shepherd_error(1, "can't open \"osjobid\" file");
}
if(sge_switch2start_user() == 0) {
# if defined(IRIX)
{
/* The following block contains the operations necessary for
* IRIX6.2 (and later) to set array session handles (ASHs) and
* service provider info (SPI) records
*/
struct acct_spi spi;
int ret;
char *cp;
shepherd_trace("in irix code");
/* get _local_ array session id */
if ((ret=newarraysess())) {
shepherd_error(1, "error: can't create ASH; errno=%d", ret);
}
/* retrieve array session id we just assigned to the process and
* write it to the os-jobid file
*/
sprintf(osjobid, "%lld", getash());
shepherd_trace(osjobid);
/* set service provider information (spi) record */
strncpy(spi.spi_company, "SGE", 8);
strncpy(spi.spi_initiator, get_conf_val("spi_initiator"), 8);
strncpy(spi.spi_origin, get_conf_val("queue"),16);
strcpy(spi.spi_spi, "Job ");
strncat(spi.spi_spi, get_conf_val("job_id"),11);
if ((ret=setspinfo(&spi))) {
shepherd_error(1, "error: can't set SPI; errno=%d", ret);
}
if ((cp = search_conf_val("acct_project"))) {
prid_t proj;
if (strcasecmp(cp, "none") && ((proj = projid(cp)) >= 0)) {
shepherd_trace("setting project \"%s\" to id %lld", cp, proj);
if (setprid(proj) == -1)
shepherd_trace("failed setting project id");
}
else {
shepherd_trace("can't get id for project \"%s\"", cp);
}
} else {
shepherd_trace("can't get configuration entry for projects");
}
}
# elif defined(CRAY)
{
char *cp;
{
int jobid;
if ((jobid=setjob(pw->pw_uid, 0)) < 0) {
shepherd_error(1, "error: can't set job ID; errno = %d", errno);
}
if (sesscntl(jobid, S_ADDFL, S_BATCH) == -1) {
shepherd_error(1, "error: sesscntl(%d, S_ADDFL, S_BATCH) failed,"
" errno = %d", sid, errno);
}
sprintf(osjobid, "%d", jobid);
}
if ((cp = search_conf_val("acct_project"))) {
int proj;
if (strcasecmp(cp, "none") && ((proj = nam2acid(cp)) >= 0)) {
shephed_trace("setting project \"%s\" to acid %d", cp, proj);
if (acctid(0, proj) == -1) {
shepherd_trace("failed setting project id (acctid)");
}
} else {
shepherd_trace("can't get id for project \"%s\"", cp);
}
} else {
shepherd_trace("can't get configuration entry for projects");
}
}
# elif defined(NECSX4) || defined(NECSX5)
{
id_t jobid = 0;
dispset2_t attr;
int value;
/*
* Create new Super-UX job
*/
if (setjid() == -1) {
shepherd_trace("ERROR: can't set jobid: %s[%d]", strerror(errno), errno);
} else {
jobid = getjid(0);
shepherd_trace("Created job with id: "sge_u32, (u_long32) jobid);
}
sprintf(osjobid, sge_u32, (u_long32) jobid);
/*
* We will use limits for the whole job
*/
set_rlimits_os_job_id(jobid);
/*
* The job will use the resources of the configured
* Resource Sharing Group (rsg)
*/
{
char *rsg_id_string;
int rsg_id;
char fsg_dev_string[256];
rsg_id_string = get_conf_val("processors");
rsg_id = atoi(rsg_id_string);
if (rsg_id) {
int fd;
sprintf(fsg_dev_string, "/dev/rsg/%d", rsg_id);
fd = open(fsg_dev_string, O_RDONLY);
if (fd <= 0) {
shepherd_trace("ERROR: can't switch to rsg%d because can't open"
"device: %s[%d]", rsg_id, strerror(errno), errno);
} else {
if (ioctl(fd, RSG_JUMP, NULL) == -1) {
close(fd);
shepherd_trace("ERROR: can't switch to rsg%d: %s[%d]",
rsg_id, strerror(errno), errno);
return;
} else {
close(fd);
shepherd_trace("switched to rsg%d", rsg_id);
}
}
} else {
shepherd_trace("using default rsg", rsg_id);
}
}
/*
* Set scheduling parameter for job
*/
if (((attr.basepri = atoi(get_conf_val("nec_basepriority"))) != NEC_UNDEF_VALUE)
&& ((attr.modcpu = atoi(get_conf_val("nec_modcpu"))) != NEC_UNDEF_VALUE)
&& ((attr.tickcnt = atoi(get_conf_val("nec_tickcnt"))) != NEC_UNDEF_VALUE)
&& ((attr.dcyfctr = atoi(get_conf_val("nec_dcyfctr"))) != NEC_UNDEF_VALUE)
&& ((attr.dcyintvl = atoi(get_conf_val("nec_dcyintvl"))) != NEC_UNDEF_VALUE)
&& ((attr.tmslice = atoi(get_conf_val("nec_timeslice"))) != NEC_UNDEF_VALUE)
&& ((attr.mempri = atoi(get_conf_val("nec_memorypriority"))) != NEC_UNDEF_VALUE)
&& ((attr.szefctmrt = atoi(get_conf_val("nec_mrt_size_effct"))) != NEC_UNDEF_VALUE)
&& ((attr.priefctmrt = atoi(get_conf_val("nec_mrt_pri_effct"))) != NEC_UNDEF_VALUE)
&& ((attr.minmrt = atoi(get_conf_val("nec_mrt_minimum"))) != NEC_UNDEF_VALUE)
&& ((attr.agrange = atoi(get_conf_val("nec_aging_range"))) != NEC_UNDEF_VALUE)
&& ((attr.spinherit = atoi(get_conf_val("nec_slavepriority"))) != NEC_UNDEF_VALUE)
&& ((attr.concpu = atoi(get_conf_val("nec_cpu_count"))) != NEC_UNDEF_VALUE)) {
if (dispcntl(SG_JID, getjid(0), DCNTL_SET2, &attr) == -1) {
shepherd_trace("ERROR: can't set scheduling parameter: %s[%d]",
strerror(errno), errno);
} else {
shepherd_trace("control parameters for active process scheduling modified");
print_scheduling_parameters(attr);
}
} else {
shepherd_trace("we do not control active process scheduling");
}
}
# else
/* write a default os-jobid to file */
sprintf(osjobid, pid_t_fmt, sid);
# endif
sge_switch2admin_user();
}
else /* not running as super user --> we want a default os-jobid */
sprintf(osjobid, "0");
if(fprintf(fp, "%s\n", osjobid) < 0)
shepherd_trace("error writing osjobid file");
FCLOSE(fp); /* Close os-jobid file */
}
# endif
return;
FCLOSE_ERROR:
shepherd_error(1, "can't close file");
}
