/*************************************************
* svr_is_request
*
* Return: svr_is_request always returns a non-zero value
* and it must call close_conn to close the connection
* before returning. PBSE_SOCKET_CLOSE is the code
* for a successful return. But which ever retun
* code is iused it must terminate the while loop
* in start_process_pbs_server_port.
*************************************************/
void *svr_is_request(
void *v)
{
int command = 0;
int ret = DIS_SUCCESS;
int i;
int err;
char nodename[PBS_MAXHOSTNAME];
int perm = ATR_DFLAG_MGRD | ATR_DFLAG_MGWR;
unsigned long ipaddr;
unsigned short mom_port;
unsigned short rm_port;
unsigned long tmpaddr;
struct sockaddr_in addr;
struct pbsnode *node = NULL;
char log_buf[LOCAL_LOG_BUF_SIZE+1];
char msg_buf[80];
char tmp[80];
int version;
struct tcp_chan *chan;
long *args;
is_request_info *isr = (is_request_info *)v;
if (isr == NULL)
return(NULL);
chan = isr->chan;
args = isr->args;
version = disrsi(chan, &ret);
if (ret != DIS_SUCCESS)
{
log_err(-1, __func__, "Cannot read version - skipping this request.\n");
close_conn(chan->sock, FALSE);
DIS_tcp_cleanup(chan);
return(NULL);
}
command = disrsi(chan, &ret);
if (ret != DIS_SUCCESS)
{
snprintf(log_buf, sizeof(log_buf), "could not read command: %d", ret);
log_err(-1, __func__, log_buf);
close_conn(chan->sock, FALSE);
DIS_tcp_cleanup(chan);
return(NULL);
}
if (LOGLEVEL >= 4)
{
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"message received from sock %d (version %d)",
chan->sock,
version);
log_event(PBSEVENT_ADMIN, PBS_EVENTCLASS_SERVER, __func__, log_buf);
}
/* Just a note to let us know we only do IPv4 for now */
addr.sin_family = AF_INET;
memcpy(&addr.sin_addr, (void *)&args[1], sizeof(struct in_addr));
addr.sin_port = args[2];
if (version != IS_PROTOCOL_VER)
{
netaddr_long(args[1], tmp);
sprintf(msg_buf, "%s:%ld", tmp, args[2]);
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "protocol version %d unknown from %s",
version,
msg_buf);
log_err(-1, __func__, log_buf);
close_conn(chan->sock, FALSE);
DIS_tcp_cleanup(chan);
return(NULL);
}
/* check that machine is known */
mom_port = disrsi(chan, &ret);
rm_port = disrsi(chan, &ret);
if (LOGLEVEL >= 3)
{
netaddr_long(args[1], tmp);
sprintf(msg_buf, "%s:%ld", tmp, args[2]);
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"message received from addr %s: mom_port %d - rm_port %d",
msg_buf,
mom_port,
rm_port);
log_event(PBSEVENT_ADMIN,PBS_EVENTCLASS_SERVER,__func__,log_buf);
}
ipaddr = args[1];
if ((node = AVL_find(ipaddr, mom_port, ipaddrs)) != NULL)
{
node->lock_node(__func__, "AVL_find", LOGLEVEL);
} /* END if AVL_find != NULL) */
else if (allow_any_mom)
{
const char *name = get_cached_nameinfo(&addr);
if (name != NULL)
snprintf(nodename, sizeof(nodename), "%s", name);
else if (getnameinfo((struct sockaddr *)&addr, sizeof(addr), nodename, sizeof(nodename)-1, NULL, 0, 0) != 0)
{
tmpaddr = ntohl(addr.sin_addr.s_addr);
sprintf(nodename, "0x%lX", tmpaddr);
}
else
insert_addr_name_info(NULL, nodename);
err = create_partial_pbs_node(nodename, ipaddr, perm);
if (err == PBSE_NONE)
{
node = AVL_find(ipaddr, 0, ipaddrs);
node->lock_node(__func__, "no error", LOGLEVEL);
}
}
if (node == NULL)
{
/* node not listed in trusted ipaddrs list */
netaddr_long(args[1], tmp);
sprintf(msg_buf, "%s:%ld", tmp, args[2]);
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"bad attempt to connect from %s (address not trusted - check entry in server_priv/nodes)",
msg_buf);
if (LOGLEVEL >= 2)
{
log_record(PBSEVENT_SCHED, PBS_EVENTCLASS_REQUEST, __func__, log_buf);
}
else
{
log_err(-1, __func__, log_buf);
}
close_conn(chan->sock, FALSE);
DIS_tcp_cleanup(chan);
return(NULL);
}
if (LOGLEVEL >= 3)
{
netaddr_long(args[1], tmp);
sprintf(msg_buf, "%s:%ld", tmp, args[2]);
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"message %s (%d) received from mom on host %s (%s) (sock %d)",
PBSServerCmds2[command],
command,
node->get_name(),
msg_buf,
chan->sock);
log_event(PBSEVENT_ADMIN,PBS_EVENTCLASS_SERVER,__func__,log_buf);
}
mutex_mgr node_mutex(&node->nd_mutex, true);
switch (command)
{
case IS_NULL: /* a ping from server */
DBPRT(("%s: IS_NULL\n", __func__))
break;
case IS_UPDATE:
DBPRT(("%s: IS_UPDATE\n", __func__))
i = disrui(chan, &ret);
if (ret != DIS_SUCCESS)
{
if (LOGLEVEL >= 1)
{
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"IS_UPDATE error %d on node %s\n", ret, node->get_name());
log_err(ret, __func__, log_buf);
}
goto err;
}
DBPRT(("%s: IS_UPDATE %s 0x%x\n", __func__, node->get_name(), i))
update_node_state(node, i);
if ((node->nd_state & INUSE_DOWN) != 0)
{
node->nd_mom_reported_down = TRUE;
}
break;
case IS_STATUS:
{
std::string node_name = node->get_name();
if (LOGLEVEL >= 2)
{
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"IS_STATUS received from %s", node->get_name());
log_event(PBSEVENT_ADMIN, PBS_EVENTCLASS_SERVER, __func__, log_buf);
}
node_mutex.unlock();
ret = is_stat_get(node_name.c_str(), chan);
node = find_nodebyname(node_name.c_str());
if (node != NULL)
{
node->nd_stream = -1;
node_mutex.mark_as_locked();
if (ret == SEND_HELLO)
{
//struct hello_info *hi = new hello_info(node->nd_id);
write_tcp_reply(chan, IS_PROTOCOL, IS_PROTOCOL_VER, IS_STATUS, DIS_SUCCESS);
hierarchy_handler.sendHierarchyToANode(node);
ret = DIS_SUCCESS;
}
else
write_tcp_reply(chan,IS_PROTOCOL,IS_PROTOCOL_VER,IS_STATUS,ret);
}
if (ret != DIS_SUCCESS)
{
if (LOGLEVEL >= 1)
{
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"IS_STATUS error %d on node %s", ret, node_name.c_str());
log_err(ret, __func__, log_buf);
}
goto err;
}
break;
}
default:
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"unknown command %d sent from %s",
command,
node->get_name());
log_err(-1, __func__, log_buf);
goto err;
break;
} /* END switch (command) */
/* must be closed because mom opens and closes this connection each time */
close_conn(chan->sock, FALSE);
DIS_tcp_cleanup(chan);
return(NULL);
err:
/* a DIS write error has occurred */
if (node != NULL)
{
if (LOGLEVEL >= 1)
{
DBPRT(("%s: error processing node %s\n",
__func__,
node->get_name()))
}
netaddr_long(args[1], tmp);
sprintf(msg_buf, "%s:%ld", tmp, args[2]);
sprintf(log_buf, "%s from %s(%s)",
dis_emsg[ret],
node->get_name(),
msg_buf);
}
else
{
/*************************************************
* svr_is_request
*
* Return: svr_is_request always returns a non-zero value
* and it must call close_conn to close the connection
* before returning. PBSE_SOCKET_CLOSE is the code
* for a successful return. But which ever retun
* code is iused it must terminate the while loop
* in start_process_pbs_server_port.
*************************************************/
int svr_is_request(
struct tcp_chan *chan,
int version)
{
int command = 0;
int ret = DIS_SUCCESS;
int i;
int err;
char nodename[PBS_MAXHOSTNAME];
int perm = ATR_DFLAG_MGRD | ATR_DFLAG_MGWR;
unsigned long ipaddr;
unsigned short mom_port;
unsigned short rm_port;
unsigned long tmpaddr;
struct sockaddr_in *addr = NULL;
struct sockaddr s_addr;
unsigned int len = sizeof(s_addr);
struct pbsnode *node = NULL;
char *node_name = NULL;
char log_buf[LOCAL_LOG_BUF_SIZE+1];
command = disrsi(chan, &ret);
if (ret != DIS_SUCCESS)
goto err;
if (LOGLEVEL >= 4)
{
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"message received from sock %d (version %d)",
chan->sock,
version);
log_event(PBSEVENT_ADMIN,PBS_EVENTCLASS_SERVER,__func__,log_buf);
}
if (getpeername(chan->sock, &s_addr, &len) != 0)
{
close_conn(chan->sock, FALSE);
log_err(errno,__func__, (char *)"Cannot get socket name using getpeername\n");
return(PBSE_SOCKET_CLOSE);
}
addr = (struct sockaddr_in *)&s_addr;
if (version != IS_PROTOCOL_VER)
{
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "protocol version %d unknown from %s",
version,
netaddr(addr));
log_err(-1, __func__, log_buf);
close_conn(chan->sock, FALSE);
return PBSE_SOCKET_DATA;
}
/* check that machine is known */
mom_port = disrsi(chan, &ret);
rm_port = disrsi(chan, &ret);
if (LOGLEVEL >= 3)
{
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"message received from addr %s: mom_port %d - rm_port %d",
netaddr(addr),
mom_port,
rm_port);
log_event(PBSEVENT_ADMIN,PBS_EVENTCLASS_SERVER,__func__,log_buf);
}
ipaddr = ntohl(addr->sin_addr.s_addr);
if ((node = AVL_find(ipaddr, mom_port, ipaddrs)) != NULL)
{
lock_node(node, __func__, "AVL_find", LOGLEVEL);
} /* END if AVL_find != NULL) */
else if (allow_any_mom)
{
char *name = get_cached_nameinfo(addr);
if (name != NULL)
snprintf(nodename, sizeof(nodename), "%s", name);
else if (getnameinfo(&s_addr, len, nodename, sizeof(nodename)-1, NULL, 0, 0) != 0)
{
tmpaddr = ntohl(addr->sin_addr.s_addr);
sprintf(nodename, "0x%lX", tmpaddr);
}
else
insert_addr_name_info(nodename, NULL, addr);
err = create_partial_pbs_node(nodename, ipaddr, perm);
if (err == PBSE_NONE)
{
node = AVL_find(ipaddr, 0, ipaddrs);
lock_node(node, __func__, "no error", LOGLEVEL);
}
}
if (node == NULL)
{
/* node not listed in trusted ipaddrs list */
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"bad attempt to connect from %s (address not trusted - check entry in server_priv/nodes)",
netaddr(addr));
if (LOGLEVEL >= 2)
{
log_record(PBSEVENT_SCHED, PBS_EVENTCLASS_REQUEST, __func__, log_buf);
}
else
{
log_err(-1, __func__, log_buf);
}
close_conn(chan->sock, FALSE);
return PBSE_SOCKET_CLOSE;
}
if (LOGLEVEL >= 3)
{
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"message %s (%d) received from mom on host %s (%s) (sock %d)",
PBSServerCmds2[command],
command,
node->nd_name,
netaddr(addr),
chan->sock);
log_event(PBSEVENT_ADMIN,PBS_EVENTCLASS_SERVER,__func__,log_buf);
}
switch (command)
{
case IS_NULL: /* a ping from server */
DBPRT(("%s: IS_NULL\n", __func__))
break;
case IS_UPDATE:
DBPRT(("%s: IS_UPDATE\n", __func__))
i = disrui(chan, &ret);
if (ret != DIS_SUCCESS)
{
if (LOGLEVEL >= 1)
{
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"IS_UPDATE error %d on node %s\n", ret, node->nd_name);
log_err(ret, __func__, log_buf);
}
goto err;
}
DBPRT(("%s: IS_UPDATE %s 0x%x\n", __func__, node->nd_name, i))
update_node_state(node, i);
if ((node->nd_state & INUSE_DOWN) != 0)
{
node->nd_mom_reported_down = TRUE;
}
break;
case IS_STATUS:
if (LOGLEVEL >= 2)
{
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"IS_STATUS received from %s", node->nd_name);
log_event(PBSEVENT_ADMIN, PBS_EVENTCLASS_SERVER, __func__, log_buf);
}
if ((node_name = strdup(node->nd_name)) == NULL)
goto err;
unlock_node(node, __func__, "before is_stat_get", LOGLEVEL);
ret = is_stat_get(node_name, chan);
node = find_nodebyname(node_name);
if (ret == SEND_HELLO)
{
struct hello_info *hi = (struct hello_info *)calloc(1, sizeof(struct hello_info));
write_tcp_reply(chan, IS_PROTOCOL, IS_PROTOCOL_VER, IS_STATUS, DIS_SUCCESS);
hi->name = strdup(node_name);
enqueue_threadpool_request(send_hierarchy_threadtask, hi);
ret = DIS_SUCCESS;
}
else
write_tcp_reply(chan,IS_PROTOCOL,IS_PROTOCOL_VER,IS_STATUS,ret);
if(node != NULL)
node->nd_stream = -1;
if (ret != DIS_SUCCESS)
{
if (LOGLEVEL >= 1)
{
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"IS_STATUS error %d on node %s", ret, node_name);
log_err(ret, __func__, log_buf);
}
free(node_name);
goto err;
}
free(node_name);
break;
default:
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"unknown command %d sent from %s",
command,
node->nd_name);
log_err(-1, __func__, log_buf);
goto err;
break;
} /* END switch (command) */
/* must be closed because mom opens and closes this connection each time */
close_conn(chan->sock, FALSE);
if(node != NULL)
unlock_node(node, __func__, "close", LOGLEVEL);
return PBSE_SOCKET_CLOSE;
err:
/* a DIS write error has occurred */
if (node != NULL)
{
if (LOGLEVEL >= 1)
{
DBPRT(("%s: error processing node %s\n",
__func__,
node->nd_name))
}
sprintf(log_buf, "%s from %s(%s)",
dis_emsg[ret],
node->nd_name,
netaddr(addr));
unlock_node(node, __func__, "err", LOGLEVEL);
}
else
{