/**
* @brief
* Open a connection with a pbs server.
* Do not allow TCP to block us if Server host is down
* At this point, this does not attempt to find a fail_over Server
*
* @param[in] server - specifies the server to which to connect
* @param[in] tout - timeout value for select
*
* @return int
* @retval >= 0 index to the internal connection table representing the
* connection made.
* @retval -1 error encountered in getting index
*/
int
pbs_connect_noblk(char *server, int tout)
{
int out;
int i;
pbs_socklen_t l;
int n;
struct timeval tv;
fd_set fdset;
struct batch_reply *reply;
char server_name[PBS_MAXSERVERNAME+1];
unsigned int server_port;
struct addrinfo *aip, *pai;
struct addrinfo hints;
struct sockaddr_in *inp;
short int connect_err = 0;
struct sockaddr_in sockname;
pbs_socklen_t socknamelen;
#ifdef WIN32
int non_block = 1;
struct sockaddr_in to_sock;
struct sockaddr_in from_sock;
#endif
#ifndef WIN32
int nflg;
int oflg;
#endif
/* initialize the thread context data, if not already initialized */
if (pbs_client_thread_init_thread_context() != 0)
return -1;
if (pbs_loadconf(0) == 0)
return -1;
/* get server host and port */
server = PBS_get_server(server, server_name, &server_port);
if (server == NULL) {
pbs_errno = PBSE_NOSERVER;
return -1;
}
/* Reserve a connection state record */
if (pbs_client_thread_lock_conntable() != 0)
return -1;
out = -1;
for (i=1;i<NCONNECTS;i++) {
if (connection[i].ch_inuse) continue;
out = i;
connection[out].ch_inuse = 1;
connection[out].ch_errno = 0;
connection[out].ch_socket= -1;
connection[out].ch_errtxt = NULL;
break;
}
if (pbs_client_thread_unlock_conntable() != 0)
return -1; /* pbs_errno set by the function */
if (out < 0) {
pbs_errno = PBSE_NOCONNECTS;
return -1;
}
/* get socket */
#ifdef WIN32
/* the following lousy hack is needed since the socket call needs */
/* SYSTEMROOT env variable properly set! */
if (getenv("SYSTEMROOT") == NULL) {
setenv("SYSTEMROOT", "C:\\WINNT", 1);
setenv("SystemRoot", "C:\\WINNT", 1);
}
connection[out].ch_socket = socket(AF_INET, SOCK_STREAM, 0);
if (connection[out].ch_socket < 0) {
setenv("SYSTEMROOT", "C:\\WINDOWS", 1);
setenv("SystemRoot", "C:\\WINDOWS", 1);
connection[out].ch_socket = socket(AF_INET, SOCK_STREAM, 0);
}
#else
connection[out].ch_socket = socket(AF_INET, SOCK_STREAM, 0);
#endif
if (connection[out].ch_socket < 0) {
connection[out].ch_inuse = 0;
pbs_errno = ERRORNO;
return -1;
}
/* set socket non-blocking */
#ifdef WIN32
if (ioctlsocket(connection[out].ch_socket, FIONBIO, &non_block) == SOCKET_ERROR)
#else
oflg = fcntl(connection[out].ch_socket, F_GETFL) & ~O_ACCMODE;
nflg = oflg | O_NONBLOCK;
if (fcntl(connection[out].ch_socket, F_SETFL, nflg) == -1)
#endif
goto err;
/* and connect... */
strcpy(pbs_server, server); /* set for error messages from commands */
memset(&hints, 0, sizeof(struct addrinfo));
/*
* Why do we use AF_UNSPEC rather than AF_INET? Some
* implementations of getaddrinfo() will take an IPv6
* address and map it to an IPv4 one if we ask for AF_INET
* only. We don't want that - we want only the addresses
* that are genuinely, natively, IPv4 so we start with
* AF_UNSPEC and filter ai_family below.
*/
hints.ai_family = AF_UNSPEC;
hints.ai_socktype = SOCK_STREAM;
hints.ai_protocol = IPPROTO_TCP;
if (getaddrinfo(server, NULL, &hints, &pai) != 0) {
CLOSESOCKET(connection[out].ch_socket);
connection[out].ch_inuse = 0;
pbs_errno = PBSE_BADHOST;
return -1;
}
for (aip = pai; aip != NULL; aip = aip->ai_next) {
/* skip non-IPv4 addresses */
if (aip->ai_family == AF_INET) {
inp = (struct sockaddr_in *) aip->ai_addr;
break;
}
}
if (aip == NULL) {
/* treat no IPv4 addresses as getaddrinfo() failure */
CLOSESOCKET(connection[out].ch_socket);
connection[out].ch_inuse = 0;
pbs_errno = PBSE_BADHOST;
freeaddrinfo(pai);
return -1;
} else
inp->sin_port = htons(server_port);
if (connect(connection[out].ch_socket,
aip->ai_addr,
aip->ai_addrlen) < 0) {
connect_err = 1;
}
if (connect_err == 1)
{
/* connect attempt failed */
pbs_errno = ERRORNO;
switch (pbs_errno) {
#ifdef WIN32
case WSAEWOULDBLOCK:
#else
case EINPROGRESS:
case EWOULDBLOCK:
#endif
while (1) {
FD_ZERO(&fdset);
FD_SET(connection[out].ch_socket, &fdset);
tv.tv_sec = tout;
tv.tv_usec = 0;
n = select(connection[out].ch_socket+1, NULL, &fdset, NULL, &tv);
if (n > 0) {
pbs_errno = 0;
l = sizeof(pbs_errno);
(void)getsockopt(connection[out].ch_socket,
SOL_SOCKET, SO_ERROR,
&pbs_errno, &l);
if (pbs_errno == 0)
break;
else
goto err;
} if ((n < 0) &&
#ifdef WIN32
(ERRORNO == WSAEINTR)
#else
(ERRORNO == EINTR)
#endif
) {
continue;
} else {
goto err;
}
}
break;
default:
err:
CLOSESOCKET(connection[out].ch_socket);
connection[out].ch_inuse = 0;
freeaddrinfo(pai);
return -1; /* cannot connect */
}
}
freeaddrinfo(pai);
/* reset socket blocking */
#ifdef WIN32
non_block = 0;
if (ioctlsocket(connection[out].ch_socket, FIONBIO, &non_block) == SOCKET_ERROR)
#else
if (fcntl(connection[out].ch_socket, F_SETFL, oflg) < 0)
#endif
goto err;
/*
* multiple threads cant get the same connection id above,
* so no need to lock this piece of code
*/
/* setup connection level thread context */
if (pbs_client_thread_init_connect_context(out) != 0) {
CLOSESOCKET(connection[out].ch_socket);
connection[out].ch_inuse = 0;
/* pbs_errno set by the pbs_connect_init_context routine */
return -1;
}
/*
* even though the following is communication with server on
* a connection handle, it does not need to be lock since
* this connection handle has not be returned back yet to the client
* so others threads cannot use it
*/
/* send "dummy" connect message */
DIS_tcp_setup(connection[out].ch_socket);
if ((i = encode_DIS_ReqHdr(connection[out].ch_socket,
PBS_BATCH_Connect, pbs_current_user)) ||
(i = encode_DIS_ReqExtend(connection[out].ch_socket,
NULL))) {
pbs_errno = PBSE_SYSTEM;
return -1;
}
if (DIS_tcp_wflush(connection[out].ch_socket)) {
pbs_errno = PBSE_SYSTEM;
return -1;
}
reply = PBSD_rdrpy(out);
PBSD_FreeReply(reply);
/*do configured authentication (kerberos, pbs_iff, whatever)*/
/*Get the socket port for engage_authentication()*/
socknamelen = sizeof(sockname);
if (getsockname(connection[out].ch_socket, (struct sockaddr *)&sockname, &socknamelen))
return -1;
if (engage_authentication(connection[out].ch_socket,
server,
server_port,
&sockname) == -1) {
CLOSESOCKET(connection[out].ch_socket);
connection[out].ch_inuse = 0;
pbs_errno = PBSE_PERM;
return -1;
}
/* setup DIS support routines for following pbs_* calls */
DIS_tcp_setup(connection[out].ch_socket);
pbs_tcp_timeout = PBS_DIS_TCP_TIMEOUT_VLONG; /* set for 3 hours */
return out;
}
/**
* @brief
* establishes connection to host
*
* @param[in] momhost - The name of the MOM host to connect to.
*
* @return int
* @retval connection slot success
* @retval -1 error
*
*/
static int
pbs_connect2mom(char *momhost)
{
int conn, sd;
#if defined(__hpux)
struct sockaddr_in mom_addr;
struct hostent *hp;
#else
struct addrinfo *aip, *pai;
struct addrinfo hints;
struct sockaddr_in *inp;
#endif
#if 0
struct passwd *pw;
#endif
/* initialize the thread context data, if not already initialized */
if (pbs_client_thread_init_thread_context() != 0)
return -1;
if (pbs_loadconf(0) == 0)
return -1;
if (pbs_client_thread_lock_conntable() != 0)
return -1;
/*
* Find an available connection slot.
*/
for (conn=1; conn<NCONNECTS; conn++) {
if (connection[conn].ch_inuse)
continue;
break;
}
if (conn >= NCONNECTS) {
pbs_errno = PBSE_NOCONNECTS;
goto err;
}
/*
* Ensure we have something valid to connect to.
*/
if (!momhost)
momhost = "localhost";
if (strlen(momhost) <= 0)
momhost = "localhost";
#if defined(__hpux)
hp = gethostbyname(momhost);
if (!hp) {
pbs_errno = PBSE_BADHOST;
goto err;
}
#else
memset(&hints, 0, sizeof(struct addrinfo));
/*
* Why do we use AF_UNSPEC rather than AF_INET? Some
* implementations of getaddrinfo() will take an IPv6
* address and map it to an IPv4 one if we ask for AF_INET
* only. We don't want that - we want only the addresses
* that are genuinely, natively, IPv4 so we start with
* AF_UNSPEC and filter ai_family below.
*/
hints.ai_family = AF_UNSPEC;
hints.ai_socktype = SOCK_STREAM;
hints.ai_protocol = IPPROTO_TCP;
if (getaddrinfo(momhost, NULL, &hints, &pai) != 0) {
pbs_errno = PBSE_BADHOST;
goto err;
}
for (aip = pai; aip != NULL; aip = aip->ai_next) {
/* skip non-IPv4 addresses */
if (aip->ai_family == AF_INET) {
inp = (struct sockaddr_in *) aip->ai_addr;
break;
}
}
if (aip == NULL) {
pbs_errno = PBSE_BADHOST;
goto err;
} else
inp->sin_port = htons(pbs_conf.mom_service_port);
#endif
#if 0
/*
* Prepare for authentication before we make the connection.
*/
pbs_current_uid = getuid();
if ((pw = getpwuid(pbs_current_uid)) == NULL) {
pbs_errno = PBSE_SYSTEM;
goto err;
}
strcpy(pbs_current_user, pw->pw_name);
#endif
/*
* Establish a connection.
*/
sd = socket(AF_INET, SOCK_STREAM, 0);
if (sd < 0) {
pbs_errno = PBSE_PROTOCOL;
goto err;
}
#if defined(__hpux)
mom_addr.sin_family = AF_INET;
mom_addr.sin_port = htons(pbs_conf.mom_service_port);
memcpy((char *)&mom_addr.sin_addr, hp->h_addr_list[0], hp->h_length);
if (connect(sd, (struct sockaddr *)&mom_addr, sizeof(mom_addr)) < 0) {
close(sd);
pbs_errno = errno;
goto err;
}
#else
if (connect(sd, aip->ai_addr, aip->ai_addrlen) < 0) {
close(sd);
pbs_errno = errno;
freeaddrinfo(pai);
goto err;
}
freeaddrinfo(pai);
#endif
#if 0
/*
* Have pbs_iff authenticate the connection.
*/
if (PBSD_authenticate(sd)) {
close(sd);
pbs_errno = PBSE_PERM;
return -1;
}
#endif
/*
* Setup DIS support routines.
*/
DIS_tcp_setup(sd);
pbs_tcp_timeout = PBS_DIS_TCP_TIMEOUT_VLONG;
/*
* Register the connection slot as in use.
*/
connection[conn].ch_inuse = 1;
connection[conn].ch_errno = 0;
connection[conn].ch_socket= sd;
connection[conn].ch_errtxt = (char *)NULL;
/* setup connection level thread context */
if (pbs_client_thread_init_connect_context(conn) != 0) {
close(connection[conn].ch_socket);
connection[conn].ch_inuse = 0;
/* pbs_errno would be set by the pbs_connect_init_context routine */
goto err;
}
if (pbs_client_thread_unlock_conntable() != 0)
return -1;
return conn;
err:
(void)pbs_client_thread_unlock_conntable();
return -1;
}
/**
* @brief
* Makes a PBS_BATCH_Connect request to 'server'.
*
* @param[in] server - the hostname of the pbs server to connect to.
* @param[in] extend_data - a string to send as "extend" data.
*
* @return int
* @retval >= 0 index to the internal connection table representing the
* connection made.
* @retval -1 error encountered setting up the connection.
*/
int
__pbs_connect_extend(char *server, char *extend_data)
{
struct sockaddr_in server_addr;
struct sockaddr_in my_sockaddr;
int out;
int i;
int f;
char *altservers[2];
int have_alt = 0;
struct batch_reply *reply;
char server_name[PBS_MAXSERVERNAME+1];
unsigned int server_port;
struct sockaddr_in sockname;
pbs_socklen_t socknamelen;
#ifdef WIN32
struct sockaddr_in to_sock;
struct sockaddr_in from_sock;
#endif
#ifndef WIN32
char pbsrc[_POSIX_PATH_MAX];
struct stat sb;
int using_secondary = 0;
#endif /* not WIN32 */
/* initialize the thread context data, if not already initialized */
if (pbs_client_thread_init_thread_context() != 0)
return -1;
if (pbs_loadconf(0) == 0)
return -1;
/* get server host and port */
server = PBS_get_server(server, server_name, &server_port);
if (server == NULL) {
pbs_errno = PBSE_NOSERVER;
return -1;
}
if (pbs_conf.pbs_primary && pbs_conf.pbs_secondary) {
/* failover configuered ... */
if (hostnmcmp(server, pbs_conf.pbs_primary) == 0) {
have_alt = 1;
/* We want to try the one last seen as "up" first to not */
/* have connection delays. If the primary was up, there */
/* is no .pbsrc.NAME file. If the last command connected */
/* to the Secondary, then it created the .pbsrc.USER file. */
/* see if already seen Primary down */
#ifdef WIN32
/* due to windows quirks, all try both in same order */
altservers[0] = pbs_conf.pbs_primary;
altservers[1] = pbs_conf.pbs_secondary;
#else
(void)snprintf(pbsrc, _POSIX_PATH_MAX, "%s/.pbsrc.%s", pbs_conf.pbs_tmpdir, pbs_current_user);
if (stat(pbsrc, &sb) == -1) {
/* try primary first */
altservers[0] = pbs_conf.pbs_primary;
altservers[1] = pbs_conf.pbs_secondary;
using_secondary = 0;
} else {
/* try secondary first */
altservers[0] = pbs_conf.pbs_secondary;
altservers[1] = pbs_conf.pbs_primary;
using_secondary = 1;
}
#endif
}
}
/* if specific host name declared for the host on which */
/* this client is running, get its address */
if (pbs_conf.pbs_public_host_name) {
if (get_hostsockaddr(pbs_conf.pbs_public_host_name, &my_sockaddr) != 0)
return -1; /* pbs_errno was set */
}
/* Reserve a connection state record */
if (pbs_client_thread_lock_conntable() != 0)
return -1;
out = -1;
for (i=1;i<NCONNECTS;i++) {
if (connection[i].ch_inuse) continue;
out = i;
connection[out].ch_errno = 0;
connection[out].ch_socket= -1;
connection[out].ch_errtxt = NULL;
connection[out].ch_inuse = 1; /* reserve the socket */
break;
}
if (pbs_client_thread_unlock_conntable() != 0)
return -1; /* pbs_errno set by the function */
if (out < 0) {
pbs_errno = PBSE_NOCONNECTS;
return -1;
}
/*
* connect to server ...
* If attempt to connect fails and if Failover configured and
* if attempting to connect to Primary, try the Secondary
* if attempting to connect to Secondary, try the Primary
*/
for (i=0; i<(have_alt+1); ++i) {
/* get socket */
#ifdef WIN32
/* the following lousy hack is needed since the socket call needs */
/* SYSTEMROOT env variable properly set! */
if (getenv("SYSTEMROOT") == NULL) {
setenv("SYSTEMROOT", "C:\\WINNT", 1);
setenv("SystemRoot", "C:\\WINNT", 1);
}
connection[out].ch_socket = socket(AF_INET, SOCK_STREAM, 0);
if (connection[out].ch_socket < 0) {
setenv("SYSTEMROOT", "C:\\WINDOWS", 1);
setenv("SystemRoot", "C:\\WINDOWS", 1);
connection[out].ch_socket = socket(AF_INET, SOCK_STREAM, 0);
}
#else
connection[out].ch_socket = socket(AF_INET, SOCK_STREAM, 0);
#endif
if (connection[out].ch_socket < 0) {
connection[out].ch_inuse = 0;
pbs_errno = errno;
return -1;
}
/* and connect... */
if (have_alt) {
server = altservers[i];
}
strcpy(pbs_server, server); /* set for error messages from commands */
/* If a specific host name is defined which the client should use */
if (pbs_conf.pbs_public_host_name) {
/* my address will be in my_sockaddr, bind the socket to it */
my_sockaddr.sin_port = 0;
if (bind(connection[out].ch_socket, (struct sockaddr *)&my_sockaddr, sizeof(my_sockaddr)) != 0) {
return -1;
}
}
if (get_hostsockaddr(server, &server_addr) != 0)
return -1;
server_addr.sin_port = htons(server_port);
if (connect(connection[out].ch_socket,
(struct sockaddr *)&server_addr,
sizeof(struct sockaddr)) == 0) {
break;
} else {
/* connect attempt failed */
CLOSESOCKET(connection[out].ch_socket);
pbs_errno = errno;
}
}
if (i >= (have_alt+1)) {
connection[out].ch_inuse = 0;
return -1; /* cannot connect */
}
#ifndef WIN32
if (have_alt && (i == 1)) {
/* had to use the second listed server ... */
if (using_secondary == 1) {
/* remove file that causes trying the Secondary first */
unlink(pbsrc);
} else {
/* create file that causes trying the Primary first */
f = open(pbsrc, O_WRONLY|O_CREAT, 0200);
if (f != -1)
(void)close(f);
}
}
#endif
/* setup connection level thread context */
if (pbs_client_thread_init_connect_context(out) != 0) {
CLOSESOCKET(connection[out].ch_socket);
connection[out].ch_inuse = 0;
/* pbs_errno set by the pbs_connect_init_context routine */
return -1;
}
/*
* No need for global lock now on, since rest of the code
* is only communication on a connection handle.
* But we dont need to lock the connection handle, since this
* connection handle is not yet been returned to the client
*/
/* The following code was originally put in for HPUX systems to deal
* with the issue where returning from the connect() call doesn't
* mean the connection is complete. However, this has also been
* experienced in some Linux ppc64 systems like js-2. Decision was
* made to enable this harmless code for all architectures.
* FIX: Need to use the socket to send
* a message to complete the process. For IFF authentication there is
* no leading authentication message needing to be sent on the client
* socket, so will send a "dummy" message and discard the replyback.
*/
#if !defined(PBS_SECURITY ) || (PBS_SECURITY == STD )
DIS_tcp_setup(connection[out].ch_socket);
if ((i = encode_DIS_ReqHdr(connection[out].ch_socket,
PBS_BATCH_Connect, pbs_current_user)) ||
(i = encode_DIS_ReqExtend(connection[out].ch_socket,
extend_data))) {
pbs_errno = PBSE_SYSTEM;
return -1;
}
if (DIS_tcp_wflush(connection[out].ch_socket)) {
pbs_errno = PBSE_SYSTEM;
return -1;
}
reply = PBSD_rdrpy(out);
PBSD_FreeReply(reply);
#endif /* PBS_SECURITY ... */
/*do configured authentication (kerberos, pbs_iff, whatever)*/
/*Get the socket port for engage_authentication() */
socknamelen = sizeof(sockname);
if (getsockname(connection[out].ch_socket, (struct sockaddr *)&sockname, &socknamelen))
return -1;
if (engage_authentication(connection[out].ch_socket,
server,
server_port,
&sockname) == -1) {
CLOSESOCKET(connection[out].ch_socket);
connection[out].ch_inuse = 0;
pbs_errno = PBSE_PERM;
return -1;
}
/* setup DIS support routines for following pbs_* calls */
DIS_tcp_setup(connection[out].ch_socket);
pbs_tcp_timeout = PBS_DIS_TCP_TIMEOUT_VLONG; /* set for 3 hours */
/*
* Disable Nagle's algorithm on the TCP connection to server.
* Nagle's algorithm is hurting cmd-server communication.
*/
if (pbs_connection_set_nodelay(out) == -1) {
CLOSESOCKET(connection[out].ch_socket);
connection[out].ch_inuse = 0;
pbs_errno = PBSE_SYSTEM;
return -1;
}
return out;
}
