/**
* @brief
* pbsTcl_Init - Function to initialize Tcl interpreter based on the environment.
*
* @param[in,out] interp - Interpreter for application.
*
* @return int
* @retval TCL_OK : everything looks good.
* @retval TCL_ERROR : something got wrong!
*/
int
pbsTcl_Init(Tcl_Interp *interp)
{
if (Tcl_Init(interp) == TCL_ERROR)
return TCL_ERROR;
#if TCLX
if (Tclx_Init(interp) == TCL_ERROR)
return TCL_ERROR;
#endif
fullresp(0);
add_cmds(interp);
Tcl_SetVar(interp, "tcl_rcFileName", "~/.tclshrc", TCL_GLOBAL_ONLY);
return TCL_OK;
}
int pbsTcl_Init(
Tcl_Interp *interp) /* Interpreter for application. */
{
if (Tcl_Init(interp) == TCL_ERROR)
return TCL_ERROR;
#if TCLX
#if TCL_MINOR_VERSION < 5 && TCL_MAJOR_VERSION < 8
if (TclX_Init(interp) == TCL_ERROR)
{
#else
if (Tclx_Init(interp) == TCL_ERROR)
{
#endif
return TCL_ERROR;
}
#endif /* TCLX */
#ifndef __cplusplus
fullresp(0);
#endif
add_cmds(interp);
Tcl_SetVar(interp, "tcl_rcFileName", "~/.tclshrc", TCL_GLOBAL_ONLY);
return TCL_OK;
}
int main(int argc, char *argv[])
{
chk_file_sec_stderr = 1;
Tcl_Main(argc, argv, pbsTcl_Init);
return 0;
} /* END main() */
/*
* Find an entry for the resources for the requested host in the list of
* existing resources, or create a new one for that host and return it.
*/
Resources *
schd_get_resources(char *exechost)
{
char *id = "schd_get_resources";
Resources *rptr, *new_rsrcs;
int rm;
char *response = NULL;
int badreply = 0;
int cpus_avail = 0;
size_t pmem_avail = 0;
char hpm_ctl[64];
struct sigaction act, oact;
unsigned int remain; /* Time remaining in any old alarm(). */
time_t then; /* When this alarm() was started. */
#ifdef NODEMASK
Bitfield cpy;
int i, j;
#endif /* NODEMASK */
/*
* Check for a local copy of the resources being available already.
* If so, just return a reference to that Resources structure.
*/
if (schd_RsrcsList != NULL)
{
for (rptr = schd_RsrcsList; rptr != NULL; rptr = rptr->next)
if (strcmp(rptr->exechost, exechost) == 0)
return (rptr);
}
schd_timestamp("get_rsrcs");
/*
* No cached resource information for 'exechost'. Need to query the
* host for its information.
*/
if ((new_rsrcs = (Resources *)malloc(sizeof(Resources))) == NULL)
{
(void)sprintf(log_buffer, "Unable to alloc space for Resources.");
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
DBPRT(("%s: %s\n", id, log_buffer));
return (NULL); /* Can't get the information - nowhere to store it. */
}
memset((void *)new_rsrcs, 0, sizeof(Resources));
act.sa_flags = 0;
act.sa_handler = connect_interrupt;
sigemptyset(&act.sa_mask);
remain = 0;
then = 0;
/*
* Set the alarm, and maintain some idea of how long was left on any
* previously set alarm.
*/
if (sigaction(SIGALRM, &act, &oact) == 0)
{
remain = alarm(GETRSRCS_CONNECT_TIME);
then = time(NULL);
}
if ((rm = openrm(exechost, 0)) == -1)
{
(void)sprintf(log_buffer,
"Unable to contact resmom@%s (%d)", exechost, pbs_errno);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
/*
* Turn off full response. Responses will be received in the order in
* which they are sent.
*/
fullresp(0);
/* Build a list of all the resources about which we want information. */
addreq(rm, "loadave");
addreq(rm, "availmem");
addreq(rm, "physmem");
addreq(rm, "ncpus");
#ifdef NODEMASK
addreq(rm, "availmask");
#endif /* NODEMASK */
if (schd_MANAGE_HPM)
{
(void)sprintf(hpm_ctl, HPM_CTL_FORMAT_STR, HPM_CTL_QUERY_STR);
addreq(rm, hpm_ctl);
}
/* Get the values back from the resource monitor, and round up. */
/* Receive LOADAVE response from resource monitor. */
response = getreq(rm);
if (response != NULL)
{
new_rsrcs->loadave = atof(response) * schd_FAKE_MACH_MULT;
(void)free(response);
}
else
{
(void)sprintf(log_buffer, "bad return from getreq(loadave), %d, %d",
pbs_errno, errno);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
/* Receive AVAILMEM response from resource monitor. */
response = getreq(rm);
if (response != NULL)
{
new_rsrcs->freemem = schd_val2byte(response);
new_rsrcs->freemem *= schd_FAKE_MACH_MULT;
(void)free(response);
}
else
{
(void)sprintf(log_buffer, "bad return from getreq(freemem), %d, %d",
pbs_errno, errno);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
/* Receive PHYSMEM response from resource monitor. */
response = getreq(rm);
if (response != NULL)
{
pmem_avail = schd_val2byte(response);
pmem_avail *= schd_FAKE_MACH_MULT;
(void)free(response);
}
else
{
(void)sprintf(log_buffer, "bad return from getreq(realmem), %d, %d",
pbs_errno, errno);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
/* Receive NCPUS response from resource monitor. */
response = getreq(rm);
if (response != NULL)
{
cpus_avail = atoi(response) * schd_FAKE_MACH_MULT;
(void)free(response);
}
else
{
(void)sprintf(log_buffer, "bad return from getreq(ncpus), %d, %d",
pbs_errno, errno);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
#ifdef NODEMASK
/* Receive available nodes from resource monitor. */
response = getreq(rm);
if (response == NULL)
{
(void)sprintf(log_buffer, "bad return from getreq(availmask), %d, %d",
pbs_errno, errno);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
else
{
if (schd_bits2mask(response, &new_rsrcs->availmask) != 0)
{
if (schd_str2mask(response, &new_rsrcs->availmask) != 0)
{
(void)sprintf(log_buffer, "can't parse availmask '%s'", response);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
}
(void)free(response);
}
#endif /* NODEMASK */
if (schd_MANAGE_HPM)
{
/* Receive HPM_CTL response from resource monitor. */
response = getreq(rm);
if (response != NULL)
{
if (strcmp(response, HPM_CTL_USERMODE_STR) == 0)
new_rsrcs->flags |= RSRCS_FLAGS_HPM_USER;
else if (strcmp(response, HPM_CTL_GLOBALMODE_STR) == 0)
new_rsrcs->flags &= ~RSRCS_FLAGS_HPM_USER;
else
{
(void)sprintf(log_buffer, "bad response '%s' for '%s@%s'",
response, hpm_ctl, exechost);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id,
log_buffer);
badreply = 1;
goto bail;
}
}
else
{
(void)sprintf(log_buffer, "bad return from getreq(%s), %d, %d",
hpm_ctl, pbs_errno, errno);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
}
/*
* NOTE: response will be free()'d in bail. Be sure to explicitly free()
* response if more getreq() calls are added before the code below.
*/
bail:
if (response != NULL)
(void)free(response);
/* Disconnect from the resource monitor. */
if (rm >= 0) /* resmom handle "0" is valid in RPP. */
closerm(rm);
/* And unset the alarm and handler. */
alarm(0);
sigaction(SIGALRM, &oact, &act);
/* Reset the old alarm, taking into account how much time has passed. */
if (remain)
{
DBPRT(("%s: old alarm had %d secs remaining, %d elapsed, ", id,
remain, (time(NULL) - then)));
/* How much time remains even after the time spent above? */
remain -= (time(NULL) - then);
/*
* Would the previous time have already expired? If so, schedule
* an alarm call in 1 second (close enough, hopefully).
*/
if (remain < 1)
remain = 1;
DBPRT(("reset to %d secs\n", remain));
alarm(remain);
}
/*
* Verify all the data came back as expected; if not, abort this
* iteration of the scheduler.
*/
if (badreply)
{
(void)sprintf(log_buffer,
"Got bad info from mom@%s - aborting sched run", exechost);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
DBPRT(("%s: %s\n", id, log_buffer));
free(new_rsrcs);
return (NULL);
}
/* Make a copy of the hostname for the resources struct. */
new_rsrcs->exechost = schd_strdup(exechost);
if (new_rsrcs->exechost == NULL)
{
(void)sprintf(log_buffer, "Unable to copy exechost %s to rsrcs",
exechost);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
DBPRT(("%s: %s\n", id, log_buffer));
free(new_rsrcs);
return (NULL);
}
new_rsrcs->nodes_total = NODES_REQD(cpus_avail, pmem_avail);
#ifdef NODEMASK
/* Copy the availmask schd_FAKE_MACH_MULT times to match avail cpus. */
BITFIELD_CPY(&cpy, &(new_rsrcs->availmask));
for (i = 2; i <= schd_FAKE_MACH_MULT; i++)
{
for (j = 0; j < (cpus_avail / schd_FAKE_MACH_MULT / 2); j++)
BITFIELD_SHIFTL(&cpy);
BITFIELD_SETM(&(new_rsrcs->availmask), &cpy);
}
#endif /* NODEMASK */
if (schd_RsrcsList == NULL)
{
schd_RsrcsList = new_rsrcs; /* Start the list. */
}
else
{
for (rptr = schd_RsrcsList; rptr->next != NULL; rptr = rptr->next)
/* Find the last element in the list. */ ;
rptr->next = new_rsrcs;
}
/* Next pointer for the tail of the list points to nothing. */
new_rsrcs->next = NULL;
return (new_rsrcs);
}
/*
* Find an entry for the resources for the requested host in the list of
* existing resources, or create a new one for that host and return it.
*/
Resources *
schd_get_resources(char *exechost)
{
char *id = "schd_get_resources";
Resources *rptr, *new_rsrcs;
int rm;
char *response;
int badreply = 0;
int cpus_avail = 0;
int cpus_tot = 0;
struct sigaction act, oact;
unsigned int remain; /* Time remaining in any old alarm(). */
time_t then; /* When this alarm() was started. */
/*
* Check for a local copy of the resources being available already.
* If so, just return a reference to that Resources structure.
*/
if (schd_RsrcsList != NULL)
{
for (rptr = schd_RsrcsList; rptr != NULL; rptr = rptr->next)
if (strcmp(rptr->exechost, exechost) == 0)
return (rptr);
}
schd_timestamp("get_rsrcs");
/*
* No cached resource information for 'exechost'. Need to query the
* host for its information.
*/
if ((new_rsrcs = (Resources *)malloc(sizeof(Resources))) == NULL)
{
(void)sprintf(log_buffer, "Unable to alloc space for Resources.");
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
DBPRT(("%s: %s\n", id, log_buffer));
return (NULL); /* Can't get the information - nowhere to store it. */
}
memset((void *)new_rsrcs, 0, sizeof(Resources));
act.sa_flags = 0;
act.sa_handler = connect_interrupt;
sigemptyset(&act.sa_mask);
remain = 0;
then = 0;
/*
* Set the alarm, and maintain some idea of how long was left on any
* previously set alarm.
*/
if (sigaction(SIGALRM, &act, &oact) == 0)
{
remain = alarm(GETRSRCS_CONNECT_TIME);
then = time(NULL);
}
if ((rm = openrm(exechost, 0)) == -1)
{
(void)sprintf(log_buffer,
"Unable to contact resmom@%s (%d)", exechost, pbs_errno);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
/*
* Turn off full response. Responses will be received in the order in
* which they are sent.
*/
fullresp(0);
/* Build a list of all the resources about which we want information. */
addreq(rm, "mppe_app");
addreq(rm, "mppe_avail");
/* Get the values back from the resource monitor, and round up. */
/* Receive MPPE_APP response from resource monitor. */
/* returns the total number of Application PEs configured */
response = getreq(rm);
if (response != NULL)
{
cpus_tot = atoi(response) * schd_FAKE_MACH_MULT;
}
else
{
(void)sprintf(log_buffer, "bad return from getreq(ncpus), %d, %d",
pbs_errno, errno);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
/* Receive MPPE_AVAIL response from resource monitor. */
/* returns the largest contiguous block of APP PEs */
response = getreq(rm);
if (response != NULL)
{
cpus_avail = atoi(response) * schd_FAKE_MACH_MULT;
}
else
{
(void)sprintf(log_buffer, "bad return from getreq(ncpus), %d, %d",
pbs_errno, errno);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
new_rsrcs->freemem = MB_PER_NODE * schd_FAKE_MACH_MULT;
bail:
/* Disconnect from the resource monitor. */
if (rm)
closerm(rm);
/* And unset the alarm and handler. */
alarm(0);
sigaction(SIGALRM, &oact, &act);
/* Reset the old alarm, taking into account how much time has passed. */
if (remain)
{
DBPRT(("%s: old alarm had %d secs remaining, %d elapsed, ", id,
remain, (time(NULL) - then)));
/* How much time remains even after the time spent above? */
remain -= (time(NULL) - then);
/*
* Would the previous time have already expired? If so, schedule
* an alarm call in 1 second (close enough, hopefully).
*/
if (remain < 1)
remain = 1;
DBPRT(("reset to %d secs\n", remain));
alarm(remain);
}
/*
* Verify all the data came back as expected; if not, abort this
* iteration of the scheduler.
*/
if (badreply)
{
(void)sprintf(log_buffer,
"Got bad info from mom@%s - aborting sched run", exechost);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
DBPRT(("%s: %s\n", id, log_buffer));
free(new_rsrcs);
return (NULL);
}
/* Make a copy of the hostname for the resources struct. */
new_rsrcs->exechost = schd_strdup(exechost);
if (new_rsrcs->exechost == NULL)
{
(void)sprintf(log_buffer, "Unable to copy exechost %s to rsrcs",
exechost);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
DBPRT(("%s: %s\n", id, log_buffer));
free(new_rsrcs);
return (NULL);
}
new_rsrcs->nodes_total = cpus_tot;
new_rsrcs->nodes_alloc = cpus_tot - cpus_avail;
if (schd_RsrcsList == NULL)
{
schd_RsrcsList = new_rsrcs; /* Start the list. */
}
else
{
for (rptr = schd_RsrcsList; rptr->next != NULL; rptr = rptr->next)
/* Find the last element in the list. */ ;
rptr->next = new_rsrcs;
}
/* Next pointer for the tail of the list points to nothing. */
new_rsrcs->next = NULL;
return (new_rsrcs);
}
void
start_tcl(void)
{
char *id = "start_tcl";
char buf[BUFSIZ];
int fd;
int tot, len;
interp = Tcl_CreateInterp();
if (Tcl_Init(interp) == TCL_ERROR)
{
sprintf(log_buffer, "Tcl_Init error: %s",
Tcl_GetStringResult(interp));
log_err(-1, id, log_buffer);
die(0);
}
#if TCLX
#if TCL_MINOR_VERSION < 5 && TCL_MAJOR_VERSION < 8
if (TclX_Init(interp) == TCL_ERROR)
{
#else
if (Tclx_Init(interp) == TCL_ERROR)
{
#endif
sprintf(log_buffer, "Tclx_Init error: %s",
Tcl_GetStringResult(interp));
log_err(-1, id, log_buffer);
die(0);
}
#endif
add_cmds(interp);
if (initfil)
{
int code;
code = Tcl_EvalFile(interp, initfil);
if (code != TCL_OK)
{
char *trace;
trace = (char *)Tcl_GetVar(interp, "errorInfo", 0);
if (trace == NULL)
trace = (char *)Tcl_GetStringResult(interp);
fprintf(stderr, "%s: TCL error @ line %d: %s\n",
initfil, interp->errorLine, trace);
sprintf(log_buffer, "%s: TCL error @ line %d: %s",
initfil, interp->errorLine,
Tcl_GetStringResult(interp));
log_err(-1, id, log_buffer);
die(0);
}
sprintf(log_buffer, "init file %s", initfil);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER,
id, log_buffer);
}
if ((fd = open(bodyfil, O_RDONLY)) == -1)
{
log_err(errno, id, bodyfil);
die(0);
}
sprintf(log_buffer, "body file: %s", bodyfil);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
if (body)
free(body);
if ((body = malloc(BUFSIZ)) == NULL)
{
log_err(errno, id, "malloc");
die(0);
}
for (tot = 0; (len = read(fd, buf, sizeof(buf))) > 0; tot += len)
{
if ((body = realloc(body, tot + len + 1)) == NULL)
{
log_err(errno, id, "realloc");
die(0);
}
memcpy(&body[tot], buf, len);
}
if (len == -1)
{
log_err(errno, id, bodyfil);
die(0);
}
body[tot] = '\0';
close(fd);
#if TCL_MAJOR_VERSION >= 8
if (body_obj == NULL)
{
body_obj = Tcl_NewStringObj(body, tot);
Tcl_IncrRefCount(body_obj);
}
else
{
Tcl_SetStringObj(body_obj, body, tot);
}
#endif
}
int
addclient(name)
char *name;
{
static char id[] = "addclient";
struct hostent *host, *gethostbyname();
struct in_addr saddr;
if ((host = gethostbyname(name)) == NULL)
{
sprintf(log_buffer, "host %s not found", name);
log_err(-1, id, log_buffer);
return -1;
}
if (numclients >= START_CLIENTS)
{
pbs_net_t *newclients;
newclients = realloc(okclients,
sizeof(pbs_net_t) * (numclients + 1));
if (newclients == NULL)
return -1;
okclients = newclients;
}
memcpy((char *)&saddr, host->h_addr, host->h_length);
okclients[numclients++] = saddr.s_addr;
return 0;
}
/*
* read_config - read and process the configuration file (see -c option)
*
* Currently, the only statement is $clienthost to specify which systems
* can contact the scheduler.
*/
#define CONF_LINE_LEN 120
static
int
read_config(file)
char *file;
{
static char *id = "read_config";
FILE *conf;
int i;
char line[CONF_LINE_LEN];
char *token;
struct specialconfig
{
char *name;
int (*handler)();
} special[] =
{
{"clienthost", addclient },
{ NULL, NULL }
};
#if !defined(DEBUG) && !defined(NO_SECURITY_CHECK)
if (chk_file_sec(file, 0, 0, S_IWGRP | S_IWOTH, 1, 0))
return (-1);
#endif
if ((conf = fopen(file, "r")) == NULL)
{
log_err(errno, id, "cannot open config file");
return (-1);
}
while (fgets(line, CONF_LINE_LEN, conf))
{
if ((line[0] == '#') || (line[0] == '\n'))
continue; /* ignore comment & null line */
else if (line[0] == '$') /* special */
{
if ((token = strtok(line, " \t")) == NULL)
token = "";
for (i = 0; special[i].name; i++)
{
if (strcmp(token + 1, special[i].name) == 0)
break;
}
if (special[i].name == NULL)
{
sprintf(log_buffer, "config name %s not known",
token);
log_record(PBSEVENT_ERROR,
PBS_EVENTCLASS_SERVER,
msg_daemonname, log_buffer);
return (-1);
}
token = strtok(NULL, " \t");
if (*(token + strlen(token) - 1) == '\n')
*(token + strlen(token) - 1) = '\0';
if (special[i].handler(token))
{
fclose(conf);
return (-1);
}
}
else
{
log_record(PBSEVENT_ERROR, PBS_EVENTCLASS_SERVER,
msg_daemonname,
"invalid line in config file");
fclose(conf);
return (-1);
}
}
fclose(conf);
return (0);
}
void
restart(sig)
int sig;
{
char *id = "restart";
if (sig)
{
sprintf(log_buffer, "restart on signal %d", sig);
log_close(1);
log_open(logfile, path_log);
}
else
{
sprintf(log_buffer, "restart command");
}
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
Tcl_DeleteInterp(interp);
if (configfile)
{
if (read_config(configfile) != 0)
die(0);
}
start_tcl();
}
void
badconn(msg)
char *msg;
{
static char id[] = "badconn";
struct in_addr addr;
char buf[5*sizeof(addr) + 100];
struct hostent *phe;
addr = saddr.sin_addr;
phe = gethostbyaddr((void *) & addr, sizeof(addr), AF_INET);
if (phe == NULL)
{
char hold[6];
int i;
union
{
struct in_addr aa;
u_char bb[sizeof(addr)];
} uu;
uu.aa = addr;
sprintf(buf, "%u", uu.bb[0]);
for (i = 1; i < (int)sizeof(addr); i++)
{
sprintf(hold, ".%u", uu.bb[i]);
strcat(buf, hold);
}
}
else
{
strncpy(buf, phe->h_name, sizeof(buf));
buf[sizeof(buf)-1] = '\0';
}
sprintf(log_buffer, "%s on port %u %s", buf, ntohs(saddr.sin_port), msg);
log_err(-1, id, log_buffer);
return;
}
unsigned int
server_command()
{
static char id[] = "server_command";
int new_socket;
int i;
torque_socklen_t slen;
unsigned int cmd;
pbs_net_t addr;
slen = sizeof(saddr);
new_socket = accept(server_sock,
(struct sockaddr *) & saddr, &slen);
if (new_socket == -1)
{
log_err(errno, id, "accept");
return SCH_ERROR;
}
if (ntohs(saddr.sin_port) >= IPPORT_RESERVED)
{
badconn("non-reserved port");
close(new_socket);
return SCH_ERROR;
}
addr = (pbs_net_t)saddr.sin_addr.s_addr;
for (i = 0; i < numclients; i++)
{
if (addr == okclients[i])
break;
}
if (i == numclients)
{
badconn("unauthorized host");
close(new_socket);
return SCH_ERROR;
}
if ((connector = socket_to_conn(new_socket)) < 0)
{
log_err(errno, id, "socket_to_conn");
return SCH_ERROR;
}
if (get_4byte(new_socket, &cmd) != 1)
{
log_err(errno, id, "get4bytes");
return SCH_ERROR;
}
return cmd;
}
/*
* lock_out - lock out other daemons from this directory.
*/
static void lock_out(fds, op)
int fds;
int op; /* F_WRLCK or F_UNLCK */
{
struct flock flock;
flock.l_type = op;
flock.l_whence = SEEK_SET;
flock.l_start = 0;
flock.l_len = 0; /* whole file */
if (fcntl(fds, F_SETLK, &flock) < 0)
{
(void)strcpy(log_buffer, "pbs_sched: another scheduler running\n");
log_err(errno, msg_daemonname, log_buffer);
fprintf(stderr, log_buffer);
exit(1);
}
}
int main(argc, argv)
int argc;
char *argv[];
{
char *id = "main";
int code;
struct hostent *hp;
int go, c, errflg = 0;
int lockfds;
int t = 1;
char *ptr;
pid_t pid;
char *cp, host[100];
char *homedir = PBS_SERVER_HOME;
unsigned int port;
char path_priv[_POSIX_PATH_MAX];
char *dbfile = "sched_out";
int alarm_time = 180;
struct sigaction act;
caddr_t curr_brk = 0, next_brk;
extern char *optarg;
extern int optind, opterr;
fd_set fdset;
#ifndef DEBUG
if (IamRoot() == 0)
{
return (1);
}
#endif /* DEBUG */
glob_argv = argv;
if ((cp = strrchr(argv[0], '/')) == NULL)
cp = argv[0];
else
cp++;
msg_daemonname = strdup(cp);
port = get_svrport(PBS_SCHEDULER_SERVICE_NAME, "tcp",
PBS_SCHEDULER_SERVICE_PORT);
while ((c = getopt(argc, argv, "L:S:d:i:b:t:p:a:vc:")) != EOF)
{
switch (c)
{
case 'L':
logfile = optarg;
break;
case 'S':
port = (unsigned int)atoi(optarg);
if (port == 0)
{
fprintf(stderr,
"%s: illegal port\n", optarg);
errflg = 1;
}
break;
case 'd':
homedir = optarg;
break;
case 'i': /* initialize */
initfil = optarg;
break;
case 'b':
bodyfil = optarg;
break;
case 't':
termfil = optarg;
break;
case 'p':
dbfile = optarg;
break;
case 'a':
alarm_time = strtol(optarg, &ptr, 10);
if (alarm_time <= 0 || *ptr != '\0')
{
fprintf(stderr,
"%s: bad alarm time\n", optarg);
errflg = 1;
}
break;
case 'c':
configfile = optarg;
break;
case 'v':
verbose = 1;
break;
case '?':
errflg = 1;
break;
}
}
if (errflg || optind != argc)
{
static char *options[] =
{
"[-L logfile]",
"[-S port]",
"[-d home]",
"[-i init]",
"[-b body]",
"[-t term]",
"[-p output]",
"[-a alarm]",
"[-c configfile]",
"[-v]",
NULL
};
int i;
fprintf(stderr, "usage: %s\n", argv[0]);
for (i = 0; options[i]; i++)
fprintf(stderr, "\t%s\n", options[i]);
exit(1);
}
/* Save the original working directory for "restart" */
if ((oldpath = getcwd((char *)NULL, MAXPATHLEN)) == NULL)
{
fprintf(stderr, "cannot get current working directory\n");
exit(1);
}
(void)sprintf(path_priv, "%s/sched_priv", homedir);
#if !defined(DEBUG) && !defined(NO_SECURITY_CHECK)
c = chk_file_sec(path_priv, 1, 0, S_IWGRP | S_IWOTH, 1, 0);
c |= chk_file_sec(PBS_ENVIRON, 0, 0, S_IWGRP | S_IWOTH, 0, 0);
if (c != 0) exit(1);
#endif /* not DEBUG and not NO_SECURITY_CHECK */
if (chdir(path_priv) == -1)
{
perror(path_priv);
exit(1);
}
(void)sprintf(path_log, "%s/sched_logs", homedir);
(void)strcpy(pbs_current_user, "Scheduler");
/* The following is code to reduce security risks */
/* start out with standard umask, system resource limit infinite */
umask(022);
if (setup_env(PBS_ENVIRON) == -1)
exit(1);
c = getgid();
(void)setgroups(1, (gid_t *)&c); /* secure suppl. group ids */
c = sysconf(_SC_OPEN_MAX);
while (--c > 2)
(void)close(c); /* close any file desc left open by parent */
#ifndef DEBUG
#ifdef _CRAY
(void)limit(C_JOB, 0, L_CPROC, 0);
(void)limit(C_JOB, 0, L_CPU, 0);
(void)limit(C_JOBPROCS, 0, L_CPU, 0);
(void)limit(C_PROC, 0, L_FD, 255);
(void)limit(C_JOB, 0, L_FSBLK, 0);
(void)limit(C_JOBPROCS, 0, L_FSBLK, 0);
(void)limit(C_JOB, 0, L_MEM , 0);
(void)limit(C_JOBPROCS, 0, L_MEM , 0);
#else /* not _CRAY */
{
struct rlimit rlimit;
rlimit.rlim_cur = RLIM_INFINITY;
rlimit.rlim_max = RLIM_INFINITY;
(void)setrlimit(RLIMIT_CPU, &rlimit);
(void)setrlimit(RLIMIT_FSIZE, &rlimit);
(void)setrlimit(RLIMIT_DATA, &rlimit);
(void)setrlimit(RLIMIT_STACK, &rlimit);
#ifdef RLIMIT_RSS
(void)setrlimit(RLIMIT_RSS , &rlimit);
#endif /* RLIMIT_RSS */
#ifdef RLIMIT_VMEM
(void)setrlimit(RLIMIT_VMEM , &rlimit);
#endif /* RLIMIT_VMEM */
}
#endif /* not _CRAY */
#if !defined(NO_SECURITY_CHECK)
c = 0;
if (initfil)
{
if (*initfil != '/')
{
(void)sprintf(log_buffer, "%s/%s", path_priv, initfil);
c |= chk_file_sec(log_buffer, 0, 0, S_IWGRP | S_IWOTH, 1, 0);
}
else
{
c |= chk_file_sec(initfil, 0, 0, S_IWGRP | S_IWOTH, 1, 0);
}
}
if (bodyfil)
{
if (*bodyfil != '/')
{
(void)sprintf(log_buffer, "%s/%s", path_priv, bodyfil);
c |= chk_file_sec(log_buffer, 0, 0, S_IWGRP | S_IWOTH, 1, 0);
}
else
{
c |= chk_file_sec(bodyfil, 0, 0, S_IWGRP | S_IWOTH, 1, 0);
}
}
if (termfil)
{
if (*termfil != '/')
{
(void)sprintf(log_buffer, "%s/%s", path_priv, termfil);
c |= chk_file_sec(log_buffer, 0, 0, S_IWGRP | S_IWOTH, 1, 0);
}
else
{
c |= chk_file_sec(termfil, 0, 0, S_IWGRP | S_IWOTH, 1, 0);
}
}
if (c) exit(1);
#endif /* not NO_SECURITY_CHECK */
#endif /* not DEBUG */
if (log_open(logfile, path_log) == -1)
{
fprintf(stderr, "%s: logfile could not be opened\n", argv[0]);
exit(1);
}
if (gethostname(host, sizeof(host)) == -1)
{
char *prob = "gethostname";
log_err(errno, id, prob);
perror(prob);
die(0);
}
if ((hp = gethostbyname(host)) == NULL)
{
char *prob = "gethostbyname";
log_err(errno, id, prob);
perror(prob);
die(0);
}
if ((server_sock = socket(AF_INET, SOCK_STREAM, 0)) < 0)
{
char *prob = "socket";
log_err(errno, id, prob);
perror(prob);
die(0);
}
if (setsockopt(server_sock, SOL_SOCKET, SO_REUSEADDR,
(char *)&t, sizeof(t)) == -1)
{
char *prob = "setsockopt";
log_err(errno, id, prob);
perror(prob);
die(0);
}
saddr.sin_family = AF_INET;
saddr.sin_port = htons((unsigned short)port);
memcpy(&saddr.sin_addr, hp->h_addr, hp->h_length);
if (bind(server_sock, (struct sockaddr *)&saddr, sizeof(saddr)) < 0)
{
char *prob = "bind";
log_err(errno, id, prob);
perror(prob);
die(0);
}
if (listen(server_sock, 5) < 0)
{
char *prob = "listen";
log_err(errno, id, prob);
perror(prob);
die(0);
}
okclients = (pbs_net_t *)calloc(START_CLIENTS, sizeof(pbs_net_t));
addclient("localhost"); /* who has permission to call MOM */
addclient(host);
if (configfile)
{
if (read_config(configfile) != 0)
die(0);
}
lockfds = open("sched.lock", O_CREAT | O_TRUNC | O_WRONLY, 0644);
if (lockfds < 0)
{
char *prob = "lock file";
log_err(errno, id, prob);
perror(prob);
die(0);
}
lock_out(lockfds, F_WRLCK);
#ifndef DEBUG
lock_out(lockfds, F_UNLCK);
if ((pid = fork()) == -1) /* error on fork */
{
char *prob = "fork";
log_err(errno, id, prob);
perror(prob);
die(0);
}
else if (pid > 0) /* parent exits */
exit(0);
if ((pid = setsid()) == -1)
{
log_err(errno, id, "setsid");
die(0);
}
lock_out(lockfds, F_WRLCK);
freopen(dbfile, "a", stdout);
setvbuf(stdout, NULL, _IOLBF, 0);
dup2(fileno(stdout), fileno(stderr));
#else
pid = getpid();
setvbuf(stdout, NULL, _IOLBF, 0);
setvbuf(stderr, NULL, _IOLBF, 0);
#endif
freopen("/dev/null", "r", stdin);
/* write schedulers pid into lockfile */
(void)sprintf(log_buffer, "%d\n", pid);
(void)write(lockfds, log_buffer, strlen(log_buffer) + 1);
#if (PLOCK_DAEMONS & 2)
(void)plock(PROCLOCK); /* lock daemon into memory */
#endif
sprintf(log_buffer, "%s startup pid %d", argv[0], pid);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
sprintf(log_buffer, "%s using TCL %s (%s)", argv[0],
TCL_VERSION, TCL_PATCH_LEVEL);
fprintf(stderr, "%s\n", log_buffer);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
fullresp(0);
sigemptyset(&allsigs);
act.sa_flags = 0;
sigaddset(&allsigs, SIGHUP); /* remember to block these */
sigaddset(&allsigs, SIGINT); /* during critical sections */
sigaddset(&allsigs, SIGTERM); /* so we don't get confused */
act.sa_mask = allsigs;
act.sa_handler = restart; /* do a restart on SIGHUP */
sigaction(SIGHUP, &act, NULL);
act.sa_handler = toolong; /* handle an alarm call */
sigaction(SIGALRM, &act, NULL);
act.sa_handler = die; /* bite the biscuit for all following */
sigaction(SIGINT, &act, NULL);
sigaction(SIGTERM, &act, NULL);
start_tcl();
FD_ZERO(&fdset);
for (go = 1; go;)
{
unsigned int cmd;
FD_SET(server_sock, &fdset);
if (select(FD_SETSIZE, &fdset, NULL, NULL, NULL) == -1)
{
if (errno != EINTR)
log_err(errno, id, "select");
continue;
}
if (!FD_ISSET(server_sock, &fdset))
continue;
cmd = server_command();
if (cmd == (unsigned)SCH_ERROR || cmd == (unsigned)SCH_SCHEDULE_NULL)
continue;
if (sigprocmask(SIG_BLOCK, &allsigs, &oldsigs) == -1)
log_err(errno, id, "sigprocmaskSIG_BLOCK)");
if (verbose)
{
sprintf(log_buffer, "command %d", cmd);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER,
id, log_buffer);
}
switch (cmd)
{
case SCH_SCHEDULE_NEW:
case SCH_SCHEDULE_TERM:
case SCH_SCHEDULE_TIME:
case SCH_SCHEDULE_RECYC:
case SCH_SCHEDULE_CMD:
case SCH_SCHEDULE_FIRST:
alarm(alarm_time);
#if TCL_MAJOR_VERSION >= 8
/* execute compiled body code for TCL-8 */
code = Tcl_EvalObj(interp, body_obj);
#else
code = Tcl_Eval(interp, body);
#endif
alarm(0);
switch (code)
{
case TCL_OK:
case TCL_RETURN:
break;
default:
{
char *trace;
char codename[20];
switch (code)
{
case TCL_BREAK:
strcpy(codename, "break");
break;
case TCL_CONTINUE:
strcpy(codename, "continue");
break;
default:
strcpy(codename, "<unknown>");
break;
}
trace = (char *)Tcl_GetVar(interp, "errorInfo", 0);
if (trace == NULL)
trace = (char *)Tcl_GetStringResult(interp);
fprintf(stderr, "%s: TCL interpreter return code %d (%s) @ line %d: %s\n",
bodyfil, code, codename,
interp->errorLine, trace);
sprintf(log_buffer,
"%s: TCL error @ line %d: %s",
bodyfil, interp->errorLine,
Tcl_GetStringResult(interp));
log_err(-1, id, log_buffer);
die(0);
}
}
break;
case SCH_CONFIGURE:
case SCH_RULESET:
restart(0);
break;
case SCH_QUIT:
go = 0;
break;
default:
log_err(-1, id, "unknown command");
break;
}
if (connector >= 0 && server_disconnect(connector))
{
log_err(errno, id, "server_disconnect");
die(0);
}
connector = -1;
if (verbose)
{
next_brk = (caddr_t)sbrk(0);
if (next_brk > curr_brk)
{
sprintf(log_buffer, "brk point %p", next_brk);
log_record(PBSEVENT_SYSTEM,
PBS_EVENTCLASS_SERVER,
id, log_buffer);
curr_brk = next_brk;
}
}
if (sigprocmask(SIG_SETMASK, &oldsigs, NULL) == -1)
log_err(errno, id, "sigprocmask(SIG_SETMASK)");
}
if (termfil)
{
code = Tcl_EvalFile(interp, termfil);
if (code != TCL_OK)
{
char *trace;
trace = (char *)Tcl_GetVar(interp, "errorInfo", 0);
if (trace == NULL)
trace = (char *)Tcl_GetStringResult(interp);
fprintf(stderr, "%s: TCL error @ line %d: %s\n",
termfil, interp->errorLine, trace);
sprintf(log_buffer, "%s: TCL error @ line %d: %s",
termfil, interp->errorLine,
Tcl_GetStringResult(interp));
log_err(-1, id, log_buffer);
die(0);
}
sprintf(log_buffer, "term file: %s", termfil);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER,
id, log_buffer);
}
sprintf(log_buffer, "%s normal finish pid %d", argv[0], pid);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
(void)close(server_sock);
exit(0);
}
/*
* Find an entry for the resources for the requested host in the list of
* existing resources, or create a new one for that host and return it.
*/
Resources *
schd_get_resources(char *exechost)
{
char *id = "schd_get_resources";
Resources *rptr, *new_rsrcs;
int rm;
char *response = NULL;
int badreply = 0;
int local_errno = 0;
struct sigaction act, oact;
unsigned int remain; /* Time remaining in any old alarm(). */
time_t then; /* When this alarm() was started. */
/*
* Check for a local copy of the resources being available already.
* If so, just return a reference to that Resources structure.
*/
if (schd_RsrcsList != NULL)
{
for (rptr = schd_RsrcsList; rptr != NULL; rptr = rptr->next)
if (strcmp(rptr->exechost, exechost) == 0)
return (rptr);
}
schd_timestamp("get_rsrcs");
/*
* No cached resource information for 'exechost'. Need to query the
* host for its information.
*/
if ((new_rsrcs = (Resources *)malloc(sizeof(Resources))) == NULL)
{
(void)sprintf(log_buffer, "Unable to alloc space for Resources.");
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
DBPRT(("%s: %s\n", id, log_buffer));
return (NULL); /* Can't get the information - nowhere to store it. */
}
memset((void *)new_rsrcs, 0, sizeof(Resources));
act.sa_flags = 0;
act.sa_handler = connect_interrupt;
sigemptyset(&act.sa_mask);
remain = 0;
then = 0;
/*
* Set the alarm, and maintain some idea of how long was left on any
* previously set alarm.
*/
if (sigaction(SIGALRM, &act, &oact) == 0)
{
remain = alarm(GETRSRCS_CONNECT_TIME);
then = time(NULL);
}
if ((rm = openrm(exechost, 0)) == -1)
{
(void)sprintf(log_buffer,
"Unable to contact resmom@%s ", exechost);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
/*
* Turn off full response. Responses will be received in the order in
* which they are sent.
*/
fullresp(0);
/* Build a list of all the resources about which we want information. */
addreq(rm, "loadave");
addreq(rm, "availmem");
addreq(rm, "physmem");
addreq(rm, "ncpus");
addreq(rm, "tmpdir");
addreq(rm, "arch");
/* Get the values back from the resource monitor, and round up. */
/* Receive LOADAVE response from resource monitor. */
response = getreq_err(&local_errno, rm);
if (response != NULL)
{
new_rsrcs->loadave = atof(response);
(void)free(response);
}
else
{
(void)sprintf(log_buffer, "bad return from getreq(loadave), %d, %d",
local_errno, errno);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
/* Receive AVAILMEM response from resource monitor. */
response = getreq_err(&local_errno, rm);
if (response != NULL)
{
new_rsrcs->freemem = schd_val2byte(response);
(void)free(response);
}
else
{
(void)sprintf(log_buffer, "bad return from getreq(freemem), %d, %d",
local_errno, errno);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
/* Receive PHYSMEM response from resource monitor. */
response = getreq_err(&local_errno, rm);
if (response != NULL)
{
new_rsrcs->mem_total = schd_val2byte(response);
(void)free(response);
}
else
{
(void)sprintf(log_buffer, "bad return from getreq(realmem), %d, %d",
local_errno, errno);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
/* Receive NCPUS response from resource monitor. */
response = getreq_err(&local_errno, rm);
if (response != NULL)
{
new_rsrcs->ncpus_total = atoi(response);
(void)free(response);
}
else
{
(void)sprintf(log_buffer, "bad return from getreq(ncpus), %d, %d",
local_errno, errno);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
/* Receive TMPDIR response from resource monitor. */
response = getreq_err(&local_errno, rm);
if (response != NULL)
{
new_rsrcs->tmpdir = schd_val2byte(response);
(void)free(response);
}
else
{
(void)sprintf(log_buffer, "bad return from getreq(tmpdir), %d, %d",
local_errno, errno);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
/* Receive ARCH response from resource monitor. */
response = getreq_err(&local_errno, rm);
if (response != NULL)
{
new_rsrcs->arch = schd_strdup(response);
(void)free(response);
}
else
{
(void)sprintf(log_buffer, "bad return from getreq(arch), %d, %d",
local_errno, errno);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
badreply = 1;
goto bail;
}
bail:
/* Disconnect from the resource monitor. */
if (rm >= 0) /* resmom handle "0" is valid in RPP. */
closerm(rm);
/* And unset the alarm and handler. */
alarm(0);
sigaction(SIGALRM, &oact, &act);
/* Reset the old alarm, taking into account how much time has passed. */
if (remain)
{
DBPRT(("%s: old alarm had %d secs remaining, %d elapsed, ", id,
remain, (time(NULL) - then)));
/* How much time remains even after the time spent above? */
remain -= (time(NULL) - then);
/*
* Would the previous time have already expired? If so, schedule
* an alarm call in 1 second (close enough, hopefully).
*/
if (remain < 1)
remain = 1;
DBPRT(("reset to %d secs\n", remain));
alarm(remain);
}
/*
* Verify all the data came back as expected; if not, abort this
* iteration of the scheduler.
*/
if (badreply)
{
(void)sprintf(log_buffer,
"Got bad info from mom@%s - skipping this node", exechost);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
DBPRT(("%s: %s\n", id, log_buffer));
free(new_rsrcs);
return (NULL);
}
/* Make a copy of the hostname for the resources struct. */
new_rsrcs->exechost = schd_strdup(exechost);
if (new_rsrcs->exechost == NULL)
{
(void)sprintf(log_buffer, "Unable to copy exechost %s to rsrcs",
exechost);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
DBPRT(("%s: %s\n", id, log_buffer));
free(new_rsrcs);
return (NULL);
}
if (schd_RsrcsList == NULL)
{
schd_RsrcsList = new_rsrcs; /* Start the list. */
}
else
{
for (rptr = schd_RsrcsList; rptr->next != NULL; rptr = rptr->next)
/* Find the last element in the list. */ ;
rptr->next = new_rsrcs;
}
/* Next pointer for the tail of the list points to nothing. */
new_rsrcs->next = NULL;
return (new_rsrcs);
}
int main(
int argc,
char *argv[])
{
char *id = "main";
struct hostent *hp;
int go, c, errflg = 0;
int lockfds;
int t = 1;
pid_t pid;
char host[100];
char *homedir = PBS_SERVER_HOME;
unsigned int port;
char *dbfile = "sched_out";
struct sigaction act;
sigset_t oldsigs;
caddr_t curr_brk = 0;
caddr_t next_brk;
extern char *optarg;
extern int optind, opterr;
extern int rpp_fd;
fd_set fdset;
int schedinit(int argc, char **argv);
int schedule(int com, int connector);
glob_argv = argv;
alarm_time = 180;
/* The following is code to reduce security risks */
/* move this to a place where nss_ldap doesn't hold a socket yet */
c = sysconf(_SC_OPEN_MAX);
while (--c > 2)
(void)close(c); /* close any file desc left open by parent */
port = get_svrport(PBS_SCHEDULER_SERVICE_NAME, "tcp",
PBS_SCHEDULER_SERVICE_PORT);
pbs_rm_port = get_svrport(PBS_MANAGER_SERVICE_NAME, "tcp",
PBS_MANAGER_SERVICE_PORT);
strcpy(pbs_current_user, "Scheduler");
msg_daemonname = strdup("pbs_sched");
opterr = 0;
while ((c = getopt(argc, argv, "L:S:R:d:p:c:a:-:")) != EOF)
{
switch (c)
{
case '-':
if ((optarg == NULL) || (optarg[0] == '\0'))
{
errflg = 1;
}
if (!strcmp(optarg, "version"))
{
fprintf(stderr, "version: %s\n", PACKAGE_VERSION);
exit(0);
}
else
{
errflg = 1;
}
break;
case 'L':
logfile = optarg;
break;
case 'S':
port = atoi(optarg);
if (port == 0)
{
fprintf(stderr,
"%s: illegal port\n", optarg);
errflg = 1;
}
break;
case 'R':
if ((pbs_rm_port = atoi(optarg)) == 0)
{
(void)fprintf(stderr, "%s: bad -R %s\n",
argv[0], optarg);
return 1;
}
break;
case 'd':
homedir = optarg;
break;
case 'p':
dbfile = optarg;
break;
case 'c':
configfile = optarg;
break;
case 'a':
alarm_time = atoi(optarg);
if (alarm_time == 0)
{
fprintf(stderr,
"%s: bad alarm time\n", optarg);
errflg = 1;
}
break;
case '?':
errflg = 1;
break;
}
}
if (errflg)
{
fprintf(stderr, "usage: %s %s\n", argv[0], usage);
exit(1);
}
#ifndef DEBUG
if (IamRoot() == 0)
{
return (1);
}
#endif /* DEBUG */
/* Save the original working directory for "restart" */
if ((oldpath = getcwd((char *)NULL, MAXPATHLEN)) == NULL)
{
fprintf(stderr, "cannot get current working directory\n");
exit(1);
}
(void)sprintf(log_buffer, "%s/sched_priv", homedir);
#if !defined(DEBUG) && !defined(NO_SECURITY_CHECK)
c = chk_file_sec(log_buffer, 1, 0, S_IWGRP | S_IWOTH, 1, NULL);
c |= chk_file_sec(PBS_ENVIRON, 0, 0, S_IWGRP | S_IWOTH, 0, NULL);
if (c != 0) exit(1);
#endif /* not DEBUG and not NO_SECURITY_CHECK */
if (chdir(log_buffer) == -1)
{
perror("chdir");
exit(1);
}
(void)sprintf(path_log, "%s/sched_logs", homedir);
(void)sprintf(path_acct, "%s/%s", log_buffer, PBS_ACCT);
/* The following is code to reduce security risks */
/* start out with standard umask, system resource limit infinite */
umask(022);
if (setup_env(PBS_ENVIRON) == -1)
exit(1);
c = getgid();
(void)setgroups(1, (gid_t *)&c); /* secure suppl. groups */
#ifndef DEBUG
#ifdef _CRAY
(void)limit(C_JOB, 0, L_CPROC, 0);
(void)limit(C_JOB, 0, L_CPU, 0);
(void)limit(C_JOBPROCS, 0, L_CPU, 0);
(void)limit(C_PROC, 0, L_FD, 255);
(void)limit(C_JOB, 0, L_FSBLK, 0);
(void)limit(C_JOBPROCS, 0, L_FSBLK, 0);
(void)limit(C_JOB, 0, L_MEM , 0);
(void)limit(C_JOBPROCS, 0, L_MEM , 0);
#else /* not _CRAY */
{
struct rlimit rlimit;
rlimit.rlim_cur = RLIM_INFINITY;
rlimit.rlim_max = RLIM_INFINITY;
(void)setrlimit(RLIMIT_CPU, &rlimit);
(void)setrlimit(RLIMIT_FSIZE, &rlimit);
(void)setrlimit(RLIMIT_DATA, &rlimit);
(void)setrlimit(RLIMIT_STACK, &rlimit);
#ifdef RLIMIT_RSS
(void)setrlimit(RLIMIT_RSS , &rlimit);
#endif /* RLIMIT_RSS */
#ifdef RLIMIT_VMEM
(void)setrlimit(RLIMIT_VMEM , &rlimit);
#endif /* RLIMIT_VMEM */
}
#endif /* not _CRAY */
#endif /* DEBUG */
if (log_open(logfile, path_log) == -1)
{
fprintf(stderr, "%s: logfile could not be opened\n", argv[0]);
exit(1);
}
if (gethostname(host, sizeof(host)) == -1)
{
log_err(errno, id, "gethostname");
die(0);
}
if ((hp = gethostbyname(host)) == NULL)
{
log_err(errno, id, "gethostbyname");
die(0);
}
if ((server_sock = socket(AF_INET, SOCK_STREAM, 0)) < 0)
{
log_err(errno, id, "socket");
die(0);
}
if (setsockopt(server_sock, SOL_SOCKET, SO_REUSEADDR,
(char *)&t, sizeof(t)) == -1)
{
log_err(errno, id, "setsockopt");
die(0);
}
saddr.sin_family = AF_INET;
saddr.sin_port = htons(port);
memcpy(&saddr.sin_addr, hp->h_addr, hp->h_length);
if (bind(server_sock, (struct sockaddr *)&saddr, sizeof(saddr)) < 0)
{
log_err(errno, id, "bind");
die(0);
}
if (listen(server_sock, 5) < 0)
{
log_err(errno, id, "listen");
die(0);
}
okclients = (pbs_net_t *)calloc(START_CLIENTS, sizeof(pbs_net_t));
addclient("localhost"); /* who has permission to call MOM */
addclient(host);
if (configfile)
{
if (read_config(configfile) != 0)
die(0);
}
lockfds = open("sched.lock", O_CREAT | O_TRUNC | O_WRONLY, 0644);
if (lockfds < 0)
{
log_err(errno, id, "open lock file");
exit(1);
}
lock_out(lockfds, F_WRLCK);
fullresp(0);
if (sigemptyset(&allsigs) == -1)
{
perror("sigemptyset");
exit(1);
}
if (sigprocmask(SIG_SETMASK, &allsigs, NULL) == -1) /* unblock */
{
perror("sigprocmask");
exit(1);
}
act.sa_flags = 0;
sigaddset(&allsigs, SIGHUP); /* remember to block these */
sigaddset(&allsigs, SIGINT); /* during critical sections */
sigaddset(&allsigs, SIGTERM); /* so we don't get confused */
act.sa_mask = allsigs;
act.sa_handler = restart; /* do a restart on SIGHUP */
sigaction(SIGHUP, &act, NULL);
act.sa_handler = toolong; /* handle an alarm call */
sigaction(SIGALRM, &act, NULL);
act.sa_handler = die; /* bite the biscuit for all following */
sigaction(SIGINT, &act, NULL);
sigaction(SIGTERM, &act, NULL);
/*
* Catch these signals to ensure we core dump even if
* our rlimit for core dumps is set to 0 initially.
*
* Chris Samuel - VPAC
* [email protected] - 29th July 2003
*
* Now conditional on the PBSCOREDUMP environment variable
*/
if (getenv("PBSCOREDUMP"))
{
act.sa_handler = catch_abort; /* make sure we core dump */
sigaction(SIGSEGV, &act, NULL);
sigaction(SIGBUS, &act, NULL);
sigaction(SIGFPE, &act, NULL);
sigaction(SIGILL, &act, NULL);
sigaction(SIGTRAP, &act, NULL);
sigaction(SIGSYS, &act, NULL);
}
/*
* Local initialization stuff
*/
if (schedinit(argc, argv))
{
(void) sprintf(log_buffer,
"local initialization failed, terminating");
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
exit(1);
}
if (getenv("PBSDEBUG") == NULL)
{
lock_out(lockfds, F_UNLCK);
#ifdef DISABLE_DAEMONS
pid = getpid();
#else
if ((pid = fork()) == -1)
{
/* error on fork */
perror("fork");
exit(1);
}
else
if (pid > 0) /* parent exits */
{
exit(0);
}
if ((pid = setsid()) == -1)
{
perror("setsid");
exit(1);
}
#endif /* DISABLE_DAEMONS */
lock_out(lockfds, F_WRLCK);
if (freopen(dbfile, "a", stdout) == NULL)
{
perror("opening lockfile");
exit(1);
}
setvbuf(stdout, NULL, _IOLBF, 0);
dup2(fileno(stdout), fileno(stderr));
}
else
{
setvbuf(stdout, NULL, _IOLBF, 0);
setvbuf(stderr, NULL, _IOLBF, 0);
pid = getpid();
}
if (freopen("/dev/null", "r", stdin) == NULL)
{
perror("opening /dev/null");
exit(1);
}
/* write scheduler's pid into lockfile */
(void)sprintf(log_buffer, "%ld\n", (long)pid);
if (write(lockfds, log_buffer, strlen(log_buffer) + 1) != (ssize_t)(strlen(log_buffer) + 1))
{
perror("writing to lockfile");
exit(1);
}
#if (PLOCK_DAEMONS & 2)
(void)plock(PROCLOCK); /* lock daemon into memory */
#endif
sprintf(log_buffer, "%s startup pid %ld", argv[0], (long)pid);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
FD_ZERO(&fdset);
for (go = 1;go;)
{
int cmd;
if (rpp_fd != -1)
FD_SET(rpp_fd, &fdset);
FD_SET(server_sock, &fdset);
if (select(FD_SETSIZE, &fdset, NULL, NULL, NULL) == -1)
{
if (errno != EINTR)
{
log_err(errno, id, "select");
die(0);
}
continue;
}
if (rpp_fd != -1 && FD_ISSET(rpp_fd, &fdset))
{
if (rpp_io() == -1)
log_err(errno, id, "rpp_io");
}
if (!FD_ISSET(server_sock, &fdset))
continue;
cmd = server_command();
if (sigprocmask(SIG_BLOCK, &allsigs, &oldsigs) == -1)
log_err(errno, id, "sigprocmaskSIG_BLOCK)");
alarm(alarm_time);
if (schedule(cmd, connector)) /* magic happens here */
go = 0;
alarm(0);
if (connector >= 0 && server_disconnect(connector))
{
log_err(errno, id, "server_disconnect");
die(0);
}
next_brk = (caddr_t)sbrk(0);
if (next_brk > curr_brk)
{
sprintf(log_buffer, "brk point %ld", (long)next_brk);
log_record(PBSEVENT_DEBUG, PBS_EVENTCLASS_SERVER,
id, log_buffer);
curr_brk = next_brk;
}
if (sigprocmask(SIG_SETMASK, &oldsigs, NULL) == -1)
log_err(errno, id, "sigprocmask(SIG_SETMASK)");
}
sprintf(log_buffer, "%s normal finish pid %ld",
argv[0],
(long)pid);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
close(server_sock);
exit(0);
} /* END main() */
